open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I have problems with the number of iterations of GRACJG (BIEF) for the Tracer (temperature, modeled as a source term). With an ACCURACY FOR DIFFUSION OF TRACERS of 1.0E-6 I get for example
GRACJG (BIEF) : 0 ITERATIONS, PRECISION RELATIVE: 0.2440890E-09.

If I increase the accuracy to 1.0E-10 or even 1.0E-20, I get
GRACJG (BIEF) : 1 ITERATIONS, PRECISION RELATIVE: 0.000000.

Do I have to worry about the results if it does only one iteration?

I used the following settings:
TYPE DES SOURCES =1
FORME DE LA CONVECTION =14;5;14;1 /TYPE OF ADVECTION
SOLVEUR POUR LA DIFFUSION DES TRACEURS =1 /SOLVER FOR DIFFUSION OF TRACERS
SOLVEUR =3 / 1 tested as well
COEFFICIENT D'IMPLICITATION DES TRACEURS = 0.6 /IMPLICITATION COEFFICIENT OF TRACERS
COEFFICIENT DE DIFFUSION DES TRACEURS = 1.E-6 /COEFFICIENT FOR DIFFUSION OF TRACERS
COEFFICIENT DE DIFFUSION DES VITESSES = 1.0E-4
PRECISION DU SOLVEUR =1.E-4
PAS DE TEMPS = 1.0 (means a Courant number of 0.7; I tested dt 2s as well with a Courant number of 1.4)

If I change the TYPE OF ADVECTION for the Tracer to 1 or 5 then I get
GRACJG (BIEF) : MAX D'ITERATIONS ATTEINT: 5000 PRECISION RELATIVE: NaN.

Thanks in advance,
Beate

Hello

The fact that you have only one iteration is, probably, induced by:
1-your initial conditions are close to the solution of the problem
2-the accuracy you asked for the solver of tracer is not very restrictive
Therefore, there is no problem with that.
Usually we recommand the use of better accuracy(typically 10^-14). For "smooth" problems, with reasonable time steps, this is not expensive in CPU time.
A final remark, the use of scheme 14 is recommanded for problems with wetting&drying sequences. If, you do not have dry zones, you can use scheme 4 (or 5) for tracer. Scheme 1 (methods of characteristics) is not recommanded since it is not mass conservative. However, scheme 5 should work well only for problems without drying zones (which could explain the NaN).

I hope that this helps

with my best regards

Riadh ATA

Hello,

thanks a lot for your reply and your explanations!
As you supposed, my initial conditions are close to the solution of the problem. Unfortunately, they have to remain.
Increasing the accuracy (PRECISION POUR LA DIFFUSION DES TRACEURS) up to 1.E-40 does not change anything. I just get one iteration with a precision of 0.
Whereas the iteration of the advection (Tracer) seems to be alright, here I get
CVTRVF_POS (SCHEMA 13 OU 14) : 10 ITERATIONS.

There are still two points I do not understand properly:
1. If I get it right, it is no problem if the calculation of the diffusion of the tracer needs only one iteration. Is it also no problem if it needs 0 iterations?
In cases with less difference of the temperatures, I do get 0 iterations instead of 1 iteration. Is there another possibility to increase the iterations?
2. I was told before (by J.M. Hervouet, discussing a different case: “weir: free surface upstream weir lower than downstream”) that for the solver it is important not to get 0 iterations. Is there a difference what is acceptable for the solver and what is acceptable for the solver of the tracer?

Thanks in advance,
Beate

Hello

It is useless to decrease the accuracy up to 1E-40, try to stand close to the machine precision (of 1E-15). In this range of accuracy, convergence could be spoiled by the numerical noises.
Yes I agree that it is a bit problematic if there is 0 iteration, but only in some specific cases. Actually, 0 iteration means that solution at time tn does not evolve during time step DT, moreover, nothing in the model itself (hydrodynamic varibales, physical and numerical parmeters, boundary conditions, etc). But, even in these conditions, to obtain 0 iteration of the solver, you need to have very small time step and a very loose accuracy for the solver.
I don't know if you are in this very particular situation or not, but, you can increase the time step gradually (by multiplying it by 2 for instance) and observe the convergence of the solver. If after several attemps, the number of iterations is unchanged, then there is a problem and in that case we can look at your case more in depth.

With my best regards

Riadh ATA

Hello,

Thank you for your quick answer!

As you advised, I increased the time step, beginning with dt=1s up to 8s. The iterations of the tracer diffusion remain with 1 iteration (ETAPE DE DIFFUSION DES TRACEURS). Instead, the iterations of the ETAPE DE DIFFUSION-PROPAGATION rise with increasing time step. For dt=8s it needs actually 100 iterations, whereas the one for the tracer still says 1 iteration. (The accuracy of the tracer was set to 1.E-14.)

I attach the steering file and the log-file of the calculation with dt=1s. I prefer this time step, because here I get a Courant number of 0.7. I suppose one of my tracer settings is set false. In general, the results of the surface and the tracer are plausible; I can not find any areas of big mistakes.

Thanks in advance,
Beate

Hello

I can not detect any obvious incoherence. Thus, go on with your simulation and let us know if your results are physically acceptable or not.

With my best regards
Riadh

Hello Riadh,

Thanks a lot for your help!

Best regards,
Beate