Python Library for Monte Carlo Simulations with Ab Initio and Machine-Learned Interatomic Potentials

Area of Science:

• Computational chemistry and materials science.
• Development of simulation software and algorithms.

Background:

• The simulation community requires transparent, reproducible, usable, and extensible (TRUE) Monte Carlo (MC) frameworks.
• Integrating ab initio methods and machine-learning interatomic potentials (MLIPs) into MC simulations is crucial for advancing computational studies.

Purpose of the Study:

• To introduce ASE-MC, a Python library enhancing the Atomic Simulation Environment (ASE) with MC simulation capabilities.
• To provide a flexible and extensible framework for diverse MC simulations using various energy engines.

Main Methods:

• Developed ASE-MC, a Python library integrating MC algorithms with ASE.
• Demonstrated flexibility through simulations of liquid water, biphenyl dihedral angles, and ammonia adsorption on Pt(111).
• Incorporated ab initio and MLIP engines, grand canonical MC with cavity bias, and custom MC move additions.

Main Results:

• Showcased the ability to combine ASE's system-building and calculation tools with MC algorithms.
• Successfully performed simulations in canonical, isothermal-isobaric, and grand canonical ensembles.
• Highlighted the flexibility in engine choice, MC ensemble, and custom move integration.

Conclusions:

• ASE-MC offers a concise Python scripting approach for complex MC workflows.
• The library facilitates reproducible MC simulations, enabling easier application to new research systems.
• This framework supports the transparent and extensible sampling of configurational space in materials simulations.