symplectic-geometry-lecture-notes

Symplectic geometry
Gilles Castel
June 11, 2020

Contents
1 Symplectic linear algebra 3

1.1 Symplectic vector spaces . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Subspaces and reduction . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Compatible complex structures . . . . . . . . . . . . . . . . . . . 6
2 Symplectic manifolds 9
2.1 Vector bundles, vector fields, differential forms . . . . . . . . . . 9
2.2 Symplectic manifolds . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Basic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Hamiltonian vector fields . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Hamiltonian Mechanics . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Almost Kähler Manifolds . . . . . . . . . . . . . . . . . . . . . . 18
3 Submanifolds and Normal Forms 20

3.1 Isotopes, homotopy operators and tubular neighbourhoods . . . . 20
3.2 Darboux’ theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Lagrangian submanifolds . . . . . . . . . . . . . . . . . . . . . . . 29
4 Review of Lie groups 38

4.0 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.1 Lie algebra cohology . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Adjoint representation . . . . . . . . . . . . . . . . . . . . . . . . 41
5 Lie groups 44
5.1 Lie group actions . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Moment maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.3 Examples of moment maps . . . . . . . . . . . . . . . . . . . . . 48
5.4 Existence and uniqueness of moment maps . . . . . . . . . . . . . 53
6 Symplectic reduction 58
6.1 Symplectic reduction theorem . . . . . . . . . . . . . . . . . . . . 58
6.2 Examples of symplectic reduction . . . . . . . . . . . . . . . . . . 62
6.3 Reduction in stages . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.4 The convexity theorem . . . . . . . . . . . . . . . . . . . . . . . . 66
1
Lecture 1: Introduction
wo 12 feb 10:38
• Charlotte Kirchhofflukat
• Main text: Ana Cannas da Silva Lectures on symplectic geometry
CONTENTS 2
Chapter 1
Symplectic linear algebra
1.1 Symplectic vector spaces

Definition 1. A symplectic vector space is a pair (V, Ω) st.
• V is a real vector space

• Ω:V ×V →R
– antisymmetric
– ker Ω = {0}.
Exercise. (V, Ω). Then Ω̃ : V → V ∗ : v 7→ Ω(v, ·) is an isomorphism.

Exercise. Let A an n × n matrix, Ω(v, w) = v T Aw bilinear form. Show Ω is non
degenerae if det A 6= 0.
Exercise. A ∈ Mn×n (R) such that AT = −A. If n is odd, then det A = 0.
Corollary. A symplectic vector space is even dimensional.
Example (Standard example of symplectic vector space). Consider R2n with stan-
dard basis coordinates (e1 , . . . , en , f1 , . . . , fn ), with Ωcan the unique antisym-
metric bilinear form s.t.
Ω(ei , ej ) = 0 Ω(fi , fj ) = 0 Ω(ei , fj ) = δij .
The matrix representation is given by

T 0 I
Ωcan (v, w) = v w.
−I 0

Exercise. Show that this is a symplectic vector space.
3
Definition 2. Let (V, Ω) be a symplectic vector space, W ⊂ V a subspace.

The symplectic orthogonal to W is
W Ω := {v ∈ V | Ω(v, w) = 0 ∀w ∈ W }.
Exercise. What is the symplectic orthogonal to the following spaces in (R2n , Ωcan )
• (R2n )Ω = {0}, because Ω is non-degenerate.
• {0}Ω = R2n , because Ω is bilinear.
• span(e1 , . . . , en )
• span(e1 , f1 )
• span(e1 , . . . , en , f1 )
Exercise. (W Ω )Ω = W for all subspaces W ⊂ V .
Exercise. Show that dim W + dim W Ω = dim V.
Exercise. W Ω ∩ W = 0
iff (W, Ω|W ×W ) is a symplectic vector space
iff (W Ω , Ω|W Ω ×W Ω ) is a symplectic vector space.
Definition 3. Let (V1 , Ω1 ), (V2 , Ω2 ) be symplectic vector spaces. A linear

symplectic morphism is a linear isomorphism φ : V1 → V2 such that Ω1 =
φ ∗ Ω2 .
Exercise. Consider the standard symplectic vector space. A ∈ M2n×2n . Then

A is a linear symplectic morphism if

0 I 0 I
AT A= .
−I 0 −I 0
This forms a group: the symplectic group.
Proposition 1. If (V, Ω) is a symplectic vector space, then there exists a
φ : (R2n , Ωcan ) → (V, Ω),
for some n.
Proof. We know that dim V = 2n, for some n. Finding φ is equivalent to

finding basis e1 , . . . , en , f1 , . . . , fn st.
Ω(ei , ej ) = 0 Ω(fi , fj ) = 0 Ω(ei , fj ) = δij .
Base case dim V = 2. Pick any e1 6= 0. As Ω is non-degenerate, there is
CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 4

some f1 such that Ω(e1 , f1 ) 6= 0. Rescale and we’re done.

Induction Assume the claim holds for all k < n. Pick a pair (e1 , f1 ) such
that Ω(e1 , f1 ) = 1. Then W = span(e1 , f1 ) is a symplectic subspace.
We know that dim W Ω = 2n − 2. We also know that, because W is
a symplectic subspace, we must have W ∩ W Ω = 0, and W Ω is itself
also a symplectic vector space. By the inductive hypothesis, there
is a basis (e2 , . . . , en , f2 , . . . , fn ) of W Ω , which satisfy the relations.
Now, we have
Ω(e1 , ei ) = 0 Ω(e1 , fi ) = 0 Ω(f1 , ei ) = 0 Ω(f1 , fi ) = 0
So (e1 , . . . , en , f1 , . . . , fn ) is a basis, and it satisfies all he relations.
Corollary. Any two symplectic vector spaces are symplectormorphic if their

dimensions agree.
1.2 Subspaces and reduction

Definition 4. Let W ⊂ V be subspace. We call it
• symplectic iff W ∩ W Ω = 0.
• isotropic iff W ⊂ W Ω . (The symplectic form restricts to 0)

• coisotropic subspaces iff W Ω ⊂ W
• Lagrangian iff W Ω = W
Exercise. Are the subspaces from exercise 6 isotropic, coisotropic, lagrangian?

Exercise. If W is
• isotropic, then dim W ≤ dim V

2
• coisotropic, then dim W ≥ dim V

2
• lagrangian, then dim W = dim V

2
Exercise. If W ⊂ V coisotropic. Then Wred := W

WΩ is even dimensional.

Proposition 2. Let W ⊂ V coisotropic and let π : W → Wred = WWΩ . Then

there exists a bilinear form Ωred such that (Wred , Ωred ) is a symplectic
vector space and
Ωred (π(v1 ), π(v2 )) = Ω(v1 , v2 ).
This process is called the reduction of symplectic vector spaces.
Proof. Note that π is surjective.
Ωred (ṽ, w̃) = Ω(v, w), π(v) = ṽ, π(w) = w̃.
Why is this will defined? Suppose π(v1 ) = π(v2 ), π(w1 ) = π(w2 ). So

v1 − v2 ∈ W Ω . So w1 − w2 ∈ W Ω .
Ω(v1 , w1 ) = Ω(v1 − v2 + v2 , w1 − w2 + w2 )
= Ω(v1 − v2 , w1 − w2 ) + Ω(v2 , w1 − w2 ) + Ω(v1 − v2 , w2 ) + Ω(v2 , w2 )
= Ω(v2 , w2 ).
Antisymetry is okay, the only thing left to check if the non-degeneracy.

Take z ∈ Wred such that Ωred (z, z 0 ) = 0 for all z 0 ∈ Wred . We want to show
that z = 0. Pick v ∈ W such that π(v) = z. So Ω(v, w) = 0 for all w ∈ W ,
so that means that in the reduced vector space, z = 0.
1.3 Compatible complex structures

Definition 5. Let V be a real vector space. A complex structure on V is an
endomorphism J that squares to −1.
Exercise. Any such J induces a structure of a C-vector space!

We can define action of C on the vector: i · v := Jv. Indeed i · (i · v) = −v.
Definition 6. If (V, Ω) is a symplectic vector space. Then we call a complex

structure J compatible with the symplectic structure iff the bilinear form
g(v, w) = Ω(v, Jw)
is a scalar product:
• g(v, w) = g(w, v)
• g(v, v) ≥ 0 if v 6= 0.
Exercise. If J is some complex structure on V . Show that g(v, w) is symmetric

iff J is a linear symplectormorphism from (V, Ω) → (V, Ω).

Lecture 2
Example. Consider (R2n , Ωcan ). Then the can. compatible complex structure is
defined as
J(ei ) = fi J(fi ) = −ei .

0 I
J= ,
−I 0

I 0
g(v, w) = v T w.
0 I

Exercise. Show all symplectic vector spaces admit a compatible complex struc-
ture.
Exercise. Show there are infinitely many complex structures on R2n that are
compatible with Ωcan . Idea: QT JQ with a Q in the symplectic linear group.
Exercise. If V is a vector space and J : V → V a complex structure and a scalar
product. Then we can define
Ω(v, w) = g(v, −Jw).
Show this is symplectic iff g(Jv, Jw) = g(v, w).
Exercise. Let J1 , J2 be compatible complex structure on (V, Ω). Assume Ω(v, J1 w) =
Ω(v, J2 w), then J1 = J2 .
So if we have two of (Ω, J, g), we can always determine the other one.
Exercise (Recommended!). U (n) = {A | A† A = 1}. Consider the isomorphism
of real vector spaces
Re ⊕ Im : Cn → R2n .
Show that the action of U (n) on R2n given by this isomorphism is by symplec-
tormorphism.
Proof. Let U = A + iB. Then

A −B v
(A + iB)(v + iw) = = ...
B A w
Check that matrices of this form obey M T ΩM = Ω, for Ω = Ωcan.
Lecture 2
wo 19 feb 11:06
Proposition 3 (Lagrangian Grassmannian). The set
Lag(R2n ) = {W ⊂ R2n | W is Lagrangian}
has the structure of a smooth manifold. The Lagrangian is canonically

U (n) a
diffeomorphic to O(n) .
a O(n) is not a normal subgroup, so we don’t get a group.

Lecture 2
Proof. Suppose W ⊂ V Lagrangian, M ∈ Symp(Ωcan ), then M W is still

Lagrangian. U (n) acts on the Lagrangian Grassmannian. Show:
• The group action is transitive (one orbit)
Rn ⊂ Cn is Lagrangian after mapping with <⊕= (indeed, span of ei ).
Consider W ⊂ Cn , a real subspace, which is Lagrangian w.r.t Ωcan .
Pick an orthonormal basis of W w.r.t. the normal scalar product
on R2n . Preimage (< ⊕ =)−1 (w1 , . . . , wn ) is orthonormal w.r.t. the
hermitian inner product on Cn . (Exercise!)
Therefore, (w1 , . . . , wn ) is a unitary matrix. Also, it maps Rn ⊂ Cn
to W . This proves the transitivity.
• Stabiliser is O(n). For this, using the proof of the previous part, we
can simply consider the matrices that preserve the space Rn ⊂ Cn .
These are of course the real unitary matrices O(n).
Basic facts of Lie groups: U (n)/O(n) is smooth. This proves the state-
ment: simply steal the smooth structure from U (n)/O(n)

Chapter 2
Symplectic manifolds
2.1 Vector bundles, vector fields, differential forms
Definition 7. A k-form on a manifold M is a section of Λk T ∗ M . We write

Γ(Λk T ∗ M ) = Ωk (M ), Ω0 (M ) = C ∞ (M ), Ω• (M ) = ⊕k∈Z Ωk (M )
Exercise. Show there is a natural bijection
Ωk → {C ∞ alt. multilin. maps X(M ) × X(M ) . . . × X(M ) → C ∞ (M )}
Punchline: the set of differential forms is a differential graded commutative

graded.
Definition 8. A R-algebra is a (real) vector space A with a bilinear function

called ‘multiplication’ A × A → A : (a, b) 7→ a · b = ab which is associative.
It is called commutative when a · b = b · a.
Example. C ∞ (M ) is a commutative algebra.
Definition 9. Let A be an algebra. A derivation is an R-linear map D :

A → A that is compatible with · as follows:
D(a · b) = D(a) · b + a · D(b).
Remark. The set of all derivations on an algebra A, Der(A) forms a Lie algebra.
[D1 , D2 ] = D1 ◦ D2 − D2 ◦ D1 .
Remark. Der(C ∞ (M )) = X(M ).
9
Lecture 2
Definition 10. A Lie algebra (which is not an algebra in the above sense) is a
real vector space with an antisymmetric bilinear map [·, ·] : L×L → L called
the ‘Lie bracket’ which is not associative, but obeys the Jacobi identity:
[a, [b, c]] + [c, [a, b]] + [b, [c, a]] = 0,
or equivalently,
[a, [b, c]] = [[a, b], c] + [b, [a, c]],
which kinda looks like derivations.
Example. Let (A, ·) be an algebra. Then we can define

[a, b] = a · b − b · a,
which makes (A, [·, ·]) a Lie algebra.
Example. X(M ) forms a Lie algebra with the Lie bracket of vector fields.
Definition 11. A Z-graded vector space, is a vector space with a composi-

tion into a direct sum of subvectorspaces:
M
V = Vn .
n∈Z
Note that while this sum looks infinite, it is not in most of the cases.
Elements of Vn are called homogeneous elements of degree k.
Remark. Geometrically, we’re not interested in f + dx, or dx + dx ∧ dy or some-

thing like this, but these things live in Ω• . So these sums are formal.
Example. R[X] is a Z-graded vector space. 10x2 has homogeneous of degree 2,
...
Definition 12. A Z-graded algebra A is an algebra that is a Z-graded vector

space with the additional condition that for all k, j ∈ Z: Ak · Aj ⊂ Ak+j ,
i.e. the degrees behave additively w.r.t. the multiplication.
Definition 13. A is graded commutative if for all a ∈ Ak , b ∈ Aj :
a · b = (−1)kj (b · a).
Remark. If we use a trivial grading, and we put everything in A0 , then this

is the usual commutativity. If we use a proper grading however, this becomes
more interesting.
Proposition 4. Ω• is a graded commutative algebra.
CHAPTER 2. SYMPLECTIC MANIFOLDS 10

Lecture 2
Definition 14. A degree k-derivation of a Z-graded algebra A is a linear

map D : A → A such that
D(Aj ) ⊆ Aj+k D(ab) = D(a)b + (−1)kj aD(b),
where j is the degree of a. We write Derk (A).
Remark. Derivations form a graded Lie algebra. If D ∈ Derk (M ) and D0 ∈

Derl (A). Then we define
[D, D0 ] := D ◦ D0 − (−1)kl D0 ◦ D ∈ Derk+l (A).
Definition 15. Lie derivative of a function f ∈ C ∞ (M ) w.r.t. X ∈ X(M ) is
LX (f ) = X(f ).
Definition 16. Lie derivative of a vector field Y ∈ X(M ) w.r.t. X ∈ X(M )

is
d
LX (Y )|p = ((φ−t
X )∗ Y ) = [X, Y ]p .
dt 0
Definition 17. Lie derivative of a differential form α ∈ Λ(M ) w.r.t. X ∈

X(M ) is
d
LX (α)|p = ((φtX )∗ α).
dt 0
Definition 18 (de Rham differential). d : Ωk (M ) → Ωk+1 (M ) for α ∈

Ωk (M ),
k
X
(dα)(X0 , . . . , Xk ) = (−1)i Xi (α(X0 , . . . , X
ci , . . . , Xk ))
i=0
X
(−1)i+j α([Xi , Xj ], X0 , . . . , X
ci , X
cj , . . . , Xk ).
0≤i<j≤k
Definition 19. Interior product: iX : Ωk (M ) → Ωk−1 (M )
iX (α)(Y1 , . . . , Yk−1 ) = α(X, Y1 , . . . , Yk−1 )).

Lecture 3
Proposition 5. LX , d, iX are graded derivations of degree 0, +1, −1 resp,

i.e. they distribute nicely. The following identities hold:
1. [d, d] = 2d2 = 0
2. [iX , iY ] = iX iY + iY iX = 0a
3. [LX , LY ] = LX LY − LY LX = L[X,Y ]
4. [d, iX ] = diX + iX d = LX (Cartan’s magic formula b )

5. [d, LX ] = 0
6. [LX , iY ] = LX iY − iY LX = i[X,Y ]
a α(X, Y, . . .)= −α(Y, X, . . .)
b Useful for calculating LX α
Proof. Exercise.
1. Easy 4. Tedious
2. Easy 5. Easy
3. Fine 6. Easy
https://math.stackexchange.com/questions/2504737/proving-wikipedias-formula-on-interior
https://www.staff.science.uu.nl/~ban00101/lecnotes/lieder.pdf
2.2 Symplectic manifolds

Definition 20. A symplectic manifold is a pair (M, ω) of a smooth manifold
and a symplectic form ω ∈ Ω2 (M ), i.e.
• dω = 0 (closed)
• ωp : Tp M × Tp M → R is non-degenerate (i.e. (Tp M, ωp ) is a symplec-

tic vector space.)
• If degenerate, but rank is constant and ω is closed: pres ymplectic

• If not closed, then we call ω almost symplectic
Lecture 3
wo 26 feb 11:06

Lecture 3
Definition 21. Let f : (M1 , ω1 ) → (M2 , ω2 ) be a smooth map between two

symplectic manifolds. Then f is a symplectomorphism if it’s a diffeomor-
phism and f ∗ (ω2 ) = ω1 .
Example. Let M = R2n with the canonical basis with as coordinates (q1 , . . . , qn , p1 , . . . , pn ).
Then X
ω= dqi ∧ dpi
Then ωp = Ωcan .
Exercise. Show that there is a linear symplectomorphism (Tp R , ωp ) → (R , Ωcan )
2n 2n
Exercise. Let M be a 2-dimensional manifold. Let ω ∈ Ω2 (M ) be a volume

form. Then (M, ω) is a symplectic manifold.
Definition 22. Let Q be a manifold. Consider π : T ∗ Q → Q. The tauto-

logical one-form θ ∈ Ω1 (T ∗ Q) is defined by
θξ∈T ∗ Q (v ∈ Tξ (T ∗ Q)) = θ(v) := hξ, π∗ (v)i := ξ(π∗ (v)).
Proposition 6. The tautological one form θ is the unique one form with
the property: For all one forms α ∈ Ω1 (Q), we can consider it as α : Q →
T ∗ Q : q 7→ αq , (α)∗ (θ) = α.
Proof. • θ satisfies this. h(α∗ θ), vi = hθ, α∗ vi = hαq , π∗ α∗ vi = hαq , Id ◦ vi =

hαq , vi.
• θ is the only thing that satisfies this.
Suppose τ ∈ Ω1 (T ∗ Q) which also satisfies these properties. Let ξ ∈
T ∗ Q, v ∈ Tξ (T ∗ Q) such that w = π∗ v 6= 0. Then there exists an
α ∈ Ω1 (Q) such that α∗ w = v. Then
hτ, vi = hτ, α∗ wi
= hα∗ τ, wi
= hα, wi = hα∗ θ, wi = hθ, α∗ wi = hθ, vi ,
so these things agree on vectors that are not vertical. But these not-
vertical vectors form an open dense subset, so the forms agree.
Exercise. Show: if π∗ v = w for v ∈ T (T ∗ Q), then there exists α ∈ Ω1 (Q) such

that α∗ w = v.
Let U ⊂ Q and let q1 , . . . , qn be coordinates. Then ∂ ∂
∂q1 , . . . , ∂qn form a

Lecture 3
coordinate vector frame ∈ Γ(T Q|U ). Define p1 , . . . , pn ∈ C ∞ (T ∗ Q|U ) by

∂
pi (ξ) = ξ, ∀ξ ∈ Tp∗ Q|U .
∂qi p
Then
(q1 ◦ π, . . . , qn ◦ π, p1 , . . . , pn )
are coordinates for T ∗ Q|U . We often abuse notation and write (q1 , . . . , qn , p1 , . . . , pn ).
Lemma 1. In these coordinates, θ = pi dqi .

P
Proof. Take a point in T ∗ Q, ξ = ai (dqi )x . Then take v ∈ Tξ (T ∗ Q).

P
Write qi = qi ◦ π. Then dqi = π dqi . Note that pi (ξ) = ai .
∗
h(pi dqi )ξ , vi = hpi (ξ)π ∗ (dqi )x , vi

= hπ ∗ ai (dqi )x , vi
= hai (dqi )v , π∗ vi
= hξ, π∗ vi
= hπ ∗ ξ, vi
= hθξ , vi .
Corollary. ωT ∗ Q = −dθ (minus sign is a convention) is a symplectic form

on T ∗ Q.
Proof. Non degenerate because of expression pi dqi , and closed because

P
exact. In local coordinates,
X
−dθ = dqi ∧ dpi .
2.3 Basic properties

Proposition 7. • dim M is even
• M is orientable. ω ∧ · · · ∧ ω is a volume form.a
a Why doesn’t this vanish immediately? Because ω is a 2-form! Example, R4 , ω =
dx1 ∧ dy1 + dx2 ∧ dy2 , then ω ∧ ω = 2dx1 ∧ dy1 ∧ dx2 ∧ dy2 .

Lecture 3
Proposition 8. Let M 2n , ω be a compact manifold without boundary. Then

for all k = 1, . . . , n : [ω k ] ∈ H 2k (M ) is non-zero!. There’s a power of ω in
all even degree homology classes which is not zero!
Example. When you find a manifold with H 2k = 0 for some k, the manifold
cannot have symplectic form. (If M is compact without boundary)
Exercise (Important). Proof this!
• Use stokes to show that ω n is not exact.
• Show that if [ω k ] = 0, then [ω k+1 ] = 0.
Corollary. A sphere S 2n admits a symplectic structure iff n = 1.
Proof. Indeeed, H k (S 2n ) = R[k = 0 or k = 2n]
2.4 Hamiltonian vector fields

Let
ω̃ : T M → T ∗ M : v 7→ iv (ω).
Definition 23 (Hamiltonian vector field). If f ∈ C ∞ (M ). The Hamiltonian

vector field of f is the unique vector field Xf such that iXf ω = df , or in
other words, Xf = ω̃ −1 (df ).
Definition 24 (Symplectic vector field). Y ∈ X(M ) is a symplectic vector

field iff LY ω = 0, i.e. a vector field whose flow is a symplectomorphism.
Remark. If Y is symplectic. 0 = LY ω = d(iY ω) + iY dω = d(iY ω) = 0. I.e.

when we put Y in the symplectic form, we get a closed form. Hamiltonian says
when we put it in the symplectic form, we get an exact form.
Exercise. Show Y ∈ X(M ) is symplectic if iY ω is closed. Show Y ∈ X(M ) is
Hamiltonian if iY ω is exact.
Lemma 2. If Y1 , Y2 ∈ X(M ) symplectic vector fields. Then [Y1 , Y2 ] is sym-

plectic again, but not only that, it is a Hamiltonian vector fields!
[Y1 , Y2 ] = Xω(Y2 ,Y1 ) !

Lecture 3
Proof. i[X,Y ] = LX iY − iY LX . So
i[Y1 ,Y2 ] ω = LY1 iY2 ω − iY2 LY1 ω

= LY1 iY2 ω
= d(iY1 iY2 ω) + iY1 (diY2 ω)
= d(iY1 iY2 ω)
= d(ω(Y2 , Y1 )).
Remark. Lie subalgebra
Proposition 9. Short exact sequence:

f 7→Xf
0 → R → C ∞ (M ) −−−−→ XHamil. (M ) → 0.
Definition 25 (Poisson bracket). The Poisson bracket of f, g ∈ C ∞ (M ) is
{f, g} := ω(Xf , Xg ) = (df )(Xg ) = LXg (f ).
Example. Consider (R2n , ω = dqi ∧ dpi ). Then Xqi = − ∂p ∂

i
, Then Xpi = ∂qi .
∂
In particular,
{pi , pj } = {qi , qj } = 0 {pi , qj } = δij .
X ∂f ∂g ∂f ∂g
{f, g} = − .
∂qi ∂pi ∂pi ∂qi

Exercise. Show that
• [Xf , Xg ] = X{f,g}
• {f, g} = −{g, f }
• {f, {g, h}} + {h, {f, g}} + {g, {h, f }} = 0
• {f, gh} = {f, g}h + g{f, h}.
?
• {f, ·} is a derivation on smooth functions. Note {f, ·} = ±LXf
• {f, g} = 0 if f is constant.
Exercise. Suppose ω1 , ω2 symplectic forms on M . Show that ω1 = ω2 iff Poisson
brackets agree.
Example. Consider R2n with any coordinates x1 , . . . , x2n . Suppose we have two
Poisson brackets of ω1 , ω2 . Show that the symplectic forms agree if the Poisson
brackets agree on the coordinate functions! Hint: use Taylors theorem.

Lecture 3
2.5 Hamiltonian Mechanics

Definition 26. A Hamiltonian triple (M, ω, H) is a symplectic manifold
(M, ω) with H ∈ C ∞ (M ) that we call the Hamiltonian function.
Definition 27. An f ∈ C ∞ (M ) is a conserved quantity if
LXH f = {f, H} = 0.
Example (Point particle). Consider point particle of mass m in R3 with potential

V : R3 → R. q0 ∈ R3 is the starting position, v0 ∈ Tq0 R3 is it starting velocity.
We want to find q(t), q : R → R3 . Newton: mq̈ = F = −∇V . q(0) = q0 ,
q̇(0) = v0 . Consider its momentum: mq̇(t), which is in the tangent bundle (but
we can connect cotangent bundel and tangent bundle)
q̇ = p/m, q(0) = q0
ṗ = −∇V, p(0) = mv0 .
Then
1 2
H(q, p) = p + V (q).
2m
The previous equations can then we written as
q̇i = {qi , H} ṗi = {pi , H},
so (q, p) is an integral curve of XH .

In the case of a harmonic oscillator, we have V (q) = 21 kq 2
Figure 2.1: Example of the pendulum

Lecture 4

Remark. Why are momenta covectors and not vectors? TODO. https://math.
stackexchange.com/questions/1324814/why-is-the-momentum-a-covector
Lecture 4
wo 04 mrt 11:04
Example. Let (Q, g) be a Riemannian manifold. Consider k · k2 : T Q → R :
v 7→ g(v, v). Let g [ : T Q → T ∗ Q : v 7→ g(v, ·). Let H : T ∗ Q → R, H = kg ] k =
k(g [ )−1 k. Then (T ∗ Q, Ωcan. , H) is Hamiltonian triple. Integral curves of this
system are geodescics.
Definition 28 (Symmetry). A symmetry of (M, ω, H) is a Hamiltonian vec-

tor field Xf such that
LXf H = 0 ⇔ {f, H} = 0.
Theorem 1 (Noether’s theorem). Let Xf be a symmetry of (M, ω, H). Then

f is conserved by integral curves of XH . In other words: integral curves of
XH lie inside level sets of f .
Proof. If LXf H = 0, then {f, H} = 0, or LXH f = 0.
2.6 Almost Kähler Manifolds

Definition 29 (Almost complex structure). An almost complex structure on
M is a vector bundle morphism J : T M → T M such that J 2 = −IdT M .
Definition 30 (Almost Kähler (AK) manifold). An almost Kähler manifold

is a triple (M, ω, J) such that (M, ω) is a symplectic manifold and J is an
almost complex structure such that ω and J are compatible, i.e.
g(v, w) := ω(v, Jw)
is positive definite and symmetric.
Remark. Every symplectic manifold admits a compatible almost complex struc-

ture.
Proposition 10. Suppose (M, ω, J) is AK and N ,→ M is a submanifold

such that J(T N ) = T N , then N is a symplectic submanifold.

Lecture 4
Proof. Restriction is automatically closed (pullback of closed is closed). So

we only need to show that ωT N is non degenerate. Suppose v ∈ Tx N such
that
ω(v, w) = 0 ∀w ∈ Tx N.
In particular, ω(V, Jw) = 0 for all w, as J(T N ) ⊂ T N . This implies that
ω(v, Jv) = 0. This implies that v = 0, because this is g(v, v).
Definition 31 (Complex structure). An almost complex structure is called

complex if the Nijenhuis tensor vanishes. Let X, Y ∈ X(M )
X(M ) 3 NJ (X, Y ) := [JX, JY ] − [X, Y ] − J ([JX, Y ] − [X, JY ]) .
Exercise. Show that NJ ∈ Γ Λ2 T ∗ M ⊗ T M . In other words, NJ (X, Y )p de-

pends only on Xp , Yp !
Hint: show that it is C ∞ (M ) linear. (Hint: look at the proof that the torsion
tensor is a tensor!)
Theorem 2 (Newlander-Nirenberg). Let J be an almost complex structure

on M . Then NJ = 0 if and only if there exists an atlas covering the entire
manifold φi : Ui ⊂ Cn → M such that the transition maps are holomorphic
and when J is restricted to Ui , it’s the standard complex structure on Cn :
J(z1 , . . . , zn ) = (iz1 , . . . , izn ).
(This is not multiplication by a scalar, when we view it as R2n .)
Proof. A lot of analysis.

Chapter 3
Submanifolds and Normal

Forms
3.1 Isotopes, homotopy operators and tubular neigh-

bourhoods
Isotopes
Fix a manifold M , I = [0, 1].
Definition 32. An isotopy is a smooth family of diffeomorphisms {ρt : M →

M }t∈I , i.e. I × M → M is smooth.a
a Note I × M is a manifold with boundary!
Definition 33. A time dependent vector field is a smooth family of vector

fields, {X t }t∈I , i.e. I × M → T M is smooth.
There is a correspondence between time dependent vector fields an isotopies.
Given ρt an isotopy. Then we can define
d
(X s )p := ρt (q) q := (ρs )−1 (p).
dt t=s
20
Lecture 4
γ p = (ρs )(q)
Figure 3.1: Isotopy
Remark. The correspondence
{isotopies} → {t.d. vector fields}
is injective. Not surjective: not all vector fields have a flow (only complete
vector fields.)
Exercise. Show this is surjective is M is compact. (Any vector field on a compact
manifold has a flow?)
Exercise (Bit tricky). Suppose ρt is an isotopy and ρt+s = ρt ◦ ρs for all 0 ≤
t, s, t + s ≤ 1. Then the corresponding X t is time independent.
Proposition 11. What if both forms are time dependent and we also have
an isotopy. Let αt be a smooth family of k-forms and ρt is an isotopy.a

d d
((ρt )∗ αt ) = ρ∗t LXt αt + αt .
dt dt
The left hand side: in a point, it is the usual derivative in a vector space.
a Actually this also holds when we just have a family of smooth functions that are
not necessarily diffeomorphisms not necessarily isomorphisms.
Proof. Idea of proof. Local coordinates: work, but lot of work. Better idea:

d d d
(ρ∗t αt ) = ρ∗t αt + ρ∗t αt .
dt t=t0 dt t=t0 dt t=t0
Enough to show this for αt time independent.
• Take the case in local coordinates when α = dxi . Prove this.

• Show if it holds for α and β then it holds for α ∧ β. Use that the Lie
derivative is a derivation w.r.t. the wedge product.
• Show that if it holds for α, then it holds for f α where f ∈ C ∞ (M ).
CHAPTER 3. SUBMANIFOLDS AND NORMAL FORMS 21

Lecture 5
ρt (S 1 )
t t ∂
∂t
Xt
(ρt (p), t)
M = S1
p X0
Figure 3.2: both time dependent formula
https://math.stackexchange.com/questions/2608048/time-derivative-of-a-pullback-of-a-time
Homotopy operator
Definition 34 (Homotopy operator). Let fi : M → N for i = 0, 1 be smooth.

A homotopy operator between f0 and f1 is a linear map Q : Ωk (N ) →
Ωk−1 (M ) such that
f1∗ − f0∗ = d ◦ Q + Q ◦ d : Ω• (N ) → Ω• (M ).
Exercise. Suppose H : I × M → N is a smooth function st H|i×M = fi for

i = 0, 1. Consider It : M → I × M : M → {t} × M . Show that
Z 1
Q := It∗ ◦ i ∂ ◦ H ∗ dt
∂t
0
is a homotopy operator from f0 to f1 . Idea: pull back to the big manifold,

contract with ∂t
∂
, and than integrate along all level sets. Hint: use the formula
from Proposition 11.
https://math.stackexchange.com/questions/2643279/homotopy-invariance-of-de-rham-cohomology

Lecture 5
Lecture 5
wo 11 mrt 11:10
Aim of the upcoming sections:
• Mozer theorem
• Darboux theorem
• Weinstein Lagrangian neighbourhood theorem.
Proposition 12 (Tubular neighborhood theorem). Let M be a smooth man-

ifold. Let X ⊂ M be a compact, embedded submanifold. For p ∈ X, there
exists a neighbourhood
T M |X
Op ⊂ V ⊂ N X := of the zero section.
TX
such that for all p ∈ X, V ∩ Np X is convex (‘ball around zero section of all
fiber) and there exists a diffeomorphism φ : V → φ(V ) such that
φ
V M
i
0
X
commutes
Proof. This proof is not examinable. We use Riemannian geometry for

this.
Definition 35. Riemannian metric g on M is a fiberwise metric for

TM.
Proposition 13. All manifolds admit a Riemannian metric.
Proposition 14. A Riemannian metric induces a metric on M which

makes (M, d) a metric space.
Z 1 p
dg (p, q) = inf g(γ 0 (t), γ 0 (t)) dt.
γ(0)=p,γ(1)=q 0
The metric space is not always complete.

Lecture 5
Definition 36. The energy functional on paths
1 1
Z
E(γ) = g(γ 0 (t), γ 0 (t)) dt.
2 0
Geodesics minimize this energy functional and locally minimize arc

length.
Theorem 3. Le p ∈ M , then there exists a neighbourhood U and ε > 0

such that for all q ∈ U and for v ∈ Tq M that are short (kvkg < ε), there
is a unique geodesic γv which goes through the point q and γv0 (0) = v.
Definition 37. For any p ∈ M the exponential map at p is
expp : V ⊂ Tp M → M : v 7→ γv (1).
Definition 38. Let E → M be a vector bundle with metric g. Write
Eε = {e ∈ E | kek < ε}.

Now, on to the proof! Choose a Riemannian metric g. Define
T M |X
E = {v ∈ Tp M | p ∈ X ⊂ M, v ∈ (Tp X)⊥ } ∼
= NX = .
TX
Define
Uε = {q ∈ M | dg(q, X) < ε}.
If ε is small enough, then we claim that Uε and Eε are diffeomorphic.
Indeed,
Proposition 15. For small enough ε, exp is a diffeomorphism.
Proof. V ⊂ E for set of non-criticala points, Clearly 0-section ⊂ V .

Write V1 = V ∩ E1 . This is compact and contains 0-section.
Lemma 3. If A is a compact metric space, X0 ⊂ A closed and

f : A → B a local homeomorphism s.t. f |X0 is bijective, then f
is bijective on ta whole neighborhood of X0 .
a Critical point: where de differential is degenerate
Using this lemma, the exponential map is bijective on a whole

neighbourhood.

Lecture 5
What we now want to show is that the image of E is actually Uε . By

definition, it’s clear that exp( Eε ) ⊂ Uε .
Converse. Let q ∈ Uε , p ∈ X closest to q. Pick γ a geodesic of length
< ε such that γ(0) = p, γ(1) = q. So
g(γ 0 (0), Tp X) = 0 ⇒ expp (γ 0 (0)) = q.
Remark. Normal bundle of a submanifold:

T M |X
NX = .
TX
Remark. Must be compact. Otherwise, consider (x, |x|
1
) ⊂ R2 . Then ε-tubes do
not exist. We need compactness for a bound on ε.
M
X
Figure 3.3: Tubular Neighborhood
Proposition 16. Let U ⊂ M a tubular neighborhood π : U → X. Then
• i∗ : H • (U ) → H • (U ) is an isomorphism.
• If β ∈ Ωk (U ), dβ = 0 and i∗ β = 0 (so 0 on X) Then there exists
ξ ∈ Ωk−1 (U ) such that
dξ = β ξ|X = 0.
Proof. • π : U → X . Homotopy equivalent π ◦ i = idX , i ◦ π ' idU .

H : I × U → U, (t, v) 7→ tv. (We use identification as vector bundle)

Lecture 5
Then H0 = i ◦ π and H1 = idU .

• H : I × U → U induces a chain homotopy operator
Q : Ω• (U ) → Ω•−1 (U )
from Id∗U and (i ◦ π)∗ . Assume β ∈ Ωk? (U ) such that i∗ β = 0.
((ι ◦ π)∗ − Id∗U )(β) = dQβ − Q

dβ.

Now
(π ∗ ◦ i∗ )β = 0,
so we get
β = dQ(β),
so we’ve shown already that β is exact. Now,
Z 1
Q(β) : = (It∗ ◦ i∂t ◦ H ∗ )(β) dt
0
Z 1
= ft∗ (iV β)dt,
0
where V := H∗ (∂t ), ft = H|{t}×i(X)

H|I×i(X) does not depend on t. So
Q(β)|X = 0 V |i(X) = 0.
Remark. Restriction of a form: β|X is not the same as pulling back with the
embedding. When pulling back, we also restrict tangent vectors to be along the
submanifold.
3.2 Darboux’ theorem

Theorem 4 (Moser). Let M be compact, {ωt } smooth family of symplectic
forms such that [ωt ] ∈ H 2 (M ) is constant. Then there exists an isotopy ρt
such that
ρ∗t ωt = ω0 ,
i.e. constant in particular
Proof. Recall, on compact manifolds,
{isotopies} ↔ {time-dependent vfs}.

Lecture 6
d ∗ d
ρt ωt = ρ∗t (LVt ωt + ωt ),
dt dt
where ρt is flow of Vt . We need to find Vt such that
d
LVt ωt + ωt = 0,
dt
as ρ0 = Id. Note that ωt − ωt0 is always exact by assumption. Fact (we
don’t prove this): ∃ηt ∈ Ω1 (M ) such that dηt = dt
d
ωt . a
Take Vt the unique time dependent vector field such that iVt ωt +ηt = 0.b
This implies that, using diX + iX d = LX and dω = 0,
d
0 = d(iVt ωt + ηt ) = LVt ωt + ωt .
dt
Then the flow of Vt , ρt satisfies ρ∗t ωt = ω0 .
a So we claim that when we take the limit t0 → t, the difference is still exact.
b Why does this exists? ω is non degenerate, we can invert it. So it is ωt−1 (−ηt ).
Lecture 6
wo 18 mrt 10:03
Theorem 5 (Local normal form). Let M be a manifold, i : X → M an
embedded compact submanifold. Let ω0 , ω1 be symplectic forms such that
ω0 |X = ω1 |X .
Then there exists tubular neighborhoods U0 , U1 of X and a diffeomorphism

relating these tubular neighborhoods, f : U0 → U such that f |X = IdX
and
f ∗ ω1 = ω0 .
Proof. Let U ⊂ X be a tubular neighbourhood. Recall that X ,→ U

induces an isomorphism on cohomology. And there exits an η ∈ Ω1 (U )
such that dη = (ω1 − ω0 )|U and η|X = 0. Set ωt = ω0 + t(ω1 − ω0 ). Note
that ωt |X = ω0 |X , because the two forms agree on X. We can shrink U
sufficiently such that this form is always non degenerate for all t. Indeed,
ω0 and ω1 are non degenerate, and they agree on X, so in particular ωt |X
is non degenerate for all t, and non degeneracy is an open condition.
d d
ωt = ω0 + t(ω1 − ω0 )
dt dt
= ω1 − ω0
= dη.

Lecture 6
Let Vt be the unique time dependent such that
iVt ωt + η = 0.
Because η|X = 0, Vt |X = 0. Now, consider the flow of Vt , let’s call it

ρt . As the manifold is compact, there is a nice tubular neighbourhood U0
where this flow exists. Take U1 := ρ1 (U0 ) ⊂ M , which is also a tubular
neighborhood of X, as Vt |X = 0. Then we claim the diffeomorphism is
exactly
f := ρ1 |U0 .
Prove this claim as an exercise, i.e. prove
f ∗ ω1 = ω0 .
Remark. Why local normal form. This show that if two symplectic manifolds
agree on a submanifold, then they will always agree in a neighbourhood around
that submanifold!
Theorem 6 (Darboux). Let (M 2n , ω) a symplectic manifold. For all p ∈ M ,

there exists an open neighbourhood U (p) ⊂ M and coordinates
(q1 , . . . , qn , p1 , . . . , pn ) ∈ C ∞ (U ).
such that ω|U = dqi ∧ dpi .

P
Remark. This means that every invariant in symplectic geometry must be global.
In Riemannian geometry, such thing, we cannot do! We can put the metric in
standard form at one single point, but not in a neighbourhood around a point!
The curvature prevents this.
Proof. As seen in one of the previous lectures, there exists a linear sym-
plectomorphism
Lp : (Tp M, ωp ) → (R2n , Ωcan ),
for all p. Let x1 , . . . , xn , y1 , . . . , yn ∈ C ∞ (U ) local coordinates around p.
such that under Lp ,
∂ ∂
= ei , = fi , Ω(ei , fj ) = δij .
∂xi p ∂y i p
This is just in one specific point p. Now we want to extend this to a full
neighbourhood. Write φ : R2n ⊃ V → U ⊂ M for a coordinate chart

Lecture 6
corresponding to (xi , y i ). So
φ∗ ω|p = Ωcan .
Now, we apply the local normal form theorem! to {0} ⊂ V , which is a

zero dimensional submanifold. So we have a symplectomorphism
f : (V0 , Ωcan ) → (V1 , φ∗ ω),
V0 , V1 ⊂ R2n open neighbourhoods around 0 ∈ V . Let U = (φ ◦ f )(V0 ).

And then (φ ◦ f )−1 : M → R2n forms coordinates.
3.3 Lagrangian submanifolds
Definition 39. A submanifold X ⊂ (M 2n , ω) is Lagrangian if Tp X ⊂ Tp M

is a Lagrangian subspace for ωp for all p ∈ M . Or equivalently, X is
n-dimensional and i∗ ω = 0.
Exercise. If (M 2 , ω)is a two dimensional symplectic manifold, then any one-

dimensional submanifold is Lagrangian.
Proposition 17. Let Q be a smooth manifold. Consider (T ∗ Q, ωcan = −dθ).

Take a α ∈ Ω1 (Q). Then graph α is Lagrangian submanifold iff dα = 0
Proof. Let α : Q ,→ T ∗ Q. Recall the tautological one form, θ ∈ Ω1 (T ∗ Q)

uniquely characterised by α∗ θ = α. Therefore
dα∗ θ = dα
α∗ dθ = dα
∗
−α ωcan = dα,
so this vanishes pricely when dα vanishes.
Example. Consider (R2n , Ωcan ). You can view R2n = ∼ T ∗ Rn . Given a function
f ∈ R , then
n

∂f ∂f
df : Rn → R2n : (q1 , . . . , qn ) 7→ q1 , . . . , qn , (q), . . . , n (q) .
∂q1 ∂q
Then d(df ) = 0, so the graph of df would be Lagrangian.

Exercise. If L ⊂ (R × R , Ωcan ) such that π1 : R
n n 2n
→ R is diffeo on L, then
n
L is as in the previous example.

Lecture 6
Definition 40. Let X ⊂ Q. Then the canonical bundle of X is

G
(T X)0 = (Tx X)0 := {ξ ∈ Tx∗ Q | x ∈ X and ξ|Tx X = 0} ⊂vec. subbundle T ∗ Q.
x∈X
This is the annihilator at each point.

(T X)0 ⊂ T ∗ Q is a submanifold of half the dimension of Q, no matter
the dimension of Q.
Proposition 18. (T X)0 is a Lagrangian submanifold with respect to ωcan =

−dθ.
Proof. Exercise.
Corollary. We can take the submanifold X = {q}. Then TX = 0, so the

annihilator is Tq∗ Q is a Lagrangian submanifold. Every fiber of the vector
bundle is itself a sub Lagrangian manifold of the vector bundle T ∗ Q, which
is not compact.
The following proposition shows that every symplectomorphism between two
symplectic manifolds can be represented as a Lagragian inside a bigger symplec-
tic manifold.
Proposition 19. Let (M1 , ω1 ) and (M2 , ω2 ). Construct (M × N, π1∗ ω1 −

π2∗ ω2 ), where π1 , π2 are the projection on M, N resp.
Given a symplectomorphism f : M → N , then
graph f := {(m, f (m) ∈ M × M )} ⊂ M × N
is Lagragian with respect to π1∗ ω1 − π2∗ ω2 .
Exercise. Show that this form is nondegenerate.

“Everything is a Lagragian!” — A. Weinstein
Proof. Note dim graph f = 1

2 dim(M × N ).
dim(M × N ) = dim M + dim N = 2 dim M since M ∼

= N.
And dim M = dim graph. We need to show that graph f is isotropic iff
f ∗ ω2 = ω1 (symplectomorphism). Set Ω = π1∗ ω1 − π2∗ ω2 . Let f : M →
M × N : m 7→ (m, f (m)). This f is characterised by two properties:
• π1 ◦ f = IdM

Lecture 6
• π2 ◦ f = f
So
∗ ∗ ∗
f ∗ ω2 = f π2∗ ω2 = f (π1∗ ω1 − Ω) = ω1 − f Ω.
∗
This equation gives the result. Indeed the graph being isotropic if f Ω = 0,
which will be precicely the case when f ∗ ω2 = ω1
Around any Lagrangian submanifold, the symplectic form can be brought
to a familiar standard form.
Theorem 7 (Weinstein). Let (M, ω) be a symplectic manifold. Let L ⊂ M

be a Lagrangian compact manifold. Then there exists a tubular neighbour-
hood U ⊂ M of L, and V ⊂ T ∗ L of the zero section. such that they are
symplectomorphic, i.e.
f : (U, ω|U ) → (V, ωcan |V ).
Proof. We’re doing the Riemannian geometry proof here, you can find other
proofs elsewhere. Choose a compactible almost structure J : T M → T M
for ω and write g = ω(·, J·).a Then we can a find a neighborhood U ⊂
(T L)⊥ ⊂ T M such that φ : exp |(T L)⊥ : U → M is an embedding.
Note, we can also embed U in T ∗ L as follows:
F : U ,→ T ∗ L : v 7→ ω(v, ·).
Since T L are maximal isotropic subspaces, if we put in vectors which are

orthogonal to L, then for each one of these vectors, we get exactly one 1-
form. So this is an isomorphism of vector bundles from the normal bundle
to T ∗ L. This is smooth and also an embedding of the tubular set U . Now,
for all p ∈ L:
T0p U ∼
= Tp L ⊕ Tp L⊥ induced by g.
Now, (dφ)0p : Tp L⊕(Tp L)⊥ → Tp M is an canonical identification for p ∈ L.
(d exp0 = Id )
Note that (Tp L)⊥ = J(Tp L), because g(v, w) = ω(v, Jw). If g(v, w) = 0
for all v ∈ Tp L iff ω(v, Jw) = 0, so Jω ∈ Tp L (Tp L maximal isotropic
subspace w.r.t ω, so if something pairs to zeros with all elements of Tp L,
then it is itself in Tp L?) So w ∈ J(Tp L) as J 2 = −Id.
So we find that Tp L and J(Tp L) are Lagrangian subspaces of Tp M w.r.t.
ω.
Thus:
(φ∗ ω)(v ⊕ w, v ⊕ w0 ) = ω(v, w0 ) + ω(w, v 0 ),
where the other terms vanish because they are in Lagrangian subspaces.

Lecture 7
On the other hand, we have
T0p (Tp∗ L) = Tp L ⊕ Tp∗ L,
as every tangent space to a vector bundle splits up in the tangent space to

the base and the fiber.
ωcan (v ⊕ α, v 0 ⊕ α0 ) = α(v 0 ) − α0 (v).
Remember that F is a bundle map, so it is linear in each fiber, so its

derivative is the map itself in each fiber:
(dF )0p : Tp L ⊕ (Tp L)⊥ −→ Tp L ⊕ Tp∗ L

v ⊕ w 7−→ v ⊕ ω(w, ·).
So this map is compatible with the splitting on the origin. We use this fact
to compute the pullback:
F ∗ ωcan (v ⊕ w, v 0 ⊕ w0 ) = ωcan (v ⊕ ω(w, ·), v 0 ⊕ ω(w0 , ·))

= ω(w, v 0 ) − ω(w0 , v)
= ω(v, w0 ) + ω(w, v 0 ).
Comparing this with φ∗ ω, we get F ∗ ωcan = φ∗ ω. So we found two neigh-

bourhoods U, V and a symplectomorphism.
a Note that in the proof of the tubular neighbourhood theorem, we pick a metric at
random, but here we specifically want this metric.
Remark. This seems lucky! Crucial that we choose the correct Riemannian
metric.
Lecture 7
za 28 mrt 14:51
Recall: if two symplectic forms agree on a submanifold, we can make them
agree in a neighbourhood with a symplectomorphism. Corollary of this is the
Darboux Theorem.
Today: local normal form stuff!
Definition 41 (Presymplectic manifold, form). A presymplectic form on M

is ω ∈ Ω2 (M ) s.t.
• dω = 0
• Kernel distribution ker ωp = {X ∈ Tp M | iX ω = 0} has the same
dimension for all p ∈ M
We call (M, ω) a presymplectic manifold.

Lecture 7
A symplectic form is a presymplectic form where the kerel distribution has

dimension 0.
Lemma 4. Let (M, ω) be a presymplectic manifold. Then the set ker ω ⊂

T M is an involutive subbundle of T M .a
a Smooth because ω is smooth and subbundle because dimension is the same.
Proof. Remember, involutive means all combinations of sections have van-

ishing Lie bracket. We need to show: X, Y ∈ Γ(ker ω) ⇒ [X, Y ] ∈ Γ(ker ω)
i[X,Y ] ω = [LX , iY ]ω = LX iY ω − iY LX ω = −iyLX ω,
where we used iY ω = 0, and now writing the derivative with Cartan’s magic
formula, and using dω = 0, we get
−iY diX ω = 0.
Theorem 8 (Frobenius). A subbundle V ⊂ T M is involutive iff it is the

tangent bundle of a regular foliation F on M .
Definition 42 (Regular foliation). A regular foliation is a partition F

of M
into smooth manifolds of a constant (usually lower) dimension: M = i Li
where Li is connected, regularly immersed submanifold. The Li are called
the leaves of the foliation.
Figure 3.4: Regular Foliation
Foliations hard, subbundles good!

Remark. This means the presymplectic manifold (M, ω) are equipped with a
foliation. G
M= Li with Tp Li = ker ωp .

Lecture 7
Remark. Note that M/ ∼ where p ∼ q if they are in the same leaf, then this is
not necessary a smooth manifold!
Exercise. Suppose ∼ is an equivalence relation on smooth manifold M . Show
that there exists at most one smooth structure on M/ ∼ with makes M → M/ ∼
a smooth submersion.
Proposition 20 (Presymplectic reduction). Let (M, ω) be a presymplectic

manifold with equivalence ∼ from the kernel foliation. Suppose N := M/ ∼
is smooth such that π : M → N is a submersion. Then there exists a
symplectic structure ωred on N such that π ∗ ωred = ω.
Proof. π : M → N is a submersion. Therefore, because of surjectivity, we

can define define
ωpred (π∗ X, π∗ Y ) = ωp (X, Y ) ∀X, Y ∈ Tp M, p ∈ M,
which is of course saying that ω = π ∗ ωred . So if ωred exists, it is like this.

Remaining to show: is this well-defined?
It is, because ker π∗ = ker ω. So when x, x0 ∈ Tp M such that they are
identified under π ∗ , so x − x0 ∈ ker π∗ = ker ω. Then
ωred (π∗ (x), π∗ (w)) = ωred (π∗ (x − x + x0 ), π∗ (w)) = ωred (π∗ (x0 ), π∗ (w)).
Now we’ve showed that this is well defined and it fully determines ωred .
Now we show that ωred is a symplectic form.
Closed π ∗ : Ω0 (N ) → Ω0 (M ) is injective (because π is a surjective sub-
mersion) So
0 = dω = d(π ∗ ωred ) ⇒ dωred = 0.
Non degeneracy. SUppose Y ∈ Tp N such that ωred (X, Y ) = 0 for all

X ∈ Tp N . Then
ω(X̃, Ỹ ) = 0 ∀X̃ ∈ Tp̃ M,
where π(p̃) = p and π∗ Ỹ = Y . This means that Ỹ is in the kernel

distribution, which then means that π∗ (Ỹ ) = 0. This proves the
claim!
Next up: Interesting examples of presymplectic manifolds!

Lecture 7
Definition 43. Let (M, ω) symplectic manifolds. A submanifold X ⊂ M

is called coisotropic if Tp X ⊂ (Tp M, ωp ) is a coisotropic subspace for all
p ∈ M , i.e.
(Tp X)ω ⊂ Tp X.
Recall also that this implies that dim X ≥ 1
2 dim M . It contains it’s own
symplectic complement.
Example. Lagrangian are maximal coisotropic subspaces. But on these, ω pulls

back to zero, so they don’t have a symplectic structure any more. Higher di-
mensional submanifold come with an intrinsic structure!
Example. All submanifolds of codimension 1 are coisotropic.
Proof. Tp X coisotropic iff Ann(Tp X) ⊂ Tp∗ M is isotropic w.r.t ω −1 .

If codimension of X is one, then annihilator has dimension 1, and any
one dimensional subspace w.r.t an alternating form is isotropic.a
a Indeed, span X ⊂ (span X)Ω = {v | ω(v, x) = 0} as ω(x, x) = 0.
Proposition 21 (Coisotropic reduction). Let i : X → (M, ω) be a coisotropic

submanifold. Then
• (X, i∗ ω) is a presymplectic manifold
• If the foliation given by i∗ ω is such that X/ ∼ is smooth, then there

exists a unique symplectic form ωred on X/ ∼ such that π ∗ ωred = i∗ ω,
where π : X → X/ ∼.
Proof. (a) is an exercise, part (b) follows from the previous proposition.
Example. Consider (R4 , dq1 ∧dp1 +dq2 ∧dp2 ). Then N := {p2D = 0}

E is coisotropic.
The presymplectic form on N is dq1 ∧ dp1 , and the kernel is ∂
∂q2 . The reduced
space is (R , dq1 ∧ dp1 ).
2

Theorem 9 (Gotay, 1982). Let (N, ω) be a presymplectic manifold. Then:
• There exists (M̃ , ω̃) and embedding (N, ω) ,→ (M̃ , ω̃) such that i∗ ω̃ =
ω and N is embedded as a coisotropic submanifold.
• Given two such embeddings i1 , i2 there exists neighbourhoods U1 , U2
around the embedded N ’s a symplectomorphism φ such that i1 =
i2 ◦ φ.
So every presymplecic manifold embeds coisotropically into a symplectic
manifold. So the examples we’ve discussed before are the only examples in

Lecture 7
some sense! This is like a neighbourhood theorem for coisotropics! But a little
weaker than Weinstein for examples, as that one gives a standard form. We
don’t yet have a standard form
Proof. This is the rough idea.
Existence Write E = ker ω ⊂ T N . Trick of the proof: we want to equip

the dual bundle to E, Tot E ∗ with a symplectic form ω̃ such that the
embedding of the zero section O : N ,→ E ∗ is the coisotropic embed-
ding, and O∗ ω̃ = ω. This would give us a non-compact manifold, but
that’s okay. Recall:
T0p E ∗ ∼
= Tp N ⊕ Ep∗ .
Furthermore,
0 → E → T N → T N/E → 0
is a short exact sequence of vector bundles. One can prove that this
implies we can split T N , i.e. we can find a map T N/E → T N , and
G ⊂ T N such that T N = E ⊕ G. (Not canonical!!) So we get
T0p E ∗ ∼
= Ep ⊕ Gp ⊕ Ep∗ .
This can be equipped with a symplectic structure as follows. First,

Ep ⊕ Ep∗ caries a canonical linear symplectic form:
ωE (a ⊕ α, b ⊕ β) = β(a) − α(b).
Write π : E ∗ → N . Now, set ΩG = π ∗ ω + ωE , clearly smooth and

non-degenerate.a
So in conclusion, (T E ∗ |N , ΩG ) is a symplectic vector bundle.b
Now we use the extension theorem (another result of Weinstein)
Lemma 5. Let P be a manifold and N ⊂ P a closeda subman-

ifold. Suppose (T P |N , Ω) is symplectic vector bundle such that
Ω|T N (a differential form on N ) is closed. Then Ω extends to a
symplectic form on a tubular neighbourhood of N in P
a According to splitting π ∗ ω is non-degenerate form on G : π ∗ ω(a ⊕ g ⊕ α, b ⊕ g 0 ⊕ β) =
ω(π∗ (a ⊕ g ⊕ α), π∗ (b ⊕ g 0 ⊕ β)) = ω(a ⊕ g, b ⊕ g 0 ) = ω(g, g 0 )

b Vector bundle of symplectic vector spaces in the sense of symplectic linear algebra
a compact without boundary
Now P = E ∗ and Ωg and we can apply this theorem because when

pulling it back, it is closed (presymplectic). Since ΩG |N = π ∗ ω + ωE ,
it is obvious that N is coisotropic with respect to it.

Lecture 7
Local (i.e. around N ) uniqueness Assume that (N, ω) ,→ (M, Ω) is

coisotropic embedding. Consider the symplectic vector bundle (T M |N , Ω|N ).
Using the tubular neighbourhood embedding for N , we can split
T M |N ∼
= TN ⊕ E∗
|{z}
∼
= E ⊕ G ⊕ E∗,
∼
=N N viaΩ|N
where N N is the normal bundle of N and the further splitting is done

by choosing a complement. Now, we can characterize Ω|N by three
properties:
1. (Ω|N )|G⊕E = ω
2. (Ω|N )|E⊕E ∗ = ωE (because we identified N N with E ∗ using Ω)
3. GΩ|N = E ⊕ E ∗
So we find that the symplectic vector bundle (E ⊕G⊕E ∗ , Ω) depends
only on ω and the choice of G, and not on the ambient manifold.
So if we have two coisotropic embeddings ii : (N, ω) ,→ (M̃i , ω̃i )
= (E ⊕ G ⊕ E ∗ , Ω) ∼
we get linear symplectomorphism (T M1 ω̃1 |N ) ∼ =
(T M2 , ω̃2 |N ). Because in both cases, the symplectic form will be
characterized by these properties.
Now, we need to apply a more general Local normal form theorem
Lemma 6. If X ,→ M compact embedded submanifold and M

is equiped with ω0 , ω1 symplectic forms such that
(T M |X , (ω0 )|X ) ∼
= (T M |X , (ω1 )|X )
as symplectic vector bundles, then there exists neighborhoods of

∼
=
X and f : U0 −→ U1 such that f ∗ ω1 = ω0 .
This is the local normal form theorem, but here, the symplectic forms
don’t have to agree, they just have to induce isomorphic symplectic
vector bundles!
This proves the theorem.

Chapter 4
Review of Lie groups
Lecture 8: Review of Lie groups, Lie-algebra cohomology,

the adjoint representation
wo 01 apr 17:57
4.0 Introduction
Definition 44. A Lie group is a smooth manifold G which is equipped with
a smooth function m : G × G → G : (g, h) 7→ g · h such that (G, m) is a
group.
Definition 45. A Lie group homomorphism F : G → H is a smooth function

which is also a group homomorphism.
Remark. The inverse map g 7→ g −1 is automatically smooth.

Example. Given V a vector space. Take G = GL(V ). More generally G ,→
GL(V ) is a submanifold closed under matrix multiplication.
• n × n real matrices with determinant 1, SL(n)

• n × n real matrices which preserve the standard Euclidian metric O(n)
• n × n complex matrices which preserve the standard Hermitian metric
U (n)
• 2n × 2n real matrices which preserve the standard symplectic form Sp(n)
38
Lecture 8: Review of Lie groups, Lie-algebra cohomology, the adjoint
representation
Definition 46 (Left invariant vector field). Suppose G is a Lie group. Let

Lg : G → G : h 7→ g ·h be the left translation map. A vector field X ∈ X(G)
is called left invariant if for all h ∈ G,
(dh Lg )Xh = Xg·h .
Remark. • XL (G) ≤ X((G)) is a Lie subalgebra (i.e. Commutator of two

left invariant vector fields is again a left invariant vector field.)
• Given X ∈ XL (G), then Xg = (de Lg )Xe .
• Consequence: XL (G) ∼
= Te G Therefore g := Te G is a Lie algebra.
Example. Consider GL(Rn ) and gl(Rn ). GL(Rn ) ⊂ gl(Rn ) open1 , where gl(Rn )
are all matrices, not generally invertible.
Because this is an open subset of a vector space, we can decompose its
tangent space as
T GL(Rn ) ∼= gl(Rn ) × GL(Rn ).
Therefore,
TI GL(n, R) ∼
= gl(Rn ).
Given B ∈ gl(n, R). Then left invariant vector field M → T M is
GL(n, R) → gl(n, R) × GL(n, R) : A 7→ (A · B, A).

Exercise. Covering space of a Lie group is also a Lie group. Suppose G is a Lie
group and G̃ is the covering space of G. Fix an element p ∈ G̃ which sits over the
identity of G. Show that G̃ has a unique Lie group structure for which p is the
identity and which makes the covering map G̃ → G a Lie group homomorphism.
(Remember that covering space can be tough of as homotopy classes of
paths, and what happens when you multiply two paths? How does the notion
of homotopy of a path interact with the group structure.)
Exercise. Suppose B ∈ gl(Rn ) is a matrix and X : GL(Rn ) → T GL(Rn ) is the
corresponding left-invariant vector field. Show that the flow φtX of X satisfies
∀A ∈ GL(n, R) : φtX (A) = AetB .
Exercise. Recall that if X and Y are vector fields on a manifold, then the Lie
bracket [X, Y ] can be defined as follows:
d √ √ √ √
−
[X, Y ]p := φX t
◦ φ−
Y
t
◦ φXt ◦ φY t (p).
dt 0
Suppose X and Y are left-invariant vector fields on GL(Rn ) which satisfy

XI = A and YI = B. Use the formula above to show that [X, Y ] = AB − BA.
Remark: Lie groups → Lie algebras is functorial.
1 The property of being invertible comes from determinant which is smooth
CHAPTER 4. REVIEW OF LIE GROUPS 39

representation
Proposition 22. Suppose Φ : G → H is a homomorphism of Lie groups.

Then de Φ : Te G → Te H is a homomorphism of Lie algebras.
This means that every G ≤ GL(Rn ) has a corresponding g ≤ gl(Rn ).
Remark. Lie’s second theorem: If G is simply connected, then Φ 7→ de Φ is
surjective, so this is kinda the reverse of the previous result.
Remark. Lie’s third theorem: every lie algebra is isomorphic to the lie algebra
of some lie group. In other words: G 7→ Te G is surjective up to isomorphism.
4.1 Lie algebra cohology

The deRham cohomology comes from
∞
M
Ω(M ) := Γ(Λk T ∗ M ).
k=0
The Lie algebra cohomology comes from

∞
M
∧g∗ = ∧k g∗ .
k=0
Recall: ∧ g is the set of all multilinear antisymetric maps gk → R.

k ∗
Definition 47. The Chevally-Eilenberg differential is the degree 1 map:
dg : ∧g∗ → ∧g∗
defined by the following formula for all η ∈ ∧k g∗

X
dg (η)(v1 , . . . , vk+1 ) := (−1)i+j η([vi , vj ], v1 , . . . , v̂i , v̂j , . . . , vk+1 ),
1≤i≤j≤k+1
Remark. The deRham differential looks kinda similar:
X X
ddR (α)(. . .) = (−1)i LVi α(. . . V̂i ) + (−1)i+j+1 α([Vi , Vj ], . . . , V̂i , V̂j ).
i<j
Exercise. Let g be a Lie algebra and d the CE differential. This forms a complex!
Steps:
• Suppose η ∈ g∗ . Show that d2 η = 0. (Follows from Jacobi identity)
• Show that d is a graded derivation: η ∈ ∧k g∗ , ζ ∈ ∧l g∗ , then
d(η ∧ ζ) = dη ∧ ζ + (−1)k η ∧ dζ.
(Hint: Since both sides of the equations are linear maps, if suffices to show
this for basis vectors. And then induction)

representation
• Use the previous parts to show that d2 = 0. (Hint: First apply this to
some basis vectors)
Definition 48. Given a Lie algebra g, the Chevally-Eilenberg cohomology

of g is the sequence of groups {HCE
n
(g)}n∈N
n {η ∈ ∧n g∗ | dη = 0} ker d
HCE (g) := = .
{dη | η ∈ ∧n−1 g∗ } im d
Exercise. Take g = R2 with x, y unit vectors and [x, y] = x.

Let u, v dual basis to x, y. Then
du(x, y) = −u([x, y]) = −u(x) = −1 dv(x, y) = −v(x) = 0.
The kernel of d : g∗ → ∧2 g∗ is one-dimensional, it is spanned by v. (u is not

in the kernel) As a consequence HCE
1
(g) ∼
= R. On the other hand, the kernel of
d : ∧ g → ∧ g is everything, because ∧3 g∗ is trivial. The kernel is everything,
2 ∗ 3 ∗
but we also mod out by everything, since that was the image, so HCE 2
(g) = 0.
4.2 Adjoint representation

Definition 49 (representation). Suppose G is a Lie group and V is a vector
space. A representation on V is a Lie group homomorphism R : G →
GL(V ).
Remark. A Lie group representation are a special kind of group actions. Take
A : G × V → V : (g, v) 7→ R(g)v.
Definition 50. Let g be a Lie algebra and V a vector space. A representa-

tion of g on V is a lie algebra homomorphism:
ρ : g → gl(V ) := {linear endomorphisms of V }.
Remark. If R : G → GL(V ) is a representation then
ρ := de R : Te G → gl(V ) is a representation.
If you are given a Lie algebra representation, ρ, then we can look at forms
on the Lie algebra which take values in the vector space:
∧k g∗ ⊗ V := {antisymmetric maps gk → V }.

representation
Definition 51. The Chevally-Eilenberg differential with values in a repre-

sentation ρ : g → gl(V ) is defined as follows. For η ∈ ∧k g∗ ⊗ V :
k+1
X X
(dρ η)(. . .) := (−1)i ρ(vi )(η(. . . , v̂i ))+ (−1)i+j η([vi , vj ], . . . , v̂i , v̂j ).
i=1 1≤i≤j≤k+1
This is similar to the deRham differential, but we’ve replaced the Lie derivative
with the action by representation of vi .
One notable representation is the following:
Definition 52 (Adjoint representation). Let G be a Lie group and g := Te G

its associated Lie algebra. Then:
• G acts on itself by
Cg (h) := ghg −1 .
This conjugation map is a Lie group homomorphism (whereas left
translation is not.)
• The Lie group homomorphism Ad : G → GL(g) defined by
Ad(g)v := (de Cg )v
is called the adjoint representation. So this is a map from the Lie

group to linear transformations of the tangent space at identity.
• The lie algebra homomorphism ad : g → gl(g) defined by
!
ad(v)w := ((de Ad)v)w = [v, w]
is called the adjoint representation of g. This is just the derivative of

Ad.
Exercise. Let g be a Lie algebra. Using the formula ad(v)w = [v, w]. Show that
ad : g 7→ gl(g)
is a Lie algebra homomorphism.

You should need to use the Jacobi identity.
Exercise. Let G = GL(Rn ) and g = gl(Rn ). Verify that
(de Ad)(M )N = [M, N ] = M N − N M.
Exercise. Suppose R : G → GL(V ) and ρ : g → gl(V ) are representations of

a Lie group and Lie algebra on V . Show that the following formulas define
representations on V ∗ .

Lecture 9: Lie group actions, Moment maps
∀g ∈ G, η ∈ V ∗ , v ∈ V R∗ (g)(η)(v) = η(R(g −1 )v)

∗
∀z ∈ g, η ∈ V , v ∈ V ρ∗ (g)(η)(v) = −η(ρ(z)v).
Definition 53. Given a representation of a Lie group or Lie algebra on a

vector space V . The representation obtained in the previous exercise on V ∗
are called the dual representations. The dual representation to the adjoint
representation is called the coadjoint representation.

di 21 apr 10:31

Chapter 5
Lie groups
5.1 Lie group actions

Definition 54. Let G be a Lie group and M a manifold. An action of G on
M is a smooth function:
G×M →M (g, m) 7→ g · m.
such that ∀g, h ∈ G and m ∈ M we have that g · (h · m) = (gh) · m.
Remark. Equivalently, it is a ‘smooth homomorpism’ G → Diff(M ) := diffeo-

morphisms of M . But Diff(M ) is k∞-dimensional, which makes it hard to make
this definition fully rigorous. This can be thought of as a non-linear represen-
tation (compare with → GL of a vector space).
Remark. Given g ∈ G, we usually write φg : M → M to denote the diffeomor-
phism associated to g.
Definition 55. The orbit of p ∈ M is G · p ⊂ M .
Definition 56. The isotropy group at p is the set
Gp = {g ∈ G | g · p = p} ≤ G.
Definition 57. Let g be a Lie algebra. A infinitesimal action of g on a

manifold M is a Lie algebra homomorpism:
ρ : g → X(M ).
Remark. You can think of X(M ) as the differentiation of Diff(M ).
44
Proposition 23. Suppose Φ : G × M → M is a Lie group action. Consider

the function
d
ρ : g → X(M ) ρ(v)p := Φ(exp(−tv), p).
dt t=0
defines an infinitesimal action of g. Here exp(tv) denotes the flow of the

left invariant vector field associated to v.
The function ρ is called the infinitesimal generator of the action.
Remark. Why the minus sign? This is an artifact of some of the conventions
we chose (use of left-invariant vector fields).
Proof. First note that T (G × M ) ∼

= T G × T M . Using the chain rule,
ρ(v)p = d(e,p) Φ(−v, 0).
Therefore ρ is linear (this is just applying a linear function to −v) To show

that it is a Lie algebra homomorphism, we claim:
ρ(v) = (dΦ)∗ (X−v , 0) (Xv right invariant vf).
Right invariant vector fields actually lie algebra anti homomorphism, which
is why we add a minus sign in fount of v. Because submersion (or some-
thing like that), (dΦ)∗ preserves Lie bracket. These two facts show that
(dΦ)∗ (X−v , 0) is a Lie algebra homomorphism.
To see this, first notice that for all g, h, p
Φ(h, Φ(g, p)) = Φ(Rg (h), p) ⇒ de,g·p Φ(w, 0) = d(g,p) Φ(de Rg (w), 0),
where the left hand side is the associativity, and the rhs is the derivative
w.r.t. h and p, leaving g fixed. Here w is a tangent vector at the identity,
and Rg is right multiplication with g.
Note that the rhs is de definition of a right invariant vector field. In-
serting w = −v, we get
ρ(v)g·p = d(e,g·p) Φ(−v, 0) = d(g,p) Φ(de Rg (−v), 0),
which is to say
ρ(v) = (dΦ)∗ (X−v , 0).
Exercise. Suppose G has an action on M . Show that for all p ∈ M the set
G · p ⊂ M is an immersed submanifold of M , so its differential is injective.
Exercise. Suppose G has an action on M and ρ : g → X(M ) is the associated
CHAPTER 5. LIE GROUPS 45

Lie algebra action. Show that for all p ∈ M
Tp (G · p) = {ρ(v)p | v ∈ g}.
So not only immersion, but its tangent spaces arises as the image of the Lie
algebra under the infinitesimal action.
Example. Consider that action of (R, +) on C by rotation.
t · z := eit z.
The Lie algebra is just R. If we use the coordinates decomposition z = x + iy,

the infinitesimal action takes the form

∂ ∂
ρ(v) = v y −x v ∈ R.
∂x ∂y
Note that ρ(v) is clockwise rotation, while eit z is counter clockwise rotation!
This is a consequence of the minus sign in the definition of infinitesimal action.

5.2 Moment maps

Exercise. Suppose G is a Lie group and G acts on (M, ω) by symplectomor-
phisms. Let ρ : g → X(M ) be the infinitesimal generator. Show that for
all v ∈ g, the vector field ρ(v) is a symplectic vector field. In other words,
Lρ(v) ω = ω.
If symplectic actions give rise to symplectic infinitesimal generators, can we
make sense of a notion of Hamiltonian infinitesimal generators? This is the idea
behind moment maps.
Suppose for all v ∈ g, there exists f ∈ C ∞ (M ) such that ρ(v) = Xf . Fix a
basis: v1 , . . . , vk and functions f1 , . . . , fk such that ρ(vi ) = Xfi .
Then we can define a linear function
µ : g → C ∞ (M ) : ci vi 7→ ci fi .
There is a natural bijection between the following sets:
{smooth functions J : M → g∗ } ↔ {linear functions µ : g → C ∞ (M )}.
µ(v)(p) = hJ(p), vi v ∈ g, p ∈ M.
Therefore, we get a smooth function J : M → g∗ .
Remark. If you have an infinitesimal action with this property, all the informa-
tion is in J, i.e. you can completely recover ρ.

Definition 58 (Moment mamp, Hamiltonian action). Suppose G acts on M

by symplectomorphisms with infinitesimal action ρ : g → X(M ). A func-
tion J : M → g∗ is called a moment map if
• ρ(v) = −Xµ(v) for all v ∈ g, where µ(v) ∈ C ∞ (M ) : p 7→ hJ(p), vi
smooth.
• J(g·p) = Ad∗g J(p) for all g ∈ G. This property is called G equivariant
w.r.t the coadjoint action of G on g∗ .
An action is said to be Hamiltonian if there exists a moment map.
So the first part says that the action is by Hamiltonian vector fields, but it
turns out the second property is also very useful to have.
Exercise. Recall that the Poisson bracket is a natural Lie bracket on the smooth
functions of a symplectic manifold:
{·, ·} : C ∞ (M ) × C ∞ (M ) → C ∞ (M ) : {f, g} 7→ Xf (g).
Suppose G acts on a symplectic manifold: (M, ω) and J : M → g∗ is a momen-

tum map for this action. Show that the function µ : g → C ∞ (M ) defined above
is a Lie algebra homomorphism.
This exercise should give you a hint as for why there is a minus sign in the
definition of moment map. Going from a function to a hamiltonian vector field
is a lie algebra anti homomorphism for our conventions. So it swaps the Lie
bracket when you go from functions to vector fields. So if we want both µ and
ρ to be a Lie algebra homomorphism, then, we’re required to put a minus sign
in ρ(v) = −Xµ(v)
Corollary. Suppose Gacts on a symplectic manifodl (M, ω) and J : M → g∗

is a moment map for this action. Then we get the following commuting
diagram of Lie algebras:
C ∞ (M )
µ f 7→−Xf
ρ
g X(M )
If G is connected, the converse holds!

So when you have a moment map, a lot of the structure of the symplectic
manifold gets reflected in the Lie algebras.
Note that this commuting diagram is like an infinitesimal version of condition
two in the Definition of moment map. This commuting diagram is equivalent
to condition 2 iff G is connected.

Lecture 10
Example. Take G = (R, +) and M = T ∗ R3 with
v · (q, p) = (q + v, p),
so the action is translating the position. Then the infinitesimal generator be-
comes (mind the sign swap):
∂
ρ(v1 , v2 , v3 ) = −v i .
∂q i
Recall that ∂
∂q i = Xpi . Then we can define:
µ : R3 → C ∞ (R3 ) : µ(v1 , v2 , v3 ) = v i pi .
Alternatively,
J : R3 × R3 → R3 : J(q, p) = p.
So the moment map is the projection to the momentum coordinates!
This example shows why the moment map is called the moment map. In
similar rotational context, the moment map is the angular momentum!
Lecture 10
di 28 apr 15:52
5.3 Examples of moment maps

Example. Consider (R, +) acting on M . Then φt : M → M is an isotopy. By
differentiation such a thing you get a vector field V ∈ X(M ). If there exists an
f ∈ C ∞ (M ) such that V = −Xf , then f : M → R = R∗ is actually a moment
map.
Proposition 24. Suppose (M, ω) is asymplectic manifold where ω = dα.

Let G act on M and ρ : g → X(M ) be the associated infinitesimal action.
If we assume that the action of G preserves α, then G acts by symplecto-
morphisms and
J : M → g∗ : p 7→ v 7→ (iρ(v) (α))(p) = αp (ρ(v)|p )
is a moment map for this action.
Proof. Denote by J : M → g∗ be the moment map and by µ : g → C ∞ (M )

the dual map. To show that J is a moment map, we need:
ρ(v) = −Xµ(v) J(g · p) = Ad∗g J(p).
• Definition of hamiltonian vector field says something about contract-

Lecture 10
ing it with ω. So let’s do that with ρ(v)!
i−ρ(v) ω = −iρ(v) dα = diρ(v) α − Lρ(v) α = d(iρ(v) α) = dµ(v),
were we used ω = dα, Cartans magic formula. For the last step,
remember that the action preserved the one form, so Lρ(v) α = 0. So
we have that i−ρ(v) ω = dµ(v). This is the definition of hamiltonian
vector field.
• Now equivariance.
J(p · v)(v) = (iρ(v) α)(g · p)

Same thing as pullback diffeomorphism associated to v
= (φ∗g (iρ(v) α))(p)
Now, we have
φ∗ iρ α = φ∗ α(ρ, ·, ·) = α(ρ, dφ(·), dφ(·)) = α(dφ(dφ−1 ρ), dφ(·), dφ(·)) = idφ−1 ρ φ∗ α,
so this becomes
= idφ−1 φ∗ α(p)
g ρ(v) g
The action preserves α, so

= idφ−1
g ρ(v)
α(p)
We claim that dφg ρ(v) = ρ(Adg v), so
= iρ(Adg−1 v) α(p)
Definition of moment map
= J(p), Adg−1 v
Definition of coadjoint action gives
= Ad∗g J(p)v .
Example. Suppose G acts on a manifold Q. Consider the lifted action of G on

T ∗ Q:
g · η := φ∗g−1 η q ∈ Q, η ∈ Tq∗ Q.
Note that the base point of η is q and that of g · η is g · q. Show that this action
preserves the tautological 1-form.

Lecture 10
Corollary. Suppose Q is a smooth manifold and G is a Lie group action on

Q. Then the associated action of G on T ∗ Q is Hamiltonian.
Example. Take R3 with cylindrical coordinates r, θ, z. The 2-form is symplectic

on the unit sphere: dθ ∧ dz.1 Take action of S 1 on S 2 :
t · (r, θ, z) = (r, θ + t, z).
The Lie algebra of S 1 is R and the infinitesimal action is

∂
ρ : R → X(R3 ) : v 7→ −v
∂θ
and
J : S 2 → R : (θ, z) 7→ z
is a moment map for this action.2
Proof. Note that
J : S 2 → (g → R) : (θ, z) 7→ (a 7→ az)
µ : g → (S 2 → R) : a 7→ ((θ, z) 7→ az).
We check iρ(v) ω = −d(µ(v)). Indeed:

∂
iρ(v) ω = (dθ ∧ dz)(ρ(v), ·) = (dθ ∧ dz) −v , · = −vdz(·),
∂θ
and
−d(µ(v)) = −d((θ, z) 7→ vz) = −vdz.
So when you contract ρ(v) with ω, we get dz, which is exact, so we find
that ρ(v) is Hamiltonian.
Here we see that the moment map is kinda like the angular momentum?
We think of angular momentum being orthogonal to the plane of rotation,
and here we project on the z-component.
Important example:
1 manifold is two dimensional, so automatically closed.
2 Equivariance for free because Lie algebra is abelian.

Lecture 10
Proposition 25 (Kirillov-Kostant-Souriau, KKS symplectic form). Let g be a

Lie algebra and let O ⊂ g∗ be an orbit of the conjoint action of the Lie
group. Let
ρ : g → X(g∗ )
denote the infinitesimal generator of this action. Then O has a canonical
symplectic structure ω given by the formula:
ωη (ρ(v)η , ρ(w)η ) := hη, [v, w]i η ∈ g∗ , v, w ∈ g.
Remark. How should think of this? You want a two form, which is something
that when you give it two vectors, it gives you a vector. So a two form, evaluated
at some point in O ⊂ g∗ , you want to come up with a number given two vectors
tangent to the orbit. But a vector tangent to the orbit is exactly a vector that lies
in the image of the infinitesimal generator. So the ingredients we’re provided
with are two elements of the lie algbra and a point of the orbit, which is a
covector. The most natural thing you can do is hη, [v, w]i. The remarkable thing
with this is this is actually a symplectic form! Astounding! Deep connection
between symplectic geometry and Lie theory.
Given any Lie algebra, the coadjoint orbits are symplectic manifolds.
Exercise. Let G be a Lie group and suppose that g is its Lie algebra. Let
ρ : g → X(g) be the infinitesimal generator of the coadjoint action of G on g∗ .
Show that ρ satisfies the following equation:
hρ(v)η , wi = hη, [v, w]i η ∈ g∗ , v, w, ∈ g.
To makes sens of the above formula, we have utilized the identification Tη g∗ ∼

= g∗
Proof. Note that for all points η ∈ O and a vector ṽ ∈ Tη O, there exists
an element of you lie algebra, v ∈ g such that ρ(v)η = ṽ. So we can always
treat vectors as coming from the Lie algebras.
• First we need to show non degeneracy. Using the exercise, if iρ(v) ω =

0, then:
∀w ∈ g : 0 = ωη (ρ(v)η , ρ(w)η ) = hη, [v, w]i = hρ(v)η , wi .
But if 0 = hρ(v)η , wi is true for all w, then ρ(v)η is zero. So this

tangent vector (we think of it as an arbitrary tangent vector to the
orbit) is zero. This proves nondegenerate: iρ(v) ω = 0 ⇒ ρ(v) = 0.
• Closed? Use definition of differential of a two form Let η ∈ g∗ and
v, w, z, ∈ g. Then
X
(dω)η (ρη (v), ρη (w), ρη (z)) = Lρ(v) ω(ρ(w), ρ(z))−ωη ([ρ(v), ρ(w)]η , ρ(z)η ),
c.p. v, w, z

Lecture 10
where c.p. stands for cyclic permutations.

Two terms:
– Let g(t) = exp(−tv) : (−ε, ε) → G Note that we’re simply taking
the Lie derivative of a function. We can take the derivative by
plugging in curve whose tangent vector is ρ(v).
d
Lρ(v) ω(ρ(w), ρ(z))(η) = ω(ρ(w), ρ(z))(g(t) · η)
dt 0
Definition of KKS symplectic form
d
= hg(t) · η, [w, z]i
dt
0
d
= g(t) · η, [w, z]
dt 0
= hρ(v)η , [w, z]i
By previous exercise
= hη, [v, [w, z]]i .
Because the sum is over cyclic permutation, these terms become

zero because of the Jacobi identity.
– On the other hand, using the fact that ρ is a Lie algebra homo-
morphism and the definition of the KKS symplectic form, you
get
ωη ([ρ(v), ρ(w)], ρ(z)) = hη, [[v, w], z]i .
Therefore, after cyclicly permutin, we get that the sum of terms
is also zero!
Therefore dω = 0.
Most simple non-trivial example:

Exercise. Let G = SO(3) and g = so(3). Equip so(3) with the following metric:
* 0 x z
 
0 a c +

−x 0 y  , −a 0 b  = xa + yb + zc.
−z −y 0 −c −b 0
Equip so(3)∗ with the dual metric to the one defined above. Show that
the coadjoint orbits of G are the unit spheres relative to this metric. Finally
compute the KKS symplectic form in the x, y, z coordinates. You should get
the standard symplectic form on the unit sphere in R3 .
Why is this in the section about moment maps?

Lecture 10
Proposition 26. Consider the coadjoint action of a Lie group G on the dual
of its Lie algebra g∗ . Let O be an orbit of this action and ω the KKS form
on O. Then the inclusion O ,→ g∗ is a moment map for the transitive
action of G on (O, ω).
This is kind of amazing!
Proof. There’s no need to check equivariance condition, because the map

is the inclusion. So it’s obviously compatible with the coadjoint action. We
only need to show that ρ(v) = −Xµ(v) , or −iρ(v) ω = dµ(v), where
µ(v)(η) := hη, vi .
To prove that −iρ(v) ω = dµ(v), we contract both sides by a tangent vector

ρ(w) at some point η ∈ O. Left hand side:
def.
iρ(w) iρ(v) ωη = ωh (ρ(v), ρ(w)) = hη, [v, w]i .
Right hand side:
−iρ(w) (dµ(v))(η) = −Lρ(w) µ(v)(η) = − hρ(w)η , vi = hη, [v, w]i ,
where we used that this contraction of a smooth function is the same as

the lie derivative of a smooth function, which is the same thing as the
infinitesimal generator applied to v.a
This concludes the proof.
a This step is similar as the one in the previous proof
5.4 Existence and uniqueness of moment maps

Example. Let M = T 2 = S 2 × S 2 . dθ ∧ dψ symplectic form. Note that θ, ψ are
not coordinates as they take values in S 1 and not in R. The group S 1 acts on
M via rotation:
t · (θ, ψ) := (θ + t, ψ).
The infinitesimal generator is ρ(v) = −(v)∂θ . Since −∂θ is not Hamiltonian,
there is no moment map. Indeed, when we contract it with the symplectic
form, we get −dψ, which is not exact.
Suppose the infinitesimal generator is always a Hamiltonian vector field. Is
it always possible to find a moment map?

Lecture 10
Proposition 27. Suppose G is a compact Lie group and acts on (M, ω) by

symplectomorphisms. Let ρ be the infinitesimal generator of this action and
suppose that for all v ∈ g there exists f ∈ C ∞ (M ) such that ρ(v) = −Xf .
Then there exists a moment map for this action!
Proof. Sketch of a proof. By choosing a basis of g one can construct
µ̃ : g → C ∞ (M ) such that ρ(v) = −Xµ̃(v) ,
by just transporting coefficients. But µ is probably not G equivariant. To

fix this, we make use of an ‘averaging’ principle that exists for compact Lie
groups:
Lemma 7. Suppose G is a compact Lie group and G has an action on

a manifold M . Then G naturally acts on C ∞ (M ) by the push-forward
operation:
g ∈ G, f ∈ C ∞ (M ), p ∈ M (g · f )(p) := f (g −1 p).
For all smooth functions f ∈ C ∞ (M ), there exists another smooth

function favg ∈ C ∞ (M ) which satisfies the following properties:
• favg is invariant under the action of G. Equivalently favg is
constant on the orbits of G
• favg lies in the closed convex hull of G · f . (This says that it’s
sorta like a convex combination of all the functions you can get
by g · f for all g)
By thinking of µ̃ as µ̃ : g × M → R. Then action adjoint in first
component, and as above in second. Then µ = µ̃avg is the desired moment
map.
• ρ(v) = −Xµ(v) : convex combinations of µ’s will satisfy this equation,
and as µ̃avg lies in the convex hull, this is the case.
• The fact that this mu is invariant under this action corresponds pre-
cisely to the fact that J is G-equivariant.
How much choice do we have in picking a moment map?

Lecture 10
Proposition 28. Suppose G is a Lie group acting on (M, ω) and M is con-

nected. Suppose J1 and J2 : M → g∗ are moment maps for this action.
Then
• There exists a c ∈ [g, g]◦ such that J1 = J2 + c
• If H 1 (g) = 0, then J1 = J2 .
Here, [g, g] := {[v, v] | v ∈ g} and A◦ is the annihilator of A, which are all
the covectors anihilating this set.
Proof. Note that by definition of Lie algebra cohomology:
0 = H 1 (g) := {η ∈ g∗ | dg η = 0 ⇔ ∀v, w ∈ g : hη, [v, w]i = 0} = [g, g]◦ .
Note that because this is H 1 , there are no exact things. So going back to
the definition, being closed corresponds to annihilating commutators. So
the first property implies the second property.
Proof of the first property.
Firs, we know that for all v ∈ g, we have Xµ1 (v) = Xµ2 (v) . This implies
that µ1 (v) = µ2 (v) + c(v) for c : g → R ⊂ C ∞ (M ), where ⊂ is i : R →
C ∞ (M ) : c 7→ (p 7→ c). This is because the kernel of the map that sends
something to its Hamiltonian vector field is price cloy the locally constant
functions. Because the manifold is connected, these are globally constant
functions.
To finish, we need to show that c annihilates commutators, i.e.
∀v, w ∈ g c([v, w]) = 0.
(Why?) Calculating this, we get

def
c([v, w]) = µ1 ([v, w]) − µ2 ([v, w])
As µ are moment maps, they are homomorphisms of Lie algebras,
= {µ1 (v), µ1 (w)} − {µ2 (v), µ2 (w)}
= {µ2 (v) + c(v), µ2 (w) + c(w)} − {µ2 (v), µ2 (w)}
Now, poisson bracket with constant is zero, so
= {µ2 (v), µ2 (w)} − {µ2 (v), µ2 (w)}
= 0.
∀f ∈ C ∞ (M ) {f, 1} = {f, 1 · 1} = 1 · {f, 1} + 1 · {f, 1} = 2{f, 1} = 0.
Now, we know how many degrees of freedom we have.

Lecture 10
Proposition 29. Suppose G acts by symplectomorphisms on (M, ω) and

H 1 (g) = 0 and H 2 (g) = 0. We assume that G and M are connected. Then
there exists a unique moment map.
This is bit different from our previous existence result. Then, we assumed
that the infinitesimal generator took values in Hamiltonian vector fields. But
here, we don’t have that assumption.
Proof. We only need to show existence. Uniqueness is by the previous

proposition.
Since the bracked of symplectic vector fields is Hamiltonian:
ρ([v, w]) = [ρ(v), ρ(w)] is Hamiltonian: vector field..
Since H 1 (g) = [g, g]◦ = 0, we know that [g, g] = g. This means that
everything is the bracket of something. This means that ρ(u) = ρ([v, w])
for some v, w, which is a Hamiltonian vector field.
∀v ∈ g : ∃f ∈ C ∞ (M ) : ρ(v) = −Xf .
So we can choose a linear function
µ : g → C ∞ (M ) such that ρ(v) = −Xµ(v) .
But µ is probably not G equivariant. Trick! Note that we cannot use

our previous avg trick, because group might not be compact. We need to
modify µ. Consider the following smooth function:
c(v, w) := {µ(v), µ(w)} − µ([v, w]).
This function measures the failure of µ to be a Lie algebra homomorphism.

Remember, when G is connected, the equivariance condition is the same as
µ being a Lie algebra homomorphism.
So in principle, this is just a smooth function, but it turns out this is
actually always a cont function. We can show this by showing that c lies
in the kernel of the hamiltonian map, i.e.
ρ = −Xµ ⇒ Xc(v,w) = 0,
which tells you that c(v, w) is a locally constant function on M . Because M

is connected, this is constant. So c is an anti-symmetric function g×g → R,
or equivalently c ∈ ∧2 g∗ .
Exercise. Show that dg c = 0. As a hint: You wil need to use the fact that
constant functions are in the kernel of the Poisson bracket.

Lecture 10
Since dg c = 0 and H 2 (g) = 0, there exists ` ∈ g∗ such that dg ` = c. So

for all v, w ∈ g we have
`([v, w]) = c(v, w).
Take µ(v) = µ(v) + `(v). Then:
ρ(v) = −Xµ = −Xµ µ Lie algebra morphism.
Why?
• This new µ still satisfies the first moment map condition. Because
`(v) should be thought of as spitting out constant functions, so taking
the Hamiltonian of µ(v) is the same as taking that of µ(v) + `(v).
• Why Lie algebra morphism? Exactly because of the definition of `.
Check this yourself!

Chapter 6
Symplectic reduction
Lecture 11: Symplectic reduction theorem, examples

wo 06 mei 10:23
6.1 Symplectic reduction theorem
Given a symplectic manifold M and a Lie group G which acts on it. Then
quotient M/G does not necessarily inherit a symplectic structure. It may not
even be even dimensional!
Example. Consider M = R4 and G = (R, +) acting by translation
t · (x, y, z, w) 7→ (z + t, y, z, w).
Then M/G = R3 which is not even even dimensional. And this action is even
Hamiltonian!
The goal of symplectic reduction is to use moment maps to construct new
symplectic manifolds out of group actions!
Definition 59. The isotropy Lie algebra at p ∈ M is defined to be
∀p ∈ M gp := {v ∈ g : ρ(v)p = 0} = ker ρp .
This Lie algebra is the Lie algebra of the isotropy group:
Gp := {g ∈ G : g · p = p}.
58
Lemma 8 (6.1.1). Suppose G acts on (M, ω) and J : M → g∗ is a moment

map for this action. For p ∈ M consider the differential
dp J : Tp M → TJp g∗ ∼
= g∗ .
Then
• ker dp J = (Tp (G · p))ω , so ω(ker dp J, Tp (G · p)) = 0.
• Im dp J = (gp )◦ , where g◦p = {η ∈ g∗ : ρ(v)p = 0 ⇒ hη, vi = 0}
Remark. The first bullet point implies that the tangent space to the fibers of J
are symplectic orthogonal with respect to the orbits of G.
Proof. To prove this lemma, we first claim that:
∀x ∈ Tp M, v ∈ g : ωp (ρ(v)p , x) = hdp Jx, vi .
(“dp Jx” = push the vector down to the Lie algebra) To see why the formula
is true, we refer to the following calculation:
ω(ρ(v), x) = ix iρ(v) ω
= ix d µ(v)
= Lx µ(v)
d
= µ(v)(p(t)) p0 (0) = x
dt t=0
d
= hJ(p(t)), vi
dt t=0
= hdp Jx, vi ,
in the last step we used that hJ(p(t)), ·i is a linear map and the derivative
of such a thing is its differential (?).
• Proof of (1) Given x ∈ ker dp J then for all y ∈ Tp (G · p) we have:
∃v ∈ g : ω(x, y) = ω(x, ρ(v)p ) = hdp Jx, vi = h0, vi = 0.
As being tangent to an orbit is by assumption lying in the image of

ρ. This shows that ker dp J ⊂ Tp (G · p)ω . Now assume x ∈ Tp (G · p)ω :
∀v ∈ g : 0 = ω(x, ρ(v)) = hdp Jx, vi ⇒ dp Jx = 0.
As by assumption, being orthogonal to the orbit means that for all

v ∈ g, ω(x, ρ(v)) = 0.
CHAPTER 6. SYMPLECTIC REDUCTION 59

Exercise. Complete the proof of the previous Lemma, which should be pretty
straightforward.
Definition 60. Recall that a regular value of a smooth function f : M → N

is a point n ∈ N such that for all m ∈ M we have that
dm f : Tm M → Tn N
is surjective. When n is a regular value, we know automatically that f −1 (n)

is smooth.
Lemma 9. • η ∈ g∗ is a regular value of J if and only if:
∀p ∈ J −1 (η) ⊂ M gp = 0 (i.e. Gp is discrete).
So isotropic Lie algebra at all inverse images is 0.

• In this case, J −1 (η) is smooth, and there is a well defined Lie group
action of Gη on J −1 (η). Here Gη is the isotropy group at point η and
here we’re thinking of the isotropy group of the coadjoint action on
g∗ , because η ∈ g∗ .
• Finally, ker i∗ ω = T (Gη · p). Here, i : J −1 (η) ,→ M is the inclusion.
So the kernel of the restriction of ω to the fiber J −1 (η) is actually
just the tangent space to the orbits of this action.
So the form you get is not necessarily symplectic any more, but its
kernel is actually precisely the orbits of that action.
Proof. • η is a regular value iff dp J is surjective for all p ∈ J −1 (η).

From the previous lemma we know that Im dp J = g◦p . Therefore,
Im dp J is surjective iff g◦p = g∗ , which is true if and only if gp = 0.
Now as gp is the Lie algebra of the isotropy group Gp , this implies
that this group is discrete, because only discrete groups have trivial
Lie algebras.
• The action of Gη on J −1 (η) is well defined since:
∀g ∈ Gη , ∀p ∈ J −1 (η) : J(g · p) = g · J(p) = J(p).
Regular value, so J −1 (η) is smooth. Follows from equivariance of

moment map: J(g·p) = g·J(P ) If you have element g which stabilizes
η, then J(p) = η and η is preserved by action of g, so g · J(p) = J(p),
so the result still lies in the fiber.

• We use Lemma 6.1.1:
ker i∗ ω = ker dJ ∩ (ker dJ)ω = ker dJ ∩ T (G · p) = T (Gη · p).
Here we used that the kernel of a symplectic form can be thought of

as the intersection of that subspace with it’s symplectic complement.
In the last step: we have directions that are tangent to the orbit, but
also tangent to the fibers of J, so we get the elements of G which stay
inside the fibers of J. Which is precisely the stabiliser of J(p) = η.
Although M/G may not be a symplectic manifold, the symplectic reduction

theorem says that J −1 (η)/Gη inherits the structure of a symplectic manifold.
Theorem 10 (Marsden-Weinstein 1975). Let G be a Lie group acting on a

symplectic manifold (M, ω) with a moment map J : M → g∗ . Suppose
η ∈ g∗ is regular value of J and N := J −1 (η)/Gη is a smooth manifold.
Then there exists a unique symplectic form ωred on N such that π ∗ ωred =
ω|J −1 (η) where π : J −1 (η) → N is the quotient map.
i π
(M, ω) ←−− (J −1 (η), i∗ ω = π ∗ ωred ) −−→ (J −1 (η)/Gη , ωred ).
Proof. J −1 (η) is a pre-symplectic manifold and the proof is basically the

same as the pre-symplectic reduction theorem.
∀v, w ∈ Tp J −1 (η) ωred (dp πv, dp πw) := ω(v, w).
• For this definition to make sense, it needs to be independent of v and

w. Suppose dp πv1 = dp πv2 . Then v1 = v2 + u for some u ∈ ker dp π.
ω(v1 , w) = ω(v2 + u, w) = ω(v2 , w) + ω(u, w) = ω(v2 , w),
because tangent to orbit en kernel of π are orthogonal.

One little problem: the p’s in the definition could be different p’s that
projects to the same point. For this you need that the value of π is
invariant of which element from the orbit you pick.
• For non-degeneracy, suppose we have v ∈ Tp J −1 (η) such that for all
w ∈ Tp J −1 (η), ωred (dp πv, dp πw) = 0. Then:
∀w ∈ Tp J −1 (η) ω(v, w) = 0 ⇔
v ∈ (Tp J −1 (η))ω = (ker dp J)ω = Tp (G · p) ⇒ v ∈ ker dp π.

So dp πv = 0.

• Closed? Submersion and you pull back closed two form, you get
a closed two form, because pullback operation is compatible with
differential.
If the action of a compact Lie group is free, i.e. ∀p ∈ M Gp = {e}, then

M/G is smooth! This fact we assume without proof. Therefore, we have the
following corollary:
Corollary. Suppose G is a compact Lie group which acts on a symplectic

manifold (M, ω). Suppose further that the action of G = G0 on J −1 (0) is
free. Then J −1 (0)/G is smooth and there is a unique symplectic form ωred
such that π ∗ ωred = ω|J −1 (0) .
Proof. The only thing to check is that 0 is a regular value of J. Since the
action on J −1 (0) is free, it follows that gp = 0. Therefore, dp J is surjective
for all p ∈ J −1 (0).
6.2 Examples of symplectic reduction

Example. Suppose our symplectic manifold is Cn+1 . Consider the action of the
unit circle S 1 ⊂ C :
ζ · (z1 , . . . , zn+1 ) := (ζz1 , . . . , ζzn+1 ).
If we take polar coordinates zi = ri eiθi on Cn+1 , the symplectic form becomes

n+1
X
ri dri ∧ dθi .
i=1
This is a non-degenerate form, even though it looks like it vanishes at 0. This

is an artefact of the coordinates being singular at the origin. The infinitesimal
generator ρ : R → X(Cn+1 ) of the action is
X ∂
ρ(t) = −t .
i
∂θi
Now, i ∂θ is Hamiltonian. Why? Because when we contract the symplectic

P ∂
i
form by this vector field, we get ri dri = d( 21 ri2 ) Therefore, we have a
P P
moment map:
n+1
1X 2
J : Cn+1 → R : (r1 , θ1 , . . . , ri+1 , θi+1 ) 7→ − r .
2 i=1 i

So J is basically the radius squared (upto factor − 21 ). So then:

X
J −1 (−2)/S 1 = {(z1 , . . . , zn , zn+1 ) | |zi |2 = 1}/S 1 = CP n .
So S 1 is the stabiliser of the covector, but the Lie algebra is abelian, so the
stabilisor is everything, so we are modding out the whole unit circle. Another
way to think of it: what part of the group that’s acting preserves this level set?
And that’s everything.
Note that it was very important that we consider the fiber here. If we would
just consider Cn+1 /S 1 , we would end up with an odd-dimensional manifold.
Another natural example of symplectic reduction comes from any group
action.
Proposition 30. Suppose we are given an action of G on a manifold Q.

Let ρ : g → X(M ) be the infinitesimal generator of the action. Recall the
action of G on the cotangent bundle T ∗ Q. This action admits a moment
map:
∀ζ ∈ Tp∗ Q hJ(ζ), vi := hζ, ρ(v)p i .
Assume that Q/G is a smooth manifold. Then J −1 (0)/G is symplectomor-
phic to T ∗ (Q/G)
Proof. Sketch of proof. First define h : J −1 (0) → T ∗ (Q/G) :
∀v ∈ Tπ(p) (Q/G) hh(η), vi := hη, ṽi ,
where ṽ ∈ Tp Q is such that dp πṽ = v. Think of h to be a push forward

map. ‘And we define it like a splitting of this’
Exercise. Show that h : J −1 (0) → T ∗ Q is a well-defined surjective submer-
sion and that the fibers of h coincide with the orbits of G.
Therefore the following commutes:
J −1 (0)
h
quotient
J −1 (0)/G T ∗ (Q/G)
and the bottom left right arrow because the fibers are the same. Now we
need to check that these arrows are a symplectomorphism
Exercise. Recall the tautological 1-form α on T ∗ (Q/G). Let α̃ denote the
tautological 1-form on T ∗ Q. Show that:
α̃|J −1 (0) = h∗ α.

Lecture 12
From this exercise, we conclude that we have a symplectomorphism

between J −1 (0)/G and T ∗ (Q/G).
Why? So we have a tautological one form on T ∗ (Q/G). And the exercise
says that when we pull it back, we get the tautological one form on J −1 (0).
So the same holds for the symplectic form. So unique(?) then also the
left-down arrow and hence symplectomorphic?
Lecture 12
di 12 mei 10:28
6.3 Reduction in stages
Definition 61. Suppose G1 , G2 are Lie groups that act on M . The actions
are said to commute if
∀g ∈ G1 , h ∈ G2 , p ∈ M g · (h · p) = h · (g · p).
Exercise. Suppose G1 and G2 have well defined actions on (M, ω) with moment
maps J1 : M → g1 and J2 : M → g2 resp. Show that there is a well defined
action of the product group G1 × G2 on M . Show that the dual of the Lie
algebra of G1 × G2 is canonically isomorphic to g∗1 ⊕ g∗2 . Finally, show that
J1 ⊕ J2 : M → g∗1 ⊕ g∗2 is a moment map for this action.
Given commuting Hamiltonian actions, one can define a Hamiltonian action
on the reduced spaces. So the following lemma says that when you first perform
Hamiltonian reduction w.r.t one of the action, then you still have a Hamiltonian
action on the reduced space with the other group.
Lemma 10 (6.3.2). Suppose G1 and G2 are compact groups with commuting

Hamiltonian actions on (M, ω). Let J1 and J2 be moment maps for this
action. If G1 acts freely on J1−1 (0) a then:
• G2 acts on J1−1 (0)/G1 by g · (G1 · p) := G1 · (g · p).
| {z }
∈J1−1 (0)/G1
• The moment map descends to a function J2 induces a map
J2 : J1−1 (0)/G1 → g∗ .
So there is a well defined function when we restrict J2 to the fiber

and then looking at what it does to orbits
• This function is a moment map. J2 is a moment map for the action
of G2 on J1−1 (0)/G1
a Because G , G are compact groups and free action of compact group gives a man-
1 2
ifold.

Lecture 12
Proof. Sketch of proof.
• A simple calculation shows that J1 is G2 invariant. This implies that

the action of G2 on J1−1 (0) is well defined. The fact that the actions
commute implies that the action descends to a well defined actionon
J1−1 (0)/G1
• Since J2 is G1 invariant it descends to a smooth function
• Exercise.
Remark. Mind the distinction between equivariant and invariant! J1 is G2 in-

variant, but G1 equivariant.
Theorem 11 (6.3.3). Assume we are in the same situation as the previous

lemma. Furthermore, assume that J := J1 ⊕ J2 is a moment map for the
action of G1 × G2 and the action of G1 × G2 on J −1 (0, 0) is free. In that
case:
−1
J2 (0)/G2 ∼ = J −1 (0, 0)/(G1 × G2 ),
canonical symplectomorphism.
In other words, if we can split a Lie group as a direct product, then we can
perform symplectic reduction in two stages. For the proof and more detail, see
24.3 in Lectures on symplectic geometry, Cannes da Silva.
Example. Consider C3 with the following two actions of G1 = G2 = S 1 .
eiθ · (z1 , z2 , z3 ) := (eiθ z1 , eiθ z2 , eiθ z3 ) eiθ ? (z1 , z2 , z3 ) := (eiθ z1 , z2 , z3 ).
These two actions commute and have moment maps:

3
!
1 X 1 1
J1 (z1 , z2 , z3 ) = 1− zi zi J2 (z1 , z2 , z3 ) = − z1 z1 .
2 i=1
2 2
1
The level set J1−1 (0) is the 5-dimensional unit sphere. As we saw inn a
previous lecture, J1−1 (0)/G1 = CP 2 . This way, we get:

1 1 z1 z1 −1
J2 ([z1 : z2 : z3 ]) = − P J2 (0)/G2 = CP 1 .
2 2 zi zi
Note that this function is well defined.

1 You shouldn’t worry about the 1−. Moment maps are not unique, and group abelian, so
moment maps can differ by constant. Here we tried to arrange it such that the level set of 0
is exactly the sphere of radius one! And on the right we have that the level set gives sphere
of norm 21 .

Lecture 12
6.4 The convexity theorem

Definition 62. Let V be a vector space and U ⊂ V be a subset. U is said
to be convex if for all v, w ∈ U and t ∈ [0, 1] then tv + (1 − t)w ∈ U . The
closed convex hull Convex(U ) is defined to be the intersection of all closed
convex subsets of V which contain U .
Theorem 12 (6.4.2: Atiyah-Guillemen-Sternberg 1982). Let (M, ω) be a com-

pact symplectic manifold and Tk act on M with moment map J : M → g∗ .
Then
J(M ) = Convex {fixed points of Tk } ,

where
{fixed points of T} = {p ∈ M | g · p = p for all g ∈ Tk }.

Why is this interesting? This tells you about the structure of the manifold.
This J can be thought of as giving you the shadow of the manifold on g∗ . For
example, if k = 2, then it tells you the manifold projects onto a convex subset
of the two dimensional plane. This can teach you about geometry of manifold
itself. You can learn a lot this way about manifolds with torus actions (there is
even a classical classification theorem using this.)
The other reason of this is because sometimes people are interested in fixed
points of group actions, e.g. fixed points of Hamiltonian flows.
Corollary. The action has at least one fixed point.
Remark. (So this way we can see in a convoluted way: acting by rotation on a
torus is not a Hamiltonian action, because it doesn’t have fixed points!
For a proof by Ana Rita Pires, see http://pi.math.cornell.edu/~apires/
convexitytalk.pdf
Remark. You should remember the statement of this theorem! The proof is
involved, uses a inversion of Darboux’s theorem, and also a bit of morse theory!
Example. Recall the action of S 1 on S 2 where the symplectic form in cylindrical
coordinates is given by dθ ∧ dz.
ψ · (θ, z) := (θ + ψ, z) J(θ, z) = z.
The fixed points of the action are the north and south pole pn , ps . So, as S 1 is
a torus, and M is compact,
J(S 2 ) = Convex({J(pn ), J(ps )}) = Convex({−1, 1}) = [−1, 1].

Lecture 12
Example. Consider the action of T2 on CP 2 , where
(θ1 : θ2 ) · [z1 : z2 : z3 ] := [eiθ1 z1 : eiθ2 z2 : z3 ].
The fixed points of the action are
[1 : 0 : 0] [0 : 1 : 0] [0 : 0 : 1].
The moment map for this action is
(|z1 |2 , |z2 |2 )
J : CP 2 → R2 : [z1 : z2 : z3 ] 7→ − .
2|(z1 , z2 , z3 )|2
You can compute this by first looking at the action of θ1 and then, because sum
of, . . . , you get (. . . , . . .).
(− 12 , 0) = J([1 : 0 : 0])
(0, 0) = J([0 : 0 : 1])
(0, − 12 ) = J([0 : 1 : 0])
Figure 6.1
This picture gives you a good look at the regular values of the moment map.
In this case, they will be the interior points of this triangle.
Now, an application of this:
Theorem 13 (6.5.6: Schur-Horn 1954). Fix some real numbers (I1 , . . . , In ) ∈

Rn . Consider the function
p : {A ∈ Herm(n) : Eigval(A) = {I1 , . . . , In }} −→ Rn

A 7−→ (A11 , . . . , Ann ) = Diag(A).
The image of p is
Convex({(Iσ(1) , . . . , Iσ(n) ) | σ ∈ Sn }).

Lecture 12
Proof. Idea of the proof. Show that p is the moment map for a Hamiltonian
Tn action. How? The first step is to show that there is an isomorphism
∗
u(n) with Herm(n) and that under this identification the coadjoint repre-
sentation of U (n) on u(n)∗ is by conjugation.
And we know stuff about coadjoint actions! Orbits are symplectic man-
ifolds. And coadjoint action is Hamiltonian action, where moment map is
just the inclusion.
And every Hermitian matrix can be diagonalized by a unitary matrix.
It follows that
 
I1 0
{A ∈ Herm(n) : Eigval(A) = {I1 , . . . , In }} = U (n) ·  ..
. ,
 
0 In
so it equal to the coadjoint orbit of the diagonal matrix. Note that this
dot is not matrix multiplication, but conjugation! So the domain of p is a
symplectic manifold! And there is also a sorta moment map by including
it into the set of hermitian matrices.
But to apply the theorem, we need a torus action. How can we find
such a thing? Well, inside of U (n) there is a torus Tn with Lie algebra t
(diagonal matrices.)
Restricting the Lie group U (n) to just the torus, it’s no longer transitive,
but that action is Hamiltonian. The Hamiltonian action is inclusion in
hermitian matrices, and then projection to t∗ . The projection map g∗ →
t∗ = Rn corresponds to
 
A11 · · · A1n
 .. .. ..  7→ (A , . . . , A ).
 . . .  11 nn
An1 ··· Ann
So p is the moment map for a Hamiltonian action. So the image is

convex. Furthermore, the only elements stabilized by the torus action are
the diagonal elements of the set
{A ∈ Herm(n) : Eigval(A) = {I1 , . . . , In }}.
These are exactly diagonal matrices with as diagonal entries I1 , . . . , In ,

permuted in any order. Taking the moment map then, we find
Convex({(Iσ(1) , . . . , Iσ(n) ) | σ ∈ Sn }).

symplectic-geometry-lecture-notes

Uploaded by

symplectic-geometry-lecture-notes

Uploaded by

Symplectic geometry

June 11, 2020

1 Symplectic linear algebra 3

3 Submanifolds and Normal Forms 20

4 Review of Lie groups 38

• Main text: Ana Cannas da Silva Lectures on symplectic geometry

Symplectic linear algebra

1.1 Symplectic vector spaces

• V is a real vector space

Exercise. (V, Ω). Then Ω̃ : V → V ∗ : v 7→ Ω(v, ·) is an isomorphism.

Corollary. A symplectic vector space is even dimensional.

Ω(ei , ej ) = 0 Ω(fi , fj ) = 0 Ω(ei , fj ) = δij .

The matrix representation is given by

Definition 2. Let (V, Ω) be a symplectic vector space, W ⊂ V a subspace.

Definition 3. Let (V1 , Ω1 ), (V2 , Ω2 ) be symplectic vector spaces. A linear

Exercise. Consider the standard symplectic vector space. A ∈ M2n×2n . Then

Proposition 1. If (V, Ω) is a symplectic vector space, then there exists a

φ : (R2n , Ωcan ) → (V, Ω),

Proof. We know that dim V = 2n, for some n. Finding φ is equivalent to

Ω(ei , ej ) = 0 Ω(fi , fj ) = 0 Ω(ei , fj ) = δij .

Base case dim V = 2. Pick any e1 6= 0. As Ω is non-degenerate, there is

CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 4

some f1 such that Ω(e1 , f1 ) 6= 0. Rescale and we’re done.

Ω(e1 , ei ) = 0 Ω(e1 , fi ) = 0 Ω(f1 , ei ) = 0 Ω(f1 , fi ) = 0

So (e1 , . . . , en , f1 , . . . , fn ) is a basis, and it satisfies all he relations.

Corollary. Any two symplectic vector spaces are symplectormorphic if their

1.2 Subspaces and reduction

• isotropic iff W ⊂ W Ω . (The symplectic form restricts to 0)

Exercise. Are the subspaces from exercise 6 isotropic, coisotropic, lagrangian?

• isotropic, then dim W ≤ dim V

• coisotropic, then dim W ≥ dim V

• lagrangian, then dim W = dim V

Exercise. If W ⊂ V coisotropic. Then Wred := W

CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 5

Proposition 2. Let W ⊂ V coisotropic and let π : W → Wred = WWΩ . Then

Proof. Note that π is surjective.

Ωred (ṽ, w̃) = Ω(v, w), π(v) = ṽ, π(w) = w̃.

Why is this will defined? Suppose π(v1 ) = π(v2 ), π(w1 ) = π(w2 ). So

Antisymetry is okay, the only thing left to check if the non-degeneracy.

1.3 Compatible complex structures

Exercise. Any such J induces a structure of a C-vector space!

Definition 6. If (V, Ω) is a symplectic vector space. Then we call a complex

g(v, w) = Ω(v, Jw)

Exercise. If J is some complex structure on V . Show that g(v, w) is symmetric

CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 6

Proof. Let U = A + iB. Then

Check that matrices of this form obey M T ΩM = Ω, for Ω = Ωcan.

Lag(R2n ) = {W ⊂ R2n | W is Lagrangian}

has the structure of a smooth manifold. The Lagrangian is canonically

CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 7

Proof. Suppose W ⊂ V Lagrangian, M ∈ Symp(Ωcan ), then M W is still

CHAPTER 1. SYMPLECTIC LINEAR ALGEBRA 8

2.1 Vector bundles, vector fields, differential forms

Definition 7. A k-form on a manifold M is a section of Λk T ∗ M . We write

Exercise. Show there is a natural bijection

Ωk → {C ∞ alt. multilin. maps X(M ) × X(M ) . . . × X(M ) → C ∞ (M )}

Punchline: the set of differential forms is a differential graded commutative

Definition 8. A R-algebra is a (real) vector space A with a bilinear function

Example. C ∞ (M ) is a commutative algebra. 

Definition 9. Let A be an algebra. A derivation is an R-linear map D :

D(a · b) = D(a) · b + a · D(b).

Remark. Der(C ∞ (M )) = X(M ).

[a, [b, c]] + [c, [a, b]] + [b, [c, a]] = 0,

Example. Let (A, ·) be an algebra. Then we can define

Definition 11. A Z-graded vector space, is a vector space with a composi-

Remark. Geometrically, we’re not interested in f + dx, or dx + dx ∧ dy or some-

Definition 12. A Z-graded algebra A is an algebra that is a Z-graded vector

Definition 13. A is graded commutative if for all a ∈ Ak , b ∈ Aj :

Remark. If we use a trivial grading, and we put everything in A0 , then this

Proposition 4. Ω• is a graded commutative algebra.

CHAPTER 2. SYMPLECTIC MANIFOLDS 10

Definition 14. A degree k-derivation of a Z-graded algebra A is a linear

D(Aj ) ⊆ Aj+k D(ab) = D(a)b + (−1)kj aD(b),

where j is the degree of a. We write Derk (A).

Remark. Derivations form a graded Lie algebra. If D ∈ Derk (M ) and D0 ∈

[D, D0 ] := D ◦ D0 − (−1)kl D0 ◦ D ∈ Derk+l (A).

Example. C ∞ (M ) is a commutative algebra.