A PROOF OF THE GAUSS-BONNET THEOREM

AARON HALPER
Abstract. In this paper I will provide a proof of the Gauss-Bonnet Theorem.
I will start by briey explaining regular surfaces and move on to the rst
and second fundamental forms. I will then discuss Gaussian curvature and
geodesics. Finally, I will move on to the theorem itself, giving both a local and
a global version of the Gauss-Bonnet theorem. For this paper, I will assume
that the reader has a knowledge of point-set topology, analysis in R
n
, and
linear algebra.
Contents
1. Introduction 1
2. Regular Surfaces 1
3. The First Fundamental Form 3
4. Orientation 4
5. The Gauss Map 4
6. The Second Fundamental Form 6
7. geodesics 6
8. The Local Gauss-Bonnet Theorem 9
1. Introduction
The Gauss-Bonnet theorem relates the sum of the interior angles of a triangle
with the its Gaussian curvature, an intrinsic quantity of the geometry of the space
that the triangle is drawn on. The theorem has numerous applications within and
without its native eld of dierential geometry. In order to understand the Gauss-
Bonnet theorem we must rst understand some basic dierential geometry. To this
end, we start with the most basic idea in dierential geometry, a regular surface.
2. Regular Surfaces
Denition 2.1. A subset S R
3
is a regular surface if for every point p S there
is a neighborhood V R
3
and a function f which maps an open set U R
2
onto
V S R
3
which has the following properties:
(1) f is dierentiable.
(2) f is a homeomorphism.
(3) For every q U, the dierential df
q
: R
2
R
3
is one-to-one.
Throughout this paper we denote the partial derivative of f with respect to u as
f
u
=
f
u
. Note that f is a vector-value function.
Date: DEADLINE AUGUST 22, 2008.
1
2 AARON HALPER
We will discuss (3), often called the regularity condition, once we have the fol-
lowing denition.
Denition 2.2. The function f dened as above is called a parametrization in
a neighborhood of p. The neighborhood V S of p S is called a coordinate
neighborhood.
Condition (3) is more familiar if we compute the matrix of the linear map df
q
in the standard, or canonical bases e
1
= (1, 0), e
2
= (0, 1) of R
2
with coordinates
(u, v) and f
1
= (1, 0, 0), f
2
= (0, 1, 0), f
3
= (0, 0, 1) of R
3
, with coordinates (x, y, z).
Let q = (u
0
, v
0
). Then e
1
is tangent to the curve u (u, v
0
) where the image
of this curve under f becomes the curve
u (x(u, v
0
), y(u, v
0
), z(u, v
0
)).
This image curve lies on our surface S and at f(q) has the tangent vector
(
x
u
,
y
u
,
z
u
) =
x
u
.
Here we compute the derivatives at (u
0
, v
0
) and a vector is indicated by its compo-
nents in the basis f
1
, f
2
, f
3
. By the denition of dierential,
df
q
(e
1
) = (
x
u
,
y
u
,
z
u
) =
x
u
.
Similarly, we nd
df
q
(e
2
) = (
x
v
,
y
v
,
z
v
) =
x
v
.
Combining this into a single matrix, we see that
df
q
=
_
_
_
_
_
_
_
_
_
_
x
u
x
v
y
u
y
v
z
u
z
v
_
_
_
_
_
_
_
_
_
_
Condition (3) is equivalent to requiring that the two column vectors of the above
matrix be linearly independent, or that the vector product
x
u

x
v
= 0,
or nally that one of the Jacobian determinants
(x, y)
(u, v)
=

x
u
x
v
y
u
y
v

,
(y, z)
(u, v)
,
(x, z)
(u, v)
.
is non-zero.
We dened surfaces as subsets of R
3
. We do this by covering a surface S in R
3
with
embeddings of open sets in R
2
or charts. Condition (1) is very natural if we are
A PROOF OF THE GAUSS-BONNET THEOREM 3
to do dierential geometry on S. Condition (2) makes sure charts only overlap in
2dimensional subsets, so that the tangent plane at a point is unique. We will now
show that condition (3) guarantees that the set of tangent vectors to parametrized
curves of S at a point p makes up a plane.
Denition 2.3. A tangent vector to S at a point p S is the tangent vector

(0)
of a dierentiable parametrized curve : (, ) S with (0) = p.
Proposition 2.4. Let f : U R
3
S be a parametrization of a regular surface
S and let q U. The vector subspace of dimension 2,
df
q
(R
2
) R
3
,
coincides with the set of tangent vectors to S at f(q).
We will not prove this proposition, but we will note that the plane df
q
(R
2
),
which passes through f(q) = p, does not depend on the parametrization f. We will
denote this plane the tangent plane to S at p and write it as T
p
(S). It is easy to
see that f
u
and f
v
span T
p
S.
3. The First Fundamental Form
Besides dierentiability, surfaces carry further geometric structures, the most
important of which is called the rst fundamental form.
Denition 3.1. By restricting The natural inner product , on R
3
to each tan-
gent plane T
p
(S) of a regular surface S, we get an inner product on T
p
(S). We
call this inner product on T
p
S the rst fundamental form and denote it by I
p
. So
I
p
v, w = v, w.
Thus the rst fundamental form tells us how the surface S inherits the natural
inner product of R
3
. We want to write it in terms of {f
u
, f
v
}, a basis associated to
a parametrization f(u, v) at p. To do this, we remember that a tangent vector w
T
p
(s) is the tangent vector to a parametrized curve (t) = f(u(t), v(t)), t (, )
with p = (0) = f(u
0
, v
0
). Then, keeping in mind that u

and v

are the respective

derivatives of u and v,
I

(0),

(0) = f
u
u

+ f
v
v

, f
u
u

+ f
v
v

= f
u
, f
u
(u

)
2
+ 2f
u
, f
v
u

+f
v
, f
v
(v

)2
= E(u

)
2
+ 2Fu

+ G(v

)
2
where we dene
E(u
0
, v
0
) = f
u
, f
u
(3.2)
F(u
0
, v
0
) = f
u
, f
v
(3.3)
G(u
0
, v
0
) = f
v
, f
v
. (3.4)
To round o the section, we give the following
Denition 3.5. A parametrization is orthogonal if F(u, v) = 0.
4 AARON HALPER
4. Orientation
Denition 4.1. A regular surface S is orientable if it is possible to cover S with a
family of coordinate neighborhoods so that if a point p S is in two neighborhoods
of this family, then the change of coordinates has positive Jacobian at p. The choice
of family that satises this condition is called an orientation of S, and S is called
oriented. If it is not possible to nd such a family then S is called nonorientable.
Given a parametrization f(u, v) at p, we have a denite choice of a unit normal
vector N at p by the rule
N(p) =
f
u
f
v
|f
u
f
v
|
(p).
(It may worth to say that the normal space is 1dimensional.)
Taking a second parametrization f

(u

, v

) at p, we see that
f

u
f

v
= (f
u
f
v
)
(u, v)
(u

, v

)
,
where
(u, v)
(u

, v

)
is the Jacobian of the coordinate change. From this we can see
that N will not change its direction if the Jacobian is positive, and change its
direction if the Jacobian is negative. So we can see from this that a surface is
orientable if N keeps its direction no matter how it is moved around the surface.
Example 4.2. On the M obius strip, we cannot nd a dierentiable eld of unit
normal vectors that are dened on the entire surface. Intuitively, we can see this
by taking a vector eld N around the middle circle of the gure and noticing that
it would come back as N, which contradicts the continuity of N. This is because
we cannot decide which side of the surface we are on since we can go continuously
to the other side without breaking the surface.
5. The Gauss Map
We denote S
2
as the unit sphere, i.e. S
2
= {(x, y, z) R
3
| x
2
+ y
2
+ z
2
= 1}.
Denition 5.1. Let S R
3
be a surface with an orientation N. The Gauss map
is dened to be N : S S
2
R
3
is dened to be p N(p).
It is easy to see that the Gauss map is dierentiable. The dierential dN
p
is a
linear map from T
p
(S) to T
N(p)
(S
2
). We can identify T
p
S and T
N(p)
S since they
are parallel planes, hence dN
p
is a linear map on T
p
(S). We now are ready for the
following
Proposition 5.2. The dierential dN
p
: T
p
(S) T
p
S of the Gauss-map is self-
adjoint.
Proof. We need to show that dN
p
(w
1
), w
2
= w
1
, dN
p
(w
2
) for a basis {w
1
, w
2
}
of T
p
(S). To do this, let f(u, v) be a parametrization of S at p and {f
u
, f
v
} the
associated basis of T
p
(S). If (t) = f(u(t), v(t)) is a parametrized curve in S, with
(0) = p, we get
A PROOF OF THE GAUSS-BONNET THEOREM 5
dN
p
(

(0)) = dN
p
(f
u
u

(0) + f
v
v

(0)
=
d
dt
N(u(t), v(t))

t=0
= N
u
u

(0) + N
v
v

(0);
in particular, dN
p
(f
u
) = N
u
and dN
p
(f
v
) = N
u
. Thus, in order to prove that
dN
p
is self-adjoint, we only need to show that
N
u
, f
v
= f
u
, N
v
.
We can see this by taking the derivatives of N, f
u
= 0 and N, f
v
= 0, relative
to v and u respectively, and get
N
v
, f
u
+N, f
uv
= 0
N
u
, f
v
+N, f
vu
= 0.
Hence,
N
u
, f
v
= N, f
uv
= N
v
, f
u
.

For a parametrization f(u, v) at a point p S with (t) = f(u(t), v(t)) a

parametrized curve on S, with (0) = p, the tangent vector to (t) at p is

= f
u
u

+ f
v
v

, and
dN(

) = N

(u(t), v(t)) = N
u
u

+ N
v
v

.
But N
u
and N
v
belong to T
p
(S), so we can write them in terms of our parameters
N
u
= a
11
f
u
+ a
21
f
v
, (5.3)
N
v
= a
12
f
u
+ a
22
f
v
, (5.4)
hence
dN(

) = (a
11
u

+ a
12
v

)f
u
+ (a
21
u

+ a
22
v

)f
v
,
which gives us
dN
_
u

_
=
_
a
11
a
12
a
21
a
22
__
u

_
Thus, in the basis {f
u
, f
v
}, dN is given by the matrix (a
ij
); i, j = 1, 2
Denition 5.5. The determinant of dN
p
is the Gaussian curvature, K, of S at a
point p.
6 AARON HALPER
6. The Second Fundamental Form
Now that we have the self-adjoint, linear map dN
p
, we can associate with it a
quadratic form which we imaginatively call the second fundamental form.
Denition 6.1. The quadratic form II
p
is dened in T
p
(S) by
II
p
(v) = dN
p
(v), v, and is called the second fundamental form of S at p.
Just like with the rst fundamental form, we now proceed to write the second
fundamental form in the basis {f
u
, f
v
}. Now
II
p
(

) = dN(

),

= N
u
u

+ N
v
v

, f
u
u

+ f
v
, v

= e(u

)
2
+ 2fu

+ g(v

)
2
,
where
e = N
u
, f
u
= N, f
uu
, (6.2)
f = N
v
, f
u
= N, f
uv
= N
u
, f
v
, (6.3)
g = N
v
, f
v
= N, f
vv
, (6.4)
since N, f
u
= N, f
v
= 0. We can use the coecients e, f, g to nd the values
of a
ij
, giving us
f = N
u
, f
v
= a
11
F + a
21
G, (6.5)
f = N
v
, f
u
= a
12
E + a
22
F, (6.6)
e = N
u
, f
u
= a
11
E + a
21
F, (6.7)
g = N
u
, f
u
= a
12
F + a
22
G. (6.8)
Or in matrix form,

_
e f
f g
_
=
_
a
11
a
12
a
21
a
22
__
E F
F G
_
.
From this it is clear that
K = det(a
ij
) =
eg f
2
EGF
2
.
7. geodesics
Denition 7.1. Given a dierentiable vector eld w in an open U S with
p U, take a y T
p
(S). Consider the parametrized curve : (, ) U, with
(0) = p,

(0) = y, and let g(t), t (, ), be the restriction of of the vector eld

g to the curve . Then projecting
dw
dt
(0) onto T
p
(S) forms the covariant derivative
at p of the vector eld w relative to the vector y, which we denote by
Dw
dt
(0) or
(D
y
w)(p).
The covariant derivative is the vector eld analogue of the usual derivative in
the plane. It is easy to verify that the covariant derivative does not depend on the
choice of .
A PROOF OF THE GAUSS-BONNET THEOREM 7
Denition 7.2. A vector eld along a parametrized curve S is parallel if
Dw
dt
= 0 for all t I.
Denition 7.3. A nonconstant, parametrized curve : I S is geodesic at t I
if the eld of the tangent vectors

(t) is parallel along at t, i.e.

D

(t)
dt
= 0. is
called a parametrized geodesic if is geodesics for every t I.
Denition 7.4. Let w be a dierentiable eld of unit vectors along a parametrized
curve : I S on an oriented surface S. Since w(t), t I, is a unit vector eld,
dw
dt
(t) is normal to w(t), hence
Dw
dt
= (N w(t)).
The real number = (t), denoted by
_
Dw
dt

, is called the algebraic value of the

covariant derivative of w at t.
Denition 7.5. Let C be an oriented regular curve contained on an oriented
surface S, and let (s) be a parametrization of C, in an neighborhood of p S, by
the arc length s. The algebraic value of the covariant derivative
_
D

(s)
ds
_
= k
g
of

(s) at p is called the geodesic curvature of C at p.

Lemma 7.6. Let a, b be dierentiable functions in I with a
2
+ b
2
= 1 and
0
be
such that a(t
0
) = cos
0
and b(t
0
) = sin
0
. Then the function dened by
=
0
+
_
t
t0
(ab

ba

) dt
has the properties that cos (t) = a(t), sin (t) = b(t) for t I, and (t
0
) =
0
.
Proof. To prove this we need only show that
(a cos )
2
+ (b sin )
2
= 2 2(a cos + b sin)
is zero everywhere. Or in other words, that
A = a cos + b sin = 1
.
By using the fact that aa

= bb

and the denition of , we get

A

= a(sin)

+ b(cos )

+ a

cos + b

sin
= b

(sin )(a
2
+ b
2
) a

(cos )(a
2
+ b
2
) + a

cos + b

sin
= 0
This tells us that A(t) = constant, and since A(t
0
) = 1, we have proved the
lemma.
Lemma 7.7. Let v, w be dierentiable vector elds along the curve : I S,
with |w(t)| = |v(t)| = 1, t I. Then
_
Dw
dt
_

_
Dv
dt
_
=
_
d
dt
_
where is the function given in the previous lemma.
8 AARON HALPER
Proof. Take the vectors v = N v and w = N w. Then
w = (cos )v + (sin ) v, (7.8)
w = N w = (cos )N v + (sin )N v = (cos ) v (sin )v. (7.9)
Dierentiating (7.8) we get
w

= (sin )

v + (cos )v

+ (cos )

v + (sin ) v

Taking the inner product of this with (7.9) and using the fact that v, v

=
v, v = 0 we get
w

, w = (sin
2
)

+ (cos
2
)v

, v + (cos
2
) (sin
2
)v

, v
=

+ (cos
2
)v

, v (sin
2
)v

, v.
but v, v = v, v

, so
w

, w =

+ (cos
2
+ sin
2
)v, v

+v, v

.
Thus,
_
Dw
dt
_
=
_
Dw
dt
_
N w, w =
dw
dt
, w = w

, w =

+v

, v =
d
dt
+
_
Dv
dv
_
,
which proves the lemma.
Proposition 7.10. Let f(u, v) be an orthogonal parametrization of a neighborhood,
and w(t) a dierentiable eld of unit vectors along the curve f(u(t), v(t)). Then
_
Dw
dt
_
=
1
2

EG
_
G
u
dv
dt
E
v
du
dt
_
+
d
dt
where
0
is the angle from from f
u
to w in the given orientation.
Proof. Let e
1
=
fu

E
, e
2
=
fv

G
be the unit vectors tangent to the coordinate curves.
Then e
1
e
2
= N, and by lemma 7.7 we have
_
Dw
dt
_
=
_
d
dt
_
+
_
Dv
dt
_
(7.11)
where e
1
(u(t), v(t)) is the restriction of the eld e
1
to the curve f(u(t), v(t)). We
know that
De
1
dt
=
_
de
1
dt
, N e
1
_
=
_
de
1
dt
, e
2
_
= (e
1
)
u
, e
2

du
dt
+(e
1
)
v
, e
2

dv
dt
But since F = 0, we have f
uu
, f
v
=
1
2
E
v
, giving us
(e
1
)
u
, e
2
=
__
f
u

E
_
u
,
f
v

G
_
=
1
2
E
v

EG
.
Similarly,
A PROOF OF THE GAUSS-BONNET THEOREM 9
(e
1
)
v
, e
2
=
1
2
G
u

EG
.
Putting all of this back into (7.11), we get
_
Dw
dt
_
=
1
2

EG
_
G
u
dv
dt
E
v
du
dt
_
+
d
dt
,
exactly what we wanted.

8. The Local Gauss-Bonnet Theorem

Theorem 8.1 (Local Gauss-Bonnet). Given an orthogonal parametrization f :
U S of an oriented surface S, where U R
2
is homeomorphic to an open disk
and f is compatible with the orientation of S, let R f(U) be a simple region of S,
and let : I S be so that R = (I). If is positively oriented, parametrized
by arc length s, and if (s
0
), ..., (s
k
) and
0
, ...,
k
are respectively the vertices and
external angles of , then
k

j=0
_
sj+1
sj
k
g
(s) ds +
__
R
K d +
k

j=0

j
= 2
where k
g
is the geodesic curvature of the regular arcs of and K is the Gaussian
curvature of S.
Proof. Let u = u(s), v = v(s) be the expression of the parametrization of in the
parametrization f. By proposition 7.10 we have
k
g
=
1
2

EG
_
G
u
dv
ds
E
v
du
ds
_
+
d
j
ds
where
j
(s) is the dierentiable function which measures the positive angle from
f
u
to

(s) in [s
j
, s
j+1
]. By integrating the above expression in every interval
[s
j
, s
j+1
], and adding the results we obtain
k

j=0
_
sj+1
sj
k
g
(s) ds =
k

j=0
_
sj+1
sj
_
G
u
2

EG
dv
ds

E
v
2

EG
du
ds
_
ds +
k

j=0
_
sj+1
sj
d
j
ds
ds
Now, the Gauss-Green theorem states the following: If P(u, v) and Q(u, v) are
dierentiable functions in a simple region A R
2
, whose boundary is given by
u = u(s), v = v(s), then
k

j=0
_
sj+1
sj
_
P
du
ds
+ Q
dv
ds
_
ds =
__
A
_
Q
u

P
v
_
dudv.
Applying this theorem, we get
k

j=0
_
sj+1
sj
k
g
(s) ds =
__
f

1(R)
__
E
v
2

EG
_
v
+
_
G
u
2

EG
_
u
_
dudv+
k

j=0
_
sj+1
sj
d
j
ds
ds
Now since we have an orthogonal parametrization, i.e. F = 0,
10 AARON HALPER
__
f

1(R)
__
E
v
2

EG
du
dt
_
v
+
_
G
u
2

EG
_
u
_
dudv =
__
f

1(R)
K

EG dudv =
__
R
K d
And from topology we know the Theorem of Turning Tangents which tells us
that
k

j=0
_
sj+1
sj
d
j
ds
ds =

j=0
k(
j
(s
j+1
)
j
(s
j
)) = 2
k

j=1

j
Since the curve is positively oriented, the sign should be plus. Putting all of this
together gives us
k

j=0
_
sj+1
sj
k
g
(s) ds +
__
R
K d +
k

j=0

j
= 2

Theorem 8.2 (Global Gauss-Bonnet). Let R S be a regular region of an oriented

surface and let C
1
, ..., C
n
be the closed, simple, piecewise regular curves which form
the boundary R of R. Suppose that each C
j
is positively oriented and let
1
, ...,
p
be the set of all external angles of the curves C
1
, ..., C
n
. Then
n

j=1
_
Cj
k
g
(s) ds +
__
R
K d +
p

j=1

j
= 2(R)
where s denotes the arc length of C
j
, and the integral over C
j
means the sum of
integrals in every regular arc of C
j
. Also, = F E + V is the Euler-Poincare
Characterization where for a given triangulation, F denotes the number of faces, E
denotes the number of edges, and V denotes the number of vertices of the triangu-
lation.
Proof. By a theorem in Topology, we know that we can take a triangulation J of
the region R with the property that every triangle T
j
is contained in a coordi-
nate neighborhood of a family of orthogonal parametrizations compatible with the
orientation of S. By making the boundary of every triangle of mathfrakJ is op-
positely oriented, we get opposite orientations in the edges that adjacent triangles
share. To every triangle we apply the Local Gauss-Bonnet Theorem, add them up,
remembering that each interior side is described twice in opposite orientations,

j
_
Cj
k
g
(s) ds +
__
R
K d +
F,J

j,k=1

j,k
= 2F
4 where F denotes the number of triangles of J, and
j1
,
j2
,
j3
are the external
angles of the triangle T
j
.
The interior angles of the triangle T
j
we shall denote
jk
=
jk
. From this,
we see that

j,k

j,k
=

j,k

j,k

j,k
= 3F

j,k

j,k
.
I will now introduce the following notation:
A PROOF OF THE GAUSS-BONNET THEOREM 11
E
e
= number of external edges of J,
E
i
= number of internal edges of J,
V
e
= number of external vertices of J,
V
i
= number of exteranl vertices of J.
Since the curves C
i
are closed, E
e
= V
e
. It is also clear that
3F = 3E
i
+ E
e
hence

j,k

j,k
= 2E
i
+E
e

j,k

j,k
.
Now either the external vertices are vertices of some curve C
i
, which well call
V
ec
, or they are vertices introduced by the triangulation, which well call V
ej
. So
we can write V
e
= V
ec
+V
ej
. And the sum of the angles around each internal vertex
is 2, so we can write

j,k

j,k
= 2E
i
+E
e
2V
i
2V
ei

j
(
j
)
Adding and subtracting E
e
to the right hand side of this equation, and using
the fact that E
e
= V
e
, we have

j,k

j,k
= 2E
i
+ 2E
e
2V
i
V
ei
V
ec
+

j
= 2E 2V +

j
.
Finally, we can put it all back together again to get
n

j=1
_
Cj
k
g
(s) ds +
__
R
K d +
p

j=1

j
= 2(F E + V )
= 2(R),
which is exactly what we wanted to prove.