Machine learning accelerated random structure searching: Application to yttrium superhydrides.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Atomistic simulations and Crystal Structure Prediction (CSP) algorithms, coupled with Density Functional Theory (DFT), have advanced the search for superhydrides.
• Increasingly complex systems pose computational challenges due to an exponential growth in search space.
• Efficient sampling strategies are crucial for sustainable computational cost in materials discovery.

Purpose of the Study:

• To develop and validate an efficient active-learning strategy for discovering complex crystal structures.
• To overcome the computational limitations of traditional CSP methods for complex materials.
• To explore novel superhydride phases for hydrogen storage and high-temperature superconductivity.

Main Methods:

• An active-learning process integrating machine learning potentials with DFT simulations.
• Application to tin crystal structures under varying pressures as a proof of concept.
• Development of atomic environment analysis, clustering algorithms, and an x-ray spectra-based metric for structure selection.
• Exploration of yttrium superhydrides (YHₓ) under pressure.

Main Results:

• Successfully retrieved known tin crystal phases, including the unique α phase, and predicted phases under pressure (20 and 100 GPa).
• Identified the known YH₆ structure (Im-3m) and explored more complex YHₓ systems.
• Demonstrated the effectiveness of new selection methods in guiding the search towards experimentally relevant structures.
• Discovered new, complex YHₓ phases previously inaccessible to CSP algorithms.

Conclusions:

• The proposed active-learning strategy significantly enhances the efficiency and scope of CSP for complex materials.
• The approach effectively handles systems of increasing complexity, pushing back the 'exponential wall' in computational materials discovery.
• This method holds great promise for accelerating the discovery of advanced materials for hydrogen storage and superconductivity.