Machine-Learning-Assisted Materials Discovery from Electronic Band Structure.

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Traditional materials discovery relies on time-consuming trial-and-error methods.
• Machine learning (ML) offers powerful pattern recognition for revolutionizing materials discovery.
• Electronic band structure data contains crucial information about material properties.

Purpose of the Study:

• To explore the application of ML techniques in materials discovery using band structure data.
• To identify patterns and relationships within a large dataset of material band structures.
• To demonstrate the utility of ML clustering for aiding novel material identification.

Main Methods:

• Retrieved band structure data for 63,588 materials from the Materials Project database.
• Grouped data into 85 batches based on band path in the first Brillouin zone.
• Applied feature selection, engineering, and noise reduction before training three ML clustering algorithms.

Main Results:

• Successfully trained ML clustering algorithms on processed band structure data.
• Validated ML models by comparing the properties of materials within identified clusters.
• Demonstrated the potential of ML to categorize and analyze vast materials datasets.

Conclusions:

• ML clustering is a viable approach to analyze complex band structure data for materials discovery.
• This method can significantly enhance the efficiency of identifying promising new materials.
• The findings pave the way for data-driven approaches in materials science research.