Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Nanavaty

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

AN INTRODUCTION TO DIFFERENTIAL FORMS, STOKES’

THEOREM AND GAUSS-BONNET THEOREM

ANUBHAV NANAVATY

Abstract. This paper serves as a brief introduction to differential geome-


try. It first discusses the language necessary for the proof and applications of
a powerful generalization of the fundamental theorem of calculus, known as
Stokes’ Theorem in Rn . Further, geometry in R3 will be discussed to present
Chern’s proof of the Poincaré-Hopf Index Theorem and Gauss-Bonnet The-
orem in R3 , both of which relate topological properties of a manifold to its
geometric properties. Only a working knowledge of multivariable calculus is
needed to understand this paper. All other concepts are introduced and dis-
cussed.

Contents
1. Introduction 1
2. Differential Forms and Manifolds 2
2.1. Differentiable Manifolds 2
2.2. Differentiable Forms 4
2.3. Exterior Derivatives 6
2.4. Integration of Forms 8
3. Stokes’ Theorem 10
3.1. Applications 13
4. Riemannian Manifolds and Geometry in R3 14
4.1. Cartan’s Structure Equations in Rn 15
4.2. Curvature in R3 18
5. The Gauss-Bonnet Theorem 20
6. Conclusion 26
Acknowledgments 26
References 26

1. Introduction
We first introduce the concept of a manifold, which leads to a discussion of
differential forms, the exterior derivative and pull-back map. We then discuss
integration of forms in Rn in order to state and prove Stokes’ Theorem in Rn . A few
applications of Stokes’ Theorem are also stated and proved, such as Brouwer’s fixed
point theorem. In order to discuss Chern’s proof of the Gauss-Bonnet Theorem in
R3 , we slightly shift gears to discuss geometry in R3 . We introduce the concept
of a Riemannian Manifold and develop Elie Cartan’s Structure Equations in Rn to
define Gaussian Curvature in R3 . The Poincaré-Hopf Index Theorem is first stated
1
and proved, and the concept of the Euler number is introduced in order to end
with a proof of the Gauss-Bonnet Theorem in R3 . A few important implications of
the theorem are then mentioned. Most of the definitions, along with proofs of the
propositions and theorems have been adapted from Do Carmo’s Differential Forms
and Applications [1], along with Pressley’s Elementary Differential Geometry [2].

2. Differential Forms and Manifolds


We begin with the concept of a differentiable manifold. A generic theme in
differential geometry is that we associate seemingly ‘unknown’ objects, such as
manifolds, with ‘known’ objects, such as Rn , so that we can study the local behavior
of the object using concepts such as differential forms.

2.1. Differentiable Manifolds.


Definition 2.1. A differentiable manifold is a set M along with a set of injective
maps fα : Uα → M , such that Uα ⊂ Rn are open, and:
S
(1) M = α fα (Uα )
(2) For all α, β such that fα (Uα )∩fβ (Uβ ) = W 6= Ø, it must be that fα−1 (W ), fβ−1 (W )
are both open in Rn . Further, fα−1 ◦ fβ and fβ−1 ◦ fα are differentiable.
(3) {(fα , Uα )}, called the set of charts or coordinate systems, is maximal in
regards to both (1) (2)
In other words, an n dimensional manifold, denoted M n is a set that locally
‘looks’ like Rn . The second condition makes sure that if two coordinate systems
overlap, then points that are ‘close together’ in one system are mapped to points
that are also ‘close together’ in the other system. The last condition serves the
purpose of inducing a topology on M n , meaning that open sets on M can now be
defined using open sets in Rn . From this definition, it is clear that Rn is a mani-
fold (as we can just map the topology on Rn to itself), but there are examples of
manifolds whose global properties are drastically different from Rn . For the sake of
simplicity, we only look at the cases where the manifold satisfies Haussdorf’s axiom
and has a countable basis1.

We will now describe the concept of the tangent space of a manifold. In the
manifold Rn , so to find a tangent vector to a curve
α : I = [a, b] → Rn
at point p ∈ [a, b] we calculate the derivative α0 (p), which is in fact the tangent
vector. However, it is not obvious how to do such a thing on an arbitrary manifold.
Note the following definition.
Definition 2.2. Take p ∈ M n , and let α : I → M n be smooth such that α(0) =
p ∈ M ,2. and consider D = {ϕ : M → R | ϕ is linear}. We define the tangent
vector to α at p as a map α0 (0) : D → R defined by α0 (0)(ϕ) = dt
d
(ϕ ◦ α)|t=0 .

1View [1] for a statement of Haussdorf’s axiom and the definition of a countable basis.
2Note that for any α such that p ∈ α(I), we can reparametrize α such that α(0) = p

2
The definition of the tangent vector is intuitively the same as the one we had
for Rn . In order to why this is the case, set M = Rn and take a curve α : I → Rn
such that α(t) = (α1 (t), . . . , αn (t)). For any linear f : Rn → R, we can calculate
the tangent vector at α(0) in the following way using the chain rule:
d X ∂f (α(0))
f (α(t))|t=0 = αi0 (0)
dt i
∂x i
hX ∂ i
= αi0 (0) f (α(0))
i
∂xi
d dα
As desired, dt f (α(t)) is dependent on dt .

Definition 2.3. For M n , the tangent space at a point p ∈ M n is denoted as


Tp M = {α0 (0) | α : I → M n , α(0) = p}.
In other words, the tangent space at a point is the set of all tangent vectors at
that point. This next proposition allows for us to better understand the tangent
space at p.
Proposition 2.4. For p ∈ M n , dim(Tp M n ) = n. Further, if p ∈ fα (Uα ), then
∂ ∂
span({ ∂x j
| j ∈ [n]}) = Tp M n . The set { ∂x j
} is the set of partial derivative
∂ ∂
operations of Rn with respect to fα such that ∂xj ϕ := ∂xj (ϕ ◦ fα )

Proof. If p ∈ fα (Uα ) such that Uα ⊂ Rn and p = f (0, . . . , 0), then for a curve
α : I → M n with α(0) = p we have that f −1 (α(0)) = (x1 (t), . . . , xn (t)). For
simplicity’s sake3, we conflate f −1 ◦ α with α, so that α(t) = (x1 (t), . . . , xn (t)), x ∈
Rn . This allows us to state that ϕ ◦ α(t) = ϕ(x1 (t), . . . , xn (t)) for all ϕ : M → R.
We can now deduce that
d
(2.5) a0 (0)ϕ = (ϕ ◦ α)|t=0
dt
d
(2.6) = ϕ(x1 (t), . . . , xn (t))|t=0
dt
n 
X ∂ψ  0
(2.7) = xi (0) We can now write a0 (0) as:
i=0
∂x i 0
n
X  ∂ 
(2.8) a0 (0) = x0i (0)
i=0
∂xi 0
Since there are n terms in the summation, it follows that the tangent space is

n-dimensional and { ∂x i
} is a basis for the space. 

Proposition 2.9. If M = Rn , then for p ∈ Rn , Tp (Rn ) = Rn .


It is now useful to discuss the concept of orientability.
Definition 2.10. A differentiable manifold is orientable if there exists a differen-
tiable structure {(fα , Uα )} such that for all α, β where fα (Uα ) ∩ fβ (Uβ ) = W 6= Ø,
the determinant of the differential map (known as the jacobian J) of fβ−1 ◦ fα is
positive.

3This is a standard practice in differential geometry

3
In other words, let F = fβ−1 ◦fα : Uα → Uβ . It follows, by multivariable calculus,
that for p ∈ Uα , the differentiable map, or jacobian, is defined by DF (p) : Tp (Uα ) →
Tf (p) (Uα ). If det(dF ) > 0 for all p, this intuitively means that the tangent spaces of
the two charts cannot have opposite orientations, and so there is a fixed orientation
at each p ∈ M .
2.2. Differentiable Forms. Now, we can slightly shift gears to discuss forms on
manifolds.
Definition 2.11. The dual space at p is denoted as the set of linear functions
(Tp M )∗ = {ϕ : Tp M → R}. Further we define
k
^
(Tp M )∗ = {ϕ : [(Tp M )∗ ]k → R}
as the set of all real linear functions that take k elements4 of Tp M and are k-linear
and alternate5
The definitions of a 1-form and 0-form follow.
Definition 2.12. An exterior 1-form is a function ω which maps p ∈ M n to
ω(p) ∈ (Tp M )∗ .
In other words, the 1-form assigns to each point p a real linear function on M n .
We can find a basis for these forms in the following manner: we know that lo-
cally, p is parametrized by some Uα ⊂ Rn such that p ∈ fα (Uα ). Consequently,
we can assign to p coordinates of Rn by the functions xi : Rn → R such that
xi (fα−1 (p)) = (fα−1 )i (p). In other words, we project the i’th component of p ac-
cording to the coordinate chart Uα . For convenience sake, we conflate fα−1 (p) ∈ Rn
with p ∈ M , because we assume that we are working under a single chart Uα , so
locally, f (Uα ) can be conflated with Rn .

Consider the differential maps (via multivariable calculus) {d(xi )}. We can see
that
∂xj
d(xi )p (ej ) = = δij
∂xi
Above, the {ej } are the canonical basis in Rn . Using our knowledge of linear
algebra, we can see that the set serves as a basis for (Tp M )∗ , so we can write the
1-form as
X n
(2.13) ω= ai dxi
i=1
n
Where ai : M → R are functions.
Such functions ai are called 0-forms. If each ai is differentiable, then ω is called
a differentiable 1- form. In order to define forms of higher degree, we need to
introduce a new concept.
Definition 2.14. A wedge product of two linear functionals is denoted by ∧ and
defined by (ϕ1 ∧ ϕ2 )(v1 , v2 ) = det(ϕi (vj )), where ϕ1 , ϕ2 ∈ (Tp M )∗ and v1 , v2 ∈
Tp M . By properties of the determinant, we can see that dxi ∧ dxi = 0 and dxi ∧
dxj = −dxj ∧ dxi if i 6= j.
4[(T M )∗ ]k = (T M )∗ × · · · × (T M )∗ (k-times)
p p p
5Alternate means that that if (for example) ϕ ∈ V2 (T M )∗ , then ϕ(v , v ) = −ϕ(v , v )
p 1 2 2 1
4
In M n , one can think of the wedge product as the area of the parallelogram
formed between (ϕ1 (v1 ), ϕ1 (v2 )) and (ϕ2 (v1 ), ϕ2 (v2 )). Now, for the generic defini-
tion of a k − f orm
Definition 2.15. An exterior differentiable k-form is a function ω which maps
Vk
p ∈ M n to ω(p) ∈ ( Tp M )∗
It also follows that we can write
Xn
(2.16) ω= ai1 <···<ik (dxi1 )p ∧ · · · ∧ (dxik ) where i1 , . . . ik ∈ [n]
i1 <···<ik

Where (dxi1 )p ∧ · · · ∧ (dxik )p is a wedge product of 1-forms, and each aI : M n → R


is a differentiable map. Similarly to the basis for the 1-form, one can check that
Vk
{(dxi1 )p ∧ · · · ∧ (dxik )p | i1 , . . . ik ∈ [n]} serves as a basis for ( Tp M )∗
Example 2.17. If ω is a 2-form on a 3-manifold, then locally in most generic form,
ω = a12 (dx1 ∧ dx2 ) + a13 (dx1 ∧ dx3 ) + a23 (dx2 ∧ dx3 )
We can more succinctly write (2.16) as
Xn
ω= aI dxI
I
We can also naturally define wedge product between two forms
Pn Pn
Definition 2.18. If ω = I aI dxI is a k-form and ϕ = J aJ dxJ is an s-form,
then we define a (k + s)-form by :
X
ω∧ϕ= aI bJ (dxI ∧ dxJ )
IJ
Here, the IJ sum of the product dxI ∧ dxJ is the sum of the wedge products of all
possible combinations of basis terms of each form.
Now that we have an understanding of forms, we can now relate a k-form on
one manifold to another k-form on another manifold if given a differentiable map
between the two manifolds. In particular:
Definition 2.19. Given manifolds M n and N m and differentiable map f : N m →
Vk Vk
M n , the pull-back map is defined as f ∗ : k∈[n] (Tp M n )∗ → k∈[n] (Tp N m )∗ ,
S S
n ∗ m
such that if ω is a k-form in M , then f ω is a k-form in N given by
(f ∗ ω)(p)(v1 , . . . , vk ) = ω(f (p))(dfp (v1 ), . . . , dfp (vk ))
Where p ∈ N m , v1 , . . . , vk ∈ Tp N m , and dfp : Tp N m → Tp M n is the differential
map. If g is a 0-from in M n then we define
f ∗g = g ◦ f
Proposition 2.20. If ω and ϕ are k-forms in M n , g is a zero form in M n , and
we are given f : N m → M n , then
(1) f ∗ (ω + ϕ) = f ∗ ω + f ∗ ϕ
(2) f ∗ (gω) = f ∗ (g)f ∗ (ω)
(3) If ϕ1 , . . . , ϕk are 1-forms on M n then f ∗ (ϕ1 ∧ · · · ∧ ϕk ) = f ∗ (ϕ1 ) ∧ · · · ∧
f ∗ (ϕk )
We leave these proofs to the reader. It should be noted, however, that the
pullback map seems to be a reasonable definition due to these properties.
5
2.3. Exterior Derivatives. Now comes an important concept, central to Stokes’
theorem:
Definition P
2.21. An exterior differential of a k-form ω is a k + 1 form dω, such
n Pn
that if ω = I aI dxI , then dω = I d(aI ) ∧ dxI
From multivariable calculus, we know that for any 0-form aI : Rn → R, the
derivative of a is given as
n
X ∂a
da = dxi
i=0
∂xi
which agrees with the exterior derivative.
Example 2.22. If the 1-form ω = xydx + x2 y 2 dy, then:
dω = d(xy) ∧ dx + d(x2 y 2 ) ∧ dy
= (ydx + xdy) ∧ dx + (2xy 2 dx + 2yx2 dy) ∧ dy
= x(dy ∧ dx) + 2xy 2 (dx ∧ dy) (because dx ∧ dx = 0)
2
= (2xy − x)(dx ∧ dy) (because dx ∧ dy = −dy ∧ dx)
We will now prove some minor properties of the exterior derivative to arrive at
a significant conclusion.
Proposition 2.23. (1) For any k-forms, ω1 , ω2 , d(ω1 ) + d(ω2 ) = d(ω1 + ω2 )
(2) If ω is a k-form and ϕ is an s-form, then d(ω ∧ ϕ) = dω ∧ ϕ + (−1)k ω ∧ dϕ
(3) For any k-form ω, d2 (ω) = d(dω) = 0.
Proof. (1) We know from multivariable calculus that for 0-forms a, b : Rn → R,
that by distributivity
P of the derivative
P operator, d(a + b) = da + db. Now,
let ω1 = I aI dxI and ω2 = I bI dxI . Consequently,
X X 
d(ω1 + ω2 ) = d aI dxI + bI dxI
I I
X 
=d (aI + bI )dxI
I
X
= d(aI + bI ) ∧ dxI
I
X X
= d(aI ) ∧ dxI + d(bI ) ∧ dxI = d(ω1 ) + d(ω2 )
I I
Proving (1).
(2) Let ω and ϕ be as described. By definition 1.17,
X 
d(ω ∧ ϕ) = d aI bJ (dxI ∧ dxJ )
IJ
X
= d(aI bJ ) ∧ dxI ∧ dxJ
IJ
X X
= d(aI )bJ ∧ dxI ∧ dxJ + aI d(bJ ) ∧ dxI ∧ dxJ
IJ IJ
X X
= (d(aI ) ∧ dxI ) ∧ bJ dxJ + (−1)k aI d(xI ) ∧ (dbJ ∧ dxJ )
IJ IJ
k
= dω ∧ ϕ + (−1) ω ∧ dϕ
6
Thus proving (2).
(3) We first prove the proposition for a 0-form. Suppose f : Rn → R is a
0-form. The following must hold:
n
X ∂f
d(f ) = dxi
i=0
∂xi
n n
X ∂ h X ∂f i
d(d(f )) = dxi ∧ dxj
j=0
∂xj i=0 ∂xi
n X
n
X ∂2f
= dxi ∧ dxj
j=0 i=0
∂xi ∂xj
X ∂2f ∂2f
= − dxi ∧ dxj (because dx ∧ dy = −dy ∧ dx)
i<j
∂xi ∂xj ∂xj xi
 ∂2f ∂2f 
=0 because =
∂xj xi ∂xi ∂xj
Now, for the general case k-form, and due to (1), we only need to consider
when ω = aI dxI . By definition dω = daI ∧ dxI , and so

d2 ω = d(daI ∧ dxI )
= d2 aI ∧ dxI + (−1)daI ∧ d2 xI from (2)
=0

d2 aI = 0 from the case for 0-forms. It must be that d2 ω = 0 for all ω.




Note that the exterior derivative is the generalized form of the derivative in
multivariable calculus which we can use for in wider variety of cases in the following
sense: In multivariable calculus, the derivative of a 0-form in Rn is a 1 × n matrix,
which can correspond to a 1-form. Now, we can take derivatives of higher forms.
it is also interesting to note that the pullback map and the exterior derivative are
related in a straightforward but important way.

Proposition 2.24. If f : M n → N m , and ω is a k-form in N m , then d(f ∗ ω) =


f ∗ (dω). In other words, if two forms are related by a pullback map, then their
derivatives are similarly related.

Proof. Let ω : N m → R be a 0-form in N m . In a chart, the following holds:


 X ∂ω 
f ∗ (dω) = f ∗ dxi
i∈m
∂xi
X ∂ω ∂f
= dyi (by definition of pullback)
ij
∂xi ∂yi
X ∂(ω ◦ f )
= dxi
i
∂xi
= d(f ∗ (ω))
7
aI dxI be a k-form in N m .
P
Now let ω = I
X
f ∗ (dω) = f ∗ ( d(aI ) ∧ dxI )
I
X
= f ∗ (d(aI ) ∧ dxI ) (prop. 1.19)
I
X
= f ∗ (d(aI )) ∧ f ∗ (dxI )
I
X
= d(f ∗ aI ) ∧ f ∗ (dxI ) (from the 0-form result)
I
X
= d(f ∗ ( aI dxI )) = d(f ∗ ω)
I

Which concludes the proof. 

2.4. Integration of Forms. Forms are also useful because we can integrate them
along manifolds. Before we discuss integration, we discuss representation of forms.
Definition 2.25. If fα : Uα ⊂ Rn → M n , and ω is a k-form in M n defined at
p ∈ fα (Uα ), then the representation of ω at Uα is ωα such that
ωα = fα∗ ω
Proposition 2.26. If fα : Uα ⊂ Rn → M n , fβ : Uβ ⊂ Rn → M n such that
p ∈ fα (Uα ) ∩ fβ (Uβ ), then for any k-form ω defined at p,
(fβ−1 ◦ fα )∗ ωβ = ωα
Proof. We use the definition of a pull-back map.
(fβ−1 ◦ fα )∗ ωβ (v1 , ..., vk ) = ωβ (dfβ−1 (p)) d(fβ−1 )p (v1 ), . . . , d(fβ−1 )p (vk )


= ω(p) (dfβ ◦ dfβ−1 )p (v1 ), . . . , (dfβ ◦ dfβ−1 )p (vk )




= ωα (v1 , ..., vk )


Definition 2.27. Let M n be a compact oriented manifold, and ω = a(x1 . . . xn )dx1 ∧


· · · ∧ dxn be a n-form defined on A ⊂ M n , and define K to be the support of
ω given as K the closure of {p ∈ M n | ω(p) 6= 0}. Since K ⊂ M , it follows
that K is compact. If K ⊂ Vα , where Vα = fα (Uα ) for some α, we then denote
ωα = fα∗ (ω) = aα dx1 ∧ · · · ∧ dxn , and define
Z Z Z Z
ω= ω= ωα = aα dx1 . . . dxn
A Vα Uα

We will avoid issues of convergence by assuming that M n is compact throughout


Proposition 2.28. The definition of integration of forms on a single chart is well-
defined. In particular, if the support K of ω is contained in both Vα = fα (Uα ) and
Vβ = fβ (Uβ ), then
Z Z
ωα = ωβ
Uα Uβ
8
Proof. Due to the fact that K ⊂ Vα ∩ Vβ we can shrink Vβ so that Vβ = Vα . Now,
we know by 1.25 that (fβ−1 ◦ fα )∗ ωβ = ωα , so by definition of pull-back maps, let
F = fβ−1 ◦ fα , then we have
F ∗ ωβ = ωα
det(dF )aβ dx1 ∧ · · · ∧ dxn = ωα

Where aα (y1 , . . . yn ) = aβ (F1 (y1 , . . . yn ), . . . , Fn (y1 , . . . , yn )), and yi ∈ Uβ , yi ∈ Uα .


We also know that by the substitution of variables formula given in multivariable
calculus, it follows that
Z Z

aα dx1 ∧ · · · ∧ dxn = det(dF ) aβ dx1 ∧ · · · ∧ dxn
Uα Uβ
n
Since M is oriented, det(dF ) > 0, so we have that
Z Z
ωα = ωβ
Uα Uβ

Suppose K of ω is not covered by a single chart, meaning that there does not
exist fα (Uα ) such that K ⊂ fα (Uα ). Since K is compact, if we are given an open
covering of patches {fα (Uα )}α∈Λ of K, then there exists a finite open subcover of K
given by {fα (Uα )}α∈[n] . We use this fact as motivation for the following definition.
Definition 2.29. Given a finite covering {Vα } of a compact manifold M , a parti-
tion of unity subordinate to {Vi } is a finite family of differentiable real functions
{ϕ1 , . . . , ϕn } on M such that:
Pn
(1) i ϕi (x) = 1, for all x ∈ M

(2) 0 ≤ ϕi (x) ≤ 1, for all x ∈ M , i ∈ [n]

(3) For all i ∈ [n], there exists Vi ∈ {Vα } such that the support Ki of ϕi is
contained in Vi .
Note that if ω is defined on M , the support of ϕi ω is contained in Vi . This
allows us to define the integration of a form across multiple surface patches in the
following way:6
Definition 2.30. Given a finite covering {Vα } of a compact manifold M , and a
subordinate partition of unity {ϕ1 , . . . , ϕn }, we define
Z Xn Z
ω= ϕi ω
M i Vi

It follows in straightforward manner that this integral is well defined, and does
not depend on the covering. We have at present introduced and discussed most of
the language needed to understand Stokes’ Theorem.

6We will assume the existence of smooth partitions of unity without proof.

9
3. Stokes’ Theorem
Stokes’ Theorem is about manifolds with boundaries, which is larger than the
class of manifolds. For example, a cylinder of radius r and length d oriented in the
z − axis is not a manifold, because, roughly speaking, the edges of the cylinder do
not locally ‘look’ like Rn . However, the area around the edges does look like the
half plane, H 2 = {(x1 , x2 ) | x1 ≤ 0}. This motivates the following definition:
Definition 3.1. A differentiable manifold with a regular boundary is a set M along
with a set of injective maps fα : Uα ⊂ H n → M , H n = {(x1 , x2 , . . . , xn ) | x1 ≤ 0}
(H n is the n-dimensional half-plane) such that each Uα is open in H n . Further,
S
(1) M = α fα (Uα )
(2) For all α, β such that fα (Uα ) ∩ fβ (Uβ ) = W 6= Ø, then fα−1 (W ), fβ (W ) are
both open in Rn . Further, fα−1 ◦ fβ and fβ−1 ◦ fα are differentiable.
(3) {Uα , fα }, is maximal in regards to both (1) (2)
The above definition is similar to that of a differential manifold, except for the
fact that Rn is replaced by H n . Intuitively, the edge of the half plane is what results
in a boundary. We can further explore this in the following proposition
Lemma 3.2. For a manifold M and p ∈ M , if there exists α such that p =
fα (0, x2 , . . . , xn ) where xi ∈ R, then p is on the boundary of M , denoted ∂M .
Further, if there exists β such that p ∈ fβ (Uβ ), then ‘p’ is still on the boundary, or
p = fβ (0, x20 , . . . , xn0 ) for some xi0 ∈ R.
Proof. Assume, for the sake of contradiction that there where exists α such that
p = fα (0, x2 , . . . , xn ), but there also there exists β such that p = fβ (x10 , x20 , . . . , xn0 )
such that x1 6= 0. If W = fα (Uα ) ∩ fβ (Uβ ), define f : fβ−1 (W ) → fα−1 (W ) such
that f (x) = fα−1 ◦ fβ (x), for x ∈ fβ−1 (W ). It follows that f is both bijective and
differentiable. By the inverse function theorem, f −1 will take a neighborhood V ⊂
Uα such that fα−1 (p) ∈ V to another neighborhood U ⊂ Uβ such that fβ−1 (p) ∈ U .
This in turn implies that there exists points (x1 , . . . , xn ) ∈ Uα such that x1 > 0,
which is a contradiction because Uα ⊂ H n . Therefore, the boundary is well defined,
because it does not change with parametrization. 

The above proof suggests that the boundary of a manifold is also a manifold. To
be precise:
Lemma 3.3. Given M n with a boundary, ∂M , it follows that ∂M is a manifold
of dimension (n-1). Also, if M n is oriented, so is ∂M .
Proof. If {fα , Uα } is a differentiable structure on M n , then consider {fα0 , Uα0 } where
Uα0 = Uα ∩ {(x1 , . . . , xn ) | x1 = 0}, and fα0 : Uα0 → M n such that fα0 (x) = fα (x)
for all x ∈ Uα0 . Due to {fα , Uα } being a differentiable structure, it must be that
{fα0 , Uα0 } is a differentiable structure on ∂M , and since x1 = 0, it follows that ∂M
is n − 1 dimensional.

To see the orientation, we know that, by definition of orientation on M there exists a


differentiable structure {fα , Uα } such that for α, β where fα (Uα )∩fβ (Uβ ) = W 6= Ø,
(3.4) det(fα−1 ◦ fβ ) > 0
10
As in the first part of the proof, consider {fα0 , Uα0 }. Due to (2.4), for all α, β where
fα0 (Uα0 ) ∩ fβ0 (Uβ0 ) = W 0 6= Ø, it must be that det(fα0−1 ◦ fβ0 ) > 0, so ∂M is also
oriented due to the orientation of M . 

Now, we can finally state and prove Stokes’ Theorem.

Theorem 3.5. Consider a differentiable compact manifold M with boundary ∂M ,


and a n-1 form ω defined on M . Let i : ∂M → M be the inclusion map, defined as
i(x) = x, for all x ∈ ∂M . If follows that7

Z Z
dω = i∗ ω
M ∂M

Proof. Two major cases arise with regards to the closed support K (of ω).

(1) If there exists V = fα (U ) such that K ⊂ V .


Let
X
ω= aj dx1 ∧ · · · ∧ dxj−1 ∧ dxj+1 ∧ · · · ∧ dxn
j

Then,
n
X ∂aj 
dω = (−1)j−1 dx1 ∧ · · · ∧ dxn
j=1
∂xj

Now, suppose thatR K ∩ ∂M = Ø. It follows that by definition of the


inclusion map i, ∂M i∗ ω = 0. Since aj : U → R, we extend each aj to H n
in the following manner.

(
aj (~x) if ~x ∈ U
aj (~x) =
0 if ~x ∈ H n \ U

Consider a parallelepiped Q ⊂ H n , such that

Q = {[x11 , x21 ]×· · ·×[x1n , x2n ]} where x1i ≤ xi ≤ x2i , for (x1 , . . . , xi , . . . , xn ) ∈ f −1 (K)

In other words, Q is the smallest parallelepiped containing f −1 (K). Note


that Q exists because K is compact (and so is f −1 (K)). Therefore, we

7There is a slight abuse of notation here. In this case, we conflate ω with ω , as by definition
α
we can only integrate ωα .
11
have:
Z Z X n
∂aj 
dω = (−1)j−1 dx1 . . . dxn
M U j=1
∂xj
n Z 
X
j−1 ∂aj 
= (−1) dx1 . . . dxn
j=1 Q ∂xj
n Z x21 Z x2j Z x2n  ∂a 
j
X
= (−1)j−1 ··· ··· dx1 . . . dxn
j=1 x11 x1j x1n ∂xj
n Z x21 Z x2n Z x2j  ∂a 
j
X
= (−1)j−1 ··· ··· dxj dx1 . . . dxj−1 dxj+1 . . . dxn
j=1 x11 x1n x1j ∂xj
Xn Z h i
= (−1)j−1 aj (x1 , . . . , x1j , . . . , xn ) − aj (x1 , . . . , x0j , . . . , xn ) dx1 . . . dxj−1 dxj+1 . . . dxn
j=1 Q

=0
For the last step, it was realized that
aj (x1 , . . . , x1j , . . . , xn ) = aj (x1 , . . . , x0j , . . . , xn ) = 0
Now, consider when K ∩ ∂M 6= Ø. In this case, we know that if p ∈ M
is on the boundary, then p = f (0, x1 , . . . , xn ), for all parametrization maps
f . Thus, if
X
ω= aj (x1 , . . . , xn )dx1 ∧ · · · ∧ dxj−1 ∧ dxj+1 ∧ · · · ∧ dxn
j

Then,
i∗ ω = a1 (0, . . . , xn )dx2 ∧ · · · ∧ dxn
We construct a parallelepiped Q ⊂ H n similar to the previous subcase,
except that
x11 ≤ x1 ≤ 0, while x1i ≤ xi ≤ x2i for i ≥ 2 where (x1 , . . . , xi , . . . , xn ) ∈ f −1 (K)
So, as in the previous subcase, Q is the smallest parallelepiped that contains
f −1 (K). Now,

Z n Z 
X ∂aj 
dω = (−1)j−1 dx1 . . . dxn
M j=1 Q ∂xj
Z h i
= a1 (0, . . . , xn ) − a1 (x11 , . . . , xn ) dx2 . . . dxn
Q
n
X Z h i
j−1
+ (−1) aj (x1 , . . . , x1j , . . . , xn ) − aj (x1 , . . . , x0j , . . . , xn ) dx1 . . . dxj−1 dxj+1 . . . dxn
j=2 Q
Z
= [a1 (0, . . . , xn )]dx2 . . . dxn (as aj (x1 , . . . , xij , . . . , xn ) = 0 for j ≥ 1, i ∈ {1, 2})
Q
Z
= i∗ ω
∂M
12
(2) We can finally prove the general case. Given the differential structure
{fα , Uα } on a compact manifold M , take an open covering of M {Vβ }β∈Λ ,
where Vβ = fβ (Uβ ) for some β. There exists a finite subcover {Vα } ⊂
{Vβ }β∈Λ of M . Now let {ϕ1 . . . , ϕm } be a differentiable partition of unity
subordinate to {Vα }. For an n − 1 form ω, we have that ϕj ω is an n − 1
form
P completely contained in Vj , which is the case P first discussed. Since
ϕ j = 1, differentiating both sides gives us j dϕj = 0. Recall that
Pj
ϕ
j j ω = ω. Using these facts, we find that:
X X X
d(ϕj ω) = dϕj ω + ϕj dω
j j j
X
= ϕj dω = dω
j

As a result,
Z m Z
X
dω = ϕj ω
M j=1 M
Xm Z
= i∗ (ϕj ω)
j=1 ∂M
Z m
X
= i∗ (ϕj ω)
∂M j=1
Z
= i∗ ω
∂M

3.1. Applications. Stokes’ Theorem appears in many forms. It is, in fact, the gen-
eralized form of the fundamental theorem
R of calculus. This can be seen in the first
case of the proof, when we compute M dω. This theorem will be central to proving
the Gauss-Bonnet theorem, but we now look at some of its other applications.
Corollary 3.6. Green’s Theorem. If M = R2 , ω = P dx + Qdy, then for the region
R bounded by the closed curve ∂R = C,
Z Z  Z
∂Q ∂P 
− dxdy = (P dx + Qdy)
R ∂x ∂y C
Proof. By straightforward calculation,
 ∂Q ∂P 
d(P dx + Qdy) = − dx ∧ dy
∂x ∂y
The result follows by Stokes’ theorem. 
Corollary 3.7. Brower’s Fixed Point Theorem for Smooth Maps. Let Dn ⊂ Rn
be the ball Dn = {p ∈ Rn | |p| ≤ 1}, where |p| is the euclidean norm. For every
differentiable map f : Dn → Dn there exists fixed point, q ∈ Dn such that f (q) = q.
Proof. We must first prove two lemmas:
Lemma 3.8. If M n is a compact oriented differentiable manifold, then there exists
a differential n-from ω such that every representative form is positive everywhere
on M
13
Proof. Since M is oriented, there exists a differentiable structure {fα , Uα } such that
for all α, β where fα (Uα ) ∩ fβ (Uβ ) = W 6= Ø, the determinant of the differential
map of f −1 ◦ g is positive. Further, since M is compact, for any open cover, there
exists a finite open subcover of M , which we denote as {Vi }. Subordinate to {Vi }
there exists a partition of unity {ϕi }.
Now, on each Vi we define ωi = 1dx1 ∧ · · · ∧ dxn . Due to orientablility, if any
elements in the covering overlap, the form will stillP be positive due to the positive
determinant. Since {Vi } is a covering of M , ω = i ϕi ωi is positive and defined
globally on M . 

Lemma 3.9. For any compact, oriented, differentiable manifold M n with boundary
∂M , there does not exist a differentiable map f : M → ∂M such that f |∂M is the
identity.
Proof. By contradiction. Suppose that f : M → ∂M exists such that f |∂M is the
identity. By Lemma 2.3, ∂M is also oriented and has dimension n-1. Take the
(n-1)-form given in the previous lemma, ω = dx1 ∧ · · · ∧ dxn−1 . Since dω = 0, and
f |∂M is the identity, df ∗ (ω) = f ∗ (dω) = 0. Also,every representative form of ω is
positive, so ω = i∗ f ∗ (ω) and
Z Z
ω= i∗ (f ∗ ω) 6= 0
∂M ∂M

But, by Stokes’ Theorem,


Z Z Z
i∗ (f ∗ ω) = d(f ∗ ω) = f ∗ (dω) = 0
∂M M M
R
So ∂M
ω = 0 which is a contradiction. Therefore, no such f exists. 

Now, we can prove Corollary 3.7. Suppose, for the sake of contradiction, that
there exists such a function f : Dn → Dn such that f (q) 6= q. Consider the half-line
(or a ray) starting from f (q) and passing through q. We know that only one such
line exists for each q, because f (q) 6= q. We also know that this ray will intersect
∂Dn at a unique point r (notice that if q ∈ ∂Dn , then q = r). Define g : Dn → ∂Dn
such that g(q) = r. It then follows that g|∂Dn is the identity mapping, which is a
contradiction by 2.9. Therefore there must exist a fixed point. 

Stokes’ Theorem, as we can see, can be used to prove some important theorems.
Now, in order to discuss the Gauss-Bonnet theorem in R3 , we must first discuss
important concepts related to geometry in R3

4. Riemannian Manifolds and Geometry in R3


We first need to introduce vector fields, a concept similar to that of forms:
Definition 4.1. Given a differentiable manifold M , A differentiable vector field is
a function X that assigns p ∈ M to X(p) ∈ Tp M . In other words, X is a function
which assigns a point in M to another vector in its tangent space. Recall that a
tangent vector is a function which assigns a real linear function on M to an element
of R. In order to ensure that the vector field is differentiable, we require for all
p ∈ M and all linear ϕ : M → R that X(p)(ϕ) : M → R is differentiable.
14

It naturally follows that if fα : Uα → M , and Xi = ∂x i
, i ∈ {1, . . . , n} is the
associated basis of the parametrization, then a vector field X in fα (Uα ) can be
written as X
X= ai Xi
i
P
Note that each vector
P field corresponds to a 1-form. In other words, X = i ai Xi
corresponds to ω = i ai dxi .
Definition 4.2. A Riemannian manifold is a manifold M , along with a choice,
for each p ∈ M , of a Riemannian metric inner product h , ip , such that for any
differentiable vector fields X, Y we have that p 7→ hX(p), Y (p)ip is differentiable in
M . Hence, the inner product at p ∈ M is defined in Tp M .
From now on, we write hX(p), Y (p)i = hX, Y ip . If we take M = Rn , we define,
for p ∈ Rn , that n
P if x, y ∈ Tp R where x = (x1 , . . . , xn ) and y = (y1 . . . , yn ),
then hx, yi = i xi yi , which is known as the dot product. The concept of the
inner product is useful, because we can use it to quantify similarity between the
positions of two vectors, which in turn can give us a frame of reference. The next
few definitions further this notion.
4.1. Cartan’s Structure Equations in Rn .
Definition 4.3. Given U ⊂ Rn , an orthonormal moving frame is a set of vector
fields {e1 , . . . , en } such that for p ∈ U , we have that
 hei , 
ej ip = δij . Further, we
define the set of forms {ω1 . . . ωn } such that ωi (p) ej (p) = δij as the coframe
associated with {ei }. Thus, {(wi )p } is the dual basis of {(e1 )p }
Recall that Tp Rn ⊂ Rn , and ei : Rn → Rn , implying that the differentiable map
d(ei ) : Rn → Rn is linear. Intuitively, this map describes how the frame is rotated
as it moves from point to point in Rn . It would now be useful to create forms that
relate to the differential map.
Definition 4.4. Given the orthonormal moving frame {ei } and its coframe {ωi }
we define the connection forms ωij of d(ei ) : Rn → Rn such that for p, v ∈ Rn ,
X
d(ei )p (v) = (ωij )p (v)ej
j

In more general terms, X


dei = ωij ej
j

Because dei is a linear map, it follows that wij are 1-forms.


Proposition 4.5. In the indices i, j, the connection forms are anti-symmetric,
meaning that ωij = −ωji .
Proof. We know that hei , ej i = δij . Differentiating both sides, we get that:
hdei , ej i + hei , dej i = 0
DX E D X E
ωik ek , ej + ei , ωjk ek = 0
k k
ωij + ωji = 0
ωij = −ωji
15


Now we can state and prove Elie Cartan’s structure equations in Rn

Theorem 4.6. Let {ei } be a moving frame in U ⊂ Rn , let {wi } be its coframe,
and let {ωij } be the connection forms. The following relation holds:
X
dωi = ωk ∧ ωki
k

X
dωij = ωik ∧ ωkj
k

Where i, j, k ∈ {1, . . . n}.

Proof. Let a1 = (1, 0, . . . , 0), . . . an = (0, . . . , 1) be the canonical basis in Rn , and


let xi : U → R be such that for all y = (y1 , . . . , yn ) ∈ U , xi (y) = yi . Since each xi
is a 0-form, each dxi is a 1-form. In other words, xi projects the i’th component of
the vector. Further, dxi (aj ) = δij . Therefore {dxi } is the coframe of {ai }. Then,
for some arbitrary orthonormal moving frame {ei } and its coframe {ωi }, we have
that
X
(4.7) ei = βij ai
j

Further,
X
(4.8) ωi = βij dxj
j

Note that
X
(4.9) dei = ωik ek
k
X X X
(4.10) = ωik ( βkj ai ) = ωik βkj aj (3.7)
k j jk
P
Since dei = j dβij aj , we have that
X
(4.11) dβij = ωik βkj
k

Now differentiating 3.8,


X
dωi = dβij ∧ dxj
j
X
= ωik βkj ∧ dxj
k
X
= βkj dxj ∧ ωki
k
X
= ωk ∧ ωki
k
16
Differentiating 3.11, we get the second equation:
X X
d(dβij ) = 0 = dωik βjk − ωik ∧ dβjk
k k
X X X
dωik βjk = ωik ∧ ωks βsj
k k s
X
ωir = ωik ∧ ωkr (multiplying by the inverse matrix of βij )
k

Therefore, in R3 , the structure equations are as follows:
dω1 = ω12 ∧ ω2
dω2 = ω21 ∧ ω1
dω3 = ω13 ∧ ω32
dω12 = ω13 ∧ ω23
dω13 = ω12 ∧ ω21
dω23 = ω21 ∧ ω13
Now, before we can properly discuss manifolds in R3 , we must state and prove
a lemma that will be important in defining curvature in three dimensional space.
Lemma 4.12. (Cartan’s Lemma) Let V n be a vector space (such as Rn ), and
{ω1 , . . . , ωr | r ≤ n} be linearly independent
P 0-forms. If therePexists another set
of 0-forms {ϕ1 , . . . , ϕr } for which j ωj ∧ ϕj = 0, then ϕi = j aij ωj , such that
aji = aij
Proof. From linear algebra, we know that we can extend the set of linearly inde-
pendent forms to have a complete basis {ω1 , . . . , ωn } of V n . Therefore, given any
form ϕi , X X
ϕi = aij ωj + bij ωj
j≤r j>r
P
Let l = {r, r + 1, . . . , n}. If j ωj ∧ ϕj = 0, then
X X
0= aij ωi ∧ ωj + aij oi ∧ ωj
j≤r j>r
X X
= (aij − aji )ωi ∧ ωj + bil ωi ∧ ωl
i≤j i>l

However, since ωs ∧ ωt is linearly independent for s < t, it must be that bij = 0.


Further, this implies that aij = aji . 
We now claim that the structure equations are unique, in particular,
Proposition 4.13. Consider U ⊂ Rn , and let ω1 , . . . , ωn be a set of differential
1-forms defined in U . Assume that there exists a set of 1-forms {ωij | i, j ∈
{1, . . . , n}} such that
X
ωij = −ωji dωj = ωk ∧ ωkj
k

Then, {oij } is unique.8


8We leave this proof to the reader

17
4.2. Curvature in R3 . Now, the manifolds that we will explore exist in R3 , but
locally ‘look’ like R2 . The concept of immersion crystallizes this idea.
Definition 4.14. Given a manifold M n an immersion of a manifold M n → Rm
for some m ≥ n is a smooth function f : M n → Rm such that the differential dfp is
injective for all p ∈ M .
We will only be dealing with the immersion of 2-dimensional manifolds into R3 .
For M 2 let f : M 2 → R3 be an immersion. We then define the metric at p ∈ M 2
for y, z ∈ Tp M 2 as:
hy, zip := hdfp (y), dfp (z)if (p)

The left hand side is defined by the right hand side, which is the inner product of
R3 . Now for each p ∈ M 2 , there exists a neighborhood U containing p such that
f |U is an immersion. Let f (U ) = V . We can find a moving frame {e1 , e2 , e3 } for
V , such that e1 , e2 are tangent to V , and e3 is normal to V . We can accordingly
generate the coframe and the connected forms as given in (3.12-7).

We can also create pull-back maps, f ∗ (ωi ) and f ∗ (ωij ) and create new structure
equations of M 2 . Observe that f ∗ ω3 = 0, because, for y ∈ Tp M,
f ∗ ω3 (y) = ω3 (dx(y)) = ω3 (a1 e1 + a2 e2 ) = 0
Notice, that we can now use both 3.14 to satisfy the conditions necessary for Car-
tan’s lemma, giving us9
ω3 = 0 = ω13 ∧ ω32
d(0) = 0 = ω1 ∧ ω13 + ω2 ∧ ω23
Therefore,
ω13 = h11 ω1 + h12 ω2
ω23 = h21 ω1 + h22 ω2
with hij being differentiable real functions on U . Now, we know that de3 = ω13 e1 +
ω32 e2 , which implies that
 
h11 h12
de3 =
h21 h22
We can think about the map e3 as the orientation of the immersion, as each vector
in the field points perpendicularly to the manifold. Therefore, de3 is a map which
describes the orientation of the planes that are tangent to the manifold. This is
what motivates definitions of curvature.
 
h11 h12
Definition 4.15. Given p ∈ M , such that de3 (p) = − , the Gaussian
h21 h22
Curvature K of M in p is defined as
det(de3 ) = h11 h22 − h12 h21
= h11 h22 − h212

9For simplicity’s sake, we will now write f ∗ (ω ) = ω because U is imbedded in R3 by f , so


i i
essentially U ⊂ R3 .
18
This is due to the fact that h12 = h21 . By manipulation of the structure equations
in R3 , it follows that
ω12 = −(h11 h22 − h212 )(ω1 ∧ ω2 ) = −K(ω1 ∧ ω2 )
From the definition of curvature, we can prove an important theorem:
Theorem 4.16. Given M 2 , if f, f 0 : M 2 → R3 are immersions that both have the
same induced metrics with respective curvature functions of K and K 0 , then for all
p ∈ M , K = K 0 . In other words, the curvature is only dependent on the metric,
and not the immersion function.
Proof. We denote with a prime all of the entities related to f 0 . Now, for p ∈ U ⊂ M
we can find a moving frame {e1 , e2 }. It follows that {df (e1 ), df (e2 )} is the moving
frame of f (U ) = V , and {df 0 (e1 ), df 0 (e2 )} for f 0 (U ) = V 0 . Due to (3.13), ω1 =
ω10 ,ω2 = ω20 and ω12 = ω12
0
. Therefore
−K(ω1 ∧ ω2 ) = −K 0 (ω10 ∧ ω20 )

Notice that if v1 , v2 are linearly independent vectors at M , then ω1 ∧ω2 (v1 , v2 ) =
area(v1 , v2 ). That is why ω1 ∧ ω2 is known as the area element, as it generates the
area of the parallelogram generated by the two vectors.
Note that, given a moving frame, there must exist a unique antisymmetric 1-form
ω12 that obeys the structure of equations in R3 . Just take
ω12 (e1 ) = dω1 (e1 , e2 )
ω12 (e2 ) = dω2 (e1 , e2 )
One can check that it satisfies the necessary properties.
Our next goal is to show that this concept of curvature is intrinsic to the proper-
ties of the manifold, in particular, that it is not dependent on the choice of moving
frame. First, we must consider the choice of moving frames. if {e1 , e2 } and {e1 , e2 }
are both frames with the same orientation, then
e1 = f e1 + ge2
e2 = −ge1 + f e2
Where f, g : M → R are differential functions such that f 2 + g 2 = 1. This result
comes from both linear algebra and analysis.
Lemma 4.17. At a point p ∈ M , if {e1 , e2 } and {e1 , e2 } are both frames with the
same orientation, then
ω12 = ω12 − τ
Where τ = f dg − gdf . Further, if K and K are the respective curvatures of the
frames, then K = K. This shows that the curvature does not depend on the choice
of frame.
Proof. As a result of f, g as defined before the lemma,
ω1 = f ω1 − gω2
ω2 = gω1 + f ω2
Differentiating ω1 , we have that
dω1 = df ∧ ω1 + f dω1 − dg ∧ ω2 − gω2
19
By structure of equations in R3 , and because ω 12 = −ω 21 ,
dω1 = ω12 ∧ ω2 + (f df + gdg) ∧ ω1 + (gdf − f dg) ∧ ω2
But since f + g 2 = 1, differentiating gives us that f df + gdg = 0 and therefore:
2

dω1 = (ω 12 − τ ) ∧ ω2
Similarly for ω2 , we find that
dω2 = −(ω 12 − τ ) ∧ ω1
Combining these equations gives us:
ω12 = ω 12 − τ
Now, we can calculate that dτ = 0, which gives us that dω12 = dω 12 . Therefore,
K(ω 1 ∧ ω 2 ) = K(ω1 ∧ ω2 )
and so K = K. 

Technically τ is the differential map of the angle function between the two frames
ei and ei . Intuitively, τ measures the rate of change of the angle between the two
frames. Now that we have shown that the Gaussian curvature of a manifold is
remarkably independent, we can discuss the Gauss-Bonnet Theorem.

5. The Gauss-Bonnet Theorem


We will only prove the Gauss-Bonnet theorem for two-dimensional manifolds
immersed in R3 . First, we prove an equivalent theorem, known as the Poincaré-
Hopf index theorem, and show its equivalence to Gauss Bonnet. Both theorems
relate topological properties of the manifold to its geometric properties. With
regards to topology, we would like to define the index of a vector field at a point
on a manifold.
Definition 5.1. Consider a differential vector field X defined on M . We define
p ∈ M as a singular point if X(p) = 0. Further, p is isolated if there exists a
neighborhood Vp ⊂ M containing p that contains no other singular point.
The number of such isolated points is finite, since M is compact. We also choose
V to be homeomorphic to a disk in R2 , because integration is easier. We will now
develop a topological property corresponding to an isolated point.
Proposition 5.2. Consider a differential vector field X defined on M and consider
the set of isolated points {q ∈ M | X(q) = 0}. Now, we define the moving frame
X(p)
at Vq \ q such that (e1 )p = kX(p)k and (e2 )p as orthogonal to e1 and preserving the
orientation of M . Now, arbitrarily choose another moving frame {e1 , e2 }. There
will consequently exist two sets of connection forms and coframes for each of the
moving frames. From lemma 3.22, ω 12 − ω12 = τ . Thus, for any closed curve C
bounding a compact region of V containing p:
Z
τ = 2πI
C

Where I is known as the index of X at p.


20
In order to see why this is true, note that as we start at p and travel around the
curve, each ei and ei must end up in the same place after any full rotation. This
implies that each moving frame performs a rotation of some integer of 2π. Since
ei always points in the direction of the vector field, as the vector field ‘rotates,’ so
does ei . Now, since τ is the differential of the angle between ei and ei , integrating
τ along a closed curve would have to give us some integer multiple of 2π, because
it would be the difference in rotations between ei and ei , which are each multiples
of 2π.
Example 5.3. Below are a few examples of vector fields with different indices at
isolated singularities:

Note that, when defining I, we chose the frame {ei }, a Riemannian metric, and
the closed curve C. We will now show that I does not depend on these choices.
Lemma 5.4. The index of X at an isolated point p ∈ M does not depend on the
closed curve C that contains a compact subset of Vp .
R
Proof. Take two such closed curves, label them C1 , C2 . Let C1 τ = 2πI1 and
R
C2
τ = 2πI2 , and denote ∆ as the region bounded by the two curves. By calcula-
tion, we have
Z Z
2π(I1 − I2 ) = τ− τ
ZC1 C2

= τ
Z∂∆
= dτ (Stokes’ Theorem)

=0 (dτ = 0)
Therefore, I1 = I2 . If C1 and C2 intersect, we can choose another curve C3 that
does not intersect either curve, and apply the above method. 
Now, we show that the index is independent of the choice of frame {e1 , e2 } in
the following way.
21
Lemma 5.5. Suppose we are given a vector field X and {ei }. Consider B(r, p),
or in other words, a disk of radius r centered at an isolated point p. Let S(r, p) =
∂B(r, p). The following relation must hold:
Z
1
lim ω 12 = I
r→0 2π S(r,p)

Proof. First, we must prove that such a limit exists. Choose an arbitrary sequence
Z Z
ω 12 , . . . , ω 12 . . .
S(r1 ,p) S(rn ,p)

such that lim {rn } = 0. By Stokes’ theorem,


n→0
Z Z Z
ω 12 − ω 12 = dω 12 = 0 as ri , rk → 0
S(ri ,p) S(rk ,p) B(ri ,p)\B(rk ,p)

So, the above sequence is a Cauchy sequence. Since an arbitrary sequence was
1
R
chosen, it must be that lim 2π ω exists. Let I denote this limit. Now,
S(r,p) 12
r→0
consider Z Z
ω 12 − ω 12
S(r1 ,p) S(r2 ,p)
R
for r1 , r2 > 0. Fix r1 and let r2 → 0. It follows that S(r2 ,p) ω 12 = 2πI, and so by
Stokes’ Theorem
Z Z
ω 12 − 2πI = dω 12
S(r1 ,p) B(r1 ,p)
Z Z
ω 12 = − K(ω 1 ∧ ω 2 ) + 2πI
S(r1 ,p) B(r1 ,p)

From Lemma 3.22, we have that ω 12 = ω12 + τ , and so


Z Z Z
ω 12 = ω12 + τ
S(r1 ,p) S(r ,p) S(r1 ,p)
Z 1
= dω12 + 2πI
B(r1 ,p)
Z Z
− K(ω 1 ∧ ω2 ) + 2πI = − K(ω1 ∧ ω2 ) + 2πI
B(r1 ,p) B(r1 ,p)

Lemma 3.22 also tells us that K(ω 1 ∧ ω 2 ) = K(ω1 ∧ ω2 ), and so I = I. 


Thus, Lemma 3.22 is heavily responsible for proving the independence of the
index from the choice of moving frame.
Lemma 5.6. The index is not dependent on the metric of M
Proof. Let h, i1 and h, i2 be two arbitrarily chosen metrics on M . Define a function
dependent on t ∈ [0, 1] such that
h, it = th, i1 + (1 − t)h, i0
It can be seen that h, i is a valid inner product on M that varies smoothly with p.
Let I0 , It , I1 be the respective indices. From the previous two lemmas, we can see
22
that It is a smooth function. Since It can only be an integer, it must be that It = c
for all t ∈ (0, 1). Therefore, due to continuity, I0 = It = I1 . 

Now we are ready to state and prove the Poincaré-Hopf Index Theorem in R3 .
Theorem 5.7. Consider an oriented differentiable compact manifold M 2 . Let X be
a differential vector defined field on M with isolated singularities p1 , . . . , pk , whose
respective indices are I1 , . . . , Ik . For all Reimannian metrics on M ,
Z X k
Kσ = 2πIi
M i

Where σ = ω1 ∧ ω2 is the area element.


S
Proof. As previously discussed, in M \ i {pi } consider the frame where (e1 )p =
X(p)
kX(p)k and e2 is perpendicular to e1 . Now consider the collection of balls Bi , where
each pi ∈ Bi , and Bi contains no other isolated point. Since ω 12 = −Kω 1 ∧ ω 2 , we
have
Z Z
S
Kω 1 ∧ ω 2 = − S
dω 12
M \( i Bi ) M \( i Bi )
Z
= S
ω 12 (Stokes’ Theorem)
i ∂Bi
k Z
X
= ω 12
i ∂Bi

In theSsecond step above, the changeSin sign is due to the fact that the orientation of
M \( i Bi ) is the exact opposite of i Bi . In other words, for each ball, the outside
of the ball has an orientation opposite to the inside of each ball.
R Now, from lemma
3, we know that for any frame and coframe {ω1 , ω2 } that M \(S Bi ) Kω 1 ∧ ω 2 =
R i
S
M \( i Bi )
Kω1 ∧ ω2 . Let ri denote the radius of each Bi . Due to lemma 4.4, it
must be that Z
lim ω 12 = 2πIi
ri →0 ∂Bi
Therefore, S
as all of the radii approach 0 we have (for the purposes of integrating)
that M \ ( i Bi ) = M , and so
Z Z k
X
Kω 1 ∧ ω 2 = Kω1 ∧ ω2 = 2πIi
M M i


Now, in order to prove Gauss-Bonnet, we introduce a seemingly new topolog-


ical concept of the Euler number of a 2-manifold that arises from the concept of
triangulation
Definition 5.8. Given a compact oriented 2-manifold M , a triangulation of the
manifold is a collection of curvilinear triangles {Ti }, such that
S
(1) i Ti = M
(2) For all i 6= j, Ti ∩ Tj is either a vertex, an edge, or empty
23
Example 5.9. For the surface of a sphere, D2 , a possible triangulation is taken
from [2] and is as follows:

Proposition 5.10. Every compact 2-manifold admits a triangulation10

Definition 5.11. Given a triangulation of a compact oriented 2-manifold M , let


V be the number of vertices, A be the number edges, and let F be the number of
triangles. We define the Euler number of the manifold to be χ(M ) = V − A + F

Example 5.12. For the triangulation of the sphere as given in example 5.9, we
have that V = 6, A = 8 and F = 4, so

χ(D2 ) = V − A + F = 2

At first glance, definition seems to be dependent on the given triangulation.


However, this is not true, as will be shown by the following lemma.

Lemma 5.13. Consider a compact oriented differentiable 2-manifold M . For all


triangulations on M , and for any vector field defined on M we have that
X
χ(M ) = Ii
i

P P
Above, i Ii is the sum of the indices of the vector field. Since we know that i Ii
is the same for all vector fields, it follows that χ(M ) must be the same for all
triangulations.
P
Proof. By our proof of the Poincaré-Hopf Theorem, we already know that i Ii is
the same for all vector fields. Therefore, we just need to prove the lemma for one
vector field. Given an arbitrary triangulation of M , we define a vector field such
that the field has isolated singularities with I = 1 at the vertices and midpoints
of the edges. Further, the field also has isolated singularities with I = −1 at the
centroid of each triangle. For example, given the subset of the triangulation:

10This proof is beyond the scope of this paper, a proof is given here [3]

24
Above, χ(M ) = 5 − 8 + 4 = 1. The vector field is then defined as follows:

− 5 and 14 − 17 have an index of 1, while 6 − 13 have an


As we can see, points 1P
index of −1. Therefore i Ii = 1 = χ(M ), proving the lemma. 

Now, the proof of the Gauss-Bonnet theorem is almost immediate.

Theorem 5.14. Consider an oriented differentiable compact manifold M 2 . For


any Riemannian metric on M ,
Z
Kσ = 2πχ(M )
M
R Pk
Proof. By the Index theorem, we already have that M Kσ = i 2πIi . By the
Pk R
previous lemma, we have that i 2πIi = 2πχ(M ), so M Kσ = 2πχ(M ). 
R
Now, we can more easily calculate the calculate M
Kσ for many objects.
25
Example
R 5.15. For the sphere, as we had previously shown, χ(D2 ) = 2, so
D2
Kσ = 4π.

6. Conclusion
The Gauss-Bonnet theorem surprisingly implies that the Euler number does not
depend on the vector field X, and the integral of the curvature with respect to the
area element does not depend on the metric. Therefore, each of these concepts are
inherent to the structure of the manifold.
Another interesting consequence of this theorem is that two compact manifolds
without boundary M 2 and N 2 are diffeomorphic, in other words, there exists a
smooth function with a smooth inverse between M 2 and N 2 , if and only if χ(M 2 ) =
χ(N 2 ). In other words, if we can smoothly identify one manifold with another, then
their Euler numbers are the same. This proof is primarily due to Shiing Shen Chern,
who in fact used this technique to prove a generalized version of the Gauss-Bonnet
theorem that holds for higher dimensional manifolds.
Acknowledgments. It is a pleasure to thank my mentor, Ronno Das, for his
invaluable help in learning the material and writing this paper. I would also like
to greatly thank Professor May for editing my paper and giving me the chance to
participate in the apprentice program.

References
[1] Manfredo do Carmo, Differential Forms and Applications, Springer 1971
[2] Andrew Pressley, Elementary Differential Geometry, Springer-Verlag 2012
[3] Yiwang Chen, Triangulations of surfaces, https://faculty.math.illinois.edu/ ruiloja/Math519/chen.pdf

26

You might also like