open TELEMAC-MASCARET The mathematically superior suite of solvers

Issues

When non convergence occurs between list outputs, all we see is that non conergence occurred and which iterative routine it was in. Is it possible to also pass through a text tag that indicates if it is tracer diffusion, or k-epsilon or what ever. The same goes for the advective routines. It is very difficult to determine what phase of the timestep calculation causes the error.

In some places, eg advective (eg murd3d) error messages are printed which include the 3D cell index, which is effectively useless. In 3D it would be much better to break this down into the x,y,plane or at the least the global 2d index and plane. This is because some of these errors are due to poor geometry and better location info helps check that out.

Can the error printing method when spread across multiple processors, transfer the major error messages to the master processor so that the messages appear on the listing that comes to the screen. Having to go through each processor listing to find the error is just another hurdle to jump.

Additions and Fixes for these would make the system easier to setup and manage.

Cheers
john

Hello John

In deed your first remark is ineteresting. It can be easy to add a clear message that indicates which part of the algorithm is not converging. I'll check that soon.
However, the maximum iteration is reached generally when the time step is big and this is a clear message to decrease it; whatever the part of the algorithm is concerned.
For your last remark, you can find easily the corresponding 2D node number since the numerotation starts from the bottom plane from 1 to NPOIN2 (2D mesh nodes) and then increases like j=(iplane-1)*NPOIN2+i. The unique and useful information is the global number of the node.

I hope that this helps
with my best regards

Riadh ATA

Regarding the 2d element number - specifically the grid is essentially defined in the 2d, with the generalised vertical grid. So instabilities may occur due to unusual grid shapes or local severe variations in the grid depths etc. So in these cases being able to home in on the 2d location helps the user look at what has been defined. You can also easily give the vertical grid location also.

In cases of maximum iteration type errors, the fact that you got an error and which process is important as you can change the performance by varying the numerical method, eg iterative matrix solver and preconditioning etc. So knowing the process where the problem arises allows easier checking of the computational parameters. There are more things than just reducing timesteps that can be varied.

Cheers