Collaboration on Machine-Learned Potentials with IPSuite: A Modular Framework for Learning-on-the-Fly.

Area of Science:

• Computational materials science
• Machine learning in physics
• Scientific software development

Background:

• Machine learning potentials (MLPs) have rapidly advanced due to algorithmic and hardware improvements.
• The deployment infrastructure for MLPs has not kept pace, leading to fragmented development.
• A unified platform is needed to bridge the gap between MLP research and practical application.

Purpose of the Study:

• Introduce IPSuite, a Python package to integrate diverse MLP methods.
• Establish a collaborative infrastructure to improve reproducibility in MLP development.
• Facilitate data management for sharing and deploying MLP models in simulations.

Main Methods:

• Developed IPSuite as a Python-driven software package.
• Integrated six state-of-the-art machine learning approaches for interatomic potential fitting.
• Incorporated methods for training data selection, ab initio calculations, and learning-on-the-fly strategies.

Main Results:

• IPSuite provides a unified platform connecting various MLP methods and algorithms.
• The package offers a collaborative infrastructure promoting reproducibility.
• Its data management facilitates model sharing and deployment for simulations.

Conclusions:

• IPSuite addresses the need for a cohesive infrastructure in the rapidly evolving field of MLPs.
• The software promotes collaboration, reproducibility, and efficient deployment of MLPs.
• IPSuite supports a comprehensive workflow from model development to simulation integration.