Parallel Implementation of Nonadditive Gaussian Process Potentials for Monte Carlo Simulations.

Area of Science:

• Computational Chemistry
• Materials Science
• Statistical Mechanics

Background:

• Gaussian process potentials offer a data-efficient approach for molecular simulations.
• Implementing these potentials in large-scale simulations requires efficient parallelization strategies.
• Accurate modeling of interatomic interactions, including three-body effects, is crucial for reliable simulation outcomes.

Purpose of the Study:

• To develop and implement a parallel programming strategy for Gaussian process potentials in molecular simulations.
• To focus on optimizing the calculation of three-body nonadditive energy within this framework.
• To demonstrate the scalability and efficiency of the proposed method.

Main Methods:

• A parallel programming approach was designed to distribute molecular simulation tasks across multiple processing units.
• The strategy specifically addresses the distribution of pair and triplet interactions for general potentials.
• The implementation was tested using argon in a simulation box, performing full box and atom displacement calculations.

Main Results:

• The parallel strategy effectively distributes pair and triplet interactions, applicable to both additive and nonadditive energies.
• Significant speed-up was observed: a 4-fold increase with five processes, reaching 20-fold with 40 processes, and 30-fold with 120 processes.
• The method proved effective for Monte Carlo simulations, as demonstrated by full box and atom displacement calculations.

Conclusions:

• The presented parallel programming strategy enables efficient implementation of Gaussian process potentials in molecular simulations.
• The method demonstrates excellent scalability and provides substantial computational speed-up, making complex simulations more feasible.
• This approach is broadly applicable to various potentials and simulation types, advancing the field of computational molecular modeling.