This action might not be possible to undo. Are you sure you want to continue?

# WENO scheme with subcell resolution for computing nonconservative Euler equations with applications to one-dimensional compressible

two-medium ﬂows Tao Xiong1 , Chi-Wang Shu2 and Mengping Zhang3 Abstract High order path-conservative schemes have been developed for solving nonconservative hyperbolic systems in [32, 7, 6]. Recently, it has been observed in [3] that this approach may have some computational issues and shortcomings. In this paper, a modiﬁcation to the high order path-conservative scheme in [7], based on the high order ﬁnite volume WENO scheme with subcell resolution and utilizing the exact Riemann solver to catch the right paths at the discontinuities, is proposed to improve its computational performance and to overcome some of the shortcomings. An application to one-dimensional compressible two-medium ﬂows of nonconservative or primitive Euler equations is carried out to show the eﬀectiveness of this new approach.

Keywords: path-conservative schemes; high order ﬁnite volume WENO scheme; subcell resolution; nonconservative hyperbolic systems; primitive Euler equations; two-medium ﬂows; exact Riemann solver

Department of Mathematics, University of Science and Technology of China, Hefei, Anhui 230026, P.R. China. E-mail: jingt@mail.ustc.edu.cn 2 Division of Applied Mathematics, Brown University, Providence, RI 02912, USA. E-mail: shu@dam.brown.edu. Research supported by ARO grant W911NF-11-1-0091 and NSF grant DMS-1112700. 3 Department of Mathematics, University of Science and Technology of China, Hefei, Anhui 230026, P.R. China. E-mail: mpzhang@ustc.edu.cn. Research supported by NSFC grant 11071234.

1

1

1

Introduction

Recent years have seen a growing interest in developing numerical algorithms for solving compressible multicomponent ﬂows. The dynamics of inviscid multicomponent ﬂuid may be modeled by the Euler equations. However, computations often run into unexpected diﬃculties due to nonphysical oscillations generated at the vicinity of the material interface, while such oscillations do not arise in single-ﬂuid computations when a nonlinearly stable scheme, such as the essentially non-oscillatory (ENO) or weighted ENO (WENO) scheme [16, 38, 29, 20], is used. The underlying mechanisms have been analyzed and several methods have been developed to overcome these diﬃculties, e.g. in [23, 22, 10, 1, 2, 25, 5]. There are mainly two approaches to circumvent these oscillations. One is still based on the conservative Euler equations, and the other is to write the Euler equations in nonconservative or primitive form. The ghost ﬂuid method (GFM) developed in [13] with the isobaric ﬁx technique in [14] has provided an attractive and ﬂexible way to treat the two-medium ﬂow model for conservative Euler equations. The GFM treats the material interface as an internal boundary, and by deﬁning ghost cells and ghost ﬂuids, the two-medium ﬂow can be solved via two respective single-medium GFM Riemann problems. This method is simple and it easily extends to multi-dimensions, and it can maintain a sharp interface without oscillations. Variants of the original GFM and their applications can be found in, e.g. [25, 26, 27, 46] and the references therein. Later, these techniques are used to develop the Runge-Kutta discontinuous Galerkin (RKDG) ﬁnite element method for multi-medium ﬂow in [33, 34, 51, 47]. The GFM, even though approximating directly the conservative Euler equations, is in general not a conservative method because the ﬂux at the interface is double-valued. Recently, a conservative modiﬁcation to the GFM using the ﬁfth order ﬁnite diﬀerence WENO scheme with third order Runge-Kutta time discretization has been studied in [28], which attempts to reduce the conservation error of the GFM without aﬀecting its performance. The other approach is based on the observation that erroneous pressure ﬂuctuations 2

are generated by the conservative equations, hence a better approximation can be obtained when writing the equations in a nonconservative (primitive) form [21, 22]. In these papers a scheme for the nonconservative Euler formulation is proposed, with consistent correction terms to remove the leading order conservation errors. This scheme can completely eliminate spurious oscillations at the material interface, yielding clean monotonic solution proﬁles. This work has been extended to solve two dimensional problems in [35]. In this method, the correction terms depend heavily on the corresponding conservative numerical scheme, and only second order nonconservative scheme has been provided and applied to shocks of weak to moderate strengths, it is less justiﬁable for high-resolution schemes with narrow shock transition and it is not justiﬁed in cases of strong shocks [18]. It is noted in these papers that, for nonconservative hyperbolic systems, the shock relationships are not uniquely deﬁned by the limiting left and right states, but also depend on the viscous path connecting the two states. The correct shock capturing lies in getting correctly the viscous path. The theory developed by Dal Maso, LeFloch and Murat [11] gives a rigorous deﬁnition of nonconservative products, associated with the choice of a family of paths. Later, people have paid much attention to the development of numerical schemes for solving nonconservative hyperbolic systems, see for example [32, 7, 8, 6] and references therein. A high order Roe-type scheme based on the reconstructed states has been provided in [7] for the one-dimensional case, and then extended to two dimensions in [6], but only with applications to shallowwater systems. Other work based on this high order method has been carried out for solving two-phase ﬂow models in [45, 12, 42]. A limitation of this approach has been pointed out recently in [3] when applied to nonconservative Euler equations. The problem is related to the eﬀective choice of a correct path in the nonconservative high order scheme. It appears that, for nonconservative Euler equations, using the Roe linearization to choose a path as in [43, 7, 6] might end up in converging to a weak solution of a diﬀerent path [3]. In this paper, we focus on adapting high order schemes for solving nonconservative Euler equations, still based on high order Roe-type ﬁnite volume schemes as those in [7, 6]. For

3

A summary of our algorithm to aid 4 . In Section 5. with the sharp left and right states at the discontinuities. or ENO and WENO schemes. they may not land on the correct path and hence may lead to convergence to erroneous weak solutions on diﬀerent paths. we ﬁrst describe the high order Roe scheme for the nonconservative hyperbolic system. shock capturing schemes such as monotone. Later. we use the exact Riemann solution to catch the right path that connects the two states. but at discontinuities. however in a nonconservative formulation. In smooth regions. Realizing this diﬃculty. which unfortunately is generic with all shock capturing schemes. they are not [21]. we present the level set function to track the material interface. as the correct capturing of the shock speed is sensitive to the accuracy of the numerically achieved path. total variation diminishing (TVD). In Section 6. As a result. our basic idea in this paper is to use Harten’s subcell resolution technique [15] to sharpen the discontinuities and eﬀectively eliminate (or at least signiﬁcantly reduce) the transition points. Then we introduce the two medium ﬂow model for nonconservative Euler equations in Section 3. In Section 2. the right path should recover the weak solution of the conservative Euler formulation with density. Harten’s subcell resolution idea [15] is based on ENO schemes with a Lax-Wendroﬀ time discretization procedure in a cell-averaged framework. smear discontinuities with one or several transition points. As is well known.the nonconservative or primitive Euler equations. These transition points are necessary for conservation. this subcell resolution idea has been used in solving advection equations with stiﬀ source terms. Recently. In Section 4. the convergence towards the correct weak solution based on the originally desired path seems to be restored. we describe the computation of the integral term in the high order Roe scheme in detail. we illustrate how to apply the WENO reconstruction with subcell resolution to the high order Roe scheme. In this paper. momentum and energy as its variables. the conservative and nonconservative Euler equations are equivalent. this idea is extended in [39] to both ﬁnite volume and ﬁnite diﬀerence ENO schemes with Runge-Kutta time discretization. to obtain correct shock speed on coarse meshes [48].

2 2 i = 1.1) reduces to a classical conservation law Wt + F (W )x = 0 A conservative ﬁnite volume semi-discretization of the system (2. t) is a N-component state vector and A(W ) is a N × N matrix.. a nonconservative hyperbolic system reads Wt + A(W )Wx = 0.2) is Wi′ (t) = with the numerical ﬂux + Gi+1/2 = G(Wi− +1/2 (t). 2 2 2 2 The cells. The system is supposed to be hyperbolic. that is when A(W ) = ∂F/∂W . 2.1) where W = W (x. 2 2 2 ∆x ≡ xi+ 1 − xi− 1 . Wi+1/2 (t)). we follow the procedure in [7] to deﬁne the high order Roe scheme for solving nonconservative hyperbolic systems. i. 2 High order Roe scheme for nonconservative hyperbolic systems In this section.2) 1 (Gi−1/2 − Gi+1/2 ). Nx . A(W ) has N real eigenvalues and a full set of N linearly independent eigenvectors.e.3) (2. For the one-dimensional case. 2 2 1 xi ≡ (xi− 1 + xi+ 1 ).4) 5 . ∆x (2... .implementation is given in Section 7. x ∈ Ω ⊂ R. We use a uniform grid a = x 1 < x 3 < · · · < xNx − 1 < xNx + 1 = b. xi+ 1 ]. (2. which is the Jacobian of a ﬂux function F (W ). (2. t>0 (2. cell centers. and the uniform cell size are denoted by Ii ≡ [xi− 1 . Concluding remarks follow in the last section. In the case of systems of conservation laws. and one-dimensional numerical examples are provided to demonstrate the eﬀectiveness of our approach in Section 8.

Wi+1/2 (t)) (2. 6 . ± ± Λ± i+1/2 = diag ((λ1 )i+1/2 .8) (2. (λN )i+1/2 ) (2. and Λi+1/2 is the diagonal matrix whose diagonal entries are the corresponding eigenvalues of Ai+1/2 . The intermediate matrices are deﬁned by − Ai+1/2 = A(Wi+ +1/2 (t).10) where Ri+1/2 is a N × N matrix with each column as a right eigenvector of Ai+1/2 .where Wi (t) is used to approximate the cell averaged value W i (t).9) and −1 ± A± i+1/2 = Ri+1/2 Λi+1/2 Ri+1/2 . which is deﬁned as 1 W i (t) = ∆x xi+1/2 W (x. · · · . t)dx.3) can be written as Wi′(t) = − 1 + − (A+ i−1/2 (Wi−1/2 (t) − Wi−1/2 (t)) ∆x (2. xi−1/2 and Wi± +1/2 (t) are the reconstructed states associated to the cell average sequence {Wj (t)}. The semi-discrete high order Roe scheme for (2.3) with the numerical ﬂux (2.5) + − + − +A− i+1/2 (Wi+1/2 (t) − Wi+1/2 (t)) − F (Wi−1/2 (t)) + F (Wi+1/2 (t))) which is equivalent to the conservative scheme (2.4) deﬁned to be + − − Gi+1/2 = F (Wi− +1/2 (t)) + Ai+1/2 (Wi+1/2 (t) − Wi+1/2 (t)) + − + = −F (Wi+ +1/2 (t)) + Ai+1/2 (Wi+1/2 (t) − Wi+1/2 (t)) (2.7) and the Roe property A(W + − W − ) = F (W + ) − F (W − ) has been used. 2 − |Ai+1/2 | = A+ i+1/2 − Ai+1/2 (2.6) = Here 1 − + − ((F (Wi+ +1/2 (t)) + F (Wi+1/2 (t)) − |Ai+1/2 |(Wi+1/2 (t) − Wi+1/2 (t))).

that include density.2) . p)T . (2. we describe the two-medium inviscid compressible ﬂow model for the nonconservative or primitive Euler equations. As pointed out in [21]. such that lim + Pit (x) = Wi+ −1/2 (t). the choice of the primitive set of variables.11). (3. lim − Pit (x) = Wi− +1/2 (t). 3 Two-medium ﬂow model for nonconservative Euler equations In this section. (2. provides a model better suited for computations of propagating material fronts and results in clean and monotonic solution proﬁles. so we consider the one dimensional primitive Euler equations Wt + A(W )Wx = 0 with W = (ρ. velocity and pressure.1) can be easily generalized from (2. 0 γp u 7 (3.12) Wi′ (t) = − 1 (A+ (W + (t) − Wi− −1/2 (t)) ∆x i−1/2 i−1/2 + − +A− i+1/2 (Wi+1/2 (t) − Wi+1/2 (t)) + xi+1/2 xi−1/2 A(Pit (x)) d t P (x)dx) dx i (2.13) with the function Pit (x) satisfying (2.We introduce Pit (x) as any smooth function deﬁned in the cell Ii . the Harten-Hyman entropy ﬁx technique [17.12) Now the numerical high order Roe scheme for solving the nonconservative system (2. dx (2.11) x→xi−1/2 x→xi+1/2 Then. In order to obtain entropy-satisfying solutions.5) can be rewritten as Wi′ (t) = − 1 (A+ (W + (t) − Wi− −1/2 (t)) ∆x i−1/2 i−1/2 + − +A− i+1/2 (Wi+1/2 (t) − Wi+1/2 (t)) + xi+1/2 xi−1/2 d F (Pit (x))dx). 44] can be applied to this scheme.1) u ρ 0 A(W ) = 0 u ρ−1 . u.

where e is the speciﬁc internal energy per unit mass. Nx . The ρu2 . 33] is expressed as ρe = (p + Nw B )/(Nw − 1) where B = B − A. linear version ut + f (u)x = 0. We would like to reconstruct u− i+1/2 and ui−1/2 in each cell from the sequence of cell averages {uj }.1) based on the big stencil Si ≡ {Ii−r . The extension + to the nonlinear and system cases will follow. γ is the ratio of speciﬁc heats..13) with the third order TVD Runge-Kutta time discretization [38]. We follow the procedure in [39] to describe how to apply the subcell resolution technique of Harten [15] to the scheme (2. WENO Reconstruction Algorithm: Given the cell averages {uj } of a function u(x): uj = 1 ∆x xj +1/2 u(ξ )dξ. 8 u+ i−1/2 = j =0 ωj (xi−1/2 )pj (xi−1/2 ) (4. Ii .. to describe the WENO reconstruction with the subcell resolution technique. We ﬁrst consider the 1D. xj −1/2 j = 1.0E 5 Pa and B = 3. scalar. with f (u) = au and a > 0. .2) . 37]. a k -th (k = 2r + 1) order accurate + approximation to the boundary values u− i+1/2 and ui−1/2 in the cell Ii is reconstructed as r r u− i+1/2 = j =0 ωj (xi+1/2 )pj (xi+1/2 ).Here ρ is the density.3) 4 WENO reconstruction with subcell resolution In this section. we will describe how to use the WENO reconstruction with subcell resolution to reconstruct Wi± +1/2 (t) from the cell averages {Wj (t)}. A = 1. with the following algorithm. u is the velocity.4) (3.. Ii+r }. · · · .15. 2. (4. 13. Nw = 7. p is the pressure.31E 8 Pa. The WENO reconstruction is described in detail in [20. · · · . 25. (3. total energy is given by E = ρe + 1 2 The γ -law equation of state (EOS) used for ideal gases is given as ρe = p/(γ − 1) and Tait EOS used for the water medium [9.

the technique to treat negative weights in [36] needs to be applied.3) The nonlinear weight ωj (x) is close to zero when a discontinuity is located in the stencil Sij . · · · . while still reconstructing values in the cell Ii (e. ∆− ui )|. an ) = s min1≤i≤n |ai |. the linear weights dj (x) may be negative. if s = sign(a1 ) = · · · = sign(an ) otherwise . let θi = (ui+1 − ui )/(ui+1 − ui−1 ). and they satisfy ωj (x) > 0. in the following subcell resolution algorithm).4) (II) For any “critical interval” Ii . Subcell Resolution Algorithm: First. xi+1/2 ) by σi ≥ σi+1 . σi > σi−1 .2). in each Runge-Kutta time cycle. so as to avoid involving much information from any stencil Sij which contains discontinuities. Then. we perform: + (III) Let the cell Ii boundary values u− i+1/2 and ui−1/2 be deﬁned as usual. at the beginning of every Runge-Kutta cycle: (I) Deﬁne the “critical intervals” (intervals containing discontinuities) Ii = (xi−1/2 . unless Ii or (for the second and third Runge- 9 . using the standard WENO reconstruction algorithm (4.where each pj (x) is a reconstruction polynomial that uses the cell averages in the small stencil Sij ≡ {Ii−j . where σi = |m(∆+ ui . Ii−j +r } ⊂ Si . at certain internal points x ∈ Ii (reconstruction at those points are needed in Section 5).2). the linear weights dj (xi+1/2 ) and dj (xi−1/2 ) are positive. and use xi−1/2 + θi ∆x as an approximation to the discontinuity location inside the cell Ii . The nonlinear weights {ωj (x)}r j =0 are calculated from r the polynomials {pj (x)}r j =0 and the linear weights {dj (x)}j =0 at each ﬁxed point x. The linear weights may also become negative if the stencil Si is changed to Si+1 or Si−1 . Remark: For the cell boundaries (4. ∆+ ui = ui+1 − ui .g. j =0 (4. However. · · · . In these cases. 0. r ωj (x) = 1. ∆− ui = ui − ui−1 and m is the minmod function which is deﬁned to be m(a1 . (4.

and ui+1/2 is the WENO reconstruction with the stencil Si+1 and evaluated at xi+1/2 . we always set it to be a “critical interval”. −. Notice that here for ui+1/2 the technique for treating negative weights needs to be used.1 a∆t (4.old where u− i−1/2 is the standard WENO reconstruction with the stencil Si−1 .old u+ i−1/2 = ui−1/2 (L) (L) (R) with ξi = min (1 − θi )∆x . ui+1/2 is also the standard WENO reconstruction with the stencil Si−1 . we deﬁne u− i+1/2 = (1 − ξi )ui+1/2 + ξi ui+1/2 . the subcell resolution algorithm is simply applied in each local characteristic ﬁeld. but evaluated at xi+1/2 of the cell Ii . to avoid obtaining a nonphysical. The only diﬀerence in our current situation is that we just use A(Wi ) as the Roe average matrix to do the characteristic decomposition when deﬁning if Ii is a “critical interval”. (c) For nonlinear systems. and A Wi+ +Wi− +1/2 +1/2 2 Wi−1 +Wi 2 for the left cell boundary at xi−1/2 .Kutta stages) Ii−1 is a “critical interval”. The detailed procedure can be found in [39]. then we remove the one with a smaller σi from the list of critical intervals.5) (4. and we use A Wi +Wi+1 2 (R) (L) (R) to do the characteristic decomposition for the right cell boundary at xi+1/2 . the subcell resolution is used to get sharp left and right states at the 10 . and when Ii−1 is a “critical interval” and ξi−1 < 1. entropy condition violating solution. we choose the stencil Si+1 for u− i+1/2 . (d) It is pointed out in [15. However if the material interface is located in the cell Ii . If Ii is a “critical interval”. Remark: (a) The case for a < 0 is easily obtained by symmetry. (b) If two adjacent cells are both “critical intervals”. we choose the stencil Si+2 for ui+1/2 if ξi < 1 and the negative weight treating technique needs to be used here as well. Special caution is needed when one tries to sharpen a (nonlinear) shock. For the second and third Runge-Kutta stages. The Roe average to deﬁne the intermediate matrix in (2. In our approach. 39] that the subcell resolution technique should be applied only to sharpen contact discontinuities.9) is Ai+1/2 = A .6) .

with eigenvalues λL and λR corresponding to the left and right states respectively. ω2.13) xi+1/2 xi−1/2 A(Pit (x)) d t P (x)dx dx i (5. According to this. At a nonlinear shock.discontinuities. 2 ].1) by xi+1/2 xi−1/2 A(Pit (x)) d t P (x)dx = dx i G ωj A(Pit (sj )) j =1 d t P (s j ). 5. (5. positions {sj } and associated weights {ωj } for a G-point quadrature in the interval [− 2 we can replace the analytical path integral (5. dx i (5. near discontinuities.2) In our numerical experiments. However.4 = . all simple wave models for conservative Euler equations and primitive Euler equations are equivalent. we use the four-point Gauss-Lobatto quadrature rule: √ 1 5 5 1 s2. we apply the subcell resolution algorithm in both the linearly degenerate ﬁeld and genuinely nonlinear ﬁelds [44]. and discontinuities at the material interface. but for the genuinely nonlinear ﬁelds. ω1. our choice of the path is divided into three parts: the smooth case. 2 10 6 6 11 . this is also needed so as to get a more accurate shock speed which heavily depends on the left and right states.4 = ∓ . 5 Choice of the path and evaluation of the path integral In smooth regions.3) s1. discontinuities in a single medium. In the computation for Euler equations of compressible gas dynamics.1) can be computed via a high order accurate Gauss-Lobatto quadrature rule. the integral term in (2.1 The smooth case In the smooth case. they are not equivalent [21]. we only apply the subcell resolution for the case λL ≥ λR to avoid sharpening a rarefaction wave. Given the 1 1 .3 = ∓ .3 = .

a contact discontinuity. and Pit (s3 ) as Pit (s4 ) corresponding to the local characteristic ﬁeld at xi+1/2 . Pit (s4 ) = Wi− +1/2 (5. d P t (x) dx i can be approximated by Q(x). Note here for the smooth case.2 Discontinuities in a single medium At the discontinuities. since all zero. and Pit (x) in this part is just the rarefaction wave line. The exact Riemann solution for the compressible Euler equations contains four constant states connected by a rarefaction wave or shock wave.For Pit (s1. It can be computed as the following seven parts: (a) The four constant-state parts. and another rarefaction wave or shock wave. In this case of the “critical interval”. where Pit (s2 ) is obtained in the same way as Pit (s1 ) corresponding to the local characteristic ﬁeld at xi−1/2 .3 ).1).11) we already have Pit (s1 ) = Wi+ −1/2 . Otherwise. the integral in these four parts are . 44] to obtain the exact Riemann solution between the two states. we do not need the subcell resolution in the WENO reconstruction. if WL and W∗L are connected by a shock wave. Since we have {Pit (sj )}4 j =1 . (5. (b) If WL and W∗L are connected by a rarefaction wave. from (2. then the integral path for this part needs to satisfy the Rankine-Hugoniot jump condition of the conservative Euler equations. 12 d P t (x) dx i = 0. we use the exact Riemann solution to get the integral path for the integral term (5. The four constant states can be denoted as WL . we can use the exact Riemann solver [40. we can also obtain Pit (s2.4 ).4) following the same procedure for obtaining Pit (s1.2). we similarly use the 4-point Gauss-Lobatto quadrature rule (5. the derivative of the Lagrangian d P t (s j ) dx i interpolation polynomial based on {Pit (sj )}4 j =1 . Denote the left and right states Wi+ −1/2 and Wi+1/2 to be WL and WR . Then can be replaced by Q(sj ) in 5. W∗L . W∗R and WR .4 ). the cell is deﬁned as a “critical interval”. and we have obtained the left − and right states Wi+ −1/2 and Wi+1/2 from the WENO reconstruction with subcell resolution − in Section 4.3).

4. The level set equation for the one-dimensional case is φt + uφx = 0. which can be easily veriﬁed. We only need to make sure that these paths exist. Since W∗L and W∗R have the same pressure p and velocity u and diﬀerent densities ρ∗L and ρ∗R . Apart from W∗L and W∗R which are connected by a contact discontinuity. (c) For the contact discontinuity part between W∗L and W∗R .1) 13 .3 Discontinuities at the material interface We always set the cell at the material interface as a “critical interval”. 6 Tracking the moving medium interface In this section. a contact discontinuity. we do not need to know the exact integral paths for satisfying the Rankine-Hugoniot jump condition. Similar results can be obtained for the connection between W∗R and WR . with σ being the shock speed related to the two states WL and W∗L . also containing four constant states connected by a rarefaction wave or a shock wave. and another rarefaction wave or shock wave [25.1) can be computed similar to the second case of discontinuities in a single medium. in order to get the integral results along those paths. 5. 31] to track the moving ﬂuid interface. we describe how to use the level set equation [41. the integral path also needs to satisfy the Rankine-Hugoniot jump condition. for the shock wave and contact discontinuity.and the integral result is σ (W∗L − WL ). 0)T . Remark: In (b) and (c) above. 24]. (6. the integral result is simply (u(ρ∗R − ρ∗L ). 0. 27. the left part is for medium one with the ratio of speciﬁc heats γ1 and the right part is for medium two with the ratio of speciﬁc heats γ2 . as the exact Riemann solution at the material interface is almost the same as that for the single component compressible Euler equations. The integral term (5.

The solution φ = φ0 from solving the level set equation (6. A serious distorted level set function φ = φ0 may lead to signiﬁcant errors for t > 0. 37] with the third order TVD Runge-Kutta time discretization [38] to solve the level set equation (6.The interface is tracked as the zero level set of φ. We use the Godunov Hamiltonian and the ﬁfth order ﬁnite diﬀerence WENO discretization in [19. We can now summarize our algorithm to advance one Euler forward time step (still denoted 14 . 3 (7. where S (φ) = √ φ φ2 +(∆x)2 (6. 4 2 ∆tL(W (2) ). which can be written as Wt = L(W ).1). 7 Algorithm summary Our basic semi-discrete scheme is (2. with the initialized φ(x) to be the signed normal distance to the material front. and we take ∆τ = ∆x/10 in the experiments [41].2) is the approximate sign func- tion. using the velocity u coming from the Euler equations. The stopping criterion for this iteration is e1 < ∆τ (∆x)2 . W (2) = W n+1 3 n W + 4 1 n W + = 3 1 (1) W + 4 2 (2) W + 3 1 ∆tL(W (1) ).13). For this reason. but it needs not be the distance function for t > 0. where e1 is the L1 diﬀerence of φ between two consecutive iteration steps. We use a ﬁfth order ﬁnite diﬀerence WENO method [20.1) has the zero level set as the material interface. φ0 (x) is reinitialized to be a signed normal distance function to the interface by solving the following eikonal equation to steady state φτ = S (φ0 )(1 − |φx |) through iterating the pseudo-time τ . This equation is solved concurrently with the nonconservative Euler equations (3.1) Notice that this Runge-Kutta method is simply a convex combination of three Euler forwards. 50] to solve (6.2).1). It is discretized in time by the third order TVD Runge-Kutta method [38]: W (1) = W n + ∆tL(W n ).

Reinitialize φn+1/2 by solving (6. With ﬁfth order WENO reconstruction in Section 4 (r = 2) and 15 . and ρj . Here (7.1) as described in Section 5.1). P (φ) can be formulated simultaneously. we compute the integral term (5. Now we have advanced one Euler forward time step. we ﬁrst reconstruct Wi+ −1/2 and Wi+1/2 as described in Section 4. velocity and pressure at time level tn . Step 2.2) where cn j = n n n n γpn j /ρj is the sound speed.1) with the ﬁfth order ﬁnite diﬀerence WENO spatial discretization. Compute the new time step size based on the CFL condition: n ∆t = CFL ∆x/ max (|un j | + cj ) 1≤j ≤Nx (7. given the velocity u = un from the Euler equations.4) is written from (6. 8 Numerical experiments In the following. Deﬁne the new interface position from the zero level set of φn+1 . and denote this solution by φn+1 . pj are the density. Step 3. We can then formulate the right side of (7. Taking W n and φn as the initial condition. respectively. and the updated φ by φn+1/2 . − based on Wi+ −1/2 and Wi+1/2 .3).2) with φ0 = φn+1/2 .as from tn to tn+1 ) in the following steps. solve Wt = L(W ). φ t = P (φ ). Step 1. (7. Denote the updated W by W n+1.4) for one time step using the Runge-Kutta time discretization (7. uj . This step needs to be done only at the beginning of the whole Runge-Kutta cycle. Step 4. then.3) (7. At each Runge− Kutta time cycle. we show several one-dimensional numerical examples to demonstrate that our approach can improve the performance of the high order Roe scheme for the nonconservative Euler equations.

velocity. which will be tested in Example 1.1. Since the solution to this problem is smooth.87 5.93E-07 1.53E-10 order – 4.02 4.19E-08 3. 2π ] with periodic boundary conditions.1. respectively.5.07 16 . The CFL number is taken as 0. t=1. we do not need to apply the subcell resolution. as listed in Table 8.05 5. the Roe scheme can achieve ﬁfth order accuracy for smooth solutions. u(x) = 1. the ﬁfth order accuracy can be achieved for this path-conservative scheme applied to the single component nonconservative Euler equations. The exact solution is ρ(x) = 1 + 0. p0 (x) = 1 (8. we ﬁrst test the accuracy for a smooth solution to a single component nonconservative Euler equations with initial conditions ρ0 (x) = 1 + 0. Table 8. 1]. length and time are kg/m3 .8 sin(x − t).04 5. u0 (x) = 1.02E-04 5.1: Accuracy test for Example 1 with initial data (8. the computational domain for all other examples are taken as [0.four-point Gauss-Lobatto quadrature rule in Section 5. Except for Example 1.08 5.99 L∞ error 3.1). Nx 20 40 80 160 320 L1 error 2.1. The computational domain is divided with Nx = 100 uniform grids.28E-05 3. p(x) = 1. and the initial material interface for the two-medium ﬂow problems is located at x = 0.74E-04 1.1) on a domain [0.76E-10 order – 5. Example 3 and Example 5. 8. pressure.1 Example 1 In this example.1.8 sin(x). P a. Units for density. m/s. With ﬁfth order WENO reconstruction in Section 4 (r = 2) and four-point Gauss-Lobatto quadrature rule for the integral term in the smooth case in Section 5.96E-06 1.03 5.81E-07 5.58E-09 1. m and s.

contact discontinuity and shock wave are all captured well.9 ( 14 3 .5.1 against the exact solution in Fig. p) = 41 . We also use this simple example to illustrate that in our approach.2.3) We compute this problem on a domain [0. 8. we list the results for two cases: one is obtained when we use the exact Riemann solution but without subcell resolution. We can see that neither of them can catch the correct shock solution. to demonstrate that we obtain the correct entropy solution by our non-conservative scheme. We can see that our algorithm can capture the correct shock location in this case. (8.3. 8. We can see that the rarefaction wave. 8.8. the other is obtained when we use subcell resolution but with the line path [43. 3. with the initial condition: (ρ.3 Example 3 This example is a right moving shock for a single medium ﬂow. velocity u and pressure p are plotted at t = 0. u. 8. 3] instead of the exact Riemann solution.2) The computed density ρ.5.571) for x ≤ 0. on a very reﬁned mesh Nx = 1000. Here γ = 1. (8. for x > 0. 17 .5. In Fig. 0. 0.17 against the exact solution in Fig. p) = (0. and the minor glitches on the left constant state would not grow with a much reﬁned mesh (ﬁgures on the right).528) (0.698. The computed density ρ. 1].1.4 is used. 0.5. the subcell resolution and the exact Riemann solution are both necessary in order to catch the right path in this non-conservative scheme. velocity u and pressure p are plotted at t = 0. u.5. 1) for x > 0. 0. Here γ = 5/3 is used. for x ≤ 0. which is the standard Lax shock tube problem.2 Example 2 In this example. 10) 41 (1. A similar example (in Lagrangian form) is used in [3] to demonstrate potential problems of nonconservative path-based Roe-type and Lax-Friedrichs (LxF)-type schemes. with the initial condition: (ρ. we now show a single medium ﬂow problem.445.

4 0.8 0.6 0.5 Pressure 2 1.6 0.4 0.6 0.5 Velocity 1 0.2 0 0. velocity and pressure for Example 2.4 0.4 1.2 0. t = 0.4 0.8 1 x 4 3.2 0.17.1: Density.5 3 2.1.5 1 0. Solid line: the exact solution.8 1 x Figure 8.8 1 x 1.5 0 0 0. Symbol: the numerical solution.6 0. 18 .2 0.5 0 0 0.2 1 Density 0.

5 -0.5 0 0.5 0. 3 x 3 2.2 0.6 0.8 1 x .2 0. Symbol: the numerical solution. t = 0.4 0. 12 x 10 10 8 Pressure Pressure 0 0.6 0. velocity and pressure for Example 3.5 2 2 Velocity 1.1.8 1 1.4 0.4 0.8 1 6 5 4 2 0 0 0 0.4 0. 19 . x Figure 8.8 1 x .8 1 x .4 0.8 1 2 1 1 0 0 0 0.4 4 3 3 Density 2 Density 0 0.6 0.2 0.2 0. Right: mesh Nx = 1000.2 0.6 0.4 0.5 0 0 -0.5 Velocity 0 0.6 0. Left: mesh Nx = 100.2: Density.5 2.6 0.5 1 1 0. Solid line: the exact solution.2 0.

6 0.8 1 2 1.8 1 x .5 1 1 0 0.4 0.6 0.2 0. velocity and pressure for Example 3.5 2.3.4 0.2 0.8 1 Velocity 1 0 0 0 0. x Figure 8.2 0. t = 0.1.4 0.2 0.3: Density.4 0.2 0.8 1 x .6 0.8 1 x . 20 . x 2 2 1 Velocity 0 0.2 0.5 1.6 0.6 0.5 Density 2 Density 0 0.5 3 3 2. Nx = 1000. Left: exact Riemann solution without subcell resolution. Right: subcell resolution and with the line path instead of the exact Riemann solution.5 3. Symbol: the numerical solution. Solid line: the exact solution. 12 x 12 10 10 8 8 Pressure Pressure 0 0.4 0.8 1 6 6 4 4 2 2 0 0.4 0.6 0.

1 × 104 . 0. we also show the results obtained with a second order method. 39]. The solid line is the solution obtained by the ﬁfth order method with Nx = 2000 points. p. 28]. u. The initial discontinuity is right on the material interface and located at x = −4. 1 × 105 . 0. 0. γ ) = (1. which can be considered as a converged reference solution.4) (1 + 0.5. u.5. 1. for x > 0. For comparison.4) The computed density ρ.5. 1. We can see that the rarefaction wave. 8.8 on a domain [−5. especially in the region with ﬁne structures. the contact discontinuity and the shock wave are all captured well for this two-medium ﬂow. (8.8. which is a simple model for shock-turbulence interactions. We plot the densities obtained by the second order and the ﬁfth order methods with Nx = 200 in Fig. We can ﬁnd that the two methods can both capture the correct solution.4. This is is similar to the results in [47]. 8. 1. 5].4) (0. on this relatively coarse mesh. p. by replacing the ﬁfth order WENO reconstruction with subcell resolution in Section 4 with a second order ENO reconstruction [15.5 Example 5 This example is used to demonstrate the advantage of high order methods as in [47] for two medium ﬂows. and replacing the four-point Gauss-Lobatto quadrature rule for the smooth integral term in Section 5. 10.333333. 1. however the result of the ﬁfth order method is in better agreement with the converged reference solution than the second order method.125.0007 against the exact solution in Fig. (8. 33. velocity u and pressure p are plotted at t = 0.2) for x ≤ 0. It contains both shocks and ﬁne structures in smooth regions.4 Example 4 This is an air-helium shock tube problem taken from [13. 21 . for x > −4. the left and right states for the initial discontinuity are (ρ.629369.2 sin(5x).857143. 2.1 with a trapezoid rule. 8. γ ) = (3.5) We compute the example up to time t = 1. with the initial condition: (ρ. 5/3) for x < −4.

8 1 x Figure 8.4 0.6 0.2 0 0 0. Symbol: the numerical solution. t = 0.2 1 0.4 0.6 0.6 0.8 1 x 350 300 250 200 Velocity 150 100 50 0 -50 0 0.1. Solid line: the exact solution.2 0. 22 .4 0.8 1 x 100000 80000 Pressure 60000 40000 20000 0 0 0.8 Density 0.6 0.0007.2 0. velocity and pressure for Example 4.2 0.4: Density.4 0.

1. velocity u and pressure p are plotted at t = 0. 1 × 105 .05. also taken from [13.0012 against the exact solution in Fig.5 4. p. 33.6 Example 6 This is a problem of a shock wave refracting at an air-helium interface with a reﬂected weak rarefaction wave taken from [13. 0. γ ) = (1. u.5 4 3. Symbol: the numerical solution with Nx = 200. 1. 3.5.2817 10 . 8. γ ) = (1.05 < x ≤ 0. 1. velocity and pressure for Example 5. 0 . The strength of the shock for this example is pL /pR = 1.5. 8.5 Density 2. 1 × 10 . with the initial condition as: √ 6 5 for x ≤ 0. the computed results compare well with the exact solutions. 28]. (8.7) The computed density ρ. 1 × 10 .05 < x ≤ 0.7. 0.5.5 1 0. 8.5 2 1.5.4) for 0.6.5: Density. 1. 5 (0.5 0 -4 -2 0 2 4 Density 3 3 2. 5/3) for x > 0.05. Right: ﬁfth order method. (1 .5 2 1. 1. velocity u and pressure p are plotted at t = 0. The computed results still compare reasonably well to the exact 23 .5.0005 against the exact solution in Fig. t = 1.5 5 4. 3333 . p.5 × 105 . 5 (0. with the initial condition: √ 105 . u.1379. 0.7 Example 7 This example is the same as Example 6.3333.4) for x ≤ 0. Solid line: the converged reference solution for the ﬁfth order method with Nx = 2000. 1 × 10 . (8.1379. 5/3) for x > 0. x Figure 8.4) (ρ.5 0 -4 -2 0 2 4 x . (4.5 1 0.5 4 3. 8. without oscillation around the material interface for the density.5 × 10 . Left: second order method. 0. 3535 5 (ρ. 1.6) The computed density ρ.4) for 0. 28] only by increasing the strength of the right shock wave to pL /pR = 15. 33.

24 .5 0 0 0.2 0.0012. Solid line: the exact solution.4 0.6 0.8 1 x Figure 8.8 1 x 150 100 Velocity 50 0 0 0. Symbol: the numerical solution. velocity and pressure for Example 6.6 0.6 0.4 0.8 1 x 160000 140000 Pressure 120000 100000 0 0.1.5 1 Density 0.2 0.4 0. t = 0.2 0.6: Density.

1. γ ) = (1270. it is more pronounced since the shock wave is much stronger compared to Example 6. 1.05.8) The computed density ρ.9.5. 33. for 0.5. u. u. for x > 0. 0 . 1.4) (3. It appears that our subcell resolution procedure is still not accurate enough to capture the strong generated shock during the instant when the original shock wave impacts on the material interface.0017 against the exact solution in Fig. similar results can also be observed in [13] using the standard scheme from [30].00016 against the exact solution in Fig.5 × 105 . The small glitches on the left constant state can also be observed for the methods in [13. for x > 0. If we increase the strength of the shock to pL /pR = 15. Even though the initial pressure in the gas is extremely high and a very strong shock is generated in the water.9) The computed density ρ.4) (1000.4) (1 . with the initial condition as: √ 105 .solutions. 3535 (ρ. 1. 8. 7. velocity u and pressure p are plotted at t = 0.5. The computed results show that there are slight oscillations near the left shock region.5. 1. which were explained in [13] to be due to the (mis)capturing of the perfect shock initial data by a shock capturing scheme. and there is still no oscillation around the material interface for the density with this stronger strength of the shock. In this example.9 Example 9 This is a gas-water shock tube problem with very high pressure in the gaseous medium taken from [33. 1 × 105 .8. p. the code would blow up. 0. 28].15) for x ≤ 0.249) for x ≤ 0. 0. 1 × 105 . 0. 8.1538. (8.05 < x ≤ 0. γ ) = (1. (8. 25 . 28]. 1 × 105 . but with a reﬂected shock wave [13. p. our computed results still compare well to the exact solutions. 33].8 Example 8 This is a problem of a shock wave refracting at an air-helium interface. 3333 . 0. The initial condition is given as (ρ. 8 × 108 . 8. velocity u and pressure p are plotted at t = 0. 8.

6 0.0005. 26 .2 0.8 1 x 1500 1000 Velocity 500 0 0 0. t = 0.5 4.8 1 x Figure 8.5E+06 Pressure 1E+06 500000 0 0 0. Solid line: the exact solution.6 0.4 0.5 1 0.2 0.5 2 1. Symbol: the numerical solution.5 0 0 0. velocity and pressure for Example 7.4 0.7: Density.8 1 x 1.5 4 3.4 0.5 Density 3 2.2 0.6 0.

2 0.6 0.5 4.8: Density.0017.2 0. velocity and pressure for Example 8.5 1 0 0.8 1 x 100 Velocity 50 0 0 0.4 0. 27 .4 0.6 0.8 1 x 160000 Pressure 140000 120000 100000 0 0.5 4 3.6 0.2 0. Symbol: the numerical solution.8 1 x Figure 8.5 2 1. t = 0.4 0. Solid line: the exact solution.5 Density 3 2.

2 0.4 0.9: Density. t = 0.2 0. Solid line: the exact solution.6 0.6 0.4 0.8 1 x 300 250 200 Velocity 150 100 50 0 -50 0 0.00016.8 1 x 8E+08 6E+08 Pressure 4E+08 2E+08 0 0 0.8 1 x Figure 8.2 0.1300 1200 Density 1100 1000 900 0 0.6 0.4 0. Symbol: the numerical solution. velocity and pressure for Example 9. 28 .

originally studied in [49] for the underwater explosions. (8.8. which do not necessarily land on the desired paths.4) (1000. has been shown to signiﬁcantly improve the capturing of the correct discontinuity waves by deﬁning the correct path in the path integral at discontinuities. 28] can solve this problem well.15) for x ≤ 0. p.5. Our discussion is restricted to the one-dimensional case. however the errors are still reasonably small considering that we are using a nonconservative scheme on this example with very strong discontinuities. u. Preliminary study on the ex29 . 1 × 105 .10) The computed density ρ. 33. we have investigated using the WENO reconstruction with a subcell resolution technique when applied to high order Roe-type schemes for nonconservative Euler equations. as the most probable reason for the failure of schemes to converge to the correct weak solution on the desired paths. the proposed subcell resolution technique within a Runge-Kutta time discretization seems to have robustness problems for very strong discontinuities. 0. An improvement on the robustness of the algorithms when transition points are removed or signiﬁcantly reduced constitutes our future work.10. for x > 0. 1. 7. with the initial condition given as (ρ.10 Example 10 This example increases the energy of the explosive gaseous medium in Example 9. 8.5. which is taken from [13. 33. The subcell resolution to sharpen discontinuities. γ ) = (1630.81 × 109 . 0.0001 against the exact solution in Fig. We begin to see some discrepancies between the exact solution and the numerical solution. especially during interactions of such discontinuities. velocity u and pressure p are plotted at t = 0. 28]. 9 Concluding remarks In this paper. The methods in [13. 7. While this paper identiﬁes the smearing of discontinuities by shock capturing schemes with transition points at the discontinuities. in order to remove or signiﬁcantly reduce transition points at discontinuities.

6 0. velocity and pressure for Example 10. t = 0.4 0. 30 .8 1 x 8E+09 6E+09 Pressure 4E+09 2E+09 0 0 0.2 0.4 0.4 0. Symbol: the numerical solution.6 0.6 0.8 1 x Figure 8.2 0.0001.10: Density. Solid line: the exact solution.1600 1400 Density 1200 1000 0 0.8 1 x 1200 1000 800 Velocity 600 400 200 0 -200 0 0.2 0.

D. 125 (1996). [4] D. Journal of Scientiﬁc Computing. has indicated limitations for some of the complex discontinuity wave patterns. Par´ es. 2759-2763. Ferreiro. Journal of Computational Physics. 150-160. A. Journal of Computational Physics. Acknowledgments We would like to thank Dr.A. Fern´ andez-Nieto. Karni. High order extensions of Roe schemes for two-dimensional nonconservative hyperbolic systems. Computations of compressible multiﬂuids. 289-308. Castro.H. Karni.M. [3] R. 269-277. 31 . Sethian. 594-623.J. 67-114. 169 (2001). References [1] R. A comment on the computation of non-conservative products. [5] E. 229 (2010). Wei Liu for helpful discussions. 39 (2009).tension of our methodology to two-dimensional problems (not reported in this paper). A fast level set method for propagating interfaces. implemented in a dimension by dimension fashion (not dimension splitting). Abgrall and S. Journal of Computational Physics. [6] M. [2] R. E. Journal of Computational Physics. Abgrall and S. A pressure-invariant conservative Godunov-type method for barotropic two-ﬂuid ﬂows. Journal of Computational Physics. Koren. Garc´ ıa-Rodr´ ıguez and C. van Brummelen and B. Adalsteinsson and J. Abgrall. How to prevent pressure oscillations in multicomponent ﬂow calculations: a quasi-conservative approach. This may be due to the complication of two-dimensional waves and the limitation of the subcell resolution technique to capture accurately the correct states and the corresponding paths at the discontinuities. 118 (1995). 185 (2003). Further investigation of eﬀective multi-dimensional algorithm in this context also constitutes our future work.A. J.

Chen and C.J. Par´ es. 545-578. [12] M.P. Mathematics of Computation. 70-88. A simple extension of the Osher Riemann solver to nonconservative hyperbolic systems. International Journal for Numerical Methods in Fluids. [8] M. P. 457-492. Journal of Computational Physics. High order ﬁnite volume schemes based on reconstruction of states for solving hyperbolic systems with nonconservative products. Mu˜ noz-Ruiz and C.P. Journal of Scientiﬁc Computing. Dumbser and E. 137 (1997). 83 (1989). M.F. [11] G. P. 529-541. Merriman. Saurel. Deﬁnition and weak stability of nonconservative products. ENO schemes with subcell resolution. [10] J. [9] T. LeFloch and F. Merriman and S. Castro. T. Gallardo and C. On the Riemann problem for liquid or gas-liquid media. Why many theories of shock waves are necessary: Convergence error in formally path-consistent schemes. A non-oscillatory Eulerian approach to interfaces in multimaterial ﬂows (the Ghost Fluid Method). Journal of Computational Physics. Cocchi and R.M. Journal of Computational Physics. 483-548. Castro.-P. Par´ es. [13] R. Application to shallow-water systems. Harten. Journal of Computational Physics. 11031134. 74 (1995). Cooke.L.[7] M. Aslam. 48 (2011). Osher. 265298. An isobaric ﬁx for the overheating problem in multimaterial compressible ﬂows. Marquina and B. 32 . Murat. A Riemann problem based method for the resolution of compressible multimaterial ﬂows. [14] R. 152 (1999). 18 (1994). Journal of Math Pures Application. LeFloch. Fedkiw. A. J. 75 (2006). [15] A.-J. 148-184.H.J. Fedkiw. Dal Maso. B. Toro. 148 (1999). Journal of Computational Physics. 227 (2008). 8107-8129.

Harten. Journal of Computational Physics. Khoo and K. Larrouturou. [18] T. 231-303. Osher and S.C. Part I: a new methodology with applications to 1D gas-gas and gas-water cases. Computers & Fluids. 29 (1992). III. Jiang and D. Viscous shock proﬁles and primitive formulations. 30 (2001). B.M. Khoo and C. Multicomponent ﬂow calculations by a consistent primitive algorithm. Mathematics of Computation. B. 112 (1994). 315-337. B. The simulation of compressible multi-medium ﬂow. 2126-2143. [19] G. [22] S. Shu. Jiang and C. Journal of Computational Physics. 1592-1609.G. 71 (1987).-W.G. Journal of Computational Physics. S. 291-314. Journal of Computational Physics.-P. Hyman. Liu. Computers & Fluids.C. [26] T. Yeo. 204 (2005). 235-269. Engquist. Wang. B. [24] T. Eﬃcient implementation of weighted ENO schemes. 62 (1994). Self adjusting grid methods for one-dimensional hyperbolic conservation laws.S. 50 (1983). Hou and P. Harten and J. Journal of Computational Physics. The simulation of compressible multi-medium ﬂow.C. How to preserve the mass fractions positivity when computing compressible multi-component ﬂow. 193-221. Karni. The ghost ﬂuid method for compressible gaswater simulation. Peng. Part II: applications to 2D underwater shock refraction. 31-43. 30 (2001). Chakravarthy. 497-530.G. 95 (1991). SIAM Journal on Scientiﬁc Computing. 126 (1996).W. [20] G. Yeo. LeFloch. Why nonconservative schemes converge to wrong solutions: Error analysis. Uniformly high order essentially non-oscillatory schemes. 202-228. SIAM Journal on Numerical Analysis. Khoo and K. [21] S.[16] A.S. 21 (2000). 33 . [25] T. Liu. [23] B. [17] A. Weighted ENO schemes for Hamilton-Jacobi equations. Karni. 31-43.Y. Journal of Computational Physics. Liu.

Mulder.C. 3 (2008). [31] S. Journal of Computational Physics. Par´ es. Khoo and K. Journal of Computational Physics.-W.A. Level Set Methods and Dynamic Implicit Surfaces. Hu and C. [32] C. Qiu. Springer.C. Shu.S. Fedkiw. 190 (2003). T. Sethian. 200-212. Runge-Kutta discontinuous Galerkin methods for compressible two-medium ﬂow simulations: One-dimensional case. 2003. Simulations of compressible two-medium ﬂow by RungeKutta discontinuous Galerkin methods with the Ghost Fluid Method. Quirk and S. [34] J. 10 (2011). B. Osher and R. 175 (2002). T.J.G. On the dynamics of a shock-bubble interaction. [30] W. [36] J.[27] T. 108-127. Weighted essentially non-oscillatory schemes. Osher and J. 318 (1996). A technique of treating negative weights in WENO schemes.P. Osher and T. 44 (2006). 129-163. [28] W. 353-373. Karni. 651-681. Qiu. 34 . Ghost ﬂuid method for strong shock impacting on material interface.-D. 100 (1992). Journal of Computational Physics. Y. 479-504. Chan. C. Liu. Communications in Computational Physics. Liu. S. [35] J. Shu. SIAM Journal of Numerical Analysis. 115 (1994). [33] J. 1238-1248. 300-321. Liu and B. Numerical methods for nonconservative hyperbolic systems: a theoretical framework. Liu and B. Yuan and C. Journal of Fluid Mechanics. Journal of Computational Physics.-W. Journal of Computational Physics. Khoo. 222 (2007). Liu. Communications in Computational Physics.C. [29] X. 209-228. Khoo. Yeo. Computing interface motion in compressible gas dynamics. Shi. A conservative modiﬁcation to the ghost ﬂuid method for compressible multiphase ﬂows. S.

Shu (1998). Journal of Computational Physics. Shu and S. Riemann Solvers and Numerical Methods for Fluid Dynamics: A Practical Introduction. Tadmor (Editor: A. Eﬃcient implementation of essentially non-oscillatory shockcapturing schemes. 229 (2010). Toro. Osher. HLLC-Type Riemann solver for the Baer-Nunziato equations of compressible two-phase ﬂow. [40] J. Journal of Computational Physics. New York. Osher. E. B. 439-471. Cockburn.E. P. Smereka and S.F. Journal of Computational Physics. Sussman. Berlin (2009).[37] C.-W. Springer.-W. Castro. Quarteroni). 360-373. Computers & Fluids. 146-159.A. [45] S. Smoller. Berlin. 3573-3604. Lecture Notes in Mathematics. (1994). Johnson.-W. Tian. 46 (2011). Essentially non-oscillatory and weighted essentially non-oscillatory schemes for hyperbolic conservation laws. [38] C. 114 (1994).F. Toumi. 102 (1992). Shu and E. Springer. Journal of Computational Physics. Shock Waves and Reaction-Diﬀusion Equations. Tokareva and E. 122-132. volume 1697. Springer. [39] C. Eﬃcient implementation of essentially non-oscillatory shockcapturing schemes II. Toro. A weak formulation of Roe’s approximate Riemann solver.-W. 2nd Edition. In Advanced Numerical Approximation of Nonlinear Hyperbolic Equations. A level set approach for computing solutions to incompressible two-phase ﬂow. C. Journal of Computational Physics. 32-78. [41] M. 35 . 77 (1988).F. 3rd Edition. Toro and C. [44] E. Shu and S. 325-432. [43] I. 83 (1989). Osher. A path-conservative method for a ﬁve-equation model of two-phase ﬂow with and HLLC-type Riemann solver. C. [42] B.

Applied Numerical Mathematics. Shu.C. 1998. [50] T. Wang and C.C. [51] J. 278-302. Liu and B.-W. RKDG methods with WENO type limiters and conservative interfacial procedure for one-dimensional compressible multi-medium ﬂow simulations.W. An interface treating technique for compressible multimedium ﬂow with Runge-Kutta discontinuous Galerkin method. Qiu. SIAM Journal on Scientiﬁc Computing. Wang.[46] C. 229 (2010). Shu. A real-ghost ﬂuid for the simulation of multimedium compressible ﬂow.-W. 28 (2006). 190-214. Zhang. T. C. Wang. Xiong. Fast sweeping ﬁfth order WENO scheme for static Hamilton-Jacobi equations with accurate boundary treatment. 45 (2010).-W. Shu. Sj´ ogreen. Underwater Explosion Test Cases. Journal of Computational Physics. 514-536. High order ﬁnite diﬀerence methods with subcell resolution for advection equations with stiﬀ source terms. 554-580. Y. [48] W. C. Yee and B.W.G. Liu and B. Zhu. [49] A. M. Journal of Computational Physics. 61 (2011). Wardlaw. Zhang. Journal of Scientiﬁc Computing. H.G.C. IHTR 2069. Khoo. T. J.-T. 36 . 8823-8843. 231 (2012). [47] C. Khoo.