NONLINER THEORY OF UNSTABLE FLUID MIXING DRIVEN BY SHOCK WAVE

Qiang Zhang and Sung-Ik Sohn Department of Applied Mathematics and Statistics SUNY at Stony Brook Stony Brook, NY 11794-3600 Corresponding author: Qiang Zhang Telephone: 516-632-7567 Fax: 516-632-8490 Email: zhang@ams.sunysb.edu

ABSTRACT

A shock driven material interface between two uids of different density is unstable. This instability is known as Richtmyer-Meshkov (RM) instability. In this paper, we present a quantitative nonlinear theory of compressible Richtmyer-Meshkov instability in two dimensions. Our nonlinear theory contains no free parameter and provides analytical predictions for the overall growth rate, as well as the growth rates of the bubble and spike, from early to later times for uids of all density ratios. The theory also includes a general formulation of perturbative nonlinear solutions for incompressible uids (evaluated explicitly through the fourth order). Our theory shows that the RM unstable system goes through a transition from a compressible and linear one at early times to a nonlinear and incompressible one at later times. Our theoretical predictions are in excellent agreements with the results of full numerical simulations from linear to nonlinear regimes. PACS numbers: 47.20.Ma, 47.20.Ky

-21. Introduction When an incident shock collides with an interface between two materials of different densities, the material interface becomes unstable. This phenomena is known as Richtmyer-Meshkov instability and plays an important role in studies of supernova and inertial connement fusion. This interfacial instability was theoretically predicted by Richtmyer [1]. The rst experiment of this instability was performed by Meshkov [2]. Since then several experiments [3-4] and numerical simulations [5-12] on the growth of the RM unstable interfaces have been performed. Several theories have been developed by different approaches [1,13-18]. For review of current researches of Richtmyer-Meshkov instability, see [19]. Further references can be found in the proceedings of PCTM Workshops [20] and references therein. Most of previous theoretical work focused on the growth rate in linear regime. However, the growth of the RM unstable interface is nonlinear [8] and, for a long time, theories failed to give a quantitatively correct prediction for the growth rate of RM unstable interface in the nonlinear regime. Previous theoretical predictions were about twice as large as the experimental data on air-SF 6 . In this paper, we present nonlinear theory for Richtmyer-Meshkov instability for compressible uids in the case when the reected wave is a shock. The development of the unstable interface is the formation and growth of ngers, called spike and bubble. A spike is a portion of heavy uid penetrating into light uid, and a bubble is a portion of light uid penetrating into heavy uid. The ngers are the dominant structures of the RM unstable interface. The overall growth rate of RM unstable interface is dened as the growth rate of the half of the overall size of the mixing zone between the two uids. In the case of incompressible ow with innite density ratio, Hecht. et. al [17] followed Layzers approach [21] to develop a model and obtained an analytic expression for the asymptotic bubble growth rate. Our theory presented in this paper provides quantitative predictions for the overall growth rate of the unstable interface and the growth rates of spike and bubble in compressible uids of all density ratio in the case of a reected shock from early to later times. The agreements among our theoretical predictions, the results of full nonlinear numerical simulations and experimental data are remarkable. Richtmyer proposed an impulsive model to simplify the dynamics of the RM unstable interface [1]. The model approximates the incident shock as an impulse (a delta function) and the post-shocked uid as incompressible. The linear growth rate v imp of the ngers predicted by the impulsive model [1] is v imp = uAka 0 (1)

-3where u is the difference between the shocked and unshocked mean interface velocities, A = ( )/( + ) is the Atwood number and and are the post-shocked uid densities. The incident shock propagates from uid of density to uid of density . a 0 is the postshocked perturbation amplitude at the interface. Equation (1) is a widely used theoretical model for RM instability. For weak shocks, the prediction of (1) agrees quite well with the asymptotic growth rate from the linear theory for compressible uids. A systematic comparison between (1) and the solution of the linearized Euler equations showed signicant disagreement in certain physical parameter domains [16]. Even when the prediction of the impulsive model agrees with the result of the linear theory, it agrees in the regime where the nonlinearity is important and the linear theory is no longer valid. Most of the experimental and numerical simulations are conducted from the linear regime to the nonlinear regime. This is the time range which will be considered in this paper. We present two main results in this paper: (i) a general formulation of perturbative nonlinear solutions for interfacial mixing between incompressible uids (evaluated explicitly for the impulsive model through the fourth order); (ii) a quantitative nonlinear theory of compressible RM instability from early to later times. The goal of our theoretical study is (ii), which is more important. The results of (i) play essential roles in (ii). Now, we explain how the results of (i) is related to (ii). In order to develop an approximate nonlinear theory for compressible RM unstable interface, we adopt the physical picture that the dominant effects of the compressibility occur near the shocks. This inuences the material interface when the shocks are in the vicinity of the material interface, namely at early times. We assume that the initial disturbance at the interface is small. Then, at early times the compressibility is important and the nonlinearity is less important. As time evolves, the magnitude of the disturbance at the material interface increases signicantly and the transmitted shock and reected wave move away from the interface. The effects of compressibility are reduced and the nonlinearity starts to play a dominant role in the interfacial dynamics. From this physical picture, we see that at early times the dynamics of the system are mainly governed by the linearized Euler equations for compressible uids, while at later times the dynamics are mainly governed by the nonlinear equations for incompressible uids. The RM unstable system goes through a transition from a linear and compressible one at early times to a nonlinear and incompressible one at later times. Our approach is to qualitatively separate the dynamics of the RM instability into two stages corresponding to early and later times. We determine an approximate solution in each stage. Then we match the early time solution and the later

-4time solution to obtain an analytical expression which changes gradually from one to the other. The matched solutions give quantitative predictions for the overall growth rate and the growth rates of spike and bubble at compressible RM unstable interface from early to later times. The solution of the linearized Euler equations in the case of a reected shock can be found in [1] and [16]. Richtmyer developed the linear theory for the case of reected shock [1]. One of the authors and his coworker extended the linear theory to the case of a reected rarefaction wave [16]. In Section 2, we derive the nonlinear perturbation solutions for incompressible uids. These perturbation solutions serve as a generating series for Pade approximations. In Section 3, we apply Pade approximations and develop a nonlinear theory of RM instability for compressible uid. In Section 4, we present validation studies of the Pade approximations, our nonlinear theory, and the physical picture on which our theory is based. In Section 5, we present the quantitative predictions of the nonlinear theory for several cases. In Section 6, we give a summary and conclusion for our theoretical study.

2. Nonlinear Perturbation Solutions for Incompressible Fluids As we have described in the previous section, at later times, the effects of compressibility are less important. Therefore we can approximate the uids as incompressible. In this section we derive the nonlinear perturbation solutions for incompressible uids. The expansion is in terms of the disturbance at the initial interface. In Section 2.A, we derive perturbation solutions with general initial growth rate. A general formula is obtained for n-th order solutions. In Section 2.B, we demonstrate our solution method by evaluating the nonlinear solution of the impulsive model. The impulsive model has specic initial conditions for the growth rate. Explicit expressions for the growth rate of the unstable material interface is given through fourth order. Furthermore, the symmetry properties of the solutions of all orders have been determined analytically.

2.A. Governing Equations and Solution Procedure Here we derive the nonlinear solutions for incompressible systems with no external forces and with general velocity distribution along the initial interface. The governing equations for inviscid, irrotational, incompressible uids with no external forces are given by 2 (x,z,t) = 0 in material 1, 2 (x,z,t) = 0 in material 2, (2)

-5 =0 + z x x t =0 + z x x t at z = , at z = ,

(3) (4) (5)

2 2 1 1 ) ] = 0 at z = . ) +( + [( )2 + ( )2 ] [( + z x 2 z x 2 t t

Here the unprimed variables are the physical quantities in material 1 and the primed variables are the physical quantities in material 2. z = (x,t) is the position of the material interface at time t. and are densities of material 1 and 2, respectively. and are the velocity potentials in
= in material 2. Two equations given in (2) come from the incompressibility conditions. v

material 1 and 2, respectively. The velocity eld is given by = in material 1 and by v

Equations (3) and (4) represent the kinematic boundary condition that a uid particle initially situated at the material interface will remain at the interface afterwards. Equation (5) represents the dynamic boundary condition in which the pressure is continuous across the material interface. We consider the single mode RM instability only in this paper. The initial shape of the material interface is given by (x,t = 0) = a 0 cos(kx) and the initial velocity distribution along the material . . interface is given by (x,t = 0). Here a 0 is the amplitude of the initial disturbance. (x, 0) is an arbitrary single valued function of x. The impulsive model corresponds to a particular initial velocity distribution along the interface. This particular initial velocity will be derived in next section from the assumption of an impulsive force. We expand all quantities in terms of powers of a 0 k, i. e. f = fn . Here fn = n , n and n , are proportional to (a 0 k)n . Then (2)-(5) can be expressed as

n =1

2 (n) = 0 in material 1,
n =1

2 (n) = 0 in material 2, n =1
at z =

(6)

n (i) (n i) (n) (n) )=0 + z x t i =1 x

(n) , n =1 (n) , n =1

(7)

n =1

n (i) (n i) (n) (n) )=0 + z x t i =1 x

at z =

(8)

(n) (i) (n i) 1 n (i) (n i) (n) ] + + [ + z z x 2 n =1i =1 x n =1 t n =1 t

(i) (n i) 1 n (i) (n i) ]=0 + [ z z x 2 n =1i =1 x

at z =

(n) . n =1

(9)

-6-

Since the boundary conditions given in (7)-(9) hold at the position z = =

(n) , they n =1

need to be further expanded at z = 0. After expanding the equations and collecting all terms of order (a 0 k)n , we have the following equations for the n th order quantities. 2 (n) = 0 in material 1, 2 (n) = 0 in material 2, at z = 0, at z = 0, at z = 0, (10) (11)

(n) (n) = S (n) (t)cos(ikx) + i z t 0in (n) (n) = S (n) (t)cos(ikx) + i z t 0in (n) (n) = T (n) (t)cos(ikx) + i t t 0in

(12)

(13)

with the initial conditions (n) (x,t = 0) = a 0 cos(kx)1n , . (n) (x,t = 0) = (14) (15)

ai 0in

. (n)

(0)cos(ikx).

. (n) Here 1n is Kronecker delta function. a i (0) is determined by the Fourier mode decomposition of the left hand side of (15). S (n) , S (n) and T (n) are determined by the Fourier mode decomposii i i tion of the right hand sides of the following equations:
i S (n) (t)cos(ikx) = sum1 p ! 0in

1 p +1 (a) (b) p (ni ) 1 p +1 (a) p (ni ) + p x i =1 z p +1 i =1 sum2 p ! xz

at z = 0, (16) at z = 0, (17)

0in

S (n) (t)cos(ikx) = i

1 p +1 (a) (b) p (ni ) 1 p +1 (a) p (ni ) + p x i =1 z p +1 i =1 sum2 p ! xz sum1 p !

i T (n) (t)cos(ikx) = sum1 p ! 0in

p 1 p +1 (n ) ((a) (a) ) i p i =1 z t

p +1 (a) q+1 (b) p+1 (a) q+1 (b) 1 1 p!q! [( z p x z q x + z p+1 z q+1 ) 2 sum3 p +1 (a) q +1 (b) p (ni ) q (m j ) p+1 (a) q +1 (b) )] at z = 0 + i =1 j =1 z q +1 z p +1 z q x z p x (18)

where

-7sum1 : (0 < n 1 , n 2 , ...,np , p, a < n), (n 1 + n 2 + ... + np + a = n); sum2 : (0 < n 1 , n 2 , ...,np , a, b < n),(0 p < n), (n 1 + n 2 + ... + np + a + b = n); sum3 : (0 < n 1 , n 2 , ..., np , m 1 , m 2 , ...,mq , a, b <n),(0 p, q < n), (n 1 + n 2 + ... + np + m 1 + m 2 + ... + mq + a + b = n). Note that the quantities on the right hand side of (16)-(18) are already known for the n th order equations since they have orders lower than n . Equations (10)-(13) are linear equations for the n th order variables (n) , (n) and (n) which we are going to solve for. The n th order solution can be expressed as (n) (x,t) = (n) (x,z,t) = (n) (x,z,t) =
i a (n) (t)cos(ikx), 0in i b (n) (t)cos(ikx)e ikz , 0in i b (n) (t)cos(ikx)e ikz . 0in

(19) (20) (21)

After substituting (19)-(21) into (11)-(13) and using the orthogonality of different Fourier modes, we have b (n) (t) i 1 da i (t) S (n) (t)], = [ i dt ik 1 da i (t) S (n) (t)], [ i dt ik
(n) (n)

(22)

b (n) (t) = i

(23)

db (n) (t) db (n) (t) i i = T (n) (t) + i dt dt

(24)

. (n) with the initial amplitudes a (n) (0) = a 0 1n 1i and the initial growth rate a i (0). The solutions for i (22)-(24) with the initial conditions are given by a (n) (t) i 1 [ik(t t )T (n) (t ) + (S (n) (t ) S (n) (0)) + (S (n) (t ) S (n) (t ))]dt = i i i i i + 0 . (n) + a i (0)t + a 0 1n 1i , 1 . (n) 1 1 [T (n) (t )dt + (S (n) (t) S (n) (t))] + a i (0), = i i i ik ik + 0 1 . (n) 1 1 [T (n) (t )dt + (S (n) (t) S (n) (t))] a i (0) = i i i ik ik + 0
t t t

(25)

b (n) (t) i

(26)

(n) i (t)

(27)

-8for i 0. For the case i = 0, a (n) (t) = 0 from the condition of incompressibility. From (11) and 0 (12), it follows that S (n) (t) = S (n) (t) = 0. From (13), b (n) (t) and b (n) (t) are determined by 0 0 0 0
t

b (n) (t)+ b (n) (t) = b (n) (0)+ b (n) (0) + T (n) (t )dt . 0 0 0 0 0
0

Since the velocities are the gradients of the velocity potentials and all the source terms in (16)(17) involve differentiation with respect to x or/and z, the functional forms of b (n) and b (n) are 0 0 irrelevant. Therefore we will not evaluate them explicitly. Let us summarize the recursive procedure for obtaining the nonlinear solution. We progress from lower orders towards higher orders, starting from the rst order. We rst evaluate the source terms, namely the right hand sides of (16)-(18). These source terms are known from the lower order solutions. We determine S (n) , S (n) and T (n) from these source terms. Then the n th i i i order solutions are simply given by (19)-(21) with a (n) , b (n) and b (n) given by (25), (26) and i i i (27), respectively. It is easy to show that a (n) , b (n) and b (n) are polynomials of t. With respect i i i to t, the degree of a (n) is not greater than n and the degrees of b (n) and b (n) are not greater than i i i n 1.

2.B. Nonlinear Solution of Impulsive Model Let us apply the solution procedure to determine the nonlinear solution of the impulsive model through fourth order. The impulsive model approximates that the uids are at rest at t<0. An impulsive force acts on the uids at time t = 0 and results an impulsive acceleration g = u(t). Here u represents the strength of the impulse. Then (5) becomes ()g 2 2 1 1 ) ] = 0 at z = .(28) ) +( + [( )2 + ( )2 ] [( + z x 2 z x 2 t t

After integrating (28) over time from 0 to 0+ we obtain the following initial condition ()u + = 0 at z = = a 0 cos(kx) and t = 0+. (29)

For t 0+, (28) reduces to (5). Therefore our general solution procedure is valid for the impulsive model. The only thing we need to do is to determine the initial growth rate from (29). We further expand (29) at z = 0. The result is ()u(1) 1n (n) + (n) =
0 in

R (n) cos(ikx) i

at z = 0 and t = 0.

(30)

-9Here R (n) are determined by the Fourier mode decomposition of the right hand sides of the foli lowing equation
i R (n) cos(ikx) = m 1 m ! 0in = n 1

m ((n m) (n m) )[a 0 cos(kx)]m m z

(31)

at z = 0 and t = 0. From (20) and (21), (31) can be expressed as ( )ua 0 1n 1i b (n) (0) + b (n) (0) = R (n) . i i i Finally from (22),(23) and (32), we obtain the initial growth rate for the impulsive model ikR (n) + S (n) (0) + S (n) (0) . (n) i i i . a i (0) = a 0 1n 1i + + (33) (32)

Here = Auk for the impulsive model. Therefore the nonlinear solutions to the impulsive model are given by (19)-(21) and (25)-(27) with the initial growth rate given by (33). The explicit expressions through the fourth order are given by (1) = (1 + t)a 0 cos(kx), (2) = 1 Aka 2 2 t 2 cos(2kx), 0 2 1 2 3 k a 0 [(4A 2 + 1)2 t 3 + 3t 2 + 6t]cos(kx) 24 1 2 3 k a 0 [(4A 2 1)2 t 3 3t 2 ]cos(3kx), 8 1 3 4 k a 0 [4A 3 3 t 4 + 6At 2 + At]cos(2kx) 12 1 3 4 k a 0 [(8A 3 4A)3 t 4 8A2 t 3 + 3At 2 ]cos(4kx). 12 (37) (36) (34) (35)

(3) = + (4) = +

The rst and second order solutions shown above are identical to the ones obtained previously by Richtmyer [1] and Haan [14], respectively. The derivations for (34)-(37) are given in the Appendices A, B and C. Now, we examine the symmetry properties of n th order quantities. From (18) and (31), T (n) and R (n) can be expressed as i i T (n) = F (n) F (n) , i i i R (n) = G (n) G (n) . i i i The denitions of F (n) , F (n) , G (n) and G (n) are obvious from (18) and (31). i i i i (38) (39)

- 10 The n-th order source terms have the following symmetry S (n) (A) = (1)n +1 S (n) (A), i i F (n) (A) = (1)n F (n) (A), i i G (n) (A) = (1)n G (n) (A). i i The n-th order solutions have the following symmetry (n) (A) = (1)n +1 (n) (A), (n) (A,z) = (1)n (n) (A, z). (43) (44) (40) (41) (42)

In (40)-(44), remains xed. In other words, we do not change the sign of A in the denition of . The proofs for (40)-(44) are given in Appendix D. From (43) we see that (n) is an even (odd) function of A when n is odd (even). Therefore can be expressed as = a + b . Here a =

(2k +1) contains all odd orders of a 0 and is an even function of A, and b = k (2k) k =0 =1

contains all even orders of a 0 and is an odd function of A. We mention again that, here we keep xed and do not include the implicit dependence of A in . In Fig. 1, we illustrate the development of unstable interfaces for several different values of the Atwood number, using the perturbation solutions through fourth order given by (34)-(37) at t = 0.3, 0.6, 0.9 and 1.2. Figures 1(a), 1(b), 1(c) and 1(d) are for A = 0.1, 0.4, 0.7, and 1.0, respectively. In Fig. 1, as well as Figs. 2 and 3 shown later, we choose a 0 = 1/2,k = 1 and = 1. Note that the time scales in Fig. 1 are from linear stage to early nonlinear stage. Figure 1 shows that, as the Atwood number approaches to 1, the spike becomes more narrow and grows faster. In Fig. 2, we compare the shapes of a bubble and spike for A = 1. The solid curve is a spike and the dashed curve is a bubble. Figure 2 is obtained from Fig. 1(d) by ipping the bubble portion of the interface over the x-axis and translating it along the x-axis, so that the symmetry axis of the spike and bubble coincide. Figure 2 shows that, as t increases, the shape of spike becomes longer and narrower than that of bubble. In other words, the asymmetry between shapes of spike and bubble increases with time. Figures 2(a), 2(b), 2(c) and 2(d) are for the time = 0.3, 0.6, 0.9 and 1.2, respectively. In Fig. 3, we compare the interfaces of a spike and bubble for several different values of the Atwood number at time = 1.2. Figures 3(a), 3(b), 3(c) and 3(d) are for A = 0.1, 0.4, 0.7, and 1.0, respectively. Figure 3 shows that the interface becomes more and more

- 11 symmetric, as the Atwood number decreases. We see that, for xed value of , the spike for large A grows much faster than the one for smaller A, while the variation of A does not have a signicant inuence on the growths of bubble. However, for xed impulse strength (or, xed incident shock strength), the growth rates of both spike and bubble increase with A.

3. Pade Approximation and Nonlinear Theory for Compressible Fluids We have systematically derived the nonlinear perturbation solutions for incompressible uids. In this section, we construct the perturbation series for the overall growth rate and the growth rates of spike and bubble, and apply Pade approximation to extend the range of validity beyond the range of validity of series expansion itself. Then, we match the linear compressible solution at early times and the nonlinear incompressible solution at later times to arrive at a nonlinear theory for compressible uids. Let us discuss the initial conditions we are going to choose for the nonlinear solutions for incompressible uids. As we have outlined earlier, we would like to develop a nonlinear theory for compressible uids from early to late times. This will be done by matching the solution of the linear theory for compressible uids (valid at early times) and the solution of the nonlinear theory for incompressible uids (valid at later times). The growth rate determined from the linear theory for compressible uids contains a single Fourier mode only. Therefore, to be consistent with the solution of the linear theory, we choose single mode initial conditions for the nonlinear solution of incompressible uids: (x, 0) = a 0 cos(kx) and . (x, 0) = v 0 cos(kx). (45)

Here we assume that a 0 k is small and that v 0 is proportional to a 0 . Then from (15), we have . (n) a i (0) = v 0 1n 1i . (46)

v 0 will be determined later through matching. Applying the solution procedure given in Section 2.A, We have the following perturbation solutions for the initial conditions (45): (1) = (a 0 + v 0 t)cos(kx), (2) = 1 Akv 2 t 2 cos(2kx), 0 2 1 2 2 k v 0 [(4A 2 + 1)v 0 t 3 + 3a 0 t 2 ]cos(kx) 24 1 2 2 k v 0 [(4A 2 1)v 0 t 3 3a 0 t 2 ]cos(3kx), 8 (49) (47) (48)

(3) = +

- 12 (4) = + 1 3 2 k v 0 [4A 3 v 2 t 4 + 3Aa 2 t 2 ]cos(2kx) 0 0 12 1 3 2 k v 0 [(8A 3 4A)v 2 t 4 8Aa 0 v 0 t 3 + 3Aa 2 t 2 ]cos(4kx). 0 0 12 (50)

Note that, in (47)-(50), we did not use the initial conditions of the the impulsive model. This is due to the fact that the initial growth rate obtained from the impulsive model is based on the impulsive approximation to the incident shock. Such an approximation is not valid at early times for shock driven compressible uids. The impulsive model contains two approximations: the incompressible approximation and the impulsive approximation. In our approach, we applied only the incompressible approximation at the later times, but not the impulsive approximation. The spike and bubble are located at x = 0 and x = /k, respectively. Let vsp and vbb be the growth rates at the tips of spike and bubble, respectively. Then . . . . vsp = a (0,t) + b (0,t), vbb = a (0,t) + b (0,t). Here we have used the facts that a contains odd cosine Fourier modes and that b contains even . 1 cosine Fourier modes. a (0,t) represents the overall growth rate dened as v = (vsp vbb ). 2 . 1 b (0,t) represents (vsp + vbb ). Therefore we have 2 . 1 a (0,t) = v 0 a 0 v 2 k 2 t + (A 2 )v 3 k 2 t 2 + O ((a 0 k)5 ), 0 0 2 . 4 b (0,t) = Akv 2 t 2Ak 3 a 0 v 3 t 2 + k 3 (A 3 A)v 4 t 3 + O((a 0 k)6 ). 0 0 0 3 (51) (52)

The range of the validity of this perturbation solutions is quite limited. One of the standard methods to extend the range of validity beyond the range of validity of the nite Taylor series expansion series is a Pade approximation [22,23]. Applying the Pade approximation to (51), we have [18] . a (0,t) = v0 1 + v 0 a 0 k t + max{0,
2

a2 k 2 0

1 A + }v 2 k 2 t 2 0 2
2

(53)

Equation (53) is based on the P 0 Pade approximant when a 2 k 2 A 2 2 0 approximant when a 2 k 2 < A 2 0

1 and on the P 0 Pade 1 2

1 . The physical reason for choosing these Pade approximants is 2

that the overall growth rate decays at large times [8,10].

- 13 Similarly, we construct the P 1 Pade approximant for (52), and the result is 2 . b (0,t) = Akv 2 t 0 1 + 2k a 0 v 0 t +
2

4k 2 v 2 [a 2 k 2 0 0

1 + (1 A 2 )]t 2 3

(54)

Following the same proof given in Appendix D, we have also checked that, for the single mode initial conditions given by (45), the perturbations solutions of all orders satisfy the same symmetry properties given by (40)-(44). Note that the Pade approximants given by (53) and (54) satisfy the symmetry properties shown in (43). Equations (53) and (54) are approximate nonlinear solutions for incompressible uids. From the physical picture which we gave earlier, they are also approximate nonlinear solutions for compressible uids at later times. At early times, the solution is given by the linear theory for compressible uid, v lin . In order to develop a nonlinear theory for compressible uids, we need to construct expressions which smoothly match the linear solution for compressible uids at early times and the nonlinear solution for incompressible uids at later times. Furthermore, the matching should allow us to determine v 0 . We can adopt the techniques of asymptotic matching developed in boundary layer problems. In a boundary layer problem, the dynamics in a thin layer next to the boundary, called inner layer, is quite different from the dynamics in the region away from the boundary, called outer layer. One determines the solution in the inner layer, called inner solution, and the solution at the outer layer, called outer solution, separately, and match these two solutions to form matched asymptotics. Both inner and outer solutions contain integration constants which are determined by boundary conditions and matching which will be described shortly. We follow the same procedure. Since our system is an initial value problem, rather than a boundary value problem, a boundary condition is replaced by the initial conditions and the spatial variable is replaced by the temporal variable. In our case, the inner solution is the linear compressible solution and the outer solution is the nonlinear incompressible solution given by (53). A recipe to determine v 0 in (53) was proposed by Prandtl at the beginning of this century, namely by taking the large time limit of inner solution and small time limit of the outer solution, and setting them equal [24]. Therefore, we . . have the equation v lin (t ) = a (0,t 0). Here a is given by (53). This equation leads to v 0 = v = v lin (t ). Then, (53), the outer solution for the overall growth rate, becomes lin vincomp = v lin 1 + v a 0 k 2 t + max{0, a 2 k 2 A 2 + lin 0 1 2 2 2 }v lin k t 2 . (55)

- 14 Similarly, the outer solutions for growth rates of the spike and bubble can be obtained by setting . . . . v 0 to v in a + b and a + b , respectively. Equation (55) is a nonlinear incompressible lin solution with an initial growth rate given by v . For weak shocks, v in (55) can be approxilin lin mated by the linear solution of the impulsive model given by (1). Finally, following the procedure proposed by Prandtl [24] (see also chapter 2 of [25]), we add the inner and outer solutions and subtract the common part (v in our case) to arrive at matched asymptotics for the overall lin growth rate vmatch = v lin + v lin
2 1 1 + v a 0 k 2 t + max{0, a 2 k 2 A 2 + }v k 2 t 2 lin lin 0 2

v . lin

(56)

The matched solution for the spike and bubble growth rates can be obtained in a similar way. The essence of the matched asymptotic technique is to blend the inner and outer solutions smoothly. The technique proposed by Prandtl requires to calculate the asymptotic velocity of the linear theory. We would like to construct a simpler matched solution which has the same order of accuracy. The facts that (53) approaches v 0 at early times and that the growth rate of the linear theory for compressible uids approaches an asymptotic constant v at later times show that lin another way of matching can be achieved by replacing v 0 with v lin in (53) and (54). Then we have v= v lin 1 + v lin a 0 k 2 t + max{0, a 2 k 2 A 2 + 0 1 2 2 2 }v lin k t 2 (57)

for the overall growth rate, v bb = v + Akv 2 t lin 1 + 2k a 0 v lin t +

2

4k 2 v 2 [a 2 k 2 lin 0

1 + (1 A 2 )]t 2 3

(58)

for the growth rate of bubble and v sp = v + Akv 2 t lin 1 + 2k a 0 v lin t +

2

4k 2 v 2 [a 2 k 2 lin 0

1 + (1 A 2 )]t 2 3

(59)

for the growth rate of spike. Therefore we have achieved our goal of constructing approximate theories for the overall growth rates and the growth rates of the bubble and spike in compressible uids. We emphasize that the range of validity of (57)-(59) are not limited to the range of validity of generating series (47)-(50). This is well known from the theory of Pade approximation

- 15 [22,23]. It is easy to see that in the early time, or small amplitude limits, (57)-(59) approach v lin . Equations (58) and (59) show that the spike grows faster than the bubble. We comment that our nonlinear theories given by (57)-(59) contain no adjustable parameter. Equations (57)-(59) are applicable to the systems with no indirect phase inversion only. An indirect phase inversion is dened for the situation a 0 (0+)v lin (t ) < 0. (We use the notations 0 and 0+ to represent the times just before and after shock contact interaction, respectively.) For the case of reected shock, the indirect phase inversion usually does not occur [16]. The comparison between results of the matched asymptotic technique (56) proposed by Prandtl and the matching given by (57) will be presented in the next section.

4. Validation of the the nonlinear theory and physical picture In this section, we validate our nonlinear theory as well as our physical picture. Our theoretical derivations contain two major steps: one is the Pade approximation, and the other is the matching between the compressible linear solutions and the incompressible nonlinear solutions. To check the validity of Pade approximation, we consider the incompressible uids with A = 1. For this system, Hecht et. al. [17] have developed a Layzer-type [21] potential ow model for the bubble. The model assumes that the uid is incompressible and irrotational, and approximates the shape of the interface near the tip of the bubble as a parabola. An asymptotic bubble growth rate of 2/3kt has been determined analytically. The prediction of the Layzer-type potential ow model over nite time scale can be obtained numerically. It has been shown that the numerical solution of the Layzer-type model for a bubble is in excellent agreement with the result of a full nonlinear numerical simulation in the case of A = 1, (see Fig. 1 of [17]). In Fig. 4, we show the comparison among the predictions of the Layzer-type potential ow model, the perturbation solutions given by (47)-(50), and the Pade approximants given by (55). Here, the asymptotic bubble growth rate of the Layzer-type model is based on the analytic expressions 2 /3kt, while the nite time solution of the Layzer-type model is determined numerically. The parameters used here are a 0 k = v = 1/2. It is clear from Fig. 4, that the range of the validity of the perturbation solution lin is small, but the range of the validity of the Pade approximant is quite large. This is an important feature of Pade approximation . As a text book says, "The real power of Pade summation is illus trated by its application to divergent series"(see page 385 of [23]).

- 16 . . . In Fig. 5(a), we compare the predictions of the perturbation solutions, 1 and 1 + 3 (with v 0 = v ), the prediction from the Pade approximation given by (55) for incompressible system lin (the outer solution), and the result from the full nonlinear numerical simulation for the overall . . growth rate of the compressible unstable interface between air and SF 6 . Note that 2 and 4 do not contribute to the overall growth rate. The compressible interface is accelerated by a weak shock of Mach number 1.2 moving from air to SF6 . The reected wave is a shock. The initial amplitude of the perturbation is a 0 (0) = 2.4mm, the wave length is 37.5mm and the pressure ahead of the shock is 0.8 bar. These physical parameters are taken from Benjamins experiments [4]. The post-shock Atwood number is A = 0.701. Figure 5(a) shows again that the range of validity of the nonlinear perturbation solutions are very limited and the Pade approximation has suc cessful extended the range of validity. The prediction of the perturbation solution deviates from the nonlinear theory around t = 250 s. At this time the dimensionless amplitude of the ngers at the interface, k, is about 0.8. Figure 5(a) shows that the prediction of (55) is accurate even at intermediate time scales. In Fig. 5(b), we compare the results from the compressible linear theory, the incompressible nonliner theory given by (55), the matched nonlinear theory for compressible uids given by (57), and a full nonlinear numerical simulation for the overall growth rate of an air-SF 6 unstable interface. The physical parameters here are identical to the ones shown in Fig. 5(a). Figure 5(b) shows that the technique of the matching is successful. We comment that the difference between the predictions of (57) and the linear theory is due to the nonlinear effects; while the difference between the predictions of (55) and (57) is due to the compressible effects. Figure 5(b) shows that the dynamics of the RM unstable interface indeed changes from an approximately linear and compressible dominant one at early times to a nonlinear dominant and approximately incompressible one at later times. The transition occurs around the highest peak of the nonlinear growth rate. This validates the physical picture on which our theory was based. In Fig. 6, we compare the predictions of the matched asymptotics given by (56) which is derived from the matching technique proposed by Prandtl, the matched nonlinear theory given by (57), the full numerical simulation as well as the compressible linear theory and the impulsive model (1) for an air-SF6 unstable interface. Figure 6(a) is for the overall growth rate and Fig. 6(b) is for the amplitude. Since the incident shock is weak, the prediction of the impulsive model is close to the asymptotic velocity of the linear theory. Figures 6(a) and 6(b) shows that the results

- 17 of (56) and (57) indeed give same order of accuracy. The reader may compare Fig. 6(a) with Fig. 5(b) to distinguish the prediction of (56) from that of (57). Since the expression of the matched nonlinear theory given by (56) is simpler than the one given by (57), we use (56)-(58) for the predictions of the growth rates. We emphasize that our theory provides the quantitative growth rates from linear to nonlinear stages for all Atwood numbers. Although we have demonstrated that the range of the validity of the Pade approximant is signicantly larger than that of primitive perturbation expansion, the range of the validity of the Pade approximant is still not innity. Therefore, our theory may not applicable at asymptotic large times. In fact, at the time when v bb = 0, our theory is no longer valid. In reality, the unstable system becomes turbulent at very late times. The physics of uid turbulence involves much more than just the nonlinearity [26].

5. Quantitative Predictions In this section we present the quantitative predictions of our nonlinear theory for the overall growth rate and the growth rates of the bubble and spike. Figure 6(b) shows that our analytical prediction is in remarkable agreement with the result of the full non-linear numerical simulation from linear to nonlinear regimes. In experiments, it was difcult to measure the growth rate directly. Instead, one measured the amplitude of the disturbed interfaces, i.e. the half of the longitudinal distance between the spike and bubble tips. One assumed that the amplitude was a linear function of time and applied a linear regression analysis to determine the overall growth rate of the unstable interfaces. The overall growth rate determined from the experimental data was 9.2 m/s over the time period 310-750 s, (see Fig. 6(b)). When we applied the linear regression to the amplitude predicted by our theory and to the amplitude determined from numerical solution of full Euler equations, we found the identical growth rate 9.3 m/s for over that time period. Therefore, the prediction of our theory is in excellent agreement with the experimental result, as well as with the full nonlinear numerical simulation. Predictions of the growth rate from the impulsive model and from the linear theory are 15.6 m/s and 16.0 m/s, respectively. A comparison for the growth rate of Kr-Xe unstable interface is shown in Fig. 7. The interface is accelerated by a strong shock of Mach number 3.5 moving from Kr to Xe. The reected wave is a shock. The initial amplitude of the perturbation is a 0 (0) = 5 mm, the wave length is 36 mm, and the pressure ahead of the shock is 0.5 bar. These physical parameters are taken from

- 18 Zaytsevs recent experiments. The post-shock Atwood number is A = 0.184. The dimensionless perturbation amplitude a 0 (0)k is 0.87. This initial perturbation amplitude is about two times larger than the amplitude a 0 (0)k = 0.40 given in Figs. 5 and 6 for air-SF6 . Figure 7 shows that the predictions of (57) is still in quite good agreement with the results of the full non-linear numerical simulation. The experimental results are not yet published. The physical systems shown in Figs. 5, 6 and 7 belong to the parameter regime a2 k 2 > A 2 0 1 . Our theoretical predictions were based on P 2 Pade approximation in that 0 2 1 ), and the result from full numerical simulation can 2

parameter regime. A comparison between the prediction of our nonlinear theory based on P 0 1 construction (for the case of a 2 k 2 < A 2 0

be found in Fig. 3 of reference [18]. As one expected, the theoretical prediction based on P 0 con1 struction of Pade approximant is less accurate. However, our theoretical prediction is still quite good and is signicantly better than the predictions of linear theory and impulsive model [18]. In Fig. 8, we compare the effects of different initial amplitude. We consider Kr-Xe unstable interface with the same physical parameters used in Fig. 7. We choose the two initial amplitudes, a 0 (0)k = 0.87 and 0.6. The solid curves are results from the linear theory and nonlinear theory, and the dashed curves are ones from the numerical simulations. The index (i) corresponds to a 0 (0)k = 0.87 and the index (ii) a 0 k(0) = 0.6. Figure 8 shows that the predictions of (57) are in good agreement with the results of the numerical simulation for both cases and the nonlinear effect is more pronounced in the larger initial amplitude case. In Fig. 9, we compare our theoretical predictions for the the bubble and spike, i.e. (58) and (59), with the results from full numerical simulations as well as the predictions of the linear theory and Richtmyers impulsive model for air-SF 6 interface. The physical parameters are same as the ones used in Fig. 5. Figures 9(a) and 9(b) are for the growth rate and the amplitude of the bubble, respectively. Figure 9(c) and 9(d) are for the growth rate and the amplitude of the spike, respectively. The amplitudes of the bubble and spike are determined by integrating (58) and (59) over time, respectively. In Fig. 10, we consider a Kr-Xe interface. The physical parameters are same as the ones used in Fig. 7. Figures 10(a) and 10(b) are for the growth rate and the amplitude of the bubble, respectively. Figures 10(c) and 10(d) are for the growth rate and the amplitude of the spike, respectively. Figures 9 and 10 show that our theoretical predictions are in good agreement with the results from full numerical simulations, while the predictions of the linear theory for

- 19 compressible uids and the linear impulsive model are qualitatively incorrect at later times. We comment that our theory is still valid at the early stage after the formations of the roll-ups, mushroom-shaped vortex structures, at the spike. The reason is that our theory is for the tips of the spike and bubble. The roll-ups are secondary structures which do not have signicant effects on the growth rates at the early stages of roll-ups. At the later stage of roll-ups, the secondary structures do affect the growth rate of the spike. This is why the prediction for the spike growth rate becomes less accurate than that for the bubble at later times. For further improvement of the theory for RM instability, it may be necessarily to include the rotational effects in the dynamics of the spike.

6. Conclusion In this paper, a nonlinear theory for Richtmyer-Meshkov instability in compressible uids has been presented. The theory is applicable from the linear (small amplitude) regime to the moderately large amplitude nonlinear regime of the development of RM instability for systems with no indirect phase inversion. Most of full numerical simulations and experiments were conducted over this time range. Our theory may not be applicable at very late time. Our theory shows that the dynamics of RM instability is compressible and approximately linear at early time, and nonlinear and approximately incompressible at late times. This physical picture has been validated by the results from full numerical simulations. Our theory provides analytical predictions (expressed explicitly in terms of the solution of the linear theory) for the overall growth rate, as well as the growth rates of the spike and bubble, of RM unstable interface with all density ratios. Our theoretical predictions are in excellent agreement with the results from full numerical simulations for compressible uids, and the results from experiments on air-SF6 unstable interface without any adjustable parameters. In the process of deriving our nonlinear theory, we have developed the procedure to obtain the nonlinear perturbation solutions for incompressible uids with arbitrary density ratio and arbitrary velocity distribution along the initial interface. We give an analytical expression for the n th order solution. Our solution procedure is recursive, - the higher order solutions are expressed in terms of the lower order quantities. The nonlinear perturbation solution for the impulsive model are given explicitly through fourth order in this paper. The nonlinear theory presented in this paper is for Richtmyer-Meshkov instability in two dimensions for the case of a reected shock wave. Our theoretical approach can also be extended

- 20 to the case of reected rarefaction wave and the three dimensions.

Appendix A: Derivation of the First and Second Order Quantities In this appendix, we derive the rst order quantities (1), (1) and (1) , and second order quantities (2), (2) and (2) . For the rst order equations, (16)-(18) and (31) give
i S (1) (t)cos(ikx) = 0i1S (1) (t)cos(ikx) = 0i1T (1) (t)cos(ikx) = 0i1R (1) (t)cos(ikx) = 0. i i i 0i1

It follows that S (1) (t) = S (1) (t) = T (1) (t) = R (1) = 0. Then from (25)-(27), we have the rst order i i i i solution (34) and (1) = a 0 kz e cos(kx), k (1) = a 0 kz e cos(kx). k (A1)

For the second order equation, (16)-(18) and (31) give

i S (2) (t)cos(ikx) = i =0 i S (2) (t)cos(ikx) = i =0 2 2 2

2 (1) (1) (1) (1) + x x z 2 2 (1) (1) (1) (1) + x x z 2

at z = 0, at z = 0,

(A2)

(A3)

i =0

i T (2) (t)cos(ikx) = (

(1) (1) 2 (1) (1) 1 2 (1) ) [( )2 + ( )2 ] z x 2 tz tz (1) 2 (1) 2 1 ) ] ) +( [( z x 2 at z = 0, (A4) (A5)

+
i R (2) cos(ikx) = 2

i =0

(1) (1) z z

at z = 0, t = 0.

The right hand sides of (A2)-(A5) can be evaluated analytically from the rst order solution given by (34) and (A1). Then we can determine the Fourier coefcients. The non-zero Fourier coefcients are S (2) = a 2 k(t + 1), S (2) = S (2) , T (2) = 2 0 2 2 0 1 1 ( )2 a 2 , R (2) = R (2) = ( )a 2 . 0 0 0 2 2 2

From (25)-(27) and (31), we obtain the second order solution (35) and (2) = (2) = 1 1 2 a 0 [(1 + A)t + 1]e 2kz cos(2kx) Aa 2 (t + 1), 0 2 2 1 1 2 a 0 [(1 A)t + 1]e 2kz cos(2kx) + Aa 2 (t + 1). 0 2 2 (A6) (A7)

- 21 Appendix B: Derivation of the Third Order Quantities In this appendix, we derive the third order quantities (3) ,(3) and (3) . From (16)-(18), the source terms for the third order equations are given by
i S (3) (t)cos(ikx) = ( 3

i =1

2 (1) (2) 1 3 (1) (1)2 2 (2) (1) ) + + 2 z 3 z 2 z 2 2 (1) (1) (2) (1) (1) (2) ) + +( x x xz x x 2 (1) (2) 1 3 (1) (1)2 2 (2) (1) ) + + 2 z 3 z 2 z 2 2 (1) (1) (2) (1) (1) (2) ) + +( x x xz x x 2 (2) (1) 2 (2) ) tz tz 2 (1) (2) 2 (1) 3 (1) (1)2 3 (1) 1 ) ) + ( ( tz tz 2 tz 2 tz 2 (1) 2 (1) (1) (1) 2 (1) (1) 2 (1) (1) (1) 2 (1) ) + ) + ( + z z 2 x xz z xz x z 2 (1) (2) (1) (2) (1) (2) (1) (2) ) + ) + ( + z z x x z z x x at z = 0. (B3) at z = 0, at z = 0, (B1) (B2)

+
i S (3) (t)cos(ikx) = ( 3

i =1

+
i T (3) (t)cos(ikx) = ( i =1 3

( (

and initial condition (31) is

3 (2) i R (3) cos(ikx) = [ z i =1

2 (1) 2 (1) 1 (2) ][cos(kx)]2 . ]cos(kx) + [ 2 2 2 z z z

(B4)

From the rst order solution given by (34) and (A1), and the second order solution given by (35) and (A6)-(A7), we can evaluate the right hand sides of (B1)-(B4) explicitly. From these equations we can determine S (3) , S (3) , T (3) , and R (3) , The non-zero source terms are i i i i S (3) (t) = 1 S (3) (t) = 3 1 2 3 k a 0 [(2A + 3)2 t 2 + (4A + 6)t + 3], 8 1 2 3 k a 0 [(18A + 9)2 t 2 + (12A + 18)t + 9], 8

S (3) (t) = S (3) (t, A A), i = 1, 3, i i T (3) (t) = 1 1 3 2 ka 0 [( )A(3t + 1) + ( + )(t + 1)], 2

- 22 T (3) (t) = 3 R (3) = 1 R (3) = 3 1 3 2 ka 0 [( )A(t + 1) + ( + )(t + 1)], 2 1 3 ka 0 ( + ), 8 3 3 ka 0 ( + ). 8

From our solution formulae (25) we obtain (36) for (3) , and from (26) and (26), we have (3) = 1 3 ka 0 [( 4A 2 + 2A + 2)2 t 2 + (4A + 4)t + 1]e kz cos(kx) 8 + (3) = 1 3 ka 0 [(12A 2 + 18A + 6)2 t 2 + (12A + 12)t + 9]e 3kz cos(3kx), 24 (B5)

1 3 ka 0 [(4A 2 + 2A 2)2 t 2 + (4A 4)t 1]e kz cos(kx) 8 + 1 3 ka 0 [( 12A 2 + 18A 6)2 t 2 + (12A 12)t 9]e 3kz cos(3kx). 24 (B6)

Appendix C: Derivation of the Fourth Order Quantities In this appendix, we derive the fourth order quantities (4) ,(4) and (4) . From (16)-(18), the source terms for the fourth order equations are given by
i S (4) (t)cos(ikx) = ( i =1 4

2 (1) (3) 3 (1) (1) (2) 1 4 (1) (1)3 2 (2) (2) + + + 6 z 4 z 2 z 3 z 2 2 (1) (1) (2) (2) (1) (3) 1 3 (2) (1)2 2 (3) (1) ) + +( )+ + x x xz x x 2 z 3 z 2 2 (1) (2) 1 3 (1) (1)2 2 (2) (1) (3) (1) ) + + + x x xz 2 xz 2 xz at z = 0,

+(
i S (4) (t)cos(ikx) = ( 4

i =1

2 (1) (3) 3 (1) (1) (2) 1 4 (1) (1)3 2 (2) (2) + + + 6 z 4 z 2 z 3 z 2 2 (1) (1) (2) (2) (1) (3) 1 3 (2) (1)2 2 (3) (1) ) + +( )+ + x x xz x x 2 z 3 z 2 2 (1) (2) 1 3 (1) (1)2 2 (2) (1) (3) (1) ) + + + x x xz 2 xz 2 xz at z = 0,

+(
i T (4) (t)cos(ikx) = ( i =1 4

2 (2) (2) 2 (2) ) tz tz 2 (3) (1) 2 (3) 2 (2) (1)2 3 (2) 1 ) ) + ( ( tz tz 2 tz 2 tz 2

- 23 1 2 2 (1) (1) xz
2

[ +

1 2

(2) x

(1) 2 (1) (2) xz x

2 (1) (2) (1) 1 (1) 3 (1) (1)2 (1) 2 (2) (1) (1) (3) ] + + + 2 x xz 2 x xz x x xz x 2 (1) (1) xz
2

1 + [ 2 +

1 + 2

(2) x

(1) 2 (1) (2) x xz

2 (1) (2) (1) 1 (1) 3 (1) (1)2 (1) 2 (2) (1) (1) (3) ] + + + 2 x xz 2 xz x x x x xz 2 (1) (1) z 2
2

1 [ 2 +

1 + 2

(2) z

(1) 2 (1) (2) z z 2

2 (1) (2) (1) 1 (1) 3 (1) (1)2 (1) 2 (2) (1) (1) (3) ] + + + 2 z z z z z z 3 z 2 z 2 2 (1) (1) z 2
2

1 + [ 2 +

1 + 2

(2) z

(1) 2 (1) (2) z z 2

2 (1) (2) (1) 1 (1) 3 (1) (1)2 (1) 2 (2) (1) (1) (3) ] + + + 2 z z 3 z z z z z 2 z 2

at z = 0. From (31) we have

4 (3) i R (4) cos(ikx) = [ z i =0

2 (2) 2 (2) 1 (3) ][cos(kx)]2 ]cos(kx) + [ 2 2 2 z z z

2 (1) 2 (1) 1 ][cos(kx)]3 . [ 2 2 6 z z

The non-zero Fourier coefcients are S (4) (t) = 2 S (4) (t) = 4 1 4 3 a 0 k [(4A 2 + 18A + 2)3 t 3 + (12A 2 + 30A + 6)2 t 2 + (12A + 3)t + 4], 12 1 4 3 a 0 k [(30A 2 + 24A + 2)3 t 3 + (18A 2 + 36A + 6)2 t 2 + (12A + 15)t + 8], 6

S (4) (t) = S (4) (t, A A), i = 2, 4, i i T (4) (t) = 0 T (4) (t) = 2 1 2 4 2 k a 0 ( )[(2A 2 + 2)2 t 2 + (4A 2 + 4)t + 3], 8

1 2 4 2 k a 0 [( )((20A 2 + 2)2 t 2 + (8A 2 +4)t + 5) + ( + )A(182 t 2 + 20t + 4)], 8

- 24 T (4) (t) = 4 1 2 4 2 k a 0 [( )((7A 2 + 2)2 t 2 + (6A 2 + 4)t + 2) + ( + )A(92 t 2 + 10t + 2)]. 4

1 R (4) = k 2 a 4 ( ), 0 0 8 R (4) = 2 R (4) = 4 5 2 4 k a 0 ( ), 24 1 2 4 k a 0 ( ). 3

From our solution formulae (25) we obtain (37) for (4) , and from (26) and (26), we have (4) = 1 2 4 k a 0 [( 16A 3 + 4A 2 + 18A + 2)3 t 3 24 + (12A 2 + 30A + 6)2 t 2 + 3t A + 4]e 2kz cos(2kx) + 1 2 4 k a 0 [(16A 3 + 30A 2 + 16A + 2)3 t 3 24

+ (18A 2 + 24A + 6)2 t 2 + (15A + 15)t + 8]e 4kz cos(4kx) + (4) = 1 2 4 k a 0 [(2A 3 + 2A)3 t 3 + (6A 3 + 6A)2 t 2 + 9At + 3A ], 24 (C1)

1 2 4 k a 0 [(16A 3 + 4A 2 18A + 2)3 t 3 24 + (12A 2 30A + 6)2 t 2 + 3t + A + 4]e 2kz cos(2kx) + 1 2 4 k a 0 [( 16A 3 + 30A 2 16A + 2)3 t 3 24

+ (18A 2 24A + 6)2 t 2 + ( 15A + 15)t + 8]e 4kz cos(4kx) 1 2 4 k a 0 [(2A 3 + 2A)3 t 3 + (6A 3 + 6A)2 t 2 + 9At + 3A ]. 24 (C2)

Appendix D: Proofs for (40)-(44) Here we provide proofs for (40)-(44). From (20) and (21), it is easy to see that and have opposite sign for their z dependence. We prove (40)-(44) by induction. From (34) and (35) as well as the fact that S (1) = S (1) = T (1) = R (1) = 0, we can see (40)-(44) hold for k = 1. i i i i

- 25 As an induction hypothesis, we assume (40)-(44) hold for all k n 1. Let us prove the case k = n, for n 2. We prove (40) rst. We examine the rst term on the right hand side of (17),

sum1 p !

1 p +1 (a) (A) p (ni ) (A) i =1 z p +1

a +p +1+

1 p ! (1) sum1

(ni + 1) p +1 (a) (A) i =1

z
p +1

i =1

(ni )

(A)

= (1)(n +1) [

1 p +1 (a) (A) p (ni ) (A)]. i =1 z p +1 sum1 p !

(D1)

The rst equality in (D1) follows from the induction hypothesis (a) (A) = (1)a (a) (A) and
(ni )

(A) = (1)

(ni + 1) (ni )

(A).

for all a,ni < n, and the fact that the z dependence of (a) and (a) has an opposite sign (see (18) and (31)). The second equality in (D1) follows from the fact

ni + a = n, (see the denition of i =1

sum1). Comparing the nal expression of (D1) with the rst summation on the right hand side of (16), we see that the rst summations from the right hand sides of (16) and (17) indeed satisfy the relation (40). The proof for the relation between the second terms on the right hand sides of (16) and (17) is similar (keeping in mind that
i =1 p

ni + a + b = n; see the denition of sum2).

p q

Therefore

we have proven that (40) is true for k = n. Following the similar procedure, we have checked that (41) and (42) are hold for k = n also. There we need the fact denition of sum3.) Now we prove that (43) is true for k = n. From (38) and (39) and the identities 1 = (1 + A) and 2 + (25) can be written as a (n) (A) i 1 = {ik(t t )[(1 + A)F (n) (A,t ) (1 A)F (n) (A,t )] i i 20 + (1 + A)S (n) (A,t ) + (1 A)S (n) (A,t ) + ik[(1 + A)G (n) (A) (1 A)G (n) (A)]}dt i i i i = 1 {ik(t t )[[(1)n F (n) (A,t ) F (n) (A,t )] + A[(1)n F (n) (A,t ) + F (n) (A,t )]] i i i i 2 0
t t

ni + j m j + a + b = n. (See the i =1 =1

1 = (1 A), 2 +

(D2)

- 26 -

[(1)n S (n) (A,t ) S (n) (A,t )] A[(1)n S (n) (A,t ) + S (n) (A,t )] i i i i + ik[(1)n G (n) (A) G (n) (A)] + ikA[(1)n G (n) (A) + G (n) (A)]}dt . i i i i The second equality in (D3) follows from (40)-(42). From (D3) it is easy to see that a (n) (A) = (1)n +1 a n (A). i i From (D4) and (19), we see that (43) holds for k = n, n 2. Finally, let us prove (44). From (38)-(42) and (D2), (26) can be expressed as b (n) (A) i 1 = {[(1)n F (n) (A,t ) F (n) (A,t )] + A[(1)n F (n) (A,t ) + F (n) (A,t )]}dt i i i i 20 + 1 {[(1)n G (n) (A) G (n) (A)] + A[(1)n G (n) (A) + G (n) (A)]} i i i i 2 1 (1 + A)[(1)n S (n) (A) + S (n) (A)]. i i 2ik (D5)
t

(D3)

(D4)

Similarly, (36) can be expressed as b (n) (A,t) = i 1 {[(1)n F (n) (A,t ) F (n) (A,t )] + A[(1)n F (n) (A,t ) + F (n) (A,t )]}dt i i i i 2 0 1 {[(1)n G (n) (A) G (n) (A)] + A[(1)n G (n) (A) + G (n) (A)]} i i i i 2 (D6)
t

1 (1 A)[(1)n S (n) (A) + S (n) (A)]. i i 2ik

From (D5) and (D6), we see that b (n) (A) = (1)n b (n) (A). Then from (20) and (21), we see that i i (44) holds for k = n, n 2. This completes our proofs for (40)-(44).

References [1] R. D. Richtmyer, "Taylor instability in shock acceleration of compressible uids", Comm. Pure Appl. Math. 13, 297, (1960).

- 27 [2] E. E. Meshkov, "Interface of two gases accelerated by a shock wave", Fluid Dyn. 4, 101, (1969). [3] A. N. Aleshin, E. V. Lazareva, S. G. Zaytsev, V. B. Rozanov, E. G. Gamalii and I. G. Lebo, "Linear, nonlinear, and transient stages in the development of the RichtmyerMeshkov Instability", Sov. Phys. Dokl. 35, 159, (1990). [4] R. Benjamin, D. Besnard, and J. Haas, "Shock and reshock of an unstable interface", LANL report, LA-UR 92-1185, (1993). [5] K. A. Meyer and P. J. Blewett, "Numerical investigation of the stability of a shockaccelerated interface between two uids", Phys. Fluids 15, 753, (1972). [6] L. D. Cloutman, and M. F. Wehner, "Numerical simulation of Richtmyer-Meshkov instabilities", Phys. Fluids A 4, 1821, (1992). [7] T. Pham and D. I. Meiron, "A numerical study of Richtmyer-Meshkov instability in continuously stratied uids", Phys. Fluids A 5, 344, (1993). [8] J. Grove, R. Holmes, D. H. Sharp, Y. Yang and Q. Zhang, Quantitative theory of Richtmyer-Meshkov instability, Phys. Rev. Lett. 71, 3473, (1993). [9] K. O. Mikaelian, "Growth rate of the Richtmyer-Meshkov instability at shocked interfaces", Phys. Rev. Lett. 71, 2903, (1993). [10] R. L. Holmes, J. W. Grove and D. H. Sharp, "A numerical investigation of RichtmyerMeshkov instability using front tracking", J. Fluid Mech. 301, 51, (1995). [11] U. Alon, J. Hecht, D. Ofer and D. Shvarts, "Power laws and similarity of Rayleigh-Taylor and Richtmyer-Meshkov mixing fronts at all density ratios", Phys. Rev. Lett. 74, 534, (1995). [12] D. L. Youngs, "Numerical simulation of turbulent mixing by Rayleigh-Taylor instability", Laser and Particle Beams 14, 725, (1994). [13] G. Fraley, "Rayleigh-Taylor stability for a normal shock wave-density discontinuity interaction", Phys. Fluids 29, 376, (1986). [14] S. W. Haan, "Weakly nonlinear hydrodynamic instabilities in inertial fusion", Phys. Fluids B 3, 2349, (1991). [15] R. Samtaney and N. J. Zabusky, "On shock polar analysis and analytical expressions for vorticity deposition in shock-accelerated density-stratied interfaces", Phys. Fluids A 5, 1285, (1993).

- 28 [16] Y. Yang, Q. Zhang and D. H. Sharp, "Small amplitude theory of Richtmyer-Meshkov instability", Phys. Fluids A 6, 1856, (1994). [17] J. Hecht, U. Alon and D. Shvarts, "Simple potential ow models of Rayleigh-Taylor and Richtmyer-Meshkov bubble fronts", Phys. Fluids A 6, 4019, (1994). [18] Q. Zhang and S.-I. Sohn, "An analytical nonlinear theory of Richtmyer-Meshkov instability", Phys. Lett. A 212, 149, (1996). [19] V. Rupert, "Shock-interface interactions: current research on the Richtmyer-Meshkov problem", in Shock Waves, Proceedings of the 18th International Symposium on Shock Waves, edited by K. Takayama, (Springer-Verlag, Berlin, 1992). [20] PCTM Workshops, 1988-1995: Proc. First Intl Workshop on the Physics of Compressible Turbulent Mixing, Princeton, 1988, edited by W. P. Dannevik, A.C. Buckingham and C. E. Leith; Proc. Second Intl Workshop on the Physics of Compressible Turbulent Mixing, Pleasanton CA, 1989, compiled by V. Rupert; Proc. Third Intl Workshop on the Physics of Compressible Turbulent Mixing, Abbey of Royaumont, France, 1991; Proc. Fourth Intl Workshop on the Physics of Compressible Turbulent Mixing, Cambridge, England, 1993; Proc. Fifth Intl Workshop on the Physics of Compressible Turbulent Mixing, SUNY at Stony Brook, NY, 1995; [21] D. Layzer, "On the gravitational instability of two superposed uids in a gravitational eld", Astrophys. J. 22, 1, (1955). [22] A. Pozzi, Applications of Pade Approximation Theory in Fluid Dynamics, (World Scientic Publishing Co., 1994). [23] C. M. Bender and S. A. Orszag, Advanced Mathematical Methods for Scientists and Engineers, (McGraw-Hill, 1978). [24] L. Prandtl, "Fluessigkeiten bei sehr kleimer reibung, III", International Math. Kongress, Heidelberg, Teubner, Leipzig, 484, (1905). [25] P. A. Lagerstrom, Matched Asymptotic Expansions, (Springer-Verlag, New-York, 1988). [26] W. D. McComb, The Physics of Fluid Turbulence, (Oxford Science Publications, 1990).

- 29 Captions Figure 1. The predictions of the perturbation solutions through fourth order at time t = 0.3, 0.6, 0.9 and 1.2. Solutions for different values of the Atwood number are shown here. Figure 1(a), 1(b), 1(c) and 1(d) are for A = 0.1, 0.4, 0.7 and 1.0, respectively. Figure 2. Comparison of the shapes of a spike and bubble for A = 1. The solid curve is a spike and the dashed curve is a bubble. Figure 2 is obtained from Fig. 1(d) by ipping the bubble portion of the interface over the x-axis and translating it along the x-axis, so that the symmetry axis of the spike and bubble coincide. Figures 2(a), 2(b), 2(c) and 2(d) are for the time = 0.3, 0.6, 0.9 and 1.2, respectively. Figure 3. Comparison of the shapes of a spike and bubble for different values of Atwood number at time = 1.2. The solid curve is a spike and the dashed curve is a bubble. Figures 3(a), 3(b), 3(c) and 3(d) are for the A = 0.1, 0.4, 0.7 and 1.0, respectively. Figure 4. Comparison of the perturbation solutions given by (47)-(50), the Pade Approxi mation given by (55), and the Layzer-type potential ow model [17] for A = 1. The asymptotic bubble growth rate of the Layzer-type model is based on the analytic expressions 2/3kt, while the nite time solution of the Layzer-type model is determined numerically. The parameters here are a 0 k = v = 1/2. The curves labeled a, b, c and d correspond to lin . . . . . . . . . . 1, 1 2, 1 2 + 3, and 1 2 + 3 4 , respectively. Figure 5. Comparison of the predictions for the overall growth rates of the compressible unstable interface between air and SF6 . A shock of Mach number 1.2 incidents from air to SF6 . The dimensionless preshocked interface amplitude is 0.40. Figure 5(a) is the com. . . parison of predictions of the perturbation solutions, 1 and 1 + 3 , the prediction from the Pade approximation given by (55), and the result from the full nonlinear numerical simula tion. Figure 5(b) is the comparison of the results from the compressible linear theory, the incompressible nonliner theory given by (55), the matched nonlinear theory for compressible uids given by (57), and a full nonlinear numerical simulation. Figure 5(b) shows that the RM unstable system indeed changes from a compressible and approximately linear system to an a nonlinear and approximately compressible system at late times. Figure 6. Comparison of the results of the matched asymptotics (56), the matched nonlinear theory given by (57), a full nonlinear numerical simulation, the linear theory, and Richtmyers impulsive model for air-SF6 unstable interface. (a) is for the growth rate and (b) is for the amplitude. The physical parameters here are identical to the ones in Fig. 5.

- 30 The comparison shows that (56) and (57) have same accuracy. Figure 7. Comparison of the results of the linear, nonlinear, and Richtmyers impulsive theories, and of a full nonlinear numerical simulation for Kr-Xe unstable interface. A shock of Mach number 3.5 incidents from Kr to Xe. The dimensionless preshocked interface amplitude is 0.87. (a) is for the growth rate and (b) is for the amplitude. Figure 8. Comparison of the effects of different initial amplitude for Kr-Xe unstable interface. The physical parameters are same as the ones in Fig. 7 except initial amplitudes. The solid curves are results from the linear theory and nonlinear theory, and the dashed curves are ones from the numerical simulations. The curves labeled (i) correspond to a 0 k = 0.87 and the curves labeled (ii) correspond to a 0 k = 0.6. Figure 9. Comparison of the predictions of the nonlinear theory, i.e. (58) and (59), and full numerical simulations, for the growth rates and amplitudes of bubble and spike for air-SF 6 interface, as well as the predictions of the linear theory and Richtmyers impulsive model. The physical parameters are same as the ones used in Fig. 5. (a) and (b) are for the bubble. (c) and (d) are for the spike. Figure 10. Comparison of the predictions of the nonlinear theory, i.e. (58) and (59), and full numerical simulations, for the growth rates of bubble and spike at Kr-Xe interface, as well as the predictions of the linear theory and Richtmyers impulsive model. The physical parameters are same as the ones used in Fig. 7. (a) and (b) are for the bubble. (c) and (d) are for the spike.