Documentos de Académico
Documentos de Profesional
Documentos de Cultura
by
VINCENT EDWARD FA TICA
B.S., Syracuse University, 1972
ABSTRACT OF DISSERTATION
August 1990
ABSTRACT
Chapter one provides an introduction to and an overview of the
work.
The overview is condensed in this abstract.
For the
convenience of the reader, the basic notions and terms of graph theory
are presented in chapter two.
Chapter three gives an account of most of the major known
results on graphs which are edge-crticial with respect to independence
number. Chapter four presents three characterizations of these edgecritical graphs.
The following is shown in chapter five.
For the task of
decomposing a graph into disjoint subgraphs so that each inherits a
maximum independent set of vertices from such a set in the given
graph, so that the sum of their independence numbers equals that of
the given graph, and so that none of them may be further decomposed,
the edge-critical graphs are necessary and sufficient to serve as the
decomposing subgraphs.
by
VINCENT EDWARD FA Tl CA
DISSERTATION
(
August 1990
Approved - - - - - - - - - - - - - -
Date
(
\
Copyright 1990
VINCENT EDWARD FA TICA
(
ii
To my parents
John Joseph Fatica
and
Jean Moore Fatica
and to my grandmother
Carolyn Bertha Moore
in memory of her
iii
Contents
Chapter
page
iv
who is unfamiliar with the basic notions and terms of the subject, these are
provided in the next chapter. For more in-depth presentations of the rudiments
of graph theory, the reader is referred to books by Berge ([B]), Bondy and
Murty ([BM]), Harary ([H2D, and Graver and Watkins ((GW]).
given
The treatment
emphasis on its
symbolism.
mathematical
Edge-
critical graphs, per se, were studied for a brief time, by an apparently small
number of researchers, among them such notables as Lovasz and Gallai.
collection
The
theorems presented in chapter 3 are indeed nearly all of the major results in
the area.
The others
Chapter 5 asks and answers the following vaguely stated question: "Can a
graph be taken apart, or decomposed, into smaller, perhaps more manageable
graphs are, in a very natural way, both necessary and sufficient to serve as
the "pieces".
of chapter 5.
cube, the dodecahedron, and the icosahedron, and take them apart (as much as
we can) in the manner of chapter 5.
is particularly striking.
They
by subgraphs in such a
vertex
At
the
same
time
as
we
are
trying
to
prove
the
above-mentioned
consequence
of
9.2),
and
his
proof
uses
advanced
notions and
The
elementary proof presented here will make the proof of this extension of
Konig's theorem accessible to a greater audience.
Chapter 11
may
be considered an
epilogue.
It
notion which proved to be the seed from which all of these researches grew.
small gesture of this author's appreciation for the inspiration, assistance, and
encouragement that Jack Graver has provided him.
We shall use the simple term "graph" to connote what should be more
properly termed "finite, simple, undirected graph without loops".
purposes, a
For our
which are called the vertices of G, paired with another set E(G) (E for short),
whose elements, called the edges of G, are doubleton subsets of V.
The sets V
and E, aside from the stipulations that V be finite, and that each element of E
be a doubleton subset of V, are arbitrary.
(V, E) to
denote the graph G whose vertex set is the set V, and whose edge set is the
set E.
For a pair u
V, v
V of vertices of a graph G
(V, E), if e
{u, v}
E, (i.e., e is an edge of G), we shall say that the vertices u and v are adjacent,
or that u is adjacent to v, and vice versa.
the vertex u (or v), is incident with, or is an endpoint of, the edge e; we shall
also say that u and v are neighbors.
to be adjacent.
As an example,
K 4 may be pictured
d
b
In the Illustration, the heavy dots labelled a, b, c, and d represent the
vertices of K4 , and a line drawn between two dots reflects the fact that the
vertices represented by these dots comprise an edge of K4
Illustration 2:
l<:i_
Ks
Notice that in the case of K5 , that the lines drawn to represent the edges
were allowed to cross (their crossing is in fact of necessity, though a proof of
that fact is not within the scope of this work; the interested reader should
consult (CW]).
It is
understood that at the point where these lines cross, there is no vertex (there
not being a heavy dot there), and it should be imagined that the "edges" do not
in fact meet.
The graphs Kn (n
"complete".
C4,
Illustration 3:
C3
. Cs
graph is meant the number of edges incident with it; for a given vertex v of a
graph, its degree is denoted p(v).
Un
of G for each i
O, 1, ... , n - 1.
u,
Un =
The
The graph G
connected.
Illustration 4:
0..
---------
c
Later
in
this
connectedness.
work,
we
shall
come
across
more
refined
notions
of
A graph H
G
(V(H), E(H))
E(G)
and v
P 2 (V(H)), where
Certain types of
If G
(V, E) is a
deleting from V the vertex v, and deleting from E those edges incident (in G)
with v.
Similarly, if G
And too, if G
(V,
In a graph G
(of
vertices
of
G),
and
the
cardinality
of
maximum
with
this
notation, and
for
the sake
of conciseness,
maximum
IVCG) I
(i)
v(G)
(ii)
e(G)
IE(G) I, and
(iii)
6'(G)
v(G) - 2o:.(G), (v - 2a. for short), which has by a few authors been
work is dedicated.
10
Can we restrict our attention to any particular, hopefully small and welldefined, class of graphs to serve as these "pieces"?
Definition:
A graph G
every edge e of G.
That is, G is edge-critical if the removal of any one of its edges causes
the independence number to increase.
The edge-critical graphs were first considered by Zykov ([Z]) in 1949, and
the list of those who have studied them since then is short, though it includes
some of the most prominent names in graph theory.
11
We will present below many (in fact, almost all) of the basic results on
edge-critical graphs.
later chapters.
3.1
and let U
Let G
~anned
by the
3.2
3.3
isolated vertices (ie: without vertices of degree 0), then for every independent
set S of G, IN(S) I
3.4
(ibid)
IS J.
[ N(S)
U N(V) ]
vES
p(v)
3.5
v(G) - 2cx(G) + 1
o(G) + 1.
An edge-
v(G) - 2cx(G)
We shall adopt
12
rn
l:ln
v(G) - 2a.(G)
n.
n r n
(n
0, 1, 2, ... ).
The fact that a connected graph has no isolated vertices, together with 3.5
above provide the range of values of n in the definition of An.
For convenience, we will let ;:1_ 1 denote the collection of graphs whose
only member is the graph K 10 the graph with one vertex and no edges.
graph is edge-critical and connected, both trivially, and indeed o(K 1)
(This
-1).
3.6
3.7 (ibid)
A1
{C 3 , C 5 , C 7 ,
... }
13
Illustration 5:
3.8 (ibid)
K/s.
Though it is not the last in our list, investigation into the edge-critical
graphs seems to have come to an end with the following finite basis theorem.
3.9
For each of 6
1, 2, 3, ... there
exists a finite collection of graphs in A 8 with the property that all the graphs
in t.. 0 can be obtained from these by the replacement of some or all of their
edges by mutually disjoint paths of odd length.
Note:
We've seen in 3.7 that the single graph K3 provides such a "finite basis"
for A 1, and in 3.8 that the single graph K4 provides one for .6. 2
Such
Lovasz finishes the proof of 3.9 by deducing that a graph in the "basis"
for a given
to be sharp.
o may
In the case 6
vertices.
In
14
As 3.8 and 3.9 suggest, this process of odd-subdivision (and its "reversal")
preserve the property of edge-criticality.
3.10
a) If G = (V, E)
E.
f and v
E.
w, then the graph G' obtained from G by the deletion of the vertex
v and the edges {u, v} and {v, w}, and the subsequent identification
of the vertices u and w, is also in f; moreover, o(G')
b) If G
(V, E)
E.
r,
E.
o(G).
tices v 1 and v 2 , while N(v) is partitioned into the non-empty sets N(v 1)
and N(v 2 ), and if a new vertex u, and the edges {u, v 1} and {u, v 2 } are
added, then the resulting graph, G', is also inf; moreover o(G')
Before we
can appreciate
the
following
beautiful
theorem
o(G).
of Gallai,
characterizing the edge-critical graphs which are connected, but not highly
connected, we must elaborate on the notion of connectedness itself.
15
Gallai's
theorem,
below,
characterizes
the
edge-critical
graphs
of
3.11
{uJJ vJ be an edge of G 1
e and let G2
b)
Let G
Then G
Vu
of
G~
with u 2 and v 2 of G2
A.
Then {u, v}
is not an edge of G (cf., 3.1) and the graph G - {u, v} has exactly two
components. Let
N~'
1, 2, let
subgraph of G spanned by
N~
Ni
Gi has
the property that the graph Gi obtained from it by adding the edge
{u, v} belongs to A; and the other one (and only the other one) has
the property that the graph, Gi, obtained from it by identifying u and
v belongs to A.
c)
In both a) and b)
v(G)
v(G 1) + v(G 2 )
cx.(G)
o(G)
0CG1) + 0CG2) - 1.
and
16
edge-critical graphs in terms of the local interaction between a-sets and some
particular sets of vertices.
A;
To
set indusion
I
0 for
every O'.-set I of G - v;
(c)
Proof:
(a implies b)
Suppose G
N(v)
If N(v)
0,
then, being connected, G is the graph K1' whose only vertex is v; in this case
(b) follows trivially.
Let
Suppose N(v)
0 and let u
cx(G).
17
(ii)
Every o:-set of G - v meets N(v), for if one, say J, did not, J U {v}
would be an independent set in G with
IJ
{v}I
which is absurd.
!JI + 1
a(G)
0 since v
<x(G)
Let I be an a.-set of G - e.
I ll S
And,
G.
cx(G - v)
II I
cx(G)
1 (as in (i)).
Moreover,
So suppose w
Cc implies a)
G.
N(v).
If
N(v)
{w}, for I is
This is as desired.
N(v)
{w}.
Let e
I U {v} is independent in G - e
and thus
a(G - e) :;:;:;
implying G
II
U {v} I
II I + 1
a(G) + 1
t:... 0
The following definition will facilitate the statement and proof of our
third characterization of connected edge-critical graphs.
18
Definition:
I of G, I
4.2
A connected
graph
is edge-critical if and
only if it
has no
proper
hereditary subgraphs.
Proof
subgraph H of G is hereditary.
Case
1:
V{H)
If V(H)
=
I.
Let I be an a-set of G - e.
Since G is edge-
But I is independent in
H. This is absurd.
Case
~:
If V(H)
V(G) - V(H).
V(K)I
V(H) I
ryJH)
a(G) - a(H).
Since G is edge-critical,
II I
cx(G)
!CI - {v})
V(H)I
II
+ 1.
Certainly
II
V(H) I
19
o(G).
Then, G - e is itself a
20
answer them.
Definition:
Let G
(V, E) be a graph.
(c)
U VCH;)
Note:
i=l
Va.CG).
Of great importance is that the properties (a), (b), (c) above imply that if
k
2:cx.(Hi).
i=l
Definition:
Definition:
An a-decomposition of a
Note:
connected.
21
5.1
Proof:
the
the
kind
of decomposition
5.2
Proof:
That
(Note
G0
e0
e0)
ex(G 0 ), and
Gi - e 1 This cannot
continue ad infinitum.
with V(GN)
ex(G).
N, GN will be edge-critical,
5.3
If
(Lemma)
collection H 1,
is
connected
and
edge-critical,
then
there exists no
vertex sets span H, and which has the property that o:.(H)
Proof:
2:)x(Hi).
ex(H) + 1.
III
But
(
\
5.4
Proof:
If {H 1,
edge-critical, while not being a preferred ex-decomposition, this would imply the
existence, for at least one of the Hit of a collection of subgraphs, the existence
of which is precluded by 5.3.
5.5
critical subgraphs of G.
23
Proof:
This shows that the connected edge-critical graphs are sufficient for the
task of decomposing graphs in the sense of what we've called a preferred
cx-decomposi ti on.
24
The Platonic solids have been of interest for a very long time, and more
recently, their vertices and edges have become interesting examples of graphs.
We digress here to view these well known graphs in light of the foregoing
remarks.
1.
V(G).)
Illustration 6: Tetrahedron (K 4 )
2.
25
Illustration 7: Octahedron
I\
I
In the illustration, all the edges, both solid and broken, are edges of the
octahedron. The two subgraphs (triangles, three-cycles) show a preferred
a-decomposition of the octahedron into two edge-critical subgraphs, each with
3.
Illustration 8 depicts the cube (solid and broken edges) together with a
preferred a-decomposition (solid edges only) into 4 edge-critical subgraphs,
each with
o=
0.
26
f':
I
I
//1
'
"
" "-
/
I
k
(
4.
o\
8
=- 4
;b
/
V=-
'
"
"
""
I
~
27
v:: /,;),_
o<. :.
3.
=-
Referring to the
(ii)
contains 2 of the x's, then, of the v's and y's, it may contain none of
one and at most 1 of the other;
(iii) contains 3 of the x's, it may contain none of the v's and none of the y's.
3.
28
removal of the edge {xl> x 2 }, for, by virtue of the symmetry of the icosahedron,
this edge may represent any of its edges.
Moreover,
v - 2<X
12 - 2(3)
6.
5.
o=
2.
(Recall that
~2
is the
consistent.
V
<:>(.
= .2..0
8'
:::.
d =t
(
29
_., .-
--
-- -
- - - -
- - - -
--
........
'
'+ '
' '
'\
G-
... ... 7
10
\
\
'
'
--
l
'
'
--
- -
:-
- -
-- -
_,.
..-
/'
'>/'
10 and
o=
8. Since
;;...
30
Before
continuing
along
more
abstract
lines,
we
will
give
few
7.1
set inclusion among the subsets of V(G) which have non-empty intersection with
every a-set of G.
V(G')
V(G) U {v},
(i.e., add the new vertex v to G, and connect it with edges to each vertex in
U).
Then,
G'
Proof:
of G.
cx(G')
ll.
and o(G')
o(G)
+ 1.
Since G' is connected, we need only to show that the removal of any
31
(The argument that follows is similar to the one presented in the proof of 4.1.)
Some a.-set of G contains v 11 for if not, the set U - {vJ would contradict
the minimality of U.
contains some other element of U, the set U - {v 1} would again contradict the
minimality of U.
This is as
Sets such as the set U in the preceding theorem do not in general make
themselves obvious.
corollary to 7 .1.
7 .2
(Corollary to 7 .1)
V(G) then
the graph G' obt ined from G by the addition of a new vertex u, and new edges
from u to each of the members of {v} U N(v) is also connected and edgecritical.
Proof:
guarantees that the set {v} U N(v) has the necessary properties to serve as the
set U in the construction of 7 .1.
A-1' A 0 , A 1,
32
k~
i>
>
+
0
1<.:i.
}(3
+
l<.'f
Another example is given in Illustration 13.
Illustration 13:
7 .3
... ,
(i)
the vertices v 1, v 2 ,
... ,
v Zk+H y 1,
... ,
G of length 2k + 1,
(ii)
(iii)
Yk+l
1, ... , k),
33
For k
Cb)
Illustration 14:
v7
v,
Proof:
(a)
v(G)
(2k + l) + (k + 1) + 1
3k + 3. o:.(G)
(x.)
k + 1, for
(b)
(c)
+ 3 - 2(k +
1)
k + 1.
(b)
34
(c)
Vi
The graphs resulting from this construction are (2 but not 3)-connected;
indeed, the vertex Yk+i has degree 2.
7.4
v 2 k+H xi> x 2 ,
... ,
Let the vi
be the vertices of a complete graph K2 k+l and let the x 1 be the vertices of a
XzHZk+l
A 2 k+P
1 and k
2.
35
Illustration 15:
-x.,
'X-7
k::- I d =3
f<=;L
:?:
2k + 1.
cf =-S
1, 2, ... , 2k + 1} is independent in G.
Thus
v's, and if such a set contains one of the v's, then it may contain only 2k of
the x's (since the inclusion of one of the v's precludes the inclusion of an
antipodal pair of x's).
6k + 3, o(G)
Thus cx(G)
(6k + 3) - 2(2k + 1)
2k + 1. Since v(G)
(2k + 1) + (4k + 2)
2k + 1.
{xll
As we have
supposed without loss of generality that {xll x 4k+2} was removed, this
independent set could be taken to be
36
for 2k + 1 of the x's appear, but not x 2 and x2k+ 3 , the x's which neighbor v 1
(b)
The
If one of the edges {xH v J} is removed from G we may find the desired
independent set of cardinality 2k + 2 by simply choosing vJ along with the
2k + 1 vertices xH XHz xH 4 ,
the other neighbor of v J among the x's (which is XHzk+r) is not chosen. D
... ,
x 6 , YH y 2 , y 3
Let the
v's (respectively, the x's, the y's) be the vertices of a triangle (respectively, a
chordless 6-cycle, a triangle).
Xzi-1' Xw and XzHl and for i
For i
=
1, 2, 3, let
Yi
and x 2 H 2 (the subscripted indices on the x's being considered modulo 6).
is the icosahedron (pictured below).
This
37
Illustration 16:
V =I~
d ::. b
The construction
(respectively,
the
x's,
the
y's)
11
be
the
vertices
of
chordless
A8
4-cycle
Xw
and
38
Illustration 17:
Proof:
4.
or
(c) I contains 2 of the v's, in which case I contains at most 2 of the x's.
Case 2: If I contains 1 of the y's, then either
(a) I contains 3 of the x's and none of v's,
(b) I contains 2 of the x's and at most 1 of the v's, or
39
4.
symmetry of G.
(1)
(2)
(3)
(4)
{yl> x 3 }, representing typical edges from one of the 4-cycles to the 8-cycle.
{yl> Y2,
(2)
(3)
{y 1, x 2 , x 6 , v 2 , v 4 } and
(4)
Lastly, o(G)
v(G) - 20'..(G)
16 - 2(4)
8. D
....
40
7 .6
.. .,
x2k+i y 1, y 2 ,
.. .,
Yn+I> (k
2: 1).
and the y's be each the vertices of a chordless odd cycle of length 2k + 1, and
let
X;
also have as neighbors y,_ 1, Yi, and Yi+i (subscripts modulo 2k + 1).
Then
E A2k+2
2 and k
~I
j
k=::i.
Proof of
7.~:
.. ., {x 2 k+H
) d=6
Clearly, v(G)
Yzk+ 1}.
<>
1~ = 3
4k + 2.
{xH
And since {x 2 , x 4 ,
... ,
x 2 k} is independent in G,
41
k.
Therefore,
o(G)
v(G) - 2o:.(G)
(4k + 2) - 2k
2k + 2.
(2)
{xu y 1} and
(3)
...,
X21J,
(2)
...,
Xn:-1},
(3)
{x1'
.l2k+2
42
Vo.(G) will not be considered in this work, and it seems to be, in general, a
difficult one.
Definition.
. ,
Hk} of
U V(Hi).
i=l
ex.CG)
II I
.u o n vmi) I
t=l
Definition.
L: II n
i=l
VCH;) I ~ :Lex.CH;).
i=l
1'
It
i=l
will be seen that with some knowledge of the kinds of subgraphs which G has,
we can put very definite limits on the subgraphs needed to serve in an
43
o:-covering.
there
(ii)
These
differences
are
exemplified
by
the
following
simple
ex.:rnple,
showing a graph ({a)), its ex-decomposition ((b)), and two ex-coverings ((c) and (d))
by edge-critical subgraphs:
..
(a..)
(b)
..
(C)
:i..
~
(d)
:;:.
Note that the ve.:-tex x, which is not in Va does not appear in the ex.de~omposition.
Note also that the edge {w, x} in (c), and the triangle in (d),
44
8.1
Proof:
Let G be a graph.
trivially.
edges without increasing oJG) until this can no longer be done yields a subgraph
of G whose components, which are each edge-critical, constitute an a-covering
of G.
8.::.
Proof:
also an a-decomposition of G.
The desired
result is immediate. D
8.1 and 8.2 show, respectively, the sufficiency and the necessity of the
connected edge-critical graphs for the task of providing coverings of graphs
which respect vertex independence in the sense of our a.-coverings.
It is
unfortunate indeed that this class is so large, and that so little is known about
it.
In light of this, one might hope that for a given graph, it would be possible
The
first theorem along these lines, a celebrated theorem of Konig (see [BJ, [BMJ,
[H2], or [GWJ) well predates any consideration of edge-critical graphs or of
coverings in general which respect vertex independence in the sense of our a.coverings.
45
Definition:
into sets V 1 and V2 so that every edge of G has one of its vertices in V 1 and
the other in V2
Konig's Theorem:
Let J&
be a covering of G [in our sense of covering] whose elements are each an edge
of G.
a_(G)
1%1
is minimum, then
1%!.
(
'
Putting these last two observations together, and noticing that the first
of them allows us to handle graphs with isolated vertices, we may restate
Konig's theorem as
46
or, as
8.3.
Konig's Theorem:
If G contains no subgraph in
L}.u
then G has an
i.}. 0
8.4.
If G contains no subgraph in
i.}.. 0
L}.k
U ... U
A.k+i
L}.k-l
might be
tempted
L}. 3 , ..
1).
But this would be begging the question, for Berge's proof of 3.2 in fact
uses Konig's theorem.
47
does not rely on Konig's theorem, and when applied to an edge-critical graph
(where all the edges are critical) it may be stated:
This is the same as Berge's result, but without the work "chordless," and
suffices to allow the offhand deduction of Konig's theorem (as above) without
committing the fallacy of petitio principii.
8.5
(Conjecture)
.O.k-1
8.6
(Conjecture)
If G
E .O.k-l
To see that 8.6 implies 8.5, suppose that 8.6 holds, and that the graph G
contains no subgraph in .O.k.
subgraph in Ak+i-
48
which is
/:ik-l
2.
The
49
of a graph, this abuse of the language will not give rise to any ambiguity.
When we speak of a vertex v in this sense, let it be understood that we mean
the subgraph ({v}, ), and for an edge e
{e}).
9.1
(Conjecture)
of A. 2 ), then G has an a.-covering whose elements are vertices, edges, and odd
cycles (ie: elements of A.- 1 U A. 0 U A. 1).
To
this
end
we focus
our attention on
more
9.2
(Conjecture) If G
Unfortunately, our efforts toward this end will not be entirely successful;
that is, 9 .2 will not be proved.
50
advanced
theoretical
regular matroids).
notions and
techniques (most
Presently, we
discuss this
notably
the
theory of
just-mentioned extension of
Konig's Theorem, while making apparent our choice of the name very-odd-K 4 for
an element of
~2
Illustration 18:
where the wriggled lines denote pairwise vertex disjoint paths, and where the
four "faces" (triangles) of this "tetrahedron" are bound by cycles of odd length.
The graphs which we have called very-odd-K 4 's comprise a proper subset
of the collection of odd-K 4 's.
The labels o (for odd) and e (for even) refer to the number of
~2
51
9.3
(
In the next chapter, we will give a graph theoretic proof of 9.3.
A subset
of that proof will provide [and this will be duly noted at the proper time] a
proof of the following more technical result.
9.4
If G
conjectured 9.2.
X which
52
the graphs X or Y, and show that in either case, Z (from whose supposed
existence we wish to derive a contradiction) must have some very particular
properties.
9.5
If for some k
f::.k
connected, and which contains no very-odd-K 4 [resp. no odd-K 4 ], then for some j
with k
f::.J
!:roof:
(Refer to 3.11)
o2
which are (2 but not 3)-connected and which contain no very-odd-K 4 's [resp. no
odd-K/s].
o,
these, let G be one with a minimum number of edges, and let o(G)
(i.e., G
Ak).
Let G! and
G~
be as in 3.ll(b).
53
Illustration 20:
odd u-v
p<:..-t-h I~
G;
\
I
.1
\
/
I
G_,_
G~
is the
one with the property that the graph G 1 obtained from it by adding the edge
{u, v} is also in A, and that G2 is the one with the property that the graph G2
obtained from it by identifying u and v is again in A.
Since k
o(G)
6'(G 1) + o(G 2 )
(o(G 1), o(G)) are (k, 1), (k - 1, 2), (k - 2, 3), ... , (2, k - 1), (1, k).
(i)
54
G2 .
not using the vertex which results from the identification of u and v,
then this subgraph would also be a subgraph of G; and if 0 2 contains
such a subgraph which does use this vertex, then, in G, this subgraph,
with u and v split, together with a path of even length in
Gi connecting u
and v would again contradict G's containing no very-odd-K 4 's [odd-K.1 's].
The existence of such a path is guaranteed, again by virtue of 3.2, for if
e 3 is an edge of G 1 not equal to {u, v} but incident with u, then in 0 11 the
adjacent edges e 3 and {u, v} must lie on an odd cycle. The removal of the
edge {u, v} from this cycle provides the desired path of even length from
u to v in
Gf .
The minimality properties of G imply that if o(G;) > 2 then G 1 will be 3connected, and will therefore provide the graph whose existence is asserted in
If neither 6(0 1) > 2 nor 6(0 2 ) > 2, then it must
6(G 2 )
2 in which case G 1
~3)
E ~2
and G2
E ~2,
has a subgraph in
and the
~ 2,
which
is contrary to hypothesis. D
55
If
there
exists
counterexample
to
9.2,
let
it
have
the
following
X is edge cri ti ca 1
(b)
X is 3-connected
(c)
6(X) > 2
(d)
X contains no very-odd-K 4
(e)
6(X) is minimum among all graphs with properties (a), (b), (c), and (d)
(f)
e(X) is minimum among all graphs with properties (a), (b), (c), (d),
and (e).
If
there
exists
counterexample
to
9.4
let
it
have
the
following
Y is edge-critical
(b)
Y is 3-connected
(c)
6(Y) > 2
(d)
Y contains no odd-K 4
(e)
6(Y) is minimum among all graphs with properties (a), (b), (c), and (d)
(f)
e(Y) is minimum among all graphs with properties (a), (b), (c), (d),
and (e).
X or if Z
Y.
X.
Where a
Y, it is given in [bracketed]
56
9.6
Proof:
3.
In [S3J, Suranyi has proved the following lemma, which we will exploit
greatly, and which is stated below in terms appropriate to this discussion.
9.7
If Z
Ilk for k
::e:
1, and if x is a vertex of Z,
Observe that
o(Z - x)
9.8
v(Z - x) - 2a.(Z - x)
v(Z) - 1 - 2a.(Z)
o(Z) - 1.
Proof:
tl 0 U fl 1 U ... U fl 0 czi-i
for if one did, so also would Z, and thus by the minimality conditions, (e) and
(f) on Z must lie in !l 0 U fl 1
57
possessing a
9.9
Proof:
:?:
3.
Illustration 21:
-....
'-
'
'
r-/1/-
....
'
'9
.z:
/
'
/
l.
'
"
_j
'
.......
58
9.10
and z, it cannot
Proof:
Illustration 22:
(w)
-- -
-.w
9.11
3.
59
Proof:
9.12
Proof:
The sum of the degrees of the vertices of a graph must be even (being
Since in Z all
the degrees are 3, 3v(Z) must be even, requiring that v(Z) be even and
consequently that o(Z)
9.13
60
Illustration 23:
Ii ew
1':;)0/o..
+ed
.e~e.s
Then y and z divide this cycle into two y - z paths, one containing an
odd number of edges, the other containing an even number, greater than 2, of
edges.
The
remainder of the vertices along this path may be paired, as is shown by the
wriggled lines.
Now consider the odd cycle formed by the first of the above mentioned
paths and the edges {x, y} and {x, z}, together with the pairs of vertices just
mentioned.
original cycle of Z - x (for x is now included, while w is omitted), and have the
same value of v - 2o:. as did that cycle, namely 1.
consisting of this cycle and these edges has the same 6 as did the original cycle
of Z - x.
Now if, in the collection of edges and odd cycles which comprises Z - x,
we replace the original odd cycle with this new one and these isolated edges, we
61
9.14
Z contains no triangles.
Proof:
Illustration 24).
cycles must contain the edge {y, z}, they must be the same cycle.
Illustration 24:
(
'
odd path
\.._
____~, '-
........
__
......
This is contrary
(
9.15
62
Proof:
common
neighbor
on
cycle
unless that
9.16
Proof:
cycle is a triangle
- x for which no
- x which neighbor x.
Let x be any vertex of Z, and choose any Z -x for which the sum of
If there exists an
Illustration 25:
-t-h~ jde..
This even arc together with e forms an odd cycle, while the vertices
along the arc with an odd number of edges may be divided into adjacent pairs.
In Z - x, the cycle C can be replaced by this new odd cycle and the edges
formed by these adjacent pairs of edges.
63
(i)
(ii)
9.17
For at least one choice of the vertex x of Z, and for at least one
Proof:
Suppose on the contrary that for every choice of the vertex x, and
neighbors of x have another common neighbor on that cycle, and that 9.15
implies that the third neighbor of x be on some other odd cycle of Z - x.
**
following
the illustration number indicate that all else, not pictured, but whose existence
has at that point been asserted, is as it is in Illustration 26.
64
Illustration 26:
c..:i..
Ci;.
\
65
Let x be a vertex of Z, and from among all the realizations of
which the odd cycles neighboring x are chordless in Z (see 9.16), choose one for
which the number of isolated edges is a minimum.
components are the odd cycles Ci, i
Since p(w 2 )
Observe that w 3 is not on the cycle C 1, for this would imply that Z contains a
very-odd-K 4 [odd-K 4 ], namely the one shown in Illustration 27, where the dashed
line denotes the arc along C 1 from w3 to u 1 that has an odd number of edges.
Illustration 27:
(Very-odd-K 4 if w3 is on C 1)
c~
66
one of the isolated edges (see Illustration 28). This is contrary to 9.9.
Illustration 28**:
contradicting 9.9)
At this point we have just completed the initial step of the induction
argument.
Wzn+1'
with certain
properties which will be stated presently, it must be the case that Z has
numerous previously unmentioned additional vertices, and in particular one
among them which we will call
Wz 71 + 3 ,
.. .,
Wzn+i
This will
67
W2n
has neighbors
U2n
and
V2n
On
C2n
and
Uz 71
and
V2n
common neighbor x 2 n.
1.
For the sake of clarity, we present the induction step in terms of the
subscripts of the very next iteration, that is, the deduction of the existence of
w5 , satisfying the induction hypothesis, from the existence of w 3
The very
same arguments, with the appropriate subscripts, will allow the deduction of the
existence of w2 ,,+ 3 from that of w2 n+i
By hypothesis, in any
v 2 ) neighbor u 3 or v 3
as a neighbor.
68
Z - w2 ,
which cannot
happen.
This
realization
of
Z - w2
is depicted
in
Illustration 29.
C..z.
u.%.
Vi..
\
\
\
1C::. /)(._z
U.3
'X-3
WJ
c.3
x.
69
70
c.3
So x 4 has two "new" vertices as neighbors, call them u 4 and v 4 , and these
that the vertices u 4 , w'l, and v 4 must lie on some previously unmentioned cycle
of Z - x, call it C 4
Now, w5 is
71
c'l.
w~/
Finally, w5 cannot be on one of the isolated edges of Z - x (as w3 could
not have been), for there would then exist a realization of Z - w4 , as depicted
in Ill ustra ti on 34 in which in Z, w4 neighbors one of the isolated edges,
contrary to 9.9.
73
Illustration 34**:
(w 5
W4
contradicting 9.9)
C::i..
....
... ,
This implies
0
74
Illustration 35: (with possibly isolated edges and additional odd cycles).
CL_
~\
75
We are now
in a position
(9.3)
contain no odd-K 4 but which do not have a-coverings by vertices, edges, and
odd cycles.
. ,
U V(H;) then
!=l
a(G) s 2)x.CHi)
i=l
contain no odd-K 4 , would have no ex-covering by vertices, edges, and odd cycles
(since such would be an a-covering of G), and would have fewer edges than G.
This is contrary to our choice of G, and thus, for every edge e of G,
a.(G - e) > ex.CG). That is: G is edge-critical.
76
>
2, and
such a graph will not only complete this proof but also provide a proof of 9.4,
mentioned earlier, and again later in this chapter.]
In 9.5 it was shown that if such a graph exists, then one exists which is
3-connected.
It now follows that there exists a graph satisfying the conditions defining
the graph Y of Chapter 9, and in Chapter 9 (see 9.17) it was shown that such a
graph necessarily contains three disjoint odd cycles, CH C 2 , and C 3 , and a
vertex x adjacent to each of these three cycles, as shown.
Illustration 36:
C.;.
C3
~/
One version of Menger's Theorem ([MJ, see [BJ), a well known result in
graph theory, states:
77
generality that another of these paths avoids one of the cycles C 1 and C 3 , say
C3
If the third of them avoids the cycle C 1 then the claim is verified; if it
Illustration 37:
"-
"' '
LC
I J '2.
"
,,,
/
'\.
I
'-.,.
/ /
/
'f'-x.
The vertices where these paths meet C 2 , together with x's neighbor on C 2 ,
divide C 2 into three arcs.
78
Case 1:
Illustration 38:
,,
-- -
...._
,~3
......
\
\
'
I'
2 1
to y 1 and from
22 - 21 -
y1
2 3
-
22
x and
22 -
z3
y3
79
Illustration 39:
~,
...,,.
- - - -.,-r..
.........
\
I
odd
\
/
z1
y1
x and
z~
- z3
y3
Illustration 40:
Cases 3 and 4:
from y 2 to
z9
80
Illustration 41:
~I
,,,.-
I
I
ever;
z;\
- - -.,
evej
odd
'
,, I
/
~)
_j3
In this case, the arcs along C 1 and C 3 are chosen so as to make the
z2
z1
y1
y3
of type III shown below, and also providing the contradiction which completes
the proof.
Illustration 42:
To close this chapter, we recount what has indeed been shown in our
attempt to prove conjecture 9.2 (and consequently, conjec:ture 9.1).
81
(9.4)
subgraph an odd-K 4
10.1
3, then G contains as a
Proof:
10.2
o among
o>
2, then o(G)
4 (and
3, then
Proof:
edges.
If o(G) < 2, G is an
3, G
82
(V, E) be a tree.
for v
V,
if v
otherwise;
b)
for e
{u, v}
E,
if {u, v} n I rf- 0 for every
cx.-set I of G, while hG(u)
otherwise.
It is this very notion, that of the function hG, which has proved to be
the seed from which grew all of the work in this thesis.
83
for v
V,
hG(v)
(ij)
for e
{u, v}
and
E,
To see that
they also coincide on the edges of G, observe that hG(e), as originally defined
(1)
(2)
O;
(b)
hG(v) = 0, or
1 while hG(u)
1 and hc(v)
0.
In [GY] (p. 137), Graver and Yackel have shown that for a tree G
(V, E),
84
11.1
Let G
(V, E) be a tree.
hG(v)
IEI
Then
2:::C-U'g,(v)
i=O
and
i=l
hG(v)
where E'
a_(G) -
cx(G - v)
containing i edges.
above
summations.
difference,
since
it
contributes
equally
to
each
of
the
two
IEI
.
2.::(-1/g;(v)
i=O
as desired.
85
Furthermore,
where E'
g~
G - u - v containing i edges.
zero
to
the
above,
being
counted
equally
in
the
first
two
contributes equally to the first, third, and fourth summations, and not at all to
the second.
!El
.
- I:C-lYgi(e),
i=l
86
87
BIBLIOGRAPHY
[A]
[B]
C. Berge: "Graphs and Hypergraphs'', North-Holland, AmsterdamLondon, American Elsevier, New York, 1970.
[BHP]
L.W. Bein eke, F. Harary, and M. Plummer: "On the critical lines of a
graph", Pacific Journal of Mathematics, 22, no. 2, (1967), 205-212.
[BM]
[Gl]
[G2J
(G3]
[GW]
88
[GY]
[Hl]
[H2]
[Kl
[Ll]
[L2]
[M]
[Sl]
89
[S2J
[S3J
[W]
[ZJ
90
INDEX
adjacent ............................. 5
o:.(G) ................................... 9
a-covering ....................... 42
a.-decomposi ti on ................ 20
ex-set ................................. 9
a.-support ......................... 20
incident ................................. 5
independence number .............. 9
independent ........................... 8
independent set ...................... 9
isolated vertex .................... 11
K,, .........................................6
k-connected ......................... 14
Konig's Theorem ............. 44, 45
C,, ...................................... 7
chord ................................. 8
circuit ............................... ?
complete graph ................... 6
connected ........................... 7
connectivity ..................... 14
covering ........................... 42
cube ................................ 26
cut-set ............................. 11
cycle ................................. 7
N(v) ...................................... 8
neighbor ................................ 5
neighbor set .......................... 8
octahedron, .......................... 25
odd-K 4 ................................. 50
odd subdivision of K 4 ........... 12
deficiency .......................... 9
degree ................................ 7
D.. ..................................... 12
L'.ln ................................... 12
o(G) ...................................9
dodecahedron .................... 28
path ...................................... 7
platonic graph ...................... 24
preferred ex-decomposition .... 20
E(G) ...................................5
p(v) ....................................... 7
e(G) ...................................9
edge ................................... 5
edge-critical graph ............ 10
endpoint ............................. 5
finite basis theorem .......... 13
G - e ................................. 8
G - S .................................8
G - v ................................. 8
graph ................................. 5
hG .................................... 82
hereditary ........................ 18
span .................................... 11
subgraph ............................... 8
tetrahedron ......................... 24
tree ..................................... 82
V(G) ...................................... 5
v(G) ......................................9
V a(G) .................................. 20
valence .................................. 7
vertex ................................... 5
very-odd-K 4 ......................... 12
Z - x ................................... 56
icosahedron ...................... 26
91
BIOGRAPHICAL DAT A
Elementary School
Saint Patrick's School
Ravena, New York
High School
The Vincentian Institute
Albany, New York
Higher Ed uca ti on
B.S., Syracuse University 1972
M.S., Syracuse University 1977
M.Ph., Syracuse University 1985