Extracting an Empirical Intermetallic Hydride Design Principle from Limited Data via Interpretable Machine Learning

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Designing metal hydrides with specific thermodynamic properties is challenging due to their complex structures and chemistries.
• Existing machine learning models predict properties but lack transparency, hindering rational material design.
• A need exists for simple, physics-based design rules for metal hydrides.

Purpose of the Study:

• To identify simple, physics-based design rules for metal hydride thermodynamic properties.
• To establish a structure-property relationship for intermetallic hydrides.
• To enable the rational design of novel metal hydrides for hydrogen storage.

Main Methods:

• Utilized a gradient boosting tree regressor to analyze feature importance.
• Identified a key volume-based descriptor influencing metal hydride equilibrium H2 pressure.
• Validated the descriptor's applicability across various compositions, substitutions, and structures.

Main Results:

• Discovered a strong dependence of metal hydride equilibrium H2 pressure on a volume-based descriptor.
• This descriptor can be calculated solely from the elemental composition of intermetallic alloys.
• The identified structure-property relationship holds across diverse intermetallic hydride systems.

Conclusions:

• A simple, physics-based descriptor enables prediction of metal hydride properties.
• This facilitates targeted design of novel intermetallic hydrides for hydrogen storage.
• A previously uninvestigated intermetallic is predicted to form a low-stability hydride.