(4dr) Switching the Lights on: The Vision of a ML Model for Enhanced Photocatalysis

My poster showcases an overview about the research that I have been conducting so far both as a PhD at the University of Bremen, Germany, and as a current Postdoc at the University of Stanford, in order to act as a springboard for a future research proposal. The journey starts with Density Functional Theory (DFT) studies on TiO₂, used as a photocatalyst for the removal and degradation of major water pollutants and concludes with extensive investigations through Density Functional Tight Binding (DFTB) for transition metal catalysts.

Both works want to highlight the importance of scaling in quantum chemistry, which is still nowadays a non-trivial computational problem. While several recent studies on Machine Learning (ML) interatomic potentials are an appealing alternative to more traditional methods, in principle being able to handle system sizes and timescales comparable to those addressed by Force Fields (FF) while maintaining accuracy close to that of DFT, poor focus has been reported on the extension to excited state quantum physics. In fact, ultrafast spectroscopy and more in general photo-catalysis still rely on very computationally expensive methods such as perturbation theory or Time-Dependent DFT (TD-DFT), which limit the system size to small molecules and the time scales to a couple of picoseconds.

A key challenge in photocatalysis, though, is understanding the interplay between the electronic and structural properties of the catalyst under operating conditions. By combining TD-DFT and TD-DFTB, the aim of my research idea is to develop a comprehensive framework that captures the dynamics of excited states and their impact on catalytic activity. This approach will allow for the simulation of real-time processes, providing deeper insights into the mechanisms of photocatalysis and guiding the design of more efficient catalysts.