Title: Solving the electronic Schrödinger Equation with Deep Learning
Abstract: The electronic Schrödinger Equation represents one of the most fundamental models in Physics. It allows to model all properties of Matter (Molecules or Solids) from first principles. An efficient numerical solver for the electronic Schrödinger equation would thus enable us to replace costly and time consuming experiments by numerical simulations and revolutionize the design of materials or drugs with specific desirable properties. The search for efficient numerical solvers therefore constitutes one of the most important problems in computational chemistry. To this end, Deep Learning based approaches have led to great recent successes, enabling the currently largest simulations of molecules to within chemical accuracy. Mathematically these methods are however not yet well understood. In this talk I will provide an introduction to Deep Learning based methods for solving the electronic Schrödinger equation and demonstrate their impressive empirical results. After that I will formulate a number of interesting open mathematical problems whose solution could potentially lead to further more systematic improvements.
(Host: Markus Bachmayr)