algorithm - Eliminating symmetry from graphs -
i have algorithmic problem in have derived transfer matrix between lot of states. next step exponentiate it, large, need reductions on it. contains lot of symmetry. below examples on how many nodes can eliminated simple observations.
my question whether there algorithm efficiently eliminate symmetry in digraphs, way i've done manually below.
in cases initial vector has same value nodes.
in first example see b
, c
, d
, e
receive values a
, 1 of each other. hence contain identical value, , can merge them.
in example spot, graph identical point of view of a
, b
, c
, d
. respective sidenodes, doesn't matter inner node attached. hence can reduce graph down 2 states.
update: people reasonable enough not quite sure meant "state transfer matrix". idea here is, can split combinatorial problem number of state types each n
in recurrence. matrix tell how n-1
n
.
usually interested value of 1 of states, need calculate others well, can next level. in cases however, multiple states symmetrical, meaning have same value. it's quite waste calculate of these, want reduce graph until nodes "unique".
below example of transfer matrix reduced graph in example 1.
[s_a(n)] [1 1 1] [s_a(n-1)] [s_f(n)] = [1 0 0]*[s_f(n-1)] [s_b(n)] [4 0 1] [s_b(n-1)]
any suggestions or references papers appreciated.
brendan mckay's nauty ( http://cs.anu.edu.au/~bdm/nauty/) best tool know of computing automorphisms of graphs. may expensive compute whole automorphism group of graph, might able reuse of algorithms described in mckay's paper "practical graph isomorphism" (linked nauty page).
Comments
Post a Comment