Speaker
Description
The energy landscape paradigm provides a powerful framework for understanding the structure, dynamics, and properties of disordered systems such as glasses, yet navigating these complex, high-dimensional landscapes remains computationally challenging. This talk presents three recent advances in leveraging machine learning and differentiable simulations to explore energy landscapes of disordered materials. First, we will discuss a graph reinforcement learning framework, namely, StriderNet, that learns policies to displace atoms toward low-energy configurations on rough, non-convex landscapes. Evaluated on binary Lennard-Jones particles, calcium silicate hydrates, and disordered silicon, StriderNet outperforms classical optimization algorithms with remarkable transferability to system sizes an order of magnitude beyond training conditions. Second, we demonstrate approaches for visualizing high-dimensional energy landscapes through dimensionality reduction and interpretable machine learning, revealing hidden patterns in configurational space and providing intuitive understanding of energy barriers, basins, and transition pathways. Finally, an end-to-end differentiable molecular dynamics framework exploits automatic differentiation to compute gradients of macroscopic properties, including elastic constants, vibrational density of states, and transport coefficients, with respect to force field parameters, enabling inverse design of force fields from target material properties. Together, these advances demonstrate how AI algorithms are transforming our ability to understand and engineer disordered materials through their energy landscapes.