Spectral operator representations

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Machine learning in materials science often focuses on atomic geometry, which is insufficient for learning spectral properties.
• Learning intrinsic material properties like band gaps requires methods beyond simple atomic environments.

Purpose of the Study:

• To develop a general machine learning framework based on electronic structure descriptors.
• To apply this framework for material similarity assessment and accelerated screening.

Main Methods:

• Developed a novel framework utilizing electronic structure descriptors.
• Leveraged natural symmetries and interpretability of physical models.
• Applied the framework to material similarity and screening tasks.

Main Results:

• A model trained on 217 materials achieved 75% accuracy in labeling promising transparent conducting materials.
• The framework demonstrates effectiveness in accelerated materials screening.
• The approach shows promise for learning complex material properties.

Conclusions:

• Electronic structure descriptors offer a more promising approach for learning complex material properties compared to atomic geometry.
• The developed framework facilitates efficient materials discovery and screening.
• This work advances the application of machine learning in predicting spectral and intrinsic material properties.