Re: [boost]: Hypergraph

From: Francois Faure (Francois.Faure_at_[hidden])
Date: 2001-06-22 03:50:32

Hi,
I am currently using BGL for mass-spring networks simulations. It works
fine.
I would like to extend the approach to finite element computations.
Has anyone already done something like this ?
- vertices represent mesh nodes
- edges store numerical relations between nodes (for matrix
computations)
- cliques would represent elements

So, my problem is to represent cliques efficiently for this application.
I have thought about an external container storing lists of
vertex_descriptors (the cliques) cross-linked with node properties
giving access to the relevant cliques.

The idea of "hypernodes" is interesting but handling two kinds of nodes
within the same graph is complex. For example, I expect vertices() to
return a sequence of mesh nodes, not a mixture of mesh nodes and
hypernodes. Similarly for edges.

Maybe the hypernodes should better be stored in a separate graph. In
this graph, a node would store a list of vertex_descriptors of the other
graph. Edges would represent element adjacency.

Additionally, I am interested in identifying disjoint sets because they
allow handling the associated equation systems separately, thus more
efficiently.

  FF.

-- 
Francois Faure, iMAGIS-GRAVIR/IMAG (CNRS,INRIA,INPG,UJF)
655 av. de l'Europe, 38330 Montbonnot, France
Francois.Faure_at_[hidden]  Tel [phone]  Fax [phone]
http://www-imagis.imag.fr/~Francois.Faure