Efficient grand canonical global optimization with on-the-fly-trained machine-learning interatomic potentials

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Characterizing nanostructured materials in reactive environments is complex due to intricate structures and chemical changes.
• Predicting stable material structures requires global optimization, but existing methods are computationally intensive, especially with first-principles calculations.
• Efficiently exploring configurational and compositional spaces for nanostructured materials remains a significant challenge.

Purpose of the Study:

• To develop and evaluate a computationally efficient grand canonical global optimization algorithm.
• To identify stable structures and chemical states of nanostructured materials under specific reaction conditions (pressure, temperature).
• To reduce the computational cost associated with first-principles energy evaluations in materials discovery.

Main Methods:

• Implementation of a grand canonical global optimization algorithm.
• Leveraging on-the-fly trained machine-learning interatomic potentials (MLIPs) using sparse Gaussian process regression.
• Incorporation of the smooth overlap of atomic positions (SOAP) descriptor.
• Application of the ab initio thermodynamics framework to approximate Gibbs energy for environment-aware optimizations across multiple stoichiometries.

Main Results:

• Demonstrated computational efficiency of the developed algorithm.
• Successfully identified stable structures and chemical states for targeted nanostructured systems.
• Validated the algorithm's ability to reproduce known literature examples.
• Reduced the number of required first-principles energy calculations significantly.

Conclusions:

• The developed algorithm offers a computationally efficient approach for materials characterization under reactive conditions.
• This method accelerates the discovery of nanostructured materials with predictable stable structures and chemical states.
• The integration of MLIPs and ab initio thermodynamics provides a powerful framework for environment-aware materials optimization.