Notes on Triple Deck.

P.-Y. Lagr ee CNRS & UPMC Univ Paris 06, UMR 7190, Institut Jean Le Rond Alembert, Bo te 162, F-75005 Paris, France pyl@ccr.jussieu.fr ; www.lmm.jussieu.fr/lagree October 14, 2013

Abstract We present here briey the famous triple deck theory. In this framework, boundary layer separation is possible without singularity. Some numerical experiments describing the ow over bumps or wedges without or with ow reversal in various asymptotic r egimes are presented.

Introduction.

Let us present a summing up of the preceding chapters. We presented what happens in a wall layer when a shear ow is disturbed, we next saw the perturbation of a Poiseuille ow. We observed in that case that perturbations can exist in the core ow (the Main Deck), the perturbations are expressed as a perturbation of the stream lines trough a function A. We then presented the Blasius boundary layer, we emphasized the inuence of the displacement thickness and the retroaction with the ideal uid in the framework of Interacting Boundary Layer Theory.

2
2.1

Triple Deck
Overview

In the ftys Lighthill [4] and Landau among a lot of others began to understand that boundary layer separation will be explained by new scales and a strong displacement of the boundary layer. This occurring at a small longitudinal scale, but larger than the boundary layer itself. Then, simultaneously Neiland [6], Messiter [5] and Stewartson [13] in 69 proposed a new asymptotic structure in three decks (gure 1). First, there is the basic boundary layer, which is now the Main Deck. this layer is disturbed near the wall where the velocity is the smaller, the

Triple Deck

length of this layer is small. Perturbations in this lower layer called Lower Deck are transmitted through the Main Deck. In this layer the perturbation acts as a displacement of the stream lines (with a function called A(x)). This deexion of the stream lines is transmitted to the ideal uid layer : the Upper Deck. This deexion creates a disturbance of pressure, and this disturbance of pressure will be transmitted back in the lower deck promoting the velocity disturbances. So that we will deal with a coupled system of equation: a disturbance of pressure creates a disturbance of stream lines which in turn creates a disturbance of pressure.
(STEWARTSON K. & WILLIAMS P.G.1969 "Self induced separation", Proc Roy. Soc A 312, 181- 206.)

1 3 ! R -3/8

"

-4/8

!"

-5/8 R

-3/8

La synthse se fait en partant de la solution de Blasius et en la perturbant l'chelle !3 dgage. Comme on est une chelle plus courte que celle du dveloppement de la couche limite, le profil de Blasius (not U0) ne varie pas en x l'chelle considre. On a pour le frottement parital (en y=0): dU0/dy=# !-4. et #=.3321 on crit le dveloppement suivant: u=U0+!u1+...; v=!2v1+... p = !2p1

Figure 1: Left, the triple deck scales. Right, triple decker ship of the line from

HMS victory brochure Porthmouth (vaisseau de ligne ` a trois ponts). In german Dreierdeck-Theorie, a french translation of Triple Deck Theory may be Triple Pont instead of Triple Couche. En substituant dans les ENS: on en dduit que la solution est une perturbation non visqueuse
des quations, et qu'il n'y a pas de variation transverse de pression: d$ % u1= A(x)U0', et v1= U et p =0 dx 0 %y 1

-dA/dx (x en ! 3.....) Retenons que la solution dans le "pont principal" ("Main Deck ") est une perturbation non visqueuse de la solution de Blasius.

2.2

Scales

2.2.1

Main Deck

1.2.2.

Prs de la paroi on constate que la dveloppement de la solution de pont principale donne: u=#y + ! A(x)#

pont infrieur The classical way to look at Triple Deck is to consider perturbations of the Boundary Layer. The rst idea to introduce is the existence of a perturbation of small length compared to the boundary layer development itself. 3

12 novembre 2004 "3DEATC"

We have the basic non dimensional Blasius prole y ) in the boundary UB ( layer, where y is the transverse variable scaled by L/ Re. Now suppose that at longitudinal scale say x3 there is a perturbation of this basic prole. We will call Main Deck the region considered which is of relative scale x3 but which is of boundary layer scale in the transverse direction. As this scale is small, the boundary layer as not evolved, and at rst order VB = 0. So, suppose that at longitudinal scale say x3 there is a perturbation of this basic prole of magnitude , then: u = UB ( y ) + u 1 - IV . 2-

Triple Deck

In order to retain all the terms in the incompressibility and in the total derivative equation, u = UB ( y ) + u 1 , v = x3 Re v 1

u 1 longitudinal equation of momentum (UB 1 UB ), is of order /x3 . The x + v previous analysis show that the relevant pressure term is in 2 /x3 which is negligible as are the viscous terms. This small value of pressure may be considered here as a rst hypothesis that we will verify after. The system to solve is then

u 1 v 1 u 1 + = 0, (UB +v 1 UB ) = 0, x y x

p 1 = 0. y

1 2 (v By elimination we nd: UB y UB ) = 0, the classical notation is then to introduce a function of x say A(x) introduced as a constant of integration, such as u 1 = A(x)UB ( y ) and then v 1 = A (x)UB ( y)

is solution of the system. With this description, the velocity is not zero but A(x)UB (0) on the wall, so we have to introduce a new layer to full t the no slip condition.

2.2.2

L'quilibre diffusif/ convectif doit tre respect pour assurer l'adhrence la paroi: Lower Deck !u !2 u!x ~ R-1 !y 2 u.

prs de la paroi: "u

~ u ~ #3/#.

Near the wall the velocity prole is linear, the order of magnitude of the variation of sur velocity must be the same than the basic ow in order to obtain produit une chelle rapide x 3, separation. "u"u/x3~"u/(#3/#)2.
Ce qui s'crit avec les ordres de grandeur prcdents et compte tenu du fait que l'accident se Cette expression fournit l'ordre de grandeur de l'chelle rapide en fonction du rapport des couches: The purpose of the lower deck is to 3introduce a layer in which this perx3~(#3/#) =$3. turbation On ofconstate velocity will be annihilated. So the scale velocity is , then facilement ensuite que la pression est en $2, on admet (dans of cette analyse rapide as the velocity of le the boundary layer isy linear near the wallauit is natural to mais on peut montrer) qu'elle ne varie pas en et quelle encore inchange travers du est "Pont Deck . Cette de #3 de la guess that thePrincipal" lower Main deck will by perturbation of size L/ Re . couche limite produit une dflexion des lignes de courant #3=$#. L'angle de dflexion correspondant est donc: The behavior of the velocity in the 3 Main Deck is . $#/$

Figure 2:

u = UB ( y )le+ Aparfait (x)Ucomme y ). B ( Cette perturbation est alors ressentie par fluide une bosse de longueur $3 et d'paisseur $#. Le fluide parfait linaris rtroagit donc avec $3 comme chelles transverses et longitudinales ("Pont Suprieur" Upper Deck ) cette bosse d'angle $#/$3. La perturbation - IV . 3- l'angle de la bosse en $#/$3. Or l'ordre de pression de fluide parfait est donc proportionnelle de grandeur de la pression compatible dans le Pont Infrieur est $2, donc pour qu'il y ait
rtroaction, il faut que ces deux pression soient gales: $2=$#/$ 3 Ce qui donne le paramtre magique: $=#-1/4=R-1/8.

1.2.

synthse: dveloppements asymptotiques

Triple Deck

We look at it near the wall. For y 0 the Blasius prole is linear near the wall UB ( y ) UB (0) y and then the velocity is UB (0) y + A(x)UB (0), written in the inner variables of the lower deck this is (as y = y ) (y + A(x))UB (0). So we deduce that in the lower deck the velocity should match to this quantity: lim u = (y + A(x))UB (0)
y +

The convective diusive equilibrium of the Navier Stokes equations u x Re1 2 y 2

written with the longitudinal x3 and transversal Re1/2 scales reads: 1 Re1 x3 (Re 1/2 )2 so that the longitudinal scale is : x3 = 3 . The pressure is of order 2 , and the transverse equations of momentum gives as in the classical boundary layer: p =0 y so, the pressure does not depend on y and is constant across the lower deck. The nal system is then: u v u u dp 2 u + = 0, u +v = + . x y x y dx y 2 With no slip condition at the wall (u = v = 0), the entrance velocity prole u(x , y ) = UB (0)y , and the matching condition with the Main Deck: u(x, y ) = (y + A)UB (0). Note, that the system is parabolic, there is no output condition needed to solve it.

2.3

Upper Deck

The disturbed velocity in the Main Deck is : u = UB ( y ) + A(x)UB ( y ); v = 2 1 A (x)UB ( y)

Re

- IV . 4-

Triple Deck

and for the pressure p =0 y Now let us see what happens at the top of the Main Deck, for y : u = 1; v = 2 Re 1 A (x),

there is no more longitudinal perturbation of the velocity at order , but there is a transverse velocity, a kind of blowing velocity at the edge of the Main Deck. Note that the pressure remains the same order 2 . Therefore we look at a layer of longitudinal size x3 = 3 and of same 1 thickness in which we have a blowing velocity at the wall of order Re and 2 2 a pressure of order . To have a consistent problem both should be equal so that we obtain the nal magic parameter: = Re1/8

3
3.1

The various r egimes

Upper Deck, coupling relation incompressible

The velocity at the top of the Main Deck is then the velocity at the bottom of the upper deck: A . Depending on the ideal uid r egime, one may compute the pressure. For a incompressible ow on has the Hilbert relation: p= 1
dA dx

One see that there in the equations one can remove U0 in the equations in changing the scales, say u multiplied by U , p is multiplied by P , x by X and y by Y , so by invariance (ux u versus x p) P = U 2 , and (ux u versus 2 u) gives U = X/Y 2 . At innity u U y + ... so that X = U Y 3 . The y 0 0 pressure displacement relation tells that the pressure is proportional to A , so U 2 = Y /X which is (X/Y 2 )2 = Y /X or X 3 = Y 5 . But remember that X = U0 Y 3 , then X = U0 5/4 so, nally x3 = (U0 )5/4 Re3/8 , 3 = (U0 )3/4 Re5/8 u3 = (U0 )1/4 Re1/8 , v3 = (U0 )3/4 Re1/4 , = (U0 )1/2 Re1/4 so that the nal system is independent of the base ow.

- IV . 5-

Triple Deck

3.2

Upper Deck, coupling relation supersonic

For a compressible supersonic ow, one has the Ackeret formula: p = M dA 2 M 1 dx

again, changing the scales, one can remove the U0 : x3 = C 3/8 (U0 )5/4 (M 2 1)3/8 Re3/8 , ... then the relation is p = A .

3.3

Upper Deck, coupling relation transcritical

In transcritical ows a new parameter K = (M 2 1)(U0 )2/5 C 1/5 Re1/5 and 3 A )2/3 p = ( 2 +1 see Bodoniy, Bartels & Rothmayer, Bodonyi and Kluwick [2].

3.4

Upper Deck, coupling relation sub/supercritical

In the case of water ow, the ideal uid response was for the pressure the disturbance divided by F 1. So by rescaling: x = (x /L1)UB (0)5 |F r 1|3 /(Re3/8 ), y = (y /L)UB (0)2 |F r 1|1 /(Re5/8 ), 2 ))U (0)2 |F r 1|2 /(Re2/8 ), p = (p /(U0 B p = A for subcritical ows (F < 1) and p = A for supercritical ows (F > 1).

3.5

Jet Flow

Nearly the same conguration may exist for a wall jet of thickness = Re1/2 near the wall l: 1 2 u 2u u 2 2 ( 2) x3 x Re y
so that x3 = 3 and u = U0 + A(x)U0 and v = x A (x)U0 3

U0

v p 2 p1 2 = , so the scale is (A ( x))U0 , 2 x y y x3

so that = Re1/7 which gives x3 = Re3/7 and = Re9/14

p( x, 0) = A ( x)
0

2 U0 (y )dy

- IV . 6-

Triple Deck

3.6

The various r egimes, canonical system

The canonical system is: u u dp 2 u u v + = 0, u +v = + . x y x y dx y 2 With at the wall (u = v = 0), at the entrance u(x , y ) = y , and at the innity u(x, y ) = (y + A). p=
1
dA

xdx d incompressible case

p = A supersonic p = A hypersonic case p = A uvial. p = A torrential. p = A mixed convection. A = 0 pipes, Couette. p = A pipes, wall jets.

3.7

Linearised solution: self induced solution

We can look at a linearised solution of the canonical system. The linearised system is: u1 v1 u1 dp1 2 u1 + = 0, y + v1 = + . x y x dx y 2 With at the wall (u1 = v1 = 0), at the entrance u1 (x , y ) = 0, and at the innity u1 (x, y ) = A1 . We test eKx solutions on the linearized system, with K > 0. u1 = eKx (y ), v1 = eKx (y ), p1 = eKx P

with (0) = (0) = 0 and say () = 1 so that A1 = eKx ; as the incompressibility is fullled, the momentum is Ky (y ) (y ) = KP + so
2 (y ) y 2

2 (y ) , y 2

(1)

= Ky (y ), and as (0) = KP , so is K 2/3 Ai(K 1/3 y )P/Ai (0)

K 1/3 P Ai (0) y 0

and =

Ai( )d so that we deduce () =

K 1/3 3Ai (0) P

- IV . 7-

Triple Deck

The supersonic case allows then an eigen solution K = (3Ai (0))3/4 with K = 0.827 This exponential is the rational explanation of the observed self induced separation. The supercitical case and the hypersonic case and the mixed convection case allow then an eigen solution K = (3Ai (0))3 with K = 0.47 The jet case (or pipe) allows then an eigen solution K = (3Ai (0))3/7 with K = 0.89 The incompressible, the uvial and the couette or symmetrical pipe cases do not allow this self induced solution.

3.8

The Prandtl transform

There is a trick called Prandtl tranform which allows to change the bumpy wall in a at one. One writes y = y f (x) and keeps x = x. Then, as x = x f ( x ) and = 0 + continuity equation becomes y y y u+ (v f u) = 0 x y and as the total derivative: u u + v u = u u + (v f u) u x y x y

so the Prandtl transform is: y = y f (x) , x = x, u = u and v = (v f u) so that system is invariant: u v x + y = 0, (2) u dp 2 u u u +v = + . x y dx y 2 u = v = 0 on y = 0, u y when x , and u y +f (x) when y . The sole dierence lies in the boundary condition at the top. - IV . 8-

Triple Deck

3.9

Linearised solution explicit solution in Fourier space

We can look at a linearised solution of the canonical system. The linearised system is in Prandtl transform: u1 v1 u1 dp1 2 u1 + = 0, y + v1 = + . x y x dx y 2 With at the wall (u1 = v1 = 0), at the entrance u1 (x , y ) = 0, and at the innity u1 (x, y ) = A1 + f1 . We are looking for solutions in Fourier space so we test eikx solutions on the linearized system. u1 = eikx (y ), v1 = eikx (y ), p1 = eikx Pk f1 = fk eikx , A1 = ak eikx

with (0) = (0) = 0 and then () = ak + fk so that; as the incompressibility is fullled, the momentum is iky (y ) (y ) = ikPk + so
2 (y ) y 2

2 (y ) , y 2

(3)

= iky (y ), and as (0) = ikPk , so is (ik )2/3 Ai((ik )1/3 y )Pk /Ai (0)
(ik)1/3 P Ai (0) y 0

and =

Ai( )d so that we deduce () = ak + fk =

(ik)1/3 3Ai (0) P

The relation between the perturbation of pressure and the displacement is then in Fourier space: F T [p] = F T [(A + f )] where = (3Ai (0))1 (ik )1/3 . The supersonic case pf = 1/(ik ) The subsonic case the pressure displacement relation is pf F T [p] = F T [(A)] with pf = 1/ |k | The supercitical case is such that pf = 1 The uvial case is such that pf = 1 The no displacement case pf = 0. The no displacement case pf = 0. To plot the gures 3.10 one uses then the following relations: F T [p] = (ik )2/3 F T [f ] and F T [ ] = Ai(0)F T [p]. pf Ai (0) (4)

and the linearized perturbation of the skin friction ( ) - IV . 9-

Triple Deck

3.10
3 2.5 2

Plots of linearised solutions

Hilbert 3 f(x) (x) p(x) f(x) (x) p(x) subsonic 2.5 2

1.5

1.5

0.5

0.5

-0.5

-0.5

-1 -4 -2 0 x 2 4

-1 -4 -2 0 x 2 4

supersonic 3 f(x) (x) p(x) 3

p=-A f(x) (x) p(x)

2.5

2.5

1.5

1.5

0.5

0.5

-0.5

-0.5

-1 -4 -2 0 x 2 4

-1 -4 -2 0 x 2 4

A=0 3 f(x) (x) p(x) 3

p=A f(x) (x) p(x)

2.5

2.5

1.5

1.5

0.5

0.5

-0.5

-0.5

-1 -4 -2 0 x 2 4

-1 -4 -2 0 x 2 4

Figure 3: Friction distribution and pressure over a bump in 6 cases, linear solution. Top left the Hilbert case, just to compare. Top right the subsonic
1 case p = xdx d . Middle left, the supersonic p = A case. Middle right, p = A case. Bottom left, the A = 0 case. Bottom right, the p = A case.
dA

- IV . 10-

Triple Deck

3.11

Supersonic case

Supersonic case p = dA dx , ow over a wedge.

4 =0.5 =1.5 =2.5 =3.5 wedge

3.5

2.5

p(x)

1.5

0.5

0 -10

-5

0 x

10

Figure 4: pressure distribution over a wedge

1.5

=0.5 =1.5 =2.5 =3.5 wedge

(x)

0.5

-0.5

-10

-5

0 x

10

Figure 5: Friction distribution over a wedge

- IV . 11-

Triple Deck

3.12
p=
1

Subsonic case
dA

xdx d subsonic/ incompressible case with the Hilbert integral.

1 =1.0 =2.0 =3.0 =4.0 bump

0.5

p(x)

-0.5

-1

-1.5

-2 -6 -4 -2 0 x 2 4 6

Figure 6: pressure distribution over a bump

=1.0 =2.0 =3.0 =4.0 bump

(x)

-6

-4

-2

0 x

Figure 7: Friction distribution over a bump

- IV . 12-

Triple Deck

3.13

Transsonic case
3 p = ( A )2/3 2 +1

Figure 8: Friction distribution pressure distribution over a expansive wedge

- IV . 13-

Triple Deck

3.14

Various scales on a bump

Smith et al. [12] and Roget et al. [7] showed that a bump which is at the triple deck size is at a kind of interaction of several eects.

B H NS H=1/X H=X
non lin

NS 5/3 H=X
non lin

C
d ! !2 ! ! ! u = " p + 2 u; u+ v u u +v dx !y !x !y !x !y u(x , 0) = v( x , 0) = 0 u ( x , y " > # )" > y + A. p =" 1 $

& +xdf" %

/ dx

d%

TC

p="

1 $

&

" dA / dx + df / dx d% x "%

lin

lin

X
d ! ! ! ! u +v ! u = " p + 2 u u+ ; v u dx !x !y !x !y !y u(x , 0) = v( x , 0) = 0 u ( x , y " > # )" > y + f ( x ).
2

! !2 ! ! d ! u+ p + 2 u; v u u+v u= " !y !x dx !x !y !y u( x ,0) = v( x , 0) = 0 u( x , y" > #) " > y + A . dA dx + df / dx p="1 " / d% x "% $

&

Figure 9: various scales converging in the Triple Deck We saw that the triple deck corresponds to a very special size for a bump. So we may ask what happens if we change the size of the bump. Imagine that the bump is no more of size O(1) in the Triple Deck scale but is smaller or larger. How changes teh longitudinal scale if we want to have always the maximum number of terms in the equations. Let say that y is changed by Y y , so, in the triple deck if we change y Y y then in order to have the non linear viscous balance, we have: u Y u, x Y 3 , p Y 2 p and A Y A. With this transformation lower deck equations are the same. Let use H the size of the bump as the parameter of height. H is not Y , and f Hf The ideal uid relation (using Prandtl transform): p= 1 A f d x

- IV . 14-

17 mai 2006 "3DEATC"

20

1/3 H=X

! !2 ! ! d ! u=" u+ v u u +v p + 2 u; !y !x dx !x !y !y u(x , 0) = v( x , 0) = 0 u ( x , y " > # )" > A .

Triple Deck

has the following rescaling: (Y 2 )p = (Y 2 ) 1 A 1 d + (HY 3 ) x f d x

If now Y is large (this is a large bump X = Y 3 ), and if H is large, the displacement contribution decreases. for Y 2 = HY 3 i.e. H = Y 5 or H = X 5/3 . The largest size of the bump is the boundary layer itself, so it gives a maximum size of Re3/10 . If now Y is small (small bump X = Y 3 ), and if H is large, the displacement contribution decreases. (Y 4 )p = 1 A 1 d + (HY 1 ) x f d x

so H = Y and we have A + f = 0. this is the no displacement case. there is a more subtle case, as the matching relation is u(x, ) = y + A + f , then if H is large, we may imagine that A is large (change A HA) and as y Y y (with Y << H ), the lower Deck is broken in two parts one where u goes from 0 to A and another one where u = A + f (u is of order H >> Y ) so if we change A HA, x Xx,u Hu and p H 2 p. The ideal uid relation (using Prandtl transform): p= has the following rescaling: (H 2 )p = (H/X ) 1 A 1 d + (H/X ) x f d x 1 A f d x

so that H = 1/X . We then have to solve: A A 1 = x x A f d x

- IV . 15-

Triple Deck

Link with IBL

The IBL formulation emphasizes on the displacement thickness, 1 = (Re1/2 )

0

(1 u(x, y ))dy

we have to decompose it into two parts as we cross the lower and the main decks. Let introduce Y 1 = (Re1/2 )(
0 Y

(1 u ( x, y ))dy +
Y

(1 u ( x, y ))dy )

the rst integral is estimated near the wall, so the Lower Deck description ( y = y ) is valid there, but a good idea is to write the velocity u(x, y ) = UB (0)(y + A) + uc where uc is a correction:
Y / Y / Y

(
0

(1 u ( x, y ))dy ) = (
0

(1 (UB (0)(y + A)))dy

0

uc dy )

the second one is in the Main Deck (1 u(x, y ))dy =

Y Y

(1 UB ( y ) A(x)UB ( y ))dy .

Re summing the two integrals and changing the order of the terms allows then write: 1 = (Re1/2 ){[(
0 / Y / Y

(1 (UB (0)(y )))dy +

(1 UB ( y ))dy ]+
/ Y

+[(
0

((UB (0)(A)))dy +

(A(x)UB ( y ))dy ] 2
0

uc dy )}

so that we recognise : 1 = (Re1/2 ){

0 / Y

(1 UB ( y ))dy +
0

(A(x)UB ( y ))dy 2
0

uc dy )}.

or 1 = (Re1/2 ){
0

(1 UB ( y ))dy A(x) O(2 )}.

the A contribution of the triple deck is the perturbation of the displace ment thickness 0 (1 UB ( y ))dy . So the IBL technique based on 1 is justied by the triple deck analysis.

Alembert Paradox and Kutta condition

At this point, it is time to introduce the famous dAlembert Paradox. The triple deck structure is a possible response to solve it. The Kutta condition does not exist, the ow can turn round the trailing edge, but this is at a Re3/8 scale, it is so small when Re that this gives the Kutta condition. - IV . 16-

Triple Deck

Conclusion

In this chapter we presented the Triple Deck scales and equations. We showed that there is an interactive problem between a thin layer near the wall and a layer of ideal uid through the displacement of the stream lines A. In the thin layer, Prandtl equations are valid with new scales, and a dierent matching condition involving this displacement function A. The upper layer Euler small disturbance theory applies, the layer in between is the boundary layer which is passive and only transmits the perturbations of A and pressure p. This framework allows to understand boundary layer separation and self induced separation. The pressure deviation relation pressure p displacement A allows a large variety of various coupled problems...

References
[1] Cebeci T. & Cousteix J. (1999): Modeling and computation of boundary layer ows, Springer Verlag. [2] Bodonyi and Kluwick, (1977): Freely interacting transonic boundary layers. Phys. Fluids 20 (1977), pp. 1432143 [3] Gajjar J. & Smith F.T. (1983): On hypersonic self induced separation, hydraulic jumps and boundary layer with algebraic growth, Mathematika, 30, pp. 77-93. [4] M.J. Lighthill (1953) : On boundary-layer and upstream inuence : II. Supersonic ows without separation. Proc. R. Soc., Ser. A 217 :478507 [5] A.F. Messiter (1970): Boundarylayer ow near the trailing edge of a at plate. SIAM J. Appl. Math., 18 :241257, [6] Neiland V. Ya (1969): Propagation of perturbation upstream with interaction between a hypersonic ow and a boundary layer, Mekh. Zhid. Gaz., Vol. 4, pp. 53-57. [7] C. Roget, J. Ph. Brazier , J. Cousteix and J. Mauss (1998): A contribution to the physical analysis of separated ows past three-dimensional humps European Journal of Mechanics - B/Fluids Volume 17, Issue 3, May-June 1998, Pages 307-329 [8] Ruban A.I. & Timoshin S.N. (1986): Propagation of perturbations in the boundary layer on the walls of a at channel, MZG 2, pp. 74-79. [9] Saintlos S. & Mauss J. (1996): asymptotic modelling for separating boundary layers in a channel, Int. J. Engng. Sci.,Vol 34, No 2, pp201211.

- IV . 17-

Triple Deck

[10] Smith F. T. (1976): Flow through constricted or dilated pipes and channels, part 1 and 2, Q. J. Mech. Appl. Math. vol 29, pp 343- 364 & 365- 376. [11] Smith F. T.(1982): On the high Reynolds number theory of laminar ows, IMA J. Appl. Math. 28 207-81. [12] Smith FT, Brighton PW M., Jackson PS & Hunt JCR On boundary layer ow past wo dimensional obstacles, JFM 110, pp 1-37 (1981). [13] Stewartson K. & Williams P.G. (1969): Self - induced separation,Proc. Roy. Soc. London, A 312, pp 181-206. [14] Stewartson K. (1974): Multistructured boundary layer on at plates and related bodies Advances Appl. Mech. 14 p145-239. [15] Sychev V. V. , Ruban A. I. , Sychev V. V. & Korolev G. L. (1998): Asymptotic theory of separated ows, Cambridge University Press. [16] http://cis.jhu.edu/tilak/blsep.html The web page of this text is:
http://www.lmm.jussieu.fr/lagree/COURS/CISM/

The last version of this le is on:

http://www.lmm.jussieu.fr/lagree/COURS/CISM/TriplePont CISM.pdf

This le consist in notes for the CISM Course ASYMPTOTIC METHODS IN FLUID MECHANICS: SURVEY AND RECENT ADVANCES, Advanced School coordinated by Herbert Steinr uck Udine, September 21 25, 2009. P .-Y . L agr ee s contribution is:
http://www.lmm.jussieu.fr/ lagree/COURS/CISM/IVIIBL CISM.pdf

- IV . 18-