Very-Large-Scale GPU-Accelerated Nuclear Gradient of Time-Dependent Density Functional Theory with Tamm-Dancoff...

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• High-Performance Computing

Background:

• Modern graphics processing units (GPUs) offer significant computational power.
• Accurate calculation of molecular properties requires efficient theoretical methods.

Purpose of the Study:

• To present a high-performance, multi-GPU implementation of analytical nuclear gradients for Kohn-Sham time-dependent density functional theory (TDDFT).
• To demonstrate the algorithm's efficiency on large-scale systems.

Main Methods:

• Implementation of multi-GPU algorithms for TDDFT nuclear gradients using the Tamm-Dancoff approximation (TDA) and Gaussian-type atomic orbitals.
• Development of GPU-efficient algorithms for derivatives of electron repulsion integrals and exchange-correlation functionals.
• Application to a large biomolecule (green fluorescent protein) with explicit solvent.

Main Results:

• The TDA-TDDFT gradient calculation for a 4353-atom system was performed.
• Favorable parallel efficiencies were achieved on 256 Nvidia A100 GPUs, with >70% efficiency up to 64 GPUs and 31% with 256 GPUs.

Conclusions:

• The developed multi-GPU implementation effectively leverages high-performance computing resources.
• This method significantly accelerates quantum chemistry calculations for large molecular systems.