Hello!

What is the maximum number of iterations one should expect to reach convergency. I tried up to 5000 but not reached to convergency. Is there any guide line one can follow to determine the number of iterations one can tray to see if convergency can be achieved.

Hi

Your question is too broad, keep in mind that the itertion number is different for each case and also it depends of the turbulence model selected, relaxation factors, discretization and grid type.

You should provide more infromation about your case!!

Regards

Alex Munoz

Dear Munoz

I agree with you. But I just wonder rougly under extreme cenario what the case could be. I am enquiring not problem specific answer. I am traying to get a guide line so that I can get a feeling to say well this is too much iteration attempt let me do some adjustemnt.

You can even tell me from your previous experience the highest number of iteration which leads to convergence in CFD analysis you encountered.

It's difficult to give general guidelines because the number of iteration you need to obtain a converged solution depends on a lot of factors (models, boundary and initial conditions). FLUENT considers a solution converged if all scaled residuals reduce of three order of magnitude (six for energy). But if you set a good initial solution for your problem, you couldn't notice such a reduction because you're already close to the converged solution.

However, for steady calculations I think you should stay around 1000-2000 iterations. While for unsteady calculation, the manual suggests 20 iteration per time step, but sometimes I had to increase to 30-40 at the beginning of the calculation.

5000 iterations seem too much, maybe your problem is ill posed. Check boundary conditions. Or maybe you're initial solution is extremly good, so you can't have a big reduction of residuals. Check their behaviour to understand that (are they stable/reducing or they vary a lot?).

Hi and good work

ap

Hi

I have follow ap's answers not only for this case but also in other cases. He seems to me like a senior CFD user. Therefore, I am completely agree to his comments.

I only want to added some ideas like

Too many iteration is sign of poor mesh or wrong mesh selection (structured, unstructured, hexahedral tgrid, etc)

I have run cases where only 600 iterations are enough, However, I have read that some researchers use 5000 iterations for the same case. In my opinion the diferences could be the goals of your project.

In my case I just want the velocity field, in the researchers case, they want the velocity, turbulent kinetic energy (TKE)and its dissipation rate(TDR). Now, you can understand that the iterations number depends of your objectives.

Boundary conditions are fundamental in whatever case, after all, what a CFD code does is projection of the boundary condition inside the domain. For instance, right boundary condition for x-velocity, TKE and TDR leads to reduce the number of iteration to half or less.

Now, If you are running an unsteady case, a few iteration per time step could tell you that the spatial resolution is not enough to capture the periodicity of the flow. Too many iteration per time step could tell you that the time resolution is too high an you need to reduce the time step size instead of increasing the number of iteration per time step.

I hope I help you! I know that you try to keep the confidenciality of your proyect. However, sooner or later you have to provide some information in order to have progress.

Anyway, If you need more help you can send me an e-mail. Do not forget that "ap" knows a lot about CFD.

Best regards

Alex Munoz

Thanks for compliments Alex.

Hi

ap

Alex Munoz and ap thank you for your valuable explanation on the subject.

Solomon

[Aeroma]

Originally Posted by ap
;105586

It's difficult to give general guidelines because the number of iteration you need to obtain a converged solution depends on a lot of factors (models, boundary and initial conditions). FLUENT considers a solution converged if all scaled residuals reduce of three order of magnitude (six for energy). But if you set a good initial solution for your problem, you couldn't notice such a reduction because you're already close to the converged solution.

However, for steady calculations I think you should stay around 1000-2000 iterations. While for unsteady calculation, the manual suggests 20 iteration per time step, but sometimes I had to increase to 30-40 at the beginning of the calculation.

5000 iterations seem too much, maybe your problem is ill posed. Check boundary conditions. Or maybe you're initial solution is extremly good, so you can't have a big reduction of residuals. Check their behaviour to understand that (are they stable/reducing or they vary a lot?).

Hi and good work

ap

Hey, Is it possible to iterate a solution too much that the solution becomes wrong while it was correct and converged? I have a supersonic ramjet simulation and everything is stable at 5500 iterations and temperature jumps while mass flow lines crossed paths and seperated. I would conside it converged when mass flows are cery close and everything is stable but after some point it changes.

Best

Ertugrul