Extreme-scale turbulent dispersion: particle tracking on GPUs for studies of Lagrangian intermittency

A Lagrangian view of turbulence naturally lends itself to studies of dispersion and transport of particles by turbulent flows. A particularly complex issue is the nature of Lagrangian intermittency, which is known to be stronger than its Eulerian counterpart but is less understood. Investigations in the Lagrangian perspective require tracking of fluid particles or so-called passive tracers. Results of high fidelity for theory and modeling require high Reynolds number, well-resolved velocity fields, reliable sampling, and an accurate interpolation scheme for fluid properties at instantaneous particle positions. In this talk, we first discuss the development of a GPU-accelerated particle tracking algorithm on the world's first Exascale supercomputer, Frontier. A dynamic local particle decomposition and a ghost-layer approach for cubic spline coefficients used for interpolation provide high performance at particle counts up to several hundred millions at grid resolutions up to $32768^3$. We use the new simulations to study the effects of resolution on Lagrangian intermittency, as well as the physics of anomalous inertial range scaling for the Lagrangian velocity increment, which has direct consequences for modeling.