Lecture 5, version April 18, 2023 page 1

Math 530 Spring 2022, Lecture 5: Bridges and

dominating sets
website: https://www.cip.ifi.lmu.de/~grinberg/t/22s

1. Simple graphs (cont’d)

1.1. The longest path trick
Here is another proposition that guarantees the existence of cycles in a graph
under certain circumstances. More importantly, its proof illustrates a useful
tactic in dealing with graphs:
Proposition 1.1.1. Let G be a simple graph with at least one vertex. Let d > 1
be an integer. Assume that each vertex of G has degree ≥ d. Then, G has a
cycle of length ≥ d + 1.

Proof. Let p = (v0 , v1 , . . . , vm ) be a longest path of G. (Why does G have a

longest path? Let’s see: Any path of G has length ≤ |V | − 1, since its vertices
have to be distinct. Moreover, G has at least one vertex and thus has at least
one path. A finite nonempty set of integers has a largest element. Thus, G has
a longest path.)
The vertex v0 has degree ≥ d (by assumption), and thus has ≥ d neighbors
(since the degree of a vertex is the number of its neighbors).
If all neighbors of v0 belonged to the set {v1 , v2 , . . . , vd−1 } 1 , then the

number of neighbors of v0 would be at most d − 1, which would contradict the

previous sentence. Thus, there exists at least one neighbor u of v0 that does
not belong to this set {v1 , v2 , . . . , vd−1 }. Consider this u. Then, u ̸= v0 (since a
vertex cannot be its own neighbor).
Attaching the vertex u to the front of the path p, we obtain a walk
p′ := (u, v0 , v1 , . . . , vm ) .
If we had u ∈ / {v0 , v1 , . . . , vm }, then this walk p′ would be a path; but this
would contradict the fact that p is a longest path of G. Thus, we must have
u ∈ {v0 , v1 , . . . , vm }. In other words, u = vi for some i ∈ {0, 1, . . . , m}. Consider
this i. Since u ̸= v0 and u ∈ / {v1 , v2 , . . . , vd−1 }, we thus have i ≥ d. Here is a
picture:

v0 v1 v2 ··· vi = u ···

1 If d − 1 > m, then this set should be understood to mean {v1 , v2 , . . . , vm }.

Lecture 5, version April 18, 2023 page 2

Now, consider the walk

c := (u, v0 , v1 , . . . , vi ) .

This is a closed walk (since u = vi ) and has length i + 1 ≥ d + 1 (since i ≥ d). If

we can show that c is a cycle, then we have thus found a cycle of length ≥ d + 1,
so we will be done.
It thus remains to prove that c is a cycle. Let us do this. We need to check
that the vertices u, v0 , v1 , . . . , vi−1 are distinct, and that the length of c is ≥ 3.
The latter claim is clear: The length of c is i + 1 ≥ d + 1 ≥ 3 (since d > 1
and d ∈ Z). The former claim is not much harder: Since u = vi , the vertices
u, v0 , v1 , . . . , vi−1 are just the vertices vi , v0 , v1 , . . . , vi−1 , and thus are distinct
because they are distinct vertices of the path p. The proof of Proposition 1.1.1
is thus complete.

1.2. Bridges
One question that will later prove crucial is: What happens to a graph if we
remove a single edge from it? Let us first define a notation for this:

Definition 1.2.1. Let G = (V, E) be a simple graph. Let e be an edge of G.

Then, G \ e will mean the graph obtained from G by removing this edge e.
In other words,
G \ e := (V, E \ {e}) .

Some authors write G − e for G \ e.

Theorem 1.2.2. Let G be a simple graph. Let e be an edge of G. Then:

(a) If e is an edge of some cycle of G, then the components of G \ e are

precisely the components of G. (Keep in mind that the components are
sets of vertices. It is these sets that we are talking about here, not the
induced subgraphs on these sets.)

(b) If e appears in no cycle of G (in other words, there exists no cycle of G

such that e is an edge of this cycle), then the graph G \ e has one more
component than G.
Lecture 5, version April 18, 2023 page 3

Example 1.2.3. Let G be the graph shown in the following picture:

a (1)

(where we have labeled the edges a and b for further reference). This graph
has 4 components. The edge a is an edge of a cycle of G, whereas the edge b
appears in no cycle of G. Thus, if we set e = a, then Theorem 1.2.2 (a) shows
that the components of G \ e are precisely the components of G. This graph
G \ e for e = a looks as follows:

and visibly has 4 components. On the other hand, if we set e = b, then

Theorem 1.2.2 (b) shows that the graph G \ e has one more component than
G. This graph G \ e for e = b looks as follows:

and visibly has 5 components.

Proof of Theorem 1.2.2. We will only sketch the proof. For details, see [21f6, §6.7].
Let u and v be the endpoints of e, so that e = uv. Note that (u, v) is a path of
Lecture 5, version April 18, 2023 page 4

G, and thus we have u ≃G v.

(a) Assume that e is an edge of some cycle of G. Then, if you remove e from
this cycle, then you still have a path from u to v left (as the remaining edges of
the cycle function as a detour), and this path is a path of G \ e. Thus, u ≃G\e v.
Now, we must show that the components of G \ e are precisely the compo-
nents of G. This will clearly follow if we can show that the relation ≃G\e is
precisely the relation ≃G (because the components of a graph are the equiva-
lence classes of its ≃ relation). So let us prove the latter fact.
We must show that two vertices x and y of G satisfy x ≃G\e y if and only if
they satisfy x ≃G y. The “only if” part is obvious (since a walk of G \ e is always
a walk of G). It thus remains to prove the “if” part. So we assume that x and y
are two vertices of G satisfying x ≃G y, and we want to show that x ≃G\e y.
From x ≃G y, we conclude that G has a path from x to y (by the second
proposition in Lecture 4). If this path does not use2 the edge e, then it is a path
from x to y in G \ e, and thus we have x ≃G\e y, which is what we wanted to
prove. So we WLOG assume that this path does use the edge e. Thus, this path
contains the endpoints u and v of this edge e. We WLOG assume that u appears
before v on this path (otherwise, just swap u with v). Thus, this path looks as
follows:
( x, . . . , u, v, . . . , y) .
If we remove the edge e = uv, then this path breaks into two smaller paths
( x, . . . , u) and (v, . . . , y)
(since the edges of a path are distinct, so e appears only once in it). Both of
these two smaller paths are paths of G \ e. Thus, x ≃G\e u and v ≃G\e y.
Now, recalling that ≃G\e is an equivalence relation, we combine these results to
obtain
x ≃G\e u ≃G\e v ≃G\e y.
Hence, x ≃G\e y. This completes the proof of Theorem 1.2.2 (a).
(b) Assume that e appears in no cycle of G. We must prove that the graph G \
e has one more component than G. To do so, it suffices to show the following:
Claim 1: The component of G that contains u and v (this component
exists, since u ≃G v) breaks into two components of G \ e when the
edge e is removed.
Claim 2: All other components of G remain components of G \ e.
Claim 2 is pretty clear: The components of G that don’t contain u and v do
not change at all when e is removed (since they contain neither endpoint of e).
Thus, they remain components of G \ e. (Formalizing this is a nice exercise in
formalization; see [21f6, §6.7].)
It remains to prove Claim 1. We introduce some notations:
2 We say that a walk w uses an edge f if f is an edge of w.
Lecture 5, version April 18, 2023 page 5

• Let C be the component of G that contains u and v.

• Let A be the component of G \ e that contains u.

• Let B be the component of G \ e that contains v.

Then, we must show that A ∪ B = C and A ∩ B = ∅.

To see that A ∩ B = ∅, we need to show that u ≃G\e v does not hold (since A
and B are the equivalence classes of u and v with respect to the relation ≃G\e ).
So let us do this. Assume the contrary. Thus, u ≃G\e v. Hence, there exists a
path from u to v in G \ e. Since e = uv, we can “close” this path by appending
the vertex u to its end; the result is a cycle of the graph G that contains the
edge e. But this contradicts our assumption that no cycle of G contains e. This
contradiction shows that our assumption was wrong. Thus, we conclude that
u ≃G\e v does not hold. Hence, as we said, A ∩ B = ∅.
It remains to show that A ∪ B = C. Since A and B are clearly subsets of C
(because each walk of G \ e is a walk of G, and thus each component of G \ e
is a subset of a component of G), we have A ∪ B ⊆ C, and therefore we only
need to show that C ⊆ A ∪ B. In other words, we need to show that each c ∈ C
belongs to A ∪ B.
Let us show this. Let c ∈ C be a vertex. Then, c ≃G u (since C is the
component of G containing u). Therefore, G has a path p from c to u. Consider
this path p. Two cases are possible:

• Case 1: This path p does not use the edge e. In this case, p is a path of
G \ e, and thus we obtain c ≃G\e u. In other words, c ∈ A (since A is the
component of G \ e containing u).

• Case 2: This path p does use the edge e. In this case, the edge e must be
the last edge of p (since the path p would otherwise contain the vertex u
twice3 ; but a path cannot contain a vertex twice), and the last two vertices
of p must be v and u in this order. Thus, by removing the last vertex from
p, we obtain a path from c to v, and this latter path is a path of G \ e (since
it no longer contains u and therefore does not use e). This yields c ≃G\e v.
In other words, c ∈ B (since B is the component of G \ e containing v).

In either of these two cases, we have shown that c belongs to one of A and B.
In other words, c ∈ A ∪ B. This is precisely what we wanted to show. This
completes the proof of Theorem 1.2.2 (b).
We introduce some fairly standard terminology:

3 Indeed,the path p already ends in u. If it would contain e anywhere other than at the very
end, then it would thus contain the vertex u twice (since u is an endpoint of e).
Lecture 5, version April 18, 2023 page 6

Definition 1.2.4. Let e be an edge of a simple graph G.

(a) We say that e is a bridge (of G) if e appears in no cycle of G.

(b) We say that e is a cut-edge (of G) if the graph G \ e has more compo-
nents than G.

Corollary 1.2.5. Let e be an edge of a simple graph G. Then, e is a bridge if

and only if e is a cut-edge.

Proof. Follows from Theorem 1.2.2.

We can also define “cut-vertices”: A vertex v of a graph G is said to be a cut-
vertex if the graph G \ v (that is, the graph G with the vertex v removed4 ) has
more components than G. Unfortunately, there doesn’t seem to be an analogue
of Corollary 1.2.5 for cut-vertices. Note also that removing a vertex (unlike
removing an edge) can add more than one component to the graph (or it can
also subtract 1 component if this vertex had degree 0). For example, removing
the vertex 0 from the graph

3 0 1

results in an empty graph on the set {1, 2, 3, 4}, so the number of components
has increased from 1 to 4.

1.3. Dominating sets

Here is another concept we can define for a graph:

Definition 1.3.1. Let G = (V, E) be a simple graph.

A subset U of V is said to be dominating (for G) if it has the following
property: Each vertex v ∈ V \ U has at least one neighbor in U.
A dominating set for G (or dominating set of G) will mean a subset of V
that is dominating.

4 When we remove a vertex, we must of course also remove all edges that contain this vertex.
Lecture 5, version April 18, 2023 page 7

Example 1.3.2. Consider the cycle graph

2
3

4
5
C5 = ({1, 2, 3, 4, 5} , {12, 23, 34, 45, 51}) = .

The set {1, 3} is a dominating set for C5 , since all three vertices 2, 4, 5 that
don’t belong to {1, 3} have neighbors in {1, 3}. The set {1, 5} is not a dom-
inating set for C5 , since the vertex 3 has no neighbor in {1, 5}. There is no
dominating set for C5 that has size 0 or 1, but there are several of size 2, and
every subset of size ≥ 3 is dominating.

Here are some more examples:

• If G = (V, E) is a simple graph, then the whole vertex set V is always
dominating, whereas the empty set ∅ is dominating only when V = ∅.
• If G = (V, E) is a complete graph, then any nonempty subset of V is
dominating.
• If G = (V, E) is an empty graph, then only V is dominating.
Clearly, the “denser” a graph is (i.e., the more edges it has), the “easier” it
is for a set to be dominating. Often, a graph is given, and one is interested in
finding a dominating set of the smallest possible size5 . As the case of an empty
graph reveals, sometimes the only choice is the whole vertex set. However, in
many cases, we can do better. Namely, we need to require that the graph has
no isolated vertices:
Definition 1.3.3. Let G be a simple graph. A vertex v of G is said to be
isolated if it has no neighbors (i.e., if deg v = 0).

An isolated vertex has to belong to every dominating set (since otherwise,

it would need a neighbor in that set, but it has no neighbors). Thus, isolated
vertices do not contribute much to the study of dominating sets, other than
inflating their size. Therefore, when we look for dominating sets, we can restrict
ourselves to graphs with no isolated vertices. There, we have the following
result:
5 Supposedly, this has applications in mobile networking: For example, you might want to
choose a set of routers in a given network so that each node is either a router or directly
connected (i.e., adjacent) to one.
Lecture 5, version April 18, 2023 page 8

Proposition 1.3.4. Let G = (V, E) be a simple graph that has no isolated

vertices. Then:

(a) There exists a dominating subset of V that has size ≤ |V | /2.

(b) There exist two disjoint dominating subsets A and B of V such that
A ∪ B = V.

One proof of this proposition will be given in homework set #2 exercise 4.

Another appears in [17s, §3.6].
Next, we state a rather surprising recent result about the number of dominat-
ing sets of a graph (which, I believe, appeared as a Putnam problem a couple
decades ago):

Theorem 1.3.5 (Brouwer’s dominating set theorem). Let G be a simple graph.

Then, the number of dominating sets of G is odd.

Three proofs of this theorem are given in Brouwer’s note [Brouwe09]. Let me
show the one I like the most. We first need a notation:

Definition 1.3.6. Let G = (V, E) be a simple graph. A detached pair will

mean a pair ( A, B) of two disjoint subsets A and B of V such that there exists
no edge ab ∈ E with a ∈ A and b ∈ B.

Example 1.3.7. Consider the cycle graph

3 2

4 1

5 6
C6 = ({1, 2, 3, 4, 5, 6} , {12, 23, 34, 45, 56, 61}) = .

Then, ({1, 2} , {4, 5}) is a detached pair, whereas ({1, 2} , {3, 4}) is not (since
23 is an edge). Of course, there are many other detached pairs; in particular,
any pair of the form (∅, B) or ( A, ∅) is detached.

Let me stress that the word “pair” always means “ordered pair” unless I say
otherwise. So, if ( A, B) is a detached pair, then ( B, A) is a different detached
pair, unless A = B = ∅.
Here is an attempt at a proof of Theorem 1.3.5. It is a nice example of how
to apply known results to new graphs to obtain new results. The only problem
is, it shows a result that is a bit at odds with the claim of the theorem...
Lecture 5, version April 18, 2023 page 9

Proof of Brouwer’s dominating set theorem, attempt 1. Write the graph G as (V, E).
Recall that P (V ) denotes the set of all subsets of V.
Construct a new graph H with the vertex set P (V ) as follows: Two subsets
A and B of V are adjacent as vertices of H if and only if ( A, B) is a detached
pair. (Note that if the original graph G has n vertices, then this graph H has 2n
vertices. It is huge!)
I claim that the vertices of H that have odd degree are precisely the subsets
of V that are dominating. In other words:

Claim 1: Let A be a subset of V. Then, the vertex A of H has odd

degree if and only if A is a dominating set of G.

[Proof of Claim 1: We let N ( A) denote the set of all vertices of G that have a
neighbor in A. (This may or may not be disjoint from A.)
The neighbors of A (as a vertex in H) are precisely the subsets B of V such
that ( A, B) is a detached pair (by the definition of H). In other words, they are
the subsets B of V that are disjoint from A and also have no neighbors in A (by
the definition of a “detached pair”). In other words, they are the subsets B of
V that are disjoint from A and also disjoint from N ( A). In other words, they
are the subsets of the set V \ ( A ∪ N ( A)). Hence, the number of such subsets
B is 2|V \( A∪ N ( A))| .
The degree of A (as a vertex of H) is the number of neighbors of A in H.
Thus, this degree is 2|V \( A∪ N ( A))| (because we have just shown that the num-
ber of neighbors of A is 2|V \( A∪ N ( A))| ). But 2k is odd if and only if k = 0.
Thus, we conclude that the degree of A (as a vertex of H) is odd if and only if
|V \ ( A ∪ N ( A))| = 0. The condition |V \ ( A ∪ N ( A))| = 0 can be rewritten as
follows:

(|V \ ( A ∪ N ( A))| = 0)
⇐⇒ (V \ ( A ∪ N ( A)) = ∅)
⇐⇒ (V ⊆ A ∪ N ( A))
⇐⇒ (V \ A ⊆ N ( A))
⇐⇒ (each vertex v ∈ V \ A belongs to N ( A))
⇐⇒ (each vertex v ∈ V \ A has a neighbor in A)
⇐⇒ ( A is dominating) (by the definition of “dominating”) .

Thus, what we have just shown is that the degree of A (as a vertex of H) is odd
if and only if A is dominating. This proves Claim 1.]
Claim 1 shows that the vertices of H that have odd degree are precisely the
dominating sets of G. But the handshake lemma (the first corollary in Lecture
2) tells us that any simple graph has an even number of vertices of odd degree.
Applying this to H, we conclude that there is an even number of dominating
sets of G.
Lecture 5, version April 18, 2023 page 10

Huh? We want to show that there is an odd number of dominating sets of G,

not an even number! Why did we just get the opposite result?
Homework: Find the mistake in our above reasoning! (I will also reveal it
next time.)

References
[17s] Darij Grinberg, Notes on graph theory, draft of two chapters, 4th April
2022.
https://www.cip.ifi.lmu.de/~grinberg/t/17s/nogra.pdf

[21f6] Darij Grinberg, Math 235 Fall 2021, Worksheet 6: Graphs and some of
their uses, 6 April 2022.
http://www.cip.ifi.lmu.de/~grinberg/t/21f/lec6.pdf

[Brouwe09] Andries E. Brouwer, The number of dominating sets of a finite graph is

odd, http://www.win.tue.nl/~aeb/preprints/domin2.pdf .