3

Connectivity

3.1 Cutpoints and Bridges

Among connected graphs, some are connected so slightly that removal of a single
vertex or edge will disconnect them. Such vertices and edges are quite important.
A vertex x is called a cutpoint in G if G - x contains more components than G
does; in particular if G is connected, then a cutpoint is a vertex x such that G - x
is disconnected. Similarly, a bridge (or cutedge) is an edge whose deletion increases
the number of components. As an example, consider the graphs in Figure 3.1; the
left-hand graph has cutpoints u and y and cutedge uy, while the right-hand graph has
cutpoint x and no cutedge.

t u v u v

wQkj x y z
I7V1
w x y

Fig. 3.1. Graphs whose cutpoints and cutedges are listed

A minimal collection of edges whose deletion disconnects G is called a cutset in

G. A cutset partitions the vertex-set V(G) into two nonempty components, say A and
B, such that the edges joining vertices in A to vertices in B are precisely the edges of
the cutset. This is the set of edges [A, B], as defined in Section 1.2, and we refer to
"the cutset [A, B]." (The two sets A and B are not uniquely defined - for example,
if there is an isolated vertex in G, it could be allocated to either set - but the cutset
will be well defined.) We omit set brackets when no confusion arises: for example, if
A = {x, y}, we might write [A, B] or [xy, B], but not [{x, y}, B].
If A is the set of all vertices of G other than x, so that A = V(G)\{x}, then
the cutset [A, x], consisting of all edges incident with the vertex x, is called a trivial
cutset.
44 3. Connectivity

Example. Consider the graph

abc d

G=I e
I I I
f g h
Three of the cutsets in this graph are
C = [ae, bcdfgh] ,
D = [abef, cdgh],
E = [b, acdefgh].
As graphs, the cutsets and their complements are
0-----0 0 0 o 0-----0 0 o
0-----0 0 0 o 0-----0 0 o
C D E

G\C
DO G\D G\E
(with vertex labels omitted). E is a trivial cutset.
The notation [A, B] can be used to represent the set of edges with one endpoint in
A and the other in B, even when A and B are not complements. In that case [A, B]
does not necessarily disconnect the graph. In the example graph G, [bf, cg] is the cut-
set D, but [ab, cg] is not a cutset. On the other hand, when A and B are complements,
[A, B] might not be a cutset: its removal will always disconnect G, but the set of edges
might not be minimal. Again referring to our example, [bf, bcdfgh] consists of the
four edges ab, be, ef, fg, but the deletion of ab and ef is sufficient to disconnect G.

Lemma 3.1. An edge xy in a connected graph G is a bridge if and only if it belongs

to no cycle in the graph.

Proof. Recall that a graph is connected if and only if any two vertices have a walk
joining them. Write H for the graph resulting when xy is deleted from C.
(i) Suppose xy is a bridge. G is connected but H is not. So there must exist vertices
wand z that are joined by a walk in G but not in H. Then xy must lie on every walk
from w to z in G. Consider a walk from w to z that includes xy. If xy lies on a cycle
in G, then delete the edge xy from the walk and replace it by the rest of the cycle. The
result is a walk from w to z in H, which is a contradiction. So xy belongs to no cycle
in G.
(ii) Suppose xy is not a bridge. Then H is connected. Select a walk from x to y
in H. By Theorem 2.1, this walk contains an x-y path; since xy is not an edge of H,
it is not an edge of the path, and the union of edge xy with the path is a cycle that
contains xy. D
 3.1 Cutpoints and Bridges 45

The deletion of a cutpoint from a connected graph may yield a graph with any
number of components. As an example, consider the star K1,n; deletion of its central
vertex yields n components. However, the situation with bridges is different.

Lemma 3.2. The deletion of a bridge from a connected graph yields a graph with
exactly two components.

Proof. Suppose G is a connected graph with bridge xy; write H for G with xy deleted.
Define H (x) to be the set of all vertices connected to x in H, and H (y) to be the set of
all vertices connected to y. If w is any vertex that does not lie in H(x), then the walk
from x to w in G must have contained the edge xy, and w must still be connected to
y. So H(x) U H(y) equals the vertex-set of G. The induced subgraph of H spanned
by H (x) is connected: if wand z are any two members of H (x), there is a walk from
w to z via x in H. Similarly (H(y») is connected. So H has at most two components.
Since xy is a bridge, H is not connected, so it has exactly two components. 0

Exercises 3.1

(lilvtj
A3.1.1 Find all cutpoints in the following graphs.

3.1.2 Find all cutpoints in the following graphs.

(ii) (iii)

3.1.3 What is the maximum number of bridges in a graph on v vertices?

A3.1.4 Prove that a graph in which every vertex has even degree can have no bridge.
H3.1.5 Prove that if x is a cutpoint of G, then x is not a cutpoint of G.
A3.1.6 List all cutsets in the graphs in Figure 3.2.

I /f
(ilc~e
(ii) a C

b~d
Fig. 3.2. Find all cutsets in these graphs

3.1.7 Repeat the preceding exercise for the graphs in Figure 3.3.
 46 3. Connectivity

(i) a b (ii) a b

~ rsJ
f g c d
(iii) a b (iv) a b c d

l/L rzs:sJ
c d e e f g h

Fig. 3.3. Find all cutsets in thesegraphs

3.1.8 Let x be a vertex of a connected graph G. Prove that the following statements are
equivalent:
(i) x is a cutpoint of G.
(ii) There exist vertices y and z. neither equal to x, such that x lies on every path
from y to z.
(iii) There exists a partition of V (G) \ {x} into subsets Y and Z such that for any
vertices y E Y and z E Z, x lies on every y-z path.
A3.1.9 The center C(G) of a finite graph G of radius R was defined in Section 2.2 to
consist of all those vertices x that have eccentricity e(x) = R; the graph G is
called self-centered if C(G) = G. Show that a connected self-centered graph can
contain no cutpoint.
A3.1.10 In the graph G, given in the example earlier, which of the following are cutsets?
(i) [ae, bcf] (ii) [abf, cdh]
(iii) [ade, bf] (iv) [ade, be]
(v) [ad, bee] (vi) [abed, efgh]

3.2 Blocks

A graph is called nonseparable if it is connected, nontrivial and contains no cutpoints.

A block in a graph G is a maximal nonseparable subgraph - that is, a nonseparable
subgraph that is not properly contained in any other nonseparable subgraph of G. A
nonseparable graph is itself often called a block. K2 is a block, but obviously no other
block can contain a bridge.
Any graph can be considered as a collection of blocks hooked together by its cut-
points. The other vertices are often called internal to their blocks, or simply internal
vertices.
Example. Partition the following graph into blocks.
 3.2 Blocks 47

P, Q

~
P,
x
'P
2
z
Y X
c§:b 'P
2
z
b
Y

(a) (b)

Fig. 3.4. Proof that (i) :::} (ii) in Theorem 3.3

The partition is

D~D~D
Theorem 3.3. Suppose G is a connected graph with at least three vertices. Then the
following are equivalent:
(i) G is a block.
(ii) Any two vertices of G lie on a common cycle.
(iii) Any vertex and edge of G lie on a common cycle.
(iv) Any two edges of G lie on a common cycle.

Proof. We prove (i) ::::} (ii), (ii) ::::} (iii) and (iv) ::::} (i). (The proof that (iii) ::::} (iv) is
similar to the proof that (ii) ::::} (iii).)
(i) ::::} (ii) Assume G is a block. Suppose x and y are distinct vertices of G, and
write X for the set of all vertices other than x that lie on a cycle passing through x.
Since G has at least three vertices and no cutpoint, it contains no bridge. So every
vertex adjacent to x is in X, and X is not empty.
Assume y is not in X; we shall derive a contradiction. Select a vertex z in X
such that the distance d (y, z) is minimal; let Po be a shortest y-z path, and write PI
and P2 for the two disjoint x-z paths that make up a cycle containing x and z. (See
Figure 3.4(a).) Since z is not a cutpoint, there will be an x-y path not containing z (see
Exercise 3.1.8); say Q is such a path. Let b be the vertex nearest to x in Q that is also
in Po, and a the last vertex in the x-b section of Q that lies in PI U P2 ; without loss of
generality we can assume a is in PI. This is illustrated in Figure 3.4(b).
We now construct two x-b paths R and S. To form R, follow PI from x to a and
Q from a to b. S consists of P2 followed by Po from z to b. Then R U S is a cycle
containing x and b, whence b is in X. The only vertex in Po U X is z, so b = z, and z
is in Q - a contradiction.
(ii) ::::} (iii) Select a vertex x and an edge yz of G. Let C be a cycle containing x
and y. If z is also on C, the required cycle is constructed from the edge yz together
with a y-z path that is part of C. Otherwise, select a z-x path that does not contain
y (this must be possible since y is not a cutpoint). Let a be the point of P n C that
 48 3. Connectivity

.:
y
P
x C
Fig. 3.5. Proof that (ii) =} (iii) in Theorem 3.3

is nearest to z. Then a cycle is formed as follows: take edge yz, followed by the z-a
section of P , and the a-y path of C that includes x . (See Figure 3.5.)
(iv) =} (i) Suppo se x is a cutpoint in G, and p is an edge containing x . From (iv),
p lies in a cycle, so x is on a cycle. But this contradicts Lemma 3. 1. Therefore G
contains no cutpoint, so it certainly contains no bridge. 0

The block graph B(G) of G has as its vertices the blocks of G; two vertices are
adjacent if the corresponding blocks have a common vertex .

Theorem 3.4. [64] A graph H is the block graph of som e graph if and only if every
block of H is complete.

Proof. Let H = B (G ), and assume there is a block Hi of H that is not complete. Then
there are two vertices in Hi that are nonadj acent and lie on a shortest common cycle C
of length at least 4. But the union of the block s of G corresponding to the points of Hi
that lie on C is then connected and has no cutpoint, so it is itself contained in a block,
contradicting the maximality property of a block of a graph.
On the other hand, let H be a graph in which every block is complete. Form B(H ),
and then form a graph G by adding to each vertex Hi of B(H ) a number of pendant
edges equal to the number of vertices of the block Hi that are not cutpoints of H. Then
it is easy to see that B(G ) is isomorph ic to H . 0

Exercises 3.2
A3.2.1 Partition the follow ing graphs into block s.

(i) (ii)

O) OTI
3.2.2 Partit ion the following graphs into blocks.

(ii)

A3.2.3 Prove that a connect ed graph with at least two edges is a block if and only if any
two adjacent edges lie on a cycle.
 3.3 Connectivity 49

3.2.4 The square G 2 of the graph G was defined in Exercise 2.1.11. If G is a nontri vial
connected graph, prove that G 2 is a block.
HA3.2.5 Write b( G) for the number of blocks of G , and be (x ) for the number of blocks of
G that contain the vertex x . If G is connected, prove that

b (G ) - 1 = :L)bcCx) - I]

(the sum is taken over all vertices x of G) . [63]

3.2.6 G is a nontrivial connected graph. C(G) is the number of cutpoints of G; c(B) is
the number of cutpoints of G that are vertices of the block B.
(i) Prove that c(B ) = C (B ) if and only if B = G.
(ii) Prove that
C (G ) - I = 2:)c(B ) - 1]
where the sum is over all blocks B of G. [51]
3.2.7 A graph G is a cr itic al blo ck if G is a block but G - p is not a block for any edge
p of G. A chord of G is an edge joining two vertices that lie on a cycle but are not
adjacent in the cycle . If G is a critical block with v( G) :::: 4, prove:
(i) G has no chords.
(ii) G contains no subgraphs isomorphi c to K 3 •
(iii) v( G) :s e(G) :s 2v( G) - 4.
(iv) If G is not a cycle, and if all vertices of degree 2 are deleted from G, the
resulting graph is disconnected. [93]

3.3 Connectivity

Generalizing the idea of a cutpoint, we define the connecti vity K (G) of a graph G to be
the smallest number of vertices whose removal from G results in either a disconnected
graph or a single vertex . (The latter special case is included to avoid problems when
discussing complete graphs. ) If K( G) :::: k, then G is called k-connected. The edge-
connectiv ity K ' (G) is defined to be the minimum number of edges whose removal
disconnects G (no special case is needed ). In other words , the edge-connectivity of
G equals the size of the smallest cutset in G . From the definition, it is clear that the
connecti vity and edge-c onnectivity of a graph is at least as great as that of any of its
subgraphs.
The following theorem is due to Whitne y [133]. Recall that 8(G ) denote s the min-
imum degree of vertices of G .

Theorem 3.5. For any graph G,

K(G) :s K ' (G ) :s 8(G).
Proof. It is clear that K '(G) :s 8(G), becau se one can disconnect G by removing all
edges incident with anyone given vertex.
Suppose T = [X , Y] is a cutset of minimal size in G , where X U Y = V (G ) and
X n Y = 0. Then K' (G ) = ITI.
50 3. Connectivity

If every vertex in X is adjacent in G to every vertex in Y , then the number of edges

in G is at least K '(G ) ::: IXI·IYI ::: v - I = 8(K v ) , where v = IV(G)I . But trivially
K '(G) ~ 8(G) ~ 8(K v ) , so G = K v , and in this case the theorem is easily seen to be
true.
So let us assume that there exist vertices x E X and y E Y that are not adjacent.
Define
S = {p : p E Y, px E T} U {q : q EX , q =1= x, qy E n.
Then G - S is a subgraph of G - T . Both x and y are vertices of G - S, and they are
in different components of G - T, so they are in different components of G - S, and
G - S is not connected. Therefore K(G ) ~ IS[. But /SI ~ ITI, since each vertex of S
is incident with at least one edge of T, and each edge of T is incident with exactly one
vertex in S. Therefore K(G ) ~ lSI ~ ITI = 8( G ). 0

Fig. 3.6. A graph G with K(G ) = I, K ' (G) = 2, B(G) = 3

It is easy to see that all combinations of strictne ss are possible in Theorem 3.5:
both of the inequalities can be strict, or one of them, or neither. For example, Figure
3.6 shows a graph G with K(G) < K ' (G ) < 8(G). Graph H in Figure 3.7 has K(H) =
K'(H) = 8(H), while K has K(K) < K'(K) = 8(K ). Chartrand and Harary [24]
proved that if e, m and n are any integer s such that 0 < e ~ m ~ n, then there is a
graph with K(G) = e, K ' (G ) = m and 8(G ) = n. However, if the minimum degree
is restricted in terms of the number of vertices in the graph, the amount of freedom
in assigning connectivities is considerably less; see Exercises 3.3.5 and 3.3.6. Apart
from the trivial case of K I , a graph is I-connected if and only if it is connected. All
2-connected graphs are blocks, and K 2 is the only block that is not 2-connected. So
we have the following corollary to Theorem 3.3.

H K

Fig. 3.7. K(H ) = K I ( H) = B(H ) = 2; K(K ) = I, K I ( K ) = B(K) = 2

 3.3 Connectivity 51

Theorem 3.6. G is 2-connected if and only if every two vertices of G lie on a cycle.

Corollary 3.7. All Hamiltonian graphs are 2-connected.

Dirac [36] proved that if G is n-connected, then any n vertices lie on a cycle, but
the converse is obviously false for n > 2. In fact, the characterization of n-connected
graphs in general is a difficult problem. Tutte found a characteri zation of 3-connected
graphs; for details see [116].

Exercises 3.3
A3.3.1 Find the connectivity and edge-c onnectivity of the following graphs .

3.3.2 Find the connectivity and edge-connectivity of the following graphs.

A3.3.3 Find examples, different from those in the text, of graphs with:
(i) K(G) = K /(G ) = 8(G);
(ii) K(G) < K'(G) = 8(G );
(iii) K(G) = K'(G) < 8(G).
3.3.4 Find a graph G with K(G) = 2, K/(G) = 3, 8(G) = 3.
3.3.5 Prove that if 8(G) ::: v(G) - 2, then K(G) = 8(G). Find a graph with 8(G) =
v(G) - 3 and K(G) < 8(G ).
A3 .3.6 Prove that if 8(G) ::: !v (G) , then K'(G) = 8(G). Find a graph with 8(G) =
L~ v (G) - l ] and K /(G ) < 8( G).
3.3.7 A-s an extension of Exercise 3.3.6, prove that for every v ::: 3 there exists a graph
on v vertices with 8(G ) = L!v - IJ and K /(G ) < 8(G ).
3.3.8 Prove that if 8( G ) ::: !(v(G) + k - 2), then K(G ) ::: k.