Solutions to exercises on symplectic geometry

Peter Hochs

Contents
1 Symplectic linear algebra 2

2 Flows of vector fields 4

3 Lie derivatives 9

4 Poisson brackets 12

5 Examples of symplectic manifolds 15

5.1 Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Cotangent bundles . . . . . . . . . . . . . . . . . . . . . . . . 16
5.3 Smooth projective manifolds . . . . . . . . . . . . . . . . . . 18

6 Darboux theorem 20

7 Groups and actions 27

7.1 Proper maps . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.2 Proper and free actions . . . . . . . . . . . . . . . . . . . . . 28

8 Lie groups 32
8.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8.2 The exponential map . . . . . . . . . . . . . . . . . . . . . . . 35
8.3 The Lie bracket . . . . . . . . . . . . . . . . . . . . . . . . . . 40

9 Infinitesimal actions 43

10 The slice lemma 46

11 Hamiltonian group actions 53

12 Symplectic reduction 56

1
13 Cotangent bundles 61
13.1 Momentum map . . . . . . . . . . . . . . . . . . . . . . . . . 61
13.2 Symplectic reduction . . . . . . . . . . . . . . . . . . . . . . . 64

14 Coadjoint orbits 68

15 Smooth projective manifolds 72

1 Symplectic linear algebra

In Exercises 1.1–1.4, we work out the details of the proof of Lemma 1.2 in
the lecture notes. This will be used in the proof of the Darboux theorem
in Section 6, and in particular shows that any symplectic vector space is
even-dimensional.
Let (V, ω) be a symplectic vector space. Fix a nonzero e1 ∈ V .

Exercise 1.1. Show that there is an f1 ∈ V such that

ω(e1 , f1 ) = 1.

Proof. If for all v ∈ V ,

ω(e1 , v) = 0,
then e1 would equal zero, by nondegeneracy of ω. Hence there is a v for
which this expression is nonzero. Take
v
f1 := .
ω(e1 , v)

Exercise 1.2. Show that the restriction of ω to

U = Re1 + Rf1

is nondegenerate.

Proof. Let v = ae1 + bf1 ∈ U be given. If ω(v, w) = 0 for all w ∈ U , then in

particular

a = ω(v, f1 ) = 0;
b = −ω(v, e1 ) = 0.

So v = 0, and the map U → U ∗ defined by ω is injective.

2
 We now work out some details of the third paragraph on page 5 of the
lecture notes.

Exercise 1.3. Let b be a nondegenerate bilinear form a finite-dimensional

vector space V . Let U ⊂ V be a subspace such that b is nondegenerate on
U . Prove that
V = U ⊕ U b,
and that b is nondegenerate on U b . To be more precise, show that

(a) dim U + dim U b = dim V ;

Hint: Apply the rank-nullity theorem to the map

ψ : V → U∗

given by
ψ(v) = b(v, −)|U .

(b) U ∩ U b = {0};

(c) b is nondegenerate on U b .

Proof. (a) By the rank-nullity theorem, one has

dim im ψ + dim ker ψ = dim V.

Since b induces an isomorphism V ∼

= V ∗ , the map ψ is surjective. So

dim im ψ = dim U ∗ = dim U.

Also, ker ψ = U b , so the claim follows.

(b) Let v ∈ U ∩ U b . Then b(v, w) = 0 for all w ∈ U , so v = 0 by

nondegeneracy of b|U .

(c) Let v ∈ U b , and suppose b(v, w) = 0 for all w ∈ U b . Since V = U + U b ,

it follows that b(v, w) = 0 for all w ∈ V . By nondegeneracy of b, we
therefore have v = 0.

Exercise 1.4. Use induction on the dimension of V to conclude from the

preceding exercises that Lemma 1.2 is true.

3
Proof. The vectors e1 and f1 in Exercise 1.1 are linearly independent. In-
deed, if f1 were a multiple of e1 , then ω(e1 , f1 ) = 0. So the dimension of V
must be at least 2. If dim V = 2, then {e1 , f1 } is a basis with the required
properties.
Next, suppose dim V = n > 2, and that the claim holds for all vector
spaces of dimension smaller than n. Choose vectors e1 , f1 as in Exercise
1.1. Set U := Re1 + Rf1 . Then by Exercise 1.3, the restriction of ω to
U ω is nondegenerate. The dimension of U ω is n − 2, so by the induction
hypothesis, there is a basis {e2 , . . . en , f2 , . . . fn } of U ω that has the required
properties.
Since, by Exercise 1.3, one has V = U ⊕U ω , the vectors {e1 , . . . , en , f1 , . . . , fn }
form a basis of V . And because

ω(e1 , v) = ω(f1 , v) = 0

for all v ∈ U ω , the required properties of the basis {e1 , . . . en , f1 , . . . fn }

follow.

2 Flows of vector fields

Exercise 2.1. (a) Let v ∈ X(R2 ) be given by

v(x, y) = (1, 0).

What is the flow of v?

(b) The same question for

v(x, y) = (y, 0).

Proof. (a) The flow φt : R2 → R2 along v over time t is determined by


d 
φt (x, y) = v(φs (x, y)) = (1, 0).
dt t=s

So
φt (x, y) = (t + a, b),
for certain numbers a, b depending on x and y. Because φ0 is the
identity map, we must have a = x and b = y. So

φt (x, y) = (t + x, y).

4
 (b) Split up the flow φt : R2 → R2 along v over time t into coordinates:


φt (x, t) = φt,1 (x, y), φt,2 (x, y) .

Then for all x, y, s:


d 


φt,1 (x, y), φt,2 (x, y) = v(φs (x, y)) = φs,2 (x, y), 0 .
dt t=s


So 
d 
φt,2 (x, y) = 0,
dt t=s
while φ0,2 (x, y) = y. Hence φt,2 (x, y) = y for all t.
Now 
d 
φt,1 (x, y) = φs,2 (x, y) = y,
dt t=s
so φt,1 (x, y) = yt + a for some number a (which may depend on x and
y). Since φ0,1 (x, y) = x, we conclude that a = x, so

φt (x, y) = (yt + x, y).

Exercise 2.2. Let v ∈ X(R2 ) be given by1

      
x −y 0 −1 x
v = = .
y x 1 0 y

(a) Draw a picture of the vector field v, and guess what the flow curves2
of v are.

(b) Let φt be the flow of v over time t. Prove that

    
x cos(t) − sin(t) x
φt = .
y sin(t) cos(t) y

What does this mean for the shape of the flow curves?
1
In this exercise we denote points in R2 by column vectors, so we can apply matrices
to them from the left.
2
A flow curve of a vector field v on a manifold M is a curve γ of the form γ(t) = φt (m),
for an m ∈ M .

5
 (c) Let f ∈ C ∞ (R2 ), and consider the two-form

ω := f dy ∧ dx ∈ Ω2 (R2 ).
 
x
Consider a point m = on the x-axis, and two vectors
0
   
v1 w1
v= ,w = ∈ Tm R2 = R2 .
v2 w2

Compute that
 
x cos(t)
(φ∗t ω)m (v, w) =f (v2 w1 − v1 w2 ).
x sin(t)

Proof. (a) The flow curves look like circles.

(b) We compute
      
d  cos(t) − sin(t) x d  x cos(t) − y sin(t)
= 
dt t=s sin(t) cos(t) y dt x sin(t) + y cos(t)
 t=s 
−x sin(s) − y cos(s)
=
x cos(s) − y sin(s)
  
− sin(s) − cos(s) x
=
cos(s) − sin(s) y
   
0 −1 cos(s) − sin(s) x
=
1 0 sin(s) cos(s) y
   
0 −1 x
= φs
1 0 y
  
x
= v φs
y

Also,  
cos(t) − sin(t)
sin(t) cos(t)
is the identity matrix for t = 0.
The flow curves are indeed circles, since for all t and x,
   
x cos(t)
φt =x
0 sin(t)

6
 (c) By definition of pullbacks of differential forms, we have
(φ∗t ω)m (v, w) = ωφt (m) Tm φt (v), Tm φt (w) .

Since φt : R2 → R2 is a linear map, it is equal to its own tangent map.

I.e.
  
cos(t) − sin(t) v1
Tm φt (v) =
sin(t) cos(t) v2
 
v1 cos(t) − v2 sin(t)
=
v1 sin(t) + v2 cos(t)
   
∂ ∂
= (v1 cos(t) − v2 sin(t)) + (v1 sin(t) + v2 cos(t)) ,
∂x m ∂y m
and similarly for w. Therefore,


ωφt (m) Tm φt (v), Tm φt (w) =
    
∂ ∂
f (φt (m))dy∧dx (v1 cos(t) − v2 sin(t)) + (v1 sin(t) + v2 cos(t)) ,
∂x m ∂y m
    
∂ ∂
(w1 cos(t) − w2 sin(t)) + (w1 sin(t) + w2 cos(t))
∂x m ∂y m
 
x cos(t)
=f ·
x sin(t)


(v1 sin(t)+v2 cos(t))(w1 cos(t)−w2 sin(t))−(v1 cos(t)−v2 sin(t))(w1 sin(t)+w2 cos(t))
Working out the latter expression, using the fact that cos2 (t)+sin2 (t) =
1, leads to the answer.

The next exercise will later be an illustration for the exponential map of
a Lie group.
Exercise 2.3. Let Mn (R) be the vector space of real n × n matrices. For
any X ∈ Mn (R), consider the vector field X
e on Mn (R) given by
eg := g · X
X ∈ Mn (R) = Tg Mn (R),

for g ∈ Mn (R). Prove that the flow φt along X

e over time t is given by
∞ j j
X t X
φt (g) = getX := g , (1)
j!
j=0

for all g ∈ Mn (R). (Here X j is the jth matrix power of X.)

7
 Hint: You may use the following result in analysis. If (fj )∞j=0 is a se-
quence of smooth maps
fj : R → Rn ,
and if the series ∞
P P∞ 0
P∞ the series j=0 fj con-
j=0 fj converges pointwise, while
verges uniformly, then the function defined by j=0 fj is differentiable, and
 ∞ ∞
d  X X
fj = fj0 (s). (2)
dt t=s
j=0 j=0

In your solution, you may use the fact that the series (1) converges
uniformly onP compact subsets of R2 . Since (2) is a local statement, it is
enough that ∞ 0
j=0 fj converges uniformly on compact sets.

Proof. First, note that for all j ≥ 1,

tj X j sj−1 X j sj−1 X j−1

d 
= j = X.
dt t=s j! j! (j − 1)!
Hence the series
 
∞ ∞ j−1 j−1 ∞  j j ∞ j j
tj X j
 
X d  X s X X s X X s X
= X= X = X
dt t=s j! (j − 1)! j! j!
j=0 j=1 j=0 j=0

converges uniformly. In the last step, we have used the fact that right
multiplication by X is linear and continuous, and hence commutes with the
infinite sum.
The result mentioned in the hint therefore implies that
∞ ∞ j j
d  X tj X j
 
d  tX
X s X
e = = X. = esX X
dt t=s
 dt t=s
 j! j!
j=0 j=0

We conclude that for any g ∈ Mn (R),

 
d  d 
ge = g  etX = gesX X = X
tX egesX .
dt t=s dt t=s
Since in addition, ge0X = g, we see that (1) indeed satisfies the definition of
the flow along X.
e

Exercise 2.4. Consider the vector field v ∈ X(R2 ) given by

v(x, y) = (x2 + 1)(y 2 + 1), y ,

for (x, y) ∈ R2 . Let φt be the flow along v over time t.

8
 (a) Show that
φt (0, y) = tan((y 2 + 1)t), y ,

for t, y ∈ R for which the right hand side is defined.

(b) Conclude that there is no single t 6= 0 such that φt is defined as a map
defined on all of R2 .
Proof. (a) We see directly that φ0 (0, y) = (tan(0), y) = (0, y) for all y ∈ R.
Furthermore, we compute for all y, s ∈ R,

d 
φt (0, y) = (y 2 + 1) 1 + tan2 ((y 2 + 1)s) , 0



dt t=s



= v φs (0, y) .

(b) The flow along v starting at a point of the form (0, y) cannot be ex-
tended continuously beyond the domain
 
−π π
t∈ , .
2(y 2 + 1) 2(y 2 + 1)
Any t 6= 0 falls outside this domain, for a certain y ∈ R.

3 Lie derivatives
Exercise 3.1. Let M be a smooth manifold, f ∈ C ∞ (M ) a smooth function,
and v a vector field on M . Prove that
Lv (f ) = v(f ).
Proof. Let φt be the flow along v over time t. At every point m ∈ M , one
has

d 
(Lv f )(m) =  (φ∗t f )(m)
dt t=0

d 
=  f (φt (m))
dt t=0
  
d 
= dm f φt (m)
dt t=0
= dm f (vm )
= v(f )(m).

9
 The following fact is useful for computing the derivative of a function
with respect to a real parameter that plays two different roles in the defini-
tion of the function. It is often called a chain rule, although the chain rule
does not necessarily have to be applied to prove it.

Exercise 3.2. Let M be a smooth manifold,3 and let

F : R2 → M

be a smooth map. Prove that

  
d  d  d 
F (t, t) =  F (t, 0) +  F (0, t).
dt t=0 dt t=0 dt t=0

Proof. Consider the curve

∆ : R → R2
given by ∆(t) = (t, t). Its derivative at any point is the vector (1, 1).
We compute
 
d  d 
F (t, t) =  F (∆(t))
dt t=0 dt t=0
= T(0,0) F (∆0 (0))
= T(0,0) F (1, 1)
= T(0,0) F (1, 0) + T(0,0) F (0, 1),

by linearity of the tangent map T(0,0) F . Now

   
d  d 
T(0,0) F (1, 0) = T(0,0) F (t, 0) =  F (t, 0),
dt t=0 dt t=0
d

and similarly T(0,0) F (0, 1) = dt t=0
F (0, t). This finishes the argument.

Exercise 3.3. Let M be a smooth manifold, φ : M → M a diffeomorphism.

Let α ∈ Ωp (M ) be a p-form, and let v ∈ X(M ) be a vector field. Prove that

φ∗ (iv α) = iφ∗ v φ∗ α.
3
This exercise can be applied in the solutions of some other exercises,Vwhere the man-
ifold M is often a vector bundle over another manifold N , such as M = T ∗ N .

10
Proof. For all m ∈ M and v1 , . . . , vp−1 ∈ Tm M , one has

φ∗ (iv α) m (v1 , . . . , vp−1 ) = (iv α)φ(m) Tm φ(v1 ), . . . , Tm φ(vp−1 )

= αφ(m) vφ(m) , Tm φ(v1 ), . . . , Tm φ(vp−1 ) . (3)

Now
vφ(m) = Tm φ Tφ(m) φ−1 (vφ(m) ) = Tm φ (φ∗ v)m .

Hence (3) equals

αφ(m) Tm φ φ∗ v , Tm φ(v1 ), . . . , Tm φ(vp−1 ) = iφ∗ v φ∗ α m (v1 , . . . , vp−1 ).

m

Exercise 3.4. Let M be a smooth manifold. For a vector field v on M ,

denote the Lie derivatives of differential forms by v by Lv . Let iv be contrac-
tion of differential forms by v. Prove that for all differential forms α ∈ Ω(M )
and all vector fields v, w on M ,

i[v,w] α = Lv iw α − iw Lv α. (4)

Deduce from this equality that for all functions f ∈ C ∞ (M ),

[v, w](f ) = v(w(f )) − w(v(f )).

Hint: See the paragraph below the proof of Theorem 2.1, on page 17 of
the lecture notes. Explain all steps in the argument given there in detail.

Proof. Let vector fields v, w on M be given, and let α ∈ Ωp (M ) be a p-form.

Denote the flow along v over time t by φvt , and similarly for w.
By Exercise 3.3, one has

d 
Lv (iw α) =  (φvt )∗ (iw α)
dt t=0

d 
=  i(φvt )∗ w (φvt )∗ α.
dt t=0
By Exercise 3.2, this equals
 
d  d 
i v ∗ α +  iw (φvt )∗ α.
dt t=0 (φt ) w dt t=0
Since at every point m, the map
^p ^p−1
∗ ∗
(iw )m = iwm : Tm M→ Tm M

11
 V ∗
is linear, it commutes with derivatives of curves in Tm M . Hence
   
d  d 
iw (φvt )∗ α = iw (φv )∗ α = iw Lv α.
dt t=0
 dt t=0 t

Similarly, at every point m ∈ M , and for all v1 , . . . , vp−1 ∈ Tm M ,

   
d  d  v ∗


i(φ v )∗ w α (v 1 , . . . , v p−1 ) = α m ((φ t ) w) m , v 1 , . . . , v p−1
dt t=0 t m dt t=0
  
d  v ∗
= αm ((φ t ) w) m 1, v , . . . , v p−1
dt t=0


= (i[v,w] α)m v1 , . . . , vp−1 ,

where we have used linearity of αm in its first entry. We finally conclude

that
Lv (iw α) = iw Lv α + i[v,w] α,
as claimed.
The claim for functions follows by taking α = df for a function f ∈
C ∞ (M ), and using Exercise 3.1 and the definition

u(f )(m) = dm f (um ) = (iu df )(m)

for any vector field u and point m ∈ M .

4 Poisson brackets
Exercise 4.1. Let (M, ω) be a symplectic manifold, and let {−, −} be the
Poisson bracket defined by ω. Prove that for all functions f, g ∈ C ∞ (M ),
one has
[vf , vg ] = v{f,g} .

Hint: This exercise is Theorem 4.2 in the lecture notes. A short proof is
given just above that theorem, so it is only necessary to fill in some details,
and briefly mention why all equalities are true.

Proof. Let f, g ∈ C ∞ (M ) be given. Since Lie derivatives commute with the

exterior derivative, one has

d{f, g} = d(vf (g)) = d(Lvf (g)) = Lvf (dg) = −Lvf ivg ω.

12
By Exercise 3.4, the latter expression equals

−i[vf ,vg ] ω − ivg Lvf ω.

By Cartan’s formula for the Lie derivative,

Lvf ω = d(ivf ω) + ivf (dω).

Since dω = 0 and also d(ivf ω) = −ddf = 0, we conclude that

−Lvf ivg ω = −i[vf ,vg ] ω.

Therefore,
d{f, g} = −i[vf ,vg ] ω,
which means that
[vf , vg ] = v{f,g} ,
by definition of Hamiltonian vector fields.

A Lie bracket on a real vector space V is a map

V ×V →V; (v, w) 7→ [v, w],

such that

1. (bilinearity) for all u, v, w ∈ V , λ ∈ R:

[u + λv, w] = [u, w] + λ[v, w];

2. (anti-symmetry) for all v, w ∈ V ,

[v, w] = −[w, v];

3. (Jacobi-identity) for all u, v, w ∈ V ,

[[u, v], w] + [[v, w], u] + [[w, u], v] = 0.

Exercise 4.2. Let (M, ω) be a symplectic manifold. Prove that

(a) the Poisson bracket {−, −} defined by ω is Lie bracket on C ∞ (M );

(b) for f, g, h ∈ C ∞ (M ), one has

{f, gh} = g{f, h} + {f, g}h.

13
Remark 4.1. • In general, a Poisson bracket on a smooth manifold M
is defined as a Lie bracket on C ∞ (M ) with the property in part (b).
So Exercise 4.2 shows that the bracket defined by the symplectic form
is in fact a Poisson bracket in this sense.
• Exercise 4.1 is exactly the statement that the map f 7→ vf is a Lie
algebra homomorphism from C ∞ (M ) to the Lie algebra of vector fields
on M .
Proof. (a) Bilinearity of the Poisson bracket follows from bilinearity of ω
and from the linear dependence of df on the function f .
Anti-symmetry of the Poisson bracket follows from the equality
{f, g} = vf (g) = dg(vf ) = ω(vf , vg ),
and anti-symmetry of ω.
The Jacobi identity for the Poisson bracket follows from Exercise 4.1.
Indeed, for f, g, h ∈ C ∞ (M ), that exercise implies that
{{f, g}, h} = v{f,g} (h)
= vf (vg (h)) − vg (vf (h))
= {f, {g, h}} − {g, {f, h}}.

(b) We will use the product rule: for all f, g ∈ C ∞ (M )

d(gh) = gdh + hdg. (5)
This may be used without proof (it is a standard property of the
exterior derivative), but for completeness’ sake, we give a proof here.
First of all, note that the tangent map of the multiplication map
µ : R2 → R; µ(x, y) = xy,
is given by

d 
T(x,y) µ(v, w) =  (x + tv)(y + tw) = yv + xw,
dt t=0
for all x, y, v, w ∈ R. By the chain rule, one therefore has for all
f, g ∈ C ∞ (M ) and m ∈ M ,
dm (f g) = dm (µ ◦ (f, g))
= T(f (m),g(m)) µ ◦ Tm (f, g)
= T(f (m),g(m)) µ ◦ (dm f, dm g)
= g(m)dm f + f (m)dm g.

14
 This proves (5).
Pairing the one-form d(f g) with a vector field v, one therefore obtains
v(f g) = f v(g) + gv(f ).

This in particular implies that for all f, g, h ∈ C ∞ (M ),

{f, gh} = vf (gh) = gvf (h) + hvf (g) = g{f, h} + {f, g}h.

5 Examples of symplectic manifolds

5.1 Vector spaces
Exercise 5.1. Consider the situation of Exercise 1.4. Now view V as a
smooth manifold, equipped with coordinates xj , ξj : V → R, such that for
all v ∈ V ,
Xn
v= xj (v)ej + ξj (v)fj .
j=1
We view the symplectic form ω on the vector space V as a symplectic form
on the manifold V by identifying all tangent spaces to V with V . Prove that
Xn
ω= dξj ∧ dxj .
j=1

So in particular, ω = dθ is exact, hence closed, for

Xn
θ= ξj ∧ dxj .
j=1
∂ ∂
Proof. The vector fields ∂xj and ∂ξj associated to the coordinates xj and ξj
are given by    
∂ ∂
= ej ; = fj ,
∂xj v ∂ξj v
for all v ∈ V . Hence for all k, l, and all v ∈ V , one has
    
∂ ∂
ωv , = ω(ek , el ) = 0;
∂xk v ∂xk v
    
∂ ∂
ωv , = ω(fk , fl ) = 0;
∂ξk v ∂ξl v
    
∂ ∂
ωv , = ω(fk , el ) = 1.
∂ξk v ∂xl v

15
Also, one has
 
n     
X ∂ ∂
 dξj ∧ dxj  ,
∂xk v ∂xl v
j=1
v
n            
X ∂ ∂ ∂ ∂
= dv ξj dv xj −dv ξj dv xj
∂xk v ∂xl v ∂xl v ∂xk v
j=1

= 0,

since      
∂ ∂
dv ξj = dv ξj = 0.
∂xk v ∂xl v
Similarly,  
n     
X ∂ ∂
 dξj ∧ dxj  , = 0.
∂ξk v ∂ξl v
j=1
v
Finally,
 
n     
X ∂ ∂
 dξj ∧ dxj  ,
∂ξk v ∂xl v
j=1
v
n            
X ∂ ∂ ∂ ∂
= dv ξj dv xj − dv ξj dv xj
∂ξk v ∂xl v ∂xl v ∂ξk v
j=1
n
X
= δjk δjl − 0
j=1

= δkl .

So the forms ω and nj=1 dξj ∧ dxj take the same values on the vector fields
P
∂ ∂
∂xk and ∂ξk . Since these span all tangent spaces, and the two forms are
bilinear, this shows that they are equal.

5.2 Cotangent bundles

Exercise 5.2. Prove the local equality
n
X
θ= ξj dxj
j=1

16
above Definition 3.6 in the lecture notes. Conclude that ω = dθ is indeed a
symplectic form on M = T ∗ N .

Proof. Note that in general, any one-form β ∈ Ω1 (T ∗ N ) locally equals

X
β|T ∗ U = aj dxj + bj dξj ,
j

for functions aj , bj ∈ C ∞ (T ∗ N ), where U ⊂ N is an open set where the

coordinates x are defined. The claim is that for β = θ,
 
∂
aj = θ = ξj ;
∂xj
  (6)
∂
bj = θ = 0.
∂ξj

To prove these relations, we will use the facts that, as vector fields on U ,
∂ ∂
Tπ ◦ = ;
∂xj ∂xj
(7)
∂
Tπ ◦ = 0.
∂ξj

Let us prove these last two equalities. Consider the local coordinates

(x, ξ) : T ∗ U → V ⊂ R2d

(with d the dimension of N and V the image of (x, ξ)). Note that

π ◦ (x, ξ)−1 = x−1 ◦ p : V → U,

with p : R2d → Rd projection onto the first d coordinates. Hence

T π ◦ (Dx, Dξ)−1 = Dx−1 ◦ T p : T V → T U.

Here we denote the derivative (tangent map) of a chart by the letter D.

Now on T ∗ U ,
∂
= (Dx, Dξ)−1 (ej );
∂xj
∂
= (Dx, Dξ)−1 (ed+j );
∂ξj

17
where {e1 , . . . , e2d } is the standard basis of R2d . On U , one has

∂
= Dx−1 (ej ) = Dx−1 ◦ T p(ej ),
∂xj

so (7) follows.
Finally, by (7), we have for all η ∈ T ∗ U ,
     
∂ ∂ ∂
θη = η Tπ ◦ =η = ξj (η),
∂xj ∂xj ∂xj

and    
∂ ∂
θη = η Tπ ◦ = 0.
∂ξj ∂ξj
This proves (6).

5.3 Smooth projective manifolds

Exercise 5.3. Let h be a Hermitian form on a complex vector space V .
Write h = B + iω, with B and ω real-valued. Prove that for all v, w ∈ V ,

B(iv, w) = −ω(v, w).

Proof. Let v, w ∈ V be given. Then

B(iv, w) + iω(iv, w) = h(iv, w)

= ih(v, w)
= −ω(v, w) + iB(v, w).

This implies the claim. (And also that ω(iv, w) = B(v, w).)

Exercise 5.4. In this exercise, we will make it plausible that the standard
Hermitian form on Cn+1 indeed induces a symplectic form on the projective
space Pn (C). (See Example 3.19 in the lecture notes.)
Let h = B + iω be the standard Hermitian form on Cn+1 :
n+1
X
h(v, w) := vj w̄j ,
j=1

for v, w ∈ Cn+1 . Consider the sphere S 2n+1 as the unit sphere in Cn+1 =
R2n+2 , and the restricted two-form ω|S 2n+1 on this sphere. Fix a point
p ∈ S 2n+1 .

18
 (a) Consider the submanifold
U (1)p = {eiα p; α ∈ R} ⊂ S 2n+1 .
Show that
Tp (U (1)p) = iRp.
(b) Let ker ωp be the subspace of all vectors v ∈ Tp S 2n+1 such that for all
w ∈ Tp S 2n+1 ,
ωp (v, w) = 0.
Prove that
ker ωp = iRp = Tp (U (1)p).
The conclusion is that, intuitively, the directions in which ω is degenerate
are the tangent spaces to the sets U (1)p. Passing from S 2n+1 to Pn (C), one
therefore gets rid of all these degenerate directions, and obtains a nonde-
generate form. This will be made more precise in the proof of the Marsden–
Weinstein theorem that symplectic reduction is well-defined.
Hint: For part (b), use the fact that a tangent space to a sphere can be
identified with the orthogonal complement of its base point on the sphere:
Tp S 2n+1 = {v ∈ Cn+1 ; Bp (v, p) = 0}.
Also, use Exercise 5.3.
Proof. (a) The tangent space Tp (U (1)p) equals the set of derivatives γ 0 (0)
of all curves γ in U (1)p such that γ(0) = p. This space contains all
vectors of the form 
d 
eitλ p = iλp,
dt t=0
for λ ∈ R. So iRp ⊂ Tp (U (1)p). Since both spaces are one-dimensional,
they are actually equal.
(b) Let v ∈ Tp S 2n+1 ,→ Cn+1 be given. By Exercise 5.3, one has v ∈ ker ωp
if and only if for all w ∈ Tp S 2n+1 ,
B(iv, w) = 0. (8)
This means that iv is in the orthogonal complement of Tp S 2n+1 . Since
Tp S 2n+1 is the orthogonal complement of p, we see that (8) is equiva-
lent to iv ∈ Rp. We conclude that
v ∈ ker ωp ⇔ iv ∈ Rp,
as required.

19
6 Darboux theorem
In the lecture notes, Moser’s proof of the Darboux theorem (Theorem 3.24)
is given. In this set of exercises, we work out the details of this proof.
Let (M, ω) be a symplectic manifold, and fix a point x ∈ M . Let V ⊂ M
be a coordinate neighbourhood of x, with a chart

κ : V → W ⊂ R2n ,

such that κ(x) = (0, 0). Then

ω0 := (κ−1 )∗ (ω|V )

is a symplectic form on W . We write

ν := (ω0 )(0,0)

for the symplectic form (ω0 )(0,0) on T(0,0) W = R2n . Let

{e1 , . . . , en , f1 , . . . , fn }

be a basis of R2n as in the linear Darboux lemma (Lemma 1.2 in the lecture
notes), with respect to this form ν. For j = 1, . . . , n, consider the linear
coordinates
yj , ηj : R2n → R
determined by
n
X
v= yj (v)ej + ηj (v)fj ,
j=1

for all v ∈ R2n .

Let ω1 ∈ Ω2 (W ) be the translation-invariant extension of ν to all of W ,
i.e. for all p ∈ W and v, w ∈ Tp W = R2n ,

(ω1 )p (v, w) = ν(v, w).

Exercise 6.1. Prove that

X
ω1 = dηj ∧ dyj .
j

20
Proof. For p ∈ W and v, w ∈ Tp W = R2n , one has

(ω1 )p (v, w) = ν(v, w) =

X 
yj (v)yk (w)ν(ej , ek )+ηj (v)yk (w)ν(fj , ek )+yj (v)ηk (w)ν(ej , fk )+ηj (v)ηk (w)ν(fj , fk )
j,k
X

= yj (v)ηj (w) − ηj (v)yj (w) .
j

Since the functions yj , ηj on W are linear, they are equal to their own
derivatives. Hence
(dηj ∧ dyj )(v, w) = (ηj ∧ yj )(v, w)
= yj (v)ηj (w) − ηj (v)yj (w).

This proves the claim.

The forms ω0 and ω1 on W are not the same in general; we only know
they are the same at (0, 0). In the remainder of the proof, we will deform
ω1 to ω0 in a small enough neighbourhood of (0, 0), in a way that allows us
to prove the Darboux theorem.
For r > 0 we will write Br for the open ball in R2n around (0, 0) of radius
r. Since W is open and contains (0, 0), there is an ε > 0 such that Bε ⊂ W .
Since Bε is contractible, and

d(ω1 − ω0 ) = dω1 − dω0 = 0,

there is a one-form λ ∈ Ω1 (Bε ) such that on Bε ,

ω1 − ω0 = dλ.

(This is the Poincaré lemma.)

Exercise 6.2. Show that one may choose λ such that λ(0,0) = 0.

Proof. Let µ ∈ Ω1 (Bε ) by a one-form such that

ω1 − ω0 = dµ.

Let µ0 be the translation invariant extension of µ(0,0) to Bε . Set

λ = µ − µ0 .

Since dµ0 = 0, the claim follows.

21
 For t ∈ [0, 1], set

ωt := (1 − t)ω0 + tω1 ∈ Ω2 (Bε ).

Exercise 6.3. Prove that there is a ζ with 0 < ζ < ε, such that for all
t ∈ [0, 1], the form ωt is nondegenerate on Bζ .

Hint: In any set of local coordinates {al }2n

l=1 (such as {yj , ηj }), the form
ωt can be written as
2n
X
(ωt )p = (Ml,m )(p, t)dal ∧ dam .
l,m=1

This form is nondegenerate at p if and only if the matrix

M (p, t) := (Ml,m (p, t))2n

l,m=1

has nonzero determinant. Now use the function

D : [0, 1] × Bε → R

given by
D(t, p) = det M (p, t).
and the tube lemma in topology.

Proof. By continuity of the function D, the set

D−1 (R \ {0})

is open. Since the forms ω0 and ω1 both equal ν at (0, 0), one has

(ωt )(0,0) = ν

for all t. This form is nondegenerate, so

[0, 1] × {(0, 0)} ⊂ D−1 (R \ {0}).

Since [0, 1] is compact, there is a ζ > 0 such that

[0, 1] × {(0, 0)} ⊂ [0, 1] × Bζ ⊂ D−1 (R \ {0}).

For p ∈ Bζ , the matrix M (p, t) has nonzero determinant for all t, so ωt is

nondegenerate there.

22
Exercise 6.4. Prove that for all t ∈ [0, 1], there is a unique vector field vt
on Bζ such that
ivt ωt + λ = 0. (9)
Furthermore, the map
(t, p) 7→ vt (p) (10)
from [0, 1] × Bζ to R2n , is continuous (smooth on ]0, 1[×Bζ ).

Proof. Since ωt is nondegenerate on Bζ , at every p ∈ Bζ it defines a linear

isomorphism
(ω̃t )p : Tp Bζ → Tp∗ Bζ .
Hence (9) is equivalent to

vt (p) = −(ω̃t )−1

p (λp ).

The map (10) is continuous/smooth, because (ωt )p depends continu-

ousy/smoothly on t and p, and λ is smooth.

Next, consider the family of smooth maps4

φ : [0, 1] × Bζ → R2n

given by
φ(t, p) = φt (p),
such that for all s ∈ [0, 1] and p ∈ Bζ ,

d 
φt (p) = vs (φs (p)).
dt t=s

Exercise 6.5. Prove that there is a δ > 0 such that for all t ∈ [0, 1],

φt (Bδ ) ⊂ Bζ .

Hint: Analogously to Exercise 6.3, use the continuous map

φ : [0, 1] × Bζ → R2n .
4
Smooth on ]0, 1[×Bζ ; continuous on [0, 1] × Bζ . Existence of such a family of maps
follows from a slight generalisation of the proof of existence of flow of vector fields.

23
Proof. Since vt is continuous in t, so is φt , and hence the map φ is continuous.
Therefore the set
φ−1 (Bζ )
is open. It contains [0, 1] × {(0, 0)}, because vt (0, 0) = 0 for all t, so that
φt (0, 0) = (0, 0). Hence, by compactness of [0, 1], there is a δ > 0 such that

[0, 1] × {(0, 0)} ⊂ [0, 1] × Bδ ⊂ φ−1 (Bζ ).

This δ has the required properties.

The following property of the maps φt is important to Moser’s trick.

Exercise 6.6. In this exercise, we prove that for all differential forms α ∈
Ω(Bζ ), one has 
d 
φ∗ α = φ∗s Lvs α. (11)
dt t=s t
Note that for constant families of vector fields vt , this relation is easier to
prove.
In this exercise, you may use all properties of Lie derivatives.

(a) Prove that (11) holds if α = f ∈ C ∞ (M ) is a smooth function.

(b) Prove that if (11) holds for a differential form α ∈ Ω(M ), then it also
holds for dα.

(c) Prove that if (11) holds for two differential forms α, β ∈ Ω(M ), then
it also holds for α ∧ β.

(d) Conclude from parts (a)–(c) that (11) holds for all α ∈ Ω(Bζ ).

Proof. (a) For a function f ∈ C ∞ (Bζ ), one has for all p ∈ Bζ

   
d  ∗ d 
φ f (m) =  f (φt (m))
dt t=s t dt t=s
= dφs (m) f (vs (φs (p)))
= v(f )(φs (m))
= (φ∗s Lvs f )(m).

24
 (b) Suppose (11) holds for a form α. Then
 
d  d 
φ dα =  dφ∗t α
∗
dt t=s t dt t=s

d 
= d  φ∗t α
dt t=s
= dφ∗s Lvs α
= φ∗s Lvs dα.

(c) Suppose (11) holds for α, β ∈ Ω(M ). Then

 
d  ∗ d 
φ∗t α ∧ φ∗t β .


φt (α ∧ β) = 
dt t=s
 dt t=s

By Exercise 3.2 (which extends immediately to differentiation at other

values than t = 0), this equals
 
d  ∗ ∗ d 
φ∗s α ∧ φ∗t β = φ∗s Lvs α ∧ φ∗s β + φ∗s α ∧ φ∗s Lvs β



φt α ∧ φs β + 
dt t=s
 dt t=s
= φ∗s Lvs α ∧ β + α ∧ Lvs β

= φ∗s Lvs (α ∧ β).

(d) All differental forms on Bζ (indeed, on any manifold), are sums of

wedge products of functions and differentials of functions.

We are now ready for the key step in Moser’s proof of the Darboux
theorem.

Exercise 6.7. Prove that for all t ∈ [0, 1],

φ∗t (ωt |Bζ ) = ω0 |Bδ . (12)

Hint: See the proof of Theorem 2.12 in the lecture notes.

Proof. By Exercise 3.2, one has for all s ∈ [0, 1],

  
d  d  d 
φ ωt =  φt ωs +  φ∗s ωt
∗ ∗
dt t=s t dt t=s dt t=s

25
By Exercise 6.6, the first of these terms equals φ∗s (Lvs ωs ). The second term
equals
 
d  d 
φs ωt =  (1 − t)φ∗s ω0 + tφ∗s ω1 = φ∗s (ω1 − ω0 ).
∗
dt  t=s dt t=s

Therefore, 
d 
φ∗ ωt = φ∗s (Lvs ωs ) + φ∗s (ω1 − ω0 ).
dt t=s t
By Cartan’s formula for the Lie derivative Lvs , closedness of ωs , and the
definition of the form λ, the latter expression equals

φ∗s (d(ivs ωs ) + ivs dωs + ω1 − ω0 ) = φ∗s (d(ivs ωs ) + dλ)

= φ∗s d(ivs ωs + λ).
d
φ∗ ω = 0 for all s, so φ∗t ωt is

Since ivs ωs + λ = 0, we conclude that dt t=s t t
independent of t. Because φ0 is the identity map, this implies (12).

Exercise 6.7 finally allows us to prove Darboux’s theorem. Consider the

coordinates
xj := yj ◦ φ1 ◦ κ;
ξj := ηj ◦ φ1 ◦ κ

on U := κ−1 (Bδ ).

Exercise 6.8 (Darboux theorem). Prove that

X
ω|U = dξj ∧ dxj .
j

Proof. We compute
X  X
(κ−1 )∗ dξj ∧ dxj = d(ξj ◦ κ−1 ) ∧ d(xj ◦ κ−1 )
j j
X
= d(ηj ◦ φ1 ) ∧ d(yj ◦ φ1 ).
j

By Exercise 6.1, one has

X
ω1 = dηj ∧ dyj ,
j

26
and Exercise 6.7 implies that in particular

φ∗1 ω1 = ω0

on Bδ . We conclude that
X 
(κ−1 )∗ dξj ∧ dxj = φ∗1 ω1 |Bδ
j

= ω0 |Bδ
= (κ−1 )∗ (ω|U ).

Exercise 6.9. Compute the Poisson bracket in Darboux coordinates.

7 Groups and actions

7.1 Proper maps
Exercise 7.1. Let X, Y, Z be topological spaces, and ϕ : X → Y × Z
a continuous map. Show that ϕ is proper if and only if for all compact
subsets CY ⊂ Y and CZ ⊂ Z, the set ϕ−1 (CX × CY ) is compact.
Proof. If ϕ is proper, then ϕ−1 (CX × CY )is compact because CX × CY is
compact.
Conversely, suppose that ϕ−1 (CX ×CY ) is compact for all compact CY ⊂
Y and CZ ⊂ Z. Let C ⊂ Y × Z be any compact subset. Then there are
compact subsets CY ⊂ Y and CZ ⊂ Z such that

C ⊂ CY × CZ .

Indeed, one may take the projections of C onto Y and Z. These are contin-
uous, so their images are compact. By assumption, the set ϕ−1 (CX × CY )
is compact. Since ϕ−1 (C) is a closed subset if this set, it is compact as
well.

Exercise 7.2. Let G be a locally compact topological group, H < G a

subgroup. Consider the action by H on G by right multiplication, and let
G/H be the orbit space of this action. Consider the quotient map

q : G → G/H,

given by q(g) = gH. Prove that q is proper if and only if H is compact.

27
 Hint: If H is compact, let C ⊂ G/H be a compact subset. Show that
there are finitely many relatively compact open sets U1 , . . . , Un ⊂ G such
that
n
[
q −1 (C) ⊂ H · Uj .
j=1

Proof. If H is noncompact, then q −1 ({eH}) = H is noncompact, so q is not

proper.
Conversely, suppose H is compact. Let C ⊂ G/H be compact. Consider
a cover {Uj }j∈J of q −1 (C) by relatively compact open subsets. Since C is
compact, the cover {q(Uj )}j∈J has a finite subcover {q(Uj )}j∈I (i.e. with
I ⊂ J a finite subset). Then {q −1 q(Uj )}j∈I covers q −1 (C). Now q −1 q(Uj ) =
H · Uj , so [ [
q −1 (C) ⊂ H · Uj ⊂ H · Uj.
j∈I j∈I

Since the action is continuous, the sets H · U j are compact. Their finite
union is also compact. Since q −1 (C) is closed, it is compact as well.

7.2 Proper and free actions

Exercise 7.3. Prove that a proper action has compact stabilisers.

Proof. Let G be a topological group acting properly in a topological space

X. Let x ∈ X. Consider the map

ϕ : G × X → X × X,

given by
ϕ(g, x) = (x, gx).
Since ψ is proper, the set

ψ −1 ({(x, x)}) = Gx × {x}

is compact, so the stabiliser Gx is compact.

Exercise 7.4. Let G be a topological group acting on a topological space

X. Prove the following statements.

(a) If G is compact, the action is proper.

(b) If X is compact, the action is proper if and only if G is compact.

28
Proof. Consider the map

ϕ : G × X → X × X,

given by
ϕ(g, x) = (x, gx).

(a) Suppose G is compact, and let C ⊂ X × X be compact. Then

ϕ−1 (C) = {(g, x); (x, gx) ∈ C} ⊂ G × C.

So ϕ−1 (C) is a closed subset of the compact set G × C, and hence

compact.

(b) If G is compact, then the action is proper by part (a). Conversely,

suppose the action is proper. By compactness of X, the set

G × X = ϕ−1 (X × X)

is then compact. So G must be compact.

Exercise 7.5. Let G be a topological group, H < G a closed subgroup.

Consider the action by H on G by right multiplication. Let G/H be the
orbit space of this action. Consider the action by G on G/H defined by

g · (g 0 H) = gg 0 H,

for g, g 0 ∈ G.

(a) Show that the action by H on G by right multiplication is proper, so

that G/H is a Hausdorff space.

(b) If G is Lie group, show that G/H has a smooth manifold structure.

(c) Determine the stabiliser of any point gH ∈ G/H.

(d) Show that the action by G on G/H is free if and only if H is the trivial
group.

(e) Show that the action is proper if and only if H is compact.

29
Proof. (a) Consider the usual map ϕ : H × G → G × G given by ϕ(h, g) =
(g, gh). Let C1 , C2 ⊂ G be compact. By Exercise 7.1, it is enough to
show that ϕ−1 (C1 × C2 ) is compact for all such sets. Now
ϕ−1 (C1 × C2 ) = {(h, g) ∈ H × G; g ∈ C1 , gh ∈ C2 } ⊂ C1−1 C2 × C1 .
By continuity of the multiplication and inversion maps, the latter set
is compact. The set ϕ−1 (C1 × C2 ) is closed in H × G, and because H
is closed, it is closed in G × G as well. Therefore, as a closed subset of
a compact set, is is compact.
(b) The action by H on G is free, and we showed in part (a) that it is
proper. Hence G/H is a smooth manifold.
(c) Let gH ∈ G/H and g 0 ∈ G be given. Then g 0 gH = gH if and only if
there is an h ∈ H such that g 0 g = gh, i.e.
g 0 = ghg −1 .
Hence the stabiliser of gH is
GgH = gHg −1 .

(d) All stabilisers are trivial if and only if gHg −1 is trivial for all g ∈ G,
i.e. H is trivial.
(e) If H is noncompact, then all stabilisers are noncompact. Therefore,
by Exercise 7.3, the action is not proper.
Conversely, suppose H is compact. Let C1 , C2 ⊂ G/H be given. Then
ψ −1 (C1 × C2 ) = {(g, g 0 H) ∈ G × G/H; g 0 H ∈ C1 , gg 0 H ∈ C2 }.
By Exercise 7.2, the subsets q −1 (C1 ) and q −1 (C2 ) of G are compact.
Now g ∈ G satisfies
gg 0 H ∈ C2
for some g 0 H ∈ C1 , if and only if gg 0 ∈ q −1 (C2 ) for some g 0 ∈ q −1 (C1 ).
So
g ∈ q −1 (C1 )−1 q −1 (C2 ).
By continuity of multiplication and inversion, this set is compact. We
conclude that the closed subset
ψ −1 (C1 × C2 ) ⊂ q −1 (C1 )−1 q −1 (C2 ) × q −1 (C1 )

is compact as well. By Exercise 7.1, this shows that ψ is proper.

30
Exercise 7.6. Let T2 = R2 /Z2 be the two-torus. For λ ∈ R, define the
subgroup
H := {(a, λa) + Z2 ; a ∈ R} < G.
Consider the action by H on G by multiplication (addition). For which
values of λ is this action proper? And when is it free? What is the relation
with Exercise 7.5 (a)?
√
Hint: Draw a picture of H for λ = 1/2, and try to draw one for λ = 1/ 2.
To draw these pictures, it is easiest to depict T2 as the square [0, 1]2 , with
appropriate identifications on the boundaries.

Proof. Multiplication actions by subgroups are always free.

By Exercise 7.4 (b), the action by H on T2 is proper if and only if H is
compact. Consider the map

ϕ:R→H

given by ϕ(a) = (a, λa) + Z2 . Let a, b ∈ R and suppose that ϕ(a) = ϕ(b).
Then there are k, l ∈ Z such that

(a − b, λ(a − b)) = (k, l).

That is,

1. a = b and k = l = 0; or

2. λ = l/k ∈ Q and a − b = k.

This implies that for irrational λ, the map ϕ is injective. Since it is also
surjective, continuous, and it has a continous inverse, we see that H ∼
= R is
noncompact and the action is not proper.
If λ = l/k is rational, then for all a ∈ R and m ∈ Z, one has

ϕ(a + km) = a + km, λ(a + km) + Z2

= (a, λa) + (km, lm) + Z2

= ϕ(a).

Hence the map ϕ induces a surjective map

R/kZ → H

by
a + kZ 7→ ϕ(a).

31
Since the circle R/kZ is compact, the image H if this continous map is
compact. Hence the action is proper.
In Exercise 7.5 (a), we saw that multiplication actions by closed sub-
groups are proper. For rational λ, the subgroup H is compact, hence closed.
For irrational λ, H is noncompact. Hence H is not closed, since otherwise
it would be compact by compactness of T2 . (In fact, for irrational λ, the
subgroup H is dense in T2 .)

8 Lie groups
8.1 General
Exercise 8.1. Let G be a Lie group, and consider the multiplication map

m : G × G → G.

Prove that its derivative at the identity element is give by addition:

Te m(X, Y ) = X + Y,

for all X, Y ∈ g.

Proof. For X, Y ∈ g, Exercise 3.2 implies that


d 
Te m(X, Y ) =  exp(tX) exp(tY )
dt t=0
 
d  d 
=  exp(tX) +  exp(tX)
dt t=0 dt t=0
= X + Y.

Exercise 8.2. Consider the determinant map

det : GL(n) → R.

Prove that
TI det = tr
is the trace map.

32
Proof. For X ∈ Mn (R) = Lie(GL(n)), one has

d 
TI det(X) =  det(I + tX).
dt t=0

Working out the deteminant of

 
X11 + t · · · X1n
I + tX = 
 .. .. .. 
. . . 
Xn1 ··· Xnn + t

explicitly, one finds that

det(I + tX) = 1 + t tr(X)

plus terms of order 2 and higher in t.

Exercise 8.3. Consider the Lie group

SL(n) = {g ∈ GL(n); det(g) = 1}.

(a) Prove that the determinant map det : GL(n) → R is a submersion at

I.

(b) Prove that the Lie algebra sl(n) of SL(n) equals

sl(n) = {X ∈ Mn (R); tr(X) = 0}.

Proof. (a) By Exercise 8.2, the tangent map of det at I is the trace map.
This map is surjectuve onto R.

(b) By the submersion theorem, SL(2) is a smooth submanifold of GL(n)

in a neigbourhood of I. (By Lemma 2.13 in the notes on Lie groups,
or the more powerful Theorem 2.16, it is a smooth submanifold every-
where.) Furthermore, one has

TI SL(n) = (TI det)−1 (0) = {X ∈ Mn (R); tr(X) = 0}.

Exercise 8.4. Consider the Lie group

O(n) = {g ∈ GL(n); g T g = I}.

33
Here g T is the transpose of g, and I is the identity matrix.
Let Sn (R) be the vector space of all symmetric real n × n matrices,5 and
let SGL(n) be the open subset

SGL(n) := Sn (R) ∩ GL(n).

(a) Prove that the map f : GL(n) → SGL(n) given by

f (g) = g T g

has tangent map

TI f : Mn (R) → Sn (R)
given by
TI f (X) = X T + X.

(b) Prove that f is a submersion at I.

(c) Prove that the Lie algebra o(n) of O(n) equals

o(n) = {X ∈ Mn (R); X + X T = 0}.

Proof. (a) Since SGL(n) is an open subset of the vector space Sn (R), its
tangent space at I equals Sn (R). Tranposing matrices is a linear map
from Mn (R) to Mn (R) in general. So by Exercise 8.1 and the chain
rule, the claim follows.

(b) If Y ∈ §n (R), then

Y = TI f (Y /2),
so TI f is surjective.

(c) By the submersion theorem, SL(2) is a smooth submanifold of GL(n)

in a neigbourhood of I. (By Lemma 2.13 in the notes on Lie groups,
or the more powerful Theorem 2.16, it is a smooth submanifold every-
where.) Furthermore, one has

TI O(n) = (TI f )−1 (0) = {X ∈ Mn (R); X + X T = 0}.

5
The symbols Sn (R) and SGL(n) are not standard.

34
Exercise 8.5. Consider the Lie group

SO(n) = SL(n) ∩ O(n) = {g ∈ GL(n); g T g = I and det(g) = 1}.

Prove that the Lie algebra so(n) of SO(n) equals

so(n) = {X ∈ Mn (R); X + X T = 0} = o(n).

(In fact, orthogonal matrices have determinant ±1. The group O(n)
has two connected components: SO(n), and the subset of matrices with
determinant −1.)

Proof. One has

so(n) = sl(n) ∩ o(n) = {X ∈ Mn (R); X + X T = 0 and tr(X) = 0}.

Since antisymmetric matrices have zeros on the diagonal, they have trace
zero. So
so(n) = {X ∈ Mn (R); X + X T = 0} = o(n).

8.2 The exponential map

Exercise 8.6. Consider the Lie group G = R, with addition as the group
operation. Fix an element X ∈ g = R.
(a) What is the left invariant vector field vX associated to X?
(b) What is the differential equation defining the flow curve αX along vX
starting at e = 0?
(c) What is the solution of this equation?
(d) What is exp(X)?
Proof. (a) Let g ∈ G = R. The map lg : G → G is given by

lg (h) = g + h,

for all h ∈ G. So
T0 lg : Te G → Tg G
is the identity map on R. Therefore,

vX (g) = T0 lg (X) = X,

at any point g ∈ G.

35
 (b) By definition of flow, αX satisfies

d 
αX (t) = vX (αX (s)) = X,
dt t=s

for all s ∈ R. Also,

αX (0) = 0.

(s) The solution of this equation is

αX (t) = tX,

for all t.

(d) By definition of the exponential map,

exp(X) = αX (1) = X.

That is, the exponential map is the identity map form the Lie algebra
g = R to the Lie group G = R.

Exercise 8.7. Consider the Lie group G = R+ of positive real numbers,

with multiplication as the group operation. Fix an element X ∈ g = R.

(a) What is the left invariant vector field vX associated to X?

(b) What is the differential equation defining the flow curve αX along vX
starting at e = 1?

(c) What is the solution of this equation?

(d) What is exp(X)?

Proof. (a) Let g ∈ G = R+ . The map lg : G → G is given by

lg (h) = gh,

for all h ∈ G. So
T1 lg : Te G → Tg G
is multiplication by g (on R). Therefore,

vX (g) = T1 lg (X) = gX,

at any point g ∈ G.

36
 (b) By definition of flow, αX satisfies

d 
αX (t) = vX (αX (s)) = αX (s)X,
dt t=s
for all s ∈ R. Also,
αX (0) = 1.

(d) The solution of this equation is

αX (t) = etX ,

for all t.

(d) By definition of the exponential map,

exp(X) = αX (1) = eX .

That is, the exponential map is the usual exponential map from the
Lie algebra g = R to the Lie group G = R+ .

Exercise 8.8. Let G < GL(n) be a linear group, and let X ∈ g ⊂ Mn (R).
Use Exercise 2.3 to show that
∞
X
X Xj
exp(X) = e := .
j!
j=1

Proof. For g ∈ G, the map lg : G → G is the restriction to G of the linear

map
g· : Mn (R) → Mn (R),
given by left matrix multiplication by g. The tangent map at the identity
matrix I of this map equals the same map

g· : TI Mn (R) = Mn (R) → Tg Mn (R) = Mn (R).

Restricting this map to TI G, we see that

TI lG : TI G → Tg G

is given by left matrix multiplication by g.

Let X ∈ g be given. By the preceding argument, the left invariant vector
field vX on G is given by

vX (g) = gX ∈ Tg G ⊂ Mn (R).

37
By Exercise 2.3, the flow curve αX along this vector field, starting at I, is
given by
αX (t) = etX .
Hence
exp(X) = αX (1) = eX .

Exercise 8.9. Consider the Lie group SO(2).

(a) Show that the Lie algebra of SO(2) equals

   
0 −1
so(2) = t ;t ∈ R .
1 0

(b) Prove that for all j ∈ N:

 2j
0 −1
= (−1)j I;
1 0
 2j+1  
0 −1 j 0 −1
= (−1) .
1 0 1 0

(c) Prove that

    
0 −1 cos t − sin t
exp t = .
1 0 sin t cos t

Proof. (a) This follows from Exercise 8.5, and the  fact that
 all antisym-
0 −1
metric 2 × 2 matrices are scalar multiples of .
1 0

(b) One directly verifies that

 2
0 −1
= −I.
1 0

The other claims follow from this equality.

38
 (c) By Exercise 8.8, one has
   X∞ j  j
0 −1 t 0 −1
exp t = .
1 0 j! 1 0
j=0

By part (b), this equals

∞ ∞
(−1)j t2j (−1)j t2j+1
 
X X 0 −1
I+ .
j! j! 1 0
j=0 j=0

Now we use the Taylor series for the sine and cosine functions:
∞
X (−1)j t2j+1
sin(t) = ;
j!
j=0

∞
X (−1)j t2j
cos(t) = .
j!
j=0

The claim follows.

Exercise 8.10. Let G and H be Lie groups, with Lie algebras g and h, unit
elements eG and eH , and exponential maps expG and expH , respectively.
Let
ϕ:G→H
be a Lie group homomorphism.
Fix an element X ∈ g. Consider the left invariant vector field vTeG ϕ(X)
on H associated to TeG ϕ(X) ∈ h.
(a) Prove that for all s, t ∈ R,


ϕ(expG ((s + t)X)) = lϕ(expG (sX)) ϕ(expG (tX)) .

(b) Let αTHe ϕ(X) : R → H be the flow curve along the vector field vTeG ϕ(X) ,
G
starting at the unit element eH . Prove that for all t,

αTHe ϕ(X) (t) = ϕ(expG (tX)). (13)

G

(c) Conclude that

expH (Te ϕ(X)) = ϕ(expG (X)),
i.e. Lemma 3.9 in the lecture notes on Lie groups is true.

39
Proof. (a) By Lemma 3.6(b) and the fact that ϕ is a group homomor-
phism, we have

ϕ(expG ((s + t)X)) = ϕ(expG (sX) expG (tX))

= ϕ(expG (sx))ϕ(expG (tX))


= lϕ(expG (sX)) ϕ(expG (tX))

(b) For t = 0, we have

ϕ(expG (0 · X)) = ϕ(eG ) = eH = αTHe ϕ(X) (0).

Furthermore, by part (a) and the definition of the vector field vTeG ϕ(X) ,
 
d  d 
ϕ(expG (tX)) =  ϕ(expG ((s + t)X))
dt t=s dt t=0

d 
=  lϕ(expG (sX)) ϕ(expG (tX))
dt t=0
= TeH lϕ(expG (sX)) (TeG ϕ(X))
= vTeG ϕ(X) (ϕ(expG (sX))).

Therefore, the curve t 7→ ϕ(expG (sX)) satisfies the defining differential

equation of the flow curve αTHe ϕ(X) .
G

(c) Take t = 1 in (13).

8.3 The Lie bracket

Exercise 8.11. Let G be a Lie group, and g = Te G its Lie algebra. For
X ∈ g, we denote the associated left invariant vector field by vX . In this
exercise, we will show that for any two X, Y ∈ g, one has

v[X,Y ] = [vX , vY ], (14)

the Lie bracket of the vector fields vX and vY . The conclusion is that the
space of left invariant vector fields with the usual Lie bracket of vector
fields, is isomorphic, as a Lie algebra, to g with the Lie bracket defined via
the adjoint action.
Fix X, Y ∈ g and g ∈ G. For any Lie algebra element Z ∈ g, we denote
the flow along vZ over time t by φZ t .

40
 (a) Prove that 
d 
[vX , vY ]g =  Te φX


−t ◦ lφX (g) (Y ).
dt t=0 t

(b) Prove that, for all s, t ∈ R

(φX
−t ◦ lφX
t (g)
)(exp(sY )) = gαX (t)αY (s)αX (−t).

Hint: At the start of the proof of Lemma 3.2 in the notes on Lie
groups, it is shown that for all Z ∈ g, h ∈ G and u ∈ R,

φZ
u (h) = hαZ (u). (15)

(c) Conclude from parts (a) and (b) that

 
d  d 
[vX , vY ]g =  gαX (t)αY (s)αX (−t).
dt t=0 ds s=0

(d) Prove that for all t ∈ R,


d 
Ad(exp(tX))Y = αX (t)αY (s)αX (−t).
ds s=0

(e) Prove that

 

d  d 
v[X,Y ] g
=  gαX (t)αY (s)αX (−t),
dt t=0 ds s=0
completing the proof of (14).
Proof. (a) Consecutively applying the definitions of the Lie bracket of vec-
tor fields, of the Lie derivative of vector fields, of pullbacks of vector
fields and of the vector field vY and finally the chain rule, we see that


[vX , vY ]g = LvX (vY ) g

d  ∗
(φX


=  t ) vY g
dt t=0

d  X


=  TφX (g) φ−t (vY )φX (g)
dt t=0 t t

d 
=  TφX φX ◦ Te lφX (Y )
dt t=0 t (g) −t t (g)

d
=  Te φX


−t ◦ lφX (g) (Y ).
dt t=0 t

41
 Here we have also used that φX X
−t is the inverse of φt , since

φX X X X X
−t ◦ φt = φ0 = φt ◦ φ−t

is the identity map.

(b) Applying (15) several times, we see that for all s ∈ R,
(φX )(exp(sY )) = φX X


−t ◦ lφX
t (g) −t φt (g) exp(sY )

= φX


−t gαX (t)αY (s)
= gαX (t)αY (s)αX (−t).

(c) By parts (a) and (b),


d 
Te φX


[vX , vY ]g = −t ◦ lφX (g) (Y )
dt t=0
 t
 
d  d 
=  (φX ◦ l X )(exp(sY ))
dt t=0 ds s=0 −t φt (g)
 
d  d 
=  gαX (t)αY (s)αX (−t).
dt t=0 ds s=0

(d) Because of the equality αZ (u) = exp(uZ) for all Z ∈ g and u ∈ R, we

have
  
d 
Ad(exp(tX))Y = Ad(exp(tX)) exp(sY )
ds s=0

d 
= exp(tX) exp(sY ) exp(−tX)
ds s=0

d 
= αX (t)αY (s)αX (−t).
ds s=0

(e) Using the fact that


d 
ad(X)Y =  Ad(exp(tX))Y,
dt t=0
we compute


v[X,Y ] g
= Te lg ([X, Y ])
= Te lg (ad(X)Y )
  
d 
= Te lg Ad(exp(tX))Y
dt t=0

d

=  Te lg Ad(exp(tX))Y ,
dt t=0

42
 since Te lg is a linear map. Now by part (d), the latter expression
equals
     
d  d  d  d 
Te lg αX (t)αY (s)αX (−t) =  gαX (t)αY (s)αX (−t).
dt t=0 ds s=0 dt t=0 ds s=0

9 Infinitesimal actions
Exercise 9.1. Consider the natural action by SO(3) on the two-sphere
S 2 ⊂ R3 by rotations.
(a) Prove that the matrices
 
0 0 0
Rx :=  0 0 −1 
0 1 0
 
0 0 −1
Ry :=  0 0 0 
1 0 0
 
0 −1 0
Rz :=  1 0 0 
0 0 0
form a basis of the Lie algebra so(3) = Lie(SO(3)).
Hint: Use Exercise 8.5.
(b) Generalise Exercise 8.9 to show that for all t ∈ R,
exp(tRx ) ∈ SO(3)
is rotation over angle t around the x-axis, and similarly for the expo-
nentials of tRy and tRz . (Give explicit expressions for these exponen-
tials.)
(c) Let  
x
m =  y  ∈ S2
z
be given. Let (Rx )S 2 , (Ry )S 2 and (Rz )S 2 be the vector fields on S 2
induced by Rx , Ry and Rz via the infinitesimal action. Compute
(Rx )S 2 (m), (Ry )S 2 (m) and (Rz )S 2 (m).

43
 (d) For j = 1, 2, 3, let ej be the jth standard basis vector of R3 . Conclude
that in particular,

(Rx )S 2 (e1 ) = 0;
(Ry )S 2 (e1 ) = e3 ;
(Rz )S 2 (e1 ) = e2 .

In other words, the vector fields induced by Rx , Ry and Rz indeed

point “in the direction of the action”.

Proof. (a) The lie algebra so(3) consists of all antisymmetric 3 × 3 matri-
ces. These are of the form
 
0 −a −b
 a 0 −c  = aRx + bRy + cRz ,
b c 0

for a, b, c ∈ R. The matrices Rx , Ry and Rz are linearly independent,

so they indeed form a basis.

(b) Analogously to Exercise 8.9, one has

∞ j j
X t Rx
exp(tRx ) =
j!
j=0
∞
X t2j 2j t2j+1
= Rx + R2j+1 .
(2j)! (2j + 1)! x
j=0

Now  
0 0 0
Rx2j = (−1)j  0 1 0 
0 0 1
if j ≥ 1, and
   
1 0 0 0 0 0
Rx0 = I =  0 0 0  +  0 1 0  .
0 0 0 0 0 1

Hence
     
∞ 2j 1 0 0 ∞ 2j 0 0 0 1 0 0
X t X t
Rx2j =  0 0 0 + (−1)j  0 1 0  =  0 cos(t) 0 .
(2j)! (2j)!
j=0 0 0 0 j=0 0 0 1 0 0 cos(t)

44
 Similarly,
Rx2j+1 = (−1)j Rx
for all j, so
 
∞ ∞ 0 0 0
X t2j+1 X t2j+1
R2j+1 = (−1)j Rx =  0 0 − sin(t)  .
(2j + 1)! x (2j + 1)!
j=0 j=0 0 sin(t) 0
Hence  
1 0 0
exp(tRx ) =  0 cos(t) − sin(t)  ,
0 sin(t) cos(t)
which is rotation around the x-axis over angle t.
In the same way, one finds
 
cos(t) 0 − sin(t)
exp(tRy ) =  0 1 0 ,
sin(t) 0 cos(t)
and  
cos(t) − sin(t) 0
exp(tRz ) =  sin(t) cos(t) 0  .
0 0 1

(c) A direct comuptation yields


d 
(Rx )S 2 (m) =  exp(tRx )m
dt t=0
  
 1 0 0 x
d 

=  0 cos(t) − sin(t)   y 
dt t=0
0 sin(t) cos(t) z
 
0
=  −z 
y
 
0 0 0
=  0 0 −1  m
0 1 0
= Rx · m.
Similarly, one finds that
(Ry )S 2 (m) = Ry · m and (Rz )S 2 (m) = Rz · m.

45
 (d) This follows from explicit matrix calculations.

Exercise 9.2. To generalise Exercise 9.1, let G < GL(n) be a linear group,
and let M ⊂ R2 be a G-invariant submanifold. Consider the natural action
by G on M . Prove that, for all X ∈ g and m ∈ M ,
XM (m) = X · m.
Here XM is the vector field on M induced by the infinitesial action, and the
dot in X · m denotes the product of the matrix X and the vector m.
Proof. We compute

d 
XM (m) =  exp(tX)m
dt t=0
∞
d  X tj j

=  X · m.
dt j!
t=0 j=0

As in Exercise 2.3, we note that the series

∞ j
X t
Xj · m
j!
j=0

converges, while the series of derivatives

∞
tj j

X d 
X · m = X · m.
dt t=0 j!

j=0

converges uniformly (because only one term is nonzero). Hence the deriva-
tive and the infinite sum may be interchanged, and one finds that
∞
tj j

X d 
XM (m) = X · m = X · m.
dt t=0 j!
j=0

10 The slice lemma

The next set of exercises is about a proof of the slice lemma (Lemma 13.7
in the notes on Lie groups). Since we focused on left actions, we will state
this result for left actions rather than right actions.6
6
But recall that any left action α corresponds to a right action β (and vice versa) via
β(g) = α(g −1 ).

46
 Let G be a Lie group acting smoothly, properly and freely on a smooth
manifold M . We will prove the following result:

Lemma 10.1 (Slice lemma). For every m ∈ M , there is a submanifold S

of M containing m, such that the map

ϕ:G×S →M

given by ϕ(g, s) = g · s for all g ∈ G and s ∈ S, is an equivariant diffeomor-

phism onto an open, G-invariant neighbourhood U of M .

The slice lemma is the key step in the proof that proper, free, smooth
actions have smooth quotients. Indeed, it implies that locally, M/G looks
like the smooth manifold S. More on this in Exercise 10.7.
Fix a point m ∈ M . Consider the map

αm : G → M

given by αm (g) = g · m, for g ∈ G.

Exercise 10.1. Prove that αm is injective.

Proof. Let g, g 0 ∈ G, and suppose αm (g) = αm (g 0 ). Then g −1 g 0 m = m, so

by freeness of the action,

g −1 g 0 ∈ Gm = {e}.

Hence g = g 0 .

Let g be the Lie algebra of G, and consider the tangent map

Te αm : g → Tm M.

Exercise 10.2. Let X ∈ g be in the kernel of Te αm :

Te αm (X) = 0.

We will show that X = 0, so that Te αm is injective.

Consider the curve c : R → M given by

c(t) = exp(tX) · m

for t ∈ R.

(a) Prove that c0 (0) = 0.

47
 (b) Prove that c0 (t) = 0 for all t ∈ R.
(c) Prove that X = 0.
Proof. (a) One has
 
0 d  d 
c (0) =  exp(tX) · m =  αm (exp(tX)) = Te αm (X) = 0.
dt t=0 dt t=0

(b) For all t ∈ R, one has


0 d 
c (t) = exp((t + s)X) · m
ds s=0

d 
= exp(tX) exp(sX) · m
ds s=0
  
d 
= Texp(sX)m exp(tX) exp(sX) · m .
ds s=0

By part (a), we have


d 
exp(sX) · m = 0,
ds s=0
and the claim follows.
(c) By part (b), the curve c is constant. So
exp(tX) · m = m
for all t ∈ R. I.e.
exp(tX) ∈ Gm = {e}
for all t. Since exp is a local diffeomorphism, it is injective on a
neighbourhood of e. Hence tX = 0 for all t in a neighbourhood of 0,
so X = 0.

Choose a linear subspace s ⊂ Tm M such that

Tm M = s ⊕ Te αm (g).
Since Te αm is injective, s has dimension dim M − dim G. Choose a subman-
ifold S 0 ⊂ M containing m, such that
Tm S 0 = s.
Define the map ϕ : G × S 0 → M by ϕ(g, s) = g · s, for g ∈ G and s ∈ S 0 .

48
Exercise 10.3. (a) Prove that for all X ∈ g and v ∈ s,

T(e,m) ϕ(X, v) = v + XM (m),

where XM is the vector field on M induced by X via the infinitesimal

action.

(b) Prove that T(e,m) ϕ is bijective.

(c) Prove that T(e,s) ϕ is bijective for all s in an open neighbourhood S 00

of m in S 0 .

(d) Prove that T(g,s) ϕ is bijective for all g ∈ G and s ∈ S 00 .

Proof. (a) Let γ : R → S 0 be a curve with γ(0) = m and γ 0 (0) = v. Then


d 
T(e,m) ϕ(X, v) =  exp(tX)γ(t)
dt t=0
 
d  d 
=  γ(t) +  exp(tX) · m
dt t=0 dt t=0
= v + XM (m).

Here we have used Exercise 3.2.

(b) We first show that T(e,m) ϕ is injective. Let v ∈ Tm S 0 and X ∈ g be

such that T(e,m) ϕ(X, v) = 0. Then by part (a),

v + XM (m) = 0.

Now XM (m) = Te αm (X), v ∈ s, and

s ∩ Te αm (g) = {0}.

Hence v = 0 and Te αm (X) = 0. By injectivity of Te αm , this implies

that X = 0. So T(e,m) ϕ is indeed injective.
Since dim s = dim M − dim G, the vector spaces

T(e,m) (G × S 0 ) = g × s

and Tm M have the same dimension. Since T(e,m) ϕ is injective, it is

therefore surjective as well.

49
 (c) Since T(e,m) ϕ is bijective, the inverse function theorem states that ϕ
is a local diffeomorphism at (e, m). Hence there is an open neighbour-
hood O of e in G, and an open neighbourhood S 00 of m in S 0 , such
that
ϕ : O × S 00 → M
is a diffeomorphism onto its image. In particular, the tangent map of
ϕ is bjiective on O × S 00 . This set contains all points of the form (e, s),
for s ∈ S 00 .

(d) The map ϕ is equivariant, in the sense that for all g, h ∈ G and s ∈ S 00 ,

ϕ(g · (h, s)) = ϕ(gh, s) = gh · s = g · ϕ(h, s).

In other words, for all g ∈ G,

g ◦ ϕ ◦ (lg−1 × idS 00 ) = ϕ

Hence, by the chain rule, for any g ∈ G and s ∈ S 00 ,

T(g,s) ϕ = T(g,s) g ◦ ϕ ◦ (lg−1 × idS 00 ) = Ts g ◦ T(e,s) ϕ ◦ T(g,s) (lg−1 × idS 00 ).

The three maps in the latter composition are all bijective, hence so is
T(g,s) ϕ.

We conclude that, by the inverse function theorem, ϕ : G × S 00 → M is

a local diffeomorphism onto its image. So there are open neighbourhoods O
of e in G, and S 000 of m in S 00 , such that

ϕ : O × S 000 → M

is a diffeomorphism onto its image. (We use the same notation for ϕ and its
000
restrictions to various subsets.) Furthermore, if the closure S of S 000 is not
compact, we can always replace S 000 be a smaller neighbourhood of m, such
000
that S is compact. Assume this has been done.
Set
000 000
C := {g ∈ G; gS ∩ S 6= ∅}.
This set is compact by properness of the action, and Lemma 13.3 in the
notes on Lie groups. Set
C0 := C \ O.

50
Exercise 10.4. Prove that there is an open subset S ⊂ S 000 such that for
all g ∈ C0 ,
gS ∩ S = ∅.

Hint: Use Lemma 13.6 in the notes on Lie groups.

Proof. The set C0 is a closed subset of C, hence compact itself. Also, if m

would be an element of C0 m, then one would have gm = m for a g ∈ C0 .
But the action is free, so this would imply g = e. Since e ∈ O is not in C0 , we
conclude that m 6∈ C0 m. By Lemma 13.6, there is an open neighbourhood
U of m in M such that
C0 U ∩ U = ∅.
Take S := S 000 ∩ U .

Exercise 10.5. Prove that ϕ : G × S → M is injective.

Proof. Suppose g1 , g2 ∈ G and s1 , s2 ∈ S, such that

ϕ(g1 , s1 ) = ϕ(g2 , s2 ).

Then s2 = (g2−1 g1 )s1 , so g2−1 g1 ∈ C. By Exercise 10.4, one has g2−1 g1 6∈ C0 .

So
g2−1 g1 ∈ C \ C0 ⊂ O.
Therefore, we have
ϕ(g2−1 g1 , s1 ) = ϕ(e, s2 ),
with g2−1 g1 ∈ O, e ∈ O, and s1 , s2 ∈ S ⊂ S 000 . Because ϕ is injective on
O × S 000 , this implies that

g2−1 g1 = e and s1 = s2 ,

as required.

Set
U := ϕ(G × S).

Exercise 10.6. Finish the proof of Lemma 10.1, by showing that U is a

G-invariant open neighbourhood of m, and

ϕ:G×S →U

is an equivariant diffeomorphism.

51
Proof. Since ϕ is injective on G × S, and its tangent maps are bijective on
this set, it is a diffeomorphism onto its image U . Hence U is open. Since
m = ϕ(e, m), U contains m. If g, h ∈ G and s ∈ S, then by equivariance of
ϕ,
g · ϕ(h, s) = ϕ(gh, s) ∈ U,
so U is G-invariant.

As mentioned above, the slice lemma can be used to prove that M/G
has the structure of a smooth manifold. Indeed, by the slice lemma, M can
be covered by open subsets Uj for which there are submanifolds Sj of M
and equivariant diffeomorphisms

ϕj : G × Sj → Uj ,

given by ϕj (g, s) = g · s for g ∈ G and s ∈ Sj . Consider the induced

homeomorphisms on orbit spaces

χj := (ϕj )−1 ∼
G : Uj /G → (G × Sj )/G = Sj .

By shrinking the Sj , one may assume that Sj is diffeomorphic to an open

subset of Rn . Then the maps χj form a smooth atlas on M/G, if the
transition maps χk ◦ χ−1
j are smooth where they are defined.

Exercise 10.7. Let j, k be such that Uj ∩ Uk 6= ∅. Use Lemma 12.4 in the

notes on Lie groups to show that the map

χk ◦ χ−1
j : χj (Uj /G ∩ Uk /G) → χk (Uj /G ∩ Uk /G)

is smooth.

Proof. Consider the diagram

ϕ−1
k ◦ϕj
G × Sj ⊃ ϕ−1 / ϕ−1 (Uj ∩ Uk )
j (Uj ∩ Uk ) k ⊂ G × Sk
p2 p2 p2 p2
  χk ◦χ−1  
Sj ⊃ χj (Uj /G ∩ Uk /G)
j
/ χk (Uj /G ∩ Uk /G) ⊂ Sk .

The projection maps p2 onto the second factor are smooth submersions, and
ϕ−1 −1
k ◦ ϕj is a diffeomorphism. Hence Lemma 12.4 implies that χk ◦ χj is a
smooth map.

52
11 Hamiltonian group actions
Exercise 11.1. Consider the manifold M = R2 , with coordinates (q, p).
Let ω = dp ∧ dq be the standard symplectic form on M . Consider the Lie
group G = R, acting on M be translation in the q-direction:
g · (q, p) = (q + g, p),
for g ∈ G, (q, p) ∈ M .
Prove that this action is Hamiltonian, by showing that the map µ : M →
R, given by
µ(q, p) = p
is a momentum map. Here we identify g∗ ∼ = R, using the fact that any linear
function on g = R is given by multiplication by a real number.
Proof. Since the group G is abelian, the coadjoint action is trivial:
Ad∗ (g)ξ = ξ
for all g ∈ G and ξ ∈ g∗ . Equivariance of µ therefore means that for all
g ∈ G and (q, p) ∈ M ,
µ(g · (q, p)) = µ(q, p).
Since both sides equal p, µ is indeed equivariant.
Let X ∈ g = R. Then the function µX ∈ C ∞ (M ) is given by
µX (q, p) = Xp
for all (q, p) ∈ M . Hence
dµx = Xdp.
On the other hand, note that for all (q, p) ∈ M ,
  
d  d  d 
XM (q, p) =  exp(tX)·(q, p) =  (tX)·(q, p) =  (q+tX, p) = (X, 0).
dt t=0 dt t=0 dt t=0
Here we have used Exercise 8.6. So
∂
XM = X .
∂q
Hence, for all v ∈ X(M ),
 
∂
(iXM ω)(v) = (dp ∧ dq) X , v
∂q
   
∂ ∂
= dp X dq(v) − dp(v)dq X
∂q ∂q
= Xdp(v).

53
We conclude that indeed,

dµx = Xdp = −iXm ω.

Exercise 11.2. Let a Hamiltonian action by a Lie group G on a symplectic

manifold (M, ω) be given. Let µ : M → g∗ be a momentum map. Let
H < G be a closed subgroup, and consider the map

p : g∗ → h∗

given by restricting linear functions on g to h.

Prove that the action by H on (M, ω) is Hamiltonian, with momentum
map µH = p ◦ µ.

Exercise 11.3. Let a Hamiltonian action by a Lie group G on a symplectic

manifold (M, ω) be given. Let µ : M → g∗ be a momentum map. Let
N ⊂ M be a G-invariant symplectic submanifold, and consider the inclusion
map
ι : N ,→ M.
Then, by assumption, ι∗ ω is a symplectic form on N .
Prove that the action by H on N is Hamiltonian, with momentum map
µN = ι∗ µ.

Exercise 11.4. Let (M, ω) be a symplectic manifold, and let a Hamiltonian

action by a Lie group G on (M, ω) be given. Let µ : M → g∗ be a momentum
map. Let X ∈ g, and let vµX be the Hamiltonian vector field of µX .

(a) What is the Hamiltonian vector field of the function µX , for X ∈ g?

(b) Prove that for all X, Y ∈ g,

{µX , µY } = −µ[X,Y ] .

Proof. (a) By definition of momentum maps, one has

dµX = −ω(XM , −).

This exactly means that XM is the Hamiltonian vector field of µX .

54
 (b) By equivariance of µ, and by part (a), we have for all X, Y ∈ g and
m ∈ M,

{µX , µY }(m) = XM (µY )(m)


d 

=  µY exp(tX) · m
dt t=0

d 
Ad∗ (exp(tX))µ(m) (Y )


= 
dt t=0

d 

=  (µ(m)) Ad(exp(−tX))Y
dt t=0
  
d 
= (µ(m)) Ad(exp(−tX))Y
dt t=0
= (µ(m)) (− ad(X)Y )
= −µ[X,Y ] (m).

Exercise 11.5. Let ϕ : G → H be a Lie group homomorphism between

two Lie groups G and H. Let (M, ω) be a symplectic manifold, and let
a Hamiltonian action by H on (M, ω) be given. Let µH : M → h∗ be a
momentum map for this action.
Consider the action by G on M given by

g · m := ϕ(g) · m,

for g ∈ G and m ∈ M . Prove that this action is Hamiltonian, with momen-

tum map µG : M → g∗ given by

µG (m) (X) = µH (m) (Te ϕ(X)).

Remark 11.1. This exercise generalises Exercise 11.2, where ϕ is the in-
clusion map of a closed subgroup. (The roles of G and H are interchanged
between the two exercises.)
Proof. The action by G is symplectic, because the action by H is.
Let us check that µG is G-equivariant. For g ∈ G, m ∈ M and X ∈ g,
the fact that µH is H-equivariant implies that

µG (g · m) (X) = µH (ϕ(g) · m) (Te ϕ(X))

= Ad∗ (ϕ(g))µH (m) (Te ϕ(X))

= (µH (m)) Ad(ϕ(g)−1 ) ◦ Te ϕ(X) .

55
Now by the chain rule, one has

Ad(ϕ(g)−1 ) ◦ Te ϕ = Te Cϕ(g)
H


−1 ◦ ϕ ,

H
where Cϕ(g) −1 in H. Since ϕ is a homomor-
−1 denotes conjugation by ϕ(g)
phism, we have
H G
Cϕ(g) −1 ◦ ϕ = ϕ ◦ Cg −1 .

Hence
µG (g · m) (X) = (µH (m)) Te (ϕ ◦ CgG−1 (X)

= (µH (m)) Te ϕ ◦ Ad(g −1 )(X)

= Ad∗ (g)µH (m) (Te ϕ(X))

= Ad∗ (g)µG (m) (X).

Next, we compute dµG

X , for X ∈ g. First, note that for all m ∈ M ,

µG H
X (m) = µTe ϕ(X) (m).

Also, by Exercise 8.10,

 

d  d 
Te ϕ(X) M (m) =  exp(Te ϕ(X))·m =  ϕ(exp(X))·m = XM (m).
dt t=0 dt t=0

(Here the left hand side refers to the vector field induced by the infinitisimal
action by h, while on the right hand side, the infinitesimal action by g is
used.) We conclude that

dm µG H




X = dm µTe ϕ(X) = −ωm Te ϕ(X) M (m), − = −ωm XM (m), − .

12 Symplectic reduction
In this set of exercises, we prove a slight simplification of Theorem 5.17 in
the lecture notes.
Let (M, ω) be a symplectic manifold, and let G be a Lie group. Let a
Hamiltonian action by G on (M, ω) be given, and let µ : M → g∗ be a
momentum map. Let ξ ∈ g∗ be a regular value of µ, i.e. for all m in the
nonempty subset µ−1 (ξ) ⊂ M , the tangent map

Tm µ : Tm M → g∗

56
is surjective. Then µ−1 (ξ) is a smooth submanifold of M , by the submersion
theorem.
ιξ : µ−1 (ξ) ,→ M
be the inclusion map.
Let Gξ be the stabiliser of ξ with respect to the coadjoint action, and
suppose Gξ acts properly and freely on µ−1 (ξ). Then
Mξ := µ−1 (ξ)/Gξ
is a smooth manifold, as we saw in Section 10. Let
πξ : µ−1 (ξ) → Mξ
be the quotient map.
Theorem 12.1 (Marsden–Weinstein). There is a unique symplectic form
ωξ on Mξ , such that
πξ∗ ωξ = ι∗ξ ω ∈ Ω2 (µ−1 (ξ)). (16)
Note that the quotient map πξ is a submersion.7 Hence the tangent map
Tm πξ : Tm µ−1 (ξ) → TGm Mξ
is surjective for all m ∈ µ−1 (ξ). Hence (16) indeed determines ωξ uniquely.
Indeed, any two tangent vectors in TGm Mξ are of the form
Tm πξ (v), Tm πξ (w) ∈ TGm Mξ ,
for v, w ∈ Tm µ−1 (ξ). (Though these may be equal even if v and w are
different.)
Exercise 12.1. Show that for all for v, w ∈ Tm µ−1 (ξ),


(ωξ )Gm Tm πξ (v), Tm πξ (w) = ωm (v, w), (17)
where we consider Tm µ−1 (ξ) as a subspace of Tm M via the map Tm ιξ .
Proof. The relation (16) implies that
(ωξ )Gm Tm πξ (v), Tm πξ (w) = (πξ∗ ωξ )m (v, w)

= (ι∗ξ ω)m (v, w)

= ωm (Tm ιξ (v), Tm ιξ (w))
= ωm (v, w).

7
See the proof of Theorem 12.5 in the notes on Lie groups.

57
 It therefore remains to show that

1. ωξ is well-defined by (17);

2. ωξ is nondegenerate;

3. ωξ is closed.

The first two points will follow from the facts in the following exercise.

Exercise 12.2. Fix m ∈ µ−1 (ξ).

(a) Show that

Tm (G · m) = {XM (m); X ∈ g}. (18)

(b) Show that
ωm

Tm µ−1 (ξ) = Tm (G · m) ,
where the superscript ωm denotes the annihilator (orthogonal com-
plement) of a space with respect to ωm . (See page 5 in the lecture
notes.)

(c) Show that

Tm (Gξ · m) = ker(Tm πξ ).
Hint: use the slice lemma to prove that these spaces have equal di-
mensions.

Proof. (a) Note that for all X ∈ g and m ∈ M ,


d 
XM (m) =  exp(tX)m ∈ Tm (G · m),
dt t=0

since exp(tX)m ∈ G · m for all t. Hence

Tm (G · m) ⊃ {XM (m); X ∈ g}.

We have seen in Exercise 10.2 that the map X 7→ XM (m) from g to

Tm M is injective. And the map g 7→ g · m from G to M is injective as
well. So both sides of (18) have the same dimension (the dimension of
G), and they must be equal.

58
 (b) By the submersion theorem, one has

Tm µ−1 (ξ) = ker(Tm µ).

Now for all v ∈ Tm M and X ∈ g,

Tm µ(v) (X) = dm µX (v) = −ωm (XM (m), v).

So v ∈ ker(Tm µ) if and only if it is orthogonal to all tangent vectors

of the form XM (m), for X ∈ g. By part (a), this yields the desired
result.

(c) Note that Tm (Gξ · m) equals the vector space of all derivatives γ 0 (0)
of curves γ in Gξ · m with γ(0) = m. Since πξ (Gξ · m) is a single point
in M/G, for such curves, the composition πξ ◦ γ is constant. Hence
γ 0 (0) ∈ ker(Tm πξ ). We conclude that Tm (Gξ · m) ⊂ ker(Tm πξ ).
We finish the proof by showing that these spaces have equal dimen-
sions. Indeed, since Gξ acts freely on µ−1 (ξ), we have

dim Mξ + dim Gξ = dim µ−1 (ξ).

(This follows from the slice lemma.) By the rank-nullity theorem, we

therefore have

dim ker(Tm πξ ) = dim µ−1 (ξ) − dim Mξ = dim Gξ = dim(Gξ · m).

In the last equality, we have used that Gξ acts freely on µ−1 (ξ), so (for
example by the slice lemma) there is a diffeomorphism Gξ · m ∼ = Gξ .

Exercise 12.3. Prove that ωξ is well-defined by (17). I.e. if m ∈ µ−1 (ξ),

v, w ∈ Tm µ−1 (ξ), and Tm πξ (v) = 0, then ωm (v, w) = 0. (And similarly if
Tm πξ (w) = 0.)
Proof. If Tm πξ (v) = 0, then by part (c) of Exercise 12.2, one has

v ∈ Tm (Gξ · m) ⊂ Tm (G · m)

By part (b) of Exercise 12.2, one therefore has

ωm (v, w) = 0

for all w ∈ Tm µ−1 (ξ). By antisymmetry of ω, one also gets ωm (w, v) = 0 for
all such w.

59
Exercise 12.4. Let W be a vector space, and b a bilinear form on W . Let
V ⊂ W be a linear subspace. Prove that

(V b )b = V.

Hint: For the inclusion (V b )b ⊂ V , consider a vector w ∈ (V b )b , and the

annihilator of the space U := V + Rw.

Proof. If v ∈ V , then by definition of the annihilator V b , one has

b(v, w) = 0

for all w ∈ V b . Hence V ⊂ (V b )b .

Conversely, suppose that w ∈ (V b )b . Write U := V + Rw. Then V ⊂ U ,
so U b ⊂ V b . And vice versa, if x ∈ V b , then b(x, w) = 0, so x ∈ U b . Hence
V b = U b , so that V = U . We conclude that w ∈ V .

Exercise 12.5. Prove that ωξ is nondegenerate.

Proof. Let v ∈ Tm µ−1 (ξ), and suppose that for all w ∈ Tm µ−1 (ξ),


ωGm Tm πξ (v), Tm πξ (w) = 0.

We will show that Tm πξ (v) = 0. In Exercise 12.1, we saw that

ωGm Tm πξ (v), Tm πξ (w) = ωm (v, w)

for all w ∈ Tm µ−1 (ξ). This equals zero for all such w if and only if

ω
ω
v ∈ (Tm µ−1 (ξ))ωm = Tm (G · m) m m ,

by Exercise 12.2 (b). By Exercise 12.4, the right hand side equals Tm (G·m).
So there is an X ∈ g such that v = XM (m), by Exercise 12.2 (a). Since
v ∈ Tm µ−1 (ξ) = ker Tm µ, equivariance of µ implies
 
d  d 
0 = Tm µ(XM (m)) =  µ(exp(tX)m) =  exp(tX)ξ = Xg∗ (ξ).
dt t=0 dt t=0

We conclude that X ∈ gξ , so

v = XM (m) ∈ Tm (Gξ · m) = ker Tm πξ .

So indeed Tm πξ (v) = 0, and ωξ is nondegenerate.

Exercise 12.6. (a) Argue that ωξ is closed if and only if πξ∗ (dωξ ) = 0.

60
 (b) Prove that ωξ is closed.

Proof. (a) Since πξ is a submersion, its tangent map is surjective. This

implies that the pullback

πξ∗ : Ω(Mξ ) → Ω(µ−1 (ξ))

is injective. Hence, if πξ∗ d(ωξ ) = 0, then dωξ = 0. (The converse

implication follows from linearity of πξ∗ .)

(b) Because dω = 0, we have

πξ∗ (dωξ ) = d(πξ∗ ωξ )

= d(i∗ξ ω)
= i∗ξ (dω)
= 0.

13 Cotangent bundles
Let N be a smooth manifold, and consider its cotangent bundle T ∗ N . Let
θ ∈ Ω(T ∗ N ) be the tautological one-form, and let ω := dθ be the standard
symplectic form on T ∗ N .
Let G be a Lie group acting smoothly on N . The induced action on T ∗ N
is given by
(g · η)(v) = η(Tg·n g −1 (v)), (19)
for g ∈ G, n ∈ N , η ∈ Tn∗ N and v ∈ Tg·n N .

13.1 Momentum map

Exercise 13.1. (a) Prove that θ is G-invariant.

(b) Prove that the action is symplectic.

(c) Prove that for all X ∈ g,

d(iXT ∗ N θ) = −iXT ∗ N ω.

Here XT ∗ N is the vector field on T ∗ N induced by X via the infinites-

imal action.

61
Proof. (a) By (19), the cotangent bundle projection map π is G-equivariant.
Let g ∈ G, n ∈ N , η ∈ Tn∗ N , and v ∈ Tη (T ∗ N ). Then

(g ∗ θ)η (v) = θg·η Tη g(v)

= (g · η) Tg·η π ◦ Tη g(v)
= η Tg·n g −1 ◦ Tgη̈ π ◦ Tη g(v)

= η(Tη (g −1 ◦ π ◦ g)(v))
= η(Tη π(v))
= θη (v).

Here we have used equivariance of π.

(b) One has for all g ∈ G,

g ∗ ω = g ∗ (dθ) = d(g ∗ θ) = dθ = ω.

(c) By Cartan’s formula for the Lie derivative, we have for all X ∈ g∗

0 = LXT ∗ N θ = d(iXT ∗ N θ) + iXT ∗ N (dθ) = d(iXT ∗ N θ) + iXT ∗ N ω.

Consider the map µ : T ∗ N → g∗ given by

µ(η) (X) = η XN (n) ,

for n ∈ N , η ∈ Tn∗ N and X ∈ g. Let π : T ∗ N → N be the cotangent bundle

projection.

Exercise 13.2. (a) Prove that for all X ∈ g, n ∈ N and η ∈ Tn∗ N ,

Tη π XT ∗ N (η) = XN (n) ∈ Tn N.

(b) Prove that for all X ∈ g,

µX = iXT ∗ N θ ∈ C ∞ (T ∗ N ).

(c) Prove that for all X ∈ g,

dµX = −iXT ∗ N ω.

62
Proof. (a) Let X ∈ g, n ∈ N and η ∈ Tn∗ N . Since π is equivariant, we
have
  

d 
Tη π XT ∗ N (η) = Tη π exp(tX)η
dt t=0

d 

=  π exp(tX)η
dt t=0

d 
=  exp(tX)n
dt t=0
= XN (n).

(b) Let X ∈ g, n ∈ N and η ∈ Tn∗ N . Then because of part (a),

(iXT ∗ N θ)(η) = θη XT ∗ N (n)


= η Tη π(XT ∗ N (n))
= η(XN (n))
= µX (η).

(c) By part (c) of Exerise 13.1 and part (b) of this exercise, we conclude
that for all X ∈ g,

dµX = d(iXT ∗ N θ) = −iXT ∗ N ω.

It remains to check equivariance of µ.

Exercise 13.3. (a) Prove that for all X ∈ g, n ∈ N and g ∈ G,

Tn g(XN (n)) = (Ad(g)X)N (g · n).

(b) Prove that µ is equivariant.

Proof. (a) By Exercise 8.10, we have


d 
Tn g(XN (n)) =  g exp(tX)n
dt t=0

d 
=  exp(t Ad(g)X)g · n
dt t=0
= (Ad(g)X)N (g · n).

63
 (b) Because of part (a), we see that for X ∈ g, n ∈ N , η ∈ Tn∗ N and
g ∈ G,



µ(g · η) (X) = (g · η) XN (gn)
= η Tn g −1 (XN (gn))

= η (Ad(g)−1 X)N (n)

= µ(η) Ad(g)−1 (X)

= Ad∗ (g)(µ(η)) (X).

13.2 Symplectic reduction

We have seen that the action by G on T ∗ N is Hamiltonian, with momentum
map µ. We now suppose that G acts properly and freely on N . Then N/G is
a smooth manifold. Let θG ∈ Ω1 (T ∗ (N/G)) be the tautological one-form on
T ∗ (N/G), and let ωG := dθG be the standard symplectic form on T ∗ (N/G).
We are going to construct a symplectomorphism

(T ∗ N )0 , ω0 ∼ = T ∗ (N/G), ωG .

Here (T ∗ N )0 , ω0 is the symplectic reduction at zero of T ∗ N by the action

by G.
Consider the quotient map q : N → N/G. For n ∈ N , the tangent map
Tn q dualises to
(Tn q)∗ : TG·n
∗
(N/G) → Tn∗ N,
(Tn q)∗ (ζ) (v) = ζ Tn q(v) ,

∗ (N/G) and v ∈ T N .
for n ∈ N , ζ ∈ TGn n

Exercise 13.4. Prove that for all n ∈ N

(Tn q)∗ TG·n

∗
(N/G) ⊂ µ−1 (0).

∗ (N/G) and X ∈ g. Then

Proof. Let n ∈ N , ζ ∈ TG·n

µ (Tn q)∗ ζ (X) = (Tn q)∗ ζ (XN (n))

= ζ Tn q(XN (n)) .
Now  
d  d 
Tn q(XN (n)) =  q(exp(tX)n) =  G · n = 0.
dt t=0 dt t=0

64
 Consider the quotient map p : µ−1 (0) → M0 = µ−1 (0)/G, and the
inclusion map ι : µ−1 (0) ,→ T ∗ N . Then, for all n ∈ N , we have the diagram
(Tn q)∗ 
∗ (N/G)
TG·n / µ−1 (0)  ι / T ∗ N.

p
p◦(Tn q)∗ & 
(T ∗ N )0

Consider the map

Ψ : T ∗ (N/G) → (T ∗ N )0
given by
Ψ(ζ) = p ◦ (Tn q)∗ (ζ),

∗ (N/G).
for all ζ ∈ TG·n
Exercise 13.5. Show that the map Ψ is well-defined, in the sense that for
∗ (N/G) and g ∈ G,
all n ∈ N , ζ ∈ TG·n

p ◦ (Tg·n q)∗ (ζ) = p ◦ (Tn q)∗ (ζ).

Proof. Let n ∈ N , g ∈ G and ζ ∈ TG·n ∗ N . Then for all v ∈ T

g·n N , we have

(Tg·n q)∗ ζ (v) = ζ Tg·n q(v)

= ζ (Tn q ◦ Tg·n g −1 )(v)

= g · (Tn∗ q)ζ (v).

Hence (Tg·n q)∗ ζ = g · (Tn∗ q)ζ, and the claim follows.

We first show that Ψ is a diffeomorphism.

Exercise 13.6. (a) Let f : X → Y be a submersion between smooth
manifolds X and Y . Prove that, for each x ∈ X, the map

(Tx f )∗ : Tf∗(x) Y → Tx∗ X,

defined in the same way as Tn∗ q above, is injective.

(b) Prove that Ψ is injective.
Proof. (a) Let x ∈ X, and ζ ∈ Tf (x) Y such that (Tx f )∗ ζ = 0. Then for
all v ∈ Tx X,
0 = (Tx f )∗ ζ (v) = ζ Tx f (v) .

Because Tx f is surjective, all tangent vectors in Tf (x)Y are of the form

Tx f (v), with v as above. Hence ζ = 0.

65
 (b) Let n ∈ N . Then the map (Tn q)∗ : TG·n (N/G) → Tn∗ N is injective
by part (a). Let ζ, ζ 0 ∈ TG·n
∗ (N/G), and suppose that Ψ(ζ) = Ψ(ζ 0 ).

Then there is a g ∈ G such that

(Tn q)∗ ζ 0 = g · (Tn q)∗ ζ.

Since the left hand side is in Tn∗ N and the right hand side is in Tg·n∗ N,

we must have g · n = n. Since the action is free, this means that g = e.

Hence (Tn q)∗ ζ 0 = (Tn q)∗ ζ, and ζ = ζ 0 by injectivity of (Tn q)∗ .

Next, we use the fact that (T ∗ N )0 is a vector bundle over N/G. The
vector bundle projection map is induced by the equivariant cotangent bundle
projection π : µ−1 (0) → N .

Exercise 13.7. Prove that, at all G · n ∈ N/G, the map Ψ is a linear

isomorphism
∗ ∼
=
→ (T ∗ N )0 G·n .


ΨG·n : TG·n (N/G) −

Proof. The map ΨG·n is linear, and injective by Exercise 13.6. Since the
vector spaces that ΨG·n have the same dimension, it is also surjective.

The above exercise implies that Ψ is a bijection. It follows from the

theory of vector bundles that Ψ, and its inverse, are smooth. Hence Ψ is
indeed a diffeomorphism.
It remains to show that the diffeomorphism Ψ is in fact a symplectomor-
phism. Fix a point n0 ∈ N . We will prove the equality

Ψ∗ ω0 = ωG .

locally, in a neighbourhood n0 . Since n0 is an arbitrary point in N , this

proves the claim.
Since the action by G on N is proper and free, the slice lemma implies
that there is a G-invariant neighbourhood U of G · n0 in N/G such that
there is an equivariant diffeomorphism
∼
=
τ : q −1 (U ) −
→ U × G.

This allows us to choose a smooth map

σ : U → q −1 (U )

66
such that σ(G · n) ∈ G · n for all G · n ∈ U . For example, one can set

σ(G · n) := τ −1 (G · n, e).

Using σ, we define the map

Tσ∗ q : T ∗ (N/G) → T ∗ N

by
Tσ∗ q(η) = (TσG·n q)∗ η,
∗ (N/G). By definition of the map Ψ, we have
for all η ∈ TG·n

Ψ|U = p ◦ Tσ∗ q

on U .
Now consider the diagram

T ∗ (N/G) ⊃ U
(Tσ∗ q)
/ µ−1 (0)   ι / T ∗ N.

p
Ψ ' 
(T ∗ N )0

Exercise 13.8. (a) Prove that, on U ,

q ◦ π ◦ Tσ∗ q = πG |U ,

where π : T ∗ N → N is the cotangent bundle projection of N , and

πG : T ∗ (N/G) → N/G is the cotangent bundle projection of N/G.

(b) Prove that
∗

ι ◦ Tσ∗ q θ = θG |U .

(c) Prove that

(Ψ∗ ω0 )|U = ωG |U .

Proof. (a) Let η ∈ U , suppose η ∈ TG·n ∗ N . Suppose n = σ(G · n). (Other-

wise, replace n by σ(G · n)). Then, since (Tn q)∗ η ∈ Tn N ,

(q ◦ π ◦ Tσ∗ q)(η) = q ◦ π (Tn q)∗ η

=G·n
= πG (η).

67
 (b) Let G · n ∈ U be given. As above, suppose n = σ(G · n). Let η ∈
∗ (N/G) and v ∈ T (T ∗ (η)). Then by part (a),
TG·n η

∗

ι ◦ Tσ∗ q θ η (v) = θTσ∗ q(η) Tη (Tσ∗ q)(v)

= Tσ∗ q(η) TTσ∗ q(η) π ◦ Tη (Tσ∗ q)(v)

= η (Tn q) ◦ TTσ∗ q(η) π ◦ Tη (Tσ∗ q)(v)

= η Tη (q ◦ π ◦ Tσ∗ q)(v)




= η Tη πG (v)
= (θG )η (v).

(c) By definition of the symplectic form ω0 , we have

p∗ ω0 = ι∗ ω.

Hence, on U (where we omit the restriction |U to U in the notation),

we have
Ψ∗ ω0 = (p ◦ Tσ∗ q)∗ ω0
= (Tσ∗ q)∗ p∗ ω0
= (Tσ∗ q)∗ ι∗ ω
= d(ι ◦ Tσ∗ q)∗ θ
= dθG
= ωG .

14 Coadjoint orbits
Let G be a Lie group, with Lie algebra g. Let g∗ be the dual vector space
of g. Consider the coadjoint action (representation)

Ad∗ : G → GL(g∗ )

defined by
Ad∗ (g)ξ (X) = ξ(Ad(g −1 )X),

for all g ∈ G, ξ ∈ g∗ and X ∈ g. This induces the infininitesimal coadjoint

action
ad∗ : g → End(g),

68
given by
ad∗ = Te Ad∗ .
That is,

ad(X)ξ (Y ) = −ξ([X, Y ]),
for all X, Y ∈ g, ξ ∈ g∗ .

Definition 14.1. A coadjoint orbit of G is an orbit of the coadjoint action

Ad∗ . That is, a subset O of g∗ of the form

O = Ad∗ (G)ξ = {Ad∗ (g)ξ; g ∈ G},

for some ξ ∈ g∗ .

Exercise 14.1. Let O be a coadjoint orbit of G, and let ξ ∈ O be a point

on O. Show that8

Tξ O = {Xξ ; X ∈ g} = {ad∗ (X)ξ; X ∈ g} ⊂ g∗ .

Proof. For all ξ ∈ O and X, Y ∈ g, we have


d
Ad∗ (exp(tX))ξ (Y )


Xξ (Y ) = 
dt t=0

d 

=  ξ Ad(exp(−tX))Y
dt t=0
= −ξ(ad(X)Y )
= (ad∗ (X)ξ)(Y ).

Let O be a coadjoint orbit.

Definition 14.2. The Kostant–Kirillov–Souriau two-form ω on O is given

by
ωξ (Xξ , Yξ ) := −ξ([X, Y ]),
for all ξ ∈ O and X, Y ∈ g.

We will show that the form ω is a well-defined symplectic form on O.

8
In this section, we write Xξ instead of XO (ξ) for the value at ξ of the vector field XO
on O induced by the infinitesimal action.

69
Exercise 14.2. (a) Prove that the above expression for ω is well-defined.
I.e. if ξ ∈ O, X, Y ∈ g and Xξ = 0, then −ξ([X, Y ]) = 0, and similarly
for Y .

(b) Prove that the form ω is nondegenerate.

(c) Prove that the form ω is closed.

Hint: For part (c), show that for all X ∈ g ,→ C ∞ (g∗ ),

iX ω = −dX.

Then show that

iX dω = 0
for all such X.

Proof. (a) We first show that ω is well-defined. Let X, Y ∈ g, and suppose

Xξ = 0. Then

ξ([X, Y ]) = (ad∗ (X)ξ)(Y ) = Xξ (Y ) = 0.

By antisymmetry, it also follows that ξ([X, Y ]) = 0 if Yξ = 0. By

bilinearity, this implies that ω is well-defined.

(b) A very similar argument shows that ω is nondegenerate. Indeed, let

X ∈ g, ξ ∈ O, and suppose

ωξ (Xξ , Yξ ) = 0

for all Y ∈ g. Then

(ad∗ (X)ξ)(Y ) = Xξ (Y ) = 0

for all Y , so Xξ = 0. Hence ω is nondegenerate.

(c) To see that ω is closed, note that for all X ∈ g,

iX ω = −dX,

for all
X ∈ g ,→ C ∞ (g∗ ).
Indeed, for all Y ∈ g and ξ ∈ g∗ , one has

dξ X(Yξ ) = Yξ (X) = ad∗ (Y )ξ (X) = ξ([X, Y ]) = −ωξ (Xξ , Yξ ).

70
 Hence
d(iX ω) = −ddX = 0.
And by G-invariance of ω,

iX dω = iX dω + d(iX ω) = LX ω = 0,

for all X ∈ g. So dω = 0.

Exercise 14.3. Show that the symplectic form ω is G-invariant.

Proof. Let g ∈ G, ξ ∈ O and X, Y ∈ g be given. Then

(g ∗ ωξ (Xξ , Yξ ) = ωgξ̇ Tξ g(Xξ ), Tξ g(Yξ ) .

Now
  
d 
Tξ g(Xξ ) = Tξ g exp(tX)ξ
dt t=0

d 
=  g · exp(tX)ξ
dt t=0

d 
=  exp(t Ad(g)X)g · ξ
dt t=0
= (Ad(g)X)g·ξ ,

and similarly for Y . Hence

(g ∗ ωξ (Xξ , Yξ ) = ωgξ̇ (Ad(g)X)g·ξ , (Ad(g)Y )g·ξ

= −(g · ξ) [Ad(g)X, Ad(g)Y ]
= −ξ Ad(g −1 )[Ad(g)X, Ad(g)Y ]




= −ξ [X, Y ]
= ωξ (Xξ , Yξ ),

since Ad(g) is a Lie algebra homomorphism.

The coadjoint action by G on O is in fact Hamiltonian.

Exercise 14.4. Prove that the inclusion map

µ : O ,→ g∗

is a momentum map for the coadjoint action by G on O.

71
Proof. Equivariance of the inclusion map is immediate.
Let X ∈ g and ξ ∈ O. Then

µX (ξ) = ξ(X).

Hence for all Y ∈ g,


d 
dξ µX (Yξ ) =  µX (Ad∗ (exp(tY ))ξ)
dt t=0

d 
=  ξ(Ad(exp(−tY ))X)
dt t=0
  
d 
=ξ Ad(exp(−tY ))X
dt t=0
= ξ([X, Y ])
= −ωξ (Xξ , Yξ ).

Alternatively, note that µX = X, where X is viewed as a linear function

on g∗ . We saw above that
iX ω = −dX.

15 Smooth projective manifolds

Consider the complex vector space Cn , equipped with the standard hermi-
tian form h:
n
X
h(z, z 0 ) = zj z̄j0 ,
j=1

for z, z 0 ∈ Cn . Then
h = B + iω,
with B the standard inner product on R2n ∼= Cn , and ω the standard sym-
plectic form on R2n ∼
= Cn .
The natural action by U(n) on Cn preserves h (by definition), hence also
B and ω. Hence the action is symplectic.

Exercise 15.1. Prove that the Lie algebra of U(n) is

u(n) = {X ∈ Mn (C); X ∗ + X = 0}.

Here X ∗ is the conjugate transpose of X.

72
 Hint: see Exercise 8.4.
We will show that the map

µ : Cn → u(n)∗ ,

given by

i
µ(z) (X) = h(Xz, z),
2
n
for z ∈ C and X ∈ u(n), is a momentum map for this action.

Exercise 15.2. Prove that the map µ is equivariant.

Proof. Let z ∈ Cn , X ∈ u(n) and g ∈ U(n). Then

i
µ(g · z) (X) = h(Xgz, gz).
2
For linear groups such as U(n), we have

Xg = g Ad(g −1 )X,

for group elements g and Lie algebra elements X. Hence, since g ∈ U(n)
preserves h,
i i
h(Xgz, gz) = h(g Ad(g −1 )Xz, gz)
2 2
i
= h(Ad(g −1 )Xz, z)
2
= µ(z)(Ad(g −1 )X)
= Ad∗ (g)µ(z) (X).

Exercise 15.3. Prove that for all z ∈ Cn , v ∈ Tz CN ∼

= Cn and X ∈ g,

(dµX )z (v) = −ω(XCn (z), v).

Hint: Use Exercise 9.2. What is the simplest curve γ in Cn with γ(0) = z
and γ 0 (0) = v?

Proof. Let z ∈ Cn , v ∈ Tz CN ∼
= Cn and X ∈ g. Note that

d 
v =  (z + tv).
dt t=0

73
Hence

d 
(dµX )(v) =  µX (z + tv)
dt t=0

d  i

=  h X(z + tv), z + tv
dt t=0 2

i d  
2

= h(Xz, z) + t h(Xz, v) + h(Xv, z) + t h(Xv, v)
2 dt t=0
i

= h(Xz, v) + h(Xv, z) .
2
Now, since X ∗ = −X, one has

h(Xv, z) = −h(v, Xz) = −h(Xz, v).

Hence
i
i

h(Xz, v)+h(Xv, z) = h(Xz, v)−h(Xz, v) = − Im(h(Xz, v)) = −ω(Xz, v).
2 2
By Exercise 9.2, we have Xz = XCn (z), and we are done.

Exercise 15.4. Let G be a Lie group, and ρ : G → U(n) a Lie group

homomorphism. Prove that the induced action by G on Cn is Hamiltonian,
and give a momentum map.

Hint: Use Exercise 11.4. Your answer may be short.

Proof. By Exercise 11.4, the action is indeed Hamiltonian, and a momentum

map is
i
(µG (z))(X) = (µU(n) (z))(Te ρ(X)) = h(Te ρ(X)z, z)
2
for z ∈ Cn and X ∈ g.

74