Machine Learning-Driven First-Principles Modeling of Far-Infrared Materials

Defended on 30 September 2025 at Université Paris-Saclay (C2N), under the supervision of Frédéric Aniel, with Anne-Sophie Grimault-Jacquin and Nicolas Zérounian.

The thesis develops BMach, a Bayesian optimization framework that calibrates Hubbard U within DFT+U against hybrid (HSE, HSEsol) and G0W0 references using Gaussian process regression. It applies the framework to InSb, to CuPt-ordered $\mathrm{InAs}_x\mathrm{Sb}_{1-x}$ alloys (supercell and VCA, with a G0W0 bowing parameter of 0.838 eV) and to the CdTe/InSb(001) band alignment, which comes out type-I and in agreement with experiment.