Article ID Journal Published Year Pages File Type
10139583 Journal of Computational Physics 2018 21 Pages PDF
Abstract
GPUSPH is an implementation of Weakly-Compressible Smoothed Particle Hydrodynamics with an explicit predictor-corrector integration scheme, that takes advantage of the parallel nature of the method to run on Graphic Processing Units (GPUs). Despite the massive speed-up granted by the use of GPUs, the application of GPUSPH to the simulation of highly viscous fluids is still problematic, due to the severe time-stepping restrictions imposed by the explicit integration scheme when the viscous term becomes dominant. This is an issue in the simulation of lava flows, where the thermal-dependent rheology can lead to kinematic viscosities in the order of 104 m2s−1 or more at low temperatures. To overcome this limitation, we introduce a semi-implicit integration scheme, where only the viscous part of the momentum equation is solved implicitly. Here we show the significant advantages of our approach in terms of simulation run times as well as better quality of the results over the fully explicit scheme.
Related Topics
Physical Sciences and Engineering Computer Science Computer Science Applications
Authors
, , , , , , , ,