Machine-Learning Spectral Indicators of Topology.

Area of Science:

• Condensed matter physics
• Materials science
• Quantum chemistry

Background:

• Topological materials discovery is crucial but experimentally challenging.
• X-ray absorption spectroscopy (XAS) probes local atomic structure and bonding, key to topology.
• The theory of topological quantum chemistry (TQC) links material properties to topology.

Purpose of the Study:

• To develop a machine learning model for predicting topological material classes using XAS data.
• To overcome experimental challenges in determining material topology.
• To accelerate the discovery of novel topological materials.

Main Methods:

• Computed X-ray absorption near-edge structure (XANES) spectra for over 10,000 inorganic materials.
• Trained a neural network (NN) classifier to predict topological class from XANES signatures.
• Leveraged the quantitative agreement between experimental and computational XAS.

Main Results:

• Achieved high F1 scores: 89% for topological and 93% for trivial classes.
• Demonstrated NN's ability to predict topological class directly from XANES data.
• Validated the potential of XAS as a topological indicator.

Conclusions:

• Machine-learning-augmented XAS is a powerful tool for topological materials discovery.
• The method can identify topological properties in challenging materials like amorphous compounds.
• Enables in situ studies of field-driven topological phase transitions.