Pólya-Burnside counting

A. M. Dawes

1 Groups.
1.1 Permutations.
Definition 1.1 A permutation or symmetry of a set X is a 1-1 onto function σ : X →
X. SX denotes the set of all permutations of X; in the special case where X = {1, 2, . . . , n}
we write Sn .

Example. S4 has 4! = 24 elements. We may represent them in function table notation as

! ! !
1 2 3 4 1 2 3 4 1 2 3 4
, , , etc.
1 2 3 4 1 2 4 3 1 3 2 4

If σ and π are permutations of X, the composition π ◦ σ (π followed by σ)! is also a

1 2 3 4
permutation of X. To continue the example above, suppose σ = and π =
2 3 4 1
!
1 2 3 4
. Then
3 4 2 1
! !
1 2 3 4 1 2 3 4
π◦σ = , and σ◦π = .
4 1 3 2 4 2 1 3

Definition 1.2 We use e to denote the identity permutation on a set X (that is, e(x) = x
for all x ∈ X). For σ ∈ SX , the inverse of σ is the unique permutation σ −1 with σ ◦ σ −1 =
σ −1 ◦ σ = e.
!−1 !
−1 1 2 3 4 1 2 3 4
Continuing the example, σ = = .
2 3 4 1 4 1 2 3
Notation. Composition is often denoted by simple juxtaposition, and the value of a per-
mutation π applied to an element x of X is denoted xπ or by π(x). Thus in the example
above, 1π = 3, σ(2) = 3, πσ(2) = 1, and σ −1 (4) = 3.

Definition 1.3 For σ ∈ Sn , and 1 ≤ x ≤ n, the orbit of x under σ consists of the set
{x, σ(x), σ 2 (x), . . .}. When restricted to the orbit of x, σ gives a permutation of that set with
a special structure, called a cycle. More generally, if a1 , a2 , . . . , ar are distinct elements of
{1, 2, . . . , n} then the permutation π given by

π(a1 ) = a2 , π(a2 ) = a3 , . . . , π(ar ) = a1

and π(x) = x for x 6∈ {a1 , . . . , ar }, is a cycle of length r.

1
Such a cycle is written as (a1 a2 . . . ar ); in the example above we write σ = (1234) and
π = (1324). Note that this notation does not explicitly show the value of n; thus (1324)
could also denote an element of S5 , for example. A cycle of length 2 is called a transposition.
To see that the process of repeatedly applying σ to an element x eventually leads back to
x, note that the orbit of x is a finite set (since it’s a subset of {1, 2, . . . , n}). Thus if we
continue applying σ, we eventually find σ i (x) = σ j (x) for some i < j (pigeonhole principle).
Then we use σ −1 to find x = σ j−i (x), as required. For infinite sets X this won’t necessarily
i
work, and in this case the orbit of x under σ is defined to be {σ! (x) : i ∈ Z}.
1 2 3 4 5 6 7
Example. For the permutation σ = the orbit of 1 is {1, 3}; the
3 4 1 7 5 2 6
orbit of 2 is {2, 4, 7, 6}; the orbit of 3 is {3, 1}, etc.. Furthermore note that σ can be
written (13)(2476)(5). Often cycles of length 1 are omitted, giving σ = (13)(2476); such a
composition is called a product. Now observe that the cycle (2476) can be written as a
product of transpositions (24)(27)(26). It follows that σ = (13)(24)(27)(26).

Theorem 1.1 Any element of Sn can be written as a product of disjoint cycles (uniquely
except for order); furthermore any cycle can be written as a product of transpositions. Thus
any permutation can be written as a product of transpositions.

Proof: To write σ ∈ Sn as a product of disjoint cycles, write the orbit of 1 as a cycle;

then choose the least x not in the orbit of 1 and write the orbit of x as a cycle, etc. The
representation is unique up to order because the orbits form a partition of X. For the second
result, note that the cycle (a1 a2 a3 . . . ar ) may be written (a1 a2 )(a1 a3 ) · · · (a1 ar−1 )(a1 ar ).

The decomposition of a permutation into transpositions is not unique; any transposition is

its own inverse, so pairs (ij)(ij) can be inserted anywhere in the product. For a less obvious
example, consider that (2476) can also be written as (7624) = (76)(72)(74). However, there
is one useful result:

Theorem 1.2 If σ ∈ Sn is written as a product of r transpositions, and also as a product

of s transpositions, then (−1)r = (−1)s (that is, r and s are both even or both odd).

Proof: First we introduce some concepts useful here and later. Let I(σ) be the number of
inversions of σ; that is, if σ is written in usual function table notation, I(σ) is the number
of pairs in the second
! row of σ which are reversed from their natural order. For example, for
1 2 3 4
π= , I(π) = 5. We say the parity of σ is even or odd based on the value of
3 4 2 1
I(σ); the example π has odd parity (or more briefly, π is an odd permutation). The parity
of a transposition (ij) is always odd; to see this, we may assume i < j. Then the number
of inversions is 1 (for i and j themselves) plus 2 for each k between i and j (since each such
k will be out of order with both i and j). Furthermore for π, σ ∈ Sn the parity of π ◦ σ is
determined by the parities of π and σ by the usual rule for adding integers. More formally,
(−1)I(π◦σ) = (−1)I(π) · (−1)I(σ) . To prove this, consider the effect of π ◦ σ on some pair i and
j, where i < j. There are four cases:

2
 1. i, j remain in natural order throughout,

2. π leaves i and j in natural order, but σ then inverts them,

3. π inverts i and j, and σ doesn’t change them back,

4. i, j change twice, ending up back in natural order.

Let mq denote the number of times case q occurs as i and j vary (with i < j); then I(π) =
m3 + m4 , I(σ) = m2 + m4 , and I(π ◦ σ) = m2 + m3 . Thus I(π ◦ σ) = I(π) + I(σ) − 2m4 ,
giving the result.
To prove the theorem from these preparations, observe that if π is written as a product of
transpositions, then the number of transpositions must agree with the parity of π.

For practical calculations of parity it is tedious to count inversions; the proof of the theorem
provides a short method if we note that the parity of any cycle of length r is even if r is
odd, odd if r is even. Alternatively, note that if σ is expressed in disjoint cycle notation with
k cycles (counting all cycles of length 1), then the parity of σ is the same as the parity of
n − k.

Example. For σ ∈ S7 as above, with σ = (13)(2476), the parity is odd + odd = even.
Alternatively, σ = (13)(2476)(5), writing in the hidden cycle of length 1, so n− the number
of cycles is 7 − 3 = 4 which is even.

1.2 Exercises.
1. Calculate the products
! !
1 2 3 4 1 2 3 4
(a) ◦ (c) (12345)(234)
4 3 2 1 2 4 3 1
!6
1 2 3 4 5 6 7
(b) (d) (42)(15)(362)
4 7 5 6 3 1 2

In each case express the result as a product of disjoint cycles, then as a product of
transpositions. Find the parity of each of the permutations involved. Also find the
inverse of each permutation. What rule do you observe relating products and inverses?

2. Express (123)(456) as the power of a single cycle in S6 . What is the general result
here?

3. Show that every even permutation can be written as a product of 3-cycles (cycles of
length 3).

3
1.3 Abstract groups.
Definition 1.4 A group (G, ·) is a set G with a binary operation · on G such that

1. · is associative; (a · b) · c = a · (b · c) for all a, b, c ∈ G;

2. there is an identity e ∈ G with e · a = a · e = a for all a ∈ G;

3. for each a in G there is an inverse a−1 in G with a · a−1 = a−1 · a = e.

The definition does not require identity and inverse elements to be unique, but it is easy to
show that this follows from their existence. Our primary example of a group is the set SX
with the operation of composition, and in a certain sense all groups can be considered in
terms of permutations. However our main reason for introducing the above definition is that
we require the notion of a subgroup.

Definition 1.5 If (G, ·) is a group and H is a nonempty subset of G, then (H, ·) is a

subgroup of (G, ·) if

1. for all a, b ∈ H, the product a · b is in H;

2. for all a ∈ H, the inverse a−1 is in H.

Examples. Our work with permutations provides many natural examples:

Cyclic groups. If σ ∈ Sn , then the set {σ, σ 2 , σ 3 , . . .} is a subgroup of Sn . For example,

consider the permutation (123) in S3 . The corresponding subgroup is {(123), (132), e}.
A useful fact becomes apparent here: if H is finite then condition (1) in the definition
is sufficient to guarantee that H is a subgroup. To prove this, we may use a pigeonhole
and cancellation argument like the one in the discussion of orbits above. Since all
cyclic groups with n elements are essentially the same, we use the notation Cn for such
a group.
In general, if a is any element of a (possibly infinite) group G, the cyclic subgroup
generated by a has elements ai for i ∈ Z (a0 = e), and is denoted by hai. a is a
generator for the subgroup.

4
Alternating groups. The set of all even permutations in Sn , denoted An , is a subgroup
of Sn .

Dihedral groups. We consider the set D2n of all symmetries of a regular n-gon free to
move in three dimensions.
Caution: D2n is denoted Dn in some texts.
For example, the equilateral triangle has six symmetries: three rotations and three
reflections. In this case we have the group S3 . For a more enlightening example,
consider the symmetries of the square:
The group D8 of the square consists of the identity e, three rotations (by 90 degrees,
180 degrees, and 270 degrees counter-clockwise), and four reflections (one in each of
the vertical and horizontal axes, and two keeping diagonally opposite corners fixed).
Let g denote the rotation by 90 degrees. Then hgi = {e, g, g 2 , g 3 } is a subgroup of
D8 , the group of proper rotations of the square. Now let h be the reflection keeping
vertex 1 (and vertex 3) fixed. Then all the other reflections can be obtained from g
and h.

@ 4 3 In terms of permutations,
@
g = (1234)
@
@
@ g2 = (13)(24)
@
@ g◦h g3 = (1432)
@
@ h = (24)
g◦h = (14)(23)
@
@
h @
@ g2 ◦ h
g2 ◦ h = (13)
1 g ◦h 2 @
3
g3 ◦ h = (12)(34)

In general the group Dn may be obtained in similar fashion: let g be a counter-clockwise

rotation by 2π/n radians, and let h be the reflection which fixes vertex 1 and the center.
Then hgi is Cn , a cyclic subgroup of proper rotations of the n-gon, and the rest of the
symmetries are obtained by composing these with h. It is easy to see that D2n has 2n
elements—any symmetry is determined by the images of vertices 1 and 2. For vertex
1 there are n choices; if the image of 1 is r + 1 (where 0 ≤ r ≤ n − 1), then the image
of 2 must be r or r + 2. If 2 goes to r + 2 the permutation is g r ; if 2 goes to r the
permutation is g r h. Furthermore we note that g r h = hg −r = hg n−r (since g n = e).

Returning to D8 we may summarize the group operation in a group table:

5
 ◦ e g g2 g3 h gh g2h g3h
e e g g2 g3 h gh g2h g3h
g g g2 g3 e gh g2h g3h h
g2 g2 g3 e g g2h g3h h gh
g3 g3 e g g2 g3h h gh g2h
h h g3h g2h gh e g3 g2 g
gh gh h g3h g2h g e g3 g2
g2h g2h gh h g3h g2 g e g3
g3h g3h g2h gh h g3 g2 g e

1.4 Cosets.
Definition 1.6 For H a subgroup of G and a ∈ G, the left coset of H by a is the set
aH = {ah : h ∈ H}. Similarly the right coset of H by a is Ha = {ha : h ∈ H}.

For example, consider the subgroup H = {e, h} of D4 above. We tabulate the cosets of H
as a varies over D4 :

a aH Ha
e H H
g {g, gh} {g, g 3 h}
g2 {g 2 , g 2 h} {g 2 , g 2 h}
g3 {g 3 , g 3 h} {g 3 , gh}
h H H
gh {gh, g} {gh, g 3 }
g2h {g 2 h, g 2 } {g 2 h, g 2 }
g3h {g 3 h, g 3 } {g 3 h, g}

We note two phenomena: first, left cosets are in general different from right cosets (this is
not too surprising since the group operation here is not commutative); and secondly that
the left cosets partition the group, as do the right cosets. To put this second observation
another way, the right (for example) cosets induce an equivalence relation ≡H on G, namely
a ≡H b iff ab−1 ∈ H. To see this, suppose a and b are in the same coset, say Hc; then
a = h1 c and b = h2 c for some h1 , h2 ∈ H, and so ab−1 = h1 cc−1 h−1 −1
2 = h1 h2 ∈ H by the
closure properties of subgroups. Conversely, if ab−1 = h for some h ∈ H then a = hb, and it
follows that a ∈ Hb. Since we certainly have b ∈ Hb, a and b belong to the same coset of
H. It is straightforward to verify that ≡H is in fact an equivalence relation. Noting further
that in the example all cosets have the same size, we are led to the following result:

Theorem 1.3 (Lagrange) If G is a finite group and H is a subgroup of G then |H| divides
|G|.

6
Proof: We first show that any right coset Ha has the same size as H. Clearly a coset cannot
have more elements than the subgroup; if it has fewer then (by the PHP) we have h1 a = h2 a
for some h1 6= h2 ∈ H. However multiplying both sides by a−1 we find the contradiction
h1 = h2 . Furthermore every g ∈ G appears in at least one coset (namely Gg). Thus the
right cosets partition G and we have a disjoint union

G = Ha1 ∪ Ha2 ∪ · · · ∪ Hak

for some a1 , a2 , . . . , ak ∈ G. Then |G| = k · |H| as required.

The proof of the theorem actually gives more. If we define the index of H in G to be |G : H|
= the number of distinct right cosets of H in G, then

|G : H| = |G|/|H|.

The number of elements in a group G is known as the order of G, often written o(G) (we
will continue to use |G|). In the example above, |G| = 8, |H| = 2, and |G : H| = 4. The
four right cosets are H itself, {g, g 3 h}, {g 2 , g 2 h}, and {g 3 , gh}.

1.5 Exercises.
1. Prove that inverses are unique in groups.

2. Prove that for any group G of even order, there is some non-identity element g ∈ G
with g 2 = e.

3. Prove that any group of prime order is cyclic. [Hint: consider the cyclic subgroup
generated by any non-identity element of the group.] Conversely, show that if n is
prime then Cn has n − 1 generators. In general how many generators does Cn have?

4. Give the left and right cosets of the subgroup {e, g, g 2 , g 3 } in D4 . Find another sub-
group of size four and repeat the exercise.

5. Prove that |An | = |Sn |/2. [Hint: consider the coset An (12).]

6. In each case either find a subgroup of the required order in the given group, or show
there is none:
(a) order 10 in S5 ; (b) order 3 in D14 ; (c) order 5 in A6 .

7
2 Burnside’s Theorem.
In this section we begin to apply groups to counting problems. Here are two simple examples
of the kind of problem we consider.
Example. (Colouring the vertices of the square.) Suppose the vertices of the square are
to be coloured using two colours (red and white). How many different ways are there to do
this, if two colourings are considered equivalent if one can be obtained from the other (a) by
a proper rotation; (b) by any symmetry of the square?
Example. (Switching circuits.) How many different functions are there from Z23 to Z2 , if two
functions are considered equivalent if one can be obtained from the other by permuting the
inputs (variables)? For example, x1 x2 + x3 is equivalent to x2 x3 + x1 , but not to x1 + x2 + x3 .

2.1 The action of a group on a set.

Definition 2.1 The group (G, ·) acts on the set X if for each g ∈ G and x ∈ X there is
some g(x) ∈ X such that
(i) (g1 g2 )(x) = (g1 (g2 (x)) for all g1 , g2 ∈ G and x ∈ X;
(ii) e(x) = x for all x ∈ X.
If in addition g1 (x) = g2 (x) for all x ∈ X implies g1 = g2 , G is said to act faithfully on X,
and G may be considered as a subgroup of SX .

In our applications, all group actions will be faithful and all sets will be finite, so all groups
we use can be taken to be subgroups of some appropriate Sn . In the first example above,
the two appropriate groups are (a) C4 and (b) D8 (subgroups of S4 ).
In the second example we use a subgroup H ' S3 of S8 , corresponding to those permutations
of the elements of Z23 = {000, 001, 010, 011, 100, 101, 110, 111} obtained by permuting the
inputs under S3 . Numbering the elements of Z23 as 1, 2, . . . , 8 we tabulate the elements of
S3 and corresponding permutations of Z23 :

S3 H
e (1)(2)(3)(4)(5)(6)(7)(8)
(12) (1)(2)(35)(46)(7)(8)
(13) (1)(25)(3)(47)(6)(8)
(23) (1)(23)(4)(5)(67)(8)
(123) (1)(253)(467)(8)
(132) (1)(235)(476)(8) .

For example, transposing inputs 1 and 2 leaves 000 and 001 unchanged, transposes 010
and 100, transposes 011 and 101, and leaves 110 and 111 unchanged. We have listed the
permutations in complete cycle decomposition, including the trivial 1-cycles, because this is
important in later work.

8
Definition 2.2 If G acts on X and x ∈ X, then the stabilizer of x is
Stab x = {g ∈ G : g(x) = x}
i.e. the set of permutations g in G which leave x unchanged, xg = x.
For x ∈ X, the orbit of x under G is
Orb x = {g(x) : g ∈ G}
i.e. the set of all images xg of x under all permutations g in G.
Note that this definition of Orb x is a generalization of the previous one.

Example. (continued) In the group of all symmetries of the square, the stabilizer of 2 is
{e, g 2 h}; in the second example above, the stabilizer of 011 is {e, (23)(67)}.
Once again in the square, the orbit of 2 is {1, 2, 3, 4} (all vertices can be reached from vertex
2); in the second example, the orbit of 011 is {011, 101, 110}.
Theorem 2.1 If G acts on X and x ∈ X, then Stab x is a subgroup of G. If furthermore
G is finite then |G : Stab x| = |Orb x|.
Proof: Let K = Stab x. To see that K is a subgroup, first note that K 6= ∅ since e ∈ K.
Next check the two closure requirements: if k1 ∈ K and k2 ∈ K, then xk1 = x and xk2 = x.
Thus x(k1 k2 ) = (xk1 )k2 = xk2 = x, so k1 k2 ∈ K. For inverses, suppose k ∈ K. Then
xk = x, so (xk)k −1 = xk −1 ; since kk −1 = e, we have x = xk −1 , and so k −1 ∈ K.
For the second part of the theorem, we establish a 1–1 correspondence between the right
cosets of K and Orb x by matching Kg with xg. To see that this does in fact give a bijection
note that Kg1 = Kg2 ⇐⇒ g1 g2−1 ∈ K ⇐⇒ xg1 g2−1 = x ⇐⇒ xg1 = xg2 .
Note that the second part of the theorem may be restated as |G| = |Stab x||Orb x| for any
x ∈ X.

Example. (Proper rotations of the cube.) We find |G|, where G is the group of symmetries
of the cube that can actually be carried out in three dimensions.
5r r7
@ @
@ @
@6
r @r8

1r r3
@ @
@ @
2@r @r 4

The stabilizer of vertex 1 is Stab 1 = {e, (235)(476), (253)(467)} (if 1 is fixed, then so is 8;
thus we are left with the identity and two non-trivial rotations about the axis through 1 and
8). The orbit of 1 is Orb 1 = {1, 2, 3, 4, 5, 6, 7, 8} because any vertex can be reached from 1
by some rotation. Therefore |G| = |Stab 1||Orb 1| = 3 · 8 = 24.
Note that the group H above (permuting the inputs of the switching functions) contains
improper rotations (reflections) of the cube.

9
2.2 Burnside’s theorem.
If G acts on X, the orbits partition X; thus to count the number of non-equivalent elements
of X, we can count the number of orbits. To do this, we consider fixed points from the point
of view of the group G.

Definition 2.3 For G acting on X, and g ∈ G, Fix g is the set of elements of X fixed by
g, i.e.
Fix g = {x ∈ X|xg = x}.

Example. In the group H above acting on Z23 , let h = (35)(46). Then Fix h = {1, 2, 7, 8}.

Theorem 2.2 (Burnside, 1911) If a finite group G acts on a finite set X, and N is the
number of orbits of X under G, then
1 X
N= |Fix g|.
|G| g∈G

Proof: Consider the set S = {(x, g)|x ∈ X, g ∈ G and xg = x}. This set can be counted
in two ways: first by counting for each permutation g ∈ G the number of elements of X fixed
by g, and second by counting for each x ∈ X the number of elements of G that fix x. Thus
X X
|S| = |Fix g| = |Stab x|.
g∈G x∈X

We now count the sum over X in a carefully chosen order. Choose a set of representatives
x1 , x2 , . . . , xN , one from each orbit. If x is in the same orbit as xi , then Orb x = Orb xi (the
orbits partition X), and so by theorem 2.1 |Stab x| = |Stab xi |. Then

X N
X X N
X
|Fix g| = |Stab x| = |Orb xi ||Stab xi | = N · |G|
g∈G i=1 x∈ Orb xi i=1

using theorem 2.1 again.

10
Example. We return to the problem of counting 2-colourings of the square. First consider
only the group C4 (proper rotations), and for each σ ∈ G calculate |Fix σ|.

σ |Fix σ|
e 16 (all colourings)
g 2 (all red or all white)
g2 4 (opposite corners must agree)
g3 2

Thus there are (1/4) · (16 + 2 + 4 + 2) = 6 different colourings of the vertices of the square
using red and white.
To do the calculations in the case where all permutations in D8 are allowed, we condense
the table by grouping permutations that have the same cycle structure (the second column
of the table gives the number s of such permutations):

cycle structure number, s |Fix σ| s · |Fix σ|

e = (1) 1 16 16
g = (1234) 2 2 4
g 2 = (13)(24) 3 4 12
h = (24) 2 8 16

Thus there are (1/8) · (16 + 4 + 12 + 16) = 6 different colourings (why do we get the same
answer?)

Example. For the switching circuits with three inputs, we find

cycle structure number, s |Fix σ| s · |Fix σ|

e = (1) 1 28 = 256 256
(35)(46) 3 26 = 64 192
(253)(467) 2 24 = 16 32

and there are (1/6) · (256 + 192 + 32) = 80 different circuits.

2.3 Exercises.
1. How many ways can the square be coloured with three colours (red, white, and black)
if proper rotations are allowed? if all symmetries are allowed?

2. A necklace is a colouring of the vertices of a regular n-gon, where two colourings

are equivalent if one can be obtained from the other by a symmetry in D2n (coloured
beads are placed on a circle; the circle may be rotated and also turned over). Find
the number of different types of necklace that can be made from the following beads,
assuming all beads are used.

11
 (a) Three black and three white beads.
(b) Three black, three white, and three red beads.
(c) Three black, three white, and two red beads.
(d) Ten beads total, selected from a bag containing ten black and ten white beads.

3. A benzene ring is a ring of six carbon atoms at the vertices of a hexagon.

Different organic compounds may be
C C made by attaching another group of

A atoms (a radical) to each of these six

A carbon atoms. How many different

A
compounds can be made by attaching
C C
one H, CH3 , or OH radical at each ver-
A



tex? Assume all the C–C bonds are
A
A
equivalent, and that the rings may be
C C turned over.

4. How many ways can the vertices of a regular tetrahedron be coloured, using 2 colours?
3 colours? n colours?

5. In definition 5.2.1 above (the action of a group on a set) it is claimed that faithful
group actions may be regarded as subgroups of SX . Prove that this is so; i.e. prove
that each g ∈ G acts as a permutation (1–1 onto function) on X.

6. How many ways can the vertices of a cube be coloured with 2 colours? with n colours?

3 Cycle index and pattern inventory.

In the previous sections we found that the cycle structure of permutations played an impor-
tant rôle in counting problems. Our goal in this section is to find the generating function
for the number of different ways to m-colour a set X acted on by a group G, and to this end
we introduce a polynomial representation of the cycle structure for σ in G.

Definition 3.1 For σ ∈ Sn , the cycle structure term of σ is a product of the variables
xi , with one copy of xi for each cycle of length i in σ (including 1-cycles). The cycle index
polynomial PG of a subgroup G of Sn is the sum of the cycle structure terms for σ ∈ G,
divided by the order of G:
1 X
PG (x1 , x2 , . . . , xn ) = cycle structure term of σ.
|G| σ∈G

Example. For D8 , the group of the square, we tabulate the cycle structure terms:

12
 cycle structure of σ ∈ D8 cycle structure term
e = (1)(2)(3)(4) x41
g = (1234) x4
g 2 = (13)(24) x22
g 3 = (1432) x4
2
h = (1)(24)(3) x1 x2
g ◦ h = (14)(23) x22
g 2 ◦ h = (13)(2)(4) x21 x2
g 3 ◦ h = (12)(34) x22
1 4
and the cycle index of D8 is (x1 + 2x21 x2 + 3x22 + 2x4 ).
8
Example. The group H above (in the switching circuit example) is a subgroup of S8 ; we
extend the previous table describing H to include cycle structure terms:

cycle structure of σ ∈ H cycle structure term

(1)(2)(3)(4)(5)(6)(7)(8) x81
(1)(2)(35)(46)(7)(8) x41 x22
(1)(25)(3)(47)(6)(8) x41 x22
(1)(23)(4)(5)(67)(8) x41 x22
(1)(253)(467)(8) x21 x23
(1)(235)(476)(8) x21 x23
1 8
The cycle index of H is (x1 + 3x41 x22 + 2x21 x23 ).
6

13
Definition 3.2 The pattern inventory of the m-colourings of a set X acted on by a group
G is the generating function for the number of non-equivalent m-colourings of X.
That is, the pattern inventory is a polynomial I(c1 , c2 , . . . , cm ) (representing colours 1 to m)
such that the coefficient on ci11 ci22 · · · cimm is the number of colourings of X with i1 elements of
colour 1, i2 elements of colour 2, etc. Note that each term must satisfy i1 +i2 +· · ·+im = |X|.

Example. Let c1 and c2 be variables representing red and white, for the 2-colourings of the
square acted on by D8 . The number of colourings for each possible combination of i1 and i2
is:
i1 i2 number of colourings
4 0 1
3 1 1
2 2 2
1 3 1
0 4 1
and the pattern inventory is I = c41 + c31 c2 + 2c21 c22 + c1 c32 + c42 .

Most pattern inventories are difficult to find directly like this, and indeed there would be little
interest in such calculations—instead we wish to find inventories using Burnside’s theorem
and then use the inventories to find the number of colourings. To motivate the eventual
result, we consider a particular case.
Example. In the switching circuit problem, let σ = (1)(2)(35)(46)(7)(8). For illustration
we will find the number of functions which take the value 1 exactly 5 times (and value 0
exactly 3 times) and are fixed by σ. To do this we must find the number of ways to distribute
five 1s and three 0s over the eight points so that points 3 and 5 get the same value, and
points 4 and 6 get the same value. This is the same as the coefficient on c51 c32 in the expansion
of (c1 + c2 )(c1 + c2 )(c21 + c22 )(c21 + c22 )(c1 + c2 )(c1 + c2 )—in this case we find the number of
ways to select either c1 or c2 in each factor for a total of five c1 s and three c2 s. The cycle
structure term for σ is x41 x22 ; we see that to get the required number of functions we may
evaluate the coefficient of c51 c32 in the polynomial that results from substituting c1 + c2 for x1
and c21 + c22 for x2 in the cycle structure term for σ. The expansion gives
(c1 + c2 )4 (c21 + c22 )2 =
c81 + 4c71 c2 + 8c61 c22 + 12c51 c32 + 14c41 c42 + 12c31 c52 + 8c21 c62 + 4c1 c72 + c82
and we conclude that there are 12 functions taking value 1 exactly five times and fixed by
σ.
To apply Burnside’s theorem we carry out a similar calculation for each σ ∈ H. As before,
we condense the work by noting that some permutations have the same cycle structure. The
table below gives the appropriate coefficients on cp1 cq2 for each possible cycle structure in H,
taking into account the number s of such permutations.

14
 σ∈H s polynomial c81 c71 c2 c61 c22 c51 c32 c41 c42 c31 c52 c21 c62 c1 c72 c82
e 1 (c1 + c2 )8 1 8 28 56 70 56 28 8 1
(35)(46) 4 2 2
3 3(c1 + c2 ) (c1 + c2 ) 3 2 12 24 36 42 36 24 12 3
(253)(467) 2 2(c1 + c2 )2 (c31 + c32 )2 2 4 2 4 8 4 2 4 2
6 6 24 54 96 120 96 54 24 6

To find the pattern inventory, we divide the sum indicated in the bottom row of the table
by |H|, giving

I = c81 + 4c71 c2 + 9c61 c22 + 16c51 c32 + 20c41 c42 + 16c31 c52 + 9c21 c62 + 4c1 c72 + c82 .

There is one function with constant value 1, there are four essentially different functions
with exactly 7 ones and 1 zero, etc. Note that the total number of different functions is

1 + 4 + 9 + 16 + 20 + 16 + 9 + 4 + 1 = 80

as before.

Theorem 3.1 (Pólya, 1937) If a subgroup G of Sn with cycle index PG (x1 , x2 , . . . , xn ) acts
on a set X with |X| = n then the pattern inventory for the non-equivalent m-colourings of
X is m m m
!
c2i , . . . , cni
X X X
I(c1 , c2 , . . . , cm ) = PG ci ,
i=1 i=1 i=1

Proof: As above, we use Burnside’s theorem to obtain the coefficient of ci11 ci22 · · · cimm where
i1 is the number of elements of colour 1, i2 is the number of colour 2, etc. This coefficient
will be
1 X
|Fix σ|
|G| σ∈G
where we consider only those colourings with the correct number of each colour. Once again,
|Fix σ| is found by substituting the sums as shown into the cycle structure term for σ. The
division by |G| is part of the definition of PG .

15
 Example. Find the number of necklaces which can be made using five black and four white
beads.
First note that the appropriate group is D18 , acting on the regular 9-gon. D18 has four kinds
of permutations: the identity, single 9-cycles, products of 3 disjoint 3-cycles, and reflections.
σ ∈ D18 s polynomial coefficient of c51 c42
 
9
e 1 (c1 + c2 )9 5
= 126
(123456789) 6 6(c91 + c92 ) 0
(147)(258)(369) 2 2(c31 + c32 )3  
0
4
(1)(29)(38)(47)(56) 9 9(c1 + c2 )(c21 + c22 )4 9 2 = 54
18 180
180
Thus the number of such necklaces is = 10.
18
Example. The faces of a cube are coloured with two faces red, two faces blue, and two
faces green. In how many ways can this be done?
The appropriate group here is the group G of proper rotations of the cube, considered as
a subgroup of S6 , permuting the six faces of the cube. For convenience we reproduce the
diagram of the cube, and label the faces a, b, c, d, e and f as given in the table beside the
diagram:

vertices face
5r r7 1, 2, 3, 4 a
@ @
@ @ 1, 2, 5, 6 b
@r6 @r8
1, 3, 5, 7 c
1r r3 3, 4, 7, 8 d
@ @ 2, 4, 6, 8 e
@ @
2@r @r 4 5, 6, 7, 8 f

We describe the proper rotations of the cube in terms of permutations of a, b, c, d, e, f . In

addition to the identity (a)(b)(c)(d)(e)(f ) there are three kinds of rotation:

5r r7 5r r7 5r r7
@ @ @ @ @ @
@ × @ @ @+ @ @
@6
r @r8 @6
r @r 8 @6r r8

@

 
1r r3 1r r3 1r  r3

@ @ @ @  @ @
@ × @ +
@ @ @ @
2@r @r 4 2@r @r 4 2@r @r 4

16
 1. rotations about an axis through the centres of two opposite faces—there are three
choices for the pair of faces, and then three non-identity rotations. Two are of form
(bcde), one of form (bd)(ce).

2. rotations by π about an axis through the midpoints of two diagonally opposite edges—
there are six such pairs of edges, giving permutations of form (ab)(ce)(df ).

3. rotations by 2π/3 and 4π/3 about an axis through two diagonally opposite vertices
such as 1 and 8. There are four ways to choose the pair of vertices, and two rotations
for each, giving eight symmetries of form (abc)(def ).

The table gives the number of each kind of symmetry, the associated polynomial in c1 , c2
and c3 , and the coefficient on c21 c22 c23 :

σ∈G s polynomial coefficient of c21 c22 c23

  
6 4
(a)(b)(c)(d)(e)(f ) 1 (c1 + c2 + c3 )6 2 2
= 90
(a)(bcde)(f ) 6 6(c1 + c2 + c3 )2 (c41 + c42 + c43 ) 0
(a)(bd)(ce)(f ) 3 3(c1 + c2 + c3 )2 (c21 + c22 + c23 )2 18
(ab)(ce)(df ) 6 6(c21 + c22 + c23 )3 36
(abc)(def ) 8 8(c31 + c32 + c33 )2 0
24 144

144
giving = 6 distinct colourings.
24
If only the total number of m-colourings is required, we have the following corollary which
is simpler both to state and to apply:

Theorem 3.2 For G and X as above, the number of non-equivalent m-colourings of X is

PG (m, m, . . . , m)

Proof: For the total number of m-colourings we require the sum of the coefficients of the
inventory. This is obtained by setting c1 = c2 = · · · = cm = 1, thus making all the sums in
the statement of the previous theorem equal to m.

Example. If four colours are available, in how many ways can the vertices of the square be
coloured if proper rotations are allowed? if all symmetries are allowed?
The cycle index of C4 is 41 (x41 + x22 + 2x4 ); thus there are 14 (44 + 42 + 2 · 4) = 70 different
colourings if only proper rotations are allowed.
For D8 we find PD8 (4, 4, 4, 4) = 18 (44 + 2 · 42 · 4 + 3 · 42 + 2 · 4) = 55 different colourings.
(Find an example of two four-colourings of the vertices of the square which are equivalent
under D8 but not under C4 .)

17
Example. How many ways can the faces of a cube be coloured red, blue, or green, if any
number of each colour many be used?
From previous work on the group of rotations of the cube we find the cycle index is
1  6 
PG (x1 , x2 , . . . , x6 ) = x1 + 6x21 x4 + 3x21 x22 + 6x32 + 8x23 .
24
Thus the number of three-colourings (obtained by setting x1 = · · · = 3) is (36 + 6 · 32 · 3 +
3 · 32 · 32 + 6 · 33 + 8 · 32 )/24 = 57.

3.1 Exercises.
1. How many ways can the faces of a cube be coloured with three faces red, two white,
and one blue?

2. Re-do exercises 2.3 (1–4, 6) using pattern inventories.

3. How many non-equivalent ways are there to colour the edges of a cube with 2 colours?
m colours? with six edges red and six edges blue?

4. (Switching circuits with four inputs). Permuting the co-ordinates of the vertices of the
4-dimensional cube Z24 gives a group K isomorphic to S4 acting on Z24 .

(a) If the vertices of the cube are labeled {1, 2, . . . , 9, A, B, . . . , G} give the cycle
structure terms for representative permutations in K.
(b) What is the cycle index of K?
(c) Find the number of non-equivalent (under permutation of variables) switching
circuits with four inputs (functions from Z24 to Z2 ).

5. Let X = {1, 2, 3, 4}2 , i.e. X is the set of all ordered pairs (i, j) with 1 ≤ i, j ≤ 4.
S4 induces a group G acting on X by permuting the co-ordinates (for example, the
transposition (12) in S4 corresponds to a permutation of X which interchanges (1,1)
and (2,2), interchanges (1,2) and (2,1), interchanges (1,3) and (2,3), etc.).

(a) Give the cycle structure terms for representative permutations in G, and find the
cycle index PG .
(b) What is the number of non-equivalent 2-colourings of X under such permutation of
co-ordinates? What is the number of non-equivalent (4 × 4) matrices with entries
either 0 or 1, if two matrices are equivalent when one can be obtained from the
other by applying the same permutation to both rows and columns? What is
the number of non-equivalent binary relations on {1, 2, 3, 4}, if two relations are
equivalent when one can be obtained from the other by re-labeling (i.e. by a
permutation of {1, 2, 3, 4})?

18
6. Let Y be the set of all 2-subsets of {1, 2, 3, 4}, i.e. Y = {{1, 2}, {1, 3}, . . . , {3, 4}}. As
in the previous exercise, S4 induces a group G acting on Y (for example, (12) in S4
corresponds to a permutation of Y which fixes {1, 2}, interchanges {1, 3} and {2, 3},
etc.).

(a) Find the cycle index of G.

(b) What is the number of non-equivalent 2-colourings of Y ? What is the number of
non-equivalent symmetric (4 × 4) 0,1-matrices with diagonal entries all 0? What
is the number of non-isomorphic simple graphs on 4 points? (Recall that a graph
is simple if it has no multiple edges or loops.)

7. What is the number of non-isomorphic graphs on 4 points with no multiple edges (but
possibly loops)?

19