open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,
i am aware of the option PRECONDITIONING which does the process of accelerating the convergence while solving AX=B to determine the unknown vectors.

I would like to know the difference between the following two options
i) Diogonalized preconditioning with the condensed matrix
ii) Crout's preconditioning per element

I also wish to know the term CONDENSED matrix

Thanks in advance
Sathya

Hello,

The condensed matrix is a diagonal with all all columns summed. This preconditioning is used when the diagonal itself may be negative, but except with SUPG I do not see other cases where this could happen.

Crout preconditionning per element is for the storage of matrices = 1, i.e. EBE (Element By Element). As the edge-based storage (1) is much faster, it is hardly used now. This preconditionning consists in solving the linear system inside every triangle (assuming that other points are solved). A drawback is that the effect in parallel is not exactly the same as in scalar, so we tend to avoid it now for this lack of reproducibility.

With best regards,

Jean-Michel Hervouet