Abstract
Monte Carlo methods and machine learning for rare event sampling
Perovskites are highly promising materials for emerging technological applications. Among them, strontium titanate (STO) stands out as a transition metal oxide perovskite that is extensively studied both, experimentally and theoretically. Understanding its surface properties is crucial for potential applications as catalyst or substrate for compound materials.
Following recent experimental advances in preparing a well-defined (1×1) termination of STO(001), my work aims to provide a comprehensive theoretical analysis of its electronic surface properties through means of density functional theory (DFT) calculations.
Main topics will include the investigation of the weakly-polar character of STO(001) and a compensation mechanism thereof, via experimentally observed Sr surface defects. Further, small polaron formation resulting from localization of excess charge on this surface is analyzed.
In all domains of natural sciences, we encounter rare events such as phase transitions, chemical reactions, or protein folding. Molecular simulations of these processes can lead to invaluable insights into their kinetics, thermodynamics, and mechanisms. However, the timescale disparity between the fastest molecular motions and the rate at which these processes occur renders it nearly impossible to observe rare events in unbiased equilibrium simulations. For these cases, transition path sampling is a method that focuses the computational resources on the ensemble of reactive trajectories that connect arbitrary states. The ensemble is sampled by repeatedly generating a new path from a point on a previous or initial path in the so-called shooting move. While this significantly reduces the computational resources necessary to obtain a reactive path, the shooting move introduces correlations between subsequently generated paths. Reducing these correlations is vital for any efficient path sampling scheme, particularly in complex systems with different reaction channels and many short-lived intermediate states. This thesis focuses on developing shooting schemes with the primary goal of reducing correlations between sampled paths. That includes the proposal of new shooting schemes using generative models, advancing replica exchange methods for path sampling, and revisiting shooting methods that evolve the shooting point from trial to trial.
Defense committee:
Jutta Rogal, New York University, USA (reviewer)
Titus Sebastian van Erp, Norwegian University of Science and Technology, NO (reviewer)
Christoph Dellago (supervisor)
Thomas Pichler (chair)