Machine-learned interatomic potentials for modelling nanoscale fracturing in silica and basalt

At the nanoscale, fracturing creates surface area and flow pathways, which control the rates of fluid-rock interactions. However, how fractures form at the nanoscale remains enigmatic. Here, we implement molecular dynamics simulations to reproduce fracture propagation in quartz and basalt. These simulations require large systems and long simulation times and are therefore currently depending on interatomic potentials. In the recent years, machine learning approaches have been established as a way to fit interatomic potentials, where the potentials are trained with quantum-mechanical data obtained from ab initio molecular dynamics simulations. We have developed machine-learned interatomic potentials for silica and basalt that allow using molecular dynamics simulations to simulate fracture propagation at the nanoscale. The interatomic potentials reproduce the mechanical properties of bulk silica and basalt sand have also been trained to account for fracture propagation. First, we trained a potential on silica to verify the fitting procedure, and then we used the same procedure to train an interatomic potential for basalt. By training the potential with water and carbon dioxide as fluids, we aim to study how a dynamic fracture damage basaltic glass and how the water and carbon dioxide enter these fractures in the wake of rupture. Our results are relevant for carbon mineralization where a coupling between dissolution of the basalt and precipitation of carbonate minerals can lead to nanofracturing of the rock.