Convergence criteria

  • Last Post 07 May 2019
akin3rd posted this 05 May 2019


I am running some simulations to get the lift and drag coefficients on an airfoil.

Now that I have been able to get a high-quality mesh, my residuals do not converge as early as they used to (around 2000 iterations previously). the convergence criteria for the residuals are the same (1e-6). although the lift and drag coefficients do not change much after about 1500 iterations with this new mesh (the first two digits after the decimal point stay the same), Is it okay if I just take the values as they are after 2500 iterations or do I need to iterate some more?

also, what number of iterations is good?


rwoolhou posted this 07 May 2019

It depends......  As you refine the mesh you're going to pick up more details in the flow field. In your case you may have found a small transient: what does the residual graph look like?  

What we want in a solution is convergence. Depending on the monitors, mass conservation etc you may be able to take data now. Without looking at the model (which I can't!) I can only given general guidance. How does the lift/drag vary? Ie please plot against iteration number. 

Solutions can converge in 100 iterations, or can take 50k and up. It all comes down to what you're modelling, mesh and settings.