MATH 257A Symplectic Geometry

Shintaro Fushida-Hardy
381B Sloan Hall, Stanford University, CA

This document contains course notes from MATH 257A (taught at Stanford, fall 2019)
transcribed by Shintaro Fushida-Hardy. This was largely live-TeXed during lectures, but
some additional theorems, propositions, thoughts and mistakes have been added.
Contents

1 Symplectic geometry 4
1.1 Lecture 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Admin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Lecture 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Different levels of symplectic structures . . . . . . . . . . . . . . . . 6
1.3 Lecture 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 Naturality of the tautological form . . . . . . . . . . . . . . . . . . . 8
1.3.2 Converse to the naturality result . . . . . . . . . . . . . . . . . . . . 9
1.4 Lecture 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.1 Sub-objects of symplectic vector spaces . . . . . . . . . . . . . . . . 11
1.4.2 Sub-objects of symplectic vector bundles and manifolds . . . . . . . 12
1.5 Lecture 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.1 More Lagrangian examples . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.2 Normal form theorem for symplectic vector spaces . . . . . . . . . . 14

2 Complex geometry 17
2.1 Lecture 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Corollaries of the normal form theorem . . . . . . . . . . . . . . . . 17
2.1.2 Complex structures on symplectic vector spaces . . . . . . . . . . . . 17
2.2 Lecture 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 2-out-of-3 property . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Lecture 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 Exploration of complex vector bundles following the 2-out-of-3 property 22
2.3.2 Integrability of almost complex manifolds . . . . . . . . . . . . . . . 25
2.4 Lecture 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.1 Stein Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Kähler manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Lecture 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5.1 K3 surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1
 2.5.2 Dolbeault cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.3 Construction of symplectic manifolds . . . . . . . . . . . . . . . . . . 32
2.6 Lecture 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.6.1 Symplectic reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3 Physics 36
3.1 Lecture 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.1 Solutions to PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.2 Hamiltonian vector fields . . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.3 Poisson bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Lecture 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.1 Various isotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Lecture 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Symplectic isotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4 Lecture 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.1 Hamiltonian isotopies . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.2 Infinitesimal group actions . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 Lecture 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.1 Lie theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.2 Adjoint and coadjoint representations . . . . . . . . . . . . . . . . . 46
3.6 Lecture 17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6.1 Coadjoint action examples . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6.2 Moment and comoment maps . . . . . . . . . . . . . . . . . . . . . . 49
3.7 Lecture 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.1 Moment map examples . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7.2 Symplectic reduction revisited . . . . . . . . . . . . . . . . . . . . . 51

4 Local and global invariants 52

4.1 Lecture 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.1 Darboux’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Lecture 20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.1 More theorems proved with Moser’s trick . . . . . . . . . . . . . . . 54
4.3 Lecture 21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.1 Extension theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Lecture 22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.1 Applications of the extension theorem . . . . . . . . . . . . . . . . . 57
4.4.2 Gromov non-squeezing . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.5 Lecture 23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.5.1 Gromov non-squeezing: proof idea (in only one lecture?!) . . . . . . 59
4.6 Lecture 24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.6.1 Lagrangians in linear symplectic reductions . . . . . . . . . . . . . . 61

2
4.7 Lecture 25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.7.1 Lagrangians in general symplectic reductions . . . . . . . . . . . . . 62
4.7.2 Symplectic category . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.8 Lecture 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.8.1 Exact Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.9 Lecture 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.9.1 Poincaré’s last geeometric theorem (Umut’s talk) . . . . . . . . . . . 66
4.9.2 Wavefronts (Jordi’s talk) . . . . . . . . . . . . . . . . . . . . . . . . 66
4.10 Lecture 28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.10.1 Surgery theory (Eric’s talk) . . . . . . . . . . . . . . . . . . . . . . . 67
4.10.2 Soft Lagrangian obstructions (My talk) . . . . . . . . . . . . . . . . 67
4.10.3 h-principle (Alec’s talk) . . . . . . . . . . . . . . . . . . . . . . . . . 69

3
Chapter 1

Symplectic geometry

1.1 Lecture 1
1.1.1 Admin
Most probable course outline:

• First half of the course: First 8 lectures of Weinstein - “Lectures on Symplectic

Manifolds”

• Next quarter of the course: Gromov non-squeezing, possibly following Gromov’s

paper.

• Final quarter of the course: “Symplectic rigidity: Lagrangian submanifolds” by

Audin-Lalonde-Polterovich.

Prerequisites:

• Familiarity with vector fields and flows on smooth manifolds.

• Familiarity with differential forms and integration, and basic de Rham theory.

• Transversality theorem.

Other resources:

• McDuff-Salamon (big, good reference book)

• Da Silva (easy to read, similar to Weinstein)

• Arnol’d - “Methods of Classical Mechanics” (for the physically minded)

4
1.1.2 Motivation
Symplectic geometry is the study of symplectic manifolds, that is, the study of smooth
manifolds equipped with a closed non-degenerate 2-form. More explicitly, a symplectic
manifold is the data (M, ω), where ω satisfies the following properties:

1. ω ∈ Ω2 (M ), i.e. ω is an anti-symmetric bilinear form on Tp M for each p in M, which

varies smoothly on M .

2. ω is non-degenerate. This means for each p ∈ M , v ∈ Tp M non-zero, there exists

v 0 ∈ Tp M such that ω(v, v 0 ) 6= 0.

3. ω is closed. This means dω = 0, where d is the exterior derivative.

This last condition is difficult to motivate. However, all three conditions are necessary for
Darboux’s theorem to hold, which can be thought of as motivation.

Theorem 1.1.1 (Darboux). Let (M, ω) be a symplectic manifold, and p ∈ M . Then there
are local coordinates {x1 , . . . , xn , y1 , . . . , yn } such that ω = dx1 ∧ dy1 + · · · + dxn ∧ dyn .

This result classifies all symplectic manifolds locally: the only local invariant up to
symplectomorphism (defined later) is dimension. On the other hand, familiar structures
such as Riemannian metrics have non-trivial local invariants (e.g. curvature).
Why study symplectic manifolds?

1. They arise naturally in other areas of maths:

(a) Classical mechanics, or more specifically, Hamiltonian mechanics. Given a man-

ifold X of “coordinates in space”, the cotangent bundle T ∗ X defines the “phase
space”. The Hamiltonian is a smooth function

H : T ∗ X → R.

A symplectic structure allows the Hamiltonian to describe time evolution (dy-

namics) on X.
(b) Complex geometry. Any affine variety which is also a complex manifold (more
generally, a Stein manifold) has a natural symplectic structure which is unique
up to symplectomorphism.
(c) Lie groups/Lie algerbas. Let G be a Lie group and g its Lie algebra. Recall the
adjoint representation G → Aut(g), which sends each g ∈ G to Ad(g) = dΦg ,
where Φg is the inner automorphism of g. The coadjoint representation is then
a map G → Aut(g∗ ), defined by g 7→ Ad(g−1 )∗ . The orbit of each F ∈ g∗ under
the coadjoint action is a submanifold of g∗ , and in fact carry natural symplectic
structures.

5
 (d) Representation varieties in low dimensional topology.
2. Worth studying for its own sake: On one hand, there are no local invariants, but
on the other hand, there are beautiful and complicated global invariants. For closed
symplectic manifolds, a simple global invariant is volume. Despite the lack of local
invariants, symplectic manifolds are still “interesting”, in the sense that the choice
of ω is important. We will observe that symplectic structures usually have large
symmetry groups, while metrics generally have small symmetry groups.
3. Finally, there is input from other fields.

1.2 Lecture 2
1.2.1 Different levels of symplectic structures
1. Symplectic vector space
2. Symplectic vector bundle
3. Symplectic manifold

Definition 1.2.1. Let V be a finite dimensional vector space, and Ω : V × V → R a

bilinear map. Note that the data of Ω is equivalent to that of a linear map Ω e : V → V ∗,
defined by Ω(v)
e : w 7→ Ω(v, w). (V, Ω) is a symplectic vector space if Ω is antisymmetric
and non-degenerate (in the sense that Ω e is an isomorphism). Note that there is not yet
enough structure to make sense of Ω being closed. In this picture, a symplectomorphism
is any bijective linear map which preserves the symplectic structure. Explicitly, a linear
bijection T : (V1 , Ω1 ) → (V2 , Ω2 ) is a symplectomorphism if Ω1 (v, w) = Ω2 (T v, T w) for all
v, w ∈ V1 . More concisely, this means Ω1 = T ∗ Ω2 .
Example. Let U be a real vector space. Then U ⊕U ∗ has a natural symplectic structure de-
fined by Ω(u+u∗ , v+v ∗ ) = v ∗ (u)−u∗ (v). Suppose we choose a basis {e1 , . . . , en , f1 , . . . , fn }.
Then it is clear that Ω agrees with Darboux’s theorem. Therefore, all symplectic vector
spaces arise as (U ⊕ U ∗ , Ω) for some U .
Definition 1.2.2. Let X be a manifold, and π : E → X a vector bundle. E is a symplectic
vector bundle if it is equipped with a smoothly varying symplectic form Ωx (as in definition
1.2.1) defined on the fibres Ex for each x ∈ X. More formally, smoothly varying means
that Ω is a smooth section of the tensor bundle E ∗ ⊗ E ∗ → X. Here a symplectomorphism
is a vector bundle isomorphism which pulls back the symplectic form.
Example. Recall that all symplectic vector spaces arise as direct sums U ⊕ U ∗ . Do all
symplectic vector bundles arise as Whitney sums P ⊕ P ∗ → X?
No. The sphere S2 admits a symplectic structure on its tangent bundle. However, any
line bundle on S2 is trivial, so if the tangent bundle of S2 cannot be a sum bundle.

6
Definition 1.2.3. Let X be a manifold. A symplectic manifold is the data (X, ω) where ω
is a symplectic structure on T X → X as in definition 1.2.2, which is also closed (dω = 0).
In this picture a symplectomorphism is a diffeomorphism which pulls back the symplectic
form.

Example. Let V be a vector space, considered as a manifold. Then the manifold T ∗ V is

diffeomorphic to V × V ∗ , and so it has a natural symplectic structure as follows: At any
point (v, v ∗ ) in T ∗ V , T(v,v∗ ) (T ∗ V ) = V ⊕V ∗ which carries a natural symplectic form Ω(v,v∗ ) .
Define ω to be this symplectic form at each tangent space. Since it does not depend on
the base point, it is exact.
Consider local coordinates (q1 , . . . , qn , p1 , . . . , pn ), with the qi induced by ei and pi
induced by e∗i . The qi represent position, while pi represent momentum. Then

ω = dp1 ∧ dq1 + · · · + dpn ∧ dqn . (1.1)

Example. Now suppose X is a manifold. Then T ∗ X has a natural symplectic structure

as in the previous example. One method of construction is to consider an atlas of X. This
induces an atlas of T ∗ X. Each of these charts is diffeomorphic to a subset of R2n , so they
can be endowed with the natural symplectic structure from the above example. Moreover,
they can be shown to glue correctly.
Another approach is to construct the so-called tautological 1-form. Consider the pro-
jection map π : T ∗ X → X. Each point px in T ∗ X is of the form (x, ξ) where ξ ∈ Tx∗ X.
Thus, define λ ∈ T ∗ (T ∗ X) pointwise by

λ(x,ξ) = ξ ◦ dπ = dπ ∗ ξ. (1.2)

Given a vector v ∈ T (T ∗ X), λ(x,ξ) first projects v onto Tx X, and then since ξ ∈ Tx∗ X, the
above formula defines a 1-form. The map λ is called the tautological 1-form. In fact, dλ
defines a symplectic form, called the canonical 2-form.
Closedness of dλ is clear. To prove that dλ is non-degenerate and antisymmetric, we
compute in coordinates. Let {Uα } be an atlas of X. Choose any point (x, ξ) in T ∗ Uα . In
local coordinates, (x, ξ) = {q 1 , . . . , q n , p1 , . . . , pn } as in the previous example. Choose a
vector v ∈ T(x,ξ) (T ∗ X), and choose αi , β i such that v(x,ξ) = αi ∂q∂ i + β i ∂p ∂
i . Then

  ∂ ∂ 
λ(x,ξ) (v) = ξ dπ αi i + β i i
∂q ∂p
 ∂ 
= pj α i i
∂q
i
= pi α .
∂
The last line follows from the fact that pj are induced by the dual basis to ∂q i
. On the

7
other hand,
 ∂ ∂ 
pj dq j (v) = pj dq j αi i + β i i
∂q ∂p
i
= pi α .
Since these must hold for all (x, ξ) and v, it follows that λ(x,ξ) = pi dq i in local coordinates.
But now dλ = dpi ∧ dq i , which is a non-degenerate antisymmetric form.
This last example is very significant because it is defined globally on T ∗ X, for any X.

1.3 Lecture 3
1.3.1 Naturality of the tautological form
Let ϕ : M → N be a diffeomorphism. Then there is a natural map dϕ : T M → T N , and
in turn, a natural map dϕ∗ : T ∗ N → T ∗ M . Explicitly, dϕ and dϕ∗ are defined by
dϕ(v)(f ) = v(ϕ∗ f ) = v(f ◦ ϕ), dϕ∗ (α) = α ◦ dϕ. (1.3)
In fact, dϕ∗ is a diffeomorphism. This is immediate if dϕ is a diffeomorphism. The latter
is easily seen to be injective. For surjectivity, let u ∈ T N . Consider v = u ◦ (ϕ−1 )∗ ∈ T M .
Then dϕ(v)(f ) = u ◦ (ϕ−1 )∗ (ϕ∗ f ) = u(f ), since (ϕ−1 )∗ = (ϕ∗ )−1 .
Hereafter we denote dϕ∗ by Φ. Recall that T ∗ M and T ∗ N are equipped with tauto-
logical 1-forms, λM and λN . We claim that dΦ∗ λM = λN . (Here is a helpful? diagram for
myself.)
dπN πN

T (T ∗ N ) T ∗N TN N
dΦ Φ = dϕ∗ dϕ ϕ

T (T ∗ M ) T ∗M TM M

πM
dπM

Theorem 1.3.1. Let ϕ : M → N be a diffeomorphism, and define Φ : T ∗ N → T ∗ M as

above. Then dΦ∗ λM = λN , where λM,N are the tautological forms on T ∗ M, T ∗ N .
Proof. Choose any v ∈ T(z,ζ) (T ∗ N ). Choose the unique (x, ξ) such that ϕ(x) = z and
Φx ζ = ξ. Then
(dΦ∗ λM )(z,ζ) v = λM Φ(z,ζ) (dΦ(v))
= λM (x,ξ) (dΦ(v))
= ξ(dπM (dΦ(v)))
= ζ ◦ dϕx (dπM (dΦ(v))).

8
But by the commutative diagram above, dϕ ◦ dπM ◦ dΦ = dπN . Hence we have

(dΦ∗ λM )(z,ζ) v = ζ ◦ dϕx (dπM (dΦ(v))) = ζ(dπN (z,ζ) (v)) = λN (v)

as required.

This can also be derived in coordinates. Let pi dq i be the tautological form of T ∗ N , and
qi the tautological form of T ∗ M . Since dqi are contravariant while pi are covariant,
pei de
qi is pulled back to pi dq i .
pei de

Corollary 1.3.2. Let ωM,N denote the canonical 2-form on T ∗ M, T ∗ N respectively. Then
with ϕ, Φ as above, dΦ∗ ωM = ωN .

In fact, a converse also holds. That is, if Φ : T ∗ N → T ∗ M is a diffeomorphism such

that dΦ∗ λM = λN , then Φ = dϕ∗ for some diffeomorphism ϕ : M → N . To prove this, we
introduce a certain canonical vector field.

1.3.2 Converse to the naturality result

Definition 1.3.3. Let X be a smooth manifold. The Liouville vector field VX on T ∗ X is
defined by ιVX ω = λ. Since ω is non-degenerate, this is well defined.
∂ i ∂
We now determine V in coordinates. Initially suppose V = v i ∂p i + u ∂q i , and let
∂ i ∂
W = wi ∂p i + s ∂q i . Then we require ω(V, W ) = λ(W ), so in coordinates,

 ∂ ∂ ∂ ∂ 
(dpj ∧ dq j ) v i i + ui i , wi i + si i = v j sj − uj wj
∂p ∂q ∂p ∂q
 ∂ ∂ 
= pj sj = (pj dq j ) wi i + si i .
∂p ∂q
∂
Therefore, V = pi ∂p i . This is a “fibre-wise radial vector field”. That is, given any (x, ξ) ∈

T X, V(x,ξ) is a vector in T(x,ξ) (T ∗ X) which points radially away from (x, 0).
∗

We now make some observations regarding V . Let M, N be manifolds, and Φ : T ∗ N →

T M a diffeomorphism. Since ω and λ are preserved by the pullback dΦ∗ , and Φ is a
∗

diffeomorphism, the pushfoward of VN to VM is well defined, and

dΦVN = VM . (1.4)

Secondly, Let ϕtN,M denote the flows of −VN,M respectively. Then by equation 1.4, for all
t,
Φ ◦ ϕtN = ϕtM ◦ Φ. (1.5)
We are now ready to prove that Φ arises as dϕ∗ for some diffeomorphism ϕ : M → N .
This is done in three steps.

9
 (1) Let ZN be the zero section of T ∗ N . That is, the collection of all points (z, 0) for
z ∈ N . By the coordinate expression of the Liouville vector field, we see that it vanishes
exactly on zero sections. But now by equation 1.4, VM must vanish on Φ(ZN ). Therefore
Φ(ZN ) is contained in ZM (where the latter is the zero section of T ∗ M .) Since Φ is a
diffeomorphism, arguing with the inverse shows that ZN is mapped onto ZM .
(2) We now show a stronger result, that Φ maps cotangent fibres to cotangent fibres. By
our explicit form of the Liouville vector field, given any (z, ζ) in T ∗ N , we have ϕtN (z, ζ) →
(z, 0) as t → ∞. Let Ψ : ZN → ZM be the restriction of Φ to the zero sections as above.
We wish to show that Φ sends (z, ζ) to the fibre above Ψ(z, 0). But this is true by equation
1.5, since
ϕtM (Φ(z, ζ)) = Φ(ϕtN (z, ζ)) → Ψ(z, 0) as t → ∞,
and ϕtM is purely radial, so Φ(z, ζ) must lie above Ψ(z, 0).
(3) The remainder of the proof is a coordinate calculation. Let pi , qi denote canonical
coordinates on T ∗ M , and pei , qei denote canonical coordinates on T ∗ N . Let ϕ : M → N be
the diffeomorphism induced by Ψ. Since we showed above that cotangent fibres map to
cotangent fibres, we must have qei = ϕ(qi ). On the other hand, there is some map ρ such
that pi = ρ(e
pi , qei ). This gives

q i = pei d(ϕ(qi )) = pei dϕ ◦ dqi = (dϕ∗ pei )dq i .

pi , qei )dq i = pi dq i = pei de
ρ(e

pi , qei ) = dϕ∗ pei . Since Φ sends pei to pi , this proves that Φ = dϕ∗ . In
But this forces ρ(e
summary,
Theorem 1.3.4. Let Φ : T ∗ N → T ∗ M be a diffeomorphism such that dΦ∗ λM = λN . Then
Φ = dϕ∗ for some diffeomorphism ϕ : M → N .
Example. In the above theorem, is not enough that Φ pulls back symplectic forms. (That
is, it is essential that it pulls back the tautological 1-form.) To see this, consider the
cotangent bundle of the circle, T ∗ S1 = R/Z × R, with p, q coordinates. Consider a Dehn
twist T : T ∗ S1 → T ∗ S1 , with the latter space expressed in pe, qe coordinates. The twist
leaves p coordinates unchanged, while q gets mapped to qe = q + f (p). Therefore

p ∧ de
de q = dp ∧ (dq + df dp) = dp ∧ dq.

That is to say, T is a diffeomorphism which preserves the canonical symplectic form.

However, to see that T does not arise as dϕ∗ for some ϕ : S → S, it suffices to show that
T does not preserve the tautological form. We have

q = p(dq + df dp) 6= pdq,

pede

since df is non-zero for a Dehn twist. In fact, this example is of great importance, since
Dehn twists can be embedded in most surfaces (as their support is an annulus which can
be made arbitrarily small).

10
1.4 Lecture 4
1.4.1 Sub-objects of symplectic vector spaces
Definition 1.4.1. Let (V, Ω) be a symplectic vector space, and W ⊂ V a subspace. Then

W Ω := {v ∈ V : Ω(v, w) = 0 for all w ∈ W }

is the symplectic complement or orthogonal of W .

Proposition 1.4.2. dim W Ω + dim W = dim V .

Proof. Define f : V → W ∗ by v 7→ Ω(v, ·). Then ker f = W Ω . Non-degeneracy of Ω

ensures that im f = W ∗ . Therefore by the rank nullity theorem,

dim W + dim W Ω = dim W ∗ + dim W Ω = dim im f + dim ker f = dim V.

Proposition 1.4.3. W = (W Ω )Ω .

Proof. We first prove that W ⊂ (W Ω )Ω . Let w ∈ W . Choose v ∈ W Ω . Then Ω(w, v) = 0,

so w ∈ (W Ω )Ω . In fact, by proposition 1.4.2, we have dim W = dim(W Ω )Ω , so equality
follows.

These two propositions raise the question of whether or not W ⊕ W Ω = V holds in

general (as it does for inner products). We find that this is not true, giving rise to the
following definitions:

Definition 1.4.4. Let W be a subspace of (V, Ω).

• W is coisotropic if W Ω ⊂ W .

• W is isotropic if W ⊂ W Ω .

• W is Lagrangian if it is isotropic and coisotropic.

• W is symplectic if W ∩ W Ω = {0}.

Example. Suppose W ⊂ V has dimension 1. Choose w ∈ W . Then for any other w0 ∈ W ,

there exists a real number r such that Ω(w, w0 ) = Ω(w, rw) = rΩ(w, w) = 0. Since w is
orthogonal to all w0 ∈ W , we have W ⊂ W Ω , that is, W is isotropic. Similarly, every
codimension 1 subspace is coisotropic.

Example. Consider (V ⊕ V ∗ , Ω) where Ω is the canonical symplectic form. Then V and

V ∗ are clearly Lagrangian subspaces. This raises the question of: what are all of the
Lagrangian subspaces of V ⊕ V ∗ ?

11
Proposition 1.4.5. W ⊂ V is Lagrangian if and only if W is (co)isotropic and half-
dimensional. (i.e. dim W = 21 dim V .)

Proof. If W is Lagrangian, it is clearly half-dimensional and both isotropic and coisotropic.

Conversely, suppose W is isotropic and half-dimensional. Then by 1.4.2, W is Lagrangian.

When is the graph of a linear map f : V → V ∗ Lagrangian? Since the dimension of the
graph is the rank of f , by proposition 1.4.5, G(f ) must be isotropic and injective. That
is, we must have Ω(v ⊕ f (v), w ⊕ f (w)) = 0 for any v ⊕ f (v), w ⊕ f (x) in G(f ), with f
injective. But the former property is exactly the requirement that

f (w)v = f (v)w.

Therefore, the graph of a linear map f : V → V ∗ Lagrangian if fe : V × V → R defined

by fe : (v, w) 7→ f (v)w is a symmetric bilinear form. A Lagrangian subspace arising as a
graph is called graphical. The previous example shows that not every Lagrangian subspace
is graphical (namely V ∗ ).

1.4.2 Sub-objects of symplectic vector bundles and manifolds

Definition 1.4.6. Let E → M be a symplectic vector bundle, and P ⊂ E a subbundle.
P is coisotropic (isotropic, Lagrangian, symplectic), if for every x ∈ M , the fibre Px ⊂ Ex
is a coisotropic (isotropic, Lagrangian, symplectic) subspace.

Definition 1.4.7. Let M be a symplectic manifold, and N ⊂ M a submanifold. N is

coisotropic (isotropic, Lagrangian, symplectic), if T N ⊂ T M |N is a coisotropic (isotropic,
Lagrangian, symplectic) subbundle.

Example. By proposition 1.4.5, a submanifold N of a symplectic manifold (M, ω) is

Lagrangian if ω|N = 0 and dim N = 12 dim M .

If a vector bundle is of the form E⊕E ∗ → X, then E and E ∗ are Lagrangian subbundles.
When is a symplectic vector bundle of the form E ⊕ E ∗ ?

Theorem 1.4.8. Let P → X be a symplectic vector bundle. Then P → X is symplecto-

morphic to E ⊕ E ∗ → X for some E if and only if P → X admits a Lagrangian subbundle.

The “only if” direction is clear; we prove this theorem in lecture 7.

Remark. There exist symplectic vector bundles which admit no Lagrangian subbundles,
but do admit half-rank subbundles. This is explored at a later point.

What are some Lagrangian submanifolds in symplectic manifolds?

12
Example. Let Σ be a surface equipped with an area-form, and L any one dimensional
submanifold. Then L is Lagrangian. Note that an area form on a surface is a symplectic
form. Let x ∈ L. We have that Tx L is a half dimensional subspace of Tx Σ. Moreover, it is
a one dimensional subspace, so it is isotropic. Therefore it is a Lagrangian subspace. Thus
L is a Lagrangian submanifold.

Example. Let M = T ∗ X be the symplectic manifold equipped with the canonical form
dλ. Let Z ⊂ T ∗ X denote the zero section, Z = {(x, 0) ∈ T ∗ X}. Then Z is a Lagrangian
submanifold. To see this, note that

dλ = dp1 ∧ dq 1 + · · · + dpn ∧ dq n .

Restricting to Z forces each p to be identically zero. Thus dλ|Z = 0, so Z is isotropic.

Since Z is half-dimensional, Z is Lagrangian. Similarly, any cotangent fibre is Lagrangian.

Proposition 1.4.9. Let s : X → T ∗ X be a section of the canonical projection. Then

s∗ λ = s, where λ denotes the tautological form on T ∗ X.

Proof. Choose a point x ∈ X. Then s(x) = (x, ξ) for some ξ ∈ Tx∗ X. The tautological
form at s(x) is given by ξ ◦ dπ. But then for any v ∈ T (T ∗ X), we have

(s∗ λs(x) )(v) = λ(s(x)) (dsv) = ξ(dπ(dsv)) = ξ(v).

The third equality is because π ◦ s = idX , so idT X = d(π ◦ s) = dπ ◦ ds. Since ξ(v) is sx (v),
we are done.

In fact, the tautological one-form is the unique horizontal one-form that “cancels” a
pullback. It follows that if ω is the canonical form, then

s∗ ω = s∗ (dλ) = d(s∗ λ) = ds.

1.5 Lecture 5
1.5.1 More Lagrangian examples
In lecture 4, we defined what it means for a submanifold of a symplectic manifold to be
Lagrangian. We can generalise this definition by stating it in terms of embeddings.

Definition 1.5.1. Let (M, ω) be a 2n dimensional symplectic manifold. Let f : N → M

be an embedding of an n-manifold. f is a Lagrangian embedding if f ∗ ω = 0.

This implies that im f is isotropic and half dimensional, hence a Lagrangian submani-
fold.

13
Example. As a corollary of 1.4.9, we obtain a new class of Lagrangian submanifolds. Let
s be a section of T ∗ X, and suppose s is closed. Then s is a Lagrangian embedding of X,
since s∗ ω = ds = 0. Therefore im s is a Lagrangian submanifold. In fact, since s is a
section of the cotangent bundle, im s can be interpreted as the graph of a function.
This example is surprisingly deep, as the calculation also shows that im s is a Lagrangian
submanifold if and only if s is closed. There is a natural bijective correspondence between
Lagrangian submanifolds of T ∗ X which project smoothly onto X and closed 1-forms on
X.
We now create a family of graphical Lagrangian submanifolds. Recall that a symmetric
bilinear form V × V → R gives rise to a graphical Lagrangian subspace of V ⊕ V ∗ .
Let (X1 , ω1 ), (X2 , ω2 ) be symplectic manifolds, and ϕ : X1 → X2 a symplectomorphism.
The canonical symplectic structure on the product space is π1∗ ω1 + π2∗ ω2 . However, we can
also define a twisted product by
e 2 = (X1 × X2 , π1∗ ω1 − π2∗ ω2 ).
X1 ×X

Moreover, the graph G(ϕ) is a Lagrangian submanifold of X1 ×X e 2 . To see this, define the
embedding Φ : X1 → X1 ×X2 by Φ : x → (x, ϕ(x)), so that im Φ = G(ϕ). Then we need
e
only show that Φ is a Lagrangian embedding, i.e. it pulls back the symplectic form to 0.
We see that
Φ∗ (π1∗ ω1 − π2∗ ω2 ) = ω1 − ϕ∗ ω2 = ω1 − ω1 = 0,
so Φ is a Lagrangian embedding as required.
In fact, the calculation shows that the converse also holds. If the graph of a diffeomor-
phism ϕ : X1 → X2 is not Lagrangian in X1 ×X e 2 , then ϕ is not a symplectomorphism.
This gives some insight into the philosophy of Weinstein that “symplectomorphisms are
special cases of Lagrangian submanifolds”.

1.5.2 Normal form theorem for symplectic vector spaces

At the beginning of lecture 1, Darboux’s theorem was stated as motivation for the re-
quirement that a symplectic form be closed. We now prove Darboux’s theorem in the
algebraic setting, that is, every symplectic vector space V arises as W ⊕ W ∗ equipped with
the canonical form, for some half dimensional vector space W . The proof approach is as
follows: first we show that V admits a Lagrangian subspace W . Then we show that W
admits a Lagrangian complement, W ⊕ W 0 = V . Finally we use this to construct the
canonical basis of V .
First choose any 1-dimensional subspace W1 of V . Then W1 is isotropic. If W1 is
half-dimensional, we are done. Otherwise, choose any w ∈ W1Ω such that w 6∈ W1 , and
define W2 to be the span of W1 and w. Since w ∈ W1Ω , it is orthogonal to both W1
and its own span, giving W2 ⊂ W2Ω . Inductively this process continues until W = Wn is
half-dimensional and isotropic, thus Lagrangian.

14
 Next, suppose W ⊂ V is Lagrangian. Let w10 ∈ V \ W . Then W10 , (the span of w10 ), is
isotropic and satisfies W10 ∩ W = {0}. If W10 is half-dimensional, we are done. Otherwise
choose w20 ∈ V \ {W ∪ W10 }. As above, this is still isotropic and trivial intersects W .
Note that 1.4.2 ensures the existence of w20 . Inductively following this process, eventually
W 0 = Wn0 is half dimensional, at which point it is also a Lagrangian subspace and has trivial
intersection with W . Since they are both half-dimensional, it follows that W ⊕ W 0 = V .
Now consider the map α : W 0 → W ∗ defined by α(w0 ) : w 7→ Ω(w, w0 ). If w0 is an
element of ker α, then w0 belongs to W Ω = W . Since W intersects W 0 trivially, w0 = 0. But
now α is an injection between two equal-dimensional vector spaces, so it is an isomorphism.
This induces an isomorphism ϕ : W ⊕ W 0 → W ⊕ W ∗ . It remains to prove that ϕ is in fact
a symplectomorphism from (W ⊕ W 0 , Ω) to (W ⊕ W ∗ , ΩW ).
For any w ⊕ w0 , v ⊕ v 0 ∈ W ⊕ W 0 , we have

(ϕ∗ ΩW )(w ⊕ w0 , v ⊕ v 0 ) = ΩW (ϕ(w ⊕ w0 ), ϕ(v ⊕ v 0 ))

= ΩW (w ⊕ α(w0 ), v ⊕ α(v 0 ))
= α(v 0 )w − α(w0 )v
= Ω(w, v 0 ) − Ω(v, w0 ) = Ω(w ⊕ w0 , v ⊕ v 0 ).

The last equality comes from the fact that W ⊕ W 0 is a Lagrangian splitting, so the cross
terms vanish. Therefore ϕ is a symplectomorphism. In summary, we have the following
theorem.
Theorem 1.5.2. Let (V, Ω) be a symplectic vector space. Let W ⊂ V be a Lagrangian
subspace. (Such a subspace exists.) Then (V, Ω) is symplectomorphic to (W ⊕ W ∗ , ΩW )
(where ΩW denotes the canonical symplectic form on W ⊕ W ∗ ).
The above theorem statement doesn’t give any insight as to the amount of freedom
we have in choosing the symplectomorphism between V and W ⊕ W ∗ . We find that
topologically there is only one choice of symplectomorphism (in the sense that the collection
of choices is a contractible space).
To see this, first note that the above proof establishes the following bijective correspon-
dence:
 
Symplectomorphisms
Lagrangian complements to
  n o
V → W ⊕ W ∗ which extend ←→
W inside V .
the identity W → W .
 

But a Lagrangian complement to W inside V ∼ = W ⊕ W ∗ projects onto W ∗ , so it

∗
is a graphical Lagrangian over W . Thus it corresponds to a symmetric bilinear form
W ∗ × W ∗ → R. In summary we have the bijective correspondence
 
Symplectomorphisms  
Symmetric bilinear forms
 
∗
V → W ⊕ W which extend ←→
W ∗ × W ∗ → R.
the identity W → W .
 

15
But the right hand side is a contractible space, since given any two symmetric bilinear
forms g and g 0 , (1 − t)g + tg 0 is a symmetric bilinear form.

16
Chapter 2

Complex geometry

2.1 Lecture 6
2.1.1 Corollaries of the normal form theorem
We can now classify all skew symmetric forms on vector spaces:
Corollary 2.1.1. Let (V, Ω) be a vector space equipped with a skew-symmetric form. Then
there exists a basis e1 , . . . , en , f1 , . . . , fn , w1 , . . . , wm of V such that for all i, j,

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

Proof. Let W = {w ∈ V : Ω(w, ·) = 0}. Choose any Ve ⊂ V such that W ⊕ Ve = V . Then

Ω restricted to Ve is a symplectic form. Let e1 , . . . , en , f1 , . . . , fn be the canonical basis for
Ve , and w1 , . . . , wm any basis for W .

Corollary 2.1.2. Suppose Ω is a skew-symmetric bilinear form on a vector space of di-

mension 2n. Then Ω is non-degenerate if and only if Λn Ω 6= 0.
Proof. This follows immediately from expressing Ω in terms of the basis in the above
corollary.

Corollary 2.1.3. If ω is a symplectic form on a manifold of dimension 2n, then Λn ω is

a non-vanishing 2n-form, i.e. a volume form.

2.1.2 Complex structures on symplectic vector spaces

Definition 2.1.4. Let V be a real vector space. A complex structure on V is a linear
map J : V → V such that J 2 = id. Let (V, Ω) be a symplectic vector space. A complex
structure J on (V, Ω) is said to be
• almost compatible with Ω if Ω = J ∗ Ω (i.e. Ω(v, w) = Ω(Jv, Jw) for all v, w ∈ V ),

17
 • tame if Ω(v, Jv) > 0 for all non-zero v ∈ V ,

• compatible if it is almost compatible and tame.

As with the various sub-objects in the previous two lectures, the definition of a com-
patible complex structure extends to symplectic vector bundles and symplectic manifolds
in the obvious way. For example, if (M, ω) is a symplectic manifold, then a compatible
almost complex structure on M is a smooth section J of End(T M ) → M such that for each
x ∈ M , Jx2 = −idTx M , and Jx is compatible with ωx .

Remark. We referred to J above as an almost complex structure rather than a complex

structure. This is because a smooth manifold is called complex if it has charts mapping
into Cn with holomorphic transition maps. Every complex structure can be realised as
an almost complex structure, but there exist 4-real-dimensional almost complex manifolds
which are not complex. It is still open as to whether there are such manifolds in higher
dimensions.

The definition of compatibility above gives rise to the famous 2-out-of-3 property of
symplectic structures, complex structures, and inner products.

Definition 2.1.5. An inner product g is said to be compatible with a complex structure

J if and only if g = J ∗ g.

Proposition 2.1.6. Let (V, Ω, J) be a symplectic vector space equipped with a complex
structure. Then Ω(·, J·) is an inner product if and only if J is compatible with Ω. Moreover,
Ω(·, J·) is compatible with J.

Proof. Suppose J is compatible. Then for any v, w ∈ V ,

Ω(v, Jw) = Ω(Jv, J 2 w) = −Ω(Jv, w) = Ω(w, Jv).

Therefore Ω(·, J·) is symmetric. Moreover, since J is tame, Ω(·, J·) is positive definite, so it
is an inner product. The converse is similar. Compatibility of Ω(·, J·) with J is immediate
from the almost-compatibility of Ω with J.

Proposition 2.1.7. Let (V, g, J) be an inner product space equipped with a compatible
complex structure. Then g(J·, ·) is a symplectic form which is compatible with J.

Proof. Let v, w ∈ V . Then

g(Jv, w) = g(J 2 v, Jw) = −g(v, Jw) = −g(Jw, v),

so g(J·, ·) is skew-symmetric. It is non-degenerate since the induced map V → V ∗ is defined

by −eg ◦ J, where ge is the isomorphism from V to V ∗ given by v 7→ g(·, v), and J is an
automorphism of V . Compatibility is again immediate.

18
 The final direction is to prove that given a symplectic vector space equipped with an
inner product, there is a canonical complex structure which is compatible with both the
symplectic structure and inner product. This takes more work - we must introduce polar
decompositions.

Proposition 2.1.8. Let (V, g) be a real inner product space. Let T : V → V be a linear
bijection. Then there exist unique P, Θ : V → V such that

• P is positive definite and self-adjoint (i.e. g(P v, v) > 0 whenever v ∈ V is non-zero,

and g(P v, w) = g(v, P w) for all v, w ∈ V ).

• Θ is isometric (i.e. g(Θv, Θw) = g(v, w) for all v, w ∈ V ).

• T = P Θ.

Proof. Consider the isomorphism T T ∗ : V → V (where V ∗ is the adjoint of V ). T T ∗ is

self adjoint and positive definite. By the spectral theorem, T T ∗ admits a positive definite
square root - i.e. a map P : V → V such that P 2 = T T ∗ . Moreover, P is self-adjoint. Now
define Θ = P −1 T . Θ is an isometry if Θ∗ Θ = id, since we then have

g(Θv, Θw) = g(v, Θ∗ Θw) = g(v, w).

This is indeed the case, because

Θ∗ Θ = (P −1 T )∗ (P −1 T ) = T ∗ P −1 P −1 T = T ∗ (P 2 )−1 T = T ∗ (T T ∗ )−1 T = id.

Clearly this construction gives T = P Θ.

We next show uniqueness. Suppose T = P 0 Θ0 for some isometry Θ0 and positive definite
self adjoint P 0 . Then T T ∗ = (P 0 Θ0 )(P 0 Θ0 )∗ = P 0 Θ0 Θ0∗ P 0 = P 02 . Since positive definite
self adjoint operators admit unique positive definite square roots, P 0 = P , and this forces
Θ0 = Θ.

Proposition 2.1.9. Let (V, g) be a real inner product space, and T : V → V a linear
bijection. Let T = P Θ be the polar decomposition of T . if T is normal (T T ∗ = T ∗ T ) then
T = ΘP .

Proof. Since T is normal, we have

P 2 = T T ∗ = T ∗ T = (Θ∗ P )P Θ = Θ∗ P 2 Θ.

Since positive definite self adjoint matrices admit unique positive definite square roots, it
must be the case that P is the unique positive definite square root of Θ∗ P 2 Θ. However,
we see that Θ∗ P Θ is positive definite since g(Θ∗ P Θv, v) = g(P Θv, Θv) > 0 for all v 6= 0,
and moreover, it is a square root of Θ∗ P 2 Θ. It follows that P = Θ∗ P Θ, so ΘP = P Θ.

19
2.2 Lecture 7
2.2.1 2-out-of-3 property
In lecture 6 we showed that symplectic + complex ⇒ inner product, and inner product
+ complex ⇒ symplectic. We now complete the result by establishing that symplectic +
inner product ⇒ complex.
More precisely, we establish that given a symplectic vector space (V, Ω) there is a canon-
ical map Θ which admits a canonical section s, giving rise to the following correspondence:
 
Θ Complex structures on V ,
{ Inner products g on V . } )−−
−−−−−*
−
s compatible with Ω.

Moreover, given any g, we show that Θ(g) is compatible with g.

Remark. While this is referred to as a correspondence, note that in general, many inner
products may map to the same complex structure.
Suppose g is an inner product on (V, Ω). Define Kg : V → V by g(Kg v, w) = Ω(v, w)
for all v, w ∈ V . Since g and Ω are non-degenerate, this gives a well defined map. Since
Kg is a linear bijection, it has a unique polar decomposition (by proposition 2.1.8) into
Kg = Pg Jg , where Jg is isometric and Pg is positive definite and self adjoint. Define the
map Θ by
Θ : g 7→ Jg .
We claim that θ is the desired map in the above correspondence. First we show that for any
g, Jg is indeed a complex structure compatible with Ω. Observe that Kg is skew adjoint
in the sense that

g(Kg v, w) = Ω(v, w) = −Ω(w, v) = −g(Kg w, v) = g(v, −Kg w)

for all v, w ∈ K. This implies that Kg is normal, so Pg Jg = Jg Pg . In particular, this

implies that Jg = Pg−1 Kg = Kg Pg−1 . But then

Jg = Kg Pg−1 = −Kg∗ (Pg∗ )−1 = −(Pg−1 Kg )∗ = −Jg∗ ,

so Jg2 = −Jg∗ Jg = −id. Jg is almost compatible, since for any v, w ∈ V ,

Ω(Jg v, Jg w) = g(Kg Jg v, Jg w) = g(Pg Jg v, w) = g(Kg v, w) = Ω(v, w).

Jg is tame, since for any v ∈ V non-zero, Pg positive definite gives

Ω(v, Jg v) = g(Kg v, Jg v) = g(Pg Jg v, Jg v) > 0.

Therefore Jg is a compatible complex structure as required. In fact, by construction, Jg is

isometric, so it is compatible with any metric in Θ−1 (Jg ). Finally, we claim that the map

20
s defined by J 7→ Ω(·, J·) is a section of Θ. Choose any J compatible with Ω. Let Θ e be
defined by g 7→ Kg (so that Θ(g) is the isometric part of Θ(g).)
e Then for any v, w ∈ V ,

Ω(−J Θ(s(J))v,
e w) = Ω(Θ(s(J))v,
e Jw) = s(J)(Θ(s(J))v,
e w) = Ω(v, w).
It follows that J Θ(s(J))
e = −id, so J = Θ(s(J)).
e In particular, it also follows that Θ(s(J))
e
is isometric, so Θ(s(J))
e = Θ(s(J)) by the uniqueness of the polar decomposition. Therefore
s is a section of Θ. This result can be summarised by the following theorem:
Theorem 2.2.1 (2-out-of-3). Let (V, g, ω, J) be a symplectic inner product space equipped
with a complex structure. Suppose g(·, ·) = ω(·, J·). Then any one structure (g, ω, or J) is
uniquely determined by the other two structures.
Corollary 2.2.2. The space C of compatible complex structures on a symplectic vector
space (V, Ω) is contractible.
Proof. The space of all inner products on V is contractible (since it is convex). Choose
any inner product g on V . Using the correspondence above, define ϕ : C × [0, 1] → C by
ϕt : J 7→ Θ ◦ ((1 − t)s(J) + tg).
This is a contraction, since ϕ0 = idS , ϕ1 = Θ(g).

Corollary 2.2.3. Every symplectic vector bundle (symplectic manifold) admits a compat-
ible complex structure. Moreover, any two compatible complex structures are homotopic.
There is also a canonical compatible complex structure that can be equipped on a
symplectic vector space, given canonical coordinates. Suppose (V, Ω) is symplectic, with
canonical basis e1 , . . . , en , f1 , . . . , fn . Then define J component-wise by
Jei = fi , Jfi = −ei .
It is easy to see that J is a compatible complex structure.
One can also ask: in what way does a choice of compatible complex structure add
rigidity to a symplectic vector space? One answer is that it provides Lagrangian comple-
ments: Suppose W ⊂ V is a Lagrangian subspace. Then JW ∩ W is trivial, and JW is
Lagrangian. Therefore JW ⊕ W = V . To see this, consider the inner product g defined by
g(·, ·) = Ω(·, J·). Since W is Lagrangian, we have
W ⊂ {v ∈ V : Ω(v, w) = 0 for all w ∈ W }
= {v ∈ V : g(v, w) = 0 for all w ∈ JW }.
Because g is an inner product, this shows that W ∩ JW is trivial. To see that JW is
Lagrangian, we observe that
JW = {v ∈ V : g(v, w) = 0 for all w ∈ W }
= {v ∈ V : Ω(v, w) = 0 for all w ∈ JW } = JW Ω .

21
As a corollary, we can finally give a simple proof that a symplectic vector bundle admits a
Lagrangian subbundle if and only if it splits into the canonical form.
Theorem 2.2.4. Let P → X be a symplectic vector bundle. If P admits a Lagrangian
subbundle L, then P arises as a sum bundle L ⊕ L∗ equipped with the canonical symplectic
form.
Proof. Suppose P admits a Lagrangian subbundle L. Since every real vector bundle admits
a metric, by the 2-out-of-3 property, P admits a complex structure. But now JL is a
complementary Lagrangian subbundle to L. By the proof of the normal form theorem,
L ⊕ JL is isomorphic to L ⊕ L∗ .

2.3 Lecture 8
2.3.1 Exploration of complex vector bundles following the 2-out-of-3
property
In the previous lecture it was shown that every symplectic vector bundle admits a com-
patible complex structure, and any two such structures are homotopic. It turns out that
this still holds when we interchange the roles of the symplectic and complex structures. In
fact, it follows that any two symplectic vector bundles are isomorphic if and only if their
corresponding complex vector bundles are isomorphic.
Proposition 2.3.1. Every symplectic vector bundle admits a compatible complex struc-
ture. Moreover, any two compatible complex structures are homotopic. (Proved previ-
ously.)
Proposition 2.3.2. Every complex vector bundle admits a compatible symplectic struc-
ture. Moreover, any two compatible symplectic structures are homotopic.
Proof. This is a counterpart to correspondence between inner products and compatible
complex structures established in lecture 7. Let (V, J) be a complex vector space. It is
easy to verify that the following (not necessarily bijective) correspondence holds, where s,
defined by Ω(·, ·) 7→ Ω(·, J·), and Φ is constructed in a similar manner to the correspondence
in lecture 7:
 
Φ Symplectic structures on V ,
{ Inner products g on V . } ) −−
−−−−−*
−
s compatible with J.
An analogous construction to the proof of corollary 2.2.2 shows that the space of compatible
symplectic structures on V is contractible. The desired result follows.

Theorem 2.3.3. Let (E1 , ω1 , J1 ) and (E2 , ω2 , J2 ) be symplectic vector bundles over X
equipped with compatible complex structures. The vector bundles are isomorphic as sym-
plectic vector bundles if and only if they are isomorphic as complex vector bundles.

22
Proof. First suppose (E1 , ω1 , J1 ) and (E2 , ω2 , J2 ) are isomorphic as complex vector bundles.
Then there is a vector bundle isomorphism Φ : E1 → E2 such that Φ∗ J2 = J1 . Therefore
Φ∗ ω2 and ω1 are both compatible with J1 . It follows by the previous proposition that there
is a one parameter family ωt of compatible structures, with ω0 = Φ∗ ω2 . But now we can
find a smooth family of vector bundle isomorphisms Ψt : E1 → E1 such that Ψ∗t Jt = J1
(McDuff-Salamon). In particular, it follows that Φ ◦ Ψ0 is a vector bundle isomorphism
preserving the complex structure. The converse is very similar.

Corollary 2.3.4. Suppose (E, ω) is a symplectic vector bundle and J1 , J2 are two com-
patible complex structures. Then (E, J1 ) and (E, J2 ) are isomorphic as complex vector
bundles.

Example. Suppose E ⊕ E ∗ → X is a symplectic vector bundle. By the above, up to

isomorphism, there is only one complex structure on the vector bundle. What is the
associated complex structure? It is in fact the complexification of E → X:
Given a real vector bundle E → X, the complexification of E is the vector bundle
E ⊕ E → X equipped with the complex structure J 0 : (e, ee) 7→ (−e e, e), which we denote
EC → X. Equivalently, the complexification is the tensor bundle E ⊗ C → X. Complex
multiplication is then defined by α(v ⊗ β) = v ⊗ (αβ).
At the end of lecture 7 we showed that if L is a Lagrangian subbundle of P , then L⊕JL
is a Lagrangian decomposition of P . Therefore by the previous theorem, every compatible
complex structure on E ⊕ E ∗ is isomorphic to the complexification EC ∼ = E ⊕ JE.

The above result has a “reverse analogue” for symplectic subbundles:

Proposition 2.3.5. Let (V, Ω, J) be a symplectic vector space equipped with a tame
complex structure (not necessarily compatible). A subspace W ⊂ V is symplectic if JW ⊂
W.

Proof. Let v ∈ W ∩ W Ω . Suppose JW ⊂ W . Then Jv ∈ W , so Ω(v, Jv) = 0. Thus

v = 0.

Corollary 2.3.6. Let N be an almost complex submanifold of a symplectic manifold

(M, ω, J) equipped with a tame almost complex structure. Then N is symplectic. In partic-
ular, if N is an almost complex submanifold of an almost complex manifold which admits
a compatible symplectic form, then N is symplectic.

At the end of the previous lecture, we proved that a symplectic vector bundle splits as
E ⊕ E ∗ if and only if it admits a Lagrangian subbundle. A natural follow up question is
whether or not there are symplectic vector bundles which admit half dimensional subbun-
dles but no Lagrangian subbundles. This can be answered in the affirmative using Chern
classes. We now proceed with a number of results whose proofs will be omitted, since they
are beyond the scope of the course (at least for now).

23
 The Chern classes are invariants associated to complex vector bundles. These have
corresponding Chern numbers which are numerical invariants. In the case of complex
vector bundles over oriented closed 2-real-dimensional manifolds (which we call surfaces),
the first Chern number completely captures the first Chern class. According to McDuff-
Salamon, the Chern number can be defined axiomatically as follows:

Theorem 2.3.7. There exists a unique map c1 : E → Z called the first Chern number,
which assigns an integer to every complex vector bundle E over a surface Σ and satisfies
the following axioms:

1. Two complex vector bundles over Σ are isomorphic if and only if they have the same
rank and first Chern number.

2. Whenever ϕ : Σ2 → Σ1 is a smooth map between surfaces and E → Σ1 is a complex

vector bundle,
c1 (ϕ∗ E) = deg(ϕ)c1 (E).

3. If E1 , E2 are complex vector bundles over Σ,

c1 (E1 ⊕ E2 ) = c1 (E1 ) + c1 (E2 .)

4. The first Chern number of the tangent bundle of Σ is the Euler characteristic of Σ.

By theorem 2.3.3, the first Chern number can also be defined as an invariant of sym-
plectic vector bundles over surfaces. Moreover, the axioms imply that the first Chern
number vanishes if and only if the vector bundle is trivial. Thus the first Chern number
is an obstruction to triviality. We now use the first Chern number to show that there
exist symplectic vector bundles admitting half-dimensional subbundles, but no Lagrangian
subbundles.

Proposition 2.3.8. The first Chern number of a complexification of a real vector bundle
over a surface is zero.

Every surface Σ has an orientation reversing automorphism ϕ. Let E → Σ be a complex

vector bundle. Then ϕ∗ E = E, the “conjugate bundle”. Therefore by the second axiom
above, c1 (E) = −c1 (E). Now suppose E is a complexification of a real vector bundle R,
so that J : R ⊕ R → R ⊕ R is defined by J : (e, ee) 7→ (−e e, e). Then the conjugate complex
structure is J : (e, ee) 7→ (e
e, −e). Clearly E and E are isomorphic complex vector bundles,
by the bundle map (e, ee) 7→ (e e, e). Therefore by the first axiom above, c1 (E) = c1 (E). It
follows that c1 (E) = 0.

Example. Let T be the tangent bundle of S2 , and RC the complexification of the trivial
real line bundle over S2 . Then T admits a symplectic structure (e.g. the normalised area
form of S2 ) and RC admits a compatible symplectic structure. Therefore T ⊕ RC → S2

24
is a symplectic vector bundle, and it clearly admits a half-dimensional subbundle. On
the other hand, if T ⊕ RC → S2 admits a Lagrangian subbundle L, then it splits as
L ⊕ L∗ by theorem 2.2.4. But then any compatible complex structure is isomorphic to
the complexification LC → S2 , and so its first Chern number must vanish. But we have
c1 (T ⊕ RC ) = c1 (T ) + c1 (RC ) = 2, so T ⊕ RC does not admit Lagrangian subbundles.

2.3.2 Integrability of almost complex manifolds

Definition 2.3.9. An almost complex manifold is a smooth manifold whose tangent bundle
is equipped with a complex structure. A complex manifold is a smooth manifold covered
by charts mapping into open subsets of Cn , whose transition maps are biholomorphic.

Proposition 2.3.10. Every complex manifold can be realised as an almost complex man-
ifold. The converse is not true in general.

To see this, suppose M is a complex manifold with local coordinates z j = xj + iy j .

Then one can define a complex structure by
∂ ∂ ∂ ∂
J j
= j, J j
= − j.
∂x ∂y ∂y ∂x
Conversely, there exist almost complex manifolds in real-dimension 2 which are not induced
from holomorphic structures. We now investigate when an almost complex manifold does
in fact arise from a holomorphic structure.
By the Frobenius integrability theorem, a subbundle of the tangent bundle of the man-
ifold arises as the union of tangent spaces of a foliation if and only if the subbundle D
satisfies [X, Y ] ∈ D for each X, Y ∈ D. This gives rise to the following definitions.

Definition 2.3.11. Let M be a manifold and D a subbundle of T M . Then D is called a

distribution, and is said to be integrable if [D, D] ⊂ D.

Suppose (M, J) is an almost complex manifold. Consider the complexification T MC

of T M - note that this is 4n-real-dimensional. Complexifying J gives a C linear map
JC : T MC → T MC .
Define T1,0 M := {X − iJC X : X ∈ T M }, T0,1 M := {X + iJC X : X ∈ T M }. Then for
any V ∈ T1,0 M , we have

JC V = JC (X − iJX) = JX + iX = i(X − iJX).

Therefore T1,0 M is the subbundle of T MC consisting of eigenvectors of JC which correspond

to the eigenvalue i. Similarly, T0,1 M is the subbundle of T MC consisting of eigenvectors
corresponding to eigenvalue −i. Moreover, it is easy to show that T MC = T1,0 M ⊕ T0,1 M .
These are called the holomorphic and antiholomorphic tangent bundles of (M, J). The
celebrated Newlander–Nirenberg theorem is can be stated as:

25
Theorem 2.3.12. Let (M, J) be an almost complex manifold. Then J is induced from a
holomorphic structure if and only if the distribution T 0,1 M ⊂ T MC is integrable.

This is highly non-trivial and not in the scope of this course. (In fact, the previous
couple of pages in these notes are just me binge-typing notes about things that weren’t
actually covered in the lecture.)

Theorem 2.3.13. An almost complex manifold is induced from a holomorphic structure

if and only if the Nijenhuis tensor N J (X, Y ) = [X, Y ] + J[JX, Y ] + J[X, JY ] − [JX, JY ]
vanishes.

Proof. We take for granted that N J is truly a tensor, and prove that T 0,1 M is integrable if
and only if N J vanishes. Suppose A + iJA, B + iJB are sections of T 0,1 M . Then we have

[A + iJA, B + iJB] = [A, B] − [JA, JB] + i([JA, B] + [A, JB]).

Therefore T 0,1 M is integrable if and only if J([A, B]−[JA, JB]) = [JA, B]+[A, JB], which
is equivalent to the vanishing of the Nijenhuis tensor.

Remark. In symplectic geometry, almost complex manifolds (as opposed to complex man-
ifolds) are sufficient 95% of the time.

2.4 Lecture 9
2.4.1 Stein Manifolds
The complex manifold n
P C i has ai canonical symplectic structure, namely through the identi-
fication with (R , i dx ∧ dy ). In particular, complex submanifolds of Cn have induced
2n

symplectic structures by corollary 2.3.6.

Definition 2.4.1. A complex manifold X is Stein if there exists a proper holomorphic

embedding of X into Cn , for some n. Recall that a map is proper if the preimage of a
compact map is compact.

Example. Subsets of Cn that arise as zero sets of polynomials (i.e. varieties) are Stein.
However, not all Stein manifolds are affine.

Remark. In general there may be many proper embeddings of X into Cn , each resulting
in a different symplectic form. However, these are not so different as shown in the following
result:

Theorem 2.4.2 (Gromov-Eliashberg). Any two symplectic structures on X arising from

proper embeddings into Cn are symplectomorphic.

26
 What are some notable properties of Stein manifolds? Firstly they necessarily have
infinite volume. A more interesting property is that they satisfy the weak Lefschetz theorem.
That is, if X 2n is Stein, then Hk (X, Z) = 0 for all k > n.
In fact, it was Shown by Eliashberg that Stein manifolds have a purely symplectic
characterisation: all Stein manifolds can be obtained by Weinstein handle attachments.
He called such manifolds Weinstein manifolds and constructed a functor

Wein : Stein → Weinstein.

2.4.2 Kähler manifolds

We showed earlier that every symplectic manifold admits a compatible complex structure.
A particularly well behaved case is when we can equip a symplectic manifold with an
integrable compatible complex structure.
Definition 2.4.3. Let (M, ω, J) be a symplectic manifold equipped with a complex struc-
ture. M is a Kähler manifold if J is integrable.
Example. Cn is Kähler when equipped with the standard structures. Any almost complex
submanifold of a Kähler manifold is Kähler. In particular, Stein manifolds are Kähler.
Proposition 2.4.4. Let (M, g, J) be a Riemannian manifold equipped with a compatible
complex structure. Let ω(·, ·) = g(J·, ·). Then ∇J = 0 if and only if (M, ω, J) is Kähler.
Proof. Suppose ∇J = 0. Since ω is non-degenerate and skew-symmetric, it suffices to show
that dω = 0, and that the Nijenhuis tensor vanishes. First note that

∇(Jv) = (∇J)(v) + J∇v = J∇v.

In other words, J commutes with ∇. It follows that

dω(u, v) = dg(Ju, v) = g(∇(Ju), v) + g(Ju, ∇v)

= g(J∇u, v) + g(Ju, ∇v) = ω(∇u, v) + ω(u, ∇v).

Therefore, given a third vector field w, we have

wω(u, v) = ω(∇w u, v) + ω(u, ∇w v).

But now by the formula for the exterior derivative on two forms,

dω(u, v, w) = uω(v, w) + vω(w, v) + wω(u, v) − ω([u, v], w) − ω([v, w], u) − ω([w, u], v)
= uω(v, w) + vω(w, v) + wω(u, v) − ω(∇u v, w) + ω(∇v u, w)
− ω(∇v w, u) + ω(∇w v, u) − ω(∇w u, v) + ω(∇u w, v)
= 0.

27
Next we show that the Nijenhuis tensor N J vanishes. Again by expressing each [ϕ, ψ] as
∇ϕ ψ − ∇ψ ϕ, we immediately have
N J = J(∇X J)Y − J(∇Y J)X + (∇JY J)X − (∇JX J)Y.
Therefore if ∇J = 0, N J vanishes.
Conversely, suppose (M, ω) is symplectic, and that J is integrable. To show that
∇J = 0, one can work in local coordinates. It is intuitively clear since a holomorphic
structure is locally unchanging.

Another example of a Kähler manifold is (CPn , ω), where ω is the Fubini Study metric.
The remainder of the lecture is dedicated to constructing the Fubini Study metric.
Definition 2.4.5. The complex projective space CPn is the quotient of S2n+1 ⊂ Cn+1
obtained by identifying all points of S2n+1 which lie on the same complex line. Equivalently,
CPn is the quotient of S2n+1 by the diagonal group action of the multiplicative group U (1).
(Here U (1) ∼
= S1 is the circle group, {eiθ : θ ∈ [0, 2π)}.)
If π denotes the quotient map, then π : S2n+1 → CPn is a circle bundle over CPn . In
fact, this is exactly the Hopf fibration! Also note that S2n+1 inherits a metric from Cn+1
(namely the round metric g ◦ ). Let the distribution D on S2n+1 be defined by
Dx = {v ∈ Tx S2n+1 : v is orthogonal to the fibres of π}.
For each x ∈ S2n+1 ,
dπx |Dx : Dx → T[x] CPn
is now a surjection. We can define a metric h on T CPn by h(u, v) = g ◦ (e u, ve) for any lifts
u
e and ve by dπ. To see that this is well defined, we note that the circle metric is invariant
under the U (1) action. h is called the Fubini-Study metric on CPn . Moreover, let J denote
the ambient complex structure on Cn+1 . Then each Dx is closed under J, and J|Dx is
invariant under the U (1) action. Therefore J|Dx descends to a complex structure on CPn .
One can show that J is covariantly constant on CPn to conclude that CPn is a Kähler
manifold with the Fubini-Study metric and induced complex structure.
Corollary 2.4.6. Let A be a complex projective variety. Then A is a Kähler manifold.
More generally, all complex submanifolds of CPn are Kähler.

2.5 Lecture 10
2.5.1 K3 surfaces
In the previous lecture we introduced Kähler manifolds, and remarked observed that all
complex submanifolds of Cn and CPn are Kähler. One might then ask - are there Kähler
manifolds that are not projective? It turns out there are many such examples, and some
arise as K3 surfaces.

28
Definition 2.5.1. A K3 surface X is a compact connected complex surface with trivial
canonical bundle and irregularity zero. Here the canonical bundle refers to the 2nd exterior
power (in general it is the nth exterior power of a complex n-manifold). The irregularity
is the hodge number h0,1 , or equivalently, the dimension of the sheaf cohomology group
H 1 (X, OX ).

Unfortunately understanding this definition is beyond the scope of the course. How-
ever, I attempt to write some exposition to understand the hodge number. Some notable
properties are as follows:

Proposition 2.5.2. Some properties of K3 surfaces:

• K3 surfaces exist.

• Every K3 surface is a Kähler manifold.

• Any two K3 surfaces are diffeomorphic.

• By Shing-Tung Yau’s solution to the Calabi conjecture, every K3 surface is a Calabi-

Yau manifold. (i.e. the Kähler metric is Ricci flat.)

• A K3 surface is projective if and only if it is algebraic. That is, it can be embedded

in CPn if and only if it arises as a variety.

Since every K3 surface is diffeomorphic, we might ask - how much choice do we have in
the complex structure? It turns out the space of K3 surfaces is 20 dimensional. Moreover,
the space of algebraic K3 surfaces can be shown to be 19 dimensional. It follows that there
are K3 surfaces which cannot be holomorphically embedded in CPn .

Example. An example of a K3 surface is the Fermat quartic, defined to be the zero set of
x4 + y 4 + z 4 + w4 . This can be embedded in CP3 .

2.5.2 Dolbeault cohomology

For a long time it was believed that every symplectic manifold was Kähler. In this section
we prove that there exist non-Kähler symplectic manifolds, by constructing the Kodaira-
Thurston manifold. However, to appreciate the construction, we must first develop some
Dolbeault theory.
Recall from lecture 8 the holomorphic and anti-holomorphic tangent bundles, T1,0 M
and T0,1 M , of an almost complex manifold (M, J), defined by

T1,0 M = {X − iJX : X ∈ T M }
T0,1 M = {X + iJX : X ∈ T M }.

29
Moreover, we had T MC = T1,0 M ⊕ T0,1 M . An easy way to see this is to consider the maps
1
π1,0 : T M → T1,0 M, π1,0 : v 7→ (v − iJv)
2
1
π0,1 : T M → T0,1 M, π0,1 : v 7→ (v + iJv).
2
Then π1,0 and π0,1 are (real) vector bundle isomorphisms, which satisfy π1,0 ◦ J = iπ1,0 ,
and π0,1 ◦ J = −iπ0,1 . This extends to a complex vector bundle isomorphism
(π1,0 , π0,1 ) : T MC → T1,0 M ⊕ T0,1 M
by extending π1,0 and π0,1 to projections on T MC . Similarly, we decompose the cotangent
bundle, and higher exterior powers: define Λ1,0 M and Λ0,1 M by
Λ1,0 M := {ξ ∈ Λ1 MC : ξ(v) = 0 for all v ∈ T0,1 M }
Λ0,1 M := {ξ ∈ Λ1 MC : ξ(v) = 0 for all v ∈ T1,0 M }.
It can be shown that Λ1,0 M = {ξ − iξ ◦ J : ξ ∈ Λ1 M }, and Λ0,1 M = {ξ + iξ ◦ J : ξ ∈ Λ1 M }.
This gives rise to maps π 1,0 and π 0,1 as above, such that
(π 1,0 , π 0,1 ) : Λ1 MC → Λ1,0 M ⊕ Λ0,1 M
is an isomorphism.
We then define Λk,0 M and Λ0,k M to the kth exterior powers of Λ1,0 M and Λ0,1 M
respectively. Finally we define Λk,l M to be ΛL
k,0 M ⊗ Λ0,l M .

For vector spaces, we have Λ (U ⊕ V ) = ki=0 Λi U ⊗ Λk−i V . Therefore

k

k
M k
M
Λk MC = Λk (Λ1,0 M ⊕ Λ0,1 M ) = Λi,0 M ⊕ Λ0,k−i M = Λi,k−i M.
i=0 i=0

Proposition 2.5.3. Let (M, J) be an almost complex manifold. Then J is induced from
a complex structure if and only if
d(Γ(Λp,q M )) ⊂ Γ(Λp+1,q M ⊕ Λp,q+1 M ).
Proof. We use the characterisation that J is induced from a complex structure if and only
if T0,1 M is integrable. We first prove that T0,1 M is integrable if and only if d(Γ(Λ1,0 M )) ⊂
Γ(Λ2,0 M ⊕ Λ1,1 M ). In other words, we must show that T0,1 M is integrable if and only if
the Λ0,2 M component of dξ vanishes, for any section ξ on Λ1,0 M . By definition, the Λ0,2 M
component of dξ vanishes if and only if dξ(u, v) = 0 for all u, v ∈ Γ(T0,1 M ). But since
dξ(u, v) = uξ(v) − vξ(u) − ξ([u, v]) = −ξ([u, v])
for u, v sections of T0,1 M , the Λ0,2 M component of dξ vanishes if and only if T0,1 M is
integrable, as required.
This now extends to the fact that T0,1 M is integrable if and only if d(Γ(Λp,q M )) ⊂
Γ(Λp+1,q M ⊕ Λp,q+1 M ), by applying the product rule.

30
Definition 2.5.4. On a complex manifold, the holomorphic and antiholomorphic differen-
tial operators ∂ : Λp,q M → Λp+1,q M and ∂ : Λp,q M → Λp,q+1 M are defined by projecting
the image of the exterior derivative d.
By the above proposition, it is clear that d = ∂ + ∂ on complex manifolds. This also
2
gives that ∂ 2 , ∂ , and ∂∂ + ∂∂ all vanish. Hence for every k, we have a chain complex
∂ ∂ ∂ ∂
Λk,0 M Λk,1 M Λk,2 M Λk,3 M ···
which we call the Dolbeault cocomplex. Along with it, we have Dolbeault cohomology groups

k,l ker ∂ : Λk,l → Λk,l+1

HDol (M ) = .
im ∂ : Λk,l−1 → Λk,l
The Dolbeault cohomology is a useful tool in Kähler geometry.
Theorem 2.5.5. On a compact Kähler manifold (M, ω, J), the Dolbeault cohomology
groups satisfy
(M ; C) ∼ k,l
M
m
HdR = HDol (M ).
k+l=m

H k,l
In particular, ∼
= H l,k , and each H k,l is finite dimensional. This is known as the Hodge
decomposition.
Let bm denote the Betti numbers of the de Rham cohomology groups. We write hk,l
k,l
to denote the dimensions of the Dolbeault cohomology groups HDol (M ). These are called
the Hodge numbers.
Remark. In the definition of a K3 surface, we required that h0,1 = 0. In other words, we
require that every closed antiholomorphic 1-form arises as ∂f for some smooth function f .
Corollary 2.5.6. All odd Betti numbers of a Kähler manifold are even.
This is an immediate consequence of the Hodge decomposition, since
X m
X
2m+1 k,l
b = h =2 hk,2m+1−k .
k+l=2m+1 k=0

A useful diagram is the so-called Hodge diamond :

..
.

h2,0 h1,1 h0,2

h1,0 h0,1

h0,0

31
Example. Earlier in the lecture we introduced K3 surfaces. One can show that all K3
surfaces have the following Hodge diamond:
1

0 0

1 20 1

0 0

Example. The Kodaira-Thurston manifold is an example of a closed symplectic manifold

which is not Kähler. It is constructed to be the quotient of (R4 , ω) (where ω is the standard
symplectic form) by the relations
(x1 , . . . , x4 ) ∼ (x1 + 1, x2 , x3 + x4 , x4 )
(x1 , . . . , x4 ) ∼ (x1 , x2 + 1, x3 , x4 )
(x1 , . . . , x4 ) ∼ (x1 , x2 , x3 + 1, x4 )
(x1 , . . . , x4 ) ∼ (x1 , x2 , x3 , x4 + 1).
Therefore it is something like a 4-torus with a Dehn twist. One can visualise the Kodaira-
Thurston manifold as a torus-bundle over the torus, where the base space is restricted to
the x1 and x2 coordinates. Then a loop in the x1 direction corresponds to a Dehn twist
upstairs, while the x2 direction has no monodromy.
One can show that b1 = 3 for the Kodaira-Thurston manifold. Since b1 is odd, it cannot
be a Kähler manifold.

2.5.3 Construction of symplectic manifolds

The fact that the above “counter example” was required to show that not all symplectic
manifolds are Kähler highlights an underlying problem in symplectic geometry: the con-
struction problem. There is currently no method of constructing a “generic symplectic
manifold”.
Two operations that exist are the symplectic blow up and symplectic sum. The sym-
plectic blow up topologically corresponds to the usual blow up, in which, essentially, a
sphere of a neighbourhood is replaced with a projective plane. One can show that a sym-
plectic structure is induced.
The symplectic sum is a connected sum which inherits the symplectic structure. Sup-
pose and M1 , M2 are symplectic 2n + 2 manifolds, and V a symplectic 2n-manifold, em-
bedded in M1 and M2 via ι1 , ι2 . Moreover, suppose the Euler classes of the normal bundles

32
are opposite: e(NM1 V ) = −e(NM2 V ). Then M1 and M2 are “topologically gluable”, and
it was proven by Gompf that there is a canonical isotopy class of symplectic structures
on the connected sum (M1 , V )#(M2 , V ). By using symplectic sums, Gompf proved the
following theorem:

Theorem 2.5.7 (Gompf). Arbitrary finitely presented groups arise as fundamental groups
of closed symplectic 4-manifolds.

2.6 Lecture 11
2.6.1 Symplectic reduction
A couple of lectures ago, we constructed the Fubini-Study metric on CPn . This is in fact
a special case of a more general procedure, which we call symplectic reduction.

Example (Symplectic reduction of symplectic vector spaces). Symplectic reduction in the

linear setting is easy. Let (V, Ω) be a symplectic vector space, and W ⊂ V any subspace.
Then W/W ∩ W Ω is naturally symplectic. To see this, note that the kernel of the quotient
map π : W → W ∩ W Ω is exactly W ∩ W Ω , so one can define ΩW by ΩW (u, v) = Ω(u0 , v 0 )
for any lifts u0 and v 0 by π. This is non-degenerate since all non-degeneracy is precisely
removed by the quotient.
Usually symplectic reduction is applied to coisotropic subspaces, in which the reduction
is W/W Ω (since W ∩ W Ω = W Ω ).

Example (Symplectic reduction of symplectic vector bundles). This case is still easy, but
requires an additional subtle assumption. If X → M is a symplectic vector bundle, and
Y ⊂ X a submanifold, it is not generally true that Y ∩ Y Ω has constant rank. Therefore it
is not generally a vector bundle, so neither is the quotient Y /Y ∩ Y Ω . However, by adding
this as an additional premise, symplectic reduction extends to symplectic vector bundles.

The case of symplectic manifolds is the most involved, since specifying a subbundle of
the tangent bundle (i.e. a distribution) does not immediately guarantee the existence of a
submanifold whose tangent bundle is the prescribed subbundle.

Proposition 2.6.1. Suppose (M, ω) is a symplectic manifold, and X ⊂ M a submanifold.

Suppose T X ∩ T X ω has constant rank (and hence defines a distribution in T X). Then
T X ∩ T X ω is integrable.

Proof. Let u, v be vector fields in the distribution D = T X ∩ T X ω . Let ωX = ω|X . If

ωX ([u, v], ·) = 0, then by definition [u, v] ∈ D, which will prove that D is integrable. Hence
it will be shown that ωX ([u, v], ·) = 0.

33
 Suppose η is a vector field on X. The explicit formula for the exterior derivative of a
2-form gives

dωX (u, v, η) = uωX (v, η) + vω(η, u) + ηω(u, v)

− ωX ([u, v], η) − ωX ([v, η], u) − ωX ([η, u], v).

The first three terms and last two terms immediately vanish since u, v ∈ T X ω , which leaves

0 = dωX (u, v, η) = ωX ([u, v], η)

where the left side vanishes since symplectic forms are closed. Therefore D is integrable.

By the Frobenius integrability theorem, it follows that T X ∩ T X ω is the tangent space

of some foliation F in X.

Example. If X is coisotropic, this process gives the characteristic foliation. One can also
observe that the leaves are isotropic. In particular, if X has codimension 1, it is automat-
ically coisotropic, and T X ∩ T X ω is one dimensional. This is automatically integrable.

The condition that T X ∩ T X ω has constant rank is not yet enough to define symplectic
reduction for manifolds. This requires one more premise:

Proposition 2.6.2 (Symplectic reduction for symplectic manifolds). Let (M, ω), X, and
F be as above. Suppose there is a smooth manifold B (the leaf space) such that each
leaf of F is given by the fibre of a surjective submersion π : X → B. Suppose moreover
that the fibres are connected. Then B has a canonical symplectic structure ωr such that
π ∗ ωr = ωX . (B, ωr ) is the symplectic reduction of X.

Proof. By the “linear case”, for each x ∈ X, Tx X/Tx X ∩ Tx X ω has a linear symplectic
structure. Let x, x0 ∈ X be lifts of b ∈ B. By construction the solid arrows in the diagram
below are isomorphisms (by recalling that the tangent space at x of a leaf in the foliation is
Tx X ∩ Tx X ω ). Therefore, if the dashed line preserves the symplectic structure, a canonical
non-degenerate skew symmetric form is induced on Tb B.
Tx X/Tx X ∩ Tx X ω Tx0 X/Tx0 X ∩ Tx0 X ω

Tb B
To prove that the dashed line preserves the symplectic structures on the quotients, we begin
with a lemma: suppose u is a vector field on X, tangent to T X ∩ T X ω . Then Lu ωX = 0.
(That is, the flow of u preserves ωX .) This follows from the Cartan magic formula:

Lu ωX = ιu dωX + d(ιu ωX ).

34
The first term on the right vanishes by closedness of ω, and the second term vanishes by
the choice of u.
We now return to the original claim. Suppose x, x0 are lifts of b which lie in the same
submersion chart U . One can construct a vector field supported in U which flows x to x0
in time 1. By the previous lemma, ωX is preserved. More generally, given any lifts x and
x0 , there is a path between them (by connectivity). Cover this path in submersion charts
and choose a finite subcover. The result now follows inductively from the single-cover case.
This shows that the dashed line does in fact preserve ωX , so a non-degenerate skew
form ωr is induced on B. Finally to see that it is closed (and hence a symplectic form),
choose any section s of π. Then

dωr = d(s∗ ωX ) = s∗ (dωX ) = 0.

Example. Recall the construction of the Fubini-Study metric on CPn from two lectures
ago. Riemannian and holomorphic structures were used to determine that it was Kähler,
but symplectic reduction provides an easy method for working in the symplectic setting.

35
Chapter 3

Physics

3.1 Lecture 12
3.1.1 Solutions to PDEs
A heuristic from linear algebra is as follows: given k equations and n unknowns, if k < n,
a solution is guaranteed to exist, while if k > n, if the equations are genuinely distinct,
we cannot find solutions. Applying this to PDEs, we have the following intuition: given
n functions (in some number of variables) and k partial differential equations in these
functions, if k ≤ n, we can solve the system and vice versa.
Unfortunately the heuristic is wrong. consider the differential equation dα = β, where
α ∈ Ω2 (R4 ), β ∈ Ω3 (R4 ). Expressing this in index notation, there are 6 functions and 4
equations. However, existence of solutions is classified into the following cases:

• If dβ 6= 0, there are no solutions.

• If dβ = 0, there are infinitely many solutions.

Another example is to consider a symplectic manifold (M, ω), and let ϕ : M → M be

a smooth map. Then ϕ∗ ω = ω can be expressed as a system of differential equations, with
O(n) functions and O(n2 ) equations. The heuristic suggests there are no solutions, but
usually there do exist solutions.

3.1.2 Hamiltonian vector fields

In the previous lecture, the construction of symplectic reduction for manifolds relied on
a certain vector flow preserving the symplectic structure. Such vector fields are also of
importance outside of this construction.

Definition 3.1.1. A vector field v ∈ Γ(T M ) is symplectic if Lv ω = 0.

36
 Again by the Cartan magic formula, this is equivalent to the condition that ιv ω is a
closed 1-form. In fact, by non-degeneracy of ω, every 1-form arises as ιu ω for some u.
Hence the Cartan magic formula establishes the following one-to-one correspondence.

{ Symplectic vector fields. } ←→ { Closed 1-forms. }

Definition 3.1.2. A vector field v is Hamiltonian if ιv ω is exact. (Every Hamiltonian
vector field is automatically Symplectic.)
Hamiltonian vector fields can be realised as energy preserving vector fields by the
following construction:
Proposition 3.1.3. Let H : M → R be a smooth map (the “Hamiltonian”). Then the
vector field XH defined by
ω(XH , ·) = dH
is a Hamiltonian vector field.
Since ω is non-degenerate this truly defines a vector field. Moreover, Xf is tautologically
Hamiltonian. By this construction, two functions give rise to the same Hamiltonian vector
field if and only if they differ by a locally constant function.
Proposition 3.1.4. Let H : M → R. The flow of XH preserves H. In particular, the
level sets of H (i.e. energy levels) are preserved.
Proof.
XH (H) = dH(XH ) = ω(XH , XH ) = 0.

Suppose c is a regular value of H, i.e. a value such that dHx is surjective (equivalently,
non-zero) for each x in the preimage of c. By a dimensionality argument, H −1 (c) has
codimension 1, so it is a coisotropic submanifold, and T H −1 (c) ∩ T H −1 (c)ω is integrable.
Therefore H −1 (c) has a characteristic foliation as discussed in the previous lecture. This
is called the characteristic line field and depends on the geometry of H −1 (c) but not on H
itself.
To see this, choose v to be orthogonal to H −1 (c). Then
0 = dH(v) = ω(XH , v),
so v is ω-orthogonal to XH , and the integral curves of XH follow the leaves of the charac-
teristic foliation.
It is well known that symplectic geometry supposedly models Hamiltonian mechanics.
Now that Hamiltonian vector fields have been introduced, it is only natural that Hamilton’s
equations are derived. Consider (R2n , dpi ∧ dq i ), H : R2n → R. Then
∂H i ∂H i
ιXH ω = dp + i dq .
∂pi ∂q

37
 ∂ i ∂ ∂H
Let XH = f i ∂p i i
i + g ∂q i with a view to solving for f and g . This gives fi = ,g
∂q i i
= − ∂H
∂pi
,
so any integral curve (p(t), q(t)) of XH satisfies
∂H
q̇i (t) = − (p(t), q(t))
∂pi
∂H
ṗi (t) = (p(t), q(t)).
∂q i
These are the famous Hamilton equations! (up to a difference in sign convention.)
In particular, suppose we choose H = p2 /2m + V (q). Then the above equations become
pi
q̇i (t) = −
m
∂V
ṗi (t) = i (q(t)).
∂q
∂V
Thus ∂q i
(q(t)) = −mq̈i (t). Or, more concisely, F = ma.

Proposition 3.1.5. Let V ⊂ R2n be compact, and H : R2n → R the Hamiltonian.

Consider the time evolution of V by the flow φt of XH . Then

vol(φt (V )) = vol(V )

for all t.

Proof. Since XH preserves ω, in particular it preserves ω n .

3.1.3 Poisson bracket

Definition 3.1.6. Let f, g ∈ C ∞ (M ). The Poisson bracket {f, g} of f and g is the smooth
function defined by {f, g} = ω(Xg , Xf ).

Proposition 3.1.7. Let u, v be symplectic vector fields. Then ω([u, v], ·) = d(ω(v, u)).
I.e. [u, v] is the Hamiltonian vector field of ω(v, u).

Proof. On one hand, the Leibniz rule gives

Lu (ω(v, ·)) = (Lu ω)(v, ·) + ω(Lu v, ·) = ω([u, v], ·).

On the other hand, the Cartan magic formula gives

Lu (ω(v, ·)) = ιu (d(ω(v, ·))) + d(ω(v, u)) = d(ω(v, u)).

Proposition 3.1.8. Given any two smooth functions f, g, X{f,g} = [Xf , Xg ].

38
Proof. Set u = Xf , v = Xg . Then by the previous proposition,

ω([Xf , Xg ], ·) = d(ω(Xg , Xf )) = d{f, g}.

By definition of Hamiltonian vector fields, it follows that X{f,g} = [Xf , Xg ].

Proposition 3.1.9. The Poisson bracket satisfies the Jacobi identity.

Proof. By the previous proposition,

{f, {g, h}} = −ω(Xf , X{g,h} ) = −ω(Xf , [Xg , Xh ]).

Another way to write this is

{f, {g, h}} = ω(X{g,h} , Xf ) = d{g, h}(Xf ) = Xf {g, h} = Xf ω(Xh , Xg )

Combing these two expressions,

2({f, {g, h}} + {g, {h, f }} + {h, {f, g}}) = dω(Xf , Xh , Xg ) = 0.

In fact, bilinearity and skew symmetry are immediate, so the Poisson bracket is a Lie
bracket.

3.2 Lecture 13
The fruits of the previous lecture can be summarised in a single exact sequence of Lie
algebras:

ι SympVect(M )
0 H 0 (M ) C ∞ (M ) SympVect(M ) 0
HamVect(M )
d
π
{closed 1-forms} H 1 (M )

Exactness is immediate: H 0 (M ) corresponds to locally constant functions, so ι is simply

an inclusion map, implying exactness at H 0 (M ). Moreover, since H 0 (M ) corresponds to
locally constant functions, these are exactly the kernel of the exterior derivative, giving
exactness at C ∞ (M ). Exactness at SympVect(M ) is immediate since the image of d is the
space of exact 1-forms, which is precisely quotiented out by π. Finally exactness at H 1 (M )
is the most obvious of all, since quotient maps are surjective.
The less immediate statement is that this is an exact sequence of Lie algebras. H 0 (M )
and C ∞ (M ) are equipped with the Poisson bracket, SympVect(M ) with the Lie bracket of

39
vector fields, and H 1 (M ) the trivial Lie bracket. Since ι is an inclusion, it is immediately
a Lie algebra homomorphism. The map C ∞ (M ) → SympVect(M ) sends f 7→ Xf . By a
proposition at the end of the previous lecture, {f, g} 7→ X{f,g} = [Xf , Xg ], hence C ∞ (M ) →
SympVect(M ) is also a Lie algebra homomorphism. Finally the map SympVect(M ) →
SympVect(M )/HamVect(M ) is a Lie algebra homomorphism, since it was shown that
[u, v] is Hamiltonian whenever u and v are symplectic. This means [u, v] is mapped to
zero, which is exactly the result of the trivial bracket applied to the images of u and v.

3.2.1 Various isotopies

Given two maps, a homotopy, loosely speaking, is a continuous deformation of one map to
the other. This continuous deformation does not need to preserve any structure. On the
other hand, an isotopy between two maps must preserve additional structure assigned to
the two maps. For example, a homotopy H of simple closed curves is just a continuous
deformation from one curve to the other, so at an arbitrary time t, Ht is generally just a
closed curve. However, if H is an isotopy of simple closed curves, then Ht must also be
simple for all t.
Since it can be ambiguous to call all isotopies an isotopies irrespective of context, more
precise terms are introduced during this lecture.
Definition 3.2.1. Let Φ : M × [0, ε] → M be smooth. Denote Φ|M ×{t} by Φt . Φ is a
diffeotopy if Φ0 = id, and Φt is a diffeomorphism for all t.
There is a correspondence between diffeotopies and time dependent vector fields: sup-
pose Φ : M × I → M is a diffeotopy, where I is a closed interval. This induces a diffeomor-
phism Φ e : M × I → M × I, defined by Φ(x,e t) = (Φt (x), t). (Thus it is a diffeomorphism
preserving constant time slices.) It follows that each (y, t) ∈ M × I belongs to a unique
“curve” Φ({x}
e × I). This can be formalised: for each x ∈ M , let γx : I → M × I be defined
by γx (t) = Φ({x}
e × I). The tangent vectors of these curves are given by the pushforward of
the unit vector field ∂/∂t in I. Hence the collection of all γx gives a non-vanishing vector
field
XΦ : M × I → T (M × I)
where each (x, t) is sent to the pushforward of the unit vector field ∂/∂t by γx . But then
projecting back down by M × I → I recovers the unit vector field, so there is a map
V : M × I → T M such that
XΦ (y, t) = (V (y, t), ∂/∂t) ∈ Ty M × R = Ty,t (M × I).
V is the time dependent vector field corresponding to Φ. More explicitly, one can write
∂Φ −1
V (y, t) = (Φ (y), t).
∂t t
Conversely, given certain restrictions on time dependent vector fields, there exist dif-
feotopies giving rise to them.

40
Theorem 3.2.2. Let V be a time dependent vector field V : M × I → T M , with bounded
velocity. (i.e. M has a complete metric, and there is a constant K such that |V | is bounded
by K.) Then V generates a unique diffeotopy Φ of M satisfying

∂Φ
(x, t) = V (Φ(x, t), t).
∂t
Note that the “initial condition” of this ODE is embedded in the definition of a dif-
feotopy: Φ0 = id.

Definition 3.2.3. Let Φ : M × I → M be a diffeotopy. Then

• Φ is a symplectic isotopy if Φ preserves the symplectic structure at each t. (I like to

call these symplectopies.)

• If f : M → M is a diffeomorphism which arises as the time-1 map of a [0, 1]-dependent

Hamiltonian vector field, f is a Hamiltonian diffeomorphism.

• If Φ is Hamiltonian at each t, Φ is a Hamiltonian isotopy.

The above definitions were concise, but will be made more explicit for my own benefit:
Suppose V is a time dependent vector field. If there exists a smooth family of smooth
functions Ht : M → R such that
ιVt ω = dHt ,
then the diffeotopy corresponding to V is called a Hamiltonian isotopy. A symplectomor-
phism ϕ is called a Hamiltonian diffeomorphism if there exists a Hamiltonian isotopy Φ
such that ϕ = Φ1 .

Remark. Evidently the two definitions given above for Hamiltonian diffeomorphisms differ
from one another. It is non-trivial to show that they agree, but they are in fact equivalent
definitions. Similarly the two definitions of Hamiltonian isotopies are equivalent.

Definition 3.2.4. Symp(M ) and Ham(M ) denote the groups of symplectomorphisms and
Hamiltonian diffeomorphisms respectively. Symp0 (M ) denotes the identity component of
Symp(M ).

It is immediate that Ham(M ) ⊂ Symp(M ). However, stronger results can be obtained:

we have
Ham(M ) C Symp0 (M ) C Symp(M )
as (infinite dimensional) Lie groups. On the Lie algebra level, this corresponds to

C ∞ (M )
HamVect(M ) ∼
= SympVect(M ).
locally constant functions

41
Definition 3.2.5. Let Σ be two dimensional, and u : Σ → (M, ω). The (symplectic) area
of u is defined by Z
Area(u) := u∗ ω.
Σ
Symplectic (and related) isotopies can be characterised by their effects on symplectic areas,
as seen in the following chain of propositions.
Proposition 3.2.6. ϕ : M → M is a symplectomorphism if and only if ϕ preserves
symplectic areas.
Proof. First suppose ϕ is a symplectomorphism. Then
Z Z Z
∗ ∗ ∗
(ϕ ◦ u) ω = u ϕ ω= u∗ ω.
Σ Σ Σ
Conversely, if ϕ is not a symplectomorphism, choose some x and vectors v, v 0 such that
ωx (v, v 0 ) > ωϕ(x) (dϕx v, dϕx v 0 ). Then for a sufficiently small disk D with an embedding
u : D → M , u(0) = x, we have
Z Z
∗
u ω> (ϕ ◦ u)∗ ω.
D D

Proposition 3.2.7. Let Φ : M × I → M be a diffeotopy. This is a symplectopy if and

only if symplectic areas are preserved at each t, if and only if symplectic areas traced by
contractible loops are zero for all t.
Proof. First some terminology is clarified. The “symplectic area traced by a contractible
loop at t” refers to the symplectic area of S1 × [0, t] ,→ M × I → M .
The equivalence of the first two statements is clear. it remains to prove that symplectic
areas are preserved at each t if and only if symplectic areas traced by contractible loops are
zero for all t. Choose γ : S1 → M contractible, i.e. such that there exists u : D → M with
∂u = γ. Define f : D × [0, ε] → M such that (z, t) 7→ Φt (u(z)). Then by Stokes’ theorem,
Z Z
∗
(∂f ) ω = f ∗ dω = 0.
∂(D×[0,ε]) D×[0,ε]

But the left side can be decomposed into three pieces:

Z Z Z Z
∗ ∗ ∗
(∂f ) ω = f ω− f ω+ f ∗ ω.
∂(D×[0,ε]) {ε}×D {0}×D [0,ε]×∂D

The first two terms cancel for any ε and contractible loop if and only if Φ preserves all
areas. But the first two terms cancel if and only if the last term vanishes, as required.
Proposition 3.2.8. Let Φ : M × I → M be a diffeotopy. This is a Hamiltonian isotopy if
and only if symplectic areas traced by loops are zero for all t. (Not necessarily contractible
loops.)

42
3.3 Lecture 14
3.3.1 Symplectic isotopies
In this lecture we continue from the previous lecture. We proved equivalence between
properties (1), (3), and (4).

Proposition 3.3.1. Let Φ : M × [0, ε] → M be a diffeotopy. The following are equivalent:

1. Φ is a symplectopy.

2. Φ corresponds to a [0, ε]-dependent vector field which is symplectic for every t ∈ [0, ε].

3. Φ preserves symplectic areas. More precisely, if Σ is an oriented compact manifold

with boundary, and u : Σ → M is smooth, then Area(u) = Area(Φt ◦ u) for any t.

4. Symplectic areas traced by contractible loops vanish. More precisely, if γ : S1 → M

is a loop such that there exists u : D → M with ∂u = γ, then S1 × [0, t] → M defined
by (θ, k) 7→ Φk (γ(θ)) has zero area.

It remains to prove the equivalence between (1) and (2). Let Φ be a diffeotopy, and V
the corresponding time dependent vector field. Then ∂t ∂
Φ(x, t) = Φ∗t Vt . Observe that Φ is
symplectic if and only if
∂ ∗
Φ ω = 0.
∂t t
But one can show that
∂ ∗
Φ ω = Φ∗t (LVt ω),
∂t t
which shows that Φ is symplectic if and only if Vt is symplectic for each t. Due to technical
difficulties, this lecture didn’t cover any more content.

3.4 Lecture 15
3.4.1 Hamiltonian isotopies
Let H, G : M × [0, ε] → R be smooth. They give rise to time dependent Hamiltonian vector
fields, which in turn give rise to Hamiltonian flows ϕtH , ϕtG : M × [0, ε] → M .

Lemma 3.4.1. At any t, ϕtH ◦ ϕtG : M × {t} → M is the Hamiltonian flow of the smooth
map Ht + Gt ◦ (ϕtH )−1 : M × {t} → R.

Proof. Let XH , XG be the corresponding time dependent vector fields. By the chain rule,
d t
(ϕ ◦ ϕtG )(x) = XH t0 (ϕtG0 (x)) + (ϕtH0 )∗ XGt0 (ϕtG0 (x)).
dt H t=t0

43
This shows that XH t + (ϕtH )∗ XGt generates the flow ϕtH ◦ ϕtG . But for any t, we see that
(ϕtH )∗ XGt = XGt ◦(ϕtH )−1 from the following calculation: (The ts have been dropped for
clarity.)

ω((ϕH )∗ XG , V ) = (ϕH )∗ ω(XG , (ϕ−1

H )∗ V )
= ω(XG , (ϕ−1
H )∗ V )
= dG((ϕ−1
H )∗ V )
= (ϕ−1 ∗ −1
H ) dG(V ) = d(G ◦ ϕH )(V ).

Therefore XH t + (ϕtH )∗ XGt = XH t + XGt ◦(ϕtH )−1 , so the corresponding Hamiltonian is

Ht + Gt ◦ (ϕtH )−1 .

Recall the series of equivalent conditions for an isotopy to be symplectic. Similar results
exists for Hamiltonian isotopies:

Proposition 3.4.2. Let Φ : M × [0, ε] → M be a diffeotopy. The following are equivalent:

1. Φ is a Hamiltonian isotopy.

2. The corresponding time dependent vector field is Hamiltonian.

3. For every loop γ : S1 → M , the area traced by u : S1 × [0, t] → M is zero at every

time.

The third condition holds for all loops, rather than just contractible loops, distinguish-
ing it from symplectopies. The following diagram illustrates the difference.

γ1
γ
A
γ2 B

• The isotopy taking γ to γ2 is tracing out a non-zero area. However, γ is not con-
tractible. Hence the isotopy might be a symplectopy, although it cannot be a Hamil-
tonian isotopy.

44
 • The isotopy taking γ to γ2 might be a Hamiltonian isotopy if A = B. Otherwise it
cannot be Hamiltonian.

The equivalence between (1) and (2) was not proven in the lecture, although it was remarked
that 2 ⇒ 1 is immediate while the converse is non-trivial. Condition (3) is now analysed.
Recall the gradient lemma:

Lemma 3.4.3. A 1-form is exact if and only if the line integral of the form depends only
on the endpoints of the curve, or equivalently, if the integral around any smooth closed
curve is zero.

Let Vt be the time dependent vector field corresponding to the diffeotopy Φ : M ×

[0, ε] → M . Let u : S1 × [0, ε] → M be defined by (θ, t) 7→ Φt (γ(θ)). Then du dt (θ, t) =
Vt (Φt (γ(θ))). Define α to be the 1-form αt (·) = ω(Vt , ·). Then
Z Z tZ  du du  Z tZ  du  Z tZ 
∗
u ω= ω , dθdt = αt dθdt = (u|{t}×S1 )∗ αt dt.
S1 ×[0,t] 0 S1 dt dθ 0 S1 dθ 0 S1

By definition, Vt is Hamiltonian if and only if αt is exact. But by the previous lemma, αt is

exact if and only if the term on the right vanishes for all u and t. Hence Vt is Hamiltonian
for each t if and only if the area traced by u vanishes, as required.

Remark. Let Φ : [0, ε] × M → M be a symplectopy. This determines a canonical element

of H 1 (M, R) which represents the “area”. If the flux is zero at all times, Φ is Hamiltonian.

3.4.2 Infinitesimal group actions

Definition 3.4.4. Let g be a Lie algebra. An infinitesimal action on M is a Lie algebra
homomorphism g → Γ(T M ). More specifically, an infinitesimal symplectic action is a Lie
algebra homomorphism g → SympVect(M ). Finally, a Hamiltonian action is a Lie algebra
homomorphism g → C ∞ (M ), where the latter is equipped with the Poisson bracket.

Assume the image of g → SympVect(M ) lies in HamVect(M ). Does this imply that
the action arises from a Hamiltonian action? Not necessarily! There is an obstruction the
existence of the lift below:
g
∃?

C ∞ (M ) f 7→Xf
HamVect(M )

Precisely, the obstruction is the second Lie algebra cohomology with coefficients in H 0 (M, R).
More concretely, choose a basis g1 , . . . , gn of g, and lift their images in HamVect(M ) to
C ∞ (M ). It is not immediate that [gi , gj ] agrees with {fi , fj } where gi 7→ fi .

45
3.5 Lecture 16
3.5.1 Lie theory
Definition 3.5.1. A Lie group is a smooth manifold G equipped with a compatible group
structure. That is, multiplication and inversion must be smooth.

Proposition 3.5.2. Each Lie group G has a corresponding Lie algebra

Tid G ∼
= {v ∈ Γ(T G) : (mg )∗ v = v} = {left invariant vector fields}.

Here mg is the map mg : h 7→ gh. Given any v ∈ Tid G, the corresponding left-invariant
vector field is defined by ξv (g) = (mg )∗ v. The Lie bracket on left invariant vector fields is
exactly the Lie bracket of vector fields.

Proposition 3.5.3. There is a one-to-one correspondence

{connected simply connected Lie groups} ←→ {finite dimensional Lie algebras}.

Lemma 3.5.4. Let f : G → H be a Lie group homomorphism. Then df is a Lie algebra

homomorphism.

Proof. Let v ∈ Tid G, f∗ v ∈ Tid H. These correspond to left invariant vector fields ξv on G
and ξf∗ v on H. These vector fields are f -related:

(ξf∗ v ◦ f )(g) = (mf (g) )∗ f∗ v = f∗ (mg )∗ v = (f∗ ξv )(g).

Since f -relatedness is preserved by Lie brackets, df preserves Lie brackets. Thus df is a

Lie algebra homomorphism.

3.5.2 Adjoint and coadjoint representations

A Lie group action is a smooth group action G × M → M . There is a correspondence
between Lie group actions and infinitesimal group actions:

{G action on M } ←→ {infinitesimal action g → Γ(T M )}.

There is always a map from the left to right, namely

d
a : G × M → M 7−→ a(exp(tv), x) .
dt t=0

If G is connected and simply connected, and there is a basis of g consisting of complete

vector fields, given any infinitesimal action, there is a Lie group action inducing it.

46
Definition 3.5.5. Let Φg denote the inner automorphism of g for each g ∈ G. By an
earlier lemma, dΦg is a Lie algebra homomorphism. In fact, it is an automorphism. The
adjoint representation is
Ad : G → Aut(g)
defined by g 7→ dΦg .
The adjoint representation induces infinitesimal actions: consider Tg to be a manifold.
Then there is a canonical action ad : g → Γ(Tg ) defined by v 7→ [v, ·]. This is induced from
Ad in the sense that ad = d Ad, where Γ(Tg ) is identified with the Lie algebra corresponding
to the Lie group Aut(g). Thus given elements x, y ∈ g,
adx y = [x, y].
There is a very evident parallel with the Lie derivative, in which
LX Y = [X, Y ]
for all vector fields X and Y on a manifold.
Definition 3.5.6. The coadjoint representation is K : G → Aut(g∗ ) defined by
g 7→ Ad(g −1 )∗ .
The coadjoint representation is of interest since its orbits admit symplectic structures
as we now explore.
Definition 3.5.7. Let M be a closed manifold. A Lie algebra structure on C ∞ (M ) is a
Poisson structure if it satisfies the following Leibniz rule:
[f, gh] = g[f, h] + h[f, g].
For example, C ∞ (M ) equipped with the Poisson bracket is a Poisson structure.
Proposition 3.5.8. A Poisson structure is equivalent to a bivector field π ∈ Γ(Λ2 M )
satisfying [π, π] = 0 (where the brackets denote the Schouten bracket, a generalisation of
the Lie bracket).
Let π be an alternating bi-vector field. This corresponds to a map
e : T ∗ M → T M,
π
and its image defines a “distribution” corresponding to a singular foliation by the Frobenius
integrability theorem. The leaves of this foliation are symplectic.
Example. Consider the manifold g∗ . Each g ∈ G defines an orbit in g∗ via the coadjoint
representation. These leaves are naturally symplectic.
More explicitly, at each a ∈ g∗ , Ta g∗ is isomorphic to g. Thus one can define π ∈ Γ(Λ2 g)
by πa (g, g 0 ) = a([g, g 0 ]) for g, g 0 ∈ g. This defines a symplectic structure on each coadjoint
orbit.

47
3.6 Lecture 17
3.6.1 Coadjoint action examples
In the previous lecture, we observed that orbits of the coadjoint action of G on g∗ have
natural symplectic structures. Are the orbits manifolds?

Proposition 3.6.1. Suppose G acts on a smooth manifold M . Its orbits are images of
injective immersions.

Proof. Let p ∈ M . Then Stab(p) := {g ∈ G : gp = p} ⊂ G is a closed subset. The map

ι : G/ Stab(p) ,→ M, [g] 7→ gp

is an injective smooth map whose image is the orbit of p. By the quotient manifold theorem,
the domain is a manifold. ι has constant rank so it is an immersion. If G is compact, the
orbits are submanifolds.

Theorem 3.6.2 (Kirilov-Kostant-Souriau). Every homogeneous symplectic manifold of a

lie group G is, up to a possible covering, a coadjoint orbit of a central extension of G.
(Homogeneous symplectic means it admits a transitive G action by symplectomorphisms.)

Example. Let G = SU(2). Then g = su(2). Identify g with

n  ai z
 o
∗
g = : a ∈ R, z ∈ C .
−z −ai

Consider the G action on g by conjugation. This corresponds to a rotation in R3 , so the

orbits are spheres.

Example. Let G = SL(2, R), so that g = sl(2, R). sl(2, R) can be identified with real
matrices with trace zero. There are six types of conjugacy classes (i.e. orbits of the
coadjoint action), as follows:
           
0 0 0 0 0 1 λ 0 0 −λ 0 λ
, , , , ,
0 0 1 0 0 0 0 −λ λ 0 −λ 0

More precisely, the above is a complete list of representatives of distinct conjugacy classes,
where λ > 0 is arbitrary. Identifying (2, R) with R3 , the orbits are the submanifolds in the
following figure:

48
  
0 λ
−λ 0

 
0 1
0 0

   
0 0 λ 0
0 0 0 −λ

 
0 0
1 0

 
0 −λ
λ 0

3.6.2 Moment and comoment maps

Recall that Hamiltonian actions, introduced at the end of lecture 15, were Lie algebra
homomorphisms g → C ∞ (M ), where the latter is equipped with the Poisson bracket.

Definition 3.6.3. A symplectic Lie group action G × M → M is a Hamiltonian Lie group

action if the corresponding infinitesimal action g → SympVect(M ) lifts to a Hamiltonian
action g → C ∞ (M ). The lift g → C ∞ (M ) is called the comoment map.

Proposition 3.6.4. The comoment map α is G-equivariant.

Proof. First this statement is unpacked. Recall that G acts on g by the adjoint action. G
also has a natural action on C ∞ (M ) by precomposition by the left multiplication map. We
must show that for each g ∈ G and v ∈ g,

α(Ad(g)v) = α(v) ◦ mg = m∗g α(v) = α(mg ∗ v).

Observe that if α(v) generates the flow ϕt , then α(Ad(g)v) generates the the flow mg ◦ ϕt ◦
m−1
g . On the other hand, in lecture 15 it was shown that the Hamiltonian vector field of
H ◦ ϕ−1 is ϕ∗ XH . Therefore α(v) ◦ mg has the Hamiltonian vector field mg ∗ Xα(v) . The
result follows.

There is also a moment map.

49
Definition 3.6.5. The map µ : M → g∗ defined by

µ : x 7→ (g 7→ α(g)(x))

is the moment map. This is G-equivariant by the equivariance of the comoment map.

Remark. Hamiltonian actions can be defined via the moment map.

Example. Suppose O ⊂ g∗ is a coadjoint orbit. Then it is a manifold, and the moment

map O → g∗ is simply the inclusion map.

3.7 Lecture 18
3.7.1 Moment map examples
Example. Consider the sphere S2 ⊂ R3 , with the standard area form as its symplectic
structure. Consider rotation along the z-axis. This is Hamiltonian since all areas are
preserved (so it is symplectic), and all loops are contractible. The standard area form, in
coordinates, is given by
ω = dzdθ.
Hence the generator of the rotation (S1 action) is ∂/∂θ. The corresponding Lie algebra is
R. The moment map µ : S2 → R is exactly projection onto the z coordinate.
This example gives rise to Archimedes’ theorem: the horizontal projection of a sphere
onto a cylinder is area-preserving. Of course, this is obvious if one knows that an area form
is given by dzdθ. This is sometimes used as a map projection.

Example. SO(3) acts on S2 with a Hamiltonian action. This is in fact a coadjoint orbit
of SO(3), so by the general theory, the moment map is the inclusion µ : S2 ,→ g∗ ∼
= R3 .

Remark. Moment and comoment maps are not unique. For example, in the z-axis rotation
of S2 , the map R → C ∞ (M ) given by 1 7→ z + 1 is a comoment map. This gives a
corresponding distinct moment map.

Remark. 0 ∈ g∗ is always a coadjoint orbit.

Proposition 3.7.1. The moment map is unique up to a constant of integration. More

precisely, if µ1 , µ2 are moment maps of a G-action, then µ1 − µ2 ∈ [g, g]0 ∼= H 1 (g, R).
Therefore if g is semisimple, the moment map is unique. On the other and, if g is abelian,
for any c ∈ g∗ , µ + c is a moment map.

Example. R3 acts on R3 by translation. This incudes a Hamiltonian group action on

T ∗ R3 . The corresponding moment map is T ∗ R3 → R3 , which is given by projection onto
the p-coordinate. (Exercise: flesh out details.)

50
Example. SO(3) acts on R3 by rotations. This incudes a Hamiltonian group action on
T ∗ R3 . The corresponding moment map is T ∗ R3 → R3 , which is given by (p, q) 7→ p × q.
The former corresponds to linear momentum, while this example is angular momentum.
(Exercise: flesh out details.) See recent book by Peter Woit.

3.7.2 Symplectic reduction revisited

Proposition 3.7.2. Let G × M → M be a Hamiltonian action with moment map µ :
M → g∗ . Consider
dµp : Tp M → Tµ(p) g∗ ∼
= g∗ .
Then

ker dµp = (Tp Orb(p))ω

im dµp = Ann(LieAlg(Stab(p))) ⊂ g∗ .

Proof. Let α : g → Γ(T M ) be the corresponding infinitesimal action. Let p ∈ M . Then

ωp (α(X)p , v) = dµp (v)(X)

for any X ∈ g, v ∈ Tp M by definition. Thus w ∈ ker dµp if and only if it lies in the
symplectic orthogonal of all α(X)p . This is exactly the symplectic orthogonal of Tp Orb(p).
The image is similar.

This looks rather simple but it has an important consequence. A number of lectures ago,
symplectic reduction was introduced in a general setting. The usual setting of symplectic
reduction is as follows:

Theorem 3.7.3 (Marsden-Weinstein-Meyer). Let G y M be a Hamiltonian group action,

and µ : M → g∗ a corresponding moment map. Assume G is compact and acts freely on
µ−1 (0). Then dµp is surjective for all p ∈ µ−1 (0), so 0 is a regular value, thus µ−1 (0)
is a submanifold. In fact, µ−1 (0) is coisotropic. It has a characteristic foliation given
by G-orbits. Therefore µ−1 (0)/G admits a symplectic structure. This is the symplectic
reduction of M by G, denoted M//G.

51
Chapter 4

Local and global invariants

4.1 Lecture 19
4.1.1 Darboux’s theorem
Theorem 4.1.1. Let (M, ω) be a connected symplectic manifold. For any p, q ∈ M , there
is a Hamiltonian isotopy Φ : M × I → M such that Φ1 (p) = q.
Intuitively this theorem says that any two points in a symplectic manifold locally look
the same. This theorem is used to prove Darboux’s theorem.
Proof. Choose a smoothly embedded path γ : [0, 1] → M such that γ(0) = p, γ(1) = q, and
let vt = γ 0 (t) ∈ Tγ(t) M . Next we wish to find a [0, 1] dependent Hamiltonian vector field
which is supported near γ, which restricts to vt . How do we do this?
Consider a neighbourhood N of γ(t). Choose a diffeotopy Ψ : M × I → M such that Ψ
is the identity map outside of N , and Ψt (γ(0)) = γ(t). Fix an embedding E : Bε (0) ,→ M
which sends 0 to γ(0), and consider Ψt ◦ E. At each time t, solve the following in Bε (0)
for a linear Hamiltonian:
XHt (γ(t)) = vt .
Cut them off appropriately (i.e. multiply by smooth cutoff functions). This gives a time
dependent Hamiltonian vector field. This corresponds to the desired Hamiltonian isotopy.

Theorem 4.1.2 (Darboux’s theorem · Normal form theorem). Let (M, ω) be symplectic,
and x ∈ M . Then there exists a coordinate system p1 , . . . , pn , q1 , . . . , qn such that ω =
dpi ∧ dq i in the domain of the coordinates.
Proof. Let y 6= x. Suppose there exists a symplectic structure ω 0 on M such that ω 0 = ω
near x, and ω 0 = dp ∧ dq near y in some coordinate system. Then by the previous theorem,
ω 0 is of the form dp ∧ dq near x, but this is exactly ω near x.
Therefore we consider the following picture:

52
 y

dp ∧ dq ? ω

We wish to somehow interpolate and obtain a symplectic form on M .

Remark. In general, adding symplectic forms does not give a symplectic form. If ρ1 +ρ2 =
1 is a partition of unity, ρ1 ω1 + ρ2 ω2 is not closed in general.
To combat this issue, we wish to patch together exact two forms (rather than closed
two forms). Given dλ1 , dλ2 , the two-form d(ρ1 λ1 + ρ2 λ2 ) is clearly closed. However, it is
difficult to ensure non-degeneracy: we must have bounds on ρ1 , ρ2 , dρ1 , and dρ2 . Thus we
must find a one form θ compactly supported near y such that

• ω + dθ = dp ∧ dq in a neighbourhood of y,

• dθ is small.

This is possible since ω − dp ∧ dq|y = 0.

A more local (and arguably better) argument uses the Moser trick which we explore in
the following lecture.

4.2 Lecture 20
Here we present another proof of Darboux’s theorem using the Moser trick.

Proof. Set up: (M, ω) is symplectic, x ∈ M , with a neighbourhood U1 around x with p, q

coordinates such that ωx = dp ∧ dq. Consider tω + (1 − t)dp ∧ dq. This is a symplectic form
for all t ∈ [0, 1] for some neighbourhood U2 ⊂ U1 of x. We wish to find a time dependent
vector field Vt : [0, 1] → Γ(T U2 ) such that:

53
 1. Vt (x) = 0 for all t ∈ [0, 1]. By general ODE theory, there then exists a neighbourhood
U3 ⊂ U2 such that the flow of Vt leaves U3 contained in U2 .

2. If the flow is denoted by ϕt , then ϕ∗t ωt = ω for all t.

d ∗
The second condition is equivalent to dt ϕt ωt = 0. But

d ∗ d
ϕt ωt = ϕ∗t LVt ωt + ϕ∗t ( ωt ).
dt dt
d
(This follows from fixing each term and differentiating the other term.) Moreover, dt ωt =
−LVt ωt = ω − dp ∧ dq. On the other hand, LVt ωt = d(ιVt ωt ) by the Cartan magic formula.
Now suppose ω − dp ∧ dq is closed (as we are still trying to construct a vector field). Then
dθ = d(ιVt ωt ). What if we define θ = ιVt (ωt )? This will uniquely define Vt .
Thus it remains to find a θ such that dθ = dp ∧ dq − ω. We simply need θ(x) = 0. This
can be achieved by choosing any θ and modifying by a constant, or by using integration as
in the proof of the Poincaré lemma.

Loosely speaking, the Moser trick is the process of reducing a problem to the “ap-
parently more difficult” problem of finding a vector field, or one form etc, in the above
fashion.

4.2.1 More theorems proved with Moser’s trick

Proposition 4.2.1. Let (M, ω) be closed. Suppose ωt is a [0, 1] dependent family of
symplectic forms, with ω0 = ω, and [ω] = [ωt ] ∈ HdR 2 (M ) for every t. Then there exists
∗
a diffeotopy ϕt such that ϕt ωt = ω. In particular, (M, ω) is symplectomorphic to (M, ωt )
for each t.

Proof. Note that every vector field is complete. We use the Moser argument: i.e. attempt
to construct ϕt as the flow of Vt as in the previous proof:
d ∗ d
0= ϕt ωt ⇐= d(ιVt ωt ) = − ωt .
dt dt
Note that unlike the previous proof, the term on the right hand side is not constant.
d d
However, by the premise, [ dt ωt ] = dt [ωt ] = 0. Thus we wish to choose ζt ∈ Ω1 (M ) such
that
d
1. dt ωt = dζt

2. ζt depends smoothly on t.

54
How do we find ζt ? Not entirely trivial: cleanest method is to choose a metric, and use
the Hodge theorem. (Using the Hodge star, define the adjoint d∗ , and use this to find ζt .)
Another method is to choose a covering of M and patch together local primitives. A third
method is to consider the bundle of all possible ζt and find a smooth section. Intuitively,
all of the methods work since there is a contractible space of choices.

Remark. The above theorem does not say that if ω0 and ω1 belong to the same cohomology
class, then there is a smooth family ωt that connects them. The straight line homotopy ωt
might be degenerate for some t.
Theorem 4.2.2. Let Σ be a closed surface, with ω0 , ω1 two area forms. Then (Σ, ω0 ) and
(Σ, ω1 ) are symplectomorphic if and only if the two areas are equal.
Proof. Key idea for the proof: For every t ∈ [0, 1], observe that ω0 and ω1 have the same
sign (as the top cohomology is 1-dimensional: area forms essentially live in (0, ∞). Then
tω0 + (1 − t)ω1 is a symplectic form for each t. Result now follows from Moser.

More results can be proven using Moser:

Theorem 4.2.3 (Extension theorem). Let P be a manifold, and N ⊂ P a closed subman-
ifold.
1. Let Ω be a skew-symmetric bilinear form on TN P := T P |N , whose restriction to
T N is a closed 2-form. Then Ω extends to a closed skew-symmetric 2-form in a
neighbourhood of N . If Ω is symplectic, the extension is symplectic.
2. Let Ω0 , Ω1 be symplectic structures on P whose restriction to TN P are equal. Then
there are neighbourhoods U, V ⊂ N and a symplectomorphism f : (U, Ω0 ) → (V, Ω1 )
such that f |N = id and df |TN P = id.
Remark. The second result is similar to Darboux’s theorem, in which we prescribed that
ω = dp ∧ dq at a point x. Here we give this prescription to an entire submanifold.

4.3 Lecture 21
4.3.1 Extension theorem
Recall that at the end of the previous lecture, we stated the extension theorem. We give
the proof here. The key is the homotopy formula from de Rham theory.
Proposition 4.3.1 (Homotopy formula). Let f, g : X → Y , and F : X × I → Y a
homotopy between them. Then f ∗ , g ∗ : Ω∗ (X) → Ω∗ (Y ) are chain homotopic. More
explicitly, define h : Ω∗ (Y ) → Ω∗ (X) by
Z 1
ω 7→ (i∗t ι∂/∂t F ∗ )ωdt
0

55
where it is the inclusion X → X × I, x 7→ (x, t). Then h is the chain homotopy.
Proof. This follows from
R 1the Cartan magic formula. Namely, one can prove that f ∗ − g ∗ =
d ∗
dh + hd by computing 0 dt Ft ω.
Corollary 4.3.2. Let E → M be a a vector bundle, with a disk-subbundle D ⊂ E, equipped
with the zero-section z. Then there is a homotopy between z ◦ π and id from D to D,
explicitly given by
((v, m), t) 7→ (tv, m).
If ω is a form on D, by the homotopy formula above,
ω − (z ◦ π)∗ ω = dhω + hdω.
Now for any α ∈ Ω∗ (D), hα Tz(M ) D
= 0, since ι∂/∂t F ∗ α z(M )×I
= 0. This is because
F ∗ α(v
ι∂/∂t 1 , . . . , vk ) = α(F∗ ∂/∂t, F∗ v1 , . . .), and F∗ ∂/∂t = 0 on z(M ) × I. Moreover,
dhα T = 0 if α = 0, so by the homotopy formula,
z(M ) D T D
z(M )

α = α − π ∗ z ∗ α = hdα + dhα = hdα.

We are now ready to prove the extension theorem:
Proof. (Part one of the extension theorem.) Given Ω on TN P , we need to extend this to a
form in a neighbourhood P of N . We can assume that P is tubular by general differential
topology.
Since P is a tubular neighbourhood of N , it is a disk subbundle of the normal bundle
of N . Choose any extension of Ω to P , and choose any connection. Observe that at each
point (x, v) in P , T(x,v) P ∼
= (TN P )x . Call this extension of Ω α. By the homotopy formula,
α − π ∗ z ∗ α = dhα − hdα.
Then α − hdα is closed by observing that the two other terms above are closed. Moreover,
dα vanishes on N , so α − hdα also extends Ω. Thus we have obtained a closed extension
of Ω to P , as required.
(Part two of the extension theorem.) Consider Ω0 and Ω1 on P , which have the same
restriction on N . By the homotopy formula, we have a primitive of their difference: Ω0 −
Ω1 = dη. Note that tΩ0 +(1−t)Ω1 is symplectic in a neighbourhood of N . Proceed exactly
as in the Darboux theorem and apply the Moser trick. That is, we must find a vector field
Vt with ιVt ωt = η.
More explicitly, we require f (representing the time-1 flow) to satisfy df T P = id. We
N
need dϕtVt : Tn P → Tn P, n ∈ N to be the identity. Check that this follows from dη = 0.
Remark. “This lecture will get more confusing as we keep going, with normal bundles of
the subbundle of the normal bundle, there will be like two more layers...”
In fact, the lecture finished prematurely so we didn’t observe the confusion remarked
above.

56
4.4 Lecture 22
4.4.1 Applications of the extension theorem
Turned up to the lecture late! Oh no!
Last week we looked at the extension theorem, which states that if N ⊂ P is a sub-
manifold, then
1. if Ω on TN P is closed on T N , then it has a closed extension to a neighbourhood of
N.

2. if Ω0 = Ω1 on TN P , then there is a neighbourhood of N in which they are symplec-

tomorphic.
Some applications of this theorem are as follows:
Example. Darboux’s theorem: Let E be the normal bundle of N ⊂ P . By the tubular
neighbourhood theorem, there is an isomorphism TN E → TN P which fixes the subbun-
dle T N . Therefore there exist open neighbourhoods U ⊃ N ⊂ E, V ⊃ N ⊂ P , and a
diffeomorphism U → V , which induces the above isomorphism.
Example. Let L ⊂ M be a Lagrangian submanifold. Then there is an isomorphism be-
tween its normal bundle and cotangent bundle. To see this, choose a Lagrangian comple-
ment subbundle to T L ⊂ TL M . There is an isomorphism TL M ∼ = T ∗ L. This is formalised
by the Weinstein neighbourhood theorem:
Theorem 4.4.1 (Weinstein neighbourhood theorem). Let L ⊂ M . There exist neighbour-
hoods U ⊃ L ⊂ T ∗ L, V ⊃ L ⊂ M , such that there is a symplectomorphism between U and
V which fixes L.
Example. Let C ⊂ M be isotropic. Here the normal bundle and (TC M, Ω) depends on
more than just C: by linear algebra, if V is a symplectic vector space, and W ⊂ V is
isotropic, then W Ω is coistropic. The normal direction is as follows:
1. Choose any complement to W in W Ω , say S.

2. S Ω is also symplectic.

3. W ⊂ S Ω is Lagrangian.

4. Choose Lagrangian complement to W in S Ω .

Now V is symplectomorphic to W ⊕W ∗ ⊕S, where each W, W ∗ , S have canonical symplectic
structures. In each case, the freedom in the choice of sympletic structure is contractible.
Now TC M ∼ = S(C) ⊕ T C ⊕ T ∗ C, where S(C) ∼ = T C Ω /T C. Thus we have a symplec-
tomorphism TC M/T C ∼ ∗
= S(C) ⊕ T C. By the extension theorem, this is a symplectomor-
phism between neighbourhoods of C. The neighbourhoods of C correspond to symplectic

57
vector bundles over C with rank dim M − 2 dim C. i.e. all neighbourhoods of isotropic
submanifolds look like S(C) ⊕ T ∗ C.
Remark. “Many years after Weinstein’s work, someone published a two page paper on
the above result which has thousands of citations but it’s basically a homework problem!”
Example. What do neighbourhoods of symplectic submanifolds look like? How about
neighbourhoods of coisotropic submanifolds?

4.4.2 Gromov non-squeezing

We established in the previous few lectures that there are no local invariants of symplectic
manifolds. How about global invariants? Gromov discovered a global invariant (other than
volume):
Theorem 4.4.2 (Gromov non-squeezing). Suppose B(r) ⊂ R2n is a ball of radius r. Let
Z(R) = D(R) × R2n−2 be a cylinder, where D(R) ⊂ R2 is the disk of radius R. Then if
B(r) embeds symplectically into Z(R), then R ≥ r.
Conversely, a slight modification can remove the obstruction all together.
Theorem 4.4.3 (Gutn, Polterovich). Let Σ be T 2 with an open disk removed. Suppose Σ
has area 1. For all r > 0, B(r) symplectically embeds in Σ × R2 .
Observe that if T 2 \ D is replaced with S2 \ D, we are in the realm of Gromov non-
squeezing. This shows a dichotomy between full rigidity and zero rigidity.
Remark. Differential topology is very soft, while Riemannian geometry is very rigid.
Symplectic geometry is in between, sometimes it is soft and sometimes it’s rigid.
Remark. For any R, there are symplectic embeddings B n (R) ⊂ Rn with arbitrarily large
percentage (volumewise) of B n (R) in Z(1). These are called Katak embeddings.
In the following lectures, a sketch of Gromov non-squeezing will be given. First some
definitions are required:
Definition 4.4.4. Let (Σ, j) be a Riemann surface, and (M, J) an almost-complex mani-
fold. A map u : Σ → M such that

du ◦ j = J ◦ du

is called a (j, J)-holomorphic curve. (Or just J-holomorphic.)

If one chooses holomorphic coordinates z = s + it on Σ, then the above condition can
be re-written as
∂u ∂u
=J◦ .
∂t ∂s

58
4.5 Lecture 23
4.5.1 Gromov non-squeezing: proof idea (in only one lecture?!)
Theorem 4.5.1 (Gromov, 85). Let B(r) ⊂ R2n be an open ball of radius r > 0. Let
Z(R) := D2 (R) × R2n−2 ⊂ R2n be the cylinder of radius R > 0. Then there is a symplectic
embedding ϕ : B(r) ,→ Z(R) if and only if r ≤ R.

Remark. Intuitively, any time symplectic structures appear to have rigidity that con-
tradicts Riemannian rigidity, it can be thought of as a J-holomorphic curve adding an
obstruction.

Proof. Assume there is an embedding ϕ : B(r) ,→ Z(R). Let ε > 0. Then ϕ(B(r − ε)) ⊂
D2 × [−A, A]2n−2 . There is a symplectic embedding D2 (R) ,→ S 2 (πR2 + ε). This gives a
symplectic embedding

e : B(r − ε) ,→ S 2 (πR2 + ε) × T 2n−2 .

ϕ

The latter space is equipped with two form σ ⊕ τ . The actual theorem we prove is the
following:
Theorem 4.5.2. Let (M, τ ) be symplectic with dimension 2n − 2. If there is exists a
symplectic embedding ϕ : B(r) ,→ S 2 (a) × M , with S 2 (a) × M equipped with σ ⊗ τ , then
πr2 ≤ a.
non-squeezing is now a corollary: If π(r − ε)2 ≤ πR2 + ε for all ε > 0, then r ≤ R.
Recall the definition of a pseudo-holomorphic, or J-holomorphic curve: u : (Σ, j) →
(X, J) is J-holomorphic, if du ◦ j = J ◦ du.
The above theorem is proven using the following two lemmas:
Lemma 4.5.3. (Main lemma) Suppose π2 (M ) = 0. There exists an ω-compatible almost
complex structure J on S 2 × M such that ϕ∗ J = i on B(r), and a J-holomorphic curve
u : CP1 → S 2 × M such that

1. [u] = [S 2 × {pt}] ∈ H2 (S 2 × M ).

2. ϕ(0) ∈ im(u).

Lemma 4.5.4. (Monotonicity lemma) (Minimal surface theory) Suppose u e : (Σ, j) →

(B(r), i) is non-constant, proper, pseudo-holomorphic, with 0 ∈ im(e u). Then the area
e is greater than or equal to πr2 . Visually this corresponds to the interesting result
of u
that the way to slice a sphere with a surface that passes through the origin in a way that
minimise area is to use a plane. Any perturbation will have slightly more area coming from
additional wobbliness.

59
 It follows that, if we assume the above two lemmas, then

πr2 ≤ Area(e
u) ≤ Area(u) ≤ a,

where the first inequality comes from the monotonicity lemma and the second comes from
the symplectic embedding. The last equality is because
Z
u∗ ω = ω([S 2 × pt]) = σ([u]) = a.

Remark. Note that the previous two non-main lemmas are slightly false, but they contain
the right ideas.
The idea of the proof of the main lemma is as follows: Let

J = {ω-compatible almost complex structure on S 2 × M },

fJ = {u : CP1 → S 2 × M : uJ-holomorphic, u(∞) =

suppose it is W k,2 . For J ∈ J , let M
2
ϕ(0), [u] = [S × pt]}. We establish a correspondence

fJ = ∂ −1 ∂J [
M J (0) ⊂ BJ ←→ EJ = W k−1 ΓhomC (T CP1 , u∗ T (S 2 × M )).
u

Here BJ is the Banach manifold consisting of maps in W k,p rather than smooth maps,
allowing us to use methods from analysis. ∂ J = J ◦ du ◦ j + du is the antiholomorphic
differential. Thus it vanishes if and only if u is J-holomorphic. We wish to study the
linearisation
Du ∂ J : Tu BJ → Γu∗ T (S 2 × M ) → EJu .
This is a Cauchy Riemann operator, allowing us to use the Riemann Roch theorem!
−1
Implicit function theorem: If Du ∂ J is surjective for all u ∈ ∂ J (0), then M
fJ is smooth
with dimension 4. (The dimension comes from the Riemann Roch theorem: nχ(CP1 ) +
c1 ([u]) − 2n = 2n + 4 − 2n = 4.
Observe that Aut(CP1 , ∞) has dimension 4, since the automorphisms of CP1 are Mobius
transformations which have dimension 6 (mapping any three points to any three points),
but now we are forced to fix a point (any two points go to any two points). Thus

fJ / Aut(CP1 , ∞)
MJ = M

is smooth with dimension zero. Since [u] = [S 2 × pt], MJ is compact.

A proof outline for the remainder of the proof is as follows:

1. J0 = iCP1 ×iT 2n−2 is Fredholm regular, so the implicit function theorem applies. Thus
|MJ0 | = 1.

60
 2. There exists J1 Fredholm regular such that ϕ∗ J1 = i. (Existence of J1 is very lucky,
due to the choice of homology class [S 2 × pt].)

3. For a generic path Jt , t ∈ [0, 1] from J0 to J1 , M = ∪t∈[0,1] {t} × MJt has the structure
of a compact 1-dimensional manifold, with ∂M the disjoint union of MJ0 and MJ1 .
This gives a cobordism from MJ0 to MJ1 . But if MJ0 consists of a single point, the
existence of the cobordism ensures that MJ1 is non-empty (since their cardinalities
must agree mod 2).

4. Since MJ1 is non-empty, and J1 satisfies the conditions of the main lemma, the result
follows.

4.6 Lecture 24
We began with a summary of the main points in the proof of Gromov non-squeezing,
namely the two lemmas mentioned in the proof from the previous lecture.

Remark. Symplectic geometry has two eras... one is pre-Gromov, one is post-Gromov.

A point that needs justifying from the above proof: why is there a J-holomorphic curve
passing through every point in S 2 (a) × T 2 (A) for generic J?

Remark. Riemannian geometry is a huge field... 50% of it is only known by Gromov.

4.6.1 Lagrangians in linear symplectic reductions

Recall symplectic reduction: for symplectic vector spaces V , given Q ⊂ V a subspace, the
symplectic reduction of Q is Q/Q ∩ QΩ . In a more specific setting, given a coisotropic
Q ⊂ V , the symplectic reduction is simply Q/QΩ .

Lemma 4.6.1. Suppose L ⊂ V is Lagrangian, and W ⊂ V coisotropic. Then the image

of L ∩ W → W/W Ω is a Lagrangian subspace of W/W Ω .

Proof. It must be shown that LW (the image of L ∩ W ) is isotropic and coisotropic. It is

immediately isotropic. For the reverse direction, choose any [v] ∈ LW Ω , i.e. any v with
ω(v, l) = 0 for all l ∈ L ∩ W , so v ∈ (L ∩ W )Ω .
Observe that (L ∩ W )Ω = LΩ + W Ω = L + W Ω . Choose any v ∈ (L ∩ W )Ω . This
decomposes as v1 + v2 in L + W Ω . Since W Ω ⊂ W , it must be the case that v2 ∈ W . Thus
v1 = v − v2 lies in W . This gives v ∈ (L ∩ W ) + W Ω , so v ∈ LW as required.

61
 Suppose L and W are transverse; L + W = V . Suppose V has dimension 2n, so that L
has dimension n. Then W has dimension n + k, and L ∩ W has dimension k, with W/W Ω
having dimension (n + k) − (n − k) = 2k. The image of L ∩ W → W/W Ω is isomorphic to
L ∩ W/L ∩ W Ω and has dimension k, so L ∩ W Ω = 0. Taking these into account, Umut
suspects that there should be a simpler proof of the above lemma in this special case.
In summary, the following has been established: suppose V is a symplectic vector space,
W ⊂ V is coisotropic, and L ⊂ V is Lagrangian.

• Let LW denote the image of L ∩ W → W/W Ω . Then LW ⊂ W/W Ω is a Lagrangian

subspace.

• If in addition L is transverse to W , then LW ∼

= L ∩ W.

4.7 Lecture 25
4.7.1 Lagrangians in general symplectic reductions
We continue the discussion from the previous lecture, in the vector bundle setting manifold
settings.
Symplectic vector bundle case: Let E → M be a symplectic vector bundle, and Q ⊂ E
a coisotropic subbundle. This gives a symplectic vector bundle Q/QΩ → M . Let L ⊂ E
be a Lagrangian subbundle. We make the assumption that L ∩ Q has constant rank, so
that it defines a subbundle. Then the image L ∩ Q → Q/QΩ is also a subbundle. To see
this, observe that at any x in the base space, the image of the quotient is isomorphic to
Lx ∩ Qx /(Lx ∩ QΩ x ), but

Lx ∩ QΩ Ω Ω Ω
x = Lx ∩ Qx = (Lx + Qx ) ,

and the fibres of the last object forms a bundle if Lx ∩ Qx does. Therefore the image of
L ∩ Q is a vector bundle; the reduced Lagrangian subbundle.
Symplectic manifold case: Suppose M is a symplectic manifold, C ⊂ M coisotropic,
and L ⊂ M Lagrangian. We make the following assumptions:

• C is foliated by isotropic leaves.

• The leaf space X is itself a manifold, e.g. the leaves are the fibres of a submersion
C → X.

Then X has a natural symplectic structure. (See earlier lecture on integrability for details.)
To obtain stronger results, we make more assumptions:

• L intersects C “cleanly”, i.e. L∩C is a submanifold, and for all a ∈ L∩C, Ta (L∩C) =
Ta L ∩ Ta C.

62
Note that this assumption is weaker than transversality.
We wish to obtain a Lagrangian inside X, given a smooth map g : L ∩ C ,→ C → X.
This corresponds to the differential

Ta L ∩ Ta C → Ta C/(Ta C)Ω ,

where the codomain is established by the vector bundle setting. This has a Lagrangian
image: Since g has constant rank, by the constant rank theorem, the image is an immersed
Lagrangian. If L and C are transverse, then L ∩ C → X is an immersion.

Remark. The above map is not an embedding, since L can intersect a leaf more than
once.

4.7.2 Symplectic category

Definition 4.7.1. A Lagrangian correspondence, or canonical relation from (M, ωM ) →
(N, ωN ) is a Lagrangian submanifold in the twisted product M ×N
e .
∗ ω −
Recall that the twisted product is M × N equipped with the symplectic form πM M
∗ ω .
πN N

Example. Suppose ϕ : M → N is a symplectomorphism. Then the graph G(ϕ) ⊂ M N

e is
a Lagrangian correspondence.

Example. Suppose G is a Hamiltonian action on M , with moment map µ : M → g∗ .

Assume the action of G on µ−1 (0) is free. Then µ−1 (0) ,→ M × M G is a Lagrangian
correspondence. (The above map is injective, and its image lifted to M ×M
e is Lagrangian.)
−1
More explicitly, µ (0) is a coisotropic submanifold with leaves given by the orbits of G.

Example. An example of the above example: S2n−1 → Cn × CPn−1 .

Proposition 4.7.2. Lagrangian correspondences can be composed: If L1 ⊂ P1 ×P

e 2 , L2 ⊂
P2 ×P
e 3 , then there is
L1 ◦ L2 ⊂ P1 ×P
e 3.

Proof. L1 ×L2 ⊂ (P1 ×Pe 2 )×(P2 ×P

e 3 ) is Lagrangian. Consider the diagonal ∆P2 = −P2 ×P
e 2.
Then P1 × ∆P2 × P3 is a coisotropic submanifold of (P1 ×P2 ) × (P2 ×P3 ). The reduction
e e
of this coistropic submanifold is exactly P1 ×P
e 3 . By the earlier discussion of symplectic
reduction for manifolds, if the intersection of L1 × L2 with P1 × ∆P2 × P3 is clean, this
gives an immersed Lagrangian in P1 ×Pe 3.

63
4.8 Lecture 26
4.8.1 Exact Lagrangians
Definition 4.8.1. Let (T ∗ M, dλtaut ) be the canonical symplectic manifold. A Lagrangian
L ⊂ T ∗ M is called exact if λ|L is exact.

Example. • The zero section of T ∗ M is exact. In fact, this is very much the case as
λ|ZM ≡ 0.

• Suppose ϕ : T ∗ M → T ∗ M is a Hamiltonian diffeomorphism. Then ϕ(ZM ) is exact.

This follows from a flux argument: Choose a Hamiltonian isotopy and compute
integral of λ over 1-cycles on ϕ(ZM ) via Stokes’ theorem.

Nearby Lagrangian conjecture Any compact exact Lagrangian in T ∗ M is Hamiltonian-

isotopic to the zero section.

Remark. The fact that this is an open problem is an example of the general construction
problem in symplecitc geometry.

Definition 4.8.2. Let π : E → M be a submersion. (E.g. the projection map of a vector

bundle.) Let S : E → R be a function. Inside T ∗ E, there is a coisotropic submanifold

N := {(e, ξ) : ξ|ker dπe = 0}.

Then there is an inclusion

N ,→ T ∗ E → T ∗ M,
where N ⊂ T ∗ E is coisotropic with isotropic leaves given by the fibres of N → T ∗ M .
Now if the graph G(dS) in T ∗ E is transverse to N , then there is a reduced Lagrangian
submanifold ΣS := ιS : N ∩ G(dS) → T ∗ M . This is the Lagrangian generated by S.
There is an injection ιS : ΣS ,→ T ∗ E → E, allowing ΣS to be considered as a subman-
ifold of E. ΣS corresponds to “vertical critical points” of S. ιS records the horizontal part
of dS at a point of ΣS ⊂ E.

In coordinates, let E = Rn+k , M = Rn , (ξ, η) 7→ η. Then

ΣS = {(ξ, η) : ∂S/∂ξ(ξ, η) = 0},

and the inclusion map is given by ΣS ,→ T ∗ Rn ∼

= Rn × Rn , (ξ, η) 7→ (∂S/∂ξ(ξ, η), η).

Remark. This is always an immersion since Lagrangian reduction is an immersion, but it

need not be an embedding.

64
Example. S : Ra × Rx → R, (a, x) 7→ a3 /3 + (x2 − 1)a. Then
ΣS = {a2 + (x2 − 1) = 0} ⊂ R2 , ιS (a, x) 7→ (2xa, x).
Therefore the immersion sends circles at each a to a figure-8 with a double point at the
origin.
In this example, “N ” (which was seen in the definition of generating functions) is
{(a, x, pa , px ) : pa = 0}.
Lemma 4.8.3. ι∗S λtaut = d(S|ΣS ), so ιS is an exact Lagrangian immersion. This is trivial
when computing in coordinates.
Theorem 4.8.4. Every Lagrangian L ⊂ T ∗ M (for L, M compact) that is Hamiltonian-
isotopic to the zero section ZM admits a generating function. In fact, this generating
function is special; it can be chosen such that S : RN × M → R and S is quadratic at
infinity (in the RN direction).
Proof. The following will be a construction that doesn’t work immediately, but contains
inspirational ideas. Let γ : [0, 1] → T ∗ M, γ(t) = (p(t), q(t)). The action functional, where
Ht : T ∗ M × [0, 1] → R is fixed, is given by
Z Z Z 1
A(γ) = γ ∗ λtaut + Ht (γ(t))dt = p(t)q̇(t) + Ht (p(t), q(t))dt.
[0,1] [0,1] 0

Claim: consider E := {γ : [0, 1] → T M : γ(0) ∈ ZM }. Then E → M, γ 7→ π(γ(1)) is a

bundle, where A : E → R is the generating function. Then A “generates” the time 1 image
of ZM under the flow of Ht .
Exercise for the reader: If γ(t) = (p(t), q(t)) 7→ (p(t) + δp(t), q(t) + δq(t)), then A(γ) 7→
A(γ) + δA, where δA is given by
Z Z
δA = δp(t)(q̇(t) + ∂Ht /∂p(p(t), q(t))) − δq(t)(ṗ(t) − ∂Ht /∂q(p(t), q(t))) + p(1)δq(1).

In this formula, δq(1) is zero for vertical directions. That is, γ ∈ ΣS if and only if γ satisfies
Hamilton’s equations. Moreover, if ΣS → T ∗ M is given by γ 7→ (p(1), q(1)), the image is
precisely the image of ZM under the Hamiltonian flow of Ht . This is explored further in
McDuff-Salamon.

4.9 Lecture 27
In subsequent lectures, each student (as well as the lecturer!) will give a 10 minute presen-
tation on a topic from Symplectic rigidity: Lagrangian submanifolds they found interesting.
• Monday: Umut, Jordi
• Wednesday: Eric, Alec, Shintaro
• Friday: Jared, Darius, Juan

65
4.9.1 Poincaré’s last geeometric theorem (Umut’s talk)
Theorem 4.9.1. Let A = S1θ × [0, 1]x be an annulus (with area form dθdx). If ϕ : A → A
is an area-preserving diffeomorphism which “twists boundaries in opposite directions”, then
ϕ has at least two fixed points.

This theorem was motivated by Poincaré’s study of the 3-body problem. The idea was
to consider a Hamiltonian motion in 4-dimensional phase space, and to then restrict to an
energy level. Each energy level is 3-dimensional. We wish to understand the energy levels
near a given energy level.
Claim: Nearby periodic orbits are in correspondence with fixed points of the Poincaré
return map. This is an area preserving map which can, without loss of generality, be
considered on the annulus. More precisely, the linearisation of the problem lives in the
annulus.
This is captured by the following diagram:

R×I S1 × I

ϕ
e ϕ

R×I S1 × I

We have a “boundary twisting property” which can be formalised as:

ϕ(
e θe + 1, x) = ϕ(
e θ,
e x) + (1, 0).

We wish to translate Poincaré’s problem into one of fixed points of Hamiltonian diffeo-
morphisms. Assume for simplicity that near the boundary of the annulus ϕ e is a translation
in θ.
e
Yasha’s trick: consider two strips placed side by side. Consider any arc on the annulus
from one boundary component to another. On one of these strips, apply ϕ. e Glue the strips
together where two arcs meet the boundary, consider area change. This is in a video by
Yasha.
Claim: Every symplectomorphim of the torus which fixes a parallel and moves a merid-
ian by zero flux has at least three fixed points. This is solved by a Moser type argument.
Monday: Arnol’d conjecture.

4.9.2 Wavefronts (Jordi’s talk)

Consider Lagrangian immersions (rather than assuming Lagrangian submanifolds).

Definition 4.9.2. Let M be an n-manifold, and f : M → Cn . If df is injective and its

image is a Lagrangian submanifold, then f is a Lagrangian immersion.

66
 Recall: canonical form dp ∧ dq is exact; ω = dλ. Whenever f ∗ ω = 0, f ∗ λ is itself closed.

Definition 4.9.3. A Lagrangian immersion f : M → Cn is called exact if f ∗ λ is exact,

i.e. there exists F : M → R, f ∗ λ = dF .

Idea: f × F : M → Cn × R, Legendrian immersion into a contact manifold.

We are ready to consider wavefronts: These will help understand n-dimensional sub-
manifolds in Rn × Rn as n-dimensional submanifolds of Rn × R.

π = Re(f × F ) : M → Rn × R.

Definition 4.9.4. A wavefront is the image of the Lagrangian immersion

X X X
dz − yj dxj , (f × F )∗ (dz − yj dxj ) = 0, z = yj xj .
j j j

Example: Whitney immersion. f : Sn → Cn , (x, a) 7→ (1 + 2ia)x. Sends a “double tear

drop” to a “figure 8”. Somehow reversible!

4.10 Lecture 28
4.10.1 Surgery theory (Eric’s talk)
See Jared Marx-Kuo for lecture notes (was late to class).

4.10.2 Soft Lagrangian obstructions (My talk)

A big general question in symplectic geometry is the following:

Which n-manifolds L admit Lagrangian embeddings into Cn ?

By a Lagrangian embedding, we mean the image of L is an embedded Lagrangian subman-

ifold of Cn , where Cn has the standard compatible symplectic structure.

The standard symplectic structure ω of Cn is defined by

ω(zj , wj ) = <(zj )=(wj ) − =(zj )<(wj ).

Using the fact that g(−, −) = ω(−, i−) is an inner product on Cn , one can prove that a
submanifold L ⊂ Cn is Lagrangian if and only if T L ⊕ iT L = T Cn .

Definition 4.10.1. A totally real embedding of a manifold L into Cn is an embedding

such that Tx L and iTx L are transverse. (Thus every Lagrangian embedding is totally
real.) Totally real embeddings are referred to as soft Lagrangian embeddings.

67
We can now discuss soft and hard obstructions to the existence of Lagrangian embeddings
in Cn . “Soft obstructions” refer to obstructions to the existence of soft Lagrangian embed-
dings. These tend to be topological without requiring symplectic rigidity results. On the
other hand, “hard obstructions” are specifically obstructions to Lagrangian embeddings.
These truly require a consideration of the symplectic structure, e.g. Maslov classes and
symplectic area classes.

A cool observation is that S3 has no soft obstructions, even though it admits no La-
grangian embedding in C3 . In that sense, soft Lagrangian embeddings are truly softer
than Lagrangian embeddings.
Definition 4.10.2. Let f : L → Cn be an immersion of an n-real dimensional manifold L.
There is a regular homotopy invariant d(f ) defined to be the signed count of double points
of the immersion. (For n even and L oriented, there is a natural signed count. Otherwise,
blackbox.)
Theorem 4.10.3 (Whitney, classical theorem). Suppose n = 2k is even, and let f : L →
Cn be a Lagrangian immersion with normal crossings. Then

d(f ) = (−1)k+1 χ(L)/2.

(equality is mod 2 if L is non-orientable.)

Theorem 4.10.4. A closed connected n-manifold L admits totally real embeddings if and
only if it is U -parallelisable and
1. if n is even, χ(L) = 0. (Equality is mod 4 if L is not orientable.)
2. If n ≡ 1 mod 4 and L is orientable, χ
b(L) = 0.
Here χb(L) is the Kervaire semi-characteristic which essentially plays the same role as
the Euler characteristic.

I’ve introduced a definition without actually defining it. What is U -parallelisability? This
is a complex version of the requirement that a manifold has a trivial tangent bundle. An
n dimensional manifold M is parallelisable if

TM ∼
= M × Rn .
M is U-parallelisable if
T M ⊗R C ∼
= L × Cn .
As we know, most spheres have tangent bundles which are non-trivial. The only coun-
terexamples are S1 , S3 , and S7 . (E.g. S2 has a non-trivial tangent bundle by the hairy ball
theorem.) Using this fact we can attempt to determine which spheres have Lagrangian
embeddings.

68
 1. All spheres of dimension n 6= 1, 3, 7 have no soft Lagrangian embeddings (and hence
no Lagrangian embeddings).

2. S1 has a Lagrangian embedding, since the inclusion map S1 ,→ C is a Lagrangian

embedding.

3. By considering the forgetful functor from Lagrangian immersions to topological im-

mersions, one can conclude that S7 has no totally real immersions, but S3 does have
totally real immersions.

This is noteworthy, since S3 has a totally real immersion in C3 , but no Lagrangian em-
bedding into C3 . One can see this by using a hard Lagrangian obstruction. I.e. noting
that every closed Lagrangian submanifold of Cn has a non-vanishing symplectic area class,
hence a non-trivial first cohomology group. This automatically bars Sn from having a
Lagrangian embedding for all n > 1. In particular, this holds for n = 3.

4.10.3 h-principle (Alec’s talk)

Recall: If f : L → Cn is Lagrangian, then T L ⊕ iT L ∼
= T Cn . This gives an isomorphism
T L ⊗ C → Cn , so any Lagrangian immersion gives rise to a U -parallelisation.

Proposition 4.10.5. All Lagrangian immersions have U -parallelisations.

h-principle is the converse.

Theorem 4.10.6. The set of Lagrangian regular homotopy classes of Lagrangian immer-
sions L → Cn is a 1-1 correspondence with homotopy classes of U -parallelisations.

What are some nice examples?

Any n-dimensional C vector bundle with n-dimensional base....
Out of time! Next lecture angrily stormed into the lecture theatre.

69