A Novel Matrix-Free Approximate Elemental Inverse for Efficient Preconditioning of Discontinuous Galerkin Spectral Elements
Abstract
Efficient matrix-free preconditioning remains a challenge for discontinuous Galerkin spectral element discretisations of elliptic problems, particularly for general element shapes with collapsed-coordinate bases, such as prisms, tetrahedra and pyramids. Multigrid provides a promising framework, but its efficiency depends strongly on the effectiveness and computational cost of the smoother. For modal bases, point Jacobi can provide insufficient smoothing, whereas applying block Jacobi through stored dense inverses incurs substantial storage and computational costs. We propose a matrix-free approximate elemental inverse, an approximation to the inverse of each elemental diagonal block, constructed from inverse modal-to-nodal transforms and an element-dependent sparse matrix. Its application exploits generalised sum-factorisation and partial sum-factorisation in collapsed coordinates, avoiding the explicit storage and multiplication of large dense matrices. The method is evaluated as both a standalone preconditioner and a multigrid smoother on hexahedral, prismatic, tetrahedral and pyramidal elements. Although the proposed approximation generally increases iteration counts compared to exact block Jacobi, the cost per iteration is far lower. This trade-off yields substantial reductions in total solution time for hexahedral and prismatic elements, while achieving comparable performance for tetrahedral and pyramidal elements. This study can also serve as a practical guide for tuning the convergence of the discontinuous Galerkin interior-penalty Poisson solver.
Related
- Preconditioning for the time-dependent Stokes equations: stage-weighted Radau IIA and weighted discontinuous Galerkin time-stepping
- A Discontinuous Galerkin Framework for Efficient Simulation of Cyclic Voltammetry Models
- Discrete hypocoercive estimates for discontinuous Galerkin methods: application to the Vlasov-Poisson-Fokker-Planck system
- The discontinuous Galerkin method for solving the two-dimensional transport equation with distributed-order memory kernel on triangular meshes
- Hybridizable Discontinuous Galerkin methods for the Keller-Segel model
BibTeX
@misc{xia2026novel,
title = {{A Novel Matrix-Free Approximate Elemental Inverse for Efficient Preconditioning of Discontinuous Galerkin Spectral Elements}},
author = {Boyang Xia and David Moxey},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.31982},
eprint = {2610.31982},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.31982},
}
RIS
TY - UNPB
TI - A Novel Matrix-Free Approximate Elemental Inverse for Efficient Preconditioning of Discontinuous Galerkin Spectral Elements
AU - Boyang Xia
AU - David Moxey
PY - 2026
DA - 2026/10/22
PB - arXiv
DO - 10.48550/arxiv.2610.31982
UR - https://arxiv.org/abs/2610.31982
AB - Efficient matrix-free preconditioning remains a challenge for discontinuous Galerkin spectral element discretisations of elliptic problems, particularly for general element shapes with collapsed-coordinate bases, such as prisms, tetrahedra and pyramids. Multigrid provides a promising framework, but its efficiency depends strongly on the effectiveness and computational cost of the smoother. For modal bases, point Jacobi can provide insufficient smoothing, whereas applying block Jacobi through stored dense inverses incurs substantial storage and computational costs. We propose a matrix-free approximate elemental inverse, an approximation to the inverse of each elemental diagonal block, constructed from inverse modal-to-nodal transforms and an element-dependent sparse matrix. Its application exploits generalised sum-factorisation and partial sum-factorisation in collapsed coordinates, avoiding the explicit storage and multiplication of large dense matrices. The method is evaluated as both a standalone preconditioner and a multigrid smoother on hexahedral, prismatic, tetrahedral and pyramidal elements. Although the proposed approximation generally increases iteration counts compared to exact block Jacobi, the cost per iteration is far lower. This trade-off yields substantial reductions in total solution time for hexahedral and prismatic elements, while achieving comparable performance for tetrahedral and pyramidal elements. This study can also serve as a practical guide for tuning the convergence of the discontinuous Galerkin interior-penalty Poisson solver.
ER -