Convergence - scaled vs unscaled residuals
I am running an unsteady case with the Discrete Phase Model and injections at different timings. I also have a chemical reaction occuring some time after a particle has been injected. I use the segregated solver since I have had very much trouble getting the coupled solver to work at all (due to divergence).
The thing is, as things starts to happen in the system, residuals build up and more iterations are needed in the timesteps around that point (which is expected and OK). After that, however, the scaled residuals for continuity and one of my species starts to increase slowly. When examining the unscaled residuals, I can see that they are in fact constant. My interpretation is that as less things happen, the initial solution is more and more OK for each timestep, causing the scaled residuals to increase while the unscaled residuals actually are the same.
The problem is that this prevents me from judging convergence from the scaled residuals (as is standard in FLUENT). I plan to re-run my calculation and monitor and check unscaled residuals instead. I would just like some input on this. Am I moving in the right direction? Or have I misunderstood something? Does this sound like a good thing to do?
All comments are very welcome and appreciated!
Re: Convergence - scaled vs unscaled residuals
I just realized that while it is true that the scaled residual of continuity is compared to the worst/largest residual during the five first iterations, the scaled residual of any species mass fraction seem to be compared to a scaling factor representative of the flow rate of the species through the domain. This would imply that my reasoning is correct for continuity but not for mass fraction residuals, as I see it? Are there other clever ways of examining the mass fraction residuals? Normalization??
|All times are GMT -4. The time now is 10:52.|