Machine Learning for Materials Design

Atomistic Molecular-Spin Dynamics

Atomistic Molecular-Spin Dynamics

We bridge the divide between atomistic simulations at the microscopic scale and material simulations at the mesoscopic scale. By combining high-fidelity data generation with machine-learning models, we create interatomic potentials for high-performance molecular-spin dynamics simulations. This methodology couples lattice degrees of freedom with electronic spins. We utilize this method for tackling a wide range of applications, including the analysis of material strength, the investigation of transport properties in nanoscale systems, and the simulation of magneto-structural phase transitions relevant for developing ultrafast magnetic storage devices.

Dr Attila Cangi

Bartosz Brzoza
Karan Shah
Dr Lenz Fiedler
Dr Kushal Ramakrishna

Magneto-Structural Phase Transitions in Iron

We have developed a data-driven approach to create magneto-elastic machine learning models to simulate dynamics of ions and the electronic spin simultaneously. These models help us understand magneto-structural phase transitions in materials. We build these models by combining a collective atomic spin model with a machine learning interatomic potential. Both parts of the model are fine-tuned using data from first-principles calculations.

We tested our approach by applying it to alpha-iron, a material that undergoes magneto-structural phase transitions. Our results show how our method accurately predicts several material properties, such as bulk modulus, magnetization and specific heat, and the Curie temperature during the transition from a ferromagnetic to a paramagnetic state.

Simulating the heat transfer along a slab of iron using atomistic molecular-spin dynamics.