Gianluca Seaford

My Background:

I am a first-year PhD student in the EPSRC CDT in Modelling of Heterogeneous Systems (HetSys II) at the University of Warwick, and a member of both the Hine groupLink opens in a new window (under the Warwick Theory Group), and the Robertson groupLink opens in a new window (under the Warwick Microscopy Group). My research combines electronic-structure theory, stochastic/Langevin molecular dynamics, and machine-learned interatomic potentials (MLIPs) to investigate electrochemical phenomena across multiple length- and time-scales within the grand-canonical ensemble.

In particular, I study the early stages of metal nucleation at solid–liquid interfaces, focusing on surface diffusion of adsorbates and clusters and adsorption–desorption kinetics, using grand-canonical DFT (GC-eDFT) with the linear-scaling DFT package ONETEPLink opens in a new window. I use this data to produce MLIP surrogate models to extend simulations beyond the cost limitations of ab-initio methods. I am particularly interested in applying equivariant message-passing neural networks (MPNNs), such as the MACELink opens in a new window many-body neural network, to reproduce reliable dynamics in systems too large or complex for conventional DFT.

Before joining HetSys I completed my Master's Degree in Mathematics and Physics at Warwick. My dissertation focused on performing constrained minimum-energy conical intersection (MECI) searches on machine-learned potential energy surfaces (ML-PESs). I used ORCA to perform linear-response time-dependent DFT (LR-TD-DFT) calculations to calculate the ground and first-excited states for mycosporine-glycene. This data was used to train an ensemble of MACE surrogate models for each state. The committee-averaged ML-PES was used for the constrained MECI search. This work was supervised by Prof. Nicholas Hine and resulted in a high first-class award.

Current Research:

My current focus is on training multi-head MLIP surrogate models of the grand free-energy landscape of solid–liquid interfaces under applied potential bias. These models are trained on grand-canonical DFT (GC-eDFT) data and constrained using in situ liquid-cell TEM observations (in collaboration with the Robertson Group). System dynamics, including both diffusive and adsorption–desorption dynamics, will be investigated with the aim of understanding early-stage metal nucleation on graphene.

This research is funded through the EPSRC CDT in Modelling Heterogeneous Systems (HetSys II) through the EPSRC grant EP/Y035429/1.