Sei sulla pagina 1di 28

ANNUAL

REVIEWS

Further

Quick links to online content

LAMINAR SEPARATION
By S. N. BROWN AND K. STEWARTSON
Department of Mathematics, University College London

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

INTRODUCTION

T he p henomenon of separation is one of the most interesti ng features


of the motion of an i ncompressible fluid past a bluff body at high Reynolds
number. Here the mai n stream, which has hitherto been in close contact with
the body, suddenly, and for no obvious reason, breaks away, and down
stream a region of eddyi ng flow, which is usually turbul ent even if the flow
elsewhere is laminar, is set up . Originating with Prandtl (1), repeated
careful experiment and observation reveal th at in two-di mensional flow the
breakaway of the main stream occurs at or very near the point S on the body
where the skin friction T vanishes (also conventionally referred to as the
separation poi nt) . This suggests that the p henomenon is closely related to
the boundary layer upstream of S. Such an intimate relation has not of
course been established beyond a shadow of doubt, either experimentally or
theoretically. However, the contrary view, that separation and boundary
layer are unrelated, leaves the breakaway problem as a complete mystery,
with no clue to i ts resolution given by experiment or theory. The view we
adop t here is that the two are closely linked, and we hope to show that it
forms the basis of a theory which gives a depth of understandi ng of the
p henomenon and promises more in the future.
The main problem we have to face, if we believe that the boundary layer
is intimately associated with separation, is how its thickness can change
abruptly from 0 (1/2) upstream of S to 0 (1) downstream, especially as the
Reynolds nu mber R may be scaled out of the boundary-layer equations.
The only way in which this can happen is by breakdown of the solution of
the boundary-layer equations, for otherwise the boundary layer would have
a finite thickness in the scaled variables upstream or downstream of S. The
possibility of such a singulari ty was first noted by Goldstei n (2) i n his study
of the solution near a poi nt of zero skin friction, and in the first part of the
article we shall discuss the develop ments that have grown out of his basi c
idea. A few preli minary remarks are appropriate.
First, i n any study of the boundary layer, it should always be borne
in mind that such an i nvestigation is an asymptotic theory valid, at best,
only in the limit R 00. Thus when we assert that the boundary layer is
singular at separation S we do not imply that, in a real flow at finite Reynolds
number, the skin fricti on T is nonanalytic at S. That would be ridiculous.
What we mean is that the function
lim (RII2r)
Roo

45

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

46

BROWN & STEWARTSON

which is generally bounded upstream of S, is nonanalytic at S. There is


then no contradiction between this theory and results such as Dean's (3)
showi ng that the N avier-Stokes equations are analytic at S. The limiti ng
processes are different. Second, the nonanalyticity of T at S does not imply
that the main stream breaks away from the bluff body there; other possi
bilities could occur. However it is clear that a dramatic change takes place i n
the boundary layer a t the point o f zero skin friction, and this is not incon
sistent with the view that it increases rapidly in thickness j ust downstream ;
by what means is still unknown. Third, at present there is no adequate rig
orous theory of the solutions of the N avier-Stokes equations for flow past
bluff bodies at high Reynolds number; indeed, if we are correct in believing
that the limit flow as R--'> 00 contains pathological features (see paragraph 2
of Part I, below) , then, the prospects for developing one are gloomy. Fur ther ,
fr om the practical aspect the possible onset of instability at values of R100
cannot be ignored. Failing a rigorous theory we are restricted to consistency
arguments of the type: formulate an asymptotic expansion, and if every
term is obtainable, i f only in principle, then the expansion is the correct one.
S uch an argument enables us to exclude the classical irrotational flow as
a limit solution as R--'> 00. For if it were a limit solution, then to make i t
a uniformly valid first te rm o f a n asymptotic expansion, the boundary
layer would have to be joined to it. However , the boundary layer has a
singularity at S and we have a contradiction. No other mathematical argu
ment for excluding this irrotational flow is k nown.
Although we are inclined to believe that the singularity plays a decisive
role in the breakaway of the main stream from the body, it is not essential
that the solution of the boundary-layer equations always be si ngular at S.
It i s quite possible for the solution to be regular there, and indeed it m ay b e
s o in such phenomena a s long or shor t separation bubbles . However, i t i s
when w e turn t o compressible fluids, and in particular t o supersonic flows,
that we see the most spectacular examples of regular separ ation. This occurs
in the free-inter action zone of shock-wave boundary-layer interactions: here
the flow separates quite suddenly and spontaneously, and in addition the
boundary layer downstream , while thickening in comparison with upstream
is, for some way at least, still fairly thin. The mechanism seems to be that
the boundary layer thickens, producing an exter nal pressure rise which
reinforces the thickening process, until separation occurs. I t appears that, in
general, the boundary laye r m ust exert a significant influe nce on the main
stream outside if regular flow at S is to be preserved. Downstream a new and
challenging problem is posed to computers and theoreticians, for the velocity
profile is partly reversed, permitting upstream propagation of disturbances.
Established methods of numerical integr ation b ecome unstab le and the
solution is probably not unique without some sor t of additional boundary
condition. We shall investigate regular flows in the second part of the
article.
-Considerations of space unfortunately prevent us from discussing the

LAM I NA R SEPARATION

47

whole theory, and so we shall largely confine ourselves to exact analyses


in either t he mathematical or numerical sense. However, in the second part
we shall include some approxim ate studies, partly because of t he lack of
precise t heories, and partly because of the interest and controversy they have
generated.

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

PART I. FLOWS SINGULAR AT SEPARATION

Two-dimensional incompressible flows.-The possibility of the occurrence


of a singularity in the solution of t he steady laminar boundary-layer equa
tions at separation is not surprising. These equations and boundary condi
tions are t he result of a double limiting process as t he kinematic viscosity v
tends to zero and t he time t tends to infinity. That t hey are not applicable to
t he post-separation flow past , for example, a circular cylinder (in which region
it is doubtful whether a steady solution exists) is t herefore understandable.
Attempts to clarify t he situation of infinite time and small but nonzero
viscosity have been made by reintroducing the time dependence and t racing
t he flow pattern from its initial, say impulsively-started, state. The work of
Blasius (4), and Goldstein & Rosenhead (5) , is now classic. At first t he bound
ary layer is t hin, and the displacement t hickness is O( yivt). Separation first
occurs at t he rear stagnation point when U",t/a=O(1), where U", is the
undisturbed fluid velocity and a t he radius of the cylinder, and then S, t he
point of vanishing skin friction, moves upstream. Although t he initial-value
problem reveals no peculiarity in the character of the point S, we infer t hat
its limiting position for large time divides t he flow about t he cylinder into
two parts: upst ream of S t he flow is potential except for a t hin viscous layer
whose displacement t hick ness, when divided by V1/2, tends to a fi nite l imit as
vt-+ <Xl, while downstream of S the notion of a boundary layer fails and t here
is probably no steady sol ution. S uch a prediction is not inconsistent with t he
conclusion of Proudman & Johnson (6), who have made an important
contribution towards t he understanding of t he flow near the rear stagnation
point of such a cylinder. Their solution of the unsteady Navier-Stokes
equations shows t hat at large times t here is an inner boundary layer of
reversed flow and the stagnation point is one of attachment. Outside this
inner layer there is a region in which the velocity changes sign to that of the
main stream . This region, in which the fluid accommodates itself to t he inner
attaching-stagnation-point flow and to the outer, main-stream rear-stagna
tion-point flow, grows exponentially with time, and leads to a displ acement
t hick ness t hat is o (aR-l/2eT), where t he Reynolds number R= U",a/v and
T= U",t/a. Thus downstream of separation t he displacement thick ness
cannot be expected to tend uniformly to a fi nite limite as t-+ <Xl. The situa
tion is illustrated in Figure 1.
The ultimate flow for large time will be very complex, and one might
conjecture that t he flow pattern resembles that of Figure 2. Once t he un
steady flow in t he neighbourhood of t he rear stagnation point Tl of the
cylinder has reversed, and the position of zero sk in friction, which at

48

BROWN & STEWARTSON

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

FIG. 1. Streamlines in unsteady rear s tagnation flow.

first coincided with it, has moved upstrea m to Plo say, then TI is a point of
a ttachment and the boundary layers associated with it must eventually
themselves include a region of reversed flow star ting at P2 Between PI and
P2 there might be another point of a ttachment T2 w hich initia tes boundary
layers that separate at Pa. The for m of the limit solution, and even the
dependence of the len gth of the resultant wa ke on Reynolds number, is in
doubt, and although it is an interesting proble m we shall not discuss it
fur ther here. A fuller discussion is given by Stewar tson (7) . It is, however ,
o bserved tha t if the boundary layer is relevant to the limit steady solution of
the Navier-Stokes e qua tions for the flow past a bluff body as R-t 00, then i t
is unlikely that the Kirchhoff free-strea mline model correctly describes such
flows. For example, the mode l predicts that w hen the body is a circular
cylinder the pressure gradient upstrea m of separation is entirely favourable .
When the flow is two-dimensional and incompressible , the development
of a steady la minar boundary-layer with a prescribed pressure gradient is
well understood. If x , y denote coordinates parallel and normal to the wall
and u, v the corresponding components of velocity, the equations of motion
are

att av
att
att
1 dp,
a2tt
-+-=0 , tt-+v-=---+v
ax oy
ax
ay
PI dx
ayZ

1.1.

with u = v = 0 on y = 0 for an i mper meable wall, and U-'> UI (x) as y-'> 00 Here
PI is the pressure, PI the density, and UI(X) the main-stream velocity

FIG. 2.

49

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

related by PI + Cl/2)pl UI2 = constant. Since the governing equations are


parabolic it is also necessary to specify an inital profile for u at some station
of x, usually a stagnation point of the external flow. By analogy with the heat
equation, the flow properties at any point will be independent of conditions
farther downstream as long as u remains positive, and the solution exists and
is u nique [Nickel (8), Oleinik (9)]. The forces acting on the fluid in the
boundary layer are three in number, the main stream in the direction of
positive x, the skin friction in the direction of negative x, and the pressure
gradient. Thus, if the pressure decreases in the direction of flow , the skin
friction cannot vanish and the boundary-layer solution continues unim
peded. However if the pressure gradient is positive, zero skin friction is a
possibility, reversed flow sets in, and the computation may be expected to
break dow n if only because of incompleteness of the bou ndary conditions .
Further discussion of this dilemma may be fou nd in Par t I I .
With a presc ribed positive pressure gradient the solution o f the boundary
layer equations usually has no opportunity to feel any void in boundary
conditions after separation, for in general it exhibits a singularity at the
point of vanishing skin friction, and this marks the end point of the calcula
tion. It should be borne in mind that E quations 1.1 also admit of a solution
regular at the point 5, X = X., y = O , at which u=8u/8y=0. Differentiation
of the equations with respect to y gives [Curie (10)]

ax

((CiU)2)
2

Ciy

Ci4u
ay4

w hen y = O . If 84ujay4 vanishes at 5, the skin friction


nonzero then

au)
-

Ciy

._0

ex

(x, -

1.2.
is

regular, but if

X)I/2

it is

1.3.

From consideration of Equation 1.2 alone a third possibility would appear to


b e that 84u/8y4 is infinite at separation. However we shall see that this does
not arise, and that the analytic solution implies either regularity or a sin
gularity of the form (Equation 1.3), w hile numerical investigations favou r
the latter.
An expansion for the stream function if; that leads to an expression for the
skin friction of the form of Equation 1.3 was first proposed by Goldstein (2).
In it
if;

(x.

- X)3/4 ntO (x. - x)n/'1n ( (x. -Y x)1/4)

1.4.

the exponents 3/4, 1/4 being appropriate if, as is to be expected, the pressure
gradient is a controlling feature of the flow near separation. The b oundary
conditions on the functions in of Equation 1.4 are, w hen there is no suction,
in CO) =In/(O) = 0, where the prime denotes differentiation with respect to
TJ=y/(x.-x)li4, together with the requirement that no in must b e exponen-

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

50

B ROWN & STEWARTSON

tially large as TJ-.7 00 in order to locate any singularity on the wall. Later
Stewartson (11) showed that to satisfy the imposed conditions it is necessary
t o add terms in I n x to the expansion (Equation 1.4) . In t he two-dimensional
problem, of w hich a lucid and comprehensive account, including the effects
of su ction, is given by Terrill (12),16 must be replaced by 15+ F6 In x, and
similarly with subsequent terms when required. The final solu tion contains
an infinite nu mber of arbitrary constants, w hich presu mably are determined
by the flow u pstream of S. For a circular cylinder with Ul(X) = U", sin (x/a)
t he numerical work of Terrill gives, for appropriate choice of t he leading
arbitrary constants in Equation 1.4, a convincing match with t he analytic
solution valid in the immediate neighbou rhood of S, and shows, almost con
clusively, that t here is a singularity at separation. Of cou rse it is analytically
possible that the coefficients of all the irregular terms in Equation 1.4 are
zero, and it may be that, w hen given t he main-stream velocity, this non
singular form is the one adopted by the fluid, whatever the conditions at the
initial station. However there is no firm evidence to support this point of
view.
I n addition to the numerical investigations of Hartree (13) , Leigh (14) ,
and Terrill (12) , in which it was not possible to carry the integrations past
the separation point, further su pport for the existence of a singularity is
given by the series-expansion methods of Howarth (15) , and Gortler (16).
The expansions failed to converge near separation, and Gortler also noted
that b ecause of the singularity in the solution at separation w hen x0 . 16
for the main stream Ul(X) = 1/(1+x) , the series for the case U1(x) = l/(l-x)
also has a limited radius of convergence since, although there can be no sepa
ration, the associated series will have a singularity on the negative real axis
due to its similarity to that for U1(x) = 1/(1+x). Further supporting evi
dence in favour of a singularity is given by the solutions of the boundary
layer equations that arise w hen U , (x ) C( xm (m constant) . The velocity u is
given by u= U1(x) f'(TJ) w here TJ C( yx(m-n/2 and f(TJ) satisfies t he Falkner
Skan (17) equation
fill + ff" + (1
m

f'2)

1.5.

with f(O) =f'(O) = 0 and f'( 00 ) =1. The solution for which the skin friction is
zero hasf"(O) =0 with m=mo= -0.0904 and f"(0) C( (m-mo)'/2 near m=mo.
This branch point at m=mo divides the solutions for which f"(TJ) >0 for all7J
from those that have a region of reversed flow near the wall. The latter solu
tions, fou nd by Stew artson (18), are discussed i n Part II, as they are believed
t o have application to post-separation flow . Even though main streams of the
form xm are largely academic, since they become infinite as either X-.70 or
X-.7 00, and there is no rigorous confirmation of the assu mption that an actual
main stream can be replaced by a succession of such profiles with different
values of m as x increases, t he above resu lt is certainly not inconsistent with
the prediction of a singularity at separation.

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

51

Two-dimensional compressible flows.-The conclusion that the point of sepa


ration is singular if the pressure gradient is prescribed for a two-di mensio nal
i nco mpressible bou ndary layer cannot unreservedly be carried over to co m
pressible flow. However , if the wall is ther mally i nsulated, and the Prandtl
number is unity, the IlIingwor th-Stewartso n transfor mation (19, 20 ) reduces
the momentu m e quation to i ncom pressible for m, and all the theory of in
compressible boundary layers has i mmediate application. The effec t of
Prandtl nu mber is expected to be of degree r ather than c harac ter , and it is
o nly necess ary to recall that the x,y coordinates are altered by the transfor
mation, and that the mai n stream of the corresponding inco mpressible flow
is proportional to the Mach number of the co mpressible flow. However if
heat transfer is per mitted, the momentu m and energy equations cannot be
u ncoupled. Cooling the wall may be expected to delay separation while heat
ing would encourage ear lier breakaway as the lighter fluid more easily loses
contact with the bou ndary. A similar expansion to Equation 1 .4 in this sit
uation [S tewar tso n (21)] leads to the resu lt that if the heat transfer is non
zero at S, the point of vanishing s ki n fr iction, none of the singular ter ms can
occur, and this point mus t be a regu lar point. A singularity is possible if the
heat transfer vanis hes at S, even if it is nonzero elsewhere.
This r ather surprisi ng co nclusion, namely, that heat transfer precludes
a singularity at S, has given rise to an interes ting co ntroversy. Early co mpu
tations by Poots (22) su pported it as do the similarity solutions [Stewartson
(20 ) ; Cohen & Reshotko (23)]. However , Williams (24) , who r epeated Poots's
calculation bu t to a higher accuracy, found c lear evidence of a singularity at
S. Failing an error in the analytic study there are two possible interpretatio ns
of the contr adiction. One alternati ve is that the singularity in the nu merical
wor k is o nly apparent and is s moothed out within a distance of S so s mall
that it is inaccessible to the co mputer. It is noted in this co nnectio n that the
assu mptio n of a singularity i n the analytic s tudy does no t i mmediately lead
to a contradiction, sugges ting that the struc ture is r ather subtle. The o ther
possibility is that the heat transfer ac tually vanishes at S, in which case the
analysis is inapplicable . However , i n Williams' s tudy (24) , although the heat
transfer is i ncreasing r apidly at S, he is of the opinion that it is still sub
s tantially negative at S. A so mewhat mor e equivocal co nclusion was drawn
by Sells (25) in a par allel study.
Ano ther interes ting counter-example to Ste wartso n's analytical resul t
has been given b y M er ki n (26) who considers the convection problem of a
semi-infinite ver tical plate i n a parallel u nifor m s tream. If x measures dis
tance along the plate with origin at the leading edge, which is at the lowest
point, and y measures distance normal to the plate, with u,v the corres pond
ing components of velocity, the equations of motion are as Equation 1 . 1
with g{3(T T",) replacing (l/PI) (dPI/dx), and the equatio n for the te mper
ature Tis
aT
aT
a2T
U-+V-=K1.6.
ax
ay'
iJy
-

BROWN & STEWARTSON

52

I n the above g deno tes gravity, K the coefficient of conductivity of the fl uid,
and {3 that of thermal expansion. T he temperature of the fl uid at large dis
tances f ro m the plate is T"" while t he plate is kept at constant temperature
Tw. The bo undary conditions on the velocity are

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

0:

u = v =

0;

y ->

00: tt ->

U.

1 .7 .

Near the leading edge the flow i s that o f Blasi us, while the effect of buoyancy
comes i n as x i ncreases. If the wall is heated with Tw > T", the buoyancy acts
like a favo urable pressure gradient and the flow does not separ ate. However,
for a cooled wall the buoyancy has a reverse effect and separatio n occurs.
M erki n shows numerically that the position of zero ski n friction is given by
g(3(T", - T.,)x
1.8 .
= 0.1924
UI
and that the approach to separation is singular, even tho ugh the heat trans
fer does not vanish at that poi nt.
Separation of three-dimensional boundary layers.-In three dimensio ns a
boundary layer starts fro m either an attachment point or line o n the surface,
and the bo undary conditions must i nclude specificatio n of the veloci ty
co mponents parallel to the boundary, and the temperature, at all poi nts of
the normals at attachment. With this infor mation, together with the usual
conditio ns on the surface y 0 and as y-> 00, the calculation can proceed by
a marching process into a zone D which contai ns o nly streamlines that have
entered it ei ther fro m the main stream or by crossing the normals at attach
ment. Such a pr6cedure was adopted by Der & Raetz (27). Experi mental
evidence indicates that the downstream edge of such a zone coi ncides with
the breakaway of the main stream from the wall, and we may i nter pret this
=

as three-dimensional separation.

I t is possible that one is hindered i n any i nvestigation of three-di men


sio nal separation by preconceived ideas carried over fro m two-dimensional
theory. In two dimensions the flow separates at, or very near, t he poi nt
where the streamwise skin friction vanishes, and if we consider it as a three
dimensional flow then there is a whole line of poi nts o n which bo th co m
ponents of the ski n frictio n vecto r (T""T.) vanish. In truly three-dimensio nal
flow such a line cannot be expected to occur, and both co mponents of skin
fri ction will, i n general, vanish only at isolated poi nts o n the surface. It is,
however, too restrictive to fix attention o n these poi nts alo ne i n any attempt
at a mathematical definitio n of separation, since many flows appear to sepa
rate along a li ne on which the parallel co mponent of ski n frictio n is no t every
where zero. Stewartson (28) defines a separ ation li ne as a curve on the body
dividing those points that are accessible to the streamlines entering the zo ne
at attachment from those points that are inaccessible from attachment. This
definition has the advantage of embracing both two- and three-di mensional
flows and is appropriate whether t he separatio n skin friction is singular or
otherwise.

LAMINAR SEPARATION

53

A useful aid to understanding in the study of three-dimensional separa


tion is the concept of limiting streamlines as discussed by, among others,
Maskell (29) and Lighthill (30). These are the skin-friction lines or the
streamlines nearest the wall in the boundary layer and they have the equa
tion
dx

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

Tz

dz
Tz

, y

1.9.

where x,z are coordinates in the surface. The "singular points" of the differ
ential equation (1.9) are the points where both components of skin friction
vanish, and these will be saddle or nodal points of attachment or separation.
Lighthill (30) presents illustrations of possible topographical configurations
of the limiting streamlines on various surfaces, and defines a line of separa
tion as "a skin friction line which issues from both sides of a saddle point of
separation and, after embracing the body, disappears into a nodal point of
separation." However an alternative point of view also stems from the fol
lowing observation by Lighthill. If h is the distance between two adjacent
limiting streamlines then, by considering a streamtube of rectangular section
whose base is the portion of surface between these streamlines and whose
height is y, he notes that
1.10.

where c is the volume flow along the streamtube and is constant. Equation
1 . 10 shows that there are two mechanisms that can cause the streamlines to
leave the surface. The ordinate y increases if either 7,,=7;,=0 and both
components of skin friction vanish, or if h becomes very small and the lim
iting streamlines run close together. Thus the flow may also separate along
the envelope of the limiting streamlines, and we have a definition of separa
tion additional to the one proposed above. Lighthill warns the researcher
against an uncritical acceptance of this latter view of separation. We are
inclined to favour it however, as it is supported by experiment and by the
few numerical investigations that have been undertaken, and it is consistent
with many of the flow fields in the informative illustrations of Maskell (29).
These alternative proposals for separation are illustrated for the case of a
symmetrically disposed ellipsoid in Figures 3a and 3b.
The question whether the envelope of the limiting streamlines coincides
with a singularity in the solution of the three-dimensional boundary-layer
equations is presumably subordinate to the question whether the separation
line is an envelope of limiting streamlines. However it is helpful to consider
them in conjunction. We note that if algebraic singularities are permitted,
the skin-friction lines may meet the envelope at finite points, but if the skin
friction is regular they do not intersect their envelope. The two questions
can at present be answered only for a few cases of quasi-two-dimensional
flow. Support for the concept of the separation line as the envelope of the
limiting streamlines is supplied by a numerical integration of the equations
for the incompressible flow of a conical main stream past an inclined cone by

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

54

BROWN & STEWARTSON

FIG.3a. One possibility for limiting streamlines for a symmetrically disposed ellipsoid.

Cooke ( 3 1 ) . In this case the s eparation line is the line through the apex on
which the transverse component of s kin friction vanishes , and is , as shown
by Brown (32) for the closely related problem of a delta wing, a line of singu
larities of the type p roposed by Goldstein ( 2) . This conclusion is also sup
ported by the nu merical evidence.
Another three-dimensional theoretical and numerical investigation has
recently been u ndertaken by Banks & Williams ( 33) . The p roblem is that of
a yawed cylinder with a main stream that has components U1(x) = 1 -x,
WI= 1, and one of the ques tions the authors set ou t to answer is whether the
spanwise s kin friction Tz vanishes along the line on which the chordwise s kin
friction T ", is zero. The numerical results , which are matched convincingly
with a Goldstein type expansion about the position of zero chordwise skin
friction in spite of an infinite value of iJ4w/iJy4 , indicate that Tz;eO so that the
-

FIG. 3b.

Alternative possibility for limiting streamlines for a


symmet rica ll y d ispo sed e llip soid .

5S

LAMINAR SEPARATION

components of skin fric tion are not zero simultaneously. Here again the
separation line is the envelope of the limiting streamlines.
Little progress has as yet been made with the problem of general three
dimensional separation . Apar t from the enormous computational difficulties
of three (or in the heat-transfer situation four) interdependent equations,
there is the added complica tion that the limiting streamlines will not , in
gener al , be parallel to the main strea m. This presents new problems for the
marching process mentioned earlier , since the direction for the integra tion
procedure is no longer well defined.
Flows with suction and i njection. Intuiti ve ly one would expec t tha t it
would be possible to delay separation by drawing fluid into the plate , and
that ex pelling fluid from it into the boundary layer would advance separa
tion , and indeed the evidence available indicates tha t this is so. When the
main-stream and suction velocities are suitably related powers of x , the
boundary-layer equa tions reduce to the Falkner-Skan Equation 1 . 5 with

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

f(O) = C,

f'(O) =0,

1'( J) = 1

1.11.

Here C >0 corresponds to suction and C<O to injection. The discussion of


the solution of the equation with these boundary conditions given by Jones
& Watson (34) shows that for m <0 and a retarded main strea m the effect of
increasing C is to decrease the value of m a t which f" (0) =0. Also if m >0 then
1"(0) >0 for all finite C and the skin friction remains positive. Although no
deduc tion about the progress of a real boundary layer proceeding from at
tach ment to separation can be made directly from the properties of the a bove
similarity solutions, the conjecture that suction delays se para tion is con
firmed by the n u merical investigation of TerriII ( 1 2) for U1(x) U",sin(x/a) .
In addition, the analysis of Nickel (8) establishes that n o amoun t of injection
can cause separation in an accelerated flow.l
A related problem is the deter mination of the strength of the suction
velocity Vo required to prevent separation altogether in a retarded flow.
When the suction is large , though of order R-l/2 to be compati ble with the
assumptions of boundary-layer theory, the velocity distrib ution within the
boundary layer i s
=

UX) =

exp

(-:OY)

1.12.

This is the asymptotic suction profile, which of course does ot separa te.
1

It is interesting to speCUlate on the possibility of singularities in an accelerated

flow. If a fixed disc is place d normal to the axis of a potential vortex, the main stream
is azimuthal and proportional to l/r. A similarity form of the equations and boundary
conditions exists but has no solution. Since the relevance of the similarity equations
would be expec ted to be confined to the immediate neighbourhood of r=O, does the
nonexistence of a sol ution imply that the flow has separated in some sense? From
physical considerations some sort of eruption at the centre certainly seems plausible.

56

BROWN & STEWARTSON

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

Per turbations about this solution for various forms of the main stream were
considered by Watson (35). For UI(X) = Uo(1-x/l) (which is the main
stream appropriate for flow i n the neighborhood of the rear stagnation poin t
o f a cylinder, 1 being a representative length), the associated s imilarity equa
tion has no solution for any value of the suction velocity, and we deduce th at
no amount of suction will prevent separation. In this case Equation 1.12
cannot be a uniform ly valid firs t approxim ation to th e solution , th ough it is
not clear where th e expansion breaks down. A numerical investigation by
Williams (24) indicates th at, with th is linearly r etarded main stream, the
separation point x. is given by

exp

kvo

(J 1']

1.13.

where a, k are constants and k:::::: 2.1. No analytical confirmation of this resu lt
has yet been obtained, but it is in direct contradiction with the r esults of
M orduchow & Reyle (36) who predict that a finite value of the suction
velocity is sufficient to prevent separation. However we believe th at their
Pohlhausen tech nique, involving though it did seventh-degr ee polynomial
velocity profiles , is too blunt an ins trument to settle such a delicate question.
For a two-dimensional boundary layer with a retarded main s tream the
effect of suction or injection on the singularity at separation is to leave es
sentially unaltered the for m proposed by Goldstein (2). This is demonstrated
by Terrill (12). However , if the main stream has cons tant velocity U and
fluid is injected into the boundary layer , Catherall, Stewar tson & Williams
(37) encou ntered a singularity at separation of an entirely different ch ar
acter. I n this situation there is no possibility of a solution regular at the
point of zer o s kin friction, which is u nlike the retarded-main-stream case,
where the coefficients of all th e singular terms i n th e expansion about that
point can be zero. The appropriate form for the s kin friction as XX8 is found
by matching the solution i n a region in which u,...x
., .-x to an inner region in
which u,...(., X.-X)2 and is
T'"

( x* )2
In x*

1.14.

where x* is the nondimens ional distance from separation. The numerical


r esults of the i nves tigation gave x.::::::O.7456 UlI/Vo2 as th e position of s epara
tion, but s ince it is necessary that both x*l and j lnx* j 1 for th e
validity of the result of Equation 1 . 14, it was impossible to match the limit
ing for m of the s kin fr iction to this result in an entirely satisfactory m anner
as the computation could not be extended to such s mall values of x.-x.
Nevertheless th e numerical solution shows plainly that dT/dxO as xx.,
instead of tending to infinity as in the case of adverse-pressure-gradient
separation. Also uO as XXB for any finite value of y on th e separation line;
both of these conclusions are consistent with th e analysis .
Fluid properties in the neighbourhood of separation.-The expansion of

57

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

Equation 1.4 is an asymp totic expansion valid in the immediate neighbour


nood of t he separation point , and it has been found p rofitable to m atch it near
t he wall to t he numerical solution valid upstream of this point, g iving , at
least in t he incompressible case, very satisfactory agreement. In addition to
this expansion, however, Goldstein (2) presented an intermediate expansion
which effected a complete analytic match between this solution and the main
stream. Apart from t he importance of the result in any estimation of t he
effect of the separating boundary layer on the flow outside, the work is of
considerable interest as being one of the earliest ex amples of the use of t he
method of matched asymptotic expansions. If u.(y) is t he separation p rofile,
so that
as

u->u,(y)

for y

x->x,-

>0

1.15.

then from Goldstein's intermediate expansion it follows that the velocity


upstream is given by
u(x, y)

r(x)

tl.(Y) + -/-

duo
+ O(x* In x*)

dPI dx dy

1.16.

Here r(x) is defined by r(x) = JJ.au/ay)v=o, and x* is the nondimensional dis


tance to separation. There is no formal difficulty in continuing t he expansion
of Equation 1 . 16 since the possible singularity in u.(y) at x= x. [ noted by
Stewartson ( 1 1)] that arises from a term y ln x* in the series for the stream
function corresponding to Equation 1.4 is in fact spurious. The term that is
logarithmic in x. - x is exactly cancelled by another at the next stage of the
inner expan sio n and the result is a term that is logarithmic in y so u is finite
on the separation line.
Comparison of Equation 1 . 16 with the table of velocity p rofiles g iven by
Terrill (12) for a main stream UI(X) Uoosin (x/a) is favourable. Zero skin
friction occurs at x.= 1.822983 a and a neig hbouring velocity profile is tabu
lated at X= 1.821 a where r(x) cos (x/2a) =0.066 (av/Uoo)I/2 dpl/dx. From
t he velocity profiles t hemselves, which are p resented at equal intervals of
7)= (Uoo/av)1/2 y cos (x/2a) one finds that {u(x,y) -u.(y)} /(du./d7) varies from
0.064 to 0.059 as 7) increases. This error is felt to be well within t he bounds of
the approximation of Equation 1.16. From Equation 1.16 a relation between
the disp lacement thicknesses at the two points may also be found. I f we
define

OI(X)

then
OI(X) = OI(X.)

J (1 - UI(x) ) dy
o

_U_

r(x)
- dPl/dx
+ O(x*
--

In

x*)

1. 17 .

1.18.

which is in agreement with Terrill's results. Reference to Equation 1.18 will


be made again at the beginning of Part II where it will be used to demon
strate the alternative possibilities of flows regular and flows singular at
separation.

S8

BROWN & STEWARTSON

The main conclusion from the analy tic study of flow in the neighbour
hood of separation with an adverse pressure gradient is that the streamwise
component of velocity re mains finite as X--'fX.-, but the normal component
develops a singularity all along the separation line. A rather diffe rent picture
is presented by the solution near separation when the pressure gradient is
zero but instead there is uniform injection across the boundary (See sec tion
on "Flows with suction and injection"). Here a parallel analysis shows that as

X--'fX.-

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

u ""

F(y

g(x,

g(x)

In T(X)

1 .19.

Here r ex) is defined in Equation 1 .14 and F is similar to, but not identical
with, the mixing profile of Chapman (38 ) . Agree ment with the n umerical
solution [Catherall, Stewartson & Williams (37)] is quite good, and we see
that the displacement thickness now has a logarithmic singularity at sepa
ration. Thus, in spite of the weaker singularity in r (x) , the effect is much
more dramatic than in the adverse pressure gradient situation , for the
boundary layer i s blown off the wall by the injec ted fluid.
Flows with severe pressure gradient.-We conclude Part I with a few re
marks concerning a situation that occasionally arises in incompressible flow,
but is more usually associated with the interaction between a shock wave and
the boundary layer when the main stream is supersonic. The situation occ urs
if the prescribed pressure , after varying moderately slowly ove r a substantial
length of the boundary layer, suddenly rises steeply and possibly provokes
separation. For example , if x 1 represents the trailing edge of a thin plate ,
with a main stream of the form
.
1
1.20.
VI(.1:)
u + aG -b=

( X)
-

a, b, U constants and bal, the flow is almost that given by the B lasius
solution except that dUI/dx may be large when l-x--'b. Since the rapid
variations take place over a short distance the perturbations in almost the
whole boundary layer may be regarded as inviscid, the viscous forces be
ing negligible . Hence a slip velocity is induced at the wall which is reduced
to zero by an inner viscous layer. S uppose for example that a,....."b2/a Then the
inner layer has thickness Oav)I/2) and is li kely to separate . Howeve r, if
ab2/a, no separation occ urs in thi s region, while if ab2/a se paration is
provoked early on . Thus even a very small pressure rise causes separation if
it occurs sufficiently rapidly . The notion of an inner boundary layer in this
connection was first introduced by Stewartson (39) and Lighthill (40) in
their studies of shock-wave boundary-layer interactions, and an approximate
theory using these concepts was given by Stratford (41 ) . Stratford finds that
the distance to separation satisfies
1 .21 .

LAMINAR SEPARATION

S9

where K=0.0076, Cp=2(PI-PO)/P1UO\ and Po and Uo are the pressure and


velocity at some conve nient point of the external flow. It is noted that if
K=0.0104, as Curle & Skan (42) have shown, Equation 1.21 agrees well with
many accurate computations even when the pressure gradient is not severe .

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

PART

II.

FLOWS REGULAR AT SEPARATION

Introduction.-T he i nvestigations reviewed i n the previous section


strongly indicate that in general an incompressible boundary layer is singu
lar at separation when the press ure gradient is prescribed. It is, however, not
necessary in principle that a singularity must occur. Curle's formula, E qua
tion 1.2, shows that it is linked with the nonvanishing of the fourth deriva
tive of the separation profile at y = 0, and so, if this derivative does vanish,
the solution will be regular. Alternatively, we have already seen that near
x =x.

2 .1 .
s o that i f dT/dx is infinite a t separation so is the derivative o f the displace
ment thickness OI(X). Hence we can ensure regularity by requiring OI(X) to
be regular. I n fact, the first numerical integration through the point of sepa
ration was carried out, by Cathe rall & Mangler (43), using this result. Their
integrations were started at stagnation with a prescribed pressure gradient,
but at an appropriate point near separation they stopped s pecifying the
pressure gradient a priori and i nstead s moothly joined the displace ment
thickness to a parabolic or cubic form. From this point on the pressure gradi
ent was regarded as one of the u nknowns and determined step-by-step nu
merically. T hey found that the solution passed s moothly through separation
and a region of reversed flow was set up downstream: it was even possible to
achieve reattachme nt. However, they found that the numerical procedure
developed i nstabilities downstre am of separation that could easily get out of
hand.
These difficulties are not surprising, for the governing differential equa
tion is parabolic and therefore locally like the heat-conduction equation with
+x playing the part of time if u >0 and -x if u<O. Consequently, j ust as
the solution upstream of separation depends on initial conditions from
which the integration is started, in the reversed-flow region we expect the
solution to depend on some downstream conditions. T he situation may be
illustrated by Fig. 4. If u,v are given on AF and u is given on B D , then the
solution in ABCS is uniquely s pecified once u is also given on AB. In SEFS
however, propagation of disturbances in the sense of x decre asing is possible ,
and so on physical grounds we would expect that in addition to knowing u
on AB we s hould also have to know it on FE before the solution could be
completed in CDFS. Friedrichs (44) has made a rigorous study of a class of
differe ntial e quations that include these parabolic equations of mixed type

60

BROWN & STEWARTSON

Y/
A

u>O

0
u>O

E
F

IX

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

FIG. 4.

as a spe cial case , and his work strongly su pports the physical argument.
Again Gevrey (45) has m ade a detailed study o f model equations o f mi xed
type and has proved existe nce theorems (not qu ite com plete since a clear
conditio n that the data m ust satisfy was no t given) . Further, Stewartson
( 1 1) has examined the structu re of the solu tion near S and sho wn that a
nonuniqueness o ccurs in general, as an infinite set of fu nctions, each of which
has an essential si ngularity at S, may be added to the analytic solutio n
withou t violating the bou ndary conditio ns on SF, SC and C D .
Clearly, then, the problem of determi ni ng the solution downstream of a
regular separation is a fo rmidable one, but hopefully no t im possible , and we
are e ncouraged by progress that has recently been made in related studies.
Ban (46) has investigated the flow behind a weak sho ck advancing along a
flat plate, for which the governing equation is parabolic and of mixed type.
He develops a successful stable numerical procedure using a method nor
mally appro priate to elli pti c equations. A related problem is the im pulsive
motion of a flat plate. This has recently been solved numerically by Hall
(47) . The fundamental pro blem is in three variables bu t these can be reduced
to two and then the governing equatio n is parabolic and of mixed type .
Hall's method is to solve the o ri ginal equation, making use o f the similarity
properties of the solution to speed up the integration. It m ay well be that
these two studies will be adapted to the separatio n problem, and either it
would then be regarded as elli ptic, o r an extra coo rdi nate z would be added
and the integration carried out in terms of x, y, z until the solutio n becomes
inde pendent of z. It is noted that Babenko et al. (48) have had conside rable
success wi th the second method i n their study of supersonic flow past cones.
Hall and Ban were fortunate, however , in having thei r boundary conditions
fully specified. In general the condi tions on EF are not known. The prob
lems of most interest have o nly a finite region of reversed flow followed by
a region in which u >0; it remains to be seen whether the existence of re
attachment is a furthe r com plication o r a salvation.
Interaction between shock wat'es and laminar boundary layers.-A third
method for preserving a solution regular at separation , and the o ne of most
practical significance , is to relate the displacement thickness to the external
pressure gradient. I f the relationship is such that a singularity at separation
implies a singularity in the pressure gradient there, then we have a contra-

61

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

diction because dPl/dx must be finite at separation. Hence the solution must
be regular.
Such a situation occurs in the theory of the interactio n between shock
waves and boundary layers. Provided the incident shock is stro ng enough to
provoke separation, the positio n o f separ atio n S occu rs som e distance u p
stream of the point.of interaction of the main shock and the wall, and fu rther
the structure of the bou ndary l ayer near S is independent of the strength of
the incident shock. Separatio n, i n fact, is apparently sel f-induced in a free
interactio n zone. How does this corne about? Roughly speaking it stems from
the fact that the pressure gradient in the main stream , which is supersonic ,
depends on the displacement thickness throu gh the relationship
dPt
dx

p,U,2

d251

2 .2 .

(M,2 - 1)1/2 dx2

which follows from the linearized theory of su personic flow. Here M 1 is the
M ach number of the u ndisturbed u niform inviscid flow and I d20tldx21 1.
Normally 0 1 is determined from the Blasius solutio n and d20l/dx2<0 so that
the pressure gradient is favourable, which does not encourage separation.
If this can be reversed however , an instability sets in, leading to separatio n.
Two requirements must be met: first , the initial sign of d20l/dx2 must be
r eversed, and this may be achieved by the inevitable small irregularities in
the initial boundary layer; seco nd, the do wnstream pressure must show a
rise when com pared with its u pstream value, and this is also a feature o f
shock-wave interactio ns.
I n order to put this argument o n a sou nd basis, the key point we must
bear in mind is that the pressure rise to separatio n is sm all, so that the pres
sure gr adient is large in the free-interaction zone. So , from Equation 2.2,
the distance downstream over which it takes place must be small. Conse
quently the viscous forces i n the boundary layer are negligible in this region
except when Y/Vl/2 is small, i.e. , right at the wall even o n the scale of the
boundary layer. In the remainder of the boundary layer the changes occur
through inviscid processes; consequently a slip velocity occurs as Y/V'/20,
which is reduced to zero by the inner boundary layer. A start o n a theo ry
o n these lines was made by Lighthill (40). He treated the whole problem as a
linear perturbation of a Blasius flow, which he recognised as specifically
excluding the possibility of separatio n. Never theless he was able to o btain
the initial pressure rise in the form
2.3.

where
K-'

U,3/2
(O.78)3/4(MI2 - 1)3/8

(T'" )3/4 (!lvH4


TI
-

P'"-1/4

Here T is the tem perature, the suffix w denotes conditions at the wall ,
an arbitrary constant, and T is the Blasius skin frictio n.

2.4.
Xo

is

BROWN & STEWARTSON

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

62

A number of o ther authors have considered this problem and for an ac


count of the earlier work the reader is referred to Stewartson (28). We shall,
with o ne exception, not go over this ground again but will discuss the more
recent work in so me detail, but we state, very firmly, that those methods
not based on the ideas laid down by Lighthill are theoretically unsound and
canno t be regarded as contributing to the understanding o f the pheno meno n
on the basis of the N avier-Stokes equations. To be sure, the agree ment be
tween theory and experime nt is often quite good, but this in itself is no t a
sufficient basis for acce pting the theory as correct.
The most important of the earlier wor k is due to Gadd (49). He properly
divides the boundar y layer into two layers, treating the outer as if it were
inviscid and co ntrolled by a pressure gradient that is independent of y. The
inner is treated by Stratford's (41) method, and the matching o f the two
zo nes enables the pressure to be fo und and the solution to be co mple ted. I n
detail the approach i s a d hoc and intuitive, and needs to be p u t o n a more
secure mathematical basis. One of the present authors (K. S.) and P . G .
Williams (67) have done this, b u t space limitations prevent its inclusion here
and the results will be published elsewhere. Gadd's principal results are that
at separatio n the pressure is given by
1. ByM,2
P.
2 .5 .
p; 1 + 2(M,2 1)'/4R'/4
=

and the scale of the interaction in the x direction is


(Ml2

1)3/8R3I8

2.6.

where R UIX./P is the Reynolds number. A compariso n between his theory


and experi ment in o ne case is given in Figure 5. These formulae are in quali
tative agreement with the rigorous theory and Equation 2 .5 agrees well with
experiment. Lewis, Kubota & Lees (50) have carried out a correlation of
experimental data with Curle's (51) for m for the pressure:
=

2 .7 .
and report good agreement. However Curle's theory i s based o n approximate
methods more appropriate to a slowly varying pressure and cannot be re
garded as self-consistent. It wo uld be interesting to examine how the agree
ment is modified by using Equation 2.6 in 2.7, but unfortunately the data
are not available in sufficient detail. Gadd's theory predicts that the point o f
separation is passed without a singularity and this i s confirmed by the
rigorous theory, essentially because of Equatio n 2.2. We may, therefore , in
principle extend the boundary-layer solution into a region o f reversed flow,
but no wholly satisfactory integration procedure is then available. Two
approximate methods o f dealing with this regio n have been extensively used
in recent years each with a certain measure of success. They will be discussed

63

LAMINAR SEPARATION
0-08

0-06

)(

Cp

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

0 -04

O-(]'}
"

FIG.

?x )(x

,,

___"

0-2

0- 4

I
06

oe

JC

)(

)(

JC

lC

,,

./

X/X

lC

JC

OJ::

1-0

1-2

1 -,4-

1'6

5. Pressure coefficient distribution-comparison between theoretical curve and


experimental points [from (49)].

briefly below beginning with the o ne developed by Abbo tt, Holt & Nielsen
(52) and by Nielsen, Lynes & Goodwin (53, 54).
The essential idea in Nielsen's procedure is due originally to Dorodnitsyn
(55), and co nsists of co nverting the differential e quation of the bou ndary
layer into a number of integral relatio ns by multiplying the e quation by
un, n = O, 1, 2, . . . , where u=u/ U" and integrating with respect to y across
the bou ndary layer. These integrals in turn can be reduced to integrals with
respect to u in which au/ay is the only unknown fu nctio n, and the technique
is to assume a co nvenient relatio nship between au/ay and u, involving fu nc
tions of x to be fou nd, in order to reduce the integral relations to a set o f
ordinary equations in x. I n that part of the flow field for which u >0, the
authors assume that
ou (1 - 1) (U + a3) 1 12
2.9.
oy
ao + alU + a2u2
and use four integral equations (n 0, 1, 2, 3) to set up the o rdinary differen
tial equations for ao(x), aleX) , a2(x) , a3(x). The special form chosen in Equa
tion 2.9 e nsures that the principal properties of au/ay are satisfied, and yet
the necessary integrations al'e as simple as possible. I n that part of the flow
field for which u <0, u is assumed to be a polynomial in y, usually a cubic,
continuity of a2u/ay2 across the line tt = be in g required if possible. Com pari
so n with the classical bou ndary-layer compu tatio ns, of Leigh (14) fo r e x
am ple, shows good agreement u p to separation and, fo r an insulated wall, a
free-interactio n zone is possible provided the calculatio n is started by giving
the pressure a small increment !J.p. As !J.p---"O it is fou nd that the numerical
solutions seem to co nverge to a limit solutio n that is in good agreement with
experiment as far as separatio n. Beyond separation the induced pressure
reaches a maximum and, if the computation is co ntinued far enou gh, u
=

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

64

BROWN & STEWARTSON

changes sign again near the wall. neither of which is reasonable o n physical
grou nds. Nevertheless when the au thors apply the m ethod to a number o f
problems where the principal shock i s indu ced b y a wedge, encouraging
agreement with experiment is fou nd for the surface pressure. A typical ex
ample is given in Figure 6. In a later paper (54) the method is applied to
non-insu lated walls and it is co ncluded that for sufficiently cooled walls
[T",/Tadlab <0.1331 no free interaction can o ccu r. This result is in direct con
tradiction with the theories o f Lighthill and Gadd. In ou r view, the contradic
tio n is a consequ ence of the coarseness of the method of integral relations
which prima facie does not seem to be well adapted to cope with the special
features of free interactions, namely, that they are almost entirely inviscid.
Although we welcome the good agreement with experim ent in the quoted
cases for insulated walls, the unrealistic features of the results m entioned
above, together with the almost total neglect of the possibility of upstream
propagation of disturbances, lead us to the conclusion that is it a rather un
sound basis for an u nderstanding of the phenomeno n u nder co nsideratio n.
The seco nd approximate method is that of Lees & Reeves (56, 5 7) and
their associates. It is similar to that developed by Nielsen in that it m akes
use of integral relations i nstead of dealing directly with the differential
equation. In its sim plest form (56) two such relatio ns are used, corresponding
to n = 0, 1 in Nielsen's method, and the wall is supposed to be thermally in
sulated so that the tem perature is known in terms of the velocity. However,
instead of assuming a velocity pro file such as that defined by Equation 2.9,
2.2

INVIS O WEDGE
PRESSURE

.0

0 '"

,.,.

REATTACHMENT
POINT

'f
//
0

...
, I
..

..

..0

r '"

.2

. 0 0 0 O 'To o

SE.PARATION POINT

FIG. 6.

0
...f'l..

';p '8

.,.,..,.

.0&

.12

"

' 1 1,00
.16

.20

.2'

.2&

.32

.36

Experimental and theoretical pressure distribution for a shock-wave bound


ary-layer i nteract ion

induced by a wedge

[from (53)).

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

65

they make use of the self-similar solutions of the boundary-layer equations


first i nvestigated by Falkner & Skan ( 17) . These solutions arise when the
main-stream velocity IX xm and the equations can be reduced to one ordinary
differential equation. It is found that these equations have acceptable solu
tions only if m > -0.0904, in which case there are two for any value of m <0,
in one of which there is a region where u <0 [Hartee (58) , Stewartson ( 18)].
If m >0, the solution is unique. In trying to approximate to a real boundary
layer flow by a sequence of such self-similar solutions there are essentially
two parameters at our disposal, m , and 0/ which is closely related to a dis
placement thickness and fixes the scale in the y direction. Thus we write
u(x,y) = U1(x) F(m,y/o/)

2.10.

Originally, when similarity solutions were used, m was linked with the ex
ternal pressure gradient on p hysical grounds but in the Lees-Reeves method
this is abandoned and m, like 0/, is regarded as a function of x and the or
dinary differential equations satisfied by m and 0/ are derived from the inte
gral relations mentioned earlier. The authors actually replace m by a para
meter a defined as follows:
a(x) =

o. au
-

Ul ay

_o

if

au

ay

_o

2.11.

<0

2.12.

(0, being the boundary-layer thickness2) and


a(x)

Yo
0,

= -

if

-oy)

au

.-0

(Yo being defined by u(y o) = 0).

I n Equation 2. 1 1 , 0 a (x) 3.9 , with a = 1.58 for Blasius flow. In Equa


tion 2.12, 0 a (x) 1, and a (x) defines the curve separating the reversed flow
region u <0 from the forward-flow region. With these definitions, all the
integrals appearing in the two integral relations can be expressed as products
of Oi with known functions of a. Thus formally they set up two differential
equations expressing 0/, a as functions of x and the external pressure gradient.
(They actually use the shape parameter as dependent variable instead of a,
but this is a detail.) Finally the pressure gradient is related to the boundary
layer structure by computing the slope e of the streamlines at the outer edge
of the boundary layer, and relating this to the pressure on the assumption
that the Prandtl-Meyer relation between Ml and e is appropriate in the
external inviscid flow field. The constant in this relationship is determined
from the initial inviscid flow field upstream.
The method of integration of these very complicated equations of Lees &
Reeves is to begin at separation with trial values of 0/ and then to integrate
upstream. By requiring that as a ..... 1 .58, dMl/do/ and da/do/ O so that
beyond the interaction the boundary layer is a self-preserving Blasius flow,
the correct values of 0/ and e at separation and also the correct integral

.....

2 This is not defined precisely but would appear to be the point in the boundary
layer at which u = 0.99 U1

BROWN & STEWARTSON

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

66

curve, which carries the solution through separation, are found. Then a
switch is made to the similar solutions with reversed flow, a is defined by
(2.12), and the integration proceeds straightforwardly. Without any further
modifications to the boundary conditions the pressure continues to increase
monotonically and eventually reaches a plateau. This is physically more
satisfactory than the corresponding result by Nielsen's method. One possible
explanation is the form of the similarity solutions with reversed flow which
consist, particularly as mO-, of a large region of very slowly moving re
versed flow, separated from the inviscid outer flow by a thin shear layer,
whose structure approaches Chapman's (38) mixing layer, and such solutions
are in line with observation. Another is that the principal integral properties
of the similar solutions are monotonic functions of a . In their paper Lees &
Reeves went on to achieve a spectacular success with the shock-wave boun
dary-layer i nteraction problem. In this problem, at the shock i mpingement
one must change the formula relating the pressure to the boundary-layer
displacement thickness by determining the constant in the Prandtl-Meyer
l aw from the i nviscid solution far downstream, while a, M 1 and 0: are con
tinuous. At reattachment a change is made to the velocity profiles without
reversed flow and the integral curve is chosen for which dMI/do;" and
da/do"O as a_1.58 (Blasius value) . Clearly, a very large amount of itera
tion is required, for in addition the position of separation relative to shock
i mpingement must be guessed and the whole solution repeated until condi
tions far downstream are in line with the requirements of inviscid theory.
One of the comparisons made with experiment is shown in Figure 7 ; others
may be found in Lewis, Kubota & Lees (50) . The agreement on the whole is
2.2
2.0

Po

1.8
1.6
1.4
1.2
1.0

FIG.

-0

0.5

1.0

1.5

I
2.0

2.5

7. Experimental and theoretical pressure distribution for a shock-wave bound


ary-layer interaction [from (56)J.

LAM I NA R SEPARATION

67

so good that criticism of the method must be muted. Nevertheless, we make


a few points. First, the difficulty about getting through separation is not

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

intrinsic to the problem, as Gadd's work has shown. Second, the dichotomy

between the use of the similarity solutions and their physical significance
tends to put the method on an empirical basis, where the test is "does it
work ? " , and in consequence the value of the method as a means of under
standing the phenomena is diminished. Third, al thoug h the use of the similar
ity solutions with reversed flow is probably the best we can do at present to
cope with the main dead-water region downstream of separation, they are
unsatisfactory near separation because the rapidity with which the pressure
rises here means that viscous forces can be neglected in the major part of the
boundary layer. Finally one might have doubts because of the complexity
of the computations involved ; however, several workers have repeated them
in a variety of problems and confirmed the good agreement with experiment.
Hankey & Cross (59) have endeavoured to simplify the method, reducing it
to some very simple equations. The agreement with experiment is main
tained at low supersonic speeds but the simplified version is definitely inferior
at higher values of the incident Mach nu mber [Fitzhugh (60)J.
When generalisations of the method are attempted, however, it becomes
much less attractive. If the wall is maintained at a constant temperature
instead of being thermally insulated, self-similar solutions of the momentum
and energy equations may also be obtained, the appropriate ordinary dif
ferential equations h av ing been obtained by Stewartson (20) and integrated
in a variety of circumstances by Cohen & Reshotko (23) . It would therefore
appear at first sight that it would be a simple matter to extend the theory to
cover such a case, and indeed Lees & Reeves obtained the variations of the
key integral properties of the similar solutions as functions o f a for a highly
cooled wall, but no explicit comparison has been made with experiment.
Holden (61) has g enerali sed the whole method by using an energy integral
and introducing a third parameter in addition to a and 0/. There are now
three ordinary differential equations to solve and the iteration procedure to
find the boundary-layer development upstream of separation is greatly
complicated. He claims good agreement with experiment in a few cases at a
large value of M1( "" 10) but in view of the tremendous amount of computa
tion i nvolved, it seems sensible to wait for confirmation by independent
workers before accepting his conclusions. In addition of course, we know that
in the free-interaction zone heat transfer has only a scaling effect (Equation
2.6) so that the difficulties experienced using the Lees- Reeves method must
be intrinsic to it. Again, the similarity solutions with heat transfer do not
appear to be quite so efficient at representing boundary-layer flows as do
those for i nsulated walls-witness the diffi c u l ti es with separa ti o n described
in Part I .
The method has also been applied to a discussion o f the boundary layer
on a smooth blunt body, and the near-wake flow at supersonic speeds, by
Grange, Klineberg & Lees (62) . In our view the paper is unsatisfactory,

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

68

BROWN & STEWARTSON

revealing two defects in the method. The first is that the similarity solutions
i mply that a must be less than 3.9, whereas in the actual boundary-layer
cases i nvestigated a apparently must exceed this value. The authors deal
with this difficulty by setting a equal to 3.9 downstream of the point where it
first reaches this value, until separation is reached where a changes discon
tinuously. This discontinuity is forced on the method by the structure of the
ordinary differential equations it uses, and appears to have nothing whatso
ever to do with the real flow. They associate the discontinuity with a sharp
pressure rise, of 24 per cent in one case quoted, and associated fall in boun
dary-layer thickness. Actually one can virtually ignore the interaction be
tween the boundary layer and the main stream until close to separation, so
that the boundary-layer development can be computed by standard methods.
Without interaction the boundary layer becomes singular at separation, and
the leading question is whether the singularity can be removed by taking
into account the interaction, which becomes i mportant for the first time
near separation. The reader is referred to the original papers for a discussion
of the super-critical and sub-critical regions in the approximate method ; we
merely add the comment that the comparison with experiment is not con
vincing enough to overcome our serious reservations.
Wake studies.-The theory of the development of a supersonic laminar
wake behind a bluff body or a backward-facing step has attracted increasing
attention in recent years. Perhaps the most interesting of the investigations
that have been carried out are due to Lees and his associates (57, 62) , and to
Baum & Denison (63) , and we shall discuss aspects of their work here. First
however we consider the most significant of the earlier work.
This has largely been concerned with wakes from backward-facing steps,
or their eq uivalent, and the p roblems p resented are ext rem ely difficult be
cause the shear layer coming off the step separates the inviscid flow outside
from a slowly moving mass of fluid in the dead-air region. Later on, this
shear layer either reattaches to the wall downstream of the step or, in the
case of base flows, amalgamates with another shear layer. Lighthill (40)
in the paper already quoted has considered the problem of boundary-layer
development around a convex corner. He was concerned with small flow
deflections, whereas in practice the flow deflection may not be small, but the
general structure of the flow is probably the same i n the two cases, namely,
the principal expansion is inviscid with an inner boundary layer to take care
of the resultant slip velocity. In an earlier paper Lighthill (64) suggested,
but later (40) withdrew, the suggestion that some over-expansion takes
place. It is i nteresting to note that recent experimental work [Hama (65)]
on flow around a right-angled bend reveals an upstream effect and, past the
corner, an over-expansion, followed by a substantial lip-shock. A reappraisal
and possible extension of Lighthill's theory would be helpful in understand
ing this phenomenon.
A simple theory of base flow has been given by Chapman, Kuehn & Larson
(66) for flows in which the initial main flux through the boundary layer is

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LAMINAR SEPARATION

69

FIG. 8. Flow past a concave wall with leading-edge separation.


negligible. A possible configuration is shown in Figure 8 in which the corner
at A is sufficiently sharp to induce separation at the leading edge O. In Chap
man's model the fluid is practically at rest near the wall between 0 and B :
between this dead-air region and the main stream there is a thin shear layer
in which the fluid velocity varies from zero to its main-stream value and
whose properties are now classic. This shear layer entrains fluid from the
main stream which provides part of the boundary layer downstream of B ;
i t also entrains fluid from the dead-air region which must be returned near B .
The streamline separating the two entrainments i s found t o be the one on
which the velocity u* =0.587 UIr where U1 is the mainstream velocity. This
streamline must be the stagnation streamline at B. Chapman now assumes
that in the neighbourhood of B the total pressure remains constant on the
dividing streamline, so that
PDA

P2

)-/ (r-l)
1
( 1 + )' - -M*2
2

where PDA is the pressure in the dead-air region, P2 is the pressure on the wall
downstream of B and M* is the Mach number of the streamline. This theory
raises a nu mber of questions, (e.g., is the fluid in the dead-air region almost
at rest and is the flow near reattachment so simple?), but the agreement with
experiment is so good that one is inclined to believe that it contains the
principal features controlling the dead-air zone, and that it may well be
worth exploring the assumptions further from a more rigorous standpoint.
This, however, has not been done and the main contribution to this topic
has been to apply the Lees-Reeves method, taking into account an initial
main flux into the shear layer at 0 which would arise if OAB were part of a
backward-facing step. Also the method has been applied to the base flow of
a body by an obvious extension. The principles of the method are as for the
interaction problem described earlier, the main difference being that in the
base-flow problem the similarity profiles used downstream of the rear stag
nation point are those satisfying a zero stress condition on the boundary
[Stewartson (18)]. For - 1/5 <m < -0.0904 the corresponding velocity
profiles are positive, but, for -0 .0904 <m <0 , there is a region of reversed

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

70

BROWN & STEWARTSON

flow which rapidly increases in length as m-'>O-. The problems of integrating


the ordinary differential equations are again formidable, involving iterative
procedures, and the reader is referred to the original papers [Reeves & Lees
(57), Grange, Klineberg & Lees (62)] for details. In comparisons with experi
ments some agreement on trends was found but, overall, we are of the opinion
that the method cannot be justified as an improvement on Chapman's by
these comparisons, and the previously stated reservations still apply.
Stimulated by one characteristic of the ordinary differential equations
obtained by Lees & Reeves, namely, that just downstream of the wake stag
nation point they develop a saddle-point singularity, which means the avail
able parameters must satisfy a condition to pass through this point, Baum &
Denison (63) have attempted a fi nite-difference numerical integration of the
boundary-layer equations from the wake stagnation point. I n fact they go
further, permi tti n g transverse pressure variations (controlled by inertia
terms only) and relating the external pressure to the velocity outflow as in,
but more generally than, Equation 2.2. Their initial profiles contain a param
eter M( l , 0) which is the Mach nu mber at the outer edge of the boundary
layer, and they find that the solutions obtained have the main characteristic
of a saddle point, i.e., except for one value of M(l, 0) they become physically
unacceptable. This lends some support to the claim that the curious proper
ties of the equations in the Lees- Reeves method have some basis in physics.
We make a final point about the recent attempts to obtain theories of
wake flows and interaction problems. These methods lay tremendous em
phasis on massive computation and ad hoc arguments to lead to their results,
little attempt being made to j ustify the assumptions on mathematical
grounds. While we recognise the need for such a computational effort in the
end, it seems to us of great importance that a rational approach be adopted
to make sure, for example, that terms neglected in the approxi mation really

are much smaller than those retained. Until this is done, and even now it
is possible in part, it will be difficult to convince the detached and possibly
skeptical reader of their value as an aid to understanding.

LAM I NAR SEPARATION

71

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

LITERA TURE C I TED

1. Prandtl, L., Proc., Third Intern. Math.


Con gr. Heidelberg, 484-9 1 (1904)
2 . Goldstein, S., Quart. J. Meeh. Appl.
Math. I, 43-69 (1948)
3. Dean, W. R., Proc. Cambridge Phil. Soc.
46, 293-306 (1950)
4. Blasius, H., Z. Math. Phys. 56, 1-37
(1908)
5 . Goldstein, S., Rosenhead, L., Proc.
Cambridge Phil. Soc. 32, 392-401
(1936)
6. Proudman, r., Johnson, K., J. Fluid
Mech. 12, 1 6 1 -68 (1962)
7. Stewartson, K., Fluid Dynam. Trans. ,
3, 1 2 7-46, Warsaw (1967)
8. Nickel, K., Arch. Ration. Mech. A nal.
Z, 1-31 (1958)
9. Oleinik, O. A., Zh. Vychisl. Matem.
Fiz, 3, 489-507 (963)
10. Curle, N., The Laminar Boundary
Layer Equatior:s, p. 34. (Oxford,
1962)
1 1 . Stewartson, K., Quart. J. Meeh, A ppl.
Math. 11, 399-4 10 (1958)
12. Terrill, R. M ., Phil. Trans. A, 253, 55100 (1 960)
13. Hartree, D. R., A eron. Res. Council
Rept. Mem., 2426 ( 1939)
1 4. Leigh, D . C . F., Proc. Cambridge Phil.
Soc. 5 1 , 320-32 (1955)
15. HO'warth, L., Proc. Roy. Soc. A, London ,

164, 547-79 (1938)


1 6. Gortler, H., J. Math. Mech. 6, 1-66
(1957)
17. Falkner, V. M., Skan, S. W., A eron.
Res. Council Rept. Mem. , 1 3 1 4
( 1930)
1 8. Stewartson, K., Proc. Cambridge Phil.
Soc. 50, 454-65 (1954)
19. Illingworth, C. R., Proc. Roy. Soc. A ,
London, 199, 533-58 (1949)
20. Stewartson, K . , Proc. Roy. Soc. A , 200,
London, 84-100 (1949)
2 1 . Stewartson, K, J. Fluid Mech. 12, 1 1 728 (1962)
22. Poots, G., Quart. J. Meeh. Appl Math.
13, 57-84 (1960)
23. Cohen, C. B., Reshotko, E . , Nat!.
A dvis. Comm. A eron. Rep. 1 293
(1956)
24. Williams, P. G., Private communica
tion (1965)
25. Sells, C . C . L., Private communica
tion (1968)
26. Merkin, ]. H., Submitted to J. Fluid
Mech. (1969)
27. Der, J. Jr., Raetz, G . S., Inst. A erol
Space Sci. Preprint no. 62-70 (1962)
28. Stewartson, K., Theory of Laminar

29.
30.

31.
32.

Boundary Layers in Compressible


Fluids (Oxford, 19M)
Maskell, E . C., A cron. Res. Council
Rept. Mem. 1 8063 (1955)
Lighthill, M. J., Laminar Boundary
Layers 1-1 1 3 (Rosenhead, L., Ed.,
Oxford, 1963)
Cooke, J. c., A eron. Res. Council Rept.
Mem., 3530 (1967)
Brown, S. N., Phil. Trans. Roy. Soc.
London Ser. A , 257, 409-44 (1965)

33. Banks, W. H . H., Williams, P. G., To


be published (1968)
34. Jones, C. W., Watson, E . ]., Laminar
Boundary Layers 198-257 (Rosen
head, L., Ed., Oxford, 1963)
35. Watson, E . ]., A eron. Res. Council
Rept. Mem., 2 6 1 9 ( 1952)
36. l'vlorduchow, M., Reyle, S. P., Intern. J .
Heat _Hass Transfer 10, 1 2 33-54
(1967)
37. Catherall, D., Stewartson, K., Wil
liams, P. G., Proc. Roy. Soc. A ,
London, 284, 3 70-96 (1965)
38. Chapman, D . R., Nat!. A dvis. Comm.
A eron. Rep. 958 (1 950)
39. Stewartson, K., Proc. Cambridge Phil.
Soc. 47, 545-53 (1951)
40. Lighthill, M . J., Proc. Roy. Soc. A ,
2 17, London, 478-507 (1953)
41. Stratford,

B. S . , A eron. Res. Council

Rept. Mem., 3002 ( 1954)


42. Curle, N., Skan, S. W., A eron. Quart. 8,
25 7-68 (1957)
43. Catherall, D., Mangler, K W., J.
Fluid. Mech. 26, 1 63-82 (1 966)
44. Friedrichs, K., Comm. Pure A ppl.
Math. 1 1 , 333-4 1 8 (1958)
45. Gevrey, M., J. Math. Ser. 6, 10, 105-48
( 1 9 1 4)
46. Ban, S. D . , Interaction Region in the
Boundary Layer of a Shock Tube,
Doctoral

Thesis,

Case

Inst.

of

Techno!. (Cleveland Ohio, 1967)


47. Hall, M . G., To be published (1 968)

48. Babenko, K. I . , Voskresenskiy, G. P.,


Lyubimov, A. N., Rusanov, V. V.

Nat!. A eron. Space A dmin. TT F380 (1966)


49. Gadd, G. E., J. A eron. Sci. 24, 759-71
(1957)
50. Lewis, J . E . , Kubota, T., Lees, L.,
A IA A J. 6, 7-14 (1 968)
5 1 . Curle, N . , A eron. Quart. 12, 309-36
(1961)
52. Abbott, D . E., Holt, M . , Nielsen, J. N.,
ASD TDR-62 -963 (1962)
53. Nielsen, J . N., Lynes, L. L., Goodwin,
F. K, Technical Report A FFDL-

72

BROWN & STEWARTSON


TR-65-107 Wright-Patterson
Force Base, Ohio (1965)

Air

54. Nielsen, J. N., Lynes, L. L., Goodwin,

Annu. Rev. Fluid Mech. 1969.1:45-72. Downloaded from www.annualreviews.org


Access provided by CAPES on 09/15/16. For personal use only.

F. K., AGARD Conf. Proc. No. 4


(Separated Flows), 31-68 (1966)

55. Dorodnitsyn, A . A., A dvan. A eron. Sci.


3, 207-19. von Karman, Th., Ed.,
(Pergamon, 1962)
56. Lees, L., Reeves, B . L., A IAA J. 2,
1907-20 (1964)
57. Reeves, B. L. Lees, L., A IAA J. 3,
2061-74 ,1965)
58. Hartree, D . R., Proc. Cambridge Phil.
Soc. 33, 223-39 (1937)
59. Hankey, W. L., Cross, E. J. Jr., A IAA
J. 5, 651-54 (1967)

60. Fitzhugh, H., Private Communication


(1968)
61. Holden, M. S., A GARD Conf. Proc. No.
4 (Separated Flows), 147-80 (1966)
62. Grange, J-M., Klineberg, J. M., Lees,
L., A IA A J. 5, 1089-96 (1967)
63. Baum, E., Denison, M . R., A IA A J.
5, 1224-30 (1967)
64. Lighthill, M. J., Quart. J. Mech. Appl.
Math. 3, 303-25 (1950)
65. Hama, F. R., A JA A J. 6, 2 12-19 (1968)
66. Chapman, D . R., Kuehn, D. M . ,
Larson, H . K . , Nail. Advis. Comm.
A eron. Rep. 1356 :1958)
67. Stewartson, K., Williams, P. G., Sub
mitted to Proc. Roy. Soc., Ser. A . ,
(London) (1969)

Potrebbero piacerti anche