A Peculiar Convergence Issue
 

Hi people,

I have a question regarding convergence.

I use a CFD software which is very specialised on Marine applications. It is not actually a fully RANSE approach, it is rather based on Rankine source panel method, i.e. on potential flow theory for the free surface and it is coupled with boundary layer analysis for the viscous force components.
The problem has as follows:
If the convergrence criteria are left at their default values, my cases do not converge.
I notice that the criterion that fails is the free surface elevation (waveheight residual if you like). And so, I raise it by 100% (just to experiment) and as soon as it converges, I manually call the software to execute one more iteration. Then I notice that the result after this iteration are different to the results of the same iteration with original setup.
Then, I keep asking for more and more iterations (one at a time of course) to be completed only to find out that the waveheight residual is actually decreasing even beyond the original criterion value (with the original setup it would increase and eventually totally diverge).
Why does the software miss these solutions with the original setup and happens fo find them with the manual iterations? Where would you turn your attention to? Would you alter the relaxation factors to make it "search better"? Do sou suspect another source of error?
The worrying bit is that the results do change quite substantially during those final manual iterations and so this has to be addressed properly.

Thanks for your time.

Panos