Analysis and Preconditioning of a Probabilistic Domain Decomposition Algorithm for Elliptic Boundary Value Problems

We perform a rigorous numerical analysis of the PDDSparse algorithm [Bernal, Morón-Vidal & Acebrón, Comp.Math.& App. 146 (2023)] for domain decomposition of partial differential equations. PDDSparse is a novel, recently introduced parallelisation paradigm for solving large-scale elliptic bou...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Journal of scientific computing Ročník 105; číslo 1; s. 17
Hlavní autori: Bernal, Francisco, Morón-Vidal, Jorge
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: New York Springer US 01.10.2025
Springer Nature B.V
Predmet:
Algorithms
Approximation
Boundary conditions
Boundary value problems
Computational Mathematics and Numerical Analysis
Computer simulation
Decomposition
Domain decomposition methods
Heuristic
Knots
Linear algebra
Mathematical analysis
Mathematical and Computational Engineering
Mathematical and Computational Physics
Mathematics
Mathematics and Statistics
Matrix algebra
Monte Carlo simulation
Numerical analysis
Partial differential equations
Preconditioning
Simulation
Stiffness matrix
Theoretical
M-matrix
Feynman-Kac formula
65Y05
High performance computing (HPC)
65F50
Numerical PDE
GMRES
Condition number
65D05
65C05
Domain decomposition
Preconditioning
65F08
65N75
ISSN:0885-7474, 1573-7691
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:We perform a rigorous numerical analysis of the PDDSparse algorithm [Bernal, Morón-Vidal & Acebrón, Comp.Math.& App. 146 (2023)] for domain decomposition of partial differential equations. PDDSparse is a novel, recently introduced parallelisation paradigm for solving large-scale elliptic boundary value problems on supercomputers, which can be described as a Feynman-Kac formula for domain decomposition. At its core lies a linear sparse system for the solutions on the interfaces, whose entries can be generated via Monte Carlo simulations. The (slightly random) system matrix, G ~ ( ω ) , can be thought of as equivalent to the Schur complement of the stiffness matrix in standard substructuring domain decomposition. Asymptotically, G ~ ( ω ) is shown to be near a nonsingular M-matrix G , i.e. G ~ ( ω ) + E = G where || E ||/|| G || is small. Combining stochastic calculus and matrix algebra, we prove that G is stable in the sense that its condition number grows moderately with respect to the discretisation parameters of the PDE (nodes and subdomains). This is experimentally confirmed with examples. Furthermore, the rich algebraic structure of the stylised matrix G paves the way for a tailored preconditioner to the numerically feasible matrix G ~ ( ω ) . Specifically, the truncated Neumann series of G - 1 is augmented with an incomplete Arnoldi procedure, for a well behaved approximation to G ~ - 1 ( ω ) . The preconditioner is fast to set up and to run, parallelisable, and efficient, despite the fact that G ~ ( ω ) is indefinite and asymmetric. These findings are again supported by numerical evidence. Summing up, this paper puts the PDDSparse algorithm on a sound theoretical footing, and equips it with a very promising preconditioning scheme. Both of them were prerequisites to successfully tackle, in future work, the massively parallel simulations that PDDSparse is intended for.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0885-7474
1573-7691
DOI:10.1007/s10915-025-03043-4