on January 20, 2026 at 10:15 am
Modelling dusty fluid for the exascale
L.Sewanou, G.Laibe , B.Commer¸con
November 18, 2025
Dust grains pervade the universe and play a crucial role in the formation of protoplanetary disks and stars. Yet numerically modelling a gas–dust mixture with multiple grain species remains challenging, as both memory requirements and computational costs grow rapidly with the number of dust sizes included. At the same time, the arrival of exascale supercomputing offers the opportu- nity to simulate dusty astrophysical environments at unprecedented resolution and within feasible time frames, supported by new algorithms and numerical techniques.
In this seminar, I will introduce a novel scheme based on Taylor-series scaling-and-squaring matrix exponentiation on GPUs to integrate a first-order linear ODE. I will highlight the key aspects of incorporating this method ef- ficiently into multi-GPU codes and present benchmark results demonstrating its performance—paving the way toward high-resolution simulations of dusty astrophysical systems.
In the second part of my talk, I will turn to algorithms for solving Poisson’s equation in grid-based astrophysical codes. I will present recent progress on implementing a self-gravity module in the Shamrock code, including the Poisson solvers developed so far, initial test results, and future directions.
Atrium, Bâtiment ESPRIT
français
English