Machine Learning Potential Energy Surfaces: Validation, Application and Uncertainty Quantification

M. Meuwly

In this talk I will discuss the construction, validation, use and uncertainty quantification of machine learning-based potential energy surfaces (PESs). Examples to-be-discussed include reactions relevant to atmospheric chemistry and combustion. A primary focus is on using dynamics studies to compute observables that can be compared with measurements for validating the PESs. Secondly, the use of uncertainty quantification to further improve such PESs in a targeted fashion will be discussed.