Institute of Mechanical SystemsOpen OpportunitiesStudying the long-term diffusion of solutes in metals is crucial for a variety of present and futuristic engineering applications. This includes the design of safe and compact solid-state hydrogen reservoirs for automobile applications, designing corrosion-resistant materials for nuclear applications, and much more. The time scales involved in such mass diffusion processes for potential applications range from seconds to minutes. However, most state-of-the-art atomistic techniques can simulate an ensemble of atoms as large as some micrometers and for a real-time of some microseconds at best. Hence, the computational modeling of atomistic mass diffusion presents many challenges, which is why the design of these devices has relied on experiments. This project deals with an emerging class of atomistic simulation techniques based on statistical mechanics, which aims to track the relevant statistics of the ensemble rather than tracking all atomic positions and momenta. In such a statistical framework with multiple atomic species, every atomic site ceases to be a pure species and is instead identified by probabilities of finding different types of species at that site.
In order to introduce mass transport in such a setting, one needs to update the concentrations of different species at the atomic sites based on a phenomenological model, or by an atomistically informed master equation for the site probabilites. We are more interested in the latter approach, which involves computing the energy barriers and minimum energy pathways needed for atoms of different types to hop from one site to another. As this computation needs to be done for every possible atomic hop in the ensemble, the concentration update becomes computationally expensive. In this project, we plan to bypass this by employing graph neural networks (GNNs) to learn the hopping energy barriers as a function of local atomic environments and using a pre-trained GNN to update the site probabilities, which would enable us to reach higher time scales relevant for potential applications. - Engineering and Technology
- Master Thesis
|
|