1.

I. DISTRIBUTION SPACES

§1. Motivation and overview

1.1. Introduction.
In the study of ordinary differential equations one can get very far by
using just the classical concept of differentiability, working with spaces of
continuously differentiable functions on an interval I ⊂ R:
dj
C m (I) = { u : I → C | dxj u
exists and is continuous on I for 0 ≤ j ≤ m }.
(1.1)
The need for more general concepts comes up for example in the study of
eigenvalue problems for second order operators on an interval [a, b] with
boundary conditions at the endpoints a, b, by Hilbert space methods. But
here it usually suffices to extend the notions to absolutely continuous func-
tions, i.e., functions u(x) of the form
Z x
u(x) = v(y) dy + c, v locally integrable on I (1.2)
x0

(integrable on compact subsets of I); c denotes a constant. Here v is regarded

d
as the derivative dx u of u, and the fundamental formula
Z x
d
u(x) = u(x0 ) + dy
u(y) dy (1.3)
x0

still holds.
But for partial differential equations one finds when using methods from
functional analysis that the spaces C m are inadequate, and there is no good
concept of absolute continuity in the case of functions of several real variables.
One can get some ways by using the concept of weak derivatives: When u
∂
and v are locally integrable on an open subset Ω of Rn , we say that v = ∂x j
u
in the weak sense, when
Z Z
∂
− u ∂xj ϕ dx = vϕ dx, for all ϕ ∈ C0∞ (Ω); (1.4)
Ω Ω

here C0∞ (Ω) denotes the space of C ∞ functions on Ω with compact support
in Ω. (The support supp f of a function f is the complement of the largest
 1.2

open set where the function is zero.) This criterion is modeled after the fact
∂
that the formula (1.4) holds when u ∈ C 1 (Ω), with v = ∂x j
u.
Sometimes even the concept of weak derivatives is not sufficient, and the
need arises to define derivatives that are not functions, but are more general
objects. Some measures and derivatives of measures will enter. For example,
there is the Dirac measure δ0 that assigns 1 to every Lebesgue measurable set
in Rn containing {0}, and 0 to any Lebesgue measurable set not containing
{0}. For n = 1, δ0 is the derivative of the Heaviside function defined in (1.8)
below. In the book of Laurent Schwartz [S 1961] there is also a description of
the derivative of δ0 (on R) — which is not even a measure — as a “dipole”,
with some kind of physical explanation.
For the purpose of setting up the rules for a general theory of differen-
tion where classical differentiability fails, Schwartz brought forward around
1950 the concept of distributions: a class of objects containing the locally
integrable functions and allowing differentiations of any order.
This book gives an introduction to distribution theory, based on the work
of Schwartz and of many other people. Our aim is also to show how the
theory is combined with the study of operators in Hilbert space by methods
of functional analysis, with applications to ordinary and partial differential
equations. In some chapters of a more advanced character, we show how the
distribution theory is used to define pseudodifferential operators and how
they are applied in the discussion of solvability of PDE, with or without
boundary conditions. A bibliography of relevant books and papers is col-
lected at the end.
Plan.
Part I gives an introduction to distributions.
In the rest of Chapter 1 we begin the discussion of taking derivatives in the
distribution sense, motivating the study of function spaces in the following
chapter.
Notation and prerequisites are collected in Appendix A.
Chapter 2 studies the spaces of C ∞ functions (and C k functions) needed
in the theory, and their relations to Lp spaces.
The relevant topological considerations are collected in Appendix B.
In Chapter 3 we introduce distributions in full generality and show the
most prominent rules of calculus for them.
Part II connects the distribution concept with differential equations and
Fourier transformation.
Chapter 4 is aimed at linking distribution theory to the treatment of
partial differential equations (PDE) by Hilbert space methods. Here we
 1.3

introduce Sobolev spaces and realizations of differential operators, both in

the (relatively simple) one-dimensional case and in n-space, and study some
applications.
Here we use some of the basic results on unbounded operators in Hilbert
space that are collected in Chapter 12.
In Chapter 5, we study the Fourier transformation in the framework of
temperate distributions.
Chapter 6 gives a further development of Sobolev spaces as well as appli-
cations to PDE by use of Fourier theory, and shows a fundamental result on
the structure of distributions.
Part III contains more advanced material, primarily on pseudodifferential
operators (ψdo’s), a generalization of partial differential operators containing
also the solution operators for elliptic problems.
Chapter 7 gives the basic ingredients of the local calculus of pseudodiffer-
ential operators.
Chapter 8 shows how to define ψdo’s on manifolds, and how they in the
elliptic case define Fredholm operators, with solvability properties modulo
finite-dimensional spaces. (An introduction to Fredholm operators is in-
cluded.)
Chapter 9 takes up the study of boundary value problems by use of Fourier
transformation. The main effort is spent on an important constant-coefficient
case which, as an example, shows how Sobolev spaces of noninteger and
negative order can enter. Also, a connection is made to the abstract theory
of Chapter 13. This chapter can be read directly after Parts I and II.
In Chapter 10 we present the basic ingredients in a pseudodifferential
theory of boundary value problems introduced originally by L. Boutet de
Monvel; this builds on the methods of Chapters 7 and 8 and the example in
Chapter 9, introducing new operator types.
Chapter 11 shows how the theory of Chapter 10 can be used to discuss
solvability of elliptic boundary value problems, by use of the Calderón pro-
jector, that we construct in detail.
Part IV gives the supplementing topics needed from Hilbert space theory.
Chapter 12, departing from the knowledge of bounded linear operators
in Hilbert spaces, shows some basic results for unbounded operators, and
develops the theory of variational operators.
Chapter 13 gives a thorough presentation of certain families of extensions
of closed operators, of interest for the study of boundary value problems for
elliptic PDE and their positivity properties.
 1.4

Chapter 14 establishes some basic results on semigroups of operators, rel-

evant for parabolic PDE (problems with a time-parameter), and appealing
to positivity and variationality properties discussed in earlier chapters.
Finally, there are three appendics. In Appendix A, we recall some basic rules
of calculus and set up the notation.
Appendix B gives some elements of the theory of topological vector spaces,
that can be invoked when one wants the correct topological formulation of
the properties of distributions.
Appendix C introduces some function spaces, as a continuation of Chapter
2, but needed only later in the text.
1.2 On the definition of distributions.
The definition of a weak derivative ∂j u was mentioned in (1.4) above.
Here both u and its weak derivative v are locally integrable functions on
(Ω). Observe that the right hand side is a linear functional on C0∞ (Ω), i.e.,
a linear mapping Λv of C0∞ (Ω) into C, here defined by
Z
Λv : ϕ 7→ Λv (ϕ) = vϕ dx. (1.5)
Ω

The idea of Distribution Theory is to allow much more general functionals

than this one. In fact, when Λ is any linear functional on C0∞ (Ω) such that
Z
− u∂j ϕ dx = Λ(ϕ) for all ϕ ∈ C0∞ (Ω); (1.6)
Ω

we shall say that

∂j u = Λ in the distribution sense, (1.7)

even if there is no function v (locally integrable) such that Λ can be defined

from it as in (1.5).
Example 1.1. Here is the most famous example in the theory: Let Ω = R
and consider the Heaviside function H(x); it is defined by

1 for x > 0

H(x) = (1.8)
0 for x ≤ 0 .

It is locally integrable on R. But there is no locally integrable function v

such that (1.4) holds with u = H:
Z Z
d
− H dx ϕ dx = vϕ dx, for all ϕ ∈ C0∞ (R). (1.9)
R
 1.5

For, assume that v were such a function, and let ϕ ∈ C0∞ (R) with ϕ(0) = 1
and set ϕN (x) = ϕ(N x). Note that max |ϕ(x)| = max |ϕN (x)| for all N ,
and that when ϕ is supported in [−R, R], ϕN is supported in [−R/N, R/N ].
Thus by the theorem of Lebesgue,
Z
vϕN dx → 0 for N → ∞, (1.10)
R

but on the other hand,

Z Z ∞ Z ∞
d 0
− H dx ϕN dx = − N ϕ (N x) dx = − ϕ0 (y) dy = ϕ(0) = 1.
R 0 0
(1.11)
So (1.9) cannot hold for this sequence of functions ϕN , and we conclude that
a locally integrable function v for which (1.9) holds for all ϕ ∈ C0∞ (R) cannot
exist.
A linear functional that does match H in a formula (1.6) is the following
one:
Λ : ϕ → ϕ(0) (1.12)
(as seen by a calculation as in (1.11)). This is the famous delta-distribution,
usually denoted δ0 . (It identifies with the delta-measure mentioned earlier.)
There are some technical things that have to be cleared up before we can
define distributions in a proper way.
For one thing, we have to look more carefully at the elements of C0∞ (Ω).
We must demonstrate that such functions really do exist, and we need to
show that there are elements with convenient properties (such as having the
support in a prescribed set and being 1 on a smaller prescribed set).
Moreover, we have to describe what is meant by convergence in C0∞ (Ω),
and provide it with a topology. There are also some other spaces of C ∞ or
C k functions with suitable support or integrability properties that we need
to introduce.
These preparatory steps will take some time, before we begin to introduce
distributions in full generality. (The theories that go into giving C0∞ (Ω) a
good topology are quite advanced, and will partly be relegated to Appendix
B. In fact, the urge to do this in all details has been something of an obstacle
to making the tool of distributions available to everybody working with PDE
— so we shall here take the point of view of giving full details of how one
operates with distributions, but tone down the topological discussion to some
statements one can use without necessarily checking all proofs.)
The reader is urged to consult Appendix A (with notation and prerequi-
sites) before starting to read the next chapters.