Introduction to machine learning potentials for atomistic simulations.

Area of Science:

• Computational Chemistry
• Materials Science
• Physics

Background:

• Machine learning potentials (MLPs) have significantly advanced atomistic simulations.
• MLPs are increasingly integral to computational scientists' methodologies.
• The field requires accessible guidance for developing and applying these powerful tools.

Purpose of the Study:

• To provide a comprehensive overview and introduction to machine learning potentials.
• To offer a systematic guide for developing MLPs, including key components and approaches.
• To showcase practical applications and recent advancements in MLP development.

Main Methods:

• Review of chemical descriptors and regression models for MLP development.
• Discussion of data generation and validation strategies.
• Exploration of historical models (e.g., high-dimensional neural network potentials, Gaussian approximation potentials) and recent advancements.

Main Results:

• A structured guide for creating and implementing MLPs.
• Insights into the evolution from earlier models to current state-of-the-art techniques.
• References to expert reviews, open-source software, and practical examples.

Conclusions:

• Machine learning potentials offer powerful capabilities for atomistic simulations.
• This work lowers the barrier for researchers to adopt and utilize MLPs.
• MLPs are poised to push the boundaries of scientific discovery in simulations.