open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

When I ran 3d model without salinity for 4 days, it worked well and took 15 hours. However, when I add the tracer into the cas, the model ran very slowly, it took 10 hours to ran just 300 steps(TIME STEP=30.), and it is running still.

Another question. If I add the tracer into the cas and set VERTICAL/HORIZONTAL TURBULENCE MODEL=3, whatever the other keywords I set,the model will erupt after maybe 71 steps,2886 steps...
It seems that the tracer can't get along well with k-epsilon model.
Who can tell me the reason?

With best regards,
Liwei

Hello,

When you add the salinity it is an active tracer which triggers buoyancy effects, then the computation is more difficult. An important point is the initialisation of salinity. If you start from no salinity and enter it through the boundaries, it may create large currents and vertical velocities, maybe instabilities. Our best results with stratifications are obtained with the k-epsilon model (case of the Berre lake in France for example, with creation of stratifications and wind which destroy them). Depending on your mesh size, it could be that the time step of 30 s is too large.

In view of your listing, I confirm that the k-epsilon model does work. However there is 0 iteration for the diffusion of epsilon, it means that either you have converged to a steady state or that you do not have enough accuracy. You should then try : ACCURACY FOR DIFFUSION OF K-EPSILON : 1.E-8
Default value is 1.E-6.

The message on the preconditioning 17 is a mystery for me, as you have no such value in your steering file, as you ask only preconditioning 2 for all variables. Did you already change this ? There is no preconditioning 17 or 34 as default value, to my knowledge, so this would be to clarify. This impossibility of preconditioning 17 is probably due to dry zones ?

With best regards,

Jean-Michel Hervouet

Hello Jean-Michel Hervouet,

Thanks for your detailed explanations.Sorry for misleading you about the message on the preconditioning 17, attached are two diffferent exmaples, exmaple_1 with preconditioning 2 and example_2 with preconditioning 34.And both the example_1 and example_2 have bad results.

My 3d model start with the depth-averaged salinity according to the historical salinity data (the salinity is "warm start").And I have set ACCURACY FOR DIFFUSION OF K-EPSILON=1.E-15

There are dry zones in my modeled domain, and I set:
SCHEME FOR ADVECTION OF TRACERS =14
SCHEME FOR ADVECTION OF VELOCITIES =14
SCHEME FOR ADVECTION OF K-EPSILON =14

With best regards,
Liwei

Hello Liwei,

We would need all the files to run your case, both for the problem of parallelism and to see why the free surface goes to infinity, which is certainly a numerical instability.

With best regards,

Jean-Michel Hervouet

Hello Jean-Michel Hervouet,

Thanks for your help. Attached are all of my files.

With best regards,
Liwei

Hello,

I see no geometry file...

Regards,

JMH

Hi Jean-Michel Hervouet,

Sorry for missing it.

Hello,

I tested your case. The combination of density law and tidal flats crashes when the submerged walls in your domain become dry, at the beginning of the computation. It seems that you can go further with a time step of 15 s instead of 30 s, so reducing the time step could be a solution, however I had no time to run the complete case.

I noticed that during the crash the salinity also goes to infinity, which I do not like. I'll look at that, but this may be due to the diffusion on a very distorted mesh when the free surface is itself going to infinity.

I tried with preconditionings set to 2, and it works in parallel.

With best regards,

Jean-Michel Hervouet

Hi Jean-Michel Hervouet,

Thanks for your help. When I reduce the time step, my model works.But I have another question, my parallel mode doesn't work, because of the following error.

With best regards,
Liwei

Hello,

Parallelism works here with your model, so there is no mistake in your Fortran, it is certainly a problem of configuration of mpi, but sorry I am not a specialist of mpi.

With best regards,

Jean-Michel Hervouet