CHARACTERS OF FINITE ABELIAN GROUPS

KEITH CONRAD

1. Introduction
The theme we will study is an analogue on finite abelian groups of Fourier analysis on
R. A Fourier series on the real line is the following type of series in sines and cosines:
X X
f (x) = an cos(nx) + bn sin(nx).
n≥0 n≥1

This is 2π-periodic. Since einx = cos(nx) + i sin(nx) and e−inx = cos(nx) − i sin(nx), a
Fourier series can also be written in terms of complex exponentials:
X
f (x) = cn einx ,
n∈Z

where the summation runs over all integers (c0 = a0 , cn = 21 (an − bn i) for n > 0, and
cn = 12 (a|n| + b|n| i) for n < 0). The convenient algebraic property of einx , which is not
shared by sines and cosines, is that it is a group homomorphism from R to the unit circle
S 1 = {z ∈ C : |z| = 1}:
0 0
ein(x+x ) = einx einx .
We now replace the real line R with a finite abelian group. Here is the analogue of the
functions einx .
Definition 1.1. A character of a finite abelian group G is a homomorphism χ : G → S 1 .
We will usually write abstract groups multiplicatively, so χ(g1 g2 ) = χ(g1 )χ(g2 ) and
χ(1) = 1.
Example 1.2. The trivial character of G is the homomorphism 1G defined by 1G (g) = 1
for all g ∈ G.
Example 1.3. Let G be cyclic of order 4 with generator γ. Since γ 4 = 1, a character χ of
G has χ(γ)4 = 1, so χ takes only four possible values at γ, namely 1, −1, i, or −i. Once
χ(γ) is known, the value of χ elsewhere is determined by multiplicativity: χ(γ j ) = χ(γ)j .
So we get four characters, whose values can be placed in a table. See Table 1.

1 γ γ2 γ3
1G 1 1 1 1
χ1 1 −1 1 −1
χ2 1 i −1 −i
χ3 1 −i −1 i
Table 1.

1
2 KEITH CONRAD

When G has size n and g ∈ G, for each character χ of G we have χ(g)n = χ(g n ) = χ(1) =
1, so the values of χ lie among the nth roots of unity in S 1 . More precisely, the order of
χ(g) divides the order of g (which divides |G|).
Characters on finite abelian groups were first studied in number theory, which is a source
of many interesting finite abelian groups. For instance, Dirichlet used characters of the group
(Z/(m))× to prove that when (a, m) = 1 there are infinitely many primes p ≡ a mod m. The
quadratic reciprocity law of elementary number theory is concerned with a deep property
of a particular character, the Legendre symbol. Fourier series on finite abelian groups have
applications in engineering: signal processing (the fast Fourier transform [1, Chap. 9]) and
error-correcting codes [1, Chap. 11].
To provide a context against which our development of characters on finite abelian groups
can be compared, Section 2 discusses classical Fourier analysis on the real line. In Section
3 we will run through some properties of characters of finite abelian groups and introduce
their dual groups. In particular, we will see that a finite abelian group is isomorphic to
its dual group, but not naturally, and it is naturally isomorphic to its double-dual group
(Pontryagin duality). Section 4 uses characters of a finite abelian group to develop a finite
analogue of Fourier series. In Section 5 we use characters to prove a structure theorem for
finite abelian groups. In Section 6 we look at duality on group homomorphisms. Characters
are used in Section 7 to factor the group determinant of a finite abelian group.
Our notation is completely standard, but we make two remarks about it. For a complex-
valued function f (x), the complex-conjugate function is usually denoted f (x) instead of
f (x) to stress that conjugation creates a new function. (We sometimes use the overline
notation also to mean the reduction g into a quotient group.) For n ≥ 1, we write µn for
the group of nth roots of unity in the unit circle S 1 . It is a cyclic group of size n.
Exercises.

1. Make a character table for Z/(2) × Z/(2), with columns labeled by elements of the
group and rows labeled by characters, as in Table 1.
2. Let G be a finite nonabelian simple group. (Examples include An for n ≥ 5.) Show
the only group homomorphism χ : G → S 1 is the trivial map.

2. Classical Fourier analysis

This section on Fourier analysis on R serves as motivation for our later treatment of
finite abelian groups, where there will be no delicate convergence issues (just finite sums!),
so we take a soft approach and sidestep the analytic technicalities that a serious treatment
of Fourier analysis on R would demand.
Fourier analysis for periodic functions on R is based on the functions einx for n ∈ Z.
Every “reasonably nice” function f : R → C that has period 2π can be expanded into a
Fourier series X
f (x) = cn einx ,
n∈Z
where the sum runs over Z and the nth Fourier coefficient cn can be recovered as an integral:
Z 2π
1
(2.1) cn = f (x)e−inx dx.
2π 0
This formula for cn can be explained by replacing f (x) in (2.1) by its Fourier series and in-
tegrating termwise (for “reasonably nice” functions this termwise integration is analytically
 CHARACTERS OF FINITE ABELIAN GROUPS 3

justifiable), using the formula

Z 2π (
1 imx −inx 1, if m = n,
e e dx =
2π 0 0, 6 n.
if m =
An important link between a function f (x) and its Fourier coefficients cn is given by Par-
seval’s formula Z 2π
X
2 1
|cn | = |f (x)|2 dx.
2π 0
n∈Z
Rather than working with functions f : R → C having period 2π, formulas look cleaner
using functions f : R → C having period 1. The basic exponentials become e2πinx and the
Fourier series and coefficients for f are
X Z 1
(2.2) f (x) = cn e 2πinx
, cn = f (x)e−2πinx dx.
n∈Z 0

Parseval’s formula becomes

X Z 1
2
(2.3) |cn | = |f (x)|2 dx.
n∈Z 0

Note cn in (2.2) is not the same as cn in (2.1).

In addition to Fourier series there are Fourier integrals. The Fourier transform of a
function f that decays rapidly at ±∞ is the function fb: R → C defined by the integral
formula Z
f (y) =
b f (x)e−2πixy dx.
R
The analogue of the expansion (2.2) of a periodic function into a Fourier series is the Fourier
inversion formula, which expresses f in terms of its Fourier transform fb:
Z
f (x) = fb(y)e2πixy dy.
R
Define a Hermitian inner product of two functions f1 and f2 from R to C by the integral
Z
hf1 , f2 i = f1 (x)f 2 (x) dx ∈ C,
R
Plancherel’s theorem compares the inner product of two functions and the inner product of
their Fourier transforms:
(2.4) hfb1 , fb2 i = hf1 , f2 i.
In particular, when f1 = f2 = f the result is
Z Z
2
|f (y)| dy =
b |f (x)|2 dx,
R R
which is called Parseval’s formula since it is an analogue of (2.3).
The convolution of two functions f1 and f2 from R to C is a new function from R to C
defined by Z
(f1 ∗ f2 )(x) = f1 (t)f2 (x − t) dt
R
and the Fourier transform turns this convolution into pointwise multiplication:
1 ∗ f2 (y) = f1 (y)f2 (y).
f\ b b
4 KEITH CONRAD

2
Example 2.1. √ A Gaussian is a function of the form ae−bx , where b > 0. For example, the
−(1/2)x 2
Gaussian (1/ 2π)e is important in probability theory. The Fourier transform of a
Gaussian is another Gaussian and the convolution of two Gaussians is another Gaussian:
Z r
−bx2 −2πixy π −π2 y2 /b
(2.5) ae e dx = ae
R b
and
r
2 2 π 2
f1 (x) = e−b1 x , f2 (x) = e−b2 x =⇒ (f1 ∗ f2 )(x) = e−(b1 b2 /(b1 +b2 ))x .
b1 + b2
The formula (2.5) says that a highly peaked Gaussian (large b) has a Fourier transform
that is a spread out Gaussian (small π 2 /b) and vice versa. More generally, a function and
its Fourier transform can’t both be highly localized; this is a mathematical incarnation of
Heisenberg’s uncertainty principle from physics.
2
When b = π, (2.5) tells us that ae−πx is its own Fourier transform. Functions equal to
2
their Fourier transform are called self-dual, and e−πx is the simplest nonzero example.
A link between Fourier series and Fourier integrals is the Poisson summation formula:
for a “nice” function f : R → C that decays rapidly enough at ±∞,
X X
(2.6) f (n) = fb(n),
n∈Z n∈Z
2
where fb(y) = R f (x)e−2πixy dx. For example, when f (x) = e−bx (with b > 0), the Poisson
R
summation formula says
r
X π
−bn2 2 2
X
e = e−π n /b .
b
n∈Z n∈Z

To prove the Poisson summation formula, we use Fourier series. Periodize f (x) as
X
F (x) = f (x + n).
n∈Z

Since F (x + 1) = F (x), write F as a Fourier series: F (x) = n∈Z cn e2πinx . Then

P

Z 1
cn = F (x)e−2πinx dx
0
Z 1 X !
= f (x + m) e−2πinx dx
0 m∈Z
XZ 1
= f (x + m)e−2πinx dx
m∈Z 0
XZ m+1
= f (x)e−2πinx dx
m∈Z m
Z
= f (x)e−2πinx dx
R
= fb(n).
 CHARACTERS OF FINITE ABELIAN GROUPS 5

Therefore the expansion of F (x) into a Fourier series is equivalent to

X X
(2.7) f (x + n) = fb(n)e2πinx ,
n∈Z n∈Z

which becomes the Poisson summation formula (2.6) by setting x = 0.

If we replace a sum over Z with a sum over a one-dimensional lattice L = aZ in R, where
a 6= 0, the Poisson summation formula becomes
X 1 X b
f (λ) = f (µ),
|a| ⊥
λ∈L µ∈L

where L⊥ = (1/a)Z is the “dual lattice”:

L⊥ = {µ ∈ R : e2πiλµ = 1 for all λ ∈ L}.
For example, Z⊥ = Z.
There are several conventions for the definition of the Fourier transform as well as the
inner product and convolution of functions. Tables 2 and 3 collect a number of different 2π-
conventions. The first two columns of Tables 2 and 3 are definitions and the other columns
are theorems.

fb(y) hf1 , f2 i f (x) hfb1 , fb2 i

f (x)e−2πixy dx 2πixy dy
R R R
f1 (x)f 2 (x) dx R f (y)e
b hf1 , f2 i
RR RR
f (x)e−ixy 1 ixy dy
R
R dx R f1 (x)f 2 (x) dx 2π R f (y)e
b 2πhf1 , f2 i
1 −ixy 1 √1 fb(y)eixy dy
R R R
√
2π R
f (x)e dx √
2π R 1
f (x)f 2 (x) dx 2π R
hf1 , f2 i
Table 2.

fb(y) (f1 ∗ f2 )(x) 1 ∗ f2 (y)

f\
−2πixy
R R
R f (x)e dx R f1 (y)f2 (x − y) dy f1 (y)fb2 (y)
b
−ixy dx
R R
R f (x)e R f1 (y)f2 (x − y) dy
fb1 (y)fb2 (y)
√1 f (x)e−ixy dx √1
R R
2π R
f (t)f2 (x − t) dt fb1 (y)fb2 (y)
2π R 1
Table 3.

√1 −ixy
R
When the Fourier transform is defined using fb(y) = 2π R f (x)e dx, the function
e−πx2 is no longer self-dual, but e −(1/2)x2
is self-dual. You need to know how the Fourier
transform is defined to say that a particular function is self-dual.
Exercises.

1. Without dwelling on analytic subtleties, check from Fourier inversion that fb(x) =
b
f (−x) (if the Fourier transform is defined suitably).
2. If f is a real-valued even function, show its Fourier transform is also real-valued and
even (assuming the Fourier transform of f is meaningful).
6 KEITH CONRAD

3. For a function f : R → C and c ∈ R, let g(x) = f (x + c). Define the Fourier

h(y) = R h(x)e−2πixy dx. If f has a Fourier transform,
R
transform of a function h by b
show g has Fourier transform gb(y) = e2πicy fb(y).
4. The Poisson summation formula over R was obtained by setting x = 0 in (2.7).
Conversely, show that (2.7) for the function f follows from the Poisson summation
formula for the function g(t) = f (t + x).
5. Assuming the Fourier inversion formula holds for a definition of the Fourier trans-
form as in Table 2, check that for all α and β in R× that if we set
Z
(Ff )(y) = α f (x)e−iβxy dx
R
for all x then Z
β
f (x) = (Ff )(y)eiβxy dy.
2πα R
(If β = 2πα2 then these two equations are symmetric in the roles of f and Ff except
for a sign in the exponential term.) Considering Ff to be the Fourier transform of
2
f , show e−(1/2)βx is self-dual.

3. Finite Abelian Group Characters

We leave the real line and turn to the setting of finite abelian groups G. Our interest shifts
from the functions einx to characters: homomorphisms from G → S 1 . The construction of
characters of these groups begins with the case of cyclic groups.
Theorem 3.1. Let G be a finite cyclic group of size n with a chosen generator γ. There
are exactly n characters of G, each determined by sending γ to the different nth roots of
unity in C.
Proof. We mimic Example 1.3, where G is cyclic of size 4. Since γ generates G, a character
is determined by its value on γ and that value must be an nth root of unity (not necessarily
of exact order n, e.g., 1G (γ) = 1), so there are at most n characters. We now write down
n characters.
Let ζ be an nth root of unity in C. Set χ(γ j ) = ζ j for j ∈ Z. This formula is well-defined
(if γ j = γ k for two different integer exponents j and k, we have j ≡ k mod n so ζ j = ζ k ),
and χ is a homomorphism. Of course χ depends on ζ. As ζ changes, we get different
characters (their values at γ are changing), so in total we have n characters. 
To handle characters of non-cyclic groups, the following lemma is critical.
Lemma 3.2. Let G be a finite abelian group, H ⊂ G a proper subgroup, and χ : H → S 1 a
character of H. For g ∈ G − H, there is an extension of χ to a character on hH, gi.
Proof. We want to extend χ to a character χ e of hH, gi.
What is a possible value for χe(g)? Since g 6∈ H, χ e(g) is not initially defined. But some
nonzero power of g is in H (e.g., g |G| = 1 ∈ H), and on these powers χ is defined. Pick
d ≥ 1 minimal with g d ∈ H. That is, d is the order of g in G/H, so d = [hH, gi : H]. If
there is a character χe on h, H, g that extends χ on H then χ e(g) must be an d-th root of
χ(g d ) since we must have χ
e(g)d = χ e(g d ) = χ(g d ). That is our clue: define χe(g) ∈ S 1 to be
d d
a solution to z = χ(g ):
(3.1) e(g)d = χ(g d ).
χ
 CHARACTERS OF FINITE ABELIAN GROUPS 7

Once we have chosen χ e on hH, gi by

e(g) to satisfy (3.1), define χ
(3.2) e(hg i ) := χ(h)e
χ χ(g)i .
This formula covers all possible elements of hH, gi, but is χ e well-defined? Perhaps H and
hgi overlap nontrivially, so the expression of an element of hH, gi in the form hg i is not
unique. We have to show this doesn’t lead to an inconsistency in the value of χ e in (3.2).
0 0
Suppose hg i = h0 g i . Then g i−i ∈ H, so i0 ≡ i mod d since d is the order of g in G/H.
0 0
Write i0 = i + dd0 , so h = h0 ai −i = h0 g dd . The terms h, h0 , and g d are in H, so
0 0
χ(h0 )e
χ(g)i = χ(h0 )e
χ(g)i χ
e(g)dd
0
= χ(h0 )e
χ(g)i χ(g d )d since χ
e(g)d = χ(g d )
0
= χ(h0 g dd )e
χ(g)i
χ(g)i .
= χ(h)e
e : hH, gi → S 1 is a well-defined function and it is easily checked to be a homo-
Therefore χ
morphism. It restricts to χ on H. 
Theorem 3.3. For a finite abelian group G and subgroup H, each character of H can be
extended to a character of G.
Proof. Let χ : H → S 1 be a character of H.
Since G is finite, it has a finite generating set {g1 , . . . , gk } (e.g., {gi } could be a listing
of all the elements of G). Therefore we can build up a tower of subgroups from H to G by
adjoining the elements gi one at a time:
H ⊂ hH, g1 i ⊂ hH, g1 , g2 i ⊂ · · · ⊂ hH, g1 , . . . , gk i = G.
Each step along this tower has the form Hi ⊂ hHi , gi i, where H0 = H. By applying Lemma
3.2 at each step of the tower, χ can be extended as a character from H to H1 to H2 , and
so on up to Hk = G. 
Let’s refine this to count the number of extensions of a character from H to G.
Theorem 3.4. For a finite abelian group G and subgroup H, each character of H can be
extended to a character of G in [G : H] ways.
Proof. We will induct on the index [G : H]. The result is clear when [G : H] = 1, i.e.,
H = G, so suppose [G : H] > 1 and the theorem is proved for characters on subgroups of
index smaller than [G : H].
Pick g ∈ G with g 6∈ H, so
H ⊂ hH, gi ⊂ G.
1
To extend a character χ : H → S to G, we at least need to be able to extend χ to a character
e on hH, gi. Let’s count the number of ways to do that. Then we will use induction to
χ
count the number of extension of each character from hH, gi all the way up to G.
Let d be the smallest positive integer such that g d ∈ H. An extension of χ on H to
a character χe on hH, gi is determined by χe(g), and this value has to satisfy the condition
e(g)d = χ(g d ). Each number in S 1 has d different d-th roots in S 1 , so there are d potential
χ
values for χ
e(g). The proof of Lemma 3.2 shows all of them really work.
The number of choices of χ e extending χ is the number of choices for χ e(g), which is
d = [hH, gi : H]. Since [G : hH, gi] < [G : H], by induction on the index there are
[G : hH, gi] extensions of each χe to a character of G, so the number of extensions of a
character on H to a character on G is [G : hH, gi][hH, gi : H] = [G : H]. 
8 KEITH CONRAD

Theorem 3.5. If g 6= 1 in a finite abelian group G then χ(g) 6= 1 for some character χ of
G. The number of characters of G is |G|.
Proof. The cyclic group hgi is nontrivial, say of size n, so n > 1. In S 1 there is a cyclic
subgroup of order n, namely the group µn of n-th roots of unity. There is an isomorphism
hgi ∼
= µn , which can be viewed as a character of hgi. By Theorem 3.3, this character of hgi
extends to a character of G and does not send g to 1.
To show G has |G| characters, apply Theorem 3.4 with H the trivial subgroup. 
We have used two important features of S 1 as the target group for characters: for each
d ≥ 1 the dth power map on S 1 is d-to-1 (proof of Theorem 3.4) and for each n ≥ 1 there
is a cyclic subgroup of order n in S 1 (proof of Theorem 3.5).
Corollary 3.6. If G is a finite abelian group and g1 6= g2 in G then there is a character of
G that takes different values at g1 and g2 .
Proof. Apply Theorem 3.5 to g = g1 g2−1 . 
Corollary 3.6 shows the characters of G “separate” the elements of G: different elements
of the group admit a character taking different values on them.
Corollary 3.7. If G is a finite abelian group and H ⊂ G is a subgroup and g ∈ G with
g 6∈ H then there is a character of G that is trivial on H and not equal to 1 at g.
Proof. We work in the group G/H, where g 6= 1. By Theorem 3.5 there is a character of
G/H that is not 1 at g. Composing this character with the reduction map G → G/H yields
a character of G that is trivial on H and not equal to 1 at g. 
It is easy to find functions on G that separate elements without using characters. For
g ∈ G, define δg : G → {0, 1} by
(
1, if x = g,
(3.3) δg (x) =
0, if x 6= g.
These functions separate elements of the group, but characters do this too and have better
algebraic properties: they are group homomorphisms.
Remark 3.8. Nowhere in the proof of Theorem 3.4 did we use the finiteness of G. What
mattered was finiteness of [G : H]. Infinite abelian groups like Z or Zn can contain finite-
index subgroups, so it’s worth noting that we really proved that for each abelian group G,
a character on a finite-index subgroup H extends in [G : H] ways to a character on G.
This result for finite-index subgroups of infinite G has applications to Hecke characters in
algebraic number theory.
Using Zorn’s lemma (the axiom of choice), not only the finiteness of |G| but also the
finiteness of [G : H] can be removed: a character of a subgroup H of an abelian group G
can be extended to a character of G (but the counting aspect with [G : H] is no longer
meaningful). In particular, Corollaries 3.6 and 3.7 are true for all abelian groups G.
Our definition of a character makes sense on nonabelian groups, but there will not be
enough such characters for Theorem 3.5 to hold if G is finite and nonabelian: a homomor-
phism χ : G → S 1 must equal 1 on the commutator subgroup [G, G], which is a nontrivial
subgroup, so such homomorphisms can’t distinguish elements in [G, G] from each other. If
g 6∈ [G, G] then in the finite abelian group G/[G, G] the coset of g is nontrivial so there is a
 CHARACTERS OF FINITE ABELIAN GROUPS 9

character G/[G, G] → S 1 that’s nontrivial on g. Composing this character with the reduc-
tion map G →TG/[G, G] produces a homomorphism G → S 1 that is nontrivial on g. There-
fore [G, G] = χ ker χ, where the intersection runs over all homomorphisms χ : G → S 1 .
This gives a “natural” explanation of why the commutator subgroup is normal in terms of
kernels of homomorphisms: kernels are normal and the intersection of normal subgroups is
normal. We put the word natural in quotes because appealing to the group G/[G, G] in
part of the argument means we had to use the normality of [G, G] anyway. (Using Zorn’s
lemma as in Remark 3.8, the intersection formula for [G, G] applies to all groups, not just
finite groups.)
Definition 3.9. For a character χ on a finite abelian group G, the conjugate character is
the function χ : G → S 1 given by χ(g) := χ(g).
Since each complex number z with |z| = 1 has z = 1/z, χ(g) = χ(g)−1 = χ(g −1 ).
Definition 3.10. The dual group, or character group, of a finite abelian group G is the
set of homomorphisms G → S 1 with the group law of pointwise multiplication of functions:
(χψ)(g) = χ(g)ψ(g). The dual group of G is denoted G.
b

The trivial character of G is the identity in G

b and the inverse of a character is its conjugate
character. Note G b is abelian since multiplication in C× is commutative.
Theorem 3.5 says in part that
(3.4) |G| = |G|.
b
In fact, the groups G and G
b are isomorphic. First let’s check this on cyclic groups.

Theorem 3.11. If G is cyclic then G ∼

=Gb as groups.

Proof. We will show G

b is cyclic. Then since G and G
b have the same size they are isomorphic.
Let n = |G| and γ be a generator of G. Set χ : G → S 1 by χ(γ j ) = e2πij/n for all j. For
b we have ψ(γ) = e2πik/n for some integer k, so ψ(γ) = χ(γ)k . Then
other characters ψ ∈ G,
ψ(γ j ) = ψ(γ)j = χ(γ)jk = χ(γ j )k ,
which shows ψ = χk . Therefore χ generates G.
b 
Lemma 3.12. If A and B are finite abelian groups, there is an isomorphism A
\ ×B ∼
= A×
b B.
b
Proof. Let χ be a character on A × B. Identify the subgroups A × {1} and {1} × B of
A × B with A and B in the obvious way. Let χA and χB be the restrictions of χ to A and
B respectively, i.e., χA (a) = χ(a, 1) and χB (b) = χ(1, b). Then χA and χB are characters
of A and B and χ(a, b) = χ((a, 1)(1, b)) = χ(a, 1)χ(1, b) = χA (a)χB (b). So we get a map
(3.5) A
\ ×B →A
b×B
b
by sending χ to (χA , χB ). It is left to the reader to check (3.5) is a group homomorphism.
Its kernel is trivial since if χA and χB are trivial characters then χ(a, b) = χA (a)χB (b) = 1,
so χ is trivial. Both sides of (3.5) have the same size by (3.4), so (3.5) is an isomorphism. 
Theorem 3.13. If G is a finite abelian group then G is isomorphic to G.
b
Proof. The case when G is cyclic was Theorem 3.11. Lemma 3.12 extends easily to several
factors in a direct product:
(3.6) (H1 × · · · × Hr ) b ∼
=Hb1 × · · · × H
br .
10 KEITH CONRAD

When Hi is cyclic, H bi ∼
= Hi , so (3.6) tells us that that dual group of H1 × · · · × Hr is
isomorphic to H1 × · · · × Hr . Every finite abelian group is isomorphic to a direct product
of cyclic groups, so the dual group of a finite abelian group is isomorphic to itself. 
Although G and G b are isomorphic groups, there is not a natural isomorphism between
them, even when G is cyclic. For instance, to prove G ∼ = Gb when G is cyclic we had to
choose a generator. If we change the generator, then the isomorphism changes.1
The double-dual group G b is the dual group of G.
b Since G and G b are isomorphic, G and
b

G
b are isomorphic. However, while there isn’t a natural isomorphism from G to G, b there is
b

a natural isomorphism from G to G. b The point is that there is a natural way to map G to
b
its double-dual group: associate to each g ∈ G the function “evaluate at g,” which is the
function Gb → S 1 given by χ 7→ χ(g). Here g is fixed and χ varies. This is a character of G,
b
since (χ1 χ2 )(g) = χ1 (g)χ2 (g) by definition.

Theorem 3.14. Let G be a finite abelian group. The homomorphism G → G

b associating
b
to g ∈ G the function “evaluate at g” is an isomorphism.
Proof. Since a finite abelian group and its dual group have the same size, a group and its
double-dual group have the same size, so it suffices to show this homomorphism is injective.
If g ∈ G is in the kernel then every element of Gb is 1 at g, so g = 1 by Theorem 3.5. 
Theorem 3.14 is called Pontryagin duality. This label actually applies to a more general
result about characters of locally compact abelian groups. Finite abelian groups are a
special case, where difficult analytic techniques can be replaced by counting arguments.
The isomorphism between G and its double-dual group given by Pontryagin duality lets us
think about every finite abelian group G as a dual group (namely the dual group of G).b
The isomorphism in Pontryagin duality is natural: it does not depend on ad hoc choices
(unlike the isomorphism between a finite abelian group and its dual group).
To illustrate Pontryagin duality, consider the following theorem.
Theorem 3.15. Let G be a finite abelian group and m ∈ Z.
a) For g ∈ G, g m = 1 if and only if χ(g) = 1 for every χ ∈ G
b that is an mth power in
G.
b
b) For g ∈ G, g is an mth power in G if and only if χ(g) = 1 for every χ ∈ G
b satisfying
m
χ = 1G .
Proof. a) If g m = 1 and χ = ψ m for some ψ ∈ G
b then
χ(g) = ψ m (g) = ψ(g)m = ψ(g m ) = ψ(1) = 1.
Conversely, suppose χ(g) = 1 whenever χ = ψ m for some ψ ∈ G. b Then for all ψ ∈ G
b we
m m m
have 1 = ψ (g) = ψ(g ), so g = 1 by Theorem 3.5.
b) If g = xm for some x ∈ G then for every χ ∈ G
b such that χm = 1G we have
χ(g) = χ(x)m = χm (x) = 1.
Conversely, assume χ(g) = 1 for all χ such that χm = 1G . Such χ are identically 1 on the
subgroup Gm of mth powers in G. Conversely, every character of G that is trivial on the
1If G is trivial or of order 2, then it has a unique generator, so in that case we could say the isomorphism
G∼
=Gb is canonical.
 CHARACTERS OF FINITE ABELIAN GROUPS 11

subgroup Gm has mth power 1G (why?). Therefore χ(g) = 1 for all χ in G

b that are trivial
m m
on G , so g ∈ G by Corollary 3.7. 
Since Theorem 3.15 is a theorem about all finite abelian groups, by Pontryagin duality
we can swap the roles of G and G
b in the theorem. Part a is equivalent to
χm = 1G ⇐⇒ χ(g) = 1 for every g ∈ G such that g = xm for some x ∈ G
and part b is equivalent to
b ⇐⇒ χ(g) = 1 for all g ∈ G such that g m = 1.
χ is an mth power in G

Exercises.

1. Let’s find the characters of the additive group (Z/(m))r , an r-fold direct product.
(a) For k ∈ Z/(m), let χk : Z/(m) → S 1 by
χk (j) = e2πijk/m ,
so χk (1) = e2πik/m . Show χ0 , χ1 , . . . , χm−1 are all the characters of Z/(m) and
χk χl = χk+l .
(b) Let r ≥ 1. For r-tuples a, b in (Z/(m))r , let
a · b = a1 b1 + · · · + ar br ∈ Z/(m)
be the usual dot product. For k ∈ (Z/(m))r , let χk (j) = e2πi(j·k)/m . Show the
functions χk are all the characters of (Z/(m))r and χk χl = χk+l .
2. Show the following are equivalent properties of a character χ: χ(g) = ±1 for all g,
χ(g) = χ(g) for all g, and χ2 = 1G .
3. Describe the error in the following bogus proof of Theorem 3.4. Let m = [G : H]
and pick a set of coset representatives g1 , . . . , gm for G/H. Given a character χ on
H, define χ
e on G by first picking the m (= [G : H]) values χ e(gi ) for 1 ≤ i ≤ m and
then writing each g ∈ G in the (unique) form gi h and defining χ e(g) = χe(gi )χ(h).
This defines χ
b on G, and since we had to make m choices there are m characters.
4. Let G be a finite abelian group of order n and g ∈ G have order m. Show
Y
(1 − χ(g)T ) = (1 − T m )n/m .
χ∈G
b

5. For finite nonabelian G, show the characters of G (that is, homomorphisms G → S 1 )

separate elements modulo [G, G]: χ(g1 ) = χ(g2 ) for all χ if and only if g1 = g2 in
G/[G, G].
6. This exercise will give an interpretation of characters as eigenvectors. For a finite
abelian group G and g ∈ G, let Tg : L(G) → L(G) by (Tg f )(x) = f (gx).
(a) Show the Tg ’s are commuting linear transformations and each character of G
is an eigenvector of each Tg .
(b) If f is a simultaneous eigenvector of all the Tg ’s, show f (1) 6= 0 (if f (1) = 0
conclude f is identically zero, but the zero vector is not an eigenvector) and then
after rescaling f so f (1) = 1 deduce that f is a character of G. Thus the characters
of G are the simultaneous eigenvectors of the Tg ’s, suitably normalized.
(c) Show the Tg ’s are each diagonalizable. Deduce from this and parts (a) and
(b) that Gb is a basis of L(G), so |G|
b = dim L(G) = |G|. (This gives a different proof
that G and G have the same size.)
b
12 KEITH CONRAD

7. For a subgroup H of a finite abelian group G, let

H ⊥ = {χ ∈ G
b : χ = 1 on H}.

These are the characters of G that are trivial on H. For example, G⊥ = {1G } and
{1}⊥ = G.b Note H ⊥ ⊂ G b and H ⊥ depends on H and G.
Show H ⊥ is a subgroup of G, [ and G/(H
b it is isomorphic to G/H, ⊥ ) ∼ H.
b = b In
⊥
particular, |H| = [G : H].
8. Let G be finite abelian and H ⊂ G be a subgroup.
(a) Viewing H ⊥⊥ = (H ⊥ )⊥ in G using Pontryagin duality, show H ⊥⊥ = H.
(Hint: The inclusion in one direction is easy. Count sizes for the other inclusion.)
(b) Show for each m dividing |G| that
|{H ⊂ G : |H| = m}| = |{H ⊂ G : [G : H] = m}|
by associating H to H ⊥ and using a (fixed) isomorphism of G with G. b
(c) For a finite abelian group G, part b says the number of subgroups of G with
index 2 is equal to the number of elements of G with order 2. Use this idea to count
the number of subgroups of (Z/(m))× with index 2. (The answer depends on the
number of odd prime factors of m and the highest power of 2 dividing m.)
(d) Show, for a prime p, that the number of subspaces of (Z/(p))n with dimension
d equals the number of subspaces with dimension n − d.
9. For a finite abelian group G, let G[n] = {g ∈ G : g n = 1} and Gn = {g n : g ∈ G}.
Both are subgroups of G. Prove G[n]⊥ = (G) b n and (Gn )⊥ = G[n]
b in G.
b

4. Finite Fourier series

We will introduce the analogue of Fourier series on finite abelian groups.
Let G be a finite abelian group. Set
L(G) = {f : G → C},
the C-valued functions on G. This is a C-vector space of functions. Every f ∈ L(G) can
be expressed as a linear combination of the delta-functions δg from (3.3):
X
(4.1) f= f (g)δg .
g∈G

Indeed, evaluate both sides at each x ∈ G and we get the P same value. The functions δg
span L(G) by (4.1) and they are linearly independent: if g ag δg = 0 then evaluating the
sum at x ∈ G shows ax = 0. Thus the functions δg are a basis of L(G), so dim L(G) = |G|.
The next theorem is the first step leading to an expression for each δg as a linear combi-
nation of characters of G, which will lead to a Fourier series expansion of f . It is the first
time we add character values.
Theorem 4.1. Let G be a finite abelian group. Then
( (
X |G|, if χ = 1G , X |G|, if g = 1,
χ(g) = χ(g) =
g∈G
0, if χ 6= 1G , 0, 6 1.
if g =
b χ∈G
P
Proof. Let S = g∈G χ(g). If χ is trivial on G then S = |G|. If χ is not trivial on G, say
P P
χ(g0 ) 6= 1. Then χ(g0 )S = g∈G χ(gg0 ) = g∈G χ(g) = S, so S = 0.
 CHARACTERS OF FINITE ABELIAN GROUPS 13

The second formula in the theorem can be viewed as an instance of the first formula via
Pontryagin duality: the second sum is a sum of the character “evaluate at g” over the group
G,
b and this character on G b is nontrivial when g 6= 1 by Pontryagin duality. 
Theorem 4.1 says the sum of a nontrivial character over a group vanishes and the sum
of all characters of a group evaluated at a nontrivial element vanishes, so the sum of the
elements in each row and column of a character table of G is zero except the row for the
trivial character and the column for the identity element. Check this in Table 1.
Corollary 4.2. For characters χ1 and χ2 in G
b and g1 and g2 in G,
( (
X |G|, if χ1 = χ2 , X |G|, if g1 = g2 ,
χ1 (g)χ2 (g) = χ(g1 )χ(g2 ) =
g∈G
0, 6 χ2 ,
if χ1 = 0, 6 g2 .
if g1 =
b χ∈G

Proof. In the first equation of Theorem 4.1 let χ = χ1 χ2 . In the second equation of Theorem
4.1 let g = g1 g2−1 . (Alternatively, after proving the first equation for all G we observe that
the second equation is a special case of the first by Pontryagin duality.) 
The equations in Corollary 4.2 are called the orthogonality relations. They say that
the character table of G has orthogonal rows and orthogonal columns when we define
orthogonality of two n-tuples of complex numbers P as vanishing of their Hermitian inner
product: in Cn : h(z1 , . . . , zn ), (w1 , . . . , wn )i := nk=1 zk wk .
Example 4.3. Let G = (Z/(m))× . For a ∈ (Z/(m))× and p a prime not dividing m,
(
1 X 1, if p ≡ a mod m,
χ(a)χ(p) =
ϕ(m) 0, if p 6≡ a mod m,
χ mod m

where the sum runs over the characters of (Z/(m))× . (Since p is prime, p not dividing m
forces p to be in (Z/(m))× .) This identity was used by Dirichlet in his proof that there are
infinitely many primes p ≡ a mod m.
By the second equation in Corollary 4.2 we can express the delta-functions in terms of
characters:
X 1 X
χ(g)χ(x) = |G|δg (x) =⇒ δg (x) = χ(g)χ(x).
|G|
χ∈G
b χ∈G
b

Substituting this formula for δg into (4.1) gives

 
X 1 X
f (x) = f (g)  χ(g)χ(x)
|G|
g∈G χ∈G
b
XX 1
= f (g)χ(g)χ(x)
|G|
b g∈G
χ∈G
X
(4.2) = cχ χ(x),
χ∈G
b

where
1 X
(4.3) cχ = f (g)χ(g).
|G|
g∈G
14 KEITH CONRAD

The expansion (4.2) is the Fourier series for f .

Equation (4.3) is similar to the formula for the coefficient cn of einx in (2.1): an integral
over [0, 2π] divided by 2π is replaced by a sum over G divided by |G| and f (x)e−inx is
replaced by f (g)χ(g). The number e−inx is the conjugate of einx , which is also the relation
between χ(g) and χ(g). Equation (4.2) shows G b is a spanning set for L(G). Since |G| b =
|G| = dim L(G), G b is a basis for L(G).

Definition 4.4. If f ∈ L(G) then its Fourier transform fb ∈ L(G)

b is given by
X
fb(χ) = f (g)χ(g).
g∈G

By (4.2) and (4.3),

1 X b
(4.4) f (x) = f (χ)χ(x).
|G|
χ∈G
b

Equation (4.4), called the Fourier inversion formula, lets us recover f from fb.
Remark 4.5. Classically the Fourier transform of a function R → C is another function
R → C. The finite Fourier transform, however, is defined on the dual group instead of on
the original group. We can also interpret the classical Fourier transform to be a function of
characters. For y ∈ R let χy (x) = eixy . Then χy : R → S 1 is a character and fb(y) could be
R
viewed as fb(χy ) = R f (x)χy (x) dx, so fb is a function of characters rather than of numbers.
Example 4.6. Let f = δg . Then fb(χ) = χ(g) = χ(g −1 ). Notice f vanishes at all but one
element of G while fb is nonzero on all of G.
b
P
Example 4.7. Let f = ψ be a character of G. Then fb(χ) = g ψ(g)χ(g) = |G|δψ (χ), so
fb = |G|δψ . Here f is nonzero on all of G and fb is nonzero at only one element of G.
b
The Fourier transform on R interchanges highly spread and highly peaked Gaussians.
Examples 4.6 and 4.7 suggest a similar phenomenon in the finite case. The next theorem is
a general result in that direction (a finite version of Heisenberg uncertainty). It is the only
time (outside Appendix B) when we will use inequalities with characters of finite abelian
groups.
Theorem 4.8. Let f : G → C be a function on a finite abelian group G that is not identically
zero. Then
(4.5) | supp f | · | supp fb| ≥ |G|,
where supp denotes the support of a function (the set of points where the function is
nonzero).
Proof. We expand f into a Fourier series and make estimates. Since
X 1
f (x) = fb(χ)χ(x),
|G|
χ∈G
b

we have
X 1 | supp fb|
(4.6) |f (x)| ≤ |fb(χ)| ≤ max |fb(χ)|.
|G| |G| χ∈G
b
χ∈G
b
 CHARACTERS OF FINITE ABELIAN GROUPS 15

By the definition of fb(χ),

X
(4.7) |fb(χ)| ≤ |f (g)|.
g∈G

Let m = maxg∈G |f (g)|, so m > 0 since f is not identically zero. Then (4.7) implies
|fb(χ)| ≤ m| supp f |, and feeding this into (4.6) yields
m| supp f || supp fb|
|f (x)| ≤ .
|G|
Maximizing over all x ∈ G implies m ≤ m| supp f || supp fb|/|G|. Divide both sides by m
and the desired inequality drops out. 
In Examples 4.6 and 4.7, inequality (4.5) is an equality, so Theorem 4.8 is sharp.
Since L(G) is spanned by the characters of G and the delta-functions, linear identities
in L(G) can be verified by checking them on characters or delta-functions. For example,
define a Hermitian inner product on L(G) by the rule
1 X
(4.8) hf1 , f2 iG = f1 (g)f 2 (g) ∈ C.
|G|
g∈G

We will prove Plancherel’s theorem for G:

1 b b
hf1 , f2 iG = hf1 , f2 iGb
|G|
for all f1 and f2 in L(G). (Compare to (2.4).) To check this identity, which is linear in
both f1 and f2 , it suffices to check it when f1 and f2 are characters. By Corollary 4.2, for
characters χ1 and χ2 of G we have
(
1, if χ1 = χ2 ,
hχ1 , χ2 iG =
0, if χ1 6= χ2 .
b = |G|δχ (Example 4.7),
Since χ
(
1 X 1, if χ1 = χ2 ,
hc c2 iGb = |G|hδχ1 , δχ2 iGb =
χ1 , χ δχ1 (χ)δ χ2 (χ) =
|G| 0, 6 χ2 .
if χ1 =
χ∈G
b

This verifies Plancherel’s theorem for G. The special case where f1 = f2 = f is a single
function from G to C gives us Parseval’s formula for G:
X 1 X b
(4.9) |f (g)|2 = |f (χ)|2 .
|G|
g∈G χ∈G
b

Let’s look at Fourier transforms for functions on a cyclic group. By writing a cyclic
group in the form Z/(m), we can make an isomorphism with the dual group explicit: every
character of Z/(m) has the form χk : j 7→ e2πijk/m for a unique k ∈ Z/(m) (Exercise 3.1).
The Fourier transform of a function f : Z/(m) → C can be viewed as a function on Z/(m):
X X
(4.10) fb(k) := f (j)χk (j) = f (j)e−2πijk/m .
j∈Z/(m) j∈Z/(m)

This is like viewing the Fourier transform of a function on R as a function on R.

16 KEITH CONRAD

Example 4.9. Let f : Z/(8) → C be a function with period 2 having values 1 and 2. See
Table 4. The Fourier transform of f vanishes except at 0 and 4, which are the multiples of
the frequency of f (how often the period repeats).

n 0 1 2 3 4 5 6 7
f (n) 1 2 1 2 1 2 1 2
fb(n) 12 0 0 0 −4 0 0 0
Table 4.

Example 4.10. Let f : Z/(8) → C have the periodic values 5, 3, 1, and 1. Both f and its
Fourier transform are in Table 5. Now f has frequency 2 (its period repeats twice) and the
Fourier transform vanishes except at 0, 2, 4, and 6, which are multiples of the frequency.

n 0 1 2 3 4 5 6 7
f (n) 5 3 1 1 5 3 1 1
f (n) 20 0 8 + 4i 0 4 0 8 − 4i 0
b
Table 5.

Example 4.11. Consider a function f : Z/(45) → C with the four successive repeating
values 1, 8, 19, 17 starting with f (0) = 1. It is not a periodic function on Z/(45) since 4
does not divide 45, but the sequence 1, 8, 19, 17 repeats nearly 11 times. (The value of
f (44) is 1.) A calculation of |fb(n)|, the absolute value of the Fourier transform of f , reveals
sharp peaks at n = 0, 11, 22, 23, and 34. See a plot of |fb(n)| below.

0 11 22 23 34
The red peaks are cut off because the lowest red bar would be around three times as tall as
the highest black bar. Peaks in |fb(n)| occur approximately at multiples of the approximate
frequency!
 CHARACTERS OF FINITE ABELIAN GROUPS 17

As Examples 4.9 and 4.10 suggest, the Fourier transform of a periodic function on Z/(m)
knows the frequency of the original function by the positions where the Fourier transform
has nonzero values (Exercise 4.4). For nearly periodic functions on Z/(m), the approximate
frequency is reflected in where the Fourier transform takes on its largest values. This idea
is used in Shor’s quantum algorithm for integer factorization [2], [3, Chap. √
17], where it is
convenient to redefine the Fourier transform (4.10) by dividing the sum by m, which has
the effect of making the Fourier transform a unitary operator on the functions Z/(m) → C.
See Exercise 4.10.
Exercises.

1. Let G be a finite abelian group, H be a subgroup of G, and K be a subgroup of G.

b
Show
( (
X |H|, if χ is trivial on H, X |K|, if χ(g) = 1 for all χ ∈ K,
χ(h) = χ(g) =
h∈H
0, otherwise, χ∈K
0, otherwise.

2. Let f : Z/(8) → C take the four values a, b, c, and d twice in this order. Compute
fb(n) explicitly and determine some values for a, b, c, and d such that fb(n) is nonzero
for n = 0, 2, and 6, but fb(4) = 0.
3. For a subgroup H of a finite abelian group G, let δH be the function that is 1 on
H and 0 off of H. Show the Fourier transform of δH is δc H = |H|δH ⊥ . How do the
supports of δH and its Fourier transform compare with the inequality (4.5)?
4. Let H be a subgroup of a finite abelian group G.
(a) Suppose f : G → C is constant on H-cosets (it is H-periodic). For χ ∈ G b with
⊥
χ 6∈ H , show fb(χ) = 0. Thus the Fourier transform of an H-periodic function on
G is supported on H ⊥ .
(b) If f : Z/(m) → C has period d where d | m, show fb: Z/(m) → C is supported
on the multiples of m/d. (See Examples 4.9 and 4.10.)
5. Find the analogue of Exercise 2.3 for functions on a finite abelian group.
6. Let G be a finite abelian group and H be a subgroup. For a function f : G → C,
Poisson summation on G says
1 X 1 X b
f (h) = f (χ),
|H| |G| ⊥
h∈H χ∈H

where H ⊥ is as in Exercise 3.7. Prove this formula in two ways:

P a) Copy the classical proof sketched in Section 2 (start with the function F (x) =
h∈H f (xh), which is H-periodic so it defines a function on G/H) to obtain
1 X 1 X b
(4.11) f (xh) = f (χ)χ(x)
|H| |G| ⊥
h∈H χ∈H

for all x ∈ G and then set x = 1.

b) By linearity in f of both sides of the desired identity, verify Poisson summation
directly on the delta-functions of G. (Corollary 3.7 and Example 4.6 will be useful.)
7. Let Tg : L(G) → L(G) be as in Exercise 3.6.
(a) Show Tdg f = χ(g)f for all f ∈ L(G).
b
(b) Show for all f1 and f2 in L(G) that hTg f1 , Tg f2 iG = hf1 , f2 iG .
18 KEITH CONRAD

8. Let f ∈ L(G), so fb is in L(G)

b and fbb is in L(G).
b
b
(a) Viewing G b(g) = |G|f (g −1 ).
b as G by Pontryagin duality, show fb
b
(b) For a subgroup H in G, define the H-average of f and the H-cutoff of f to
be the following functions on G:
(
1 X f (g), if g ∈ H,
AvgH (f )(g) = f (gh), CutH (f )(g) =
|H| 0, if g 6∈ H.
h∈H

Check the identity Avg \

H (f ) = CutH ⊥ (f ) of functions on G and then take the
b b
Fourier transform of both sides to get an identity of functions on G, which will be
(4.11) as x varies. (This shows that Poisson summation is essentially equivalent
to the fact that the Fourier transform exchanges the operators AvgH on L(G) and
CutH ⊥ on L(G),
b or equivalently CutH on L(G) and AvgH ⊥ on L(G).) b
9. Let G be a finite abelian group. For f1 and f2 in L(G), define their convolution
f1 ∗ f2 : G → C by
X
(f1 ∗ f2 )(g) = f1 (h)f2 (gh−1 ).
h∈G
(a) Show δg ∗ δh = δgh , so L(G) under convolution is a commutative C-algebra
isomorphic to the group ring C[G].
(b) Show f\1 ∗ f2 (χ) = f1 (χ)f2 (χ), so the Fourier transform turns convolution into
b b
pointwise multiplication.
(c) Show δg ∗ f = Tg−1 (f ) and χ ∗ f = fb(χ)χ in two ways: by a direct calculation
or by computing the Fourier transform of both sides and using (b).
(d) For each χ ∈ G, b the function hχ : L(G) → C given by hχ (f ) = fb(χ) is a
C-algebra homomorphism by (b). Does every C-algebra homomorphism from L(G)
to C arise in this way?
10. On every finite
p abelian group G, rescale the definition of the Fourier transform by
dividing by |G|:
1 X
fb(χ) := p f (g)χ(g).
|G| g∈G
Verify the following new versions of Fourier inversion and Plancherel’s theorem:
f (g) = √1
P b
χ f (χ)χ(g) and hf1 , f2 iG = hf1 , f2 iG
b b b.
|G|
Check that this new Fourier transform sends convolution on L(G) p (Exercise 4.9)
to multiplication only if we redefine convolution using division by |G|:
1 X
(f1 ∗ f2 )(g) := p f1 (h)f2 (gh−1 ).
|G| h∈G

5. Structure of finite abelian groups

We will now put characters to work by using them to prove that every finite abelian
group is a direct product of cyclic groups. This result was already used in the proof of
Theorem 3.13, that G ∼ = G,
b but that work will not be used here, so no circular reasoning
will occur.
The following theorem shows that every cyclic subgroup of maximal size in a finite abelian
group can be split off as a direct factor. Characters get used in an essential way in the proof.
 CHARACTERS OF FINITE ABELIAN GROUPS 19

Theorem 5.1. Let G be a finite abelian group and let g ∈ G have maximal order in G.
There is a subgroup H ⊂ G such that G ∼
= H × hgi.
Proof. Let n be the order of g. The subgroup hgi of G is cyclic of order n. In S 1 there
is a cyclic subgroup of order n, namely µn . Since cyclic groups of the same order are
isomorphic, there is an isomorphism hgi → µn , so g is mapped to a root of unity of order
n. This isomorphism can be viewed as a character of hgi. Extend this to a character of G
(Theorem 3.3), so we have a character χ : G → S 1 such that χ(g) has order n. The image
χ(G) contains µn , and it turns out to be no larger.
Claim: χ(G) = µn .
Since χ(G) is a finite subgroup of S 1 , it is cyclic (all finite subgroups of S 1 are cyclic).
Therefore χ(G) = hχ(γ)i for some γ ∈ G. Since χ(G) contains µn , χ(G) = µnn0 where
n0 ≥ 1. Thus χ(γ) has order nn0 . Let γ have order d in G, so γ d = 1 in G and thus
χ(γ)d = 1 in S 1 . That implies nn0 | d, so nn0 ≤ d. Since n is the maximal order of the
elements in G, d ≤ n. The relations nn0 ≤ d and d ≤ n imply n0 = 1, so χ(G) = µn . This
proves the claim.
Set H = ker χ. Then H ∩ hgi = {1} since χ is one-to-one on hgi by construction. For each
x ∈ G, χ(x) ∈ χ(G) = µn = χ(hgi), so χ(x) = χ(g j ) for some j. Therefore h := xg −j is in
H and x = hg j . This proves that the multiplication map H × hgi → G where (h, g j ) 7→ hg j
is surjective. It is a homomorphism and its kernel is trivial, so G ∼ = H × hgi. 
Theorem 5.2. Every finite abelian group G is isomorphic to a product of cyclic groups:
G∼= Z/(n1 ) × Z/(n2 ) × · · · × Z/(nk ).
Proof. Induct on |G|. The result is clear when |G| = 1. When |G| > 1, let n be the maximal
order of the elements of G, so G ∼
= H ×Z/(n) by Theorem 5.1. Since |H| < |G|, by induction
H is isomorphic to a direct product of cyclic groups, so G is also isomorphic to a direct
product of cyclic groups. 
Theorem 5.2 can be refined: G is a direct product of cyclic groups with the extra feature
that n1 | n2 | · · · | nk . To prove this, use the fact that the order of each element of a finite
abelian group G divides the maximal order of the elements of G (Theorem A.3).
Exercises.

1. What is the structure (as a direct product of cyclic groups) of the finite abelian
groups whose nontrivial characters all have order 2?
2. Mimic the proof of Theorem 5.1 to decompose (Z/(20))× (of size 8) and (Z/(45))×
(of size 24) into a direct product of cyclic groups.
3. Show by an explicit counterexample that the following is false: if two subgroups H
and K of a finite abelian group G are isomorphic then there is an automorphism of
G that restricts to an isomorphism from H to K.
4. For a finite abelian group G, show the maximum order of the elements of G and
the number |G| have the same prime factors. (Hint: If g has order n and there is
an element h of prime order p where p - n, what is the order of gh?)
This is false in general for nonabelian G, as shown in the table below where g(n)
(called Landau’s function) is the maximal order of an element of Sn . For all n ≥ 3
in the table, some prime factor of n! does not divide g(n).
5. Let G be a finite abelian group and F be a field containing a full set of |G|th roots of
unity. (That is, the equation x|G| = 1 has |G| solutions in F .) Define characters of
20 KEITH CONRAD

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
g(n) 1 2 3 4 6 6 12 15 20 30 30 60 60 84 105
Table 6. Maximal order of elements of Sn

G to be group homomorphisms χ : G → F × and write the set of all such characters

as G.
b
a) Construct a character table for Z/(4) and (Z/(2))2 when F is the field Z/(5).
b) Prove every lemma, theorem, and corollary from Section 3 for the new meaning
of G.
b There is no longer complex conjugation on character values, but the
inverse of χ is still the function g 7→ χ(g −1 ) = χ(g)−1 . (Hint: For each d
dividing |G|, xd = 1 has d distinct solutions in F × , which form a cyclic group.)
c) Prove Theorem 4.1 and Corollary 4.2 for F -valued characters of G.
d) Set L(G, F ) to be the functions G → F . This is an F -vector space in the
same way that L(G) is a complex vector space. For each function f ∈ L(G, F ),
define its Fourier transform fb ∈ L(G,b F ) by fb(χ) = P −1
g∈G f (g)χ(g ). Prove
the Fourier inversion formula and Plancherel’s theorem in this context. (Note:
If the field F has characteristic p then 1/|G| in the Fourier inversion formula
makes sense in F since p doesn’t divide |G| – why?)
e) Check everything you have done goes through if the assumption that x|G| = 1
has a full set of solutions in F is weakened to xm = 1 having a full set of
solutions in F , where m is the maximal order of the elements of G. For example,
if G = (Z/(2))d then m = 2 and we can use F = Z/(3).

6. Dual Homomorphisms
The set Hom(G1 , G2 ) of all homomorphisms from the abelian group G1 to the abelian
group G2 forms an abelian group under pointwise multiplication
Theorem 6.1. Let G1 and G2 be finite abelian groups. For a homomorphism f : G1 → G2 ,
set f ∗ : G b 1 by f ∗ (χ) = χ ◦ f . Then f ∗ is a group homomorphism and the map sending
b2 → G
∗
f to f gives a group isomorphism
Hom(G1 , G2 ) ∼
= Hom(G
b2 , G
b 1 ).
b 2 , then for g and g 0 in G1 we have
Proof. If f : G1 → G2 is a homomorphism and χ ∈ G
χ(f (gg 0 )) = χ(f (g)f (g 0 )) = χ(f (g))χ(f (g 0 )),
so f ∗ (χ) := χ ◦ f lies in Gb 1 . Thus we get the map Hom(G1 , G2 ) → Hom(G b2 , G
b 1 ) as
0 ∗ ∗ 0 ∗ ∗
advertised. Check (f f ) = f (f ) , so f 7→ f is a homomorphism.
Repeating this idea leads to a group homomorphism Hom(G b 1 ) → Hom(G
b2 , G b 1 ,G
b b 2 ). By
b
Pontryagin duality it is a homomorphism Hom(G b 1 ) → Hom(G1 , G2 ) and the composite
b2 , G

Hom(G1 , G2 ) → Hom(G b 1 ) → Hom(G1 , G2 )

b2 , G
turns out to be (after unwinding definitions, left to the reader) the identity function. There-
fore our original map Hom(G1 , G2 ) → Hom(G b2 , G
b 1 ) is a group isomorphism. 

The homomorphism f ∗ : G
b2 → G
b 1 is called the dual homomorphism to f .
 CHARACTERS OF FINITE ABELIAN GROUPS 21

Exercises.

1. For a homomorphism f : G1 → G2 , show (ker f )⊥ = im f ∗ in G

b 1 and (im f )⊥ =
∗
ker f in Gb2 .
2. Show the isomorphism Hom(G, G) ∼ = Hom(G,
b G)
b in Theorem 6.1 coming from the
identity map on G associates g 7→ g in Hom(G, G) with χ 7→ χm in Hom(G,
m b G).
b

7. Abelian group determinants

Consider a square n × n matrix where each row is a cyclic shift of the previous row:
 
X0 X1 X2 . . . Xn−1
 Xn−1 X0 X1 . . . Xn−2 
 
 Xn−2 Xn−1 X0 . . . Xn−3 
(7.1)  .
 .. .. .. . . .. 
 . . . . . 
X1 X2 X3 . . . X0
Its determinant is called a circulant. When n is 2 and 3, the circulants are
 
   X0 X1 X2 
 X0 X1 
 = X02 − X12 and  X2 X0 X1  = X03 + X13 + X23 − 3X0 X1 X2 .


 X1 X0  
 X1 X2 X0 
These factor as
(X0 + X1 )(X0 − X1 ) and (X0 + X1 + X2 )(X0 + ωX1 + ω 2 X2 )(X0 + ω 2 X1 + ωX2 ),
where ω = e2πi/3 .
If we think about the variables Xi as being indexed by Z/(n) then the (i, j) entry of (7.1)
is Xj−i . More generally, for a finite group G, index a set of variables Xg by G and form the
matrix indexed by G × G where the (g, h) entry is Xgh−1 . (The circulant is the determinant
of the matrix (Xj−i ) = (Xi−j )> .) The determinant is called the group determinant of G:
(7.2) ∆(G) = det(Xgh−1 ).
This is a homogeneous polynomial of degree |G| with integer coefficients. A circulant is the
group determinant of a cyclic group.
Circulants of order 2 and 3 are products of linear factors with roots of unity as coefficients.
Dedekind and Burnside each proved the same property for the group determinant of every
finite abelian group, but Dedekind’s approach revealed more structure in the factors: the
roots of unity in a given linear factor are the values of one of the characters of the group!
Theorem 7.1 (Dedekind). If G is a finite abelian group then its group determinant factors
into linear factors over the complex numbers:
 
Y X
det(Xgh−1 ) =  χ(g)Xg  .
χ∈G
b g∈G

Proof. We will realize (Xgh−1 ) as the matrix for a linear transformation and then find its
diagonalization to compute its determinant.
22 KEITH CONRAD

Let V = C[G] be the group ring of G. For each v ∈ V , define the linear map Lv : V → V
to be (left) multiplication by v: Lv (x) =
Pvx. We will compute the matrix for Lv with
respect to the basis G of V . Writing v = g∈G ag g we have for each h ∈ G
X X
Lv (h) = ag gh = agh−1 g,
g∈G g∈G

so the matrix for Lv with respect to the basis G is (aP

gh−1 ).
Another basis for C[G] is the set of formal sums g∈G χ(g)g, one for each character χ
of G: the number of such sums has the right size to be a basis, and for a linear relation
 
X X
cχ  χ(g)g  = 0
χ g∈G

P
in C[G] we get χ cχ χ(g) = 0 for all g (the coefficient of each g is 0), so every cχ is 0 by
Fourier inversion.
This new basis of C[G], indexed by the characters, consists of eigenvectors for Lv :
!   !
X X X
Lv χ(h)h =  ag g  χ(h)h
h∈G g∈G h∈G
 
X X
=  ag χ(h) k
k∈G gh=k
 
X X
=  ag χ(g −1 )χ(k) k
k∈G g∈G
  !
X X
=  ag χ(g −1 ) χ(k)k .
g∈G k∈G

Since det(Lv ) is the product of the eigenvalues of Lv for an eigenbasis,

 
Y X
det(agh−1 ) =  ag χ(g) .
χ∈G
b g∈G

If we interchange the roles of χ and χ in this product then we obtain

 
Y X
det(agh−1 ) =  ag χ(g) .
χ∈G
b g∈G

Q P
Thus the multivariable polynomials det(Xgh−1 ) and χ∈G
b g∈G χ(g)Xg are equal on all of
Cn , so they must be the same polynomial. 
 CHARACTERS OF FINITE ABELIAN GROUPS 23

Example 7.2. Taking G = Z/(n) and ζn = e2πi/n ,

 
 X0 X1 . . . Xn−1 
 n−1
Y n−1
!
 Xn−1 X0 . . . Xn−2  X
..  = ζnjk Xk

 .. .. ..
 .
 . . .  j=0 k=0
 X1 X2 . . . X0 
n−1
Y
= (X0 + ζnj X1 + · · · + ζn(n−1)j Xn−1 ).
j=0

Applications of the factorization of the group determinant for abelian (not necessarily
cyclic) groups can be found in [4, §5.5].
If G is a nonabelian group then its group determinant has an irreducible factor with
degree greater than 1. Studying irreducible factors of the group determinant for nonabelian
G led Frobenius to discover representation theory and the correct extension of the notion
of a character to (finite) nonabelian groups.
Exercises.

1. Check the factorization of the group determinant for Z/(4).

2. Compute and factor the group determinant of Z/(2) × Z/(2). P
3. If G is nonabelian, show the polynomial ∆(G) in (7.2) is divisible by g∈G Xg , and
χ(g)Xg for each homomorphism χ : G → S 1 .
P
more generally by

Appendix A. All orders and maximal order

We will use characters of finite abelian groups to prove that in a finite abelian group G,
the order of each element of G divides the maximal order of the elements of G. Although
this result can proved in a shorter way without characters, the point here is to see characters
used.
Lemma A.1. For positive d and m with d | m, the natural reduction (Z/(m))× → (Z/(d))×
is onto: when (a, d) = 1, there is b such that b ≡ a mod d and (b, m) = 1.

Proof. Let de be the product of the prime powers in m whose primes divide d, so m = dn
e
with (d, n) = 1. (For example, if m = 90 and d = 6 then d = 18 and n = 5.) Then d | d.
e e e
By the Chinese remainder theorem we can find b ∈ Z satisfying
b ≡ a mod d,
e b ≡ 1 mod n.

Then b ≡ a mod d and b is relatively prime to m since it is relatively prime to d (a factor

of d)
e and to n. 
Lemma A.2. Let G be a finite abelian group. If x ∈ G has order m and y ∈ G has order
n then there is a character χ : G → S 1 such that χ(x) has order m and χ(y) has order n.
Proof. The subgroup hxi is cyclic of order m, so there is an isomorphism χ : hxi ∼
= µm . In
particular, χ(x) has order m. Following the proof of Lemma 3.2, we can extend χ to a
character on hx, yi (and then all the way up to G) by sending y to a solution z ∈ S 1 of the
equation z d = χ(y d ), where d ≥ 1 is minimal such that y d ∈ hxi. We will show z can be
picked to have order n in S 1 .
24 KEITH CONRAD

Since y n = 1 ∈ hxi, d divides n. Then y d has order n/d, so χ(y d ) has order n/d
because χ is one-to-one on hxi. Write χ(y d ) = e2πi`/(n/d) = e2πi`d/n , where (`, n/d) = 1.
By Lemma A.1, there is an `0 ≡ ` mod n/d such that (`0 , n) = 1. Since `0 d ≡ `d mod n,
0 0
χ(y d ) = e2πi` d/n . Set z = e2πi` /n , which has order n. Since z k = χ(y d ), we can set
χ(y) = z. 

Lemma A.2 does not extend to more than two arbitrary elements in G. For instance,
if G = µ2 × µ2 then no character on G sends all three non-identity elements in G to −1.
(Why?)
Theorem A.3. Let G be a finite abelian group. The order of each element in G divides
the maximal order of the elements of G.
Proof. We will show that when G has elements of orders m and n, there is an element of
G of order [m, n]. Then if m is the maximal order among all the elements of G, and n is
the order of some element of G, [m, n] is the order of some element so [m, n] ≤ m. That
inequality can only occur when [m, n] = m, so n | m, which is what we want to show.
By Lemma A.2, there is a character χ on G such that χ(g) has order m and χ(h) has
order n. Write χ(g) = e2πia/m and χ(h) = e2πib/n , where (a, m) = 1 and (b, n) = 1.
The roots of unity e2πi/m and e2πi/n are in χ(G). For instance, letting aa0 ≡ 1 mod m,
0 0 0
χ(g a ) = χ(g)a = e2πiaa /m = e2πi/m . The argument for e2πi/n is similar. Write mu + nv =
(m, n) for some integers u and v, so the equation mn = [m, n](m, n) can be rewritten as
1 (m, n) mu + nv 1 1
= = =u +v .
[m, n] mn mn n m
Thus
e2πi/[m,n] = (e2πi/n )u (e2πi/m )v ∈ χ(G),
say e2πi/[m,n] = χ(t). The order of t in G is divisible by the order of χ(t) in S 1 , so t has
order divisible by [m, n]. Thus, raising t to a suitable power, we obtain an element of G
with order [m, n]. 

Appendix B. Functions of Two Variables

When analyzing a function of several variables, it is a common theme to decompose it
into a sum of products of functions of one variable. For instance, to solve a PDE like the
heat equation ∂t u − c∂x2 u = 0, first separable solutions of the form u(x, t) = g(x)h(t) are
classified. It is too much to hope that a general solution is separable, but in nice situations
there are theorems guaranteeing that P a general solution can be written as an infinite series
of separable solutions: u(x, t) = n≥1 gn (x)hn (t). This is where expansions in Fourier
series first appeared in mathematics.
Using characters, and in particular Parseval’s formula, we will give an example of a
function of two variables that is provably not a sum of products of functions of one variable.
Lemma B.1. Fix a positive integer N . For vectors (z1 , . . . , zN ) and (w1 , . . . , wN ) in CN ,
   1/2 !1/2
 N N N
√

X X X
e−2πijk/N zj wk  ≤ N  |zj |2  |wk |2
 

 .
j,k=1  j=1 k=1
 CHARACTERS OF FINITE ABELIAN GROUPS 25

Proof. Write the double sum as an iterated single sum:

N N N N
!
X X X X
−2πijk/N −2πijk/N
e z j wk = e wk zj = fb(j)zj ,
j,k=1 j=1 k=1 j=1

where f : Z/(N ) → C by f (k) = wk . The right side brings in the Fourier transform of f ,
where we think about fb as a function on Z/(N ) by identifying Z/(N ) with its own dual
group as in Exercise 3.1.
Using the Cauchy–Schwarz inequality,
   1/2  1/2
N  N N
X X X
|fb(j)|2   |zj |2  .


 fb(j)zj  ≤ 
 j=1  j=1 j=1

By Parseval’s formula on Z/(N ), j=1 |f (j)| = N k=1 |f (k)|2 = N N

PN b 2 PN P 2
k=1 |wk | . 
Theorem B.2. It is impossible to write
X
e2πixk = gn (x)hn (k),
n≥1

where
P x ∈ [0, 1], k ∈ Z, the functions gn : [0, 1] → C and hn : Z → C are each bounded, and
n≥1 ||gn ||sup ||hn ||sup < ∞, where || · ||sup is the sup-norm on bounded functions.

Proof. Assume there is a series expansion

X
e2πixk = gn (x)hn (k)
n≥1
P
for all x ∈ [0, 1] and k ∈ Z, where c := n≥1 ||gn ||sup ||hn ||sup < ∞. Then the series is
absolutely convergent for all x and k. Pick N ≥ 1 and x1 , . . . , xN ∈ [0, 1]. Then
 
XN N
X X X X N
e−2πijk/N e2πixj k = e−2πijk/N  gn (xj )hn (k) = e−2πijk/N gn (xj )hn (k).
j,k=1 j,k=1 n≥1 n≥1 j,k=1

Then
   
 N   N 
 X −2πijk/N 2πix k  X X −2πijk/N


 e e j 
 = 
 e gn (xj )hn (k)
j,k=1  n≥1 j,k=1 
 
X  X N 
−2πijk/N

≤ 
 e gn (xj )hn (k)
n≥1 j,k=1 
 1/2 !1/2
X√ XN XN
≤ N |gn (xj )|2  |h(k)|2
n≥1 j=1 k=1
PN PN
by Lemma B.1. Since j=1 |gn (xj )|2 ≤ N ||gn ||2sup and k=1 |hn (k)|2 ≤ N ||hn ||2sup ,
 
 N
 X√ √ √

X
−2πijk/N 2πixj k 
N ||hn ||sup ≤ N 3/2 c.


 e e  ≤ N N ||gn ||sup
j,k=1  n≥1
26 KEITH CONRAD

Now set xj = j/N :

N
X N
X
−2πijk/N 2πixj k
e e = e−2πijk/N e2πijk/N = N 2 ,
j,k=1 j,k=1

so N 2 ≤ N 3/2 c for all N ≥ 1. This is false when N is large enough (N > c2 ). 

2πixk
P
This theorem says we can’t write e = n≥1 gn (x)hn (k) where the functions gn and hn
P
are bounded and the series converges absolutely (since convergence of n≥1 ||gn ||sup ||hn ||sup
implies absolute convergence). Could there be such a series representation of e2πixk that is
conditionally convergent?

References
[1] A. Terras, “Fourier Analysis on Finite Groups and Applications,” Cambridge Univ. Press, Cambridge,
1999.
[2] P. Shor, Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum
Computer, [Link]
[3] W. Trappe and L. Washington, “Introduction to Cryptography with Coding Theory,” Prentice-Hall,
Upper Saddle River, NJ 2002.
[4] L. Washington, “Introduction to Cyclotomic Fields,” 2nd ed., Springer-Verlag, New York, 2000.