A graph is perfect if the vertices of any induced subgraph H can be
colored, in such a way that no two adjacent vertices receive the same color,
with a number of colors (denoted by
(H)) not exceeding the cardinality
(H) of a maximum clique of H. This definition leads to two
interesting problems : to determine which graphs are perfect and to produce
an optimal coloring of such perfect graphs.
Notions not defined below may be found in [2]. A graph is
minimal imperfect if all its proper induced subgraphs are perfect but it
is not. It is an easy task to check that an odd chordless cycle (usually
called a hole) of length at least five, as well as its complement
(usually called an anti-hole), are minimal imperfect graphs. A graph G
is said to be Berge if neither G nor
contain an odd hole
and Claude Berge conjectured that a graph is perfect if, and only if, it is a
Berge graph (Strong Perfect Graph Conjecture, SPGC [3]). Berge also
conjectured that a graph is perfect if, and only if, its complement graph is
pefect [3]. This last conjecture is an easy consequence of the
following theorem due to Lovász [11] :
One can deduce, from this theorem, that in a minimal imperfect graph
G = (V, E), with
(G) =
and
(G) =
, one has (see
Lovász [11], Padberg [16]) :
Bland, Huang and Trotter [4] defined a graph to be partitionable (or
(
,
)-partitionable) if there exist two
integers
,
2 such that (1) and (2) above hold.
For a graph G, we denote by NG(x) the set of vertices of G adjacent to vertex x of G and by
G(x) the neighbourhood graph of x induced by NG(x); when there can be no confusion, we shall write
N(x) = NG(x).
As usual, for any graph G = (V, E), we will note n = | V| and m = | E|.
In a minimal imperfect graph, the neighbourhood of any vertex has some
interesting properties :
It is interesting to notice that it is not yet known if
(v) must
be connected for any vertex v of a minimal imperfect Berge graph (see Lubiw
[12]). In [1] we have proposed a conjecture, weaker than Lubiw's
one, dealing with neighbouhood's structural properties in a minimal imperfect
Berge graph. In order to formulate this conjecture, we need introduce some
definitions.
We call complete join of two (vertex disjoint) graphs
A = (VA, EA) and
B = (VB, EB), the graph with the vertex set
VA
VB and the edge set
EA
EB
{ab| a
VA, b
VB}. Let
be the family of
bipartite graphs (possibly reduced to a single vertex) and let
*
be the family, containing
, such that for any two graphs G1
and G2 in
*, the complete join and the disjoint union of
G1 and G2 are in
*. It is usefull to notice that if
G
*, then every induced subgraph H of G (denoted by
H
G) also satisfies
H
*.
Lastly, we need to recall a well known property of minimal imperfect graphs due to Chvátal. A star-cutset of a graph G is a disconnecting set C (i.e. such that G - C has at least two connected components) containing a vertex adjacent to all other vertices of C.
Moreover, one can extend star-cutset lemma for any partitionable graph (see Maire's PhD Thesis [15] for a proof of this fact).