Field-aware atomistic models
Machine-learning interatomic potentials for dielectric and ferroelectric response, where polarization, Born charges, polarizability, and spectra are obtained by differentiating a shared scalar functional.
Research
Physics-informed learning for materials response
The recurring thread is to make models respect the variables that matter physically: applied fields, crystal structure, phonons, charge carriers, and the response functions that connect them.
Generative materials design
Property-conditioned crystal generation and structure recovery workflows, including CrystaLLM-pi tools for inverse design and diffraction-conditioned prediction.
Polarons and path integrals
Variational path-integral theory for charge-carrier mobility, electron-phonon coupling, optical conductivity, and high-throughput screening of polar semiconductors.