78th Annual Meeting of the APS Division of Fluid Dynamics (Nov 23 — 25, 2025)

V016: Particle advection in high-speed crossflow jets

Particle advection underpins a broad spectrum of applications, ranging from flow visualization to the extraction of derived post-processing quantities, including so-called 'flying sensors.' The computational efficiency of particle advection hinges upon the optimization of several tightly coupled components: (i) data input/output, (ii) the owning-cell localization strategy, (iii) velocity field interpolation, and (iv) particle trajectory integration.Building upon prior work [1,2], we have developed a novel owning-cell locator specifically tailored to rectilinear grid topologies, thereby enabling highly scalable particle-tracking capabilities. The algorithm exploits the inherent spatial regularity of structured meshes to drastically reduce the complexity of particle-cell association, a central bottleneck in Lagrangian transport solvers. The resulting reduction in search overhead facilitates near-ideal scalability across modern GPU architectures.Recent large-scale demonstrations underscore the efficacy of the method: we successfully advected 20 billion particles over a direct numerical simulation (DNS) mesh comprising 52 million control volumes, utilizing four NVIDIA H200 GPUs. These results highlight the algorithm's ability to sustain extreme-scale Lagrangian computations with minimal degradation in performance, thereby extending the frontier of particle-based analysis in turbulent flow simulations. The present visualization illustrates the dynamics of Lagrangian 'seeded' particles embedded within a subsonic crossflow interacting with a sonic jet. [1] Lagares, C. and Araya, G. 2023. 'A GPU-Accelerated Particle Advection Methodology for 3D Lagrangian Coherent Structures in High-Speed Turbulent Boundary Layers' Energies 16, no. 12: 4800. https://doi.org/10.3390/en16124800[2] Lagares C. and Araya G. Aquila-LCS: GPU/CPU-accelerated particle advection schemes for large-scale simulations. SoftwareX, 27, 101836 2024.