Re: [femm] Re: Self and mutual inductances of coupled microstrip lines in RF frequency range

[David Meeker <dmeeker@xxxxxxxx>]

Samuel wrote:

> Thanks for this useful information, Dave.
>
> May I ask you one more question? :-) To obtain full L-matrix of
> multiple signal strips in semiconducting dielectric stack the problem
> must be solved N times where N is the number of strips, each time 1A
> net current should apply to the next strip while keeping all other
> strips at zero net current. 
> It means that we keep solving the same equation N times replacing
> only right hand sides.
>
> Have you any idea how to make your iterative solver to memorize most
> of results from the initial solution and reuse them for the following
> solutions? Direct solver does this trick when it stores results of LU
> decomposition and reuse them for multiple RHS.

I haven't thought that much about simultaneously solving for multiple 
RHS vectors with an iterative scheme. You can use the same 
preconditioner for all the RHS vectors--this would be an advantage if an 
Incomplete Choleski preconditioner was used, because some effort is 
required to obtain the decomposition. However, I'm using the SSOR 
preconditioner, which is relatively trivial to obtain. If we were 
solving a time-stepping problem, where the change from step to step were 
very small, the solution from one time step would be a good initial 
starting point for the iteration that computes the next time step. 
However, it isn't clear to me that the solutions for different 
collections of currents in your inductance problem are close to one 
another. I think that one could generalize SSOR-PCG to many RHS 
vectors, but it isn't immediately clear to be whether it would be faster 
than solving them one at a time.

Dave.

--
David Meeker
dmeeker@xxxxxxxx
http://femm.berlios.de/dmeeker