[Hamps]

Hi all,

I'm trying to simulate non-premixed MILD combustion in a laboratory combustor. I'm running steady state using the Realizable k-epsilon model for turbulence and Eddy Dissipation Concept (EDC) for turbulence-chemistry interaction. I am using a well known global 4-step methane-air mechanism with 7 species to try and investigate different model constant setups. Reaction rates are calculated using the Chemkin solver with ISAT.

My simulations run fine on my local machine (although slow), however, when I run them on our cluster, things start to get weird. The simulation will run for a while, sometimes less than an hour, sometimes over ten hours at 100 cores. The residuals are dropping nicely without any large oscillations when suddenly the solver crashes, giving the following error:

ellu_radii: bad singular value

I have been in touch with our cluster support and they say it's most likely some memory-related problem with Fluent, as opposed to the cluster itself. Oddly, I have run simulations without chemistry, as well as with the non-premixed combustion model without any issues previously.

My desktop runs windows and the cluster is Linux, so to start simulations on the latter I have done the following:

1. Write case and data files as non-binary (turn off binary checkbox in file-write)

2. Upload to cluster

3. Change linebreaks from dos to unix by running the dos2unix command on all files (.jou, .cas, .dat, as well as chemkin files)

4. Run simulation

Is there some step I am missing when converting from Windows to Unix files, or is anybody else familiar with the error? A google search doesn't yield anything, although the name indicates that it is related to some singular matrix in the solver.

Would be grateful for any insight!