Learning Overlap Optimization for Domain Decomposition Methods
نویسندگان
چکیده
The finite element method is a numerical simulation technique for solving partial differential equations. Domain decomposition provides a means for parallelizing the expensive simulation with modern computing architecture. Choosing the sub-domains for domain decomposition is a non-trivial task, and in this paper we show how this can be addressed with machine learning. Our method starts with a baseline decomposition, from which we learn tailored sub-domain overlaps from localized neighborhoods. An evaluation of 527 partial differential equations shows that our learned solutions improve the baseline decomposition with high consistency and by a statistically significant margin.
منابع مشابه
OPTIMAL DECOMPOSITION OF FINITE ELEMENT MESHES VIA K-MEDIAN METHODOLOGY AND DIFFERENT METAHEURISTICS
In this paper the performance of four well-known metaheuristics consisting of Artificial Bee Colony (ABC), Biogeographic Based Optimization (BBO), Harmony Search (HS) and Teaching Learning Based Optimization (TLBO) are investigated on optimal domain decomposition for parallel computing. A clique graph is used for transforming the connectivity of a finite element model (FEM) into that of the cor...
متن کاملUpdating finite element model using frequency domain decomposition method and bees algorithm
The following study deals with the updating the finite element model of structures using the operational modal analysis. The updating process uses an evolutionary optimization algorithm, namely bees algorithm which applies instinctive behavior of honeybees for finding food sources. To determine the uncertain updated parameters such as geometry and material properties of the structure, local and...
متن کاملOptimized Schwarz Methods with Overlap for the Helmholtz Equation
For the Helmholtz equation, simple absorbing conditions of the form ∂n − iω were proposed as transmission condition (TC) in Schwarz methods first without overlap in [4], and later also with overlap, see [3, 12]. More advanced TCs can also be used, see e.g. [11, 14, 2]. Furthermore, parameters can be introduced into TCs and then optimized for rapid convergence, which led to the so called optimiz...
متن کاملExperimental Optimization of Parallel 3D Overlapping Domain Decomposition Schemes
Overlapping domain decomposition is a technique for solving complex problems described by Partial Differential Equations in a parallel framework. The performance of this approach strongly depends on the size and the position of the overlap since on the one hand more overlap increases the computational costs for the single subdomain but on the other hand it accelerates the iterative procedure of...
متن کاملNearwell local space and time refinement in reservoir simulation
Abstract. In reservoir simulations, nearwell regions usually require finer space and time scales compared with the remaining of the reservoir domain. We present a domain decomposition algorithm for a two phase Darcy flow model coupling nearwell regions locally refined in space and time with a coarser reservoir discretization. The algorithm is based on an optimized Schwarz method using a full ov...
متن کامل