Deep generative neural networks for learning the dynamics of nanoparticles in liquid phase electron microscopy

Motion and interaction of nanoparticles near heterogeneous surfaces play a key role in nanoscale transport processes involved in nanomedicine, environmental remediation, electronics, and sensing. Liquid phase transmission electron microscopy (LPTEM) has emerged as a promising technique for single particle tracking at the nanoscale, enabling us to visualize and characterize the motion and interaction with unprecedented spatiotemporal resolution. Yet, understanding how nanoparticles move in the heterogenous chamber of the LPTEM and in interaction with the electron beam of the microscope has remained elusive. Here, we discuss our recent work on developing a physics-inspired deep generative neural network to learn the dynamic stochastic diffusion of nanoparticles in a heterogeneous environment of the LPTEM from a large dataset of experimental and simulated spatiotemporal trajectories. By traversing the latent space of trajectories, we identify a transition in the dynamic behavior of nanoparticles at different timescales reminiscent of the crossover in the dynamics of colloidal particles in glassy systems.