Hello,
This relates to the Kawamura test case (forced channel flow with heat transfer).
Periodicity is used the temperature is decomposed in bulk gradient and deviation, as described in this forum post http://code-saturne.org/forum/viewtopic.php?f=2&t=1612.
I've been trying to get this simulation running for some time and have been using a minimum working example approach. That is to say no turbulence model (iturb = 0) is used and a steady algorithm (idtvar = -1) is used.
I'm finding that the solution is taking a very large number of pseudo time steps to converge, one would expect the solution to converge almost immediately. The error seems to be caused by the normal residual not meeting the solver precision at each step.
I've tried making changes to many options and have run out of ideas as to what could be causing this behaviour.
Attached is the listing file and the user source terms.
Kind Regards,
Martyn
slow to resolve steady state temperature
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
mkendrick
slow to resolve steady state temperature
- Attachments
-
- listing.txt
- (71.91 KiB) Downloaded 213 times
-
- cs_user_source_terms.f90
- (5.92 KiB) Downloaded 221 times
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: slow to resolve steady state temperature
Hello,
I'll check your file later, but did you try the other "pseudo-stead" algorithm (idtvar = 2) ?
The idtvar = -1 option is not used much, and seems to provide mixed results (I'll let colleagues contradict me if they read this an I am wrong)
Regards,
Yvan
I'll check your file later, but did you try the other "pseudo-stead" algorithm (idtvar = 2) ?
The idtvar = -1 option is not used much, and seems to provide mixed results (I'll let colleagues contradict me if they read this an I am wrong)
Regards,
Yvan
-
mkendrick
Re: slow to resolve steady state temperature
I've tried running from SRC using idtvar = -1 and also running from the GUI and selecting 'steady algorithm' which I believe relates to idtvar. It would appear also that selecting between SIMPLE and SIMPLEC changes the value of idtvar. I have tested with both SIMPLE and SIMPLEC with the same behaviour.
-
mkendrick
Re: slow to resolve steady state temperature
Further to my previous email. Changing SIMPLE/SIMPLEC does indeed change idtvar between -1 and 2, respectively. This is done automatically if using the GUI.
Attached is the listing with idtvar = 2
The problem was still not converging as expected.
Attached is the listing with idtvar = 2
The problem was still not converging as expected.
- Attachments
-
- listing_idtvar2.txt
- (84.18 KiB) Downloaded 225 times
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: slow to resolve steady state temperature
Hello,
I am not sure whether by "many iterations", you are talking about time steps or linear solver iterations.
The KAWAMURA test case being quite specific (due to its periodic aspect), it is dominated by diffusion, so the Jacobi solver used for scalars is a bad fit (especially if you use large time steps).
Could you try using the BiCGSTAB solver (instead of the "automatic" choice) for all linear solvers except for pressure ? And if that fails, try GMRES. In version 4.0, the BiCGSTAB's stopping criteria is better, and we also have a BiCGSTAB2, but hopefully, that test should be enough.
Regards,
Yvan
I am not sure whether by "many iterations", you are talking about time steps or linear solver iterations.
The KAWAMURA test case being quite specific (due to its periodic aspect), it is dominated by diffusion, so the Jacobi solver used for scalars is a bad fit (especially if you use large time steps).
Could you try using the BiCGSTAB solver (instead of the "automatic" choice) for all linear solvers except for pressure ? And if that fails, try GMRES. In version 4.0, the BiCGSTAB's stopping criteria is better, and we also have a BiCGSTAB2, but hopefully, that test should be enough.
Regards,
Yvan