Hello!

I have a question concerning which are the real standard convergence limits in FLUENT 6.1. I think the manual contradicts itself on this issue...?

The manual says in one place that it is 10^-3 for all residuals except energy and P-1, for which it is 10^-6. Then, later on the same page, it says that using normalized instead of scaled residuals (which, according to another chapter, is the same thing if not changing the normalization factor or the number of iterations to look for maximum residual) it is simply 10^-3 (i.e. not different for energy and P-1).

And, when trying to reach a steady-state solution with the coupled implicit solver for a simple fluid flow problem, I see that the residuals for continuity, energy and species follow each other closely. This means that I will have to wait a long time if I want energy to go below 10^-6. If solving the same problem with the segregated solver instead, the energy residual drops very quickly.

So - what is the standard value for convergence for the scaled energy residual in FLUENT 6.1 when using the coupled implicit solver?

Thanks in advance!

/Henrik

Now I see that it seems 10^-3 is most appropriate, at least if using the normalized residuals:

"If you have provided a very good initial guess, the residuals may not drop three orders of magnitude. In a nearly-isothermal flow, for example, energy residuals may not drop three orders if the initial guess of temperature is very close to the final solution." (from the 6.1 manual)

/Henrik

You need to monitor more than just your residuals to judge convergence! You need to monitor what ever it is that you're trying to get out of the model. If you're trying to get body forces, then monitor the body forces, if you're trying to get temperatures, then monitor temperatures, or pressures, or velocities, etc...

I have a case that ran last night and came back this morning and Fluent said it was "converged" based on the residuals, but I have a monitor of the body forces and it's in no way converged (I was in a rush to set it up and go home last night... forgot to change the convergence criteria).

There are lots of monitor options (use surface monitors to get flow properties, like max/min/avg temp, pressure, velocity, etc... use force monitors for body forces like lift, drag, pitching moment).

Any value you plan on reporting, you should be monitoring. Residuals are a good guide to help you judge if the solution is converging, but it should not be your judge for actual convergence. Its way too risky to assume it's converged based on residuals.

Good luck, Jason

Thank you for your answer!

Yes, indeed it is very important to monitor whatever it is you intend to calculate from your simulation.

My problem is, however, that I intend to run an unsteady simulation. I will not be around to judge convergence for each time step. I will of course check the results very carefully, but I need to set some critera to allow the time stepping to go on unattended... Which makes things a bit more complicated than a steady-state calculation, for example.

Besides, I do not really know what quantity to monitor during time stepping in my model... I simulate a spray into a gas flow, along with some chemical reactions... Any ideas?

Thanks again,

Henrik