[EternalSeekerX]

Hello peeps,

So I am running the Turbulent Onera M6 case in seriel on an arm device. Reading the config file it shows the convergence criteria and max iterations. At what iteration should it converge? It shows max iteration of 999999 and I'm currently at 162800 and it's been more than 36 hours. Should I expect it go way longer?

Also I notice that the solver writes to file from time to time. I opened up paraview to see the solution so far. I was able to visualize many parameters, but I dont see them change for each time step, is that normal?

Example:

[Situla]

Hi!


I am really not an expert, but to me it looks like the simulation has converged. At the end of the day, the convergence criterion is up to the user, and I still see some fluctuation or trend in what I assume is the cL, but it is only the last digit.
Personally I prefer using Cauchy criteria (while also paying attention to residuals), and based on those your results are probably fine.


This also explains why the intermediate results you look at in Paraview do not seem to change.


In your config file, you should find a convergence section where one you can define criteria other than just iteration count. From my experience, simply setting a fixed number of iterations is not a good idea. Either you will waste time with an already converged solution, or you will abort before it is converged. Both is generally bad. I recommend switching the convergence criterion to residual or Cauchy.


Having said that, I want to add that those are recommendations for your next simulations, because the one I see here looks converged to me and thus already completed.


Kind regards,
Martin

[EternalSeekerX]

Originally Posted by Situla

Hi!


I am really not an expert, but to me it looks like the simulation has converged. At the end of the day, the convergence criterion is up to the user, and I still see some fluctuation or trend in what I assume is the cL, but it is only the last digit.
Personally I prefer using Cauchy criteria (while also paying attention to residuals), and based on those your results are probably fine.


This also explains why the intermediate results you look at in Paraview do not seem to change.


In your config file, you should find a convergence section where one you can define criteria other than just iteration count. From my experience, simply setting a fixed number of iterations is not a good idea. Either you will waste time with an already converged solution, or you will abort before it is converged. Both is generally bad. I recommend switching the convergence criterion to residual or Cauchy.


Having said that, I want to add that those are recommendations for your next simulations, because the one I see here looks converged to me and thus already completed.


Kind regards,
Martin

I see. I exited the run around 170000 iterations. I didn't think of touching the config file as it was a tutorial, I am only limited to serial runs on my arm device (mpi won't work without root access) and and such maybe that is why it took longer. I did check the config file again and by default it is set to use Cauchy criteria, I think the issue was that the targeted residual was set to be something really small (1E6) but it would need to reach that target for atleast 100 iterations. Still I was worried because every other tutorial runs i have tried all converged with run times between mins to hours.

I was able to run the Aeroelastic tuturial which took 8hrs but met the convergence criteria. I was even able to edit the file to output a vtk file and then render it.

https://youtu.be/H8ZqP3zP3P8

Its pretty cool SU2 runs on arm as well.

[bsotelo]

Originally Posted by EternalSeekerX

Its pretty cool SU2 runs on arm as well.

I'm trying to install SU2 on ARM with parallel and MPI? Were you able to do this? or did you only keep it serial? I wonder what to watch out and hoping it will just work with C/C++ compiler tools.