Convergence Criteria for first Timestep in transient run

aerost · February 14, 2013, 21:11

Hello,

Does anyone know if it is possible to have a different convergence criteria for the first timestep of a transient simulation?

I would like to converge the first timestep for a minimum of ~500 iterations, then allow each subsequent timestep to end at a maximum of say ~50 iterations. Is this possible?

THANKS!

ghorrocks · February 15, 2013, 06:47

You could try making the max number of iterations a CEL expression.

But 500 coeff loops for an iteration is crazy. This is FAR too much. Even 50 per iteration is too much. A far better way is to use adaptive time stepping homing in on 3-5 coeff loops per iteration. Much faster and more accurate.

aerost · February 18, 2013, 20:46

Thank you. I think you are right after a bit of research. These transient sims are pretty time intensive so it's not the time to be learning by trial and error.

My specific case is a turbomachinery case and I have found that the solution takes many timesteps to converge towards a realistic and periodic one. As a result I end up throwing out many of the first few timesteps. That works, but I'd prefer to find a way to converge to a useful solution in the first couple of timesteps to save time and space.

At any rate... I am trying an approach with more timesteps and a lower number of iterations to see if the results come out acceptably... thanks again.

ghorrocks · February 19, 2013, 05:01

If you do not care about the initial transient then do not bother accurately simulating it. Just do it approximately and quickly and use a more accurate setup when the flow has matured.

But the adaptive timesteps with 3-5 coeff loops per time steps is almost always the best option available. You are wasting your time by doing anything else.

February 14, 2013, 21:11	Convergence Criteria for first Timestep in transient run	#1
aerost New Member Join Date: Jul 2010 Posts: 7 Rep Power: 16	Hello, Does anyone know if it is possible to have a different convergence criteria for the first timestep of a transient simulation? I would like to converge the first timestep for a minimum of ~500 iterations, then allow each subsequent timestep to end at a maximum of say ~50 iterations. Is this possible? THANKS!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Transient run taking too long time	Alexander Tham Keat Fu	CFX	3	June 20, 2012 07:53
Force can not converge	colopolo	CFX	13	October 4, 2011 23:03
Convergence criteria	PrandtlGlawert	FLUENT	1	August 13, 2009 19:06
Convergence Criteria	edwin	FLUENT	1	February 14, 2008 20:24
Error when changing timestep in transient run	Tony	CFX	4	December 8, 2004 15:13

February 15, 2013, 06:47		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,871 Rep Power: 144	You could try making the max number of iterations a CEL expression. But 500 coeff loops for an iteration is crazy. This is FAR too much. Even 50 per iteration is too much. A far better way is to use adaptive time stepping homing in on 3-5 coeff loops per iteration. Much faster and more accurate.

February 18, 2013, 20:46		#3
aerost New Member Join Date: Jul 2010 Posts: 7 Rep Power: 16	Thank you. I think you are right after a bit of research. These transient sims are pretty time intensive so it's not the time to be learning by trial and error. My specific case is a turbomachinery case and I have found that the solution takes many timesteps to converge towards a realistic and periodic one. As a result I end up throwing out many of the first few timesteps. That works, but I'd prefer to find a way to converge to a useful solution in the first couple of timesteps to save time and space. At any rate... I am trying an approach with more timesteps and a lower number of iterations to see if the results come out acceptably... thanks again.

February 19, 2013, 05:01		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,871 Rep Power: 144	If you do not care about the initial transient then do not bother accurately simulating it. Just do it approximately and quickly and use a more accurate setup when the flow has matured. But the adaptive timesteps with 3-5 coeff loops per time steps is almost always the best option available. You are wasting your time by doing anything else.