You are on page 1of 7

Proof of the Poincare-Hopf Theorem

Han Baek Yu(2008-11008)


December 8, 2010

Let M be a compact smooth manifold and V be a smooth vectorfield on M


with only isolated zeros. Then M has only finitely many zeros, since otherwise
the infinite sequence of zeros {zi } have a convergent subsequence on compact
space M , and hence M must possess an accumulating zero.

Definition 1 The vector fields V on M and V 0 on N correspond under f if the


derivative dfx carries V (x) into V 0 (f (x)) for each x ∈ M .

If f is a diffeomorphism, the clearly V 0 is uniquely determined by V . The


notation
V 0 = df ◦ V ◦ f −1
will be used.

Lemma 1 Any orientation preserving diffeomorphsim f of Rm is smoothly iso-


topic to the identity.

We may assume that f (0) = 0. Since the derivative at 0 can be defined as


f (tx)
df0 (x) = lim ,
t→0 x
it is natural to define an isotopy F : Rm × [0, 1] → Rm by
f (tx)
F (x, t) = 0<t≤1
t
F (x, 0) = df0 (x).

To prove F is smooth, even as t → 0, we write f in the form

f (x) = x1 g1 (x) + · · · + xm g1 (m),

where g1 , · · · , gm are suitable smooth functions and note that

F (x, t) = x1 g1 (tx) + · · · + xm g1 (tm),

1
for all values of t. This shows that f is isotopic to linear mapping df0 , which is
clearly isotopic to the identity map.


Lemma 2 Suppose that the vector field V on U corresponds to

V 0 = df ◦ V ◦ f −1

on U 0 under a diffeomorphism f : U → U 0 . Then the index of V at an isolated


zero z is equal to the index of V 0 at f (z).

We may assume that z = f (z) = 0 and that U is convex. If f preserves ori-


entation, than proceeding exactly above, we construct a one-parameter family
of embeddings ft : U → Rm with f0 = id, f1 = f , ft (0) = 0. Let Vt denote
the vector field dft ◦ V ◦ ft−1 on ft (U ), which corresponds to V on U . These
vector field are all defined and nonzero on a sufficiently small sphere centered
at 0. Hence the index of V 0 = V0 at 0 must be equal to the index of V 0 = V
at 0, since the index is an integer valued function so that the smooth mapping
which sends V 0 to V does not change integer value. This proves Lemma for
orientation preserving deffeomorphisms.

To consider diffeomorphisms wich reverse orientation it is sufficient to con-


sider the special case of a reflection ρ, since we can construct a smooth isotopy
from any orientation-reversing diffeomorphism to the reflection similarly in the
Lemma 1. Then
V 0 = ρ ◦ V ◦ ρ−1 ,
0
so the associated function V = V 0 (x)/||V 0 (x)|| on the -sphere satisfies
0
V = ρ ◦ V ◦ ρ−1 .
0
Then we see the degree of V equals the degree of V . Since the degree of V and
0
V , V 0 and V are the same, this completes the proof of Lemma 2.


Lemma 3 The index of V at a nondegenerate zero z is either 1 or −1 according


as the determinant of dVz is positive or negative.

Consider V as a diffeomorphism from som convex neighborhood U0 of z into


Rn . We may assume that z = 0. If V preserves orientation, we have observed
that VU0 can be deformed smoothly into the identity without introducing any
new zeros, in Lemma 1 and 2. Thus the index is clearly equal to 1.

If V reverses orientation, then similarly V can be deformed into a reflection,


and hence the index is −1.


2
Definition 2 The vector field V is non-degenerate at z if the linear transfor-
mation dVz is non-singular.

Lemma 4 Suppose that x is a zero of v and U is a small neighborhood of x in


M containing no other zero of v. Let D be a closed disk in U , whose interior
contains x. Then there exists a vector field v1 that equals v outside of D, with
only finitely many zeros in D, all of which are non-degerate. Any such v1 has
X X
Indx (v) = Indx (v1 ).
x∈v −1 (0) v1 (x)=0

Since we only have finitely many zeros on of V , we may assume D has only
one degenerate zero z0 . We also assume that U is contained in some chart, so
we can work in local coordinates. Pick a bump function ρ that is 1 near x and
0 outside of D, and define
V1 (z) = V (z) + ρ(z)a,
n
where a ∈ R . If a is small enough, V1 can be zero on where ρ = 1, that is,
near x. Now if −a is a regular value of V , we get that all the zeros of V1 will
be non-degenerate, and if a is also small enough, V1 will still have finitely many
zeros. Since V and V1 both restrict to the same function on ∂D, we only need
to show that the summation of index of V 0 inside D equals to the index of V
at z0 . As we did above, we define a one-parameter family of vector fields Vt
defined by
Vt (z) = V (z) + ρ(z)at 0<t≤1
V0 = V (Z).
So we can smoothly deform V to V1 , and hence the summation of index in D
agrees. This proves Lemma 4.


The above lemma tells us that we only need to consider a vector field with
non-degenerate zeros.

Corollary 1 (The Splitting Lemma). If V is a vector field with only finitely


many zeros, there exists a vector field W with only finitely many zeros, all of
which are non-degenerate, and with
X X
Indx V = Indx W.
x∈V −1 (0) x∈W −1 (0)

Let us call a non-degenerate zero z of a smooth vector field V on manifold M


a ’source’ if V , regarded as a diffeomorphism from some convex neighborhood
U0 of z into Rm is identity map, and a ’sink’ if it is a identity map with minus
sign, and a ’saddle’ if it is a reflection map. Then we can simplify our vector
field little bit more.

3
Corollary 2 If V is a vector field with only finitely many zeros, there exists a
vector field W with only finitely many zeros, all of which are either source, sink
or saddle, and with
X X
Indx V = Indx W.
x∈V −1 (0) x∈W −1 (0)

The vector field V near a non-degenerate zero z can be regarded as a deffieo-


morphism from a small neighborhood U to Rn . Then dVz either preserves or
reverses the orientation, and we have seen that such diffeomorphism can be
smoothly deformed into identity or reflection. Such smooth deformation does
not change the summation of index. Thus we can deform all the non-degenerate
zeros into either a source, or saddle. But we may leave the sinks of V . So we
get the desired result.


By Rado’s theorem, a triangulation exists for compact manifold M . Since


M is compact metric, the open covering {Uα } of M consists of coordinate neigh-
borhood Uα has a Lebesgue number  > 0 such that any subset V of M with
diameter less than  has a open cover Uα such that V ∈ Uα . We can reduce
the maximum diameter of triangle on M as much as we want by subdivision, so
that we can always work with coordinate neighborhood for any given triangle.
Since a compact manifold has an embedding into some Euclidean space, we shall
consider only compact manifold in Rn .

Definition 3 Let M be a compact smooth manifold in Rn , and V be a smooth


vector field on M . Let {Ti } be a triangulation of M . Then assign 1 to each
vertex, -1 to each edge, 1 to each face, and in general assign (−1)k to each
k-dimensional face. We call such assignment of numbers the numbering on
{Ti }. By shifting of the numbering of {Ti } we mean the re-assignment of all
numbers such that each numbers of the numbering of {Ti } is assigned to each
n-dimensional face.

Lemma 5 Let M be a compact smooth manifold in Rn , and V be a non-


vanishing smooth vector field on M . Then there is a triangulation {Ti } of
M with a shifting such that the summation of all numbers assigned to each
n-dimensional face is zero.

Let M be a compact smooth manifold in Rn , and V be a non-vanishing


smooth vector field on M . Let {Ti } be a given triangulation on M . Note that
each triangle Ti is a diffeomorphic image of Euclidean n-dimensional tetrahe-
dron T 0 . Note that if we subdivide T 0 we get a subdivision of T by the image of
subdivided T 0 under the diffeomorphism from T 0 to T . Then notice that each
face of T is smooth since it is a diffeomorphic image of corresponding face of
T 0 . For instance, an 1-dimensional edge of T is a smooth curve on M . Hence
if we subdivide enough so that each sub-triangle in T 0 becomes so small, and

4
it should be almost identical to a Euclidean n-dimensional tetrahedron. Then
note that since V is non-vanishing, it looks like constant vector field in a very
small neighborhood on M . Thus by subdividing enough, we can make each sub-
triangle of T 0 be almost congruent to an Euclidean n-dimensional triangle and
at the same time the vector field V looks like a constant vector field on each
sub-triangle in T . Let us just denote this finer triangulation by {Ti }. Then
since we have only finitely many triangles on M , there is some way to slightly
move or rotate each triangle so that no face of any triangle Ti is parallel to
the local vector field V |Ti , which is almost constant. For 2-dimensional version,
this is to say that no edge of any triangle is parallel to the constant vector field
near it. Therefore for any triangle T and its face σ, V (x) at all x ∈ σ points
identically either inward or outward of T . In other words, all V (x) on σ points
at the interior of some triangle Ti . Now we give a shifting of the numbering of
triangulation {Ti } using this fact : we assign the number (−1)k assigned to a
k-dimensional face σ to the triangle Ti at which the vectors V (x) on σ points.
Note that we do not shift the (−1)n assigned to a triangle itself. In other words,
we think of the vector field V as wind blowing on M , and we let it blow each
numbers assigned to each faces into a triangle Ti . Note that this shifting, which
we may call ”blowing”, is the desired one. This can be verified very easily for
2 or 3 dimensional case, and in similar way for higher dimension. This proves
lemma 7.


Corollary 3 Let M be a compact smooth manifold with a smooth non-vanishing


vector field on it. Then χ(M ) = 0.

For any given triangulation of M , the summation of all ±1s assigned to every
face calculates the Euler characteristic χ(M ) = k0 − k1 + k2 − · · · + (−1)n kn
where kr is the number of r-dimensional faces in the triangulation. But by
Lemma 7, the summation equals zero. Therefore χ(M ) = 0.


Note that this tequenic can be used as a way of calculating the index of
source, sink and saddle of a smooth vector field. That is, we enclose each source,
sink or saddle by convex polygonal neighborhood, and number each faces of it,
and then blow the numbers by the vector field on the convex neighborhood. To
define the blowing, the vector field on each faces should behave well-identically
point inward or outward. For that, we choose a small  ball around each zeros
and deform the vector field, if we needed, to perfectly symmetric identity map,
minus identity map or reflection map. Now we take a n-dimensional cube inside
the  neithborhood to enclose each zero, and observe that the vector field abound
a source identically point outward on the cube, and the vector field around a
sink identically point inward on the cube. So the numbering on the cube is
well-defined. We assume n to be even for now. Also observe that if we blow
the numbers on the cube around a souce, all the numbers are blown outward of
the cobe, so that the only remaining number is the (−1)2m = 1 assigned to the
n-dimensional cube, and the summation is 1. For sink, all numbers are blown

5
inside of the cube, and the summation is equal to the Euler characteristic of
n-dimensional cube, which is 1. The cube works for saddle too. The vector field
around saddle looks like xn = 1/r where r = (x21 + x2 + · · · + x2n−1 )1/2 , and
2-dimensional case can be easily checked : all the numbers on vertices blown
out, and only two (-1) on four edges blown in, so that the numbers sums up to
−1 − 1 + 1 = −1. I left the higher dimensional case and the case of n odd.
We summerize our observation as a lemma.

Lemma 6 Let V be a smooth vector field on a smooth manifold in Rn , and


let x0 , y0 , z0 be a source, sink and saddle of V respectively. Then there are n-
dimensional polihedron neighborhood B(x0 ), B(y0 ) and B(z0 ) of each zeros in
M such that the numbering on each n-dimensional polygon is well-defined and
after blowing along the vector field, the numbers inside each polygon sums up to
the index of each zeros.

Theorem 1 (Poincare-Hopf ) Let M be a compact smooth manifold and V be


an arbitrary smooth vector field on M with only isolated zeros. Then the sum
of the indices at the zeros is equal to the Euler characteristic χ(M ).

By lemma 6, we may assume that every zero of V is either a source, a sink


or a saddle. We chose a convex polygonal neighborhood described in lemma
8 around each zeros and cut them out. Now we have a compact manifold M 0
with non-vanishing vector field V 0 , and lemma 7 says there is a triangulation
of M 0 with a shifting such that the summation of all numbers assigned to each
n-dimensional face(triangles of the triangulation) is zero. Now we paste the
polygonal neighborhoods back to M 0 , and we have a sort of cell division on
M such that each zero is enclosed by a polygonal neighborhood and outside it,
triangulated. To make it a proper cell division, we subdivide each faces of the
polygonal neighborhood to agree with the triangles outside them. Note that this
further subdivision of a face of the polygonal neighborhoods does not change the
summation of numbers, which were set to be the index of the zero inclosed in
the polygon, since the vectors V (x) on a face of each polygon were set to point
identically point inward or outward so that all the numbers of new sub-faces
point identically inward or outward. In this way, we have a proper cell division
of M such that each zeros are enclosed in a polygon, and after blowing along the
vector field the numbers sums up to the index of each zero in those polygons,
and sums up to zero in the triangles which does not enclose zeros. Therefore
the summation of all numbers is equal to the summation of the indices at the
zeros of V equals the Euler Characteristic χ(M ).


6
References
Books : Topology from the differentiable viewpoint, J.W Milnor
Euler’s Gem : The Polyhedron Formula and the Birth of Topology, David S,
Richeson
Papers : The Poincare-Hopf Theorem, Alex Wright and Kael Dixon

You might also like