Colloquia

Electrons in materials can interact in ways that lead to remarkable behaviors, from magnetism to metal–insulator transitions. Simulating these correlated dynamics is a major challenge, since they unfold across many time and length scales. In this talk, I will describe how machine learning can help overcome these barriers. By training neural networks to act as “force fields” that capture the influence of electrons on slower degrees of freedom, we can perform large-scale dynamical simulations of functional electron materials with both speed and accuracy. These simulations reveal new types of nonequilibrium relaxation processes that lie beyond conventional theories. I will also discuss recent progress in extending this framework to nonequilibrium settings, where machine learning enables us to model nonconservative forces such as spin-transfer torques that underlie spintronic devices. Finally, I will touch on how these ideas generalize to the dynamics of emergent order-parameter fields, opening new directions for the study of quantum materials.