Multiscale Light-Matter Dynamics in Quantum Materials: From Electrons to Topological Superlattices

Light-matter dynamics offers a way to study topological quantum materials which hold promise in developing ultralow-power, ultrafast devices. However, to study quantum mechanics (QM), one needs to solve the time-dependent Schrödinger equations in time-dependent density functional theory (TDDFT), which remains a scaling challenge when trying to simulate over vast spatiotemporal scales. We turn to Exaflop computers to address this by leveraging their increased heterogeneity and low-precision compute, thereby achieving multiscale/Multiphysics simulations. We employ a Divide-Conquer-Recombine algorithm to divide the problem into spatial and physical subproblems of small dynamic ranges, which are mapped onto best-characteristics-matching hardware units, and a metamodel-space algebra to minimize communication and precision requirements. Using divide-and-conquer Maxwell-Ehrenfest-surface hopping (DC-MESH) and excited-state neural-network quantum molecular dynamics (XS-NNQMD) we scale up to 60,000 GPUs of Aurora for a 15.4 million-electron and 1.23 trillion-atom PbTiO₃ material, achieving 1.87 EFLOP/s.