Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

LECTURE 7

The next step is to show that in a connected manifold we can always deform isotopically
the identity to map a point to another one.
Lemma 59 (Homogeneity Lemma). Given X connected manifold (with or without boundary),
x and y in X \ ∂X, then there exists h : X → X diffeomorphism smoothly isotopic to the
identity such that h(x) = y.
Remark 60. Thus apart from points at the boundary, which are different, on a connected
manifold all points are essentially the same from the viewpoint of isotopy.
Proof. Given two points x and y in X \ ∂X the existence of h as in the lemma defines an
equivalence relation: the reflexivity is clear with h identity, the symmetry is clear with h−1 ,
the transitivity follows from composing h1 and h2 and composing the corresponding isotopic
homotopies to the identity. The equivalence classes partition X and, since X is connected,
it is enough to prove that each equivalence class is open to show X is one equivalence class.
To prove that the equivalence class of a given point x is open, we will prove that it contains
an open set around x. More precisely, given a local parametrisation ϕ : U → U with U open
around 0 in Rn (n = dim X) and U open around x in X, we have an open ball B(0, r) ⊂ U and
we will prove that ϕ(B(0, r)) is in the equivalence class of x. Note that we have used the fact
that x is in the interior of X and not on the boundary when exhibiting a parametrisation
by an open set of Rn and not Hn . Then given any u ∈ B(0, r) ⊂ Rn , we can construct
F : Rn → Rn smooth that is the identity on Rn \ B(0, r), maps B(0, r) into itself, and so
that F(0) = u. Consider ζ : Rn → R smooth that is positive on B(0, r) and zero elsewhere
and consider the ordinary differential equation Ḟ(t, w) = ζ(F(t, w))u with initial condition
F(0, w) = w. The Picard-Lindelöf theorem you have seen last year (aka Cauchy-Lipschitz
theorem) proves that a smooth solution exists for all t ≥ 0 since the Lipschitz constant of ζ
is bounded uniformly on Rn . This solution satisfies F(t, w) = w for any w ∈ Rn \ B(0, r)
and t ≥ 0, and when one considers the initial data w = 0 then F (t, 0) = f (t)u for some
scalar function f (t) because the right hand side of ODE is colinear to u, and f (t) satisfies
f ′ uniformly positive as long as f (t) ≤ 1 (because ζ is uniformly positive on B(0, |u|)) so
reaches 1 in some finite time t0 > 0. Then F (·) = F(t0 , ·) answers our problem.
We can now combine the two previous tools to produce the important structural result:
Theorem 61 (Degree modulo 2). Given X compact manifold without boundary, f : X → X
smooth and x and y regular values of f then #f −1 (x) = #f −1 (y) modulo 2. The common
value is the degree modulo 2 of f , denoted deg2 (f ), and it is invariant under smooth homotopy.
Example 62. The fundamental example is that the identity map on a compact manifold
without boundary has degree modulo 2 equal to 1, while any constant map has degree modulo
2 equal to 0, so these maps cannot be smoothly homotopic.
Proof. Consider x and y two regular values of f : X → X. The homogeneity lemma implies
there is h : X → X diffeomorphism isotopic to identity so that h(x) = y. Then by composition
y is also a regular value of h◦f and, since h◦f is homotopic to f , the homotopy lemma implies
#(h ◦ f )−1 (y) = #f −1 (y) modulo 2, which means #f −1 (x) = #f −1 (y) modulo 2. If g is
smoothly homotopic to f , Sard’s theorem implies there is x regular value of both f and g (since
they are dense open in X) and the homotopy lemma then implies #f −1 (x) = #g −1 (x).

15
We can now prove the main result of this chapter:

Theorem 63 (Brouwer’s fixed point theorem). Given n ∈ N∗ , any continuous map f : B̄n →
B̄n from the close unit ball to itself must has at least one fixed point.

Proof. We first prove the smooth Brouwer’s fixed point theorem, i.e. assuming that f is
smooth. We argue by contradiction: if f has no fixed point, then we can construct a smooth
so-called retraction g from B̄n to Sn−1 , i.e. g : B̄n → Sn−1 smooth so that g|Sn−1 is the
identity. For x ∈ Bn define g(x) ∈ Sn−1 as the point where the half-line starting from f (x)
and directed towards x hits the boundary Sn−1 . It is not hard to write Cartesian formula
for g in terms of f and deduce smoothness. Then let G : Sn−1 × [0, 1] → Sn−1 be defined
by G(x, t) = g(tx), then G is a smooth homotopy from the constant function g(0) to the
identity on the compact manifold without boundary Sn−1 , which is absurd by the previous
example. To deduce now the continuous Brouwer’s fixed point theorem, one can argue by
approximation. Consider f : B̄n → B̄n continuous with no fixed point, then by compactness
there is ϵ > 0 so that ∥f (x) − x∥ ≥ ϵ for all x ∈ B̄n , and the Stone-Weierstrass theorem
shows there is P : B̄n → Rn polynomial on each component so that ∥f − P ∥∞ < ϵ/2. Then
Q = P/(1 + ϵ/2) is valued in B̄n and ∥f − Q∥∞ < ϵ. But Q is smooth and cannot have a
fixed point (as Q(x) = x implies ∥f (x) − x∥ < ϵ) which contradicts previous point.

2.8 Extensions and remark


2.8.1 Intersection number
We can generalise the notion of degree modulo 2 to pre-images of submanifolds Z instead of
points (i.e. 0-manifolds). The restriction of such generalization is that we nevertheless need
to ensure that f −1 (Z) remains a finite set of points by a dimension condition.

Definition 64. Given X compact manifold without boundary, Y a manifold, Z a submanifold


of Y closed and without boundary so that dim X + dim Z = dim Y , f : X → Y a smooth map
with f ⋔ Z, we define the intersection number modulo 2 of f with Z, denoted I2 (f, Z),
as the cardinal of the finite set f −1 (Z) modulo 2. In the particular case where X is a compact
submanifold of Y without boundary and X and Z have complementary dimension in Y and
f is the canonical inclusion, we denote the intersection number modulo 2 by I2 (X, Z).

Remark 65. Check that under these assumptions, f −1 (Z) (or X ∩ Z) is indeed a finite set.

Lemma 66 (Homotopy lemma for intersection numbers). Given X, Y , Z and two maps f
and g as in the previous definition, if f and g are homotopic then I2 (f, Z) = I2 (g, Z).

Proof. This is a technical but easy extension of the homotopy lemma (exercise).

Remark 67. In fact, we can define I2 (f, Z) for a map f which is not necessarily transversal
to Z: using that transversality is generic, we can find a map g in the smooth homotopy class
of f such that g ⋔ Z and define I2 (f, Z) := I2 (g, Z).
Remark 68. In the case when X submanifold of Y and f inclusion as described in the
definition, this means that I2 (X, Z) is invariant under “deformation” of X, where deformation
means precisely an homotopy on the inclusion. In particular if I2 (X, Z) = 1, it means that
no matter how X is deformed, it will always intersect Z. For instance if Y = S1 × S1 is

16
the 2-torus, and X = S1 × {(0, 1)} and Z = {(0, 1)} × S1 then I2 (X, Z) = 1 and we cannot
smoothly pull the circles apart.
Remark 69. If dim X = dim Y /2 (so dim Y is even) we can take Z = X and define I2 (X, X)
the self-intersection number modulo 2. For instance the “essential line” (middle line) in a
Möbius strip has I2 = 1, unlike the middle line of a bracelet which has I2 = 0. This cannot
be changed by smooth deformation: the Möbius strip has some intersection property, which
is linked with the fact that a curve has to do two rounds to be back at the same point, and
with the fact it is not orientable (see later).

2.8.2 Intrinsic definition of smooth manifolds


The Möbius strip can be defined simply by considering the equivalence classes of points in the
rectangle when identifying two opposite edges after reversing one. The Klein bottle can be
defined by identifying two opposite edges of the rectangle after reversing one, and identifying
the two other opposite edges without reversal. Such examples suggest that it is possible to
define manifolds without any reference to an ambient space RN . However, one needs to be
careful with what defines an acceptable atlas, and whether two atlases are equivalent (the
class of equivalence, in the case of smooth manifolds, is called a “smooth structure”). We give
a brief idea of the definition, but this will be discussed fully in the Part III course “Differential
Geometry”.

Definition 70. Given n ∈ N∗ , an n-dimensional smooth manifold is a second countable


Hausdorff space X and a collection of parametrisation/charts, i.e. local homeomorphisms
ϕ : U → ϕ(U ) with U open set of Rn and ϕ(U ) open set of X, so that each point x ∈ X
belongs to the image ϕ(U ) of some parametrisation ϕ, and so that when two parametrisations
(ϕ, U ) and (ψ, V ) intersect then their transition map ϕ◦ψ −1 : ψ(U ∩V ) → ϕ(U ∩V ) is smooth,
and finally so that the collection is maximal with respect to the previous properties. This
collection is called a smooth atlas.

This definition allows to define smooth maps between manifolds exactly as we have done
in this chapter, using that the transition maps are smooth. Then a smooth map f : X → Y is
an immersion if Df is injective everywhere, a submersion if Df is surjective everywhere,
and an embedding if it is an immersion and an homeomorphism onto its image. Note that
the definition of the tangent space requires a bit of care, since there is no ambient space for
it to live in, and it gives rise to the so-called tangent bundle.
A natural question is whether such abstract manifolds can always be embedded in RN
for some N ∈ N∗ and therefore approached by the tools of this chapter. Whitney proved it
for N = 2n (which is optimal), but a simpler result of Whitney proves it for N = 2n + 1 and
can be found in textbooks.

17

You might also like