open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I like to compare some of the Finite Volume Schemes (HLLC and WAF) with the standard schemes. I figured out how to run both new schemes, but have now some questions

• I set the Courant number (HLLC and WAF scheme) to 0.5 since I have not fully understood the numeric behind the schemes. What are the maximum/optimum Courant numbers?
• I use the option "GRAPHIC PRINTOUT PERIOD". Because of the variable time-step it is hard to compare the schemes. Is it possible to adopt the Courant number near the printout times to get fixed printout times?
• Is it possible to simulate sediment transport (Sisyphe) in combination with these schemes?
• Exist any validation cases for these schemes?

I would be glad for any advice.

Best regards,
Leo

Hello Leo
Sorry for this very late reply, I was not aware about your post because I was not assiduous on the forum these last weeks.
If you are still interested by the answers to your questions, these are some elements:
1- The Courant Number is a real between 0 and 1 and it is used to ensure the stability of the scheme. The bigger the Courant number is, the larger is the time step. Thus, you can choose a value of 0.9 which usually gives nice results.
2- The use of a CFL number implies a variable time step. Thus, the use of GRAPHIC PRINTOUT PERIOD will give you result outputs of different times. There is no option for a fixed time step so far, however, this option will be in the next release. What you can do to right now, is to edit the subroutine caldt.f or just to comment the line "call caldt" in the subroutine resolu.f.
Be carefull, the use of a fixed time step could cause unstabilities, it is your full responsability.
3- The new validation document of Telemac-2d includes several cases with the option finite volume (using HLLC or WAF).

I hope that this still helps

With my best regards

Riadh ATA

Thank you Riadh!

I've tested the different schemes and used a Courant number of 0.8, a higher Courant number like 0.9 will increase the speed so I will increase my Courant number.

I think the modification in caldt/resolu.f is not necessary. I realized that the time step for explicit models is in general so small that it is not a real problem to compare the results with an implicit model. Only for the dambreak problem it was interesting. Since I have the analytical solution (thank you ) it is no longer a problem.

I've played a little bit with the HLLC scheme and added two openMP statements to increase the speed at the loop over all edges.


With more than two cores I get an speed-up. But I have to checke if the results are equal.


Best regards,

Leo

Hello Leo

It could be very interesting to have more informations about the speed up using OMP and to compare it with the MPI based parallelisation (that I'm achieving these days).
You will be able to use the parallel version next august with the new release of Telemac.

With my best regards

Riadh

I played a little bit with openmp but I realized that it is quite tricky to get performance and accuracy. Even when I used !$omp critical for the shared data (e.g. CE) there are large differences compared to the serial version of my problem.