COMPARING PARTITIONED COUPLING VS. MONOLITHIC BLOCK PRECONDITIONING FOR SOLVING STOKES-DARCY SYSTEMS

COMPARING PARTITIONED COUPLING VS. MONOLITHIC BLOCK PRECONDITIONING FOR SOLVING STOKES-DARCY SYSTEMS

PID (if applicable): arxiv:2108.13229

Problem Statement

Instationary, coupled Stokes-Darcy two-domain problem: coupled systems of free flow adjacent to permeable (porous) media Two approaches ( partioned coupling and monolitihic block preconditioning) are compared against each other and with the direct solving.

Object of Research and Objective

Analysis of the runtime and memory behavior of the two methods " partioned coupling" and "monolitihic block preconditioning" in comparison to the direct solver. The motivation for this is to avoid problems when using simple direct solvers (sparse direct solvers for the (linearized) subproblems): bad parallel scaling, untrustworthy solution with bad conditioning.

Procedure

Perform three types of solving a Darcy-Stokes system, namely Partioned Coupling, Monolithic Block Preconditioning, and Direct Solving, with subsequent comparison of runtime and memory behavior.

Involved Disciplines

Environmental Systems, Mathematics

Model

Stokes flow in the free-flow domain Darcy’s law for the porous domain

Discretization

• Time: first-order backward Euler scheme
• Space: finite volumes
  • Porous Domain (Darcy): mit two-point flux approximation for pressure
  • Free Flow domain (Stokes): staggered grid for pressure and velocity, upwind scheme for approximation of fluxes

Variables

Name	Unit	Symbol
Pressure (Dirichlet pressure)	-	$p$
Velocity (Neumann velocity)	-	$v$

Process Informationen

Process Steps

Name	Description	Input	Output	Method	Parameter	Environment	Mathematical Area
Partioned Coupling Solving	Numerical Solution of the Problem with Partioned Coupling	LGS; $S^{ff}$: $v^{pm}*n$; $S^{pm}$: $p^{ff}$	LGS_solved_cp, runtime_cp	Solver (UMFPACK, PD-GMRS, Bi-CGSTAB), Preconditioner (Uzawa, AMG), Coupling (preCICE, Picard iteration; inverse least-squares interface quasi-Newton)		DuMux, preCICE, DUNE	Numerical Mathematics
Monolithic Block-Preconditioning Solving	Solving with Preconditioner+Solver	LGS	LGS_solved_mlbp, runtime_mlbp	Solver: PD-GMRES, Preconditioner: AMG, Uzawa, ILU(0), Block-Jacobi, Block-Gauss-Seidel		DuMux	Numerical Mathematics
Direct Solving	Solving with Preconditioner+Solver	LGS	LGS_solved_direct, runtime_direct	Solver: UMFPACK		DuMux
Comparison	Comparison of the three solution approaches with regard to runtime, scalability and memory requirements	runtime_cp, runtime_mlbp, runtime_direct	Graph comparison of the runtime, Paper		Degrees of Freedom		Numerical Mathematics

Applied Methods

ID	Name	Process Step	Parameter	implemented by
wikidata:Q7001954	Dirichlet-Neumann coupling	Coupling
wikidata:Q1683631	Picard iteration mit fixed-point iteration	Coupling
wikidata:Q25098909	inverse least-squares interface quasi-Newton	Coupling
wikidata:Q1069090	block-Gauss-Seidel method	Preconditioner
wikidata:Q2467290	Umfpack	Solver
wikidata:Q56564057	PD-GMRES	Solver	k (subiteration parameter, determined automatically), tolerance: relative residual...
wikidata:Q56560244	Bi-CGSTAB	Solver	tolerance: relative residual...
wikidata:Q1471828	AMG method	Preconditioner
wikidata: Q17144437	Uzawa-iterations	Preconditioner
wikidata:Q1654069	ILU(0) factorization	Preconditioner

Software used

ID	Name	Description	Version	Programming Language	Dependencies	versioned	Published	Documented
sw:8713	preCICE	Library for coupling simulations	v2104.0	Core library in C++; Bindings for Fortran, Python, C, Matlab; Adaptors dependant on simulation code (Fortran, Python, C, Matlab)	Linux, Boost, MPI, ...	https://github.com/precice/precice	https://doi.org/10.18419/darus-2125	https://precice.org/docs.html
sw:14293	DuMux	DUNE for Multi-{Phase, Component, Scale, Physics, …} flow and transport in porous media		C++, python-bindings, utility-skripts in python	Linux, DuNE (C++ Framework), cmake (module chains), package-config, compiler, build-essentials, dpg	https://git.iws.uni-stuttgart.de/dumux-repositories/dumux	https://zenodo.org/record/5152939#.YQva944zY2w	https://dumux.org/docs/
sw:18749	ISTL	Iterative Solver Template Library” (ISTL) which is part of the “Distributed and Unified Numerics Environment” (DUNE).		C++	Linux, DuNE (C++ Framework)	https://gitlab.dune-project.org/core/dune-istl	https://doi.org/10.1007/978-3-540-75755-9_82	https://www.dune-project.org/modules/dune-istl/

Hardware

ID	Name	Processor	Compiler	#Nodes	#Cores
		AMD EPYC 7551P CPU		1	1

Input Data

ID	Name	Size	Data Structure	Format Representation	Format Exchange	Binary/Text	Proprietary	To Publish	To Archive
	LGS	O($2*10^6$) Matrix size	Data structure in DUNE/DuMux			numbers	open	?	?

Output Data

ID	Name	Size	Data Structure	Format Representation	Format Exchange	Binary/Text	Proprietary	To Publish	To Archive
	Runtime comparison of the direct solver Umfpack and iterative solvers with partitioned coupling and block preconditioning	small	statistics			numbers		?	?
	Runtime comparison of the best performing solver configurations	small	statistics			numbers		?	?
arxiv:2108.13229	Paper	O(KB)			pdf	text	open	yes	yes

Reproducibility

Mathematical Reproducibility

Yes, by all parameters

Runtime Reproducibility

Yes, for same input samples

Reproducibility of Results

Due to floating point arithmetic, no bitwise reproducibility

Reproducibility on other Hardware

a) Serial Computation

b) Parallel Computation

Transferability to

a) similar model parameters (other initial and boundary values)

b) other models

Legend

The following abbreviations are used in the document to indicate/resolve IDs:

doi: DOI / https://dx.doi.org/

sw: swMATH / https://swmath.org/software/

wikidata: https://www.wikidata.org/wiki/