Elsevier

Journal of Computational Physics

Volume 375, 15 December 2018, Pages 1238-1269
Journal of Computational Physics

An efficient, second order accurate, universal generalized Riemann problem solver based on the HLLI Riemann solver

https://doi.org/10.1016/j.jcp.2018.09.018Get rights and content

Highlights

  • A universal HLLI Riemann solver for very general hyperbolic systems.

  • A spatial-temporally coupled GRP solver containing sufficient physics.

  • A useful building block to design higher order numerical schemes.

Abstract

The Riemann problem, and the associated generalized Riemann problem, are increasingly seen as the important building blocks for modern higher order Godunov-type schemes. In the past, building a generalized Riemann problem solver was seen as an intricately mathematical task because the associated Riemann problem is different for each hyperbolic system of interest. This paper changes that situation.

The HLLI Riemann solver is a recently-proposed Riemann solver that is universal in that it is applicable to any hyperbolic system, whether in conservation form or with non-conservative products. The HLLI Riemann solver is also complete in the sense that if it is given a complete set of eigenvectors, it represents all waves with minimal dissipation. It is, therefore, very attractive to build a generalized Riemann problem solver version of the HLLI Riemann solver. This is the task that is accomplished in the present paper. We show that at second order, the generalized Riemann problem version of the HLLI Riemann solver is easy to design. Our GRP solver is also complete and universal because it inherits those good properties from original HLLI Riemann solver. We also show how our GRP solver can be adapted to the solution of hyperbolic systems with stiff source terms.

Our generalized HLLI Riemann solver is easy to implement and performs robustly and well over a range of test problems. All implementation-related details are presented. Results from several stringent test problems are shown. These test problems are drawn from many different hyperbolic systems, and include hyperbolic systems in conservation form; with non-conservative products; and with stiff source terms. The present generalized Riemann problem solver performs well on all of them.

Introduction

The Riemann solver has been a building block of Godunov-type schemes since their inception (Godunov [40]). In the Riemann problem, two constant states coming together at a zone boundary provide the input variables for the Riemann solver. The resolved state and flux from the Riemann solver are the desired output variables that we seek from the Riemann solver. Kolgan [46] and van Leer [66] presented the first functional second order Godunov-type scheme. It was based on endowing each zone with a linear profile. Consequently, the input variables for the Riemann solver were not just states from both sides of a zone boundary but also their gradients in one-dimension. The Riemann solver then has to evaluate not just the resolved state and flux but also their variation with time in order to produce a time-centered, second order flux at the given zone boundary. This explains to us the concept of a generalized Riemann solver. The generalized Riemann problem solver can be characterized as a machine that accepts left and right states along with their higher order gradients at a zone boundary. It produces as output the resolved state and the resolved flux and also their time-evolution to the desired order of accuracy.

Progress in the design of second order accurate GRP solvers picked up with the work of Ben-Artzi and Falcovitz [9], [10] and Ben-Artzi [11]. Concerning the GRP solver for real materials, please refer to [52]. A generalization to any general hyperbolic system in self-similar variables was made by LeFloch and Raviart [7] and Bourgeade et al. [19]. However, the work of LeFloch and Raviart is more of a systematization of GRP methodology. The textbook by Ben-Artzi and Falcovitz [13] also helped popularize GRP solvers. The field of GRP design got its second spurt of momentum from the work of Ben-Artzi, Li and Warnecke [14] for the Euler system and Ben-Artzi and Li [15] for general systems. A GRP for shallow water equations was also constructed by Li and Chen [48]. In Qian, Li and Wang [57] a third order accurate GRP solver was also designed for general hyperbolic systems and in Wu, Yang and Tang [68] for the Euler system. For the exact GRP for the Euler system, it has proved to be quite a daunting proposition to go past third order of accuracy. The application of the GRP solver to relativistic fluid dynamics was due to Yang and Tang [69], Wu and Tang [67]. As shown in Li and Wang [51] the analytic and nonlinear derivation of the GRP solver can cope with very extreme challenges such as high temperature and large ratio of density for multi-material flows because thermodynamical effects are precisely merged into the design of the solver.

Approximate solutions to the GRP were also attempted under the rubric of ADER schemes by Toro et al. [65], Titarev and Toro [63], [64], Montecinos and Toro [54] and Castro and Toro [25]. An ADER scheme was also constructed for the MHD system by Taube et al. [62]. Goetz and Iske [41] and Goetz and Dumbser [42] also designed a GRP that was based on the method of LeFloch and Raviart. The second order ADER is the acoustic version of GRP in Ben-Artzi and Falcovitz [13], Ben-Artzi, Li and Warnecke [14], Ben-Artzi and Li [15] and Han et al. [45]. It is fair to observe that with a few exceptions most of these GRP solvers have been restricted to the Euler system. This is because the analytic constructions for the GRP in one way or the other involve the Cauchy–Kovalevskaya procedure. Alternatively, they require a detailed solution of the variation of the flow variables within a rarefaction fan or at a shock. The analytical skills that are involved in the construction of the previously-mentioned GRP solvers have always been a natural barrier in this sort of work.

Many of the above works attempted to construct exact GRP solvers for the associated Riemann problems. This proves to be a formidable task, especially when the hyperbolic system becomes very large, although they can be done in principle (see Ben-Artzi and Li [15], Qian, Li and Wang [57], Li and Sun [50]). One can always ask whether there is a general-purpose strategy for constructing approximate GRPs for entire classes of hyperbolic systems? Based on the application of approximate shock jumps at the wave boundaries of approximate Riemann solvers, Goetz, Balsara and Dumbser ([43]; GBD hereafter) showed that it is possible to design an approximate GRP solver out of the HLL, HLLC and HLLD Riemann solvers. The spatial gradients in the resolved states were found by a least squares procedure. This provided a considerable simplification in the construction of GRP solvers. The HLL-GRP solver is applicable to all manner of conservation laws, but it is rather dissipative in its treatment of intermediate waves. The HLLC-GRP only applies to the Euler system. The HLLD-GRP only applies to the MHD system. In summary, while the work of Goetz, Balsara and Dumbser [43] made it possible to apply the GRP philosophy to several approximate Riemann solvers, it was still specific to a small set of hyperbolic systems. Obtaining a general-purpose GRP solver that applies to any hyperbolic system, whether in conservation form or in non-conservative form, still eluded Goetz, Balsara and Dumbser [43]. This was especially true if one wished to resolve intermediate waves in the hyperbolic system.

What is truly desired is to start with an approximate Riemann solver that is universal and complete and build a GRP version of it. By universal we mean that it applies to any hyperbolic system, whether it is in conservation form or it has non-conservative products. By complete we mean that it resolves the full family of intermediate waves. An approximate Riemann solver that is universal and complete has been presented by Dumbser and Balsara [35]. This Riemann solver is called HLLI because it is built on top of the HLL Riemann solver and it can handle any number of intermediate waves; hence the name “HLLI”. The HLLI Riemann solver follows in the path of the HLLEM Riemann solver (Einfeldt [38], Einfeldt et al. [39]) and was also motivated by recent self-similar formulations of the multidimensional Riemann solver (Balsara [5], Balsara and Dumbser [6], Balsara and Nkonga [7]). The first goal of this paper is to present a second order accurate GRP version of the HLLI Riemann solver of Dumbser and Balsara [35]; we refer to this as the HLLI-GRP solver.

The strategy used in the design of the HLLI-GRP in this paper builds on the approach of GBD. Thus, given states and their higher derivatives on either side of a zone boundary, GBD were able to obtain derivatives of the resolved HLL state within the Riemann fan. This was accomplished with a least squares procedure for conservation laws. In this paper our second goal is to extend that concept to include hyperbolic systems with non-conservative products. Thus a generalized form of shock jump conditions are derived and those jump conditions are used to obtain the gradient of the resolved state within the Riemann fan even when the hyperbolic system has non-conservative products. We then extend the HLLI Riemann solver construction so as to make it second order in time. We, therefore, obtain a flux form HLLI-GRP that applies to conservative systems. However, we also obtain a fluctuation form HLLI-GRP that applies to hyperbolic systems with non-conservative products.

The importance of the GRP in tackling problems involving stiff source terms has also been discussed by BenArtzi [11] and BenArtzi and Birman [12]. In recent years, this importance has also been stressed by Montecinos and Toro [54] and Goetz and Dumbser [42]. Such stiff sources arise in reactive flow calculations where the right temperature can initiate combustion in a large reaction network. Such terms also occur in radiation hydrodynamics, which seeks to couple the radiation field with the hydrodynamic equations. In all such situations, one seeks a GRP solver that is stiffly stable. In this paper our third goal is to address the treatment of stiff source terms in the HLL-GRP and HLLI-GRP solvers. Our approach is based on an innovative ADER scheme that is reported in Balsara et al. [8]. We show how this ADER scheme can be adapted so that it can be invoked within the Riemann fan! The result is that the HLL-GRP and HLLI-GRP solvers inherit all the good A-stable properties of the ADER scheme that is embedded in the GRP solution process.

While Riemann solvers were first developed in the context of hyperbolic conservation laws, they also apply to the treatment of hyperbolic systems with non-conservative products. We follow the DLM theory of Dal Maso, Le Floch and Murat [31], which has been further developed by Parés et al. [56], [55], Castro et al. [21], [22], [23] and Dumbser et al. [36], [37], [34]. The path-conservative schemes proposed in that context are designed to replicate the conventional Riemann solvers, like the Roe Riemann solver, the HLL Riemann solver and the HLLC Riemann solver, when the system is conservative. These schemes effectively derive their inspiration from the parameter vector of Roe (and also some ideas by Osher and Solomon) which says that an effective Roe matrix can be derived by integrating the characteristic matrix through a path of physical states that connect one state to another state in the Riemann problem. The choice of that path is not fully specified, resulting in some ambiguity in the treatment of non-conservative systems; see Castro et al. [24] and Abgrall and Karni [1]. Since the original HLLI Riemann solver was also developed for non-conservative systems, an additional goal of this paper is to make an HLLI-GRP version of it.

The present paper on the HLL-GRP and HLLI-GRP solvers is generally useful for any scheme that wants to reduce the number of stages in the Runge–Kutta procedure. Such a step overcomes many limitations of the Butcher barriers that have plagued Strong Stability Preserving Runge–Kutta (SSP-RK) schemes. Consequently, Li and Du [49] presented a two stage scheme that is fourth order accurate in space and time. This is done by relying on a second order GRP. Likewise, Christlieb et al. [29] and Grant et al. [44] presented analogous two stage schemes that are potentially fourth order accurate. These schemes, therefore, occupy an intermediate position between the SSP-RK schemes (Shu and Osher [60], [61]) and the modern ADER schemes (Dumbser et al. [32], Balsara et al. [2], [4]). They are easier to implement than ADER schemes, yet they do not need to repeat the reconstruction step and Riemann solvers at so many sub-stages like the SSP-RK scheme. GRP solvers have also been used to guide the motion of the mesh in ALE codes (Boscheri, Balsara and Dumbser [17], Boscheri, Dumbser and Balsara [18]). It is not the purpose of this paper to document the many advanced applications of the GRP that we design here. The goal of this paper is to thoroughly document the HLL-GRP and HLLI-GRP solvers for conservative and non-conservative systems and to show that they work well on a range of hyperbolic systems.

The plan of this paper is as follows. In Section 2 we describe the HLLI-GRP solver for conservation laws at second order. In Section 3 we formulate the HLL-GRP and HLLI-GRP solvers for hyperbolic systems with non-conservative products. We also describe the process for obtaining the gradient of the resolved state in the Riemann solver when non-conservative products are present. In Section 4 we describe the HLLI-GRP-based scheme for hyperbolic systems in conservation form as well as hyperbolic systems with non-conservative products. In Section 5 we show how stiff source terms can be included in our formalism. Section 6 presents several test problems. Section 7 draws some conclusions.

Section snippets

An HLLI-GRP solver for conservation laws at second order

In this section we focus on a derivation of the HLLI-GRP solver for conservation laws at second order of accuracy. The derivation is split into three sub-sections. In Sub-section 2.1 we cast the conservation law into similarity variables and find a solution within each constant state. In Sub-section 2.2 we show how this contributes to the formulation of an HLL-GRP for conservation laws. In Sub-section 2.3 we show how this can be transcribed to the formulation of an HLLI-GRP for conservation

An HLLI-GRP solver for hyperbolic systems with non-conservative products at second order

In this section we focus on a derivation of the HLLI-GRP for hyperbolic systems with non-conservative products at second order of accuracy. The derivation is split into three sub-sections. In Sub-section 3.1 we cast the hyperbolic system into similarity variables and find a solution within each constant state. In Sub-section 3.2 we show how this contributes to the formulation of an HLL Riemann solver for hyperbolic systems with non-conservative products. In Sub-section 3.3 we show how to obtain

HLLI-GRP-based scheme for hyperbolic systems

The easiest way to derive the fluctuation form of the HLL-GRP is to start with the conservation law given by eqn. (2.1) and cast it into fluctuation form. For this reason, Sub-section 4.1 describes the use of the HLLI-GRP in the solution of a one-dimensional hyperbolic conservation law. Sub-section 4.2 describes the formulation and use of the HLLI-GRP in the solution of a hyperbolic system in non-conservative form.

Inclusion of stiff source terms in the GRP solver

Several physical problems involving hyperbolic systems also require the inclusion of stiff source terms. In Sub-section 5.1 we provide a quick description of a strategy for including stiff source terms. In Sub-section 5.2 we show how this strategy is incorporated for conservation laws involving stiff source terms. In Sub-section 5.3 we show how this strategy is incorporated for hyperbolic systems with non-conservative products and stiff source terms.

Test problems

To illustrate the versatility of our HLLI-GRP we show that it works well for several different hyperbolic systems of practical interest. The hyperbolic systems we pick include the Euler equations, the MHD equations, the shallow water equations with spatially varying bathymetry and the compressible Navier Stokes equations. The Euler and MHD equations provide examples of a hyperbolic system in conservation form. The shallow water equations serve as an example of a non-conservative system. When

Conclusions

While the generalized Riemann problem has been extensively studied for certain select systems of hyperbolic equations, a general purpose and easily-implementable strategy for obtaining the GRP for any hyperbolic system has been missing in the research literature. This deficiency stems from the fact that the underlying Riemann problems can be vastly different for different hyperbolic systems. If a GRP solver is to be built out of an exact Riemann problem solver, we will necessarily be forced to

Acknowledgements

DSB thanks the hospitality of Institute of Applied Physics and Computational Mathematics, Beijing when this work started. Jiequan Li is supported by NSFC with nos. 11771054 and 11371063, and by Foundation of LCP at Institute of Applied Physics and Computational Mathematics, Beijing.

DSB also acknowledges support via NSF grants NSF-ACI-1533850, NSF-DMS-1622457, NSF_ACI-1713765 and NSF-DMS-1821242. Several simulations were performed on a cluster at UND that is run by the Center for Research

References (69)

  • M. Ben-Artzi et al.

    Computation of reactive duct flows in external fields

    J. Comput. Phys.

    (1990)
  • M. Ben-Artzi et al.

    A direct Eulerian GRP scheme for compressible fluid flows

    J. Comput. Phys.

    (2006)
  • M. Brio et al.

    An upwind differencing scheme for the equations of ideal magnetohydrodynamics

    J. Comput. Phys.

    (1988)
  • W. Boscheri et al.

    Lagrangian ADER-WENO finite volume schemes on unstructured triangular meshes based on genuinely multidimensional HLL Riemann solvers

    J. Comput. Phys.

    (2014)
  • A. Bourgeade et al.

    An asymptotic expansion for the solution of the generalized Riemann problem. II. Application to the equations of gas dynamics

    Ann. Inst. Henri Poincaré, Anal. Non Linéaire

    (1989)
  • P. Colella et al.

    The piecewise parabolic method (PPM) for gas dynamical simulations

    J. Comput. Phys.

    (1984)
  • M.J. Castro et al.

    Why many theories of shock waves are necessary: convergence error in formally path-consistent schemes

    J. Comput. Phys.

    (2008)
  • C.E. Castro et al.

    Solvers for the high-order Riemann problem for hyperbolic balance laws

    J. Comput. Phys.

    (2008)
  • W. Dai et al.

    Extension of the piecewise parabolic method to multidimensional ideal magnetohydrodynamics

    J. Comput. Phys.

    (1994)
  • M. Dumbser et al.

    A unified framework for the construction of quadrature-free one-step finite-volume and discontinuous Galerkin schemes

    J. Comput. Phys.

    (2008)
  • M. Dumbser et al.

    A new, efficient formulation of the HLLEM Riemann solver for general conservative and non-conservative hyperbolic systems

    J. Comput. Phys.

    (2016)
  • M. Dumbser et al.

    ADER schemes on unstructured meshes for non-conservative hyperbolic systems: applications to geophysical flows

    Comput. Fluids

    (2009)
  • M. Dumbser et al.

    FORCE schemes on unstructured meshes II: non-conservative hyperbolic systems

    Comput. Methods Appl. Mech. Eng.

    (2010)
  • B. Einfeldt et al.

    On Godunov-type methods near low densities

    J. Comput. Phys.

    (1991)
  • C.R. Goetz et al.

    A family of HLL-type solvers for the generalized Riemann problem

    Comput. Fluids

    (2018)
  • E. Han et al.

    An adaptive GRP scheme for compressible fluid flows

    J. Comput. Phys.

    (2010)
  • Ph. LeFloch et al.

    An asymptotic expansion for the solution of the generalized Riemann problem. I. General theory

    Ann. Inst. Henri Poincaré, Anal. Non Linéaire

    (1988)
  • J. Li et al.

    Remark on the generalized Riemann problem method for compressible fluid flows

    J. Comput. Phys.

    (2007)
  • J. Li et al.

    Thermodynamical effects and high resolution methods for compressible fluid flows

    J. Comput. Phys.

    (2017)
  • G.I. Montecinos et al.

    Reformulations for general advection–diffusion–reaction equations and locally implicit ADER schemes

    J. Comput. Phys.

    (2014)
  • J. Qian et al.

    The generalized Riemann problems for compressible fluid flows: towards high order

    J. Comput. Phys.

    (2014)
  • C.-W. Shu et al.

    Efficient implementation of essentially non-oscillatory shock capturing schemes

    J. Comput. Phys.

    (1988)
  • C.-W. Shu et al.

    Efficient implementation of essentially non-oscillatory shock capturing schemes II

    J. Comput. Phys.

    (1989)
  • V.A. Titarev et al.

    ADER schemes for three-dimensional nonlinear hyperbolic systems

    J. Comput. Phys.

    (2005)
  • Cited by (23)

    • Implementation of the HLL-GRP solver for multidimensional ideal MHD simulations based on finite volume method

      2023, Journal of Computational Physics
      Citation Excerpt :

      The finite volume method is widely used for MHD numerical simulations due to its easier handling of complex meshes and good conservation [36]. In this paper, we make the first attempt to implement the approximate HLL-GRP solver [18,19] into multidimensional MHD numerical simulations in the context of finite volume method. Different from the attractive genuine multidimensional schemes, e.g., [37–43], which model multidimensional wave propagation as faithfully as possible, the rotation schemes used in this paper, e.g., [44,25,45–49], rotate the normal flux to the x-direction based on the rotational invariance [50] of the MHD equations.

    • Optimal, globally constraint-preserving, DG(TD)<sup>2</sup> schemes for computational electrodynamics based on two-derivative Runge-Kutta timestepping and multidimensional generalized Riemann problem solvers – A von Neumann stability analysis

      2020, Journal of Computational Physics
      Citation Excerpt :

      The GRP solvers, therefore, make it feasible to design SSP-TS schemes for any hyperbolic system. In Goetz et al. [33] and Balsara et al. [19] we presented this novel class of one-dimensional GRP solvers which are capable of making SSP-TS schemes practicable. However, please realize that FVTD and DGTD schemes for CED require multidimensional Riemann solvers; therefore, TDRK-based DGTD schemes for CED require the existence of multidimensional GRP solvers.

    • A simplified Cauchy-Kowalewskaya procedure for the local implicit solution of generalized Riemann problems of hyperbolic balance laws

      2020, Computers and Fluids
      Citation Excerpt :

      Goetz et al. [17] have shown that approximate Riemann solvers can be obtained from the well-known HLL solver. See also [4,16] where universal approximate GRP solvers based on HLL method and the inclusion of intermediate waves have been reported. ADER approach allows the flexibility for incorporating the finite element approach into the finite volume framework.

    View all citing articles on Scopus
    View full text