Exploring the frontiers of condensed-phase chemistry with a general reactive machine learning potential

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Atomistic simulations are crucial for drug design and materials discovery.
• Machine learning interatomic potentials (MLIPs) offer a computationally efficient alternative to ab initio methods.
• A general reactive MLIP is needed for broad applications in chemistry without refitting.

Purpose of the Study:

• To develop a general reactive MLIP applicable to a wide range of chemical reactions.
• To enable high-throughput in silico reactive chemistry experimentation.
• To validate the MLIP's accuracy against experimental and traditional computational methods.

Main Methods:

• Automated sampling of condensed-phase reactions was employed.
• Development of a general reactive MLIP named ANI-1xnr.
• Application of ANI-1xnr to diverse chemical systems including nucleation, ring formation, biofuel additives, combustion, and molecule formation.

Main Results:

• ANI-1xnr demonstrated high accuracy across five distinct chemical systems.
• The MLIP closely matched experimental data and previous computational studies.
• The model proved effective for simulating reactive chemistry involving C, H, N, and O elements.

Conclusions:

• ANI-1xnr represents a significant advancement in general reactive MLIPs.
• The developed MLIP facilitates efficient in silico exploration of reactive chemistry.
• This tool has broad applicability in computational chemistry and materials science.