N I+1 I I I I I 1 1 I 1 1 I I+1 I+1 N N I I I I 1 1 I 1 1 I 1 1 I I+1 N N N
N I+1 I I I I I 1 1 I 1 1 I I+1 I+1 N N I I I I 1 1 I 1 1 I 1 1 I I+1 N N N
Solutions 5
7. Prove that subgroups and quotient groups of nilpotent groups are nilpotent (your proof should work for
infinite groups). Give an example of a group G which possesses a normal subgroup H such that H and G/H
are nilpotent but G is not nilpotent.
Solution: Let G be nilpotent, so that for some n > 0 we have Gn = 1 where G0 = G and Gi+1 = [Gi , G].
Let H G be a subgroup of G. Then we claim that H i Gi for all i. Indeed, H G by assumption and
we may assume inductively that H i Gi . Then since
{xyx1 y 1 : x H i , y H} {xyx1 y 1 : x Gi , y G},
it follows that H i+1 Gi+1 so that if Gn = 1 then also H n = 1, i.e. H is nilpotent. Now let : G K be
any surjective group homomorphism. We assert that (Gi ) = K i as sets. Indeed, suppose inductively that
(Gi ) = K i . Observe that
{xyx1 y 1 : x K i , y K} = {xyx1 y 1 : x(Gi ), y (G)}
= {(hgh1 g 1 ) : h Gi , g G}
= (Gi+1 )
since is a homomorphism. Hence if Gn = 1 then K n = (Gn ) = 1. Therefore, any homomorphic image of
G is nilpotent. If H C G, we may take to be the natural map G G/H, so that G/H is nilpotent.
Let G = S3 and H = h(123)i ' Z/3. Then clearly H C G and G/H ' Z/2. Since H, G/H are abelian,
they are both nilpotent (the commutator subgroup of an abelian group is trivial). However, G has trivial
center so that Zn (G) = 1 for all n whence G is not nilpotent. Many groups G with this property may be
constructed via the semidirect product. For example, let K = Z/(pn ) for any odd prime p and n 1. Let
G = Aut(K) n K with : Aut(K) Aut(K) an isomorphism. We see that K, G/K are nilpotent (since
they are both abelian) but since is an isomorphism, the center of G is trivial (indeed, if for any g 6= 1
we can find some h Aut(K) with h(g) 6= g, i.e. hgh1 6= g) so that Zn (G) = 1 for all n, hence G is not
nilpotent.
12. Find the upper and lower central series for A4 and S4 .
Solution: Let G = S4 and H = A4 . We showed on HW 3, 5.4, #4 that G1 = [G, G] = H and H 1 =
[H, H] = V4 , the Klein four group. Now [H, H] [H, G] = G2 C G1 . But we also showed on HW 3 that the
only normal subgroups of H containing V4 are V4 and H. Observe, however, that (23)(124)(23)(142) = (234),
so that G2 contains a 3-cycle, and hence properly contains V4 . It follows that G2 = H and hence that Gn = H
for all n 1. Now observe that (123)(12)(34)(132)(12)(34) = (13)(24) and (132)(13)(24)(123)(13)(24) =
(12)(34). Since H 2 = [A4 , V4 ] is a subgroup of V4 containing two 2 2-cycles, it must be all of V4 , so that
H n = V4 for all n 1.
Now the center of S4 is trivial since all elements of a given cycle type are conjugate (so if , have any
given cycle type then there exists g S4 with gg 1 = ; since there are at least two distinct elements of
any given cycle type, cannot be in Z(S4 )). It follows that Zi (G) = {1} for all i 1. Similarly, Z(H) C H
and so Z(H) = V4 or {1} since H is not abelian. But we have seen that [H, V4 ] = V4 so that Z(H) 6= V4 ;
hence Z(H) = {1} whence Zi (H) = {1} for all i 1.
13. Find the upper and lower central series for Sn and An for n 5.
Solution: Recall that An is simple for n 5 and that we showed on HW 3, 5.4, #5 that [Sn , Sn ] = An ,
and since [An , An ] C An is nontrivial (An is not abelian) we have [An , An ] = An . Since
An = [Sn , Sn ] [Sn , An ] [An , An ] = An ,
1
2
we have equality throughout and the lower central series for Sn , An are
Skn = An
Akn = An
for all k 1 and n 5.
Certainly Z(Sn ) An = Z(An ) C An . It follows that Z(An ) = {1} and since the image of Z(Sn )
in Sn /[Sn , Sn ] is trivial, we see that Z(Sn ) An (since An is the kernel of the natural map Sn
Sn /[Sn , Sn ]). Therefore, Z(Sn ) = {1}. Thus, the upper central series for Sn and An are
Zk (Sn ) = {1}
Zk (An ) = {1}
for all k 1.
1. Compute the upper and lower central series for G = U3 (R), the group of strictly upper triangular
invertible matrices in GL3 (R).
Solution: Let
1 a b 1 x y
g = 0 1 c h = 0 1 z .
0 0 1 0 0 1
Then straightforward computation gives
1 0 za xc
ghg 1 h1 = 0 1 0
0 0 1
so that g Z(G) if and only if for all x, z we have za xc = 0, i.e., a = c = 0. It follows that Z(G) consists
of all matrices of the form
1 0 b
0 1 0 ,
0 0 1
which form an abelian group B isomorphic to R under addition. But it is also evident from these calculations
that [G, G] = B since every element of B is a commutator and B is already a group. Now since G1 = [G, G] =
Z1 (G) = Z(G), we have G2 = [G, G1 ] = [G, Z(G)] = {1} since every commutator is trivial. Finally, since
Z1 (G) = [G, G], we know that G/Z1 (G) is abelian and hence Z2 (G)/Z1 (G) = Z(G/Z1 (G)) = G/Z1 (G) so
that Z2 (G) = G. Thus, the upper and lower central series of G are given, respectively, by
Z0 (G) = 1 Z1 (G) = B Z2 (G) = Z
0 1
G =G G =B G2 = {1}.
The remaining exercises in this assignment develop some basic notions in representation theory. We fix
throughout a group G and a field F . An (linear) representation of G on a vector space V over F is a
homomorphism : G GL(V ). Concretely, were representing elements of G by endomorphisms of a
vector space, but it must be stressed that might not be injective. The case of infinite G (e.g., Lie groups)
and infinite-dimensinal V (e.g., function spaces) are extremely fundamental, but intuition is developed by
first understanding finite G and dim V < . Well often refer to the data of (V, ) as a representation space
of G. The most classical example is G = Sn with its representation on F n through permutations of the
standard basis. Also, when dim V = 1 then a representation of G on V is just a group homomorphism
: G F .
2. We say that a representation space (V, ) is irreducible if V 6= 0 and there do not exist any G-stable
subspaces aside from 0 and V .
(i) Prove that the natural representation of GL(V ) on V is irreducible when 0 < dim V < .
3
Solution: Suppose that we had a nontrivial GL(V ) stable subspace W V . Let w W be any nonzero
vector. Then since GL(V ) acts transitively on the one dimensional subspaces of V , for any nonzero v V
there exists g GL(V ) such that gw = v. It follows that W = V and hence V is irreducible.
(ii) Consider the permutation representation of G = S3 on F 3 . If 3! = 6 6= 0 in F , show that the
hyperplane H : x1 + x2 + x3 = 0 and the line spanned by (1, 1, 1) are complementary G-stable subspaces on
which G acts irreducibly. If 2 = 0 in F , show the same claims. If 3 = 0 in F (e.g., F = F3 ), show that there
is no line in F 3 complementary to H which is G-stable, and that the action of G on H is also not irreducible
(find a G-stable line in H).
Solution: Clearly H is stable under the action of S3 (if x1 +x2 +x3 = 0 then for any S3 we also have
x(1) +x(2) +x(3) = x1 +x2 +x3 = 0). It is similarly obvious that the line spanned by (1, 1, 1, ) is S3 -stable.
Since the line is one-dimensional, it must be irreducible. Suppose we have a S3 stable line in H spanned by
(x, y, z). Then it is not difficult to see that we must have x = y = z. Since our purported line is in H, we
have 3x = 0, and since 3 6= 0 F we have x = 0 whence no such stable line exists and H is irreducible. To
see that H and the line spanned by (1, 1, 1) are complementary, observe that any v = (x, y, z) F 3 may be
written in the form v = (x , y , z ) + (1, 1, 1), where = (x + y + z)/3 and clearly (1, 1, 1) is
in our line and (x , y , z ) H. Here we are crucially using that 3 6= 0 in F . Thus, F 3 splits as
H L where L is spanned by (1, 1, 1). Observe that this argument also works when 2 = 0 F . The case
3 = 0 F is a little different. It follows from what we have said above that the line L spanned by (1, 1, 1)
and H are both S3 -stable (since our argument does not incorporate char F in any way). However, in this
case, L is contained in H since 1 + 1 + 1 = 0. Thus, H is not irreducible. Now suppose the span of (x, y, z)
is another stable line. Then as before, we must have x = y = z so that the only S3 -stable line in F 3 is L.
Therefore, there is no line in F 3 complementary to H.
(iii) Changing the ground field can make a big difference. For example, consider the representation
of Z/3 on R2 in which 1 Z/3 acts by the counterclockwise rotation through an angle of 2/3. Write
down the matrix for this and prove this representation is irreducible. Over C2 , the same matrix defines a
representation of Z/3 but use eigenvalues to explicitly determine two G-stable lines and prove these are the
only two such lines in the representation space.
Solution: From linear algebra we know that the linear transformation of R2 corresponding to rotation
about the origin through radians corresponds to the matrix (in the standard basis)
cos sin
r() =
sin cos
so that we have
1 1 3
(1) = r(2/3) = .
2 3 1
We easily compute the characteristic polynomial c(x) of (1) and find c(x) = x2 + x + 1 has no real roots.
(Observe that c(x) = (x3 1)/(x 1)). If there were any stable lines in R2 , then (1) would have a real
eigenvalue, which is evidently not the case.
Since the c(x) has distinct eigenvalues in C, it has 1-dimensional eigenspaces over C and these are the
only stable lines (as a stable line is acted upon through a scaling action). Since the C2 is 2-diml, there are
two such lines.
Direct computation shows that the lines spanned by
1 1
v1 = v2 =
i i
are stable. (Indeed, v1 , v2 are eigenvectors of (1) with eigenvalues e2i/3 , e2i/3 respectively).
(iv) If (V, ) and (V 0 , 0 ) are irreducible, prove that any G-compatible linear map T : V V 0 is either
zero or an isomorphism (note that ker T and imT are G-stable!).
Solution: Let T : V V 0 be any G-compatible linear map. We claim that ker T and Im T are G-stable.
Indeed, let v ker T . Then since T is G-compatible, we have T ((g)v) = 0 (g)(T v) = 0 (g)(0) = 0 so
4
that (g)v ker T . Similarly, if v 0 = T v then 0 (g)v 0 = 0 (g)T v = T (g)v Im T . Hence, Im T and ker T
are G-stable subspaces of V, V 0 respectively. Therefore, if V is irreducible then ker T = {0} or V and T is
injective or the zero map, while if V 0 is irreducible then Im T = V 0 or {0} so that T is either surjective or
the zero map. It follows that if V, V 0 are both irreducible then T is either an isomorphism or the zero map.
(v) If (V, ) is irreducible, dim V is finite, and F is algebraically closed (so characteristic polynomials
make sense and have roots in F ), show that the only G-compatible linear endomorphisms T : V V (i.e.,
T (g) = (g) T for all g G) are scalar multiplications. Hint: T has some nonzero eigenspace.
Solution: Let W T be any nonzero eigenspace of V with associated eigenvalue F (here is where
we use the fact that F is algebraically closed). Then W is the kernel of the linear map (T ) : V V .
Since W 6= {0}, (T ) is not an isomorphism. Since V is irreducible, we have by part (ii) that T = on
all of V ; that is, T is scalar multiplication.
(ii) In general, we wont be so lucky, so just pick any linear : V W which lifts the identity on W
(make this by using bases to make a complementary subspace away from which were projecting). Now we
average! Using that W is G-stabe and |G| 6= 0 in F , it makes sense to form the F -linear map
1 X
0 = (g) (g 1 ).
|G|
gG
Using the fraction out front and that lifts the identity on the G-stable W , prove that 0 restricts to the
identity on W and is G-equivariant. Deduce that ker 0 provides the desired G-stable complement. This
averaging trick is due to Maschke.
Solution: Let : V W be any linear map with |W = id. Observe that for w W we have
1 X
0 (w) = (g) (g 1 )(w)
|G|
gG
1 X
= (g)(g 1 )w
|G|
gG
1 X
= w
|G|
gG
= w,
5
so that 0 restricts to the identity on W . We now claim that 0 is G-equivariant. Indeed, for any h G we
have
1 X
0 (h) = (g) (g 1 )(h)
|G|
gG
1 X
= (h)(h1 g)(g 1 h)
|G|
gG
= (h) ,
so that 0 is G-equivariant. By part (i), we see that ker 0 provides the desired G-stable complement of W .
P
(iii) Consider the permutation representation of Sn on F n , with W the hyperplane xj = 0 and n! 6= 0
in F . Find an Sn -stable complement to W .
Solution: Clearly the line L in F n spanned by v 0 = (1, 1, 1, . . . , 1) is stable under the action of Sn .
Moreover, since n! 6= 0 in F , for any v = (x1 , x2 , . . . , xn ) F n we let = (x1 + x2 + + xn )/n and define
w = v v 0 . Then w W and v = w + v 0 so that W L = F n is a direct sum decomposition of F n into
Sn -stable subspaces. Since L is one dimensional, it is irreducible.
def
4. A very important representation of a finite group G is its left regular representation. Let V = F [G] =
F eg be a vector space whose basis is indexed by the elements of G, and define reg : G GL(V ) by
reg (g) : eg0 7 egg0 . This F -linearizes the left multiplication action of G on itself as a set. Its a big space
(think of G = Sn )!
(i) If (V, ) is a nonzero finite-dimensional F -linear representation of G and |G| = 6 0 in F , use induction
on dimension and Exercise 3 to show that V is a direct sum of G-stable subspaces Vi on which G acts
irreducibly. Thus, to describe all finite-dimensional representations of G up to isomorphism it is enough
in such cases to describe the irreducible ones (of course, the real art is to actually locate the Vi s explicitly
in interesting situations).
Solution: We induct on dim V . For dim V = 1, the assertion is obvious, so suppose that dim V > 1.
Let W V be a nonzero G-stable subspace of V of minimal dimension. Then certainly W is irreducible.
If W = V then we are done. Otherwise, using Problem 3 (ii), (which applies since |G| 6= 0 in F so that
we can construct the projection 0 ) we construct a G-stable subspace V 0 V with W V 0 = V . Since
W 6= {0}, we must have dim W 1 and hence dim V 0 < dim V so by induction V 0 = V2 . . . Vn is a direct
sum of irreducible G-stable subspaces. Thus, V ' W V2 . . . Vn is a direct sum of irreducible G-stable
subspaces.
(ii) Show that if (V, ) is a nonzero representation of G, then by choosing a nonzero v0 V the natural
map v0 : F [G] V defined by eg 7 (g)(v0 ) is a map of representation spaces (i.e., v0 is linear and
commutes with the G-actions: v0 reg (g) = (g) v0 for all g G). Prove that the image of v0 is nonzero
and G-stable, and conclude that if (V, ) is irreducible then v0 is surjective (in particular, dim V |G| is
finite!). Consequently, all irreducible G-representations are quotients of the left regular representation (over
F ).
Solution: That v0 : F [G] V is linear follows immediately from its definition (we have defined it on a
basis and extended linearly to all of F [G]). Moreover, we have
v0 reg (g)(eg0 ) = v0 (egg0 ) = (gg 0 )(v0 )
6
while
(g) v0 (eg0 ) = (g)((g 0 )(v0 )) = (gg 0 )(v0 ).
Since v0 , (g), and reg (g) are linear, and we have checked that v0 reg (g) = (g) v0 on a basis of
F [G], it follows that this relation holds on all of F [G]. Now our proof of Problem 2 (iv) shows that Im v0
is G-stable. Moreover, Im v0 is nonzero since v0 6= 0 and (g) GL(V ). Therefore, if V is irreducible, v0
must be surjective. It follows that dim V |G| for any irreducible representation V of G. Moreover, since
v0 is surjective and G-equivariant, we have F [G]/ ker v0 ' V , where the isomorphism of vector spaces
preserves the G action on eachthat is, every irreducible representation of G is a quotient the left regular
representation.
(iii) Prove that any finite-dimensional representation space (V, ) of G admits a rising chain of G-stable
subspaces
{0} = V0 V1 Vn = V
such that each Vi a G-stable subspace and Vi /Vi1 (for 1 i n) with its natural G-action is irreducible.
We call this a length n filtration of V by G-stable subspaces. By a Jordan-Holder style of argument, prove
that any two such filtrations with irreducible successive quotients have the same length and give rise to the
same collection of irreducible successive quotient representations (up to isomorphism), perhaps in different
orderings.
Solution: We proceed by induction on dim V . Suppose that for all r < n and W of dimension r there
exists a rising chain of G-stable subspaces with irreducible succesive quotients, and let dim V = n. If V is
irreducible, then we are done, so suppose that V is reducible. Let V1 be any nonzero G-stable irreducible
subspace of minimal dimension. Then dim V /V1 < dim V so we may apply our induction hypothesis to
conclude that there exists a chain of G-stable spaces
{0} = W1 W2 . . . Wm = V /V1
with irreducible successive quotients. Now let Vi be the preimage under the natural map V V /V1 of Wi
(since certainly V1 is the preimage of W1 = {0}, our notation is consistent). Observe that Vi V1 and that
Vi must be G-stable since Vi /V1 , V1 are both G-stable This gives a chain
{0} = V0 V1 . . . Vm = V
of G-stable spaces with successive quotients Vi /Vi1 which maps injectively to Wi /Wi1 via the map V
V /V1 (which is evidently G-compatible). Since Wi /Wi1 is irreducible by assumption, it follows from Problem
2 (iv) that either Vi /Vi1 = {0} or Vi /Vi1 ' Wi /Wi1 is irreducible. In the former case, we may delete the
term Vi1 from the chain.
Notice that we have not used that |G| 6= 0 in F anywhere. Indeed, one might be tempted to use part (i) to
say that V ' ni=1 Wi with each Wi G-stable and irreducible, and then define Vk = ki=1 Wi . This argument
only works given that we have such a direct sum decomposition to begin with, which ultimately depended
on the existence of a G-equivariant projection V Wi . In general, it is false that every finite dimensional
representation V decomposes as a direct sum of irreducible G-stable subspaces Wi . In other language, V is
not in general a semisimple G-module.
We now show that any two such filtrations with irreducible successive quotients have the same length and
give rise to the same collection of irreducible successive quotient representations. Let
(1) {0} = V0 V1 . . . Vr = V
(2) {0} = W0 W1 . . . Ws = V
be any two such filtrations, and suppose that Vi /Vi1 , Vi /Vi1 are irreducible. We will proceed by induction
on min(r, s). If min(r, s) = 1 then V is irreducible, and we must have r = s and both quotients are isomorphic
to V . Therefore, suppose that the result holds for all r, s with min(r, s) < r0 . Now let V admit two filtrations
of the form (1), (2) with r = r0 and s = s0 where we assume without loss of generality that r0 s0 .
Observe that the natural maps Vr1 /(Vr1 Ws1 ) V /Ws1 and Ws1 /(Vr1 Ws1 ) V /Vr1
are injections (if an element of Vr1 dies in the quotient V /Ws1 then it must be in Ws1 and hence
7
Vr1 Ws1 ). Moreover, we may suppose that Vr1 Ws1 is a proper subspace of both Vr1 and Ws1 :
if this were not the case then one of Ws1 , Vr1 would be contained in the other. Suppose, for example,
that Vr1 Ws1 . Then Ws1 /Vr1 is a G-stable subspace of V /Vr1 . Irreducibility of V /Vr1 forces
Ws1 = Vr1 . We then have two filtrations of the common space Vr1 = Ws1 . Now Vr1 has a filtration
of exact length r 1 < r = min(r, s), so by our induction hypothesis (r 1 < r = min(r, s)) we know that
all filtrations of Vr1 = Ws1 (with irreducible successive quotients, as always in this discussion) have exact
length r 1 and give rise to the same collection of irreducible quotient representations. It follows that the
two filtrations of V have the same length and give rise to the same collection of irreducible factors.
Therefore, we may assume that V 0 = Vr1 Ws1 is a proper subspace of Ws1 and Vr1 . Again,
V /Ws1 , V /Vr1 are irreducible and Vr1 /V 0 , Ws1 /V 0 are G-stable, so that the injections Vr1 /V 0
V /Ws1 and Ws1 /V 0 V /Vr1 are isomorphisms (since V 0 is a proper subspace of V , we see that
Vr1 /V 0 6= {0} and similarly for Ws1 /V 0 ). If V 0 = {0} then we see that Vr1 , Ws1 are irreducible
and V /Vr1 ' Ws1 and V /Ws1 ' Vr1 so that r = s = 2 and the pairs of successive quotients {V /V1 , V1 }
and {V /W1 , W1 } agree up to isomorphism (more precisely, W1 ' V /V1 , V1 ' V /W1 ). Thus, suppose
V 0 = Ws1 Vr1 is nonzero.
As before, we use our inductive hypothesis and assume that all filtrations of Vr1 have the same length
and give rise to the same collection of irreducible quotient representations (up to isomorphism). But V 0 is a
nonzero G-stable proper subspace of Vr1 with Vr1 /V 0 ' V /Ws1 irreducible. Thus, picking any filtration
for V 0 (by (iii)) we get a filtration for Vr1 . We conclude that all filtrations of V 0 have exact length r 2
and give rise to the same collection of irreducible quotient representations.
However, V 0 is a proper subspace of Ws1 with Ws1 /V 0 ' V /Vr1 irreducible, so that Ws1 has a
filtration of exact length (r 2) + 1 = r 1, namely the filtration passing through V 0 , so again by induction,
all filtrations of Ws1 have exact length r 1 and give rise to the same collection of irreducible quotient
representations Since we have been given a filtration of Ws1 of exact length s 1, we conclude that
s 1 = r 1 whence r = s so that any two filtrations of V have the same length.
Now let
0
{0} = V00 V10 . . . Vr3 0
Vr2 =V0
be any filtration of V 0 . We have seen that any other filtration has the same length and gives rise to the same
isomorphism classes and multiplicities of irreducible quotient representations. It follows that the filtration
(1) has successive quotient representations
0
R1 = {Vi+1 /Vi0 for 0 i r 3, Vr1 /V 0 , V /Vr1 },
while the filtration (2) has successive quotient representations
0
R2 = {Vi+1 /Vi0 for 0 i r 3, Wr1 /V 0 , V /Wr1 }.
However, we showed earlier that in this situation,
V /Vr1 ' Ws1 /V 0 and V /Wr1 ' Vr1 /V 0 .
It follows that the multi-sets R1 , R2 contain the same list of irreducible quotient representations up to
isomorphism. That is, any two filtrations of V have the same length and give rise to the same isomorphism
classes and multiplicities of irreducible quotient representations. This completes the proof.
(iv) Using (iii), deduce that up to isomorphism there are only finitely many irreducible representations of G
on F -vector spaces of finite dimension. The first real theorems in representation theory provide systematic
ways to explicitly determine these irreducibles, and in real life one certainly wants to realize them in
geometric ways (rather than just as abstract creatures) if at all possible. The effective version of this
finiteness result makes it possible to determine the structure of various molecules and crystals, given enough
knowledge about the symmetry. Ask your friends in physical chemistry about this.
Solution: By part (ii), every representation of G on a finite dimensional vector space V is a quotient of
the left regular representation F [G]. Since dim F [G] = |G| is finite, by part (iii) we know that F [G] admits
8
a filtration
(3) {0} = V0 V1 . . . Vn = F [G]
of G-stable subspaces with irreducible successive quotients, and that any two such filtrations have the same
length and give rise to the same collection of quotient representations. Thus, every irreducible representation
of G on a finite dimensional vector space V occurs as one of the successive quotients of (3) and hence there
are, up to isomorphism, only finitely many finite dimensional irreducible representations of a finite group G.