"Machine Learning Augmented DFT to Discover New Perovskite Photocatalysts" 

Maitreyo Biswas, MSE PhD Candidate 

Advisor: Professor Arun Mannodi-Kanakkithodi

WebEx Link

ABSTRACT

Perovskite materials have emerged as potential candidates for next-generation photovoltaics and photocatalysis due to their exceptional optoelectronic properties and structural versatility. The vast compositional space of perovskites offers immense potential for tailoring materials to specific applications such as hydrogen generation via water splitting. However, the traditional approach of exploring the chemical space through high throughput density functional theory (DFT) calculations is hindered by computational constraints, particularly when dealing with complex alloys and increasing supercell size. This report examines recent advancements in accelerating perovskite discovery for photocatalytic applications through Machine Learning (ML) augmented DFT approaches. ML techniques, including composition-based regression models and graph neural networks (GNNs), have been leveraged to rapidly screen and identify promising perovskite candidates with optimal properties such as band gaps, band alignments, and thermodynamic stability. For rational photocatalyst design I reviewed how ML-assisted screening has facilitated the discovery of perovskites with suitable band edge positions and enhanced catalytic activity for water splitting. We also discuss how these ML-based approaches provide insights into composition-property relationships, guiding experimental efforts and accelerating the development cycle of new materials. By showcasing the synergy between ML and DFT in the context of perovskite discovery for photocatalysis, this report elucidates the transformative impact of data-driven approaches in advancing renewable energy technologies.