2022 NSF Cyberinfrastructure for Sustained Scientific Innovation Principal Investigator Meeting
Published July 18, 2022 | Version v1
Poster Open

Collaborative Research: SI2-SSI: ELSI-Infrastructure for Scalable Electronic Structure Theory

Authors/Creators

  • 1. Duke University
  • 2. University of California Berkeley
  • 3. Lawrence Berkeley Laboratory
  • 4. Argonne National Laboratory
  • 5. Univ. Valencia, Spain
  • 6. Riken, Japan
  • 7. University of Bonn, Germany
  • 8. ICMAB-CSIC Barcelona, Spain
  • 9. Aalto University, Finland
  • 10. Strathclyde University, UK
  • 11. Fritz Haber Institute of the Max Planck Society, Berlin
  • 12. The Molecular Sciences Software Institute, Blacksburg, VA
  • 13. University of Valencia, Spain

Description

The overarching goal of the ELSI (ELectronic Structure Infrastructure) project is the creation of an open-source software infrastructure that integrates and helps support several leading approaches to solve or to circumvent the eigenvalue problem of electronic structure theory, particularly Kohn-Sham density-functional theory (DFT). Electronic structure theory continues to be one of the most widely important applications of scientific computing. ELSI fosters a stable, sustainable software infrastructure to solve the key bottleneck of DFT in a code-agnostic way. We review ELSI and recent developments including (1) ELPA2-GPU, a scalable, cross-node eigenvalue solver for GPU architectures, (2) a successful, robust mixed-precision approach that approximately halves the eigenvalue solver time in production DFT, and (3) very large DFT calculations without any specific sparsity approximations, carried out on Google's Tensor Processing Units (TPUs) together with a team at Google.

Files

ELSI_poster_2022.pdf

Files (3.1 MB)

Name Size Download all
md5:c86c3559750444f9e7a32d48ef0cb3a0
3.1 MB Preview Download

Additional details

Funding

U.S. National Science Foundation
Collaborative Research: SI2-SSI: ELSI-Infrastructure for Scalable Electronic Structure Theory 1450280