Dokumen - Tips Rate Controlled Separations

RATE-CONTROLLED SEPARATIONS
RATE-CONTROLLED
SEPARATIONS
PHILLIP C. W ANKA T
Schoo/ of Chemica/ Engineering, Purdue University,
West Lafayette, Indiana, USA
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

First edition 1990
Reprinted 1994
© 1994 Springer Science+Business Media Dordrecht

Originally published by Chapman & Hall in 1994
Softcover reprint of the hardcover 1st edition 1994
ISBN 978-94-010-4585-8 ISBN 978-94-011-1342-7 (eBook)

DOI 10.1007/978-94-011-1342-7
Apart from any fair dealing for the purposes of research or private study, or
criticism or review, as permitted under the UK Copyright Designs and Patents
Act, 1988, this publication may not be reproduced, stored, or transmitted, in
any form or by any means, without the prior permission in writing of the
publishers, or in the case of reprographic reproduction only in accordance with
the terms of the licences issued by the Copyright Licensing Agency in the UK,
or in accordance with the terms of licences issued by the appropriate
Reproduction Rights Organization outside the UK. Enquiries concerning
reproduction outside the terms stated here should be sent to the publishers at the
Glasgow address printed on this page.
The publisher makes no representation, express or implied, with regard to the
accuracy of the information contained in this book and cannot accept any legal
responsibility or liability for any errors or omissions that may be made.
A Catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data available
® Printed on acid-free text paper, manufactured with ANSI/NISO Z39.48-
1992 and ANIS/NISO Z39.48-1984 (Permanence of Paper)
PREFACE
Separations have always been very important in chemical engineering.

This importance has recently escalated with the imminent emergence of new
industries in biotechnology and high-performance materials. Separations will
continue to remain important in bulk chemical manufacturing, petroleum pro-
cessing, and the other standard areas of chemical engineering interest.
The development of new industries requiring the expertise of chemical

engineers leads to problems and opportunities for chemical engineering educa-
tion. Chemical engineering students need to be prepared for both the "known
future" and the "unknown future." The known future includes the use of stan-
dard chemical engineering separation methods such as distillation and absorp-
tion which will remain important for many years. The unknown future
involves the use of many relatively new separation methods such as adsorption,
chromatography, electrophoresis, membrane separations.
A major question for chemical engineering education is what to teach. In

the area of separations my personal answer has been to require undergraduates
to study classical separations including distillation, adsorption and extraction.
Then an elective course on newer methods which require a mass transfer
analysis should be made available to seniors and graduate students. I would
not mind if this second course were required of graduate students; certainly,
that would be preferable to an additional distillation course.
My first book, Equilibrium-Staged Separations, was my response for the

required undergraduate course. This book is my response to both the proposed
second course, and to practicing chemical engineers who missed this material
when they were in school.
v
VI
This book, Rate-Controlled Separations, includes separation processes

which require a rate analysis for complete understanding. This includes most
of the newer separation methods. The style of this book is similar to the style
of Equilibrium-Staged Separations and problem solving is emphasized
throughout. However, a higher level of mathematical analysis is required, and
an understanding of mass transfer is assumed.
The plan for this book is to start with crystallization which is essentially
equilibrium based. Sorption separations which can be (but seldom are)
operated as equilibrium staged systems are discussed next. Then membrane
separations which are inherently rate processes are discussed. Finally, a
progress report on selection and sequencing of separations is presented.
This book has been used in an elective course for seniors and graduate
students at Purdue over a period of several years. My experience has been that
the students have no great difficulty with crystallization even though this is
their first exposure to population balances. The students find that the chapters
on sorption separations are more difficult; possibly because of the inherent
batch nature of these processes. The average and better seniors and all of the
graduate students have been able to work their way through these difficulties.
The membrane methods seem to pose no unusual difficulty for the students.
Many people have helped me with the writing of this book. My students
have been most helpful in helping me develop clear methods to explain the
separation methods. My teaching assistants (when I had this luxury), Sung-Sup
Suh and Narasimhan Sundaram, have been very helpful in solving problems
and finding errors. Professors Linda Wang and David Graves used parts of this
book in their courses. Their comments were very helpful and have been incor-
porated into the text. Mr. Steve Leeper was very helpful in providing refer-
ences on membrane separations. Dr. Scott Rudge did a very thorough review
of Chapter 11, and he helped me appear to be reasonably knowledgeable in the
area of electrophoresis. Professors Lowell B. Koppel and William R.
Schowalter taught me my undergraduate and graduate courses in separations
and awakened my interest in the field. Professor C. Judson King's book con-
vinced me there were interesting research problems in separations. Professor
L.B. Rogers who was in Chemistry at Purdue gave me my introduction to
VII
chromatography when I audited his course. My interest in problem solving was

sparked by Professors Richard Noble and Donald Woods.
The keyboarding of this book was done by Mrs. Karen Parsons and Ms.
J an Gray. Their cheerfulness in the face of endless revisions is appreciated.
Ms. Barbara Hildebrandt patiently drew and redrew the figures.
Finally, my wife Dot has supported me when I thought I would never

finish, and my children Charles and Jennifer have provided light to my life.
Phillip C. Wankat
CONTENTS
Preface ................................................................................................................. v
1. Introduction .................................................................................................... 1
1.1. Part I: Crystallization.......................................................................... 1

1.2. Part II: Sorption and chromatography .................................................3
1.3. Part III: Membranes.............................................................................. 6
1.4. Part IV: Selection and sequencing ....................................................... 7
1.5. Problem solving .....................................................................................8
1.6. Summary - Objectives ......................................................................... 10
PART I. CRYSTALLIZATION
Nomenclature Chapters 2 to 5 ............................................................................. 13
2. Crystallization and Precipitation from Solution - Equilibrium Analysis ...... 19
2.1. Solubility .............................................................................................20

2.2. Equipment types .................................................................................. 28
2.3. Yield calculations based on solubility diagram ................................... 34
2.4. Phase equilibria and single stage systems ...........................................38
2.4.1. Binary systems ........................................................................ 38
2.4.2. Ternary systems...................................................................... 45
2.5. Fractional crystallization .................................................................... .52
2.5.1. Fractional crystallization processes with temperature
swings ...................................................................................53
2.5.2. Processes using counter-current cascades ..............................62
IX
x
2.6. Precipitation ......................................................................................... 67

2.7. Summary - Objectives ......................................................................... 70
References .......................................................................................... 71
Homework .......................................................................................... 73
3. Nucleation and Crystal Growth .................................................................... 80
3.1. Crystal shapes ..................................................................................... 80

3.2. Nucleation ........................................................................................... 84
3.2.1. Homogeneous nucleation ...................................................... 84
3.2.2. Heterogeneous and secondary nucleation ............................. 86
3.3. Crystal growth ....................................................................................92
3.3.1. Adsorption theory .................................................................. 92
3.3.2. Mass transfer theory .............................................................. 95
3.4. Summary - Objectives ........................................................................ 98
References .........................................................................................99
Homework ...................................................................................... 100
4. Population Balances and Crystal Size Distributions .................................. 103
4.1. Basic population balance equations ................................................ 103

4.2. Distributions for size independent growth ...................................... 106
4.3. Sieve analysis of CSD ..................................................................... 114
4.4. Effect of nucleation and growth kinetics on CSD .......................... .122
4.5. Seeding ............................................................................................ 126
4.6. Equipment modifications ................................................................. 128
4.6.1. Fines removal ...................................................................... 128
4.6.2. Classified product removal .................................................. 138
4.6.3. Methods of suspending solids ............................................. 140
4.7. CSD for complex crystallization kinetics ....................................... .142
4.8. Batch crystallization ........................................................................ 146
4.9. Downstream processing methods .................................................... 151
4.10. Summary - Objectives .................................................................... 152
References ..................................................................................... 152
Homework ..................................................................................... 154
XI
5. Crystallization from the Melt ...................................................................... 160
5.1. Fundamentals of melt crystallization ................................................ 161

5.1.1. Equilibrium ........................................................................... 161
5.1.2. Heat and mass transfer ......................................................... 163
5.2. Slurry processes ................................................................................ 165
5.2.1. Continuous staged processes ............................................... .l66
5.2.2. Batch staged crystallization .................................................. 169
5.2.3. Partial melting and drainage ................................................ .l7l
5.2.4. Column crystallization .......................................................... 176
5.3. Continuous solid phase ..................................................................... 183
5.3.1. Normal freezing .................................................................... 183
5.3.2. Zone melting......................................................................... 186
5.4. Summary - Objectives ...................................................................... 196
References ........................................................................................ 197
Homework ....................................................................................... 199
PART II. SORPTION AND CHROMATOGRAPHY
Nomenclature Chapters 6 to 10 ...................................................................... .207
6. Basics of Sorption in Packed Columns ...................................................... 217
6.1. Packing structure .............................................................................. 218

6.2. Adsorbents ........................................................................................222
6.3. Adsorption equilibrium ....................................................................228
6.4. Solute movement ..............................................................................239
6.5. Effects of changing thermodynamic variables ................................ .251
6.6. Theoretical assumptions and limitations of solute movement
theory ...............................................................................................266
6.7. Basic mass transfer equations ...........................................................268
6.8. Basic energy balances ......................................................................273
6.9. Local equilibrium theory ..................................................................275
6.10. Summary - Objectives ..................................................................... 277
References ...................................................................................... 277
Homework .....................................................................................280
Xli
7. Linear Theories of Sorption and Chromatography ..................................... 288
7.1. Types of chromatography .................................................................288

7.2. Application of solute movement theory to chromatography ........... .296
7.3. Superposition in linear systems ....................................................... .306
7.4. Linear dispersion model ................................................................... 308
7.5. Linear staged models for chromatography ...................................... .313
7.6. Use of Gaussian solution .................................................................. 316
7.7. Van Deemter equation ...................................................................... 326
7.8. Rosen's solution ............................................................................... 331
7.9. Application of the linear models ..................................................... .334
7.10. Programming and column switching .............................................. .336
7.11. Large-scale chromatography ........................................................... 341
References .......................................................................................... 349
Homework .......................................................................................... 354
8. Non-Linear Theories and Packed Bed Adsorption Systems ...................... 365
8.1. Mass transfer zone approach ............................................................ 365

8.2. Constant pattern solutions ................................................................370
8.3. Mass transfer correlations ................................................................376
8.4. Optimizing fixed bed systems ..........................................................383
8.5. Interacting adsorption systems ........................................................ .393
8.5.1. Interacting solutes ................................................................ 394
8.5.2. Adiabatic adsorbers ............................................................ .400
8.5.3. Velocity effects .................................................................... 404
8.6. Adsorption - Desorption Operations ............................................... .405
8.6.1. Thermal desorption of gases ............................................... .406
8.6.2. Activated carbon solvent recovery ..................................... .412
8.6.3. Processing liquids using thermal regeneration ................... .419
8.6.4. Pressure swing and vacuum swing adsorption ................... .421
8.6.5. Regeneration with purge and desorbent ............................. .431
8.6.6. Systems without regeneration ............................................. .435
Xlii
8.7. Design considerations ....................................................................... 437

8.8. Summary - Objectives ......................................................................438
References ........................................................................................440
Homework ........................................................................................445
9. Ion Exchange ..............................................................................................452
9.1. Basics of ion exchange .................................................................... .452

9.2. Ion-exchange resins ......................................................................... .455
9.3. Binary ion exchange equilibrium .................................................... .459
9.4. Ion movement theory ........................................................................463
9.5. Applications ..................................................................................... .475
9.6. Applications without exchange: Ion exclusion .............................. .484
9.7. Mass transfer in ion exchange systems ........................................... .487
9.8. Equipment design considerations .................................................... .489
9.9. Summary - Objectives ......................................................................490
References ........................................................................................491
Homework ...................................................................................... .494
10. Moving Bed and Simulated Moving Bed Sorption Separations ............... .499
10.1. Single solute recovery: Staged systems ......................................... .499

10.2. Moving bed systems ........................................................................ 510
10.3. Moving bed fractionation ............................................................... .521
10.4. Simulated moving bed fractionation .............................................. .524
10.5. Design considerations ...................................................................... 534
References ...................................................................................... .538
Homework ...................................................................................... .541
Nomenclature Chapter 11 .................................................................................. 547
11. Electrophoretic Separation Methods ..........................................................550
11.1. Theory of electrophoresis ................................................................ 550

11.1.1. Basic concepts of electrophoresis .................................... 551
11.1.2. Forces in electrophoresis ................................................. 554
XIV
11.1.3. Complicating factors in electrophoresis .......................... 561

11.2. Methods for electrophoresis ........................................................... .564
11.2.1. Moving boundary electrophoresis ................................... 564
11.2.2. Gel, membrane and paper electrophoresis ..................... .565
11.2.3. Zone spreading in zonal electrophoresis ......................... 570
11.2.4. Affinity electrophoresis and immunoelectrophoresis ..... .575
11.2.5. Free solution and capillary electrophoresis .................... .576
11.2.6. Sequential two-dimensional methods ............................. .579
11.2.7. Continuous two-dimensional electrophoresis .................. 582
11.3. Isotachophoresis (Displacement electrophoresis) .......................... 589
11.4. Isoelectric focusing ......................................................................... 594
11.4.1. History and basic ideas ................................................... .596
11.4.2. Theory of isoelectric focusing ......................................... 597
11.4.3. Methods of preparative isoelectric focusing ................... 602
11.5. Electrochromatography .................................................................. 605
References ..................................................................................... 608
Homework ..................................................................................... 615
PART III. MEMBRANES
Nomenclature Chapters 12 and 13 .................................................................... 623
12. Introduction to Membrane Separations ...................................................... 630
12.1. Basic concepts ................................................................................ 630

12.2. Membrane modules ........................................................................ 636
12.3. Gas permeation ............................................................................... 643
12.4. Reverse osmosis (RO) .................................................................... 654
12.5. Ultrafiltration (UP) ......................................................................... 673
12.6. Dialysis ...........................................................................................687
12.7. Electrodialysis (ED) ....................................................................... 693
12.8. Pervaporation .................................................................................. 703
12.9. Liquid membranes .......................................................................... 716
xv
12.10. Summary - Objectives ................................................................... 719

References .................................................................................... 719
Homework .................................................................................... 727
13. Detailed Theories for Membrane Separations ............................................733
13.1. Approximate analysis of concentration polarization ......................734

13.1.1. Basic equations ................................................................ 734
13.1.2. Mass transfer correlations ............................................... 737
13.2. Gel formation and fouling ..............................................................745
13.3. Completely mixed systems with concentration polarization .......... 749
13.3.1. Completely mixed RO and UP with concentration
polarization and no gel formation ................................... 749
13.3.2. Completely mixed UP with gel formation .......................753
13.4. Countercurrent gas permeation with no concentration
polarization ................................................................................. 758
13.5. Exact solution for concentration polarization in laminar flow ....... 762
13.6. Transport inside the membrane ...................................................... 770
13.6.1. Basic irreversible thermodynamics ................................. 770
13.6.2. Solution-diffusion membranes (RO and gas
permeation) ...................................................................... 773
13.6.3. Frictional model for porous membranes (UF) ................. 778
References ...................................................................................... 783
Homework ...................................................................................... 785
PART IV. SELECTION AND SEQUENCING
Nomenclature Chapter 14 .................................................................................. 793
14. Selection and Sequencing of Separations ...................................................795
14.1. General characteristics of the separation problem .........................795

14.2. Overview of separation methods ....................................................798
xvi
14.3. Energy criteria ................................................................................812

14.3.1. Minimum work ................................................................ 812
14.3.2. Equivalent work requirements ......................................... 815
14.4. Strategy of process design and synthesis .......................................822
14.5. Heuristics ........................................ ;...............................................824
14.5.1. G. General and ftowsheet heuristics ...............................825
14.5.2. D. Design heuristics ........................................................ 826
14.5.3. C. Component and composition heuristics .....................828
14.5.4. S. Separation specific heuristics .....................................829
14.6. Evolutionary methods ..................................................................... 835
14.7. Summary - Objectives ....................................................................838
References ......................................................................................838
Homework ......................................................................................842
Appendix. Answers to Selected Problems .......................................................847
Index ..................................................................................................................853
Chapter 1
INTRODUCTION
This textbook considers separation processes where mass transfer rates need to
be included for a complete analysis. Included in the book are crystallization,
adsorption, chromatography, ion exchange, electrophoresis and membrane
separations. We will start with equilibrium-based separation processes and
gradually switch to rate processes. This will be done by starting with crystalli-
zation and finishing with membrane separations.
1.1. PART I: CRYSTALLIZATION
Crystallization is a complex, multifaceted subject On one level of sophistica-

tion, crystallization can be considered entirely as an equilibrium stage separa-
tion process. This is done in Chapter 2 which could be included in a book on
equilibrium staged separations. The coverage in Chapter 2 is concerned with
crystallization from solution. Crystallization is done from a solution containing
a solvent such as water or alcohol. This is the most common type of industrial
crystallization. It is used for production of a large variety of inorganic salts and
organic compounds. Usually, only a single equilibrium stage is required. The
equilibrium diagrams discussed in Chapter 2 are similar to the enthalpy-
composition diagrams used for distillation and the triangular diagrams used for
extraction. Thus the principles will often be familiar although the diagrams
may be more complex. In ternary systems such as two salts plus water, a single
equilibrium stage operating at one temperature can produce a single pure salt
plus a solution containing both salts. Fractional crystallization can be used to
produce both salts as pure products. Fractional crystallization usually uses two
equilibrium stages operating at different temperatures. These cycles are
explained using triangular equilibrium diagrams.
2
When temperature has a large effect, a large amount of crystals can be

obtained by cooling the solution. When temperature has a small effect cooling
will have little effect. Crystallization then requires removal of solvent by eva-
poration or changing the solubility by adding another component (salting out).
When there is a moderate temperature effect, both cooling and evaporation are
used in a vacuum crystallizer. The equipment types used for these processes
will be explained in Chapters 2 and 4.
The product from a crystallization is often sold as a solid (e.g. table salt
or sugar). The size distribution and the shape of the crystals is very important
in customer acceptance of the product. (It is annoying to buy a bag of sugar
which is a single solid lump). The crystal size distribution (CSD) and the crys-
tal shapes are not determined by eqUilibrium. Thus, if we stop at Chapter 2 we
will miss a very important part of crystallization. To determine crystal size dis-
tributions, nucleation and growth rates are studied in Chapter 3.
Chapter 4 develops population balances (a count of the number of crys-

tals in a given size range) and from this crystal size distributions are generated.
The population balance is used first to develop the crystal size distribution for a
simple case. From this several other distributions are calculated. Use of exper-
imentally determined crystal size distributions to determine nucleation and
growth rates is explored, and the effects of equipment variations on CSD are
determined.
The second major type of crystallization is melt crystallization. In melt

crystallization no solvent is present. Instead, the solid is melted and then
slowly crystallized to purify it This method is used for purification of organic
chemicals, pharmaceuticals, and in the semi-conductor industry. Although
many of the principles of melt crystallization are similar to crystallization from
solution using cooling, the equipment is often very different Some of the
methods which will be discussed in Chapter 5 are normal freezing, zone melt-
ing, and counter-current column crystallizers.
The chapters on crystallization assume familiarity with equilibrium stage

concepts and mass transfer. No prior knowledge of crystallization is required.
The structure of the information is shown in Figure 1-1. This figure shows the
3
Chapter 2
Solubility
Phase Equilibria 8
Single Stage Calc.
r
~
Chapter 2
r
Chapter 3
Fractional Nucleation &
Crystallization Crystal Growth
, r
Chapter 4
Chapter 5
Population Balances
&CSD Melt Crystallization
Figure 1-1. Structure of knowledge for Part I. Crystallization.
path required to reach a particular topic. Chapter 2 is the basic chapter required
for all other chapters. Note that some topics, such as fractional crystallization,
although very interesting, are not prerequisites for later material.
1.2. PART II. SORPTION AND CHROMATOGRAPHY
In Chapters 6 to 10 we will switch gears and look at separation processes such

as adsorption, chromatography, and ion exchange which separate based on the
sorption of solutes onto a solid. These sorption separations all involve interac-
tion with a solid but the mechanism of the interaction can be different. The
separation is usually done in packed columns in a cyclic fashion. These separa-
tion techniques are thus usually quite different than equilibrium staged separa-
tions. The classical sorption separations use a column packed with small
porous particles. The particles remove solutes from the fluid by adsorption or
4
ion exchange. Since the solid is usually not moved, the solid will eventually
become saturated. A separate desorption or regeneration step is required to
remove and recover the desired solute. Operation is inherently time-dependent,
and is not steady-state as are the equilibrium staged processes.
In Chapter 6 we will first look at a physical picture of what happens in

adsorption and chromatography. Equilibrium expressions will then be
considered. A simple theory of solute movement in the column will be
developed for linear and then for non-linear equilibrium. This will allow us to
determine when the column will be saturated. Then the rigorous mass and
energy balances will be presented, and the solute movement results will be
rigorously derived
Chapter 7 discusses linear theories for chromatography and adsorption.
The solute movement theory developed in Chapter 6 is used to explain chroma-
tographic operations. The mass transfer equations are solved for systems with
linear isotherms. Several linear chromatographic theories are compared.
In Chapter 8 several commercially significant packed bed adsorption
processes are explored. The solute movement theory is used to delineate the
basic principles of the process. The empirical mass transfer zone (MTZ)
approach based on experimental data and theoretical constant pattern solutions
are developed. Applications such as drying and pressure swing adsorption are
discussed.
The topic of Chapter 9 is ion exchange. Electroneutrality and equili-
brium will be discussed first The properties of ion exchange resins are
explored Then the solute movement theory will be adapted to ion exchange.
Several applications such as water softening and demineralization are dis-
cussed.
In Chapter 10 the development of counter-current and simulated

counter-current processes for adsorption, ion exchange and chromatography
will be discussed. These methods make the operation steady-state and the
separations become analogous to equilibrium separations.
Figure 1-2 shows the structure of knowledge for Chapters 6 to 10.

Chapter 6 is the key chapter required for all other chapters. Chapters 7 to 10
Chapter 6
Equilibrium
Solute Movement Theory
Mass and Energy Balances
J
!
Chapter 7 Chapter 9
!
Chapter 8
Chromatography
-! Ion Exchange Basicslt---,
Linear Dispersion Model MTZ Approach
Equilibrium
Staged Model Constant Pattern
Ion Movement
Van Deemter Eq. Complicated Situations
Ion Exclusion
Rosen Model Desorption Methods
r+-. Mass Transfer
Large Scale Applications
I Q
"
For Mass Transfer
-0
! ~
i3
and for Problems 0-
m
3
en
For Ion Exclusion and for Problems
t
Chapter 10
Staged Analysis
Moving Bed
Fractionation I"""'"
5MB
Figure 1-2. Structure of knowledge for Chapters 6 to 10. Sorption separations.

V1
6
are essentially independent of each other except that some problems will use
analysis methods developed in previous chapters. These chapters assume fami-
liarity with mass transfer concepts, but no previous knowledge of adsorption,
chromatography or ion exchange. Chapters 2 to 5 are not prerequisites for
Chapters 6 to 10.
Chapter 11 is an introduction to electrophoretic separations. These

methods use migration of charged species in an electrical field to achieve
separation. These methods have some similiarity to chromatographic separa-
tion methods; thus, Chapters 6 and particularly 7 are prerequisites for Chapter
11. Since electrophoretic methods are not currently used for large-scale
separations, this chapter is a progress report
1.3. PART ill. MEMBRANES
Chapters 12 and 13 consider membrane separation methods such as gas

permeation, reverse osmosis, ultrafiltration, dialysis, electrodialysis, prevapora-
tion, and liquid membranes. The membrane separators, except dialysis and
liquid membranes, can not operate as equilibrium separators. The separation is
based on different rates of mass transfer through the membrane. These
chapters assume familiarity with mass transfer concepts, but no previous
knowledge of the membrane separators is required. Chapters 2 to 11 are not
prerequisites for these two chapters.
Chapter 12 is an introduction ~o all of the membrane separators. First the

basic concepts of membrane separators including membrane structure, flux, and
geometry are discussed. Then each type of membrane separation is considered
separately. First the method is defined and then the process is explained
theoretically, but at a fairly simple level. This explanation includes permeabil-
ity for gas permeators, osmotic pressure and concentration polarization for
reverse osmosis, gel layer theory for ultrafiltration, electrochemical reactions
and energy consumption in electrodialysis and diffusion in prevaporation.
Chapter 13 treats some of the theories for membrane separation in more

detail. Chapter 12 is a prerequisite for this chapter. The theories of concentra-
7
tion polarization, gel formation and fouling in reverse osmosis and

ultrafiltration are explained in detail. The effects of bulk flow patterns in the
membrane module are modeled. Finally, irreversible thermodynamics is used
to analyze transfer of solute inside the membrane.
1.4. PART IV: SELECTION AND SEQUENCING
Chapter 14 discusses selection and sequencing of separation methods. There

are two very distinct problem categories. First, for new chemicals the problem
may be finding any method or combination of methods which will do the
desired separation. Second, if several separation methods will work the prob-
lem is determining the best method or sequence of methods to do the separa-
tion.
Three categories of methods have been developed for selection and

sequencing. Heuristics are rules-oj-thumb used to get one started. Evolution-
ary methods are techniques to move towards an optimum solution from the
base case. Algorithmic approaches involve complete design of all alternatives.
These three approaches are often used sequentially in the order listed. Unfor-
tunately, these methods have been developed in detail for distillation, and to a
lesser extent for extractive and azeotropic distillation, absorption, and extrac-
tion. Since very little research has been done on selection and sequencing of
the separation methods covered in this book, Chapter 14 represents an unpro-
ven trial.
1.5. PROBLEM SOLVING
Engineers are problem solvers. Thus the ability to solve separation problems is
critically important. To help you develop this ability example problems are
solved in the text and homework problems are given at the end of each chapter.
An explicit problem solving strategy is often useful. The modification of

the strategy developed at McMaster University (Woods et ai., 1975) is used for
8
many of the examples. The discussion of this procedure closely follows that in
Wankat (1988). The steps in problem solving are:
o. I want to and I can
1. Define the problem
2. Explore or think about it

3. Plan
4. Do it
5. Check
6. Generalize
Step 0 is a motivation and confidence step. It is a reminder that you got

this far in your education because you can solve problems. The more different
problems you solve, the better problem solver you will become. Remind your-
self that you want to learn how to solve separation problems and you can do it.
In step 1 define the problem. Make sure that you clearly understand all
the words. Draw the system and label its parts. List all the known variables
and constraints Describe what you are asked to do. If you cannot define the
problem clearly, you will probably be unable to solve it.
In step 2 you explore and think about the problem. What are you really
being asked to do? What basic principles should be applied? Can you find a
simple limiting solution that gives you bounds to the actual solution? Is the
problem over- or underspecified? Let your mind play with the problem. Then
go back to the Define step to make sure that you are still looking at the problem
in the same way. If not, revise the problem statement and continue. Experi-
enced problem solvers always include an Explore step even if they don't expli-
citly state it
In step 3 the problem solver plans how to subdivide the problem and
decides what parts to attack first The appropriate theory and principles must
be selected, and mathematical methods chosen. The problem solver assembles
9
required resources such as data, paper, and calculator. While doing this, new
subproblems may arise; you may find there are not enough data to solve the
problem. Recycle through the problem-solving sequence to solve these sub-
problems.
Step 4, Do it, is often the first step that inexperienced problem solvers
try. In this step the mathematical manipulations are done, the numbers are
plugged in, and an answer is generated. If your plan was incomplete, you may
be unable to carry out this step. In that case, return to the Explore or Plan steps
and recycle through the process.
In step 5, check your answer. Is it the right order of magnitude? Does

the answer seem reasonable? Have you avoided blunders such as plugging in
the wrong number or incorrectly punching the calculator? Is there an alterna-
tive solution method that can serve as an independent check on the answer? If
you find errors or inconsistencies, recycle to the appropriate step and solve the
problem again.
The last step, Generalize, is important but is usually neglected. In this

step you try to learn as much as possible from the problem. What have you
learned about the physical situation? Did including a particular phenomenon
have an important effect, or could you have ignored it? Generalizing allows
you to learn and become a better problem solver.
I strongly encourage you to use this or another organized strategy and

write down each step as you do homework problems. In the long run the use of
an organized strategy will improve your problem-solving ability.
To become proficient in solving problems you must work on the home-

work questions and problems. The Homework at the end of each chapter is
organized into a series of sections as follows.
A. Discussion Problems. These are questions requesting an essay type

answer. The last question in each chapter asks for development of a key
relations chart. This is your one-page summary of everything (key facts,
equations, diagrams. etc.) which you need to trigger your memory. The
key relations chart is an excellent way for you to summarize and review
the chapter.
10
B. Generation of Alternatives. These problems ask you to brainstonn and

develop alternatives. There is no one correct solution to these problems.
C. Derivations. These problems ask you to derive equations in the text or

equations that are not given.
D. Single-Answer Problems. These are typical homework problems with a

single answer.
E. More Complex Single Answer Problems. These are more complex prob-
lems or derivations.
F. Problems Requiring Other Resources. Some of these questions ask for a

critical review of a classical paper. Others are problems which would
have been in category D except some of the required data must be
obtained from other resources.
Some chapters do not have categories B or E or F.
1.6. SUMMARY - OBJECTIVES

At the end of this chapter you should be able to
1. Briefly discuss the separation methods covered in this book.
2. List and define the steps in the problem solving strategy.

Part I
CRYSTALLIZA TION
NOMENCLATURE CHAPTERS 2 TO 5
a interfacial area/volume
a,b empirical constants in Eq. (2-1b), or Eq. (3-15)
linear width of metastable zone, Eq. (4-67)
A preexponential constant, Eq. (3-2)
A cumulative area/vol of crystals

A crystal surface area
A area of heat exchanger, m 2
A cumulative area/vol of crystals
cross sectional area
total area/volume of crystals
b drying up points
B(L) birth function, Eq. (4-59)
B rate of fonnation of nuclei
C fulcrum point
Constants of integration for size dependent growth, Eqs. (4-61) and

Table 4-3.
c concentration, kg anhydrous crystals/kg water
c* equilibrium concentration
salt concentration or amount
C flow rate crystals
C constant of integration, Eq. (4-46b)
D column diameter, m
D(L) death function, Eqs. (4-59)
13
14
D diffusivity
e entrainment, Eq. (5-15)
f solidification rate, cm/min
F feed rate, kg/hr
F flow rates, kg/hr
g fraction of change frozen
G linear growth rate, Eq. (3-5), m/s
h heat transfer coefficient
order of nucleation, Eq. (3-4)
k Boltzmann's constant = 2.3085 x 10-23 JIK
kA area shape factor, Eq. (4-20)
kf film mass transfer coefficient
~ linear distribution coefficient, Eq. (5-1)
kN rate constant for nucleation
1<r rate constant
ks solubility product, Eq. (2-3b)
kv volume shape factor
K constant, Eq (2-12)
Ko overall coefficient
I length of solid melted to form zone, m
L liquid flow rate or amount
L characteristic dimension of crystal
L cumulative length m/m3

15
LD dominant size mass distribution = 3Ot, m
Lr size of fines, m
Lp size product crystals, m
Ls size seed crystals, m
m mass solid deposited
M mixed stream flow rate or amount
M melt flow rate, kg/min
M cumulative mass of crystals/volume
MF cumulative mass of fines/volume
MT magma density
MW molecular weight
n moles water per mole hydrate
n population density
n overall growth rate order, Eq. (3-16)
n° population density of nuclei
nL population density large crystals, Eq. (4-56b)
fir "order" for growth
I1s population density small crystals, Eq. (4-56a)
N kg solvent/kg salt
N cumulative number crystals/volume
N stirrer speed, rpm
NT total number of particles/vol.

p product flow rate, kg/s
16
Q volumetric discharge rate m3Is
r radius of particle or crystal, m
R ratio retention times, Eq. (4-44a)
R gas constant
S kg salt in solution/hr
S c/c* = supersaturation ratio
Sc kg/hr crystals (hydrate fonn)
t time
leol period of collection of magma, s
T temperature °C or K
To melting point temperature
T* temperature at equilibium
Tcold coolant temperature
TM magma temperature
T. solid temperature
U overall heat transfer coefficient
V volume of magma, m 3
V kg/hr vapor evaporated
VM molar volume of crystal
VI volume magma sieved, m3
W water flow rate, kg/hr
W·1 weight of crystals in sieve fraction i, kg
Wp flow rate of product crystals, kg/hr

17
WI flow rate of seed crystals, kg/hr
x mass fraction in solid
X kg moles soluble salt./kg total salts
y mass function impurity in the melt
y mass or mole fraction salt in solution
Yso!ute mass or mole fraction solute in vapor
Y kg soluble salt./kg total salts
z distance in zone melting, cm
z increased fractional rate of removal of product
Greek
Ct. exponent in birth and death functions, Eqs. (4-59)
f3 exponent in birth and death functions, Eqs. (4-59)
f3 fraction of nuclei surviving point fines trap
f3 constant in Eq. (2-12)
y constant for size dependent growth, Eqs. (3-15) and (4-60b)
o film thickness
~ delta or difference point
~c concentration supersaturation
maximum allowed supersaturation
activation energy in Arrhenius Eq. (3-17)
fraction crystals melted.
free energy change for formation of nuclei
Lp-L.
18
M.i size range sieve fraction i,m
~T supercooling temperature
"- molar heat of fusion
PF feed density, kg/m3
Ps solid density, kg/m 3
0' interfacial tension
't VIQ, drawdown or retention time, S
'tf retention time fines, s
'tL retention time large crystals, s
'tp retention time product, s
'ts retention time smaIl crystals, s

Chapter 2
CRYSTALLIZATION AND PRECIPITATION FROM

SOLUTION-EQUILIBRIUM ANALYSIS
In the process of crystallization from solution the desired solute is dissolved in

a solvent along with some impurities. The solution is made supersaturated by
cooling and/or evaporation of solvent. Crystallization occurs usually on added
seed crystals or on very small crystals (nucleii) broken off of other crystals.
The crystals will grow and become larger until they are removed (harvested)
from the crystallizer. Often the crystals will be pure and contain almost no
impurity. This large separation factor is one of the reasons which makes cry-
stallization a desirable separation method. After harvesting, the crystals are
separated from the adhering solvent (mother liquor), and the crystals are
prepared as a final product. A second reason why crystallization is often the
separation method of choice is it can produce uniform crystals of the desired
shape. This is important when a solid product is desired (e.g. with common salt
or sugar).
Precipitation is a related process since solutes dissolved in a solvent pre-

cipitate out. However, the precipitate is usually amorphous and will have a
poorly defined shape and size. Precipitates are often aggregates of several
species and may include salts or occluded solvent. The precipitate is not pure
and has a significantly lower separation factor than crystallization. Thus pre-
cipitation serves as a "rough cut" to either remove impurities or to concentrate
and partially purify the desired product. Precipitation is an important separa-
tion method in production of many biochemicals and in mineral processing.
Since many of the basic concepts of solubility and equilibrium are the same for
crystallization and precipitation, these methods are considered together.
In this chapter we will explore solid solubility and phase equilibria for
crystallization systems. The solubility diagrams will be used to explain and
19
20
classify the different ways of doing crystallization commercially. Equilibrium

methods for crystallization calculations will be developed, and fractional cry-
stallization will be discussed. Finally, precipitation processes will be explored.
Equilibrium and equilibrium calculations are very important in crystalli-

zation design. However, equilibrium methods alone are not sufficient in crys-
tallizer design as they can be for other equilibrium staged separations. Equili-
brium calculations can tell the total mass of crystals produced, but equilibrium
calculations cannot tell the size and number of crystals produced. For example,
a mixture of one million particles each one mm in diameter has the same mass
as 100 million particles each 0.1 mm in diameter. These two mixtures will be
two very different products. To include the crystal size distribution in the
design calculations, crystallization and nucleation rates and population balances
must be included. These topics are the subject of Chapters 3 and 4.
2.1. SOLUBILITY
Solids have widely different solubilities in solvents. Since water is by far the
most common solvent in crystallization, we will focus on the solubility of com-
pounds in water. The effect of temperature on the solubility of different com-
pounds is shown in Figures 2-1a and b and in Table 2-1 (Mullin, 1972). It is
common to list concentration in mass ratio units such as kg of the anhydrous
salt per 100 kg of water as shown in Figures 2-1a and b. Figure 2-1a shows the
solubility of systems where only one crystalline phase is deposited over the
range indicated. The materials with water attached such as CUS04 ·SH2 0 are
hydrates. The crystal is formed from a compound which consists of one
molecule of CUS04 and 5 molecules of H 2 0. This is the stable form of the
solid from 0 to 100°C. Crystals with no water are called anhydrous (e.g.
NaCI). Note that this type of diagram is slightly misleading since concentra-
tions are plotted in terms of the anhydrous salt, but the stable form may be a
hydrate. Obviously it is easy to calculate the kg of CUS04 ·SH2 0 which are
soluble from the kg of anhydrous salt which are soluble. Crystals can also
change phase from one hydrate form to another or to the anhydrate. The phase
transitions are illustrated in Figure 2-1b. A similiar figure for biochemicals is
shown by Belter et al (1988).
21
100 100
(0) 0", 0 (\J (b)
~ ~
80 80 <f Q,
0 0 +-
N
(/) C
I
u0 N N
Qi
.60 60 (1) ~ u
c
...::::J
CI) -0
Qi
0 u
c
~ 40 40
a.
E
~
20 20
0 20 60 80 100 0 20 40 100
Concentration, Ib 1100 Ib water Concentration, Ib 1100 Ib water
Figure 2-1. Solubility diagrams. a Systems without phase change. b. Sys-
tems with crystal phase changes. (Mullin, 1972). Reprinted
with pennission of Butterworths. Copyright 1972.
For ideal mixtures the solubility can be predicted with reasonable accu-
racy using the Vant'Hoff equation.
A, I 1 (2-1a)
lny= - ( - - - )
R To T
where y is the mole fraction of non-solvated solute in solution, A, is the molar

heat of fusion of the solute at it's melting point To, and R is the gas constant.
Although Eq. (2-1a) does not work for non-ideal systems, an empirical
modification will often fit the data
(2-1b)
Remember to use the absolute temperature in these equations. For many sys-
tems such as CuS04oSH20 a plot of Iff versus In y will give a straight line. If
a phase transition occurs, there will be two straight lines intersecting at the
transition point. Some highly soluble salts such as sodium acetate do not
satisfy Eq. (2-1b) and do not give straight lines on the Iff vs In y plot. Tem-
perature increases can cause either marked increases in solubility as shown for
sodium acetate or even a decrease in solubility (calcium acetate). This
22
decrease in solubility is called an inverted solubility. The temperature

coefficient of solubility controls the type of crystallizer which can be used.
This is discussed in the next section.
A considerable amount of solubility data has been collected in the litera-

ture. Standard references are Seidell (edited by Linke) (1958, 1965) and
Stephen and Stephen (1963). Mullin (1972) also has fairly extensive solubility
tables. A limited amount of solubility data is presented in Table 2-1. Thermo-
dynamic methods for estimating solubility are discussed by Walas (1985).
Figures 2-1a and b show the saturation curves for the salts. For example,
at 30°C twenty-five kg of anhydrous copper sulfate can be dissolved in 100 kg
of water (see Table 2-1). This solution is stable and can be held indefinitely
without crystallization. If a crystal of CuS04oSH20 (the stable form of the
solid at 30°C) is added, the crystal and the solution will be in equilibrium.
Regions to the left of the saturation curves are unsaturated while regions to the
right are supersaturated.
If the copper sulfate solution (with no crystals present) is either cooled or

water is removed, the solution will become supersaturated. These processes
are illustrated in Figure 2-2. Even though the solution is supersaturated no
crystals may form. If a crystal were present it would grow, but crystals do not
form spontaneously in this metastable region. If the cooling or water removal
are continued until the system passes from the metastable into the labile
Equilibrium: T vs c'
Stable
(Unsaturated)
Temperature
Labile
(Unstable)
Concentration
Figure 2-2. Miers diagram showing metastable region.
23
(unstable) region then spontaneous formation of crystals does occur. Once

crystals are present, the concentration of salt in solution will decrease as the
crystals grow and the concentration will reapproach the equilibrium curve. The
transition from the metastable region to the labile region is shuwn as a band
since it depends on experimental conditions such as the degree of mixing. This
is a kinetic transition, not an equilibrium transition. The plot shown in Figure
2-2 is often called a Miers diagram. Tavare (1987) summarizes studies on the
metastable region.
Similar metastable regions can occur in vapor-liquid and liquid-liquid

systems; however, there is a considerable difference in the size of the meta-
stable regions. In crystallization and precipitation the metastable region is
much larger. For instance, Miers found that a 53% solution of sodium nitrate in
water could be supercooled by 14°C before spontaneous crystallization
occurred.
Supersaturation is the driving force for crystallization and precipitation.

Supersaturation can be defined as the concentration driving force
(2-2a)
.1.c = c - c*
or the supersaturation ratio
S=~ (2-2b)
c*
Supersaturation can also be measured in terms of the supercooling.

(2-2c)
The solvent used, the addition of co-solvent and impurities can have a
major effect on the solubility. The impurity may:
1. have no effect (this is unusual),
2. react with the solute,
3. cause the solute to precipitate (called salting out), or
4. cause the solution to become unsaturated (salting in).

N
~
Table 2-1. Solubilities of Selected Compounds (Mullin, 1972). grams anhydrous compuound
per 100 g water.
Solubility °C Stable
Hydrate
Compound Formula 0 10 20 30 40 60 80 100 0---25°C
Ammonium (N14h S04 71 73 75.4 78 81 88 95.3 103.3

bicarbonate
barium hydroxide Ba(OH)z 1.6 2.5 3.9 5.6 8.2 21 101 8
boric acid H3B03 2.7 3.6 5.0 6.6 8.7 14.8 23.8 40.3
calcium acetate Ca(C2H302)z 37.4 36.0 34.7 33.8 33.2 32.7 29.7 2
copper chloride CuCl2 69 71 74 76 81 98 2
copper sulphate CUS04 14.3 17.4 20.7 25.0 28.5 40.0 55.0 75.4 5
ferric chloride Fe03 74.4 81.9 91.8 526 540 6
ferrous chloride Fe02 61 64 68 73 77 89 100 106 6,4
lead chloride Pb02 0.67 0.81 1.0 1.2 1.5 2.0 2.6 3.3
lead nitrate Pb(N°3h 39 48 57 66 75 95 115 139
lithium carbonate LiC0 3 1.54 1.43 1.33 1.25 1.17 1.01 0.85 0.72
magnesium chloride Mg02 52.8 53.5 54.5 56.0 57.5 61.0 66.0 73.0 6
potassium chloride KO 27.6 31.0 34.0 37.0 40.0 45.5 51.1 56.7
potassium chromate K2Cr04 58.2 60.0 61.7 63.4 65.2 68.6 72.1 75.6
potassium nitrate KN03 13.3 20.9 31.6 45.8 63.9 110 169 247
potassium sulphate K 2S04 7.4 9.2 10.9 13.0 14.8 18.2 21.4 24.2
silver acetate AgC2 H30 2 0.72 0.88 1.04 1.21 1.41 1.89 2.52
silver nitrate AgN03 122 170 222 300 376 525 669 952
sodium acetate NaC2H3 0 2 36.3 40.8 46.5 54.5 65.5 139 153 170 3
sodium bicarbonate NaHC~ 6.9 8.2 9.6 11.1 12.7 16.4 decomp
sodium chloride NaCl 35.7 35.8 36.0 36.3 36.6 37.3 38.4 39.8
sodium hydroxide NaOH 42.0 51.5 109 119 129 174 340 4,3'12
sodium nitrate NaN03 73 80 88 96 104 124 148 180
sodium sulphate Na2S04 4.8 9.0 19.4 40.8 48.8 45.3 43.7 42.5 10
Acetamide CH3 CONH2 138 175 230 310 440 850
Alanine (D) CH3CHNH2COOH 12.7 14.2 15.8 17.6 19.6 24.3 30.0 37.3
Alanine (DL) CH3 CHNH2COOH 12.1 13.8 15.7 17.9 20.3 26.3 33.9 44.0
Benzoic Acid C 6 HSCOOH 0.17 0.20 0.29 0.40 0.56 1.16 2.72 5.88
Citric Acid C3~OH(COOHh 96 118 146 183 215 277 372 526
Fructose C 6 H 12 0 6 75 80 85 90
Glucose C 6 H 12 0 6 46 70 92 120 160 280 440
Lactose C 12 H22 0 n 12.2 15.0 19.5 25.2 33.3 57.5 102 153
Sucrose C 12 H 22 0 22 179 190 204 219 238 287 362 487
IV
VI
26
In many cases it is difficult to predict if an impurity will cause salting out or

salting in. However, if the added impurity has a common ion with the original
solute, salting out will almost always occur. For example, if KCI is added to a
saturated solution of NaCI, then NaCI will precipitate out.
The behavior of common ions can be predicted from the solubility pro-
duct (Belter et ai, 1988; Mullin, 1972). Consider the dissociation of a salt
(2-3a)
For a sparingly soluble salt where dissociation is complete and the activity
coefficient is unity, the solubility product is
(2-3b)
where Cp+ and c~ are ionic concentrations in any consistent set of units. The
solubility product ks usually follows an Arrhenius temperature dependance. In
the example with KCI and NaCI, cr is the common ion. Adding KCI increases
Ccr. Since k NaCl is constant, increasing Ccr requires that CNa+ decrease which
can occur only by precipitation of NaCl. Salting out can also be explained
using ternary diagrams (see Example 2-3).
The solubility diagrams, Figures 2-1, and Table 2-1, are for systems with
a single solute. When a second component is added or present in considerable
amount, these simple diagrams are no longer adequate. More detailed phase
diagrams are discussed in Section 2.4.
Figure 2-3. Cooling crystallizers. a. Indirect cooling with external circula-

tion. b. Indirect cooling with internal circulation. c. Double-
pipe scraped surface. (Garrett, 1961), Reprinted with permis-
sion from Ind. Eng. Chern., 53, 623 (1961), Copyright 1961,
American Chemical Society. d. Direct cooling (Bennett,
1988). Reprinted with permission from Chern. Eng., 95(8) 118
(1988). Copyright 1988, McGraw-Hill Book Co.
27
a b
Feed
Suspension
Feed Magma
c CRYSTALLIZATION TUBE
COOLANT
Refrigerant vapor
d
Body __
/ Boiling surface
/
Mother liquor
Baffle,
-, Settling zone
Refrigerant
Feed
Product
slurry
28
2.2. EQUIPMENT TYPES
The solubility behavior of the system controls the type of equipment which can
be used for commercial crystallization. This choice can be explored using the
solubility diagrams discussed in the previous section. Most crystallization
operations can be classified into one of five categories based on how the super-
saturation is obtained. More details on the equipment are in Chapter 4 and in
several reviews (Bennett, 1984, 1988; Mullin, 1972; Nyvlt, 1982; Moyers and
Rousseau, 1987; Singh, 1979).
Cooling crystallizers obtain supersaturation by cooling the hot saturated

solution. This method is applicable to systems where temperature has a large
effect on solubility (for example, ammonium alum and sal soda,
NaZC03 • lOHzO). The method is not useful for systems like NaCI where tem-
perature has little effect on solubility. A typical cooling crystallizer with an
external heat exchanger is shown in Figure 2-3a while a jacketed tank, cooling
crystallizer is shown in Figure 2-3b. A variety of trough type cooling crystal-
lizers are also available with a variety of mixers. Scraped surface, double pipe
heat exchangers (Votators) are used particularly for producing seed crystals
(Figure 2-3c) Garrett (1961) and for viscous solutions such as crystallizing
petroleum wax. Cooling can also be done by direct contact with an immiscible
refrigerant. One method of doing this is shown in Figure 2-3d (Bennett, 1988).
Cooling crystallizers are very desirable when they work since they have a high
yield and low energy consumption.
Evaporative crystallizers supersaturate the solution by removing solvent

through evaporation. These crystallizers are used when temperature has little
effect on solubility (e.g. NaCl) or with inverted solubilities (e.g. calcium ace-
tate). The crystallizer may be as simple as a shallow open pan heated by an
open fire. This ancient technology is still in use for making maple syrup and
fishery salt. For manufacture of common salt an evaporative crystallizer con-
taining calandria (steam chest) is often used (see Figure 2-4a (Mullin, 1972».
The magma (mixture of crystals and solution) circulates by dropping through
the central downcomer and then rises as it is heated in the calandria. At the top
some of the solution evaporates increasing the supersaturation causing crystal
growth. Forced circulation evaporative crystallizers are also commonly used.
29
One type is shown in Figure 2-4b (Bennett, 1988). These systems are obvi-
ously very similar to evaporators without crystallization (e.g., McCabe et ai.,
1985; Mehra, 1986; Perry and Green, 1984). Spray evaporative crystallizers
shown in Figure 2-4c (Bennett, 1988) are often useful when contact with air
will not cause contamination problems. This cycle is very similiar to the forced
circulation system.
The evaporators are often connected together to make a multi-effect eva-

porator as shown in Figure 2-4d. Here the vapor from the first unit is used as
the steam to heat the second unit. This reduces the steam requirements per kg
of product. The pressure must be varied as shown in the figure so that V will
be at a temperature greater than the boiling temperature in the second stage. A
variety of ways of connecting the different stages have been developed.
Any method for selectively removing the solvent could be used to super-
saturate the solution. For example, Azoury et al (1987) used reverse osmosis to
supersaturate aqueous calcium oxalate.
Vaccum crystallizers utilize evaporation to both concentrate the solution

and to cool the mixture. Thus both of the methods used to cause supersatura-
tion in evaporative and cooling crystallizers are used. The operating principles
can be understood by referring to Figure 2-5a (Bennett, 1988). The hot liquid
feed is at a higher pressure and temperature than the crystallizer. When this
feed enters the crystallizer, it flashes and part of the feed evaporates. This
flashing causes adiabatic cooling of the liquid (this is very similar to flash distil-
lation). Crystals and mother liquor, containing any non-volatile impurities, exit
the bottom of the crystallizer. Other variants of vacuum crystallizers have been
developed. One example is the draft-tube baffled type shown in Figure 2-5b
(Bennett and Van Buren, 1969). Because of the vacuum equipment required to
operate at pressures from 5 to 15 mm Hg, the vacuum crystallizers are consid-
erably more complex than the other types of crystallizers. Vacuum crystalliz-
ers are most common in large systems (more than 100,000 gal/day).
Salting-out processes add either a liquid diluent (anti-solvent) or a salt

with a common ion pair to cause precipitation. For example, the effect of
adding methanol on the solubility of common salts is shown in Figure 2-6
(Lozano, 1976). Addition of methanol to a saturated aqueous solution will
30
a -Vapour
@
@
Feed-
-Steam
Condensate
Barometric condenser_
b Noncon-
Cooling-water - - - _ densable-gas
Water
Steam inlet
_ Swirl breaker
Heat
exchanger __
Condensate
Circulation Feed
Product
pump slurry
31
Humidified air
c
___ -Blower
Spray Feed
chamber --
Hot water
Air Air
Heating element
Cold
water
Product slurry
d v v
s F F
Figure 2-4. Evaporative Crystallizers. a. Natural circulation with calandria

(Mullin, 1972). With permission of Butterworths. Copyright
1972. b. Forced circulation, Swenson type. (Bennett, 1988). c.
Spray evaporative crystallizer (Bennett, 1988), Pans c and b
reprinted with permission from Chern. Eng., 95(8), 118 (1988),
Copyright 1988, McGraw-Hill Book Co. d. Multi-effect sys-
tem.
W
Barometric condenser , __......... Water IV
To vacuum
equipment
Barometric
condenser
"
"
" 2-stage II r To hotwell

,1.1 A ____ /,_,
steam-jet
ejector I
r-""""r--'
Slurry
discharge
Steam
Feed
'- Propeller agitator
Purge
Condensate
~}---Dump valve
(a) (b)
Figure 2-5. Vacuum crystallizers a. Batch (Bennett, 1988). Reprinted with permission from Chern. Eng.,
95(8), 118 (1988). Copyright 1988, McGraw-Hill Book Co. b. Swenson draft-tube baffled
(Bennett and Van Buren, 1969). Reprinted with permission from Chern. Eng. Prog. Syrnp. Ser.
65 (95) 44 (1969). Copyright 1969, American Institute of Chemical Engineers.
33
20
A
" I
00
~ --.JS..f.L
'\;... A - MgGI2
8 - NoGI
i\ \ G -
o -
KGI
'\
80 N02S04
~ E - MgS04
'"""
;0 F - K2 S0 4
60 ~ B
5 ~
O~ ~i'z
~
40\,~
~
30
20
\
K~~~
'\
"'b
'U~ ~ "-
~ ~'-...J.
°0 fO 20 XJ 40 50 60 70 80 90 I00
METH4NOL IN SOLUTION, wr %
Figure 2-6. Effect of methanol on solubility of common salts at 30°C
(Lozano, 1976). Reprinted with pennission from Ind. Eng.
Chem. Proc. Des. Develop., 15, 445 (1976). Copyright 1976,
American Chemical Society.
often cause precipitation of the salt. Addition of alcohol has been used to salt
out Alz(S04h and to decrease the viscosity of the slurry. The crystallizer can
be a simple stirred tank, but additional equipment is required to recover the
methanol. A common use of salting out is the addition of CNH4h S04 to pro-
tein solutions to selectively precipitate different proteins. Sodium chloride has
been used commercially to salt out NH4 Cl. More details of salting-out methods
used in precipitation operations are in Section 2.6.
34
In reaction crystallization a solid product results from the reaction of

gases and/or liquids. This process is commonly used in coke oven plants, the
pharmaceutical industry, and in the production of some fertilizers. For exam-
ple, ammonium sulfate is produced by reacting ammonia and sulfuric acid in a
crystallizer. Reaction systems may produce a precipitate instead of well-
defined crystals.
Both batch and continuous crystallizers are used. Generally, a produc-

tion rate greater than about 50 ton/day is required to justify continuous opera-
tion (Bennett, 1988). The continuous operations tend to have higher yield and
require less energy than batch, but batch is more versatile.
2.3. YIELD CALCULATIONS BASED ON SOLUBILITY DIAGRAM
When relatively pure solutions are crystallized or precipitated, the yield

calculations can be done using the solubility diagram. A single stage process is
almost always sufficient and can be done either batch or continuously. This is
easiest to illustrate with two short examples.
Example 2-1A. Yield Calculation for Cooling Crystallizer.
A cooling crystallizer is used to crystallize sodium acetate, NaCzH 3 0 z ,

from a saturated aqueous solution. The feed is initially saturated at
40°C while the crystallizer is cooled to O°C. If we initially dissolved
the anhydrous salt in 100 kg of water/hr, how many kg of crystals/hr are
collected?
Solution
In a cooling crystallizer there is essentially no loss of solvent. From

Table 2-1 the feed solubility is 65.5 kg/1oo kg water while the product
solubility at O°C is 36.3 kg/1oo kg water. However, the amount of
35
water available is reduced since the stable form of the crystals at O°C is
a hydrate with 3 moles of water (see Figure 2-1b or Table 2-1). A mass
balance on the anhydrous sodium acetate is,
MWanhydrous (2-4a)
Se (kg crystals/hr)( MW ) + c out W out = cin Win
hydrate
where Sc is the kg of crystals/hr (hydrate form); Cout and ein are concen-
trations of anhydrous Na z C z H 3 0 Z per kg water, and W out and Win are
the kg of liquid water/hour. A water balance is
n MWwater (2-4b)
W out + Se( MW ) = Win
hydrate
where n is the moles of water per mole of hydrate (3 in this example).

Since cin, Cout , Win and the molecular weights are known, this
represents two equations in the two unknowns Sc and W out • Solving for
Sc (see Problem 2-Cl),
( MWhydrate )(
Cin - Cout )Win
MWanhydrous (2-5a)
Sc = -----'---------
1 - cout ( MWhydrate - 1)
MWanhydrous
For this example MWanhydrous = 82.04, MWhydrate = 136.09, Cin = 0.655

kg anhydrous/kg water, cout = 0.363 kg anhydrous/kg water and Win =
100 k/hr. Thus,
s = (1.66)(0.655 - 0.363)100 = 63708 k tal /h

c 1 _ 0.363(1.66 _ 1) . g crys s our
These crystals are NaC z H 3 0 Z ·3H z O. The mass of anhydrous material

recovered is,
MWanhydrous
Recovery, Anhydrous = Sc = 38.405 kg/hour
MWhydrate
36
If the crystals were recovered directly as anhydrous material (e.g.

with boric acid or lead chloride), Eq. (2-Sa) can still be used. Now the
ratio of molecular weights is unity and the equation simplifies consider-
ably to,
(2-Sb)
Equations should never be used blindly. There may not be

enough water available to fonn enough hydrate to crystallize all the
material. Equation (2-Sa) will predict incorrect results in this case (see
Problem 2-DI2).
Example 2-1B. Evaporative or Vacuum Crystallizers.
In evaporative and vacuum crystallizers some of the solvent is removed.

This solvent removal must be included in the yield calculations. 1000
kg/hour of water is mixed with 280 kg/hour of copper sulphate at 40°C.
The solution is cooled to IDoC and 38 kg/hour of water are evaporated
in the process. How many kg/hour of crystals were collected?
Solution
From Figure 2-1A or Table 2-1 we notice that the stable crystal fonn at
IDoC is CuS04oSHzO. A copper sulphate mass balance is,
MWanhydrous
(Sc,kg crystals/hr)( MW ) + CoutWout + VYsolute = Cin Win
hydrate
Usually Ysolute = O. The water balance is
n MWwater
Wout+V+Sc(MW ) = Win
hydrate
37
where V is the kglhour of vapor evaporated, n = 5, and the vapor is

assumed to be pure water. Solving for Se,
MWhydrate
Win MW [Cin - cout (1- V/Win)]
S = anhydrous (2-6a)
c MW
1- ( hydrate - 1)
cout MW anhydrous
Note that Eq. (2-6a) reduces to Eq. (2-5a) when V = O. For a salt such
as NaCI which crystallizes in the anhydrous fonn Eq. (2-6a) reduces to
(2-6b)
For this example, MWanhydrous = 159.61, MWhydrate = 249.69, Cin

= 0.280 kg anhydrous/kg water, Cout = 0.174 kg anhydrous/kg water, V
= 38 kg/hour, and Win = 1000 kg/hr. Thus from Eq. (2-6a) we obtain,
1000(1.564)[0.280 - 0.174(1 - 1~)]

Se = 1- 0.174(1.564 _ 1) = 195.3 kglhr crystals
These crystals are CuS0405H20. Thus the mass of anhydrous crystals

is,
MWanhydrous
Anhydrous mass = Sc( ) = 124.88 kglhr
MWhydrate
If this operation had been done in a cooling crystallizer (with V = 0)

then Sc = 183.84 kg/hr.
The value of V (which was given in this problem statement) for a

vacuum crystallizer can be detennined from an energy balance. This is
considered later. The vapor flow rate controls the diameter of the crys-
tallizer. This calculation is essentially the same as for a vertical flash
drum (e.g. see Wankat, 1988, Chapter 3).
38
Note in this example that the initial solution was not saturated.
This did not change the calculation procedure. The final solutions are
assumed to be in equilibrium, and thus are saturated.
The % recovery in this example is 124.88/280 x 100 = 44.6%

which is low. The recovery can be increased by starting with a
saturated solution at a higher temperature, operating the crystallizer at a
lower temperature, or adding a second crystallizer such as a scraped
surface crystallizer at very low temperatures.
Remember that these calculations assume that impurity concentrations

are very low and do not affect the solubilities. If there are other components
present then the phase diagrams considered in the next section are needed. The
equilibrium calculations also do not give any information about the crystal
sizes produced. For this the methods of Chapters 3 and 4 are required.
2.4. PHASE EQUILIBRIA AND SINGLE STAGE SYSTEMS
Phase equilibria for two component systems can be illustrated on temperature-

composition, enthalpy-composition, and x-y diagrams. These equilibrium
diagrams for solid-liquid systems are similar to those for vapor-liquid systems,
but are often more complex. The principles remain the same. After presenta-
tion of two-component eqUilibrium data, three-component systems will be dis-
cussed. Three component systems can be shown on triangular diagrams.
These presentations are similar to those for liquid-liquid extraction, but again
the liquid-solid systems tend to be more complex. Data for many systems is
given by Purdon and Slater (1946) and Seidell (1958, 1965).
2.4.1. Binary Systems
For binary systems Gibbs phase rule gives
F=C-P+2=2-2+2=2
39
200
T,OC 160
120
80L-----~------L-----~----~
o 50 100
mole % anthracene
Figure 2-7. Solid solution system: Phenanthrene and anthracene.

Modified from figure in Walas (1985). Reprinted with permis-
sion of Butterworths, Copyright 1985.
degrees of freedom. In crystallization pressure is usually constant and there is

one remaining degree of freedom. Graphically, temperature-composition and
enthalpy-composition diagrams are convenient ways to represent the equili-
brium data.
From the point of view of equilibrium diagrams the simplest binary sys-
tems are solid solutions. These systems are relatively common in melt crystall-
ization. An example is the system phenanthrene-anthracene shown in Figure
2-7 (Walas, 1985). Cooling a liquid slowly to the liquidus line (point A) pro-
duces a solid (point B) in equilibrium with the solution. If the feed is cooled to
a point in the two-phase region (point F) at equilibrium there will be a mixture
of crystals (at pt. G) and solution (at pt. E). Further cooling to the solidus line
(pt. H) will cause complete crystallization of a solid solution. The solid lines in
Figure 2-7 are predictions made by UNIFAC (see Walas, 1985). For many
solid-liquid systems excellent predictions can be made of the equilibrium data.
This type of system requires a multi-stage contactor for purification and is dis-
cussed in more detail in Chapter 5.
A second type of phase diagram is the simple eutectic system illustrated

in Figure 2-8 for the system naphthalene-phenol (Walas, 1985). This system
could be considered as solution crystallization with phenol as the solvent or as
40
80
70
u
0 60
...
~
CI)
~ 50
...
0
CI)
o observed
~40
CI)
- calculated
f-
30 o
20
0 0.2 0.4 0.6 0.8 1.0
mole fraction napthalene
Figure 2-8. Simple eutectic binary system: Phenol and naphthalene
(Walas, 1985). Reprinted with Pennission of Butterworths.
Copyright 1985.
melt crystallization of a mixture of phenol and naphthalene. The eutectic is an

equilibrium mixture which freezes at a constant composition. This is the point
with the lowest freezing temperature (30°C). In the region from zero to the
eutectic the crystals are pure phenol while in the region from the eutectic to
100% the crystals are pure naphthalene. For example, a mixture which is 60
mole % naphthalene would first fonn crystals of pure naphthalene at about
62°C. If this mixture was cooled to 40°C, then crystals of pure naphthalene
would be in equilibrium with a solution containing about 21 mole %
naphthalene. The relative amounts of the solution and the crystals can be found
from mass balances or the lever-arm rule.
If the cooling is continued to the solidus line, the entire mass freezes.
The solution remaining will freeze at the eutectic concentration which is
approximately 17 mole % naphthalene. Thus the solid will contain crystals of
pure naphthalene and crystals of the eutectic composition. This final freezing
occurs at the eutectic temperature (-30°C). The eutectic is a physical mixture
not a chemical compound. It can be considered as analogous to an azeotrope in
distillation since it represents an equilibrium mixture of constant composition.
41
40
Solution
20
uo
Q)
.... y
....
:;)
C....
Q) 0 0
a. N
I
N
I
~ -20 V
+
C\J
+
~ Ice + solution
°
ON
I N
(!) I
H V
o 02 04 06 08 10
Mass fraction Mn (N0 3 )2
Figure 2-9. Phase diagram for system Mn (N03)2-water (Mullin, 1972).
With permission of Butterworths. Copyright 1972.
To obtain pure naphthalene crystals the feed concentration must be greater than
the eutectic composition; otherwise, pure phenol crystals will be obtained.
Note that a single stage system is capable of producing one pure product.
Methods to produce two pure products are discussed in Section 2.5.1.
Aqueous systems tend to be considerably more complex than solid solu-

tions or simple eutectics. Figure 2-9 illustrates the system Mn(N03h - water
(Mullin, 1972). Systems of a salt in water are complicated by the presence of
hydrates. From A to B the liquidus line separates solution from a mixture of
pure ice and solution. At point B the eutectic (or cryohydric point in aqueous
solutions) is reached. Then from B to C solution will crystallize to form pure
hexahydrate, Mn(N03h·6H20. Point C represents the 2S.8°C and 0.624 mass
fraction. Point C is congruent since solidification occurs with no change in
composition. Points D and F are additional eutectic points. Point E represents
the congruent melting point for Mn(N03h'4H20 (37.1°C, 0.713 wt. frac).
Continuation of the equilibrium from F to G causes dihydrate to crystallize out.
Point G is a transition point where the dihydrate decomposes to form monohy-
drate and water. Thus this is an incongruent melting point since the composi-
tion changes. This diagram represents a rather simple example for salt solu-
42
600r-,--.-,.-,-------------~----~~--__,
c:
o
:;:
::>
o
II>
N02CO,
02 04 06 08 10
Concentrotion-weight froction of No 2 C0 3
Figure 2-10. Enthalpy-concentration diagram for NA2 C03-water (Mullin,

1972). p = 1 atm. Reprinted with permission of Butterworths.
Copyright 1972.
tions. Some compounds will form many hydrates, eutectics, congruent and
incongruent melting points. A commonly cited example is MgS04 in water
(Bennett, 1984; McCabe et al. 1985; Osburn, 1963).
Enthalpy-composition diagrams are complex, but contain all the informa-

tion needed for energy balance and equilibrium calculations. When available,
enthalpy-composition diagrams are extremely useful for calculations. Unfor-
tunately, the diagrams are only available for a few systems (e.g. see Mullin,
1972), are difficult to prepare, and are not computer-friendly. One example of
enthalpy-composition diagrams will be given to explain their use.
43
Figure 2-10 (Mullin, 1972) shows the enthalpy-composition diagram for

Na2C03-water. The zero point for enthalpy is arbitrarily set as pure, liquid
water at O°C. The regions are marked on the diagram. The plateaus with three
phases in equilibrium have
F=C-P+2=2-3+2=1
degree of freedom which is used to set the pressure. (One atmosphere in Fig-
ure 2-10.) There are no additional degrees of freedom. Thus temperature is
constant, and the compositions of the two solid phases and the solution (eutec-
tic) in equilibrium are constant in these three phase regions. When two phases
are in equilibrium, there remains one degree of freedom after setting the pres-
sure. Thus, isotherms are shown in these regions.
Mass and energy balances can be done either using the data from the
diagram, or directly on the diagram itself. The procedure is similar to that used
for distillation on Ponchon-Savarait diagrams. This procedure is illustrated in
Example 2-2.
Example 2-2. Crystallization Calculation with Enthalpy-Composition

Diagram
100 kg of an aqueous solution which is 0.2 wt. fraction Na2C03 at 60°C

is cooled to 100 e. What products result? How much of each product
results? How much energy must be removed?
Solution.
The pertinent portion of Figure 2-10 is blown-up in Figure 2-11. The

feed (point F) is cooled to 10°C (point M). This is assumed to happen
without loss of water. Point M is in a two-phase region and will have
crystals and solution in equilibrium. The crystals are Na2C03olOH20
(point C) which is 37.0% Na3 C03 while the liquid is at point L which is
44
h=160- F
C
x=0.12 0.2 Xc =0.37
No 2 C0 3 ·IOH 20
Figure 2-11. Expansion of Figure 2-10 for Example 2-2 (not to scale).
approximately 12% NazC03 • The equilibrium is represented by the 10°

isotherm. At equilibrium the mass balances are
100=M=L+C
20 = MxM = LXL + CX(:
This can also be written as the lever ann rule,
Solving for L and C we have C =32 kg and L =68 kg.
Energy removed is
kJ
(100 kg)(hF - hM)- = 100(160-(-90» = 25,000 kJ
kg
Note that the solution and the crystals have different enthalpies even
though they are at the same temperature. Continuous calculations will
be very similar to these batch calculations, but will be on a per unit time
basis.
45
If some water is removed, point M will be more concentrated than the

feed. If the amount of water removed is known, than point M can be deter-
mined and the calculation can be completed.
In a vacuum crystallizer pure water vapor is removed. The concentration

of point M and the amount of cooling can be determined from mass and energy
balances. The diagram shown in Figure 2-10 is at one atmosphere which does
not correspond to the pressure of vacuum crystallizers. Thus, the vapor-liquid
portion of Figure 2-10 will not be applicable. However, the liquid and solid
parts will be affected only very slightly by changes in pressure, and Figure 2-10
can be used to closely approximate the liquid and solid equilibrium at reduced
pressure. The vapor enthalpy can be determined from the steam tables. Thus
the enthalpy-composition diagram can be used to estimate results for vacuum
crystallizers. This calculation is explored in Problem 2-D4.
2.4.2. Ternary Systems
For ternary systems with two phases in equilibrium and constant pressure there
are two remaining degrees of freedom. To represent the effect of temperature
and composition a three-dimensional diagram would be required. (see Figure
2-13 as an example.) Since this is inconvenient, the equilibrium is often illus-
trated at constant temperature or with temperature as a parameter. To represent
all three components we can use a triangular diagram similar to those used for
extraction. Two representations for the system KN0 3 - NaN03 - H2 0 at
50 0 e are shown in Figures 2-12a and b (Mullin, 1972). Figure 2-12c shows the
behavior at 25° and 100 °e.
On all three diagrams, points a and c represent the solubilities of pure

KN0 3 and NaN0 3 in water, respectively. As NaN0 3 is added to a saturated
solution of KN0 3 , the solubility of KN03 decreases along line abo This is the
salting-out effect discussed earlier. In the region a-b-KN0 3 pure KN0 3 will be
in equilibrium with solution. Thus equilibrium tie lines are straight lines
through the KN0 3 vertex to the saturated liquid line abo If KN03 is added to a
saturated solution of NaN03 , the solubility of NaN0 3 decreases along line cb
and pure NaN0 3 will crystallize out. Region c-b-NaN03 represents mixtures
46
KN0 3
Solution NaN0 3 t Solution
0.8
Mass 0.6
frae
KN03 0.4
0.2
H2 0 0.2 0.4 0.6 0.8 NaN0 3

Mass frae KN0 3
0.8
Feed----- a
0.6
0.4
a
0.2
° 0.2 0.4 0.6 0.8 NaN0 3

Mass frae KN0 3
Figure 2-12. Equilibrium data for System KN0 3 - NaN0 3 - H2 0. a Equi-

lateral triangular diagram at 50°C. b. Right triangular diagram
at 50°C. c. Data at 25 and 100°C (M and L are for Example
2-3). Data replotted from Mullin (1972).
47
of pure NaN0 3 and solution. Tie lines are straight lines through the NaN0 3
vertex to line cb. KN0 3 and NaN03 crystallize as anhydrous salts. If one of
the salts crystallized as a hydrate the tie lines would go from the composition of
the hydrate to the saturated liquid line. This is illustrated later.
The effect of removing water is easily illustrated from any of these
diagrams. For example, if we start with feed F shown in Figure 2-12b and
remove pure water, we will follow the dilution line H2 0 - F until the line inter-
sects curve ab where pure KN0 3 starts to crystallize out. Further removal of
water (say to pt. G) will produce a mixture which will separate into pure KN0 3
solid in equilibrium with solution of composition H. The dotted line to the
KN0 3 vertex serves as a tie line in the a-b-KN0 3 region. When the line H2 0-
F-G intersects the straight line KN0 3 -b, the solid KN03 will be in equilibrium
with solution of composition b. Removal of further water does not change the
solution composition and b is known as the drying up point. The behavior of
the system in the region b-c-NaN03 will be similar when water is removed,
except pure NaN0 3 will be deposited.
In the region KNOr b-NaN03 both salts are in equilibrium with solu-
tion of composition b. Since temperature and pressure are fixed and 3 phases
are in equilibrium, there are no remaining degrees of freedom. The relative
amounts of each phase can change, but compositions and temperature are set in
this region. By adding solvent (water), mixtures in this region can be moved
into either the region a-b-KN0 3 or b-c-NaN0 3 allowing removal of pure
KN03 or NaN0 3 , respectively.
Temperature effects can be seen by comparing Figures 2-12b and 2-12c.

Decreasing the temperature decreases the solubility of both salts and of mix-
tures of the salts. Thus in this system supersaturation can be caused by cooling,
and/or by water removal and/or by salting out. Because of the variation of the
saturation curves with temperature, fractionation of the two salts can be
obtained using temperature cycles. This is explored in Section 2.5.1.
Mass balance and equilibrium calculations can be done directly on the

triangular diagrams by a procedure similar to that used for extraction, or by
extracting the equilibrium information from the diagram and doing the mass
balances separately. These procedures are illustrated in Example 2-3.
48
Example 2-3. Crystallization of a Ternary System.
We wish to crystallize KN0 3 from a saturated solution of KN03 in

water at 100°C by both cooling to 25°C and by salting out by adding
NaN0 3 • The feed is 100 kg/hr of saturated KN03 solution and 20 kg/hr
of pure NaN0 3 are added. Find the yield of crystals and the final
mother liquor concentration.
Solution
A. Define. The process is sketched in the figure.
XKNO
3
=0.712
I--_~ L kg/hr
Xi
We wish to find C, L and the composition of the mother liquid Xi'
B. Explore. We can assume that the crystals and mother liquor are in
equilibrium at 25°C. Thus Figure 2-12c can be used.
C. Plan. First add the NaN0 3 to the feed to find the mixing point M.
This can be calculated graphically or analytically. Then this mixture
separates into KN0 3 crystals and mother liquor. The concentration of
the liquor can be read from Figure 2-12c while mass balances (graphi-
calor analytical) can be used to calculate C and L.
D. Do It. Adding F plus NaN0 3 gives M = 120 kg/hr,
XM,KNO. = ~~~ = 0.5933, xM,NaNO. = ~~ =0.167

This is shown on Figure 2-12c. Then the tie line is from the KN03 ver-
tex through M to ab at 25°C. Mother liquor is
49
XNaNO. - 0.32, XKNo. - 0.22 (see Figure 21-12c). To find C and L use
mass balances
Overall: C + L = M = 120
NaN0 3 : (C)(O.O) + (L)(0.32) = (120)(0.167)
Solving for L, we have L= °O~3~ 120 = 62.63 kg/hr
and thus C = 57.37 kg/hr.
E. Check. The KN0 3 mass balance can serve as a check:
C(1.0) + L(0.22) = M(0.593)
which is, 71.15 = 71.2 which is OK.
This check is really closer than can be expected considering the size of
the figure. Larger figures will in general be more accurate.
F. Generalization. The mass balances and equilibrium calculations can

be conveniently done on triangular diagrams for single stage contact.
This example showed first a mixing operation to find M followed by an
equilibrium tie line to find Xi and mass balances to find C and L. The
calculations are essentially the same as for single-stage extraction cal-
culations on a triangular diagram. Only the form of the equilibrium
data differs. This example also illustrated salting out when a common
ion is present.
Note in this example that the mother liquor and crystals exit together.
There will always be some mother liquor exiting with the crystals. Some crys-
tallizers also remove an overflow mother liquor which is free of crystals. As
shown in the example, the crystals must be separated from the mother liquor.
This can be done in a centrifuge or filter and the crystals are usually washed to
remove any adhering mother liquor which will contaminate the pure crystals.
50
).£...--: - - - - - - - - - - - -
,/ !.-C:.~ ___ - --SOA -------

/ ~c.,.
-------------- ---~---
10
~.::: :=-~=-------------IO°
o~~~----~--~~--~~--------------------------
("10 ) No z SO.
Figure 2-13. Three-dimensional plot for equilibrium for the system

NazSOcMgSOcwater (Fitch, 1970). Reprinted with pennis-
sion from Ind. Eng. Chem .. 62, (12), 6 (1970). Copyright
1970, American Chemical Society.
Double salts can also occur in ternary systems. When a variety of

hydrates and double salts are fonned the equilibrium data becomes complex.
An example for MgSOcNazS04-water is shown in Figure 2-13 (Fitch, 1970).
Figure 2-13 shows a three dimensional plot for this system with temperature as
the third axis. Note that three different hydrates and two double salts are
shown.
Three-dimensional plots are not convenient for calculations. Projections

at different temperatures are shown in Figures 2-14 a and b (Fitch, 1970). At
50°C (Figure 2-14a) the hexa-hydrate, MgS0406HzO, the double salt astrakan-
ite, Na2S04°MgS04°4H20, and anhydrous Na2S04 will be the solid phases.
At 20°C (Figure 2-14b) the hydrate is MgS04-7H20 while Na2S0clOH20
precipitates instead of Na2S04' At 20°C astrakanite still precipitates while at
10° and O°C no astrakanite fonns. The tie lines are shown in Figures 2-14. By
51
(a) MgS04
-
- -"- =. -- -- -~~~=~-~~-~-~---~
(b) MgSO.
As trokontte
1<-L------"--_"----'~~.,.N_O_2_S0_4_10_H_2_O_ _ -------> N0 2 50 4
Figure 2-14. Projections for system Na2S0cMgSOcwater. (Fitch 1970).

a. 50°C isotherm b. 20°, 10° and 0° isotherms. Reprinted with
permission from Ind. Eng. Chern., 62, (12), 6 (1970). Copy-
right 1970, American Chemical Society.
52
changing temperatures, cycles to fractionate the Na2S04 and MgS04 are

developed in the next section.
Solid solutions also occur in three component systems. The solid solu-
tion may cover the entire range of salt concentrations as does the system Pb
(N03h - Ba(N03h - water. This system is discussed later (see Figure 2-21).
The solid solution may also be discontinuous as is the system KBr - KCI -
water. Systems which form solid solutions cannot be purified in a single equili-
brium stage; thus, countercurrent cascades are used.
Note that the 3 component salt systems all have a common ion such as
Cl- or N03". A system such as KCI-NaN0rwater is really a four component
system. In water the double decomposition reaction
KCI + NaN03 = NaCI + KN0 3
occurs rapidly. Thus all four salts can be present. Since the composition can
be expressed as the composition of three of the salts plus water, this is a four
component system from the point of view of the phase rule. Alternately, it can
be considered a 5 component system with one independent reaction. Then F =
C - P + 2 - NRX • This will give the same degrees of freedom. These recipro-
cal salt pair systems are important in commercial applications. Equilibrium
data can be extremely complex and is discussed by Mullin (1972), Purdon and
Slater (1946), and Fitch (1970).
This completes the discussion of phase equilibrium and equilibrium cal-

culations for simple, single stage solution crystallizers. Most crystallization
from solution is done in crystallizers which approximate a single equilibrium
stage. Fractional crystallization requires more complex processes or counter-
current equipment and is discussed next.
2.5. FRACTIONAL CRYSTALLIZATION
For systems with two or more salts it is usually desirable to recover both salts
as pure compounds. For instance, in Example 2-3 a method for producing pure
53
NaN0 3 was illustrated, but no mention was made of what to do with the
saturated solution containing NaN0 3 , KN0 3 and water. Processes which
separate the solutes into two or more pure products are fractional crystalliza-
tion systems.
When solid solutions are not formed, processes using single equilibrium
stage crystallizers with various temperature cycles can be used. When solid
solutions are formed, fractional crystallization schemes require counter-current
cascades. Fractional crystallization can also be done by selective seeding (see
Section 4.5.).
2.S.1. Fractional Crystallization Processes with Temperature Swings
Fractionation cycles for systems without double salts are easily developed
(Fitch, 1970). The basic idea for a batch system is to operate at one tempera-
ture to produce pure crystals and solution at the drying-up point. Then add (or
remove) water and change the temperature to produce pure crystals of the other
component and solution at the second drying-up point. Then remove (or add)
water and readjust the temperature to harvest the first component as pure crys-
tals and return to the first drying-up point. This procedure is illustrated in Fig-
ure 2-15. Start at the drying-up point at temperature T 1 , b(Tl) on Figure 2-15a.
a b
B
Figure 2-15. Batch fractional crystallization for ternary systems without

double salts. a Schematic of basic process b. Application to
system KN0 3 - NaN0 3 -water
54
Then add water (following line b(Td - H20) to point a. Change the tempera-
ture to T2 and harvest product which is a hydrate of component A in equili-
brium with solution b(T2). After removing the hydrate, remove water from
solution b(T2) to point c. Change the temperature to Tl and harvest crystals B
in equilibrium with solution beT1). This returns us to our starting point having
harvested pure hydrate and pure B.
This basic scheme with minor modifications can be used for any ternary
system which does not form double salts or solid solutions. A simple
modification is illustrated in Figure 2-15b for the KNO r NaN03 - water sys-
tem. Starting at b(1000), water is added to reach point a and the system is
cooled to 25°C. Pure KN0 3 crystals are harvested in equilibrium with b(25°).
Water is then removed to reach point c, and the mixture is heated to 100°C.
Pure NaN0 3 crystals are harvested in equilibrium with the solution b(I000).
Now the process can be repeated.
The process can be made continuous by adding a feed stream. The feed
can be added to either solution b(Tl) or b(T2). Both approaches should be
investigated. The continuous process is illustrated in Example 24.
Example 2-4. Continuous Fractional Crystallization of

NaC03-NaCI-water.
A feed of 100 kg/hr which is 30 wt % Na2C03 and 45 wt % NaCI is to

be fractionally crystallized by alternating the temperature between 0°
and 30°C. The drying-up points are (Fitch, 1970),
Anhydrous NaCI and Na2C03olOH20 hydrate are desired products.

Find the yields of each product and the amounts of water added and
evaporated when the feed is added to solution b(300).
55
Figure 2-16. Fractional crystallization system for Example 2-4.
Solution
A. Define. The process is sketched in Figure 2-16. In practice, several

of the steps such as mixing solution b(30), F, and water will be done
simultaneously, and the evaporation will probably be done in a vacuum
crystallizer. We wish to find the flow rates of Na2C03'lOH20, NaC!,
water added and removed, and the recycle flow rate b(300).
B and C. Explore and Plan. Required equilibrium data are given in the
problem statement. External mass balances can be used to find product
flow rates. The equilibrium diagram can be sketched (see Figure 2-17)
and used as a guide. First find fulcrum point c at the intersection of
lines H 20 - b(O) and NaCI - b(30). Since the flow rate of NaCI can be
calculated and point c serves as a fulcrum, the flow rate of stream b(30),
F b (30), can be calculated.
Fb(30) NaCI'c
=
c'b(30)
With Fb (30) and the feed rate known, point M can be found.
FM
= -====-
M'b(30)
and FM = F b (30) + F. Point a is found by the construction shown. Then

water addition to get from M to point a can be found from the lever-arm
rule.
M'a
=
56
or mass balances. Then Fa = FH.o + F M • Flow rate of b(OO) can be

found from a mass balance
Fb(O) = Fa - Fhydrate
Finally, the water removal to go from b(OO) to point c can be found

from the lever arm rule.
FH.o,out b(O)'c
Fb(o) H 20'c
or from a mass balance (see Figure 2-16):

FH.o,out = Fb(o) - F NaCl - F b (30)
D. Do It. Start with external balances,
Since the water flow rates are unknown, we need external balances for
Na Cl and Na2C03. All NaCI goes out in NaCI product,
FNaCl = 0.45F = 45 kg/hr
While Na2C03 exits in the Na2C03 product,
Na2C03 out = 0.3F = 30 kg/hr

However, this product is a hydrate.
MWhydrate
FNa.CO"10H.O = ( MWanhydrous )(Fanhydrous)
= (285.99 )(30) = 8107 kg/hr

105.83 .
57
0.6
wt. fraction
Nael
0.4
0.2
Water 0 ~--'---L------'-----'S--L---'
o 0.2 0.4
Hydrate
Figure 2-17. Solution to Example 2-4.
Plot points b(O) and b(30) which are the drying up points. Then plot
NaCI = 1.00 and the hydrate, xNa.CO,.lOH,O = 0.370. Draw line H2 0 -
b(OO) and extend to extension with line NaCI - b(300). Intersection is
point c. See Figure 2-17. From the lever arm rule,
F =F NaCI·c = (45) (9.25) = 32 6

b(30) NaCl (1275)·
c·b(30) .
To findM.
FM = F + Fb(30) = 100 + 32.6 = 132.6

and the lever-arm rule is,
Fb(30) 32.6 F·M

--=--= (see Figure 2-17)
F 100
M-b(30)
58
Draw lines b(O)-Hydrate and M-Water. Intersection is point a. Then

water addition flow rate is,
M·I = (132.6) 6.4 = 157.2 kglhr

5.4
Then,
and Fb(o) = Fa - Fhydrate = 289.3 - 81.1 = 208.2 kglhr

Finally, water out from the mass balance is,
= 208.2 - 45 - 32.6 = 130.6 kglhr
E. Check. The external water balance can be used as one check.
180.16
100(0.25) + 157.7 = 130.6 + 81.07( 285.99 )
which is 182.2 compared to 181.7. This is OK since it is within the

accuracy of the graph. A second check is to solve everything analyti-
cally. This is done in Problem 2-El. Results agree within the accuracy
of the graphical method.
F. Generalization. Although the details of the process will differ, the

method employed here can be used for other systems which do not form
double salts or solid solutions. Note that the calculation started with the
drying-up solution, b(OO), which was not recycled. This allowed calcu-
59
lation of fulcrum point c and then the recycle stream flow rate Fb (30)'
Then straight-forward mixing calculations could be used. This some-
what round-about approach is required whenever the flow rate of the
recycle stream is unknown. Again the idea of this procedure can be
applied to other systems.
An alternate method of doing this fractional crystallization is to

add the feed to solution b(OO) instead of b(300). This alternate process
is explored in Problem 2-D7, and is probably preferable.
The fractional crystallization problems can also be set up for analytical

solution instead of graphical solution. This requires quite a bit of algebra, but
is computer friendly. For example, for Example 2-4 one would do the mass
balances in the same order after solving the external mass balances. The equa-
tion for removal of water from solution b(OO) and for the separation of mixture
c into pure crystals and solution b(300) must be solved simultaneously. This
gives 6 equations and 6 unknowns. Once the flow rate Fb(30) is known, FM and
the coordinates of point M are easily found from mass balances for the mixing
operation. Then Fa and FH.o,in can be found from the mass balances for mixing
in water and the mass balances for the separation of stream a. If the flow sheet
is known, this procedure is straight-forward but a bit tedious. However, once
the general equations have been developed, it is extremely convenient to study
the effect of changing variables such as the feed composition. The analytical
solution methods are explored in Problems 2-E1 and 2-E2. The graphical
approach remains very useful for developing workable processes and for guid-
ing the analytical calculations.
Note that the fractional crystallization process shown in Figure 2-16 is

much more complicated than a simple crystallization. However, the fractional
crystallization does produce two pure products and no saturated solutions
which must be disposed of.
In the examples up to now the drying up points have been used as the
saturated solutions. This will maximize the yield of the pure products, but is
not always the best choice. For the KCI- NaCI- water system shown in Figure
2-18 (Fitch, 1970) it is more economical to use an operating point (a) which
60
Kel
\
\
.
\
\
F
/
I /
\ /
\ //
/
/
/
\ b(2) //
,,_~c
-- - .. - -. Noel
Figure 2-18. Cycle for fractional crystallization of KCI and NaCI without
evaporation or dilution. (Fitch, 1970). Reprinted with permis-
sion from Ind. Eng. Chern., 62 (12),6 (1970). Copyright 1972,
does not maximize yield instead of evaporating water to reach drying-up point
bel). The fractional crystallization cycle can be followed by starting at b(2).
After cooling the solution to 20°C, a crop of KCI crystals can be removed. A
saturated solution at 20°C (point a) results. The solution at point a is mixed
with fresh feed and heated to 80°C to reach point c. KCI is leached from this
mixture leaving NaCI and saturated solution b(2). This cycle is used commer-
cially and avoids the need for evaporation or dilution.
The fractional crystallization of systems with double salts (e.g.

MgS04 - Na2S04-water shown in Figures 2-13 and 2-14) can be more
complex. At 50°C (Figure 2-14a) the astrakanite is congruently soluble
(adding water leads to a saturated solution in equilibrium with the astrakanite).
It is possible to separate MgS04'6H20 and astrakanite, or to separate Na2S04
and astrakanite, but a complete separation is not possible at 50°. Isotherms
where astrakaniteeither is not congruently soluble or where astrakanite ceases
to exist must be found. Figure 2-14b shows that 0° and 10° the double salt
61
~o
!
Astrokanite
Astrokcnte
DIssolver
No-Rich 50·
Feed
!LlQuore
Glauber Salt
Crystolliz. Mg-Rch
O·
1 Fi
LIQuor bb)
Evaporate at 50 0
.....
Separate
Crystallizing
Solids
Astrakanite
1 1
Na 2 S04 " 10 H2O
Seoarate
Astrokanite -
Llquabll) 1
MgS04" 7H~
Crystalhzer
1
- Separate
MgS04 7H 2 O
1
MgS0 4 7H 2 0
Figure 2-19. Process flowsheet for fractionation for Na2C04 and MgS04.
(Fitch, 1970). Reprinted with pennission from Ind. Eng.
Chern .• 62, (12), 6 (1970). Copyright 1970, American Chemi-
cal Society.
astrakanite no longer fonns. Thus one possibility is to develop a fractional cry-

stallization cycle using these two temperatures. Several such cycles are possi-
ble (see Problem 2-B2) and will be similar to the cycles illustrated in Figures
2-15 to 2-18.
An alternative is to develop a process cycling between 50° and O°C and

using the astrakanite fonned at 50°C as a recycle stream. A process ftowsheet
62
for this process is shown in Figure 2-19 (Fitch, 1970). This cycle requires three
crystallizers; but will have lower evaporation costs than cycling between 0° and
lOoC.
Trace components are usually present in the feeds to fractional crystalli-

zation processes. Unless these trace components are removed in a liquid purge
stream, they will slowly build up in the liquid phase until it is saturated. Then
the trace components are no longer traces and these components must be
included in the phase equilibria. For simplicity, we will use purge streams to
keep the concentration of trace components low.
More complex ternary, four-component and many-component systems

are discussed by Fitch (1970). As the number of components increases the
number of possible process configurations increases. The basic principles
developed for ternary systems can still be used for these complex processes.
2.5.2. Processes Using Counter-Current Cascades
When solid solutions occur, a single equilibrium stage will not give either
solute as a pure component. In this case counter-current cascades are required
to obtain the desired fractionation. An example of a system which forms a
solid solution throughout the entire range is Pb(N03h - Ba(N03h - water.
Pb(N03h is the more soluble salt. A schematic of the counter current cascade
for fractionation of these salts is shown in Figure 2-20. Each box is assumed to
be an equilibrium stage. The more soluble Pb(N03h tends to concentrate in
Filter
Evaporator
Sj
T
Water
. :GJ:
Yj
... :[j]: ..
C j _1
Xj_1
Feed
Ba(N0 3 )2
Product
-U2 -zs:;- Pb(N0 3 l2
Product
Figure 2-20. Countercurrent cascade for fractional crystallization of solid

solution.
63
the liquid. The less soluble Ba(N03h tends to stay in the crystals. The net
result is movement of Pb(N03h to the right and Ba(N03h to the left. The
solution leaving the right end of the cascade is sent to an evaporator where
water is removed. A portion of the crystals formed are taken as Pb(N03h pro-
duct while the remainder are refluxed. A portion of the crystals leaving the left
end of the cascade are withdrawn as Ba(N03h product. The remaining crys-
tals are dissolved in water and refluxed (stream YN+l).
The counter-current cascade is very similar to fractional extraction

columns (e.g., Wankat, 1988). The differences between overall and component
flow rates of solid and liquid phases in each section of the cascade will be con-
stant. Thus we can define difference points in the same way as for extraction.
The resulting difference points can be plotted on triangular diagrams. Stages
can now be stepped off on the triangular diagrams as done in extraction (e.g.,
Wankat, 1988). Unfortunately, this construction squeezes the difference points
together near the solvent vertex.
An alternate construction is to plot the solvent to solute ratio, N, versus

the fraction of salts which are the more soluble salt, Pb(N03h, in the solid and
solution. This alternate is shown for crystallization in Mullin (1972). More
detail is given in books and articles on extraction (Smith, 1963; Treybal, 1980;
Wankat, 1982). Let Nand Y be mass ratios in solution,
N = kg solvent = kg water (2-7a)

kg salts kg Pb(N03h + kg Ba(N03h
Y = kg more soluble salt = kg Pb(N03h (2-7b)

total salts kg Pb(N03h + kg Ba(N03h
and X is the mass ratio in the solid phase.
X = kg more soluble salt = kg Pb(N03h (2-7c)

total salts kg Pb(N03h + kg Ba(N03h
The equilibrium data (Mullin, 1972) are shown in these units in Figure 2-21
where both a McCabe-Thiele type diagram and a Ponchon-Savarit type
diagram (similar to triangular diagrams) are shown. Equilibrium tie lines on
64
X or Y
0 0.2 0.4 0.6 0,8 1,0
12
10
6
N
4
SI
2
0
Co
~2 I X
PbINO~2 prod
XProd I
I I
1 I
I I
I. 0 "------:-1-"-----+----:::±==-~
0,8
Y 0,6
0.4
0.2
o ,,--,---,---,--,,--,---,---,---,--,..-J
o 0,2 0.4 0,6 0.8 1.0
X
Figure 2-21. Fractional crystallization of solid solution Pb(N03h -
Ba(N03h - water. Data from Mullin (1972).
65
the top diagrams can be obtained from the McCabe-Thiele diagram by the same
methods used for enthalpy-composition or triangular diagrams (e.g., Wankat,
1988). The top diagram in Figure 2-21 is a transformation of the triangular
diagram into the new set of variables.
These mass ratios are related to mass fractions,
Yw (2-8a)
N=-------
YPb(NO,). + YBa(NO,).
YPb(NO,). (2-8b)
y=-------
YPb(NO,). + YBa(NO,).
XPb(NO,). (2-8c)
x=-------
where x is mass fraction in the solid and Y is mass fraction in the solution.
Since these mass ratio units are on a water-free basis, all flow rates must
be converted to a water-free basis. Flow rate C is kg salt in the crystals per
hour, and is already in this basis. Flow rate S is kg salt in solution/hr. S can be
related to the total solution flow rate
(2-9)
S = (Total solution flow rate) x (YPb(No,). + YBa(NO,»)
In these units the differences in passing streams to the right of the feed in Fig-
ure 2-21 are,
(2-10a)
Sj - C j- 1 = ~l = constant
where PPb(NO,). is the kg/hr of product. The Pb(N03h and water balances are
(2-lOb)
(2-lOc)
66
Similar equations are valid for ~2 on the left side of the feed in Figure 2-21.
These equations allow calculations of the coordinates of ~l and .12, Since
these equations are the same as mixing equations, they indicate that there is a
straight line from the two passing streams (Sj and Cj-l) and the ~ point.
In the evaporator only water is removed. Since the ratios are on a

water-free basis, Y 1 = Xo, and ~l must be at this value of the abscissa. The
location of ~l can be determined from the external reflux ratio, CO/PPb(No,)•.
From the lever-arm rule
Co ~lSl (2-11)
-=---=
PPb(NO')2 COSl
and the ~l point can be found. The ~2 point can be found where the extension
of the feed line ~lF intersects the vertical line through XBa(NO,lz,product. This
use of the feed line is the same use as in extraction and distillation. The ~
points are shown in Figure 2-21.
Stages can be stepped off on the Ponchon diagram using the same pro-
cedure used for triangular diagrams, or the McCabe-Thiele diagram can be
used. In Figure 2-21 the McCabe-Thiele diagram is used to find tie lines and
the ~ points are used for the operating lines. Figure 2-21 shows an example
where the Pb(N03h product is 93% Pb(N03)2 while the Ba(N03h product is
5% Pb(N03h. The feed is a 50 % mixture of both salts in crystal form, and a
reflux ratio of 2.0 is used. A total of 3 equilibrium stages are required and the
optimum feed is the second stage. If the McCabe-Thiele diagram is used to
step off stages, curved operating lines must be generated from the Ponchon or
triangular diagram (an example of this procedure is in Wankat, 1988).
Counter-current processes for fractional crystallization are not a common
separation technique. There are only a few solid solutions which form a solid
solution continuously over the entire range as Pb(N03h - Ba(N03h - water
does. Other systems such as KBr - KCI - water form solid solutions over part
of the range but there is a discontinuity in this range. A complete separation
can be obtained only by operating two cascades at different temperatures and
cycling between the temperatures (Fitch, 1976).
67
Another problem with counter-current fractional crystallization is mass

transfer in crystallization is not analogous to distillation or extraction. There is
essentially no mass transfer between the inside of a crystal and solution. To
obtain equilibrium each stage in Figure 2-20 requires dissolution of the crystals
followed by recrystallization. When solubility is temperature dependent, this
can be done by heating crystals plus liquid to dissolve the crystals and then
recooling. For other systems additional solvent may have to be added to dis-
solve the crystals. Then to cause crystallization the solvent has to be eva-
porated. These steps make the process more expensive. In addition, separation
of the crystals from the mother liquor is usually not perfect. The entrained
mother liquor reduces the stage efficiency. Fractional crystallization of solid
solutions is probably applicable either when only a few stages are required or
when the products are very valuable.
2.6. PRECIPITATION
Precipitation is an important industrial process which is closely related to cry-

stallization from solution. In precipitation a solution is made supersaturated so
that dissolved solutes precipitate out. Precipitation differs from crystallization
since the precipitate is usually amorphous, has a poorly defined size and shape,
is often an aggregate, and is usually not pure. Precipitation and crystallization
are often complementary processes since precipitation is used for a "rough cut"
while crystallization is used for final purification. Precipitation is commonly
used in mineral processing and for the production of photographic chemicals,
paints, polymers, pharmaceuticals, and will undoubtedly be an important unit
operation in the emerging biotechnology industry.
Equipment for precipitation is often very similiar to that used for crystall-
ization. Both batch and continuous operations are used, although batch is prob-
ably more common since many precipitations are done for relatively low
volume products. Precipitation operations usually have a mixing operation
where various reagents are added to make the solution supersaturated. This is
followed by a holding period which allows for an induction period before
nucleii form and a latent period before the supersaturation starts to decrease
68
markedly. These periods are shown schematically in Figure 2-22 (Mullin,

1972). Once the precipitate starts to grow the desupersaturation often occurs at
a constant rate as shown in Figure 2-22. Towards the end of the
de supersaturation period the rate decreases as the concentration approaches
equilibrium. During this period aging often occurs. During aging small crys-
tals and precipitates redissolve and the solute is redeposited onto the larger pre-
cipitates. The result is a uniform crop of fairly large precipitate which is rela-
tively easy to separate from the solution by centrifugation (Bell et al, 1983;
Belter et al, 1988) or filtration (Belter et al, 1988). The kinetics of these steps
are discussed in Chapter 3.
It is convenient to classify precipitation processes based on the method

used to produce the supersaturation.
1. Cooling and solvent evaporation. Supersaturation can be caused by cooling

and or evaporation of solvent. This is very similiar to crystallization and the
equipment is very similiar.
2. Reaction. Some products are made by mixing reagents which react to form a
precipitate. Examples are ammonium sulphate, sodium bicarbonate, and
ammonium diuranate (Mullin, 1972).
3. Salting out. High salt concentrations often cause the precipitation of solute.
This effect was explored for common ions in Eq. (2-3) and in Example 2-3.
Salting out is also commonly used for protein precipitation (Bell et aI, 1983;
Belter et al, 1988). The Hofmeister series lists the effectiveness of anions as
citrate> phosphate> sulphate> acetate = chloride> nitrate> thiocynate
Cations follow the order

ammonium> potassium> sodium.
For the high salt concentrations required for salting out, the solubility of a pro-
tein usually follows the empirical Cohn equation,
(2-12)
69
I L
Solute
concentratio
c' - - --E
Time
Figure 2-22. Desupersaturation curve. I is induction, L is latent period, D is

constant rate desupersaturation, and E is equilibrium.
where Cs is the salt concentration, and K and ~ are constants. Constant ~

depends on the protein, pH and temperature, but is usually independent of the
salt. The slope of the salting out curve, K, depends on the salt and the protein.
Note that Eq. (2-12) is valid only at high salt concentrations. At low salt con-
centrations the protein solubility becomes constant. Since relatively large
amounts of salt are required, an inexpensive salt is desired. The combination of
cheapness plus the Hofmeister series leads to <N"H4hS04 as the most common
salt for protein precipitation. In fact, ammonium sulfate precipitation is so
common that other salts may not be considered.
4. Antisolvents. Antisolvents or nonsolvents can be added to cause precipita-

tion. This was illustrated in Figure 2-6. For proteins it is common to add
ethanol, acetone or polyethylene glycol to cause precipitation. The precipita-
tion is preferentially done at low temperature to prevent denaturation of the
protein, and at ionic strengths in the range from 0.05 to 0.2 M (Belter et al,
1988). A major application of precipitation by addition of an anti solvent is the
Cohn ethanol precipitation method used for the fractionation of blood plasma
proteins. The Cohn fractionation procedure varies the mole fraction ethanol and
pH in a strictly prescribed sequence to produce several (usually 6) fractions
which contain different blood proteins (Bell et al, 1983).
5. pH. The solubility of many compounds will be effected by pH. Proteins

70
have a net charge of zero at a particular pH called the isoelectric point or pI.
Each protein has a unique pI. Protein solubility goes through a minimum at the
pI. For proteins precipitation is usually done by adjusting the pH to the isoelec-
tric point, and then adding salt or an antisolvent. pH changes are also com-
monly used to precipitate metal ions from acidic sulphate solutions in the min-
ing industry.
6. Temperature increases. Increasing the temperature usually increases the

solubility; however, for proteins temperature increases may cause denaturation
which can drastically decrease solubility. Details of this method are discussed
by Belter et al (1988).
Precipitation can be carried out in one step to remove a large number of

compounds in a single precipitate. This is the easiest application and is most
common. Fractional precipitation requires a series of steps with each step
optimized to precipitate one component. Since any solute left in solution dur-
ing one step will probably precipitate in the next step, it is not possible to do
extremely sharp purifications by fractional precipitation. In addition, in most
cases any added salts or other chemicals have to be removed before the product
can be used.
In this chapter equilibrium calculation methods for crystallization from solution

have been explored. At the end of this chapter you should be able to do the fol-
lowing:
1. Explain and classify the types of crystallizers and precipitators used based
on how supersaturation is achieved.
2. Plot solubility data and do yield calculations. Discuss the factors which
effect solubility.
3. Use temperature-composition and enthalpy-composition diagrams for

crystallizer calculations for binary systems.
4. Use ternary diagrams for crystallizer calculations.

71
5. Develop and do calculations for fractional crystallization processes for ter-

nary systems without and with solid solutions.
REFERENCES
Azoury, R., W.G. Robertson, and J. Garside, "Generation of supersaturation

using reverse osmosis," Chemical Engr. Research Design, 65 (4), 342 (1987).
Bell, DJ., M. Hoare, and P. Dunnill, "The formation of protein precipitates and
their centrifugal recovery," in A. Fiechter, (Ed.), Advances in Biochemical
Engineering/Biotechnology, 26, Downstream Processing, Spring-Verlag, Ber-
lin, 1983, 1-72.
Belter, P.A., E.L. Cussler, and W.-S. Hu, Bioseparations. Downstream Pro-
cessing for Biotechnology, Wiley-Interscience, New York, 1988, Chapters 8
and 10.
Bennett, RC., "Crystallization from solution", in RH. Perry and D.W. Green
(Eds.), Perry's Chemical Engineer's Handbook, 6th ed., McGraw-Hill, New
York, 1984, p. 19-24.
Bennett, RC., "Matching crystallizer to material," Chem. Eng., 95 (8), 118

(May 23, 1988).
Bennett, RC. and M. Van Buren, "Commercial urea crystallization," Chem.

Eng. Prog. Symp. Ser., 65 (95), 44 (1969).
Fitch, B., "How to design fractional crystallization processes," Ind. Eng. Chem.,
62 (12), 6 (Dec. 1970).
Fitch, B., "Design of fractional crystallization processes involving solid solu-

tions," AIChE Symp. Ser., 72 (153), 79 (1976).
Garrett, D.E., "Crystallization equipment applic(!tions," Ind. Eng. Chem., 53,

623 (1961).
72
Lozano, J .A.F., "Recovery of potassium magnesium sulfate double salt from

seawater bittern," Ind. Eng. Chem. Process Des. Develop., 15,445 (1976).
McCabe, W.L., J.C. Smith, and P. Harriott, Unit Operations of Chemical

Engineering, 4th ed., McGraw-Hill, New York, 1985, Chapt. 28.
Mehra, D.K., "Selecting evaporators," Chem. Eng., 93 (3), 56 (Feb. 3, 1986).

Moyers, C.G., Jr. and RW. Rousseau, "Crystallization operations," in RW.
Rousseau (Ed.), Handbook of Separation Process Technology, Wiley, New
York, 1987, Chapt. 11.
Mullin, J.W., Crystallisation, 2nd ed., Butterworths, London, 1972.
Nyvlt, J., Industrial Crystallization, 2nd Ed., Verlag Chemie, Weinheim, 1982.
Osburn, J.O., "Crystallization," in RH. Perry, C.H. Chilton, S.D. Kirkpatrick

(Eds.), Chemical Engineer's Handbook, 4th edition, McGraw-Hill, New York,
1963, 17-7.
Perry, RH. and D.W. Green (Eds.), Perry's Chemical Engineers' Handbook,
6th ed., McGraw-Hill, New York, 1984.
Purdon, F.F. and V.W. Slater, Aqueous Solution and the Phase Diagram,
Arnold, London, 1946.
Seidell, A., Solubilities of Inorganic and Metal Organic Compounds, 4th edi-
tion (W.P. Linke, Ed.), Vol. I, Van Nostrand, New York, 1958. Vol. II, Ameri-
can Chemical Society, Washington, D.C., 1965.
Singh, G., "Crystallization from solutions," in P.A. Schweitzer (Ed.), Handbook

of Separation Techniques for Chemical Engineers, McGraw-Hill, New York,
1979, Section 2.4.
Smith, B.D., Design of Equilibrium Stage Processes, McGraw-Hill, New York,

1963.
73
Stephen, H. and T. Stephen, Solubilities of Inorganic and Organic Compounds

(in 5 parts), Pergamon, London, 1963.
Tavare, N.S., "Batch crystallization: A review," Chern. Eng. Commun .. 61,259

(1987).
Treybal, R.E., Mass Transfer Operations, 3rd ed., McGraw-Hill, New York,
1980.
Wankat, P.C., "Advanced graphical extraction calculations," in J.M. Calo, and

EJ. Henley, (eds.), AIChE Modular Instruction, Series B, Stagewise and Mass
Transfer Operations, vol. 3, Extraction and Leaching, Module B 3.3, p. 17,
AIChE, New York, 1982.
Wankat, P.C., Equilibrium-Staged Separations, Elsevier, New York, 1988.
HOMEWORK
A. Discussion Problems
AI. How does an evaporative crystallizer operated below one atmosphere

pressure differ from a vacuum crystallizer?
A2. Sketch the entire process including methanol recovery if methanol were
added to salt out KCl (see Figure 2-6).
A3. Crystallization processes which utilize only energy removal are usually
preferred to those involving addition of a salt or anti solvent if all things
are equal. Explain why. Give some examples of when the addition of a
salt or antisolvent is desireable.
A4. A process for fractional crystallization of KCI and NaCI is shown in Fig-
ure 2-18. Sketch the flow-sheet for this process.
74
AS. Define the following tenns.
a. hydrate
b. anhydrous
c. Miers diagram
d. metastable
e. labile
f. salting out
g. eutectic
h. congruent melting point
1. incongruent melting point
j. reciprocal salt pair
k. double salt
A6. Why is it better to plot the concentration in Figures 2-1 as mass ratios in
units, kg anhydrous salt/lOO kg water, instead of in units kg hydrated
salt/lOO kg water? Focus on what happens at the phase change points.
A7. Explain how solubility data and temperature-composition diagrams

correspond.
A8. Why are evaporation costs significantly higher at lOoC than at SOo C?
A9. Construct your key relations chart for this chapter. That is, on one page
list all facts, equations, figures, etc. which you would want to help you
solve problems on a test.
B. Generation of Alternatives
B 1. In fractional crystallization many alternate processes are often available.

For the fractionation for KCI from NaCI shown in Figure 2-18 several
alternative processes using evaporation and addition of water can be
75
used. Sketch the flow sheets for at least 3 such processes. Then sketch
the processes on a triangular diagram.
B2. Develop at least two different cycles for fractionating MgS04 and
Na2S04 alternating between O°C and lOoC (see Figure 2-14b). Sketch
both the process flow sheet and the process on a triangular diagram.
C. Derivations
C1. Derive Eq. (2-5a). Note that

MWhydrate = MWanhydrous + n MWwater'
C2. Derive Eq. (2-6a).
C3. Show that Eq. (2-9) can also be written as
S = (Total solution flow rate)

N+l
C4. Write the Equations for .12 for the cascade in Figure 2-21. Develop the
coordinates of .12,
C5. Prove that points .11 (XL\" NL\), S/Yj , Nj ) and q-1 (Xj-1, 0) are on a
straight line in Figure 2-21.
D. Problems
Dl. Determine if the following compounds in water satisfy Eq. (2-1b). If Eq.
(2-1 b) is satisfied determine the constants a and b. Data are in Table 2-1.
b. copper chloride
c. silver nitrate
d. sucrose
e. sodium sulphate
76
D2. Determine the yield of crystals deposited at equilibrium and the yield of
anhydrous crystals for the following problems. Data are in Table 2-l.
a. A saturated sugar solution (sucrose) is cooled from 80° to 20°e.

100 kg of water are present.
b. A saturated sucrose solution at 80°C has 1/2 of its water evaporated

and the solution is then cooled to 20°C. Initially 100 kg of water
are present.
c. A copper chloride solution initially saturated at 75°C is cooled to

10°C. 100 kg of water are present. (See Problem 2-D1b for inter-
polation).
D3. 100 kg of a solution which is 0.60 wt. fraction Na2C03 and 0040 wt frac-
tion water is cooled from an initial temperature of 80°C. Data is in Fig-
ure 2-10.
a. If the mixture is cooled to 3504 °C what three phases are present at

equilibrium? If 20,000 kJ have been removed how many kg of each
of these 3 phases are present.
b. If the mixture is cooled to O°C, what two phases are in equilibrium?

How much of each phase is present? How many kJ had to be
removed to cool the original solution from 80°C to O°C?
D4. A vacuum crystallizer is operated so that the mixed magma has an

overall concentration of 30 wt % Na2C03 and 70 % water and is at
20°e. The entering feed is a 25 wt % Na2C03 solution and is under
pressure so that it remains a liquid until it flashes in the crystallizer. Feed
rate is 100 kg/hr. Data is in Figure 2-10.
a. What crystals are formed? How many kg of steam are removed and
how many kg of magma mixture are present?
b. Data for the partial pressure of aqueous Na2C03 at 20°C is (perry &
Green, 1984, p. 3-73).
77
wt. % Na2C03 I0 5 10 15
p,mmHg 17.5 17.2 16.8 16.3
Estimate the pressure of the crystallizer.
c. From the steam tables saturated steam at 20°C has an enthalpy of -

2537 kJ/kg. (perry & Green, 1984, p. 3-238). What is the required
enthalpy of the feed? What temperature is it at? Is 1 atmosphere
pressure high enough to prevent it from boiling?
d. How many kg of crystals are produced?
D5. We have a solid mixture of 80 % KN0 3 and 20 % NaN03 at 50°C (see

Figures 2-12 a,b). What would you do to precipitate out pure KN0 3? If
there is initially 100 kg of mixture, what is the maximum amount of
KN03 which can be recovered. What is the concentration of solution in
equilibrium with the pure KN03 ? Do this problem at 50°C. Do not use
a fractional crystallization cycle.
D6. What happens if the feed to Example 2-4 also contains 0.02 wt %
NaN0 3 ? Where would you add a purge stream? If we desire to keep the
maximum concentration in the liquor below 1.0 wt % NaN0 3, what
purge flow rate should be used?
D7. Redo Example 2-4, but with feed added to the solution b(OO) instead of
solution b(300). A flow chart of one way of doing this is shown below.
Find crystal product flow rates, the water flow rates and recycle flow rate
b(OO).
Note that you must determine whether water is added or removed from
mixture M.
78
D8. Citric Acid (C3140H (COOHh) is dissolved in 10 kg water at 100°C to

form a saturated solution. This solution is cooled to 20°C and simultane-
ously 0.8 kg of water are evaporated. Assuming the solution is saturated
at 20°C, how many kg of hydrated crystals, C3140H (COOHh • IH20
are formed? How many kg of anhydrous C314 OH(COOHh will result?
D9. We wish to separate Pb(N03h from Ba(N03h. The feed is 55 wt %

Pb(N03h and 45 wt % Ba(N03h on a water free basis, and enters stage
N. The feed rate is 100 kg/In of total salts in enough water so that the
mixture is a saturated solution. A reflux ratio of 3.0 (lb crystals
returned)/(lb crystals taken as Pb(N03h product) is used. A crystal pro-
duct which is 90 wt % Pb(N03h is desired.
a. How many stages are needed?
b. How much water is required in the feed?
c. What is the composition of the Ba(N03h crystals withdrawn from
stage N?
Data are in Figure 2-21.
DlO. We have 100 kg of a mixture which is 20 kg MgS04 (anhydrous), 10 kg

Na2S04 and 70 kg water. We evaporate 20 kg water. Operation is at
50°C. Data are in Figure 2-14a.
a. What are the composition of salt and of solution which result after
crystallization is complete?
b. How many kg of each are obtained?
Dl!. At a particular pH bovine serum albumin (BSA) in (NH4h S04 has con-
stants in the Cohn equation of approximately K = 7.65 and ~ = 21.6
where Cs is in moles/liter and c * is protein solubility in g/liter (Belter et
al, 1988). The initial charge is a saturated solution of BSA at Cs = 2.6M
and we desire a 99.9% yield of BSA. What final salt concentration
should be used?
D 12. Repeat Example 2-1A, but for the case where the solution is initially at
100°C. Show that Eq. (2-5a) gives an impossible result. Explain why.
79
E. More Complex Problems
El. Solve Example 2-4 with analytical equations instead of graphically.

That is, solve the mass balances and find Fb(o), Fc, Fb(30)' FW,out, FM , Fa'
and FW,in' Do this first in algebraic form without numbers. This would
then be useful for a programmable calculator or computer solution.
Then use the numbers of Example 2-4 and redo that example.
Hint: Read the paragraph following Example 2-4 which outlines the
procedure.
E2. Redo Problem 2-D7 but do it analytically instead of graphically. Fol-

low the suggestions of Problem 2-El.
Chapter 3
NUCLEATION AND CRYSTAL GROWTH
The crystal size distribution (CSD) and the shape of the crystals depends on
kinetics and mass transfer in the crystallizer. Equilibrium data is needed to
determine the driving forces for nucleation and crystal growth, but equilibrium
data by itself is not sufficient to predict the CSD or crystal shape. In this
chapter we will first look briefly at crystal shapes. Next, the imperfectly under-
stood area of nucleation will be explored, and some empirical expressions for
nucleation rates will be presented. Then theories of crystal growth and mass
transfer will be developed. These theories will be used in Chapter 4 to predict
crystal size distributions. While reading this chapter you should not think of
this as the "truth". Instead, consider this as part way down the path to scientific
knowledge.
This chapter continues to focus on crystallization and precipitation from

solution. However, many of the fundamentals developed here are also applica-
ble to melt crystallization.
3.1. CRYSTAL SHAPES
Every chemical compound has a unique crystal shape; however, the crystal
shape depends enormously on the conditions in the crystallizer. This statement
seems like a contradiction but is not. The unique aspect of the crystal is that
the angles between adjacent faces are constant. The sizes of the faces can
change, but the angles do not change and are characteristic of the substance.
This is known as Haiiy's law. These angles can be measured by an instrument
80
81
[OJ
(0) Tabular (b) Prismatic (c) ACicular
(d)
· / branch
Pnmary
Main
!
Secondary branch
stem
Figure 3-1. Shapes of a hexagonal crystal (a, b and c). a. Tabular. b.

Prismatic. c. Needle or acicular. d Dendrites. Mullin (1972).
Reprinted by permission of Butterworths. Copyright 1972.
called a goniometer. With constant angles but different sizes of the faces, the
shape of a crystal can vary enormously. This is illustrated in Figure 3-1 (Mul-
lin, 1972) which shows three crystals which are all hexagonal. This variation
in shape is called the crystal habit. The habit is affected by the rate of crystalli-
zation, impurities, agitation, solvent used, degree of supersaturation or super-
cooling and so forth. Note that some substances illustrate polymorphism which
means the substance can crystallize into two or more unique forms. A good
example is carbon which can crystallize into graphite or diamonds.
82
Crystallography (the study of crystals) is a well-developed science

(Bunn, 1961; Mullin, 1972; Wyckoff, 1963-68). Most of the details of crystal-
lography are beyond the scope of a book on separations. Thus, only a very
brief account will be given here. A more detailed, but still brief, presentation is
given by Mullin (1972).
There are 32 possible classes of crystals based on crystal symmetry.
These 32 classes can be broken down into seven systems: regular (5 classes),
tetragonal (7), orthorhombic (3), monoclinic (3), triclinic (2), trigonal (5) and
hexagonal (7). The first six systems can be defined by referring to 3 axes while
the hexagonal system requires 4 axes. The faces of the crystals can be num-
bered in terms of the Miller indices. More than 80% of the elements and sim-
03 011
oil III
121
liD 110
2io 100
210 010
ooi 120
Iii IIi
Sucrose (monoclinic) Copper sulphate (triclinicl CalCite (trigonal)
101 001
001
Iii
110
lio
120
a
a a a 02i
130 ,2 ~
--. oli
loi 102 Iii
100
Ammonium sulphate Sodium thiosulphate Sodium chlorate

(orthorhombicl (monoclinic) (regular)
Figure 3-2. Characteristic crystal forms. Miller indices of the faces are
given. (Mullin, 1972). Reprinted with permission of Butter-
worths. Copyright 1972.
83
pIe inorganic compounds belong to either the regular or hexagonal systems.

Some characteristic crystal forms are shown in Figure 3-2 (Mullin, 1972).
Two or more substances which crystallize in almost identical forms are

isomorphous. There will be small but measurable differences in the angles
(Haiiy's law says the angles are unique for each substance). Since the crystal
shapes are so similar, these substances often form solid solutions. Isomorphs
can also grow on top of crystals of the other substance.
Enantiomorphs are crystals of the same substance which are mirror

images of each other. Compounds which are enantiomorphs are often optically
active. These compounds are often of considerable interest in biological sys-
tems where one form may be biologically active while the other form is not.
Specific additives can be very useful in controlling both crystal morphol-

ogy and purity (Davey, 1987; Randolph and Larson, 1988). Since impurities
can be incorporated only through certain faces of the crystal, crystal purity is
linked to crystal morphology. For example, optically active isomers will form
pure D and pure L crystals, but separating the two types of crystals is very
difficult. Suppose a small quantity of a resolved additive, say a D isomer,
which inhibits crystallization is added. Then the D crystals will not grow and
only the L crystals will form (Davey, 1987). Similiar separations based on the
solid state chemistry can be developed.
In industrial crystallization the growth conditions are usually far from

ideal. Thus, the individual crystals are also usually far from ideal. Only one-
half or one-quarter of the maximum number of faces may develop. Combina-
tions of different classes often occur. Crystals also often grow as composites
formed by aggregation or intergrowth. Most crystals will have imperfections
which may change the shape. Crystallization past phase transitions usually
damages the crystals. Finally, when crystallization is very rapid dendrites (see
Figure 3-1d) will often form. An example of dendrites is frost on windows.
All of these variations from ideal conspire to make the shapes of industrially
produced crystals considerably more complex than the ideal forms shown in
Figure 3-2.
84
3.2. NUCLEATION
The steps in crystallization are:
1. Supersaturate or supercool.
2. Fonn crystal nuclei.
3. Grow crystals.
Nucleation is the least well understood of these. Part of the difficulty in under-
standing nucleation is there are several types of nucleation. In primary nuclea-
tion no crystals are involved in the nucleation. Homogeneous primary nuclea-
tion occurs in the bulk liquid phase with no solid surfaces present Heterogene-
ous primary nucleation occurs on a solid such as a dust particle or the vessel
wall. Secondary nucleation is heterogeneous nucleation induced by existing
crystals.
3.2.1. Homogeneous Nucleation
Homogeneous nucleation occurs in the absence of crystals or any foreign parti-

cles. Since irregularities in the vessel wall or any crystals left over from previ-
ous crystallizations can serve as nucleation sites, the vessel must be meticu-
lously cleaned and polished. In addition, the vessel must be closed since nor-
mal atmospheric dust will often cause heterogeneous nucleation.
The current picture of homogeneous nucleation is as follows: In a highly

supersaturated solution a large number of solute units (atoms, molecules or
ions) can aggregate together to fonn a cluster or embryo. Bringing together the
solute units and expelling the solvent molecules requires work. This work or
energy of nucleation serves as an energy barrier which controls the nucleation
kinetics. If the cluster can reach a certain critical size, the free energy change
upon further growth will be negative. Then the cluster is stable and serves as a
nucleus for further growth. If the cluster cannot reach the critical size, it redis-
solves.
85
The solubility of small particles can be related to the equilibrium solubil-

ity by the Oswald-Freundlich or Kelvin equation
(3-1)
where c is the solubility of the small particle, (j is the interfacial tension

between solid and fluid, and r is the radius of the particle. Equation (3-1)
predicts that c > c· for small particles and that solubility increases as the parti-
cle radius decreases. Thus, there is a tendency for the bigger particles to grow
and the smallest ones to redissolve. This occurs because the system adjusts
=
itself to a minimum total surface free energy. Since etc· S, the supersatura-
tion ratio, Eq. (3-1) indicates that very small particles require a certain degree
of supersaturation or they will redissolve. This is one limit on the critical
radius to form a nucleus.
Since there is an energy barrier to the formation of a nucleus, Arrhenius
type kinetic equations for the rate ~ of nucleation can be developed (Garside,
1985; McCabe and Smith, 1976; Mullin, 1972; Randolph and Larson, 1988).
The rate of formation of nuclei, BO, defined as the number of nuclei
formed/(time volume) is
Bgomogen = dN
dt
I nuclei =A exp(-.~G/kT)
(3-2)
where N is the cumulative number of crystals per unit volume which in this
equation is the number of nuclei per unit volume, L\G is the free energy change
upon formation of nuclei, k is Boltzmann's constant = 1.3805 x 1O-23 J/K, and
the preexponential factor A is on the order of 1025. Starting with Eq. (3-2), the
eventual result is,
(3-3)
where VM is the molar volume of the crystal which is the reciprocal of the
molar density. Equation (3-3) predicts an extremely rapid increase in the
86
nucleation rate after some critical degree of supersaturation S is reached. At

the critical supersaturation ~G switches from a positive to a negative value and
spontaneous nucleation can occur. Experiments agree with this initially, but
then may show a maximum and a decrease in solubility. This equation is
difficult to use since (J, the solid-liquid surface energy, is difficult to measure.
Homogeneous nucleation is well established as a real phenomena which

will occur under well-controlled laboratory conditions. Equations (3-1) and
(3-3) show that a considerable degree of supersatuation is required to have
homogeneous nucleation. When homogeneous nucleation does occur, a very
large number of nuclei may be produced, and the final product may consist of a
very large number of small crystals. In properly operated commercial cooling,
evaporative, and vacuum crystallizers homogeneous nucleation seldom if ever
occurs. Homogeneous nucleation does occur in salting out and precipitation
crystallizations.
3.2.2. Heterogeneous and Secondary Nucleation
In heterogeneous nucleation the crystal forms around a foreign object such a

dust particle or a crack in the vessel wall. The foreign object lowers the sur-
face energy because the solid is wetted by the solvent. The surface energy (J
can drop from around 80 to 100 ergs/cm 2 to as low as 2 to 3 ergs/cm2 • Equa-
tion (3-3) is again applicable but with the absolute value of the argument of the
exponential decreased by as much as 4 to 5 orders of magnitude. This obvi-
ously greatly increases the rate of nucleation at any supersaturation and
decreases the critical supersaturation. The most effective particles will be in
the range of 0.1 to 1.0 J..U1l.
In industrial systems solutions are never completely free of suspended

solids, cleaning of vessels is often less than perfect, and interior surfaces are
often incompletely polished. Thus heterogeneous nucleation will occur at
much lower supersaturations than homogeneous nucleation.
In secondary nucleation the object the crystal builds on is a very small

crystal of the solute. This may be added on purpose as seed crystals to induce
87
nucleation. The seed crystals can be the substance being crystallized or a sub-
stance which is isomorphous to the solid. Seeding is commonly used for pro-
duction of borax from Trona ore, for crystallization of alumina, for production
of sugar, and in citric acid crystallization. Citric acid is an example where
nucleation is difficult if not seeded, but large crystals are easy to grow if
seeded. Seed crystals may also be added accidentially as atmospheric dust or
as crystals retained in tiny cracks from previous batches. When solutions are
seeded, there is often a lag or induction period before secondary nucleation
starts.
The presence of macroscopic growing crystals can also cause secondary
nucleation by several different mechanisms (Nyvlt, 1982; Randolph and Lar-
son, 1988; Strickland-Constable, 1972).
1. Initial breeding. Added seed crystals may have crystal dust on their
surfaces. The crystal dust arises when the crystals are dried and retained
mother liquor is dried onto the crystal forming a dust. When added to the
crystallizer, the dust can wash off and serve as additional nuclei if the
dust particles are larger than the critical radius. This can be an important
nucleation mechanism in batch crystallizers.
2. Needle or dendrite breeding. Very rapid crystallization will cause the

growth of needles and/or dendrites. Small crystals are easily broken off
which will serve as nuclei.
3. Polymerization breeding. Crystals will often grow as agglomerates.

These can break apart easily leading to nuclei.
4. Contact or collision breeding. The agitator or other parts of the vessel

can knock off very small particles from a crystal. These small pieces
serve as nuclei. Experiments show that different materials of construc-
tion (e.g. stainless steel versus polyethylene) will cause different nuclea-
tion rates. The rate with stainless steel is significantly higher. The
required contact energy appears to be rather low, but repeated rapid con-
tacts do not produce more nuclei. This is probably the most important
88
mechanism for secondary nucleation in continuous crystallizers (Garside,

1985; Rousseau, 1980).
5. Fluid shear breeding. Fluid shear by itself can apparently cause

nucleation, but this does not appear to be an important mechanism.
In all cases the nuclei will grow only if they are larger than the critical radius.
Thus secondary nucleation depends on the supersatuation. Since mixing is
never perfect, the vessel will have higher supersaturation in some localities.
Nucleation is most likely to occur in these places such as at cold walls or at the
liquid surface when evaporation is used. Equation (3-3) is valid locally;
although, for design it is often used for the entire volume. It is possible to have
nuclei formed in one part of the vessel and then redissolve in other parts if the
supersaturations are very different. This will not happen in a well-designed
crystallizer.
Equation (3-3) can be derived from fundamental principles, but is only

accurate within an order of magnitude. For design, empirical and semi-
empirical equations are used. For heterogeneous and secondary nucleation the
effective or apparent nucleation rate can often be correlated using the semi-
empirical equation.
(3-4)
where (.1c)max is the maximum allowed supersatuation and i is the "order" of

nucleation. Note that i is applied to a concentration difference and has no fun-
damental significance. For aqueous solutions the "order" lies between 2 and 9.
Low molecular weight salts are at the lower end of this range. For secondary
nucleation Eq. (3-4) again appears to be valid, but the order is significantly
lower and is in the range from 0 to 3. Many commercial crystallizers appear to
operate with orders in this range.
The nucleation rate is often correlated in terms of the linear growth rate
of the crystal, G. This eliminates the need to know the system supersaturation
which is often difficult to measure. G is defined as
dL (3-5)
G=-
dt
89
where L is a characteristic dimension of the crystal. Then BO is
(3-6)
In this equation N is the number of nuclei per unit volume and n° (given by the
term in parenthesis) is the population density of nuclei. Usually, the linear
growth rate is proportional to the supersaturation, G = leo (~c), while BO is
related by Eq. (3-4). Then equations
(3-7)
will correlate the data. A more general expression has been used by Garside
and Shah (1980) in their review of experimental data on nucleation. This
expression is
(3-8a)
BO = KR(T, hydrodynamics, impurities) CMrY Gi
where MT is the magma density in grams of crystals per liter and i,j are empiri-
cal constants. The dependence on magma density is expected when secondary
nucleation occurs. The exponent i can range from -3.5 to +5 although most
values are positive. The hydrodynamic term is sometimes separated out from
KR • Then
(3-8b)
KR = K'R (T, impurities) ~
where N is the stirrer speed in rpm and k is an empirical constant. The expres-
sions and terms in this paragraph will be extensively used later when popula-
tion balances and crystal size distributions are discussed. If this paragraph is
not clear don't panic. Wait until after population balances and CSD have been
covered.
Some of the correlations based on experimental data for nucleation are

given in Table 3-1 (Garside and Shah, 1980). Note that the "constant" i for the
same crystal often shows considerable variation between investigators. The
negative i values are suspect since it is difficult to postulate a mechanism to
1.0
o
Table 3-1. Nucleation Kinetics from Aqueous Solution. Extracted from Garside and Shah (1980).
System Temp. Range RangeG Range (c-c*) Eq. for BO

Range MT (m/s x 108) (g°hyd/gHzO) (#/L s)
(0C) (gil) x 103
Cooling and evaporation systems
Ammonium sulphate 22 30-75 4.3-12.2 2.09 x101ZMTG1. 5

Ammonium sulphate 34 6. 14xlO-llN7.84M~·98G1.ZZ
Potassium chloride 12-30 14-25 11.7-20.7 2.86xlO z1 G 2.55 at21°C

Potassium chloride 32 60-250 2-12 7 .17x 1039M~14G4.99
Potassium sulphate 29 5-28 0.3-2.5 0.6-6.1 4.07xlO-5M~4G-l
Potassium sulphate 30 23-108 1.2-9.5 2.8-7.1 1.67x103MT Go
Potassium sulphate -30 1-7 1.7-11.8 2.62xl03N2.5M~.5G°.54
Citric Acid 16-24 1.1-3.7 30-130 1.09x1010M~g4GO.84 at 20°C
Salting Out Systems
Ammonium 27 110 1.4-3.3 5.74x 1037 G4.0

Sulphate/MeOH
Sodium chloride/ 27 33 1.7-2.8 2.02x1075 G 9.o

Ethanol 3.0-5.8 4.54x10 14 G1.72
" -13 2.3-3.9 32.5-53.5 1.29x1065 G7.9
\0
92
give negative i. This variation in i and in the absolute value of BO (which can
vary by several orders of magnitude) are an indication of the difficulty in repro-
ducibly measuring nucleation rates. The values of i for salting out tend to be
higher since nucleation is probably primary. The references for each system
are given by Garside and Shah (1980). Nyvlt (1982) reports other values and
Tavare (1987) reviews the literature.
At this point it may appear that the highest possible nucleation rate would
be desired. This is not the case. If a huge number of nuclei are formed, then
the crystals cannot grow very large. (Remember 109 particles 0.1 mm in diam-
eter have the same mass as 106 particles 1 mm in diameter). Since relatively
large crystals are usually desired, nucleation must often be supressed. For
example, urea nucleates readily, but it is difficult to grow nice crystals. When
seed crystals are added, it may be desirable to have a zero nucleation rate so
that each seed crystal will become as large as possible.
3.3. CRYSTAL GROWTH
Crystal growth is a complex subject which like nucleation is not completely

understood. On a microscopic level growth rates are irregular; however, aver-
aged over a large number of crystals on a macroscopic scale growth rates are
regular. Mass transfer rate limitations are usually used for design, but mass
transfer theory cannot explain all phenomena. Thus, it will be useful to first
explore another theory of crystal growth before discussing mass transfer.
3.3.1. Adsorption Theory
Adsorption layer or kinetic theories have proven to be quite fruitful in explain-

ing crystal growth. (Garside, 1985; Ohara and Reid, 1973; Moyers and
Rousseau, 1987; Mullin, 1972; Randolph and Larson, 1988; Tavare, 1987).
These theories postulate that there is an "adsorbed" layer of units (atoms, ions
or molecules) which is loosely held to the crystal face. These adsorbed units
are free to move on the two-dimensional surface, but they have essentially lost
93
one degree of freedom. If the surface is perfect and has no kinks, dislocations
or crystallized units, the unit must crystallize at an edge or as a monolayer
island nucleus. This two-dimensional nucleation requires a considerable super-
satuation, but the energy required is considerably less than for a three-
dimensional nucleation. The approximate ratio of energies is (Mullin, 1972)
Energy two-dimensional nucleus __1_ S = 1.1

Energy three-dimensional nucleus 50 '
1
- - S= 10
1.2'
At lower supersaturations the energy required for this two-dimensional nuclea-

tion is considerably less than for normal nucleation. Thus, existing crystals can
grow under conditions where three-dimensional nucleation will not occur.
Once the initial two-dimensional nucleation has occurred, growth is easy since
the units can fill in the face. Since each unit will fit with other units on at least
two sides, the energy for growth is significantly less than that for two-
dimensional nucleation.
Crystal growth will usually occur at supersaturation levels lower than can
be explained by the two-dimensional nucleation theory. This happens because
growth is much faster at any imperfection (kinks, pits, or dislocations) where
the surface is not a perfect plane. Since crystals are usually not perfect, growth
usually proceeds by a "filling-in" process. With linear imperfections the crystal
will heal rapidly, and another two-dimensional nucleation would be required.
Two-dimensional nucleation is not required if the crystal grows by a screw-
dislocation. Crystal growth with a screw-dislocation mechanism is illustrated
in Figure 3-3. Now the crystal can grow without healing the dislocation, and
additional two-dimensional nucleations are not required. Equations resulting
from this and similiar theories are discussed elsewhere (Garside, 1985; Moyers
and Rousseau, 1987; Ohara and Reid, 1973; Randolph and Larson, 1988).
The adsorption theory helps to explain several experimental observa-

tions. First, rates of dissolution of crystals are invariably faster than rates of
crystallization for equivalent differences from equilibrium saturation. Dissolu-
tion does not require two-dimensional nucleation. Since this is the slowest step
94
Figure 3-3. Growth by screw dislocation.
in growth, dissolution would be expected to proceed at a higher rate. Second,

small crystals often have linear growth rates, G in Eq. (3-5), which are
significantly lower than large crystals. Small crystals are more likely to be per-
fect than large crystals. Thus the small crystals are much more likely to require
two-dimensional nucleation. Large crystals are less likely to be perfect, and are
more likely to be damaged by hitting the impeller or baffles. Thus, the large
crystals can grow by healing kinks, pits and dislocations. Third, two crystals of
the same size may have different growth rates although all conditions appear to
be the same. This is called growth rate dispersion (Garside, 1985; Rousseau,
1980; Tavare, 1987). If one of the crystals happens to be more perfect than the
other it will have a lower growth rate.
The adsorption theory is often incorporated into mass transfer theories.

The adsorption theory explains the crystallization once a unit is at the surface.
Mass transfer explains movement of the unit to the surface.
_--+-'7'"/ --- c (bulk cone)

Crystal
Apparent
;1<-----,-- Ci
S'lB',"""",. (""i'b,',.
) I- Hypothetical _I
film - r
concentration)
fA". thickness 8
Adsorbed layer
Figure 3-4. Schematic of mass transfer process.
95
3.3.2. Mass Transfer Theory
Mass transfer or diffusion theory is similar to mass transfer theories for other
processes. However, in crystallization the picture of the process is somewhat
different and different empirical expressions are used. The mass transfer pro-
cess is shown schematically in Figure 3-4. The steps in the process are
1. Bulk diffusion of solvated units through the film.
2. Diffusion of solvated units in the adsorbed layer.
3. Partial or total desolvation.
4. Surface diffusion of units to growth site.
5. Incorporation into lattice.
6. Counter diffusion of solvent through the adsorbed layer.
7. Counter diffusion of solvent through the film.
Since it is difficult to include all these steps in a theory, usually a simplified

model is used which includes step 1, combines steps 2 to 5, and ignores steps 6
and 7.
The mass transfer across the film is,
(3-9)
where m is the mass of solute transferred across the film and thus m is also the
mass of solid deposited. The film mass transfer coefficient is often written as
kf = % where 0 is the unknown film thickness. Next, steps 2 to 5 are com-
bined as a single surface "reaction" or rearrangement of "order" I1r with rate
constant k r •
dm • n, (3-10)
- =lr A(c· -c)
dt Ar 1
Realize that this is an empirical equation and does not correspond to steps 2 to
5 in detail.
96
In general, it is very difficult or impossible to detennine the interfacial

concentration Ci' If the surface reaction is of first order, nr = 1, Eqs. (3-9) and
(3-10) can be combined to remove Ci,
For crystals where the faces are growing at the same rate, the mass and area of
the crystal can be written as
(3-12)
where kv and kA are shape factors (see Eqs. (4-20». Note that Eqs. (3-12) are
not valid for crystals such as needles or plates where faces grow at different
rates. Substituting Eqs. (3-12) into Eq. (3-11), we obtain
_dL_ = (_k_A_) ~(c_-_c:-,:*),-- (3-13)

dt 3pskv llkf + llkr
This simplifies to
G= dL =k(c-c*)=k1c (3-14a)
dt
where G is the linear growth rate and k is a combined rate constant. For a con-
stant supersaturation ~c, this is
(3-14b)
M-=G~t
where M- is the increase in linear size of the crystal. When Eq. (3-14b) is
valid, the linear growth rate does not depend on crystal size. This is the
McCabe ~ L law (McCabe, 1929; McCabe et al.• 1985; Mullin, 1972; Ran-
dolph and Larson, 1988).
The McCabe M- law is an important simplification which will be used in

the next section. Because of the assumptions involved in the derivation
97
(surface "reaction" is first order, all faces grow at same rate, constant
coefficients, constant supersaturation in all parts of the crystallizer), it should
not be surprising that many systems do not satisfy this law. Examples are the
crystallization of potassium alum and of potassium sulfate. When the McCabe
& law is not valid, growth is size dependent. An empirical expression known
as the Abegg-Stevens-Larson (ASL) equation (Abegg et ai., 1%8)
(3-15)
is often used. In Eq. (3-15) b < 1 and yare constants, and G nuc1ei is the growth
rate of the nuclei. This result can often be simplified by letting y= (Gnuc1ei'tfl •
This reduces the number of parameters and simplifies determination of the
crystal size distribution. Other models for size dependent growth are summar-
ized by White et ai., (1976).
When nr is not a simple integer, it is not possible to solve explicity for

the growth rate. In these cases an empirical approach is to write the rate as
dm • n (3-16)
- =KoA(c-c)
dt
where n is the "overall growth-rate order." For many inorganic salts n is in the
range 1.5 to 2 (Mullin, 1972). Equation (3-16) allows for correlation of data,
but it has no fundamental significance when n:t; 1.
Temperature will affect both diffusion and the crystal rearrangement pro-
cess. An Arrhenius type equation would be expected for both Ko and k,.. Thus
plots of log KG versus Iff and log ley versus Iff will give straight lines. If
either mass transfer or surface reaction controls, then the Arrhenius equation
will be valid for the linear growth rate. In these cases
dE 1 1 (3-17)
G = G(To)exp[--( - - - ) ]
R T To
When both mass transfer and surface reaction are important, Eq. (3-17) is not
valid. Arrhenius plots for crystal growth data will thus usually give curved
instead of straight lines.
98
Dissolution appears to be a much simpler process than crystallization.

The overall dissolution rate is usually first order with respect to the concentra-
tion difference,
(3-18)
Dissolution is usually significantly faster than crystallization (5 times quicker is

not unusual). This probably occurs because two-dimensional nucleation and
rearrangement to fonn the crystal lattice are not required.
Growth rates in the range of 1O~ to 10-7 mls make it easy to grow crys-
tals in the mm size range. Growth rates in the range of 10-9 mls can take many
days to reach this size range. According to Eq. (3-14a), G can be increased by
increasing supersaturation. Unfortunately, nucleation is extremely sensitive to
supersaturation, Eq. (3-4). If excessive nucleation occurs, large crystals cannot
be grown unless the nuclei are destroyed (see the next chapter). In practice, it
may be difficult to grow large crystals of materials with low growth rates with
reasonable residence times.
3.4. SUMMARY -OBJECTIVES
This brief chapter covered crystal shapes, nucleation and crystal growth. At the
end of this chapter you should be able to:
1. Discuss what is unique about crystal shape and define tenns used.
2. Explain the types of nucleation, discuss when they are likely to occur,
and calculate nucleation rates from empirical correlations.
3. Explain the adsorption layer theory of crystal growth.
4. Explain applications of mass transfer theory as applied to crystallization,

and discuss the approximations used.
5. Be able to derive the McCabe & law and explain the assumptions
involved in this derivation.
99
REFERENCES
Abegg, C.F., J.D. Stevens, and M.A. Larson, "Crystal size distributions in con-
tinuous crytalizers when growth rate is size dependent," AlChE Journal, 14,
118 (1968).
Bunn, C.W., Chemical Crystallography, 2nd ed., Oxford, Clarendon Press,

1961.
Davey, RJ., "Looking into crystal chemistry," The Chem. Engr. (London), No.
443,24 (Dec. 1987).
Garside, J. and M.B. Shah, "Crystallization kinetics from MSMPR crystalliz-

ers," '1nd. Eng. Chem. Process Design Develop., 19," 509 (1980).
Garside, J., "Industrial crystallization from solution," Chem. Eng. Sci., 40, 3
(1985).
McCabe, W.L., "Crystal growth in aqueous solutions," Ind. Eng. Chem., 21, 30
(1929).
McCabe, W.L. and J.C. Smith, Unit Operations of Chemical Engineering. 3rd
ed., McGraw-Hill, New York, 1976, Chapt. 28.
McCabe, W.L., J.e. Smith, and P. Harriott, Unit Operations of Chemical

Engineering, 4th ed., McGraw-Hill, New York, 1985, Chapt. 28.
Moyers, C.G., Jr. and R.W., Rousseau, "Crystallization operations," in R.W.

Rousseau (Ed.), Handbook of Separation Process Technology, Wiley-
Interscience, New York, 1987, Chapter 11.
Mullin, J.W., Crystallization, 2nd ed., Butterworth, London, 1972.
Nyvlt, J., Industrial Crystallization, 2nd ed., Verlag Chemie, Weinheim, 1982.
100
Ohara, M., and R.C. Reid, Modeling Crystal Growth Rates from Solution,
Prentice-Hall, Englewood Cliffs, NJ, 1973.
Randolph, A.D. and M.A. Larson, Theory of Particulate Processes. Analysis

and Techniques of Continuous Crystallization. 2nd ed., Academic Press, New
York, 1988.
Rousseau, RW., "Crystallization: a review of recent developments," Chern.

Tech .. 10. 566 (1980).
Strickland-Constable, RF., "The breeding of crystal nuclei - a review of the

subject," AlChE Symp. Ser .• 68. (121), 1 (1972).
Tavare, N.S., "Batch crystaIlizers: A review," Chern. Eng. Commun .. 61 259

(1987)
White, E.T., L.L. Bendig, and M.A. Larson, "The effect of size on the growth
rate of potassium sulfate crystals," AlChE Symp. Ser.• 72. (153) 41 (1976).
Wyckoff, RW.G., Crystal Structures. (5 vols.), 2nd ed., Interscience, New

Yorlc,1963-1968.
HOMEWORK
AI. In any field it is important to understand the jargon. Define the following
terms.
a. primary nucleation
b. homogeneous nucleation
c. heterogeneous nucleation
d. secondary nucleation
101
e. two-dimensional nucleation
f. isomorph
g. G
h. BO
1. screw -dislocation
j. McCabe~law
A2. Why is empirical Eq. (3-15) often used instead of the usual mass transfer
Eq. (3-9)?
A3. Explain how different processes can cause secondary nucleation.
A4. Explain why the Arrhenius relation is invalid for crystal growth rates
when both mass transfer and surface rearrangement are important.
AS. The "order" for secondary nucleation in Eq. (3-4) is usually less than the
"order" for primary nucleation. What is the significance of this? Which
will have higher ~c critical?
A6. Explain why Eqs. (3-12) are not valid for plates, needles and dendrites.
A7. Why do small crystals often grow slower than large crystals?
A8. Construct your key relations chart for this chapter.
C. Derivations
C1. Derive Eq. (3-13) from Eqs. (3-11) and (3-12).
D. Problems
D 1. Assuming that G oc ~c, detennine if the Arrhenius equation is valid for

(Nl4hS04 crystallization. If it is valid, detennine ~. Note: c*
depends on T.
102
Data (Mullin 1972): 30°C, S = LOS, G =5.0 x 10-7 mls

60°C, S = LOS, G = 8.0 x 10-7 mls
90°C, S = 1.01, G =6.0 x 10-8 mls
D2. Is G & for sucrose crystallization? If the relationship is linear, deter-

DC
mine the proportionality constant
Data (Mullin, 1972): 30°C, S = 1.13, G =2.2x 10-8 mls

30°C, S = 1.27, G =4.2 x 10-8 mls
70°C, S = 1.09, G = 1.9 x 10-8 mls
70°C, S = LIS, G =3.0x 10-7 m/s
D3. Estimate the nucleation rate for K 2 S04 at 30°C at a supersaturation ratio
S = 1.07. MT is 25 gil.
a. Use the equations in Table 3-1.
b. Assume homogeneous nucleation and use Eq. (3-3). Estimate cr as

80 ergs/cm 2 . The density of K2 S04 crystals is 2.66 g/cm 3 •
c. Compare parts a and b.
Data: For S = c/c * = 1.07, G = 8.4 x 10-8 m/s and growth was size
dependent (Mullin, 1972).
F. Problems Requiring Other Resources
F1. Read the classical paper by McCabe (1929) where the McCabe &.., law
was first developed. Write a report on this short paper. Compare the
development of the &.., law in McCabe's paper to the development in
this chapter.
Chapter 4
POPULATION BALANCES AND CRYSTAL SIZE DISTRIBUTIONS
Population balances will be introduced and will be used to derive crystal size
distributions (CSD) for some simple growth kinetics. The crystal size distribu-
tion tells the number of crystals at each size range. This is important for pro-
duct quality control and for ease of downstream processing. The CSD will be
used to predict physically measurable quantities such as the sieve analysis.
Methods for using an experimentally measured CSD to infer the nucleation and
growth rates will be developed. The effect of equipment modifications on the
CSD and more details of equipment for crystallization from solution will be
discussed. Finally. the effect of the CSD on downstream processing will be
briefly explored.
Population balances are the major theoretical tool used for predicting and
analyzing crystal size distributions. The development of population balances in
the 1960's was a major breakthrough which helped move crystallization from
an art to more of a science. Population balances and CSD have been reviewed
in many books and review papers (Garside. 1985; Larson and Randolph, 1969;
McCabe et al., 1985; Mullin, 1972; Randolph and Larson, 1988; Rousseau,
1986). The most detailed source is Randolph and Larson (1988) while Garside
(1985) has an extensive bibliography. Ramkrishna (1985) explores the general
mathematical framework and other applications of population balances in con-
siderable detail.
4.1. BASIC POPULATION BALANCE EQUATIONS
The population balance is a balance of the number of crystals within a given

size range. The idea of the concept is easy to see from an analogy. Suppose
103
104
we wish to determine the number of children in a school who are between 100
cm and 110 cm tall for a two month period. This number can be determined by
a population balance on children in this height range. This balance can be writ-
ten for the two month time period,
(4-1)
Initial No. children in range + No. children growing into range
+ No. children entering school in range = final No. children in range
+ No. children growing out of range + No. children in range leaving school
The population balance differs from a mass balance in that only the item bal-
anced (children in this example) within a given size range (100 to 110 cm) are
included. In addition, the item (children or crystals) can grow into or grow out
of the size range.
The population balance for crystallization can be obtained by changing

the words (No. of children) with (No. of crystals), and replacing "school" with
"crystallizer." In addition, we will assume there is no crystal breakage. The
size range is now for crystals within a range L2 - Ll = M... The size range is
usually determined by a sieve or screen analysis.
Population balances for crystallizers are usually written in terms of the

population density, n, instead of the number of crystals. This population den-
sity is defined as,
(4-2)
L is a measure of the linear size of the crystal, and N is the cumulative number
of crystals per unit volume at the given size L. The units on n are number of
crystals/m-m3 • The population density is shown schematically in Figure 4-1.
The population density can be related to a count from a screen analysis, particle
size analyzer, or count under a microscope by integrating,
L2
J ndL = No. of crystals in sizes from Ll to LJvolume (4-3)
L}
105
L
Figure 4-1. Schematic representation of JX.>pulation density.
In terms of the population density, Eq. (4-1) for crystals can be written as
(terms are in the same order)
(4-4)
Om is the volumetric feed rate while Qout is the volumetric discharge rate, m3/s.
M. = Lz - L1 is the size range. 0 is the average population density over the
size range. n1 and n2 are shorthand for the average population densities over
the range of sizes which will grow into and out of the range M., respectively.
G 1 and G z are the average growth rates of crystals growing into and out of the
range M.. V = volume of magma (mixture of crystals and solution) which is
assumed to be constant. The magma volume is assumed to be well-mixed so
that n and ~c are constant throughout the volume.
Rearranging Eq. (4-4) gives
Taking the limit as M.~O and as .1t~O
d(Gn) - ) + V-do (4-6)

V- - + (QoutnOilt
- - Qinnin =0
dL dt
106
This equation is the unsteady state population balance. In continuous operation

the assumption of steady-state operation is usually made. Then the population
balance simplifies to
d(Gn) _ (4-7)
V~ + Q(llout - nin) = 0
where we have also assumed that the volumetric feed and discharge rates are
equal. This implies that the density (kg/m 3 ) is constant.
Solution of Eqs. (4-6) or (4-7) gives an equation for the population den-
sity distribution. From the population density distribution a variety of other
distributions (number, length, area and mass) are easily generated. These crys-
tal size distributions are developed in the next section.
4.2. DISTRIBUTIONS FOR SIZE INDEPENDENT GROWTH
In this section the CSD will be developed for steady-state crystallizers where
the crystal growth rate does not depend on crystal size. This will be done for
an idealized crystallizer known as the Mixed Suspension, Mixed Product Remo-
val (MSMPR) crystallizer for the simplest possible case. The following
assumptions will be made for the ideal MSMPR:
Steady-state
No product classification - crystals removed have same CSD as
those in magma.
Constant magma volume.
Uniform supersaturation, .1c.
No crystal breakage.
Product crystals in equilibrium with mother liquor - no change in
CSD after withdrawal and Dout = n.
=
No crystals in feed, Din O.
Growth is independent of size. Thus G = constant since .1c is con-
stant.
Many of these assumptions will be relaxed in later sections.

107
The steady-state population balance, Eq. (4-7) can now be simplified.

The draw down or retention time is -c = V/Q. Since the linear growth rate Gis
constant, Eq. (4-7) simplifies to
(4-8)
~+~=O
dL G1:
which upon integration is
(4-9)
n = nO exp(-l::...)
G1:
where n° is the population density of nuclei.
Equation (4-9) is the result we have been working towards. Taking the
natural logarithm of both sides
(4-10)
In n = - l::... + In nO
G1:
A plot of In n versus L is a straight line with a slope of - l/(G-c) and an inter-

cept of In nO. This is illustrated in Figure 4-2. If n(L) is determined experi-
mentally, nO and G can be determined from this plot. This will be illustrated in
Example 4-2.
If G and nO are known, N and other distributions can be obtained starting

with Eq. (4-9). The physical significance of these other distributions is easier
to see than the physical significance of the population density. The first distri-
bution is the cumulative number of particles at a given size L. From Eq. (4-2)
and (4-9),
L
J J
L L
(4-11a)
N= ndL= nO exp( - -)dL
o 0 G-c
which is
L (4-11b)
N = nOG-c[1 - exp( - - ) ]
G-c
108
o
18
o
In n
16
I
- Gr
14
o 0.04 0.08 0.12

L
Figure 4-2. Plot of population density following Eq. (4-10). Numerical

values are for Example 4-2.
This distribution was illustrated in Figure 4-1. As L~, the exponential term
becomes zero and
(4-12)
where NT is the total number of particles per unit volume. Note that with the
assumptions made for the ideal MSMPR each nucleus grows into one crystal in
the product
The cumulative length per unit volume is the length of crystals per unit
volume if all crystals up to size L were laid end-to-end. Within a given size
109
range, the end-to-end length per volume is LndL. Then the cumulative length
per volume is
L L L
= JLndL = JLnO exp( -
(4-13a)
L G1:)dL
o 0
which is
L =n'G+'+ -exp( - ~ )]-LexP(- ~t)} (4-13b)
As L4°o the total cwnulative length per unit volume of all the crystals is
(4-14)
The cumulative area of crystals per unit volume, A, is of more interest.

The cumulative area gives an idea of the area available for growth. The total
area also gives an idea of the area on which mother liquor and impurities can
adsorb. The cumulative area per unit volume can be calculated as
(4-15a)
where kA is a shape factor [see Eq. (4-20b)]. This becomes,
(4-15b)
The total cumulative area per unit volume is found as L400.
(4-16)
110
The cumulative mass of crystals per unit volume, M, is obviously of

interest since it gives the weight of crystals up to a given size. The total mass
per unit volume will give the concentration of crystals in the product. The
cumulative mass per unit volume is
L L
M = kvPc Jo L 3ndL = kvPc JL 3n exp( -
0
o ~)dL
~
(4-17a)
where kv is the volumetric shape factor [Eq. (4-20a)] and Pc is the density of
the crystals. This becomes
M~ 6kvp,n G<{(Gt)'
O
(4-17b)
- [(Gt)' + (Gt)'L + ~ GtL' + !

L' ]exP( - ;)}
The total mass of crystals per volume is determined as L~oo.
(4-18)
This total cumulative mass of crystals per unit volume can also be determined
from a mass balance. For a steady state process
(4-19)
where Sc is the mass of crystals produced per time. Sc can be detennined from
steady-state equilibrium yield calculations such as Eqs. (2-5a) to (2-6b). Thus
Eqs. (4-18) and (4-19) serve as a check on the population distribution or on the
screen data. If nO is known, comparison of Eqs. (4-18) and the yield calcula-
tion allows detennination of G.
The shape factors kv and kA are detennined by comparing the volume or

surface area of the crystals to the equations
(4-20a)
Volume = kvL3
(4-20b)
111
where L is a length parameter. For cubes L =length of side, ky = 1.0 and kA =

6.0. For spheres L = diameter, k y = Tt/6 and kA = Tt. For octahedrons L =
length of a side, ky = 12(3 and kA = 2..[3. The shape factors are readily calcu-
lated for other regular geometric crystals. For platelets and needles the shape
factors must be determined experimentally.
The distributions developed here are related to a mathematical method

called moment analysis. The zeroth moment is
L
JndL
N 0 L (4-21)
-NT = - - = 1- exp( - - )
JndL
DO Gt
o
The first moment is
L
f LndL
L 0 L L (4-22)
-L-
T
- -- = 1- (1 + -Gt-)exp ( - -Gt-)
f LndL
o
The second moment is
L
fL 2 ndL
_A__ 0 = 1- [1 + _L_ + !_(L_)2]exp( __L_) (4-23)
AT fL2ndL G't 2 Gt Gt
o
The third moment is
JL3ndL
M 0 (4-24)
MT
=
JL3ndL
0
= 1- [l + - L + -I(L 21L3 L
- ) + - ( - ) ]exp( - - )
G't 2 G't 6 Gt G't
112
1.0
0.8
Cumulative
0.6
M
MT 0.4
0.2
0
0 4 6 8 10
L/Gr
Figure 4-3. Dimensionless cumulative and differential mass distributions

from Eqs. (4-24) and (4-25b).
In Eqs. (4-21) to (4-24) the variable L/(G't) is a dimensionless crystal size. G't
is the size of a crystal which has grown for a period equal to the residence time.
Using the dimensionless crystal size, L/(G't), the CSD can be analyzed without
knowing the dependence of G on supersaturation. The ratio MiMT is plotted in
Figure 4-3. Table 4-1 is a short table of values for MiMT. Complete tables of
values of MiMT are given by Randolph and Larson (1988) and Singh (1979).
Obviously, values of MiMT can be easily determined from Eq. (4-24) with a
calculator.
Table 4-1. Values of Cumulative Mass Distribution [Eq. (4-24)]
L/G't MiMT L/G't

o o 5.00 0.735
1.00 0.019 5.51 0.799
1.74 0.099 6.00 0.849
2.00 0.143 6.68 0.900
2.29 0.199 7.00 0.918
2.76 0.299 7.74 0.950
3.00 0.353 8.00 0.958
3.21 0.400 8.50 0.970
3.67 0.500 9.00 0.979
4.00 0.567 9.50 0.985
4.17 0.599 9.99 0.990
4.76 0.700
113
In a sieve analysis a series of sieves with different size openings are

stacked together with the largest openings at the top. A charge of crystals is
added to the top. After a period of shaking, the stack is opened and particles
remaining between each pair of screens are collected. These particles will be
in the size range from Ll to ~ where Ll is the size of the screen opening for
the lower screen and ~ is the size of the opening in the upper screen. The
easiest analysis of these fractions is to weigh each fraction. Thus the basic raw
data is a differential mass distribution. The theoretical differential mass distri-
bution is easily determined by differentiating the cumulative mass distribution.
Differentiating Eq. (4-17b), we obtain
(4-25a)
In terms of the weight fraction. M/MT' in this size range the differential mass
distribution is,
(4-25b)
The dimensionless differential mass distribution is shown in Figure 4-3.
Figure 4-3 shows that the maximum of the differential mass balance
occurs at L/(G't) = 3. This is called the dominant size. The median size for the
cumulative mass distribution is at L/(G't) = 3.67. Thus one half of the particles
have a weight greater than the particles at this size.
Differential area, length and number distributions can also be defined.

The dominant sizes for these distributions are L/G't = 2. 1 and 0, respectively.
The median values for the cumulative distributions can be defined by setting
NAT. L/LT and N/NT equal to 1/2. For example, the median of the cumulative
number distribution is at L/(G't) = 0.693. Thus one half the particles are larger
than L = 0.693 G't.
Use of the distributions will be illustrated in Examples 4-1 and 4-2.

114
4.3. SIEVE ANALYSIS OF CSD
Experimental CSD's are usually detennined by sieve analysis, particle size

analyzer, or counting under a microscope. Since the principles to relate the
theoretical CSD to the experimental CSD are similiar, we will discuss only the
sieve analysis in detail. Mullin (1972) has an exhaustive analysis of sieving.
We would like to convert M into a predicted sieve analysis. Usually standard
sieves are used. An abbreviated table of sieve sizes is given in Table 4-2 (perry
and Green, 1984). We will first discuss the sieve analysis of experimental data
and then prediction of sieve analysis from the weight distribution. For on-line
measurements electronic zone-sieving devices (e.g. Coulter Counters) and
Iaser-light-scattering instruments (e.g. from Leeds & Nothrup) are used.
Often G and n° are not available in the literature. In this case laboratory
Table 4-2. Standard Sieve Sizes
Sieve Opening Tyler Sieve Opening Tyler

(mm) mesh (microns) mesh
8.0 21/2 841 20

5.66 31/2 707 24
4.76 4 595 28
4.00 5 500 32
3.36 6 420 35
2.83 7 354 42
2.38 8 297 48
2.00 9 250 60
1.68 10 210 65
1.41 12 177 80
1.19 14 125 115
1.00 16 88 170
63 250
44 325
37 400
Source: Perry and Green (1984)

115
or larger scale experiments can be done to detennine the kinetics of the pro-
cess. This can be done by screening the crystals and experimentally detennin-
ing a weight distribution. The data collected is:
Wi = weight of crystals collected in sieve fraction i.

&i = size range of sieve fraction i.
= larger sieve opening - smaller sieve opening (see Table
4-2).
L =average size of sieve fraction i (Table 4-2).

= 1(1 (larger sieve opening + smaller sieve opening).
leol = Period in which magma was sieved was collected.
or Vs = Volume of magma sieved
't =(VIQ)crysu.llizcr =crystallizer retention time.
From the sieve data MT is easily detennined from the total weight of crystals
collected as
(4-26)
where Qtcol is equal to the volume of magma which was sieved.
The values of nO and G can now be found two different ways. One
method is to generate an experimental cumulative mass distribution. The data
of Wi versus ~ is a differential mass distribution. The experimental cumula-
tive mass distribution is
sieve frac.i
(4-27)
Mexp,i = L (Wi)Ns
mctim under smallest size
Then Mexp.i or Mexp.dMT.exp can be plotted versus ~.

116
The theoretical value of MlMT versus (L/Gt) is known from Eq. (4-24),
Figure 4-3 or Table 4-1. By comparing experimental and theoretical values, we
can determine G and n°. For example, suppose one of the data points
corresponds to (MexpIMT,exp)i =0.4. Then from Table 4-1, 4/Gt =3.21. Since
~ and 't are known, this allows us to calculate G. If <MexpIMT,exp)i is a number
such as 0.442, the complete tabulations of M/MT (Randolph and Larson, 1988;
Singh 1979) can be used, or program Eq. (4-24) on your calculator. This pro-
cedure can be repeated for each sieve fraction. Then, average G according to
the weight fraction of particles collected in each fraction.
- L(GiWJ
G=---- (4-28)
I: Wi
all fractions
Equation (4-28) assumes growth is size independent. This is not an additional

assumption since it was already made to determine the population density func-
tion. The population density of nucleii can now be found from Eq. (4-18).
Thus
(4-29)
The second approach is to calculate experimental values of the popula-

tion density for each sieve fraction. G and n° can then be determined from Fig-
ure 4-2. To do this we first calculate MT,cxp from Eq. (4-26). Then for each
sieve fraction i
(4-30)
To generate Figure 4-2, In ni is plotted versus 4

for each fraction. A regres-
sion analysis will then give the best values of nO and G.
Obviously, the two approaches should closely agree with each other.
117
Once nO and G are known the nucleation rate BO is readily calculated from Eq.
(3-6), BO = nOG. To determine the functional dependence of BO as shown in
Table 3-1 a series of experiments with different residence times and different
stirrer speeds is done.
If G and equilibrium are known, the sieve analysis can be predicted. To

predict the weight distribution from the theoretical mass distributions, we can
first rearrange Eq. (4-26) to predict the total weight of crystals collected.
WT = ~ Wi =MTV s (4-31)
all fractioos
The weight Wi in a given sieve fraction is related to the differential mass bal-
ance, Eq. (4-25a) and (4-25b)
(4-32)
The weight Wi can be calculated for each sieve fraction which will be col-
lected. This is illustrated in Example 4-1. Note that we can calculate Wi
without knowing the shape factor, kv. This is useful since kv will be unknown
if the crystals are not ideal.
Example 4-1. CSD for Size Independent Growth.
Potassium chloride is being crystallized in a cooling crystallizer. The

crystallizer approximates an MSMPR. Feed is initially saturated at
100°C and the crystallizer operates at 20°C. Feed is 10,000 kg/In of
total solution. The crystallizer has a volume of 2.0 m3 . Assume that the
growth rate is 4.0 x 10-7 mls. Predict the total mass distribution, M, as a
function of crystal size L. If 0.1 m3 of magma is sieved predict the
weight of crystals collected between 24 and 28 mesh. Assume the
system is operating in a high-yield mode where ~c- 0 (equilibrium) in
the outlet liquid.
118
Solution
A. Define. The system is sketched in the figure.
Cool
Feed
IO,OOOkg/hr -
100°C
Saturated KCI
Plot M vs L and find Wi collected between 24 and 28 mesh.
B. Explore. Growth rate for this system is size independent (Mullin,

1972). The solution of the steady-state population balance Eq. (4-7) is
Eq. (4-9), and the total mass distribution is given by Eq. (4-17b). Thus
the theory developed in this section can be used.
C. Plan. External mass balances can be used to determine the crystal

flow rate, Sc. Table 2-1 shows that KCI crystallizes as anhydrous
material. Thus Sc can be found from Eq. (2-5b)
where Win is the water flow rate entering the crystallizer, and em and
c out are given in Table 2-1 at 100 and 20°C, respectively. Win can be
determined from a balance on the inlet stream
F
F=cin Win + Win or Win = -1--
+ Cin
The volumetric feed rate Q =F/Pe and 't =V/Q. Then from Eq. (4-19),
MT = Sc/Q, and M can be calculated from Eq. (4-24) even though nO
and k y are unknown.
To determine the weight collected between 24 and 28 mesh we

will use Eq. (4-32)
119
D. Do It. From Table 2-1, cin = 0.567 kg KCl/kg water, Caut = 0.340 kg
KCl/kg water. Then
W. = F = 1O,OOOkg/hr = 6382 kg water/hr

m 1 + cin 1 + 0.567
and Sc = (Cin - cout)W in = (0.567 - 0.340) (6382) = 1449 kg/hr
The feed density is -1208 kg/m3 (extrapolated using data from Mullin
(1972) or from Perry and Green, (1984».
Then Q = ~ = 10,000 kg/hr = 8.278 m3jhr

PF 1208 kg/m 3
't =:i. = ~ = 0.2416 hr= 870 sec

Q 8.278
Sc 1449 3
and MT = - = - - = 175 Okg/m
Q 8.278 .
M can be determined from Table 4-1 or Figure 4-3 with MT =

175.0 and Gt = (4.0 x 10-7 )(870) = 3.48 x lO-4m. To find M multiply
M/MT by 175. To find value of L (in meters) multiply L/G't by 3.48 x
10-4.
The dominant size of the mass distribution is
3G t = (3) (3.48 x 10-4) = 1.04 x 1O-3m = 1.04 mm
and the median size of the mass distribution is
[1]
3.67 G 't = (3.67)(3.48 x 10-4) = 1.28 x 10-3 m = 1.28 mm
The median size for the number distribution is
(0.693) (G t) = 2.41 x lO-4m = 0.241 mm

120
The differential mass distribution can also be obtained from Figure 4-3
using the same multipliers.
The mass of crystals collected in a sieve analysis can be predicted

using Eq. (4-29). From Table 4-2 a 24 mesh screen has an opening of
0.707 mm while a 28 mesh screen has an opening of 0.595 mm. Then
the average
L = 0.5 (0.707 + 0.595) = 0.65 mm =0.65 x 10-3 m
For this L, L/Gt = 1.871. Since & = 0.707 - 0.595 = 0.112 mm, Eq.
(4-32) becomes
This is a fraction ;~ = (17;·~~.1) = 0.32 of all the crystals collected
[WT is calculated from Eq. (4-31)]. A complete sieve analysis can be

predicted by calculating Wi for all pairs of sieves.
E. Check. Since n° is unknown and the external balances have already

been used, a good check is difficult.
F. Generalization. When the linear growth rate is independent of size,

the dimensionless solution for M shown in Figure 4-3 and Table 4-1 can
be used. Then the engineer's job is to determine MT and Gt so that the
dimensional values can be determined. If MT and Gt are constant, the
distribution will be unchanged even when other variables are changed.
For example, if F = 1000 kg/hr and V = 0.2 m3 , Win, Sc and Q are all
reduced by a factor of 10. However, G, t, and MT remain constant and
121
the distribution is unchanged. This is helpful in scaling crystallizers to

larger sizes. The growth rate for KCI crystals is rapid compared to that
of many other crystals. Thus, the residence time used in this example is
low compared to that required for other crystallizations.
Example 4-2. Kinetics from Experimental CSD
Urea was crystallized in a commercial Swenson DTB crystallizer. For

one run the following data was obtained (Bennett and Van Buren, 1969)
Tyler Screen Sizes Magma

't,hrs 14 20 28 35 48 65 Density, gIL
3.97 19.0 38.5 63.5 81.5 98.0 100 404
The screen numbers are cumulative weight %. Urea, NHzCONH z• has

Pc = 1.33 g/cm3 • The crystals had kv - 1.0. Determine G and nO.
Solution
We will illustrate use of Eq. (4-30).

~ = .1NJ~ where &i = 4 - Li- 1 is easily determined from Table 4-
2. Then from Eq. (4-30)
.1N - (Magma density)(Weight fraction)

i - -3
Pc kv Li
W·
where Weight fraction crystals in size fraction i = -'-.
LWi
The weight fraction can be determined from the cumulative weight frac-
tion as
(Weight fraction)i = (Cum. Wt. fraC.)i - (Cum Wt. frac)i-l

122
Since Magma density, Pc and kv are known,
LW _ 303.8 (Weight fraction)

i - -3
L·1
We can now generate the following Table.
cum. wt.
Mesh L [ M- frae. frae. LWi n lnn
(in 1000)
14 .119 0.19 .19
14/20 .0841 .10155 .0349 0.385 .195 56,562 1,621 14.3
20/28 .0595 .07180 .0246 0.635 .250 205,162 8,340 15.94
28f35 .0420 .05075 .0175 0.815 .180 418,306 23,903 16.99
35/48 .0297 .03585 .0123 0.980 .165 1,087,793 88,438 18.30
48/65 .0210 .02555 .0087 1.00 .020 372,929 42,865 17.57
Now plot In n vs L. This result is shown in Figure 4-2.
-1 -1
G= = (-53.5)(3.97) = 0.0047 cm/hr
(slope) 't
Intercept = 19.8 = In nO, which gives nO = 3.97 x 108
Notes: 1. Value of nO depends very much on how line is drawn.

2. The first mesh size (14) does not give a data point.
Thus it is desireable if cumulative fraction for first
screen is small. Use of a 12 mesh screen in addition to
14 mesh would have given another data point.
3. Obviously, data scatters with small screen openings.
4.4. EFFECT OF NUCLEATION AND GROWTH KINETICS ON CSD

Even when the McCabe M- law is valid, the kinetics of nucleation and growth
can vary enormously. This was shown by Eq. (3-8) and the correlations in
Table 3-1. In this section we will consider how nucleation and growth kinetics
affect the CSD (Larson and Randolph, 1969; Randolph and Larson, 1988).
123
This section will illustrate how the equations can be manipulated to compare
the effects of changing variables. The growth rate will be assumed to be
independent of size. Size dependent growth is treated in Section 4.7.
Consider two cases where secondary nucleation does not occur. Equa-
tions (3-7) are valid and nO = kNGi - l . First, compare two crystallizers where
the suspension densities are equal, but the residence times 'tl :I: 't2' Setting
MT,l = MT ,2 and substituting in Eq. (4-18), we obtain
(4-33a)
Canceling out the common terms and using nO = kNGi-l to remove n1 and n~,
this becomes
... 4G3+i _ ...4G3+i (4-33b)

"I 1 -"2 2
or
G2 _ 'tl 4/(3+i) (4-34a)

--(-)
G1 't2
Substituting nO = kNGi-l back into Eq. (4-34a), we have
o 't 4(i-l)
~=(_._l )3+1 (4-34b)
n1 't2
Clearly, the exponent i has a major effect on the growth and nucleation
rates in the two crystallizers. If i=l, n1 = n~ and G 2 = G1 ('ttft2)' Thus growth
rates depend inversely on the residence time and n° is independent of the
residence time. Since G is proportional to ~c, the supersaturation difference ~c
must also depend inversely on residence time. If residence time is halved, ~c
doubles and G doubles.
If i > 1.0, G and n° increase as residence time increases. Then from Eq.
(3-6), BO= nOG, the nucleation rate increases. When i=2, the exponents on
Eqs. (4-34a and b) are equal. For i > 2 the exponent on Eq. (4-34b) is larger
and nO grows faster than G. Thus as i becomes larger it is more and more
difficult to grow large crystals since too many nuclei are formed. This tends to
be the case for homogeneous and heterogeneous nucleation.
124
The nucleation rate BOincreases when the residence time is increased for
i> 1.0. Then from Eqs. (3-6) and (3-4) the supersaturation l:1c must increase.
Increasing the residence time increases the supersaturation at constant magma
density.
A second case without secondary nucleation compares two crystallizers
with equal residence times, but different feed concentrations and hence dif-
ferent magma densities. Again we start with Eq. (4-18), MT = kvPc nO(Gt)4.
Set 'tl = 't2 equal for the two crystallizers, and cancel out common terms.
Then,
(4-35)
Substitute in the expression n° = kNGi-l and solve first for G2/G 1 ,
(4-36a)
and then solve for n2/nY.
(4-36b)
Growth rates increase as magma density increases, and if i > 1 the

nucleation population density, nO, increases as magma density increases even
though secondary nucleation does not occur. If i = 1, n° is constant while G
and hence BO both increase as magma density increases.
The dominant size of the mass distribution, LD = 3G'tl, as easily deter-
mined from Eq. (4-36a).
L M
~ = (~)1/(i+3) (4-36c)
LD ,I MT,I
Thus sizes are increased at higher magma density if secondary nucleation does
not occur.
125
In n
Figure 4-4. Effect of magma density for size independent growth. 'tl = 't2,
MT,2 = 2MTl , i = 3. Calculated: nz/n? = 1.2599 and G2/G 1 =
1.1225.
The population density distribution (and other distributions) can easily be

determined for both cases since G and nO have been calculated. Substituting
the expressions for G and n° into Eq. (4-9) or (4-10) will give the population
density. The effect of changing MT is illustrated in Figure 4-4.
Secondary nucleation is often important in commercial crystallizers. As
a third case consider two crystallizers with equal residence times, but
MT,l < MT,2' When secondary crystallization occurs, suspended solids are a
source of nuclei. Then if the McCabe & law, Eq. (3-14a), is valid G = ~c.
Equation (3-8) then becomes,
(4-37a)
and Eq. (3-6) gives,
(4-37b)
With equal residence times Eq. (4-35) can again be derived. After substituting
in Eq. (4-37b) for nO, and solving for G2/G 1 , we obtain
(4-38)
126
In n
L
Figure 4-5. Effect of magma density on population density when secon-
dary nucleation occurs and j = 1. Mn > Mn
Experimental results often show BO oc Mr (that is,j =1). Then
GZ=G 1 ,
n2
~=--
Mr.z 0--1) (4-39)
nl M r ,l
Equations (4-39) should be compared to Eqs. (4-36a) and (4-36b). When

secondary nucleation occurs with BO oc Mr 0=1), the growth rate does not
depend on magma density and the nucleation population density depends
linearly on the magma density. This is a greater dependence than shown in Eq.
(4-33b). Neither nO nor G depend on i while in the absence of secondary
nucleation they do. The population density graph with secondary nucleation is
shown in Figure 4-5. Since G 1 = G z , the slopes are equal. Since nucleation
depends on magma density, it is difficult to obtain large crystals with high yield
(which implies high Mr) when secondary nucleation occurs.
4.5. SEEDING
Problems can often be avoided by seeding the crystallizer. With added seeds
the metastable region can be avoided. Thus the crystallizer can operate at
lower supersaturations with no homogeneous nucleation and with small nuclea-
tion rates from other secondary nucleation processes. Sometimes seeding is the
only practical way to grow crystals of the desired size range. If there is no
additional nucleation, the weight of seeds required per hour, Ws is
(4-40)
127
where Wp = weight of product/hour, L. and 4 are the mean particle sizes of

seeds and products, respectively.
When there is no additional nucleation, the final product distribution can
be determined from the CSD of the seed crystals. If growth is size indepen-
dent, M.growth =Gt is the same for all crystals. Then for a crystal of a given
initial size Ls,i' the final size is Ls,i + .6Lgrowth. The mass of the crystals is
PckV(Ls,i + M.growth)3. With no nucleation the number of crystals per volume
is constant. Thus
or
dWp = ( L . + M.growth )3 dW.

I,!
(4-41 a)
Ls,i
In terms of sieve fractions
T .+M. wth
W p,1. = [ ~,1 gro]3 W . (4-41b)
S,l
Ls,i
Equation (4-41a) or (4-41b) can be used to predict the differential weight distri-
bution. The total weight is
Ws,tot L· + M. (4-42a)
W p,tot = fo (
8,1
L .
S,1
growth 3
) dWs
or in terms of sieve fractions
L·+M. th
Wp,tot = ~
~
(5,1 groW)3 W5,1. (4-42b)
all fIacs Ls,i
and the magma density is
Wp,tot
MT = --
(4-43 a)
Vs
Since G and M. are unknown, the solution proceeds by first calculating

128
MT from mass balances and equilibrium. Equation (4-19) can be used to calcu-
late MT except we should correct for the weight of seeds added. Then
(4-43b)
Usually, this correction will be small. Then Eq. (4-43a) gives Wp,lOl' A trial-
and-error procedure is used to find a M..growtb (and hence a G) which will give
this value of Wp,lOl from Eq. (4-42). The shape of the product screen analysis
will be different than the seed distribution since the relative increase in mass is
greater for small particles than for large ones. A detailed example is given by
Foust et al (1980). Experimental data may vary significantly from the distribu-
tion predicted since the McCabe M.. law may not be valid.
Seeding can also be used to provide selective nucleation for fractional
crystallization of a solution saturated in two or more solutes. (Rousseau and
O'Dell, 1980). A modification of this procedure is used commercially for
separation of racemic mixtures of amino acids. Assume we have a saturated
mixture of the D and L isomers. If we cool the mixture but stay in the meta-
stable region, neither isomer will crystallize. Selectively seeding with one of
the isomers (e.g. the D isomer) will cause crystallization of that isomer. If con-
tact nucleation occurs, a large number of nuclei will form. The mixture will
approach saturation in the D isomer, but will remain supersaturated in the L
isomer. After harvesting the D isomer crystals, removing any D isomer nucleii
with a membrane or by heating followed by cooling to supersaturate the L iso-
mer again; the L isomer can be crystallized out by seeding with the L isomer.
Obviously, an operation such as this requires very careful control of conditions
to prevent unwanted primary nucleation.
4.6. EQUIPMENT MODIFICATIONS

How does modifying the equipment affect the CSD? This is an important ques-
tion for the design of crystallizers. In this section we will look at the removal
of crystal fines, use of classified product removal and methods of suspending
the solids. The analysis of CSD will again assume that growth rates are
independent of crystal size.
4.6.1. Fines Removal

The purpose of fines removal is to remove excessive numbers of small particles
so that a limited number can grow to larger sizes. The fines are removed by
129
Clear liquor
-----1=::;;==I=::fun==t=--- overflow
Mixed magma
Steam
Fines
i
Figure 4-6. Methods for fines removal. a. partially mixed. b. fines trap.
some type of classification scheme. This is possible since the fine particles tend
to move with the fluid while larger particles will settle out. Different
classification methods are illustrated in Figure 4-6 (Mullin, 1972; Saeman,
1961). Figure 4-6a illustrates the situation where fines classification occurs
because the crystallizer is not well mixed. Fines can be removed near the top
of the crystallizer. By adjusting the level of the withdrawal tube the size of
fines removed can be varied. The optimum size of fines to remove appears to
be approximately 1/20 to 1/10 the mean product size. Once removed the fines
are redissolved by heating and/or by adding solvent. This liquid is then recy-
cled to the crystallizer. A different type of fines trap is shown in Figure 4-6b.
Fines are collected at the level where particles larger than Lt have a settling
velocity greater than the fluid velocity. The fines are then destroyed by heating
the mixture. Fines dissolution can also occur in superheated regions inside the
crystallizer since the small particles are more soluble than large particles.
The CSD can be analyzed using the population balances (Larson and
Randolph, 1969; Randolph and Larson, 1988). Removal of fines decreases the
130
residence time of most of the small particles. Some of the fines will escape the
trap and grow to become product. Surprisingly, only 0.1 to 1.0% of the fines
are allowed to grow to become product. If we let
R = Product retention time > 1 (4-44a)

Fines retention time -
then if Qm is the flow rate into the crystallizer, the withdrawal rate of fines by
both destruction and in the product is RQm. The retention times for fines and
product are,
(4-44b)
v (4-44c)
'tp=-
Qm
where Lr is the maximum size of fines withdrawn.

The crystallizer with fines removal must have two population balances -
one for fines and one for product Equation (4-8) can be written twice.
dn n (4-45a)
-+-=0
dL Gtr
dn n (4-45b)
-+--=0
dL Gt p
Integrating these equations, the solutions are
(4-46a)
L (4-46b)
n = C exp( - - ) L > Lr
Gt p
where C is a constant of integration (see Example 4-3). Since the distribution

is continuous, at L =Lr the population densities are equal, nf =np'
The distribution in Eqs. (4-46) can be compared to the distribution
131
a
nO
I --I
G1T
-R
-I
I ,
GzT
In nO I
I
I
I
I
LF
L
b
2
1.0
0.8
M
MT
0~~2--~4--~--~8---1~0~
L,mm
Figure 4-7. Comparison of crystallizer with fines removal (2) to crystal-
lizer without fines removal (1). a. Population density distribu-
tion. b. Dimensionless cumulative weight distribution. Case 1
is from Example 4-1 while Case 2 is from Example 4-3.
without fines removal. This is shown in Figure 4-7a for the population density,
and in Figure 4-7b for the dimensionless cumulative weight distribution. Since
the fines are of negligible mass, MT,l = MT ,2' This condition forces
1 1 R (4-47a)
--<--<--
G2 't G1't G 2 't
Thus the crystallizers have different growth rates and
(4-47b)
G with fines removal > G wlo fines removal
132
The same production (MT,l = MT,Z) occurs with fines removal but on a much
smaller number of larger particles. Since the total surface area is less for the
large particles, the growth rate must be higher. Figure 4-7b shows that a
significant increase in the mean particle size also occurs. The dominant parti-
cle size for the mass distribution will also increase significantly.
With higher growth rates with fines removal, the supersaturation must be
higher. This will normally cause more nucleation and one would expect
n2 > n]'. In this case fines removal is analyzed for an idealization called a point
fines trap. A point fines trap destroys only nuclei.
We will compare crystallizer 1 to crystallizer 2 which has a point fines
trap. The analysis is similar to the analyses of the previous section. The two
crystallizers have the same V and Q. Hence, t1 = tz. Since nuclei represent an
insignificant mass, the magma densities are equal, MT, = MT•. Then from Eq.
(4-18)
(448)
where ~ is the fraction of nuclei surviving the fines trap. Thus, ~n2 is the popu-
lation density of surviving nuclei. The value of ~ can be as low as 0.001 in
practice. Next, the nucleation kinetics found by combining Eqs. (3-6) and (3-
8),
(449)
are substituted into Eq. (448). After dividing out equal terms and rearranging,
the result is
(4-50a)
The dominant size of the mass distribution, LD = 3Gr, is
Ln,z = (1- )l/(3+i) (4-50b)

L n ,1 ~
Since ~ is usually quite small, l/~ is large. Thus both the linear growth
133
rate G 2 and the dominant size Lo. are increased substantially. The degree of
improvement is reduced as the order of nucleation i increases.
Both analysis procedures show that fines traps can significantly increase
the median and dominant crystal sizes. Fines traps are often used on a variety
of different crystallizers when large crystals are desired. Fines traps are rela-
tively expensive to operate, but they are a universally valid method of control-
ling nucleation.
Example 4-3. CSD with Fines Destruction
Potassium chloride is crystallized in a cooling crystallizer with a fines

trap. The feed is saturated at 100°C and has a total flow rate of 10,000
kg/hr. The crystallizer operates at 20°C and has a volume of 2.0 m3 •
The fines trap is operated to remove crystals where Lr s 0.3 mm. The
product retention time is ten times the fines retention time. Determine
the growth rate in this crystallizer. Predict the total mass distribution,
M, as a function of crystal size L.
Solution
A. Define. This is the same system as in Example 4-1 except now there
is a fines trap. We want to find G and plot M versus L.
B. Explore. Equations (4-46) give the population density function. The

equilibrium and external mass balance calculations are the same as in
Example 4-1. Thus 'tp = V/Q = 'tw/o fines removal = 870 sec from Example
4-1. The magma density is also the same as without fines destruction,
and was obtained in Example 4-1 as MT = 175.0kglm3 • The growth
rate will be higher and needs to be calculated.
C. Plan. The residence time for fines 'tp = V/(RQin) = 'tf/R where R =
10. Thus 'tf = 87 sec. The mass distribution M can be generated from
the definition, Eq. (4-17a), and the population density functions, Eqs.
(4-46). Thus some derivation is required. For small crystals,
L
M = kvPc Jo L noexp( -
3 r~
\J~f
)dL, L s Lr
134
Thus for L s Lt, M is the same as Eq. (4~17b). The cumulative mass of
fines Mf at L = Lt is,
which from Eq. (4~17b) is
(4-51)
Me =6kvp,nOGte{(Gtr)'
- [(Gte)' + (Gtel'Lr + ~ (Gte)L[ + !Ll Jex ~)}

P( -
Since G is unknown this cannot be determined immediately.

For crystals where L > L f , the cumulative population distribution
is
L
M = Mf + kvPc f C exp( - GL)dL L> Lf (4-52a)
Lr 'tp
which becomes
(4-52b)
135
The total magma density MT is obtained by letting L ~ 00 in Eq. (4-

52b).
(4-53a)
Then, Eq. (4-52b) simplies to
(4-52c)
The value of kvnopc can be estimated from the value of MT found for
the MSMPR without fines destruction (Example 4-1) and from Eq. (4-
18).
Thus,
k P nO - 175 = 175 = 1.9887xlO15

v C - 6(Gt)kMPR 6(3.48xlO-4)4
Assume this is unchanged for the system with fines destruction.
For the system with fines destruction n is continuous at L = Lr. Setting

Eq. (4-46a) equal to (4-46b) at L = L f ,
(4-54)
136
Substituting M f and kvPcC into Eq. (4-53a)
1 3] exp( - -Lr
+ -G'tpLf - ) + (G'tf)4 - [(G)4
'tf
6 G't f
(4-53b)
Since MT is known, this represents one equation with G being the only
unknown. Solve for G. Then can find M for L $ Lr from Eq. (4-17b),
M f from Eq. (4-51) and M for L > Lr from Eq. (4-52c).
D. Do It. 'tp =870, 'tr =87, Lr = 3 x lO-4m
This is equal to the term in brackets in Eq. (4-53b). A trial and error
solution is required. The result is
G = 9.251 x 10-7 mls
This equation is very sensitive to the value of G used. It is best to do a

problem like this on a programmable calculator or a computer.
The value of kvPcC can now be calculated from Eq. (4-54).
kv cC = 1.9887 x 1015 exp(-1.7275) = 6.9444 x 1013

P exp(-Q.3727)
Calculate M f from Eq. (4-51), M f = 0.2561. Note that M f is a small

137
fraction of MT • M can be calculated from Eq. (4-52c). This equation

becomes
M = 175 - (4.1666x1014)(8.048xl~)[5.2134xl0-!O
+ 6.4776x1O-7 L + 4.024x1O-4L2 + 61 L3]exp( -L -4)

8.0484xlO
This is valid for L ~ Le. Programming this on a calculator we obtain:
L (m) M
0.0003 0.2561 0.0015
0.001 6.728 0.038
0.002 41.985 0.24
0.002953 87.5 = Median 0.50
0.0003 89.595
0.004 127.91 0.73
0.005 151.68 0.87
0.006 164.32 0.94
0.007 170.38 0.97
0.008 173.11 0.989
0.009 174.25 0.996
0.010 179.71 0.998
0.011 174.89
0.012 174.96
0.013 174.986
0.02 174.9999 1.0
This distribution is plotted in Figure 4-7b. Values of M below Lr can be

found from Eq. (4-17b).
E. Check. There are some consistency checks. As expected

G=9.251 x 10-7 > G w / o = 4.0 x 10-7 • Since fines have very little weight,
the small value of Mr/MT is expected. The median size is L = 0.00295
where M =0.5 MT • This is at L =3.67 (hp.
F. Generalization. This is a rather complex example. It illustrates the

steps necessary to determine Mr. M and MT for fines removal. Similar
138
steps can also be used for detennining the analogous quantities for
classified product removal, or for other complex situations. The
methods for detennining the constant of integration and for detennining
G from MT can also be used in other problems. Solution of the equa-
tion for G and generation of the cumulative mass distribution are
straightforward on a programmable calculator or computer, but difficult
without these tools.
If an experimental crystal size distribution is available, then G
and n° can be detennined. First, calculate the experimental values of
MT and Mi from Eqs. (4-26) and (4-27). The theoretical values for MT
and Mi are given by Eqs. (4-53b) and (4-52c) with C from Eq. (4-54).
Relating (MiIMT )cxpt'% to (M/MT )thcory gives an equation in Gi . The
average G can be found from Eq. (4-28) and nO can then be found from
Eq. (4-53b).
Note that the median at L = 3.67 Gtp is what would be expected
for a simple distribution. This occurs since Mf is so small. Note that
the median = 2.95 mm is significantly greater than the median = 1.28
mm found without fines removal (Example 4-1). Fines removal works!
It is interesting to compare this crystallizer to an equivalent
MSMPR with the same growth rate and same magma density. This
comparison (see Problem 4-D6) shows that the cumulative mass distri-
butions are essentially the same. However, the equivalent MSMPR
must have a nuclei population density significantly less than the crystal-
lizer with fines destruction. Removal of fines is a practical way of con-
trolling nucleation.
4.6.2. Classified Product Removal

Removal of crystals of the desired size by classifying the crystals and with-
drawing these crystals at a higher rate is another possible equipment
modification. This can be done in an elutriation leg where small crystals are
fluidized by the upflowing liquid. Large crystals have higher settling velocities
and thus can be concentrated. Product classification also often occurs inadver-
tently if mixing is inadequate to keep particles suspended.
If we could perfectly classify the product, then product classification
would produce a product with larger crystals. However, in practice the smal-
lest crystals are not separated, but occur in the product at approximately the
139
same concentration as in the mixed magma. The concentration of large parti-

cles in the elutriation leg will be higher than in the mixed magma. Thus, a rea-
sonable approximation is that classified product removal accelerates the remo-
val of larger particles without affecting the removal of small particles.
All crystals with L ~ r" are removed at a rate zQ where z ~ 1. Crystals
with L < r" are removed at a rate Q which is the volumetric flow rate through
the crystallizer. Then the residence times are,
(4-55a)
'ts = VIQ, L < r"
(4-55b)
The population balance Eq. (4-8) can be solved for each residence time.
L (4-56a)
ns = nOexp( - - G ), L < r"
2'tS
L (4-56b)
nL = C exp( - - G ) , L ~ r"
2'tL
where G 2 is the linear growth rate in the crystallizer with classified product
removal. At L = r" the population densities are equal, ns = nL. This allows
calculation of the constant of integration C.
The population densities are plotted in Figure 4-8 where they are also
compared with an MSMPR. The MSMPR is assumed to have the same nuclea-
tion population density n° and the same production of crystals. With the same
productivity, MT ,
(4-57)
must be the same for the two crystallizers. This will require that the MSMPR
have a slope intermediate between the two slopes of the classified product
removal case. Thus G 2 > G 1 since 'tl = 'tMSMPR. Since the classified product
removal system has more small crystals (larger n for small L) and fewer large
crystals, the MSMPR has a larger mean and dominant size for the weight distri-
butions. Thus the MSMPR produces larger crystals! The classified product
140
In n
~ .....
. . . . . . . . MSMPR
L
L
L
Figure 4-8. Comparison of MSMPR and crystallizer with classified product
removal.
withdrawal system does have a tighter distribution of product sizes. These

results agree with experiments.
Sharper distributions with larger crystals can be obtained by using both a
fines trap and classified product removal. In this case there will be three solu-
tions to the population balance (see Problems 4-C7 and 4-D8 and Figure 4-13).
Unfortunately, these modifications tend to increase the tendency towards insta-
bility (Garside, 1985; Randolph and Larson, 1988, Randolph et al., 1977). Cry-
stallization of KCI with classified product removal and fines destruction
showed low order cycling of CSD. Such cycling causes undesirable variations
in product quality and causes higher operating costs.
Another method of increasing magma density and average product size is
called clear-liquor advance. A quiescent region of the settler is used to with-
draw clear liquor. The product crystals are then drawn off at a much lower
rate. This method effectively increases residence times, increases the average
size, decreases the supersaturation, and increases MT. The limit to using
clear-liquor advance is the increase in secondary nucleation as MT increases.
4.6.3. Methods of Suspending Solids

Crystallizers can be classified based on the method of suspending the growing
crystals. One classification scheme for doing this is (Singh, 1979).
1. Circulating magma. Growing crystals are circulated through the zone
141
where supersaturation occurs. May also have fines removal and classified
product withdrawal. This type of equipment can be an MSMPR.
Examples were shown in Figures 2-3 a, b; 2-4 a, b; and 2-5 a. More
examples of mixing methods are discussed below.
2. Circulating liquor. Some type of crystal settling or screen is used so that
only clear liquor is recycled. The method is used in Krystal (or Oslo) type
crystallizers. Some authors (e.g. Saeman, 1961) see no advantage to this
method.
3. Scraped Surface. The wall of the heat exchanger is scraped with some
type of blade. This was illustrated in Figure 2-3c and d. This method
keeps the surface of the exchanger clean and gives high heat transfer rates.
However, scraping the crystals off the walls breaks up the larger crystals.
Thus only small crystals are produced. The effect of breakage of crystals
causes "death" in one size range and "birth" in smaller size ranges (see
the next section). These crystallizers are used to produce seed crystals, for
viscous mixtures, and when very low temperatures are desired.
4. Tank Crystallizers. Both agitated or static tanks are used with cooling to
the atmosphere or through a jacket. Static tanks will not be well mixed
and a wide CSD will result. For example, ammonium alum crystals in the
size range of +6 mesh to 2 inches are produced in tank crystallizers. Tank
crystallizers are used for small throughputs of relatively easy to crystallize
materials.
Mixing in crystallizers is very important, and is one of the major problems in

scale-up. If the crystals and solvent have close to the same density, the crystals
will be easy to suspend. Crystallizers equipped with mixers and baffles work
well for these systems. If an external heat exchanger is used, an external pump
is required. By returning the circulating liquid tangentially, much of the
desired mixing can be achieved. The turn-over rate of the suspension must be
high enough. This can be calculated by determining the sedimentation rate of
crystals, and limiting the distance they can fall in the time required to mix the
entire volume.
Draft tube or conduit mixing is commonly used if the crystals are
difficult to suspend. This is illustrated in Figure 4-9 (Nyvlt, 1982; Saeman,
1961). The fast rise, slow fall requires the least energy, works very well with
difficult to suspend crystals, and works well with fines traps. An air-lift can be
used instead of the mixer (this is called a Pachuca type). The air lift also pro-
vides for some evaporation. Conduit mixing with fast rise can also be used
142
a b
Figure 4-9. Draft tube mixing methods. a. Fast rise-slow fall; b. Fast fall-
slow rise.
with an external pump directing return liquor into the conduit. The fast fall,
slow rise is used on the Oslo-Krystal design. It is also useful when a trap is
used for fines removal, or when classification of product in the bottom of the
crystallizer is desired. Examples of these crystallizers are shown by Mullin
(1972), Nyvlt (1982), Rousseau (1986), and Singh (1979).
Scale-up of the mixing can be difficult. Unless growth is remarkably
fast, all regions of the mixer should have about the same supersaturation. In a
ten liter laboratory mixer a tracer bead may go through the entire mixing circuit
in a few seconds. A commercial-scale crystallizer could easily be IO feet tall
and 15 feet in diameter. The time to traverse the mixing circuit must be longer
if reasonable velocities are to be used. The half-life for supersaturation ranges
from a few seconds to several minutes while circulation times can range from
0.2 seconds in a 0.2 m diameter tank to about 35 seconds in a 3 meter diameter
tank (Garside, 1985). Thus, the two tanks can have significantly different
supersaturations and hence different nucleation and growth kinetics.
Generally, the large scale system is designed to be flexible so that its operation
can be adjusted to give acceptable results.
4.7. CSD FOR COMPLEX CRYSTALUZATION KINETICS

Crystallization can be much more complex than the situations discussed up to
now. Two other commercially important cases will be briefly discussed here.
The first is the inclusion of the birth and death of crystals in a given size range.
The second is determination of CSD for size dependent growth.
143
In real crystallizers there is invariably some attrition of crystals. If the

attrition results in contact nucleation, this effect can be accounted for in the
nucleation kinetics used in the population balance (see Eqs. (4-37) to (4-39».
Since the pieces removed from the large crystal are tiny, no other adjustments
in the population balance are required. In some crystallizers (e.g. scraped sur-
face) and with some types of crystals large crystals are often fractured into two
or more pieces. This results in the removal ("death") of crystals in one size
range, and the simultaneous "birth" of two or more crystals in smaller size
ranges. Including birth and death in the population balance is simple in a for-
mal sense, but difficult to do in a form useful for calculation. Equation (4-8) is
easily modified to include birth and death of crystals. The result is (Randolph
and Larson, 1988).
dn n (4-58)
G - + - = B(L) - D(L)
dL 't
where B(L) and D(L) are the birth and death functions which are obviously
related to each other.
To solve Eq. (4-58) functions for B(L) and D(L) must be developed.
Two-body death and birth effects are usually postulated Experimentally,
larger crystals are more likely to break. Thus a power-law death model is often
assumed.
(4-59a)
D(L) =kn L~
where k is a rate constant and f3 represents the size selectivity. Since contact of
crystals will be proportional to magma density, it would also be reasonable to
include magma density in the death function.
(4-59b)
D(L) = k n L~ M¥
The birth function is obviously related to the death function since mass must be
conserved. Depending on the assumptions made, various birth functions can be
derived. If a power law is used for the death function, the birth function will
also have a power law form.
The population balance Eq. (4-58) has been solved for the power law
death function, Eq. (4-59a), with a related birth function (see Randolph and
Larson, 1988). The results show a significant narrowing of the CSD with a
144
'0
.?:
~
iii
z
ILl 10"1
0
Z
0
fi
...J
:::>
0..
0
0..
en
en
ILl 1(J2
...J
Z
0
iii
z
ILl
~
0
DIMENSIONLESS SIZE (xl
Figure 4-10. Population density plot for size dependent growth (Abegg et ai,
1968). Reprinted from AIehE Journal 14, 118 (1%8) with
permission. Copyright 1%8, AlChE.
downward shift in the dominant size. The population density of the largest size
crystals plummets. Thus, when large crystals are desired, breakage needs to be
minimized. This can be done by using gentle mixing and by making impellers
and baffles from a soft material such as polyethylene.
A second situation which commonly occurs is size dependent growth. In
this case, the empirical model shown in Eq. (3-15) is often used. The solution
of the population balance Eq. (4-7), along with Eq. (3-15) is (Abegg et al.,
1968).
(4-60a)
where
(4-60b)
145
Table 4-3. Integration Constants for Size Dependent Growth Eqs. (4-61a,
b).
b C1 C2
0.9 5,841,000 1,681

0.8 6205 57.00
0.7 539.0 19.03
0.6 140.9 10.75
0.5 57.75 7.431
0.4 29.93 5.688
0.3 17.86 4.622
0.2 11.70 3.904
0.1 8.177 3.389
0.0 6.000 3.000
-0.1 4.569 2.696
-0.2 3.583 2.452
-0.4 2.356 2.084
-0.6 1.659 1.819
-0.8 1.221 1.618
-1.0 0.9341 1.461
-2.0 0.3438 1.000
Source: O'Dell and Rousseau (1978)
Equation (4-60a) agrees with Eq. (4-9) if b = 0 and 'Y= 1I(Gnuclei't). This value
for 'Y will be used in the following analysis.
When growth rate increases with size (this is the common case), b is
positive. Examples are the growth of K2S04 and Na2S04 • lOHzO crystals.
Equation (4-60a) is plotted in Figure 4-10. Faster growth as size increases
results in more crystals in this larger size range compared to size independent
growth (b = 0). This is usually desirable. Note in Figure 4-10 that all the
curves converge together for L/(G't) < 2. Thus size independent growth models
give good results for small crystals. Abegg et aI .• (1968) obtained good fits
with experimental data for both Na2S04 • 10 H20 and alum.
The weight distribution can be determined from Eq. (4-60a) or the
parameters for the size dependent growth, Eq. (3-15) can be determined experi-
mentally from weight distributions. The weight distributions are complicated
146
functions involving definite integrals (O'Dell and Rousseau, 1978). The

magma density and dominant size of the weight distributions can be calculated
from
(4-61a)
(4-61b)
where C 1 and C z are calculated from definite integrals and are given in Table
4-3 (O'Dell and Rousseau, 1978). Equations (4-61) are valid when
Y= lI(Gnuclei 1:) in Eq. (3-15).
4.S. BATCH CRYSTALLIZATION

Many crystallizations and precipitations are done as batch operations particu-
larly for biochemicals. Batch processes will be the choice if the yearly produc-
tion rate is fairly low or the equipment must be used to produce several dif-
ferent products. Unfortunately, batch processes usually produce a poorer qual-
ity product than continuous processes since the crystals are usually smaller.
Better products can be obtained by controlling the cooling rate.
The population balance for unsteady state systems was derived in Eq.
(4-6) while on the way to the continuous results. Equation (4-6) with birth and
death functions added would be a logical starting point to develop CSDs for
batch crystallizers. Unfortunately, although it has been tried (see Garside, 1985
and Tavare, 1987, for references) this approach has not proven to be extremely
useful. Instead, an analysis incorporating the seeding analysis of Section 4.5
has been useful in predicting how to control the crystallizer temperature.
Batch crystallization usually starts with a hot, saturated solution which is
seeded and then cooled and/or evaporated to cause supersaturation. The rate at
which the solution is cooled will have a major impact on the CSD. The "nor-
mal" or "uncontrolled" method of cooling would start cooling the hot saturated
solution with the cooling fluid at its usual temperature. From the heat transfer
equation
(4-62)
Q = UA (TM - Tco1d )
we can estimate the rate of cooling. This equation is more complicated than it
147
6~~~----r--------'--------'---------r---.
50
u
•
q,
.
...
III
~
"0
~ 40
Q.
E
~
o 50 100 150 200

Time. t. min'
Figure 4-11. Cooling curves determined for batch crystallization of potas-

sium sulphate. From top down: Controlled cooling - theory,
Eq. (4-72); Controlled cooling -- simplified theory, Eq. (4-74);
Linear; Uncontrolled cooling. (Mullin and Nyvlt, 1971).
Reprinted with permission from Chem. Eng. Sci., 26, 369
(1971). Copyright 1971, Pergammon Press.
initially appears. The magma temperature, TM, decreases during the crystalli-
zation and the overall heat transfer coefficient U is probably not constant
because of fouling and the increase in viscosity of the magma. However, Eq.
(4-62) can easily be used to estimate the initial rate of cooling when TM = TF
and the heat transfer surface is clean. Q will be large initially, and will
decrease as both U and (TM - T cold) decrease. The cooling curve calculated for
this case is curve d in Figure 4-11 (Mullin and Nyvlt, 1971). This rapid initial
removal of heat will rapidly cool the magma temperature and supersaturate the
solution. The result is likely to be the production of a large number of nucleii
instead of sustained growth on the controlled number of seed crystals. This
uncontrolled nucleation then causes a CSD with a very large number of rather
small crystals. This is illustrated in Figure 4-12 (Jones and Mullin, 1974).
Much better results can be obtained by controlled cooling (Mullin and
148
Q)
·55 80
.
- -. -..
'--
~
o
~ 60 .....
~
Q)
a.
size -----,,-----
---- -
.....
~ ,,
\
\
\
20
,
::l \
E \
::l
U
,
\
°0~~~5~0~0--L-~10~0~0--L-~~~
Crystal size, )Jm
Figure 4-12. Experimental CSD for batch crystallization using different

cooling methods. methods. • Uncontrolled . . Fol-
lowing Eq. (4-74). (Jones and Mullin, 1974). Reprinted with
permission from Chern. Eng. Sci. 29, 105 (1974). Copyright
1974, Pergamon Press.
Nyvlt, 1971; Mullin, 1972; Jones and Mullin, 1974; Belter et ai, 1988). In con-
trolled cooling the cooling rate will be slow at first to minimize excessive
nucleation and will be fast after the growing crystals have reduced the supersa-
turation (see Figure 4-11). The improvement in CSD which can be obtained is
illustrated in Figure 4-12.
To determine the temperature versus time or cooling curve desired for
controlled cooling we start with a supersaturation balance (Mullin and
NyvIt,1971; Mullin, 1972).
(4-63)
llc is the supersaturation defined by Eq. (2-2a), A is the total area, Kg and kN
are the rate constants for growth and nucleation from Eqs. (3-16) and (3-4),
respectively, and nand i are orders of growth and nucleation, respectively. The
left hand side of this equation is the change in supersaturation with time. On
149
the right hand side the first term represents the change in the saturation concen-
tration with time. Since c· is a function of temperature and we are cooling the
solution, this corresponds to the creation of supersaturation according to
dc· dc· dT (4-64)

Tt= dT dt
where -dT/dt is the cooling rate. The term dc· /dT can be determined from the
solubility data such as Figure 2-1. The second and third terms in Eq. (4-63)
correspond to the reduction of supersaturation caused by growth and nuclea-
tion, respectively.
Usually, we seed the batch and want to minimize other nucleation. To do
this we operate within the metastable region illustrated in Figure 2-2. If we
asswne that we are successful in eliminating nucleation, the last term in Eq.
(4-63) will be zero.
The left-band-side ofEq. (4-63) can be written as
d(Llc) d(Llc) dT (4-65)

--=--
dt dT dt
Now Eq. (4-63) becomes,
dT KgA(Llc)n
-- (4-66)
dt dc· d(Llc)
-+--
dT dT
The metastable zone width in Figure 2-2 probably depends upon tem-
perature. Thus, the supersaturation Llc will depend upon temperature. If we
asswne a linear dependance, we have
(4-67)
Often the metastable zone width will not change (a2 = 0), and d(Llc)/dT = O.
We are now left with only the term KgA(Llc)n to evaluate. From Eq. (3-
16) this term equals dm/dt which can be related to the crystal size by Eq. (3-
12).
(4-68)
150
With no nucleation the seed crystals will grow to the final size.
(4-69)
The summation over a series of growth rates occurs because G can be a func-
tion of both temperature and crystal size. Substituting Eqs. (4-69) and (4-68)
into Eq. (4-66) we obtain
t-I
3kv p,G[Lsecd + 1:G]2
dT i=O (4-70)
-=--~-----
dt dc" d(tic)
-+--
dT dT
The volumetric shape factor kv can be difficult to predict If the seeds grow
without changing shape, we can use Eq. (3-12) to detennine kvPs from the
known weight and size of the seed crystals.
Wseed (4-71)
kvPs = L~
Substitution of this result into Eq. (4-70) gives

t-I
3 Wseed G [Lseed + 1: G]2
dT
-=-..-------.,----
i=O (4-72)
dt
dC" + d(tic)
[ dT dT
1L3 seed
This equation gives the desired cooling rate as a function of time. Numerical
integration of this equation gives the temperature versus time results shown in
Figure 4-11.
Equation (4-72) is somewhat awkward to use. A considerably simpler
form can be obtained by making additional assumptions, but the engineer must
be aware that these assumptions may not be valid for a particular system. If the
linear growth rate does not depend on temperature or crystal size, the summa-
tion in Eq. (4-72) becomes G. Often the supersaturation is constant and
d(tic)/dT = O. Equation (4-72) then simplifies to
dT 3 Wseed G 2 (4-73)
-d-t = (dc*/dT) L~ (Lseed + G)
151
If dc*/dT can be assumed to be constant, this equation is easily integrated to

obtain the required controlled temperature as a function of time.
(4-74)
The cooling curve obtained using this simplified result is compared in Figure
4-11 to the result obtained by numerical integration of Eq. (4-72). The
improvement in CSD when temperature is controlled following Eq. (4-74)
instead of being uncontrolled is shown in Figure 4-12. An example of the use
of Eq. (4-74) is given by Belter et al (1988).
4.9. DOWNSTREAM PROCESSING METHODS

The magma exiting from the crystallizer contains both crystals and mother
liquor. Some downstream processing steps are required to recover the crystals.
These steps are affected by the size and shape of the crystals as well as the pro-
perties of the liquor.
The most common downstream processing steps are filtration and centri-
fugation. To remove the mother liquor, which contains impurities, the crystals
are often washed with solvent or water. However, the wash step depends on
the size distribution and shape of the crystals. Regular shape prismatic crystals
(Figure 3-1B) are much easier to filter without excessive pressure drops than
either tabular (Figure 3-1A) or needle forms (Figure 3-1C). Washing of the
prismatic crystals is also easier since the filter cake is more porous and the
surface-to-volume ratio is lower. These steps are also simpler with larger crys-
tals with tight CSD. Centrifugation is an alternative to filtration, and the same
comments hold.
The crystals may require a final drying step to remove any remaining sol-
vent or mother liquor. This is again easiest to do for large prismatic crystals
with tight CSD.
Screening or sieving or other size classification methods may be required
to produce a product within the desired size range. This step is reduced or
eliminated if the crystallizer can produce a tight CSD of the desired median
size. Screening and sieving separate based on the second largest dimension.
Referring to Figure 3-1, both tabular and needle forms will bounce on the
screen until the smallest cross-sectional area has a chance to slip through the
152
screen opening. For example, needles which are much longer than the screen
opening will readily pass through. If accurate sizing is required, it may be
necessary to grow prismatic crystals.
Bulk storage, movement and packaging of crystals are all made easier
when it is free flowing and not dusty. Large, prismatic crystals with a tight
CSD are most likely to be free flowing. Crystals will produce less dust if
mother liquor is removed before drying; otherwise, drying of the mother liquor
often deposits a dust of solute on the crystals.
In summary, unless the product requires tabular or needle form crystals,
prismatic shaped crystals are best for downstream processing.
4.10. SUMMARY-OBJECTIVES
This chapter has focused on the theoretical aspects of population balances and
crystal size distributions. The objectives are:
1. Be able to derive the population balance.
2. Solve the population balance for size independent growth. Generate the
number, area and weight distributions.
3. Use experimental weight distributions to determine crystallizer kinetics.
4. Determine the effect of changing residence time and magma density on
crystallizer kinetics with and without secondary nucleation.
5. Explain the effect of and do calculations for fines destruction,. classified
product withdrawal, and different mixing procedures.
6. Utilize CSD for size dependent growth.
7. Determine the cooling curve for batch crystallization.
REFERENCES
Abegg, C.F., J.D. Stevens, and M.A. Larson, "Crystal size distributions in con-
tinuous crystallizers when growth rate is size dependent," AIChE Journal, 14,
118 (1968).
Bennett, R.C. and M. Van Buren, "Commercial urea crystallization," Chern.

Engr. Prog. Symp. Ser., 65 (95),44 (1969).
153
Canning, T.F., "Interpreting population density data from crystallizers," Chern.

Eng. Prog. 66. (7) 80 (1970).
Foust, A.S., L.A. Wenzel, C.W. Clump, L. Maus, and L.B. Andersen, Princi-
ples of Unit Operations, 2nd ed., Wiley, New York, 1980, 526-529.
Garside, J. and M.B. Shah, "Crystallization kinetics from MSMPR crystalliz-

ers," Ind. Eng. Chem. Process Design Develop .. 19. 509 (1980).
Garside, J., "Industrial crystallization from solution," Chem. Eng. Sci .• 40. 3
(1985).
Larson, M.A. and A.D. Randolph, "Size distribution analysis in continuous cry-
stallization," Chem. Eng. Prog. Symp. Ser .• 65 (95), 1 (1969).
McCabe, W.L., J.C. Smith, and P. Harriott. Unit Operations of Chemical

Engineering. 4th ed., McGraw-Hill, NY, 1985, Chapt. 28.
Moyers, C.G., Jr. and R. Rousseau, "Crystallization operations," in R.

Rousseau, (Ed.), Handbook of Separation Process Technology, Wiley, New
York,1987,Chapt.11.
Mullin, J.W., Crystallization. 2nd ed., Butterworth, London, 1972.
Nyvlt, J., Industrial Crystallization. 2nd ed., Verlag Chemie, Weinheim, 1982.
O'Dell, F.P. and R.W. Rousseau, "Magma density and dominant size for size
dependent crystal growth," AlChE Journal. 24. 738 (1978).
Ramkrishna, D., "The status of population balances," Rev. Chem. Engr., 3 (1),
49 (1985).
Randolph, A.D. and M.A. Larson, Theory of Particulate Processes. Analysis

and Techniques of Continuous Crystallization. 2nd Ed., Academic Press, New
York,1988.
Randolph, A.D., J.R. Beckman, and Z.I. Kraljevich, "Crystal size distribution
dynamics in a classified crystallizer," AlChE Journal, 23, 500 (1977).
Rousseau, R.W. and F. P. O'Dell, "Separation of multiple solutes by selective

nucleation," Ind. Eng. Chern. Process Des. Develop., 19,603 (1980).
154
Saeman, W.C., "Crystallization equipment design," Ind. Eng. Chem., 53, 612
(1961).
Singh, G., "Crystallization from solutions," in P.A. Schweitzer, (Ed.), Hand-

book of Separation Techniques for Chemical Engineers, McGraw-Hill, New
York, 1979, Section 2.4.
Tavare, N.S., "Batch crystallization: A review," Chem. Eng. Commun., 61,259

(1987).
HOMEWORK
A. Discussion Problems.
Al. Why is the median size for the dimensionless number distribution,
L/(Gt) = 0.693, less than that for the dimensionless weight distribution,
L/(Gt) = 3.67?
A2. Explain the concept of a population balance in your own words.
A3. Define the following terms.

a. MSMPR
b. population density
c. moments
d. 't
e. L/(Gt)
f. dominant size
g. point fines trap
h. classified product removal
i. clear liquor advance
j. birth and death
155
A4. Sketch the expected plot of In n vs L to compare a crystallizer 1 without

a point fines trap to crystallizer 2 which has a point fines trap.
A5. Why is a weight average used to determine Gin Eq. (4-28)?
A6. When secondary nucleation does not occur and '[2 > '[1 but MT •1 = MT •2 ,
Eqs. (4-34) relate growth rates and nucleation population densities.
Sketch the graph of In n versus L if,
a. i= 1
b. i =3
Comment on your graphs.
A 7. Construct your key relations chart for this chapter.
C. Derivations
Cl. Derive Eqs. (4-11b), (4-13b), (4-15b), and (4-17b).
C2. Derive Eqs. (4-2Sa and b).
C3. Prove that the dominant sizes for differential mass, area, length and
number distributions are U(Gt) = 3,2,1 and 0, respectively.
C4. Prove that the median size for N/NT is at L/(Gt) = 0.693. Find the
median of the cumulative area distribution, NAT'
CS. Show that Eq. (4-49c) follows from Eq. (4-49b).
C6. Complete the details of the derivation of Eqs. (4-47a and b) for a point
fines trap.
C7. Combinations of fines destruction and classified product removal can be

analyzed using the population balances. Solve the population balance,
Eq. (4-8), for this combination. A population density plot for an indus-
trial crystallizer (a Swenson DTB) producing KCI is shown in Figure
4-13 (Canning, 1970). Interpret this plot in terms of the population den-
sity function. What is happening in each zone? What are the values of
Gt, R and z?
156
8 .n·
c:
9 7
~
ZON F I
>- ~
t: 6
VI
Z "I:s>... Z N 2
1&1 ""'<:: "<l
o 5
z
o 'R
~ ZONE 3
~
r"
"12
LI L
o 2 4 6 8 10 12 14 16
CRYSTAL SIZE. MM
Figure 4-13. Experimental population density results for KCI crystallizer

with fines removal and classified product withdrawal (Canning,
1970). Reprinted with permission from Chem. Eng. Prog., 66
(7),80 (1970). Copyright 1970, AIChE.
D. Problems
Dl. For the crystallization of urea in a large draft tube, baffled evaporative
crystallizer the following data were obtained.
G = 0.0240 mm/hr, 't = 6.70 hours, and In nO = 11.45 where n° is

No./mm-L. See 4-D2 and 4-D5 for other data.
Predict the screen analysis if the following screens are used: 14,20,28,
35,48,65.
D2. For Problem 4-Dl determine the distribution for n, N, L, A and M. Plot
157
these distribution. Assume kv = 1.0 and kA =6.0. Do this for the con-
tinuous distributions.
03. For Problem 4-01 determine the sieve analysis distributions for n, N, L
and A.
04. A cooling crystallizer is used to crystallize KCl. At 60°C the following
Tyler screen data are obtained.
Screen 14 16 20 24 28 32 42 60 80 115
Wt,kg 0.001 0.029 0.136 0.139 0.276 0.210 0.209 0.049 0.012 0.002
KCl forms cubic crystals with Pc = 1.98 glml. At 60°C the saturated
fluid has a density of PF = 1.198 glml. V, = 10 liters. Residence time in
crystallizer is 15 min. Find G and n°.
05. An evaporative crystallizer is used to crystallize urea. The following
data are obtained (Bennett and Van Buren, 1969)
Crystallizer
't, hrs Tyler Screen Sizes Density,
14 20 28 35 48 65 gIL
a) 5.9 36.0 62.0 80.5 92.5 98.5 100 510
b) 3.12 26.2 51.5 76.2 94.0 99.5 100 317
where the screen numbers are cumulative weight % measured experi-

mentally. Urea, NH 2 CONH2 , has Pc = 1.33 glcm3 • The crystals are
"chunky", use kv = 1.0. Find G and n° for parts a and b.
06. A sodium acetate crystallizer receives a saturated solution at 60°C. This
solution is seeded with 0.2 mm average size crystals and cooled to lO°C.
What is the mean size of the product crystals Lp? Ifthe residence time is
't = 21 hours, what linear growth rate G is required? Use a basis of 100
kg of entering water to which 0.25 kg of seeds are added. Oata are in
Figure 2-1 and Table 2-1. Why would this crystallizer be difficult to
operate?
158
D7. Example 4-3 showed that the crystallizer with a fines trap had a linear
growth rate ofG = 9.251 x 10-7 m/S.
a. Calculate the cumulative mass distribution M for an MSMPR
operating with this G value, 't == 'tp = 870s, and MT =MT,with trap =
175 kg/m 3 • Compare these results to the distribution found in
Example 4-3.
b. What value of (kvPcnO) will this MSMPR have?
D8. When we crystallize potassium chloride without a fines trap, we find that
MT = 175 kg/m 3 , 't = 870 sec, G =4.0 x 10-7 mIs, and the order of
nucleation i = 3.5. We desire the CSD to have a median size of 3 mm for
the cumulative mass distribution when we use a point fines trap. Calcu-
late the fraction of nucleii which can survive the point fines trap to
achieve this median size. Residence times and magma densities for the
crystallizers with and without the point fines trap are identical.
D9. Glauber salt (Na2S04. lOH20) demonstrates size dependent growth.

Abegg et al .• (1968) found a good fit to data when
Y= G 1. 1 8 (em-I), b=0.2
= 000
nuclCl't .
Predict the magma density MT and the dominant size Ln. Crystal density
Pc = 1.46 g/cm 3 , kv = 1(2.
E1. A classified product removal is used to remove crystals larger than 1.0
mm at a rate 4 times faster than removal of small crystals. Otherwise,
the crystallizer is the same as the MSMPR explored in Example 4-1.
Assume that the population density of nuclei and the magma densities
are the same for the MSMPR and the classified product removal system.
Determine G, the median size of the weight distribution, and the cumu-
lative weight distribution for the classified product removal system.
Compare your results to the MSMPR.
159
E2. The product from Problem 4-D4 (crystals + solution) will be fed to a
second cooling crystallizer operating at 20°C. Assume that the feed to
this crystallizer (at 60°) is a saturated mixture and that the product
magma is in equilibrium at 20°C. Assume no nucleation occurs.
Residence time is 15 min.
a. Predict MT
b. Predict the crystal size distribution
c. Predict G.
Saturated KCI solution at 20°C has a density of 1.174 glm!.
Chapter 5
CRYST ALLIZATION FROM THE MELT
In crystallization from the melt a liquid mixture is separated by crystallization.

No solvent is present. For example, a mixture of impure xylenes can be
purified by crystalization since the p-xylene crystallizes at a much higher tem-
perature. Since both melt and solution crystallization are crystallization
processes, there are many similarities. For example, many of the phase equili-
brium behavior illustrated in Chapter 2 and nucleation and growth processes
discussed in Chapter 3 are similar.
There are also significant differences between solution and melt crystalli-
zation which will be highlighted in this chapter. In solution crystallization a
variety of methods of causing supersaturation are used to induce crystallization.
In melt crystallization cooling is almost always used. Most industrial processes
for solution crystallization use a single equilibrium stage while devices with the
equivalent of many stages are often used in melt crystallization. Since there is
no solvent in melt crystallization, there is no need for solvent removal and
recovery, and there is no possibility of contamination by the solvent. However,
there is also no way to reduce the viscosity, increase the diffusivity or remove
energy by solvent evaporation. In addition, the chemicals being purified must
be stable at the melting point - there is no way to significantly lower the operat-
ing temperature. For systems such as salts where the melting points are very
high, crystallization from solution is less expensive. Lower operating tempera-
tures usually make the separation easier. Distribution coefficients can be orders
of magnitude more favorable at the lower temperatures possible in solution cry-
stallization.
These comparisons show why crystalization from solution is much more

common for bulk separations. Melt crystallization becomes advantageous
160
161
when the presence of a solvent would be detrimental. This occurs in two situa-
tions: 1. When the melting point is at a convenient temperature and equili-
brium is favorable for separation, melt crystallization is used for bulk separa-
tions. Examples are purification of p-xylene and separation of dichloroben-
zenes. If crystallization from solution were used, a solvent other than water
would be required and the solvent would have to be recovered. The solvent
recovery costs make crystallization from solution more expensive than melt
crystallization for these chemicals. 2. When an ultrapure product is desired,
solvent would be an impurity which would have to be removed. Several dif-
ferent melt crystallization processes such as normal freezing and zone melting
have been developed for ultrapurification. Ultrapure organics, metals and sem-
iconductors are all purified this way.
In the first section of this chapter the fundamentals of melt crystallization
will be briefly discussed. This section will build on Chapters 2 and 3. Then
methods for doing melt crystallization will be explored. First, slurry crystalli-
zation method where the crystals are in a slurry of crystals and liquid melt will
be discussed. Then, ultrapurification methods such as normal freezing and
zone melting are discussed. In these processes the solid is continuous and is
not in a slurry.
5.1. FUNDAMENTALS OF MELT CRYSTALLIZATION
5.1.1. Equilibrium
Phase diagrams for three melt systems were shown in Figures 2-7 and 2-8.
Solid solution behavior for phenanthrene and anthracene was shown in Figure
2-7. Eutectic behavior for naphthalene, phenol and naphthalene, I-naphthol
was illustrated in Figure 2-8. Note in Figure 2-8 that quite good predictions
can be made using activity coefficients predicted with computer routines
(Walas, 1985). In ultrapurification only a small section of the diagram (up to a
few percent impurity) is needed.
Equilibrium data for melt systems is available in a variety of sources.

The Landolt-Bomstein (1956) tables summarize data for many binary and a
162
Table 5-1. Selected distribution coefficients, Eq. (24-1), for inorganics

(Shoemaker and Smith, 1967) and organics (Zief, 1967).
Compound T,oC Impurities (ki)

Al 659.7 Co (0.14), Cr (0.8), Ca (0.06), Fe (0.29), Ga «0.1),
La «0.1), Mn (1), Sc (0.17), Sm (0.67), W (0.32)
Fe 1535 C (0.29), 0 (0.022), P (0.17), S (0.04 to 0.06)
GaAs 1238 Cd (<0.2), Cu (<0.002), Fe (0.003), Ge (0.018)
H2 O 0 D2 0 (1.021), HF (10-4), NH3 (0.17), NIJ..F (0.02)
Sn 231.9 Ag (0.03), Bi (0.5), In (0.5), Sb (1.5), Zn (0.05)
Anthracene Anthraquinone (0.005), carbazole (> 2.0),
fluorene (0.1), phenanthrene (0.06), tetracene (0.06)
Naphthalene 2-naphthol (1.85 and 2.3)
2-Naphthol Naphthalene (0.4)
Phenol p-nitrophenol (0.22)
Pyrene Anthracene (0.125), 1, 2-benzanthracene (1.95)
Cyclohexane Methylcyc10pentane (0.6)
Benzene Cyclohexane (- 0.15)
Hexadecanol Octadecanol (0.75)
few ternary systems. Timmermans (1960) also has data on a large variety of
binary systems. Phase diagrams for metals are available in Hanson (1958) and
Rhines (1956). Data on oxides and ceramics are given by Alper (1970-78) and
Levin et al. (1964,1969). Calculations are summarized by Walas (1985).
Since melt crystallization methods, particularly normal freezing and zone

melting, often operate with only a percent or less impurity, equilibrium is often
represented as a distribution coefficient The distribution coefficient is usually
defined as
~ = concentration of impurity i in solid (5-1)

concentration of impurity i in melt
Any consistent set of units can be used for the concentration measurement.
Distribution coefficients for a large number of inorganic materials (Shoemaker
163
and Smith, 1967) and a few organic materials (Zief, 1967) have been tabulated.
Selected values are given in Table 5-1. For most impurities k < 1.0 and the
impurity is excluded from the crystal. In some cases such as eutectic systems k
« 1.0 and exclusion is essentially complete for systems at equilibrium.
The distribution coefficient can be related to the complete phase
diagrams. For a solid solution system, illustrated by Figure 2-7, ki can be
obtained at any temperature. For instance, if phenanthrene is considered to be
the impurity, ki at 200°C is
~ _ 1-0.95 mole frac anthracene :;:; 0.05 :;:; 0.18

1-0.72 0.28
If the liquidus and solidus lines are straight, ~ will be constant. Otherwise, ~
will depend on concentration. For a eutectic system such as Figure 2-8 the
solid will be pure at eqUilibrium. Thus ki = O. The apparent distribution
coefficient will be greater than zero because of inclusions, imperfect washing
and other non-idealities. At low concentrations, the distribution coefficients of
different impurities are often independent However, this is not true for sem-
iconductors where large intemctions often occur (Wilcox and Estrin, 1982).
S.1.2. Heat and Mass Transfer
Heat and mass transfer processes are usually very important in melt crystalliza-
tion. Heat transfer is usually much more important in melt crystallization than
in solution crystallization. The tempemture and composition profiles are shown
schematically in Figure 5-1. The solid tempemture is usually somewhat greater
than the liquid temperature since the latent heat of solidification must be dissi-
pated. The solid is usually assumed to be at its melting point. In very slow
processes such as normal freezing and zone melting the amount of supercool-
ing, Ts - T bulk, may be only a few tenths of a degree.
If ki < 1, impurity is rejected at the growing interface of the crystal.

Thus there will be a tendency for impurity to build up at the interface unless
either mixing is vigorous or diffusional mass transfer is mpid. Usually the
164
a b
Cj
I
I
I
I - - - - - Tbu1k
j--Film -----I
t
Adsorbed
Layer
Figure 5-1. Schematic of temperature and composition profiles in melt cry-

stallization. a. Temperature. b. Concentration of impurity
with ki < 1 .
interface can be assumed to be at equilibrium; thus, Ie;. = Cs/Ci is the actual

equilibrium constant. The apparent equilibrium constant k. pp = Cs/Cbulk will
always be closer to unity than Ie;..
The heat transfer analysis for a growing crystal can be treated formally in
an approach similar to the mass transfer theory given in Eqs. (3-9) to (3-17).
The heat transfer across the films is,
dq (5-2a)
dt =hA(Ti-n
where A is the area of the crystal and h is the film heat transfer coefficient.
This film coefficient can be written as h = kcoodlOr where kcond is the thermal
conductivity and Or is the hypothetical film thickness. Unfortunately, Or is not
known and Or is not equal to the film thickness for mass transfer. It is often
convenient to express the heat transfer in terms of an overall heat transfer
coefficient U, since the interface temperature Ti is not required.
dq (5-2b)
dt=UA(T,-n
The solid temperature T, will essentially be the melting point. Thus T, is

known while Ti is nol
165
The energy which must be dissipated is the latent heat of fusion, A..
Then,
dq =A. dm (5-3)
dt dt
where m is the mass of solid deposited. The latent heats are readily available
(Mullin, 1972; Nyvlt, 1982; Perry and Green (1984). The rate of mass transfer,
dm/dt, was given by Eqs. (3-9) to (3-11) or Eq. (3-16) depending on the surface
rearrangement.
Ifheat transfer controls the rate of growth of the crystal, then setting Eqs.
(5-2b) and (5-3) equal we obtain
dm UA(Ts -1') (5-4)

--
dt A.
Heat transfer will often control in processes such as zone melting and normal
freezing where there is very little impurity and the impurity has little effect on
the kinetics of crystali1ization. When large amounts of impurity are present,
mass transfer will often be limiting and the analysis of Chapter 3 is valid. For
in-between situations both heat and mass transfer need to be included (Meyer
and Shen, 1970).
S.2. SLURRY PROCESSES
In slurry processes melt crystallization is operated with discontinuous crystals

in a continuous liquid. The three methods of slurry operation are staged sys-
tems, end-feed columns and center feed columns. These systems are com-
monly used for bulk separations although they can be used for ultrapurification
of eutectic systems.
The melt crystallizer must compete with other bulk separation processes
such as distillation or crystallization from solution. The latent heat of fusion
for organic compounds is approximately 1/3 the heat of vaporization. Thus, a
166
crystallization process will require less energy than distillation all other things
being equal. However, distillation has much higher mass and heat transfer
rates; thus, distillation columns are much cheaper per equilibrium stage. Cry-
stallization is favored over distillation for bulk separations only when distilla-
tion does not work well. Distillation may not be the separation of choice when:
1. The relative volatility is close to one. Separation factors for crystallization
may be much higher. 2. Azeotropes are formed. Eutectics may not occur, or
may be less of a problem. 3. Very high temperatures or thermal decomposi-
tion occur in distillation. Crystallization operates at lower temperatures. 4.
The product is desired as a crystalline solid. A crystallizer may be able to do
the required separation and produce the desired product in one step. 5. An
existing crystallizer can be used. Capital expenses will be very low if existing
equipment can be used. 6. A pure product is being further refined to higher
purity. For example, distillation is usually very energy intensive when going
from 99.9% to 99.99% purity.
5.2.1. Continuous Staged Processes
When equilibrium is favorable, a single equilibrium stage may be sufficient.

For example, for both eutectic systems shown in Figure 2-8 pure naphthalene
, . . - - - - -..... Recycle
Yit M
C,X;
Figure 5-2. Single stage continuous, steady-state crystallizer.

167
can be produced if the feed concentration is greater then the eutectic concentra-
tion. If the concentration of impurity (phenol or I-naphthol in Figure 2-8) is
quite small, a large amount of relatively pure naphthalene can be produced
with a minimal amount of eutectic. The eutectic would then be separated by
other methods (solution crystallization, distillation, extraction, etc.) and be
recycled. For solid solution systems (see Figure 2-7) a single equilibrium stage
will be sufficient only if the feed is already quite pure. The equilibrium calcu-
lations are essentially the same as those for solution crystallization (see Chapter
2).
If the distribution coefficient is constant, single stage calculations are

straightforward. For the steady-state continuous well-mixed system shown in
Figure 5-2 the mass balances are
(5-5a)
F=M+C
(5-5b)
where F is the feed flowrate, M is the melt flow rate, and C is the flow rate of
crystals in any set of consistent units. To be specific we will treat flow rates F,
M, and C in kg/hr. The weight fractions of impurity in the crystals and melt are
Xi and Yi> respectively. Then the equilibrium expression is
(5-5c)
where we assume that the impurities are independent. Combining Eqs. (5-6) to
remove unknown M and Yi, the solution is
~YF,i
X;.=----- (5-6)
1- ~(1-~)
F
This gives the concentration of the crystals. The adhering melt will con-
tain more impurity. This can be removed either by partial melting and drainage
or by reflux. These methods are discussed later.
168
)(
Figure 5-3. McCabe-Thiele diagram for single stage crystallization. Either

continuous steady-state or batch crystallization with completely
mixed crystals.
An alternate analysis method for continuous crystallization is to use a

McCabe-Thiele diagram as shown in Figure 5-3. The mass balance Eq. (5-5b)
can be solved for Yi to develop the operating equation.
(5-7)
The intersection of the operating line and equilibrum curve gives Yi and Xi'
This analysis is obviously applicable to nonlinear equilibria as long as the
solutes are independent
If one equilibrium stage is not sufficient, crystallizers can be cascaded to

achieve additional separation. For solid solutions continuous operation will be
essentially the same as the counter-current cascade used for solid solutions for
solution crystallization (Figure 2-20). Between each stage the crystals and melt
must be separated from each other, and the crystals must be redissolved and
recrystallized. Thus, the process is expensive unless there are very few stages.
Equilibrium staged calculations can be done on an enthalpy-composition
diagram (ponchon-Savarit diagram) which will look very similar to Figure 2-
21. Since this process will be rare in industrial practice, the development of
this analysis procedure is left as Problem 5C-l.
169
5.2.2. Batch Staged Crystallization
Batch crystallization calculations can be done several different ways. The sim-
plest operation is to load the empty crystallizer with Mo kg of feed mixture of
weight fraction Yi.O' Then cool the charge and crystallize until the desired
weight fraction, g, of the charge has frozen. The melt, (1- g)Mo kg, is then
drained off and gMo kg of crystals remain. Assuming a homogeneous well-
mixed crystal phase, equilibrium Eq. (5-6c) is valid at the end of the batch.
The mass balance on impurity at the end of the batch is
(5-8)
gMo Xi + (1 - g) Mo Yi == Mo Yi.O
Assuming that the impurities are independent and combining Eq. (5-6c) and
(5-8), we obtain
ki Yi.O (5-9)
~ = --:~:--.-+-:":"(I-_g"7"")
Note that if everything is frozen (g = 1) then Xi =Yi.O and no separation is

achieved.
A McCabe-Thiele analysis can be used for batch crystallization if the

crystals are homogeneous and well mixed. Equation (5-7) is applicable if we
interpret C and M as the kg of crystals and melt produced, and Xi and Yi are the
weight fractions of the crystals and melt when the batch is done. With this
interpretation, the McCabe-Thiele diagram is again shown by Figure 5-3.
Equation (5-9) or a McCabe-Thiele analysis are almost always

oversimplifications. Crystals are usually not homogeneous. When the first
layers of the crystal are formed. they will be in contact with solution of lower
impurity concentration. As impurity is rejected its concentration in solution
increases. Thus later layers will be less pure. To solve this problem we need a
time dependent equation. Let dm/dt = rate of growth of crystals in kg/hr.
dm/dt is set by the heat transfer rate and is assumed to be constant Then the
unsteady state mass balance is
- [dm
dt
1x. == d(MyJ
1 dt
(5-10)
170
where M is the kg of liquid melt in the crystallizer at time t. M can be related

to the crystal growth rate by an overall mass balance.
(5-11a)
Again defining g as the weight fraction of melt which has frozen
g = Mo - M = 1 _ M = (dm/dt) t (5-11b)
Mo Mo 110
Equation (5-11a) was substituted in for M/Mo to give the last equation in Eq.
(5-11b). From Eq. (5-11b) we obtain both
(5-11c)
M=Mo(1-g)
and
Mo (5-11d)
dt = (dm/dt) dg
These last two equations plus the equilibrium Eq. (5-6c) are substituted into Eq.
(5-10). After rearrangement,
dxi dg (5-12a)
-=(1-~)-
~ (I-g)
Integrating Eq. (5-12a) and applying the initial condition that the very first
crystal is in equilibrium with the initial liquid,
(5-12b)
Xi =~ Yi.O, for g = 0
we obtain
(5-13)
171
This equation gives the instantaneous solid concentration as a function of Yo, k

andg.
The average solids concentration can be obtained by integrating Eq. (5-

13).
fo Xi dg (5-14)
Xi,avg = - - -
g
=
Note that when the entire melt is frozen, (g = 1,) xi,.vg = Yi,o which is the
expected result Comparison of Eqs, (5-14) and (5-9) shows that the progres-
sive freezing calculation where the crystal is non-homogeneous results in a
lower Xi,avg that the completely mixed crystal. This analysis for a non-mixed
crystal turns out to be the same analysis as for normal (or progressive) freezing
(see Section 5.3.1).
The development leading to Eqs. (5-9) and (5-14) assumes equilibrium

throughout the process. If the freezing rate is not very slow, then equilibrium is
probably not reached. Equations (5-9) and (5-14) can still be used if an effec-
tive distribution coefficient ketr is used. The effective distribution coefficient
depends on all of the operating conditions. Equation (5-9) and (5-14) also
assume that all of the contaminated melt can be removed from the crystals.
This is often a problem when ultrapure product is required.
5.2.3. Partial Melting and Drainage
Additional purification can be obtained by partially melting the crystals and

draining off the melt (this is also called sweating). Since the impurity lowers
the freezing point, crystal layers with higher impurity concentration will melt
first Draining this melt washes the interstitial melt from the crystals, and can
result in a considerable increase in purity. The freezing and melting can be
done batch-wise in the same vessel. This "Proabd" process has been used com-
mercially and is discussed later (see Section 5.3.1). An alternative is to do the
172
freezing and thawing in separate vessels. This process has also been commer-
cialized for purification of benzene, naphthalene and other organics (Molinari,
1967b).
The effect of partial melting and drainage can be seen by a simple addi-
tion to the equilibrium model for batch crystallization. When harvested, the
crystals will entrain some weight fraction, e, of the melt even after the crystals
are allowed to stand and drain.
e = weight melt entrained (5-15)

weight of crystals
This entrained melt is assumed to be in equilibrium with the outer layer of the
crystal.
(5-16)
Ifthis melt is not removed, the melt plus crystals have a total concentration of
(5-17)
which becomes
eYi.O (1 -g )
ki-l
g -
+ Yi.O [1 ( I-g)ki] (5-18)
Yi.tot = e+l
This is the product concentration which will be observed if the melt is not
removed.
It is desirable to remove the entrained liquid since it is concentrated in

impurity. This can be done by partially melting the crystals and allowing them
to drain. Assume we crystallize a weight fraction gl of the initial charge. Then
x.vg and Ytot are given by Eqs. (5-14) and (5-18) with g = gl' Now melt a por-
tion of the crystals ~g. Assume that the drainage occurs by displacing the old
173
liquid with the new melt If ilg ~gl' with perfect displacement all of the old
entrained liquid will be removed. The fraction of crystals collected is now
(5-19)
These crystals have an average composition of,
(5-20)
xi,avg2 = -Yi,O [1 - (1 - g2 )ki]
g2
This equation is obtained from the original crystallization since the interior of
the crystals is unaffected by the partial melting.
The new melt entrained on the crystals can be a mixture of old melt plus
the newly melted crystal layer. If all the old melt is removed, we can assume
that the composition of the entrained material is the same as the melted crystal.
Thus
(5-21)
gl
Yi,avg = ilg ilg
The product composition will now be,
(eYi,avg + Xi,avg2) (5-22)

Yi,prod = (e + 1)
where the fraction entrained is assumed to still be e. Excess liquid was drained
off. This equation becomes
(5-23)
Yi,prod = (e + I)
Although less product is collected after partial melting and drainage, it is

significantly purer. This is explored in Example 5-1.
174
Prediction of the entrainment e will be difficult The entrainment will

depend on the crystal shape, the crystal packing, viscosity, swface tension and
the processing steps. Entrainment can be reduced by blowing an inert gas
through the bed of crystals, by pulling a vacuum or by centrifuging the bed. To
ensure removing all of the old entrained melt, a fraction of crystals
significantly larger than eg2 will often be melted. This reduces the yield, but
gives a purer product Partial melting and drainage can also be used with con-
tinuous processes (see Problem 5-C3).
Batch processes can be cascaded to increase the purity. After harvesting,

the crystals are melted and then recrystallized. This reduces the initial concen-
tration of impurity for the second crystallization, and will result in purer pro-
duct crystals. The melt from this batch will be near the composition of the ini-
tial charge. Thus, this melt can be recrystallized to produce more crystals.
These can then be recrystallized to form the final product. Methods for cou-
pling batch processes to increase purity and yield are explored further in Prob-
lem 5-Bl.
Example 5-1. Batch melt crystallization with partial melting.

Ten kilograms of a pyrene mixture containing 4 wt % anthracene is
purified in a batch crystallizer. 60% of the original charge is frozen.
After filtration, 0.08 weight fraction melt adheres to the crystals.
a Find the product concentration.

b. If 1 kilogram of the crystals are melted and the melt is drained off,
find the new product concentration.
Solution
a The results of the initial freezing can be found from Eq. (5-14). Yo =
0.04, g = 0.6, k = 0.125 (from Table 5-1). Then the anthracene concen-
tration is
x.vg = ~ [1 - (l_g)k] = ~~ [1- (0.4)°·125] = 0.0072

175
The entrained melt is given by Eq. (5-16),
y = yo(1_g)k-l = (0.04)(0.4r~·875 = 0.0892
Then the mix of entrained melt plus crystals is given by Eq. (5-17)
= ey + Xavg = (0.08)(0.0892) + 0.0072 = 0.0133

Yttt e+ 1 1.08
Note that the adhering melt causes a considerable increase in the impur-
ity concentration of the product.
b. ~g = 0.1 since 1 of the original 10 kg is melted. Then g2 is
The remaining 5 kg of crystals have a composition given by Eq. (5-20).
0.04 0066
] = o.
0.125
XAvg.2 = 0.5 [1 - (0.5)
Assuming that the old entrained melt is washed away by the newly
melted material, the new adhering melt composition will be given
approximately by Eq. (5-21).
Yavg =
The product composition combining this melt with the crystals is given
by Eq. (5-22).
eYavg+ x avg.2 = (0.08)(0.0101) + (0.0066) = 0 0069

Yprod = e+1 1.08 .
176
This product is obviously much purer than the product obtained without
drainage. In fact, this product is theoretically purer than the crystals
before partial melting. This occurs because the most impure layers
were melted off.
5.2.4. Column CrystaUization
An alternate method for slurry melt crystallization when a single equilibrium

contact is not sufficient is to have continuous contact between the solid and
liquid in a column. Some method such as pulsed flow or a screw conveyor is
used to move the crystals countercurrent to the liquid.
Two classes of column crystallizers have been developed. In the end

feed system a slurry is fed to the column. The column operates as a stripping
column to obtain further purification of the crystals. The largest application of
end-feed columns has been in the Phillips process for purifying p-xylene by
melt crystallization. This process is shown in Figure 5-4 (McKay, 1967). The
first scraped surface crystallizer is operated at a low temperature to increase the
recovery of the p-xylene. The crystals from the crystallizer are recovered in a
rotary filter and then melted. This melt serves as the feed to the second scraped
surface crystallizer which produces the slurry feed for the purification column.
Note that this feed has already been through a two stage crystallizer. The crys-
tal bed is moved downwards when the piston of the pulse unit retracts. Crystals
are pulled into the melter. Some of the melt is removed as product while the
remainder is refluxed. This reflux is warmer than the crystals and partially
melts some of the crystals. The latent heat for melting is supplied by freezing
some of the reflux. This partial melting of the most impure layers of the crys-
tals (which have a lower melting point) and refreezing of the relatively pure
reflux is one of the major reasons for the purification. The other major reason
for purification is the contaminated melt is washed off by the countercurrent
flow of much purer melt produced by the reflux. This washing is the main
function of the countercurrent motion. The impure melt is removed through a
filter or screen near the top of the column.
177
MOTHER LIQUOR
rEED XYLENES e.!!.,. P- XYLENE
22 °10 P- XYLENE
SCRAPED-SURFACE
CHILLER REFRIGERATED
WITH ETHYLENE
3JfC
• MOTHER
LIQUOR
85-88°70 P- XYLENE
CRYSTAL PURIFICATION
COLUMN
RECYCLE -Ieoc
99+°70 P-XYLENE
PRODUCT STORAGE
Figure 5-4. Phillips process for purifying p-xylene (McKay, 1967).

Reprinted with permission from Zief, M. and Wilcox, W.R.
(Eds.). Fractional Solidification, Marcel Dekker, NY, 1967.
Copyright 1967, Marcel Dekker.
The xylene system is an extremely favorable system for melt crystalliza-

tion. The melting points of ortho, meta and para xylenes are -46.9°C, -2S.2°C
and IS.3°C, respectively. The melting points are sufficiently far apart to make
crystallization relatively simple, and the system does not form solid solutions.
Thus the crystals are essentially pure p-xylene and the column removes
entrained melt. Since the boiling points are very close, distillation is not feasi-
ble. Recently, adsorption using the Parex process (see Chapter 10) has been
used for most new p-xylene purification plants. Existing crystallization plants
continue to operate since their operating costs are low. Other products such as
beer, fruit juices and other food products have been purified with end-feed
column crystallizers. When the crystals are ice, the column is inverted and the
solids move upward.
178
Several different designs for end feed crystallizers have been developed
(Albenino et al.• 1967; Henry and Moyers, 1984; McKay, 1967; Moyers and
Rousseau, 1987). Most of the large columns use the pulsed system shown in
Figure 5-4. The pulses tend to prevent the formation of incrustations, and they
sweep the screens free. Piston systems have also been used particularly for
columns with ice. The piston systems use a piston to push the crystal bed
through the column. The piston systems appear to be more difficult to scale up.
Center feed columns use a liquid melt as feed and form crystals inside
the column. Schematically, the column can be arranged as shown in Figure 5-
5a. The analogy to a distillation column is obvious. At the bottom, crystals are
melted producing a product with the higher melting point. The reflux at the
bottom of the column is equivalent to boilup. At the top of the crystallizer rela-
tively pure lower melting product is withdrawn. Some of the melt is frozen and
returned to the column as reflux. Conceptually, it should be easy to separate a
binary mixture into two relatively pure products.
Unfortunately, the analogy between crystallization and distillation breaks

down when one looks at hydrodynamics or mass transfer. For most systems the
melt and the crystals have approximately the same density. This is another
advantage of solution crystallization. Usually Pc>Pr in solution crystallization
while Pc-Pr in melt crystallization. Thus gravity is usually insufficient to force
rapid countercurrent flow of the melt and the crystals. Some positive means of
forcing the crystals countercurrent to the melt is required. This can be done in
a centrifuge, with a piston, in a pulsed column, or with a variety of mechanical
conveyors. For center feed columns mechanical conveyors such as spirals are
often used (Albenins, 1967; Henry and Moyers, 1984, Mullin, 1972; Moyers
and Rousseau, 1987). A variety of specialized columns such as the Brodie or
Schildknecht differ by the mechanical means used to move the crystals. A Bro-
die column is shown in Figure 5-5b. Exact design of the spiral and scale-up is
difficult. Most commercial applications are relatively small systems. The size
limit of a Brodie crystallizer is about 30 million pounds per year. Preliminary
design calculations for a Brodie crystallizer are discussed by Moyers and
Rousseau (1987). The 4C crystallizer (counter-current cooling crystallizer)
developed by Tsukishima Kikai (TSK) in Japan and licensed by C.W. Nofs-
179
lt
C M
b Recovery section
======~
_C
t
~
Coolant
Coolant
C Residue
~ Purifying section
,.--'\-+--~Steam L = Liquid
C = Crystals
Liquid product
Figure 5-5. Center feed melt crystallizer. a. Schematic b. Brodie type

column.
inger in the U.S. is based on the Brodie crystallizer (Anon, 1988). It uses
several tanlHype scraped surface crystallizers and has a capacity about five
times that of a Brodie crystallizer.
Mass transfer in a crystallizer differs from that in distillation. As has

been mentioned before, molecules inside the crystal do not exchange mass with
the melt Thus an impurity can easily be carried to the melter where it mayor
180
may not be refluxed. The reflux does cause melting of the outer layers of the
crystals and refreezing of the relatively pure reflux. The main function of the
countercurrent motion seems to be to wash the interstitial melt from the crys-
tals. Thus eutectic systems where the crystals are expected to be pure can be
purified easily in a column crystallizer. Solid solutions which require a large
number of contacts are much more difficult to purify. This is evident when one
determines HElP or HTU values. These values are often large, and may not be
constant when the column is made longer.
Both types of column crystallizers can be designed using either a staged

analysis with an experimentally measured HElP or a mass transfer analysis.
The staged analysis will be similar to that used in Chapter 2 for solution cry-
stallization. The mass transfer analysis is preferred since it is closer to the
actual process.
For eutectic systems the mass transfer analysis for end-feed crystallizers
is straightforward if solid of eutectic composition is not present. Since equili-
brium predicts that the crystals are pure (see Figure 2-8), the main purpose of
the column is to wash impurities from the liquid adhering to the crystals. For a
plug flow system the rate of mass transfer of the adhering liquid is
M dy (5-24a)
A dz = ka(Yad - y)
c
where M is the melt flow rate in kglhr, ~ is the column cross sectional area, k
is the mass transfer coefficient based on the interfacial area per volume, a,
between crystals and liquid. The adhering melt has a weight fraction Yad'
Rearranging this equation,
M Yout d
L-fo dz - f
L
(5-24b)
y
~ka Yin (YarY)
where we have assumed that M/(~ka) is constant. This assumption and later
assumptions in the mass balances can be approximately valid for end feed
181
Yad IN ,eC
'r -- ----,
I I
I :
I I
I I
I I b y= Yad
eC M
-+-r+ __ J
I I
L_
C
x
Y YOUT
Yad,OUT
eC
1\4
Melter
YIN . . . . . : , : . - - - - - - - - - - - -
Yad,OUT Yad
Figure 5-6. End-feed column crystallizer. a. Schematic. b. Operating

diagram.
columns which are adiabatic except at the melter. Equation (5-24b) can be
written as
(5-24c)
L = (HTU)(NTU)
where
Yout d
M
HTU=Acka' NTU- J Y
(5-24d)
Yin Yad - Y
To determine the NTU we must find Y-Yad as a function of y. This can

be done with a mass balance for the end-feed column shown schematically in
Figure 5-6a. The adhering melt is entrained at a weight fraction e defined by
Eq. (5-15). The mass balance using the balance envelope shown in Figure 5-6a
is
(5-25a)
eCYad,in + CXm + My = eCYad + Cx + MYout
182
where we have assumed that crystal and melt flow rates, C and M, are constant.
For eutectic systems Xm = x = O. Solving for Y
eC eC (5-25b)
Y= M Yad + Yout - M Yad,in
This is shown in Figure 5-6b where the difference (y - Yad) is indicated. Solv-
ing Eq. (5-25b) for Yad and substituting this into Eq. (5-24d), we obtain
(5-26)
The usual specifications are Yad,in and Xproduct. Since the product is
melted to produce reflux,
(5-27a)
Yin = Xproduct
Since the crystals are pure, the product concentration is related to materialleav-
ing the crystallizer by
eC Yads,out (5-27b)
Xproduct = C (1 + e)
From this equation we can detennine Yads,out. We can then detennine Yout from
an external balance around the crystallizer. With pure crystals this is
eC (5-27c)
Yout = M (Yad,in - Yad,out) + Yin
These calculations give us the limits needed to detennine NTU.
Use of this analysis requires an estimate of the amount of melt adhering

to the liquid. It also requires an estimate of HTU. In column crystallizers there
183
is considerable mixing and dispersion which is not included in this analysis.

More detailed analyses and comparison with experiments (Albertins et al.,
1967; Meyer and Shen, 1970) show that an effective dispersion coefficient
should be included in the expression for HTU. Since the analysis is linear, this
is mathematically valid. The analysis developed here can be used with the pro-
viso that ka must be determined from experimental crystallization data, or that
the HTU must be based on experimental data. In some small crystallizers puri-
fying a eutectic a total HTU of 12.3 cm was measured while a solid solution
had an HTU of 3.3 cm. (Albertins et al., 1967).
The analysis of center feed columns is much more complex. Flow rates
are not constant. The column is not adiabatic. Usually solid solutions are
separated and the crystal composition plus the adhering melt must be con-
sidered. This analysis is beyond the scope of this book.
5.3. CONTINUOUS SOLID PHASE
Entrained melt can drastically decrease the purity when we desire to ultrapurify
(see Example 5-1). Entrainment is a problem with slurry processes because of
the very large surface area of the crystals. Entrainment can be drastically
decreased by operating with a continuous solid phase (a single ingot) since
there is very little surface area. Of course, reducing the surface area also
decreases mass transfer rates. Thus the continuous solid phase systems are
used only when ultrapurity (e.g. 99.9+%) is required.
A variety of clever methods for melt crystallization with a continuous

solid phase have been developed (Henry and Moyers, 1984; Keller and
Muhlbauer, 1981; Moyers and Rousseau, 1987; Prann, 1966; Wilcox and Est-
rin, 1982; Zief and Wilcox, 1967; Zief, 1969). The two main variants, normal
freezing and zone melting, will be discussed.
5.3.1. Normal Freezing
Normal or progressive freezing is a process in which a melt is slowly frozen in

such a way that a single solid ingot is formed. It is desirable to form a flat
184
a b c
Wall
~SOlid
t
Refrigerated
Bath
JE Cool
Melt
Figure 5-7. Nonnal freezing. a. Refrigerated. b. Heated. c. On wall of

heat exchanger.
interface since the entrainment of melt will be greatly reduced. Two ways to
do this in the laboratory are shown in Figure 5-7 a and b. The melt is slowly
lowered into a refrigerated bath or removed from a furnace so that a single
solid ingot fonns. The rate of movement is a few cm/hour or less. More rapid
movement of the container causes the growth of crystals with facets or den-
drites. These greatly increase the entrainment of melt and decrease the purity.
Stirring of the melt will help remove impurity from the interface. The process
can be done on the wall of a heat exchanger as shown in Figure 5-7c. The
planar interface grows slowly into the melt. Laboratory equipment is discussed
by Wilcox and Estrin (1982), Zief and Wilcox (1967), and Wynne and Zief
(1967).
The geometry can be changed significantly. The freezing process can be

done horizontally or from the walls of the container into the center. Further
purification can be obtained by partial melting and draining of the ingot. For
the system shown in Figure 5-7a the tube would be turned upside down and
slowly warmed to achieve partial melting and drainage.
A variety of phenomena can reduce the separation (Wilcox, 1969). A

completely planar interface is desired. If back diffusion of the impurity into the
melt is slow, impurity will become concentrated at the interface. This change
in impurity concentration causes a change in freezing point. At the interface
the freezing point is lowest since impurity concentration is highest. Although
the temperature increases as one moves into the melt (the melt is heated), the
185
increase in freezing point may be quicker. Then the melt will be supercooled a
short distance from the interface. If this occurs, the interface can break down
and structures such as grooved cells or dendrites will form which can trap melt
The result will be a large increase in the impurity level. The solutions to this
problem are to freeze slowly and to stir the melt Other phenomena which can
adversely affect normal freezing are adsorption of impurity, variations in the
freezing rate, and gas bubbles trapped at the interface.
A commercial progressive freezing process was developed by Proabd

(Molinari, 1%7a). Proabd observed that naphthalene in drums was always
purer near the walls than in the center of the drum. This can be explained as a
normal freezing process. The drum is filled with molten naphthalene. As it
sets in the warehouse the drum slowly cools. The naphthalene freezes first
along the walls and excluded impurities collect in the still molten center. Pro-
abd used this observation to develop his process. The commercial equipment
consists of a tank filled with finned tube heat exchangers. The charge is slowly
cooled and most or all of it is frozen. The cooling rate is carefully controlled to
prevent growth of dendrites and to maximize the segregation of impurity. The
charge is slowly reheated and melt is drained from the vessel. The first melt is
the most impure. As melting continues the melt becomes purer and purer. A
variety of fractions can be collected. The Proabd process is available from
BEFS Technologies (Mulhouse, France) and 31 plants have been installed
mainly for purification of coal derived chemicals such as naphthalene and
anthracene (Anon, 1988).
Another commercial process which is essentially a normal freezing pro-

cess followed by partial melting and drainage is the MWB process developed
by Sulzer Bros. Ltd (Saxer and Papp, 1980). In this process the melt is slowly
frozen on the inside of long vertical tubes. A thickness of product from 5 to 20
mm is built up during this normal freezing step. The system is then heated to
slightly above the melting point to partially melt the crystals. The drainage
which is high in impurity is stored. Then the entire crystal mass is melted and
drained off. A new charge is then fed to the crystallizer and the process is
repeated. The products from each step are stored separately and fed at different
steps in the cycle to simulate a 3 to 5 stage countercurrent process. A variety
186
of organics such as naphthalene, benzoic acid and fatty acids (with a solvent
present) are commercially separated in 20 plants worldwide using this equip-
ment (Anon, 1988).
The analysis for normal freezing is essentially the same as for the pro-
gressive freezing of crystals in a batch system, Eqs. (5-10) to (5-14). Melting
and drainage are described by Eqs. (5-15) to (5-23). Since application of these
equations to normal freezing is straight forward, no further amplification will
be given. More detailed theories are discussed by Wilcox (1967).
5.3.2. Zone Melting
Zone melting is a technique for purification by melt crystallization. The

method can be understood by referring to Figure 5-8a. An ingot of solid is
heated locally so that a portion melts. As the heater is slowly moved along the
bar, the molten zone also moves. At point m the solid melts while at f the crys-
tal freezes. If impurity is excluded from the solid, Ie; < I, the impurity will con-
centrate in the molten zone. As this wne moves the impurity is swept to the
ends of the bar. If any impurity has ~ > I, it will concentrate at the other end
of the bar. Thus the center of the bar may be the purest Additional passes of
the molten zone will sweep more and more impurity to the ends of the bar. The
result is eventually an purification of most of the ingot.
Heater
a Molten Zone
Solid Solid
b
z=o z
x--J
-y
I .£
m
xo
z=L
C
z=o ....
X-Iy I
Figure 5-8. Zone Melting. a. Schematic of apparatus, b. Bar when z < L -
I, c. Bar when z > L -1.
187
This method has been extensively studied and used for ultrapurification
of a wide variety of materials. Zone melting was invented by Pfann, and has
since been extensively reviewed (Pfann, 1966; Wilcox and Estrin, 1982; Keller
and Muhlbauer, 1981; Mullin, 1972; Zief, 1967, 1969; Zief and Wilcox, 1967;
Shoemaker and Smith, 1967; Wynne, 1967; Moates and Kennedy, 1967). The
method is easily automated in the laboratory for routine ultra purification. A
home-made unit is sketched by Hudgins (1971), while more complicated units
are discussed by Pfann (1966) and Wynne and Zief (1967). The process is rou-
tinely used for ultrapurification of materials particularly semiconductors.
Between 100 and 200 metric tons of floating zone silicon single crystals are
produced every year (Keller and Muhlbauer, 1981). Although the scale of
operation is small compared to most other chemical engineering separations,
the economic value is significant.
A large number of variations have been developed. For instance, it is

easy to see that multiple zones can be used instead of the single zone shown in
Figure 5-8a Operation is often done vertically instead of horizontally. A wide
variety of container materials and shapes are used. In fact, the container can be
entirely removed in the floating zone method (Keller and Muhlbauer, 1981;
Pfann, 1966; Wynne, 1967). This technique is almost necessary for materials
melting above 1400o C. The ingot can be moved instead of the heaters. Con-
tinuous methods have also been developed (Pfann, 1966; Moates and Kennedy,
1967).
A variety of heating sources can be used. Below about SOOoC electrical

resistance heating is both convenient and inexpensive. For metals and sem-
iconductors induction heating is useful since the container does not have to be
heated. Induction heating also provides stirring of the molten zone which
increases separation. Radio frequency (RF) induction heating is used for
large-scale floating zone purification of silicon (Keller and Muhlbauer, 1981).
Since zone refining is used for ultrapurification, the initial charge is usu-
ally 99 wt % or more pure. The zone velocity can range from 0.1 to over 100
cm/hr. Typical values are 1 cm/hr for organics, 2.5 cm/hr for metals and 20
cm/hr for semiconductors (Henry and Moyers, 1984). Purification is higher at
lower movement rates. Multiple zone passes are usually employed. Com-
188
pounds purified by zone melting are surveyed by Shoemaker and Smith (1967)
and Zief (1967).
The separation phenomena occurring in zone melting are very similar to
those in normal freezing. Too rapid movement of the zone can cause the for-
mation of dendrities or crystallite facets. This drastically decreases the separa-
tion. Dendrite growth is also a significant problem if impurity concentration is
beyond one percent or so. Mixing helps to decrease these problems. Perhaps
the major advantage of zone melting compared to normal freezing is the ease of
using multiple zones. This allows many simultaneous passes of the molten
zone and significantly increases the separation.
The molten zone should be stirred to mix the zone and increase mass
transfer rates. For low temperature compounds a mechanical stirrer can be
inserted in the zone. Care must be taken to avoid contamination. Induction
stirring is convenient since it can be done as part of the heating. Electrical con-
ductors can be stirred magnetically. Low melting point compounds can be
stirred by pumping melt through an external heat exchanger.
The equilibrium analysis of batch zone melting is similar to that for nor-
mal freezing. The heater and hence the molten zone moves at a constant rate of
f cm/min. We can relate f to the solidification rate dm/dt in kg/min as,
(5-28a)
Note that dm/dt is also the rate at which the molten zone moves in kg/min.
Since the linear zone movement rate f is kept constant, dm/dt is constant Since
tf = z, from Eq. (5-28a) and a mass balance
(dm/dt)t z (5-28b)
Mo =/
where I is the length of solid melted to form the zone, and 110 is the kg of melt
in the molten zone (check the units on Eq. (5-28b) to understand it better). Dif-
ferentiating Eq. (5-28b) we obtain
Mo dz (5-28c)
d -
t- (dm/dt)l
which will be useful to simplify the mass balance.

189
The solute mass balance must be done in two parts. First, consider Fig-
ure 5-8b where z < L - l. The mass balance is
dm (Xo - x) = Mo -dy
-
(5-29)
dt dt
where x and y are the impurity weight fractions. If we assume that the zone is
well mixed and that the freezing interface is in equilibrium with the melt, we
can use the equilibrium expression y = x/k in Eq. (5-29). Substituting in Eq.
(5-28c) we obtain
k dx (5-30a)
I (xo - x) = dz ' z < L - 1
Before the start of the operation, x = Xo for all z. When the first part is melted,
y = Xo (the melting interface is not in equilibrium with the melt). Then the first
material refrozen is in equilibrium with y, or x = ky = no. Thus the boundary
condition is
(5-30b)
The solution to Eqs. (5-30a,b) is
- X = 1- (1-k)e-kzJI (5-31)
Xo
During the end of the zone pass when z > L - I, the situation is shown in
Figure 5-8c. Now the mass balance is
- [dm
dt
1 X= d(My)
dt
(5-32)
since the size of the zone is shrinking. Note that Eqs. (5-32) and (5-10) are the
same. The final step is essentially a normal freezing. Equations (5-11c) and
(5-11d) can be used to write the equation in terms of g. Then Eq. (5-32)
simplifies to Eq. (5-12a). The solution will be given by Eq. (5-13),
x = Yo k(1_g)k-l, Z ~ L-l (5-13)

190
..0
rTI, I: ~"t-TC--.nTIl l l11::~;IIIf

TITTT_-_n1TI1 1 lIT fITt ' t+lrr11:1:~:::I~~*llmTlllfttlrTl
II::~:II':::=~
i..:c:--. '
""\
0
~ ~QI
i'l--1"'~,.....
..JJII++++ +-tll+++1 ++---lllj.J +j " " " Htttt-I++--f+I+++++--l-----1
\ '\:::::~ "'<}c
~. . . .
1\112- ... .\, "- . . . . ., 1'1 ...... N ., • • • '" N ••
2
OIl lOt 1'1 .., ... N •••
2
'If'll fill ... 0
g 5! b 0
co
c , d
n 20
, . I
, L ,
2"
..
,
,-
~;~ ~
'l\ \
2
/, ,
", 1.4
:1\ \
,\ \ \
0
, n ,
~~ '1\ \ \
. k=I.2
'r--
, L , \ L/l .. o
n=I-IO
:~ fL.)- ./ ,.....- \\ \
....- V ~
'[7 V L /' ./"' V V/ W IA
\
·l\ \ \\
V V ~~\ 10 \
'z //'Z /' V~
'/ :~ ~~ .
o ,~ V~
00' L "
V ~~ ~ I I
!~~ ~~ K
00;
,,
o 00
, t---~~ ~ ....
>< 006
, , '-
>,,0 I )
-..... ./ >< .
>< '/ ,. , ~
0.
•
'-':: ~
' 7 L L,,/ // /
0.'
\ 1'--"'-~~~
'; / • \ \
L,'/ /1, ~~ l"\\ \
k o.!) ,
10
'V ~ ~7 // lit
n =
=
1-20 0.' •
\\
"" "l'\.\ \ \
0°069"L '\.\ \
0000 : 0. ,
000' , "
ULTI ...... TE '\
0006 , L , : PI5TAIBUTION",
Q.4
/ • , rz
0.00< ' ULTIMATE
I DISTRIBUTION ",
.• /L I 0.3 \
'L I 0.2
2
0.002
I
'v ,, I
I
0.00
/
, , - . . 0" . - - 10
zle z/f.
Figure 5-9. Relative solute concentration versus dimensionless zone length. n is the number of zone passes
(pfann, 1966). Reprinted with permission from Pfann, Zone Melting, 2nd ed., Wiley, N.Y.,
' -0
1966. Copyright 1966, Wiley. -
192
where Yo is the concentration of the molten zone when the normal freezing
starts. The value of Yo can be found from Eq. (5-31) with z =L-l
(5-33)
Yo = ~ = ~[1- (1- k)e-k(L-l)/I]
k k
When z > L - 1, the parameter g is related to z by
z - (L -1) (5-34)
g=-~-..:...
1
These results are for a single pass. For the next zone pass, Xo in Eqs. (5-
29) and (5-30a) is the x from the previous zone pass. Thus Xo will be given by
Eqs. (5-31) and (5-13). This makes Eqs. (5-29) and (5-30a) considerably more
complicated. These equations have been solved numerically. This solution is
shown in Figure 5-9 (Pfann, 1966) (see also Problem 5-El).
Equation (5-31) and Figure 5-9 show that for k < 1.0 the purest material
is at the starting end of the bar. Separation improves as k and zone length 1
decrease. As the number of passes n increases the separation becomes better.
As n -+ 00 an "ultimate distribution" results. The ultimate distribution is
approached when (Wilcox and Estrin, 1982)
(5-35)
It appears that one would want to decrease Z and do a lot of passes. Unfor-
tunately, I is limited by heat transfer to a minimum of approximately the diame-
ter of the rod. Usually, LIZ is in the range from 4 to 10.
The purified ingot is usually harvested by cutting off the end of the bar.
If some impurities have k > 1, both ends of the rod can be cut off. The result-
ing average composition is
Z2
J xdz (5-36)
Zl
X.vg = (Z2 - Zl)
193
where Zl and Z2 are the locations where the ends are cut off. If all ki < 1 and
the front end is not removed, then Zl = O. For a single pass x is given by Eqs.
(5-31) and/or (5-13). IfZ2 < L -I, the result for a single pass is
(1 - k) (-Z2k11
_Xo [1k (Z2 - ZlkllJ + 1] (5-37)
x avg - _ Zl) e - e
For multiple passes the numerical solution for x (e.g. Figure 5-9) can be substi-
tuted into Eq. (5-36). The integration can then be done numerically. This is
illustrated in Example 5-2.
This equilibrium analysis ignores mass transfer resistance and poor mix-
ing in the molten wne. An effective distribution constant kerr can be used
instead of k to include mass transfer effects. When the film theory for mass
transfer is solved, the value of kerr is (Pfann, 1966),
(5-38)
where 0 is the film thickness, D is the diffusivity of impurity in the melt, and f
is the linear freezing rate in cm/sec. olD is the mass transfer coefficient.
Although 0 is not known it can be measured by experiments. D can also be
measured and usually lies in the range from 10-5 to 10-4 cm 2/sec. The value of
o was estimated as 0.1 cm in a horizontal zone of volume 1 cubic inch when
mixing was caused by natural convection. With moderate stirring 0 was
estimated as 0.01 cm while with vigorous stirring 0 was estimated as 0.001 cm
(Pfann, 1966). These values can be used to estimate the actual separation.
Wilcox (1967) lists experimental values of 0 for both zone melting and normal
freezing. More detailed theories are discussed by Wilcox (1967). Detailed
analyses of the hydrodynamics and mass transfer during single crystal growth
are reviewed by Brown (1988).
194
Example 5-2. Nonnal Freezing and Zone Melting
A bar of tin (Sn) contains 0.002 wt fraction indium (In).
a. Ifnonnal freezing is done and 75% of the tin is frozen, calculate

the average crystal concentration.
b. Zone melting is done with one pass and L/I = to. If the bar is cut
at Z2 = 0.75 L, find the average product concentration.
c. Zone melting is done with ten passes and L/I = to. If the bar is
then cut at q = 0.75 L, estimate the average product concentra-
tion.
Solution
A. Define. Three separate problems as delineated in the problem

statement
B. Explore. Part a is simple use of Eq. (5-14). Part b is use of Eq. (5-
37) since q = 0.75 L < L -I = 0.9L. Part c requires the numerical
integration of Eq. (5-36).
C. Plan. Parts a and b were explained in the Explore step. For part c
the numerical integration can be done in several ways. For example,
with equal increments of size &,
x avg = - 1 ~x [ z = 2
q 1=1
1
U\z &
D. Do It. (a) For nonnal freezing Eq. (5-14) becomes
x avg = ~ [l-(I-g)k] = ~~2 [1-(0.25)°·5] = 0.001333

195
(b) One pass wne melting. Eq. (5-37) becomes,
= [(1 - k)1 (-z2k/1 _ -zlk//) + 1]

Xavg Xo k( ) e e
z2 - zl
Since z2 = 0.75L, Zl = 0, L/I = 10, k =0.5, this is
x = 0 002[ (0.5)(0.1) (e-o· 75 (IO)(0.5) - 1) + 1] = 0 00174

avg . (0.5)(0.75) .
Cc) Ten pass zone melting. If we do a numerical integration with

L\zl = ~Z2 = ~Z3 = .2, and L\z4 = .15, then the numerical calculation is,
Xo z z Z Z
Xavg = 0.75 (0.2[x( 7=1) + xC 7=3) + x( 7=5)] + 0.15x( 7=6·75)}
= (0.002 )[0.2(0.041 + 0.11 + 0.22) + 0.15(0.43)] = 0.00037

0.75
where the x values were detennined from Figure 5-9.
E. Check. The nonna! freezing results can be checked by calculating x

at g = 0.75. Then y = x/k and the check is the mass balance
(0.75 Xavg) + 0.25 y = Xo •
Unfortunately, the check is more work: than the original calculation.

The zone melting calculations are more difficult to check. The
numerical integration can be checked by doing the integration with a
different method. For example, using Simpson's rule (e.g. see Mickley
et al. 1957)
(Z2 - Zl) [ ]
6 X/Zl +4X/(Zl+Z2)/2+x/Z2 (5-39)
Xavg = - - - - - -Z2--Zl
-------
196
we obtain
= (0.75/6)(0.002)[OJ)()21 + 4(0.15) + 0.70] =0 00044
x avg 0.75 .
where 0.0021 is x./X{) at z/l = 0, 0.15 is x/X(j at z/l = 3.75, and 0.70 is
x/xo at z/l = 7.5. This disagrees with the other integration. The reason
for this disagreement is the extreme sensitivity of the calculation to the
amount of impurity near z = 0.75 L. Better agreement would be
obtained by applying Simpson's rule in several steps.
F. Generalization. One perhaps surprising result is that normal freezing

gives a better separation than one pass zone melting. This is generally
true. The two methods become equal when L/l = 1. When the zone
length equals the ingot length, zone melting becomes a normal freezing
method. With additional passes zone melting produces a significantly
better separation than normal freezing. Since the concentrated end of
the bar has a major effect on the average purity level, particularly for
small ki' a small grid should be used for the numerical integration at this
end.
5.4. SUMMARY-OBJECTIVES
This chapter has been a brief introduction to melt crystallization. At the end of
this chapter you should be able to:
1. Discuss the effect of heat and mass transfer in melt crystallization.
2. Do equilibrium calculations for continuous and batch crystallizers

including the effect of progressive freezing.
3. Do equilibrium calculations for partial melting and drainage.
4. Explain column crystallization systems and do mass transfer calculations

for end feed columns.
5. Explain and do equilibrium calculations for normal freezing.
6. Explain and do equilibrium calculations for zone melting.

197
REFERENCES
Albertins, R., W.C. Gates, and J.E. Powers, "Column crystallization," in M.

Zief and W.R. Wilcox, (Eds.), Fractional Solidification, Vol. I, Marcel Dekker,
New York, 1967,343-367.
Alper, A.M. (Ed.), Phase Diagrams, Material Science and Technology,

Academic Press, New York, Vols. 1 to 5,1970-78.
Anon., "CPI warms up to freeze concentration," Chern. Engr., 95, (6),24 (April
25, 1988).
Brown, R.A., "Theory of transport processes in single crystal growth from the
melt," AIChE Journal, 34,881 (1984).
Hanson, M., Constitution of Binary Alloys, 2nd ed., McGraw-Hill, New York,
1958.
Henry, J.D. and C.G. Moyers, Jr., "Crystallization from the melt," in R.H.
Perry, and D.W. Green, (Eds.), Perry's Chemical Engineer's Handbook, 6th
ed., McGraw-Hill, New York, 1984,p. 17-3 to 17-12.
Hudgins, R.R., "Zone Refining. A student experiment," Chern. Engr. Educ., 5,

138, (Summer, 1971).
Keller, W. and A. Muhlbauer, Floating-Zone Silicon, Marcel Dekker, New

York,1981.
Landolt-Bornstein, Zahlenwerte unci Funktionen, Vol. 2, Pts. a,b,c, Vol. 3,

Springer, Berlin, 1956.
Levin, E.M., C.R. Robbins, and H.F. McMurdie, Phase Diagrams for Ceram-
ists, American Ceramic Society, Columbus, Ohio, 1964 and 1969.
McKay, D.L., "Phillips fractional-solidification process," in M. Zief, and W.R.

198
Wilcox (Eds.), Fractional Solidification, Vol. I., Marcel Dekker, New York,
1967,427-439.
Meyer, W. and P.K. Shen, "Design models for separation of eutectic systems
by continuous countercurrent columnar crystallization," AIChE Meeting, Chi-
cago, NOv. 29, 1970.
Moates, G.H. and J.K. Kennedy, "Continuous-zone melting," in M. Zief and

W.R. Wilcox (Eds.) , Fractional Solidification, Marcel Dekker, New York,
1967,273-342.
Molinari, J.G.D., "The Proabd refiner," in M. Zief and W.R. Wilcox (Eds.) ,
Fractional Solidification, Vol. I, Marcel Dekker, New York, 1967a, 393-400.
Moyers, C.G., Jr. and RW. Rousseau, "Crystallization operations," in RW.

Rousseau (Ed), Handbook of Separation Process Technology, Wiley, 1987,
Chapt 11.
Molinari, J.G.D., "Newton Chambers' process", in M. Zief and W.R Wilcox

(Eds.) , Fractional Solidification, Vol. I., Marcel Dekker, New York, 1967b,
401-407.
Mullin, J .W., Crystallization, 2nd ed., CRC Press, Boca Raton, FL, 1972.
Pfann, W.P., Zone Melting, 2nd ed., Wiley, New York, 1966.
Rhines, N.F., Phase Diagrams in Metallurgy, McGraw-Hill, New York, 1956.
Saxer, K. and A. Papp, "The MWB crystallization process," Chern. Eng. Prog.,
76 (4), 64 (1980).
Shoemaker, C.E. and R.L. Smith, "Survey of inorganic materials," in M. Zief

and W.R. Wilcox (Eds.), Fractional Solidification, Vol. I., Marcel Dekker,
New York, 1967,624-648.
Walas, S.M., Phase Equilibria in Chemical Engineering. Butterworth Pub.,

Boston, 1985, Chapt. 8.
199
Wilcox, W.R., "Mass transfer in fractional solidification", in M. Zief and W.R.

Wilcox (Eds.), Fractional Solidification, Marcel Dekker, New York, 1967,47-
112.
Wilcox, W.R., "Fractional solidification phenomena," Separ. Sci., 4,95 (1969).
Wilcox, W.R. and J. Estrin, "Crystallization and precipitation," in P J. Elving,

E. Grushka, and I.M. Kolthoff (Eds.), Treatise on Analytical Chemistry, Part I,
Vol. 5, 2nd ed., Wiley-Interscience, New York, 1982, Chapt. 8,281-370.
Wynne, E.A., "Batch zone melting," in M. Zief and W.R. Wilcox (Eds.), Frac-
tional Solidification, Marcel Dekker, New York, 1967,237-255.
Wynne, E.A. and M. Zief, "Laboratory scale apparatus," in M. Zief and W.R.
Wilcox (Eds.), Fractional Solidification, Vol. 1, Marcel Dekker, New York,
1967,191-236.
Zief, M. and W.R. Wilcox (Eds.), Fractional Solidification, Vol. 1., Marcel
Dekker, New York, 1967.
Zief, M., "Survey of organic materials," in M. Zief and W.R. Wilcox (Eds.)
Fractional Solidification, Vol. I, Marcel Dekker, New York, 1967,649-678.
Zief, M., (Ed.) Purification of Inorganic and Organic Materials, Marcel

HOMEWORK
AI. Construct a table contrasting and comparing solution and melt crystall-
ization.
200
A2. Suppose ki < 1 in Eq. (5-7). What effect does the flow rate of melt, M,
have on the purity of the crystals? What happens in the limit as M =
O? What happens if kt =0 and M =O? Discuss this from both physi-
cal and mathematical viewpoints.
A3. Explain the process of progressive freezing followed by partial melting

and drainage in physical tenns. Why is XAvg,2 given by Eq. (5-20)?
One could calculate Yavg as
g2
J xdg (5-21a)
g2+Bg
Yavg = Bg
where Bg = eg2 instead of using Eq. (5-21). Explain Eq. (5-21a) phy-
sically. Does (Eq. (5-21) or Eq. (5-21a) predict a purer product?
A4. Why do batch slurry crystallization and nonnal freezing give the same
result for crystal composition? What are the advantages and disadvan-
tages of each process?
A5. What are the advantages of zone melting compared to nonnal freez-
ing?
A6. Construct your key relations chart for this chapter.
B1. A series of batch crystallizers can be cascaded to increase both yield

and purity. Explore various ways to do this. Try not to produce any
product streams of intennediate purity. (In other words, produce only
a high purity product and a concentrated waste stream.)
B2. We wish to purify pyrene which contains 0.8 wt % anthracene and 0.2
wt % 1,2-benzanthracene as impurities. A batch nonnal freezing
apparatus is available. Develop a process to do this purification.
201
C. Derivations
C1. Develop the analysis procedure for a staged counter-current cascade

for a binary melt crystallization system which fonns a solid solution.
The cascade will be similar to Figure 2-20. Plot enthalpy (both liquid
and solid) as the ordinate versus mole fraction in both the liquid and
solid phases. This is similar to Figure 2-21 with enthalpy replacing N.
A plot of mole fraction in the liquid versus mole fraction in the solid (a
McCabe-Thiele plot) is useful to obtain equilibrium tie lines. Develop
the ~ points using two mass balances and the energy balance.
C2. Single stage, batch crystallization can be treated in the same way as
simple Rayleigh batch distillation. That is, assume that crystals are
removed as soon as they are fonned. Set up and solve the equation
equivalent to the Rayleigh equation. Assume that y = x/k. Show that
the results are the same as Eqs. (5-13) and (5-14).
C3. The mass fraction impurity of the crystals for a single stage, steady
state, continuous crystallizer is given by Eq. (5-6). a). If a weight
fraction, e, of melt is entrained, find the average product concentration.
b). If some fraction of the crystals are melted (fraction> e) and the
excess liquid is drained, find the new product concentration.
C4. Derive Eq. (5-28b). Explain why I is defined as the length of solid
melted to fonn the zone instead of as the length of the heater.
C5. Derive Eq. (5-37).
D. Problems
D 1. A continuous single-stage, steady-state, equilibrium crystallizer is

crystallizing 2-naphthol containing 0.007 wt frac naphthalene. Data in
Table 5-1.
a. If we crystallize 60% of the feed, calculate the concentration of

the crystals and the concentration of the melt. Assume there is no
entrainment.
202
b. IT 9% by weight of melt is entrained with the crystals calculate

the average product concentration.
D2. One hundred kilograms of a benzene mixture containing 0.45 wt %
cyclohexane is purified in a batch crystallizer. We freeze 72% of the
original charge and filter the crystals. Entrainment is 0.06 wt fraction.
Data are in Table 5-1.
a. What is the product concentration?
b. IT we melt 15% of the crystals and drain off the melt, find the new
product concentration. Entrainment is unchanged.
D3. We are purifying naphthalene in a commercial batch crystallizer that
functions as a normal freezer. The naphthalene charge is 1000 kg and
is 99.87 wt % naphthalene. The impurity is unknown. IT one half of
the initial charge is frozen and the crystals are sampled (after centri-
fuging to remove melt), we find the average weight fraction is 99.93
wt % naphthalene. IT we melt off and drain 125 kg of crystals, we find
that the purity of the product (including entrained melt) increased to
99.932 wt % naphthalene. Find ~ and e.
04. We wish to purify pyrene containing 0.005 wt fraction anthracene and
0.005 fraction 1,2-benzanthracene.
a. If we freeze 70% of the initial charge in a batch stirred tank, and
collect the crystals plus entrained melt (assume e = 0.09). what is
the product concentration?
b. IT we freeze 70% of the initial charge, remove excess melt, and
then melt and drain off another 15% (~g = 0.15) of the melt. what
is the concentration of the collected product (crystals plus
entrained melt) for e = 0.09?
c. The product from part b is melted and then reprocessed by freez-

ing 30% of the product and collecting crystals plus entrained melt
(e = 0.09). What is the concentration of the non-entrained melt?
What fraction of the original feed is collected as non-entrained
melt?
Equilibrium data is in Table 5-1.
203
D5. We are purifying an unknown impurity from azobenzene in a screw-

type, end-feed, counter-current column crystallizer. The azobenzene
system is a eutectic system (crystals are pure azobenzene). The HTU
was estimated as 12.3 cm and entrainment as e = 0.12. M/C = 0.13.
The crystals enter after leaving a crystallizer where the melt contains
0.009 wt fraction impurity. The final product (crystals plus entrained
melt) should contain 0.00002 wt fraction impurity. The final product
is melted and some of this is refluxed. Find the crystallizer length
required.
D6. Repeat Parts a and b of Example 5-2 except do not assume equili-
brium. For the molten metal assume D - 3x1o-5 cm 2 /s. Calculate the
separation for natural convection (0 - 0.1 cm), moderate stirring (8 =
.01 cm) and vigorous stirring (0 = 0.001 cm). The linear rate of move-
ment of the zone is f = 2.5 crn/hr.
D7. We wish to use zone melting to purify 99.8% pure (wt. %) anthracene
containing fluorene as it's only impurity. A zone melting system with
the zone length 1/10 of the ingot length is used. Data are in Table 5-1.
a Mter 6 passes, 20% of the ingot is cut off and discarded. Calcu-
late the average composition of fluorene in the ingot.
b. Predict the weight fraction fluorene at z!L = 0.95 after one pass.
D8. Zone melting is done to purify an aluminum bar containing 0.003 wt
frac scandium (Sc) as an impurity. The zone melting has L/l = 10.
After n passes we cut off 20% of the rod. We desire an average
impurity weight fraction of 0.00003 or less. How many passes are
required? What is the actual average wt frac impurity in the bar?
What are the highest and lowest impurity wt fracs in the bar? Assume
equilibrium operation. Use Figure 5-9 to estimate the solutions. Data
are in Table 5-1.
El. Repeat Problem 5-D8 except solve Eqs. (5-28) to (5-34) and Eq. (5-36)
by numerical integration for L/I = 10 and k = 0.17. Assume the molten
204
zone is completely mixed. Note that the boundary condition equivalent

to Eq. (S-30b) for passes 2 to n should include melting a length I of the
bar, mixing the zone, and crystallizing a solid at z = O. Thus boundary
condition is
Xj+l = k Xj.lvgfromz=Otoz=l z=o

where j is the number of the pass.
FI. A laboratory zone melting apparatus is being used to ultrapurify

naphthalene. The apparatus is 1 cm wide and 1 cm deep. The heater
moves at 1.1 cm/hr. The melt temperature is measured to be 84.SSoC.
Estimate the overall heat transfer coefficient U between the solid and
the melt
Part II
SORPTION AND CHROMATOGRAPHY

NOMENCLATURE CHAPTERS 6 TO 10
a, b constants in Langmuir Eqs. (6-12) and (6-13) or constants in Eq.

(6-15) or (6-18).
activities, Chapter 9.
3p external surface area/volume, m2 /m 3
A van Deemter eddy dispersion term, Eq. (7-39)
A(T) equilibrium parameter, Eq. (6-4) or (8-17b)
Ac cross sectional area of column, m2
Anole cross-sectional area of hole in sieve plate
preexponential constant, Eq. (64b)
wall area for heat transfer, m2
b equilibrium constant, Eq. (10-15)
b,b' constants in quadratic Eqs. (9-22) and (9-24)
B van Deemter molecular diffusion term, Eq. (7-40)
c solute concentration of fluid, kg/m 3
c* equilibrium concentration
concentration after wave has passed
Cchange concentration changed by thermal wave, Eq. (6-39b)
equlibrium concentration
CF feed concentration
~.pore concentration i in pore
normality, equivalents/m 3 in solution, Chapter 9
Cmax maximum concentration in peak
207
208
cR. concentration on resin, equivalents/m3 , in Chapter 9
Or total equivalents/m 3 in solution, Chapter 9
C van Deemter mass transfer term, Eqs. (7-41)
CP.f heat capacity of fl uid
Cp,s heat capacity of solid
Cp,w heat capacity of wall
d* ~/dm, Eq. (1O-26b)
de coating thickness
<\, particle diameter, m
D column diameter, m
D eff effective diffusivity, Eqs. (9-33) and (9-34)
Dm molecular diffusivity, cm 2 /s
D mp molecular diffusivity in pore
D. surface diffusivity
Dr thermal diffusivity
e volume fraction particles in fluidized bed
erf error function, Eqs. (7-11)
ED eddy diffusivity, cm 2 /s
EDT thermal eddy diffusivity
F feed rate, L/min
F Faraday's constant, 96500 amp·sec/equiv
Fi moles of feed in pulse
g acceleration due to gravity

209
G fluid flow rate, kg solute-free fluid/hr
G mass velocity of gas
h heat transfer coefficient across film, mls
h height of packing, m
lumped parameter particle heat transfer coefficient
wall heat transfer coefficient
H height equivalent to a theoretical stage, m
lITU height of transfer unit, Eq. (lO-12c)
~H heat of adsorption
j stage number
total flux
flux due to electrical transference, Eq. (9-28)
diffusive flux
k' relative retention or capacity factor, Eq. (7-32a)
rate constants for adsorption and desorption in Langmuir expres-

sion, Eq. (6-5).
rate constants, Eq. (7-35)
constant, Eq. (6-5)
effective thermal diffusivity
film diffusion coefficient, mls
lumped parameter mass transfer coefficient
surface diffusion coefficient
K constant in pressure drop Eq. (8-23)

210
K true equilibrium constant in Eqs. (9-8) and (9-13).
K equilibrium constant in BET Eq. (6-10)
= kdk..l equilibrium constant in Langmuir expression, Eq. (6-6).
preexponential factor, Eq. (6-8)
KAB equilibrium constant or selectivity coefficient, Eqs. (9-10) and (9-

11).
fraction of interparticle volume species can penetrate, Eq. (6-3).
KDB equilibrium constant for divalent-monovalent exchange, Eqs. (9-

14) and (9-16b).
exclusion factor. Zero for excluded ions and 1.0 for non-excluded,
Chapter 9.
K1 Langmuir equilibrium constant for multicomponent system
length of travel of pulse, or packing height between ports, m
L column length, m
LMfZ length of mass transfer zone in column, m
m linear equilibrium constant
1/m constant in LRC Eq. (6-11)
mass of stationary phase per stage
n number of adsorbates
n exponent in Eq. (1O-32a)
n constant in Eq. (7-44)
n moles of gas
1/n exponent in Freundlich isotherm, Eq. (6-14)
N loading of molecular sieve, kg/kg adsorbent

211
N number of stages
maximum loading
Peclet number based on <ip, Eq. (7-14b)
NTU number of transfer units, Eq. (lO-12c)
p pressure, kPa.
P partial pressure of solute, kPa.
PH high pressure
PI- low pressure
axial Peclet number, Eq. (7-12a)
product flow rates, m3/min or kg/min
q amount of solute adsorbed, kg/kg adsorbent
CIa amount adsorbed after wave has passed
amount adsorbed from pure gas
= q (Rp), surface adsorbate loading

average amount adsorbed, Eq. (6-43b)
q* equilibrium amount adsorbed
~x monolayer coverage of adsorbent in Langmuir isotherm, Eq. (6-6),

kg/kg adsorbent
'lmono amount adsorbed for monolayer coverage
Chat saturation capacity of adsorbent
Q volumetric flow rate, m3/min
r radial direction
R gas constant
R resolution, Eq. (7-30)

212
(LILMrz;)new/(L(LMIZ)old
~Pnew/~Po1d
pore radius, m
RT crossover ratio, Eq. (8-51)
Re Reynold's number, Ee Pr v<lp/ll
S solids flow rate, kg clean adsorbent plus fluid in pores/hr
Sc Schmidt number, J.1/Pr Dm
Sh Sherwood number, kf~/Dm
t time, min
tA,band bandwidth in displacement chromatography, min
breakthrough time, Eq. (8-2)
time center of breakthrough curve exits
period of feed pulse, min
time for mass transfer zone to exit column, min
retention time of non-retained peak, min
time between switching port locations, min
retention time, min
temperature, °C or K
equilibrium temperature
T, Ts average temperatures, Eqs. (6-52)
T., Tb temperature after and before wave
T e , TH cold and hot temperatures
Tf , Ts temperature of fluid and solid

213
Tw. Trer temperature of wall and reference
UA, UB, Us solute wave velocity, m/min, Eqs. (6-23) or (6-24) .
llco-ion velocity of coion, Eq. (9-26b)
~ ionic mobility, Eq. (9-29)
UNe velocity of non-charged species, Eq. (9-26a)

lip pattern velocity
llpwse average pulse velocity, Eq. (10-10)
llport.lvg average velocity port movement, Eq. (10-23)
Us.cc observed countercurrent solute velocity, Eq. (10-19)
Usb shock wave velocity m/min, Eq. (6-30).

Ustm velocity of steam wave, Eq. (8-52)
lltb thennal wave velocity m/min, Eq. (6-36).
lltotal ion ion wave velocity, Eq. (9-20).
v interstitial fluid velocity, m/min
v* v/vrer, Eq. (1O-26b)
Vbo!e velocity through hole in sieve plate
Vmf minimum fluidization velocity, m/min
vsolid superficial solid velocity, m/min
vsuper superficial fluid velocity, m/min
V volume solution fed to column, m3 or L.
V volume solution required to saturate column, Eqs. (7-12c) and (9-

27b)
Ve elution volume of non-adsorbed species, m 3 or L.
Vec,feed extracolumn volume on feed side, m3 , Eq. (8-32)

214
V Feed volume feed gas, m3
VF volume feed pulse
Vi internal void volume, m 3 or L
Vo external void volume, m3 or L
VM volume mobile phase/stage
Vpeak volume peak exits
Vpurge volume purge gas, m3
VR volumetric flow rate resin, L/hr
Vs volumetric flow rate solution, L/hr
Vp vapor pressure
width of species i peak
w weight of column wall per length, kg/m
see Eq. (7-47)
mole fraction in solid phase, Eq. (6-16),
equilibrium mass fraction in fluid
cJVr, equivalent fraction ions in solution or CA /CAF in adsorption
general breakthrough solution
mole fraction in gas phase, Eq. (6-16)
equilibrium mass fraction on solid
CR,JCR,tot, equivalent fraction ions on resin, Chapter 9.
Ystm mole fraction water in steam
Yw,b water mole fraction in vapor before steam wave

y mass ratio solute in fluid, kg solute/kg solute-free fluid
215
z axial distance in column, m
~ valence of ion
Greek
a.AB separation factor, Eq. (6-16)
0.21 A2 /A1 , selectivity
f3A Eqs. (8-58)
f3i empirical constant, Eq. (6-15)
Y VpurgeNf=l in PSA
Yi activity coefficients, Eq. (9-9)
!l. difference calculation
Ee external void fraction (between particles), Eq. (6-1a)
Ep intraparticle void fraction (within a particle), Eq. (6-1b)
£T total porosity, Eq. (6-1c)
e fraction of surface covered by adsorbed solute
e = t-z/v, Eq. (7-47).
Aw latent heat of vaporization of water
J..l viscosity
v see Eq. (7-53)
PB bulk particle density, kg/m 3 , Eq. (6-2a)
Pf fluid density, kg/m 3
Pp particle density including fluid in pores, kg/m 3 , Eq. (6-2b)
Ps structural solid density kg/m3

t
a standard deviation, Eqs. (7-27).

216
0'1>0'2 Eqs. (7-37)
't tortuosity factor
't see Eq. (7-53)
"C see Eq. (8-6)
cp electrical potential
<I>i,j parameter for calculation of mixture viscosity, Example 8-1
X see Eq. (7-53)
X volumetric fraction of swelled resin taken up by water

Chapter 6
BASICS OF SORPTION IN PACKED COLUMNS
What phenomena are involved in adsorptive and chromatographic

separations? How do these phenomena combine to produce the desired separa-
tion? How can we simply predict the separation which will occur? In this
chapter we will try to answer these and other questions.
Adsorption is a separation technique which uses a solid (the adsorbent)

to adsorb a solute or adsorbate on its surface. Highly porous adsorbents with
large surface areas are used. Usually, the adsorbent is held in a packed column
as illustrated in Figure 6-1. This is a batch process consisting of a feed step fol-
lowed by a desorption or regeneration step. The regeneration is done by
increasing temperature, or dropping prt".ssure, or using a chemical desorbent.
Adsorption is used commercially for a wide variety of separations such as
hydrogen purification, production of oxygen and nitrogen from air, drying
gases and organic liquids, removal of pollutants from air and water, and
purification of biochemicals. Various adsorption processes are discussed in
detail in Chapter 8.
Chromatography is a similar separation process which is used for frac-

tionating feeds into several components. A pulse of feed is injected into the
column and the products are eluted with a carrier gas or solvent. The result is a
series of separated peaks (for example, see Figure 7-1). Chromatography is
most commonly used as an analytical method, but large scale chromatography
is becoming a more common commercial method for sugar separations, in
biotechnology and for purification of pharmaceuticals. Chromatographic
separations are explored in detail in Chapter 7.
Ion exchange is a process where ions in solution are exchanged for ions
held to a resin. The best known application is water softening where calcium
217
218
Feed
Distributor
~- Holddown Plate
} Po,k,d S."ioe
~ Support Plate
Product
Figure 6-1. Packed bed system.
and magnesium ions are removed and replaced with sodium ions. A variety of
other ion exchange separations are also done commercially. The usual opera-
tion is done in fixed beds, and follows a cycle of load, regenerate and wash.
Ion exchange is the subject of Chapter 9.
Adsorption, chromatography and ion exchange are also done commer-

cially in moving beds, staged systems and simulated moving beds. These
applications are explored in Chapter 10.
6.1. PACKING STRUCTURE
Commercially significant sorption operations (including adsorption, chroma-

tography, ion exchange, ion exclusion, etc.) use solid particles (sorbents) which
are highly porous and have large surface areas per gram of sorbent. The sor-
bent particles are commonly packed in a column as illustrated in Figure 6-1. In
general the particles will be of different sizes and shapes. They will pack in the
column and have an average interparticle (between different particles) porosity
of Ee. Since the particles are usually porous, each particle has an intraparticle
219
(within the particle) porosity fp which is the fraction of the particle which is
void space. The interparticle or extra particle porosity, Ee , is defined as
volume between particles (6-1a)

Ec = total volume of packed bed
while the intraparticle or particle porosity, Ep, is
volume of fluid inside all particles (6-1b)

fp = total volume of all particles (solid plus fluid)
The total porosity, ET, of the bed is the sum of the voids within the particles and
between particles.
(6-1c)
Note that all three porosities are dimensionless. The different porosities are
shown in Figure 6-2.
In a poorly packed bed Ee may vary considerably in different parts of the

column. This can lead to poor flow distribution or channeling and will
decrease the separation. If the solid particles are uniform, fp will be the same
for all particles. Gel beads used for ion exchange have fp = 0 while 3 porosi-
ties can be defined for molecular sieves and some activated carbons.
Sorbents are provided in bulk. The bulk density PB including the fluid in
the pores and the fluid between particles is easy to determine. The bulk density
.--------
Ac
Figure 6-2. Porosites in packed bed.
220
is related to the particle density pp which includes the solid plus fluid in the
pores.
(6-2a)
The particle density can be estimated from the structural or crystalline density
PI which is the density of crushed and compressed solid containing no pores.
(6-2b)
pp =(1-t;,)PI + t;, PI
When supplied in bulk the fluid is usually air. Thus the fluid density, Pf, terms
are often ignored in estimating pP and PI from I'D.
The pores are not of uniform size. Large molecules such as proteins or
synthetic polymers may be sterically excluded from some of the pores. The
fraction of volume of pores which a molecule can penetrate is called Kd • For a
non-adsorbed species, Kd can be determined from
(6-3)
where Ve is the elution volume, (that is the volume of fluid at which the species
exits from the column), Vo is the extemal void volume between the particles,
and Vi is the internal void volume. When the molecules are small and can
penetrate the entire interparticle volume, Ve = Vi + Vo and ~ = 1.0. When the
molecules are large and can penetrate none of the interparticle volume, Ve = V0
and Kd =O. Exclusion phenomena are used commercially for separations.
Examples are listed in Table 6-1. The zeolites are discussed in Section 6-2.
Sephadex is a cross-linked dextran gel, Biogel P is a cross-linked polyacrylam-
ide, Sepharose and Bio-gel A are cross-linked agarose gels. These are dis-
cussed in Chapter 7 and in more detail by Janson and Hedman (1982) for size
exclusion chromatography (SEC). Styragel is a cross-linked polystyrene
(Bidlingmeyer and Warren, 1988).
For the system shown in Figure 6-2, fluid containing solute flows in the
void volume outside the particles. The processes which occur during a separa-
tion are as follows: (1) The solute diffuses through an external film to the parti-
cle. Here the solute may (2) sorb on the external surface or (3) (more likely)
Table 6-1. Separations using Sterie Exclusion
(Collins, 1968; Janson and Hedman, 1982; Bidlingmeyer and Warren, 1988).
Packing Molecules Molecules Typical
Included Excluded Applications
3A HzO,NH 3 ,He CI4, COz , CzHz, OZ, Drying cracked gas,

Zeolite diameter <3 ~ CzHsOH, HzS ethylene, butadiene,
diameter >3 ~ and ethanol
4A HzS,C0 2, C3 H6 , C 3 Hg, compressor oil Drying natural gas,

Zeolite C2HsOH,C4~ diameter >4 ~ liquid paraffins, and
diameter <4 ~ solvents. C02 removal
5A n-Paraffins, n-olefins, Iso-compounds, all 4 n-Paraffin recovery

Zeolite n-C 4 H 9 OH + carbon rings from naphtha and
diameter <5 ~ diameter >5 ~ kerosene
lOX Iso-paraffins, iso- Di-n-butylamine and Aromatic separation

Zeolite olefins larger
diameter <8 ~ diameter >8 ~
13X Di-n-butylamine (C 4 F9h-N Desulfurization,

Zeolite diameter <1 ~ ° diameter> 1 ~ ° simultaneous HzO
and CO2 removal
Sephadex G-25 Salts, low MW compds Proteins and Desalting-SEC

<1000 to 5000 compds >5000
Bio-Gel P6 Salts, low MW compds Proteins and Desalting-SEC
<1000 to 6000 compds >6000
Sephadex G-50 <1500 to 30,000 >30,000 Insulin-SEC
Sephadex G-150 <5000 to 300,000 >300,000 Protein fractionation
Sephadex G-200 <5000 to 600,000 >600,000 Protein fractionation
Bio-Gel P-200 <30,000 to 200,000 >200,000 Protein fractionation
Bio-Gel P-300 <60,000 to 400,000 >400,000 Protein fractionation
Sepharose 2B <7x104 to 4.107 >4x107 Protein fractionation

Sepharose 4B <6x104 to 2x107 >2x107 Protein fractionation
Sepharose 6B <104 to 4x106 >4x106 Protein fractionation
Bio-Gel A-1.5m <Hf to 1.5x106 >1.5x106 Protein fractionation
Styragel <100 to 2000 >2000 SEC-Analytical

222
diffuse in the stagnant fluid in the pores. Ifthe pores are tight for the solute this
diffusion will be hindered. (4) The solute finds a vacant site and then sorbs by
physical or electrical forces or by a chemical reaction. (5) While sorbed the
solute may diffuse along the surface. This is called surface diffusion. (6) The
solute desorbs, (7) diffuses through the pores, and (8) diffuses back across the
external film and into the moving fluid A given molecule may sorb and desorb
many times during its stay inside a single particle.
While in the moving fluid the solute is carried along at the average inter-
stitial fluid velocity, v, until the solute diffuses into another particle and the
whole process is repeated. As far as migration down the column is concerned,
the particle is either moving at the interstitial velocity, v, of the fluid or it has a
velocity of zero when it is inside a particle. The average rate of movement of
the solute depends on the relative amount of time the solute spends inside parti-
cles. Thus the solute movement depends on Ec" £p, Kc!, v, and the sorption
equilibrium.
6.2. ADSORBENTS
The most commonly used adsorbents are activated carbon, zeolite molecular
sieves, sliea gel, organic polymers and activated aluminas. In chromatography
a variety of specialized packing materials are used. These are discussed in
Chapter 7. The polymeric resins used for ion exchange are discussed in
Chapter 9.
Based on tonnage used and market value activated carbon is the most
commonly used adsorbent. Activated carbon is made by first carbonizing a
carbonaceous starting material such as wood, coal, petroleum coke, or coconut
shells. The material is then partially gasified in a mild oxidizing gas such as
CO2 or steam. This activation process creates the desired porosity and surface
area, and oxidizes the surface. Different grades of activated carbon for both
gas and liquid separations are made (Bonsol et ai, 1988). Activated carbon
properties are listed in Table 6-2. Properties of a variety of commercially
available carbons used for water treatment are tabulated by Faust and Aly
223
Table 6-2. Properties of Activated Carbon Adsorbents

(Keller et ai, 1987; Lydersen, 1983; Ruthven, 1984,
Venneulen et al .. 1984; Yang, 1987)
Property Liquid-Pha<;e Vapor-Pha<;e

Wood Base Coal Base Coal Base
Mesh Size (Tyler) -100 -8 + 30 -6 + 14
Bulk density Pn (kg/m 3 ) 250-300 400-700 500-530
500 (avg)
Ash (%) 7 8 4-8
Ep -0.8 0.6-0.85 0.6-0.85
Pore vol. cm 3/g 0.15 to 0.50
Surface area (km 7 /kg) 0.8-1.8 0.9-1.2
Tortuosity, 't (Eq. 6-44b) 5-65
Adsorption Capacity, Vapor-phase carbon:
CC1 4 0.4 to 0.7 kg adsorbate/kg
nC4 (250 mm Hg, 25°) 25 wt%
water (250 mm Hg, 25°C) 5-7 wt %
(1987). Activated carbon is used for water and air purification, gas masks, sol-
vent recovery, pollution control, decolorizing sugar, gold recovery. alcohol
purification, adsorption of gasoline vapors from automobiles and a host of other
applications.
Activated carbon has an essentially nonpolar surface. Thus, it does not

strongly adsorb water. Because of this, activated carbon is commonly used for
aqueous separations and activated carbon can be used to process humid gases.
Water does adsorb, but it is by capillary condensation and is rather weak.
Since activated carbon has a very large internal surface area, it usually has a
higher capacity for nonpolar and weakly polar organics than other adsorbents.
However, since the pores are smaIl, KD < 1 for relatively large molecules and
the diffusion rates may be very low (it is not unusual to take 30 days for
activated carbon to come to equilibrium). The maximum capacity of activated
carbon for a variety of organics is listed in Table 8-3. Finally, since the heat of
224
Table 6-3. Commercially Available Zeolites
(Not all are used as adsorbents). (Vaughan, 1988).
Reprinted with permission from Chern. Engr. Prog., 84 (2), 25 (Feb. 1988).
Copyright 1988, Amer. lnst. Chern. Engrs.
Pore Composition Capacity (wt. %)
Zeolite Size Si/Al Cation H2O nC6 H14 C6 H12 Vendors
Faujasite
X 7.4'A 1-1.5 Na 28 14.5 16.6 GLPTU
Y 7.4 1.5-3 Na 26 18.1 19.5 GLPTU
US-Y 7.4 >3 H 11 15.8 18.3 GLPTU
A 3 1.0 K,Na 22 0 0 GLPTU
A 4 1.0 Na 23 0 0 GLPTU
A 4.5 1.0 Ca,Na 23 12.5 0 GLPTU
Chabazite 4 4 °N° 15 6.7 GU
Clinoptilolite 4,0 5.5 °N° 10 1.8 0 A
Erinoite 3.8 4 °N° 9 2.4 0 AM
Ferrierite 5.5.4.8 5-10 H 10 2.1 1.3 T
Ltype 6 3-3.5 K 12 8 7.4 LTU
Mazzite 5.8 3.4 Na,H 11 4.3 4.1 U
Mordenite 6.7 5.5 °N° 6 2.1 2.1 AU
Mordenite 6.7 5-6 Na 14 4.0 4.5 PTU
Mordenite 6.7 5-10 H 12 4.2 7.5 PTU
Offretite 5.8 4 K,H 13 5.7 2.0 U
Phillipsite 3 2 °N° 15 1.3 0 A
Silicalite 5.5 00 H 10.1 0 U
ZSM-5 5.5 10-500 H 4 12.4 5.9 M
A = Anaconda Minerals, Letcher Minerals, Double Eagle mining, and others.

G = W.R. Grace & Co.; L = Laporte PLC; M = Mobil Oil Co.; P = PQ Corp.; T = Toyo Soda; U =
Union Carbide Corp.
°N° = mineral zeolite; cations variable, and usually Na, K, Ca, Mg.
Note: Many small companies will mine and supply mineral zeolites on a custom basis as ground or
sized rock, but they offer no process or fabrication services.
225
adsorption is modest, the energy requirements for regeneration are often mod-
est. However, some adsorbates such as large organics are almost impossible to
desorb and must be burned off.
A new type of carbon adsorbent is the carbon sieve. Carbon sieves differ
from activated carbon since the pores are smaller and the pore distribution is
more tightly controlled. Because of the tightly controlled pore size, gas
molecules may have widely varying diffusion rates. Carbon sieves are used
commercially for separating oxygen and nitrogen based on the different diffu-
sion rates (see Ruthven, 1984; or Yang, 1987). This is of interest since it is
apparently the only rate based adsorption separation being used commercially.
Zeolite molecular sieves can also separate molecules based on size

although the most common applications are based on differences in adsorption.
Zeolites are porous crystalline aluminosilicates of general formula
Cx/n [(Al 02)x (Si02 )y] zH20

In this formula x, y, n, z are integers and y ~ x. n is the valence of cation C
while z is the number of water molecules per unit cell. By changing x, y and
the cation C a very large number of zeolites have been synthesized (Breck,
1974, Ruthven 1984,1988; Vaughan, 1988; Yang, 1987). A number of natur-
ally occurring zeolites are also used commercially. These are listed in Table
6-3 (Vaughan, 1988) while other physical properties are listed in Table 6-4.
Zeolites are useful and in some ways unique since a rigid, three-
dimensional cage structure with exact dimensions is formed. Molecules which
are too large are excluded and do not adsorb. The most common separations
which utilize steric effects are given in Table 6-1. The 3A zeolites are used for
drying reactive gases since the reactive gas is excluded from the zeolite (Zeol-
ites are also used as catalysts). The water is not excluded and is strongly
adsorbed. The structural details of the zeolites are discussed in detail else-
where (Breck, 1974; Ruthven 1984,1988; Yang, 1987).
Zeolites are also used for a variety of separations where the steric proper-
ties are not used. Examples would include the following:
Drying air to low dew points (see Table 8-2 and Figure 8-11)
226
Table 6-4. Physical Properties of Zeolite Adsorbents

(Keller et al.• 1987; Kovach, 1979; Ruthven, 1988;
Vermeulen et. al.• 1984; Yang, 1987)
Nominal Equil. Largest

Zeolite Pore size HzOcap. Molecule
Type (nm) PD, kglm 3 £p (wt %) included
3A 0.3 670-740 0.30 20-23 Hz, HzO

4A 0.4 610-720 0.32 22-28.5 Cz~, Xe
5A 0.5 600-720 0.34 21.5-28 CF4 , n-paraffins
lOX 0.8 641 28-36 isoparaffins,
benzene
13X 0.8-1.0 580-710 0.38 28.5-36 <1O~
Mordenite 0.3-0.8 720-800,880 12
Chabazite 0.4-0.5 640-720 20
Tortuosity, 't, Eq. (6-44b) varies from 1.7 to 4.5
Drying natural gas to low dew points

Drying liquids (see Figure 8-14)
COz removal
HzS removal
Separation of alcohol and water
Sugar separations
Separation of Oz from N z in air
Separation of xylenes and ethylbenzene
Static desiccant in packaging and homes
The zeolite adsorbents can be fine-tuned for a particular separation by changing

the structure and the cation.
Silica gel is an amorphous solid made up as spherical particles of col-

loidal silica, SiOz . It is produced from the following reaction
NazSi03 + 2 HC! + n HzO~2 NaCI + SiOz • n HzO + HzO

227
Table 6-5. Physical Properties of Silica Gel.

(Keller et al., 1987; Kovach, 1979;
Lydersen, 1983; Vermeulen et al., 1984; Yang 1987)
Particle Size 0.1 to 3.0mm

Bulk density 400 to 830 kg/m 3
(700 to 820 common)
Ep 0.38 to 0.55
Pore diameter (nm) 1-40 (depending on grade)
2-5 and - 14 are common
Surface area m2 /g 400-830 (800 common)
Adsorption Capacity 0.3 to 0.6 kg HzO/kg
Reactivation Temp 130-280°C
Tortuosity, 't, Eq. (6-44b) 2-6
The silica surface has a high affinity for water and the most common use of sil-
ica gel is for drying gases and liquids (see Table 8-2 and Figure 8-6). Silica gel
is cheaper and easier to regenerate than zeolites (see Table 8-1), but cannot
produce as dry a gas particularly at high temperatures. Since silica gel can be
damaged by liquid water, it is important to avoid the presence of liquid water.
The physical properties of silica gels are listed in Table 6-5. Other properties
are shown in Figure 8-11 and are listed in Table 8-2.
Activated alumina, AlZ0 3 , is made by dehydrating or activating alumi-

num trlhydrate. The surface has a strong affinity for water, but the alumina is
not harmed by immersion in liquid water. The most common application is
drying both gases and liquids. The alumina surface can be modified to pre-
ferentially adsorb other compounds. At high water concentrations adsorbent
loadings are higher than for zeolites. Physical properties are listed in Table 6-6
while other properties are given in Table 8-2 and Figure 8-11.
Organic polymer resins have been used for many years for ion exchange.
More recently these resins in uncharged form have been used for adsorption of
organics dissolved in water (Fox, 1985). In this application the organic poly-
mers compete with activated carbon. Although more expensive, the polymers
228
Table 6-6. Physical Properties of Activated Alominas

(Keller et al., 1987; Kovach, 1979;
Lydersen, 1983; Vermeulen et al., 1984; Yang, 1987)
Particle Size 14/28 mesh to 1/2 in spheres

Bulk Density 720-880 kg/m3 (800 avg)
fp 0.50-0.57
Pore diameter (nm) 1-14 depending on grade
Surface area (m 2/g) 200-390 (320 avg.)
Adsorption Capacity 0.07 to 0.25 kg H 2 0/kg
Regeneration Temp. ISO-315°C
Tortuosity, 't, Eq. (6-44b) 2-6
appear to be economical for treating concentrated waste streams. Resin details

are discussed in Chapter 9.
A variety of other adsorbents are used for special cases. These include
clays for purification of vegetable oils, calcium silicate for fatty-acid removal,
activated bauxite which is similar to activated alumina, Fuller's earth which has
the same uses as clays and used to be the most commonly used adsorbent, and
bone char which is used in sugar refining. There are also a variety of
proprietary adsorbents which have been developed for very particular applica-
tions such as adsorbing traces of mercury vapor from air.
6.3. ADSORPTION EQUILIBRIUM
Since most adsorption processes are based on differences in the equilibrium

adsorption of different chemicals, adsorption equilibrium is extremely impor-
tant. Solute adsorbed on the solid surface is usually assumed to be in equili-
brium with the solute in the fluid contained in the pores. At equilibrium the
adsorbed solute concentration, q, is related to the solute concentration in the
fluid, c. Usually q is determined in units such as moles/kg adsorbent or kg/kg
adsorbent while c is in units such as moles/m3 solution or kg/m3 solution.
229
Equilibrium is strongly affected by temperature. Constant temperature equili-

brium isothenns are usually measured.
Many fonns of equilibrium isothenns have been developed. The sim-

plest is the linear or Henry's law fonn
(6-4a)
q =A(T)c
where A(T) is the equilibrium constant which usually follows an Arrhenius

function of temperature.
(6-4b)
A(T) = Ao exp (-dH/RT)
The linear fonn is shown in Figure 6-3a. The linear form is often approximated
at very low concentrations. This fonn is extremely convenient to use in
theories and for this reason has been used where the data shows some curva-
ture. Equation (6-4b) shows how adsorbents can be thermally regenerated. At
higher temperatures less adsorbate is bound to the surface. The desorbed
material must concentrate in the fluid and hence can be swept out of the bed.
A second isothenn form which is commonly used is the Langmuir isoth-

erm. This isothenn is appealing because a simple physical picture can be used
to develop the isothenn. Assume that a gas is being adsorbed on a solid surface
and that a maximum of a monolayer of this gas can be adsorbed. The number
of molecules striking the surface is proportional to the partial pressure of the
solute, p. Only the molecules striking a bare surface can adsorb and not all of
these will adsorb. Thus the rate of adsorption is
Adsorption rate = kl (1- 9) P
where 9 is the fraction of surface covered by adsorbed solute and (1 - 9) is the

fraction of surface which is bare. kl is an adsorption rate constant. The rate of
desorption is assumed to depend on the amount adsorbed but not on the con-
centration in the gas phase.
Desorption rate = k...l 9

230
6 I
I
I
I
-15'3· :I I
0·0· :
,/ - -
x_x_-_x_x~
+20·0·
01
"C
GI
4
/
xl
.c
'-
5l j Adsorption of ethyl chloride on charcoal (wt=10gm)
(after Goldman and Po/anyil
"C
IV Constants: n = 1
III
E,-EL= 1780 cal
::: 2
vm =5'14gm
l:
o 100 200 300

Pressure (mm)
Figure 6-3. Isothenn shapes. a. Linear. b. Langmuir isothenns for ethyl

chloride on charcoal (Brunauer et ai, 1938). c. BET isothenns
for nitrogen and argon (Brunauer et ai, 1938). d. Freundlich.
e. Exchange adsorption for ethylene-propylene on silica gel
(Lewis et ai, 1950b). f. Gas-liquid chromatography. Items b
and c reprinted with pennission from 1. Am. Chern. Soc .• 60,
309 (1938). Copyright 1938, American Chemical Society.
Item e reprinted with permission from Ind. Eng. Chern .. 42,
1319 (1950). Copyright1950, American Chemical Society.
231
c 500~--~--~------------~----r----n
Adsorption of Nitrogen and Argon

on Fe- AIi>3 Catalyst 954
Wt.= 50·4gm.
400~--+---~--~---+---4----~--1H
x Aat -195'8 0
a.. vm=128'4cm 3
....
\/I Er E L=700cal
....111 n=7
E 300~--4-~-+----~--~--~----+----H
u
N2at -183 0
vm =124·7cm 3
EI-EL=900cal
n=6
o 100 200 300 400 500 600 700

Pressure (mm)
Figure 6-3.-contd.
232
e 1-0
.... -;/
0·6 ~ ~~
~
/
UI
111
" ~
'fu
~
V
V
CI
~
~ 0·6 I~
J:
U
N ~f V
~
0
:;; 0·4
ej
1
'{j
V
u
111
....
V
V
L.
'0 0·2 f
~
V
o
o
V 0·2 04 0-6 0·6 1·0
Mol fraction C2H4 in adsorbate
Silica gel Columbia G carbon
o - O·C • - 25·C
6 - 25·C
o - 40·C
C
Figure 6-3.-contd.
where lei is the desorption rate constant At equilibrium the desorption and
adsorption rates are equal. Solving for a at equilibrium we obtain,
(kilk- i ) P
1 + (kilk- i ) P
233
Since the adsorbate concentrate on the solid, q, must be proportional to e
q-
_k
2
e-
_-k2 (kdlcd p
----
(6-5)
1 + (kdlcdp
When the surface is fully covered, e = 1, and q = 'IMAx which is the monolayer
coverage. Thus Eq. (6-5) can be written
(6-6a)
where KA = kdlc l is the sorption equilibrium constant.
Equation (6-6a) is the Langmuir isotherm (Langmuir, 1918). The shape

of Eq. (6-6a) is shown in Figure 6-3b (Brunauer et ai, 1938). This is known
as afavorable isotherm. The Langmuir isotherm is applicable for dilute, single
component adsorption with an inert gas present or for pure gases at low pres-
sures. In concentration units the Langmuir equation is,
(6-6b)
At very low partial pressures the Langmuir equation is linear
(6-7a)
or at very low concentrations
(6-7b)
If we plot q/<IF vs CICF where <IF is the amount adsorbed in equilibrium

with fluid at the feed concentration, the equilibrium curve goes from 0 to 1.0.
The shape is the same as vapor-liquid equilibrium with constant relative volatil-
ity. The term (1 + KA CF) is equivalent to the relative volatility. Thus larger
(1 + KA CF) indicates a more favorable adsorption step.
234
Since KA is a sorption equilibrium constant. we would expect its tem-

perature dependence to follow an Arrhenius form.
KA = Kl exp (- !~ ) (6-8)
where MI is the heat of adsorption. Since adsorption is exothermic, MI is

negative. Thus, Eq. (6-8) correctly predicts that KA and the amount adsorbed
decrease as temperature is increased.
The Langmuir isotherm is often extended to multicomponent systems.

The resulting isotherm equation is
Qj=
n
(6-9)
1 + 1: (Kj Pj)
j=l
where n is the number of adsorbates. This equation shows competitive adsorp-

tion where less solute A is adsorbed when other solutes are present. Although
very convenient, the multicomponent Langmuir isotherm often does not fit
experimental data. Note that Clmax must be the same for all solutes or the result
is not thermodynamically consistent (LeVan and Vermeulen, 1981).
The Langmiur forms assume a maximum of a monolayer coverage. In

some gas adsorption systems additional layers of adsorbate can form on top of
the monolayer. Brunauer et al (1938) derived several isotherms which have
become known as the BET isotherms. The simplest form of the BET isotherm
for one adsorbate is
_q_ = _ _ _ _ _KD~p_ _ _ _
(6-10)
Clmooo P
[(VP) + (K-l)PJ [1- (VP) ]
where Clmono is the amount adsorbed for monolayer coverage and VP is the
vapor pressure of pure adsorbate. The shape of this isotherm depends on the
value of K and is shown in Figure 6-3c for adsorption of nitrogen or argon at
different temperatures (Brunauer et al., 1938). Note that if K» 1 and P «
235
(VP), Eq. (6-10) reduces to the Langmuir form. The BET isotherm is the basis
of a method for determining the surface area of the packing (e.g. see Yang,
1987).
In zeolite molecular sieves the entire pores are filled with adsorbate. In
this case adsorption is not by surface layers, but occurs by pore filling. The
loading rate correlation (LRC) is a simple extension of the Langmuir equation
where <Iron is replaced by the maximum attainable loading N max • For gas
adsorption this equation is (Lee, 1972),
N (6-11)
a typical zeolite molecular sieve isotherm will be similar to a Langmiur isoth-

erm except it is likely to be sharper. Other isotherm equations are also com-
monly used for zeolites (Lee, 1972; Ruthven, 1984; Yang, 1987).
For liquids it is common to generalize the Langmuir isotherm to,
ac (6-12a)
q= l+bc
where a and b do not have to be <lMAxKM and KM, respectively. The constants
a and b can be fit to experimental data. This is easiest to do if Eq. (6-12a) is
rearranged to,
c b 1 (6-12b)
-=-c+-
q a a
If the Langmuir isotherm is followed, a graph of c/q versus c will give a

straight line with a slope of b/a and an intercept of l/a (see Example 6-1). The
mUlticomponent form ofEq. (6-12a) is
<t=--n-- (6-13)
1 + 1:(bj Cj)
j=1
This is a convenient form to use but it may not fit experimental data.
236
A better fit to experimental data for liquids can often be obtained with the
Freundlich equation.
(6-14)
q = A(T) dfn n>1
The Freundlich isothenn is shown in Figure 6-3d. Unfortunately, the

Freundlich equation does not approach a linear isothenn for very dilute solu-
tions, and it does not approach a limiting asymptotic value observed for many
systems. The Freundlich isothenn is commonly used to fit data for the adsorp-
tion of organic pollutants from aqueous solution unto activated carbon. Tabu-
lations of A and n are given by Dobbs and Cohen (1980) and Faust and Aly
(1987). An empirical combination of the Langmuir and Freundlich equations
often gives a better fit to data (Fritz and Schluender. 1974).
(6-15)
where ~o. bio • ~i' ~j and bij are empirical constants. Equation (6-15) is not
thennodynamically consistent and needs to be used with caution.
In exchange adsorption two adsorbates are present and there is no ear-

rier. The surface of the adsorbent is always saturated. The two adsorbates then
exchange with each other for sites on the adsorbent. This is illustrated in Fig-
ure 6-3e for ethylene-propylene mixtures on silica gel at 25°C (LeWis et al.,
1950b). Note that this looks similar to vapor-liquid equilibrium data. For
binary systems equilibrium can often be represented by a constant separation
factor, nAB'
(6-16)
where y and x are mole fractions in the gas and solid phases, respectively. This
is analogous to a relative volatility. To predict the actual amount adsorbed, Eq.
237
(6-16) is solved simultaneously with the correlation (Lewis et ai, 1950b),
QA 'lB (6-17)
-+-=1
CIA qB
where CIA, q~ are the amount absorbed from a pure gas and QA, 'lB are the
amount adsorbed from the mixture. Equation (6-17) is often valid even if the
separation factor nAB is not constant
In gas-liquid chromatography !he mechanism responsible for separation

is essentially absorption of the solute into the stationary liquid phase not
adsorption at a surface. At low concentrations the isotherms follow a Henry's
law or linear relationship. Absorption is favored as more solute disolves in the
stationary phase. The shape can be fit by an isotherm of the form.
N=
ap b>O
(6-18)
1-bp
where N is the loading of solute on the packing. This shape isotherm is shown
to Figure 6-3f. Absorption systems usually follow Henry's law for much
higher ranges of p than adsorption systems.
More details of adsorption equilibrium are in Ruthven (1984) and Yang

(1987). A review listing sources for gas adsorption data is available (Valen-
zuela and Myers, 1984). Valenzuela and Myers (1989) give references and
show data for a very wide variety of pure gases, gas mixtures and liquids.
Faust and Aly (1987) give data for removal of organics from water with
activated carbon.
Example 6-1. Langmuir Isotherm
Tsou and Graham (1985) measured the sorption of the protein bovine
serum albumin (BSA) on an ion exchanger, DEAE Sephadex A-50.
The conditions were: T = 20 ± 1°C, pH = 6.9, 0.005 M phosphate
238
buffer. The results for c and q given below were read from a figure
given by Tsou and Graham (1985).
c, mg protein q, mg protein
per g solution per g dry resin c/q
0 0 0
0.05 5767 0.867 x 10-5
0.10 6000 1.667 x 10-5
0.20 6778 2.951 x 10-5
0.30 7578 3.959 x 10-5
0.50 7611 6.569 x 10-5
0.50 7544 6.628 x 10-5
0.70 8422 8.311 x 10-5
0.90 7789 11.55 x 10-5
Does this sorption follow a Langmuir isotherm? What are the values of
a and b in Eq. (6-12a)?
Solution
Even though this is ion exchange and not adsorption, we can use the
Langmuir isotherm. Equation (6-12b) shows that a plot of c/q versus c
should be a straight line with a slope = b1a and an intercept of l/a.
Values of c/q are calculated in the table. The plot is shown in the
figure. The line drawn is the best line from a linear regression routine
on my calculator. The resulting parameter values are
b/a = 1.2111 x 10-4 l/a = 3.933 x 10-6
and the equilibrium expression is
254,259 c
q = 1 + 30.79 c
Note that there is some scatter in the data; however, the duplicate data
point is quite close.
239
10
o
8
c/q X 10- 5
OL-____ ~ ____
J __ _ _ _ ~ ____ ~ ____ ~
o 0.2 0.4 0.6 0.8

c
This example shows that unfamiliar systems (e.g. ion exchange of

a protein) can easily be analyzed using the Langmuir isotherm. The
values of q are very high compared to normal adsorption. This is possi-
ble since the gel swells in water and one gram of dry resin swells to fill
64.5 ml of column. The values of a and b differ slightly from those
obtained by Tsou and Graham (1985) [268,925 and 32.7J. This prob-
ably occurs because of inaccuracy in reading the data points.
6.4. SOLUTE MOVEMENT
As solutes migrate through the bed they can be in the mobile fluid (external
void volume, Ee ~ az), in the stagnant fluid inside a particle (void volume
(1- Ee)Ep ~ az), or sorbed to the particle. (Refer to Figure 6-2.) The only
solutes which are moving towards the column exit are those in the mobile fluid.
Consider the movement of an incremental mass of solute added to a segment of
240
the bed shown in Figure 6-2. Within this segment this incremental amount of
solute must distribute to form a change in fluid concentration, ~c, and a change
in the amount of solute adsorbed, ~q. The amount of this increment of solute
in the mobile fluid compared to the total amount of the solute increment in this
segment is,
Amt in mobile fluid

Total amt. in segment
Amt in mobile fluid (6-19)

=---------------------------------
Aml in: (Mobile fluid + stationary fluid + sorbed)
The individual terms on the right hand side of Eq. (6-19) can be determined.
For instance,
(6-20)
Amount solute increment in mobile fluid =
(Vol. column segment) (frac which is mobile fluid) (conc. moles/liter)
After determining each term in Eq. (6-19), we obtain
Aml in mobile fluid (6-21)

Total aml in segment
=
The ~ term is not included in the sorption term (last term in the denominator)
since q is an experimentally measured quantity which will already include any
steric exclusion effects. The solid density, P" is included in Eq. (6-21) to make
the units balance. If q and c are in the same units the solid density is not
included in Eq. (6-21). Ac is the cross sectional area, and z is the axial dis-
tance.
241
Ifthe fluid has a constant interstitial velocity, v, then the average velocity
of the solute in the bed lis is,
(6-22a)
u. = v x (fraction solute in mobile phase)
Assuming a random process of adsorption, desorption and diffusion in and out

of the stagnant fluid, the solute wave velocity becomes
amount solute in mobile phase

lis = v x [ total amount solute in column
1 (6-22b)
or, after rearrangement
v (6-23)
lis (T) = 1 + [(l-Ec)/Ecl Ep~ + [(I-Ec)/EcJ (l-Ep)p. (6q/6c)
Eq. (6-23) represents a crude, first order description of movement of solute in
the column. It gives an "average" velocity, but because of the randomness of
diffusion and equilibrium there will be spreading of zones. With a few addi-
tional assumptions Eq. (6-23) can be used to predict the separation in the sys-
tem.
The most important assumption, and the assumption least likely to be

valid, is that the solid and fluid are locally in equilibrium. Then 6q will be
related to 6c by the equilibrium adsorption isotherm. This assumption allows
us to ignore mass transfer effects. The second assumption is that dispersion
and diffusion are negligible; thus, all of the solute will travel at the same
average solute velocity. These assumptions greatly oversimplify the physical
situation, but they do allow us to make simple predictions. As long as we don't
believe these predictions must be exactly correct, the simple model which
results can be extremely helpful in understanding these separation techniques.
With linear equilibrium, from Eq. (6-4a) 6q/6c = A(T), and Eq. (6-23)
becomes
(6-24)
242
For Eq. (6-24) the solute wave velocity is the same as the average solute velo-
city. Equation (6-24) allows us to explore the behavior of solute in the column
for a variety of operating methods.
Several facts about the movement of solute can be deduced from Eq. (6-
23) or Eq. (6-24). The highest possible solute velocity is v, the interstital fluid
velocity. This will occur when the molecules are very large and
~ = A(T) = 0.0. For small molecules Kd = 1.0, and with porous packings
small molecules always move slower than the interstitial velocity even when
they are not adsorbed. If adsorption is very strong the solute will move very
slowly. When the adsorption equilibrium is linear, Eq. (6-24) shows that the
solute velocity does not depend on the solute concentration. This is important
and greatly simplifies the analysis for linear equilibria. If the equilibrium is
nonlinear, tJ.q/tJ.c will depend on the fluid concentration and Eq. (6-23) shows
that the solute velocity will depend on concentration. Since linear equilibrium
is usually valid at very low concentrations, Eq. (6-24) is a limiting case which
will be valid at low concentrations.
A convenient graphical representation of the solute movement is
obtained on a plot of axial distance, z, versus time. Since the average solute
molecule moves at a velocity of lIs (1'), this movement is shown as a line with a
slope lIs. This is illustrated for a simple chromatographic separation in Figure
6-4. Figure 6-4a shows the feed pulse while Figure 6-4b shows the solute
movement in the column. The product concentrations predicted are shown in
Figure 6-4c. Note that this simple model does not predict dispersion or zone
spreading, but does predict when the peaks exit. The solute movement
diagrams will be used repeatedly; thus. you should study Figure 6-4 until you
understand it.
If desired, zone spreading can be included, but this will conceptually
complicate the model. The advantage of this model is that it is simple and can
be used to understand a variety of methods of operation. More detailed models
including mass transfer effects are discussed in Chapters 7 and 8.
If the equilibrium isothenn is nonlinear the basic structure developed

here is still applicable. but we must use Eq. (6-23) instead of (6-24). The solute
velocity now depends on both temperature and concentration. Once a specific
243
CA/CA
or 0
CB/C Bo Of.-----'-f~--
Time
Figure 6-4. Solute movement model for isothermal chromatography. a.

Feed pulse. b. Trace of solute movement in column. c. Product
concentrations.
isotherm is determined it can be substituted into Eq. (6-23). For example, if the
Freundlich isotherm Eq. (6-14) is used, then
1
. .1q aq A(T) (n- 1) (6-25)
lim - = - I T = - - C
Ac-+O.1C ac n
and
v
lIs = - - - - - - - - - - - - - - - - - - 1 - (6-26)
A(T) (--1)
1 + [(I-Ee)/EeJ Ep~ + [(I-Ee)/EeJ (l-Ep) Ps - - c n
n
The result for other isotherms such as the Langmuir isotherm is easily
developed (see Problem 6-C4). Remember that acIf<k is the slope of the equili-
244
Time
Figure 6-5. Diffuse waves. a Inlet concentration. b. Solute movement. c.
Outlet concentration.
brium isothenn at c. Thus dq(c)c can be detennined from data even if an equa-
tion is not known.
A diffuse wave (that is the solute is diffused or spread out) occurs when a
concentrated solution is displaced by a dilute solution. This is illustrated in
Figure 6-5 where the outlet wave concentrations are calculated. As long as the
concentration is cH. all solute waves move at the same velocity. lIs (CH). When
the concentration decreases. the solute wave velocity will decrease and a dif-
fuse wave or fan is generated. This results in the zone spreading shown in Fig-
ure 6-5c. Note that this spreading is due to the shape of the isothenns and is
proportional to the distance traveled. Thus this is called a proportional pattern.
The diffuse wave shape can be detennined by arbitrarily picking several con-
centrations between CH and CL. For each concentration Us is calculated from
Eq. (6-23) or the fonn written for a specific isothennal such as Eq. (6-26) for a
245
c
Lr---~--------~n-------
Time
.
Figure 6-6. Shock wave analysis. a. Inlet concentration. b. Physically
impossible solute waves following Eq. (6-23). c. Shock wave
following Eq. (6-30). d. Outlet concentration.
Freundlich isotherm. The solute movement line is drawn, and from this one
point on the outlet concentration profile is determined (see Figure 6-5 and
Example 6-2).
If we try the reverse (dilute solution displaced by a concentrated solu-

tion) then the limit in Eq. (6-25) does not exist since dC has a finite value. A
shock wave occurs. Another way of looking at this is Eq. (6-26) predicts a
lower slope for dilute systems than for concentrated. When a concentrated
solution displaces a dilute solution (Fig. 6-6a) , the theory predicts that the
solute lines overlap and two different concentrations occur simultaneously (Fig.
6-6b). This is physically impossible. To avoid this problem the mass balance
246
must be done on a finite section of the column of length t:.z. If the shock wave
moves at a velocity llsh, then the time required for the shock wave to move the
distance t:.z is
(6-27)
A mass balance for period ~t over segment t:.z is,
(6-28)
IN - OUT - ACCUMULATION = 0
or
(6-29)
where 1 refers to conditions before the shock wave and 2 to after the shock.
Now we select the time interval ~t = t:.zlush so that the shock has passed
through the entire section. Solving for the shock wave velocity, we obtain
This is shown in Figure 6-6c and the outlet concentration is calculated in Figure
6-6d. Note that this result can also be obtained directly from Eq. (6-23) if we
determine ~qf~c for the discrete step.
The shock wave velocity depends on q and c after and before the shock
wave. Usually equilibrium between solid and fluid is assumed. Then q and c
in Eq. (6-30) are related by the appropriate isotherm equation. For favorable
isotherms (shapes shown in Figures 6-3b and 6-3c)
(6-31)
Because of Eq. (6-31) shock waves sharpen up and retain their shape. The
more the curvature of the isotherm the stronger the tendency for the shock
247
wave to sharpen. For the Langmuir isotherm (1 + KA Cp) is a measUre of the

tendency to sharpen. The larger (1 + KA cp) the sharper the observed waves
will be. In actual practice dispersion and finite rates of mass transfer smooth
out the shock wave shown in Figure 6-6d. A balance is reached between the
sharpening effect of the isotherm and the dispersive effects. A constant pattern
S-shaped curve results. This shape does not change regardless of how long the
column is. This concept will be discussed further in Chapter 8. Shock and dif-
fuse waves can intercept each other and interact. This concept is discussed in
Section 7.2.
The physical reasons for diffuse and shock waves can be clarified with a
simple analogy. Suppose we have a line of cars moving on a one lane highway.
If the cars are ordered in terms of speed with the fastest car in front and the
slowest car last, the line will spread out as the cars move down the highway.
This is a diffuse wave. If the slow cars are in front, all of the cars must move
slowly. This fesult is similar to a shock wave.
Example 6-2. Solute Movement for Nonlinear Isotherms
The adsorption of acetic acid from aqueous solution onto

activated carbon was extensively studied by Baker and Pigford (1971).
The properties are:
Ps = 1.820 glcm 3 , KcJ = 1.0, Cp,s = 0.25 cal/g °C

Ee = 0.434 Cp,r = 1.00 cal/g °e
Ep = 0.57 Pr = 1.0 g/cm 3
Adsorption follows a Freudlich isotherm, Eq. (6-14), with the following

values:
Temperature A(T) n
4°C 3.646 3.277
60 0 e 3.019 2.428
where c is in gmoles/liter and q is gmoles/kg dry carbon.

248
a. An initially clean column (c = 0, q = 0) is fed with an acetic acid

solution containing 0.25 gmoles/liter acetic acid. Superficial fluid velo-
city is 5 cm/min and the column is 2 meters long. Operation is at 60°C.
Predict the shape of the solute wave and when it will exit the column.
b. Mter the column is saturated with c = 0.25 gmoles/liter, the acetic

acid is removed with a solution with 0.05 gmoles/liter acetic acid at a
superficial fluid velocity of 15 cm/min. Predict the shape and time the
solute wave will leave the column.
Solution
A. Define. Sketches of the two parts are shown in the figure.
a b
t=O DO t=O t)O
L=2
c=O.25 cF =O.05
Vsupe ,= 15
Find breakthrough profiles for each part
B. Explore. The Freudlich isotherm is non-linear and has a favorable

shape. Thus, we would expect a shock wave during part a and a diffuse
wave during part b.
C. Plan. Use shock wave Eq. (6-30) for part a. Condition 1 before the
shock wave is c = 0, q = O. Condition 2 after the shock wave has
~ = CF = 0.25 and <12 is in equilibrium with cz. Plot this wave on a z vs
tdiagram.
249
For part b the diffuse wave velocity is detennined from Eq. (6-
26). The solute velocity is calculated for several concentrations from c
=0.25 to e =0.05. The values are plotted on a z vs t diagram, and the
outlet concentration profile is determined from this.
D. Do It. a. Shock Wave.
Interstitial velocity = v = vsuper = ~ = 11.52 em/min

Ec 0.434
1 1
qz = A(60)C2n(60) =3.019(0.25) 2.428 = 1.706
From Eq. (6-30)
11.52
=------------------------------------
0566 0566 (1.82)(1.7Q6-{))
1 + 0:434 (0.57)(1.0) + 0:434 (0.43) (0.25-0)
llsh = 1.32 em/min
The shock wave exits at lout = L/Ush = 200/1.32 = 151.5 min. The outlet
profile looks like Figure 6-6d with Clow = 0, Chigh = 0.25 and toot = 151.5
minutes.
b. Diffuse wave.
Interstitial velocity v = VI: = 0.~;4 = 34.56 em/min

250
Diffuse wave velocity for Freundlich isothenn is given by Eq. (6-26).
v
lis = ----------------:1-
1-£ l-Ee A(T) (-1)
1 + (e:-)EpKc! + (e:-)(1-Ep)ps-n - c n
34.56
=----------------------
1+( :~~ )(.57)(1.0)+( :~~ )(.43)(1.82)( ;:~~~)c( 2.4;8-1 )
We can detennine Us and lout for several concentrations,
L 200cm
c Us, cm/min lout=-=
Us Us
0.25 7.53 26.9 min

0.20 6.93 28.9
0.15 6.18 32.4
0.10 5.22 38.3
0.05 3.81 52.5
Each one of these waves can be plotted starting from t =0, Z = 0 to t =

lout at z = L. The result will look like Figure 6-5b. The outlet concen-
tration profile can be plotted as shown in Figure 6-7.
E. Check. The outlet profiles have the shapes we expect. This is a

visual check.
F. Generalization. For favorable isothenns we expect shock waves

when feeding the column with a more concentrated feed, and diffuse
waves when the column is eluted with a less concentrated stream. Note
that the solute wave velocity decreases as concentration decreases and
this is quite marked at the low velocities. In fact, the Freundlich isoth-
251
0.25
0.20
0.15
COUT
0.10
0.05
20 25 30 35 40 45 50 55
Time
Figure 6-7. Results from Example 6-2.
enn predicts Us =0 when c = O. This is usually not observed in experi-

ments. Thus the Freundlich isothenn should not be used for diffuse
wave calculations when c = O. The reason for this incorrect prediction
is the empirical Freundlich isothenn is not a good fit of experimental
data for very low concentrations.
Note that plotting the solute movement diagram is not necessary

for simple problems. For more complex problems the solute movement
diagram is very useful to help understand the problem and to avoid
errors. The solute movement diagram also provides a graphical
representation which many people find useful.
6.S. EFFECTS OF CHANGING THERMODYNAMIC VARIABLES

Changes in temperature, pH, ionic strength or solvent concentration are often
used to help desorb and elute the solute. Changes in these variables will
change the equilibrium constants in the isothenn equations. For common
adsorbents the amount of material adsorbed decreases as temperature is
increased. Thus A(T) and KA are monotonically decreasing functions of tem-
perature regardless of the validity ofEqs. (6-4b) and (6-8). One can either use
a jacketed bed and change the temperature of the entire bed (called the direct
252
mode) or one can change the tempemture of the inlet stteam and have a tem-
perature wave propagate through the column (called the tmvelling wave mode).
For most of the other elution methods (pH, ion strength, etc.) the tmvelling
wave mode is used. Elution may be done either co-flow or counter-flow to the
feed direction.
The velocity at which temperature moves in the column (the thermal

wave velocity) can be obtained from an energy balance. If we can ignore the
heat of adsorption and heat of mixing, and assume the column is adiabatic, then
the average velocity of the thermal wave is
u
Ih
= [ energy in mobile phase
total energy in segment
1
V
(6-32)
The fraction of energy stored in the mobile phase in a segment of the column
is,
Energy in mobile phase

Total Energy in segment
= Energy in mobile phase (6-33)

(Energy in: Mobile + Stagnant + Solid + Wall)
The numerator is
(6-34a)
Energy in mobile phase = (6z Ac) Ee Pf CP,f (TrTIci')
while the denominator is
Total energy = 6zAc [(Ee + (l-Ee)tp) PfCp,f(TrTn{) ]
(6-34b)
[ + (I-Ec)(l-tp)PsCp,s (T.-Trcr) ] + (6zW)Cp,w(T w-TIci')
253
The Cp values are the heat capacities while W is the weight of column wall per
length and T w is the wall temperature. Ifwe have local thermal equilibrium
(6-35)
and the term (T - Tref) divides out Then combining Eqs. (6-32) and (6-34) we
obtain the thennal wave velocity,
(6-36)
Note that with the simplifying assumptions made here 11th is independent of
temperature. Comparison of Eqs. (6-36) and (6-24) show they have a similar
form but there is an additional term in Eq. (6-36) to account for thermal storage
in the column wall, and effectively Kct == 1.0 for energy changes. Just as Eqs.
(6-23J and (6-24) represented the movement of the average solute molecule,
Eq. (6-36) represents the average rate of movement of the thermal wave. A
more exact analysis is needed to include dispersion and heat transfer rate
effects.
On a graph of axial distance z versus time t the t.hennal wave will be a

straight line with a slope 11th. This is illustrated in Figure 6-8a. The outlet tem-
perature profile is shown in Figure 6-8b. Note that this behavior is exactly the
same as for a solute with a linear isothenn.
To complete the analysis we need to consider the change in solute con-

centration when the adsorbent temperature is changed. Since the following
analysis is subtle, it needs to be followed closely. The effect of temperature
changes on solute concentration can be determined by a mass balance on a dif-
ferential section of column /}.z over which the temperature changes during a
time interval At. The appropriate differential section is shown in Figure 6-9a
for a typical liquid system where
(6-37a)
254
0"--------
Figure 6-8. Thennal wave. a Trace of thennal wave in column. b. Product

temperature.
Then the mass balance is
(6-38)
where 1 refers to conditions before the temperature shift and 2 to after the shift
Note that Eq. (6-38) is an application of Eq. (6-28) and is very similar to Eq.
(6-29) used for shock waves.
To ensure that all material in the differential section undergoes a tem-

perature change the control volume is selected so that ~t = Ilz/Uth. The mass
balance then becomes
(6-39a)
255
a to to +Dot b to to + Dot
cl,TI cl,T I cl,TI CChange,"T;
Doz cl,TI c 2 ,T2 Doz CChange;r1 c 2 ,T 2
c 2 ,T2 c 2 ,T2 c 2 ,T2 c 2 ,T2

t t t t
Figure 6-9. Differential sections for mass balance when temperature
changes. ~t= &/Uth. a. llut > lis (T2. Cz). uth > lis (Tit ct>. b.
Us (Th. CF) > Us (Te, CF) > Uth·
In this equation Cz and CJ2 are the unknowns. If we assume that solid and fluid
are locally in equilibrium and that the equilibrium isotherm is linear, then Eq.
l/ 1
(6-39a) reduces to
(6-40)
C(Th) [1 1 [1 1
c(Tc) = Us (Tc) - uth lis (Th) - Uth
,In a typical liquid system where Eq. (6-37a) is satisfied, C(Th) > c(Tc).
Thus, the solute is concentrated during elution. This concentration is calculated
from Eq. (6-39a) or (6-40), and the solute movement diagram can be plotted as
shown in Figure 6-1Oa. Concentrations are shown in Figure 6-1Ob. Note that
the overall mass balance will be satisfied. The expected shape of the solute
movement diagram for a particular example is shown in Example 6-3. If the
equilibrium constant A does not change very much, lIs(Th) "" lIs(Tc) and there
will be little change in concentration during elution. Since A is not usually
strongly dependent on temperature, large temperature changes are required.
An alternative is to use a different elutant which has a major effect on the
equilibrium constant. Eqs. (6-39) and (6-40) are still valid but with UE1utant
replacing Uth.
In the direct mode the entire column is heated or cooled simultaneously

through a jacket. In this case Uth is essentially infinite. Equation (6-40) for
linear isotherms simplifies to
(6-41)
256
a
r
I
COO' Ib Tc
CI CF
.
t
C Cchange
----~~
Us (Tc)~ .,.:/f::: / I
T, ~~{ I
" Uth I Th I
.,..- / I
I
" CF US (TM)I
I
It
L,
C I
d
CI = CF "'""I
t
Figure 6-10. Effects of thermal waves for linear isotherms. a,b. Liquid sys-
tems where Uth > Us (T2, C2) > Us (fl' CF). c,d. Dilute gas sys-
tems where Us (T2, C2) > (T I, CF) > Uth·
For the usual adsorbent, Am decreases (solute desorbs) as temperature

increases. Thus Us increases and, as expected, Eq. (6-41) predicts that the
solute concentration increases as temperature increases.
Equations (6-39) and (6-40) were derived for a typical liquid system
where Eq. (6-37a) is valid. In dilute gas systems it is common to have
(6-37b)
257
A mass balance similar to Eq. (6-38) using the balance envelope shown in Fig-
me 6-9b can again be developed. The result (see Problem 6-C6) is,
(6-39b)
where Cchange is the concentration change caused by the temperature change.

This concentration has now moved ahead of the thermal wave. In Eq. (6-39a)
the unknowns are C2 and <l2(c2,T2) while in Eq. (6-39b) the unknowns are
Cchange and q (cchange,T1). Thus the differences in Eqs. (6-37a) and (6-37b)
cause an important but subtle change. Equations equivalent to Eqs. (6-40) and
(6-41) can also be derived for the case given in Eq. (6-37b) (see Problem 6-
C6). This change is illustrated in Figures 6-1Oc and d. Note that with gas sys-
tems the changed (increased for a temperature increase) concentration wave
comes out ahead of the thermal curve while in liquid systems (Figme 6-10b)
the concentration waves comes out after the thermal wave.
A third case occms when
(6-37c)
and the temperatme in the feed is increased. Now the solute waves for linear
isotherms will intersect at the thermal wave as shown in Figure 6-11a. Since
solute waves from both cold and hot sides intersect the thermal wave, the solute
concentration must increase. Since the theory does not include zone broaden-
ing, this predicts an infinite concentration. To prevent this physically impossi-
ble result non-linear isotherms must be used. Now the concentration builds up
ahead of the thermal wave so that
Obviously, this can occur only with non-linear isotherms. The mass balance
envelope will look like Figure 6-9b and the mass balance is given by Eq. (6-
39b) (see problem 6-C7). This case is known asfocusing (Wankat, 1986), and
258
a
Lf---- - - - -~- ---
,
z 1
b
Lt---- - - - - - - ~~
Figure 6-11. Effect of thermal waves when focusing occurs, Eq. (6-37c). a.
Physically impossible characteristic diagram for linear systems.
b. Characteristic diagram for nonlinear systems. c. Concentra-
tion profile for nonlinear system.
is illustrated in Figures 6-11 b and c. The concentration in the region moving

ahead of the thermal wave can be calculated from Eq. (6-39b). Since this con-
centrated wave is displacing a dilute stream, there will be a shock wave which
can be calculated from Eq. (6-30). Although rare, systems which focus c~m
give large separation factors (see Problem 6-DIO). Focusing is more likely to
occur when thermodynamic variables other than temperature are changed.
Separation methods can be devised which are based on focusing (see Wankat,
1986, and Section 7.10).
259
a +Reverse flow
C
cF
0
L
b
cF c conc
0
I
1---
C 1 Tdesorb
T 1
C 1
or c conc
Top of column
T Bottom of column
TF cF
JT c
0
c
tR tl t?
Time
Figure 6-12. Solute movement model for adsorption followed by counter-

current elution with a hot fluid (Uth > lis): a. Inlet concentration
and temperatures. b. Trace of ~lute and temperature move-
ment in bed. c. Product concentrations and temperatures.
Figure 6-12 illustrates elution using high temperatures for counter-flow

desorption. In this case a single solute is adsorbed. The feed flow is continued
until just before solute breakthrough (that is when solute appears in the outlet)
occurs. Then counter-current flow of a hot fluid is used to remove solute.
Upon reversing the flow the fluid first exits at the feed temperature and the feed
concentration. When the thennal wave breaks through, the temperature and
concentration both jump. Since the adsorption equilibrium constant is lower at
high temperature, the solute velocity can be significantly greater at the higher
temperature. In actual practice the outlet temperature and concentration waves
will be S-shaped curves because of dispersion.
260
Example 6-3. Temperature Effects in Adsorption Columns.
A well insulated one meter long column containing activated car-

bon is saturated with an aqueous solution containing 0.4 g moles/liter
acetic acid. Temperature is initially 4°C. The bed is eluted with a feed
containing 0.04 gmoles/liter acetic acid at 60°C and a superficial velo-
city of 10 em/min. Find the outlet temperature and composition
profiles.
Solution
A. Define. The process is sketched in the figure.
L= I m D t=O
e =O.4g moles/L
T=4°C
t>0
TF = 60°, vsuper = 10em/min

c=0.04
Find breakthrough profiles for temperature and composition.
B. Explore. Equilibrium, heat capacity, porosity and density data is

available in Example 6-2. We expect a simple thermal wave in the
column. This wave will cause an increase in acetic acid concentration
since acetic acid will desorb at 60°C. The behavior of this more con-
centrated wave depends on the wave velocity compared to Uth. This
will have to be calculated. The concentrated region is displaced by the
hot dilute feed and a diffuse wave will result
C. Plan. Calculate the thermal wave velocity from Eq. (6-36). Since
no information is given on column diameter and wall effects, we will
assume wall heat storage is negligible. This will be a good assumption
for commercial systems, but not so good for laboratory columns. Next
261
calculate C2 (60°) from Eq. (6-38) assuming C2 and qz are in equili-

brium at 60°C. The equation is non-linear since equilibrium is
represented by the Freundlich isothenn. Then detennine U. (C2. 60°C)
from Eq. (6-26). If U. (C2. 60°C) < Ulh. the concentration at a point in
the column will be 0.4 until the hot wave passes when it will jump to ~.
Finally. determine the diffuse wave from c2 to 0.04 gmole/L.
D. Do It. With : CP •w = O. Eq. (6-36) is

c
vaupc:r 10 .
where v = - - = - - = 23.04 cm/mm
Ee 0.434
thus.
llth = 23.04 11.53 c~

1+ 0.566 (0.57)+ (0.566)(0.43)(1.82)(0.25) mm
0.434 (0.434)(1.0)(1.0)
The thennal wave exits at 1m =~ = 100 = 8.67 min.

Ulh 11. 53
To detennine the effect of the temperature shift we need ql (4°C) in

equilibrium with Cl = 0.4. From the Freundlich isothenn
ql (4°) = A(4°)cfln = (3.646)(0.4)1/3.277 = 2.757 gmoles/kg
At 60°C, qz = A(60)d ln = 3.019 C¥L428

262
Substituting this into Eq. (6-38), we have
Substituting in the values this equation becomes,
f(C2) = 1.337 C~.419 - 0.0732 C2 - 1.192 = 0
The second term will be small. If we ignore this term for the moment
we have
= ( 1.192 )11.4119 = 0 757

C2 1.337 .
since the second term is subtracted, f(0.757) = (--0.0732)(0.7511) = -

0.055. Thus try ~ = 0.80 as first guess and calculate f(.80) =- 0.031.
This gives two guesses. Using some convergence procedure such as
Newtonian or secant approach we find C2 = 0.857. As expected C2 > C1.
Now we can calculate Us (C2' 60°) from Eq. (6-26) which is,
23.04
(1 1)
1 0.566 (0 57)(1 0)+ 0.566 (043)(1 82) 3.019 (0.857) 2A28
+ 0.434' . 0.434' . 2.428
= 7.354 cm/min
Since Us (60°, ~) < Uth, we have the simple case outlined in the PIan.
For diffuse waves we use Eq. (6-26), but with varying concentrations.
263
These are easily calculated,
c Us (60°) lout = LIu.
0.857 7.354 13.60

0.8 7.222 13.85
0.7 6.964 14.36
0.6 6.665 15.00
0.5 6.311 15.85
0.4 5.880 17.01
0.3 5.334 18.75
0.2 4.596 21.76
0.1 3.460 28.90
0.04 2.266 44.14
The total solute movement diagram and outlet concentration profile are
100
z cm
50
I
201 30 40 50
10 t, min
1
1.0
I
I
0.857
C 9 moles/L
0.5
cinit
0.04
0
10 20 30 40 50
t, min
Figure 6-13. Results of Example 6- 3,
264
E. Check. The mass balance can be checked to see if IN-OUT =

ACCUMULATION after the last wave has exited. This balance
checks.
F. Generalization. Regeneration at high temperature is used because

high concentrations and more rapid removal of the diffuse wave are
achieved. The pattern achieved here with the thermal wave exiting with
and causing the concentration wave is common for many liquid sys-
tems. However, if0. (Th' cz) > uth > 0. (Te' CI) the pattern of the solute
movement diagram and the outlet concentration profile will be quite dif-
ferent. This is illustrated in Example 6-4.
Example 6-4. Focusing.
Repeat Example 6-3, but for a special high heat capacity

activated carbon with Cp,s = 2.2 cal/gOC.
Solution
Again assume W Cp,w/Ac = O. Replace Cp,. = 0.25 with Cp,s = 2.2.
Then, Uth = 5.777 cm/min and lut = 17.31 min. Solute velocity at c = 0.4
is given by Eq. (6-26), which gives
Us (c = 0.4, 4°C) = 2.266 cm/min

Us (c = 0.4, 60°C) = 5.88 cm/min
Thus, Eq (6-37c) is satisfied and focusing occurs. Equation (6-39b)

should be used to determine the effect of the temeprature change. For
this balance Cz = 0.4, Tz = 60°C and from the isotherm
qz = 3.019 C~f2·428 = 2.07

265
Cch and qch (Cch, T = 4°C) are unknown, but q is given by

qch = (3.646) C~{!·Z77
Then Eq. (6-39b) simplifies to
- 1.615 C~ro52 + 0.9744 Cch + 0.52715 = 0
Note that this nonlinear equation is significantly different than the non-
linear equation in Example 6-3. Since the unknown condition is now at
4°C instead of 60°C the exponent on cch has changed. This equation
was solved by trial-and-error. The result is Cch = 1.22
This high concenttation displaces fluid with C1 = 0.4 and T = 4°C. This
gives a shock wave. Calculating
qch = 3.646 (1.22)113.277 = 3.874
we find,
~q = 3.8744 - 2.757 = 1.362

~c 1.22- 0.4
Ush = 0.566 ~;: = 7.352 cm/min

1 + 0.434 (0.57) + 0.434 (0.43) (1.82) (1.362)
where the values of v, fe, f.p and PI are from Examples 8-1 and 8-3.
This shock wave exits in
tah =Lfush = /~2 = 13.60 min
The results are shown in Figures 6-11 band c. Note that the concentra-
266
tion after the thermal wave immediately drops back to CF since a solute
wave at CF and T = 60°C exits at this point.
Obviously, the behavior illustrated in Examples 8-3 and 8-4 is very dif-
ferent (compare Figures 6-11 and 6-13). This occurred when only one
variable, CP,s., was changed; however, this changed the system from a
non-focusing to a focusing system.
6.6. THEORETICAL ASSUMPTIONS AND UMITATIONS OF

SOLUTE MOVEMENT THEORY
This completes the solute movement theory. As presented, this is not a

rigorous mathematical model but was based on simple physical ideas. The
equations developed here can be rigorously derived from the governing partial
differential equations by making a group of assumptions called the local equili-
brium assumptions and then using a mathematical method called the method of
characteristics (see Section 6.9). Although some of the assumptions required
for the rigorous development have been mentioned in passing, it is helpful to
list them explicitly (Table 6-7).
If any of these assumptions are invalid the predictions can be way off.
The most critical assumptions are the last three. Assumptions 14 and 15 cause
the predicted outlet concentrations and temperatures to show sharp jumps
instead of the experimentally observed S-shaped curves. Alternate mathemati-
cal models which are more realistic but much more complex are covered in
Chapters 7 and 8. Assuming linear equilibrium can also cause physically
impossible predictions, but fortunately this assumption is easily relaxed.
As we have seen, this model greatly oversimplifies the actual fluid flow
and heat and mass transfer processes occurring in the column. Because of this
the predicted separation is usually better than that obtained in practice. What is
this model good for?
The solute movement model does include the reasons why separation
occurs but not many of the phenomena which limit separation. The model is
267
Table 6-7. Common Assumptions to Simplify Mass and Energy Balances
Assumption Comments
1. Homogeneous packing Valid if careful-Poor

(no channeling) packing can invalidate
2. Negligible radial gradients May not be valid if high heat loss
3. Neglect thermal and pressure diffusion Usually OK
4. No chemical reactions except for sorption Usually OK
5. Neglect kinetic and potential energy terms Usually OK
6. No radiant heat transfer Only absolutely true for isothermal
Usually included with other heat transfer
7. No electrical or magnetic fields Usually OK
8. No changes of phase except sorption Usually OK
9. Parameters are constant except for equilibrium Never strictly true,
but usually close enough
lOa. Velocity is constant across cross-section OK if no channeling, no
viscous fingering, no radial gradients
lOb. Velocity is not a function of z OK for liquids, exchange adsorption,
very dilute gases
11a. Column is adiabatic, or OK large columns with insulation
11 b. Column is at controlled constant temperature OK small columns with jacket
12. Heat of adsorption term is negligible OK for liquids, exchange adsorption,
dilute gases
13. Solutes do not interact OK for dilute systems
14. Negligible thermal and mass Reasonable for liquids. Often
axial dispersion and diffusion lumped with mass transfer terms
15. Heat and mass transfer rates Never strictly true.
are very high so that fluid Reasonable in diffuse waves.
and solid are locally in equilibrium Obviously wrong in shock waves
268
simple and can thus be used to analyze rather complex processes. The model is
a good guide for setting operating variables. Since this model predicts the best
possible separation, the model can be used to determine if, at its best, a separa-
tion scheme is of interest. Since the predictions made are qualitatively correct,
as long as the model predictions are interpreted in a qualitative or at best semi-
quantitative sense the model is very useful. In addition, the prediction for dif-
fuse waves is often good and can be used for design.
6.7. BASIC MASS TRANSFER EQUATIONS
The separation in adsorption, chromatography and ion exchange occurs

because of differences in rates of solute or ion movement. However, as the
experimental results of the previous section show the solute movement theory
is far from the entire picture. The solute movement theory assumes that solid
and fluid are locally in equilibrium. When equilibrium is the dominant
phenomenon, the solute movement theory agrees with experimental data.
Unfortunately, for shock waves resistance to mass transfer is always important.
The result is a constant pattern, mass transfer zone which moves at the shock
wave velocity. Mass transfer resistances will also spread out diffuse waves.
However, equilibrium effects dominate except when mass transfer is very slow,
and the solute movement theory results are often adequate for diffuse waves.
Mass transfer of solute in a packed bed can be thought of as occuring in

several steps. First, the solute diffuses from the bulk fluid to the surface of the
particles. This is often treated as occuring across a film. Next, the solute dif-
fuses in the fluid through the pores. The solute then adsorbs and is held on the
solid. Once on the solid the solute may diffuse along the solid or it may even-
tually desorb. Once desorbed, the solute may diffuse through the pores back
into the bulk fluid, or the solute may diffuse to another site where it will adsorb.
Since the pores in a particle are not straight (see Figure 6-2 for a schematic
representation) the diffusing solutes must follow a torturous path. This picture
is complex and all theories require experimental data to predict results.
A fundamental study of mass transfer can start with the differential solute
269
mass balance for the packed bed. For the two porosity model this solute mass
balance is,
(6-42)
where Dm is the molecular diffusivity and ED is the eddy diffusivity. Usually

ED :> D m' Equation (6-42) assumes the packing is homogeneous, there are no
radial gradients, there are no chemical reactions other than adsorption or ion
exchange, and there are no phase changes other than adsorption. The first term
is the accumulation of solute in the mobile fluid, the second term is the accu-
mulation in the stagnant fluid in the pores, and the third term is the accumula-
tion of solute adsorbed to the solid. The terms Ci,pore and <t are the volume
average values inside the porous particles.
_ 3 R, (6-43a)
Ci,pore = R3
p
J0 C;,pore fl dr
_ 3 R,
-u,pore = -R 3
n·
P
I
n. fl dr
"U
(6-43b)
The 3/R~ term comes from the volume of the spherical particle. Term 4 in Eq.
(6-42) is the term for (out-in) due to bulk fluid movement, and the fifth term is
the term for (out-in) due to axial dispersion and diffusion.
To use Eq. (6-42) we must relate the average amount adsorbed <t to the
fluid concentration Ci' Many ways of doing this have been proposed (e.g. Ruth-
ven, 1984; Yang, 1987). Generally, fluid concentration in the pores Ci,pore is cal-
culated and CJi is assumed to be in equilibrium with the local pore concentra-
tion. Mass transfer inside the particles occurs due to some combination of nor-
mal (Fickian) diffusion in the pores, Knudsen diffusion in the pores, and sur-
270
face diffusion. For spherical particles the equation for nonnal pore diffusion is
(6-44a)
where D mp is the effective molecular diffusivity in the pore. The effective pore
diffusivity is usually related to the molecular diffusivity,
f.p Dm (6-44b)
D mp = --.:.---
't
where 't is the tortuosity factor. The pellet porosity Ep in the numerator takes
into account the solid volume which is unavailable for diffusion. The tortuosity
is the ratio of the actual distance of travel divided by the particle radius. Exper-
imental ranges for't are given in Tables 6-2, 64, 6-5 and 6-6. If there is drasti-
cally hindered diffusion or retardation of the solute, 't can be as high as 100 and
D mp can be two orders of magnitude lower than D m' The safest approach is to
use experimental data to detennine D mp' although expressions for estimating
effective diffusivities are discussed by Ruthven (1984).
For gas systems with high loadings of adsorbate, surface diffusion is an

important mechanism and can account for as much as 80% of the total flux
(Yang, 1987). In surface diffusion the adsorbate does not desorb, but instead
diffuses along the surface. We can define a surface diffusion coefficient, D s ,
by analogy to Fick's law
Flux=-D dq (6-45 a)
8dr
Unfortunately, D 5 shows a strong dependence on surface coverage, and Eq.

(645a) is difficult to use in models. It is more common to write the surface
diffusion effect as a lumped parameter expression,
(6-45b)
where ks is the surface diffusion coefficient and q* is the equilibrium value of

271
surface concentration. For linear systems q = AC and Eq. (6-45b) can be writ-
ten as
(6-45c)
Correlations for ks are discussed in Section 8.3.
The pore fluid concentration at the wall ci,pore ~) can be related to the
fluid concentration outside the pores, Ci, by the mass transfer equation across
the surface film.
(646)
In this equation k f is the film mass transfer coefficient in units such as mls and
~ is the external surface area per unit particle volume (m 2jm 3 ). For spherical
particles ~ = 3/Rp • The left hand side of Eq. (646) is the accumulation of
solute in the particle while the right hand side is the transfer rate across the sur-
face film.
Equations (6-42) to (6-46) must be solved simultaneously with the

appropriate equilibrium isotherm, initial and boundary conditions. This set of
equations is difficult to solve. Although the Rosen solution does solve a similar
set of equations for linear isotherms (see Section 7-8), it is common to drasti-
cally simplify the equations before solving. We will often assume that a
lumped parameter expression similar to Eq. (6-46) is adequate.
(647)
In this equation the entire particle is treated as having a constant amount

adsorbed <t, and the stagnant fluid inside the pores is assumed to be in equili-
brium with the solid. For gas systems the accumulation of adsorbate in the
stagnant fluid is often ignored. Since the distributed concentrations Ci and ~
have been lumped, this is called a lumped parameter mass transfer equation.
272
Thus c7, the fluid concentration in equilibrium with q, replaces Ci,pore. Equa-
tion (6-47) would be correct if mass transfer across the film were the control-
ling step. However, Eq. (6-47) is often used in other situations. Then lcm is the
effective mass transfer coefficient. Correlations for lcm are discussed in Section
8.3.
Alternate forms of the mass transfer equation and the solute mass bal-
ance are used when a single porosity model is used. The solute mass balance
is,
where Pp is the particle density of solid including the fluid in the pores and <Ii
now includes both adsorbed solute and solute in the stagnant fluid in the pores.
Terms 1 and 2 are the accumulation terms while term 3 is the convection term
and term 4 includes axial dispersion and diffusion. Equations (6-43) to (6-45)
are unchanged. The coupling equation for mass transfer across the film
becomes
(6-49)
When the lumped parameter mass transfer expression is used, this equation is
(6-50)
Equations (6-48) to (6-50) are the appropriate equations for non-porous solids
and for gel-type ion exchange resins. These equations are somewhat simpler
than Eqs. (6-42), (6-46) and (6-47), and thus the single porosity equations are
often used when the two or three porosity models would be physically more
realistic.
273
6.8. BASIC ENERGY BALANCES
For non-isothermal systems an energy balance for both phases and an energy
transfer equation are required. We will assume that no electrical or magnetic
fields are present, radiant heat transfer is negligible, viscous heating can be
neglected, kinetic and potential energy changes are small, and density is con-
stant. Then the energy balance for both phases is
_.
In Eq. (6-51) T and Tamb are the stagnant fluid and ambient temperatures,
respectively, and hw is the heat transfer coefficient for the column walls.
Terms 1, 2, 3, and 7 represent accumulation in the mobile fluid, the stagnant
fluid, the solid and the column walls, respectively. Term 4 is (out-in) with the
flowing fluid while term 6 is (in-out) due to heat transfer through the column
The volume average values T and

_.
walls. Finally, term 5 represents the axial dispersive flow of energy.
1'. are determined by equations

analogous to Eqs. (643a,b)
(6-52)
The temperatures inside the particles can be estimated from a diffusion equa-
tion
(6-53)
where r and Ts are locally in equilibrium (T. = Ts) and ke is the effective
274
thennal diffusivity including radiation effects. Equations (6-51) and (6-53) are
coupled by film diffusion.
where h is the heat transfer coefficient in units such as m/s. These equations
are analogous to the mass transfer expressions. Equations (6-51) to (6-54) are a
fonnidable set. Suprisingly, the equivalent single porosity set for heat transfer
with no adsorption can be solved with a few simplifications (see Problem 7-E3
and Section 7.8.).
A relatively simple fonn of the energy transfer equation is obtained by

assuming that an overall linear driving force model is adequate. Then the
energy balance for the solid is,
(6-55)
In this equation IIp is the particle heat transfer coefficient, 3p is the particle sur-
face area per unit volume, and Mi.ds is the heat of adsorption, kcal/gm adsor-
bate. Usually Eq. (6-55) is satisfactory.
The single porosity fonns of Eqs. (6-51), (6-53), (6-54) and (6-55) are
often used. These are easily obtained by setting Ep = 0, replacing Ps with Pp '
and replacing Cp,s with CP,p which is the particle heat capacity including solid
and ft uid in the pores.
Complete solution of Eqs. (6-42), (6-47), (6-51) and (6-55) or of the

corresponding single porosity equations is difficult. The equations are coupled.
That is, the mass balances depend on the energy balance equations since equili-
brium is temperature dependent. The energy balance is coupled to the mass
balances through the last term in Eq. (6-52),
275
6.9. LOCAL EQUILIBRIUM THEORY
How do these mass transfer and energy equations relate to the solute
movement theory? The solute movement theory can be derived from the mass
transfer and energy transfer equations if a variety of additional assumptions are
made. These assumptions are all the assumptions listed in Table 6-7. The
results are drastically simplified partial differential equations. With local
equilibrium (assumption # 15) the diffusion equation (6-44) and the mass
transfer Eqs. (6-47) or (6-50) are not required. When the assumptions in Table
6-7 are used, the solute mass balance for a single solute, Eq. (6-42), becomes
Note that with local equilibrium q = q and c = cpore. From the chain rule,
aq aq ac aq aT (6-57)
-=--+--
at ac at aT at
Substituting this into Eq. (6-56) and rearranging we obtain
where lis is given by Eq. (6-23) with fj,q/fj,c replaced by aq/ac.
Appropriate initial and boundary conditions are,
(6-59a)
c=co(z), t=O
(6-59b)
For isothermal systems or for square temperature waves, iIT(at = 0, and the
right-hand-side of Eq. (6-58) is zero. Then, Eq. (6-58) is easily solved by the
method of characteristics (Sherwood et ai., 1975; Wankat, 1981). A simple
276
approach to the method of characteristics is to compare Eq. (6-58) to the total

derivative of concentration with respect to time.
ac
-+--=-
ac dz de (6-60)
at az dt dt
If
dz (6-61)
-=lls
dt
then, the left hand side of Eq. (6-58) is the same as the left hand side of Eq. (6-
60). This forces (when aT/at = 0)
dc =0 (6-62)
dt
Thus concentration is constant along the lines represented by Eq. (6-61). This
result is exactly the same as the solute movement theory developed earlier
using a physical argument
The local equilibrium or solute movement model is extremely simple

compared to other adsorption models. This simple model includes the first and
fourth terms in Eq. (642). The second and third terms are also included but
with the assumption of local equilibrium. This assumption means that Eq. (6-
47) is automatically satisfied. The fifth term in Eq. (6-42) is ignored. Any
equilibrium form can be used. The model thus includes diffuse waves due to
nonlinear equilibrium, but does not include mass transfer and dispersion which
also broaden zones. Extensions of this model can be made to relax assump-
tions lOb to 14 in Table 6-7 (see Sections 7.4 and 8.5).
The mass transfer equations will be solved for systems with finite disper-
sion and mass transfer resistances in Chapters 7 and 8. Chapter 7 considers
systems where the isotherm is linear and then applies the solutions to chroma-
tography. Chapter 8 uses both the solute movement theory and non-linear solu-
tions of the mass transfer equations to examine a variety of adsorption separa-
tions.
277
A simplified solution for the energy balance can be obtained by assuming

local equilibrium T = 1'* =Tw =T. =Tr , assuming the heat of adsorption term
is negligible, assuming hw is zero, and ignoring axial dispersion and diffusion.
These assumptions decouple the mass and energy balances, and make Eq. (6-
55) automatically satisfied. The mathematical manipulations to solve Eq. (6-
51) for this local equilibrium theory are essentially the same as those used to
solve the mass balances (see Problem 6-C9). The results are exactly the same
as those derived earlier for the solute movement theory by physical arguments.
6.10. SUMMARY· OBJECTIVES
At the end of this chapter you should be able to meet the following objectives:
1. Discuss which adsorbents are appropriate for a particular separation.
2. Use various equilibrium isotherm forms and determine the isotherm

parameters from equilibrium data.
3. Use the solute movement theory for both linear and non-linear isotherms.
4. Explain the movement of energy waves and the effect of temperature on

solute adsorption. Calculate thermal wave velocities and concentration
changes.
5. Write and identify the terms in the mass and energy balances and the mass
and energy transfer equations.
6. Show that the local equilibrium model is essentially the same as the solute
movement theory.
REFERENCES
Baker, B. and R.L. Pigford, "Cycling zone adsorption: quantitative theory and
experimental results," Ind. Eng. Chem. Fundam., 10,283 (1971).
Bidlingmeyer, B.A., and Warren, F.V., Jr., "Column watch: Small-molecule gel
278
permeation chromatography: A technique for everyone," LC-GC, 6, 780

(1988).
BonsaI, R.C., J.-B. Donnet, and F. Stoeckli, Active Carbon, Marcel Dekker,
New York, 1988.
Breck, D.W., Zeolite Molecular Sieves, Wiley, New York, 1974.
Brunauer, S., P.R. Emmett, and E. Teller, "Adsorption of gases in multimolecu-

lar layers", J. Am. Chern. Soc., 60, 309 (1938).
Collins, J., "Where to use molecular sieves," Chern. Eng. Prog., 64 (8), 66
(1968).
Dobbs, R.A. and Cohen, J.A., "Carbon adsorption isotherms for toxic organ-
ics," EPA, Cincinnati, OR, 45268, EPA-600/8-80-023, April, 1980.
Faust, S.D. and D.M. Aly, Adsorption Processes for Water Treatment, Butter-
worths, Boston, 1987.
Fox, C.R., "Industrial wastewater control and recovery of organic chemicals by

adsorption," in F.L. Slejko (Ed.), Adsorption Technology, Marcel Dekker, New
York, 1985, 167.
Fritz, W. and E.V. Schluender, "Simultaneous adsorption equilibria of organic

solutes in dilute aqueous solution on activated carbon," Chern. Eng. Sci. 29,
1279 (1974).
Janson, J.C. and P. Redman, "Large-scale chromatography of proteins," in

Fiechter, A. (Ed.), Advances in Biochemical Engineering, Vol. 25, Springer-
Verlag, Berlin, 1982,43-99.
Kovach, JL., "Gas-phase adsorption," in P.A. Schweitzer (Ed.), Handbook of

Separation Techniques for Chemical Engineers, McGraw-Hill, New York,
1979, Sect. 3.1.
279
Langmuir, 1., "The adsorption of gases on plane surfaces of glass, mica and pla-
tinum," J. Am. Chem. Soc., 40, 1361 (1918).
Lee, M.N.Y., "Novel separations with molecular sieves adsorption," in N.N. Li

(Ed.), Recent Developments in Separation Science, Vol. I, CRC Press, Boca
Raton, FL, 1972,75.
LeVan, M.D. and T. Vermeulen, "Binary Langmuir and Freundlich isotherms

for ideal adsorbed solutions," J. Phys. Chem., 85, 3247 (1981).
Lewis, W.K., E.R. Gilliland, B. Chertow, and D. Bareis, "Vapor-adsorbate

equilibrium. III." J. Am. Chern. Soc. 72, 1160 (1950a).
Lewis, W.K., E.R. Gilliland, B. Chertow, W.P. Cadogan, "Adsorption equili-

bria: hydrocarbon gas mixtures," Ind. Eng. Chern., 42, 1319 (1950b).
Lydersen, A.L., Mass Transfer in Engineering Practice, Wiley, New York,

1983, Chapt. 7.
Ruthven, D.M., Principles of Adsorption and Adsorption Processes, Wiley-

Imerscience, NY, 1984.
Ruthven, D.M., "Zeolites as selective adsorbents," Chern. Eng. Prog., 84 (2),

42 (Feb. 1988).
Sherwood, T.K., R.L. Pigford, and C.R. Wilke, Mass Transfer, Chapl 10,
McGraw-Hill, N.Y., 1975.
Tsou, H.-S. and E.E. Graham, "Prediction of adsorption and desorption of pro-
tein on dextran based ion-exchange resin," AlChE Journal, 31, 1959 (1985).
Valenzuela, D. and A.L. Myers, "Gas adsorption equilibria," Separ. Purific.

Methods, 13, 153 (1984).
Valenzuela, D.P. and A.L. Myers, Adsorption Equilibrium Data Handbook,

Prentice-Hail, Englewood Cliffs, NJ., 1989.
280
Vaughan, D.E.W., "The synthesis and manufacture of zeolites," Chem. Engr.

Prog.,84 (2),25 (Feb. 1988).
Venneulen, T., M.D. LeVan, N.K. Hiester, and G. Klein, "Adsorption and ion
exchange," in Perry, R.H. and D.W. Green (Eds.), Perry's Chemical
Engineers' Handbook, 6th ed., Section 16, McGraw-Hill, N.Y., 1984.
Wankat, P.C., "Cyclic separation techniques," in Rodrigues, A.B. and Tondeur,

D. Percolation Processes, Theory and Application, Sijthoff and Noordhoff,
Alphen aan den Rijn, Netherlands, 1981,443-516.
Yang, R.T., Gas Separation by Adsorption Processes, Butterworths, Boston,

1987.
HOMEWORK
AI. If V col is the volume of an empty column, relate Ee and Ep to

V colt Ve , Vi> and Vo.
A2. You suspect that adsorption of a particular solute from a liquid satisfies
the Freundlich isotherm. What would you plot to determine if the data
does satisfy the Freundlich form? How could you determine A and n in
Eq. (6-14)?
A3. Show that (1 + KA CF) in the Langmuir isotherm is equivalent to the

relative volatility.
A4. Explain each term in Eq. (6-21) in detail.
A5. For the isotherm shapes shown in Figures 6-3b or 6-3d show that Eq.
(6-31) is valid. Do this based on slopes of the isotherm at a given con-
centration.
281
A6. Explain in your own words the formation of shock and diffuse waves.
When does each occur?
A7. What conditions are required for Eqs. (6-6a), (6-6b) , and (6-12a) to
become linear?
AS. Equations (6-42) and (6-4S) are alternate expressions for the mass bal-
ance. Show how these equations are related. Compare the meaning of
q for these two models. Which equation would be appropriate for:
a. ion exchange
b. size exclusion chromatography
c. adsorption of a gas with a porous adsorbent
d. adsorption of a liquid with a porous adsorbent
A9. Why do liquid systems usually satisfy Eq. (6-37a) while dilute gas sys-
tems satisfy Eq. (6-37b)?
A 10. Workers entering adsorption columns containing zeolites always wear

breathing apparatuses. Explain why.
All. Develop your key relations chart for this chapter. A key relations chart
lists everything (equations, words, figures) you want to know on one
page.
C. Derivations
Cl. Show that Eq. (6-16) has the same form as the Langmuir equation (6-
12a) except it is now in terms of mole fractions instead of q and c.
C2. Many models use a single porosity, ee' and treat the packing material as
a solid or gel without pores. This model is appropriate for ion exchange
(see Chapter 9) since gel resins which do not have internal pores are
often used. The model has also been used for adsorbents with porous
particles because it is simpler. The equilibrium expression q = fCc) must
now include solute contained in the pore fluid if there are pores. Use
282
the single porosity model to derive the solute wave velocity which is
(6-23a)
Note that when Ep = 0, Pp = Ps.
C3. Separation of fructose and glucose often uses a gel type ion exchange
resin in the calcium fonn. Separation is based on different complexa-
°
tion equilibrium of the fructose and glucose with the calcium (this is not
ion exchange). If Ep = and equilibrium is of the fonn Cli =Ai Ci where
Cli and ~ are both in the units g sugar/lOO mI, derive the equation for
solute velocity. HINT: Read problem 6-C2 first
C4. Derive the expression for the diffuse wave solute velocity for Langmuir
equilibrium (Eq. (r 12a).
C5. Derive Eq. (6-40) for linear isotherms starting with Eq. (6-39a).
C6. Derive Eq. (6-39b) for the case shown in Figure 6-9b and in Eq. (6-
37b). Also derive the equations for linear isotherms equivalent to Eqs.
(6-40) and (6-41).
C7. Derive the equation equivalent to Eq. (6-39b) when Eq. (6-37c) is valid.
Show that the equation equivalent to Eq. (6-40) for linear isotherms
predicts negative concentrations.
C8. Making the assumptions listed in Table 6-7, derive Eq. (6-58).
C9. Solve Eq. (6-51) for the local equilibrium model.
D. Single Answer Problems
D 1. Data for the adsorption of propane on silica gel at 0, 40 and 100°C are
283
given by Lewis et al. (1950a) and are repeated below. Over what pres-
sure range is the Langmuir isotherm a good fit? Calculate the values of
C1max, KA and £lif, (q is gram moles per kg adsorbent, p is mm Hg).
O°C 4Q°C 100°C
p q p q p q
16.6 0.2137 10.1 0.0418 96.4 0.0531
37.7 0.3960 27.9 0.0900 119.0 0.1471
64.4 0.5678 46.7 0.1407 406.5 0.2087
93.2 0.7307 92.2 0.2580 601.5 0.2781
129.3 0.9010 136.9 0.3352 753.7 0.3360
218.4 1.259 204.0 0.4470
298.8 1.520 282.0 0.568
429.4 1.881 373.0 0.6871
587.1 2.241 462.6 0.7876
762.6 2.582 554.9 0.904
643.0 0.9908
768.9 l.128
D2. The adsorption of acetic acid from aqueous solution onto activated car-
bon was extensively studied by Baker and Pigford (1972). Their data is
available in Example 6-2. We wish to do an experiment at a superficial
velocity of 10 cm/min (superficial velocity = Ec v). A column 100 cm
long is packed with activated carbon at 4°C. The bed is saturated with a
fluid containing 0.020 moles per liter. A feed of concentration 0.50
moles/liter at 4°C is introduced into the column for 10 minutes and is
followed by fluid of concentration 0.020 moles/liter at 4°C. Predict the
complete outlet concentration profile.
D3. Repeat Problem 6-D2, but with feed and following stream (concentra-
tions 0.50 and 0.020) at 60°C instead of 4°C. Bed is initially at 4°C.
284
04. Ching and Ruthven, A1ChE Symp. Ser., 81, (242), 1, (1985) found that
the equilibrium of fructose and glucose on ion exchange resin in the cal-
cium fonn was linear for concentrations below 5 g/100 ml. Their
equilibrium expressions are: qo =0.51 co, <IF =0.88 CF, at 30°C where
both q and c are in g/100 ml. For this resin Ep = 0 and Eo = 0.4. A
chromatographic column one meter long is operated with water flowing
at a superficial velocity of 15 cm/min.
a. Ifa very short feed pulse (- 1 sec) is input, at what time will the
fructose and glucose peaks exit the column?
b. Assuming no zone spreading, what is the longest feed pulse (in

minutes) which can be input which will just separate fructose
and glucose in this column? At what time can the next feed
pulse be inttoduced? What % of the time can feed be introduced
to the column?
Note: See problem 6-C3 before doing this problem.
D5. We have a 60 cm long column packed with activated carbon. The

column initially is filled with distilled water at 4°C. A feed at 4°C con-
taining 0.25 moles acetic acid per liter is inttoduced at t = 0 minutes.
This continues for 20 minutes. After 20 minutes, heating is used as
specified in parts a and b. The same feed concentration is used but T =
60°C. Throughout the process the superficial velocity is 10 cm/min.
Predict the outlet concentration profiles. Data is given in Example 6-2.
a. The heating at 20 minutes is done through a jacket. Assume the

entire column is heated instantaneously to 60°C.
b. The column is insulated. Heating occurs because of the hot feed

(at 60°C) only (a thermal wave is generated).
c. Compare parts b and c.
D6. Estimate an average separation factor, aEP = (YE/XE)/(YP/Xp), for the

ethylene-propylene equilibrium data shown in Pigure 6-3e.
285
D7. A column packed with activated carbon is initially at 20°C. The
column is 1 meter in diameter and 4 meters long. The carbon has the
same properties as the carbon in Example 6-2. As part of the testing
procedure the column is filled with water at 20°C. At 8:00 am. a water
feed at 40°C is turned on and enters the bottom of the column.
Superficial velocity is 20 cm/min. At 8:08 this water is turned off.
Flow is reversed and water at O°C enters the top of the column at a
superficial velocity of 10 em/min. At 8:20 this stream is shut off.
Water again enters the column from the bottom. This water is at 60°C
and has a superficial velocity of 20 cm/min. Predict the temperature
profile in the bed at 8:25 a.m.
D8. The data listed below were obtained by Reich, Ziegler, and Rogers, Ind.
Eng. Chern. Process Des. Develop.• 19. 336 (1980) for the adsOIption of
methane on Type BPL activated carbon at 301.4 K. P is partial pressure
in psia and q is mg moles/g carbon.
a. Does this data satisfy the Langmuir equilibrium isotherm? If

yes, what are good values of Q.max and KA? If no, can the Lang-
muir isotherm be used for some range of pressures? For what
range of partial pressures is the Langmuir a good fit, and what
values of Q.max and KA should be used?
b. We have a column at 250 psia filled with hydrogen (which does

not adsorb) and start adding a feed which is 20% methane in
hydrogen at 250 psia. Predict at what time the shock wave will
exit the column. The column is 2 meters long. Superficial
velocity is 80 cm/min. Temperature is 301.4 K. Use the carbon
properties listed in Example 6-2. Solve this problem using the
data and using the appropriate Langmuir fit obtained in part a.
Compare your results. Note: You must rederive Eq. (6-30) or
change the equilibrium data to have the units work.
c. Once the column is totally saturated with the 20% methane feed,
the column is desorbed with pure hydrogen. Temperature, velo-
city, and pressure are the same as in part b. Predict the outlet
concentration profile.
286
P q
19.1 0.765
39.8 1.304
67.8 1.825
98.3 2.256
129.8 2.604
168.6 2.953
209.4 3.256
249.3 3.501
D9. A column packed with alumina initially is saturated with pure cyclohex-
ane. At t = 0 a stream containing c = 0.01 gmole/L anthracene in
cyclohexane is input at a superficial velocity (Ee v) of 30 cm/min. After
10 minutes a stream containing c = 0.02 gmole!L is input at the same
velocity. Data are given in Example 7-1.
a. If the colwnn length is Ll = 75 cm, predict the outlet concentra-

tion profile.
b. If the column length is Lz = 150 cm, predict the outlet concen-

tration profile.
c. It is often tempting to linearize the nonlinear isothenn. This can

be done by using a linear isothenn calculated with
c;:: CAvg = 0.01 or c = 0.02. For example, if c = 0.02 then q =
0.0512. The linearized isothenn is set so that q = 0.0512 when c
= 0.02. Thus, the linear equation is q = 2.56 c. Use the linear
isothenn to redo Parts a and b. Compare your result with the
solutions to Parts a and b.
DIO. Fructose adsorption on a dihydroxyborylphenyl succinamyl derivative

of aminoethyl cellulose is very pH dependent. Equilibriwn isothenns at
25°C are (Busbice and Wankat, J. Chromatogr.,1l4, 369, 1975):
0.0582 c
pH 8.0:
q = 1 + 1.52 c
287
0.000165 c
pHS:
q = 1 + 1.52 c
where q is gig adsorbent and c is glLiter. The pH wave velocity was

measured as UpH =0.40 v. The feed concentration Cf = 2 gIL. Ep =
0.45, Ee =0.4, Ps = 1100 gIL, Kd = 1.0. L = 50 cm, v =20 cm/min
The column is initially saturated with feed solution (c = CF) at pH 8.0.

At t = 0 we input feed solution at pH 5.0. Predict the outlet concentra-
tion profile. Assume local equilibrium and use solute movement theory.
Chapter 7
LINEAR THEORIES OF SORPTION AND CHROMATOGRAPHY
In this chapter we will focus on linear theories for solving the mass transfer
equations. These results are applicable for both adsorption and chromatography
when the equilibrium is linear. Since chromatography is often operated at low
concentrations where the equilibrium is linear, theories and applications for
chromatography will be explored in this chapter.
The chapter starts with an introduction to a variety of types of analytical

chromatography. Next, the application of solute movement theory to chroma-
tography is briefly explored, followed by development of superposition princi-
ples in linear systems. Various linear solutions including a dispersion model, a
staged model, Gaussian solutions, the Van Deemter equation, and the Rosen
solution are discussed in Sections 7.4 to 7.8. After presenting programming
and column switching, we then briefly discuss large scale applications of
chromatography.
7.1. TYPES OF CHROMATOGRAPHY
Chromatography refers to a broad range of separation methods used to separate

complex mixtures. Because of the very broad range of methods included in the
general term, it is relatively easy to confuse methods. In this chapter chroma-
tography will refer to a system with a fixed bed of particles which has a fluid
(gas or liquid) flowing through it. This fluid is called the carrier gas or solvent.
A feed pulse is input into the flowing fluid. The components of the feed pulse
separate because they have different solute velocities. The solute movement
theory of Chapter 6 can be used to explain the separation. In this section we
will briefly discuss a variety of analytical chromatography methods.
288
289
Chromatography is an extremely powerful analytical tool for separating

and analyzing complex mixtures. A schematic of an analytical chromatograph
is shown in Figure 7-1. For liquid systems a pump is used to push the fl uid
through the column. A pulse of feed is injected into the system as was shown
in Figure 6-3a. The column is often enclosed in an oven to control the tem-
perature. After the column, the detector analyzes the stream for some parame-
ter such as refractive index or ultra-violet (UV) absorbance which can be
related to concentration. The purpose of the column is to separate the mixture
into peaks which contain only one component in addition to the solvent. Then
the detector response will be directly related to concentration. This is illus-
trated in Figure 7-1b where the chromatogram for separation of a protein sam-
ple by analytical ion exchange chromatography is shown. (perkins et al, 1987).
The ordinate in Figure 7-1b is the UV absorbance at 280 nm. For dilute sam-
ples this absorbance is directly proportional to the protein concentration. Note
that the separation is quite sensitive to pH. In analytical systems the column is
a small but critical part of the entire system. By changing the chemistry of the
column, the separation can be vastly different. Analytical applications are usu-
ally at low concentrations where both the isotherm and the detector responses
are linear.
In gas systems a carrier gas such as helium or hydrogen is used. The sol-
vent tank and pump are then replaced with a gas cylinder under pressure.
Injection is often done with a syringe into a small heater which vaporizes the
liquid sample. The most common detectors are thermal conductivity detectors
and flame ionization detectors (FID).
Two types of gas chromatographs with packed beds are used. In gas
adsorption chromatography an adsorbent is used. This method is appropriate
for separating mixtures which are normally gases. It has the disadvantage that
the isotherms are often non-linear even at low concentrations (see Figure 6-3b).
This causes the peaks to be skewed and makes analysis more difficult. Com-
mon adsorbents include zeolites, silica gel and activated alumina (Supina,
1985; Zweig and Sherma, 1972) which were discussed in Chapter 6.
The more common type of analytical gas chromatograph is gas-liquid

chromatography (GLC). In gas-liquid chromatography an inert porous solid is
290
C
a
I
u Oven
m
n
Feed Injection
Solvent Pump
Tank
b B B
(b) 0
(0)
C
0
A
A
C
~
c:
a
JJ
~
JJ
c(
c
c
I
5
Time (min)
Figure 7-1. Analytical Chromatograph. a Equipment schematic. b. 1so-

cratic ion exchange chromatography of 1 mg/mL protein stan-
danis at (a) pH 5.85 and (b) pH 6.5. Column: 10 cm x 4.6 mm
7-5J.1ITI WCX, 3OO-A pore size; mobile phase: 0.05 M phos-
phate with 0.2 M NaCI; flow rate: 1.5 mL/min; detection: UV
280 nm, 0.08 AUFS; injection: 1O!J.L. Peaks: A = ovalbumin,
B =conalbumin, C = cytochrome c, D = lysozyme. (perkins et
ai, 1987). Reprinted with permission from LC-GC, 5, 421
(1987). Copyright 1987, Aster Pub. Co.
291
coated with a viscous, high boiling liquid. This stationary liquid phase is what
does the separation. An extensive list of GLC stationary phases is given by
Zweig and Sherma (1972). The solid should be inert, have a high porosity and
be inexpensive. Diatomaceous earth is most commonly used although a variety
of other materials are available (Supina, 1985; Zweig and Sherma, 1972). The
solutes in the feed can dissolve (or condense) in the stationary liquid phase and
at a later time vaporize into the flowing gas. Separation is essentially based on
relative volatility and is really an absorption-stripping operation. Equilibrium
is often given by Eq. (6-18). At low concentrations the equilibrium reduces to
Henry's law, Eq. (6-4a). Since the porous solid has a very thin coating of sta-
tionary liquid and the area for mass transfer is large, the rate of mass transfer is
high. Thus the HETP will be low and the column has the equivalent of a large
number of equilibrium stages. GLC was invented by James and Martin (1952)
and revolutionized analytical chemistry. One major advantage of gas chroma-
tographs in analytical systems is the detectors are very sensitive and work for
any compound. Modern systems are covered by Grob (1985) and McNair and
Bonelli (1969). GLC has been tried for large scale separations but has been
significantly less successful. One problem is temperature stability since even
slow vaporization of the stationary phase can, over time, significantly change
the column characteristics. This loss of stationary phase also contaminates the
product.
Capillary gas chromatography is a commonly used analytical method. A
glass or fused silica capillary with a coating of adsorbent or high boiling sol-
vent on the wall is used. Since the amount of stationary phase is small, the
capacity is limited. However, the open capillary has little resistance to mass
transfer and very sharp separation are observed. In addition, the open tube has
a very low pressure drop and hence very long coiled columns (100 m or more)
can be used. The result is a device which can rapidly separate complex mix-
tures. The capillary GC would be difficult to scale up for large-scale applica-
tions.
A variety of liquid chromatography systems have been developed. The

oldest technique is liquid adsorption chromatography (see Johnson and Steven-
son (1978) or Snyder and Kirkland (1979». In this case an adsorbent is used.
This is a common method for large scale applications and for separation of
292
organic compounds. Silica gel and activated alumina are the most common
adsorbents, and were discussed in Chapter 6. Details of chromatographic pack-
ings are given by Zweig and Sherma (1972). In large-scale applications liquid
adsorption chromatography is often cheaper than other chromatography
methods since the adsorbents are quite inexpensive.
In liquid-liquid chromatography (LLC) a stationary liquid phase is coated
onto an inert, porous solid. The separation is thus essentially an extraction
operation. LLC was invented by Martin and Synge (1941) for which they
received the Nobel prize in chemistry in 1951. Modem applications are dis-
cussed by Johnson and Stevenson (1978) and Snyder and Kirkland (1979).
This method is very useful for separating non-volatile compounds which can-
not be separated by gas chromatography. The same porous solids are used as
supports as in GLC. A large number of different stationary phases and
modified mobile phases are used to take advantage of chemical interactions.
Bleeding which is the slow loss of stationary phase into the mobile phase can
be a problem and causes contamination of the product.
Modem liquid chromatography differs from LLC in two major ways.
First, bonded phases are usually used. The stationary phase is now chemically
attached to the inert solid. Very often silica is used as the solid. Great care has
to be taken to cover all the active sites to prevent undesired adsorption. For
large molecules special large-pore silicas with pores up to 300 nm in diameter
are used. The coating still acts as a very thin liquid phase, but there is no loss
of stationary phase by bleeding. The most common bonded phases use C8 or
C I8 compounds attached to silica gel. Water is the most common solvent. For
historical reasons this method is called reverse phase chromatography. The
second major difference is that now short columns containing very small diam-
eter packings are operated at high velocity and high pressure drops. The result-
ing high performance liquid chromatography (HPLC) has the equivalent of a
huge number of equilibrium stages (100,000 is not uncommon). Modem
analytical columns are often 10 to 25 cm long, 4 to 8 mm id, use packings from
3 to 10 JlIl1 (1 Jlm is 10-{i m) in diameter, and have pressure drops of several
thousand psi. One advantage of liquid chromatography is that changing the
solvent can have a major effect on the distribution coefficients and hence on the
separation (Schoenmakers, 1986). See Johnson and Stevenson (1978) or
293
Snyder and Kirkland (1979) for details of analytical applications and details
about bonded phase packings. Large scale applications are discussed later.
One particular type of liquid chromatography which has been extensively

used for protein purification is size exclusion chromatography, SEC, (also
called gel filtration or gel permeation chromatography). In this system there is
no adsorption. Thus ACT) in Eq. (6-24) and (t1q/t1c) in Eq. (6-23) are zero.
The only term which is different for different solutes is Kd,i which depends on
steric exclusion. Usually, the packing is a gel with a well defined distribution
of pores. The method is particularly useful for separation of large molecules
(Kd = 0) from small molecules (Kd = 1.0). The large molecules exit the column
first. This is done commercially for purification of plasma proteins and as one
step in purification of many other proteins, and analytically for protein separa-
tions and for analyzing polymer molecular weight distributions. A typical cali-
bration curve for proteins is shown in Figure 7-2 (Andrews, 1964). Equation
(6-3) can be used to determine Kd from Ve' Details of size exclusion chroma-
tography are given by Determan (1969), Janson and Hedman (1982), and Yau
et al (1979). Size exclusion chromatography of biological compounds often
uses crosslinked polymer gels such as agarose, polydextrans (Sephadex) and
polyacrylamides (Biogel) which are compressible. These gels can only operate
at low pressure drops and hence low flow rates (from 10 to 200 cm/hr depend-
ing on the gel). Molecular weight cutoffs for these packings were given in
Table 6-1. Extensive tables on the properties of SEC packings are given by
Janson and Hedman (1982) and Zweig and Sherma (1972).
Supercritical fluid chromatography is a new method (Gere, 1983). The

carrier fluid is a supercritical fluid. Compared to liquids, supercritical fluids
have a solubility and density that are roughly half as large. The diffusivity is
roughly an order of magnitude larger while the viscosity is over an order of
magnitude lower. This gives significantly lower L\p and HETP. Carbon diox-
ide spiked with a modifier to enhance selectivity is the usual carrier. Applica-
tions of supercritical fluid chrometography in analytical chemistry are becom-
ing common. Large-scale applications have not been announced yet.
The types of chromatography discussed up to now usually operate in a

migration mode. In migration chromatography the solutes sorb and desorb
294
220
Soya-bean trypsin inhibitor
220 200
Cytochrome c dimer
! -g
Ij
200 Chymotrypsinogen 180
Sucrose
'":' 180 1W §
Ovalbumin
::
::: 160
,-I::
Serum albumin 140 l
-.: 140
• Serum albumin dimer 120 t
r t1I
~
rOb~~~MgIObt j ': S
-" Pseudomonas
?: 120 cytoch rome c-551
?- Cytochrome c
fo 100
:... Ribonuclease
80 (X-Lactalbumin
60 Myoglobin
300 2 X 10 3
Molecular weight
Figure 7-2. Plots of elution volumes, V c against log (molecular weight) for
SEC of proteins on Sephadex G-75 (.) and G-lOO (0) columns
[2.4 cm. x 50 cm.; equilibrated with 0.05 M-tris hydrochloride
buffer, pH 7-5, containing KCI (0.1 M)]. (Andrews, 1964).
Reprinted with permission from Biochem. J .• 91, 222 (1964).
Copyright 1964.
many times and all solutes move at a finite solute velocity given by Eqs. (6-24)
or (6-23). The solute movement diagram for migration chromatography was
Ion exchange and affinity chromatography are usually operated with a

rather different operating method called on-off chromatography (Wankat,
1986). In qffinity chromatography the packing is modified by chemically
attaching a compound which has a specific biochemical affinity for the desired
molecules. The affinity matrix should be inert, readily accessible to the
molecules being purified, rigid or close to rigid, easy to chemically modify and
cheap. Agarose (a high molecular weight polysaccharide) has been most
commonly used although it is not cheap. A variety of attachment procedures
are used (Dechow, 1989; Gribnau et aI, 1982). The resulting affinity chroma-
295
tography packing can be very specific for the desired molecule. Thus, theoreti-
cally, one can pull out one molecule from a million undesired molecules. In
actual practice, the presence of a large number of impurities can foul the pack-
ings. Thus affinity chromatography has been used as one of the last steps in
commercial processing of biochemicals (Darbyshire, 1981; Dechow, 1989; Hill
and Hirtenstein, 1983). Large scale applications are illustrated in Section 7.11.
Ion exchange chromatography is also used for purification of biologicals and is
usually used early in the process since ion exchange has high capacities and the
resins are inexpensive. In the on-off mode of operation a molecule stays in the
liquid phase until it finds an open site for sorption (either by specific affinity or
electrical forces). Sorption is so strong that once sorbed the molecule does not
come off. In order to remove the sorbates the conditions of the feed must be
changed. Usually this means eluting with a solution of changed pH, ionic
strength or high concentration of a compound with a specific attraction for the
sorbate. This desorption can be done in several steps to remove one species at
a time. The solute movement theory for on-off chromatography is developed in
the next section. Ion exchange is discussed in Chapter 9.
There are also non-column chromatographic methods which are used for
analysis. In paper chromatography the column is replaced by a sheet of paper.
Solvent rises in the paper due to capillary forces. A spot of feed is placed on
the paper. Separation occurs because of different migration rates of the solutes.
One advantage of paper chromatography is a large number of feed spots can be
analyzed simultaneously. This method was developed by Consden et al (1944)
and modem applications are discussed by Smith (1969). Thin-layer chroma-
tography was popularized by Stahl (1965). A glass or plastic plate is coated
with a thin-layer of adsorbent such as silica gel or alumina. Operation is very
similiar to paper chromatography. In counter-current distribution (CCD)
(Craig and Craig, 1956) a series of extraction tubes is arranged so that only the
top or only the bottom phase is transferred. A pulse of feed is input into one
test tube and is eluted with additional solvent. Counter-current distribution is
thus a liquid-liquid extraction that is operated in the same way as chromatogra-
phy, and is very similiar to the staged model discussed in Section 7.5. A
modem adaption of CCD called countercurrent chromatography was
developed by Ito and his coworkers (e.g. to et ai, 1974). The method winds a
296
helical coil in a centrifuge. Spinning the coil keeps the phases separate and the
droplets of immiscible liquid will migrate through the coil. The result is essen-
tially the same as CCD, but the apparatus is not as cumbersome. All of the
types of chromatography discussed in this paragraph are commercially avail-
able as analytical instruments.
7.2. APPLICA TION OF SOLUTE MOVEMENT THEORY TO

CHROMATOGRAPHY
First, what does the solute movement theory tell us about migration chromatog-
raphy? The solute velocity for linear systems was given by Eq. (6-24). The
solute movement was shown graphically in Figure 6-4b. For a very small feed
pulse the peak maximum exits at a retention time tR,i given as,
(7-1)
For pulses which are fairly wide the retention time must be corrected. The
center of the feed pulse will form the peak maximum. Thus
(7-2)
tR = Llus + tp/2
where tp is the period of the feed pulse. For differential pulses Eq. (7-2)
reduces to Eq. (7-1). In chromatography single porosity models are often used.
An alternate expression for Us is given later in Eq. (7-32b).
The solute movement theory (or a retention time argument) can be used
to determine the linear equilibrium parameter from an experiment. Suppose a
Figure 7-3. Separation of two components in linear chromatography.

Widths W1 =40'1 and W2 =40'2 where 0'1 and 0'2 are standard
deviations of the peaks.
297
chromatograph is run in the laboratory with a very small pulse of feed and the
result shown in Figure 7-3 is obtained. The non-retained peak which exits at
time tNR shows when a small, non-retained molecule will exit. Thus tNR is a
measure of the void volume between particles and inside the particles. An
example is the air peak in gas-liquid chromatography. The retention time for
solute 1, tR,l, and solute 2, tR,2, can be measured. Assume that 1 is a chemical
which has a known eqUilibrium constant A l . Then the equilibrium constant for
solute 2 can be determined as,
tR,2 - tNR A2 (7-3)
--.:.--- = - = a.2l
tR,l - tNR Al
Equation (7-3) is easily derived from the solute movement theory (see Problem
7-Cl).
The selectivity, ~l' is the ratio of the equilibrium constants and is analo-
gous to the relative volatility used for vapor-liquid equilibrium. Note that the
cF
0
I
I
b I
I
L
I I
C
_
C ~:--======:
CF~~
lL----=====
__ ______
:
. t
Figure 7-4. Solute movement theory for one independent, non-linear solute
with a wide feed pulse. Shock and diffuse waves are indepen-
dent a. Feed pulse. b. Solute movement diagram. c. Outlet
concentration.
298
selectivity is easily determined from the experiment shown in Figure 7-3 even
if neither Al nor A z are known.
For independent non-linear isotherms the solute movement theory can be
used to predict the outlet concentration. Two situations can occur. When the
input pulse is broad, the peak: will exit at the feed concentration. This is illus-
trated in Figure 74. Note that the shock and diffuse waves do not interact; in
effect, they are isolated by the plateau at c = CF' Thus, the calculation of these
waves and the outlet concentration profile is a straightforward application of
the methods in Section 6.4.
For narrower pulses the shock and diffuse waves intersect before they
q
C2-C 1
C
b
CF
d
• I
Figure 7-5. Intersection of shock and diffuse waves for a narrow feed puse.
Illustrated for Example 7-1. a Chords on isotherm. b. Feed
pulse. c. Solute movement diagram. d Outlet concentration.
299
exist the column. Now C2 and qz in Eq. (6-30) are decreased and the slope of
the chord (qz - ql)/(C2 - cd increases (see Figure 7-5a). The result is llsh
decreases and the shock wave is curved as shown in Figure 7-5c. The shock
wave velocity needs to be calculated step-by-step or a mass balance over the
cycle (IN == OUT) can be used to determine when the shock wave exists. This
is illustrated in Example 7-1.
Figures 7-4 and 7-5 both assume that other solutes in the feed are
independent. If the solutes interact such as with a multicomponent Langmuir
isotherm, Eq. (6-9), then the relatively simple picutre shown in Figures 7-4 and
7-5 changes dramaticaly. This more complicated non-linear situation is dis-
cussed in Section 8.5.1.
For on-off or feed-elute chromatography systems the equilibrium isoth-
erm can be approximated as the step function shown in Figure 7 -6a (Wankat,
a
on qoo
off
c
F II: 2 13 1 4
1
'O~ _----'-I,~lK__":____'~.
C
1 , - - - 1
2 13 1 4 1 t
Figure 7-6. On-off chromatography. a. Approximate equilibrium isoth-

enn. b. Solute movement diagram. 1. Wash step. 2. C at
feed concentration. 3. Concentrated C exiting after wave of
elutant E. 4. Diffuse wave of C. c. Outlet concentration.
300
1986). The normal operating cycle consists of a feed step, a short wash step to
remove non-adsorbed material, elution, and another wash step. During the feed
the solute will move in a shock wave with the velocity given by Eq. (6-30).
This is illustrated in Figure 7-6b (Wankat, 1986). In the wash step the
adsorbed solute (C) quickly moves through the column until open sites are
found. After sorption, no further movement occurs. Weakly or non-adsorbed
solutes, A, are washed out of the column. Counter-flow elution is illustrated
although co-flow elution is also commonly used. The elutant will move in a
wave (E) in a fashion similar to a thermal wave. The solute first jumps to a
high value when it is desorbed and then moves as a diffuse wave (for example,
Eq. (6-26». The outlet profile is shown in Figure 7-6c. If there are several
adsorbed solutes, the elution can be done in steps with changes in elution con-
centration to remove different solutes. The final wash step is used to remove
the elutant so that the column is ready for the next feed pulse. These systems
can be analyzed in more detail using the non-linear approaches of Chapter 8.
Example 7-1. Solute Movement Theory for a Narrow Pulse
A 50 cm column is packed with activated alumina. The column is ini-

tially filled with pure cyclohexane solvent. At t = 0 a pulse of 0.009
gmole/L anthracene is input for 5 minutes. Superficial velocity is 15
cm/min.
Data Ee = 0.42, Ep = 0 (a single porosity model was used).

num: q =
E quil1'b' 223 c
5 were h ies/kg and
q = gmo c =igmo
lL es
1+ 7 c
The bulk density (measured in air) is PB = 0.85 kg/L.
KD = 1.0, Pr (cyc1ohexane) = 0.78 kg/L.
Solution
We need to determine the curved shock wave illustrated in Figure 7-5c.

This will be illustrated first with a step-by-step calculation of ush and
then by use of an overall mass balance.
301
A. Preliminary Calculations.
The particle density Pp can be determined from Eq. (6-2a)
= PB - Ee Pair = 0.85 - (0.42) (- 0.0) = 1.465

Pp 1- Ee 0.58
where the weight of the air is neglected.

From Eq. (6-2b) we see Ps = Pp when Ep = O.
v
Velocity v = super = 35.71 cm/min
Ee
Before the intersection of the shock and diffuse waves we can deter-
mine Ush and Us. For a single porosity model,
V
Ush =---- ---
1- Ee ~q
1+--p -
Ee p ~c
The shock wave goes from Cz = 0.009 to Cl = 0.0.
Then ql = 0 and qz = 22 (0.009) = 0.04526

1 + (375 (0.009)
35.71 35.71 = 3.196 c~
Ush = 1 + 0.58 (1.465) ~q = 1 2023 [0.04526] mm
0.42 ~c +. 0.009
Before the intersection with the shock wave the diffuse wave velocity is
v
= - - -v- - - = - - - - - - - - -
Us
1 1 - Ee aq l-E
l+ _ _e P
a
+ -E-
e- Pp -a-c Ee p (1 + bc)z
where a = 22 and b = 375. Thus,
35.71
Us =------
1+ 44.5
(1 + 375 cf
302
If there were no shock wave, the exit times are tR (c) = ~.

Us (c)
The following table is easily generated.
Diffuse Wave if No Shock Wave Exits
c Us cm/min tR = 50/Us ,min tF + tR

0.009 10.7415 4.655 9.655
0.008 9.446 5.294 10.294
0.007 8.141 6.142 11.142
0.006 6.851 7.298 12.298
0.005 5.594 8.938 13.938
0.004 4.395 11.377 16.377
0.003 3.290 15.196 20.196
0.002 2.300 21.744 26.744
0.001 1.456 34.350 39.350
0 0.785 63.698 68.698
B. Differential Analysis: Step-by-Step Changes in Ush'
The shock wave and the solute wave at c = 0.009 intersect at z = ZI
when
Ush (0.009 to 0.0) tl = Us (0.009) (t - tF)
tF Us (0.009) (5) (10.7415)

Then tl = =
Us (0.009) - Ush (0.009 to 0.0) 10.7415 - 3.196
tl = 7.118 min and Zl = Ush t= 22.75 cm

This is shown in Figure 7-5c. The shock wave will now curve. From
Cz = 0.009 to Cz = 0.008 the average Cz seen by the shock wave is Cz =
0.0085. Then
_ 22 (0.0085)
qz = 1 + (375) (0.00851) = 0.04466
303
and
35.71
Ugh (0.00S5 to 0) = - - - - [___- -__ ] = 3.071 cm/min
0.04466
1 + 2.023 0.00S5
Note that the shock wave slows down. This shock will intersect the dif-
fuse wave concentration c = O.OOS at t2, Z2 where
Z1 + Ush (t2 - td = US (O.OOS) (t2 - tF)
which gives
Z1 + llg (O.OOS) tF - Ush t1

t2 = --llg-(-",O----:.O....".O-,,-S)---Us-h--
t = 22.75 + (9.446)(5) - (3.071)(7.11S) = 7.54S min

2 (9.446 - 3.071)
Then, Z2 = US (O.OOS) (t2 - tF) = 24.07 cm
From this point we can do another step. From C2 = O.OOS to 0.007 the
average concentration C2 = 0.0075. Then
ijz = (22) (0.0075) = 0.0433

1 + 375 (0.0075)
and
35.71
Ush (0.0075 to 0) = [ ] = 2.S176
0.0433
1 + 2.034 0.0075
304
Shock slows down more. This shock intersects c = 0.007 at t3, Z3 cal-
culated from
Solving for t3, we obtain
Z2 + Us (0.007) tp - ush t2 .
t3 = = 8.173 mm
Us (0.007) - ush
and Z3 = Us (0.007) (t3 - tp) = 25.83 cm
This tedious procedure can be continued until the shock wave exits at Z
= L. Note that relatively small steps must be used to have sufficient
accuracy. Obviously, this approach is convenient to solve on the com-
puter.
C. Integral Analysis: Overall Mass Balance
Fortunately, a less tedious integral analysis using a mass balance over

an entire cycle can be used. From t = 0 to t = 00 we must have In = Out
since Accumulation = O.
In = Ac vsuper cp tp = (15) (0.009) 5 Ac = 0.675 Ac
Out = Ac vsuper (Area under diffuse wave from lsh to t = 00)
A trial-and-error integration of the area under the diffuse wave is

required, but this is much easier than the previous calculation. The dif-
fuse wave at c = 0 exits at tp + 63.698 = 68.698 minutes. c = 0 from
this time to t = 00. Thus, we integrate backwards from t = 68.698
minutes to lsh' This is trial-and-error to find a lsh for which Area =
0.675 Ac/15 Ac = 0.045. We set up the following table using the dif-
fuse wave table for values
305
Interval C ~t c~t
C = 0 to 0.001 0.0005 68.698 - 39.35 0.01467

=29.348
c = 0.001 to 0.002 0.0015 12.6 0.0189]

Check #1 Area = 0.03358
c = 0.002 to 0.003 0.0025 6.548 0.01637

Check # 2 Area = 0.04995
This is too far (lsh = 20.196 at c = 0.003). We could take smaller inter-
vals to get Area exactly equal 0.045. Instead, will obtain an approxi-
mate answer with linear interpolation.
~
1m 20.196 + [ ~-:;:::;~~::1 1(26.744 - 20.196) ~ 22.54 min
This procedure gives the outlet concentration profile shown in Figure
7-5d and it gives lsh. The method can be expanded to give the shape of
the shock wave shown in Figure 7-5c. (see Problem 7-D13). The dotted
line shown in Figure 7-5d is the shape of the diffuse wave if the diffuse
wave had not intersected the shock wave. For a Langmuir isotherm the
integration of the diffuse wave can also be done analytically.
Notes
The results shown in Figures 7-4c and 7-5d oversimplify what is actu-
ally observed. Instead of a shock wave a constant pattern wave results
(see Sections 8.1 and 8.2). The center of the constant pattern wave is at
the shock wave. The diffuse wave results are often quite accurate.
Thus the extensive tailing shown in Figures 7-4c and 7-5d due to the
non-linearity is often the controlling effect. Profiles with similar shapes
are commonly encountered in practice. The two procedures illustrated
here can be applied to other situations (e.g. Problem 7-E4).
306
7.3. SUPERPOSITION IN LINEAR SYSTEMS
Systems with linear equilibrium are favorites to study theoretically. There are
several reasons for this. At low enough concentrations adsorption and chroma-
tography systems approach the linear isotherm, Eq. (6-4a). Analytical chroma-
tography and removal of trace impurities by adsorption are almost always
operated in the concentration region where isotherms are linear. Finally, the
mathematical manipulations are greatly simplified for linear isotherms. Unfor-
tunately, most commercial applications of adsorption, large-scale chromatogra-
phy, and ion exchange are not operated in the linear concentration range. How-
ever, the linear theories are still useful as a limiting case valid at low concentra-
tions.
When the isotherm is linear, the solute mass balances, Eqs. (6-42) or (6-
48) and the mass transfer Eqs. (6-44), (6-47) or (6-50) will be linear. An
extremely useful property of linear equations is superposition. If we have a
solution for some simple operations, the results for more complex operations
can be obtained by adding the solutions. For example, suppose that Eqs. (6-
48), (6-50) and (6-4a) are simplified with some of the assumptions listed in
Table 6-7. These equations are then solved for the breakthrough curve when
an initially clean column has a step input of feed added. This resulting break-
through solution is of the form,
(7-4)
The function XA (z,t) can be any of the solutions for linear equilibrium which
we will discuss later.
Now we wish to use this solution to determine the elution from an uni-
formly loaded column where CA/CA,F = 1 at t = O. Subtracting the breakthrough
solution from the uniformly loaded column solution, we obtain the solution for
elution
(7-5)
XA,elution = 1 - X A (z, t)
Superposition can also be used to find the solution for a pulse input of
307
feed. If the column is initially clean we have first a breakthrough solution, and
then after time tF a step down. The result is
(7-6)
In chromatography the pulse is often very small. Eq. (7-6) can be written as
{X(z.!) - X(Z.!-IF)}
xA.Pulse = tF ~------~
tF
Taking the limit as tF ~ 0
aX(z,t) (7-7)
XA,differential pulse = tF at
is the solution for a differential pulse. Many other situations can be solved by
superposition once the basic breakthrough curve XA (z,t) is known. The use of
superposition will be clarified by studying Example 7-2.
For linear systems superposition can also be used when several solutes
are separated if the solutes do not interact with each other. Solve for each
solute independently and superimpose the results. This was done in Figure 6-4
for the solute movement theory solution for linear chromatography. The super-
position principle is valid for more complex theories which include mass
transfer and dispersion effects.
The idea of superposition may seem trivial to you. However, superposi-

tion is not valid for non-linear systems. For example, the solute movement
theory solution for breakthrough when the isotherm has a Langmuir shape is a
shock wave. Thus XA in Eq. (7-4) is a shock. The diffuse wave for elution of
the loaded column is not given by Eq. (7-5).
One other result of superposition for linear systems is that variances

(standard deviations squared) add. Thus the effect of any phenomena which
increases the zone spreading can be added into the solution by adding vari-
ances. Unfortunately, once zone spreading has been observed it is not possible
308
to determine which of several possible phenomena caused it. This property of

linear systems will be used in some of the linear theories.
Example 7-2. Superposition.
Determine the outlet concentration in terms of the breakthrough solu-

tion for the following input.
o 10 20 30 40
Solution
We have an initially uniformly loaded column followed by a step down

to 0.75 Cp followed by another step down to 0 cp. From superposition
solution is,
COUl = Cp - (1- 0.75)cp X(z,t-lO) - (0.75 - O)cp X(Z,t- 25)
The first term on the right hand side is the initially loaded column. The
second term is the step down to a value of 0.75 Cp and the third term is
the step down from 0.75 cp to zero. The solution X(z, t) could be the
error function solution given in Eq. (7-10), or it could be any other solu-
tion for a step input.
7.4. LINEAR DISPERSION MODEL
Lapidus and Amundson (1952) started with Eqs. (6-4a), (6-48) and (6-50) and
considered two cases. In the first case they assumed very rapid mass transfer
309
so that solid and fluid were in equilibrium. Then Eq. (6-50) is not needed.
Equation (6-4a) was substituted into the single porosity mass balance, Eq. (6-
48). After rearrangement, the resulting equation is,
where x = c/cF. For a step input the boundary conditions used were
x = 1 for z = 0, t> 0
x = 0 for t = 0, Z > 0
x = 0 for Z ~ 00, t> 0 (7-9)
For sufficiently long times the solution is (Lapidus and Amundson, 1952;
Lightfoot et al., 1962)
vt
z---- ---
l-e
1 + __e ppA(T)
ee
(7-10)
4(Dm + ED)t ]172
[ l-e
1 + __e ppA(1)
ee
The error function, erf, is the definite integral
2 a
J
erf(a) = - erf(-a) = 1i2 exp(-~2)d~
1t 0
(7-11a)
Since the error function is a definite integral it represents a number. Values for
the error function are available in many computers and have been tabulated in
many handbooks such as the CRC Handbook of Physics and Chemistry. A
310
Table 7-1. Error function values. For negative a, erf(a) is negative.

a erf(a) a erf(a) a erf(a)
0.0 0.0 0.48 0.50275 0.96 0.82542

0.04 0.04511 0.52 0.53790 1.00 0.84270
0.08 0.09008 0.56 0.57162 1.10 0.88021
0.12 0.13476 0.60 0.60386 1.20 0.91031
0.16 0.17901 0.64 0.63459 1.30 0.93401
0.20 0.22270 0.68 0.66378 1.40 0.95229
0.24 0.26570 0.72 0.69143 1.50 0.96611
0.28 0.30788 0.76 0.71754 1.60 0.97635
0.32 0.34913 0.80 0.7421 1.70 0.98379
0.36 0.38933 0.84 0.76514 1.80 0.98909
0.40 0.42839 0.88 0.78669 2.00 0.99532
0.44 0.46622 0.92 0.80677 3.24 0.99999
short list of erf is given in Table 7-1. The value of the error function can be
approximated to within 0.0005 by the following formula (Vermeulen et ai.,
1984),
(7-11b)
erf( la I) = [1- (1 + 0.27841a 1 + 0.23141a 12 + 0.0781IaI 4 )-4]
Equation (7-10) is one example of the breakthrough solution used in the super-
position results given in Eqs. (7-4) to (7-7). Thus Eq. (7-10) can be used to
generate the solution for a variety of situations other than breakthrough from an
initially clean column.
It is convenient to write Eq. (7-10) in terms of the Peclet number
(7-12a)
where the term (Dm + En) is treated as an effective axial diffusivity. To

rewrite Eq. (7-10) we define
(7-12b)
311
Cout
•
v
Figure 7-7. Breakthrough solution for Lapidus and Amundson model, Eq.
(7-13).
where V is the volume of solution fed, and
(7-12c)
V is the volume of solution required to saturate a column of length z. In these

variables Eq. (7-10) becomes
(7-13)
This equation also represents the breakthrough curve, X (z,t). Note that the
cross-sectional area term, Ac, divides out.
What does the breakthrough curve look like? The results of plotting Eq.
(7-13) are shown in Figure 7-7. When the Peelet number becomes infinite (the
sum of molecular diffusivity and axial dispersion go to zero), the solution is the
same as the solute movement solution for linear isotherms. At smaller Peelet
numbers (higher dispersion) there is zone spreading around the wave center
predicted by the solute movement theory.
The dispersion solution, Eqs. (7-10) or (7-13), usually gives a very good
fit to experimental data for systems with linear isotherms when an effective
dispersion coefficient is measured. This gives an effective Peelet number.
Once the effective Peelet number has been determined, Eq. (7-10) or (7-13)
plus superposition can be used to predict the results for a variety of operating
conditions.
312
Why is this solution able to fit experiments so well when no mass

transfer terms are included?
The answer to this question is based on the properties of linear systems.

Both dispersion and mass transfer cause zone spreading. The resulting zone
spreading for a linear system is identical regardless of which process caused it.
Thus a model with no mass transfer resistance can use an effective dispersion
coefficient to fit experimental data. The effective dispersion includes molecu-
lar diffusion, dispersion and mass transfer effects. Other models including only
mass transfer and no dispersion can also fit the experimental data. Note that the
ability of a model to fit experimental data does not prove that the model is
correct or has included all important terms.
Values of the dispersion coefficient without adsorption and no internal

mass transfer can be determined from the correlation developed by Chung and
Wen (1968). Their correlation was based on columns packed with non-porous
inert beads. Thus there is no mass transfer included.
(7-14a)
Ee N Pe = 0.2 + 0.011 Re°.48
where the Peclet number is defined with respect to the particle diameter,
(7-14b)
and the Reynolds number is
Ee Pr v <lp (7-14c)
Re=---"-
J.l
Be sure to note that the Peclet numbers in Eqs. (7-12a) and (7-14b) were
defined differently. Equation (7-14a) will predict a value for (Dm + ED) lower
than that observed when mass transfer is present. The difference is a measure
of the importance of mass transfer.
The solution for a differential pulse can easily be obtained by substituting

313
Eq. (7-13) into Eq. (7-7). The error function can be differentiated by using the
definition in Eq. (7-11a). The result for a differential pulse is,
_ VF P~1/2 (1 + 'IN)
XAdifferential pulse - -4 (-1t) --'----'---~ exp
[- Pe z (V - '1)2] (7-15)
(V V)lJ2 4V V
In chromatography sorption is often quite strong and V is approximately equal

to V in all parts of the elution zone. If we make the assumption that V is
approximately equal to V in the preexponential term, Eq. (7-15) first simplifies
to
XAdifferentialpulse
1 VF
="2 V 1t
[p~ ]1J2 exp [-pez(V
4VV
- '1)2] (7-16a)
A further simplification results if V V:: '12 in the exponential term. This result
is
XAdifferential pulse
1 VF
="2 V [1t ]
Pez 1J2
exp
[ -P~(V - V)
-
4 '12
2] (7-16b)
This is a Gaussian distribution and is obviously easier to use than either Eq.
(7-15) or the error function solutions. The peak maximum occurs when V = V
(the exponential term is zero). At the peak maximum the concentration is
1J2
1 V F [ P~ ] (7-16c)
(XAdifferential pulse)max :::::"2 V 1t
Application of these equations is discussed in Example 7-3.
7.5. LINEAR STAGED MODELS FOR CHROMATOGRAPHY

Although chromatographic columns are continuous contact systems and are not
staged, staged models have been used since the pioneering work of Martin and
314
Carrier
Figure 7-8. Schematic of staged model.
Synge (1941). (e.g. see King, 1980, Giddings, 1965; or Wankat, 1986). His-
torically, this occurred because countercurrent distribution (a staged system)
was a precursor of the chromatographic theories. The chromatographic column
is broken into a series of N stages each with a height H.
(7-17)
H=L/N
The staged model is shown schematically in Figure 7-8. The unsteady state
mass balance for stage j for any component is,
(7-18)
In this equation V is the cumulative volume of mobile phase flowing in the

column, VM is the volume of mobile phase per stage and Ms is the mass of sta-
tionary phase per stage. For a single porosity model we can easily relate VM
and Ms to our usual variables.
(7-19a)
(7-19b)
For linear equilibria, q = Ac, Eq. (7-18) simplifies to
dCj Cj-l - Cj (7-20)

dV
= VM+AiMs
for each component i.
Equation (7-20) is readily solved for different boundary conditions. For

315
chromatography a pulse of feed of Pi moles of each solute is introduced to

stage 0 at volume O.
(7-21a)
VMCo',1 + ~A'qo ,1. = P.l'
.l.Y.J.s V=O
and all other stages initially contain no solute.
(7-21b)
Cj,i = 0, j~ 1 and V = 0
The solution for each solute is then the Poisson distribution which is
(7-22a)
in volume units, or in time units
(7-22b)
Continuous distributions are easier to use than the discontinuous Poisson

distribution. If the number of stages is large, Eqs. (7-22) can be approximated
with Gaussian distributions. At the outlet j = N and the results are,
where Vpeak = (VM + Ai Ms)N, or
(7-23b)
where the retention time tR = L/lls i, = NH/Us,i. The predicted shape and proper-
ties of the Gaussian distribution are discussed in the next session.
316
The staged model is convenient, but is not a good model for predicting
what will happen when variables are changed. The staged model does not give
any indication of how to determine the height of an equilibrium stage, H. For a
predictive model a theory which includes mass transfer effects must be
included.
7.6. USE OF GAUSSIAN SOLUTION
For long columns and very small pulses the linear chromatography solution for
both dispersion and staged models can be written as
(7-24)
where x is the deviation from the peak maximum and (J is the standard devia-
tion in the same units. This is a Gaussian solution. Giddings (1965) developed
a definition of H as
H= d(~) (7-25a)
dz
where z is the distance migrated. For a uniform column of length L this

becomes
H= ~ (7-25b)
L
Then,
(7-26a)
N=L/H
or
(7-26b)
With this extension Hand N are defined without referring to the hypothetical
staged model. N is a measure of the efficiency of the column.
317
Any set of consistent units can be used for 0' and x. In time units
(7-27a)
which agrees with Eq. (7-23b). In length units
(7-27b)
and in volume units
(7-27c)
which corresponds to Eqs. (7-16b) and (7-23a).
The zone spreading measured by O't or 0'1 is proportional to the square

root of the column length. Equation (7-24) is a Gaussian shape as shown in
Figure 7-9. The peak maximum exits at retention time tRi given by the solute
movement theory in Eqs. (7-1) or (7-2).
The peak maximum for a differential pulse can be determined from the
dispersion model, Eq. (7-16c), and from staged models (see Problem 7-C3).
The results are
(7-28a)
A detailed comparison of these results shows that the two models give the same
results if,
(7-28b)
This comparison requires a conversion of the two different nomenclatures (see

Problem 7-C8). If N is known the effective dispersion can be calculated or
vice versa.
318
Recorder
Response ~--- Peak Maximum - - - - - - f / >
%100
60.7
50.0
13.4
4.4
Figure 7-9. Gaussian response for linear chromatography. Constants for

Eq. (7-29):
Width Constant
Wcr 4
Wh 5.54
W4C1 16
WScr 25
Wt 16
The value of N and hence H (from Eq. (7-17» can easily be determined
from experiments. The value of N is given from the staged model as,
2
peak maximum
[ ] (7-29)
N = (constant) width
where the appropriate widths and constants are shown in Figure 7-9. The peak
maximum should be measured from the center of the feed pulse. Note that any
convenient set of consistent units can be used for the peak maximum and
width. One convenient set of measurements is to measure distances on a strip
chart recorder. The most convenient calculation is the width at the half height.
In linear systems variances (standard deviations squared) add. In dis-

tance units the variance O"r, is proportional to HL. Thus the H values caused by
different zone spreading phenomena (mass transfer, dispersion, extra-column)
319
will add. Since the purely dispersive effects can be estimated from the Chung
and Wen correlation Eq. (7-14) and the mass transfer effects have been corre-
lated (in Eqs. (8-15) to (8-19», the effect of extra-column zone spreading can
be estimated. In a well designed system the extra column effects (mixing in
injectors, fittings, distributors, valves, detector, etc.) should be small.
The purpose of chromatography is to separate. In analytical linear

chromatography with differential feed pulses the separation of two peaks is
often reported as a resolution. The resolution is defined as
R = 2(tR.2 - tR.1) (7-30)

WI +w2
where the terms were shown in Figure 7-3. Note that tR and w are either in
time units or distance on the strip chart, and R is dimensionless. An R = 1.0
represents a separation where the two peak maxima are separated by 40". For
linear isotherms where the peaks are Gaussian this is about a 2% overlap. The
retention times can be determined from Eq. (7-1) while the widths can be found
as 40"t where O"t is given by Eq. (7-27a). The resulting "fundamental equation"
of linear chromatography is,
(7-31)
where CXzl is the selectivity defined in Eq. (7-3), and k' is the arithmetic average
relative retention. The relative retention is
(7-32a)
Note that a2I = k~~ = A2/AI . For a single porosity model (see Problem 6-
C2), us.i becomes
V
us.i = - - , - (7-32b)
1 + k'i
320
Equation (7-31) assumes that N is the same for both components. Substituting
Eqs. (7-32a) and (7-32b) into (7-31), we obtain (Giddings. 1965)
(7-33)
where ~us = Usl - Us2 is the difference in migration velocities and Us is the
geometric average of the solute velocities. Equation (7-33) is easy to general-
ize to other systems such as electrophoresis (see Chapter 11). This equation
shows that resolution depends on column efficiency (NVl/4) times selectivity
(~us/us).
The resolution increases as ~1' k' or N increases. Changing the selec-

tivity, Cl21, has the most effect. For Cl21 near 1.0, the number of stages (or
column length) must be very large to have a reasonable resolution. N starts to
increase very rapidly as ~1 becomes less than about 1.15. For example,
increasing ~1 from 1.1 to 1.2 doubles R at constant N or reduces N by ...J2 at
constant R. For large-scale applications Cl21 greater than 1.5 to 2.0 is desirable
since N and hence column length will not be excessive. Large Cl21 is desirable
since H can be large. The Van Deemter equation discussed later shows that
this means a high velocity and hence a high throughput can be used.
Increasing the average relative retention k' also increases resolution, but
the solute velocities are decreased. Increasing k' from - 0 to 2 dramatically
increases R or decreases N. An optimum k' seems to be between 4 and 6.
Once k' > 6 increases in k' have little effect on resolution, but the solvent
required increases. Most separations have a k' range from 1 to 10. If the ratio
of k' for the last and first peaks is greater than about 30, a satisfactory isocratic
elution (no change in solvent) separation is unlikely (Dolan, 1987). Either pro-
gramming or column switching (see Section 7.10) should be tried. For mix-
tures of an organic solvent and water, k' decreases by a factor from 2 to 3 for
each 10% decrease in the organic fraction (see Figure 7-12, Dolan, 1987). For
example, if a component shows a k' of 20 for a 80:20 methanoVwater solvent,
k' will probably be from 6 to 10 for a 70:30 methanoVwater solvent.
Increasing N also increases resolution. If H is decreased by decreasing

321
particle diameter <lp or increasing diffusivity, N will increase. Since decreasing

~ and increasing L both increase pressure drop, there are limits to how much
either of these can be changed. When large increases in resolution are needed,
the chemistry of the adsorbent should be changed to increase selectivity.
Example 7-3. Linear Chromatographic System
A pulse of anthracene was input into a chromatographic column using

isopropanol as the solvent. Column ID:= 0.75 cm and the packed length
was 80.0 cm. Ee was estimated as 0.37. N was 266.67. When a pulse
was run, the peak maximum exited at 93.79 min. Solvent flow rate was
Q:= 0.72 ml/min. We wish to build a 50.4 cm column of the same ID
and the same packing. Assume H is unchanged. For a 5 minute pulse
predict the outlet concentration profile for anthracene. Low concentra-
tions are used and isotherms may be assumed to be linear.
Solution
A. Define. We wish to calculate concentration versus time for this
pulse input.
B. Explore. Since N was estimated, the value of H can be calculated.

From this H, N and the Peclet number can be calculated for the shorter
column. One method of solution is to use the linear dispersion model
solution for breakthrough, Eq. (7-13), and from superposition find the
solution for a pulse, Eq. (7-6). An alternate solution is to use the Gaus-
sian chromatography solution, Eq. (7-24); however, this will be approx-
imate since the pulse may not be small enough to be considered dif-
ferential. Although the isotherm value is not given, the necessary
parameters can be estimated from the measured retention time.
C. Plan. We will do two solutions. First, Eqs. (7-13) and (7-6) will be
used. This involves determining the Peelet number for the 50.4 cm
column. After V is determined from Eq. (7-12c), values of the break-
through solution and the pulse solution can be determined.
322
In the second solution method we will assume that the pulse is

small enough that Eq. (7-24) is applicable. Mter determining the solute
velocity from the retention time, the retention time in the 50.4 cm
column can be estimated from Eq. (7-2). If x in Eq. (7-24) is the devia-
tion in time from this retention time, 0' must be O't given by Eq. (7-27b).
The profile can now be easily calculated.
D. Do It. The cross sectional area is
2 ' 2
A = 1tD = 1t(0.75) = 0442 cm2
c 4 4 .
Then the superficial velocity vsuper = ~ = ~:422 = 1.629 cm/min
and the interstitial velocity v = Vsuper/Ee = 1.629/0.37 =4.403
The H value is: H =LIN = 80/266.67 =0.3 cm

For the 50.4 cm column, N =L/H =50.4/0.3 =168
and from Eq. (7-28b) the Peclet number is, Pe = 2N = 336.
The solute velocity can be determined from Eq. (7-1)
Us = L/tR = 80/93.79 = 0.853 em/min
D2. Calculation for Dispersion Model:
To calculate V Eq. (7-12c) is used. The term in brackets is con-

veniently determined from us'
323
Thus
and
v= A"L[ee + (1-ee)PpA(T)] = (0.442)(50.4)(1.91) = 42.55 cm 3
The breakthrough solution is now easily determined in terms of

volumes from Eq. (7-13). The pulse solution from Eq. (7-6) is,
-c = X(pe z ' V) - X(pe z ' V - Vp)

Cp
where the volume of feed is: Vp = (0.72 rnl/min)(5 min) = 3.60 ml.
Substituting Eq. (7-13) into the pulse expression we have,
(7-34)
which is
~ =!
CA,P 2
{ rf [18.33(V-42.55)]_
e 2 (42.55V)1!2 e
rf [ 18.33 (V-46.15) ]}
2 (42.55V-153.18)1!2
Now we need to calculate. This is easiest to do on a calculator or com-

puter which has the error function built in, or by use of Eq. (7-11b). In
324
the absence of such a tool Table 7-1 or the more complete tables of the
normal curve of error (e.g. Handbook of Chemistry and Physics) can be
used. The calculation can be done in terms of either volume V or time t
since these are obviously related.
V = Qt = 0.720t or t = 1.389 V
The resulting concentrations are shown in the table below and in Figure
7-10.
Eq. (7-34) Eq. (7-16b) or Eq. (7-24)

V t clcF clcF
35.0 48.615 0.0056 0.0093
38.0 52.782 0.0684 0.0773
40.0 55.56 0.190 0.200
42.55 59.1 0.374 0.390
44.0 61.12 0.417 0.435
44.29 61.52 0.437 = Cmax
46.15 64.1 0.354 0.372
50.0 69.45 0.113 0.0964
54.0 75.0 0.00553
D3. Linear Chromatography Solution for Differential Pulse.
For the alternate solution the retention time when the peak max-
imum exits for a differential pulse would be,
t~,dif. pulse = u:-L = 0.853

50.4
=
5909 .
. mm
Since the pulse is 5 minutes long the peak will exit 2.5 minutes later
(see Eq. (7-2».
tR = 59.02 + 2.5 = 61.52 min

325
From Eq. (7-27b)
crt = (LH)1/2 = (50.4)(0.3)]1/2 = 4.56 min

Us 0.853
Now from Eq. (7-24)
where CMAX is given by Eq. (7-28a).
cp Vp 1/2
CMAX = . r --=- (pe) = 0.437 Cp
2 'J1t V
Thus Eq. (7-24) becomes,
~
Cp
= 0 437
• exp
[-(t-61.52)2]
41.59
The results at different times are listed in the table and are shown in
Figure 7-10.
E. Check. The results of the two solution methods are not in perfect
agreement, but they are reasonably close. Thus although the pulse is
not a differential pulse, this assumption is reasonable, and the two
models check each other. Equation (7-15) can also be used instead of
Eq. (7-24). This result is also reasonably close (see Problem 7-D12).
We can also use Eq. (7-29) to estimate N from Figure 7-10. The value
of N calculated using the width at half height agrees within 2.5%.
F. Generalization. For differential pulses the linear dispersion model

and the linear chromatography model are interchangeable. For finite
pulses Eq. (7-6) will be more accurate. However, Eq. (7-28) can often
be used if the retention time for the peak maximum is adjusted to follow
326
[!]
o
0.4
o
0.3
0
0.1 [!]
0
I:J) [!]
30 35 40 45 50 55
V,cm 3
Figure 7-10. Solutions to Example 7-3 for linear chromatography. 0 Disper-

sion model. m Differential pulse model.
the center of the feed pulse using Eq. (7-2). Without this adjustment the
linear chromatography model would predict a peak maximum at 59.09
minutes which is considerably in error. Obviously, Eq. (7-26) is
simpler to use. Note that Cmax < CF. Eq. (7-28a) will predict cmax > CF
for large pulses. In this case the peak will have saturated at Cmax = CF
and the Gaussian solution cannot be used. As a very rough rule of
thumb, if cmax > 0.5 CF the Gaussian solution should not be used.
7.7. VAN DEEMTER EQUATION
At the low concentrations usually used in analytical chromatography the isoth-

erms are linear. In large-scale applications of chromatography more concen-
trated solutions are usually separated and non-linear solutions are usually
required. However, the linear chromatographic solutions are very useful for
the insight they give on the importance of mass transfer and dispersion.
To predict zone spreading, mass transfer and dispersion effects must be

327
included. Lapidus and Amundson (1952) solved a second case where the rate
expression is given as,
(7-35)
When equilibrium is linear, this equation reduces to Eq. (6-50). The solutions
obtained are very difficult to use. A major simplification was made by Van
Deemter et al. (1956) who assumed linear equilibrium, a very small (differen-
tial) pulse of feed, and a long column. Their solution is,
Ci = Fi
exp
[ - [.!::.v (I + k /)_t]2] (7-36)
Ac VEe (1-k') "-i2n(cry +cr~) 2(1 + k')2 (cry +cr~)
where Fi is the moles of feed input, A; is the linear equilibrium constant, k' is
given by Eq. (7-32) and
(7-37a)
(7-37b)
Van Deemter, Zuiderweg and Klinkenberg (1956) then compared this

result to the continuous flow staged model written for a chromatographic sys-
tem, Eq. (7-23b). They found that the height of an equilibrium plate, H, is
given as
B (7-38)
H=A+ -+Cv
v
Equation (7-38) is known as the Van Deemter equation. This equation results
from the property of linear systems that variances add. H is proportional to the
variance (standard deviation squared) for the chromatographic system.
328
The A term is caused by eddy dispersion and is the flow contribution to

the plate height.
(7-39)
The B term is caused by molecular diffusion
(7-40)
where y is a labyrinth factor since diffusion paths are not straight. In gas sys-
tems the B term can be important, but in liquid systems it is often negligible.
The C term includes the mass transfer resistances, and can be expanded as
(Giddings, 1965; Snyder and Kirkland, 1979; and see Section 8.3).
(7-41a)
~ is due to extraparticle mass transfer (often called film diffusion).
(7-41b)
The constants a and b are approximately a = 1.4 and b = 2/3 (see Section 8.3).
Note that in film diffusion there is a velocity dependence. Csm is due to diffu-
sion in the stagnant mobile phase (often called pore diffusion).
(7-41c)
Finally, C s is due to diffusion in the stationary liquid phase coating the solid or
due to diffusion on the solid,
(7-41 d)
where de is either the average thickness of the coating or is <lp for adsorbents,
and D s is either the diffusivity in the stationary coating or the diffusivity on the
solid.
329
Figure 7-11. Plot of Van Deemter equation.
The Van Deemter equation is plotted in Figure 7-11 for a gas system
where the B term is significant. Note that there is a fluid velocity which minim-
izes H. Typical values for gas chromatography are A = 0 to 1 mm, B - 10
(mm)2/s, and C = 0.001 to 0.01 s which gives HETPmin = 0.5 to 2 mm and vrnin
= 1 to 10 mVs (Moody, 1982). Unfortunately, Vmin is much too low for com-
mercial applications. Usual operation is at significantly higher velocities where
the C term is dominant. Usually, pore diffusion is the controlling mass transfer
resistance. Thus combining Eqs. (7-38), (7-41a) and (7-41c), we obtain,
vd 2 (7-42)
H ex. --p + constant (high v)
Drop
The amount of zone spreading, which is measured by H, can be reduced by

decreasing particle diameter or increasing the molecular diffusivity. These
effects are discussed in detail later.
A variety of other equations for the plate height have been developed
(Giddings, 1965; Grushka et al, 1975; Horvath and Lin, 1978). The modified
Van Deemter equation
B 1 (7-43)
H= -+Csm v+ - - - - - -
v l/A + l/(Cro v)
is usually considered more accurate than the Van Deemter equation. Other
forms such as the Knox equation (Grushka et aI, 1975) are considered even
330
more accurate. A very simple fonn to correlate the effect of velocity is the
Snyder equation (Grushka et al., 1975).
(7-44)
H=ky'l
where k and n are constants for each packing material.
Detailed analysis of the zone spreading phenomena shows that the

theoretical minimum value of His H = 2 ~ (Grushka et al., 1975). H will be in
the range from 2.0 ~ to 2.5 ~ under ideal conditions and from 3.0 dp to 3.5 dp
for real samples (Dolan, 1987). These ranges can be used to rapidly estimate
H. For example, for 10 J.!m particles we would expect H to be in the range
from 20 to 35 J.!m.
In this development the column diameter never appears. Thus, for a

well designed and well packed column there are no theoretical reasons why H
should increase as the column diameter increases (in fact, H is often observed
to decrease as column diameter increases since there are less wall effects).
This serves as the basis for scale-up of chromatography to larger sizes. In
over-simplified tenns, keep length and particle diameter constant and increase
column diameter. Note that column diameter is proportional to the square root
of throughput (Wankat and Koo, 1988; McDonald and Bidlingmeyer, 1987).
The column must be designed so that there is no dead volume where mixing
can occur. The feed must be evenly distributed across the column. The
column must be packed so that all sections are unifonn and channeling is
minimized. Failure to follow these principles of good design will cause H to
increase. Since N is inversely proportional to H, Eq. (7-31) shows that resolu-
tion will decrease if H increases. The zone spreading outside the column
caused by fittings, the distributor, valves, the detector etc. is called extra-
column effects. Since d2 values add in linear systems,
(7-45)
where cr;ot is the actually observed zone spreading. Since crr,col = HL while
crl,ec is independent of length, crr,ec can be detennined by doing two experi-
ments with the same apparatus but with different column lengths. (see Problem
331
7-DI4). In a well-designed apparatus cr~ «cr~ol' Note also that the extra-
column effects cause zone spreading (cr~), but they do not cause separation
(N). If good design principles can be followed, theoretical analysis shows that
productivity (kg product/hr-kg packing) is significantly higher with short, fat
columns packed with small diameter particles and operated with rapid cycles
(Wankat and Koo, 1988).
7.8. ROSEN'S SOLUTION
Rosen (1952, 1954) developed an extraordinary solution for linear sorption sys-
tems. Assuming dispersion to be negligible and velocity to be constant, he
simplified Eq. (6-48) to
(7-46)
By using the change of variables
(7-47)
he obtained the partial differential equation
ac
-=-~
a- (7-48)
ax as
This partial differential equation is coupled to the diffusion equation by

the film diffusion equation (6-49). Substituting in the linear isotherm, Eq. (6-
4a) and Eqs. (7-47), Rosen obtained the modified film diffusion equation
(7-49)
332
where CIs = q(Rp) is the surface concentration of adsorbed material. Rosen

assumed sperical particles, ~ = 3/Rp • The diffusion equation (6-44a) can be
written as
~= Dmp ~
[r(aq)j (7-50)
ae rar ar
where the same substitutions used to obtain Eq. (7-49) were used.
The average amount adsorbed, q, can be related to the distribution within

the particles, q, by Eq. (6-43b). In Rosen's variables this is
_ 3 R,
q(x, e) = - 3
Rp 0
J
q(r,x,e) r dr
(7-51)
Finally, Rosen solved the problem for an initially clean bed with a step
input of feed. These initial and boundary conditions are,
(7-52a)
q(r,x,O) = 0, QO
c(O, e) = ° for e = t:5; °and c(O, e) = CF for e =t ~ °

(7-52b)
The solution for other boundary and initial conditions can be obtained by
superposition.
Rosen (1952) then solved Eqs. (7-48) to (7-52b). The mathematical

details are complex and will be skipped. The exact solution was written in
terms of an infinite series which, unfortunately, often has very slow conver-
gence. Rosen (1954) presents numerical solutions.
Rosen (1954) lumped variables in the form
3Dmp Az(I-Ee)p p Dmp A(I-Ee)PP 2Dmp (t-z/v) (7-53)

X= ,v= R k ,'t=--::--!--
R~ VEe p f R~
333
He showed that for large X (long columns) an asymptotic solution is
(7-54)
For X> 50, Eq. (7-54) is within 1% of the exact solution.
To explore the effect of controlling resistances, we can rearrange Eq.

(7-54) for long columns to,
(t-Z/v) 1
VE -
e Az(I - Ee)Pp
(7-55)
~=! l+erf
cp 2
If,
(7-56a)
film diffusion controls and the value of D mp is unimportant. This occurs when
kf is small or DmpA is large. Equation (7-55) is simplified for film diffusion
control by setting d~/(60 Pp (l-Ee) D mpA) = O. Note that strong adsorption
(large A) favors film diffusion control.
If,
(7-56b)
334
then pore diffusion controls. This is the more common case. Equation (7-55)
is easily simplified by setting <lp/(6kf ) = O.
Equations (7-54) and (7-55) are error function solutions. Thus they
predict S-shaped curves which will be similar to the dispersion solutions dis-
cussed in Section 7.4. However, there are differences between Eqs. (7-55) and
(7-10). The argument of the error function depends linearly on tin Eq. (7-55)
while it depend linearly on z in Eq. (7-10). Thus, by careful experimentation it
is possible to determine if axial dispersion or mass transfer is the dominant
mechanism. Equation (7-55) clearly shows that the total resistance,
[dp /(6kf ) + d~/(60 Pp (l-Ee) DmpA)), is important. For long columns any com-
bination of pore diffusion and film diffusion which gives the same total resis-
tance will give the same curve. Thus, from a single experiment one cannot del-
ineate the contribution due to pore and film diffusion. The film diffusion
coefficient can be estimated from Eqs. (8-15) or (8-16) and Dmp can be
estimated from Eq. (6-44b).
Rosen's solution can also be applied to pure heat transfer with no adsorp-
tion (see Problem 7-E3). A solution similar to Rosen's has been obtained for
zeolites with both macropore and micropore diffusion (Cen and Yang, 1986).
Carta (1988) solved essentially the same equations as Rosen but for the
periodic input of square waves in production chromatography.
7.9. APPLICATION OF THE LINEAR MODELS
The linear models have the unique advantage (or disadvantage depending on
how you look at it) that they all predict similar results if they use the same total
variance. Thus, once the results have been generated, it is difficult to determine
what caused the total variance without independent determination of axial
dispersion or mass transfer coefficients. This was illustrated in Example 7-2
where two models gave essentially the same results. Thus the dispersion
model, Eq. (7-13), can model systems where mass transfer is important if the
Peclet number is adjusted.
335
Likewise, mass transfer models which do not include dispersion such as

Rosen's model and other models (Hougen and Marshall, 1947; Lightfoot et al.,
1962; Sherwood et al., 1975; Thomas, 1944, 1948; Hines and Maddox, 1985;
and Vermeulen et ai., 1984) can model systems where dispersion is important
if the mass transfer coefficient is adjusted. The Hougen and Marshall (1947)
model assumes that the solute held in the stagnant fluid can be ignored. The
result uses J functions which are usually presented in tabular form (Hines and
Maddox, 1985; Sherwood, et al., 1975; Vermeulen et al., 1984). The J func-
tions are related to the error function. The Thomas model (Thomas, 1944,
1948; Sherwood et ai., 1975; Lightfoot et al., 1962; Vermeulen et al., 1984)
based on a kinetic reaction equation can also fit the data well. In the linear
limit (the model was derived for Langmuir equilibrium) the Thomas solution
for breakthrough is a complementary error function solution which gives
results very similar to Eq. (7-13). The Thomas model for Langmuir equili-
brium is the basis for many mass transfer solutions for ion exchange. This
development is covered in detail elsewhere (e.g. Lightfoot et al., 1962; Sher-
wood et al., 1975; and Vermeulen et al., 1984).
Since the mass transfer solutions are more complicated to use than the
relatively simple dispersion solution, the dispersion solution can be used
instead. If a chromatographic experiment is run, N can easily be determined
from Eq. (7-29). Then the Peclet number equals 2N. The effect of changing
conditions such as particle diameter, velocity or temperature can be determined
by calculating H from Eq. (7-17) and then using the Van Deemter equation or
other correlations to predict new values for H. Then N and Pez can be recalcu-
lated and the dispersion solution can be used. Note that the dispersion solution
is now applicable to a variety of operating approaches such as breakthrough or
elution, not just the pulse solution. Another advantage of the dispersion model
is it can easily be written in a two-porosity form (see Problem 6-C5) and thus
can correctly model systems where steric exclusion is important.
For more concentrated solutions the isotherms usually become non-

linear. For non-linear systems superposition is not valid and variances do not
add. Thus, it is only for linear systems that one type of solution can be used to
model a system with very different phenomena.
336
7.10. PROGRAMMING AND COLUMN SWITCHING
Programming or gradients are methods used to more rapidly remove slowly

moving components. These terms are often used interchangeably and refer to
changing a thermodynamic variable which reduces the equilibrium constant.
For example, in analytical gas chromatography temperature programming is
used to remove strongly adsorbed species. This is done by heating the entire
column in an oven at a steady rate. Since slow moving species are in the
column a lot longer than faster species, the slower species are heated to higher
temperatures. According to the Arrehenius relationship, Eq. (6-4b), the linear
equilibrium constant ~ will decrease significantly and the species will move
faster. Naturally, the column must be cooled before the next feed pulse. Tem-
perature programming is considered in more detail in Problem 7-E2.
In liquid chromatography systems a gradient of some chemical concen-

tration is often used to decrease the equilibrium constant. (Schoenmakers,
1986; Snyder and Kirkland, 1979). This is also called solvent programming.
In ion exchange, pH and ionic strength changes are often used to remove
species from the resin (see Chapter 9). In adsorption and bonded-phase
chromatography changes in the solvent polarity are commonly used. For
example, if the solvent is water, methanol is often added to help elute strongly
adsorbed organic compounds. This is a common method in large-scale
chromatography of biologicals. The effect of methanol on the capacity factor
k' is illustrated in Figure 7-12 (Majors, 1988). Note the logarithmic depen-
dence of k' on methanol concentration. Gradients can be done in a step-wise or
continuous fashion. After all solutes have been removed, the gradient forming
chemical must be washed from the column before the next feed pulse.
The reason for programming is easily explained using the system illus-
trated in Figure 7-13a. The column length or the size of the feed pulse is set by
the most difficult separation which is the separation of the two key species A
and B. The time to input the next feed pulse and hence the throughput of feed
is set by species C and A. Thus, the easiest separation controls throughput!
The spreading of the solutes can be predicted by Eq. (7-34) or (7-26). Inside
the column 0'1> Eq. (7-27b), will be approximately the same if the H values are
similar (this is commonly the case). The spread of the peaks exiting the
337
10
::- 1.0
30 40 50 60 70 80 90
0/0 methanal
Figure 7-12. Capacity factor k' as a function of methanol concentration in

reversed-phase chromatography. Column: ODS reversed-
phase; solvent: methanol/water; solutes: DBBP = 4,4'-
dibromo-biphenyl, DBB = p- dibromobenzene, BPA =
bisphenol A. (Majors, 1988). Reprinted with permission from
LC-GC, 6,16 (1988). Copyright 1988, Aster Publishing Co.
column can be estimated from Eq. (7-27a) which can be written as

(7-57)
Species C will be spread significantly more than A and B as it exits the column
since its velocity is lower.
Solvent programming is illustrated in Figure 7-13b. Now a step addition

of solvent S is added after the key components have separated. The velocity of
338
Solvent
Figure 7-13. a. lsocratic elution chromatography illustrating the "general

elution problem." b. Solvent programming result.
this solvent is easily determined from Eq. (6-23). Since the solvent is chosen
so that C is less adsorbed at high S concentrations, species C moves
significantly faster. In addition to exiting sooner, the C peak will be spread
much less. This is referred to as compression. After removing S from the
column (which may be difficult if S is strongly adsorbed), the next feed pulse
can be input. Comparison of Figures 7-13a and b shows that there can be a
considerable increase in throughput. In addition, component C is obtained as a
much more concentrated species. It is possible to have cout > CF (see Problem
7-A9). Both step gradients (Figure 7-13b) and continuous gradients are used.
Step gradients are more convenieat in large scale systems while continuous
gradients are more common in analytical applications (McDonald and Bidling-
meyer, 1987).
One particular application of solvent programming which can produce

very large increases in solute concentration is focusing. To explain this pro-
cedure consider the case where a pH gradient is used and assume that the
equilibrium constant for solute A decreases as pH increases. The pH wave
moves at a velocity upH in the column. At (PH)1, uA(pHd = upH. Then
(7-58)
339
f r
pH
(~J ° °J
100
12 OU) 0",
00 o
gO/-
80
~o
10
~ 00 I
o J.-ooJ
E 60 8
u
0
- pH~-1 r------;
40 6
4 e. \
20 • e'P.
-e-e-- ~O
/1 0
I
2 0
0
0 6
4000 5000 6000 7000
time.s
Figure 7-14. Displacement chromatography of amino acids (GLU = L-
glutamic acid; VAL = L-valine; LEU = L-Leucine) using 0.1
M NaOH as displacer on Dowex 50W-X8 cation exchange
resin. Solid line is result of model. (Carta et al. 1988).
Reprinted with pennission from AIChE Symp. Ser .. 84 (264).
54 (1988). Copyright 1988. Amer. Inst. Chern. Engrs.
Solute at pH < pH} will be overtaken by the pH wave while solute at pH> pH}
will overtake the pH wave. The result is focusing at the pH wave. Focusing
was illustrated for thennal waves in Figure 6-11. The result with a pH or sol-
vent gradient will be very similar.
Displacement chromatography is a particular application of gradients

which also uses focusing. The column is first equilibrated with a solvent which
either does not adsorb or is very weakly adsorbed. The feed pulse is followed
by a displacer which is more strongly adsorbed than all of the components in
the feed. Equations equivalent to Eq. (7-58) will be satisfied for each com-
ponent. Each component is focused and sharply separated from the other com-
ponents. This is illustrated in Figure 7-14 (Carta et al. 1988). The displacer
comes out last. Before the next feed pulse can be input the displacer must be
removed from the column. This is often easy to do in ion exchange chromatog-
raphy (see Chapter 9) but will be difficult in adsorption or partition chromatog-
340
: : A Product : B Product
I I
I I
I I A
I I
I I
Time or Volume
Figure 7-15. Chromatogram illustrating mixed recycle and recycle of frac-
tions Rl, R2, R3. (Wankat, 1986).
raphy when an adsorbate or solute is used as the displacer (see Problem 7-

AlO). Since each component is focused, the concentrations can be quite high.
The solutes thus interact (e.g. as in the multicomponent Langmuir isotherm, Eq.
(6-13». This increases the focusing effect, but in general makes theoretical
calculations significantly more difficult. Displacement chromatography is a
special case which can be conveniently analyzed (see Section 8.5.1.).
Solvent programming and displacement chromatography require addi-

tional chemicals, makes solvent reuse more difficult, and may require excessive
washing of the column if the desorbent isotherm is nonlinear. Two alternatives
which have been used commercially are recycle and column switching.
In recycle the solutes are not completely separated in the column.

Instead, the overlapping peaks exiting the column are recycled to the feed end.
This is illustrated in Figure 7-15 (Wankat, 1986). The advantages of this are
the column can be a lot shorter and much more concentrated products are col-
lected. Usually, the entire recycle stream is stored and mixed together. This
loses the partial separation which has already been obtained. Better results are
obtained if the recycle is done in fractions (Rl,R2 ,R3 ) which retain the partial
separation.
Column switching (Guiochon and Colin, 1986; Wankat, 1986) is illus-

trated in Figure 7-16 for the three component system illustrated in Figure 7-
13a. The feed is input as a pulse into the first column. This column is sized to
separate component C. The partially separated A and B peaks are sent to
341
A and B
f---i><I-- C
-t><J- F
s
Figure 7-16. Simple column switching system.
column 2 where they are further separated while component C is withdrawn as

product from column 1. Obviously, this allows the next feed pulse to be input
sooner and results in significantly less zone broadening of the C peak. The
columns can be packed with different sorbents. Use of the linear dispersion or
Gaussian models is straightforward for column switching (Agosto et ai, 1989).
Solvent programming, recycle and column switching are easily extended
to separations with more than three components. These methods can be com-
bined (Agosto et ai, 1989). The first column may be called a guard column.
The guard column protects the main column from very strongly adsorbed
species. Regeneration of the guard column is often done counter-flow to the
feed. Many other applications of column switching have been developed
(Wankat, 1986).
7.11. LARGE-SCALE CHROMATOGRAPHY
Large-scale applications of chromatography are similar yet quite different than

analytical applications. You may have used a gas-liquid chromatograph in
342
chemistry laboratory. A pulse of feed was injected with a syringe in a flowing

gas stream, and the outlet concentrations of the separated products were shown
on a recorder. Mter the last product exits, a new pulse of feed was injected.
The same ideas are used in large-scale systems, but the objectives have
changed. In large-scale chromatography we want to separate and collect the
products. Throughput should be as high as possible and the products should be
as concentrated as possible. Chromatography is only one of many ways to
separate materials and currently it is fairly expensive. McDonald and Bidling-
meyer (1987) suggest using chromatography only as a last resort. Chromatog-
raphy has become increasingly popular for production of chemicals which can-
not be distilled. In biotechnology and pharmaceutical production the com-
pounds cannot be vaporized and liquid chromatography is often the preferred
separation method.
Several different types of chromatographic separations have been used
for large-scale separations (McDonald and Bidlingmeyer, 1987; Wankat,
1986). Operation in the liquid phase appears to be preferred since many
biochemicals cannot be vaporized. In liquid adsorption chromatography the
basis for separation is adsorption. The adsorbent can be a polar compound
such as silica gel operating in a non-polar solvent such as hexane. In reverse-
phase operation water or an aqueous phase is the solvent while a non-polar
adsorbent is used. The most popular reverse phase pacldngs are C 8 or CI8
hydrocarbons bonded to silica gel. The majority of large scale applications
using migration chromatography are currently reverse-phase. Examples
include purification of steroids, hormones, dyes, antibiotics and various other
pharmaceuticals (McDonald and Bidlingmeyer, 1987; Wankat, 1986).
In actual practice the peaks spread and solutes overlap considerably. To
increase the product purity some of the outlet material may be recycled. These
points are illustrated in Figure 7-17 which shows the chromatographic
purification of sugars from molasses (Heikkila, 1983). Note that the sucrose is
not completely purified and that there are two separate recycle streams. The
regeneration of the sorbent is done by feeding fresh solvent (water in Figure 7-
17). This removes the solutes from the column, but also dilutes the product.
The chromatographic system can fractionate several solutes whereas most
adsorption systems produce only one product.
343
200
(~,
~
/
f
Cone. \
100 Y ~
giL
/'
0
30 60
NS
II
R
S RS
R
Time
Figure 7-17. Separation results for chromatographic separation of cane

molasses. (Heikkila, 1983). T = total, S = sucrose, NS = non-
sugars, RS = reducing sugars, R = recycle. Reprinted with spe-
cial pennission from Chern. Eng., 50 (Jan. 24, 1983). Copy-
right 1983, McGraw-Hill.
One of the most popular methods for large-scale liquid chromatography

is to use a reverse phase system with a C-8 or C-18 coating bonded to a porous
silica gel in an HPLC system. Packing as small as 10 J.Ul1 diameter has been
used, but larger packings are often used to reduce the pressure drop. An exam-
ple is shown in Figure 7-18 (Shih, et aI., 1983). The elution was done with
water until the desired MK peak had exited. After this, the column was regen-
erated with methanol. Since the MK was quite valuable, no recycle stream was
used. Shih et al. (1983) noted that the packings vary from batch-to-batch since
the process to make these packings is complex. They also noted that some sil-
ica leached into the water and had to be removed downstream. Other large-
scale applications are reviewed by Wankat (1986). These systems can be pur-
chased as turn-key units from several manufacturers.
Large scale size exclusion chromatography is commonly used industri-

ally for desalting proteins. This is the removal of small salt and buffer
molecules from the proteins. Hydrophilic gel packings which do not adsorb or
344
100
90
80
70 MK
Peak
60
DIversion
50 Through A
3-Woyvalve
I----t
40
30
20 c:
0
10 ~2
0 T
-Time
Figure 7-18. Large-scale chromatographic separation of semisynthetic anti-

biotic (MK) in 15-cm I.D. column. (From Shih et al. (1983).
Reprinted with permission from Chern. Eng. Prog .• 79(10). 53.
1983. Copyright 1983. American Institute of Chemical
Engineers.
de~ature the protein are often used. Since these gels tend to be compressible,
special design procedures are often required (Janson and Hedman, 1982).
Rigid particles are used where they are compatible with the chemicals being
separated since much higher flow rates and pressure drops can be used. (Hag-
nauer, 1987).
The usual pressure drop equations are given in Eqs. (8-22) to (8-24).
Section 8.5 argues that adsorption systems will have higher productivity if the
columns are short and fat, are packed with small diameter particles, and cycling
is rapid. These designs can have the same pressure drop, same throughput of
feed, and the same separation as more typical designs. Similar arguments can
be made for chromatography (Wankat and Koo, 1988). When expensive pack-
ings such as bonded phases are used, the reduction in amount of packing can
345
10 mM PO. pH 7.5 10 mM PO. pH 7.5

~ 1 M KCI
2 10
,~
I Q
1, :
3-HBDH
8
';j
'I
E
'"
~ 6
"",,
? ~
I
I
'c ~
I
:> 1
:e.
I
'1 ~ q
:I:
a
al 0
....
I '"
I
~
,
CI)
:I: I I
M < I
o
I
9
'0
~ A 280
2 '-:I 'J-O.c' t:)
~-- __L-~~~~~~"~~.• ~.~-M~~__~____

8 16 20 28 36 44 52
Fraction number
Figure 7-19. Pmification of 3-hydroxybutyrate dehydrogenase by affinity

chromatography (Darbyshire, 1981). Reprinted with
permission from Topics in Enzyme and Fermentation Biotech-
nology, Section 5, Wiseman, A., Ed., Ellis Horwood, Chiches-
ter, 1981. Copyright 1981, Ellis Horwood Ltd.
result in significant cost reductions. Guiochon and Colin (1986) also suggest
small diameter packings, but for different reasons.
The on-off, or feed-elute cycle is commonly used for large scale ion
exchange (see Chapter 9) and affinity chromatography. These processes can be
operated with essentially irreversible sorption during the feed step. Thus the
column is loaded with a very large feed pulse. Then nonadsorbed material is
washed from column, and the desired product is eluted. Additional separation
can be obtained by eluting with step changes in the elutant concentration. An
example of a large scale (1.8 liter column) for the affinity chromatography
purification of an enzyme is shown in Figure 7-19 (Darbyshire, 1981). Note
346
Feed Feed
~A~-r
'I I :
I : : Packing .L
'I I) I I
I fI I
1.... I I I
~---
1---.:::----1
I --3)
r-- ~I
'T t: -...~ -)
..... ~-- ---- --
Products
Figure 7-20. Schematic of annular flow column.
that most of the proteins (which absorb at 280 nm) pass through the column
without adsorbing. The 1 M KCI was used to desorb the desired enzyme. The
product pulse is quite sharp, is significantly purer than the feed, and is much
more concentrated than the feed. The use of a 1.8 liter column may not appear
to be large-scale, but for many affinity systems it is. The product enzymes may
be worth thousands of dollars per kilogram, but the market is small. Dar-
byshire (1981), Dechow (1989), and Hill and Hirtenstein (1983) contain more
information on large scale affinity chromatography.
On-off chromatography does not require long columns or small diameter

packings since separation is based on gradient elution not migration. A large
area for flow is desireable since larger volumes of fluid can be processed.
These considerations have lead researchers to geometries other than cylindrical
columns. Annular flow columns shown schematically in Figure 7-20 are avail-
able commercially. Flow can be radially inward as in Figure 7-20 or radially
outward. These columns have the desired short bed which reduces pressure
drop and large cross-sectional area to process large volumes of feed. Feed
capacity can be increased by stacking several units on top of each other. Sys-
tems similar to this are used commercially for ion exchange and affinity
chromatography (Chen and Hou, 1985). The extreme extension of this
geometry is to use a membrane hollow-fiber for affinity chromatography
347
(Brandt et ai, 1988). Note that other geometries could be used (see Problem 7-
B3).
The solute movement theory can be used for a qualitative understanding

of the chromatographic separation. This was illustrated in Figures 6-4 and 7-6.
Unfortunately, spreading of the zones is always important and must be included
in design. For linear systems the models discussed in this chapter are appropri-
ate. For non-linear systems the diffuse wave can be approximately predicted
from the solute movement theory. The mass transfer models for non-linear sys-
tems discussed in Chapter 8 are used for the shock wave.
A word about economics is in order. Large-scale chromatography is in

use for sucrose purification. Sucrose selling price fluctuates, but is usually less
than $O.60/kg. It is obvious that when conditions are favorable large scale
chromatography can be used for products with modest selling prices. These
favorable conditions are:
1. An inexpensive packing material with a long life, high capacity and high
selectivity (ex > 2) is available.
2. The solvent is either very inexpensive (e.g. water) or is easy to recover.
3. The feed concentration is relatively high.
4. An efficient large-scale chromatographic system is used.
5. Distillation is not feasible or is very expensive.
If some of the first four conditions are not met, large scale chromatography
may be feasible only for much more valuable products.
Finally, it should be noted that chromatography is rarely the last separa-
tion method used. The desired product exits as a dilute solution in a solvent.
Usually this solution needs to be concentrated by membrane separation (see
Chapter 12) or evaporation. If the product is desired as a solid, which is usu-
ally the case for pharmaceuticals, then an additional crystallization or precipita-
tion step (see Chapters 2 to 4) is needed. Thus chromatography solves one
separation problem but adds one or more additional problems.
348

In this chapter we have considered chromatographic separations and linear
theories to analyze these separations. At the end of this chapter you should be
able to:
1. Describe the following chromatographic terms:

a. Liquid adsorption chromatography
b. LLC
c. Bonded-phases
d. HPLC
e. Gas-solid chromatography
f. GLC
g. Size exclusion chromatography
h. Affinity chromatography
i. Programming
j. Column switching
k. Reverse-phase
2. Use the solute movement theory to explain both migration and on-off
chromatography.
3. Explain and use superposition in linear systems.
4. Use the linear dispersion model.
5. Explain the Van Deemter equation. Use this equation to qualitatively

predict the effects of changing different variables and to correlate data.
6. Determine the HElP from experiments. Predict the expected resolution.
7. Use the Gaussian model obtained from staged or continuous contact

theory to predict chromatographic separations.
8. Use Rosen's model when the simplified long column solution is appropri-
ate.
9. Explain and solve problems using gradients or column switching.

349
REFERENCES
Agosto, M., N.H.L. Wang, and P.C. Wankat, "Moving withdrawal liquid
chromatography of amino acids", Ind. Eng. Chem. Res. 28, 1358 (1989).
Andrews, P., "Estimation of the molecular weights of proteins by Sephadex gel

filtration," Biochem. J., 91,222 (1964).
Brandt, S., R.A Goffe, S.B. Kessler, J.L. O'Connor, and S.E. Zale,
"Membrane-based affinity technology for commercial scale purifications,"
Bio/Technology, 6, 779 (1988).
Carta, G., "Exact analytic solution of a mathematical model for chromato-

graphic operations," Chem. Engr., Sci., 43, 2877(1988).
Carta, G., M.S. Saunders, J.P. DeCarli, and J.H.B. Vierow, "Dynamics of
fixed-bed separations of amino acids by ion-exchange," AIChE Symp. Ser., 84
(264), 54 (1988).
Cen, P.L. and R.T. Yang, "Analytic solution for adsorber breakthrough curves
with bidisperse sorbents (zeolites)," AIChE J., 32,1635 (1986).
Chen, H.-L. and K.c. Hou, "Bioseparation by cartridge chromatography," in

G.T. Tsao (Ed.), Annual Reports on Fermentation Processes, Vol. 8, Chapt. 3,
Academic, Orlando, FL, 1985.
Chung, S.F. and c.y. Wen, "Longitudinal dispersion of liquid flowing through
fixed and fluidized beds", AIChE Journal, 14, 857 (1968).
Consden, R., AH. Gordon, and AJ.P. Martin, "Qualitative analysis of proteins:
a partition chromatographic method using paper," Biochemical J., 38, 224
(1944).
Craig, L.C. and D. Craig, in A Weissberger (Ed.), Technique of Organic

Chemistry, VoL 3, Interscience, New York, 1956, 149-332.
350
Darbyshire, J., "Large scale enzyme extraction and recovery", in A. Wiseman

(Ed.), Topics in Enzyme and Fermentation Biotechnology, Vol. 5, Ellis Hor-
wood Ltd., Chichester, England, 1981, Chapt. 3.
Dechow, F.J., Separation and Purification Techniques m Biotechnology,

Noyes, Park Ridge, N.J., 1989.
Determann, H., Gel Chromatography, 2nd ed., Springer-Verlag, Berlin, 1969.
Dolan, J.W., "Shortcuts for LC measurements," LC-GC, 5, 1030 (1987).
Gere, D.R., "Supercritical fluid chromatography," Science, 222, 253 (1983).
Giddings, J.C., Dynamics of Chromatography, Part I, Principles and Theory,

Marcel Dekker, New York, 1965.
Gribnau, T.C.J., J. Visser, and R.J.F. Nivard (Eds.), Affinity Chromatography

and Related Techniques, Elsevier, Amsterdam, 1982.
Grab, R.L. (Ed.), Modern Practice of Gas Chromatography, Wiley, New York,
1985.
Grushka, E., L.R. Snyder, and J.H. Knox, "Advances in band spreading
theories," J. Chroma tog. Sci., 13, 25 (1975).
Guiochon, G. and H. Colin, "Theoretical concepts and optimization in prepara-

tive scale liquid chromatography," Chromatography Forum, 21 (Sept. - Oct.
1986).
Hagnauer, G., "Preparative size exclusion chromatography," in B.A. Bidling-

meyer (Ed.), Preparative Liquid Chromatography, Elsevier, Amsterdam, 1987,
Chapt.8.
Heikkila, H., "Separating sugars and amino acids with chromatography,"

Chem. Eng., 90 (2) 50 (Jan. 24, 1983).
351
Hill, E.A and M.D. Hirtenstein, "Affinity chromatography: its application to

industrial scale processes," in Advances Biotechnological Processes, Vol. 1, A.
Mizraki, and A.L. Van Wezel, (Eds.), Alan R Liss, Inc., New York, 1983,31-
66.
Hines, AL. and RN. Maddox, Mass Transfer. Fundamentals and Applications,
Prentice Hall, Englewood Cliffs, NJ, 1985, Chapt. 14.
Horvath, C. and H.-J. Lin, "Band spreading in liquid chromatography, general

plate height equation and a method for the evaluation of the individual plate
height contributions," J. Chromatogr., 149,43 (1978).
Hougen, O.A and W.R. Marshall, "Adsorption from a fluid stream flowing
through a stationary granular bed," Chem. Eng. Prog. 43(4),197 (1947).
Ito, Y., RE. Hurst, RL. Bowman, and E.K. Achter, "Countercurrent chroma-
tography," Separ. Purific. Methods, 3, 133 (1974).
James, A.T. and AJ.P. Martin, "Gas liquid partition chromatography: the
separation and micro-estimation of volatile fatty acids. Formic acid to
dodecanoic acid," Biochem. J., 50,679, (1952).
Janson, J.C. and P. Hedman, "Large-scale chromatography of proteins," in A

Fiechter (Ed.), Advances in Biochemical Engineering, Vol. 125, Chromatogra-
phy, Springer-Verlag, Berlin, 1982,43.
Johnson, E.L. and R Stevenson, Basic Liquid Chromatography, Varian, Palo

Alto, CA, 1978.
King, CJ., Separation Processes, 2nd ed., McGraw-Hill, New York, 1980.
Lapidus, L. and N.R Amundson, "Mathematics of adsorption in beds. VI. The

effect of longitudinal diffusion in ion exchange and chromatographic columns,"
J. Phys. Chem., 56, 984 (1952).
352
Lightfoot, E.N., R.J. Sanchez-Palma, and D.O. Edwards, "Chromatography and

allied fixed bed separations processes," in H.M. Schoen (Ed.), New Chemical
Engineering Separation Techniques, Interscience, New York, 1962, p. 9-181.
McDonald, P.D. and B.A. Bidlingmeyer, "Stragegies for successful preparative

liquid chromatography," in B.A. Bidlingmeyer, (Ed.), Preparative Liquid
Chromatography, Elsevier, Amsterdam, 1987, Chapt.1.
McNair, H.M. and EJ. Bonelli, Basic Gas Chromatography, 5th ed., Varian,
Palo Alto, CA, 1969.
Majors, RE., "Reader's questions on HPLC columns," LC-GC, 6, 16 (1988).
Martin, A.J.P. and R.L.M. Synge, "A new form of chromatogram employing
two liquid phases, Biochem. J., 35,1358 (1941).
Moody, H.W., "The evaluation of the parameters in the Van Deemter equa-
tion," J. Chem. Educ., 59, (4), 290 (April 1982).
Perkins, RV., V.J. Nau, and A. McPartland, "Automatic pH gradient genera-

tion for protein and peptide HPLC separations" LC-GC, 5,419 (1987).
Rosen, J.B., "Kinetics of a fixed bed system for solid diffusion into spherical
particles," J. Chem. Phys., 20,387 (1952).
Rosen, J.B., "General numerical solution for solid diffusion in fixed beds," Ind.
Eng. Chem., 46, 1590 (1954).
Schoenmakers, PJ., Optimization of Chromatographic Selectivity, Elsevier,

Amsterdam, 1986.
Sherwood, T.K., RL. Pigford, and C.R Wilke, Mass Transfer, McGraw-Hill,
New York, 1975.
Shih, C.K., C.M. Snavely, T.E. Molnar, J.L. Meyer, W.B. Caldwell, and E.L.
353
Paul, "Large-scale liquid chromatography system," Chern. Eng. Prog., 79 (10),

53 (1983).
Smith, 1. (Ed.), Chromatographic and Electrophoretic Techniques, Vol. 1,

Chromatography, 3rd ed., Interscience, New York, 1969.
Snyder L.R. and J.J. Kirkland, Introduction to Modern Liquid Chromatogra-

phy, 2nd ed., Wiley, New York, 1979.
Stahl, E. (Ed.), Thin-Layer Chromatography, A Laboratory Handbook,

Academic Press, New York, 1965.
Supina, W.R., "Packed columns/column selection in gas chromatography," in

Grab, R.L., (Ed.), Modern Practice of Gas Chromatography, 2nd ed., Wiley,
New York, 1985, Chapter 3.
Thomas, H.e., Heterogeneous ion exchange in a flowing system," J. Am.

Chern. Soc., 1664 (1944).
Thomas, H.e., "Chromatography: A problem in kinetics," Annals New York

Academy of Science, 49, 161 (1948).
Van Deemter, J.J., F.J. Zuiderweg, and A. Klinkenberg, "Longitudinal diffu-

sion and resistance to mass transfer as causes of nonidea1ity in chromatogra-
phy," Chern. Eng. Sci., 5, 271 (1956).
Vermeulen, T., M.D. LeVan, N.K. Hiester, and G. Klein, "Adsorption and ion
exchange," in R.ll. Perry and D. Green (Eds.), Perry's Chemical Engineering
Handbook, 6th ed., McGraw-Hill, New York, 1984, Section 16.
Wankat, P.e., Large Scale Adsorption and Chromatography, CRC Press, Boca
Raton, FL, 1986.
Wankat, P.e. and Y.M. Koo, "Scaling rules for isocratic elution chromatogra-
phy," AIChE Journal, 34, 1006 (1988).
354
Yau, W.W., J.J. Kirkland, and D.D. Bly, Modern Size-Exclusion Chromatogra-
phy, Wiley, New York, 1979.
Zweig, G. and J. Sherma, (Eds.), Handbook of Chromatography, Vol. 2, CRC

Press, Cleveland, 1972.
HOMEWORK
AI. In your own words explain superposition. How can superposition be

used with several solutes?
A2. Explain the physical measuring of the boundary conditions in Eq. (7-9).
A3. Explain how the dispersion model with no mass transfer terms can fit
column breakthrough data where mass transfer resistances are impor-
tant.
A4. When might the B term in the Van Deemter equation be negligible?
What type of chromatography is most likely to have a negligible contri-
bution from molecular diffusion. What does Figure 7-11 look like when
the B tenn is negligible?
A5. In a poorly designed chromatograph the HETP can be significantly

greater than predicted by the Van Deemter equation. List several possi-
ble causes.
A6. Explain physically why large molecules exit before small molecules in
size exclusion chromatography. Convert Figure 7-2 to a plot of ~
versus log (MW).
A7. When similar chemicals are separated, the HETP will be approximately
the same. This implies that the zone widths inside the columns are
approximately the same. However, the widths observed on a strip chart
recorder are quite different. Explain why.
355
A8. Explain how to use the Lapidus and Amundson model when v = 0 and
only diffusion occurs.
A9. Explain how the outlet concentration of species C in solvent program-

ming (Figure 7-13b) can be greater than its feed concentration.
A10. Displacers used for displacement chromatography with adsorption or

partition systems are usually used in high concentrations. Thus, they
illustrate strong favorable, non-linear adsorption. Explain why (using
solute movement theory) it will be difficult to remove the displacer and
reequilibrate the column with solvent.
All. Develop your key relations chart for this chapter.
B 1. Develop a problem for linear systems not done in this chapter and show
how the solution can be obtained by superposition.
B2. Brainstorm methods for scaling up capillary gas chromatography to

separate tons of feed.
B3. Develop alternatives to the annular flow column shown in Figure 7-20
which will have short flow paths and high surface areas.
C. Derivations
Cl. Derive Eq. (7-3) using solute movement theory.
C2. Derive the linear isotherm, effective dispersion model solution Eq. (7-
16c) for a differential pulse starting with Eqs. (7-7) and (7-13).
C3. For the staged model for chromatography, show that Cmax is given by
Eq. (7-28a). Also, show that Eq. (7-22a) satisfies Eqs. (7-20) and (7-
21).
356
C5. The Lapidus and Amundson dispersion model with local equilibrium
was originally solved for a single porosity model. With the assumption
of local equilibrium c = c .. (or x = x*), the two porosity model can also
be used. By inspection and analogy determine the two porosity equa-
tions equivalent to Eq. (7-10) and to Eqs. (7-12c) and (7-13). The result
you should obtain equivalent to Eq. (7-10) is,
CA 1
x = -- =- 1 - erf
CAP 2
C6. The Lapidus and Amundson dispersion model with local equilibrium
can easily be applied to study the zone spreading of thermal waves.
Develop the appropriate equation for a single porosity model. Do this
for a step input from Tinitial to Tp. Use the dimensionless variable
'tT = (T - Tinitial)/(Tp - Tinitia1 ). List the necessary assumptions.
C7. Although not commonly used, the staged model can be applied to tem-
perature pulses. Develop this staged energy balance model for adia-
batic equilibrium stages including the wall heat capacity. Write in
terms of the dimensionless temperature 'tT = (T - Tinitial)/(Tp - Tinitial)'
Solve for a pulse of volume Vp of temperature Tp input to stage a at V
=0.
C8. Show that Eq. (7-28b) is valid.
C9. Plot N vs ~l for k' = 5, and R = 1.0. What does this mean in terms of
column length or allowable velocity?
357
CI0. Write a computer program and use Eq. (7 -11 b) to calculate erf(a) values
for the values of a given in Table 7-1. Compare these numbers with the
exact values given in Table 7.1.
CIL Assume that Eqs. (7-53) and (7-54) are correct (trust me). Derive Eq.
(7-55).
D 1. The solute velocity for linear systems is easily determined from pulse
chromatography experiments. The following data were obtained from a
bed 50.4 cm long.
Solute Retention time, minutes

Naphthalene 43.826
anthracene 59.086
pyrene 81.29
a. Find the solute velocities.
b. If the feed pulse is to be 20 minutes long, how long must the

column be to just separate the components? Use the solute move-
ment model.
c. For the column length determined in part b at what time can the
next pulse be input to just separate the components?
d. Use the solute movement theory to design a column switching sys-

tem and repeat parts b and c.
D2. Component A has AA = 3.22 where AA = QA/CA. In a chromatograph

experiment a small pulse is injected. The peak maximum for com-
ponent A exits at 12.5 minutes while a non-retained tracer (qtracer = 0)
exits at 2.8 minutes. Component B in the same experiment exits at 15.6
minutes. Assume isotherms are linear and molecules are small so that
Kd = 1.0. What is the value of AB ?
358
D3. We have done the analysis of an adsorption column and obtained a

solution for the breakthrough curve in the form
The analysis used linear isotherms. We now wish to utilize this analysis
to determine what happens in vacancy chromatography in a column of
length L. In vacancy chromatography the column is initially saturated
at Co and then a "negative pulse" of concentration zero is input followed
by feed of concentration co. Thus feed input is:
1.0
o~--------~~------------.
o t,
What is the solution for this input in terms of the breakthrough curve
solution, X(z,t)?
D4. We have the solution for breakthrough for a linear system.
CA = CA,F X(z,t)
In terms of X, what is the solution for the following input profile?
o 10 25
D5. The purification of hydrogen is a process which can be carried out over
Columbia grade L, 18-20 mesh coconut activated carbon. It is desired
to remove ethylene from a 1% ethylene feed at 150°F. Predict the time
359
of breakthrough from a 20 cm long tubular column and draw the break-

through curve which results from the single porosity linear dispersion
model developed by Lapidus and Amundson. The system properties are
Ee = 0040, Pp = 0.52 g/cm 3 , equilibrium q = 117.78 c, vsuper = 10 cm/s,
~ =0.10 cm, Pez = 336, and Ac = 1.0 cm 2 .
D6. For the system of Problem 7-D5, predict the outlet concentration profile
if a 10 second pulse of the 1% ethylene feed is input into a 100 cm long
column. Other conditions are the same. (Note that Pez does change.)
You may assume the outlet curve is Gaussian.
D7. In Example 7-3, the linear dispersion model and the Gaussian chroma-
tography solution are used to calculate the outlet concentration profile
of anthracene. It is now desired to see if this same system can separate
a solution of anthracene and a molecule of similar size and chemistry.
The second component has a linear adsorption coefficient which is 10%
greater than that of anthracene. Flow rates, column id, and packing are
the same as in Example 7.3.
a. Will a reasonable separation occur in an 80 cm long column?

(Calculate the resolution R.)
b. How long does the column need to be to obtain a proper separa-

tion? (That is R = 1.0).
c. If the answer to part a is no, what value of A(T) is required for

the second component for a reasonable separation (R = 1.0) to
occur in an 80 cm long column?
Assume: 1. (Dm + ED) are the same for both components.

2. Both components have linear isotherms which are unaf-
fected by each other.
D8. A liquid chromatograph using 20 J..lI11 silica gel is separating acetona-

phthalene (A) from dinitronaphthalene (D). k' values are k'A = 5.5, k'D
= 5.8 in a solvent which is 23% methylene chloride, 77% pentane.
With an interstitial velocity of 1.0 cm/sec, H is measured as 0.12 cm.
We desire a resolution of R = 1.0. What column length is required?
360
Plot the shape of the outlet acetonaphthalene concentration profile,

CA/CA,max vs. time, for an infinitesimal pulse in a column which gives R
= 1.0.
D9. Repeat Problem 6-D2, but for a pulse which is five minutes long.
DlO. Repeat Example 7-3 except for a feed pulse which is 12 minutes long.
a. Use the Lapidus and Amundson dispersion model.
b. Try to use the Gaussian solution. Why doesn't this work?
Dl1. A liquid chromatograph using 20~m silica gel is separating acetona-

phthalene (A) from dinitronaphthalene (D). k'A::: 5.5, k'o::: 5.8. At
interstitial velocity v = 1.0 cm/s, H =0.12 cm. At interstitial velocity v
= 2.0 cm/s, H ::: 0.21 cm. For this liquid system the B term in the Van
Deemter equation is negligible. Since pore diffusion controls, the C
term is
csm d~
C::: -----':....
Dsm
For an infinitesimal pulse we desire a resolution of R ::: 1.0 at v = 0.5

cm/s in a 100 cm column. What particle diameter should be used?
Note: Include both A and C terms. Assume k~ and k~ are constant.
D12. In Example 7-3 the Gaussian approximation Eq. (7-16c) or (7-24) was
used instead of the somewhat more accurate Eq. (7-15). Recalculate the
outlet concentration profile using Eq. (7-15). Compare your results with
Example 7-3.
Dl3. The integral approach illustrated in Example 7-1 gives a point on the
shock wave at z ::: L. Extend this method to obtain other points on the
shock wave. Find c and lsh for z ::: 30 cm. HINT: Set L = 30 cm and
adjust the tables in Example 7-1.
361
D14. In an attempt to determine extra column effects in a chromatograph we

do two experiments and determine N for a solute with a linear isotherm.
Experiment 1: Ll =50 cm, Nl =710

Experiment 2: Lz =25 em, N z = 330
Except for column lengths, the experiments are identical. Determine

Heol and crt,ec.
D15. We are separating methane from helium in an adsorption column

packed with 5A zeolite molecular sieve. The particles are 40 to 60 U.S.
mesh (250 to 420 J.Ull). The column was 100 cm long and interstitial
velocity is 5 cm/s. The feed is 1.1 vol% methane. Use Rosen's model
to predict the breakthrough curve if the column is initially clean.
Data: A = 23.90 where q and c are in gmoie/cm 3 ,

Dmp =0.11 cmz Is, kr =2763.0 cm/s,
ee = 0.525 (Cen and Yang, 1986).
Note: Wateh your units since q and e are in the same units.
D16. We are testing a column with dilute glucose solutions. The packing is
an ion exchange resin in calcium form. Predict the outlet concentration
profile from a 1 meter long column. Initially the column contains pure
water.
At t =0 start input of CF = 3 g glucose/l00 ml.

At t = 1 minute start input of CF = 5 g glucose/IOO ml.
Data: fp =0, ee =0.4, H =0.4 cm, Equilibrium: qG =0.51 cG

where qG and CG are in g/100 ml.
Superficial velocity = 10 cm/min.
362
E. More Complex Single Answer Problems
E1. For the fructose-glucose separation discussed in Problem 6-D4, H 10 =

cm. (Ihis is very large because of extra column mixing). Assume the
feed concentrations of fructose and glucose are equal. Use properties
listed in Problem 6-04.
a. If we desire a resolution of R = 1.0 for a differential pulse

input, determine the column length required.
b. A properly designed analytical system without extensive mix-

ing will have H - 0.06 cm. Predict required L for R = 1.0 for a
properly designed analytical system with a differential pulse
input.
c. A 100 cm column is used with a superficial velocity of 15

em/min. Based on the solute movement theory (Problem 6-04)
there should be perfect separation of the sugars if the feed
pulse is 1.48 minutes long and the waiting period between
pulses is 1.48 minutes. Suppose we include dispersion effects
(H = 0.06 cm) and use tF = 1.3 minutes and tcycle = 3.0 minutes
(a 1.7 minute waiting period). Predict the outlet concentration
profiles for three feed pulses. For the products from the second
pulse determine the glucose and fructose purities if the cut
points are chosen so that xFructose :::: XGlucose where x = C/CF' Use
the Lapidus and Amundson dispersion model. Why isn't the
Gaussian solution valid?
E2. (Derivation) In analytical gas chromatography it is common to use tem-

perature programming to more rapidly elute strongly adsorbed species.
In this procedure the feed pulse is input at temperature To. The column
is operated at this temperature while the fast moving peaks are eluted.
At t =10 the column oven is heated at a constant rate aT/at = b. This
continues until all components have eluted. The oven door is then
opened, the system is cooled and the next pulse is input. Equilibrium is
usually linear and follows an Arrehenius dependence.
363
a. Approximate the temperature profile as a series of step changes

ilT at time interval ilt so that ilT/ilt = b (the heating rate). With
this approximation aT/at = 0 in Eq. (6-58) except at the discrete
steps and the solute movement solutions developed in Chapter 6
are applicable. Determine the solute path and estimate the con-
centration.
b. Use the local equilibrium model to determine the solute path for
the temperature gradient aT/at =b (Note: in Eq. (6-58) aT/at::f:. 0.)
c. Estimate the concentration along the solute path for Part b.
E3. (Derivation) With suitable assumptions Rosen's model can be adapted

to heat transfer. Derive the appropriate equations which match Eqs.
(7-46), (749), (7-50), (7-51), and (7-52). Define 'tT as in Problem 7-
C6. Determine the lumped variables corresponding to Eq. (7-53). The
asymptotic solution remains Eq. (7-54), but with different definitions
for the variables. Do this for pure heat transfer with no adsorption. List
your other assumptions.
E4. A 1 meter column is packed with activated alumina. The column ini-
tially contains c = 0.01 gmole/L of anthracene in cyclohexane. At t = 0
a stream of pure cyclohexane at a superficial velocity (ee v) of 20
em/min is input. After 10 minutes a stream containing c = 0.011
gmole/L of anthracene in cyclohexane is added at the same velocity.
Predict the outlet concentration profile. Data are in Example 7-1.
Fl. The article by Van Deemter et al., (1956) is a classical paper in chroma-
tography. Read this rather long paper and write a two page critical
review.
F2. Another long classical paper is that by Martin and Synge (1941). They
used a staged model that was rather different than the one developed
364
here. If the number of stages is large, show that the two staged models
are equivalent.
F3. We are doing size exclusion chromatography of a protein from salts.

The protein is quite large and is totally excluded from the packing (Kd =
0). A one meter long column packed with particles with an average
diameter of 30 J.llll (30 x 10-6 meters) was used. Operation was at
25°C. Estimated value of Ee = 0040. Superficial velocity was 10
cm/min. Assume fluid has the properties of water. Predict the outlet
concentration profile ifat t = 0 a step input from 0 to CF gIL of protein is
introduced. Note: The Chung and Wen correlation given in Eq. (7-14)
is accurate for totally excluded species.
Chapter 8
NON-LINEAR THEORIES AND PACKED BED

ADSORPTION SYSTEMS
At higher concentrations adsorption isotherms for most chemicals become non-

linear. In this region the linear solutions of Chapter 7 are not valid. The non-
linear solutions given in Chapter 6 are useful, but do not include mass transfer
and dispersion effects. This chapter will present simple models for one com-
ponent isothermal, non-linear adsorption. The ramification of these results on
design will be explored. When more than one adsorbate is present, the system
is usually coupled since the adsorption of one component affects the adsorption
of all other components. Nonisothermal systems also show coupled behavior
since temperature affects the amount adsorbed. Since the coupled systems are
much more complex than uncoupled systems, the models for these systems will
not be developed. Instead, only the major results for coupled systems will be
discussed.
The basic packed bed system was shown in Figure 6-1. The operation is
cyclic with the feed (adsorption) step followed by a regeneration (desorption)
step, and a preparation step to prepare the column for the next feed step. Dur-
ing the feed step, adsorption continues until breakthrough occurs. Then the
packing is regenerated usually with counter-flow of a hot fluid or by purging at
low pressure. Usually the column is not completely regenerated but some
solute is left in the column. This is done since significantly less desorbent is
required. A variety of operating modes are used for adsorption and regenera-
tion and will be explored in this chapter using the theories of Chapters 6 and 7.
8.1 MASS TRANSFER ZONE APPROACH
In actual practice shock waves will spread due to the dispersion and mass
transfer terms in Eq. (6-42). However, this spread is opposed by the isotherm
365
366
o ~--------------~------ __ ~ ____ -1-L~ _______

L
Length in column
Figure 8-1. Concentration profiles inside column at different times for cQn-
stant pattern system.
effect which tends to fonn shock waves. The net result of these two QPposing
forces is to fonn a constant pattern wave which does not change shaw as it
moves in the column. This is shown in Figure 8-1 where the concentration of
the fluid inside the bed is shown at different times. Once fonned the wave does
not change shape as it moves through the column. This behavior will occur
whenever the solute movement theory predicts a shock wave.
A common industrial design procedure uses constant pattern ~ysis

based on experimental data. This mass transfer zone approach is based on the
original work of Michaels (1952) in ion exchange. What is actually measured
are the breakthrough curves shown in Figure 8-2 (Weyde and Wicke, 1948).
This is the outlet concentration which results from a step input in concentration
(Figure 8-2a) or a step decrease in concentration (Figure 8-2b). Note in Figure
8-2a that the breakthrough curves for the longer columns have exactly the same
shape. This is constant pattern behavior. Very short columns have a different
shape breakthrough curve since the constant pattern is not fully developed. For
the elution of a saturated column shown in Figure 8-2b, the waves become
more spread for longer columns. This is the proportional pattern behavior
predicted for nonlinear isothenns in Chapter 6. The mass transfer zone or MTZ
is the region where concentration is changing and thus mass transfer is occur-
ring. For constant pattern behavior the length of the M1Z is constant. The
length of the M1Z in time units is easily measured from the breakthrough curve
(Figure 8-3). Usually the M1Z is arbitrarily measured from a concentration of
0.05 CF to 0.95 CF , since it is hard to tell exactly where the S-shaped pattern
starts and ends. We wish to use this data to detennine the length of the M1Z
367
0.5
o ~~~~~L-~ ____~~________L-________L-______~
1.0
O.S
10 15 20 25
Time, min
Figure 8-2. Adsorption and elution of CO 2 on carbon. Reprinted from

Weyde and Wicke (1940). a. Adsorption breakthrough curves.
b. Desorption elution curves. vsuper = 4.26 cm/s.
1.0
0.95
0.5
oogL_______ .-.::=-_____ ~<!ffiIl~_ __.JL__ _ _ ____1~
Figure 8-3. Breakthrough curve for step input.

368
inside the column. The mass transfer zone must move at the shock wave velo-
city Ush. Then,
(8-1a)
Note that since the shape is not changing, all parts of the wave move at the
same velocity. The shock wave velocity can be estimated from Eq. (6-30) or it
can be. found experimentally. The stoichiometric center of the wave (c = 0.5cF
for a symmetric wave) must also be part of the local equilibrium solution (see
Figure 8-3). Then
L (8-1b)
Usb = --
lccntcr
The advantage of determining Usb from experiments, Eq. (8-1b), is the equili-
brium isotherm does not have to be known.
In typical industrial operations adsorption is stopped when breakthrough

just starts. This is illustrated by the right side of Figure 8-1. The breakthrough
time, Ibn can be related to column length, Ush and L MTZ . If breakthrough is
defined as c = 0.05 CF, the center of a symmetric wave has traveled a distance
L - (1/2) LMTZ (see Figure 8-1). Since the wave center is also the shock solu-
tion from the solute movement theory, we can calculate the breakthrough time.
(8-2)
This equation allows us to calculate the breakthrough time as the column length
is varied. Since the concentration in the mass transfer zone is varying, the sor-
bent is not saturated through the entire length of the bed. Only the shaded part
of the bed in Figure 8-3 is used. Thus the fractional bed utilization can be
determined as,
(8-3a)
369
100
80
% Bed 60
Used
40
20
Figure 8-4. Fraction of bed used as a function of L/LMTZ for constant pat-
tern step. Excerpted by special pennission from, Lukchis,
G.M., Chern. Eng. 80, (13), 111, (1973). Copyright 1973,
McGraw-Hill.
The ratio (Area not used)/(TotaI Area in M1Z) is the fraction of the mass
transfer zone which is not useful for adsorbing solute at the saturation concen-
tration. This ratio can be detennined from Figures 8-1 or 8-3. The areas are
shown in Figure 8-3. If the breakthrough curve is symmetric the ratio is 1{2.
Then Eq. (8-3a) simplifies to:
(8-3b)
Frac bed util. = 1 - .S(LMTZ / L)
In both Eqs. (8-3a) and (8-3b) LMTZ must be in length units. The fractional bed
use for symmetric beds is shown in Figure 8-4 (Lukchis, 1973).
The amount of solute which can be held by the bed is
(8-4)
Capacity = (Frac bed util.) x (Mass of adsorbent) X'lsat
where 'lsat is the saturation capacity of the adsorbent for the given feed concen-
tration. The saturation capacity is found from the appropriate equilibrium
isothenn calculated at the feed concentration, CF, or from experiment. The
saturation capacity depends on the adsorbent used, the feed concentration, and
370
the temperature. The fractional bed utilization can be increased by increasing

L or decreasing the mass transfer zone length. The mass transfer zone length
will be shorter for rapid mass transfer rates or small size adsorbent particles.
The mass transfer zone tends to be quite large for large molecules such as pro-
teins which have low rates of mass transfer.
A reasonable empirical design approach for favorable type adsorption

isothenns (Figures 6-3b or 6-3d) is as follows:
1. For the conditions of the feed step, detennine llsh , L MTZ , and Ckt. This
can be done with a single experiment in a laboratory column.
2. Design the adsorption step using the mass transfer zone approach.
Optimum design and theoretical detennination ofLMTZ is discussed later.
3. Detennine the equilibrium isothenn under the conditions of desorption.
4. Design the desorption step (proportional pattern wave) using the solute
movement theory.
5. Check the desorption step design with a laboratory experiment.
This combined approach uses the MTZ method where it is valid and it uses the
solute movement theory where it is approximately valid.
8.2. CONSTANT PATTERN SOLUTIONS
In this section we will see how the constant pattern shape can be predicted
theoretically from a mass transfer analysis. Constant pattern shapes occur
when the solute movement theory predicts a shock wave. The isothenn effect
which tends to fonn a shock wave is balanced by the dispersive effects of mass
transfer resistances and dispersion tenns. The net result is a constant pattern
wave which moves at the shock wave velocity, Ush. This pattern will always
fonn when the conditions are right (Cooney and Lightfoot, 1965).
The constant pattern shape greatly simplifies the mathematical analysis.

A listing of constant pattern adsorption models is given by Sircar and Kumar
(1983). The usual starting equation is the single porosity solute mass balance,
371
Eq. (6-48), ignoring the dispersion and diffusion term. The system is assumed
to be isothermal. Equation (6-48) simplifies to:
(8-5)
Since the constant pattern wave has a constant shape, concentration must
depend only on a single variable. This variable can be a measure of the time or
distance from the center of the pattern. The variable 't,
(8-6)
't =t - z/llp
is a measure of time from the center of the pattern which moves at velocity IIp.
Since the constant pattern wave depends only on 't, the partial differential equa-
tion, Eq. (8-5), can be reduced to an ordinary differential equation. After a few
manipulations, the result is,
(8-7a)
Now the solution is obvious,
v pp(1-Ee)q (8-7b)
(1 - -)c + = constant
IIp Ee
If the bed is initially clean then c = q = 0, and the constant must be zero. Since
any uniform initial state can be reached from a clean bed, the constant will
always be zero.
We can write Eq. (8-7b) for the conditions after the wave has passed, ca
and qa. Dividing Eq. (8-7b) by this form and rearranging, we obtain,
(8-8a)
372
For relatively concentrated gases V ':I' Va since adsorption or desorption changes

the gas velocity. For liquids, dilute gases, and exchange adsorption velocity is
essentially constant and Eq. (8-8a) simplifies to
-=-
c q (8-8b)
In the remainder of the development we will assume that velocity is constant.

Then IIp = Ush' Equation (8-8b) states how c and q must be related because of
the mass balance. Since equilibrium has not been assumed, c and q are not in
equilibrium. The conditions after the wave, c. and qa' will be in equilibrium.
So far, we haven't used the mass transfer expression Eq. (6-50). We will
substitute the mass balance result, Eq. (8-8b), and Eq. (8-6) into Eq. (6-50).
The result is
q. dc .. (8-9)
(1-£e)PP - - =- Oem 3p)(c - c)
c. dt
Since there is a constant pattern, c must depend on a single variable and an
ordinary differential equation results. To integrate this equation we need a
relationship for c... The usual method is to assume that this stagnant fluid is in
equilibrium with the solid. However, remember c ':I' c·. With c· in equilibrium
with q, the isotherm can be used to replace c· in Eq. (8-9). Then Eq. (8-8b) can
be used to replace q with c. This can be done for any isotherm which will give
a constant pattern wave.
For example, assume that equilibrium can be fit with a generalized Lang-
muir expression of the form given in Eq. (6-12a). Then a constant pattern wave
will result when a step input of feed is used. Solving for c" from the Langmuir
equation and substituting in Eq. (8-8b), we obtain
c.. = q = c(q.fe.) (8-10)

a - bq a - bc(q./c.)
Substitution ofEq. (8-10) into (8-9) gives an ordinary differential equation.
(I-Ee)( ~) de = Oem )[c _ e(q./ca ) ] (8-11a)

c. Pp dt 3p a - bc(q./c.)
373
The conditions after the wave are the feed values: Ca = CF and qa = QF.
Thus qalea = QF/cF = a/(1 + bCF)' Substitution of this ;esult into Eq. (8-11a) and
simplifying gives
[
1 + beF - be 1
dc = kmap(1 + b CF) (8-11b)
c(b CF - be) d't Pp (1 - Ee)a
The integration of Eq. (8-11 b) is straightfOIward. After some algebra the result
is,
(8-11c)
The column is assumed to be initially uniformly saturated at Cinit. At

time zero a feed of concentration CF is input To have a constant pattern wave
with a Langmuir isotherm we must have CF > cinit. The mass transfer zone can
be found by evaluating Eq. (8-11c) from
(8-11d)
to
(8-11e)
The result is
(8-12a)
_ 1 + be F In [0.05 [0.95 cinit + 0.05 CF

beF 0.95 0.05 Cinit + 0.95 CF
lJ}
374
This equation can be interpreted two ways. If we fix the length at the column
outlet, Z = L, then
(8-12b)
If we fix the time at breakthrough, t = lbr> then
(8-12c)
Note that both tMrZ and LMrZ are inversely proportional to Ocm~) since ~'t is
inversely proportional to k.n ~.
The shape of the constant pattern curve can be obtained by evaluating the
constant in Eq. (8-11c) or by evaluating (Eq. 8-11c) from a known point
(c,zhnown to an arbitrary (c,z). It is often convenient to do this from the center
point c'12 = Y2 (CF + Cinitial) at 't = 't'12 (Lapidus and Rosen, 1954). This integra-
tion assumes the profile is symmetric, but is usually accurate enough. Equation
Eq. (8-11c) gives,
(8-13)
The use of Eqs. (8-12a,b,c) are illustrated in Example 8-1. Typical constant
pattern profiles are shown in Figures 8-1, 8-2b and 8-3.
One reason for illustrating this derivation in detail is the same steps can
be followed for other eqUilibrium and mass transfer forms. Any equilibrium
form which gives a shock wave with the solute movement theory can be used
to replace Eq. (8-10). In addition, a variety of mass transfer expressions can be
375
used instead of Eq. (6-50). Thus the constant pattern solution for almost any
type of equilibrium and a variety of mass transfer expressions can easily be
developed. This analysis is also applicable to ion exchange (see Problem 9-
DS). Unfortunately, this solution is restricted to the constant pattern part of the
cycle.
When the equilibrium theory predicts a diffuse wave, a proportional pat-

tern wave results as shown in Figure 8-2b. Since the pattern is not constant,
Eq. (8-5) cannot be used to simplify Eq. (6-48) to Eq. (8-6). The Thomas solu-
tion (Thomas, 1944, 1948; Sherwood et ai., 1975; Vermeulen et ai., 1984) can
be used for the proportional pattern part of the cycle. However, it is complex
and restricted to a Langmuir isotherm and a particular type of kinetics. Since
the equilibrium solute movement predictions of the diffuse wave are often quite
accurate particularly for gas systems, we will use the simple solute movement
results for the proportional pattern part of the cycle.
A second limit on the constant pattern solution is the assumption that the
column is isothermal. Commercial gas adsorbers are usually adiabatic and
have non-negligible values of ~Hads which often leads to significant tempera-
ture increases in the bed during adsorption. During desorption the bed cools
down. Since a temperature increase will decrease adsorption equilibrium (see
Eq. (6-8», these effects should be included. Approximate (Leavitt, 1962) and
analytical (Sircar and Kumar, 1983) constant pattern methods have been
developed for adiabatic systems. Adiabatic systems are discussed in more detail
in Section 8.5.2.
In actual practice the MTZ-LUB approach is often used instead of

theoretical calculations. This is done since trace components or small tempera-
ture changes can have large effects on the breakthrough curve and the MTZ. In
addition, since the constant pattern solution ignores dispersion and is based on
a lumped parameter mass transfer expression it is not exact. The constant pat-
tern theory is very useful for predicting the effect of changing operating condi-
tions such as velocity, temperature, particle diameter, feed concentration, etc.
A semi-empirical procedure where a theoretical result such as Eq. (8-12a) is
used to guide experiments and to help correlate data is a better design approach
than an entirely empirical design approach.
376
8.3. MASS TRANSFER CORRELATIONS
In most cases the lumped parameter Eqs. (6-47) or (6-50) are adequate for
modeling packed beds. In order to use the theory a good estimate of the mass
transfer coefficient kut~ is required. kut is an "effective" mass transfer
coefficient For spherical particles
6 (8-14)
~=
C\>
where <lp is the particle diameter. For non spherical particles an equivalent
diameter is used.
The mass transfer coefficient is usually correlated in terms of dimension-

less groups. A variety of these correlations have been developed (Ruthven,
1984; Sherwood et al., 1975; Yang, 1987). Wakoa and Funazkri (1978) corre-
lated the film coefficient kf using
(8-15a)
where the dimensionless groups are
(8-15b)
Sherwood number
Sc = Jl Schmidt number (8-15c)

PrDm
Re = Pr vEe <lp Reynolds number

(8-l5d)
Jl
At low Reynolds numbers the dispersion terms must be corrected for. Wakao
and Funazkri (1978) also review a large amount of data to measure kr . An
alternate correlation for the film transfer coefficient is (Sherwood et ai., 1975)
(8-16)
377
The results obtained from Eqs. (8-15a) and (8-16) will differ somewhat from
each other. Since Eq. (8-15) was corrected for axial dispersion, this result
should be used if the model includes axial dispersion. If the model ignores
axial dispersion a better fit will probably be obtained with Eq. (8-16).
For diffusion inside the pores we really want to solve Eq. (6-44a) with
appropriate boundary conditions. This was done in Section 7.8 for Rosen's
solution. The appropriate way to sum resistances is shown in Eq. (7-55). Since
Clp = 6/dp, the term dp/6kf is I/(kf ~). Then the other term in the summation in
Eq. (7-53) is l/(k Clp)pore (see Eq. 8-19a». Thus, we would expect an
equivalent solution when
(8-17a)
where A is the linear equilibrium constant. For nonlinear systems
A= [~l avg
(8-17b)
The equivalence of the solutions to Eqs. (6-44a) and (6-50) disappears for very
non-linear systems, at very short times, and for rapid cycling separations. In
these situations Eq. (6-44a) should be solved.
For surface diffusion comparison of the solution of Eq. (6-45a) to the

solution of the lumped parameter Eq. (6-45b) show a good correspondence
when
(8-18a)
We often wish to employ a surface diffusion coefficient in equations such as

Eq. (6-50) and (6-45c). Then,
(8-18b)
378
This result is also used for nonlinear isothenns by using A defined in Eq. (8-
17b). Note the similiarity of the results for pore and surface diffusion.
The effective mass transfer coefficient ~8p can be related to the indivi-
dual coefficients by the usual sum of resistances model if either pore or surface
diffusion control transfer inside the particle. For pore plus film diffusion,
(8-19a)
For use in Eqs. (6-50) or (8-9) this is,
(8-19b)
For liquid systems the surface diffusion tenn is always neglected, and Eqs. (8-
19a) and (8-19b) are appropriate. When surface plus film diffusion occur
(8-19c)
which is
(8-19d)
If both surface and pore diffusion are important, the sum-of-resistances model
cannot be used, and the diffusion equations must be solved simultaneously.
Obviously, to use the correlations in this section we must know both

designer specified properties, v and <lp, and physical properties; Ee, pp,A, Pf,
D m , D mp' D s and~. The designer specified properties are under the control of
the engineer. The properties of the adsorbent; Ee, Pp and A; can be estimated
from manufacturer's data, Tables 6-1 to 6-6, and equilibrium data (see Section
6.3). The viscosity can often be estimated from data in Perry and Green (1984)
or by methods in Reid et al. (1977). Diffusivities can be significantly more
379
difficult to estimate. Methods for estimating the molecular diffusivity are dis-
cussed by Perry and Green (1984), Reid et al. (1977) and Sherwood et al.
(1975). Extensive tables of parameters for the parameters in theories, and
references for experimental values of D m are given by Marrero and Mason
(1972). The effective pore diffusivity Dmp can be estimated from Eq. (6-44
a,b), although experimental data is usually required to determine the tortuosity.
Surface diffusivities depend on surface coverage and are discussed by (Ruth-
ven, 1984) and (Yang, 1987). Estimation of these values is illustrated in Exam-
ple 8-1.
Example 8-1. Constant Pattern
Sircar and Kumar (1983) fit data for carbon dioxide, methane, nitrogen
and hydrogen adsorption onto BPL activated carbon to a Langmuir
isotherm of the form.
where y is the mole fraction adsorbate and q is the moles adsorbate/g

adsorbent and adsorbate in the pores is included in q. For methane the
parameter values are CImax = 3.65xlO-3 ; and at 1 atmosphere,
TOC KA
25 0.4179
45 0.2550
75 0.1343
95 0.0928
Data: Ee = 0.39, ET = 0.763, PB = 0.484 g/cm 3 , and 't = 8.
We wish to operate a one meter long adsorption column packed with 1

mm diameter BPL activated carbon particles. The column is initially
filled with helium which does not adsorb. Total pressure is 1 atmo-
sphere and temperature is 25°C. A step change in feed gas from an ini-
tially pure feed gas to 5 mole % methane is made at t = O. If the
superficial velocity is 55 cm/s, predict the width of the mass transfer
zone.
380
Solution
We must first convert the equilibrium data to the form q = ac/(1 + be).
For the gas c = Y CT where CT is gmoles/vol. For an ideal gas CT = p/RT
and c = Y p/RT.
Thus Y =CRT'/P = (298.16)(82.057)
(1) C=
24466 12
. C
CT = p/RT == 1/(82.057)(298.16) == 4.087xlo-5 gmoles/cm3
The equilibrium expression is
(CJmaxK RT/p)c
q = 1 + (KRT/p)c
Thus a = CImex KRT/p

:; (3.65xlO- 3 )(0.4179)(82.057)(298.16)/1.0 = 37.32
and b = KRT/p = 10,224.4.
We wish to go from Cunt = 0 to
CF = CTYF =(4.087 x10-5 ) (0.05) =2.044 xl~ gmoles/cm 3 •
Pp = 0.484/0.61 = 0.793 g/cm3
Since the mass balance was solved without a dispersion term, Eqs. (8-
17) to (8-19b) are appropriate.
~ = 6!~ = 6/0.1 cm = 60 cm- 1
Average mole fraction in M1Z in 2.5% Cf4
MW = 0.975(4) + 0.025(16) = 4.30
Pf = ~ = (4.087xlo-5 )(4.3) = 1.757xlO-4g/cm3

381
F ac dq a
or q = 1 + be ' de = (1 + be)2
InEq.(8-19b),A= [~~l = a
(1 + be)avg
2
avg
Cavg = Yavg CT = (0.025)(4.087xlO-s) = 1.022xlO--6

A = (0.7093) (37.32) = 28.986
Pp [(1 + (10,224.4)(1.022xlO-6)f
Reynolds number Re = Pf(V ee>dp/J..L

From Perry and Green (1984, p 3-248) at 25°C and 1 atm,
J..LHe = 0.000186 poise, J.l.cH. = 0.00011 poise
The mixture viscosity can be estimated (Bird et aI., 1961, p.24)
Js [1
2
MW -1/2 1/2 MW 1/4
where 4'., ~ + MW; 1 1+ [~ 1 [MW; 1[ ]
let He =I, C~ =2
i j MW.jMW·
1 J J..LJJ..Lj «I>ij ~x·«I>··
J 1)
1 1 1.000 1.00 1.00
1.039
2 0.25 1.69 2.548
2 1 4.0 0.591 0.3767

0.541
2 1.000 1.00 1.00
382
. = (0.975)(0.000186) + (0.025)(0.00011) = 0 0001796

J.lmix 1.039 0.541·
R = (0.0001757)(55)(0.1) = 5 38
e (0.0001796) .
Dm for C~ in He at 25°C is given in Sherwood et al. (1975, p23)
Dm = 0.675 cm 2 js at 1 abn
From E q. (6-1) - ET - Ee = 0.763 - 0.39 = 0 61

c , Ep - 1 _ Ee 0.61 .
Eq. (6-45a) Dmp = Ep~m = (0.61)~0.675) = 0.0514 cm2/s

Use Eq. (8-16) even though Reynolds number is a bit low.
(1.17)(55)°·585 (0.0001757)0.252(0.675)213
kf = = 24.27cm/s
(0.1)°.415 (0.000 1796)°·252
kf~ = (24.27)(60) = 1456.2
r
From Eq. (8-17a) Ckap)pore = (60)(0.0514)(28.986)(0.61) = 5453
, (0. I?
From Eq. (8-19a), 1<",... ~ [ 14;6.2 + s153 ~ 1149.3 S-l
Can now determine A't from Eq. (8-12a).
6. (0.61)(0.793)(37.32) {l [0.05]
't = (1149.3)[1 + (1O,224.4)(2.044xlO~)] n 0.95
_ [1 +(1O,224.4)(2.044xl0~)]
(10224.4)(2.044xlO~)
In [[ 0.05] [0.05]]}
0.95 0.95
I..rz ~ at ~ (0.0154{ -2.944) - (-287.66) } ~ 4.381

383
To find LMfZ = (L\'t) IIp , need IIp = llsh'

(eev)
L\q q(CF) - 0
where - = O' Note that q vs c equilibrium data was
L\c CF-
obtained in a manner consistent with a one porosity mass balance.
From the Langmuir isotherm,
(37.32)(2.044xlO--6) = 7.472xlO-5
1 + (10224.4)(2.044xl(J 6)
then, L\q = 7.472xl0~ = 36.556

L\c 2.044x 10
55
hence, IIp = 0.39 + (0.61)(0.793)(36.556) = 3.043 cm/s
and LMfZ = (4.381)(3.043) = 13.33 cm
This is quite short since mass transfer is rapid in gas systems.
We stop the feed step when breakthrough just starts. This occurs
when the front of the mass transfer zone reaches the end of the column.
From Eq. (8-2) this is
tF = (L - 0.5 LMfZ)/lIp = (100 - 6.665)/3.043 = 30.67s
Note that more than half of this example was involved with correlations
to determine physical properties.
8.4. OPTIMIZING FIXED BED SYSTEMS
How long should the column be? What size range of particles should be used?
How rapidly should the bed cycle? These questions are all part of the optimi-
zation process. A complete optimization requires a complete economic
384
analysis for each particular case. Since this is beyond the scope of this section,
we will only partially answer the question by giving general guidelines. The
argument will parallel that presented by Wankat (1987) for adsorbers and Wan-
kat and Koo (1988) for chromatography. We will show that beds should be
short, small particles should be used. and the beds should cycle rapidly.
As the column becomes longer the M1Z becomes a smaller fraction of

the total bed length. Thus the bed utilization increases and more concentrated
streams result on desorption. For symmetric mass transfer zones the fractional
bed utilization is easily calculated from Eq. (8-3) or Figure 8-4 (Lukchis,
1973). Once the bed length is two or three times longer than the M1Z, further
increases in bed utilization are quite modest. Since pressure drop increases as
L increases, further increases in column length are usually not justified. When
the M1Z is very long, the column-in-series arrangement shown later in Figure
8-15 should be used.
If L is to be set at 2 to 3 times LMI'Z. we need to know LMI'Z' For a

Langmuir isotherm LMI'Z can be determined from Eqs. (8-12a and c). The term
which has the most effect on LMTZ is the inverse dependence on ~~. The
mass transfer coefficient can be estimated from Eqs. (8-19) and the area per
unit volume.~, from Eq. (8-14). When film diffusion controls
(8-20a)
which can be determined from Eqs. (8-15a) or (8-16). When pore diffusion
controls
(8-20b)
When surface diffusion controls.
(8-2Oc)
For liquid systems packed bed adsorbers and chromatographs often have mass
transfer inside the particle (pore diffusion) controlling and ~8p is inversely
385
proportional to d~. By varying particle diameter, practically any km8p and

hence any value of LMI'Z can be obtained. The fractional bed use depends on
the ratio of LiLMI'Z' This ratio can easily be detennined. For instance, when a
lumped parameter mass transfer expression is adequate and pore diffusion con-
trols, Eqs. (8-12a), (8-12c) and (8-20b) give
--0.--
L L (8-21)
LMI'Z v d~
This result is also valid for surface diffusion control.
Pressure drop in the packing is important, particularly when large

volumes of gas are processed. For rigid particles the pressure drop can be
calculated from
Ap __Jl_v_,uper~L_ (8-22)
- Kd 2
P
where Ap is in Newtons/m 2 • For laminar flow, which is the usual case in

packed bed operation, the permeability K can be estimated from (Bird et ai.,
1960).
(8-23)
Equations (8-22) and (8-23) are generally valid for
~ Pr vsuper _1_ < 10 (8-24)

Jl l-ec
which is usually satisfied in packed bed adsorbers.
Decreasing particle diameter increases Ap as shown by Eq. (8-22). This

is the usual argument against decreasing dp. However, if we keep Vsuper and the
ratio (L/d~) constant then Ap is constant. If pore or surface diffusion controls
(which is often the case), keeping Vsuper and L/d~ constant will give a constant
386
I
I
I
~~~-I-
I
I
I
I
Figure 8-5. Solute movement diagram for diffuse wave. ~ = 2L1 , W2 =

2Wlo and t2 = 2tl. Thus, W21L2 = wdLlo and t21L2 = tIlL I .
ratio of ULMrZ (Eq. 8-21). Thus the fractional bed utilization during the con-
stant pattern step will be constant This assumes that £e and the equilibrium
isotherm do not change as <lp is varied.
During desorption a diffuse wave usually results. Now equilibrium is the

controlling factor and particle diameter has little effect. The diffuse wave
spread is proportional to the column length. As L decreases, the ratio (diffuse
wave width/L) remains constant. This is illustrated in Figure 8-5 which shows
a spreading diffuse wave at two different lengths. Thus, if the column is scaled
with both v and (L/d~) constant, we will have the same bed utilization, the
same separation, and the same pressure drop regardless of the column length.
As the column becomes shorter it will saturate sooner. Thus we must

make all parts of the cycle shorter. To scale properly keep the ratio (cycle
time) x v/L constant. This is illustrated in Figure 8-5 and Example 8-2. The
amount of feed processed per cycle is less, but since there are proportionally
more cycles per hour the total amount of feed processed per hour is unchanged.
Thus with small particles we use a short bed and obtain the same total separa-
tion and the same throughput of feed. The adsorbent productivity (kg
product./kg adsorbent, hour) is inversely proportional to L, and will be quite
large for columns packed with small particles.
The scaling procedure can be extended to other situations (pore diffusion

does not control, turbulent flow, decrease 6p, etc.) by allowing column diame-
ter and hence velocity to vary (Wankat, 1987). This extension is done by
387
taking the ratios of conditions for old and new designs. We assume that the old
design is adequate but is not optimized. Thus, from Eq. (8-22),
1 ~Pold [ Vold ] [Lold ] [dp,new]2 (8-25)

Rp = ~Pnew = Vnew Lnew dp,old
where we have assumed that Ee and fluid properties are constant. If ~ = 1.0,
the pressure drop will be the same for the two designs.
For a Langmuir isotherm we can find LMfZ from Eqs (8-12a) and (8-
12c). Talking the ratios for the old and new designs, we obtain.
(8-26a)
When RN = 1.0, the fractional bed use (see Figure 8-4) will be constant in the
two designs. If pore or surface diffusion controls, Eq. (8-26a) becomes
(8-26b)
The interstitial velocity v can be related to the volumetric flow rate Q,
(8-27)
Then the ratio of velocities is
(8-28)
Inserting Eq. (8-28) into Eqs. (8-25) and (8-26), we obtain
_1 = [~] [Dnew]2 [~] [dp,new]2 (8-29)

Rp Onew Dold Lnew dp,old
388
For a Langmuir isotherm we can find LMTZ from Eqs (8-12a) and (8-
12c). Taking the ratios for the old and new designs, we obtain
1
RN = [Qold
~w
] [Dnew
Dold
]2[ Lncw
Lold
][kmkut3p.oId
IIp,new
] (8-30a)
When pore diffusion controls, Eq. (8-308) simplifies to
(8-30b)
When pore diffusion controls, Eqs. (8-29) and (8-30b) can be considered as two
equations with two unknowns and four knowns. The six variables are l/Rp,
l/RN' (~ew/QoId)' (Dnew/Dold)' (Lnew/Lold), and (<ip,ncw/<ip.oId)' Any four of
these can be selected as known. A case study approach to solving Eqs. (8-29)
and (8-30a) when pore diffusion does not control is illustrated in Example 8-2.
Once the new design has been determined, the adsorbent volume is 1tD2L/4.
Then
(Adsorbent Volume)new = [Dnew]2 [Lnew ] (8-31)

(Adsorbent Volume)old DaId LaId
Many commercial adsorption systems use particles in the range of 1 mm.

Based on the argument in this section these particles are probably too large, and
the columns and cycles are probably too long. There are some practical rea-
sons why smaller particles may be difficult to use. Some of these arguments
are:
1. Careful packing procedures are required with small particles. Methods to

do this are used commercially.
2. A tighter sieving of small particles is required if axial dispersion is impor-

tant
3. The classical rules of thumb for length to diameter have been violated.
The columns are now short and fat instead of long and skinny. This will
389
require a redesign of the end fittings to prevent excessive volumes (see

below). Better fluid distribution may be required.
4. Piping and values must be redesigned to reduce volume (see below).
5. The valves must be designed for more rapid cycles. More accurate timing
is required.
6. Since the packed column will act as an efficient filter, the feed must be
free of suspended solids. However, since cycles are shorter there is less
time for suspended solids to build up and clog the bed. Feeds containing
suspended solids can be processed in fluidized beds (see Chapter 10).
7. The desired small particle size may not be commercially available, and
when they are available may be more expensive than large particles. In
addition, the lifetime of small particles may be shorter.
All of these practical arguments are true, but they can usually be taken care of
by proper design. There are other conditions where these scaling methods may
be useful (Wankat, 1987). Scaling of chromatography systems is quite similiar
(Wankat and Koo, 1988), but will not be developed here.
The analysis in Eqs. (8-20) to (8-31) has been concerned only with the
packing. As the length of packing L decreases the volume of fluid outside the
packing can become important Referring to Figure 8-6, we see that the extra-
column volume on the feed side is
(8-32)
Vec,feed = V valve 1 + V pipe + V distributor + V frit
The equation for Vec,product will be similar.
Head Support or
Space Frit
(
Pure
Packing -\---CKI------ Carrier
Waste or I----- L - - - . I Purge

Regeneration Distributor Fluid
Product
Extracolumn volume in adsorption system.
390
Consider a very simple two step cycle: Feed and Regenerate. At the end
of the feed step, the fluid in volume Vec,fccd is unprocessed feed. This material
will reduce the recovery of carrier and will contaminate the regeneration pro-
duct In addition, the pores of the column contain unprocessed feed. Assuming
that the mass transfer zone is symmetric, we obtain
(8-33)
as the volume of unprocessed feed in the pores. The fraction of feed not pro-
cessed per cycle is
Vec,fccd + Vpores,feed (8-34)

frac. feed unprocessed = Q
tF
where QtF is the volume of feed during the feed step.
To consider the effect of this unprocessed feed, we can take the ratio of
Eq. (8-34) for the new and old designs.
(frae. feed unprocessed)new (Vec,feed + Vp«eS,fccd)new tp,old (8-35)

(frac. feed unprocessed)old
=-----"-----
(Vec,feed + Vpores,fccd)old tF,new
We will consider the simple case where pore diffusion controls and
Qold = Q,ew. Then, the previous analysis showed we want Vnew = VoId,
Dnew = Doid , (L/d~)new = (L/d~)Old' and tF,new/tr,old = LnewILold·
First, consider the case where Vec,fccd « Vpore,feed for both designs (this
is often true for long columns). Then, Eq. (8-35) becomes
(frae. feed unprocessed)new (1 - 0.5 LMrZIL)new (8-36)

-:---::-------- - - 1
frac. feed unprocessed)old - (1 - 0.5 LMTZiL)old -
This equation equals one since the scaling procedure makes (LMrZIL)new =
(LMTZiL)old' This result shows that the pore volumes scale appropriately. If
Vcc,feed is not negligible compared to Vpores,fccd; then Eq. (8-35) equals 1 if
(Vcc,feed)new Lnew (8-37)

=--
(Vcc,fccd)old
391
As the volume of packing is reduced the volume outside the packing must also
be reduced.
Actually, Eq. (8-37) is more stringent than practical requirements partic-

ularly for dilute systems. For dilute systems Qtr is often quite large since the
shock wave moves very slowly. Equation (8-34) shows that the fraction of
feed unprocessed will be small. Practically, changes in this fraction may be
unimportant For concentrated systems changes in the fraction of feed unpro-
cessed is probably important. Then careful design is required to reduce
V cc,feed'
Example 8-2. Application of Scaling Methods
We have available 0.05 cm diameter BPL activated carbon parti-

cles. We wish to repeat the separation in Example 8-1 with the same
Ap, the same feed throughput, and the same or better fractional bed use.
Find the column diameter, length, feed period, and fluid velocity for the
new design.
Solution
Since both film and pore diffusion are important, we will use a
case study approach with (<lp,new/<lp,old) = 0.5. We want Rp = 1,
RN ~ 1.0, and Onew/Qold = 1. Equations (8-25) and (8-26a) need to be
satisfied. In Example 8-1 ~llp = 1149.3 S-l was calculated from Eq.
(8-19a) where the individual values of kf and kpore were found from
Eqs. (8-16) and (8-17a), and IIp is from Eq. (8-14). Note that the ratio
~ ~,oldlkm IIp,new) does not simplify when both pore and film diffusion
are important.
Trial-and-Error Case Study Approach:
a Guess vnew
b. Calculate kfllp. Ckap)pon: and ~llp
c. Find (Lold/Lnew) from Eq. (8-25) with Rp =1
d. Calculate RN from Eq. (8-26a)
e. IfRN = 1, are finished. If not, start over.
392
Step a. If pore diffusion controlled, vnew = voId' If film diffusion con-

trols, vnew = 0.75 VoId (Wankat and Koo, 1988). Since Example 8-1
showed that pore diffusion was more important than pore diffusion, try
VlUper,ncw =0.8 vlUper,oId =(0.8)(55) =44 cm/s.

Step h. Following Example 8-1, 8p =6/flp =6/0.05 = 120.
(1.17)(v )0.585(0.0001757)°·252(0.675)2(3
k - super - 3 104(v )0.585
r- (0.05)0.415 (0.0001796)0.252 -. super
Thus, kf = 3.104(44)°.585 = 28.40 cm/s and kr3p = 3408.1. From Eq.

(8-17a),
flr-<» = (60)(0.0514)(28.986)(0.61) = 21 811.8

\---p pore (0.05)2 '
r
Note that (k8p)pore is not a function ofvsuper. Then from Eq. (8-19a),
1<,., ... = [34~8.l + 21,;11.8 = 2947.6
Step C. Lold = [vncw ]

Lncw Vold
[~'Old
f1p,ncw
]2_1 = (0.8)(4)(1.0) = 3.2
Rp
Thus, Lncw = 1.0/3.2 = 0.3125 m
1
Step d. RN =(0.8)(3.2) [1149.3]
2947.6 =0.998
or RN = 1.002 which is close enough. The new diameter can be deter-

mined from Eq. (8-28)
[ Dncw]2
Doid
= [~]/[~]
Vncw Qncw
= [1.]/(1.0)= 1.25
.8
Dncw = 1.118 Dold

393
The new value of LMrZ can be found.
Thus LMTZ,new =Luewn.52 =31.25n.52 =4.16 cm.

The velocity of the mass transfer pattern is
vnew
IIp,new = IIp.old - - = (3.043)(0.8) = 2.434 crn/s
VoId
The feed period can now be determined from Eq. (8-2),
tP,new = (L - 0.5 LMrZ)/llp = (31.25 - 2.08)12.434 = 11.98 s
Then,
tP,new =0.391= [Lnew

tp.old Lold
]1 [vVoIdnew ] = (L/v)new
(L/V)old
The new design is shorter and cycles faster, but has the same separation,
same ~p and same throughput. Obviously, the new design uses less
packing.
8.5. INTERACTING ADSORPTION SYSTEMS
This series of chapters only skim the surface of theories for adsorption,
chromatography and ion exchange systems. The theories were kept simple
(remember, this is relative to other theories) by discussing only isothermal,
linear adsorption with several solutes or isothermal, non-linear, single solute
adsorption. In this section we will briefly look at the results of more complex
experiments and theories where there are several interacting solutes or the
column is adiabatic. Most of the theories required for these analyses are
beyond the scope of this book; however, references for further study will be
provided.
394
8.5.1. Interacting Solutes
In general, when there are several solutes they will interact. Thus the equili-
brium for one solute depends on the concentration of all solutes. The simplest
equilibrium form which expresses this dependence is the multicomponent
Langmuir expression,
CJMAxKjcj
qj = N (8-38)
1 + l:(KiCi)
i=1
When the concentration of component i increases, the amount of component j

adsorbed decreases (i :f;: j). There is competition for sites. Equation (8-38) can
be adapted to ion exchange also.
The behavior of coupled systems has been extensively studied. The

results for adsorption of n-pentane and n-hexane from iso and cyclic hydrocar-
bons which are not sorbed by the zeolite are shown in Figure 8-7 (Lee, 1972)
are typical. The least adsorbed solute, n-pentane, is first adsorbed and is then
desorbed as the more strongly adsorbed solute, n-hexane, displaces it. Since it
~.
:~ 1.00 ,.--- i--=-C>o
o_p"..!. LUB ---- :'r',..a'""'"' ~
~D
;/°r· "'"hT''''' 1<00' ~

~ ~ 0.75
-'
• 0- Pentone concentration
" V- -
f•• d
"'0_He.one concentration
n.Hexcne stOleh,ometrlc Ironl In
oJ o.SO
In I••d
I I 0
+ "_hexon.
'T'O
,.,-Pentone - "'Pentone
"""'j'"""<00
0.25
. o
.,(""'.0
~ o 10 15 20 25 30 35 40 45 50
A.dS'Hplion Time, Minutes
Figure 8-7. Adsorption of n=pentane and n-hexane on zeolites at 615°F.

Feed is 0.4 mole % n-butane, 25.9% n-pentane, 23.9% n-
hexane and 49.8% iso and cyclic hydrocarbons. (Lee, 1972).
Reprinted with permission from N.N. Li (Ed.), Recent
Developments in Separation Science, Vol. I, CRC Press, Boca
Raton, FL, 75-112, 1972. Copyright 1972, CRC Press.
395
is being pushed off the adsorbent, the concentration of n-pentane rises to a

value above its feed concentration. This behavior often is referred to as roll-
up. The n-pentane concentration remains at this plateau value until the more
strongly adsorbed n-hexane starts to breakthrough. Both solutes then go to
their feed concentrations.
Three plateaus (initial concentrations of zero, intermediate plateau with

high n-pentane concentration, and feed concentrations) are shown in Figure 8-
7. In general, for each step change there will be N + 1 plateaus where N is the
number of adsorbates. (In ion exchange there will be N plateaus where N is the
number of counterions). The two transition zones between the plateaus can be
either constant pattern (shock waves) or proportional pattern (diffuse waves).
The transitions shown in Figure 8-7 are both constant pattern. When this
column is eluted, the two transitions would both be proportional pattern. When
there are more components, the patterns can become very complex.
Theoretical calculation for multicomponent systems is difficult. For each

solute in adsorption there is an independent mass balance, Eq. (6-42) or (6-48),
and a mass transfer relation, Eq. (6-47) or Eq. (6-50). (For ion exchange there
are N-l independent equations since electroneutrality must be satisfied.) All of
these equations are coupled by the equilibrium relationship, Eq. (8-38). Thus,
all of the equations must be solved simultaneously. This problem has been
extensively studied using a complex form of equilibrium theory CAris and
Amundsen, 1973; Helferrich and Klein, 1970; Lightfoot et al., 1962; Rhee,
1971; Ruthven, 1984; Yang, 1987). Although rather complex, this theory is
still oversimplified and predicts shock waves instead of the S-shaped constant
pattern waves. The theory does correctly predict the plateaus, the transition
locations and the types of transitions. To include mass transfer effects numeri-
cal models are usually used (Ruthven, 1984; Yang, 1987).
For displacement chromatography (see Section 7.10) there is a simple

and useful local equilibrium result (Glueckauf, 1946; Rhee and Amundson,
1982; Wankat, 1986). In displacement chromatography the column is first
equilibrated with carrier C. Then a feed pulse of A plus B in C is added. This
is followed by a strongly adsorbed displacer D in carrier C. Finally, the
column must be reequilibrated by eluting with carrier C.
396
The strength of adsorption follows D > B > A > C. Thus the displacer
pushes out B which pushes out A which pushes out pure C. The result is a
series of bands: pure C followed by A in C followed by B in C followed by D
in C. This was illustrated in Figure 7-13. Each band is separated by a shock
wave. The solute movement diagram is shown in Figure 8-8. The dotted lines
represent transient behavior which requires the more detailed theory.
The band widths of A and B must become constant to satisfy a mass bal-
ance. This is also observed experimentally. Since the band widths are con-
stant, the shock wave velocities must be equatl.
(8-39)
llsh,l = llsh,2 = Ush ,3
The concentrations CA,band and CB,band are unknown; however, the concentra-
tions of D are known and the amount adsorbed can be calculated.
II <lD q (CO,F , CA = 0 , CB = 0) - 0 (8-40)

--=
CO,F - 0
Note that <lD can be calculated despite the coupled isotherm (Eq. 8-38) since D
and B never occur simultaneously. We can now calculate Ush,3
V
llgh.3 =--------------- (8-41)
l-Ec l-Ec II <lD
1 + ~ tp KcJ,o + ~ (l-Ep) Ps - -
Ec Ee II Co
From Eq. (8-39) this is also the value ofush.2' We can write Ush.2
V
ush.2 = ---------------:----:----:- (8-42)
l-Ec (l-Ec) (l-tp)ps q (CB.band) - 0
1 + - - Ep KcJ,B + ------=--
Ee Ec CB,band - 0
The unknowns are now ~ (CB,band) and cB.band' These unknowns are related by
Eq. (8-42) and the isotherm which is Eq. (8-38) or any other coupled, favor-
able isotherm. Solving the isotherm and shock velocity equations simultane-
397
----7"-
z c•.•
/
~/ 1
Fe,bOOd CO,F
/ 1//
Curved, / Y/.
Shock V /b
// /D
Figure 8-8. Solute movement diagram for displacement chromatography.
ously, we obtain cB,band and <In (CB,band)' This is illustrated for a Langmuir
isotherm in Example 8-3. A very similiar calculation can be done for com-
ponent A to determine CA,band and <lA (CA,band)'
Since band concentrations are known, we can use a mass balance to find
the band widths.
(8-43a)
(8-43b)
CB,F tp = CD,band fB,band
The times the bands breakthrough are easily determined.
(8-44a)
lo,br = lJU.h,3 + lp
(8-44b)
fB,br = tD,br - la,band
(8-44c)
tA,br = tD,br - la,band - tA,band
398
The removal of displacer D with pure C gives a diffuse wave. This wave is
easily calculated using the methods of Chapter 6. This step may be a major
expense of displacement chromatography because of excessive tailing. In ion
exchange chromatography column regeneration may be easy (see Problems 9-
D6 and 9-D7).
Note that the only important detail we cannot calculate is the column
length necessary to generate the pure bands. This calculation of the dotted
lines in Figure 8-8 requires more detailed theories (Rhee and Amundson,
1982).
Displacement behavior will not occur if the displacer concentration is too

low or D is not strongly enough adsorbed. If A moves faster than Uah.3, then
the A band will not follow these equations. llsh,A (CA) will be greater than llsh,3
if
(8-45a)
This dilutes the A band. Ush,S (CA,F) will be greater than ush,3 if
(8-45b)
and a separate A peak forms. Ifus (CA = 0) > ush,3. the A peak will move com-
pletely away from the B band (they will be separated by carrier), and the A
peak will have a diffuse tail. This occurs if
(8-45c)
Equations (8-45) assume Kd,A = ~,D'

Finite mass transfer rates and axial dispersion will broaden the sharp
shock waves predicted by the equilibrium theory. The effects of axial disper-
sion and finite mass transfer rates have been extensively modeled by solving
the coupled partial differential equations numerically (Phillips et ai, 1988).
399
Example 8 3. Displacement Chromatography
Displacement chromatography has been extensively studied for separa-

tion of small biomolecules. We input a 6 minute feed pulse containing
10 mM each of components A and B. The displacer D is input at 40
mM. Predict the band concentrations, the band widths, and the break-
through times.
Data: KDi = 1.0, Ee = 0.4, vsuper =1 cm/min, L = 25 cm.

.. . ~ Ci
Equihbnum: ~ = ~
1 + ~ bj Cj
where q is mM/ml and c is mM/ml.
aA = 3.297 , b A = 0.0905 (mM)-t

aB = 5.740, bA = 0.1247 (mM)-t
aD = 9.203, bn = 0.1501 (mM)-t
(from Phillips et al, 1988).
Solution
The shock wave velocity for displacer can be found from Eqs.
(8-40) and (8-41). From the equilibrium expression, we obtain
~ qo = qo (CD, CA = CB = 0) = 52.56 mM/g
~ qo = 52.56 = 1.314
~CD 40
For the units of this problem, Eq. (8-41) becomes
Ush,D = 0.4 + (0.6) (0.4) (1.0~ + (0.6) (0.6) (1.314) = 0.8984 cm/min
400
The shock waves for A and B must have the same value. From the
Langmuir isothenn
q (Ci,hand , Cj..i = 0) (847)

=----
~,band 1 + bi Ci,hand
Combining this result with Eq. (8-42), we obtain after some algebra.
(848)
Plugging in the numbers, we obtain CB,band = 27.01 mM and CA,band =

16.673 mM. From Eqs. (8-43),
tA,band = IF cA,F/cA,band = (6.0) (10.0)/(16.673) = 3.60 min
and !B band = (6.0) (10.0) 1(27.01) = 2.22 min.

From Eq. (8-44a), to,hr = 25/0.8984 + 6 = 33.83 min
From Eq. (8-44b), !B,hr =33.83 - 2.22 =31.61 min.
From Eq. (8-44c), tA,hr = 33.83 - 2.22 - 3.60 = 28.01 min.
Note that the six minute feed band is compressed to 5.82 minutes.
8.5.2. Adiabatic Adsorbers
Until now we have assumed that the column is isothennal or has step changes
in temperature. Large columns are often adiabatic not isothennal since a sub-
stantial amount of energy may be released when the solute adsorbs. The same
amount of energy is later required to desorb the solute. Thus the temperature
of the column rises during adsorption and drops during desorption.
To study adiabatic systems the energy balances are required. The energy
balance for both phases was given in Eq. (6-51) while the energy transfer equa-
401
tion is Eq. (6-55). The mass and energy balances and the equilibrium relation-
ship are all coupled. The coupling of the energy balance with the mass balance
occurs in the heat of adsorption term of Eq. (6-55). Equilibrium is coupled to
the energy balance since the equilibrium parameters depend on temperature
(Eq. (6-8». The mass balances are coupled to equilibrium since the mass
transfer rate and the amount adsorbed depends on the equilibrium.
Previously, we uncoupled the equations by assuming the column was

either isothermal or had negligible heat of adsorption effects. For an isothermal
system the energy balances are not required. When the effect of heat of
adsorption is negligible, the energy balance can be solved separately. Then the
temperature profile can be used to find the equilibrium conditions at any point
in the column. This latter method was done for thermal waves in Chapter 6 and
in Problems 7-C6, 7-C7, and 7-E3.
The effect of heat of adsorption will be negligible when:
1. Miadl; - O. This is true for exchange adsorption, displacement adsorp-

tion, and ion exchange. In these systems the sorbed species is replaced
by another sorbed species. The adsorption and desorption effects bal-
ance each other. These processes are usually isothermal.
2. Dilute gas systems. Now dq/dt - 0 since there is little adsorbate to

adsorb. Although Mi. ds may be substantial, the term
(l-Ep)(l-Ec)PsMf.dsdqf<)t in Eq. (6-55) remains small.
3. Most liquid systems. In liquid systems the volumetric heat capacity

(Pfc;,f) is large. The energy released by the heat of adsorption is either
removed by the flowing liquid or is soaked up by the solid and stagnant
fluid with very little increase in temperature. Thus operation is close to
isothermal.
4. Pressure swing adsorption systems with very rapid cycles. Temperature

does increase during adsorption and decrease during desorption. When
the cycle period is on the order of a few seconds or less, the temperature
increases and decreases remain very small. These operations are usually
treated as isothermal even though strictly speaking they are not.
402
r-------------------,35o
330
T
(K)
310
0.1
~-~L-_~ _ __ J _ _ _ L_ _ _ ~ ____ ~290
o 20 40 60
Z, eM
Figure 8-9. Concentration and temperature profiles for adiabatic adsorption

(Yang, 1987). Bed initially filled with H2 • Feed 50% H2 , 50%
Cf4. Reprinted with permission from R.T. Yang, Gas Separa-
tion by Adsorption Processes, Butterworths, Boston, 1987.
Copyright 1987. Butterworths.
Gas systems where items 1, 2 and 4 do not apply will have significant
temperature effects. This is illustrated in Figure 8-9 (yang, 1987). Note the
large increase in temperature due to the heat of adsorption. Because of the cou-
pling the thermal wave moves slightly ahead of the concentration wave.
Although gas mole fraction y is constant after the concentration wave, q will
vary since T varies. Since all the equations are coupled, the exact solution of
the partial differential equations is difficult. Equilibrium solutions can be
obtained using methods very similar to those used for interacting solutes
(Sweed, 1981; Yang, 1987). Complete solutions including heat and mass
transfer resistances require numerical methods (Ruthven, 1984; Basmadjian,
1983; Yang, 1987). The mass transfer zone will be wider in adiabatic systems
403
than in isothermal systems. Thus, it is important to include temperature effects

either in theoretical calculations or in laboratory experiments.
The behavior of gas systems can be studied qualitatively using the solute
movement theory of Chapter 6. When a gas is adsorbed, its density approaches
that of a liquid. Thus, ifadsorption is strong there will be much more adsorbate
on the solid phase than in the fluid. In Eq. (6-23) this means the first two terms
in the denominator will be very small compared to the third term. Then,
(8-49)
In most gas systems at modest pressure (CPfPf) <: (PsCPs ). Thus the
amount of energy stored in the solid is much greater than the amount in the gas
phase. Then the first two terms in the denominator of Eq. (6-36) will be small
compared to the third term. In commercial systems the column diameter is
large, W/Ac is small, the wall effect is small, and the fourth term in the denom-
inator of Eq. (6-36) can be neglected. Thus,
(8-50)
Taking the ratio of Eqs. (8-49) and (8-50) we have,
(8-51)
RT is the cross-over ratio (Garg and Ausikaitis, 1983). If RT » 1 then the

thermal wave moves ahead of the solute wave, while if RT « 1 the thermal
wave lags behind the solute wave. In both these cases the mass and energy bal-
ances are essentially uncoupled. The temperature in the region of the solute
wave can be estimated.
The most interesting and complex case occurs when RT is approximately

1.0. Now the solute and thermal waves travel together and no simplifications
404
are possible. This is the case in Figure 8-9 (yang, 1987). Although theoreti-
cally complex, this operation can be advantageous. Since the loading step is
stopped before breakthrough of the solute wave, the thermal wave also remains
in the column. The energy released during adsorption is now available for
desorption. This principle is commonly used in PSA systems where the cycles
are only a few minutes long. Recently, operation with RT near one has been
used in thermally regenerated adsorbers drying ethanol vapor (Ladisch et al.,
1984; Garg and Ausikaitis, 1983). The energy required for regeneration can be
significantly decreased. Unfortunately, the theory of these systems is beyond
the scope of this book.
Adiabatic adsorbers can be designed using the MIZ-LUB approach.

Care must be taken that the laboratory columns are equivalent to the large scale
system. As the column diameter increases the ratio of wall surface area to
column volume decreases. Thus the wall heat transfer term and the wall
storage term (the right hand side of Eq. (6-51» both become less important. In
commercial systems with large diameters both these terms are usually negligi-
ble. Thus even without insulation the column will effectively be adiabatic. In
small diameter laboratory columns the wall effects are relatively much more
important. Even with insulation it is difficult to make the wall heat transfer
term (term 6 in Eq. (6-51» negligible. Thus the laboratory and commercial
columns are not equivalent The solution to this problem is to use a large diam-
eter laboratory column (at least 3 inches in diameter) and to insulate the
column carefully. Then the experimentally measured MIZ will include heat
effects. Since the heat effects are approximately the same in the laboratory and
commercial columns, this MIZ can be used to design the larger column.
8.5.3. Velocity Effects
Up to now we have usually assumed that the fluid interstitial velocity v is con-
stant For liquids, the density of the adsorbed material is approximately equal
to the liquid density. In addition, changes in pressure or temperature do not
markedly change the density. Thus the velocity will remain approximately
constant
405
For gases there can be large velocity changes. The density of the
adsorbed material is approximately the same as a liquid and is much greater
than the gas density. Thus, as material adsorbs the gas volume decreases since
solute is removed from the gas phase. The velocity must be lower ahead of the
solute wave than it is behind the solute wave. This causes self-sharpening or
shock type behavior even for linear isotherms. Upon desorption the opposite
happens. The gas volume increases, velocity increases, and a spread or diffuse
wave results. This velocity effect has been called the sorption effect. and it can
be calculated quantitatively (Ruthven, 1984; Wankat, 1986).
The sorption effect causes shock and diffuse waves under the same con-
ditions a Langmuir isotherm will. Thus they reinforce each other and are
difficult to tell apart. For gas-liquid chromatography (OLC) where the isoth-
erms have the opposite curvature (they are unfavorable), the sorption effect
causes a shock wave when the isotherm causes a diffuse wave and vice versa.
Thus the sorption and isotherm effects tend to cancel each other in OLC. By
tuning the column temperature, OLC peaks can be made to be sharp and sym-
metrical for concentrated systems where both sorption and isotherm effects are
important (Roz et al.• 1976). This will be true for any sorbent with an unfavor-
able isotherm.
In gas systems temperature and pressure changes will also change the
density and hence the velocity. If temperature increases, the gas velocity
behind the thermal wave increasys. Thus hot waves are compressive. For a
temperature increase from 20 to 80°C the velocity increase is about 20%.
Cooling waves will be dispersive by the same percentage. Changes in pressure
will have similar but opposite effects. The pressure effects occur very rapidly
and are probably important only with very rapid cycles (1 second or less).
8.6. ADSORPTION - DESORPTION OPERATIONS
Adsorbing the desired solute is often fairly easy. The trick of economical
operation is the desorption step. The most commonly used desorption methods
are thermal desorption, pressure swing desorption, purge gas desorption and
desorption using a desorbent. These methods will be considered separately.
406
8.6.1. Thermal Desorption of Gases
To have continuous treattnent of a feed stream at least two columns are

required. A fairly typical arrangement is shown in Figure 8-10. The feed gas
passes through the operating adsorber where the solute is adsorbed. The
treated gas is one product A small portion of this gas is heated and used to
regenerate the column in counter-flow. This can be considered a type of reflux.
The concentrated gas from desorption is sent to a further separation scheme,
exhausted or burned. The operation is timed so that when the first column
starts to breakthrough the second column has been regenerated. The columns
are then switched and feed goes to the second column while the first column is
desorbed. Often the desorbed column is cooled before the feed is switched to
it. Many more complex cycles and multi-column arrangements are used (Ruth-
ven, 1984; Wankat, 1986a; Yang, 1987). Note that Figure 8-10 represents a
batch operation and not steady state. Feed is processed continuously because
there is always a clean column to send the feed to.
The solute movement theory for this type of operation was shown in Fig-
Feed
r - - - - Concentrated
Gas
J
Adsorb Desorb
Heat
Treated
Gas
Figure 8-10. Two column system with thermal desorption.
407
ure 6-12 for a system with linear isotherms and a negligible heat of adsorption
term. This figure is for one of the two columns. The linear solute movement
theory does not illustrate all of the important features of the cycle shown in
Figure 8-11. To do this more complex theories are required. The non-linear
isotherm theory with negligible L~Hadsdq!dt is illustrated in Figure 8-11. Note
that the column is not completely regenerated but some solute is left in the
column. This is called a heel and may be quite large. This diffuse wave is then
resharpened by the shock wave during the next cycle. The shock wave in Fig-
ure 8-11 is curved. This occurs because the concentrations upstream of the
shock wave are continually changing; thus, the last term in the denominator of
Eq. (6-23) changes. In practice, the shock wave will result in a constant pat-
tern wave. The M1Z approach can be used to predict the shape of this wave.
Partial regeneration means that the entire equilibrium capacity of the column is
not available for adsorption. However, partial regeneration is usually much
cheaper than complete regneration since an excessive amount of hot gas would
be required to completely remove the slow diffuse wave.
Figures 6-12 and 8-11 have assumed that the term 6Hads dq/dt in Eqs.
(6-53) or (6-55) is negligible. This term will be important in many gas streams
and Figures 6-12 and 8-11 will not be valid. A more complex form of the local
r:
F Concentrated
j
Cold
Diffuse
z Wave
Thermal Wave Hot
t
Treated Gas
t
Hot
Gas
Figure 8-1l. Solute Movement diagram for column shown in Figure 8-10
for non-linear isotherm. The term MIads dq!dt is assumed to be
negligible.
408
Table 8-1. Characteristic Temperatures To for Various Systems at 1 atm.

Basmadjian (1975a). Reprinted with permission from Ind. Eng.
Chern. Process Des. Develop., 14, 328 (1975). Copyright 1975,
System
C02-C~-5A molecular sieves -230
H20-air-5A molecular sieves >600
H20-air-silica gel -250
H2S-CH4-5A molecular sieves -400
Acetone-air-activated carbon -300
a Values are accurate only to ±1O% because of uncertainities in the isotherm

slopes and heat capacity values.
equilibrium theory can be used to analyze adiabatic adsorbers when the heat of
adsorption term is important (pan and Basmadjian, 1971; Basmadjian et al.,
1975a,b). The term ~H.ds dqldt couples the mass and energy balances, and
thus a mathematical uncoupling of the equations is required. The results of
these theories shows that efficient desorption occurs above a characteristic tem-
perature To. The characteristic temperature is the temperature at which the
slope of the adsorption isotherm at the origin equals the ratio of heat capacities
CPs/Cpf. Characteristic temperatures for several systems are given in Table 8-1
(Basmadjian, 1975a). Below the characteristic temperature regeneration is
inefficient and large volumes of purge gas are required. Above the characteris-
tic temperature the regeneration is slightly more efficient, but the reduction in
purge gas volume is very modest. Since heating the purge gas is expensive,
regeneration is usually done slightly above the characteristic temperature. The
use of too low a regeneration temperature also decreases the amount adsorbed
in equilibrium studies (Joshi and Fair, 1988). The regeneration temperature
required for essentially complete equilibrium regeneration is below the charac-
teristic temperature.
The basic system shown in Figure 8-10 can be used for a variety of
separations (Keller, 1982; Kohl and Riesenfeld, 1979; Ruthven, 1984; Wankat
409
1986), but operation by thennal swing is economical only for feeds with low
adsorbate concentrations (usually under 5 mole %). With higher feed concen-
tratios the energy costs for regeneration become excessive (Keller, 1982).
Thennal swing cycles do have the advantage of providing high (90 - 99%)
removal and recovery of adsorbate. Systems similar to Figure 8-10 are used for
removal of traces of S02, mercury and NO x from waste gas streams. Zeolite
molecular sieves are used for adsorbing S02. During desorption the concentra-
tion can rise as high as 4% compared to the feed concentration of 2000 to 4000
ppm. A 4% S02 stream can be reused to make sulfuric acid. Thus a major
advantage of adsorption with thennal desorption is the de sorbed solute is at a
much higher concentration than in the feed.
Drying of gases with adsorbents is a very common industrial and labora-
tory practice. You probably did this in chemistry laboratory using a solid such
as Drierite which changes color from blue to pink when water is adsorbed.
Industrially, activated alumina, silica gel and molecular sieve zeolites are used
for adsorption drying. Typical adsorbent properties are listed in Table 8-2
(Basmadjian, 1984). Equilibrium isothenns for drying air using several adsor-
bents are shown in Figure 8-12. If very low water concentrations are required,
molecular sieve zeolites are preferred. Molecular sieve zeolites are more
expensive than silica gel and activated alumina, and Table 8-1 shows the zeol-
ites require a higher regeneration temperature. The cheaper adsorbents are
often used for less demanding applications, or when the feed gas is very con-
centrated. A layer of cheap adsorbent such as silica gel followed by molecular
sieves is useful to achieve low moisture constant at low cost particularly for
concentrated feeds. Basmadjian's (1984) review contains data and infonnation
on adsorbent drying.
Regeneration of silica gel and activated alumina is typically done at 100
to 200°C while zeolite molecular sieves require 200 to 300 °C or higher (see
Table 8-1 for specific examples). The energy required for the molecular sieve
systems is also higher. Counter-flow regeneration, shown in Figures 8-10 and
8-11, uses less energy than co-flow regeneration; however, co-flow regenera-
tion (see Problem 8-A 7) is simpler and has lower capital costs. Both are used
industrially. A short cooling period (co-flow to the feed step) is often used.
Fast cycles increase the adsorbent productivity and smaller equipment is
....o+>-
Table 8-2. Properties of Commercial Desiccants: Manufacturers' Data

(Basmadjian, 1984).
Reprinted with permission from A.S. Mujumdar (Ed.), Advances in Drying
, Vol. 3
Copyright 1984, Hemisphere Pub. Co.
Density (g/crn 3 )
Type Particle Bulk Particle Surface Pore Pore Specific Maximum
size,mm area,m2 /g Volume, Diameter, heat heat of
crn3 /g run J/gK
Activate d alumina
1. Alcoa F-l <12 0.85 1.42 250 0.40 2.6
2. Alcoa H-151 3-6 0.85 1.38 360 0.43 4.3
3. Laporte Actal 6-12,3-6 0.64 1.15 275 0.50 2.8 0.88 2.88
4. Rhone-Poulenc 2-5,5-10 0.77 345,315 0.40
5. Kaiser A-201 <12 0.75 1.4 350 0.46 5.2
Silica gel
6. Davison 03 2 0.72 1.2 750 0.43 2.2 0.92 3.26
7. Davison 59 2-7 0.40 340 1.15 14
8. BASFE 3-5 0.75 750 0.43 2.3 1.0
9. BASFWF 1-2 0.46 370 0.97 10.6 l.0
Molecular sieves
10. unde 4A 1.6,3.2 0.66 0.4 l.0 4.19
11. linde 13X 1.6,3.2 0.61 1.0 1.0 4.19
12. Davison 513 (4A) 2-5 0.72 1.65 0.4 0.96 4.19
13. Davison 542 (13X) 2-5 0.69 LO 0.96 4.19
14. BayerT143 (4A) 1.5-2.5 0.71 1.15 0.4 0.92 3.77
15. Bayer W894 (9X) 1-4 0.65 L05 0.9 0.92 3.77
~
412
.60
/'
O.94-{).98
at 100% R.H .
.50
.40
~
"0
In
~
"0.30
o-£
Cl
13)( Sie~eS 11 1 .13)

Cl
~ ~ ~ 00 ro 00 ~ 100
R.H.%
Figure 8-12. Typical adsorption capacities for removal of water vapor with
fresh, activated adsorbents. Numbers refer to Table 8-2.
(Basmadjian, 1984). Reprinted with permission from A.S.
Mujumdar (Ed.), Advances in Drying, Vol. 3. Copyright 1984,
Hemisphere Pub. Co.
required. Slow cycles usually have longer adsorbent life and are somewhat
simpler to control. The trend has been to shorter cycle periods since major
reductions in capital cost can be achieved.
8.6.2. Activated Carbon Solvent Recovery
A modification of the basic system is used for solvent recovery using activated
carbon. This has been an extremely common industrial practice since the
413
F~Co" 0r--__.
Condense
i~Solvent
~water
Clean Air Steam Hot Air Cold Air

(Optional Steps)
Figure 8-13. Activated carbon system for solvent recovery using steam
desorption.
1920's (e.g. see Mantell, 1950). This is illustrated in Figure 8-13. The air
stream containing low amounts of the solvent (from painting, printing,
adhesives, coating operations, etc.) is slightly compressed, filtered, cooled to
80 to 90°F and then sent to an activated carbon adsorber. Activated carbon is
an excellent adsorbent for a large variety of solvents particularly at the low
concentrations encountered in solvent recovery. Concentrations are low since
for safety reasons the solvent must usually be kept below 1/4 to 1!2 the lower
explosion limit (LEL). Typical values for the LEL of common solvents are
(vol %): 3.3; acetone, 2.6; n-butyl alcohol, 1.7; chloroform, non-flammable;
ethanol, n-hexane, 1.2; methanol, 6.7; methyl ethyl ketone, 1.9; and toluene,
1.2. The air exhausted is clean. Before breakthrough, the adsorbers are
switched and the carbon is regenerated using steam or hot gas. Steam is con-
venient since condensation of the steam rapidly heats the adsorber, serves as a
purge gas, and the steam minimizes fire hazards. The outlet vapor is con-
densed. If the solvent is immiscible with water the two liquid phases can be
separated as shown in Figure 8-13 and the solvent can be reused. The steps to
cool and dry the carbon are optional. They are often included since hot
activated carbon may catalyze oxidation of the solvent, and water in the bed
can interfere with solvent adsorption. Addition of a cooling step usually
increases capapcity by about 4%. Usually the MTZ is quite short. Thus shal-
low (from one to three feet) wide beds can be used. The beds are usually held
in horizontal cylindrical vessels. A variety of other adsorber geometries such as
414
cylindrical canisters, pleated cells, flat cells, rotating cylinders and fluidized
beds are used (Turk, 1968; Wankat, 1986). Low pressure drop is important
since operation of the blower is a major expense. Typical pressure drops are
approximately 0.5 inch water per inch of bed. More details are provided by
Fulker (1972) and Turk (1968).
Activated carbon is useful for recovering non-reactive solvents with

molecular weights in the range of 45 to 200. Lower molecular weight com-
pounds tend to adsorb too weakly, and heavier compounds are difficult to
desorb. If traces of heavy compounds are present, a guard bed (a small bed in
front of the main bed) is useful to protect the main bed. The guard bed may be
periodically replaced instead of trying to regenerate it. If the solvent is misci-
ble or partially miscible with water, distillation is usually used to recover the
solvent from water. The distillation system may be more expensive than the
adsorbers. Usually, the value of the recovered solvent is expected to pay for
the capital and operating costs.
The adsorption step is relatively easy to model. A shock wave results

since equilibrium is favorable and very non-linear. The M1Z method can be
applied if adjustments for temperature increases are made. Application of the
theory to the desorption step requires some modification of the thermal wave
velocity to include the latent heat of steam from condensation. The resulting
steam wave velocity is (see Problem 8-C5)
(8-52)
Yw,b and Ystm are the water mole fractions in the vapor before the thermal wave
and in the steam, respectively. Aw is the latent heat of vaporization of water. If
steam is not used and there are no condensible vapors, Ystm = Yw,b = 0, and Eq.
(8-52) reduces to Eq. (6-36). The condensing steam wave moves considerably
slower than the gas velocity v, but considerably faster than a thermal wave cal-
culated from Eq. (6-36).
415
The diffuse wave which results from desorption is spreadout. Thus it is

common to not completely regenerate the bed (as in Figure 8-11). This means
that the working capacity is only a fraction of the equilibrium capacity (about
20-33%). For many solvents this gives a working capacity of around 7 to 9 kg
of solvent per 100 kg of carbon. This can be used for rough sizing of the
adsorber during the feed step. The usual range for steam use is 1 to 5 pounds
of steam per pound of adsorbate. Typically, less than 10% of the energy in the
steam is used for removal of the adsorbate. About 5% is used for heating the
carbon and the steel shell. The rest exits with the hot gas while slow desorption
of solvent occurs. A more complete design is usually done using experimental
data. Examples of the maximum capacities for different solvents are given in
Table 8-3. (Marc hello , 1976; Turk 1968). A conservative first estimate of the
carbon requirements can be made by assuming the working capacity is 20% of
the maximum.
The maximum capacities in Table 8-3 assume that relative humidity of

the air is low. The activated carbon does not adsorb water in the same way it
adsorbs organic solvents. However, at high relative humidities the water vapor
can condense in the capillaries. This condensed water blocks these pores and
reduces the capacity of the carbon. As a rough rule of thumb, relative humidi-
ties below 50% do not cause problems (Turk, 1968). For higher relative humi-
dities, it may be necessary to partially dry the air by condensation or adsorp-
tion, or to operate at a higher temperature where the relative humidity is lower.
Small scale activated carbon solvent recovery adsorbers (less than about
10,000 pounds of carbon) are usually purchased as packaged units. These sys-
tems show an economy of scale and become cheaper per pound of carbon as
the size increases. In the exponential cost equation for scale-up the exponent is
0.48 (Vatavuk and Neveril, 1983). Larger systems must be custom designed.
The exponent becomes 1.20 and there is no economy of scale.
There are many alternatives to steam desorption (Wankat. 1986). All
these methods have the advantage of not adding water to the system.
1. Hot gas desorption can be used. This operation will be very similar to the
thermal desorption methods discussed previously. The major disadvan-
tage of this approach is that large volumes of hot gas are required to
416
Table 8-3. Maximum capacity of activated carbon for various solvents

from air at 20°C and 1 atm. (Marchello, 1976; Turk, 1968)
Adsorbate Maximum Capacity, kg/kg Carbon

Carbon tetrachloride, CC4 0.45
Butyric acid, C4Hg02 0.35
Amyl acetate, C7 H 140 2 0.34
Toluene, C7 Hg 0.29
Putrescene, C4H 12 N 2 0.25
Skatole,C9~N 0.25
Ethyl mercaptan, CzH6S 0.23
Eucalyptole, ClOH1gO 0.23
Ethyl acetate, C4Hs 0 2 0.19
Sulfur dioxide, S02 0.10
Acetaldehyde, C 214 0 0.07
Methyl Chloride, CH3 CI 0.05
Formaldehyde, HCHO 0.03
Chlorine, Cl2 0.022
Hydrogen sulfide, H2S 0.014
Ammonia, NH3 0.013
Ozone, 0 3 decomposes to O 2
transfer the required energy into the adsorber. This method is used com-
mercially when it is important to avoid adding water to the system and in
isolated locations where steam is not available. Applications include
recovery of easily hydrolyzed esters which react with water and recovery
of solvents which form homogeneous azeotropes with water.
2. Vacuum desorption is used commercially for very concentrated gas

streams (3 to 50 vol. %) such as the gasoline vapors leaving storage tanks
when the tanks are filled. The purpose of pulling a vacuum is to lower the
solute partial pressure and thus regenerate the adsorbent. This is shown in
Figure 8-14. When the column is evacuated, the non-adsorbed gases are
417
Useful
Capacity
q
i5
Figure 8-14. Isothenn illustrating vacuum desorption.
removed first. Thus, Ysolute ~ 1.0 and the partial pressure during evacua-
tion is
(8-53)
Pvac = Ysolute Ptot,low - Ptot,low
whereas PF = YF Ptot,F' If the feed mole fraction is low, PF will be low.

Then the final pressure, Ptot,low, must be very low in order to have
Pvac < pp. The useful capacity (change in amount adsorbed over the cycle)
will be quite small. Thus, vacuum desorption in general has been limited
to concentrated feeds. For concentrated feeds vacuum desorption can be
quite advantageous since the adsorbate can be recovered at quite high pur-
ities.
3. An alternative which can be useful for solvents which are difficult to

separate from water is to use steam for desorption and then incinerate this
waste stream. The incinerator can be used as a boiler to produce the steam
required for desorption.
4. The carbon can be regenerated in a kiln. This is commonly used in waste

water treatment where high molecular weight organics are adsorbed (e.g.
Hutchins, 1979, or Perrin, 1981), but would be unusual in gas treatment
5. A solvent or supercritical carbon dioxide can be used to extract the adsor-

bate from the carbon. These methods are currently experimental.
418
0.01 L:--:----L.---.-..JL.-~:...L_L.L.J_,l;:__:__---L-...l....-...l....-L....l.....l....L~::;____----'l...----'-----'--...I.......J.-'--'-~
10-3 10-2 10-1
Fractional Saturation
Figure 8-15. Equilibrium isothenns at ambient temperature for water

adsorption from organic liquids on 4A molecular sieve. Dotted
line is the water vapor isothenn at 25°C. Key: 1. Methylene
chloride (1400), 2. Chlorofonn (970), 3. Carbon tetrachloride
(84),4. Freon 12 (150 at 38°C), 5. Freon 22 (1800 at 38°C), 6.
1,2, dichloroethane (1600), 7. Chlorobenzene (300), 8. Isopro-
pyl alcohol, 9. Acetone, 10. Ethylacetate, 11. Pyridine, 12.
Xylene (440), 13. Diethylether (12600), 14. Methyl Alcohol,
and 15. Ethanol. Numbers in parenthesis are solubility of water
in ppm at 2O-25°C. Components not followed with a number
are miscible with water (Basmadjian, 1984). Reprinted with
pennission from A.S. Mujumdar (Ed.), Advances in Drying,
Vol. 3. Copyright 1984, Hemisphere Pub. Co.
419
8.6.3. Processing Liquids using Thermal Regeneration
Adsorption is also commonly used for processing liquid streams. Packed beds
are used for dilute feeds while simulated moving beds (see Chapter 10) can be
used for concentrated systems. Drying of solvents is a major application.
Figure 8-15 shows the equilibrum uptake of water on 4A zeolite molecular
sieves for a variety of solvents (Basmadjian, 1984). Liquid feed is usually
introduced with upwards flow to avoid entrapping gas (Keller, 1982). These
systems are regenerated by first draining the column, and then using hot gas
with downwards flow to heat the adsorbent. Before the next feed step the bed
is normally cooled with gas. Because of vaporization of residual solvent in the
bed, energy requirements are relatively high; however, water concentration in
the solvents are usually low. Thus the cost per kg of solvent treated is usually
modest, and is often cheaper than drying by distillation.
A second application for liquid adsorption is the treatment of drinking
water and waste water with activated carbon (For example, see Faust and Aly,
1987). A grade of activated carbon different from that used for gases is com-
monly used. Parts per million of contaminants can be removed. Since the feed
concentrations are low and the equilibrium capacity is high, a very slow mov-
ing shock wave results. Thus very long adsorption cycles (several months) can
be used. Unfortunately, the mass transfer rates are often very low and the M1Z
can be several feet long. Thus long columns or columns-in-series are required.
The columns in series idea is illustrated in Figure 8-16. In Figure 8-16a the
first column is almost saturated and much of the M1Z is in column 2 which is
fresh. Column 3 is unloaded and freshly regenerated carbon is added. In step
B the first column has been saturated so the carbon can be removed and fresh
carbon can be added. Column 2 is now the lead column while column 3 does
the final cleanup. This columns-in-series approach is commonly used when-
ever the M1Z is long. Stirred tanks are also used (see Chapter 10).
Activated carbon is commonly used for treating drinking water in homes
and in bottling plants. In addition to removing traces of impurities, the carbon
will remove traces of chlorine. The chlorine does not adsorb but reacts with
the carbon,
(8-54)
420
a b
D
Feed Feed
Empty
and
add
I new
carbon
2 2
Treated Water
EmptY[j
and
add 3 3
new
carbon
Treated Water
Figure 8-16. Use of columns in series for waste water treatment with
activated carbon.
Note that the product water will be slightly acidic. This is not normally a prob-
lem in a bottling plant. In bottling plants fairly pure water is treated and the
carbon will last from six months to two years.
In home treatment systems the carbon bed is usually under-the-sink or at

the faucet (Anon, 1983). The major problem with home units is homeowners
have no way of telling when they are exhausted and may leave the unit on-
stream too long (see Problem 8-Bl).
Another major use of activated carbon is waste water treatment. The

compounds removed from waste water are usually in very low concentrations
and may have relatively high molecular weights. Examples are chlorinated
hydrocarbons, phenol derivatives, and humic acids. It is usually not economi-
421
cal to try and recover these impurities. The usual method of regeneration is to
unload the adsorber and send the spent carbon to a kiln (Faust and Aly, 1987;
Hutchins, 1979; Perrich, 1981). In the kiln the adsorbed material is burned off
the carbon. Naturally, some of the carbon also burns and this carbon must be
replenished. Design of systems for wastewater treatment is difficult There are
usually a large number of compounds present, and some of them may be
unknown. Also, the feed concentration usually fluctuates considerably.
Current design procedure is to use laboratory or pilot plant studies to determine
the M1Z under the loading conditions. The M1Z or LUB approach is then
used for design. Regeneration conditions are also determined empirically.
8.6.4. Pressure Swing and Vacuum Swing Adsorption
Pressure swing adsorption (PSA) regenerates the column by dropping the pres-
sure and using a portion of the pure product gas as a low pressure purge gas.
Vacuum swing adsorption (VSA) regenerates the column at vacuum pressures
where the amount adsorbed is quite low. A purge step is often not used in
VSA. PSA is illustrated in Figure 8-17a for the simple two-column system
known as the Skarstrom cycle (Skarstrom, 1959). When the right column is
close to breakthrough, the feed is switched to the left column which has been
regenerated. To achieve complete regeneration the volumetric purge-to-feed
ratio, 'Y, must be greater than 1.0. The purge to feed ratio uses volumes calcu-
lated at the conditions of each column.
VPurgc (8-55)
'Y = - - > 1.0 , usually 1.50 > 'Y> 1.05
V Fced
When 'Y> 1, a larger volume of gas sweeps through the column during the
purge step than during the feed step, and the column is regenerated. The
expansion of gas from high pressure, PH, to the low pressure, PL, allows the
designer to have 'Y> 1.0 and still produce product If the ideal gas law is
obeyed the ratio of moles of purge to moles of feed is,
f1,urge PL Vpwgc PL (8-56)

nfced = PH Vfced ='Y PH
Note that the purge can be considered a type of reflux.
422
High Pressure
a Product
~~
v ......
t
Desorb PL PH Adsorb
~ t
Low Pressure Feed
Waste
b Repressurize Feed Slowdown Purge
t
F F
Waste at P L
Figure 8-17. Pressure swing adsorption system. a. Two column system. b.

Steps in simple PSA cycle.
Regeneration occurs because of the movement of the solute wave during

the purge step. This movement can easily be calculated from Eq. (6-23) or (6-
24) for linear systems. The major reason for the movement is the sweeping of
a large volume of expanded clean purge gas (high v in Eqs. (6-23) and (6-24»
through the bed. A secondary effect is the reduction of partial pressure when
total pressure is reduced (see Figure 8-14).
423
Each column in Figure 8-17a goes through the four steps shown in Figure
8-17b.
1. Repressurization ':.i( compression. Raise pressure from PL to PH. Use

either feed gas or pure product gas. Use of product gas is advantageous
since it provides more reflux.
2. Feed. Adsorb at PH.
3. Decompression or blowdown. Drop pressure from PH to PL. In the sim-

ple Skarstrom cycle blowdown is done counter-flow to the feed. A co-
flow decompression step can also be used.
4. Purge. Use product gas at PL to regenerate the column in a counter-flow

direction.
The solute movement theory can be used to describe the feed and purge steps
which are at constant pressure. The compression and decompression steps
require solution of the partial differential equations governing the system.
(Chan et ai, 1981; Ruthven, 1984; Yang, 1987). The theory for linear isoth-
erms shows that the solute wave velocity for the feed and purge steps is given
by Eq. (6-24). The waves also move during the repressurization and depressur-
ization steps. The results then look like Figure 8-18. The dotted lines in Figure
8-18 indicate that the exact path is not known when the pressure varies, but the
ends of the step are known.
Purge Gas
Y=0
L~________ t~____ ~~~______________
t ~
2 !l
3 5
7 6
O~------~.--------~L-------~l----~~
T '-YF
Feed YProd
Time
Figure 8-18. Solute movement theory for PSA. Numbers refer to Example
8-4.
424
The change in locations of the solute waves during compression and
decompression for both PSA and VSA can be determined from,
Za1ier = ( Pafter ) -jJ A (8-57)

Zbefore Pbefore
where the axial distance Z must be measured from the closed end of the column
and
Ec + (I - Ec)Ep + (1 - Ec)(1 - Ep)PsAincrt (8-58a)

PA = Ec + (l - Ec)ep + (l - Ec)(l - Ep)PsAA
where the equilibrium is q = Ac. If equilibrium is of the form q = Ap where p

is the partial pressure, then
PA = Ec + (l-Ec)Ep + (1-Ec)(l-f-p)Ps Amcrt P~Pf (8-58b)

Ec + (I-Ec)Ep + (I-Ec)(l-E)Ps AAPtotiPf
Pi
where is the molar density of the gas. If the inert gas does not adsorb, Amcrt
= O. The term PA is the ratio of total amount of inert gas which could be stored
in the column to the total amount of solute A which could be stored. Compres-
sion and decompression steps also change the mole fraction of solute in the gas.
This change is,
YA,after = [ Pa1ier J~A - 1 (8-59)

YA,before Pbefore
The use of these equations and the characteristic diagram is ilustrated in Exam-
ple 8-4.
In practice the waves are spread by dispersion and slow mass transfer. In
usual operating procedures the column is not completely regenerated and only
part of the equilibrium capacity can be used. This is illustrated in Figure 8-19.
Keeping the mass transfer zone inside the column will make the waste gas
more concentrated, and less purge gas is used. The working capacity of the bed
425
Load
L
z
Figure 8-19. Working capacity in PSA.
is represented by the shaded region in Figure 8-19. The working capacity may
be only a few percent of the total capacity of the bed. Very steep, highly favor-
able isotherms will make the adsorption wave very sharp, but the desorption
wave will be very diffuse and spread significantly. The opposite is true for
unfavorable isotherms. The optimum isotherm shape will be close to linear.
Of course, in most cases the engineer must live with the adsorbents which are
readily available.
Pressure swing adsorption systems are used commercially for a variety of

gas separations. These include drying air and natural gas, hydrogen
purification, separating straight chain hydrocarbons, and purifying nitrogen and
oxygen from air. Zeolite molecular sieves, activated alumina, and silica gel are
used for drying. Activated carbon or zeolite molecular sieves are used for
hydrogen purification while various zeolites are used for the other separations.
A wide variety of PSA and VSA processes are used and are reviewed in
detail elsewhere (Ruthven, 1984; Tondeur and Wankat, 1985; Wankat, 1986;
Yang, 1987). One and two column systems (Figure 8-17) are commonly used
for small-scale applications such as laboratory driers. In larger installations it
is very important to have the waste gas as concentrated as possible to avoid
loosing product in the waste stream. Multibed systems are used to include
pressure equalization steps to recover product gas. A variety of more complex
systems using a variety of steps have been developed. In addition to or instead
426
of the 1. repressurization. 2. feed. 3. blowdown, and 4. purge steps, the follow-

ing steps have been used.
5. Pressure equalization involves using the co-current blowdown gas from

one column to partially repressurize another column. This step increases
the recovery of product, but leads to more complex cycles (Ruthven.
1984; Wankat. 1986; Yang. 1987).
6. Vacuum regeneration leads to VSA cycles. VSA would often consist of

steps I, 2. 3 and 6. VSA has the advantages of lower energy use and
recovery of a purer adsorbate product The disadvantages are larger
columns with longer operating cycles. Vacuum regeneration is used to
obtain both oxygen and nitrogen from air where it has lower operating
costs than PSA, but higher capital costs.
7. A high-pressure rinse step using a reflux of adsorbate can be useful to

remove nonadsorbed gas to produce a purer adsorbate product. This step
would normally be done in the same direction as the feed step.
8. Some rapid cycle processes require a delay step to allow time for mass
transfer.
The designer has a huge number of permutations and combinations of these

eight steps to choose from. This allows the designer to use his or her creativity
in developing PSA processes.
PSA systems typically operate with gas velocities in the range from 0.01
to 0.5 m/sec. The typical cycle lasts for a few minutes. The adsorber heats up
during the feed step and then cools during desorption. For air drying the typi-
cal temperature swing is from 2 to 4°C. For more concentrated feeds the tem-
perature swing can be greater than 50°C. Since the cycles are rapid, part or all
of the temperature wave stays in the column. Thus part of the energy released
during adsorption is stored in the column and is available for desorption. In
concentrated systems such as adsorption of nitrogen from air to produce oxy-
gen, the thermal wave will pass out of the column during the feed step, but will
427
100
90
80
70
60
50
u. 40
!..
011 30
'-
::l
"Iii 20
8.
E
10
....
011 0
::l
C)
-10
-20
-30
-40
-50
-60
-70L----'-_..L---'-_-'---_'----'-_..L---'-----'
o 0·5 1·0 2{) 3{) 4·0 5·0 6{) 7·0 S'O
Bed depth (feet from support screen)
Figure 8-20. Temperature profiles in an industrial-scale zeolite bed for PSA

air separation. (A) feed air at 30 and 40 oF; (B) feed preheated
to 1OO°F; (C) with heating elements inserted. in the beds near
the inlet (Collins, 1977).
stay in the column during purge. This occurs because of the reduced gas den-
sity during purge. The result is a steady state bed temperature well below
ambient. This is illustrated in Figure 8-20 (Collins, 1977).
Example 8-4. PSA solute movement analysis
We wish to remove acetylene from hydrogen gas by pressure swing

adsorption. The feed gas is 0.005 mole fraction acetylene. Superficial
velocity is 45 em/sec during the feed step. The column is one meter
long. The high pressure is 10 atm while the low pressure is 1 atm. Pick
428
a purge to feed ratio of y = 1.1. The cycle is symmetric since a typical 2

column PSA system is used. The cycle is as follows.
Repressurize with feed o sec t02 sec

Feed step at PH 2 sec to 60 sec
Blowdown 60 sec to 62 sec
Purge atPL 62 sec to 120 sec
Properties of carbon: PI = 2.1 glcc, Kc! = 1.0

Ep = 0.336, Ee = 0.43
Equilibrium for acetylene at low partial pressures and 150°F is q = 35.6

c where q is gmoles/g adsorbent and c is gmoles/cc fluid. Hydrogen
does not adsorb. Use the solute movement theory to predict the outlet
mole fraction of acetylene during the purge step (62 sec to 120 sec).
What is the minimum column length to have a pure H2 product?
Solution
A. Define. The four step cycle was shown in Figure 8-17b. We wish
to find the outlet concentration profile during the purge step.
B. Explore. The data required for the solute movement theory is

presented in the problem statement We need to trace the movement of
solute during the four steps as shown in Figure 8-18. Once the effect of
pressurization and depressurization on the acetylene mole fractions has
been determined, the outlet concentration profile is easily plotted.
C. Plan. Calculate ~A from Eq. (8-58a). Then the movement of the

solute wave during repressurization is found from Eq. (8-57). During
the feed step, solute velocity is found from Eq. (6-24) and the net move-
ment of the wave is easily determined. Two solute waves move as
shown in Figure 8-18. During blowdown Eq. (8-57) is used for both
waves. Solute waves during purge are again calculated using Eq. (6-
429
24), but velocity must be corrected for the increased purge-to-feed ratio.
Finally, Eq. (8-59) is used to calculate the mole fractions.
D. Do It.
0.43+(0.57)(0.336)+0
= 0.43+(0.57)(0.336}+(0.57)(0.664)(0.21)(35.6) = 0.0215
The value of ~A is small since AA is large.
During repressurization Eq. (8-57) is used with Z measured from the

closed end of the column. Thus zbcfore = 100, and
Zafter== 1OO( 10) ...M215 =9 5.2
which is 4.8 em from the open end. This is point 1 on Figure 8-18.
Feed step. Interstitial velocity,

v = Vsuper/Ee =45/0.43 = 104.65 em/s. From Eq. (6-24),
V
UA = ----------
l-Ee l-E
1+e:-Ep+~(1-Ep)PsAA
= 104.65 == 1.556 em/s
1+ [ ~:~~ ] (0.336}+ [ ~:~~ ](0.664)(2.1)(35.6)

During the 58 seconds of the feed step the acetylene travels
(uA)(58) = 90.24 em
Net distance for wave starting at t = 0 is 90.24 + 4.8 = 95.04 em. This
430
is the minimum column length. Wave starting at t = 2 sec travels to

90.24 cm. These are points 2 and 3 in Figure 8-18.
Blowdown step. Use Eq. (8-57) again measuring z from the closed end
of the column. Then to determine point 4 in Figure 8-18,
(z after )4 = 4 •% (_1_)-{).215
10 = 5 .21 cm
which is 94.79 cm from the feed end of column. Calculation of point 5

is similar.
which is 89.68 cm from feed end.
Purge Step. During the purge step the interstitial velocity increases
since y = 1.1. The new interstitial velocity is,
vpurge = YVfeed = (1.1)(104.65) = 115.1 cm/sec
The new solute velocity will be
UApurge =YllAfeed = (1.1)(1.556) = 1.712 cm/sec

The breakthrough times (points 6 and 7 on Figure 8-18) are
~ = ~ = 94.789 = 55.35 sec

UA 1.712
Total time ~ is 55.35 + 62 sec = 117.37 sec. For point 7,
t = 89.68
7 1.712
=52.38 sec
Total time t7 is 52.38 + 62 = 114.38 sec

431
Concentrations. From 62 to 114.38 sec acetylene has undergone one

pressure change from 10 to 1 aunosphere. Since the initial mole frac-
tion was the feed value, Eq. (8-59) gives
YProd = YA,after = YF( 110 )-0.9785 = (0.005)(9.517) = 0.0476
From 114.38 sec to 117.37 sec the solute wave was first pressurized and
then depressurized. These results approximately cancel and
YProd - YF = 0.005. This result is not exact because the exact nature of
the repressurization and blow down steps is not known. From 117.37 to
120 sec pure H2 exits and YProd = O.
E. Check. Since the exact behavior during repressurization and blow-

down are not known, an exact mass balance check is not possible.
F. Generalization. This equilibrium analysis is not complete since the

details of the blowdown and repressurization steps are not known. To
study these steps in detail would require a detailed study of the gas flow
and pressure in the column. The equilibrium analysis also does not
include mass transfer and dispersion effects. More complex models are
discussed by Ruthven (1984) and Yang (1987). Instead of completely
cleaning the bed, we would expect a concentration profile similar to that
shown in Figure 8-19. Qualitatively, the results are typical for PSA of
dilute gases. The basic equations can be used to study a variety of PSA
and VSA cycles.
8.6.5. Regeneration with Purge and Desorbent
In cases where thennal or pressure swing adsorption cycles are inadequate a

purge stream (usually not adsorbed) or a desorbent stream (which is a strongly
adsorbed component) is used for the regeneration step. This type of operation
is common in chromatography (Chapter 7), in ion exchange (Chapter 9), and in
simulated moving bed systems (Chapter 10).
432
A common application with gas systems is separating medium molecular

weight (C lO to C20 ) straight chain hydrocarbons from branched and cyclic
hydrocarbons. The straight chain hydrocarbons can diffuse into 5A zeolite
molecular sieves while the branched and cyclic molecules are too large to fit
into the cages. Thus the adsorption part of the cycle is simple. Feed is intro-
duced until the straight chain hydrocarbons start to breakthrough. Desorption
is more difficult. Once inside the zeolite the linear paraffins are strongly
adsorbed even above 350°C. Higher temperatures cannot be used since ther-
mal decomposition becomes excessive. Pressure and vacuum swing operations
are useful for the lower molecular weight paraffins « C lO ), but do not work
well for the ClO to C20 range. The problem with PSA and VSA systems is it is
difficult to provide the energy required for desorption during the low pressure
purge step.
A variety of commercial purge gas and desorbent processes with dif-

ferent desorption steps have been developed (see Keller, 1982; Ruthven, 1984;
Wankat, 1986; Yang, 1987). A general flow sheet is shown in Figure 8-21. In
the adsorption step the linear molecules are adsorbed while other molecules are
not. In the displace step the non-adsorbed material contained in the external
pores is pushed out of the column with desorbent. This step is short. The
desorption step is done counter-flow to the feed step. Different commercial
processes differ mainly in the desorbent used. For low molecular weight n-
paraffins a non-adsorbed gas such as N2 or H2 at 600 to 800°F is used. This
operation is known as purge gas stripping. The mixture of adsorbate and
Non-Ads+D
Products
Adsorb Displace Desorb Separation
( condensation
or distillation)
Figure 8-21. Apparatus for desorption by purge gas stripping or displace-
ment adsorption.
433
desorbent must then be separated. The separation shown in Figure 8-21 would
be done by partial condensation in a device similar to flash distillation. For the
harder to desorb C 10 to C20 hydrocarbons a desorbent which is adsorbed is used
(this is called displacement adsorption). Desorbents have included n-pentane,
n-hexane, ammonia and alkylamines. The separation scheme required depends
on the desorbent used but is often distillation. Because an additional separation
scheme is required, purge gas and desorbent systems are more complicated
than PSA or thermal swing. Thus purge gas and desorbent systems are used
only when necessary. Distillation will probably be preferred if the relative
volatility is greater than 1.2 to 1.3 (Keller, 1982).
In liquid systems a liquid desorbent can be used to extract the adsorbate

from the solid. Usually, a solvent which does not adsorb, but has a high affinity
for the adsorbate is used. Polymeric resins have been used to recover organics
from water (Fox 1979; Fox and Kennedy, 1985; Faust and Aly, 1987) with
desorbent regeneration. A variety of flow sheets have been developed for this.
One for phenol recovery is shown in Figure 8-22a. Either acetone or methanol
is used as a desorbent. The adsorption steps shown in Figure 8-22a consist of
adsorption of the waste water feed, then adsorption of water pushed out of
column during the start of the desorption, and then superloading which is dis-
cussed next. The column is desorbed with a solvent such as acetone which
essentially pushes the water out in plug flow. Once the acetone and phenol
start to breakthrough, the column effluent is sent to a distillation column for
separation. The column is reconditioned for the next cycle with a water wash
which displaces the weakly adsorbed acetone. The effluent from this step is
also sent to the distillation.
The bottoms from the distillation column is an heterogeneous azeotrope

which splits into two phases. The phenol phase contains a small amount of
water. If necessary, this phenol can be dried with a desiccant. The water phase
contains about 10% phenol. What can be done with this very contaminated
stream?
The clever answer is to superload the adsorption column (Fox, 1979; Fox
and Kennedy, 1985). This concept can be understood from the isotherm shown
in Figure 8-22b. The resin near the feed inlet is saturated at concentration CF.
434
a
Waste Water wash
Desorbent
recycle
Treated
Phenol
10% Phenol
Separator Distil
Adsorb Desorb and
wash
Figure 8-22. System for phenol recovery using acetone as desorbent. a. Pro-
cess. b. Isotherm illustrating superloading.
However, since the concentration of the superload stream Cs is much greater

than CF, the resin has considerable capacity remaining to adsorb this stream.
This superload capacity is shown in Figure 8-22b. The solute movement theory
can be adapted to study the system shown in Figure 8-22a. This is left as Prob-
lem 8-C8.
The desorption and purge cycles can produce both a pure non-adsorbed
product and a pure strongly adsorbed product This is a major advantage of
these cycles. The major disadvantage of these cycles is they are complex. The
processes shown in Figures 8-21 and 8-22a are considerably more complex
than the other cycles shown in this chapter. Additional separation devices such
as distillation are required.
435
8.6.6. Systems Without Regeneration
Adsorbents are used in a variety of applications where they are discarded

instead of being regenerated. This is economical for some consumer applica-
tions, hazardous material adsorption, medical use, and for treating very dilute
fluids.
Consumer applications include foods, building materials and water treat-

ment. Very small canisters of a mixture of activated carbon and silica gel are
included in some food products such as instant coffees. The carbon will adsorb
any objectional organic vapors while the silica gel adsorbs water vapor. Pack-
ets of silica gel or zeolites are included with some food products and some con-
sumer goods such as electronics to adsorb moisture. Packets of zeolites are
also enclosed in some thennal pane windows to adsorb moisture to prevent fog-
ging. Activated carbon canisters for home water treatment are usually dis-
carded when breakthrough occurs (Anon, 1983).
Adsorbents have been used in the nuclear industry. In these applications

the spent adsorbent is often encased in concrete or glass and is then buried.
One example of current interest is the adsorption of radon gas by activated car-
bon. This can be used both for cleanup and for analysis of radon concentra-
tions in buildings. Venniculite has be used for adsorption of radioactive cal-
cium and strontium from waste water (Wankat, 1986).
Gas masks are another major application of adsorbents where because of

the hazardous nature of the adsorbates the adsorbent is not regenerated.
Instead, a new canister is used to replace the old canister. Since it is desireable
to adsorb a variety of noxious vapors, gas masks are usually designed with
series of adsorbents and reactants (e.g. see Mantell, 1950). Gas mask design is
obviously an area for experts.
Activated charcoal has a long history of use in medical applications.
Two main uses are for treatment of poisoning and digestive tablets. Carbon is
also used in animal food to adsorb disease-causing organisms and toxins. Med-
ical applications are discussed in detail by Cooney (1980).
For adsorption of very dilute gases or liquids the use of non-regenerated

adsorbents is often attractive even on a large-scale. The non-regenerated sys-
436
VAPOR LADEN
AIR
PREFILTER
~AANULAA
ADSORBENT
, FINAL FILTER
~.:,:" (BALSlON GRADE BO)
l·j':" I /\.
~ ,fr'
,., JI/I
.' /
/
POLYPROPYLENE
END CAP
BUNA-N
SEAL
Figure 8-23. Canister adsorption system used for removal of trace contam-
inants from air. Balston (1986). Reprinted with pennission of
Balston, Inc., 703 Massachusetts Ave., Lexington, MA 02173.
Copyright 1986, Balston, Inc.
tern is less expensive since the regeneration facilities and extra plumbing are
not needed. When on-stream life is one or two years, regeneration facilities are
used so seldom that they may not pay for themselves. Non-regenerated adsorp-
tion systems are used for removal of trace contaminants from both air and
water, and for removal of traces of solvent with activated carbon. Guard
columns for removal of heavy organics are often designed to have the adsor-
bent discarded and replaced with fresh adsorbent. As a rule-of-thumb non-
regenerated systems should be considered for concentrations less than 2 ppm.
The adsorbent in a non-regenerated system is often held in a canister or car-
tridge as shown in Figure 8-23 (Balston, 1986). Additional canisters in parallel
437
are used for higher flow rates. Canisters are commercially available prepacked
with a variety of adsorbents.
8.7. DESIGN CONSIDERATIONS
The packed column system was shown schematically in Figures 6-1 and 8-6.
The support and holddown plates need to be sized to prevent movement of the
adsorbent. Movement of the adsorbent is likely to cause excessive attrition and
adsorbent loss. The support is usually a screen, net or frit Care should be
taken to prevent clogging of the support since this will cause excessive pressure
drops.
The distributor should add the feed mixture evenly across the packing.
This helps prevent channeling and uneven velocities which will reduce separa-
tions. An even withdrawal at the bottom of the column is also desired. The
effects of the volume of the piping, distributors etc. was analyzed in Section
8.4.
Both upwards and downwards flow are used. Upward flow may fluidize
the particles in the bed causing attrition. Fluidization will not occur if (Ledoux,
1948),
G2
---<0.0167
(8-60)
Pf Ph dp g
where G is the mass velocity of the gas in and g is the acceleration due to grav-
ity. Fluidization can also be prevented with a holddown plate. The advantage
of upward flow is channeling is probably less of a problem particularly for
liquids. Downward flow is useful when a liquid must be drained from the
column prior to regeneration.
Pressure drop and particle diameter effects were considered in Section

8.4. Pressure drop is a major consideration particularly in gas systems. Opera-
tion of large blowers and compressors is expensive and a maximum pressure
drop is often set Usual design methods have used fairly large particles to keep
.1p modest (see Eq. (8-22)). However, small diameter particles have shorter
mass transfer zones. When pore diffusion controls, the advantages of both
438
modest pressure drop and short mass transfer zones can be obtained by keeping
L/d~ constant. This means short colwnns packed with small diameter particles
with rapid cycling should be used. This is not yet a standard design procedure.
Standard design uses particles in the range of 0.1 to 10.0 mm with fairly long
colwnns (a few meters) and long cycles (minutes in PSA to days for thermal
cycles).
If Ap, <lp and L are set, then vsuper can be determined from Eq. (8-22).
This allows determination of the appropriate column diameter since vsuper can
also be calculated from the required feed rate, F kg/hr.
F (8-61)
v =--
super Ac Pf
This equation allows calculation of Ac and hence diameter.
Diameter = -V ~ v F P =
super f
4 F ilL
7t Pf (Ap)specK d~
(8-62)
Note that feed rate F affects the diameter, but has little effect on column length
unless a rule of thumb is used to set the ratio L/Diameter. Usually, two or more
colwnns are used in parallel so that feed can be continuously processed while
one column is desorbed.
At the end of this chapter you should be able to satisfy the following objectives.
1. Describe the following:
a. Thermal adsorption systems
b. PSA
c. Purge gas stripping
d. Desorbent Operation
439
2. Use the linear solute movement theory to explain adsorption separation

techniques including:
a Trace contaminant removal
b. PSA
c. Purge gas stripping
3. Use the nolinear solute movement theory to explain adsorption separation

a. Drying gases
b. Activated carbon solvent recovery
c. Activated carbon wastewater treatment
d. Superloading
4. Use the M'IZ-LUB approach to design constant pattern problems.
5. Derive and use constant pattern solutions.
6. Solve displacement chromatography problems using local equilibrium

theory with coupled isotherms.
7. Discuss qualitatively:
a. the optimization of fixed bed systems.
b. the effect of coupled adsorbates.
c. the effect of heat of adsorption.
d. the use of adsorbents in non-regenerated systems.
8. Calculate column diameter and pressure drop, and determine the dimen-
sion of systems which will give the same separation, throughput and pres-
sure drop.
440
REFERENCES
Anon, "Waterfilters," Consumer Reports. 48 (2), 68 (Feb. 1983).
Aris, R., and N.R. Amundson, Mathematical Methods in Chemical Engineer-

ing. Vol. 2, First Order Partial Differential Equations with Applications. Pren-
tice Hall, Englewood Cliffs, NJ, 1973.
Balston, Inc., Filter Products, Bulletin P-11OB, Lexington, MA, 1986.
Basmadjian, D., "The adsorption drying of gases and liquids," in Advances in

Drying. Vol. 3, Mujumdar, A.S., (Ed.), Hemisphere Pub. Co., Washington,
DC, 1984, Chapt. 8.
Basmadjian, D., K.D. Ha, and C.-Y. Pan, "Nonisothennal desorption by gas
purge of single solutes in fixed-bed adsorbers. 1. Equilibrium theory," Ind. Eng.
Chem .• Process Des. Develop .. 14, 328 (1975a).
Basmadjian, D., K.D. Ha, and D.P. Proulx, "Nonisothennal desorption by gas
purge of single solutes from fixed bed adsorbers. II. Experimental verification
of equilibrium theory," Ind. Eng. Chem. Process Des Devlop,. 14. 340 (1975b).
Bird, R.B., W.E. Stewart, and E.N. Lightfoot, Transport Phenomena. Wiley,
~,1960,p. 196-200.
Chan, Y.N., F.B. Hill, and Y.W. Wong, "Equilibrium theory of a pressure
swing adsorption process," Chem. Eng. Sci., 36. 243 (1981).
Collins, J.J., U.S. Patent, 4026, 680 (1977).
Cooney, D.O., Activated Charcoal: Antidotal and Other Medical Uses, Marcel
441
Cooney, D.O. and E.N. Lightfoot, "Existence of asymptotic solutions to fixed-

bed separations and exchange equations," Ind. Eng. Chem. Fundam., 4, 233
(1965).
Faust, S.D. and O.M. Aly, Adsorption Processes for Water Treatment, Butter-
worths, Boston, 1987.
Fox, CR., "Removing toxic organics from waste water," Chem. Eng. Prog. 75,
(8) 70 (1979).
Fox, C.R. and D.C. Kennedy, "Conceptual design of adsorption systems," in

Adsorption Technology, Slejko, F.L., (Ed.), Marcel Dekker, NY, 1985,91-165.
Fulker, R.D., "Adsorption" in G. Nonhebel (Ed.), Processes for Air Pollution

Control, 2nd. ed., CRC Press, Boca Raton, FL, 1972, Chapt. 9.
Garg, D.R. and J.P. Ausikaitis, "Molecular sieve dehydration cycle for high
water content streams," Chem. Eng. Prog. 79 (4),60 (1983).
Glueckauf, E., "Contributions to theory of chromatography," Proc. Roy. Soc.

London, A186, 35 (1946).
Helferrich, F. and G. Klein, Multicomponent Chromatography, Marcel Dekker,

NY,1970.
Hutchins, R.A., "Activated-carbon systems for separation of liquids," in P.A.

Schweitzer (Ed.), Handbook of Separation Techniquesfor Chemical Engineers,
McGraw-Hill, New York, 1979, Section 1.13.
Joshi, S. and J.R. Fair, "Adsorptive drying of toluene," Ind. Eng. Chem. Rsch,
27,2078 (1988).
Keller, G.E. II, Adsorption, Gas Absorption and Liquid-Liquid Extraction:

Selecting a Process and Conserving Energy, (Manual 9 in series on Industrial
Energy-Conservation, E.P. Gyftopoulos, Editor-in-Chief), MIT Press, Cam-
bridge, MA 1982.
442
Kohl, A. and F. Riesenfeld, Gas Purification, 3rd, ed., Gulf Pub. Co., Houston,
1979, Chapter 12.
Ladisch, M.R., M. Voloch, J. Hong, P. Bienkowski, and G.T. Tsao, "Cornmeal

adsorber for dehydrating ethanol vapors," Ind. Eng. Chern. Process Des.
Develop., 23, 437 (1984).
Lapidus, L. and J.B. Rosen, "Experimental investigations of ion exchange

mechanisms in fixed beds by means of an asymptotic solution," Chern. Eng.
Prog. Symp. Ser. 50, (14),97 (1954).
Leavitt, F.W., "Non-isothermal adsorption in large fixed beds," Chern. Eng.

Prog., 58 (8),54, (1962).
Ledoux, E., "Avoiding destructive velocity through adsorbent beds," Chern.

Eng., 118 (March 1948).
Lee, M.N.Y., "Novel separations with molecular sieves adsorption," in N.N. Li

(Ed.), Recent Developments in Separation Science, Vol. I, CRC Press, Boca
Raton, FL, 1972,75-112.
Liapis, A.T. and D.W.T. Rippen, "The simulation of binary adsorption in con-
tinuous countercurrent operation and a comparison with other operating
modes," AIChE Journal, 25, 455 (1979).
Lukchis, G.M., "Adsorption systems. Part I: Design by mass transfer-zone con-

cept," Chern. Eng., 80 (13), Ill, (June 11, 1973).
Mantell, C.L., "Adsorption," in J.H. Perry (Ed.), Chemical Engineer's Hand-

book, 3rd ed., McGraw-Hill, NY, 1950, Chapt. 14.
Marchello, J.M., Control of Air Pollution Sources, Marcel Dekker, NY, 1976,
232-250.
443
Marrero, T.R and E.A. Mason, "Gaseous diffusion coefficients," J. Phys.

Chem. Ref Data, 1, 3-118 (1972).
Michaels, A.S., "Simplified method of interpreting kinetic data in fixed bed ion
exchange," Ind. Eng. Chem., 44, 1922 (1952).
Pan, C.-Y., and D. Basmadjian, "An analysis of adiabatic sorption of single

solutes in fixed beds, equilibrium theory," Chem. Eng. Sci., 2645 (1971).
Perrich, J.E. (Ed.), Activated Carbon Adsorption for Wastewater Treatment,

CRC Press, Boca Raton, FL, 1981.
Perry, RH., and D.W. Green, (Eds.) Perry's Chemical Engineers' Handbook,
6th ed., McGraw-Hill, New York, 1984.
Phillips, M.W., G. Subramanian, and S.M. Cramer, "Theoretical optimization

of operating parameters in non-ideal displacement chromatography," J.
Chromatogr., 454, 1 (1988).
Reid, RC., J.M. Prausnitz, and T.K. Sherwood, The Properties of Gases and
Liquids, 3rd ed, McGraw-Hill, New York, 1977.
Rhee, H.-K., "Equilibrium theory of multicomponent chromatography," in Per-

colation Processes, Theory and Applications, A.E. Rodrigues and D. Tondeur
(Eds.), Sijthoff and Noordhoff, Alphen aan den Rijn, The Netherlands, 1981,
285-328.
Rhee, H.-K. and N.R Amundson, "Analysis of multicomponent separation by

displacement development," AIChE Journal, 28, 423 (1982).
Roz, B., R Bonmati, G. Hagenbach, P. Valentin, and G. Guiochon, "Practical

operation of prep-scale gas chromatogaphic units," J. Chromatog. Sci., 14, 367,
(1976).
444
Ruthven, D.M., Principles of Adsorption & Adsorption Processes. John Wiley

& Sons, New York, 1984.
Sherwood, T.K., R.L. Pigford, and C.R. Wilke, Mass Transfer, McGraw-Hill,
New York, 1975.
Sircar, S. and R. Kumar, "Adiabatic adsorption of bulk binary gas mixtures:

analysis by constant pattern model," Ind. Eng. Chern. Process Des. Dev., 22,
271 (1983).
Skarstrom, C.W., "Use of adsorption phenomena in automatic plant-type gas

analyzers," Annals. New York Acad. Sci., 72 (13) 751 (1959).
Sweed, N.H., "Nonisothermal and nonequilibrium fixed bed sorption," in Per-

colation Processes, Theory and Applications. in A.B. Rodrigues and D. Ton-
deur, (Eds.), Sijthoff and Noordhoff, Alphen aan den Rijn, The Netherlands,
1981,329-362.
Turk, A., "Source control by gas-solid adsorption and related processes," in

A.c. Stem (Ed.), Air Pollution, Vol. 3, 2nd ed., Academic Press, New York
1968,497-519.
Tondeur, D. and P.C. Wankat, "Gas purification by pressure-swing adsorption,"

Separ. Purif. Methods. 14. 157 (1985).
Vatavuk, W.M. and R.B. Neveril, "Pan XIV: Costs of carbon adsorbers,"
Chern. Eng .• 131 (Jan. 24,1983).
Wakao, N., and T. Funazkri, "Effect of fluid dispersion coefficients on

particle-to-fluid mass transfer coefficients in packed beds," Chern. Eng. Sci .• 33,
1375 (1978).
Wankat, P.C., "Cyclic separation techniques," in A.B. Rodrigues, and D. Ton-

deur (Eds.) Percolation Processes, Theory and Application. Sijthoff and
Noordhoff, Alphen aan den Rijn, Netherlands, 1981,443-516.
445
Wankat. P.C., Large Scale Adsorption and Chromatography. CRC Press, Boca
Raton, FL, 1986.
Wankat. P.C., "Intensification of sorption processes," IEC Rsch. 26. 1579

(1987).
Wankat. P.C. and Y.M. Koo, "Scaling rules for isocratic elution chromatogra-
phy," AIChE Journal. 34, 1006 (1988).
Weyde and Wicke, Kolloid z.. 90 156 (1940).
Yang, R.T., Gas Separation by Adsorption Processes. Butterworths, Boston,

1987.
HOMEWORK
AI. Explain how feed can be continuously treated in a batch adsorption pro-
cess by using 2 colwnns. How would 3 colwnns be useful?
A2. Explain how each solute movement line in Figure 8-11 is calculated.
Sketch the concentration profile (concentration in column versus axial
distance) at the end of the feed step and at the end of regeneration.
A3. Sketch the concentration profiles in the columns which are adsorbing in
Figure 8-16. Explain why columns in series allows more complete use
of the bed.
A4. Explain in your own words how PSA works. Why is temperature often
close to constant?
A5. Figures 8-21 and 8-22a show recycle streams. Why might recycle be
useful?
A6. Explain in your own words the formation of constant pattern and pro-
portional pattern behavior. Why must the constant pattern wave move
at a velocity llm when v is constant?
446
A7. Sketch a co-flow regeneration system for drying a gas. Develop the
solute movement diagram for a nonlinear system for co-flow. Compare
this with Figures 8-lO and 8-11, and show that counter-flow regenera-
tion requires less energy.
A8. Explain why a cooling step is often used in drying operations.
A9. The carbon canisters used on automobiles to prevent evaporative losses

of gasoline from the gas tank were developed based on the Skarstrom
process. However, pressure is constant. Explain how these canisters
work.
AlO. Activated carbon adsorbers for solvent recovery can be regenerated

with vacuum instead of steam.
a. Use an equilibrium diagram to explain how this works.
b. Why does this process work best when the feed gas is very concen-
trated >3%?
c. Systems with vacuum drawn co-flow and counter-flow to the feed

have been used commercially. Sketch these two systems discuss
the advantages and disadvantages of each.
d. Sometimes a short purge step with hot gas is added to the vacuum
purge step. What is the advantage of this? Would co-flow or
counter-flow vacuum be advantageous?
These systems are used for treating the gas saturated with gasoline
when storage tanks are filled.
All. Why is it advantageous to adsorb and then desorb a solvent before

incineration instead of just incinerating the feed gas?
A12. Explain how water can condense in a capillary even though the relative
humidity of the air is less than 100%.
AB. a. Explain the function of superloading.
b. The idea of superloading can be generalized. Suppose we have two

streams of different composition, but both contain the same solute.
447
Should the streams be fed separately or mixed together? If fed

separately, which stream should be fed first? Explain your answers.
A14. An adsorbent can be included in a muffler to prolong the life of the

muffler by preventing corrosion. How does this work? How is the
adsorbent regenerated?
A15. Develop your key relations chart for this chapter.
B 1. Activated carbon is commonly used in water filters for home use.
These filters are typically mounted below the sink or on the countertop.
Examples of these systems can be seen at a hardware store, a plumbing
shop or a Sears store. One major difficulty with these units has been the
owner cannot tell when the unit is exhausted (in other words, he can't
tell when breakthrough occurs) except by tasting the water. Thus he or
she is likely to either discard the cartridge containing the carbon too
soon or keep it installed too long. Generate a variety of ways which
could be used to cheaply and easily determine when a new cartridge
should be installed (see Anon, 1983, for details of these units). Could
the homeowner easily regenerate the canister at home?
B2. Your boss has read a copy of this book. He wants to apply the process
intensification ideas of reducing particle diameter to an adsorption sys-
tem for removal of one adsorbate. List all the reasons you can think of
why this might be a bad idea. Briefly (1 sentence) explain each reason
on your list.
B3. Temperature swings in PSA processes can be 50°C or more for concen-
trated feeds. This is detrimental. Brainstorm at least five ways to
reduce these temperature swings.
C. Derivations
Cl. Derive Eq. (8-7a) from Eqs. (8-5) and (8-6).
C2. Derive Eq. (8-8b) from Eq. (8-7b).

448
C3. Derive Eq. (8-9) from Eqs. (6-50) and (8-8b).

C4. Develop the constant pattern solution for ~'tMTZ if the Freudlich isoth-
erm, Eq. (6-14), is used with Eqs. (8-9) and (8-8b).
C5. Derive Eq. (8-52). Hint: Treat the steam as a shock and do an energy
balance on an element Az.
C7. If Uth > lIs in drying a gas, show that a cooling step is not required after
the regeneration step. Why might a very short cooling step still be
desirable before introducing the feed?
C8. Develop the solute movement diagram for superloading. The cycle
steps are: 1. feed, 2. superload, 3. counter-flow solvent, and 4. water
wash. Equilibrium is shown in Figure 18-22b for steps 1 and 2.
Assume solvent is not adsorbed. Assume that adsorption still occurs
when solvent is present, but is approximately 1/4 its value from water.
Note that the colwnn must be long enough to not breakthrough during
both steps 1 and 2.
D. Single Answer Problems.
D1. A 0.6 m long laboratory column operates at a superficial velocity of 15

cm/min. A breakthrough curve is measured. The center of the sym-
metric breakthrough curve is at 57.6 minutes while the width is 8.2
minutes. The laboratory system is designed with minimal extracolumn
volume in connecting tubes, valves, detector, etc. The solute is known
to follow a Langmuir type isotherm. We wish to operate a system with
the same size particles. We want a 90% bed utilization in the unit.
Assume pore diffusion controls.
a. If we operate at the same velocity, how long should the large unit
be? What is ~r ?
b. If we operate at a superficial velocity of 20 cm/m in , how long

should the unit be? What is It,r?
D2. We are adsorbing anthracene from cyclohexane onto activated alumina.
449
For a step input from c = 0 to c = 0.012 gmole/L predict LMfZ·

Superficial velocity Ee v = 25 cm/min. Assume pore diffusion controls.
Particles are 0.5 mm in diameter. Use a one porosity model (Ep 0).=
22c
Data: Ee = 0.4, q = , Pp = 1.47 kgIL, Pr cyclohexane = 0.78
1 + 375c
kgIL, KD = 1.0, Dmp = 8.2 X 10-7 cm 2 /s (this is diffusion inside pores).
q is in gmole/kg and c is in gmoles/L. Use Eq. (8-17a).
D3. We have a 100 cm long column packed with 0.1 mm diameter rigid par-
ticles. The fluid is water with a viscosity of ~ = 1 cpo Ee = 0.4. Find the
superficial velocity for a pressure drop of 100 cm water. Watch your
units. This corresponds to SEC with compressible gels.
D4. We have an adequate design using 1.0 mm diameter particles and 75%
of the bed is used (LILMrZ = 2). We wish to increase fractional bed
utilization to 85% and reduce pressure drop to 0.9 t.pold' Volumetric
flow rate is unchanged. We wish the new length to be 1/2 the old
length. What particle diameter is required? What is the required ratio
of (DnewlDold)? Assume pore diffusion controls.
D5. Your boss has decided that the ~p for Example 8-1 will be too high. He
wants Rp = 0.75. For the same particle size, same throughput, and the
same fractional bed use as Example 8-1, determine DnewlDold, LnewlLbed
and vncw/Vold'
D6. Repeat Example 8-4 with the following changes in the cycle:
Repressurization is done with pure hydrogen product gas
from 0 to 2 sees. (Counterflow to the feed).
Feed Step 2 sec to 63 sec.

Blowdown 63 to 65 sec.
Purge atPL 65 to 126 sec.
Compare the results with Example 8-4.
D7. A PSA system is being used to remove traces of methane from hydro-
gen. The feed is 1.1 % methane and the remainder is H2 • Feed pressure
is 8.9 atm. Column operates at 28.2°C. At this temperature and at low
450
partial pressure equilibrium is linear, q = 0.6 Pm. where q is gmoles

Cf4/kg carbon and Pm. is the partial pressure of Cf4 in atmospheres.
Hydrogen does not adsorb. We wish to design a simple two bed PSA
system to produce 5 nines hydrogen (99.999% pure). Low pressure is
1.1 atm. Use y = 1.15. Superficial velocity during the feed step is 50
cm/s. The cycle is symmetric. Repressurization and blowdown both
require 5 s. Feed and purge steps are 60 s. Repressurize the column
with feed gas.
a How long should the column be to prevent breakthrough?

b. Predict the outlet Cf4 concentration in the waste gas during purge.
Carbon Properties: Ps = 2.1 kg/L, Ko = 1.0, Ep = 0.336, Ee = 0.43
Assume that the ideal gas law can be used to find Pf.
Watch your units in equations for solute velocity.
D8. We are testing the adsorption of unknown pollutants from water onto
activated carbon. Superficial velocity in the columns is 12 cm/min.
The following results were obtained.

Col. A. 20 cm long column. 1m- = 180 min, tMI'Z = 200 min
Col. B. 40 cm long column. It,r =460 min, tMfZ = 200 min
Both breakthrough curves were symmetric. What is the fractional bed
utilization for column B? We want to design a column which will have
90% bed utilization at a superficial velocity of 20 cm/min. Assume that
pore diffusion controls. Determine L and It,r.
D9. Suppose in Example 8-3 that we want CA,band = 25 mM. What value of
Co is required? What are new values of cB,band, tA,band and ~,band?
Other parameters are same values set in statement for Example 8-3.
Fl. We wish to recover toluene from an air stream using activated carbon.
The toluene concentration is 0.30 vol. %. Air flow rate is 50,000 ft3
451
SlP/min at 90°F. Relative humidity is 10%. We will operate each

adsorber for 2 hours for the feed step. Each column will have a 2 ft.
bed depth.
Carbon propertie.s: PB:::: 515 kg/m 3 , Ep:::: 0.75, Ee:::: 0.43,

Cps :::: 0.25 cal/gOC , KD :::: 1.0
a. Estimate the amount of carbon required for each adsorber. (see

Table 8-3).
b. If the superficial velocity of the steam during regeneration is 75

ftlmin, estimate the time required for the steam wave to break-
through. The steam used is saturated steam at one atmosphere.
Chapter 9
ION EXCHANGE
Ion exchange involves the exchange of one ion for another. A common
application is water softening where Ca+2 and Mg+2 ions are removed and
replaced by Na+ ions. Other common applications include water demineraliza-
tion, sugar refining, hydrometallurgy applications, and a variety of biological
separations such as protein fractionation. Ion exchange resins are also used for
adsorption, complexations and exclusion separations where ions are not
exchanged.
After a short introduction to the basics of ion exchange, we will discuss

ion exchange resins and equilibrium. Then ion movement theory and applica-
tions will be discussed followed by a brief introduction to mass transfer in ion
exchange systems. Finally, design will be briefly considered.
9.1. BASICS OF ION EXCHANGE
For monovalent ion exchange the ion exchange reaction for cations (positive
ions) is
(9-1)
In this reaction R- represent the fixed negatively charged ionic sites on the
resin. The oppositely charged ions, A+ and B+ are the counterions which are
exchanging. Ion X- of the same charge as the fixed group is called the co-ion.
The co-ions do not appear to enter into the reaction, but in some cases may
effect the exchange equilibrium. Equation (9-1) is for monovalent ion
exchange where all mobile ions have a single charge. An example would be
452
453
adding KCI to a resin in the Na+ form. The counterions are K+ and Na+ while
Cl- is the co-ion. Anion exchange will be similiar except anions X- and y-
will exchange and be counterions while cation A+ is the co-ion (see Problem
9-A2).
Concentrations in the fluid, C;, and in the resin, CR,i, are usually both
measured in equivalents per liter or equivalents/m3 based on the total column
volume. Note that some authors use actual resin volume for CR, and the two
definitions will differ by a (I-Ee) term.
The total ionic concentration in solution outside the resin is
(9-2)
while the total concentration of counterions in the resin phase
(9-3)
Once the resin has been manufactured, the total resin concentration is fixed.
This is the concentration of negative charges in Eq. (9-1). Because of the prin-
ciple of electroneutrality (positive and negative charges match), Eq. (9-3) must
be satisfied. Thus as counterion B+ leaves the resin it must be replaced or
exchanged by another ion, A+. (Hence the name ion exchange). The total
ionic concentration, CT, can be changed by adding a more concentrated solution
to the column. Until this happens Eq. (9-2) must always be satisfied. Increas-
ing the ionic concentration in the feed will cause an ion wave in the column.
This is a wave where there is a change in the total ionic concentration CT. After
the ion wave, Eq. (9-2) is again satisfied, but with a new value for~.
After complete exchange following reaction (9-1), the resin will be

entirely in the R-A+ form. To regenerate the resin the reaction must be
reversed. This can be done by adding a concentrated solution of B+X-. Thus a
complete ion exchange cycle consists of a loading step (A+ goes on the resin),
a regeneration step (A+ removed from the resin), and wash step where excess
B+X- is removed from the column.
454
For exchange of a divalent cation with a monovalent cation the reaction

is
(9-4)
Now the divalent ion takes up two fixed sites on the resin. An example is the
removal of Ca from water by exchanging it for Na+. Regeneration can be done
with a concentrated NaCI solution.
Equations (9-2) and (9-3) are still valid if CD and CRB are replaced by CD
and CRD. These equations are valid since the units are in equivalents. In equili-
brium equations it will be convenient to define the equivalent fractions of ions
in solution
(9-5)
and in the resin phase
(9-6)
Note that both equivalent fractions must sum to one.
(9-7)
One other interesting phenomena is that the concentration of co-ion, X-,

inside the resin is less than the co-ion concentration outside the resin. This
Donnan exclusion occurs because the co-ions are repelled by the fixed charges
on the resin. The lower CT, the stronger the Donnan exclusion and the lower
the co-ion concentration inside the resin. Donnan exclusion has an important
effect on ion exchange cycles. Regeneration is usually done at high concentra-
tions where a significant amount of co-ion is able to diffuse into the resin.
When the wash solution where CT is very low is added, the ionic concentration
in solution drops rapidly. The co-ions plus their corresponding counterions are
forced out of the resin by the exclusion effect Thus the effective mass transfer
rate during washing is much higher than would be expected.
455
Non-ionic molecules such as sugars or alcohols have no charge and are

not excluded from the resin. Since the co-ion is excluded, the ionic and non-
ionic species can be separated by this ion-exclusion process. This process is
used for sugar purification and does not involve exchange of ions.
9.2. ION-EXCHANGE RESINS
Although different materials are used, the most popular backbone for ion
exchange resins is polystyrene. To make the resin insoluble the polystyrene is
cross-linked with divinylbenzene (DVB). The greater the percentage of divi-
nylbenzene the less the resin will swell when ions are exchanged, but the resin
will be tight and have low mass transfer rates. The range available commer-
cially is from 2 to 12% DVB. A common compromise is to use 8% DVB. The
polymer matrix serves to hold the functional side groups which give the resin
its fixed charge. By adjusting the chemistry of these functional side groups the
behavior of the resin can be changed. This is similar to changing the chemistry
of solvents in solvent extraction.
Resin beads are made in two forms: gel and macroporous. The macro-
porous resins are polymerized in the presence of a third component which is
insoluble in the polymer. After removal of the precipitate, large pores remain
in the beads. These pores make the inside of the beads more accessible to ions.
Macroporous resins can be particularly useful for large ions such as proteins.
Unfortunately, the macroporous resins are more expensive, have lower capa-
city, and are harder to regenerate than gel resins. For both types of beads the
external porosity is typically Ee = 0.38 to .40. Physical properties of gel type
resins are given in Table 9-1. Properties of macroporous resins are given by
Faust and Aly (1987).
Acidic resins have negative fixed charges and can exchange cations as in
Eqs. (9-1) and (9-4). Basic resins have positive fixed charges and can
exchange anions. Exchangers are also classified as strong or weak. Strong
resins are fully ionized and all the fixed groups are available to exchange coun-
terions. Benzene-sulfonic acid groups on a polystyrene-DVB polymer is the
\Jl
0\
"""
Table 9-1. Physical Properties of Ion-Exchange Resins. (Miller et a11984, Vermeulen et ai, 1984)
PB,wet % Swelling Max pH Wet Max

Resin (drained) due to T,oC range Exchange Flow Regenerant
kg/L exchange cap eq(L m/h
Polystyrene HCI
-Sulfonic acid or H2 SO4
4%DVB 0.75-0.85 10-12 120- 0-14 1.2-1.6 30 or NaCI
8-10% DVB 0.77-0.87 6-8 150 0-14 1.5-1.9 30
Polyacrylic acid 0.70-0.75 20-80 120 4-14 3.3-4.0 20 11 0% of theory

(gel) HCI, H2 SO4
Polystyrene
Quaternary 0-7 -20 60- 0-14 1.3-1.5 17 NaOH
Ammonium 80
Polystyrene
- tert-amine 0.67 8-12 100 0-7 1.8 17 NaOH
(gel)
457
a. b. I. II.
-CH - CH 2- -CH - CH - -~H-CH -

@2 @2
@
Ie I I
S03 CH 2 CH 2
I I
CH 3-N-CH 3 CH -N-CH
3 10 3
10
CH CH 2
3
I
CH 2
1
OH
c. CH - CH 2 - i - - t - Ct - CH 2 d. CH -CH -
12 2
/
1
\
@ I
n»m
@
r
HO 0 CH - CH - CH 2
/ 2 I
n N
/'
CH 3
"-CH
3
Figure 9-1. Ion exchange monomers. a. Strong acid. Benzene sulfonic

acid. b. Strong base. Quaternary ammonium structures. c.
Weak acid. Polyacrylic acid with DVB crosslink. d. Weak
base. Tertiary amine on polystyrene.
most common strong acid resin (Figure 9-1a). This is essentially an immobil-
ized sulfuric acid. The commercial resins have dry weight capacities of 5.0 ±
0.1 eq/kg. This corresponds to a typical wet capacity of CRT = 2.0 eq/L;
although this number will vary as the degree of swelling changes (see Table 9-
1). These resins are very stable and can commonly have 20 or more years of
service.
The two most common strong base resins are also based on polystyrene-
DVB polymers. Now the positively charged fixed groups have a quaternary
458
ammonium structure. Two types are shown in Figure 9-1 b. Both of these
resins are fully ionized and are essentially equivalent to sodium hydroxide.
Typical wet capacities are in the range of CRT = 1.0 to 1.4 eq/L. The strong
base resins can degrade, and temperatures above 60°C are detrimental particu-
larly at high pH's. See Table 9-1 for other properties.
Weak acid and weak base resins are only partially ionized at most pH's.
In effect, this often results in a lower exchange capacity, but makes regenera-
tion easier. Since the weak resins can operate near the stoichiometric require-
ments of Eqs. (9-1) or (9-4), they require less regenerant than the strong acid
and base resins. This can be a major advantage; however, the weak resins are
not always applicable.
Weak acid resins are usually copolymers of divinylbenzene and acrylic

or methacrylic acid. The polyacrylic acid is shown in Figure 9-1c, and proper-
ties are listed in Table 9-1. The weak acid resins have a large number of acid
groups which will be partially ionized. These resins can be titrated as shown in
Figure 9-2. At low pH's these resins are not useful since the carboxylic acid
groups are not ionized and their effective capacities are zero. The weak acid
resins have two other disadvantages compared to strong acid resins. These
resins swell or contract when ions are exchanged. This swelling can be as high
as a 90% increase in volume when H+ is replaced by the larger Na+ ion. Swel-
ling can cause excessive pressure drop, resin rupture, and equipment breakage.
The resins are chemically stable, but may break from repeated swelling and
shrinking cycles. The weak acid resins also are "tight" and ions diffuse slowly.
Thus mass transfer resistances inside the resin are high, and the mass transfer
zone will tend to be long.
A variety of weak-base resins are commercially available. One type uses

the same polystyrene-DVB polymer as strong resins for a backbone, but con-
tains tertiary amine groups (see Figure 9-1d). The amine group can be titrated
to give pH curves which are approximately the mirror image of Figure 9-2.
The weak base resins are fully ionized at low pH's and not ionized at high pH.
The weak base resins can all suffer from oxidation and organic fouling. With a
wide variety of resins available, usually one can be found which will be stable
under the loading and regeneration conditions.
459
3.0
...J
"-
g 2.5
->.
'0
c
0..
2.0
C
t.>
Q)
1.5
CI
c:
c No
.&:
t.>
1.0
><
W
0.5
7911
Solution pH
Figure 9-2. Titration curves for weak acid resins in O.03M NaCI. A)
Acrylic acid, B) Methacrylic acid. (Anderson, 1979; p. 1-378).
Reprinted with permission from P.A. Schweitzer (Ed)., Hand-
book of Separation Techniques for Chemical Engineers, 1979.
Copyright 1979, McGraw-Hill.
A variety of speciality and selective resins are also available for particu-
lar applications. (Anderson, 1979; Calmon, 1979; Streat and Cloete, 1987).
Some of these resins have a very high selectivity of one ion because of selec-
tive chemical reactions or complexations. Additional details on ion exchange
resins are available in sources such as Anderson (1979); Arden (1968), Calmon
and Gold (1979), Dechow (1989), Dorfner (1972), Helfferich (1962), Kunin
(1960), Naden and Streat (1984), Rodrigues (1986), and Streat and Cloete
(1987).
9.3. BINARY ION EXCHANGE EQUILIBRIUM
For the exchange of monovalent ions shown in Eq. (9-1), the law of mass
action gives an equilibrium constant
(9-8)
460
a b
Figure 9-3. Equilibrium curves for ion exchange. a. Monovalent exchange,
Eq. (9-11). b. Divalent-monovalent exchange, Eq. (9-15a).
Parameter is KDB CRT/cT
where the a;. are activities. Although this K is a true constant, this equation is
not easy to work with since the activities are not known. Activities can be
related to concentration by
(9-9)
a=~
where y is the activity coefficienL Substituting Eq. (9-9) into Eq. (9-8), assum-
ing X- concentrations and activity coefficients are constant and rearranging,
(9-lOa)
The concentration equilibrium "constant" or selectivity coefficient, K AB , will be

constant only when the activity coefficients are constant. This is more likely to
be true in dilute instead of concentrated solutions.
If we use Eqs. (9-5) and (9-6) the equilibrium can be written in terms of
the equivalent fractions of ions, x and y.
(9-lOb)
461
Table 9-2. Approximate Selectivity Constants (Anderson, 1979).

Strong Acid Resins. B = Li+ Strong Base Resin. B = Cl-
8%DVB Medium Moisture
Ion A KAB 10nD KDB Ion A KAB Ion D Knn
Li+ 1.0 UOr- 2.5 OIr (fype 1) 0.05-0.07 S04- 0.15

H+ 1.3 Mg++ 3.3 F 0.1 CO--
3 0.03
Na+ 2.0 Zn++ 3.5 CH3 COO- 0.2 HP0 4 - om
NH! 2.6 Co++ 3.7 HC03" 0.4
K+ 2.9 Cu++ 3.8 OH- (fypc II) 0.65
Rb+ 3.2 Cd++ 3.9 Br03" 1.0
Cs+ 3.3 Ni++ 3.9 cr 1.0
Ag+ 8.5 Mn++ 4.1 CN- 1.3
Ca++ 5.2 N0 2 1.3
Sr++ 6.5 HS0 4 1.6
Rb++ 9.9 Br- 3
Ba++ 11.5 N03" 4
r 8
Equation (9-7) for binary exchange was used for the last equality in Eq. (9-
lOb). Note that KAB is essentially the same definition as a relative volatility or
as the separation factor for exchange adsorption (Eq. 6-16). If the activity
coefficients are constant, then solving for YA gives,
(9-11)
In this form Eq. (9-11) is the same form as the Langmuir isotherm, Eqs. (6-6)
and (6-12a). This is an important result. If KAB is constant, theories developed
for Langmuir adsorption can be used for binary monovalent ion exchange and
vice versa. KAB will be constant for dilute systems where all the 'Yi = 1.
The equilibrium curves for Eq. (9-11) are shown in Figure 9-3a.
Approximate values for KAB for both anions and cations are given in Table 9-2.
462
With weak base resins the resin structure can have a major effect on KAB (Clif-
ford, 1982). Equilibrium data can be checked for fit to Eq. (9-11) as in Eq. (6-
12b) and Example 6-1.
For exchange of two monovalent ions
(9-12)
KcA = KcIJIKAB
Thus, Table 9-2 can be used to find selectivities for a variety of pairs of ions.
If divalent and monovalent ions are exchanging as in.Eq. (9-4) the mass
action law gives an equilibrium constant
(9-13)
If activity coefficients are constant the concentration equilibrium constant,

KDB , is
(9-14a)
Introducing the equivalent fractions of ions from Eqs. (9-5) and (9-6), we
obtain
cRT)_ YD xB2
KDB ( - (9-14b)
---2-
CT xD YB
Equilibrium now depends on the groups (KDB CRT/CT) and will change when
the feed concentration is changed. This effect is utilized in water softening and
is illustrated in Example 9-2. Substituting in the summation Eqs. (9-7), we have
(9-1Sa)
or
yfi
---
1 xfi (9-1Sb)
l-YB (KDB CRT/~) (l-xB)
463
Equation (9-15a) is plotted in Figure 9-3b. Note that the curves are symmetric
around the y = x axis. These equations are also valid for anion exchange where
the charge signs on D, B and R are changed in Eq. (9-4). Note in Figure 9-3b
that the shape and relative selectivity can be changed by changing the fluid
concentration, Or. For more accuracy Eqs. (9-15a or b) are easily solved if
either xo or Yo are known (see Example 9-2). A nomograph is also available
(Vermeulen et oJ .• 1973).
Approximate values of KOB for strong resins are given in Table 9-2. A
large number of mass-action constants are reported by Marcus and Howery
(1975), and have been correlated by Vermeulen et al .• (1984). For exchange of
two divalent ions, E and D,
(9-16a)
while for the exchange of a monovalent ion A, and a divalent ion D,
(9-16b)
Thus Table 9-2 can be used to estimate the selectivity for a variety of ion
exchange systems.
9.4. ION MOVEMENT THEORY
The solute movement theory developed in Chapter 6 can easily be adapted to

binary ion exchange. The most common resins used are gels in which the ions
move by diffusion. Because of electroneutrality, there is no volume where ions
can sit without being attached to the resin. Thus the development in Eqs. (6-
19) to (6-23) is modified to have Ep = 0 and to use the ion exchange equilibrium
in terms of equivalent fractions x and y. The result for any ion is
(9-17)
464
where y and x refer to the equivalent fractions of the ion being studied. KE is
an exclusion factor to include the effects of Donnan exclusion and electroneu-
trality. If the ion is excluded KE = 0 while otherwise KE = 1. For the coun-
terions, A+, B+ and D++ in Eqs. (9-1) and (9-4), KE = 1. For the co-ions, X-,
KE is zero at low concentrations. The (I-Ee) P. term which appears in the
denominator in adsorption calculations (Eq. (6-23», is not required in ion
exchange since CRT is defined in terms of the total volume, and is in
equi valen ts/liter.
Equation (9-17) can represent either shock or diffuse waves. Figure 9-3
shows that when KAB > 1 or when (KOB CRT/Or) > 1 the equilibrium isotherms
have a favorable shape. Then shock waves will result if a solution with low
fraction of species A (or D) is displaced by a solution with a high fraction of A
(or D). This is exactly the same conditions which give a shock wave for
adsorption (see Eq. (6-30) and Example 6-2). For shock waves Eq. (9-17) is
(9-18)
where a refers to after the shock wave and b to before. Both before and after
the shock wave liquid and resin are assumed to be in equilibrium.
For a favorable shape isotherm a diffuse wave will result if a solution of

high fraction species A (or D) is displaced by a solution of low fraction species
A (or D). For the diffuse wave !J.y/tlx is determined from the derivative and
Eq. (9-17) becomes,
(9-19)
The derivative, dy/dx, can be determined from the appropriate equilibrium

equation, such as Eqs. (9-11) or (9-15), or from the slope of the equilibrium
curve at the desired x value.
For monovalent ion exchange equilibrium data (YA vs XA) does not
465
depend on the total resin capacity, CRT, or the total fluid concentration, CT.
However, both shock and diffuse waves depend on the ratio CRT/CT. If CRT/CT
is high the resin capacity is large compared to the ionic concentration in the
fluid. Waves move slowly since sites on the resin are filled slowly. If CRT/cT is
low, the resin is easily saturated. The waves move quicldy in the column.
Regeneration usually uses this phenomenon. The ionic concentration of the
regenerant may be orders of magnitude higher than the feed concentration. The
column will then quicldy saturate with the regenerating ion and regeneration is
fast.
When the ionic concentration of the liquid is changed, an ion wave will
pass through the column. A simplified picture of the ion wave is that the resin
is already holding all the counterions it can. Counterions can exchange accord-
ing to Eqs. (9-1) or (9-4), but more equivalents cannot be held up or supplied to
the fluid. Thus, for an overall balance the ions are excluded and KE = 0 in Eq.
(9-17). The resulting ion wave moves at the interstitial fluid velocity.
(9-20)
Utotal ion = V
For concentrated solutions this picture is modified since some co-ions can
penetrate into the resin. The co-ions take counterions with them. Thus the
effective capacity of the resin is greater than CRT and Urotal im < v. This effect is
usually not large, and will be ignored in our calculations.
For monovalent systems the ion wave does not affect equilibrium, thus
xib = xia and Yib = Yia. The ion wave does change shock and diffuse wave velo-
cities following Eqs. (9-18) and (9-19), respectively. The pattern of shock,
diffuse and ion waves for monovalent ion exchange is illustrated in Example
9-1 and in Figure 9-4.
Example 9-1. Ion movement with monovalent ion exchange
We wish to remove K+ from a liquid using a Duolite C20 (sul-

fonated polystyrene with 8% DVB - see Figure 9-1a) cation exchange
resin at 20°C. The resin is initially in the Na+ form and will be regen-
erated with both 0.2 N and IN NaCl. The initial feed to the column is
466
0.2N solution with XK = 0.70. The superficial velocity is 5.0 cm/min. A

15 cm laboratory column is used. A pulse of feed is introduced for 15
minutes followed by 5 minutes of 0.2 N NaCI and finally 1.0 N NaO.
Determine the outlet concentration profiles for both K+ and Na+.
Solution.
A. Define. The steps of the process are sketched below.
+ I 0.2~ L~ 0~2~1 L I..ONJ I_

No xK =0.95'-------" xNc .I.OlL--------'!xNc .I.OlL-.- - - - - ' I
Initial Feed I st 2nd
Regeneration Regeneration
B. Explore. To determine the outlet concentration profile we need to

develop the ion movement diagram. This requires calculation of
Ush, Us and ulota! ion. Both Usb and Us require eqUilibrium data. After the
ion wave, both Usb and Us will change.
C. Plan. The equilibrium data has been measured at 20°C by Bailly

and Tondeur (1981). Equilibrium data for a 0.2 N solution was fit very
well by a constant equilibrium constant, KKN. = 1.54 (Note this is close
to the approximate result from Table 9-2. KKN. = KKLiIKNaLi = 1.45).
We will assume there is no concentration effect on KKN•. Porosity was
approximately Ee = 0.4. The total exchange capacity was 2.38
equivalents!liter.
Since KKN. > 1.0 and the bed initially has a zero concentration of
K+, a shock wave will result when the feed is introduced. Equation (9-
18) can be used to calculate Usb. For this step Ks = 1, YbK = XbK = 0,
and the fractions after the shock are in equilibrium following Eq. (9-11)
with XaK = xFeed = 0.70.
The first regeneration is at the same value of CT so an ion wave is

not generated. A diffuse wave following Eq. (9-19) will result. The
derivative (dy/dx) comes from Eq. (9-12) and is,
dYK KKN. (9-21a)

---
dxK [1 + (KNa-1)xK]2
467
and the resulting diffuse wave velocity is,
(9-21b)
As expected UK depends on XK'
The second regeneration step causes an ion wave which follows

Eq. (9-20). When this wave intersects the diffuse and shock waves,
these waves change since CT has increased to 1.0 N. The wave slopes
are recalculated from Eqs. (9-21b) and (9-18).
D. Do It Interstitial velocity v = Vsuper = 54 = 12.5 c~ . Then the

Ee O. mm
shock wave velocity is,
From equilibrium,
(0.7)(1.54) = 0.782
1+(1.54-1)0.7
Other values are: Ee = 0.4, CRT =2.38, CT = 0.2, KE = 1.0, YK.b = XK.b =
O.
Then Ush = _ _ _ -:--_1_2~.5_.."..--__:_ = 0.365 cm/min

1 + _1_ [2.38] [0.782-0]
0.4 0.2 0.7-0
This is plotted on the z vs t diagram in Figure 9-4.

468
The diffuse wave for the first regeneration step uses Eq. (9-21b)
with Ee =0.4, CRT =2.38, CT = 0.2, KE = 1.0, and KKN. = 1.54.
12.5
Thus UK = ---45-.-8-15--
1+----_:=_
[1 + 0.54xKf
Values can be determined for different XK as shown below.
XK UK (~=0.2) UK (~ = 1.0) cm/min

0.7 0.497 2.146
0.5 0.424 1.871
0.3 0.358 1.605
0.1 0.296 1.352
0.0 0.267 1.240
This diffuse wave is plotted on Figure 9-4.
The total ion wave is Uu,tal ion = V = 12.5 starting at t = 20 minutes. This
is plotted on Figure 9-4a At the intersection points of the total ion
wave with the diffuse and shock waves these wave velocities change.
Thus we must recalculate Ush with CT = 1.0.
and the diffuse wave velocity UK(CT = 1.0),
12.5
UK = ---=--~--
1+ 8.163
[1 + 0.54XK]2
The diffuse wave values are given in the table above. These changes in
slope are shown in Figure 9-4a
The solute movement diagram can serve as a guide for the exact
calculations. For example, for the shock wave point 1 on Figure 9-4a
can be calculated as the point of intersection of shock and ion waves.
Figure 9-4. Solute movement and concentration profiles for Example 9-1.
Ion exchange reaction:
K+ + R-Na+ + Cl- ~ Na+ + R- K+ + Cl-
a. Solute movement diagram. b. Outlet profiles.
That is,
llsh(er = 0.2)t = Utotalion(t - 20)
or t:: 20.60 min and z = Utotal ioo (20.6 - 20) = 7.5 cm. Then point 2 at z
=L = 15 is the continuation of the shock wave but at a new value of CT
and thus a new shock velocity.
(15 - 7.5) = Ush(t - 20.60)
which gives t = 25.19 minutes. Similar calculations can be done for the
diffuse wave at different concentrations.
470
The outlet concentration and K+ fractions are plotted in Figure 9-

4b. The Na+ fractions are easily detennined from
E. Check. A basic check is the mass balance on K+. Does
Mass Kin =Mass Kout

over the entire cycle? At first, Figure 9-4 may seem to violate this.
However, the mass balance
is satisfied since the outlet total concentration is 5 times the inlet total
concentration.
F. Generalization. In general, ion movement and regeneration is faster

at higher total fluid concentrations. Thus regeneration is usually done at
higher total concentrations. The use of the ion movement diagram as a
guide for exact algebraic calculations is generally useful. Of course, the
ion movement diagram is then unnecessary, but serves to keep the
bookkeeping straight.
Note that both K+ and Na+ have shock waves at the same time
and have diffuse waves at the same time.
If the column in Example 9-1 were longer, the shock and diffuse waves
would intersect. The peak concentration would decrease below XF = 0.7. Also,
the shock wave would become curved since x. and y. in Eq. (9-18) decrease.
For divalent-monovalent ion exchange following Eq. (9-4), equilibrium

depends on Or. If a divalent ion such as Mg++ or Ca++ is exchanged for a
monovalent ion such as a Na+, the feed step will usually have (KOB CRT/CT) >
1.0. This occurs because CT is low and KOB is usually> 1.0. Thus, equilibrium
is favorable and a shock wave results during the loading step. During the
471
regeneration step a concentrated NaCI solution is used. This greatly increases

CT and now it is possible to have (KDB CRT/CT) < 1. Figure 9-3b shows that the
isotherm is now unfavorable. Since the divalent ion is being removed with a
solution which has a lower fraction XMg or XCa' a shock wave again results.
This makes regeneration much easier. This behavior is exploited in water
softening and will be explored further in Example 9-3.
When the total ion wave passes through the system, a mass balance on a
segment of the column shows that YD. = YDb' This balance is essentially the
same as the balance for thermal waves, Eq. (6-39a), but with Utotal ims = v
replacing uth (see Problem 9-C3). Although fractions on the resin are constant,
the equivalent fractions in the liquid change since Or changes. Equations (9-
15) can be used to calculate XD and XB after the total ion wave passes. These
calculations are illustrated in Examples 9-2 and 9-3.
Mass transfer and dispersion have the same qualitative effect on the
predicted breakthrough curves for ion exchange as they did for adsorption.
Thus the shock waves will be spread out and will form a constant pattern wave
as in Figures 8-1 to 8-3. This spreading can be predicted from the mass
transfer zone and constant pattern arguments discussed in Chapter 8. For
higher values of the particle diffusivity or smaller particles, mass transfer will
be more rapid and the breakthrough curve will approach the shock wave shape.
For diffuse waves the isotherm shape is usually the controlling effect Thus the
predicted diffuse waves will agree with experimental data except when mass
transfer is very slow (for example, with proteins).
The simple theory presented here is not applicable to systems with more
than two ions. More complex theories (for example see Helferrich and Klein,
1970; Aris and Amundsen, 1973; or Tondeur and Bailly, 1986) are required
since equilibrium depends on the concentration of all ions. These theories are
beyond the scope of this book.
Example 9-2. Effect of Total Ion Wave
A 100 cm long bed packed with a strong acid resin is saturated with a
feed with Or = 0.1 equivalents/liter. xH = 0.4 and XMg at 0.6. At t = 0
472
we feed a mixture with xH = 0.4, xMg = 0.6 but Or = 0.8. Superficial

velocity is 15 cm/min, fe = 0.40, co-ion is cr, CRT = 2, and Table 9-2
can be used for selectivities. Determine the effect of this increase in ion
concentration.
Solution
An ion wave will pass through the column at velocity
Uim =V =vsU{X%/£c + 15/0.4 = 37.5 cm/min

It takes ~m = Llllim = 100/37.5 = 2.6667 min to exit the column. We
can do equilibrium calculations to determine the effect of the ion wave.
From Eq. (9-16b) and Table 9-2,
Before the wave: KMgH CRT/Ct = (1.953) (2.0)/(0.1) = 39.05
After: K Mg H CRT/Or = (1.953) (2.0)/0.8 =4.882
Equation (9-15a) can be used to find Yo before the ion wave. Since Xo
=0.6 is known, we can solve for Yo. The result is a quadratic
(9-22a)
where
b=- [
2+
K;B ~RT 'D 1 (9-22b)
The solution is
Yo
-b- ...Jb2=4
= ---:2:----
(9-22c)
473
where we select the minus sign in the quadratic fonnula since

o ~ Yo ~ 1. For this case
b [
=- 2+ 1]
(39.05) (0.6)/(0.4)2
=-20068
. 3
and YMg,b = 0.9207

After the ion wave has passed,
(9-23)
Yo,. = YO,b
Now Yo,. is known and we solve Eq. (9-15a) for xO,.' The result is
again a quadratic
(9-24a)
X5 + b' Xo + I = 0
where
KOB CRT (1 - yo)2] (9-24b)

b / =- [ 2 + - - - - - -
CT Yo
and the solution is again given by Eq. (9-22c). Now
b' = - [2 (4 882) (0.0793)2] = - 2 03334

+. 0.9207 .
Note that the parameter KOB CRT/~ has changed values because of the
ion wave. The result from Eq. (9-22c) is XMg,a :;:; 0.8333.
The equivalent fraction of Mg++ has increased.
This stream is now pushed out with a feed containing CT :;:; 0.8
and XMg = 0.6. The result will be a diffuse wave. The ion velocity is
given by Eq. (9-19) which is
v
UO,dif =-------- (9-25a)
1 CRT dyo
1 + - - - KE - -
Eo CT dxo
474
a
L
Z, cm
50
0
0 5 10 15
t
0.8 b
0.6
I
J
cMg
ouf 0.4
0.2
0
0 5 10 15
t, min
Figure 9-5. Solution to Example 9-2. a. Ion movement diagram. b. Outlet
Mg+ concentration in equivalents/liter.
We can detennine dyo/dxo by differentiating Eq. (9-15a). After some

algebra. we obtain
(1 - YO)3 KOB CRT (9-25b)

--=
dxo l+yo CT
When xMg = 0.8333, we know YMg = 0.9207. Then
dYMg (x = 0.8333) = (1- 0.9207)3 (4.882) 1 + 0.8333 = 0.5017

dx Mg Mg 1 + 0.9207 (1 - 0.8333)3
and
UMg,dif = 1 (2~7.5 = 9.0678

1 + 0.4 (0.8) (1.0)(0.5017)
475
Other values are listed in the table.
dYMg
xMg YMg UMg t= L/UMg CMg
dxMg
0.8333 0.9207 0.5017 9.0678 11.028 min. 0.6666
0.700 0.85035 0.5567 8.371 11.9455 min. 0.56
0.6000 0.792 0.6128 7.7636 12.881 min. 0.48
The ion-movement diagram and outlet concentration profile are shown

in Figure 9-5. Outlet concentration is cMg = XMg CT.
If CT after the wave was high enough so that K MgH CRT/Or < 1, then a
shock wave instead of a diffuse wave would follow the total ion wave.
This is illustrated in Example 9-3.
9.5. APPLICATIONS
The most common application of ion exchange is water softening. "Softening"

is the removal of "hard" ions such as Ca++ or Mg++ with "soft" ions such as
Na+. The hard ions have inverse solubility curves; that is, they become less
soluble in hot water. When hard water is heated, the calcium and magnesium
precipitate out and form a scale. This is not only unsightly it also decreases the
heat transfer rate. Hard ions also tend to form a scum with soaps which makes
washing clothes, dishes or hands more difficult When the Ca++ and Mg++ are
exchanged for Na+, none of these problems occur. However. the removal of
Ca++ may be detrimental in drinking water since the Ca++ seems to decrease
the incidence of heart disease (Anon, 1985). Softening is also useful for beet
sugar syrups to prevent scaling of pipes and evaporators.
Water softening usually uses a strong acid polystyrene resin with around
8% DVB for cross linking. Packed columns are used and regeneration is done
with concentrated salt (NaCl) solutions. The entire process is usually
automated. A good idea of this process can be obtained by looking at a home
water softener (try the local Sears store or a Sears catalog).
476
Water softeners take advantage of the change in equilibrium for

divalent-monovalent exchange when Or is increased. Both loading and regen-
eration steps will be operated with shock waves (see Example 9-3). Regenera-
tion is usually done counterflow to the feed direction. Feed flow would usually
be downwards while regeneration is done upwards. Regeneration is usually
completed in a few hours while the loading step lasts several days. Regenera-
tion is followed by a short wash step to remove excess salt solution from the
bed.
Example 9-3. Analysis of Water Softener Cycle.
We wish to soften a water containing 2.0 meq/L Ca++ and 9.0

meq/L Na+ (meq is milliequivalents). Superficial velocity of feed is 10
cm/min. A 2 meter column is used. Regeneration uses 25 wt % NaCI
at a superficial velocity of 0.5 cm/min. A strong acid resin with a total
wet capacity of CRT = 2.0 eq/L and equilibrium parameters given by
Table 9-2 should be used. Determine the maximum feed period, the
time required for regeneration, and the time required for washing. Also
determine the outlet concentration profile during regeneration and wash.
Solution.
A. Define. The process is shown below. Note that counterflow regen-

eration is chosen.
vsuper=O.5
2 meters
26 wt% Wash
Noel Water
477
We are to determine the period for each step.
B. Explore. Data is available in Table 9-2; although. we must

remember this is only approximate. Several different units are given
and must be put into the same units. Since KcaNa CRT/Or probably> 1.0
during the feed. the maximum feed period occurs when the shock wave
starts to exit Thus ~ = LlUm. This does not include the M1Z and thus
overpredicts the feed time. The large increase in cT. during regeneration
probably makes CKcaNa CRT/CT) < 1, and thus a shock wave occurs dur-
ing regeneration also. The wash step removes excess excluded salt
from the bed and; ignoring dispersion. moves at the velocity utotal ion =
V.
C. Plan. Preliminary calculations will put all concentrations in units of

eq./L.
Feed step. Calculate Xc. in feed. Calculate KcaN. CRT/Or and check if >
1.0. Determine Yea and Uab from Eq. (9-18) and then ~ = L/Um.
Regeneration. Calculate CKcaN. CRT/CT) and check if < 1.0. Determine
effect of ion waves on Yea and Xc. before Uab' Calculate Usb and
~ =L/Um·
Wash. Calculate lwasb = L/1ltotal ion
Can use following data values. Ee = 0.4, KcaNa = (KCaIJ/(KNaLY = 1.3.
D. Do It. For/eed step CT =2 + 9 = 11 meq/L = 11 x 10-3 eq!L
XF.Ca = 2/11 =0.1818, XF,Na =0.8182

v = Vsuper/Ee = 10/0.4 = 25 cm/min
CKcaK CRT) = (1.3)( 2.0 3) = 236.4 which is:> 1.0

'T Ilxl0-
Mter shock wave x•.c. = XF,C. =0.1818

478
Calculate Yea+> from Eq. (9-15a) or Eq. (9-22)
Ya.C. _ rv _ CRT) x.,C.

- ,"'''CaK 2
(l-Ya,C.)2 ~ (1- x.,c.)
== (236.4) 0.1818 == 64.2

[l-Q.1818f
Solving for Y"c. using Eq. (9-22), we obtain Ya,Ca == 0.8827.

Because of the low concentration of Ca++ in the feed and the very high
value of (Kcax: CRT/cT), the resin contains a much higher fraction of
Ca++ than the feed.
v
1lsb==----------
1 CRT K Y.,Ca - Yb,C.
1+ - -- E ----'----
Ec CT X.,C. - xb,Ca
Before the shock the column is in Llte sodium form. Thus Yb,C. == xb,Ca ==
o
Ush == 25 = 0.0113 cm/min
1+ (1)(2.0)(1.0)(0.8827)
(0.4) (1.1x1o-2) (0.1818)
where KE = 1.0 since ions are not excluded. Note the low velocity of
the shock wave. This occurs because the resin has a high capacity com-
pared to the liquid concentration, and the resin is selective for Ca++.
200cm .
IF = 00 3 I ' = 17,664 mm =294.4 hours
. 11 cmmm
Regeneration: v = vsupcr/Ec =0.5/.4 = 1.25 cm/min

To convert 25 wt % to equivalentslL need the density. At 25°C,
PI. = 1.194 g/ml (perry and Green, 1984, p. 3-83). Pick a basis of 1
liter of solution. Then 1.194 kg solution x 0.25 = 310.4 g NaCl. Since
479
MWNaCl = 58.45, we have 310.4/58.45 = 5.311 moles which is also

5.311 equivalents. Thus CT = 5.311 equivalents/liter. XCa,Reg = 0 since
regenerant is entirely NaCl.
Ratio KeaNacRT/CT = (1.3)(2.0)/(5.311) = 0.49 which is less than 1.0.
Thus, we do get a shock wave when material concentrated in Ca++ is
removed with regenerant.
The total ion wave moves at velocity v = 1.25 cm/min and takes
200 cm/(1.25 cm/min) = 160 min to exit.
When the ion wave passes, Ya.Ca = Yb.C. = 0.8827, but xc.
changes since CT has changed. Use Eqs. (9-15a) and (9-24) to calculate
Xa.Ca
= 0.8827 _1_ = 130.92

(0.1173)2 (0.49)
We obtain x•. c. = 0.9167 which is a large increase. Before the ion

wave, fluid exits with CT = 1.1 x 10-2 and xb.Ca = 0.1818. Immediately
after the ion wave fluid exits with CT = 5.311 and xa.Ca = 0.9167. This
continues until the shock wave when Xc. drops to zero. For the shock
wave Yb.Ca = 0.8827 and xb.Ca = 0.9167. After the shock Ya.Ca=xa.C.=O.
v
lIm=-------::------::-
1 CRT [Ya.ca - Yb.C.j
1+- --KE
Ee CT X•.C. - Xb.C.
lIm = 1.25 0-0 882 = 0.656 cm/min

1+ (0~4)( 5~~1 )( 0--0:916;)
· h k
Th IS . 200 cm = 305.0 minutes which is
s oc wave reqwres 0.656 cm/min
145.0 minutes after the total ion wave exits. Thus solution with CT =
5.311 and Xc. = 0.9167 exits for 145.0 minutes.
480
Wash: The wash step has v = 1.25. Since the salt contained in the void
volume is excluded (the resin is always at capacity), KE =0 and Uwuh =
v = 1.25. (This can also be considered another total ion wave which
gives the same result). The wash step requires 160 minutes; however,
we can start the wash step so that the wash wave exits shortly after the
regeneration shock wave exits. This saves time and regenerant solution.
Summary: Feed Sl.ep 17,764 min = 294.4 hours

Regeneration 305.0 - 160 minutes = 145.0 minutes =
2.42 hours (see Figure 9-5).
Wash step 160 minutes = 2.667 hours
Total 299.48 hours
294.4
% on stream time = 299.48 x 100 =98.3%
The complete solute movement diagram is shown in Figure 9-6.
xCa =0.1818 xc~0.9167

F c T=l.lxI0-2 c T=5.311 F
L=200r-~~----~~!~~~
u sh
__~__ - L__ - '__ ~ __ T-1-
z
em
Uwash
xCa=o
ush CT =0
OO~~~~2~~3~VVVVVV~29=3=-~~-7~~-2~9t~8~2~99-.5~
Time, hours 294.4

Wash
Regeneration
Figure 9-6. Solute movement diagram for water softener for Example 9-3.
Ion exchange reaction:
481
E. Check. Since complete regeneration was done, Ca* in should equal

Ca* out over a cycle.
During feed: ea++ in = vsuper Ac Xc. CT tF

= (10) Ac (0.1818)(1.1 x lO-2)(17,764) = 355.2 Ac
Ca++out = vsuper Ac[(Xca CT t~ore total ion wave
+ Xc. CT ltrom ion wave to Shock wave]
= 0.5 Ac[(0.1818)(1.1 x lO-2)(16O) + (0.9167)(5.311)(145.0)] = 353.3 Ac
In = Out within the accuracy of the calculation.
F. Generalization. This example shows a very large increase in both

Xc. and ~ when regeneration with a concentrated NaCl solution is
used. In addition, the concentrated NaCI solution fonns a shock wave
which makes regeneration faster and more efficient Note that the
regeneration and wash steps take much less time than the loading step
even though the superficial velocity is much lower in the regeneration
and wash steps. This large shift in concentration, the speed of regenera-
tion and the fonnation of shock waves during both feed and regenera-
tion steps are all reasons why water softening has been widely adapted.
In addition, the low Xc. and CT of the feed are very favorable. With
higher Xc. and ~ in the feed, the feed shock wave would move faster
and the column would load quicker.
Demineralization (or deionization) is also a common industrial ion

exchange process. In demineralization the cations are exchanged for H+ and
anions are exchanged for aIr. The W and aIr react to fonn water. Thus,
unlike water softening, this process removes ions from the solution. The cation
and anion exchange resins are separately regenerated with acid and base. One
application of deionization is in the sugar industry where sugar yields can be
decreased by first deionizing. A second application is polishing (final
482
purification to remove traces of ions) of condensate for high pressure boilers.

(e.g. see Streat, 1988)
Demineralization can be done in two beds packed separately with cation

and ion exchange resins. However, it is more efficient to use mixed beds. A
mixed bed contains both anion and cation resins during the feed step. After the
feed step is completed, the resins must be separated before regeneration. This
can be done by fluidizing the bed with upward water flow. Since different den-
sity and size resins are used, the two resins will stratify and can be separated.
The ion movement theory plus the M1Z analysis can easily be applied to dem-
ineralization systems. Design of the mixed beds requires detailed hydro-
dynamic data.
Ion exchange is extensively used for recovery of metal ions. This is done
both in mining operations and in industrial reprocessing such as plating opera-
tions. Examples are recovery of uranium, copper, gold, silver, rare earths, and
transuranic elements (for example, see Naden and Streat, 1984 or Streat, 1988).
These applications often use moving bed systems which are discussed in
Chapter 10.
Ion exchange chromatography is a common separation method in

biotechnology (Dechow, 1989; Janson and Helman, 1982; Wankat, 1986).
Amino acids, peptides and proteins all have charged groups and can be
exchanged on cation or anion exchange resins. Resins with large pores must be
used. The operational method for ion exchange chromatography usually uses
an on-off cycle which was briefly discussed in Chapter 7. In on-off
chromatography the equilibrium constant is very large under the feed condi-
tions. Thus, a protein or amino acid which exchanges stays on the resin. There
is no migration after an empty site has been found. The off part of the cycle
requires changing the conditions so that the protein or amino acid is less
attracted to the resin. This can be done by changing either the salt concentra-
tion and/or the pH. By using a salt or pH gradient proteins or amino acids can
be removed one-by-one. Thus much of the separation may be achieved during
the regeneration step. Batch operation in stirred tanks is common since loading
can be done in an agitated state which will not clog from particulates (see
Chapter 10).
Elution profiles from DEAE-Sepharose CL -68 of a production run
00"., Column KS370/15.1I, 161 1 Sample apphcatoan

Flow rate 24 I/h 2 Sodtum acetate pH 5 2. I 0025
Eluent. Sodoum acetate butters 3 Sodium acetate pH 4 5. I 0025
Sample 4 2 I~res contaIning ca 500 9 albumon 4 Elution of albumin fraction
Product 20 htres contaonong ca 500 9 albumIn 5 Sodtum acetate pH 4 O. I 0 '5
6 Sodtum acetate pH 5 2. I 0025
[JAibumlO traction
I
o 100 200 300 400 500
Elutoon volume. htr.
Figure 9-7. Use of ion exchange chromatography for commercial purification of albumin. (Curling et al.
1980). By permission ofPharmacia Fine Chemicals AB, Uppsala, Sweden.
~
00
w
484
An example of the commercial use of ion exchange chromatography for

purification of blood plasma proteins is shown in Figure 9-7. Three complete
cycles are shown. The first peak (from 1 to 2) is unretained material. Mter
washing the column, the albumin is eluted (4 to 5) by decreasing the pH. The
strongly held impurities are removed by further decreasing the pH and increas-
ing the ionic strength. The column is then washed with same buffer solution
(#6) as is used for the feed. This prepares the column for the next feed step.
The DEAE Sepharose CL-6B is a special cation exchange resin with very large
pores (Janson and Hedman, 1982).
The ion movement theory can be applied to on-off chromatography

(Wankat, 1986). Shock (constant pattern) waves occur during the loading step.
The waves during regeneration depend on the details of the exchange. If the
amino acid, peptide or protein has more than one charged group which
exchanges, shock waves can also occur during regeneration.
Very large scale applications of ion exchange may not use packed bed
systems. In these cases the moving bed systems discussed in Chapter 10 are
used.
9.6. APPLICATIONS WITHOUT EXCHANGE: ION EXCLUSION
Ion exchange resins are also used for separations where ion exchange
does not occur. For example, (the polystyrene-DVB resins will adsorb many
organic compounds from water. Macroporous resins are normally used. These
adsorption applications were covered in Section 8.6.5.
A second non-exchange application is to separate using ion exclusion

(Neuman et ai, 1987; Wheaton and Bauman, 1953). This is illustrated in Fig-
ure 9-8 for the separation of sugar (glucose) from sulfuric acid (Neuman et ai,
1987). Non-charged molecules such as sugars can penetrate into the resin since
they are not excluded. Co-ions such as S042 will be excluded from a cation
resin by Donnan exclusion effects. Ifa cation exchange resin in the H+ from is
used, the non-charged sugar molecules have a larger volume available to them
than the S042. Because of electroneutrality, the W must migrate with the
485
S042. The result is a chromatographic separation similar to SEC, but the

mechanism for separation is different Note that the sulfuric acid exits first
Ion exclusion is used for separating sugars from salts and acids, separating gly-
cerol from salts, and purification of ethylene glycol by removing salts.
The non-charged (NC) molecules will have a velocity given by
v
UNC = --1---- (9-26a)
1 + - X Kci,NC
Ee
where X is the volumetric fl'dction of the swelled resin taken up by water, and
Kci.Nc is fraction of the volume available to the non-charged molecules. Note
that at high concentrations X and ~,NC will be a function of the ionic concen-
tration. The factor (X Kd,Nc) can be treated as a distribution coefficient which
needs to be determined experimentally. At low concentrations the co-ions are
excluded by the Donnan exclusion effect and stay in the external void fraction
Ee·
(9-26b)
l1co-ion = v
Obviously, the co-ions move faster and can be separated from the non-charged
molecules. Because of electroneutrality, a counterion must move with the
excluded co-ions.
An example of ion exclusion chromatography is shown in Figure 9-8

where sulfuric acid is separated from glucose (Neuman et al., 1987). The
strong cation exchange resin had Ee = 0040 and X = 0.766 in pure water and X =
0.701 in 0.82 M sulfuric acid. In pure water UNC = OA7v and separation is rela-
tively easy. As the sample size increases, the local value of Or increases and
Donnan exclusion becomes less effective. Thus, as illustrated in Figure 9-8,
the separation deteriorates considerably with larger feed volumes.
Ion exclusion at low concentrations is essentially a linear system with a

resin concentration given by,
(9-27a)
486
TEMPERATURE =55°C
1.0
O.S 0.10 Vo (0 )
06
0.4
0.2
0.0 L-_-'-----<Dl--'--.iI2lI>---L-_---""'-_----L_---'
00 05 1.0 1.5 2.0 2.5 3.0
1.0 ~----------------,
0. 25Vo (b)
OS
...
U 0.6
"-
U 04
02
00 L-_.L---(£(ll.-<lII!llo!::.L..02m:~-----.:;.!lnL_ __l
00 0.5 1.0 15 20 2.5 3.0

I0 ~----------------,
0.50Vo (c)
OS
06
04
02
o0 L-_"-----~'-LDl"'-----'-----=:u----'-""W----.J
00 05 10 15 2.0 2.5 30
VIVO
Figure 9-8. Ion exclusion chromatography separation of sulfuric acid and

glucose. Vo =void volume =Ee Vcol . a. VF =0.10 Ee Vcol . b.
VF = 0.25 Ee Vcol . C. Vp = 0.50 Ee Vcol • (Neuman et al.,
1987). Reprinted with permission from Reactive Polymers, 5,
55 (1987). Copyright 1987, Elsevier.
Thus, at low concentrations the local equilibrium dispersion model, Eqs. (7-12)
and (7-13) can fit the outlet concentration profiles. To do this we modify V
from Eq. (7-12c) to
(9-27b)
V= (Acz) [Ee + Kci,NC X]
The (l-Ee) term in Eq. (7 -12c) disappears since CR is based on total bed
volume. Equations (7-12a), (7-12b) and (7-13) are unchanged. Superposition
principles are valid and the equations for differential pulses (Eqs. (7-15) and
487
(7-16a,b» are valid. For small pulses and hence low concentrations this dispet-
sion model fit the glucose data shown in Figure 9-8a (Neuman et aI, 1987). For
larger pulses (higher concentrations in the column) X ~,NC varies and the
model did not fit as well. This model is applied to ion exclusion in Problem 9-
D8. With appropriate data Rosen's model (Section 7.8) can also be used to
model ion exclusion at low concentrations.
9.7. MASS TRANSFER IN ION EXCHANGE SYSTEMS
Mass transfer in ion exchange systems is considerably more complex

than in adsorption. The flux due to electrical transference is (Helferrich, 1962)
(9-28)
where <I> is the induced electrical potential, <1 is the concentration in moles/liter,
Zj. the electrical valence and ~ the ionic mobility. For ideal solutions the
mobility can be determined as
(9-29)
where F is Faraday's constant. Equation (9-29) is also valid inside the ion
exchange resin.
The net flux is the sum of diffusional flux and the electrical transference
flux.
F (9-30)
J.1 =J.1 dif +J.1 e1cc =-D·(V'r.
1 "'l
+z·c·-V'AI.)
1 1 RT 'I'
This is known as the Nemst-PIanck equation. Equation (9-30) is solved along

with the condition of electroneutrality,
(9-31)
488
and the condition of no current flow (equal fluxes of equivalents)
(9-32)
Even very small deviations from electroneutra1ity will produce strong electrical
fields. If one counterion diffuses faster than the other counterion, the excess
flux produces an electrical field. This field slows down the faster counterion
and speeds up the slower counterion. The net result is the two fluxes (in
equivalents) will be equal, and Eq. (9-32) is satisfied
Combining Eqs. (9-30) to (9-32) and removing <1>, the result is
(9-33)
This can be considered as
(9-34)
where the effective diffusivity is given by the term in brackets in Eq. (9-33).
Note that the effective diffusivity is concentration dependent. If CA <: CB then
D ef:f =D A while if CB <: CA. D ef:f =DB. Thus the ion of lowest concentration
controls the transfer.
As the ionic mobilities or diffusivities approach each other there is less

effect of the electrical field. If the diffusivities are equal, then D ef:f =D A = DB.
Typical diffusion constants in gel type resins are on the order of 10-7 cm 2 /s
Some values are given by Vermeulen et ai, (1984). For small concentration
changes the effective diffusivity will be approximately constant.
In solutions for packed beds we often use a lumped parameter expression

such as Eq. (6-50) instead of solving the diffusion equation for the particles. If
the effective diffusivity is approximately constant, ~ will be constant and Eq.
(6-50) can be applied to ion exchange. In this case the solutions obtained for
adsorption in Chapter 8 also apply to ion exchange. The value of ~ needs to
be calculated from Eq. (8-19a) with kc determined from Eqs. (8-15) or (8-16)
489
and ~ from Eq. (8-17a). For ion exchange D mp in Eq. (8-17a) is replaced by
D off from Eqs. (9-33) and (9-34).
If the effective diffusivity is not constant, then k.n will be concentration
dependent. Unfortunately, this is often the case. For instance, if an ion
exchanger is exchanging ions with different diffusivities, the rates (or k.n
values) for loading and elution will be different Exchange rates are faster
when the faster species is initially on the resin (Helferrich, 1962). Theoretical
methods for these cases are beyond the scope of this book.
9.8. EQUIPMENT DESIGN CONSIDERATIONS
The design of packed bed ion exchange columns is similar to that of adsorption
columns, except the large volume changes which may occur must be designed
for (see Table 9-1). The simplest way to allow for expansion and contraction is
to leave free space above the packing. When the resin swells, it moves up to
occupy the free space. For easy separations such as water softening this
approach works well and is inexpensive. The extra mixing which occurs in the
free space and because of resin movement is relatively unimportant. The free
space can either have a water or an air dome (Anderson, 1979). Water dome
operation is simpler, but involves more mixing. Some free space is also useful
for backwashing the column. If the feed step is done with downward flow, the
column will fluidize when backwashed. This is useful for removing suspended
solids and other crud which will tend to clog the column.
Distribution is very important to prevent channeling, and to minimize
dilution. If the feed mixture is denser than the wash liquid, downard flow of
the feed with a water dome works well. The denser feed will flow from the dis-
tribution pipes and layer directly above the resin. There will be minimal mix-
ing of the feed with the liquid in the water dome. The distributors should be 1
to 2 inches above the resin when the resin is fully expanded. Feeds which are
less dense than the wash liquid will mix with the liquid in water dome opera-
tion causing significant dilution. Air dome operation is used to minimize dilu-
tion. In an air dome the liquid layer is kept 1 to 2 inches above the resin layer.
This requires continual adjustment since the resin bed moves up and down.
Since it is much simpler to operate, water dome operation is more common.
490
For more difficult separations a free space may be detrimental. The

column expansion of the resin can be accommodated for with floating heads or
with balloons (Dorfner, 1972).
Typical beds are from 1 to 3 meters long although special short beds a
few cm long using fine mesh resins are available (Eco- Tech, 1983-84; Brown
and Fletcher, 1988). The wider the bed the more important fluid distribution is.
Once resin swelling has been designed for, scale-up is straightforward. The
period of cycles may have to be greater than some minimum time such as 20
minutes to allow the resin time to swell and shrink during the cycle. Since the
heat effects balance in an exchange reaction, large systems can be designed
directly from laboratory data. If care is taken to have good liquid distribution,
the large scale system will often preform better than the lab unit
For chromatographic separations of biologicals large volume changes are

usually not observed. Resins with large pores (- 1000 'A for proteins) are
required. These resins are often compressible and special designs are used
(Janson and Hedman, 1982).

At the end of this chapter you should be able to meet the following objectives.
1. Define the terms used in ion exchange.
2. Derive the equilibrium expressions from the mass action law. Explain the
difference in equilibrium between monovalent exchange and divalent -
monovalent exchange.
3. Use ion movement theory for binary ion exchange. Determine the effects
of changing total ion concentration.
4. Describe and use ion movement theory for the following processes:
a. Water softening
b. Demineralization
c. On-off ion exchange chromatography

491
5. Use both ion movement and dispersion models for ion exclusion.
6. Discuss the effect of electrical fields on mass transfer. Apply the solutions
of Chapter 8 to ion exchange when the effective diffusivity is constant.
REFERENCES
Anderson, R.E., "Ion-exchange separations" P.A. Schweitzer (Ed.), in Hand-
book of Separation Techniques for Chemical Engineers. McGraw-Hill, NY,
1979, Section 1.12.
Anon., "Hard water may lower heart disease risk," Chem. Eng. News. 30, (Sept.
24, 1985).
Arden, T.V., Water Purification by Ion Exchange. Plenum, New York, 1968.
Aris, R. and N.R. Amundson, Mathematical Methods in Chemical Engineering

Vol. 2, First Order Partial Differential Equations with Applications, Prentice
Hall, Englewood Cliffs, NJ, 1973.
Bailly, M. and D. Tondeur, "Two-way chromatography. Flow reversal in non-

linear preparative liquid chromatography", Chem. Eng. Sci., 36. 455 (1981).
Brown, CJ. and CJ. Fletcher, "The Recofto short bed ion exchange process,"
in M. Streat (Ed.), Ion Exchange for Industry, Ellis Horwood Ltd., Chichester,
England, 1988, 392-403.
Calmon, C. and H. Gold, (Eds.), Ion Exchange for Pollution Control, CRC
Press, Boca Raton, FL, 1979.
CalmoD, C., "Specific ion exchangers" in C. Calmon and H. Gold (Eds.), Ion-
Exchange for Pollution Control, Vol. II, CRC Press, Boca Raton, Florida,
1979,151.
492
Clifford, D., "Multicomponent ion-exchange calculations for selected ion

separations," Ind. Eng. Chem. Fundam., 21, 141 (1982).
Curling, I.M., I.H. Gerglof, S. Ericksson, and I.M. Cooney, "Large scale pro-
duction of human albumin by an all-solution chromatographic process", Ioint
meeting of the 18th Congress of the International Society of Heamatology and
the 16th Congress of the International Society of Blood Transfusion, Montreal,
Quebec, Canada, Aug. 16-22, 1980.
Dechow, FJ., Separation and Purification Techniques in Biotechnology,

Noyes, Park Ridge, N.J., 1989.
Dorfner, K., Ion Exchangers, Principles and Applications, Ann Arbor Sci.
Pub., Ann Arbor, MI, 1972.
Eco-Tec, Ltd., "Eco-Tee ion exchange systems", 1983 and "Recofto - A break
through in water deionization systems," 1984,925 Brock Rd. South, Pickering
(Toronto), Ontario, Canada, LIW 2X9.
Faust, S.D. and O.M. AIy, Adsorption Processes for Water Treatment, Butter-
worths, Boston, 1987, Chapt 10.
Helfferich, F., Ion Exchange, McGraw-Hill, NY, 1962.
Helfferich, F. and G. Klein, Multicomponent Chromatography, Marcel Dekker,

NY,1970.
Janson, I.-C. and P. Hedman, "Large scale chromatography of proteins," in A.

Fiechter (Ed.), Advances in Biochemical Engineering, Vol. 125, Chromatogra-
phy, Springer-Verlag, Berlin, 1982,43-99.
Kunin, R., Elements of Ion Exchange, Reinhold, NY, 1960.
Marcus, and Howery, Ion Exchange Equilibrium Constants, Butterworth, Lon-

don, 1975.
493
Miller, S.A., J.D. Darji, and A.W. Michalson, "Liquid-solid systems - ion
exchange and adsorption equipment," in R.H. Perry and D. Green (Eds.),
Perry's Chemical Engineer's Handbook, 6th ed., McGraw-Hill, New York,
1984, 19-40 to 19-48.
Naden, D. and M. Streat, (Eds.), Ion Exchange Technology, Ellis Horwood

Ltd., Chichester, England, 1984,563-578.
Neuman, R.P., S.R. Rudge, and M.R. Ladisch, "Sulfuric acid-sugar separation
by ion exclusion," Reactive Polymers, 5,55 (1987).
Rodrigues, A. (Ed.), Ion Exchange: Science and Technology, Martinus Nijhoff

Publishers BV, Dordrecht, The Netherlands, 1986.
Sherwood, T.K., R.L. Pigford, and C.R. Wilke, Mass Transfer, McGraw-Hill,
NY, 1975, Chapt. 10.
Streat, M. (Ed.)., Ion Exchange for Industry, Ellis Horwood, Chichester, Eng-
land, 1988.
Streat, M. and F.L.D. Cloete, "Ion exchange," in R. Rousseau (Ed.), Handbook

of Separation Process Technology, Wiley, New York, 1987, Chapt., 13.
Tondeur, D. and M. Bailly, "Design methods for ion exchange processes based
on the equilibrium theory," in A. Rodrigues (Ed)., Ion Exchange: Science and
Technology, Martinus Nijhoff Publishers BV, Dordrecht, The Netherlands.
1986,147-198.
Vermeulen, T., G. Klein, and N.K. Hiester, "Adsorption and ion exchange," in
R.H.Perry and C.H. Chilton, (Eds.), Chemical Engineer's Handbook, 5th ed,
McGraw-Hill, NY, 1973, Section 16.
Vermeulen, T., M.D. LeVan, N.K. Hiester, and G. Klein, "Adsorption and ion
exchange," in R.H. Perry and D. Green (Eds.), Perry's Chemical Engineer's
Handbook, 6th ed., McGraw-Hill, NY, 1984, Section 16.
494
Wankat, P.C., Large-Scale Adsorption and Chromatography, CRC Press, Boca

Raton, FL, 1986.
Wheaton, R.M. and W.C. Bauman, "Ion exclusion. A unit operation utilizing
ion exchange materials," Ind. Eng. Chern., 45, 228 (1953).
HOMEWORK
AI. As in any field, ion exchange has its own jargon. Define the following
terms.
a) counter-ion
b) co-ion
c) electroneutrality
d) Donnan exclusion
e) cation
t) anion
g) strong and weak exchangers
h) selectivity coefficient
i) water softening
j) demineralization
k) mixed beds
1) water dome and air dome operations
01) ion exclusion
n) effective diffusivity
A2. Write the reactions analogous to Eqs. (9-1) and (9-4) for anion
exchange.
A3. Table 9-2 gives KAB and KDB with selectivities compared to Li+ and
a-. Explain how to determine selectivities with respect to other
cations or anions. For example, what is KDB for Ca++ - Na+?
A4. In Example 9-1 and Figure 9-4 the shock and diffuse waves will inter-
sect if the column were longer. What happens to the peak concentration
495
when this occurs? If the column is very long what is the ultimate peak
shape?
A5. Why must cation and anion exchange resins in a mixed bed be
separated before regeneration?
A6. The equilibrium expressions for monovalent-monovalent, and divalent-

monovalent exchange were derived. What does the equilibrium expres-
sion for divalent-divalent exchange look like?
A7. In binary ion exchange if one ion has a diffuse wave the other ion must
also have a diffuse wave. The same is true for shock waves. Explain
why this is true.
A8. Explain the differences between size exclusion chromatography and ion
exclusion chromatography.
C. Derivations
Cl. Equilibrium expressions for monovalent ion exchange and for divalent
ion exchange are derived in the text Derive the equilibrium equation
corresponding to Eqs. (9-11) and (9-15a) for trivalent-monovalent ion
exchange with the following reaction:
C3. Derive the result that YD,. == YD,b when a total ion wave passes through a
column undergoing divalent-monovalent exchange.
C4. For divalent-divalent ion exchange show that the equilibrium expression
is the same as for monovalent-monovalent ion exchange (that is Eq.
(9-11» if KAB is constant
496
C5. For divalent-monovalent exchange show that Eq. (9-16b) is correct
C6. Derive the equilibrium expression for the general ion exchange reac-
tion,
m~+ + nR;Bm+ + mnA = MR; + nBm+ + mnX-
where m and n are positive integers.
01. A strong base (fype II) ion exchange resin has a resin capacity of CRT
= 1.2 eq/L. The total ionic concentration of the solution is 2.0 eq/L.
Generate the eqUilibrium curves for
a. C~ , OIr exchange
b. S04"2 , NO) exchange
Use the approximate selectivity constants in Table 9-2.
02. A strong base resin is being used to exchange NO) with cr. The resin
capacity is CRT = 1.25 eq/liter. The column is initially in the NO) fOnTI.
At time t = 0 minutes a feed which has XNo, = 0.6 enters. This contin-
ues for 10 minutes. Then a feed with xNO, = 0 enters. The total ionic
concentration is ~ = 0.85 eq!liter throughout the process. The column
is 0.55 meter long. Superficial velocity is 12 cm/min. Ee = 0.4. Predict
the outlet concentration profiles for NO) and for cr. Use Table 9-2 for
equilibrium data. Use ion movement theory.
03. A strong acid ion exchanger is exchanging Ni+2 and W. Initially CT =

0.1 eq/L and xNi = 0.2. Displace this with a fluid with ~ = 5.1 eq/L and
XNi = 0.2. Column is 75 cm long and superficial velocity is 10 cm/min.
Predict the outlet concentration profile using ion movement theory.
Data: CRT = 2.1 eq/L, Ee = 0.4, KE = 1.0, and selectivity constants are in
Table 9.2.
497
D4. A strong acid resin is being used to exchange Ca++ and Mg++. The
resin has a capacity of CRT = 2.0 eq/liter. The column is initially in the
calcium fonn. The column is 1 meter long, has a superficial velocity of
10 cm/min, and Ee = 0.39. The fluid has a total ionic concentration of
~ = 1.0 eq/liter which is constant At t = 0 a feed which has an
equivalent fraction of 0.95 Mg++ and 0.05 Ca++ is introduced into the
column. This continues until t = 30 min when the feed becomes 0.05
fraction Mg++ and 0.95 Ca++. Use equilibrium from Table 9-2. Predict
the concentration profile for calcium using ion movement theory.
D5. Lapidus and Rosen (Chern. Eng. Prog. Syp. Ser. 50 (#14), 97 (1954))
studied the breakthrough of Na+ on Dowex 50X4 resins initially in the
W fonn. For Na+ concentrations> 120 x 10-3 meq/ml constant pattern
behavior was observed. Equilibrium data could be fit by the equation n
= 60.6c/(1 + 60.6c) where c = meq Na+ in solution/ml and n = meq Na+
on resin/mt. The mass transfer expression was
dn •
- =kf~ (c-c )
dt
where kf~ was measured as 3.0 ml/(min)(meq) for 50 to 100 mesh par-
ticles (dp = 0.446 mm). With a flow rate of 11.0 ml/min in a 15 cm
long, 1.4 cm ID column containing 11.65 grams of packing, cleF = 1/2
when 335 ml had passed through. Feed concentration CF = 173 X 10-3
meq/ml. Ee = 0.4.
a Calculate L MrZ •
b. Determine the shape of the constant pattern.
Use the Constant Pattern equations, but be careful with your units.
D6. The use of displacement chromatography in ion exchange chromatogra-

phy can be quite advantageous. The displacing ion can often be
removed with little or no diffuse wave by increasing CT, This high con-
centration buffer can then be removed with a wash solution. We wish
498
to separate a feed containing Na+ and NHt by displacement chromatog-

raphy. The column is initially preequilibrated with W. The displacing
ion is Ag+. Fluid concentration throughout the cycle is Or = 004 eq/L.
A strong acid resin with CRT = 2.1 eq/L is used. The feed is introduced
for 10 minutes. The feed is 40 mole % Na+ and 60 mole % NHt. Co-
ion is C1-. The column is 75 cm long, has Vsuper = 20 cm/min and
ee = 0.39.
Predict the outlet concentration profiles. Use Table 9-2 to calcu-
late selectivities. See Section 8.5.1. for a description of displacement
chromatography.
D7. A strong-acid resin bed is saturated with an Ag+ solution XAG = 1.0 at
CT = 004 eq/L. We wish to remove this Ag+ solution with a concen-
trated acid solution (W) with CT = 2.5 eq/L and XH = 1.0. Predict the
outlet concentration profile of Ag+ in equivalents/liter. Information:
Co-ion is C1-, vsuper = 20 cm/min, e,. = 0.39, CRT = 2.1 eq/L, Column
length = 75 cm. Use Table 9-2 to predict selectivities.
Note: This is the column regeneration step for the displacement

chromatography system in Problem 9-D6. However, the two problems
can be solved independently.
D8. Neuman et al (1987) studied the ion exclusion chromatography separa-

tion of glucose and sulfuric acid on Amberlite IR-1l8 in the W form.
At 55°C they determined the following data for glucose: vsupcr = 0.61
cm/min, L = 61 cm, Vcol = AcL = 309 cm 3 , Ee = 00401, X = 0.766, ~.NC
= 0.345, (Dm + ED) = 0.672 cm2/min, VF = 1204 cm 3. (The units in
Neuman et al. (1987) are somewhat different than the units in this
chapter, and the reported values have been adjusted accordingly). Use
the Lapidus and Amundson linear dispersion model to predict the glu-
cose outlet concentration. Compare your result with Figure 9-8a.
Chapter 10
MOVING BED AND SIMULATED MOVING BED

SORPTION SEPARATIONS
In most chemical engineering unit operations continuous counter-current flow

is used since it is usually the most efficient way to operate. This is the case for
equilibrium staged separations. Unfortunately, counter-current movement of
solid and fluid is difficult to achieve without extensive mixing of the solid parti-
cles. This difficulty has catalyzed considerable research in this area. Counter-
current schemes have included flow in open columns, the hypersorption process
where solids flow was controlled by the opening and closing of holes in sieve
trays, moving belt schemes, countercurrent flow of gas and solids in sieve tray
columns with downcomers, pulsed operation in sieve tray columns without
downcomers, and magnetically stabilized moving bed systems. The idealized
analysis of all these systems will be similar.
In some cases counter-current movement has been abandoned, and simu-

lated counter-current flow has been used. The simulated counter-current sys-
tem uses a series of packed beds and moves the location of all feed and product
ports. A close approximation to continuous counter-current movement can be
obtained with the simulated systems.
10.1. SINGLE SOLUTE RECOVERY: STAGED SYSTEMS
Three different types of staged systems with countercurrent flow of solids and
fluid are currently used commercially for removal of one solute. Since the bed
on each stage is fluidized, particulates in the feed will pass through without
clogging. Sieve tray columns with downcomers are used commercially for sol-
vent recovery from dilute air streams using activated carbon (Anon, 1977;
Wankat, 1986). An example is shown in Figure 10-1. The solid is fluidized on
499
500
~ Adsorbent How
~GasHow
[
i[
[
AdSOQJ110r. <
,
l Steam 10r
- Gas Hft hot-!
f
_~~ Preheat ~ng tube
ho",09
i
De"QrpllQn j
~ct!or; ~
/
1 t
Recovered
solvent
Conden!iate
Figure 10-1. Staged counter-current adsorber with downcomers (Anon,

1977), Reprinted with special permission from Chern. Engr. 39,
(Aug. 29, 1977). Copyright 1977, McGraw-Hill.
each stage and flows like a liquid on the stage and into the downcomer. These
systems were first commercialized in the 1950's, but were plagued by exces-
sive attrition and loss of the carbon. A special very hard, spherical activated
carbon is now used and it is claimed that this prevents excessive attrition.
Stage efficiencies as high as 90% have been reported in commercial units.
The second type of staged system (called a Cloethe-Streat contactor)

used commercially uses inlermittanl operation of sieve tray columns but
without downcomers (Cloethe, 1984; Dechow, 1989; Naden and Streat, 1984;
Slater, 1981; Streat, 1980; van der Wiel and Wesselingh, 1989; Wankat, 1986;
SOl
a t b c +
.t~~ •.
'l~ - -
-
t.:-. -: t ~.
Lb --
F.~.~·
-
y-.-
-
t ~
Figure 10-2. Intennittant counter-current flow of solid and fluid. a. Solid
fluidized while fluid flows. b. Fluid flow stops while solid set-
tles. c. Pulse of fluid carries solid down to next stage.
Wesselingh and van der Meer, 1986). These systems are used for large-scale,
continuous, countercurrent ion exchange for recovering metals and could be
applied in biotechnology. During upflow of the liquid the solids on each stage
are fluidized. To transfer the solids down the fluid flow is stopped, the solids
are allowed to settle, and then a pulse of fluid is used to push the solids down.
This type of operation is illustrated in Figure 10-2. The time dependence of
concentration in a four stage Ooethe-Streat contactor is shown in Figure 10-3
(Dodds et al, 1973). Each stage repeats the same concentration profile each
cycle (this is a cyclic steady state.) The operation is not continuous, but closely
approximates continuous countercurrent operation. The number of stages
required or the separation obtained can be estimated from staged models such
as a McCabe-Thiele analysis. A variety of other intennittent, pulsed movement
systems with packed beds during liquid flow are used for continuous ion
exchange (see the next section).
One advantage of the pulsed, intennittent flow contactors is solids and
liquid flows are decoupled. Liquid flow rates are chosen to fluidize the solid
resin without eluting it from the bed. For loading columns (columns receiving
the feed) linear fluid velocities in the range from 15 to 23 m/hr are used com-
mercially in uranium processing (Cloethe, 1984). Any desired flow rate of
502
0019
3 5
STAGE 3
c 3
00240~------~~~--------*"--------~30S
c 2
00340~----------;1~0----------2';;'0;----------~35
t
,,0040: 51AGE -' ~

...... ~--------c'=---------_='=,___
- ~-
u_____
0.0391- *-
I
________
o 10 t 20 305
Figure 10-3. Concentration profiles for a four-stage intermittent fludized bed

ion exchanger. (Dodds et ai, 1973). Reprinted with permis-
sion from Chern. Eng. Sci .• 28, 1233, (1973). Copyright 1973,
Pergamon Press.
solids can be obtained by adjusting the time for liquid flow in step A of Figure
10-2. Ratios of liquid flow rate/solid flow rate as high as 300 are commonly
used in the uranium mining industry. Such a large ratio would be very difficult
to maintain in a continuous countercurrent system. These systems can also
treat feeds containing up to 10% suspended solids. Design details and
503
hydraulic considerations are discussed in section 10.5 and in more detail by

Wesselingh and van der Meer (1986).
The third type of staged system uses mixers followed by filters, screens,
or settlers to separate the resin and liquid. These systems are used for waste
water treatment with powdered activated carbon where only one or two stages
are required. Agitated staged systems are also used in processing of gold and
uranium ores where the feed can contain up to 40% finely suspended solids
(Bansal et al, 1988). In this latter case large resin particles are used so that the
resin is easily separated from the suspended solids. Similar systems are used
for the batch ion exchange purification of biologicals such as proteins
(Dec how , 1989). The stirred tank does not clog with particulates such as cell
debris. After the resin is loaded, it is often transferred to a column for gradient
elution of the desired solutes. Treybal (1980) discusses agitated tank adsorbers
in considerable detail.
All of these staged systems can be designed using the procedures used
for other countercurrent staged systems. If the system is isothermal, the design
is conveniently done on a McCabe-Thiele diagram which is very similar to the
McCabe-Thiele diagrams for absorbers and strippers. (e.g. Brian, 1972; Ruth-
ven and Ching, 1989; Treybal, 1980; Wankat, 1988) A steady-state staged
adsorber is shown schematically in Figure 10-4. Since the flow rate of clean
adsorbent plus fluid in the pores, S kglhr, is constant and the flow rate of carrier
gas or non-adsorbed solvent, G kglhr, will often be constant, it is convenient to
use mass or mole ratio units. Then solute mass ratio on the adsorbent is q kg
solute/kg clean adsorbent plus pore fluid, and in the fluid the mass ratio is Y kg
solute/kg solute-free fluid. In these units the adsorbent flow rate and the carrier
gas or solvent flow rate are constant
(10-1)
With these constant overall flow rates, the solute mass balance for the balance
envelope shown in Figure 10-4 is
(10-2)
Sqo + GYj+1 = GY 1 + SC}j
This balance includes solutes in the pores with q. Thus eqUilibrium must
504
G,'Yf
p, T constant
Figure 10-4. Schematic of staged adsorber.
include the solute in the pores with the amount of solute adsorbed which is
essentially a single porosity model. The ratios q and Y can also be in units of
moles solute/kg without changing Eqs. (10-1) to (10-3). Equation (10-2) is
easily put in the form of an operating line by solving for Yj+1'
Y .... 1
J'
= -GS <J.i
S
+ (Y1 - -G qo)
(10-3)
This is the equation for a straight line since S/G is constant The McCabe-
Thiele diagram is now easily developed on a Y vs q diagram. This is shown in
Figure 10-Sa. Note that the equilibrium curve requires both isobaric and isoth-
ermaloperation. Stage efficiencies can be included in the analysis.
The loaded solid from the adsorber is then sent to a regenerator. Since
less solute is adsorbed at higher temperatures and lower pressures, the regen-
eration is done by operating at higher temperature and/or lower pressure. Thus
the mass balances are given by Eqs. (10-1) and (10-2), and the operating line is
again Eq. (10-3). Only the concentrations and the equilibrium curve change.
The result is shown in Figure lO-Sb. The outlet mass ratio Y 1 from the regen-
erator is usually much higher than the inlet mass ratio into the adsorber. For
505
1j ,p
YI
~----------------------~---------
q
Figure 10-5. McCabe-Thiele diagram for adsorption. a. Loading step. b.
Regeneration step.
506
gases, this high concentration allows recovery of the solute by condensation or

absorption. If the solute isn't valuable the concentrated stream can be
incinerated. The adsorbent from the regenerator is returned to the adsorber.
Note that the combination of an adsorber and a regenerator is similar to the
combination of an absorber and a stripper in a gas treating plant.
If the equilibrium curve is straight in either the adsorber or regenerator,

then the analysis can be done with the Kremser equation. This analysis is simi-
lar to that used for absorbers or extractors and is illustrated in Example 10-1.
More details are given by Wankat (1988), or Brian (1972).
Fluid flow rates in regenerators are usually much lower than in adsorbers.
With very low gas flow rates the stages will not fluidize properly. Thus regen-
erators usually use compact moving beds such as those shown in the bottom of
Figure 10-1. For liquid systems the intermittent fluidized bed system shown in
Figure 10-2 is commonly used for the loading step in ion exchange. The regen-
eration is usually done in a compact moving bed. The staged model and
McCabe-Thiele diagrams can be applied to compact moving beds if an HE1P
is measured. Then the length of moving bed required is
(10-4)
(N) (HE1P) =h
Although this method is commonly used, a theoretically more satisfying
approach is to use the mass transfer analysis discussed in Section 10.2.
For ion exchange and exchange adsorption the systems will be isother-
mal. Stage-by-stage or iterative matrix methods are easily employed for these
isothermal systems. If the selectivity Kij is constant, multicomponent systems
are easily analyzed using a non-iterative algorithm (Wesselingh and van der
Meer, 1986) which is the same algorithm used for distillation with constant
relative volatilities (e.g. see Wankat, 1988, Chapter 8). If the operation is not
isothermal, then the simple analyses shown in Figures 10-5a and b are not
applicable. In this case energy balances must be included. An exact stage-by-
stage calculation requires a trial-and-error solution which will be similar to
those for non-isothermal absorbers (Treybal), 1980; Smith, 1965). Alterna-
tively, the matrix approach used for absorbers can be applied to staged
adsorbers (e.g. Wankat, 1988, Chapt 15).
507
a b
,
q
x
Figure 10-6. Batch staged sorption. a. Apparatus. L,xo and S,qo are initial
charges. b. McCabe-Thiele solution.
McCabe-Thiele diagrams can be used to predict the fin~ equilibrium

conditions for batch stirred tanks. Consider the system shown in Figure 10-6a.
L is the charge in kg of liquid to the tank of initial mass fraction x kg solute/kg
solution and S is the charge in kg of resin of initial mass ratio qo kg solute/kg
resin. Then the mass balance is,
(10-5)]
Lxo + S qo = Lx + Sq
Solving for q, we obtain
L L (10-6)
q=--x+Qo+-Xo
S S
This is a straight operating line as plotted in Figure 1O-6b. Mter a long period
of operation, equilibrium will be reached. Then the final values, <ltin and Xfin
are at the intersection of the operating line and the equilibrium curve (see Fig-
ure 10-6b). If mass transfer is slow and equilibrium is not reached, a mass
transfer rate analysis is required. Equations (10-5) and (10-6) and Figure 1O-6b
are also applicable to a continuous steady state stirred tank if L and S are flow
rates.
Example 10-1. Staged analysis
We wish to sorb glucose onto an ion exchange resin in a staged contac-

tor. Feed concentration is 11g/L. We wish an outlet concentration of
508
1.0 gIL. Inlet resin contains 0.25 g glucose/L. The ratio of volumetric
flow rates of resin/solution is 2.5 (L resin/hr)/(L solution/hr). Find the
number of equilibrium contacts required.
Solution
We will illustrate the solution using both the McCabe-Thiele
method and the Kremser equation. From Problem 6-04 equilibrium is q
= 0.51 c where q and c are in gIL. Assuming that volumetric flow rates
of resin, V R L/hr, and of solution, Vs L/hr, are constant, we obtain the
mass balance
(10-7)
for the column shown in Figure 10-4. This leads to the operating equa-
tion,
(10-8)
where VRNs = 2.5, qo = 0.25, Cl = 1.0. Point (qo, Cl) is on operating

line. Equilibrium is c = 1.961 q.
Staged Solution
The equilibrium and operating lines are shown in Figure 10-7. Step off
stages. Need 7 equilibrium stages.
Kremser Solution
The Kremser equation is easily applied to this problem since
operating and equilibrium lines are straight. Many forms of the
Kremser equation are available (e.g. see Wankat, 1988). In the units of
509
10
c, giL 6
o~--~--~--~--~----~~
qlN 2 3 4 5 6
q, giL
this problem the Kremser equation for the number of equilibrium con-
tacts N is (modification of Eq. (15-30) from Wankat, 1988),
N = In [ rII ---
m Vs]
V R
[CN+1- c.7 ] + -
Cl -Cl
m Vs
-
VR
1/ [In -
R]
V-
Vsm
(10-9)
where m = c/q = 1.961 and c; = mqo = (1.961)(0.25) = 0.490. Then
In (I - 0.7844) [ II~~~~4": ]+ 0.7844

N= :-; 6.81
In (1.2749)
Generalization. Within the accuracy of the graph the two solutions

agree. Varioos fonns of the Kremser equation can be used. The
Kremser equation is much more convenient when N is specified
because outlet concentrations can be calculated directly whereas the
510
McCabe-Thiele solution will be trial-and-error. Note that the Kremser

equation requires constant flow rates and linear equilibrium while the
McCabe-Thiele solution can solve problems with non-linear equili-
brium. The regeneration step is explored in Problem 10-D5.
10.2. MOVING BED SYSTEMS
Several types of moving bed systems are commonly used commercially. With
granular activated carbon for waste water treatment pulsed beds are often used.
An example is shown in Figure 10-8a (Storm, 1981). The bed acts as a packed
bed during the upward liquid flow step. Then intermittently, the solid is pulsed
downward and part of the carbon is removed and sent to a kiln for regeneration.
The solute movement theory for operation of a pulsed bed is illustrated in Fig-
ure 1O-8b. The feed is stopped (time fp) before breakthrough. The carbon
removal pulse has a length Ip. This pulse moves the mass transfer zone back to
the starting position, and the feed is started again. This requires that
(10-10)
where fp is the feed period between pulses. Note that at time fp all of the car-
bon removed is completely saturated. Since the pulsing introduces additional
mixing, pulses are usually quite small. Typically, L/lp is about 20. The result-
ing operation is then very close to a continuous countercurrent contact. The
pulsed beds can usually be operated at considerably higher flow rates (5 to 9
gpm/ft2) than beds in series (3 to 7 gpm/ff) or single beds (1 to 4 gpm/ft2)
(Hutchins, 1981). Thus, although they are somewhat more complex than a
packed bed, the pulsed bed systems can be smaller and cheaper.
The Asahi system shown in Figure 10-9 (Arden, 1968; Dechow, 1989;
Rodrigues, 1986; Slater, 1980; Streat, 1981; Wankat, 1986) is commonly used
for ion exchange systems. During the feed step (Figure 10-9a), the upward
flowing water pushes the resin against the top grid and a dense packed bed is
formed. Treated water exits at B, and ball valve D is pushed shut which
prevents addition of more resin. A small amount of feed water is used to push
resin out of cone C into the resin hopper for the next unit. To pulse the resin
511
a ~
~
CHARGE CHAMBER
(TYPo'
SCREENED ~
~ ~
OVERFLOW .......... EFFLUENT TAKEoOFF
~-=~~....- ____ -. HEADER
TYPICAL INFLUENT I
EFFLUENT SEPTA -T"'P",_
VESSEL CONSTRUCTION
316SS CARBON STEEL
W/RUBBER OR EPOXY
LINING
NOZZLE (TYPo'
FEED INLET - - - 1 ...... -,-----"'1

+. . .
- - I N L E T HEADER
L
WATER INLET
PULSE TRANSFER
T b
Adsorb
Desorb
o tp
Figure 10-8. Pulsed bed system. a Equipment for activated carbon waste-
water treatment (Storm, 1981). Reprinted with permission
from J.R. Perrich (Ed.)., Activated Carbon Adsorption for
Wastewater Treatment, 1981. Copyright 1981, CRC Press. b.
Solute movement diagram.
512
Resin to
succeeding unit
(a) (b)
Figure 10-9. Asahi system for moving bed adsorption and ion exchange.
(Alders, 1968). a. Fluid flow step. b. Addition of new resin.
Reprinted with permission.
downwards, valve A is shut (Figure 1O-9b). Resin now drops due to gravity
and valve D opens allowing fresh resin into the vessel. Since a normal ion
exchange cycle consists of feed, regeneration and wash steps; three columns
are required for the Asahi system. Since the three columns have different func-
tions, they are optimized separately. The loading column has a high flow rate
and operates with a slowly moving constant pattern wave. This column can be
wide and short. The regeneration column will have a very low liquid flow rate,
and this column is narrow. Since a diffuse wave occurs during regeneration,
the regeneration column is usually quite long. Washing can be done in a single
mixed tank. The Asahi system has been used extensively for water treatment
and in sugar refining.
An alternate moving bed ion exchange contactor is the Higgins or

Chem-Seps system (Dechow, 1989; Higgins and Chopra 1979; Rodrigues,
1986; Slater, 1981; Streat, 1980; Wankat, 1986) shown in Figure 10-10. All
operations are done simultaneously in the loop. During the liquid flow step the
513
O•• rf1ow
wlte,
Resin
reservOir
section
Treatment
Ie-CUO"
B.ckwu"
Pul ..
Raw water In Pulse section
Treated water out Regen.,.n.

waste out
Rinse watf'r
Rinse
section
Regentr'"t
Resin \teflon
flo ..
Regen"lnt In
Valve poSitions during cycles

Run cycle Pulse cycle
Valves open Valv,s closed Valves open Valves closed
A,WA,IN,EF B,C,D,PU B,C,D,PU A,WA,IN,EF,RI,RE
RI. RE
Figure to-tO. Chem.-Seps semi-continuous ion exchange system. (Higgins

and Chopra, 1979). Reprinted with permission from CaImon,
C. and Gold., H., (Eds.), Ion Exchange for Pollution Control,
Vol. 2. Copyright 1979, CRe Press.
514
resin acts as a packed bed in the treaunent, rinse and regenerant sections, but is
fluidized in the resin reservoir section where the resin is backwashed to remove
suspended solids. The solids movement step uses a counterflow pulse of water
to push the resin. The Chem-Seps system has been extensively used for water
treatment and hydrometallurgical processes.
A variety of other systems have been developed. Regenerators have
quite low liquid flow rates and satisfactory intennittent operation can be
obtained in empty pipes (Wankat, 1986). Pulsed upward flow of the solid
appears to work well. Fluidized beds can be used when affinity is very high,
but there tends to be considerable mixing if the bed is not stabilized
(Lochmuller et ai, 1988). Truly continuous plug flow of solids can be obtained
in magnetically stabilized fluidized beds. (Rosensweig, 1979; Lucchesi et ai,
1979; Burns and Graves, 1985, 1987). In these systems either magnetic parti-
cles are used as the sorbent or a mixture of magnetic beads and a non-magnetic
sorbent are used. The flow of the particles is stabilized with a magnetic field.
The advantages of this system are that true plug flow without mixing can be
attained and particulates in the feed will not clog the bed as they will in packed
or dense moving beds. Unfortunately, the magnetically stabilized systems have
proven to be difficult to scale-up and have not yet been commercialized. This
technology has been applied both to gas and liquid systems and to adsorption
and chromatography systems. It's major use may be in medium-sized equip-
ment for biotechnology. Other designs are reviewed by Slater (1981), Streat
(1980) and Wankat (1986).
The countercurrent moving bed systems can be analyzed using a mass

transfer analysis. This analysis is more realistic than a staged analysis when
the system has continuous contact between solid and fluid. Since the mass
transfer approach for moving bed systems is similar to the mass transfer
analysis applied to vapor-liquid systems (e.g. see Treybal, 1980; or Wankat,
1988, Chapter 19), only an abbreviated development will be given here. The
analysis will assume flow is continuous, and thus will ignore the discrete nature
of most of the contactors used commercially. Other presentations of this
analysis are given by Treybal (1980) and Wankat (1986). Burns and Graves
(1985) use different forward and reverse rate coefficients. Ruthven (1984)
includes axial dispersion and gives the solution for linear isothenns.
515
The column is shown schematically in Figure lO-11a and is assumed to

be isothermal. The fluid phase flows at a constant flow rate L kglhr and flows
countercurrent to the solid phase with a flow rate S kg/hr. For a differential
element of height dz the mass transfer expression is
(10-11)
where x is the adsorbate mass fraction in the fluid and YE is the fluid concentra-
tion in equilibrium with the solid. The overall mass transfer coefficient K is
based on the interfacial area per unit volume, "P, between solid and fluid. K
can be estimated from a sum of resistances model such as Eq. (8-19a), while 3p
is given by Eq. (8-14) for spherical particles.
Rearranging Eq. (10-11) we have
h xin
h- Jdz - AclCap
L J dy
(x - XE)
(10-12a)
o "out
where we have assumed that L/AclCap is constant. Equation (1O-12a) is often

written as
(1O-12b)
h = (HTU)(NTU)
where
Xin
HTU= AclCap
L
NTU= J dy (1O-12c)
To determine the NTU, (x - XE) must be determined as a function of x.

The equilibrium value is determined from the equilibrium isotherm. The x
value can be found from a mass balance using the mass balance envelope
516
a
YoupL S
,qlN
r- -,
I I b
I I
I I
I I
I I
I I
t-.- ~t~IdZ
I
L Y
Your
Y
Figure 10-11. Mass transfer analysis for moving bed systems. a Schematic
of column. b. Determination of y - YE. c. Graphical integra-
tion to determine NTV.
517
shown in Figure 10-11a. The operating equation resulting from this mass bal-
ance is,
S S (10-13)
X= L q+xout - L <tin
Both the operating fUld equilibrium curves can be plotted on a McCabe-Thiele
type diagram of x versus q. This is shown in Figure 10-11b for adsorption.
The value of (x - XE) is easily detennined as the difference between the operat-
ing and equilibrium curves. The NTU integration can be done graphically,
numerically, or using Simpson's rule. The graphical integration is shown in
Figure 10-11c. Simpson's rule will be
Xin
NTU= J X-XE
dx
=
Xout
[
Xin - Xout
6
1[ x - 1 XE
I
Xin + x-
4
XE
Xin+Xout + _1_
2 x- xE
IXout 1 (10-14)
Additional accuracy can be obtained if more complicated forms of Simpson's
rule are used, or if the integral is divided into parts.
If the equilibrium is linear of the form
(10-15)
x=mq+b
then the integration can be done analytically. This result is known as the Col-
burn equation.
NTU = [
1-
~lIn {(1-
5
mL) [ Xin - (mQm + b)
S Xout - (mQm + b)
1+ mL}
S
(10-16)
The Colburn equation for adsorbers is similar to the Colburn equation

developed for absorbers (e.g. see Wankat, 1988, Chapter 19).
518
To use Eq. (1O-12b) for design the value of the HTU must be known.
This requires a value for Kap. The value of Kap can be determined experimen-
tally or from correlations such as Eqs. (8-15) and (8-17a). In correlations the
velocity should be the relative velocity between the solid and fluid.
This analysis includes fluid in the pores with the solid phase. Thus the
equilibrium isotherms must also include solute in the pore fluid along with the
adsorbed solute. An alternate analysis is to treat the adsorber as a three phase
contactor (solid, pore fluid and external fluid) (Wankat, 1986). In most cases
the analysis presented here will be adequate. The most serious limitation of
this analysis is the isothermal assumption. Non-isothermal systems can be
modeled numerically (Liapis and Rippin, 1979).
Example 10-2. Moving Bed Mfinity Chromatography
Data for moving bed systems is relatively scarce. Burns and Graves
(1987) present the following results for human serum albumin (HSA)
sorption by Cibacron Blue F3GA attached to calcium alginate-
magnetite beads in a magnetically stabilized fluidized bed.
Feed Concentration = 0.2 mg HSA/ml solution
Input solids = 0.0 mg HSA/g solids
=
Fluid flow rate = L 10 mVmin
Solids flowrate = S = 0.5 g/min
L kci
K=S ~= 1.6
where kci and Ie. are desorption and adsorption rate constants.
Percent HSA adsorbed = 56%
Bed Length, h = 8.0 cm
Find NTU and HTU
Solution
At the low concentration given here adsorption will be linear.

Since K = 1.6, kcilk. = (1.6) (0.5/10.0) = 0.08 g/m!. This leads to a
519
Langmuir expression (see Section 6.3), but at low concentrations will

be linear. If we write equilibrium as Eq. (10-15) then b = 0 and m = c/q
=kcilk. =0.08 g/ml.
Also: <lin =0, mLlS = (0.086.~1O·0) = l.6, Cin =0.2,
Cout = (1-0.56) (0.2) = 0.088 mg/g.
We can solve this either graphically or with the Colburn equation. We
will illustrate both. The equations are valid in the units of this problem
if S and L are constant. For these dilute solutions they will be.
Graphical Solution: Plot c versus q as shown in Figure 10-12.

Equilibrium line: c =mq =0.08 q
. l'me: c = L
Ope raung S q + Cout - L
S qin
'
Slope = S/L = 0.05

Goes through point (c out , <lin) = (0.088,0). Stop at cin =0.2 mg/ml.
C'N
0.2 -- I - - ---.-- --,--- ----,--
0.15
Equilibrium
C mg/ml C-C E
0.1
COUT q'N
0.05
o~--~~--~---~--~~-~
o 0.5 1.0 1.5 2.0
q. mg/g

520
In these units Eq. (10-14) becomes
Cin
NTU= J dc
C-CE
Cout
em + Caul
h
were 2 = 0.2 +20.088 =.
0 144
From the graph we can determine values of C - CE. Then
- 0.112 [_1_
NTU - 3
4
0.088 + (0.144 - 0.0896)
+ 1
(0.2 - 0.179)
1=247
.
and
HTU = h/NTU = 811.47 = 3.24 cm
Colburn Equation Solution: Equation (10-16) in these units is:
NTU- [ 1mL
- 1-5
1 {[1 - -1
In mL
S
[cin-(mQm+b)
cout-(mQm+b)
1+ -
mL}
S
(10-17)
521
Using Eq. (1O-12b) we have

HIU = hj(NTU) = 8/2.4 = 3.333 cm
This is a reasonable height which does not indicate excessive backmix-
ing of the resin. Note that the results of the two solutions are essentially
equivalenL
10.3. MOVING BED FRACTIONATION
A counter-current system for fractionating two components is shown in Figure

10-13. The solids flow down the column while the fluid flows up. The less
strongly adsorbed solute A moves up in zone 1 while strongly adsorbed solute
B moves down in this zone. Thus zone 1 purifies solute A. Zone 2 removes
solute A from B and thus purifies solute B. In zone 3 solute B is des orbed with
desorbent D. Zone 4 serves to remove solute A from the desorbent so that
desorbent can be recycled. The desorbent could be water or a solvent.
The solute movement theory can be applied to this system. The solute
wave velocities calculated from Eq. (6-23) or (6-24) were with respect to a sta-
Solids
4
A+D,~
A+B,F
2
B+D,~
3
Fluid
o
Figure 10-13 . Counter-current separator for fractionation of two solutes.
522
tionary solid. The appropriate fluid velocity is then the interstitial fluid velocity
relative to the solid. Thus
vsupa: (10-18)
V= - - + IVlolid I
Ee
where vsupc:r is the superficial fluid velocity and Vso\id is the superficial solid
velocity. Now Us calculated from Eqs. (6-23) or (6-24) is the solute velocity
with respect to the solid The solute velocity which an observer will see is
obtained by subtracting the solids velocity
(10-19)
u..cc = Us - IVsolid I
u..ccis positive when the solute flow is up the column and negative when it
flows down.
In a counter-current column the solids velocity is the same in all zones

but the superficial fluid velocity varies from wne to zone since feed is added
and products are withdrawn. If we set Vsupcr.3 as velocity in zone 3, then for
relatively dilute systems
(10-20)
Vsupcr.l = Vsupcr.2 + F/(PfAc)
where the product and feed flow rates in kg/min are shown in Figure 10-13.
Since vaUpel" changes, u..cc will change from zone to zone. In addition, if the
desorbent affects the adsorption of solute then the equilibrium constant A(T)
will vary from zone to zone and u..cc will change. This latter effect is very use-
ful, but is not necessary to make the counter-current column work.
To achieve the separation indicated in Figure 10-13 we want solute A to

move upward in zones 1 and 2 and downward in zone 4. Thus
(10-21)
UA ce.l , UA ce.2 > 0 > UA ec,4
523
s 4
o
Time
Figure 10-14. Solute movement diagram in continuous countercurrent
column for linear isotherms.
Solute B should move downwards in zones 1 and 2 and upwards in zone 3.

Thus
(10-22)
UB <:<:,3 > 0 > uB ce,l , uB ce.Z
Equations (10-21) and (10-22) are an important result since they control the
operation of the continuous counter-current column. These equations are valid
for both linear and nonlinear isotherms. For nonlinear systems the solute velo-
cities are obviously concentration dependent (see Problem 1O-C5). If the sys-
tem will work, there are ranges of values for P 1 , Pz and D for a given feed flow
rate which will satisfy these inequalities. In actual practice it is desirable to
choose the flow rates so that all the inequalities are as large as possible.
The appropriate solute waves are shown in Figure 10-14. In the ideal
case at steady state there will be no solute A in zones 4, 2 or 3 and no solute B
in zones 1,3 and 4. Because of dispersion and finite mass transfer rates solute
A will appear in zones 4 and 2, and B will be in zones 1 and 3. The size of the
zones required depends on these dispersion and mass transfer rate effects. In
addition, any axial solid or fluid mixing caused by non-perfect flow will require
524
a larger column. Extreme mixing or channeling can destroy the desired separa-
tion.
The solute movement theory does not include mass transfer effects. This
can be done using the steady-state counter-current theory of mass transfer dis-
cussed earlier. A staged analysis can also be used. When there are two adsor-
bates and a diluent. the analysis will be similar to fractional extraction analysis
(e.g. see Brian, 1972; or Wankat, 1988, Chapter 16).
In exchange adsorption (see Figure 6-3e) two adsorbates compete for

sites and no diluent is present. A moving bed apparatus can be used to frac-
tionate these solutes. This was done commercially in the 1950's using the
Union Oil Co. Hypersorption process (Berg, 1951; Treybal, 1980; Wankat,
1986). This separation had difficulty competing with vacuum distillation and is
no longer used. The process was complex both conceptually and mechanically.
Since the process is no longer in use, we will not consider it further (see Prob-
lem IO-Fl for further study).
Currently, there are no announced moving bed systems being used com-
mercially for fractionation. If separation is easy, other separation techniques
such as distillation have been cheaper. If separation is difficult, mixing of the
solids must be minimized. Unfortunately, this has been very difficult. New
developments such as the magnetically stabilized fluidized beds discussed
briefly in Section 10.2 may eventually solve these problems.
10.4. SIMULATED MOVING BED FRACTIONATION
An alternative to moving bed systems is to simulate counter-current movement.

This is done with a series of packed bed sections by switching the location of
all feed and product withdrawal ports. This is illustrated in Figure 10-15. At
time t1 the product port and all other ports are switched. An observer located at
the product port "sees" the solid move down at this time. From time t1 to t2
the fluid flows upward. At time t2 the ports are switched again and the observer
"sees" the solid move downward. The net result is the observer "sees" an
intermittant counter-current motion of solid and fluid.
525
t t t
/~ /// /// Product
f--
/// /// Pro ///
/// Pro duct ///
- duct
///
r---+-
/// /// ///
t t t
tl to t2 t2 to t3
Figure 10-15. Simulated countercurrent system.
The first simulated counter-current system was the Shanks system which
has been applied to leaching, adsorption and ion exchange. The Shanks system
uses a series of columns with plumbing arranged so that feed can be input and
product withdrawals removed from any column. Thus the counter-current
separator shown in Figure 10-13 can be simulated. A modem adaption of the
Shanks process simulating the system shown in Figure 10-13 has been exten-
sively commercialized by Universal Oil Products (UOP) (Broughton, 1984-85;
Dechow, 1989; Ruthven and Ching, 1989; Wankat, 1986; Storti et ai, 1989).
UOP used a pilot plant scale system which is a series of columns for scaling up
their commercial scale units. The commercial UOP simulated counter-current
process, Sorbex, uses a single column with many packed sections and has a
rotating valve for distributing feed, desorbent and products. This is shown
schematically in Figure 10-16. The UOP process was first commercialized as
Molex for separation of linear paraffins from branched-chain and cyclic hydro-
carbons using 5A molecular sieves (see Table 6-1). Since then, processes for
p-xylene purification (parex), olefin separation (Olex), and separation of fruc-
tose from glucose (Sarex) have been commercialized. Pilot plant scale separa-
tions for a variety of other problems have been demonstrated. A large number
of patents have been granted on simulated moving bed systems.
The Universal Oil Products system works very well, but is complex and
expensive. The rotary valve in particular is expensive since it must be custom
526
.----..... A
D
.---..... 8
F
Recycle
Figure 10-16. Universal Oil Products Simulated Moving Bed (SMB) System.
designed and machined. Simpler and hence cheaper systems are also commer-
cially available and are commonly used for separation of fructose from glucose.
One such arrangement is shown in Figure 10-17 (Wankat, 1986). Here there is
one column per zone. Thus, the approach to a truly countercurrent system is
only approximate, but the system is simple. The system shown in Figure 10-17
requires 4 solenoid valves per column. In commercial columns a fifth valve
may be added to allow bacldlushing the resin.
The solute movement theory can be used to analyze the simulated

counter-current system in two ways. First, if the observer fixes himself at one
of the outlet or inlet ports then he sees the solid and entrained fluid transferred
downwards in pulses. This observer then sees solids movement in pulses. The
avelage solids velocity this observer sees is given by Eq. (10-10) and the
analysis procedure is essentially the same as for a pulsed counter-current sys-
tem (see Figure 1O-7b).
In the second analysis the observer fixes himself on the ground and he
sees the solid as stationary. With the fluid flowing up the column, he sees all
the inlet and outlet ports move up the column at discrete times. When a port
*r--~*2---r--*---r*
527
A -+--4 j 3 4
B
5 (\ 7 8
<;) 10 11 12
D ~~--~------7---L------7--~------'-~
F ~)__~*_1_3________*~14_________*~15________~*16
Figure 10-17. Arrangement for 5MB with one column per zone and four
zones. (Wankat. 1988).
Zone Inlet Valve Open Outlet Valve Open
I 13,14,15 or 16 1,2,3 or 4
n none none
III 9,10,11 or 12 5,6,7 or 8
IV none none
Reprinted with permission from P.C. Wankat, Large Scale
Adsorption and Chromatography, CRC Press, 1986. Copyright
1986, CRC Press.
reaches the top of the column, it recycles back to the bottom. In between the
shifting of port locations, the adsorber is a fixed bed system. Thus the solute
wave velocity can be determined from Eqs. (6-23) or (6-24). The fluid veloci-
ties in each section will differ. The superficial fluid velocities are given by Eq.
(10-20) and the interstitial velocity v equals vsuper/Ee. The shifting of ports does
not shift the solute waves, but does change the wave velocities since it changes
528
RECYCLE STREAMS
I 3 14
.L _B Prod_ 3 _ _
I ~-.L
Z 3 1 4 14
2_ _BProd_l_ ~ _1_ ~ _1 ____ 1_
2 3 3 4 I4 II
B Prod l i D 1 1 A Prod
- -I 3-1 --I 4" T~-I~
3_ _I _ D_ _I _ -.L ~ A Pro6Y ~ _ 1_
3 1 4 1 4 117 1 I 12
D I I A Prod 1 A 1 A+ B 1
I--t~ Recycled
Time
Figure 10-18. Solute movement is simulated countcurrent system for linear

isothenns.
which zone the solute is in. This is illustrated in Figure 10-18. If the desorbent
changes the equilibrium constants, this will also change the solute velocities.
Note in Figure 10-18 that the movement of both species is up, but the
more strongly adsorbed solute B moves down with respect to the port locations.
Feed would be introduced continuously at the port marked A + B, but was illus-
trated for only one time period. The zone numbers corresponding to Figure
10-13 are shown on Figure 10-18. The fluid velocities and hence the solute
velocities are different in each zone. When fluid reaches the top of the cascade
it is recycled to the bottom. Thus the solute waves are also recycled. If the
timing of the switches is appropriate, solute A will appear in zones 1,2 and the
lower section of zone 4, and solute B will appear in zones 1, 2 and the upper
section of zone 3. Solute A goes into zone 4 since only a portion of the fluid is
withdrawn as product Pl' Solute B goes into zone 3 because of the switching
of ports. Dispersion and mixing effects will naturally spread out the solute
529
waves in each zone. Figure 10-18 needs to be studied carefully since the simu-
lated countercurrent motion is somewhat subtle.
To achieve the desired separation we desire to have solute A move up in

zones 1 and 2 faster than the port movement and slower than the port move-
ment in zone 4. Solute B should move slower than the port movement in zones
1 and 2 and faster in zone 3. The average velocity of port movement is
(10-23)
where lport is the packing height between ports and fp.m is the time between
switches of ports. The conditions to achieve separation are then
(10-24)
(10-25)
UB,3 > \lport,avg > UB,l , ~,2
These conditions follow the same order as Eqs. (10-21) and (10-22).
How close is a simulated counter-current system to a truly counter-

current separator? Although the answer to this depends on the chemical system
and the column length, Liapis and Rippin (1979) found that the simulated sys-
tem had an adsorbent utilization from 79% to 98% that of the truly counter-
current system. With a single zone system they found that from two to four
sections were sufficient and that two to four column switches were required to
reach a periodic concentration profile. An example of experimental results
from a simulated countercurrent system are shown in Figure 10-19 (Neuzil and
Jensen, 1978). The average profiles are quite smooth and look very much like
the expected shape for a continuous countercurrent system. Because of this
similarity, 5MB systems are occasionally designed as if they were continuous
contact systems, and either the staged or HTU-NTU methods developed previ-
ously can be used. A preliminary design can also be done using a staged
analysis to determined N in each section and then h from Eq. (10-4). This
requires an estimate of HETP. The advantage of this approach is existing mul-
ticomponent calculation methods developed for absorption and extraction can
be adapted to the 5MB. Determination of the number of subsections required
530
LEGEND
o FRUCTOSE
~DEXTROSE
o POLYSACCHARIDES
ZONE ZONE ZONE ZONE

IV I II III
Figure 10-19. Compositions in column for simulated countercurrent system

from Sarex pilot plant. Neuzil and Jensen (1978). Reprinted
with permission.
for each zone requires a detailed time-dependent model (Ruthven and Ching,
1989; Storti et ai, 1989).
Comparison of simulated counter-current and truly counter-current sys-

tems is of considerable interest Both systems at steady state can at best do a
complete binary separation. Partial separation of additional components can be
obtained with side withdrawals. The simulated counter-current system could
also be extended to more complex cycles where part of the bed is temporarily
operated as a batch chromatograph. The simulated moving bed system is actu-
ally a fixed bed system. Thus flooding (unintentional upwards entrainment of
solid) will not be a problem, but excessive pressure drop may result for small
particles or viscous solutions. The fixed bed will have a lower Ee and hence a
higher capacity than truly counter-current systems, but this will be offset by the
distribution regions between sections. In moving beds the actual movement of
solids requires means for keeping the bed stable and may result in excessive
attrition, but allows for easy solids replacement or external reactivation. Both
systems have mechanical difficulties to overcome. In the simulated moving
bed these difficulties are the valving, timing and distributors while in an actual
531
moving bed they involve moving the solids without mixing. Currently, the
simulated counter-current systems have been the preferred choice for large-
scale adsorption installations to fractionate two solutes.
Example 10-3. Fructose-Glucose Separation in 5MB
We wish to use the local equilibrium model to estimate feasible operat-

ing conditions for a 5MB to separate a dilute fructose and glucose mix-
ture using an ion exchange resin in the calcium form. Do this for a 4
zone system with one section per zone. We arbitrarily choose !port = 5
minutes since this is a convenient time. Design for a 5% glucose, 5%
fructose feed at a feed velocity
vp = _F_ = 1.0 em/min.

PfAc
The superficial velocity in zone 3 should be vsuper.3 = 20 em/min. Use

water as the solvent
Solution
A. Define. The 5MB will have the 4 zones shown on Figure 10-13.
We need to find product velocities, superficial velocities in each zone,
Iport, and product concentrations.
B. Explore. This problem is quite open-ended since many different

answers will satisfy inequalities (10-24) and (10-25). We will look at
the most restrictive of these inequalities and use Eqs. (10-20) to deter-
mine values which work. Equilibrium data was given in Problem 6-D4
and u. was calculated in Problem 6-C3.
C. Plan. Calculate Up,3 (solute velocity of glucose in zone 3). Estimate

a port velocity which satifies Eq. (10-25). Then estimate a UPI value to
satisfy Eq. (10-25) and calculate vsupcr,l and vsuper,2' Calculate
532
110.1 and 110.2 and check if Eq. (10-24) is satisfied. Detennine a 110,4
which also satisfies Eq. (10-24). Then detennine vsupc:r,4. The product
withdrawal rates can be detennined from Eqs. (10-20). This procedure
gives a set of feasible operating conditions, and does not give optimum
operating conditions.
D. Do It From the solution to Problem 6-C3 and data in Problem 6-

04,
Vsupcrj vsupc:rJ
upJ = Eo + (1- £0) Ap = 0.4 + (0.6) (0.88) = 1.078 vsupcrj
Vsuperj = vsupc:rj = 1416 v .

Uoj = Eo + (1 - Eo) Ao 0.4 + (0.6) (0.51) . super.)
In zone 3 vsuper.3 = 20 cm/min and UP.3 = 21.56

From Eq. (10-25) 21.56> llport,avg
Arbitrarily, pick llport,avg = 20 cm/min
Then,lport = (uport,avg) lport = (20 cm/min) (5 min) = 100 cm
If we choose UP.l = 18.5, Eq. (10-25) will be satisfied.
Then vsupc:r,1 = 1~~;8 = 17.16 cm/min

vsupc:r,2 = vsupcr.l - Vp = 16.16 cm/min
Up.2 = (1.078 (16.16) = 17.42 cm/min
For glucose: 110.1 = (1.416) vsupc:r,1 = 24.30 cm/min

110,2 = (1.416) vsupc:r,2 = 22.88 cm/min
Eq. (10-24) is satisfied so far, but must also have
Uo,4 < uport,avg = 20

533
Arbitrarily pick 00.4 = 19.0 cm/min. Then
1Ig,4 19.0
vsupcr,4 =- -=
1.416
1416
.
= 1342
. cm/mm
.
Product withdrawals can be detennined from Eq. (10-20).
Pd(PtAc) = VlUpcr.1 - Vsupcr.4 =3.74 cm/min

Pz/(PfAc) = vsupcr,3 - vsupcr,2 = 3.84 em/min
Assuming that fructose and glucose are completely separated, the glu-
cose product is diluted to (S%) (1.0)/(3.74) 1.34%.=
Fructose Product is (S%) (1.0)/(3.84) = 1.30%.
E. Check. A check of solute velocities shows that Eqs. (10-24) and

(1O-2S) are satisfied.
F. Generalization. This is clearly not the only solution and not the only
solution method. To see the effect of some of the arbitrary choices
made in the solution, I tried the following alternate set:
llport,avg = 21, lport = lOS cm, and uF.1 = 20.5
This gives Vsupcr.1 = 19.02, vsuper.2 = 18.02, and UF.2 = 19.42.
For glucose: 00.1 = 26.93, 110.2 = 2S.S

Letting 00.4 = 20.5 then vsupcr,4 = 14.48
Products: Pd(PfAc) = 4.S4; 1.10% glucose
P 2/(PfAc) = 1.98; 2.S3% fructose
This choice will be harder to operate since solute velocities are

significantly closer to the port velocity. If the isotherms have any non-
linearity the initial choice is more likely to work. However, there is less
overall dilution and significantly less fructose dilution with this set.
534
10.5 DESIGN CONSIDERATIONS
Each of the contacting methods discussed in this chapter has different design
considerations. Fluidized plate columns such as the Cloethe-Streat contactor
must operate within certain hydrodynamic limits (Wesselingh and van der
Meer, 1985). The lower limits are the fluid velocity must be greater than the
minimum fluidization velocity, and the velocity in the holes of the sieve plate
must be greater than the settling velocity of the largest resin particles. If we
define
d* =dp/dnt , dnt = [
y2
Pr
]1/3 (lO-26a)
g (pp - Pr)
v
* = vIvref • vref =
[g(Pp - Pr)V ]1/3 (lO-26b)
Pr
where v is the kinematic viscosity and g is the acceleration due to gravity, then
the settling velocity is
(1O-27a)
v~ =0.25d* 3 < d* < 30
and
(10-27b)
v:cwm g = d*2/18 d* < 1 (Re < 1)
Equation (1O-27a) is in the normal range of diameters for ion exchange parti-
cles and leads to vsettling from 2 to 3 cm/s.
The average hole velocity is
(10-28)
Since usually At.olJAc. s 0.05, the hole velocity is significantly greater than the
superficial velocity. If Vhole is less than vsettling given by Eqs. (10-26) and (10-
27) weeping will occur. Either the number or size of holes can be decreased.
535
Wesselingh and van der Meer (1985) suggest 0.3 cm holes in laboratory
columns and 1.2 cm holes.in larger systems.
The minimum fluidization velocity depends on the shape and porosity of

the packed bed. The superficial minimum fluidization velocity, Vrnf, can be
determined by equating drag and gravity forces. This velocity can be deter-
mined from the solution of
150 (1 - e)J.1 Vrnf 1.75 Pf v~ (1O-29a)
(Pp-Pf)g== d~e3 + ~e3
At low Reynolds numbers the last term is negligible and
(pp - Pf) g d~e3 (1O-29b)

Vmf == 150 (I-e) J.1
A good rough estimate which is easier to use is (Wesselingh and van der Meer,
1985),
(1O-29c)
vmf == 0.02 vsettling
The upper limits to successful operation are flooding and insufficient

holdup of particles. Flooding occurs when resin particles are carried out of the
bed
(10-30)
This limit is usually not a problem since holdup becomes too small before Eq.
(10-30) is satisfied. As the fluid velocity increases, the fluidized bed expands
and holdup decreases. Thus, extremely high stages may be required to hold
enough resin for mass transfer. Wesselingh and van der Meer (1985) suggest
that holdup is too low if
(10-31)
The pressure drop of the bed is the same as the pressure drop in a packed
bed when vsupe:r < vrnf. Thus, Eqs. (8-22) and (8-23) can be used to calculate
~p. Once the bed fluidizes dP = dP (Vrnf) is a constant
536
The volume fraction particles, e, can be estimated from the Richardson

and Zaki relationship (Wesselingh and van der Meer, 1985).
vsuper Vsolid n-1 (1O-32a)

-- - -- = V1c:uling (1-e)
l-e e
where vsolid is the superficial velocity of the solids and n can be determined
from
2
5-n [ d· ] (1O-32b)
n-2.5 = -=;-
This last equation leds to n - 5 for small particles and n - 2.5 for large particles.
Wesselingh and van der Meer (1985) give a graphical solution for Eqs. (10-
32). At the minimum fluidization velocity e - 0.6.
The Cloethe-Streat contactor is somewhat unique in that detailed design

procedures have been published in the open literature. Fortunately, Eqs. (10-
26) to (10-32) are valid for single fluidized beds such as those occassionally
used for on-off chromatography or ion exchange. With the exception of one
old paper (Ermenc, 1961), design details for the staged counter current
adsorber with downcomers shown in Figure 10-1 have not been published.
Hole velocities and fluidization characteristics will be similar to the Cloethe-
Streat contactor. What is unique is design of the downcomer. This must be
done carefully to avoid fluid bypassing the stages and flowing up the downco-
mer. In addition, resin attrition can be minimized by designing the system to
have smooth edges.
In stirred tank systems the particles are suspended with an impeller.
Design details for these systems are discussed by Treybal (1980).
Less information is available about the design of moving bed systems,

partially because these systems can be difficult to scale-up. Without a doubt,
Exxon has amassed significant information on the design of magnetically sta-
bilized fluidized beds, but most of this information is confidential. Since the
project has been dropped, there was apparently difficulty in scaling-up to very
large sizes. Bums and Graves (1985) present some sketchy details from their
537
laboratory experiments. Moving beds with intennittant or pulsed solids flow

are apparently easier to scale-up and are extensively used (see Figures 10-8a,
10-9 and 10-10). Details for pulsed bed design are sparse, but Hutchins (1981)
suggests that
(10-33)
L=201p
and a fluid superficial velocity of 5 to 9 gpm/f~ be used during the loading

step.
Fluidized beds, moving beds and stirred tanks often have different resin
requirements than packed beds. Large diameter, dense resins have higher d*
values and hence higher settling velocities. This means that higher fluid veloci-
ties and smaller diameter vessels can be used. Unfortunately, larger diameter
particles have lower mass transfer rates. Thus, there is a tradeoff between
hydrodynamics and mass transfer.
The 5MB is really a packed bed, and the packed sections are designed as
packed beds. The regions between packed sections and the valve network
shown in Figures 10-16 and 10-17 have to be designed to minimize
extracolumn volume and mixing. Because it is a fixed bed, the 5MB will filter
out suspended solids and eventually clog if there are suspended solids in the
feed. The moving bed systems all have the advantage that some suspended
material can be processed.
10.6 SUMMARY - OBJECTIVES
The purpose of this chapter was to introduce you to moving bed and simulated
moving bed adsorption systems. At the end of this chapter you should be able
to meet the following objectives.
1. Explain how the following processes work.

a Staged moving bed system with downcomers.
b. Intermittantlyoperated sieve trays without downcomers.
c. Asahi and Chem-Seps processes.

d. Simulated counter-current system.
538
2. Use the linear solute movement theory to explain adsorption separation

a Moving bed systems.

b. Pulsed moving bed systems.
c. Simulated moving bed systems.
3. Use a stage model for design of staged moving bed systems and for
approximate design of both continuous contact moving bed and simulated
moving bed systems.
4. Use the mass transfer analysis to design isothermal moving bed systems.
REFERENCES
Anon., "Beaded carbon ups solvent recovery," Chem. Eng., 39 (Aug. 29,
1977).
Arden, T.V., Water Purification by Ion Exchange, Butterworths, London, 1968.
Berg, C., "Hypersorption design. Modern advancements," Chem. Eng. Prog.,

47,585 (1951).
Bon sal, R.c., I.-B. Donnet, and F. Stoeckli, Active Carbon, Marcel Dekker,
New York, 1988.
Brian, P.L.T., Staged Cascades in Chemical Processing, Prentice-Hall, Engle-

wood Cliffs, Nl, 1972.
Broughton, D.B., "Production-scale adsorptive separations of liquid mixtures

by simulated moving bed technology," Separ. Sci. Technol., 19, 723 (1984-85).
Burns, M.A. and DJ. Graves, "Continuous affinity chromatography using a

magnetically stabilized fluidized bed," Biotechnology Progress, 1 (2), 95
(1985).
539
Burns, M.A. and DJ. Graves, "Application of magnetically stabilized fluidized

beds to bioseparations," Reactive Polymers, 6,45 (1987).
Cloethe, FL.D., "Comparative engineering and process features of operating

continuous ion exchange plants in South Africa," in Naden, D. and Streat, M.
(Eds.), Ion Exchange Technology. Ellis Horwood Pub., Chichester, England,
1984,661-678.
Dechow, FJ., Separation and Purification in Biotechnology, Noyes, Park

Ridge, NJ, 1989.
Dodds, R, P.I. Hudson, L. Kershenbaum, and M. Streat, "The operation and

modelling of a periodic countercurrent, solid-liquid reactor," Chem. Eng. Sci .•
28, 1233 (1973).
Ermenc, E.D., "Designing a fluidized adsorber," Chem. Eng .• 68, (11), 87,
(1961).
Ford, M.A., "The simulation and process design of NIMCIX contactors for the
recovery of uranium," in D. Naden and M. Streat (Eds.), Ion Exchange Tech-
nology. Ellis Horwood Pub., Chichester, England. 1984, p. 668 to 678.
Higgins, 1.R. and RC. Chopra. "Municipal waste effluent treatment," in C. Cal-
mon and H. Gold (Eds.), Ion Exchange for Pollution Control, Vol. II, CRC
Press, Boca Raton, FL, 1979, Chapt 7.
Hutchins, R.A., "Development of design parameters," in J.R Perrich (Ed),

Activated Carbon Adsorption for Wastewater Treatment. CRC Press, Boca
Raton, FL, 1981, Chapt. 4A.
Liapis, A.I. and D.W.T. Rippin, "The simulation of binary adsorption in con-
tinuous counter-current operation and a comparison with other operating
modes," AlChE Journal. 25.455 (1979).
540
Lochmuller, C.H., C.S. Ronsick. and L.S. Wigman, "Fluidizied bed separators
reviewed: A low pressure drop approach to column chromatography,"
Preparative Chromatography, I, 93 (1988).
Lucchesi, P J., W.H. Hatch, F.X. Mayer, and R.E. Rosensweig, "Magnetically
stabilized beds. New gas solids contacting technology," Proc. 10th World
Petroleum Congress, Vol. 4, SP-4, Heyden & Sons, Philadelphia, 1979,49.
Naden, D. and M. Streat (Eds.), Ion Exchange Technology, Ellis Horwood

Pub., Chichester, England, 1984.
Neuzil, R.W. and R.H. Jensen, "Development of the Sarex process for the
separation of saccharides," paper 22d, AIChE meeting, Philadelphia, PA (June
6, 1978).
Rodrigues, A. (Ed.), Ion Exchange: Science and Technology, Martinus Nijhoff

Pub., Dordrecht, The Netherlands, 1986.
Rosensweig, R.E., "Magnetic stabilization of the state of uniform fluidization,"

Ind. Eng. Chem. Fundam..18, 260 (1979).
Ruthven, D.M., Principles of Adsorption and Adsorption Processes, Wiley-

Interscience, New York. 1984, ChapL 12.
Ruthven, D.M. and C.B. Ching, "Counter-current and simulated counter-

current adsorption separation proceses," Chem. Eng. Sci. 44, 1011 (1989).
Slater, M.J., "Recent industrial-scale applications of continuous resin ion

exchange systems," J. Separ. Proc. Technol .• 2 (3),2 (1981).
Storm, D.W., "Contacting systems," in J.R. Perrich (Ed), Activated Carbon

Adsorption for Wastewater Treatment, CRC Press, Boca Raton, FL, 1981,
Chapter 5.
541
Storti, G., M. Masi, and M. Morbidelli. "On countercurent adsorption separa-

tion processes," in A.E. Rodrigues et al (Eds.), Adsorption: Science and Tech-
nology, Kluwer Academic Publishers, The Netherlands, 1989,357-381.
Streat, M., "Recent developments in continuous ion exchange," J. Separ. Proc.

Technol., 1 (3), 10 (1980).
Treybal, R.E., Mass Transfer Operations, 3rd ed., McGraw-Hill, New York,
1980.
van der Wiel, J.P. and J.A. Wesselingh, "Continuous adsorption in biotechnol-
ogy," in A.E. Rodrigues et al (Eds.), Adsorption: Science and Technology,
Kluwer Academic Publishers, The Netherlands, 1989.
Wankat, p.e., Large Scale Adsorption and Chromatography, CRC Press, Boca
Raton, FL, 1986.
Wankat, P.C., Equilibrium-Staged Separations, Elsevier, New York, 1988
Wesselingh, J.A. and A.P. van der Meer, "Counter-current ion exchange," in A.
Rodrigues, (Ed.), Ion Exchange: Science and Technology, Martinus Nijhoff
Publishers BV, Dordrecht, The Netherlands, 1986,289-318.
HOMEWORK
AI. In your own words explain how the intermittant transfer system shown
in Figure 10-2 gives counter-current flow. What advantages might this
method of operation have compared to using continuous fluidized beds.
A2. In your own words explain how a simulated counter-current system

works. Why does the separation become better as each zone is divided
into more column sections?
542
A3. Study Figure 10-18. Relate solute movement in Figure 10-18 to zones
in Figure 10-13 and to concentration profiles shown in Figure 10-19.
A4. Relate Eqs. (10-21) and (10-22) to Figures 10-13 and 10-18.
AS. Show how to include a Murphree efficiency based on the fluid phase in
Figures 1O-5a and 10-5b.
A6. Show that the units in Eq. (10-2) are correct for a solute mass balance.
A7. Explain why the curvature of the isotherms in Figures 1O-5a and 1O-5b
are the opposite of those shown in Figures 6-3b and d even though they
could represent the same adsorption system.
A8. Write a time schedule for switching the valves in Figure 10-17 to simu-
late a moving bed.
A9. Sketch a system similar to Figure 10-17, but with 8 columns (2 per
zone). Explain the time schedule used for switching the valves. What
are the advantages and disadvantages of this system compared to Figure
lO-l7?
A 10. Despite considerable research and development on moving bed and

simulated moving bed systems, packed bed adsorbers and chromato-
graphs remain much more common. Discuss why packed beds are
more commonly used.
All. The Cloethe-Streat design for intermittant countercurrent flow of solids

and fluid is used commercially for ion exchange processing of uranium
solutions. What are the advantages of this design compared to using
packed beds? What are the advantages of this designed compared to
using a column with sieve trays and downcomers?
A12. Film mass transfer rates in fluidized beds are usually lower than those in
packed beds. Explain why.

543
C. Derivations
Cl. Derive the solute movement diagram for simulated countercurrent

motion if the observer fixes his frame of reference at one of the product
ports. (Remember that the product port shifts location every few
minutes).
C2. Convert the Kremser equation to the appropriate units for a staged
adsorber with a linear isotherm.
C3. (Challenging!) The staged and solute movement theories agree qualita-
tively. Show that this is true for complete removal of a single solute in
a moving bed system for linear isotherms.
Hint: Start by comparing the slopes of the operating and equilibrium
lines in Figure to-5a to achieve Y1 - o.
C4. Derive the equation to relate Y to c.
C5. Expand Eqs. (10-21) and (10-22) or Eqs. (10-24) and (10-25) to include
the concentration dependence of the solute velocities. Using a typical
favorable equilibrium isotherm, explain why satisfaction of this set of
equations is more difficult for nonlinear systems than for linear systems.
Dl. A staged adsorber (as shown in the top section of Figure to-I) is used
to adsorb propane from hydrogen on a silica gel at 40°C. The adsorber
operates at a total pressure of 2 atmospheres. Inlet silica gel is pure (q =
0). Inlet gas is 0.10 mole fraction propane. Outlet gas should contain
0.005 mole fraction propane. Use a S/G = 1.5 x (S/G)min, and determine
the number of equilibrium stages required. S = kg/hr clean silica gel, G
=kg/hr of H2 •
Equilibrium data was given in Problem 6-D1 for a single porosity
model.
D2. A Cloethe-Streat intermittant contact fluidized bed system is to be used

for recovery of U 3 0 S on a strong acid resin initially in the W form.
544
The feed is 0.10 g!liter of U 3 0 g in aqueous solution. An outlet concen-

tration of 0.002 gil is desired. The recycled resin from the elution
column contains 1.0 gIliter of U 3 0 g • Equilibrium data (Ford, 1984) are
given below,
c,gIl 0.0030 0.0111 0.0330 0.0560 0.1010

CR, gil 8.2 14.7 25.0 30.6 38.0
If (liquid flow rate, liter!hr)/(solid flow rate, liter!hr) = 300, find
a. The outlet concentration of the resin.
b. The number of equilibrium stages required (Assume the flows are

continuous).
D3. We wish to use a DEAE Sephadex A-50 ion exchanger to sorb bovine
serum albumin in a batch stirred tank system. The tank is charged with
10 kg of dry resin which is initially clean. The total liquid charge is
60,000 kg and is 1.2 x 10-3 weight fraction prutein which is 1.2 mg
protein/g solution in an aqueous solution. Find the final concentration
in the tank and on the resin after equilibrium is attained. Equilibrium
data is given in Example 6-1. W-lat fraction of the protein is recovered
on the resin?
D4. A staged contractor is being used to soften water. The feed contains 2.0
meq/L Ca+2 and 9.0 meq/L Na+. A strong acid resin with CRT = 1.7
eq/L is used. A Murphree stage efficiency based on liquid concentra-
tions of 80% is expected. The system has four real stages. Entering
resin is entirely in the Na+ fonn. We desire an outlet product which is
0.01 meq/L Ca+2 • What ratio (liquid flow rate/solids flow rate) is
required?
Equilibrium data is available in Table 9-2.
D5. The resin from Example 10-1 is regenerated in a staged contactor with 4
equilibrium stages. The inlet resin concentration is qo = 4.52 gIL while
outlet is <IN = 0.25 gIL. Entering solution is pure hot water. At the
higher temperature of operation equilibrium is q = 0.05 c. What value
of VR/Vs is required?
545
D6. Example 10-2 presented one set of data from Burns and Graves (1987).
For another experiment under different condtions they found:
Feed concentration = 1.0 mg HSA/ml solution
Input solutes = 0 mg HSA/g solids
Fluid flowrate = 10.0 mVmin
Solids flowrate = 0.5 g/min
L kci
K= S k;=3.1
% HSA adsorbed = 25%
Bed length, h = 5.0 cm
Find NTU and HTU. Speculate on why H1U increased.
D7. We wish to regenerate calcium alginate beads which contain human

serum albumin (HSA) in a moving bed column. Regeneration is done
by changing pH and ionic strength so that equilibrium is linear and
becomes m = c/q = 1.2. The inlet regenerant is pure, Cin = 0 mg
HSAlml and inlet solids have 'lin = 2.24 mg HSA/g adsorbent Solids
flow rate S = 0.5 g/min and L = 1.0 mVmin. We desire qout = 0.05 mg
HSA/g adsorbent H1U = 10.0 cm. Find the height of the column and
Cout in mg HSA/ml.
Note: This problem represents the regeneration of the beads from
Example 10-2.
D8. A 4 zone 5MB is being used to separate pyrene and anthracene. The
system will have one colwnn per zone. Each of the 4 columns is 150
em long. The value of lp = 937.5 s. Eo = 0.37. Measurements in a
packed bed give solute velocities: 1lp = 0.0664 v and UA = 0.1066 v.
Isotherms are linear. The following superficial velocities are used:
Vsuper,3 == 1.0 cm/s , vsuper,2 = 0.666 cm/s ,

vsuper,l = 0.814, Vsuper,4 == 0.444
a. Does this satisfy Eqs. (10-24) and (1O-25)?
b. Determine the time required and the distance travelled for pyrene
and anthracene to exit the 5MB.
c. Determine the product velocities where vProd == PIPe Ac.
546
D9. We wish to fluidize a polystyrene-sulfonic acid ion exchange resin. The

spherical beads have a diameter of 0.7 mm. The resin density is 0.82
g/cm 3 when they are wet but have been drained. X = 0.75. Fluid is
water at 25°C where Pf - 1.0 g/cm3 and J.I. = 0.94 cpo The acceleration
due to gravity at sea level is g = 980 crn/s2 •
Calculate: a. Reference velocity and diameter
b. d*
c. Vsculing
d. If vsuper =2.0 crn/s and VsoJids =0, determine e.
Fl. The hypersorption process is an interesting historical example of a pro-

cess which was used commercially, but is now obsolete. Study the
hypersorption process (Berg, 1951; Treybal, 1980; Wankat, 1986) and
write a report on this process. In your report answer the following
questions:
Why is the process no longer used?

What were the major problems with hypersorption?
Could modem adsorbents or techniques be used to make hypersorption
competitive?
If you were employed by a major petroleum or petrochemical company,
would you recommend a major research program to develop a modem
hypersorber?
Why or why not?
NOMENCLATURE CHAPTER 11
a angle in two-dimensional system (Figure 11-7b)
c solute concentration
<1,0 bulk molar concentration
Cp specific heat
D efI effective axial dispersion coefficient, cm2 Is
e electron charge, 1.602x 10-19 coulombs
E electric field strenght, volts/m
fc friction coefficient
F force, Newtons
F Faraday's constant
g acceleration due to gravity, m2 Is
Gr Grashof number, Eq. (11-12)
h characteristic length, e.g. Ry, or 1f2 distance beteen plates, m
I current, amps
k Boltzmann constant, 1.386 x 10-23 JIK
kT thermal conductivity
L distance traveled, Eq. (11-32)
L length, m
ni maximum number of peaks in development direction i
nw maximum number of peaks in two-dimensional development
N number of plates
NA Avagadro's number, 6.02 x 1023
547
548
P -dJ.l/dz, Eq. (1144)
P power, watts
Q net charge, coulomb
r radial direction, m
R resolution
R resistance, ohms
Ra Rayleigh number, Eq. (11-12)
Rp radius, m
t time,s
lexper time of experiment, s
T absolute temperature, K
u migration velocity, cm/s
electroosmosis velocity
solute velocity from Eq. (6-22)
buffer velocity, cm/s
v voltage, volts
w width, m
y direction between electrodes
Yi,net net lateral movement, Eq. (11-33)
Yrecyc!e distance recycle stream shifted
Yi location of peak center
valence
migration distance or location of peak center

549
Greek
~ coefficient volumetric expression, Eq. (11-11)
6R p shear boundary is at Rp + ~p
I'lT characteristic temperature difference, K
e dielectric constant of medium
eo permittivity free space, 8.85 x 10-12 C 2/Jm
~ zeta potential, value 'If at Rp + I'lRp
~wa1l zeta potential of wall
11 viscosity
1C Debye-Huckel constant, Eq. (11-6), m- 1
A:i ionic equivalent conductivity, Eq. (11-9d)
J.l electrophoretic mobility, m2 N s
J.losm electrophoretic mobility of layer next to wall
p fluid density, kg/m 3
p average fluid density
(J'c electric conductivity, (ohm· cmr1
(J'l,i standard deviation in length units for component i
'If electrical potential
"'0 electric potential at surface of particle

Chapter 11
ELECTROPHORETIC SEPARATION METHODS
Electrophoresis is a method which uses the different migration rates of charged

molecules or particles in an electric field for separation. The method was first
developed by Arne Tiselius (1937) who received the Nobel prize in 1948
because of the tremendous impact of electrophoresis on biochemical separa-
tions. A large number of variants of electrophoresis have been developed and
will be introduced in this chapter. These methods are used extensively for
laboratory scale biochemical separations such as protein and polynucleotide
separations. Despite numerous efforts, scale-up has had minimal success
mainly because of thermal convection caused by Joule heating of the system.
First, the basic concepts and theory of electrophoresis will be presented

followed by a discussion of phenomenon which complicate the simple picture.
Then, some of the wide variety of electrophoresis processes will be briefly
explained and possibilities for scale-up will be explored. lsotachophoresis
which is essentially a displacement chromatography version of electrophoresis
will be explained theoretically and scale-up will be discussed. lsoelectric
focusing which has developed as a separate technique will be explained.
Finally, electrochromatography will be briefly described. Chapter 7 is a prere-
quisite for this chapter since chromatographic approaches to zone spreading
will be used. Since electrophoretic methods are not currently used on a large-
scale, this chapter should be considered a progress report.
11.1. THEORY OF ELECTROPHORESIS
The theory of electrophoresis is well understood. but can rapidly become quite
complex. We will first start with a physical picture of the basic reasons why
550
551
Table 11-1. SI Units for Electrical Terms
Charge, Q: Coulomb (C) = Amp - second

Coulomb =Joule/volt = NmN
Dielectric constant, E: dimensionless
Electric field, E: Volt/meter
Energy: Joule (1) = Newton meter
Force, F: Newton (N) = kg m/s2
Power,P: Watt (W) = Joule/second
Pressure, p: Pascal (pa) = N/m2
Viscosity,11: Pascal-second = Ns/m 2
Permittivity free space: Eo = 8.85 x 10-12 C2 jJm
1 QQ'
Coulomb's law: F=-- - -
4 7t Eo er2
electrophoresis works. Then a simplified mathematical treatment of the forces

involved in electrophoresis will be presented, and finally the complicating fac-
tors which reduce the separation will be identified.
We will use SI units for electrical terms. Great care must be taken when
comparing with equations in other units since the equations may differ. Some
of the SI conversions are listed in Table 11-1. The proportionality factor
1/(47t Eo) for Coulomb's law is required for SI units, but does not occur in other
units.
11.1.1. Basic Concepts of Electrophoresis
Electrophoresis is essentially based on different rates of migration in an electric

field. Charged molecules or particles have a force exerted on them by the
electric field. If the net charge on the particle is Q, then the force is QE where
E is the electric field strength. Because of this force, the particle will accelerate
towards one of the electrodes. Particles with a net positive charge move
towards the cathode (the negative pole) while particles with a net negative
552
charge move towards the anode (the positive pole). The acceleration caused by
this electrophoretic attraction is opposed by friction, electrophoretic retardation
and a relaxation effect At steady state the sum of the forces is zero and the
particle moves at a constant migration velocity u. This is usually written as
(11-1)
where the migration velocity u is in m/s or cm/s, the field strength E is in

volts!m or volts/cm, and the electrophoretic mobility J.L is in mZ/voltos or
cmz/voltos. The electrophoretic mobility can be either positive or negative
depending upon the sign of the net charge on the molecule. The essential pur-
pose of the theory is to predict J.L.
Electrophoresis is useful for separating a variety of biochemicals because

these molecules have net charges. For example, the general formula for an
amino acid is
NH z - C -COOH
I
R
where R is a functional group. At low pH values the a-amino group will be in

the protonated form, NHj, while at high pH values the carboxylic acid group
will be in the form COO-. Since both groups can be ionized simultaneously,
the amino acid can have a positive, negative or neutral charge. In addition, the
R group may be charged (see Lehninger, 1970, for details). Amino acids are the
basic building block of proteins; thus, a protein will usually contain a large
number of positive and negative charges. The net charge on the protein will
determine which electrode the protein migrates towards. There is one pH value
where the net charge on the protein is zero and J.L = O. This is the isoelectric
point or pI of the protein. A short list of pI values for amino acids and proteins
is given in Table 11-2. Isoelectric focusing uses a pH gradient in the column
and focuses each protein at its isoelectric point.
A second major group of biochemicals which are often separated by

electrophoresis are the polynuc1eotides such as RNA and DNA. These
molecules are phosphoric acid esters of 5-carbon sugars with a base attached.
553
There are only five principal bases in polynuc1eotides all of which are purines
or pyrimidines (see Lehninger, 1970 for details). Since the phosphate groups
are usually very negatively charged, the polynuc1eotides usually migrate to the
anode. The polynuc1eotides often have an almost constant charge-to-size ratio,
and thus sieving gels are used to utilize frictional forces (see Section 11.2.2.).
Table 11-2. Molecular Weights and Isoelectric Points

(Blackshear, 1984; Lehninger, 1970; Reeder, 1987; Righetti, 1983)
Compound Mol. Wt. pI(25°C)
Aspartic Acid 2.77

L-glutamic acid 3.22
~-aspartyl-histidine 4.94
Triglycine 5.59
Histidine 7.47
Lysine 9.74
Arginine 10.76
Ovalbumin 43,000-45,000 4.70

Bovine albumin 68,000 4.95 ± 0.02 (4°C)
~- Lactoglobulin 18,400 5.14 (A)/5.39 (B)
5.45 ± .02 (B at 4°C)
Mysosin 2,000,000--2,120,000 6.2- 6.6
Myglobin (horse) 16,950-17,800 7.33 ± 0.01
7.58 ± 0.02 (4°C)
Ribonuclease 13,700 8.88± 0.03
Cytochromec 11,700-12,500 9.28 ± 0.02
Human fetal brain cells 4.38

Escherichia coli 5.6
Normal liver cells 6.5
554
11.1.2. Forces in Electrophoresis
A relatively simple theory of electrophoresis can be developed by considering a

force balance. The development given here will follow that of Hiemenz (1986)
while more detailed analyses are given by Overbeek and Bijsterbosch (1979)
and O'Brien and White (1978).
The force which causes electrophoretic migration is the electrophoretic

attraction force.
(11-2)
Electrophoretic Attraction = QE
where Q is the net charge on the molecule or the particle. The primary force
which deaccelerates the particle is the friction force
(11-3a)
Friction Force = - fc u
where fc is the friction coefficient For spherical particles at very low Reynolds
numbers in free liquids fc can be determined from Stokes law
(11-3b)
where Tl is the viscosity of the medium and Rp is the radius of the particle or
molecule. In stabilizing media such as gels or paper fc needs to be determined
experimentally.
For isolated ions we can set the sum of these two forces to zero. Solving
for velocity, we have
QE (ll-4a)
U=--
fc
which for spherical ions is
QE (11-4b)
U=
61tTlRp
555
Since the charge on an ion is Q = z e where e is the electron charge,

1.602)( 10- 19 coulombs, and z is the valence, Eq. (11-4b) become
zeE (11-4c)
u=---
67t11Rp
Comparing Eqs. (11-4c) and (11-1), we obtain the mobility of an isolated ion
(ll-4d)
Note that viscosity should be in Pa·s (Ns/m2) = 0.001 centipoise so that units
wOIk
For charged colloidal particles the picture is significantly more compli-

cated, and there are additional forces. (See Hiemenz, 1986, for a detailed
explanation). In order to carry an appreciable current there must be small elec-
trolytes in solution. The electric field exerts a force on these ions; however,
since the ions are quite small and are hydrated the force is transferred to the
water. This causes an electroosmotic flow of water with respect to the particles
which retards the particle movement To detennine this electrophoretic retar-
dation force we must look at the environment of large charged molecules or
particles. This environment is shown schematically in Figure 11-11 for a nega-
tively charged spherical particle (the picture will be sirniliar for a positively
charged particle). To have electroneutrality the negatively charged particle
must have an appropriate number of positively charged ions associated with it.
These positively charged counterions distribute around the particle. The conbi-
nation of negative and positive charges is called the double layer. The electri-
cal potential has a value '1'0 at the particle surface. However, there is a layer of
liquid one or two molecules thick which is bound to the particle and moves
with the particle. The fluid starts to move at the shear boundary (Rp + ~p in
Figure 11-1). The potential has a value C(the zeta potential) at the shear boun-
dary. The zeta potential is the value observed in experiments and Ccontrols the
electrophoretic mobility of particles. Usually Cis measured by measuring the
electrophoretic mobility of the particles.
556
Potential '"
'" ="'0 exp(-I )
r-oo
r
Figure 11-1. Potential around a negatively charged particle. a. Schematic.
b. Potentials at different values of 1(.
The zeta potential is unknown but is proportional to the electric potential

at the surface of the sphere '1'0. The electrical potential drops from a value of
'1'0 at the surface of the particle to zero at an infinite distance according to the
Poisson-Boltzmann distribution
(II-5a)
'I' = '1'0 exp [-lC (r-~)]
which is shown in Figure ll-Ib. The potential at the surface of the sphere. '1'0'
is
(11-5b)
where Q is the total charge on the particle and SI units are being used. The
term lC
(11-6)
557
is the Debye-Huckel constant. The summation tenn is twice the ionic strength.
In this equation e is the elementary charge which is 1.602 x 10-19 coulombs,
NA is Avogadro's number 6.02 x 1023, Ci,o is the bulk molar concentration of
the ion, Z is the charge number of the ions including the sign, E is the dielectric
constant of the medium, k is the Boltzmann constant, 1.38066 x 10-23 JIK, T is
the absolute temperature, and Eo = 8.85 x 10-12 Coulomb2 /Joule m. The units
of x: are m-1 • This can be seen by writing out the SI units in Eq. (11-6).
'/2
(lOOO ~) (C? (_1_) ( mole)
m3 mole L
x:=
(C2 /lm) [ ~ 1K
The thickness of the double layer is imprecisely set at 1/x: (see Figure 11-1 b).
Then KRp is the ratio of the particle diameter to the thickness of the double
layer. Typical values of 1/x: in aqueous solution are (Overbeek and Bijster-
bosch, 1979): c = lO-5 M, 1/K = 1.0 x lO-7 m; c = lO-3 M, 1/ K = 1.0 x lO-8 m,
and c = lO-1 M, I/K = 1.0 x lO-9 m.
The electrophoretic retardation force can be detennined several different

ways. The general solution first developed by Henry is
(1l-7a)
Electrophoretic Retardation Force = [41t Eo ECR p f(x: Rp) - QJE
Values for f(K Rp) at various values of ~ are given by Hiemenz (1986). For
small particles with ~ < 0.1 this simplifies to Huckel's solution.
(1l-7b)
Electrophoretic Retardation Force = [41t Eo E~p - QJE
For large particles with ~ > 100 Henry's solution simplifies to

Smoluchowski's solution
Electrophoretic Retardation Force = [(1.5) 41t Eo E~ - Q JE

(1l-7c)
558
There is a fourth force which occurs because the double layer is distorted
since the charged particle and the ion atmosphere move in opposite directions.
The distorted atmosphere exerts a force on the particle which is known as the
relaxation effect. If we are willing to assume that the relaxation effect is negli~
gible, we can easily determine J.1. This assumption is usually made in
simplified treatments, but more exact numerical solutions include the relaxation
effect (Overbeek and Bijsterbosch, 1979). We sum Eqs. (11-2) and (1l-1), and
solve for J.1. The general result is
(1l-8a)
while if ~ < 0.1 we obtain
(11-8b)
and if x:Rp > 100 we obtain
(ll-Bc)
For free solution systems at low Reynolds numbers Eq. (11-3) can be inserted
forfe in Eqs, (l1-B). The general result is
(1l-9a)
while if1C Rp < 0.1 we nave
(1l-9b)
and if 1C Rp > 100 we obtain
Eo £ ~ (11-9c)
J.1=--
11
Note that some references present these equations in cgs units. The cgs result
can be obtained by dividing Eqs. (11-9) by 41t Eo.
559
Table 11-3. Mobilities in Aqueous Solution

(Everaerts et al, 1976; Karol and Karol, 1978)
Ion
H+ 36.2 x 1~
C~ 8.13 x 10-4
Rb+ 8.03 x lQ-4
K+ 7.67 x lQ-4
NH:t 7.6 x 10-4
CH3 NH! 6.0 x 10-4
(CH3 hNH! 5.4 x 10-4
(CH3h~ 4.8 x 10-4
(CH3 )4 W 4.7 x 10-4
Li+ 4.02 x lQ-4
(C ZH S )4 W 3.6 x 10-4
(C4 H9 )4 W 2.0 x 10-4
N0 3 -7.4 x 10-4
cr -7.9 x 10-4
Br- - 8.1 x 10-4
OH- - 20.5 x 10-4
Typical values for Jl in free solution are 5 x 10-4 cmZjsV (5

x 10-8 mZjsV) for small ions and 1 x lO-S cmZjsV (1 x 10-9 mZjsV) for proteins.
Table 11-3 lists values for small ions (Everaerts et aI, 1976; Karol and Karol,
1978). Note that Jl decreases as the size of the ion increases. Hydrogen and
hydroxyl ions are unique in their high mobilities. For small ions mobilities are
usually not reported in the literature. Instead, ionic equivalent conductivities,
~, are usually reported and are related to mobility by (Newman, 1973)
(1l-9d)
560
where F is Faraday's constant, Z;. is the charge number of species i, and ~ is

the ionic equivalent conductance, mho· cm 2/equiv.
From these equations we can deduce quite a bit about the electrophoretic
behavior of a molecule or particle. From Eqs. (11-9) and (11-1) it is clear that
particles are retarded by high viscosities, and particles with higher CJX>tentials
move faster. The dependence of the CJX>tential on operating variables can be
deduced from Eqs. (II-Sa) to (11-6). Remember that Cis the value of 'I' at the
shear boundary (r=~+~). The value of '1'0 and hence Cincreases as the
charge Q increases or the particle radius Rp decreases. For practical purposes C
and hence J.l and u are proportional to the charge to size ratio Q/Rp. Since the
net charge Q on a molecule is very dependent on the pH, C, J.l, and u are very
dependent on pH. Both the thickness of the double layer, 11K, and Cdecrease
rapidly as the concentration or the valence of the electrolytes increase. Increas-
ing temperature increases 1/K and hence ~ and decreases the viscosity 11, and
thus both J.l and u are increased. Of course, temperature is limited by the stabil-
ity of the molecule and convection effects.
Example 11-1. Double Layer Thickness and Zeta Potential
a. Estimate the double layer thickness in a 0.005 M solution of NaCl

at 25°C.
b. A lO-6m diameter particle has a measured mobility of 1
x 10-9 m 2/sV in a 0.005 M aqueous solution of NaCl at 25°C.
What is the value of C?

Solution
Part a. At 25°C, E = 78.30 for water (Dean, 1985, 10-99).
Then, from Eq. (11-6)
K=
[ (1000) (1.602 x 10-19 )2 (6.02 x 1023 ) (0.01)
(8.85 x 10-12 ) (78.30) (1.38066 x 10-23 ) (298.16)
l
~
561
which gives
1( = 2.327 x 108 m-1 and ~ = 4.297 x 10-9 m

1(
Part b. 1( ~ = (2.327 x 108 m- 1) (1 x lO4jm) =232.7

Use Eq. (1l-9c). Rearranging, we have
For water at 25°C, 11 = 0.890 x 10-3 N ~ (Dean, 1985, p. 10-99).

m
[0.89 x 10-
3 N: 1 (1 x 10-9 m2/sV)
~= m = 0.001284 N Jm
(8.85 x 10-12 C2/Jm) (78.30) V C2
which is
~ = 0.001284 +-
C V
= 1.284 x 10-3 V
where we have used the conversions in Table 11-1.
It should be noted that it is important to check the units. Also, the

stoichiometry and equilibrium of the dissociation must be considered
when calculating L (Ci zr) for more complex ions.
11.1.3. Complicating Factors in Electrophoresis
Unfortunately, the operation of elecrophoretic separations is not as simple as

the preceeding section may imply. The first complicating factor is Joule or
resistance heating. The power generated in a resistance is
(11-10)
562
where I is the current in amps, V is the voltage in volts, and R is the resistance
in ohms. Then the power consumption P is in joules/sec = watts. This power is
dissipated by heating the electrophoresis cell. This I2R heating is undesireable
since the local differences in temperature causes local differences in density
and hence convection cells. In addition, large increases in temperature may
destroy fragile biochemicals.
Convection caused by Joule heating can destroy the separation developed

by different migration rates. The temperature gradient can be related to density
changes by assuming that density is linear in temperature
(11-11)
p=p-p~(T-T)
where ~ is the coefficient of volumetric expansion. The occurrence of thermal

convection depends upon the value of the Rayleigh number (for example, see
Ostrach, 1977).
(11-12)
The amount of convection depends upon the value of the Grashof number.
Gr = p2 ~ g h3 ~T = Pg h3 ~p (11-13)
T\2 T\2
The second equation in both Eqs. (11-12) and (11-13) is obtained by substitut-
ing in Eq. (11-11). In these dimensionless groups p is the average fluid density,
g is the acceleration due to gravity, h is a characteristic length such as the half
distance between parallel plates or the pore radius, kT is the thermal conduc-
tivity, Cp is the specific heat of the fluid, O"c is the electrical conductivity in
l/(ohm-cm), and ~T is a characteristic temperature difference. The Grashof
number is the ratio of buoyancy to viscous forces while the Rayleigh number is
the Prandtl number (Cp T\/kT) times the Grashof number. For very small Ray-
leigh numbers there will be no convection, while for larger Rayleigh numbers
convection occurs. The critical Rayleigh number depends on the geometry
(Ostrach, 1977). The convection becomes stronger as the Grashof number
increases.
563
Many ingenious methods have been developed to control or prevent con-

vection. Efficient cooling helps reduce L\T and prevents thermal degradation.
Biochemicals are often electrophoresized near 4°C since they are stable at this
temperature and water has a density maximum at 4°C which means that ~ = O.
Usually, the difference between plates is kept small (small h), or the electro-
phoresis is done in small tubes or in the pores of a gel, membrane, paper or
capillary. Another method of keeping Ra and Or small is to increase the
viscosity 11. Rotation of the apparatus or the fluid has been used as a way of
stabilizing the fluid so that convection cells do not occur. The critical Rayleigh
number is larger in horizontal equipment than in vertical equipment; thus, hor-
izontal chambers can have the plates further apart and still not have convection
(Ostrach, 1977). Finally, electrophoresis has been done in space where g is
very small. Equipment to keep both Or and Ra small is explored in Section
11.2. where a variety of different electrophoresis methods are explored.
The introduction of a gel or membrane to stop convection can add an

additional complicating factor. This is sieving of the molecules or particles.
The migration of molecules in a gel is hindered since the pores of the gel are
not straight. If the pores are not a lot larger than the molecule there will be
additional frictional forces between the molecule and the gel which will retard
the migration of the molecule. The larger the molecule the more this sieving
effect hinders the migration. If the molecule is too large it cannot enter the gel
at all and u = O. These sieving effects can be very helpful since they allow one
to separate molecules with the same charge to size ratio. The application of
sieving effects is discussed in Section 11.2.2.
Electroosmosis is another complicating factor in electrophoresis. Fixed

charges on the chamber walls or on the gel are subject to a force because of the
electric field. Since the charges are fixed, the fluid moves. This fluid move-
ment is electroosmosis (Vanderhoff and Micale, 1979). The electroosmosis
velocity can easily be calculated. For instance, for electrophoresis between two
parallel walls which are charged, 1.1 can be determined from Eq. (11-9c). Sub-
stituting this into Eq. (11-1), we obtain
(11-14)
564
Equation (11-9c) is used because the wall can be considered as a particle with
infinite radius and hence KRp is very large. The effect of electroosmosis is to
reduce the separation. The amount of reduction depends upon the geometry
and will be discussed in Section 11.2. Electroosmosis can be reduced or elim-
inated by using walls or gels which have no residual charge. This can be done
by coating the walls with a material such as methylcellulose which has a very
low zeta potential.
11.2. METHODS FOR ELECTROPHORESIS
Electrophoresis has been blessed with a huge number of variants designed to

avoid some of the problems discussed in the previous sections. Two major
variants, Isotachophoresis and Isoelectric focusing, have to a large extent
become considered as separate techniques. These methods are discussed in
Sections 11.3 and 11.4. In this section a number of variants of the basic elec-
trophoresis method will be discussed. Many of these methods are probably of
interest only as analytical methods, and thus will be discussed rather briefly.
Those methods which may have applications in large scale separations will be
discussed in more detail.
11.2.1. Moving Boundary Electrophoresis
Moving boundary electrophoresis is the original method developed by Tiselius

(1937) for separating proteins. It is now used mainly for measuring electro-
phoretic mobilities since more recent methods are easier to use and provide
better separations. The apparatus is shown schematically in Figure 11-2. The
cell is U-shaped with rectangular cross-section. Protein solution in buffer is
initially added to one ann and the bottom of the cell while the other ann is
filled with pure buffer solution. The voltage is applied and electrophoresis
proceeds. Each protein migrates according to Eq. (11-1) where Jl is determined
by Eq. (11-9a, b, or c). Since the protein solution is denser than the buffer, the
density gradient is stable and convection is prevented.
565
G)-
--- Final t- - - Initial
--- Initial "" -- Final
Figure 11-2. Moving boundary electrophoresis.
Analysis is based on detennining the migration of bands of different den-

sity. This is usually done by using an optical system which measures the
deflection of light as it passes through liquid with a gradient of the index of
refraction. The commonly used methods are schlieren and interference optics.
These methods are specialized and expensive (for a brief introduction see Bier,
1974).
Only the fastest moving component in each direction can be recovered as

a pure component This limitation is one of the major reasons that moving
boundary electrophoresis has been replaced by other electrophoretic methods.
The chromatographic analog of moving boundary electrophoresis is frontal
analysis which is also rarely used.
11.2.2. Gel, Membrane and Paper Electrophoresis
One common method of preventing convection is to do the electrophoresis in a

stabilizing porous media such as a gel, a membrane, or on paper. These
methods are usually operated as zone electrophoresis which is analogous to
elution chromatography. Since zone electrophoresis is quite similiar regardless
of the porous media used, we will explain the method for the most common
approach which is gel electrophoresis. A laboratory gel electrophoresis cell is
shown in Figure 11-3a. A thin layer of gel is polymerized onto a support plate.
The gel is connected to the electrode compartments by filter paper bridges. The
support plate, which is usually Mylar or glass, is placed on a coolant plate so
566
a Cover
Sample Filter paper
(±) ~=~:::~~:=:=:=37/L
Anode ......... """""L--l-c
Bridge e
Cathode
Cooling Plate
Support
Buffer
Plate
o
Cone
Figure 11-3. Zonal gel electrophoresis. a. Apparatus. b. Observed concen-

trations. Al and A z are anions, 0 = zero charge, CI and C z are
cations.
that Joule heating can be removed. The gel layer is kept quite thin (usually less
than Smm) to have better heat dissipation and for increased mechanical
strength. The entire apparatus is covered to control evaporation and for safety
reasons (high voltages may be used). A variety of modifications to the basic
apparatus are used, but they do not change the function of the device. These
modifications include the use of vertical slabs or tubes, and two-dimensional
methods (see Sections 11.2.5. and 11.2.6.).
In zonal gel electrophoresis a small sample is placed in a sample well in

the center of the gel slab. When the current is turned on, the solute molecules
start to migrate towards the anode or the cathode. Since the width of the feed
sample is quite small, each solute migrates as a zone with a velocity ~ given by
Eq. (11-1). The zones eventually become separated from each other. The
operation is stopped before the fastest migrating species reaches the bridges to
the electrode compartments. In analytical applications the products are often
detected directly on the gel by UV absorbance, fiuoresence, or other methods
(see Andrews, 1986; or Melvin, 1987; or Righetti, 1983). For preparative
567
applications the solutes need to be Ittovered from the gel. Recovery can be
done by scrapping some of the gel into a beaker, by blotting onto paper, or by
eluting electrophoreticall y or with flow. The final concentrations observed will
have a Gaussian shape as shown in Figure 11-3b. (see Section 11.2.3).
The stabilizing media should have certain properties. It should be inert

and not react with the species being separated. There should be no residual
charges on the media since charges may ion exchange with the species being
separated and will cause electroosmotic flow. The pores of the media must be
at least as large as the molecules being separated or the molecules will not enter
the media. If the pores are only slightly larger than the largest molecules being
separated, then the larger molecules will be retarded by frictional forces and
these molecules will no longer follow Stoke's law. However, this sieving
effect is useful since sieving separates on the basis of size while electrophoresis
without sieving separates on the basis of the charge to size ratio. To use
sieving, gels with known and different pore sizes are polymerized.
The most common electrophoresis in use today is polyacrylamide gel

electrophoresis (PAGE). Polyacrylamide is an excellent anti-convective gel. It
is easy to polymerize in place, and it can be polymerized with a variety of pore
sizes. In addition, there is very little residual charge on the polymer. To make a
PAGE plate, acrylamide (see Figure 11-4a) is polymerized directly as a slab on
the support plate. The acrylamide is cross-linked with NN'methylene bisa-
crylamide (bis). An initiator such as ammonium or potassium persulphate and
a catalyst such as N,N,N',N', tetramethylenediamine (TEMED) are required.
The polymerization is a vinyl type polymerization which produces a random
coil polymer. The pore size can be controlled by varying the total monomer
concentration T (% acrylamide + % 'bis') or the degree of cross-linking C (%
'bis' / % T x 1(0). Reducing %T increases the pore size, but the maximum
attainable size is about 80 nm. Gels more dilute than this become mechanically
unstable. Pores also become larger if %C is increased to 30%C which
corresponds to gels in the range of 200 to 250 nm. This is done at a constant
%T which is often about 15w/v% acrylamide. The pores become larger at high
crosslinker concentrations because a "bead-like" structure is formed instead of
the usual three-dimensional lattice. Recipes for the gels and buffer solutions are
568
b. CH 2 =CH
I
C=O
a. CH2 -CH
I
NH
I I
C=O CH2
I I
NH2 NH
I
C=O
I
CH2 =CH
C. CH3 - (CH2)1O - CH2 OS03 NA+
Figure 11-4. Chemicals used in gel electrophoresis. a. acrylamide monomer.

b. NN' methylene bis acrylamide (bis.). c. Sodium dodecyl sul-
phate (SDS). d. Agarobiose (1,3 linked ~-D galactopyranose
and 1,4 linked 3,5 anhydro-a-L- galactopyranose).
given by Andrews (1986) and Blackshear (1984). The gel slab can be polymer-
ized with a gradient in pore size so that molecules will slow down and eventu-
ally stop when they reach a limiting pore size. This is called pore limit electro-
phoresis (Gianazza and Righetti, 1979).
A modification of PAGE which is commonly used is to use a stacking gel

which is also called disc gel electrophoresis (Ornstein, 1964; Davis 1964). In
this case a very dilute stacking gel (2.5 w/v% acrylamide) is placed on top of
the separating gel (15 % acrylamide) in a vertical tube. Thus the gel is discon-
tinuous (disc). The sample is placed between a leading chloride buffer which
migrates rapidly and a trailing glycinate buffer which migrates slowly. In the
stacking gel this combination leads to the electrophoresis equivalent of dis-
placement chromatography. (In other words isotachophoresis occurs in the
569
stacking gel. The theory behind this method is explored in Section 11.3.) The
sample is then concentrated in the stacking gel. The separating gel operates at
a higher pH which causes the glycine to dissociate and hence move faster. In
the separating gel the glycine move immediately behind the chloride and the
sample molecules separate as in zone electrophoresis. Andrews (1986) and
Blackshear (1984) give recipes for the gels and buffer solutions.
Another common method used for PAGE is to treat proteins with sodium
dodecyl sulphate (SDS) (see Figure 11-4c). This method was first developed
by Maizel (1966) and modem methods are spelled out by Andrews (1986) and
Blackshear (i984). When a solution of proteins at pH 7 is treated with 1 w/v%
SDS and O.IM mercaptoethanol to destroy disulfide bonds, the polypeptide
chain is unfolded and the protein is converted to a rod-like molecule. What is
remarkable is that with very few exceptions 1.4 grams of SDS will be bound to
each gram of protein. The result is that each protein is converted into a com-
plex with the same charge-to-mass ratio, and all complexes have essentially the
same electrophoretic mobilities in free solution. The SDS-protein complexes
are different sizes and can be separated on the basis of protein size by using the
sieving property of polyacrylamide gels. SDS-PAGE is commonly used as an
analytical method and to estimate relative masses. The method is also used to
look for subunits since breaking the disulphide bonds breaks apart subunits.
Other gels are also used for electrophoresis. Agarose gels are used
because they can have larger pores than polyacrylamide gels and thus can
separate larger molecules. Agarose is a polymer of D-galactose and 3,6 anhy-
drogalactose. The repeating unit in agarose, agarobiose, is shown in Figure
11-4d. The alternating sugar groups are substituted with sulphate, methoxyl,
pyruvate, and carboxyl groups, and this substitution has a major effect on the
gel properties. Chains are held together by hydrogen bonds. For vertical slabs
an agarose concentration as low as 0.8 w/v% can be used. Molecules as large
as 50x106 molecular weight can migrate into the gel. For horizontal slabs con-
centrations as low as 0.2 w/v% are mechanically stable and will admit
molecules as large as 150xl~ molecular weight into the gel. Agarose gels are
commonly used for electrophoresis of DNA. The agarose gels do contain sul-
phate and carboxylic acid groups which are charged and can cause elec-
troosmosis. This can be minimized with an alkali pretreatment. In addition,
570
the properties of agarose gel are very temperature dependent and precise tem-
perature control is important to obtain reproducible results. Agarose gels are
used in pulsed electrophoresis which is discussed in Section 11.2.5.
Polyacrylamide-agarose composite gels are sometimes used for electrophoresis
of RNA, DNA, and large proteins. The composites have intermediate proper-
ties.
The first gel used for gel electrophoresis was starch (Smithies, 1955).
Starch is cheap and easy to use, but is not commonly used today. It is difficult
to control pore size with starch and there are negative side chains which can
interact with the protein and cause electroosmotic flow. Recipes for starch gel
electrophoresis are given by Andrews (1986).
Filter paper was the first anti-convective medium used for zone electro-
phoresis. The filter paper replaces the gel in Figure 11-3a. This method is con-
venient and simple, but has disadvantages. The paper has fixed negative
charges which interact with proteins and cause electroosmotic flow. An
advance over paper is the use of cellulose acetate membranes since the
hydroxyl groups have been esterified. Cellulose acetate membranes are advan-
tageous for analysis since the membrane can be made transparent with a
mineral oil of the appropriate refractive index.
11.2.3. Zone Spreading in Zonal Electrophoresis
Zonal electrophoresis is similiar to elution chromatography. The linear zone

spreading analysis of Section 7.6. has been extended to electrophoresis (Gid-
dings, 1966; Jorgenson and Lukacs, 1981). A modification of their analyses
will be presented here. For linear systems with a pulse of feed we expect a
Gaussian solution such as Eq. (7-24). This is
(z- ~)2l (11-15)

C; = Cmax,i exp [- 2 at
where O'l,i is the standard deviation of the electrophoresis system and ~ is the
location of the peak maximum. The width of the zone is then 40'1,i. This is
571
shown in Figure 11-3b. The net velocity observed by a molecule is the sum of
electrophoretic migration and electroosmotic flow.
(11-16a)
For homogeneous solutions E = V/L where L is also the maximum length of

migration. Then Eq. (11-16a) becomes
(1l-16b)
unet = J..l V /L + J.l.osm V/L
The migration or retention time for a species migrating an average distance ~

is
~L (11-17a)
tR.i = ZJ~ = ~ + J..losm) V = lexper
where the migration distance is given by
(11-17b)
Because batch electrophoresis is not normally run as an elution system, the

retention time of all peaks is equal to the experimental time, lexpen and is the
same. If the peaks are eluted out the end of the column, then Zi = Land lR.i =
L/~.
The Einstein relationship for spreading due to diffusion can be extended

to all the linear dispersion processes in the electrophoresis system.
(11-18)
O'f,i = 2 D eff,i tR.i = 2 D eff,i lexper
where Deff is the effective axial dispersion coefficient in the electrophoresis
system. Combining Eqs. (11-17a) and (11-18), we obtain
(11-19)
572
Equation (11-19) can be combined with Eq. (7-26b) with a migration distance
~ to obtain.
2 2
I...
+J4sm
IJo'i
)VZ.
1
[~~ + JJum ~ ]2lcxper (1l-20)
Ni = Zdcri = W L
eff,i
= W
eff,i
Note that as long as the value of V/L is high a large N is achieved. The time of
the experiment. lcxpcr. is usually set so that the fastest migrating species travels
a distance L. This is often a dye added as a marlcer.
L L2
lcxper = - - =
Urnax
-;[-----"'J-
+ ~m V
Jl.max
(11-21a)
Then Eq. (11-20) becomes
(11-21b)
This result is somewhat suprising since it says that the efficiency of the electro-
phoretic system (Ni) does not depend on the migration length~. Note that if
we let the marker dye determine L or lcxpcr. then the marker we choose helps
determine N. Equation (ll-21b) implies that the marlcer should have a mobility
close to that of the sample. The easiest way to increase N in electrophoresis is
to increase the applied voltage V.
Since N depends on the experimental time. a somewhat better com-

parison is the rate of generation of plates in plates/sec.
Generauon I
. 0 f pates NI' 2 D eff.l·
= - - = -::-----'---"""]""2
t...., r ~ +~ ~
(11-22)
where Eq. (11-20) was used. The higher the generation of plates the more
efficient the device is.
573
The resolution R between different zones can also be determined. Reso-

lution is again defined by Eq. (7-30). Equation (7-33) gives a convenient way
to detennine R. Substituting Eqs. (11-20) and (11-16) into Eq. (7-30), we
obtain
(11-23)
which simplifies when lex.per is given by Eq. (11-21a) to
(11-24)
In these equation J..I.i is the electrophoretic mobility of species i; and jl, N, and
Deff,;are the arithmetic average values. Note that resolution can be increased
by increasing V.
This analysis of zone spreading is also applicable to free-flow and capil-

lary electrophoresis (see Section 11.2.4.).
Example 11-2. Batch Electrophoresis
A batch electrophoresis experiment is done in PAGE where J..losm =

O.
The measured mobilities and diffusivities of the two proteins of interest
are:
JlA = 1.050 x 10-5 , J..IB = 1.033 x 10-5 cm 2Ns,
The experiment is run at E = 125 volts/em for 2.5 hours. Assume the
width of the feed well is very small. Determine the location of the
peaks (distance from feed well), the peak widths, and the resolution of
the two proteins.
574
Solution
This problem is a straight forward application of the theory
developed in this section. The migration distance is:
which is
and
ZB = (1.033 x 10-5 ) (125) (2.5) (3600) = 11.62 cm

From Eq. (11-18),
0-[; = 2 D err lR,i = (2) (1.5 x 10- 7) (2.5) (3600) = 0.0027 cmz
which gives 01,i = 0.0520 cm which is the same for both proteins.
Width = 4 01 = 4(0.052) = 0.208 cm
Then From Eq. (11-23),
1 V texper Ih 1 [
texper]'h
R = 4" L [ 2 D err ]
(Jll - IJ.z) = 4" E 2 D eff (Jll - IJ.z)
which is
R = 125 [ 2.5 x 3600 ] 'h (1 050 x 10-5 _ 1 033 x 10-5 ) = 0.9202

4 2 (1.5 x 10- 7) ' •
The resolution can be increased by increasing E and by increasing texper
(assuming the plate is long enough so that L > ZA)'

575
11.2.4. Affinity Electrophoresis and Immunoelectrophoresis
Affinity electrophoresis and immunoelectrophoresis are two related methods

which are commonly used for biochemical analysis. In affinity electrophoresis
the gel contains immobilized ligands which can interact with the migrating
species (Horejsi, 1984; Andrews, (1986). The ligand is often chosen to have a
specific biochemical affinity for the species of interest, and thus is similiar to
affinity chromatography (see Section 7.1.). Agarose gel is normally used
because the chemistry of binding a large variety of compounds to agarose is
well known. Usually, the affinity interaction retards the movement of the
desired species. Thus, affinity electrophoresis would be a useful complement
to a normal electrophoresis run since the affinity electrophoresis experiment
could separate compounds which would form a single zone in normal electro-
phoresis.
A variety of immunoelectrophoresis methods are used for clinical ana-

lyses. In classical immunoelectrophoresis a slab containing 1 to 2% agarose is
formed and the proteins (antigens) are separated as in typical gel electro-
phoresis. Then, suitable antibodies are placed in a trough parallel to the path of
migration of the species. The antibodies then diffuse through the gel to the
antigen zones. When antibodies and antigens react they form precipitate if
there is sufficient but not excess of the two reactants. This is the equivalence
region shown in Figure 11-5a. Arcs of precipitates are observed. These arcs
and the various steps in the process are illustrated in Figure 11-5b. If a single
antibody is used it will react preferentially with a single antigen and the result-
ing pattern will be simple. If a broad spectrum of antibodies are used it will
react with most or all of the proteins and a very complex pattern of precipitate
arcs will form. Introductions to classical immunoelectrophoresis are given by
Andrews (1986), Bier (1974) and Melvin (1987).
Rocket electrophoresis is a modification of classical electrophoresis

developed by Laurell (1966). In rocket electrophoresis the agarose sheet con-
tains the antibody of interest before the electrophoresis is started. As the elec-
trophoresis proceeds precipitate forms whenever the amounts of antigen and
antibody are in the equivalence region. The result are the patterns shown in
Figure 11-5c which have been called "rockets". The height of a rocket depends
576
a
Excess Excess
Antibody Antigen
Amount
Precipitate
Amount Protein ( Antigen)
b
Antigen Zones Feed Well
e ~~ ~ 0 ~~ 8
Trough Preci pitate
Arc
j~ ~ ~ ~ ~ ~ ~
Standards Samples
Figure 11-5. Immunoelectrophoresis. a. Precipitation between antigen and

antibodies. Amount of antibody is fixed. b. Classical immu-
noelectrophoresis. c. Rocket electrophoresis.
upon the antigen concentration in the feed. Thus, the method can be made
quantitative by running a series of standards of known antigen concentration at
the same time as the unknown samples are run.
11.2.5. Free Solution and Capillary Electrophoresis
Free solution electrophoresis is done without a supporting medium. Obvi-

ously, moving boundary electrophoresis is an example of free solution electro-
577
phoresis. Free solution electrophoresis can also be done with zonal develop-
ment and in small capillaries. The application of free solution electrophoresis
of most interest to chemical engineers is continuous two-dimensional electro-
phoresis which is discussed in Section 11.2.7.
The use of a free solution is essential for electrophoresis of very large
molecules such as some proteins, RNA, and DNA. (Molecules of molecular
weight less than 150 x 106 are usually done in agarose gels.) In addition, the
electrophoresis of particles must be done in free solution. This is of particular
importance for the separation of whole cells. In free solution electrophoresis
there is no gel or paper to stabilize flow. To prevent convection thin chambers
are normally used. This makes the h3 term in the Rayleigh number, Eq. (11-
12), small, and makes efficient cooling easier so that ~T is also small. Since
the critical Rayleigh number is higher for horizontal plates than for vertical
plates (Ostrach, 1977), the plates are usually laid horizontally.
The zone spreading analysis developed in Section 11.2.3. is applicable to

zonal free solution electrophoresis. Special coatings can be applied to the walls
of the chamber to keep J.I.o.m and hence llosm small. Then, Eqs. (11-23) and
(11-24) show that higher resolutions will be obtained with high applied vol-
tages. The effective axial diffusivity is kept small by use of thin chambers.
Microscope electrophoresis is a free solution method which does electro-

phoresis of colloidal particles between two microscope slides (Bier, 1974). The
migration of particles is directly observed with a microscope. This method was
in use before moving boundary electrophoresis was developed and is still used
for determining J.1 of particles.
Capillary electrophoresis is currently a hot area of research for analytical

chemists (Gordon et ai, 1988; Jorgenson, 1987). Commercial capillary electro-
phoresis systems became available in late 1988. In capillary electrophoresis
(also called capillary zone electrophoresis and high-performance capillary elec-
trophoresis) capillaries which are 25 to 75 J.lffi in inside diameter are used.
With capillaries of this small size cooling is quite efficient and the Rayleigh
number is quite small. In addition, the walls act to inhibit convective motion.
Capillaries above 80 J.lffi show a significant decrease in performance. With
capillaries smaller than 80 J.lffi operation can be done without convection even
578
at voltages as high as 30,000 volts. To dissipate the Joule heating caused by

30,000 volts capillaries up to one meter long are used. Obviously, at this high
voltage stringent safety precautions are necessary. According to Eqs. (11-20)
and (11-23), both N and R will be high at these high voltages. Experiments
routinely show 300,000 to 400,000 plates if very small samples with concentra-
tions as low as 10-5 M are used, and up to one million plates have been
observed. The analysis uses a sample from 1 to 10 nL, and the analysis time
can be as short as ten minutes. Short analysis time is very important when
many samples need to be analyzed. The method has been extremely successful
with small molecules, but significantly less successful with proteins since pro-
teins adsorb to the walls of the capillary. This can be controlled by modifying
the walls of the capillary by appropriate surface treatment. Appropriate surface
coatings are also useful to control electroosmosis. Electroosmosis is often
desireable since it acts as a "pump" which causes fluid flow with an almost flat
velocity profile (Gordon et aI, 1988). Protein adsorption can also be controlled
by operating under basic conditions where both the wall and the proteins are
negatively charged. High concentrations of salts in the buffer will also reduce
interactions with the wall. Currently, protein analysis must be considered as
qualitative not quantitative. These methods are all part of current research pro-
grams (Gordon et aI, 1988; Jorgenson, 1987).
Large scale application of free solution electrophoresis is currently done

in two-dimensional systems (see Section 11.2.7.). Large-scale application of
•••
a b
I -0
I
- 0
••
), l .-.
-.~
•
fa
samPletz-
yt
- z
Figure 11-6. Sequential two-dimensional separation. a. First development.

b. Second development
579
capillary electrophoresis would be of considerable interest, but has major obs-

tacles to overcome before it becomes a reality. Because of the small sample
size and low concentration required for high N in small capillaries, a very large
number of capillaries would have to be bundled in parallel. Each of these
capillaries has to be essentially identical or the separation observed in each will
be different The capillary walls must be quite thin or the volumetric efficieny
will be rather low (porosity will be very low). One possible way around these
problems would be to form the capillaries in a microporous material such as the
polymer blocks being used for chromatography. In addition, methods of elut-
ing the separated components would need to be developed.
11.2.6. Sequential Two-Dimensional Methods
Even with a very large N, molecules with the same electrophoretic mobilities
will not separate. In addition, biological mixtures often have thousands of
components which is more than can be separated and observed in a single elec-
trophoretic experiment. To overcome these problems a variety of sequential
two-dimensional methods have been developed for analysis. The basic idea
behind sequential two-dimensional development is illustrated in Figure 11-6.
The sample is developed in the first direction using an electrophoretic or
chromatographic technique. Then the plate is developed in the second direc-
tion using a different technique which utilizes a somewhat different mechanism
for separation. For example, if the first development uses ordinary electro-
phoresis the second development could use SDS-PAGE. Then the first
development is essentially separating on the basis of the charge to size ratio
while the second development is separating on the basis of size only.
Molecules with the same size-to-charge ratio will not separate during the first
development, but will separate during the second development if the molecular
sizes are different It is important that different experimental conditions be
used for the two runs; otherwise, nothing is gained from doing development in
two directions.
Two-dimensional methods can also separate many more compounds than

one-dimensional systems as long as the two developments are independent.
580
Giddings (1984) showed that the maximum number of compounds n2D that can
be detected by a two-dimensional method is
(11-25)
where n1 and n2 are the maximum number of peaks which can be detected by
the two separate developments. If the two developments are not independent
n2D will be significantly less than predicted by Eq. (11-25).
Sequential two-dimensional separations can be traced back to Haugaard

and Kroner (1948) who did paper chromatography followed by paper electro-
phoresis. Since paper chromatography separates by a combination of adsorp-
tion and ion exchange while paper electrophoresis separates on the basis of ion
exchange and electrophoresis. the two developments are close to independent.
A modification of this method which uses paper electrophoresis followed by
paper chromatography has become commonly known as fingerprinting and is
used for producing peptide maps (Ingram. 1956; Melvin. 1987). A large
variety of similiar sequential two-dimensional methods have been reviewed by
Smith (1968. 1979). These include paper electrophoresis followed by transfer-
ring the spots to a starch gel and then doing starch gel electrophoresis, starch
gel electrophoresis followed by paper or thin layer chromatography, cellulose
acetate electrophoresis followed by electrophoresis on DEAE paper, and other
methods.
O'Farrell (1975) acheived a major breakthrough in sequential two-
dimensional separations by doing iso-electric focusing first followed by SDS-
PAGE. Since isoelectrical focusing separates based on the isoelectric point
(see Table 11-1 and Section 11.4.) and SDS-PAGE on molecular size. the two
developments are independent. This was the first sequential two-dimensional
method with completely independent developments. In addition, both methods
are high resolution methods with high values of n1 and n2 in Eq. (11-25). The
predicted value of n2D is as high as 5000. A large number of related techniques
have since been developed (Andrews. 1986; Gianazza and Righetti, 1979; Les-
ter et al, 1981; Dunn and Burghes, 1983). These include isoelectric focusing
followed by a variety of electrophoretic methods such as: PAGE, gel pore-
gradient electrophoresis, immunoelectrophoresis. and isotachophoresis.
58t
A one or two-dimensional method which is rather different than the

methods discussed up to now is pulsed gel electrophoresis which is being
developed as a method for separating very large DNA molecules (Schwartz,
1987). The basic idea in pulsed gel electrophoresis is to change the direction of
the field so that the DNA molecules must change direction. In the two-
dimensional pulsed gel system there are two pairs of electrodes which are
arranged perpendicular to each other. Every time the field is switched (or
pulsed) the long DNA molecule changes direction. Different size molecules
can adjust to the pulsing with different relaxation times. Smaller molecules can
follow the change in direction quicker and thus should move faster. Thus, the
method can be tuned to separate different size molecules. This method has
caused considerable excitement since it is capable of separating yeast chromo-
somes. The method was first announced in 1983 and a commercial instrument
is available for analytical separations.
There are many other possible combinations of two-dimensional

development methods. Giddings (1984) estimated that there are from let to
106 distinguishable two dimensional methods. Obviously, not all of these com-
binations will be advantageous. Problem ll-BI encourages you to generate
some ideas of your own.
Zone spreading in sequential two-dimensional systems can be described

by an extension of the methods used for one-dimensional systems (Giddings,
1984). The variance for a one dimensional system is given by Eq. (11-18). For
a two dimensional system there will be variances in both directions.
(11-26)
where subscript 1 refers to the first development. After both developments, the
total variances will be
(1l-27a)
(1l-27b)
582
The solution for the total concentration is the sum of the Gaussians in the y and
z directions.
(y - YJ2 (z - ZJ 2 ] (11-28)
Ci (y,z) = ci,max exp [ - 2 O'~ - 2 0';
where n is the total moles of solute and Yi and ~ are the location of the peak
(or spot) center. Note that Yi =~.1 t1 and Zi =Ui.2 t2. Equation (11-28)
represents an elliptical spot on the y-z plane. The widths of the spot are
Wy =4 O'y and W z = 4 O'z. The four effective dispersion coefficients can be
different. Since developments 1 and 2 may represent completely different
processes on different media, it is reasonable that 0'1 and 0'2 will be different.
These effective dispersion coefficients can be measured from experiments.
11.2.7. Continuous Two-Dimensional Electrophoresis
Continuous two-dimensional electrophoresis is the current focus of efforts to

scale up electrophoresis so that it can be used for commercial separations. The
basis of continuous two-dimensional electrophoresis is illustrated in Figure 11-
7a. The feed is introduced at the top of the plate and flows downward while
subject to an electric field which is perpendicular to the flow direction. In the
simplest version of the process all species have the same vertical velocity. The
horizontal veolcities are given by Eq. (11-1) and depend upon the electro-
phoretic mobilities. The net velocity of each component is the vector addition
of the vertical and horizontal velocities (see Figure 11-7b).
u (11-29)
tana=--
Vbuffc:r
The result is a continuous, steady state, multicomponent separation. The one-

dimensional zone spreading analysis developed in Section 11.2.3. can be
applied with slight modifications to the continuous two-dimensional apparatus.
The separation is occurring only in the y direction in Figure 11-7a. The x
direction is a non-selective displacement. Thus, the maximum number of com-
583
y
a b
2h
Products
Figure 11-7. Continuous two-dimensional electrophoresis. a. Apparatus. b.
Vector addition of velocities.
ponents which can be separated is n1. When Eqs. (11-14) to (11-24) are
applied to the continuous two-dimensional apparatus the retention time is the
same for all components.
(11-30)
Now, u is the velocity in the y direction, E = V/W, cJl is measured in the y

direction, and Zj is the distance of migration in the y direction which is Zi =
Di tR· Although the transformation of the one-dimensional equations to the
continuous two-dimensional form is left as Problem ll-C1, the final result for
resolution is given below
'h
1 [
L ]
V (11-31)
R= 4" 2 D eff Vbuffer (Jl1 - J.12) w
Comparison with Eq. (11-23) shows that there is considerable similarity

between one-dimensional and continuous two-dimensional electrophoresis.
The basic process shown in Figure 11-7a has been used for paper electro-
phoresis (Durrum, 1951; Strain and Sullivan, 1951). If there is no adsorption or
584
ion exchange, the results are two-dimensional electrophoresis and the theory
presented earlier can be used to predict the location of peaks and the amount of
zone spreading. If ion exchange or adsorption occur, the separation is a combi-
nation of these effects plus electrophoresis. The paper in Figure 11-7a can
obviously be replaced with a variety of gels (Smith, 1968). If sieving occurs in
the gel, then separation will be a combination of size and charge effects. One
can also use size exclusion chromatography particles (this is different than hav-
ing a gel since the space between particles is available to all molecules) to pro-
duce a combination of size exclusion chromatography and electrophoresis
(Epstein, 1977). The devices with paper and gels have relatively low flow
rates, and apparently have not been scaled-up to much larger sizes.
The major scale-up efforts have been made for continuous, two-
dimensional electrophoresis in free solution. Early versions of this method
were demonstrated by Svensson and Brattsten (1949) in Sweden and Grassman
and Hannig (1950) in West Germany. The flow rates can be significantly
greater than in gel or paper, and particles can be separated. The basic
apparatus is the same as Figure 11-7a. This process has attracted considerable
attention and has been extensively reviewed (Hannig, 1978; McCann et al,
1973; Just and Werner, 1979; Kolin, 1979; and Wankat, 1984-85). Scale-up is
difficult because of the need to control convection and the relatively low
resolving power of the system at modest voltages. Early efforts at scale-up
were aimed at the basic apparatus shown in Figure 11-7a. Since no stabilizing
medium is used, a thin chamber (0.6 to 1.5 mm) with cooling on both sides is
used. The apparatus has usually been used vertically, but horizontal operation
is probably preferable since the critical Rayleigh number is much higher. Use
of chambers with a width less than 5mm will stabilize the flow, but limits the
separation which can be acheived. Special 'fan' collectors need to be used for
these very narrow designs (Kolin, 1979). Currently obtainable volumetric
throughputs using the basic apparatus are about 4 mL/hr. A commercial
apparatus called the Elphor YaP 21 is available from Bender and Hobein of
Munich (Mosher et al, 1987). This device is quite versatile and can also be used
for isotachophoresis and isoelectric focusing (see Sections 11.3. and 11.4.).
Devices similiar to Figure 11-7a have been flown in the space shuttle as a
way to prevent thermal convection since gravity is very small (Morrison et al,
585
1984). Both the Rayleigh and Grashoff numbers, Eqs. (11-12) and (11-13),
will be small since g is small. This approach works, but is rather expensive.
Once it became clear that the basic geometry had almost insunnountable
difficulties, a variety of clever methods were devised to overcome these
difficulties. Perhaps the first attempts to stabilize the flow was the addition of a
density gradient (Philpot, 1940; Mel, 1964). Chemicals such as sucrose or
cesium chloride are added to the carrier fluid in layers of increasing density.
Migration is then normal to the density gradient in either a vertical or an hor-
izontal device. If the added density gradient is steep enough, thermal convec-
tion can be prevented. A related approach is to add a thickening agent such as
methylcellulose to increase the viscosity, 11 (Dobry and Finn, 1958). This
reduces the Rayleigh number in Eq. (11-12) and may prevent convection. If
convection occurs it will be reduced since the Grashof number in Eq. (11-13)
becomes smaller. However, increasing viscosity also decreases J..li (see Eq.
(11-9c», and probably decreases resolution (Eq. (11-31». Unfortunately, both
these methods require addition of chemicals to the separating mixture and these
chemicals must eventually be removed.
The fluid can also be stabilized by superimposing a rotational flow per-
pendicular to the migration direction. The most direct way of doing this was
developed by Philpot (1973). Philpot's device is shown schematically in Fig-
ure 11-8 (philpot, 1973; Beckwith and Ivory, 1988; Mosher et al, 1987). Car-
rier fluid flows vertically between two concentric cylinders which are separated
by a gap of about 3 to 5mm. The inner cylinder is the cathode and is fixed.
The outer cylinder serves as the anode and rotates at speeds up to 150 rpm.
This rotation produces a steady angular flow in the annulus which is superim-
posed on the vertical parabolic flow. Feed is introduced into the carrier along
the entire circumference of the inner cylinder. Migration of the molecules is
radially outward, and product is collected at different radial positions but
around the entire circumference of the annulus. Despite the 5mm distance
between cylinders, 29 separate fractions are collected in the commercial Bios-
tream device. Since the feed is introduced along the entire circumference of
the inner cylinder, the feed rate which can be processed is much higher than
with the parallel plate device shown in Figure 11-7a. This device can process
protein solutions at throughputs greater than two liters per hour which
586
Products
8 c±)
Migration Fixed"""-
Routes
Rotor
)
Figure 11-8. Philpot-Harwell rotating electrophoresis device.
Figure 11-9. Endless fluid belt electrophoresis (Kolin, 1979). Reprinted with
permission from P.G. Righetti, C.J. van Oss, and J.W. Van-
derhoff (Eds.), Electrokinetic Separation Methods,
Elsevier/North Holland Biomedical Press, Amsterdam, 1979.
Copyright 1979, Elsevier/North Holland Biomedical Press.
587
corresponds to a maximum of 150g of protein per hour. On the other hand, the
resolution and the number of components which can be separated is higher in
the parallel plate device. About a 20°C temperature increase is observed in the
flowing liquid, but thermal convection is totally prevented. To prevent thermal
degradation of proteins the feed is first chilled to about 2°C. The short 15 to 60
second residence time also helps to prevent protein degradation. Beckwith and
Ivory (1988) presented a detailed analysis of the solute trajectories. The
Philpot-Harwell device is commercially available from Harwell as the Bios-
tream separator. Unfortunately, further scale-up of the device is likely to be
difficult.
Kolin (1979) devised an alternate method of superimposing a rotational
flow on the flowing axial fluid. Kolin put a magnet inside the loop of fluid to
produce a rotational flow which represses thermal convection. Several dif-
ferent devices were developed; one of the more sophisticated, a racetrack
design, is shown in Figure 11-9 (Kolin, 1979). The combination of rotational
flow and particle migration causes a helical flow path for all particles. The
addition of a carrier flow in the same direction as the electrical field serves to
change the pitch of all particles. The particles can be made to form several
loops; then
(11-32)
Distance Traveled, L = n x circumference
where n is the number of turns traveled. This is equivalent to recycling the

fluid. According to Eq. (11-31), larger L increases resolution. Thus, this dev-
ice has excellent resolution but limited throughput. Further scale-up will prob-
ably be difficult.
A rather different approach is to use continuous free flow two-

dimensional electrophoresis with a series of recycles to produce what is effec-
tively a continuous, steady-state, countercurrent electrophoresis (Gobie et al,
1985). A schematic of this device is shown in Figure 11-10. Feed is intro-
duced in the center and purge at the two ends. These are the only locations
where fresh solution is added; thus, products withdrawn from the bottom will
not be excessively diluted. Liquid leaving the bottom of the apparatus is recy-
cled to the top where it provides the buffer flow. The recycle streams are
588
Low Mobility Ports Recycle High Mobility

Product Lines Product
Figure 11-10. Recycle two-dimensional electrophoresis operating as a con-

tinuous countercurrent electrophoresis system. Modified from a
Figure in Gobie et aI. (1985).
shifted one or more ports back. This shift is important since it essentially pro-
duces the countercurrent motion. The solute retention time for one pass is
given by Eq. (11-30) and the solute migration velocity in the y direction is
given by Eq. (11-16a) where E = V/W. The net lateral movement of a solute i
is
(11-33)
Yi,net = Ui.net L/vbuffer - Yrecycle
where Yrecyc!e is the distance the recycle stream is shifted. IfYi,net > 0, then the
net solute migration is to the right These solutes will eventually exit as high
mobility product. All solutes with Yi,net < 0 (includes positive charge and no
charge) will have net migration to the left. These solutes eventually end up as
low mobility effluent The regenerators shown in Figme 7-9 have the same
purposes as zones 3 and 4 in counter-current adsorbers (Figure 10-11). The
right regenerator where Yi,net < 0 for all components is analogous to zone 4
while the left regenerator where Yi,net > 0 for all components is analogous to
zone 3. These regenerators allow the use of reflux at both ends and elute
without excessive dilution. The slope of the left regenerator must be adjusted if
there are positively charged solutes in the feed The device shown in Figure
11-10 acts as a steady-state, countercurrent separator and separates the feed
589
into two products. Thus, it is a binary separator. Because of the large amount
of recycle, separation should be quite sharp even for solutes with close to the
same mobility. This approach is still experimental, and is it not clear if this
apparatus can be scaled-up.
11.3. ISOT ACHOPHORESIS (DISPLACEMENT ELECTROPHORESIS)
Isotachophoresis can be considered as a displacement form of electrophoresis,

and it has an analogous relationship to zonal electrophoresis as displacement
chromatography does to elution chromatography. The theory of isotacho-
phoresis was developed in 1897 by Kohlrausch, but serious developmental
work did not begin until the 1940's when A. J. P. Martin and F. M. Everaerts
collaborated on the process (see Everaerts et al, 1976; and Everaerts and
Verheggen, 1987). The method did not become popular until the late 1960's.
Currently, it is a fairly common analytical technique and has considerable
potential as a large-scale method. When stacking gels are used in electro-
phoresis (see Section 11.2.2.), the stacking gel is operating as an isotacho-
phoresis system.
In isotachophoresis the sample is placed between a leading (L) and a

trailing (1) electrolyte. The pH of the mixture is adjusted so that all solutes
(such as amino acids or proteins) have Lie same charge. To be specific we will
assume they are all cations. The leading electrolyte is chosen to have a higher
mobility than all the solutes and the trailing electrolyte is chosen to have a
lower mobility than all the solutes. Thus, L is analogous to the solvent and T to
the displacer in displacement chromatography. Recipes for leading and trailing
electrolytes are given by Everaerts et al (1976) and Holloway and Battersby
(1984). After the current is turned on, a series of zones are formed with sharp
boundaries between the zones. The zones are in the order L, fastest solute, next
fastest solute, ... , slowest solute and then T, (See Figure ll-lla). Each zone is
concentrated and contains essentially pure solute in solution plus the coun-
terion. Operation is with constant current. Since the mixture is concentrated
and is not homogeneous, the field strength E is not constant but varies from
zone to zone. The field strength is lowest in zone L and highest in zone T, (see
590
a
81: T llG
Anode Cathode
~
Distance
y
T
1
8
Anode
G
Cothod, j
L
W ~I
Figure 11-11. Isotachophoresis of cations. a Schematic of bands. b. Electric

field strength. c. Continuous two-dimensional system.
Figure l1-l1b). Since Joule heating is proportional to E, the temperature

becomes highest in the tenninating electrolyte zone. Thus, vertical equipment
should be arranged with L on the bottom, T on top, and downwards migration.
lsotaccphoresis can be done in paper, gels, capillaries, and free solution
(Andrews, 1986; Everaerts et aI, 1976; Everaerts and Verheggen, 1987; Hollo-
way and Battersby, 1984; Mosher et aI, 1987). Paper, gel and capillaries are
operated in batch mode while free solution is usually operated continuously.
Modified electrophoresis equipment is lJSually used.
The theory of isotachophoresis is presented in considerable detail by

Everaerts et aI (1976). In this section we will present a simplified theory for
dilute systems with completely ionized salts which closely follows the
591
development of Karol and Karol (1978). In isotachophoresis pure bands with

constant concentrations are observed experimentally. To satisfy a mass balance
each band must have constant velocities
(11-34)
which is analogous to Eq. (8-39) developed for displacement chromatography.

Substituting in Eq. (11-1), we obtain
(11-35)
Equation (11-35) is known as the isotachophoretic condition. This equation

differentiates isotachophoresis from the other electrophoretic methods. Since
the mobilities are in order J.lL > J.l1 > J.l2 > ... > J.lT, the field strengths in each
band must be in the reverse order, ~ > En> ... > E1 > EL, (see Figure l1-1Ob).
Any solute molecule diffusing into a slower band experiences a field strength
which is higher, and thus the solute is accelerated back into the correct band.
Thus there is a focusing effect which keeps the bands sharp despite diffusional
effects.
The total current I is the sum of the current carried by cations and anions.
(11-36)
where z is the valence, e = 1.6 x 10-19 Coulomb, and ze is the charge on each
ion. N is Avogadro's number 6.022 x 1023 , Ac is the cross-sectional area of the
apparatus, and c is the concentration in g moles/cm3 • Note that eN = F,
Faraday's constant. Electroneutrality requires that
(11-37)
Substituting this condition into Eq. (11-36), we obtain
(11-38)
which becomes
(11-39)
592
when we substitute in Eq. (11-1). Since the current I is constant, Eq. (11-39) is
valid for i = L,I,2, ... ,T. Setting Eq.(11-39) equal for components Land i, and
substituting in Eq. (11-35) to remove the field strengths, we obtain
(11-40)
This result, which is valid for isotachophoresis of dilute systems with com-
pletely ionized salts, allows the calculation of the concentration in each band.
Obviously, if CL is fairly high then Ci in the band will also be high. Note that Ci
does not depend on the feed concentration of component i. Thus, if the feed is
dilute rather high concentration ratios can be achieved. In fact, the solubility of
proteins places a limit on the value of CL which can be used (Holloway and
Battersby, 1984). The band widths can be determined from a mass balance
which is
(11-41)
CF,i (feed width) = Ci.band (bandwidth)
This is analogous to Eq. (8-43) developed for displacement chromatography.
This solution is oversimplified since it ignores the transient period before

the bands are fully formed. Thus it is again analogous to the theory used for
displacement chromatography where the transient period was also ignored.
Everaerts et al (1976) give a more detailed analysis.
Because of electroneutrality there can be no gaps between bands. If one

or more bands are quite narrow it can be difficult to recover the solute as a pure
component. Spacers can be advantageous in this case. A spacer is an ion with
a mobility in-between the mobilities of the two solutes of interest. The spacer
will then separate these solutes. Since the spacer concentration is controlled by
the experimenter, Eq. (11-41) shows that the spacer band can be any desired
width. Unfortunately, finding appropriate spacers may require a lot of trial and
error. A variation of the spacer method is the field step method (Mosher et al,
1987). In this method the feed is introduced into a buffer of low conductivity
which is sandwiched between two buffers of high conductivity. The field
strength is high in the low conductivity buffer and thus the solutes migrate
593
quickly in this buffer. In the high conductivity buffers the field strength and the
solutes slow down; thus, solute accumulate at the two interfaces. This
approach is thus a binary separation.
Isotachophoresis has considerable potential as a preparative method. The

capacity is much greater than in conventional electrophoresis or isoelectric
focusing (Karol and Karol, 1978). Because of the focusing and concentrating
effects, rather large feed pulses can be used in batch operation. In continuous
operation the feed port can be quite wide. A commercial free-flow electro-
phoresis apparatus, the Elphor VaP 21, is readily adaptable to continuous,
free-flow isotachophoresis (see Section 11.2.7.). In continuous isotacho-
phoresis as shown in Figure 11-l1c the feed is introduced between the leading
and tenninating electrolyte ports (Mosher et al, 1987). Electrophoretic migra-
tion is in the y direction while everything flows at the same velocity in the x
direction. Capacity of the commercial instrument is up to 5g protein per hour.
In the field step method the Elphor VaP has processed proteins at the rate of
30g per hour.
Example 11-3. Continuous Two-Dimensional Isotachophoresis
Continuous two-dimensional isotachophoresis is run in an apparatus

similar to Figure 11-11c. The leading electrolyte is 0.04 N H+. The
trailing electrolyte is 0.01 N Li+. The coion is cr. The feed contains
O.OlN RbCI and 0.01 N KCI. The apparatus has L = W = 12 cm. The
trailing electrolyte is input from y =0 to y =2.5 cm (measured from the
anode), the feed is then input for 7.0 cm, and then the leading electro-
lyte (see Figure l1-11c). Find the Rb+ and K+ concentrations in the
bands and the band widths once steady state has been obtained. Data is
available in Table 11-3.
Solution
Equation (11-40) can be applied. From Table 11-3,

~L = 36.2 x 10-4, ~Rh = 8.03 x 10-4, ~K = 7.67 x 10-4, ~Li = 4.02 x 10-4
594
and IJlcll = 7.9 x 10-4 cm 2Ns. Then from Eq. (11-40),
= (0 04) (36.2 + 7.9) (8.03) Ql = 0.0248 N

CRb,band • (8.03 + 7.9) (36.2) (1)
c = (004) (36.2 + 7.9) (7.67) (1) = 0 0240 N

K,band • (7.67 + 7.9) (36.2) (1) .
From Eq. (11-41),
bandwidth RbCl = (0.01) (2.5 cm) / (0.0248) = 1.0081 cm
bandwidth KCI = (0.01) (2.5) / (0.0240) = 1.0415 cm
The order of the ions will be H+, Rb+, K+, and then Li+. Detennination
of other details about the operation such as the distance required to
reach steady state and the exit locations of the bands require more com-
plicated theories (Evereaerts et ai, 1976).
11.4. ISOELECTRIC FOCUSING
Isoelectric focusing is electrophoresis in a pH gradient. Remember that the

size and sign of the charge on proteins depends on the pH, and at the isoelectric
pH (called the pI) the protein has no net charge. This is shown in Figure 11-
l2a. Note that at pH values lower than the pI the proteins have a positive
charge, and they will migrate towards the cathode. At pH values higher than
the pI the proteins have a negative charge and will migrate towards the anode.
For isoelectric focusing the pH gradient is set up with low pH values at the
anode and high pH values at the cathode (see Figure 11-12b). Thus, in a sta-
tionary pH gradient proteins will migrate until they reach pH = pI and then they
stop since there is no force on the protein. At this point the mobility J..L = 0
since ~ = O. Since proteins from both higher and lower pHs migrate to the pI
position, the protein is focused at the pI. The final separation achieved is an
595
a
+
Charge Of----!t------+--+----
4 5 16 7 8 \,0
I pH I I
I I
-I
I
b I I
z I
pH 4 5 I6 7 8 19 110
I
81 18
I I
Anode I B C Cathode
e I I I
I I
I
Jv\
cone. I
A
Distance
Figure 11-12. Isoelectric Focusing. a. Effect of pH on charge of proteins. b.

Focused bands. c. Concentrations.
equilibrium separation with each protein focused at its pI. This is shown
schematically in Figures 11-12b and 11-12c. Molecules which differ by as lit-
tle as 0.02 pH units can be separated. The resulting bands can be very concen-
trated compared to the feed concentrations. lsoelectric focusing is a very
powerful separation method which has been extensively used for separation of
biological compounds.
We will first briefly consider the history of isoelectric focusing since that
will clarify the difficulties which had to be overcome to make the technique
successful. Then, a simple zone spreading theory will be presented. Finally, a
variety of preparative methods for isoelectric focusing will be reviewed.
596
11.4.1. History and Basic Ideas
Righetti (1983) has presented a detailed and very readable account of the his-
tory of isoelectric focusing. In 1954-1955 A. Kolin developed a method for
focusing ions in a continuous pH gradient stabilized with a sucrose density gra-
dient. His pH gradients were produced by placing the sample between acidic
and basic buffers, applying the electric field and allowing diffusion to produce
the pH gradients. Unfortunately, the pH gradients were transient and long-tenn
focusing runs were not possible.
In the late 50's and early 60's Harry Svensson (now Rilbe) did funda-
mental theoretical work which predicted the requirements of a stable isoelectric
focusing system (Svensson, 1961, 1962a, 1962b; Rilbe, 1973). To generate a
stable pH gradient a carrier ampholyte was required. A carrier ampholyte is:
1. Amphoteric. That is, the sign of the charge on the molecule is

pH dependent. Amphoteric compounds will reach an equilibrium
position in a pH gradient.
2. A carrier. This means that the species can carry current which
requires that it be a good conductor, and that it can 'carry' pH
which requires that it be a good buffer. The value of !pI - pKll
should be small.
3. In addition, the carrier ampholyte should be quite soluble, have

no effect on the sample, and be easily separable from the sample.
Whilst developing the theory Svensson searched through chemical catalogs for
appropriate compounds. He was partially successful, but no appropriate com-
pounds were found in the very important range between pH 5 and 7. The first
seven compounds listed in Table 11-2 are from Svensson's list (Svensson,
1962a). Svensson then found that peptides fonned from cleaving proteins
could be used to fonn a partially successful carrier ampholyte.
Svensson encouraged Olof Vesterberg to join in the search for a suitable

597
carrier ampholyte. In 1964 Vesterberg was successful. He coupled acrylic

acid to a series of amines to produce a series of compounds with pI and pK
values in the range from 3.5 to 10 (Vesterberg, 1969). Subsequent work
extended the pH range from 2 to 11. These carrier ampholytes are commer-
cially available as Ampholine from LIrn. Later, Serva and Pharmacia
developed different carrier ampholytes.
Once carrier ampholytes were available applications immediately fol-

lowed. Since much of the equipment was developed from existing electro-
phoresis designs, equipment development was rapid. Preparative systems
include gel, density gradient, free flow and recycling systems. These are dis-
cussed in Section 11.4.3.
In 1982 Righetti's group (Bjellqvist et al, 1982) announced the develop-

ment of immobilized pH gradients. Acrylamide derivatives of the structure
CH2 = CH-C~NH-R
where R is either a carboxylic or tertiary amino group will form a series of

buffers with different pK values (Dunn, 1987). When these buffers are mixed
with acrylamide and Bis (see Section 11.2.3.), they can be polymerized to form
a gel with a pH gradient. Since the compounds forming the pH gradient are
part of the solid matrix, the pH gradient is completely stable. These gels are
available as Immobiline gels from LIrn. The use of immobilized pH gradients
has generated considerable excitement. but is currently only of use with PAGE.
Thus, carrier ampholytes must be used with other gels and in free solution. The
use of carrier ampholytes and immobilized gels simultaneously has proven to
be advantageous since the carrier ampholytes increase the rather low conduc-
tivity of the immobilized pH gradients (Righetti et al, 1987).
11.4.2. Theory of Isoelectric Focusing
Once focusing is complete. isoelectric focusing is a steady state process where

the electrical focusing effect is just balanced by diffusion. The mass balance at
598
steady state for any component which focus is (Righetti, 1983; Svensson, 1961)
d (q.lE) = ~ D dc (11-42)
dz dz dz
This equation is not valid for compounds which do not focus (neutral or not
amphoteric compounds). It is convenient to set z=O at the center of the focused
band with positive z towards the cathode (see Figure 11-12b). Doing the first
integration we obtain
de (11-43)
cJ.lE=D-
dz
where the constant of integration must be zero since the protein concentration
is zero outside the focused zone. The left hand side is Cu which is flow of pro-
tein into the zone due to the electrophoretic force. The right hand side is the
diffusion away from the zone.
Before doing the next integration, we need to know how J.l varies with
distance z. Define
dJ.l -dJ.l d(pH) (11-44)
p=-"dZ= d(pH) ~
Because of the choice of axis, d(PH)/dz is positive. The value of d(pH)/dz can
be determined experimentally by locating marker compounds such as those
listed in Table 11-2. dJ.LId(pH) is always negative since C in Eqs.(11-9)
decreases from positive to negative as pH increases. Thus, p in Eq. (11-44) is
an inherently positive quantity. p is considered to be a constant since the
focused bands are quite narrow. Then dJ.LId(pH) will be approximately constant
and d(pH)/dz can be linearized. At the center of the band (z=O) J.l = O.
Integrating Eq. (11-44), we obtain
(11-45)
J.l = - pz
After substituting this result into Eq. (11-43), we obtain
(11-46)
599
If p, E and D are all constant in the focused zone, Eq. (11-46) can be
integrated.
-PEZ 2 } (11-47)
C= cmax exp { 2D
When z=O, C = Cmax which is the maximum concentration in the focused zone.
Equation (11-47) is a Gaussian distribution and is shown in Figure l1-12c.
Comparing this result with Eq. (7-24), we see that the standard deviation
for isoelectric focusing is
(11-48a)
Substituting in Eq. (11-44), we also have
0"= (11-48b)
The standard deviation will be small for molecules which have low diffusivities
and high values of dWd(pH). These conditions are both valid for proteins
which means that isoelectric focusing will give sharp bands for proteins. How-
ever, D is the effective diffusivity in the isoelectric focusing system. Thus, it is
important to stabilize the system to prevent convection. This is done with a
gel, density gradient, thin chambers, or other methods.
The conditions required for adequate resolution can be derived from this
Gaussian solution (Righetti, 1983; Rilbe, 1973). If the peak to peak distance
between two adjacent zones is equal to three times the average value of cr, then
the peaks will overlap but will be definitely resolved. This leads to the differ-
ence in pH of the two peaks being
~pH = ~ d(pH) = 3cr d(pH) (11-49)

dz dz
600
Substituting in Eq. (11-48b), we obtain the following condition for adequate

resolution.
i5 [d(PH)/dz] (11-50)
!J. pI = 3
E [ - d J.l/d(pH)]
This result shows that the required difference in pI values of the two proteins is
less for molecules with low diffusivity and high values of -{ij.J/d(pH), and less
for systems with shallow pH gradients and high field strengths. The estimated
minimum value of !J.pI is about 0.02 pH units with carrier ampholytes. One
advantage of immobilized pH gradients is that quite shallow pH gradients can
be generated to maximize the resolution of proteins with close to the same pI.
The minimum value of !J.pI may be as low as 0.002 pH units (Dunn, 1987).
Example 11-4. Isoelectric Focusing
An IEF experiment is done with a combined immobilized pH gra-

dient on PAGE and carrier ampholytes. Operation is at steady state
at 25°C and E = 80 V/cm. We observe horse myglobin at 5.12 cm
from the anode and histidine at 7.84 cm from the anode. We wish to
separate two quite similar proteins whose pI is about 7.4. When we
do IEF of one of these proteins the peak width is 0.21 cm. What is
the minimum value of !J.pI for which we can still separate the two
proteins?
Solution
The pH gradient can be detennined from the two marker compounds

horse myglobin and histidine. From Table 11-2
pImyglobin = 7.33 and pIhist = 7.47.
d(pH)
Th en ~ = 7.47 -7.33 00515 H .
7.84 _ 5.12 =. p. uOlts/cm.
I
601
Since the proteins we want to separate are similar. D I [ ;~~ ] will

be approximately equal for them. We can combine width = 4cr with
Eq. (11-4Sb),
(11-51)
width = 4
Then,
(11-52)
From the measured width and pH gradient,
~_D_~ = [0.21 cm]2 SO ~ (0.0515) pH units = 0.01136

4 em em
[
- d~~ ]
Now Eq. (11-50) gives,
~pI = 3 .
A
'VI(- dWd(PH)
i5 d(pH)/dz
E
or
~pI = 3 ..../(0.01136) (0.0515/S0) =O.OOSI pH units

Note that this is within the guidlines for the minimum ~pI value
when immobilized pH gradients are used.
602
11.4.3. Methods of Preparative Isoelectric Focusing
A variety of methods for preparative IEF have been developed. Most of these
methods are based on modifications of electrophoresis systems. In fact, many
commercial electrophoresis systems are designed so that electrophoresis, isota-
chophoresis and isoelectric focusing can all be done in the same apparatus.
Gel isoelectric focusing is commonly used for small-scale preparative

applications (Andrews, 1986; Dunn, 1987; Fawcett, 1973; Radola, 1984;
Righetti, 1983; Righetti et al, 1987). In IEF sieving is undesireable since siev-
ing will slow down the attainment of the steady-state. Thus, a gel with large
pores is desired. Either PAGE polymerized to form a continuous layer or beads
of Sephadex or Biogel are used. The continuously polymerized PAGE systems
allow the use of immobilized pH gradients, but the recovery of products is
more difficult (Dunn, 1987; Radola, 1984). The PAGE systems dominate
analytical applications, but are less commonly used for preparative applica-
tions. Although columns can be used, slabs are preferable since cooling is
more efficient. A variety of units are commercially available (Radola, 1984).
Gel thicknesses of up to about 5mm are used since beyond this thickness ther-
mal gradients start to skew the zones. Granulated gels such as Sephadex G-200
or Biogel are commonly used for preparative IEF. Recovery of the proteins is
easier, and these gels can be used in two-dimensional devices such as those dis-
cussed next (see Figure 11-13).
Two-dimensional free-flow electrophoresis equipment is easily adaptable

to isoelectric focusing (Fawcett, 1973; Just and Werner, 1979; Righetti et al,
1980; Righetti, 1983). A comparison between free-flow electrophoresis and
free-flow isoelectric focusing is shown in Figure 11-13 (Fawcett, 1973).
Ampholyte is added to all the entering streams in Figure 11-13b and focuses as
it flows down the column. Prefocusing the ampholyte will speed up the focus-
ing of the sample. Note that free flow isoelectric focusing can be advantageous
since a wider band can be fed and the products are concentrated into narrow
bands while in electrophoresis the bands are spread. Obviously, more than two
solutes can be separated. The free-flow apparatus is also very good for separat-
ing cells or particles (Just and Werner, 1979). Several variants of the basic
apparatus have been developed and are discussed in detail by Righetti (1983).
603
Figure 11-13. Diagrammatic representation of continuous flow systems. a.

electrophoresis, b. Isoelectric focusing. Liquid flows at right
angles to the electric field. From Fawcett (1973).
Density gradients can be used to stabilize IEF from convection (Fawcett,

1973; Righetti, 1983). Density gradients can be done in batch systems, and
commercial instruments based on Svensson's design are available. However,
the main interest for chemical engineers is in continuously flowing two-
dimensional systems. One way of doing this is illustrated in Figure 11-14
(Fawcett, 1973). A sucrose density gradient is used to increase the density at
i
G
L
!.
Base ~~
s~~~~:@3 ~ P""
--:::J
A "d" --"""""L C
~---r-------------------t=
sucrose---' G
Figure 11-14. Continuous-flow density gradient isoelectric focusing
apparatus. Cathode along the top, anode in side limb. Density
gradient flows from left to right Modified from Fawcett
(1973).
604
the bottom of the system. If the system is cooled from the bottom only, the
temperature gradient will help stabilize the system.
One problem with the devices shown in Figures 11-12b and 11-14 is the
residence time may be too short to complete focusing. If the flow rate is
decreased, throughput obviously drops. Ifthe voltage is increased, cooling can
be a problem. One solution to this problem is to use the continuous flow, recy-
cling IEF apparatus shown in Figure 11-15 (Righetti et ai, 1980; Righetti, 1983;
Bier et ai, 1986; Mosher et ai, 1987). This device uses a compact focusing cell
with a cathode to anode distance of 3 em. The cooling is done in a separate
heat exchanger which cools the recirculating streams. This arrangement effec-
tively uncouples the focusing and cooling functions which are done simultane-
ously in Figures 11-13 and 11-14. The flow channels in the focusing cell can
Figure 11-15. Continuous flow, recycling isoelectric focusing.

605
be separated by PVC filters or Nylon screens. Operation is batch since a sam-

ple is added and recirculated until separation has been acheived; then, the pro-
ducts are withdrawn. Recently, it has been found that separation is improved if
a thin film of fluid is focused with no membranes, filters or screens (Mosher et,
1987). This modification allows for a continuous pH gradient instead of the
step gradients obtained with ten separate channels. The continuous flow, recy-
cling IEF system is available commercially.
11.S. ELECTROCHROMATOGRAPHY
If one is doing electrophoresis with a stabilizing medium, it is very appealing to

try chromatography simultaneously. This simultaneous electrophoresis and
chromatography is called electrochromatography, and was tried by both Dur-
rum (1951) and Strain and Sullivan (1951). These experiments used paper as a
support medium in a two-dimensional apparatus. Because of the charged
groups on the paper, some ion exchange chromatography is almost unavoid-
able. The combination of chromatography and electrophoresis can be used to
separate compounds which would be difficult to separate by electrophoresis
alone. Nerenberg and Pogojeff (1969) applied an axial electrical field to a
Sephadex size exclusion column. With this one-dimensional system they found
a significant increase in resolving power compared to SEC alone. Epstein
(1977) did combined SEC and electrophoresis in a continuous two dimensional
system. He found that the Sephadex gel helped to stabilize the system and
prevent convection. (Sephadex beads are commonly used in IEF to prevent
convection, see Section 11.4.3.) None of these methods appear to have been
adopted on a large scale.
O'Farrell (1985) developed the counteracting chromatographic electro-

phoresis method. This method has caused considerable interest particularly
among chemical engineers since the method appears to have promise for large
scale separations. An extensive review and analysis has been published by
Locke and Carbonell (1989).
The apparatus for counteracting chromatographic electrophoresis is

606
Upper Electrode/Electrolyte
Constant
Voltage or
Current
Power
Supply Polyacrylamide Gel
Particles
Polyacrylamide
Plug
Lower Electrode
Electrolyte
Chamber
Figure 11-16. Apparatus for counteracting chromatographic electrophoresis

(Locke and Carbonell. 1989). Reprinted with permission from
Separation and Purification Methods, 18. 1 (1989). Copyright
1989. Marcel Dekker.
shown in Figure 11-16 (Locke and Carbonell; 1989). The pH is adjusted so

that the desired proteins have a negative charge; then, all the desired proteins
migrate upwards to the anode. At the same time a downwards flow of feed and
buffer solution is superimposed on the system. The net velocity of a protein
molecule is then
(11-53)
unet =~E+~sm E-us
where Us is given by Eq. (6-22) with the adsorption parameter A O. Usually. =

the electroosmotic flow will be negligible, J..I.osm = O. The top layer of particles,
polyacrylamide in Figure 11-16. is chosen so that all of the desired proteins are
excluded from the pores, ~ = O. The flow rate and electrical field are adjusted
so that the net movement of desired protein is downwards in this layer. The
second layer of particles, agarose in Figure 11-16, is chosen so that the desired
607
protein can enter all the pores, Kc! = 1. With the proper choice of velocity and
electrical field the net velocity of the desired protein given by Eq. (11-53) will
be positive in the agarose. The result will be focusing of the desired protein at
the interface of the two gels. As the protein concentration increases
significantly in the focused zone, the electrical conductivity will increase. Thus,
the electrical field E will locally decrease and J..LE in Eq. (11-53) will decrease.
At high enough protein concentrations the net velocity will become zero or
negative. The result is the development of an equilibrium zone containing
quite concentrnted protein in the agarose gel layer. As more protein is added
the size of this equilibrium zone will increase.
The engineer can adjust operating conditions so that proteins which are
not desired will not focus at the gel boundary. If these proteins have a positive
charge, there is no upward force and they will be swept downwards to the
cathode. Negatively charged proteins will not focus if they have a net upwards
velocity in the polyacrylamide layer which can happen if Kd > O. These pro-
teins also will not focus if they have a net downwards velocity in the agarose
layer which can occur if Kc! < 1. The result is the equilibrium zone will be both
significantly concentrated in the desired protein, and will be somewhat purified.
This method will probably not separate very similiar proteins.
An obvious extension which O"Farrell (1985) tried is to have a series of

layers of gels with different sixe exclusion limits. These layers can easily be
generated by mixing two gels with different pore sizes. Different proteins will
then focus at different borders. O'Farrell showed that the protein could be
moved from one equilibrium zone to another by changing the voltage.
The method appears to have promise as a preparative technique since

O'Farrell (1985) reported that the capacity was 1000 times that of electro-
phoresis. Collection of the proteins could be done through side taps placed at
the borders of different gel layers. Feed can be added for quite a long time if
the equilibrium zones have a concentrntion significantly greater than the feed
concentrntion. This will usually be the case since protein feeds are often quite
dilute. One possible problem is precipitation of the protein if the concentrntion
becomes too high. Continuous operation could be achieved by continuously
withdrawing a small side stream from each equilibrium zone.
608
This chapter is an introduction to electrophoretic separation methods and a pro-

gress report on large-scale applications. At the end of this chapter you should
be able to:
1. Understand and explain the basic concepts of electrophoresis, isotacho-

phoresis, isoelectric focusing, and electrochromatography.
2. Predict the zone spreading for electrophoresis, isotachophoresis, and
isoelectric focusing.
3. Be familiar with analytical applications of the electrophoretic methods.
4. Explain the advantages and disadvantages of various preparative methods.
REFERENCES
Andrews, AT., Electrophoresis, 2nd ed., Oxford University Press, Oxford,

1986.
Beckwith, J.B. and C.F. Ivory, "The influence of diffusion on elution profiles in
the Philpot-Harwell electrophoretic separator", Chern. Eng. Commun. 54, 301
(1987).
Bier, M., "Electrophoresis", in B. Karger, L.R. Snyder and C. Horvath, (Eds.),

An Introduction to Separation Science, Wiley, New York, 1974, Chapter 17.
Bier, M., N.B. Egen, G.E. Twitty, R.A Mosher, and W. Thormann, "Prepara-
tive electrophoresis comes of age," in CJ. King and J.D. Navratil (Eds.),
Chemical Separations, Vol. I, Principles, Litarvan Literature, Denver, 1986, p.
133-151.
BjeUqvist, B., K. Ek, P.G. Righetti, E. Granazza, A Gorg, R. Westermeier, and

W. Postel, J. of Biochem. Biophys. Methods, 6, 317 (1982).
609
Blackshear, Pl., "Systems for polyacrylamide gel electrophoresis", in W.B.

Jakoby (Ed.), Methods in Enzymology, Vol. 104, Enzyme and Related Tech-
niques, Part C, Acadf'.mic Press, Orlando, FL, 1984, p. 237-255.
Davis, BJ., "Disc electrophoresis II. Method and application to human serum
proteins," Annals N.Y. Acad. Sci., 12] ,404 (1964).
Dean, J.A. (Ed.), Lange's Handbook of Chemistry, 13th ed., Marcel Dekker,
New York, 1986.
Dobry, R. and R.K. Finn, "Engineering problems in large-scale

electrophoresis", Chern. Engr. Prog., 54,(4) 59 (1958).
Dunn, M.J., "Electrophoresis in polyacrylamide gels", in J.W. Jorgenson and

M. Phillips (Eds.), New Directions in Electrophoretic Methods, ACS Symp.
Series No. 335, ACS, Washington, DC, 1987, Chapt. 2.
Dunn, M.J. and A.H. Burghes, Electrophoresis, 4, 97-116 (1983).
Durrum, E.L., "Continuous electrophoresis and ionophoresis on filter paper"

Journal ACS, 73,4875 (1951).
Epstein, 1., "An apparatus for continuous separation of macromolecules by

combined electrophoresis and gel filtration", Chromatographia, 10 (2), 65
(1977).
Everaerts, P.M., J.L. Beckers, and Th.P.E.M. Verheggen, Isotachophoresis.

Theory, Instrumentation and Applications, Journal of Chromatography Library,
Vol. 6, Elsevier, Amsterdam, 1976.
Everaerts, P.M. and Th.P.E.M. Verheggen, "Capillary isotachophoresis", in

J.W. Jorgenson and M. Phillips (Eds.), New Directions in Electrophoretic
Methods, ACS Symp. Ser. No. 335, ACS, Wa3hington, DC, 1987, Chapt 14.
610
Fawcett, J.S., "Isoelectric focusing", in E. Reid (Ed.), Methodological Develop-

ments in Biochemistry, 2. Preparative Techniques, Longmans, London, 1973,
Chapter 7.
Gianazza, E. and P.G. Righetti, "Electrophoresis in gels of graded porosity and

two-dimensional techniques", in P.G. Righetti, CJ. van Oss, and J.W. Vanderh-
off (Eds.), Electrokinetic Separation Methods, Elsevier/North Holland Biomed-
ical Press, Amsterdam 1979, p. 293-311.
Giddings, J.C., "Generation of variance, theoretical plates, resolution, and peak

capacity in electrophoresis and sedimentation", Separ. Sci., 4, 181 (1969).
Giddings, J.C., "Two-dimensional separations: concept and promise", Anal.

Chem .• 56, 1258A (1984).
Gobie, W.A., J.B. Beckwith and CF. Ivory, "High resolution continuous flow
electrophoresis", Biotechnol. Prog .• l, (1), 60 (1985).
Gordon, MJ., X. Huang, S.L. Pentoney, Jr., and R.N. Zare, "Capillary electro-
phoresis," Science. 242,224 (1988).
Grassman, W. and K. Hannig, "Ein Enfaches Verfahren zur Kontinuirlichen

Trennung von Stoffgemischen Auf Filterpapier Durch Electrophorese"
Naturwissenschaften. 37, 397 (1950).
Hannig, K. "Continuous free-flow electrophoresis as an analytical and prepara-

tive method in biology",J. Chromatogr .. 159, 183 (1978).
Haugaard, G. and T.D. Kroner, "Partition chromatography of amino acids with

applied Voltage", Journal ACS. 70,2135 (1948).
Hiemenz, P.C., Principles of Colloid and Surface Chemistry, 2nd ed., Marcel
611
Holloway, C.J. and R.V. Battersby, "Preparative isotachophoresis" , in W.B.

Jakoby (Ed.), Methods in Enzymology, Vol. 104, Purification and Related
Techniques, Part C, Academic Press, Orlando, 1984, Chapt 15.
Horejsi, V., "Affinity electrophoresis", in W.B. Jakoby (Ed.), Methods in Enzy-

mology, Vol. 104, Enzyme Purfication and Related Techniques, Part C,
Academic Press, Prlando, FL, 1984, p. 275-281.
Ingram, V.M., "A specific chemical difference between the globins of normal
human and sickle-cell anemia haemoglobin", Nature, 178, 792 (1956).
Jorgenson, J.W., "Capillary zone electrophoresis", in J.W. Jorgenson and M.

Phillips (Eds.), New Directions in Electrophoretic Methods, ACS Symp. Ser.
No. 335, ACS, Washington, DC, 1987, Chapt. 13.
Jorgenson, J.W. and K.D. Lukacs, "Zone electrophoresis in open-tubular glass

capillaries", Anal. Chem.53, 1298 (1981).
Just. W.W. and G. Werner, "Cell separation by continuous flow electrophoresis

and isolectric focusing", in P.G. Righetti, CJ. van Oss, and J.W. Vanderhoff
(Eds.), Electrokinetic Separation Methods, Elsevier/North Holland Biomedical
Press, Amsterdam (1979), 143-167.
Karol, P.J. and M.H. Karol, "Isotachophoresis", J. Chem. Educ .• 55, 626
(1979).
Kolin, A., "Endless belt continuous flow deviation electrophoresis", in P.G.

Righetti, CJ. van Oss, and J.W. Vanderhoff (Eds.), Electrokinetic Separation
Methods, Elsevier/North Holland Biomedical Press, Amsterdam (1979), 169-
220.
Laurell, C.-B., "Quantitative estimation of proteins by electrophoresis in

agarose gels continuing antibodies", Anal. Biochem.,15, 45 (1966).
Lehninger, A.L., Biochemistry, Worth, New York, 1970.

612
Lester, E.P., P.F. Lemkin, and L.E. Lipkin, "New dimensions in proteins
analysis", Anal. Chern .. 53, 390A (1981).
Locke, B.R and RG. Carbonell, "A theoretical and experimental study of
counteracting chromatographic electrophoresis," Separ. Purific. Methods. 18, 1
(1989).
McCann, G.D., J.W. Vanderhoff, A. Strickler and T.I. Sacks, "Separation of

latex particles according to size by continuous flow electrophoresis", Separa-
tion Purific. Methods. 2, 153 (1973).
Maizel, J.V., Jr., "Acrylamide-gel electrophorograms by mechanical fractiona-

tion: radioactive adenovirus protein", Science 151, 989 (1966).
Mel, H.C., "Stable-flow boundary migration and fractionation of cell mixtures.

I. Apparatus and hydrodynamic feedback principles", J. Theoret. Bioi .. 6, 307
(1964).
Melvin, M., Electrophoresis, in Analytical Chemistry by Open Learning Series,

Wiley, Chichester, England, 1987.
Mosher, RA., W. Thormann, N.B. Egen, P. Couasnon, and D.W. Sammons,

"Recent advances in preparative electrophoresis", in J.W. Jorgenson and M.
Phillips (Eds.), New Directions in Electrophoretic Methods, ACS Symp. Ser.
No. 335, ACS, Washington, DC, 1987, Chapt 16.
Morrison, D.R., G.H. Barlow, and C. Cleveland et al., Adv. Space Res.. 4, 67
(1984).
Nerenberg, S.T. and G. Pogojeff, "Preparative electrochromatography and elec-

trophoresis using adopted Sephadex and other columns," Am. J. CUn. Path ..
51, 728 (1969).
Newman, J.T., Electrochemical Systems, Prentice-Hall, Englewood Cliffs, N.J.,

1973, Chapt 11.
613
O'Brien, R.W. and L.R. White, "Electrophoretic mobility of a spherical col-

loidal particle," J. Chem. Soc. Faraday Transactions, II, 74, 1607 (1978).
O'Farrell, P.H. "High resolution two-dimensional electrophoresis of proteins",

J. Bioi. Chem., 250, 4007 (1975).
O'Farrell, P.H. "Separation techniques based on the opposition of two

counteracting forces to produce a dynamic equilibrium," Science, 227, 1586
(March 29, 1985).
Ornstein, L., "Disc electrophoresis I. Background and theory," Annals N.Y.

Acad. Sci., 121,321 (1964).
Ostrach, S., "Convection in continuous-flow electrophoresis", J. Chromatogr.,

140, 187 (1977).
Overbeek, J.Th.G. and B.H. Bijsterbosch, "The electrical double layer and the
theory of electrophoresis", in P.O. Righetti, C.J. van Oss and J.W. Vanderhoff
(Eds.), Electrokinetic Separation Methods, Elsevier/North-Holland Biomedical
Press, Amsterdam, 1979, p. 1-32.
Philpot, J.StL., "Apparatus for continuous-flow preparative electrophoresis", in

E. Reid (Ed.), Methodological Developments in Biochemistry, 2. Preparative
Techniques, Longmans, London, 1973, Chapter 8.
Philpot, J .StL., "The use of thin layers in electrophoretic separations", Trans.

Faraday Soc., 39, 38 (1940).
Radola, BJ., "High-resolution preparative isolectric focusing", in W.B. Jakoby

(Ed.), Methods in Enzymology, Vol. 104, Enzyme Purification and Related
Techniques, Part C, Academic Press, 1984, Chapter 13.
Reeder, DJ., "Development of electrophoresis and electrofocusing standards",

in J.W Jorgenson and M. Phillips (Eds.), New Directions in Electrophoretic
Methods, ACS Symp. Ser. No. 335, ACS, Washington, DC, 1987, Chapt 7.
614
Righetti, P.G., Isolectric Focusing: Theory, Methodology and Applications.

Elsevier Biomedical, Amsterdam, 1983.
Righetti, P.G., E. Gianazza, and K. Ek, "New developments in isoelectric

focusing", J. Chromatogr .• 184, 415, (1980).
Righetti, P.G., C. Gelfi, and E. Gianazza, "Immobilized pH gradients: recent

developments", in J.W. Jorgenson and M. Phillips (Eds.), New Directions in
Electrophoretic Methods, ACS Symp. Ser. No. 335, ACS, Washington, DC,
1987, Chapl 3.
Rilbe, H., "Historical and theoretical aspects of isoelectric focusing," Ann. N.Y.
Acad. Sci. 209, 11, (1973).
Schwartz, D.C., "Pulsed electrophoresis", in J.W. Jorgenson and M. Phillips

(Eds.), New Directions in Electrophoretic Methods, ACS Symp. Ser. No. 335,
ACS, Washington, DC, 1987, Chapt. 12.
Smith,!. (Ed.), Chromatographic and Electrophoretic Techniques, Vol. II.

Zone Electrophoresis, 2nd Ed., William Heineman Ltd., 1968.
Smith, 1., "Zone electrophoresis on paper, thin layers and pevikon block", in
P.G. Righetti, CJ. van Oss, and J.W. Vanderhoff (Eds.), Electrokinetic Separa-
tion Methods, Elsevier/North Holland Biomedical Press, Amsterdam (1979), p.
33-53.
Smithies, 0., "Zone electrophoresis in starch gels: group variations in the

serum proteins of normal human adults", Biochem. J. 61,629 (1955).
Strain, H.H. and Sullivan, J.C., "Analysis by electromigration plus chromatog-

raphy", Anal. Chern. 23, 816 (1951).
Svensson, H., "Isoelectric fractionation, analysis, and characterization of

ampholytes in natural pH gradients. 1. The differential equation of solute con-
centrations at a steady state and its solution in simple cases", Acta Chem.
Scand. 15,325 (1961).
615

ampholytes in natural pH gradients. IT. Buffering capacity and conductance of
isoionic ampholytes", Acta Chem. Scand.16, 456 (1962a).

ampholytes in natural pH gradients. ITL Description of apparatus for electro-
lysis in columns stabilized by density gradients and direct determination of
isoelectric points",Arch. Biochem. Biophys. Suppl. 1, 132 (1962b).
Svensson, H. and 1. Brattsten, "An apparatus for continous flow electro-

phoresis", Arkiv. Kemi.1, (47),401 (1949).
Tiselius, A., "A new apparatus for electrophoretic analysis of colloidal mix-
tures", Trans. Faraday Soc., 33,524 (1937).
Vanderhoff, J.W. and F.J. Micale, "Influence of electroosmosis", in P.G.

Righetti, CJ. van Oss, and J.W. Vanderhoff, (Eds.), Electrokinetic Separation
Methods, Elsevier/North-Holland Biomedical Press, Amsterdam, 1979, p. 81-
93.
Vesterberg, 0., "Synthesis and isoelectric fractionation of carrier ampholytes",

Acta Chem. Scand., 23, 2653 (1969).
Wankat, P.C., "Two-dimensional separation processes", Separ. Sci. Technol.,

19,801 (1984-85).
HOMEWORK
AI. Define the following terms.
a. Moving boundary electrophoresis
b. Zone electrophoresis
616
c. Isoelectric focusing
d. Isotachophoresis
e. Paper electrophoresis
f. SDS-PAGE electrophoresis
g. Immunoelectrophoresis
h. Rocket electrophoresis
i. Fingerprinting
j. Capillary electrophoresis
k. Free flow electrophoresis
1. Recycle electrophoresis
m. pI
n. Electrical double layer
o. Zeta potential
p. Electroosmosis
q. Carrier ampholyte
A2. Explain why the "rockets" shown in Figure ll-Sc form.
A3. Why is isolectric focusing a more robust separation method than electro-
phoresis?
A4. Explain why the two developments should be independent in sequential

two-dimensional systems?
AS. Why is two-dimensional electrophoresis currently the system of choice for

scale-up?
A6. Continuous operation of counteracting chromatographic electrophoresis

would not completely purify the desired proteins. Explain why there
would be more impurity than in batch operation.
617
A 7. Develop your key relations page for this chapter.
B 1. Brainstorm at least 20 different sequential two-dimensional development
methods. A matrix may be useful for doing this.
C. Derivations
Cl. Transform the one-dimensional zone spreading Eqs. (11-15) to (11-24) to

the continuous two-dimensional system shown in Figure 11-7a
C2. Equation (11-15) can be derived by solving the appropriate partial dif-
ferential equations. The Lapidus and Amundson dispersion model (see
Section 7.4) can be used to develop the solution. Do this derivation.
NOTES: l. Use Eq. (7-10),
2. A(T) = 0 and v = \l;.,net (Eq. (11-16b)).
3. Use superposition Eq. (7-7) and the definition of erf in Eq.

(7-11a).
C3. Units are a problem and need to be practiced.
a. Show that '1'0 in Eq. (11-5b) is in volts.
b. Show that the units in Eq. (l1-4b) are correct
c. Show that the units in Eq. (11-9b) are correct.
Dl. A 1 x lO-9 m diameter particle has a measured mobility of 1 x 10-8 mZ/sV

in a 0.0001 M aqueous solution of Naz S04 at 25°C. What is the value of
~?
D2. A protein is run in a PAGE electrophoresis apparatus for 3 hours at 2000

volts. The plate is 20 cm long. The center of the protein spot is measured
at 8.54 cm from the feed well and the spot width is 0.103 cm. Estimate J.l
and D for the protein at the conditions of this experiment.
618
03. Tiselius (1937) reported the following data on two proteins: A = R. Phy-
cocyan and B = R. Phycoerythrin.
L;-
pIA =4.85 pIB =4.25
[ d~ 10.2. Hr' (a, pI) [d~]. =-14.2. Hr' (a'pl)

u is in cm2/s volt.
The proteins are to be separated in a continuous flow, two-dimensional

free solution apparatus (see Fig. 11-7). The apparatus dimensions are L =
10 cm, W =5 cm and h =0.3 cm. The feed entry is 0.1 cm wide. Voltage
is 900 volts. Average fluid velocity Vbuffer = 0.2 cm/s. The apparatus is
horizontal and has efficient cooling from the bottom. Assume there is no
thermal convection and no electroosmosis.
a) At pH 4.25 estimate ~A and ~B' Determine the angles aA and aB'

Estimate the width of the zones at the exit and the locations of the
peak maximums if D = 4 x 10-7 cm2/s for both proteins. If the
observed width is 0.9 em, calculate D eff.
b.) Repeat part A at pH 4.55. Assume the values of d w'd(pH) are

constant.
04. A sequential 20 experiment is done for a protein which has D y,l - 10-8
and D y.2 - 2 x 10-7 cm2 Is. Assume D y,1 = D z,l and D z.2 = 5 x 10-6. The
first development is for 4 hours and the second is for 10 hours.
a. What are the widths wyand W z of the spot?
b. If the spot center is at Y = 5.0 cm, Z = 10.6 cm, plot the constant con-
centration line where c = 0.5 Cmax on a y,z plane.
05. We wish to separate various ammonium ions by isotachophoresis. NH! is

used as the leading ion and (C4~)4 N+ as the trailing ion. Initial concen-
tration of NH! and (C4~)4 W are both 0.03 N. The feed contains O.OIN
CH3NHt, 0.0025N (CH3 )3 NW and O.OO6N (C2Hs )4 W. The feed band
is 2 cm long. The experiment is operated as a batch experiment in free
619
solution cr is the anion. Once the bands have been fonned. detennine
the concentrations of CH3NHt. (CH3h NW. (C 2 H s )4 W and (C4H 9 )4
N+. Also. find the widths of the CH3 NHt. (CH3h NH+ and (C 2 Hs )4 N+
bands. Data is in Table 11-3.
D6. An isotachophoresis experiment is done to detennine ~N.' The leading

electrolyte is NHt of concentration 0.01 N. Trailing electrolyte is
(C2 HS)4 W initially at 0.005 N. The co-ion is cr. The feed band is 3
cm wide and contains 0.005 N NaCl. After the experiment is at steady
state. the Na+ band is 1.8703 cm wide. What is ~N" (in cm2 /Vs)? Data is
in Table 11-3.
07. The two proteins in problem 11-03 are to be separated in a batch IEF sys-
tem. The system length is 10 cm. pH varies linearly from pH 3.0 to pH
5.0 over this length. Note that this is a very easy IEF separation. Assume
Deff = 0.0004 cm 2/s which is high for an IEF system.
a) Detennine the field strength E required to achieve an acceptable

separation (Eqs. 11-49 and 11-50).
b.) At the field strength calculated in part A detennine crA. crB and the
widths of the two zones. (width = 4cr).
c.) IfE = 80 volts/cm, detennine crA and crB and the widths.
08. An isoelectric focusing experiment was done on a cooled plate. The fol-
lowing data were obtained at steady state at E = 100 V/cm:
Observed Compound Distance from anode, em
Ovalbumin 6.2
~-aspartyl-histidine 7.0
Evans blue (PI = 5.35) 8.5
Congo red (pI = 5.80) 10.1
See Table 11-2 for additional data on these compounds.

620
a We wish to separate two similar proteins whose pI is about 5.2,

D = 2 x 10-7 cm 2/s,
dJl -5 cm 2
d(pH) =- 1.5 x 10 s V (PH unit)
What difference in pI values is required to obtain two resolved

peaks?
b. If another protein is observed at 9.1 cm, what is its pI?
D9. A continuous flow, two-dimensional, free solution, IEF system is used for
separating proteins. The carrier ampholyte is prefocused. Assume that
steady state is obtained. Dimensions: L = 20 cm, W = 10 cm, 2h = 0.5
cm. There are 50 outlets and each covers 0.2 cm of the width. From a
given outlet the product is well mixed. Operation is at 25°C. The pH gra-
dient varies from 3 at the anode to 10 at the cathode. Assume the gradient
is approximately linear. Assume cooling is adequate and there is no con-
vection. Ports are numbered starting with 1 at the anode.
a. Which port(s) does protein A of Problem ll-D3 exit from if V = 375

volts?
b. Which port(s) does protein B of Problem 11-D3 exit from if V = 375

volts?
Part III
MEMBRANES
NOMENCLATURE CHAPTERS 12 AND 13
a,a' tenns in quadratic equation
a constant for diffusivity, Eq. (12-75)
a constant for osmotic pressure, Eqs. (12-18b,c)
A area of membrane, m2 or cm2
Acs cross-sectional area for flow, m2 or cm 2
Am area of a membrane in ED stack, m2
b constant, Eq. (12-64)
b, b' tenns in quadratic equation
c, c' tenns in quadratic equation
c concentration. Either weight or molar or equivalents (ED) gIL,

glcc, gmoles/L, g equivalents/L
Cb bulk concentration
cg gel concentration
cin inlet concentration
em . concentration in membrane
COUI outlet concentration
Cp penneate concentration
Czrn solute concentration in membrane

,
C2rn solute concentration in membranes on high pressure side
"
c2rn solute concentration in membrane on low pressure side
CZp solute concentration in penneate
Cw wall concentration
623
624
CZ w solute concentration at wall
heat capacity liquid, caVgOC
C;,V heat capacity vapor, cal/gOC
d,D diameter, cm or m
dcq Equivalent diameter, Eq. (13-12).
~ particle diameter in gel layer, m
D Diffusivity, cm zIs
solvent diffusivity in membrane, cmzls
solute diffusivity in membrane, cm zIs
parameter for diffusivity, Eq. (12-75)
energy, joules
Fanning friction factor, Eq. (13-lOb,c)
friction factors, Eqs. (13-86) and (13-87)
flow rate, gmoles/time or g/time
inlet flow rate, kg molesls or m3 Is
outlet flow rate, kg molesls or m 3 Is
permeate flow rate, cm 3 Is, m 3 Is or kg molesls
FRecycle recycle flow rate
F Faraday. 96,500 amp'sec/equiv
h half height between parallel plates, em
H hydrodynamic permeability, Eq. (13-96)
~Hf latent heat of freezing, Eq. (12-19)
i current density, amp/m z

625
I current, amps
jn Colburnj-factor, Eq. (13-10a,b)
J asy asymptotic volumetric flux, Eq. (13-19)
J icit initial flux
J,olv, J 1 volumetric solvent flux, cm3 /cm 2s or L/m2 day
J:01v , J~ mass solvent flux, g/cm2s
J so1ute , J 2 solute flux. g/cm 2s
hdif solute flux due to diffusion
k mass transfer coefficient, cm/s
k!mute feed mass transfer coefficient
k~ute dialysate mass transfer coefficients
kT total mass transfer coefficient in dialysis
KA solubility parameter, Eq. (12-15)
Kg gel permeability. Eq. (12-38)
~lute solute permeability
Ksolv solvent permeability
K} =Cl m/c1p' solvent distribution coefficient

K2 solute distribution coefficient, Eq. (13-78)
L length, cm or m
Lu, L;j phenomenological coefficients, Section 13.6
m exponent for flux decay, Eq. (13-19)
m mobility, Section 13.6
M =Cw/Cb' concentration polarization modulus

626
MW molecular weight
n total moles solute, Section 12.5.
n number of tubes or cells
n constant in Eq. (12-18c)
N number of molecules or particles
Rate of transfer, cm3 /s
pressure, atm, bar, psi, mmHg, kPa
high pressure
low pressure
permeate pressure
effective pressure driving force across membrane, Eq. (13-94)
partial pressure
partial pressure on liquid side for pervaporation
P2 partial pressure on vapor side for pervaporation

p permeability of gas, cm3 (STP) cm/cm 2 s (cm Hg)
molar flow rates on high pressure and permeate sides
permeate molar flow rate
charge,coloumbs
volumetric flow rate, cm3/s or L/hr

dialysate volumetric flow rate
feed volumetric flow rate
r radial direction
R radius, cm
R resistance per cell, ohms

627
R retention
R gas constant
RO inherent retention = 1 - c;,/cw
Rapp apparent retention = 1 - Cp/Cb
Renergy gas constant in energy units, (Example 13-1)
~ pore radius, cm
~ress gas constant in pressure units, (Example 13-1)
Re Reynolds number =d Ub p/~
S* Henry's law constant, Eq. (12-76)
Sc Schmidt number =~ p/D
t time, s or days
fg gel layer thickness, cm
tm membrane thickness, cm
T absolute temperature, K
T f , T; freezing temperature of solution and pure solvent, K
Tin inlet temperature
Tout outlet temperature
Tref reference temperature
u velocity in x direction, cm/s

u center of mass velocity of pore fluid
llb bulk velocity, cm/s
llm inlet average velocity, em/s
v velocity in y or r direction, cm/s
v·1 partial molar volume, Eq. (13-63)

628
velocity through wall, cm/s
partial molar volume of solvent, gmoles/cm 3
V.olvcnt partial molar volume of solvent, gmoles/cm 3
v voltage, volts
v volume, cm3 or L
VP vapor pressure
w width, cm
x axial distance, cm
x distance into membrane, cm
x liquid mole fraction
X-l force on component i, Section 13.6
Y direction perpendicular to parallel plates
Y gas or vapor mole fraction
Y' y/h
gas mole fraction on high pressure side
Yin inlet mole fraction
Yout outlet mole fraction
permeate mole fraction
mole fraction on high pressure side
Y2 mole fraction on permeate side
Y2.0 eqs. (13-44a,c)
external force on component i, Section 13-6
z feed mole fraction

629
Greek Letters
n selectivity, KsolvlK.olute in RO
nAB selectivity, P A/PB, in gas permeation

,
nAB selectivity, Eq. (12-81), in pervaporation
'Yw fluid shear rate, Eqs. (13-15) and (13-17)
a boundary layer thickness, em
£ porosity
11m manifold efficiency, ED
115 semipermeability membrane efficiency, ED
11w water transfer efficiency, ED
a cut = Fp/Fin
6rot total cut in recycle system, Fp/Fnew
A latent heat of vaporization, cal/g or cal/gmole

A tortuosity
J..l viscosity, poise or pascal-sec.

J..li chemical potential of species i
v kinematic viscosity = Wp
~ efficiency of current utilization, Eq. (12-67)
~ v! h x/3Um D 2, Section 13.5

7t osmotic pressure, atm
7t Pi,3.141592654 ....
P density, g/cm 3
A
P molar density, gmoles/cm 3 or kgmoles/m 3

ro angular velocity, radians/s
Chapter 12
INTRODUCTION TO MEMBRANE SEPARATIONS
A variety of membrane separation processes such as reverse osmosis,

ultrafiltration, gas permeation, and electrodialysis are becoming increasingly
popular in@industry. This popularity is due to the simplicity of the processes,
the gentle nature of the separation (high temperatures and phase changes are
not required), the usual low energy requirements, and the often low capital and
operating costs. In this chapter all of the membrane separation processes will
be introduced. Simple equations for understanding and designing the processes
will be presented. The emphasis will be on understanding the physical nature
of the processes. Chapter 13 builds on the basics presented in this chapter and
presents more detailed analyses of the processes occurring in membrane
separators.
12.1. BASIC CONCEPTS
A membrane is a physical barrier between two fluids. In over 99% of the cases
of current industrial interest the membrane is made from a polymer. This poly-
mer is cast or spun or extruded to form a continuous film without holes in the
desired geometry. The simplest geometry used for membrane separators is the
stirred cell shown in Figure 12-1. The fluid enters into the upper chamber. Part
of the feed stream is transferred through the membrane to the lower chamber.
With a perfect membrane only species A will transfer through the membrane
while species B will remain in the upper chamber. Either species may be the
desired product. The stirrer is used to improve mass transfer rates and prevent
the buildup of species B at the membrane surface.
630
631
~
~
,- I
~
-
Membran e
t +
Cp
III
Figure 12-1. Stirred cell membrane system
The flux through the membrane can be written as
Flux = Transfer rate (12-1)

Transfer area
= Permeability
Thickness
(Dri'
vmg 'Force)
Flux is defined either as J = (Volume)/(area)(time) or as J' =

(mass)/(area)(time). Obviously, these two fluxes are related by J' = Jpp where
Pp is the permeate density. The flux depends linearly on both the permeability
(a constant which indicates how "tight" the membrane is) and the driving
force. The driving force depends upon the type of separation. For instance, in
gas permeation the driving force is the pressure difference across the mem-
brane. The membrane can separate two gases because the permeabilities are
very different Note that, in general, both gases do transfer through the mem-
brane, but one of them transfers at a much higher rate. Thus, most membrane
separators are not equilibrium processes but are rate processes. The flux also
depends inversely upon the thickness of the membrane. The thinner the mem-
brane the higher the flux.
High fluxes are usually very desirable since the transfer area required
will be lessened. Thus we want a high driving force. In the cases of gas per-
meation, reverse osmosis (used for purifying water) and ultrafiltration (used for
concentrating large molecules) this means a high pressure drop across the
membrane. The membrane must be mechanically strong to withstand these
high pressures. Mechanical strength requires a thick membrane. However,
632
a Feed Phase
Thin
Skin
Porous
Support
Layer
Permeate
Figure 12-2. Asymmetric membrane. a. Schematic. b. Scanning electron

microscope photograph. (Warashina et ai, 1985). Part b
reprinted with permission from Chern. Tech, 558 (1985).
Copyright 1985, American Chemical Society.
633
Eq.(12-1) shows that thick membranes will have low fluxes and hence large
are$ will be required to achieve the desired transfer rate. A fundamental prob-
lem which stopped the commercial use of membranes for many years was how
to make a membrane which was thin enough to have high fluxes while at the
same time it was mechanically strong.
The solution to this problem was found in the late 1950's and early
1960's by Sidney Loeb and S. Sourirajan (Loeb and Sourirajan, 1960, 1963).
Their solution was to make a membrane which was asymmetric or anisotropic.
This type of membrane is illustrated in Figure 12-2. The membrane was cast to
have a very dense thin layer or skin on one side. This layer was typically 0.1 to
1.0 microns thick. The thin skin does the actual separation of the species.
Thus in Eq.(12-1) the thickness term is very small. The thin skin is supported
by a much thicker porous layer. This layer provides the required mechanical
strength, but it is so porous that no separation occurs in this layer. These mem-
branes can be operated at pressure drops in excess of 1000 psig. The Loeb-
Sourirajan membrane was made from cellulose acetate for desalination of
water by reverse osmosis. Methods for casting these membranes are reviewed
by Kesting (1985) and Soltanieh and Gill (1981). The use of asymmetric mem-
branes has been extended to a variety of other polymers and to other membrane
separation methods. The history of membrane separations is discussed by
Lonsdale (1982). Not all membrane separators use asymmetric membranes.
The exceptions will be discussed in the individual sections.
A variety of different polymers have been used commercially for mem-

brane systems. The polymers used can be classified into three categories: rub-
bery polymers, glassy polymers, and ion exchange membranes. Rubbery poly-
mers such as silicone rubber operate above the glass transition temperature, Tg ,
of the polymer. As a generalization, rubbery polymers have high fluxes of
organics and low flux of water. Glassy polymers such as polycarbonate operate
below Tg • These polymers can be either amorphous or partially crystalline.
Glassy polymers are selective for water. Ion exchange membranes are essen-
tially ion exchange resins (see Chapter 9) made in membrane form. Ion
exchange membranes are very selective for water.
The most common structures are shown in Figure 12-3 (Kesting, 1985;
634
b c
(-C- y- )
\ C=N n
(-S02~r@-o~~?)
H ~ cf H n
f
) -CF2,C-) n
(-(CF2 CF2,
O-(C~ yF )y-OC~C~S02F
to-poj
g
CF3
CH 3 n
Figure 12-3. Polymers used for membrane separators. a. Cellulose (D,UF).

b. Polyacrylonitrile (D,UF). c. Poly sulfone (GP,RO,UP,ED). d.
Polybenzimidazolone (RO). e. Polyamide (from lcrephlhalic
acid m-arnino benzarnide and m-phenylene diamine) (RO,UF).
f. Nation (Donnan dialysis). g. Silicone (GP). Key: GP = Gas
Permeation, D = Dialysis, RO = Reverse Osmosis, UP =
Ultrafiltration, ED = Electrodialysis
635
Pusch and Walch. 1982; Soltanieh and Gill. 1981; Ward et al., 1985). Addi-
tional examples of polymers used for membranes are discussed by Kesting
(1985). Lloyd and Meluch (1985). Finken (1985). Pusch and Walch (1982), and
Ward et al., (1985). For example. Figure 12-3 shows only one of the possible
polyamides (Kesting, 1985; Pusch and Walch, 1982). The cellulose (Figure
12-3a) and cellulose acetate membranes were the firstcommercialIy successful
membranes. Although still commonly used, considerable research has gone
into developing membranes with better chemical and thermal resistances.
Polysulfone (Figure 12-3c) and Nafion (Figure 12-3f) have outstanding resis-
tances to chemicals. Polystyrene cross-linked with divinylbenzene (Figures 9-1
a and b) also has excellent chemical resistance and is used as a backing for
composite membranes and as the basis for electrodialysis membranes.
With a high flux rate species B which does not transfer through the mem-
brane may build up to a high concentration at the surface of thel membrane.
This is called concentration polarization and is illustrated in Figure 12-4.
Since the transfer rate of species B through the membrane is low, the permeate
concentration is low, cp «Cb' Species A (which might be the solvent) has a
high flux through the membrane. Thus. there must be a convective flow
towards the membrane. This convective flow carries B to the membrane sur-
face. Since the flux of B is low. the concentration of B will build up at the
membrane wall. Thus, a concentration gradient is produced. This buildup of B
I
+ Back Diffusion
Convective Flow
I
---:--- cb
-~·I
y=8
Membrane
Figure 12-4. Build-up of retained species at membrane wall (concentration
polarization).
636
is counteracted by diffusion into the bulk fluid. If diffusion rates are high as is
true in most gas systems, then the concentration buildup will be modest and
concentration polarization will not be a problem.
In liquid systems diffusivities are small and Cw > Cb. Concentration

polarization is detrimental for several reasons. The high Cw value means that
any liquid that leaks through the membrane will be at a high concentration of B
and thus will increase Cpo The high wall concentration also increases the
osmotic pressure (see Section 12.4.) which reduces the driving force for
transfer of solvent High concentrations of solute can also cause gelling or
fouling at the membrane surface which will drastically decrease flux (see Sec-
tion 12.5.). To avoid concentration polarization we can use stirring, turbulence,
high shear rates, or very thin channels.
U.2. MEMBRANE MODULES
The general design requirements for any membrane system can now be listed.
1. Thin active layer of membrane.
2. High permeability for species A and low permeability for
species B.
3. Stable membrane with long service life.
4. Mechanically strong.
5. Large surface area of membrane in small volume.
6. Concentration polarization eliminated or at least controlled.
7. Easy to clean if necessary.
8. Inexpensive to build.
9. Low operating cost
This list of desirable design conditions is general. Items specific to each of the
membrane devices will be discussed later. A historical perspective on mem-
brane modules with many early advertisements is given by Kremen (1986).
Currently, spiral wound and hollow fiber systems have most of the market.
The simplest geometry for a membrane separator is the stirred cell shown
in Figure 12-1. Stirred cells are commonly used in laboratories, but the mem-
brane surface area is small. Thus for large scale commercial applications other
637
a b
Porous Tube
O
------------r-.
I \
( \
I I
I I
\ I
\ I
- - - - - - - - - ~ ...'
Membrane Coating
(Inside or Outside)
----..:: Spacers
Reten~
c
... :~
Permeate
Permeate channel
r Potting IIIt Permeate

d
Feed--(]lilililililililllll: ·
L Hollow fibers
Distribution
Figure 12-5. Geometries of commercial membrane systems. a. Plale-and-
frame. b. Tubular. c. Spiral wound. d. Hollow fiber.
geometries are required. The most commonly used commercial systems are
shown in Figure 12-5. The plate-and-frame system is similar to a plate-and-
frame filter press with a series of flat membrane sheets. The area per volume
ratio is significantly higher than stirred cells but lower than spiral wound or
hollow fiber systems. The sheets can be quite close together to reduce concen-
tration polarization. The major advantage of plate-and-frame systems is they
638
can be taken apart for cleaning if necessary. Their major application has been
in the food industry where fouling is a major problem. The plate-and-frame
geometry is also used for electrodialysis.
Cylindrical geometries are also used. In the tubular system shown in

Figure 12-5b the membrane is coated on the inside or outside of the porous
support tube. Typical inside diameters are in the range from 1 cm to 1 inch. A
bundle of tubes can be packaged together as in a shell-and-tube heat exchanger,
but the area/volume ratio is low. The main advantage of tubular systems is
they can be mechanically cleaned by forcing a sponge ball through the tubes.
Thus tubular systems compete with plate-and-frame systems for dirty or foul-
ing applications. Design details of these systems are given by Stana (1977).
The spiral wound system shown in Figure 12-5c is a way of increasing

the area/volume ratio of flat plat devices. Two feed channels, two membranes,
and a permeate channel are layered and wrapped around a central porous tube
several times. Spiral wound systems made from cellulose acetate are com-
monly used for reverse osmosis. This design is a good method for achieving
large area/volume (about 300 f~ per ft3 ) with membrane materials which can-
not easily be extruded into hollow fibers. Concentration polarization is con-
trolled by using thin channels and wrapping the membrane around plastic net-
ting which induces turbulence. Unfortunately, these systems cannot be used for
treating very dirty or fouling systems since they are not easy to clean. Design
details are given by Kremen (1977). Spiral wound is one of the two most
popular configurations.
The hollow fiber system shown in Figures 12-5d and 12-6 is also one of
the two most popular designs. The fibers are extruded from polymers such as
Nylon or polysulfone. A typical reverse osmosis hollow fiber would have a
inner diameter of 42 microns, an outer diameter of 85 microns, and a 0.1 to 1.0
micron skin on the outer surface (Applegate, 1984). For gas permeation the
fibers are made somewhat larger to reduce the pressure drop inside the tubes.
For ultrafiltration the fibers are much larger (500 to 1100 micron i.d.) and the
skin is on the inside of the fiber. These numbers indicate that a considerable
amount of engineering design can be done with the fibers. The fibers can have
fairly thick walls and with flow radially inward large pressure gradients can be
639
- 1126mm------·---1
-+ Solution outlet
tS9-m;-
diameter
Case
Hollow-liber membrane
Tube.eat
Figure 12-6. Details of hollow-fiber module. (Warashina et ai, 1985).

Reprinted with permission from Chern. Tech., 558 (1985).
handled. The major advantage of the hollow fiber systems is that a huge
number of fibers (4.5x106 ) can be packed inside a 25.4 cm (10 inch) diameter
cartridge. Thus the area/volume ratio is very high (about 5000 ft2 per ft3) and
costs are kept down. A major disadvantage of hollow fibers is particulate
solids can permanently clog the tiny fibers. Thus dirty fluids must be
thoroughly cleaned before being sent to the hollow fiber separator. Details of
hollow fiber systems are given by Orofino (1977) and Caracciolo et al., (1977).
All of the basic systems are usually built in a few basic modular sizes. A
"module" is a complete unit such as a hollow fiber separator capable of pro-
cessing a given amount of fluid with a given separation. The details of a hol-
low fiber module used for ultrafiltration are shown in Figure 12-6 (Warashina
et al., 1985). A single module is often not sufficient for the required flow rate
and/or separation. Thus modules are cascaded in a variety of ways to produce
the desired separation. One advantage of modular construction is the modules
640
- -~
a b
$ OOOOOOOIOUI.
,OOOOOOtOOlOL
FIN
~:~: 1 001 OOOOfOOOOOO
YIN
C d
~'I~iQt~ ~ ~ -- i-- ~
! ~
-- ---
e f
Figure 12-7. Idealized flow patterns for flow inside module. a. Completely
mixed. b. Mixed on feed (high pressure side). c. Mixed on per-
meate (low pressure) side. d. Cross-flow. e. Co-current. f.
countercurrent.
can be constructed in a plant instead of being field constructed. This is usually

significantly cheaper. One disadvantage of modular construction is that there is
probably very little economy of scale.
The flow patterns inside the module have a large effect on the outlet con-
centrations and flow rates of the high and low pressure products. Several ideal-
ized flow patterns are shown in Figure 12-7. The completely mixed case, Fig-
ure 12-7a, is the easiest to analyze but the least advantageous for separation.
The system which is mixed on the feed side, Figure 12-7b, is close to the flow
pattern which occurs for most hollow fiber systems for gas permeation and
reverse osmosis. Figure 12-7c shows the idealized flow pattern of a hollow
fiber system for ultrafiltration where the feed is inside the fibers. The solution
is well-mixed on the permeate side. Plate-and-frame and spiral wound systems
can have a modification of the cross-flow system, Figure 12-7d, or they could
641
be modifications of the co-current flow system, Figure 12-7e or the counter-

current flow system, Figure 12-7f. The countercurrent flow pattern will be
most advantageous and has been used to design experimental fractionation sys-
tems.
Once the flow pattern is known external mass balances can be combined
with Eq. (12-1) to solve for the outlet concentrations and flow rates. For a
completely mixed system Eq. (12-1) is the same everywhere. Mass balances
for perfectly mixed systems will be illustrated throughout this chapter. These
are the easiest balances to solve, and since perfectly mixed gives the least
separation the design will be conservative. For the other flow patterns the driv-
ing force varies throughout the module and an integration is required. These
external mass balances are developed in Chapter 13.
Cascade schemes are shown in Figure 12-8. When the capacity of a sin-
gle module is insufficient, the obvious solution is to connect two or more
modules in parallel (Figure 12-8a). If a single module is unable to remove the
desired amount of permeable species A, the series arrangement shown in Fig-
ure 12-8b can be used. The parallel-series arrangement is useful when both
high flow rates and high recoveries of A are desired. Note in Figure 12-8c that
the number of modules in parallel can be decreased as more fluid is removed.
A commonly used alternative is the recycle system in Figure 12-8d. This sys-
tem keeps flow rates high to decrease concentration polarization and it allows
for a high recovery of the permeable species. Recycle systems are commonly
used in batch operation to process liquids. Recycle systems can also be put in
parallel to increase the capacity. When the selectivity of the membrane is too
low to produce a pure enough permeate, the permeate in series arrangement
shown in Figure 12-8e can be used. The engineer would prefer to obtain a
better membrane instead of staging the permeate since permeate staging
requires an extra pump or compressor. Thus operating costs will be high. Note
that a recycle stream will often be used for the stream that does not pass
through the membrane. If both streams need to be put in series the counter-
current cascade shown in Figure 12-8f can be used. This countercurrent cas-
cade is similar to other countercurrent cascades discussed in this book except
that additional energy must be input before each stage. This occurs because the
rate process is unable to reuse the energy from stage-to-stage. Because of this,
642
l:
~ ~
fA
~B d41r'
fA
1
e
~~
~ Recycle
~ C A
~
Figure 12-8. Cascades for membrane separators. a. Parallel. b. Series. c.
Parallel-series. d. Recycle. e. Permeate in series. f. Counter-
current fractionation. g. Batch. Species A is more permeable
and species B is less permeable.
operating costs will be high and the use of this countercurrent cascade in a
membrane process will be unusual. In the batch operation shown in Figure
12-8g recovery in a single pass is insufficient, and operation is continued until
enough of the more permeable species has been recovered. Batch systems are
common in applications such as food processing where frequent cleaning is
required. Modifications of these cascades used for gas permeation are shown
in Figure 12-11.
In the remainder of this chapter the different membrane separation sys-

tems will be considered individually. The presentation starts with gas permea-
tion since in many ways it is the simplest of the membrane processes. Then the
liquid systems reverse osmosis, ultrafiltration and dialysis are considered.
Electrodialysis uses electrical forces to separate liquids containing ions, and
643
thus this process introduces new concepts. Following this, pervaporation which
involves vaporizing a liquid at the membrane surface will be considered.
Finally, the experimental liquid-membrane systems will be briefly discussed
12.3. GAS PERMEATION

In gas permeation two or more gas species are separated based on different per-
meabilities in a membrane. Although any of the devices shown in Figure 12-4
could be used, hollow fiber and spiral wound systems are used commercially.
The hollow fiber systems have the advantage of very large surface to volume
ratios. The hollow fibers can have an i.d. up to 200 microns in diameter. The
larger the i.d. the lower the pressure drop inside the tubes ( a surprisingly
important design consideration), but the lower the area/volume ratio. The wall
thickness can be from 25 to 250 microns depending upon the pressures which
must be withstood (Henis and Tripodi, 1983). The skin is from 0.1 to 1.0
microns thick and is formed on the outside of the hollow fiber. Thus flow is
radially inward since this allows the fiber to withstand much higher pressures.
Shell-and-tube separators similar to Figure 12-6 are used. The hollow fiber
systems are currently being used commercially for recovery of hydrogen and
carbon dioxide, and nitrogen generation from air.
The spiral wound configuration is also used commercially for carbon

dioxide recovery using cellulose acetate membranes. This configuration has
the advantages of a lower pressure drop in the direction of flow, and the tech-
nology to build the system is less complex. However, the area/volume ratio is
much lower than in hollow fiber systems. A variety of both hollow fiber and
spiral wound applications are discussed by Spillman (1989).
The basic flux equation written in words in Eq. (12-1) becomes
(12-2)
where h is the volumetric flux, Fp,A is the steady state volumetric transfer rate
of species A through the membrane, A is the membrane area, PAis the permea-
bility of the membrane for species A. the driving force L\p is the change in the
644
partial pressure of the species across the membrane, and lm is the thickness of
the membrane skin. Obviously, consistent units must be used. The usual units
for permeability are cm3 [S1P)crn/cm2 s (cm Hg). Since the partial pressure is
the mole fraction times the total pressure, PA = YAPtot, Eq. (12-2) can also be
written as
J A = Fp,A = PAw!'Yh,A - PpYp,A) (12-3)

A lm
1n this equation Pp is the total permeate pressure, Ph is pressure on high pres-

sure side, Yh,A is mole fraction A on high pressure side and Yp,A is the mole
ftaction A in the permeate. This equation often has to be applied point-by-
point since Y and p can vary. With a large pressure difference across the mem-
brane, it is possible to have Yp,A>Yout,A and still have transfer of this species
through the membrane. In order to obtain separation of different gases the per-
meability of the membrane must be significantly different for the two gases.
The selectivity of the membrane
(12-4)
should be greater than 20 or preferably 40.
For the completely mixed system in Figure 12-7a the external mass bal-
ance at steady state is
(12-5)
FinYin = FoutYout + Fpyp
where F is the molar flow rate and y is the mole fraction. This mass balance is
often written as
(12-6)
Yin =(1- 0) Yout + 0 yp
where the cut 0 is the permeate flow rate divided by the feed flow rate.
(12-7)
645
The cut is usually one of the design parameters. With a specified, we can solve
Eq. (12-6) for Yom
Yin - a yp (12-8)
Yout = 1- a
For gas permeators concentration polarization is usually not a problem

since diffusivities are high and can probably be neglected. Thus the flux equa-
tions apply to bulk: concentrations since the wall concentration equals the bulk
concentration at each point The external mass balance, Eq.(12-5), and the
flux, Eq. (12-3), can be solved simultaneously for the completely mixed separa-
tor shown in Figure I2-6a (Hwang and Karnmermeyer, 1975). For a binary gas
Eq. (12-3) can be written for both species A and B. In molar units the result for
component A is,
(I2-9a)
while for component B,

A
(12-9b)
F p (1 - yp ) = P B1m
ApB [PH(1- Yom) - pdI - yp) ]
where PA and PB are the molar densities at the permeate conditions, and A is
the area for mass transfer. Dividing Eq. (12-9a) by (12-9b), we obtain
yp
A
PA
Yout - [~ 1 yp
(12-10)
-"--- = ClAB -,..- I:'H
----~---''-----
1- yp PB PL
(1 - Yout) - - (1 - Yp)
PH
Substituting in Eq. (12-8) we have
,..
Yp PA (12-11)
--=ClAB -A-
I- Yp PB 1 _ Yin - a yp _ PL (1 _ Y )
I-a PH p
646
Once fractions are cleared, Eq. (12-11) becomes
(12-12)
where
a= [_9 + PL
1-9 PH
1[a PB~A
AB -1]
(12-13a)
PA
b=(I-aAB)- A [
PL
-+--+-- ---
PH
9
1-9
Yin
1-9
1 1
1-9
(12-13b)
PB
A
1
A
PA (12-13c)
c=aAB -A- Yin
[ 1-9
PB
and then
-b± ~bz-4ac (12-14)

yp =
2a
The root which gives yp between zero and one is used. For ideal gases
PA/PB = 1 and the equations simplify. Alternate manipulations of these equa-
tions are shown in Example 12-1.
The penneate concentration increases as the membrane selectivity

increases, the pressure ratio PH/PI.. increases, or the cut 9 decreases. These
effects are illustrated in Section 13.4. We will see that the completely mixed
module provides the worst separation and countercurrent is best.
Example 12-1. Gas Penneation.
We wish to remove COz contaminating CJ4. At 35°C and PH = 20 atm

the penneability of cellulose acetate membrane is (Chern et ai, 1985) P co.
= 15 x 10-10 and PCH,. = 0.48 x 10-10 [cc (STP) cm]/[cmzs cm Hg]. The
647
unit 10-10 [CC (STP) cm]/[cm 2 scm Hg] is defined as 1 Barrer. The gas is
perfectly mixed on both sides of the membrane. The high pressure feed
gas is 90 mole % C~ and 10 mole % CO2 , The permeate pressure is 1.1
atm, and the cut e = 0.5. Determine the membrane selectivity, Yp,eo" the
CO2 flux and the Cf4 flux if Lm = 1 J..UTI.
Solution
..
SeIecUVlty, Pea, 15
a.co..~ = - - = - - = 31.25. The permeate mole frac-
Pea. 0.48
tion Yp = Yp,eo, can be determined from Eqs. (12-12) to (12.14). From
Eq. (12-13),
a= r 0.5
L1- 0.5
+ Q]
20
(31.25 - 1) = 31.91
b~ (- 30.25) [;~ + 1+ ~:! ]- 2 ~ -39.96
~
c (31.25) [~:! ] ~ 6.25
Where we have assumed PAIPB = 1.0. Then from Eq. (12-14)
_ 39.96± ..J1596.8 - (4) (31.91) (6.25)

yp - 63.82
Since O:s; yp :s; 1, use the minus sign. yp = 0.183. Now using mass bal-
ance Eq. (12-8)
Yout
= 0.1 - 0.5 (0.183)
0.5
=0.0168
648
The CO2 flux can be found from Eq. (12-3).
15 x 1010 [cc (S1P) cm 1

co. -
cm2 scm Hg
J - ---~:-----........
(1~) (10-4 cm/1 ~)
[
76cm Hg
1 aun
1
x [(20) (0.0168) - (1.1) (0.183)]aun = l.535 x 10-4 cc S21P
cm s
where Yh,CO. = Yout since the module is well mixed.
Jeu. - [ 0.481~~0- 10
1(76) [20 (0.9832) - (1.1) (0.817)1 - 6.85 x 10-'
Notes: l.) Most of the CO2 is removed, yet permeate gas is still
mainly C~. This is important since driving force for CO2
transfer remains positive.
2.) Since the module is perfectly mixed, Yh,A = Yout in Eq.

(12-3).
3.) JCll. > J co• despite much higher CO2 permeability. This
occurs because of the much higher driving force for C~
4.) Appropriate conversion factors are required.
5.) Concentration polarization (see next section) is assumed to

be negligible.
6.) Note that transfer of CO2 is "uphill" in the sense that CO2
mole fraction is higher on the permeate side. The pressure
drop allows this direction of mass transfer since
Pin,co. > Pp,co •.
7.) Since other flow configurations such as counter-current

will give higher Yp and lower Yout. this design is conserva-
tive.
649
A large literature has developed on the subject of membrane penneabilities

(Chern et at., 1985; Finken, 1985; Hwang and Karnmenneyer, 1975; Kesting,
1985; Koros and Chern, 1987; Rogers. 1985; Stannett et at., 1979; Stern, 1976;
Stem and Frisch. 1981). A simplified picture following Baker and Blume
(1986) will be presented here. The penneability is the product of the solubility
of the gas in the membrane and the diffusivity of the gas in the membrane.
(12-15)
PA =KAD A
The solubility parameter K can be considered an Henry's law constant which

links the concentration of a species in the membrane to the partial pressure in
the adjacent gas. For simple gases the solubility increases as the gas diameter
increases because the gas becomes easier to condense. This is illustrated in
Figure 12-9a where the Lennard-Jones collision diameter is used as a measure
of the size of the gas. As the diameter of the penneate molecule increases, the
diffusivity nonnally decreases becauses large molecules interact more with the
polymer chains. This effect is illustrated in Figure 12-9b. Figures 12-9a and b
are for natural rubber membranes. The values of K and D depend upon the
polymer, but the general behavior will be similar to that shown in these figures.
The penneability is the product of K and D. This product is shown in Figure
12-9c for several different polymers. The diffusivities of most gases in rubbery
polymers are approximately the same; thus, selectivity comes from solubility
differences. Ll glassy polymers both diffusivity and solubility differences are
important in detennining selectivity. Listings of penneabilities for several
gases in a variety of polymers are given by Koros and Chern (1987), Pusch and
Walch (1982), Stannett et at. (1979), and Stem (1986).
Figure 12-9c helps to explain why purification of hydrogen from carbon

monoxide, nitrogen or methane; and separation of carbon dioxide are currently
the major industrial successes. Hydrogen and carbon dioxide have significantly
greater penneabilities than the gases they are separated from since H2 has a
high diffusivity and CO2 has a high solubility. Development of membranes for
other separations is more difficult since more specific interactions between the
penneant and the polymer are probably required to achieve high selectivities.
One way to do this is to use "carriers" which will complex with the desired gas
to help transport it across the membrane (Baker and Blume, 1986).
650
,. 140
,.. 15 150 ..
E%
U E
III
120
'C02
EU ~C02 ......
~ ~
-, 100
r
C E
~
....
Q
lOr
-I
,
tH2 100 III U
u'
i!
=
~.
u%
8 80
iIII
'u \ U = il-;' 60
§~ ~~ r:
1U
5 \
\
50 e.!!
i =
tEEu
Q, U
40
I!
c
~~H'
C
0 Co CH.
'iii .!!!. ::.
=
= H. 02 JiI.::: 2 .. E
.~!
20
C OJ. H4 0 Cu
3.5 4.0 4.5
III ..
%CII 0
0'2.5 3.0 u
C
0 2.5 3.0 3.5 4.0 4.5
Lennard-Jones collision diameter (i) !
'-'
Lennard-Jones cOllision diameter (i)
a b
CO CH.
,
He H. O. N. C02
10,000
~-~---
Silicone
1,000 rubber /
.-- - 'V
!
Natural f
rubber I
.... ~ I.
)... '\/ L
Poly->(" \. /I
c
Co
styrene~ \ -- /
:::. Cellulose _"- \ 1/
acetate '\ -/
"J
0.1
o 2.5 3.0 3.5 4.0 4.5
Lennard-Jones collision diameter (i)
Figure 12-9. Penneation of gases in polymer membranes (Baker and Blume,

1986). a. Diffusivity and solubility in natural rubber. b. Per-
meability in natural rubber. c. Penneabilities in four polymers.
Reprinted with pennission from Chern. Tech., 232 (1986).
651
Obtaining a membrane with high permeability is not sufficient to have an

economical commercial separator. In addition, we must be able to form the
membrane into a geometry which will be able to withstand fairly high pres-
sures, will have high area/volume ratios, and will have a thin active skin with
no holes. Spiral wound cellulose membranes and hollow fiber poly sulfone
membranes satisfy the first two requirements. The third requirement is much
more difficult. The thinner the skin the more likely there will be imperfections
or pores which are much larger than the gas molecules being separated. The
gas flows through these pores by convective flow driven by the pressure gra-
dient across the membrane. Unfortunately, this gas does not undergo the
separation mechanism and the concentration of this gas will be the same as that
on the high pressure side of the membrane. Although significant strides have
been made in producing more perfect membrane skins, this approach has not
yet made thin skinned membranes which are perfect enough for hydrogen
purification. A surface porosity as low as 1~ will lower the separation factor
of H2 compared to CO for poly sulfone membranes from 40 to approximately 5
(Henis and Tripodi, 1983). This value is not high enough to obtain pure hydro-
gen.
Two methods are used in commercial separators to circumvent this prob-

lem. For CO2 purification cellulose acetate membranes are used with a fairly
thick skin. This reduces the permeability, but with CO2 this is less critical
since CO2 has a very high permeability (see Figure 12-9c). The second solu-
tion used for H2 purification is to coat an asymmetric polysulfone membrane
which inherently has a high separation factor with a layer of silicone rubber
(Henis and Tripodi, 1983, Finken, 1985; Schell, 1985) and Stem (1986). The
silicone rubber layer serves to seal the defects. These resistance model (RM)
membranes are illustrated schematically in Figure 12-10. The silicone rubber
has a very high permeability (but low selectivity). The permeabilities [cm 3
(STP) cm/cm 2 s cm Hg] of silicone rubber are 5.2 x 10-8 and 2.5 x 10-8 for H2
and CO, respectively. The polysulfone has permeabilities of 1.2 x 10-9 and
3.0 x 10-11 for H2 and CO, respectively (Finken, 1985). The RM membrane
will have its permeability reduced slightly, but has a selectivity> 30 which is
high enough for commercial H2 purification. Note in Figure 12-10 that it is
difficult to determine exact values for 1m for either layer. Experimentally, it is
652
Silicone Rubber
Defect
}-Thin Skin
Polysulfone
Support Layer
Figure 12-10. Resistance model (RM) gas permeation membrane.
easiest to measure the ratio Pltm and to not separate the tenns. Other commer-
cial applications are discussed by Schell (1985) and Stem (1986).
A first cut at design of a gas penneator can be made once the penneabili-
ties, pressure drop, thickness of the active layer, and area per volume of the
spiral wound or hollow fiber system are known. To detennine the amounts and
compositions of both high and low pressme products the flow patterns inside
the separator and the product cut must be specified. These details are discussed
in Chapter 13.
The single stage systems shown in Figures 12-5 to 12-7 are often not
optimal and modifications of the cascades shown in Figure 12-8 are often used.
Three cascades used commercially for gas penneation are shown in Figure
12-11 (Spillman, 1989). The two-stage process with recycle shown in Figure
12-11a is used to improve recovery of the slower penneating species at the cost
of additional compression and membrane area. If the concentration of the per-
meating species is very high, the "premembrane" shown in Figure 12-11b can
be useful for bulk removal. According to Eq. (12-3) a low penneate pressure
will decrease membrane area or increase product recovered; however, if
recompression is required a low penneate pressme increases recompression
costs. One compromise is the two stage process shown in Figure 12-11c where
the penneate pressures are different. Since gas compression is done in stages,
the two penneate pressures are chosen to match the compression scheme.
Currently. gas penneators are speciality items which would be ordered

from the manufacturer. The penneators are often ordered as a turn-key device.
That is, the purchaser specifies required operating conditions including the inlet
653
Membrane
~~~~~~--~8 E
Feed
Membrane
Corr4:lress
Permeate
b
Bulk Removal Ptrification
Permeate
c 60% Hydrogen
20% Nitrogen _ _ _ _ ....-,
20% Hydrogen
20% Inerts r-
,--_ _ _ - , 4;~%N:~r~r~:n
2,000 psi
PURGE GAS FUEL GAS
1,000 psi
90% Hydrogen 88% Hydrogen

6% Nitrogen 70/0 Nitrogen
4% Inerts 50/0 Inerts
86% Hydrogen Recovery

Figure 12-11. Cascades for gas permeation (Spillman, 1989). a. Two stage
with recycle LO improve recovery of the slower permeating
species. b. Three stage showing addition of a "premembrane"
for bulk removal. c. Two stage with different permeate pres-
sures. Reprinted with permission from Chern. Eng. Prog ..
85(1),41 (1989). Copyright 1989, Amer. Inst. of Chern. Eng.
654
gas composition and flow rates, temperature and pressure, and the required
outlet gas composition for both high and low pressure gases. The manufacturer
then designs and builds a system which he guareentees will meet the required
specifications. Unfortunately, this arrangement is not as full-proof to the pur-
chaser as it sounds. If the actual operating conditions are different than
specified, the guareentee may be voided. In addition, the supplier will tend to
over-design to be sure he meets the specifications. Finally, tum-key units tend
to be more expensive than systems where the buyer does the design. It is best
if the buyer knows as much as possible about the separator even when he is
buying a tum-key system. This allows the buyer to write a more advantageous
contract and to do a better job bargaining. Additional factors which can be
important in design are discussed in Chapter 13.
The economics of gas permeation separation compared to competing
processes is discussed in detail by Spillman (1989). He concludes:
1. Membranes are efficient concentrators, but are not efficient in producing

100% purity.
2. Only optimized systems including multi-stage processes such as those

shown in Figure 12-11 should be used in economic comparisons.
3. Recompression costs are very important and must be included in the

analysis.
4. Since the technology is changing very rapidly, economic analyses can

become rapidly outdated.
5. Hybrid systems (membranes plus another separation) are often advanta-

geous. This occurs since membranes work best at low concentrations of
the permeating species while systems such as absorption work best at high
concentration. Examples of hybrid systems for air separation are given by
Beaver et al (1988) and Gienger and Ray (1988).
12.4. REVERSE OSMOSIS (RO)
Reverse osmosis (RO) is a process which reverses the normal direction of

osmosis by increasing the pressure of the concentrated stream. Obviously, this
655
definition is useless unless we know what osmosis is. Thus we need to digress
slightly into the thennodynamics of membranes.
Osmosis is the flow of solvent through a semipenneable membrane from

the less concentrated to the more concentrated region. A "semipenneable
membrane" is a membrane which allows solvent to pass but completely
prevents the flow of certain solutes or ions in solution. The osmotic flow is a
natural occurrence as the system tends to come to equilibrium and equalize
chemical potentials. The osmotic flow can be decreased by applying a pressure
to the concentrated solution. The higher the applied pressure the less the
osmotic flow. When the flow stops, the applied pressure is the osmotic pres-
sure which is given the symbol1t. At higher pressures flow will be reversed
and we will transfer solvent from the concentrated to the dilute solution. This
is reverse osmosis.
As long as the membrane is perfectly semipermeable, the osmotic pres-

sure is a thennodynamic property of the solution and is independent of the
membrane properties. The osmotic pressure can be measured directly,
although it is often much more convenient to estimate it from other thenno-
dynamic quantities. Osmotic equilibrium requires that the chemical potentials
of the solvent on both sides of the membrane are equal.
(12-16)
Jll.solvent = Jl2,solvent
Note that the equilibrium condition is on solvent only. Since solute does not
pass through the membrane, solute ~s not in equilibrium. Thermodynamic
arguments can now be used to derive equations which allow calculation of the
osmotic pressure (for example, see Reid, 1966). Assuming that the liquid is
incompressible and that activities can be estimated from vapor pressure meas-
urements, the result is
ltVsolvent
VPpure solvent
= TIT In [ VP .
1 (12-17)
soluuoo
where vsolvent is the partial molar volume of the solvent Equation (12-17) is
usually quite accurate, but is not always convenient to use since it requires
656
Table 12-1. Osmotic pressure of aqueous sucrose

solutions at 30°C (Reid, 1966).
7t in atmospheres
c, gmoles/liter Van'tHoff Eq. (12-17) Experimental

Eq. (12-18a) Data
0.991 20.3 26.8 27.2
1.646 30.3 47.3 47.5
2.366 39.0 72.6 72.5
3.263 47.8 107.6 105.9
4.108 54.2 143.3 144.0
5.332 61.5 199.0 204.3
vapor pressure data of solutions. A simpler result is obtained by assuming that

the solution is dilute and that Raoult' s law is valid. This result is the van' t Hoff
equation,
(12-18a)
7t=cRT
where c is the molar concentration of solute in solution (moles/volume).

Although very convenient, the van't Hoff equation is often incorrect. This is
illustrated in Table 12-1 which presents the calculated and experimental values
for the osmotic pressure of aqueous sucrose solutions (Reid, 1966). For this
system Eq. (12-17) is quite accurate while the van't Hoff equation seriously
underestimates the osmotic pressure. If the solution dissociates or associates
in solution Eq. (12-17) remains valid, but Eq. (12-18a) should not be used.
Although the van't Hoff equation may be invalid it is often useful to assume a
linear dependence on concentration
(12-18b)
7t=ac
where a is an empirically determined constant, and c may be in any desired

units. For macromolecules the osmotic pressure increases much faster than
657
linearly, and can often be represented as (Wijmans et ai., 1984),
(12-18c)
where a is a constant and n> 1. Often n is approximately 2.
The osmotic pressure of solutions is not widely tabulated and vapor pres-
sure data may not be available. Fortunately, the osmotic pressure can be
related to freezing point depression which is easy to measure and is widly tabu-
lated (e.g. in Handbook of Chemistry and Physics). The osmotic pressure is
related to freezing point depression by (Reid, 1966),
(12-19)
where T; is the freezing point temperature of pure solvent, Tf is the freezing

point temperature of solution of the desired mole fraction and .1 Hf is the latent
heat of freezing of the solvent. The ratio ~lRenergy is the ratio of gas con-
stants in pressure and energy units. This conversion is required to obtain 1t in
pressure units (see Problem 12-F2). Note that both Eqs. (12-18a) and (12-19)
predict 1t depends linearly on T. Use of Eq. (12-19) is illustrated in Example
13-1.
The flow in reverse osmosis is driven by pressure differences. Although

concentration polarization is usually important in reverse osmosis (see Figure
12-4), for the moment we will assume that the solutions are perfectly mixed so
that there is no concentration buildup. Then Cw = Cb (This development follows
that of Blatt et ai, 1970). Equation (12-1) becomes
K.olv
Jso1v = --(.1p - .11t)
(12-20a)
1m
where K.olv is the permeability of the membrane to the solvent (usually water),
.1p is the pressure drop across the membrane, and .11t is the osmotic pressure
difference across the membrane. Note that the pressure driving force is
reduced by the osmotic pressure difference. As the solution becomes more
658
----L)- -IT I . ··l.111-

3.0
(a)
1 - ;
i
-~ - .. -
I
Reference IS 800 pSlg, 25·C, 30% recovery i
, and 30,000 ppm NaCI feed
I
2. To find productivity at any other concentration i ! I
---
-~I--
.. __
and pressure, multiply data in Table I by the ~T-~
J~_
._.
l.-
appropriate factor
..J.--
---
II: 2.0
V V
--
~ I
·. f-
~
>-
~ -~
..
N()CI leeO c.-- l--
.-- -
l-- V
--
I-
:;
t .\,'!l~: ....--V V
....
~
o
o
II:
~k
k:
. r::
...- ...
...
j:; . \O,~ ...-::. .. '
V k-
~
I
---
..- i-"'""
1.0
p. .
Q.
~ :. I
~~
.. 009911\
L.;..rl
....
... 'l:,
00 . ;::;...,.-- V
~.,
. . .. 11\
I--
:.;- ~ ~
" ,.
·
... '!)o,oo~j...--
,. t'
.. L--I-
"-j • 1" " • I • •
..
--
~.
V~ r-
k ~ l-- ttO~
>-:: ...
...
.. "1'
· t~ •• , t
V
o
400 500 600 700 800
PRESSURE, psig
(b)
_. --,+ . . . ' ....... -,,:, ---'-,. .
-= =
~
~.:: :-::-: HI. Reference is 800 psig,' 30·/~ recovery ,

. ... . .. land 30,000 ppm NaCI feed. i-- h---,' .~.~. --·'~--lf~· I~
~I'-=:~: I~2. To find salt passage at any other can- I-+Lt::t=--.:We-:-- •::=;
~ 2.0 1---., -;w .J centration and recovery, multiply data - '. - 40 r'jt:-t:-:::.:. ..
t; ~= ~ ~ in Table 1 by the appropriate factor. ~f-W+. ~:.. Jyt::~ --:-.
~ :::.... ::-.: h ++T ::tixf-:;:;=:-:- _ . ::::.f:-'-"' -+j- ~,Itil: :-:.I.::-::;;II::r:::
-./
~ t:::""-R::- !+H=t-:H-'FH+t~ ~¥ - ;:;- : +,(--= .:: j/': 3oo4=lr :-":::
~_ ~ -+-4-+1-+-+-+
r=;=1---t+.:...J.
. ,.. ~. fEf,,+t. ;-~f=I '.~" -I--~.-. V_'
:::. .
_
:
Q. 1--. !=h.
:.::e::-:~~~jrr: ':-:'t=F '.~c::t= ". h+T1 It+==--.::±

o .... -- 1. .. - - - L.. tn:: ,f+. 1+4+ i Lg+- . !~~ . . ' .
o 10,000 20,000 30,000 40,000
FEED CONCENTRATION, ppm NoCI
659
concentrated, 1t increases and higher pressures are required to achieve accept-

able fluxes.
Real membranes are partially permeable. Thus there will be some solute
flux through the membrane. Neglecting concentration polarization, the flux of
solute is
(12-21)
where Ksolutc is the permeability of the membrane to the solute, and c p is the
permeate concentration. Conservation of mass requires that
(12-22)
which relates the two flux equations. Experimentally the rejection coefficient
(12-23)
is often reponed. Combining Eqs. (12-20a) to (12-23), we obtain the rejection

coefficient RO with no concentration polarization
RO = Ct.(~p - ~1t) (12-24a)

1+ Ct.(~p - ~1t)
Figure 12-12. Reverse osmosis results (Caracciolo et ai, 1977). a. Effect of

pressure and feed concentration on production from a DuPont
B 10 hollow fiber RO system for desalinating water. Produc-
tivity at reference conditions (referred to as Table 1) is 1500
gpd. b. Effect of feed concentration and recovery on salt pas-
sage for a DuPont B 10 hollow fiber RO system for removing
NaCI from water. Salt passage at reference (referred to as
Table 1) is 1.5 wt%. Reprinted with permission from S.
Sourirajan (Ed.), Reverse Osmosis and Synthetic Membranes
(1977). Copyright 1977, National Research Council of
Canada, Onawa.
660
where a is the selectivity,

(12-24b)
a=KsolvlKsolute
For a perfectly impermeable membrane the selectivity is infinite and R= 1. For

membranes with finite selectivity some increase in the observed rejection
coefficient can be obtained by increasing ~p. This also increases the solvent
flux. This pressure effect is illustrated in Figure 12-12a (Caracciolo et ai,
1977) for a hollow fiber membrane.
Concentration polarization is usually important in reverse osmosis.

Equation (12-20) remains valid except we must now calculate ~1t using the
concentrations across the membrane. Referring to Figure 12-4,
~1t=1t(cw)-1t(cp). Since cw>Cb, M increases and solvent flux decreases as con-
centration polarization increases. Equation (12-21) changes since the bulk fluid
concentration Cb is replaced by the wall concentration Cw. This is usually writ-
ten as
(12-25)
where M is the polarization modulus defined as
(12-26)
As M increases the solute flux increases since any solution which leaks through
the membrane is more concentrated.
The solvent flux, Eq. (12-20a), can be written in terms of M if we assume

that 1t is proportional to concentration (Eq. (12-18b». This makes the osmotic
pressure difference
and Eq. (12-20a) becomes
(12-27)
661
Note that this predicts that solvent flux decreases as M increases. Combining
Eqs. (12-22), (12-23), (12-25) and (12-27), we obtain an expression for the
rejection coefficient when there is concentration polarization.
(12-28)
As expected R decreases as M increases except when the selectivity ex is

infinite. Equation (12-28) reduces to Eq. (12-24a) when M=1. Additional
manipulations of these equations are illustrated in Example 12-2.
Example 12-2. Reverse Osmosis of Na2C03
An experimental RO membrane is being used to produce a waste stream

which is Na2C03 solution. The pure water flux at 30°C is measured as
J so1v = 1500 L/m2 day when hop =25 abn. With the Na2C03 solution,
operation is at 30°C with Ph = 50 abn and PI.. = 1 abn. Both permeate
and feed sides of the membrane are assumed to be perfectly mixed.
a. For an ideal membrane (R = 1) find Jso1v when M = 1 and when M

= 1.2.
b. For a real membrane we measure cp = 0.005 when M = 1. Find

J so1v when M = 1 and when M = 1.2.
Solution
A. Define. The problem is clearly defined in the problem statement.
B. and C. Explore and Plan. Although the definition is clear, the path
to solution is not. We obviously need to determine osmotic pressure.
Equation (12-17) is accurate. Vapor pressure data (perry and Green,
1984), p. 3-73) at 30°C is:
wt% Na2 C03 0 5% 10% 15% 20% 25% 30%
VP, mm Hg 31.82 31.2 30.4 29.6 28.8 27.8 26.4

662
From the pure water flux we can find Ksolv/t.n from Eq. (12-20a). Then
for part a J so1v is obtained from Eq. (12-20a) when M = 1.0 and from
Eq. (12-27) when M = 1.2. For the real membrane with M = I, cp can
be found from Eq. (12-23). Then Jso1v is determined from Eq. (12-20a).
We can also use the measured value of R to find a = K.olvlK.olute from
Eq. (12-24a). This gives us K.olute/tn,. This value is needed when M =
1.2. Combining Eqs. (12-22), (12-25) and (12-27) we can solve for Cp.
Once Cp is known Eq. (12-27) is used to find Jso1v .
D. Do It. Eq. (12-17), X = -RT-

vwater
In [ VP
VPwater
.
solution
1
(MW)w 18.016 glgmole
V water = Pw = 0.996547 glee = 18.095 cc/gmole
For 5 wt % solution at 30°C (Cb =0.05),
[ 82.057 cc atm ] (303.16K)

gmole K [ 31.824]
x= In =27.2 atm
(18.095 ee/gmole) 31.2
Assuming linear dependence, Eq. (12-18b),
c
x = 0.5 (27.2) = 544.5c, or a = 544.5
Part a. Ideal (R = 1). For pure water Eq. (12-20a) becomes
K.olv _ J so1v _ 1500 _ 60 L/( 2 da

- -------- ,m yatm )
t.n Ap 25
When M = I, Ap = 50 - 1 =49 atm, Ax = 27.2 atm (x (c p ) = 0). Note

Ax is the same everywhere since the membrane module is perfectly
mixed.
Ksolv 2
Jsolv = - - (Ap - Ax) = (60) (49 - 27.2) = 1306.6 LIm day
tn,
663
Ksolv
When M = 1.2, J solv =- - [~p - M 1t (Cb)]
fro
J so1v :;;:; (60) [49 - (1.2) (27.2)] =979.9 L/m2 day
Pan b. Real Membrane. For M = 1 we observe c p = 0.005. Rearrang-
ing Eq. (12-23a), we obtain
R =1- .2 = 1 - 0.005 = 0.9

Cb 0.05
For a linear dependence of osmotic pressure,
1t (cp ) = (0.1) 1t (5%) = 2.72 attn

Equation (12-20a) can be written as,
(12-20b)
This equation is valid everywhere since Cb and cp are constants.
Jso1v = 60 [49 - (27.2 - 2.72)] = 1469.9 L/m2 day
Note that solvent flux of the real membrane is greater than that of the
ideal membrane since ~1t is decreased.
Rearranging Eq. (12-24a) to solve for a,
Ksolv R (12-24c)
a = Ksolute = -:-----:---:-:---=:-:-
(~p - ~1t) (1 - R)
Then, a = (24.~)~0.1) = 0.367

Ksolute = Ksolv/tm = ~ = 163.32 L
fro a 0.367 m2 day
664
For M = 1.2 we need to solve Eqs. (12-22), (12-2S) and (12-27) simul-
taneously. The resulting equation is
(12-29)
where a is defined in Eq. (12-18b). Equation (12-29) is easily solved

with the quadratic formula. The three constants for the quadratic for-
mula are
a Ksolv = (S44.4) (60) = 32.669

t.n
and
K.olv ( A
- - LlP-
M 1t (» Ksolute
Cb + - -
t.n t.n
= 60 [49 - (1.2) (27.2)] + 163.3 = 1143.2
and
K.olv
- - - M Cb = - (163.32) (1.2) (O.OS) =- 9.8
1m
From the quadratic formula,
- 1143.2 ± ....j(1143.2)2 - 4(32,669) (- 9.8)

cp = 2 (32,699)
To have JX>sitive cp must use + sign. cp = 0.00712. Then from Eq.

(12-27) and (12-15),
J I01v = 60 {49 - [(1.2) (27.2) - (0.00712) (S44.5)]} = 1212.S
JlOlute = (163.3) [(1.2) (O.OS) - 0.00712] = 8.64

665
E. Check. Can use Eq. (12-22) as one check. For Part b with M = 1.2
(the most difficult problem),
8.64 = Jso1ute = cp Jso1v = (0.00712) (1212.5) = 8.63 which is OK.
F. Generalize. There are several parts of this detailed example:
1. Note that the units on the gas constant R controls the units of 1t.
2. The value K.olv/lm determined from data can be used under dif-
ferent conditions.
3. As M increases Jsolv decreases.
4. Note that Jsolv increased as R decreased since permeate osmotic

pressure increased.
5. K.oluteltm calculated from observed retention can be used under

different conditions.
6. Calculation of cp by simultaneous solution of mass balances when

R < 1.0 and M > 1.0 using previously determined values of
K.olute/tm and Kaolvltm, differed from the method used in Example
12-1 since different information was given.
7. If we do not have perfect mixing, Cb and c;, will both vary. Thus
1t (Cb) and 1t (c;,) vary which means the flux varies as we go down
the membrane. Mass balances for other situations are illustrated in
Section 13.3.
8. If the cut e = Fp/Fin is specified, then the required feed concentra-

tion can be determined from the analogue to Eq. (12-8),
(12-30)
9. This system was a continuous, steady state process. Batch opera-

tion is considered in Eqs. (12-43) to (12-46).
666
Determination of the polarization modulus M is important When rejec-

tion is complete (R=l), a steady state, one-dimensional mass balance on the
feed side of the membrane is (see Figure 12-4),
(12-31)
This equation is a statement that the movement of solute to the membrane by

convection (the bulk flow through the membrane) is equal to the movement of
solute away by diffusion. At the membrane wall (y=O) c equals cw , and in the
bulk fluid which starts at y=O, c equals Cb' With constant diffusivity the solu-
tion to Eq. (12-31) is
(12-32a)
Since D/o = k, the mass transfer coefficient, this result is often written as
M=-=exp
CW
Cb
-S01V
- [J 1
k
(12-32b)
Although considerably over-simplified and difficult to use without a value for

film thickness 0, or mass transfer coefficient k, these results do show that con-
centration polarization increases as:
1. Solvent flux increases. Thus highly permeable membranes have

more of a problem with concentration polarization.
2. The boundary layer thickness 0 increases. The boundary layer

thickness can be decreased by decreasing the channel diameter or
width or by promoting turbulence.
3. The diffusivity increases. This shows that large solutes will have
worse concentration polarization.
More detailed theories are discussed in Chapter 13.

667
The effect of feed concentration on the productivity of an hollow fiber

RO permeator was shown in Figure 12-12a (Caracciolo et ai, 1977). This con-
centration effect is due to concentration polarization and the increased osmotic
pressure at higher feed concentrations. Salt passage as a function of feed con-
centration and the recovery (which is the same as cut) is shown in Figure 12-
12b (Caracciolo et ai, 1977). As feed concentration or recovery increase, the
salt passage increases. This would be expected, but the nonlinear increase
illustrated is a surprise. This effect is due to the worsening of concentration
JX>larization at increasing concentrations. Typical rejection coefficients for a
variety of membranes are listed in Table 12-2 (Pusch and Walch, 1982). More
detailed tables for a variety of inorganic and organic compounds are compiled
by Eisenberg and Middlebrooks (1986).
The spiral wound and hollow fiber configurations are commonly used for
RO. The common membrane polymers include cellulose acetate, polyamides
(see Figure 12-3), and a composite membrane developed by FilmTec. The
composite membrane is a crosslinked polyamide barrier layer supported by a
microporous polysulfone coating on a JX>lyester web. This complex membrane
has high flux, high salt rejection, a pH range from 4 to 11, and a wide range of
temperature stability. Other membranes are listed in Figure 12-3 and Table
12-2.
The skin of the RO membranes have pores which are several Angstroms
or larger in diameter. These pores are larger than the hydrated ions, yet the
ions are excluded. The separation is clearly not a sieving effect. The current
physical model of the separation effect can be summarized as follows (Kesting,
1985). The water molecules are preferentially attracted to the hydrophilic
membrane surface. An ordered water sheath of thickness t (- ~) exists at the
membrane surface and repels ions. This water sheath pours into pores. If the
pore is less than 2t (- 10~) in size the ordered water sheath will be continuous
and ions will be excluded. In the occasional large pores (imperfections) with
diameter > 2t, the ordered water sheath does not fill the pore. In the center of
these pores ordinary water plus ions pass. The observed rejection coefficient is
the sum of the almost pure water from the small pores and the contaminated
water from the larger pores. These membranes are often called "solution-
diffusion" membranes.
0\
0\
00
Table 12-2. Rejection coefficients and fluxes for a variety of RO membranes (Cadotte et al, 1988; Pusch
and Walch, 1982). HF = hollow fiber, PF =plate and frame, SW = Spiral wound.
Membrane Module Manufacturer R Flux Test

% L/m2 day Conditions
Cellulose-3 acetate HF DowChem 97 200 28 bar, 0.3% NaCI
pH 6-8, 35°C
Cellulose-2.5 acetate PF DDS ~99 ~400 40 bar, 1% NaCl
pH 2-8, 30°C
Cellulose-2.5 acetate SW Millipore 90 14 bar, 0.1 % NaCI
Cellulose 2-acetate HF Toyobo 93 30 bar, 0.2% NaCI
pH 3-7, 30°C
Cellulose acetate Tube UOP 97 450 40 bar, 0.5 % NaCI
pH 3-7, 45°C
Cellulose-2.8 acetate SW UOP ~99 400 70 bar, 3.5% NaCl
pH 4-7.5, 35°C
Aromatic Polyamide (B9) HF DuPont 95 50 28 bar, 0.15% NaCl
pH4-11,35°C
Aromatic Polyamide (B 10) HF DuPont 98.5 40 56 bar, 3.5% NaCI
pH 5-9, 35°C
Composite PF FilmTec 98 1800 40 bar, 1% NaCI
$60o C
Composite, polyamide SW Film Tech 99.5 20 gfd 1.4 :rv1Pa, dilute NaCI
FT30
Polyfuran, composite SW Osmonics 99 200 56 bar, pH 0.5-10.5
75°C
Composite SW Toray 99.7 700 400 bar, 0.25% NaCI
pH 4-12, 40°C
Composite SW UOP 99.2 600 63 bar, 3.5% NaCI
pH 2-12, 45°C
Composite, polyamide SW Film Tech 50 (NaCI) 20gfd 0.7 :rv1Pa, dilute
XP45 (Nanofilter) 97.5 (MgS04 )
Composite, polyamide SW Film Tech 75 (NaCI) 20gfd 0.4 MPa, dilute
NF70 (Nanofilter) 97.5 (MgS04 )
0-
0-
\0
670
As shown in Figure 12-12a the required pressure depends on the feed

concentration and also on the recovery of water. For brackish water (1500
mg/L dissolved solids) 1t is about 104 kPa (15psi) and a typical pressure range
is from 2,760 to 3, 137 kPa (400 to 600 psig). For a typical seawater of 35,000
mg/L dissolved solids the osmotic pressure is about 2,415 kPa (350 psi). Since
a 50% recovery will approximately double the osmotic pressure, operating
pressures from 5,515 to 6895 kPa (800 to 1000 psig) are often used for seawa-
ter desalination (Applegate, 1984).
The effect of recovery on salt passage was illustrated in Figure 12-12b.

Increasing the recovery decreases the average water flux since the average salt
concentration on the high pressure side of the membrane is raised. This, in
tum, increases the osmotic pressure and decreases the flux. Typical recoveries
are about 33% for seawater and up to 90% for low salinity brackish water
(Kesting, 1985). In under the sink units for treating tap water recoveries as low
as 10% are used. These low recoveries decrease the purchase price of the unit,
but raise the operating cost since more water is thrown away. However, the
average consumer is much more aware of the initial purchase cost than the
water costs. Calculating the effect of recovery or the product concentrations
requires a knowledge of the flow geometry, the degree of concentration polari-
zation, and the external mass balances. Details of the calculation procedures
are in Chapter 13.
The reverse osmosis module needs a variety of supporting equipment to
function properly (Applegate, 1984; Cheremisinoff and LaMendola, 1983;
Eisenberg and Middlebrooks, 1986; Hwang and Kammermeyer, 1975; Porter,
1979; and Sourirajan, 1977). This is illustrated in Figure 12-13 where a water
treatment plant is illustrated. The raw water must be stored to avoid temporary
interruptions in supply. If required, chemicals will be added to adjust the pH,
control biological growth, prevent calcium carbonate or other scaling, or aid in
coagulation and settling. For very dirty waters a settling basin and/or a prefilter
would be added to Figure 12-13. The water is then sent to a micron sized filter
to remove particulates which could clog the membrane. Then a high pressure
pump is used to boost the pressure. The pressure control valve is used for pres-
sure control and the high pressure switch is for safety in case a valve is inad-
vertently closed or a pipe becomes clogged. A parallel-series arrangement of
671
Permeate
Product
Storage
F
Pressure Bypass
Control
Valve Waste Water
to Disposal
Figure 12-13. Complete RO system showing auxiliary equipment
penneators with bypass lines is shown in Figure 12-13. The bypass lines are
used to mix in raw water if less pure water is required. The water which does
not pass through the membranes is more concentrated in the dissolved solids.
Usually, this stream is a waste stream which must be disposed of. The product
water is sent to a storage tank where additional chemicals may be added. For
example, if the water is for drinking chlorine or ozone will be added here.
Chlorine is detrimental to the membranes and thus is added after the mem-
branes unless it is required to stop biological growth. When treating highly
fouling materials such as food products, a cleaning cycle must be added. This
requires shut-down of the equipment, and it requires quite a bit of hardware in
addition to that shown in Figure 12-13. Details of cleaning cycles and the
chemicals used for cleaning are given by Eisenberg and Middlebrooks (1986).
Unfortunately, this can greatly increase the costs.
Membranes have a finite lifetime which depends upon the operating con-
ditions. Membrane replacement is usually treated as an operating cost. In
addition, there is usually a long tenn flux decay caused by membrane compac-
tion and fouling of the membrane. To take this into account the initial flux of
the membrane is usually not used in the design calculations. Instead the flux
after a few months of operation is used since the additional decay after this
time is quite slow. Obviously, the membrane must be protected from chemical
upsets since the membranes can be destroyed quite quickly.
There are, of course, other ways to purify water such as distillation,

adsorption, ion exchange, and electrodialysis. The method used depends upon
672
the economics of the particular case. The economics of RO are often favor-
able, and RO is used very extensively for treating brackish water, seawater, and
tap water. One of the largest RO units is at Ras Abu Janjur, Bahrain, which
produces 12.7 million gallons of potable water per day from brackish well
water. The system consists of seven trains of hollow-fiber RO units with 2100
membrane modules per train. RO applications and economics are discussed by
Belfort (1984), Cheremisinoff and LaMendola (1983), Applegate (1984), and
Eisenberg and Middlebrooks (1986).
Reverse osmosis has also been used for separation of organics from
water. Although this is not yet a common industrial separation, it may be
important in particular cases. For example, dilute £-caprolactam solutions are
concentrated commercially using RO. If the organic recovered dissolves poly-
mers or if the module needs to be steam sterilized, ceramic membranes can be
used (Hsieh, 1988). Data is presented by Eisenberg and Middlebrooks (1986),
Leeper (1986), and Sourirajan (1977).
Nanofiltration is a variant of reverse osmosis using membranes which

pass small solutes but retain somewhat larger molecules such as sucrose.
Nanofiltration membranes would typically have NaCI rejections from 20 to
70% and organic cutoffs in the 200 to 500 MW range. (Cadotte et ai, 1988)
This cutoff range corresponds to molecules with a diameter of approximately
10 ~ which is one nanometer. These membranes are thus significantly looser
than RO membranes but tighter than ultrafiltration membranes (minimum cut-
off about 10,000 MW). The nanofiltration membranes operate at pressures in
the range 0.4 MPa to 0.7 MPa which is from 1/4 to 1/2 a seawater RO mem-
brane. Some results are given in Table 12-2. The rejection mechanism is the
same as for RO, and the membranes can be thought of as loose RO membranes.
There are many applications where the ability to discriminate between

salts and moderate molecular weight organics is useful. For instance,
nanofiltration membranes can desalt whey or sugars. Nanofiltration mem-
branes can also be made with 50% rejection of NaCI and 97.5% rejection of
MgS04 • These then are an alternative to water softening by ion exchange (see
Section 9.5). Nanofiltration is a new development which can be expected to
grow in the future.
673
Osmotic concentration can be considered a variant of osmosis. In

osmotic concentration a dilute stream we wish to concentrate is separated from
a concentrated salt solution by a reverse osmosis membrane such as cellulose
acetate. Due to osmosis water migrates from the dilute stream to the concen-
trated salt stream. This concentrates the dilute stream and simultaneously
dilutes the salt stream. The energy for this process comes from the energy
required to evaporate water to reconcentrate the salt solution. If a waste brine
stream (say from an RO plant) is available, than the energy required for con-
centration may be practically free.
12.5. ULTRAFILTRATION (UF)
Ultrafiltration (UP) is used for retaining larger solutes than in RO, but many of
the operating characteristics are very similar to RO. The osmotic pressure in
UP is invariably quite low since the molar concentration of the high molecular
weight molecules separated by UP is quite low even when the weight concen-
trations are high. Because a large osmotic pressure does not have to be over-
come, much lower operating pressures are used in UP than in RO. Typical
operating pressures are in the range from 10 to 100 psig.
The membranes used in UP are usually asymmetric polymers used in one
of the geometries shown in Figure 12-5. A variety of polymers such a cellulose
acetate, aromatic polyamide, polysulfone, polyvinyl chloride, polyacrylonitrile
and polycarbonate have been used to cast membranes (see Figure 12-3).
Polysulfone has a wide temperature tolerance (to 93° C) and a wide pH range
(0.5 to 13) and is probably the most commonly used membrane material
(Applegate, 1984). The membranes are formed so that the dense skin is
microscopically porous. Thus the separation mechanism can be considered (in
a somewhat over-simplified form) to be sieving of the molecules. The mem-
branes can be tailor made to fix the desired pore size distribution and thus pro-
vide for different molecular weight cutoffs. The molecular weight cutoff is
loosely defined as the molecular weight below which species begin to pass
through the membrane. The values for molecular weight cutoffs can range
from 1000 to 80,000. Values for a variety of membranes are listed in Table
12-3. Molecular weight cutoffs should be employed with discretion since siev-
0\
--.I
~
Table 12-3. Molecular weight cutoffs and fluxes ofUF systems. (pusch and Walch, 1982; and
Warashina et ai., 1985). HF:;: hollow fiber, PF:;: plate and frame, SW:;: spiral wound
Membrane Module Manufacturer MW Flux Test

L/m2 day Conditions
Cellulose 2-aeetate SW Abeor IS,OOO 6000 3.S bar, SOoC,

pH 3-7.S
Polyvinylidene fluoride SW Abeor 10,000/20,000 7S00 3.S bar, 90°C,
pH 2-9
Cellulose acetate PF DDS 6000 to 6S,000 2-7000 S bar, SO°C,
pH 2-8
Polysulfone PF DDS 8000/20,000/ 1-7000 3 bar, 80°C,
6S,000 pH 0-14
Polysulfone PF Dorr-Oliver 10,000 10,000 2 bar, 70°C,
pH 2-12
Polysulfone SW Millipore 80,000 2000 5 bar, 32°C,
pH 3-11
Polysulfone PF Osrnonics 1000/20,000 4000/12,000 3.5 bar, 93°C,
pH 0.5-13
Cellulose acetate Tube Patterson Candy 1000-20,000 pH 3-6, 30-S0°C
Co-poly acrylonitrile/ PF Rhone-Poulenc 20,000 7000 2 bar, 40°C
Methallyl sulfonate pH 1-10
Polyacrylonitrile HF Asahi Chern. 13,000 6380 to 19,100 3 kg/crn 2 , 25C
pH 2-10
Polyacrylonitrile HF AsahiChern. 6,000 14,900 3 kg/crn 2 , 25C
pH 1-14
0'1
-.I
VI
676
Approximate molecular dlometer, A
08
.: 06
>
c
Q)
~ 04
02
, Nominal molecular welCJht
exclusion of membrane
Molecular welCJht
Figure 12-14. Solute retention on Amicon Diaflo membranes (porter, 1979).
Reprinted with permission from P.A. Schweitzer (Ed.), Hand-
book of Separation Techniques for Chemical Engineers (1979).
Copyright 1979, McGraw-Hill.
ing depends on size and shape of the hydrated molecule, and molecular weight
is only a rough guide to size and shape. In addition, there is considerable reten-
tion of molecules of lower molecular weight. Roughly, a sharp cut-off UP
membrane will have R = 0 when the molecular weight is 1(2 the molecular
weight which has R = 0.9. Diffuse cutoff membranes have a much broader
range of fractionation. Typical retention data are shown in Figure 12-14
(porter, 1979). A comparison of Tables 12-2 and 12-3 is interesting. Note that
the UP membranes with their much larger pores have considerably larger fluxes
at lower pressures than the RO membranes. Details of UP membranes are dis-
cussed by Blatt (1976), Cooper(1979), Kesting (1985), Lloyd (1985), Porter
(1979), and Pusch and Walch (1982).
Concentration polarization, gel formation and fouling are extremely

important in UP and need to be included in any detailed analysis. We will first
look at the equations ignoring these effects, and then include them. Solvent
flux without concentration polarization can be written as Eq. (12-lOa or b).
677
For the large molecules separated by UP the molar concentration will be quite
low even if the weight concentration is high. Thus, the osmotic pressure differ-
ence ~1t will be small and is usually ignored. The resulting solvent flux equa-
tion is,
Kso1v
J I01v = - - ~p
(12-33)
tm
For small and intennediate macromolecules (5000 or 100,000 daltons) it may
be necessary to include the osmotic pressure tenn. This is discussed further in
Chapter 13.
For sieve type membranes flow of both solvent and solute is in pores.
Let R represent the fraction of solvent flux carried by pores which exclude the
solute. Then l-R of the solvent flux is in pores which do not exclude solute.
With no concentration polarization the liquid passing through these pores will
have a solute concentration Cb. This gives a solute flux of
(12-34)
In addition, Eq. (12-22) must remain valid since it represents a mass balance.
Combining Eqs. (12-22), (12-33) and (12-34) and solving for R, we obtain
Eq.(12-13). Thus in sieve type membranes the solute rejection can also be con-
sidered as the fraction of solvent flux in pores which reject solute molecules.
This analysis predicts that without concentration polarization the solvent

flux is proportional to ~p and is independent of the feed concentration. In addi-
tion, the rejection coefficient should be constant These results are a reasonable
first approximation, but are drastically oversimplified. R is usually constant for
very dilute systems, but may decrease for more concentrated systems. Some
pores may pass solute molecules slowly so that the concentration in the pores is
less than Cb, but the cut-off is not perfect.
Concentration polarization can cause an increase in solute concentration

as shown in Figure 12-4, or it can cause a gel or cake layer to fonn. Assum-
ing that there is no gel or cake layer fonned, the solute flux equation becomes
(12-35)
J I01uIe =MCb (1 - R) J I01v
678
y=8
Figure 12-15. Ultrafiltration system with gel layer at membrane wall.
where M is again the concentration polarization mooulus, Eq. (12-26). Solving

Eqs. (12-22) and (12-35) simultaneously, we obtain the apparent rejection
coefficient Rapp '
cp (12-36)
R
app
= 1- -
Cb
=1- M (1 - R)
Unless R=1 (all pores completely reject solute), concentration polarization

decreases the apparent rejection coefficient. When M=I, R.pp=R. This non-
gelling concentration polarization is predicted to have no effect on the solvent
flux since osmotic pressure is negligible.
Unfortunately, many of the .materials which the engineer desires to

ultrafilter (paint pigments, foods, polymers, minerals, and so forth) form gels,
cakes or slimes at the wall. This gel cake or slime can form at concentrations
cg below 1 wt% for polysaccharides to 50 vol% for suspensions of polymer lat-
tices (Blatt et al, 1970). The concentration profile for UP with gel formation is
shown in Figure 12-15. This gel layer acts as a porous solid which causes an
additional resistance to flow. The solvent flux is now,
l\p
J lalV = --.-!;....-- (12-37)
_tm_+_ls
Kaalv Kg
679
where tg is the thickness of the gel layer and Kg is the permeability of the gel
layer. Since 19 can vary throughout a run, Eq. (12-37) is usually not useful for
predictions. Often, the gel layer will have far more resistance to flow than the
membrane and the gel will control the solvent flux. In these cases it does not
matter what type of membrane is used as long as it retains the solute and causes
a gel to form. The value of Kg can be estimated from the Carman-Kozeny
equation (Blatt et ai, 1970).
(12-38)
Note that the gel permeability will be very sensitive to the particle diameter ~
in the gel layer. If e=O.5 and the viscosity is 1.0 cp, then Kg = 3xlO-ll for one
micron particles and Kg = 2x10- 16 for 30 Angstrom particles. Thus a one
micron layer of these gels with a pressure drop of 100 psi across the gel would
have solvent fluxes of 200 and 0.0013 cm 3 /cm 2 sec, respectively. Obviously, a
gel layer can have a major effect on the solvent flux.
A simplified steady state, one-dimensional mass balance is still given by

Eq. (12-31). Now the concentration at the wall (y =-tg) is Cw =c g • The boun-
dary conditions for the mass balance (see Figure 12-15) are at y = 0, C =c g and
at y =0, C =Cb. The solution to this is
(12-37)
Since cg and Cb are constants, the solvent flux is the variable. This is different
from Eq. (12-32) where Cw was unknown. The solvent flux is set by the rate at
which the solute can diffuse back from the gel layer to the bulk fluid. Increas-
ing the pressure drop will cause a brief temporary increase in the solvent flux.
However, during this period of increased flux transfer of solute to the gel layer
is greater than the rate of transfer of solute back into the fluid. The gel layer
must then become thicker. According to Eq. (12-37), J I01v will then decrease.
This decrease in J I01v continues until the gel layer is thick enough so that the
solvent flux has obtained the value predicted by Eq. (12-29). In other words,
680
the engineer has no control over the flux rate except by increasing the dif-
fusivity (increase temperature) or decreasing ~ (stir or use thin channels). For
high molecular weight polymers diffusivities are quite low and the solvent flux
is low if gels form. The mass transfer coefficient k can be estimated for many
flow geometries. This analysis is done in Chapter 13.
Example 12-3. Protein Ultrafiltration.
We wish to ultrafilter a 2 wt % aqueous solution of casein at 25°C. The

casein gels with cg = 0.14 .. We do two experiments as follows:
Experiment a: Run pure water and obtain Jsolv = 7000 L/m2 day with
~p= 5 atm.
Experiment b: Run casein solution at ~p = 5 atm and obtain J501v =

233 L/m 2day, R = 1.
a. Calculate the mass transfer coefficient k and Kg/tg for Experiment b.
b. If we increase ~p to 7 atm, determine the peak value of Jsolv , the
final value of J solv, and the final value of Kg/tg.
Solution
a. From the pure water data we can find Ksolv/1m from Eq. (12-33).
K.olv = J so1v = 7000 = 1400 L

1m ~p 5 m2 day atm
From the casein data and Eq. (12-39),
~. n3 L
k = In (Cg/Cb) = In (0.14/0.02) = 119.74 m2 day
Rearranging Eq. (12-37) and using the casein data,
-tg = -~p - -1m- = - 5 - -1- = 0.0207

Kg J so1v Ksolv 233 1400
681
Thus, Kg/tg =48.20 L/m2 day atm
b. If we increase AP to 7 atm, we will have a momentary increase in

water flux since tg does not increase immediately. Using Kg/tg found in
part a and Eq. (12-37) we have,
Jso1v = ___ 7_ _ _ = 326.2 L/m2 day

1 1
--+--
1400 48.2
Mter a short period of time, tg increases and J so1v is given by Eq. (12-
39). Since k, c g and Cb are unchanged from part a, J so1v = 233. Solving
for tg/kg
-tg = -Ap - -1m- = - 7 - -1- =0.0293

Kg Jsalv K.olv 233 1400
and Kg/tg = 34.096. Thus, tg has increased by 40.8%.
Note that in both parts the resistance of the gel, fglKg, is significantly
greater than the membrane resistance, tmlKm. This is commonly the
case when proteins are ultrafiltered.
Although simple and very appealing, there are significant problems with
the gel layer model (Fane, 1986; Porter, 1979; Wijmans et aI., 1984). The con-
centration cg found by extrapolation may depend upon equipment geometry.
The measured cg may be too low and independent experiments show no gel-
ling. In addition, c g may be estimated as > 100% which is physically impossi-
ble. The mass balance, Eq. (12-31) does not include convection in the axial, x,
direction and the boundary condition at y = 5 is hypothetical. However, the
model is useful for correlating data. Modifications of this model are discussed
in Chapter 13.
The presence of a gel layer may have conflicting effects on the rejection
of solutes. If the gel layer is very viscous, but remains fluid so that molecules
682
diffuse in the layer, solute molecules will continue to pass through the mem-
brane. Since Cw>Cb, the fluid passing through pores which do not reject solute
will be concentrated. This will cause the apparent rejection coefficient to drop.
This situation is similar to concentration polarization without a gel layer. The
second case occurs when the gel layer has a solid-like nature and solute
molecules are essentially immobile. Now the rejection coefficient can increase
since the gel layer traps the solute molecules and very few of them can diffuse
into the pores.
The fonnation of a gel layer also has a large effect on the separation of
mixtures containing several solutes of widely different size. Suppose we wish
to separate a low molecular weight compound from a very high molecular
weight polymer. When run separately, the low molecular weight compound
has R=O while the polymer has R=l. What happens when the mixture is
ultrafiltered? Our first guess is probably to use superposition and assume each
molecule is unaffected by the presence of the other molecule. If no gel layer is
fonned, this assumption is probably close to true. Then the small molecules
have R=O and the polymer has R=1. The separation is easy. If a gel layer
fonns, the gel serves as a membrane which has different retention characteris-
tics than the original membrane. The gel layer may retain the small molecules.
If this happens both solutes may have retentions near 1.0 and separation will
not be obtained. Thus gel layers change the fundamental nature of UF.
Once fonned, the gel layer may not come off easily. In this case the
membrane is said to be fouled. Fouling often occurs when the solute is
adsorbed to the membrane or when charges OR the membrane attract colloidal
particles of opposite charge. The entire UF installation is often controlled by
the presence of gel layers and fouling. Fouling can also occur from precipita-
tion of salts, and is discussed in detail by Fane (1986) and Potts et al. (1981).
In non-fouling or moderate fouling applications hollow fiber and spiral

wound configurations are commonly used. The hollow fibers have a 0.1
micron skin on the inside of the membrane. The fibers are from 500 to 1,100
microns inside diameter. Since pressures used in UF are much lower than in
RO the membranes do not have to withstand very high pressures. Thus the
high pressure liquid can be on the inside of the membrane. This has the advan-
683
tage of reducing concentration polarization since velocities are higher and

boundary layers thinner than for systems with the high pressure liquid on the
outside of the fibers. The systems can be cleaned by washing with high velo-
city clean water or with chemical solutions. In many applications a dilute caus-
tic wash is a very effective cleaning agent if the membrane can withstand the
high pH. Hollow fiber membranes can also be cleaned by back-washing with a
clean liquid. Back-washing forces liquid through the porous support and then
through the skin. Any particles attached to the skin will tend to be washed off
into the bulk fluid.
In highly fouling applications plate-and-frame and tubular membranes

are often used. In food plants the membranes are cleaned daily and sometimes
every few hours. Obviously, the cleaning cycle becomes very important in
controlling costs. Biotechnology applications of UP with an emphasis on pro-
tein recovery are discussed by Belter et al (1988).
All of the cascade arrangements shown in Figure 12-8 can be used.

When gel formation or fouling are severe, the cascade design as well as the
module design are controlled by the concentration polarization. Designs which
keep liquid velocities on the feed side of the membrane high are favored.
These are the parallel-series and recycle cascades shown in Figures 12-8 c and
d. In particular, the recycle system is often used since it is readily adaptable to
the batch operation shown in Figure 12-8g which is required when cleaning
must be frequent. Recycle systems are often cascaded in series to increase the
recovery of the permeate or in parallel to increase capacity.
The batch system shown in Figure 12-8g is easily analyzed for simple
cases. A mass balance on the entire system (tank + pipes + UP module) is
dV (12-40)
-=-AJ
dt 1
SOY
Where V(t) is the volume of liquid in the tank + pipes + UP module. If osmotic
pressure is negligible and no gel forms, Jso1v is given by Eq. (12-33). Then
dV A Ksolv .1p (12-41)

-=-----
dt t.n
684
The initial condition is V =V 0 for t =O. Assuming that ~p is kept constant, we

obtain the solution to Eq. (12-41).
(12-42)
If a gel forms then J. olv is a constant given by Eq. (12-39). Now Eq.
(12-40) becomes
dV (12-43)
- = -A k In(c ICb)
dt g
Since R = 1, the total moles of solute in the system, n, must remain constant
(12-44)
n = c(t) V(t) = constant
Then Eq. (12-43) becomes
dV (12-45)
""(it = - A k In (cg VIn)
The initial condition remains V = V 0 when t = O. The solution to Eq. (12-45) is
+ 1
(3)(3!)
[(In V)3 - (In V )3] + ...
o = -A kt In (cg/n)
(12-46)
which may be inconvenient to use since convergence may be slow.
If osmotic pressure is important, 1t is proportional to concentration, and

no gel forms; then Eq. (12-27) should be used for Jso1v with 1t ('1» = 0('1> = 0
since R = 1) (Belter et ai, 1988). Assuming 1t is linear with respect to concen-
tration, 1t =ac. Since R = 1, Eq. (12-44) remains valid. Then,
(12-47)
685
Substituting Eq. (12-47) into Eq. (12-40), we obtain
dV
dt
= _ A K.olv l\p
1m
[1 - [al\pMn 1~V 1 (12-48)
with the initial condition V = V0 when t = 0. The solution to Eq. (12-48) is (see
Problem 12-C4),
V _ V _ a Mn In aMn
vo - - -
l\p
[ V _ a Mn
1= _ A
K.olv
l\
P t (12-49)
o l\p 1m
l\p
Note that when a = 0, Eq. (12-49) reduces to Eq. (12-42). Equation (12-49) is
also valid for batch operation of RO when R = 1.
Ultrafiltration and reverse osmosis systems are often used together. For
example, cheese whey is the left over liquid from making cheese. The whey is
approximately 1% protein, 5% lactose (milk: sugar), 1% salts, and the
remainder is water. The whey has a very high Biological Oxygen Demand
(BOD) and is expensive to dispose of properly. One processing route which
utilizes all of the components of whey is shown in Figure 12-16. The protein is
recovered by UP. This protein concentrate is quite valuable and can either be
dried and sold or it can be added wet to cottage cheese or ice cream. The per-
meate from the UP system is sent to RO where the sugar and salts are concen-
trated by removing water. This is desirable since it makes both the downstream
processing steps smaller and cheaper. The concentrated solution can then be
fermented with yeast to produce a dilute ethanol solution. The ethanol is
recovered by distillation or other separation scheme. The waste liquor contain-
ing about 3% salts is sent to evaporators to recover the salts which can be used
as fertilizer. Note that the UP system is first. The UP system serves to protect
the RO system from particulates and protein which can cause fouling and gel
formation. Both UP and RO systems would be cleaned at least once a day.
Somewhat more complex combined UP and RO systems can be used for

686
Protein
Concentrate
F
r====t-- Permeate
(Water)
Storage
Fermentation
Tanks ''----95wt%
Ethanol
Disti Ilation
Water Vapor
Evaporators
L---=r- Stea m
Conc.
Salt
Solution
Figure 12-16. Simplified flow diagram for whey processing.
processing fruit juices (Stana, 1977). In concentration of fruit juices the pro-
cessor wants to recover all of the pulp, the sugar, and the volatile flavor
ingredients. The final product may be from 20 to 25% pulp and 20 to 25%
sugars. This product is very viscous and has a very high osmotic pressure.
(Open a can of frozen orange juice and let it melt. This is the desired final pro-
duct) A high recovery of sugar is desired; thus, the permeate leaving the RO
system should be close to zero concentration. A 50 wt% solution of sugar has a
very high osmotic pressure (see Table 12-1), and any leakage through the
membrane will cause significant contamination of the waste stream. Thus a
single RO module will be difficult to use and the membrane arrangement used
in Figure 12-16 for whey processing is not a good scheme. Instead, the pro-
cessing scheme shown in Figure 12-17 can be used (Stana, 1977). Note that
687
Concentrate
40-50%
Pulp
Raw
Juice
UF
Recycle 10-15% sugar
Figure 12-17. Scheme for processing fruit juices. (Stana, 1977).
the UF system is again placed first to protect the RO systems. Since the UF
membrane does not retain the sugar, the sugar concentration is the same in the
permeate and the high pressure products. Thus, although 7t is significant, d7t is
approximately zero and the usual UF equations can be used. The first RO sys-
tem produces the desired low concentration waste water, but the sugar solution
is concentrated to only 25%. This keeps M reasonable. The second RO sys-
tem produces the desired sugar solution, but the permeate is a 10 to 15% solu-
tion which is recycled to the first RO system. This keeps d7t modest in the
second RO system also. The pulp and sugar are then mixed to produce the
desired final product.
Other applications of UF include removal of colloidal matter in water

treatment, pigment recovery in painting operations, concentration of oil emul-
sions, depyrogenating pharmaceutical process water, and recovery of dyes.
Applegate (1984), Eykamp and Steen (1987), and Klinkowski (1978) discuss
these applications.
12.6. DIALYSIS
Dialysis is a process where small molecules diffuse through a membrane

because of a concentration driving force. Large molecules are excluded. The
process differs from UF since the solute movement in dialysis is caused by dif-
fusion while in UF the major flux is that of solvent caused by pressure gra-
dients. In UF the small molecules are carried along with the solvent flux. In
dialysis the flux of small molecules is independent of the solvent flux and may
688
be in the opposite direction from the solvent flux. Dialysis also usually does
Rot use asymmetric membranes, but instead uses homogeneous membranes
without skins.
The major commercial use of dialysis is hemodialysis for removal of
waste products such as urea and creatine from patients who have had kidney
failure. This application is commonly known as the artificial kidney even
though hemodialysis is not a good mimic of the way the kidney functions (Col-
ton and Lowrie, 1981; Ward et ai, 1985, Lysaght et ai, 1986). Other commer-
cial applications include the recovery of caustic in the manufacture of rayon,
salt removal in pharmaceuticals manufacturing, and recovery of spent acid in
the metal industry. Except in medical applications, the engineer is more likely
to be involved with RO, UP, or electrodialysis than with dialysis. Both plate-
and-frame and hollow fiber dialyzers are available (Lacey, 1979; Klein et ai,
1987).
As small solutes pass through the microporous membrane, concentration
gradients will develop on both sides of the membrane. This is illustrated in
Figure 12-18. The flux of the permeable solutes through the membrane can be
written as
Ksolute (12-50)
J solute = (c m h - cm '!)
lm .
where 'K.oIute is the permeability of the membrane to the nonexcluded solute.

As shown in Figure 12-18, the concentration driving force is the difference in
solute concentrations across the membrane. If both sides are very well mixed
the membrane concentrations will be the same as the bulk concentrations. In
most systems there will be a resistance to mass transfer on both sides of the
membrane and concentration profiles will build up. When this occurs, the
values of Cm.h and cm,! will not be known. An overall driving force is more
convenient The usual mass transfer expression is (Hwang and Kammermeyer,
1975)
(12-51)
where Cb and Cd are the bulk fluid concentrations on the feed and dialysate
sides, and k~lute is the total or overall mass transfer coefficient. If the solvent is
689
Flux of small
--l-+- molecules
- - - c b Small molecules
Dialysate
Side
Feed Side
- - - - - Macromolecules
Membrane
Figure 12-18. Concentration gradients in dialysis.
stagnant, the overall mass transfer coefficient can be found from a sum of resis-
tances.
1 1m 1 1 (12-52)
--- +--+--
k;olute - Ksolute k~olute k~olute
where k!olule and k~lute are the mass transfer coefficients on the feed and
dialysate sides of the membrane. The individual mass transfer coefficients can
be estimated from standard correlations such as those in Chapter 13. The three
terms on the right hand side of Eq. (12-52) are often the same order of magni-
tude.
As illustrated in Figure 12-18 there can be concentration polarization of

macromolecules if the solute flux is high enough. In some cases this may be
severe enough to form a gel which can then interfere or stop the transfer of
small molecules. The concentration polarization can be controlled by using
thin channels or turbulence.
Solvent flux (see Eq. (12-27» can be in either direction in Figure 12-18.
If the pressures on both sides of the membrane are equal, the solvent will flow
from the dialysate to the feed side because of osmosis. This osmotic flow can
be drastically increased by the concentration polarization of macromolecules,
and in turn will tend to reduce the concentration polarization. Unfortunately,
the osmotic flow also decreases the flux of the small solutes since the convec-
tive flow is in the opposite direction to the solute diffusive flux. The solvent
690
flow can be stopped or reversed by applying a positive pressure on the feed side
of the membrane. Dialyzers are often operated with essentially no solvent flux,
or with a small flux of solvent into the dialysate.
Any of the flow patterns shown in Figure 12-7 can be used in a dialyzer.
Hemodialyzers are usually countercurrent If the volumetric flow rates of
liquids on the feed <IF and dialysate sides Qd and the overall mass transfer
coefficient kT can be assumed to be constant, a variety of steady-state flow case
can be solved (Michaels, 1966). The correct average concentration difference
in Eq. (12-51) is the logarithmic mean of the inlet and outlet concentration.
(12-53)
where 1 and 2 are the two ends of the dialyses. For example, if countercurrent
flow is used,
(12-54a)
(12-54b)
Solving Eqs. (12-51), (12-53) and (12-54) with the mass balance
(12-55)
where NA is the amount transferred

(12-56)
we can obtain the result for a counter-current dialyszer.
(12-57)
691
In the derivation of Eq. (12-57) the volumetric flow rates Qp and Qd were
assumed to be constant. At equilibrium, cP,out = Cd,in' Thus the denominator on
the LHS of Eq. (12-57) is the maximum possible concentration change in the
feed stream while the numerator is the actual concentration change. The LHS
of Eq. (12-57) is thus the fraction of maximum solute removal which is
attained.
Results for other geometries are also easily obtained Michaels (1966).
For parallel flow of feed and dialysate the result is,
I-exp -~ [1+ ~1 (12-58)
=
1 + Qp/Qd
If the dialysate is completely mixed and the feed is in plug flow, the
result is
(12-59)
The counter-current system is superior. Dialysis is also used in batch processes

(Klein et ai, 1987).
Membranes used include cellulose, cellulose acetate, and polyacryloni-

trile, but cellulose has been most common. For hemodialysis (dialysis of
blood) isotropic cellulose hydrogels with a molecular weight cutoff of approxi-
mately 1000 are the most common membranes. The membranes are fonned by
the cuprammonium process and the membranes are homogeneous and do not
have a skin. These membranes easily pass the low molecular weight waste pro-
ducts such as urea (MW=60) and creatinine (MW=I13), but do not pass high
molecular weight proteins which are desirable. In addition, the membranes will
692
pass amino acids and salts which it is desirable to retain. This latter problem is
solved by using a dialysate fluid which is a physiological electrolyte solution
containing the appropriate concentrations of the salts. Since salt concentrations
are the same on both sides of the membrane, there will be no transfer of the
salts. In hemodialysis a positive pressure is applied to the feed (blood) side in
order to remove some water (which can be considered the solvent). Equipment
used in medical applications must satisfy a large number of constraints such as
preventing blood clotting. Details of hemodialysis are discussed by Colton and
Lowrie (1981), Lysaght et al (1986), and Ward et al (1985). Other membrane
separations used in artificial organs are discussed by Lysaght et al (1986).
In the usual dialysis process exclusion of large solutes is based on steric
effects. In ion exchange dialysis anion exchange membranes are used. The
membranes are the same as those used in electrodialysis (see the next section).
These membranes will exclude cations (ions with positive charge), but the
exclusion of H+ cations is poor (see Chapter 9 for a discussion of Ion
Exchange). Thus, if a solution of acid and metal salts is on one side of the
membrane and water or a dilute acid solution is on the other side of the mem-
brane, the anions (ions of negative charge) readily transfer through the mem-
brane. To keep electroneutrality cations must also transfer. Since the only
cation which can transfer through the membrane is H+, the acid is separated
from the metal ion. This is illustrated in Figure 12-19a. Ion exchange dialysis
is used commercially to recover metal ions in the plating industry. The systems
use plate-and-frame arrangements similar to those used in electrodialysis
except no current is required. Water and feed solutions are alternated in the
different compartments. Other geometries could also be used.
Donnan dialysis is another variant using ion exchange membranes. The
stack utilizes all cation or anion exchange membranes. For purposes of this
explanation we will assume that all of the membranes are cation exchange
membranes since our purpose is to concentrate a valuable cation. A dilute
solution containing the valuable cation (for example Cu++) is circulated in the
odd-numbered compartments. In the even-numbered compartments a concen-
trated solution of cheap acid is circulated. One cell is illustrated in Figure 12-
19b. The W ions will transfer through the membranes due to the concentration
driving force. To keep electroneutrality, either the anion must transfer through
693
a Dilute b dilute conc.
Acid
* +
! CuS0 4 acid
l
Metal
Salt
Anion conc.
+
l
dilute CUS04
H+
H2 S0 4
Metal
Salt Acid
t \ Cation
Exchange
\ ~ Membrane
* Anion
Exchange
Membrane
Figure 12-19. Special dialysis systems. a. Ion exchange dialysis. b. Donnan

dialysis.
the cation exchange membrane (which it cannot do because of Donnan

exclusion), or Cu++ must transfer through the membrane in a direction opposite
its concentration driving force. Thus the valuable cation (Cu++) is concentrated
at the price of diluting the acid. The copper can be concentrated from 30 ppm
Cu S04 to as high as 4000 ppm CUS04' The cost of the separation is the dilu-
tion of the sulfuric acid.
12.7. ELECTRODIALYSIS (ED)
Electrodialysis is dialysis in the presence of an electrical field. The driving

force for electrodialysis is the electrical potential difference. This driving force
pushes cations through cation exchange membranes and anions through anion
exchange membranes. The arrangement of membranes shown schematically in
Figure 12-20 is used. Cation and anion exchange membranes are alternated. In
694
Desalted solution
NaCI solution to be treated

A= anion - permeable membrane
C = cation - permeable membrane
Figure 12-20. Schematic of electrodialysis.
the presence of the electrical field the cations will migrate towards the cathode
and anions towards the anode. The cations can pass through the cation
exchange membranes but not through the anion exchange membranes. The
anions can pass through the anion exchange membranes but not through the
cation exchange membranes. Because of the alternation of the cation and anion
exchange membranes the result is to concentrate ions in the odd numbered
compartments and to dilute the ions in the even numbered compartments in
Figure 12-20. In this way a desalted water and a concentrated brine are
formed. In commercial systems from 100 to 600 unit cells (pairs of mem-
branes) are in a stack. Reviews of electrodialysis are given by Applegate
(1984), Hwang and Kammermeyer (1975), Klein et al (1985), Komgold
(1984), Lacey (1979), Mintz (1963), Solt (1976), and Spiegler (1984).
Ion exchange membranes can be thought of as ion exchange resins which

are formed as membranes. The same polymers used for ion exchange resins
(see Chapter 9) are used for some of the commercially available ion exchange
membranes. The most common cation exchange membrane has been a sul-
fonated polystyrene cross linked with divinylbenzene. The same basic polys-
tyrene backbone is used for anion exchange with a CH2~(CH3h group cou-
pled to the benzene ring. Other membranes such as Nafion (Figure 12-3g) are
695
used in special cases. Ion exchange membranes are discussed in detail else-
where (Kesting, 1985; Flett, 1983; Klein et al, 1987; Komgold, 1984; Lacey,
1979; Pusch and Walch, 1982; Solt, 1976; Strathmann, 1985). Lists of sup-
pliers and membrane properties are given by Komgold (1984), Spiegler (1984),
and Strathmann (1985).
The usual reaction at the cathode is
(12-60a)
The OIr fonned must be neutralized in the cathode compartment. The

cathode can be made from tantalum, niobuim, or titanium all of which must be
coated with platinum. In the anode the usual reaction is
(12-60b)
while if Cl- is present the reaction will be
(12-60c)
The anode can be made from the same materials as the cathode or from stain-
less steel. The anode and cathode compartments are separated from the stack
by membranes. These compartments are vented to remove the gases fonned.
Since the stack is in series, the current flow through each membrane and
each compartment must be the same. The reason for using a large number of
unit cells is to use this current many times and thus produce more desalted
water. The current is carried by the ions. The fraction of current carried by an
ionic species is the "transference number". This fraction can be different for
the different ions since current is the product of the charge times the velocity
times the number of ions transferred. Small ions such as W will carry more
current than large ions such as Cl- since the H+ have a higher velocity. In a
solution such as KCI where the ions are approximately the same size and thus
have the same velocity, the transference numbers will be approximately equal.
696
cations onions
I I
I Well Mixed 1
I Zone I
1 I
1 1 1 1
18 81 18 81
1 1
Ion Concentration
I I
I I
Cation Anion
Exchange Exchange
Membrane Membrane
Figure 12-21. Concentration polarization in electrodialysis.
In solution in the comparonents both ions carry current. In the mem-

branes one ion is excluded; thus, the other ion must carry all of the current.
The transference number of the cation in the cation exchange membrane is
essentially 1.0 while its transference number is essentially zero in the anion
exchange membrane. Suppose that K+ is carrying half the current in the com-
paronent. Suddenly, in the cation exchange membrane K+ must carry all of the
current. This requires that more K+ be transferred through the membrane than
is transferred in solution. Referring to Figure 12-21, we see that the cations are
diluted on the right hand side of the cation exchange membrane and concen-
trated on the left hand side of the membrane. Since concentration gradients
have been developed, there will be diffusion of the cation towards the cation
exchange membrane on the right hand side of this membrane (see Figure 12-
21). This diffusion provides part of the cations needed to carry the current
through the membrane. On the other side of the membrane the cations diffuse
away from the membrane. Since electroneutrality must be satisfied, the total
ion concentration will be diluted when the cation concentration is diluted and
concentrated when the cation is concentrated Exactly the same phenomena
occurs at the anion exchange membrane where the anion transfers all of the
current. This picture of concentration polarization is oversimplified since it
does not include osmotic flow of water and other effects (Hwan¥ and Kammer-
meyer, 1975).
697
Table 12-4. Basic electrical equations and units
V(volts) = I(amps) R(ohms) (12-61a)
I(amps) = Q(charge in coloumbs) /t(s) (l2-61b)
E(joules) = V(volts) I(amps) t(s) (l2-61c)
F = 96 500 coloumb = 96 500 amp sec (l2-6ld)

, equivalent ' equivalent
Concentration polarization has important ramifications for the operation

of electrodialysis. If one increases the current density (current per area), then
the rate of electrical transport also increases. This will require that the dif-
fusive transport must also increase. This requires an increase in the concentra-
tion gradient and the ionic concentrations at the membrane surface will
decrease. If these concentrations approach zero at the membrane, the ions will
be unable to carry all the current. This point is called the limiting current den-
sity. The resistance at the membrane surfaces will be high and energy require-
ments will increase. At this point water will be ionized and W and OH- will
transfer through the membranes to carry the current. This changes the pH in
the compartments, and represents energy which is not used to separate the salts
from the water. The changes in pH can cause precipitation of salts. The
transfer of OH- through the anion exchange membrane can cause dimensional
changes in the membrane. Because of these adverse effects, operation is con-
ducted at a current density below the limiting current density. This upper limit
on the current density puts a lower limit on the area of membranes which must
be used.
A brief review of electrical equations and units is given in Table 12-4.

The energy consumption in electrodialysis can be calculated by combining Eqs.
(12-61a) and (12-61c) for n cells.
(12-62)
E=J2nRt
698
where E is the energy consumption in joules, I is the current in amps, n is the

number of unit cells in the stack, R is the resistance per unit cell in ohms, and t
is the time in second. The resistance in the dilute and concentrated parts of the
cell differs. Thus, R is the average cell resistance. Alternately, we can calcu-
late E from,
(12-63)
if each chamber is completely mixed. The resistance is usually approximately

inversely proportional to concentration,
(12-64)
R=b/c
where b is a constant, and c is in equivalents/liter. After some algebraic mani-

pulation, we obtain
(12-65)
where R(CF) is the resistance per cell determined at the feed concentration. For
example, if we recover a product water which is one half as concentrated as the
feed this result is
The current requirements can be determined from the required concentra-

tion changes.
(12-66)
where F is Faraday's constant which is 96,500 coulombs/equivalent, Q is the

volumetric flow rate of solution in liters/sec, and ~ is the current utilization.
The concentration difference must be in gram equivalents per liter. The most
important term in this equation is the current utilization~. The current utiliza-
tion is directly related to the number of cells n in the stack and the efficiency of
utilization of the current.
(12-67)
~ = nll sllw 11m = n (Elec. Eff.)
699
where n is the number of unit cells, 11, is the efficiency due to the semipennea-
bility of the membrane (co-ions transfer through a membrane that they should
be excluded from), l1w is the efficiency related to water transfer through the
membrane, and 11m is the efficiency since some of the current invariably leaks
through the manifold holding the membranes. All of the efficiencies are less
than 1.0 and the overall electrical efficiency is often about 0.9. However, n
often varies from 100 to 600. Thus, ~ will usually be significantly greater than
1.0. An important exception to this occurs as the feed concentration becomes
high (3 to 5 mol/liter). As this happens 11, and hence Elec. Eff. approach zero
since the coions are not strongly excluded from the membrane. At these high
concentrations ~ becomes small and energy requirements become too large for
electrodialysis to be economical.
The membrane area A for the cation and anion membrane can be
estimated from
(12-68)
where i is the current density in amp/cm 2 • The membrane area A is related to

the area of each individual membrane Am by
(12-69)
A=nAm
The area Am is set by the dimensions of the plate-and-frame module which are
standardized by the manufacturer. Substituting in
(12-70)
plus Eqs. (12-67) and (12-69) into (12-68), we obtain
(12-71)
n=
1115 11m l1w
Mintz (1963) presents a more detailed analysis.

700
The major commercial application of electrodialysis has been for desalt-

ing brackish water (1000 to 5000 mgIL of total dissolved salt) to produce pot-
able water (less than 500 mgIL). ED is also used for producing salt from sea-
water in Japan, for desalting cheese whey, and for recovering metals from plat-
ing baths. In water treatment ED appears to be competitive with RO and ion
exchange in the range from 1000 to 3000 mgIL IDS. Typical ED systems
operate with current densities from 6 to 20 milliamp/cm 2 • The cell would typi-
cally be 40 cm wide by 100 cm long and 0.1 cm thick (Mintz, 1963). The sys-
tem would be designed for 50% demineralization. One major advantage of ED
is that significantly less pretreatment is usually required than will be required
for RO or ion exchange. The raw water does need to be filtered before the ED
system, and the water may need to be treated chemically.
Many ED plants use polarity reversal to reduce scaling caused by precip-

itation of salts. Polarity reversal involves periodically switching the polarity of
the electrodes so that the cathode becomes the anode and vice versa. This
switches the direction of ion transfer. The pmpose of this is to prevent scaling
by causing precipitated salts to dissolve when the concentration drops after the
direction of ion transfer is reversed. Unfortunately, when the polarity is
reversed both streams will have a high salt concentration and must be sent to
waste for a minute or two. In a commercial ED system using polarity reversal
approximately 13% of the production is lost. Thus the polarity reversal system
must be larger than a normal ED system. Since scaling can also be prevented
by adding acid, the economics of polarity reversal depends upon acid costs.
Since more than 500 of the 1000 commercial ED plants use polarity reversal, it
appears to be economical in many cases (Applegate, 1984).
Example 12-4. Electrodialysis of Brackish Water
a A laboratory electrodialysis unit is available to determine cell

resistance and electrical efficiency for desalting a NaCI solution.
The following experiment is done: Q = 30 L/min, n = 50, CF = 20
gIL, Cdil = 0.8 gIL. We measure I = 21.5 amp and E = 60,670
joules in one minute. Determine the resistance per unit cell at eF,
and the electrical efficiency.
701
h. Assuming that R and electrical efficiency are unchanged for a

commercial unit, design a system for Q = 300 L/min going from Cr
= 2.0 gIL to Cdil = 0.49 gIL if ~ = 0.4 m2 and i = 20 amp/m2.
Find I, n, and E in 1 minute.
Solution
a Rearranging Eqs. (12-66) and (12-67),
(EI Eff) = F Q (CF - Cdil)

ec.. NI
Since MWNaC1 = 58.45,
- g [ 1 equiv ]_ eq
CF - Cdil - (1.2 L) 58.45 g - 0.0205 L
96,500 am.: s 30 ':;n ][~i: ][0.0205 i ]

(E1ec. Eff.) = -"-____ .L.....>~_~~_~"___ _____"_
(50) (21.5 amps)
= 0.920
Assuming that cell resistance is inversely proportional to concentration
Eq. (12-65) is valid. Rearranging,
R(CF)~ t~n [2:F~ - [c: r]

R C _ 60,670 joules
( F) - (60 s) (21.5 amp)2 (50)
2(0.8) -
[ 2.0
[.2.!]2]
2.0
= 0.028 ohm/unit cell

702
Note that the units work since
joule = (volt) (amp) (s) = (amp)2 (ohm) (s)
b. We can use Eq. (12-70) to find I.
Then from Eq. (12-71),
[96,500 amp sec] 300 ~ 1 min] [2.0 - 0.49 e q ]

eq mm 60 sec 58.45 L
n=------------~----~~----~~--------~
(8 amp) (0.92)
where 118 11m 11n = elee. eff. = 0.92. n = 1693.6 or 1694.

From Eq. (12-65),
(8)2 (1694) (60) (0.028)

E= = 423,604 joules
2 (0.49) _ [0.49]2
2.0 2.0
Since there are 300 Ljmin, this is 1412 joules/L = 337.5 callL.
Notes: 1.) One of the hardest parts of ED calculations is the use of

unfamiliar units. Thus it is important to cheek units and
refer to Table 12-4.
2.) Equations (12-62) to (12-66) assume that electrode resis-

tance is small compared to nR. If n is small, this may not
be true. n = 50 is probably large enough.
3.) If the cell design and size or spacer design are changed,
R and Elec. Eff. may not be the same for laboratory and
commercial units. Also, if the high concentrations are
very different 115 may be different in the two units.
703
12.8. PERVAPORATION
Pervaporation is a membrane process where there is liquid on one side of the

membrane and a vapor on the other side. The liquid diffuses through the mem-
brane and vaporizes. This is illustrated in Figure 12-22. The partial pressure in
the vapor phase is usually reduced by pulling a vacuum or using a sweep gas.
Any of the modular geometries shown in Figure 12-5 could be used although
the hollow fiber and spiral wound configurations are most likely. Unlike the
other membrane processes discussed up to now, pervaporation is not currently
a well established separation method. However, the method has considerable
promise and commercial units are available (Sander and Soukup, 1988).
We usually want the least concentrated species to penneate preferentially

through the membrane. This reduces the required membrane area and the load
on the vacuum pump. For removal of trace organics rubbery polymers such as
polyether block amide (PEBAX) are used. For removal of water from azeo-
tropes such as the ethanol-water azeotrope either glassy polymers or ion
exchange membranes are used.
Theoretically, pervaporation is the most complex membrane separation
process. On the liquid side concentration polarization may be important partic-
ularly for the removal of trace organics from water. The membranes operate
by a solution-diffusion mechanism. Since the polymer is highly swollen on the
VP Vacuum
2
---p
Figure 12-22. Profiles for pervaporation.

704
liquid side and dry on the vapor side, the diffusivity varies markedly across the
membrane. The evaporation at the vapor side may have a major effect on the
observed selectivity. For the separation of trace organics from water, the rela-
tive volatility of the trace organic is often very high even though the boiling
point of the organic is higher than water's boiling point. This occurs because
these are very nonideal solutions with high activities. Finally, because of eva-
poration the process is nonisothermal. The temperature change affects solubil-
ity, diffusivity, and evaporation. Currently, a complete model including all
these effects is not available.
Pervaporation must be analyzed by using a concentration dependent dif-

fusivity. For the one dimensional system shown in Figure 12-22 the flux is
(12-72)
Since flux is proportional to dcJdx, the steady state form of Fick's second law
becomes,
(12-73)
Boundary conditions are
(12-74)
'1 = '1,1 at x=O , Ci = Ci.2 at x=tm
Concentration polarization on the liquid side may be important, and must be

estimated to calculate Ci,l' To solve this set of equations a functional form for
the diffusivity as a function of all solutes is required.
Unfortunately, solution of these coupled equations is very difficult and

has only been done analytically for a pure fluid (Li and Long, 1969; Hwang and
Kammermeyer, 1975). This has been done for an exponential dependence of
diffusivity on concentration which is appropriate for polymer membranes.
(12-75)
D(c) =Do exp (ac)
705
The parameters Do and a are functions of temperature and the properties of the
polymer.
Ifequilibrium follows a form of Henry's law, at the wall
(12-76)
where S* is the Henry's law constant and is a form of solubility. Then the flux
is
(12-77)
Note that Eq. (12-77) does not fit the standard form given in Eq. (12-1). If we
force fit pervaporation into the form of Eq. (12-1) using ~P as the driving force,
we obtain a "permeability" P of
J1m
P= - = -
Do
[exp(aS "Pl)-exp(aS "P2)]
(12-78)
~P Mp
This result shows that the permeability is a function of the partial pressures, in
addition to the diffusivity and the solubility. Since both solubilty and dif-
fusivity usually increase as the temperature increases, the permeability and per-
meation rate increase with temperature increases. The selectivity between dif-
ferent species may decrease as temperature increases, although for commercial
alcohol-water pervaporation there is little decrease in selectivity (Sander and
Soukup, 1988).
For multicomponent systems the logical first attempt is to assume super-

position. Then the partial pressures become
(12-79)
Pz = YiP
The flux and concentration profiles are assumed to be unchanged by the pres-
ence of the other diffusing components. Unfortunately, superposition is seldom
valid. The presence of a second component affects the permeation rate of the
706
30
25
.s
u
20
...."
c:
-2 15
"<-
"
0-
Il
til
o 10
o
Weight fraction of benzene in liquid
o = permeation 40° e • =equilibrium 40 0 e
t:l. = permeation 5Qoe • = equilibrium 50° C
o = permeation 60 0 e • = equilibrium 60 0 e
Figure 12-23. Separation factors for separation of benzene and isopropyl

alcohol mixtures through polyethylene by pervaporation. (Car-
ter and Jagannadhaswamy, 1964). Reprinted with permission
from Symposium on The Less Common Means of Separation,
Institution of Chemical Engineers, London, 1984.
first component. Analysis becomes significantly more complicated and is

beyond the scope of this book. The selectivity can be a complex function of
concentration and temperature. This is illustrated in Figure 12-23 (Carter and
Jagannadhaswamy, 1964). Note that selectivity can increase or decrease as
mole fraction changes and as temperature increases depending on the system.
The selectivity can be significantly higher than the vapor-liquid equilibrium
selectivity.
If selectivity and flux are known from laboratory data, we can design the
707
Figure 12-24. One pass pervaporation system.
pervaporation system without knowing the details of the diffusion. Design

does depend upon the details of the flow geometry. The simplest case, com-
pletely mixed liquid and vapor (Figure 12-7a), will be considered here. Hwang
and Kammermeyer (1975) can be consulted for the analysis for other flow
geometries.
A schematic of a pervaporation system is shown in Figure 12-24 for a

binary separation. Fin, FL and Fp are flow rates in moles/hr while z, X out and yp
arc mole fractions of the more permeable component. Defining the cut e as
0= Fp/Fin, the mass balances will be similar to Eqs. (12-5) to (12-7). The
result is
-
z-Oyp (12-80)
Xout - I-a
Since a is usually determined based on energy balance considerations (see Eqs.

(12-88) to (12-90b), Eq. (12-80) is one equation with two unknowns.
An additional equation can be obtained from selectivity data, a~
(12-81)
Note that this definition is similar to the definition of relative volatility in

vapor-liquid equilibrium although pervaporation is not an equilibrium system.
Solving for y, we obtain
(12-82)
708
Equations (12-81) and (12-82) apply to local values at the same location of the
membrane. For a perfectly mixed system
(12-83)
yP =Y , Xout = X
and Eq. (12-82) becomes
(12-84)
yP = 1 + (ClAB
, - 1) Xout
The global character of this expression for perfectly mixed systems greatly
simplifies the remaining analysis.
Equations (12-80) and (12-84) can be combined. Upon clearing fractions

this becomes
(12-85)
This equation can obviously be solved for yp with the quadratic formula (see
Example 12-5). Equation (12-85) is linear in z and 9; thus, if yp and 9 are
known
(12-86)
or if yp and z are known
Cl~ (Cl~ 1) z Yp - Yp
9= (12-87)
Z- -
, 2
(ClAB - 1) (Yp - yp)
These equations are obviously valid and easy to use if Cl~ is constant.
However, Figure 12-23 shows that Cl~ is a function of liquid mole fraction x.
The equations are still valid, but must be used in a trial-and-error fashion. For
709
example, if z and e are given we can proceed as follows:
1. Guess Xout.guess
2. Detennine (l~ (Xo.lt ) from data.
3. Calculate yp from the solution of Eq. (12-85).
4. Calculate Xout,calc from Eq. (12-80).
5. Check: Is Xout.guess = Xout,ca1c? If not, return to step 1.
The value of the cut is often controlled by the energy balance. For the system
shown in Figure 12-24 the energy balance is,
where A is the latent heat of vaporization. With thennal equilibrium in a com-

pletely mixed system, Tp = Tout. Since the reference temperature is arbitrary,
we can pick T ref = Tp = Tout. Then, the energy balance simplies to
(12-89)
which is
(12-90a)
or
(12-90b)
The high temperature, Tin, is limited by the stability of the membrane. The low
temperature, Tout = T p' is limited by the need to have a vapor on the penneate
side. Thus, if T p is decreased a very low pressure may be required. Since
latent heats are significantly greater than specific heats, the amount of energy
710
Recycle
Permeate Recycle
Product I Product 2
Figure 12-25. Pervaporation system coupled with distillation.
required to vaporize the permeate will not be available in the feed unless the
permeate rate is low. For removal of trace organics permeate rates will be low
and sufficient energy is usually available in the feed. For breaking azeotropes
conentrations are usually significantly higher and heat effects become
important. A recycle stream and/or interstage heaters (see Figure 12-25) are
used when necessary to provide sufficient sensible heat for the vaporization.
Additional heat input into the pervaporation unit is provided by the hot distil-
late streams. The value of e per pass is low, but e..r for the entire unit shown in
Figure 12-25 can be high. It is desireable to keep the temperature drop quite
modest (5 to lO°C) since fluxes are significantly higher at higher temperatures
(Sander and Soukup, 1988). Thus a large number of stages or high recycle rate
is desirable. The use of these equations for a one-pass system is illustrated in
Example 12-5. Recycle systems are left to Problem 12-C3.
What are the possible advantages of pervaporation as compared to other

membrane processes? In pervaporation the driving force .1p can be increased
by decreasing the downstream pressure by drawing a vacuum or using a sweep
gas. In liquid permeation similar increases in the driving force usually require
high upstream pressures. To use gas permeation the entire vapor stream must
be vaporized instead of just the permeate as in pervaporation. If the feed is
already vaporized, gas permeation will be a competing process. Pervaporation
711
often has much larger selectivities but lower fluxes than RO. For trace oganics
the high activity of the organic gives high relative volatilities and thus both
high selectivities and high fluxes. In addition, pervaporation will not be limited
by osmotic pressure.
Pervaporation has been extensively studied as a method of breaking

azeotropes, and the method is now available commercially for the ethanol-
water azeotrope (Sander and Soukup, 1988). Leeper (1986) has collected per-
vaporation data for separations of alcohols from water and from a variety of
organics. Some of this data is shown in Table 12-5. It is useful to extract as
much information as possible from this data. For example, the fluxes in Table
12-5 are considerably less than those in Tables 12-2 and 12-3 for RO and UP.
Fluxes and selectivities in Table 12-5 vary significantly. In general, high selec-
tivity membranes will have low fluxes and high flux membranes will have low
selectivities. The cellophane data shows that additives can have a major effect
on the selectivities. The acetone water data shows that downstream pressure
has little effect as long as a vapor is present. Although not shown in Table 12-
5, upstream pressure usually increases flux only slightly and has little effect on
selectivity. Flux usually increases and selectivity usually decreases as tempera-
ture increases although there are exceptions (eg. the cellophane data). Leeper
(1986) also presents RO data for the same separations. Reverse osmosis results
show higher fluxes and lower selectivities.
In breaking azeotropes pervaporation will usually be coupled with distil-

lation columns. One way of doing this is shown in Figure 12-25 (Goldblatt and
Gooding, 1986; Leeper, 1986; Sander and Soukup, 1988). Two or more stages
are used in series. The liquid is heated between the stages. Stages are shown
operating under a vacuum since this technique is more common than using a
sweep gas. Efficient permeate condensation is important since it greatly
reduces the load on the vacuum pump. The vaccum pump normally represents
the biggest energy cost. Refrigeration systems are used commercially to reduce
the load on the vacuum pump (Sander and Soukup, 1988). The permeate recy-
cle is required since permeate is unlikely to be pure. The method shown in Fig-
ure 12-25 could be used for separation of isopropyl alcohol and water. Optimi-
zation of the system shown in Figure 12-25 will be quite difficult.
-....J
N
Table 12-5. Pervaporation data (Leeper, 1986).
Separation Membrane Cone. A T, of P2, Selectivity Flux

Problem in feed (wt%) mmHg (lAB Ib/ft2/hr
Ethanol (B) Cellophane 50 77 0.94 0.21

from water (A) Cellophane 50 113 2.0 0.43
Cellophane 50(1) 77 24.6 0.08
Cellophane 5(Ji2) 77 0.63 0.12
Cellulose Acetate 50 68 0.1 2.0 0.22
Polyvinylaleohol 10-50 104 0.1 220-75
Polyvinylalcohol 50-90 104 0.1 -75
Polyacrylonitrite 9-94 68 0.1 16-35 0.0001
Polysulfone 50 68 0.1 332 0.001
Polysulfone, Asymmetric 50 68 0.1 3.0 0.025
Polysulfone, Composite 50 68 0.1 19.0 0.008
Aromatic polyamide 4.5 90-130 0.5 0.59-0.36 1-2
Aromatic polyamide 18.7 70-130 0.5 0.56-0.36 0.3-2
Silicone rubber 8-12 86-140 0.14 0.02-0.05
Silicone rubber 45-46 86-140 0.23-0.22 0.04-0.10
Silicone rubber 81-87 86-140 0.67-0.71 0.04-0.16
n-Butanol(B) Cellophane 10-24 140 20 10-15 0.2-0.4
from Water (A) Cellulose 2.5 acetate 85 140 20 45 0.2
Acetone (A) Polypropylene 45 230 10 2.5 0.094

from Water (B) Polypropy lene 45 230 49 2.8 0.094
Polypropylene 45 230 488 2.8 0.092
(1) Includes 2 wt% sodium citrate. (2) Includes 2 wt% benzoic acid.
-...J
W
714
Distillation column 1 is used to approach the azeotrope concentration

while column 2 is used to distill on the other side of the azeotrope. In some
cases the retentate stream may be of high enough purity that the second distilla-
tion column is not required. This is the case for breaking the ethanol water
azeotrope (Leeper, 1986; Sander and Soukup, 1988). The commercial ethanol
water pervaporator uses a composite poly (vinyl alcohol) membrane which has
a very high selectivity for water. The permeate recycle stream is still required
for this simplified ftowsheet Of course, there are alternatives for separating the
ethanol water azeotrope such as azeotropic distillation, extraction, adsorption,
and gas permeation. The choice will depend on the economics of the particular
case. Sander and Soukup (1988) state that capital costs were higher for perva-
poration than the use of azeotropic distillation, but utility costs were about 1/3
those of azeotropic distillation.
Example 12-5. Pervaporation
We wish to remove water from n-butanol using a cellulose 2.5 acetate

membrane. Feed 90 mole % n-butanol. Feed is at 60°C. Selectivity,
a~ = 43; Flux = 0.2Ib/ff hr. Maximum permissible temperature drop
is 30°C. Find the cut e and the permeate and liquid mole fractions. If
feed rate is 100 lb/hr, find the membrane area.
Data: A.B = 141.6 cal/g, Aw =9.72 kcal/gmole
cp,B (45°C) = 0.625 cal/gOC, Cpw (45°C) = 1.0 cal/gOC
MWB =74.12, MWw = 18.016
Solution
We can find the cut a from Eq. (12-90b). First, we need consistent
units.
For 90 mole % butanol, A. =(0.1) 9.72 + (0.9) 10.5 =10.4.

715
Since liquid heat capacities are close to linear, cp was estimated at

T.vg = [Tin + Toot J12 =(60 + 30)12 =45°C
c
p.B
= 0.625 caV °C [
g
cal
1000 kca1
1[74.12
gmole
g 1= 0.046 kcal
gmole °C
Then, Cp,F = (0.1) (0.0180) + (0.9) (0.046) = 0.0435.

Eq. (12-83b) gives,
[Tin - Tout J Cp.F _

(30°C) [0.0435 kal
gmoleOC
1
/.. - 10.4 kcal/gmole
= 0.125
Penneate mole fraction can be found by solving Eq. (12-85).
where a = (a~ - 1) e = (43 - 1) (0.125) = 5.25

b = - [1 - e + (a~ - 1) z + a~ e ] = - [0.875 + (42) (0.1) +
(43) (0.125)] =- 10.45
C = a~z =(43) (0.1) =4.3
Note that z = 0.1 refers to water mole fraction since a' = a~. Then yp
is also water mole fraction.
yp = 10.45 ± 4.35
10.5 0581 mo Ie fracllon
=. . water
where the minus sign is used since yp < 1.0.

716
From Eq. (12-80),
= z - e yP = 0.1 - (0.125)(0.581) = 0.031.

x w•out 1- e (1 - 0.125)
l00lb/hr
Area = Feed Rate/Flux = 2 = 500 ft2
0.2Ib/hr ft
Notes: 1. e is small despite the significant temperatme drop. Thus,

recycle is often required.
2. yp is increased significantly, but liquid product still contains

3% water. A cascade or recycle system is probably needed.
3. This separation can also be done by RO, distillation, adsorp-

tion and extraction. The choice will depend on economics.
4. In this example A should be calculated at X<xlt and at the

vacuum pressure. Both these corrections will be small.
There are two alternatives to pervaporation in addition to RO. One is to

do gas permeation with the vapor stream. For reasons that are not completely
understood, this method gives selectivities and fluxes which are significantly
less than those observed for pervaporation. The second alternative is perstrac-
tion which is liquid permeation into a carrier liquid on the permeate side. Per-
vaporation has been more extensively studied than perstraction since recovery
of permeate is normally easier from a vapor than from a liquid. In perstraction
the permeate is usually recovered from the high boiling carrier liquid by distil-
lation. Perstraction may be advantageous in some cases since a vacuum pump
is not required.
12.9. LIQUID MEMBRANES
A liquid membrane has a layer of liquid which serves as the separation medium
instead of the solid polymer used in the other membrane methods discussed
717
a b ~ Feed Phase
o 00 0
Feed Phase o 00 0 Recelvlng
..
0 0 00 0
o 0 0 Phase
0
Liquid 0
Membrane
o
Liquid Exploded
View
Membrane
Receiving Phase
Figure 12-26. Liquid membrane systems. a Supported liquid membranes. b.

Emulsion liquid membranes. c. Facilitated transport.
previously. Two types of liquid membrane systems have been extensively stu-
died. In supported liquid membranes the liquid is held in a porous matrix
which serves to support the liquid (Danesi, 1984-85; Noble et ai, 1986; Ward,
1970; Noble and Way, 1987; Way et ai, 1982, 1985). The equipment and
operating procedures for supported liquid membranes are very similar to those
for the other membrane separations discussed in this chapter. In emulsion
liquid membranes a double emulsion is formed with the receiving liquid encap-
sulated by an immiscible material which is also immiscible with the outer fluid
(Cahn and Li, 1976; Li, 1971; Marr and Kopp, 1982; Noble et ai, 1986; Way et
ai, 1982). The emulsion liquid membrane systems are operated in a manner
very similar to liquid-liquid extraction. Liquid membrane systems have been
an area of considerable research, but only waste water treatment and gas sen-
sors for ion selective electrodes are commercial (Noble and Way, 1987). In the
future these separation techniques may become important as commercial
separation devices.
Supported liquid membranes are shown schematically in Figure 12-26a.

The liquid membrane phase is held in a porous matrix which is usually a poly-
mer. The liquid membrane itself must be chosen to be immiscible with both
718
the feed phase and the receiving phase. The feed and receiving phases can be
essentially the same liquid and in many cases will be an aqueous solution. The
liquid membrane serves to separate the feed and receiving phases. In order to
increase the capacity of the receiving phase, it will often contain a chemical
which will react with the diffusing solute. The supported liquid membrane sys-
tems can use one of the geometries shown in Figure 12-5. One of the major
problems with this type of liquid membrane is bleeding which is loss of the
liquid membrane as it dissolves into the feed and receiving phases. Supported
liquid membrane systems have apparently not been commercialized yet.
The emulsion liquid membrane system is shown schematically in Figure
12-26b. The system is a double emulsion with the receiving phase distributed
within the liquid membrane drops which are dispersed within the feed phase.
The emulsion type systems are fairly simple to make and have a very large sur-
face area. Normal extraction equipment is used for the contacting. Mter the
extraction, the liquid membrane usually must be broken to recover the solute.
The receiving phase usually contains a chemical to react with the solute to
increase the capacity. Emulsion type systems have been extensively designed
and piloted for a variety of commercial processes such as copper recovery,
phenol removal from waste water, and hydrocarbon separations.
Both types of liquid membrane systems can use facilitated transport
(Cussler, 1971; Goddard, 1977; Noble et ai, 1989). Facilitated transport is
shown schematically in Figure 12-26c. A carrier contained within the liquid
membrane reacts with the solute and then diffuses across the membrane. In the
receiving phase the carrier-solute complex is broken and carrier diffuses into
the receiving phase. The carrier then diffuses back across the membrane to
pick up another "load". This mechanism is similar to the carrying of oxygen in
blood by hemoglobin. In the second mechanism shown in Figure 12-26c a
second molecule of A is back-transported across the membrane. The back-
transported molecule supplies the energy for the process (Cussler, 1971).
Examples of this process are using W to supply energy to separate cations.
Facilitated transport is of interest since it can provide much faster mass transfer
rates with higher selectivities than the liquid membrane alone, and because
facilitated transport appears to be important in transfer across cell membranes
in living systems. In terms of a solution-diffusion membrane, facilitated tran-
719
sport involves increasing the solubility and/or the diffusion rate of the solute.
Noble et al (1989) list a large number of systems which have been developed
for facilitated transport. Note that these facilitated transport chemical systems
can often be used to advantage in other separation processes such as absorption
or extraction, and these alternatives may be cheaper than liquid membranes.
1. Explain and describe the following membrane separation

processes: Gas permeation, reverse osmosis, ultrafiltration,
dialysis, electrodialysis, pervaporation, and liquid membranes.
Clearly delineate between the different processes.
2. Determine the flux for the different membrane separation tech-

niques.
3. Describe concentration polarization, explain what can be done to

reduce it, and estimate the effect of concentration polarization on
flux and rejection coefficients.
4. Describe gelling and fouling, and estimate the effects on flux and
rejection. Explain why flux is determined by back-diffusion when
a gel layer exists.
5. Explain the effects of flow geometry qualitatively. Do this quanti-

tatively for perfectly mixed systems.
REFERENCES
Applegate, L.E., "Membrane separation processes," Chern. Eng., 91 (12) 164

(June 11, 1984).
720
Baker, R.W. and I. Blume, "Permselective membranes separate gases," Chem-

tech. 16. 232 (1986).
Beaver, E.R., P.V. Bhat, and D.S. Sarcia, "Integration of membranes with other
air separations technologies," AIChE Symp. Ser .• 84 (261), 113 (1988).
Belfort, G. (Ed), Synthetic Membrane Processes. Fundamentals and Water

Applications, Academic Press, Orlando, FL, 1984.
Belter, P.A., E.L. Cussler, and W.-S. Hu, Bioseparations. Downstream Pro-
cessingfor Biotechnology, Wiley-Intersciences, 1988, Chapt 9.
Blaisdell, C.T. and K. Kammermeyer, "Countercurrent and co-current gas

separation," Chem. Eng. Sci., 28. 1249 (1973).
Blatt, W.F., A. Dravid, A.S. Michaels, and L. Nelson, "Solute polarizatio,' and
cake formation in membrane ultrafiltration: Causes, consequences and COl trol
techniques," in J.E. Flinn (Ed.), Membrane Science and Technology, Plenum
Press, NY, 1970,47-97.
Blatt, W.F., "Principles and practice of ultrafiltration," in P. Meares (Ed),

Membrane Separation Processes, Elsevier, Amsterdam, NY, 1976, Chapt. 3.
Cadotte, J., R. Forester, M. Kim, R. Petersen, and T. Stocker, "Nanofiltration

membranes broaden the use of membrane separation technology," Desalina-
tion, 70, 77 (1988).
Cabn, R.P. and N.N. Li, "Hydrocarbon separation by liquid membrane

processes," In P. Meares (Ed.), Membrane Separation Processes, Elsevier,
Amsterdam, 1976, Chapt.9.
Caracciolo, V.P., N.W. Rosenblatt and V J. Tomsic, "Du Pont's hollow fiber
membranes," in S. Sourirajan (Ed.), Reverse Osmosis and Synthetic Mem-
branes, Ottawa, Canada, National Research Council of Canada, 1977, Chapt.
16.
721
Carter, J.W. and B. Jagannadhaswamy, "Liquid separations using polymer

films," Proceedings Symposium on the Less Common Means of Separation,
Institution Chemical Engineers, London, 1964,35-42.
Cheremisinoff, P.N. and A.R. LaMendola, "Industrial applications of reverse

osmosis," in N.P. Cheremisinoff and D.S. Azbel (Eds.), Liquid Filtration, Ann
Arbor Science, 1983, Chapt. 14.
Chern, R.T., WJ. Koros, H.B. Hopfenberg, and V.T. Stannett, "Material selec-
tion for membrane-based gas separations," in D.R. Lloyd (Ed.), Materials Sci-
ence of Synthetic Membranes, Am. Chern. Soc., Washington, DC, 1985, Chapt.
2.
Colton, c.K. and E.G. Lowrie, "Hemodialysis: Physical principles and techni-
cal considerations," The Kidney, Vol. 2, 2nd ed., Philadelphia, W.B. Saunders
Co. 1981,2425-2489.
Cooper, A.R. (Ed.), Ultrafiltration Membranes and Applications, Plenum Press,

NY,1979.
Cussler, EL., "Membranes which pump," AIChE Journal, 17, 1300 (1971).
Danesi, P.R., "Separation of metal species by supported liquid membranes,

Separ. Sci. Technol., 19, 857 (1984-85).
Eisenberg, T.N. and EJ. Middlebrooks, Reverse Osmosis Treatment of Drink-

ing Water, Butterworths, Boston, 1986.
Eykamp, W. and J. Steen, "Ultrafiltration and reverse osmosis," in R.W.

Rousseau (Ed.), Handbook of Separation Process Technology, Wiley-
Interscience, New York, 1987, Chapt 18.
Fane, A.G., "Ultrafiltration: Factors influencing flux and rejection", in RJ.

Wakeman (Ed.), Progress in Filtration and Separation 4, Elsevier, NY,
101-180 (1986).
722
Flett, D.S. (Ed), Ion Exchange Membranes, Ellis Horwood Ltd., Chichester,
England. 1983.
Finken, H., "Asymmetric membranes for gas separations," in D.R. Lloyd (Ed.),
Materials Science of Synthetic Membranes, Amer. Chern. Soc., Washington,
D.C., 1985, Chapt. 11.
Gienger, J.K. and R.J. Ray, "Membrane-based hybrid processes," AIChE Symp.
Ser.,84 (261), 168 (1988).
Goldblatt, M.E. and C.E. Gooding, "An engineering analysis of membrane-

aided distillation",AIChE Symposium Series, 82 (248) 51 (1986).
Goddard, J.D., "Further applications of carrier-mediated transport theory - A

survey," Chern. Eng. Sci., 32,795 (1977).
Henis, J. M.S. and M.K. Tripodi, "The developing technology of gas separating
membranes," Science, 220 (4592), 11 (April 1, 1983).
Hsieh, H.P., "Inorganic membranes," AIChE Symp. Ser., 84 (261), 1 (1988).
Hwang, S.-T. and K. Kammermeyer, Membranes in Separations, Wiley, NY,

1975.
Kesting, R.E., Synthetic Polymeric Membranes. A Structural Perspective, 2nd

ed., Wiley, NY, 1985.
Klein, E., R.A. Ward, and R.E. Lacey, "Membrane processes-Dialysis and elec-
trodialysis", in R.W. Rousseau (Ed.), Handbook of Separation Process Tech-
nology, Wiley, New York, 1987, Chapt. 21.
Klinkowski, P.R., "Ultrafiltration: an emerging unit-operation," Chem. Eng.,

85 (11) 165 (May 8, 1978).
Komgold, E., "Electrodialysis - Membranes and mass transpon", in G. Belfon

723
(Ed.), Synthetic Membrane Processes. Fundamentals and Water Applications,

Academic Press, Orlando, FL 1984, Chapt. 6.
Korns, W J. and RT. Chern, "Separation of gaseous mixtures using polymer

membranes," in RW. Rousseau (Ed.), Handbook of Separation Process Tech-
nology, Wiley-Interscience, NEw York, 1987, Chapl20.
Kremen, S.S., "Membrane systems", in M.B. Chenoweth (Ed), Synthetic Mem-

branes, MMI Press, Midland, MI, 1986,53-87.
Kremen, S.S., "Technology and engineering of ROGA spiral-wound reverse

osmosis membrane modules," in S. Sourirajan (Ed.), Reverse Osmosis and Syn-
thetic Membranes, Ottawa, Canada, National Research Council of Canada,
1977, Chapt. 17.
Lacey, RE., "Dialysis and electrodialysis," in P.A. Schweitzer (Ed.), Hand-

book of Separation Techniques for Chemical Engineers, McGraw-Hill, NY,
1979, Section 1.14.
Leeper, S.A., "Membrane separation in the production of alcohol fuels by fer-

mentation," in W.C. McGregor (Ed.), Membrane Separations in Biotechnology,
Marcel Dekker, NY, 1986, Chapt. 8.
Li, N.N. and RB. Long, "Permeation through plastic films," AIChE Journal,
15, 73 (1969).
Li, N.N., "Separation of hydrocarbons by liquid membrane permeation," Ind.

Eng. Chern. Process Des. Develop., 10, 215 (1971).
Lloyd, D.R (Ed.), Materials Science of Synthetic Membranes, Amer. Chern.

Soc.," DC, 1985.
Lloyd, D.R and T.B. Meluch, "Selection and evaluation of membrane materi-
als for liquid separations," in D.R. Lloyd (Ed.), Materials Science of Synthetic
Membranes, Amer. Chern. Soc., Washington, DC, 1985, Chapt. 3.
724
Loeb, S. and S. Sourirajan, "Seawater demineralization by means of an osmotic

membrane," Advan. Chem. Ser., 38,117 (1963).
Loeb, S. and S. Souriraj an , "Seawater demineralization by means of a semi-

permeable membrane," University of California at Los Angeles Engineering
Report No. 60-60, 1960.
Lonsdale, H.K., "The growth of membranes technology," J. Membrane Sci., 10,

81 (1982).
Lysaght, MJ., D.R. Boggs, and M.H. Taim isto , "Membranes in artificial
organs", in M.B. Chenoweth (Ed), Synthetic Membranes, MMI Press, Mid-
land, MI, 1986,100-117.
Marr, R and A. Kopp, "Liquid membrane technology - A survey of

phenomena, mechanisms and models," Internat. Chem. Eng., 22 (1),44 (1982).
Michaels, A.S., "Operating parameters and performance criteria for hemodi-

alyzers and other membrane-separation devices", Trans. Amer. Artij. Intern.
Organs, 12,387 (1966).
Mintz, M.S., "Electrodialysis, principles of process design," Ind. Chern., 55(6)

18 (June 1963).
Noble, RD., C.A. Koval, and JJ. Pellegrino, "Facilitated transport membrane
systems," Chern. Engr. Prog., 85 (3), 58 (March 1989).
Noble, R.D. and J.D. Way (Eds.), Liquid Membranes, Theory and Applications,
ACS Symp. Ser., No. 347, ACS, Washington, D.C., 1987.
Noble, RD., J.D. Way and A.L. Bunge, "Liquid membranes," in Y. Marcus
(Ed.), Ion Exchange and Solvent Extractions, vol. 10, Marcel Dekker, NY.
Orofino, T.A., "Technology of hollow fiber reverse osmosis systems," in S.

725
Sourirajan (Ed.), Reverse Osmosis and Synthetic Membranes, Ottawa, Canada,

National Research Council of Canada, 1977, Chapt 15.
Porter, M.C., "Membrane filtration," in Schweitzer, P.A. (Ed.), Handbook 0/

Separation Techniques/or Chemical Engineers, McGraw-Hill, NY, 1979, Sec-
tion 2.1.
Potts, D.E., R.C. Ahlert and S.S. Wang, "A critical review of fouling of reverse
osmosis membranes," Desalination, 36,235 (1981).
Pusch, W. and A. Walch, "Synthetic membranes-preparation, structure, and

applications," Angew Chern. Int. Ed. Engl., 21,660 (1982).
Reid, C.E., "Principles of reverse osmosis," in U. Merten (Ed.), Desalination by

Reverse Osmosis, MIT Press, Cambridge, MA 1966, Chapt 1.
Rogers, C.E., "Permeation of gases and vapours in polymers," in J. Comyn

(Ed.), Polymer Permeability, Elsevier, London, 1985, Chapt. 2.
Sander, U. and P. Soukup, "Design and operation of a pervaporation plant for

ethanol dehydration," J. Memb. Sci., 36,463 (1988).
Schell, W J., "Commercial applications for gas permeation membrane sys-

tems," J. Memb. Sci., 22, 217 (1985).
Solt, G.S., "Electrodialysis," in P. Meares (Ed.), Membrane Separation

Processes, Elsevier, Amsterdam, NY, 1976, Chapt. 6.
Soltanieh, M. and W N. Gill, "Review of reverse osmosis membranes and tran-

sport models," Chern. Eng. Commun., 12, 279 (1981).
Sourirajan, S. (Ed.), Reverse Osmosis and Synthetic Membranes. Theory-

Technology - Engineering, National Research Council of Canada, Ottawa,
Canada, 1977.
726
Spiegler, K.S., "Electrodialysis", in RH. Perry and D. Green (Eds.), Perry's

Chemical Engineer's Handbook, 6th ed., McGraw-Hill, N.Y., 1984, 17-37 to
17-47.
Spillman, RW., "Economics of gas separation membranes", Chern. Eng.

Prog .. 85(1), 41 (Jan. 1989).
Stana, RR, "Westinghouse membrane systems," in S. Sourirajan (Ed.),

Reverse Osmosis and Synthetic Membranes. Ottawa, Canada, National
Research Council of Canada, 1977, Chapt 18.
Stannett, V.T., W J. Koros, D.R. Paul, H.K. Lonsdale, and RW. Baker,
"Recent advances in membrane science and technology," Adv. Polymer Sci ..
32.69 (1979).
Stem, S.A., "New developments in membrane processes for gas separations",

in M.B. Chenoweth (Ed.) , Synthetic Membranes, MMI Press, Midland, MI,
1986, 1-37.
Stem, S.A. and H.L. Frisch, "The selective permeation of gases through poly-
mers," Ann. Rev. Mater. Sci., 11, 523 (1981).
Stem, S.A., "The separation of gases by selective permeation," in P. Meares

(Ed.), Membrane Separation Processes. Elsevier, NY, 1976, Chapt. 8.
Strathmann, H., "Membrane separation processes," J. Memb. Sci .• 9, 121

(1981).
Strathmann, H., "Electrodialysis and its application in the chemical process

industry," Separ. Purific. Methods, 14, 41 (1985).
Warashina, T., Y. Hashino and T. Kobayashi, "Hollow-fiber ultrafiltration,"

Chern. Tech .• 15, 558 (1985).
Ward, R.A., P.W. Feldhoff and E. Klein, "Membrane materials for therapeutic
727
applications," in D.R. Lloyd (Ed), Materials Science of Synthetic Membranes,

Amer. Chern. Soc., Washington, DC, 1985, Chapt. 5.
Ward, WJ., "Analytical and experimental studies of facilitated transport,"

AIChE Journal, 16,405 (1970).
Way, J.D., RD. Noble and B.R Bateman, "Selection of supports for immobil-
ized liquid membranes," in Lloyd, D.R (Ed.), Materials Science of Synthetic
Membranes, ACS, Washington, DC, 1985, Chapt 6.
Way, J.D., RD. Noble, T.M. Flynn and E.D. Sloan, "Liquid membrane tran-
sport: A survey," J. Memb. Sci., 12, 247 (1982).
Wijmans, J.O., S. Nakao, and c.A. Smolders, "Flux limitation in ultrafiltration:

Osmotic pressure model and gel layer model," J. Memb. Sci., 20, 115 (1984).
HOMEWORK
AI. How does polymer structure affect selectivity?
A2. What are the advantages and disadvantages of each module shown in
Figure 12-5?
A3. In pervaporation the permeate side is usually under a vacuum. Explain

why. Would pulling a vacuum on the permeate side of a gas permea-
tion system be useful? Explain why.
A4. In gas permeation too high a selectivity can be detrimental and may
actually limit the removal of the permeating gas from the feed. Explain
this phenomenon.
A5. How can RO and UF complement each other?
A6. Contrast concentration polarization without a gel to concentration polar-

ization with a gel.
728
A7. Contrast how concentration polarization develops in RO and in ED.
A8. Figures 12-4 and 12-15 show no concentration polarization on the per-
meate side. Figures 12-18 and 12-21 show mass transfer on both sides
of the membrane. Explain these differences.
A9. Why is ED more economical for separation of relatively dilute solutions

than for concentrated solutions?
AlO. Sketch a pervaporation system to break the ethanol-water azeotype

when column 2 in Figure 12-25 is not required. Assume water is the
preferred permeate. Why might a recycle stream be required for the
permeator?
A 11. Develop your key relations chart for this chapter.
B 1. The RM membrane is a clever and commercially successful solution to
the problem of how to make essentially defect free gas permeation
membranes. Brainstorm at least 5 other solutions to this problem.
C. Derivations
Cl. Show that Eq. (12-32a) is a solution to Eq. (12-31) and the appropriate
boundary conditions.
C2. Show that Eq. (12-39) is a solution to Eq. (12-31) and the appropriate
boundary conditions.
C3. A simple pervaporation unit with a recycle stream is shown below.

Assume that the following variables are known: Fnew , Xnew , T new , R, aT,
TR , Cp,in, cp,new, cp,R, a', A, Tout. The module is perfectly mixed and
operates at thermal equilibrium.
a. Calculate a
b. Assuming a is known, solve for yp in terms of known variables.

729
C4. Derive Eq. (12-49) from Eq. (12-48).
C5. Derive Eq. (12-46) from Eq. (12-45).
D. Single-Answer Problems
Dl. Estimate the permeabilities for Hz and CO of a RM membrane for gas

permeation if both membranes are perfect The top layer is 1J.U1l sil-
icone rubber. The second layer is polysulfone with a 1.2J.U1l thin skin.
Estimate aH..co. Data are in the text How will surface pores affect a?
D2. Data for a hollow fiber RO module from Dow Chemical is given in
Table 12-2. Details of how the test were done are not complete.
a. If there was no concentration polarization, calculate Ksolv/tM for this
system.
b. IfM = 2 during the experiment, determine K.olv/tM'
You can use the van't Hoff equation.
D3. An ultrafiltration membrane is first tested in a stirred cell where there is

no concentration polarization. The experimental values obtained in the
table are obtained. Then the same membrane is used in a spiral wound
module where there is concentration polarization. Values listed are
obtained. Assume membrane thickness and permeabilities are the same.
Calculate the expected solvent flux and the concentration polarization
modulus in the spiral wound membrane system. No gel layer forms.
Cb(conc. cp(conc. Jso1v•

solute) solute) L/mz day dP, bars
Stirred Cell 10 gIL 0.25 gIL 7000 3.5
Spiral Wound 6g/L 1.0 gIL 4.0
730
04. Ultrafiltration of a dextran solution gives the following data
J, mVcm2 min 0.055 0.045 0.030 0.018

wt. frac dextran 0.01 0.02 0.05 0.10
Estimate the weight fraction at which dextran gels.
D5. A countercurrent dialyzer is using an experimental membrane to

remove urea from an aqueous feed. The entering feed contains lOgIL
urea and outlet should be 19/1iter. Feed rate is 1 L/min. Inlet dialysate
is pure water and flow rate is 3 L/min.
a. Determine the overall transfer rate NA in g/min.
b. Determine required value of kT A.
c. If A = 100 m2 calculate kT.

D6. We wish to desalinate 400 liters/min of a brad"ish water containing 1.5
gIL salt to 0.5 gIL salt in an ED system. The system has 200 cells. The
efficiencies are estimated as: TI. = 0.98, Tlw = 0.99, TIM = 0.99. The
average resistance of each cell is 0.03 ohms. Determine,
a. The current in amps and the voltage in volts.
b. The energy required in cal/liter water.
c. The power in kilowatts.
D7. An experimental RO cellulose-acetate membrane with an active layer 6

microns thick achieved a flux of 0.33 x 10-4 g/cm 2 sec for a 40 atmo-
sphere pressure difference with 0.1 N NaCl in water (n=4.9 atm). What
was the water permeability and list the units if:
b) This membrane had RO = 0.96. What was the salt permeability for
731
D8. We are ultrafiltering a flexible chain polymer. In a stirred cell with no

concentration polarization RO = 0.996 and K.ruvltm = 2000 L/(m2 day
atm). In a hollow fiber system gelling occurs at cg = 80 gIL. The gel
layer appears to be completely mobile. In an experiment where a gel
layer formed with Cb = 26 gIL and .1p = 6 atm we find Jso1v = 82.83
L/m2 day.
a In the stirred cell find Jso1v and Jsolule if .1p = 5 atm and Cb =20g/L.
b. In the hollow fiber system find J so1v , RaP!' and J so1uIe if .1p = 5 atm
and Cb = 2Og/L. Other conditions are the same as in previous
experiment
D9. In Example 12-1 we obtained Yout = 0.0168. Suppose we want Yout =

0.010. What value of e is required? Other specifications are the same
as in Example 12-1.
D1O. A cellulose acetate membrane is separating a CO2-N2 mixture. The

feed gas is 40 mole % N2 and 60 mole % CO 2, The permeate pressure
is 770 cm Hg while the feed pressure is 10,000 cm Hg. The effective
membrane thickness is 1.5 J..U11. The membrane module can be assum~
to be completely mixed on both sides. If the cut e = FpIFin = 0.55,
determine Yout and yp' Assume permeate is an ideal gas. Permeability
data from Figure 12-9 gives PN, - 0.3 x 10-10 and Peo, - 20 x 10-10
Dl1. We are separating oxygen and nitrogen by gas permeation with a sil-
icone rubber membrane. The membrane module is perfectly mixed.
°
Inlet gas is 21 mole % 02' We desire a permeate product which is 27
mole % 2 , Membrane has a selectivity (lAB =P AIPB = 2.1. Pressure
ratio is PL/PH = 0.35. Treat the gases as ideal gases. What cut e must
be used?
D12. Blood plasma proteins are known to gel at a concentration of cg = 0.2

weight fraction. The proteins can be treated as particles approximately
30 x 10-10 meters diameter. With pure water the membrane has a flux
of 2.5 x 10-5 m 3 /m 2 sec at a transmembrane pressure drop of 68.96 kPa.
At an operating pressure drop of 103.44 kPa with a plasma solution the
732
flux is 0.416 x 10-5 m3 /m 2 sec. Assume the gel layer has a porosity of E
=0.5.
a) What is the gel thickness at steady state?
b) If pressure is doubled, what is the gel thickness at steady state?

Note: Watch your units on viscosity. Assume jJ. =~.ter at 25°C.
D13. We are separating oxygen and nitrogen with a composite membrane,

ao. r.. = Po.lPN• = 5.4. We have set up a membrane cascade of per-
fectly mixed modulus to produce a final permeate product of yp = 0.95
oxygen. In the last module PIiPH = 0.25 and e =0.60. What must be
the inlet mole fraction to this last module (feed contains only O2 and
N 2 )?
FI. Estimate the osmotic pressure of acetic acid, CH3 COOH, in a 25 wt %

aqueous solution at 25°C.
a. Use van't Hoff Eq.
b. Use Eq. (12-17).
c. Use Eq. (12-19).

Data is in Perry's and the Handbook of Chemistry and Physics.
F2. (Derivation). It is unusual to have a ratio of gas constants in an equa-

tion such as in Eq. (12-19). Show in the derivation (see Reid, 1966)
how this occurs. Note that Reid (1966) does not show this since he is
not concerned with units.
Chapter 13
DETAILED THEORIES FOR MEMBRANE SEPARATIONS
The basic concepts, definitions, and theories for a variety of membrane separa-
tors were presented in Chapter 12. In order to cover the entire area of mem-
brane separations while at the same time keeping that chapter a reasonable
length some of the more involved analyses were not included. These more
involved analyses will be discussed in this chapter. Chapter 12 is a prerequisite
for this material.
The organization of this chapter is to divide the membrane separator into

three parts. The first part (Sections 13.1, 13.2, and 13.5) is the transfer from the
bulk fluid to the membrane wall. This part discusses concentration polariza-
tion, gel formation and fouling. The second part (Sections 13.3 and 13.4) con-
siders the overall flow pattern in the membrane module such as those illustrated
in Figure 12-6. The third part (Section 13.6) treats the transport of solvent and
solute inside the membrane.
Concentration polarization was discussed in Chapter 12 and a simple

theory was presented there. However, the equations were left in terms of the
boundary layer thickness B or the mass transfer coefficient k, and no mention
was made of how to determine B or k. The determination of k using correla-
tions will be done in Section 13.1. The gel model for UP was also discussed in
Chapter 12. A more complete analysis will be done in Section 13.2.
The bulk flow patterns in the membrane modules have a large effect on
separation. In Chapter 12 fluid on both sides of the membrane was assumed to
be perfectly mixed. In Section 13.3 perfectly mixed modules with concentra-
tion polarization are studied. Countercurrent flow is studied in Section 13.4.
The transfer phenomena inside the membrane was essentially ignored in
733
734
Chapter 12. In Section 13.6 an irreversible thennodynamics analysis of this

transfer is developed. The solution-diffusion model for transfer inside of RO
membranes and a frictional model appropriate for transfer inside of porous UF
membranes will be presented.
13.1. APPROXIMATE ANALYSIS OF CONCENTRATION POLARIZA·

TION
In this section a somewhat more detailed model of concentration polarization

than that presented in Eqs. (12-25) to (12-32) will be developed. The analysis
in this section uses mass transfer correlations to find k.
13.1.1. Basic Equations.
The usual picture of concentration polarization was shown in Figure 12-4. The
solute mass balance was given by Eq. (12-31) when the rejection is 1.0 and cp
= O. When the rejection is not 1.0. the mass balance can be written as:
dc (13-1)
]solv cp +]solv c +D -
dy
=a
where the first tenn represents solute which passes through the membrane. The
boundary conditions are
(13-2a)
c=cw at y=O
(13-2b)
c = Cb at y= 0
Since Cp =(l-RO)cw • the solution to this set of equations is
Cw exp (Jsolvo/D) (13-3a)

M=-=---------
Cb RO + (l-RO) exp (Jsolv B/D)
735
which reduces to Eq. (12-32a) when RO = 1.0. In this result RO = 1 - cp/cw is

the intrinsic rejection in the absence of concentration polarization.
To use this result we must be able to estimate o.

Since 0 controls the
boundary condition. This can be done by assuming that 0 is the same as the 0
for a semi-impermeable wall which has R=1. This is a reasonable estimate for
low flux membranes since the solute flux through the membrane is much less
than the solute flux parallel to the membrane. Then the mass transfer
coefficient to the wall is
(13-4)
k=D/o
This mass transfer coefficient k can be calculated from mass transfer correla-
tions for different flow regimes. This is the subject of Section 13.1.2. Substi-
tuting Eq. (13-4) into (13-3a), we obtain
Cw exp Osolvlk) (13-3b)

M=-=---------
Cb RO + (1 - RO) exp Osolvlk)
JI01v is given by Eq. (12-27). Note that Equations (12-27) and (13-3b) are valid
at each point of the membrane since, in general, Cb, c w, cP ' J solv , and M vary
throughout the membrane module. For a mixed system Cb = COul is constant and
hence Cw, cP ' M, and Jsolv are constant. When RO = 1, simultaneous solution of
Eqs. (12-27) and (13-3b) gives J so1v and M. If RO < 1, we must simultaneously
solve Eqs. (12-22), (12-25), (12-27) and (13-3b). For high flux membranes
Eqs. (13-4) and (13-5) are both suspect.
Equations (13-3), (13-4) and the expressions for k are convenient to use
when Jsolv is constant and specified. If Jso1v is unknown, these equations must
be solved simultaneously with Eq. (12-27). Unfortunately, the polarization
modulus M is required to calculate J so1v in Eq. (12-27) and Jso1v is required to
determine M in Eq. (13-3). Thus an iterative trial-and-error solution is often
required Fortunately, if Jso1vlk «1.0 an approximate solution can be
developed (Rao and Sirkar, 1978). This solution is particularly simple for ideal
semi-impermeable membranes where RO = 1. Assume that the osmotic pres-
sure is a linear function of concentration following Eq. (12-18b) which is rea-
736
sonable for dilute solutions. When J101v/k « 1.0, the exponential tenn in Eqs.
(12-32a) and (13-3) can be expanded as
cw J 1 JIOIV
, (13-5)
M =-=exp - - - 1 +--
( so v )
Cb k k
where we have inserted RO=1. Setting cp = a (R0 = 1) and 7t(c w ) = ac w in Eq.

(12-27), we obtain
(13-6)
Equations (13-5) and (13-6) can now be solved simultaneously for J so1v '
Ksolv
--(~P-acb)
J lmv = -1m- - - - - (13-7)
1 Ksolvacb
+ 1m k
When using this equation, the assumption that JI01v/k« 1.0 should be checked.
Equation (13-7) is useful when Cb is known. If Cb is not known, we need to
include mass balances (see Section 13.3.2.).
IfRO < 1 and/or (JlOlvlk) is not very small, the result from Eq. (13-7) can
be used as the first guess for J101v ' Then Eq. (13-3b) can be solved for M. Now
the solvent flux equation with RO < I, Eq. (12-27), can be solved simultane-
ously with Eqs. (12-22) and (12-25). This value of J solv can be used in Eq.
(13-3b) and the procedure can be continued until there is convergence.
Once M and JI01v have been detennined the required membrane area and
module length can be found. An external mass balance for the module is
(13-8)
J lmv A Cp + (F - J I01v A)cout =F Cp
In this equation F is the volumetric feed rate to the module (e.g. liter/hr) and
cout is the concentrated product on the feed side of the membrane. Since cout is
737
usually given, Eq. (13-5) can be solved for A. For tubular or hollow fiber sys-
tems,
(13-9)
A = (xLd)n
where n is the number of tubes of diameter d and length L.
13.1.2. Mass Transfer Correlations

For turbulent flow the Nernst film theory can be used to estimate the value of k.
Figure 12-4 can again be used. The boundary layer is assumed to contain all of
the concentration gradient while the bulk flow is turbulent and well mixed. The
mass transfer coefficient in turbulent flow can be written in terms of the
Chilton-Colburn j factor.
. SC-2/3
k = UbJD (13-lOa)
where Sc = WpD is the Schmidt number. In round tubes
(13-lOb)
where f is the Fanning friction factor which in turbulent flow can be estimated
from
(13-1Oc)
f = 0.0791 Re-1/4
where Re = dUbp/1J. is the Reynolds number. This allow estimation of k and o.

An alternate method is to use the standard correlation for mass transfer in
round pipes in turbulent flow (Blatt et al., 1970; Porter, 1979).
(13-11)
k = 0.023 ~ ReO. 83 Sc 1/3
Fully developed turbulent flow on tubes will certainly occur for Re > 20,000
and in UP devices appears to occur at Re = 2000 (porter, 1979). These results
can be substituted into Eq. (13-3) to estimate the concentration polarization
modulus in turbulent flow. Note that concentration polarization in turbulent
738
flow does not depend upon the distance down the tube. For turbulent flow
between parallel sheets the same correlation can be used except the equivalent
diameter of the channel deq should be used
deq = 4 cross-sectional area = 4 2hw (13-12)

wetted perimeter 2w + 2h
where 2h is the gap between the sheets and w is the width.
For turbulent flow in a stirred vessel (Blatt et al., 1970),
0.33 [ 2] 0.75
(13-13)
k=0.0443 ~;
[ ] 00:
00
where d is the vessel diameter in cm, is the stirrer speed in radians/sec, v is
the kinematic viscosity in cm 2 /sec, and D is the diffusivity in cm 2 /sec. This
form is useful for laboratory stirred tanks, but the stirred tank configuration is
unlikely to be used on a large scale. An example calculation is part of Example
13-2.
Turbulent flow can be important in tubular (Figure 12-5b), stirred tanks
and plate-and-frame (Figure 12-5a) modules. Qualitatively, Eq. (l3-3b) shows
that M decreases if the mass transfer coefficient k increases. This can be
achieved with high velocities ~, high diffusivities, and low viscosities.
Operating at high temperature increases D and decreases)J.. Increasing Jsolv
increases concentration polarization. Thus high flux membranes have more
concentration polarization.
For laminar flow conditions the average mass transfer coefficient can be
estimated from (Blatt et al. 1970; Porter, 1979)
(13-14)
where Yw is the fluid shear rate at the membrane surface and L is the length of
the flow channel. The fluid shear rate in laminar flow in a round tube is
4~ (13-15)
Yw=T
739
and thus the average mass transfer coefficient in round tubes is
k= 1.295 [ U~ 2]1/3 (13-16)
For flow between parallel plates with a spacing of 2h
3 Ub (13-17)
YW=-h-
and the mass transfer coefficient is
(13-18)
These laminar flow correlations are used when the concentration polarization
layer is thin, which holds when the axial distance is much less than the entrance
length (see Example 13-2). These correlations be used when a gel does not
form to estimate Jso1v and Musing Eqs, (13-3b) and (13-7), or when a gel forms
(Section 13-2).
Unfortunately, these correlations do not cover many applications which

are commercially significant. RO and gas permeation in hollow fibers usually
have the feed on the outside. The appropriate mass transfer coefficient then
depends on the flow on the shell side of the module. Spiral wound systems are
not just parallel plates wound in a spiral; instead, the spacers are designed to
promote mass transfer. For both these cases experimental data is required.
Fortunately, the manufacturers have much of the required data.
Example 13-1. Concentration Polarization in Turbulent Flow
A 0.4 gmole/L NaCI solution in water is being purified by RO with a

composite polymer membrane. The membrane has J501v = 750 L/m 2 day
with pure water and M' = 70 atm. With a 0.4 gmole/L solution, R O =
0.995 when M = 1 and M> = 70. An existing tubular module with 1 cm
740
id tubes is being used. The average fluid velocity Ub = 1300 cm/sec and
oM> =50 atm. The cut S =0.10. Operation is at 18°C.
Find the average concentration polarization modulus M, the average
solvent flux Jaolv , Couto and cpo
Data:
at 18°C: DNaC! (0.4 gmoles/L) = 1.17 x 10-5 cm 2 /s
DNaC! (0.8 gmoles/L) = 1.19 x 10-5 cm2 /s
(Sherwood et aI. 1975, p. 37)
From Handbook Chemistry and Physics (p. 0-175 of 48th Ed):

Freezing point depression: T; - Tf = 1. 19°C for 0.346 gmole/L
= 1.49°C for 0.435 gmole/L
MIr,w = 1436 ca1Igmole, T;,w =O°C =273.16 K
For 2 wt% NaCI solutions: p = 1.01442 (l0°C), p = 1.01112 (25°C)

4 wt% NaCI: p = 1.0292 (l0°C), p = 1.0253 (25°C)
(perry and Green, 1984, p. 3-83)
Viscosities at 18°C: J.lw = 1.1 cp = 0.011 poise

Jl (25 wt% NaCI) = 2.41 cp
(perry and Green, 1984, p. 3-252)
Solution
A. Define. In problem statement

B and C. Explore and Plan. In order to determine the mass transfer
coefficient from the correlations we need the physical properties. Obvi-
ously, some interpolation will be needed. The osmotic pressure can be
determined from the freezing point depression data with Eq. (12-19). If
the flow is turbulent we will use Eq. (13-11) to find k. Since R is close
to 1.0, we will first assume R = 1 and cp =O. Then from Eq. (12-30).
Cout =cin/(I-S) =0.4/0.9 =0.444 gmoles/L.

The average Cb = (Cin + crut )/2 = 0.422 gmoles/L. Assuming Jso1vik «
1, we can determine J. olv from Eq. (13-7). Now M can be found from
741
Eq. (13-3b) and R and hence S, from the solution to Eq. (12-29). We
can then check that Coo.ll is not changed significantly.
D. Do It In a tube with turbulent flow, radial mixing is fairly uniform

(high k), but axial mixing is incomplete. Thus Cb (and 7t) vary from em
»'
(7t (Cin» to Cout (7t (caut Average conditions are at Cb. Thus, we will
do all calculations at Cb = 0.422 gmoles/L.
1. Calculate 7t. Eq. (12-19) is
= (1.45) (1436) (291.16) [82.057] = 18.69 attn

(18.05) (271.71) (273.16) 1.9872
where T; - Tf is estimated at 0.422 gmole/L by linear interpolation
Then Tf = T; - 1.45 =2.73.16 - 1.45 =371.71 K

T = 18°C =291.16 K
cc attn]
The factor [82.057 gmole K / [ 1.9872
gmole Cal]
K .
converts two different
values of the gas constant. The values ~ and Renc::rgy depend on the
units of .1~, v.olvent, and the desired pressure units for 7t.
Since osmotic pressure data for NaCI is readily available (perry and
Green, 1984, p. 17-23), the purpose of this calculation is illustrative.
Interpolating between concentrations and extrapolating to 18°C using
the tabulated data, we find 7t = 18.57 which is a 0.6% difference.
Assuming 7t is linear with concentration, a = 7t/c = 18.69/0.422 = 44.29
atm/gmole/L.
742
2. Calculate k.
A 0.422 gmole/L solution is about 2.4 wt% (same Table as Freez-
ing point depression). Linearly interpolating, p(WOe) - 1.0174 and p
(25°C) - 1.0140. Then doing another linear interpolation p(l8°C) -
1.0156 glee. [This is not the best way to estimate density, but since the
differences are small it will not cause much error].
Viscosity of a mixture can be estimated from (Reid et ai, 1977, p. 462)
where the Xi are mole fractions. Here component 1 is pure water and
component 2 is 25 wt% solution. From the same table which gives
freezing point depression:
Water. 998.2 gIL = 55.4 gmoles/L

0.422 gmoles salt/L: has - 990.83g water/L = 55.0 gmoIe water/L
25 wt% has 5.086 mole NaCl/L and - 891.5 gwater/L = 49.48 gmoles/L
of water.
Assuming volumes add, we need 0.422/5.086 = 0.083 fraction of 25
wt% and 0.917 fraction of pure water.
In /Jmix =(0.917) In 0.011 + (0.083) In 0.024 =- 4.445

/Jmix = 0.0117 poise
Then, Sc = WpD = 0.0117/(1.0156)(1.17 x 10-5 ) = 984.6
Re = D Ub pill = (1)(1300)(1.0156) = 112,844

0.0117
which is clearly tubulent.
From Eq. (13-11)
743
3. Solvent flux.
From the pure water data:
Ksolv = Jso1v = 750 = 10.71 L/m2 day atm

trn Ap 70
= 1.24 x 10-5 cm/s atm

Assuming J 501v lk« 1.0, we can use Eq. (13-7),
J -
(1.24 x 10-5 ) (50 - (44.29) (0.422» = 0.000386 cm/s
101v - (1.24 x 10-5 ) (44.29) (0.422)
1+ 0.0418
Which is 333.48 2/m2day
Check: Jso1vlk = 0.000386/0.0418 =0.0092« 1.0.
4. Solute.
Now use actual RO = 0.995 in Eq. (13-3b)
M= exp (Jsolvlk) = exp (0.0092) = 1.009

RO + (l-RO) exp (Jsolvlk) 0.995 + 0.005 exp (0.0092)
which is very modest due to the highly turbulent system.

The value of Ksolute /1m can be obtained from the salt rejection data with
M = 1. From Eq. (l2-24C),
Ksolv RO 0.995
0.= = =- - - - - -----
Ksolute (AP- An) (l-RO) (70 - (44.29) (0.4» (1- 0.995)
0.= 3.806
Th Ksolute = Ksolv/1m = 1.24 x 10-5 = 3 25 10~ /

en 1m 0. 3.806 . x cm s.
744
Now the solution to Eq. (12-29) gives~. Constants for the quadratic
formula are
a Ksolv = (44.29) (1.24 x 10-5) = 0.000549

1m
- - up- M 1t (»
Ksolv ( A
Cb +
Ksolute
1m 1m
= (1.24 x 10-5) [50 - (1.009) (18.69)] + 3.26 x 1O~ = 0.000389
Ksolute M Cb =- (3.26 x 1O~) (1.009) (0.422) = -1.387 x 1~

1m
From quadratic formula,
_ - 0.000389 ± ~(0.000389? - 4(0.000549) (- 1.387 x 1O~)

cp = 2 (0.000549)
Use the plus sign to obtain a positive value ofcp •
~ = 0.0039 gmoles/L
Note that, as expected, c p is quite low. This is an average value of cp

since average values for Cb and 1t were used.
E. Check. The solute flux J so1ute = cp Jsolvent = 1.515 x 1O~ which is

small compared to Jsolvent. Thus ignoring this flux in Eq. (13-7) is rea-
sonable. Recalculation of Coot using Eq. (12-30) gives
_ em - e cp = 0.4 - (0.1) (0.0092) = 0443.

Coot - 1- e 0.9 .
where Cp = ~ was used. This is essentially unchanged from our initial

guess based on Cp = O.
745
F. Generalizations. 1. In this example average values of M. J101v • and

S, were found. Since concentration varies along the tube, an exact cal-
culation requires an integration along the length of the tube. Usually,
calculation of average conditions will be accurate enough for both tur-
bulent and laminar flow.
2. Much of the effort in this example was involved with determining
physical properties. This is often true.
3. Section 13.3.2 develops equations for completely mixed modules
with concentration polarization.
13.2. GEL FORMATION AND FOULING
Gel formation was discussed in Chapter 12 following Eqs. (12-37) and (12-39).
In this section more details of the calculation procedures will be given. The
formation of a gel depends mainly upon the solute type and concentration
although the membrane characteristics and hydrodynamics also affect gel for-
mation (Blatt et ai, 1970; Fane, 1986). Rigid chain, solvated macromolecules
such as polysaccharides can gel at concentrations well below 1 wt %. Flexible
chain, linear macromolecules often gel in the range of 2 to 5 wt%, while highly
structured spheroidal macromolecules such as proteins and nucleic acids gel in
the range from 10 to 30 wt %. Colloids can also form gels. Submicron pig-
ments or minerals gel in the range from 5 to 25 vol % while polymer lattices
require from 50 to 60 vol %.
For membranes with R=I, the solute mass balance was given in Eq. (12-
31) and the solution was given in Eq. (12-39),
cg (12-39)
J I01v = k In-
Cb
The mass transfer coefficient can be estimated from the mass transfer theories
discussed in Section 13.1.2. Experimental UP data can be used to easily esti-
mate k and cg • If J solv is determined at a series of bulk concentrations, Cb. a
plot of J solv versus In Cb will have a slope of -k and an intercept on the concen-
tration axis of In cg (see Problem 12-D4). The resulting k and cg values are
746
then useful for correlating data. Changes in conditions such as velocity or tem-
perature will change the mass transfer coefficient and can be correlated with
Eqs. (13-11), (13-13), (13-16) or (13-18).
Should one employ a turbulent or a laminar flow system? For a given

volumetric flow rate the mass transfer rate can be maximized by maximizing
the shear rate. For tubular and plate-and-frame systems the flow can always be
made turbulent by recycling liquid. This will give high k values but also results
in high pressure drops along the length of the module. If the value of k/.1p is
found for both turbulent and laminar flow, one finds that this ratio is much
larger for laminar flow. (See problem 13-Cl). Thus, when energy costs are
included in the calculation laminar flow is usually preferable. When UP is
done in hollow fibers the feed flows inside the fibers, and flow almost has to be
laminar. It is fortunate that laminar flow is desirable.
Although very appealing the gel formation theory often does not agree
with experimental data (Fane, 1986; Le and Howell, 1984; Porter, 1979; Wij-
mans et ai, 1984; Zydney and Colton, 1986). For instance, the model does not
include any affect of the membrane, velocity, or feed concentration on cg , but
experimental results show such effects. Since cg is usually much less than the
solubility limit, it is difficult to determine fundamentally exactly what cg is
measuring. Solute-solute and solute-membrane interactions are important
experimentally, but are not included in the model. Experiments with colloids
show fluxes which can be one or two orders of magnitude greater than expected
(porter, 1979). (This is very helpful and helps explain why the earliest com-
mercial applications of UP were for processing colloids, but it is a major prob-
lem for the theory.) Finally, the theory predicts a limiting flux which is
independent of.1p while experiments often show a.1p dependence on the limit-
ing flux.
Many attempts have been made to explain these problems. For example,
in medium molecular weight applications the osmotic pressure can be impor-
tant and should be included in the model (Fane, 1986; Wijmans et ai, 1984).
Unfortunately, models with osmotic pressure or with gels have very similar
predictions and it is difficult to determine which is correcL For macro-
molecules> 100,000 daltons the osmotic pressure cannot have much influence.
747
Porter (1979) suggested that the colloid data can be explained by the tubular
pinch effect. This is the tendency of small particles flowing in small rubes to
migrate away from the tube walls. This effect will increase the mass transfer of
particles away from the wall and will increase the solvent flux. Although this
effect definitely exists, it does not appear to produce a large enough effect to
completely explain the data (Le and Howell, 1984). Other explanations for
high fluxes with particles are the particle diffusivity is much higher than
Brownian diffusivity because it is caused by particle-particle interaction
(Romero and Davis, 1988; Zydney and Colton, 1986). Other models which
include variable diffusivity or cake aging or adsorption or leaky pores all seem
to be able to explain only part of the phenomena. The mass balance, Eq. (12-
31), does not include axial convection of solute and the boundary condition c =
a
Cb at y = is hypothetical. Shen and Probstein (1977) and Trettin and Doshi
(1980) improved the model by including convection in the x direction (e.g. see
Eq. (13-48» and writing the boundary condition as c = Cb at Y = 00. They also
allowed for a concentration dependent diffusviity. Their results for the limiting
flux agreed better with experimental data for protein UP than Eq. (12-39). At
the current time gelling in UP must be considered only partially explained.
Future research will eventually unravel the complexities involved. The simple
gel formation model still serves as a simple physical picture and as an excellent
method for correlating data.
Fouling can occur in both RO and UP although it is often worse in UP
since dirtier streams are often processed. Fouling is a plugging or coating on
(external) or in (internal) the membrane which is partially irreversible. That is,
in normal operation the fouled membrane will stay fouled until a separate
cleaning step is employed. After clean-up part or all of the original flux will be
recovered. Fouling is complex because it can be caused by many different
phenomena such as precipitation, adsorption, electrostatic attraction, biological
growth, chemical reaction or polymerization.
In RO fouling can be classified as inorganic, particulate, or biological
(potts et aI, 1981). Inorganic fouling occurs when compounds such as CaC03 ,
Caso 4 , MgC03 , and silica or iron precipitate out onto the membrane. Since
the purpose of RO is to remove water and since the concentration of salts is
highest at the membrane wall, it should not be surprising that precipitation can
748
be a problem. The higher the water recovery the more likely precipitation is to
be a problem. Fortunately, precipitation can often be controlled by adding sul-
furic acid to adjust the pH in the range 4 to 6, and by adding compounds which
bind calcium (e.g. Calgon). Fouling from particulates can be caused by parti-
culates in the feed. This is relatively easy to control by prefiltering the feed. A
more serious problem is the presence of both inorganic and organic colloids.
Methods for solving these problems will be very specific for each case since
the chemistry will differ from case to case. Some membranes such as cellulose
acetate are susceptible to biological growth. This can be controlled by operat-
ing in the pH range from 4 to 6, removing dissolved oxygen, or chlorinating.
All of the above types of fouling can also occur in UF. In addition, since
macromolecules are often ultrafiltered, polymer precipitates or gels may form
which can be attached to the membrane by adsorption or electrostatic forces.
The polymer may also react to form a cross-linked gel layer. Generally speak-
ing, the higher the molecular weight of the solute the worse fouling will be.
Modules should be designed to have no stagnant regions since fouling is invari-
ably worse in these locations. Separate cleaning steps are often employed.
These include pulsing clean water, back-flowing clean water through the mem-
brane, mechanical cleaning with sponge balls or other methods, and chemical
cleaning such as a caustic wash. Fouling may also be reduced by changing the
membrane properties. For instance, electrostatic attraction can usually be
reduced by giving the membrane a negative charge since fouling colloids are
often negatively charged.
Flux decline in both RO and UP often follows an exponential decline and

then levels off to a low asymptotic value.
(13-19)
where Jso1v(t), Jinit , and Jasy are the fluxes at time t, initially and asymptotically.
The time t is often measured in days, and JlO1v(day 1) = Jinito The empirical
constant m is negative. An m = -0.1 corresponds to about a 45% decline in
flux in one year while an m = -0.03 corresponds to about a 16% flux decline in
one year. If the membrane is chemically or mechanically cleaned, much of the
original flux can be restored. The decline will then start again from this
749
restored flux. Every time the membrane is cleaned some flux is. pennanently
lost. Thus the membranes will have a finite life. This lifetime can vary from a
few months to several years depending on the service conditions. Membrane
replacement should be included as an operating cost. Membrane systems are
usually designed with an average flux. For a parallel cascade membranes can
be replace in a staggered pattern so that some membranes are always fresh and
others are near the end of their useful life.
13.3. COMPLETELY MIXED SYSTEMS WITH CONCENTRATION

POLARIZA nON
In Chapter 12 the mass balances for completely mixed systems were con-
sidered. In this section we will first develop the external mass balances for
reverse osmosis and ultrafiltration with concentration polarization. The possi-
ble occurrence of a gel will be included in Section 13.3.2.
13.3.1. Completely Mixed RO and UF With Concentration Polarization

and No Gel Formation
In reverse osmosis and ultrafiltration concentration polarization must be

included in the analysis. This will be done for a completely mixed flow case
using the approximation developed by Rao and Sirkar (1978) in Eq. (13-5). In
RO and UP rejection is quite high and it makes more sense to write equations
in tenns of rejections instead of selectivity as used in gas penneation. The
development in this section will include osmotic pressure and is valid for RO
and for UP when a gel does not fonn.
The balances will be done for the system shown schematically in Figure
13-1. The overall external mass balance is
(13-20)
Fin = Fout + Fp
while the external solute balance is
(13-21)
Fin em = Fout Coot + Fpcp
750
~N,CIN
Figure 13-1. Schematic of completely mixed RO or UF with concentration
polarization.
The volumetric permeate flowrate can be determined from the flux as Fp = JA.
For a perfectly mixed container Coot = Cb' Equations (13-20) and (13-21) can
be solved to find Couto
(13-22)
which is Eq. (12-30). Note that these equations are essentially the same as Eqs.
(12-5) to (12-8) since the external balances are unaffected by what happens
inside the system. The permeate concentration is related to the wall concentra-
tion
(13-23)
where RO is the intrinsic rejection which will be assumed to be constant Com-

bining Eqs. (13-23) and (13-22), we obtain
(13-24)
The wall concentration can be removed from Eqs. (13-23) and (13-24) by sub-
stituting in Eq. (13-3b). The results are
(13-25)
751
and, after some algebra,
Cb = ------------
em
(13-26)
8(I-RO) exp (Jsolvlk)
(1-8) + - - - - - - -
RO + (l-RO) exp (JlOlvlk)
Note that if RO =1.0, cp=O and Cotlt=cuJ(I-8). When RO =1, the flow patterns
and solvent flux are unimportant for determining concentrations.
For R O <1 Eqs. (13-25) and (13-26) are very useful if Jso1v is known.
Unfortunately, J so1v is usually not known. The solvent flux must now be deter-
mined from the flux Eq. (12-27). We will assume that the osmotic pressure
depends linearly on concentration so that Eq. (12-18b) is valid. Combining
Eqs. (12-27), (12-18b) and (13-23), we have
(13-27)
The wall concentration can be removed using Eq. (13-3b), and this result can
be solved simultaneously with Eq. (13-26). The equations are nonlinear and a
closed form solution cannot be obtained. If the exponential term is approxi-
mated as in Eq. (13-5), an algebraic solution can be obtained. The wall con-
centration becomes
(13-28)
which gives
(13-29)
After some algebra the bulk concentration is
(13-30)
752
These two equations can now be solved simultaneously. After some manipula-
tions the following quadratic equation results.
(13-31)
a' Jtolv + b' Jso1v + C' = 0
where
(13-32a)
(13-32b)
, _ Kaolv aRc (1 ORO) K.olv A (13-32c)

c - --.:- cin - - --.:- up
After determining k from the appropriate mass transfer correlation, J solv can be
found from the quadratic formula.
- b' ± ""/b'2 - 4a'c' (13-33)

J so1v = 2a'
Once Jso1v is known, Cp and Cb can be calculated from Eqs. (13-25) and (13-26).
If gel formation is possible, calculate Cw from Eq. (13-28). If Cw < c g a gel
won't form and this solution is correct. If Cw > cg , a gel forms and this solution
is not correct (see Section 13.3.2). Also, this solution is based on Jso1vlk < < 1
and this restriction should be checked. If the assumption is invalid, we need to
solve Eqs. (13-3b), (13-26) and (13-27) simultaneously. Equation (13-33) can
be used for a first guess.
IfRo=1 the solvent flux simplifies to
(13-34)
for R=1
753
It is interesting to compare this result to the solvent flux without concentration

polarization and with RO=1. With no concentration polarization Cw = Cb while
cp = O. Solving Eqs. (12-18b), (13-6) and (13-24) simultaneously for these
conditions we obtain,
(13-35)
Then the ratio of solvent fluxes with RO=1 is
J solv 1
-----=--=--------
w. COllC pol.
J101v w/o conc. 1 + em a Kaolv

pol.
forRo = 1 (13-36)
k(1-9)tro
The flux without concentration polarization is always higher, and thus is an

upper limit to the actual flux. As the mass transfer coefficient becomes very
large the two solvent fluxes approach each other.
The flow pattern used in RO and UP is usually not completely mixed.

However, stirred tanks and systems with a very large recycle (see Problem 13-
06) are often close to perfectly mixed. At higher concentrations 7t is usually
not linearly dependent upon concentration. In addition, the pressure drop on
the high pressure side decreases PH and thus decreases ~p (Hwang and Kam-
menneyer, 1975). Complete design calculations are qUite complex and require
computer solutions. These design calculations are discussed by Harris et al
(1976), and Sourirajan (1970,1977).
13.3.2. Completely Mixed UF with Gel Formation

If a gel forms, the analysis in Section 13.3.1 needs to be modified. Equations
(13-23) and (13-24) become
(13-37)
(13-38)
754
where we assume that the solute is mobile in the gel and does not change RO.
Since cg is known, Cp and Cb can be calculated immediately.
When a gel is formed Eq. (13-3b) becomes
(13-39)
which reduces to Eq. (12-39) when RO = 1.0. We can determine k from the
appropriate correlation and then calculate l solv • For laminar flow Eqs. (13-16)
and (13-18) are convenient for first estimates of the average value of k.
If c g is high, cin is low, or 9 is low, we may not know in advance if a gel

will form. In this case, we can calculate Cw using the method in Section 13.3.1.
A gel will form if Cw > c g , and Eqs. (13-37) to (13-39) should be used. This
procedure will be illustrated in Example 13-2.
Example 13-2. Concentration Polarization in Completely Mixed UP
A 10 cm diameter stirred tank with a mixer operating at 4000 rpm is

being used to ultrafilter albumin. With pure water and ~p = 5.0 atm, the
flux is 2500 L/m2 day. RO = 0.992. Operation of the stirred cell will be
at 18°C with ~p = 4.2 atm.
Data: Albumin (porter, 1979), D =6)( 10-7 cm2/s, cg =0.45 (wt. frac),
MWalbumin = 70,000.
Water (perry and Green, 1984): J..lw = 0.011 poise, Pw = 0.9986 g/cm 3 •
Assume v = wp is approximately constant.
Find Cp, Cb and J I01v for feed concentrations of 0.04 and 0.20 wt frac.
albumin ifa = 0.5.
Solution
We will first assume a gel does not form and use Eqs. (13-32) and
(13-33) to find J 101v • Then we will find Ct, from Eq. (13-26) and Cw from
755
Eq. (13-28). If C w < cg , this solution is correct If Cw > cg we need to

use Eqs. (13-37) to (13-39). The mass transfer coefficient k for a stirred
tank can be found from Eq. (13-13). To use this equation,
v = IJ/pw = 0.011/0.9986 g/cm 3 = 0.011 cm2/s

which we will (for now) assume to be constant.
OJ = 4000 rpm I 21t radians I 1 min I = 418.8 radians/s

revolution 60s
Then from Eq. (13-13),
k = (0.0443) [6 x 1~0-7] [ 0.011 ] 0.33 [(418.8) (10)2]0.75

6 x 10-7 0.011
= 0.00594 cm/s
From the pure water flux we can find k,.olv/tm
k,.olv J solv 2500 2

-- = -- = -5 0 = 500 L/m day atm = 0'(lOO578 cm/s atm
1m ~p .
To estimate a in Eq. (12-18b) we will use Eq. (12-18a). Since the solu-
tion will be approximately doubled in concentration, use c = 8 wt%
albumin. Then for 1 liter have approximately: 960 g water = 53.3
gmoles water and 83.5 g albumin = 0.00119 gmoles albumin.
Thus
1t= [0.001~9gmOle] [11itef] [82.057ccatm] (291.16K)

liter 1000 cc gmole k
= 0.0285 atm
from Eq. (12-18a). We can use Cin as 0.08 wt. frac in Eq. (12-18b).
756
Then
1t 0.0285 atrn
a =- = 0 .08 wt fraC =0.356 atrn/wt frac.
c
Since 1t/Ap = (0.0285/4.2) x 100 = 0.68%, we can ignore osmotic pres-

sure for the 4 wt% feed and set a = O. Remember that the van't Hoff
Eq. (12-18a) is highly unreliable; however, the basic conclusion that
osmotic pressure can be ignored is probably valid.
Now, we can detennine J so1v from Eq. (13-33), assuming Jso1v/k « 1.

From Eq. (13-32),
a' = (I-RO)/k = 0.008/0.00594 = 1.347
b' = 1 - (0.5) (0.992) - (0.0005789) [ 0~!4 right)(4.2) + 0 =0.5007

c' = 0 - [1 - (0.5) (0.992)] (0.000578) (4.2) = - 0.001224
From Eq. (13-33)
J _ - 0.5007 ± -/(0.5007)2 - (4) (1.347) (-0.001224)

solv - (2)(1.347)
= 0.00243 cm/s
where the positive sign is used to make J ao1v positive. To check the
assumption calculate
Jsolv/k = 0.00243/0.00594 = 0.409
There will be some error in the calculation (see Notes) and a second
tria' would be required for accurate results.
From Eq. (13-16) we find Cb
Cb =
em = 1.9763 Cin
0.5 + 0.5 (0.008) exp (0.409)
0.992 + (0.008) exp (0.409)
757
while from Eq. (13-38)
(1.9763) em (1.409)
Cw :::: 0.992 + (0.008) (1.409) :::: 2.775 Cin
and from Eq. (13-23)
c p :::: Cw (1 - RO) :::: (2.775 Cin) (0.008) :::: 0.0222 cin
Note that since osmotic pressure is negligible, em does not affect the
calculation of J so1v and the other concentrations can be detennined as
functions of inlet concentration.
4 wt% feed: Cb:::: 0.0791, Cw :::: 0.111 < c g , c p =0.00089
20 wt% feed: Cw :::: 0.555 > cg
Thus, for the 20 wt% solution we need to use the solution when a gel
forms. With a gel formed we use Eqs. (13-37) to (13-39).
Cp :::: (0.45) (0.008) :::: 0.0036
:::: 0.2 - (0.5) (0.45) (0.008) :::: 0 3964

Cb (0.5) .
J so1v ::::
(0.992) (0.45) / (0.3964)
(0.00594) In [ 1 _ (0.008) (0.45) / (0.3964)
1:::: 0.00076 cm/s
Notes:
A. There are several approximations involved in this solution.
1. The worst approximation is v:::: v water in Eq. (13-13). The

albumin solutions will be significantly more viscuous, v will
758
be higher, and k will be lower than predicted. We need

viscosity data.
2. Jso1v/k is really too large to use the approximation that
exp (Jsolv/k) = 1 + Jso1v/k, and R :#:- 1.0. Using calculated
J.olv/k = 0.409, exp (Jsolv/k) = 1.505 which is 6.8% greater
than 1 + Jao1v/k = 1.409. Another trial should be done. For
UP this approximation will often be suspect.
3. Ignoring osmotic pressure in reasonable for the 4 wt% feed.

For the 20 wt% feed osmotic pressure is unimportant
because a gel fonns.
4. When the gel fanned, we assumed R was unchanged. We
need data to check this assumption. If the gel is rigid, R may
become 1.0.
B. The fonnation of the gel reduces Jsolv by a factor of 3.
C. Even with vigorous stirring M = Cw/Cb =2.775/1.9763 = 1.4 which

is significant.
13.4. COUNTERCURRENT GAS PERMEATION WITH NO CONCEN-

TRATION POLARIZATION
The effect of flow patterns in gas penneators is discussed in great detail by

Hwang and Kammenneyer (1975). We will follow their analysis for the coun-
tercurrent gas penneator shown schematically in Figure 13-2. Consider the
1-----1
~ L _____
:::::f1..J
~
Figure 13-2. Schematic of countercurrent gas penneator.
759
separation of a binary gas stream. The differential mass balances for

components A and B are
(13-40a)
"
-d [ql (l-Yl) ] = -d [<U(l-Y2)] = P~B [(l-Yl)PH - PL (l-Y2)] (13-40b)
In these equations ql and qz are the flow rates in moles/hour on the feed and
permeate sides of the membrane, and Yl and Y2 are the mole fractions on the
feed and permeate sides of the membrane, respectively. A is the membrane
area PA and PB are the permeabilities of the membrane. The second equation
p,
in each set comes from the flux Eq. (12-3). The molar density, is required to
convert the volumetric flux JAto a molar flux. These two mass balances can be
rearranged. Adding Eqs. (13-40a) and (13-40b) we obtain
(13-41)
Rearranging Eq. (13-40b) we have a second equation.
(13-42)
A third equation is also required. This can be obtained from the overall
and component mass balances for the mass balance envelope shown in Figure
13-2.
(13-43a)
(13-43b)
760
Removing the unknown flowrate 'l2 and solving for Y2, we obtain
(13-43c)
Equations (13-41), (13-42), and (13-43c) are solved simultaneously. The boun-
dary conditions are:
(13-44a)
(13-44b)
In Eq. (13-44a) Y2.0 is the penneate mole fraction corresponding to Yout. This
penneate mole fraction can be obtained by taking the ratio of flux Eq. (12-3)
written for components A and B at the outlet. This result is
PL
Yout - PH Y2.0
Y2.0 PA (13-44c)
- - = aAB - --------
l-Y2.0 PB PL
l-Yout - - (I-Y2.o)
PH
Note that Eq. (13-44c) is essentially the same as Eq. (12-10) for a perfectly
mixed system.
Obviously, a numerical solution of Eqs. (13-41), (13-42), and (13-43c) is

required. This was done by Blaisdell and Kammenneyer (1973) for oxygen
and nitrogen penneation through a silicone rubber membrane. Results are
shown in Figure 13-3. The results in Figure 13-3 are the same for all flow
configurations at 9 = O. If we desire a concentrated oxygen product, operation
should be at low cut (very little penneate product). If a concentrated nitrogen
product is desired, a high 9 with countercurrent flow should be used. The
=
selectivity, a 2.05, used in these calculations is low. The best currently avail-
able oxygen membrane has a selectivity of about 7.5. This membrane is used
for N2 production. The countercurrent flow arrangement is best since the aver-
age driving force is highest while completely mixed is worst since the average
driving force is lowest (see Problem 13-C4). Balances for flow arrangements
II
0.29
"'~'
0.28 \':0:,
, '.,
.... ',
,
'\ ...... ',
0.27 \. .....
', '"
0.14
'\,".. ,, ,
~ '\, U
<l>
ClJ '~~
'"
"..... ,,
0.26 "\, ". ~
E
ClJ ,,', ".
,"
a.
\', .••.. " c
" 012
<l>
OJ)
c \ '..... ,
ClJ
" 0.25
OJ) ~ 0.10
>-
x ". c
0
'. \.'<-'." \ , 0
c \ ':,.... \ .;::;
0 u
.;::; 0.24 Countercurrent .' \ ".'\;" ,\ - - - Countercurrent
u .::'" 008[
.::'" ---- Cross-flow " '';;.,'. \ - - - - Cross-flow
\
........ Cocurrent " 'i:. \ 0.06 ....•..... Cocurrent
0.23
,,
___ ._ Constant yp '. \ •...'\;." \ \ \ ----- Constantyp
_ •. _ Well mixed
",:\.\ ---- .. Well mixed
0.22
'\.\
, \\' '" 0.02
~-'\,~-~
0.21
I I I I
0
°T0 0.2 04 06 08 10
,~
Cut, !:I Cut
(a) (b)
Figure 13-3. Product concentration for air separation (ex =2.05, prJPH = 0.359, YF = 0.209). a. Permeate con-
centrations. b. Low pressure product concentrations. (Blaisdell and Kammermeyer, 1973).
Reprinted with permission from Chern. Eng. Sci., 28, 1249 (1973). Copyright 1973, Pergamon
--.J
Press. 0\
762
other than completely mixed and countercurrent are developed by Hwang and
Kammermeyer (1975).
13.5. EXACT SOLUTION FOR CONCENTRATION POLARIZATION

IN LAMINAR FLOW
In laminar flow both concentration and velocity boundary layers will form.
Since diffifusivities are low, the concentration boundary layer will be thinner at
the beginning of the tube. Both boundary layers become thicker as one
progresses down the tube. If the tube is long enough (or the channel is thin
enough) both boundary layers will eventualy fill the tube. Thus, one should
expect two solutions for laminar flow: an entrance region solution and a fully
developed far-downstream solution. The solutions have been generated for the
case where the velocity profile is fully developed immediately, but the concen-
tration profile is not. The solute was assumed to be completely rejected. Solu-
tions for both parallel plates and round tubes are available (Sherwood et ai,
1965.
The geometry for RO between two parallel sheets is shown in Figure
13-4a For laminar flow the velocity must first be calculated from the following
Navier-Stokes equations,
(13-45a)
(13-45b)
and the continuity equation
(13-45c)
The boundary conditions are: a) no slip at the wall,
(13-46a)
u=O at y=±h
763
a
t t t t t
yLx tv .. u th
th -
• tv w
• • •
b
2R
\
Figure 13-4. Geometries for concentration polarization analysis. a. Parallel

plates, b. Round tubes.
b) v equals solvent withdrawal rate at wall,
(13-46b)
v= Vw at y=±h
and c) symmetry at the centerline.
au (13-46c)
v=-=O
ay at y=O
The flow solution obtained by Berman (1953) is
-u= -
Uin
3 [ 1 -Vw
2
--x (1-y')
Urn h
1 - h- [ 2-7y'2_7y'4 J
2 [ 1 -Vw
420 v
1(13-47a)
and
v y' ,z vwy ,2 I(j (13-47b)
- = - (3 - y ) - - - (2 - 3y - y )
Vw 2 280 v
764
where Ujn is the average fluid velocity at the channel inlet, and y' = y{h. This
solution reduces to the usual solution for laminar flow between flat plates when
vw =0.
These solutions for velocity can be inserted in the steady state mass bal-
ance for solute. When rejection is perfect, R = I, this balance is,
a [uc - D ax
ax ac] + ay
a [vc - D ac] = 0
ay
(13-48)
The solute boundary conditions are: a) concentration equals Cin at inlet to RO

system,
(13-49a)
c=cin at x=O
b) at the wall the flux of solute to the wall equals the backward difussion of salt
c Vw
ac
= D ay at y= ± h
(13-49b)
and c) symmetry at the centerline.
(13-49c)
~=O at y=O
oy
The solution to Eqs. (13-47) to (13-49) is difficult; however, the equa-

tions can be simplified (Sherwood et al., 1965). For typical RO solvent fluxes,
the terms divided by 420 v and 280 v in Eqs. (13-47) can be ignored. Also,
since the axial concentration gradient will be small, the axial dispersion term,
- Doc/ox, can be neglected in Eq. (13-48). The solution for the entire range of
variables was obtained numerically by Sherwood et al. (1965), and is shown in
Figure 13-5.
Sherwood and his coworkers also obtained asymptotic solutions for cer-
tain regions. The solutions are written in terms of the dimensionless variable
(13-50)
765
l:::,. a =0.0677
o a =0.27
o a =0.50
10.0
M-I
o.I 1..-----'L-...J'---'-J'-'---'--...L-L-'-'---'----L-.<-L-.l----1.----1.--'--I...J..-"----"-,--'-'
0.001 0.01 0.1 1.0 10.0 100.0
~ =8L/3a 2
Figure 13-5. Solutions for laminar Bow between parallel plates (Sherwood
et ai, 1965). Reprinted with permission from Ind. Eng. Chem.
Fundam .. 4, 113 (1965). Copyright 1965, Amer. Chern. Soc.
Very close to the inlet (~ < < I),
(13-51)
M = 1 + 1.536(~)1!3
For larger values of ~ the approximate solution for the entrance region is
(13-52)
M = 6 + ~ - 5 exp(--.J'fj3)
Very far downstream ( large ~ ) the boundary layer will completely fill the
tube. Once this occurs M will be constant.
(13-53)
The locations of validity of these equations are shown in Figure 13-5.

To determine if the far downstream solution Eq. (13-53) or the entrance
region solution Eq. (13-52) should be used, calculate M with both equations.
The smaller M should be used. Very close to the channel entrance, ~ < < I, use
the larger value of M calculated from Eqs. (13-51) and (13-52).
766
A similar but more complicated analysis can be done for laminar flow in
round tubes. The geometry is shown in Figure 13-4b. The Navier-Stokes
equations were solved by Yuan and Finkelstein (1956). Sherwood et ai. (1965)
used this velocity solution in the solute mass balance. The approximate solu-
tion for the entrance region for v! x R/(4llm D2) s 0.02 is
(13-54)
and for v! x R/(4 Uin D2) > 0.02
M= v3XR2
w + 6 - 5 exp [v3
- ( w XR 2 )112 1 (13-55)
4 Uin D 12 llm D
Note that the concentration polarization depends upon the axial distance x.
Far downstream the boundary layer will completely fill the tube. Once
this occurs M will be constant. This far-downstream solution was obtained
numerically and was presented in graphical form. This result is shown in Fig-
ure 13-6 (Sherwood. et ai, 1965). To determine if the entrance region solu-
tions, Eqs. (13-54) or (13-55), or the far-downstream solution (Figure 13-6)
should be used. calculate M by both methods. The smaller M should be used.
If the average polarization modulus M is desired, this can be estimated

by doing the calculation at x = L/2. If either the solvent velocity through the
tube wall Vw or the solvent flux J so1v are given, the calculation of M(x) or M is
straightforward for both geometries. However, if M, J so1v and Vw are unknown
the problem is trial-and-error. This involves guessing M and then calculating
Jso1v from Eq. (12-27) or Eq. (13-6) if R=1. Then calculate M from the
appropriate solution. If this M does not agree with the M used to calculate
Jsolv' the procedure is repeated.
Although usually quite accurate, these results for both turbulent and lam-
inar flow are approximate for a variety of reasons. The no-slip boundary condi-
tion at the wall is not strictly true for a porous solid since there can be lateral
fluid movement in the pores (Fane, 1986). Fluid properties P,1l and v and
767
100
80
60 -
40
0
20
M-I 10
8
6
0.02 0.04 0.06 0.10 0.2 0.4 0.6 0.8 1.0
..b
vwR
Figure 13-6. "Far-downstream" solution for round tubes. (Sherwood et ai,
1965). Reprinted with permission from Ind. Eng. Chern. Fun-
dam .. 4, 113 (1965). Copyright 1965, Arner. Chern. Soc.
solute diffusivity vary if the polarization modulus M is large. Mass transfer

coefficients are correlated for systems with impermeable walls, but will be
increased by the flux through the walls. Since concentration polarization
increases as the concentration boundary layer develops, the solvent fiux and
hence Vw usually decreases with increasing x. Including this variation is
significantly more complex than the constant Vw case presented here (see
Matthiasson and Sivik, 1980, for references).
Example 13-3. Concentration Polarization in Laminar Flow
Estimate the average concentration polarization modulus M for a dilute

solution which is being concentrated in an RO system. The properties
768
are D = 5.2 x 10-7 cm 2 /sec, P = 1.01 glee, J.L = 0.0095 cm 2/sec, CL = 0.99
gH 2 0/ml (solvent concentration). The operation is inside round hollow
fibers which are 0.01 cm in radius. Bulk v~locity at the inletis 3.2
cm/sec. Flux rate is 1.8 x 10-4 glsec cm 2 • The totaIlength of each hol-
low fiber is 50 cm. Assume RO = 1.0 and Ppc:rmeate = 1.0.
If the hollow fibers are increased to 500 cm long, estimate the average
concentration polarization.
Solution
First, we need to check the Reynold's number.
R = DUb = 2R Ub = 2(0.1)(3.2) = 6.74

e v v (0.0095)
Flow is laminar. For average value of M use x = L/2 = 25 cm.

Wall velocity,
vw _- - - -_ 1.8 x 10-4 -
flux _ 1. 80 x 10-4 Cm/S
Pw 1.00
and parameter for abscissa of Figure 13-3 is
_D_ =
Vw R
1
5. 2x 0- 7
(1.8 x 10 ) (0.01)
=0.2888
Far downstream result from Figure 13-6, M-l - 3.0 or M = 4. Also,

need entrance region solution. Parameter value
v! x R ] = (1.8 x 1Q-4)3 (25) (0.01) = 0.421

[
4 llm D2 (4) (3.2) (5.2 x 10-7 )2
From Eq. (13-55),
[- [- 3 - ]'h] =2.98
0.421
M=0.421+6-5exp
Since entrance region result is smaller, use that value of M.

769
For L = 500, x = L/2 = 250 cm. Since abscissa in Figure 13-6 is

unchanged, Mfar downstream =4.0.
For entrance region
v3
w
R
X
= (0.421) -250 = 4.21
4lljn D2 25
M = 4.21 + 6 ~ 5 exp [ ~ [ 4.~1 r] = 8.68
Since the far downstream solution is smaller, M =4 when L =500 cm.
Note that this method of calculating an average using x = L(2 is reason-

able if all of the system is in the entrance region or most of the system is
far downstream. The value of M calculated is always somewhat too
large which makes this design approach conservative. A more accurate
prediction can be made by calculating M at several points and then
integrating numerically to find an average.
As a check on the solution for L = 50 we can use Eqs. (13-27) and (13-
3b).
k = 1.295 [ u~~2]1/3 = 1.295 [ (3.2lo~~·~ ~;~-7)2]1/3 = 0.0001555 cm/s
For R O = 1, M = exp (Jsolvlk)

J 501v -- 1.8 x 10-4 g/s 3cm
2 _ 18
-. x
10-4 I
cm s
(1.00 g/cm )
M=ex [ 1.8 X lO-4]=3.18

P 1.555 x 10-4
% Dif = 3.1~~~.98 x 100 = 6.7%

770
Note that 0 = D/k = 5.2 x 10-7 /1.555 x 10-4 = 3.34 x 10-3 cm < R.
If we apply this correlation when L = 500,
k =0.00007218, M =12.108
and 0 = 0.0072 which is close R = 0.01. This closeness of Rand 0 is a

signal to not use this mass ttansfer correlation.
13.6. TRANSPORT INSIDE THE MEMBRANE
In this section two models for ttansport inside the membrane will be developed.
The structure of irreversible thermodynamics will be used for these models and
will be briefly presented first Then the solution-diffusion model appropriate
for RO membranes will be developed in a simplified form. Finally, a frictional
model appropriate for UP membranes which have small pores will be dis-
cussed.
13.6.1. Basic Irreversible Thermodynamics
Irreversible thermodynamics has been extensively applied to study the ttansport

inside the membrane (Hwang and Kammermeyer, 1975; Johnson et ai, 1966,
Lakshminarayanaiah, 1969; Merten, 1966; Soltanieh and Gill, 1981). The flux
of species i through the membrane, Ji is
(13-56)
Ji = Flux i = Ln (Force on i) + L ~j (Force on j)
Thus the flux of species i is proportional both to the force on that species and to
the force on other species. This is usually written as
J.=T··X+
1 ~ 1
~ T··XJ
~ '-'iJ
(13-57)
jU,.l)
where Xi is the force on species i, Ln is the phenomenological coefficient for

the movement of species i caused by the force on that species, and Lij is the
771
phenomenological coefficient for the movement of species i caused by the

force on another species j.
A thennodynamic analysis based on the continuous increase in entropy

can be used to show that
(13-58)
and
(13-59)
Equation (13-59) states that a species will move in the direction of the force
applied on it Equation (13-58) requires that ~j =0 if Lii = O. Thus, if a com-
ponent will not move through a membrane due to forces acting directly on the
component then it will not pass through the membrane at all. Thennodynami-
cally, ~j can be either positive or negative since Eq. (13-58) only limits the
absolute magnitude of the coupling coefficients L ij . In practice all known Lij
are ~ O.
One other condition on the phenomenological" coefficients is Onsager's
reciprocal relationship.
(13-60)
~j = Lji
This states that coupling of two species is the same regardless of which is
experiencing the force and which flux we are considering. This relationship is
based on microscopic reversibility.
The validity of these general statements concerning the phenomenologi-

cal coefficients requires the appropriate choices of the forces Xi' Under isoth-
ennal conditions the forces are
(13-61)
where Ili is the chemical potential of species i and Yi is the external force on
component i. The external force in ion exchange membranes is an electrical
force. In porous membranes where separation is based on sieving the Yi are
frictional forces between the component and the membrane.
772
The chemical potential term can be expanded in terms of concentration

gradients and pressure gradients.
grad Ili = [ dlli

dC' 1 grad Ci + Vi grad p
(13-62)
1 p,T
where Ci is the concentration of species i in weight per volume and Vi is the

partial molar volume of species i,
(13-63)
where N is the number of molecules or particles. Equation (13-62) can be

integrated for the solvent (which is usually called species 1).
Remembering that the osmotic pressure difference across a membrane is the

pressure difference that exists when.1J.!l = 0, then Eq. (13-64) becomes
(13-65)
Substituting this result into Eq. (13-64), we obtain
(13-66)
where we have assumed that Vl is constant so that we could do the last integra-
tion in Eq. (13-64). Note that &t is from one membrane interface to the other.
That is, from c'm to C"m in Figure 13-7.
It will be our purpose to first determine the appropriate terms for the
external forces, to then arrange the equations in a form which allows experi-
mental determination of the coefficients Ln and ~j' and finally to predict the
773
phenomenological coefficients from other properties such as diffusivities and

solubilities.
13.6.2. Solution-Diffusion Membranes (RO and Gas Permeation)
=
Assume that there is no coupling of fluxes (Lij 0), and that external forces are
unimportant (Yi = 0). This assumes that convective flow of water or carrier gas
through pores or pinholes is negligible. This is appropriate for solution-
diffusion membranes used in reverse osmosis and gas permeation. Combining
Eqs. (13-57), (13-61), and (13-66), we obtain
(13-67)
where we have approximated grad J..li as ~Itm. J' is the mass flux in units such
as g/cm 2 s, and J'solv = J solv Psolv. This result agrees with Eq. (12-20a) if we
relate
_ Kao1v Psolv
L 11- (13-68)
VI
The minus sign in Eq. (13-67) comes in because the +x direction shown in Fig-
ure 13-7 is in the direction of osmotic flow for the irreversible thermodynamic
argument. The solvent density appears because mass and volumetric fluxes are
MEMBRANE
SUPPORT
e"2m
Figure 13-7. Schematic of concentration profiles for RO membrane.

774
being compared. Since.rut refers to the difference from concentration c'm to

c"m, concentration polarization and solubilities are included. If 1t is propor-
tional to concentration, Eq. (13-67) will have the same form as (12-27). Thus,
after a lot of work we find that the solvent flux is given by the same equation
we have been using all along. The advantage of the irreversible thermodynam-
ics development is we can now relate the phenomenological coefficients and
the fluxes to more fundamental quantities.
Assume that sorption and desorption at the membrane surface are rapid
so that diffusion controls. Then Lu can be related to the concentration within
the membrane Cim and the mobility IDjm .
(13-69)
The mobility is the velocity per unit force on one mole of particles. In dilute
solutions the mobility can be related to the diffusivity.
(13-70)
mim =D 1m /RT
where R is the gas constant and D 1m is the solvent diffusivity within the mem-
brane. Since the diffusivity and the solvent concentration are often approxi-
mately constant, Lll is often approximately constant. Now the solvent flux is
(13-71)
]'solv =
Remember that .rut refers to the osmotic pressure difference across the mem-
brane. The solvent permeability can be estimated by comparing Eqs. (13-71) to
(13-67) and (13-68)
(13-72)
This allows estimation of permeabilities, and if permeability is determined

experimentally it allows extrapolation to other conditions. This result also
shows that Kso1v will be approximately constant in dilute solutions where the
775
solvent concentration and diffusivity are approximately constant. In concen-

trated solutions Ksolv is not constant (Soltanieh and Gill, 1981).
The flux of solute (species 2) can also be estimated. In RO systems

external forces and the coupling of fluxes can usually be ignored. Then com-
bining Eqs. (13-57), (13-61) and (13-62) gives
(13-73)
where J 2 is a mass flux which is the same as in Chapter 12. Since the last term
is usually negligible, we obtain
(13-74)
The phenomenological coefficient Lzl can again be related to the mobility and
for dilute solutions to the diffusivity.
D'}m (13-75)
~2 = m:an C']m = - - '2m
RT
Since c'}m can change significantly in the membrane, L22 may not be constant.
For dilute solutions
[ ~l
dC:an T,p
RT (13-76)
Combining Eqs. (13-74) to (13-76), we have
(13-77)
where we have approximated dc'}m/dx as AC'}m/tm. In this equation AC'}m is the

concentration difference within the membrane. This is illustrated in Figure
13-7.
776
The relationship between the solute concentration in the membrane and

in solution depends upon the solubility. We will assume that the membrane
surfaces are in equilibrium with the solution and that the distribution coefficient
is constant.
c~ C"2m (13-78)
Kz = - - = - -
CZ w cp
Substituting this into Eq. (13-77), we have
D 2m K z D 2m Kz (13-79)
Jsolute = ---(czw-cZp) = - (M CZb - cp )
t.n t.n
This result agrees with Eq. (12-25) and shows that
(13-80)
Note that this result also agrees with Eq. (12-15) used for gas permeation which
is also a solution-diffusion process.
It is interesting to use these results to study the change in salt concentra-

tion across a RO membrane. The salt concentration on the permeate side, cZ,p
can be calculated from
CZp
[
g salt
cm 3
1 = J'
J, [~l
[g water 1
c
Ip
[g water
cm 3
1 (13-81)
101v cmz s
Then from Eqs. (13-71), (13-79) and (13-81), concentration ratio is
(13-82)
Assuming that cZw » CZp and defining Kl = Clm I Cl p as the water distribution
coefficient in the membrane, we obtain
(13-83)
777
From Eq. (13-83) we can estimate the properties required to obtain a given con-
centration ratio. Note that the concentration ratio in RO membranes is propor-
tional to (Ap - Ax). More separation is achieved by increasing Ap.
Example 13-4. Water Desalination.
What membrane properties are required to desalinate sea water to

potable water using RO?
Solution
A. Define. This problem is not well defined Typical operating condi-

tions are Ap = 100 attn with an average brine concentration of 5 wt %.
Potable water is 500 ppm or 0.05 wt %. We wish to find the required
value of Dim Kl / D2m K 2 •
B and C. Explore and Plan. We can use Eq. (13-83) to estimate the
required properties. We will assume concentration polarization is negli-
gible. The osmotic pressure of 5 wt % brine is about 37 attn while the
osmotic pressure of 0.05 wt% is about 0.4 atm. The value of vdRT is
l
~= 1(8.095 cc/gmole ::.:: 0.000752 atm-l
RT 82.057 (cc atm)
gmoleK
1
(293.16 K)
Solving Eq. (13-83)
D. Do It. If concentration polarization is negligible,
CZw c2b
-=-=100
c2p c2p
Ap - A1t = 100 - (37 - 0.4) = 63.4 attn

778
and
DIm Kl =100 [ 1 ] [_1_] = 2100

D 2m K2 0.000752 63.4
For typical RO membranes Kl is slightly less than 1.0 and

D 1m =D water in membrane - 2.5 to 16 x 10-7 cm 2fs for high selectivity cel-
lulose acetate. (Soltanieh and Gill, 1981). Thus, a typical required salt
permeability is
DIm Kl
D2m K2 = 2100 = 1.2 to 7.6 x 10-10 cm 2fs
E. Check. These results agree with typical values of K2 in cellulose

acetate membranes from 0.017 to 0.039 and typical values of D 1m from
0.8 to 5 x 10-9 cm 2fs (Soltanieh and Gill, 1981).
F. Generalization. Obviously, these results can be changed for other

operating conditions and other membranes. For less selective cellulose
acetate membranes D 1m is higher. Values of diffusivities and distribu-
tion coefficients for a variety of membranes are tabulated by Soltanieh
and Gill (1981). If concentration polarization is important, .11t and
Czw = M C2b increase. Both these effects increase the value of
DIm KIf D2m K 2 . When RO is used to produce Ultrapure water,.11t is
negligible and significantly higher values of C2.w f cz.P can be obtained
(see Problem 13-D8).
13.6.3. Frictional Model For Porous Membranes (UF)
In ultrafiltration separation appears to be based mainly on a sieving mechanism

and not on solution and diffusion. The frictional model assumes that solvent
flow through the membrane is mainly laminar convective flow. Thus the sol-
vent flow can be estimated as Poiseuille flow. The solute flux depends upon the
convective flow (that is coupled to the solvent flow), diffusion, and friction. In
this model coupling between fluxes and external forces (friction) is important.
779
The solute flux is due to the sum of contributions from the convective
flow and diffusion.
(13-84)
Jz = c2m U + J2djf
where u is the center-of-mass velocity of the pore fluid. The term c2rn u
represents -~1 dJlddx. The diffusive flux can be written as LzZX 2 • Lz2 can
again be related to the mobility, and X z is the force from Eq. (13-61). Thus,
aJl2] dC2m
J2di[ = m2 c 2rn [- [ aC2m p,T ~ + Y2
1 (13-85)
where Y2 is the frictional force on the solute due to friction against the mem-
brane. This external force can be related to the solute velocity within the pores
as
(13-86)
where f23 is a friction factor between the solute (2) and the membrane (3). The
mobility mz can also be related to a friction factor except now it is the friction
between the solute and solvent
(13-87)
mz = l/fz1 = D21 / RT
where (2) is solute and (1) is solvent. Combining these equations and solving
for J 2, we obtain
(13-88)
For dilute solutions Eq. (13-76) is valid and we have
(13-89)
780
As u approaches zero this gives us diffusion with a drag factor applied. The
solute flux can also be related to the solvent flow as
(13-90)
J 2 = E U c2p
where E is the porosity of the membrane and ~p is the permeate concentration

in the fluid. Assuming that the distribution coefficient given by Eq. (13-78) is
constant for the porous membrane, Eqs. (13-89) and (13-90) can be combined
and integrated to find C2. The integration is from x=O where c'2m=K2C2w to
X= - lm). where c2m"=K2c2p. ). is the tortuosity which takes into account the
fact that the pores are not straight, ). > 1. The result after some manipulation is
(13-91)
which allows one to predict the permeate concentration once the solvent velo-
city is known. For high negative velocities Eq. (13-91) simplifies to,
C2w K2 f21 (13-92)

~=
(f23 + f2d E
The solvent velocity due to Poiseuille flow in the pores without any fric-
tion would be,
dp .1Pm (13-93)
E u=-H - --H--
dx lm).
where H is the hydrodynamic permeability of the membrane and .1Pm is the

effective pressure driving force.
(13-94)
The terms .11t' and .11t" are the osmotic pressure differences at the two
membrane-liquid interfaces.
~lt' =1t [ c':-J- 1t(c,.)

(13-95a)
~1t" = c'~2m J-
1t [ 1t(c2p)
(13-95b)
781
The hydrodynamic permeability in Poiseuille flow can be estimated from

£R2 (13-96)
H=--P
8J.l
where ~ is the pore radius and J.l is viscosity.
In addition to the pressure gradient there is a frictional force which must

be added to the pressure gradient. Then,
u = -.!!.
E
[ddxP+ Y E(MWh
c2m 1
2
(13-97)
where Y2 is the force per mole of pore fluid given by Eq. (13-86), and
C7m/e(MWh is the molar concentration of solute in the pore fluid. The product
of these is the force per volume of the pore fluid. Combining Eqs. (13-86),
(13-92), and (13-97), we obtain
u =- H
E
[.11mpA. +
m fn J2
E(MW)2
1 (13-98)
The solute flux J2 is given by Eq. (13-90). If this equation is substituted into
Eq. (13-98) and the result is solved for u, we obtain
u =- H
E
[ H
1 + (MW)
~23
2E
c2p
j.11mpmA. (13-99)
Note that u is negative because of the chosen direction for x. Solution of Eqs.
(13-91) and (13-99) simultaneously allows prediction of the solvent velocity
and the permeate concentration through the membrane. (See Problem 13-C3).
In Eq. (13-91) C2w is considered to be known since it can be calculated from the
concentration polarization analysis. The solute flux can then be determined
from Eq. (13-90). The solvent flux can be found from Eq. (13-81) which after
substituting in Eq. (13-90), is
(13-100)
J'lOlv = EU Cl p
782
This model appears to be a reasonable physical model for sieve type

membranes with intrinsic rejections in the range 0 < RO < 1. The theory agrees
qualitatively with experimental data. Unfortunately, predicting the necessary
friction factors, f21 and f23 , and the tortuosity A is difficult Thus this model
should probably be considered a qualitative model which helps to explain the
transport processes inside sieving type membranes. It can serve as a guide to
qualitatively predict or explain the effect of various variables and to help extra-
polate data.
1. Predict the polarization modulus for both laminar and turbulent flow
when no gel layer is formed using mass transfer correlations.
2. Predict the mass transfer coefficient and the solvent flux for UP when a
gel forms. Determine whether or not a gel does form. Discuss the problems
with the gel formation theory.
3. Explain the mass balances for a completely mixed RO systems with con-
centration polarization. Estimate the solvent flux for this system.
4. Explain the mass balances for countercurrent gas permeation and explain
qualitatively why countercurrent operation is superior to other flow patterns.
5. Use the exact solutions for laminar flow to predict the concentration
polarization modulus in tubes or between flat plates.
6. Discuss the application of irreversible thermodynamics to models of

transport inside of membranes, and explain qualitatively both the solution-
diffusion and the frictional models.
783
REFERENCES
Berman, A.S., "Laminar flow in channels with porous walls," J. Appl. Phys.,
24, 1232 (1953).
Blaisdell, C.T. and K. Kammermeyer, "Countercurrent and co-current gas

separation," Chern. Eng. Sci., 28, 1249 (1973).
Blatt, W.P., A. Dravid, A.S. Michaels, and L. Nelson, "Solute polarization and
cake formation in membrane ultrafiltration: Causes, consequences and control
techniques," in J.E. Flinn (Ed.), Membrane Science and Technology, Plenum
Press, NY, 1970,47-97.
Fane, A.G., "Ultrafiltration: Factors influencing flux and rejection," in RJ.

Wakeman (Ed.) Progress in Filtration and Separation 4, Elsevier, NY, 101-
180,1986.
Harris, FL., G.B. Humphreys and K.S. Spiegler, "Reverse osmosis

(hyperfiltration) in water desalination," in P. Meares (Ed.) Membrane Separa-
tion Processes, Elsevier, Amsterdam, 1976, Chapt. 4.
Hwang, S.-T. and K. Kammermeyer, Membranes in Separations, Wiley, NY,

1975.
Johnson, J.S., L. Dresner and K.A. Kraus, "Hyperfiltration (Reverse Osmosis),"

in K.S. Spiegler (Ed.), Principles of Desalination, Academic Press, NY, 1966,
345-439.
Lakshrninarayanaiah, N. Transport Phenomena in Membranes, Academic

Press, NY 1969.
Le, M.S. and I.A. Howell, "Alternative model for ultrafiltration," Chern. Eng.
Res. Des., 62,373 (1984).
Matthiasson, E. and B. Sivik, "Concentration polarization and fouling", Desali-

nation, 35, 59 (1980).
784
Merten, U. "Transport properties of osmotic membranes," in U. Merten (Ed.),

Desalination by Reverse Osmosis. MIT Press, Cambridge, MA, 1966, 15-54.
Porter, M.C., "Membrane filtration," in Schweitzer, P.A. (Ed.), Handbook 0/

Separation Techniques/or Chemical Engineers. McGraw-Hill, NY, 1979, Sec-
tion 2.1.
Potts, DE., RC. Ahlert and S.S. Wang, "A critical review of fouling of reverse
osmosis membranes," Desalination. 36. 235 (1981).
Rao, G. and K.K. Sirkar, "Explicit flux expressions in tubular reverse osmosis
desalination," Desalination. 27. 99 (1978).
Romero, C.A. and RH. Davis, "Global model of crossflow microfiltration

based on hydrodynamic particle diffusion," J. Membrane Sci .. 39, 157 (1988).
Shen, J.J.S. and RF. Probstein, "On the prediction of limiting flux in laminar
ultrafiltration of macromolecular solutions," Ind. Eng. Chern. Fundarn., 16,459
(1977).
Sherwood, T.K., P.L.T. Brian, RE. Fisher and L. Dresner, "Salt concentration
at phase boundaries in desalination by reverse osmosis," Ind. Eng. Chern. Fun-
darn .• 4. 113 (1965).
Soltanieh, M. and W.N. Gill, "Review of reverse osmosis membranes and tran-
sport models," Chern. Eng. Commun .• 12. 279 (1981).
Sourirajan, S., Reverse Osmosis. Logos Press, London 1970.
Sourirajan, S. (Ed.), Reverse Osmosis and Synthetic Membranes. Theory -

Technology - Engineering. National Research Council of Canada, Ottawa,
Canada, 1977.
Trettin, D.R. and M.R Doshi, "Limiting flux in ultrafiltratio of macromolecular

solutions," Chern. Eng. Commun .• 4, 507 (1980).
785
Wijmans, J.G., S. Nakao, and C.A. Smolders, "Flux limitation in ultrafiltration:

Osmotic pressure model and gel layer model," J. Memb. Sci .• 20. 115 (1984).
Yuan, S.W. and A.B. Finkelstein, "Laminar pipe flow with injection and suc-
tion through a porous wall," Transactions ASME. 78.719 (1956).
Zydney, A.L. and C.K. Colton, "A concentration polarization model for the
filtrate flux in cross-flow microfiltration of particulate suspensions," Chern.
Eng. Commun., 47, 1 (1986).
HOMEWORK
AI. UF systems will almost always show concentration polarization. Under

what conditions would you expect this to lead to gel formation? When
would fouling be likely?
A2. Why are there multiple solutions for concentration polarization in lam-
inar flow?
A3. Theoretical analyses of turbulent flow are much more complicated than
laminar flow analysis. However, the concentration polarization analysis
for turbulent flow is simpler than the exact analysis for laminar flow.
Explain this paradox.
A4. Qualitatively, why is a countercurrent flow pattern superior to com-

pletely mixed?
A5. How can you determine if the solution obtained with Eqs. (13-3b) and
(13-16) or (13-18) is reasonable?
786
A6. On one page develop your key relations chart for this chapter. This will
be a challenge because you will have room for only the more important
equations.
C. Derivations
C1. Derive the ratio of k/.1p for both laminar and turbulent flow. Compare
these results.
C2. Derive Eq. (12-57) for a countercurrent dialyzer.
C3. If a constant b < < 1, then exp (b) - l+b. Using this approximation,
simplify Eq. (13-91) when (- u lmAf21IRn < <1. Then solve Eqs. (13-
91) and (13-99) simultaneously to determine c2p' Note u < O. Note that
this assumption of small b will often not be valid.
C4. Prove that the lowest driving force for a countercurrent gas permeation
module equals the driving force for a completely mixed module.
D. Single-Answer Problems
Dl. We are doing an ultrafiltration experiment with a high molecular weight

solute in water. The bulk concentration is 0.005 weight fraction. The
osmotic pressure can be ignored. A 1.5 cm id tubular membrane is
used. Feed to each tube is 100 ml/sec. The membrane has a pure water
flux of 10.5 x 10-4 g/sec cm 2 at a pressure of 38 atmospheres. The
inherent rejection coefficient (M=I) for the membrane is RO = 0.972.
Under the conditions of this experiment (M > 1) the measured rejection
coefficient is R = 0.85. Estimate the solute diffusivity.
Data: p = 1.0 g/cm 3 , J.l =0.0105 poise, no gel is formed.
D2. We are ultrafiltering a dextran solution in a laminar thin-channel sys-

tem. The channel consists of two flat plates 0.1 cm apart. Fluid flows at
3.0 cm/min and is recycled through a 20 cm length of the channel. The
dextran is known to gel at -30 wt %. The solvent flux is measured as a
787
function of dextran concentration. Kinematic viscosity v = 8.96 cm 2 /s.

For the data given in problem 12-04 estimate the dextran diffusivity.
D3. We wish to ultrafilter a 2 wt % aqueous solution of casein at 25°C. The

cg for casein is approximately 14 wt %. The channel is to be 25 cm
long and is a rectangular slit 0.025 cm deep. The bulk velocity is 50
cm/sec. The pressure drop across the membrane is 40 psi and casein
diffusivity is approximately 10-7 cm 2 /sec.
a. Predict the water flux at steady state.
b. If RO = 0.99 (without concentration polarization) and assuming

that the casein is mobile in the gel, estimate R when a gel layer
fOnTIs.
04. Estimate the average concentration polarization modulus for a 4 wt%

NaCI solution. D = 1.61 x 10-6 cm 2/sec. Ph = 1.025 glcc, Sc = 560,
CL = 0.987 gH2 0/ml. Operation is in round tubes 0.05 cm in radius, 50
cm long, and with a bulk velocity at the inlet o( 2.0 cm/sec. Flux rate is
3 x 10-4 glsec cm2 • Use the exact solution for laminar flow.
D5. We wish to do a reverse osmosis experiment on a 0.98 % salt solution.

The osmotic pressure is 6.958 atmospheres and 1t = be is valid. Opera-
tion is at 45 atmospheres. For pure distilled water a flux of 8.5 x 10-4
glsec cm2 is observed. Salt retention without concentration polarization
is 0.998. Tubular membranes 2 cm in diameter are used. Each tube is
80 cm long. Feed to each tube is 1.0 glsec. Including concentration
polarization find the average water flux.
Physical properties: D = 1.6 x 10-6 cm 2 /sec, P = 1.01 glcc, Sc = 560,
CL = 0.99g H2 0/ml (solvent concentration)
06. We are ultrafiltering a polymer in a hollow fiber system. The hollow

fibers are 100 cm long and 0.1 cm id. The polymer gels at cg = 0.015
wt frac. Inlet velocity Um = 3.5 cm/s and em = 0.001 wt frac. Flux rate
before gelling occurs is 3.2 X 10-4 cm 3/cm2 s. RO = 1.0.
a. Estimate the distance down the tube at which the polymer starts to
gel (± 0.1 cm).
788
b. What is the flux after gelling occurs?
DATA: p = 1.01 glee (assume constant)

1.1 = 0.03 poise (assume constant)
D = 5 X 10-7 cmz/s
D7. An experimental gas permeation system with flat plates is being tested.
Rejection, R O = 0.96. The flow path is 100 cm long and the plates are
0.02 cm apart. The high pressure side is at ten atmospheres and the low
pressure side is at one atmosphere. The total gas flux through the mem-
brane is J = 1.0 x 10-3 cc (SzTP). The inlet gas velocity is 100 cm/s.
cm s
Operation is at 25°C and the system is an ideal gas. Find the average
concentration polarization modulus.
DATA: Average MW on high pressure side = 40

D = 0.42 cm3 /s
1.1 = 1.4 x 10-4 poise
D8. We wish to use a cellulose acetate membrane to purify tap water for an
electronics planL The tap water contains 120 ppm IDS (total dissolved
salts) and we wish a product which is 1 ppm or less . .1p = 110 atm.
a Can we do this with existing membranes?

b. If we can do it, what cut a should be used if the RO module is per-
fectly mixed? Set c p = 1 ppm
c. Predict J:01v and J.olut.e if 1m = 1.2 x 10-4 cm for part b.
d. If M = 2.0, set cp = 1 ppm and find a, J~v and Jso1ut.e.

DATA: Operate at 18°C. Assume properties of water (see Example
13-1). Use p. 17-23 in Perry and Green (1984) to estimate 1t. Assume
salts can be treated as if they are NaCI. Kz = 0.025, D 2m =
=
2.5 x 10-9 cm3/s, D w 1.0 x 1O~ cmz/s, Kl 0.98. =
Ignore concentration polarization in parts a, b, and c.
789
D9. Merten (1966) reports values for transfer of sucrose through a cello-
phane membrane at 25°C. At infinite dilution, E = 0.67 and
If- UE =0.9 x 10-4 cm/s at L1Pm = 1 atm, determine:
b. ~
Note: Watch your units.
Part IV
SELECTION AND SEQUENCING

NOMENCLATURE CHAPTER 14
linear equilibrium constant for adsorption
bottoms flow rate of melted crystals
heat capacity bulk adsorbent
C;.sorb heat capacity sorbate
distillate rate
capacity factor in chromatography Eq. (7-32a)
K.olv membrane permeability
L column length. m
LMI'Z length of mass transfer zone. m
M concentration polarization modulus
rate adsorbate product
p pressure
p permeate rate
<kvg average adsorbed phase loading
<kat adsorbate loading at saturation

Q heat supplied or removed
heat for condenser and melter
heat for reboiler
heat required for regeneration
R gas constant
LID. external reflux ratio

external reflux ratio for melter
membrane thickness
793
794
T temperature, K
To surroundings temperature, K
Tc cold (refrigeration) temperature
TRG temperature regeneration gas is supplied at
Ts steam temperature
Tim supply temperature for melter
Weq equivalent work
WmID minimum reversible work
W rev reversible work
WRO equivalent work for RO
~j mole fraction component i in phase j
Yi mole fraction in vapor
Greek
CX;j separation factor, Eq. (14-1)
'Yi activity coefficient
.6.H ads heat of adsorption
power cycle efficiency
Carnot cycle efficiency for refrigeration
latent heat of vaporization
latent heat of freezing
factor defined in Eq. (14-21)
1t osmotic pressure
factor for excess regeneration gas

Chapter 14
SELECTION AND SEQUENCING OF SEPARATIONS
Throughout this book we have been discussing mainly known results - we now
move into the unknown area of selection and sequencing of separations. This
is an area which has been extensively studied for distillation and to a lesser
extent for extractive and azeotropic distillation - all equilibrium staged
processes. Very little research has been done for the rate controlled separations
which are the subject of this book. In this chapter we will discuss what has
been done, and extend by analogy the results obtained for equilibrium staged
processes to rate controlled processes. The results obtained are logical, but are
not proven; thus, they should be used carefully.
We will first consider the general characteristics of separation problems.

Then, an overview of separation methods will be presented. The purpose of
this overview is to aid in the selection of separation methods which will work.
Energy criteria for evaluating separation processes will be presented. After a
brief introduction to the strategy of process design and synthesis, heuristics
(rules of thumb) for developing separation schemes will be discussed. Evolu-
tionary methods which are used to improve the initial sequences will be briefly
introduced. Finally, needed research will be outlined.
14.1. GENERAL CHARACTERISTICS OF THE SEPARA nON PROB-

LEM
There are so many possible types of separation problems that it is hard to be

specific about your immediate problem which becomes the separation problem.
795
796
However, there are a few abstract ideas which are useful in looking at the
separation problem.
1. Invariably, there are several (n) components present and the

initial problem is multicomponent Often only one or two of
these components are valuable.
2. Most real problems will require several methods in series.

Often, one or more separation methods capable of separating
most or all of the components are readily available. There is a
strong temptation to immediately use these methods. Resist
this temptation.
3. The separation can be split into n - 1 binary pairs. These

binary pairs are not unique since they depend upon the physi-
cal properties used by different separation methods (see
Section 14.2). For any given set of n - 1 binary pairs there
will be one separation which is most difficult This most
difficult separation may be different for different choices of
separation methods.
4. If we know how to do all of the separations then the problem
becomes one of choosing the appropriate set of separation
methods. If we do not know how to do all of the separations,
then the initial problem is to find some set of separation
methods which will do the separation. The next problem is
detennining the optimum set of separation methods.
5. The appropriate separation method depends upon where you
are in the life cycle of the product If you are racing to be first
on the market and the product can demand a high price, then
the series of separation processes which can be put on stream
fastest has a definite edge. For more mature products the cost
of the separation methods becomes much more important and
different separation processes may be optimum.
6. The scale of operation is also important. A process which is
economic on a small scale may not be economic on a large
797
scale and vice versa. For example, cyrogenic distillation of

air is currently the most economic air separation method at
very large scale, but either adsorption or membrane processes
or more economic at small scale.
7. The concentration of the feed stream affects the separation

costs. The effect of the naturally occuring concentration on
IOr----------------------------------------------,----~
Vitamin 8-1: •
\\ined gold •
Gold from
Penicillin _
sea water
o
• Bromme from sea water
•
Sulfur from
stack gas (cost)
-,
Wat~r from ~J v.'at~r
-4 0 12 14
-log (.:on,~ntration. mole fraction)
Figure 14-1. Relation between pure product cost and raw material concen-
tration (Sherwood et ai, 1975). Reprinted with permission
from T.K. Sherwood, RL. Pigford and C.R. Wilke, Mass
Transfer, McGraw-Hill, 1975. Copyright 1975, McGraw-Hill.
798
price is illustrated in Figure 14-1 (Sherwood et al, 1975). In

general, products which are dilut.e cost more to separate.
Gold is an interesting example of this. Gold can be recovered
from sea water, but this is not currently economic since mined
gold is more concentrated and hence is cheaper.
8. Every industry has its favorite separation methods. Since

these methods are known to work within that industry, there is
a stong predisposition to use those methods.
9. The criteria used to select a separation method for research

and development (either academic or industrial) are different
than the criteria used to select separation methods to solve a
problem. The focus of this chapter is on solving separation
problems. Bravo et al (1986) and Fair (1987) discuss the
results of a study to evaluate separation processes as candi-
dates for further research with respect to significant energy
savings.
14.2 OVERVIEW OF SEPARATION METHODS
The purpose of this section is to provide an overview of separation methods

and a very preliminary guide on methods of selecting separation methods
which will work.
There are a variety of ways to classify separation methods. One

approach is shown in Figure 14-2. This book has been concerned with diffu-
sional separation processes where the separation is at a molecular level.
Mechanical separation processes involve the separation of bulk phases. Real
problems will often require both mechanical and diffusional separation
methods. The mechanical separation processes are covered in detail elsewhere
(Evans, 1980; Jacobs and Penney, 1987; Ludwig, 1979; Perry and Green, 1984;
Schweitzer, 1979).
799
Chemical
Mechamcal DiffusIOnal Modifications
Centrifuge D,L isomer sep.

Cyclone Ester production
Decanter
Demlster
Electrostatic Separator
Emulsion Separator
T
Heterogeneous
Expression Homogeneous
Filtration Mass Spectrometer
Flotation
High-gradient Magnetic
Impingement Separator
~ Pressure Diffusion
Thermal Diffusion
Ultracentrifuge
E
Magnetic Electrophoresis
Sedimentation IEF
Non-Equilibrium Equilibrium
Scrubber ITP
Sink-float
Membrane or Barrier Non·Membrane Gas-Solid Liquid-Solid Gas-Liquid
Kinetic Adsorption Adsorption Absorption

Electrodialysis Adsorption
Freeze Drying Molecular Sieve Distillation
Pervaporation Molecular
!
Molecular Sieve Clathration Azeotropic
Gas permeation Distillation
Sublim/Desublim Crystallization Extractive
RO
Precipitation Flash
UF
Zone Melting Reactive
Gas Diffusion
Ion Exchange Steam
Chromatography
Ion Exc'~s'on Vacuum
Affinity
Leaching Evaporation
Capillary
Washing Foam Frac!.
Electrochromatography
Drying Solids Stripping
GLC
Liquid-Liquid
GSC
Extraction
Ion Ex.
Dual Temp. Exchange
LLC
Liquid Membrane
HPLC
LSC 3-phase
PC Dialysis
SEC
TLC
Figure 14-2. Classification of separations.

800
The third major category, chemical modification, involves doing a reac-

tion which simplifies the separation. In the production of esters and MTBE in
distillation columns the reaction and the separation are done simultaneously. In
D, L amino acid separations one of the isomers first reacts to fonn an ester, and
then the remaining isomer in the acid fonn is easily separated from the ester.
The diffusional separation processes can be further subdivided into

homogeneous (one phase present) and heterogeneous (two or more phases
present). Electrophoretic processes (Chapter 11) are considered to be homo-
geneous since the actual separation occurs in a single phase-the addition of a
second phase as a anticonvective medium is optional. The heterogeneous
separation processes have been further subdivided into equilibrium and none-
quilibrium separations. Nonequilibrium processes without a membrane or bar-
rier present are relatively Tare. Kinetic adsorption was discussed in Chapters 6
and 8. The nonequilibrium processes with a membrane include all the mem-
brane processes discussed in Chapters 12 and 13. The equilibrium processes
have been subdivided based on the phases present except for chromatography
which is listed separately. I have included chromatography as an equilibrium
process since the major reasons for separation in these methods are equilibrium
differences even though the column may never be at eqUilibrium. The equili-
brium processes also include crystallization methods (Chapters 2 to 5), adsorp-
tion and ion exchange (Chapters 6 to to), and the equilibrium staged processes
such as distillation, extraction and absorption which are covered in many other
books. The most commonly used diffusional separations are the equilibrium
processes.
Figure 14-2 has over 70 separation methods listed, and it does not
include all methods. In addition most methods have many variants which are
not listed, and many methods can be done with a variety of solvents, adsor-
bents, additives, or membranes. The result of all these variations is there are
literally thousands of separation methods available if one includes all the varia-
tions. The engineer's job is to select methods which will solve the problem
economically.
There are many other approaches to classifying separation methods

which may be useful in selecting the appropriate methods. Since separations
801
are based on physical property differences, a classification based on the physi-

cal property differences can be very useful. One way of doing this is shown in
Table 14-1 where the sizes of species being separated is also indicated. Some
separations are listed more than once in Table 14-1 since different property
differences can be used. For example, absorption can be due to vapor pressure
differences (Physical absorption) or due to reversible or irreversible chemical
reaction.
The separations listed in Table 14-1 are also marked as to whether they
use an energy (E) or mass (M) separating agent. Separations using an energy
separating agent use heat (e.g. crystallization) or pumping energy (e.g. gas per-
meation) to effect the separation. Separations using a mass separating agent
add an additional species such as a solvent or purge gas to the mixture to be
separated. Note that the insoluble membrane, adsorbent or packing material is
not considered to be a mass separating agent. However, the solvent or carrier
gas in chromatography and the purge gas or displacer in adsorption is a mass
separating agent The disadvantage of adding a mass separating agent is that
usually the mass separating agent has to be removed before the product can be
used. Thus, a separation scheme using an energy separation agent is usually
required in addition to the mass separating agent processes. Separations such
as precipitation or adsorption can use either an energy or a mass separating
agent depending on how they are operated. Separations such as electrochroma-
tography use both energy and mass separating agents. Gradient chromatogra-
phy systems have two or more mass separating agents.
The separation factor obtainable with the different separations listed in

Figure 14-2 and Table 14-1 is a very important first indication of the feasiblity
of a given separation method. For equilibrium processes the inherent separa-
tion factor is defined as (King, 1980)
(14-1)
where Xi,1 is the mole fraction of component i in phase 1. Phases 1 and 2 are
assumed to be in equilibrium. For distillation Cli,j is the relative volatility. For
00
Table 14-1. Property Differences for Separations otv
(King, 1980; Null, 1987; Rudd et al., 1973)
Property
Differences Molecules Macromolecule Particles
GLC(M)
flash dist. (E)
vapor distillation (E)
pressure absorption (M)
(reI. volatility) sublimation/desublimation (E)
stripping (M)
evaporation (E)
high vacuum distillation (E)
reI. volatity azeotropic dist. (M+E)

+ chern. interact extractive dist. (M+E)
reactive dist (M+E)
distribution extraction (M)

2 liquids liquid membranes (M)
f- two-aqueous phase extraction (M) -t
-LLC(M)-
melting point crystallization (E)
zone melting (E)
solubility crystallization
absorption/stripping (M)
leaching (M)
precipitation (E or M)
solubility + gas permeation (E)

diffusi vity reverse osmosis (E)
pervaporation (E)
liquid membrane (M)
molecular sieves (M) filtration!

size & shape f- size exclusion chromatography (M) -7 demister
clathration (M) centrifuge
(filtering)
f- ultrafiltration (M) -7 micro filtration (M)
00
ow
00
Property ~
electric ~ ion exchange (M) ~
charge ~ ion exchange chromatography (M) ~
ion exclusion (M)
electric charge ~ electrodialysis (E) ~
& mobility ~ electrophoresis (E) ~
~ isotachophoresis (E) ~
electric mobility ~ immunoelectrophoresis (M+E) ~
& chern. rxn ~ affinity electrophoresis (M+E) ~
surface GSC (M)

adsorption ~ TLC(M) ~
f- LC (M) ~
adsorption (M or E)
~ foam fractionation (M) ~
adsorption and shape molecular sieves (M)
adsorption & charge f- paper chromatography (M) ~

adsorption & diffusivity carbon sieves (M)
hydrophobic interaction f- hydrophobic chromatography (M) ---7
rate of evaporation molecular distillation (E)
diffusivity gas diffusion (M + E)
electric mobility & f- electrochromatography (M + E) ---7

SIZe (w. SEC)
electric mobility f- electrochromatography (M) ---7

& adsorption (w. adsorption)
reversible absorption/stripping (M)

chern. rxn liquid membrane (M)
extraction (M)
dual-temperature exchange (M)
f- affinity chromatography (M) ---7
irreversible chern. rxn absorption (M)

00
oVI
00
Property 0
0'\
density & size ultracentrifuge (E) settling

centrifuge
(sedimentation)
cyclone
sink-float
decanter
high gradient
magnetic mag. separation magnetic separator
surface wettability floatation
electrical electrostatic
conductivity separator
thermal diffusivity thermal diffusion (E)
isoelectric point (- isolectric focusing (M+E) ~
molecular wt. (- SDS electrophore~is (M+E) ~

807
linear adsorption or chromatographic systems ~,j is the ratio of equilibrium

constants.
~ k~ (14-2)
~.=-=-,
J AJ k.J
For eutectic crystallization systems in the concentration range where a pure

species is in equilibrium with the eutectic the inherent separation factor is
infinite.
For non-equilibrium processes the inherent separation factor can always

be determined experimentally. The inherent separation factor is again defined
by Eq. (14-1), but Xi, 1 is now the mole fraction of species i in product 1, and the
two product streams are not in equilibrium with each other. In certain cases
where simplifying assumptions can be made ~j can be estimated. For exam-
ple, for reverse osmosis irreversible thermodynamics arguments were used in
Chapter 13 to estimate the change in salt concentration across a membrane. If
Eq. (13-83) is combined with Eq. (14-1) and we assume that the water distribu-
tion coefficient Kw = Kl is the same on both sides of the membrane, we obtain
cwp / cwwaste Dw m Kw Vw (~p - ~1t) (14-3)

aws= ' , =
, cg,p / cs,wutc D S,In Ks RT
where terms are defined in Chapter 13.
The value of the separation factor required before the separation method
is of interest depends on several factors. The ease of multi staging is very
important. In this context multistaging means that the separation acheived can
be multiplied many times and the separation agent can be reused. A packed
bed chromatograph and a packed bed distillation column are both multistage
systems even though no physical stages are present Equilibrium processes
such as distillation, adsorption and chromatography can easily be operated with
many stages, and the separation agent can be reused many times with low addi-
tional capital investment. Thus the separation acheived with a low separation
808
factor can easily be multiplied many times in an economical fashion. Other

equilibrium processes such as crystallization can be staged, but the staging is
not nearly as convenient Membrane processes can be staged (operated in
series), but the low pressure product needs to be repressurized after each stage
and a new module is needed. Thus staging is not easy. The easier staging, the
lower the inherent separation factor can be. As a rough guide the ease of stag-
ing is
adsorption, chromatography> distillation> crystallization> membrane (14-4)
In addition to ease of staging we prefer energy separation processes com-

pared to mass separation agent processes since an additional separation is not
automatically required. This choice favors distillation, crystallization and
membrane processes.
Ultimately, the choice of separation method usually depends upon

economics (occassionally government regulations control the choice). Energy
use is important and the low energy use of the membrane processes is one of
their most appealing features; however, energy is not the only important factor.
Keller (1982) and Ho and Keller (1987) show that for distillation the per year
cost of capital is equal or greater than the energy cost. This is significant since
distillation is normally considered to be a low capital, high energy process.
Thus, capital costs must be included. Energy criteria for selecting separation
processes are discussed in more detail in Section 14.3. Since capital costs are
important, simple systems are favored This favors distillation, absorption, and
membranes. Very complex processes such as displacement adsorption and
electrophoresis tend to be used only when necessary. Some rough guidelines
for chosing processes are given in Section 14.5.
Distillation keeps appearing in the list of possible alternatives. Distilla-

tion is an energy separating agent process which is easy to stage and simple to
construct The design principles are well understood and many companies will
bid to design and build distillation systems. Distillation has good economy of
scale and can be operated in very large sizes. Two pure products can be
recovered, and the principles for sequencing columns have been extensively
809
studied. Although energy use may be high, the energy is usually supplied by
low pressure steam which is often very cheap. The net result is that if distilla-
tion can be used it should be considered as one of the alternatives. Guidelines
for when distillation cannot be used are discussed in Section 14.5.
Example 14-1. Generation of Possible Separation Processes
Generate possible separations to separate ethanol from water.
Solution
A. Define. This problem was purposely ill-defined. The choice of

separation process will depend upon:
a. The purpose of the separation. Do we want pure ethanol or

the azeotrope, or is it a pollution control problem?
b. The feed concentration.
c. The scale of operation.
d. Presence of other components.
To further define the problem asswne a fermentation process is

used to produce approximately an 8 wt% ethanol in water mixture.
We want a water waste stream which can be discarded and pure
(99.9%) ethanol. Both small and large plants should be con-
sidered. For now, the trace contaminants such as glycerol will be
ignored.
B. and C. Explore and Plan. There are obviously a variety of

ways to proceed. For this example we will consider the physical
property differences listed in Table 14-1 to develop possible
separation methods.
810
D. Do It We will list the physical property and the separation

methods that result
1. Relative Volatility: Vapor-liquid equilibrium data is readily

available (e.g. Seader. 1984; Wankat. 1988). At low concen-
trations the relative volatility is quite high (- 12); however.
there is an azeotrope at 0.894 mole fraction ethanol at 1 atm
pressure. Thus distillation can produce pure water and the
azeotrope. and is a commercial method. The azeotrope con-
tains less water as the pressure is decreased (Seader. 1984).
At 70 mm Hg the azeotrope disappears. Thus distillation at
70 mm Hg could be used. but does not appear to be economi-
cally feasible (Black. 1980).
2. Relative volatility and chemical interaction. Azeotropic distil-

lation is commonly used to break the azeotrope. A hydrocar-
bon entrainer such as pentane or benzene can be used to form
a heterogeneous. ternary azeotrope overhead (e.g.• Black.
1980; Seader. 1984; Wankat, 1988). Extractive distillation
can also be used (Black, 1980). but unlike azeotropic distilla-
tion is not a commercial process.
3. Distribution between two liquids. Equilibrium data is avail-

able in several sources such as Leeper and Wankat (1982);
and Roddy (1981). Liquid-liquid extraction can be used to
either extract the ethanol from the azeotrope (Leeper and
Wankat. 1982) or extract ethanol from the feed. The first may
be economical while the second does not appear to be
economically feasible.
4. Melting point. Data is available in the Handbook of Chemis-

try and Physics and Lange's Handbook of Chemistry. Addi-
tion of ethanol to water causes a significant freezing point
depression (-51.3°C at 71.9 wt% ethanol). Freezing can be
used for removing water at low ethanol concentrations
although a countercurrent wash column (see Chapter 5) will
811
be needed. This method might be useful for breaking the

azeotrope, but more data is needed.
5 Solubility and diffusivity. Membranes can be fabricated

which preferentially pass either water or ethanol. Some data
was given in Chapter 12. Reverse osmosis can be used to
preconcentrate a dilute feed, but by itself cannot produce pure
ethanol. Pervaporation units are available commercially for
breaking the azeotrope (see Chapter 12).
6. Surface adsorption. Adsorbents will show preference for

either ethanol or water. Activated carbon preferentially
adsorbs ethanol and is used for ethanol recovery from air
while activated alumina, com grits, starches and molecular
sieves prefer water. Cornmeal is used commercially for
breaking the azeotrope (Ladisch, et al., 1984). Operation is in
the vapor phase.
7. Adsorption and shape. 3A molecular sieves exclude ethanol

and adsorb water (Garg and Ausikaitis, 1983). This method is
used commercially for breaking the azeotrope operating with
a vapor.
8. Rate of evaporation. Water will evaporate more rapidly and

thus molecular distillation could be used, but will be very
expensive.
9. Chromatography. Both GLC and LLC work with a variety of

packings and are used for analysis. They are likely to be
expensive, but could be considered for breaking the azeo-
trope.
E. Check. Data can be checked from other sources, but this is

probably not necessary for an initial list like this.
F. Generalization. This list is not inclusive. It certainly does not

include all possible solvents, entrainers and adsorbents. A detailed
812
literature search will generate additional infonnation. This list

does show that there are a variety of possibilities several of which
compete commerica11y. Also, use of an organized list such as
Table 14-1 is a good method to generate possible separation
schemes.
14.3. ENERGY CRITERIA
Energy requirements for separation processes are discussed in many sources

(Ho and Keller, 1987; Keller, 1982; King. 1980; Null. 1980; Robinson and Gil-
liland, 1950). In this section we will first discuss minimum work requirements
for separations. and then compare the equivalent work required for some
separation processes. Before starting. we will reiterate that energy use is only
one of the criteria used in selecting a separation process.
14.3.1. Minimum Work
The minimum reversible work required for the complete separation of a mix-
ture of ideal gases is (Keller. 1982)
(14-5)
If the feed, PF. and product, pp. pressures are equal this result simplifies to
c
(14-6)
Wmin,T,p =-RT 1: (Yi,F In Yi,P)
i=l
For the separation of a mixture of nonidealliquids the result is
Wmin,T,p = -RT ~ [Xi.F In (Yi Xi.F)]

(14-7)
1=1
These results are easily derived by postulating a reversible process. One

way of doing this is to postulate the existence of perfect semipenneable mem-
branes which will pass one component but none of the other components
813
}----..- I at T,pp
2 at T, Pp
Feed _
PF ,T, Yi,F
1------< } - - - . n at T, Pp
\Membranes
Figure 14-3. Reversible membrane separation process for calculating

minimum work.
(Keller, 1982). This process is illustrated in Figure 14-3. After passing

through the membranes, the gases are completely pure but are at reduced pres-
sures.
(14-8)
Pi = Yi,F PF
Each of these gas streams is compressed isothermally to the product pressure

pp. For an ideal gas the reversible work required to isothermally compress one
mole from Pinitial to Pfinal is
(14-9)
W rev =RT In (Pfinal / Pinitial)
Since Plinal = Pp and Pi,initial = Pi in Eq. (14-8), the reversible work for one gas is
(14-10)
W~v,i = RT In (Pp / Yi,F PF)
The total reversible (and hence minimum) work for the separation is the sum of
Wrev,i for all components. The result is Eq. (14-5). More detailed develop-
ments which include partial separation and nonisothermal operation are given
by King (1980).
The minimum work required for any separation is easily determined.

This is illustrated in Example 14-2.
814
Example 14-2. Minimum Work of Separation
We wish to separate a liquid mixture of ethylbenzene and styrene

at 110°C. Determine Wmin,T,p for mole fractions ethylbenzene of
0, 10,20, ... , 100 in the feed.
Solution
Since the liquid can be assumed to be ideal, ~ = 1. Then Eq. (14-

7) becomes
(14-11)
Wmin,T,p = -RT [XE In XE + (I-XE) In (I-XE)]
for this ideal binary mixture. For this problem,

R = 1.9872 cal/gmole K, T = 383.16 K.
As an example, for a 0.1 mole fraction ethylbenzene feed
W min ,T,p(O.1O) = -(1.9872)(383.16)[0.1 In 0.1+0.9 In 0.9]

= 247.52 cal/gmole feed
The remaining results are sirniliar and are listed in the Table.
W'nun,T ,p Wmin,prod
xE,F (cal/gmole feed) (cal/gmole EB prod)
0 0
0.1 247.52 2475.2
0.2 381.01 1905.1
0.3 465.12 1550.4
0.4 512.44 1281.1
0.5 527.77 1055.5
0.6 512.44 854.1
0.7 465.12 664.5
0.8 381.01 476.3
0.9 247.52 275.0
1.0 0 0
815
Note that the minimum work required to separate a 0.0 and a 1.0
mole fraction feed are both zero since these feeds are already pure.
The minimum work required for separation per mole of feed is
symmetric around a feed mole fraction of 0.5. For nonidealliquids
this will not be true. The third column in the table is of consider-
able interest since it shows the minimum energy required per mole
of ethylbenzene product. As expected, this value decreaBl!s as the
feed becomes more concentrated in ethylbenzene.
The energy requirements calculated in Example 14-2 s.~m quite reason-

able. The minimum work of separation is easily calculated and gives a simple
first approach to compare the difficulty of different separations. Unfortunately,
the fractional efficiency (WminIWactJJal) of most separation processes ranges
from a few percent to about twenty percent (Keller, 1982). Since distillation is
a benchmark process which other processes are compared to, it is interesting to
look at the maximum efficiency of distillation processes. This is shown in Fig-
ure 14-4 (Ho and Keller, 1987) for a simple distillation column producing two
pure products at minimum reflux. Actual columns must operate above the
minimum reflux ratio and the real efficiency will be 5 to 20% lower than the
values shown in Figure 14-4 (Ho and Keller, 1987).
Further research is required on the minimum energy requirements for

some of the more complex processes. In addition, research is required on how
the minimum energy requirements can be incorporated into a search for the
best separation method for a given problem.
14.3.2. Equivalent Work Requirements
Equations (14-5) to (14-11) determine the minimum mechanical work require-

ment. Many separations use heat instead of mechanical work as the energy
source. In order to compare different processes we will calculate the
equivalent work which is the work which could have been obtained by a rever-
sible heat engine. This equivalent work can be calculated from (Null, 1980)
Weq=Q [ T-T
TO 1Ep (14-12)
816
.-
Q)
Q)
0.
E 6
-
0
u
o~
>, c-
uo
c'';::;
.!!:! ~ 4
.5,? ro
:t;c.
Q)Q)
VI
E
~
E 2
'xro
~
o 0.2 0.4 0.6 0.8 1.0

Mole fraction of more volatile component
in feed
Figure 144. Maximum energy efficiency for simple distillation (Ho and
Keller, 1987). Reprinted with permission from Y.A. Liu, H.A.
McGee and W.R. Epperly (Eds.), Recent Developments in
Chemical Process and Plant Design, Wiley-Interscience, 1987.
Copyright 1987, Wiley-Interscience.
where Q is the heat supplied at absolute temperature T and the surroundings are
at To. The efficiency Ep is the power cycle efficiency compared to a Carnot
cycle. From this definition of equivalent work Null (1980) derives approximate
equations for the energy requirements of several commercial separation
processes. The following development is based on Null's (1980) development.
Consider first a simple distillation column with one feed, a condenser,

and a reOOiler. The approximate energy requirements are
(14-13)
where ~ and Qc are the reOOiler and condenser heat loads, respectively; Dis
the distillate product rate; A. is the latent heat of vaporization of the distillate
product; and RD = LID is the external reflux ratio. Combining Eqs. (14-12) and
817
",0
"
~Q
.,
>
~ ..,
c
"-
0'
Q)
~ N
100 200 300 400 500 600
(37.8) (93.3) (149) (204) (260) (316)
T s' F (el
Figure 14-5. Equivalent work requirement for normal distillation. To =
37.8°C (Null, 1980). Reprinted with permission from Chern.
Eng. Prog., 76(8),42 (1980). Copyright 1980, American Insti-
tute of Chemical Engineers.
(14-13), we can determine the equivalent work required for simple distillation.
(14-14)
where Ts is the temperature of the stearn used to supply the heat. Note that Ts
> Treboiler since some temperature difference is necessary for heat transfer. To
is the temperature of the cooling water used. The predicted equivalent work
values normalized by DA. are given in Figure 14-5 (Null, 1980). Null (1980)
also presents results for somewhat more complicated distillation columns.
818
Although the relative volatility does not appear explicitly in Eq. (14-14) or Fig-
ure 14-5, a. is an important parameter in determining Weq since a. controls the
required external reflux ratio. The value of Weq for distillation can now be
compared to the results for other processes.
Melt crystallization is analogous to distillation with the solid crystals tak-

ing the place of the liquid in distillation, and the melt taking the place of the
vapor. Processes were shown in Figures 5-4, 5-5, and 5-6. Consider the center
feed system shown in Figure 5-5. In the melter the approximate amount of
energy required to melt all the crystals is
(14-15)
where Be is the bottoms product of melted crystals, At- is the latent heat of
freezing, and RM is the external reflux ratio for the melter. If the heat is sup-
plied to the me Iter at temperature TIM where TIM> TM• then the equivalent
work for the heated end of the crystallizer is
(TsM - To) (14-16)

Weq.M = Be Ep (RM + 1) At- - . T
.. --
sM
where To is the ambient temperature of the surroundings. At the cooled end of

the crystallizer approximately the same amount of heat is removed. Usually.
refrigeration will be required to form the crystals for reflux. The refrigeration
is required at temperature Te. Then the equivalent work for cooling is
(14-17)
where ER is the Carnot cycle efficiency of the refrigeration system. The total
equivalent work for the melt crystallization process is the sum of Eqs. (14-16)
and (14-17).
W =B (R +1)'L[Ep(TSM-To)+(To-Te)] (14-18)
eq.cry e M "'f T1M T0 E
R
819
Based on energy considerations only, we can now compare crystalliza-

tion to simple distillation by comparing the equivalent work terms from Eqs.
(14-18) and (14-14). Crystallization will be preferred on an energy basis if
Weq.ery < Weq.diA which occurs if
Null (1980) shows some step-by-step simplifications. If Be = D (same amount

of product), T. = TaM (normally Ts > TIM), and cooling can be done with nor-
mal cooling water (Te = To), then Eq. (14-19) becomes
(14-20)
Since latent heats of freezing are typically about 1/5 of the latent heat of vapor-
ization, Eqs. (14-19) and (14-20) often show an energy advantage for crystalli-
zation. However, Fair (1987) notes, "The specific energy advantages of cry-
stallization have not been a factor in its selection for commercial use." These
results do show the potential of crystallization particularly if energy costs
escalate.
Temperature swing adsorption of gases is another process which is

amenable to this type of analysis. Assuming that the column is adiabatic and
that the energy required to heat the column itself is small, we obtain a
simplified energy balance for the energy required for regeneration.
where Cp•sorb and Cp,B are the heat capacities of the sorbate and the bulk adsor-
bent in kJ/kgmole K and kJ/kg adsorbent K, respectively, and 'lavg is the aver-
age adsorbed phase loading at the end of the feed step. For a symmetric mass
transfer zone,
(14-22)
<lavg = <laat (L - 0.5 LMfZ) / L
820
'l'in Eq. (14-21) is a factor to account for the use of excess regeneration gas,
Tads is the average bed temperature at the end of the feed step, Trcg is the aver-
age temperature of the bed after the regeneration step, Mlads is the heat of
adsorption in kJ/kgmole, Nads is the average rate the adsorbate product is pro-
duced in kgmole/hr. Then; is the total regeneration heat required per kg mole
adsorbed.
Combining Eqs. (14-12) and (14-21), we obtain the equivalent work

required
(14-23)
where TRG is the supply temperature of the regeneration gas, TRG > Treg. This
result can be compared to Eq. (14-14) for distillation. Adsorption uses less
equivalent work when
(14-24)
Usually, MladJI > A., and hence C;/A »1. Adsorption typically adsorbs the least
volatile componenL Thus, NadJI/D corresponds to BID if either adsorption or
distillation can do a separation. Adsorption will be favored if BID is small
which means that a small amount of nonvolatile material is to be removed.
Adsorption is also favored if distillation is difficult and a large reflux ratio is
required. Null's (1980) numerical results are shown in Figure 14-6.
Reverse osmosis directly uses mechanical energy and the work required
is easily calculated. The mechanical work required is the volumetric flow rate
times the pressure increase. Although the entire feed stream is pressurized, part
of the mechanical work can be recovered from the high pressure waste stream.
Only the permeate product is completely degraded. Assuming that the permeate
pressure and feed pressures (before pressurizing) are the same, we obtain
(14-25)
where P is the volumetric flow rate of permeate product and ~p is the pressure
821
"...0
0
N
-
G)
110
>
-
co
u. co
(/)
..... -
c
.2
Q.
:!
N
0
(/) 0
".c
c(
110
(J
.c co
~
G) .,
>
0
..a N
co
0.1
0
0
Ix: 500 600
200 300 400
(93.3) (149) (204) (260) (316)
Ts ,F(C)
Figure 14-6. Comparison of equivalent work requirements for temperature
swing adsorption with regeneration at 316°C versus simple dis-
tillation (Null, 1980). Reprinted with pennission from Chern.
Eng. Prog .• 76(8), 42 (1980). Copyright 1980, American Insti-
tute of Chemical Engineers.
drop across the membrane. We have also assumed that the efficiency of energy
recovery from the high pressure waste stream is 100%. Obviously, ~p must be
greater than the osmotic pressure difference. From Eq. (12-27) ~p can be
related to the volumetric solvent flux Jso1v = PIA and the osmotic pressure
difference
Ap __
LJ.
PIA
~--'-~- + M 1t (Cb) - 1t (cp) (14-26)
K so1v 11m
where 1t is the osmotic pressure, A is the membrane area, and M is the concen-
tration polarization modulus. Combining these equations, we obtain
(14-27)
822
The work required is inversely proportional to the permeability of the mem-

brane which is a rate constant Thus, the work requirements depend upon the
membrane rate properties in addition to physical properties. This differs from
previous results obtained for equilibrium processes where rate properties do not
enter into the equivalent work expressions. Note that one must be careful with
units in Eqs. (14-25) and (14-17). In SI units WRO will be in J/S.
With reasonable values for fluxes and permeabilities the work calculated
from Eqs. (14-25) or (14-27) is very low when compared to the equilibrium
processes which involve a phase change (note that there is no latent heat term
in these equations). Thus, from an energy point of view reverse osmosis and
the other membrane separations are very appealing. However, RO is not a
complete separation process since the reject stream is a concentrated solute in
solvent, and cannot be pme solute. The concentration of the reject stream is
limited by the osmotic pressure which increases rapidly at high concentrations.
If a complete separation of solute and solvent is required then RO can be
advantageously coupled with another separation method in a hybrid process.
Further research is required to determine the equivalent work require-

ments for more complex separation processes. Also, methods for using the
equivalent work results to aid in the selection of the appropriate separation pro-
cess are needed.
14.4. STRATEGY OF PROCESS DESIGN AND SYNTHESIS
The appropriate strategy to use for process design and synthesis has been
extensively studied recently (Douglas, 1988; Rudd et al, 1973; Westerberg,
1982). We will follow Westerberg (1982) and present some guidelines without
examples. Detailed examples are given in these references, and in Blumberg
(1988) for liquid-liquid extraction.
Westerberg (1982) presents five guidelines for attacking design prob-

lems.
1. Evolve from simple to complex calculations.

Use approximate calculation methods initially. So many changes
823
will be made in the design that the use of complex and time-
consuming detailed calculation methods is unecessary. Douglas
(1988) discusses short-cut calculation methods for equilibrium
staged processes. Detailed calculation methods should be used in
the later stages of design.
2. Quickly develop a complete feasible solution.

This is a depth first approach and requires that one avoid back-
tracking to improve the design of subproblems. The first feasible
design should be considered to be a learning experience. It shows
where the easy and where the difficult steps are. It may even show
that the entire project probably cannot be done economically with
existing technology. It serves as a guide for later backtracking to
improve steps. Douglas (1988) considers this step in detail.
3. Develop approximate criteria for screening among alternatives.

It is not possible to determine if one has a good design until one
has a design. It is also impossible to evaluate the criteria which
will be used to assess the design without a completed design. To
avoid this dilemma one should use aproximate criteria and heuris-
tics (see Section 14.5.) to obtain an initial design.
4. Alternate between global and detailed solutions.

A global approach is needed to keep the overall goal of the design
in mind. This goal is then partitioned into subgoals which are
further subdivided into additional subgoals until one obtains
subgoals which can be designed without further partitioning. This
is known as a top-down approach. Once the lowest level subgoals
have been identified, detailed design using approximate, short-cut
methods is done. Initially, the purpose of this design is to find
those subgoals which preclude obtaining an economic solution.
This is called a bottom-up approach. If any impossible subgoals
are found one returns to the global design and makes appropriate
changes. Some sort of strategy such as heuristics or an evolution-
ary approach (see Section 14.6.) should be used for generating the
next global design.
824
5. When trying to prove that concepts will not wolk, use optimistic conditions.
Most initial design concepts will not work. During the screening
process the designer's job is to prove things will not work or can-
not be economic. This type of comparison is convincing only
when optimistic guesses are made (e.g. make the optimistic
guesses that there is no entrainment in a crystallization process or
that there is no concentration polarization in a membrane separa-
tion). If one were conservative in making the guesses, the failure
of the design might be due to the guess and one does not have a
clear reason for rejecting the design.
14.5 HEURISTICS
Heuristics are rules of thumb which allow us to make preliminary choices in

complicated situations without doing a large number of detailed calculations.
The use of heuristics for general selection of separation processes has been
considered by Douglas (1988), Hendry et al (1973), Keller (1982), Kelly
(1987), King (1980), Liu (1987), Nishida et al (1981), and Rudd et al (1973)
among others. The heuristics have been most highly developed for distillation.
Some work on heuristics has been done for modifications of distillation such as
azeotropic and extractive distillation, and for absorption and extraction. Only
the heuristics for sequencing distillation systems have been rigorously tested.
The other heuristics appear to be valid, and have received some modest testing.
One problem with heuristics is that different heuristics often conflict with
each other. Then the proper choice of heuristic is unclear. All of the heuristics
should be preceeded with the words, .. IT all things are equal ... " Liu (1987) has
rank ordered heuristics for distillation and modified distillation systems.
In this section a large number of possible heuristics for the selection and
sequencing of separations will be listed and the reasoning for the heuristic will
be explained. These heuristics were developed either by generalizing the
heuristics developed for the equilibrium staged separations or from comments
in the literature. Except for the distillation heuristics, these heuristics have not
been rigorously tested and should always be checked with calculations.
825
14.5.1. G. General and Flowsheet Heuristics
10. Always generate more than one alternative separation method or

sequence.
Although not usually explicitly stated, this idea is inherent in the ideas of pro-
cess synthesis and good problem solving practice.
20. In developing a flow sheet select the separation schemes first before
adjusting temperatures or flow rates.
This heuristic is stated by Rudd et al (1973) and Kelly (1987). Compared to
temperature and flow rate adjustment separations are much more expensive and
require more energy. Thus, the separation problems should receive priority.
30. Reduce the separation load by splitting streams, blending or mixing as

appropriate.
This heuristic applies to the development of flow sheets and was stated by
Rudd et al (1973) and Kelly (1987).
40. Favor schemes which give the smallest product sel

This heuristic states that components which will eventually be in the same pro-
duct should not be separated. This heuristic was listed by Kelly (1987), King
(1980), Liu (1987), Nath and Motard (1981) and Thompson and King (1972)
and others. Nath and Motard (1981) consider this the primary heuristic while
Liu (1987) ranks it third. If extended to chromatography it implies that com-
ponents which will be lumped together as a waste stream should not be
separated.
50. Do not produce purer products than required.

This can be considered as a corollary of heuristic 40.
6G. Do not remix partially separated streams.

If components are ultimately to be separated, any partial separation should be
retained since remixing is an increase in entropy. Although not explicitly
stated elsewhere, this heuristic is a generalization of common practices such as
the use of two-feed distillation columns, "super-loading" in adsorption
columns, and segregated recycle in chromatography.
826
7G. Consider using the standard separation methods of the industry.

These methods may not be optimum. but they will probably give a working
base case. This heuristic is often standard operating procedure. but should
always be used in conjuction with heuristic #lG.
8G. Favor separation methods which are known within the company.
The separation methods which the company is familiar with will require the
least design and development time. These methods will also be easiest to sell
to management. Heuristic #lG should not be forgotten when applying this
heuristic.
9G. For new products look for separations which can be put on stream quickly.
With new products the company which gets to the market first often captures a
large proportion of the market Thus. separation schemes which will take a
long time to develop are not favored for the first plant Known separation
processes will be favored (See heuristics 7G and 8G).
1OG. Consider hybrid plants when one separation method by itself is not ideal
for a given separation problem.
Hybrid plants use two or more separations for the same problem. An example
would be concentrating a stream by reverse osmosis before doing evaporation
even though evaporation by itself could do the separation. Hybrid plants can
utilize the strength of a separation in the concentration range where it is best,
and avoid weaknesses.
14.5.2. D. Design Heuristics
ID. Favor separations using energy separation agents.

These separations do not add anything to the mixture to be separated. This
heuristic is explicitly stated by Liu (1987) and is first in his ordered list. This
heuristic leads to heuristic IS (favor distillation).
2D. If a mass separating agent is used remove it early and do not use a mass
separating agent to recover a mass separating agent.
827
This heuristic is commonly stated (Kelly, 1987; Liu, 1987; Nath and Motard,
1981; Rudd et aI, 1973). It probably does not strictly apply when solutes must
be dissolved in a solvent so that they can be processed. Rudd et al (1973) also
suggest considering the use of a chemical already present in the process even if
it is not optimum based on its properties alone.
3D. Prefer processes with only one and not two mass separating agents.
The use of an additional mass separating agent such as a gradient in chroma-
tography makes the downstream recovery more difficult However, Mazsaroff
and Regnier (1986) note that gradients are required for elution of
macromolecules when separation is due to surface forces.
4D. Avoid excursions in temperature and pressure.

Both very high and very low temperatures should be avoided, but when neces-
sary go high not low (Kelly, 1987; King 1980; Liu, 1987; Rudd et aI, 1973).
This implies that vacuum distillation and distillation with refrigeration are less
likely to be economical. It also favors PSA over VSA. This heuristic also
shows a preference for liquid chromatography instead of gas chromatography
since LC operates at lower temperatures. Both pervaporation (which uses a
vacuum) and low temperature crystallization are contraindicated.
5D. Do the cheapest separation next.

This heuristic was developed by Thompson and King (1972) and is also used
by Douglas (1988). The reasoning is that every separation method used
reduces the volume of material to be processed; thus the more expensive
separations will process less material and will cost less. Although the use of
this heuristic alone may not give optimum flowsheets, it does appear to gen-
erate reasonably good flowsheets quickly. However, this heuristic was tested
only with distillation type separations. Heuristics 7S and 8S for adsorption
and chromatography follow this heuristic.
6D. Avoid solids and if they are present remove them early.
Solids cause processing problems. Separations involving solids (e.g. adsorp-
tion and chromatography) are difficult to operate as countercurrent processes.
828
70. When a solid material needs to be processed consider putting the sample
into solution and removing insolubles to be a separation step.
This heuristic is explicitly stated for chromatography by McDonald and
Bidlingmeyer (1987), but can be extended to any process such as extraction,
crystallization, or adsorption where solids are first dissolved.
80. Consider side withdrawals fer sloppy splits.

Sloppy splits are problems where the product does not have to be of high purity.
This heuristic was developed by Tedder and Rudd (1978) for distillation, but
can be extended to other separations. For example, a side withdrawal for
column switching of chromatography (see Figure 7-16) is roughly analogous to
a side withdrawal in distillation. Side withdrawals would be relatively easy to
use in fractional extraction or crystallization. A single cascade with a side
withdrawal will be quite inexpensive when the side withdrawal can produce the
desired purity.
14.5.3. C. Component and Composition Heuristics
IC. Remove corrosive, hazardous, and unstable compounds early.

This heuristic is commonly stated (Douglas, 1988; Kelly, 1987; Liu, 1987;
Rudd et ai, 1973). Corrosive materials often require expensive materials of
construction, and overall cost can be reduced by removing these components
early. Hazardous materials may require special conditions and also cost more
to process. Unstable compounds decompose less if they are removed early. In
particular, reactive monomers and proteases should be removed early.
2C. Do the difficult separations last.

The difficult separations will require the longest column. It usually makes
sense to do these separations with no other compounds present unless these
other components change the equilibrium and make the separation easier. This
heuristic has been commonly presented for both equilibrium stage separations
(Douglas, 1988; Kelly, 1987; Liu, 1987; Rudd et ai, 1973) and for chromatog-
raphy (Guiochon and Colin, 1986; McDonald and Bidlingmeyer, 1987; Wan-
kat, 1986). Nath and Motard (1981) state that the "easiest separation should be
done first", which is essentially the inverse.
829
3C. Do high recovery and high purity separations last.

This heuristic is closely related to heuristic 2C and is based on the same reason·
ing.
4C. Remove most plentiful component next.

This heuristic was developed for distillation (Douglas,1988; Kelly, 1987; Liu,
1987; Rudd et al, 1973) but is probably applicable to other separations as well.
Use of this heuristic reduces the load on downstream processes.
5C. If component compositions are similiar, favor 50/50 splits or equal size
parts.
This was also derived for distillation and serves to balance the distillation
column (Douglas, 1988; Kelly, 1987; Liu, 1987; Rudd et aI, 1973). Its use for
other separations is not clear.
6C. In organic systems remove water early.

In organic systems water often causes significant problems due to immiscibility
(in distillation), freezing out (in crystallization), or strong adsorption (in
adsorption). Thus, removing the water early even though it may be only a trace
component is often good practice.
14.5.4. S. Separation Specific Heuristics
IS. Favor distillation unless a < <Xcritical.

This heuristic is commonly stated, but the value of Clcritical varies. Null (1987)
uses Ucntical = 1.05, Nath and Motard (1981) and Douglas (1988) use Ucritical =
1.10, for comparison with adsorption Ruthven (1984) uses Clcritical = 1.25 while
Keller (1982) uses a range from 1.2 to 1.5.
2S. In distillation remove products as distillates.

This heuristic is commonly stated (Douglas, 1988; Kelly, 1987; Rudd et aI,
1973). The reason is that most degradation products have high boiling points
and will appear in the bottoms. Thus, distillate products are more likely to be
pure. This heuristic can be extended to other separations (see Problem 14·
All).
830
3S. If the product is sold as a solid the last purification step should probably be
crystallization.
Crystallization can be used both to pnxluce crystals with the desired size distri-
bution and as a final polishing step.
4S. Start considering adsorption instead of distillation when have:
a. (l < 1.2 to 1.5.
b. An azeotrope forms.
c. lightgases
d. a.ds is very high

e. feed is dilute « 10 to 25% of product). This is particularly important if
the product is the heavier component
f. product is thennally unstable
g. there are two sets of compounds with boiling point overlap. Note that
extraction is also commonly used in this case.
This set of heuristics is from Keller (1982).
5S. Consider temperature swing adsorption if adsorbate concentration in the

feed is < 5 mole%.
Keller (1982) notes that energy costs in TSA can be very high and they
increase as the adsorbate concentration in the feed increases.
6S. Consider displacement adsorption only when a.ds > 2 and (ldist < 1.2 to
1.3.
Keller (1982) notes that displacement adsorption processes are quite complex,
and the process should be considered as a last resort.
7S. In adsorption and chromatography consider methods with cheap pac kings
first
This heuristic is related to heuristic 5D. For adsorption this heuristic favors
activated carbon. For chromatography it favors plain silica, activated
alumina, or ion exchange chromatography systems.
831
8S. In adsorption and chromatography use expensive packings last.

This is common practice particularly with HPLC and affinity chromatography
where the packing can be very expensive. Using the expensive packing last
means that the solution is cleaner and is less likely to foul the packing.
9S. Chromatography is expensive - use it only when necessary.

This heuristic is stated by McDonald and Bidlingmeyer (1987).
lOS. In large scale liquid chromatography always optimize the solvent sta-
tionary phase system.
McDonald and Bidlingmeyer (1987) note that the selectivity ex is the most
important parameter, and increasing ex by optimization is well worth the
effort. However, Kopaciewicz and Regnier (1983) note that solvents need to
be compatible for chromatography columns in series or an extra separation
will be required between columns. Solvent selection is also very important in
absorption (Keller, 1982), azeotropic and extractive distillation (Van Winkle,
1967; Wankat, 1988), and extraction ( Blumberg, 1988; King, 1981).
lIS. Consider large-scale gas-liquid chromatography instead of distillation if

ex < 1.10. If ex> 1.2 prefer distillation or vacuum distillation.
This heuristic was stated by Bonmati et al (1984). However, note that liquid
chromatography will also be a competing process and LC has been more suc-
cessful for large-scale applications.
12S. Do ultrafiltration or microfiltration before reverse osmosis.

This heuristic is implied by many process flow sheets. The reasons are that
UP and microfiltration are better able to handle feeds containing particulates,
and they have has higher fluxes. Thus the UP or microfiltration system will
protect the RO system and will decrease the volume which has to be pro-
cessed by the RO system.
13S. Consider RO for preconcentration of dilute aqueous streams in a hybrid

process.
This is one of the classical uses of hybrid processes and is often done with RO
followed by evaporation.
832
14S. If hydrogen or carbon dioxide need to be purified, consider gas permea-

tion.
With currently available membranes these two gases have the highest per-
meabilities, and these applications have now been widely commercialized.
As was noted at the beginning of this section many of these heuristics have
not been rigorously tested. In addition, an ordering of these heuristics for
nondistillation applications is needed. This implies that considerable research
is needed. In the meantime the heuristics must be used with caution.
Example 14-3. Use of Heuristics
Use heuristics to generate preliminary separation fiowsheets for

Example 14-1.
Solution
Following heuristic IG, we will generate several sequences.

Heuristics 2G to 6G, 8G and 9G are not applicable. Heuristic 8G
suggests:
Distillation to produce waste water and azeotrope (also supported
by ID, IS and 2S).
Adsorption or azeotropic distillation for breaking the azeotrope.
Heuristics 9G and 13S suggest preconcentrating feed with RD.

Heuristic 2D suggests removing mass separating agent early. This
can easily be done in the azeotropic distillation since the azeotrope
is heterogeneous.
Heuristics 3D and 65 suggest adsorption should be either tempera-
ture or pressure swing.
Heuristics 5D and 2C lead to distillation first, followed by breaking
the azeotrope.
Heuristics 70, 8D, IC and 3C are not applicable.
Heuristic 4C suggests removing water first.
Heuristic 2S is satisfied by distillation taking the azeotrope as over-
head, but is not satisfied by the usual azeotropic distillation.
833
Heuristic 5S suggests TSA may have high energy costs for break-
ing the azeotrope. An alternative such as PSA should be con-
sidered.
Heuristic 9S probably rules out chromatography.
Heuristic 12S would be applicable if solids were present in the
feed.
The net result are the flow sheets shown in Figure 14-7. Figure
l4-7a is a fairly standard distillation column followed by azeotro-
pic distillation. Reverse osmosis is optional as a preconcentrator
(Note: because of the unusual shape of the VLE curve this precon-
centration has little effect on the reflux ratio RD' Then according
to Eq. (14-13) 0& is not changed much by preconcentration.
Although suggested by the heuristics RO is probably not useful for
ethanol-water and is not shown on the other flow sheets. Precon-
centration by RO can be useful for other systems.) Figure l4-7b
uses distillation followed by liquid phase adsorption. This requires
a drain step, evaporation of liquid, and then thermal regeneration.
Since energy requirements will be quite high the vapor phase
adsorption alternative shown in Figure 14-7c should be considered.
The adsorption can be done as TSA or PSA. TSA is done com-
mercially, but from heuristic 5S we should also consider PSA.
This will require operating the distillation column under pressure.
This may be advantageous since a smaller diameter distillation
column can be used although equilibrium is less favorable.
Other known processes can be substituted for breaking the

azeotropes. Examples are extraction and pervaporation. Heuristic
4D suggests pervaporation may not be advantageous; however,
since it is commercially available it should be considered.
The heuristics give several reasonable flowsheets. Genera-

tion of a final flowsheet will require further economic comparis-
ons. Note that in plants there are also solids and trace components
to be removed.
834
a
Water + EtOH
Azeotropic
Distillation
EtOH
Water
b EtOH
Drain
Adsorption
F
c EtO
nT
PSA or TSA
Regeneration
Go.
F
835
14.6. EVOLUTIONARY METHODS
Evolutionary methods are used to systematically modify an existing flow sheet

to improve it. Although evolutionary methods do not guarentee that the
optimum flowsheet will be generated, they are excellent insurance against
catastrophic failure. Evolutionary methods are discussed by Liu (1987), Nath
and Motard (1981), Nishida et al (1981), Seader and Westerberg (1977), and
Stephanopoulos and Westerberg (1977) among others.
To use an evolutionary method one must first generate an initial

sequence. The initial flowsheet is important. The better this flowsheet is, the
easier it will be to apply evolutionary methods and find a result close to the
optimum. Generation of the initial flowsheet is usually done using heuristics
(Liu, 1987; Nath and Motard, 1981; Nishida et ai, 1981; and Seader and
Westerberg, 1977). This is one reason that heuristics have a prominent place in
this chapter. An alternative to using heuristics is to use the solution to a simi-
liar problem solved in the past. Conversion of flow sheets for similar liquid-
liquid extraction problems is discussed by Blumberg (1988).
The second step is to develop the evolutionary rules. Local evolutionary

methods look only at separation problems which are neighbors (Liu, 1987;
Seader and Westerberg, 1977; Stephanopoulos and Westerberg, 1976). The
local methods have the advantage of reducing the number of calculations which
have to be done, but at the price of limiting the search field. The two evolu-
tionary rules used for the local evolutionary method are (Seader and Wester-
berg, 1977):
1. Interchange the relative position of two neighboring subprob-

lems.
2. Substitute an alternate separation method.
Figure 14-7. Possible ftowsheets for Example 14-3. a Distillation followed

byazeotropic distillation with optional RD. b. Distillation fol-
lowed by liquid phase adsorption. c. Distillation followed by
vapor phase adsorption (TSA or PSA).
836
Application of the rules requires a strategy. The strategy developed by

Seader and Westerbel'g (1977) consisted of the following:
1. Generate each possible neighbor using the two rules. This will
result in a lot of possibilities.
2. Keep a neighbor as a candidate if the heuristic used to choose

the original separation did not clearly discriminate between the two
choices. or ifa competing heuristic would have selected the neigh-
bor.
3. Evaluate the neighboring flowsheets until a locally "best"

flowsheet is found
4. Consider neighbors which are plausible. but were not included

in strategy item 2.
It should be evident that this method can involve a lot of work. Thus, one
needs to use short-cut estimation methods in the evaluation of flowsheets.
A global evolutionary method was developed by Nath and Motard (1981)

and was used by Liu (1987). Unfortunately, this development included only
distillation and extractive distillation; however, the method can probably be
expanded to include other separation processes. Nath and Motard (1981) used
heuristics in the following order:
1 (4G). Favor the smallest product set

2 (IS). Favor distillation.
3 (Inverse of 2C). Easiest separation first.
4 (2D). Do not use a mass separating agent to recover a mass
separating agent
5. (IS). Separations with a < aunn are unacceptable.
6 (4D). Set operating pressure close to ambient.
7. Use user set split fractions.
=
8. Use lID 1.3 (L/D)min.
837
These heuristics are used in the order listed to generate the initial fiowsheet.
The following evolutionary rules are then applied:
1. Challenge Heuristic 1 (smallest product set).

To obtain the smallest product set one may have to use a mass
separating agent Relaxing heuristic 1 may result in a fiowsheet
without the mass separating agent, and this may be a better
fiowsheet
2. Challenge neighboring separations if the difficulty of separation

is about equal or if refrigeration is required for condensation of
distillate.
Nath and Motard (1981) define a Coefficient of Difficulty of
Separation to aid in the application of this rule.
3. Challenge heuristic 2 (favor distillation).

This rule allows one to consider mass separating agent processes.
4. Examine neighbors to see if removal of a mass separating agent

should be delayed.
Heuristic rule 3 (easiest separation) will almost always force the
removal of the mass separating agent next If the mass separating
agent can be used in more than one separation it is advantageous to
leave it in the process longer.
5. Challenge heuristic 3 (easiest separation) if the following

separation is much more difficult than the one being looked at.
Removing a component may make the next separation very
difficult because of composition effects on the equilibrium parame-
ters.
Liu (1987) notes that rules 1,3, and 4 may be very valuable in developing less
expensive designs.
These evolutionary rules can be applied in a variety of ways. The stra-
tegy chosen by Nath and Motard (1981) makes this a global method. Evolu-
tionary rule 1 is done first. Then rules 2 to 5 are treated equally. When a
838
change proves to be advantageous, the downstream process is eliminated and

regenerated. All parts of the process are checked with rules 2 to 5. Nath and
Motard (1981) give examples, but they include only distillation and extractive
distillation.
The evolutionary approach or some other systematic approach to improv-

ing ftowsheets are very useful in developing better ftowsheets. Unfortunately,
the evolutionary approaches need to be developed and validated for separations
other than distillation and extractive distillation. This requires more research.
1. Discuss the type of separation problems which occur.
2. Develop tables of properties to compare separation alternatives for a given

problem.
3. Use energy considerations to compare separation alternatives.
4. Discuss and use different strategies for developing ftowsheets.
5. Use heuristics to develop an initial flowsheet Discuss the rationale for each
heuristic.
6. Discuss and use evolutionary methods.
7. List and discuss the areas which need further research in selecting a separa-
tion method.
REFERENCES
Bonmati, R. t G. Chapelet-Letourneax, and G. Guiochon, "Gas chromatography:

A new industrial process of separation. Application to essential oils", Separ.
Sci. Technol.,19, 113 (1984).
839
Black, C., "Distillation modelling of ethanol recovery and dehydration

processes for ethanol and gasohol," Chern. Eng. Prog .. 76 (9), 78 (1980).
Blumberg, R., Liquid-Liquid Extraction, Academic Press, London, 1988.
Bravo, J.L., J.R. Fair, J.L. Humphrey, CL. Martin, A.E. Seibert and S. Joshi,
Fluid Mixture Separation Technologies for Cost Reduction and Process
Improvement, Noyes Publications, Park Ridge, N.J., 1986.
Douglas, J.M., Conceputal Design of Chemical Processes, McGraw-Hill, New

York, 1988.
Evans, FL., Jr., Equipment Design Handbook for Refineries and Chemical
Plants, 2nd ed., Gulf Pub. Co., Houston, TX, 1980.
Fair, J.R., "Energy-efficient separation process design", in Y.A. Liu, H.A.

McGee and W.R. Epperly (Eds.), Recent Developments in Chemical Process
and Plant Design, Wiley-Interscience, New York, 1987, Chapt. 3.
Garg, D.R. and J.P. Ausikaitis, "Molecular sieve dehydration cycle for high
water content streams," Chern. Eng. Prog .• 79 (4) 60 (1983).
Guiochon, G. and H. Colin, "Theoretical concepts and optimization in prepara-

tive scale liquid chromatography", Chromatography Forum. 21, (Sept.-Oct,
1986).
Hendry, J.E., D.P. Rudd, and J.D. Seader, "Synthesis in the design of chemical
processes," AIChE Journal. 19, 1 (1973).
Ho, F.-G. and G.E. Keller, II, "Process integration", in Y.A. Liu, H.A. McGee,
and W.R. Epperly (Eds.), Recent Developments in Chemical Process and Plant
Design, Wiley-Interscience, New York, 1987, Chapt 4.
Jacobs, L.J., Jr., and W.R. Penney, "Phase Segregation", in R.W. Rousseau
840
(Ed.), Handbook of Separation Process Technology, Wiley-Interscience, New

York, 1987, Chapt 3.
Keller, G.B., n, Adsorption, Gas Absorption, and Liquid-Liquid Extraction:

Selecting a Process and Conserving Energy, Manual 9 in Industrial Energy-
Conservation, The MIT Press, Cambridge, MA, 1982.
Kelly, R.M., "General Processing Considerations:' in R.W. Rousseau (Ed),

Handbook of Separation Process Technology, Wiley-Interscience, New York,
1987, Chapt 4.
King, CJ., Separation Processes, 2nd ed., McGraw-Hill, New York, 1980.
Kopaciewicz, W. and F.E. Regnier, "A system for coupled multiple-column

separation of proteins", Anal. Biochem.129, 472 (1983).
Ladisch, M.R., M. Voloch, J. Hong, P. Bienkowski and G.T. Tsao "Commer-

cial adsorber for dehydrating ethanol vapors," Ind. Eng. Chem. Process Des.
Develop., 23, 437 (1984).
Leeper, S. and P.C. Wankat, "Gasohol production by extraction of ethanol from

water using gasoline as solvent," Ind. Eng. Chem. Process Des. Develop., 21,
331 (1982).
Liu, Y.A., "Process Synthesis: Some simple and practical developments," in

Y.A. Liu, H.A. McGee, Jr., W.R. Epperly (Eds.), Recent Developments in
Chemical Process and Plant Design, Wiley-Interscience, New York, 1987,
Chapt6.
Ludwig, E.E., Applied Process Design/or Chemical and Petrochemical Plants,

2nd ed., Gulk Pub. Co., Houston, TX, 1979.
Mazsaroff, 1. and F.B. Regnier, "An economic analysis of performance in

preparative chromatography ofproteins",l. Liqd. Chromatogr. 9,2563 (1986).
841
McDonald, P.D. and B.A. Bidlingmeyer, "Strategies for successful preparative

liquid chromatography," in B.A. Bidlingmeyer (Ed), Preparative Liquid
Chromatography, Elsevier, Amsterdam, 1987.
Nath, R. and R.L. Mothard, "Evolutionary synthesis of separation processes",

AIChE Journal, 27, 578 (1981).
Nishida. N., G. Stephanopoulos and A.W. Westerberg, "A review of process

synthesis," AIChE Journal, 27, 321 (1981).
Null, H.R., "Energy 'eCOnomy in separation processes," Chem. Engr. Prog., 76

(8),42 (Aug. 1980).
Null, H.R. "Selection of a separation process," in R.W. Rousseau (Ed.)., Hand-

book of Separation Process Technology, Wiley-Interscience, New York, 1987,
Chapt 22.
Perry, R.H. and D.W. Green (Eds.), Chemical Engineers' Handbook, 6th ed.,
McGraw-Hill, New York, 1984.
Robinson, C.S. and E.R. Gilliland, Elements of Fractional Distillation, 4th ed.,
McGraw-Hill, New York, 1950.
Roddy, J.W., "Distribution of ethanol-water mixtures to organic liquids," Ind.

Eng. Chem. Process Des. Develop., 20,104 (1981).
Rudd, D.F., G.J. Powers, and J.J. Siirola, Process Synthesis, Prentice-Hall,
Englewood Cliffs, NJ., 1973.
Ruthven, D.M., Principles of Adsorption and Adsorption Processes, Wiley,

New York, 1984.
Schweitzer, P.A. (Ed.), Handbook of Separation Techniques for Chemical

Engineers, McGraw-Hill, New York, 1979.
842
Seader, J.D., "Distillation," in RH. Perry and D.W. Green (Eds.), Perry's
Chemical Engineers' Handbook, 6th ed., McGraw-Hill, New York, 1984, Sec-
tion 13.
Seader, J.D. and AW. Westerberg, "A combined heuristic and evolutionary
strategy for synthesis of simple separation sequences", AIChE Journal, 23, 951
(1977).
Sherwood, T.K., RL. Pigford and C.R. Wilke, Mass Transfer, McGraw-Hill,
New York, 1975, Chapt. 1.
Stephanopoulos, G. and A W. Westerberg, "Studies in process synthesis - II",

Chem. Eng. Sci., 31, 195 (1976).
Tedder, D.W. and D.F. Rudd, "Parametric studies in industrial distillation. Part
I. Design Comparisons", AIChE Journal, 24,303 (1978).
Thompson, RW. and C.J. King, "Systematic synthesis of separation schemes,"

AIChE Journal, 18,941, (1972).
Van Winkle, M., Distillation, McGraw-Hill, New York. 1967.
Wankat, P.C., Large-Scale Adsorption and Chromatography, CRC Press, Boca

Raton, FL, 1986.
Wankat, P.C. Equilibrium-Staged Separations, Elsevier, New Yorle, 1988.
Westerberg, AW., "Design research. Both theory and strategy", Chem. Eng.
Educ., 16, 12 (1982) and 16,62 (1982).
HOMEWORK
A Discussion Problems
A I.Define the following terms
a. Minimum work
843
b. Equivalent work
c. Heuristic
d. Evolutionary method
e. Local evolution
f. Global evolution
g. Separation factor
h. Diffusional separations
i. Mechanical separations
j. Hybrid process
k. Neighbor
j. Top-down!Bottom-up
A2. When one is developing new separation processes, distillation is often con-
sidered "the enemy". Explain why.
A3.Heuristics can be misleading. For example, heuristic 4D could be mislead-
ing when high pressure gas streams are processed. Explain why one prob-
ably wants to do the separation at high pressure.
A4.Based on an equivalent work analysis crystallization is a very attractive

separation process. Why isn't crystallization used more often?
A5.Explain the differences and similarities between local and global evolution.
A6.Expand on the idea of using flowsheets for similar problems to generate an

initial flowsheet How similar must the problems be? What are some
sources of flowsheets?
A7.0n Figure 14-1 "Water from sea water", vitamin B-12, and "gold from sea
water" all falloff the empirical straight line. Hypothesize why each of
these three is different
844
A8.Why is adsorption favored compared to distillation when a small amount of

nonvolatile material is to be removed?
A9.List and discuss the areas requiring additional research.
AlO. What are "optimistic conditions" for comparison of the following separa-
tion processes?
a. zone melting
b. UP
c. ED
d. LC for protein purification
All. Extend heuristic 2S to the following separations:
a. crystallization
b. adsorption
c. membrane separations
A12. Westerberg (1982) suggests that theories evolve from simple to complex.
Explain how one can do this for
a. Chromatography
b. Reverse osmosis
c. Solution crystallization
A13. Compare depth-first with breadth-first approaches. Why should both

depth and breadth be included? List and discuss the areas requiring addi-
tional research.
B1. Hybrid processs allow the designer to stretch his/her imagination. Mem-
845
brane processes in particular seem to be very useful in hybrid processes.

Develop two novel hybrid processes involving membranes.
B2. Suppose you work for a company whose business is selling processes.
Brainstorm and then pare down a list of separations which should be
explored further by R&D. Explain your final list of 5 items.
B3. Generate possible separations to recover acetic acid from a dilute aqueous
solution.
B4. Use heuristics to generate preliminary separation flowsheets for Problem

14-B3.
C. Derivations
C1.Develop a local evolutionary process which uses Nath and Motard's (1981)
evolutionary rules.
C2.Derive Eq. (14-19).
C3.Derive Eq. (14-14).
C4.Many of the heuristics in this chapter were developed from material in pre-
vious chapters. Generate two additional S heuristics based on information
in previous chapters.
D. Problems
D1.We wish to separate a gas mixture of ethylbenzene and styrene at 110°C.

Determine Wmin,T for mole fractions of ethylbenzene of 0,10,20, ... ,100 in
the feed. Ratio Pp/PF = 1.5.
D2.Determine the equivalent work requirement for a melt crystallization sys-

tem purifying p-xylene from a mixture of para and meta xylenes. The feed
is 40 mole % p-xylene, crystal product is 99.9 mole % p-xylene. Operation
of the crystallizer is very slightly above the eutectic temperature (221 K).
846
Assume the melt product is essentially at the eutectic composition (13 mole
% p-xylene). Pressure is 101.3 kPa Determine the ideal equivalent work
(ep = eR = 1) for melt crystallization if a lOoC approach temperature is used
in the melter and the crystallizer. The ambient temperature To is 18°C.
Melter reflux ratio RM = 0.5. Use a feed rate of 100 kglhr as a basis.
Xylene Data (King, 1980, p. 740): MW = 106.16, Ap = 340 kJ/kg, AM =

343 kJ/kg, Tboiling,p = 411.8 K., Tboiling,m = 412.6 K, Tr,M = 225.4 K, Tr,p =
286.6 K,
(perry and Green, 1984, p. 3-123) Ar,M =26.045 calIg, Ar,p =38.526 cal/g
D3.Find the equivalent work in calI(kg product) required for the RO system in
=
Example 12-2b for M 1.2. Watch your units. Use a water density of 1
kg/liter.
Fl. Different chromatographic methods used for protein purification have dif-
ferent solvent requirements. Read Kopaciewicz and Regnier (1983) and
develop heurispcs for sequencing chromatography columns for protein
purification.
F2. Extend Nath and Motard's (1981) "CoefficienL of Difficulty of Separation"

to center-feed melt crystallization of solid solutions.
APPENDIX. ANSWERS TO SELECTED PROBLEMS
Chapter 2
2-D 1. CUS04 ·5 H20 , a = -1610 , b = 1.727
2-D3. a. Na2 C0 3 ·7 H 20(45.9 kg) , Na2 C03 • H 20(40,4 kg) and

eutectic (13.6 kg)
b. Na2C03 • 7 H20 (65.0 kg) , Na2C03· H 20 (35.0 kg)
30,500 KJ
2-D5. Add water. 66.8 kg KN0 3 , Point b = 0.28 KN0 3
2-D7. F(b(O) = 116.97 kg/hr , F(b(30)) = 188.08 kg/hr.

Win,l = 16.11 kg/hr , W in ,2 = 9.96 kg/hr.
Fhydrate = 81.07 kg/hr
2-Dl1. 3.503 M
Chapter 3
3-D2. Linear. 30°C proportionality = 7.514 x 10-10

70°C proportionality = 6.334 x 10-9
3-D3. a. BO= 41.75 #/cm 3 • s , b. BOzO
Chapter 4
4-D1. 65 mesh, M = 21.9

48 mesh, M =58.6
35 mesh, M = 134.1
28 mesh, M = 254,4
20 mesh, M =385
4-D5a. G = 0.0041 cm/hr , nO = 1.61 x 108 #/cm • L
4-D6. G =0.0821 mm/hr
4-D9. MT = 0.28 gm/cm 3, Ln = 0.31 mm

847
848
4-E1. G = 6.068 x 10-7 mls

Median size 0.9016 mm.
Chapter 5
5-D2. a. Ytot = 0.00178

b. YProd = 0.00102
5-D3. ~ =0.4525 , e=0.1243

5-D5. L = 239.1 em
5-D7. b. x = 0.0024
Chapter 6
6-D2. e = 0.02 until t = 24.6 min then c = 0.50 until t = 25.5 min. At this
point get a diffuse wave.
c = 0.34 at t = 28.0 min
c=0.10 at t= 42.0 min
c = 0.02 at t= 92.2 min
6-D4. a. 1:0 = 4.71 min, tF = 6.19 min

b. tfeed = 1.48 min
!:next = 2.96 min (1.48 min waiting period)
6-D6. CXEP = 7.36
6-D8. b. tOllt = 14.12 min based on data

tOllt = 13.46 min based on Langmuir fit.
Chapter 7
7-Dl. a. UN = 0.0115, UA = 0.00853, up = 0.0062 mlmin

b. L = 0.66 m
c. wait 49.1 minutes after end of pulse
7-D2. AB = 4.25
7-D3. x = 1 - X (L,t) + X(L, t - t1)
849
7-D5. C/CF = 0.5 at 74.29 minutes
7-D7. a. R = 0.317
b. L=799cm
7-D11. ~ = 5.97 J.Ull
7-D13. tsh = 10.15 min, C = 0.0052
7-D14. Heol = 0.0651 cm, cr~ = 0.266 cm 2
Chapter 8
8-D1. a. L = 42.71 cm, Ibr = 36.9 min

b. L = 56.94 cm, Ibr = 36.9 min
8-D3. vsuper = 0.0116 cm/s
8-D4. dp,new = 0.639 mm, DnewIDold = 1.166

8-D7. a. =268.4 em
L
b. Cmethane = 0.0036 gmolejL
8-D8. frac bed uti! col B = 0.82

For 90% uti!, L = 119.1 em, tbr = 900 min.
Chapter 9
9-D3. From 0 to 3 minutes, CT = 0.1 , XNi = 0.2.

From 3 to 6.10 min, CT = 5.1 , XNi = 0.78
Mter 6.10 min, eT = 5.1 , XNi =0.2
9-DS. a. LMIZ = 8.13 em
9-D6. In bands YNa = 1 and YNH4 = 1.
tNa,band = 4 min, tNH4,band =6 min
Chapter 10
10-D 1. N = 4 stages
1O-D3. 88% of protein is recovered.
850
1O-D5. VRN s = 11.93

1O-D7. h = 23.4 cm. Cout = 1.095 mg/ml
Chapter 11
ll-D1. C= 19.265 mV
ll-D2. ~ = 7.907 x 10-6 cm2 /sV
D = 3.07 x 10-8 cm 2/s
ll-D4. a. Wy =0.485 cm , Wz =2.40 cm

ll-D6. ~ = 5.12x 10-4 cm 2/sV
ll-D7. a. E = 16.39 V/cm
b. O'A= 1.0935 cm , WA =4.375 cm
c. O'A = 0.495 cm , WA = 1.98 cm
lI-D8. a. .1pJ = 0.018 pH units

b. pI = 5.53
Chapter 12
12-D1. (lHZ.C02 = 39.3

12-D3. J so1v = 8000 L/m2 day, M = 6.67
12-D4. cg = 0.314
12-D6. a. 1= 57.27 amp , V = 343 volts
b. E = 705.5 cal/L
c. Power = 19.68 kilowatts
12-D9. e =0.160
12-DIO. yp =0.889 , Yout =0.247
12-DI2. a. tg = 6.98x 10-5 cm
b. tg = 1.48 x 10-4 cm
851
Chapter 13
13-D1. D = 3.09x 10-8 em 2/s

13-D4. M = 5.23
13-D6. a. Gelling starts at 18.93 em

b. J = 3.2x 1(Jl em/s
13-D7. M ~ 1.0
13-D9. a. C2p/C2w =0.771

b. Rp = 5.6 x 10-7 em
Chapter 14
14-D1. YEF =0 , W min = 308.7 eal/gmole feed

YEF =0.1 , Wmin = 556.25 eal/gmole feed
14-D2. Weq,cry = 5.27 Keal/kg feed

INDEX
A water treatment, 419-420

A term, 327-329 Adiabatic theory, 400-404, 407408
Absorption, 505,506, 517, 799, Adsorbents,222-228
802-807, 824, 831 Adsorption on crystals, 92-94
Acetaldehyde, 416 Adsorption, See also Specific topics,
Acetamide, 25 4-6,217-287,799,804,808,810,
Acetic acid, 247-251, 260-267 827-834
Acetonaphthalene,359-360 equilibrium, 228-239
Acetone, 408,413,418,433434, Affinity chromatography, 294-295,
711,713 345-346,518-521,799,804,831
Acetylene, 427-431 Affinity electrophoresis, 575-576,804
Acid recovery, 688 Agarobiose,568-569
Acidic resin, 455-459 Agarose, 220-221, 294, 483-484,
Acrylamide,567-568 569-570, 575,606
Activated alumina, 227-228, 300-305, Aging, 68
408-410,412 Air drying, See Drying
Activated carbon, 222-223, 225, 232, Airdome, 489-490
247-251,285-286,367,379-383, Airlift, 141
391-393,408,419-420,434-435, Air purification, 412-417, 434-435
810,830 Air separations, 226,425,654,797
capacity, 223,415-416 Alanine, 25
powdered, 503 Albumin, 754-758
pulsed bed, 510-511 Alkylamines, 433
solvent desorption, 417 Alum,145
solvent recovery, 412-417,499-500 Alumina, See Activated alumina
thermal regeneration, 260-267 Aluminum, 162
853
854
Amino acids 128,482, 552-553, 589 Azeotropic distillation, 795, 799, 802,
Ammonia, 416, 433 810,824,831-834
Ammonium bicarbonate, 24
Ammonium diuranate, 68 B
Ammonium sulfate, 33, 34, 68, 69, B term, 327-329
90-91, 101-102 Backftush, See Backwash
Ampholytes, 596-597, 660,602 Backwash, 489, 526
Amyl acetate, 416 Barium ac~tate, 21
Analytical chromatography, Barium hydroxide, 24
See Chromatography, Analytical Barium nitrate, 52, 62-67
Anhydrous,20,35 Basic resin, 455-459
Anions, 453, 461 Batch crystallization, 34, 105, 146-151
Anisotropic membrane, Bed utilization, See Fractional bed use
See Asymmetric membrane Benzene, 162, 172
Annular flow, 346-347 Benzene-sulfonic acid, 455-457
Answers, 847-851 Benzoic acid, 25, 186
Antibiotics, 343-344 BET isotherm, 231, 234-235
Anthracene, 39,162, 174-176, 185, Biochemicals, 342-346,482-484,490,
300-305,321-326,448-449 550,579
Antibodies, 575-576 Biogel, 220-221, 602
Antigens, 575-576 Biostream, 585-587
Antisolvent, 33, 69 Bis, 567-568
Arginine, 553 Bleeding, 292, 718
Argon, 231 Blood proteins, 69, 482-483, 688,
Arrhenius kinetics, 85, 97 690-692, 731-732
Arrhenius relationship, 229, 234 Blowdown, 423-424, 427-431
Artificial kidney, See Hemodialysis Boltzmann constant, 557
Asahi system, 510, 512 Bonded phase, 292-293
ASL equation, 97, 14 Bone char, 228
Aspartic acid, 553 Boric acid, 24
Astrakanite, 50-51 Boundary layer, 635, 666, 678-680,
Asymmetric membrane, 632-635, 651, 696, 734-735
667,673,682 Bovine serum albumin (BSA), 78,
Attrition, 437, 500, 536 237-239,553
Azeotropes, 710-711, 714 Brackish water, See Water, brackish
855
Breaktlrrough, 306,311, 366-370, Cellulose, 634-635, 691

Bromine, 797 Cellulose acetate membrane, 570,633,
Butane adsorption, 223 635,638,643,646-648,650,
Butanol, 413, 713-716 667-668,673-675,691, 712-716,
Butyric acid, 416 748, 778
Center feed columns, 178, 183
Centrifuge, 151,799,803
c Cesium chloride, 585
C term, 327-329 Channeling, 219,489
Calcium acetate, 21,24 Characteristic temperature, 408
Calcium carbonate, 747 Charcoal, 230
Calcium sulfate, 747 Chemical modification, 799-800
Canister systems, 414, 435-437, 446 Chem-Seps process,
Capacity, 369,456-458 See Higgins process
Capillary chromatography, 291, 799 Chlorine, 416, 419-420, 671
Capillary condensation, 223,415 Chlorobenzene,418
Capillary electrophoresis, Chloroform, 418
See Electrophoresis, capillary Chromatofocusing, See focusing,
Caprolactam,672 Chromatography, See also specific
Carbon adsorption systems, topics, 4-6,217,288-364,807-808,
See Activated carbon and 810,825,827-828,830-831
Carbon molecular sieves analytical, 288-296, 306, 316
Carbon dioxide, 226, 293, 367, 408, large-scale, 306, 320, 330-331, 336,
643,646-651,831 341-347
Carbon monoxide, 649-651 linear, 297, 306-334
Carbon molecular sieves, 225, 804 nonlinear, 298-306
Carbon tetrachloride, 223, 418 resolution, 319-320
Carman-Kozeny equation, 385, 679 scaling, 389
Carrier ampholytes, See Ampholytes solute movement theory, 296-305,
Carriers, 649 395-400
Cascades, membrane, theories, 296-335, 347
See Membrane, cascades Chung and Wen correlation, 312
Casein, 680-681, 787 Citric acid, 25, 87, 91
Cations, 452, 461 Classification, See Separations,
Cellophane membrane, 711-713, 789 classification
856
Classified product removal, 138-140, Copper sulfate, 20, 21, 24, 36-38,
155-156, 158 692-693
Clatlrration,799,803 Costs, 796-798, 808
Clay, 228 carbon adsorption systems, 415
Cleaning, See Membrane, cleaning large-scale chromatography, 437
Clear liquor advance, 140 membrane, 641-642, 654, 672, 700,
Cloethe-Streat contactor, 500-503, 714
534-536 Counter ion, 452
CO2 , See Carbon dioxide Counteracting chromatographic
Cohn equation, 68 electrophoresis, 605-607
Cohn fractionation, 69 Counter-current chromatography,
Co-ion, 452 295-296
Colburn equation, 517,520-521 Counter-current distribution, 295-296,
Colloids, 678, 687, 745-748 314
Column crystallization, 176-183 Countercurrent moving bed, See
Column design, 330-331, 344-345, Moving bed
383-393,437-438,489-490,537 Countercurrent processes, 62-67,
Column diameter, 330, 387-388,438 521-524,587-589,758-762
Column-in-series, 419420 Coupled systems, 365, 393405
Column switching, 340-341, 828 Cross-over ratio, 403404
Compression, 338 Cryohydric point, 41
Concentration polarization, 635-636, Crystal birth, 141-144
638,657-666,676-685,688-689, Crystal death, 141-144
696-697, 703-704, 733-745, Crystal dissolution, 93-94, 98
749-758, 762-770, 773-774 Crystal growth, 92-98, 122-126
Congo red, 619 Crystal habit, 81
Congruent melting, 41 Crystal morphology, 2, 80-83
Constant pattern, 247, 305, 366, Crystal shape, 80-83
370-375, 379-388, 394-395 Crystal size distribution, 2,80, 103-159
Contact breeding, 87-88 Crystallization, 799, 803, 808,828-830
Controlled cooling, 146-151 batch, 34,105,146-151,169-176
Convection,562-563,577,584-585 continuous,32,34
Cooling curve, 146-151 equilibria, 19-26,38-52,161-163
Copper,482,692,693, 718, 797 fractional, 52-67, 77, 79, 128
Copper chloride, 24 kinetics, 84-98, 122-126
857
knowledge structure, 3 Cyclohexane, 162,448-449

melt, 2,161-204,818-819 Cyclone, 799, 806
methanol effect, 33 Cyrogenic distillation, 797, 827
nucleation, 84-93 Cytochrome C, 553
pH effects, 69-70
reaction, 34, 68 D
solution, 1,2, 19-159 Dead volume, See Extracolumn effects
T effects, 20-25,46-47,50-51, DEAE Sephadex, 237-239
53-59, 70, 97 Debye-Huckel constant, 556-557
ternary, 45-67 Decanter, 799,806
Crystallizers, Deionization, See Water
Brodie, 178-179 demineralization
column, 176-183 Delay step, 426
cooling, 2, 26-28, 34-36 Demister, 799, 803
evaporative, 2, 28-31, 36-38 Demineralization, See Water
modifactions, 128-142 demineralization
Pachuca, 141 Dendrites, 81, 83, 87
piston, 176, 178 Density gradient electrophoresis, See
reactive, 34 Electrophoresis, density gradient
scale-up, 142 Density, packing, 219-220, 223,
scraped surface, 27,141 226-228,301,380,410-411,456
spray, 31 Depressurization, See Blowdown
staged,I66-176 Desalting, 658-659, 668-670, 694-695,
vacuum, 2,29,32, 36-38,45 700-702
yield, 34-38, 48-49, 118-119 Design, See Column design and
Crystallography, 82-83 Membrane, design
CSD, See Crystal size distribution Desorbent regeneration, 431-434
Cumulative area, 109-111, 113 Desorption, 386, 405-434
Cumulative length, 108-109, 111, 113 Dextran, 730, 786-787
Cumulative mass distribution, Dextran gels, See Sephadex
110-113,117-121,133-137 Dialysis, 687-693, 799
Cumulative number crystals, 104, Dichlorobenzene, 161
107-108,111,1113 Dichloroethane, 418
Current utilization, 698-699 Diethylether, 418
Cut, 644-645, 665, 707-709, 760-761 Difference point, 63-66, 75
858
Differential mass distribution, 112-113 Donnan dialysis, 692-693

Diffuse tails, See Diffuse waves Donnan exclusion, 454-455, 463-464,
Diffuse waves, 244-245, 247-251, 484-487
262-263,298-305,386,398,405, Double layer, 555-558
407,464,466-469,473-475 Double salt, 50-51, 60-62
Diffusion, 269-270, 273, 328, 331-334, Drainage, 171-176
377-383,571,598-599 Drierite, 409
Diffusional separation, 798-800 Driving force, 631,671,688,693,
Diffusivity, 382,488, 573, 649-650, 705,710
704-705 Drying, 151,225-227,408-412,
Dinitronaphthalene,359-360 418-421,425,799
Direct mode, 251, 255-256 Drying up point, 47, 53
Disc gel electrophoresis, See Dual-temperature exchange, 799, 804
Electrophoresis, stacking gel DVB, See Divinylbenzene
Displacement chromatography, Dye recovery, 687
339-340,355,395-401,497-498,
589,591-592
Displacement electrophoresis, E
See isotachophoresis e, Electron charge, 555, 557
Dispersion correlation, 312 Economics, See Costs
Dispersion model, 308-313, 322-327, Effective diffusivity, 270, 273, 379,
341,356,486-487,617 488-389
Dissolution, See Crystal dissolution Effective distribution coefficient, 193
Distillation, 166, 178,347,433-434, Effective mass transfer coefficient,
506,524,685-686,710-711,714, 376-378,391-393
795,801,802,807-810,815-821, Efficiencies, 698-701, 815-816
824-838 Electric charge, 804
Distribution, 437, 802 Electric potential, 555, 693
Distribution coefficient, 162-163, 193, Electrical terms, 551
485 Electrochromatography, 550,605-607,
Divalent ion exchange, 454, 460-463, 799,801,804
470-481 Electrodes, 551-552,694-695
Divinylbenzene, 455-458 Electrodialysis, 635,692-702,799,
DNA, 552, 569, 577, 580 804
Dominant size, 113, 119, 124, 146 Electrolytes, 589-590, 692
859
Electroneutrality, 453-455, 484485, 571-573,591-594,598-601,606,

487-488,591-592,696 804
Electroosmosis, 555,563-564, Electrophoretic retardation force,
571-572,578 555-557
Electrophoresis, 6, 320, 547-589, Electrostatic separator, 799, 806
617-618,799,804,808 Emulsion liquid membrane, 717-718
capillary, 577-579 Emulsion separator, 799
continuous, 582-590, 593, 602-604 Enantiomorphs, 83
density gradient, 585,603-604 End feed columns, 176-178, 180-183
disc gel, 568-569 Energy balance, 273-274, 277, 709-710
displacement, See Isotachophoresis Energy criteria, 808, 812-822
free solution, 576-579, 584 Energy reduction, 798
gel, 565-570, 573, 584,602 Energy separating agent, 801-806,
helical flow, 587 808,826
magnetic flow stabilization, Energy wave, See Thermal wave
586-587 Enthalpy-composition diagram,
membrane, 565,570 42-45,201
microscope, 577 Entrainment, 172-176, 181-184
moving boundary, 564-565 Equilibrium data, 38-52, 162,228-239,
Page, See Electrophoresis, 412,459463,466
polyacrylamide gel Equilibrium isotherms, 228-239, 412
paper,565,570,580,583-584 Equilibrium staged models, See
pH gradient, See Isoelectric focusing Staged models
polyacrylamide gel, 567-569, 573, Equilibrium staged separations, 1, 2,
597,602 795,807,823,828
pore limit, 568, 580 Equivalent work, 815-822
pulsed gel, 581 Erf, See Error function,
recycle, 586-589,604-605 Error function, 309-310, 333-334
rotating, 585, 587 Escherichia coli, 553
stacking gel, 568-569,589 Esters, 416, 799
starch gel, 570, 580 Ethanol, 404, 413, 418,685-686,
two-dimensional, 578-589, 593, 711-714,809-812,832-834
602-605,618 Ethylmercaptan, 416
zonal, 565-567, 570-574 Ethylacetate, 416, 418
Electrophoretic mobility, 552-560, Ethylbenzene, 814-815
860
Ethylchloride, 230 Fines removal, 128-138, 155-156

Ethylene, 230, 232, 358-359 Fines trap, 129-130, 132, 140
Ethylene glycol, 485 Fingerprinting, 580
Eucalyptole, 416 Flash distillation, 37, 799, 802
Eutectic, 39-45, 180 Floating zones, 187
Evans blue, 619 Flotation, 799, 806
Evaporation, 685-686,799,802,831 Flowsheet, 835-838
Evolutionary methods, 823, 835-838 Fluid shear breeding, 88
Exchange adsorption, 232, 236-237, Fluidization, 437, 535-536, 546
401 Fluidized beds, 499-510,514,534-536
Exclusion, See Sterie exclusion and Flux, membranes, 631, 643-644,
Donnan exclusion 657-669,674-675,677-681,
Expression, 799 688-689,703-705, 736, 743,
Extracolumn effects, 319, 330-331, 748-758,770,773-781
389-391 Foam fractionation, 799, 804
Extraction, 63, 295, 524, 718-719, 799, Focusing, 257-258, 264-266, 286-287,
802,804,810,822,824,828,831, 338-340
833, 835 Food processing, 642, 683, 685-687
Extractive distillation, 795, 799, 802, Formaldehyde, 416
831,836,838 Fouling, 458, 636, 638, 676, 682-683,
733, 747-748
F Fractional bed use, 368-370, 385-388
Facilitated transport, 718-719 Fractional crystallization, 52-67, 77,
Faraday, 697-698 79, 128
Fatty acid, 185,228 Fractionation systems,
Favorable isotherm, 233, 246 continuous adsorption, 521-524
Feed concentration, 797-798 simulated moving beds, 524-533
Ferric chloride, 24 Free-flow electrophoresis, See
Ferrous chloride, 24 Electrophoresis, free-flow
Fickian diffusion, 269-270 Freeze drying, 799
Field step method, 592 Freezing point depression, 567,
Film mass transfer, 271-272,274,328, 740-741
331-334,376-384,391-393,666, Freon, 418
679-681 Freundlich isotherm, 231,236,243,
Filtration, 151,670-671, 799, 803 247-251,260-267
861
Friction factors, 554-558, 737, 779-782

Gel formation, membranes, 636,
Frontal analysis, 565 676-685,689,733, 745-747,
Fructose, 25, 284, 286-287,362 753-758
Gel permeation chromatography, See
Fructose-glucose separation, 284, 362,
525-526, 529-533 Size exclusion chromatography
Fruit juice, 685-687 Gel resin, 455-456
Fundamental equation of General characteristics separations,
chromatography, 319 795-798
Gibbs phase rule, 38,43, 52
Glass transition temperature, 633
G Glassy polymers, 633, 649
Gallium arsenide, 162 Glauber salt, 61, 158
Gas adsorption, 402-418, 421-433, GLC, See Gas-liquid chromatography
435-438 Glucose, 25, 284, 361-362,484-487,
Gas chromatography (GC), 289, 291, 498,507-510
328-329, 799,804 Glutamic acid, 553
Gas diffusion, 799, 804 Glycerol, 485
Gas-liquid chromatography (GLC), Glycine, 568-569
232,237,289,291,405,799, Gold, 482, 503, 797
802,831 Gradient elution, See Programming
Gas masks, 435 and Gradients
Gas permeation, 631, 638, 640, Gradients, 36-340, 482-484, 827
643-654,726,758-762,773-777, Graphical integration, 516-517
799,831 Grashof number, 562, 584-585
Gasoline, 416 Growth of crystals, 92-98, 122-126
Gas solid chromatography (GSC), 289, Growth rate dispersion, 94
799,804 Guard bed, 314,414,436
Gaussian distribution, 313, 315-326,
341,566-567,570-574,582,599 H
Gelconcentration,cg ,678-681, H,SeeHETP
745-746 Hagen-Poiseuille equation, 780-781
Gel electrophoresis, See Hard water, See Water softening
Electrophoresis, gel Hauy's Law, 80-81
Gel filtration, See Size Heat of adsorption, 273-274,400-404
exclusion chromatography Heat transfer, 146-148, 163-165
862
Heatless adsorption, See HTU analysis, 180-183,514-521,539

Pressure swing adsorption Human serum albumin, 518-521
Heel, 407, 424-425 Humic acid, 420
Height equivalent to theoretical plate, Hybrid processes 654, 685-687,
See HElP 710-711,714,822,826,831
Helium, 361 Hydrate, 20, 34-38
Hemodialysis, 688, 690-692 Hydrogen, 358-359,402,425,432,
Henry's law, 229-230, 237,649-650, 643,649-651,653,831
705 Hydrogen sulfide, 226, 416
Heterogeneous nucleation, 84, 86-92, Hyperfiltration, See Reverse osmosis
123-125 Hydrophobic chromatography, 805
Heterogeneous separations, 799-800 Hypersorber, 499, 524, 546
HETP, 314, 318, 327-330, 506
Heuristics, 823-837 I
Hexadecanol, 162 Ideal gas, 380,421,646-648,812
Hexane, 344-359,433 IEF, See Isoelectric focusing
Higgins process, 512-514 Immobilized pH gradient, 597
High gradient magnetic separation Immunoelectrophoresis, 575-576, 580,
(HGMS), 799, 806 804
High performance liquid Impingement separator, 799
chromatography (HPLC), 292, 799, Incineration, 417, 506
831 Incongruent melting, 41
Histidine, 553,600-601 Initiator, 567
Homework, 9,10, See also Answers Initial breeding, 87
Hollow fibers, 636-640, 643, 667-669, Intensification, See Scaling
675,682-683,688,703,737,739, Interacting systems, See Coupled
746, 762-770 systems
Homogeneous nucleation, 84-86, Intermittent moving bed, See Pulsed
123-125 moving bed
Homogeneous separations, 799-800 Interparticle porosity, See Porosity
Hot gas desorption, See Thermal swing Intraparticle porosity, See Porosity
adsorption Inverted solubility, 22, 28
HPLC, See High performance liquid Ion exchange chromatography,
chromatography 289-290,295,482-484,799,804,
H2 S, See Hydrogen Sulfide 830
863
Ion exchange dialysis, 692-695 Kelvin equation, 85

Ion exchange, 217-218, 452-498, 799, Key relations chart, 9
804 Kidney, artificial, 688
continuous operation, 500-503, Kiln, 417, 421,510
510-514 Kinetic adsorption, 225, 799-800, 804
equilibrium, 459-463, 466 Kremser equation, 506, 508-510
mass transfer, 487-489
resins, 284, 455-459 L
Ion exchange membranes, 633, 692 Labile, 22,23
693, 771 Lactoglobulin, 553
Ion exclusion, 454-455, 484-487, 498, Lactose, 25
799,804 Laminar flow, 738-739, 746, 762-770
Ion movement, 463-481, 484-485 Langmuir isotherm, 229-230, 232-235,
Ion wave, 453, 465,468,471-475,480 237-239,247,305,372-374,
Ionic equilivalent conditivities, 559-560 379-383,461
Iron, 162, 747 coupled isotherms, 234-235, 394
Irreversible thermodynamics, 770-782 Lapidus and Amundson model,
Isocratic, 320 308-313,322-327,356,486-487,
Isoelectric focusing, 550,580, 594-605, 617
619-620, 799, 806 Large-scale chromatography, 306,320,
Isoelectric point, 70, 552-553, 594-595, 330-331,336,341-347,483-484
597,600,806 Latent period, 67-70
Isomorphous, 83 Layered beds, 409
Isopropanol, 418, 711 Leaching, 799, 803
Isotachophoresis, 550, 589-594, Lead chloride, 24
618-619,799,804 Lead nitrate, 24, 52, 62-67
Isotachophoretic condition, 591 Length of unused bed, See Mass
Isotherms, See Equilibrium isotherms Transfer Zone
ITP, See Isotachophoresis LC, See Liquid chromatography
Lever arm rule, 40, 44, 55-58
J Limiting current density, 697
j-factor, 737
Linear chromatography See
Joule heating, 561-563, 566, 578,590
Chromatography, linear
K Linear dispersion model, See
Kd ,220 Dispersion model
864
Linear growth rate, 88-89, 96-97, Magnesium, 797

106-107 Magnesium carbonate, 747
Linear isotherm, 299-230, 237, 288 Magnesium chloride, 24, 33
solute movement with, 241-243, Magnesium sulfate, 33, 50-51, 60-62
423-431 Magnetically stablized fluidized beds,
sorption effect, 405 514,518-521,524,536-537,545
Linear parafins, See Straight chain Magnetic separator, 799
hydrocarbons Magnetic resin, 514, 518
Linear systems, 271, 288,306-335, Manganese nitrate, 41
485-486 Maple syrup, 28
Linear theory of chromatography, 288, Mass balance, 666, 679, 734-735
297,306-334 Mass distribution, See Cumulative
Liquid adsorption chromatography, mass distribution
291-292 Mass ratios, 63-66, 503-505
Liquid chromatography (LC), 291-296, Mass separating agent, 801-806, 808,
799,804,831 826-827,836-837
Liquid-liquid chromatography, 292, Mass spectrophotometer, 799
799 Mass transfer, 67,94-98, 163-165,
Liquid membranes, 716-719, 799, 268-272,275-276,454,487-489,
802-804 514-421,666,688-691
Lithium carbonate, 24 Mass transfer correlations, 376-383,
Loading ratio correlation (LRC), 235 518, 737-745, 755, 764-770
Local equilibrium theory, 266-268, Mass transfer resistance, 328, 458, 689
275-277 Mass transfer zone (MTZ), 365-370,
Loeb-Sourirajan membrane, See 375,384,402,404,407,421,497
Asymmetric Membrane McCabe-Thiele diagram, 64-66,
Lower explosion limit, 413 168-169,503-510
LUB, See Mass transfer zone, McCabe M.law, 96-97,102,122
Lumped parameter, 271-272,274, Mechanical separations, 798-799
372, 376,488-489 Median size, 112-113, 119
Lysine, 553 Medical applications, 435
Melt crystallization See
Crystallization, melt,
M Melting, partial, See Partial melting,
Macroporous resin, 455, 484 Melting point, 803
865
Membrane Separations, 623-789, See Methanol, 33, 320, 336-337,413,418,

also particular type 433
Membranes, 6,7, Methyl cellulose, 585
batch, 683-685 Methyl chloride, 416
cascades, 641-642,652-653, 710, Methylene chloride, 418
808 Methylethyl ketone, 413
ceramic, 672 Microfiltration, 803, 831
cleaning, 637-639, 642,671,683, Miers diagram, 22, 23
747-749 Migration veocity, 552, 554-561
coflow, 691, 701 Migrational chromatography, 243,
composite, 667-669, 714 293-294
costs, See Costs, membrane Miller indices, 82
countercurrent, 641-642, 648, Minimum work, 812-815
690-691,758-762 Mixed module, See Membrane,
design, 636, 753 mixed module
flux data, 658, 668-669 Mobility, 487-488, 774-775, 779,
ion exchange, 633, 692-695, 771 See also Electrophoretic mobility
life, 671, 748-749 Module, membrane, 636-643, 733
mixed module, 640, 644-648, Molasses, 342-343
661-665,691,707-709,714-716, Molecular distillation, 799,804,810
749-758, 760-761 Molecular sieves, See Zeolite
module, See Module, membrane Molecular weight cutoff, 673-676
permeability, See Permeability Molecular weight data, 553
recycle, 641-642, 652-653, 683, Moment analysis, 111-113
687,710-711,714,728-729 Monolayer coverage, 229-234
resistance model, 651-652 Monovalent ion exchange, 452-453,
sieving, 673, 676,770-771, 778-782, 459-462, 464-469
786 Moving bed, 499-524, 530, 534-538,
Membrane chromatography, 346-347 543-546
Mercaptoethanol,569 comparison with 5MB, 530-531
Metal ions, 482, 501,692-693, 700 magnetically stabilized, 514,
Metastable region, 22, 23, 126, 149 518-521,524,536-537,545
Methane, 285-286, 361, 379-383, staged, 499-510, 536
391-393,402,408,449-450, Moving boundary electrophoresis,
646-650 564-565
866
MSMPR,I06-113 o
M1Z, See Mass transfer zone (M1Z) Olefin, 525
MWB process, 185-186 On-off chromatography, 294-295,
Multilayer adsorption, 234 299-300,345-346,482-484
Multistaging, 807-808 On sager relationship, 771
Myglobin, 553, 600-601 Optical isomers, 83, 128
Myosin, 553 Optimization, 383-393
Osmosis, 654-657, 673
N Osmotic concentration, 673
Nafion, 634-635, 694 Osmotic pressure, 654-657, 673, 677,
Nanofiltration, 669,672 684,686,740-741,746,772,780
Naphthalene, 39,40, 162, 172, Ostwald-Freundlich equation, 85
185-186 Ovalbumin, 553
Naphthol,162 Oxygen, 649-650, 760-761, 797
Natural rubber, 649-650 Ozone, 416
Navier-Stokes equations, 762-763, 766
Needle breeding, 87 p
Nernst-Planck equation, 487 Packed beds,
<N"H4h S04, See Ammonium sulfate breakthrough, See Breakthrough
Nitrogen, 231,432, 643, 649-650, 653, canister systems, See Canister
760-761 Systems
Nomenclature, 13-18,207-216, desorbent regeneration, See
547-549,551,623-629,793-794 Desorbent regeneration
Nonisothermal theories, 365, 375, energy balances, 273-274
400-404,407-408,506,518 intensification, 383-393
Nonlinear isotherm, 229-239 mass transfer equations, 268-272
solute movement with, 242-251, nonregenerative design, 435-437
298-305,407 particle diameter, 344-345, 383-393
Nonlinear theories, 365-375, 379-405, regeneration, 365, 405-434
507,523 scaling, 383-393
Normal freezing, 161, 171, 183-186, schematic, 218, 365
189,192,194-196 simulated moving beds, 524-533
Nuclear industry, 435 solute movement theory, See Solute
Nucleation, 84-93, 122-126 movement theory
Nylon, See Polyamide thermal regeneration, 406-417
867
PAGE, See Electrophoresis, Plate-and-frame, 637-638, 640,

polyacrylamide gel 668-669,674-675,683,688,692,
Paper chromatography, 295, 605, 799, 694,699-700,738,746,762-765
804 Plate generation, 573
Paper electrophoresis, See Plate height, See HETP
Electrophoresis, paper Plate theories, See Staged theory
Parex, 177 Plateau, 394-395
Partial melting, 171-176, 184-186,200 Point fines trap, 132
Pb(N03h, See Lead nitrate Poiseuille flow, 780-781
Peak maximum, 313, 317, 326 Poisson distribution, 315
Peak width, 318-319 Poisson-Boltzmann distribution, 556
Peclet number, 310-312, 317 Polarity, chemical, 336-337
Penicillin, 797 Polarity reversal, 700
Pentane, 394-395,433 Polarization, See
Permeability, gas, 631,643-644,646, Concentration polarization
652, 776 Polarization Modulus, 661-666,
Permeability, liquid, 657-665, 679, 677-678, 734-736, 765-770
680-681,688,705,774-775 Polishing, 481-482
Permittivity free space, 551 Polyacrylamide gel electrophoresis,
Perstraction, 716 See Electrophoresis,
Pervaporation 703-716, 728-729, 799, polyacrylamide gel
803,810,827,833 Polyacrylic acid, 456-459
pH effects, 69-70, 286-287, 289 Polyacrylonitrile, 634, 673, 675, 691,
pH gradient, 286-287,336,338-339 712
pH marker, 553, 598 Polyamide, 634, 638, 667, 669, 673,
Pharmaceuticals, 342 713
Phase equilibria, See Equilibrium data Polycarbonate, 633,673
Phenanthrene, 39 Polyether block amide (PEBAX), 703
Phenol,39,40, 162,420,433-434,718 Polyfuran, 669
Phenomenological coefficients, Polymeric adsorbent, 227-228,433
770-771, 773-776, 779 Polymeric resin, 433, 455-459
Phillips process, 176-177 Polymorphism, 81
Philpot-Harwell device, 585-587 Polynucleotide, 550, 552-553
pI, See Isoelectric point Polypropylene, 713
Plasma proteins, See Blood proteins Polysaccharide, 678, 745
868
Polystyrene, 455-457, 465, 475, 484, Programming, 336-340

635,650,694 Propane, 282-283
Polysulfone, 634-635,638,651-652, Proportional pattern, 366-370, 375
673-675, 712-713 Propylene, 230, 232
Polyvinylalcohol,712 Proteins, 68-70, 289-290, 293-294,
Polyvinylchloride, 673 345-346,482-484,490,503,
Polyvinylidene fluoride, 674 518-521 550,552-553,559,569,
Ponchon diagram, 42-45, 62-67,201 573-574,577-578,589,592,
Population balance, 2, 103-159 600-601,606-607,617-620,680,
Population density, 89, 104-108, 116, 745
125, 126, 131, 139-140, 144-145 PSA, See Pressure swing adsorption,
Pore diffusion, 269-270, 328, 332-334, Pulse solution, 306-307, 313, 323
377-384,391-393 Pulsed moving beds, 500-503,
Porosity, 218-219, 223, 226-228 510-514,537
Potassium chloride, 24, 33, 59-60, Purge, 422-423
90-91,117-121,133-138,156-158 Purge gas stripping, 431-433
Potassium chromate, 25 Putrescene, 416
Potassium nitrate, 25, 45-49,53-54 Pyrene, 162, 174-176
Potassium sulfate, 25, 90-91, 102, 145 Pyridine, 418
Precipitation, 19,67-70, 799, 803
Preface, v, vi
Preparative-chromatography, See
Q
Quaternary ammonium, 456-458
Large-scale chromatography,
Prerequisites, v, vi
Pressure diffusion, 799 R
Pressure drop, membranes, 631-633 Radial Flow, See Annular flow
Pressure drop, sorbents, 344, 385-387, Radioactive compounds, 435
414,437-438,449,535 Radium, 797
Pressure equalization, 426 Rate processes, 631
Pressure swing adsorption, 421-432, Rayleigh convection, See Convection
827, 832-834 Rayleigh distillation, 201
Primary nucleation, 84-86, 123-125 Rayleigh number, 562, 584-585
Proabd process, 171, 185 Rayon manufacture, 680
Problem answers, 847-851 Reactive distillation, 799, 802
Problem solving, 7-10 Reciprocal salt pairs, 52
869
Recycle, 340, 342-343, 641-642, Rosen solution, 271,331-334,363,

652-653 377,487
Recycling electrophoresis, See Rubber, See Natural Rubber
Electrophoresis, Recycling Rubbery polymers, 633, 649
Reflux, 176,421
Reflux ratio, 66
Regeneration, 365,405-434 S
desorbent, See Desorbent Salt, common, 28, 700
regeneration Salting in, 23, 26
ion exchange, 456, 465-481 Salting out, 23, 26, 29, 33,45,68-69,
moving bed, 504-506, 545 91
purge gas, See Purge gas stripping Salts, solubility, 24-26, 68
thermal, See Thermal regeneration Saturation, 22-23
Rejection coefficient, 659-661, Scaling, 383-393, 437-438
667 -669, 677-678, 681-682 Scraped surface crystallizers, 27, 141
Relative retention, 319-320 Screen analysis, See Sieve analysis
Relaxation effect, 558 Scrubber, 799
Repressurization, 423-424, 427 -431 Screw dislocation, 93-94
Resin, See Ion exchange resin SDS, See Sodium dodecyl sulphate
Resistance, electrical, 697-698 SDS-PAGE, 569, 579
Resolution, 319-320, 573-574,583, Sea water, 670, 700, 777-778, 797
599-601 SEC, See Size-exclusion
Retention, 676 chromatography
Retention time, 296-297,571-572,583 Secondary nucleation, 84, 86-92,
Reverse osmosis, 638, 640, 654-673, 125-126
685-687,711,739-745,747-758, Sedimentation, See Settling
762-770,773-778,799,807,810, Seeding, 86-87,126-128,146-151
820-822,831,833-834 Selection, 7, 791-795
Reverse-phase chromatography, 292, Selectivity, 297-298, 319-320,
342-344 460-463,644,659-660,706-707,
RNA, 552, 577 712-713
Rinse step, 426 Separation factor, 236, 706, 801, 807
RO, See Reverse osmosis Separations, 1-851
Rocket electrophoresis, 575-576 classification, 798-806
Roll-up, 395 diffusional, 798-800
870
equilibrium staged, See Sink-float, 799, 806

Equilibrium staged separations Size dependent growth, 97, 144-146,
general characteristics, 795-798 158
mechanical, 798-790 Size exclusion, 220-221, 225, 293-294
overview, 798-812 Size exclusion chromatography (SEC),
scale, 796-797 293-294,343-344,799,803
Sephadex, 220-221,293-294,602,605 Skarstrom pressure swing adsorption,
Sepharose, See Agarose 421-423,427-431
Sequencing, 7, 791-795 Skatole, 416
Settling, 534-535,799, 806 Sloppy split, 828
Shanks system, 525 Slurry crystallization processes, 161,
Shape factors, 110-111, 150 165-183
Shear boundary, 555-556 5MB, See Simulated moving bed
Shock wave, 245-251,264-266, Snyder equation, 330
298-305,368,370,383,396-400 S02, See Sulfur dioxide
405,407,464,466-469,476-481 Sodium acetate, 21, 25, 34
Sieve analysis, 113-122, 127-128 Sodium bicarbonate, 25, 68
Sieve tray columns, 499-503, 534-536 Sodium carbonate, 28,42-59, 77,
Sieving membrane, See Membrane, 661-665
sieving Sodium chromate, 21
Sieving, electrophoresis, 563-569, 584, Sodium chloride, 21, 25, 28, 33, 54-60,
602 91,478-479,658-659,668-669,
Silica, 292, 747 700-702, 739-745, 787-788
Silica contamination, 343 Sodium dodecyl sulphate, 568-569, 806
Silica gel, 226-227, 230, 232, 282-283, Sodium hydroxide, 25
359-360,408-409,411-412,434 Sodium nitrate, 23, 25, 45-49,53-54
Silicon, 187 Sodium sulfate, 21,25,33,50-51,
Silicone rubber, 633-634,650-652, 60-62, 145, 158
713, 760-761 Solid solutions, 39, 52, 62-67
Silver, 482 Solubility, 803
Silver acetate, 25 crystallization, 19, 26, 68-70
Silver nitrate, 25 membranes, 649, 705
Simpson's rule, 195,517,520 Solubility product, 26
Simulated moving bed (SMB), 499, Solute movement theory, See also
524-533, 545 Local equilibrium theory, 239-268,
871
296-305,347,395-400,403,407, Staged theory, See Staged model

423-431 Steam distillation, 799
assumptions, 266-268 Steam regeneration, 412-416
co-flow and counter-flow Steric exclusion, 220-221, 225,
regeneration, 300 434-435
formal mathematical development, Straight chain hydrocarbons, 434-435,
275-277 525
fractionation, 521-524 Strategy of design, 822-824
ion exchange, 463-481, 484-485 Stirred cell, 630-631, 636, 738
pulsed bed, 511 Stirred tanks, 419, 482,503,536
simulated moving beds, 526-533 Stokes Law, 544, 567
solvent recovery with activated Stripping, 799, 802-805
carbon, 414-415 Strong resin, 455-458
solute wave velocity, 241-251, Styragel,220-221
281-282 Styrene, 814-815
thermal wave velocity, 252-254, Sublimation, 799, 802,
403 Sucrose, 25, 102,342-343,347,585,
Solution-diffusion membranes, 667, 603,656,789
718-719,770,773-778 Sugar, 87,226,228,452,475,481,
Solvent programming, See Gradients 512,672,686-687
Solvent recovery, 412-417, 499-500 Sulfur, 797
Sorbex, 525-526 Sulfur dioxide, 416
Sorption, 3-6 Sulfuric acid, 484-487
Sorption effect, 405 Supercritical chromatography, 293
Space, electrophoresis applic., 563, Superloading, 433-434, 446-448, 825
584-585 Superposition, 306-308, 313, 323, 332,
Spacers, 592 358,486-487,682,705
Spiral wound, 636-638, 640, 643, Supersaturation, 22-23, 28, 67-70,
667-669,674,675,682,703,739 85-86,93-94, 148-149
Staged adsorber, See Moving bed, Supported liquid membrane, 717-718
staged Surface diffusion, 270-271, 377-379,
Staged crystallizer, See Crystallizer, 384
staged Suspended solids, 140-142
Staged model, 1,2,313-316,503-510, Sweating, 171-176
529,536,543-544 Swelling, 456, 458, 489-490
872
T Tortuosity, 223, 226-228, 270, 379,

Tank crystallizer, 141 780-782
Temperature profiles, 402, 427 Total ion wave, See Ion wave
Temperature programming, 146-151, Transference number, 695-696
336, 362-363 Traveling wave, 252-266
Temperature swing adsorption, See Trivalent exchange, 495-496
Thermal swing adsorption Tubular membrane, 638,668,675,
Tert-amine, 456-458 683,737-746,762-770
Thermal convection, See Convection Tubular pinch effect, 747
Thermal diffusion, 273, 799,806 Turbulent flow, 737-746
Thermal desorption, See Thermal Two-dimensional methods, 578-590,
regeneration 593,602-605,617-618,620
Thermal regeneration, 229 Two-dimensional nucleation, 92-93
gases, 256-257,406-416 Tyler mesh sizes, 114
liquids, 259, 418-421
drying, 409-412
solute movement analysis, 253-266
u
UF, See Ultrafiltration
solvent recovery with activated
carbon, 412-417 Ultracentrifuge, 799, 806
Ultrafiltration, 638, 640, 673-687,
steam, 412-417
Thermal swing adsorption, 406-412, 745-758,770,799,803,831
419-421,819-821, 830,832-834 Ultrapurification, 161, 183, 187, 778
Thermal wave, 251-266, 356,402-403 Unfavorable isotherm, 232, 405
linear isotherms effect on, 255-256 UNIFAC, 39
regeneration, 253-266 Units, 551, 697
solute, effect on, 253-266 UOP, 525-526
velocity, 252-254, 261, 264 Uranium, 482,501-503,543-544, 797
zone spreading, 356 Urea, 121-122, 156-157
Thin-layer chromatography, 295, 799,
804 V
Thomas solution, 335, 375 Vacancy chromatography, 358
Three-phase, 518, 799 Vacuum desorption, 416-417, 421,
Tin, 162, 194-196 426,446
TLC, See Thin-layer chromatography Vacuum distillation 524, 799, 802,
Toluene, 413, 416 827,831
873
Vacuum swing adsorption (VSA), 421, Weak resin, 455-459

424-426,432,827 Whey, 672,685-686,700
Van Deemter equation, 326-331, 363
Vant'Hoff equation, 21, 656, 756
x
Vapor pressure, 655, 661,802 Xylene, 160-161, 176-177,226,418,
Vermiculite, 435
525, 845-846
Velocity changes, 404-405
Viscosity, 381-382, 555, 742
y
Vitamin B-12, 797
Volume average, 269, 273 Yield calculations, 34-38,48-49,
VSA, See Vacuum swing adsorption 118-119
W Z
Wall effect, 330 Zeolite, 220-221, 224-226,360,
Washing, 176, 180-183, 799 394-395,408-409,411-412,
Wastewater treatment, See water 418-419,434-435,525,799,
treatment 803-804, 810
Water, brackish, 670, 672, 700 Zeolite molecular sieves, See Zeolite
Water demineralization, 452, 481-482 Zeta potential, 55-561
Water dome, 489-490 Zone melting, 161, 183, 186-196,799,
Water softening, 217-218, 452, 462, 803
471,475-481,489,672 Zone refining, See Zone melting
Water treatment, 419-421, 435, 447, Zone spreading, 305-306, 317, 329,
503,510-512,670-672,717 336-338,356,471,550,570-574,
Water vapor, 408 577,581-583,597-601

Dokumen - Tips Rate Controlled Separations

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

Dokumen - Tips Rate Controlled Separations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dokumen - Tips Rate Controlled Separations

Uploaded by

Copyright:

Available Formats

RATE-CONTROLLED SEPARATIONS

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

© 1994 Springer Science+Business Media Dordrecht

ISBN 978-94-010-4585-8 ISBN 978-94-011-1342-7 (eBook)

Separations have always been very important in chemical engineering.

The development of new industries requiring the expertise of chemical

A major question for chemical engineering education is what to teach. In

My first book, Equilibrium-Staged Separations, was my response for the

This book, Rate-Controlled Separations, includes separation processes

chromatography when I audited his course. My interest in problem solving was

Finally, my wife Dot has supported me when I thought I would never

1.1. Part I: Crystallization.......................................................................... 1

Nomenclature Chapters 2 to 5 ............................................................................. 13

2. Crystallization and Precipitation from Solution - Equilibrium Analysis ...... 19

2.1. Solubility .............................................................................................20

2.6. Precipitation ......................................................................................... 67

3. Nucleation and Crystal Growth .................................................................... 80

3.1. Crystal shapes ..................................................................................... 80

4. Population Balances and Crystal Size Distributions .................................. 103

4.1. Basic population balance equations ................................................ 103

5. Crystallization from the Melt ...................................................................... 160

5.1. Fundamentals of melt crystallization ................................................ 161

PART II. SORPTION AND CHROMATOGRAPHY

Nomenclature Chapters 6 to 10 ...................................................................... .207

6. Basics of Sorption in Packed Columns ...................................................... 217

6.1. Packing structure .............................................................................. 218

7. Linear Theories of Sorption and Chromatography ..................................... 288

7.1. Types of chromatography .................................................................288

8. Non-Linear Theories and Packed Bed Adsorption Systems ...................... 365

8.1. Mass transfer zone approach ............................................................ 365

8.7. Design considerations ....................................................................... 437

9. Ion Exchange ..............................................................................................452

9.1. Basics of ion exchange .................................................................... .452

10.1. Single solute recovery: Staged systems ......................................... .499

Nomenclature Chapter 11 .................................................................................. 547

11. Electrophoretic Separation Methods ..........................................................550

11.1. Theory of electrophoresis ................................................................ 550

11.1.3. Complicating factors in electrophoresis .......................... 561

PART III. MEMBRANES

Nomenclature Chapters 12 and 13 .................................................................... 623

12. Introduction to Membrane Separations ...................................................... 630

12.1. Basic concepts ................................................................................ 630

12.10. Summary - Objectives ................................................................... 719

13. Detailed Theories for Membrane Separations ............................................733

13.1. Approximate analysis of concentration polarization ......................734

PART IV. SELECTION AND SEQUENCING

Nomenclature Chapter 14 .................................................................................. 793

14. Selection and Sequencing of Separations ...................................................795

14.1. General characteristics of the separation problem .........................795

14.3. Energy criteria ................................................................................812

Appendix. Answers to Selected Problems .......................................................847

1.1. PART I: CRYSTALLIZATION

Crystallization is a complex, multifaceted subject On one level of sophistica-

When temperature has a large effect, a large amount of crystals can be

Chapter 4 develops population balances (a count of the number of crys-

The second major type of crystallization is melt crystallization. In melt

The chapters on crystallization assume familiarity with equilibrium stage