open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi,

I am running a t3d case with a hot water point source.
the results show an unphysical spatial distribution of the tracer (see image file)
any thoughts? the .cas file is also attached (in zip)


cheers,

now with the files

Hi,

Most probably it is irrelevant to your problem but in your CAS file

ORIGINAL DATE OF TIME : 2010;19;09

looks strange. The manual says:

Lastly, the two keywords ORIGINAL DATE OF TIME(formatYYYY;MM;DD)

Best regards,

Hello,

Which release do you use?
How many triangles do you have in the 2D mesh?

Have you tried to decrease the time step and/or the number of cores used for your computation?

Why do you use HYDROSTATIC INCONSISTENCY FILTER = YES?

Why do not you use TREATMENT OF NEGATIVE DEPTHS = 2 to always ensure positive depths? In that case, you have to set MASS-LUMPING FOR DEPTH = 1.

To accelerate and possibly facilitate your computation, you can use:
PRECONDITIONING FOR DIFFUSION OF TRACERS = 34 as well as PRECONDITIONING FOR DIFFUSION OF K-EPSILON like for velocities.

You can try to use characteristics for the advection of velocities and k-epsilon:
SCHEME FOR ADVECTION OF VELOCITIES = 1
SCHEME FOR ADVECTION OF K-EPSILON = 1
SCHEME OPTION FOR ADVECTION OF VELOCITIES = 1
SCHEME OPTION FOR ADVECTION OF K-EPSILON = 1

FYI, if using:
- k-epsilon turbulence model, DAMPING FUNCTION is useless (only for mixing lenth models over the vertical),
- DENSITY LAW = 1, STANDARD VALUES FOR TRACERS is useless (only for DENSITY LAW = 4).

Hope this helps,

Chi-Tuan

Thanks Chi-Tuan,

I will give it a go.
I was wondering why you think it has something to do with the number of cores? (I was using 120 cores for a 338810 2D elements x 22 layers). I have a 72-hour limit on the campus cluster, so I have to speed up the computation somehow.

Cheers,

If you run your computations and the number of prisms is e.g. less than 1,000 per subdomain, the parallel run may be not efficient and also be wrong sometimes.

In your case, you are far from this case. If it is possible (and depending on the cluster you use), my classical estimation of good number of cores is 10,000 elements (triangles in 2D, prisms in 3D) per subdomain/core, sometimes a little bit less (e.g. 5,000). If possible, I would try 720 cores or more with your set up (338,810 2D elements x 22 layers).

Hope this helps,

Chi-Tuan

I agree about the 10K/core. I ran some tests a while ago and got that number roughly.
Unfortunately, I have little chance of getting 720 cores without waiting a week in the queue.