Determination of potential energy surface global minimum for solid-solid interface with Bayesian optimization approach

The following thesis analyzes the use of statistical tools and active learning in determining the potential energy surface of solid-solid interfaces. Within the broader field of research related to tribology, this thesis focuses on atomistic calculation methods. The work presented here stems from the need to go beyond the high-symmetry points method to calculate the adhesion energy for solid-solid interfaces, which is inefficient forstudying poorly symmetric systems. In this regard, this thesis analyzes the application of Bayesian optimization and Gaussian processes employed in the BOSS code, to determine potential energy surfaces and compute their global minimum values, which allow to evaluate adhesion energy. Starting from the validation of the Bayesian approach on simple and symmetric interfaces, such as the homogeneous Cu(111)(1x1)-Cu(111)(1x1), the study continues by applying it to increasingly complex cases: the heterogeneous interfaces Au(111)(1x1)-C(111)(1x1), Au(111)(1x1)-amorphous carbon and Au(111)(1x1)C(111)(2x1). For each interface analyzed, part of the study is devoted to the search for improved methods which exploit symmetry features enjoyed by slab surface, in order to reduce the computational cost of simulations.