Lecture 8 - Preservation of The Jordan Decomposition and Levi's Theorem
Lecture 8 - Preservation of The Jordan Decomposition and Levi's Theorem
Lecture 8 - Preservation of The Jordan Decomposition and Levi's Theorem
Oct 2, 2012
Theorem 1.1 Assume g ⊂ gl(V ) is a semisimple linear Lie algebra. Given any x ∈ g, the
abstract and usual Jordan decompositions coincide.
Pf. Assume xs = Ps (x) is the usual Jordan decomposition. Because T r(xs ) = T r(x) = 0,
we know xs ∈ sl(V ). Consider the subalgebra g0 ⊆ gl(V ) generated by g and xs . Since g0
is a g-module, it is completely reducible by Weyl’s theorem, so there is some submodule M
with
g0 = g ⊕ M. (1)
To see that M = {0}, let y ∈ M , and note that [y, g] = 0. Thus y acts by constant
multiplication on every irreducible g-submodule of W ⊆ V . Finally we prove that T r|W (y) =
0, forcing y to be zero.. To see this, first note that T r|W (xs ) = T r|W (x) = 0. The final
equality is due to the fact that x ∈ [g, g] so is a linear combination of brackets of operators
that preserve W , so has trace zero. Then since g0 is generated, via brackets and linear
combinations, of elements of g along with xs , all of which preserve W , we have T r|W (y) = 0
for all y ∈ g0 .
Pf. We have
m
M
g = gi (2)
i=1
1
P
where the gi are simple. We can write x = xi uniquely where xi ∈ gi . Each xi has its own
Jordan decomposition xi = xi,s + xi,n where xi,s , xi,n ∈ gi . By the mutual commutativity
of the gi , we therefore have that
m
X m
X
x = xi,s + xi,n (3)
i=1 i=1
0 0
m
X m
X
0
x = xi,s + xi,n . (4)
i=1 i=1
2 Levi’s Theorem
Theorem 2.1 Assume g is a finite dimensional Lie algebra over an algebraically closed
field of characteristic 0, and let r be its radical. Then there is a semisimple subalgebra l ⊆ g
so that l ∩ r = {0} and g = r + l.
Pf.1 The proof goes by imitation with the proof of Weyl’s theorem.
Step I. We first reduce to the case that r does not properly contain any ideal of g. So
assuming the theorem is true when r contains no ideal that is also an ideal in g, then let
a ⊂ l be such an ideal that is maximal in r. We have
where l ⊂ g has the properties that [l, r] ⊆ a, l ∩ r ⊆ a, and l/a is semisimple so Rad(l) = a.
One can then make an induction argument on the dimension of g; since dim(l) < dim(g)
we have a splitting l = a + l0 . Therefore
g = r + a + l0 = r + l0 (7)
since a ⊂ r.
1 This proof was taken from Fulton-Harris.
2
Step II. We next reduce to the case that, in addition to the conclusion from step I, r
is also abelian. If [r, r] 6= {0}, then actually a = [r, r] = r ∩ [g, g] is an ideal in g properly
contained in r, so we are back in the situation of Step I.
Step III. In addition to the conclusions from steps I and II, we can reduce to the case
that r 6= Z(g). For if that were the case, the adjoint action passes to a natural action of the
semisimple algebra g/Z(g) on g. Since Z(g) ⊆ g is a submodule, Weyl’s theorem gives the
existence of a complimentary submodule l and
g = l ⊕ Z(g) (8)
where both l and of course Z(g) are ideals.
Step IV. Due to steps I-III, we assume r is abelian, has no ideals that are also ideals
of g, and that r 6= Z(g).
3
where [x, ϕ(y)] ∈ [r, r] = {0} and [x, y] ∈ r so ϕ([x, y]) = c[x, y]. Thus also r.c ⊆ A.
Now consider the g-modules B/A ⊆ C/A. These both have the property of being
cancelled by the radical r, so these are in fact g/r-modules. Since g/r is semisimple, Weyl’s
theorem provides a complimentary submodule D/A (necessarily 1-dimensional) so
Clearly g/r acts trivially on D/A, so there is a non-zero map ϕ ∈ C with ϕ|r = 1 but
g.ϕ ∈ A. Given this ϕ, set
l = x ∈ g x.ϕ = 0 . (16)
0 = x.ϕ = − ad x (17)
Then x is central, so Fx is an ideal of g, so we must have r = Fx. But then r = Z(g), which
was assumed not to be the case. Thus l ∩ r = {0}.
Finally we show that g = l ⊕ r in the vector space sense. Because g.ϕ ∈ A, if y ∈ g then
there is some yr ∈ r with y.ϕ = ad yr . Because yr .ϕ = −ad yr we have have (y + yr ).ϕ = 0
so y = (y + yr ) − yr where y + yr ∈ l and −yr ∈ r.
A reductive Lie algebra is any Lie algebra that can be written in the form
g = g1 ⊕ · · · ⊕ gn (18)
where each gi is irreducible under ad g (meaning each is an ideal in g with no proper sub-
ideals). It is easy to see that each gi must be simple. Note that, unless [Rad(g), Rad(g)] =
{0} then the ideal Rad(g) is never an irreducible g-module, and also that unless the Levi
decomposition
g = l + Rad(g) (19)
is a Lie algebra direct sum (meaning l is also an ideal), then g is not reductive. Thus g is
reductive if and only if Rad(g) = Z(g).