Augmented FETI-DP Method Based on Polynomial Preconditioning

Date of Submission: 
January 20, 2005
Report Number: 
05-002
Report PDF: 
Abstract: 
To foster the absolute scalability (A-scalability) in solving large-scale non-linear dynamic structural problem on large-processor-count high-performance computing (HPC) systems, this paper examines a scalable implementation of an augmented finite element tearing and interconnecting of dual-primal version (AFETI-DP) method based on generalized lieast-squares polynomial preconditioning. Compared to previous work, the AFETI-DP concerned is featured by utilizing finite element based iterative method to solve the coarse problem and a well load-balance general purpose (applicable to 2D and 3D problems) virtual corner selection strategy to obtain optimal computational performance. From the point of view of scalability, the corresponding experimental results for IBM-SP2 and Cray-X1 are presented and critically assessed.