J . Fluid Mech. (1972), vol. 54, part 2, pp.

241-261 241
Printed in Great Britain

The mechanics of an organized wave in turbulent

shear flow. Part 2. Experimental results
By A. K. M. F.HUSSAINI AND W. C. REYNOLDS
Department of Mechanical Engineering, Stanford University

(Received 18 October 1971)

Results on the behaviour of controlled wave disturbances introduced artificially

into turbulent channel flow are reported. Weak plane-wave disturbances are
introduced by vibrating ribbons near each wall. The amplitude and relative phase
of the streamwise component of the induced wave is educed from a hot-wire
signal, allowing the wave speed, the attenuation characteristics and the wave
shape to be traced downstream. These results form a basis for evaluation of
closure models for the dynamical equations describing wave components in
shear-flow turbulence.

1. Recapitulation
This paper is a direct continuation of Hussain & Reynolds (1970a), hereafter
referred to as I, and summarizes all the key experimental results given in our
report Hussain & Reynolds ( 1 9 7 0 4 , hereafter referred to as R. Connexions with
theoretical models will be discussed in part 3 (Reynolds & Hussain 1972). The
problem at hand is the description of the behaviour of controlled periodic
(sinusoidal) disturbances artificially introduced into a turbulent channel flow.
Motivations for this work are elaborated in I and R, and include the current
interest in possible wave theories of turbulent shear flow (Landahl 1967).
The basic apparatus (see figure 2 of I)involves an air flow channel with a gap of
2.5 inches in which vibrating ribbons located in the flow on opposite sides of the
channel are used to introduce sinusoidal disturbances of a controlled frequency
and phase into fully developed two-dimensional turbulent channel flow (see
figure 1).The objective is to measure the propagation and decay characteristics
of disturbances of chosen frequencies, and to relate these measurements to
theoretical treatments. The basic flow was described in I$ and is documented in
greater detail in R. Particular attention was paid to the elimination of residual
large-scale motions arising from the entrance conditions, and the long length of
this apparatus (length/gap ratio of 230) gives a degree of turbulence-structure
development not previously achieved.
The ribbon vibration is produced by passing an alternating current through
the ribbons, which are located in a steady magnetic field. The ribbons may be

t Present address : Department of Mechanical Engineering, University of Houston,

Houston, Texas.
$ In figure 4 of I the ordinate should be (0, 2.0) instead of (0, 20).
16 PLM 54
 242 A . K . H.P.Hussain and W . C. Regnolds

4==%
End-wall boundary layer

two-dimensional
channel flow

Small-scale --+
turbulencz
45 in 0 ‘
ometer

11 Controlled periodic
disturbance

z
7 Channel half-width, u”= 1.25 in.

FIGURE
1. Schematic diagram of test flow.

2. Measured ( a )2 amplitude and ( b ) G phase distributions a t 25 Hz.

FIGCRE
 A . K . M . F . Hussain and W . C . Reynolds

L
0

3. Measured (a) ii amplitude and ( b ) .iiphase distributions a t 50 Hz.

FIGURE

moved so as to produce either symmetric or antisymmetric disturbances in the

disturbance velocity field. The ribbon current forms a useful reference signal for
analysis of the hot-wire data. Care was taken to maintain a fixed ribbon amplitude
for all runs a t a given frequency, thus permitting surveys taken on different days
to be properly related.
It is convenient to represent the velocity field as a superposition of three
components:
u, = u,+?zi+u;. (1.1)
Here ;lii is the mean (time-averaged) contribution, (6,is the periodic contribution
(the organized wave) and u{ correspondsto the turbulent motion. Straightforward
time-averaging determines Z7:.I n order to measure 4, we use a phase-averaging
technique (seeI); the phase a,verageis defined as the average a t a given phase in
the cycle of the basic wave (i.e. the reference signal). A wave form eductor, a
device which can simultaneously determine the phase average a t 100 points in
the cycle of the basic wave, is used for this process. The eductor samples the
signal at 100 points in each cycle and computes a running average a t each of
these points. I n our experiments the wave component is very weak, so a large
ensemble of cycles (typically LO5)is required. The phase average of the signal then
gives the sum of the mean and the organized wave. Denoting the phase average
by (>,we have
(Ui) = ui + G,. (1.2)
I n effect, the phase-averaging process rejects the random background turbu-
lence and extracts the organized motions from the total signal.
A linearized constant-temperature hot-wire anemometer is used to measure
the streamwise velocity a t selected points downstream of the ribbon. The
 The mechanics of an organized wave. Part 2 245

I
\ I
\ \ /
\ /b \

a
. .
\ \
. oc,
II c

.
/ '

.
\

I
.
T
I- i-
/ \ \

, \
4
'
,
'
\

I \

. I
'\

- T \/'I
 A . K . M . P.Hussain and W . C . Reynolds

4. Measured (a)ii amplitude and ( b ) .iiphase distributions a t 75 Hz.

FIGURE

hot-wire signal is first ampliiied with an ax. coupled amplifier, which rejects the
d.c. component, and the amplified signal is then fed to the wave form eductor,
which provides the desired phase-averaged signal as an output. The educed
wave form is in general slightly non-sinusoidal, and the sinusoidal component is
extracted by analysing the continuous educed signal with a lock-in amplifier. A
schematic diagram of the circuitry is given in I; for more details the reader should
refer to R.
Suppose that we represent the reference signal (ribbon voltage) by
e, = a cos (wt) (1.3)
and the educed hot-wire signal corresponding to the streamwise component 6,
at a fixed point by
B, = IGll cos (wt- q5). (1.4)
Then the signal analysis processes outlined above give us both the amplitude
I.iill and phase q5 of B, locally.
We denote the distance downstream from the ribbon by x,the distance from
one wall by y and the chaiinel half-width by S (figure 1). All measurements
reported here were taken a t the channel centre-height (z = 0) with a centre-line
velocity Uo = 21*9ft/s, corresponding to a Reynolds number UoS/v of 13800.
Spanwise surveys indicate that the waves remain quite two-dimensional as they
progress downstream, hence we have treated the disturbance wave as two-
dimensional. For documentahion see R.
I n I we reported partial data at 100 Hz. Here we report the data for 25, 5 0 , 7 5
and 100Hz. For more detail see R.
 The mechanics of an organized wave.Part 2 247

/\

,
/

/ \
/
I d I \

\
' IB' /

\
\

JP
+--t-tt
-
c
24 8 A . K . M . F.Hussain and W . C . Reynolds

0 0.2 0.4 0.6 0.8 1d 0 0.2 0.4 0.6 0.8 1.0

FIGURE
5. Measured ( a ).ii amplitude and (a) .ii phase distributions at 100 Hz.

2. Wave data
The data are given in figures 2-5. The phase plots show q5vs. y/6 a t selected
streamwise stations XIS. Note that the origin of these phase plots corresponds to
the channel centre-line (y/6 := 1) and that the circumference of the polar plots
corresponds to the wall (y/6 =: 0). The bars on the data indicate the experimental
uncertainty as estimated from selected repetitions of the experiments. These data
were taken with the ribbons vibrating in the same direction a t the same time;
this produces a disturbance which is antisymmetric in the streamwise fluctuation
Q and symmetric in the cross-stream fluctuation 6. Surveys in the range
1 < y/s < 2

confirm this antisymmetry in G (see R).

The magnitudes of the amplitudes should be noted. Observe that the r.m.s.
disturbance amplitude is typically one-thousandth of the centre-line mean
velocity U,, or a few hundredths of the turbulence 1.111.5. velocity. Thus we are
able to recover an extremely weak organized component from a background of
finite fluctuations.
Let us first concentrate on the amplitude behaviour. I n each case the distur-
bance amplitude generally decreases in the downstream direction, although in
some cases there is a momentary increase at particular y positions. By analogy
with linear stability theory, one might expect that the amplitude curves would
retain their shape, provided o l course that the ribbons excite a single disturbance
eigenmode. However, there are significant changes in the shapes of the amplitude
curves in each case, although in each case there is an apparent tendency towards
self-similarity a t the most downstream stations. Again calling on stability theory
for insight, we recognize that a superposition of several eigenmodes having the
same frequency, but each decaying a t a different rate in the streamwise direction,
would not be self-similar but would approach this behaviour far downstream,
The mechanics of an organized wave. Part 2 249

\ 18' /
250 A. K . M . F . Hussain and W . C. Reynolds

0.3L
0
I
4
' ' 8
I I -''
12 16 20 24 28 32 36 40
46
FIGURE6 . Peak of zi amplitude a t different streamwise stations.
Re = 13800; 0, 5 0 H z ; 0,
2 5 H z ; 0, 75Hz; v,
100Hz.

where the most slowly decaying mode would dominate. Thus, we should not be
surprised to observe changes in the amplitude distribution as the disturbance
progresses downstream.
Figure 6 shows the maximum disturbance amplitude as a function of x for the
four test frequencies.Note that the streamwise decay is most rapid for the highest
frequencies and that the maximum amplitude (which at different x stations
occurs at different y points) decays exponentially after an initial region near the
ribbon. Exponential decay is indeed the behaviour one would expect from small
amplitude disturbance theory, and one therefore suspects that the disturbance
should be modelled adequately by a linear theory. Figure 7 shows the streamwise
decay at fixed y stations for each of the four frequencies. The dashed line corre-
sponds to an exponential de,cayfor subsequent use in comparing the data with
linewized disturbance theories.
The phase distributions (see figures 2 (b)-5 ( b ) ) also show a somewhat self-
similar nature. The sudden phase change that takes place midway across the
flow results from vorticity shed by the oscillating ribbon. I n linearized distur-
bance theories, a normal-mode disturbance will travel downstream with a self-
preserving phase distribution, offset by an angle related to the disturbance
propagation speed. Since the data determine the phase angle only within
0" < q5 < 360°, some interpretation is needed to relate properly the phases at two
streamwise stations. By giving due consideration to the probable magnitude of
the disturbance propagation velocity (see $ 3 for more detail) we arrived at the
 The mechanics of an organized wave. Part 2 25 1

6.0
5.0
4.0
3.0
c)

B 2.0 5
-
1

-
X
X

s 1.0
-z 0.8
1% 0.6
0.5
0.4
0.3
0.2

2
c1

related phase distributions shown, for two y stations, in figure 8. Note that the
phase advances most slowly for the lowest frequency disturbance. The linearity
of these curves strongly suggests a constant disturbance propagation velocity, as
would be expected from a linearized disturbance theory.
Some insight as to the presence of higher harmonics in the educed signal was
obtained from oscillascope pictures taken a t each point. These pictures also
provided a convenient visual check on the measured amplitudes and phases, and
on the stability of the educed pattern. The higher harmonics are important only
very near the ribbon; in the region where exponential decay is observed they
seem unimportant. For photographs and Fourier analyses of particular wave
forms see R.
252 A . K . M . F . H u s s a i n and W . C. Reynolds

t /

4s
FIGURE 8. Phases of .ii at constant distance from the wall.

y/S = 0.16
ylS = 0.56
100Hz
v
v
+
75Hz

0
50Hz
0
0
25Hz
rn
0

3. Single-mode analysis
From the discussion above it is clear that the disturbance wave shapes are not
exactly preserved in the streamwise direction, and hence the disturbance waves
are not pure modes. A proper analysis of the data will therefore require a multiple-
mode analysis. However, for the purpose of approximate analysis of the data and
evaluation of average propagation characteristics, the disturbance will now be
treated as though it were exactly a single mode. It is also assumed that the
wave amplitudes are sufficiently small to be treated by linear theory. All quanti-
ties will be treated as dimensionless, using the channel half-width S and the
centre-height (z = 0) continuity velocity U, as the reference scales. Note that
for this flow UJU, = 0.881, where U, is the velocity on the channel centre-line.
It is usual to represent the two-dimensionalnormal-mode perturbation wave as

where * denotes a complex conjugate, Q(y) = [G(y),O(y),01 is the two-dimensional

complex eigenmode shape, a = a,+iai is the (complex) wavenumber, a, being
the streamwise wavenumber and as the growth factor, and c = c,+ic, is the
(complex) wave speed. Alternatively, one may write (3.1 a) as

ii = &{Q(y)ei(az-ot)+ conjugate), (3.lb)

 The mechanics of a n organized wave. Part 2 253

25 Hz 50 Hz 75 Hz 100 Hz
56 f 7' 107 f 9" 156 f 11' 204 f 12"
0.981 f 13 yo 1-87f 8.5 yo 2.72 7 yo 3-57 6.1 yo
0.0727 f 10 yo 0.122 5 12 yo 0.196 14 yo 0.285 f 16 Yo
0.847 1-692 2.54 3.39
0.861 13 yo 0.904 f 8.5 yo 0.93 f 7 yo 0.946 f 6.1 yo
0.0638 f 10 yo - 0.0591 k 12 yo - 0.0672 f 14 Yo - 0.0755 f 16 yo
6.40 3.36 2-31 1.76
0.862 0.905 0.934 0.949
0.465 0.420 0.451 0.501
TABLE1. Wave data. All quantities are dimensionless and normalized using
& a n d Urn;U,&/V = 13800, U,lU,, = 0.881.

where o is the (real) circular frequency of the oscillation. Note that

c = 0101. (3.2)
The actual wave phase velocity is
v, = o/a,. (3.3)
Within the approximation mentioned above, it is possible to estimate the
propagating wave characteristics (a,, ai,
c, and ci) from the data. For two stations
x, and x2,both at the same distance y from the wall,

Hence

Values of aias obtained from the average decay estimates shown in figure 7 are
given in table 1. Positive values of ai,as found in this experiment, are qualita-
tively consistent with the decaying-wave model of Landahl(l967) and the wall-
pressure correlation measurements of Willmarth & Wooldridge (1962) and Bull
(1963) on a flat plate and Corcos (1964) in a pipe.
From (3.1) it also follows that

w 2 , Y,M m l , Y,t ) = exp [ - aik2 - x1) + M X 2 -2111. (3.6)

If the phase of .iiis denoted by 4, (3.6) gives
$(xz,Y7 t ) = $&I, Y,t ) +a,(% - XI) (3.7)
and thus a, = (#2- M ( x 2 - 21). (3.8)
Values of a, deduced from the phase data as interpreted in figure 8 are given in
table 1.
Once a,,and aiare known, c can be calculated from (3.2) and V, from (3.3).
Also of interest are the wavelength
h = 2;rr/a,, (3.9)
254 A . K . M . I?. Hussain and W . C . Reynolds

YP
FIGURE 9. Inferred ( a ) d amplitude and ( b ) d phase distributions a t 25Hz. -, x/S = 4;
--.-..,
- - , x I S = 6 ; - - - - , ~ / & = 8 ; - - - * - - - , x/S=10;
-...-, z/S= 18; -....-, XIS = 22.
x / S = 1 2 ; -..- , x/S = 14;

1.9

0 0.5 1 .O
YlS
FIGURE 10. Inferred ( a )6 amplitude and ( 6 ) d phase distributions at
50 He. Notation as in figure 9.
 The mechanics of an organized wave. Part 2 255
and the attenuation factor per wavelength,

a = aih. (3.10)

Values of c, V,, h and a are also included in table 1.

Two features of the data deserve comment. First, we note that the convection
velocity increases slightly with frequency over the range investigated. Second,
the attenuation per wavelength is, within the experimental uncertainty, inde-
pendent of the frequency, with ctila, w 0.07 over the range of frequencies investi-
gated.
Some additional remarks on the determination of the phase differences may be
helpful. I n the polar phase diagrams (figures 2 (b)-5 ( b ) )one can read the phase
difference between stations clockwise or counterclockwise, and in each case one
can also add or subtract multiples of 360". To understand how we unravelled
this riddle, consider figure 2(b) and concentrate on the phase distribution a t
stations x/6 = 8 , 1 0 , 1 2 and 14. The distributions are similar, the phase difference
between two successive stations being about 120" clockwise. The corresponding
phase velocity from (3.3) is V , = 0.81. If the angle is instead taken counter-
clockwise, i.e. as 240°, we have V, = 0.405. If 360" were added in clockwise
direction we would obtain V , = 0.2, and if 360" were added in the counterclock-
wise direction V, = 0.16. It seems reasonable that these large-scale disturbances
travel at a speed close to the channel mean speed (which corresponds to V , = l ) ,
hence the clockwise interpretation giving a phase difference of about 120" is
selected.
The above comparisons lead one to favour strongly the measurement of phase
between stations in the clockwise direction. Once this has been decided upon, the
same system is followed for other frequencies. I n this way the angles measured
for 50Hz and 75Hz are approximately 215" and 310" respectively. However,
at 100 Hz the clockwise direction measurement has to be carried out in a slightly
different way to keep the angles consistent. I n this case an extra 360" must be
added while measuring the phase differencebetween two successivestations. This
yields an average phase difference between two stations of about 410" and a
correspondingphase velocity V , = 0.94. Without this addition of 360" the angle is
only 50" and the corresponding phase velocity would be V, = 0.078.
*
For a two-dimensional disturbance the continuity equation is

aqax + aqay = 0. (3.11)

Now, in the single-mode analysis,

aiilax = ia&eiax-iwt + conjugate. (3.12)

Using this in (3.11), and writing

v" = O(x,y)e+t + conjugate, (3.13)

we have (3.14)
Ju
256 A . K . lli. P.Hussain and W . C. Reynolds

YP
FIGURE
12. Inferred ( a ) d amplitude and ( b ) d phase distributions a t
100 Hz. Notation as in figure 9.
 The mechanics of an organized wave. Part 2 257

40

32

24

16
I
.
0

-
1

X 8
m 0
b
0

-8

-It

-24

-32
0 0.5 1.0
YP YP
FIGURE 13. Inferred ?u distribution. (a)25 Hz, ( b ) 50 Hz,
( c ) 75 Hz and ( d ) 100 Hz. Notation as in figure 9.
258 A . K . M.B. Hussain and W . C. Reynolds
Although one should be careful not to take the calculation too seriously, we did
compute ij using (3.14) and smoothed curves through the measured Q data; the
results are shown in figures !3-12. The striking feature is the apparent peakiness of
v" near the wall for the higher frequency disturbances. This is, indeed, quite
similar to the behaviour obtained from solutions of the Orr-Sommerfeld equation
for small amplitude disturbances on a laminar shear flow. Note that the ij curves
are not self-similar near the ribbon but seem to be progressing towards a self-
similar shape in the downstream direction. This would be characteristic of a
disturbance formed from a superposition of a lightly damped and a more heavily
damped eigenmode.
Armed with both 4 and i; data, one might be tempted to compute the distur-
bance Reynolds stress distribution. I n view of the problems arising from a one-
mode treatment, one should yield to this temptation only after issuing a con-
siderable warning to the readers, which is implicit in this discussion. The wave
Reynolds stress is
-
Qv" = B(GB*+&*B). (3.15)

Having cautioned fiercely against substantiative use of these results, we note that
the wave Reynolds stresses are given in figure 13.
It should be clear that a direct measurement of v" would be preferable to the
calculations outlined above. I n addition to those of the wave Reynolds stress,
one would also like to have direct measurements of the wave-induced fluctuations
in the turbulent Reynolds stresses:
__
Fij = <u;u;>- u;u;. (3.16)

Such measurements are now in progress at Stanford.

4. Relationships to 'natiiral' flows

In attempting to compare these data with parameters deduced from various
filtered correlation data obtained in 'natural' (undisturbed) flows, two points
must be kept in mind. First, the typical filtered correlation measurements (e.g.
Wills 1964) include contributions from waves of all obliquities, not just the two-
dimensional waves. Second, the waves studied here are relatively long in com-
parison with those studied in most previous experiments. These facts make
comparisons with most previous data very difficult. However, Landahl (1967)
pointed out that Corcos's (1964) wall-pressure data indicate that the wall-pressure
cross-power spectral density is dominated by waves that are nearly normal to
the stream. Consequently, it does make some sense to compare our two-dimen-
sional wave data t o such pressure data. Although the data for boundary layers
are more extensive, comparison with pipe flows seems more reasonable, and we
shall use Corcos's (1964) pipe data for this comparison.
One of the most striking observations in Corcos's data, also observed by
Willmarth & Wooldridge (1962), is that the amplitude of the wall-pressure cross-
power spectral density depends only upon the parameter wg/:/vc, where 6 is the
separation distance in the streamwise direction. I n figure 14 we have replotted
 The mechanics of an organized wave. Part 2 259

-
-
-
a
u3" 0.4
- - -

0.2 I I I , I I I I I I
0 2 4 6 8 10
dK
14. Comparison of wall-pressure cross-power spectral density.
FIGURE
-, Corcos's data; - - -, equation (4.7).

Corcos's data on semi-log paper; the striking similarity with figure 6 should be
noted. For long separation distances the behaviour is essentially exponential,
which is consistent with our small-disturbance model.
If we view the turbulence in a pipe as being composed of a superposition of
weak waves and assume that the pressure will be dominated by long two-
dimensionalwaves of the type studied here, a direct quantitative comparison can
be made. We represent the contribution of a single wave to the wall pressure by
f l = fj; exp { - ayx + iapx + iot}+ conjugate. (4.1)
Here the superscript n denotes the nth wave. It seems reasonable to assume that
the waves are uncorrelated, i.e. that

(92 fig*) = An anme (44

Here ( ) denotes an ensemble average. Then the autocorrelation of the wall
pressure is

= IAnI 2 exp { - 2arx - a?[ - ia,"[- ~ w " T +) conjugate. (4.3)

n

Now, the harmonic decomposition with respect to r defines the cross-power

spectral density. We see that the amplitude of the cross-power spectral density
should behave like (see Landahll967)

IS,,(W)~ N x IAn12e-Lf5. (4.4)

[ -5a:~)
n
This may be written as
~ x p p ( u ) lN Cn.lAnl2exp (4.5)

I 7-2
260 A . K . M . F . Hussain and W . C . Reynolds

‘?0.6
s 0.4
0.2
0 2 4 6
fdl a
FIGURE
15. Comparison of canvection velocities for with Corcos’s data for pipe flow.
-, Corcos’s data based on wall pressures; 0,present data.

The experiments deal with the ‘least-damped’ modes, which of course will
c.
dominate lSpplfor large For these modes we find = 0.07, and hence for
large E we would expect IS,,,I to depend only upon the parameter w[/V,,as indeed
is found experimentally. For large 5 the least-damped mode should dominate,
and hence we would expect

IS,, I N constant x exp { - 0-07wLJQ as 5 co.

--f (4.7)

The dashed line in figure 184 shows this asymptotic behaviour, which is in re-
markable agreement with CIorcos’s data.
Corcos also measured the convection velocity of the wall-pressuredisturbances.
If we again assume that the two-dimensional waves dominate the wall pressure,
then a comparison can be made. Figure 15 shows Corcos’s data; f i s the frequency
(Hz), d is the pipe diameter and 0 is the ‘mean discharge velocity’ for the pipe.
Using 2Sinstead of d and Urnfor g , the present data are also shown. Note that the
present experiments are confined to relatively low frequencies and that values
comparable with Corcos’s are found, although the trends with frequency seem to
be in opposite directions. This might be due to dominance of more strongly
oblique waves a t the higher frequencies.
The close association between our data for two-dimensionalwavesand Corcos’s
wall-pressuredata provides considerable support for the notion that it is useful to
view shear-flow turbulence as waves (Landahl 1967).

5. Concluding remarks
The results given here provide the first real means of testing model equations
purporting to describe the behaviour of wavelike disturbances in a turbulent
shear flow. Although the ii data are reported with some confidence, the inferred
fi data should be used with caution, pending direct experimental measurement.
The work described in this paper is now being extended at Stanford using a
better technique (cross-correlation)for sifting out the periodic component of the
hot-wire signal, and in due course we shall be able to provide theoreticians with
good measurements of both the wave components ii, and the wave-induced
Reynolds stress fluctuations Fij for a well-documented turbulent channel flow.
 The mechanics of an organized wave. Part 2 261
Additional interpretation of these data, in relationship to pending theoretical
models, is given in part 3 of this investigation (Reynolds & Hussain 1972).

This work is sponsored by the Mechanics Branch of the Air Force Office of
Scientific Research and by the National Science Foundation; their assistance is
gratefully acknowledged.

REFERENCES
BULL,M. K. 1963 Properties of the fluctuating wall-pressurefield of a turbulent boundary
layer. University of Southampton A.A.S.U. Rep. no. 234.
CORCOS,G. M. 1964 J. Fluid Mech. 18, 353.
HUSSAIN, A. K. M. F. & REYNOLDS, W. C. 1970a J. Fluid Mech. 41, 241.
HUSSAIN, A. K. M. F. & REYNOLDS, W. C. 1970b The mechanics of a perturbation wave
in turbulent shear flow. Department of Mechanical Engineering Rep,, Stanford Uni-
versity, FM-6.
LANDAHL, M. T. 1967 J. Fluid Mech. 29, 441.
REYNOLDS, W. C. & HUSSAIN, A. K. M. F. 1972 J. Fluid Mech. 54, 263.
WILLMARTH, W. W. & WOOLDRIDGE, C. E. 1962 J. Fluid Mech. 11, 187.
WILLS,J. A. B. 1964 J . Fluid Meoh. 20, 417.