[Ember] Opposed Jet Diffusion config - solution Divergence

[vijay...@gmail.com]

Dear Ember users,

    I was trying the basic Diffusion configuration example in ember. I just changed the fuel to H2:1.Cantera version=2.6.0a4 and Ember version=1.4.0.

1. I noticed that the solution started diverging if I run longer. Find attached the script I used and plots at tEnd = 0.001 and 0.003. The default convergence criteria of heat release rate also keeps diverging. Both atmospheric pressure and high pressures exhibit this behavior.

2. Many times I notice the following error message.

QssIntegrator failed: timestep too small: dt = 7.334188e-17, tn = 7.334188e-17, dtmin = 1.000000e-16

At high pressures, this error is persisting and ultimately errors out. Is there any way to fix this?

Thanks,

Vijay

[vijay...@gmail.com]

Dear Ember users,

   It seems the globalTimestep value is too high for this condition. Once I reduced it to 1e-7 from the original 1e-5, the runs were stable. Please refer plots attached for both settings.

This brings me to next question. Is there any way to do adaptive time stepping in Ember?

Thanks

[Ray Speth]

Hi Vijay,

Thanks for the question. I have observed this instability mode in a few cases before, and at least when I looked at it, it seemed to require a combination of factors, namely (a) the diffusion flame configuration, (b) using the QSS chemistry integrator and (c) using the balanced splitting scheme. Depending on the size of the kinetic mechanism, it could be more efficient to either switch to CVODES for the chemistry integration (for small to medium sized mechanisms), or to the simpler Strang splitting method with a smaller global timestep (probably more on the order of 1e-6 to 5e-6, rather than having to go all the way to 1e-7).

Would you mind posting this case as an issue on the Ember GitHub page? While I don't anticipate having the opportunity to investigate it in the near future, it would be helpful in making sure this problem doesn't get forgotten.

Regards,

Ray

[vijay...@gmail.com]

Thanks a lot for your suggestions Ray. I tried them out. It seems to help with the global time step size. I am now able to run it at higher values. But, unfortunately, the total time (wall clock time) taken for convergence doesn't seem to come down. I still have to run longer. If I plot wdot[H2O] vs time, the value seem to increase multiple orders in magnitude with higher time step size and it takes forever to come to the original value(seen at lower global time step).

Yes. I will upload this issue in Ember GITHub page.