Partitioned Coupling vs. Monolithic Block-Preconditioning Approaches for Solving Stokes-Darcy Systems
Authors:
Jenny Schmalfuss,
Cedric Riethmüller,
Mirco Altenbernd,
Kilian Weishaupt,
Dominik Göddeke
Abstract:
We consider the time-dependent Stokes-Darcy problem as a model case for the challenges involved in solving coupled systems. Keeping the model, its discretization, and the underlying numerics for the subproblems in the free-flow domain and the porous medium constant, we focus on different solver approaches for the coupled problem. We compare a partitioned coupling approach using the coupling librar…
▽ More
We consider the time-dependent Stokes-Darcy problem as a model case for the challenges involved in solving coupled systems. Keeping the model, its discretization, and the underlying numerics for the subproblems in the free-flow domain and the porous medium constant, we focus on different solver approaches for the coupled problem. We compare a partitioned coupling approach using the coupling library preCICE with a monolithic block-preconditioned one that is tailored to different formulations of the problem. Both approaches enable the reuse of already available iterative solvers and preconditioners, in our case, from the DuMux framework. Our results indicate that the approaches can yield performance and scalability improvements compared to using direct solvers: Partitioned coupling is able to solve large problems faster if iterative solvers with suitable preconditioners are applied for the subproblems. The monolithic approach shows even stronger requirements on preconditioning, as standard simple solvers fail to converge. Our monolithic block preconditioning yields the fastest runtimes for large systems, but they vary strongly with the preconditioner configuration. Interestingly, using a specialized Uzawa preconditioner for the Stokes subsystem leads to overall increased runtimes compared to block preconditioners utilizing a more general algebraic multigrid. This highlights that optimizing for the non-coupled cases does not always pay off.
△ Less
Submitted 30 August, 2021;
originally announced August 2021.
Exa-Dune -- Flexible PDE Solvers, Numerical Methods and Applications
Authors:
Peter Bastian,
Mirco Altenbernd,
Nils-Arne Dreier,
Christian Engwer,
Jorrit Fahlke,
René Fritze,
Markus Geveler,
Dominik Göddeke,
Oleg Iliev,
Olaf Ippisch,
Jan Mohring,
Steffen Müthing,
Mario Ohlberger,
Dirk Ribbrock,
Nikolay Shegunov,
Stefan Turek
Abstract:
In the Exa-Dune project we have developed, implemented and optimised numerical algorithms and software for the scalable solution of partial differential equations (PDEs) on future exascale systems exhibiting a heterogeneous massively parallel architecture. In order to cope with the increased probability of hardware failures, one aim of the project was to add flexible, application-oriented resilien…
▽ More
In the Exa-Dune project we have developed, implemented and optimised numerical algorithms and software for the scalable solution of partial differential equations (PDEs) on future exascale systems exhibiting a heterogeneous massively parallel architecture. In order to cope with the increased probability of hardware failures, one aim of the project was to add flexible, application-oriented resilience capabilities into the framework. Continuous improvement of the underlying hardware-oriented numerical methods have included GPU-based sparse approximate inverses, matrix-free sum-factorisation for high-order discontinuous Galerkin discretisations as well as partially matrix-free preconditioners. On top of that, additional scalability is facilitated by exploiting massive coarse grained parallelism offered by multiscale and uncertainty quantification methods where we have focused on the adaptive choice of the coarse/fine scale and the overlap region as well as the combination of local reduced basis multiscale methods and the multilevel Monte-Carlo algorithm. Finally, some of the concepts are applied in a land-surface model including subsurface flow and surface runoff.
△ Less
Submitted 6 November, 2019; v1 submitted 4 November, 2019;
originally announced November 2019.
