Building on this topic, I want to check my understanding of when connectivity information is required for volume coupling. Suppose I have a hexahedral mesh for my fluid solver. I would expect that I could pass the cell centroids to preCICE as a point cloud, with which I could do just-in-time data mapping with any of the RBF interpolation methods, or just nearest neighbor. I would only need to split the cells into tetrahedra if I wanted to use nearest-projection. In other words, the information my fluid solver needs to pass to preCICE for volume coupling is analogous to what it sends for surface coupling. Is this correct?
Is there any reason to prefer nearest-projection or RBFs for coupling a fluid solver to a particle solver?
Connectivity information for volume coupling is only required for mappings:
- both
to and from meshes of a mapping using constraint="scaled-consistent-volume"
to mesh of a linear-cell-interpolaiton mapping with constraint="conservative"from mesh of a linear-cell-interpolaiton mapping with constraint="consistent"
It’s also used by <watch-intergral scale-with-connectivity="true" ... />.
Only if your mapping constrain requires it. As you do volume coupling, linear-cell-interpolation is likely what you want to use. nearest-projection is for surface coupling.
Correct
The linear-cell-interpolation mapping can be much faster and easier to debug than an RBF mappings.
@DavidSCN and @r_walloner have more experience with this though.
At the moment, just-in-time mapping only supports RBF and NN anyway. As @fsimonis already mentioned, both of these do not require connectivity information. So if you want/need to use just-in-time mapping, that would probably be the limiting factor.
As for preferring RBF over NN, in my experience RBF can yield a more stable coupling. When using NN for mapping fluid velocities to the particles, I experienced more solver crashes, especially at larger time step sizes.
Thanks @fsimonis and @r_walloner, this clarifies much and saves me a lot of work.
1 Like