CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > FLUENT

Epsilon Residual Will Not Converge

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   July 19, 2005, 14:39
Default Epsilon Residual Will Not Converge
  #1
Steve
Guest
 
Posts: n/a
I have an analysis in which all the residuals have converged (.001) except for the epsilon residual. The analysis is incompressible with the k-e turbulence model and steady state

It has run for 3000 iterations with a dynamic grid adaptation on the pressure.

I have tried to adapt on high gradients of the epsilon, but the epsilon residual still remains in the .02 range.

Any ideas?

thanks,

steve
  Reply With Quote

Old   July 19, 2005, 15:51
Default Re: Epsilon Residual Will Not Converge
  #2
Jason
Guest
 
Posts: n/a
Watching the residuals isn't a good idea for judging convergence since the residuals are normalized based on the initial residuals and averaged across the control volume. You should be monitoring forces, or local pressures or temperatures, etc... you should be monitoring whatever it is that you're trying to get out of your model and using that to judge convergence. If you're epsilon values have leveled off, your model may very well be converged. The default of 1e-03 was chosen because it is a value that gives decent convergence for a lot of cases. I wouldn't even go as far as to say this works for most cases, but it's as good a starting value as any. Even if your residuals all dropped below 1e-03, your model may not be converged. Your forces (or mass balance, or local pressures, or temperatures, etc...) may still be converging.

If you really want to drop the residuals, then you can switch to the double precision solver. Sometimes this works.

Hope this helps, and good luck, Jason
  Reply With Quote

Old   July 19, 2005, 16:01
Default Re: Epsilon Residual Will Not Converge
  #3
Steve
Guest
 
Posts: n/a
Thanks for the info. My plan was to run a steady state analysis intially, then switch to a transient analysis. From what you are saying, I should be monitoring the mass flow, pressure, etc. per iteration during the transient to ensure time step convergence as opposed to relying on the residuals to converge.

What is the purpose then of plotting the residual convergence during the anlaysis? If the real monitoring should take place on the desired output variables, then why monitor the residuals. Everything I have read thusfar says to monitor the residuals.

Thanks for the help,

steve
  Reply With Quote

Old   July 19, 2005, 17:03
Default Re: Epsilon Residual Will Not Converge
  #4
Jason
Guest
 
Posts: n/a
Sorry, what I was talking about before was for steady state solutions... things are a little different in transient conditions... I don't do a lot of work with unsteady flows, but I do a little, and I try to keep up with other people's comments on here. You are more dependent on the residuals in unsteady solutions, but you still don't have to hold strictly to the 1e-03 (it's not always possible, or at least feasible, to get 1e-03 drop for each time step). One judge of convergence is that the peak of each residual doesn't increase from time-step to time-step and that each residual drops the same amount for each time-step. What you don't want to happen is that the residuals "go crazy" as the time steps increase. This means there's something wrong (could be mesh, model setup, time step size <-seems to be the most common in transient solutions, or solver settings).

Also, for transient solutions you need to monitor whatever it is you're trying to get out of the solution (you can choose the "write to file" option for anything you're monitoring), or you can only get the data out of whatever time steps you've saved. So if you were looking for an unsteady loading, you would monitor the force you want for each time step, that way you don't have to save the case and data each time step.

Back to what I was talking about before for steady state analysis... The residuals aren't completely useless in steady state solutions. You still want them to either get really small, or to level off. One example I've seen is that the forces seemed to level off, but the residuals were increasing, and when you looked at the solution, it was unrealistic. Also, there's different techniques you can use to reach a converged solution depending on what the problem is. Lets say you're having problems getting a converged solution, and you notice that most of the residuals are behaving, but not the turbulence residuals... then you may turn off all of the equations except the turbulence model and run that for a little while, then turn the rest of the equations back on and see if that helped. Same thing for the energy equation if that's being a problem, or the flow equation if you're having problems with continuity or one of the velocity residuals.

The residuals are good guides to help you understand what's going on with your solution but you can't rely on them as a sole source to tell you if your solution is converged or not. There are three things that can happen if you rely no the residuals... the most up-beat is that you reach a converged solution. Sometimes the residuals say the model is converged and they're right (it can be common, or rare depending on what type of model you're working with)! Another thing that can happen is that the residuals level off at a higher value than expected, but the model really is converged. You wouldn't know if the model is converged or not without monitoring other values to confirm it though. And the third thing that can happen is that the residuals reach 1e-03 (or whatever you set as the "converged" value in the residuals monitor panel) but the model really isn't converged. If you were monitoring the forces you may see that the forces are still "drifting". You could see any of these depending on what you're modeling and how you set up your model (solver settings, choice of BC's, and mesh quality).

I hope this clarifies what I was saying, Jason
  Reply With Quote

Old   July 19, 2005, 17:11
Default Re: Epsilon Residual Will Not Converge
  #5
Steve
Guest
 
Posts: n/a
Crystal clear now. Thanks for the insight. Just as you explained, in my steady solution the mass flow rates and pressure that I was monitoring had converged several hundred iterations prior, however, the epsilon residual would not drop below .001. However, the epsilon residual did level off.

Thanks again for you help.

steve
  Reply With Quote

Old   July 19, 2005, 20:13
Default Re: Epsilon Residual Will Not Converge
  #6
zxaar
Guest
 
Posts: n/a
hey steve, i do have a bad news for you, in compressibel cases with steady state , i have observed that sometimes its very difficult to get a converged solution. i was once troubled by a 2d case with mach 0.95. the same case converged nicely with unsteady conditions.

the good news is that, if you can switch to FAS, with say multigrid levels 4 or 5 (they will be more than enough), and increase the pre and post iterations by one in multigrid control. you will see that the same case converges very nicely and easily.

aprat from this, i remember someone wrote a little tips (on this forum) few months ago about what should one do to get the solution converged (so i am not going to reiterate them, just search the forum).

but the main advice would be, stick to first order schemes in the start,, use very small courant number (i used 0.5), and try to keep your mesh as structured or hex elements as possible. do not adapt (i know its hard to hear this, but trhe soultion can converge mostly with out it, just make your best mesh).
  Reply With Quote

Old   July 20, 2005, 11:54
Default Re: Epsilon Residual Will Not Converge
  #7
Steve
Guest
 
Posts: n/a
Thanks for the advice. I will search this forum for the convergence tips you spoke about.

Thanks,

Steve
  Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
transsonic nozzle with rhoSimpleFoam Unseen OpenFOAM Running, Solving & CFD 7 April 16, 2014 03:38
Orifice Plate with a fully developed flow - Problems with convergence jonmec OpenFOAM Running, Solving & CFD 3 July 28, 2011 05:24
Extrusion with OpenFoam problem No. Iterations 0 Lord Kelvin OpenFOAM 6 April 12, 2011 11:24
Computational time sunnysun OpenFOAM Running, Solving & CFD 5 March 16, 2009 04:32
Differences between serial and parallel runs carsten OpenFOAM Bugs 11 September 12, 2008 11:16


All times are GMT -4. The time now is 13:35.