A machine learning route between band mapping and band structure

Area of Science:

• Solid-state physics
• Materials science
• Computational materials science

Background:

• Electronic band structure and crystal structure are key identifiers for solid-state materials.
• Current computational methods limit the extraction of quasiparticle dispersion from photoemission band mapping data.
• Databases of crystal structures are extensive, but band structure data extraction is challenging.

Purpose of the Study:

• To develop a computational pipeline for accurate band-structure reconstruction from photoemission data.
• To overcome limitations in current methods for analyzing large-scale photoemission datasets.
• To integrate machine learning with theoretical calculations for materials science.

Main Methods:

• Developed a pipeline combining probabilistic machine learning with data processing, optimization, and evaluation.
• Leveraged theoretical calculations to aid in band-structure reconstruction.
• Applied the pipeline to reconstruct all 14 valence bands of a semiconductor.

Main Results:

• The pipeline demonstrated excellent performance on benchmark and other materials datasets.
• Successfully reconstructed electronic band structures, revealing previously inaccessible momentum-space information.
• Showcased the ability to analyze both global and local structural information.

Conclusions:

• The developed pipeline offers a scalable approach for feature extraction in multidimensional data.
• Combines machine learning with domain knowledge for advanced materials analysis.
• Paves the way for integrating advanced band structure analysis with materials science databases.