Thanks to visit codestin.com
Credit goes to www.scribd.com

0% found this document useful (0 votes)
96 views10 pages

Handout 2 For MATH 323, Algebra 1: Permutation Groups and Abstract Groups

This handout discusses group theory and abstract groups, focusing on permutation groups and their relationship to polynomial equations. It begins by explaining the origins and early concepts of group theory in studying quadratic and polynomial equations. It then provides definitions of key algebraic structures like groups, fields, and their relationships. The document aims to motivate group theory as expressing symmetries and uses examples from Galois theory to illustrate applications of group concepts.

Uploaded by

vikrant singh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
96 views10 pages

Handout 2 For MATH 323, Algebra 1: Permutation Groups and Abstract Groups

This handout discusses group theory and abstract groups, focusing on permutation groups and their relationship to polynomial equations. It begins by explaining the origins and early concepts of group theory in studying quadratic and polynomial equations. It then provides definitions of key algebraic structures like groups, fields, and their relationships. The document aims to motivate group theory as expressing symmetries and uses examples from Galois theory to illustrate applications of group concepts.

Uploaded by

vikrant singh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

Handout 2 for MATH 323, Algebra 1:

Permutation groups and abstract groups


Laurence Barker, Mathematics Department, Bilkent University,
version: 30th October 2011.

These notes discuss only some aspects of the lectured material, and they are not intended to
be particularly useful as preparation for any exam.

1: The origins of group theory


Algebra is a branch of mathematics whose core topics arose from the study of equations such
as ax2 +bx+c = 0 and ax3 +bx2 +cx+d = 0 and, generally, an xn +an1 xn1 +...+a1 x+a0 = 0.
Such equations are called polynomial equations. Of course, algebra has evolved, over time,
in many different directions. Nevertheless, at an undergraduate level, the Galois theory of
polynomial equations is still of central importance, partly because it provides a motivation for
various fundamental concepts in algebra, partly because it serves as a paradigm for applications
of group theory and ring theory.
A quadratic equation is an equation having the form ax2 + bx + c = 0, where a = 0. There
is an intuitively evident symmetry in the formula for the solutions

b b2 4ac
x= .
2a
We mean to say, when there are two distinct solutions, the two solutions are given by the same
formula, differing from each other only in the choice of a sign, as if they were, so to speak,
mirror images of each other.
To express the symmetry more precisely, we shall be needing some abstract definitions.
Given a set S, a function S S S is called a binary operation on S. Let be a binary
operation on S, and let us write it as (a, b) 7 a b. We say that is associative provided
a (b c) = (a b) c for all a, b, c S. Note that, when is associative, we can write a b c
unambiguously. We say that is commutative provided a b = b a for all a, b S.
We define a field to be a set F equipped with a binary operation on F called addition and
a binary operation on F called multiplication, such that, writing the addition as (a, b) 7 a+b
and writing the multiplication as (a, b) 7 ab, the following nine conditions hold:
Zero Axiom: There exists an element 0 F such that, for all a F , we have 0 + a = a.
Negation Axiom: For all a F , there exists an element a F such that a + (a) = 0.
Additive Associativity Axiom: The addition is associative.
Additive Commutativity Axiom: The addition is commutative.
Unity Axiom: There exists an element 1 F such that 1 = 0 and 1a = a for all a F .
Inversion Axiom: For all a F , there exists an element a1 F such that a a1 = 1.
Multiplicative Associativity Axiom: The multiplication is associative.
Multiplicative Commutativity Axiom: The multiplication is commutative.

1
Distributivity Axiom We have a(b + c) = ab + ac for all a, b F .
The element 0 is unique. We mean to say that, given elements 0 and 0 such that 0+a = 0 =
0 + a for all a F , then 0 = 0 . Indeed, 0 = 0 + 0 = 0 + 0 = 0 . We call 0 the zero element
of F . Given a F , then the element a F satisfying a + (a) = 0 is unique. Indeed, given
b, b F satisfying a + b = 0 = a + b then b + a = 0 and b = b + 0 = b + a + b = 0 + b = b . We
call a the negative of a. Similarly, the element 1 is unique and, when a = 0, the element
a1 is unique. We call 1 the unity element of F and we call a1 the inverse of a.
Three examples of fields are the field C of complex numbers, the field R of real numbers,
the field Q of rational numbers. As another example, letting n be a positive integer, writing
Z/n = {0, 1, ..., n 2, n 1} and equipping Z/n with the evident addition and multiplication
obtained by taking remainders modulo n, then Z/n is a field if and only if n is prime.
When F and E are fields such that F E and such that the addition and multiplication
operations on F are restrictions of the addition and multiplication operations on E, we call
F a subfield of E and we write F E. If, furthermore, F E, then we call F a strict
subfield of E and we write F < E. For example, Q < R < C.
While we are in the mood for abstract definitions, let us introduce another one. We define a
group to be a non-empty set G equipped with a binary operation called the group operation
such that, writing the group operation as (g, h) 7 g h, the following three conditions hold:
Identity Axiom: There exists an element 1 G such that 1 g = g = g 1 for all g G.
Inversion Axiom: For all g G, there exists a g 1 G such that g g 1 = 1 = g 1 g.
Associativity Axiom: The group operation is associative.
Again, it is easy to check that the element 1 is unique and that, given g G, then the
element g 1 is unique. We call 1 the identity element of G. We call g 1 the inverse of g.
When the group operation is commutative, we call G an abelian group. This terminology
is in honour of Neils Heinrik Abel, one of the early pioneers of Galois theory.
To see two examples of abelian groups, consider a field F . Forgetting the multiplication
operation on F , then F becomes an abelian group under the addition operation. Note that,
for the group F under addition, the identity element is the zero element 0, and the inverse
of an element a of F is the element a. Forgetting the addition operation on F , and letting
F denote the set of non-zero elements of F , then F become an abelian group under the
multiplication operation. For the group F under multiplication, the identity element is the
unity element 1 of F and, given a F , the inverse of a as an element of the group F is the
inverse of a as a non-zero element of the field F .
Often, but not always, we write the operation on a group G as (g, h) 7 gh, calling the
operation multiplication and calling gh the product of g and h. Of course, this is just a
convention, and it must be avoided when it would cause confusion.
Proof of the following remark is easy, and we omit it.
Remark: Let G be a group and let H be a non-empty subset of G. Then the following
conditions are equivalent:
(a) The group operation G G G restricts to a group operation H H H.
(b) For all h, k H we have h1 H and hk H.
(c) For all h, k H, we have hk 1 H.
When the equivalent conditions in the latest remark hold, we call H a subgroup of G and

2
we write H G. If we also have H = G, then we call H a strict subgroup of G and we write
H < G. Thus, as a chain of subgroups under addition, we have Q < R < C and, as a chain of
subgroups under multiplication, we have Q < R < C .

Some of the motivations for the notion of a field are easy enough for a novice student to
grasp. For instance, the notion of a field supplies a suitable setting for many of the fundamental
concepts of linear algebra. Anyway, the student can readily appreciate that the axioms of a
field express some fundamental rules of arithmetic. Motivations for the notion of a group are
rather harder to grasp, despite the fact that, from a logical point of view, the definition of a
group is much simpler than the definition of a field. So let us spend some time indicating how
groups can arise in a meaningful mathematical scenario.
Group theory is a theory of symmetry. Usually, when groups appear in contexts of appli-
cation, they express symmetries of mathematical objects. To indicate a genuine application of
group theory, we shall touch on a few ideas from Galois theory.
It will be helpful to bear in mind the following old-fashioned way of expressing the definition
of a group. When the following four conditions hold, we call G a group under .
Closure Axiom: G is a non-empty set and, for all elements g and h of G, there is an element
g h of G.
Identity Axiom: There exists an element 1 G such that 1 g = g = g 1 for all g G.
Inversion Axiom: For all g G, there exists a g 1 G such that g g 1 = 1 = g 1 g.
Associativity Axiom: The group operation is associative.
Of course, this old-fashioned definition is equivalent to the modern definition presented
above. The two definitions amount to the same thing. In the modern definition, the Closure
Axiom has been assimilated into the definition of a binary operation. No matter which style of
definition one may prefer, the fact remains that, to confirm that some given thing is a group,
all four conditions do need to be checked: closure as well as unity, inversion and associativity.
We begin with the observation that the solutions to the quadratic equation x2 + 1 are x = i
and x = i. Complex conjugation is, in some sense, a symmetry which interchanges those two
solutions. In an geometrical way, we can view complex conjugation as a mirror symmetry of
the complex plane, with the real number line playing the role of the mirror. Thus, complex
conjugation sends each complex number to its reflection on the other side of the mirror. In
particular, the two solutions x = 1 are mirror images of each other.
Let us see what we can do to generalize that particular observation about complex conju-
gation. Given a field F , we define an automorphism of F to be a bijection : F F which
preserves the additive and multiplicative structure of F in the sense that (x + y) = (x) + (y)
and (xy) = (x)(y). The set of automorphisms of F , denoted Aut(F ), becomes a group
whose operation is composition of functions. Indeed, the composite of two automorphisms of
F is an automorphism of F , the identity function on F is an automorphism of F , the inverse
of an automorphism of F is an automorphism of F , and composition of automorphisms of F
is associative.
As an example, complex conjugation is an automorphism of C. We can regard complex
conjugation as a symmetry associated with R and C and the quadratic equation x2 1 = 0.

3
Exercise: Show that, for each of the fields Q, R, Z/p, where p is a prime, the identity
automorphism is the unique automorphism.
For simplicity of discussion, let us confine our attention to polynomial equations

an xn + ... + a1 x + a0 = 0

where n is a positive integer and the coefficients an , ..., a0 are rational numbers. When an = 0,
we say that the equation has degree n. The Fundamental Theorem of Algebra says that the
field C is algebraically closed, in other words, there exist complex numbers 1 , ..., n such
that, for all complex numbers x, we have

an xn + ... + a1 x + a0 = an (x 1 )...(x n ) .

Thus, x is a solution to the equation an xn + ... + a1 x + a0 = 0 if and only if x = j for some


1 j n. Consider the subfields L C such that all the solutions 1 , ..., n belong to E.
Let E be the intersection of all those fields. Plainly, E is a field, E owns all the solutions 1 ,
..., n , furthermore, E is the minimal subfield of C such that E owns all the solutions. The
minimality condition, here, is the condition that, given a subfield L of C such that L owns all
the solutions, then E L. Note that Q E. It can be shown that E is finite-dimensional as
a vector space over Q and, in particular, E < C. We call E the splitting field over Q for
the polynomial equation.
A polynomial equation of degree 2 is called a quadratic equation. Let us look briefly
at some quadratic equations. It is easy to show that, in the case of the equation x2 1 = 0,
the elements of the splitting field E over Q are those complex numbers that can be written in
the form u + iv where u, v Q. The automorphism group Aut(E) has precisely two elements,
namely, the identity map and the map coming from complex conjugation. Of course, the
non-identity automorphism interchanges the two solutions i to the equation.
The case of the equation x2 + 2 = 0 is very similar. We now let E be the splitting field for
x + 2 = 0 over Q. By some straightforward arguments, which
2
we omit, it can be shown that
E consists of the complex numbers having the form u + v 2 where, again, u, v Q. Let us
mention that the inverse of u + v 2 is

(u + v 2)1 = (u v 2)/(u2 2v 2 ) .

Note that the left-hand expression makes sense because u2 2v 2 = 0 by the irrationality of 2.
It is also fairly easy
to show that Aut(E) has a unique non-identity automorphism, namely,
the functionu + v 2 7 u v 2. Again, the non-identity automorphism interchanges the two
solutions 2.
Generally, it can be shown that, given any quadratic equation ax2 + bx + c = 0 with
a, b, c Q and a = 0, letting E be the splitting field over Q, then there is an automorphism
interchanging the two solutions if and only if the solutions do not belong to Q. Of course, for
the equation x2 3x + 2, the solutions are x = 1 and x = 2, the splitting field over Q is Q,
and there is no automorphism of Q interchanging the two solutions.
A polynomial equation of degree 3 is called a cubic equation. Let us examine two
equations having the form ax3 + bx2 + cx + d = 0 where a, b, c, d Q and a = 0. To avoid the
kind of degenerate behavior that we observed for quadratic equations with solutions in Q, we
shall make sure that neither of our two example exquations have any rational solutions.
As a first example, let us consider the cubic equation x3 2 = 0. The three solutions
are t and t and 2 t where t is the real cube root of 2 and = e2i/3 . Plainly, none of the

4
three solutions belongs to Q. Let E be the splitting field for this equation. Writing for the
automorphism of E coming from complex conjugation, then (t) = t and (t) = 2 t and
( 2 t) = t. Thus, writing 1 for the identity automorphism of E, and writing 2 = , we
have 2 = 1.
Now let us consider the cubic equation 3 + 2 2 1 = 0. For a contradiction, suppose
that the above cubic has a rational solution u/v, where u and v are integers with u 0 = v.
We can choose u and v such that u is as small as possible. Since 0 is not a solution to the
specified cubic equation, u 1. We have u3 + u2 v 2uv 2 v 3 = 0, hence every prime dividing
u must also divide v 3 . Hence, every prime dividing u must divide v. Similarly, every prime
dividing v must also divide u. But u is as small as possible, so u and v canot have any common
prime factors. We deduce that u = 1 and v = 1, hence u/v = 1. But this is impossible,
because 1 and 1 are plainly not solutions to the equation. We have proved that the equation
has no rational solution.
We claim that the three solutions are = c1 and = c2 and = c3 where

c1 = + 6 = 2 cos(2/7) = 2 cos(12/7) ,

c2 = 2 + 5 = 2 cos(4/7) = 2 cos(10/7) ,
c3 = 3 + 4 = 2 cos(6/7) = 2 cos(8/7) ,
and = e2i/7 . To see this, first observe that 7 = 1 and

(1 + + 2 + ... + 6 )(1 ) = 1 7 = 0

hence + 2 + ... + 6 = 1. We have

c1 c2 + c2 c3 + c1 c3 = ( + 6 )( 2 + 5 ) + ( 2 + 5 )( 3 + 4 ) + ( + 7 )( 3 + 4 )

= 2( + 2 + 3 + 4 + 5 + 6 ) = 2 .
By a similar method, it is easy to show that c1 + c2 + c3 = 1 and c1 c2 c3 = 1. Therefore

( c1 )( c2 )( c3 ) = 3 (c1 + c2 + c3 ) 2 + (c1 c2 + c2 c3 + c1 c3 ) + c1 c2 c3

= 3 + 2 2 1 .
The claim is established.
Now let E be the splitting field for the cubic equation and let be an automorphism of E.
Using the formula cos(2) = cos2 () 1, we obtain

c2 = c21 2 , c3 = c22 2 , c1 = c23 2 .

If (c1 ) = c1 then, since preserves the addition and multiplication operations on E, we have

(c2 ) = (c21 2) = (c1 )2 (2)c21 (2) .

But 1 is the unique element of E satisfying 1x = x for all x E. It follows that (1) = 1 and
(2) = (1 + 1) = (1) + (1) = 1 + 1 = 2. We have shown that if fixes c1 then fixes c2 .
Similar argument show that, if fixes any one of the solutions c1 , c2 , c3 , then fixes all three
of the solutions. We conclude that if = 1 then 3 = 1 but 2 = 1.

5
Thus, we have exhibited two cubic equations, neither of which have any solutions in Q, yet
the two equations have different symmetry properties, since only one of them gives rise to an
automorphism such that 2 = 1 = .
Galois theory is based on the following idea. Consider a field F and a polynomial equation
an xn + ... + a1 x + a0 = 0 where each ai F and an = 0. Let E be a field that is minimal,
in some suitable sense, subject to the condition that F E and there exist elements i E
such that, for all x E, we have an xn + ... + a1 x + a0 = an (x 1 )...(x n ). We call E
a splitting field for the equation. Let Gal(E/F ) denote the subgroup of Aut(E) consisting
of those automorphisms of F which satisfy (x) = x for all x F . It is not hard to see that,
given an element Gal(E/F ), then, for each index i, there exists an index j with (i ) = j ,
furthermore, is determined by the elements (i ). It follows that n! |Gal(F/E)|. In fact,
by Lagranges Theorem, below, it follows that n! is divisible by |Gal(F/E)|.
The group Gal(E/F ), called the Galois group of the polynomial equation, is to be inter-
preted as an expression of the symmetries of the equation. Properties of the equation, or rather,
properties of E as an extension of F , can be examined in terms of corresponding properties of
the group Gal(E/F ). For instance, the Fundamental Theorem of Galois Theory implies that,
in the case where F E C, there is a bijective correspondence K H between the fields
K satisfying F K E and the subgroups H of Gal(E/F ).
When n = 5, we call the equation a quintic. In that case, writing G = Gal(E/F ), then
|G| divides 120. It turns out that, when |G| = 60 or |G| = 120, the group G has a peculiar
property, called unsolvability. Using Galois theory it can be shown that, when |G| is 60 or 120,
none of the solutions to the equation can be expressed in terms of the elements of F using
addition, subtraction, multiplication, division and extraction of r-th roots, we mean, extraction
solutions to equations having the form xr a = 0 where r is a positive integer and a F .
In Algebra II, we shall find that, supposing G to be the Galois group of a quintic, if |G| = 60,
then G is a group called the alternating group of degree 5, denoted A5 , while if |G| = 120, then
G is a group called the symmetric group of degree 5, denoted S5 . As preparation for that,
we shall be proving, in Algebra I, that the groups A5 and S5 are unsolvable.

2: Lagranges Theorem and the Orbit-Stabilizer Equation


Since any genuine application of group theory requires substantial knowledge of other areas of
mathematics, we shall consider a toy application: counting the rigid symmetries of Platonic
solids. All of the material in this section has been more thoroughly discussed in class, and we
shall be giving only a quick summary below.
Recall that there are five kinds of Platonic solid: the tetrahedron, the octahedron, the
cube, the dodecahedron, the icosahedron. We understand the vertices of a Platonic solid to be
points in solid Euclidian space E3 . Each edge is connected to its nearest neighbouring vertices
by a straight edge, and the edges comprise the boundaries of flat faces.
A permutation of a set S is defined to be a bijection S S. A rigid transformation of
E is defined to be a permutation of E3 that preserves distances in the sense that d((x), (y)) =
3

d(x, y), where d(x, y) denotes the distance between points x, y E3 . A rigid symmetry of a
given Platonic solid is defined to be a rigid transformation of E3 that permutes the vertices,
we mean, restricts to a permutation of the vertices. It follows that such also permutes the
edges and the faces.
The number of rigid symmetries of a cube is 48. Indeed, there are 8 vertices, there are 6
rigid symmetries fixing a given vertex, and 8.6 = 48. Alternatively, there are 12 edges, 4 rigid
symmetries stabilizing a given edge, and 12.4 = 48. As one more argument, there are 6 faces,

6
8 rigid symmetries stabilizing a given face, and 6.8 = 48.
Similar calculations show that the number of rigid symmetries of an octahedron is 48. It
is no coincidence that this answer is the same as for the cube, because the octahedron and
the cube are duals to each other in the sense explained in class. Likewise, the dodecahedron
and the icosahedron are dual to each other, and they both have 120 rigid symmetries. Finally,
the tetrahedron, which is the dual of itself, has 4 vertices and 6 rigid symmetries stabilizing a
given vertex, hence 4.6 = 24 rigid symmetries in total.
We did not need any group theory in the calculations just above. Rather, we shall be
introducing some group theory as a way of clairifying the principles that we were applying.
We define a finite permutation group to be a pair (G, S) where S is a finite set and G
is a non-empty set of permutations of S such that the following condition holds:
Closure Axiom: For all elements g and h of G, the composite g h belongs to G.
Let us write the composite as gh = g h, and let us write the identity function on S as 1.
Remark: Let (G, S) be a finite permutation set. Then the set G is finite. We have 1 G.
For each element g of G, the inverse bijection g 1 belongs to G. The composition operation
G G G is associative.
Proof: The number of permutations of S is |S|! and, perforce, |G| is finite. So the terms of the
infinite sequence g 1 , g 2 , ... cannot be mutually distinct, and there must exist positive integers
i and k such that g i = g i+k . Hence 1 = g i g i = g i g i+k = g k , which is an element of G. Also,
g 1 = g 1 1 = g 1 g k = g k1 , which is an element of G. Finally, composition of functions is
associative and, in particular, composition as a binary operation on G is associative.
For g G and s S, we write gs = g(s). The set Gs = {g G : gs = s} is called the
stabilizer of s in G. In our above calculations for the number of rigid symmetries of a cube,
we were making use of the following principle.
Orbit-Stabilizer Equation: Let (G, S) be a finite permutation group. Suppose that, for all
s, t S there exists an element g G satisfying gs = t. Then |G| = |S||Gs |.
Proof: Fix an element s S. For each t S, let Gts = {g G : gs = t}. As a disjoint union,

G= Gts .
tS

The hypothesis on S implies that, for each t, there exists an element gt Gt . The function
Gs Gts given by g 7 gt g and the
function Gts Gs given by h 7 gt1 h are mutual inverses.
Therefore |Gs | = |Gs | and |G| = t |Gs | = |S||Gs |.
t t
The reader will have noticed that several groups appeared implicitly in the above discussion.
Given any set S, writing Sym(S) to denote the set of permuations of S, then Sym(S) becomes
a group under composition. We call Sym(S) the symmetric group on S. The set of rigid
transformations of E3 , denoted Aut(E3 ), is a subgroup of Sym(E3 ). The set of rigid symmetries
of a given Platonic solid is a subgroup of Aut(E3 ). The latest remark says that, given a finite
permutation set (G, S), then G is a subgroup of the finite group Sym(S) and, in particular, G
is a finite group. The alert reader will also have noticed something peculiar about the axioms.
In our definition of a finite permutation group, the only explicit axiom was the Closure Axiom.
Recall that the Closure Axiom is implicit in the modern definition of a group because it appears
in the definition of a binary operation. Yet all three of the axioms explicitly listed in the modern

7
definition of a group Identity, Inversion, Associativity arose not as hypotheses but as
conclusions in the latest remark.
But the proof of the latest remark breaks down if we drop the condition that S is finite. The
notion of a finite permutation group can be generalized as follows. We define a permutation
group to be a pair (G, S) where S is a set and G is a set of permutations of S such that the
following three conditions hold:
Closure Axiom: For all elements g and h of G, the composite gh = g h belongs to G.
Identity Axiom: The identity function 1 on S belongs to G.
Inversion Axiom: For all elements g of G, the inverse bijection g 1 belongs to G.
For an arbitrary permutation group (G, S), the composition operation on G is still asso-
ciative, hence the following remark.
Remark: Given a permutation group (G, S), then G is a group.
Often, in applications, the set S is a mathematical object with some structure, and the
group G acts on S in such a way as to preserve the structure. Then the elements of G can
be regarded, intuitively, as symmetries of S. The Closure Axiom says that two symmetry
operations combine to give a symmetry operation, and the Identity and Inversion Axioms
together say that symmetry operations are reversible.
The abstract notion of a group emerges when we discard the set S. When we do that,
though, we also discard the interpretation of the group operation as composition of functions.
That is why, in the abstract definition of a group, associativity has to be imposed as an axiom.
Given a group G, we define a permutation set for G to be a set S equipped with a function
: G Sym(S) such that (gh) = (g)(h) for all g, h G. We call a permutation
representation of G. The next remark says that the notion of a permutation set generalizes
the notion of a permutation group.
Remark: Let G be a group and let S be a permutation set for G with representation . Then
((G), S) is a permutation group.
Proof: We make use of an evident cancellation property for groups: given elements f, g, h G
such that f g = f h or gf = gh, then g = h.
Since (1)2 = (12 ) = (1) we have (1) = 1. Since (g)(g 1 = (g, g 1 ) = (1) = 1, we
have (g 1 ) = (g)1 . So the subset (G) of Sym(S) is closed under inverses. But (G) is also
closed under multiplication, so (G) is a subgroup of Sym(S).
A slightly different version of the Orbit-Stabilizer Equation is as follows. Much as before,
given a group G, a permutation set S for G and an element s S, we define the stabilizer of
s in G to be Gs = {g G : gs = s} as a subgroup of G. We say that S is transitive provided,
for all s, t S, we have t = gs for some g G.
Orbit-Stablizer Equation: Let G be a finite group, let S be a transitive permutation set for
G and let s S. Then S is finite and |G| = |S||Gs |.
Proof: The finiteness of S is clear. The rest of the argument is similar to the proof we gave
above for a variant of this result.

The next result can be seen as an abstract version of the Orbit-Stabilizer Equation. First,
we need some notation. Let G be a group, let H be a subgroup of G and let g G. We define

8
gH = {gh : h H} and Hg = {hg : h H}. We call gH a left coset of H in G and we call
Hg a right coset of H in G.
Lagranges Theorem: Let G be a finite group and let H G. Then |H| divides |G|.
Proof 1: We give a direct argument. Let be the relation on G such that g1 g2 provided
g2 = g1 h for some h H. Then is an equivalence relation, indeed, the reflectivity, symmetry
and transitivity properties of follow, respectively, from the identity, inversion and closure
properties of G. The equivalence classes under are precisely the left cosets of H in G.
Therefore the left cosets of H in G are mutulally disjoint.
The function H gH given by h 7 gh and the function gH H given by k 7 g 1 k are
mutual inverses. So |H| = |gH|. In other words, all the left cosets of H in G have the same
size. Therefore, writing m for the number of left cosets of H in G, we have |G| = m|H|.
Proof 2: Let S = {gH : g G}, as a transitive permutation set for G such that each f G
sends the coset gH to the coset f gH. The stabilizer of the coset H = 1H is H. The Orbit-
Stablizer Equation becomes |G| = |S||H|.

For an arbitrary group G and subgroup H, it is not hard to see that there is a bijective
correspondence gH Hg 1 between the left cosets and the right cosets. So the cardinality
of the set of left cosets is equal to the cardinality of the set of right cosets. We call that
cardinal number the index of H in G, denoted |G : H|. Thus, when G is finite, we have
|G : H| = |G|/|H|. As an infinite example, the set of even integers 2Z is a subgroup of the
additive group of integers Z, moreover, 2Z has finite index in Z and the index is |Z : 2Z| = 2.
We shall give two more versions of the Orbit-Stabilizer Equation. First, we need another
definition. Given a permutation set S for a group G, we define a relation =G on S such that,
given s, t S, then s =G t provided gs = t for some g G. It is easy to see that =G is an
equivalence relation. The equivalence class [s]G = {gs : g G} is called the G-orbit of s. As
a disjoint union,
S= [s]G
sG S

where the notation indicates that s runs over representatives of the G-orbits. Each G-orbit
[s]G is a transitive permutation set for G.
Orbit-Stabilizer Equation: Given a finite permutation set S for a finite group G, then
1
|S| = |G| .
|Gs |
sG S

Proof: By observations above, |S| = |[s]G | and each |[s]G | = |G|/|Gs |.

sG S

Actually, there was no need for us to assume that G is finite. The following generalization
of the result is not particularly useful, but it does at least provide some motivation for the
notion of a subgroup with finite index.
Orbit-Stablizer Equation: Let S be a finite permutation set for a group G. Then, for each
s S, the stabilizer subgroup Gs has finite index in G, and

|S| = |G : Gs | .
sG S

9
Proof: Again, S is the disjoint union of the G-orbits [s]G and |S| is the sum of the sizes of the
the G-orbits. So we may assume that S is transitive. Choosing an element s S and writing
H = Gs , we must show that H has finite index in G and that |S| = |G : H|.
Let L be the set of left cosets of H in G. The bijective correspondence S L given by
gs gH is well-defined because, for all f, g G the following four conditions are mutually
equivalent: the condition gs = f s; the condition f 1 gs = s; the condition f 1 g H; the
condition gH = f H. Hence |S| = |L| = |G : H|, as required.
Let us give an illustrative example. We let G become a permutation set for G such that
an element g G sends an element x G to the element gxg 1 G. The stablizer of x is the
subgroup
CG (x) = {g G : gx = xg}
which we call the centralizer of x in G. The orbit of x is

[x]G = {gxg 1 : g G} .

We call gxg 1 the conjugate of x by g, and we call [x]G the conjugacy class of x in G. Of
course, G is the disjoint union of the conjugacy classes,

G= [x]G
xG G

where x, here, runs over representatives of the conjugacy classes. Suppose now that G is finite.
Then the Orbit-Stablizer Equation says that each |[x]G | = |G : CG (x)|, whence

|G| = |[x]G | = |G : CG (x)| .
xG G xG G

In other words,
1
1= .
|CG (x)|
xG G

How it was possible for human beings to dream up the material we have been discussing?
Well, the definition of a group emerged very gradually, over the course of several decades during
the 19th century. Lagranges Theorem is so-named because, already during the 18th century,
Lagrange had observed that, when the n solutions to a polynomial equation are permuted, n!
is divisible by the number of values that can be taken by a given suitable function of those
n solutions. In Lagranges time, there was no explicit notion of a set, let alone an explicit
notion of a group. All the material in this section came about initially, in content, through
special cases which arose in applications, then subsequently, in form, through a long process
of clarification. Depite the impression that may be conveyed by the undergraduate literature,
mathematics does not progress as a logical game where abstract definitions are randomly tried
out and then aimless deductions are made.
Which version of the Orbit-Stabilizer Equation should one learn? But that is a misconceived
question, because all versions of the result express the same idea, and it is the idea that must
be learned and understood. Mathematical ideas are flexible. Indeed, the particular ideas that
we have been discussing have been evolving since the 18th century.

10

You might also like