Conference Presentation (Invited)

FZJ-2019-06910

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png The ChASE library for Large Hermitian Eigenvalue Problems

Di Napoli, E. (Corresponding author)FZJ* ; Winkelmann, J.FZJ* ; Schleife, A.

2019

Please use a persistent id in citations: http://hdl.handle.net/2128/23784

Abstract: ChASE is a new library based on an optimized version of subspace iteration with polynomial acceleration. ChASE is written in C++ using the most current concepts in software engineering which favor a simple integration in application codes and effortless portability over heterogeneous platforms. When solving sequences of Hermitian eigenproblems for a portion of their exterior spectrum, ChASE experiences a considerable speedup and outperforms direct solvers in many scenarios. ChASE performance depends on a number of internal parameters whose choice has been optimized by the automatic estimation of the necessary spectral properties of the eigenproblem. Such feature greatly simplifies its use by the average user. The library is highly scalable, ships with two distinct parallelization schemes and is easily extensible to other computing architectures.

---

Contributing Institute(s):

1. Jülich Supercomputing Center (JSC)

Appears in the scientific report 2019

---

Database coverage:

---