AI & ML
Near-optimal convergence of Widlund's method and numerical exploration of different implementations
arXiv cs.NA0 views
arXiv:2611.23829v1 Announce Type: new Abstract: For linear systems arising from the discretization of certain classes of mathematical models, the splitting of the linear system into its Hermitian and skew-Hermitian parts has a physical meaning. If, in addition, the Hermitian part is positive definite, there exist Krylov subspace methods based on three-term recurrences where the Hermitian part is used as a preconditioner. Two such Krylov subspace methods are Widlund's method and Rapoport's method. In this paper we establish a ``near-optimal'' guarantee for the convergence of Widlund's method, showing that the overall convergence of Widlund's method is nearly as good as that of Rapoport's method. Specifically, we show that for every iteration $k$, there exists an iteration $j\leq k$ where the residual norm of Widlund's method is less than $\sqrt{k+1} $ times the residual norm of Rapoport's method at iteration $k$. This bound is sharp. We also provide a computational study on various implementations of Widlund's method in the exact and inexact cases, and show that certain implementations may stagnate while others do not, i.e., we look at the robustness of the different implementations. We recommend two implementations.
Source:https://arxiv.org/abs/2611.23829PreviousSplit-and-Learn: Decoupled Learning for Reactive Transport in Heterogeneous Marine SedimentsNextAsymptotic error of the accelerated exponential Euler scheme for semilinear parabolic SPDEs driven by fractional Brownian motion
Related posts

So sicherst du einen MCP-Server mit codex sandbox ab
0dev.to

Perfiles de permisos de Codex: ejemplos de config.toml para ejecuciones seguras de agentes
0dev.to

GPTBot vs OAI-SearchBot: How to Configure robots.txt for AI Crawlers
0dev.to

ملفات تعريف أذونات Codex: أمثلة config.toml لتشغيل الوكلاء بأمان
0dev.to

Write a cover letter with AI and keep every fact yours: a nine-step loop
0dev.to
Reading project health off a download page — faster than star count
0dev.to