When using the nearest-projection mapping with ASTE, the splitting line for each sub-domain used by ASTE (the argument -n for precice-aste-partition or mpirun) during the mapping process if we use more than one MPI process is clearly visible in the relative error plot. This phenomenon appears consistently across all test functions I’ve used so far (e.g., the sum of the squared coordinates at each point or the standard sphere test function), and it occurs within relatively large error ranges. While the error in this case is very small, more complex functions exhibit larger error ranges, and the stitching lines between sub-domains remain visible. Note that this does not always result in lower errors at the interfaces between sub-domains.
I have not tested this setup on real cases however or more complex geometries but I will try later.
Is this a known issue or should I open an issue on the ASTE repository?
Update: after further testing, it appears that this problem is persistent on other platforms with the latest ASTE version (develop branch): Using 8 processes, the partitions are clearly visible.
Here are the versions used this time : Metis V5.1.0, VTK V9.4.2 and ASTE V3.3.0. The previous post was using older versions. Below another example with another test function. I don’t know how relevant this is.
preCICE cannot currently handle connectivity information between partitions as it partitions nodes instead of cells. Hence, the connectivity information is lost.
Without connectivity, the nearest-projection mapping falls back to mapping onto the nearest edge or vertex, meaning the error should be at most the error of the nearest-neighbour mapping at the partition boundaries.
