CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   rESIDUALS and rElaxation factors (http://www.cfd-online.com/Forums/fluent/74397-residuals-relaxation-factors.html)

Mohsin March 31, 2010 09:09

rESIDUALS and rElaxation factors
 
hELLO

I hope you people are doing well. My question is regarding residuals and under relaxations values. Could anyone please tell me that how much under relaxation can be given what is the miniumum limit. for example

I have given an under relaxation of 0.05 for pressure and 0.2 for others including density, body weighte, momentum. Can i furthur reduce it?what would be the impact on solution if I furthur reduce it? when i reduce the under relaxation factors more it drastically redduces and converges. is it ok? Please guide

Thanks

Chris D March 31, 2010 09:21

Under-relaxation factors should not influence your final solution. They only control how much a variable can change from one iteration to the next. This is to improve stability. Setting low values for URF's means that it will take longer to reach the final solution, but it won't influence what your final solution actually is. There's a good write-up about them in the FLUENT User's Guide.

I think a value of 0.05 is too low. Is the calculation blowing up? If so, then you might have some other problem. Without knowing anything about what you're doing, I would suggest setting a better initial condition (with FMG initialization) or making sure your grid is not the problem.

Mohsin March 31, 2010 09:30

Thank you Chris

FMG initialization is not applicable in my case because i m working on Multiphase flow using DPM model.
For solution initialization i used the initial values of velcoity at the inlet. The residual graph is sloping downwards but at some stage (after 1000 iterations) it move up as it moves up i change the under relaxation factos. so doing so I reached to my converged solution that was 10^- 3 for velocity and 10^-6 for Continuity.

while reaching to a converged solution my under relaxation factors went like this

1. Pressure=0.05
2. Density=0.3
3. Body Forces=0.3
4. Momentum=0.2
5. Turbulent Kinetic Energy=0.3
6. Turbulent Dissipation rate=0.3
7. Turbulent viscosity= 0.3

These under relaxation facotrs gave me a converged solution. I read through the user guide but it said u can reduce the under relaxation facyors to 0.1 for pressure and 0.4 for momentum and so on. But it didnt give me any informationw hether i could reduce it more and if i reduce it more what impact would it give to my solution. I hope i m able to convey my problem to you now

Mohsin March 31, 2010 09:47

Under relaxation facotrs help provide stability to the solution. Stability means the variations in the residual graph would be less and it will go down slowly means time it would take is same but straight line trend would be greater (less change in the previous value) as in the default under relaxation factors.

But why it moves down drastically initially??

Chris D March 31, 2010 09:56

The more you reduce the URF's, the less the solution changes from one iteration to the next. You could lower them more, and the only impact it should have is that it takes your solution longer to converge. You should, however, converge to the same solution, regardless of the URF's.

The reason they exist is to keep your solution from going unstable by preventing wild changes in a solution variable. If your solution is converging, then I can't think of a reason why you would want to reduce them further.

Also, I would suggest that you don't judge convergence only with the residuals. This could be misleading. Instead, you should also monitor a few physical quantities that you're interested in, like mass flow rate, drag, etc. By doing this, you might find that, although the residuals told you that you're converged, you actually aren't.

Chris D March 31, 2010 10:10

(I was writing my previous post while you were posting.)

Quote:

Originally Posted by Mohsin (Post 252494)
Stability means the variations in the residual graph would be less...

I'm not sure about this. I don't think stability has anything to do with how fast the solution converges, or even that it will converge. It only means that it won't blow up. But I think you're right about lowering the URF's causing the solution to converge more slowly.

I'm not sure what you mean by "But why it moves down drastically initially?" What moves down? The residuals?

Mohsin March 31, 2010 13:14

I mean that when you reduce URFs and then iterate U will see on the residual graph that it moves down drastically at first with straight line downwards. I dont know why it does so. I think it tries to reach to the solution very fast and if u reduce it much further than it can lead to a false solution even if the residual is showing convergence.

Chris D March 31, 2010 14:25

The sharp drop is because of how FLUENT calculates residuals. For the pressure based solve, the continuity residual is the rate of mass creation, summed over all the cells in the domain.

By decreasing the URF's, you are forcing the solution to change slowly from one iteration to the next. By forcing the solution to change slowly, you are essentially lowering the rate of mass creation, which you would see as a rapid decrease in residuals. Something similar probably happens with the other residuals.

So, when you reduce the URF's and the residuals drop to 10^-6 or something, it does not mean that you have converged. Far from that. You've only played a numerical trick on yourself. :)

Mohsin March 31, 2010 14:45

Ya Chris you are absolutely rite.

Thanks for ur time...:)

chinc October 24, 2010 01:22

Hi Mohsin,

I came across your thread. In fact many CFD users have misinterpreted the residual plot as the only convergence criteria. Do have a look at the following link which explains very clearly about issues on convergence criteria. Generally speaking, a converged iteration does not imply a correct solution.

http://www.fluent.com/about/news/new...1v10i2/a20.htm

Cheers,
ChinC


All times are GMT -4. The time now is 08:16.