How to judge convergence?
 

Dear All,

How are you all doing! I believe doing awesome!

I have a problem in understanding the convergence of my problems. I solve the model without cavitation and then with caviatation. I set my RMS residuals to 1e-06. I check for

1) Mass and Momentum to be below 1e-06
2) Turbulence to be below at least 1e-4
3) Volume fractions to be below 1e-4
4) Imbalance of all and most important is mass imbalance and P-volum imbalance almost near to zero
5) Flow - In - Mass boundary flow on in(viscor) and P-Mass boundary flow on in to that of Out - Mass boundary flow on Out (viscor) and P-Mass boundary flow on Out to be equal over the number of iterations.

Kindly let me know whether I am right or please correct me if I am wrong.

My Imbalance especially mass imbalance and P-volum imbalance oscillates at zero between 10 to 22 and I stop it at zero, when it has run long

Please some on solve me or suggest me the better way.

Best Regards
Akash A V

seems to be fine. but as suggested by ghorrocks time to time " do the sensitivity analysis"

What do you mean by 10 to 22?

Dear Sir,

Thanks for your kind response.

I usually see to that all the residuals are below the set value(1e-06), but some times I even compromise at 1e-04.

Regarding the mass imbalance and P-volum imbalance oscillates (+/-) at zero between 10 to 22% and then flantens up to zero.

I wanted to ask that, is it right thing to stop the solution when the residuals have come down the set value and before it converges fully. Or if not should we allow it run still it come to halt.

Are there some kind of divergence and bisection graphs where upon we can make out the solution is converging or not.

please let me know on this. As this is making me confused with when to stop the solution, although the residuals have come below the set values(1e-06)

Once again thank you!

Best Regards
Akash A V

better to give some time, to be on the safe side, to see that solution does not change for next 20-50 iterations, If you have difficulty in getting the residuals below 1e-04, try to reduce the time step

My preferred approach is to do a sensitivity analysis. Compare the results of a parameter you care about (eg drag, maximum pressure, whatever is important in your analysis) with a range of convergence tolerances. When the parameter asymptotes out to within a tolerance you are happy to accept then that defines your convergence critereon.

Dear All,

Thanks a lot. I am doing out sensitive test, with the mesh as well as the parameters. I see with change in the mesh size and the values things change horribly and a time I felt, what ever I did was all crap.

But soon realized that nothing much changes with these changes, only certain output changes but need to again get hold upon it based uopon sensitive tests. I will stick on to 1E-06 and is the best approach.

Reduce all residuals below 1e-06 and see to the imbalane graph and check all variable are at atleast 0.1% +/- between zero. Check other parameters like mass flow in and going out (must be all most the same) and the mass viscor, turbulence Rms all below 1e-04.

Send me your email ID I will send the convergence graphs

Please correct me if I am wrong

Best Regards
Akash A v

please provide the convergence plots here, so that other users may be benefited by your work

for mass flow rate, try to get the 0.01% or less

No Far, you cannot prescribe general convergence recommendations. That figure might work fine in your work but this is a totally different case so it is likely to be far tighter than is required. The only rational approach is to do a sensitivity analysis and establish what this specific simulation needs.

So 0.01% is tighter than required? I always enforce this condition, because I read in many paper that mass flow rate imbalance should be equal to or less than 1e-04.Though I never checked that what happens if criteria is 1e-03 or even 1e-02.

PS. In external aerodynamics CFD I put less stringent convergence criteria than for the internal flows, specially for turbo-machinery

I totally agree with the recommendation. I should rather say that every body who is involved CFD analysis must do the sensitivity analysis as it has two advantages.

1. Saves alot time in future

2. Confidence level in results is higher, and every body shall trust your CFD, even yourself!!!!!

Your 0.01% was actually the mass residual? From your initial comment I suspected you were talking about a mass imbalance. 1e-4 is a loose residual (but often OK for design purposes) but 1e-4 as a mass imbalance is very tight.

But again, the best approach is to do a sensitivity study and find out how tight your exact case requires.

Yes I always try to get this or even 1e-05 as mass imbalance

Yes it is loose residual. But I always try for 1e-07 even 1e-08, but alteast I ensure that residaul for mass, momentum, energy and turbulence quantities are alteast 1e-06