On the Stability of Two-Dimensional Parallel Flows

Author(s): C. C. Lin
Source: Proceedings of the National Academy of Sciences of the United States of America,
Vol. 30, No. 10 (Oct. 15, 1944), pp. 316-324
Published by: National Academy of Sciences
Stable URL: http://www.jstor.org/stable/87454 .
Accessed: 07/05/2014 17:15

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

National Academy of Sciences is collaborating with JSTOR to digitize, preserve and extend access to
Proceedings of the National Academy of Sciences of the United States of America.

http://www.jstor.org

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
316 PHYSICS: C. C. LIN PROC.N. A. S.

was performed only on the nonads from the fifth to the eleventh, inclusive,
since the first four sums-of-products would have reproduced a part of S516'
and not of r156. (All of the nonads after the sixth were composed entirely
of supplied zeros both for N and qL56.) The last figures of s156 were copied
on the third line from the bottom of the fifth strip so that by forming the
difference between the second and third lines the required value of rms6
came out on the bottom line. This remainder was found on Aug. 11, 1944,
to be 118 57508 80382 71696 98184 73569 85091 23773 18030 92037. Since
this residue is not zero it follows that Mi57 is composite and incidentally
that M127still retains the position of being the largest known prime number.
For every value of k from 8 to 156 the numbers on the three corresponding
strips were found to satisfy the relation s,' = N- q +- rk for each of the
moduli 103 + 1, 104 + i and 107 + 1. The author desires to announce
that he has already begun to investigate M167.
'Archibald, R. C., Scripta Mathematica,3, 112-119 (1935).
2 Lehmer, D. H., Jour. London Math. Soc., 9-10, 162-165 (1934-1935).
3 Powers, R. E., Bull. Amer. Math. Soc., 40, 883 (1934).
4 Lehmer, D. N., Amer. Math. lrMonthly,30, 67, 68 (1923).

ON THE STABILITY OF TWO-DIMENSIONAL PARALLEL FLOWS*

BY C. C. LIN
GUGGENHEIM LABORATORY, CALIFORNIA INSTITUTE OF TECHNOLOGY

CommunicatedAugust 24, 1944

1. Introduction.-Heisenberg's' remarkable contribution to the hydro-

dynamic stability of two-dimensional parallel flows has not been favorably
accepted and properly appreciated, because his paper is not completely
free from obscure points. Nor has the work of Tietjens,2 Tollmien3 and
Schlichting4 been properly estimated. As a result, the theory to account
for the instability of laminar flow at high Reynolds numbers has become
very confused, and its further development has been very much retarded.
Various authors suggest that it is necessary (1) to consider disturbances of
finite amplitudes, (2) to include the effect of compressibility or even (3)
to modify the Navier-Stokes equations. The present situation of our
knowledge may be seen from the general lectures given by G. I. Taylor5
and J. L. Synge.6
Recently the present writer carried out some investigations in an at-
tempt to clarify the situation. The theory of Heisenberg was critically ex-
amined, somewhat modified and further developed. These developments
were made with particular emphasis on the general criteria of instability

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
VOL. 30, 1944 PHYSICS: C. C. L1N 317

and their underlying physical mechanism, both in an inviscid fluid and in a

real fluid. According to the general criterion obtained, the plane Poiseuille
flow and the Blasius flow are both unstable at sufficiently high Reynolds
numbers.7 Detailed numerical calculations were also worked out in these
cases. The present article gives a brief account of the essential results ob-
tained. A detailed treatment will be published elsewhere.
2. The Equation of Orr and Sommerfeld.-In studying the stability of
two-dimensional parallel flows, we may restrict ourselves to two-dimen-
sional disturbances, according to an investigation of Squire.8 A two-di-
mensional periodic disturbance in a field of flow with main velocity w(y)
parallel to the x-axis may be represented by the stream function ~, =
p,(y)ei"(- t), and the linearized differential equation for pY(y)is

(w - c)(ap" - a2) - w"( = - (i' - 2a2' + a4). (2.1)

All the quantities involved are dimensionless: R is the Reynolds number of

the main flow, and a, c are constants describing the nature of the disturb-
ance. We shall regard a as real so that 2-r/a is the wave-length of the dis-
turbance. The real part of c gives the phase velocity while its imaginary
part (multiplied by a) gives the index of damping or magnification. The
function w(y) is regarded as defined for all complex values of y by analytic
extension. Physical considerations then lead to certain homogeneous
boundary conditions to be satisfied by so(y) at two real points y, and y2 of
the complex y-plane, which represent the co6rdinates of certain layers in
the field of flow. There are altogether four boundary conditions. Hence,
we are led to an eigenvalue problem, and a relation of the type

F(a, c, aR) = 0 or c = c(a, R) (2.2)

holds, where F is indeed an entire function of the parameters. Since sta-

bility, neutral stability or instability of the motion correspond to the real
part of -iac less than, equal to or greater than zero, the problem is to de-
termine the imaginary part ci(a, R) of c(a, R) for a set of real values of a
and R.
The actual investigation depends upon the solution of (2.1) by the
method of successive approximations. Two methods are used. First, we
introduce the variable

'= (y - yo)/e, (2.3)

where w(yo) = c, e = (aR)-l/3, and w'(yo) is positive for real values of c.

In this way, we obtain a fundamental system of four solutions sp(y) in the
form

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
318 PHYSICS: C. C. LIN PROC.N. A. S.

co

p(y) = ES exn) (X), (2.4)

n 0

where x(n)(v) are explicitly expressible in terms of the Hankel functions

H'l 2 ['2/3(iC /, o = .
{w(yo)1/3a Indeed, the series (2.4) can be
shown to be convergent for the range of variables and parameters concerned.
Secondly, we may put
co

o(y)= E (aR) " (y), (2.5)

n - 0

which gives us two asymptotic solutions. The initial approximation of (2.5)

satisfies the inviscid equation
- -
(w c)(p?(0" - a2o(0) w"(?0) = 0, (2.6)

which can be solved by expanding p(O)in power series of a2, convergent for
all values of a2. Two other asymptotic solutions are obtained in the form

p(y) = exp{f gdy}, g = (aR) /2go + gl + (aR)-t/2g,2 + .... (2.7)

Both methods of solution are essentially due to Heisenberg,' and Toll-
mien followed the same way. Heisenberg also carried out some approxi-
mate calculation for the case of channel flow, while Tollmien studied the
case of the boundary layer. Criticism of Heisenberg's work is usually as-
sociated with the multiple-valued nature of the asymptotic solutions and
the convergence of the power series in a2 used for solving (2.6). Detailed
investigations justify Heisenberg's treatment. In fact, the asymptotic
solutions hold only for certain regions of the y-plane to be determined by
comparison with the asymptotic expansions of (2.4). Indeed, from the
properties of the Hankel functions, it can be easily shown that under the
restriction

< arg (aom) < 6 (2.8)

6 6
the asymptotic expansions do not suffer any jump (which may be called
the Stokes .phenomenon or the Uebergangs-substitution). Fortunately,
such a restriction does not exclude the existence of a connected region en-
closing the points y1 and y2 where our boundary conditions are to be satis-
fied. The eigenvalue problem can therefore be conveniently treated with
the help of the asymptotic solutions. However, the region of validity of the
asymptotic expressions does not always include all the points of the real
segment (y1, y2). As a matter of fact, the asymptotic solutions hold
throughout the real interval (yl, y2) only when ci > 0. For ci < 0, there are
two points on the real axis where the Stokes jump of the asymptotic solu-
tions occur. These two points coalesce into one critical point when ci = 0.

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
VOL. 30, 1944 PHYSICS: C. C. LIN 319

Physically, the failure of the asymptotic solutions at some points on the

real axis (and therefore the failure of (p0) as a first approximation) in the
case of neutral (ci = 0) and damped oscillations (ci <0) indicates the pres-
ence of the effect of viscosity at the corresponding layers inside the fluid.
This was noted by several authors in the cases of neutral stability. How-
ever, the existence of two inner viscous layers for the damped disturbance
does not seem to have been noticed before. As we shall see in Section 3, it
has an important consequence upon the well-known criteria of Rayleigh9
and Tollmien.10
3. General Criteria of Stability and the Curve of Neutral Stability.-In the
limit of infinite Reynolds numbers, it can be shown by careful mathematical
investigation that the problem can be treated by finding a solution of (2.6)
satisfying two boundary conditions. Rayleigh' and Tollmien1? have shown
that the necessary and sufficient condition for the existence of neutral or
self-excited disturbances is that the velocity curve has a point of inflection
in the interval (yl, y2). The necessary condition is general, while the suf-
ficient condition has been proved by Tollmien only for velocity distribu-
tions of the symmetrical type or of the boundary-layer type. In fact, an
example has been shown by the present writer where a point of inflexion
exists in the velocity curve, but a neutral disturbance does not exist.
The authors mentioned above also concluded that damped disturbances
are excluded if a point of inflexion does not exist. However, their proof as-
sumes the analytical nature of the solution (0)(y) along the real axis,
which is true for self-excited disturbances, but not for damped disturbances.
Previous authors regarded a damped disturbance as the complex conjugate
of a self-excited disturbance. This is in contradiction with the above-men-
tioned drastic difference of nature of those solutions along the real axis.
Careful investigation reveals that the damped or self-excited nature of a
solution is not changed by taking the complex conjugate. Therefore
damped disturbances are not excluded by the condition w"(y) 7 0 (at least not
according to Rayleigh's proof).
Tollmien's proof to establish the existence of a point of inflexion in the
velocity curve as sufficient to insure instability is rather cumbersome, and
he has to assume that w"' does not vanish at the point of inflexion. These
points have now been improved by a more careful consideration of the
analytical nature of (2.6).
Having settled the question of hydrodynamic stability at infinite Rey-
nolds numbers, the author studied the stability in a real fluid by trying to
find out how the foregoing results should be Inodified for large but finite
values of the Reynolds number. One such result has already been obtained
by Heisenberg,l whose conclusion is as follows. If a velocity distribution
allows an inviscid neutral disturbance with finite wave-length and non-
vanishing phase velocity, the disturbance with the same wave-length is un-

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
320 PHYSICS: C. C. LIN PROC.N. A. S.

stable in the real fluid when the Reynolds number is sufficiently large. In
the present investigation, the proof has been formulated in a slightly more
satisfactory manner.
Since a point of inflection in the velocity curve is sufficient to insure in-
stability only for velocity distributions of the symmetrical and boundary-
layer types, the author has also made more attempts in these cases. In-
deed, it has been shown that in these cases the flow is always unstable for
sufficiently large Reynolds numbers, whether the velocity curve has a point
of inflection or not. The curve of neutral stability

c,(a, R) = 0 (3.1)

may belong to either of the types shown in figure 1. When the velocity
curve has no point of inflection, the two asymptotic branches of the curve
have the common asymptote a = 0 (Fig. 1 a). When there is a point of
inflection, one branch has the asymptote a = 0, while the other has the
asymptote a = as > 0 (Fig. 1 b). These results are in agreement with
those of Rayleigh, Tollmien and Heisenberg cited above.
cc
/
/ R /
/

/ /

0(C

FIGURE 1 (a) FIGURE 1 (b)

General nature of the curve of neutral stability. The dotted curve is a
curve of stability given by Synge.

Simple general formulae have also been derived to express the exact
forms of those asymptotic branches in terms of the velocity curve. The
fact that these two asymptotic branches join together to give a maximum
.a and a minimum R (instead of going to infinite a) can be inferred from a
criterion of Synge.6 It states that there is always stability (ci < 0) if

(qR)2 < (2a'2 + 1)(4a4 + l)/ac2, q= max. Iw'(y)|. (3.2)

yi < Y y2

Indeed, the minimum value of the Reynolds number on the neutral curve
marks the beginning of instability and is therefore very important. Ac-
cordingly, the following approximate formulae have been derived for the
evaluation of this minimum value:

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
VOL. 30, 1944 PHYSICS: C. C. LIN 321

R = 30w - w2dy }, for symmetrical profiles, (3.3)

R- 25wl
25, for boundary-layer profiles, (3.4)

where wl is the value of w'(y) at w = 0, and c is given by

, w(y)w"(y)
-7wl ) = 0.6 atw(y) = c. (3.5)

In these dimensionless expressions, the velocities are referred to the maxi-

mum velocity and the lengths to the distance from w = 0 to w = 1. Com-
plete numerical calculation of the curve ci(a, R) = 0 has been carried out
for the plane Poiseuille flow and the Blasius flow. The results are in gen-
eral agreement with figure 1 (a). For the channel, the critical Reynolds
number is found to be 16,000 as defined in terms of the maximum velocity
and the width of the channel; for the boundary layer, it is found to be 400
as defined in terms of the free-stream velocity and the displacement thick-
ness. The result in the latter case is in better agreement with Tollmien's.
original calculation3 than with Schlichting's later one.4 ' The maximum
value of a is even larger than Tollmien's value, while Schlichting's value is
smaller.
4. Physical Interpretations and Concluding Discussions.-We shall now
try to understand the physical mechanism underlying the above results.
Generally speaking, the investigation in an inviscid fluid brings out the
r61eof pressure and inertial forces. The subsequent investigation of viscous
forces will then settle the stability problem of a real fluid.
Equation (2.1) is the equation of vorticity. This suggests that the con-
ditidn w' = 0 in Rayleigh-Tollmien's criterion means an extremum of vor-
ticity of the main flow w(y). In fact, the physical mechanism can be
understood by following this line of thought. Let us regard the field'of flow
as vorticity distribution in parallel layers, and imagine a disturbance con-
sisting of the interchange of two fluid filaments in two different layers.
These fluid elements will retain their original vorticities, and will therefore
be present in layers with vorticities different from their own. It can be
shown that these filaments will then be pushed back to their original layers
if the vorticity gradient of the main flow has the same sign.12 The flow is
therefore stable. Indeed, if we consider an excess concentrated distribution
of vorticity &(x,y) (a "vortex") upon a small region of a given field of vor-
ticity, ~o(y), then the vortex is accelerated with an average acceleration

a = r. 2(x, y)'o'(y)dxdy (4.1)

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
322 PHYSICS: C. C. LIN PROC.
N. A. S.

in the direction of the positive y-axis, where I' is the total strength of the
vortex, v(x, y) is the y-component of velocity due to it, and the integration
is taken over the whole field of flow. A consideration of the sign of r and
Po'(y) will then substantiate the above statements. Thus, the stability in
an inviscid fluid as governed by inertial and pressure forces can be discussed
in terms of the gradient of vorticity of the main flow.
Another interpretation of the significance of the gradient of vorticity is
to be found from Kelvin's "cat-eye" picture of a flow with neutral dis-
turbance.13 If we consider the vorticity of fluid elements along a closed
stream line in that picture, we see that such a flow pattern is not possible
without diffusion of vorticity by viscous forces, if the gradient of vorticity
is finite at the critical layer where w = c.
This consideration immediately leads to a simple method of visualizing
the extent of viscous forces in controlling the stability of fluid motion. If
we consider the diffusion of vorticity from the critical layer for a period T
of a neutral disturbance, the effective distance covered would be of the
order of (vT)1/2, v being the kinematical viscosity. The ratio of this dis-
tance to that between the critical layer and the solid boundary can be
easily shown to be

s= 2 (4.2)
where z = - auo0 is directly related to the argument of the Hankel functions
of Section 2, wi being the value of v at y = yi. This parameter s has the
value 0.7 at infinite Reynolds number on the lower branch of the neutral
curves of figure 1. It decreases to about 0.5 at minimum Reynolds number
and further decreases to zero as R becomes infinite along the upper branch.
Referring to those figures, we note that for small values of s (small viscos-
ity), the effect of viscosity is essentially destabilizing, since an increase of
Reynolds number gives more stability (cf. Heisenberg's criterion of Section
3). For large values of s (large viscosity), the opposite is true. The physi-
cal mechanism can be described in the following way. It is known that the
destabilizing effect is caused by phase shifts of the disturbance, which tends
to build up its shear -pu'v' (u', v' being the components of velocity of the
disturbance, and p the density of the fluid). This, in turn, transfers the
energy of mean flow into that of small oscillations. The stabilizing effect
is due to dissipation. Now, for small values of s, we may think of the flow
as having two thin viscous layers, one at the wall, the other at the layer
where w = c; the effect of dissipation is relatively unimportant, and the
resultant influence of viscous forces is destabilizing. When s exceeds 0.5,
these two layers may be regarded as having joined each other into one vis-
cous region; the effect of dissipation becomes important, and the resultant
influence is stabilizing.

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
VOL. 30, 1944 PHYSICS: C. C. LIN 323

The viscous forces upset the decisive nature of an extremum of vorticity

for the determination of the stability of the flow of a real fluid, as one can
see from a comparison of figure 1 (a) (no extremum) and figure 1 (b) (with
extremum). Thus, there is no drastic change in the stability characteris-
tics of a boundary layer as the pressure gradient changes from a slightly
positive value to a slightly negative value (as one would expect if one tries
to apply Rayleigh-Tollmien's criterion to a real fluid). However, the ap-
pearance of the point of inflection in a velocity curve does indicate an in-
crease of instability. It can be shown, by using (3.3), (3.4) and (3.5), that
the critical Reynolds number will be decreased. Also, for a given Rey-
nolds number, the range of wave-lengths of unstable disturbances will be
increased. For a boundary layer with adverse pressure gradient, a further
indication of increasing instability is given by the decrease of wl. Indeed,
as w[ -> 0, the approximate formulae (3.4), (3.5) give c -- 1, R 0. This
is, of course, an extrapolation of those formulae, but the conclusion stands
in qualitative agreement with the usual notion that the flow would be vio-
lently unstable in such a case (separation).
5. Transition to Turbulence.-Finally, we want to touch briefly the ques-
tion of transition to turbulence. A linearized theory of hydrodynamic
stability is of course not quite adequate to account for such a phenomenon.
But if we allow our theory to be extrapolated to the beginning of non-linear
effects, and assume the subsequent development into turbulence to be very
rapid, there is the possibility of determining approximately the Reynolds
number of transition. For flow in boundary layers, it is interesting to note
that the disturbance in the free stream has an isotropic nature in two di-
mensions. There may be therefore the possibility of introducing the re-
sults of the theory of instability in boundary layers into Taylor's theory of
the transition Reynolds number,5 which appears to be quite successful
when the transition is caused by the turbulence in the free stream.
The author wishes to express his sincere gratitude to Professor Theodore
von Karman for suggesting this problem and for his invaluable help
throughout this work; to Professor Clark B. Millikan for helpful sugges-
tions and for using some of his unpublished notes, from which much inspira-
tion and suggestion has been derived; to Professors P. S. Epstein, H. Bate-
man, Dr. Hans W. Liepmann and several of his friends for many valuable
discussions and suggestions.
* The present article is a brief
report of some results contained in the author's disser-
tation approved by the California Institute of Technology for the degree of Doctor of
Philosophy.
1Heisenberg, W., Ann. d. Phys., 74, 577-627 (1924), especially p. 599.
2Tietjens,0., Zeitz.Angew.Math.Mech.,5, 200-217(1925).
21-44 (1929).
3 Tollmien,W., GbttingerNachrichten,
4 Schlichting, H., Ibid., 181-208 (1933).

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions
324 PHYSICS: G. D. BIRKHOFF PROC.N. A. S.

5 Taylor, G. I., Proceedingsof the Fifth InternationalCongressfor Applied Mechanics,

Cambridge, Mass., U.S.A. (1938), pp. 304-310, especially equation (45), p. 307.
6 Synge, J. L., SemicentennialPublicationsAmer. Math.Soc., Vol. II: Semicentennial
Addresses,pp. 227-269 (1938); especially equation (11.23), p. 258.
7 The conclusion that the plane Poiseuille flow may be unstable with respect to certain
disturbances is in contradiction with that of Noether, F. (ZaMM, 6, 232-243) (1926),
and Pekeris, C. L. (Jour. Aero. Sci., 5, 237-240) (1938). The untenable points in their
investigation are discussed in the present investigation.
8 Squire, H. B., Proc. Roy. Soc. London, A142, 621-628 (1933).
9 Rayleigh, Lord, ScientificPapers, Vol. 1, pp. 474-487.
10Tollmien, W., GottingerNachrichten,Neue Folge,1, 70-114 (1935).
11Rayleigh's treatment of this problem using broken linear velocity distributions has
been denouncedby Heisenberg, but his proof of the necessary condition is acceptable,
except for the modification to be discussed below.
12 The fact that an element with excessive vorticity is accelerated toward the region
of higher vorticity was recognized and suggested to the author by Th. v. Karman in con-
nection with the theory of vorticity transfer in a fully developed turbulent flow.
13Kelvin, Lord, Math.Phys. Papers, Vol. 4, pp. 186-187.

FLAT SPA CE-TIME AND GRA VITA TION

BY GEORGED. BIRKHOFF

DEPARTMENT OF MATHEMATICS, HARVARD UNIVERSITY

CommunicatedSeptember7, 1944

If one admits that physical events take place in a 4-dimensional space-

time continuum (an idea abandoned in current quantum-mechanical the-
ory) there are three interesting possibilities: classical space and time;
flat or electromagnetic space-time; curved space-time. The appropriate
corresponding mathematical languages are, respectively, those of 3-vec-
tors, 4-vectors and 4-tensors.
In a certain sense the flat space-time, characteristic of the so-called special
theory of relativity, is just as absolute as classical space and time, since the
co6rdinates t, x, y, z require exactly 10 arbitrary constants for their com-
plete specification in both cases. But, in the framework of flat space-time,
the fundamental electromagnetic equations of Maxwell and Lorentz lose
the artificiality which they possess in classical space and time.
The initial attempts to incorporate gravitational phenomena in flat
space-time were not satisfactory. Einstein turned to the curved space-
time suggested by his principle of equivalence, and so constructed his gen-
eral theory of relativity. The initial predictions, based on this celebrated
theory of gravitation, were brilliantly confirmed. However, the theory has

This content downloaded from 169.229.32.136 on Wed, 7 May 2014 17:15:01 PM

All use subject to JSTOR Terms and Conditions