Scaling error w/ standard drainage in distance mode

[kbruning]

I am attempting to run morphological drainage at sphere radii that exceed subdomain dimensions. I have this working at low resolution, but am getting erroneous solutions when scaling up.

I switched the standard_method for morphological drainage to mode="distance" and commented out the check_radii test in the porosimetry function. The attached files "drainage_sp.py" with "sphere_pack.in" saved in the examples folder seem to work. The sphere_pack.in was renamed to sphere_pack.txt to attach here, but should be switched back.

When scaling to voxels = (100,100,100) and reservoir_voxels = 850 or to voxels = (200,200,200) and reservoir_voxels = 1700, I get erroneous solutions. See the attached image of disconnected nonwetting fluid for the final capillary pressure state in the 200^3 case. Large reservoirs are required for very small capillary pressures, e.g. for the 200^3 case a capillary pressure of 0.10706 corresponds to a sphere radius of 1698 voxels.

sphere_pack.txt

drainage_sp.py

[kbruning]

Looks like I am not on the most up to date version of PMMoTo, this is likely my issue. I will report back or close this.

[kbruning]

I am now up to date on the main branch. Unfortunately, I am now unable to run a sphere radius that exceeds the subdomain size at all. I commented out the "check_radii" in the porosimetry function of porosimetry.py again, and tried to run the attached drainage_sp.py with distance mode at small pc. The hybrid and distance modes at larger pc = 2.0 are fine.

This input file gives this error:2025-09-09 13:28:07 [Rank 0/8] ERROR: The buffer ((46, 46, 46)) exceeds at least one dimension of the subdomain ((25, 25, 25)). Simulation stopping.

I did not see this on the drainage_img_updates branch where I could run 2x2x2 subdomains at 50^3, 100^3 and 200^3, but had solution issues with the 100^3 and 200^3.

drainage_sp.py

[tmweigand]

So i think the best approach for this is will be for you to modify the input domain. Rather than having a cut off domain and spheres, you can circumvent the issue by enlarging the domain box and adding more periodic spheres if needed. The solution will be the same.

This is my preferred approach as allowing the buffer to be larger than the subdomain size causes significant issues with how communication is handled. Non-neighboring processes will have to communicate with each other and while that may be a good feature to add down the road, it will not be straight forward.

[kbruning]

I agree that the answer would be the same. I'll try this approach.