# Applied Math Seminar: High fidelity representation of Born-Oppenheimer potential energy surfaces by machine learning

### Event Description:

Within the Born-Oppenheimer approximation, which separates the electronic motion from nuclear motion, the spectroscopy and reaction dynamics of molecular systems can be characterized by nuclear movement on potential energy surfaces. These multidimensional surfaces are continuous and single valued, but may have quite complex topological features, such as wells, saddles, and dissociation asymptotes. While the values of the potential energy at given nuclear configurations can now be obtained with high accuracy by solving the electronic Schrödinger equation, high fidelity representation of the global potential energy surface from these discrete data points remains a challenge. In this talk, we will discuss the latest developments in representing the potential energy surface using machine learning methods such as neural networks and Gaussian processes. In particular, how the permutation symmetry, which is necessary property of the potential energy surface, is imposed in these implementations will be emphasized.

Coffee and cookies will be served in the lounge at 15.00