REGULAR GRAPHS OF GIVEN GIRTH

BROOKE ULLERY
Contents
1. Introduction
This paper gives an introduction to the area of graph theory dealing with prop-
erties of regular graphs of given girth. A large portion of the paper is based on
exercises and questions proposed by L aszl o Babai in his lectures and in his Discrete
Math lecture notes.
2. Graphs
Graph theory is the study of mathematical structures called graphs. We dene
a graph as a pair (V, E), where V is a nonempty set, and E is a set of unordered
pairs of elements of V . V is called the set of vertices of G, and E is the set of
edges. Two vertices a and b are adjacent provided (a, b) E. If a pair of vertices
is adjacent, the vertices are referred to as neighbors. We can represent a graph
by representing the vertices as points and the edges as line segments connecting
two vertices, where vertices a, b V are connected by a line segment if and only
if (a, b) E. Figure 1 is an example of a graph with vertices V = {x, y, z, w} and
edges E = {(x, w), (z, w), (y, z)}.

x
Figure 1
Now we dene some relevant properties of graphs.
Denition 2.1. A walk of length k is a sequence of vertices v
0
, v
1
, . . . , v
k
, such
that for all i > 0, v
i
is adjacent to v
i1
.
Denition 2.2. A connected graph is a graph such that for each pair of vertices
v
1
and v
2
there exists a walk beginning at v
1
and ending at v
2
.
Example 2.3. The graph in Figure 2 is not connected because there is no walk
beginning at z and ending at w.
Date: August 3, 2007.
1
2 BROOKE ULLERY

x
Figure 2
From this point on, we will assume that every graph discussed is connected.
Denition 2.4. A cycle of length k > 2 is a walk such that each vertex is unique
except that v
0
= v
k
.
From this point on, we will also assume that every graph discussed has at most
one edge connecting each pair of vertices. That is, we assume that there are no
2-cycles.
Denition 2.5. A tree is a graph with no cycles.
Denition 2.6. The girth of a graph is the length of its shortest cycle.
Since a tree has no cycles, we dene its girth as inf =
Example 2.7. The graph in gure 3 has girth 3.

a

b

c

d

e
Figure 3
Denition 2.8. The degree of a vertex is the number of vertices adjacent to it.
Denition 2.9. A graph is r-regular if every vertex has degree r.
Denition 2.10. A complete graph is a graph such that every pair of vertices
is connected by an edge.
We observe that a complete graph with n vertices is n 1-regular, and has
_
n
2
_
=
n(n 1)
2
edges.
Denition 2.11. A complete bipartite graph is a graph whose vertices can be
divided into two disjoint sets X and Y , such that every vertex in X is adjacent to
every vertex in Y and no pair of vertices within the same set is adjacent.
The complete bipartite graph with | X |= x and | Y |= y is denoted K
x,y
.
Denition 2.12. The complement of a graph G, denoted

G, is a graph on the
same vertices as G, only the vertices adjacent in G are not adjacent in

G and the
vertices not adjacent in G are adjacent in

G.
REGULAR GRAPHS OF GIVEN GIRTH 3
Example 2.13. In Figure 4,

G is the complement of G.

b

c

b

c

f
G

G
Figure 4
3. Girths of Regular Graphs
Using only the denitions of the previous section and some elementary linear
algebra, we are able to prove some interesting results concerning r-regular graphs
of a given girth. We begin with two lemmas upon which the rest of the paper will
depend.
Lemma 3.1. The number of vertices of an r-regular graph with an odd girth of
g = 2d + 1 (where d Z
+
) is n, such that
n 1 +
d1

i=0
r(r 1)
i
.
We will rst show this for g = 5:
Proof of Lemma 3.1 for g = 5. Let G be an r-regular graph with girth 5. Take a
vertex v
0
of G. Let V
0
= {v
0
}. v
0
must be adjacent to r vertices. Let V
1
be the
set containing those r vertices. None of the elements of V
1
can be adjacent to one
another, because that would create a 3-cycle. Thus, each of those vertices must
be adjacent to an additional r 1 vertices. Each of these vertices must be unique
in order to avoid creating a 4-cycle. Let V
2
be the set of all vertices adjacent to
elements of V
1
except v
0
. Elements of V
2
can be adjacent to one another, creating
5-cycles. Thus, adding up the vertices in Figure 5, we have
n = 1 + r + r(r 1)
n = 1 +
1

i=0
r(r 1)
i
.

v0




V
0
(1 vertex)
V
1
(r vertices)
V
2
(r(r 1) vertices)
4 BROOKE ULLERY
Figure 5
Now we extend this to any g = 2d + 1.
Proof of Lemma 3.1. Let G be an r-regular graph with girth g = 2d + 1. Take a
vertex v
0
of G. Let V
0
= {v
0
}. v
0
must be adjacent to r vertices. Let V
1
be the set
consisting of those r vertices. None of the elements of V
1
can be adjacent to one
another without creating a 3-cycle. Now, we will create more sets of vertices, as
in Figure 5. The vertices adjacent to those in V
1
(not including v
0
) will be in the
set V
2
. Similarly, if v is a vertex, we dene V
k
= {v | v is adjacent to an element
of V
k1
and v / V
n
where n < k 1} . There can be no cycles within vertices in

d1
i=0
V
i
. Suppose there were such a cycle. Then, without loss of generality, we can
assume that the cycle is a result of an adjacency between two elements within the
same set V
n
. Then, there would be a 2n+1-cycle, and since n < d, 2n+1 must be
less than g. Thus, there can be no such cycle, which implies that each vertex of V
d
is a unique vertex of G. Each set V
n
(< n d) has a cardinality of (r 1) times
the number of vertices of V
n1
, and V
1
has r vertices. Thus, V
k
has r(r 1)
k1
vertices. So, summing the cardinalities of V
0
through V
d
we have,
n 1 +
d1

i=0
r(r 1)
i

We now give a similar lemma for graphs of even girth.

i=0
r(r 1)
i
= 2
d1

i=0
(r 1)
i
.
We will prove this in two dierent ways. In the rst proof, we will begin with a
vertex of the graph, as in the proof of Lemma 3.1, and in the second one we will
begin with two adjacent vertices.
Proof #1 of Lemma 3.2. Let G be an r-regular graph with girth g = 2d, where
d 2. Let v
0
be a vertex in G. Let V
0
= {v
0
}. v
0
must be adjacent to r vertices.
Let V
1
be the set consisting of those r vertices. None of the elements of V
1
can be
adjacent to one another without creating a 3-cycle. The vertices adjacent to those
in V
1
(not including v
0
) will be in the set V
2
. Similarly, if v is a vertex, we dene
V
k
= {v | v is adjacent to an element of V
k1
and v / V
n
where n < k 1} . There
can be no cycles within vertices in

d1
i=0
V
i
. Suppose there were such a cycle. Then,
without loss of generality, we can assume that the cycle is a result of an adjacency
between two elements within the same set V
n
. Then, there would be a 2n+1-cycle,
and since n < d, 2n + 1 must be less than g. Thus, there can be no such cycle,
which implies that there must be more vertices than just those in sets V
1
through
V
d1
. Each set V
n
(n > 0) has a cardinality of (r 1) times the number of vertices
of V
n1
, and V
1
has r vertices. Thus, V
k
has r(r 1)
k1
vertices. Thus, we have
REGULAR GRAPHS OF GIVEN GIRTH 5
counted a total of
n 1 +
d2

i=0
r(r 1)
i
vertices so far. However, each vertex in V
d1
must have r 1 additional neighbors
which are not in

d1
i=0
V
i
in order to satisfy the girth requirement of 2d and r-
regularity. That gives r(r 1)
d1
additional vertices. However, these vertices need
not be unique in order to create a 2d-cycle. Since each of the vertices can have at
most r neighbors within V
d1
, we can divide by r so that we only actually need an
additional (r 1)
d1
unique vertices. So, summing the cardinalities of V
0
through
V
d1
, and the additional vertices that are adjacent to those in V
d1
, we have,
n 1 + (r 1)
d1
+
d2

i=0
r(r 1)
i
.

Proof #2 of Lemma 3.2. Let G be an r-regular graph with girth g = 2d, where
d 2. Let a
0
and b
0
be adjacent vertices in G. Let V
1
= {a
0
, b
0
}. Each vertex in
V
1
must be adjacent to an additional r 1-vertices. These vertices must be unique
in order to prevent a 3-cycle from being created. Dene V
2
as the set of the 2(r 1)
vertices adjacent to the vertices in V
1
(not including elements of V
1
. Similar to the
rst proof, we dene V
k
= {v | v is adjacent to an element of V
k1
and v / V
n
where n k1}. There can be no cycles within vertices in

d1
i=0
V
i
. Suppose there
were such a cycle. Then, without loss of generality, we can assume that the cycle
is a result of an adjacency between two elements within the same set V
n
. Then,
there would be a 2n-cycle, and since n < d, 2n must be less than g = 2d. Thus,
there can be no such cycle, which implies (since g is even) that each vertex of V
d
is a unique vertex of G. Each set V
n
has a cardinality of (r 1) times the number
of vertices of V
n1
and V
1
has 2 vertices. Thus, V
k
has 2(r 1)
k1
vertices. So,
summing the cardinalities of V
1
through V
d
, we obtain
n 2
d1

i=0
(r 1)
i
.
Now, simple algebra shows that 1+(r1)
d1
+

d2
i=0
r(r1)
i
= 2

d1
i=0
(r1)
i
:
Let x = r 1. Then the left hand side of the equation becomes:
LHS = (x + 1)
d2

i=0
x
i
+ x
d1
+ 1
=
(x + 1)(x
d1
1)
x 1
+ x
d1
+ 1
=
1
x 1
_
(x + 1)(x
d1
1) + (x
d1
+ 1)(x 1)

=
1
x 1
(x
d
x + x
d1
1 + x
d
x
d1
+ x 1)
=
2x
d
2
x 1
= 2
_
x
d
1
x 1
_
= 2
d1

i=0
x
i
= 2
d1

i=0
(r 1)
i
= RHS

6 BROOKE ULLERY
4. Minimal Regular Graphs of Given Girth
A natural question that arises from the lemmas in the previous section is when
do the equalities hold? That is, for which pairs of g = 2d + 1 and r can we nd a
graph such that the number of vertices is
(1) n = 1 +
d1

i=0
r(r 1)
i
,
and for which pairs of g = 2d and r can we nd a graph such that the number of
vertices is
(2) n = 2
d1

i=0
(r 1)
i
?
For small values of g there are many values of r for which the equality holds. We
begin with g = 3. In this case, we must nd r-regular graphs with girth 3 and
1 + r(r 1)
0
= 1 + r vertices.
Observation 4.1. For every integer r > 1, there exists an r-regular graph with
girth 3 and exactly r + 1 vertices.
Proof. Given an integer r > 1, the complete graph with r + 1 vertices is r-regular
and has a girth of 3.
In order to further understand regular graphs of given girth, we must introduce
some linear algebra relevant to graph theory.
Denition 4.2. The adjacency matrix A of a graph G is the square matrix
whose entries are a
ij
, where
a
ij
=
_
1 if i and j are adjacent
0 otherwise.
Clearly, the adjacency matrix of a (non-directed) graph is symmetric, that is
A
T
= A, since adjacency is a symmetric relation. That is, i is adjacent to j j
is adjacent to i. We also observe that the sum of the entries of the ith row or of
the ith column is equal to the degree of the ith vertex. Therefore, the sum of any
row or any column of the adjacency matrix of an r-regular graph is r.
A result of the spectral theorem is that every symmetric matrix has an orthonor-
mal eigenbasis, and real eigenvalues. Thus, the same holds for every adjacency
matrix.
Observation 4.3. The square of an adjacency matrix has entries b
ij
, where b
ij
is
equal to the number of vertices which are adjacent to both vertex i and vertex j for
i = j, and b
ii
is equal to the degree of vertex i.
Proof. Suppose A is an n by n adjacency matrix of a labeled graph. Then, by
matrix multiplication, A
2
will have entries b
ij
, where
b
ij
=
n

k=1
a
ik
a
kj
.
REGULAR GRAPHS OF GIVEN GIRTH 7
We notice that the product a
ik
a
kj
will be nonzero if and only if k is adjacent to
both i and j, in which case a
ik
= 1 and a
kj
= 1 so a
ik
a
kj
= 1. Thus,
b
ij
=
l

k=1
1 = l,
where l is the number of vertices which are adjacent to both vertex i and vertex j.
b
ii
will thus simply be the number of vertices adjacent to i, or the degree of i, since
i shares all adjacencies with itself.
Since each entry of A
2
has a geometric interpretation within the graph, it is
reasonable to infer that the same holds for A
k
. In fact, we can state the following
lemma.
Observation 4.4. Let A be an adjacency matrix of a labeled graph G, with n
vertices. Then the entry a
ij
of A
k
, where k 1, is the number of walks of length k
from vertex j to vertex i.
Proof by Induction. The statement holds for k = 1 by the denition of an adjacency
matrix. Assume the statement holds for k = l. Then, we multiply A
l
and A to
obtain A
l+1
. It does not matter in which order we multiply them since matrix
multiplication is associative. Let the entries of A be denoted b
ij
and let the entries
of A
l
be denoted c
ij
. The entry a
ij
of A
l+1
is the dot product of the ith row of A
and the jth column of A
l
. That is,
a
ij
=
n

m=1
b
im
c
mj
.
We notice that the product b
im
c
mj
will be nonzero if and only if i is adjacent to m
and there is at least one walk of length l from j to m. this is equivalent to saying
there is at least one walk of length l + 1 from j to i whose second to last vertex is
m. Thus, if b
im
c
mj
is nonzero, it is equal to the number of walks of length l + 1
from j to i whose second to last vertex is m. Thus, summing over n, we nd that
a
ij
=

n
m=1
b
im
c
mj
is the total number of walks of length l from j to i.
Now that we have studied adjacency matrices, let us briey return to r-regular
graphs of girth 3 with r +1 vertices (complete graphs) in order to give an example.
Example 4.5. Let G be a complete graph of degree r > 1, and let A be its adjacency
matrix. Since G is complete, the entries of A are
a
ij
=
_
0 if i = j
1 if i = j.
Also, since G has r + 1 vertices, every pair of unique vertices has exactly r 1
neighbors in common, so the entries of A
2
are
a
ij
=
_
r if i = j
r 1 if i = j.
Thus, for the adjacency matrix of the complete graph, the following equality holds:
A + A
2
= rJ,
where J is the matrix of all ones.
8 BROOKE ULLERY
Now, let us look at r-regular graphs of girth g = 4 (d = 2). According to Lemma
3.2, at least n vertices are required to make such a graph, where
2
1

i=0
(r 1)
i
= 2(r 1)
0
+ 2(r 1)
1
= 2 + 2r 2 = 2r.
In fact, only 2r vertices are required, as stated in the following lemma.
Lemma 4.6. For every integer r > 1, the only r-regular graph G that has girth
g = 4 and exactly 2r vertices is the complete bipartite graph K
r,r
.
Proof. Let x be a vertex in G. x is adjacent to r vertices. Let Y be the set of
vertices adjacent to x. Let X be the set of r vertices not adjacent to x (including x
itself). No two vertices in Y can be adjacent without creating a 3-cycle. Thus, every
vertex of Y must be adjacent to every vertex in X in order for G to be r-regular.
Thus, no two vertices in X can be adjacent without creating a 3-cycle. Therefore,
G is the complete bipartite graph K
r,r
. Suppose v X and z Y . Then, the
vertices x, y, v, and z create a four cycle. Thus, G has girth 4.
By using Lemma 4.6, we can make the following observation about the adjacency
matrix of an r-regular graph of girth four with 2r vertices:
Observation 4.7. Let G be an r-regular graph of girth four with 2r vertices, and
let A be an adjacency matrix of G. Then the following equality holds:
A
2
+ rA = rJ.
Proof. Let i and j be vertices in G. Since G is a complete bipartite graph, i and j
have r neighbors in common if i is not adjacent to j and no neighbors in common
if i is adjacent to j. Thus, A
2
has zeros where A has ones, and rs where A has
zeros. Thus,
A
2
+ rA = rJ.

Now let us approach the problem of when the equality in equation (1) holds for
regular graphs of girth g = 5 (d = 2). That is, we must nd for which values of r
we can nd a graph with girth 5 and n vertices, where
n = 1 + r(r 1)
0
+ r(r 1)
1
= 1 + r + r
2
r = 1 + r
2
It turns out that there are only a nite number of values of r for which the equality
holds. In fact, there are at most 5 values: {1,2,3,7,57}. We say at most because
such a graph has not yet been discovered for r = 57. To give an idea of how complex
the graphs become as r increases, lets look at the simplest of these graphs:

a

b
Figure 6: r = 1, n = 2

e
REGULAR GRAPHS OF GIVEN GIRTH 9
Figure 7: r = 2, n = 5

j
Figure 8 (Petersen Graph): r = 3, n = 10
The graph for r = 7 is called the Homan-Singleton graph, and has 50 vertices
and 175 edges. In fact, the theorem in this section regarding graphs of girth ve is
known as the Homan-Singleton Theorem. Before we state the Homan-Singleton
Theorem, we must discuss some features of regular graphs with girth 5 for which
the equality in equation (1) holds.
Lemma 4.8. If G is an r-regular graph, has girth at least 5, and n = r
2
+ 1
vertices, the number of common neighbors of a pair x, y of vertices is zero if x, y
are adjacent and 1 if x, y are not adjacent.
Proof. For this proof, we will be directly referring to the proof of Lemma 3.1 for
g = 5 and to Figure 5. Without loss of generality, let x = v
0
.
Case 1: y V
1
. In this case, y is adjacent to x and x and y have no neighbors
in common, so the lemma holds.
Case 2: y V
2
. In this case, y is not adjacent to x, and x and y are both
adjacent to exactly one vertex in V
1
. Thus the lemma holds.
Theorem 4.9 (Homan-Singleton Theorem). If there exists an r-regular graph G
of girth greater than or equal to 5 such that the number of vertices n satises the
equality n = r
2
+ 1, then r {1, 2, 3, 7, 57}.
See appendix for proof.
5. Further Studies
It is possible to study the minimal regular graphs of girth greater than 5, however,
such studies require more complex mathematical tools than those used in this paper,
and some have even been studied with no success yet.
10 BROOKE ULLERY
6. Appendix
The following proof was given in a lecture by L aszl o Babai. I added it for
completeness of the paper.
Proof of Theorem 4.9. Let A be the adjacency matrix of G, and let

A be the adja-
cency matrix of

G. By Observation 4.3 and Lemma 4.8, we have
(3) A
2
=

A + rI,
where I is the identity matrix. Also, by the denition of the complement of a graph,
we have
(4) A +

A + I = J,
where J is the matrix of all ones. Thus, substituting equation (3) into equation
(4), we have
A + (A
2
rI) + I = J
A + A
2
(r 1)I = J (5)
As observed earlier, the sum of the elements of each row of A is r. Thus,
A 1 =
_

_
r
r
.
.
.
_

_,
where 1 is the vector of all ones. So we have A 1 = r 1, which implies that 1
is an eigenvector of A with eigenvalue r. By the Spectral Theorem (as described
earlier), A has an orthogonal eigenbasis, e
1
= 1, e
2
, . . . , e
n
. That is, e
i
e
j
for all
pairs of i, j {1, 2, . . . , n} such that i = j. Thus, 1 e
i
for all i 2. So, for i 2,
we have
Je
i
=
_

_
e
i
1
.
.
.
e
i
1
_

_ = 0.
Multiplying equation (5) by e
i
(i 2) to the right, we obtain
(6) (A
2
+ A (r 1)I)e
i
= Je
i
= 0.
We know that Ae
i
=
i
e
i
, and thus A
2
e
i
=
2
i
e
i
, so equation (6) becomes

i
e
i
+
i
e
i
(r 1)e
i
= 0,
and thus
(7)
2
i
+
i
(r 1) = 0.
Since this is a quadratic equation, we know that it has at most 2 roots. Thus, A
must have at most 3 eigenvalues; lets call them r,
1
, and
2
with multiplicities 1,
m
1
, and m
2
respectively.

1
and
2
must be roots of equation (7). Thus, we have

1,2
=
1
_
1 + 4(r 1)
2
=
1

4r 3
2
.
Let s =

4r 3. Then, s
2
= 4r 3, or , equivalently,
(8) r =
s
2
+ 3
4
.
REGULAR GRAPHS OF GIVEN GIRTH 11
Since A is an n by n matrix, we have
(9) 1 + m
1
+ m
2
= n.
We also know that the sum of the eigenvalues is equal to the trace of the matrix,
so
(10) r + m
1

1
+ m
2

2
= 0.
Since n = r
2
+ 1, we can rewrite equation (9) as
(11) m
1
+ m
2
= r
2
.
We know that

1,2
=
1 s
2
.
Without loss of generality, let

1
=
1 + s
2
and
2
=
1 s
2
.
Now we can rewrite equation (10) as
r + m
1
1 + s
2
+ m
2
1 s
2
= 0,
or
(12) 2r + s(m
1
m
2
) (m
1
+ m
2
) = 0.
Consider the case where s is irrational. Then m
1
m
2
= 0 (since m
1
, m
2
, and
r are all integers), and we have
2r (m
1
+ m
2
) = 0,
or
2r r
2
= 0,
by equation (11). Since r = 0, this equation implies that r = 2. This number does,
in fact, occur in the theorem as a possible value of r.
From now on, we will assume that s is rational, and thus s is an integer (since
it is the square root of an integer).
Substituting equation (11) into equation (12), we have
2r + s(m
1
m
2
) r
2
= 0.
Plugging in equation (8), we have
2
s
2
+ 3
4
+ s(m
1
m
2
)
(s
2
+ 3)
2
16
= 0.
Simplifying, we obtain
s
4
2s
2
16(m
1
m
2
)s 15 = 0.
Thus, s must be a divisor of 15. That is, s {1, 3, 5, 15}. Thus, by equation
(7), we have r {1, 3, 7, 57}. Including the case where s is irrational, we add 2 to
the possible values of r, which brings us to the conclusion that r {1, 2, 3, 7, 57}.

References
[1] Babai, L aszl o. Lectures and Discrete Math Lecture Notes.
[2] Biggs, Norman. Algebraic Graph Theory. 2nd ed. Cambridge: Cambridge UP, 1993.