GPU acceleration of ab initio simulations of large-scale identical particles based on path integral molecular dynamics

TL;DR

This paper demonstrates significant GPU acceleration for ab initio path integral molecular dynamics simulations of large-scale identical particles, enabling efficient simulation of systems with tens of thousands of particles.

Contribution

The authors developed an open-source GPU-accelerated PIMD code that does not rely on third-party libraries, significantly improving simulation efficiency for large particle systems.

Findings

GPU acceleration reduces simulation time substantially compared to CPU.

Able to simulate up to 40,000 particles with a 24GB GPU.

Simulation time scales roughly linearly with the number of particles.

Abstract

Path integral Monte Carlo (PIMC) and path integral molecular dynamics (PIMD) provide the golden standard for the ab initio simulations of identical particles. In this work, we achieved significant GPU acceleration based on PIMD, which is equivalent to PIMC in the ab initio simulations, and developed an open-source PIMD code repository that does not rely on any other third party library. Numerical experiments show that for a system of 1600 interacting identical bosons in a harmonic trap, using a single GPU and a single CPU, it only takes two hours to achieve satisfactory simulation accuracy. With the increase of the number of identical particles, the advantage of GPU acceleration over CPU becomes more obvious, making it possible to simulate tens of thousands of identical particles from first principles using a single GPU. For example, for a system of 10000 non-interacting bosons, numerical experiments show that it takes 23 hours to obtain a simulation that is highly consistent with the exact results. Our study shows that GPU acceleration can lay a solid foundation for the wide application of PIMD simulations for extremely large-scale identical particle quantum systems with more than 10,000 particles. Numerical experiments show that a 24GB GPU can simulate up to 40000 identical particles from first principles, and the GPU acceleration leads to a roughly linear relationship between the computation time and the number of identical particles. In addition, we have also successfully implemented simulations for fictitious identical particle thermodynamics using GPU to overcome the Fermion sign problem, which makes it promising to efficiently and accurately simulate tens of thousands of fermions based on GPU.