Question about transiet simulations in Fluent

Vitaly · February 18, 2012, 16:17

Hello,

Could someone please clarify this for me....

When doing transient simulations in fluent, I set an adaptive time stepping technique (or constant time step, does not matter). I also specify the maximum number of iterations fluent can do in a single time step.

It seems to me that if fluent did not reach a specified error threshold, it should either reduce the time step (if adaptive) and repeat Newton's method again, or just give an error (if constant time step) and stop the simulation. The user should then either ease the error threshold or manually reduce the time step.

Yet, it seems that fluent does the amount of iterations I specified, accepts the solution even if the error threshold did not get reached and proceeds to the next time step. This means that my solution is junk because it's not within error!

Someone please clarify this in case I am not understanding it correctly...

Thank You!

Vitaly

emreg · February 18, 2012, 16:20

i prefer to select changing time step size myself.

and you have to try that..

Vitaly · February 18, 2012, 16:27

Hi Emreg,

What do you mean by that?
I can either specify an adaptive time step or a constant time step.

Throughout a simulation, fluent is calculating for hundreds of thousands of time steps. Nevertheless, I want to confirm with someone on this thread that each time fluent is carrying out a time step, it is reaching the error threshold I set, or repeating the time step with a different dt.

Thanks,

Vitaly

me3840 · February 19, 2012, 16:49

Fluent will proceed to the next time step once the maximum number of iterations per timestep (which you define) is reached, regardless of your residual threshold. If you set it to switch to the next dt at e-05, and use a maximum of 20 iterations, and it reaches 20 iterations but is only at e-03, it will stop that time step and continue to the next one anyway.

Vitaly · February 19, 2012, 19:26

Hi ME3840 thanks for your response.

This seems like a terribly way of doing numerical analysis. Every time step could be accumulating error and I would not over know it, because I can't sit in front of the monitor for 3 days.

Basically, error thresholds set in Fluent are useless when presenting data because you have no idea if it was satisfied.

me3840 · February 19, 2012, 23:06

Not really. So long as you choose your timestep carefully, you should be okay. If you're just guessing, yes, you'll get pretty bad errors. The error thresholds are there to save you computer time. Why bother doing another 5-6 iterations per timestep when your solution is already converged? Fluent will not iterate until converged per timestep for the same reason - what if your setup is too poor and it never converges for a given timestep, but continues iterating? You would have wasted a large amount of computer time.

Generally you only have to watch the first few. If your timestep is chosen well, then your solution should converge in around 20 iterations. Fluent tends to speed up as the calculation goes on, so the first few usually represent the maximum amount of iterations per timestep that will be taken. If your simulation does not converge in 20 or so iterations, you need to reduce the timestep.

February 18, 2012, 16:17	Question about transiet simulations in Fluent	#1
Vitaly Member Vitaly Join Date: Jan 2012 Posts: 32 Rep Power: 14	Hello, Could someone please clarify this for me.... When doing transient simulations in fluent, I set an adaptive time stepping technique (or constant time step, does not matter). I also specify the maximum number of iterations fluent can do in a single time step. It seems to me that if fluent did not reach a specified error threshold, it should either reduce the time step (if adaptive) and repeat Newton's method again, or just give an error (if constant time step) and stop the simulation. The user should then either ease the error threshold or manually reduce the time step. Yet, it seems that fluent does the amount of iterations I specified, accepts the solution even if the error threshold did not get reached and proceeds to the next time step. This means that my solution is junk because it's not within error! Someone please clarify this in case I am not understanding it correctly... Thank You! Vitaly

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Question about modeling an elastic wall in Fluent	Mosheraa	FLUENT	2	June 24, 2011 19:47
few quesions on ANSYS ICEMCFD and FLUENT	Prakash.Paudel	ANSYS	0	August 12, 2010 13:07
Fluent Output Data Question	fossu15a	FLUENT	3	March 30, 2009 20:29
CP values higher than 1 in fluent simulations	farhan	FLUENT	8	June 22, 2007 10:24
Fluent question	Q	FLUENT	1	May 5, 2006 19:18

February 18, 2012, 16:20		#2
emreg Senior Member Emre G Join Date: May 2011 Location: Turkey Posts: 126 Rep Power: 15	i prefer to select changing time step size myself. and you have to try that..

February 18, 2012, 16:27		#3
Vitaly Member Vitaly Join Date: Jan 2012 Posts: 32 Rep Power: 14	Hi Emreg, What do you mean by that? I can either specify an adaptive time step or a constant time step. Throughout a simulation, fluent is calculating for hundreds of thousands of time steps. Nevertheless, I want to confirm with someone on this thread that each time fluent is carrying out a time step, it is reaching the error threshold I set, or repeating the time step with a different dt. Thanks, Vitaly

February 19, 2012, 16:49		#4
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 25	Fluent will proceed to the next time step once the maximum number of iterations per timestep (which you define) is reached, regardless of your residual threshold. If you set it to switch to the next dt at e-05, and use a maximum of 20 iterations, and it reaches 20 iterations but is only at e-03, it will stop that time step and continue to the next one anyway.

February 19, 2012, 19:26		#5
Vitaly Member Vitaly Join Date: Jan 2012 Posts: 32 Rep Power: 14	Hi ME3840 thanks for your response. This seems like a terribly way of doing numerical analysis. Every time step could be accumulating error and I would not over know it, because I can't sit in front of the monitor for 3 days. Basically, error thresholds set in Fluent are useless when presenting data because you have no idea if it was satisfied.

February 19, 2012, 23:06		#6
me3840 Senior Member Join Date: Nov 2010 Location: USA Posts: 1,232 Rep Power: 25	Not really. So long as you choose your timestep carefully, you should be okay. If you're just guessing, yes, you'll get pretty bad errors. The error thresholds are there to save you computer time. Why bother doing another 5-6 iterations per timestep when your solution is already converged? Fluent will not iterate until converged per timestep for the same reason - what if your setup is too poor and it never converges for a given timestep, but continues iterating? You would have wasted a large amount of computer time. Generally you only have to watch the first few. If your timestep is chosen well, then your solution should converge in around 20 iterations. Fluent tends to speed up as the calculation goes on, so the first few usually represent the maximum amount of iterations per timestep that will be taken. If your simulation does not converge in 20 or so iterations, you need to reduce the timestep.