You are on page 1of 19

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN FLUIDS

Int. J. Numer. Meth. Fluids 2011; 66:12071225


Published online 5 March 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/fld.2306

A high-resolution scheme for compressible multicomponent flows


with shock waves

Soshi Kawai1, , and Hiroshi Terashima2


1 Center for Turbulence Research, Stanford University, 488 Escondido Mall, Stanford, CA 94305-3035, U.S.A.
2 Mechanical Engineering Department, Worcester Polytechnic Institute, 100 Institute Road, Worcester,
MA 01609-2280, U.S.A.

SUMMARY
A simple methodology for a high-resolution scheme to be applied to compressible multicomponent flows
with shock waves is investigated. The method is intended for use with direct numerical simulation or large
eddy simulation of compressible multicomponent flows. The method dynamically adds non-linear artificial
diffusivity locally in space to capture different types of discontinuities such as a shock wave, contact
surface or material interface while a high-order compact differencing scheme resolves a broad range
of scales in flows. The method is successfully applied to several one-dimensional and two-dimensional
compressible multicomponent flow problems with shock waves. The results are in good agreement with
experiments and earlier computations qualitatively and quantitatively. The method captures unsteady shock
and material discontinuities without significant spurious oscillations if initial start-up errors are properly
avoided. Comparisons between the present numerical scheme and high-order weighted essentially non-
oscillatory (WENO) schemes illustrate the advantage of the present method for resolving a broad range
of scales of turbulence while capturing shock waves and material interfaces. Also the present method
is expected to require less computational cost than popular high-order upwind-biased schemes such as
WENO schemes. The mass conservation for each species is satisfied due to the strong conservation form
of governing equations employed in the method. Copyright q 2010 John Wiley & Sons, Ltd.

Received 21 April 2009; Revised 5 December 2009; Accepted 23 January 2010

KEY WORDS: compressible multicomponent flows; turbulence; shock waves; discontinuity capturing;
artificial diffusivity method; high-order methods; compact differences

1. INTRODUCTION

Understanding interface dynamics in compressible flows is of significant interest in a wide range


of both scientific and engineering applications. It is also true that the interface dynamics are often
coupled with shock waves and turbulence. The interaction between material interfaces, shock
waves and turbulence plays an important role in a broad area of applications, including supersonic
mixing and combustion, flame propagation in detonation, explosion, inertial confinement fusion,
and astrophysics.
Early computational works for such compressible multicomponent flows were dedicated to
avoiding spurious oscillations at fluid interfaces. The oscillations are usually generated by using

Correspondence to: Soshi Kawai, Center for Turbulence Research, Stanford University, 488 Escondido Mall, Stanford,
CA 94305-3035, U.S.A.

E-mail: skawai@stanford.edu

Contract/grant sponsor: The Basic Science Research Program of the Sumitomo Foundation
Contract/grant sponsor: AFOSR-MURI program; contract/grant number: FA9550-04-1-0387

Copyright q 2010 John Wiley & Sons, Ltd.


1208 S. KAWAI AND H. TERASHIMA

upwind-biased shock-capturing schemes in a strong conservative form of governing equations. The


primary reason for the spurious oscillations is the numerical artifact of smeared interfaces that
cannot maintain pressure equilibrium across the interfaces when the ratio of specific heats for each
fluid is different. Abgrall [1] and Karni [2] developed quasi- and non-conservative schemes to
prevent such numerical oscillations. Abgrall, for example, introduced a non-conservative form of
an advection equation for the ratio of specific heats that is coupled with a conservative form of the
Euler equations to maintain velocity and pressure equilibrium across the fluid interfaces. This quasi-
conservative scheme has been extended to more general equations of state [35]. However, as
expected, the quasi- and non-conservative schemes often suffer from poor mass conservation [6]
because they are not in the strong conservation form. It should be noted that although there are
several ideas to distinguish different fluids, such as using the ratio of specific heats, the mass
fraction, and level set functions, the choice does not essentially avoid the numerical oscillations.
Fedkiw et al. [7] introduced the ghost fluid method as an alternative way for computing compress-
ible multimaterial fluids using level set functions. Generally the method can prevent pressure
oscillations at the fluid interfaces by considering the ghost fluid to be thermodynamically identical
to the corresponding real fluid, extrapolating the discontinuous variables, such as entropy, to the
ghost fluid. Although the Euler equations in a conservative form are solved, the method is not
discretely conservative [8, 9] because of the extrapolation of the variables across the interface.
One of the advantages is the ease of implementation for existing solvers, whereas the accuracy is
limited to first-order at fluid interfaces due to the constant extrapolation of variables. A similar
idea of using the same equation of state across fluid interfaces was taken in the use of the pres-
sure evolution equation proposed by Karni [2] in order to capture strong shock waves with a
non-conservative scheme.
Using explicit marker points to track fluid interfaces (including shock waves and expansion
waves), Glimm et al. [10] developed the front tracking method for compressible multicomponent
flows. They construct the Riemann problems along the normal and the tangential directions at the
fluid interfaces and then extrapolate the solutions onto ghost nodes. Owing to the use of ghost nodes,
the spurious oscillations can be avoided in the method. Terashima and Tryggvason [11] recently
combined front tracking with a ghost fluid method for simulating fluid interfaces in compressible
flows while keeping their merits. Revisiting the idea of constructing the Riemann problems, several
researchers have attempted to use the exact Riemann solutions at fluid interfaces for robust and
accurate computations [1214], especially focusing on capturing gasliquid interfaces, with the
mass fraction model or the level set functions. In those methods the (exact) Riemann solutions are
introduced to nodes around the fluid interfaces as the appropriate status, resulting in suppressing
the numerical oscillations. However, introducing the Riemann solutions on specific nodes is time-
consuming work, especially when considering 3-D simulations. Generally the ghost fluid and front
tracking methods, which perform some treatments at interfaces, do not satisfy mass conservation
for each fluid due to the extrapolation procedure across the interface, although recent works [15, 16]
have been undertaken to improve the conservation property.
The numerical schemes used in the above work well in the sense of shock- and interface-
capturing for compressible multicomponent flows but are often too dissipative to resolve the
broad range of scales of turbulence and the interaction between turbulence and material inter-
faces with a reasonable grid resolution. This is primarily because of the dissipative conventional
low-order upwind-biased schemes employed in the simulations. One attempt at applying higher-
order schemes for multicomponent flows has been proposed by Johnsen and Colonius [17] using
weighted essentially non-oscillatory (WENO) schemes in a quasi-conservative form of equations.
The method does not generate the spurious oscillations at fluid interfaces. However, similar to the
numerical methods discussed above, the use of quasi-conservative schemes does not conserve the
mass of each species [6]. The lack of mass conservation for each fluid in these methods can be a
critical issue when considering mixing and combustion where mixing and chemical reactions occur
at material interfaces. Moreover, it should also be noted that Larsson et al. [18], Kawai and Lele
[19], Johnsen et al. [20] and Kawai et al. [21] have shown that the numerical dissipation introduced
by these upwind-biased schemes overwhelms a large range of scales even if the formal order of

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1209

accuracy of these schemes is high. These results illustrate that upwind-biased schemes (even if the
order of accuracy is high) are often too dissipative for use in direct numerical simulation (DNS)
or large eddy simulation (LES) to properly resolve the broad range of scales of turbulence using
limited computational resources.
In this paper, we investigate a simple methodology for a high-resolution scheme to be applied
to compressible multicomponent flows with shock waves. The method is intended for use with
DNS or LES of compressible multicomponent flows. The method dynamically adds non-linear
artificial diffusivity locally in space to capture different types of discontinuity (shock waves, contact
surfaces and material interfaces), whereas a high-order compact differencing scheme resolves the
broad range of scales in flows. A strong conservation form of the governing equations is employed
to satisfy the mass conservation for each species, which is an another advantage of the present
method compared with the quasi- and non-conservative schemes.
The performance of the present method is addressed through the one-dimensional isolated
material interface advection and two-fluid shock-tube problems and two-dimensional shockbubble
interaction and single-mode RichtmyerMeshkov instability problems. The numerical results are
compared with available experiments and earlier simulations. Although we assume that the fluid
is inviscid and immiscible throughout the paper for comparison with earlier simulations, the
extension of the proposed method to the NavierStokes equations is straightforward. Basically
only the additional process is to calculate the fluid transport coefficients (f , f , f , Df,k ). The
detailed study for simulating viscous and miscible fluids using the present method may be found
in Reference [22].

2. NUMERICAL FRAMEWORK

2.1. Governing equations


We describe the present numerical framework using the compressible NavierStokes equations
for the ease of explanation, although we assume inviscid and immiscible fluid in this study.
The compressible NavierStokes equations for an ideal non-reactive multicomponent gas are
*
+ (u) = 0, (1)
*t
*u
+ (uu+ pds) = 0, (2)
*t
*E
+ [Eu+( pds)uT ] = 0, (3)
*t
*Yk
+ (uYk )(Dk Yk ) = 0, (4)
*t
p 1
E= + uu, p = ( 1)e (5)
 1 2
where  is the total density, u is the velocity vector, p is the static pressure, E is the total energy,
T is the temperature,  is the ratio of specific heats for the fluid mixture, e is the internal energy,
 is the thermal conductivity, d is the unit tensor. Equation (4) is the transport equation for species,
where Dk is the species diffusion coefficient. We assume that the species are calorically perfect
gases in thermal equilibrium. The viscous stress tensor s (for a Newtonian fluid) is
 
s = (2S)+  23  ( u)d, (6)

where  is the dynamic (shear) viscosity,  is the bulk viscosity and S is the strain rate tensor,
S = 12 (u+(u)T ).

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1210 S. KAWAI AND H. TERASHIMA

In this study, the mass fraction Yk = k / is used to describe the fluid mixture where, for
two-fluid flows, Yk Y1 = 1Y2 . The ratio of specific heats for the mixture  is

Cp z i C p,i
 = = i , (7)

Cv i z i C v,i

where z i = Yi /Mi and Mi is the molar weight for specie i. C p,i and Cv,i are the specific heat
constants for specie i and are given by
i Ru Ru
C p,i = , Cv,i = , (8)
i 1 i 1

where Ru is the universal gas constant. We use Daltons law of partial pressures, p = i pi , and
assume that the molar weight for the species is uniform, which is the same assumption as the mass
fraction model in References [1, 8, 23]. If one assumes the uniform molar weight for the species,
Equation (7) can be rewritten as

Cp Yi C p,i
 = = i . (9)

Cv i Yi C v,i

2.2. Spatial differencing and filtering schemes


The governing equations in the strong conservative form are solved where the spatial derivatives
are evaluated using a sixth-order compact differencing scheme [24]. For any scalar quantity f , the
finite difference approximation to the first spatial derivative at node i, * f i /*l , is obtained by the
following formula:
* f i1 * f i * f i+1 f i+1 f i1 f i+2 f i2
 + + =a +b (10)
*l *l *l 2l 4l

where  = 13 , a = 14
9 and b = 9 for the sixth-order scheme. At boundary points i = 1 and 2, and
1

correspondingly at i = imax and imax1, second- and fourth-order one-sided formulas are utilized
that retain the tridiagonal form of the equation set [25].
The eighth-order (N = 4 in Equation (11)) low-pass spatial filtering scheme [24] is applied to the
conservative variables once in each direction after each final RungeKutta step to ensure numerical
stability:
N a
f fi1 + fi +f fi+1 =
n
( f i+n + f in ) (11)
n=0 2

where f is the solution vector, and f is filtered quantity. An eighth-order filter is obtained with
a0 =(93+70f )/128, a1 =(7+18f )/16, a2 =(7+14f )/32, a3 =(12f )/16, a4 =(1+2f )/128.
The f is a free parameter satisfying the inequality 0.5<f 0.5. In this range, as f is increased,
the filtering is more localized to the high wavenumbers. In the present study, f is fixed to 0.495
for all the cases examined in this paper. High-order one-sided formulas are used near-boundary
points at i = 14 and correspondingly at i = imax imax3 [26]. Detailed spectral responses of
these filters may be found in Reference [27]. The classical, four-stage, fourth-order, low-storage
form of the RungeKutta method [28] is used for temporal integration.

2.3. Localized artificial diffusivity (LAD) method


When central differencing schemes, such as high-order compact differencing schemes, are applied
to solve flows that involve steep gradients, e.g. due to shock waves, contact surfaces or material
discontinuity, non-physical spurious oscillations that make the simulation unstable are generated.
A key issue here is the identification of an efficient algorithm that ensures the removal of the
non-physical spurious oscillations without affecting the physical motions, e.g. resolved scales of
turbulence.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1211

Our discontinuity capturing scheme is based on adding grid-dependent artificial fluid transport
coefficients to the coefficients appearing in Equations (2), (3), (4) and (6) proposed by Cook [29],
 = f + , (12)

 = f + , (13)
 = f + , (14)
Dk = Df,k + Dk , (15)
where the f subscripts and asterisks denote fluid and artificial transport coefficients, respectively.
Since we assume that the fluid is inviscid and immiscible throughout the paper, the fluid transport
coefficients (f , f , f , Df,k ) are set to zero in this study. The artificial fluid transport coefficients
are intended to automatically vanish in smooth, well-resolved regions of the flow and provide
damping in non-smooth, unresolved regions to capture different types of discontinuity. According
to Reference [29]  ,  ,  and Dk serve as a multi-purpose model for subgrid-scale transport,
shock capturing, contact surface capturing and material discontinuity capturing.
Here, for generality, we describe the localized artificial diffusivity (LAD) method on a multi-
dimensional generalized coordinate system although uniformly spaced Cartesian meshes are used
in this study. We modeled the LAD based on References [19, 21] as
 
 r 
 3 * S r 2 
 = C   l l, , (16)
l=1 *lr 
 
3 *r u  H ( u)( u)2
 2 
 = C  f sw  rl l,  , f sw = (17)
l=1 *lr
( u)2 +( u)2 +
 
c 3 *r e 
 
 = C
s r
  l,  , (18)

T l=1 *l r l 
 
 r 
 3 * Yk r 
Dk = C Dk cs   l,D k  +C Yk cs [Yk 1]H (Yk 1)Yk [1 H (Yk )]Yk . (19)
l=1 *rl l 

The method is the extension of previous works on the artificial diffusivity [29, 30]. C , C , C ,
C Dk and CYk are dimensionless user-specified constants. cs is the speed of sound and H is the
Heaviside function. l refers to generalized coordinates , and
when l is 1, 2 and 3, respectively.
= 1032 is a small positive constant to prevent division by zero in the region where both u
and u are zero. l and l, are the grid spacing in the computational space and physical
space. Usually l is set to 1. We define the grid spacing l, as follows:
     
 q   e   Yk 

l, = |Dxl |, l, = Dxl 
, l, = Dxl 
, l,Dk = Dxl . (20)
|q|  |e|  |Yk | 
The length scalings used here are the natural extension based on Reference [30]. Dx l is the
local displacement vector along the grid line in the l direction and is defined as Dxl =
((xi+1 xi1 )/2, (yi+1 yi1 )/2, (z i+1 z i1 )/2)T , where the nodes in the l direction are
indexed by i. Thus, l, , l, and l,Dk are the grid spacing in the l direction perpendicular to
the shock waves, contact surfaces and material discontinuities. Yk in Equation (19) is the grid
spacing in the physical space defined by
 
3  *r Yk 
  l,D
l=1 
*rl  k
Yk =   r 2 . (21)
3 * Yk
l=1 +
*rl

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1212 S. KAWAI AND H. TERASHIMA

If r is sufficiently high, the high-wavenumber-biased (k r ) artificial diffusivities,  ,  and Dk ,


are localized in the regions of shock, contact and material discontinuities, respectively, to capture
the discontinuities by smearing them over a numerically resolvable scale. It should be noted that
the motivation of adding the artificial shear viscosity  is to model the terms of unresolved
subgrid-scale eddies in LES. Since we assume that the fluid is inviscid and immiscible (f = f =
f = Df,k = 0) and turbulence is not considered throughout the paper, we set C = 0.0. The detailed
analysis of the performance of  ,  and Dk for shock, contact and material discontinuities in
single fluids may be found in References [19, 21].
In the present study r =4 is adopted. The fourth-derivative terms in Equations (16)(19) are
evaluated using a fourth-order central differencing scheme [24]:

4
* fi f i+3 9 f i+1 +16 f i 9 f i1 + f i3 f i+2 4 f i+1 +6 f i 4 f i1 + f i2
=b +a , (22)
*l4 6l4 l4

where a = 2 and b = 1. At near-boundary points i = 1 and 2, and correspondingly at i = imax and


imax1, one-sided second-order explicit scheme [19] is used. A second-order central scheme is
used for i = 3 and imax2. The overbar in Equations (16)(19) denotes an approximate truncated-
Gaussian filter [31] to remove the cusps introduced by the high-order derivative operators.
The dimensionless user-specified constants in Equations (16)(19) are set to C = 0.0, C = 1.75,
C = 0.01, C D = 0.01 and CY = 100 in this study. The f sw in Equation (17) is the switching
function that is designed to help localize the artificial bulk viscosity only near shock waves.
The main advantages of this method are its simple formulation, low computational cost, ease of
implementation, capacity to localize artificial dissipation in regions, where an unresolved high
wavenumber content of the flowfield exists (high-resolution characteristics of a high-order compact
scheme are preserved in smooth region), and no weighting/hybrid procedures. It should also be
noted that the method may be coupled with any type of numerical schemes. Further details of the
LAD method may be found in References [19, 21].

2.4. Comments on the computational cost


There is a considerable difficulty in comparing the computational cost between different numer-
ical methods since the computational cost may vary depending on the implementation, problem,
platform, etc. Thus, qualitative comments on the computational cost are discussed in this section.
Reference [20] provided the quantitative estimates of the number of floating point operations
per grid point for popular upwind-biased high-order schemes (fifth- and seventh-order WENO
schemes [32]) and an original artificial diffusivity method [29] although the estimates are based
on single-fluid codes. They showed that the computational costs of the fifth- and seventh-order
WENO schemes for the spatial discretization are approximately 1.6 and 3.3 times higher than that
of the artificial diffusivity method. In the cost estimation of the artificial diffusivity method, the
convective terms are in a skew-symmetric form and the first and second derivatives are evaluated
by a pentadiagonal compact differencing scheme. In the present numerical framework, a strong
conservation form of governing equations is solved with a tridiagonal compact differencing scheme,
which requires approximately half of the computational cost that is required for the method
described in Reference [20]. Also in the present method, additional computational efficiency benefit
comes from approximating the double Laplacians used in the original artificial diffusivity method
with the direct implementation of fourth derivatives (Equation (22)) as discussed in Reference [19].
Therefore, in total the present numerical framework will require roughly less than 13 and 16 of the
computational costs that are required for the fifth- and seventh-order WENO schemes, respectively.
We finally note that the parallel efficiency and the cost of computing the viscous terms are
not considered in the estimation. Also as noted in Reference [20], the large difference in the
computational cost between fifth- and seventh-order WENO scheme comes from the calculation
of the smoothness indicators.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1213

3. RESULTS

3.1. One-dimensional problems


We first investigate the effect of initial interface conditions on numerical results through one-
dimensional test cases since initial start-up errors [33] due to the initial sharp numerical discon-
tinuity are expected to be severe for central differencing schemes because of the non-dissipative
characteristics. Practically, except sharp interface methods such as the front tracking method, the
interfaces on a given mesh are not sharp one-point jumps but smooth since they are either resolved,
modeled (e.g. by a subgrid-scale model for scalar dissipation) or captured in simulations. Therefore,
the smooth initial profile is a reasonable initial condition.

3.1.1. Initial interface conditions. The initial interface conditions for the sharp (one-point jump)
and smooth interface profiles considered in this study are shown in Figure 1. For the smooth
interface, any primitive quantity q (, u, p and Y ) near the initial interface is imposed by the
following formula:

q = ql (1 f sm )+qr f sm , (23)

f sm = (1+erf[R/ ])/2, (24)

where subscripts l and r denote the left- and right-interface conditions, erf is the error function, R
is the distance from the initial material interface where f sm = 0.5, and is the parameter to impart
an initial thickness to the material interface. We set = C x 104 so that the initial interface is
represented by the same number of grid points even when the grid resolution is changed, where
x is the grid spacing. Thus, the thickness of the initial interface is sharpened in physical space
while retaining the same number of grid points across the interface when the grid resolution is
increased. Figure 1 shows the smooth and sharp functions in terms of grid index and the interface
thickness for C ranging between 0.5 and 4.0. The initial interface thickness is defined by
 f sm
= 
x * f sm 
x
*x max

1 7
Initial interface thickness

6
Smooth function (fsm)

0.8

0.6 5

0.4 4

0.2 3

0 2

0 5 10 15 20 0 1 2 3 4
(a) Grid index (b) C

Figure 1. Initial interface profiles for sharp (one-point jump) and smooth interfaces. Smooth function:
f sm = (1+erf[R/ ])/2, = C x 104 . Initial interface thickness: /x =  f sm /(x(* f sm /*x))|max .
Solid line with circle, smooth interface with C = 3.0; solid line with triangle, smooth interface with
C = 1.5; solid line with cross, sharp interface; dotted lines, smooth interfaces with C = 0.54.0: (a) Smooth
and sharp functions and (b) interface thickness.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1214 S. KAWAI AND H. TERASHIMA

1.2 1.08

1.06
1
1.04
0.8
1.02

Velocity
Density

0.6 1

0.98
0.4
0.96
0.2
0.94

0 0.92
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x

1.08 1.65

1.06 1.6

1.55
1.04
Pressure

1.5

1.02
1.45

1
1.4

0.98 1.35
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x

Figure 2. One-dimensional isolated material interface advection in a uniform flow at t = 1.0. Solid line,
exact for smooth initial interface with C = 3.0; dotted line, exact for sharp initial interface; circle, smooth
initial interface with C = 3.0; triangle, smooth initial interface with C = 1.5; cross, sharp initial interface.

where  f sm is the f sm jump across the initial interface; thus,  f sm = 1.0. The thickness denotes an
approximate number of grid points across the initial interface. As C increases, the initial interface
is spread over more grid points.

3.1.2. Isolated material interface problem. The first test case is an inviscid one-dimensional
isolated interface advection in a uniform flow introduced by Johnsen and Colonius [17]. Initial
conditions are:  = 1.0, u = 1.0, p = 1.0 and  = 1.6 for |x|0.25, and  = 0.1, u = 1.0, p = 1.0
and  = 1.4 for |x|>0.25. Both the smooth and sharp interfaces across the grid points are initially
imposed. Simulations are performed on a uniformly spaced grid with 100 grid points in the region
of 0.5x<0.5 (x = 0.01). Periodic boundary conditions are applied on the boundaries in the
x-direction. This simple test case allows us to investigate the occurrence of the spurious oscillations
at fluid interfaces as reported in References [8, 17].
Figure 2 shows the comparison between the exact solution and the numerical simulations with the
smooth (C = 1.5 and 3.0) and sharp initial interfaces for density, velocity, pressure and  profiles
at t = 1.0, when the interfaces convect the distance of 1.0 and return to the original positions.
The maximum wiggles amplitudes of the density, velocity and pressure for C ranging between
0.0 (sharp initial interface) and 4.0 are shown in Figure 3. The sharp initial interface shows spurious
oscillation throughout the domain. As C increases, the amplitude of the spurious oscillations
decreases. The results of the smooth initial interface with C = 3.0 show good agreement with the

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1215

Maximum wiggles amplitude (%)


6

0
0 1 2 3 4
C

Figure 3. Maximum wiggles amplitude (%) in terms of C for the one-dimensional isolated material
interface advection in a uniform flow at t = 1.0. Solid line with circle, density; dashed line with triangle,
velocity; dotted line with cross, pressure.

1.2 1.2

1 1
Primitive variables

Primitive variables

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
(a) x (b) x

Figure 4. Initial start-up errors in primitive variables (, u and p) for smooth (C = 3.0) and sharp initial
interface at t = 0.1. Solid line, exact for density; dotted line, exact for velocity and pressure; circle, density;
triangle, velocity; cross, pressure: (a) smooth initial interface with C = 3.0 and (b) sharp initial interface.

exact solution and do not show significant spurious oscillations as reported in References [8, 17].
For C = 3.0 (initial interface thickness 5.5 /x), the maximum wiggles amplitudes are less than
0.08%.
The primitive variables (, u and p) at t = 0.1 obtained by the smooth (C = 3.0) and sharp
initial interfaces are shown together with the exact solutions in Figure 4 in order to investigate the
source of the spurious oscillations. Over this time the isolated material convects a distance of 0.1.
As clearly shown, the sharp initial interface suffers from large initial start-up errors that propagate
upstream and downstream with the characteristic speeds of u +cs , u and u cs . The start-up errors
remain for a long period of time due to the non-dissipative characteristics of the compact scheme
and contaminate the solution, whereas the smooth initial interface does not show significant start-
up errors. The large start-up errors are a common feature for central difference schemes when the
sharp discontinuity is imposed initially. The result of the smooth initial interface illustrates that

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1216 S. KAWAI AND H. TERASHIMA

4e-7 4e-7

2e-7 2e-7

Mtotal(t)/Mtotal(t=0)-1
Miso(t)/Miso(t=0)-1

0 0

-2e-7 -2e-7

-4e-7 -4e-7
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) t (b) t

Figure 5. Time-history of normalized mass conservation error. Solid line, smooth initial interface with
C = 3.0; dotted line, sharp initial interface: (a) mass of isolated material and (b) total mass.

the present method does not generate significant spurious oscillations at the interfaces if the initial
start-up errors are properly avoided.
Figure 5 shows the time-history of the error in mass conservation obtained by the smooth
(C = 3.0) and sharp initial interfaces where the t = 0 subscripts denote the value of initial condition.
As expected, both mass and total mass are conserved within the admissible error for both the
smooth (order of 1013 ) and sharp (order of 107 ) initial interface, whereas the smooth initial
interface shows smaller error compared with the sharp interface. Mass conservation is a clear merit
in the present method relative to the quasi- and non-conservative schemes [1, 6].

3.1.3. Multicomponent shock-tube problem. The one-dimensional two-fluid modified Sod shock-
tube problem introduced by Abgrall and Karni [8] is computed as the second test case. Initial
left- and right-side conditions are: l = 1.0, u l = 0.0, pl = 1.0 and l = 1.4 for x0, and r = 0.125,
u r = 0.0, pr = 0.1 and r = 1.6 for x>0. Simulations are performed on a uniformly spaced grid with
401 grid points in the region of 0.5x0.5 (x = 0.0025).
Figure 6 shows a comparison between the exact solution for the sharp initial conditions and the
results of the smooth (C = 3.0) and sharp initial conditions for the density, velocity, pressure and
 at t = 0.2. Large inaccuracies as reported in Reference [8] are not observed in all the smooth and
sharp initial conditions. The results show good agreement with the exact solutions, in which the
shock and the contact discontinuities are captured without significant spurious oscillations. Very
similar results are obtained by the smooth and the sharp initial conditions. The shock wave is
captured over approximately four grid points for both cases. The smooth initial condition alleviates
the significant over- and under-shoots at the material interface and at the edge of the expansion
region, which can be seen in the sharp initial condition case. It is known that the initial start-up
errors generated by the initial discontinuity cause these over- and under-shoots [33]. Similarly to
the isolated material interface problem, the initial start-up errors are reduced with increasing C .
Although not shown here, the Sod [34] and Lax [35] problems with (l , r ) = (1.4, 1.6) and (1.2,
1.6) were simulated using the present scheme, and the results show good agreement with the exact
solutions.

3.1.4. Multicomponent stiff shock-tube problem. The final test case in one-dimensional is the two-
fluid shock-tube problem with much larger pressure ratio of pl / pr = 1000 [8]. This test case can be
used to investigate the robustness of our numerical approach. Initial left- and right-side conditions
are: l = 1.0, u l = 0.0, pl = 100.0 and l = 1.4 for x0, and r = 1.0, u r = 0.0, pr = 0.1 and l = 1.6
for x>0. Simulations are performed on a uniformly spaced grid with 401 grid points in the region
of 0.5x0.5 (x = 0.0025).

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1217

1 1

0.8
0.8

0.6

Velocity
Density

0.6

0.4
0.4
0.2

0.2
0
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x
1
1.6

0.8
1.55
Pressure

0.6
1.5

0.4
1.45

0.2

1.4
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x

Figure 6. One-dimensional two-fluid modified Sod shock-tube problem at t = 0.2. Solid line, exact for
sharp initial condition; circle, smooth initial condition with C = 3.0; triangle, sharp initial condition.

The results of smooth (C = 3.0) and sharp initial conditions for density, velocity, pressure
and  at t = 0.035 are shown in Figure 7, compared with the exact solution for the sharp initial
condition. All the results show good agreement with the exact solution although slight overshoots
appear in the density near the shock and material discontinuities. The amplitudes of the overshoots
in the density normalized by the density jump are 1.3 and 2.0% at the shock wave and the material
interface, respectively. A similar level of the amplitude of the overshoot is also observed in the
single-fluid shock-tube problem using the LAD method [19]. Owing to the initial start-up errors,
the sharp initial condition shows the spurious oscillations at the edge of the expansion region.
The oscillations are reduced with increasing C and the smooth initial condition with C = 3.0 does
not show the significant oscillations. We note that the present numerical approach with the smooth
initial conditions with C = 3.0 was able to predict the increased pressure ratio case pl / pr = 10 000
( pl = 1000.0 and pr = 0.1) although not shown here. This indicates the robustness of the present
method.
Through the one-dimensional test problems, it has been shown that the present method captures
the unsteady shock waves and material discontinuities without significant spurious oscillations
if the initial start-up errors are properly avoided. If the sharp interface is imposed on the initial
conditions, the method suffers from start-up errors, which is a common feature of central difference
schemes. We note that the amplitude of spurious oscillations in terms of the initial smoothness
as shown in Figure 3 may be changed if a different spatial discretization scheme is used. The
required initial smoothness depends on the resolution characteristics of an employed numerical

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1218 S. KAWAI AND H. TERASHIMA

4.5
6
4

3.5 5

3 4

Velocity
Density

2.5 3
2
2
1.5
1
1

0.5 0
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x

100
1.6

80

60
Pressure

1.5

40

20

1.4
0
-0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4
x x

Figure 7. One-dimensional two-fluid stiff shock-tube problem at t = 0.035. Solid line, exact for sharp
initial condition; circle, smooth initial condition with C = 3.0; triangle, sharp initial condition.

scheme (a low-resolution scheme needs more smooth initial interface conditions). In the following
two-dimensional test cases based on the one-dimensional results, smooth material interface with
C = 3.0 (interface thickness 5.5 /x) is imposed initially to avoid start-up errors.

3.2. Shockbubble interaction


The first test case in two-dimensional is the shockbubble interaction problem of Hass and Sturte-
vants experiment [36]. This problem involves shock waves, material interfaces and their interaction
and has been used by several researchers [6, 7, 11, 3740] to demonstrate the capability of their
numerical schemes in compressible multicomponent flows.
The initial flowfield and computational domain are described in Figure 8. The computational
domain extends 2.5x/D4.0 and 0y/D1.78 where D is the diameter of the initial helium
bubble and the location of x/D = 0 is the right side of the bubble edge. Initially, a Mach 1.22
normal shock wave in air is imposed at x/D = 0.5. The center of the helium bubble is x/D = 0.5
and y/D = 0.89. Non-dimensional initial flow conditions are:

 = 1.0, u = 0.0, v = 0.0, p = 1.0/air , air = 1.4 for pre-shock air

 = 1.3764, u = 0.3336, v = 0.0, p = 1.5698/air , air = 1.4 for post-shock air

 = 0.1819, u = 0.0, v = 0.0, p = 1.0/air , helium = 1.648 for helium bubble.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1219

6.5D

pre-shock air post-shock air

Shock wave

1.78D
Helium
bubble

D 0.5D 2D
x

Figure 8. Schematic of initial flowfield and computational domain.

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure 9. Time-series snapshots of M = 1.22 shockbubble interaction with x/D = y/D = 0.0025.
Non-linear function of density gradient magnitude, = exp(||/||max ), contours from 1.0 to 1.7.
Domain size of 1.3D 0.65D is presented: (a) t = 0 s; (b) t = 32 s; (c) t = 62 s; (d) t = 102 s;
(e) t = 240 s; (f) t = 427 s; (g) t = 674 s; and (h) t = 983 s.

The smooth initial two-dimensional material interface is initially imposed for the helium bubble to
avoid initial start-up errors by using the smooth function of Equations (23) and (24) with C = 3.0.
Three different levels of mesh resolution x/D = y/D = 0.01, 0.005, 0.0025 are employed.
The corresponding number of grid points for the three mesh resolutions are 651179, 1301357
and 2601713, respectively. The computational time-step size is set to tcs,air /D = 105 x/D.
Slip-wall boundary condition is imposed on the upper and lower boundaries. The left and right
boundary conditions are extrapolated. The dimensional quantities discussed in the results are
computed by D = 50 mm and cs,air = 331.6 m/s [39] where air = 1.29103 kg/m3 and pair =
1.01325105 Pa are assumed.
Figure 9 shows the time evolution of the shockbubble interaction with the finest mesh resolution,
using the non-linear function of density gradients, = exp(||/||max ). The upper half of the
bubble is presented, whereas the simulation is performed on the whole bubble structure. During
the evolution of the shockbubble interaction, the shock and material discontinuities are captured
well without significant spurious oscillations. Although not shown here, other flow quantities,
such as density, velocity and pressure, also do not show detrimental oscillations. A qualitative
comparison can be made by comparing the simulation with the experiment of Hass and Sturtevant
[36], although precise quantitative agreement with the experiment cannot be expected because of
the inviscid and immiscible fluid used in the simulation. The simulation qualitatively reproduces
the very similar dynamics of the helium bubble as observed in the experiment. The shock wave
moves faster inside the bubble than the outside air due to the difference in the speed of sound.
The bubble interface becomes flat once the shock wave hits the bubble and a jet forms due to the
baroclinic torque, resulting in piercing of the bubble. The KelvinHelmholtz instability induces
vortex generation along the bubble interface.
Figure 10 shows the spacetime diagrams for three characteristic interface positions (upstream,
downstream and jet) obtained by the three levels of mesh resolution. The simulations by Quirk

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1220 S. KAWAI AND H. TERASHIMA

800 300
Upstream Upstream

600
Jet
200

Jet
400
Downstream

100
200
Upstream Downstream

Downstream Jet
0 0
0 50 100 150 0 20 40 60 80
(a) x-position (mm) (b) x-postion (mm)

Figure 10. Spacetime diagrams for three characteristic interface points. Solid line, x/D = 0.0025;
dashed-dot line, x/D = 0.005; dotted line, x/D = 0.01; circle, simulations by Quirk and Karni [38] and
Bagabir and Drikakis [39]; triangle, simulation by Terashima and Tryggvason [11]: (a) Overall motion
and (b) Close-up of early stage motion.

and Karni [38] and Bagabir and Drikakis [39] and simulation by Terashima and Tryggvason [11]
are also included for comparisons. The results of the three levels of mesh resolution are nearly
identical and show the convergence of the global bubble motions, suggesting that the smooth
initial interface employed in this study does not affect the global bubble motion. The results are
in excellent agreement with the earlier simulations.
Figure 11 shows the structures of the bubble interface and vorticity at t = 427, 674 and 983 s
obtained by the three levels of mesh resolution. Consistent with the previous finding [19], the
discontinuities are captured approximately by a fixed number of grid points. Thus, the discon-
tinuities are sharpened by refining the mesh. Because of the absence of physical viscosity and
diffusivity in the current simulations (most of the earlier simulations also assume that the fluid
is inviscid and immiscible), a converged solution for small scales cannot be expected. Therefore,
small-scale structures are being generated as the mesh resolution is refined. These resolved scales
highly depend on the resolution and dissipative characteristics of numerical schemes used in the
simulation.
As compared with the earlier simulation using a fifth-order finite volume WENO scheme [6], the
present method resolves the smaller vortex structure along the interface under almost the same mesh
resolution (see Figure 12 in p. 729, Reference [6]). The present method also gives similar bubble
structure compared with a fifth-order finite difference WENO scheme with more than twice the
mesh resolution (see Figure 7 in p. 132, Reference [40]). The qualitative comparison between the
present method and the high-order WENO schemes illustrates the superior performance of the
present method in resolving the small eddies along the interface although the smooth initial interface
is used in the simulation. Note that a sharp interface is initially imposed for the earlier simulations
using the WENO schemes. It should also be noted that the fifth-order WENO scheme uses seven
points in the stencil to evaluate a spatial derivative and requires higher computational cost than the
present compact scheme with the artificial diffusivity method as discussed in Section 2.4. Similar
to the one-dimensional isolated interface problem in Section 3.1.2, the error in mass conservation
for each species, Mk /Mk,t=0 1, is very small for the three levels of mesh resolution (order of
109 for x/D = y/D = 0.01 and 1012 for x/D = y/D = 0.005 and 0.0025 at t = 1000 s).

3.3. RichtmyerMeshkov instability


Another two-dimensional test case is the single-mode RichtmyerMeshkov instability problem.
This problem was performed experimentally by Brouillette and Sturtevant [41, 42] and Collins and

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1221

Figure 11. Time evolution of M = 1.22 shockbubble interaction with three different levels of mesh reso-
lution x/D = y/D = 0.0025, 0.005 and 0.01 at t = 427, 674 and 983 s (from left to right). Non-linear
function of density gradient magnitude, = exp(||/||max ), contours from 1.0 to 1.7 for upper half.
Vorticity, D/cs,air , contours from 20 to 20 for lower half. Domain size of 1.3D 1.3D is presented:
(a) x/D = y/D = 0.0025; (b) x/D = y/D = 0.005; and (c) x/D = y/D = 0.01.

Jacobs [43] under similar flow conditions. Several researchers [4, 11, 13, 44] have also simulated
this problem to demonstrate the capability of their numerical schemes.
Figure 12 shows the schematic of the initial flowfield and computational domain. The compu-
tational domain extends 0.0x/4.0 and 0.0y/1.0 where  is the initial perturbation wave-
length. The initial perturbed interface is generated by
x  y 
= 2.90.1 sin 2 +0.25 . (25)
 
Initially, a Mach 1.24 normal shock wave in air is imposed at x/ = 3.2 location. Non-dimensional
initial flow conditions are
 = 1.0, u = 0.0, v = 0.0, p = 1.0/air , air = 1.4 for pre-shock air,
 = 1.4112, u = 0.3613, v = 0.0, p = 1.6272/air , air = 1.4 for post-shock air,
 = 5.04, u = 0.0, v = 0.0, p = 1.0/air , SF6 = 1.093 for SF6 .
The smooth initial two-dimensional material interface with C = 3.0 is initially imposed to avoid
initial start-up errors. Three different levels of mesh resolution x/=y/= 128 1 1
, 256 1
and 512
are employed. The corresponding number of grid points for the three mesh resolutions are

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1222 S. KAWAI AND H. TERASHIMA

Bubble Pre-shock Post-shock air


air

Spike Shock wave

SF6 Air

Figure 12. Schematic of initial flowfield and computational domain.

Figure 13. Time evolution of RichtmyerMeshkov instability with three different levels of mesh resolution
x/ = y/ = 5121
, 256
1 1
and 128 at t = 2.75, 5.50, 8.25 and 11.00 (from left to right). Non-linear function of
density gradient magnitude, = exp(||/||max ), contours from 1.0 to 1.7. Domain size of 1.11.0
is presented: (a) x/ = y/ = 1/512; (b) x/ = y/ = 1/256; and (c) x/ = y/ = 1/128.

513128, 1025256 and 2049512, respectively. The computational time-step size is set to
tcs,air /x/ = 0.002 1281
. Periodic boundary conditions are imposed on the upper and lower
boundaries. A non-reflection boundary condition [45] is used for the left boundary. The right
boundary conditions are extrapolated.
Figure 13 shows the time evolution of the RichtmyerMeshkov instability obtained by the
simulation with the three mesh resolutions at the non-dimensional time t = 2.75, 5.50, 8.25 and
11.00. Contours of the non-linear function of density gradient magnitude, = exp(||/||max ),
are shown. After the shock wave hits the initial perturbed interface, the interface deforms due to
the baroclinic generation of vorticity and the induced velocity. During the evolution process, the
roll-up of the vortex structures along the interface is observed whereas the shock and material
discontinuities are captured without significant spurious oscillations. Similar to the shockbubble
interaction problem, the interface becomes sharper by a constant factor as the mesh is refined,
and small scales in the flow keep generating due to the absence of fluid viscosity and diffusivity
whereas the large-scale structures, such as the spike and the bubble, show a similar behavior. The
present scheme resolves the roll-up of the small vortex structures along the distorted interface with

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1223

Spike

x-position (x/)
Bubble

Thickness of mixing layer


0
0 2 4 6 8 10
t

Figure 14. Spacetime diagrams for three characteristic interface points. Solid line, x/ = 5121
; dashed-dot
line, x/ = 256
1
; dotted line, x/ = 128
1
; circle, simulations using front-tracking/ghost-fluid method [11].

increasing mesh resolution, whereas the ninth-order WENO scheme with the same mesh resolution
under similar flow conditions (but single-fluid (gamma) formulation) show the absence of these
small structures (see Figure 2 in p. 810, Reference [44]). A sophisticated level set approach with
x/ = y/ = 200 1
also does not show the small structures along the interface (see Figure 10 in
p. 516, Reference [13]). These comparisons illustrate that the proposed method has less numerical
dissipation, showing the advantage of the present method for simulating flows involving shock
waves, material interfaces, turbulence and their interactions.
The time evolution of the location of the spike, the bubble and the thickness of the mixing layer
(defined as a distance between the spike and bubble) obtained by the three mesh resolutions is
shown in Figure 14. We define the YSF6 = 0.5 location as the material interface. The result using
the front-tracking/ghost-fluid method [11], which has demonstrated its capability through several
compressible multicomponent flow problems, is also included for comparisons. The results of the
three mesh resolutions are nearly identical and are in good agreement with the simulation using
the front-tracking/ghost-fluid method.

4. CONCLUSIONS

A simple methodology for a high-resolution scheme to be applied to compressible multicomponent


flows with shock waves has been investigated. The method dynamically adds non-linear artificial
diffusivity locally in space to capture different types of discontinuities (shock waves, contact
surfaces and material interfaces), whereas a high-order compact differencing scheme resolves the
broad range of scales in flows. The performance of the present method was addressed through
the one-dimensional isolated material interface advection and two-fluid shock-tube problems and
the two-dimensional shockbubble interaction and single-mode RichtmyerMeshkov instability
problems.
Through the one-dimensional problems, it was shown that the present method captures the
unsteady shock and material discontinuities without significant spurious oscillations if initial start-
up errors are properly avoided. If a sharp discontinuity is imposed for the initial conditions, the
method suffers from start-up errors, which is a common feature for central difference schemes.
The results in the two-dimensional shockbubble interaction and the RichtmyerMeshkov insta-
bility show good agreement with several earlier simulations and experiments quantitatively and

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
1224 S. KAWAI AND H. TERASHIMA

qualitatively. The qualitative comparison between the present method and high-order WENO
schemes illustrates the superior performance of the present method in resolving the broad range of
scales in the flows even if the smooth initial interface is used in the simulation. Thus the method
has an advantage for simulating flows involving shock waves, material interfaces, turbulence and
their interactions. It should be noted that the present compact scheme with the localized artificial
diffusivity method is expected to require less computational cost than high-order WENO schemes.
Mass conservation for each species is satisfied due to the strong conservation form of the
governing equations employed in the method.

ACKNOWLEDGEMENTS
This work is supported by the Basic Science Research Program of the Sumitomo Foundation. The first
author also acknowledges support from AFOSR-MURI (Grant FA9550-04-1-0387) program. The present
code is based on the extension to the code FDL3DI provided by Dr M. R. Visbal, whom the authors
thank for this. They also gratefully acknowledge Dr E. Johnsen for providing the exact solution for the
one-dimensional two-fluid shock-tube problems and valuable discussions.

REFERENCES
1. Abgrall R. How to prevent pressure oscillations in multicomponent flow calculations: a quasi conservative
approach. Journal of Computational Physics 1996; 125(1):150160.
2. Karni S. Hybrid multifluid algorithms. SIAM Journal on Scientific Computing 1996; 17(5):10191039.
3. Shyue KM. An efficient shock-capturing algorithm for compressible multicomponent problems. Journal of
Computational Physics 1998; 142(1):208242.
4. Saurel R, Abgrall R. A simple method for compressible multifluid flows. SIAM Journal on Scientific Computing
1999; 21(3):11151145.
5. Allaire G, Clerc S, Kokh S. A five-equation model for the simulation of interfaces between compressible fluids.
Journal of Computational Physics 2002; 181(2):577616.
6. Johnsen E. Spurious oscillations and conservation errors in interface-capturing schemes. Annual Research Briefs
2008, Center for Turbulence Research, NASA Ames and Stanford University, 2008; 115126. Available from:
http://www.stanford.edu/group/ResBriefs/ARB08.html.
7. Fedkiw RP, Aslam T, Merriman B, Osher S. A non-oscillatory Eulerian approach to interfaces in multimaterial
flows (the ghost fluid method). Journal of Computational Physics 1999; 152(2):457492.
8. Abgrall R, Karni S. Computations of compressible multifluids. Journal of Computational Physics 2001;
169(2):594623.
9. Liu TG, Khoo BC, Yeo KS. Ghost fluid method for strong shock impacting on material interface. Journal of
Computational Physics 2003; 190(2):651681.
10. Glimm J, Grove J, Li X, Shyue K, Zeng Y, Zhang Q. Three-dimensional front tracking. SIAM Journal on
Scientific Computing 1998; 19(3):703727.
11. Terashima H, Tryggvason G. A front-tracking/ghost-fluid method for fluid interfaces in compressible flows.
Journal of Computational Physics 2009; 228(11):40124037.
12. Hu XY, Khoo BC. An interface interaction method for compressible multifluids. Journal of Computational
Physics 2004; 198(1):3564.
13. Nourgaliev RR, Dinh TN, Theofanous TG. Adaptive characteristics-based matching for compressible multifluid
dynamics. Journal of Computational Physics 2006; 213(2):500529.
14. Chang CH, Liou MS. A robust and accurate approach to computing compressible multiphase flow: stratified flow
mode and AUSM+ -up scheme. Journal of Computational Physics 2007; 225(1):840873.
15. Glimm J, Li X, Liu Y, Xu Z, Zhao N. Conservative front tracking with improved accuracy. SIAM Journal on
Numerical Analysis 2004; 41(5):19261947.
16. Nguyen D, Gibou F, Fedkiw R. A fully conservative ghost-fluid method and stiff detonation waves. Twelfth
International Detonation Symposium, San Diego, CA, U.S.A., Institute of Technical Physics, August 2002.
17. Johnsen E, Colonius T. Implementation of WENO schemes in compressible multicomponent flow problems.
Journal of Computational Physics 2006; 219(2):715732.
18. Larsson J, Lele SK, Moin P. Effect of numerical dissipation on the predicted spectra for compressible turbulence.
Annual Research Briefs 2007, Center for Turbulence Research, NASA Ames and Stanford University, 2007;
4757. Available from: http://www.stanford.edu/group//ctr/ResBriefs/ARB07.html.
19. Kawai S, Lele SK. Localized artificial diffusivity scheme for discontinuity capturing on curvilinear meshes.
Journal of Computational Physics 2008; 227(22):94989526.
20. Johnsen E, Larsson J, Bhagatwala AV, Cabot WH, Moin P, Rawat PS, Shankar SK, Sjogreen B, Yee HC, Zhong X
et al. Assessment of high-resolution methods for numerical simulations of compressible turbulence with shock
waves. Journal of Computational Physics 2010; 229(4):12131237.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld
A HIGH-RESOLUTION SCHEME FOR COMPRESSIBLE MULTICOMPONENT FLOWS 1225

21. Kawai S, Shankar SK, Lele SK. Assessment of localized artificial diffusivity scheme for large-eddy simulation
of compressible turbulent flows. Journal of Computational Physics 2010; 229(5):17391762.
22. Shankar SK, Kawai S, Lele SK, Numerical simulation of mixing in compressible multicomponent shock accelerated
flows using localized artificial diffusivity method. AIAA Paper 2010-0352, AIAA, January 2009.
23. Larrouturou B. How to preserve the mass fractions positivity when computing compressible multi-component
flows. Journal of Computational Physics 1991; 95(1):5984.
24. Lele SK. Compact finite difference schemes with spectral-like resolution. Journal of Computational Physics 1992;
103(1):1642.
25. Gaitonde DV, Visbal MR. High-order schemes for NavierStokes equations: algorithm and implementation into
FDL3DI. AFRL-VA-WP-TR-1998-3060, Air Force Research Laboratory, 1998.
26. Gaitonde DV, Visbal MR. Pade-type higher-order boundary filters for the NavierStokes equations. AIAA Journal
2000; 38(11):21032112.
27. Gaitonde DV, Shang JS, Young JL. Practical aspects of high-order accurate finite-volume schemes for
electromagnetics. AIAA Paper 97-0363, AIAA, 1997.
28. Fyfe DJ. Economical evaluation of RungeKutta formulae. Mathematics of Computation 1966; 20:392398.
29. Cook AW. Artificial fluid properties for large-eddy simulation of compressible turbulent mixing. Physics of Fluids
2007; 19(5):055103.
30. Mani A, Larsson J, Moin P. Suitability of artificial bulk viscosity for large-eddy simulation of turbulent flows
with shocks. Journal of Computational Physics 2009; 228(19):73687374.
31. Cook AW, Cabot WH. A high-wavenumber viscosity for high-resolution numerical method. Journal of
Computational Physics 2004; 195(2):594601.
32. Jiang GS, Shu CW. Efficient implementation of weighted ENO scheme. Journal of Computational Physics 1996;
126(1):202228.
33. Arora M, Roe PL. On postshock oscillations due to shock capturing schemes in unsteady flows. Journal of
Computational Physics 1997; 130(1):2540.
34. Sod GA. A survey of several finite difference methods for systems on non-linear hyperbolic conservation laws.
Journal of Computational Physics 1978; 27(1):131.
35. Lax PD. Weak solutions of nonlinear hyperbolic equations and their numerical computation. Communications on
Pure and Applied Mathematics 1954; 7(1):159193.
36. Hass JF, Sturtevant B. Interaction of weak shock waves with cylindrical and spherical gas inhomogeneities.
Journal of Fluid Mechanics 1987; 181:4176.
37. Picone JM, Boris JP. Vorticity generation by shock propagation through bubbles in a gas. Journal of Fluid
Mechanics 1988; 189:2351.
38. Quirk JJ, Karni S. On the dynamics of a shockbubble interaction. Journal of Fluid Mechanics 1996; 318:129163.
39. Bagabir A, Drikakis D. Mach number effects on shockbubble interaction. Shock Waves 2001; 11:209218.
40. Marquina A, Mulet P. A flux-split algorithm applied to conservative models for multicomponent compressible
flows. Journal of Computational Physics 2003; 185(1):120138.
41. Brouillette M, Sturtevant B. Experiments on the RichtmyerMeshkov instability: single-scale perturbations on a
continuous interface. Journal of Fluid Mechanics 1994; 263:271292.
42. Brouillette M. The RichtmyerMeshkov instability. Annual Review of Fluid Mechanics 2002; 34:445468.
43. Collins BD, Jacobs JW. PLIF flow visualization and measurements of the RichtmyerMeshkov instability of an
air/SF6 interface. Journal of Fluid Mechanics 2002; 464:113136.
44. Latini M, Schilling O, Don WS. Effects of WENO flux reconstruction order and spatial resolution on reshocked
two-dimensional RichtmyerMeshkov instability. Journal of Computational Physics 2007; 221(2):805836.
45. Thompson KW. Time dependent boundary conditions for hyperbolic systems. Journal of Computational Physics
1987; 68(1):124.

Copyright q 2010 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Fluids 2011; 66:12071225
DOI: 10.1002/fld

You might also like