Abstract Details
Abstracts
Author: Sebastian De Pascuale
Requested Type: Poster
Submitted: 2019-02-22 15:29:01
Co-authors: D.L. Green, R.L. Barnett
Contact Info:
Oak Ridge National Laboratory
1 Bethel Valley Rd
Oak Ridge, 37830
USA
Abstract Text:
Reduced-order models (ROM) offer the flexibility of improving compute time for plasma physics simulations at the expense of simplifying the governing system of equations. In this presentation we show the application of two decomposition techniques aimed at obtaining a representative basis for dynamic data. We test these methods on a multiscale problem with clear separation of timescales relevant to fusion plasmas: the ponderomotive modification of electron density under high power RF waves. This scenario provides a quasi-linear electrostatic force in the cold-fluid limit of kinetic theory. From numerical simulation of the 1x-1v distribution function solution to the Vlasov equation, we recover the perturbed 1D density evolution using Proper Orthogonal Decomposition (POD). POD provides a much smaller basis set over which to re-calculate the dynamics than a traditional spectral method, ranking each component by its overall weighted contribution to the simulated data. To gain further insight on the physical behavior we exploit Dynamic Mode Decomposition (DMD), which obtains a temporal eigenvalue for the linear approximation of each basis vector in the POD set. In this reduced system the DMD modes can be projected forward in time without integration. We discuss the merits of each data-driven method and the results of their application on a plasma physics problem.
Comments:
