Compressing the time series of five dimensional distribution function data from gyrokinetic simulation using principal component analysis

Abstract

Phase space structures are extracted from the time series of five dimensional distribution function data computed by the flux-driven full-f gyrokinetic code GT5D. Principal component analysis (PCA) is applied to reduce the dimensionality and the size of the data. Phase space bases in (φ,v∥,w) and the corresponding spatial coefficients (poloidal cross section) are constructed by PCA, where φ, v∥, and w, respectively, mean the toroidal angle, the parallel velocity, and the perpendicular velocity. It is shown that 83% of the variance of the original five dimensional distribution function can be expressed with 64 principal components, i.e., the compression of the degrees of freedom from 1.3×1012 to 1.4×109. One of the important findings—resulting from the detailed analysis of the contribution of each principal component to the energy flux—deals with avalanche events, which are found to be mostly driven by coherent structures in the phase space, indicating the key role of resonant particles. Another advantage of the proposed analysis is the decoupling of 6D (1D time and 5D phase space) data into the combinations of 3D data which are visible to the human eye.

DEISA: Dask-Enabled In Situ Analytics

“…The PCA is computed using the parallel incremental PCA provided by Dask [32] that is more memory efficient than the basic PCA algorithm. This analysis has not been selected for its physical significance in this specific case, but because it is representative of analytics used in real simulations [33].…”

Correlation of the Orthogonal Basis of the Core Plasma Distribution to the Divertor Footprint Distribution in LHD

“…2D matrixes of the ion saturation current I in database A and the electron pressure p e in database B were analyzed by the POD method, which does not use any artificial orthogonal function [12,13] and is generally the same or similar to other multivariable analysis methods, e.g., PCA and singular value decomposition (SVD). Such the multivariable methods have been often applied to magnetic confinement plasmas [14][15][16]. In this study, the POD analysis procedure of the matrix I is the same as the previous study [9].…”

Prospect for Experimental Investigation of Phase-Space Turbulence in Magnetically Confined Fusion Plasmas

“…Due to the limited opportunity for experimental study of phase-space turbulence, the aid of kinetic simulation works is inevitable. Recently, a direct survey of phasespace structures in kinetic simulation data was performed [41]. Figure 8 (a) shows the perturbed part of the phasespace structure in a set of fixed positions except for the toroidal angle and the parallel velocity coordinates during an avalanche event.…”

“…Fig.8(a) Survey of phase-space structure in five-dimensional kinetic simulation space, reproduced from[41], with the permission of AIP Publishing, and (b) numerical turbulence diagnostic simulator[42]. ©IOP Publishing Ltd. All rights reserved.…”