Multi-session run hangs in pressure setup (gs_setup)

[Chinthaka Jacob]

Dear Nek Community,

  I am running a two-session case where the second mesh is a strict subset of the first. The first session solves the velocity field, which is then interpolated into the second session where only passive scalars are solved. The motivation is to avoid artifacts in the velocity field near boundaries due to stabilization treatments.

  Since the second mesh is a subset, I am not using standard NekNek coupling. Instead, I run two sessions (say master and slave) via SESSION.NAME, without coupling (IFNEKNEKC = F). The velocity field is transferred from the master to the slave using Nek’s internal fgslib_findpts / fgslib_findpts_eval routines (i.e., volume interpolation based on point location and element mapping). Here is my SESSION.NAME:

2
F
slave
/full_path/...
118
master
/full_path/...
138

  The issue is that the case hangs during pressure solver setup (gs_setup). This persists with both the default XXT solver and AMG (semg_amg). Both cases completes hierarchy construction but do not proceed to time stepping (logfiles attached).

  I ran both meshes as standalone single-session cases, so the meshes themselves appear valid. At this point, the hang only occurs in the multi-session configuration, even though the sessions are not explicitly coupled. I would appreciate any insight into what in the multi-session setup could trigger this behavior during pressure initialization.

Chinthaka

[Julian Weissmann]

Dear Chinthaka,

this is a very interesting setup. I have been thinking along similar lines for geophysical flow problems, where such a decoupling of velocity and scalar transport could indeed be quite beneficial, in particular in the presence of stabilization near boundaries.

I was wondering whether you have made any further progress with this multi-session configuration. Your approach using fgslib_findpts for volume interpolation in this uncoupled setting is quite appealing, and I would be interested to understand whether this can be made robust in practice.

If you have any updates or partial workarounds, I would greatly appreciate hearing about them. Perhaps we could also be in touch offline if you are open to it.

Best,
Julian