Particle methods for geophysical flows: recent advancements in GPUSPH
Résumé
GPUSPH is a computational engine for the NavierStokes and heat equations that implements the Smoothed Particle Hydrodynamics (SPH) numerical method using Graphics Processing Units as high-performance parallel computing hardware. SPH is a Lagrangian mesh-less method which is well suited for the modeling of complex fluids, thanks to its capability to handle, in a mostly implicit manner, aspects such as a free surface, large deformation, fragmentation and the wet/dry transition. In its standard formulations, it's a highly parallelizable method, well-suited for implementation of parallel computational hardware. At the Osservatorio Etneo (Catania Section of the Istituto Nazionale di Geofisica e Vulcanologia, INGV), we designed GPUSPH to leverage these aspects for the detailed modelling of lava flows in three dimensions. More recently, through a number of international collaborations, the model has been extended with a large variety of rheological models (Newtonian, HerschelBulkley, granular) and related features covering the modeling needs of most geophysical flows (e.g. lava flows, landslides, tsunamis). Here we review the main features of GPUSPH and present the most recent progress made in the modeling and computational fields, with a showcase of its application to natural hazards.