Mixed domain decomposition for 3D heterogeneous problems - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

Mixed domain decomposition for 3D heterogeneous problems

Résumé

The topic of our presentation focuses on a non-overlapping mixed domain decomposition method. These iterative solvers may face difficulties for heterogeneous models leading to bad conditioning and poor convergence. Recent works about classical overlapping Schwarz methods or primal (BDD) and dual (FETI)non-overlapping domain decomposition methods have overcome these difficulties. Multi-preconditioning as simultaneous FETI[1] or local pre-analysis on subdomain such as GenEO approach [3] are families of method to face the heterogeneity. Our work adapts the GenEO approach to the multi-scale Latin method [2] . We try to select the responsible modes of poor convergence by a local spectral analysis with a generalized eigenvalue problem.These modes are used to build a adapted coarse problem. The particularity of the Latin method is its mixed nature, with Robin parameter also called "search direction". In our presentation we propose to illustrate our developments on 3D heterogeneous test cases. In our presentation we propose to illustrate choice of search directions directly inspired by FETI and BDD scaling. Particularly we compare them through scalability tests. [1] Gosselet,P., Rixen, D., Roux F.-X., Spillane, N.Simultaneous FETI and block FETI: Robust domain decomposition with multiple search directions.Int. J. Num. Meth. Engng.(2015)104:905–927. [2] Oumaziz, P., Gosselet, P., Boucard, P.-A., Abbas, M., A parallel noninvasive multiscale strategy fora mixed domain decomposition method with frictional contact.Int. J. Num. Meth. Engng.(2018)115:893–912. [3] Spillane, N., Rixen, D., Automatic spectral coarse spaces for robust finite element tearing and interconnecting and balanced domain decomposition algorithms.Int. J. Num. Meth. Engng.(2013)95:953–990
Fichier non déposé

Dates et versions

hal-03242975 , version 1 (01-06-2021)

Identifiants

  • HAL Id : hal-03242975 , version 1

Citer

Paul Oumaziz, Pierre Gosselet, Karin Saavedra. Mixed domain decomposition for 3D heterogeneous problems. 14th World Congress on Computational Mechanics (WCCM) -- ECCOMAS Congress 2020, Jan 2021, Paris (virtuel), France. ⟨hal-03242975⟩
50 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More