Professional Documents
Culture Documents
Definition 1.1. An algebra A is a vector space V over a field F, endowed with a binary operation
which is bilinear:
Example 1.1. The set of n × n matrices with the matrix multiplication, Matn (F) is an associative
algebra: (ab)c = a(bc).
Example 1.2. Given a vector space V , the space of all endomorphisms of V , End V , with the
composition of operators, is an associative algebra.
Definition 1.3. A Lie algebra is an algebra with product [a, b] (usually called bracket), satisfying
the following two axioms:
1. (skew-commutativity) [a, a] = 0
2. (Jacobi identity) [a, [b, c]] + [b, [c, a]] + [c, [a, b]] = 0
Example 1.3.
1. Take a vector space g with bracket [a, b] = 0. This is called an abelian Lie algebra;
3. Let A be an associative algebra with product ab. Then the space A with the bracket [a, b] =
ab − ba is a Lie algebra, denoted by A .
Exercise 1.1. Show the Jacobi identity holds in Example 1.3.3 for the following cases.
1
Proof. Expanding the Jacobi identity,
Example 1.4. A special case of example 1.3.3: glV = (End V ) , where V is a vector space, is a
Lie algebra, called the general Lie algebra. In the case V = Fn , we denote glV = gln (F), the set of
all n × n matrices with the bracket [a, b] = ab − ba.
Exercise 1.2. Show that tr[a, b] = 0 ∀a, b ∈ Matn (F). In particular, sln is a Lie algebra, called
the special Lie algebra.
Proof. XX
tr [a, b] = (aji bij − bji aij ) = 0
i j
sln is trivially a subspace by the linearity of the trace, and we have shown it to be closed under
the bracket operation. Hence, sln is a subalgebra and is therefore a Lie algebra.
Exercise 1.3. Show that oV,B is a subalgebra of the Lie algebra glV
2
Proof. Consider a, b ∈ oV,B . Then
by the bilinearity of B. Hence, oV,B is closed under the bracket. As oV,B is trivially a subspace of
glV , it is also a subalgebra and therefore a Lie algebra.
Exercise 1.4. Let V = Fn and let B be the matrix of a bilinear form in the standard basis of Fn .
Show that
oFn ,B = a ∈ gln (F)|aT B + Ba = 0
reads, in terms of the standard basis, employing summation convention (a repeated index is summed
over):
B(aij uj e~i , vk e~k ) + B(uj e~j , aik vk e~i ) = 0.
Hence, from the bilinearity of B,
1. son,B (F) if B is a non-degenerate symmetric matrix; this is called the orthogonal Lie algebra.
2. spn,B (F) if B is a non-degenerate skew-symmetric matrix; this is called the symplectic Lie
algebra.
The three series of Lie algebras sln (F), son,B (F) and spn,B (F) are the most important for this
course’s examples.
Convenient notation: If X,Y are subspaces of a Lie algebra g, then [X, Y ] denotes the span of
all vectors [x, y], where x ∈ X,y ∈ Y .
3
Definition 1.4. Let g be a Lie algebra. In the above notation, a subspace h ⊂ g is a subalgebra
if [h, h] ⊂ h. A subspace h of g is called an ideal if [h, g] ⊂ h.
Dim 1. g = Fa, [a, a] = 0 so the Abelian Lie algebra is the only one.
Dim 2. Consider [g, g] ⊂ g. Let g = Fx + Fy, then [g, g] = F[x, y]. Therefore, dim[g, g] ≤ 1.
Exercise 1.5. Let f : Matn (F) → F be a linear function such that f ([a, b]) = 0. Show that
f (a) = λtr (a), for some λ independent of a ∈ Matn (F).
By linearity of f ,
(aij bjk − bij ajk )f (eik ) = 0 ∀a, b ∈ Matn (F)
where summation convention has been used. Let a = emn , b = enn for some m 6= n. Then
f (emn ) = 0. Hence
f ([a, b]) = (aij bji − bij aji )f (eii ) = 0.
But f (eii ) = f (ejj ) ∀i, j as f (eii ) − f (ejj ) = f (eij eji − eji eij ) = 0. Hence f (eii ) = λ for some
constant λ, and f (a) = tr (a)f (eii ) = λtr (a)
4
18.745 Introduction to Lie Algebras September 14, 2010
We saw in the previous lecture that we can form a Lie algebra A− , from an associative algebra A,
with binary operation the commutator bracket [a, b] = ab − ba. We also saw that this construction
works for algebras satisfying any one of a variety of other conditions.
As Algebras of Derivations
Lie algebras are often constructed as the algebra of derivations of a given algebra. This corresponds
to the use of vector fields in geometry.
Definition 2.1. For any algebra A over a field F, a derivation of A is an F-vector space endo-
morphism D of A satisfying D(ab) = D(a)b + aD(b). Let Der(A) ⊂ glA be denote the space of
derivations of A.
For an element a of a Lie algebra g, define a map ad(a) : g → g, by b 7→ [a, b]. This map is referred
to as the adjoint operator. Rewriting the Jacobi identity as
we see that ad(a) is a derivation of g. Derivations of this form are referred to as inner derivations
of g.
Proposition 2.1.
(b) The inner derivations of a Lie algebra g form an ideal of Der(g). More precisely,
1
Solution: The derivations of A are those maps D ∈ glA which satisfy D(ab) − D(a)b − aD(b) = 0
for all a and b in A. For fixed a and b, the left hand side of this equation is linear in D, so that
the set of endomorphisms satisfying that single equation is a subspace. The set of dervations is the
intersection over all a and b in A of these subspaces, which is a subspace.
We are only left to check that the bracket of two derivations is a derivation. For any a, b ∈ A and
D1 , D2 ∈ Der(A) we calculate:
[D1 , D2 ](ab) = D1 (D2 (a)b + aD2 (b)) − D2 (D1 (a)b + aD1 (b))
= D1 D2 (a)b + D2 (a)D1 (b) + D1 (a)D2 (b) + aD1 D2 (b)
− {D2 D1 (a)b + D1 (a)D2 (b) + D2 (a)D1 (b) + aD2 D1 (b)}
= D1 D2 (a)b − D2 D1 (a)b + aD1 D2 (b) − aD2 D1 (b)
= [D1 , D2 ](a)b + a[D1 , D2 ](b).
Thus the derivations are closed under the bracket, and so form a Lie subalgebra of glA .
Show that this bracket satisfies the axioms of a Lie algebra if and only if {xi , xi } = 0, {xi , xj } =
−{xj , xi } and any triple xi , xj , xk satisfy the Jacobi identity.
Solution: If the Poisson bracket defines a Lie algebra structure for some choice of values {xi , xj },
then in particular, the axioms of a Lie algebra must be satisfied for brackets of terms xi . The
interesting question is whether the converse holds. We suppose then that the {xi , xj } are chosen
so that {xi , xj } = −{xj , xi }, and so that the Jacobi identity is satisfied for triples xi , xj , xk .
The bilinearity of the bracket follows from the linearity of differentiation, and the skew-symmetry
follows from the assumption of the skew symmetry on the xi .
At this point we introduce some shorthands to simplify what follows. If f is any function, we write
fi for the derivative of f with respect to xi . When we are discussing an expression e in terms of
three functions f, g, h, we will write CS(e) for the ‘cyclic summation’ of e, the expression formed
by summing those obtained from e by permuting the f, g, h cyclically. In particular, the Jacobi
identity will be CS({f, {g, h}}) = 0.
First we calculate the iterated bracket of monomials xi :
X
{xi , {xj , xk }} = {xj , xk }l {xi , xl } (an example of the shorthands described).
l
Now the iterated bracket of any three polynomials (or functions) f , g and h is:
X X
{h, {f, g}} = [fil gj hk + gjl fi hk ] {xi , xj }{xk , xl } + fi gj hk {xi , xj }l {xk , xl }.
i,j,k,l i,j,k,l
2
By the assumption that the Jacobi identity holds on the xi , we have (for any i, j, k):
X
CS(fi gj hk ){xi , xj }l {xk , xl } = 0,
l
for cyclicly permuting the f, g, h corresponds to cyclicly permuting the i, j, k (in the opposite order).
Thus we have: X
CS ({h, {f, g}}) = CS(fil gj hk + gjl fi hk ){xi , xj }{xk , xl }.
i,j,k,l
The remaining task can be viewed as finding the {xα , xβ }{xγ , xδ } coefficient in this expression,
where we substitute all appearences of {xβ , xα } for −{xα , xβ }, and so on. To do so, we tabulate
all the appearances of terms which are multiples of {xα , xβ }{xγ , xδ }. We may as well assume here
that α < β and γ < δ.
Given a basis e1 , . . . , en of a Lie algebra g over F, the bracket is determined by the structure
constants ckij ∈ F, defined by:
X
[ei , ej ] = ckij ek .
k
The structure constants must satisfy the obvious skew-symmetry condition (ckii = 0 and ckij = −ckji ),
and a more complicated (quadratic) condition corresponding to the Jacobi identity.
3
Definition 2.2. Let g1 , g2 , be two Lie algebras over F and ϕ : g1 → g2 a linear map. We say that
ϕ is a homomorphism if it preserves the bracket: ϕ([a, b]) = [ϕ(a), ϕ(b)], and an isomorphism if it is
bijective. If there exists an isomorphism ϕ, we say that g1 and g2 are isomorphic, written g1 ∼= g2 .
Exercise 2.3. Let ϕ : g1 → g2 be homomorphism. Then:
(b) im ϕ is a subalgebra of g2 .
(c) im ϕ ∼
= g1 / ker ϕ.
(b) The image im ϕ is a subspace, again as ϕ is F-linear. Now for any u, v ∈ im ϕ, we may write
u = ϕ(x) and v = ϕ(y) for elements x, y ∈ g1 . Then [u, v] = [ϕ(x), ϕ(y)] = ϕ([x, y]) ∈ im ϕ.
Thus the image is a subalgebra.
(c) Consider the map ψ : g1 / ker ϕ → im ϕ given by x + ker ϕ 7→ ϕ(x). We must first see that ψ
is well defined. If x + ker ϕ = x0 + ker ϕ, then x0 − x ∈ ker ϕ, so that:
Thus our definition of ψ does not depend on choice of representative, and ψ is well defined.
It is trivial to see that ψ is a homomorphism. Now suppose that x + ker ϕ ∈ ker ψ. Then
ϕ(x) = 0, so that x ∈ ker ϕ, and x + ker ϕ = 0 + ker ϕ. Thus ψ is injective, and that ψ is
surjective is obvious. Thus ψ is an isomorphism g1 / ker ϕ → im ϕ.
Definition 2.3. An algebraic group G over a field F is a collection {Pα }α∈I of polynomials on the
space of matrices Matn (F) such that for any unital commutative associative algebra A over F, the
set
G(A) := {g ∈ Matn (A) | g is invertible, and Pα (g) = 0 for all α ∈ I}
is a group under matrix multiplication.
Example 2.2. The general linear group GLn . Let {Pα } = ∅, so that GLn (A) is the set of invertible
matrices with entries in A. This is a group for any A, so that GLn is an algebraic group.
Example 2.3. The special linear group SLn . Let {Pα } = {det(xij ) − 1}, so that SLn (A) is the set
of invertible matrices with entries in A and determinant 1. This is a group for any A, so that SLn
is an algebraic group.
Exercise 2.4. Given B ∈ Matn (F) , let On,B (A) = {g ∈ GLn (A) : g T Bg = B}. Show that this
family of groups is given by an algebraic group.
4
Solution: For any unital commutative associative algebra A over F, the set
We then only have to show that the condition g T Bg = B can be written as a collection of polynomial
equations in the entries gij of the matrix g, with coefficients in F. This is obvious — we have one
polynomial equation for each of the n2 entries of the matrix, and the coefficients depend only on
the entries of B.
Definition 2.4. Over a given field F, define the algebra of dual numbers D to be
D := F[]/(2 ) = {a + b | a, b ∈ F, 2 = 0}.
Example 2.4. (1) Lie GLn = GLn (F), since (In + X)−1 = In − X. (In − X approximates
the inverse to order two, but over dual numbers, order two is ignored).
Solution: For (2), We need only prove the formula det(In + X) = 1 + tr (X). It is trivial when
n = 1, and we proceed by induction on n. Consider the matrix In + X, and the cofactor expansion
of the determinant along the final column. If i < n, the (i, n) entry is a multiple of , and so is every
entry of the ith column of the matrix obtained by removing the row and column containing (i, n).
Thus the corresponding cofactor has no contribution to the overall determinant. The determinant is
therefore 1 + Xnn multiplied by the minor corresponding to (n, n). By induction, the determinant
is (1 + Xnn )(1 + (tr (X) − Xnn )). The result follows.
For (3), the following calculation gives the result:
Proof. We first show that Lie G is a subspace. Indeed, X ∈ Lie G iff Pα (In + X) = 0 for all α.
Using the Taylor expansion:
X ∂Pα
Pα (In + X) = Pα (In ) + (In ) xij ,
∂xij
i,j
as 2 = 0. Now as Pα (In ) = 0 (every group contains the identity), this condition is linear in the
xij , so that Lie G is a subspace.
5
Now suppose that X, Y ∈ Lie G. We wish to prove that XY − Y X ∈ Lie G. We have:
(In + X)(In + 0 Y )(In + X)−1 (In + 0 Y )−1 = In + 0 (XY − Y X) ∈ G(F[, 0 ]/(2 , (0 )2 )).
6
18.745 Introduction to Lie Algebras September 16, 2010
Definition 3.1. Let g be a Lie algebra over a field F and V a vector space over F. A representation
of g in V is a homomorphism π : g → glV . In other words, it is a linear map a → π(a) from g to
the space of linear operators on V such that π([a, b]) = π(a)π(b) − π(b)π(a).
Example 3.1. Trivial representation of g in V where π(a) = 0 for all a.
Example 3.2. Adjoint representation of g in g : a → ada.
Let’s check that it is in fact a representation. We must show that
ad[a, b] = (ada)(adb) − (adb)(ada).
Applying both sides to c ∈ g, we check
[[a, b], c] = [a, [b, c]] − [b, [a, c]].
By skew-symmetry, this is just the Jacobi Identity.
Definition 3.2. The center of a Lie algebra g is denoted Z(g) = {a ∈ g|[a, g] = 0}.
Clearly, Z(g) is an ideal of g.
Exercise 3.1. Show Z(gln (F)) = FIn , Z(sln (F)) = 0 if char F - n.
Proof. This is clear when n = 1 so assume otherwise. Suppose A = (aij ) ∈ Z(gln (F)) and there
exists aij 6= 0 with i 6= j. Let B = (bij ) and consider AB = C = (cij ) and BA = C 0 = (c0ij ) . We
wish to show C 6= C 0 for some B. We have
n
X
ci1 = aik bk1
k=1
and
n
X
c0i1 = bik ak1 .
k=1
Define B by bj1 = 1 and bk1 = 0 for all k 6= j. We then have ci1 = aij 6= 0 and that the ith row of B
only has the restriction bi1 = 0. We choose the remaining entries of the row so that c0i1 6= ci1 = aij .
Therefore aij = 0 for all i 6= j. We observe that these restrictions still allow B to have trace zero
and determinant non-zero.
Now suppose without loss of generality that a11 6= a22 . Consider
n
X
c12 = a1k bk2
k=1
and let b12 = 1 and bi2 = 0 for all i 6= 1. These restrictions on B still allow B ∈ sln (F) ⊂ gln (F).
Clearly FIn ⊂ Z(gln (F)), so by above FIn = Z(gln (F)). As well, since we allowed B ∈ sln (F) and
tr(A) = na11 , if char F - n, then Z(sln (F)) = 0.
1
Proposition 3.1. The adjoint representation defines an embedding of g/Z(g) in glg .
Proof. Suppose dim Z(g) = dim g − 1 and pick any non-zero x ∈ g \ Z(g). Clearly, x commutes
with Z(g) and with αx which implies x ∈ Z(g). This is a contradiction and therefore dim Z(g) 6=
dim g − 1.
Definition 3.3. We define Heis2n+1 to be the Lie algebra with basis {pi , qi , c} where [pi , qi ] = c =
−[qi , pi ], 1 ≤ i ≤ n, and all other bracketed pairs are 0.
Exercise 3.3. Classify all finite dimensional Lie algebras for which dim Z(g) = dim g − 2. Let
dim g = n and show either g ∼ = Abn−3 ⊕ Heis3 or g ∼ = Abn−2 ⊕ h where h is the two-dimensional
non-abelian Lie algebra.
Proof. Suppose dim Z(g) = dim g − 2, then g/Z(g) may be generated by two elements, and consider
their preimages, say p and q. Let [p, q] = c 6= 0 (else p, q ∈ Z(g). Suppose c ∈ Z(g), then in this
case g ∼= Abn−2 ⊕ Heis3 . Assume otherwise, that c = z + ap p + aq q (without loss of generality,
assume ap 6= 0). We have [c, q] = ap [p, q] = ap c. Let q 0 = aqp , then c, q 0 are linearly independent
and [c, q 0 ] = c which shows g ∼
= Abn−2 ⊕ two-dimensional non-abelian Lie algebra.
Definition 3.6. A linear operator A in a vector space V is called nilpotent if AN = 0 for some
positive integer N .
Exercise 3.4. Show if dim V < ∞, then A is nilpotent if and only if all eigenvalues of A are zero.
2
Lemma 3.3. Let A be a nilpotent operator in a vector space V , then
(a) There exists a non-zero v ∈ V such that Av = 0.
(b) ad A is a nilpotent operator on glV .
Proof. (a) Consider minimal N > 0 such that AN = 0, then AN −1 6= 0. Choose a non-zero vector
v ∈ AN −1 V 6= 0. Then Av = 0.
Remark 2. ad A = LA − RA
LA (B) = AB, RA (B) = BA
LA RB = RB LA , due to the associativity of the product of operators. Hence
M
M M −j
LjA RA
X
(ad A)M = .
j
j=0
since either j ≥ N or 2N − j ≥ N .
Theorem 3.4. Engel’s Theorem Let V be a non-zero vector space and let g ∈ glV be a finite
dimensional subalgebra which consists of nilpotent operators. Then there exists a non-zero vector
v ∈ V such that Av = 0 for all A ∈ g.
Remark 3. If we assume dim V < ∞, then g is finite dimensional since dim g ≤ (dim V )2 < ∞.
Therefore Engel’s Theorem holds if we only assume dim V < ∞.
3
Corollary 3.5. Let π : g → glV be a representation of a Lie algebra g in a finite dimensional
vector space V such that π(a) is a nilpotent operator for all a ∈ g. Then there exists a basis of V
in which all operators π(a), a ∈ g are strictly upper triangular matrices.
Exercise 3.5. Construct in sl3 (F) a two-dimensional subspace consisting of nilpotent matrices,
which do not have a common eigenvector.
Then the subspace generated by A and B is two dimensional. A has eigenspace generated by
(1, 0, 0)t and B has eigenspace generated by (0, 0, 1)t . Therefore A and B have no common eigen-
vectors. Any linear combination of A, B say αA + βB has characteristic polynomial −λ3 and
therefore is nilpotent.
4
18.745 Introduction to Lie Algebras September 21, 2010
Definition 4.1. Let g be a Lie algebra over a field F. The lower central series of g is the descending
chain of subspaces
We note that
Definition 4.2. A lie algebra g is called nilpotent (resp. solvable) if gn = 0 for some n > 0 (resp.
g(n) = 0 for some n > 0).
Example 4.2. Let H3 = Fp + Fq + Fc with [c, g] = 0 and [p, q] = c. Then H32 = Fc, H33 = 0.
Example 4.3.
Exercise 4.1. Show bn is a solvable (but not nilpotent) Lie algebra and that [bn , bn ] = ηn (n ≥ 2).
Also show that ηn is a nilpotent Lie algebra.
1
We notice aik = bik = 0 if k < i and bkj = akj = 0 if k > j. Thus
j
X
cij = (aik bkj − bik bkj )
k=i
Then if i = j, cii = aii bii − bii aii = 0, implying C ∈ ηn , i.e., [bn , bn ] ∈ ηn . Define the kth diagonal
to be the set of cij where j − i = k. We show that the kth diagonal contains no nonzero entries
in bk+1
n . We have shown this to be true for k = 0. Now assume it is true up to k = m for some
m ≥ 0. Then any C = [A, B] ∈ bm+1 n is such that A, B ∈ bm
n and thus
i+m
X
ci,i+m = (aik bk,i+m − bik ak,i+m )
k=1
First we note
1. Any subalgebra of a nilpotent (resp. solvable) Lie algebra is nilpotent (resp. solvable).
2. Any factor algebra of a nilpotent (resp. solvable) Lie algebra is nilpotent (resp. solvable)
Exercise 4.2. Let g be a Lie algebra and h ⊂ g be an ideal. Show that if h is solvable and g/h is
solvable, then g is solvable too.
Proof. First we prove that all the homomorphic images of a solvable algebra are solvable. Let g1
be solvable and φ : g1 → g2 a surjective homomorphism. We show
(i) (i)
φ(g1 ) = g2
Thus if g1 is solvable, so is g2 . Now suppose h ⊆ g is a solvable ideal, say h(n) = 0, and (g/h)(m) = 0.
Consider the canonical homomorphism
2
π : g → g/h From the previous result,
The last exercise does not hold if we everywhere put “nilpotent” in place of “solvable,” as the
following example shows.
Example 4.4. Suppose g= Fa+Fb, [a, b] = b. Fb ⊂ g is an ideal, Fb and g/Fb are 1-dimensional
and hence abelian and nilpotent. But g is not nilpotent.
Theorem 4.1. (a) If g is a nonzero nilpotent Lie algebra then Z(g) is nonzero
(b) If g is a finite-dimensional Lie algebra such that g/Z(g) is nilpotent, then g is nilpotent.
Proof. (a) Take N > 0 minimal such that gN = 0. Since g 6= 0, N ≥ 2, but then gN −1 6= 0 and
[g, gN −1 ] = gN = 0, so gN −1 ⊂ Z(g).
(b) g = g/Z(g) is nilpotent, i.e., gn = 0 for some n which implies gn ⊂ Z(g), but then gn+1 =
[g, gn ] ⊂ [g, Z(g)] = 0.
Theorem 4.2. Let g be a finite-dimensional Lie algebra. Then g is nilpotent iff for each a ∈ g,
(ad a)n = 0 for some n > 0. One may always take n = dimg.
proof If g is nilpotent then gn+1 = 0 for some n. In particular, (ad a)n b = 0 for all a, b ∈ g sinc
this is a length (n + 1) commutator. For the converse: The adjoint representation gives an injective
homomorphism g/Z(g) ,→ glg and by assumption the image consists of nilpotent operators. So by
Engel’s Theorem (from last lecture), g/Z(g) consists of strictly upper triangular matrices in the
same basis. Therefore g/Z(g) is nilpotent and hence g is nilpotent as well.
Let g be n-dimensional and nilpotent with Z(g) 6= 0 so g/Z(g) is nilpotent of dimension n1 < n.
Definition 4.3. • g is 1-step nilpotent if it is abelian
B: V ×V → Z(g)
(a,b) 7→ [ã, b̃]
3
where ã and b̃ are preimages of a, b under g → V (B is an alternating form, i.e., B(x, x) = 0 for all
x).
Exercise 4.3. Show that 2-step nilpotent Lie algebras are classified by such nondegenerate alter-
nating bilinear forms.
Proof. Suppose g is a 2-step Lie algebra so g/Z(g) is abelian. Let W = Z(g) and V ∼ = g/Z(g). We
check that the form φ : (v1 , v2 ) → [v1 , v2 ] is nondegenerate and alternating. It is clearly alternating
by the definition of the bracket, and nondegenerate since if [v, v] = 0 then v ∈ Z(g) ⇒ v = 0. For
the other direction, given a triple (v, w, φ) such that φ : v × v → w, consider g = v ⊕ w. Then 2-step
0 0 0
nilpotent Lie algebras with bracket [v +w, v +w ] = φ(v, v ). This is the case because the bracket is
alternating by the definition of φ, the bracket satisfies the Jacobi identity (see the next paragraph),
and since the bracket is nondegenerate, V ∼ = g/Z(g). To check that the bracket satisfies the Jacobi
identity, check that
[v1 , [v2 , v3 ]] + [v2 , [v3 , v1 ] + [v3 , [v1 , v2 ]] = φ(v1 , φ(v2 , v3 )) + φ(v2 , φ(v3 , v1 )) + φ(v3 , φ(v1 , v2 ))
= φ(v1 , 0) + φ(v2 , 0) + φ(v3 , 0) = 0
We must show these maps are isomorphisms. Let α : g → (v, w, φ) and β : (v, w, φ) → g. We check
that αβ ∼
= 1(v,w,φ) and βα ∼= 1g . We’ve seen that β sends a triple to the Lie algebra v ⊕w where w is
0 0 0
the center of the form [v + w, v + w ] = φ(v, v ) ∈ w which gets mapped to (v ⊕ w, w, φ) ∼ = (v, w, φ)
The other direction g → (g/Z(g), Z(g)) → g/(Z(g) ⊕ Z(g)) giving a bijection.
You can show that the problem of classifying all nilpotent algebras is equivalent to problems that
are known to be impossible. However, you can classify things in some special circumstances.
Exercise 4.4. Show that if Z(g) = Fc and g is 2-step nilpotent, then g is isomorphic to H2n+1 =
(Fp1 + Fp2 + . . . + Fpn ) + (Fq1 + Fq2 + . . . + Fqn ) + Fc with [pi , qj ] = δij , [pi , pi ] = 0, [qi , qj ] = 0,
and [c, H2k+1 ] = 0.
Proof. From the previous exercise there exists a skew-symmetric form B on V := g/Fc, B(v1 , v2 ) =
[v1 , v2 ] and just as above it is easy to check that it is nondegenerate. But for any non-degenerate
skew-symmetric bilinear form B on V over any field F there exists a basis pi , qi such that B(pi , qj ) =
δij , B(pi , pj ) = 0, and B(qi , pj ) = 0. Indeed, pick arbitrary p1 ∈ V and a q1 such that B(p1 , q1 ) = 1,
and let V1⊥ be the orthocomplement to Fp1 + Fq1 in V . Continue by induction on dimV . Then look
at the preimages of these pi and qi in the space g, and note that they satisfy the same commutation
relations. This implies that H = (Fp1 + Fp2 + . . . + Fpn ) + (Fq1 + Fq2 + . . . + Fqn ) + Fc with the
desired commutation relations.
4
18.745 Introduction to Lie Algebras September 23, 2010
1 Lie’s Theorem
Notation 1.1. Let V be a vector space over a field F. We will denote by V ∗ the dual vector space.
Lemma 1.1 (Lie’s Lemma). Let g be a Lie algebra and h ⊂ g an ideal both over F algebrically
closed and of characteristic 0. Let π be a representation of g in a finite dimensional F−vector space
V . Then each weight space Vλh for the restricted representation π|h is invariant under g.
Proof. We wish to show that if v ∈ Vλh , then π(a)v ∈ Vλh , ∀a ∈ g. But this is verified if and only if
Now,
Since V is finite dimensional, then there exists N ∈ N s.t. N is the maximal integer for which
all the generators of WN are linearly indipendent. Then we have WN = WN +1 = . . . , hence
1
π(a)WN ⊂ WN .
We will consider the increasing sequence of subspaces
thus,
π(h)π(a)m v − λ(h)π(a)m v ∈ Wm−1
because it is a sum of elements in Wm−1 . This concludes the proof of the inductive step.
We know that WN is invariant both for π(a) and for π(h), ∀h ∈ h. In particular,(4) shows that
∀h ∈ h, π(h) acts on WN as an upper triangular matrix, with in the basis {v, π(a)v, . . . , π(aN )v},
λ(h) ∗ ... ∗ ∗
.. ..
0
. ∗ ... .
0
0 λ(h) ∗ ∗
.. .
..
0 0 0 .
0 0 0 0 λ(h)
which implies that λ([h, a]) = 0, since char F = 0. This concludes the proof.
Lie’s Theorem. Let g be a solvable Lie algebra and π a representation of g on a finite dimensional
vector space V 6= 0, over an algebraically closed field F of characteristic 0. Then there exists a weight
λ ∈ g∗ for π, that is Vλg 6= {0}.
2
Proof. We can suppose that g is finite dimensional, since V is finite dimensional and the represen-
tation factors in the following way:
gA π / End(V )
AA u:
AAφ i uuu
AA u
A uu
uu
π(g)
As g is solvable, also π(g) will be solvable. This follows from the fact that π(g)(n) = π(g(n) ) - by
induction on n.
We will prove Lie’s theorem by induction on the dimension of g, dim g = m.
The case dim g = 0 is trivial.
Suppose now that we have proved Lie’s theorem dim g = m − 1, we want to show that the theorem
holds also for g, dim g = m ≥ 1.
Since g is solvable, of positive dimension, g properly includes [g, g]. Since g/[g, g] is abelian, any
subspace is automatically an ideal.
Take a subspace of codimension one in g/[g, g], then its inverse image h is an ideal of codimension
one in g (including [g, g]). Thus, we have the following decomposition of g as a vector space
g = h + Fa.
Now, dim h = m − 1 and it is solvable (since an ideal of a solvable Lie algebra is solvable), hence by
inductive hypothesis we can find a non-zero weight space Vλh 6= {0}, λ ∈ h∗ . By Lie’s lemma, Vλh
is invariant under the action of π(g). In particular, π(a)Vλh ⊂ Vλh , hence (since F is algebraically
closed) there exists 0 6= v ∈ Vλh such that π(a)v = la, for some l ∈ F. We define a linear functional
λ0 ∈ g∗ on g by
λ0 (h + µa) = λ(h) + µl, ∀h ∈ h, l ∈ F.
Thus, by construction, we see that v belongs to Vλg0 . In particular Vλg0 6= {0}.
Exercise. 5.1 Show that we may relax the assumption on F in Lie’s Lemma. Show Lie’s Lemma
under the assumptions that F is algebraically closed and that dim V < char F.
In the proof, we calculated the trace of a certain endomorphism of W ⊆ V to be zero, and also to
be N q, where N = dim W and q was a quantity that we needed to show was zero. Of course then
if the characteristic of F exceeds the dimension of V , dim W is non-zero in F and we may conclude
q = 0. Given this weaker assumption, the rest goes through unaltered.
Exercise. 5.2 Consider the Heisenberg algebra H3 and its representation on F[x] given by
c 7→ Id,
p 7→ (f (x) 7→ xf (x)), ∀f (x) ∈ F[x],
d
q 7→ (f (x) 7→ f (x)), ∀f (x) ∈ F[x].
dx
Show that the ideal generated by xN , 0 < N = char F, in F[x] is invariant for the representation
of H3 and that the induced representation of H3 on F[x]/(xN ) has no weight.
3
(xN ) is a subrepresentation, as it is preserved by the action of p and c, and the action of q annihilates
the xN term when differentiating. Now consider v = a0 x0 + a1 x1 + . . . + aN −1 xN −1 , a representative
of an element of the quotient. All members of the quotient will be represented this way. Now if v
is a weight vector, its derivative must be kv for some k, so that iai = kai−1 for each i < N . On the
other hand, in order that it’s an eigenvector for p, it must have zero constant term, or have pv = 0.
These two statements show easily that v = 0, (as i 6= 0 when i = 1, . . . , N − 1). Thus there is no
weight vector.
• for all representations π of a solvable Lie algebra g on a finite dimensional vector space V over
an algebraically closed field F, char F = 0, there exists a basis for V for which the matrices
of π(g) are upper triangular;
The second statement is simply an application of the first. We prove the first by induction on the
module’s dimension. It is trivial in dimension 1. Suppose V is a module. Use Lie’s theorem to find
a weight v of V . The quotient module V /Fv, by induction, can given a basis such that g acts by
upper triangular matrices. Taking any preimages of that basis in V , and extending it to a basis of
V by including v, we obtain a basis of V on which g acts by upper triangular matrices.
Exercise. 5.4 Let g be a finite dimensional solvable Lie algebra over the algebrically closed field
F, char F = 0. Show that [g, g] is nilpotent.
If [g/Z(g), g/Z(g)] is nilpotent, so is [g, g], so using the adjoint representation, we may assume g
is a solvable subalgebra of glg . Then by the previous, there is a basis in which g consists of upper
triangular matrices. Then [g, g] consists of strictly upper triangular matrices, and is a subalgebra
of the nilpotent Lie algebras un .
4
18.745 Introduction to Lie Algebras September 28, 2010
Definition 6.1. Let A be a linear operator on a vector space V over field F and let λ ∈ F, then
the subspace
Vλ = {v | (A − λI)N v = 0 for some positive integer N}
is called a generalized eigenspace of A with eigenvalue λ. Note that the eigenspace of A with
eigenvalue λ is a subspace of Vλ .
Proposition 6.1. Let A be a linear operator on a finite dimensional vector space V over an alge-
braically closed field F, and let λ1 , ..., λs be all eigenvalues of A, n1 , n2 , ..., ns be their multiplicities.
Then one has the generalized eigenspace decomposition:
s
M
V = Vλi where dim Vλi = ni
i=1
Proof. By the Jordan normal form of A in some basis e1 , e2 , ...en . Its matrix is of the following
form:
Jλ1
Jλ2
A=
. .
.
Jλn ,
where Jλi is an ni × ni matrix with λi on the diagonal, 0 or 1 in each entry just above the diagonal,
and 0 everywhere else.
Let Vλ1 = span{e1 , e2 , ..., en1 }, Vλ2 = span{en1 +1 , ..., en1 +n2 }, ..., so that Jλi acts on Vλi . i.e. Vλi
are A-invariant and A|Vλi = λi Ini + Ni , Ni nilpotent.
From the above discussion, we obtain the following decomposition of the operator A, called the
classical Jordan decomposition
A = As + An
where As is the operator which in the basis above is the diagonal part of A, and An is the rest
(An = A − As ). It has the following 3 properties
(iii) As An = An As .
Ls
(iii) holds since V = i=1 Vλi , AVλi ∈ Vλi , and As |Vλi = λi I. Hence As An = An As .
1
Definition 6.2. A decomposition of an operator A of the form A = As + An , for which these three
properities hold is called a Jordan decomposition of A. We have established its existence, provided
that dimV < +∞, F = F̄
Proposition 6.2. Jordan decomposition is unique under the same assumptions
Lemma 6.3. Let A and B be commuting operators on V ; i.e., AB = BA. Then
Proof. (a) is immediate from definition of generalized eigenspace. (b) follows from (a) since each Vλi
is B-invariant, As |Vλi = λi Ini , therefore A and As commute on each Vλi , therefore commute.
P roof of the proposition. Consider a Jordan decomposition A = A0s + A0n , and let A = As + An be
the classical Jordan decomposition. Take the difference, we get
As − A0s = An − A0n
But A0s commutes with A0n and itself, hence with A. Hence by taking B = A0s in lemma (b), we
conclude that A0s commutes with As and An . Therefore A0n = A − A0s also commutes with As
and An . So in (2) we we have difference of commutative operators on both sides. Hence LHS is
diagonalizable and RHS is nilpotent (by the binomial formula). But equality of a diagonalizable
operator to a nilpotent one is possible only if both are 0.
Question. Is it true in general that Jordan decomposition is unique?
Exercise 6.1. Show that any nonabelian 3-dimensional nilpotent Lie algebra is isomorphic to the
Heisenberg algebra H3 .
Proof. If g is nonabelian and 3-dimensional, then Z(g) must have dimension less than 3. By a
previous exercise (3.2), dim Z(g) 6= dim g − 1, so this dimension cannot be 2. A proposition from
lecture 4 states that if g is nonzero and nilpotent, Z(g) is nonzero. Hence Z(g) is 1-dimensional.
Now by exercise 3.3, the n-dimensional Lie algebras for which Z(g) has dimension two less than
g are Abn−2 ⊕ g2 and Abn−3 ⊕ H3 , where g2 is the 2-dimensional Lie algebra Fx + Fy defined by
[x, y] = y.
Since g is nilpotent, it cannot be Ab1 ⊕ g2 , because g2 is not nilpotent. Then the only remaining
possibility is g = Ab0 ⊕ H3 = H3 .
2
Theorem 6.4. π (gaα ) Vλa ⊆ Vλ+α
a
Lemma 6.5. Suppose U is a unital associative algebra over F, and let a, b ∈ U and λ, α ∈ F. Then
N
N
(ad a − αI)j b (a − λ)N −j .
X
N
(a − α − λ) b =
j
j=0
La−α−λ = La − αI − λI (1)
= ad a + Ra − αI − λI
La−α−λ = (ad a − α) + Ra−λ
Proof of Theorem 6.4. Applying the lemma to π(g), we have the following for all g ∈ g, and thus
for all g ∈ gaα . (Recall that a ∈ g is fixed.)
N
N
(ad π(a) − α)j π(g) (π(a) − λ)N −j
N
X
(π(a) − α − λ) π(g) =
j
j=0
Apply both sides of this to v ∈ Vλa with N > dim Vλa + dim gaα . By this choice of N , either
j > dim gaα or N − j > dim Vλa . If j > dim gαa , then (ad π(a) − α)j π(g) = 0 since g ∈ gλa .
Otherwise, N − j > dim Vλa , so (π(a) − λ)N −j v = 0 since v ∈ Vλa .
This makes every term in the sum on the right zero, so (π(a) − α − λ)N π(g)v = 0. Then π(g)v is
a . Since this holds for all
a generalized eigenvector of π(a) with eigenvalue α − λ, so π(g)v ∈ Vλ+α
g ∈ gaα and v ∈ Vλa , the claimed inclusion holds.
By analogy to the definition of a generalized eigenspace, we can define generalized weight spaces of
a Lie algebra g.
Definition 6.3. Let g be a Lie algebra with a representation π on a vector space on V , and let
λ ∈ g∗ be a linear functional on g. The generalized weight space of g in V attached to λ is
n o
Vλg = v ∈ V | (π(g) − λ(g)I)N v = 0 for some N depending on g, for all g ∈ g .
Under the right conditions, a nilpotent subalgebra h ⊆ g permits decomposing V as a direct sum
of the generalized weight spaces of h, each of which is a subrepresentation of πh . The following
theorem makes this precise.
3
Theorem 6.6. Let g be a finite-dimensional Lie algebra and π its representation on a finite-
dimensional vector space V , over an algebraically closed field F of characteristic 0. Let h be a
nilpotent subalgebra of g. Then the following equalities hold.
M h
V = Vλ (2)
λ∈h∗
π ghα Vλh ⊆ Vλ+α
h
(3)
Remark. In the case of the adjoint representation, we may express these as follows.
M
g= ghα (4)
α∈h∗
h i
ghα , ghβ ⊆ ghα+β (5)
To finish, we’ll prove the inclusion (3). Suppose α, λ ∈ h∗ , and suppose g ∈ ghα . Then g ∈ gaα(a) for
all a ∈ h. By Theorem 6.4, π(g)Vλ(a)a a
⊂ Vλ(a)+α(a) for all a ∈ h. Then
\ \
a a
v∈ Vλ(a) =⇒ π(g)v ∈ Vλ(a)+α(a) .
a∈h a∈h
\
Since a
Vλ(a) = Vλh by the definition of a generalized weight space, this establishes (3).
a∈h
4
∂
Proof. Suppose p acts as ∂x , q acts as multiplication by x, and c acts as the identity on F[x]/(x2 ).
Then:
0 1 a
p(a + bx) = b + 0x = 1 x
0 0 b
0 0 a
q(a + bx) = 0 + ax = 1 x
1 0 b
1 0 a
c(a + bx) = a + bx = 1 x
0 1 b
Then making a basis on F[x]/(x2 ) using 1 and x, we can write the matrix representing some
rp + sq + tc ∈ H3 as the following matrix.
t r
s t
In a field of characteristic 2, we can drop the ± sign. By passing to the algebraic closure if necessary,
we can assume the square root of rs always exists. Thus:
√
λ(rp + sq + tc) = t + rs
(To verify that λ is not linear, observe that by this formula, λ(p) = λ(q) = 0, but λ(p + q) = 1.)
Exercise 6.3. By the example of the adjoint representation of a nonabelian solvable Lie algebra,
show that the generalized weight space decomposition fails if the Lie algebra is solvable but not
nilpotent.
Proof. Consider the Lie algebra g2 = Fx + Fy, with the bracket operation defined by [x, y] = y.
It’s apparent by induction that gk2 = [g2 , gk−1
2 ] = Fy (for k ≥ 2), so g2 is not nilpotent. However,
(1) (2)
then g2 = Fy, which is 1-dimensional, so g2 = 0, and thus g2 is solvable.
Taking x and y as the basis elements of g2 , the adjoint representation takes x and y to the following
matrices.
0 0 0 0
x 7→ y 7→
0 1 −1 0
λ(λ − 1) = 0 λ2 = 0
λ = 0 or 1 λ=0
5
The corresponding generalized eigenvectors can be found by lucky guessing. Specifically, ad x has
x with eigenvalue 0 and y with eigenvalue 1, while ad y has all of g2 with eigenvalue 0.
So we can get weight spaces V0 = span{x} and Vx∗ = span{y}, corresponding to the zero linear
functional and the linear functional defined by x 7→ 1. The vector space decomposes into the direct
sum of these weight spaces, but the representation does not! Specifically, V0 is not closed under
the action of y.
Exercise 6.4. Take g = gln (F) and h = {diagonal matrices}. Find the generalized weight space
decomposition in both the tautological and the adjoint representations, and check the inclusions
(3) and (5) in Theorem 6.6.
Proof. Suppose g = gln (F) and h ⊂ g consists of diagonal matrices. Then the generalized eigenvec-
tors of h ∈ h are actual eigenvectors, so every standard basis element of Fn is an eigenvector. Also,
given any linear combination aei +bej of more than one basis element, there is some diagonal matrix
that takes ei to ei and ej to zero, so these linear combinations are not generalized eigenvectors of
everything in h. Thus the only candidates for generalized weight spaces are the n axes, each of
which is the span of a single standard basis element of Fn .
For the axis Vi spanned by ei , the linear functional on h that takes h to the component hi,i is a
weight making Vi a weight space. Thus in the tautological representation, Fn decomposes as a
direct sum of n copies of F.
To do the same with the adjoint representation, suppose the diagonal entries of h ∈ h are hi . Then
for a ∈ g, we have:
This shows that ad h is diagonalizable, which again implies that its generalized eigenvectors are
actual eigenvectors, and so its generalized weight spaces are actually weight spaces.
Possible pairs of eigenvalues are
1. hi − hj vs. hi − hk ,
2. hi − hj vs. hk − h` ,
3. hi − hj vs. hj − hi , and
4. hi − hi vs. hj − hj .
By appropriate choice of h, we can always make distinct eigenvalues in (1) and (2), so the basis
elements ei,j of g satisfying i < j lie in distinct eigenspaces for some h, and thus they lie in distinct
candidate weight spaces. This weight space can be achieved with the linear functional λi,j taking
h to hi − hj .
Since for the theorem we assume the characteristic of F is not 2, the eigenvalues in (3) will be
distinct, so we’ll also have λi,j with i > j. Finally, both eigenvalues in (4) are always zero, so the
zero linear functional has h as its weight space.
6
Combining all of this, the generalized weight space decomposition of g in the adjoint representation
is h plus some 1-dimensional weight spaces:
M
g=h⊕ span{ei,j }
i6=j
To check the assertion of the theorem from class, first we verify for the tautological representation
that:
π ghα Vλh ⊆ Vλ+αh
Each Vλh is the span of some basis element ei , with λ corresponding to the map h 7→ hi , so we really
only need to check that, for some appropriate j:
π ghα ei ∝ ej
In the case α = 0 we should get j = i, and we do; the space gh0 consists of all diagonal matrices, so
they act on ei by scaling.
In the case αh = hk − h` , we then have ghα = ek,` , and α + λ = hk − h` + hi . We should expect zero
if i 6= `, since we only have nonzero weight spaces for λ of the form h 7→ hsomething ; and indeed this
is the case, since if i 6= ` then ek,` ei = 0. Furthermore, if i = ` then we should get the span of ek ,
which is the weight space corresponding to h 7→ hk . We verify this by observing that ek,i ei = ek .
So in fact we have equality:
h
π ghα Vλh = Vλ+α
The second assertion in (b) of the theorem is essentially the same statement for the adjoint repre-
sentation. First, if α = β = 0, then both ghα and ghβ are equal to h. Since h consists of diagonal
matrices, it’s commutative, so the bracket is zero and thus is contained in any weight space we like.
If α = 0 and β = {h 7→ hi − hj }, then we end up with [h, span{ei,j }]. Observe that:
Finally, if α maps h to hi −hj and β maps h to hk −h` , we have essentially [ei,j , ek,` ]. We need either
j = k or i = ` for α + β to be a weight, and we indeed see that if neither holds, then ei,j ek,` = 0.
Otherwise, by relabeling α and β, we can assume without loss of generality that j = k. This gives
us ei,` if i 6= ` and 0 if i = `, so either way it’s in the weight space of h 7→ hi − h` .
7
18.745 Introduction to Lie Algebras September 30, 2010
Definition 7.1. A topological space is a set 𝑋 with a collection of subsets 𝐹 (the closed sets)
satisfying the following axioms:
1) 𝑋 ∈ 𝐹 and ∅ ∈ 𝐹
4) (weak separation axiom) For any two points 𝑥, 𝑦 ∈ 𝑋 there exists an 𝑆 ∈ 𝐹 such that 𝑥 ∈ 𝑆
but 𝑦 ∕∈ 𝑆
A subset 𝑂 ⊂ 𝑋 which is the complement in 𝑋 of a set in 𝐹 is called open. The axioms for a
topological space can also be phrased in terms of open sets.
Definition 7.2. Let X = 𝔽𝑛 where 𝔽 is a field. We define the Zariski topology on 𝑋 as follows. A
set in 𝑋 is closed if and only if if is the set of common zeroes of a collection (possibly infinite) of
polynomials 𝑃𝛼 (𝑥) on 𝔽𝑛
Proof. We check the axioms. For 1), let 𝑝(𝑥) = 1, so that 𝕍(𝑝) = ∅, so ∅ ∈ 𝐹 . If 𝑝(𝑥) = 0, then
𝕍(𝑝) = 𝔽𝑛 , so 𝔽𝑛 ∈ 𝐹 . For 2), first take two closed sets 𝕍(𝑝𝛼 ) with 𝛼 ∈ 𝐼 and 𝕍(𝑞𝛽 ) with 𝛽 ∈ 𝐽 ,
and consider the family of polynomials 𝑓𝛼𝛽 = {𝑝𝛼 𝑞𝛽 ∣ 𝛼 ∈ 𝐼, 𝛽 ∈ 𝐽}. Then 𝕍(𝑓𝛼𝛽 ) = 𝕍(𝑝𝛼 ) ∪ 𝕍(𝑞𝛽 ).
Therefore 𝕍(𝑝𝛼 ) ∪ 𝕍(𝑞𝛽 ) is closed. This can be iterated for any n-fold union. For 3) closure
under arbitrary intersections is obvious. For 4), suppose 𝑥 ∕= 𝑦 ∈ 𝔽𝑛 and 𝑥 = (𝑥1 , 𝑥2 , ..., 𝑥𝑛 ). Set
𝑋 = 𝕍(𝑝1 , 𝑝2 , ..., 𝑝𝑛 ) where 𝑝𝑖 (𝑧) = 𝑧𝑖 − 𝑥𝑖 . So 𝑥 ∈ 𝑋 but y is not.
Example 7.1. If 𝑋 = 𝔽 the closed sets in the Zariski topology are precisely the ∅, 𝔽 and finite
subsets of 𝔽.
3) If a polynomial 𝑝(𝑥) vanishes on a nonempty Zariski open set then it is identically zero.
1
Example 7.2. The condition that 𝔽 be infinite is crucial. In 3), for instance, the polynomial
𝑝(𝑥) = 𝑥2 + 𝑥 on 𝔽2 vanishes at 1 and 0 but is not the zero polynomial.
Proof. For 1), We induct on 𝑛. When 𝑛 = 1, since any nonzero polynomial has at most deg
𝑃 roots, the complement of this set is infinite (since 𝔽 is). If 𝑛 ≥ 1 then any nonzero polyno-
mial 𝑝(𝑥1 , 𝑥2 , ..., 𝑥𝑛 ) can be written as a polynomial in one variable with coefficients polynomi-
als in the other variables. So we can write 𝑝(𝑥) = 𝑝𝑑 (𝑥2 , 𝑥3 , ..., 𝑥𝑛 )𝑥𝑑1 + 𝑝𝑑−1 (𝑥2 , 𝑥3 , ...𝑥𝑛 )𝑥𝑑−1 1 +
... + 𝑝0 (𝑥2 , 𝑥3 , ..., 𝑥𝑛 ) where 𝑝𝑑 (𝑥2 , 𝑥3 , ...𝑥𝑛 ) is a nonzero polynomial. By the inductive hypothesis,
we can find 𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 ∈ 𝔽 such that 𝑝𝑑 (𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 ) ∕= 0 Now fixing these values, 𝑝(𝑥) =
𝑝𝑑 (𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 )𝑥𝑑1 + 𝑝𝑑−1 (𝑥𝑜2 , 𝑥𝑜3 , ...𝑥𝑜𝑛 )𝑥𝑑−1
1 + ... + 𝑝0 (𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 ) we are back in the 𝑛 = 1
case. We can find an infinite number of values 𝑥𝑜1 such that 𝑝(𝑥) = 𝑝𝑑 (𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 )(𝑥𝑜1 )𝑑 +
𝑝𝑑−1 (𝑥𝑜2 , 𝑥𝑜3 , ...𝑥𝑜𝑛 )(𝑥𝑜1 )𝑑−1 + ... + 𝑝0 (𝑥𝑜2 , 𝑥𝑜3 , ..., 𝑥𝑜𝑛 ) ∕= 0. Thus we have found an infinite number
of points in 𝔽𝑛 where 𝑝 does not vanish. To prove the second claim of 1), just observe if 𝑆 a
collection of polynomials containing the nonzero polynomial 𝑝(𝑥), we have 𝕍(𝑆)𝐶 ⊃ 𝕍(𝑝)𝐶 so by
the first claim, we have that 𝕍(𝑆)𝐶 contains an infinite set.
2) Consider 𝕍(𝑆1 ) and 𝕍(𝑆2 ) where 𝑆1 and 𝑆2 are nonzero sets of polynomials. It suffices to prove
that 𝕍(𝑝1 )𝐶 ∩ 𝕍(𝑝2 )𝐶 is nonempty for any fixed 𝑝1 ∈ 𝑆1 and 𝑝2 ∈ 𝑆2 nonzero polynomials. But
observe that 𝕍(𝑝1 𝑝2 ) = 𝕍(𝑝1 ) ∪ 𝕍(𝑝2 ), so 𝕍(𝑝1 𝑝2 )𝐶 = 𝕍(𝑝1 )𝐶 ∩ 𝕍(𝑝2 )𝐶 but by 1) we know that the
term on the left is infinite, hence nonempty.
3) If 𝑝 vanishes on 𝕍(𝑆)𝐶 for S containing a nonzero polynomial 𝑞 then 𝑝 vanishes on 𝕍(𝑞)𝐶 . If 𝑝
were nonzero, then by 2), 𝕍(𝑞)𝐶 ∩ 𝕍(𝑝)𝐶 is nonempty. This is a contradiction, so 𝑝 is identically
zero.
Let 𝔤 be a finite dimensional Lie algebra of dimension 𝑑 and consider the characteristic polynomial
of an endomorphism ad 𝑎 for some 𝑎 ∈ 𝔤: det𝔤 (ad 𝑎 − 𝜆) = (−𝜆)𝑑 + 𝑐𝑑−1 (−𝜆)𝑑−1 + ... + det(ad 𝑎).
This is a polynomial of degree 𝑑. Because (ad 𝑎)(𝑎) = [𝑎, 𝑎] = 0, we know det(ad 𝑎) = 0 and
the constant term in the polynomial vanishes.
Exercise 7.2. Show that 𝑐𝑗 is a homogeneous polynomial on 𝔤 of degree 𝑑 − 𝑗.
Proof. By ”polynomial
∑ in 𝔤” we mean that if we fix a basis 𝑋1 , 𝑋2 , ..., 𝑋𝑑 of 𝔤 and write a general
element 𝑎 = 𝑖 𝑎𝑖 𝑋𝑖 , then 𝑐𝑘 (𝑎) is a polynomial in the 𝑎𝑖 . Observe that ad 𝑎 is a 𝑑 by 𝑑 matrix
whose entries 𝑏𝑖𝑗 are linear combinations of the 𝑎𝑖 . Now consider det𝔤 (ad 𝑎 − 𝜆). This is a
polynomial in the entries of the matrix ad 𝑎 − 𝜆. A general element in the expansion of this
determinant is 𝑏∗∗ 𝑏∗∗ ...(𝑏𝑖𝑖 − 𝜆)...(𝑏𝑗𝑗 − 𝜆). In order to get a term with 𝜆𝑘 we have to chose 𝑘 𝜆’s
and 𝑑 − 𝑘 of the 𝑏∗∗ in the expansion of this product. Therefore the coefficient of the polynomial
𝜆𝑘 is homogeneous of degree 𝑑 − 𝑘 in the 𝑏∗∗ , so it is also homogeneous of degree 𝑑 − 𝑘 in the 𝑎𝑖
since the 𝑏𝑖𝑗 are linear in the 𝑎𝑖 .
Definition 7.3. The smallest positive integer such that 𝑐𝑟 (𝑎) is not the zero polynomial on 𝔤 is
called the rank of 𝔤. Note 1 ≤ 𝑟 ≤ 𝑛. An element 𝑎 ∈ 𝔤 is called regular if 𝑐𝑟 (𝑎) ∕= 0. The nonzero
polynomial 𝑐𝑟 (𝑎) of degree 𝑑 − 𝑟 is called the discriminant of 𝔤.
Proposition 7.2. Let 𝔤 be a Lie algebra of dimension 𝑑, rank 𝑟 over 𝔽
2
2) If 𝔤 is nilpotent, the set of regular elements is 𝔤.
Proof. 1) 𝑟 = 𝑑 means that det(ad 𝑎 − 𝜆) = (−𝜆)𝑑 for all 𝑎. This is true if and only if all the
eigenvalues of ad 𝑎 are 0, which happens if and only if ad 𝑎 is nilpotent for all 𝑎. By Engel’s
theorem, this is equivalent to 𝔤 being nilpotent.
2) If 𝔤 is nilpotent, we know det(ad 𝑎 − 𝜆) = (−𝜆)𝑑 and 𝑐𝑟 = 𝑐𝑑 = 1, which does not vanish
anywhere, so all elements are regular.
3) If 𝔤 is not nilpotent, the set of regular elements is by definition 𝕍(𝑐𝑟 )𝐶 , which is the complement
of a hypersurface, defined by a homogeneous polynomial of degree 𝑑 − 𝑟 > 0 by Ex 7.2. By part 1
of the previous theorem the complement of a hypersurface is an infinite set if 𝔽 is.
Example 7.3. For 𝔤 = 𝔤𝔩𝑛 (𝔽), the rank is 𝑛 and we will find its discriminant. Also, the set of
regular elements consists of all matrices with distinct eigenvalues.
Exercise 7.3. 1) Show that the Jordan decomposition of ad 𝑎 is given by ad 𝑎 = (ad 𝑎𝑠 )+(ad 𝑎𝑛 )
in 𝔤𝔩𝑛
2) If 𝜆1 , ..., 𝜆𝑛 are eigenvalues of 𝑎𝑠 then 𝜆𝑖 − 𝜆𝑗 are eigenvalues of ad 𝑎𝑠
3) ad 𝑎𝑠 has the same eigenvalues as ad 𝑎.
Proof. For 2), If 𝜆1 , ..., 𝜆𝑛 are eigenvalues of 𝑎𝑠 observe that if 𝐸𝑖𝑗 is the matrix with 1 in the
𝑖𝑗 slot and zero elsewhere, then [𝑎𝑠 , 𝐸𝑖𝑗 ] = (𝑎𝑠𝑖𝑖 − 𝑎𝑠𝑗𝑗 )𝐸𝑖𝑗 so ad 𝑎𝑠 is diagonalizable since the
𝐸𝑖𝑗 form a basis of 𝔤𝔩𝑛 (𝔽). The eigenvalues are thus 𝜆𝑖 − 𝜆𝑗 . As for 1), we already showed
that ad 𝑎𝑠 is diagonalizable, also ad 𝑎𝑛 is nilpotent because 𝑎𝑛 is. Finally, ad 𝑎𝑠 and ad 𝑎𝑛
commute: [ad 𝑎𝑠 , ad 𝑎𝑛 ]𝑣 = [𝑎𝑠 , [𝑎𝑛 , 𝑣]] − [𝑎𝑛 , [𝑎𝑠 , 𝑣]] = −[[𝑎𝑠 , 𝑎𝑛 ], 𝑣] = 0 (since [𝑎𝑠 , 𝑎𝑛 ] = 0). Thus
by uniqueness of Jordan form, ad 𝑎 = ad 𝑎𝑠 + ad 𝑎𝑛 . As for 3), we know from Jordan canonical
form that the semisimple part of an operator and the operator itself have the same eigenvalues –
and since ad 𝑎𝑠 is the semisimple part of the operator ad 𝑎, the claim follows.
Exercise 7.4. Deduce the statement∏ of example 1.3 by showing that the rank of 𝔤𝔩𝑛 is 𝑛 and that
the discriminant is given by 𝑐𝑛 (𝑎) = 𝑖∕=𝑗 (𝜆𝑖 − 𝜆𝑗 ). Also compute 𝑐2 (𝑎) for 𝔤𝔩2 (𝔽) in terms of the
matrix coefficients of 𝑎.
∏
Proof. We must compute det(ad 𝑎 − 𝜆). This is equal ∏ to det(ad 𝑎 𝑠 − 𝜆) = ∏ 𝑖,𝑗 (𝜆𝑖 − 𝜆𝑗 − 𝜆). Note
when 𝑖 = 𝑗, we can pull out the (−𝜆), so we have 𝑖,𝑗 (𝜆𝑖 − 𝜆𝑗 −∏ 𝜆) = (−𝜆) 𝑛
𝑖∕=𝑗 (𝜆𝑖 − 𝜆𝑗 − 𝜆). Thus
we see that the rank of 𝔤𝔩𝑛 (𝔽𝑛 ) = 𝑛, and the discriminant is 𝑖∕=𝑗 (𝜆𝑖 − 𝜆𝑗 ). An element in 𝔤𝔩𝑛 is
therefore regular
∏ if and only if all its eigenvalues are distinct. Finally we compute the discriminant
𝑐2 for 𝔤𝔩2 (𝔽): 2
( ) 𝑖∕=𝑗 (𝜆𝑖 − 𝜆𝑗 ) = (𝜆1 − 𝜆2 )(𝜆2 − 𝜆1 ) = −(𝜆1 + 𝜆2 ) + 4𝜆1 𝜆2 . If we write a matrix
𝑎 𝑏
, then the discriminant is −(𝑎 + 𝑑)2 + 4(𝑎𝑑 − 𝑏𝑐). This a clearly a homogeneous polynomial
𝑐 𝑑
of degree 𝑛2 − 𝑛 = 22 − 2 = 2.
3
18.745 Introduction to Lie Algebras October 5, 2010
The fact that Ng (h) is a subalgebra follows directly from the Jacobi identity. Also note that
h ⊂ Ng (h) and the normalizer of h is the maximal subalgebra containing h as an ideal.
Lemma 8.1. Let g be a nilpotent Lie algebra and h ⊂ g a subalgebra such that h 6= g. Then,
h ( Ng (h).
Definition 8.2. A Cartan subalgebra of a Lie algebra g is a subalgebra h, satisfying the following
two conditions:
i) h is a nilpotent Lie algebra
ii) Ng (h) = h
Corollary 8.2. Any Cartan subalgebra of g is a maximal nilpotent subalgebra
Proof. This follows directly from Lemma 1 and the definition of Cartan subalgebras.
Exercise 8.1. Let g = gln (F) with char(F) 6= 2. Let h = {nn + FIn }, where nn is the subalgebra of
strictly upper triangular matrices. Then this is a maximal nilpotent subalgebra but not a Cartan
subalgebra.
Solution. First, we show that nn + FIn is not a Cartan subalgebra of g. h is not Cartan since, as
was shown earlier, [bn , nn ] ⊂ nn and thus bn ⊂ Ng (nn ) ⊂ Ng (nn + FIn ).
Now we show that nn + FI Ln is a maximal nilpotent subalgebra. Note that since In commutes with
everything, then h = nn FIn . Hence, it is nilpotent. Now, it suffices to show that h is maximal
in gln . Suppose there exists some nilpotent 0
P subalgebra nn + FIn ( h . First note that0 in fact we
have bn = Ngln (nn + FIn ), since if b = i,j cij Eij ∈ Ng (nn + FIn ) with ci0 j 0 6= 0 for i > j 0 , then
0 0 th and the
P P
Ej 0 i0 ∈ nn ⊂ nn + FIn and [b, Ej 0 i0 ] = i cij 0 Eii0 − j ci0 j Ej 0 j . Note that the (i , i )
(j 0 , j 0 )th entries are ci0 j 0 and −ci0 j 0 respectively; both of which are nonzero. Thus, [b, Ej 0 i0 ]nn + FIn ,
unless char(F) = 2. Now, by the Lemma above, we must have that nn + FIn ( h ∩ Ngln (nn + FIn ).
Thus, h0 contains some element of bn \nn + FIn ; but all elements of bn \nn + FIn have at least two
distinct eigenvalues, and thus are not ad-nilpotent. Thus we find a contradiction to Engel’s theorem
and our assumption must be wrong. We find there is no proper set containing nn + FIn ; which
implies that h is a maximal nilpotent subalgebra.
1
Proposition 8.3. Let g ⊂ gln (F) be a subalgebra containing a diagonal matrix a = diag(a1 , ..., an )
with distinct ai , and let h be the subspace of all diagonal matrices in g. Then h is a Cartan
subalgebra.
Proof. We have to prove that h satisfies the two conditions necessary to be a Cartan subalgebra.
i) We know Pthat h is abelian; and thus, it is a nilpotent Lie algebra.
ii)Let b = ni,j=1 bij eij ∈ g such
P that [b,Ph] ⊂ h (i.e. bP
∈ Ng (h)). In particular, we have that [a, b] ∈ h
for all a ∈ h. But [a, b] = [ k ak ekk , i,j eij eij ] = ij (ai − aj )bij eij , which will be non-diagonal
only if bij 6= 0 for some i 6= j. Thus, we can conclude that b ∈ h, and therefore Ng (h) = h.
Theorem 8.4. (E.Cartan) Let g be a finite dimensional Lie algebra over an algebraically closed
a
L
field F. Let a ∈ g be a regular element (which exists since F is infinite), and let g = λ∈F gλ be the
generalized eigenspace decomposition of g with respect to ada. Then h = ga0 is a Cartan subalgebra.
Proof. The proof for this theorem uses the fact that Zariski Topology is highly non-Hausdorff,
namely any two non-empty Zariski open sets have a non-empty intersection. We will also recall the
[gaλ , gaµ ] ⊂ gaλ+µ and inL
fact that L particular, if λ = 0 then [h, gaµ ] ⊂ gaµ .
Let V = λ6=0 gaλ . Then g = h V and [h, V ] ⊂ V .
Consider the following two subsets of h:
U = {h ∈ h such that adh|h is not a nilpotent operator}
R = {h ∈ h such that adh|V is a non-singular operator }
Both U and R are Zariski open subsets of h. Next we note that a ∈ R since all zero eigenvalues of
ada lie in h, hence R is non-empty.
Now, we shall prove by contradiction that h is a nilpotent subalgebra. Suppose the contrary is
true. Then, by Engel’s Theorem there exists h T ∈ h such that adh|h is not nilpotent.
T But in this
case h ∈ U ; and hence, U 6= ∅. Therefore U R 6= ∅. We now take b ∈ U R. Then adh|h is
not nilpotent and adb|V is invertible. Hence gb0 ( h, which contradicts the fact that a is a regular
element. Thus, this contradicts the assumption made; and we find that h is a nilpotent Lie algebra.
Finally, we need to proof that Ng (h) = h. Now, if b ∈ Ng (h), so that [b, h] ⊂ h, then we have that,
in particular, [b, a] ∈ h. But since a ∈ h and h is a nilpotent Lie algera, then ada|h is a nilpotent
operator. In particular, 0 = (ada)N ((ada)b) = (ada)N +1 (b). Hence b ∈ ga0 = h, which completes
the proof of the theorem.
Remark The dimension of the Cartan subalgebra constructed in Cartan’s Theorem, Theorem 4,
equals the rank of g.
Proposition 8.5. Let g be a finite dimensional Lie algebra over an algebraically closed field F of
characteristic zero and let h ⊂ g be a Cartan subalgebra. Consider theL generalized weight space
decomposition (called root space decomposition) with respect to h: g = λ∈h∗ gλ . Then g0 = h.
Proof. Since by Engel’s Theorem adh|h is nilpotent for all h ∈ h, it follows that h ⊆ g0 . But by
definition of g, for all elements h ∈ h, adh|g0 is a nilpotent operator. Hence adh|g0 /h is a nilpotent
operator for all h ∈ h. Therefore, by Engle’s Theorem, there exists a non-zero barb ∈ g0 /h which
is annihilated by all adh|g0 /h . Taking a pre-image b ∈ g of b̄ , this means that [b, h] ⊂ h (i.e.
h 6= Ng (h)), which contradicts the fact that h is a Cartan subalgebra.
2
LL
Remark g = h ( λ∈h∗ ,λ6=0 gλ ) by the latter Proposition.
Next we will apply the last couple of theorems, lemmas and propositions in order to classify all
3-dimensional Lie algebras g over an algebraically closed field F of characteristic zero.
We know that the rank(g) = 3, 2 or 1; and rank(g) = 3 if and only if g is nilpotent.
Rank(g) = 3:
We know by exercise 6.1, that any 3-dimensional nilpotent Lie algebra is either abelian or H3 . So
any 3-dimensional Lie algebra of rank(g) = 3 is either abelian or H3 .
Rank(g) = 2:
In this case dim(h) = 2. Since h must be a nilpotent Lie algebra, we can conclude that h is abelian
(otherwise h would be a 2-dimensional
L solvable Lie algebra, which is not nilpotent). Hence the
root space decomposition is g = h Fb, where [h, b] ⊂ Fb. Since h = Ng (h) then [h, b] 6= 0. Hence
there exist a ∈ h such that [a, b] = b. Also since h is 2-dimensional, then there exists c ∈ h such
that [c, b] = 0. We also know that [a, c] = 0. Thus, we can conclude that the only 3-dimensional
Lie algebra of rank 2 is a direct sum
L of a 2-dimensional non-abelian algebra and one dimensional
central subalgebra: g = (Fa + Fb) Fc.
Rank(g) = 1:
Exercise 8.2. Any 3-dimensional Lie algebra g of rank 1 is isomorphic to one of the following Lie
algebras with basis h, a, b:
i) [h, a] = a, [h, b] = a + b, [a, b] = 0;
ii) [h, a] = a, [h, b] = λb, where λ ∈ F/{0} [a, b] = 0;
iii) [h, a] = a, [h, b] = −b, [a, b] = h;
L
Solution. Let h = Fh,where h ∈ g, be a Cartan subalgebra; we have: g = h V , where [h, V ] ⊂ V ,
dim(V ) = 2, and adh is non-singular on V . We may assume one of the eigenvalues of adh|V is 1,
if we scale h accordingly.
1 1
First, suppose adh|V is not semisimple. Thus, adh|V = in some basis {a, b}. Thus,
0 1
[h, a] = a and [h, b] = a + b. Also, by using Jacobian Identity, we have [h, [a, b]] = [[h, a], b] +
[a, [h, b]] = [a, b] + [a, a + b] = 2[a, b]. But we know that the value 2 cannot be an eigenvalue of adh,
thus [a, b] = 0. Thus, this Lie alebra corresponds to (i).
1 0
Now, assume adh|V is semisimple. Thus, we have that adh|V = in some basis {a, b}. We
0 λ
then have that [h, a] = a and [h, b] = λb. By following the same procedure as above, we note that
[h, [a, b]] = [[h, a], b] + [a, [h, b]] = [a, b] + [a, λb] = (1 + λ)[a, b]. Thus, we must have that either
[a, b] = 0, which corresponds to (ii), or (1 + λ) is an eigenvalue of adh|V , case (iii). Note that for
the case of [a, b] = 0, then λ is arbitrary and uniquely defined by g up to inverting it by swapping
a and b and scaling h accordingly. In the case were (1 + λ) is an eigenvalue of adh|V , we must have
that 1 + λ = 0 and that [a, b] be a multiple of h. Thus λ = −1 and, scaling a accordingly, we may
assume that [a, b] = h. Thus, we get option (iii) with the latter case.
Exercise 8.3. Show that all Lie algebras in exercise 8.2 are non-isomorphic . Those from (i) and
(ii) are solvable, and the one from (iii) is isomorphic to sl2 (F), which is not solvable.
3
0 0 0
[h, a0 ] = 2a0 , [h, b0 ] = −2b0 and [a0 , b0 ] = h. Now if we scale and set h = h2 , a = √a 2 and b = √b 2 ;
then we get the Lie algebra of case (iii). Thus, Lie algebra (iii) is isomorphic to sl2 (F), which is
not solvable. In contrast, it is clear that the algebras (i) and (ii) are solvable by construction.
4
18.745 Introduction to Lie Algebras October 7, 2010
In the last lecture we defined Cartan subalgebras and gave a construction using regular elements.
In this lecture we will show that this construction is essentially unique by proving Chevalley’s
Theorem on conjugacy of Cartan subalgebras.
To state the theorem, we need the notion of the exponential of a nilpotent operator.
Definition 9.1. Let A be a nilpotent operator on a vector space V over a field F of characteristic
0. Define eA = I + A + 2!1 A2 + 3!1 A3 + · · · . (As A is nilpotent, this is a finite sum.)
Exercise 9.1. If A and B are commuting nilpotent operators, then we know that A + B is also
nilpotent, so eA+B , eA , and eB are all defined. Show that eA+B = eA eB . In particular, deduce that
eA e−A = I, so eA is always an invertible operator.
Every term above with k > N or ` > N vanishes, as either Ak = 0 or B ` = 0. Conversely, every
term with k ≤ N and ` ≤ N is present in the sum. Thus
N
! N !
X 1 X A k X B`
eA+B = Ak B ` = = eA eB ,
k!`! k! `!
k,`≤N k=0 `=0
Exercise 9.2. Let g be an arbitrary (not necessarily Lie) algebra over a field F of characteristic 0,
and let D be a nilpotent derivation of g. Show that eD is an automorphism of g.
Solution: We already know from Exercise 9.1 that eD is a linear isomorphism; the content of this
exercise is that it also respects multiplication in g. Let µ : g⊗F g → g denote the multiplication map.
Define linear transformations D1 and D2 on g⊗F g by D1 (a⊗b) = D(a)⊗b and D2 (a⊗b) = a⊗D(b).
It is clear that D1 and D2 commute, that they are both nilpotent, and that eD1 (a ⊗ b) = eD (a) ⊗ b
and eD2 (a ⊗ b) = a ⊗ eD (b). Next, because D is a derivation, notice that
eD (a · b) = (eD µ)(a ⊗ b) = (µeD1 +D2 )(a ⊗ b) = (µeD1 eD2 )(a ⊗ b) = eD (a) · eD (b),
as desired.
1
We are now ready to state the main result.
Theorem 9.1 (Chevalley). Let g be a finite-dimensional Lie algebra of an algebraically closed
field F of characteristic 0. Denote by G the subgroup of the group of automorphisms of g which is
generated by automorphisms of the form ead a for a ∈ g such that ad a is nilpotent. Then any two
Cartan subalgebras h1 and h2 are conjugate by G, i.e. there exists σ ∈ G such that σ(h1 ) = h2 .
Remark 9.1. The group G in Chevalley’s Theorem is almost (but not quite) the Lie group asso-
ciated to the Lie algebra g.
Before proving Chevalley’s Theorem, we give a corollary that addresses the question with which
we opened the lecture.
Corollary 9.2. Let F be an algebraically closed field of characteristic 0 and let g be a finite-
dimensional Lie algebra over F. Then any Cartan subalgebra h in g is of the form ga0 for some regular
element a ∈ g, and in particular dim h = rank g. Also, all such subalgebras ga0 are isomorphic.
Proof. Fix a regular element a ∈ g. By Chevalley’s Theorem, any Cartan subalgebra h is conjugate
to ga0 , say h = σ(ga0 ). Hence dim h = dim ga0 = rank g. But because σ is an algebra automorphism,
σ(a)
it is easy to check that σ(ga0 ) = g0 . The dimensionality of this tells us that σ(a) ∈ g is a regular
element. Finally, the last claim is immediate as conjugate subalgebras are isomorphic.
To prepare for the proof of Chevalley’s Theorem, we first prove two lemmas.
Lemma 9.3. Let h ⊆ g be a Cartan subalgebra and suppose there is a regular element a ∈ g which
is in h. Then h = ga0 .
Proof. Being a Cartan subalgebra, h is nilpotent. Thus ad a|h is nilpotent, and so h ⊆ ga0 . Now
we know from the last lecture that ga0 is a Cartan subalgebra, so in particular it is nilpotent. We
also know from the last lecture that h, being a Cartan subalgebra, must be a maximal nilpotent
subalgebra. Hence h ⊆ ga0 implies h = ga0 , as desired.
2
4. For each i = 1, 2, . . . , m, take a polynomial equation satisfied by xi over F(f1 , . . . , fm ), clear
denominators to get a polynomial over F[f1 , . . . , fm ], and let pi (f1 , . . . , fm ) be the leading
coefficient of this polynomial. Show that the set of points {y ∈ Fm : pi (y) 6= 0, i = 1, 2, . . . , m}
satisfies Lemma 9.4.
Solution: Step 1 is completely standard. For step 2, suppose for contradiction that the fi ’s are
algebraically dependent, and let F (y1 , . . . , ym ) be a nonzero polynomial of minimal degree such
that F (f1 , . . . , fm ) = 0 identically. For any j = 1, 2, . . . , m, apply the partial derivative ∂/∂xj to
the identity F (f1 , . . . , fm ) = 0. By the chain rule, this gives the equation
m
X ∂F ∂fi
· = 0.
∂yi (f1 ,...,fm ) ∂xj
i=1
m m
∂fi ∂F
Let J(x) be the Jacobian matrix ∂xj x and let ~v (y) be the row vector ∂yi (y ,...,y ) .
i,j=1 1 m i=1
Then the above equations yield ~v (f (x))J(x) = 0.
Let L be the locus of points x such that det(J(x)) 6= 0 and let L0 be the locus of points x such
that ~v (f (x)) 6= 0. At any point of L0 , the matrix J has a nonzero null vector, so it is singular: thus
L ∩ L0 = ∅. Both L and L0 are Zariski open subsets of Fm , and L is nonempty by assumption.
As the base field F is infinite, the only way for the nonempty open set L not to intersect the open
set L0 is to have L0 = ∅. But this means that the polynomials in ~v , i.e. the partial derivatives
∂F/∂yi , are polynomials in y1 , · · · , yn which vanish identically upon substitution of the fi ’s. As
deg(∂F/∂yi ) < deg F , our minimality assumption on deg F then implies ∂F/∂yi = 0. Over a field
of characteristic zero, a polynomial whose partial derivatives all vanish identically must be constant.
But F is not constant; contradiction.
For step 3, note that both F(x1 , . . . , xm ) and F(y1 , . . . , ym ) have transcendence degree m over F,
and an extension of fields with the same transcendence degree must be algebraic. Hence each xi is
algebraic over F(y1 , . . . , ym ), and in particular they generate a finite extension.
For step 4, note that the stated set is certainly a nonempty Zariski open set, as it is the nonvanishing
locus of the nonzero polynomial p1 · · · pm . We need to show that it is in the image of f , so choose
any y such that p1 (y), . . . , pm (y) 6= 0. Because the elements fi are algebraically independent, we
can define an evaluation homomorphism
ey : F[f1 , . . . , fm ] → F
mapping each fi to yi . By our assumption on y, ey maps each pi to a nonzero field element. Thus
ey extends uniquely to a map on the localization, êy : F[f1 , . . . , fm , p−1 −1
1 , . . . , pm ] → F. For each i,
it follows from the definition of pi that xi is integral over the ring F[f1 , . . . , fm , p−1 −1
1 , . . . , pm ]. We
now use the following fact from commutative algebra:
3
Example 9.1. Consider the map f : R → R given by f (x) = x2 . The image is f (R) = R≥0 ,
which does not contain a nonempty Zariski open set (because nonempty Zariski open sets in one
dimension are cofinite). Thus the algebraic closure assumption in Lemma 9.4 is necessary. (The
characteristic assumption is not necessary, though.)
Example 9.2. It is also important in Lemma 9.4 that the map f be algebraic. There is a well-
known example of a smooth (but not algebraic) map from R to the torus which does not include
any nonempty open set in its image.
Proof of Chevalley’s Theorem: Let h be any Cartan subalgebra of g. As h is nilpotent, we can define
a corresponding generalized root space decomposition g = ⊕α∈h∗ ghα . For later use, we observe that
gh0 = h by a proposition from last lecture.
We claim that for any α 6= 0 and any x ∈ ghα , ad x is nilpotent. To prove this, recall that the
root space decomposition satisfies [ghβ , ghγ ] ⊆ ghβ+γ for all β, γ. Thus for any β and any N ≥ 1,
(ad x)N ghβ ⊆ ghβ+N α . As α 6= 0 and char F = 0, {β + N α : N ≥ 1} is an infinite set of distinct
functionals. There are only finitely many nonzero root spaces ghγ , so for some N we must have
(ad x)N ghβ = 0. This holds for all root spaces gβ . Replacing N by its maximum over the finite
number of nonzero root spaces, we have (ad x)N g = 0, which proves the claim.
Our next goal is to show that there is a Zariski open subset of g consisting of images of elements of
h under the action of the group G. Let {αi } be the set of nonzero functionals α such that ghα 6= 0,
h
and let {bj }m
j=1 be a basis for ⊕i gαi consistingP of a union of bases for each gαh i . Then any element
of g has a unique expansion of the form h + m j=1 xj bj for some h ∈ h and some scalars xj ∈ F.
Define a map f : g → g by
Xm
f h + xj bj = ex1 ad b1 ex2 ad b2 · · · exm ad bm (h).
j=1
This is well-defined because each ad bj is nilpotent by the result of the last paragraph. Moreover, it
is a polynomial map. Looking to apply Lemma 9.4, we compute P the differential df at an arbitrary
point a ∈ h, evaluated at a point b + h (where h ∈ h and b = xj bj ). This is
d d h tx1 ad b1 tx2 ad b2 txm ad bm
i
df |x=a (b + h) = [f (a + t(b + h))] = e e · · · e (a + th) .
dt t=0 dt t=0
To compute this derivative, it suffices to expand the function we are differentiating to first order in
t. Doing so, we find
m m
d Y d X
df |x=a (b + h) = (I + txj ad bj ) (a + th) = (a + th) + txj (ad bj )(a) ,
dt
t=0 dt t=0
j=1 j=1
Pm
which equals h+ j=1 xj [bj , a] = h+[b, a]. Thus the linear operator df |x=a restricts to the identity
on h and to − ad a on ⊕i ghαi . On each space ghαi , the only eigenvalue
of − ad a is −αi (a). Thus if we
can find a ∈ h such that αi (a) 6= 0 for each i, then df |x=a will act invertibly on each generalized
root space and so it will be nonsingular as an operator on g.
4
It is easy to prove that this is possible. Indeed, for each i we can find ai ∈ h such that αi (ai ) 6= 0,
just because the functional αi ∈ h∗Pis nonzero. We identify the span of the ai ’s with affine space
via the association f = (fi ) 7→ a = i fi ai . For each i, the condition αi (a) 6= 0 defines a nonempty
Zariski open subset. The (finite) intersection of these over all i is then also nonempty, as desired.
We have now proven that the polynomial map f has the property that df |x=a is nonsingular at
some point a. Then by Lemma 9.4, the image of f contains a nonempty Zariski open subset of
g. Let Ωh be such a subset. Notice that, by definition, the image of f (and thus the subset Ωh )
consists of points of the form σ(h) for some σ ∈ G and h ∈ h.
The arguments so far hold for any Cartan subalgebra of g. Thus, letting h1 and h2 be two arbitrary
Cartan subalgebras, we obtain corresponding nonempty Zariski open subsets Ωh1 and Ωh2 of g. Let
Ωr be the subset of all regular elements of g; this is also a nonempty Zariski open subset. The
intersection Ωh1 ∩ Ωh2 ∩ Ωr is then nonempty. Rephrasing this, there exists a regular element x ∈ g,
elements h1 ∈ h1 and h2 ∈ h2 , and automorphisms σ1 , σ2 ∈ G such that σ1 (h1 ) = x = σ2 (h2 ). As
x is regular and σ1 is an algebra automorphism of g, h1 = σ1−1 (x) is also regular. Thus by Lemma
9.3, h1 = gh0 1 . Similarly, h2 = gh0 2 . The automorphism σ = σ2−1 σ1 ∈ G maps h1 to h2 , and so (as in
the proof of Corollary 9.2)
σ(h )
σ(h1 ) = σ(gh0 1 ) = g0 1 = gh0 2 = h2 .
This finishes the proof of Chevalley’s Theorem.
5
18.745 Introduction to Lie Algebras October 12, 2010
Definition 10.1. Let g be a Lie algebra and π a representation of g on a finite dimensional vector
space V . The associated trace form is a bilinear form on g, given by the following formula:
(a, b)V = tr (π(a)π(b))
Proposition 10.1. (i) The trace form is symmetric, i.e. (a, b)V = (b, a)V .
(ii) The trace form is invariant, i.e. ([a, b], c)V = (a, [b, c])V .
Proof. (i) follows from the fact that tr (AB) = tr (BA). For (ii), note the following:
([a, b], c)V = tr (π([a, b])π(c)) = tr (π(a)π(b)π(c) − π(b)π(a)π(c))
= tr (π(a)π(b)π(c) − π(a)π(c)π(b)) = tr (π(a)π([b, c]))
= (a, [b, c])V
Definition 10.2. If dim g < ∞, then the trace form of the adjoint representation is called the
Killing form:
κ(a, b) = tr ((ad a)(ad b))
Exercise 10.1. Let F be a field of characteristic 0. Suppose (·, ·) be an invariant bilinear form on
g. Show that if v ∈ g is such that ad a is nilpotent, then (ead v a, ead v b) = (a, b). In other words,
the bilinear form is invariant with respect to the group G generated by ead v , ad v nilpotent.
Proof.
1 1
(ead v a, ead v b) = (a + (ad v)a + (ad v)2 + · · · , b + (ad v)b + (ad v)2 b + · · · )
2! 2!
X X 1
j k
= (a, b) + ((ad v) a, (ad v) b)
j!k!
i≥1 j+k=i
1
v)j a, (ad v)k b) = 0. Note the following:
P
Thus it suffices to prove that for i ≥ 1, j+k=i j!k! ((ad
((ad v)j a, (ad v)k b) = ((ad v)(ad v)j−1 a, (ad v)k b) = −((ad v)j−1 a, (ad v)k+1 b)
Here we have used the invariance of (·, ·). So this means ((ad v)i a, b) = −((ad v)i−1 , (ad v)b) =
((ad v)i−2 , (ad v)2 b) = · · · . So:
X 1 X 1 1 X i!
((ad v)j a, (ad v)k b) = (−1)k ((ad v)i a, b) = ((ad v)i a, b) (−1)k =0
j!k! j!k! i! j!k!
j+k=i j+k=i
Pi k i
Above we have used the fact that k=0 (−1) k = 0.
1
Exercise 10.2. Show that the trace form of gln (F) and sln (F) associated to the standard represen-
tation is non-degenerate and the Killing form on sln (F) is also non-degenerate, provided char F - 2n.
Find the kernel of the Killing form on gln (F).
Proof.
P To show the trace form on gln (F) for the standard representation
P is non-degenerate, if
i,j aij eij lies in the kernel where for some r, s, ars 6= 0, then tr(( i,j aij eij )esr ) = tr(ars ers esr ) =
ars 6= 0 since tr(ei,j es,r ) = 0 unless j = s, i = r, which is a contradiction.
P To show the trace form
on sln (F) for the standard representation is non-degenerate, if x = i,j aij eij lies in the kernel
and some P ars 6= 0, then by the same argument we have P a contradiction. So say ars = 0 for r 6= s,
so x = aii eii . If ajj 6= akk for some j, k, then tr( aii eii , ejj − ekk ) = ajj − ak 6= 0, which is a
contradiction. So ajj = akk ∀j, k, so trx = na11 = 0, so a11 = 0 ( since char F - n, and x = 0 ).
For the killing form on gln (F), consider the basis of gln (F), {eij }. Then:
ad eij ad ekl (egh ) = [eij , δlg ekl − δlk egl ] = δjk δlg ekh − δki δlg ekj − δjg δhk eil + δli δhk egj
The coefficient of egh in this expansion is agh = δgi δjk δlg − δjk δhg δhi δlg − δgi δlh δjg δhk + δhj δli δhk .
So:
X X
κgln (eij ekl ) = agh = (δgi δjk δlg − δgk δhg δhi δlg − δgl δkl δjg δhk + δhj δli δhk )
g,h g,h
= nδil δjk − δkl δij − δij δkl + nδjk δil = 2nδil δjk − 2δij δkl = 2ntr(eij ekl ) − 2tr(eij )tr(ekl )
Lemma 10.2 (Cartan). Let g be a finite dimensional Lie algebra over F = F, a field of character-
istic 0 (so that Q ⊂ F). Let π be a representation of g in a finite dimensional vector space V . Let
h be a Cartan sub-algebra of g, and consider the generalized weight space decomposition of V and
the generalized root space decomposition of g with respect to h:
M M
V = Vλ , g = gα , π (gα ) Vλ ⊆ Vλ+α , [gα , gβ ] ⊆ gα+β
λ∈h∗ α∈h∗
Pick e ∈ gα , f ∈ g−α , so that h = [e, f ] ∈ g0 = h. Suppose that Vλ 6= 0. Then λ(h) = rα(h), where
r ∈ Q depends only on λ and α but not on h.
L
Proof. Let U = n∈Z Vλ+nα ⊂ V . Then dimU < ∞, and U is π(e), π(f ) and π(h) invariant. But
[π(e), π(f )] = π(h), hence trU (π(h)) = 0. Thus we have the following:
X X
0 = trU (π(h)) = trVλ+nα (π(h)) = (λ + nα)(h)dim Vλ+nα
n n
2
In the last line we have used the fact that the matrix of π(h)|Vλ+nα takes the following form:
(λ + nα)(h) ∗
(λ + nα)(h) ∗
A=
..
.
(λ + nα)(h)
X X
=⇒ λ(h)( dim Vλ+nα ) = −α(h) n dim Vλ+nα
n
P n
n dim Vλ+nα
=⇒ λ(h) = rα(h), r = − Pn
n dim Vλ+nα
P
Note in the above that Vλ 6= 0, so n dim Vλ+nα 6= 0.
Theorem 10.3 (Cartan’s criterion). Let g be a subalgebra of gl(V ), where V is a finite dimensional
vector space over F = F, a field of characteristic 0. Then the following are equivalent:
3. g is solvable.
Lemma 10.4. If g ⊂ glV , such that [g, g] = g, then (a, a)V 6= 0 for some a ∈ g.
M M
V = Vλ , g = gα
λ∈h∗ α∈h∗
P
Since [g, g] = g, and [gα , gβ ] ⊂ gα+β , we obtain that h = α∈h∗ [gα , g−α ]. Hence by Cartan’s Lemma
λ(h) = rλ,α α(h) for rλ,α ∈ Q, 6= 0 if Vλ 6= 0. The assumption that (a, a)V = 0 means that, for all
h ∈ [gα , g−α ]:
X X X
0 = (h, h)V = trV (π(h)2 ) = trVλ (π(h)2 ) = λ(h)2 dim Vλ = α(h)2 2
rλ,α dim Vλ
λ∈h∗ λ∈h∗ λ
3
In the above, we have used the fact that trVλ (π(h)2 ) = λ(h)2 dim Vλ , which follows from the Jordan
form theorem, since π(h)|Vλ can be expressed as an upper triangular matrix with λ(h)-s on the
Pfollows from this calculation that α(h) = 0, and hence λ(h) = 0 for all h ∈ [gα , g−α ].
diagonal. It
Since h = α∈h∗ [gα , g−α ], it follows that λ(h) = 0 for all h ∈ h. Since λ ∈ h∗ was arbitrary, this
means V = V0 .
Since π(gα )V = π(gα )V0 ⊂ Vα = 0 for α 6= 0, it follows that gα = 0 for α 6= 0. Hence g = g0 , and g
is nilpotent. This contradicts the fact that [g, g] = g.
Corollary 10.5. A finite dimensional Lie algebra g over F = F is solvable iff κ(g, [g, g]) = 0.
Proof. Consider the adjoint representation g → gl(g). Its kernel is Z(g). So g is solvable iff
adg ⊂ gl(g) is a solvable. But by Cartan’s criterion, adg is solvable iff κ(g, [g, g]) = 0.
Exercise 10.3. Consider the following 4-dim solvable Lie algebra D = Heis3 + Fd, where Heis3 =
Fp + Fq + Fc, with the relations [d, p] = p, [d, q] = −q, [d, c] = 0. Define on D the bilinear form
(p, q) = (c, d) = 1, rest = 0. Show that this is a non-degenerate symmetric invariant bilinear form,
but (D, [D, D]) 6= 0, so Cartan’s criterion fails for this bilinear form.
Remark. Very often one can remove the condition F = F by the following trick. Let F ⊂ F be the
algebraic closure. Given a Lie algebra g over F, let g = F ⊗F g be a Lie algebra over F.
4. ga0 is a Cartan sub-algebra for every regular element of a ∈ g for any field F.
4
1. Note [g, g] = [g, g]. To see this, [g, g] is the F-span of [ai , aj ] with certain linear relations
holding between them, so both [g, g] and [g, g] are the F-span of [ai , aj ] with certain linear
relations holding between them; and the Lie algebra structure is the same. Iterating this, we
have g(i) = g(i) . So g is solvable ↔ ∃i, g(i) = 0 ↔ ∃i, g(i) = 0 ↔ g is solvable. A similar
argument shows that [g, [g, g]] = [g, [g, g]], and more generally, gi = g i . So g is nilpotent
↔ ∃i, gi = 0 ↔ ∃i, g i = 0 ↔ g is nilpotent.
2. In Cartan’s Criterion, note that the second condition (a, a)V = 0∀a ∈ [g, g] is equivalent
to the condition (a, b)V = 0∀a, b ∈ [g, g], since (·, ·)V is symmetric (to see this, expand
(a + b, a + b)V = 0 and note it is characteristic 0). Thus by Cartan’s criterion, we have
(g, [g, g])V = 0 ↔ (g, [g, g])V = 0 ↔ (a, b)V = 0∀a, b ∈ [g, g] ↔ (a, b)V = 0∀a, b ∈ [g, g]. So the
first two conditions of Cartan’s criterion are equivalent for charF = 0. By Cartan’s criterion,
(a, b)V = 0∀a, b ∈ [g, g] ↔ (a, b)V = 0∀a, b ∈ [g, g] ↔ g is solvable ↔ g is solvable, so the
last two conditions of Cartan’s criterion are equivalent for charF = 0. For the Corollary, g is
solvable ↔ g is solvable ↔ κ(g, [g, g]) = 0 ↔ κ(g, [g, g]) = 0, so the Corollary is true for all
charF = 0.
5
18.745 Introduction to Lie Algebras October 14, 2010
1. if a, b ⊂ g are ideals, then a + b and a ∩ b are ideals, and if a and b are solvable then
a + b and a ∩ b are solvable.
Solution:
Let x ∈ g, a ∈ a, b ∈ b. Then [x, a + b] = [x, a] + [x, b] ∈ a + b since a, b are ideals. Similarly,
if y ∈ a ∩ b then [x, y] ∈ a, [x, y] ∈ b, so a ∩ b is an ideal too.
By an easy induction, (a ∩ b)(i) ⊂ a(i) , which is eventually zero if a is solvable, so a ∩ b is
solvable.
We showed earlier (in lecture 4) that if h ⊂ g and both h, g/h are solvable, then g is solvable.
Consider (a + b)/b. By hypothesis, b is solvable. Noether’s Third Isomorphism Theorem
shows (a + b)/b ∼ = a/(a ∩ b) (more explicitly, look at the homomorphisms a → a + b and
a + b → (a + b)/b, notice that the kernel of their composition is exactly a ∩ b). Since a is
solvable, a/(a ∩ b) is solvable. Hence a + b is solvable.
For part 2, given any a1 , a2 ∈ a and b1 , b2 ∈ b, [a1 + b1 , a2 + b2 ] = [a1 , a2 ] + [b1 , a2 ] + [a1 , b2 ] +
[b1 , b2 ]. Notice the sum of the first three terms is in a since it is an ideal, and the last term
is in b since it is a subalgebra, thus the entire sum is in a + b. The closure of a + b under
addition and scalar multiplication is obvious. Thus the statement is proven.
Proposition 11.1. The radical of g contains any solvable ideal of g and is unique.
Proof. If a is a solvable ideal of g, then a + R(g) is again a solvable ideal. Since R(g) is of
maximal dimension, a + R(g) = R(g) and a ⊂ R(g). For uniqueness, if there are two distinct
maximal dimensional solvable ideals of g, then by above explanation, the sum is actually
equal to both ideals. Thus we have a contradiction.
If g is a finite dimensional solvable Lie algebra, then R(g) = g. The opposite case is when
R(g) = 0.
1
Proposition 11.2. A finite dimensional Lie algebra g is semisimple if and only if either of
the following two conditions holds:
hence r(k−1) is a nonzero abelian ideal, since all the r(i) are ideals of g. Abelian ideals are
solvable, so the other direction is obvious.
Remark 11.1. Let g be a finite dimensional Lie algebra and R(g) its radical. Then g/R(g)
is a semisimple Lie algebra. Indeed, if g/R(g) contains a non-zero solvable ideal f, then its
preimage r contains R(g) properly, so that r/R(g) ∼
= f, which is solvable, hence r is a larger
solvable ideal than R(g), a contradiction.
So an arbitary finite dimensional Lie algebra “reduces” to a solvable Lie algebra R(g) and a
semisimple Lie algebra g/R(g).
Theorem 11.1. (Levi decomposition) If g is a finite dimensional Lie algebra over a field F
of characteristic 0, then there exists a semisimple subalgebra s ⊂ g, complementary to the
radical R(g), such that g = s ⊕ R(g) as vector spaces.
Remark 11.2. The open end of n goes on the side of the ideal. When both are ideals, we
use × or ⊕, and the sum is direct.
Exercise 11.2. Let h and r be Lie algebras and let γ : h → Der(r) be a Lie algebra homo-
morphism. Let g = h ⊕ r be the direct sum of vector spaces and extend the bracket on h and
on r to the whole of g by letting
for h ∈ h and r ∈ r. Show that this provides g with a Lie algebra structure, g = h n r, and
that any semidirect sum of h and r is obtained in this way. Finally, show that g = h × r if
and only if γ = 0.
2
Solution:
The bracket so defined is clearly skew-symmetric. Restricted to h or r it satisfies the Ja-
cobi identity. Take h1 , h2 ∈ h and r1 , r2 ∈ r. Then we have [h1 , [r1 , r2 ]] + [r1 , [r2 , h1 ]] +
[r2 , [h1 , r1 ]] = γ(h1 )([r1 , r2 ])−[r1 , γ(h1 )(r2 )]+[r2 , γ(h1 )(r1 )] = [γ(h1 )(r1 ), r2 ]+[r1 , γ(h1 )(r2 )]−
[r1 , γ(h1 )(r2 )] + [r2 , γ(h1 )(r1 )] = 0. We used the fact that γ(h1 ) is a derivation.
Furthermore, [h1 , [h2 , r1 ]] + [h2 , [r1 , h1 ]] + [r1 , [h1 , h2 ]] = [h1 , γ(h2 )(r1 )] − [h2 , γ(h1 )(r1 )] −
γ([h1 , h2 ])(r1 ) = γ(h1 )γ(h2 )(r1 ) − γ(h2 )γ(h1 )(r1 ) − (γ(h1 )γ(h2 )(r1 ) − γ(h2 )γ(h1 )(r1 )) = 0.
We used the fact that γ : h → Der(r) is a Lie algebra homomorphism. It follows that [, ] is
a Lie bracket. Clearly r is an ideal under the bracket, so g = h n r.
Conversely, by the Jacobi identity, any bracket on h n r is a homomorphism from h to Der(r)
(essentially we just reverse the two calculations above). Finally, if γ = 0 then [h, r] = 0 for
all h ∈ h, r ∈ r, and if [h, r] = 0 for all h, r then plainly γ(h) = 0 for all h, so g = h × r if
and only if γ = 0.
Exercise 11.3. Let g ⊂ gln (F) be a subspace consisting of matrices with arbitrary first m
rows and 0 for the rest of the rows. Find R(g) and a Levi decomposition of g.
Solution:
Write x ∈ g as (A, B), where A is the upper-left m×m matrix block and B is the upper-right
m × (n − m) block of x. Take y = (A0 , B 0 ). Then it is clear that [x, y] = ([A, A0 ], AB 0 − A0 B).
Hence, if h ⊂ glm ,→ g is an ideal of glm , then (h, 0) + R is an ideal of g, where R denotes
the set of all (A, B) with A = 0. Furthermore, h is solvable if and only if h + R is solvable,
because of [R, R] = 0 and the above identity. Notice that the radical of g obviously contains
R. Hence, the radical of g corresponds to the radical of glm , i.e. R(g) = (R(glm ), 0) + R.
But, by the notes and problem 4 below, the radical of glm is FI, the scalar matrices. Hence
R(g) = (FI, 0) + R (sum of ideals). The complement of this can obviously be the subalgebra
(slm , 0).
Theorem 11.2. Let V be a finite-dimensional vector space over an algebraically closed field
of characteristic 0 and let g ⊂ glV be a subalgebra, which is irreducible i.e. any subspace
U ⊂ V , which is g−invariant, is either 0 or V . Then one of two possibilities hold:
1. g is semisimple
Proof. If g is not semisimple, then R(g) is a non-zero solvable ideal in g. By Lie’s theorem,
there exists λ ∈ R(g)∗ such that Vλ = {v ∈ V |av = λ(a)v, a ∈ R(g)} is nonzero. By Lie’s
lemma, Vλ is invariant. Hence, by irreducibility Vλ = V . Hence a = λ(a)IV for all a ∈ R(g),
so R(g) = FI. Hence (g ∩ slV ) ∩ R(g) = 0, which proves that we have case 2, as g ∩ slV is
semisimple since it is the complement of the radical.
3
Exercise 11.4. Let V be finite-dimensional over a field F which is algebraically closed and
characteristic 0. Show that glV and slV are irreducible subalgebras of glV . Deduce that slV
is semisimple.
P
Solution: Suppose W ⊂ V, W 6= 0 is fixed by sln . Take nonzero vector w = ci vi ∈ W ,
where {vi } is a basis of V . Suppose ck 6= 0, and pick ` 6= k. Then e`k ∈ sln , and
e`k w = ck e` ∈ W . For every m 6= `, em` ∈ sln , so em` e` = em ∈ W . Hence W = V .
Since sln ⊂ gln , gln is also irreducible. It follows from the theorem that sln ∩ sln = sln is
semisimple.
Let g be a finite-dimensional Lie algebra. Recall the Killing form on g: K(a, b) = trg (ad a)(ad b).
Exercise 11.5. Let V be a finite-dimensional vector space with a symmetric bilinear form
(, ). Let U be a subspace such that the restriction (, )|U ×U is non-degenerate. Denote U ⊥ =
{v ∈ V |(v, U ) = 0}. Then V = U ⊕ U ⊥ .
Proof. 1. v ∈ a⊥ means (v, a) = 0. If b ∈ g, then ([v, b], a) = (v, [b, a]) = 0, since the form
is invariant and a is an ideal. Hence a⊥ is an ideal.
4
2. Follows from the preceeding exercise and part 1.
Definition 11.4. A Lie algebra g is called simple if its only ideals are 0 and g and g is not
abelian.
Corollary 11.1. Any semisimple, finite-dimensional Lie algebra over a field F of charac-
teristic 0 is a direct sum of simple Lie algebras.
Proof. If g is semisimple, but not simple, and if a is an ideal, then by the theorem, the
Killing form restricted to a is non-degenerate, hence g = a ⊕ a⊥ , where a and a⊥ are also
semisimple. After finitely many steps it can be decomposed into simple algebras.
5
18.745 Introduction to Lie Algebras October 19, 2010
In lecture 10, we saw that Cartan’s criterion holds without requiring the base field to be algebraically
closed. Similarly, we can deduce the semisimplicity criterion proved in the last lecture assuming
only that the base field has characteristic zero.
Exercise 12.1. Let g denote a Lie algebra over F with char(F) = 0. Show that g is semisimple if
and only if the Killing form on g is nondegenerate.
Solution: When we proved the theorem in lecture 11, we used the algebraic closure of the base field
F only to apply Cartan’s criterion. We know from exercise 10.4, however, that Cartan’s criterion
is valid assuming merely that F has characteristic 0. So the same proof applies in this case.
We should not be led to think, however, that the assumption of algebraic closure is never necessary.
Merely requiring that the base field has characteristic zero is not enough for instance to derive
Chevalley’s theorem on the conjugacy of Cartan subalgebras.
Exercise 12.2. Let g = sl2 (R). Show that there are two distinct conjugacy classes of Cartan
subalgebras given by
1 0 0 1
h1 = R and h2 = R .
0 −1 −1 0
Solution: Let {e, f, h} be the standard basis of g where [h, e] = 2e, [h, f ] = −2f , and [e, f ] = h.
We want to show that there exist two distinct conjugacy classes of Cartan subalgebras in g given
by Rh and R(e − f ).
If w = ae + bf + ch, then the characteristic polynomial of ad w is
In this lecture and the few that follow, we will study the structure of finite dimensional semisimple
Lie algebras with the aim of classifying them. This will amount to a detailed knowledge of root
space decompostions. The first step is to use the Killing form to understand Cartan subalgebras
and their actions under the adjoint representation.
Unless otherwise stated, we will assume throughout that our base field F is algebraically closed of
characteristic zero.
We begin by generalizing our notion of Jordan decomposition to an arbitrary Lie algebra.
Definition 12.1. An abstract Jordan decomposition of an element of a Lie algebra g is a decom-
position of the form a = as + an , where
(c) [as , an ] = 0 .
Exercise 12.3. Abstract Jordan decomposition in a Lie algebra g is unique when it exists if and
only if Z(g) = 0.
Solution: (⇒) Suppose c ∈ Z(g) is not 0, and that a ∈ g has abstract Jordan decomposition
a = as + an . Let a0s = as − c and a0n = an + c. We note the following facts: a = (as − c) + (an + c) =
a0s + a0n ; [a0s , a0n ] = [as , an ] = 0 ; ad a0s = ad as and ad a0n = ad an as Z(g) = ker ad . Since
ad an is nilpotent, and ad as is semisimple, we conclude that a = a0s + a0n is an abstract Jordan
decomposition. But a0s 6= as as c 6= 0. So the decomposition for a is not unique.
(⇐) Suppose that a = as + an = a0s + a0n are abstract Jordan decompositions for some a ∈ g.
Since ad (as + an ) and ad (a0s + a0n ) are commuting operators, and ad a = ad as + ad an and
ad a = ad a0s + ad a0n are Jordan decompositions, we know from a lemma of lecture 6, that
and consequently
[ad a0s , ad as ] = 0 [ad a0n , ad an ] = 0 .
Since the sum of commuting semisimple (resp. nilpotent) operators is semisimple (resp. nilpotent),
Our first result in this lecture will be to show the existence of abstract Jordan decompositions
under the assumption of semisimplicity.
Theorem 12.1. Let g be a finite dimensional semisimple Lie algebra over F.
(a) Z(g) = 0.
2
(b) All derivations of g are inner.
(c) Any a ∈ g admits a unique Jordan decomposition.
Proof. (a): One of our equivalent definitions of semisimplicity is that g has no nontrivial abelian
ideals. Hence Z(g) = 0.
ad
(b): By construction g −→ Der g with kernel Z(g). Since Z(g) = 0, this implies g ∼
= ad g.
Consider the trace form on Der g under the tautological representation. The restriction to ad g is
the Killing form on g, which is nondegenerate by semsimplicity. From a lemma of lecture 11, we
have that Der g = ad g ⊕ ad g ⊥ as a direct sum of Lie algebras.
We need to show that ad g ⊥ = 0. In the contrary case, there exists a nonzero element D ∈ ad g ⊥ .
For a ∈ g, we have [D, ad a] = 0. We recall from exercise 2.1 that [D, ad a] = ad D(a). Hence
D(a) ∈ Z(g) for any a ∈ g. Since Z(g) = 0, we conclude that D = 0 in contradiciton to our
assumption.
(c): Fix a ∈ g. Consider the classical Jordan decompostion of ad a = As + An , where As is
diagonalizable, An is nilpotent, and [As , An ] = 0.
Consider the generalized eigenspace decomposition of g with respect to ad a
M
g= gaλ .
λ∈F
Recall that gaλ = (ad a − λI)N = 0 , and [gλ , gµ ] ⊂ gλ+µ with As |gλ = λI.
We want first to show that As is a derivation. By linearity, it is enough to check this for x ∈ gaλ
and y ∈ gaµ . We have
As ([x, y]) = (λ + µ) · [x, y] = [λx, y] + [x, µy] = [As x, y] + [x, As y] .
This means that As ∈ Der g.
By part (b), all derivations are inner. So there exists as ∈ g such that ad as = As . Letting
an = a − as , we have that ad an = An . It remains to check that [as , an ] = 0. Since
ad ([as , an ]) = [ad as , ad an ] = [As , An ] = 0
part (a) gives the result.
With the previous result under our belts, let us get a firmer understanding of the Cartan subalgebras
of g from Theorem 12.1.
Choose a Cartan subalgebra h of g, and consider the generalized root space decomposition
M
g= gα with g0 = h
α∈h∗
where [gα , gβ ] ⊂ gα+β . Recall that in the case of the adjoint representation, we say generalized root
space decomposition instead of generalized weight space decomposition. It is important to make
this distinction; we will see its convenience in later lectures, as we try to better understand the
functionals α appearing in the decomposition.
3
Theorem 12.2. Let g and h be as defined, and K denote the Killing form on g.
(a) K(gα , gβ ) = 0 if α + β 6= 0.
for N 0 as g is finite dimensional. So we can choose N sufficiently large so that the opera-
tor ((ad a)(ad b))N is zero on each summand of the decomposition. Therefore (ad a)(ad b) is
nilpotent. By exercise 3.4, the eigenvalues of (ad a)(ad b) are zero implying that
By Theorem 12.1 (a), this implies [h0 , hs ] = 0 for h0 ∈ h. But since h is a maximal nilpotent
subalgebra, we conclude that hs ∈ h.
It remains to be shown that hn = 0. Note that hn = h − hs ∈ h.
Since ad h is solvable, Lie’s theorem implies that there exists a basis of glg such that the elements
of ad h are upper triangular. As ad hn is nilpotent, it is strictly upper triangular. Therefore
tr ((ad h0 )(ad hn )) = 0 for h0 ∈ h; namely K(h, hn ) = 0. By part (b), K |h×h is nondegenerate. So
hn = 0, implying that h = hs is diagonalizable.
4
For arbitrary Lie algebras, Cartan subalgebras can have an unwieldy structure, and an unpre-
dictable action under the adjoint representation. We see from the previous result, however, that
with the assumption of semisimplicity the situation is much more transparent.
We can rewrite the root space decomposition as
M
g=h⊕ gα
α∈∆
where ∆ is the set of all α 6= 0 such that gα 6= 0. Recall that gα = {a ∈ g | [h, a] = α(h)a ∀h ∈ h}.
Definition 12.2. We call α ∈ ∆ a root of g, and gα the corresponding root space.
The rest of the classification will be concerned more or less with gathering information about roots
of g. Having put the Killing form to good use in determining the structure of Cartan subalgebras,
we would like to extend it to ∆. We have a canonical linear map
ν : h 3 h 7→ K(h, •) ∈ h∗ .
Since K |h×h is nondegenerate by Theorem 12.2 (b), this implies ν is injective. As h and h∗ have
the same dimension, ν determines a vector space isomorphism.
Definition 12.3. Abusing notation, we define a bilinear form K : h∗ × h∗ → F by the rule
K(γ, µ) = K(ν −1 (γ), ν −1 (µ)). Note that K(ν(h), ν(h0 )) = ν(h)(h0 ) = ν(h0 )(h).
Theorem 12.3. (a) If α ∈ ∆, e ∈ gα , f ∈ g−α , then [e, f ] = K(e, f )ν −1 (α).
Proof. (a): We know that [e, f ] ∈ h. By Theorem 12.2 (b), K |h×h is nondegenerate. So it is enough
to show that K([e, f ] − K(e, f )ν −1 (α), h0 ) = 0 for any h0 ∈ h. We have
Note that we have used the invariance of K, and the fact that h0 ∈ h, f ∈ g−α . This gives the
result.
(b): By Theorem 12.2 (b), K |gα ×g−α is nondegenerate. So we can find e ∈ gα , f ∈ g−α such that
K(e, f ) = 1. By part (a), we know that [e, f ] = ν −1 (α). Hence
and similarly
[ν −1 (α), f ] = α(ν −1 (α))f = K(α, α)f .
Suppose to the contrary that K(α, α) = 0. By the above relations, we obtain a Lie algebra
a = Fe + Ff + Fν −1 (α)
isomorphic to Heis3 . Recall that Heis3 is solvable with center [Heis3 , Heis3 ].
5
So applying Lie’s theorem to the restriction of the adjoint representation to a, we can find a basis
of glg such that ν −1 (α) = [a, a] is strictly upper triangular. But by Theorem 12.2 (d), we know
that ν −1 (α) ∈ h is diagonalizable. So we conclude that ν −1 (α) = 0, and α = 0. This contradicts
definition 12.2, and the fact that α ∈ ∆.
Exercise 12.4. (a) Show that all derivations of the 2-dimensional nonabelian Lie algebra are inner.
(b) Find Der Heis3 . Note that not all derivations are inner.
Solution (a) Let g be the 2-dimensional nonabelian Lie algebra over F with ordered basis {e, f }
where [e, f ] = f . Choose D ∈ Der (g) with
a b
D= where a, b, c, d ∈ F.
c d
We have
P g = Heis3 over F with ordered basis {p, q, c} where [p, q] = c. Choose D ∈ Der (Heis3 ) with
(b) Let
D = 1≤i,j≤3 aij · eij . Note that
0 a13 0 −a13 0 0
[D, ad p] = 0 a23 0 , [D, ad q] = −a23 0 0 .
−a21 a33 − a22 −a23 −a33 + a11 a12 a13
Recall from exercise 2.1 that [D, ad p] = ad D(p) and [D, ad q] = ad D(q). Since ad p = e32 and
ad q = −e31 , it follows that a13 = a23 = 0 and a33 = a11 + a22 ; namely
a11 a12 0
D = a21 a22 0 .
a31 a32 a11 + a22
By checking on basis elements, we see that every matrix of this form is indeed a derivation.
Hence the derivations of g form a 6-dimensional subspace, while the inner derivations form a 2-
dimensional subspace.
Exercise 12.5. Show that Theorem 12.1 parts (b) and (c), and Theorem 12.2 part (c) hold for
char(F) = 0.
Solution (12.1 (b)): Note that Theorem 12.1 (a) makes no assumptions on F. We check that the
lemma from lecture 11, does not require that the F be algebraically closed. Hence we only use
6
algebraic closure to deduce that the Killing form on g is nondegenerate. This follows from exercise
12.1 merely assuming that F has characteristic zero.
(12.1(c)): Let g = F ⊗F g. Choose a ∈ g. We know that there exist as , an ∈ g such that a = as + an
is an abstract Jordan decomposition. Choose a basis {g1 , . . . , gn } of g, and note that {g 1 , . . . , g n } =
{1 ⊗ g1 , . . . , 1 ⊗ gn } is a basis of g.
Claim. as ∈ g :
Since char(F) = 0, this implies that F/F is a Galois extension. Recall that f ∈ F iff σ(f ) = f for
all σ ∈ Gal(F/F).
P
Let as = 1≤i≤n αi g i . Suppose to the contrary that αj ∈ F − F. There exists σ ∈ Gal(F/F) such
that σ(αj ) 6= αj . Define a map
X X
Φ:g3 βi g i 7→ σ(βi )g i ∈ g .
1≤i≤n 1≤i≤n
since [g i , g j ] = [gi , gj ] ∈ g.
Hence we check that a = Φas + Φan is an abstract Jordan decomposition of a. By exercise 12.3,
we know that Φas = as . This is a contradiction.
7
18.745 Introduction to Lie Algebras October 21, 2010
Throughout this lecture, let g be a finite dimensional semisimple Lie Algebra over an algebraically
closed field F of characteristic 0.
where,
3. The restriction K|gα ×g−α , i.e. K(a, b) with a ∈ gα and b ∈ g−α , is non-degenerate, so it
induces a pairing between gα and g−α . In particular, we have dim gα = dim g−α .
2
Now, given α ∈ ∆, pick non-zero E ∈ gα and F ∈ g−α such that K(E, F ) = K(α,α) . Let H =
2ν −1 (α)
K(α,α) . Then, we can check that:
[H, E] = 2E,
[H, F ] = −2F,
[E, F ] = H.
The choice of E and F is possible by 3 and the last claim of 4. We only verify the first equality,
the second being analogous and the third coming from 5. We have:
2α ν −1 (α)
2ν −1 (α) 2K(α, α)
[ , E] = E= E = 2E,
K(α, α) K(α, α) K(α, α)
1
where the first equality comes from ν −1 (α) ∈ h and E ∈ gα .
Lemma 1.1 (Key Lemma for sl2 (F)). Let F be a field of characteristic 0. Let π be a representation
of sl2 (F) in a vector space V over F and let v ∈ V be a non-zero vector such that π(E)v = 0 and
π(H)v = λv for some λ ∈ F (such vector is called a singular vector of weight λ).
Then:
c) If dim V < ∞, then λ ∈ Z≥0 , the vectors π(F )j v for 0 ≤ j ≤ λ are linearly independent, and
π(F )λ+1 v = 0.
Proof. a) We prove this by induction on n. For n = 0, the result is given to us. Suppose it holds
for n = k − 1 for some k > 0. Then:
Exercise 13.1. b) Again, we use induction on n. For the case n = 1, we have π(E)π(F )v =
π(F )π(E)v + [π(E), π(F )]v = 0 + π ([E, F ]) v = π(H)v = λv. Suppose the result holds for
n = k − 1 for some k > 1. Then:
c) Suppose λ ∈
/ Z≥0 . Then, the term n(λ − n + 1) is non-zero for all n ≥ 1. Hence, by induction,
we get π(F )n v 6= 0 for all n ∈ Z≥0 . But by a), this implies that all vectors π(F )n v are
2
eigenvectors of π(H) with distinct eigenvalues. This allows us to conclude dim V = ∞,
therefore proving the first claim. If now, λ ∈ Z≥0 , by the same argument we see that the
vectors π(F )n v are linearly independent for 0 ≤ n ≤ λ, and moreover, if π(F )λ+1 v 6= 0, then
by induction we see that π(F )n v 6= 0 for all n > λ + 1, and so there are infinitely many
linearly independent vectors. Hence, if dim V < ∞, then π(F )λ+1 v = 0, proving the last two
claims.
Exercise 13.2. Using the notation of the Key Lemma for sl2 (F), if v instead satisfies that π(F )v =
0 and π(H)v = λv, then we have:
c) If dim V < ∞, then −λ ∈ Z≥0 , the vectors π(E)j v for 0 ≤ j ≤ −λ are linearly independent,
and π(F )−λ+1 v = 0.
Proof. It is enough to check that the function ψ(E) = F , ψ(F ) = E, ψ(H) = −H, is an automor-
phism of aα . Indeed:
Thus, the Key Lemma shows that if π(ψ(E))v = 0 and π(ψ(H))v = λ0 v with λ0 ∈ F, then:
c) If dim V < ∞, then λ0 ∈ Z≥0 , the vectors π(ψ(F ))j v for 0 ≤ j ≤ λ0 are linearly independent,
0
and π(ψ(F ))λ +1 v = 0.
Theorem 1.2. The root space decomposition of g with respect to a Cartan Subalgebra h and the
set of roots ∆ satisfy the following properties:
3
Proof. a) Suppose that dim gα > 1 for some α ∈ ∆, then dim g−α > 1 by non-degeneracy of the
restriction K|gα ×g−α , property 3 above. Consider the adjoint representation of the subal-
−1
gebra aα = FE + FF + FH on g. In particular, recall H = 2ν (α)
K(α,α) . Since dim g−α > 1,
there exists a non-zero vector v ∈ g−α such that K(E, v) = 0. Hence, (ad E)v = [E, v] =
−1 (α),v] −2α(ν −1 (α))v
K(E, v)ν −1 (α)=0. But (ad H)v = [H, v] = 2[νK(α,α) = K(α,α) = −2K(α,α)v
K(α,α) = −2v.
Hence, dim g = ∞ by the Key Lemma, which yields a contradiction.
b) Let q be the largest integer such that β + qα ∈ ∆ ∪ {0}. Notice q ≥ 0. Pick a non-zero vector
v ∈ gβ+qα . Then, (ad E)v = 0 since it lies in gβ+(q+1)α . Also, (ad H)v = (β + qα)(H)v =
2K(α,β)
K(α,α) + 2q v. Hence, by the Key Lemma: λ := 2K(α,β) j
K(α,α) + 2q ∈ Z≥0 and (ad F ) v are non-
zero vectors for 0 ≤ j ≤ λ. But (ad F )j v ∈ gβ+(q−j)α , so β+qα, β+(q−1)α, . . . , β+(q−λ)α ∈
∆ ∪ {0}. Define p := −(q − λ) = q + 2K(α,β) 0 0
K(α,α) . Let p be the largest integer for which β − p α ∈
∆ ∪ {0}. Again, notice p0 ≥ 0. Pick a non-zero vector v 0 ∈ gβ−p0 α . Then, (ad F )v 0 = 0
2K(α,β)
and (ad H)v 0 = 0 0
K(α,α) − 2p v . By the corollary of the Key Lemma, we conclude that
−λ0 := 2p0 − 2K(α,β) 0 0 0 0
K(α,α) ∈ Z≥0 and that β − p α, β − (p − 1)α, . . . , β − (p + λ )α ∈ ∆ ∪ {0}.
Define q 0 := −(p0 + λ0 ) = p0 − 2K(α,β) 0
K(α,α) . Since q and p are the largest integers for which
β + qα ∈ ∆ ∪ {0} (resp. β − p0 α ∈ ∆ ∪ {0}), we conclude that q ≥ q 0 and p0 ≥ p. Hence,
2K(α,β) 2K(α,β)
K(α,α) = p − q ≤ p0 − q 0 = K(α,α) , showing that p = p0 , q = q 0 , p, q ∈ Z≥0 .
c) Pick the largest integers p and q such that β − pα, β + qα ∈ ∆ ∪ {0}. Pick a non-zero vector
v ∈ gβ−pα . Then, (ad F )v = 0 and (ad H)v = 2K(α,β) K(α,α) − 2p v. By the corollary of the
Key Lemma, (ad E)j v 6= 0 for 0 ≤ j ≤ 2p − 2K(α,β) K(α,α) = p + q. But q ≥ 1 since α + β ∈ ∆,
so (ad E) v is a non-zero vector. Its corresponding root is α + β, and (ad E)p v ∈ gβ , so
p+1
4
precisely the set of diagonal matrices of g. A basis for h∗ is given by {ε1 − ε2 , ε2 − ε3 , . . . , εn−1 − εn },
where εi (b) = bi for any b = diag (b1 , b2 , . . . , bn ) ∈ h and i ∈ {1, . . . , n}. We can check that:
Hence, the set ∆ := {εi − εj |i, j ∈ {1, . . . , n}, i 6= j} is a root system for g. For no pair of
roots α, β ∈ ∆ it is true that β + 3α ∈ ∆ ∪ {0}. Thus, the only possibilities for (q, p) are
(2, 0), (1, 1), (0, 2), (1, 0), (0, 1), (0, 0).
When n = 2, we can only have (q, p) = (2, 0), (0, 2). Let α := ε1 − ε2 . Then ∆ = {±α}, so
α = α + (0)α, 0 = α − (1)α, −α = α − (2)α and we have α, 0, −α ∈ ∆ ∪ {0}, giving all possible
values for p and q.
If n = 3, we can only have (q, p) = (2, 0), (0, 2), (1, 0), (0, 1). We have the pairs (2, 0) and (0, 2) by
the previous case. Now, letting β := ε1 − ε3 and γ := ε2 − ε3 so that ∆ : {±α, ±β, ±γ}, we see
that α − β, α ∈ ∆ ∪ {0} but α − 2β, α + β ∈ / ∆ ∪ {0}, and similar relations hold among α, γ and
β, γ by symmetry.
If n ≥ 4, we can only have (q, p) = (2, 0), (0, 2), (1, 0), (0, 1), (0, 0). The first four pairs come from
the previous two cases. The fifth pair (0, 0) occurs if we let δ = ε3 − ε4 and then notice that
α − δ, α + δ ∈
/ ∆ ∪ {0}.
In general, let αij := εi − εj for all i, j ∈ {1, . . . , n} with i 6= j. Then, for all multisets {i, j, k, l} ⊆
{1, . . . , n}:
if {i, j} ∩ {k, l} = 2, then αij and αkl are related by pairs (2, 0), (0, 2);
if {i, j} ∩ {k, l} = 1, then αij and αkl are related by pairs (1, 0), (0, 1);
if {i, j} ∩ {k, l} = 0, then αij and αkl are related by the pair (0, 0).
5
18.745 Introduction to Lie Algebras October 22, 2010
In this lecture we will prove more consequences of semisimplicity. Then we will prove a theorem
that is used to prove that a Lie algebra is semisimple. Lastly, we will begin to examine examples
of semisimple Lie algebras.
Recall some of the facts that we already know about semisimple Lie algebras. Let g be a finite
dimensional semisimple Lie algebra over F, an algebraically closed field of characteristic 0. Choose
a Cartan subalgebra h and consider the root space decomposition:
!
M
g=h⊕ gα , [h, h] = 0, dimgα = 1.
α∈∆
We use this decomposition to rewrite the Killing form K as a sum over root spaces. For h1 , h2 ∈ h
we have:
X
K(h1 , h2 ) = trg (ad h1 )(ad h2 ) = α(h1 )α(h2 ). (1)
α∈∆
X X
K(λ1 , λ2 ) = α(ν −1 (λ1 ))α(ν −1 (λ2 )) = K(λ1 , α)K(λ2 , α). (2)
α∈∆ α∈∆
Definition 14.1. h∗Q ⊂ h∗ is the Q-span of ∆. (Note that Q ⊂ F since charF = 0.)
1. ∆ spans h∗ over F.
Proof. 1. Suppose ∆ does not span h∗ , then there exists a non-zero h ∈ h such that α(h) = 0 for
all α ∈ ∆. This implies that [h, gα ] = 0 for all α from the definition of root space. It is also true
that [h, h] = 0, as from lecture 12 the Cartan subalgebra h is abelian. Therefore, h is in Z(g). But
g is semisimple, so Z(g) = 0, which is a contradiction.
2. From Equation (2) we have:
1
X
K(λ, λ) = K(λ, α)2 ∀λ ∈ h∗ . (3)
α∈∆
4 X 2K(λ, α) 2
= , λ ∈ ∆. (4)
K(λ, λ) K(λ, λ)
α∈∆
4 2K(λ,α)
It follows that K(λ,λ) ∈ Z for λ ∈ ∆ because by the string condition K(λ,λ) = p − q ∈ Z. This
implies that K(λ, λ) ∈ Q for λ ∈ ∆.
2K(α,β)
But since K(α,α) ∈ Z for α, β ∈ ∆ and K(α, α) ∈ Q we conclude that K(α, β) ∈ Q.
3. From part 2, K(α, β) ∈ Q, hence K(λ, α) ∈ Q for λ ∈ h∗Q α ∈ ∆. Since by part 1, ∆ spans h∗
and the Killing form is non-degenerate on h, its restriction to h∗Q is non-degenerate as well.
Hence by equation (3), K(λ, λ) ≥ 0 for λ ∈ h∗Q since it is the sum of rational squares. This proves
that K is positive and semi-definite.
It is a theorem of linear algebra that any non-degenerate positive semi-definite symmetric bilinear
form is positive definite. This proves part 3.
Every semisimple algebra is the direct sum of simple algebras. The following exercise shows that
this decomposition is unique up to permutation.
Prove that this decomposition is unique up to permutation of the summands and prove that any
ideal of g is a subsum of this sum.
Proof. We first prove the second part, take any ideal h. h ∩ sj is an ideal of the subalgebra sj . sj
is simple so this is either 0 or sj . So h = N
L L
j=1 (h ∩ sj ) = j∈A sj , which proves the second part.
We will now examine how the decomposition of a Lie algebra corresponds to a decomposition of
its root space.
Let g be the direct sum of two ideals, g = g1 ⊕ g2 , where both gi are semisimple. Consider for each
of them the root space decomposition. To do this we choose a Cartan subalgebra hi in each gi .
!
M M
g i = hi ⊕ gα g=h⊕ gα
α∈∆i α∈∆
2
Where h is h1 ⊕ h2 and ∆ = ∆1 ⊔ ∆2 .
If α ∈ ∆1 and β ∈ ∆2 and [gα , gβ ] = 0, then we have that α + β ∈
/ ∆ and α + β 6= 0. In other
words,
∆ = ∆1 ⊔ ∆2 , α+β ∈
/ ∆ ∪ 0 if α ∈ ∆1 , β ∈ ∆2 . (5)
Definition 14.2. The set ∆ in a vector space V is called indecomposable if it cannot be decomposed
into a disjoint union of non-empty subsets ∆1 and ∆2 such that equation 5 holds.
With this notion, we can give a simple way to check if a semisimple Lie algebra is simple.
L
Theorem 14.2. Let g = h ⊕ ( α∈∆ gα ) be a decomposition of a finite dimensional Lie algebra g
into a direct sum of subspaces such that the following properties hold:
3. h∗ is spanned by ∆.
Then g is a semisimple Lie algebra. Moreover if the set ∆ is indecomposable, then g is simple.
n
X
u= v λi , v λ i ∈ Vλ i , λi 6= λj . (6)
i=1
We will prove that all vλi are in U by induction on n. For the case n = 1, vλ1 = u ∈ U .
For the case, n > 1 we apply π(h) to both sides:
n
X
π(h)u = λi (h)vλi (7)
i=1
3
Now for the proof of the theorem.
Proof. We want to prove that if a is an abelian ideal of g, then a = 0. Note that since a is an
ideal, it is invariant with respect to ad h on g. Hence by the lemma, either gα is in a for some α
or h ∩ a is non-zero. This uses the fact that dimgα = 1. In the first case, [gα , g−α ] = Fhα ⊂ a,
but [hα , gα ] 6= 0 since α(hα ) 6= 0. So a contains the non-abelian subalgebra Fhα ⊕ gα , which is
impossible since a is abelian.
In the second case, where h ∩ a is non-zero, let h ∈ a, h ∈ h, h 6= 0. By condition 3, h∗ is spanned by
∆, so α(h) 6= 0 for some α ∈ ∆. Hence [h, gα ] = gα ∈ a, which again contradicts that a is abelian.
So a=0.
This proves that g is semisimple.
Now the proof of simplicity, given that ∆ is indecomposable.
Since g is semisimple, we have that g = g1 ⊕ g2 , where the gi are non-zero ideals. In the contrary
case, by our discussion before Definition 14.2, this implies that ∆ is decomposable. This is a
contradiction, so g is simple.
Exercise 14.2. Prove that a finite subset: ∆ ⊂ V \ {0} is an indecomposable set if and only if for
any α, β ∈ ∆, there exists a sequence γ1 , γ2 . . . γs such that α = γ1 , β = γs and γi + γi+1 ∈ ∆ for
i = 1 . . . s − 1.
Also for any ∆ ⊂ V \ {0}, construct its canonical decomposition into a disjoint union of indecom-
posable sets.
Example 14.1. g = sln (F), where F is an algebraically closed field with characteristic 0. Let the
Cartan subalgebra h be the set of all traceless diagonal matrices. h lies in Dn , the space of all
diagonal matrices. Denote by ǫi the following linear function on D: ǫi (A) = ai , the ith coordinate
of the diagonal of matrix A. The set {ǫi | i = 1 . . . n} is clearly a basis of D∗ . If charF ∤ n, another
basis is {ǫi − ǫi+1 , ǫ1 + . . . + ǫn | i = 1 . . . n − 1}, since it generates the first basis and has the same
size. h = {a ∈ D | (ǫ1 + . . . + ǫn )a = 0}. Therefore, h∗ = D∗ /(ǫ1 + . . . + ǫn ), and {ǫi − ǫi+1 | i 6= j}
is a basis for h∗ .
4
L
The root space decomposition of g = sln (F ) is g = h α∈∆ gα , where ∆ = {ǫi − ǫj |i 6= j} and
gǫi −ǫj = Feij .
To prove sln (F) is semisimple, we check the conditions of theorem 14.2. 1. is clearly true. 2.
[eij , eji ] = eii − ejj , (ǫi − ǫj )(eii − ejj ) = 2 6= 0. This is why F cannot be characteristic 2. 3. When
char ∤ n, the ǫi − ǫj span h, so ∆ spans h.
This implies sln (F) is semisimple for any n 6= 2 and charF ∤ n. To show it is simple, we prove that
∆ is indecomposable.
Let α = ǫi − ǫj , β = ǫs − ǫt . Using exercise 14.2, let γ1 = α, γ2 = ǫj − ǫs , γ3 = β. This is a string
from α to β as γ1 − γ2 = ǫi − ǫs ∈ ∆ and γ2 − γ3 = ǫj − ǫt ∈ ∆. This proves ∆ is indecomposable
and that sln (F) is simple.
Exercise 14.3. The above argument fails if charF | n. As sln (F) contains a non-trivial abelian
ideal, Z(sln (F)) since In ∈ Z(sln (F)). How does the argument fail?
Proof. The argument fails because ∆ does not span h∗ . When charF | n, In ∈ h and ∆ is still
{ǫi − ǫj | i 6= j}. Since (ǫi − ǫj )(In ) = 0, ∆ can’t span h∗ .
5
18.745 Introduction to Lie Algebras October 28, 2010
Recall
OV,B (F) = {a ∈ glV (F) | B(au, v) + B(u, av) = 0, for all u, v ∈ V } ⊂ glV (F)
where V is a vector space over F, B is a bilinear form : V xV → F. Choosing a basis of V and
denoting by B the matrix of the bilinear form in this basis, we proved we get the subalgebra
For different choices of basis, we get isomorphic Lie algebras on,B (F).
Now, consider the case where B is a symmetric non-degenerate bilinear form. If F is algebraically
closed and char F 6= 2, one can choose a basis in which the matrix of B is any symmetric non-
degenerate matrix.
and denote by soN (F) the corresponding Lie algebra oN,B (F).
Exercise 15.1. Show soN (F) = {a ∈ glN (F)|a + a0 = 0} where a0 is the transposition of a with
respect to the anti-diagonal.
Proof. soN (F) = {a ∈ glN F) | aT B + Ba = 0} where B is the matrix consisting of ones along the
anti-diagonal.
As B = B T , we have aT B = aT B T = (Ba)T . Viewing B as a permutation matrix, we get Ba
permutes the rows by rowi → rown−i . Transposing and reapplying B, we get B(Ba)T = a0 and
BBa = a. Hence we obtain the following sequence of implications
aT B + Ba = 0
(Ba)T + Ba = 0
B(Ba)T + a = 0
1
and finally
a0 + a = 0.
Therefore, soN (F) = {a ∈ glN (F)|a + a0 = 0}.
α 0
Example 15.2. so2 (F) = { , α ∈ F}, which is one-dimensional abelian, hence not
0 −α
semisimple.
a1
..
.
an
h= 0 . ⊂ so2n+1 (F)
−an
..
.
−a1
This is a Cartan subalgebra since it contains a diagonal matrix with distinct entries.
Case 2: N = 2n (even). Let
a1
..
.
an
h= .
−an
..
.
−a1
2
Exercise 15.2. a) Using the root space decomposition, prove that soN (F) is semisimple if N ≥ 3.
b) Show soN (F) is simple if N = 3 or N ≥ 5 by showing that ∆ is indecomposable.
Thus we have another two series of simple Lie algebras: so2n+1 (F) for n ≥ 1 (type B) and so2n (F)
for n ≥ 3 (type D).
Proof. a) We must check (1), (2), and (3) of the semisimplicity criterion.
(1) is clear for B and D and (3) is clear for B. For (3) in case D, we have roots i − j and i + j ,
adding and dividing by 2 (as char F 6= 2) gives us i , hence (3) holds.
(2) We compute [gα , g−α ] = [eij − eN +1−j,N +1−i , eji − eN +1−i,N +1−j ] = (eii − eN +1−i,N +1−i ) +
(eN +1−j,N +1−j − ejj ) = hα ∈ h. As α(hα ) 6= 0, soN is semisimple in N ≥ 3.
b) To show simple for N = 3 and N ≥ 5, we show that ∆ is indecomposable. This is clear for
N = 3. We list pairs and corresponding paths for n ≥ 3. N = 5 is done separately. For ease of
notation, we write i as i and remark that any root is connected to its negative by the path of
length one.
• (i + j) → (j + k) via (i + j, −k − i, j + k)
• (i + j) → (i − j) via (i + j, k − i, j − k, i − j)
Exercise 15.3. Show ∆so4 (F) = {1 − 2 , 2 − 1 } t {1 + 2 , −1 − 2 } is the decomposition into
decomposables. Deduce that so4 (F) is isomorphic to sl2 (F) ⊕ sl2 (F).
This is in fact a decomposition since the collection of roots distance one from 1 − 2 is 2 − 1 as
1 − 2 + 1 + 2 = 21 ∈
/ ∆ and 1 − 2 + −1 − 2 = −22 ∈ / ∆. Hence, we have a decomposition.
To show isomorphic to sl2 ⊕ sl2 . Consider the basis of h, x = e11 − e44 and y = e22 − e33 . If we
change bases to from x, y to x + y, x − y. Let
Then
[x + y, e] = 0, [x + y, f ] = 2f, [x + y, g] = −2g, [x + y, h] = 0, [g + f ] = x + y
and
[x − y, e] = 2e, [x − y, f ] = 0, [x − y, g] = 0, [x − y, h] = −2h, [e, h] = x − y
3
and finally
[x + y, x − y] = [e, f ] = [e, g] = [h, f ] = [h, g] = 0.
Therefore, we have two copies of sl2 formed by {f, g, x + y} and {e, h, x − y}, and therefore an
isomorphism.
Show always indecomposable and deduce that sp2n (F) simple for all n ≥ 1.
These Lie algebras are called type C simple Lie algebras.
Proof. We must check (1),(2), and (3) of the semisimplicity criterion. (1) is clear.
For (2), split a matrix M into its four quadrants. Label the upper left quadrant A. The upper left
half of the upper right quadrant B, with that portion of the anti-diagonal X. Finally, the upper
left half of the lower left quadrant C, with that portion of the anti-diagonal Y . We then have the
following eigenvectors.
4
• eij − eN +1−j,N +1−i if eij ∈ A
• eij if eij ∈ X ∪ Y
[eij − eN +1−j,N +1−i , eji − eN +1−i,N +1−j ] = (eii − eN +1−i,N +1−i ) + (eN +1−j,N +1−j − ejj ) = hα ∈ h
[eij + eN +1−j,N +1−i , eji + eN +1−i,N +1−j ] = (eii − eN +1−i,N +1−i ) − (eN +1−j,N +1−j − ejj ) = hα ∈ h
[eij , ejj ] = eii − ejj = hα ∈ h.
In each case α(hα ) 6= 0, and therefore (2) holds.
Finally, (3) is clear, so we only must show in-decomposablility to get simplicity.
From Exercise 15.2, when n ≥ 3, have that all pairs of the form ±i ± j are connected to i − j .
We may connect 2j to i − j as 2j + i − j = i + j ∈ ∆ and therefore through concatenation
we are done.
When n = 2 we may connect i − j to i + j as i − j + i + j = 2i ∈ ∆, so we have in-
decomposability.
When n = 1, indecomposability is clear.
Remark 1. Thus we get four series of simple Lie algebras An = sln (F)(n ≥ 1), Bn = so2n+1 (F)(n ≥
1), Cn = sp2n (F)(n ≥ 1), Dn = so2n (F), (n ≥ 4) called the classical simple Lie algebras.
Proof. a) If (·, ·) is an invariant bilinear form and I is its kernel, then I is an ideal, hence g simple
implies that either I = 0 or I = g.
b) Choose a basis of g and let Bi be the matrix of (·, ·)i in the basis. Det(Bi ) 6= 0. Consider
det(B1 − λB2 ) = det(B2 )det(B1 B2−1 − λI) = 0 if λ is an eigenvalue of B1 B2−1 . Hence the form
(a, b)1 −λ(a, b)2 is a degenerate, invariant, bilinear form as det is 0. Hence the form (a, b)1 −λ(a, b)2
is identically zero by (a), which implies (a, b)1 = λ(a, b)2 .
Corollary 15.3. If g ⊂ glN (F) is a simple Lie algebra, then the Killing from on g is proportional
to the trace form (a, b) = tr ab on g.
Example 15.3. On glN (F): (1) tr eii eij = δij (with eii basis of D), hence the induced bilinear
form on D∗ = (i , j ) = δij (2). Hence for all classical simple Lie algebras A,B,C,D, the Killing
form is a positive constant multiple of (1) and on h∗ is a positive constant multiple of (2).
5
Definition 15.1. Let V be a finite dimensional real Euclidean space, i.e. V finite dimensional
vector space over R with symmetric positive definite bilinear form (·, ·).
Let ∆ ⊂ V be a subset of V . Then the pair (V, ∆) is called a root system if:
i) ∆ finite, 0 ∈
/ ∆, ∆ spans V over R;
T
ii) (String Condition) For any α, β ∈ ∆, the set {β + jα | j ∈ Z} (∆ ∪ {0}) is a string β + pα, β +
(α,β)
(p − 1)α, ..., β − qα where p, q ∈ Z, and p − q = 2 (α,α) ;
Example 15.4. The basic example: Let g be a finite dimensional Lie algebra over an algebraically
closed field F of characteristic 0, h a Cartan subalgebra, ∆ ⊂ h∗Q the set of roots, (·, ·) the Killing
form on h∗Q which is Q-valued and positive definite.
Let V = R ⊗Q h∗Q , i.e. linear combinations of roots with real coefficients and extend the Killing
form by bilinearity. Then the pair (V, ∆) is a root system, called the g root system.
Remark 2. This construction is independent of the choice of the Cartan subalgebra h due to
Chevalley’s Theorem.
Exercise 15.5. Let (V, ∆) be a root system. Then ∆ is indecomposable if and only if there does
not exist non-trivial decomposition (V, ∆) = (V1 , ∆1 ) ⊕ (V2 , ∆2 ) where V = V1 ⊕ V2 , V1 ⊥ V2 ,
∆i ⊂ Vi , and ∆ = ∆1 ∪ ∆2 . (Hint: Use String Condition)
F
Moreover, the decomposition of ∆ = ∆i into indecomposable sets corresponds to decomposition
of the root system in the orthogonal direct sum of indecomposable root systems.
6
18.745 Introduction to Lie Algebras November 1, 2010
Recall that a root system is a pair (V, ∆), where V is a finite dimensional Euclidean space over R
with a positive definite bilinear form (·, ·) and ∆ is a finite subset, such that:
1. 0 ∈
/ ∆; R∆ = V ;
(V, ∆) is called indecomposable if it cannot be decomposed into a non-trivial orthogonal direct sum.
r = dimV is called the rank of (V, ∆) and elements of ∆ are called roots.
Definition 16.1. An isomorphism of an indecomposable root system (V, ∆) and (V1 , ∆1 ) is a vector
space isomorphism ϕ : V → V1 , such that ϕ(∆) = ∆1 , and (ϕ(α), ϕ(β))1 = c(α, β) for all α, β ∈ ∆,
where c is a positive constant, independent of α and β. In particular, replacing (·, ·) by c(·, ·), where
c > 0, we get, by definition, an isomorphic root system.
Example 16.1. Root systems of rank 1: (R, ∆ = {α, −α}), α �= 0, (α, β) = αβ. This root system
is isomorphic to that of sl2 (F), so3 (F), and sp2 (F).
Proposition 16.1. Let (V, ∆) be an indecomposable root system with the bilinear form (·, ·). Then
1. Any other bilinear form (·, ·)1 for which the string property holds is proportional to (·, ·), i.e.
(α, β)1 = c(α, β) for some positive c ∈ R, independent of α and β.
Proof. Fix α ∈ ∆. Since (V, ∆) is indecomposable, for any β ∈ ∆ there exists a sequence
γ0 , γ1 , . . . , γk such that α = γ0 , β = γk , (γi , γi+1 ) �= 0 for all i = 0, . . . , k − 1. Define c by
(α, α)1 = c(α, α). By the string property p − q = 2 (α,γ 1) (α,γ1 )1
(α,α) = 2 (α,α)1 . Hence (α, γ1 )1 = c(α, γ1 ).
2(α,γ1 ) 2(α,γ1 )1
Likewise, by the string property, (γ1 ,γ1 ) = (γ1 ,γ1 )1 . Hence (γ1 , γ1 ) = c(γ1 , γ1 ). Continuing this way,
we show that (γ2 , γ2 )1 = c(γ2 , γ2 ), . . . (β, β)1 = c(β, β). Since 2(α,β) 2(α,β)1
(α,α) = (α,α)1 , we conclude that
(α, β)1 = c(α, β) for all α, β ∈ ∆. Since ∆ spans V , we conclude that (1) holds. The same argument
proves (2).
Definition 16.2. A lattice in an Euclidean space V is a discrete subgroup (Q, +) of V , which spans
V over R, i.e. RQ = V . For example, Zn ⊂ Rn .
Proposition 16.2. If ∆ is a finite set in an Euclidean space V , spanning V over R, such that
(α, β) ∈ Q for all α, β ∈ ∆, then Z∆ is a lattice in V.
1
Proof. The only thing to prove is that Z∆ is a discrete
�r set. Choose a basis β1 , . . . , βr � of V among the
vectors of ∆. Then for any α ∈ ∆, we have α = i=1 ci βi , ci ∈ R. Hence, (α, βj ) = ri=1 ci (βi, , βj ).
But ((βi , βj ))ri,j=1 is a Gramm matrix of a basis, hence it is non-singular. Hence the ci ’s can be
computed by Cramer’s rule, so all ci ∈ Q. So Z∆ ⊂ Q{β1 , . . . , βr }. But since ∆ is finite, we
conclude that Z∆ ⊂ N1 Z{β1, . . . , βr } where N is a positive integer. But N1 Z{β1 , . . . , βr } is discrete,
hence Z∆ is discrete.
√ √
Example 16.2. {1, 2} ⊂ R, then Z{1, 2}is not a discrete set.
Corollary 16.3. If (V, ∆) is a root system, then Q := Z∆ is a lattice, called the root lattice.
Proof. By the two propositions, the corollary holds if (V, ∆) is indecomposable, hence holds for any
root system (V, ∆).
We will list four series of root systems of rank r known to us. In all cases (�i , �j ) = δij .
Type g V ∆ Q
r+1
� r+1
� r+1
� r+1
�
A slr+1 (F) { ai �i |ai ∈ R, ai = 0} {�i − �j |1 ≤ i, j ≤ r + 1} { ai �i |ai ∈ Z, ai = 0}
i=1 i=1 i=1 i=1
r
� r
�
B so2r+1 (F) { ai �i |ai ∈ R} {±�i ± �j , ±�i |1 ≤ i, j ≤ r, i �= j} { ai �i |ai ∈ Z}
i=1 i=1
�r r
� r
�
C sp2r (F) { ai �i |ai ∈ R} {±�i ± �j , ±2�i |1 ≤ i, j ≤ r, i �= j} { ai �i |ai ∈ Z, ai ∈ 2Z}
i=1 i=1 i=1
�r �r �r
D so2r (F), r ≥ 3 { ai �i |ai ∈ R} {±�i ± �j |1 ≤ i, j ≤ r, i �= j} { ai �i |ai ∈ Z, ai ∈ 2Z}
i=1 i=1 i=1
�
Remark Explanation: in case A, V is a factor space of V� = r+1 i=1 ai �i by a 1-dimensional subspace
R(�1 + . . . + �r+1 ). Notice that (�1 + . . . +�
�r+1 ) = V in the
⊥
�table, so V� = V ⊕ R(�1 + . . . + �r+1 )
r+1 r+1
with direct sum. Secondly, why is ∆A = { i=1 ai �i |ai ∈ Z, i=1 ai = 0}. Clearly, Z{�i − �j |i �= j}
is included in this set. To show the reverse inclusion, write
r+1
�
Q� ai �i = a1 (�1 −�2 )+(a1 +a2 )(�2 −�3 )+. . .+(a1 +a2 +. . .+ar )(�r −�r+1 )+(a1 +. . .+ar+1 )(�r+1 ),
i=1
where the coefficient of the last term is zero. So the reverse inclusion is also true. For case B, the
form of the root lattice is clearly correct.
Exercise 16.1. Explain the root lattices in cases C and D.
2
D. The proof for QDr is identical except 2�i ∈ ∆Dr , so write
r
�
ai �i = a1 (�1 − �2 ) + (a2 + a1 )(�2 − �3 )+
i=1
ar + . . . + a1 ar + . . . + a1 ar + . . . + a1
... + (�r − �1 ) + (�r − �2 ) + (�1 + �2 )
2 2 2
as r ≥ 3, and this belongs to Z∆Dr .
Definition 16.3. A lattice Q is called integral (respectively even) if (α, β) ∈ Z (respectively (α, α) ∈
2Z) for all α, β ∈ Q. Note that an even lattice is always integral: if α, β ∈ Q, Q even, then
(α + β, α + β) = (α, α) + (β, β) + 2(α, β), (α, α), (β, β) ∈ 2Z. Hence 2(α, β) ∈ 2Z, so (α, β) ∈ Z.
Theorem 16.5. Let Q be an even lattice in an Euclidean space V , and assume the subset ∆ =
{α ∈ Q|(α, α) = 2} spans V over R. Then (V, ∆)is a root system.
Proof. Axiom (1) of a root system is clear, as is axiom (2): if (α, α) = 2, then (nα, nα) = 2 iff.
n = ±1. It remains to show the string property. Reversing the sign of α if necessary, we may assume
(α, β) ≥ 0. Note that for α, β ∈ ∆ we have:
where (α, β) ∈ Z, (α, β) ≥ 0. So the only possibilities are (α, β) = 0, 1 or 2. In the last case, hence
q = 0, p = 2, so p − q = (α, β) = 2 and the string property is satisfied.
3
The most remarkable lattice is E8 (which is even by proposition.)
Proof.
�8 8
� 8
� 8
1 �
E8 = { ai �i | all ai ∈ Z, ai ∈ 2Z} ∪ { ai �i | all ai ∈ Z + , ai ∈ 2Z}.
2
i=1 i=1 i=1 i=1
The elements from the first set satisfying (α, α) = 2 are clearly {±�i ± �j |i �= j} as (�i , �j ) = δij ,
�
and the second set must have all ai = ± 12 else (α, α) > 2, and as 8i=1 ai ∈ 2Z, there must be an
even number of minus signs.
� �
� 1 �
|∆E8 | = |{�i + �j |i �= j}| + |{−�i − �j |i �= j}| + |{�i − �j |i �= j}| + ��{ (±�1 ± . . . ± �8 )| even number of minus signs}��
2
8.7 8.7
= + + 8.7 + 27 = 240.
2 2
�i +�j �i −�j
Clearly �i = 2 + 2 ∈ R∆E8 , and {�i } form a basis of V , hence R∆ = V .
So (R8 , ∆E8 ) is a root system by the theorem, which is called the root system of type E8 .
Exercise 16.4. Consider the following subsystem of the root system of type E8 : take ρ = ( 12 , . . . , 12 )
and let ∆E7 = {α ∈ ∆E8 |(α, ρ) = 0}, QE7 = {α ∈ QE8 |(α, ρ) = 0}, VE7 = {v ∈ VE8 |(v, ρ) = 0}.
Show that (VE7 , ∆E7 ) is a root system of rank 7, and that |∆E7 | = 126.
Proof. Clearly ∆E7 = {α ∈ QE7 |(α, α) = 2}. QE7 is an even lattice in VE7 as it is a subgroup of
QE8 and clearly RQE7 = VE7 , so (VE7 , ∆E7 )is a root system of rank 7 as VE8 = VE7 ⊕ R( 12 , . . . , 12 ),
VE7 = ( 12 , . . . , 12 )⊥ , and dimVE8 = 8.
±�1 ± . . . ± �8
∆E7 = {�i − �j |i �= j} ∪ { | 4 minus signs.}
2
Hence, � �
8
|∆E7 | = 56 + = 126
4
Exercise 16.5. Let ∆E6 = {α ∈ ∆E7 |(α, �7 + �8 ) = 0}, QE6 = {α ∈ QE7 |(α, �7 + �8 ) = 0},
VE6 = {v ∈ VE7 |(v, �7 + �8 ) = 0}. Show that (VE6 , ∆E6 ) is a root system of rank 6, and that
|∆E6 | = 72.
4
Proof. Clearly ∆E6 = {α ∈ QE6 |(α, α) = 2}. QE6 is an even lattice in VE6 as it is a subgroup of
QE7 and clearly RQE6 = VE6 , so (VE6 , ∆E6 ) is a root system of rank 6 as VE7 = VE6 ⊕ R(�7 + �8 ),
VE6 = (�7 + �8 )⊥ , and dimVE7 = 7.
±�1 ± . . . ± �8
∆E6 = {�i −�j | if i = 7, j = 8, if i = 8, j = 7}∪{ | 4 minus signs and �7, �8 have opposite signs}
2
Hence, � �
6
|∆E6 | = 6.5 + 2 + 2 = 72.
3
5
18.745 Introduction to Lie Algebras November 04, 2010
Previously, given a semisimple Lie algebra g we constructed its associated root system (V, ∆).
(The construction depends on choosing a Cartan subalgebra, but by Chevalley’s theorem, the root
systems constructed from the same g are isomorphic.) Next, given a root system we’ll construct a
Cartan matrix A, and from this we’ll eventually see how to reconstruct g.
We’ll see that to every root system there corresponds a semisimple Lie algebra, so it’s important
to know all the root systems. Last time we saw the four series Ar , Br , Cr , and Dr , and the three
exceptions E6 , E7 , and E8 . The remaining two exceptions are F4 and G2 , which we will describe
in the following exercises.
Exercise 17.1. Define:
4
M
V = Ri ; (i , j ) = δij ;
i=1
( 4 )
X 1
QF4 = ai i | all ai ∈ Z or all ai ∈ + Z ; and
2
i=1
∆F4 = {α ∈ QF4 | (α, α) = 1 or 2} .
Show that (V, ∆F4 ) is an indecomposable root system of rank 4 with 48 roots.
Solution. A simple verification shows ∆ = {±εi , ±εi ± εj : i 6= j} ∪ {1/2(±ε1 ± ε2 ± ε3 ± ε4 )}, which
has 48 elements.
All that is difficult is to check that the string property holds. For this, we tabulate enough α, β
with the corresponding data p, q, (α, β), (α, α). I will use a shorthand for writing vectors — a string
of four numbers η1 η2 η3 η4 represents the vector η1 ε1 + η2 ε2 + η3 ε3 + η4 ε4 . Depending on how many
numbers are presumed zero in such a string, we take the other numbers η to be ±1 or ±1/2 as
appropriate.
In the last line here we have written δk+ to mean 1 when there are exactly k positive ηi and zero
otherwise. While considering each calculation of p and q, that the roots actually appear in strings
1
without gaps is easily checked. Thus we have a root system. To see that it is indecomposable,
we need to note that all the roots are equivalent under the the equivalence relation generated by
α ∼ α0 when (α, α0 ) 6= 0. Note that ±εi ∼ 1/2(±ε1 ± ε2 ± ε3 ± ε4 ), so that {εi , 1/2(±ε1 ± ε2 ± ε3 ± ε4 )}
is contained in an equivalence class. Next, ±εi ± εj ∼ εi , showing that all roots are equivalent.
Thus the root system is indecomposable.
Exercise 17.2. Show that the following describes an indecomposable root system with 12 roots:
VG2 = VA2 ;
QG2 = QA2 ; and
∆G2 = {α ∈ QA2 | (α, α) = 2 or 6} .
P 2
Solution. ∆ = {(a1 , a2 , a3 ) | a1 + a2 + a3 = 0, ai ∈ Z, ai ∈ {2, 6}}. Now in order to have
square sum 2, exactly two of the ai must equal ±1, so we have ±(1, −1, 0), ±(1, 0, −1), ±(0, 1, −1).
In order to have square sum 6, exactly one must be ±2, and the other two must both be ∓1:
±(2, −1, −1), ±(−1, 2, −1), ±(−1, −1, 2).
We tabulate enough of the relevant quantities below to verify that G2 is a root system, in the style
of the previous exercise. (Here ε stands for 1 or −1.)
α β p q p − q (α, β) (α, α)
(1, −1, 0) (ε, −ε, 0) 1+ε 1−ε 2ε 2ε 2
(ε, 0, −ε) 1/2(3 + ε) 1/2(3 − ε) ε ε 2
(2ε, −ε, −ε) 3/2(1 + ε) 3/2(1 + ε) 3ε 3ε 2
(−ε, −ε, 2ε) 0 0 0 0 2
(2, −1, −1) (ε, −ε, 0) 1/2(1 + ε) 1/2(1 − ε) ε 3ε 6
(0, ε, −ε) 0 0 0 0 6
(2ε, −ε, −ε) 1+ε 1−ε 2ε 6ε 6
(−ε, 2ε, −ε) 1/2(1 − ε) 1/2(1 + ε) −ε −3ε 6
Finally, to see that G2 is irreducible, note that no two of the shorter roots are perpendicular, and
each of the longer roots is not perpendicular to one of the shorter roots.
Definition 17.1. Suppose (V, ∆) is a root system and f : V → R is a linear map such that
f (α) 6= 0 for all α ∈ ∆. Then:
(i) α ∈ ∆ is positive if f (α) > 0 and negative if f (α) < 0.
(ii) A positive root is simple if it cannot be written as the sum of two positive roots.
(iii) A highest root θ ∈ ∆ is a root where f is maximal; that is, f (θ) ≥ f (α) for all α ∈ ∆.
Notation 17.1.
• ∆+ is the set of positive roots, and ∆− is the set of negative roots.
2
(b) Every positive root is a nonnegative integer linear combination of simple roots; i.e. ∆+ ⊆
Z≥0 Π.
(d) Π is a basis of V over R and of the lattice Q over Z. Hence the integer linear combinations
from part (b) are unique.
• γ ∈ ∆+ , so α = β + γ, which contradicts α ∈ Π; or
• γ ∈ ∆− , so β = α + (−γ), which contradicts β ∈ Π.
(b) If α is simple then we’re done. Otherwise, α = β + γ for some β, γ ∈ ∆+ . Then f (α) =
f (β) + f (γ), so both f (β) and f (γ) are strictly less than f (α). Repeat this process with β
and γ until all summands are simple (which must happen in finitely many steps since ∆ is
finite), thus yielding α as a sum of simple roots.
(c) Suppose α ∈ ∆+ r Π. If α − γ ∈ / ∆ for all γ ∈ Π, then the string condition would imply
2(γ,α)
(γ,γ) ≤ 0 for all γ ∈ Π. Then by (b),
X
(α, α) = α, kγ γ ≤ 0,
γ∈Π
(d) From (b) we have that Π spans ∆+ over Z. Then since ∆ = ∆+ ∪ ∆− = ∆+ ∪ −(∆+ ) and
Q = Z∆, Π spans Q over Z and thus spans V over R.
To prove linear independence of Π,Psuppose the contrary—that there existed a nontrivial
linear combination of simple roots i ki αi = 0. Split this into positive and negative parts,
moving the negative parts to the other side to obtain
X X
γ := ai αi = bi αi ,
i i
where all ai and bi are nonnegative and ai bi = 0 for all i. Since all αi are positive, f (γ) > 0,
so γ 6= 0 and (γ, γ) > 0. However, by (a) we also have
X X
(γ, γ) = ai αi , bj αj ≤ 0,
3
Suppose the contrary—then α = γ1 + γ2 for some α ∈ ∆ and γi ∈ Z≥0 Πi . By flipping the
sign of α if necessary, we can assume α ∈ ∆+ . By (b), we can subtract simple roots until γ1
is simple. Then Π1 ⊥ Π2 implies
2(α, γ1 ) 2(γ1 , γ1 )
= = 2,
(γ1 , γ1 ) (γ1 , γ1 )
Exercise 17.3. Prove that if (V, ∆) is an indecomposable root system and f : V → R is a linear
map such that f (α) 6= 0 for all α ∈ ∆, then there exists a unique highest root θ ∈ ∆.
P
Solution. Suppose first that θ = λi αi is a highest root. We wish first to show that λi > 0 for
root systems that λi ≥ 0 for all i). Suppose on the contrary that
all i (it is a basic property of P
λj = 0. Expanding (αj , θ) = λi (αj , αi ), as all the simple roots are at obtuse or right angles,
we have (αj , θ) ≤ 0, with equality iff αj ⊥ αi whenever λi 6= 0. Now as θ + αj cannot be a root,
the string condition implies that (αj , θ) = 0. Thus αj ⊥ αi whenever λi 6= 0 and λj = 0. Thus, if
λj = 0 for some j, the simple roots decompose into disjoint perpendicular subsets {αj : λj = 0}
and {αj : λj 6= 0}. This is impossible, as the root system is indecomposable.
Now suppose we have two distinct highest roots θ and θ̃. Then P (θ, θ̃) ≤ 0, as otherwise θ − θ̃ is a
root (yet f (θ − θ̃) = 0 which is impossible). Thus 0 ≥ (θ, θ̃) = λi (αi , θ̃), showing that (αi , θ̃) = 0
(as the string condition implies that (αi , θ̃) ≥ 0 for each i). Yet then θ̃ is perpendicular to a basis
of V , so θ = 0, a contradiction.
Definition
17.2. Let Π = {α1 , . . . , αr } be the set of simple roots of ∆ (corresponding to f ). The
2(αi ,αj ) r
matrix A = (αi ,αi ) is called the Cartan matrix. We will show later that it is independent of
ij=1
choice of f .
Proposition 17.2. The Cartan matrix has all entries integers, and the following properties:
(a) Aii = 2 for all i.
Definition 17.3. If V, ∆) is indecomposable, we have a unique highest rootrθ (by the above
e = 2(αi ,αj )
exercise). Let α0 = −θ, and Π0 = {α0 , α1 , . . . , αr }. The matrix A is called the
(αi ,αi ) ij=0
extended Cartan matrix.
4
Exercise 17.4. A
e satisfies all of the properties of Proposition 17.2, except det A
e = 0.
Solution. The proof of properties (a) and (b) are exactly the same as for the standard Cartan
matrix A. However, we need to see tht the principle minors are still all positive, except for the
determinant (which is zero).
Again, à factors as à = diag(2/(αi , αi ))ni=0 · ((αi , αj ))nij=0 , and the first matrix has all diagonal
entries positive, so we only need to investigate the principle minors of Q = ((αi , αi ))ni=0 ·.
Given any proper subset I of {0, . . . , n}, I is a subset of some subset I 0 of {0, . . . , n} with n elements.
Now the αi with i ∈ I 0 are a basis of V , as the highest root is
P
λi αi with the λi nonzero. Thus
the submatrix of Q corresponding to I 0 is a Gram matrix for the inner product, and thus has
all principle minors zero. In particular, the principle minor of Q corresponding to I is nonzero
(by Sylvester’s criterion), and thus so is the corresponding principle minor of Ã. Of course, the
determinant of Q (and thus A) is zero as the αi (i = 0, . . . , n) are not linearly independent.
Definition 17.4. A r × r matrix satisfying all of the properties of Proposition 17.2 is called an
abstract Cartan matrix.
Let’s classify the abstract Cartan matrices. The only 1 × 1 such matrix is (2), the Cartan matrix
of type A1 . There are more possibilities for 2 × 2 abstract Cartan matrices A. We know that
2 −a
A = −b 2 for nonnegative integers a, b. Moreover, 4 − ab > 0, so that ab = 3. There are four
possibilities for A (up to taking the transpose):
2 0 2 −1 2 −1 2 −1
, , , and .
0 2 −1 2 −2 2 −3 2
The Dynkin diagram D(A) depicts the Cartan matrix A by a graph with r vertices in bijection
with the simple roots. For any two distinct simple roots αi and αj , the corresponding 2 × 2 Cartan
matrix will be one of the above four, or a transpose thereof. The vertices corresponding to a chosen
pair of roots are joined as follows in each case:
2 0 2 −1 2 −1 2 −2 2 −1 2 −3
0 2 −1 2 −2 2 −1 2 −3 2 −1 2
◦ ◦ ◦−−◦ ◦==>◦ ◦<==◦ ◦≡≡>◦ ◦<≡≡◦
In each diagram, the left node corresponds to the first row and column of the matrix. Note that
when the two roots in question have different lengths, the arrow points to the shorter root. The
diagram formed in this way is the Dynkin diagram D(A).
Remark. Any subdiagram of a Dynkin diagram is again a Dynkin diagram. Π is indecomposable if
and only if the Dynkin diagram is connected.
We can now calculate Cartan matrices and Dynkin diagrams for various root systems. In each of
the following root systems, we have {εi } an orthonormal set of vectors (which is enough to calculate
inner products).
For Ar , ∆ = {εi − εj : 1 ≤ i, j ≤ r + 1, i 6= j}. Letting f (εi ) = r + 1 − i, the simple roots may be
indentified easily, as they all take value 1 under f . We find:
5
For the rest of the root systems treated in this lecture, we use the same function f (although there
is no basis vector εr+1 ).
For Br , ∆ = {±εi ± εj , ±εi : 1 ≤ i, j ≤ r, i 6= j}. We find:
Π = {ei − ei+1 : 1 ≤ i ≤ r − 1} ∪ {εr }, and θ = ε1 + ε2 .
The Cartan matrices are shown below when r = 6 — the pattern can be read off easily enough.
The augmented Cartan matrices are the whole matrix, where the standard matrix is the bottom
right 6 × 6 block.
2 −1 0 0 −1 0 −1 0
0 0 2 0 0 0
−1 2 −1 0 0 0 0 0 2 −1 0 0 0 0
0 −1 2 −1 0 0 0 −1 −1 2 −1 0 0 0
A6 :
0 0 −1 2 −1 0 0
B6 : 0 0 −1 2 −1 0 0
0
0 0 −1 2 −1 0
0
0 0 −1 2 −1 0
0 0 0 0 −1 2 −1 0 0 0 0 −1 2 −1
−1 0 0 0 0 −1 2 0 0 0 0 0 −2 2
l+
Ar :
l l ls s s l l
l+
Br :
l l ls s s l l
These diagrams are the extended Dynkin diagrams, while the Dynkin diagrams are obtained by
removing the node marked with +. (This comment applies to all four Dynking diagrams shown.)
Exercise 17.5. Perform the same analysis for the root systems Cr (r ≥ 2) and Dr (r ≥ 3).
Solution. For Cr , ∆ = {±εi ± εj , ±2εi | i 6= j}, ∆+ = {εi + εj , εi − εj | i 6= j} ∪ {2εi } and Π = {αi },
where αi = εi − εi+1 (1 ≤ i < r) and αr = 2εr . Furthermore, α0 = −2ε1 .
For Dr , ∆ = {±εi ± εj | i 6= j}, ∆+ = {εi + εj , εi − εj | i 6= j} and Π = {αi }, where αi = εi − εi+1
(1 ≤ i < r) and αr = εr−1 + εr . Furthermore, α0 = −ε1 − ε2 .
2 −1 0 0 −1 0
0 0 0 0 2 0 0 0
−2 2 −1 0 0 0 0 2 −1 0
0 0 0 0
0 −1 2 −1 0 0 0 −1 −1 2 −1 0 0 0
C6 :
0 0 −1 2 −1 0 0 D6 : 0 0 −1 2 −1 0 0
0
0 0 −1 2 −1 0
0
0 0 −1 2 −1 −1
0 0 0 0 −1 2 −2 0 0 0 0 −1 2 0
0 0 0 0 0 −1 2 0 0 0 0 −1 0 2
6
Cr : l
+ l ls s s l l
l+
l
Dr : l l ls s s l
PP
PP
P
P l
Solution.
A1 × A1 : A2 :
c
c c
c c c c
c c
c
B2 : G2 :
c
c c c
c c c c
c c c c
c c c c
c c c
c
7
18.745 Introduction to Lie Algebras October 21, 2010
In the examples that follow, we will compute the Cartan matrices for the indecomposable root
systems that we have encountered earlier. We record these as Dynkin diagrams, summarized in
Figure 1. Later in the lecture, we will prove that these are actually the Dynkin diagrams of all
possible indecomposable root systems. We also compute extended Dynkin diagrams specifically for
the purposes of this proof.
In the following examples, the rank of the root system is always denoted by r, and we get simple
roots α1 , . . . , αr . The largest root, used in the extended Dynkin diagrams, is denoted by θ.
Example 1.1. Ar (r ≥ 2): This corresponds to slr+1 (F). We have (εi , εj ) = δij and ∆slr+1 (F) =
{εi − εj | i, j ∈ {1, . . . , r + 1} and i 6= j} ⊂ V , where V is the subspace of r+1
L
i=1 Rεi on which the
sum of coordinates (in the basis {εi | i ∈ {1, . . . , r + 1}}) is zero.
Take f ∈ V ∗ given by f (ε1 ) = r + 1, f (ε2 ) = r, . . . , f (εr+1 ) = 1.
Then, ∆+ = {εi − εj | i < j}, Π = {εi − εi+1 | i ∈ {1, . . . , r}} and θ = ε1 − εr+1 .
For the Cartan matrix, recall that:
2 −1 0
0 ···
−1 2 −1 ..
.
A = 0 −1 2 . . . 0 .
..
.. ..
. . . −1
0 · · · 0 −1 2
Example 1.2. Br (r ≥ 3): This corresponds to so2r+1 (F). We have ∆so2r+1 (F) = {±εi ± εj , ±εi |
i, j ∈ {1, . . . , r} and i 6= j} ⊂ V = ri=1 Rεi .
L
1
Figure 1: Dynkin diagrams and extended Dynkin diagrams of all the indecomposable root systems,
from top to bottom, these correspond to: Ar , Br , Cr , Dr , E8 , E7 , E6 , F4 , G2 .
2
The Cartan matrix is then:
2 −1 00 ···
−1 2 −1 ..
.
A = 0 −1 2
. .
.
. 0
..
.. ..
. . . −1
0 · · · 0 −2 2
Example 1.3. Cr (r L ≥ 1): This corresponds to sp2r (F). We have ∆sp2r (F) = {±εi ± εj | i, j ∈
{1, . . . , r}} ⊂ V = ri=1 Rεi . We take the same f as in the previous case. Then, Π = {εi − εi+1 |
i ∈ {1, . . . , r − 1}} t {2εr } and θ = 2ε1 . This gives the Cartan matrix:
2 −1 0 · · · 0
−1 2 −1 ..
.
A = 0 −1 2 . . . 0 .
..
.. ..
. . . −2
0 · · · 0 −1 2
Example 1.4. Dr (r ≥ 4): This corresponds to so2r (F). We have ∆so2r (F) = {±εi ± εj | i, j ∈
{1, . . . , r} and i 6= j}. Define f ∈ V ∗ by f (ε1 ) = r − 1, . . . , f (εr ) = 0.
Then ∆ = {εi ± εj | i < j}, Π = {ε1 − ε2 , . . . , εr−1 − εr , εr−1 + εr } and θ = ε1 + ε2 .
The Cartan matrix is:
2 −1 0
0 ... ...
−1 2 −1 ..
.
. . . .
0 −1 . . 0 0
A= .
.
.
.. . . 2 −1 −1
..
. 0 −1 2 0
0 . . . 0 −1 0 2
Hence, we have the Dynkin diagrams in Figure 1d.
Example 1.5. E8 : We have ∆E8 = {±εi ±εj | i, j ∈ {1, . . . , 8} and i 6= j}t{ 21 (±ε1 ± ε2 ± · · · ± ε8 )}
even number of + signs
L8
with V = i=1 Rεi .
Let f (ε1 ) = 32, f (ε2 ) = 6, f (ε3 ) = 5, . . . , f (ε7 ) = 1, f (ε8 ) = 0.
Then, ∆+ = {εi ± εj | i < j} t { 12 (ε1 ± ε2 ± · · · ± ε8 )}. Also, Π = {ε2 − ε3 , ε3 − ε4 , . . . ε7 −
even number of + signs
ε8 , 12 (ε1 − ε2 − · · · − ε7 + ε8 )} and θ = ε1 + ε2 . We get the diagrams E8 and Ẽ8 in Figure 1e.
3
Exercise 1.1. E6 : We have:
Pick f (ε1 ) = 32, f (ε2 ) = 7, f (ε3 ) = 6, . . . , f (ε7 ) = 2, f (ε8 ) = 1. We can then check that Π =
{ε2 − ε3 , . . . , ε5 − ε6 , ε7 − ε8 , 12 (ε1 − ε2 − ε3 − ε4 + ε5 + ε6 − ε7 + ε8 )}, in particular, ε1 − ε2 =
1 1
2 (ε1 − ε2 − ε3 − ε4 + ε5 + ε6 − ε7 + ε8 ) + 2 (ε1 − ε2 + ε3 + ε4 − ε5 − ε6 + ε7 − ε8 ). Also, θ = ε1 − ε6
for obvious reasons. We obtain the Dynkin diagrams in Figure 1g.
E7 : Also, ∆E7 = {α ∈ ∆E8 | α1 + α2 + · · · + α8 = 0}, so:
Again, pick f (ε1 ) = 32, f (ε2 ) = 7, f (ε3 ) = 6, . . . , f (ε7 ) = 2, f (ε8 ) = 1. Then, Π = {ε2 − ε3 , ε3 −
ε4 , . . . , ε7 − ε8 , 12 (ε1 − ε2 − ε3 − ε4 − ε5 + ε6 + ε7 + ε8 )} and θ = ε1 − ε8 for obvious reasons. We
obtain the Dynkin diagrams in Figure 1f.
Theorem 2.1. A complete non-redundant list of connected Dynkin diagrams is the following:
D(Ar )(r ≥ 2), D(Br )(r ≥ 3), D(Cr )(r ≥ 1), D(Dr )(r ≥ 4), D(E6 ), D(E7 ), D(E8 ), D(F4 ) and
D(G2 ). (See Figure 1 for these Dynkin diagrams and their extended versions.)
Remark 2.2. Note that D(A1 ) = D(B1 ) = D(C1 ), D(B2 ) = D(C2 ) and D(D3 ) = D(A3 ).
Proof. We prove that there are no other Dynkin diagrams. To do this, we find all connected graphs
with connections of the 4 types depicted in Figure 2, such that the matrix of any subgraph has
a positive determinant, in particular any subgraph must be a Dynkin diagram. Equivalently, we
require that the determinant of all principal minors of the corresponding Cartan matrix are positive.
In particular, our graphs contain no extended Dynkin diagrams as induced subgraphs, since these
have determinant 0.
4
Figure 2: The four Dynkin diagram connection types, corresponding to the four types of 2 × 2
Cartan matrix minors.
Part 1. We first classify all simply-laced Dynkin diagrams D(A), i.e. diagrams using only
or
−
connections (which correspond to a symmetric Cartan matrix A). Such a diagram
contains no cycles, since otherwise it contains D(Ãr ). It has simple edges and it may or may not
contain branching vertices. If there are no branching vertices, we get D(Ar ). If there are two
branching vertices, the diagram contains D(D̃r ), which is not possible. If there is precisely one
branching vertex, it has at most 3 branches, since D(D̃4 ) is the 4-star in Figure 3a.
Therefore, it remains to consider the case when our graph D(A) is of the form Tp,q,r , p ≥ q ≥ r ≥ 2,
depicted in Figure 3b.
If r = 3, then Tp,q,r contains D(Ẽ6 ) = T3,3,3 , which is impossible, so r = 2.
If q = 2, we have type D.
Now, assume q ≥ 3. If q > 3, then D(A) contains D(Ẽ7 ) = T4,4,2 , which is impossible, so we are
left with the case Tp,3,2 with p ≥ 3.
The case p = 3 is E6 , the case p = 4 is E7 and the case p = 5 is E8 . The case p = 6 is Ẽ8 , so Tp,3,2
with p > 5 is impossible.
Part 2. We now classify all non-simply-laced diagrams D(A), i.e. those containing double- or
triple-edge connections (corresponding to a non-symmetric Cartan matrix A). This can be done
by computing many large determinants, but we would rather argue using the following result:
Exercise 2.1. Let A be an r × r matrix, and let B (resp. C) be the submatrices of A obtained by
removing the first row and column (resp. the first two rows and two columns). We have:
5
Figure 4: A loop with one double-connection.
a) If
−a 0 · · ·
2 0
−b
A=
0 ,
B
..
.
0
then, det(A) = 2 det(B) − ab det(C).
b) If
c1 −a1 0 ··· 0 −br
−b1 c2 −a2 0 0
. . . . .
.
0 −b2 c3 . . .
A= . ,
.. .. ..
0 . . −ar−2 0
..
0 . −br−2 cr−1 −ar−1
−ar 0 ··· 0 −br−1 cr
Qr
then, det(A − εe12 ) = det(A) − ε (b1 det(C) + i=2 ai ). In particular, if b1 > 0, ai > 0 for
i = 2, . . . , r, det(C) > 0 and ε > 0, then det(A − εe12 ) < det(A).
a) This is clear, just do cofactor expansion along the first row of A to obtain directly det(A) =
2 det(B) − ab det(C).
b) This is also clear. Consider the first row of A − εe12 , write (c1 , −a1 − ε, 0, 0, . . . , 0, −br ) =
(c1 , −a1 , 0, . . . , 0, −br ) + (0, −ε, 0, . . . , 0), use the multilinearity of det to write det(A − εe12 ) =
det(A)+ε det D, where D is obtained from A removing column 2 and row 1, and thenQ do cofac-
tor expansion along the first column of D to get det(D) = −b det(C)+(−1) r (−1)r−1
Q 1 i=2 ai =
−b1 − i=2 ai , where the second term comes directly from a lower triangular matrix.
Exercise 2.1 implies that in the case of a non-simply laced diagram D(A), there are no cycles
since then det(A) < det(Ãr ) = 0, a contradiction. For example, the diagram in Figure 4 has
A = Ã4 − E12 , so det A < 0.
6
Figure 5: Possible neighbors of G2 in a diagram.
Next, looking at the extended Dynkin diagrams, if D(A) contains G2 (Figure 1i), then it must
be exactly G2 by Exercise 2.1. Indeed, otherwise, using Figure 5 we see that there is a principal
submatrix of A of the form:
2 −a 0
0 2 −1
and M 00 = 2 .
M= −b 2 −1 with a, b > 0. Let M =
−3 2
0 −3 2
Hence, by Exercise 2.1, det(M ) = 2 det(M 0 ) − ab det(M 00 ) = 2(1 − ab) ≤ 0, a contradiction. (The
matrix of the second graph in the figure is actually AT , but all the determinants are the same.)
It remains to consider the case when D(A) has only simple or double connections. Looking at the
extended Dynkin diagrams, D(C̃r ) (in Figure 1c) cannot be a subdiagram. By Exercise 2.1, the
variants with flipped arrow directions also do not work. Indeed, using the notation of the exercise,
they are obtained from the Cartan matrix A of D(C̃r ) by replacing some of A, B or C by their
transposes, which does not change any of the determinants in the calculation. Thus, Exercise 2.1a
shows that each of these variants have also determinant 0.
Therefore, D(A) may contain only one double connection. But then, we cannot have branching
points, since D(B̃r ) contains a double edge and a branching point. So, the only remaining case is a
line with a left-right double edge, having p single edges to the left, and q single edges to the right.
If p = 0, we get D(Cr ). If q = 0, we get D(Br ). If p = q = 1, we get D(F4 ). The diagram D(F̃4 )
has p = 2, q = 1; its transpose has p = 1, q = 2. Hence, if p > 1 or q > 1, D(A) contains D(F̃4 ),
which is impossible.
We have now shown that any finite-dimensional simple Lie algebra yields one of a very restricted set
of Dynkin diagrams (and hence Cartan matrices). The next step in the classification of semisimple
Lie algebras will be to give a construction associating an abstract Cartan matrix to a Lie algebra,
and hence to prove that these four classes plus five exceptional algebras are in fact the only finite-
dimensional simple Lie algebras.
We depict the plan of our actions:
Cartan h (V = h∗ , ∆) choose f ∈ V ∗
A
g
s.s. Lie alg. −−−−→ R −−−−−−−−−−→ Π −→ ⊆ abstract
root system Cartan matrix
↑ Cartan
g(A) ←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−− matrix A
7
18.745 Introduction to Lie Algebras November 16, 2010
Proof of Theorem 19.1a) Let 𝔤 be a semisimple finite dimensional Lie algebra ⊕ ⊕over 𝔽. Choose
a Cartan subalgebra 𝔥 and consider the root space decomposition: 𝔤 = 𝔥 ( 𝛼∈Δ 𝔤𝛼 ), where
Δ ⊂ 𝔥∗ is the set of roots. Choose a linear function 𝑓 on 𝔥∗𝛼 which does not vanish on Δ, and let
Δ+ = {𝛼 ∈ Δ∣𝑓∐ (𝛼) > 0} and Δ− = {𝛼 ∈ Δ∣𝑓 (𝛼) < 0}.
Then Δ = Δ+ Δ− , where Δ− = −Δ+ . Let Π ⊆ Δ+ be the set of simple roots. We will prove
that nothing
⊕ depends on 𝑓 . ⊕
Let 𝔫+ = 𝛼∈Δ+ 𝔤𝛼 and 𝔫− = 𝛼∈Δ− 𝔤𝛼 , which are obviously subalgebras of 𝔤. Then we have
the triangular decomposition ⊕ ⊕
𝔤 = 𝔫− 𝔥 𝔫+
as vector spaces.
Exercise 19.2. Show that if 𝔤 = 𝔰𝔩𝑛 , 𝔰𝔬𝑛 or 𝔰𝔭𝑛 , choosing 𝔥 to be all diagonal matrices in 𝔤; then
𝔫+ (resp. 𝔫− ) consists of all strictly upper (resp. lower) triangular matrices in 𝔤.
𝑆𝑜𝑙𝑢𝑡𝑖𝑜𝑛. For 𝔰𝔩𝑛 , we know that the root space corresponding to 𝜖𝑖 − 𝜖𝑗 is 𝐸𝑖𝑗 where 𝐸𝑖𝑗 is 1 in the
(𝑖, 𝑗) slot and zero elsewhere. In the standard ordering of the roots, 𝜖𝑖 − 𝜖𝑗 is positive if and only if
𝑖 < 𝑗. Thus the positive roots spaces correspond to strictly upper triangular matrices and negative
root spaces correspond to strictly lower triangular matrices.
For 𝔰𝔬𝑛 , we have: ⊕⊕
𝔰𝔬𝑛 (𝔽) = 𝔥 ( 𝔽𝐹𝑖𝑗 ), (2)
𝑖,𝑗
1
where 𝐹𝑖𝑗 = 𝐸𝑖𝑗 − 𝐸𝑛+1−𝑗,𝑛+1−𝑖 and 𝔥 = (𝑎1 , , ...𝑎𝑟 , −𝑎𝑟 , ... − 𝑎1 ). Then the positive roots are:
𝜖𝑖 − 𝜖𝑗 (𝑖 < 𝑗) and 𝜖𝑖 + 𝜖𝑗 (𝑖 ∕= 𝑗), if 𝑛 = 2𝑟. Thinking of the four 𝑟 by 𝑟 blocks in 𝔰𝔬𝑛 , the
𝜖𝑖 + 𝜖𝑗 (𝑖 ∕= 𝑗) root spaces fill in the upper right hand block, and the 𝜖𝑖 − 𝜖𝑗 (𝑖 < 𝑗) fill in the piece
above the diagonal for the upper left half. Thus the positive root spaces span the set of upper
triangular matrices in 𝔰𝔬𝑛 (which are by definition antisymmetric with respect to the antidiagonal).
The odd case is similar.
For 𝔰𝔭𝑛 (𝔽), we have again the Cartan subalgebra 𝔥 = (𝑎1 , , ...𝑎𝑟 , −𝑎𝑟 , ... − 𝑎1 ). Here the positive
roots are ( 𝜖𝑖 + 𝜖𝑗 (where
) 𝑖 is permitted to equal 𝑗) and 𝜖𝑖 − 𝜖𝑗 where 𝑖 < 𝑗. We know that 𝔰𝔭𝑛 (𝔽) is
𝑎 𝑏
given by such that 𝑑 = −𝑎′ , 𝑏 = 𝑏′ , 𝑐 = 𝑐′ (where ′ denotes transposition with respect to
𝑐 𝑑
the antidiagonal). The positive root spaces corresponding to 𝜖𝑖 − 𝜖𝑗 are 𝐹𝑖𝑗 = 𝐸𝑖𝑗 − 𝐸𝑛+1−𝑗,𝑛+1−𝑖
for 1 ≤ 𝑖, 𝑗 ≤ 𝑟 (𝑖 ≤ 𝑗) - these fill in the part of 𝑎 above the main diagonal in 𝔰𝔭𝑛 (𝔽) and also
the part of 𝑑 above this main diagonal by 𝑑 = −𝑎′ . The root spaces corresponding to 𝜖𝑖 + 𝜖𝑗 are
𝐹𝑖𝑗 = 𝐸𝑖𝑗 + 𝐸𝑛+1−𝑗,𝑛+1−𝑖 for 1 ≤ 𝑖 ≤ 𝑟, 𝑟 + 1 ≤ 𝑗 ≤ 𝑛, and these spaces fill in the block 𝑏. Thus
the positive root spaces span the strictly upper triangular matrices.
Continuation of the proof of Theorem 19.1. Recall that for 𝛼 ∈ Δ we can choose 𝐸𝛼 ∈ 𝔤𝛼 and 𝐹−𝛼 ∈
𝔤−𝛼 such that 𝐾(𝐸𝛼 , 𝐹𝛼 ) = 2/𝐾(𝛼, 𝛼). Then 𝐻𝛼 = 2𝜈 −1 (𝛼)/𝐾(𝛼, 𝛼) so that 𝔽𝐸𝛼 + 𝔽𝐹−𝛼 + 𝔽𝐻𝛼
is isomorphic to 𝔰𝔩2 (𝔽). Now given Π = (𝛼1 , ..., 𝛼𝑟 ), set 𝐸𝑖 = 𝐸𝛼𝑖 ,𝐹𝑖 = 𝐹𝛼𝑖 ,𝐻𝑖 = 𝐻𝛼𝑖 for 1 ≤ 𝑖 ≤ 𝑟.
Then we have:
3) [𝐻𝑖 , 𝐹𝑗 ] = −𝑎𝑖𝑗 𝐹𝑗
4) [𝐸𝑖 , 𝐹𝑗 ] = 𝛿𝑖𝑗 𝐻𝑗 (since 𝛼𝑖 − 𝛼𝑗 is not a root when 𝑖 ∕= 𝑗 because the 𝛼𝑖 are simple)
Definition 19.1. These relations on the 𝐸𝑖 , 𝐹𝑖 , and 𝐻𝑖 are called the Chevalley relations. The
𝐸𝑖 , 𝐹𝑖 and 𝐻𝑖 are called the Chevallay generators.
Definition 19.2. Denote by 𝔤 ˜(𝐴) the Lie algebra with generators 𝐸𝑖 , 𝐹𝑖 and 𝐻𝑖 (1 ≤ 𝑖 ≤ 𝑟) subject
to the Chevalley relations. This Lie algebra is infinite dimensional if 𝑟 ≥ 1 but is closely related to
𝔤: we have the surjective homomorphism 𝜙 : 𝔤 ˜(𝐴) → 𝔤 sending 𝐸𝑖 to 𝐸𝑖 , 𝐹𝑖 to 𝐹𝑖 and 𝐻𝑖 to 𝐻𝑖 .
2
⊕ ⊕
˜+ =
b) 𝔫 𝛼∈ℚ+ ˜𝛼 and 𝔫
𝔤 ˜− = ˜−𝛼 where ℚ+ = ℤ+ Π/0
𝔤 𝛼∈ℚ+
⊕
˜(𝐴) then 𝐼 = (𝔥 ∩ 𝐼) ⊕ ( 𝛼∈ℚ+ ∪−ℚ+ 𝔤
c) If 𝐼 is an ideal in 𝔤 ˜𝛼 ∩ 𝐼).
˜(𝐴) contains a unique proper maximal ideal 𝐼(𝐴) provided 𝐷(𝐴) is connected.
d) 𝔤
Proof. For a), we will first show that 1) [𝐻𝑖 , 𝔫 ˜+ + 𝔫 ˜− + 𝔥] ⊂ 𝔫 ˜+ + 𝔫 ˜− + 𝔥, 2) [𝐸𝑖 , 𝔫˜+ + 𝔫
˜− + 𝔥] ⊂
˜+ +˜
𝔫 𝔫− +𝔥 and 3) [𝐹𝑖 , 𝔫 ˜+ +˜𝔫− +𝔥] ⊂ 𝔫 ˜+ +˜ 𝔫− +𝔥. Since a general element of 𝔤 ˜(𝐴) is an iterated bracket
of the 𝐸𝑖 , 𝐹𝑖 and 𝐻𝑖 , by the Jacobi identity, 1), 2) and 3) together imply that 𝔤 ˜(𝐴) ⊆ 𝔫 ˜+ + 𝔫
˜− + 𝔥.
Therefore 𝔫 ˜+ + 𝔫 ˜− + 𝔥 is a subalgebra of 𝔤 ˜(𝐴) and must coincide with 𝔤 ˜(𝐴) since it contains all
generators. To prove 1), observe that 𝔫 ˜+ is the span of commutators of the form [𝐸𝑖1 , ..., 𝐸𝑖𝑛 ] so that
˜+ since [𝐻𝑖 , 𝐸𝑖𝑘 ] ∈ 𝔫
[𝐻𝑖 , [𝐸𝑖1 , ...𝐸𝑖𝑛 ]] = [[𝐻𝑖 , 𝐸𝑖1 ], [𝐸𝑖2 , ...𝐸𝑖𝑛 ] + [𝐸𝑖1 , [𝐻𝑖 , 𝐸𝑖2 ], ...𝐸𝑖𝑛 ] ∈ 𝔫 ˜+ by the
Chevalley relations. The same is true for 𝔫 ˜− . This proves 1). For 2), first observe [𝐸𝑖1 , ...𝐸𝑖𝑛 ] ∈ 𝔫 ˜+ .
Also we can write [𝐹𝑖 , [𝐸𝑖1 , ...𝐸𝑖𝑛 ]] = [[𝐹𝑖 , 𝐸𝑖1 ], [𝐸𝑖2 , ...𝐸𝑖𝑛 ] + [𝐸𝑖1 , [𝐹𝑖 , 𝐸𝑖2 ], ...𝐸𝑖𝑛 ] + .... By the
Chevalley relations, [𝐹𝑖 , 𝐸𝑖𝑘 ] ∈ 𝔥, so each summand in this decomposition, appealing to the Jacobi
identity again, is in 𝔫 ˜+ , so [𝐹𝑖 , [𝐸𝑖1 , ...𝐸𝑖𝑛 ]] ∈ 𝔫
˜+ . 3) is established similarly. Thus we have shown
˜+ + 𝔫
𝔫 ˜ ˜
∑− + 𝔥 = 𝔤(𝐴). To show that the sum is direct, let ℎ ∈ 𝔥 satisfy 𝛼𝑖 (ℎ)∑= 1 for all 𝑖 (writing
ℎ = 𝑥𝑖 𝐻𝑖 , finding such an ℎ is equivalent to solving the linear system 𝑎𝑖𝑗 𝑥𝑖 = 1, which is
˜
possible since det 𝐴 ∕= 0). Then ad ℎ acts by positive eigenvalues on 𝔫+ , negative eigenvalues on
˜− , and by 0 on 𝔥. This proves the sum is direct.
𝔫
To prove d), we’ll need a weaker statement than c), namely, that 𝐼 = (𝐼 ∩ 𝔫 ˜− )⊕(𝐼 ∩ 𝔫 ˜
⊕ + )⊕(𝐼 ∩𝔥). To
prove this we will invoke an earlier lemma which said that for any 𝔥-module 𝑉 = 𝜆∈𝔥∗ 𝑉𝜆 and any
𝔥-invariant subspace 𝑈 , we have 𝑈 = ⊕𝜆 𝑈 ∩ 𝑉𝜆 . Apply this ˜
⊕ to 𝔥 = 𝔽ℎ and 𝜋 = ad and 𝑉 = 𝔤(𝐴).
Since 𝐼 is an invariant subspace under 𝔥, this gives 𝐼 = 𝜆 𝐼 ∩ 𝑉𝜆 where 𝑉𝜆 is the eigenspace of
˜(𝐴) with height 𝜆. Appealing to the lemma again, the sum of the positive eigenspaces in this
𝔤
decomposition is 𝔫 ˜+ ∩ 𝐼, the sum of the negative eigenspaces is 𝔫 ˜− ∩ 𝐼, and zero eigenspace is 𝔥 ∩ 𝐼
establishing 𝐼 = (𝐼 ∩ 𝔫 ˜− ) ⊕ (𝐼 ∩ 𝔫 ˜+ ) ⊕ (𝐼 ∩ 𝔥).
Let 𝐷(𝐴) be connected and 𝐼 be a proper ideal of 𝔤 ˜(𝐴). Then 𝐼 ∩ 𝔥 = 0. Indeed, if 𝑎 ∈ 𝐼 ∩ 𝔥 is
nonzero, then 𝛼𝑖 (𝑎) ∕= 0 for some 𝑖, so that [𝑎, 𝐸𝑖 ] = 𝛼𝑖 (𝑎)𝐸𝑖 ∕= 0, so 𝐸𝑖 ∈ 𝐼. Hence 𝐻𝑖 ∈ 𝐼 by the
Chevalley relations. Also by the Chevalley relations, 𝐸𝑗 and 𝐹𝑗 are contained in 𝐼 for all 𝑗 such that
𝑎𝑖𝑗 ∕= 0. Since 𝐷(𝐴) is connected, it follows that all 𝐸𝑗 and 𝐹𝑗 are contained in 𝐼. By Chevalley
relations, this implies the 𝐻𝑖 are all contained in 𝐼, which would mean 𝐼 = 𝔤 ˜(𝐴), contradicting the
properness of 𝐼. Thus we have the decomposition 𝐼 = (˜ 𝔫+ ∩ 𝐼) ⊕ (˜𝔫− ∩ 𝐼) for any proper ideal,
hence for the sum of all proper ideals, 𝐼(𝐴). Hence 𝐼(𝐴) is the unique proper, maximal ideal.
Proof. b) Since 𝔫 ˜+ is by definition the span of commutators of the form [𝐸𝑖1 , ..., 𝐸𝑖𝑛 ], then it is
spanned by the subspaces 𝔤 ˜𝛼 where 𝛼 runs through ℚ+ (since each 𝔤 ˜𝛼 consist precisely of terms
∑
[𝐸𝑖1 , ..., 𝐸𝑖𝑛 ] where, if 𝛼 = 𝑛𝑖 𝛼𝑖 , each 𝐸𝑖 appears 𝑛𝑖 times). To see that the sum is direct, first
observe that since det 𝐴 ∕= 0, for each 𝑗 we can find ℎ𝑗 ∈ 𝔥 such that 𝛼𝑖 (ℎ𝑗 ) = 𝛿𝑖𝑗 . By the Chevalley
relations, we have [ℎ𝑗 , 𝐸𝑖 ] = 𝛿𝑖𝑗 𝐸𝑖 . Thus given an iterated bracket in 𝔫 ˜+ , [𝐸𝑖1 , ..., 𝐸𝑖𝑛 ], we have
that [ℎ𝑗 , [𝐸𝑖1 , ..., 𝐸𝑖𝑛 ]] = 𝑐[𝐸∑
𝑖1 , ..., 𝐸𝑖𝑛 ] where 𝑐 is the number of times 𝑗 appears in the index set
𝑖1 , ..., 𝑖𝑛 . On each 𝔤˜𝛼 , 𝛼 = 𝑛𝑖 𝛼𝑖 , ad ℎ𝑗 acts by the constant 𝑛𝑗 . This means each 𝔤 ˜𝛼 is a joint
eigenspace for the ad ℎ𝑖 , so that sum is direct. The same is true for 𝔫 ˜− .
c) Given the decomposition we already proved, 𝐼 = (𝐼 ∩ ˜− ) ⊕ (𝐼 ∩ 𝔫
𝔫 ˜+ ) ⊕ (𝐼 ∩ 𝔥), it is enough to show
⊕
𝐼 ∩𝔫 ˜+ = ˜𝛼 . But 𝔫+ is an 𝔥-module under ad and 𝐼 ∩ 𝔫
𝐼 ∩𝔤 ˜+ is an invariant subspace since 𝐼
3
⊕ ⊕
is an ideal. Therefore by the lemma we had earlier, 𝐼 ∩ 𝔫˜+ = ˜𝛼 since by b), 𝔫+ =
𝐼 ∩𝔤 ˜𝛼 is
𝔤
a decomposition of 𝔫˜+ by weights of ad 𝔥 acting on 𝔫
˜+ .
Continuation of the proof of Theorem 19.1. By part d) of the lemma 𝐼(𝐴) is the unique proper
˜(𝐴). Now set 𝔤(𝐴) = 𝔤
maximal ideal in 𝔤 ˜(𝐴)/𝐼(𝐴). Since 𝔤 is simple, ker (𝜙 : 𝔤
˜(𝐴) → 𝔤) is a
maximal proper ideal. By Lemma 2d, ker 𝜙 = 𝐼(𝐴). Hence 𝜙 induces an isomorphism between
𝔤(𝐴) and 𝔤. This proves the ”if” part of the Theorem 19.1a. The ”only if” part will follow once we
show the independence of 𝐴 from the choice of 𝑓 .
So far, we have shown that if 𝔤 = 𝔰𝔩𝑛 , 𝔰𝔬𝑚 , 𝔰𝔭𝑛 then 𝔤 ∼
= 𝔤(𝐴) where 𝐴 is the Cartan matrix of 𝔤. The
only remaining simple, finite dimensional Lie algebras can be 𝔤(𝐴) where 𝐴 = 𝐸6 , 𝐸7 , 𝐸8 , 𝐹4 , 𝐺2 .
Hence to complete part b) of the theorem, we need to prove that the dimension of 𝔤(𝐴) is finite in
these five cases. We will prove this by exhibiting explicit constructions of these five.
4
18.745 Introduction to Lie Algebras November 18, 2010
First consider the simply-laced case: a symmetric Cartan matrix, root system ∆, root lattice
Q = Z∆, satisfyingL∆ = {α ∈ Q : (α, α) = 2}. We will construct g, a semisimple Lie algebra,
satisfying g = h⊕( α∈∆ FEα ). We will think of h as F⊗Z Q. The brackets should be the following:
1. [h, h0 ] = 0 ∀ h, h0 ∈ h,
2. [h, Eα ] = −[Eα , h] = (α, h)Eα ,
3. [Eα , E−α ] = −α for α ∈ ∆,
4. [Eα , Eβ ] = (α, β)Eα+β if α, β, α + β ∈ ∆,
5. [Eα , Eβ ] = 0 if α, β ∈ ∆, α + β ∈
/ ∆∪0
The problem is to find non-zero (α, β) ∈ F such that g with the four brackets above is a Lie algebra
(i.e. skew-symmetry, Jacobi identity). Then automatically g will be simple with the root system
∆, by our general criterion of simplicity.
Proposition 20.1. ∃ : Q × Q → ±1 with the following properties:
Proof. Choose a set Π of simple roots {α1 , · · · , αr } (so Π is a Z-basis of Q). For each pair i, j, make a
choice of (αi , αj ) and (αj , αi ) subject to the following constraints: (αi , αj )(αj , αi ) = (−1)(αi ,αj )
(for i 6= j) and (αi , αi ) = −1. Now extend bi-multiplicatively to all pairsP of elements in Q. Now
we can verify that the relation (α, α) = (−1)(α,α)/2 works, where α = i ki αi :
Y
(α, α) = (αi , αj )ki kj
i,j
2
Y Y
= (αi , αi )ki (αi , αj )(αj , αi )ki kj
i i<j
2 (αi ,αi ) (α,α)
P Y
= (−1) i ki 2 (−1)kj ki (αi ,αj ) = (−1) 2
i<j
Remark. (α, α) = −1 if α is a root. Further, if α, β ∈ Q, we can extend the identity (αi , αj )(αj , αi ) =
(−1)(αi ,αj ) extends bi-multiplicatively to give (α, β)(β, α) = (−1)(α,β) . Alternatively, note that
1 1
(α + β, α + β) = (α, α)(β, β)(α, β)(β, α) gives us the following: (−1) 2 (α,α)+ 2 (β,β)+(α,β) =
1 1
(−1) 2 (α,α)+ 2 (β,β) (α, β)(β, α), which implies (α, β)(β, α) = (−1)(α,β) .
1
Theorem 20.2. The brackets (1) − (4) above in g, with the form defined above, gives a simple
Lie algebra of finite dimension with root system (V = R ⊗Z Q, ∆).
Proof. The skew-symmetry follows by the Remark, since (α, β) = −(β, α) if α + β ∈ ∆. It now
suffices to check the Jacobi identity when a, b, c ∈ h or Eα (α ∈ ∆). If a ∈ Eα , b ∈ Eβ , c ∈ Eγ and
α+β +γ ∈ / ∆ ∪ 0, then the Jacobi identity trivially holds since all three terms are 0. For the same,
it trivially holds when a, b, c ∈ h. Otherwise we have the following cases:
Case 1: a, b ∈ h, c = Eα . Then we have [a, [b, c]] = (α, b)[a, Eα ] = (α, b)(α, a)Eα , [b, [c, a]] =
−[b, [a, Eα ]] = −(α, b)(α, a)Eα , [c, [a, b]] = 0, so they sum up to 0, as required.
Case 2: a ∈ h, b = Eα , c = Eβ . Then we have: [a, [b, c]] = (α+β, a)[b, c]; [b, [c, a]] = −(β, a)[b, c]; [c, [a, b]] =
−(α, a)[b, c], so they sum up to 0, as required.
Case 3: a = Eα , b = Eβ , c = Eγ , α + β + γ = 0. Then we have:
(β, −β)(β, −α)α − (−α, α)(−β, α)β + (α, β)(α + β) = (β, α)α + (β, α)β + (α, β)(α + β) = 0
Exercise 20.1. Show that there are remaining two cases when α + β + γ ∈ ∆ (i) α = −β (ii)
(α, β) = −1, (β, γ) = −1, (α, γ) = 0, and check the Jacobi identity in both of them.
Proof. (i) In this case, if (α, γ) = 0, then since g is simply laced, α + γ, α − γ ∈ / ∆, so we have that
[Eα , [E−α , Eγ ]] = 0, [E−α , [Eγ , Eα ]] = 0, [Eγ , [Eα , E−α ]] = [Eγ , −α] = (α, γ)Eγ = 0, so all three
terms are 0.
WLOG, the other case is when (α, γ) = −1 (since if (α, γ) = 1 switch α with −α), so since g is
simply laced, α + γ ∈ ∆, α − γ ∈ / ∆. Here we have that [Eα , [E−α , Eγ ]] = 0, [E−α , [Eγ , Eα ]] =
(γ, α)[E−α , Eα+γ ] = (γ, α)(−α, α)(−α, γ)Eγ while [Eγ , [Eα , E−α ]] = −[Eγ , α] = (α, γ)Eγ . So
it suffices to prove that (γ, α)(−α, α)(−α, γ) + (α, γ) = 0, which follows from the fact that
(γ, α)(α, γ) = (α, γ) = −1 in this case.
(ii) If no two of α, β, γ sum to 0, then using the fact that (α+β +γ, α+β +γ) = 2, one deduces that
(α, β) + (α, γ) + (β, γ) = −2, so since none of them can be −2 (if (α, β) = −2, α = −β), after re-
ordering (α, β) = −1, (β, γ) = −1, (α, γ) = 0. Then [Eα , [Eβ , Eγ ]] = (β, γ)(α, β)(α, γ)Eα+β+γ ;
[Eβ , [Eγ , Eα ]] = 0; [Eγ , [Eα , Eβ ]] = (α, β)(γ, α)(γ, β)Eα+β+γ . So it suffices to show that we
have: (β, γ)(α, β)(α, γ) + (α, β)(γ, α)(γ, β) = 0, which is true since (α, γ) = (γ, α), (β, γ) =
−(γ, β) in this case.
Above was the simply-laced case. For the non-simply laced case, note by the following two exercises
that each non-simply laced Lie algebra can be expressed as a sub-algebra of a simply-laced one.
More precisely, type Br ⊂ Dr+1 , Cr ⊂ A2r−1 , F4 ⊂ E6 , G2 ⊂ D4 . To see this, put the following
2
orientations on the Dynkin diagrams of E6 and D4 and define automorphisms of their respective
Dynkin diagram (σ2 for E6 and σ3 for D4 ) switching the indicated vertices:
Exercise 20.2. Check that the map Ei → Eσ(i) , Fi → Fσ(i) , Hi → Hσ(i) defines an automorphism
of g̃(A), and hence of g(A).
Proof. The relation [Hσ(i) , Hσ(j) ] = 0 holds trivially, as does the relation [Ei , Fj ] = σi,j Hj (since the
map σ is a bijection). It remains to check the relation [Hσ(i) , Eσ(j) ] = aij Eσ(j) and the analogous
relation for the F ’s; this is exactly equivalent to aij = aσ(i),σ(j) , which follows from the fact that
σ is an automorphism of the Dynkin diagram and preserves the inner products of its roots. Since
g̃(A) has a unique maximal ideal, it is invariant under σ, hence σ induces an automorphism of
g(A).
Proof. It is clear that the elements in question lie in the fixed point sub-algebra. In either case,
the first Chevalley relations (that the Cartan subalgebra is abelian) is trivial. The third Chevalley
relation (about the commutator of an E and an F ) follows from the third Chevelley relation for E6
and F2 , combined with the fact that in both sets {X1 +X5 , X2 +X4 , X3 , X6 } and {X1 +X3 +X4 , X2 },
the indices of different elements are distinct. The second Chevalley relation is equivalent to saying
that the in E6 , the four elements {X1 +X5 , X2 +X4 , X3 , X6 } correspond (in terms of inner products)
to the four simple roots of F4 ; and that in D4 , the two elements {X1 + X3 + X4 , X2 } correspond
(in terms of inner products) to the two simple roots of G2 . Both of these assertions are trivial to
verify.
By these exercises, we have homomorphisms g̃(F4 ) → g(E6 )σ2 , and g̃(G2 ) → g(D4 )σ3 . This proves
that g(F4 ) and g(G2 ) are finite dimensional, completing the proof. Soon we will show that in fact,
g(E6 )σ3 = g(F4 ), g(D4 )σ3 = g(G2 ).
Using this explicit construction of simply-laced algebras, we can easily construct a symmetric
invariant bilinear form (which is unique up to constant factor). We have a bilinear form (·, ·) on Q;
3
extend it by bilinearity to h. We let (h, Eα ) = 0, (Eα , Eβ ) = 0 if α + β 6= 0, (Eα , E−α ) = −1.
Proof. It is sufficient to prove that ([a, b], c) = (a, [b, c]) when a, b, c are each either in h or of the
form Eα . If a, b, c ∈ h clearly both sides are 0. If a, b ∈ h, c = Eα , then the LHS is clearly
0, while the RHS is also 0 since (h, Eα ) = 0; a similar situation happens if b, c ∈ h, a = Eα . If
a, c ∈ h, b = Eα , then both sides are again 0 since (h, Eα ) = 0. If a = Eα , b = Eβ , c ∈ h, the LHS
is 0 unless α + β = 0, and [b, c] ∈ FEβ , so the RHS is also 0 unless α + β = 0. If α + β = 0, then
the LHS is ([Eα , E−α ], c) = −(α, c), while the RHS is (Eα , [E−α , c]) = (α, c)(Eα , E−α ) = (α, c).
By symmetry, the case where c = Eα , b = Eβ , a ∈ h follows. If a = Eα , b ∈ h, c = Eβ , the
LHS is ([Eα , b], Eγ ) = −(α, b)(Eα,Eγ ) , and the RHS is (Eα , [b, Eγ ]) = (b, γ)(Eα , Eγ ). Clearly both
quantities are equal if α + γ = 0, and both quantities are 0 otherwise. The final case is when
a = Eα , b = Eβ , c = Eγ ; here both sides are clearly 0 unless α + β + γ = 0. If this quantity is 0,
then the LHS is −(α, β), while the RHS is (β, −α − β) = −(β, α) = (α, β), where in the last
equality we use the fact that α + β is a root.
Proof. It suffices to prove that ω([a, b]) = [ω(a), ω(b)] when a, b are either in h or of the form Eα . If
both a, b are in h, then both sides are 0. If a ∈ h, b = Eα , then the LHS is ω((α, a)Eα ) = (α, a)E−α ,
while the RHS is [−a, E−α ] = (α, a)E−α . Finally, if a = Eα , b = Eβ then both sides are 0 unless
α + β is a root; if it is a root then both sides are clearly (α, β)Eα+β since (α, β) = (−α, −β).
Definition 20.1. The fixed point set of ω is a Lie algebra over R, gC , called the compact form of
g.
Exercise 20.6. If g = sln (C), then gC = sun = {A ∈ sln (C)|A = −Āt }, and ω(A) = −Āt .
Proof. In this case, it is clear that Eαi −αj = Eij if i < j, and −Eij if i > j (this is to fulfill
the condition [Eα , E−α ] = −α). Then it is clear that the automorphism ωR sends A to −At , and
consequently ω sends A to −Āt , as required.
Proposition 20.3. The restriction of the invariant symmetric bilinear form (·, ·) from g to gc is
negative definite.
P P
Proof. We can write gc = ihR + α∈∆+ R(Eα + E−α ) + α∈∆+ iR(Eα − E−α ) and these 3 spaces
are orthogonal to each other. It remains to show that it is negative-definite one each space. This is
true because (ih, ih) = −(h, h) < 0; (Eα + E−α , Eα + E−α ) = −2 < 0, (i(Eα − E−α ), i(Eα − E−α )) =
−2 < 0.
4
Finally, the restriction of the invariant bilinear form (Killing form) from g(E6 ) or g(D4 ) to gσi is
non-degenerate, hence gσi is semi-simple and thus simple. To see this, just take gc ∩ gσi , where
g = E6 or D4 . Since the Killing form is negative definite on gc , it is negative definite on gc ∩ gσi ,
and thus also on its complexification gσi . It follows that E6σ2 = F4 , D4σ3 = G2 .
5
18.745 Introduction to Lie Algebras November 23th, 2010
Let V be a finite dimensional Euclidean vector space, i.e. a real vector space with a positive definite
symmetric bilinear form (. , .). Let a ∈ V be a nonzero vector and denote by ra the orthogonal
reflection relative to a, i.e. ra (a) = −a, ra (v) = v if (a, v) = 0.
2(a,v)
Formula 21.1. ra (v) = v − (a,a) a.
Exercise 21.1. Prove: (a) ra ∈ Ov (R), i.e. (ra u, ra v) = (u, v), u, v ∈ V . (b) ra = r−a and ra2 = 1.
(c) detra = −1. (d) If A ∈ Ov (R), then Ara A−1 = rA(a) .
Definition 21.1. Let (V, ∆) be a root system. Let W be the subgroup of Ov (R), generated by
all rα , where α ∈ ∆. The group W is called the Weyl group of the root system (V, ∆) (and of the
corresponding semisimple lie algebra g).
Proposition 21.1. (a) w(∆) = ∆ for all w ∈ W . (b) W is a finite subgroup of the group Ov (R).
Remark 21.1. (a) shows the string property of the root system implies that ∆ is W -invariant. One
can show converse is true: if we replace string property by W -invariance of ∆ and that 2(α,β)
(α,α) ∈ Z,
1
then we get an equivalent definition of a root system. One has to check only for the case dimV = 2.
(see Serre)
Fix f ∈ V ∗ which doesn’t vanish on ∆, let ∆+ = {α ∈ ∆|f (α) > 0} be the subset of positive roots
and let Π = {α1 , ..., αr } ∈ ∆+ be the set of simple roots (r = dimV ). Then the reflections si = rαi
are called simple reflections.
Theorem 21.2. (a) P ∆+ \{αi }Pis si -invariant. (b) if α ∈ ∆+ \Π, then there exists i such that
htsi (α) < htα (ht i ki αi = i ki ). (c) If α ∈ ∆+ \Π, then there exists a sequence of simple
reflections si1 , ...sik such that si1 , ...sik (α) ∈ Π and also sij , ...sik (α) ∈ ∆+ for all 1 ≤ j ≤ k. (d)
The group W is generated by simple reflections.
Proof. (a) For a positive root α, si (α) = α − nαi , where n is an integer. If α 6= αi , then all
coefficients in the decomposition of simple root remain positive, except possibly for the coefficient
of αi . But in a positive root α, all coefficients are nonnegative. Hence if α is not simple, it must
have positive coefficient in front of some simple roots other than αi (other positive integer multiples
of αi are not in ∆ by definition of root system), thus si (α) should remain positive.
Exercise 21.2. Compute Weyl Group for root systems of type Br , Cr , Dr . In particular show that
for Br and Cr they are isomorphic, but not isomorphic to Dr .
Proof. ∆Br = {±i ± j |1 ≤ i, j ≤ r, i 6= j} ∪ {±i } ⊂ Rr = ⊕ri=1 Ri , with (i , j ) = δi,j . It is easy
to see that, as in the example, ri −j switches i with j , ri switches the sign of i and the other rα
are generated by the previous two. Thus the Weyl Group consisting of all elements that permute r
elements as well as switch some of their signs. So it is the semidirect product group Zr2 o Sr . The
root system for Cr is just that for Br with ±i replaced by ±2i . So of course the reflections are
exactly the same, so Weyl Groups for Br and Cr are the same.
∆Dr = {±i ± j |1 ≤ i, j ≤ r, i 6= j} ⊂ Rr = ⊕ri=1 Ri , with (i , j ) = δi,j . Now ri −j acts as before,
and ri +j switches i to −j and j to −i . Thus Weyl Group is the group of all elements that
permute r elements as well as switching an even number of their signs. Thus it is Zr−1 2 o Sr , and
it is not isomorphic to that of Br and Cr .
2
Definition 21.2. Consider the open (in usual topology) set V −∪α∈∆ Tα in V (Tα is the hyperplane
perpendicular to α). The connected components of the set are called open chambers, the closures
are called closed chambers. C = {v ∈ V |(αi , v) > 0, αi ∈ Π} is called the fundemental chamber;
C = {v|(αi , v) ≥ 0, αi ∈ Π} is called closed fundemental chamber.
Proof. We first show that C is connected. Given any u, v ∈ C, we have (αi , u) > 0 for all αi ∈ Π.
Now given any α ∈ ∆+ , α is a linear comobination of αi with positive coefficients. Thus (α, u) > 0.
Similarly, (α, v) > 0. Now any point on the straight segment connecting u and v has the form
xu + (1 − x)v for some 0 ≤ x ≤ 1. Thus it is easy to see that (α, x) > 0. Thus u, v are in a single
component, i.e. an open chamber (we only considered α being a positive root. But if it is negative,
we get similarly that (α, u), (α, v) and (α, x) are all negative).
Now, take any element u0 not in C, then by definition (αi , u0 ) ≤ 0 for some αi ∈ Π. Then obviously
u0 and u ∈ C are not in the same component as they are separated by the hyperplane Tαi . Thus C
is an entire component, i.e. an open chamber.
Theorem 21.3. (a) W permutes all chambers transitively, i.e. for any two chambers Ci and
Cj there exists w ∈ W such that w(Ci ) = Cj . (b) Let ∆+ and ∆0+ be two subsets of positive
roots, defined by linear functions f and f 0 . Then there exists w ∈ W such that w(∆+ ) = ∆0+ . In
particular, the Cartan matrix of (V, ∆) is independent of the choice of f .
Proof. (a) Choose a segment connecting points in Ci , Cj , which doesn’t intersect ∪α,β∈∆ (Tα ∩ Tβ ).
Let’s move along the segment until we hit a hyperplane Tα . Then replace Ci by rα Ci . After finitely
many steps we hit the chamber Cj .
(b) a linear function f on V can be written as fa , where fa (v) = (a, v) for fixed a ∈ V . f doesn’t
vanish on ∆ means that a ∈ / ∪α∈∆ Tα , so f = fa with a in some open chamber. If we move a around
this chamber, the set ∆+ , defined by f remains unchanged. Hence all the subsets of positive roots
in ∆ are labelled by open chamber and if w(C) = C 0 , then for the corresponding sets of positive
roots ∆+ and ∆0+ we get that w(∆+ ) = ∆0+ .
Definition 21.3. Let s1 , ...sr be the simple reflections in W (they depend on choice of ∆+ ). Any
w ∈ W can be written as a product w = si1 ...sit due to Theorem 21.2(d). Such a decomposition with
minimal possible number of factors t is called a reduced decomposition and in this case t = l(w) is
called the length of w. Note: detw = (−1)l(w) since detsi = −1. E.g. l(1) = 0, l(si ) = 1, l(si , sj ) = 2
if i 6= j, but = 0 if i = j since s2i = 1.
Lemma 21.4. (Exchange Lemma) Suppose that si1 ...sit−1 (αit ) ∈ ∆− , αit ∈ Π. Then the expression
w = si1 ...sit is not reduced. More precisely, w = si1 ...sim−1 sim+1 ...sit−1 for some 1 ≤ m ≤ t − 1.
Proof. Consider the roots βk = sik+1 ...sit−1 (αit ) for 0 ≤ k ≤ t − 1. Then β0 ∈ ∆− and βt−1 = αit ∈
∆+ . Hence there exists 1 ≤ m ≤ t − 1 such that βm−1 ∈ ∆− , βm ∈ ∆+ . But βm−1 = sim βm . Hence
by Theorem 21.2(a), βm = αim ∈ Π. Let w = sim+1 ...sit−1 , by Ex.21.1(d), it follows wsit w−1 = sim ,
or wsit = sim w. The result follows by multiplying both sides by si1 ...sim on the left.
3
Proof. In the contrary case, w(C) = C for some w 6= 1, hence w(∆+ ) = ∆+ for ∆+ corresponding
to C. Take a reduced expression w = si1 ...sit , t ≥ 1. Then w(αit ) = si1 ...sit−1 (−αit ) ∈ ∆+ . Hence
si1 ...sit−1 (αit ) ∈ ∆− . Hence by Exchange Lemma, si1 ...sit is nonreduced. Contradiction.
Transitivity is proved in Theorem 21.3 (a).