Training Equivariant Neural Network Interatomic Potentials to Model Phase Transformations in Complex Crystals
||Training Equivariant Neural Network Interatomic Potentials to Model Phase Transformations in Complex Crystals|
||Johan Klarbring <email@example.com>|
||2023-11-01 – 2024-05-01|
This project aims to train a set of recently developed equivariant graph Neural Network interatomic potentials ,  to model phase transformations in complex materials. We will primarily be interested in so-called plastic crystals, which are crystals that contain molecular units that are randomly oriented, but whose centers-of-mass retain the long-range ordering characteristic of crystalline solids. These systems often have phase transformations from an ordered low-temperature state to a high-temperature phase with disordered molecular units. Since these transformations are highly sensitive to external pressure, they can be used to produce efficient solid-state cooling devices through the barocaloric effect .
Ideally, atomistic modelling of such systems would employ DFT-based ab-initio molecular dynamics (AIMD) simulations based on density functional theory (DFT). The time- and length scales that need to be simulated are, however, prohibitively expensive with this methodology. The development of accurate machine learning interatomic potentials (MLIPs) over the last ~decade show promise in resolving this issue.
In this project, we will use DFT-based data-sets for a set of systems plastic crystals and to train and compare two classes of state-of-the-art equivariant neural network interatomic potentials, Allegro  and MACE . We will sweep a space of hyper-parameter’s looking for suitable accuracy-efficiency tradeoffs. After training on Berzelius, production MD runs will be performed on other computational resources.
Allegro and MACE have both been installed, tested and used on Berzelius in our previous proposal.
We apply for the default allocation.
 https://github.com/mir-group/allegro; https://www.nature.com/articles/s41467-023-36329-y
 https://github.com/ACEsuit/mace ; https://arxiv.org/abs/2206.07697