Multiscale Light-Matter Dynamics in Quantum Materials

PI Aiichiro Nakano, University of Southern California
Co-PI Rajiv Kalia, University of Southern California
Ken-ichi Nomura, University of Southern California
Priya Vashishta, University of Southern California
Award INCITE
Hours 1,000,000 Aurora Node-Hours
Year
Domain Materials Science
Photo-switching of a ferroelectric nanoscale structure in lead titanate

Photo-switching of a ferroelectric nanoscale structure in lead titanate. (Image: ALCF Visualization and Data Analytics Team.)

Project Summary

This project combines first-principles and AI-accelerated multiscale simulations with DOE experimental data to uncover how light drives electronic and atomic dynamics in quantum materials, enabling ultrafast ferroelectric switching and energy-efficient synthesis of high-temperature ceramics.

Project Description

To reveal the fundamental mechanisms of light-matter dynamics in quantum materials, this project is using a multiscale approach that integrates first-principles nonadiabatic quantum molecular dynamics (NAQMD) and AI-accelerated neural-network quantum molecular dynamics (NNQMD) simulations, in tandem with X-ray, electron-beam, and neutron experiments at DOE facilities. The team will use this approach to study: (1) light-induced switching of topological ferroelectric moiré superlattices for future ultrafast, ultra-low-power ferroelectric “topotronics”; and (2) photochemical pathways in polymeric ceramic precursors for energy-efficient additive manufacturing of high-temperature ceramics.

Simulating multiple field and particle equations for light, electrons, and atoms over vast spatiotemporal scales is computationally demanding. To address this, the team employs exascale-demonstrated multiscale light-matter dynamics (MLMD) simulations that leverage hardware heterogeneity and low-precision arithmetic. Divide-conquer-recombine (DCR) algorithms split problems into spatial and physical subproblems with small dynamic ranges and minimal mutual information, which are mapped to hardware units with optimal characteristics. Metamodel-space algebra (MSA) keeps key data structures on their respective hardware units, minimizing communication and precision demands.

Within the DCR/MSA paradigm, MLMD software combines first-principles DC-MESH (divide-and-conquer Maxwell-Ehrenfest surface hopping) and AI-accelerated excited-state (XS)-NNQMD modules. DC-MESH integrates Maxwell-Ehrenfest dynamics for short-time light-electron coupling on GPUs and surface-hopping NAQMD for longer-time electron-atom coupling on CPUs, with minimal GPU-CPU communication via electronic occupation numbers, while XS-NNQMD extends accessible spatiotemporal scales by orders of magnitude. In previous work using 60,000 GPUs on Aurora, the DC-MESH and XS-NNQMD modules achieved nearly perfect scalability, with 1.87 exaflops performance for DC-MESH.

Domains
Allocations

More Materials Science Projects