1 Preliminaries On Fibrations: Definition 1. Let F: X Y Be A Continuous Map of Topological Spaces, y

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

S EMINAR ON L IE GROUPS , L IE ALGEBRAS , AND

REPRESENTATIONS

Lecture 8
The Borel-Weil theorem

1 Preliminaries on fibrations

Definition 1. Let f : X → Y be a continuous map of topological spaces, y0 ∈ Y , and


F = f −1 (y0 ) ⊂ X. We say that f is a locally trivial fibration with typical fiber F if
for every y ∈ Y there is an open neighborhood U ⊂ Y s.t. f −1 (U ) is homeomorphic
to U × F over U .
Lemma 2. If F → X → Y is a locally trivial fibration then

1. If γ : [0, 1] → Y is a path and x0 ∈ X is s.t. f (x0 ) = γ(0), there exists a path


γ̇ : [0, 1] → X lifting γ with γ(0) = x0 .

2. If H : [0, 1] × [0, 1] → Y is a homotopy and γ̇0 , γ̇1 , δ̇ : [0, 1] → X are paths


lifting H(s, 0), H(s, 1), and H(0, t), then there exists a homotopy Ḣ : [0, 1] ×
[0, 1] → X lifting X with Ḣ(s, 0) = γ̇0 (s), Ḣ(s, 1) = γ̇1 (s), and Ḣ(0, t) =
δ̇(t).

Proof. DRAW PICTURES!

First consider the path γ. Cover X by open nbds U as in the definition. By


compactness of [0, 1] there exists n s.t. γ([i/n, (i + 1)/n]) is contained in a single
U . By induction we may assume that n = 1, i.e. no subdivision necessary. If
h : U × F → p−1 (U ) is a homeomorphism, then γ̇(s) = h(γ(s), x0 ) is a path
lifting γ.

Now consider the homotopy H. Again cover X by open nbds as in the defini-
tion. By compactness of [0, 1]2 there exists n s.t. H([i/n, (i + 1)/n] × [j/n, (j +
1)/n]) is contained a single U . By the first part, for each j the path s 7→
H(s, j/n) lifts to a path γ̇j/n : [0, 1] → X. By induction it is enough to lift
H|[0,1]×[0,1/n] . Further induction reduces to lifting H[0,1/n]×[0,1/n] . Thus again
no subdivision is necessary, i.e. image of H contained in a single U . Again
h : U × F → p−1 (U ) homeomorphism. We want to lift H, and we are given
Ḣ restricted to the bottom, top, and left wall of the square, call their union
C ⊂ [0, 1]2 . We now let r : [0, 1]2 → C be an arbitrary retraction. Define
Ḣ(s, t) = h(H(s, t), p2 h−1 (Ḣ(r(s, t)))). Note p1 h−1 = p.
Corollary 3. If F → X → Y is a locally trivial fibration then we have an exact
sequence
π1 (F ) → π1 (X) → π1 (Y ).

Proof. Let γ̇ represent a loop in X based at x0 . Assume its image γ in Y


is nilhomotopic, via a homotopy H. We lift this homotopy to Ḣ subject to

1
Ḣ(s, 0) = x0 = Ḣ(s, 1) and Ḣ(0, t) = γ̇(t). Then t 7→ Ḣ(1, t) is a loop in X
based at x0 whose image in Y is the constant loop at y0 . Thus H(1, t) is a loop
in F .
Lemma 4. Let G be a Lie group, H ⊂ G a closed Lie subgroup. Given a subvec-
torspace X ⊂ g complementary to h, there exists an open neighborhood U ⊂ X of 0
s.t. the map
U × H → G, (u, h) 7→ expG (u)h
is a diffeomorphism onto an open neighborhood of H in G.

Proof. Write f : X × H → G for f (x, h) = expG (x)h. Let h0 ∈ H. We have the


differential df (x, h0 ) : Tx X × Th0 H → Texp(x)h0 G. We claim there exists open
nbd U ⊂ X of 0 s.t. df (u, h0 ) is invertible for all u ∈ U and h0 ∈ H.

For this, first note that df (0, 1) : X × h → g is the identity. Choose open U so
that df (u, 1) is invertible for all u ∈ U . If Rh0 is right translation by h0 , then
we have Rh0 ◦ f ◦ (idX × Rh−1 ) = f so we see df (u, h0 ) = dRh0 (exp(u)) ◦
0
df (u, 1) ◦ (idX × dRh−1 (h0 )) and the invertibility of df (u, 1) and dRh0 imply the
0
invertibility of df (u, h0 ).

Thus f restricts to a local diffeomorphism U × H → G. It is enough to show


that, upon possibly shrinking U , f is also injective.

Since H is a closed submanifold of G and exp : g → G is a local chart around


1 ∈ G we can shrink U and choose a nbd U 0 of 0 in h s.t. exp : U × U 0 → G is a
diffeomorphism onto its image, and exp(a + b) ∈ H if and only if a = 0.

If exp(a)h ∈ H with a ∈ U and h ∈ H, then exp(a) ∈ H, and hence a = 0.

We can further shrink U and U 0 to ensure that exp(a) exp(b) is a diffeomor-


phism on U × U 0 . Let now V 0 = exp(U 0 ) and set W = exp(U ) · V 0 . Note that by
the above we have W ∩ H = V 0 .

We shrink U one last time, to ensure that if a1 , a2 ∈ U then exp(−a2 ) exp(a1 ) ∈


W . Now say exp(a1 )b1 = exp(a2 )b2 with a1 , a2 ∈ U and b1 , b2 ∈ H. Then
exp(−a2 ) exp(a1 ) = b2 b−1 0
1 =: b ∈ H ∩ W = V . Then exp(a1 ) = exp(a2 )b, but
0
since this is a diffeomorphism U × V → W we have b = 1, i.e b1 = b2 , and
a1 = a2 .
Definition 5. Let X be a top space and G a top group. An action of G on X is a cts
map G × X → X satisfying the usual axioms of group action. If X is a manifold and
G a Lie group we require the map to be smooth.
Definition 6. A locally trivial fibration f : X → Y is called a principal H-bundle
if there is an action of H on X s.t. f is H-invariant, the action of H on each fiber is
simply transitive, and the local trivialization respect the action.

Example: Let Y = S1 and H = Z/2Z. We have the trivial principal H-


bundle X0 = Y × H and the Moebius strip X1 . Then Γcts (Y, X0 ) = Z/2Z
and Γcts (Y, X1 ) = ∅.
Corollary 7. The projection G → G/H is a principal H-bundle.

2
Proof. Let g ∈ G. Choose an open nbd U as in the lemma. Then (u, h) 7→
g exp(u)h makes U × H into an open nbd of the subset gH, and hence its pro-
jection to G/H is an open nbd of the point gH. This open nbd is diffeoomorphic
to U , and its inverse image in G is still g exp(U )H and hence diffeomorphic to
U × H.

2 Review of vector bundles

Definition 8. Let M be a real manifold. A real vector bundle on M consists of the


following data:

1. A real manifold V .
2. A surjective map of real manifolds p : V → M .
3. For each x ∈ M , the structure of an R-vs on p−1 (x).

subject the the following condition, called local triviality: For every x ∈ M there
exists an open neighborhood U of x and an diffeomorphism ϕ : U × Rk → p−1 (U ) s.t.

1. p(ϕ(x0 , v)) = x0 for all x0 ∈ U .


2. Rk → p−1 (x0 ), v 7→ ϕ(x0 , v) is an isomorphism of R-vs for all x0 ∈ U .

In the same way one defines a complex vector bundle on a real manifold, or
a real vector bundle on a complex manifold, or a complex vector bundle on a
complex manifold.
Definition 9. A section of V over U ⊂ M is a function s : U → V s.t. p(s(x)) = x.

We denote the set of sections over U by Γ(U, V ). It is clearly a real resp. com-
plex vector space, with vector space operations defined point-wise. Since both
M and V are manifolds, it makes sense to talk about continuous or smooth sec-
tions. When M , and hence also V , are complex manifolds, then one can also
talk about holomorphic sections.

Example: When M = S1 , we have the trivial line bundle V0 = M × R1 , and


the Moebius strip, which is a non-trivial line bundle V1 . Then Γcts (M, V0 ) =
Ccts (S1 , R), while Γcts (M, V1 ) = {f ∈ Ccts ([0, 1], R)| limx↑1 f (x) = − limx↓0 f (x)}.

Definition 10. Let M be a manifold with an action of a Lie group G. A G-equivariant


vector bundle on M is the data of a vector bundle p : V → M and an action of G on
V s.t. p is equivariant, and moreover g : p−1 (x) → p−1 (gx) is linear.

Remark: For each f : M1 → M2 and p2 : V2 → M2 one can construct the


pull-back bundle p1 : f ∗ V2 → M1 . Then a G-equivariant vector bundle is an
isomorphism of vector bundles a∗ V → p∗2 V , where a, p2 : G × M → M .

Given a G-equivariant vector bundle V , the space of sections Γ(M, V ) carries a


natural action of G, namely (gs)(x) = gs(g −1 x).

3
Definition 11. A G-equivariant vector bundle on M is called homogeneous if G
acts transitively on M .

Assume that M is connected and has a transitive G-action. Let m ∈ M and let
H be the stabilizer of m. If we have a complex homogeneous vector bundle V
we can consider its fiber Vm = p−1 (m). Then H acts linearly on the complex
vector space Vm , which then becomes a H-representation.

There is also an inverse construction. Given a H-representation Vm , we can


consider the manifold G × Vm . It receives a smooth action of H by h(g, v) =
(gh−1 , hv). This action is free, so the quotient G × Vm /H, which we will write
G ×H Vm , is a manifold. The projection G × Vm → G induces G ×H Vm →
G/H → M . We claim that this turns G ×H Vm → M into a complex vector
bundle whose fiber over m is Vm . First, it is a locally trivial fibration, because
G → G/H is a principal bundle with fiber H. Second, we introduce a vector
space structure on the fiber over gH by choosing g ∈ gH and using the bijection
v → (g, v). If we choose a different gh ∈ gH, then the bijection gets changed
to (gh, v) = (g, π(h)v), i.e. the linear structure gets translated by the map π(h),
which is linear, so the structure doesn’t change. Finally, we have an action of G
on G ×H Vm by left multiplication on the first component. Together we get:
Proposition 12. Assume that M is connected and let m ∈ M . Then the assignment
V 7→ p−1 (m) sets up an equivalence of categories between the complex homogeneous
G-equivariant vector bundles on M and the complex representations of Gm .

3 Statement of the Borel-Weil theorem

Let G be a connected compact semi-simple Lie group, T a maximal torus. We


have the lattices Q ⊂ A ⊂ P ⊂ it∗ . Choose a Weyl chamber and let A+ be
the intersection of A with its closure. The highest weight classification theorem
asserts that sending an irreducible representation (π, V ) to its highest weight
λ ∈ A+ is a bijection G
b → A+ .

The Borel-Weil theorem supplies an inverse to this bijection. In other words,


given λ ∈ A+ the theory privdes a construction of an irreducible representation
(π, V ) with highest weight λ.

We consider ξλ : T → C× and form the complex G-equivariant line bundle


Eλ := G ×T ξ−λ on G/T . Then we may consider the spaces Γcts (G/T, Eλ ) or
Γsmooth (G/T, Eλ ). These are complex representations of G. We will soon see
that they are much too big. What saves the day is a rather unexpected (from our
point of view) twist: The real manifold G/T has a canonical complex structure,
and so does Eλ . Thus we may also consider the space Γhol (G/T, Eλ ).

Theorem 13 (Borel-Weil). The representation Γhol (G/T, Eλ ) is irreducible with high-


est weight −w0 λ.

4 Induced representations

Let H ⊂ G be Lie groups, (π, V ) a finite-dimensional representation of H.

4
Definition 14. The representation continuously/smoothly induced from (π, V ) is the
complex vector space consisting of continuous/smooth functions f : G → V satisfying
f (hg) = π(h)f (g), with G acting by right translation.
Theorem 15 (Frobenius reciprocity). Let (π, V ) and (ρ, W ) be finite-dimensional
representations of H and G, respectively. Then we have the mutually inverse isomor-
phisms
S ∈ HomG (ρ, Indπ) ↔ HomH (ρ|H , π) 3 T
defined by T (w) = S(w)(1) and S(w)(g) = T (gw) for g ∈ G and w ∈ W . Here the
induction can be taken to be either continuous or smooth.

Proof. Given S: T is defined as the composition of S and the map Indπ → π


defined by evaluating at 1. Both of these are H-equivariant and linear, and then
so is T .

Given T : The linearity of S follows from that of T and the G-action. The fact
that S(w) lies in the space of the induced representation follows from the H-
equivariance of T . The G-equivariance of S comes from the definition of the
G-action on the induced space.

Note that we could have given the same proof if we replaced continuoius in-
duction by smooth induction, or even other types of functions. The point is
that the image of ρ in Indπ is small, so allowing more or less crazy functions,
while it does change Indπ, it won’t change the image of ρ in it.
Proposition 16. The map Indπ → Γ(G/H, G×H V ) sending f to the section s(g) =
(g, f (g −1 )) is well-defined and an isomorphism of G-representations.

Proof. First we check the section s is well-defined: s(gh) = (gh, f (h−1 g −1 )) =


(gh, π(h−1 )f (g −1 )) = (g, f (g −1 )). That it is indeed a section is clear. The
linearity of the map f 7→ s is also clear. The G-equivariance follows from
gs(g −1 x) = g(g −1 x, f (x−1 g)) = (x, (gf )(x−1 )).

Convesely, given a section s we define f (g) = gs(g −1 ). This is an element of


the fiber of G×H V over 1 ∈ G/H, which is naturally identified with V , namely
via the inclusion V → G × V → G ×H V given by v 7→ (1, v).

We can now see that Γ(G/T, G ×T ξ−λ ) = IndG T ξ−λ is in fact not a finite-
dimensional C-vs. For example, when λ = 0 then we get Csmooth (G/T ), which
is rather far from the trivial 1-dimensional representation with highest weight
0.

More generally, for an irrep ρ of G the space HomG (ρ, IndG


T ξ−λ ) = HomT (ρ, −λ)
has dimension equal to the multiplicity of −λ as a weight of ρ. So every ρ that
has −λ as a weight occurs in IndG T ξ−λ .

5 Using the complex structure

Theorem 17. There exists a natural embedding G → GC of G into a complex Lie


group G s.t. its differential Lie(G) → Lie(GC ) embeds Lie(G) as a real structure of
the complex vector space Lie(GC ), i.e. Lie(GC ) = Lie(G) ⊗R C.

5
We will use this theorem now, and discuss a proof later. For now just a quick
plausibility remark: One can take a faithful representation G → GLn (C). The
resulting embedding g → End(V ) of real Lie algebras extends to an embedding
gC → End(V ) of complex Lie algebras, and we can take GC to be the closed Lie
subgroup of GLn (C) with Lie algebra gC . It has a complex structure, because
the canonical atlas we discussed earlier, given by the exponential map, is now
complex. The question of naturality is however not addressed.

We have chosen a maximal torus T and a Weyl chamber in tC , i.e. a set of


positive roots Φ+ ⊂ Φ. Recall the triangluar decomposition
M
gC = n− ⊕ tC ⊕ n+ , n+ = gα .
α∈Φ+

Recall also the commutator relation [gα , gβ ] = gα+β . It implies that n+ is a Lie
subalgebra of gC . Thus we have the closed connected Lie subgroups TC and N
of GC . Since TC normalizes n+ , it normalizes N , and we set B = TC N .

Proposition 18. The inclusion G → GC induces a diffeomorphism G/T → GC /B.

Proof. Since T ⊂ B, the map is well-defined, and smooth. Its injectivity is


equivalent to G ∩ B = T . On the level of Lie-algebras this is obvious: If σ is
complex conjugation, then σ(n+ ) = n− , hence g ∩ b = (b ∩ σ(b))σ = tσC = t.
Pick g ∈ G ∩ B. The action of Ad(g) on gC preserves both g and b, hence their
intersection t, hence also T . Thus g ∈ N (T ). Since it preserves B, it preserves
the set of positive roots, hence the chosen Weyl chamber. Since the action of
the Weyl group on the set of Weyl chambers is simple, we conclude g ∈ T .

The differential of G/T → GC /B at 1 ∈ G/T is the map g/t → gC /b. For


each α ∈ Φ− fix non-zero Xα ∈ gα . If σ denotes complex conjugation, L then
Xα + σ(Xα ) ∈ g and this set forms a basis for the complement [ α∈Φ gα ] ∩ g
of t in g, hence projects to a basis of g/t. At the same time, Xα is a basis for the
complement n− of b in gC , so Xα + σ(Xα ) forms a basis for gC /b. Thus g/t →
gC /b is bijective. Since G/T is a homogeneous space for G, the differential
of G/T → GC /B at any g ∈ G/T is bijective. It follows that this map is an
everywhere local diffeomorphism.

In particular, its image is open. But it is also closed and connected, since G is
compact and connected. So the map is surjective. Now it is a bijective every-
where local diffeomorphism, so a diffeomorphism.

We have thus endowed G/T with the structure of a complex manifold. Our
next step is to recognize Eλ as a complex manifold as well, more precisely as a
complex line bundle on GC /B.

For this we shall investigate more carefully the structure of B.

The group TC is what is called a complex torus. Its structure is the following:
We have tC = t ⊕ it. This is an abelian complex Lie algebra, i.e. simply a
complex vector space. It is endowed with a real structure. For illustration,
consider first the case tC = C = R ⊕ iR. Applying exp : C → C× we have the
following

1. exp : R → R>0 is an isomorphism of Lie groups.

6
2. exp : iR → S1 is a universal covering of Lie groups with kernel 2πiZ.
3. We have the canonical decomposition C× = R>0 × S1 .

In general we have the same behavior: We know that T = exp(t), and we let
A = exp(it) ⊂ TC . Equivalently, A ⊂ TC is the connected Lie subgroup with
Lie algebra it ⊂ tC . Then
Lemma 19. 1. exp : it → A is an isomorphism of Lie groups.

2. exp : t → T is a universal covering with kernel 2πiA(T )∗ .


3. We have the canonical decomposition TC = T × A.

The proof is given by choosing a basis of t and applying the above observations.

Now we take our λ ∈ A(T ). The character ξλ : T → C× defined by ξλ (exp(H)) =


eλ(H) extends to ξλ : TC → C× by the same formula, where we now allow
H ∈ tC . Moreover, N is normalized by TC , and we have the canonical isomor-
phism B/N → TC , so we may view ξλ as a character of B. This allows us to
consider the line bundle
GC ×B Cξ−λ
on GC /B. The natural embedding G × C → GC × C gives a commutative
diagram
G ×T Cξ−λ / GC ×B Cξ
−λ

 
G/T / GC /B

where the horizontal arrows are isomorphisms, the top being an iso of vec-
tor bundles (bijectivity by diagram chase, smoothness obvious, local diffeo by
looking at tangent spaces).

We have thus made sense of Γhol (G/T, Eλ ).

6 Holomorphic sections of the induced representa-


tion

In the same way we see that G → GC gives an isomorphism

IndG GC
T ξ−λ → IndB ξ−λ .

The inverse of this isomorphism is given by simply restricting functions along


the inclusion G → GC . We want to understand which functions f ∈ IndG T ξ−λ
correspond to functions F ∈ IndG C
B −λξ that are holomorphic.

Recall that g acts on C ∞ (G) as right-invariant vector fields. That is, given X ∈ g
and f ∈ C ∞ (G) we define Xf (g) = df (g)(dRg (1)(−X)) = dt d
|t=0 f (e−tX g). If

we consider complex-valued smooth functions C (G, C) then this action up-
grades to an action of gC by linearity: For X, Y ∈ g we have [X + iY ]f (g) =
Xf (g) + iY f (g).

7
Lemma 20. A smooth f ∈ IndG T ξ−λ corresponds to a holomorphic function F ∈
GC
IndB ξ−λ if and only if Zf = 0 for all Z ∈ n+ .

Proof. Step 0: Preliminaries.


For a moment, recall that F : C → C is holomorphic at z if and only if it is
differentiable at z as a function R2 → R2 and its differential dF (z) ∈ EndR (R2 )
is the multiplication by a complex number. Now A ∈ EndR (R2 ) is the multi-
plication by a complex number if and only if it is C-linear. But, being R-linear,
this is equivalent to it commuting with multiplication by i, i.e. with
 
0 1
.
−1 0

This observation generalizes: F : Ck → Ck is holomorphic at z ∈ Ck if and only


if it is differentiable at z as a function F : R2k → R2k and dF (z) ∈ EndR (R2k )
commutes with multiplication by i.

This further generalizes to manifolds: Since holomorphy is a local condition,


we can work within a coordinate chart, and we see that F : M → N is holo-
morphic at p ∈ M if and only if it is smooth at p and the R-linear map dF (p) :
Tp M → TF (p) N commutes with multiplication by i.

Step1: A smooth function F ∈ IndG B ξ−λ is holomorphic if and only if, for all
C

Z = X + iY ∈ n+ we have XF + iY F = 0:
We want the R-linear dF (g) : Tg GC → C to commute with i, i.e. for all Z ∈ gC
we want
dF (g)(dRg )(iZ)) = idF (g)(dRg (Z)),
equivalently
d d
|t=0 F (e−itZ g) = i |t=0 F (e−tZ g).
dt dt
Since gC = n− ⊕ tC ⊕ n+ it is enough to test this equation with Z ∈ n− , Z ∈ tC ,
and Z ∈ n+ separately. If Z ∈ n+ then e−tZ ∈ N and F (e−tZ g) = F (g), so both
sides of the above equation are zero. If Z ∈ tC , then F (e−tZ g) = eλ(tZ) F (g) and
d d
the above equation reduces to dt |t=0 eitλ(Z) = i dt |t=0 etλ(Z) , which is true since
both sides equal iλ(Z).

Thus F is holomorphic if and only if the above equation holds for all Z ∈ n− .
For such Z, we write it as Z = X + iY with X, Y ∈ g and compute

dF (g)(dRg (iZ)) − idF (g)(dRg (Z)) = dF (g)(idRg (X)) − idF (g)(dRg (X))
− idF (g)(idRg (Y )) − dF (g)(dRg (Y ))

On the other hand we have X − iY = σ(Z) ∈ n+ . Then we know from the


above computation

0 = dF (g)(dRg (σ(Z))) = dF (g)(dRg (X)) − dF (g)(idRg (Y )).

Applying this to Z and iZ and plugging into the above we get

dF (g)(dRg (iZ)) − idF (g)(dRg (Z)) = −2i(dF (g)(dRg (X)) − idF (g)(dRg (Y )))

and we conclude that F is holomorphic if and only if XF − iY F = 0.

Step 3: A smooth f ∈ IndGT ξ−λ satisfies Xf + iY f = 0 for Z = X + iY ∈ n+ if


and only if the corresponding F ∈ IndG B ξ−λ does.
C

8
Recall that f = F |G . Thus XF + iY F = 0 immediately implies Xf + iY f = 0,
while conversely Xf + iY f = 0 implies XF (g) + iY F (g) = 0 for all g ∈ G. We
would like to amplify this to all g ∈ GC . But by Proposition we have GC = B·G,
so enough to take b ∈ B, g ∈ G and compute XF (bg) + iY F (bg).

For any X ∈ g we have


d d −1
XF (bg) = |t=0 F (e−tX bg) = |t=0 F (be−tAd(b) X g) = ξ−λ (b)[Ad(b)−1 X]F (g)
dt dt
and thus, writing X b := Ad(b)−1 X,
XF (bg) + iY F (bg) = ξ−λ (b) X b F (g) + iY b F (g) .


But if Z = X + iY ∈ n+ , then also Z b = X b + iY b ∈ n+ , so the right hand side


above is zero.

7 Proof of the Borel-Weil theorem

Let w0 be the longest element in the Weyl group with respect to the given Weyl
chamber.
Lemma 21. Let V be an irrep. If γ ∈ A+ is its highest weight, then −w0 γ is the
highest weight of V ∗ . Symbolically V (γ)∗ = V (−w0 γ).

Proof. α is a weight of V if and only if −α is a weight of V ∗ . Since the set of


weights is invariant under the Weyl group, −α is a weight if and only if −w0 α
is a weight. Since −w0 preserves the Weyl chamber, we have Now α ≤ β if and
only if −w0 α ≤ −w0 β. It follows that α is the highest weight of V if and only if
−w0 α is the highest weight of V ∗ .
Lemma 22. Fix a highest wieght vector v ∗ ∈ V (−w0 λ)∗ . The homomorphism
V (−w0 λ) → C∞ (G)fin , v 7→ fv∗ ,v
is an isomorphism between V (−w0 λ) and the subspace of functions f ∈ C∞ (G)fin
satisfying Zf = 0 for Z ∈ n+ and f (tg) = ξ−λ (t)f (g) for t ∈ T .

Proof. We have the isomorphism of G × G-representations


M
V ∗  V → C∞ (G, C)fin , (v ∗ , v) 7→ fv∗ ,v ,
(π,V )∈G
b

where fv∗ ,v is the matrix coefficient fv∗ ,v (g) = hv ∗ , gvi.

According to the highest weight classification, the sum is indexed by γ ∈ A+ ,


with V = V (γ), and V ∗ = V (γ)∗ = V (−w0 γ).

The action of X ∈ g by Xf (g) = dtd


|t=0 f (e−tX g) translates to the usual action

on V and the trival action on V . We are thus looking at
V (−w0 γ)n+ =0,T =λ  V (γ).
In the left factor, n+ = 0 isolates the highest weight space, which is the 1-
dimensional representation ξ−w0 γ of T . Thus T = λ makes the result zero
unless λ = −w0 γ, in which case it is 1-dimensional. The result on the left-hand
side is thus V (−w0 λ).

9
Lemma 23. Let D(G, C) denote the algebraic dual to C ∞ (G, C)fin . Then the map
Z
C ∞ (G, C) → D(G, C), f 7→ (φ 7→ f (g)φ(g)dg)
G

is an embedding of G × G-reps.

Proof. The G × G-equivariance follows from the left- and right-invarinace of


the Haar measure. Linearity is obvious. For injectivity, enough to show kernel
is zero.

For this, note first that the linear form determined by f is in fact defined on
all of C ∞ (G, C) and is continuous for the topology on C ∞ (G, C) determined by
the sup-norm:
Z Z Z
| f (g)φ1 (g)dg − f (g)φ1 (g)dg| ≤ ||φ1 − φ2 ||∞ f (g)dg.
G G G

According to Peter Weyl, C ∞ (G, C)fin is dense in C ∞ (G, C) with respect to the
sup norm, and injectivity follows.
Corollary 24. The embeddings C ∞ (G, C)fin → C ∞ (G, C) → D(G, C) restrict to
n =0,T =λ
isomorphisms C ∞ (G, C)fin+ → C ∞ (G, C)n+ =0,T =λ → D(G, C)n+ =0,T =λ .

Proof. We have M
V ∗  V = C∞ (G, C)fin
(π,V )∈G
b

and hence Y
V  V ∗ = D∞ (G, C),
(π,V )∈G
b

leading to
Y
[V  V ∗ ]n+ 1=0,T =λ = D∞ (G, C)n+ =0,T =λ .
(π,V )∈G
b

n =0,T =λ
All except one of the factors on the right vanish, showing that C ∞ (G, C)fin+ →
D(G, C)n+ =0,T =λ is an isomorphism.

8 Tannaka-Krein duality

We now consider the smooth real valued functions C ∞ (G) on G and the smooth
complex-valued functions C ∞ (G, C). We have C ∞ (G, C) = C ∞ (G) ⊗R C. Both
of these are representations of G × G and we can take their G-finite vectors, or
equivalently G × G-finite vectors. We have C ∞ (G, C)fin = C ∞ (G)fin ⊗R C. We
shall write M := C ∞ (G)fin for short. So far we know that M is an R-algebra
with 1. We will now endow M with the structure of a Hopf-algebra. We begin
with some preparatory lemmas.
Lemma 25. Let X be a set and V a finite-dimensional R-vs consisting of functions
X → R. There exists a basis f1 , . . . , fk for V and points x1 , . . . , xk ∈ X s.t. fi (xj ) =
δi,j .

10
Proof. We have the obvious map X → V ∗ and claim that its image is generat-
ing. Indeed, let U ⊂ V ∗ be the span of its image, and consider U ⊥ ⊂ V . An
element f ∈ U ⊥ has the property that it vanishes on all of U , in particular on
the image of X, i.e. f (x) = 0 for all x ∈ X. But this means f is the zero function
on X, i.e. f = 0 ∈ V . Thus U ⊥ = 0, hence U = V ∗ .

It follows that there are x1 , . . . , xk , with k = dim(V ), s.t. their image in V ∗ is a


basis. Let f1 , . . . , fk be the dual basis.
Lemma 26. Let G and H be two Lie groups. The map M (G)R M (H) → M (G×H)
sending a ⊗ b to (g, h) 7→ a(g) · b(h) is an isomorphism of algebras.

Proof. The analogous map C ∞ (G) × C ∞ (H) → C ∞ (G × H) is clearly bilinear so


induces a morphism of R-vs. It is moreover multiplicative, so an algebra mor-
phism. If a and b are contained in the fin-dim subspaces V and W respectively,
then a ⊗ b is contained in V ⊗ W and then a · b is contained in the image of
that subvector space, which is still finite-dimensional. This gives the desired
algebra homomorphism M (G) R M (H) → M (G × H).

We now wantP to show it is bijective. First, take p ∈ M (G) ⊗R M (H) and write
it out as p = fi ⊗ gi . Then the gi are all contained in a finite-dimensional
vector space V of functions on H, so by above lemma there is a basis s1 , . . . , sk
for that vector space and points y1 , . . . , yk ∈ H with si (yj ) = δi,jP. Each gi is a
0
linear combination of s1 , . . . , sk , so collecting terms we
P 0get p = fi ⊗ si . The
image of p in M (G × H) has the property p(x, y) = fi (x)si (y). If this were
the zero function, then plugging in y = y1 , . . . , yk we see that each fi0 = 0 and
hence p is itself zero. This proves injectivity.

For surjectivity, we take f ∈ M (G × H). Then the translates {x · f |x ∈ G} span


a finite-dimensional subspace, and so do their restrictions to H. Call this space
V . Apply above lemma to get (s1 ,P . . . , sk ) basis for V and y1 , . . . , yk ∈ H. Thus
for x ∈ G we have ((x, 1)f )H = ui (x)si for certain functions ui : G → R.
Then ui (x) = xf (1, yi ) = f (x, yi ), showing that ui is smooth P and spans a finite-
dimensional subspace, i.e. ui ∈ M (G). Then f (x, y) = ui (x)si (y) shows that
f is in the image of our algebra morphism, hence it is surjective.

We are now ready to give the Hopf-algebra structure on M (G). First of all, we
have the usual algebra structure given by pointwise operations. We encode it
as
u : R → M (G), u(r)(x) = r
for the unit (i.e. structure map), and

µ : M (G) ⊗R M (G) → M (G), µ(f ⊗ g)(x) = f (x)g(x),

for the multiplication map. The usual axioms of being a (commutative) algebra
can now be expressed in terms of these functions, for example if i : M (G) →
M (G) is the identity, then µ ◦ (u ⊗ i) = i = µ ◦ (i ⊗ u) says that u(1) is multi-
plicative identity.

Next we introduce a co-algebra structure. It consists of the co-multiplication

γ : M (G) → M (G) ⊗ M (G) = M (G × G), γ(f )(x, y) = f (x · y),

the antipode
η : M (G) → M (G), η(f )(x) = f (x−1 ),

11
and the co-unit
c : M (G) → R, c(f ) = f (1).
These three are algebra-homomorphisms. Furthermore, they give M (G) the
structure of an associative co-algebra. More precisely we have the properties
of co-associativity (i ⊗ γ) ◦ γ = (γ ⊗ i) ◦ γ and co-unit (c ⊗ i) ◦ γ = i = (i ⊗
c) ◦ γ. Moreover, the algebra and co-algebra structures are intertwined by the
following properties: c ◦ u = i, µ ◦ (η ⊗ i) ◦ γ = u ◦ c.

We have yet another piece of structure on M , given by the Haar integral. It


can be encoded as a linear form J : M → R. Its invariance is equivalent to
(J ⊗ i) ◦ γ = u ◦ J, and its positivity to J(f 2 ) > 0.

Definition 27. An R-algebra (M, µ, u, γ, c, η) satisfying these properties is called an


R-Hopf algebra (with antipode). The linear form J is called a gauge.
Proposition 28. Let (M, µ, u, γ, c, η) be any Hopf algebra over R.

1. The set GM (R) of R-algebra homomorphisms M → R becomes a group as


follows: Given x, y : M → R we have x ⊗ y : M ⊗ M → R and define
x · y = (x ⊗ y) ◦ γ, x−1 = x ◦ η, and identity c.

2. The following basis of neighborhoods of c makes GM (R) into a topological group:


N (f1 , . . . , fk , ), where f1 , . . . , fk ∈ M and  > 0, defined by

x ∈ N (f1 , . . . , fk , ) : ∀i |x(fi ) − c(fi )| < .

3. If M has a gauge then GM (R) is compact.

Proof. We leave it as an exercise that GM (R) is a topological group. Given


f ∈ M and x ∈ GM (R) we shall write x(f ) or f (x) interchangeably. Note that
x 7→ f (x) is a continuous map GM (R) → R.

Let us show that GM (R) is compact if there is a gauge J on M . We claim


that for a fixed f ∈ M the set {f (x)|x ∈ GM (R)} is a bounded subset of R.
Admitting Q this claim, let Tf be a closed interval in R containing this set, and
let T = f ∈M Tf , a compact topological space by Tychonoff’s theorem. For
each f ∈ M we have the map GM (R) → Tf sending x to f (x). Putting these
maps together for all f we obtain a continuous map α : GM (R) → T . This map
is injective: If α(x) = α(y) then x(f ) = y(f ) for all f , but then by definition
x = y. An element (tf ) ∈ T is in the image of α if and only if it satisfies the
properties necessary for an R-algebra homomorphism: tu(1) = 1, tf g = tf tg ,
tf +g = tf + tg , and trf = rtf . These conditions cut out a closed subset of T , so
α is a continuous bijection of GM (R) onto a closed, and hence compact, subset
of T . Finally, one checks easily using the definition of the topology of GM (R)
that this inverse of α is also continuous, so α is a homeomorphism, and hence
GM (R) compact.

We now come to proving the claim that for a fixed f ∈ M the set {f (x)|x ∈
GM (R)} is a bounded subset of R. We first observe that ((i ⊗ x)(γ(f )))x∈GM (R)
is a subset
P of M lying in a finite-dimensional subvector space V : Indeed, if
γ(f ) = ai bi then (i ⊗ x)(γ(f )) = ai · bi (x) so V is spanned by the ai . On any
finite-dimensional subspace of M we have the positive-definite scalar Pproduct
(a, b) = J(ab). Fix an ONB f1 , . . . , fk of V . Express (i ⊗ x)(γ(f )) = sj (x)fj

12
for some sj (x) ∈ R depending on x. We have J ◦ (i ⊗ x) ◦ γ = (J ⊗ x) ◦ γ =
x ◦ (J ⊗ i) ◦ γ = x ◦ u ◦ J = J and conclude
X
J(f 2 ) = J(i ⊗ x)(γ(f )2 ) = sj (x)2 .

Thus each sj is a bounded function of x. On theP other hand we have x = cx =


(c ⊗ x) ◦ γ = c ◦ (i ⊗ x) ◦ γ which gives x(f ) = sj (x)c(fj ). Thus x(f ) is also
bounded as a function of x.
Lemma 29. If M 0 ⊂ M is a G × G-invariant and dense subspace, then M 0 = M . If
MC0 ⊂ MC is a G × G-invariant and dense subspace, then MC0 = MC .

Proof. If M 0 ⊂ M is G × G-invariant and dense, then so is M 0 ⊗R C ⊂ MC .


Moreover M 0 = M if and only if ML 0
C = MC . So enough to treat the complex
case. Assume MC 6= MC . But MC = π π ∗  π. Thus there is at least one π s.t.
0

π ∗  π is not included in MC0 . But being an irreducible G × G-representation,


it must completely avoid MC0 . By Schur orthogonality, π ∗  π is orthogonal to
MC0 , but this contradicts its density in MC .
Theorem 30. The assignments G 7→ M (G) and M 7→ GM (R) are mutually inverse
equivalences of categories between the category of compact topological groups and the
category of reduced Hopf algebras with a gauge.

Proof. If M is a reduced Hopf-algebra with a gauge let GM (R) be the corre-


sponding compact group. For every f ∈ M let f ◦ : GM (R) be the func-
tion f ◦ (x) = x(f ). Then f 7→ f ◦ gives an algebra homomorphism M →
M (GM (R)). This homomorphism is injective due to the assumption that M
is reduced (so the algebra homomorphisms M → R separate the elements of
M ). It is dense in M (GM (R)) because it saparates the points of GM (R). But it
is also GM (R) × GM (R)-invariant, so above lemma implies M = M (GM (R)).

Conversely let G be a compact topological group. For x ∈ G we obtain x◦ ∈


GM (G) (R) =: S by x◦ (f ) = f (x). Then x 7→ x◦ is a continuous group homo
G → S. It is injective because M (G) separates points. Since G is compact,
it is thus identified with the image of this injection. We want to show that
this image is all of S. We apply the previous part of the proof to obtain an
isomorphism M (G) → M (S) =: M . We claim that the normalized Haar in-
tegrals IG , IS : M → R corresponding to G and S coincide. Granting that,
since M is uniformly dense in C(S) we see that IG and IS coincide on C(S).
If G ( S we could find a non-negative 0 6= f ∈ C(S) s.t. f |G = 0, implying
IG (f ) = 0 6= IS (f ).

We now prove the claim that IG = IS . Let f ∈ M . It is contained in a finite-


dimensional S-submodule of M , which we can decompose into a direct sum
V0 ⊕ V1 , with V0 being trivial, and V1S = {0}. Then f = f0 + f1 accordingly,
where f0 ∈ V0 is a constant function, and f1 ∈ V1 . Now IS restricts trivially to
V1 – it factors through the S-coinvariants, but the norm map is an isomorphism
[V1 ]S → V1S = {0}. It follows IS (f ) = IS (f0 ) = f0 (1). We now claim that V1G =
{0}. Indeed, if not, we could decompose further f1 = f2 + f3 with f2 constant
on G. The isomorphism M (G) → M (S) implies that since the restriction of
f2 to G is constant, in fact f2 is constant on all of S. But it belongs to V1 ,
contradiction. As before we conclude that IG (f ) = IG (f0 ) = f0 (1), implying
IS (f ) = IG (f ).

13
Remark: It is known that a Hopf-algebra over a field of characteristic zero
is reduced in the sense that it has no nilpotent elements. This is a result of
Pierre Cartier. It is furthermore known that the nilradical and Jacobson radi-
cal of a finitely-generated algebra over a field coincide. This means that for a
finitely-generated Hopf-algebra over a field, the algebra homomorphisms into
the algebraic closure of the base field separate the elements. But it is still un-
clear to me that the algebra homomorphisms into the base field alone, without
considering extensions, will also separate elements.
Proposition 31. Let G be a compact connected topological group. TFAE

1. G is a Lie group.
2. G has a faithful finite-dimensional representation.
3. M is finitely generated.

Proof. We have seen that a compact Lie group has a finite-dimensional faith-
ful representation. Conversely, if G is a compact topological group and V is a
finite-dimensional faithful representation, then we can use the closed embed-
ding G → GL(V ) to endow G with the structure of a Lie group, using the fact
that a closed subgroup of a Lie group is a Lie subgroup.

Assume now that the compact topological group G has a finite-dimensional


faithful rep V . It is unitary. Choosing an orthonormal basis of V we obtain an
injective homomorphism π : G → Un for some n. Let ai,j (g) be the coefficients
of π(g) for g ∈ G. These clearly lie in MC , being matrix coefficients of V . Let
d(g) be the inverse of the determinant of π(g). Then g 7→ d(g) is a 1-dimensional
representation of G, so d ∈ M .

Let MC0 be the subalgebra of MC generated by 1, d and ai,j . We claim that MC0
is dense in C(G, C). Since it is also G × G-invariant, the above lemma would
then imply that it equals MC . By the Stone-Weierstrass theorem it is enough to
show that MC0 separates points in G and is closed under complex conjugation.
Separating points is immediate from the injectivity of π. For complex conjuga-
tion, note that since π(g) is a unitary matrix, its complex conjugate is equal to
its inverse, which by Cramer’s rule has coefficients made out of polynomials
in ai,j and d. Now Stone-Weierstrass applies and we see that MC0 is dense in
C(G, C).

We have thus shown that MC is finitely generated. Let f1 , . . . , fk be generators.


Write fj = gj + igk+j with gj = Re(fj ) ∈ M and gk+j = Im(fj ) ∈ PM . Then
g1 , . . . , g2k generate the R-algebra M : Indeed, if x ∈ M , then x = I zI g I for
i2k
some zI ∈ C, where I = (i1 , . . . , i2k ) are multi-indices and g I = g1i1 . . . g2k . But
1
P zI +σ(zI )
x = 2 (x + σ(x)), so x = I 2 gI .

Finally, assume that the compact topological group G has finitely-generated


M . Let V be a finite-dimensional subspace generating M as an algebra. There
exists a finite-dimensional and G-invariant subspace W containing V . Then W
is a representation of G. If g acts trivially on all of W , then it acts trivially on all
of M , i.e. f (xg) = f (x) for all x ∈ G and f ∈ M , in particular f (g) = f (1) for
all f ∈ M , but this contradicts the fact that M separates points.

Remark: We want to remark here more generally that given a locally compact
topological group G and a nbd U of 1 there exist an open subgroup G0 ⊂ G

14
and a compact normal subgroup K ⊂ G contained in U s.t. G0 /K has the
structure of a Lie group. This is the Gleason-Yamabe theorem. In particular, if
G is connected then G0 = G, and if further G has no small subgroups, then G
itself is a Lie group. Thus: Every connected locally compact topological group
without small subgroups is a Lie group.

Moreover, the Baker-Campbell-Hausdorff formula implies that a topological


group has at most one differentiable structure that makes it Lie.

See Terry Tao’s book on Hilbert’s Fifth Problem for more details.

9 Complexification of compact Lie groups

Let G be a compact Lie group. Then M is a finitely-generated reduced Hopf-


algebra over R. Consider MC = M ⊗R C. This is a finitely-generated reduced
Hopf-algebra over C. We can consider GC = HomC−alg (MC , C). This is a com-
plex affine algebraic group – the maximal spectrum of MC .

10 For later

Lemma 32. The map exp : n+ → N is a diffeomorphism.

Proof. We may choose a faithful representation G → GL(V ) so that tC is di-


agonal and n+ is strictly upper triangular. The compatibility of exponentials
implies the same for TC and N . This implies the injectivity of exp (look at pow-
ers of strictly upper triangular matrices).

Next, we know that exp is a local diffeomorphism everywhere from Corollary


29 of Lecture 3, because the points where its differential is not invertible are
those X for which ad(X) has an eigenvalue in 2πi(Z r {0}). But for all X ∈ n+ ,
ad(X) is nilpotent on gC , hence also on n+ .

For surjectivity, recall Theorem 6 of Lecture 3. It applies to the domain where


exp has invertible differential, which in our case is all of n+ . Thus for all X, Y ∈
n+ there is µ(X, Y ) ∈ n+ s.t. exp(X) exp(Y ) = exp(µ(X, Y )). This shows that
exp(n+ ) is a subgroup of N . But it also generates N , so must equal to N .

Thus we have a smooth bijective map that is everywhere a local diffeomor-


phism, hence a diffeomorphism.

15

You might also like