An 𝒉 ​ 𝒑 \boldsymbol{hp} Multigrid Approach for Tensor-Product Space-Time Finite Element Discretizations of the Stokes Equations

Nils Margenberg, Markus Bause, Peter Munch · 2025

Read An 𝒉 ​ 𝒑 \boldsymbol{hp} Multigrid Approach for Tensor-Product Space-Time Finite Element Discretizations of the Stokes Equations by Nils Margenberg, Markus Bause, Peter Munch (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-09159.txt.

Introduction

Solving flow equations on a mesh that covers space and time together creates some of the largest systems in computing. Nils Margenberg, Markus Bause, and Peter Munch present a solver strategy for that setting in this 2025 paper, aimed at the Stokes equations for slow fluid flow. Their method works on grids built as products of space and time meshes.

The solver is a multigrid scheme that coarsens the mesh in both geometry and polynomial order at once, all written as exact mathematical maps. Special patch-based smoothers, built around cells and corners, tame the saddle structure of the flow problem. Tests on high-performance machines pushed past a trillion unknowns while processing hundreds of millions per second. The authors credit a matrix-free library design for the speed, and frame the work as a path to space-time simulation at scale.

Catalog record

CREATORNils Margenberg, Markus Bause, Peter Munch
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-09159

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

An 𝒉 ​ 𝒑 \boldsymbol{hp} Multigrid Approach for Tensor-Product Space-Time Finite Element Discretizations of the Stokes Equations Thanks: Submitted to the editors DATE. Nils Margenberg, Markus Bause, Peter Munch Abstract We present a monolithic h ​ p hp space-time multigrid method for tensor-product space-time finite element discretizations of the Stokes equations. Geometric and polynomial coarsening of the space–time mesh is performed, and the entire algorithm is expressed through rigorous mathematical mappings. For the discretization, we use inf-sup stable pairs ℚ r + 1 / ℙ r disc \mathbb{Q}_{r+1}/\mathbb{P}_{r}^{\text{disc}} of elements in space and a discontinuous Galerkin (DG ( k ) (k) ) discretization in time with piecewise polynomials of order k k . The key novelty of this work is the application of h ​ p hp multigrid techniques in space and time, facilitated and accelerated by the matrix-free capabilities of the deal.II library. While multigrid methods are well-established for stationary problems, their application in space-time formulations encounter unique challenges, particularly in constructing suitable smoothers. To overcome these challenges, we employ a space-time cell-wise Vanka smoother. Extensive tests on high-performance computing platforms demonstrate the efficiency of our h ​ p hp multigrid approach on problem sizes exceeding a trillion degrees of freedom (dofs), sustaining throughputs of hundreds of millions of dofs per second. keywords Space-time finite elements, space-time multigrid, monolithic multigrid, matrix-free, higher-order finite elements, high-performance computing † † runningheads: An hp Multigrid Approach for space-time finite elements / Markus Bause, Nils Margenberg, Peter Munch MSC 65M60, 65M55, 65F10, 65Y05 1 Introduction Time-dependent partial differential equations, such as the nonstationary Stokes equations, greatly benefit from methods that exploit parallelism in space and time. Space-time finite element methods (STFEMs) offer a natural framework for such parallelization by treating time as an additional dimension, enabling simultaneous discretization and solution in space and time. In this work, we present a hp space-time multigrid method ( hp STMG) for tensor-product space-time finite element discretizations of the Stokes system. For the discretization in space we use the mapped version of the inf-sup stable \mathbb{Q}_{r+1}/\mathbb{P}_{r}^{\text{disc}} pairs of finite elements, with r\in\mathbb{N} . For the discretization in time, we use the discontinuous Galerkin method (DG (k) ) of order k\in\mathbb{N} . Continuous in time Galerkin methods are not studied here, due to their difficulties related to the computation of an discrete initial value for the pressure and the non-wellposedness of the discrete pressure trajectory [ 6 ] . The key novelty of our approach is the application of geometric and polynomial space-time multigrid techniques. The implementation is facilitated by the matrix-free capabilities of the deal.II library [ 2 , 28 , 35 , 17 ] . The present work builds upon the foundation established in [ 32 ] . The code is available on GitHub at https://github.com/nlsmrg/dealii-stfem . Extending multigrid methods to STFEMs poses challenges, particularly in the design of effective smoothers for the arising linear systems. However, STFEMs offer appreciable advantages: They naturally integrate spatial and temporal discretizations and handle coupled problems. Further, they facilitate duality-based and goal-oriented adaptivity in space and time [ 8 , 7 , 42 ] . Adaptive STFEMs have also been investigated in [ 29 , 49 , 12 ] . Alternative approaches, particularly for unstructured space-time meshes are discussed in [ 31 , 45 , 36 , 14 , 30 ] . The utilization of global STFEMs, i.e. the concurrent treatment of all subintervals, has the potential to fully exploit the computational resources of high-performance computers. Conversely, local STFEMs employ space-time variational discretizations as time-marching schemes by selecting a test basis supported on the subintervals. This requires less computational resources, while scalability to global formulations is maintained. Finally, higher order FE spaces facilitate improved accuracy of discrete solutions on computationally feasible grids. Parallel time integration methods have been developed to exploit parallelism in the time dimension and to overcome the sequential bottleneck of traditional time-stepping methods. A comprehensive review of such methods can be found in [ 19 ] . However, most of these methods entail a trade-off between additional computational complexity and time parallelism. An alternative is the all-at-once solution of the entire space–time system [ 13 , 44 ] . Owing to the established connection between Runge–Kutta methods and variational time discretizations [ 20 , 44 ] , these contributions relate to the present work. Space-time multigrid methods treat time as an additional grid dimension, enabling simultaneous multilevel coarsening in space and time [ 24 , 18 , 16 , 23 ] . While algebraic multigrid methods have been applied to space-time systems [ 46 , 30 ] , geometric multigrid technqiues offer advantages in computational efficiency and scalability [ 22 , 20 ] . Another approach to time parallelism that does not increase the computational complexity is stage parallelism within a single time step [ 11 , 39 , 34 ] . While the scalability of stage parallelism is constrained by the number of stages, these methods are effective in the scaling limit. Our goal in this work is the design of a hp STMG with the same grid-independent convergence seen in established geometric multigrid techniques for elliptic or stationary Stokes-type problems [ 37 ] . For stability reasons, the hp STMG is applied as a preconditioner for GMRES iterations, which has become a standard approach in multigrid frameworks. For parallel efficiency, the V -cycle form is used, with a single V -cycle per application of hp STMG. The efficiency of multigrid methods strongly depend on the smoothing operator. We employ a space-time cell-wise Vanka smoother. Additive Schwarz or Vanka-type smoothers have a wide range of applications in fluid mechanics [ 3 , 4 ] , solid mechanics [ 50 ] , fluid-structure interaction [ 15 ] , dynamic poroelasticity [ 5 ] and acoustic wave equations [ 32 ] . To ensure computational efficiency of the hp STMG, we focus on the feasibility of matrix-free implementations. In recent years, there has been work on matrix-free, high-order monolithic multigrid methods to solve Stokes (and Navier–Stokes) equations on high-performance computing architectures. The authors of [ 27 ] developed a parallel, matrix-free multigrid solver achieving “textbook multigrid efficiency”, scaling up to multiple trillions of unknowns. The authors of [ 1 ] extend a patch-based Vanka smoother to fully implicit Runge–Kutta discretizations of incompressible flows using standard Taylor–Hood elements. Jodlbauer et al. used a matrix-free monolithic geometric multigrid solver for discretizations of the Stokes equations with Taylor-Hood elements with scaled Chebyshev-Jacobi smoothers [ 25 ] . Prieto Saveedra et al. present a matrix-free solver for SUPG and PSPG stabilized equal-order discretizations of the incompressible Navier–Stokes equations, and achieve substantial speedups and reduced memory usage compared to matrix-based methods [ 41 ] . The authors of [ 48 ] propose a monolithic ph -multigrid method for stationary Stokes. In line with their findings, our experiments demonstrate that this approach outperforms geometric multigrid methods. This paper is organized as follows. In Section 2 we introduce the continuous problem and the tensor-product space time finite element discretization. We formulate the algebraic system arising from the discretization in Section 3 . In Section 4 we introduce the hp STMG algorithm, which we use as a preconditioner to a GMRES method. We verify this methodology

Read the complete text