DAlemberts Paradox

Downloaded 03/13/13 to 129.8.242.67. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.
php
SIAM REVIEW
Vol. 23, No. 3, July 1981
(C) 1981 Society for Industrial and Applied Mathematics
0036-1445/81/2303-0003501.00/0
DALEMBERTS
PARADOX*
KEITH STEWARTSON+
Abstract. Since classical inviscid theory leads to the patently absurd conclusion that the resistance
experienced by a rigid body moving through a fluid with uniform velocity is zero, great efforts have been made
during the last hundred or so years to propose alternate theories and to explain how a vanishingly small
frictional force in the fluid can nevertheless have a significant effect on the flow properties. The methods used
are a combination of experimental observation, computation often on a very large scale, and analysis of the
structure of the asymptotic form of the solution as the friction tends to zero. This three-pronged attack has
achieved considerable success, especially during the last ten years, so that now the paradox may be regarded
as largely resolved. The lecture will review these achievements in subsonic and supersonic flow, for bluff
bodies, for trailing-edge flows and for internal flows. Most of the work has been on steady two-dimensional
problems but the special difficulties in unsteady and three-dimensional flow will also be touched on.
1. Introduction. Conjectures are a vital part of the fabric of pure mathematics.

Not only have they engaged the lasting attention of many mathematicians in all periods
of history, but their resolution has often led to a deeper appreciation of the discipline
and to a broadening of its power, to say nothing of suggesting new problems. Some
conjectures, still open, are both deceptively simple and quite old: one of my own
favorites is Goldbachs conjecture, first stated in a letter to Euler (1742), that every
even number is the sum of two primes.
Theoretical fluid mechanics also has its conjectures which have long tantalized and
disconcerted scientists. Some of the most persistent have a history of over two hundred
years, as they arise from the famous paradox discovered by dAlembert in 1752. This
paradox, like Goldbachs conjecture, is extremely easy to state, namely that there is no
drag on a finite body at rest in an infinite, incompressible, inviscid fluid otherwise in
uniform motion. That such a result is to be expected can be seen from the following
argument. If p is the pressure in the fluid, p the density and q the local velocity, then, in
virtue of Bernoullis theorem,
(1.1)
P + _p q2 const,
since conditions are uniform at infinite distances from the body. Hence, the drag, i.e.,
the component of the force on the body in the direction of the main stream, is the same
if is replaced by -i. Thus, if the drag is positive for one flow it will become negative if
the direction of fluid motion at infinity is reversed, contradicting the second law of
thermodynamics. It is easy to imagine the consternation which such a conclusion
immediately caused because it is so entirely at variance with ones everyday experience.
Indeed the reduction of drag on moving bodies is a central problem for practical fluid
dynamicists such as ship designers and aeronautical engineers. The paradox haunted
theoretical workers throughout the nineteenth century and beyond, and only gradually
were they able to come to terms with it and learn that, nevertheless, theory has much to
contribute to the understanding of the way bodies move through fluids.
The mathematical conjectures associated with dAlemberts paradox can be stated
most clearly in terms of the Navier-Stokes equations which, for an incompressible
Received by the editors June 12, 1980. This paper was presented as the 1980 John von Neumann Lecture
at the SIAM 1980 Spring Meeting, Alexandria, Virginia, June 6, 1980.
?Department of Mathematics, University College London, Gower Street, London WC1E 6BT, England.
308
309
DALEMBERTS PARADOX
Downloaded 03/13/13 to 129.8.242.67. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
viscous fluid in steady motion, may be written as
(1.2)
where
-q x curl q
-grad
+q + pV2q,
div q
O,
, is the kinematic viscosity. It is generally believed by fluid dynamicists, virtually
as an act of faith, that these equations, or their equivalents for compressible fluids
(Howarth 1953) accurately describe the motion of a wide class of fluids, notably air and
water, except possibly in extreme conditions such as might occur in the interior of
shocks. In addition we must specify constitutive relations for the diffusive coefficients,
of which ,p is one, and an equation of state connecting the pressure with the density and
the temperature (for a uniform incompressible fluid it would be O const) together with
appropriate boundary conditions, for example the no-slip condition on surfaces of
contact between the fluid and a rigid body.
Consider a fixed smooth body $ of characteristic length L in an incompressible
viscous fluid which, at infinite distances from S, is in uniform motion with velocity Uoo in
the direction defined by the unit vector i. Then we may define the Reynolds number of
the flow as R UL/z, and on replacing q by q/U, distance r by r/L, equations (1.2)
are unaltered except that z, is replaced by R
In practical problems the value of R is
-3
for spermatozoa to 109 or more for
of
10
from
of
a
tremendous
values,
range
capable
vessels.
ocean-going
large
The first mathematical question which arises is whether (1.2) have solutions for all
values of R. I am not competent to give an account of the present state of the attack on
this problem and refer the interested reader to the literature (e.g., T6mam (1977)).
I take it for granted that such a solution exists and am concerned rather with the form it
takes as R-. For dAlemberts paradox follows when one makes the intuitively
reasonable assumptions that, in this limit, the viscous term u V2q may be neglected, the
no-slip condition may be replaced by a contact requirement and the resulting solution is
smooth. However, the governing equations then become the Eulerian equations, (1.1)
holds and it can be shown rigorously that the drag is zero just as dAlembert argued.
Provided the act of faith is allowed, the theoretical fluid dynamicist has an advantage
over the pure mathematicianhis theories may be observable. In fact there is a
tremendous amount of information available on the flow of air or water, but it has to be
used with care since at high Reynolds number the steady solution, if it exists, is usually
unstable and we observe the consequences, namely a partially turbulent flow. The study
of turbulent flow is of central importance to fluid mechanics, but most students will
concede that much of it has, perforce, to be empirical, and the role of the mathematician
is not easy to fit into this endeavor. There is an extensive theory of turbulent flow but
hardly any of it is relevant to the problem of estimating forces on moving bodies. This is
very disappointing to me, and it is no consolation that fundamental experimental
studies of turbulent flows are in a similar situation. It seems in fact that the practical
engineer relies on model equations to describe the turbulent phenomena and some of
his most telling concepts are derived from laminar theory.
In this review I shall concentrate largely on laminar flow and accept the fact
opportunities for comparison with experiment are limited. The aim of the theoretical
work is to obtain the correct structure of the solution of (1.2) as R-o, and so
understand dAlemberts paradox. In addition, however, we hope that the results will
be directly of use in the description of reasonable flows at the finite values of R at which
laminar flow can be achieved, and conceptually useful at higher values of R when parts
of the flow field are turbulent. I will show during the course of this lecture that some
-.
310
KEITH STEWARTSON
success can be achieved in all of these aims. Returning to the theories derived from
observation, these brief remarks indicate that from the point of view of inviscid flow
(R oo) they are not wholly relevant and must be regarded as providing clues from
which conjectures can be formulated and tested in some mathematical sense.
Thus experiment shows that the notion of a smooth inviscid flow as the limit of the
solution of (1.2) is incorrect, but does not explain why. We find that the fluid passing
close to a bluff body at the forward stagnation point does not remain near to the body as
far as the rear stagnation point, but appears to break away from the surface leaving
behind an eddying wake of almost constant pressure. This wake can extend to large
distances behind the body and is turbulent, but we may infer that in the appropriate
laminar solution the fluid also detaches from the body setting up a discontinuity surface.
On one side of this surface the fluid has originated from an infinite distance upstream of
the body but on the other it has had a different origin. The simplest structure one could
ascribe to it is that it be at rest forming, as it were, a dead-water region.
We may therefore think of there being two candidates for the limit solution of (1.2),
one (Fig. la) being smooth and leading to dAlemberts paradox. The other (Fig. lb) is
discontinuous and contains a degree of asymmetry enabling us to avoid the paradoxical
(a)
(b)
FIG. 1. Two of the candidates for the steady solution of the Navier-Stokes equations for flow past a circular
cylinder at R >> 1. (a) attached potential flow. (b) Kirchhofffree-streamline flow.
conclusion. The study of discontinuous flows was initiated by Helmholtz (1868), who
considered jets, and by Kirchhoff (1869) for finite bodies; the type of flow exemplified
by Fig. l b is thus usually referred to as Kirchhoff flow and during the course of this
lecture I hope to explain why I think it is correct. Once discontinuities are admitted
there are obviously many choices to be made. How do we choose P, the detachment
point? Is the fluid at rest in the wake? How many detachment points are there? Is the
wake finite? These questions cannot all be answered on an inviscid basis only, but for a
circular cylinder Brodetsky (1923) demonstrated that the detachment streamline is
smooth at P if the angular distance a from the forward stagnation point is 55 and
Schmieden (1929) showed that if a < 55 this streamline subsequently enters the body
while if c > 135 the streamlines cross in the wake. For further details of these and
related studies reference may be made to Birkhoff and Zarantonello (195;7).
311
DALEMBERTS PARADOX
The next important step was taken by Prandtl (1905) who introduced the concept
of the boundary layer. Mathematically the inviscid solution cannot be uniformly valid
when R oe because it fails to satisfy the no-slip condition on the body and this
nonuniformity can only be removed by introducing a layer near the body where
viscosity matters. Hopefully its thickness is zero in the limit R co. In its simplest form
this theory assumes that the inviscid flow has been correctly calculated, and gives rise to
a slip velocity Ue(X) on the body which we may assume is defined for our purposes by
y 0. This slip velocity is reduced to zero on the body in a layer of thickness O(R-a/2L).
A prima facie consistent set of equations for this layer is obtained by scaling the velocity
component normal to the body by R -a/2, so that, if (xy*) denote physical distances
along and normal to the body and (u*, v*) the corresponding velocity components, we
write
(1 3)
yR -a/2 y*, x x*
vR -a/2 =v *, u=u *
p=p *
in the boundary layer. The Navier-Stokes equations reduce to
(1.4)
Ov
Oy
Ou
Ox
Ou
Ox
--+--= 0, u--+v
Ou
Oy
10p 021g o___p

+ 2 oy 0
p Ox Oy
after we make the substitution (1.3) and let R oe. Thus the pressure is constant across
the boundary layer and given by its value just outside, as in (1.1). The corresponding
boundary conditions are, for an impermeable wall,
(1.5)
at y
together with some conditions on u at an initial station of x, usually u 0 for y > 0 at the
forward stagnation point x x0. It is now known that under these conditions the
solution of (1.4) exists in a region Xo<X <xa if due/dx >= 0 (Oleinik (1963)) and is
unique as long as u > 0 (Nickel (1958), (1973)). Partly by making use of observation,
Prandtl conjectured that the mainstream leaves the surface when the skin friction
coefficient
(1.6)
vanishes, and that this occurs in an adverse pressure gradient due/dx < 0. In some sense
this condition defines the position of P in Fig. lb. Although this was a definite step
forward and, in another context, of immense importance for the aeronautical industry
among others, it leaves open the questions of why the vanishing of r(x) is critical and of
the mechanism by which the inviscid free streamline leaves the neighborhood of the
body.
Broadly, the theoretical development up to this point had been concerned with the
investigation of candidates for the limit solution without being able to decide, on
internal evidence, which is most likely to occur. The situation was changed in 1938
when Howarth published a numerical solution of (1.4) for an external velocity Ue
1--X. He found that r vanished at x---0.120 at which point it appeared to become
singular. Further, more detailed numerical studies confirmed his conclusion; an analysis
of the nature of the singularity was made by Goldstein (1948) and Stewartson (1958),
and finally Terrill (1962) demonstrated that this analysis and the numerical studies are
completely in accord. Some details of the analytic structure are of interest here. The
solution just upstream of the point x x,, y 0 with r(x.,)= 0 may be written as an
312
KEITH STEWARTSON
expansion, of which the leading terms are

2
(1.7)
(Xs X) 1/4
, =y/,
where 0 < p < p2... and further terms may include powers of log On substituting
(1.7), together with an equivalent form for v, into (1.4) we find that UoCX: / 2 u. cc /
whatever the values of p, p2" There must, however, be a value of n such that p, 2p in
view of the special nonlinearities of (1.4), and the corresponding equation for u, leads
to a contradiction unless p 1. Thus it is relatively easy to convince oneself by
considering the first two terms of the expansion for u that, near x Xs, y O,
(1.8)
A ly 2 + A2(xs
x)(p+l)/4y +. .,
where A now and hereafter are appropriate constants, so that it superficially looks as if
the singularity can occur. In general, however, it is the computation of the next term in
the expansion that decides whether it is a viable proposition. If there is heat transfer the
structure of the singularity becomes even more complex (Buckmaster (1970), Davies
and Walker (1977)), but there is no doubt of its presence. On the other hand, for
unsteady boundary layers this difficulty has so far prevented the development of an
equivalent expansion (see Sears and Telionis (1975), Riley (1975)).
In addition to the form (1.7) valid near y 0 there is another expansion valid when
y 1 and 0 < << 1, namely
(1.9)
V
Us(y)+A3(x,-x)l/aUs(y)+...,
A3(xs X) -1/2 Us(y) +" ",
(Goldstein (1948)) which matches with (1.7) as y --> 0 and z --> oo, respectively, and is in
excellent accord with Terrills numerical studies. This result shows that when r 0 there
is a failure of the notion that the boundary layer has only a small effect on the inviscid
flow outside. At the least, the inviscid flow must be modified in a significant way near the
body at x Xs. Prandtl pointed out that experimentally the inviscid flow in fact leaves
the neighborhood of the body at this point and introduced the term separation to
describe the phenomenon. Goldstein also showed that the solution of the boundary
layer equations could not be continued downstream of separation, if a singularity occurs
and thus we have a complete breakdown of the hierarchical system. Incidentally it does
not seem possible to confine the effect of the singularity to the neighborhood of the
separation point using interactive arguments based on the triple-deck effect (Stewartson (1970)). From these studies and others (Sychev (1972)) concludes that "the scheme
of flow, assuming that the separation of the stream on the smooth surface occurs as a
result of an adverse pressure gradient distributed over a finite segment of the body
surface
apparently does not have any place in actuality." This strong statement ran
counter to the accepted views at that time (Stewartson (1974)) but does not appear to be
seriously disputed now. One thing seems certain" The classical irrotational inviscid
attached flow round a bluff body cannot be the correct limit of the Navier-Stokes flow as
R --> oo, thus disposing of a crucial concept which led dAlembert to his paradox.
2. The triple-deck. At the end of the introduction we described how a singularity
seems to arise in the boundary-layer calculations at separation whenever the external
velocity is prescribed, but there is no compelling intrinsic reason for it to do so. Can one
construct external velocity distributions which do not lead to the singularity, and how
does the boundary layer provide the necessary constraint for this? The first question was
DALEMBERTS PARADOX
313
answered by Lees and Reeves (1964) and by Catherall and Mangler (1966); it is only
necessary to avoid prescribing the external pressure gradient. For example, if it is
related to the displacement thickness, no singularity occurs at separation. The answer to
the second is provided by the theory of the triple-deck developed independently by
Neiland (1969), Messiter (1970) and Stewartson and Williams (1969), [see Neiland
(1974), Stewartson (1974), (1980), Messiter (1978) and Smith (1979) for comprehensive reviews of the progress made in using this concept] and based on seminal ideas put
forward by Lighthill (1953).
If the boundary layer is to have a significant effect on the external flow, rapid
changes must occur in it in order that the displacement thickness be sharply increased
(and hence, also, the induced normal velocity of the inviscid flow at the body). It follows
that during the interaction viscous effects must be confined to the lower part of the
boundary layer, the remainder behaving virtually as if inviscid. Then, the boundary
layer divides into two parts, usually referred to as the lower and main decks respectively. The third deck of the triplet is the region of the external flow field which is most
significantly affected by the rapid changes in the boundary layer. It is called the upper
deck and lies just above the other two decks (Fig. 2).
The conventional boundary-layer equations (1.4) hold to leading order but their
derivation requires the application of different scaling laws which we shall not attempt
LR :Yupper
]deck
boundary
layer
LR,- [ _I
/////////////ower deck
FIG. 2. Sketch
of the triple-deck region near separation on a flat plate.
to specify as yet. The no-slip condition holds at the wall and upstream of the triple:deck
the solution must be compatible with the conventional boundary layer from which it
develops. Hence,
(2.1)
uhy
asx-,
u-hyhA(x) asy
where A is the value of r at the origin of the triple-deck coordinates, according to

conventional theory, while A(x) has to be found, as indeed does p. In a sense, the
main-deck theory plays a passive role, transmitting the effective slip velocity AA(x)
through the boundary layer and converting it into a normal velocity at the outer edge,
but this is crucial for the success of the scheme. The specific structure in the main deck is
found by an application of Prandtls transposition theorem (1905). Suppose the basic
boundary layer has the profile UB(y) in the interaction region and we base L on the
distance of the origin of the triple-deck from the nose or leading edge. Then, in the
main deck,
(2.2)
u*= Ut(y)+A(x)U(y),
v*=-A(x)Un(y)R-1/2+
from (1.4), giving us an automatic match between the x-components of velocity in the
main and lower decks. As y --> ee on the main-deck scale
(2.3)
u*- U(oe),
and, independently by M/filler (1953).
v* -R-1/Zu(oo)A(x),
am indebted to Dr. A. Kluwik for this reference.
314
KEITH STEWARTSON
so that the small parameter R -1/2 in the normal velocity at the outer edge of the
boundary layer is partly compensated by the amplifying effect of the derivative with
respect to x. This normal velocity acts at the bottom of the upper deck and generates a
pressure gradient, O(R-1/:A"(x)) in nondimensional terms, which in turn drives the
lower deck. Now we may deduce the various length scales of each deck by insisting on a
balance between the various terms of the lower deck; the lateral scales of the lower,
main and upper decks are found to be respectively R-S/SL, R-I/2L, R-3/SL. Consequently, in the main deck, the velocity component normal to the wall is O(R -1/4) and
much larger than its value outside the triple-deck region.
Although the lower-deck equations are essentially the same for supersonic or
subsonic external flows, the same is not true for the upper deck, which assumes a
hyperbolic form in supersonic flow (Moo> 1 where Moo is the Mach number of the
inviscid flow just outside the triple deck) and elliptic in subsonic flow. In each case a
simple expression may be written down connecting p and A. After applying appro-
priate but algebraically complicated scales (see Stewartson (1974) for details) the
governing equations of the triple deck may be reduced to (1.4) with p 1 together with
(2.4a)
(2.4b)
-A(x) if Moo> 1,
A(]__.Xl dxa
7r 3_oo
if Moo < 1
X mX
The corresponding boundary conditions are, for an impermeable wall,
(2.5a)
(2.5b)
u=v=0 aty=0,
y as x
-oe,
y -* A (x)
as y
3. The supersonic triple-deck. We consider the case

holds. As stated, (1.4), (2.2).have the obvious solution
(3.1)
u=y,
Moo > 1
first, and so (2.4a)
p=0,
which corresponds to an undisturbed boundary layer in the triple-deck region. For

supersonic flow Lighthills linearized theory demonstrates that it is not unique there
being two others of the form
(3.2)
y :v eXf(y),
+/-
as x
-oo,
where K 0.8272... and f is related to the Airy function. The underlying physical
explanation of this phenomenon had been given earlier by Oswatitsch and Wieghardt
(1946)--in supersonic flow any change in pressure produces changes in the displacement thickness of the boundary layer which tend to reinforce the original change. We
are especially interested here in the upper sign in (3.2) corresponding to a rising
pressure--the solution properties when the lower sign is taken are discussed by
Stewartson (1970) and little further progress has been made since. The adverse pressure
gradient provokes a separation boundary layer but now, since the displacement
thickness can exert a restraining influence on the pressure, there is no singularity at
separation. Upstream of separation, the solution of (1.4) seems to be uniquely determined by the position of separation which we shall now fix at x 0. The uniqueness is
well known to experimenters and was discussed in physical terms by Chapman et al.
(1958).
DALEMBERTS PARADOX
315
Beyond separation, reversed flow sets in near the wall and standard methods of
solving (1.4) become unstable. There is a dearth of precise theoretical results for
parabolic equations when the arrow of time points both in the direction of x increasing
and x decreasing, and so, in order to obtain information about the basic solution
properties, we must rely on structural analysis and specially constructed numerical
methods. Locally a nonuniqueness can be identified just downstream of separation
(Stewartson (1958)), but it is quite likely that, nevertheless, the global solution is unique
as long as the pressure remains bounded for all x. It is found that as x eo, a plateau is
reached in which the pressure is constant (p P-1.80 as x oe) and the forward
moving part of the lower-deck boundary layer is displaced to lie above the line y Px.
Below, there is a region of slowly moving reversed flow, the interface being a relatively
thin shear layer. In physical terms, the angle of separation of the ongoing boundary
layer from the wall is small ec (R -1/4) and thereafter it is virtually independent of the
wall. For example, if further downstream the wall is convex with respect to the fluid, the
inviscid stream does not have to follow it round. Thus, we have obtained a mathematically consistent and physically reasonable description of separatiofi in a supersonic
stream. As derived, it appears to be a spontaneous evolution of the boundary layer
which can occur at an arbitrary point of the wall. However, the pressure rise must be
consistent with the feature further downstream provoking it, for example, a shock-wave
or a ramp. Since this rise depends on the Reynolds number at separation, the
consistency requirement determines the position of the separation point.
The most satisfactory method available at present for computing boundary layers
with regions of reversed flow is probably that of Reyhner and Fliigge-Lotz (1968) with
modifications due to Williams (1975), (1979). The first step is to set u Ou/Ox =0
whenever u < 0 in a standard marching program in the direction of x increasing, thus
preserving stability and leading to the first approximation to the solution. If the reversed
fl0w region is finite in extent, the integration is stopped at reattachment. It is next
carried out in the reversed direction but, in the latest version of the method (Williams
(1979)) only over that part of the flow field in which u <0, both its extent and the
mass-flux through it being fixed from the first pass. The purpose of this integration is to
FIG. 3. Streamlines in the reversed-flow region of a boundary layer when the displacement thickness varies
1.065 to a maximum of 11.6 at x 1.35 and returns to that[or the same flat plate
at x 1.884 (see Cebeci et al. (1979) for details).
from its flat-plate value at x
316
KEITH STEWARTSON
obtain an estimate for u Ou/Ox when u < 0 and this is taken as given in the third step
which is to repeat the forward integration of the first step. The procedure is continued
and only requires a few iterations to converge. In Fig. 3, we display some recent results
obtained by an earlier version of this method (Cebeci, Keller and Williams (1979))
when the displacement thickness is prescribed. Other methods available include the
alternating-direction-implicit method with artificial time (Napolitano, Werle and Davis
(1979)), the use of the unsteady triple-deck equation (Rizzetta, Burggraf and Jenson
(1978)) and global iteration with upwind differencing (Dennis (1972)).
If the reversed flow region is infinite, then one can make use of the asymptotic
structure of the solution of the lower deck, described by Neiland (1971) and by
Stewartson and Williams (1973), to start the integration in the direction of x decreasing.
It is interesting to note that Curle (1980) has applied Van Dykes method of series
truncation to the expansion of p in ascending powers of e x, of which the leading term is
given in (3.2). He obtains good agreement with results obtained by numerical integration not only in x < 0 but also in x > 0 and even obtains Poo 1.76. This strengthens the
view that the solution in x > 0 is unique and an analytic continuation of that in x < 0.
The basic ideas of the supersonic triple-deck have been extended in five principal
directions. First of all we know that the theory agrees quite well with experiment as far
as the separation pressure and plateau pressure are concerned (Stewartson (1974)), but
the predicted slope of the pressure curve at x 0 is too small. At one time it was thought
possible that an improvement might be obtained by regarding the triple-deck solution
as providing the leading terms in expansions of the dependent variables in powers of R
and log R. However, the most recent work, by Ragab and Nayfeh (1980), indicates that
very little improvement is obtained by including the next term. Indeed, Burggraf,
Rizzetta, Werle and Vatsa (1979) have examined the relationship between the tripledeck solutions and those of the full Navier-Stokes equations and concluded that the
Reynolds number must be much higher than that commonly occurring in experiments
( 10 ) before the two become satisfactorily close. These difficulties are of course part of
the price which must be paid if the solution is obtained in asymptotic form, and strictly
we have not proved that it is asymptotic even though it is most likely to be so. It is also
possible that the difficulties may be associated with the evolutionary properties of .the
basic boundary layer on which the triple-deck structure is built. A comparative study of
numerical, experimental and asymptotic approaches to the separation problem by
IZIussaini, Baldwin and MacCormack (1980) is of interest in this connection.
Secondly, the concept has been generalized to include transonic and hypersonic
boundary layers. For transonic flows, Moo 1 and modifications must be made to (2.4a)
if (Moo-1)= 0(R-1/5). Messiter, Feo and Melnik (1971.) have shown that while the
lower-deck equations are unaltered, the relation between p and A can now only be
obtained by solving
Oy2-Ox 2
in y > 0 subject to O&/Ox -p, O/c3y dA/dx at y 0. The longitudinal scale of the
interaction region is now O(R -3/) and the pressure rise at separation is O(R-1/s). The
supersonic part of the interaction (p < 1) has been looked at most recently by Bodonyi
and Kluwik (1979), with similar results to those for Moo > 1. When p > 1, the upper deck
becomes subsonic, the Oswatitsch-Wieghardt argument is reversed and the disturbance
to the basic flow must die away. Hence, p must decrease to values less than unity,
supersonic flow must be recovered, and it is an open and tantalizing question whether
the triple-deck dies away or continues to oscillate. If the latter, there may be the
317
DALEMBERTS FARADOX
glimmerings of a theory to explain the multishocks found by Ackeret, Feldman and

Rott (1947); see Howarth (1953, vol. 1, p. 473).
For hypersonic flows defined by x=O(1), where X =M6/R, the important
modification is that even without the presence of the triple-deck there is an interaction
between the boundary layer and the inviscid external flow. Indeed, the Lighthill
eigenvalue (3.2) has its counterpart in the expansion of the pressure about the leading
edge of a fiat plate (at X 0, say) in ascending powers of X, being now algebraically small
as X 0 rather than exponentially small as x -eo (Neiland (1.970), Werle, Dwoyer and
Hankey (1973), Brown and Stewartson (1975)). For a triple-deck centered at a finite
value of X it is found (Brown, Stewartson and Williams (1975a)) that the lower deck
equations are unaltered but (2.4) must be changed, in one limiting case, to
dA
(3.4)
dp
-P- Ux
where tr is a nondimensional parameter of order t- 1/4 The general features of the

compressive triple-deck, with p > 0, are similar to those when tr 0 and it seems that r
cannot be neglected in comparisons with experiment when M > 4. On the other hand,
when p < 0 and the triple-deck is expansive, the solution terminates at a finite value of p
(Brown, Stewartson and Williams (1975b)) in contrast to the situation for r 0 when
p -o at a finite value of x (Stewartson (1970)).
Unsteady effects manifest themselves principally through the lower deck adding a
term Ou/Ot to the left-hand side of the momentum equation in (1.4) and are important
when the time-scale is 0(R-1/4). Studies to date have concentrated on the linearized
form of these equations (Schneider (1974)). Lastly, the generalization to include
three-dimensional effects changes both the lower-deck and upper-deck equations. The
scale of z, the variable in the third direction, along the wall but perpendicular to the
direction of the main stream, is also R -3/8, and if w is the reduced velocity in that
direction (1.4) becomes
Ou Ov
Ox Oy
Ou
Ox
Ow
Oz
Oy
(3.5)
bl
Ow
Ow
Ow
nt- w
-l" l)
0z
0x
0y
Op 02u
Ox Oy
Ou
Oz
Ou
Op
0z
02w
" 0y
2,
with boundary conditions
(3.6a)
u-yA(x,z),
zwD(x,z) as y-o,
where oD/Ox =-Op/Oz (Smith, Sykes and Brighton (1977)). In the upper deck the
velocity potential satisfies the linearized supersonic inviscid equations and we have
02A/O 2 dse drt

3T
e) 2 (y "1)2] 1/2
taken over the region :<x for which (x-)2> (y-n) 2 (Smith (1979)).
(3.6b)
llf[(x
The studies described so far have been local in character and need to be fitted into a
global context. This may be done by regarding the triple-deck as caused by a corner, by
strong-injection, by a trailing edge, or by a shock-wave intersecting the boundary layer.
Any of these provides the downstream boundary condition needed to provoke the
evolution of the triple-deck. Corner problems have received the most attention in the
literature and here we shall concentrate on concave corners, or ramps, the studies on
convex corners having already been reviewed by Stewartson (1974). Let us suppose
318
KEITH STEWARTSON
that, with respect to Cartesian coordinates, the wall has the equation y* 0, x* < 0 and
y*=a*x*,x*>O, where o*=0(R-1/4)>0. Then, after appropriate scaling
(Rizzetta, Burggraf and Jenson (1978)), the relevant equations for determining the flow
in the lower deck near the corner are (1.4) together with the boundary conditions
u=v=0 aty=0,
allx,
p0 asx-*-co,
P a asx-co,
u-y-A(x) asyco forx<0,
(3.7)
u-yA(x)+ax asy-co forx>0,

where a O(1) is the reduced ramp angle and the x-axis now lies entirely along the
wall. We envisage therefore a Blasius boundary layer approaching the corner, interacting with the mainstream through a triple-deck, and then reforming beyond it as
another Blasius boundary layer but at an enhanced pressure due to the action of the
corner in turning the flow. Analytic solutions are possible when a << 1 and the upstream
influence of the corner clearly manifests itself through the Lighthill solution (3.2)
(Stewartson (1970)), three-quarters of the total pressure rise occurring at x =0. For
larger values of a, numerical methods must be used (Rizzetta et al. (1978)) and a reversed
flow bubble is found to develop when a > 1.57 with separation ahead of the corner and
reattachment downstream. The computational procedure is to replace (1.4) by its
unsteady form and allow the solution to evolve from some reasonable first approximation until a steady state is reached. Studies have been carried through for values of a up
to 3.5, and the variation of pressure and reduced skin friction with x are displayed in
a=3.5
separation point
reattachment pint
(a)
3.0
P 2
-20
-10
20
10
1.0
(b)
0._/y
0.5
=0
0
-0.5
-20
-10
10
20
X
FIG. 4. Compression-ramp properties in supersonic flow (a) pressure variation (b) skin
(Rizzetta et al. (1978)).
friction
variation
DALEMBERTS PARADOX
319
Fig. 4. We see that as a increases the separation point moves upstream until at the
largest values of a the initial evolution of the triple-deck is very similar to that of a free
interaction, with a plateau pressure of ---1.8 achieved well upstream of the corner.
Thereafter, the boundary layer evolves as a shear layer until, abruptly, it strikes the wall
again. The reattachment zone for large values of a has not yet been fully worked out but
it is clear that the principal interaction is inviscid, the pressure rising from 1.8 almost to
its final value of a within a distance O(a -/) (Neiland (1970), Burggraf (1973),
Messiter, Hough and Feo (1973)). Fluid has been entrained by the shear layer over the
whole of the plateau and must be returned at reattachment to the slowly moving fluid
below. On this basis Burggraf (1975) was able to conclude that the length of this region
is O(a3/:z), a result in quite good agreement with experiment. In physical terms this
means that separation occurs at a distance 0(a3/2) upstream of the corner when a* is
small but finite; i.e., the Reynolds number effect disappears. This property is characteristic of laminar boundary-layer observations. Further details of the flow properties in
the separated region are incomplete. The streamline springing from the wall at
separation cannot reattach to the wall in the inviscid region (Neiland (1970), Burggraf
(1973)) and the reattachment must occur in a quasi triple-deck region further downstream, of length O(1) with a corresponding pressure rise O(1). Its structure can be
stuaied independently of the rest of the flow field and an investigation of its properties
has been made by Daniels (1979), (1980).
The major outstanding question in the ramp problem when a >> 1 is, however, the
nature of the reversed flow field. Various suggestions have been made, usually based on
the Prandtl-Batchelor theorem that the vorticity is constant since the streamlines are
closed (Messiter (1978), but it does not seem possible to write down a consistent
description on this basis. The numerical studies indicate that the flow field is still
evolving even at a 3.5, and we cannot even be sure that the eddy remains simple as
a oo. Related studies by Cebeci et al. (1979) already mentioned, and by Leal (1973),
on the flow into a right-angled corner at finite Reynolds number (_-<800), are not
inconsistent with the notion that secondary eddies may form at large enough values of a.
Multiple eddies in closed regions have of course often been observed. Be that as it may,
the question of the eddy structure recurs over and over again in the rational theory of
high Reynolds number flows and its resolution would give a tremendous fillip to
investigators and open the way to further progress on a large scale.
Injection into a supersonic boundary layer is another example of a flow for which
the ideas of the triple-deck are relevant. Consider a flat plate of finite length 0 < x* < L,
across which fluid is being injected into the boundary layer at a uniform velocity v* over
the region 0<xo* <x<L. For v* << R-1/2Uoo the boundary layer remains virtually
undisturbed, the injected velocity being easily absorbed. At larger values of v* this is no
longer possible and separation occurs, the boundary layer being apparently blown off
the plate. For example, if xo* 0 separation occurs when
(3.8)
x* =0.7456(R-1/2Uoo/v)2L
(Catherall, Stewartson and Williams (1965)). The subsequent behavior of the boundary
layer is complicated particularly near the separation point but its main features have
been elucidated by Diver (1979). She followed the work of Pretsch (1944) and of Cole
and Aroesty (1968) in assuming that after separation the injected fluid moves slowly but
inviscidly, displacing the oncoming boundary layer away from the wall. She also
investigated the entrainment from below into the boundary layer on the assumption
that apart from being displaced it evolves independently of the injected fluid. A sketch
of the flow field is given in Fig. 5 where injection is assumed to begin at x0*, separation
320
KEITH STEWARTSON
viscous
L.E.
r6gi
xo
Xf
FIG. 5. Flow structure for a supersonic boundary layer separating and reattaching due to infection over a
finite region.
x.
Consider a
occurs at x*, injection ceases at x* and reattachment takes place at
streamline crossing the plate at a point A of the separation region and entrained into the
shear layer at B, the position of B relative to A being fixed by mass conservation. Along
this streamline we have, since viscous terms may be neglected and 0/0y* << O/Ox*,
p*+1/2p*wU*2=p*(A),
(3.9a)
and hence
O*(x*)
(3.9b)
q,*(x)
dO*(A
dO*
p wbl*
giving us another equation connecting 6" and p*, in addition to the Ackeret formula
(2.4a). Here O* is a stream function of the injected fluid defined by OO*/0y*=
p*wU *, c34,*1x* -p *wV * We may now easily infer that, whereas in the unseparated part
of the boundary layer the pressure induced by the boundary layer is only 0(R-1/2),
once separation occurs it rises to O(R-I/3). The thickness of the injected region is now
O(R-1/3L) and much larger than that of the oncoming boundary layer. The position of
separation x is fixed by classical methods and reattachment by the condition that all
the fluid injected in x < x* < x 1" is entrained. In order to achieve this it is essential that
This proves possible in
take account of the cutoff at
the solution in x* < x*<
spite of the apparent inability of disturbances to propagate upstream in any of the three
regions of flow because the solution of (3.9) near x* x* is locally not unique, in a way
roughly reminiscent of (3.2) although the nonuniqueness is now algebraic. Diver, partly
in collaboration with Stewartson (1978), considered a wide variety of problems using
this approach; the only parts of the flow field not fully understood are separation, which
appears to be much more subtle than in a conventional triple-deck, and a small
recirculating region near reattachment.
As v* increases further, the separation point moves towards x0* and when v* is
O(R -3/8 U) the strong-injection theory of Smith and Stewartson applies with separation ahead of x0* occurring through a standard triple-deck. Fuller accounts of these
theories may be found in Stewartson (1974), Smith (1979) and Napolitano (1980).
Numerical solutions at finite Reynolds number have been considered by Vatsa (1975)
and by Werle (1979). Broadly, their conclusions are in line with the rational theory, but
some of the features they find are inexplicable on this basis.
Of other applications, little work has been done so far on the shock-wave
boundary-layer problem, notwithstanding its importance, and we shall defer the
trailing-edge studies until 5 below.
The triple-deck theory is asymptotic, and in many problems of flow past bodies,
either the Reynolds number takes on only moderate values or the flow is turbulent. The
ideas may be incorporated into nonrational or semi-empirical theories to predict the
flow properties in such situations with a considerable degree of success. Thus Tu and
x.
321
DALEMBERTS PARADOX
2.2
//
1.4
corner
1.0
0.8
1.2
1.
FIG. 6. Comparison of supersonic interacting boundary-layer predictions of pressure with solutions of the
Navier-Stokes equations and with experiment]or a 10 comparison ramp. R 68,000, M 4, adiabatic walt.
Navier-Stokes Carter (1972)
boundary layer Vatsa (1975);
experiment Lewis et al. (1968).
Weinbaum (1976) have developed an approximate theory for finite Reynolds number
flows by allowing for the fact that the lower deck occupies a finite part of the boundary
layer. Werle and Vatsa (1974) noticed that the lower and main decks arise as limits of
the boundary layer equations and chose to solve the complete boundary layer equations
as a whole with a suitable interaction formula for the pressure. As Fig. 6 shows this leads
to quite good agreement with experiment. Finally Adamson, Messiter and Melnik and
their associates have adapted the ideas to study turbulent interactions. The reader is
referred to the articles by Messiter (1978) and by Adamson and Messiter (1980) for an
account of their work.
4. The incompressible triple-deck. When M < 1, and for simplicity we shall take
M 0, as typical, the interactive equation for the pressure changes to (2.4b) and the
Oswatitsch-Wieghardt argument for an eigensolution fails. Any spontaneous disturbance in the boundary layer is damped out by the induced pressure. Nevertheless, an
eigensolution can be found, provided one does not insist that p remain bounded. For
there is an eigensolution of (2.4b) given by
A 0 if x<0,
p -a(-x) 1/2
(4.1)
A -ax 1/2 if x>0,

p O,
and there is a possibility that, for a suitable choice of a, the boundary layer smooths
away the singularity at x 0 and provides compatible connecting relations between p
and A when Ixl is large. Such indeed was Sychevs opinion (1972), and he developed the
As x + it has a close
appropriate asymptotic structure for the lower deck as J
similarity to that for supersonic flow with a displaced shear layer and a slow reversed
flow below, but as x
the decay of the solution is now algebraic rather than
exponential. He also observed that in physical terms the pressure takes the form
I+
_ceR-1/16 pUoo2 (x* x ,)1/2 + const,

just upstream of the interaction at x* x*, and so the adverse pressure gradient is weak.
(4.2)
This led him to the famous conjecture that separation occurs at a pressure minimum in
the limit R m.
In connection with the existence of this eigensolution it is important to know
whether the solution can be continued smoothly through separation and the point was
322
KEITH STEWARTSON
considered by Dijkstra (1978) who demonstrated, in the context of a smooth step, that
the Goldstein singularity does not occur. Smith (1977) then carried out a numerical
solution of Sychevs problem, using the Reyhner and Flugge-Lotz (1968) approximation in the region of reversed flow, and found that consistency between the lower and
upper decks is achieved if
c =0.44.
(4.3)
In view of this approximation the answer is subject to some (and probably small) error,
but there can be no doubt that an eigenfunction exists. The method he used involved
many iterations with a small relaxation parameter. A new method for dealing with the
incompressible triple-deck has been derived by Veldman (1979) and offers hope of
substantially reducing the computational effort required for solving such problems as
Sychevs. Basically, the idea is to solve the two equations simultaneously in a series of
sweeps updating the local value of p from the Hilbert integral. A surprising feature of
the procedure is that it remains stable in the reversed flow regime.
The demonstration of the existence of Sychevs eigensolution has an important
bearing on the title of this lecture for we now have the possibility of developing a
rational theory of flow past smooth bodies at large Reynolds number which, while not
complete, is at least free from any obvious inconsistencies and contradictions. Let us, as
a paradigm, focus attention on a fixed circular cylinder of radius L placed in an
otherwise uniform stream of fluid moving with velocity U. The obvious candidate for
the limit solution is the Brodetsky form of the Kirchhoff free streamline flow and Sychev
(1967), (1972) and Smith (1977), (1979) have described many of its structural properties. In the limit solution two free streamlines spring from the cylinder at angular
distances of 55 from the forward stagnation point, the internal flow accelerating up to
these points. As x oe the free streamlines are asymptotic to a parabola II whose latus
rectum is fixed by the drag on the cylinder. When R is large but not infinite, a weak
adverse pressure gradient with a singularity at detachment of the form (4.2) is added to
the attached external flow. The Sychev eigensolution now comes into play, enabling the
main stream to leave the neighborhood of the cylinder and inserting a weak reversed
flow in the gap which opens up. The free streamlines have a curvature O(R-a/16(x*x*) -/) just downstream of detachment, formally singular at x* but soon becoming
the curvatures are finite everywhere and the free
negligible. In the limit R
streamlines come smoothly off the cylinder. In order to balance the momentum defect
at infinity with the force on the cylinder it is essential that when R is finite the free
streamlines join up at a distance B, say downstream of the cylinder. There is then a wake
bubble behind the cylinder in which the fluid motion is sustained by entrainment into
the free shear layers near the free streamlines.
Both Sychev and Smith make the reasonable assumption that the motion of the
fluid inside the wake bubble is very slow so that the pressure is almost constant there. It
follows that the bubble shape is an ellipse whose dimensions are fixed, first by the
Brodetsky parabola II which gives the shape of its nose, and second by the requirement
of momentum conservation mentioned earlier. Its length is found to be
(4.4)
0.34LR,
and the reduced pressure inside is
(4.5)
Peddy -1.37R -1/2.
In Fig. 7, we give comparisons between this theory and various studies at finite
Reynolds number, both computational and experimental; others may be found in
323
DALEMBERTS PARADOX
(a)
(b)
15
-0.4
10
5
-0.1
50 2R 160
50
2R
160
FIG. 7a. Eddy length as afunction of Reynolds number. The solid line is the qsymptotic result apartfrom an
origin shift. The crosses are experimental and the others are numerical.
FIG. 7b. Variation of the eddy pressure with Reynolds number. The solid line is the asymptotic result, the
squares are numerical values and the remainder are experimental.
Smiths second paper (1979). Although we cannot put a finger on anything intrinsically
wrong with the theory, the motion of the fluid in the bubble remains to be described and
until this has been achieved it must be regarded as a less-than-satisfactory feature. 2
Indeed the nature of the flow inside large bubbles presents a challenge to students in
many problems of asymptotic theory, but this is the prime example of a large bubble
which has not been built up from small dimensions as some parameter is varied. As we
shall see in the study of trailing edge flows, attempts to develop a Sychev-Smith bubble
lead to an interesting hysteresis.
In other cases moderately sized bubbles have been obtained which are consistent
with the principal assumptions of this theory, but it has not proved possible to construct
an analytic description of them that is capable of being generalized to the large
dimensions required. Nevertheless, it is generally agreed that the pressure variation in
such bubbles must be very small, a key requirement.
The generalization of the two-dimensional triple-deck theory to include threedimensional effects is obviously important but difficult in both the lower- and the
upper-deck calculations. Following on the successful studies on two-dimensional
humps in boundary layers (Smith (1973), Napolitano, Werle and Davis (1978)), some
extensions to three dimensions have been made. The lower-deck equations are given in
(3.5) and the Hilbert integral of the upper deck is replaced by
(4.6)
c32A/c32ddrl
p=-- f;I [(x_j)z+(y_n)2],/2.

1
Linearized solutions have been obtained by Smith, Sykes and Brighton (1977) and
indicate the existence of a "corridor" effect behind the hump where the disturbance to
the flow is concentrated. Numerical studies of the corresponding nonlinear problem
have been made by Sykes (1980), which reveal a remarkably complicated flow structure
when separation occurs.
Dr. Smith has informed me that B. Fornberg has carried out extensive studies of flow past circular
cylinders at values of R up to about 300. He finds that the linear growth of the bubble length with R is abruptly
halted at R 260. Dr. Fornberg believes that his results cast "definite doubt on (the validity of) Brodetskys
solution". In my opinion the significant departures from this solution as amended by Sychev and Smith only
occur at large distances from the cylinder and lead to such a serious inconsistency that one must rather doubt
the validity of the numerical work until it has been confirmed independently. For, if the wake is of finite length
when R 0% the force on the cylinder must then be zero. In fact the force obtained shows no signs of
approaching zero as R oo and is always greater than Brodetskys value [see Fornberg (1980)].
324
KEITH STEWARTSON
5. Trailing-edge flows. Since the limit solution as R c of the flow past a body
of finite thickness involves separation, the basic flow must include a wake region of
reversed flow almost certainly of infinite extent. It would be satisfying if such a flow
could be shown to arise naturally by evolution, as the body thickness increases, from a
flow without separation. The obvious candidate for such a study is the finite flat plate of
zero thickness defined by y*= 0,-L < x*< 0. Having obtained an asymptotic theory
for such a body it might be possible to approach the Sychev-Smith theory for bodies of
finite thickness by gradual stages. Even for this, the simplest of all finite bodies, the full
asymptotic theory is complicated, requiring the elucidation of a large number of
different r6gimes of flow just to obtain a uniformly valid first approximation.
For a symmetrically disposed plate we confine attention to the region y > 0 and
beginning with the inviscid solution u* U, v* 0, remove the inconsistencies one by
one. First, the no-slip condition is violated by this solution, and so we add a Prandtl
boundary layer in -L <x*< 0 together with a Goldstein wake (1930) in x*> 0. The
displacement effect of these shear layers is of higher order (R -1/2) except in the vicinity
of the leading edge (x*=-L) and the trailing edge (x*= 0). The inconsistency at the
leading edge requires the numerical solution of the full Navier-Stokes equations for a
semi-infinite plate (Van de Vooren and Dijkstra (1970)) and then uniform validity in
this neighborhood has been achieved. That at the trailing edge is more complicated due
to the existence of an oncoming boundary layer and a single once-and-for-all study of its
neighborhood as a whole is not possible. Instead it must be broken up into a number of
separate regions of which the most important is the principal triple-deck discussed in
essence in 2. The necessity for its appearance can be justified by examining the nature
of the singularities that arise in the shear flow, especially in the displacement thickness
as x* 0 (Stewartson (1969), (1974) and Messiter (1970)), but it is sufficient to say that
any change in the boundary condition that causes a small disturbance in the boundary
layer must be associated in some sense with a triple-deck. The structure is as described
in 2, the crucial equations controlling it being (1.4) and (2.4) while the boundary
conditions are the same as (2.5), except that (2.5a) is replaced by
Ou
v=--=0 aty=0, x>0,
u=v=0 aty=0, x<0,
(5.1)
0y
the trailing edge now being placed at x y 0. The conditions when x > 0 reflect the
smoothness of the flow downstream of the plate and across the line of symmetry. The
scale of the triple deck is LR -3/8 in all directions and the pressure variation is O(R -1/4)
compared with O(R -1/2) elsewhere in the flow field. Physically the release of the fluid
from the restraining effect of the plate generates an acceleration which causes a
dramatic decrease in the displacement thickness of the boundary layer thus setting up
the pressure gradient. Upstream of the triple-deck the solution must match with the
Blasius solution so that u y, A 0, p 0 as x -c in the lower deck. Downstream it
must match with the Goldstein wake so that, as x
(5.2)
u=(1/4x)/31/2ka(x),
wherex =y/3(2x)
/3
k,, +2k,k, =0,
kl(0)= k (0)=0, k
and hence p 0, A 0.8920x /3.

At the trailing edge itself x y 0, we must ensure that A is sufficiently smooth or
an unacceptably strong pressure gradient is generated there. As a result it turns out that
while p is smooth for x < 0, its gradient must develop a singularity as x -* 0 +. In fact
(5.3)
pp(O)+O.6133x2/3+
asxO+,
DALEMBERTS PARADOX
325
and the solution in the lower deck takes on a double-structured form. Worthy of special
note is the behavior when y ---x x which is given by
u(x) 1/3 k(x), k(O)=k (0)=0, k-18x--O asx

the last condition ensuring the relative smoothness of A at x 0. Numerical solutions of
the triple-deck equations have been carried out by Burggraf and Jobe (1973), Veldman
and Van der Vooren (1974) and by Melnik and Chow (1975) for incompressible flows
and by Daniels (1974) for supersonic flows. The drag coefficient on an adiabatic plate is
given by
Co
(5.4)
1.328
R/2
+[M_l13/SRT/S+",
on the assumption that the viscosity of the fluid is proportional to temperature, where
0 2.67 if M < 1 and 0 2.03 if M > 1. The agreement with numerical and experimental studies at finite values of R is remarkably (and sometimes fortuitously) good. As
Fig. 8 indicates, the error is small even when R- 10 and the lateral extent of the
triple-deck exceeds the length of the plate.
0.04
FIG. 8. Drag, of a flatplate as a function of Reynolds number. -according to (5.4), @ experimental, A from
Navier-Stokes equations,
from Blasius equation.
Even with the inclusion of this triple-deck into our structure, we have not
succeeded in making it uniformly valid because of the singularity in the pressure
gradient at x 0 throughout the lower and the main decks. Further substructures must
be introduced to smooth them out, and Veldman (1976) has identified a double
staircase of triple-decks needed for this purpose. At the end of the staircase is the
central region of the trailing edge of extent O(R-3/4L) in both directions when all
viscous forces are equally important. A discussion of this region was given by Stewartson (1969), but he misconceived its role in the overall problem. Veldman has suggested
that this staircase has an infinite number of steps.
Further progress towards a link-up with the Sychev-Smith model for a finite body
has been made using analytical arguments only. The flat plate may be replaced by a thin
body of thickness ratio r* having a wedge-like trailing edge. Then, if z* 0(R-1/4), the
above theory may formally be adapted (Stewartson (1974), with changes needed only in
the boundary conditions to describe the flow in the neighborhood of the trailing edge.
Since we know (Riley and Stewartson (1969)) that, according to classical boundary-layer
theory, which assumes the external flow to be attached and inviscid, separation must
326
KEITH STEWARTSON
occur within a distance 0(7"3/2L) of the trailing edge, we can expect that the solution of
the triple-deck equations will lead to reversed flow when z*R 1/4 1, but the necessary
calculations have not been carried out. Guiraud and Schmitt (1975) have gone on to
argue that the necessary link-up with the Sychev-Smith theory can be achieved when
-r*R 1/4 >> 1, but again a numerical confirmation is needed.
Once the condition of symmetry is relaxed, the numerical solutions of the corresponding problems do not, at the least, give solid support to these analytic conjectures, hs we shall now see. First of all it appears to be pointless to consider a plate of
zero thickness at an angle a* of incidence to the main stream, no matter how small a* is,
because irrecoverable separation occurs at the leading edge. This result follows from
studies by Cebeci, Khattab and Stewartson (1980), who have shown that for thin elliptic
wings, of thickness ratio r*, separation occurs close to the leading edge when a*=
O(z*) and by Cebeci, Stewartson and Williams (1980) who demonstrate that the
Goldstein singularity cannot be removed by a triple-deck argument. The inference is
that when separation occurs it must be of the Smith-Sychev type, there being a rapid
and dramatic change in the external flow as a* passes through the critical value.
Let us then suppose that the fluid is incompressible and we have an airfoil of thin
but finite thickness and with a sharp trailing edge. Let us also suppose that attached
inviscid flow is the correct limit of the solution of the full Navier-Stokes equations as
R c, except in particular neighborhoods, provided that the Kutta condition is applied
at the trailing edge. Our endeavor is to make this solution uniformly valid and obtain a
viscous correction to the lift. The significant angle of attack is a* O(R -1/16) just as in
the Sychev-Smith theory. In terms of a reduced angle of attack a, numerically
proportional to a*R 1/16, the determination of the flow near the trailing edge may be
reduced to the solution of two triple-decks, one above the airfoil and one below which
join smoothly together downstream (x > 0). The governing equations are as usual but
the boundary conditions are changed. For details reference may be made to Brown and
Stewartson (1970) but the crucial change is that
p+(x)+a(-x)l/:O asx-eo
to match with the inviscid pressure derived from an application of the Kutta condition.
It might be thought that as a increases a link-up with the Smith-Sychev theory would be
obtained first by separation occurring on the upper side of the airfoil (a > 0). The
separation point is then expected to move upstream until the change in boundary
conditions and the link-up with the triple-deck on the lower side become of vanishingly
small importance. Further, this should happen as a 0.44, as in (4.3).
Chow and Melnik (1976) have carried out the difficult task of integrating the two
sets of equations numerically and their results contain some surprises. They were able to
find solutions for 0 _-< a <_-0.45 and obtained the dependence of al, in the formula
(5.6)
CL=2ra(1
al(a)
R 3/8
a2
logR
R 1/-
O(R -1/2)),
for the lift coefficient, predicted by the theory; it varies from 0.53 when a 0 to 1.04
when a =0.45. The value of the constant a2 was later obtained by Brown and
Stewartson (1975); it arises from the wake curvature. In Fig. 9 we display the variation
of z Ou/Oy at y 0 on both sides of the plate for various values of a. The slow decay of
the triple-deck solutions towards the Blasius form for all a > 0 is immediately apparent,
but this is not all due to the interaction. More significant is the fact that separation does
not occur, and the minimum value of z in the upper triple-deck occurs at a value of x
327
DALEMBERTS PARADOX
(b)
(a)
1.0
Blasius
=0.1
=0.2
=0.3
a=0.4
a =0
-2.0
=0
--"--
-1.5
0.4-=0.3
a 0.2
1.0
-30
-10
-20
-30
-0 X
/I
J /I
-10
FIG. 9. Skin friction distribution near the trailing edge of a fiat plate. (a) Top. (b) Bottom.
which increases with a and appears to be moving towards the trailing edge, not away
from it as a match with the Sychev-Smith solution requires. Indeed the solutions
terminate when separation occurs at the trailing edge, for the solution just downstream
of it depends on the existence of k3(x), satisfying (5.2) with
(5.7)
k3 18r+/-(O)x --> 0
as X "-> +
where r+(0) are the values of r at x 0- above and below the airfoil. Such a solution
does not exist if r/(0)< 0! Chow and Melnik conjectured that r+(0)= 0 if a 0.47.
Thus the attempt to approach the Sychev-Smith solution by increasing a from zero has
to come to an end in a quite different form from that required, and possibly at a larger
value of a. The resolution of these discrepancies is one of the most important
outstanding problems in these asymptotic studies.
A similar situation occurs in the supersonic analogue (Daniels (1974)). Now the
angle of attack a* is 0(R-1/4), and again we may reduce the determination of the flow
in the neighbourhood of the trailing edge to the solution of two triple-decks, ofie above
and one below the airfoil, joining up smoothly downstream of it. The usual triple-deck
equations apply, including (2.4), but there are some changes in the boundary conditions, (5.5) now being
(5.8)
p+/-(x) + a
as x --> -oo,
where a is the reduced angle of attack (oca*R1/4). If separation occurs on the upper
side of the airfoil at large negative values of x, we must expect that a 1.8, for it takes
place through a free interaction, as described in 2. The final pressure achieved
downstream of the airfoil is the ambient pressure (p 0) and there is a large region of
fluid slowly moving upstream. However, as a increases from zero Daniels does not find
that this limit is being approached at all. The situation is in fact similar to that for
incompressible trailing edges, except that he is able to be precise about the value 2.05 of
a for which separation occurs at the trailing edge and the progression of solutions comes
to an end. Thus, whether the flow is supersonic or subsonic, it has not yet proved possible
to join the eigensolutions for the triple-deck on to a family of solutions which include
unseparated flows, even though it appears to be feasible using structural arguments.
The behavior of unsteady triple-decks has also attracted attention. Interest here
has been in the generalization of the Kutta condition to include unsteady flows and the
possible impact on the noise radiated from a finite plate when subject to a trubulent flow
field. With the definition S oo*L/Uo, where o* is the frequency of the oscillation
imposed, Brown and Daniels (1975) showed that the triple-deck formulation is
328
KEITH STEWARTSON
unaltered if $ << R +a/4, while if $ R +a/4, time may be regarded as a parameter except
in the lower deck where (1.4) must be generalized to an unsteady form
Ou
Ot
(5.9)
Ou
Ox
+ U--+ V
Ou
Oy
10p 02hi
2,
p Ox Oy
Obl OV
-t-
Ox
However, in view of the relation connecting p* and

theory; namely
e*
Oy
O.
in classical boundary layer
10p* U*q-----f-e
OU*e OU*
Ot*
OX*
OX*
Pe
(5.10)
care must be taken upstream of the triple-deck. Over the majority of this region only the
last term on the right is important, but there is a fore-deck region of length O(LS -a) just
before the triple-deck where both terms are significant. In the upper part of the tripledeck (5.10) itself must be modified. Indeed, if 1 <-S <=R 1/4, Brown and Cheng (1980)
have found the necessity for yet another precursor region, of thickness LS -3/2
wherein the lower deck adjusts to achieve the quasi-steady form required in the
triple-deck.
Daniels (1978) has studied the viscous correction to the Kutta condition in detail
when $ R +a/4 and the fluid below the plate is stagnant. This problem was earlier
examined by Orszag and Crow (1970) on an inviscid basis allowing for three trailingedge conditions, full Kutta, rectified Kutta and "no Kutta," and their conclusions have
been applied to the problem of finding the noise radiated by a flat plate (Crighton
(1972)). Daniels found that application of the full Kutta condition leads to consistent
results, while the other conditions lead to inconsistencies when the amplitude of the
oscillation is large enough (->-R-7/16L), and conjectures that in fact they may always be
excluded. Support for this view is supplied by the observations of Bechert and
Pfizenmaier (1975). For further discussion of these questions the reader is referred to
Rienstras thesis 1979).
6. Internal flows. Constriction in pipes and channels may also be treated by thie
methods of the previous sections, although, as Smith and his associates have shown, in
the cases of most interest when they are of the same dimensions as the duct, significant
changes of approach are required. Let us concentrate on channel flows and suppose that
upstream of the constriction the undisturbed flow is Poiseuille. Taking the channel walls
as y* 0, a and basing the Reynolds number on channel width, we then have
(6.1)
u*=(ay*-y*2), v*=0,
a
p.=
pU2 + const,
a
when x* is large and negative.

Consider first symmetric constrictions, so that v*= 0 when y* a/2, and let the
wall y* 0 be distorted to the form y* ahF (x), x x* / a, in the neighborhood of x 0
where F is an O(1) function and h is to be varied from vanishingly small values to O(1)
values. There are three stages in the evolution of the flow field as h increases. First, if
h O(R --a/3) a thin boundary layer also of thickness O(R -a/3) is generated near each
wall and a pressure gradient O(R -2/3) is induced to ensure that, to leading order, the
displacement effect of the boundary layer is zero; otherwise it would provoke a core
flow containing a stronger pressure gradient. We have in effect a lower-deck problem,
but with A 0, and separation can only occur on the downstream side of the constriction where the pressure gradient is adverse. Further, since the boundary-layer
329
DALEMBERTS PARADOX
problem is of the inverse type, i.e., p has to be found, separation is regular (Smith
(1976)).
As h increases to O(R -1/6) separation on the downstream side becomes complete,
and a free streamline leaves the constriction virtually at its maximum height and
thereafter is almost parallel to the channel walls. On the upstream side the displacement
effect, previously of higher order, now begins to bear on the boundary layer and an
interactive problem must be solved (Smith, (1978)). The connection between A(x) and
p(x), needed to solve (1.4) and (2.5), is found now by solving the differential equation
(6.2)
Y2)V2t -2&,
Y y* / a,
subject to
0
(6.3)
at Y
2,
as x
a
O=-1/2(hR1/6)ZF(x), --=A(x)
-oo,
at Y=0.
Upstream separation is now possible, and occurs if hR 1/6 is large enough. When
hR1/6-->oo, the position x of separation tends to minus infinity. For (6.2) has eigensolutions of the form
(6.4)
e V"XFn Y),
with Fn(0)= F,(1/2)= 0, for an ordered set of y, of which 31

it may be shown that
5.175 is the smallest, and
2
xs-----log(hR 1/6) +O(1) whenhR1/6>>l.
(6.5)
71
Hence, formally, for a severe constriction with h

(6.6)
xs
371
O(1),
log R + O(1).
The correctness of (6.6) has been verified numerically for a symmetric semi-infinite step
by Dennis and Smith (1980), as Fig. 10 shows, and also by structural analysis of the
situation when h O(1) (Smith (1979)).
o.4
0.3
XS
OJ
0.2
0.1
560
lObO
FIG. 10. Eddy length ahead of a step as a function
1500
of R. (6.6), C) numerical.
330
I,:2.
STEWARTSON
When the constriction is asymmetric the scaling laws are quite different, because the
boundary layers on the two walls are not identical and the displacement effect of the one
may be compensated for by the other without the core flow being significantly affected
(Smith (1977)). The smallest significant height of a constriction is O(aR-2/7), when a
triple-deck is set up between the two boundary layers, the core flow acting somewhat
like the passive main deck of 2. The streamwise extent of this triple-deck is O(aR 1/7),
and so extends a long way upstream. As the height of the constriction increases
separation occurs fore and aft of it, and when its height is comparable with the channel
width we may expect that the first separation occurs so far upstream that it must be
controlled by a free interaction. For each boundary layer we must now solve (1.4)
writing the pressure as pu(x) on the upper wall and pl(x) on the lower wall. The
appropriate boundary conditions are given by (2.5), with Au(x)=Al(x), and then
u*, p* have order of magnitude R -2/7, R -4/7 respectively. The relation between the
two pressures is found by integrating the y* equation of momentum across the core and
taking the perturbation in the core flow to be given by (2.2). Thus the control exerted by
the core is of higher order than in the standard triple-deck, the reason being that to
leading order there is an automatic match at its junction with the boundary layers. We
have, using (2.2), that
(6.7)
Pu -pl
A"(x)
f..o (Y- y:Z) dY
A"(x).
The same situation, with the transverse variation of pressure across the main deck
playing a crucial role, occurs in the analysis of wall jets near trailing edges (Smith and
Duck (1977), Messiter and Lifian (1976)).
The two boundary layers on the duct walls, together with the connecting formula
(6.7), have an eigensolution with a structure similar to that of (3.2) when x is large and
negative. Thus
(6 8)
A,
a ex
A --al e x
where
(6.9)
[-180 Ai(0)] 3/7 -+- 5.19,
confirming that the flow can anticipate the presence of the constriction. The further
evolution of the boundary layers must now be examined numerically. With the
constriction on Y 0, the pressure in the lower boundary layer increases, p"ovoking
separation at some value xs of x. Meantime the pressure in the upper boundary layer
decreases, strictly without limit, in the manner of the expansive free interaction in
supersonic triple-decks (Stewartson (1970)), becoming singular at x x2, where
(6.10)
Near x
X2=Xs+0.54.
x2, Pu -2{25(Xz-X)} -4. It is reasonable to identify this value of x with the

position of the constriction, bearing in mind that in physical terms the length of the
constriction is O(a) while that of the free interaction is O(aR1/7). We conclude that
when the constriction is severe separation occurs at a distance
(6.11)
0.49 (2R)1/7a
ahead of it, and note that the free interaction solution will be modified near the
constriction because the boundary conditions on (1.4) are changed. Further extensions
of this theory including a study of reattachment have recently been made by Smith and
Duck (1980).
331
DAI_,EMBERTS PARADOX
7. Unsteady boundary layers. The survey of the advances made in our understanding of high Reynolds number flows by the use of the concept of the triple-deck is by
no means complete. Additional references may be found in the review articles already
cited, and important advances in stability theory have recently been made by Smith
(1978), (1980) using this concept. Some discussion of time-dependent flows has already
been given here, but there is a controversy in the literature about the fundamental
properties of unsteady boundary layers which has important implications for the
general theory of high-Reynolds number flows and so deserves special emphasis. The
question at issue is whether the unsteady boundary-layer equations subject to a
prescribed external velocity Ue(X, t) can develop a singularity at a finite time having
been free of any singularities for < ts.
Were the answer no, then the evolution of the flow past a smooth body impulsively
0 with uniform velocity in a fixed direction through an infinite fluid
set in motion at
otherwise at rest could be described as follows. The initial flow relative to the body is
inviscid, and of the type which in steady flow would lead to DAlemberts paradox,
except for a thin layer of thickness L(t/R)I/2 near the body where viscous effects matter
>-0.322 for a circular
1 (specifically, Ut*/L
(Bar-Lev and Yang (1975)). When
local
a
the
of
radius
region of reversed
boundary
develops
layer
unsteady
cylinder
L)
For large
and
laterally.
both
and
increases
this
as
spreads
longitudinally
region
flow,
values of the rear stagnation boundary layer is exponentially thick (Proudman and
Johnson (1962)) on the boundary-layer scale, and this suggests that the inviscid flow
1/2 log R and then
must be modified by the displacement of the boundary layer when
on a time-scale which is virtually inviscid. Thereafter, the inviscid flow field and the
boundary layer evolve together until a steady state is reached which is claimed here to
be a Kirchhoff free-streamline flow. A vital step in this description has been denied by
the proposition that an unsteady boundary layer can develop a singularity at a finite
time (Sears and Telionis (1975), see also Riley (1975)). Unsteady boundary layers can
contain singularities but the only ones found for two-dimensional flows with certainty
are for semi-similar flows; e.g., Williams and Johnson (1974) considered Ue
1--x(l+t) -1 and, assuming that the velocity is a function of =x/(1 +t) and rt
y/(1 + t) 1/ only, found a singularity to occur at a finite value of Furthermore, the
solution properties near the singularity may be found as a special case of Browns (1965)
theory of singularities.
Sears suggested that a singularity can occur in the following circumstances. Let the
singularity be at x =xs(t), y ys(t). Then at this point Ou/Oy =0 and u =2s. The
Williams-Johnson singularity satisfies this criterion and indicates that the singularity
need not be centered at the wall. The unsteady boundary layer on a moving wall is
another example. Using numerical methods, some workers claim to have discovered
spontaneous occurrence of singularities, i.e., to have identified a time ts as defined
above, but most of their claims appear to have been disposed of by two detailed studies
of Cebeci (1978), (1979). For one of these, the boundary layer due to the impulsive
motion of a circular cylinder, he established that no singularity can occur for < 1.4 and
inferred from the general structure of his solution that none occurs at a greater value of
t. While the smoothness of solution is not disputed for < 1.4, two other recent studies
by Shen (1978) and Wang (1979) allege that the solution breaks down at a slightly larger
value of t. Shens numerical study, based on Lagrangian coordinates, suggests that the
singularity manifests itself just downstream of separation but as y oo, and Wang, while
agreeing on the longitudinal location of the singularity, infers that it is centered at y 0.
Thus, there is still an element of uncertainty about the existence, or otherwise, of a
t, and it is important that the controversy be settled. For if t does exist, the relatively
:.
332
I<ErrI4 STEWARTSON
straightforward calculation of the unsteady boundary layer comes to an end at

ts.
The asymptotic theory of solutions of the Navier-Stokes equations must be extended to
include mutual interaction between the boundary layer and the inviscid flow in the
neighborhood of
ts at the least, and possibly also for > t. For this purpose an
analytic description of the singularity according to conventional boundary-layer theory
must first be obtained, analogous to the Goldstein expansion (1948) and to date this has
not been achieved. If, as Sears and Telionis (1975) have suggested, it is similar to that
expansion, the experience of Stewartson (1970) would suggest that the description of
the mutual interaction between the boundary layer and the inviscid flow is likely to be a
formidable task.
The computation of unsteady boundary layers, such as that on a circular cylinder, is
made especially difficult, even apart from the vexing question of the singularity, by two
features. First, downstream of the onset of reversed flow the boundary layer thickens
dramatically, and second, above the separation point the velocity u varies very rapidly
with x. These two features may be inferred from the solution computed by Cebeci for
_<- 1. The most ambitious computational study for these problems was carried out by
Robins and Howarth (1972) for the Proudman-Johnson equation, and they required
104 points across the boundary layer at 3. A potentially attractive method for
computing very thick boundary layers is the double-structured Keller-box scheme first
applied by Smith (1974) to compute steady trailing-edge flows and by Cebeci, Stewartson and Williams (1979) for unsteady flows subject to sudden changes. The Keller-box
readily lends itself to variable-mesh grids, and the properties of the mesh can be chosen
in accord with the known analytic properties of the solution at large times. In this way an
explosive growth in boundary-layer thickness may be computed using only a limited
number of grid points.
Quite apart from its bearing on the asymptotic theory of solutions of the NavierStokes equations, the study of unsteady boundary layers, both laminar and turbulent, is
relevant to a number of important problems in engineering, notably dynamic stall (Carr,
McAlister and McCroskey (1977)). Here an airfoil slowly oscillating through a large
angle, including steady stall, is seen to cast off a large eddy at a crucial point in the cycle.
Closer study shows that just previously the boundary layer develops a large bulge which
may have originated near the leading or the trailing edges, depending on the size of the
radius of curvature of the nose. Clearly the boundary-layer properties are bound up
with the casting of the eddy, and advocates of the singularity might well associate it with
this phenomenon.
Examples of singularities in more complicated boundary layers may be given. Thus
Bodonyi and Stewartson (1977) found a singularity in the boundary layer on a disk in
counterrotation to that of the fluid surrounding it. Also, Banks and Zaturska (1979)
found one for a spinning sphere in a fluid at rest. Again, one occurs with convection
round a heated horizontal cylinder (Simpson and Stewartson (1981)). All these are,
however, associated with the onset of a boundary-layer collision in which two boundary
layers, possibly having different origins but certainly different histories, meet and,
possibly, generate a jet of fluid in a direction perpendicular to the plate. The details of
the evolution of the collision are not yet available, but the indications are that the
nonuniformity is strictly local in effect, unlike the singularity discovered by Howarth
(1938).
8. Three-dimensional boundary layers. The ultimate goal of high-Reynolds
number theories is the asymptotic structure of the flow past general bodies including
three-dimensionality. As already mentioned in this review, some steps toward this end
DALEMBERTS PARADOX
333
have already been taken, notably by Sykes (1980) and Smith (1980), but even the
classical three-dimensional boundary layer evolving under a prescribed external velocity field is still imperfectly understood. The reader may consult Lighthill (1963), Brown
and Stewartson (1969), Eichelbrenner (1973), Smith (1975) for discussions of early
work in this area, which was handicapped by a shortage of accurate numerical solutions
of the governing equations.
The principal problem can be divided into three: How should the computations be
initiated? How to continue them? How are they terminated? Again there is very little
rigorous theory to fall back on, and so we must rely on physical insight, consistency and
numerical experience to build up a core of useful knowledge for these problems.
A typical set of equations is given by (3.5), to which might be added linear and
quadratic functions of u and w for specific problems and also nonnegative scalar
functions of position multiplying each derivative. None of these modifications is of any
fundamental importance, although they may give rise to interesting local phenomena
and complicate the numerical procedure. Thus, broadly, we have to solve a pair of
parabolic equations, linked together, and with an effective arrow of time which varies in
direction across the boundary layer. The boundary conditions may be taken as no-slip
on the surface y 0, although again modifications are possible, and, as y
(8.1)
ble(X Z),
We(X Z),
where b/e, W are given so that p can be regarded as known. For initial conditions we have
a choice depending on the nature of attachment, but if it is at a point, the forward
stagnation point C where Ue We 0, we may conveniently take
(8.2)
there. Since this is the situation for a finite smooth body, we shall adopt (8.2)as
fundamental to our study of (3.5). If attachment is a line other conditions become
appropriate, but we shall not discuss them.
Usually we find that Ue and We are linear functions of x, z near C, and it was shown
by Howarth (1951) that a similarity form of the three-dimensional boundary-layer
equations can then be written down in which u, w are also linear functions of x, z whose
coefficients are functions of y only, satisfying a pair of linked ordinary differential
equations. Solutions of these equations may in turn be found if the attachment point is
nodal; i.e., all the streamlines of the external flow are directed away from the normal to
C but they are not necessarily unique (Davey and Schofield (1967)). There is, however,
a convention about the appropriate choice to be made, not fully tested. If some of the
external streamlines are pointing towards this normal, the attachment is a saddle, and
solutions can still be found provided the normal velocity outside the boundary layer is
directed towards the body (Davey (1961)).
Now that the computations have been initiated, the next question is the method to
be adopted for continuing them. The most widely used methods have been applied
mainly to the study of boundary layers on bodies with a plane of symmetry or
equivalent, on which the solution can be found without reference to the rest of the flow
field. An example is the prolate spheroid at incidence (Wang (1976)), for which the plane
of symmetry passes through the major axis and C. If we take x as the distance along the
major axis and z as the azimuthal angle, then we may also assume that w 0 and Ow/Oz
is finite on the two lines of symmetry z 0, 7r of the spheroid. Then Ow/Oz, u and v may
be found on these lines independently of the rest of the flow field (Wang (1970),
(1974b), Cebeci, Khattab and Stewartson (1980)). An interesting feature of these
334
KEITH STEWARTSON
calculations, carried out for a spheroid of thickness ratio 0.25, is that, for angle of
attachment a < 42 separation occurs towards the rear of the spheroid, whereas for
two-dimensional boundary layers over thin airfoils, separation occurs near the leading
edge except at quite small values of a (Cebeci et al. (1980)). In addition, near the leeside
line of symmetry, where one would expect the fluid in the boundary layer to be moving
towards it, a reversal sets in near the body so that fluid is moving away from this line of
symmetry. This usually occurs near a minimum of the meridional component of skin
friction and further downstream the boundary layer thickens. Also the boundary layer
absorbs fluid from the inviscid external flow near separation, whereas elsewhere and
usually in two-dimensional flow it acts as a source. Above c 42 separation occurs
near the nose on the leeside line of symmetry.
With the solution known on the plane of symmetry we may, either by using a
coordinate system based on the streamlines of the internal flow (Geissler (1975)) or
after appropriate transformation of the body-based coordinates (Cebeci et al. (1981)),
convert the computational problem into that of finding the solution of (3.5) in
0 < x, 0 < z < rr given the properties of u, w on the planes z 0, 7r and x 0. The
three-dimensional region of interest is replaced by a grid of which a typical mesh point is
(xi, Yi, zk), where i,/, k are integers each running from one to a maximum value, with
u, w known at the extreme points of the range of i, ], k and v known at y 1. Then, with the
solutions known at all grid points _-< I and all grid points k < K, we wish to find the
solution at xi, Yi, z: for all/. The Crank-Nicolson method has been applied to this
problem with some success (Wang (1974a), (1974b), (1975)) as has the Keller-box
(Cebeci, Hirsh and Kaups (1976)), but both run into difficulties when the amount of
cross-flow reversal (w < 0) becomes significant. A modification of the box method,
known as the characteristic box, may then be used to continue the computation (Cebeci
et al. (1981)). In this method the term
(8.3)
Uox--+ w-Oz
is replaced by
qiosi
where qi u2+ 142 2 and s is the streamline coordinate in the plane y Yi. The momentum equations of (3.5) are then solved as two-dimensional equations with the direction
of s. varying at each level Yi of y. Since these streamlines must pass through (xi, yi, z),
the place where they intersect the plane x xz-1 must be found by iteration, but the
resulting finite difference system can be solved as far as it would be possible for any such
computation to go.
For there is a limit to the region for which the solution can be computed, given the
boundary and initial conditions, and this is usually referred to as the accessible region
Such a region exists in two-dimensional boundary layers and comes to an end at
separation. For even if the solution is regular at separation there is a region beyond
where u < 0 and the solution here may depend on externally applied conditions further
downstream and not supplied. In three-dimensional boundary layers the accessible
region is partly bounded by the surface Ls, consisting of the normals to the body through
the separation line Is, but may also be partly fixed by another criterion, as we shall see.
There is some controversy about the nature of the separation line l in three
dimensions, Lighthill (1963) arguing that it must be a limiting streamline, i.e., its
tangent must be in the direction of the local skin-friction. While this is assuredly the case
when interactions with the external inviscid stream are taken into account, in the
present context the external velocity field is prescribed and it is possible that the
solution may become singular at ls in a manner analogous to two-dimensional
335
DALEMBERTS PARADOX
boundary layers (Brown (1965)). In that event Is must be an envelope of limiting

streamlines. In either case, Is passes through the separation points on the lines of
symmetry and the solution cannot be advanced beyond
The most extensive studies of the nature of IA, the line of intersection of the
boundary of M and the body, have been carried out by Cebeci et al. (1981 for the prolate
spheroid of thickness ratio 0.25 at an angle a of incidence to the oncoming stream. In
Fig. 11, we display lA for a --6 30 and its characteristic ok (lit. arrow, Turk.) shape,
earlier noted by Wang (1975), may be seen. In integrations from the windward side
180
(a)
90
0-1
+1
180
II/
(b
90
FIG. 11. Graphs.

accessibility line la, separation line ls
vanishing cross-flow skin friction. (a) c 6 (b) c 30
+0.5
external streamlines
line of
0), IA is found to coincide with Is which, when it occurs, is encountered just after the
cross-flow skin friction at y 0 changes sign, and there is strong evidence that it is
accompanied by a singularity of the Brown type. Thus, this part of lA is an envelope of
limiting streamlines. In integration from the leeward side the situation is not so clearcut. For c 6 there is some indication that IA is largely, if not entirely, also coincident
with ls, but now no evidence of the Brown singularity was found even though it is
believed that the computations were carried out to within a distance 0.04 of ls. For
a 30 1A definitely does not coincide with ls over the majority of the lee side, the
reason being that no computation can be carried forward beyond the external streamline through the top of the ok. Further downstream the computation depends on
knowing flow properties on the other side of the windward part of Ls, as Fig. 11 shows,
and this information is not available. The characteristic box enabled computations to be
carried out right up to this particular external streamline and, as we have seen, no
further progress is possible with a prescribed external flow. Thus, although it has been
suggested that open separation can occur in these circumstances, so that ls would be an
envelope of limiting streamline on both sides, the conclusion from the studies just
(z
336
KEITH STEWARTSON
described is that such a possibility is irrelevant when the external velocity is prescribed.
It is not possible to continue the calculation far enough to test the hypothesis.
In order to extend the computations beyond s, and indeed to make them more
relevant to real flows, an interaction theory must be developed as indicated in 3. In
addition to the studies of three-dimensional interactions already noted, special mention
should be made of the generalization by J. H. B. Smith (1977) and F. T. Smith (1978) of
the Sychev-Smith theory of two-dimensional separation to include three-dimensional
effects. They show that, as Lighthill had earlier conjectured, the separation line Is is now
a limiting streamline and the solution is smooth there. Fiddes (1980) has applied the
theory to determine separation on elliptic cones at incidence, obtaining good agreement
with observation (Rainbird, Crabbe and Jurewicz (1963)).
Another interesting feature of three-dimensional boundary layers not found in
3
classical two-dimensional theory is the occurrence of collisions. Their existence was
first revealed in Howarths (1951) study of the boundary layer on a spinning sphere. He
found that two boundary layers are initiated on the axis of rotation, move round the
sphere, collide at the equator and subsequently form a radial jet in the equatorial plane.
Bowden and Lord (1963) confirmed his conclusion in a beautiful experiment. Later,
Merkin (1976) discovered another such phenomenon in his study of convection near a
horizontal heated cylinder (see also Bathelt, Viskanta and Leidenfrost (1979)), and
Stewartson, Cebeci and Chang (1980) found yet another in the entrance region of a
curved pipe. The asymptotic structure of the collision region is incomplete, however.
Smith and Duck (1977) conclude that separation must occur near the equator of the
spinning sphere even though the pressure gradient seems favorable at first sight, and
this conclusion is supported by the experiments of Agrawal, Talbot and Gong (1978),
(1979) on the curved pipe problem. Triple-deck theory suggests that the angular
distance of separation from the equator is O(R -3/14) but it also implies that there is a
relatively large recirculating flow below the main boundary layers whose structure is not
known.
Summary. In this lecture, I have tried to give a flavor of the many important
advances that have been made during the last few years in our understanding of the flow
past bodies at high Reynolds number. Much remains to be done; in particular, the
development of a rational theory for three-dimensional, and possibly also unsteady
two-dimensional flows may be in its infancy. Nevertheless, I believe that firm foundations have been laid down on which future workers may safely build. One of the
encouraging features of the recent work is the considerable number of occasions on
which analytical results in the asymptotic theory of flow past finite bodies are in good
agreement with experimental or numerical results at finite Reynolds number, previously a very rare occurrence.
Two important fundamental questions remain. The first is the technical point of
discovering, within the assumptions of the triple-deck theory, the limiting form of the
structure of a separated region as its dimensions become arbitrarily large. Its resolution
would help with the development of post-triple-deck situations, for example, in the
ramp problem when the angle c* becomes O(1). The second question is to link up the
solution for smooth bodies to that for bodies with sharp trailing edges, especially at
incidence, when the triple-deck solution terminates with the appearance of separation at
the trailing edge. While there are compelling arguments for the correctness of the
They have been observed in two-dimensional flows induced by unsteady oscillations (Davidson and
Riley (1972)).
DALEMBERTS PARADOX
337
Sychev-Smith model, the fact remains that it is disjoint from the solutions for sharptailed bodies, all of which may be obtained by continuous transformations from that for
the finite flat plate at zero incidence. I have full confidence that the latter flows are
correctly known in essence, but until a continuous link is established, a soupcon of
doubt must remain about the solutions for smooth bodies.
The successes described here have been achieved by a combination of observation
of real flows, careful numerical integration of model flows and analytical examination of
the structure of high Reynolds number flows. Each of these methods suffers from
drawbacks. The experimenter is hampered by the difficulty of setting up the desired flow
and of making accurate measurements, and the effects of unsteadiness and especially
turbulence may diminish the value of the observations to the theoretician. The
numerical analyst finds that the computation of the solution of the Navier-Stokes
equations at Reynolds number greater than about 100 becomes very difficult due to the
occurrence of thin regions of rapid shear, sometimes near surfaces which are a priori
unknown, and of extensive wake regions. The structural analyst perforce must set up
formal expansions in descending powers of the Reynolds number R, sometimes as low
as R -1/16, which at first sight would seem to make them relevant only at ridiculously
large values of R. As this discussion has shown, that would be a pessimistic view, and an
asymptotic expansion in integral powers of e is not necessarily of value only when
e
Nobody knows for sure how large e can be until a comparison is made with
solutions at finite e. Failing a rigorous theory, confidence is established by a judicious
combination of these three approaches.
Rigorous, as opposed to rational, arguments have in fact been of little direct help
in the development of our present understanding of these flows, and perhaps this is
how it should be. A fully rational theory in which consistency of assumptions and
consequences is established as far as possible is the first step in the development of a
rigorous theory. It may be regarded as providing the conjectures which, like Goldbachs
problem, need to be proved or disproved. Without a firm appreciation of the structure
of a solution, proof of any of its properties is all the harder. For this reason I do not
.accept views sometimes expressed that nowadays the structural analyst "has his back to
the wall" or that he will not be necessary in a future in which the pure mathematician
and the numerical analyst will unify their activities. If the experience of the last ten years
is any guide, the role of the structural analyst is more important than ever, if only to
guide and check numerical work. Indeed, the challenges before him are more demanding and exciting than at any earlier time. The experimenter needs help to interpret his
data, the numerical analyst needs help to gain confidence in his output. The structural
analyst is uniquely prepared by his education to do these tasks and to provide a deeper
insight into flow properties. In order to carry out his role effectively, he would, I believe,
be well advised always to bear in mind the epigrammatic words of Richard yon Mises
(1958): "Two things are indispensable in any reasoning, in any description we shape of
a segment of reality: to submit to experience and to face the language that is used with
unceasing logical criticism."
<.
Acknowledgment. The author is grateful to S. N. Brown for substantial assistance

during the preparation of this review.
REFERENCES
J. ACKERET, F. FELDMAN AND N. Roast (1947), Investigations o]: compression shocks and boundary
layers in gases moving at high speed, translated in Nat. Adv. Comm. Aeronaut. Tech. Note 1113 from
E. T. H. Zurich, 10 (1946).
338
KEITH STEWARTSON
T. C. ADAMSON, JR. AND A. F. MESSITER (1980), Analysis of two-dimensional interactions between shock
waves and boundary layers, Ann. Rev. Fluid Mech., 12, pp. 103-138.
Y. AGRAWAL, L. TALBOT AND K. GONG (1978), Laser anemometer study offlow development in curved
circular pipes, J. Fluid Mech., 85, pp. 497-518.
W. H. H. BANKS AND M. E. ZATURSKA (1979), Collision of unsteady laminar boundary layers, J. Eng.
Math., 13, pp. 193-212.
M. BAR-LEV ANt) H. T. YANG (1975), Initial ]tow field over an impulsively started circular cylinder, J. Fluid
Mech., 72, pp. 625-647.
A. G. BATHELT, R. VISKANTA AND W. LEIDERFROST (1979), An experimental investigation of natural
convection in the melted region round a heated horizontal cylinder, J. Fluid Mech., 90, pp. 227-239.
D. BECHERT AND E. PFIZENMAIER (1975), Optical compensation measurements on the unsteady exit
condition at a nozzle discharge area, J. Fluid Mech., 71, pp. 123-144.
G. BIRKHOFF AND E. H. ZARANTONELLO (1957), Jets, Wakes and Cavities, Academic Press, New York.
R. J. BODONYI AND A. KLUWIK (1979), Freely interacting transonic boundary layers, Phys. Fluids, 20, pp.
1432-1437.
R. J. BODONYI
AND K. STEWARTSON (1977), The unsteady boundary layer on a rotating disk in a

counter-rotating fluid, J. Fluid Mech., 79, pp. 669-688.
F. B. BOWDEN AND P. G. LORD (1963), Aerodynamic resistance to a sphere rotating at high speed, Proc. Roy.
Soc. (London) Ser. A, 271, pp. 143-153.
S. BRODETSK (1923), Discontinuous fluid motion past circular and elliptic cylinders, Proc. Roy. Soc.
(London) Ser. A, 102, pp. 542-553.
S. N. BROWN (1965), Singularities associated with separating boundary layers, Phil. Trans. Roy. Soc.
(London) Ser. A, 257, pp. 409-444.

S. N. BROWN AND K. STEWARTSON (1969), Laminar separation, Ann. Rev. Fluid Mech., 1, pp. 45-72.
(1970), Trailing-edge stall, J. Fluid Mech., 42, pp. 561-584.
(1975a), A non-uniqueness of the hypersonic boundary layer, Quart. J. Mech. Appl. Math., 28,
pp. 75-90.
S. N. BROWN, K. STEWARTSON AND P. G. WILLIAMS (1975a), Hypersonic self-induced separation, Phys.
Fluids, 18 (1975), pp. 633-639.
S. N. BROWN AND K. STEWARTSON (1975b), Wake curvature and the Kutta condition, Aeronautical
Quarterly, 27, pp. 275-280.
S. N. BROWN AND P. G. DANIELS (1975), On the viscous ]tow about the trailing edge of a rapidly oscillating
plate, J. Fluid Mech., 67, pp. 743-761.
S. N. BROWN, K. STEWARTSON AND P. G. WILLIAMS (1975b), On expansive free interactions in boundary
layers, Proc. Roy. Soc. (Edin.), 74A, pp. 271-283.
S. N. BROWN AND H. K. CHENG (1981), Correlated unsteady and steady laminar trailing-edge ]tows, in J.
Fluid Mech., to appear.
J. BUCKMASTER (1970), The behaviour of a laminar compressible boundary layer on a cold wall near a point of
zero skin friction, J. Fluid Mech., 44, pp. 237-247.
O. R. BURGGRAF (1973), Inviscid reattachment of a separated shear layer, Proc. 3rd Int. Conf. on Numerical
Methods in Fluid Mech. Lecture notes in Physics 19, Springer-Verlag, Berlin, pp. 39-47.
O. R. BURGGRAF AND C. E. JOBE (1974), The numerical solution of the asymptotic equations of trailing edge
flow, Proc. Roy. Soc. (London) Ser. A, 340, pp. 91-111.
O. R. BURGGRAF (1975), Asymptotic theory of separation and reattachment of a laminar boundary layer on a
compression ramp, AGARD-CP-168 Flow separation paper No. 10.
O. R. BURGGRAF, D. RIZZETTA, M. J. WERLE AND V. N. VATSA (1979), Effect of Reynolds number on
laminar separation of a supersonic stream, AIAA J., 17, pp. 336-344.
L. W. CARR, K. W. MCALISTER AND W. J. MCCROSKEY (1977), Analysis of the development of dynamic
stall based on oscillating airfoil experiments, Report TN D-8382, National Aeronautics and Space
Administration.
J. E. CARTER (1972), Numerical solutions of the Navier-Stokes equations for the supersonic laminar flow
over a two dimensional compression corner, Report TR-R-385, National Aeronautics and Space
Administration.
D. CATHERALL, K. STEWARTSON AND P. G. WILLIAMS (1965), Viscous flow past a fiat plate with uniform
infection, Proc. Roy. Soc. (London) Ser. A, 284, pp. 270-296.
D. CATHERALL AND K. W. MANGLER (1966), The integration of the two-dimensional laminar boundarylayer equations past the point of vanishing skin friction, J. Fluid Mech., 26, pp. 163-182.
T. CEBECI, R. S. HIRSH AND K. KAUPS (1976), Calculation of three dimensional boundary layers on bodies of
revolution at incidence, Rept. MDC J7643, Douglas Aircraft Company, Long Beach, CA.
DALEMBERTS PARADOX
339
T. CEBECI (1978), An unsteady boundary layer with separation and reattachment, AIAA J., 16, pp.
1305-1306.
(1979), The laminar boundary layer on a circular cylinder started impulsively from rest, J. Comp. Phys.,
31, pp. 153-172.
T. CEBECI, H. B. KELLER AND P. G. WILLIAMS (1979), Separating boundary-layer flow calculations, J.
Comp. Phys., 31, pp. 363-378.
T. CEBECI, K. STEWARTSON AND P. G. WILLIAMS (1979), On the response of a stagnation boundary layer to
a change in extremal velocity, SIAM J. Appl. Math., 36, pp. 190-199.
T. CEBECI, A. K. KHATTAB AND K. STEWARTSON (1980), On nose separation, J. Fluid Mech., 97, pp.
435-454.
T. CEBECI, K. STEWARTSON AND P. G. WILLIAMS (1980), Separation and reattachment near the leading
edge of a thin airfoil at incidence, presented at AGARD Symposium Computation of ViscousInviscid Interacting Flows, Colorado Springs, CO.
T. CEBECI, A. K. KHATTAB AND K. STEWARTSON (1981), Three dimensional boundary layers and the ok of
accessibility, J. Fluid Mech., to appear.
D. R. CHAPMAN (1950), Laminar mixing o" a compressible fluid, Rep. 958, Nat. Adv. Comm. Aeronaut.,
Washington DC.
D.R. CHAPMAN, D. M. KUEHN AND H. K. LARSON, Investigation of separated flows in supersonic and
subsonic streams with emphasis on the effect of transition, Rep. 1356, Nat. Adv. Comm. Aeronaut.,
Washington DC.
R. CHOW AND R. E. MELNIK (1976), Numerical solutions of the triple-deck equation for laminar trailing-edge
stall, Proc. 5th Int. Conf. Num. Meth. in Fluid Dynamics, Lecture notes in Physics 59, Springer,
New York, pp. 135-144.
J. O. COLE AND J. AROESTY (1968), The blowhard problemminviscid flows with surface infection, Int. J.
Heat Mass Transfer, 11, pp. 1167-1183.
O. G. CRIGHTON (1972), Radiation properties of the semi-infinite vortex sheet. Proc. R. Soc. (London) Ser. A,
330, pp. 185-198.
S. N. CURLE (1980), private communication.
J. LE ROND DALEMBERT (1752), Essai dune nouvelle theorie de la resistance des fluides.
P. G. DANIELS (1974a), Numerical and asymptotic solutions for the supersonic flow near the trailing edge of a
flat plate, Quart. J. Mech. Appl. Math., 27, pp. 175-191.
(1974b), Numerical and asymptotic solutions ]:or the supersonic flow near the trailing edge of a flat plate
at incidence, J. Fluid Mech., 63, pp. 641-656.
(1978), On the unsteady Kutta condition, Quart. J. Mech. Appl. Math., 31, pp. 49-75.
(1979), Laminar boundary layer reattachment in supersonic flow, J. Fluid Mech., 90, pp. 289-303.
(1980), Laminar boundary-layer reattachment in supersonic flow, Part II, J. Fluid Mech., 97, pp.
129-144.
A. DAVEY AND D. SCHOFIELD (1967), Three-dimensional flow near a two-dimensional stagnation point, J.
Fluid Mech., 28, pp. 149-151.
A. DAVEY (1961), Boundary-layer at a saddle point of attachment, J. Fluid Mech., 10, pp. 593-610.
T. DAVIES AND G. WALKER (1977), On the solution of the compressible laminar boundary-layer equations
and their behaviour near separation, J. Fluid Mech., 80, pp. 279-292.
B. J. DAVIDSON AND N. RILEY (1972), Jet induced by oscillatory motion, J. Fluid Mech., 53, pp. 287-303.
S. C. R. DENNIS (1972), The motion of a viscous fluid past an impulsively started semi-infinite flat plate, J. Inst.
Math. Applics., 10, pp. 105-117.
S. C. R. DENNIS AND F. T. SMITH (1980), Steady flow through a channel with a symmetrical constriction in the
form o]" a step, Proc. Roy. Soc. (London) Ser. A, 372, pp. 393-414.
D. DIJKSTRA (1978), Separating incompressible laminar boundary-layer flow over a smooth step of small
height, Proc. 6th Intern. Conf. Num. Meth. Fluid Dynamics, Tbilisi, pp. 45-52.
C. DIVER (1979), Ph.D. Thesis, London University.
C. DIVER AND K. STEWARTSON (1978), On moderate injection into a separated supersonic boundary layer,
with reattachment, J. Fluid Mech., 88, pp. 115-132.
R. DUGAS (1950), Histoire de la mecanique, Griffon, Paris, pp. 283-287.
E. A. EICHELBRENNER (1973), Three-dimensional boundary layers, Ann. Rev. Fluid Mech., 45, pp.
339-360.
S. P. FIDDES (1980), A vortex sheet model for the separated flow past an elliptic cone, Rep. 38382,
Aeronautical Research Council, London.
B. FORNBERG (1980) A numerical study of steady viscous flow past a circular cylinder, J. Fluid Mech. 98, pp.
819-855.
340
KEITH STEWARTSON
W. GEISSLEl (1975), Calculation o[ the three-dimensional laminar boundary layer around bodies ol revolution
at incidence and with separation, AGARD-CP-168.
C. GOLDBACH (1742), letter to L. Euler, quoted in L. E. Dickson, Theory of Numbers, Vol. 1, Chelsea, New
York, 1952, p. 421.
S. GOt.DSTEIrq (1930), Concerning some solutions o]" boundary-layer equations in hydrodynamics, Proc.
Camb. Phil. Soc., 26, pp. 1-30.
S. GOt.DSTEIN (1948), On laminar boundary layer ]tow near a point o]" separation, Quart. J. Mech. Appl.
Math., 1, pp. 43-69.
J-P. GUIRAUD AND R. SCHMITT (1975), Laminar separation at a trailing edge, AGARD-CP-168, paper
No. 3.
H. HELMHOLTZ (1968), Uber diskontinuierliche Flussigkeitsbewegungen Monatsber, Berlin Akad., pp.
215-228.
L. HOWARTH (1951), The boundary layer in three-dimensional flow, Part I: Derivation o[ the equations o[[low
along a general curved sureace, Phil. Mag., 42, pp. 239-243.
L. HOWARTH (1938), On the solution ot the laminar boundary equations, Proc. Roy. Soc. (London) Ser. A,
1964, pp. 547-579.
L. HOWArTH (1951), Note on the boundary layer on a spinning sphere, Phil. Mag., 42, pp. 1308-1315.
(1953), Modern Developments in Fluid Dynamics: High Speed Flow, Vol. 1, Oxford University Press,
London, p. 50.
M. Y. HUSSAINI, B. S. BALDWIN AND R. W. MACCORMACK (1980), Asymptotic ]eatures o[ shock-wave

boundary-layer interactions, AIAA J. 18, pp. 1014--1016.
G. KnCHHOFF (1869), Zur theorie Fliissigkeitsstrahlen, J. Reine Angew Math., 70, pp. 289-298.
L. G. LEAr. (1973), Steady separated flow in a linearly decelerated ]ree stream, J. Fluid Mech., 59,
pp. 513-535.
B. L. REEVES (1964), Supersonic separated and reattaching laminar Jtows: L General theory
and applications to adiabatic boundary layer/shock wave interactions, AIAA J., 2, pp. 1907-1920.
J. E. LEWIS, T. KUBOTA AND L. LEES (1968), Experimental investigation o]: supersonic laminar twodimensional boundary layer separation in a compression corner with and without cooling, AIAA J. 6,
pp. 7-14.
M. J. LIGHTHILL (1953), On boundary layers and upstream in]tuence II. Supersonic [low without separation,
Proc. Roy. Soc. (London) Ser. A, 217, pp. 478-507.
(1963), Introduction; boundary layer theory, Laminar Boundary Layers, Chapter II, L. Rosenhead,
ed., Oxford University Press, London.
R. E. MELNIK AND R. CHOW (1975), Asymptotic theory o" two-dimensional trailing edge ]tows, NASA
Conference on Aerodynamic Analysis Requiring Advanced Computers, Report NASA SP-437,
National Aeronautics and Space Administration, pp. 177-249.
J. H. MERKIN (1976), Free convection boundary layer on an isothermal horizontal cylinder, ASME-AICLE,
Heat transfer Conferences, St. Louis.
A. F. MESSTEl (1970), Boundary layer[low near the trailing edge o] a jqat plate, SIAM. J. Appl. Math. 18,
pp. 241-257.
A. F. MESSITEI, A. FEO AND R. E. MELNIK (1971), Shock-wave strength ]or separation o]" a laminar
boundary layer at transonic speeds, AIAA J., 9, pp. 1197-1198.
A. F. MESSTER, G. R. HOUGH AND A. FEO (1973), Base pressure in laminar supersonic flow, J. Fluid
Mech., 60, pp. 605-624.
A. F. MESSITER AND A. LINAN (1976), The vertical plate in laminar tree convection: effects o]: leading and
trailing edges and discontinuous temperature, Z. Angew. Math. Phys., 27, pp. 633-651.
A. F. MESSTER (1978), Boundary-layer separation, Proc. 8th U.S. Natl. Congr. Appl. Mech., pp. 157-179.
R. YON MISES (1958), The Mathematical Theory of Compressible Flow, p.v. Academic Press, New York.
E. A. MOt.t.ER (1953), Dissertation, Universitiit G6ttingen.
M. NAPOt.ITANO, M. J. WERIE AND R. T. DAVIES (1979), Numerical techniques ]:or the triple-deck problem,
AIAA J., 17, pp. 699-705.
L. LEES
AND
M. NAPOLITANO (1980), Numerical study ol strong slot injection into a supersonic laminar boundary layer,
AIAA J., 18, pp. 72-77.
M. NArOLTANO, M. J. WERLE AND R. T. DAVIES (1978), Numerical solutions o) the triple-deck equations
/:or supersonic and subsonic [tow past a hump, AFL Rep. 78-6-42, Dept. of Aerospace Eng. Univ.
Cincinnati, Cincinnati, Ohio.
V. YA NEILAND (1969), Towards a theory o] separation o] the laminar boundary layer in a supersonic stream.
Izv. Akad. Nauk SSSR Mekh. Zhidk. Gaza, 4, pp. 33-35.
DALEMBERTS PARADOX
341
(1970), Asymptotic theory of plane steady supersonic ]tows with separation zones, Izv. Akad. Nauk.
SSSR Mekh. Zhid Gaza, 3, pp. 22-32.
(1970), Asymptotic theory ofplane steady supersonic flows with separation zones, Izv. AN SSSR Mekh.
Zhid. Gaza, 3, pp. 22-32.
(1971 ), Flow beyond the separation point of the boundary layer in a supersonic stream. Izv. Akad. Nauk.
SSSR, Mekh. Zhidk. Gaza, 3, pp. 19-25.
(1974), Asymptotic problems in the theory of supersonic viscous flows, Central Institute of Aero- and
Hydro-dynamics, Moscow, No. 1529, pp. 1-124 (In Russian).
K. NICKEL (1958), Einige Eigenschaften yon Losungen der Prandtlschen Grenzschicht-Differentialgleichungen, Arch Rational Mech. Anal., 2, pp. 1-31.
(1973), Prandtls boundary-layer theory from the viewpoint of a mathematician, Ann. Rev. Fluid
Mech., 5, pp. 405-428.
O. A. OLINIK (1963), On a system of equations in boundary layer theory, Zhurn. Vychislit. Mat. Mat. Fiz, 3,
pp. 489-507.
S. A. ORSZAG AND S. C. CROW (1970), Instability of a vortex sheet leaving a semi-infinite plate, Stud. Appl.
Math., 49, pp. 167-181.
K. OSWATITSCH AND K. WIEGHARDT (1946), Theoretical investigations on steady potential flows and
boundary layers at high speed, Rep. 10378, Aeronaut. Res. Council, London.
L. PRANDTL (1905), Ueber Flussigkeitsbewegung mit kleiner Reibung, Verhandlung des drittes internationalen
Mathematiker-Kongresses, Heidelberg (Leipzig), pp. 484-491.
J. PRETSCH (1944), Die laminate Grenzschichte bei starken absaugen und ausblasen. Unter Mitt. Deut.
Luftfahrt Rep. 3091
I. PROUDMAN AND K. JOHNSON (1962), Boundary-layer growth near a rear stagnation point, J. Fluid Mech.,
12, pp. 161-168.
W. J. RAINBIRD, R. S. CRABBE AND L. S. JUREWICZ (1963), A water tunnel investigation offlow separation
about circular cones at incidence, Aeronautical Report LR 385, National Research Council of
Canada.
S. A. RAGAB AND A. H. NAYFEH (1980), Second-order asymptotic solution for laminar separation, Physics
Fluids, 23, pp. 1101-1110.
T. A. REYHNER AND I. FLOGGE-LOTZ (1968), The interaction of a shock wave with a laminar boundary
layer, Internat. J. Non-Linear Mech., 3, pp. 173-199.
S. W. RIENSTRA (1979), Edge influence on the response of shear layers to acoustic forcing, Doctoral thesis,
Technische H6geschool, Eindhoven, the Netherlands.
N. RILEY AND K. STEWARTSON (1969), Trailing edge flows, J. Fluid Mech., 39, pp. 193-207.
N. RILEY (1975), Unsteady laminar boundary layers, this Review, 17, pp. 274-297.
D. P. RIZZETTA, O. R. BURGGRAF AND R. JENSON (1978), Triple-deck solutions for visCous supersonic and
hypersonic flow past corners, J. Fluid Mech., 89, pp. 535-552.
A. J. ROBINS AND J. A. HOWARTH (1972), Boundary-layer development at a two-dimensional rear
stagnation point, J. Fluid Mech., 56, pp. 161-171.
W. SCHNEIDER (1974), Upstream propagation of unsteady disturbances in supersonic boundary layers, J. Fluid
Mech., 63, pp. 465-485.
C. J. SIMPSON AND K. STEWARTSON (1981), On a singularity initiating a boundary layer collision, submitted
to Quart. J. Mech. App. Math.
W. R. SEARS AND D. P. TELIONIS (1975), Boundary layer separation in unsteady flow, SIAM J. Appl. Math.,
28, pp. 215-235.
S. F. SHEN (1978), Unsteady separation according to the boundary layer equation, Adv. in Appl. Mech., 18,
pp. 177-220.
C. J. SIMPSON AND K. STEWARTSON (1981), in preparation.

F. T. SMITH (1973), Laminar flow over a small hump on a flat plate, J. Fluid Mech., 57, pp. 803-824.
(1974), Boundary layer flow near a discontinuity in wall boundary conditions, J. Inst. Maths. Applics.,
13, pp. 127-143.
(1976), Flow through constricted or dilated pipes and channels, Parts and II, Quart. J. Mech. Appl.
Math., 29, pp. 343-364, 365-376.
F. T. SMITH AND P. W. DUCK (1977), Separation of jets or thermal boundary layers from a wall, Quart. J.
Mech. Appl. Math., 30, pp. 143-156.
F. T. SMITH (1977a), The laminar separation of an incompressible fluid streaming past a smooth surface, Proc.
Roy. Soc. (London) Ser. A, 356, pp. 433-463.
(1977b), Upstream interactions in channel flows, J. Fluid Mech., 79, pp. 631-655.
F. T. SMITH AND P. W. DUCK (1977), Separation offets on thermal boundary layers from walls, Quart. J.
342
KEITH STEWARTSON
F. T. SMITH, R. I. SYKES AND P. W. M. BRIGHTON (1977), A two-dimensional boundary layer encountering

a three-dimensional hump, J. Fluid Mech., 83, pp. 163-176.
F. T. SMITH (1978), Three-dimensional viscous and inviscid separation of a vortex sheet from a smooth
non-slender body, R.A.E. Technical Rep. 78095.
(1978), Flow through symmetrically constricted tubes, J. Inst. Math. Appl., 21, pp. 145-156.
(1979a), Theory of laminar streaming flows, Lecture Course on Asymptotic Methods in Mechanics,
Oct. 1979, CISM Udine.
(1979b), Private communication.
(1979c), Laminar flow of an incompressible fluid past a bluff body the separation, reattachment, eddy
properties and drag, J. Fluid. Mech., 92, pp. 171-205.
(1979d), The separating flow through a severely constricted symmetric tube, J. Fluid Mech., 90,
pp. 725-754.
(1979e), On the non-parallelflow stability of the Blasius boundary layer, Proc. Roy. Soc. (London) Ser.
A, 366, pp. 91-109.
F. T. SMITH AND P. W. DUCK (1980), On severe constricting, curving or cornering of channel flows, J. Fluid
Mech., 98, pp. 727-753.
F. T. SMITH (1980), Non-linear stability of boundary layers for disturbances of various sizes, Proc. Roy. Soc.
(London) Ser. A, to appear.
(1980), A three dimensional boundary-layer separation, J. Fluid Mech., to appear.
J. H. n. SMITH (1977), Behaviour of a vortex sheet separating from a smooth surface, R.A.E. Technical
Rep. 77058.
(1975), A review of separation in steady, three-dimensional flow, Paper 31, AGARD-CP-168.
K. STEWARTSON (1958), On Goldsteins theory of laminar separation, Quart. J. Mech. Appl. Math., 11,
pp. 399-410.
K. STEWARTSON (1968), On the flow near the trailing edge of a flat plate, Proc. Roy. Soc. (London) Ser. A,
306, pp. 275-290.
K. STEWARTSON (1969), On the flow near the trailing edge of a flat plate II, Mathematika, 16, pp. 106-121.
K. STEWARTSON AND P. G. WILLIAMS (1969), Self-induced separation, Proc. Roy. Soc. (London) Ser. A,
312, pp. 181-206.
(1970a), On supersonic laminary boundary layers near convex corners, Proc. Roy. Soc. (London) Ser.
A, 319, pp. 289-305.
(1970b), Is the singularity at separation removable? J. Fluid Mech., 44, pp. 347-364.
K. STEWARTSON AND V. G. WILLIAMS (1973), Self-induced separation II, Mathematika, 20, pp. 98-108.
K. STEWARTSON (1974), Multistructurat boundary layers on flat plates and related bodies, Adv. Appl. Mech.,
14, pp. 145-239.
(1980), High Reynolds-number flows, in Approximation Methods for Navier-Stokes Problems,
IUTAM Symposium, Paderborn, 1979.
K. STEWArTSON, T. CE3ECI AND K. CHANG (1980), A boundary-layer collision in a curved pipe, Quart. J.
V. V. SYCHEV (1967), On steady laminar flow of a fluid around a bluff body at large Reynolds number, 8th
Symposium on Advanced Problems and Methods in Fluid Mechanics, Tarda, Poland.
V. V. SYCHEV (1972), Concerning laminar separation, Izv. Akad. Nauk. SSSR Mekh. Zhidk. Gaza, 3, pp.
47-59.
R. I. SYI<ES (1980), On three-dimensional boundary layer flow over surface irregularities, Proc. Roy. Soc.
(London) Ser. A, submitted.
R. TEMAM (1977), Navier-$tokes Eq.ttations Theory and Numerical Analysis, North-Holland, Amsterdam.
R. M. TERRILL (1962), Laminar boundary-layer 1:low near separation with and without suction, Phil. Trans.
Roy. Soc. (London) Ser. A, 253, pp. 55-100.
K-M. Tu AND S. WEINBAUM (1976), A non-asymptotic triple-deck model for supersonic boundary-layer
interactions, AIAA J., 14, pp. 767-775.
A. J. VAN DE VOOREN AND D. DIJKSTRA (1970), The Navier-$tokes solution for laminar flow past a
semi-infinite plate, J. Eng. Math., 4, pp. 9-27.
V. N. VATSA (1975), Ph.D. dissertation, Univ. of Cincinnati, Cincinnati, Ohio.
A. E. P. VELDMAN AND A. I. VAN DE VOOREN (1974), Drag of a finite plate, Proc. 4th Int. Conf. on
Numerical Methods in Fluid Dynamics, Lecture notes in Physics 35, Springer-Verlag, Berlin,
pp. 422-430.
A. E. P. VEIDMAN (1976), Higher order boundary layer theory near the trailing edge of a flat plate, Rep.
TW-157, Math. Inst., Univ. of Groningen.
(1979), A calculation method for incompressible boundary layers with strong viscous-inviscid interaction, N.L.R. MP 79029U, Amsterdam.
DALEMBERTS PARADOX
343
K. C. WANG (1970), Three-dimensional boundary la yer near the plane of symmetry of a spheroid at incidence, J.
Fluid Mech., 43, pp. 187-209.
(1974a), Boundary layer over a blunt body at high incidence with an open-type ofseparation, Proc. Roy.
Soc. (London) Set. A, 340, pp. 33-55.
(1974b), Laminar boundary layer near the symmetry plane of a prolate spheroid, AIAA Journal, 12, pp.
949-95.
(1975), Boundary layer over a blunt body at low incidence with circumferential reversed flow, J. Fluid
Mech., 72, pp. 49-65.
(1976), Separation of three-dimensional viscous flow, in Viscous Flow Symposium, Lockheed Georgia
Company, Atlanta, GA.
(1979), Unsteady boundary layer separation, Martin Marietta, Baltimore Rep. MML TR 79-16c.
M. J. WERLE, D. L. DWOYER AND W. L. HANKEY (1973), Initial conditions for the hypersonicshock interaction problem, AIAA J., 11, pp. 525-530.
M. J. WERLE AND V. N. VATSA (1974), A new metlwd for supersonic boundary layer separation, AIAA J.,
12, pp. 1491-1497.
M. J. WERLE (1979), Supersonic laminar boundary layer separation by slot injection, AIAA J., 17,160-167.
J. C. WILLIAMS, III AND W. D. JOHNSON (1974), Semi-similar solutions to unsteady boundary layer flows
including separation, AIAA J., 12, pp. 1388-93.
P. G. WILLIAMS (1975), A reverse-flow computation in the theory of self-induced separation, Proc. 4th Int.
Conf. Numerical Methods in Fluid Dyn., Boulder, Colorado, pp. 445-451.
(1979), Private communication.

DAlemberts Paradox

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DAlemberts Paradox

Uploaded by

Copyright:

Available Formats

Downloaded 03/13/13 to 129.8.242.67. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.

(C) 1981 Society for Industrial and Applied Mathematics

1. Introduction. Conjectures are a vital part of the fabric of pure mathematics.

viscous fluid in steady motion, may be written as

, is the kinematic viscosity. It is generally believed by fluid dynamicists, virtually

in the boundary layer. The Navier-Stokes equations reduce to

10p 021g o___p

expansion, of which the leading terms are

of the triple-deck region near separation on a flat plate.

where A is the value of r at the origin of the triple-deck coordinates, according to

The corresponding boundary conditions are, for an impermeable wall,

3. The supersonic triple-deck. We consider the case

first, and so (2.4a)

which corresponds to an undisturbed boundary layer in the triple-deck region. For

from its flat-plate value at x

glimmerings of a theory to explain the multishocks found by Ackeret, Feldman and

where tr is a nondimensional parameter of order t- 1/4 The general features of the

with boundary conditions

02A/O 2 dse drt

u-yA(x)+ax asy-co forx>0,

A -ax 1/2 if x>0,

_ceR-1/16 pUoo2 (x* x ,)1/2 + const,

and the reduced pressure inside is

p=-- f;I [(x_j)z+(y_n)2],/2.

k,, +2k,k, =0,

and hence p 0, A 0.8920x /3.

u(x) 1/3 k(x), k(O)=k (0)=0, k-18x--O asx

triple-deck exceeds the length of the plate.

However, in view of the relation connecting p* and

in classical boundary layer

when x* is large and negative.

with Fn(0)= F,(1/2)= 0, for an ordered set of y, of which 31

5.175 is the smallest, and

Hence, formally, for a severe constriction with h

FIG. 10. Eddy length ahead of a step as a function

f..o (Y- y:Z) dY

[-180 Ai(0)] 3/7 -+- 5.19,

x2, Pu -2{25(Xz-X)} -4. It is reasonable to identify this value of x with the

straightforward calculation of the unsteady boundary layer comes to an end at

boundary layers (Brown (1965)). In that event Is must be an envelope of limiting

FIG. 11. Graphs.

Acknowledgment. The author is grateful to S. N. Brown for substantial assistance

AND K. STEWARTSON (1977), The unsteady boundary layer on a rotating disk in a

(London) Ser. A, 257, pp. 409-444.

M. Y. HUSSAINI, B. S. BALDWIN AND R. W. MACCORMACK (1980), Asymptotic ]eatures o[ shock-wave

C. J. SIMPSON AND K. STEWARTSON (1981), in preparation.

F. T. SMITH, R. I. SYKES AND P. W. M. BRIGHTON (1977), A two-dimensional boundary layer encountering

You might also like