Accelerating the Discovery of Transition Metal Borides by Machine Learning on Small Data Sets

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning in Materials Discovery

Background:

• Discovering new materials like ternary transition metal borides (MABs) is crucial but hindered by laborious traditional methods.
• Predicting material stability and structure-stability relationships requires significant experimental and computational effort.
• Accelerating the identification of promising MAB candidates is essential for advancing materials science.

Purpose of the Study:

• To develop and apply a small-data set machine learning (ML) method for efficient prediction of MAB stability.
• To accelerate the discovery of novel ternary transition metal boride (MAB) compounds.
• To establish a quantitative relationship between decomposition energy and thermodynamic stability using descriptors.

Main Methods:

• Utilized ab initio calculations to generate a dataset for training machine learning models.
• Developed three robust neural networks to predict decomposition energy (ΔHd) and assess thermodynamic stability.
• Employed composition-and-structure descriptors to unravel the relationship between ΔHd and stability.

Main Results:

• Identified three stable hexagonal M2AB2 compounds (Nb2PB2, Nb2AsB2, Zr2SB2) with negative decomposition energies.
• Discovered 75 metastable MABs with decomposition energies below 70 meV/atom.
• Validated ML model predictions through ab initio calculations of dynamical stability and mechanical properties.

Conclusions:

• The developed ML approach effectively accelerates the discovery of new compounds using limited data.
• Successfully expanded the known MAB phase family to include transition metals from VA and VIA groups.
• Demonstrated the reliability and efficiency of ML models in predicting material stability and properties.