Machine learning augmented neutron and x-ray scattering for quantum materials

Machine learning (ML) has demonstrated great power in materials science but
encounters a few challenges for spectroscopic data. In this presentation, I will offer three
examples from our works to introduce how ML can augment the spectroscopy analysis,
including elastic scattering (momentum-space), inelastic scattering (energy-space) and
absorption spectroscopy (intensity-space). On elastic scattering, we show that the manual
fitting to reflectometry spectra can be fully replaced by neural networks, which leads to an
automated fitting-free analysis, a super-resolution, and fine resolution on the magnetic
proximity effect in topological insulator heterostructures. For inelastic scattering, by building a
full 3D rotational-transnational equivariant neural network, we show that the elementary
excitations of phonon can be well captured by only using information of atomic species and
coordinates, which demonstrates great power when studying materials with large supercells
and alloy systems. Lastly, by labeling the x-ray absorption spectra with the topological class, we
show that the materials topological information - a global quantity in reciprocal space - can well
be coded implicitly in local real space information. We conclude by showing a variety of
problems machine learning may solve in neutron and x-ray spectroscopic researches.