DIHEDRAL GROUPS II

KEITH CONRAD

We will characterize dihedral groups in terms of generators and relations, and describe
the subgroups of Dn , including the normal subgroups. We will also introduce an infinite
group that resembles the dihedral groups and has all of them as quotient groups.

1. Abstract characterization of Dn
The group Dn has two generators r and s with orders n and 2 such that srs−1 = r−1 . We
will show any group with a pair of generators like r and s (except for their order) admits a
homomorphism onto it from Dn , and is isomorphic to Dn if it has the same size as Dn .
Theorem 1.1. Let G be generated by elements a and b where an = 1 for some n ≥ 3,
b2 = 1, and bab−1 = a−1 . There is a surjective homomorphism Dn → G, and if G has order
2n then this homomorphism is an isomorphism.
The hypotheses an = 1 and b2 = 1 do not mean a has order n and b has order 2, but only
that their orders divide n and divide 2. For instance, the trivial group has the form ha, bi
where an = 1, b2 = 1, and bab−1 = a−1 (take a and b to be the identity). When n = 4, the
group (Z/(8))× has generators 3 and 5 with 34 = 1, 52 = 1, and 5 · 3 · 5−1 = 3 = 3−1 .
Proof. The equation bab−1 = a−1 implies baj b−1 = a−j for any j ∈ Z (raise both sides to
the jth power). Since b2 = 1, we have for any k ∈ Z
kj
bk aj b−k = a(−1)
by considering even and odd k separately. Thus
k
(1.1) bk aj = a(−1) j bk .
Thus, any product of a’s and b’s can have all the a’s brought to the left and all the b’s
brought to the right. Taking into account that an = 1 and b2 = 1, we get
G = ha, bi
= {aj , aj b : j ∈ Z}
(1.2) = {1, a, a2 , . . . , an−1 , b, ab, a2 b, . . . , an−1 b}.
Thus G is a finite group with #G ≤ 2n.
To write down an explicit homomorphism from Dn onto G, the equations an = 1, b2 = 1,
and bab−1 = a−1 suggest we should be able send r to a and s to b by a homomorphism.
This suggests the function f : Dn → G defined by
f (rj sk ) = aj bk .
This function makes sense, since the only ambiguity in writing an element of Dn as rj sk is
that j can change modulo n and k can change modulo 2, which has no effect on the right
side since an = 1 and b2 = 1.
1
2 KEITH CONRAD

To check f is a homomorphism, we use (1.1):

0 0 0 0
f (rj sk )f (rj sk ) = aj bk aj bk
k 0 0
= aj a(−1) j bk bk
k 0 0
= aj+(−1) j bk+k
and
0 0 k j0 0
f ((rj sk )(rj sk )) = f (rj r(−1) sk sk )
k j0 0
= f (rj+(−1) sk+k )
k j0 0
= aj+(−1) bk+k .
The results agree, so f is a homomorphism from Dn to G. It is onto since every element of
G has the form aj bk and these are all values of f by the definition of f .
If #G = 2n then surjectivity of f implies injectivity, so f is an isomorphism. 
Remark 1.2. The homomorphism f : Dn → G constructed in the proof is the only one
where f (r) = a and f (s) = b: if there is any such homomorphism then f (rj sk ) =
f (r)j f (s)k = aj bk . So a more precise formulation of Theorem 1.1 is this: for any group
G = ha, bi where an = 1 for some n ≥ 3, b2 = 1, and bab−1 = a−1 , there is a unique homo-
morphism Dn → G sending r to a and s to b. Mathematicians describe this state of affairs
by saying Dn with its generators r and s is “universal” as a group with two generators
satisfying the three equations in Theorem 1.1.
As an application of Theorem 1.1, we can write down a matrix group over Z/(n) that is
isomorphic to Dn when n ≥ 3. Set
  
±1 c
(1.3) Dn =
e : c ∈ Z/(n)
0 1
inside GL2 (Z/(n)). The group D e n has order 2n (since 1 6≡ −1 mod n for n ≥ 3). Inside
De n , ( −1 0 ) has order 2 and ( 1 1 ) has order n. A typical element of D
e n is
0 1 01
    
±1 c 1 c ±1 0
=
0 1 0 1 0 1
 c  
1 1 ±1 0
= ,
0 1 0 1
so ( 10 11 ) and ( −1 0 11 1 1 −1 are conjugate by ( −1 1 ):
0 1 ) generate Dn . Moreover, ( 0 1 ) and ( 0 1 )
e
0 1
   −1  
−1 0 1 1 −1 0 1 −1
=
0 1 0 1 0 1 0 1
 −1
1 1
= .
0 1
Thus, by Theorem 1.1, D e n is isomorphic to Dn , using ( 1 1 ) in the role of r and ( −1 0 ) in
01 0 1
the role of s.
This realization of Dn inside GL2 (Z/(n)) should not be confused with the geometric
cos(2π/n) − sin(2π/n)
realization of Dn in GL2 (R) using real matrices: r = ( sin(2π/n) cos(2π/n) ) and s = ( 10 −1
0 ).

Corollary 1.3. If n ≥ 6 is twice an odd number then Dn ∼

= Dn/2 × Z/(2).
 DIHEDRAL GROUPS II 3

= D3 × Z/(2) and D10 ∼

For example, D6 ∼ = D5 × Z/(2).
Proof. Let H = hr2 , si, where r and s are taken from Dn . Then (r2 )n/2 = 1, s2 = 1, and
sr2 s−1 = r−2 , so Theorem 1.1 tells us there is a surjective homomorphism Dn/2 → H. Since
r2 has order n/2, #H = 2(n/2) = n = #Dn/2 , so Dn/2 ∼ = H.
n/2
Set Z = {1, r }, the center of Dn . The elements of H commute with the elements of
Z, so the function f : H × Z → Dn by f (h, z) = hz is a homomorphism. Writing n = 2k
where k = 2` + 1 is odd, we get f ((r2 )−` , rn/2 ) = r−2`+k = r and f (s, 1) = s, so the image
of f contains hr, si = Dn . Thus f is surjective. Both H × Z and Dn have the same size, so
f is injective too and thus is an isomorphism. 
There is no isomorphism as in Corollary 1.3 between Dn and Dn/2 × Z/(2) when n is
divisible by 4: since n and n/2 are even the center of Dn has order 2 and the center of
Dn/2 × Z/(2) has order 2 · 2 = 4. Therefore the centers of Dn and Dn/2 × Z/(2) are not
isomorphic, so these groups are not isomorphic.

2. Dihedral groups and generating elements of order 2

In Dn , we can obtain r from s and rs (just multiply: rs · s = rs2 = r), so we can use the
reflections rs and s as generators for Dn :
Dn = hr, si = hrs, si.
In group-theoretic terms, Dn is generated by two elements of order 2. They do not commute:
rs · s = r and s · rs = srs = r−1 ss = r−1 .
What finite groups besides Dn for n ≥ 3 can be generated by two elements of order 2?
Suppose G = hx, yi, where x2 = 1 and y 2 = 1. If x and y commute, then G = {1, x, y, xy}.
This has size 4 provided x 6= y. Then we see G behaves just like the group Z/(2) × Z/(2),
where x corresponds to (1, 0) and y corresponds to (0, 1). If x = y, then G = {1, x} = hxi
is cyclic of size 2. If x and y do not commute, it turns out that G is essentially a dihedral
group!
Theorem 2.1. Let G be a finite non-abelian group generated by two elements of order 2.
Then G is isomorphic to a dihedral group.
Proof. Let the two elements be x and y, so each has order 2 and G = hx, yi. Since G is
non-abelian and x and y generate G, x and y do not commute: xy 6= yx.
The product xy has some finite order, since we are told that G is a finite group. Let the
order of xy be denoted n. Set a = xy and b = y. (If we secretly expect x is like rs and y is
like s in Dn , then this choice of a and b is understandable, since it makes a look like r and
b look like s.) Then G = hx, yi = hxy, yi is generated by a and b, where an = 1 and b2 = 1.
Since a has order n, n | #G. Since b 6∈ hai, #G > n, so #G ≥ 2n.
The order n of a is greater than 2. Indeed, if n ≤ 2 then a2 = 1, so xyxy = 1. Since x
and y have order 2, we get
xy = y −1 x−1 = yx,
but x and y do not commute. Therefore n ≥ 3. Since
(2.1) bab−1 = yxyy = yx, a−1 = y −1 x−1 = yx,
where the last equation is due to x and y having order 2, we obtain bab−1 = a−1 . By
Theorem 1.1, there is a surjective homomorphism Dn → G, so #G ≤ 2n. We saw before
that #G ≥ 2n, so #G = 2n and G ∼ = Dn . 
4 KEITH CONRAD

Theorem 2.1 says we know all the finite non-abelian groups generated by two elements of
order 2. What about the finite abelian groups generated by two elements of order 2? We
discussed this before Theorem 2.1. Such a group is isomorphic to Z/(2) × Z/(2) or (in the
degenerate case that the two generators are the same element) to Z/(2). So we can define
new dihedral groups
D1 = Z/(2), D2 = Z/(2) × Z/(2).
In terms of generators, D1 = hr, si where r = 1 and s has order 2, and D2 = hr, si where r
and s have order 2 and they commute. With these definitions,
• #Dn = 2n for every n ≥ 1,
• the dihedral groups are precisely the finite groups generated by two elements of
order 2,
• the description of the commutators in Dn for n > 2 (namely, they are the powers of
r2 ) is true for n ≥ 1 (commutators are trivial in D1 and D2 , and so is r2 in these
cases),
• for even n ≥ 1, Corollary 1.3 is true when n is twice an odd number (including
n = 2) and false when n is a multiple of 4,
• the model for Dn as a subgroup of GL2 (R) when n ≥ 3 is valid for all n ≥ 1.
However, D1 and D2 don’t satisfy all properties of dihedral groups when n > 2. For
example,
• Dn is non-abelian for n > 2 but not for n ≤ 2,
• the description of the center of Dn when n > 2 (trivial for odd n and of order 2 for
even n) is false when n ≤ 2,
• the matrix model for Dn over Z/(n) doesn’t work when n ≤ 2,
Remark 2.2. Unlike finite groups generated by two elements of order 2, there is no ele-
mentary description of all finite groups generated by two elements with equal order > 2.
Theorem 2.3. If N is a proper normal subgroup of Dn then Dn /N is a dihedral group.
Therefore any nontrivial homomorphic image of a dihedral group is a dihedral group.
Proof. The group Dn /N is generated by rs and s, which both square to the identity, so
they have order 1 or 2 and they are not both trivial since Dn /N is not trivial. If rs and s
both have order 2 then Dn /N is a dihedral group by Theorem 2.1 if Dn /N if nonabelian
or it is isomorphic to Z/(2) or Z/(2) × Z/(2) if Dn /N is abelian, which are also dihedral
groups by our convention on the meaning of D1 and D2 . If rs or s have order 1 then only
one of them has order 1, which makes Dn /N ∼= Z/(2) = D1 . 
We will see what the proper normal subgroups of Dn are in Theorem 3.8; aside from
subgroups of index 2 (which are normal in any group) they turn out to be the subgroups
of hri.

3. Subgroups of Dn
We will list all subgroups of Dn and then collect them into conjugacy classes of subgroups.
Our results will be valid even for n = 1 and n = 2. Recall D1 = hr, si where r = 1 and s
has order 2, and D2 = hr, si where r and s have order 2 and commute.
Theorem 3.1. Every subgroup of Dn is cyclic or dihedral. A complete listing of the sub-
groups is as follows:
(1) hrd i, where d | n, with index 2d,
 DIHEDRAL GROUPS II 5

(2) hrd , ri si, where d | n and 0 ≤ i ≤ d − 1, with index d.

Every subgroup of Dn occurs exactly once in this listing.
In this theorem, subgroups of the first type are cyclic and subgroups of the second type
are dihedral: hrd i ∼
= Z/(n/d) and hrd , ri si ∼
= Dn/d .
Proof. It is left to the reader to check n = 1 and n = 2 separately. We now assume n ≥ 3.
Let H be a subgroup of Dn . The composite homomorphism H ,→ Dn → Dn /hri to a
group of order 2 is either trivial or onto. Its kernel is H ∩ hri.
If the homomorphism is trivial then H = H ∩ hri, so H ⊂ hri, which means H = hrd i for
a unique d | n. The order of hrd i is n/d and its index in Dn is 2n/(n/d) = 2d.
If the homomorphism H → Dn /hri is onto then H/(H ∩ hri) has order 2, so H ∩ hri has
index 2 in H. Set H ∩ hri = hrd i, so [H : hrd i] = 2. Since hrd i has order n/d, #H = 2n/d
and [Dn : H] = 2n/#H = d. Choosing h ∈ H with h 6∈ hrd i, we know h is not a power of r
since hrd i = H ∩ hri, so h is a reflection. Write h = ri s. Then H contains
n n o
rdk , rdk+i s : 0 ≤ k ≤ − 1 ,
d
which is already 2(n/d) terms, so H = hr , r si. Multiplying ri s by an appropriate power
d i

of rd will produce an rj s where 0 ≤ j ≤ d − 1, and we can replace ri s with this rj s in the

generating set. So we may assume 0 ≤ i ≤ d − 1. The subgroup hrd , ri si is nontrivial and
generated by two elements of order 2 (ri s and rd · ri s), so it is isomorphic to a dihedral
group. Since rd has order n/d, the order of hrd , ri si is 2(n/d) = 2n/d, whose index in Dn
is d.
To check the two lists of subgroups in the theorem have no duplications, first we show
the lists are disjoint. The only dihedral groups that are cyclic are groups of order 2, and
hrd , ri si has order 2 only when d = n. The subgroup hrn , ri si = hri si has order 2 and ri s
is not a power of r, so this subgroup is not on the first list.
The first list of subgroups has no duplications since the order of hrd i changes when we
change d (among positive divisors of n). If the second list of subgroups has a duplication,
say hrd , ri si = hre , rj si, then computing the index in Dn shows d = e. The reflections in
hrd , ri si are all rdk+i s, so rj s = rdk+i s for some k. Therefore j ≡ dk + i mod n, and from
d | n we further get j ≡ i mod d. That forces j = i, since 0 ≤ i, j ≤ d − 1. 
Corollary 3.2. Let n be odd and m | 2n. If m is odd then there are m subgroups of Dn
with index m. If m is even then there is one subgroup of Dn with index m.
Let n be even and m | 2n.
• If m is odd then there are m subgroups of Dn with index m.
• If m is even and m doesn’t divide n then there is one subgroup of Dn with index m.
• If m is even and m | n then there are m + 1 subgroups of Dn with index m.
Proof. Check n = 1 and n = 2 separately first. We now assume n ≥ 3.
If n is odd then the odd divisors of 2n are the divisors of n and the even divisors of 2n
are of the form 2d, where d | n. From the list of subgroups of Dn in Theorem 3.1, any
subgroup with odd index is dihedral and any subgroup with even index is inside hri. A
subgroup with odd index m is hrm , ri si for a unique i from 0 to m − 1, so there are m such
subgroups. The only subgroup with even index m is hrm/2 i by Theorem 3.1.
If n is even and m is an odd divisor of 2n, so m | n, the subgroups of Dn with index m
are hrm , ri si where 0 ≤ i ≤ m − 1. When m is an even divisor of 2n, so (m/2) | n, hrm/2 i
6 KEITH CONRAD

has index m. If m does not divide n then hrm/2 i is the only subgroup of index m. If m
divides n then the other subgroups of index m are hrm , ri si where 0 ≤ i ≤ m − 1. 
From knowledge of all subgroups of Dn we can count conjugacy classes of subgroups.
Theorem 3.3. Let n be odd and m | 2n. If m is odd then all m subgroups of Dn with
index m are conjugate to hrm , si. If m is even then the only subgroup of Dn with index m is
hrm/2 i. In particular, all subgroups of Dn with the same index are conjugate to each other.
Let n be even and m | 2n.
• If m is odd then all m subgroups of Dn with index m are conjugate to hrm , si.
• If m is even and m doesn’t divide n then the only subgroup of Dn with index m is
hrm/2 i.
• If m is even and m | n then any subgroup of Dn with index m is hrm/2 i or is
conjugate to exactly one of hrm , si or hrm , rsi.
In particular, the number of conjugacy classes of subgroups of Dn with index m is 1 when
m is odd, 1 when m is even and m doesn’t divide n, and 3 when m is even and m | n.
Proof. As usual, check n = 1 and n = 2 separately first. We now assume n ≥ 3.
When n is odd and m is odd, m | n and any subgroup of Dn with index m is some
hrm , ri si. Since n is odd, ri s is conjugate to s in Dn . The only conjugates of rm in Dn are
r±m , and any conjugation sending s to ri s turns hrm , si into hr±m , ri si = hrm , ri si. When
n is odd and m is even, the only subgroup of Dn with even index m is hrm/2 i by Theorem
3.1.
If n is even and m is an odd divisor of 2n, so m | n, a subgroup of Dn with index m
is some hrm , ri si where 0 ≤ i ≤ m − 1. Since ri s is conjugate to s or rs (depending on
the parity of i), and the only conjugates of rm are r±m , hrm , ri si is conjugate to hrm , si
or hrm , rsi. Note hrm , si = hrm , rm si and rm s is conjugate to rs (because m is odd), Any
conjugation sending rm s to rs turns hrm , si into hrm , rsi.
When m is an even divisor of 2n, so (m/2) | n, Theorem 3.1 tells us hrm/2 i has index m.
Any other subgroup of index m is hrm , ri si for some i, and this occurs only when m | n, in
which case hrm , ri si is conjugate to one of hrm , si and hrm , rsi. It remains to show hrm , si
and hrm , rsi are nonconjugate subgroups of Dn . Since m is even, the reflections in hrm , si
are of the form ri s with even i and the reflections in hrm , rsi are of the form ri s with odd
i. Therefore no reflection in one of these subgroups has a conjugate in the other subgroup,
so the two subgroups are not conjugate. 
Example 3.4. For odd prime p, the only subgroup of Dp with index 2 is hri and all p
subgroups with index p (hence order 2) are conjugate to hrp , si = hsi.
Example 3.5. In D6 , the subgroups of index 2 are hri, hr2 , si, and hr2 , rsi, which are
nonconjugate to each other. All 3 subgroups of index 3 are conjugate to hr3 , si. The only
subgroup of index 4 is hr2 i. A subgroup of index 6 is hr3 i or is conjugate to hsi or hrsi.
Example 3.6. In D10 the subgroups of index 2 are hri, hr2 , si, and hr2 , rsi, which are
nonconjugate. The only subgroup of index 4 is hr2 i, all 5 subgroups with index 5 are
conjugate to hr5 , si, and a subgroup with index 10 is hr5 i or is conjugate to hr10 , si or
hr10 , rsi.
Example 3.7. When k ≥ 3, the dihedral group D2k has three conjugacy classes of sub-
groups with each index 2, 4, . . . , 2k−1 .
 DIHEDRAL GROUPS II 7

We now classify the normal subgroups of Dn , using a method that does not rely on our
listing of all subgroups or all conjugacy classes of subgroups.
Theorem 3.8. In Dn , every subgroup of hri is a normal subgroup of Dn ; these are the
subgroups hrd i for d | n and have index 2d. This describes all proper normal subgroups of
Dn when n is odd, and the only additional proper normal subgroups when n is even are
hr2 , si and hr2 , rsi with index 2.
In particular, there is at most one normal subgroup per index in Dn except for three
normal subgroups hri, hr2 , si, and hr2 , rsi of index 2 when n is even.
Proof. We leave the cases n = 1 and n = 2 to the reader, and take n ≥ 3.
Since hri is a cyclic normal subgroup of Dn all of its subgroups are normal in Dn , and
by the structure of subgroups of cyclic groups these have the form hrd i where d | n.
It remains to find the proper normal subgroups of Dn that are not inside hri. Any
subgroup of Dn not in hri must contain a reflection.
First suppose n is odd. All the reflections in Dn are conjugate, so a normal subgroup
containing one reflection must contain all n reflections, which is half of Dn . The subgroup
also contains the identity, so its size is over half of the size of Dn , and thus the subgroup is
Dn . So every proper normal subgroup of Dn is contained in hri.
Next suppose n is even. The reflections in Dn fall into two conjugacy classes of size n/2,
represented by r and rs, so a proper normal subgroup N of Dn containing a reflection will
contain half the reflections or all the reflections. A proper subgroup of Dn can’t contain
all the reflections, so N contains exactly n/2 reflections. Since N contains the identity,
|N | > n/2, so [Dn : N ] < (2n)/(n/2) = 4. A reflection in Dn lying outside of N has order 2
in Dn /N , so [Dn : N ] is even. Thus [Dn : N ] = 2, and conversely every subgroup of index
2 is normal. Since Dn /N has order 2 we have r2 ∈ N . The subgroup hr2 i in Dn is normal
with index 4, so the subgroups of index 2 in Dn are obtained by taking the inverse image
in Dn of subgroups of index 2 in Dn /hr2 i = {1, r, s, rs} ∼ = Z/(2) × Z/(2):
• the inverse image of {1, r} is hri,
• the inverse image of {1, s} is hr2 , si,
• the inverse image of {1, rs} is hr2 , rsh.

Example 3.9. For an odd prime p, the only nontrivial proper normal subgroup of Dp is
hri, with index 2.
Example 3.10. In D6 , the normal subgroups of index 2 are hri, hr2 , si, and hr2 , rsi. The
normal subgroup of index 4 is hr2 i and of index 6 is hr3 i. There is no normal subgroup of
index 3.
Example 3.11. The normal subgroups of D10 of index 2 are hri, hr2 , si, and hr2 , rsi. The
normal subgroup of index 4 is hr2 i and of index 10 is hr5 i. There is no normal subgroup of
index 5.
Example 3.12. When k ≥ 3, the dihedral group D2k has one normal subgroup of each
index except for three normal subgroups of index 2.
The “exceptional” normal subgroups hr2 , si and hr2 , rsi in Dn for even n ≥ 4 can be
realized as kernels of explicit homomorphisms Dn → Z/(2). In Dn /hr2 , si we have r2 = 1
and s = 1, so ra sb = ra with a only mattering mod 2. In Dn /hr2 , rsi we have r2 = 1 and
s = r−1 = r, so ra sb = ra+b , with the exponent only mattering mod 2. Therefore two
8 KEITH CONRAD

homomorphisms Dn → Z/(2) are ra sb 7→ a mod 2 and ra sb 7→ a + b mod 2. These functions

are well-defined since n is even and their respective kernels are hr2 , si and hr2 , rsi.
We can also see that these functions are homomorphisms using the general multiplication
rule in Dn :
b
ra sb · rc sd = ra+(−1) c sb+d .
We have a + (−1)b c ≡ a + c mod 2 and a + (−1)b c + b + d ≡ (a + b) + (c + d) mod 2.

4. An infinite dihedral-like group

In Theorem 2.1, the group is assumed to be finite. This finiteness is used in the proof to
be sure that xy has a finite order. It is reasonable to ask if the finiteness assumption can be
removed: after all, could a non-abelian group generated by two elements of order 2 really
be infinite? Yes! In this appendix we construct such a group and show that there is only
one such group up to isomorphism.
Our group will be built out of the linear functions f (x) = ax + b where a = ±1 and
b ∈ Z, with the group law being composition. For instance, the inverse of −x is itself and
the inverse of x + 5 is x − 5. This group is called the affine group over Z and is denoted
Aff(Z). The label “affine” is just a fancy name for “linear function with a constant term.”
In linear algebra, the functions that are called linear all send 0 to 0, so ax + b is not linear
in that sense (unless b = 0). Calling a linear function “affine” avoids any confusion with
the more restricted linear algebra sense of the term “linear function.”
Since polynomials ax + b compose in the same way that the matrices ( a0 1b ) multiply, we
can consider such matrices, with a = ±1 and b ∈ Z, as another model for the group Aff(Z).
We will adopt this matrix model for the practical reason that it is simpler to write down
products and powers with matrices rather than compositions with polynomials.
Theorem 4.1. The group Aff(Z) is generated by ( −1 0 11
0 1 ) and ( 0 1 ).
In the polynomial model for Aff(Z), the two generators in Theorem 4.1 are the functions
−x and x + 1.
Proof. The elements of Aff(Z) have the form
   k
1 k 1 1
(4.1) =
0 1 0 1
or
      `  
−1 ` 1 ` −1 0 1 1 −1 0
(4.2) = = .
0 1 0 1 0 1 0 1 0 1

While ( −1 0 11
0 1 ) has order 2, ( 0 1 ) has infinite order. However, we can also generate Aff(Z)
by two matrices of order 2.
Corollary 4.2. The group Aff(Z) is generated by ( −1 0 −1 −1
0 1 ) and ( 0 1 ), which each have
order 2.
In the polynomial model for Aff(Z), these generators are −x and −x − 1.
Proof. Check ( −1 −1 11
0 1 ) has order 2. By Theorem 4.1, it now suffices to show ( 0 1 ) can
be generated from ( −1 0 −1 −1
0 1 ) and ( 0 1 ). It is their product (taken in the right order!):
−1 0 −1 −1 1 1
( 0 1 )( 0 1 ) = ( 0 1 ). 
 DIHEDRAL GROUPS II 9

Corollary 4.3. The matrices ( −1 0 −1 −1

0 1 ) and ( 0 1 ) are not conjugate in Aff(Z) and do not
commute with a common element of order 2 in Aff(Z).
Proof. Any conjugate of ( −1 0 −1 2b −1 −1
0 1 ) in Aff(Z) has the form ( 0 1 ) for b ∈ Z, and ( 0 1 ) does
−1 0
not have this form. Thus, the matrices are not conjugate. In Aff(Z), ( 0 1 ) commutes only
with the identity and itself. 
Corollary 4.2 shows Aff(Z) is an example of an infinite group generated by two elements
of order 2. Are there other such groups, not isomorphic to Aff(Z)? No.
Theorem 4.4. Any infinite group generated by two elements of order 2 is isomorphic to
Aff(Z).
Proof. Write such a group as G and its two generators of order 2 as x and y. Since G is infi-
nite, x and y do not commute. (Otherwise hx, yi = {1, x, y, xy} has only 4 elements.) Since
x−1 = x and y −1 = y, we do not need to use any exponents on x and y when writing prod-
ucts. The elements of G are strings of x’s and y’s, such as xyyxxyxyxyxyxyxxy. The rela-
tions x2 = 1 and y 2 = 1 let us cancel any pair of adjacent x’s or y’s, so xyyxxyxyxyxyxyxxy
can be simplified to
xyxyxyxyx = (xy)4 x.
Also, the inverse of such a string is again a string of x’s and y’s.
As any element of G can be written as a product of alternating x’s and y’s, there are four
kinds of elements, depending on the starting and ending letter: start with x and end with
y, start with y and end with x, or start and end with the same letter. These four types of
strings can be written as
(4.3) (xy)k , (yx)k , (xy)k x, (yx)k y,
where k is a non-negative integer.
Before we look more closely at these products, let’s indicate how the correspondence
between G and Aff(Z) is going to work out. We want to think of x as ( −1 0
0 1 ) and y as
( −1 −1 −1 0 −1 −1
0 1 ). Therefore the product xy should correspond to ( 0 1 )( 0 1 ) = ( 0 1 ), and in
11

particular have infinite order. Does xy really have infinite order? Yes, because if xy has
finite order, the proof of Theorem 2.1 shows G = hx, yi is a finite group. (The finiteness
hypothesis on the group in the statement of Theorem 2.1 was only used in its proof to show
xy has finite order; granting that xy has finite order, the rest of the proof of Theorem 2.1
shows hx, yi has to be a finite group.)
The proof of Theorem 4.1 shows each element of Aff(Z) is ( 10 11 )k or ( 10 11 )k ( −1 0
0 1 ) for some
k ∈ Z. This suggests we should show each element of G has the form (xy)k or (xy)k x.
Let z = xy, so z −1 = y −1 x−1 = yx. Also xzx−1 = yx, so
(4.4) xzx−1 = z −1 .
The elements in (4.3) have the form z k , z −k , z k x, and z −k y, where k ≥ 0. Therefore elements
of the first and second type are just integral powers of z. Since z −k y = z −k yxx = z −k−1 x,
elements of the third and fourth type are just integral powers of z multiplied on the right
by x.
Now we make a correspondence between Aff(Z) and G = hx, yi, based on the formulas
in (4.1) and (4.2). Let f : Aff(Z) → G by
   
1 k −1 `
f = zk , f = z ` x.
0 1 0 1
10 KEITH CONRAD

This function is onto, since we showed each element of G is a power of z or a power of z

multiplied on the right by x. The function f is one-to-one, since z has infinite order (and,
in particular, no power of z is equal to x, which has order 2). By taking cases, the reader
can check f (AB) = f (A)f (B) for any A and B in Aff(Z). Some cases will need the relation
xz n = z −n x, which follows from raising both sides of (4.4) to the n-th power. 
Remark 4.5. The abstract group hx, yi from this proof is the set of all words in x and y
(like xyxyx) subject only to the relation that any pair of adjacent x’s or adjacent y’s can
be cancelled (e.g., xyxxxy = xyxy). Because the only relation imposed (beyond the group
axioms) is that xx and yy are the identity, this group is called a free group on two elements
of order 2.
Corollary 4.6. Every nontrivial quotient group of Aff(Z) is isomorphic to Aff(Z) or to
Dn for some n ≥ 1.
Proof. Since Aff(Z) is generated by two elements of order 2, any nontrivial quotient group
of Aff(Z) is generated by two elements that have order 1 or 2, and not both have order 1.
If one of the generators has order 1 then the quotient group is isomorphic to Z/(2) = D1 .
If both generators have order 2 then the quotient group is isomorphic to Aff(Z) if it is
infinite, by Theorem 4.4, and it is isomorphic to some Dn if it is finite since the finite
groups generated by two elements of order 2 are the dihedral groups. 
Every dihedral group arises as a quotient of Aff(Z). For n ≥ 3, reducing matrix entries
modulo n gives a homomorphism Aff(Z) → GL2 (Z/(n)) whose image is the matrix group D en
a b
from (1.3), which is isomorphic to Dn . The map ( 0 1 ) 7→ (a, b mod 2) is a homomorphism
from Aff(Z) onto {±1} × Z/(2) ∼ = D2 and the map ( a0 1b ) 7→ a is a homomorphism from
∼
Aff(Z) onto {±1} = D1 . Considering the kernels of these homomorphisms for n ≥ 3, n = 2,
and n = 1 reveals that we can describe all of these maps onto dihedral groups in a uniform
way: for any n ≥ 1, h( 10 n1 )i C Aff(Z) and Aff(Z)/h( 10 n1 )i ∼
= Dn . This common pattern is
another justification for our definition of the dihedral groups D1 and D2 .