open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi,

I'm modeling marine submersion. I have programmed a simple tidal model in bord.f. I'm computing the model in parallel.
I have met this error and after the simulation failed :



And this is my parameter file :



Does somebody know how can I solve the problem?

Regards

Nicolas

Hello,

Actually GMRES is not suited for too easy linear systems, and will crash if given a diagonal to invert (what happens if you have no diffusion). Choose Conjugate gradient instead (1 instead of 7 for the SOLVER FOR DIFFUSION OF TRACERS).

For advection diffusion GMRES has to be used preferably with SUPG because in this case the system matrix becomes non symmetrical.

With best regards,

Jean-Michel Hervouet

Hi Jean-Michel,

I have tried GMRES (7) and conjugate gradient (1) for SOLVER and it doesn't work for both.
You are talking about SOLVER FOR DIFFUSION OF TRACERS. I haven't this parameter in my file. You advise me to take the value one for this parameter but isn't it the default value?

Regards

Nicolas

Hello,

OK let's look at it from scratch. GMRES is never given as default value in Telemac-2D, so it must be in your parameter file and you can see for which solver.

If it is just SOLVER = 7 (propagation), look the value of :

TREATMENT OF THE LINEAR SYSTEM

If it is 1 (default) it is the primitive equation option, try 2 (wave equation) with conjugate gradient (SOLVER=1)

If it is already 2, GMRES is useless and you can try the conjugate gradient (SOLVER = 1). This conjugate gradient is now programmed to return without doing nothing if the residual is 0, so it should work.


With best regards,

Jean-Michel Hervouet

Thank you Jean-Michel

I will try to modify TREATMENT OF THE LINEAR SYSTEM for my next test.
It's now running with SOLVER=3 but it's only 17 minutes simulation elapsed so I have to wait to see if it's really working.

Regards

Nicolas

Yes if you use the primitive equations options, i.e. TREATMENT OF THE LINEAR SYSTEM = 1, the conjugate gradient will not work because the system coupling h, u and v is not symmetric. This is why 3 or 7 is necessary.

Regards,

JMH

I have tried with TREATMENT OF THE LINEAR SYSTEM = 2 and SOLVER = 1 and it seems better.
It worked until I met a bathymetric problem (strong gradiant of bathymetry where it should not exist). I have modify the bathymetry to avoid the problem.
But I think I will have the same problem with dikes. I'm expected to simulate dike submersion, when the flow will come over the dikes it's the same... Is it over speeds or high number of currents that lead the simulation to failure?
How can I limit the speeds?
Which kind of values for number of currents is it reccomanded to use?

Regards

Nicolas

Hi Nicolas,

It is always not trivial to deal with dikes. The main point to success the simulation is to evoid to diecretize loosely the dyke. The better the dyke is refined, the better behaves the code.
The heigh speeds appear when very small water depths and/or partially wetted elements appear. Recently a new algorithm was implemented in Telemac-2D (V6P2) to handle efficiently such kind of elements. Thus, I encourage you to use this version if it is not already the case.
You can remote the speeds by introducing a small 'DESIRED COURANT NUMBER' comined with a 'VARIABLE TIME STEP'. However, in my opinion it is not necessary, depending on the advection scheme you are using, and if your time step is not very high, Telemac will manage internally the high speeds and the wetting and drying aspects.

I hope that this helps.
With my best regards

Riadh Ata

Hi Riadh,

I have already refined the dikes and I'm using 'DESIRED COURANT NUMBER' combined with a 'VARIABLE TIME STEP'.
When you say "a small desired courant number" that's mean which kind of value? Currently, I'm using 5 but I don't know what Telemac admits.
I'm using V6P1 and I'm waiting for the Perl version to instal V6P2.

Regards

Nicolas

Hi,

The V6P2 perl version has already been released.
Files are available on the website (the systel.ini file can be downloaded separately) or through SVN.

Fabien