Molecular scale machine-learning based simulations of dynamic fracture in rocks

In this project, we will use machine learning to develop molecular scale rock models that capture mechanical properties of rocks and rock–water interactions simultaneously, in order to understand dynamic fracture of rocks both in the presence and absence of water.


Dynamic fractures in rocks are associated with various important natural phenomena such as earthquakes, and for engineering applications such as hydrofracking and fluid injection operations for thermal energy exchange.

Brittle materials like rocks can often be described well by linear elastic fracture mechanics; the fracture properties come down to the elasticity tensor of the material and a fracture toughness, both of which can be measured experimentally.

But rocks are often wet and heterogeneous, resulting in important deviations from the rules of linear elastic fracture mechanics. Fracture propagation in brittle rocks is also highly dynamic, thus several interacting mechanisms may be at play simultaneously: A linear elastic-like fracture of the rock itself, weakening of the crack tip by water reacting with juvenile fracture surface, water phase transitions due to a sudden pressure drop during crack opening etc.

In this project, we will use machine learning to create molecular-scale models representing water–rock interactions and the structural and elastic properties of rocks simultaneously and accurately. These models will be used to explore emergent properties of dynamic fractures in rocks both in the presence and absence of water.

Working with this project you will leverage machine learning as a tool for improving state-of-the-art simulation methods. This trains a computational science skillset that is applicable both in academia and in industry. The project may include collaborations with the Center for Advanced Computing and Simulations at University of Southern California.


  • MSc in physics or fracture mechanics or a mathematically focused MSc in geosciences.
  • Candidates with documented experience in statistical or computational physics, scientific programming, molecular dynamics simulations, and experience from machine learning will be prioritized.


Professor Anders Malthe-Sørenssen

Dr. Henrik Andersen Sveinsson

Call 2: Project start autumn 2022

This project is in call 2, starting autumn 2022. 



Published Sep. 21, 2020 4:55 PM - Last modified Nov. 19, 2020 2:28 PM