AENET-LAMMPS and AENET-TINKER: Interfaces for accurate and efficient molecular dynamics simulations with machine...

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Machine-learning potentials (MLPs) offer a computationally efficient alternative to traditional quantum mechanics methods.
• Integrating MLPs with molecular dynamics and Monte Carlo simulation software is crucial for practical applications.

Purpose of the Study:

• To develop and evaluate interfaces connecting the atomic energy network (ænet) MLP package with TINKER and LAMMPS simulation software.
• To enable accurate and cost-effective simulations of large, complex systems using MLPs.

Main Methods:

• Development of two interfaces: ænet-TINKER and ænet-LAMMPS.
• Performance evaluation through scaling tests on shared-memory and distributed-memory systems.
• Demonstration of utility in simulating diffusion in liquid water and battery materials.

Main Results:

• The ænet-TINKER interface shows near-optimal parallel efficiency on shared-memory systems.
• The ænet-LAMMPS interface achieves excellent parallel efficiency on distributed-memory systems.
• Both interfaces facilitate accurate simulations with linear scaling computational cost.

Conclusions:

• The developed interfaces significantly enhance the accessibility and efficiency of MLP-based simulations.
• These tools accelerate research in areas like liquid dynamics and advanced materials development.
• Open-source availability of ænet, TINKER, and LAMMPS promotes wider adoption and innovation.