Parallel Implementation of Nonadditive Gaussian Process Potentials for Monte Carlo Simulations

Abstract

A strategy is presented to implement Gaussian process potentials in molecular simulations through parallel programming. Attention is focused on the three-body nonadditive energy, though all algorithms extend straightforwardly to the additive energy. The method to distribute pairs and triplets between processes is general to all potentials. Results are presented for a simulation box of argon, including full box and atom displacement calculations, which are relevant to Monte Carlo simulation. Data on speed-up are presented for up to 120 processes across four nodes. A 4-fold speed-up is observed over five processes, extending to 20-fold over 40 processes and 30-fold over 120 processes.

Accelerated Bayesian Inference for Molecular Simulations using Local Gaussian Process Surrogate Models

“…Local Gaussian processes (LGPs) are an emerging class of accelerated GP methods that are well-equipped to handle large sets of experimental data. These so-called “greedy” Gaussian process approximations are constructed by separating a GP into a subset of GPs trained at distinct locations in the input space. ,− Computation on the LGP subset scales linearly with the number of IVs, is trivially parallelizable, and easily implemented in high-performance computing (HPC) architectures. , State-of-the-art LGP models have been used to design Gaussian approximation potentials (GAPs), a type of machine learning potential used to study atomic − and electron structures, , as well as nuclear magnetic resonance chemical shifts with uncertainty quantification . However, to our knowledge, LGPs have not been applied as surrogate models for UQ/P on complex experimental data in computational chemistry.…”

“…46,57−59 Computation on the LGP subset scales linearly with the number of IVs, is trivially parallelizable, and easily implemented in high-performance computing (HPC) architectures. 60,61 State-of-the-art LGP models have been used to design Gaussian approximation potentials (GAPs), 62 a type of machine learning potential used to study atomic 63−65 and electron structures, 62,66 as well as nuclear magnetic resonance chemical shifts 67 with uncertainty quantification. 34 However, to our knowledge, LGPs have not been applied as surrogate models for UQ/P on complex experimental data in computational chemistry.…”