SiO2 Fracture: Chemomechanics with a Machine Learning Hybrid QM/MM Scheme

PI James Kermode, University of Warwick
Crack propagation in a two-dimensional bilayer of amorphous silicon dioxide modeled with a polarizable interatomic potential
Project Description

Understanding the chemo-mechanical phenomena that cause silicates to fracture would prove a great advantage to both enabling the process, as in large-scale mining, and eliminating it in products that rely on silicate materials. Researchers are utilizing advanced computing tools on Mira to better understand the behaviors that drive stress corrosion and chemically activated crack propagation at both the macro- and microscopic levels.

This multi-year INCITE project is pioneering simulation methodologies for predictive modelling of failure processes in oxides using a hybrid quantum mechanical/molecular mechanical (QM/MM) scheme to help describe the fracturing of silicon dioxide in a wet environment. On the method development side, the team is advancing a novel machine learning (ML) approach that significantly reduces the number of expensive QM calculations necessary per unit of simulated system time. This approach dramatically increases the efficiency of ongoing production calculations, reducing the computational cost anticipated for CY 2016 by approximately 40 percent for the same scientific milestones.

The simulations have shown, thus far, that cracks in silicon can initiate and propagate in the presence of oxygen, even if the energy supplied by the load is insufficient to create new fracture surfaces in pure systems. These results were confirmed by experiments that showed no evidence of cracking in oxygen-free conditions.

Continuing work on machine learning of QM forces is of key importance at this point in the project, as applications science and machine learning will converge to deliver much larger 3D modeling of silica/water systems. Simulation times will run much longer than is possible with current state‐of‐the‐art, first-principles molecular dynamics approaches.

The results from this project have relevance for a broad range of applications in mining, photovoltaics, and biomedical implants. They are also expected to generate fundamental insights that could help rationalize and guide future materials design and novel algorithmic developments.

Allocations