Machine-Learning Characterization of Intermittency in Relativistic Pair Plasma Turbulence: Single and Double Sheet Structures

Abstract

The physics of turbulence in magnetized plasmas remains an unresolved problem. The most poorly understood aspect is intermittency -- spatio-temporal fluctuations superimposed on the self-similar turbulent motions. We employ a novel machine-learning analysis technique to segment turbulent flow structures into distinct clusters based on statistical similarities across multiple physical features. We apply this technique to kinetic simulations of decaying (freely evolving) and driven (forced) turbulence in a strongly magnetized pair-plasma environment, and find that the previously identified intermittent fluctuations consist of two distinct clusters: i) current sheets, thin slabs of electric current between merging flux ropes, and; ii) double sheets, pairs of oppositely polarized current slabs, possibly generated by two non-linearly interacting Alfv\'en-wave packets. The distinction is crucial for the construction of realistic turbulence sub-grid models.