Simulation succeeds past first time step in serial but not in parallel

MFiX
#1

Hey all,

I have a case which runs to completion at atmospheric pressure, in both serial and parallel, just fine. When I try to run this case at low pressures, the serial computation runs to completion but the parallel simulation can’t make it past the first time step (see Serial vs Parallel.zip (215.0 KB)). It seems like the first time step has very high y-velocity residuals “V0” that decrease substantially by the second time step (from ~1 → 0.0001).

It doesn’t seem like I can patch the gas phase initial condition using the results from the serial simulation, so I don’t really know how to get around this. I also don’t understand why only the parallel simulation has this convergence issue.

Let me know your thoughts! I omitted some text in the serial output log file to make the file size smaller, the first time step successfully converges at line 231.

#2

Your case runs fine in parallel on my machine. I used 20.4 release and the following:
gnu/8.2.0
openmpi/4.0.1_gnu8.2
cmake/3.19.1

#3

I’m using mfix-20.4.2 and I typically load the modules intel/17.4 and impi/17.3 for compilation. Is this not supported?

#4

Intel is supported. I don’t have Intel 17 on my system. I was just reporting what works for me and I cannot reproduce the failure you see.

#5

I guess I’ll just treat it as a fluke then. I haven’t had this problem with my other simulations, and unfortunately we don’t have the gnu or openmpi packages you mentioned on our supercomputing cluster, so I don’t think I can test those out.