Gaussian process structural optimization with density functional theory plus ghost-Gutzwiller approximation: a case study on SrMoO3

The prediction of ground state atomic structures in strongly correlated materials is a significant challenge, as single-particle frameworks based on density functional theory (DFT) often fail to capture the interplay between electronic interactions and lattice degrees of freedom. We investigate the ground-state crystal structure of the correlated metal SrMoO3 using the density functional theory plus ghost-Gutzwiller approximation (DFT+gGA). We compute the multi-dimensional potential energy surface (PES) by training a Gaussian process regression (GPR) model on a relatively small set of DFT+gGA total energies. This approach accurately maps the energy landscape. The total energies and electronic structures for these phases show reasonable agreement with previous DFT plus dynamical mean-field theory (DFT+DMFT) calculations. Our study shows that the computational efficiency of DFT+gGA enables PES exploration for correlated materials previously impractical with more demanding many-body methods.