PySEQM 2.0: Accelerated Semiempirical Excited State Calculations on Graphical Processing Units

TL;DR

This paper introduces PySEQM 2.0, a GPU-accelerated software for rapid semi-empirical excited state calculations using CIS and TDHF methods, enabling large molecule simulations and machine learning integration.

Contribution

The paper presents a GPU-accelerated implementation of semi-empirical excited state methods in PySEQM, significantly improving computational speed and enabling machine learning applications.

Findings

Excited state calculations for large molecules can be done in under a minute.

GPU acceleration provides substantial speed-up over CPU-based methods.

The software supports machine learning integration for parameter optimization.

Abstract

We report the implementation of electronic excited states for semi-empirical quantum chemical methods at the configuration interaction singles (CIS) and time-dependent Hartree-Fock (TDHF) level of theory in the PySEQM software. Built on PyTorch, this implementation leverages GPU acceleration to significantly speed up molecular property calculations. Benchmark tests demonstrate that our approach can compute excited states for molecules with nearly a thousand atoms in under a minute. Additionally, the implementation also includes a machine learning interface to enable parameters re-optimization and neural network training for future machine learning applications for excited state dynamics.