COMPUTATION RESEARCH in BOSTON and BEYOND (CRiBB)
Date November 6, 2015
Speaker Eric Polizzi University of Massachusetts, Amherst
Topic FEAST Eigensolver: Breakthrough, Challenges and Opportunities
Abstract

The FEAST solver (www.feast-solver.org) is a new open source numerical library for solving the eigenvalue problem, which has also been recently integrated into the INTEL-MKL library. The solver is based on an innovative numerical algorithm - the FEAST algorithm - which represents a departure from "textbook approaches" for solving the eigenvalue problem.FEAST is using elements of complex analysis, numerical linear algebra and approximation theory, to offer a set of important capabilities for achieving high-performance, robustness, accuracy, and potential for linear scalability on parallel architectures. Since the first FEAST publication in 2009, we have been working on (i) establishing a solid mathematical foundation for the algorithm that successfully demonstrates its fast convergence property [SIAM. J. Matrix Anal. & Appl. 35, 354 (2014)]; (ii) advancing the algorithm development to broaden the class of eigenvalue problems that can be addressed (e.g. non-Hermitian and non-linear); as well as (iii) enhancing its multilevel parallel performances to achieve load balancing on high-end computing platform. Finally, we will also discuss how FEAST has considerably broaden the perspectives for enabling reliable and high-performance first-principle electronic structure calculations with applications in computational material nanosciences and device nanoengineering. Bio: Prof. Eric Polizzi, Associate Professor in the Department of Electrical and Computer Engineering and Department of Mathematics and Statistics, University of Massachusetts, Amherst


back