Hi, does GPUSPH support non-uniform resolution of initial particles? this could be useful to save calculation efforts.
while the feature is on our wishlist, I’m afraid GPUSPH does not support non-uniform particle resolution presently.
Is there any paper you are aware that can provide instructions on how this can be implemented?
off the top of my head I can recommend two references: https://doi.org/10.1016/j.cma.2012.12.014 describing the theory, and a description of the impemention in SPHysics that was presented at SPHERIC 2016 (available via ResearchGate).
Thanks Giuseppe! Do you think it is difficult to implement the multi-resolultion functions in GPUSPH? I plan to do some work on this, how much time do you think it may be necessary?
The biggest difficulty in introducing multiple resolutions in GPUSPH is in the handling of the support grid, that we use for several things:
- the cells are used as sorting buckets for the neighbors list construction;
- the cells are used as transfer units for multi-GPU and multi-node support;
- the cells center are used as reference position for the particle positions, to preserve a uniform precision in the entire domain.
Even ignoring point 2. (building the implementation assuming single GPU use only at first), I’m not sure what the best approach would be to handle 1. and 3. Fixed-size grid with resolution based on the largest supported influence radius? Multiple grids for different resolutions?