Solutions to Introduction to Lie Algebras and Representation

Theory by J. E. Humphreys
I. BASIC CONCEPTS
1. Denitions and rst examples
1. The cross product is bilinear and satises x x = 0. By linearity it suces to check the
Jacobian identity on the standard basis: (e
i
e
j
) j
k
+ (e
i
e
j
) j
k
+ (e
i
e
j
) j
k
= 0.
Indeed, if any two indices are equal, say i = j, then the rst term vanishes while the second
is the negative of the third by antisymmetry, and if all indices are distinct, then each term
vanishes.
Since e
i
e
i+1
= e
i+2
(indices (mod 3)), the structure constant a
k
ij
equals 0 if any two indices
are equal, and otherwise equals 1 if j i + 1 (mod 3) and 1 if j i + 2 (mod 3). That is,
a
3
12
= a
1
23
= a
2
31
= 1, a
2
13
= a
3
21
= a
1
32
= 1, and all other structure constants are 0.
2. Again checking the Jacobian identity on the basis vectors, the same argument as above veries
the case when any two are equal, so it remains to check that [[xy]z] + [[yz]x] + [[zx]y] =
0 + 0 + 0 = 0 and [[xz]y] + [[zy]x] + [[yx]z] = 0 + 0 + 0 = 0 (and in fact the rst implies the
second).
3.
ad x =

0 2 0
0 0 1
0 0 0

, ad h =

2 0 0
0 0 0
0 0 2

, ad y =

0 0 0
1 0 0
0 2 0

.
4. Since L has basis x, y with [xy] = x, we have Z(L) = 0 and hence the adjoint representation
(which has kernel Z(L)) gives an isomorphism onto ad L gl(L). Concretely, with respect
to the same basis,
ad x =

0 1
0 0

, ad y =

1 0
0 0

, ad L =

a b
0 0

: a, b F

gl(L),
and we check that ad xad y ad y ad x = ad x.
1
2. Ideals and homomorphisms
1. Let x L and Der L. For any z L, we have
[, ad x](z) = ([x, z]) [x, (z)] = (x)z +x(z) (z)x z(x) x(z) +(z)x = [(x), z],
so [, ad x] = ad((x)) ad L. Thus ad L is an ideal of Der L.
2.
3.
4. Extend a basis x of [LL] to x, y, z of L. Since x [LL] Z(L), we have [xy] = [xz] = 0.
Then 1 = dim[LL] = dim'[yz]`, so [yz] = ax for some a = 0. Replacing x by ax, the new basis
x, y, z of L satises [xy] = [xz] = 0 and [yz] = x, which determines L up to isomorphism.
We verify that [LL] = 'x` has dimension 1 and lies in Z(L).
5. L = [LL] is equivalent to the statement that every element of L is a commutator. A Lie-
algebra homomorphism sends a commutator to a commutator, so a homomorphic image of L
has the same property. Consider in particular L/I for any ideal I of L. If dim(L/I) = 1 or
2, then [L/I, L/I] has a lower dimension and cannot equal L/I. Thus L/I = 0 or 3, so L has
no proper ideal, i.e. L is simple.
For L = sl(2, F), we have dimL = 3 and [xy] = h, [hx] = 2x, [hy] = 2y, so assuming
char F = 2, L = [LL] and hence L is simple.
6.
7.
8.
9.
10.
ad x =

0 2 0
0 0 1
0 0 0

, (ad x)
2
=

0 0 2
0 0 0
0 0 0

, (ad x)
n
= 0, n 3,
ad(y) =

0 0 0
1 0 0
0 2 0

, (ad(y))
2
=

0 0 0
0 0 0
2 0 0

, (ad(y))
n
= 0, n 3,
2
so
exp ad x = 1 + ad x +
(ad x)
2
2
=

1 2 1
0 1 1
0 0 1

exp ad(y) = 1 + ad(y) +

(ad(y))
2
2
=

1 0 0
1 1 0
1 2 1

= exp ad x exp ad(y) exp ad x

=

1 2 1
0 1 1
0 0 1

1 0 0
1 1 0
1 2 1

1 2 1
0 1 1
0 0 1

= antidiag(1, 1, 1).
Hence (x) = y, (y) = x, (h) = h, as desired.
11. Call the map . Then is clearly linear and has inverse y (g
1
(y)g)
t
. Since
([xy]) = g(xy yx)
t
g
1
[(x)(y)] = (gx
t
g
1
)(gy
t
g
1
) (gy
t
g
1
)(gx
t
g
1
) = gx
t
y
t
g
1
gy
t
x
t
g
1
are equal, Aut L.
12.
3. Solvable and nilpotent Lie algebras
1. If I, J are ideals of L, then so is [IJ]. Now induct.
2.
3. Let L = sl(2, F), where char F = 2. Then [xy] = h and [hx] = [hy] = 0, so L
1
= [LL] = Fh
and L
2
= [LL
1
] = 0. Thus sl(2, F) is nilpotent.
4. Since ad : L gl(V ) is a homomorphism with kernel Z(L), it induces L/Z(L)

= ad(L).
Since Z(L) is even abelian, L solvable (resp. nilpotent) ad(L) solvable (resp. nilpotent).
5.
6. If I, J are solvable (resp. nilpotent) ideals of L, then so is I J, so (I + J)/J

= I/(I J)
implies I/(I J) solvable (resp. nilpotent) and I +J solvable (resp. nilpotent).
7. Imitate proof of Theorem 3.3. Directly: K acts on L by ad, so on L/K. By Theorem 3.3,
some x / K is killed by K , i.e. [xK] = 0, i.e. x N
L
(K).
3
8. See proof of Theorem 3.3. [Let K be the maximal proper subalgebra of L (exists assuming
dimL < ). By 7, the ideal N
L
(K) must be L, i.e. [LK] K i.e. K is an ideal of L. If
dimL/K > 1, then the preimage in L of a one-dimensional subalgebra of L/K is a proper
subalgebra of L properly containing K, a contradiction.]
9. Let L be nilpotent. By 8, we can write L = K + Fx for some ideal K of codimension 1.
Applying Theorem 3.3 to the adjoint action of K on L, K kills some non-zero element of L,
so C
L
(K) = 0 [alternatively, C
L
(K) Z(L) = 0]. Since the lower central series eventually
vanishes, we can choose n such that C
L
(K) L
n
, C
L
(K) L
n+1
, and let z C
L
(K) ` L
n+1
.
An element of L = K +Fx can be written uniquely as y +ax for y K and a F.
Dene : L L by (y +ax) = az. Clearly is linear. Since K is an ideal and [xx] = 0, we
have ([y +ax, y

+a

x]) = ((element of )K +0) = 0. Meanwhile, since z C

L
(K), we have
[(y +ax), y

+a

x] +[y +ax, (y

+a

x)] = [az, y

+a

x] +[y +ax, a

z] = aa

[zx] +aa

[xz] = 0.
Thus is a derivation. Suppose = ad w for some w L. Then [w, K] = (K) = 0 implies
w C
L
(K) L
n
, so z = (x) = [w, x] L
n+1
, a contradiction. Hence is an outer
derivation.
10. (Since K is an ideal, ad x(K) = [xK] K, so ad x[
K
makes sense.) Let x, y L. Since L/K
is nilpotent, ad(x +K) : L/K L/K is a nilpotent endomorpism, (ad x)
m
(y) K for some
m. Then since ad x[
K
is nilpotent, (ad x)
n
(ad x)
m
(y) = 0 for some n. Thus ad x is nilpotent
for all x L. By Engels theorem, L is nilpotent.
II. SEMISIMPLE LIE ALGEBRAS
4. Theorems of Lie and Cartan
1. Let L = sl(V ). Clearly Z(L) Rad L. Let B be a maximal solvable subalgebra of L. Then
[B +Z(L), B +Z(L)] = [BB], so B +Z(L) is also a solvable subalgebra, hence Z(L) B by
maximality; thus Z(L) lies in each maximal solvable subalgebra. Fix B. By Lies Theorem,
we can choose a basis of V such that B Lt, where t is the subalgebra of upper-triangular
matrices under this basis. But L t is solvable, so B = L t by maximality. Since the
transpose of B is also a maximal solvable subalgebra, Rad L lies in their intersection L d.
2. Any assumption on char F is only used in the last step, which concludes from n([xy]) = 0
that ([xy]) = 0, where n = dimV . This goes through if dimV < char F.
3. In characteristic p,
[x, y] = xy yx =

1
2
.
.
.
p 1
0

0
1
.
.
.
p 2
p 1

= x,
4
so L := 'x, y` is a two-dimensional solvable subalgebra of gl(p, F). Since y is diagonal with
distinct entries in F, its eigenvectors are multiples of the standard basis vectors, none of
which is an eigenvector of x. Thus Lies Theorem fails in characteristic p.
4.
5.
6. The base case n = 0 is clear. By induction,
( (a +b).1)
n+1
(xy) = ( a.1 b.1)
n

i=0

n
i

( a.1)
ni
x

( b.1)
i
y

=
n

i=0

n
i

( a.1)
n+1i
x

( b.1)
i
y

+
n

i=0

n
i

( a.1)
ni
x

( b.1)
i+1
y

=
n+1

i=1

n
i

n
i 1

( a.1)
n+1i
x

( b.1)
i
y

=
n+1

i=1

n + 1
i

( a.1)
n+1i
x

( b.1)
i
y

,
where for the third equality we replaced i by i 1 in the second sum two obtain like terms.
7. Let L gl(V ) be solvable. By Lies Theorem, we may choose a basis of V so that L t.
Then [LL] [tt] = n (shown earlier in the text). Clearly [LL]L nt n, so Tr([LL]L) = 0.
Together with Cartans criterion, this shows that L gl(V ) is solvable if and only if Tr(xy) =
0 for all x [LL], y L.
8. Since Tr is linear, it suces to check the hypothesis on x, y ranging over a basis of L.
Recall that L has basis x, y, z satisfying [xy] = z, [xz] = y, [yz] = 0. Then L is solvable:
L
(1)
= [LL] = 'y, z`, L
(2)
= 0. For Cartans criterion (applied to the adjoint representation),
it suces to check that () = Tr(ad ad ) = 0 for y, z and x, y. This is easy
to check directly: under this basis, we have
ad x =

0 0 0
0 0 1
0 1 0

, ad y =

0 0 0
0 0 0
1 0 0

, ad z =

0 0 0
1 0 0
0 0 0

,
and none of the products in question has even a non-zero diagonal entry.
5
5. Killing Form
1. If L is nilpotent, then ad L is nilpotent, so by Engels Theorem there exists a basis of V such
that ad(L) n, hence ad(L) ad(L) n and trivial.
2. By Cartans Criterion and its converse, L solvable ad L solvable ([LL], L) = 0 [LL]
lies in the radical of .
3. Recall that L has basis x, y satisfying [xy] = x. Under this basis, ad x =

0 1
0 0

, ad y =

1 0
0 0

, so (x, x) = (x, y) = (y, x) = 0 but (y, y) = 1. [This agrees with converse of

Cartans Criterion (on ad L): clearly [LL] = 'x`, so (x, x) = (x, y) = (y, x) = 0.]
4.
5. For , L written as column vectors in the standard basis, the text computed on p. 22
that (, ) =
t

4
8
4

. Since

4
8
4

a
b
c

4c
8b
4a

, we get
(x, ax +bh +cy) = 4c
(h, ax +bh +cy) = 8b
(y, ax +bh +cy) = 4a.
Hence the dual basis for x, h, y with respect to is
1
4
y,
1
8
h,
1
4
x.
Alternative, from
ad x =

0 2 0
0 0 1
0 0 0

, ad h =

2 0 0
0 0 0
0 0 2

, ad y =

0 0 0
1 0 0
0 2 0

,
it is manifest that the Killing form vanishes on all pairs of the standard basis vectors except
for (x, y) = 4 and (h, h) = 8.
6. Suppose is nondegenerate, i.e. S = 0. Let I be an arbitrary abelian ideal of L. To
show L semisimple, it suces to show that I S, hence I = 0; that is, we want (I, L) =
Tr(ad I ad L) = 0. Indeed, for any x I, y L, we have
L
ad y
//
L
ad x
//
I
ad y
//
I
ad x
//
[II] = 0
since I is an ideal and abelian, so (ad xad y)
2
= 0 and hence Tr(ad xad y) = 0.
7.
6
8. Write x L (uniquely) as x =

x
i
, x
i
L
i
, and let x
i
= s
i
+ n
i
be the abstract Jordan
decomposition in L
i
. Thus ad
L
i
s
i
(resp. ad
L
i
n
i
) is a semisimple (resp. nilpotent) endomor-
phism in gl(L
i
), and [s
i
n
i
] = 0. Let s =

s
i
and n =

n
i
, so x = s + n. Since L =

L
i
,
for any y =

y
i
L, y
i
L
i
, we have [sy] = [

s
i
,

y
i
] =

[s
i
y
i
]. Thus, considereing
ad : L gl(L), ad
L
s acts on each component L
i
as ad
L
i
s
i
, hence ad
L
s is semisimple. Sim-
ilarly, ad
L
n is nilpotent. Moreover [sn] = [

s
i
,

n
i
] =

[s
i
n
i
] = 0, so x = s + n is the
abstract Jordan decomposition of x by the uniqueness of such.
6. Complete reducibility of representations
1.
2.
3. Let L be solvable and L gl(V ) an irredcuible representation. By Lies Theorem, L has a
common eigenvector v V , which therefore spans an L-stable subspace 'v`. Irreducibility
forces V = 'v`, so dimV = 1.
4. Let L be semisimple. We know ad L Der L; we want to show equality. Let Der L.
Dene an L-action on the direct sum F + L by x.(a, y) = (0, a(x) + [xy]). Then F + L is
an L-module. Indeed, the linearity of the action follows from vector space axioms and the
linearity of and the bracket. Moreover,
x.x

.(a, y) = x.(0, a(x

) + [x

y]) = (0, [x, a(x

) + [x

y]]) = (0, a[x, (x

)] + [x[x

y]])
and similarly x

.x.(a, y) = (0, a[x

, (x)] + [x

[xy]]), so
x.x

.(a, y) x

.x.(a, y) = (0, a([mx, (x

)] + [(x), x

]) + [x[x

y]] [x

[xy]])
= (0, a([xx

]) + [[xx

]y]).
Identify L in F + L via y (0, y). Since x.(0, y) = (0, [xy]), L is an L-submodule of
F + L (adjoint representation). Since F + L is nite-dimensional, by Weyls Theorem, L
has an L-stable complement, i.e. there exists some (a, y) F + L with a = 0 such that
x.(a, y) '(a, y)` for all x L. Since x.(a, y) = (0, a(x) +[xy]), this implies a(x) +[xy] = 0
and so (x) = a
1
[yx] = [a
1
y, x]. Thus = ad(a
1
y) ad L, as desired.
5. (a) Since ad : L ad L gl(L) has kernel Z(L) = Rad L, we have L/ Rad L

= ad L gl(L).
Via the inclusion, L is an ad L-module. Note that an ad L-submodule of L is an ideal
of L. Since ad L

= L/ Rad L is semisimple, by Weyls Theorem, the ideal Z(L) of L has

a complement M, so that L = Z(L) M. Then [LL] [MM] M. On the other
hand, L/Z(L) = L/ Rad L is semisimple and so equals its derived group, so [LL] maps
onto L/Z(L) = M/Z(L) under the canonical map L L/Z(L). [In other words: for
any m M, m + Z(L) = [x + Z(L), y + Z(L)] = [xy] + Z(L) for some x, y L, so
7
[xy] m Z(L) with [xy] [LL] M and m M. Since the sum L = Z(L) M is
direct, this implies m = [xy], hence M [LL].] Hence [LL] = M, i.e. L = Z(L) [LL],
with
(b)
(c)
(d)
6.
7. Assuming sl(n, F) is simple, by 6, (x, y)/ Tr(xy) is a constant (when dened). Let h =
diag(1, 1, 0, . . . , 0), and take x = y = h. Then ad h kills any diagonal matrix, and
[h, e
ij
] =

2e
ij
if i = 1, j = 2
2e
ij
if i = 2, j = 1
e
ij
if i = 1, 3 j n
e
ij
if i = 2, 3 j n
e
ij
if 3 i n, j = 1
e
ij
if 3 i j = 2
0 otherwise,
where e
ij
is the matrix having 1 in the ij-entry and 0 elsewhere, so the standard basis for
sl(n, F) is a basis of eigenvectors for ad h. Then (ad)
2
has eigenvalues 4 (multiplicity 2),
1 (multiplicity 4(n 2)), and the rest 0, so (h, h) = Tr((ad h)
2
) = 8 + 4(n 2) = 4n.
Meanwhile, Tr(h
2
) = Tr(diag(1, 1, 0, . . . , 0)) = 2, so the constant is 4n/2 = 2n.
8. The adjoint action of L on L naturally induces on L-action LL by x.(yz) = [xy]z+y[xz]
and on (L L)

by (x.)(y z) = (x.(y z)) = ([xy] z +y [xz]). Hence

L. = 0 ([xy] z) +(y [xz]) = 0 for all x, y, z L
([yx], z) +(y, [xz]) = 0 for all x, y, z L
associative.
9. For x L

L, let x = s

+ n

= s + n be the Jordan decompositions in L

and in L,
respectively. The adjoint representation ad : L gl(L) also restricts to a representation of
L

, so by the preservation of Jordan decomposition, adx = ad s

+ad n

= ad s+ad n are both

Jordan decompositions in gl(L). Uniqueness forces ad s

= ad s and ad n

= ad n, and since
ad is faithful for semisimple Lie algebras, s

= s and n

= n. Thus the Jordan decompositions

in L

and in L agree.
8
7. Representations of sl(2, F)
1. Since B = 'h, x` is upper triangular, hence a solvable subalgebra of L, for any nite-
dimensional representation : L gl(V ), there exists a simultaneous eigenvector v V
for (B). Since x is nilpotent, so is (x) by the preservation of Jordan decomposition, so
(x)(v) = 0. Thus v is a maximal vector.
2.
3.
8. Root space decomposition
1.
2.
3.
4. Let H be a maximal toral subalgebra and the corresponding set of roots, so L = H
(

). Since dimL

= 1 for each , 3 = dimL = dimH +[[. Since H is abelian,

it cannot be all of L, so dimH 2. But [[ is even, which forces dimH = 1.
5. By the root space decomposition, every x L can be written uinquely as x = h +

,
where h H and x

. If x N
L
(H), then for every h

H, we have [h

x] =

(h

)x

H, so (h

)x

= 0 for all . If x

= 0 for some , then (t

, h

) = (h

) = 0 for
t

= 0 and all h

H, contrary to the nondegeneracy of the Killing form. Hence x

= 0 for
all , so that x H. Thus N
L
(H) H, hence N
L
(H) = H.
6.
7.
8.
9. Let L be a semisimple Lie algebra of dimension 3, and H a maximal toral subalgebra. Semisim-
plicity implies [LL] = L (Corollary 5.2), so L is not abelian; in particular, H = L. As in
4, dimension consideration in the root space decomposition shows that dimH = 1. Writing
= , we have L = H L

= S

= sl(2, F).
10. Follows from classication of simple Lie algebras.
11.
9
III. Root Systems
9. Axiomatics
Choice of simple roots , in rank-2 root systems are from Figure 1 (p. 44).
1. Suppose / E

. Let E

. Since

() = ', ` E

, we have ', ` E

. This
forces ', ` = 0, so P

. Thus E

.
2.

is clearly nite. Since

=
2
(,)
is just multiplication by a nonzero scalar,

spans
E since does, and 0

. Since ()

=
2()
(,)
=
2
(,)
=

, the only multiple of

in

are

.
For

,
'

` =
2

2
(,)
,
2
(,)

2
(,)
,
2
(,)
=
2(, )
(, )
= ', `.
In particular, '

` Z. Moreover,

) =

'

=
2
(, )

2(, )
(, )

2
(, )
=
2
(, )

2(, )
(, )

=
2

()
(

(),

())
= (

())

,
so

leaves

invariant. Viewing the Weyl groups as groups of permutations on and

on

, this calculation also shows that the identication

induces an isomorphism

.
A
1
, A
2
: All roots have the same length, so the coroot system is just the root system scaled
(may as well be taken to be equal).
B
2
: If the roots have lengths 1 and

2, then after scaling they will have length

2
1
2
1 = 2 and
2

2
2

2 =

2, respectively, so B

2
has type B
2
but root lengths reversed.
G
2
: Similarly, lengths 1,

3 become 2,
2
3

3, so analogous conclusion.
3.

is rotation through .
4. It is clear that the reections generate the dihedral groups. Every rank-2 root system is one
of these because they cover all the possible angles between roots.
10
5. B
2
: (, +) = 0 but ( +) = B
2
.
G
2
: (, +) > 0 but (+) = G
2
, and (, +2) = 0 but (+2) = + G
2
6. Any element of the Weyl group can be written

i
. For Aut , we have (

i
)
1
=

1
) =

(
i
)
by Section 9.2 Lemma.
7.
8. Several examples.
The -string through is , +, + 2, + 3, so r q = 0 3 = 3 = ', `.
The -string through + is ( +) , +, ( +) +, ( +) +2, so r q = 12 =
1 = ' +, `.
The -string through is , +, so r q = 0 1 = 1 = ', `.
The -string through + is ( +) , +, so r q = 1 0 = 1 = ' +, `.
9.
10.
11.
10. Simple roots and Weyl group
1.
11. Classication
1.
12. Construction of root systems and automorphisms
1.
2.
3.
13. Abstract theory of weights
1.
2.
3.
11
IV. ISOMORPHISM AND CONJUGACY THEOREMS
14. Isomorphism theorem
1.
2.
3.
15. Cartan subalgebras
1. Fix a semisimple element s sl(n, F). Viewing sl(n, F) as the subalgebra of traceless endo-
morphisms in some gl(n, F), dim
F
V = n, we may choose a basis of eigenvalues for s so that
s = diag(d
1
, . . . , d
n
) with

a
i
= 0. Clearly C(s) contains the maximal toral algebra d(n, F),
so s is regular if and only if C(s) = d(n, F). If s has an eigenvalue of multiplicity at least 2,
say a
i
= a
j
with i = j, then a sl(n, F) with a
ij
, a
ji
= 0 but otherwise diagonal commutes
with s, hence C(s) d(n, F).
Conversely, if s has all distinct eigenvalues, then a C(s) implies sa = as, so equating
coecients yields d
i
a
ij
= d
j
a
ij
. Then a
ij
= 0 whenever i = j, i.e. a d(n, F). Thus
C(s) = d(n, F), so s is regular.
2.
3.
16. Conjugacy theorems
1.
2.
3.
V. EXISTENCE THEOREM
17. Universal enveloping algebras
1.
2.
3.
12
18. Generators and relations
1.
2.
3.
19. The simple algebras
1.
2.
3.
VI. REPRESENTATION THEORY
20. Weights and maximal vectors
1.
2.
3.
21. Finite dimensional modules
1.
2.
3.
22. Multiplicity formula
1.
2.
3.
23. Characters
1.
2.
3.
13
24. Formulas of Weyl, Kostant, and Steinberg
1.
2.
3.
VII. CHEVALLEY ALGEBRAS AND GROUPS
25. Chevalley basis of L
1.
2.
3.
26. Kostants Theorem
1.
2.
3.
27. Admissible lattices
1.
2.
3.
14