Machine learning approaches are now beginning to be actively explored in theoretical chemistry, and in particular, perhaps as an important starting point, efficiently describing the energies of molecules have recently been proposed and shown reasonable performances. Yet, the need for the machine to eventually learn many-body effects as well as subtle correlations poses a significant challenge for these approaches to achieve high accuracy for general purposes. In this talk, we focus on the configuration energies of solids and some chemisorption models on them, and illustrate that the machine learning approaches can indeed be a useful alternative to the existing models to evaluate the latter energies. For both cases, we obtain mean absolute errors of 0.05-0.13 eV based on the artificial neural networks.
Back to Workshop I: Machine Learning Meets Many-Particle Problems