open TELEMAC-MASCARET The mathematically superior suite of solvers

Dear,

I am working on simulations with very low diffusion coefficients, but I have some problems with the precision.

I did multiple simulations to test the effect of different values of the diffusion coefficients on my results. I don't use advection in all the cases. I imposed a factor (like the morphological factor) to speed up my simulations. Thereby I multiplied the diffusion coëfficiënt with a factor and divided the number of timesteps with the same factor. However, there is no diffusion in the simulations with a low diffusion coëfficiënt (around 10^-10). While there is diffusion when I impose this factor. I think this has to do with the precision in Fortran.

How can I increase the relative precision of tracers? I already tried to change EPS in the subroutine Difsou, but this did not changed the precision.

Best regards,
Veerle

Hello Veerle,

It is the accuracy of linear solvers that should ne increased, otherwise they "think" that the problem is already solved. So try:

ACCURACY FOR DIFFUSION OF TRACERS : 1.D-12 to 1.D-15

and see what happens. However a diffusion coefficient of 1.D-10 will probably do nothing... are you sure it is the right thing to do, it is not what we do with a morphological factor : the time is reduced by a factor and the evolution is multiplied by the same factor.

With best regards,

Jean-Michel Hervouet

Dear Jean-Michel,

Thanks for the fast answer, I will change the ACCURACY FOR DIFFUSION OF TRACERS.

Is it not correct to multiply the diffusion coefficient and divide the number of time steps by a fixed factor, in order to speed up the simulations? I want to speed up a long simulation (of about 120 days). The system is a quite static river with a constant discharge and constant water level which don't change over time. And the diffusion of the tracers is very slow.

Are there possibilities to speed up this kind of simulations?

Best regards,
Veerle

Hello,

Yes it might work, but you mention a diffusion of 1.D-10 and I thought you were dividing the diffusion. OK I understand now that you start with a very low coefficient and then multiply it by a large number. Yes, a coefficient of 1.D-10 could be challenging the machine accuracy. Anyway your technique seems better than multiplying the evolution, because monotonicity will be kept.

Regards,

JMH