Elsevier

Journal of Computational Physics

Volume 233, 15 January 2013, Pages 339-358
Journal of Computational Physics

A space–time discontinuous Galerkin method for the incompressible Navier–Stokes equations

https://doi.org/10.1016/j.jcp.2012.08.052Get rights and content

Abstract

We introduce a space–time discontinuous Galerkin (DG) finite element method for the incompressible Navier–Stokes equations. Our formulation can be made arbitrarily high-order accurate in both space and time and can be directly applied to deforming domains. Different stabilizing approaches are discussed which ensure stability of the method. A numerical study is performed to compare the effect of the stabilizing approaches, to show the method’s robustness on deforming domains and to investigate the behavior of the convergence rates of the solution. Recently we introduced a space–time hybridizable DG (HDG) method for incompressible flows [S. Rhebergen, B. Cockburn, A space–time hybridizable discontinuous Galerkin method for incompressible flows on deforming domains, J. Comput. Phys. 231 (2012) 4185–4204]. We will compare numerical results of the space–time DG and space–time HDG methods. This constitutes the first comparison between DG and HDG methods.

Introduction

This is our third article in a series devoted to obtaining accurate and efficient numerical schemes for solving the incompressible Navier–Stokes equations on deforming domains. Examples of applications these schemes may be used for include fluid–structure interaction in wind turbine simulations [14] or flows with free-surfaces, such as water waves and sloshing [24].

Accurately solving partial differential equations by a numerical method on moving and deforming domains, however, is non-trivial. Many schemes fail to automatically preserve uniform flow on moving or deforming meshes. This requirement is used to derive the Geometric Conservation Law (GCL) [17] and in [13], [17] it is shown that the GCL is essential for the time-accuracy of the solution.

Recently, Persson et al. [20] introduced a discontinuous Galerkin (DG) method for solving the compressible Navier–Stokes equations on deformable domains. They introduced a continuous mapping between a fixed reference configuration and the time-varying domain. The Navier–Stokes equations are then written in the reference configuration, where the equations are solved using a DG method in space and Runge–Kutta method in time. This method is simple and allows the use of explicit time-stepping, however, it does not automatically satisfy the GCL and an extra equation needs to be solved to prevent loss of accuracy.

A different approach is that of the space–time DG method (see e.g. [16], [22], [25], [26]). The space–time formulation of the DG method results in a conservative numerical discretization on deforming meshes and it automatically satisfies the GCL. The method is implicit in time and so requires efficient solvers, e.g. multigrid methods [27], [28]. In this article we follow the space–time DG approach.

Our first article in this series introduced the space–time DG formulation for the Oseen equations [26]. The scheme can be made arbitrarily high-order accurate in both space and time and is well suited for hp-adaptation. The method was shown to be unconditionally stable and continuity, coercivity and stability were proven. Furthermore a full hp-error analysis was presented.

In our second article, we introduced a space–time hybridizable DG (HDG) formulation for incompressible flows [22]. We showed that the HDG method displays optimal rates of convergence not only for the velocity and pressure fields, but also for the velocity gradient. This has not yet been achieved with DG methods. Another advantage of an HDG method over the DG method is that for higher order polynomial approximations, the computational cost of the HDG method is smaller than that of the DG method. As discussed in [22], the number of unknowns in the global system of the HDG method is of the order O(pd) compared to O(pd+1) of the DG method, where p is the polynomial order and d is the dimension of the problem. It was also shown in [15] that for p>4, the HDG method can be made more efficient than the continuous Galerkin method.

In this third article, we provide the Navier–Stokes extension of [26] resulting in a space–time DG method that can directly be applied to deforming domain problems. To solve the Navier–Stokes equations, we employ a Picard iteration technique. At every iteration, we therefore need to solve the Oseen equations. Solving the Oseen equations or the Navier–Stokes equations, results in the divergence-free velocity constraint being satisfied only weakly. While this is no problem for the Oseen equations [26], for the Navier–Stokes equations it means that stability of the scheme cannot be proven, unless a post-processing or modification of the weak formulation is applied. For the analysis of the Oseen equations of the space–time DG method, see [26]. The analysis of the DG method for the Oseen equations is given in [6] and for the Navier–Stokes equations in [7], [8], [10]. From numerical results in the literature, it appears that even though we cannot prove stability without modifying the weak formulation or applying a post-processing, the scheme is stable [3], [22], [23]. Our numerical results in Section 4 also show this. We will, however, also discuss two stabilization techniques. The first adds nonconservative terms to the weak formulation [11], [12]. While this is an easy way of ensuring a stable scheme, local conservativity is lost. For the incompressible Navier–Stokes equations, this however might not be problematic as large gradients in the solution, which are associated with shocks and/or large discontinuities, are not present. We also introduce a space–time BDM projection to stabilize the method (see [7], [10] for standard DG BDM projections). The BDM projection is such that if the divergence-free velocity constraint is satisfied weakly, the BDM projected velocity field is exactly divergence free on the element and its jump in the normal direction across element faces is zero. An advantage of this stabilization method is that local conservativity is preserved. The disadvantage, however, is that it is not trivial to obtain arbitrarily high order BDM spaces in higher dimensions. Let pt and ps denote the polynomial order in time and space, respectively. We consider here only 3 dimensional space–time BDM spaces for pt=ps=1 and 2. Future work involves the expansion of space–time BDM spaces to more general situations, including ptps and 4 dimensional space–time.

As mentioned above, we recently introduced a space–time HDG method for incompressible flows [22] with the idea of reducing the globally coupled degrees of freedom in order to obtain a more efficient DG method. In this article we test this statement by comparing the space–time HDG method with the natural extension of the space–time DG method for Oseen equations [26] to the Navier–Stokes equations. Comparisons between HDG and continuous Galerkin methods have been studied in [15], however, this is the first comparison between HDG and DG methods. The comparison is made for polynomial approximations in which pt=ps+1 and pt=ps with ps=1 and ps=2. We show that for these polynomial approximations, the space–time DG method is as efficient, if not more efficient, as the space–time HDG method in terms of L2 errors of the velocity and pressure vs. the total number of degrees of freedom. In terms of CPU times and memory usage, the space–time HDG method seems slightly more efficient.

In [22] we found, for the space–time HDG method, that if we take the polynomial approximation of time equal to that of space, pt=ps, we were unable to obtain optimal rates of convergence for the pressure. Increasing the polynomial approximation in time to one order higher than that in space, pt=ps+1, restored the optimal rates of convergence of the pressure. We show a similar behavior for the space–time DG method.

The outline of this article is as follows. We introduce the space–time formulation of the incompressible Navier–Stokes equations in Section 2. In Section 3 we introduce the space–time DG method for the Navier–Stokes equations, we discuss the stability of the method and introduce the space–time BDM projections. A numerical study in two dimensions on deforming domains will be conducted in Section 4. We will demonstrate convergence properties of the method, compare results of the different stabilization methods and compare results of the space–time DG and the space–time HDG method. For completeness, a short summary of the space–time HDG method for the Navier–Stokes equations is given in Appendix A. Concluding remarks are given in Section 5.

Section snippets

Incompressible Navier–Stokes equations

Denote by ΩtRd a bounded, time-dependent flow domain at time t in d spatial dimensions. Its boundary is denoted by Ωt. Furthermore, let x¯=(x1,x2,,xd) be the spatial variables. The time-dependent Navier–Stokes equations in Ωt are given byui,t+(uiuj),j-ν(ui,j),j+(δijp),j=fi,ui,i=0,inΩt,where uRd is the velocity field, pp/ρR the kinematic pressure, ρ the fluid density, νR+ the kinematic viscosity, and fRd a prescribed external body force. We use here index notation with the summation

The space–time DG method

Divide the domain Ωtn into Nn non-overlapping spatial elements Kn, and similarly for the domain Ωtn+1. A space–time element Kn can now be obtained by connecting Kn to Kn+1 via linear interpolation in time. At curved boundaries, a higher order accurate interpolation may be necessary. The space–time domain Eh, limited to the time interval (tn,tn+1), defines a space–time slab Ehn. The tessellation Thn of Ehn consists of all space–time elements Kn. The tessellation of Eh is given by ThnThn. The

Numerical test cases

We apply now the space–time DG method to different test cases. We will determine rates of convergence, compare the different stabilization methods and give a comparison between results from a space–time DG and HDG method. We recently introduced the space–time HDG method in [22]. A short summary of the method is given in Appendix A.

In the Picard iteration scheme (2), used to solve the Navier–Stokes equations, we consider the scheme to be converged if the residual of the discrete system drops

Conclusions

We have introduced a space–time DG finite element method for the incompressible Navier–Stokes equations. To ensure stability of the method we introduced a space–time BDM-projection and the more simpler stabilization method of adding nonconservative terms to the weak formulation. From numerical experiments, using the BDM stabilization or the nonconservative stabilization improves slightly the efficiency of the method compared to No stabilization, since less Picard iterations are needed per time

Acknowledgments

Sander Rhebergen gratefully acknowledges funding by a Rubicon Fellowship from the Netherlands Organisation for Scientific Research (NWO) and the Marie Curie Cofund Action. Bernardo Cockburn was supported in part by the National Science Foundation (Grant DMS-1115331) and by the University of Minnesota Super Computing Institute.

References (28)

Cited by (87)

  • An unfitted HDG method for a distributed optimal control problem

    2024, Journal of Computational and Applied Mathematics
  • P<inf>0</inf> time/space subcell limiting DG-DGLM method for hyperbolic systems of conservation laws

    2021, Computers and Mathematics with Applications
    Citation Excerpt :

    On the other hand, large time integration is necessary in some simulations and an implicit method in time is appropriate, which requires a less strict CFL condition than the explicit method especially in high order approximation in time. For implicit time integration, space-time DG, [47,48], semi implicit (IMEX) DG, [46,20,37,1], fully implicit (BDF) DG, [14,13], have been actively applied. Apart from that, many variants of DG such as Multiscale DG (MDG), [2,3,24,36] and Hybridizable DG (HDG), [5,42–44,41,26,27,49] have been introduced to reduce the total number of degrees of freedom of DG.

View all citing articles on Scopus
View full text