https://faculty.utk.edu/Konstantinos.Vogiatzis
Title: "Connectivity in Chemistry: From Persistent Homology to Electronic-Structure-Informed Machine Learned Potentials"
Abstract: Connectivity is a fundamental concept in chemistry, linking molecular structure, electronic organization, and ultimately chemical function. In this talk, I will discuss how different representations of connectivity can be combined with artificial intelligence to address challenges ranging from molecular discovery to high-accuracy quantum molecular simulation. At the molecular level, our work uses graph-based representations and topological data analysis, particularly persistent homology, to capture structural and geometric relationships that govern chemical properties. Coupled with machine learning, these representations enable the exploration of large chemical spaces and the identification of structure–property relationships relevant to carbon capture and catalysis. Applications include the discovery of CO2-philic molecular units and the identification of molecular descriptors controlling the reactivity of transition-metal complexes for catalytic applications.
Connectivity also extends beyond molecular structure to the electronic degrees of freedom that determine chemical bonding and reactivity. Building on this perspective, we develop data-driven electronic structure methods that incorporate orbital information and electron correlation directly into machine-learning representations. One example is OrbGNN, an electronic-structure graph architecture in which quantum entanglement and electron correlation defines their connectivity.