A Novel Approach for Data Analysis Based on Visualization of Phase Space Distribution Function in Plasma Turbulence Simulations

Gyrokinetic simulations are important for analyzing magnetically confined plasmas. However, the data obtained from the gyrokinetic simulations are time-series of a five-dimensional phase space distribution function, making analyzing the transport phenomena extremely difficult because of its high dimensionality and large data size. We propose a novel method for analyzing such phase space distribution functions. First, the two-dimensional velocity space distribution function is mapped into the wavenumber space and visualized as an image. This enables us to easily capture the global features and the features of the individual velocity space distribution functions. Second, we apply similarity analysis based on the local features of images and cluster analysis based on distances between images and the velocity space distribution function. The proposed method enables us to automatically extract similar structures in the velocity space distribution function and quantify the duration of these structures.