Abstract Details
Abstracts
Author: Minglei Yang
Requested Type: Consider for Invited
Submitted: 2023-03-23 19:24:37
Co-authors: Diego del-Castillo-Negrete, Guannan Zhang
Contact Info:
Oak Ridge National Laboratory
5200, 1 Bethel Valley Rd
Oak Ridge, TN 37830
US
Abstract Text:
Particle-based computations play a key role in numerical simulations of plasmas in general and magnetically confined fusion plasmas in particular. These computations are time-consuming due to the multiscale dynamical processes involved and the need to follow large ensembles of initial conditions to avoid statistical sampling errors. Motivated by the need to overcome these computational challenges, we present a novel method to accelerate particle-based computations. Our approach is based on the use of Normalizing Flows, a powerful machine learning technique, to construct surrogate models for the fast integration of stochastic differential equation (SDE) corresponding to Fokker-Planck models of plasmas kinetics. In contrast to the computationally expensive standard Monte Carlo methods, the proposed method can directly generate samples of the SDE's final state bypassing the integration. In particular, the normalizing flow model can learn the conditional distribution of the state, i.e., the distribution of the final state conditioned to the initial state, such that the model only needs to be trained once and then used to handle arbitrary initial conditions. This feature provides significant computational savings when studying the dependence of the final state on the initial distribution. Two applications are presented. The first one is the simulation of the hot-tail generation of runaway electron (RE) resulting from the fast thermal quench during tokamak disruptions. Once trained, the proposed surrogate model can accurately simulate the production of RE for arbitrary initial electron distributions without the need to integrate the orbits. The second example considers transport in the ABC (Arnold-Beltrami-Childress) velocity field, a paradigmatic model of chaotic advection in 3D fluids, perturbed by Brownian noise modeling diffusion. Both applications are benchmarked and compared in terms of efficiency and accuracy with direct Monte Carlo simulations.
Comments: