[make]

I was reading the threads dealing with the underrelaxation factor, but I'm still wondering.

If I call lets say p.relax() while p is the geometric field of the pressure scalar then it is explicitly relaxing the geometric field by p=alpha*p + (1-alpha)*pold assumed I called p.storePrevIter()?

But if I call UEqn.relax() and UEqn is a fvMatrix then I'm setting the alpha for the main diagonal of the solution matrix means the underrelaxation for the linear solution algorithm?

So what against underrelaxing the linear solution in the inner loop of a transient run?

Like Hrvoje already wrote using SIMPLE for a transient run even needs underrelaxation.

Do I miss a point?

Matthias

[mathieu]

Hi Matthias,

" So what against underrelaxing the linear solution in the inner loop of a transient run? "

Underrelaxation is used to accelerate convergence in an iterative loop. In transient solutions, you have to choices (in pseudo code) :

1. Iterative procedure :

for all timesteps
{
do while convergence
{
solve momentum
correct pressure and momentun (SIMPLE, PISO, etc.)
}
}


2. Non iterative procedure :

for all timesteps
{
solve momentum
correct pressure and momentun (PISO, etc.)
}

Transient solvers from the standard OpenFOAM package use the non iterative procedure which require a smaller timestep to respect the CFL condition. The non iterative procedure also require a more accurate pressure-velocity coupling (like PISO). If you use relaxation in a non iterative procedure, you will change the solution itself. Hope this helps.

Mathieu

[make]

Hello Mathieu,
thanks for your reply. What you mentioned is what I understand as a relaxation of the outer loop.
Lets say I call UEqn.solve() then it is starting the linear solver till a certain convergence criteria is reached. After that pEqn.solve() is called and so on in a segregated manner.
So I don't want to underelax between this segregated steps I want to change the relaxation for the linear solver step, so during the linear solver is running.
Isn't that simply influencing the number of iteration steps I need to reach the convergence criteria of the linear solver, but the result of the linear solver keeps the same??

Did I understand the way OF underelaxes wrong?

Matthias

[mathieu]

Hi Matthias,

I'm not an expert with the code, but as far as I understand, if you want to relax the linear solver step then use UEqn.relax() (see that post : http://www.cfd-online.com/OpenFOAM_D...tml?1201846208 ). Regards,

Mathieu