open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi everyone,

I would like to know if there is a way to control the time step when Finite Volume Method is used, i mean, i am getting a small time step and a CFL number low (0.15 the higher), so i was wondering if there is a way to get a higher CFL number and a higher time step. You will find attached my steering file.

Thanks in advance,

cheers,

José

Hi
The time step value given in the steering file is the maximum value allowed for the computation so maybe you could try to give a higher value...

Regards

Hi,

do use the keywords:

VARIABLE TIME STEP = YES
DESIRED COURANT NUMBER = 0.95

the size of the time step will depend on the CFL-Number (including the velocity and water depth, gravity). So you have to say what COURANT NUMBER / CFL-Number you allow. For first order schemes CFL between 0.9-0.95 should work.

Best regards,

Leo

Hi Christophe and Leo,

Christophe, i know the time step is the maximum value allowed in FV, and if the courant number is exceeded this one changes, but, the values of time steps im getting are around 1E-4s, despite of this, i tried higher values and nothing change.

Leo, i had used this values previously and the time step is similar, another keywords?.

Now, is the solver accuracy or the solver i am using the best choice with finite volume? If you think there are other choices which could accelerate my computation, please make me note. My mesh has about 5000 nodes.

Thanks for your support,

cheers,

José

Hi,

sorry you allready had the correct values. As far as I know, the Finite Volume Scheme works only with some friction laws (e.g. HLLC with Manning ?) and turbulence models are also not available? Maybe this has changed in the last years. So you have to check this.

The best thing is if you start with no turbulence and no friction to check if the modell will run.

O.T.
The handling of friciton terms for small water depths is allways tricky an can reduce the time step size.


I don't understand why one has to choose a solver for the FVMs? They should be explicit, so no solver is needed.


Best regards,

Leo

Hello,

Not an answer but rather a comment in our experience in this case.


A few months back when messing with the "Desired Courant" keyword I also noticed it was "way too conservative" and hence increased the computation time to 2-3 times compared to its FE equivalent with virtually no significant difference in the results obtained for our case (So we sticked to FE with a fixed time-step instead)

Which led me to think that there were a couple of nodes in my domain whose quality was sub-par and they were messing the computation forcing a time step on them that was way too conservative to the other 99.99% nodes, but If I remember correctly the value of courant shown in the listing is the maximum of the domain...

Another possibility here is that the subroutine (?) in charge of the time-step scaling isn't working as expected and its rate is way too low or its relaxation is way too big to "make" a difference in a run...

Regards,

José Díaz.

Dear Leo and José,

Thanks a lot for your comments.

José, i also noticed that FV is too conservative with the courant number and i could not change that. I already tried to use FE but the results i got were not good enough for my purposes, i believe in this case i will have to use FV despite of computation time to get the results i need to, the differences in my model between FE and FV are considerable.

Cheers,

José