Topic 01-10 : Hierarchical latent-space models for battery interface dynamics

The focus of the PhD research project is to develop novel computational approaches to describe and predict the spatio-temporal evolution of battery interfaces. While descirptors are a well-established and succesful approach in other fields of science, catalysis in particular, the development of descriptors that are capable of capturing the complex dynamics and processes at battery interfaces remain to be established.

In this project, we will seek to establish the physical insights needed to identify such descriptors using data from both simulations and experiments. The goal is to develop machine learning algorithms, which can learn to map the multi-scale battery interface dynamics into multi-resolution hierarchically coupled regularized latent spaces that each encodes for structures at different length scales. This model will be integrated with a suitable coupled Markov chain model in the extended latent space to yield a complete dynamics simulator, which has validated against trajectories obtained from simulations of battery interface system dynamics at different length and time scales. Uncertainty quantification and propagation methods will be integrated within the probabilistic latent space generative models to provide predictive uncertainties that take into account both data and model uncertainties. The project will be closely linked to the activities in Battery 2030+, specifally the BIG-MAP project, which is coordinated by the main PhD advisor and where both co-advisors at the partner institution are coPIs.

The student will work closely with the BIG-MAP postdocs and researchers working on this topic, and the student will spend her/his secondments at the University of Cambridge to increase the interaction with the leading computational (Prof. Gabor Csányi) and experimental groups (Prof. Clare Grey) in the field.