Вы находитесь на странице: 1из 14

GILA III: The orbit-counting lemma and baby Polya

June 16, 2009 by Qiaochu Yuan

The orbit-stabilizer theorem implies, very immediately, one of the most


important counting results in group theory. The proof is easy enough to
give in a paragraph now that weve set up the requisite machinery.
Remember that we counted fixed points by looking at the size of the
stabilizer subgroup. Lets count them another way. Since a fixed point is
really a pair

such that

, and weve been counting them indexed

by , lets count them indexed by . We use

to denote the set of

fixed points of . (Note that this is a function of the group action, not the
group, but again were abusing notation.) Counting the total number of
fixed points vertically, then horizontally, gives the following.
Proposition:

On the other hand, by the orbit-stabilizer theorem, its true for any orbit
that

, since the cosets of any stabilizer subgroup

partition

. This immediately gives us the lemma formerly known as

Burnsides, or the Cauchy-Frobenius lemma, which well give a neutral


name.
Orbit-counting lemma: The number of orbits in a group action is given
by

, i.e. the average number of fixed points.

In this post well investigate some consequences of this result.


First, though, Id like to note that the proof is usually presented slightly
differently. I prefer this presentation because vertical and horizontal
counting is a very general technique and its good to see it spelled out
when it appears. The orbit-counting lemma is not, by any means,
a deep or technicalresult, but its value is in the change of perspective:
averaging the behavior of every group element tells us the aggregate
behavior of the group action. The emphasis on

also ties the lemma

into representation theory, since fixed points are an example of


a character, and such averaging arguments are also common in
representation theory.

Now, one example application I mentioned was about painting the faces of
a cube up to rotational symmetry. This is the example used in
the Wikipedia article on Burnsides lemma, so instead of repeating
Wikipedia Ill just add that the group in question is known as
the octahedral group. Its not too hard to list its elements explicitly and
figure out their fixed points.
I would give more examples, but the next big result I want to prove is, in
my opinion, just as easy for the general case as for any of the interesting
examples, so Ill give the examples later. Recall that the general situation
the PET deals with is that, given a set

on which a group

acts, we want

to compute the orbits of the induced group action on functions


where

is a (for now, finite) set of colors or urns. So let

denote

the number of orbits of the induced action on functions

(Once more, this is a function of the group action and not just the group.)
Orbit-counting tells us that to do this it suffices to count the number of
fixed points of a given

When does a permutation

fix a function

for some function

if and only if

? This is possible
for all

, but after a

moments thought this condition is equivalent to every slot in a cycle


having the same color. In other words, if
in the cycle decomposition of
then

denotes the number of cycles

(with respect to the group action on

),

(with respect to the group action on functions), since

each cycle can be colored independently of the other cycles. This gives the
following preliminary result which we will generalize to Polyas theorem.
Baby Polya:
In other words,

is a polynomial (which would not have been obvious

without orbit-counting) and the coefficient of


permutations in

with

cycles divided by

is the number of
. It is, of course, quite useful

to be able to describe an important aspect of the group action on


functions

using only information about the group action on

, but

to really appreciate this result we should apply it to a few examples.


Balls and urns

In this example
symmetry group

is a set of size

and it is being acted on by the full

; this is just the balls and urns problem, with slots as

the balls and colors as the urns. It can be solved with an elementary
counting argument: an arrangement of

indistinguishable balls into

distinguishable urns is equivalent to a string of


which correspond to balls and

symbols,

of

of which correspond to dividers

between urns. A typical string looks like


OOOO|O||OO|||
which corresponds to the sequence

of seven balls in seven

urns. By an elementary counting argument (which can be phrased in


terms of group actions if you like), the number of ways to do this is

.
Note that this is a polynomial in

of degree

; baby Polya tells us that it

is also equal to

and therefore that the coefficient of


number of permutations in

with

in

is the

cycles in their cycle decomposition.

This is a defining identity of the (unsigned) Stirling cycle numbers. (Giving


a direct proof of this fact is a good exercise.)
In this example, baby Polya didnt tell us a better way to count the orbits
of the group action, but rather gave us information about the group of
symmetries. In the next example, we know more about the group of
symmetries than the orbits.
Necklaces
Now the symmetry group is the cyclic group

and

is a finite

set of colors for each of the beads of a necklace. Describing the cycle
decomposition of elements of

requires some number theory, so its

instructive to look at an example. For example, when

and the group

is generated by

, the cycle decompositions of its powers are as

follows.

An easy way to describe whats going on is that


addition of
integer

is isomorphic to the

. The length of any cycle is the smallest positive

such that

. If

then we can take

more generally,

.
It follows that the number of cycles in the cycle decomposition of
is

. By baby Polya, the number of necklaces with

beads

and colors is

.
To get a nicer formula, we need to count the solutions
to

. This is equivalent to solving

which has

solutions (in the range we care about) by the definition of

thetotient function. It follows that the number of necklaces is also

.
Setting

gives a classic proof of the identity

Setting

gives the number of binary necklaces, which appears

as A000031 in the OEIS. This identity can also be proven via Mobius
inversion, another fascinating area of combinatorics I hope to discuss
someday.
Simple graphs of small order
Here

is the set of pairs of

vertices, which well denote

is the induced action of the full symmetry group

, the action

acting on pairs in the

obvious way, and we are only interested in functions

, where a

denotes that the edge is there. Unfortunately, it is very difficult to write


down the cycle decomposition of permutations acting on pairs in general,
so well have to make do with small examples.
First, it should be clear that the number of cycles induced by a given
permutation

on

depends only on its conjugacy class, since

conjugation is equivalent to relabeling the elements of each cycle. In fact,


the number of conjugacy classes in
of

, so lets take

group has

is given by the number of partitions

, which only has seven partitions. The symmetry


elements arranged in the following conjugacy classes,

which well name using a representative element:


1.

The identity, of which there is one. It has

2.

, of which there are


there are

3.

cycles, one for every edge.

. Every edge is in a cycle of order five, so

cycles.

, of which there are

. The six edges connecting

organize into two cycles, but of different sizes (check this). The remaining four
edges form another cycle, so there are
4.

, of which there are

cycles.
. The three edges connecting

form one cycle. The six edges connecting


The edge
5.

is fixed, so there are

, of which there are

to

form two cycles.

cycles.

. The six edges connecting

to

form three cycles. The remaining four edges are all fixed, so there are
6.

, of which there are


form two cycles. The edges

cycles.

. The four edges connected to


are fixed. The remaining four edges

form two cycles, so there are cycles.


7.

, of which there are


edges connecting

. The edge

is fixed and the

form their own cycle. The remaining six edges are in

their own cycle, so there are cycles.

Phew! In total, baby Polya tells us that the number of non-isomorphic (not
necessarily connected) simple graphs on

vertices is

(which doesnt seem to display entirely, unfortunately). This computes


to

, and you can find the list here. But isnt this computation so much

easier than trying to list them by hand and only finding

? This is

sequence A000088 in the OEIS, which also corroborates the value of

Stanley says that there exist explicit formulas describing the cycle
decomposition of the action of

of

, so everything weve done here

generalizes.
Now, by restricting ourselves to a few classes of permutation that we
understand well, we can compute lower bounds on the number of nonisomorphic graphs. For example, taking only the identity gives the trivial
lower bound

, which is what would happen if every graph had full

automorphism group. Taking the identity and every transposition gives


.
The OEIS gives more precise asymptotic expansions along these lines. We
can also use this technique to get an upper bound. First, note that the
reason we picked transpositions is that any non-identity permutation
induces at most

cycles in its action on edges (why?) and

transpositions are the equality case. All other permutations induce larger
cycles, so have less of them, which gives the upper bound
.
Unfortunately, the two are rather far apart (far larger than a constant
factor). Even if we attempt the casework going through other nice
conjugacy classes, I believe the gap between the lower and upper bounds
we get remains large. The upper bound gives us much more information
than the lower bound, however; asymptotically, the lower bound reduces
to the trivial lower bound, whereas the upper bound is a factor of
smaller than the trivial upper bound (the number of labeled graphs on
vertices).
Where do we go from here?
The analysis we had to do in the case of simple graphs suggests that we
might not only like to count the number of cycles in a given group action,
but how many cycles there are of each length. It would be useful, in
particular, if we could do this for the symmetric group. Our analysis also
suggests that baby Polya isnt detailed enough: if we want to compute the

number of non-isomorphic graphs with a fixed number of vertices and also


a fixed number of edges, what we need is a way to generalize baby Polya
to count functions with a specified number of uses of each color.
Both of these concerns are answered by the PET and its surrounding
machinery. As a motivating example, consider prime necklaces, i.e.
let

be prime. Then the necklace-counting result tells us that the

number of necklaces is

which, as we have seen, gives us a combinatorial proof of Fermats little


theorem. Now let

where

is an odd prime and lets count the

number of necklaces using two colors, but with the additional constraint
that there are exactly

beads of the first color and

beads of the other.

This is a natural enough question, but baby Polya isnt equipped to specify
how many of each color there are.
We can solve this problem by hand, though. There are

labeled

necklaces. Two of these necklaces are in their own orbit: they are the
necklaces consisting of alternating colors. Every other labeled necklace
has trivial stabilizer subgroup (why?), hence is in a full orbit. This gives
a combinatorial proof that

which is a special case of a special case of Lucas theorem. More generally,


its true that

and if we could generalize baby Polya to computing the number of


necklaces with
and

such that

of the beads are of one color

of the beads are of the other, we could prove this as well.

Again, this can be proven by hand (and its easier to consider a smaller
symmetry group to do it), but were going for generality here.

Whats going to happen is that while it will be tedious to compute the


number of orbits that use a particular combination of colors, it will be
very interesting to compute the number of orbits using every combination
of colors, packaged into a generating function; the particular coefficients
of this generating function will tell us about particular combinations of
colors, but its the entire function that is most easily computed. The next
post will discuss and justify the use of generating functions; Ill try to start
from the beginning and give plenty of references.
Exercises
Compute
group

for dihedral necklaces, i.e. replace

with the dihedral

Compute

for alternating balls-and-urns, i.e. replace

thealternating group

Compute directly the number of permutations in


given by a partition

with

in the conjugacy class

; in other words, the number of permutations

with a cycle decomposition consisting of cycles of lengths

Give a second proof of the orbit-counting lemma by analyzing the linear


operator
space over

on the representation of

given by the free vector

If you havent seen it before, give a combinatorial proof


that
RHS as

. (Bonus: why is it interesting to write the


? What does this have to do with the third exercise?)

Burnside's lemma: orbit-counting theorem


(cube and necklace colorings)

Contact Graeme
Home
Email
Twitter

Math Help > Counting > Burnside's lemma

Burnside's lemma, is also called Burnside's counting theorem,


the Cauchy-Frobenius lemma or the orbit-counting theorem.
Burnside's Lemma

Let G be a finite group that acts on a set X. For each g in G let


Xg denote the set of elements in X that are fixed by g.
Burnside's lemma asserts the following formula for the number
of orbits, denoted |X/G|:
|X/G| =

1 g
|X |
|G| g G

Two elements of X belong to the same "orbit" when one can be


reached from the other by through the action of an element of
G. For example, if X is the set of colorings of a cube, and G is
the set of rotations of the cube, then two elements of X belong
to the same orbit precisely when one is a rotation of the other.
Example

6 people are to select a piece each from 10 pieces of cake, of


which there are 2 pieces each of 5 different kinds. How many
ways is this possible?
Solution:
Create 4 more virtual people. Now there are 10 pieces of cake
to be distributed among 10 people. The number of
permutations of 10 things is 10!, and since there are 5
indistinguishable pairs, we divide by 2^5. 10!/2^5=113400 is
the number of elements of X.
Now we apply Burnside's lemma, where the permutation group
of order 113400 is acted on by the group of permutations of the
four virtual people, of size 4! = 24. That is, if two of the 113400

elements of X can be reached by permuting the four virtual


people, then they belong to the same orbit, and so the number
of orbits is the number of distinct answers to the original
question.
The group of permutations of 4 objects has 1 identity element,
six 2-cycles, eight 3-cycles, six 4-cycles, and 3 pairs of 2cycles, as given by the following table:
cycles

number of permutations

example

identity

(1)(2)(3)(4)

2-cycles

(12)(3)(4)

3-cycles

(123)(4)

4-cycles

(1234)

pair of 2-cycles

(12)(34)

Total

24

Next, we find the number of elements of X that are fixed under


each permutation of G, and add them up. That means, for a
given permutation of G, for example interchanging the cakes
selected by virtual persons 1 and 2, what elements of X are left
unchanged? That would be just the elements of X in which
virtual persons 1 and 2 have the same kind of cake. The
number of such elements is the number of ways for persons 1
and 2 to pick one kind of cake, and then persons 3 through 10
to choose from two pieces of each of four kinds of cake -- the
example is filled out as the second row of the following table:
cycles

number of
permutations

identity

example

Fixed elements of X under this


permutation

(# perm)*(fixed
elements in this perm)
= total fixed elements
of X

(1)(2)(3)(4)

All of them: 10!/25=113400

1 * 113400 = 113,400

6 * 12600 = 75,600

2-cycles

(12)(3)(4)

Elements of X are fixed if 1 and 2 have the


same type of cake. 5 ways to pick a type of
cake for 1,2 to share, then 8!/24 ways to pick
the other pieces. 5*8!/24=12600.

3-cycles

(123)(4)

None of the elements of X are fixed by a 3cycle, because 1,2,3 must have at least two
kinds of cake among them.

4-cycles

(1234)

No fixed elements of X.

pair of 2cycles

Total

24

(12)(34)

5 ways to pick the kind of cake for 1,2; 4


ways to pick the kind of cake for 3,4;
6!/23 ways to pick the other pieces.
5*4*6!/23=1800.

3 * 1800 = 5400

194,400

Finally, 194400/24 = 8100, which is the number of orbits of X


under permutations of the virtual people.
Same example not using Burnside's Lemma -- the hard way
(cases)

To illustrate how hard this problem is without Burnside's


Lemma, I'll show you how I solved it before this wonderful
lemma was pointed out to me...
Proof that 6 distinct people (123456) can choose a piece of
cake from 2 of each of 5 kinds (AABBCCDDEE) in 8100
ways:
For each of the 5 cases where person 1 and person 2 choose
the same kind of cake, then persons 3-6 have 204 ways to
choose among the remaining 4 kinds of cake.
For each of the 20 cases where person 1 and person 2 chose
different kinds of cake, then persons 3-6 have 354 ways to
choose among the remaining eight pieces of cake.
5*204 + 20*354 = 8100
Proof that 4 distinct people (3456) can choose a piece of
cake from 2 of each of 4 kinds (BBCCDDEE) in 204 ways:
For each of the 4 cases where person 3 and person 4 choose
the same kind of cake, then persons 5-6 have 9 ways to pick
cake from CCDDEE, namely CC, CD, CE, DC, DD, DE, EC,
ED, EE.
For each of the 12 cases where persons 3 and 4 choose
different kinds of cake, then persons 5-6 have 14 ways to pick
cake from BCDDEE, namely BC, BD, BE, CB, CD, CE, DB, DC,
DD, DE, EB, EC, ED, EE.

4*9 + 12*14 = 204


Proof that 4 distinct people (3456) can choose a piece of
cake from (ABCCDDEE) in 354 ways:
For the 2 cases where persons 3 and 4 choose A and B, then
persons 5-6 have 9 ways to pick cake from CCDDEE (see
above).
For the 12 cases where person 3 or 4 chooses A or B, and
person 4 or 3, respectively, chooses C, D, or E, then persons 56 have 14 ways to pick from BCDDEE (see above).
For the 6 cases where persons 3 and 4 each choose different
C, D, or E, then persons 5-6 have 21 ways to pick from
ABCDEE, namely AB, AC, AD, AE, BA, BC, BD, BE, CA, CB,
CD, CE, DA, DB, DC, DE, EA, EB, EC, ED, EE
For the 3 cases where persons 3 and 4 each choose CC, DD,
or EE, then persons 5-6 have 14 ways to pick from ABEE,
namely AB, AD, AE, BA, BD, BE, DA, DB, DD, DE, EA, EB, ED,
EE
2*9 + 12*14 + 6*21 + 3*14 = 354
Same example not using Burnside's Lemma -- the easy way

This puzzle can be divided into three cases:


Case 1: Only two people choose the same kind of cake as
another person.
In this case, there are C(5,1)=5 ways to choose the kind of cake
chosen by 2 people, and then 6!/2 ways to permute the cakes,
considering that one pair of cakes are indistinguishable.
5*6!/2=1800.
Case 2: Four of the people choose the same kind of cake as
another person.
In this case, there are C(5,2)=10 ways to choose the two kinds
of cake chosen by these 4 people, and C(3,2) ways for the two
remaining people to choose from among the three other cake

types. Then there are 6!/4 ways to permute the cakes,


considering two pairs of cakes are indistinguishable.
10*3*6!/4=5400.
Case 3: All six people choose the same kind of cake as another
person.
In this case, there are C(5,3)=10 ways to choose the three
kinds of cake, and then 6!/8 ways to permute the cakes,
considering three pairs of cakes are indistinguishable.
10*6!/8=900.
Adding the cases, 1800+5400+900=8100.
Internet references

Wikipedia: Burnside's lemma -- example of the 57 colorings of


the faces of a cube with three colors.
Mathworld: Cauchy-Frobenius Lemma
Dr. Math: Permutations in a Necklace -- using Burnside's
Lemma to count the number of different necklaces. (From the
same site: Polya-Burnside Lemma)
Jaap's Puzzle Page: Useful Mathematics -- Permutations, The
parity of a permutation, Disjoint cycle notation, The Order of a
permutation, Groups, Conjugation, Commutation Size of the
group, Subgroups, Centre, Supergroup, Metrics, God's
Algorithm, Counting, The number of orientations, The number
of permutations, The number of combinations, Burnside's
Lemma
Bard College Abstract Algebra Math332: Quiz2 Practice
Problems.pdf (mirror)
Haskell for Maths: Polya Counting -- How many different ways
are there to colour the faces of a dodecahedron using red,
green and blue? No, the answer's not 312...
Related pages in this website

"hyper-binary" partition

Combination identities

Вам также может понравиться