Suppressing negative massfracs

[RSuryaNarayan]

Hello,

Is there a way to suppress negative massfractions and reset them to 0? I do know that I could do this in post_restart using the state, but I am wondering if there's something already inside Pele that does this. Currently, this is causing CVODE to crash...

I observe this with methane combustion particularly. H2 with Li-Dryer works just fine!

[marchdf]

Hi, we currently do not have an option for that and have been very careful about introducing such an option. We let the user do that for the EB because the current state redistribution is not designed for algebraic constraints among variables. Typically, we've found that negative mass fractions away from the EB are a sign of something else going wrong. However, if you feel like this would solve your problem and that you can live with the consequences, then doing something in post_advance to clean the mass fractions after each time step, would work. Note that if you do that, you have to worry about mass conservation, consistency among other variables, etc.

[baperry2]

ode.atol=1e-20 should be plenty tight enough for GMRES and potentially could be quite slow. I would have tried rtol of 1e-12 and atol of 1e-14. But it would still worth trying a different solve type just in case.

[drummerdoc]

Years ago I hit the annoying issue of a built-in instability as follows: take the reaction O + O <=> O2, and assume you start with a small amount of O in a cell. If the time step is even slightly too large so that you deplete the O2 to a teeny negative amount, even 1.e-30 and even after many careful steps together there, things go bad quickly. The forward reaction goes as [O]^2, which then quickly drives [O] very negative. This one is obviously a bad actor if it's in your set, but there are likely others that have the same effect indirectly. If you can isolate a cell that goes bad, you can look for that and we can offer suggestions around it.

[RSuryaNarayan]

thanks @drummerdoc! Interestingly, I find these oscillations to get smaller (O(1e-7) i.e. 0.1ppm) when I switch to the FFCM1_Red mechanism already there in the PelePhysics repo. Just that I can't find a citation for that anywhere but it looks to work well with ReactorRK64

although, is there a way to know when CVODE crashes, which exact cell(s) are the problematic ones leading to the crash?

[drummerdoc]

More interestingly, Hai Wang, the keeper of FFCM1 doesn't recognize the Red mech at all, but did send me an updated reduced mech that I will insert into PelePhysics when I get a chance. Since the new one will be referenced and supported, it would be best if that worked for you as a better replacement. I'll let you know when I do that.

[RSuryaNarayan]

@drummerdoc thank you for the information. If you are talking about the 12 species one (FFCMy_12) mechanism here: https://www.sciencedirect.com/science/article/pii/S0010218024000890 have it imported into PelePhysics already. Will port FFCM1-21 mentioned in the paper also into PeleC and see if that works. Thanks a lot!