Atomistic global optimization X: A Python package for optimization of atomistic structures.

Area of Science:

• Computational Materials Science
• Materials Informatics
• Atomistic Simulations

Background:

• Accurate materials modeling from first principles necessitates detailed knowledge of atomistic structures.
• Determining atomic positions and identities for complex systems (e.g., nanoparticles, surfaces) is a high-dimensional global optimization challenge.
• Machine learning (ML) advancements offer potential for accelerating materials structure searches.

Purpose of the Study:

• To introduce the Atomistic Global Optimization X (AGOX) framework and code.
• To provide a customizable and efficient approach for assembling and testing global optimization algorithms in materials science.
• To facilitate the integration of ML techniques into structure search methodologies.

Main Methods:

• Development of the AGOX Python package, featuring a modular design for expressing global optimization algorithms.
• Implementation of various optimization strategies within AGOX, including random search, basin-hopping, and ML-aided approaches.
• Application of AGOX to diverse materials science problems, such as cluster structures and surface reconstructions.

Main Results:

• AGOX enables efficient construction and experimentation with diverse global optimization algorithms.
• Demonstrated successful application of AGOX to complex atomistic structure prediction tasks.
• Validated the framework's flexibility in incorporating ML-driven surrogate energy models for accelerated searches.

Conclusions:

• The AGOX framework offers a powerful and adaptable tool for advancing global optimization in computational materials science.
• Its modularity and modern programming practices streamline the development and testing of novel structure search algorithms.
• AGOX is poised to accelerate materials discovery by enhancing the efficiency of atomistic structure determination.