How do u know that one's solution has converged or it has reached the solution?

I set my residual at 0.00001 when i commenced my simulation but i noticed that the solution is fluctating after awhile so i decided to increase the residual to 0.0001 and the solution converged immediately. This makes feel that the solution is not accurate. What other criteria can I use to know if I am to confirm my solution is ok apart from validation with experimental data?

VERY ADVICE COUNTS

Rachel, this question is repeated many times in this forum and many users have answered it. i will tell you for your benefit though. residuals are just a way to achieve numerical accuracy. a decreasing residual is an indication that computed variables per iteration are increasingly converging to a value. the default criteria that FLUENT uses for residuals are adequate in my opinion (some people may not agree on this issue). in your case, you increased burden on solver by putting too stringent a criterian on residual. this is rarely needed if at all. now, best place to know about other criteria is FLUENT manual itself. the manual neatly discusses convergence issues and suggests which definition of residuals should be followed for a particular class of problems. i suggest you should go through it once. briefly, you should monitor other integrated (averaged over surface/volume) quantities of interest to you. they generally cease to change when convergence is nearing and this can happen before or after residuals cross the criteria you provided. what you expect physically to happen is also one way of looking at convergence. if you are satisfied with what you see after post-processing, you can even accept a solution that did not converge in the sense of residual reduction.

hope this will impart some direction,

Swarup.

Hi Swar, I am sure you will be able to answer to the following precisely.

I read about the 'scaled residual' and 'normalized residue' definitions. Can I ask you to explain the difference between them?

If initial guess values are very close to the expected results, then what would be the behavior of 'scaled residue' and 'normalized residue'?

Is there any specific advantage of using one over the other? if yes, what?

Thank you very much for your time in advance. I apreciate your effort.

Regards -US

I will try to answer step by step. I think it will be easier to use residual definitions for segregated solver.

Scaled residual involves use of a flow rate indicator term as scaling factor over unscaled residual. Normalized residual will on the other hand use another residual (scaled or unscaled) that is worst/maximum upto a certain number of iterations (decided by you). If you run a case for few iterations with different residuals, you will immediately find the difference. Since at the start, only one residual is available, a normalized residual will start from 1. A scaled residual will take values depending upon relative magnitude of numerator and denominator.

when your guess is very good, for a scaled residual, disparity between LHS and RHS will be less and you should get smaller unscaled and scaled residual as seen by their definitions. Normalized residual will show you the extent of drop/increase starting from 1 always. For good guesses, the worst residual will not be much different from newer ones so you should not see much variation in normalized residuals.

there is no reason why only a single definition should be preffered but you should always think about scaled residual first. it is closer to reality than any of other definitions and easy to track. As you may have read in the manual, sometimes just the extent of drop in residuals is all that we want, this at least for an approximate analysis. then you should look at normalized residual. it needs to be monitored carefully if your solution oscillates during iterations because the number M is set by you and won't change on its own. it is a good practice to track continuity residual always, especially for incompressible flows because its definition is intimately related to physical reality.

i have found that it is confusing to see so many numbers at the same time. however, they make sense when you are able to find a trend in them and are able to establish correlation amongst them. this reveals much about accuracy of model and clarifies the physics.

as a customary caution, do not just stop with residual, see other averaged quantities also.

I hope you will find something of use here.

Swarup

Thanks Swarup. I will try that. Thanks