First-principles ultrafast phonon dynamics by solving the real-time Boltzmann transport equation with adaptive multirate time stepping

ORAL

Abstract

Ultrafast lattice dynamics in materials can be accessed with pump-probe spectroscopies and used to prepare novel nonequilibrium quantum phases. Therefore, quantitative modeling of ultrafast lattice dynamics would advance nonequilibrium physics. Electron dynamics can be modeled by the electron real-time Boltzmann transport equation (rt-BTE) with first-principles electron-phonon (e-ph) collisions using an efficient parallel algorithm [1]. Yet, solving the lattice (phonon) rt-BTE with e-ph and phonon-phonon (ph-ph) collisions [2] remains challenging due to the different timescales of e-ph and ph-ph interactions. In this work, we interface the PERTURBO code [1] with the SUNDIALS library [3] to efficiently advance coupled electron and phonon rt-BTEs in time. We demonstrate a significant speed-up using adaptive step size and multirate infinitesimal (MRI) methods from SUNDIALS that enable utilizing different time step sizes for fast-evolving, cheap e-ph interaction and the slow, expensive ph-ph interaction. These advances allow us to model ultrafast lattice dynamics in real materials with moderate computational cost. Application to graphene and silicon will demonstrate the capabilities of this novel first-principles tool, which expands the scope of nonequilibrium first-principles calculations.

[1] J.-J. Zhou, et al. Comput. Phys. Commun. 264,107970 (2021)

[2] X. Tong, M. Bernardi, Phys. Rev. Research, 3 (2021)

[3] D. R. Reynolds, et al. ACM Trans. on Math. Software, 49(2), pp. 1-26 (2023)

* This work was supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research and Office of Basic Energy Sciences, Scientific Discovery through Advanced Computing (SciDAC) program under Award No. DESC0022088.

Presenters

  • Jia Yao

    Caltech

Authors

  • Jia Yao

    Caltech

  • Ivan Maliyov

    CNRS, Caltech

  • David J Gardner

    Lawrence Livermore National Laboratory

  • Carol S Woodward

    Lawrence Livermore National Laboratory

  • Marco Bernardi

    Caltech