Integral Sets and Cayley Graphs of Finite Groups

Roger C. Alperin Brian L. Peterson

Department of Mathematics Department of Mathematics
San Jose State University San Jose State University
San Jose, CA 95192 San Jose, CA 95192
[email protected] [email protected]

Submitted: Jun 12, 2011; Accepted: Feb 7, 2012; Published: Feb 23, 2012
Mathematics Subject Classifications: 05C50, 20C99, 20K01

Abstract
Integral sets of finite groups are discussed and related to the integral Cayley
graphs. The Boolean algebra of integral sets are determined for dihedral group
and finite abelian groups. We characterize the finite abelian groups as those finite
groups where the Boolean algebra generated by integral sets equals the Boolean
algebra generated by its subgroups.

1 Introduction
Integral graphs were first introduced by Harary and Schwenk in 1974 [5]. These are simple
graphs without loops with adjacency matrix having only integer eigenvalues. Since then
the interest continues in finding integral graphs of certain types: trees, cubic graphs,
Cayley graphs; see [1], [2] and their references. The integral Cayley graphs for a cyclic
group were determined in [3]. This result was rediscovered [8]; recently, there is renewed
interest in the case of abelian groups [6], [7].
We abstract some of the properties of integral Cayley graphs of abelian groups so that
we can apply similar methods to any finite group. We introduce the notion of an integral
set. The result then for the general finite abelian group is that the Boolean algebra
generated by integral sets is the same as the Boolean algebras generated by subgroups
(cf. Theorem 5.1). On the other hand for dihedral groups this algebra is the same as the
power set of the group (cf. Theorem 6.1).

2 Boolean Algebra
Suppose X is a set and F is a family of subsets then the Boolean algebra B(F) generated
by F is the lattice of subsets of X obtained by arbitrary finite intersections, unions, and

the electronic journal of combinatorics 19 (2012), #P44 1

complements of the sets in the family F. The minimal non-empty subsets of this algebra
are called the atoms. Each element of B(F) is expressible as a union of the atoms.
Consider the equivalence relation on X: a ≡ b if for every S ∈ F, then a, b ∈ S or
a, b ∈
/ S.

Lemma 2.1 The equivalence classes of this relation are the atoms of B(F)

Proof. If A is an atom then any set S either contains A or is disjoint from A. So elements
of A are equivalent. Conversely if a ∈ X and ā is its equivalence class then whenever it
meets the subset S it is contained in S; hence ā must be a minimal set (an atom). 

Theorem 2.2 The atoms for the B(FG ) where FG is the family of subgroups of the finite
group G are the sets [a] = {b | < b >=< a >}, the generators of the cyclic subgroup
generated by a, for any a ∈ G.

Proof. Let a ∈ G and its equivalence class ā from the relation afforded by FG . By
Lemma 2.1 we see that b ∈ ā iff every subgroup of G containing a also contains b so
< b >=< a >. Hence ā = [a]. 

3 Atomic Numbers
It
P is easy to see
P that Nthe number of atoms of B(FG ) (atomic number ) is Atomic(G) =
1
a∈G φ(|a|) = d
d||G| φ(d) , where Nd is the number of elements in G of order d and φ is
Euler’s totient function. For example for a cyclic group of order n, then there are φ(d)
elements of order d, so Atomic(Zn ) = d|n φ(d)
P
φ(d)
= τ (n), the number of divisors of n.

Lemma 3.1 If n1 = |G1 | and n2 = |G2 | are relatively prime, then

Atomic(G1 × G2 ) = Atomic(G1 )Atomic(G2 ).

Proof. Every divisor n = n1 n2 is uniquely of the form d1 d2 where di |ni , i = 1, 2; an

element (g1 , g2 ) ∈ G1 × G2 has order d = d1 d2 iff order(gi ) = di , i = 1, 2. Thus Nd (G1 ×
G2 ) = Nd1 (G1 )Nd2 (G2 ) and the result follows using the multiplicitivity of the φ function.


3.1 Symmetric Groups

Use the cycle decomposition P of an element of Sn : an element decomposes into ei ≥ 0
cycles of length i; note that ni=1 iei = n. The order of the element is g(e1 , . . . , en ) =
lcm{j : ej > 0, 1 ≤ j ≤ n}; let f (e1 , e2 , . . . , en ) = φ(g(e1 , . . . , en )). Then
X n!
Atomic(Sn ) =
Pn 1e1 e1 !2e2 e2 ! . . . nen en !f (e1 , e2 , . . . , en )
(e1 ,...,en ),ei ≥0, i=1 iei =n

the electronic journal of combinatorics 19 (2012), #P44 2

 For the alternating groups An we get the subsum:
X n!
Atomic(An ) =
Pbn/2c 1e1 e1 !2e2 e2 ! . . . nen en !f (e1 , e2 , . . . , en )
i=1 e2i =0 mod 2

3.2 Dihedral Groups

The dihedral group Dn consists of the cyclic group of order n and n other elements of
order 2. It follows that
Atomic(Dn ) = τ (n) + n.

3.3 Abelian Groups

By using Lemma 3.1 we may restrict to abelian p-groups, p a prime. Consider G =
Zpe1 ⊕ Zpe2 · · · ⊕ Zper , where 1 ≤ e1 ≤ e2 · · · ≤ er . The orders of elements are pf
Npf (G)
and φ(pf ) = pf − pf −1 Therefore we have Atomic(G) = 1 + efr=1 pf −p
P
f −1 . Suppose
e1 ≤ · · · ≤ es < f ≤ es+1 · · · ≤ er . Let s = s(f ). Now an element (g1 , g2 , . . . , gr ) ∈ G
has order pf iff among the components gi all have order ≤ pf and at least one has order
equal to pf . Thus the components in the first s slots are arbitrary, and the last r − s
slots have all element of order ≤ pf and some element of order greater than pf −1 . Then
Npf (G) = (pe1 +e2 +···+es )(p(r−s)f − p(r−s)(f −1) ). This gives the formula
er
X p(r−s(f ))f − p(r−s(f ))(f −1)
Atomic(G) = 1 + pe1 +e2 +···+es(f ) .
f =1
pf − pf −1

For example, Atomic(Zr2 ⊕ Zs4 ⊕ Zt3 ) = Atomic(Zt3 )Atomic(Zr2 ⊕ Zs4 ).

3t + 1
Atomic(Zt3 ) =
2
Atomic(Zr2 ⊕ Zs4 ) = 2r+s−1 (1 + 2s )

4 Boolean Algebra of Integral Sets

Suppose G is a finite group; let Ĝ be the set of characters of representations of G over
the complex numbers. P
For any subset A ⊆ G and any χ ∈ Ĝ let χ(A) = a∈A χ(a). We call A integral
if χ(A) ∈ Z for every χ ∈ Ĝ. Since every character is an integer linear combination of
irreducible characters, it suffices to check integrality of a set using only the irreducible
characters.

Proposition 4.1 Let G be a finite group. Then any atom of B(FG ) is an integral subset
of G.

the electronic journal of combinatorics 19 (2012), #P44 3

Proof. Diagonalize an element a ∈ G using the representation ρ. The element a has order
m so the eigenvalues are m-thP roots ofPunity sayP 1 ≤ j ≤ n, n = ρ(1). The character
ζj , P
n
on the atom [a] has value r,(r,m)=1 j=1 ζj = j r,(r,m)=1 ζjr . Thus the spectrum of
r

[a] for this representation in the j-th component is r,(r,m)=1 ζjr ; let ζj = ζm
s
P
where ζm is
sr
P
a primitive m-th root of unity; so the eigenvalue at the j-th position is z = r,(r,m)=1 ζm .
Hence z is an algebraic integer. Moreover z is invariant under the action of the Galois
group of Q(ζm ) over Q so z is rational. Thus z is an integer and hence any atom is
integral. 

Let B(IG ) denote the Boolean algebra generated by the integral sets; let P(G) denote
the power set of G.

Corollary 4.2 For any finite group G, B(FG ) ⊆ B(IG ) ⊆ P(G)

Corollary 4.3 Any subgroup of G is integral.

Proof. Any subgroup H is a disjoint union of some atoms of B(FG ). Each of these atoms
is integral by Proposition 4.1. Hence H is integral since the union is disjoint. 

Hence B(IG ) has its atoms contained in the atoms of B(FG ).

Theorem 4.4 If H is a subgroup of the finite group G then B(IH ) ⊆ B(IG ).

Proof. If A is an integral set for H then for any χ ∈ Ĝ its restriction to H is a character
of H and so χ(A) ∈ Z. Hence any set in the Boolean algebra generated by integral subsets
of H is a set in the Boolean algebra generated by integral subsets of G. 

Theorem 4.5 If all the irreducible representations of the finite group G are realized over
Q then any non-empty subset of G is integral. Hence B(IG ) = P(G).

Proof. Since every irreducible representation is rational all the characters are integer
valued, so χ(a) ∈ Z for all χ ∈ Ĝ and a ∈ G. Thus every subset of G is integral. Every
singleton set is an atom. 

It is well-known that the irreducible representations of Sn are rational.

Corollary 4.6 For all n ≥ 1, every non-empty subset of Sn is integral, hence B(ISn ) =
P(Sn ).

the electronic journal of combinatorics 19 (2012), #P44 4

4.1 Orthogonality of Characters in Finite Groups
As is well-known the irreducible characters form an orthonormal basis for the class func-
tions on the group G, functions which are constant on conjugacy classes. The inner
product between characters is given by
1 X 1 X
< χ1 , χ2 >G = χ1 (g)χ̄2 (g) = χ1 (g)χ2 (g −1 ).
|G| g∈G |G| g∈G

For any subgroup H ⊆ G, the representations of G restrict to H to give representations

of H and the representations of H induce to representations of G. A relation between
these is captured by the Frobenius Reciprocity Law:

< ψ, Res(χ) >H =< Ind(ψ), χ >G .

Corollary 4.7 Every non-empty subset of Q8 , the quaternion group of order 8, is integral,
hence B(IQ8 ) = P(Q8 ).

Proof. Here are the characters. One can determine this character table knowing the 1
dimensional representations; using orthogonality of the rows (weighted) we can determine
the 2 dimensional character.
There is a central element and the other non-identity conjugacy classes of size 2. There
are four 1 dimensional characters coming from the Z2 ⊕ Z2 quotient of Q8 by its center.
The first row represents the sizes of the conjugacy class.
1 1 2 2 2
χ1 1 1 1 1 1
χ2 1 1 1 -1 -1
χ3 1 1 -1 1 -1
χ4 1 1 -1 -1 1
χ5 2 -2 0 0 0

This group is not a rational group since its rational group algebra contains a copy of the
quaternion skew-field. 

4.2 Dedekind Groups and Integrality

A group in which all subgroups are normal is called a Dedekind group. This set of groups
includes the abelian groups. The non-abelian Dedekind groups are called Hamiltonian
groups; the structure theorem for finite Hamiltonian groups (due to Dedekind) says that
such a group is isomorphic to Q8 × D where D is an abelian group having no elements of
order 4 [4].
We call G a D-group when any non-normal cyclic subgroup of G is of order 2. Any
product of dihedral group Dn , n ≥ 3 is a D-group.
A group is called cyclotomic if B(FG ) = B(IG ).

the electronic journal of combinatorics 19 (2012), #P44 5

Lemma 4.8 A cyclotomic group is a D-group.

Proof. Suppose G has a non-normal cyclic subgroup H generated by an element a of

order greater than 2. By Corollary 4.3, the subgroup H is integral. Without loss of
generality we can assume sas−1 ∈ / H for some s ∈ G. Let K = H − {a} ∪ {sas−1 }.
It is well-known that if χ is a character then χ(a) = χ(sas−1 ). So K is integral. It is
easy to see now that H ∩ K ∩ [a] is also integral so H ∩ K ∩ [a] ∈ B(IG ); thus, from the
hypothesis,H ∩ K ∩ [a] ∈ B(FG ). Since H ∩ K = H − {a} then H ∩ K ∩ [a] is a proper
subset of the atom [a]; this contradicts that [a] is an atom. Thus G is a D-group. 

Theorem 4.9 If a group G is cyclotomic then it is a D-group which does not contain
Q8 , D4 or D3 .

Proof. By Lemma 4.8 G is a D-group. If G contains Q8 then any character restricts to

Q8 to give an integral character of Q8 . By Corollary 4.7 for every a ∈ Q8 , {a} is integral.
By Theorem 4.4 and B(IG ) = B(FG ), every element of Q8 is an atom in B(FG ). This is
a contradiction since Q8 has elements of order 4.
It is easy to see that the character table of D4 is the same as for Q8 . Also the characters
of D3 are integral. Thus by a similar argument any group containing D4 or D3 can not
be cyclotomic. 

Corollary 4.10 Any Dedekind group which is cyclotomic must be abelian.

Proof. A non-abelian Dedekind group contains Q8 . These can not be cyclotomic by

Theorem 4.9.

5 Abelian Groups
Theorem 5.1 For any finite abelian group D every atom of B(ID ) is integral. Hence
every subset of B(ID ) is integral and D is cyclotomic.

Proof. Consider the matrix M = (mi,j ) = (χi (aj )) where aj , j = 1, . . . , n are the
distinct elements of D and χi , i = 1, . . . , n, are the distinct characters, n = |D|. From the
orthonormality of characters, the rows of M are independent. The entries of M belong
to the field Q(ζm ), where ζm is a primitive m-th root of unity, where m is the exponent
of D.
Consider the conjugate transpose M̄ t ; the orthogonality formulas for characters is the
same as M M̄ t = |D|In or M −1 = n1 M̄ t
For each atom [a] ∈ B(FD ) let v[a] be the vector with a 1 in those locations which
are in [a], and 0s elsewhere. Thus M v[a] ∈ Zn since atoms are integral. Let V = {v ∈
Qn |M v ∈ Qn }. Hence W = spanQ {v[a] | all atoms [a]} ⊆ V .
If v ∈ V then M v = w ∈ Qn ; hence v = n1 M̄ t w. Thus vi = n1 j χj¯(ai )wj ∈ Qn .
P

the electronic journal of combinatorics 19 (2012), #P44 6

 Say b = ar , where r is relatively prime to |a|. The Galois group of Q(ζm ) over Q acts
transitively on roots of unity of the same order. Hence there is an automorphism σ taking
χj (a) to χj (b) for every j.

1
PWe can ¯
index the P
1
components of v by the elements of D. Thus va = σ(va ) =
¯
n j σ(χj (a))wj = n j χj (b)wj . Thus the b-component of v is the same as the a-
component of v. Hence v ∈ W ; so V = W .
It is also clear that the set of vectors v[a] , [a] an atom of B(FD ), is a linearly independent
set; hence it is also a basis for W . If A is integral then the vector vA with a 1 in any
component when a ∈ A, and 0s elsewhere satisfies M vA ∈ Qn so vA is a linear combination
of the v[a] . However it is clear that all coefficients in the linear combination are 0 or 1. It
follows that A is a union of atoms for B(FD ). Thus B(ID ) = B(FD ). 

6 Dihedral Groups
6.1 Representations of Dn
Let s be the generator the cyclic subgroup, sn = 1, and t the basic reflection, t2 = (ts)2 =
1.
If n is odd then the abelianization is Z2 so there are two one dimensional representa-
tions and the characters take only integral values.
If n is even then the abelianization is Z2 ⊕ Z2 so there are four one dimensional
representations and the characters take only integral values.
The rest of the irreducible representations are 2-dimensional. Let ρ be the standard
representation of Dn on the regular n-gon. The basic rotation for the generator of the
cyclic subgroup is by 2π/n. The other representations have basic rotations for the gener-
ator by 2kπ/n.
In case n is odd then the representations are different for 1 ≤ k ≤ n−1 2
. Thus there are
a total of n+3
2
irreducible representations, which is the same as the number of conjugacy
i −i
classes. There are 2 conjugacy classes consisting of the pairs {s , s }, 1 ≤ i ≤ n−1
n−1
2
.
and the two singleton classes {1} and {t}.
In case n is even then the representations are different and irreducible for 1 ≤ k ≤ n−2 2
.
n+6
Thus there are a total of 2 irreducible representations, which is the same as the number
of conjugacy classes. There are n−2 2
conjugacy classes consist of the pairs {si , s−i }, 1 ≤
n n
i ≤ n−2
2
. and the four singleton classes {1}, {t}, {s 2 } and {ts 2 }
It is easy to evaluate the characters χk for these two dimensional representations. The
value of χk on any reflection is zero and the value on sl is 2 cos( 2klπn
). If n is even then
the value of the character on the element of order 2 is 2 cos(kπ) = ±2.

6.2 Integral Sets

Theorem 6.1 For the dihedral groups Dn , n ≥ 1, then B(IDn ) = P(Dn ).

the electronic journal of combinatorics 19 (2012), #P44 7

Proof. Since the one dimensional representations are integral we can test integrality
of a set using only the irreducible two dimensional characters. Any atom of B(IDn ) is
contained in an atoms [a] ∈ B(FDn ). We can suppose that the size of [a] is greater than
1.
The Galois group of the totally real subfield of Q(ζm ) over Q, obtained as the invariants
under complex conjugation, acts transitively on cos( 2kπ m
) where gcd(k, m) = 1 and m is a
divisor of n
Also any set B ⊆ [sd ], dm = n, such that B ∩ B −1 = φ and B ∪ B −1 = [sd ] is
also integral since 2 cos( 2kπ
m
) is an algebraic integer of degree φ(m)/2 for gcd(k, m) = 1.
By taking intersections of sets of this type we can obtain any of the singleton sets, sdk ,
gcd(k, m) = 1. Thus any set is in B(IDn ). 

Theorem 6.2 If a group is cyclotomic then it is a D-group which does not contain Q8
or any Dn , n ≥ 3.

Proof. This follows immediately from Theorem 6.1 and the methods of Theorem 4.9. 

7 Spectra of Cayley Graphs

Suppose that A is a subset of the finite group G, which is closed under inversion and
does not contain the identity. P We use this as a set of generators for the Cayley graph
X = X(G, A). Let A = a∈A a, an element of the group algebra CG. It is easy to see
that the adjacency matrix of X is obtained by calculating the matrix of the multiplication
CG → CG, u → Au. Let ρ denote the regular representation of G. The character P of this
represention is then χρ and the trace of the adjacency matrix is χρ (A) = a∈A χρ (a).
The regular representation is the direct sum of χi (1) = ni copies of the i-th irreducible
representation, 1 ≤ i ≤ r.
The trace spectrum of the Cayley graph X(G, A) is the multi-set consisting of χ1 (A),
χ2 (A), . . . , χr (A), each repeated according to its multiplicity n1 , n2 , . . . , nr times,
[χ1 (A)n1 ; χ2 (A)n2 ; . . . ; χr (A)nr ].
For example, the group S3 has irreducible characters are χ1 , χ2 , χ3 ; these are the
characters of the trivial representation, the sign representation and a two dimensional
representation. The characters values on the non-trivial atoms a1 = (12), a2 = (13),
a3 = (23) a4 = (123), a5 = (132), are

χ1 (a1 ) = 1, χ1 (a2 ) = 1, χ1 (a3 ) = 1, χ1 (a4 ) = 2, χ1 (a5 ) = 2

χ2 (a1 ) = −1, χ2 (a2 ) = −1, χ2 (a3 ) = −1, χ2 (a4 ) = 2, χ2 (a5 ) = 2

χ3 (a1 ) = 0, χ3 (a2 ) = 0, χ3 (a3 ) = 0, χ3 (a4 ) = 0, χ3 (a5 ) = 0
There are 16 possible choices for an integral set which does not include the identity and
is closed under inversion. These are obtained by taking Boolean sums of the atomic trace
spectra for the atoms: a1 : [1; −1; 02 ], a2 : [1; −1; 02 ], a3 : [1; −1; 02 ], {a4 , a5 } : [4; 4; 02 ]

the electronic journal of combinatorics 19 (2012), #P44 8

Theorem 7.1 If G is a finite group and the Cayley graph X(G, A) has integer spectrum
then A is an integral set in G.

Proof. Every irreducible representation χi , 1 ≤ i ≤ r, occurs as a sub-representation of

the regular representation, and by hypothesis all the eigenvalues (of the regular represen-
tation) when summed over A are integers; hence also the irreducible sub-representation
has integer eigenvalues, when summed over A; thus all χi (A), 1 ≤ i ≤ r are integers.
Hence A is an integral set. 

The following result confirms a conjecture of [6].

Corollary 7.2 If G is a finite abelian group then the Cayley graph X(G, A) has integral
spectrum iff A is an integral set.

Proof. All the characters of a finite abelian group are one dimensional. The result now
follows from Theorem 5.1. 

7.1 Cubic connected abelian Cayley graphs

Any set A with three non-identity elements will give a cubic Cayley graph; if the set is
integral the possibilities for A (for a cyclotomic group) are:

1. three elements of order 2;

2. an element of order 2, an element of order 3 and its inverse;

3. an element of order 2, an element of order 4 and its inverse;

4. an element of order 2, an element of order 6 and its inverse.

If we also want the graph connected, then this set A will be a generating set for G. If
G is an abelian group then it is one of the following:

1. Z22 , Z32 ;

2. Z6 ;

3. Z4 , Z2 ⊕ Z4 ;

4. Z6 , Z2 ⊕ Z6 .

The classification of non-abelian cubic connected Cayley graphs with integral spectra
is given in [1] ; this can be somewhat streamlined using Theorem 7.1.

the electronic journal of combinatorics 19 (2012), #P44 9

7.2 An integral dihedral graph
Let Dn be the dihedral group having generators a, b with an = 1, b2 = 1, and ba =
a−1 b. Consider the Cayley graph with the integral set A = {a, a2 , . . . , an−1 , b}. Use the
representation of elements as ai or bai , i = 1, . . . , n  
J −I I
The adjacency matrix is given by the matrix which is similar to
  I J − I
J 0
, where J is the matrix of all 1s. Since J is of rank 1, we can easily determine
I J − 2I
the eigenvalues: (−2)n−1 , 0n−1 , n − 2, n

7.3 A non-integral dihedral graph

Let Dn be the dihedral group having generators a, b with a2 = 1, b2 = 1, and (ab)n = 1.
Consider the Cayley graph with the integral set A = {a, b}. Using the representation of
elements as 1, a, ab, aba, . . . , b, ba, bab, . . . it is easy to calculate the adjacency matrix. The
Cayley graph is a circuit of length 2n; it is well-known that the eigenvalues are integral
iff n=1, 2, 3, 4, 6 [1].

8 Cyclotomic Groups are Abelian

Dedekind classified finite groups where any proper subgroup is normal. A consequence
of Dedekind’s theorem is that either the group is abelian or it contains a Q8 subgroup.
So to prove that a cyclotomic finite group is abelian it suffices to show that all proper
subgroups are normal; then since there are no Q8 subgroups the group is abelian. To see
that any subgroup is normal we notice that it suffices to show that cyclic subgroups are
normal: for an element a in the subgroup H then xax−1 ∈< a >⊆ H for any x ∈ G so H
is normal in G.
b b−1
Lemma 8.1 The group Gb =< x, y | x2 = 1, y 2 = 1, xyx = y 1+2 > is not cyclotomic
for b ≥ 3.

Proof. The abelianization of Gb is generated by two independent elements X, Y with

only the relations 2X = 0, 2b−1 Y = 0; it follows that the abelianization of Gb is isomorphic
to Z2 ⊕ Z2b−1 .
b−2 b−2 b−2 b−1
Consider the element u = xy 2 , u2 = xy 2 xy 2 = y2 so u is an element of
order 4 in Gb . Also the image of u in the abelianization is X + 2b−2 Y and therefore every
1-dimensional character of Gb takes values ±1 on u.
Using the character identity
X
|χr (g)|2 = |C(g)|
r

with g = u we get a contribution of 2b on the left side from the 1-dimensional characters.
The element u is not central so the righthand side is exactly 2b and now again from the

the electronic journal of combinatorics 19 (2012), #P44 10

character identity above it follows that a higher dimensional character has value 0 on u.
Thus u is an integral element of order 4 and Gb is not cyclotomic by Theorem 4.4. 

Theorem 8.2 Any cyclotomic 2-group is abelian.

Proof. It suffices by Dedekind’s Theorem to show that the subgroups of order 2 are
normal. Say x is an element of order 2 and y is an element of order 2b , b ≥ 1. In case
b = 1 the subgroup generated by x, y is dihedral so it must be D2 and thus yxy −1 = x
Now suppose that b ≥ 2. We do know that the subgroup generated by y is normal so
2
xyx = y a and so also y = x(xyx)x = xy a x = (xyx)a = y a ; thus a2 = 1 mod 2b . The
case of a = −1 is impossible since then the subgroup generated by x, y is non-abelian
dihedral. The case a = 1 shows that < x > is normal.
If b = 2 then we also know from Theorem 4.9 that any group of order 8 that is
cyclotomic must be abelian.
So now we consider the other two solution a = ±1 + 2b−1 when b ≥ 3 and show that
these also are impossible for a cyclotomic group. Suppose first that a = −1 + 2b−1 , then
xy 2 x = y 2a = y −2 since 2(1 + a) = 0 mod 2b ; this gives a non-abelian dihedral subgroup
generated by x, y 2 of order 2b if b ≥ 3.
Thus we can assume that a = 1 + 2b−1 , b ≥ 3. Consider the groups
b b−1
Gb =< x, y | x2 = 1, y 2 = 1, xyx = y 1+2 >.

This group is not cyclotomic by Lemma 8.1 and hence it can not be contained in G. Thus
we have shown in all cases that G has all subgroups normal, and since it also does not
contain Q8 by Theorem 4.9 it is abelian by Dedekind’s Theorem. 

Theorem 8.3 Cyclotomic groups are abelian.

Proof. For any Sylow p-subgroup S for p odd, its cyclic subgroups are normal by Lemma
4.8; then x < a > x−1 =< a > for any a ∈ S, x ∈ G, so S is normal. Thus also S is
abelian follows from Dedekind’s Theorem since all subgroups are normal.
For p = 2 it follows from Theorem 8.2 the Sylow 2-subgroup must be abelian since
subgroups of cyclotomic groups are cyclotomic. Let T be the subgroup generated by
all elements of order 2 then any 2 of these generate a dihedral group so it is D2 so the
elements commute; thus T is elementary abelian and also normal ; say S is a Sylow 2
subgroup containing T then for any x ∈ G if a is not of order 2, x < a > x−1 ⊆< a >⊆ S
and if a is of order 2 then x < a > x−1 ⊆ T ⊆ S so S is normal.
Since all Sylow subgroups are normal and abelian, it now follows easily the group is
the direct product of its Sylow subgroups and hence the group G is abelian. 

the electronic journal of combinatorics 19 (2012), #P44 11

References
[1] A. Abdollahi, E. Vatandoost, Which Cayley graphs are integral, Electron. J. Combin.
16 (1) #R122, 17 pp., 2009.
[2] Omran Ahmadi, Noga Alon, Ian F. Blake, Igor E. Shparlinski, Graphs with integral
spectrum, Linear Algebra and its Applications, Volume 430, Issue 1, 547-552, 2009
[3] W. G. Bridges, R. A. Mena, Rational circulants with rational spectra and cyclic
strongly regular graphs. Ars Combin. 8, 143-161, 1979.
[4] M. Hall, Theory of Groups, MacMillan, 1959.
[5] F. Harary, A.J. Schwenk, Which graphs have integral spectra?. In: R. Bari and F.
Harary, Editors, Graphs and Combinatorics, Lecture Notes in Mathematics vol. 406,
Springer-Verlag, 45-51, 1974.
[6] Walter Klotz, Torsten Sandor, Integral Cayley graphs over abelian groups, Electron.
J. Combin. 17(1) #R81, 1-13, 2010.
[7] Walter Klotz, Torsten Sander, Integral Cayley Graphs Defined by Greatest Common
Divisors, Electron. J. Combin. 18(1) #R94, 1-15, 2011.
[8] Wasin So, Integral Circulant Graphs, Discrete Math. 306(1), 153-158, 2006.

the electronic journal of combinatorics 19 (2012), #P44 12