Prerequisites from Differential Geometry

appendix to the Symplectic Geometry lecture notes

Ana Cannas da Silva∗

September 2004

1 Isotopies and Vector Fields

Let M be a manifold, and ρ : M × R → M a map, where we set ρt (p) := ρ(p, t).

Definition 1.1 The map ρ is an isotopy if each ρt : M → M is a diffeomorphism,

and ρ0 = idM .

Given an isotopy ρ, we obtain a time-dependent vector field, that is, a family

of vector fields vt , t ∈ R, which at p ∈ M satisfy

d
q = ρ−1

vt (p) = ρs (q) where t (p) ,
ds s=t

i.e.,
dρt
= vt ◦ ρt .
dt
Conversely, given a time-dependent vector field vt , if M is compact or if the vt ’s
are compactly supported, there exists an isotopy ρ satisfying the previous ordinary
differential equation.
Suppose that M is compact. Then we have a one-to-one correspondence

{isotopies of M } ←→ {time-dependent vector fields on M }

ρt , t ∈ R ←→ vt , t ∈ R

Definition 1.2 When vt = v is independent of t, the associated isotopy is called

the exponential map or the flow of v and is denoted exp tv; i.e., {exp tv : M →
M | t ∈ R} is the unique smooth family of diffeomorphisms satisfying
d
exp tv|t=0 = idM and (exp tv)(p) = v(exp tv(p)) .
dt
∗ E-mail: acannas@[Link] or acannas@[Link]

1
2 1 ISOTOPIES AND VECTOR FIELDS

Definition 1.3 The Lie derivative along v is the operator

d
Lv : Ωk (M ) −→ Ωk (M ) defined by Lv ω := (exp tv)∗ ω|t=0 .
dt
When a vector field vt is time-dependent, its flow, that is, the corresponding
isotopy ρ, still locally exists by Picard’s theorem. More precisely, in the neighborhood
of any point p and for sufficiently small time t, there is a one-parameter family of
local diffeomorphisms ρt satisfying

dρt
= vt ◦ ρt and ρ0 = id .
dt
Hence, we say that the Lie derivative along vt is

d
Lvt : Ωk (M ) −→ Ωk (M ) defined by Lvt ω := (ρt )∗ ω|t=0 .
dt

Exercise. Prove the Cartan magic formula,

Lv ω = ıv dω + dıv ω ,

and the formula

d ∗
ρ ω = ρ∗t Lvt ω , (?)
dt t
where ρ is the (local) isotopy generated by vt . A good strategy for each formula is
to follow the steps:

(a) Check the formula for 0-forms ω ∈ Ω0 (M ) = C ∞ (M ).

(b) Check that both sides commute with d.

(c) Check that both sides are derivations of the algebra (Ω∗ (M ), ∧). For instance,
check that
Lv (ω ∧ α) = (Lv ω) ∧ α + ω ∧ (Lv α) .

(d) Notice that, if U is the domain of a coordinate system, then Ω• (U) is gener-
ated as an algebra by Ω0 (U) and dΩ0 (U), i.e., every element in Ω• (U) is a
linear combination of wedge products of elements in Ω0 (U) and elements in
dΩ0 (U).

♦
We will need the following improved version of formula (?).

Proposition 1.4 For a smooth family ωt , t ∈ R, of k-forms, we have

 
d ∗ dωt
ρt ωt = ρ∗t Lvt ωt + .
dt dt
 3

Proof. If f (x, y) is a real function of two variables, by the chain rule we have
 
d d  d 
f (t, t) = f (x, t) + f (t, y) .
dt dx x=t dy y=t

Therefore,  
d ∗ d ∗  d ∗ 
ρ ωt = ρ ωt + ρ ωy
dt t dx x x=t dy t y=t
| {z
 } | {z }
dω

ρ∗
x Lvx ωt  by (?) ρ∗ y
t dy 
x=t y=t

 
dωt
= ρ∗t Lvt ωt + .
dt


2 Submanifolds
Let M and X be manifolds with dim X < dim M .

Definition 2.1 A map i : X → M is an immersion if dip : Tp X → Ti(p) M is

injective for any point p ∈ X.
An embedding is an immersion which is a homeomorphism onto its image.1
A closed embedding is a proper2 injective immersion.

Exercise. Show that a map i : X → M is a closed embedding if and only if i is an

embedding and its image i(X) is closed in M .
Hint:
• If i is injective and proper, then for any neighborhood U of p ∈ X, there is
a neighborhood V of i(p) such that f −1 (V) ⊆ U.
• On a Hausdorff space, any compact set is closed. On any topological space,
a closed subset of a compact set is compact.
• An embedding is proper if and only if its image is closed.
♦

Definition 2.2 A submanifold of M is a manifold X with a closed embedding

i : X ,→ M .3

Given a submanifold, we may regard the embedding i : X ,→ M as an inclusion,

in order to identify points and tangent vectors:

p = i(p) and Tp X = dip (Tp X) ⊂ Tp M .

1 Theimage has the topology induced by the target manifold.
2A map is proper if the preimage of any compact set is compact.
3 When X is an open subset of a manifold M , we refer to it as an open submanifold.
4 3 TUBULAR NEIGHBORHOOD THEOREM

3 Tubular Neighborhood Theorem

Let M be an n-dimensional manifold, and let X be a k-dimensional submanifold
where k < n and with inclusion map i : X ,→ M . At each x ∈ X, the tangent
space to X is viewed as a subspace of the tangent space to M via the linear inclusion
dix : Tx X ,→ Tx M , where we denote x = i(x). The quotient Nx X := Tx M/Tx X
is an (n − k)-dimensional vector space, known as the normal space to X at x.
The normal bundle of X is

N X = {(x, v) | x ∈ X , v ∈ Nx X} .

The set N X has the structure of a vector bundle over X of rank n − k under the
natural projection, hence as a manifold N X is n-dimensional.
Exercise. Let M be Rn and let X be a k-dimensional compact submanifold of Rn .
(a) Show that in this case Nx X can be identified with the usual “normal space”
to X in Rn , that is, the orthogonal complement in Rn of the tangent space
to X at x.
(b) Given ε > 0 let Uε be the set of all points in Rn which are at a distance less
than ε from X. Show that, for ε sufficiently small, every point p ∈ Uε has a
unique nearest point π(p) ∈ X.
(c) Let π : Uε → X be the map just defined for ε sufficiently small. Show that,
if p ∈ Uε , then the line segment (1 − t) · p + t · π(p), 0 ≤ t ≤ 1, joining p to
π(p) lies in Uε .
(d) Let N Xε = {(x, v) ∈ N X such that |v| < ε}. Let exp : N X → Rn be the
map (x, v) 7→ x + v, and let ν : N Xε → X be the map (x, v) 7→ x. Show
that, for ε sufficiently small, exp maps N Xε diffeomorphically onto Uε , and
show also that the following diagram commutes:

exp - Uε
N Xε
@
@
@
ν @ π
@
R
X

(e) Suppose now that the manifold X is not compact. Prove that the assertion
about exp is still true provided we replace ε by a continuous function ε : X →
R+ which tends to zero fast enough as x tends to infinity. You have thus
proved the tubular neighborhood theorem in Rn .
♦
 5

In general, the zero section of N X,

i0 : X ,→ N X , x 7→ (x, 0) ,

embeds X as a closed submanifold of N X. A neighborhood U0 of the zero section

X in N X is called convex if the intersection U0 ∩ Nx X with each fiber is convex.

Theorem 3.1 (Tubular Neighborhood Theorem) Let M be an n-dimensional

manifold, X a k-dimensional submanifold, N X the normal bundle of X in M ,
i0 : X ,→ N X the zero section, and i : X ,→ M inclusion. Then there exist
a convex neighborhood U0 of X in N X, a neighborhood U of X in M , and a
diffeomorphism ϕ : U0 → U such that

ϕ
N X ⊇ U0 - U ⊆M
'
I
@ 
@
@ commutes.
i0 @ i
@
X
Outline of the proof.

• Case of M = Rn , and X is a compact submanifold of Rn .

Theorem 3.2 (ε-Neighborhood Theorem)

Let U ε = {p ∈ Rn : |p − q| < ε for some q ∈ X} be the set of points at a
distance less than ε from X. Then, for ε sufficiently small, each p ∈ U ε has
a unique nearest point q ∈ X (i.e., a unique q ∈ X minimizing |q − x|).
π
Moreover, setting q = π(p), the map U ε → X is a (smooth) submersion with
the property that, for all p ∈ U ε , the line segment (1 − t)p + tq, 0 ≤ t ≤ 1,
is in U ε .

Here is a sketch. At any x ∈ X, the normal space Nx X may be regarded as

an (n − k)-dimensional subspace of Rn , namely the orthogonal complement
in Rn of the tangent space to X at x:

Nx X ' {v ∈ Rn : v ⊥ w , for all w ∈ Tx X} .

We define the following open neighborhood of X in N X:

N X ε = {(x, v) ∈ N X : |v| < ε} .

Let
exp : N X −→ Rn
(x, v) 7−→ x + v .
Restricted to the zero section, exp is the identity map on X.
6 3 TUBULAR NEIGHBORHOOD THEOREM

Prove that, for ε sufficiently small, exp maps N X ε diffeomorphically onto

U ε , and show also that the diagram
exp
N Xε - Uε
@
@
@ commutes.
π0 @ π
@
R
X

• Case where X is a compact submanifold of an arbitrary manifold M .

Put a riemannian metric g on M , and let d(p, q) be the riemannian distance
between p, q ∈ M . The ε-neighborhood of a compact submanifold X is

U ε = {p ∈ M | d(p, q) < ε for some q ∈ X} .

Prove the ε-neighborhood theorem in this setting: for ε small enough, the
following assertions hold.
– Any p ∈ U ε has a unique point q ∈ X with minimal d(p, q). Set q = π(p).
π
– The map U ε → X is a submersion and, for all p ∈ U ε , there is a unique
geodesic curve γ joining p to q = π(p).
– The normal space to X at x ∈ X is naturally identified with a subspace of
Tx M :

Nx X ' {v ∈ Tx M | gx (v, w) = 0 , for any w ∈ Tx X} .

p
Let N X ε = {(x, v) ∈ N X | gx (v, v) < ε}.
– Define exp : N X ε → M by exp(x, v) = γ(1), where γ : [0, 1] → M is the
geodesic with γ(0) = x and dγ ε
dt (0) = v. Then exp maps N X diffeomorphi-
ε
cally to U .
• General case.
When X is not compact, adapt the previous argument by replacing ε by an
appropriate continuous function ε : X → R+ which tends to zero fast enough
as x tends to infinity.

Restricting to the subset U 0 ⊆ N X from the tubular neighborhood theorem,
π0
we obtain a submersion U0 −→ X with all fibers π0−1 (x) convex. We can carry this
fibration to U by setting π = π0 ◦ ϕ−1 :

U0 ⊆ NX is a fibration =⇒ U ⊆M is a fibration
π0 ↓ π↓
X X

This is called the tubular neighborhood fibration.

 7

4 Homotopy Formula
Let U be a tubular neighborhood of a submanifold X in M . The restriction i∗ :
` `
HdeRham (U) → HdeRham (X) by the inclusion map is surjective. As a corollary
of the tubular neighborhood fibration, i∗ is also injective: this follows from the
homotopy-invariance of de Rham cohomology.
` `
Corollary 4.1 For any degree `, HdeRham (U) ' HdeRham (X).
At the level of forms, this means that, if ω is a closed `-form on U and i∗ ω is
exact on X, then ω is exact. We will need the following related result.
Proposition 4.2 If a closed `-form ω on U has restriction i∗ ω = 0, then ω is exact,
i.e., ω = dµ for some µ ∈ Ωd−1 (U). Moreover, we can choose µ such that µx = 0
at all x ∈ X.
'
Proof. Via ϕ : U0 −→ U, it is equivalent to work over U0 . Define for every
0 ≤ t ≤ 1 a map
ρt : U0 −→ U0
(x, v) 7−→ (x, tv) .
This is well-defined since U0 is convex. The map ρ1 is the identity, ρ0 = i0 ◦ π0 , and
each ρt fixes X, that is, ρt ◦ i0 = i0 . We hence say that the family {ρt | 0 ≤ t ≤ 1}
is a homotopy from i0 ◦ π0 to the identity fixing X. The map π0 : U0 → X is
called a retraction because π0 ◦ i0 is the identity. The submanifold X is then called
a deformation retract of U.
A (de Rham) homotopy operator between ρ0 = i0 ◦ π0 and ρ1 = id is a linear
map
Q : Ωd (U0 ) −→ Ωd−1 (U0 )
satisfying the homotopy formula
Id − (i0 ◦ π0 )∗ = dQ + Qd .
When dω = 0 and i∗0 ω = 0, the operator Q gives ω = dQω, so that we can take
µ = Qω. A concrete operator Q is given by the formula:
Z 1
Qω = ρ∗t (ıvt ω) dt ,
0

where vt , at the point q = ρt (p), is the vector tangent to the curve ρs (p) at s = t.
The proof that Q satisfies the homotopy formula is below.
In our case, for x ∈ X, ρt (x) = x (all t) is the constant curve, so vt vanishes
at all x for all t, hence µx = 0. 
To check that Q above satisfies the homotopy formula, we compute
Z 1 Z 1
Qdω + dQω = ρ∗t (ıvt dω)dt + d ρ∗t (ıvt ω)dt
0 0

Z 1
= ρ∗t (ıvt dω + dıvt ω )dt ,
0 | {z }
Lvt ω
8 5 WHITNEY EXTENSION THEOREM

where Lv denotes the Lie derivative along v, and we used the Cartan magic formula:
Lv ω = ıv dω + dıv ω. The result now follows from
d ∗
ρ ω = ρ∗t Lvt ω
dt t
and from the fundamental theorem of calculus:
Z 1
d ∗
Qdω + dQω = ρt ω dt = ρ∗1 ω − ρ∗0 ω .
0 dt

5 Whitney Extension Theorem

Theorem 5.1 (Whitney Extension Theorem) Let M be an n-dimensional man-
ifold and X a k-dimensional submanifold with k < n. Suppose that at each p ∈ X
'
we are given a linear isomorphism Lp : Tp M → Tp M such that Lp |Tp X = IdTp X
and Lp depends smoothly on p. Then there exists an embedding h : N → M
of some neighborhood N of X in M such that h|X = idX and dhp = Lp for all
p ∈ X.

The linear maps L serve as “germs” for the embedding.

Sketch of proof for the Whitney theorem.
Case M = Rn : For a compact k-dimensional submanifold X, take a neighbor-
hood of the form

U ε = {p ∈ M | distance (p, X) ≤ ε} .

For ε sufficiently small so that any p ∈ U ε has a unique nearest point in X, define
a projection π : U ε → X, p 7→ point on X closest to p. If π(p) = q, then
p = q + v for some v ∈ Nq X where Nq X = (Tq X)⊥ is the normal space at q; see
Appendix A. Let
h : U ε −→ Rn
p 7−→ q + Lq v ,
where q = π(p) and v = p − π(p) ∈ Nq X. Then hX = idX and dhp = Lp for
p ∈ X. If X is not compact, replace ε by a continuous function ε : X → R+ which
tends to zero fast enough as x tends to infinity.
General case: Choose a riemannian metric on M . Replace distance by rieman-
nian distance, replace straight lines q + tv by geodesics exp(q, v)(t) and replace
q + Lq v by the value at t = 1 of the geodesic with initial value q and initial velocity
Lq v.