About residuals, and "bounding epsilon" of executing log-files
 

Dear Sir,

Do calculation presision in case the residuals are fluctuating (they are decreasing in the whole calculated time) inferior to that in case they are decreasing monotonously?
(See the attatched pictures)

And in my some calculation, "bounding epsilon" and/or "bounding k" occur in the log file. Should I handle for them not to occur (even when residuals are decreasing), even though executed calculations are finished by satisfying the convergence conditions?

In the first place, what do "bounding ***"s mean?



Sincerely,

KeiJun

Sorry, I don't understand your first question, only the second ;)
"Bounding *** error" means that the solver calculated values that are forbidden by the code, such as unphysical values. "k" and "epsilon" have to be larger than zero but sometimes numerical mess leads to values lower than zero. Now the solver realizes it and manually sets them to zero to get a physically meaningfull value. This also holds for values of "k" and "espilon" that are much too high.

Often these "k" / "omega" / "epsilon" bounding errors are produced by the laplacian scheme. I guess you have set the laplacian schemes to "Gauss linear corrected", right? Try to set them to "Gauss linear uncorreted" and see if the errors vanish.

Thank you for your thoughtful reply, and sorry for my poor English.:(

For the first question, I attached the pictures in which I wrote explanation to express my question.


Then, are calculation results in which "bounding *** error" sometime occurs imprecise and unreliable? Because k and/or epsilon values of such results are too big.

I am using "Gauss linear corrected" as you said, because my mesh quarity is not so good.
But I will try "uncorrected".


Sincerely,:):)

1) If the residuals fluctuate on their way down but the final residuals are stable, everything is fine.
2) But if the final residuals are fluctuating as much as in your pictures it could have different reasons. Maybe the case is unsteady and the steady solver can't get a solution. Or the mesh is bad, or the numerical settings are bad... I would try to run the case with safer numerical settings or a better mesh and see if that helps.
3) Yes, if you get bounding errors something is wrong with the calculation. In my experience a better mesh helps in most of the times.

What are demerits not to use "Gauss linear corrected" but to use "Gauss linear uncorrected"?

When I use "Gauss linear uncorrected" not for occuring this error, are there something for me to have to sacrifice?

"corrected" is more accurat but also less stable.
You can read in the PhD thesis of Hrvoje Jasak (google...).

Dear Sir,

I tried to calculate using "Gauss linear uncorrected", but the result (u, p, k, ε) didn't varied, despite "bounding epsilon" didn't occur.

In the mean while, I tried to do with the relaxation factor small, which lead to decrease the number of occuring of "bounding epsilon", then the result (u, p, k, ε) varied.

For the attatched pictures, the meaning of "Cond1" and "Cond2" are "large relaxation factor" and "small relaxation factor", respectively.
The result using "Gauss linear uncorrected" is the same to "Gauss linear corrected" (Cond1).

Why does relaxation factor cause the different result?

I don't know what's what.:confused::confused::confused:
What should I think about it?

Please help me!!!!:(:(:(


KeiJun

The relaxation factors don't change the result, if the solution converges in both cases. So if your result is different by changing the relaxation factors, one of the solutions did not converge.