1.

Group actions and other topics in group theory

October 11, 2014

The main topics considered here are group actions, the Sylow theorems, semi-direct
products, nilpotent and solvable groups, and simple groups.
See Preliminary remarks for some of the notation used here, especially regarding general
linear groups. Some further notation: [n] denotes the set of the first n natural numbers
1, 2, ..., n.
P k [n] denotes the set of k-element subsets of [n].

1 Group actions
1.1 Definition of a group action or G-set
Let G be a group, with identity element e. A left G-set is a set X equipped with a map
: G XX satisfying (i) (gh, x) = (g, (h, x)) for all g, h G and all x X, and (ii)
(e, x) = x for all x X. Usually we write either g x or simply juxtaposition gx for (g, x);
in the latter notation conditions (i) and (ii) become (gh)x = g(hx) and ex = x. We also call
this data a group action, or say that G acts on X (on the left).
Similarly a right G-set is a set X equipped with a map : X GX satisfying (in
the evident juxtaposition notation) x(gh) = (xg)h and xe = x. Of course the distinction
between left and right G-actions does not depend on whether we write the domain of as
G X or X G. The distinction is that in a left action gh acts by h first, then g, whereas
in a right action g acts first, then h.
Notice that up to this point, we havent even used the existence of inverses, so exactly
the same definition makes sense for left and right monoid actions. We will make little use
of monoid actions, however. One immediate advantage of the existence of inverses is that
any right action can be converted to a left action by setting g x = xg 1 ; similarly any left
action can be converted to a right action. Nevertheless it is important to pay close attention
to which side the group is acting on. If the side is not specified we always mean a left action
(an arbitrary choice on my part!). Any statement about left actions has a parallel statement
for right actions; we leave it to the reader to make the translation.

1.2 Some fundamental terminology

Let X, Y be (left) G-sets. A map : XY is a G-map or a G-equivariant map if (gx) =
g(x) for all g G, x X. Then G-sets and G-maps form a category that we will denote

1
G-Set.

The fixed-point set X G is defined by X G = {x X : gx = x g G}. This defines

a functor G-set Set in the evident way. The isotropy group Gx of a point x X is
defined by Gx = {g G : gx = x}; clearly Gx is a subgroup. Thus the fixed-points are the
points x with Gx = G. The action is trivial if every point is a fixed point. At the opposite
extreme, the action is free if Gx = {e} for all x X.

Define an equivalence relation on X by x y if there exists g G such that gx = y. A

equivalence class is called an orbit, usually denoted O. The orbit determined by a particular
x X is denoted Ox or Gx. The set of all orbits of a left action is denoted G\X; the set
of orbits of a right action is denoted X/G. This notational distinction is important because
we will often have groups acting on the left and the right of the same set X.
The action is transitive if there is only one orbit. In other words, for all x, y X there
exists a g such that gx = y. Transitive actions will be discussed in more detail in a later
section.
If O X is an orbit and x O, then the map GX given by g 7 gx factors through

=
a G-bijection G/Gx O = Gx. Hence [G : Gx ] = |Gx| (this is true even when the sets in
question are infinite, but we have in mind here the finite case).

Finally, we recall a simple but very powerful counting formula. Suppose the group G
acts on the finite set X. Then
X X
|X| = |O| = [G : Gx ],
O xG\X

where the first sum is over the orbits O of the action and the second sum means, in a
mildly abusive notation, that we are taking a fixed representative x of each orbit. This
choice of x O is arbitrary, but the sum is nevertheless well-defined since [G : Gx ] = |O| is
independent of the choice.

2 Examples
1. The symmetric group Sn acts on the left of [n] := {1, 2, ..., n} by permutions. The action
is transitive, with the isotropy group of any point isomorphic to Sn1 .
More generally, if X is any set, we let P erm X denote the group of bijections XX.
Then by construction P erm X acts on the left of X by x = (x). It is a left action
because a composition acts by first, then .

2. If G is any group, H any subgroup, then the left translation action of H on G is

defined by h g = hg for h H, g G. The right translation action is given by g h = gh.
These are both free actions. The orbit space H\G of the left action is by definition the set
of right cosets Hg, while the orbit space G/H of the right action consists of the left cosets
gH.
If G is finite, then (since every orbit has size |H|) the counting formula just says that
|G| = |H| [G : H].

2
 3. If G is any group, the left conjugation action of G on itself is given by g x = gxg 1 .
Similarly right conjugation is defined by x g = g 1 xg. The fixed-point set of either action
is the center C(G) (Z(G) is another common notation for the center). The orbits are the
conjugacy classes of G. The isotropy group of x is CG x, the centralizer of x in G.
In this case the counting formula yields the class equation. To state it we need a notation
for conjugacy classes, andsadlywe have already assigned the letter C to centers and
centralizers. I will use the non-standard notation (x) to mean the conjugacy class of x, and
Conj G to mean the set of conjugacy classes. We then have
X X
|G| = |(x)| = [G : CG x].
xConj G xConj G

Once again the notation has the obvious interpretation: We are choosing one x from each
conjugacy class, and the choice doesnt matter.

4. If G is any group, let S(G) denote the set of subgroups of G. Then G acts on S(G)
by left conjugation: g H = gHg 1 (there is also a right conjugation, of course). The
fixed-points are the normal subgroups. The orbits are conjugacy classes of subgroups. The
isotropy group of H is the normalizer NG H of H in G.

5. Let X, Y be sets and let F (X, Y ) denote the set of functions XY . If Y is a left
G-set, we get a left G-action on F (X, Y ) by (g )(x) = g((x)). If X is a left G-set, we get
a right G-action on F (X, Y ) by ( g)(x) = (gx). Note carefully that this is a right action.
However, we can always convert it to a left action by (g ? )(x) = (g 1 x). If both X and
Y are G-sets, we can get a combined left action of G on F (X, Y ) by (g )(x) = g(g 1 x).
The fixed-point set of the combined left action gg 1 is the subset of G-equivariant
maps XY , as is easily checked.

6. Let X be a set, X n the n-fold Cartesian product of X with itself. Then Sn acts on
X n by permuting the coordinates. This is a right action, given explicitly by

(x1 , ..., xn ) = (x(1) , ...x(n) ).

One could check directly that this is a right action, but easier is to note that X n = F ([n], X),
where Sn acts on the domain, on the left. So this is a special case of the previous example.
The fixed-point set is the diagonal subset of all (x, x, ..., x). Contemplation of the orbits and
isotropy groups is left to the reader.

7. Projective spaces. The purpose of this example is twofold. First of all, projective
spaces are ubiquitous in topology and geometryincluding especially algebraic geometry,
which leads me to discuss them in an algebra course. Second, in nature we are often first
confronted not with a group action or even a group, but with a set (or topological space,
etc.) X that may secretly be equipped with a useful group action and/or realization as the
orbit set of a group action. Its important to be able to recognize such structures.
Let F be a field, and let V be a finite dimensional vector space over F . (In fact the
finite-dimensionality isnt necessary, but I prefer to avoid distractions.) The projective space
P(V ) is the set of lines through the origin in V . Even though at the moment we are not giving

3
it any topology, we call it a space because that is the traditional term, and if you call it
the projective set you risk being sneered at as an ignorant yokel from the backcountry.
Thus P(V ) does not involve a group in its definition, but it is in fact crawling with groups.
First of all, it is the orbit set of the action of F on V {0} by scalar multiplication. This
is a very handy interpretation. Second, GL(V ) acts transitively on it: By elementary linear
algebra, any line can be moved to any other line by an invertible linear transformation.
This is a particularly important interpretation; it is often useful to recognize a set as a
transitive G-set (also known as a homogeneous space). To complete the picture we choose
a convenient point in the set and determine its isotropy group. Here there is no natural
choice of a line; we just pick one and call it L0 . The isotropy group H consists of invertible
transformations preserving L0 , i.e. the set of A GL(V ) such that A has an eigenvector in

=
L0 . Then there is a G-isomorphism GL(V )/H P(V ). If we want to be more explicit, we
choose a basis e1 , ..., en and take L0 = he1 i. Then H corresponds to the group of matrices
with ai1 = 0 for i > 1 (i.e. the first column is zero except for a11 ).

3 Actions preserving some additional structure

Frequently, the G-sets X one encounters are not merely sets but have some extra structure
preserved by the action. Some examples:

Example. X is itself a group, and G is acting on it via group automorphisms. In other words,
g (xy) = (g x)(g y) for all g G, x, y X. The action of G on itself by conjugation is an
action of this type. Actions via group automorphisms will be used to construct semi-direct
products later.

Example. We have a vector space V over a field F , and the G-action is linear: g (v + w) =
g v + g w, and g (cv) = cg v (c F ). This type of action is called a representation of
G over F. In this course representation will always be taken to mean finite dimensional
representation, unless otherwise specified. Note that GL(V ) acts linearly on V by definition;
we call this the standard representation of GL(V ).
Representation theory is one of the major branches of mathematics. Well consider
representation theory of finite groups in some detail, especially over C.

Example. Let F be a field, and suppose G acts on F via field automorphisms. This is
precisely the situation one studies in Galois theory.

Example. Let X be a topological space, and suppose G acts on X via homeomorphisms. In

other words, for each fixed g, the map x 7 g x is a homeomorphism. In fact we only need
to check this map is continuous, since it automatically has an inverse given by the action
of g 1 . In the topological case, however, G itself might be a topological groupi.e. both a
space and a group, with the multiplication and inverse maps continuous. In that context a
topological group action means a group action such that the action map G XX is
continuous (G X has the product topology). This is usually a much stronger assumption
than merely saying that G is acting by homeomorphisms.

4
 There are many variants on this theme: groups acting on metric spaces by isometries,
smooth Lie groups actions on smooth manifolds (studied in our Manifolds course), sim-
plicial actions on simplicial complexes...all of these are dear to my heart, but lie outside the
scope an algebra course.

4 An alternate view of group actions

Let X be a G-set. Then each g G defines a bijection XX, so we get a map :
GP erm X. Moreover, the axioms for a group action translate into the statement that
is a group homomorphism. Conversely if a homomorphism : GP erm X is given,
we get a G-action on X by g x = (g)(x). Thus G-actions on X are the same thing as
homomorphisms GP erm X. If is injective we say that the action is faithul (effective
is another commonly used term). Note that free actions are faithful, but not conversely: the
action of Sn on [n] is faithful, but certainly not free. Indeed an action is free if and only
if all isotropy groups are trivial, whereas an action is faithful if and only if the intersection
xX Gx of all the isotropy groups is trivial.
In analyzing the structure of a newly encountered group G, optimists hope to find a proper
non-trivial normal subgroup H such that H and G/H are already known, or at least more
approachable. Since kernels are always normal, Ker is at least a candidate. Moreover, even
when Ker is trivial (i.e. the action is faithul), thereby sinking the optimists strategy, we
get an injective homomorphism GP erm X that may yet prove useful for understanding
G.
Lets consider the case when G is finite and X is a finite G-set. If |X| = n, a choice of
ordering of X yields an isomorphism P erm X = Sn , so well think of as a homomorphism
GSn . But to get any use out of this method we first need to find some finite G-sets.
The most obvious candidate is G itself, with the left translation action. Since the action
is free and hence faithful, this yields an injective homomorphism : GSn . Thus every
finite group is isomorphic to a subgroup of some Sn , a result known as Cayleys theorem.
The n obtained, however, is n = |G|, and |Sn | is too large for this result to be of more than
occasional use.
To find smaller G-sets, we can choose a subgroup H and consider X = G/H with its trans-
lation action. But how do we find subgroups of a general finite G? The Sylow p-subgroups
of G and their normalizers, discussed in the next section, form the most important and most
useful family of subgroups that are defined for an arbitrary finite group. Alternatively, one
may be able to exploit special features of a particular G. As a simple, fun example, consider
GL2 F2 . It acts linearly on F22 , a vector space having a grand total of four elements. So there
are three nonzero vectors, and after choosing an ordering of them, we obtain a homomor-
phism : GL2 F2 S3 . This is especially useful, as it turns out to be an isomorphism
(exercise).

Remarks. 1. If X is a G-set with extra structure, then : GP erm X lands in the

automorphism group of this structure. In example 1 we get : GAutgrp X, in example 2
we get : GGL(V ), and so on. Indeed if C is any category, we can define a G-object in
the category to be an object X together with a homomorphism GAutC X.

5
 2. Suppose X1 , X2 are G-sets with the same underlying set X but with different G-actions.
Let 1 , 2 be the corresponding homomorphisms GP erm X. Then X1 is isomorphic to
X2 as a G-set if and only if 1 , 2 are conjugate as homomorphisms to P erm X. (Exercise.
By conjugate I mean there is a P erm X such that 1 1 = 2 . )

3. If X is a non-faithful G-set, with H = Ker , then the G-action factors through a

G/H-action: (gH)x = gx. This is well-defined since H is normal. For example, consider the
action of GL(V ) on P(V ) discussed earlier. The kernel of this action is the scalar matrices
F GL(V ), so we get an action of GL(V )/F on P(V ). This is the reason GL(V )/F is
called the projective general linear group, denoted P GLn F .

5 Transitive G-sets
Recall that a G-set X is transitive if there is only one orbit, and that in this case a choice

=
of x X yields an isomorphism of G-sets G/Gx X. Notice, however, that the choice of
x is arbitrary. This is true even for a free transitive action: then we get an isomorphism of

=
G-sets G X, where G has the left translation action, but there is no natural choice of
such an isomorphism; it depends on the choice of x.
The next result is basic.
Proposition 5.1 Let X be a transitive G-set. Then the isotropy groups form a complete
conjugacy class of subgroups of G.
Proof: Suppose x, y X. Choose g G with gx = y. Then gGx g 1 = Gy , as is readily
checked. Hence any two isotropy groups are conjugate. Conversely, let H be a subgroup
conjugate to Gx ; say gGx g 1 = H. Then H = Ggx , so every subgroup in the conjugacy class
occurs as an isotropy group.

Now suppose we have two subgroups K, H and we want to show that K is conjugate
to a subgroup of H, a problem that arises quite frequently. In the spirit of the previous
proposition we have at once:
Proposition 5.2 K is conjugate to a subgroup of H if and only if the left action of K on
G/H has a fixed point. More precisely, KxH = xH if and only if x1 Kx H.
Now lets consider the set of all orbits of the K-action on G/H. These are called the
(K, H)-double cosets, denoted K\G/H. We could just as well think of K\G/H as the H-
orbits of the right action on K\G, or more symmetrically as the orbits of the left KH-action
on G given by (k, h) g = kgh1 . More often, however, we stick with the first interpretation.
Here is an example known as the Bruhat decomposition:
Proposition 5.3 Let F be a field, and let B := Bn F . Then the (B, B)-double cosets of
GLn F are given by
a
GLn F = BwB,
wW
where W
= Sn is the Weyl group.

6
 The proof is by old-fashioned row and column reduction, and is left to the reader. Whats
not obvious is why this particular way of arranging the row/column reduction is of interest. It
turns out that it plays an important role in the structure theory of a large class of interesting
groups (not just the general linear groups), and enters into algebraic geometry and topology
via Schubert varieties and Schubert cells. For example, when F = R or F = C then the
orbit set GLn F/B is a topological space known as a flag manifold or flag variety, and the
left B-orbitswhich by the proposition are indexed by elements of W are homeomorphic
to vector spaces over F and called Schubert cells. They are very useful for studying the
geometry and topology of flag manifolds, a subject which is in itself a major industry these
days. I mention all this just to pique your curiosity, and to suggest how the humble process
of row reduction connects with beautiful, deep mathematics.

6 A fixed-point theorem, and the Sylow theorems

Throughout this section, p is a prime. Everything will be based on the following simple
fixed-point theorem:

Theorem 6.1 Let P be a finite p-group, S a finite P -set. Then |S| = |S P | mod p.

Proof: |S| = |O|, where O ranges over the P -orbits. Since P is a p-group, |O| is either
P

divisible by p or consists of a single fixed point, whence the result.

Corollary 6.2 If |S| is prime to p, then there is at least one fixed point.

As an immediate application of the corollary, we have:

Proposition 6.3 Every finite p-group G has non-trivial center.

Proof: Consider the action of G by conjugation on G {e}. By the corollary it has a

fixed-point, so G has non-trivial center.

The proposition in turn has a nice corollary:

Corollary 6.4 If G has order p2 (p a prime) then G is abelian.

Proof: By the proposition, G has a central subgroup H of order p. Hence |G/H| = p, so

G/H is cyclic. But whenever a group G has a central subgroup with cyclic quotient group,
G is abelian (why?).

The next application can be proved using the binomial theorem, but we can also deduce
it from Theorem 6.1.
 
n
Proposition 6.5 Let n = spk . Then pk
= s mod p.

7
Proof: Partition [n] into s disjoint subsets A1 , ..., As of size pk . Let Ci Sn be a cyclic
group of order pk that permutes Ai transitively and fixes the other Aj s pointwise, and take
P = C1 ... Cs . Then the action of P on P pk [n] has exactly s fixed points, namely
A1 , ..., As . Hence the proposition follows from the theorem.

Let G be a finite group, and write |G| = spk with s prime to p. A p-Sylow subgroup is
a subgroup P of order pk . The first Sylow theorem asserts that such subgroups always
exist.

Theorem 6.6 G has a p-Sylow subgroup.

Proof: The strategy is to look for a finite G-set S such that some isotropy group Gx is a
p-Sylow subgroup. Indeed, suppose we had a finite G-set S such that (i) |S| is prime to p;
and (ii) all isotropy groups Gx are p-groups. Then by (i) there is an orbit O with |O| prime
to p. Choose x O; then |Gx | = pi for some i by (ii). But |G| = |Gx | |O|, forcing i = k.
Hence Gx is a p-Sylow subgroup.
It remains to exhibit such an S. Take S to be the set of subsets of G of size pk , with
action induced by the left translation action of G on itself. Then (i) holds by Proposition 6.5.
Now let A S. Then GA acts freely on the elements of A (since the left translation action
is free), so |GA | divides pk , proving (ii).

The second Sylow theorem says that all p-Sylow subgroups are conjugate. More
precisely:

Theorem 6.7 Let H be a p-subgroup of G, P a p-Sylow subgroup. Then H is conjugate to

a subgroup of P . In particular, any two p-Sylow subgroups are conjugate.

Proof: As discussed earlier, this is equivalent to saying that the action of H on G/P has
a fixed point: if HxP = xP , then x1 Hx P and conversely. But H is a p-group and
|G/P | = s is prime to p, so this follows immediately from Theorem 6.1.

The last item of business is to say something about the set of all p-Sylow subgroups of
G. How many such subgroups are there? By the second Sylow theorem, G acts transitively
on this set by conjugation. If we fix a p-Sylow subgroup P , the isotropy group of the action
is the normalizer NG P . Hence the number of p-Sylow subgroups is [G : NG P ]. This brings
us to the third Sylow theorem:

Theorem 6.8 Let np G denote the number of distinct p-Sylow subgroups of G. Then np G
divides |G| and np G = 1 mod p.

Proof: Since np G = [G : NG P ] for any choice of p-Sylow subgroup P , np G divides |G|.

Now fix a p-Sylow subgroup P , and note that by the second Sylow theorem P is the unique
p-Sylow subgroup of NG P (since it is a normal p-Sylow subgroup of NG P ). Now consider the
left translation action of P on G/NG P . If P xNG P = xNG P then x1 P x NG P , forcing
x1 P x = P since P is the unique p-Sylow subgroup of NG P . Hence x NG P ; in other
words, the P -action has a unique fixed point. Using Theorem 6.1 we conclude

8
 np G = |G/NG P | = (G/NG P )P = 1 mod p.

Heres another interesting fact about p-Sylow subgroups:

Proposition 6.9 Let P be a p-Sylow subgroup of G, and suppose NG P H. Then NG H =

H. (In particular, this is true for H = NG P .)

Proof: Suppose gHg 1 = H. Then gP g 1 is a p-Sylow subgroup of H, so by the second

Sylow theorem there is an h H such that hP h1 = gP g 1 . Then h1 g NG P , so g H.

Many examples and applications of p-Sylow subgroups can be found in the exercises.

7 New G-sets from old: restriction, disjoint unions,

products, and induction
Change of group and restriction. Suppose X is a left G-set and : HG a homomorphism.
Then X is a left H-set with action h x = (h)x. Thus defines a functor : G-set
H-set. The case when is inclusion of a subgroup is of particular importance; in this case we
use the notation XH for X regarded as an H-set, and call X 7 XH the restriction functor.
`
Disjoint unions. Suppose X and Y are G-sets. The disjoint union X Y is a G-set in the
obvious way, and in fact is the categorical coproduct. Similarly if X is any collection of G-
`
sets indexed by a set J, J X is a G-set and is the categorical coproduct. The fixed-point
set and orbit set functors take coproducts of G-sets to coproducts of sets.

Products of G-sets. Again let X, Y be G-sets. The product X Y is a G-set via the diagonal
action: g (x, y) = (gx, gy). The product of any collection of G-sets is defined similarly,
and is the categorical product. The fixed-point functor takes products to products, e.g.
(X Y )G = X G Y G . Orbits, however, are another matter; there is no simple relationship
between G\(X Y ) and G\X, G\Y . We will see some examples later.

Balanced products. Suppose X is a right G-set, Y a left G-set. Define an equivalence relation
on X Y by (xg, y) (x, gy). The balanced product X G Y is the set of equivalence classes.
In fact this is just the orbit set of the left G-set X Y , where g acts by g (x, y) = (xg 1 , gy).
But the slight change in viewpoint can be useful and enlightening. For our immediate
purposes, the induced G-spaces below provide the most important example.

Induction. Suppose H is a subgroup of G and X is a left H-set. Then the balanced product
GH X is a left G-set with action g1 [g, x] = [g1 g, x]; we say that the G-action is induced from
the H-action. Here the brackets [] denote equivalence class in the balanced product. Thus
X 7 G H X defines the induction functor H-set G-set (the definition of the functor
on morphisms being obvious). Note that the map i : XG H X given by i(x) = [e, x] is
an H-map.
Induction has the following universal property (see the category theory notes for a general
discussion of such properties). We keep the above notation.

9
Proposition 7.1 Suppose Y is a G-space, X is an H-space and : XY is an H-map.
Then there is a unique G-map : G H XY such that the following diagram commutes:
i
X - G H X
p
p pp
p

p pp
pp
!
p
?
Y

Proof: There is no choice in the definition of : We must take ([g, x]) = g(x). Now check
that it works (in particular, check that is well-defined).

As usual, we give two alternate ways of thinking about the universal property:

Plain English version. If you want to define a G-map G H XY , it is enough (indeed

equivalent) to define an H-map XY .

Adjoint functor version. Induction H-set G-set is left adjoint to restriction G-set
H-set.
See the category theory notes for discussion of adjoint functors. In essence there is not
much to it in the present example; the universal property translates immediately to the
assertion that there is a bijection

HomG (G H X, Y )
= HomH (X, Y ).
To complete the proof that these are adjoint functors, one has to show that the above
bijection is natural in X and Y . This is easy once one has absorbed the definition of
natural transformation, but it is not essential to understand all this right away. Just use
the universal property.

There is a recognition principle for induced G-sets. Let Y be a G-set, X Y an H-

invariant subset (H G). Applying the universal property to the inclusion j : XY , we
get a G-map : G H XY ; indeed it is just ([g, x]) = gx.

Proposition 7.2 is bijective if and only if (i) for all y Y , there is a g G such that
gy X; and (ii) whenever x1 , x2 X and gx1 = x2 , we have g H.

The proof is a straightforward check; (i) gives the surjectivity and (ii) the injectivity.
Note that (ii) says that if g
/ H, then g moves every element of X to an element not in X.

Example. Let F be a field, G = GLn F , and Y the set of pairs (L, v) with L a line in F n and
v L. We have an evident G-action on Y given by g (L, v) = (gL, gv). Let L0 denote the
line spanned by the standard basis vector e1 , and let X = {(L, v) Y : L = L0 }. Then X
is invariant under H := {g GLn F : gL0 = L0 }. Now lets check conditions (i) and (ii) of
the recognition principle above: (i) is clear, since by linear algebra GLn F acts transitively

10
on the lines. And if (gL0 , gv) = (L0 , w) then g H. So we have a canonical isomorphism

=
of G-sets G H X Y . To complete this description, one should describe how H acts
on X. Identifying X with the vector space L0 , it is the linear action that pulls back scalar
multiplication along the homomorphism HF taking a matrix A to a11 , the upper left
entry: A (L0 , v) = (L0 , a11 v).
Examples of this type occur frequently in geometry and topology.

8 Semi-direct products and group extensions

8.1 Prelude on products
One of the most commonly used techniques in mathematicsand here I admit to stating the
obvious, but bear with meis to combine simple objects in some way to build complicated
objects, and conversely to understand complicated objects by breaking them down into
simpler objects. In the case of groups, for example, given groups H, K we can form a new
group by taking the product H K. Conversely, if we can decompose a given group G as a
product G = H K, where H and K are already understood, thenat least in principlewe
can understand G. To carry out this latter strategy we need a recognition principle for such
products. The reader has probably already seen this, but here is a reminder of how it works
for an arbitrary finite number of factors:

Proposition 8.1 Suppose H1 , ..., Hn are normal subgroups of G such that (i) Hi ( j6=i Hj ) =
Q

{e} and (ii) G = H1 ...Hn . Then the multiplication map m : H1 H2 ... Hn G is an

isomorphism of groups.

Proof: First note that m is a group homomorphism: For this, one needs to know that for
i 6= j the elements of Hi , Hj commute with one another. But if x Hi and y Hj , then
by normality xyx1 y 1 Hi Hj = {e}. Then m is injective by (i) and surjective by (ii),
completing the proof.

8.2 Semi-direct products

But only if we are lucky will G decompose as a product. The next best thing is a semi-
direct product, which we know describe. Suppose given groups H, K and a homomorphism
: KAut H. Equivalently, we are given a left action of K on H by group automorphisms.
Associated to this data we have the semi-direct product H o K: As a set it is just H K,
but with group multiplication defined by

(h1 , k1 ) (h2 , k2 ) = (h1 ((k1 )(h2 )), k1 k2 ).

The proof that this multiplication defines a group structure is left as an exercise. Note
that (h1 , e) (e, k2 ) = (h1 , k2 ). So there is no harm in dropping the parentheses and writing
hk in place of (h, k). Note also that the outer two factors in the displayed product just go
along for the ride; all the action takes place with the inner two. Thus the essence of the
multiplication rule is that it tells you how to commute k with h: kh = ((k)(h))k. But we

11
already know how to do this, in any group: kh = (khk 1 )k. The upshot of this discussion
is that in the semi-direct product, the automorphism (k) of H is the same thing as left
conjugation by k.
Note that by construction H is normal in H o K, and that H o K is the direct product
H K if and only if is trivial if and only if K is normal (check this!).
If is understood, we often omit it from the notation and simply write H o K. Needless
to say, abusive notation of this kind must be used with care.

Example. Suppose we ask: Are there non-abelian groups of order 21? With semi-direct
products in hand, it is easy to construct such groups explicitly: Aut C7 is cyclic of order
6, so we can choose an injective homomorphism : C3 Aut C7 (there are two such ho-
momorphisms) and form C7 o C3 . More generally, given primes p, q with p|q 1, we get
non-abelian groups of order pq this way. Indeed one can show that every group of order pq
has this form.

More importantly, many groups occuring in nature can be recognized as semi-direct

products and thereby better understood. Here is a simple recognition principle:

Proposition 8.2 Suppose G contains subgroups H, K such that:

(a) H is normal in G.
(b) H K = e.
(c) HK = G.
Then G = H o K, where (k)(h) = khk 1 .

Proof: Conditions (b),(c) imply that multiplication H KG is a bijection. Since H is

normal, the stated homomorphism is defined, and a trivial check (which we have in effect
already done above) shows that the multiplication on G is exactly the semi-direct product
multiplication.

With this criterion in hand, you soon realize that semi-direct products are everywhere.
Here are a few important examples, with details and verifications left to the reader:

Examples. 1. The dihedral group of order 2n is a semi-direct product Cn o C2 , where C2

acts on Cn as multiplication by 1.
2. Sn = An o C2 .
3. Let V be a finite dimensional vector space over a field F . The affine group Af f (V ) is
defined to by the subgroup of P erm V generated by GL(V ) and the subgroup of translations
Tv : x 7 x + v . The group of translations is isomorphic to the additive group of V , and
Af f (V ) = V o GL(V ). Here the action of GL(V ) on V by conjugation is the same as the
standard action. When V = F n , we also write Af fn F in place of Af f (F n ).
4. Consider the symmetric group Sp and the p-Sylow subgroup Cp generated by the
standard p-cycle (12...p). Then the normalizer NSp Cp is isomorphic to Af f (Fp ) and hence is
a semi-direct product of the form Cp o Cp1 .
5. Bn F = Un F o Dn F . What is the action of Dn F on Un F ?
6. Nn F = Dn F o Sn , where Sn acts on Dn F = (F )n by permuting the factors. This is
a type of semi-direct product known as a wreath product, disussed further in the exercises.

12
8.3 Group extensions
A group extension consists of group homomorphisms
i
H G K,
where is surjective and i is an isomorphism onto the kernel of . Thus without loss of
generality we can, if desired, assume that i is just an inclusion and K = G/H. In fact we
will often treat H as a subgroup and omit i from the notation.
Note that H and K alone do not determine G, even when G is abelian. For example, if
we are given a group extension C2 GC2 , then since |G| = 4 we know G is abelian, but
without further information there is no way to know whether G is C4 or C2 C2 .
The extension is central if H C(G). Note, for example, that if H = C2 then the
extension is automatically central. This leads to another example of the ambiguity inherent
in group extensions: If we have an extension C2 GC2 C2 , then it is a central extension
but G could be any of the five groups of order 8 except C8 : (C2 )3 , C2 C4 , the dihedral
group D8 , or the quaternion group Q8 . Each of these four groups fits into such an extension,
as you can easily check.
Note that a semi-direct product G := H o K fits into an extension HGK. In
fact we can characterize the semi-direct products in terms of extensions. A group extension
i
H G K splits if there is a homomorphism s : KG such that s = IdK . We
call s a splitting (or sometimes a section) of .
Proposition 8.3 If G := H o K is a semi-direct product, the extension HGK splits.
Conversely if HGK is a split group extension, then G = H o K with K acting on H by
conjugation. More precisely, if s : KG is a splitting, K acts on H by k h = s(k)hs(k)1 .
Proof: If G = H o K, define s by s(k) = (e, k). Conversely if HGK is split, choose a
splitting s : KG. Then the pair H, s(K) satisfies the recognition principle for semi-direct
products (an easy check).

Remark. Note that to give a splitting s : KG is the same thing as giving a subgroup
K 0 G such that : GK maps K 0 isomorphically to K.

Example. Let G be a group of order pq, where p, q are primes with p < q. I claim that G is
a semi-direct product of the form Cq o Cp . First of all, by the third Sylow theorem there is
a unique and hence normal q-Sylow subgroup, cyclic of order q. Hence there is an extension

Cq G Cp . Now choose a p-Sylow subgroup H. Then |H is injective, and hence
an isomorphism. This proves the claim. Note that Cp acts on Cq by some homomorphism
Cp Aut Cq = Cq1 , and hence for the action to be non-trivial we must have p|(q 1) or
equivalently q = 1 mod p. This fits with the third Sylow theorem because if H is not normal,
then there are q p-Sylow subgroups.

9 Solvable and nilpotent groups

In attempting to analyze a group G by fitting it into an extension HGKor equiva-
lently, finding a normal subgroup H with quotient Kone simple possibility we might hope

13
for is to find such an extension with both H and K abelian. Or if we are feeling especially
lucky, we might hope that in addition the extension is central. A more reasonable although
still optimistic hope is that we can build G by a finite iteration of such extensions. This
leads to the concepts solvable and nilpotent group in the respective cases.

Example. We can build S4 in two steps out of abelian groups. First we form the extension
C22 A4 C3 . Then we form the extension A4 S4 C2 .

But we need a smoother way to think about building up from extensions. This is the
subject of the next section.

9.1 Dont fight it, filter it!

An increasing filtration of a group G consists of subgroups

{e} G1 G2 G3 ....
A decreasing filtration likewise consists of subgroups

G G1 G2 G3 ...
In either case the filtration stabilizes at Gn if Gk = Gn for all k n. An increasing (resp.
decreasing) filtration is finite if Gn = G for some n (resp. Gn = {e} for some n). For finite
filtrations there is no real difference between the increasing and decreasing cases, since one
could always reverse the ordering to convert from one to the other.
In fact this definition makes sense for any kind of object with subobjects: rings and
subrings, vector spaces and sub-vector spaces, topological spaces and subspaces, etc. In
group theory filtrations are classically known as series. I prefer the term filtration
because it has a verb, to filter, that goes with it, and because it is the more widely used
term across many different categories. But I will freely use both terms, just so you get used
to them.

Example. Let p be a prime. Any abelian group A has a natural decreasing p-adic filtration
A pA p2 A ... (which need not terminate; think of A = Z), as well as a natural p-torsion
filtration A[p] A[p2 ] ... (which need not terminate; think of A = Q/Z). Incidentally, it
isnt necessary for p to be a prime here, but the prime case is by far the most important.

Example. Note that according to our definition, a filtration of a finite group need not be
a finite filtration. For example, suppose p, q are distinct primes, and G is a finite abelian
q-group. Then pG = G and hence the p-adic filtration takes the form G G G ...; it
never reaches {e}. Similarly the p-torsion filtration takes the form {e} {e} ...; it never
reaches G. Thus a finite filtration is not merely one with a finite number of distinct terms,
but one that begins at the trivial subgroup and ends at the whole group (or vice-versa).

Usually the point of filtering a group (or anything else) is to arrange it in such a way
that the quotient objects Gk /Gk1 (increasing case) or Gk /Gk+1 (decreasing case) have
some simple form that we understand; then the hope is that we can recover, perhaps by an

14
induction argument, information about G itself. This is the meaning of the motto: Dont
fight it, filter it! In our category of groups, as it stands the quotients are in general only
sets, so we make the definitions:

Definition. An increasing filtration is subnormal if Gi is normal in Gi+1 for all i, and normal
if Gi is normal in G for all i. Thus normal implies subnormal but not conversely (for a
minimal counterexample to the converse, look at our old friend A4 ). Subnormal and normal
decreasing filtrations are defined similarly. The classical terminology is subnormal/normal
series.

It will be handy to have a term for the following simple construction: Suppose HG
K is a group extension, and we are given finite filtrations (which we can assume are increas-
ing) of H and K. Then we get a filtration of G by splicing them together:

{e} = H0 H1 ... H 1 K1 ... 1 Kn = G.

We call this the splice of the two filtrations. Note that the splice of subnormal filtrations is
subnormal, and that the list of quotient groups obtained is just the union of the quotient
groups of the original filtrations.
Caution: The splice of normal filtrations need not be normal. Once again, A4 provides a
counterexample. In order for the splice to be a normal filtration, the filtration on H would
have to be invariant under the conjugation action of K on the set of normal subgroups of H.

Now, lets get on to our main examples.

9.2 Solvable groups

Abelian groups are easier to understand than general groups. In the spirit of our filtration
motto, therefore, it is reasonable to make the following definition: A group G is solvable if
it admits a finite subnormal filtration with abelian quotients. (The terminology comes from
Galois theory, where group theory originated.) We call such a filtration a solvable filtration
(or series) for G. Any abelian group is solvable, and if HGK is a group extension
with H, K abelian, then G is solvable. Before giving further examples, we establish some
basic properties of solvable groups.
First of all, there is a functorial decreasing filtration with abelian quotients of any group
G, defined recursively by G0 = G and Gi+1 = [Gi , Gi ]. We call this the commutator filtration.
Note that it is not only a normal filtration, but even a characteristic filtration; i.e. the
subgroups in question are characteristic subgroups (invariant under arbitrary automorphisms
of G). The first thing to note is that it has limited applicability; indeed for many groups it
is a useless filtration: For example, if G is a nonabelian simple group then it is the filtration
G G G.... But if it does yield a finite filtration, i.e. one that reaches {e}, then G is
solvable. We will prove the converse shortly.

Remark. By functorial we mean that any group homomorphism preserves the filtration. To
make this fit precisely into the framework of category theory, we would define a category of
filtered groups, with morphisms the filtration preserving homomorphisms, so that assigning

15
to G its commutator filtration is a functor to this new category. But theres no compelling
reason to do this at the moment.

Proposition 9.1 The following are equivalent:

a) The commutator filtration of G is finite, i.e. G(m) = {e} for some m.
b) G admits a finite normal filtration with abelian quotients.
c) G is solvable.

Proof: Clearly (a) (b) (c). Now suppose G is solvable, and let G = H 0 H 1 ...
H m = {e} be a solvable filtration. Then G/H 1 is abelian, so [G, G] H 1 . Similarly, since
H 1 /H 2 is abelian, we have G(2) [H 1 , H 1 ] H 2 . Continuing in this manner, we find that
G(k) H k for all k. Hence G(m) = {e}, proving that (c) (a).

Proposition 9.2 Solvable groups are closed under taking subgroups, quotients, extensions,
and finite products.

Proof: We first show that if G is solvable, so is any quotient group. Suppose : GH

is a surjective homomorphism. Let G G1 G2 ... be a solvable filtration for G, and
consider the filtration (Gi ) of H. It is a finite subnormal filtration, and the quotients are
abelian because (Gi )/(Gi+1 ) is a quotient of Gi /Gi+1 . Hence H is solvable. The case of
subgroups is equally straightforward, and left to the reader.
Next we show that solvable groups are closed under extensions; that is, if HGK
is a group extension and H, K are solvable, so is G. This is immediate because given
solvable filtrations for H, K we can splice them to get a solvable filtration for G (recall that
subnormality is preserved under splicing, and the quotients remain abelian because they are
in fact identical to the original quotients).
Finally, consider products. The product of two solvable groups G, H is solvable because
of the extension HG HG. Induction on the number of factors then shows that any
finite product of solvable groups is solvable.

Remark: It follows that the solvable groups can be described as the smallest class of groups
that contains the abelian groups and is closed under extensions.

Now, here is one of the most important solvable groups. We could work over a general
commutative ring, but to avoid distractions we will stick to the case of a field F . Recall that
Bn F GLn F is the Borel subgroup of upper triangular matrices.

Proposition 9.3 Bn F is solvable.

Proof: Note that B1 F = F , and that there is a surjective homomorphism : Bn F Bn1 F

given by simply projecting A Bn F onto its upper left (n 1) (n 1) block. By induction
we can assume Bn1 F is solvable, so it suffices to show Ker is solvable. Now Ker is
the right column group consisting of matrices that equal the identity in the first n 1
columns, which we will quaintly denote RCn F . It fits into an extension RCnu RCn F ,
where the second map is just projection on the nn coordinate and RCnu is therefore the

16
subgroup with ann = 1 (the superscript u if for unipotent). One easily checks that RCnu
is isomorphic to the additive group F n1 , and in particular is abelian. So RCn F is solvable
and the proof is complete.

See the exercises for a discussion of the commutator series of Bn F .

The finite p-groups are another important family of solvable groups. However, they
satisfy the even stronger property of nilpotence, as we will show in the next section.

There are a number of theorems showing that under certain restrictions on the prime
factors of n, every group of order n is solvable. Here are three such theorems, in increasing
order of difficulty (the first is by far the easiest, and is demoted to the status of proposition):

Proposition 9.4 a) If n = pq with p, q prime, then every group of order n is solvable.

b) If n = pqr with p, q, r distinct primes, then every group of order n is solvable.

Proof: Exercise. Part (a) is trivial from things weve already proved. Part (b) is a little
more interesting.

The next result is known as Burnsides pa q b theorem.

Theorem 9.5 Suppose p, q are primes and |G| = pa q b . Then G is solvable.

The proof is a beautiful application of representation theory, as we will show later.

Corollary 9.6 Every group of order < 60 is solvable.

Proof: 60 = 22 3 5 is the smallest number that is neither the product of three distinct
primes nor of the form pa q b . (Its also a nice exercise to prove the corollary directly, without
Burnsides theorem.)

The next theorem is due to Feit and Thompson in 1963.

Theorem 9.7 Every finite group of odd order is solvable.

The original proof is 200 pages long, and as far as I know has never been simplified.

9.3 Nilpotent groups

An increasing normal filtration {e} = G0 G1 ... of G is central if Gi+1 /Gi C(G/Gi ).
We say that G is nilpotent if admits a finite central filtration, i.e. one that terminates at G.
Note that nilpotent groups are solvable, since any central filtration is a solvable filtration.
On the other hand, S3 is solvable but not nilpotent.

Proposition 9.8 Every finite p-group is nilpotent.

17
Proof: Let G be a finite p-group. Then C(G) is non-trivial. By induction on order we can
assume G/C(G) has a finite central filtration; splicing with the one-step filtration {e}
C(G) yields the result.

In fact any group has a functorial, normal central filtration, sometimes called the as-
cending central series, defined recursively as follows: Let C1 = C(G). Having defined
the normal subgroup Ck , let : GG/Ck be the quotient homomorphism, and set
Ck+1 = 1 C(G/Ck ). It is clear that Ck+1 is normal.
The next proposition is analogous to Proposition 9.1.

Proposition 9.9 The following are equivalent:

a) The ascending central series of G is finite; i.e., ends at G.
b) G is nilpotent.

Proof: Clearly (a) (b). To show that (b) (a), assume given a central filtration Gi with
Gm = G and show inductively that Gi Ci . Then Cm = G, as desired.

Before proceeding further, it will be convenient to generalize the concept nilpotent

group to nilpotent group action. Suppose K, G are groups and K acts on G by group
automorphisms. We say that the action in nilpotent if G has a finite subnormal filtration
{e} G1 ... Gm = G such that (i) each Gi is invariant under the K-action, and (ii) the
induced K-action on each quotient Gi /Gi1 is trivial. We call such a filtration K-nilpotent.

Example. Let G act on itself by conjugation. This action is nilpotent if and only if G is
nilpotent.

Example. Consider a field F and take K to be the group of upper triangular unipotent
matrices Un F . For G we take the additive group F n , with its standard left Un F action.
Filter F n by the F i s (where as always, our default inclusion F i F n is in the first i
coordinates). This is automatically a subnormal (indeed normal) filtration, since F n is
abelian, and satisfies the nilpotent action conditions (i)-(ii) by definition of Un F .
Note: From the point of view of this example, it would have made more sense to use the
term unipotent action in place of nilpotent action. But such terminology conflicts are
inevitable, and one just has to live with them.

Nilpotent groups are not closed under extensions (think of S3 , for example). Our next
definition compensates for this deficiency: A group extension HGK is nilpotent if the
conjugation action of G on H is nilpotent. Any central extension is nilpotent, for example,
or more generally any extension in which the conjugation action of G on H is trivial. The
extension C3 S3 C2 is not nilpotent.

Proposition 9.10 The class of nilpotent groups is closed under subgroups, quotients, nilpo-
tent extensions and finite products.

Proof: Let G be nilpotent, H a subgroup. Since the action of G on itself by conjugation is

nilpotent, so is the restriction of this action to H, i.e. the action of H on G by conjugation.

18
But H is invariant under the latter action, and we get a finite central filtration of H by
intersecting with a given such filtration for G. The case of quotients is also straightforward,
and left to the reader.
Now suppose HGK is a group extension with K nilpotent and G acting nilpo-
tently on H by conjugation (which implies that H is nilpotent). Choose a G-nilpotent
filtration of H and a nilpotent filtration of K; splicing these yields a nilpotent filtration of
G, as desired. Details are left to the reader.
If H, G are nilpotent, then HG HG is clearly a nilpotent extension, since the
conjugation action of G H on H factors through H. So H G is nilpotent. It then
follows by induction on the number of factors that any finite product of nilpotent groups is
nilpotent.

Next is one of the most important examples. Let F be a field, and recall that the
unipotent group Un F is the group of upper triangular nn matrices with 1s on the diagonal.
Since it is a subgroup of the solvable group Bn F , it is solvable. But more is true:

Proposition 9.11 Un F is nilpotent.

Proof: We follow closely the proof already given for solvability of Bn F . As in that case,
there is a group extension

RCnu F Un F Un1 F,
where the second map is projection on the upper left (n 1) (n 1) block and RCnu F is
again the unipotent right column group; for example RC3 F consists of matrices

1 0 a
0 1 b

0 0 1

By induction we can assume Un1 F is solvable, so it suffices to show that the action of Un F
on RCnu F by conjugation is nilpotent. Since RCnu F is abelian, the action of Un F factors
through Un1 F , so what we need to show is that the conjugation action of Un1 F on RCnu F
is nilpotent. But under the evident isomorphism RCnu F = F n1 (where the entries of the
right column are ordered from top to bottom), this action corresponds to the standard linear
action of Un1 F on F n1 . We saw earlier that this latter action is nilpotent, so the proof is
complete.

Here are two more interesting facts about nilpotent groups. Both are false for solvable
groups; the reader can easily supply examples.

Proposition 9.12 Let G be a nilpotent group, H a normal subgroup. Then H C(G) 6= {e}.

Proof: Since G acts nilpotently on itself by conjugation, and H is invariant, it acts nilpotently
on H. In particular H has a non-trivial subgroup H1 on which G acts trivially, so H1
H C(G).

19
Proposition 9.13 Let G be a nilpotent group, H a proper subgroup of G. Then H is a
proper subgroup of its normalizer NG H.

Proof: If C := C(G) is not contained in H, then H 6= CH NG H. If C H, then by

induction we can assume H/C 6= NG/C H/C; it follows at once that H 6= NG H.

Now we can neatly characterize finite nilpotent groups.

Theorem 9.14 A finite group G is nilpotent if and only if G is a product of p-groups (where
p ranges over the prime divisors of |G|).

Proof: We have shown that any p-group is nilpotent. Since any finite product of nilpotent
groups is nilpotent, this proves the if.
Conversely, suppose G is nilpotent, let p divide |G|, and let P be a p-Sylow subgroup.
Then NG (NG P ) = NG P (see the section on the Sylow theorems). By Proposition 9.13
this is impossible unless NG P = G, so P is normal. So if p1 , ..., pm are the prime divisors
of G, with corresponding unique pi -Sylow subgroups Pi , the natural multiplication map
: P1 ... Pm G is an injective group homomorphism. Comparing orders, we see that
it must be an isomorphism.

The descending central series of a group G is defined recursively by C 0 = G and C k+1 =

[G, C k ]. Thus C 1 = [G, G], so this starts off the same as the commutator series. But at
the next step we take [G, [G, G]] instead of [[G, G], [G, G]]. By construction the descending
central series is a characteristic (hence normal) filtration, whose quotients are not only
abelian but satisfy C k /C k+1 central in G/C k+1 . Hence if the descending central series is
finite, i.e. ends at {e}, reversing the order yields a finite central increasing filtration of G.
This proves one direction of the following proposition; the converse is left to the reader.

Proposition 9.15 G is nilpotent if and only if the descending central series is finite, i.e.
C n = {e} for some n.

We conclude by mentioning a class of finite groups lying between the nilpotent and
solvable groups. If G is a finite group, G is supersolvable if it has a finite normal filtration
with cyclic quotients.

Proposition 9.16 For finite groups G,

nilpotent supersolvable solvable.

The second implication is immediate, while the first is left as an exercise. Note that
S3 is supersolvable but not nilpotent, while A4 is solvable but not supersolvable. One can
also check that any solvable finite group admits a subnormal filtration with cyclic quotients;
hence the insistence on normal filtrations in the definition of supersolvable groups is key.
Supersolvable groups have nice representation-theoretic properties, as we will see in [Serre],
8.5.

20
10 Simple groups and perfect groups
10.1 Simple groups
A group G is simple if it has no non-trivial proper normal subgroups. If G is abelian, then G
is simple if and only if it is cyclic of prime order (an easy exercise). Non-abelian finite simple
groups were not completely classified until 2004; the proofs occupy thousands of pages. The
formidable Feit-Thompson theorem implies that every non-abelian finite simple group has
even order, but this is just the first little step!
Note that since every group of order < 60 is solvable, there are no non-abelian simple
groups of order < 60 (indeed these two statements are equivalent).

Proposition 10.1 The alternating group An is simple for n 5.

For a short and interesting proof, see Artins undergraduate algebra text (proofs can be
found also in [Hungerford] and [Dummit-Foote]). Another infinite family of simple groups
is given by the projective special linear groups P SLn F , F a field. These are defined by
P SLn F = SLn F/C, where C := C(SLn F ) is the center. Since C(GLn F ) consists of the
scalar matrices, it is easy to see that C(SLn F ) is also just the scalar matrices with deter-
minant 1, and hence is a finite cyclic group of order dividing n. It turns out that with just
two exceptions, P SLn F is simple for all n, F . Well prove this in a later section for n = 2,
and sketch the proof for general n. By taking F to be a finite field we get an infinite family
of finite simple groups. This is as far as well go.

10.2 The Jordan-Holder theorem

Let G be a group. A Jordan-Holder filtration of G is a finite increasing subnormal filtration
with simple quotients. Such a filtration need not exist; for example, an abelian group admits
a Jordan-Holder filtration if and only if it is finite (easy exercise). On the other hand, clearly
any finite group admits a Jordan-Holder filtration.
Here is the Jordan-Holder theorem:

Theorem 10.2 If G admits a Jordan-Holder filtration, then the list of simple groups occur-
ing as the quotients is independent of the choice of filtration, up to ordering.

We will rarelyif everuse this theorem, so we wont prove it here. For a proof and
further discussion see e.g. [Hungerford]. (We will, however, prove and use the analogous
theorem for modules over a ring.)

10.3 Perfect groups

A group G is perfect if G = [G, G]. Thus G is perfect if and only if Gab = {e} if and only if
every homomorphism from G to an abelian group is trivial. Some easy facts about perfect
groups:

A group is both perfect and solvable if and only if it is trivial.

21
 Any simple non-abelian group is perfect.
Perfect groups are closed under quotients, extensions, and arbitrary products (but not
under subgroups).
Note that a perfect group need not be simple. For example, any product of non-abelian
simple groups such as A5 A5 is perfect, but certainly not simple.
We will see below that with two exceptions, SLn F is perfect (but typically has non-trivial
center, hence need not be simple). Indeed for us this is the main reason to bother introducing
the concept of a perfect group; we will use it as a stepping stone toward proving P SLn F is
(almost always) simple.

11 On SLnF and P SLnF

Recall (see Preliminary Remarks) that SLn F is the group of n n matrices over the field
F with determinant 1. Note that GLn F fits into a split extension

SLn F GLn F F .
Many basic subgroups of SLn F are defined by simply intersecting with the corresponding
subgroups of GLn F . For example, define the Borel subgroup of SLn F by SBn F = Bn F
SLn F , the upper triangular matrices with determinant 1. Although one rapidly grows tired
of putting the S in SBn F and its cousins, well do so for a little while. Thus SDn F
denotes diagonal matrices of determinant 1, SUn F = Un F (since the unipotent subgroup
already consists of determinant 1 matrices), and SNn F = Nn F SLn F . Here one can check
that SNn F = NSLn F SDn F provided, as usual, that F 6= F2 .
The one significant difference to watch out for concerns the Weyl group. One might think
that we should just take permutation matrices of determinant 1, but this turns out to be
the wrong thing to do. The way to think of it is as follows: In GLn F we have the extension

Dn F Nn F Wn F,
where Wn F = Sn and a splitting of the extension is already given by using permutation
matrices. If F 6= F2 there is an analogous extension

SDn F SNn F SWn F,

where SWn F is by definition SNn F/SDn F and is again isomorphic to the symmetric group
Sn . But unless char F = 2, the extension doesnt split: the problem is that transpositions
(as permutation matrices) have determinant 1 and there is no way to lift them to elements
of order 2 in SNn F . Its already a problem when n = 2, as you can check. When char F 6= 2,
they do however lift to elements of order 4; e.g. for n = 2 use
!
0 1
1 0
or its inverse. The upshot of this discussion is that we just have to live with the fact that
the extension SDn F SNn F SWn F = Sn usually doesnt split. In practice, fortunately,

22
this does not cause any difficulties. To illustrate, lets first clarify our definition of SWn F : If
char F = 2, it is just the permutation matrices as before, i.e. is equal to Wn F . If char F 6= 2,
SWn F = SNn F/SDn F . In all cases SWn F = Wn F = Sn . Now consider the SLn version of
the Bruhat decomposition.

Theorem 11.1 a
SLn F = SBn F wSBn F,
wWn F

where in the case char F 6= 2, w is any preimage of w in SNn F (in the characteristic 2 case
it is just a permutation matrix as usual).

Proof/Discussion: First of all, note that the expression on the right is well-defined. Any
two choices of w differ by an element of SDn F SBn F , so the double coset SBn F wSBn F
is independent of the choice. The theorem then follows easily from its GLn analogue; no
new fussing about with row reduction is needed. Certainly the indicated double cosets are
distinct and hence disjoint, by comparing with the GLn result. And if g SLn F , we can
write g = b1 wb2 with b1 , b2 Bn F by the GLn result, with det b2 = (det b1 )1 . We can write
b1 = b01 d with b01 SBn F and d Dn F . Then since w normalizes Dn F , we have

g = b01 w(d0 b2 ),
where det d0 = det d = det b1 and hence d0 b2 SBn F .

Carrying around the decorations S, n, F gets tiresome, as does the w notation. So as

long as we have firmly declared that our context is SLn F , we prefer to abbreviate the above
Bruhat decomposition in a slightly abusive but much more pleasant notation as
a
SLn F = BwB.
wW

Thus it is understood that B means SBn F , and so on.

11.1 On SL2 F
The goal of this section is to study the structure of SL2 F in detail, and in particular to
prove:

Theorem 11.2 If F is a field, then P SL2 F is a simple group except when F = F2 or

F = F3 .

In the exercises you show that P SL2 F2 = S3 and P SL2 F3

= A4 , so these two cases are
not simple. Taking F = Fq to be a finite field with q > 3, the theorem gives us a large,
potentially new class of simple groups. I say potentially new because it might happen
that some of these groups are isomorphic to alternating groups. Indeed |P SL2 F5 | = 60, and
hence P SL2 F5 = A5 (see the exercise on simple groups of order 60). It turns out that this
is the only such coincidence, however, a fact that one can check by hand for small q by just

23
checking the orders are not of the form n!/2. For example, P SL2 F7 is simple of order 168
and so is not an alternating group.
The group SL2 F is important for much more general reasons, as it is the most basic
example of a semi-simple algebraic group, or in the case F = R, C of a semi-simple Lie
group. It plays a central role in representation theory and Lie theory, as well as in certain
parts of algebraic topology, algebraic geometry, and combinatorics.

11.1.1 Some handy formulas in SL2

We will need some elementary formulas in SL2 F . They are all easy to check directly. First
some notation: If t F we set x(t) =
!
1 t
0 1

and y(t) =
!
1 0
t 1

If t F we set w(t) =
!
0 t
1
t 0

and h(t) =
!
t 0
0 t1

These satisfy:

1. w(1)x(t)w(1)1 = y(t)

2. h(t) = w(t)w(1)1

3. w(t) = x(t)y(t1 )x(t)

4. [h(s), x(t)] = x((s2 1)t)

5. h(s)x(t)h(s)1 = x(s2 t)

6. x(t)h(s)x(t)1 = h(s1 )x((s2 1)t)

Note that any of the last three formulas easily determines the other two, and that there
are analogues with x(t) replaced by y(t).
Let w = w(1).

24
11.1.2 Properties of SL2 F
Note that the center C := C(SL2 F ) is trivial if char F = 2 and is C2 (i.e. Id) otherwise.
`
The Bruhat decomposition SL2 F = B BwB immediately implies:

Proposition 11.3 B is a maximal proper subgroup of SL2 F .

Proposition 11.4 SL2 F is generated by the subgroups U, U (together), i.e. by the elements
x(t), y(t), t F .

Proof: Let K denote the subgroup generated by U, U . By the Bruhat decomposition we

know that SL2 F is generated by U , D, and w; that is, by the elements x(t), h(t), w(1). Since
w(1) K by Formula (3), and hence h(t) K by Formula (2), we are done.

Proposition 11.5 SL2 F is perfect if F 6= F2 , F3 .

Proof: By the preceeding proposition, it suffices to show that x(a), y(a) are commutators
for all a F . By Formula (1), we need only consider x(a). By Formula (4) we see that x(a)
is a commutator provided there is an s F with s2 6= 1, and this latter statement is true
if and only if |F | > 3.

The two exceptional cases are not perfect (see the exercises).

11.1.3 Simplicity of P SL2

We restate the theorem:

Theorem 11.6 If F is a field, then P SL2 F is a simple group except when F = F2 or

F = F3 .

Proof: Assume |F | > 3. We abbreviate G := SL2 F . It is equivalent to show that if H is a

normal subgroup of G, then either H C (the center) or H = G. Since H is normal, HB
is a subgroup of G, and as it contains B, by maximality of B we have either HB = B or
HB = G.
Suppose HB = B. Then H B, and as H is normal and wBw1 = B , we have
H B B = D. But by Formula (6), any subgroup of D that is normal in G must consist
of elements h(s) with s2 = 1, i.e. must lie in C.
Now suppose HB = G. Then G/H = B/(B H). Since perfect groups and solvable
groups are preserved under quotients, we conclude using Proposition 11.5 (and the assump-
tion |F | > 3) that G/H is both perfect and solvable, hence trivial. So H = G, as desired.

25
11.2 Simplicity of P SLn F : a sketch
In this section we sketch how the results on SL2 extend to SLn . In particular we sketch a
proof that P SLn F is simple for all n 3 and all F .
We use the following notation: For each i, 1 i < n, we let Gi GLn F denote the
block diagonal subgroup consisting of just one 2 2 SL2 F block in the i, i + 1 position. For
example, if n = 4 then G2 is the subgroup

1 0 0 0


0 a b 0

0 c d 0


0 0 0 1

with ad bc = 1. In each such block we have copies of the elements x(t), y(t), h(t), w(t)
defined above for SL2 F , which we denote xi (t), yi (t), etc. Similarly we have subgroups
Ui , Bi , Di , Ni etc. of Gi corresponding to U, B, D, N in SL2 F . Caution: This conflicts with
our earlier notation,in which for example Bi denoted the upper triangular matrices in GLi F .
In the displayed matrix above, B2 is the subgroup defined by c = 0.

Proposition 11.7 SLn F is generated by the subgroups Gi , and hence is generated by the
elements xi (t), yi (t), t F .

Proof: The first statement follows by the usual row/column reduction (or Bruhat decomposi-
tion) argument, since all the elementary row/column operations can be realized by repeated
left/right multiplication by elements of the Gi s. The second statement then follows by what
we proved for SL2 .

Proposition 11.8 SLn F is perfect except when n = 2, F = F2 , F3 .

Proof: The case n = 2 was proved earlier. If F 6= F2 , F3 , the general case follows immediately
from the fact that the Gi s are perfect together with the previous proposition. However, one
can give a uniform proof for all F and n 3 as follows: It suffices to show each xi (t) is a
commutator. These are all conjugate in SLn F (via a permutation of coordinates), so we can
assume i = 1, in which case it is enough to prove the result for n = 3. But x1 (t) is conjugate
to x13 (t), i.e.

1 0 t
0 1 0

0 0 1

Since [x1 (a), x2 (b)] = x13 (ab), x13 (t) is a commutator for all t and were done.

Now the other key ingredient in the P SL2 simplicity proof was that the Borel subgroup
B P SL2 F is a maximal subgroup. This is clearly not true for higher n, as there are
various block-triangular groups that contain B. Explicitly if a = (a1 , ..., ar ) is an ordered
partition of n, there is an associated parabolic subgroup Pa consisting of block triangular

26
matrices whose diagonal blocks have size a1 , ..., ar . For example if n = 6 and a = (3, 1, 2),
then Pa consists of matrices

a b c

d e f

g h i




0 0 0 j

0 0 0 0 k l


0 0 0 0 p q

where the entries are arbitrary and the three diagonal blocks have the product of their
determinants equal to 1. Note that B itself corresponds to the partition (1, 1, ..., 1), while
SLn F corresponds to (n). Note also that Pa is generated by B together with the wi s it
contains. The total number of parabolic subgroups is thus 2n1 . The surprising fact is:

Proposition 11.9 Suppose H is a subgroup of SLn F containing B. Then H = Pa for some

a.

In particular there are only finitely many subgroups containing B. Note that when n = 2
the proposition says that B is a maximal proper subgroup.

Theorem 11.10 P SLn F is simple unless n = 2 and F = F2 , F3 .

Proof: The case n = 2 was proved earlier, so assume n 3 and let H SLn F be a normal
subgroup. Then HB is a subgroup containing B, so HB = Pa for some partition a. If
Pa = B then H B, and as in the case n = 2 we find that H D, from which it follows
that in fact H C, C being the center (i.e. scalar matrices of determinant 1). The reader
can fill in the details of this step. If Pa = SLn F , then as in the case n = 2 we find that
SLn F/H is both perfect and solvable, hence trivial, so H = SLn F and were done.
The new step that remains is to show that the case Pa 6= B, SLn F cant occur. We
sketch briefly the ideas involved. Suppose Pa 6= B, SLn F . Then Pa contains some but
not all of the wi s. Hence there must exist i, i + 1 such that wi Pa and wi+1 / Pa .
Since Bwi B Pa = HB, we have (Bwi B) H 6= . Since H is normal, it follows that
1
(wi+1 Bwi Bwi+1 ) H 6= . From this one can show that wi+1 wi wi+1 HB = Pa . But this is
false, as one can check from the definition of Pa . So we have a contradiction and the proof
is complete.

Taking F to be a finite field, this yields a large family of finite simple groups. With a
very small number of exceptions, they are distinct from each other and from the alternating
groups (up to isomorphism). In fact if memory serves, the only (?) exception besides the
coincidence P SL2 F5
= A5 already mentioned is P SL2 F7 = GL3 F2 . Note these last two
groups have order 168. To get the isomorphism one can show that in fact there is only one
simple group of order 168 up to isomorphism; for an elaborate proof of this see [Dummit-
Foote].

27
12 Exercises
Note: Selected exercises will be assigned. Remember too that your mission is not merely
to find any old proof; always strive for a simple, elegant argument, and of course make full
use of the machinery that we develop. Needless to say, a simple, elegant argument may or
may not come to mind, or even be possible, but it should always be your goal.
Notice. The Feit-Thompson theorem is off-limits unless explicitly allowed!

A. G-actions.

A1. By Cayleys theorem (which is trivial, from a modern perspective), every finite group
is isomorphic to a subgroup of Sn for some n, where the n provided by the proof is n = |G|.
On the other hand, G may well be isomorphic to a subgroup of Sm for some much smaller
m (think of Sm itself, for example). Show, however, that for G = Q8 (the quaternion group
of order 8), n = 8 is the minimal n for which the conclusion of Cayleys theorem holds.

A2. Prove by first finding a suitable set on which the group in question acts:
a) GL2 F2 is isomorphic to S3 , and Af f2 F2 is isomorphic to S4 .
b) P SL2 F3 is isomorphic to A4 .
c) If G is a simple group of order 60, then G is isomorphic to A5 .

B. Wreath products.

Let H be a group, G Sn a subgroup. The wreath product H G is the semi-direct

R

product
R
H n o G, where G acts on the left of H n by permuting the factors. Thus the elements
of H G have the form (h1 , ..., hn )g, with multiplication determined by the formula

g (h1 , ..., hn ) = (hg1 1 , ..., hg1 n )g.

R
Note that the notation is defective, since H G depends not just on G but on the particular
n and the way G is embedded in Sn . But this should cause no confusion in context. In
particular the notation H Sn always means H n o Sn unless otherwise specified.
R

B1. Partition [mn] into n blocks (subsets) of equal size m, compatibly with the standard
order on [mn]. Let Smn denote the subgroup of all block-preserving permutations; by
this we mean that is allowed to permute the blocks as well as the elements within a
particular block. Then
R
= Sm Sn .

B2. Suppose G Sm and K Sn . Regard Sm Sn as a subgroup of Smn as in 2.1. Then

R

there is an isomorphism
Z Z Z Z
(H G) K
=H (G K).
R R R
Consequently expressions such as H G1 G2 ... Gr are unambiguous (assuming Gi is
given
Rr
as a subgroup of Sni ). In particular we can define the r-fold iterated wreath product
H of a subgroup H Sn . It is a subgroup of Snr .

28
 B3. Wreath products at Wimbledon. The Wimbledon tennis tournament begins with a
draw of 128 players. Player no. 1 plays no. 2, no. 3 plays no. 4, the winner of 1-2 plays
the winner of 3-4, and so on. Some care is needed in determining the draw; for instance the
two (theoretically) best players should be placed in opposite halves, say one at position 1
and the other at Rposition 128, so that if they meet at all it will be in the finals. The iterated
wreath product 7 C2 S128 can be thought of as the group of permutations of the draw
leaving it essentially unchanged, and the number of essentially distinct possible draws
is (128!)/2127 . Explain.

B4. Sylow subgroups of symmetric groups. Let n = a0 + a1 p + ... + am pm be the p-adic

expansion of n (so 0 ai < p). Then Sn has a p-Sylow subgroup P with
m Z i
P Cp )ai .
Y
= (
i=1
Ri
Here Cp is the i-th iterated wreath product of Cp , and for each i we are taking the product
of ai copies of it, 1 i m.

C. Sylow subgroups of general linear groups of finite fields. In this exercise p is a prime and
q = pd for some d, while Fq denotes a field with q elements. For any prime `, ` n denotes
the exponent of ` in the prime factorization of n.
n Q
C1. Answer/show: a) |GLn Fq | = q ( 2 ) ni=1 (q i 1)
b) The unipotent subgroup U is a p-Sylow subgroup.
c) How many p-Sylow subgroups are there?
d) What is the order of SLn Fq ? Order of P SLn Fq ?
Note: Recall that F
p is a cyclic group. This is true for any finite field, so Fq is cyclic of
order q 1. We havent proved the general case yet, but you can assume it if necessary in
part (d).

Note: Some of the remaining C problems may require a little knowledge of finite fields
beyond what has been discussed in class, and therefore may be postponed.

C2. Now let ` be a prime 6= p. Assume that ` divides q1, and that if ` = 2 then 4|(q1).
Then DW = NGLn Fq D contains an `-Sylow subgroup
R
of GLn Fq . Hence if ` (q1) = a, GLn Fq
has an `-Sylow subgroup isomorphic to C`a L, where L is an `-Sylow subgroup of Sn (cf.
3.1).
Remarks: (i). If ` is odd and ` doesnt divide q 1, the `-Sylow subgroups are of a
similar nature but the details are more complicated. To pursue this point, think about the
extension of Fq obtained by adjoining an `-th root of unity.
(ii). Why the restriction when ` = 2? One of my favorite mottos is the doubly nonsensical
4 is an odd prime. The key, and elementary, number-theoretic fact behind this motto is
the following: Suppose ` (x 1) = a 1. Then ` (xn 1) = a + ` n provided that either `
is odd or a 2. When ` = 2 and a = 1 this fails, e.g. for x = 3, n = 2.

C3. Let a = 2 (q 2 1).

29
 a) If q = 3 mod 4, the 2-Sylow subgroups of GL2 Fq are isomorphic to the semi-dihedral
group SDa+1 of order 2a+1 , defined for a 3 as follows: Let C2a , C2 have generators x, y
a
respectively, and let C2 act on C2a by y x = x2 1 . Then SDa+1 = C2a o C2 .
Suggestion: Identify F2q with Fq2 and consider the group of units of Fq2 together with the
Frobenius a 7 aq .
R
b) If q = 3 mod 4 and n = 2m, then GL2 Fq Sm (embedded in GLn Fq in the evident way)
R
contains a 2-Sylow subgroup P . Hence P = SDa+1 Q, where Q is a 2-Sylow subgroup of
Sm . What happens for n odd?

c) If q is any odd prime power, the 2-Sylow subgroups of SL2 Fq are isomorphic to the
generalized quaternion group Qa of order 2a . (Qa is defined as follows for a 3: Write the
quaternions as H = C Cj. Then Qa is the subgroup of H+ generated by j and the 2a1 -st
roots of unity in C.)

D. Balanced products and induced G-sets.

D1. Let F be a field, and let Xn Mn F denote the subset consisting of matrices with n
distinct eigenvalues, all of which lie in F . In this problem you will give two alternate ways of
thinking about Xn . Let Zn = Dn Xn , which we identify with the subset of (F )n consisting
of n-tuples with no repeated entries.
a) Let Yn denote the set of ordered n-tuples (L1 , ..., Ln ) of lines in F n such that Li = F n .
P

Then there is a natural bijection Zn Sn Yn = Xn . (Here natural is meant informally,

i.e. the definition of the bijection should flow naturally out of the given data.)

b) Note that GLn F acts on Xn by conjugation, and Zn is invariant under the restriction
of this action to Nn F . Use the recognition principle for induced G-sets to show that the
canonical map GLn F Nn F Zn Xn is a bijection.

Remark: For fans of topology, I note that for F = R, C, the set Xn is a subspace of Mn F
(with its usual topology). The other two spaces in (a), (b) have quotient topologies making
the above bijections homeomorphisms.

D2. Let H be a subgroup of G and let X be a G-set. Then there is a natural isomorphism
of G-sets G H X = (G/H) X, where the target has the product G-action. In particular,
the induced action g1 (g, x) = (g1 g, x) on G X is isomorphic to to the product action.

E. Maximal and minimal subgroups of nilpotent and solvable groups.

Maximal subgroup means maximal proper subgroup, while minimal subgroup
means minimal non-trivial subgroup. Here are four interesting facts to prove; G is al-
ways a finite group.

E1. Suppose G is solvable. Then

a) Every minimal normal subgroup is an elementary abelian p-group. (A finite abelian
p-group A is elementary if every element of A has order p, or equivalently A is a product of
of Cp s.)
b) Every maximal subgroup has prime power index.

30
 E2. Suppose G is nilpotent. Then:
a) Every minimal normal subgroup is central of prime order.
b) Every maximal subgroup is normal and has prime index.

E3. The converse of 2b holds in the form: If G is a finite group such every maximal
subgroup is normal, then G is nilpotent. (Suggestion: Show that every p-Sylow subgroup is
normal.)

F. Properties of unipotent and Borel groups.

1. Un F is generated by the subgroups Ui,i+1 for 1 i < n.

2. C(Un F ) = U1,n .
3. [Un F, Un F ] is the subgroup of elements whose (i, i + 1) entries are zero, 1 i < n.
4. Determine the ascending central, descending central and commutator series for U4 F .
(Or if feeling ambitious, do it for general n.)

Structure of Bn F .
1. If F 6= F2 , [Bn F, Bn F ] = Un F . Hence for all fields the commutator series of Bn F is
determined by that of Un F .
2. Determine the upper and lower central series of Bn F for F 6= F2 . (They wont get
far!)

M. Miscellaneous problems. These are mostly problems already suggested in the body
of the notes.

M1. Show that if either n = pq with p, q prime, or n = pqr with p, q, r distinct primes,
then every group of order n is solvable.

M2. Without using Burnsides pa q b theorem, show that every group of order < 60 is
solvable.

M3. Show that every finite nilpotent group is supersolvable.

M4. Show that an abelian group admits a Jordan-Holder filtration if and only if it is
finite.

M5. Let P be a p-Sylow subgroup of Sp (p prime). Show that NSp P

= Af f1 Fp .

31