Challenges and Opportunities for Electronic Structure Calculations at Exascale

At the RIKEN Center for Computational Science, we are currently co-designing a successor to our supercomputer Fugaku. Our goal is to enable 100x speedup on applications, including those relevant for materials and chemistry, when the machine is deployed in 2030. However, achieving this boost in scientific capability will not be straightforward. In this talk, I will present some of the challenges we are facing --- and solutions we are finding --- to accelerate simulations in an increasingly AI focused hardware ecosystem. I will highlight our progress in simulation software based on Density Functional Theory and Selected Configuration Interaction, and present future algorithmic strategies including memory partitioning, floating point emulation, coupling of heterogeneous workflows, and agentic AI.

Seminar realized in the framework of the funded project
- HORIZON-EUROHPC-JU-2022-INCO-04 - HANAMI - project nr. 101136269.