[m5edr]

what the difference in conversion when i check (or uncheck) scale box?

[LuckyTran]

Originally Posted by m5edr

what the difference in conversion when i check (or uncheck) scale box?

The "scale" option enables/disables the scaling of the residuals that are reported. If not on, the residuals are not scaled. If enabled, then the residuals are scaled.

The unscaled residuals are the sum of the imbalance (mass, momentum, energy, etc.) over all cells in the domain. The scaled residual, is the unscaled residual divided by the appropriate scaling factor.

For most variables (all but continuity), the unscaled residual (sum of imbalance) is divided by the sum of the variable. Sum of momentum imbalance divided by sum of momentum. The sum of the quantity over all cells is the scaling factor.

Continuity is treated differently. For continuity, the unscaled residual is divided by the worst residual encountered during the first five iterations.

It is explained in detail in the Fluent manual 29.15.1.1

[Far]

How you look at the unscaled residuals? What should be the criteria to judge the unscaled residuals?

[emreg]

scale is not too much important..
dont bother this.

[ghost82]

Originally Posted by emreg

scale is not too much important..
dont bother this.

Directly from Fluent's guide:

"In general, it is difficult to judge convergence by examining the residuals defined by Equation 25.18-3 since no scaling is employed. This is especially true in enclosed flows such as natural convection in a room where there is no inlet flow rate of with which to compare the residual."

"The scaled residual is a more appropriate indicator of convergence for most problems".

[ghost82]

Originally Posted by Far

How you look at the unscaled residuals? What should be the criteria to judge the unscaled residuals?

Fluent's guide says that unscaled residuals should decrease by 3 orders of magnitude, but in my opinion it is not enough to look at the residuals to judge convergence; absolute values of residuals, behaviour and monitoring of key parameters, I think these are the factors we have to look at.

Daniele

[LuckyTran]

Originally Posted by ghost82

Fluent's guide says that unscaled residuals should decrease by 3 orders of magnitude, but in my opinion it is not enough to look at the residuals to judge convergence; absolute values of residuals, behaviour and monitoring of key parameters, I think these are the factors we have to look at.

Daniele

Just to further the points made:

If the solver is initialized with the exact solution to the problem. The unscaled residuals will never decrease. The scaled residuals likewise will also not decrease. Hence searching for a reduction in three orders of magnitude of the residuals is not a very good criteria to judge convergence/accuracy.

The residuals do tell the amount of imbalance in the governing equations being solved. However, the residuals do not tell whether the solution has converged. Monitoring actual solution values is definitely needed.

The scaled residuals are suitable in my opinion for considering the amount of imbalance since the scaling factor is the sum of the variables. The scaled residuals are therefore like % imbalances and can be compared directly for different problems. The unscaled residuals can be the same for two different problems and have different meanings depending on magnitude of the solution values. Consider the laboratory flow around a cylinder versus interstellar flow around a galaxy, the difference in length and velocity scales are tremendous!

[bumper]

Hello,


i have a general question about the calculation of the scaled residual. In Fluent, there is the following defintion of it:
Please see attached picture.


Maybe someone could explain it to me becasue the documentation of Fluent doens't help me. What is calculated exactly? Is it the difference of a governing variable from the actual iteration minus the last iteration in relation to the last iteration??


Thanks for your help,


bumper

[LuckyTran]

Originally Posted by bumper

Hello,


i have a general question about the calculation of the scaled residual. In Fluent, there is the following defintion of it:
Please see attached picture.


Maybe someone could explain it to me becasue the documentation of Fluent doens't help me. What is calculated exactly? Is it the difference of a governing variable from the actual iteration minus the last iteration in relation to the last iteration??


Thanks for your help,


bumper

On the top and bottom is a summation over all cells p. Each summand is the difference difference in fluxes of the current cell and neighboring cells + sources (b). The summand in the denominator is the flux in each cell. Hence, it is the sum of all imbalances divided by the sum of all fluxes. And since the number of cells in numerator and denominator are the same, it is the average/mean imbalance divided by the avg/mean flux.

[bumper]

Okay, so to sum it up in my words:

In general, the scaled residuals compares the differences to neighboring cells in relation to all in one iteration step, right ? And with that, you can rate the convergence behaviour?

Isn't it necessary to consider the behaviour betweeen iteration steps?

[LuckyTran]

Originally Posted by bumper

Okay, so to sum it up in my words:

In general, the scaled residuals compares the differences to neighboring cells in relation to all in one iteration step, right ? And with that, you can rate the convergence behaviour?

Isn't it necessary to consider the behaviour betweeen iteration steps?

Yes it is only for 1 iteration. Residuals measure imbalance, they don't tell you anything about convergence except that a converged solution should have low residuals.



It still is necessary to compare something between iterations. What and how that is done, is up to the user to be creative.

[bumper]

Originally Posted by LuckyTran

It still is necessary to compare something between iterations. What and how that is done, is up to the user to be creative.

Do you have any recommmendations?

And can you tell me what is represented by the coefficient a in the equation of residuals? Ansys documentation says that is is a center coefficient or an influence coefficient. But what does it mean exactly?