Lie Algebra Math 7970 Solutions and Remarks

For historical background of Lie algebra and Lie group see the book review by Brian Parshall on
“Essays in the history of Lie groups and algebraic groups”, by Armand Borel. Also early history.
See a biography of Marius Sophus Lie.

1.1
Example: Verify that the cross product on L = R3 makes L a Lie algebra, i.e.,

x × y = (x2 y3 − x3 y2 , −x1 y3 + x3 y1 , x1 y2 − x2 y1 ),

where x = (x1 , x2 , x3 ) and y = (y1 , y2 , y3 ) ∈ R3 . See Exercises 1.

Questions: 1. Show that (L2’) [x, y] = −[y, x] if and only if (L2) [x, x] = 0 for all x, y ∈ L, provided
that char F 6= 2 (why char F causes trouble?).
2. Show that [x, 0] = 0 for all x ∈ L.
3. Show that if [x, y] 6= 0, then x, y ∈ L are linearly independent.

1.2
Questions: 1. Verify that the bracket [x, y] = xy − yx makes End V a Lie algebra over F.
2. What is the difference between gl(V ) and End (V )?
3. Verify eij ekl = δjk eil and equation (*) [eij , ekl ] = δjk eil − δli ekj .
4. Verify n(n, F), d(n, F), t(n, F) are Lie subalgebras of gl(n, F).
5. Is it true that [n(n, F), n(n, F)] = n(n, F)? Explain.
6. Show that [L, L] is a subalgebra of L. Is it true that [L, L] = L for general L? Give an
example if your answer is not.

o(n, F): Let char F 6= 2. Let f be a nondegenerate symmetric bilinear form on the vector space V
over F. A symmetric bilinear form f : V × V → F is bilinear and

f (v, w) = f (w, v), for all v, w ∈ V.

Nondegenerate means that f (v, w) = 0 for all w ∈ V implies v = 0.

Let (e1 , . . . , en ) be a basis of V . Then the matrix s = (f (ei , ej )) (nonsingular symmetric
P P
matrix) completely characterizes f , i.e., if v = i vi ei , w = i wi ei , and if we identify v and
w as (v1 , . . . , vn )t and (w1 , . . . , wn )t respectively, then f (v, w) = v t sw.
Now
L := o(V ) = {x ∈ gl(V ) : f (x(v), w) + f (v, x(w)) = 0 for all v, w ∈ V }.
It is a subalgebra of gl(V ).

1
 Proof. Since we use the usual bracket of gl(V ), (L1), (L2) and (L3) are clearly satisfied. We
only need to show that o(V ) is closed under bracket. Let x, y ∈ o(V ). Then for all v, w ∈ V
f ([x, y](v), w) + f (v, [x, y](w))
= f ((xy − yx)(v), w) + f (v, (xy − yx)(w))
= f (xy(v), w) − f (yx(v), w) + f (v, xy(w)) − f (v, yx(w)) bilinearity of f
= f (x(y(v)), w) − f (y(x(v)), w) + f (v, x(y(w))) − f (v, y(x(w)))
= −f (y(v), x(w)) + f (x(v)), y(w)) − f (x(v), y(w)) + f (y(v), x(w)) since x, y ∈ o(V )
= 0

Remark: (1) You can also verify it in terms of matrices (see below).
(2) In the above proof we did not use the symmetric property of f . Indeed the proof works for
nondegenerate skew symmetric bilinear form f , i.e. f (v, w) = −f (w, v) for all v, w ∈ V , i.e., a
proof for sp(V ) being a subalgebra of gl(V ). But it turns out skew symmetric nondegenerate
form f on V implies that V is even dimensional.

Remark: We can describe o(V ) in terms of matrices, i.e., o(n, F) as we see on p.3. On p.3, the form f
on V is defined by the matrix s as we just saw, i.e., f (v, w) = v t sw for all v, w ∈ V , where
Ã !

 0 I`

 if n = 2`


I` 0
 
s= 1 0 0

  



  0 0 I`   if n = 2` + 1.


0 I` 0
Notice that s is nonsingular symmetric. So
   

 0 b1 b2 

 t  t t
o(2` + 1, F) = x = −b2 m n  ∈ sl(2` + 1, F) : n = −n, p = −p, m ∈ F`×`

 

−bt1 p −mt
( Ã ! )
m n
o(2`, F) = x= ∈ sl(2`, F) : nt = −n, pt = −p, m ∈ F`×` .
p −mt

Suppose F = C. Let r be an (complex) orthogonal matrix such that rt r = s, namely

à !
e−iπ/4 I iI
r := √
2 iI I

Then the map x 7→ rxr−1 sends x ∈ o(2`, C) to a skew symmetric matrix since
sx = −xt s ⇔ rt rx = −xt rt r ⇔ (rxr−1 )t + rxr−1 = 0
and it is clearly an (algebra) isomorphism. Similar for o(2` + 1, C).

Remark: Indeed the above argument works for algebraically closed field F and char F 6= 2 since every
symmetric bilinear form on the vector space V over F has an orthogonal basis if char F 6= 2
(by induction on the dimension of V , e.g., Serge Lang, Algebra, p.575). In other word, the
symmetric matrix s over F can be written as s = ut du where d is a diagonal matrix over F,
√
then set r = du since F is algebraically closed, i.e., s = rt r.

2
 1.3
Remark: 1. An F-algebra A is a vector space over F equipped with an additional binary operation
(multiplication) from A × A to A, (x, y) 7→ xy denoted by juxtaposition. The binary
operation is bilinear which means that the following identities hold for any three elements
x, y, z ∈ A, and all scalars a, b ∈ F,
(a) (x + y)z = xz + yz.
(b) x(y + z) = xy + xz.
(c) (ax)(by) = (ab)(xy).
Multiplication is not necessarily associative.
2. End V is an F-algebra (associative) with the usual composition as the multiplication. It
is associative but not commutative (when dim V ≥ 2).
3. A Lie algebra is an F-algebra in which the bracket is the multiplication, i.e., xy = [x, y].
4. By (L1) ad x : L → L is a linear map for each x ∈ L and the adjoint representation
ad : L → gl(L) is a linear map. By the Jacobi identity, ad x ∈ Der L for each x ∈ L.
Check!

Questions: 1. Is the term “derivation” justified for a linear map δ : A → A such that δ(ab) = aδ(b) +
δ(a)b?
2. Do Exercise 11 on p.6. Hence Der A is a subalgebra of gl(A) and thus Der L is a
subalgebra of gl(L).
3. Verify that ad x ∈ Der L, i.e., ad x[y, z] = [ad x(y), z] + [y, ad x(z)]. So ad : L → gl(L)
can be considered as ad : L → Der L.
4. Is [H, K] = {[x, y] : x ∈ H, y ∈ K}?
5. Is [H, K] always a subalgebra of L if H, K ⊂ L are subalgebras of L? Explain.

1.4
Examples: The following is a realization of the Lie algebra R3 withcross product in terms
 of linear Lie
0 −x3 x2
 
algebra (Ado’s theorem). Let φ : L → L0 , (x1 , x2 , x3 ) 7→  x3 0 −x1 , where L0 is the
−x2 x1 0
algebra of 3 × 3 real matrices. It is a vector space isomorphism: Check that it is one-to-one
since L and L0 have the same dimension and that

φ(x × y)
= φ(x2 y3 − x3 y2 , −x1 y3 + x3 y1 , x1 y2 − x2 y1 )
 
0 −(x1 y2 − x2 y1 ) −x1 y3 + x3 y1
 
=  x1 y2 − x2 y1 0 −(x2 y3 − x3 y2 )
−(−x1 y3 + x3 y1 ) x2 y3 − x3 y2 0
   
0 −x3 x2 0 −y3 y2
   
=  x3 0 −x1  ,  y3 0 −y1 
−x2 x1 0 −y2 y1 0
= [φ(x), φ(y)].

3
 P
Remark: “A moment of thought on p.5”: With [xi , xj ] = k akij xk (warning akij does not mean the
power (aij )k ; k is just an index)

[xi , xi ] = 0 = [xi , xj ] + [xj , xi ] for all i, j ⇔ akii = 0 = akij + akji , for all i, j, k (1)
P
is implied by (L2). Conversely (L2) follows from (1) since if x = i ci xi , then
X X X X X
[x, x] = [ ci xi , cj xj ] = ci cj [xi , xj ] = c2i [xi , xi ] + ci cj [xi , xj ] = 0.
i j i,j i i6=j

Notice that
X X X X XX
[[xi , xj ], x` ] = [ akij xk , x` ] = akij [xk , x` ] = akij am
k` xm = ( akij am
k` )xm .
k k k m m k

So the Jacobi identity implies

X
[[xi , xj ], x` ] + [[xj , x` ], xi ] + [[x` , xi ], xj ] = 0 ⇔ (akij am k m k m
k` + aj` aki + a`i akj ) = 0, (2)
k

for all i, j, `, m. Set

f (xi , xj , x` ) := [[xi , xj ], x` ] + [[xj , x` ], xi ] + [[x` , xi ], xj ].

Then f (xi , xj , x` ) = 0 implies that f (xσ(i) , xσ(j) , xσ(`) ) = 0 for all permutations σ : {i, j, `} →
{i, j, `}, i.e. checking (2) is sufficient for the ordered triple (i, j, `). The bilinearity of the
P P
bracket and (2) imply the Jacobi identity for x, y, z: Let x = k ak xk , y = k bk xk , z =
P
k ck xk . Then

[[x, y], z] + [[y, z], x] + [[z, x], y]

X X X
= ak [xk , y], z] + ak [[y, z], xk ] + ak [[z, xk ], y]
k k k
X
= ak ([xk , y], z] + [[y, z], xk ] + [[z, xk ], y])
k
X X
= ak ai ([xk , xi ], z] + [[xi , z], xk ] + [[z, xk ], xi ])
k i
XXX
= ak ai aj ([xk , xi ], xj ] + [[xi , xj ], xk ] + [[xj , xk ], xi ])
k i j
= 0

Questions: 1. Check: [x, y] = x actually defines a Lie algebra. Notice that a112 = 1, a212 = 0.
2. Summarize the 1-dimensional and 2-dimensional Lie algebras.

4
Exercises on Section 1

1. (L1) We know that the cross product on R3 is bilinear, i.e.,

[a1 x1 + a2 x2 , y] = (a1 x1 + a2 x2 ) × y = a1 x1 × y + a2 x2 × y = a1 [x1 , y] + a2 [x2 , y]

for all x1 , x2 , y ∈ R3 and a1 , a2 ∈ R and similar for the second coordinate.

(L2) [x, y] = x × y = −y × x = −[y, x] is also well-known.
(L3) By vector triple product a × (b × c) = b(a · c) − c(a · b) we have

[x, [y, z]] + [y, [z, x]] + [z, [x, y]]

= x × (y × z) + y × (z × x) + z × (x × y)
= y(x · z) − z(x · y) + z(y · x) − x(y · z) + x(z · y) − y(z · x) = 0.

Since [e1 , e2 ] = e3 , [e1 , e3 ] = −e2 , [e2 , e3 ] = e1 , the structure constants are

a112 = 0, a212 = 0, a312 = 1,

a113 = 0, a213 = −1, a313 = 0,
a123 = 1, a223 = 0, a323 = 0.

2. Given a basis (x, y, z) of L, [x, y] = z, [x, z] = y, [y, z] = 0, the structure constants are

a112 = a212 = 0, a312 = 1,

a113 = a313 = 0, a213 = 1,
a123 = a223 = a323 = 0.

They satisfy the structure equations (see p.5)

akii = 0 = akij + akji , (the first equality is implied by (L2))

X
(akij am k m k m
k` + aj` aki + a`i akj ) = 0,
k

or for the last equality we just observe (see the note on p.5)

[x, [y, z]] + [z, [x, y]] + [y, [z, x]] = [x, 0] + [z, z] + [y, −y] = 0.

Remark: Such checking is not practical when dim L is getting large.

3. Let à ! à ! à !
0 1 1 0 0 0
x= , y= , z= ∈ sl(2, F).
0 0 0 −1 1 0
Now

ad x(x) = 0 = 0 · x + 0 · h + 0 · y
ad x(h) = [x, h] = xh − hx = −2x = −2x + 0 · h + 0 · y
ad x(y) = [x, y] = xy − yx = h = 0 · x + h + 0 · y.
 
0 −2 0
 
So the matrix of ad x with respect to the basis (x, h, y) is ad x = 0 0 1.
0 0 0

5
 Similarly

ad h(x) = −ad x(h) = 2x

ad h(h) = 0
ad h(y) = −2y.
 
2 0 0
 
Hence ad h = 0 0 0 .
0 0 −2

ad y(x) = −ad x(y) = −h

ad y(h) = −ad h(y) = 2y
ad y(y) = 0.
 
0 0 0
 
Hence ad y = −1 0 0.
0 2 0

4. Given [x, y] = x. Now ad (x)y = x and ad y(x) = −ad x(y) = −x. So

à ! à !
0 1 −1 0
ad x = , ad y = .
0 0 0 0

Hence ad : L → End (L) is injective. Thus ad L ⊂ End (L) is a Lie algebra isomorphic to L.
Via the adjoint representation ad , we have a matrix model for the 2-dimensional L (predicted
by Ado’s theorem p.4). Clearly

ad x ad y − ad y ad x = [ad x, ad y] = ad [x, y] = ad x.

See Example 10 on p.4 of Notes on Lie Algebras to see how it arises as the affine Lie algebra
of the line consisting of all real 2 × 2 matrices with second row 0.

5. It is easy to see that t(n, F), n(n, F) and d(n, F) are invariant (or stable) under the (matrix)
bracket. So they are subalgebras of gl(n, F). We have

t(n, F) = n(n, F) ⊕ d(n, F)

since each A ∈ t(n, F) is the sum of its diagonal part, diag A ∈ d(n, F), and its strictly upper
triangular part, and clearly the intersection of d(n, F) and n(n, F) is trivial, i.e., the zero space.
Notice that d(n, F)n(n, F), n(n, F)d(n, F) ⊂ n(n, F) so that

[d(n, F), n(n, F)] ⊂ n(n, F).

(here [H, K] denotes the subspace of L spanned by commutators [x, y], x ∈ H, y ∈ K. See
p.3.) So

[t(n, F), t(n, F)] = [n(n, F) ⊕ d(n, F), n(n, F) ⊕ d(n, F)]
⊂ [n(n, F), n(n, F)] + [d(n, F), n(n, F)] + [n(n, F), d(n, F)] + [d(n, F) ⊕ d(n, F)]
= [n(n, F), n(n, F)] ⊂ n(n, F)

6
 since d(n, F) is abelian. The rest is to show that n(n, F) ⊂ [t(n, F), t(n, F)], i.e., eij ∈
[t(n, F), t(n, F)] for all i < j. Since

[eij , ek` ] = δjk ei` − δ`i ekj

(equation (*) on p.2) it follows that eij = [eii , eij ] if i 6= j and we have the desired result.

1. t(n, F) has basis eij , 1 ≤ i ≤ j ≤ n. So dim t(n, F) = 1 + · · · + n = n(n + 1)/2.

2. n(n, F) has basis eij , 1 ≤ i < j ≤ n. So dim n(n, F) = 1 + · · · + (n − 1) = n(n − 1)/2.
3. d(n, F) has basis eii , 1 ≤ i ≤ n. So dim d(n, F) = n.

6. Let x ∈ gl(n, F) be a matrix of distinct eigenvalues. It is well known from matrix theory that
a matrix with distinct eigenvalues is diagonalizable. Let y −1 xy = d where d is a diagonal
matrix, i.e., x = ydy −1 .
Recall ad x ∈ End (gl(n, F)). Consider a slightly more general case. Let σ ∈ Aut (gl(n, F))
defined by σ(x) := yxy −1 , x ∈ gl(n, F), where y ∈ gl(n, F) is nonsingular. Then

ad σ(x) = σ(ad x)σ −1

(indeed it is true for all σ ∈ Aut (gl(n, F)). See p.9). It is because

ad (σ(x))(z) = ad (yxy −1 )(z) = [yxy −1 , z] = yxy −1 z − zyxy −1 = y[x, y −1 zy]y −1 .

Hence ad σ(x) and ad x have the same eigenvalues since they are similar.
Let us get back to our case. As σ(d) := ydy −1 = x, ad x and ad d are similar and have
the same eigenvalues. So it is sufficient to consider ad d where d = diag (a1 , . . . , an ) with
a1 , . . . , an distinct eigenvalues of x. Now

ad d(eij ) = (ai − aj )eij , 1 ≤ i, j ≤ n.

So eij (1 ≤ i, j ≤ n) are eigenvectors of ad d with eigenvalue ai − aj and there are n2 such

eigenvalues. They need not be distinct.

7. Each a ∈ gl(n, F) can be decomposed as

Tr a Tr a
a= In + (a − )In ∈ sl(n, F) + s(n, F).
n n
Moreover if b ∈ s(n, F) (the algebra of scalar matrices), i.e., b = λIn , and c ∈ gl(n, F), then

[b, c] = bc − cb = λIn c − cλIn = 0,

i.e., [s(n, F), gl(n, F)] = 0.

It remains to show that the intersection sl(n, F) ∩ s(n, F) is trivial with the assumption that
the characteristic of F, denoted by char F, is zero or else a prime not dividing n (note that in
general char F is either zero or a prime for a field F). Let a ∈ sl(n, F) ∩ s(n, F). So a = λIn
for some λ ∈ F and 0 = Tr a = nλ. Hence λ = 0 and thus a = 0.

7
8. Similar to o(2`, F),
(Ã ! )
m n t t
o(2`, F) = : n = −n, p = p, m ∈ F`×` .
p −mt

So
{eij − e`+j,`+i , 1 ≤ i ≤ j ≤ `} ∪ {e`+i,j − e`+j,i , ei,`+j − ej,`+i : 1 ≤ i < j ≤ `}
is a basis of o(2`, F), where eij ∈ F`×` is the standard basis element with (i, j) entry 1 and
zero elsewhere. Hence

dim o(2`, F) = `2 + `(` − 1)/2 + `(` − 1)/2 = 2`2 − `.

9. Clearly [L, L] ⊂ L. It suffices to show that L ⊂ [L, L] where L is a classical algebra with the
assumption char F = 0.

1. sl(` + 1, F), ` ≥ 1: Notice

2eij = [eii − ejj , eij ] ∈ [L, L], 1 ≤ i 6= j ≤ ` + 1

eii − ejj = [eij , eji ] ∈ [L, L], 1 ≤ i 6= j ≤ ` + 1,

motivated from Ex. 3.

2. sp(2`, F): Notice that
(Ã ! )
m n t t
sp(2`, F) = : n = n, p = p, m ∈ F`×` .
p −mt

To get m ⊕ (−mt ) consider 1 ≤ i 6= j ≤ ` and

2(eij − e`+j,`+i ) = [eii − ejj − e`+i,`+i + e`+j,`+j , eij − e`+j,`+i ] ∈ [L, L]

eii − e`+i,`+i = [ei,`+i , e`+i,i ] ∈ [L, L].

To get n consider 1 ≤ i 6= j ≤ ` and

2(ei,`+j + ej,`+i ) = [eii + ejj − e`+i,`+i − e`+j,`+j , ei,`+j + ej,`+i ] ∈ [L, L]

2(ei,`+i ∓ e`+i,i ) = [eii − e`+i,`+i , ei,`+i ± e`+i,i ] ∈ [L, L].

So ei,`+i ∈ [L, L]. To get p consider 1 ≤ i 6= j ≤ `

2(e`+i,j + e`+j,i ) = [e`+i,`+i + e`+j,`+j − ei,i − ej,j , e`+i,j + e`+j,i ] ∈ [L, L].

3. o(2`, F): Recall

(Ã ! )
m n t t
o(2`, F) = ∈ sl(2`, F) : n = −n, p = −p, m ∈ F`×` .
p −mt

In view of the computation for sp(2`, F) we only need to consider n and p. For n, if
1 ≤ i 6= j ≤ `, then

2(ei,`+j − ej,`+i ) = [eii + ejj − e`+i,`+i − e`+j,`+j , ei,`+j − ej,`+i ] ∈ [L, L].

To get p consider 1 ≤ i 6= j ≤ `

2(e`+i,j − e`+j,i ) = [e`+i,`+i + e`+j,`+j − ei,i − ej,j , e`+i,j − e`+j,i ] ∈ [L, L].

8
 4. o(2` + 1, F): Recall
  

 0 b1 b2 

 t  t t
o(2` + 1, F) = −b2 m n  ∈ sl(2` + 1, F) : n = −n, p = −p, m ∈ F`×`

 −bt p −mt 

1

and the computation is similar.

Remark: We will see in Corollary 5.4 that L = [L, L] is a property of semisimple Lie
algebra and A` − D` are simple.

11. Let δ, δ 0 ∈ Der A . We are going to show that [δ, δ 0 ] ∈ Der A. For a, b ∈ L,

[δ, δ 0 ](ab)
= (δδ 0 − δ 0 δ)(ab)
= δδ 0 (ab) − δ 0 δ(ab)
= δ(aδ 0 (b) + δ 0 (a)b) − δ 0 (aδ(b) + δ(a)b)
= δ(a)δ 0 (b) + aδδ 0 (b) + δδ 0 (a)b + δ 0 (a)δ(b) − (δ 0 (a)δ(b) + aδ 0 δ(b) + δ 0 δ(a)b + δ(a)δ 0 (b))
= aδδ 0 (b) + δδ 0 (a)b − aδ 0 δ(b) − δ 0 δ(a)b
= a[δ, δ 0 ](b) + [δ, δ 0 ](a)b,

i.e., [δ, δ 0 ] ∈ Der A. So Der A is a subalgebra of gl(A). Thus if L is a Lie algebra in which the
product of x, y is just the bracket of x, y, then Der L is a subalgebra of L.
Notice that

(δδ 0 )(ab) = δ(aδ 0 (b) + δ 0 (a)b) = δ(a)δ 0 (b) + aδδ 0 (b) + δδ 0 (a)b + δ 0 (a)δ(b).

Now set δ = δ 0 , a = b ∈ A so that δ 2 (a2 ) = 2(δ(a))2 + aδ 2 (a) + δ 2 (a)a 6= aδ 2 (a) + δ 2 (a)a,

unless (δ(a))2 = 0.

12. It is sufficient to show that if y, z ∈ L are eigenvectors of ad x corresponding to λ and µ

respectively, then [y, z] ∈ L is also an eigenvector of ad x or zero. By the Jacobi identity,

ad x([y, z]) = [x, [y, z]] = [[x, y], z] + [y, [x, z]] = [λy, z] + [y, µz] = (λ + µ)[y, z].