NDMI011: Combinatorics and Graph Theory 1

Lecture #8
Menger’s theorems and the Ear lemma
Irena Penev

In what follows, all graphs are finite, simple (i.e. have no loops and no
parallel edges), and non-null.

1 Menger’s theorems
Menger’s theorem (vertex version). Let G be a graph, and let A, B ⊆
V (G).1 Then the minimum number of vertices separating A from B in G is
equal to the maximum number of pairwise disjoint A-B paths in G.2

a4 = b1 A = {a1 , a2 , a3 , a4 }

a1 b2 B = {b1 , b2 , b3 }

a2 b3

a3
set of vertices
separating
A from B

Proof. We assume inductively that the theorem holds for graphs that have
fewer than |E(G)| edges. More precisely, we assume that for all graphs
G′ such that |E(G′ )| < |E(G)|, and all sets A′ , B ′ ⊆ V (G′ ), the minimum
number of vertices separating A′ from B ′ in G′ is equal to the maximum
number of pairwise disjoint A′ -B ′ paths in G′ . We must prove that this holds
for G as well. From now on, we let k be the minimum number of vertices
separating A from B in G.
1
A and B need not be disjoint.
2
“Pairwise disjoint” here means that no two paths have a vertex in common (and
consequently, no two paths have an edge in common).

1
 First, we claim that there can be no more than k pairwise disjoint paths
from A to B in G. Indeed, let X ⊆ V (G) be a k-vertex set separating A
from B in G, and let P be any collection of pairwise disjoint paths from A
to B. By definition, every path in P contains at least one vertex of X, and
since paths in P are pairwise disjoint, no two paths in P contain the same
vertex of X. So, |P| ≤ |X| = k, as we had claimed.
It remains to show that there are at least k pairwise disjoint paths from
A to B. Clearly, for any set X ⊆ V (G) separating A from B in G, we have
that A ∩ B ⊆ X; consequently, |A ∩ B| ≤ k. Now, if E(G) = ∅, then A ∩ B
separates A from B in G, and so |A ∩ B| = k; in this case, the vertices of
A ∩ B form k pairwise disjoint one-vertex paths from A to B, and we are
done. From now on, we assume that G has at least one edge, say xy. Let
Gxy := G/xy, i.e. let Gxy be the graph obtained from G by contracting the
edge xy, and let vxy be the vertex obtained by contracting xy.3

x y vxy

G G/xy

Now, if x or y belongs to A, then let A′ = (A \ {x, y}) ∪ {vxy }, and otherwise,

let A′ = A. Similarly, if x or y belongs to B, then let B ′ = (B \{x, y})∪{vxy },
and otherwise, let B ′ = B.
Let Y ⊆ V (Gxy ) be a minimum-sized set of vertices separating A′ from
′
B in Gxy .4 By the induction hypothesis, there are |Y | many pairwise disjoint
paths in Gxy from A′ to B ′ , and it readily follows5 that there are at least
|Y | many pairwise disjoint paths in G from A to B. So, if |Y | ≥ k,6 then
we are done. From now on, we assume that |Y | ≤ k − 1. Then vxy ∈ Y ,
for otherwise, Y would separate A from B in G,7 contrary to the fact that
|Y | ≤ k − 1. Now X := (Y \ {vxy }) ∪ {x, y} separates A from B in G,8
and we have that |X| = |Y | + 1. Note that this implies that |X| = k.9 Set
3
Formally, vxy is some (“new”) vertex that does not belong
 to V (G), and
Gxy is the graph with vertex set V (Gxy ) = V (G) \ {x, y} ∪ {vxy } and edge
set E(Gxy ) = {e ∈ E(G) | e is incident neither with x nor with y in G} ∪ {vvxy |
v ∈ V (G), v is adjacent to x or y in G}.
4
This means that for all sets Y ′ ⊆ V (Gxy ) separating A from B in Gxy , we have that
|Y | ≤ |Y ′ |.
5
Details?
6
In fact, it is not possible that |Y | > k (details?), but we do not need this stronger fact.
7
Proof?
8
Proof?
9
Indeed, since |Y | ≤ k − 1, we have that |X| ≤ k. On the other hand, since X separates
A from B in G, we know that |X| ≥ k. So, |X| = k.

2
X = {x1 , . . . , xk }.
We now consider the graph G \ xy, i.e. the graph obtained from G by
deleting the edge xy.10 Since x, y ∈ X, we know that any set of vertices
separating A from X in G \ xy also separates A from B in G;11 consequently,
any such set has at least k vertices, and so by the induction hypothesis,
there are k pairwise disjoint paths from A to X in G, call them P1 , . . . , Pk .
Similarly, there are k pairwise disjoint paths from B to X in G, call them
Q1 , . . . , Qk . We may assume that for all i ∈ {1, . . . , k}, xi is an endpoint
both of Pi and of Qi . So, P1 − x1 − Q1 , . . . , Pk − xk − Qk are pairwise disjoint
walks from A to B. But in fact, each of these walks is a path, for otherwise,
it would contain a path from A to B that contains no vertex of X.12 So,
there are at least k pairwise disjoint paths from A to B in G.

Given a graph G and distinct vertices s, t ∈ V (G), two paths from s to t

in G are internally disjoint if they have no vertices in common except the
endpoints s and t.
The following corollary is also often referred to as the vertex version of
Menger’s theorem.
Corollary 1.1. Let G be a graph, and let s, t ∈ V (G) be distinct, non-
adjacent vertices of G. Then the minimum number of vertices of V (G) \
{s, t} separating s from t in G is equal to the maximum number of pairwise
internally disjoint s-t paths in G.

The red and blue

path are internally
disjoint.
s t

set of two vertices

separating s from t

Proof. Let S = NG (s) and T = NG (t). Obviously, the minimum number of

vertices of V (G) \ {s, t} separating s from t in G is equal to the minimum
10
So, V (G \ xy) = V (G) and E(G \ xy) = E(G) \ {xy}.
11
Let us check this. Let Z be any set of vertices separating A from X in G \ xy, and
let p1 , . . . , pt , with p1 ∈ A and pt ∈ B, be a path from A to B in G. Then some vertex of
p1 , . . . , pt belongs to X; let i ∈ {1, . . . , t} be the smallest index such that pi ∈ X. Then
p1 , . . . , pi is a path from A to X in G. Furthermore, since p1 , . . . , pi contains exactly one
vertex of X, and since x, y ∈ X, we see that the path p1 , . . . , pi does not use the edge xy;
consequently, p1 , . . . , pi is a path from A to X in G \ xy, and we deduce that this path
(and consequently, the path p1 , . . . , pt as well) contains a vertex of Z.
12
Details?

3
number of vertices of V (G) \ {s, t} separating S from T in G \ {s, t}.13
Similarly, the maximum number of pairwise internally disjoint s-t paths
in G is equal to the maximum number of pairwise disjoint S-T paths in
G. By Menger’s theorem (vertex version), the minimum number of vertices
separating S from T in G \ {s, t} is equal to the maximum number of pairwise
disjoint S-T paths in G \ {s, t}. So, the minimum number of vertices of
V (G) \ {s, t} separating s from t in G is equal to the maximum number of
pairwise internally disjoint s-t paths in G. This completes the argument.

Our next goal is to prove the edge version of Menger’s theorem. The
line graph of a graph G, denoted by L(G), is the graph whose vertex set is
E(G), and in which e, f ∈ L(V (G)) = E(G) are adjacent if and only if e and
f share an endpoint in G.

f1
f1

e1 e4 e1 e4
e5 e5

e2 e3 e2 e3
f2
f2

G L(G)

Proposition 1.2. Let G be a graph, let s, t ∈ V (G) be distinct vertices of

G, let S be the set of all edges in G incident with s, and let T be the set of
all edges in G incident with t. Let X ⊆ E(G). Then X separates s from t in
G if and only if X separates S from T in L(G).

Proof. Suppose that X separates s from t in G; we must show that X

separates S from T in G. Suppose otherwise. Then there exists some path
e1 , . . . , er in L(G) that does not contain any vertex (in L(G)) from X.14 For
each i ∈ {1, . . . , r − 1}, let vi be a common vertex of ei and ei+1 .15 Then
s, v1 , . . . , vr−1 , t is a walk in L(G) from s to t that uses only edges e1 , . . . , er ,
and consequently, does not use any edge of X. It follows that there is a path
from s to t in G that does not use any edges of X, contrary to the fact that
X separates s from t in G. This proves that X indeed separates S from T
in G.
Suppose now that X does not separate s from t in G; we must show that
X does not separate S from T in L(G). Since X does not separate s from t
13
Indeed, for any set X ⊆ V (G) \ {s, t}, we have that X separates s from t in G if and
only if X separates S from T in G \ {s, t}.
14
Note that e1 , . . . , er are vertices of L(G), and consequently, edges of G.
15
Such a vertex exists because ei and ei+1 are adjacent vertices of L(G), and consequently,
they are edges of G that share an endpoint.

4
in G, we know that there is a path v1 , . . . , vr in G, with v1 = s and vr = t,
that does not use any edge of X. But now v1 v2 , v2 v3 , . . . , vr−1 vr is a path
from S to T in L(G) that does not use any vertex (in L(G)) in X. So, X
does not separate S from T in L(G).

Proposition 1.3. Let G be a graph, let s, t ∈ V (G) be distinct vertices of

G, let S be the set of all edges in G incident with s, and let T be the set
of all edges in G incident with t. Let ℓ be a non-negative integer. Then the
following are equivalent:

(i) there are ℓ pairwise edge-disjoint s-t paths in G;

(ii) there are ℓ pairwise disjoint S-G paths in L(G).

The red and blue path

are edge-disjoint.

s t

Proof. Suppose first that (i) holds, and fix ℓ pairwise edge-disjoint s-t paths
in G, say P1 , . . . , Pℓ . For all i ∈ {1, . . . , ℓ}, set Pi = v1i , . . . , vri i . Now, for all
i ∈ {1, . . . , ℓ}, set PiL = v1i v2i , v2i v3i , . . . , vri i −1 vri i (with v1i = s and vri i = t).
Clearly, P1L , . . . , PℓL are pairwise disjoint S-T paths in L(G).
Suppose now that (ii) holds, and fix ℓ pairwise disjoint S-T paths in
G, say QL L L i i
1 , . . . , Qℓ . For all i ∈ {1, . . . , ℓ}, set Qi = e1 , . . . , eri . Now, for
i
all i ∈ {1, . . . , ℓ} and j ∈ {1, . . . , ri }, let vj be a common vertex of the
edges eij and eij+1 in G, and set Qi = s, v1i , . . . , vri i −1 , t. Then Q1 , . . . , Qℓ are
pairwise edge-disjoint s-t walks in G, and we deduce that there are ℓ pairwise
edge-disjoint s-t paths in G.

Menger’s theorem (edge version). Let G be a graph, and let s, t ∈ V (G)

be distinct vertices of G. Then the minimum number of edges separating s
from t in G is equal to the maximum number of pairwise edge-disjoint s-t
paths in G.

edges separating s from t

s t

5
Proof. Let S be the set of all edges in G incident with s, and let T be the set
of all edges in G incident with t. By Proposition 1.2, the minimum number
of edges separating s from t in G is equal to the minimum number of vertices
separating S from T in L(G). By Proposition 1.3, the maximum number
of pairwise edge-disjoint s-t paths in G is equal to the maximum number
of pairwise disjoint S-T paths in G. By Menger’s theorem (vertex version),
the minimum number of vertices separating S from T in L(G) is equal to
the maximum number of pairwise disjoint S-T paths in G. We now deduce
that the minimum number of edges separating s from t in G is equal to the
maximum number of pairwise edge-disjoint s-t paths in G. This completes
the argument.

The global version of Menger’s theorem. Let G be a graph on at least

two vertices, and let k, ℓ ≥ 0 be integers.
(a) G is k-connected if and only if for all distinct s, t ∈ V (G), there are k
pairwise internally disjoint s-t paths in G.
(b) G is ℓ-edge-connected if and only if for all distinct s, t ∈ V (G), there are
ℓ pairwise edge-disjoint s-t paths in G.
Proof. We first prove (a). Suppose that G is k-connected, and let s and t be
distinct vertices of G.
Suppose first that s and t are non-adjacent. Since G is k-connected, s
and t cannot be separated by fewer than k vertices of V (G) \ {s, t}; so, by
Corollary 1.1, there are k internally disjoint paths between s and t.
Suppose now that s and t are adjacent. Set G′ = G \ st.16 By Propo-
sition 3.1 from Lecture Notes 7, G′ is (k − 1)-connected. Now s and t are
distinct and non-adjacent in G′ , and they cannot be separated (in G′ ) by
fewer than k − 1 vertices of V (G′ ) \ {s, t}; so, Corollary 1.1 guarantees that
there are k − 1 internally disjoint paths between s and t in G′ . These k − 1
paths, plus the one-edge path s, t form k internally disjoint paths in G.
Suppose now that there are k internally disjoint paths between any two
distinct vertices of G; we must show that G is k-connected.
Let us first show that |V (G)| ≥ k + 1. By hypothesis, G has at least two
vertices; fix any distinct vertices s, t ∈ V (G). Then there are k internally
disjoint paths between them, and all but possibly one of those paths have
an internal vertex;17 so these k paths together have at least k − 1 internal
vertices, and it follows that |V (G)| ≥ (k − 1) + 2 = k + 1,18 which is what
we needed.
It remains to show that for all sets X ⊆ V (G) such that |X| ≤ k − 1, we
have that G \ X is connected. Suppose otherwise, and fix some X ⊆ V (G)
16
So, G′ is the graph obtained from G by deleting the edge st.
17
If s and t are adjacent, then s, t is a path between s and t with no internal vertices.
However, any other path between s and t has at least one internal vertex.
18
We are counting the k − 1 internal vertices of our paths, plus the endpoints s and t

6
such that |X| ≤ k − 1 and G \ X is disconnected. Then G \ X has at least
two components, and we choose vertices s and t from distinct components of
G \ X. Now X ⊆ V (G) \ {s, t} separates s from t, and so by Corollary 1.1,
there can be at most |X| ≤ k − 1 internally disjoint paths between s and t
in G. But this contradicts the fact that there are k internally disjoint paths
between any two distinct vertices of G.
We now prove (b). Suppose that G is ℓ-edge-connected. Fix distinct
vertices s, t ∈ V (G). Since G is ℓ-edge-connected, s cannot be separated from
t with fewer than ℓ edges of G, and so by Menger’s theorem (edge version),
there are at least ℓ pairwise edge-disjoint paths between s and t in G.
Suppose now that G is not ℓ-edge connected. Then there exists a set
F ⊆ E(G) such that |F | ≤ ℓ − 1 and G \ F is disconnected. Since G \ F is
disconnected, it has at least two components; let s and t be vertices from
distinct components of G \ F . Now F separates s from t, and in particular, s
can be separated from t by at most |F | ≤ ℓ − 1 edges of G. So, by Menger’s
theorem (edge version), there are at most ℓ − 1 pairwise edge-disjoint paths
between s and t in G.

2 2-connected graphs and ear decomposition

A cut-vertex of a graph G is any vertex v ∈ V (G) such that G \ v has more
components than G.

cut-vertex

Recall that, for a non-negative integer k, a graph G is k-connected if

|V (G)| ≥ k + 1 and for all S ⊆ V (G) such that |S| ≤ k − 1, we have that
G \ S is connected. So, a graph is 2-connected if it has at least three vertices,
is connected, and has no cut-vertices.
Lemma 2.1. Let G be a graph on at least two vertices. Then G is 2-connected
if and only if any two distinct vertices lie on a common cycle.19
Proof. By Menger’s theorem (global version), a graph on at least two vertices
is 2-connected if and only if for any pair of distinct vertices, there are two
internally disjoint paths between them. But obviously, two distinct vertices
lie on a common cycle if and only if there are two internally-disjoint paths
between them. The result now follows.
19
Note that if G has at least two vertices, and any two distinct vertices lie on a common
cycle, then in particular, G contains a cycle, and therefore, G has at least three vertices.

7
 In this section, we give a full structural description of 2-connected graphs.
A path addition (sometimes called open ear addition) to a graph H is the
addition to H of a path between two distinct vertices of H in such a way
that no internal vertex and no edge of the path belongs to H. In the picture
below, we show how the cube graph can be constructed by starting with a
cycle of length four and then repeatedly adding open ears (the path/open
ear added at each step is in red).

The Ear lemma. A graph is 2-connected if and only if it is a cycle or can

be obtained from a cycle by repeated path addition.

Proof. We first prove the “if” (i.e. “⇐=”) part of the lemma. Clearly, cycles
are 2-connected (indeed, every cycle has at least three vertices, is connected,
and has no cut-vertices).20 Further, if a graph G can be obtained from a
2-connected graph H by adding a path, then G has at least three vertices
(because H does), and it is easy to see that G is connected and has no
cut-vertices;21 so, G is 2-connected. It now follows by an easy induction
(e.g. on the number of paths added) that any graph obtained from a cycle
by repeated path addition is 2-connected. This proves the “if” part of the
lemma.
It remains to prove the “only if” (i.e. “=⇒”) part of the lemma. Fix a
2-connected graph G. By Lemma 2.1, G contains a cycle.22 Now, let H be
a maximal subgraph of G that either is a cycle or can be obtained from a
cycle by repeated path addition.23 We must show that H = G.
First, we claim that H is an induced subgraph of G.24 If not, then there
exist distinct vertices u, v ∈ V (H) that are adjacent in G, but not in H; but
then the graph obtained from H by adding the one-edge path u, v contradicts
the maximality of H. So, H is indeed an induced subgraph of G.
20
Alternatively, this follows from Lemma 2.1.
21
Check this!
22
Indeed, G has at least three vertices (because it is 2-connected), and by Lemma 2.1,
any two of them lie on a common cycle. So, G contains a cycle.
23
This means that no subgraph H ∗ of G that either is a cycle or can be obtained from a
cycle by repeated path addition contains H as a proper subgraph.
24
A graph H is an induced subgraph of a graph G if V (H) ⊆ V (G), and for all distinct
u, v ∈ V (H), we have that uv ∈ E(H) if and only if uv ∈ E(G).

8
 u
v
H G

It remains to show that V (H) = V (G). Suppose otherwise. Then since

G is connected, there is at least one edge between V (H) and V (G) \ V (H);
fix adjacent vertices u ∈ V (H) and v ∈ V (G) \ V (H). Since both G is
2-connected, we know that G \ u is connected; consequently, there is a
path in G \ u from v to some vertex in V (H) \ {u}; let P = v, p1 , . . . , pt
(t ≥ 1) be a path in G \ u with pt ∈ V (H) \ {u}; we may assume that
p1 , . . . , pt−1 ∈ V (G) \ V (H).25 But now the graph obtained from H by
adding the path u, v, p1 , . . . , pt contradicts the maximality of H.
pt−1
pt
p1
u v
H G

This proves that V (H) = V (G). Since we already know that H is an induced
subgraph of G, it follows that H = G. This proves the “only if” part of the
lemma.

25
Otherwise, we fix a minimal index i ∈ {1, . . . , t − 1} such that pi ∈ V (H), and we
consider the path v, p1 , . . . , pi instead of v, p1 , . . . , pt .