Momentum source coefficient and convergence
 

Hi

I'm applying momentum source coefficients over a sub domain which are dependent on velocity. Without a momentum source coefficient I was not getting convergence. I've found that a momentum source coefficient of -100000 allows the model to converge, and the results are accurate. I did this based on what the manual says:

General Momentum Source
Momentum sources can be specified directly in terms of a momentum value per unit volume in a specified direction. To obtain good convergence when the source is a function of velocity, the source should be linearized by including a Momentum Source Coefficient.

However I don't understand why this works - can anyone explain? Or suggest a reference?

Why does the momentum source coefficient improve convergence?

Thanks, Matt

This required by the specific solver technology of CFX. I guess it is describbed in the manual how to linearize. If not you have to ask the service.

This is a complex question. I will have to refer you to CFD text books like "Introduction to Computational FLuid Dynamics" by Versteeg and Malalasekera.

Hi, ghorrocks.

I have 3 CEL souece terms in 3 directions.

How could I define 3 Mom. source term coefficients in diffferent Directions?

It seems that CFX can only define one Mom. source term coefficient.

Thank you!

:)

Define your own general momentum source, and then you can define any function you like, and can define a different function for X, Y or Z directions.

Sorry, I'm fail to present the problem.

I need to define 3 Mom. Source Coefficients, not 3 Mom. source terms (In fact, I have done it).

such as, partial Sx/partial x, partial Sy/partial y, partial Sz/partial z.

However, it seems can define only one Mom. Source Coefficient, not 3 components.

Yes, that is what I understood. The momentum source coefficient just gets turned into a source term and applied as a source term. So I suggesting you write the source term yourself directly using the functions described in the CFX documentation. Then you have the ability to define a different source term in each direction.

The source coefficient has no impact on the final solution. So, you are free to do as you will.

The advice to obtain a decent source coefficient is to linearize your source term with respect to the variable you are solving for, i.e. dS/dVelocity

But, because the variable is a vector, you end up with a tensor of source coefficients; therefore, your assumption that only 3 coefficients are needed is also incomplete unless you know the off-diagonals of that tensor are exactly 0.

ANSYS CFX only exposes diagonal linearization, i.e. same source coefficient for all equations of the vector variable being solved.

What do you do? Anything is possible, but you can try different approaches:
1 - Max of the diagonal in the tensor
2 - Trace of the diagonal in the tensor if all the diagonal are of the same sign
3 - Some kind of norm for the matrix
4 - One of the above times some heuristic value

Thanks ghortocks and Opaque.

As Opaque pointed out, the Mom. source Coef. should be a tensor (not a vector).

The source coefficient has no impact on the final solution.

Thus, CEL is usually unnecessary.
Giving a negative number is a simple and effective method.

Though it will work, it may not be efficient to use a single uniform negative source coefficient.

For optimal convergence, we want a dynamic negative source coefficient only where is needed, and only the mathematics/physics knows when it is needed.

Summary: an appropriate problem-dependent negative value will help you converge the setup for this specific flow conditions while the correct negative expression will consistently help you converge the problem w/o further baby-sitting/monitoring the treatment of this specific source