(370h) BASC: Applying Bayesian Machine Learning Techniques to the Search for Global Minima on Potential Energy Surfaces

Controlling molecule-surface interactions is key for chemical applications ranging from catalysis to gas sensing. We present a framework -- the Bayesian Active Site Calculator (BASC) -- for rapidly and accurately searching for the global minimum on potential energy surfaces, corresponding to stable adsorbate-surface structures. We formulate the problem as finding the minimum of a low-dimensional objective function. We present a technique using Bayesian inference that enables us to predict converged density functional theory (DFT) potential energies with fewer self-consistent field (SCF) iterations. We then discuss how we apply Bayesian Optimization to search for the global minimum of the objective function. We benchmark our method against Constrained Minima Hopping, a popular method for solving the problem at hand, and present the adsorption sites found by our global optimization method for various simple hydrocarbons on the rutile TiO₂ (110) surface.