Re: [Gnucap-devel] Interesting paper in IEEE Transactions on CAD

[al davis]

On Friday 28 July 2006 12:27, walter steffè wrote:
> I have not well understood the details of this post but I am
> very interested in the topic from a user perspective.
>
> I think that you (and the cited paper) are dealing with the
> possibility of perform a fast resimulation of the circuit
> after having made a change of a circuit part (trough a
> partial update of the LU decomposition). 

No...

The issue is accelerating the simulation of mostly linear 
circuits.  It may give you the impression that it deals 
with "fast resimulation" because a transient analysis is really 
a sequence of single time simulations.  The classic Spice 
algorithm repeats the whole solution every iteration, with 
several iterations per time step.  The paper, and the gnucap 
algorithm, speeds things up by recognizing things that do not 
change between iterations, and not doing them again.

> I am interested in 
> this capability because I would like to implement a
> design/optimization strategy for microwave components which
> is based on a Electromagnetic software I am developing. This
> software will produce an equivalent circuit that is composed
> of many subcircuits describing the different parts
> (subvolumes) of the component.
> The optimization process typically changes only a  few of the
> subcircuits but the variations are not smooth because of the
> casualty of  some algorithms (like remeshing) and because the
> equivalent circuits are not unique. So I can not make use of
> the sensitivity versus a small variation of the discrete
> parameters and I am wondering if it is possible to apply the
> partial update of the LU decomposition after having replaced
> a subcircuit with a new one (which can also have a different
> number of elements). 

Perhaps ..  Look at m_matrix.h.  It is a 
sparse, "bump-and-spike" solver that allows incremental 
changes.

> I have seen that gnucap offers some 
> commands which allow to change the value of a discrete
> component but is it possible to replace an entire subcircuit
> ?

You can change anything.