18.

745 Introduction to Lie Algebras November 30, 2010

Lecture 22 — The Universal Enveloping Algebra

Prof. Victor Kac Scribe: Aaron Potechin

Definition 22.1. Let g be a Lie algebra over a field F . An enveloping algebra of g is a pair (ϕ, U ),
where U is a unital associative algebra and ϕ : g → U− is a Lie algebra homomorphism, where U−
stands for U with the bracket [a, b] = ab − ba.

Example 22.1. Let ϕ : g → End V be a representation of g in a vector space V . Then the pair
(End V, ϕ) is an enveloping algebra of g.

Definition 22.2. The universal enveloping algebra of g is an enveloping algebra (Φ, U (g)) which
has the following universal mapping property: for any enveloping algebra (ϕ, U ) of g there exists a
unique associative algebra homomorphism f : U (g) → U such that ϕ = f ◦ Φ.

Exercise 22.1. Prove that the universal enveloping algebra is unique (if it exists).

Solution: Assume (Φ1 , U1 (g)) and (Φ2 , U2 (g)) are both universal enveloping algebras of g. Then
by the universal mapping property of the universal enveloping algebra, we have unique maps
f11 , f12 , f21 , f22 such that for i, j ∈ {1, 2}, Φi = fij ◦ Φj . Now Φi = id ◦ Φi , so by uniqueness,
−1
fii = id. Φi = fij ◦ fji ◦ Φi , so by uniqueness fij ◦ fji = fii = id. f12 = f21 , so (Φ1 , U1 (g)) and
(Φ2 , U2 (g)) are isomorphic, as needed.

Existence of universal enveloping algebras:

Let T (g) be the free unital associative algebra on a basis a1 , a2 , · · · of g and let J(g) be the two-
sided ideal of T (g) generated by the elements ai aj − aj ai − [ai , aj ]. Then U (g) = T (g)/J(g).
Define Φ : g → U (g)− by letting Φ(ai ) = the image of ai in U (g) and extending linearly.

Remark: T (g) is called the tensor algebra over the vector space g.
T (g) = F ⊕ g ⊕ (g ⊗ g) ⊕ (g ⊗ g ⊗ g) ⊕ · · · with the concatenation product.
Good linear algebra textbooks: Artin, Vinberg.

Exercise 22.2. Prove that (Φ, U (g)) is the universal enveloping algebra (i.e the universality prop-
erty holds).

Solution: First, we will show that (Φ, U (g)) is an enveloping algebra. T (g) is a unital associative
algbera, so U (g) = T (g)/J(g) is a unital associative algebra. To check that Φ is a Lie algebra
homomorphism, by linearity, it suffices to check that for all i, j, Φ([ai , aj ]) = [Φ(ai ), Φ(aj )]. But
this is clear, as Φ([ai , aj ]) − [Φ(ai ), Φ(aj )] = [ai , aj ] − ai aj − aj ai is in J(g), so it is zero in U (g).

NowQ let (U, ϕ) Q be another enveloping algebra. Define the map f : T (g) → U by taking
f ( lk=1 aik ) = lk=1 ϕ(aik ) and extending linearly. This is clearly a well-defined unital associative
algebra homomorphism.
Now for all i, j, f ([ai , aj ]−ai aj −aj ai ) = ϕ([ai , aj ])−ϕ(ai )ϕ(aj )−ϕ(aj )ϕ(ai ) = 0 because ϕ : g → U−
is a Lie algebra homomorphism. Thus, J(g) ⊆ ker(f ), so f : U (g) → U is a well-defined unital
associative algebra homomorphism.

1
For all i, (f ◦ Φ)(ai ) = f (ai ) = ϕ(ai ), so by linearity ϕ = f ◦ Φ.

Assume that f 0 : U (g) → U is another unital associative algebra homomorphism with ϕ = f 0 ◦ Φ.

Then f 0 (1) = f (1) = 1 and for all i, f (ai ) = ϕ(ai ) = (f 0 ◦ Φ)(ai ) = f 0 (ai ). 1, a1 , a2 , · · · generate
U (g) as a unital associative algebra, so f = f 0 . Thus, f is unique, as needed.

Corollary 22.1. Any representation π : g → End V extends uniquely to a homomorphism of

associative algebras U (g) → End V (so that ai 7→ π(ai )).

Theorem 22.2. Poincaré-Birkhoff-Witt (PBW) theorem: Let a1 , a2 , · · · be a basis of g. Then the

monomials (*) ai1 ai2 · · · ais with i1 ≤ i2 ≤ · · · ≤ is form a basis of U (g).

Proof. Easy part: the monomials(*) span U (g).

Proof is by induction on the pair (s, N ), where s is the degree of the monomial and N is the number
of inversions, i.e. number of pairs im , in for which m < n but im > in , lexicographically ordered,
i.e. ((s, N ) > (s0 , N 0 ) if s > s0 or s = s0 and N > N 0 )
For N = 0 there is nothing to prove.
If N ≥ 1, then in the monomial we have ait ait+1 where it > it+1 , as otherwise the monomial is
already in our set of monomials(*).
But we have the relation ait ait+1 = ait+1 ait + [ait , ait+1 ], so that in U (g):
ai1 ai2 · · · ais = ai1 · · · ait+1 ait · · · ais + ai1 · · · ait−1 [ait , ait+1 ]ait+2 · · · ais . The first term is a monomial
of degree s and N − 1 inversions, and the second term is the sum of monomials with degree s − 1.
Thus, by the inductive hypothesis, each of these monomials is generated by the monomials(*), so
ai1 ai2 · · · ais is also generated by these monomials, as needed.

The hard part: why are the monomials(*) linearly independent?

Let Bs be the vector space over F with basis bi1 · · · bis with i1 ≤ i2 ≤ · · · is .
Take B0 = F and let B = ⊕s≥0 Bs . We shall construct a linear map f : T (g) → B such that
J(g) ⊆ ker(f ) and f (ai1 · · · ais ) = (bi1 · · · bis ) if i1 ≤ i2 ≤ · · · ≤ is .
This will induce a linear map f : U (g) = T (g)/J(g) → B. Hence the monomials(*) are linearly
independent because bi1 · · · bis , i1 ≤ i2 ≤ · · · is are linearly independent.

Construction: f (1) = 1, f (ai1 · · · ais ) = (bi1 · · · bis ) if i1 ≤ i2 ≤ · · · ≤ is , and

(1) f (ai1 · · · ait ait+1 · · · ais ) = f (ai1 · · · ait+1 ait · · · ais ) + f (ai1 · · · ait−1 [ait , ait+1 ]ait+2 · · · ais )
if it > it+1

By induction on (s, N ), we can use the inversion (1) to reduce f (ai1 · · · ais ) to a sum of terms
of the form f (aj1 · · · ajs0 ), where j1 ≤ j2 ≤ · · · js0 . We just need to check that the final expression
is independent of which sequence of inversions we choose. We do this by induction on (s, N ).

Case 1:
ai1 · · · ais = ai1 · · · ait ait+1 · · · air air+1 · · · ais , where it > it+1 and ir > ir+1
Using the left inversion first gives us
f (ai1 · · · ait+1 ait · · · air air+1 · · · ais ) + f (ai1 · · · [ait , ait+1 ] · · · air air+1 · · · ais )
By the inductive hypothesis, we may use any sequence of inversions (1) to evaluate this, so using
the right inversion on each term, we get
f (ai1 · · · ait+1 ait · · · air+1 air · · · ais ) + f (ai1 · · · ait+1 ait · · · [air , air+1 ] · · · ais ) +
f (ai1 · · · [ait , ait+1 ] · · · air+1 air · · · ais ) + f (ai1 · · · [ait , ait+1 ] · · · [air , air+1 ] · · · ais )

2
Using the right inversion first gives us
f (ai1 · · · ait ait+1 · · · air+1 air · · · ais ) + f (ai1 · · · ait ait+1 · · · [air , air+1 ] · · · ais ).
By the inductive hypothesis, we may use any sequence of inversions (1) to evaluate this, so using
the left inversion on each term, we get
f (ai1 · · · ait+1 ait · · · air+1 air · · · ais ) + f (ai1 · · · [ait , ait+1 ] · · · air+1 air · · · ais ) +
f (ai1 · · · ait+1 ait · · · [air , air+1 ] · · · ais ) + f (ai1 · · · [ait , ait+1 ] · · · [air , air+1 ] · · · ais )
These expressions are the same, so we get the same result whether we begin with the left inversion
or the right inversion.

Case 2: Inversions overlap

ai1 · · · ais = ai1 · · · ait ait+1 ait+2 · · · ais with it > it+1 > it+2 .

Exercise 22.3. Show that we get the same result whether we start with the inversion on a it ait+1
or the inversion on ait+1 ait+2 .

Solution: First using the left inversion gives us

f (ai1 · · · ait+1 ait ait+2 · · · ais ) + f (ai1 · · · [ait , ait+1 ]ait+2 · · · ais )
Using the inversion on ait ait+2 in the first term, we get
f (ai1 · · · ait+1 ait+2 ait · · · ais ) + f (ai1 · · · ait+1 [ait , ait+2 ] · · · ais ) + f (ai1 · · · [ait , ait+1 ]ait+2 · · · ais )
Using the inversion on ait+1 ait+2 in the first term, we get
f (ai1 · · · ait+2 ait+1 ait · · · ais ) + f (ai1 · · · [ait+1 , ait+2 ]ait · · · ais ) + f (ai1 · · · ait+1 [ait , ait+2 ] · · · ais ) +
f (ai1 · · · [ait , ait+1 ]ait+2 · · · ais )

First using the right inversion gives us

f (ai1 · · · ait ait+2 ait+1 · · · ais ) + f (ai1 · · · ait [ait+1 , ait+2 ] · · · ais )
Using the inversion on ait ait+2 in the first term, we get
f (ai1 · · · ait+2 ait ait+1 · · · ais ) + f (ai1 · · · [ait , ait+2 ]ait+1 · · · ais ) + f (ai1 · · · ait [ait+1 , ait+2 ] · · · ais )
Using the inversion on ait ait+1 in the first term, we get
f (ai1 · · · ait+2 ait+1 ait · · · ais ) + f (ai1 · · · ait+2 [ait , ait+1 ] · · · ais ) + f (ai1 · · · [ait , ait+2 ]ait+1 · · · ais ) +
f (ai1 · · · ait [ait+1 , ait+2 ] · · · ais )

Now look at equation (1).

f (ai1 · · · ait ait+1 · · · ais ) = f (ai1 · · · ait+1 ait · · · ais ) + f (ai1 · · · [ait , ait+1 ] · · · ais ) if it > it+1
By skew-symmetry, this gives

f (ai1 · · · ait+1 ait · · · ais ) = f (ai1 · · · ait ait+1 · · · ais ) − f (ai1 · · · [ait , ait+1 ] · · · ais )
= f (ai1 · · · ait ait+1 · · · ais ) + f (ai1 · · · [ait+1 , ait ] · · · ais )

Thus, (1) holds regardless of whether it > it+1 or it+1 > it . By the inductive hypothesis, we may
use this relation freely for monomials of dimension s − 1 and it will not change the result. By
linearity, we have that for any b, c ∈ g
f (ai1 · · · bc · · · ais−1 ) − f (ai1 · · · cb · · · ais−1 ) = f (ai1 · · · [b, c] · · · ais−1 ).
Applying this, we find that the difference between the first expression and the second expression
above is
f (ai1 · · · [[ait+1 , ait+2 ], ait ] · · · ais )+f (ai1 · · · [ait+1 , [ait , ait+2 ]] · · · ais )+f (ai1 · · · [[ait , ait+1 ], ait+2 ] · · · ais )
Using skew-symmetry, this is
f (ai1 · · · [[ait+1 , ait+2 ], ait ] · · · ais )+f (ai1 · · · [[ait+2 , ait ], ait+1 ] · · · ais )+f (ai1 · · · [[ait , ait+1 ], ait+2 ] · · · ais )

3
which is 0 by the Jacobi identity.
Thus, we get the same result regardless of which inversion we start with.

By induction, for any monomial ai1 · · · ais , the evaluation of f (ai1 · · · ais ) is independent of the
sequence of inversions(1) that we use, so f (ai1 · · · ais ) is well-defined. Thus, f : T (g) → B is well-
defined.

It remains to show that J(g) ⊆ ker(f ). By linearity, it suffices to show that for all i, j, for all
A, B ∈ T (g), f (A(ai aj − aj ai − [ai , aj ])B) = 0. If i > j, this is just the equation (1). If i < j, the
eqation (1) gives
f (Aaj ai B) = f (Aai aj B) + f (A[aj , ai ]B). Using skew-symmetry,
f (Aai aj B) = f (Aaj ai B) − f (A[aj , ai ]B) = f (Aaj ai B) + f (A[ai , aj ]B).
f (A(ai aj − aj ai − [ai , aj ])B) = 0, as needed. This completes the proof.

The Casimir Element of U (g)

We assume that dim g < ∞ and g carries a non-degenerate symmetric invariant bilinear form
(·, ·) (e.g. g is semi-simple and (a, b) = K(a, b))

Choose a basis {ai } of g and let bi be the dual basis i.e. (ai , bj ) = δij . The Casimir element
P g
is the following element of U (g): Ω = dim
i=1 ai bi .

Exercise 22.4. Show that Ω is independent of the choice of the basis {ai }.

Solution: From linear algebra, to go from one basis to another, it is sufficient to use the following
three operations:
1. a0i = ai if i 6= j, a0j = caj where c ∈ F , c 6= 0
2. a0i = ai if i 6= j, i 6= k. a0j = ak , a0k = aj
3. a0i = ai if i 6= j. a0j = aj + cak , where c ∈ F
To show that Ω is independent of the choice of the basis {ai }, it is sufficient to show that Ω is
invariant under these three operations. For these three operations, the dual basis changes as follows:
1. b0i = bi if i 6= j, b0j = 1c aj
2. b0i = bi if i 6= j, i 6= k. b0j = bk , b0k = bj
3. b0i = bi if i 6= k. b0k = bk − cbj
P g 0 0 Pdim g
In all three cases, it is easily verified that Ω0 = dimi=1 ai bi = i=1 ai bi = Ω.

Lemma on dual basis:

Lemma 22.3. P For any a ∈ g writeP

(2) [a, ai ] = j αij aj , (3) [a, bi ] = j βij bj , αij , βij ∈ F . Then αij = −βji .

Proof. Taking the inner product of (2) with bj and of (3) with aj , we get
([a, ai ], bj ) = αij and ([a, bi ], [aj ]) = βij . We also have
([a, ai ], bj ) = (a, [ai , bj ]) and ([a, bi ], [aj ]) = (a, [bi , aj ]) = −(a, [aj , bi ]), hence αij = −βji .

Definition 22.3. Let g be a Lie algbera, and V be a g-module.

(We used the language of a representation π of g in V , notation π(g)V , g ∈ g, v ∈ V . A little more

4
convenient is the equivalent language of a g-module V , notation: gV )
A 1-cocycle of g with coefficients in a g-module V is a linear map f : g → V such that
(4) f ([a, b]) = af (b) − bf (a)

Example: Trivial 1-cocycle: for v ∈ V , let fv (a) = av.

Exercise 22.5. Show that fv : g → V is a 1-cocycle.

Solution: fv is clearly linear, and f ([a, b]) = [a, b](v) = a(b(v)) − b(a(v)) = af (b) − bf (a), as needed.
Thus, fv is a 1-cocycle of g.

Denote by Z 1 (g, V ) the space of all 1-cocycles of g with coefficients in V . Then by exercise 22.5,
trivial cocycles form a subspace denoted by B 1 (g, V ).

Definition 22.4. H 1 (g, V ) = Z 1 (g, v)/B 1 (g, V ) is called the first coboundary.

Note that H 1 (g, V ) = 0 just means that any 1-cocycle of g, i.e. any linear map f : g → V satisfying
(4) is trivial, i.e. of the form f = fv for some V .

Theorem 22.4. If g is a semi-simple Lie algebra over a field F of characteristic 0 and V is a

finite-dimensional g-module, then H 1 (g, V ) = 0

Lemma 22.5. If {ai } andP {bj } are dual bases of g and f is a 1-cocycle of g with values in V then
for any a ∈ g we have a( i ai f (bi )) = Ωf (a)

Exercise 22.6. Prove this using Lemma 1.

Solution: The equation (4) gives that for all a, b ∈ g, af (b) = bf (a) + f ([a, b]).
X X
a( ai f (bi )) = a(ai f (bi ))
i i
X X
= a(f ([ai , bi ])) + a(bi f (ai ))
i i
X X X
= f ([a, [ai , bi ]]) + [ai , bi ](f (a)) + a(bi f (ai ))
i i i

By the Jacobi identity, for all a, b, c ∈ g, [a, [b, c]] + [b, [c, a]] + [c, [a, b]] = 0, so
[a, [b, c]] = [b, [a, c]] − [c, [a, b]].
X X X
f ([a, [ai , bi ]]) = f ([ai , [a, bi ]]) − f ([bi , [a, ai ]])
i i i
XX XX
= f ([ai , βij bj ]) − f ([bi , αij aj ])
i j i j
XX XX
= βij f ([ai , bj ]) + αji f ([ai , bj ]) = 0 by Lemma 1
i j j i

5
 X X X
a( ai f (bi )) = [ai , bi ](f (a)) + a(bi f (ai ))
i i i
X X X
= ai bi (f (a)) − bi ai (f (a)) + a(bi f (ai ))
i i i
X X X
= Ωf (a) − bi a(f (ai )) − bi f ([ai , a]) + a(bi f (ai ))
i i i
X X X
= Ωf (a) + bi f ([a, ai ]) + a(bi f (ai )) − bi a(f (ai ))
i i i
XX X
= Ωf (a) + bi f (αij aj ) + [a, bi ]f (ai )
i j i
XX XX
= Ωf (a) + αij bi f (aj ) + βij bj f (ai ) = Ωf (a) by Lemma 1
i j i j