Date: Wednesday, Nov. 11, 2026, at 12:00 p.m.

Location: Coda, Room 230 (Google Maps link)

Lunch provided!

Title: Algorithms for Random Phase Approximation Correlation Energy Computations on Exascale Architectures

Abstract: This presentation presents massively parallel algorithms for computing the random phase approximation (RPA) correlation energy within Kohn-Sham density functional theory. Standard RPA formulations are extremely expensive for large chemical systems due to their quartic algorithmic scaling, which is typically associated with the explicit construction and eigen-decomposition of the large response function matrix. We discuss recent advancements in the SPARC electronic structure software package that permit cubic algorithmic scaling and nearly ideal speedups on the Frontier supercomputer. In particular, we use Lanczos quadrature to evaluate the trace of a matrix function on a real-space grid. We show it is possible to exploit the spatial smoothness of the RPA correlation energy density in real-space to reduce computational cost. We extend this method to multi-GPU systems by employing block Lanczos quadrature. Our matrix-free matrix-vector multiplication operator involves solving the complex symmetric Sternheimer equations in mixed precision across many complex diagonal shifts and right-hand side (RHS) vectors with coarse convergence tolerances. This is the first RPA formulation that solves the Sternheimer equations on distributed GPUs in mixed precision.We compute the RPA correlation energy for a lithium-hydride system with 512 valence electrons in less than 20 minutes on 512 MI250X GPUs. The proposed multi-GPU formulation achieves speedups of up to 74.4x in terms of node-hours across 128 nodes on the Frontier supercomputer versus a CPU-only baseline.  

Bio: Abir Haque is a PhD student in the College of Computing at Georgia Tech advised by Prof. Edmond Chow. His research focuses on developing novel high-performance computing and numerical linear algebra techniques that support scientific applications. He is currently focusing on accelerating electronic structure calculations by developing theory, algorithms, and software for the Simulation Package for Ab-initio Real-space Calculations (SPARC).

About HotCSE

HotCSE is an academic seminar series to bring Ph.D. students in Computational Science and Engineering together to discuss interesting topics. The topics consist of high-performance computing, machine learning, data analysis, simulation, computational sustainability, medical informatics, etc.

The talks have always been enjoyable and have ranged from quite informal to formal conference style talks. Either chalks or slides can be used to help people understand your talk. It is also a great forum to practice conference talks and bounce around new ideas.

Currently the talks are sponsored by the School of Computational Science and Engineering. The goal of CSE is slightly broader than that of these talks - we want to bring more people outside CSE to discuss their related work here.