Abstract
Simulating high-Reynolds-number turbulent flows remains one of the grand challenges in computational fluid dynamics due to the vast range of spatial and temporal scales dictated by the Kolmogorov cascade. While adaptive mesh refinement (AMR) techniques provide a theoretically sound mechanism to resolve localized coherent turbulent structures without the prohibitive cost of global uniform discretization, their scalability in massively parallel distributed-memory environments is severely constrained by dynamic load imbalance and communication bottlenecks. In this paper, we propose a novel parallelized AMR framework founded on a scalable, forest-of-octrees topology integrated with a localized wavelet-based multiresolution error indicator and an asynchronous dynamic load-balancing scheme governed by Hilbert space-filling curves. The mathematical formulation ensures strict conservation across refinement interfaces via flux-correction operators tailored for high-order finite-volume formulations of the compressible Navier-Stokes equations. We evaluate the proposed algorithm on rigorous benchmark problems, including the three-dimensional Taylor-Green vortex breakdown and wall-bounded turbulent channel flow at friction Reynolds numbers up to Re_tau = 590. Our computational results demonstrate excellent strong scaling efficiency exceeding 82% up to 16,384 processing cores, alongside a reduction in total grid degrees of freedom of up to 74% compared to uniform grid resolutions, without compromising turbulent kinetic energy dissipation spectra or Reynolds stress profiles.
