prescribed velocity does not drive flow in surrounding regions

[mibillen]

Hello,
I am having a strange problem using the prescribed internal velocity cookbook in the manual

When I run this cookbook the solution shows the applied velocities in the correct location but there is no flow driven by this internal velocity in the adjacent regions (see attached image) as there should be (see image at bottom of cookbook in the manual).

I showed the issue to @lhy11009 and he suggested that I run the test in /test/prescribed_velocity.prm. This parameter file does not run "as is" because the file does not include the line needed to link to the required additional library (it does run if I add that line following the cookbook example).
set Additional shared libraries = /quobyte/billengrp/Software/aspect/cookbooks/prescribed_velocity/libprescribed_velocity.so
This made us suspicious that this test is not actually being run (or maybe we don't understand something about how the tester works?).

Some additional information:

1. We've tested the cookbook on the aspect v 3.0 main branch from BEFORE (download June 25th) the pull request that @lhy11009 and @gassmoeller were working on to combine prescribed internal solutions.
2. We've also tested the cookbook on the main branch after that pull request was merged (download October 14). We find the same issue for this simple cookbook (only modified to use the new prescribed internal solution section instead of the linked library).
3. We also ran a test using the version after the pull request for a subduction model that includes buoyancy forces, and when you look at the solution, you can see the flow being driven by the buoyancy force surrounding the region where the velocities are prescribed. So, it looks like the internal velocity constraints are not being handed to the solver properly.
4. Haoyuan also adds: "And from what we looked at, the constraints are not in the solution. I remember this is not the case when I tested before my PR a month ago, where the velocity field is affected by the prescribed solution. So I assume there might be something in the PR that breaks this." I'm not sure this is true, since we had this problem with the earlier version of Aspect before the PR was merged. This could have been broken earlier, but the tests did not catch it?

I've put the figures below and the parameter files for the cookbook below,
Thanks for your help in getting this feature working again.
Magali

prescribed internal velocity cookbook result: this should entrain flow in the rest of the box:

subduction model with buoyancy forces and prescribed velocities inside the subducting plate: the solution is not affected by the prescribed velocity solution.

prescribed_velocity/corner_flow.prm parameter file: only changed path to the compiled shared library file:
corner_flow.txt

Same model, but modified to run with new prescribed internal solution merged in main:
corner_flow_main.txt

[tjhei]

This parameter file does not run "as is" because the file does not include the line needed to link to the required additional library

This is normal. Our testing framework automatically adds the required shared library to each test that has a .cc file.

Can you check if you see the same when explicitly switching from GMG to AMG as the solver?

[mibillen]

I changed to AMG and it works! So, the problem is only in the GMG solver.

[tjhei]

Fascinating! I will try to figure out how this can be fixed.