Machine Learning Approximations for Quantum Simulations of Hydrogen-Bond Dynamics

DASHH Doctoral Researcher: Yahya Saleh

Supervisors: Prof. Jochen Küpper (DESY, UHH), Dr. Andrey Yachmenev (DESY), Prof. Armin Iske (UHH)

Accurate computations of laser-induced hydrogen-bond dissociation require knowledge of thousands of excited ro-vibrational states, from the ground state all the way up to the dissociation limit. The primary goal of this project is to create machine learning-based approaches to reduce the otherwise prohibitively high costs associated with computing many excited states. Two approaches have been created to achieve this. One is using the active learning method to address the issue of the high cost of quantum chemistry calculations of potential energy surfaces. The other method lowers the cost of variational nuclear dynamics simulations by utilising normalising-flow neural networks for solutions of the Schrödinger equation.

PhD Thesis