[fjtb]

Hi guys, I'm just wondering if it is required for any CFD simulation to have a smooth RMS residual plot until it reaches a certain residual criteria, say 1E-5. I'm asking this because I've encountered several simulations where the residual plot is kind of oscillating and chaotic to some extent, but it still moves downward toward my RMS residual criteria as the simulation progressed. In the said simulations, I've checked the mesh quality and I think it's good enough (all orthogonal quality > 0.10 and all skewness < 0.90). What can you say about this? Also, can you offer some mathematical/numerical/physical explanation why is this so? Thank you in advance to those who will answer!

[Gert-Jan]

No, the way convergence is achieved should not affect convergence. Either in a straight way or through a few bumps. In line, the end solution should be affected by the initial guess.

Provided you run steady state calculation, chaotic residuals indicate that the solver is searching for a stable solution which is hard to find. If the residuals keep oscillating, your solution may be transient, and the solver will never be able to find a stationary solution.
Keep in mind that residuals are useless as long as your imbalances are not wihtin 1%. So, convergence should not only be judged from residuals.

[ghorrocks]

No, the residuals do not need to be smooth for it to be good convergence. One source of bumps in the residuals convergence is startup vorticies being convected through the domain and out the outlet. As the flow develops and these vorticies get smaller the size of the kink they put in the residuals convergence gets smaller, but it sometimes does not disappear completely. But if the vorticies are small enough they do not matter, so you can have an adequately converged solution with some of these vorticies still present.