Abstract:To extend the B-spline based s-version of the finite element method (BSFEM), which enables locally high-resolution analysis, to large-scale simulations, domain-decomposition-based parallel computation on distributed-memory systems is indispensable. However, the highly complex interactions induced by the spatial overlap of independent meshes make the construction of interprocessor communication structures extremely challenging, thereby posing a major obstacle to the application of conventional domain decomposition approaches. In this study, we propose a method that mathematically generalizes and simplifies the complex connectivity arising from the physical overlap of meshes as a weighted graph. By reducing the problem to graph theory, we successfully construct a domain-decomposition-based parallelization algorithm for BSFEM and SFEM-based methods with complex mesh structures in distributed-memory environments. To demonstrate the effectiveness of the proposed framework, numerical verification is performed using the three-dimensional steady Poisson equation, confirming the realization of parallel computation in a distributed-memory environment. Furthermore, we demonstrate that assigning weights to graph nodes effectively distributes localized computational loads and substantially reduces the total computational time. The graph-theoretic approach established in this study not only enables the parallelization of BSFEM and SFEM-based methods, but also provides a general theoretical foundation for designing parallelization strategies for other methods with complex communication structures.
From: Nozomi Magome [view email]
[v1]
Fri, 19 Jun 2026 06:28:12 UTC (9,545 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。