open TELEMAC-MASCARET The mathematically superior suite of solvers

My computational domain is the entire Northwestern European Continental Shelf and the mesh resolution is increased close to Belgian Coast and in the Scheldt estuary (Netherlands, Belgium). The main forcing is the tide imposed at the shelf break using TPXO data.

At the beginning, I was only focusing on the velocity field and I didn't notice the big wiggles on the water elevation (see result_1.png). After some reading on the forum, I understood that it should be due to the inf-sup (LBB) condition and that I had to decrease the value of the FREE SURFACE GRADIENT COMPATIBILITY.

One mention that it breaks the compatibility between depth and velocity. What are the consequences of that? In terms of mass conservation for example...

Also, I noticed that increasing the values of the IMPLICITATION FOR DEPTH and the IMPLICITATION FOR VELOCITY also helped.

My best results are obtained with (see result_2.png):

IMPLICITATION FOR DEPTH : 1
IMPLICITATION FOR VELOCITY : 1
TREATMENT OF THE LINEAR SYSTEM : 2
FREE SURFACE GRADIENT COMPATIBILITY : 0.5

But I fear that I have introduced too much numerical diffusion in my system. I know that I have to calibrate those parameters regarding measurements I have in the area of interest, but I would be interested in advices of people that would have faced the same kind of problems...

(sorry mesh file too heavy to post)

Hello
This is right, these wiggles could be caused by the pseudo inf-sup (LBB) condition (this condition is identified mathematically for the Navier Stokes equations but they are not established for shallow water equations, but we believe that it have a similar effect). Therefore, to overcome the problem, the recommended solution is the following:
- rigourously, to avoid the LBB consequences, you need to use different discretisations for water depth (H) and velocity (U,V). This can be achieved easily by using the key-word DISCRETIZATION IN SPACE =12;11
- In this case, it is recommande to use the wave equation (TREATMENT OF LINEAR SYSTEM =2 ) with a FREE SURFACE COMPATIBILITY =0.9

This is right also that the use of a fully implicit coefficients (=1) and the use of mass lumping gives more stable results because with these options you introduce more numerical diffusion. Thus, you can test the change in space discretization alone for a purpose of stabilization. This could be sufficient (but more expensive in CPU time).

I hope that this helps.
With my best regards

Riadh ATA