Chapter 27

The Gauss–Bonnet
Theorem
That the sum of the interior angles of a triangle in the plane equals π radians was
one of the first mathematical facts established by the Greeks. In 1603 Harriot1
showed that on a sphere of radius 1 the area of a spherical triangle (that is, a
triangle whose sides are parts of great circles) with angles α, β, γ is given by

α + β + γ − π.

For example, an octant of the unit sphere S 2 (1) has area π/2, and comprises the
interior of a triangle with angles all equal to π/2. Both the planar and spherical
facts are special cases of the Gauss–Bonnet2 Theorem, which will be proved in
this chapter.
The Gauss–Bonnet Theorem consists of a formula for the integral of the
Gaussian curvature over all or part of an abstract surface. For example, we
know that the curvature and area of a sphere S 2 (a) are 1/a2 and 4πa2 . Thus

1 1
Z Z
K dA = 2
dA = 2 4πa2 = 4π,
S 2 (a) S 2 (a) a a

1 Thomas Harriot (1560–1621). English mathematician, adviser to Sir Walter Raleigh. He

made a voyage to Virginia in 1585–86 and reported on the Native American languages and
customs. His telescopic observations were the earliest in England; in particular, he discovered
sunspots in 1610 and used them to deduce the period of the Sun’s rotation.
Pierre Ossian Bonnet (1819–1892). French mathematician, who made
many important contributions to surface theory, including the Gauss–
2 Bonnet Theorem. Bonnet was director of studies at the École Polytech-
nique in Paris, professor of astronomy in the faculty of sciences at the
University of Paris, and member of the board of longitudes.

901
902 CHAPTER 27. THE GAUSS–BONNET THEOREM

showing that the integral of the Gaussian curvature over a sphere is independent
of the radius a of the sphere. It is remarkable that for any ellipsoid, or more
generally for any convex surface M in R3 , we have
Z
K dA = 4π.
M

This fact derives ultimately from the presence of K in formulas for area in the
image of the Gauss mapping defined on page 333.
Still more generally, for any compact surface M, the Gauss–Bonnet Theo-
rem expresses the integral of the Gaussian curvature in terms of a topological
invariant, the Euler characteristic χ(M). This important number is always an
integer, and is expressible in terms of the number of vertices, faces and edges of
any triangulation of M.
We begin by generalizing to abstract surfaces the notions of turning angle
(given in Section 1.5) and of total signed curvature of a plane curve (given in
Section 6.1). This enables us to consider integrals of geodesic curvature in the
abstract setting, though the results of this chapter can also be appreciated by
using the extrinsic definitions given in Section 17.4.
In Section 27.2, we prove a local version of the Gauss–Bonnet Theorem,
using Green’s Theorem which is of independent interest. We improve the results
in Section 27.4. A summary of topological facts about surfaces is given in
Section 27.5 in order to establish global versions of the Gauss–Bonnet Theorem
in Section 27.6. Finally, Section 27.7 is devoted to applications.

27.1 Turning Angles and Liouville’s Theorem

Since in general there is no notion of ‘horizontal’ for a general abstract surface,
we must choose a reference vector to measure turning. In the next two lemmas,
we suppose that M is an oriented surface, α : (a, b) → M is a regular curve,
and that X is an everywhere nonzero vector field along α.

Lemma 27.1. Fix t0 with a < t0 < b. Let θ0 be a number such that
α0 (t0 ) X(t0 ) JX(t0 )
α0 (t0 ) = cos θ0 X(t0 ) + sin θ0 X(t0 ) .

Then there exists a unique differentiable function θ[α, X]: (a, b) → R such that
θ(t0 ) = θ0 and

α0 1

(27.1) = cos θ[α, X]X + sin θ[α, X] J X ,
α0 X

at all points on the curve. We call θ[α, X] the turning angle of α with respect
to X determined by θ0 and t0 .
27.1. TURNING ANGLES AND LIOUVILLE’S THEOREM 903

Proof. Corollary 1.24 on page 18 implies the existence of a unique function

θ[α]: (a, b) → R such that θ[α](t0 ) = θ0 and
α0 (t) · X(t)

 cos θ[α](t) = α0 (t) X(t) ,



(27.2)
α0 (t) · J X(t)
 sin θ[α](t) = α0 (t) X(t) .




Then (27.2) is equivalent to (27.1).

We can now generalize Lemma 1.26, page 20, to abstract surfaces.

Lemma 27.2. The turning angle θ[α, X] is related to the geodesic curvature
κg [α] of α by the formula


0
0 ∇α0 X, J X (t)
(27.3) θ[α, X] (t) = κg [α](t) α (t) −
.
X(t) 2

Proof. Using the analog of (1.23), we have

* ! +

α00 (t), J α0 (t)


α0 (t) J α0 (t)
= κg [α](t) α0 (t) .

(27.4) ∇α0 (t) 0 , 0 =
2
α (t) α (t)

0
α (t)

On the other hand, (writing α = α(t), X = X(t) and θ = θ[α, X]) we use
(27.1) to compute
 0   
α X cos θ + JX sin θ
∇α0 = ∇ 0
kα0 k α kXk
kXk0
 

−X sin θ + J X cos θ 0
= − X cos θ + J X sin θ + θ
kXk2 kXk


(∇α0 X) cos θ + (J ∇α0 X) sin θ
+
kXk


kXk0 α0 J α0 0 (∇α0 X) cos θ + (J ∇α0 X) sin θ
= − + θ + ,
kXk kα0 k kα0 k kXk
so that  0 

J α0 ∇α0 X, JX
 
α 0
(27.5) ∇α0 , =θ + .
kα0 k kα0 k kXk2
Now (27.3) follows from (27.4) and (27.5).
A practical choice for the vector field X is xu , where x is a patch on an ab-
stract surface. This will lead us to an analogue of Euler’s Theorem on page 397.
Suppose now that the surface M is oriented, and denote its metric by

(27.6) ds2 = E du2 + 2F dudv + Gdv 2 .

904 CHAPTER 27. THE GAUSS–BONNET THEOREM

We shall say (referring to page 879) that a patch x is coherent with the orien-
tation of M if the equation

−F xu + E xv
(27.7) J xu = √
EG − F2
holds, rather than one with an opposite sign.

Lemma 27.3. Let α : (a, b) → M be a regular curve whose trace is contained

in x(U), where x: U → M is a patch coherent with the orientation of M. Write
α(t) = x(u(t), v(t)) for a < t < b. Then

F
− (Eu u0 + Ev v 0 ) − Ev u0 + Gu v 0 + 2Fu u0
0
κg [α] kα k = θ[α, xu ] + 0 E √ .
2 EG − F2

Proof. We have α0 = u0 xu + v 0 xv ; thus from (27.7) it follows that


∇α0 xu , Jxu
 
1 0 0 −F xu + E xv
(27.8) = u ∇xuxu + v ∇xvxu , √ ,
kxu k2 E EG − F2

which equals

−u0 F ∇xuxu , xu + u0 E ∇xuxu , xv − v 0 F ∇xvxu , xu + v 0 E ∇xvxu , xv






√
divided by E E G − F 2 . Since ∇xuxu , xu = 12 Eu , ∇xvxu , xv = 12 Gu and




∇xuxu , xv = Fu − ∇xvxu , xu = Fu − 21 Ev ,




we see that (27.8) equals

 0
v0

1 u
√ − (F Eu − 2EFu + EEv ) + (−F Ev + EGu ) .
E EG − F2 2 2

Combining this equation with (27.3) completes the proof.

Corollary 27.4. With the hypotheses of Lemma 27.3, suppose in addition that
F = 0 in (27.6). Then

1
(27.9) κg [α] kα0 k = θ[α, xu ]0 + √ (Gu v 0 − Ev u0 ).
2 EG

Next, we find a formula for the geodesic curvature of a curve on a patch

x in terms of the geodesic curvatures of the coordinate curves u 7→ x(u, v)
and v 7→ x(u, v). We denote these geodesic curvatures by (κg )1 and (κg )2
respectively.
27.1. TURNING ANGLES AND LIOUVILLE’S THEOREM 905

Theorem 27.5. (Liouville3 ) Let M be an oriented surface, and suppose that

α : (a, b) → M is a regular curve whose trace is contained in x(U), where
x: U → M is a patch coherent with the orientation of M for which F = 0.
Then
θ0
(27.10) κg [α] = (κg )1 cos θ + (κg )2 sin θ + ,
kα0 k
where θ = θ[α, xu ] is the angle between α and xu .

Proof. Applying (27.9) to the curves u 7→ x(u, v) and v 7→ x(u, v), we obtain
√ Ev √ Gu
(κg )1 E = − √ and (κg )2 G = √ .
2 EG 2 EG
For a general curve α, we have
kα0 k kα0 k
u0 = √ cos θ and v 0 = √ sin θ,
E G
so that
Ev u0 Gu v 0
(27.11) − √ = kα0 k(κg )1 cos θ and √ = kα0 k(κg )2 sin θ.
2 EG 2 EG
Then (27.10) follows from (27.9) and (27.11).
Notice the deceptive similarity between (27.10) and Euler’s formula (13.17).

Total Geodesic Curvature

The analog of Definition 6.1 on page 154 for a curve on a surface is

Definition 27.6. Let M be an abstract surface. The total geodesic curvature of

a curve α : [a, b] → M, denoted by TGC[α], is the number
Z Z b
κg [α](t) α0 (t) dt,

(27.12) κg [α]ds =
α a

where κg [α] is the geodesic curvature of α.

The abstract definition of the geodesic curvature was given on page 880, though
if the surface lies in R3 , one may use the more elementary definition on page 541
to make sense of (27.12).

Joseph Liouville (1809–1882). French mathematician, who worked in num-

3
ber theory, differential equations, dynamics, differential geometry and
complex variables. Liouville played a large role in introducing Gauss’s
ideas to France.
906 CHAPTER 27. THE GAUSS–BONNET THEOREM

The generalization of Lemma 6.2 on page 154 is

Lemma 27.7. The total geodesic curvature of a curve on a surface remains

unchanged under a positive reparametrization, but changes sign under a negative
reparametrization.

The proofs of Lemmas 27.7 and 6.2 are essentially the same (see Exercise 8).
We also have the following generalization of Lemma 6.3, obtained by integrating
(27.3):

Lemma 27.8. Let M be an oriented surface, α: (a, b) → M a regular curve,

and X be an everywhere nonzero vector field along α. Then the total geodesic
curvature can be expressed by
Z b

∇α0X, JX
TGC[α] = θ[α, X](b) − θ[α, X](a) + dt.
a kXk2

27.2 The Local Gauss–Bonnet Theorem

First, we need some facts about piecewise-differentiable curves on surfaces. Let
M be an abstract surface, and let α: [a, b] → M be a piecewise-differentiable
curve. Associated with α is a subdivision

a = t 0 < t1 < · · · < tk = b

of [a, b] for which α is differentiable and regular on [tj , tj+1 ] for j = 0, . . . , k−1.
The appropriate left and right derivatives of α are required to exist at the end
points of each subinterval. The points α(t0 ), . . . , α(tk ) are called the vertices of
the curve α.
The following theorem is an immediate consequence of Corollary 27.4.

Theorem 27.9. Let M be an abstract surface, and suppose α: [a, b] → M is a

piecewise-differentiable curve whose trace is contained in x(U), where x: U → M
is a patch on M for which F = 0. Denote by α(t0 ), . . . , α(tk ) the vertices of α.
Then
k−1
X Z tj+1
κg [α](t) α0 (t) dt

TGC[α] =
j=0 tj

k−1
X Z tj+1  k−1
G dv Ev du
√u
X
+
θ−


= − √ dt + j+1 − θ j ,
j=0 tj 2 E G dt 2 E G dt j=0
where

θ− θ+



j = lim θ[α, xu ] α(t) and j = lim θ[α, xu ] α(t) .
t ↑ tj t ↓ tj
27.2. LOCAL GAUSS–BONNET THEOREM 907

The numbers θ + j are only well defined up to integer multiples of 2π, although
+ +
θj completely determines the number θ− −
j+1 . Typically, εj = θ j − θ j is called
the exterior angle of α at α(tj ) and ιj = π − εj the interior angle at α(tj ),
though we require 0 < ιj < 2π for these definitions to be valid.
In Section 12.4 we defined the notion of area of a closed subset of a surface;
that notion admits the following generalization:
Definition 27.10. Let M be an abstract surface, x: U → M an injective regular
patch, S ⊆ x(U) a closed subset, and f : S → R a continuous function. Then
the integral of f over S is given by
ZZ ZZ p
f dA = (f ◦ x) E G − F 2 dudv.
S x−1 (S)

Just as in Section 12.4, we need to show that this definition is geometric. The
proof of the following lemma is similar to that of Lemma 12.17 on page 374 (see
Exercise 1).
Lemma 27.11. The definition of the integral of f is independent of the choice
of patch.
We make precise the meaning of some common words from general topology.
A neighborhood of a point p of an abstract surface M is any open set containing
p. The boundary of a subset S of M is the set, denoted ∂S, consisting of points
p ∈ M for which every neighborhood of p contains both a point in S and a
point not in S. A region of M is the union of an open connected subset with
its boundary.
A key tool for proving the Gauss–Bonnet theorem is Green’s Theorem. (See,
for example, [Buck, page 406].) This can be stated as follows.
Theorem 27.12. (Green4 ) Let R ⊂ R2 be a simply connected region, and let
P, Q : U → R be differentiable functions, where U is an open set containing R.
Then ZZ  
∂Q ∂P
Z
(27.13) − dudv = P du + Q dv.
R ∂u ∂v ∂R

The line integral is performed by parametrizing the boundary in a counter-

clockwise sense in the plane. To check this is correct, we may merely take
(P, Q) = (0, u) and R = {0 6 u 6 1, 0 6 v 6 1}, so that (27.13) becomes
ZZ Z
1 dudv = u dv,
R ∂R

4 George Green (1793–1841). English mathematician, owner of a windmill in Nottingham.

Although he published only 10 papers, he made profound contributions to mathematical
physics. He invented Green’s functions and coined the term ‘potential function’. Where
Green acquired his mathematical skills remains a mystery.
908 CHAPTER 27. THE GAUSS–BONNET THEOREM

and both sides are equal to 1. More generally, the boundary of a region of an
abstract surface will be traversed in the sense determined by the operator J
defining the orientation.
We are now ready to prove the first version of the Gauss–Bonnet Theorem.

Theorem 27.13. (Gauss–Bonnet, local version) Let M be an oriented abstract

surface and let x: U → M be a regular patch coherent with the orientation of
M. Let R = x([a, b] × [c, d]), where [a, b] × [c, d] ⊆ U. Then
ZZ Z 3
X
(27.14) K dA + κg ds + εj = 2π,
R ∂R j=0

where ε0 , ε1 , ε2 , ε3 are the exterior angles at the vertices p0 , p1 , p2 , p3 of R.

Proof. First, suppose that x is a patch with F = 0. We suppose that the

vertices are chosen so that

p0 = x(a, c), p1 = x(b, c), p2 = x(b, d), p3 = x(a, d).

The portion of ∂R between p0 and p1 is then parametrized by u 7→ x(u, c), and

we may choose
θ+ −
0 = θ 1 = 0.

We explain next that the hypothesis that x be consistent with the orientation
ensures that the external angle at each vertex lies between 0 and π.

Ε2
Ε3

Ε1

Ε0

Figure 27.1: A quadrilateral with exterior angles i

27.2. LOCAL GAUSS–BONNET THEOREM 909

On the portion of ∂R between p1 and p2 , we deduce from (27.1) and (27.7)

that the turning angle θ = θ[xv , xu ] of the boundary relative to xu satisfies
√ √
E sin θ = hxv , J xu i = E G.
We are therefore free to choose 0 < sin θ+
1 < π, so that the internal angle ι1 at
p1 also lies between 0 and π. Continuing around the quadrilateral, the portion
of ∂R between p2 and p3 is parametrized by u 7→ x(u, d), and we need to choose
θ+ −
2 = θ3 = π and π < θ+ −
3 , θ 4 < 2π.

The resulting exterior angles are



 ε1 = θ+ −
1 − θ1 = θ+
1,

θ + − θ− = π − θ−

 ε2 =
2,

2 2


 ε3 = θ+ −
3 − θ3 = θ+
3 − π,

+
ε0 = θ0 − θ4 + 2π = 2π − θ−
−

4.


It is necessary to add 2π in the last line so that the internal angle ι0 = π − ε0

lies between 0 and 2π (indeed, 0 and π). Therefore,
3
X
+ + +
θ− θ− −




j+1 − θ j = 2 − θ1 + θ4 − θ3
j=0 3
X
= π − ε2 − ε1 + 2π − ε0 − ε3 − π = 2π − εj .
j=0

By Green’s Theorem (Theorem 27.12) and formula (17.6), page 534, we have
Z  
Gu dv Ev du
√ − √ dt
∂R 2 E G dt 2 E G dt
Z bZ d    
E G
= √v + √u dudv
a c 2 EG v 2 EG u
Z bZ d √ ZZ
=− K E G dudv = − K dA.
a c R

Substituting the equations above into Theorem 27.9, it follows that the total
geodesic curvature of ∂R is given by
Z X3 Z tj+1
κg [α](t) α0 (t) dt

κg ds =
∂R j=0 tj
3
X ZZ
= 2π − εj − K dA.
j=0 R

Since all terms in (27.14) are invariantly defined, the result holds for all
patches irrespective of the assumption F = 0.
910 CHAPTER 27. THE GAUSS–BONNET THEOREM

It will be convenient to rewrite (27.14) using interior angles.

Corollary 27.14. Under the hypotheses of Theorem 27.13 we have

ZZ Z 3
X
(27.15) K dA + κg ds = −2π + ιj ,
R ∂R j=0

where ι0 , ι1 , ι2 , ι3 are the interior angles at the vertices of R.

27.3 An Area Bound

This short section provides an independent application of Green’s theorem to the
area of surfaces of constant negative curvature in R3 , such as the pseudosphere
and other examples studied in Chapters 15 and 21. It uses terminology from
Section 21.2, in particular Definition 21.11 on page 690.

Lemma 27.15. Let D = [a, b] × [c, d], and let x: D → R3 be a Tchebyshef

principal patch of constant negative curvature −a−2 . Then
ZZ
dA 6 2πa2 .


(27.16) area y(D) =
D

Proof. Let ω denote the angle function of the associated asymptotic patch

y(p, q) = x(p + q, p − q),

described on page 686. Then the principal patch x has metric

ds2 = a2 (cos2 θ du2 + sin2 θ dv 2 ),

where θ = ω/2. The infinitesimal area element of y is given by

p √
dA = E G − F 2 dpdq = a4 − a4 cos2 ω dpdq
= a2 sin ω dpdq
= a2 ωpq dpdq,

the last line with the help of Lemma 21.2. By Theorem 27.12, we have
ZZ
2


area y(D) = a ωpq dpdq
D
2Z
a

= − ωp dp + ωq dq
2 ∂D

= a2 ω(a, c) + ω(b, d) − ω(a, d) − ω(b, c) .

(27.17)

Since neither cos θ nor sin θ can vanish, we can assume that 0 < θ < π/2, so
0 < ω(p, q) < π for all p, q. Hence area y(D) 6 2a2 π.


27.4. MORE COMPLICATED REGIONS 911

More generally, suppose that we can cover the surface M by one large rec-
tangle divided into a grid of smaller rectangular Tchebyshef principal patches of
the type above. When we sum the contributions arising in (27.17), these cancel
out in pairs, and we are left with only external vertices.

Corollary 27.16. Let M ⊂ R3 be a connected regular surface with constant

negative curvature −a2 . Then area(M) 6 2π a2 .

This result extends Theorem 26.24, since it follows that one cannot find any
complete surfaces of constant negative Gaussian curvature in R3 .

27.4 A Generalization to More Complicated Regions

In order to generalize Theorem 27.13, we introduce the following notion.

Definition 27.17. Let M be an abstract surface. A regular region is a compact

region whose boundary is the union of a finite number of piecewise-regular simple
closed curves that do not intersect each other.

Note that a compact regular surface can be considered as a regular region with
empty boundary.
Each of the piecewise-regular simple closed curves that constitute the bound-
ary ∂Q of a regular region Q has vertices and exterior angles; therefore, we can
speak of the vertices p1 , . . . , pm and the external angles ε1 , . . . , εm of the reg-
ular region Q. Let us also call the portion of ∂Q between consecutive vertices
an edge of Q.

Definition 27.18. A polygonal region is a regular region with a finite number

of vertices, each of which has a nonzero external angle. A triangular region is a
polygonal region with exactly 3 vertices.

It will be necessary to decompose regular regions into smaller regions.

Definition 27.19. A polygonal decomposition of a regular region Q of an ab-

stract surface is a finite collection R of polygonal regions R called faces such
that
(i) each R ∈ R is homeomorphic to a disk;

(ii) the union of all regions in R is Q;

(iii) if R1 , R2 ∈ R, then the intersection R1 ∩ R2 is either a common edge

or a common vertex of R1 and R2 .
An oriented polygonal decomposition of Q is a polygonal decomposition in which
the faces and edges are assigned orientations as follows:
912 CHAPTER 27. THE GAUSS–BONNET THEOREM

(iv) each face of R is assigned the orientation of M;

(v) each edge of a face R ∈ R, when parametrized as a curve α with respect
to R, is such that Jα0 points toward the interior of R.
If all the faces are triangular regions, then R is called a triangulation of Q.
Note that if γ is an edge of polygonal regions R1 and R2 , the orientation of γ
with respect to R1 is the opposite of the orientation of γ with respect to R2 .
As a first step to generalizing Theorem 27.13, we derive the formula for the
integral of the Gaussian curvature over a triangular region.
Corollary 27.20. Let M be an oriented abstract surface, and suppose that T
is a triangular region of M such that T ⊂ x(U), where x: U → M is a regular
patch coherent with the orientation of M. Then
ZZ Z
(27.18) K dA + κg ds = −π + ι1 + ι2 + ι3 ,
T ∂T
where ι1 , ι2 , ι3 are the interior angles at the vertices of T .
Proof. We construct an oriented polygonal decomposition of T , consisting of
quadrilaterals, as follows. Choose a point in the interior of the triangular region
and points on each of the edges between the vertices. Join the three new points
on the edges to the point in the center. Three quadrilaterals R1 , R2 , R3 are
formed; we choose them to have their orientations coherent with that of T .
Since T is contained in the trace of the patch x, the three quadrilaterals can be
chosen as images under x of ordinary quadrilaterals in R2 . Therefore, because
of additivity of integration over closed subsets of R2 and (27.15), we have
ZZ X 3 ZZ X 3 Z
(27.19) K dA = K dA = −6π + I − κg ds,
T k=1 Rk k=1 ∂Rk

where I is the sum of all interior angles.

The boundaries ∂R1 , ∂R2 , ∂R3 contribute six interior edges and six exterior
edges. The integrals Z
κg ds

cancel in pairs over the interior edges and combine in pairs over the exterior
edges; hence
X3 Z Z
(27.20) κg ds = κg ds.
k=1 ∂Rk ∂T

Included in the sum I are the interior angles ι1 , ι2 , ι3 , together with the
interior angles at the newly-created vertices. It is easily seen from Figure 27.2
that the interior angles at the newly created vertices add up to 5π, and so
(27.21) I = 5π + ι1 + ι2 + ι3 .
27.4. MORE COMPLICATED REGIONS 913

Now (27.18) follows from (27.19)–(27.21).

Ι3

R3

2Π
Π
Π
R2
R1

Π Ι2
Ι1

Figure 27.2: A triangle decomposed into three quadrilaterals

Now we can state the original version of the theorem that Gauss proved in
[Gauss2, Section 20].

Corollary 27.21. Suppose that the sides of the region T in Corollary 27.20 are
geodesics. Then ZZ
(27.22) K dA = −π + ι1 + ι2 + ι3 ,
T

where ι1 , ι2 , ι3 are the interior angles at the vertices of T .

When K = 0, we obtain the familiar theorem that the sum of the angles of
a triangle in the plane equals π. More generally:

Corollary 27.22. Suppose that the sides of the region T in Corollary 27.20 are
geodesics, and that K has the constant value λ 6= 0. Then the area of T is given
by
−π + ι1 + ι2 + ι3
(27.23) area(T ) = ,
λ
where ι1 , ι2 , ι3 are the interior angles at the vertices of T .
914 CHAPTER 27. THE GAUSS–BONNET THEOREM

When λ = 1, equation (27.23) reduces to Harriot’s formula for the area of

a spherical triangle. Thus on a unit sphere, the sum of the interior angles of
any geodesic triangle is greater than π, and the excess over π equals the area
of the triangle. By contrast, any triangle on a surface of constant Gaussian
curvature −1 has area strictly less than π. Using the upper half-plane model
(Figure 26.3 on page 883), we may choose three vertices on the real axis (thus
strictly speaking outside the upper half-plane) and join them by three semicir-
cles. The resulting asymptotic geodesic triangle has interior angles all zero and
area π relative to the Poincaré metric (26.6).

27.5 The Topology of Surfaces

In this section, we give a rudimentary account of the topology of surfaces. For
more information see [AhSa, Chapter 1], [Arm, Chapter 7], [Massey, Chapter 1],
[Springer, Chapter 5] and [FiGa]. The most important theorem we need is

Theorem 27.23. Let M1 and M2 be compact abstract surfaces. The following

conditions are equivalent:
(i) M1 and M2 are homeomorphic;

(ii) M1 and M2 are diffeomorphic;

(iii) M1 and M2 are both orientable or both nonorientable and have the same
Euler characteristic.

Suppose that (i) holds. A map Φ: M1 → M2 defining a homeomorphism will

not necessarily be a diffeomorphism; nonetheless the theorem asserts that a
diffeomorphism Ψ : M1 → M2 will exist.
Here is the definition of Euler characteristic:

Definition 27.24. Let R be a polygonal decomposition of a regular region Q of

an abstract surface. Put


 V = the number of vertices of R,

E = the number of edges of R,


F = the number of faces of R.


Then χ(Q) = F − E + V is called the Euler characteristic or Euler number of Q.

It is implicit in Theorem 27.23 that the Euler characteristic χ(M) of a

compact abstract surtface M does not depend on the polygonal decomposition
chosen. This fact really follows from the array of propositions below, details of
which can be found in the books cited on the previous page.
27.5. TOPOLOGY OF SURFACES 915

Proposition 27.25. The Euler characteristic of a regular region of an abstract

surface is invariant under the following processes:
(i) subdivision of an edge by adding a new vertex at the interior point of an
edge;

(ii) subdivision of a polygonal region by connecting two vertices by a new edge;

(iii) introducing a new vertex into a polygonal region and connecting it by an

edge to one of the vertices of the polygonal region.

The proof of proposition 27.25 is instructive and not difficult (see Exercise 2).
Clearly, it can be used to prove that any two polygonal decompositions of a
regular region of an abstract surface have the same Euler characteristic, provided
each edge of one of the two decompositions intersects each edge of the other in
only a finite number of points and a finite number of closed intervals. However,
it can happen that two polygonal decompositions intersect each other in bizarre
ways. Nevertheless:

Proposition 27.26. The Euler characteristic of a regular region of an abstract

surface is independent of the polygonal decomposition used to define it.

Proposition 27.27. A regular region of an abstract surface always admits a

triangulation.

If we take these results for granted, the classification is best understood by

listing examples in the table below. The sphere S 2 and torus T 2 are orientable
surfaces in R3 , unlike the projective plane RP2 and Klein bottle K that we met
in Chapter 11.
A polygonal decomposition of the sphere is obtained by dividing it into eight
triangles of the type mentioned on page 901. This results in 6 vertices, 12 edges
and 8 regions, whence
χ(S 2 ) = 6 − 12 + 8 = 2.

We can do the same job with far fewer regions, but Proposition 27.26 tells us
that the Euler characteristic will not change.
The Euler characteristics of T 2 , RP2 , K can be deduced from the diagrams
in Section 11.2, by merely taking F = 1 face, corresponding to the square. The
number of vertices and edges will depend on how the perimeter of the square
has been identified with itself. For example Figure 11.4 on page 337 represents
both the torus and Klein bottle with 1 vertex and 2 edges, whence

χ(T 2 ) = χ(K) = 1 − 2 + 1 = 0.

The difference of course is that T 2 is orientable, whereas K is not.

916 CHAPTER 27. THE GAUSS–BONNET THEOREM

M orientable χ(M)

sphere S 2 yes 2

projective plane RP2 no 1

torus T 2 yes 0

klein bottle K no 0

sphere with g > 1 handles yes 2 − 2g

sphere with p > 1 cross caps no 2−p

It is easy to dream up more general surfaces. Starting from the sphere, one
can attach one or more ‘handles’. If the number of handles is denoted by g (g
is called the genus), then the resulting surface Mg is homeomorphic to a torus
with g ‘holes’. There is a short cut to computing the Euler characteristic of
such a surface: adding a handle or extra hole effectively reduces it by 2, and we
obtain the famous formula χ(Mg ) = 2 − 2g. Note that the sphere has genus 0,
and the torus genus 1.

Figure 27.3: Orientable surfaces with χ = −6

Figure 27.3 shows the surface M4 . The two representations are topologically
equivalent, and were indeed created by the same program, varying only the
number of sample points. We can check the value of χ(M4 ) by counting the
27.5. TOPOLOGY OF SURFACES 917

numbers of vertices, edges and faces on the right. Each of the four polyhedral
tori contributes at least 12 vertices, at least 28 edges, and exactly 16 faces.
Inspection of the central part from different angles (there is an animation in
Notebook 27 to help, using the quaternion techniques of Chapter 23) reveals
that the whole surface admits a triangulation with

V = 62, E = 132, F = 64,

confirming that χ = 62 − 132 + 64 = −6.

Figure 27.4: A representation of a nonorientable surface with χ = −1

Compact nonorientable surfaces cannot be represented inside R3 without

self-intersections. Nevertheless, it is customary to think of a nonorientable sur-
face as a sphere with a certain number p > 1 of cross caps glued on. The
projective plane corresponds to p = 1, since we can shrink the sphere so as to
obtain the model shown in Figure 11.15 on page 346. Adding a cross cap actu-
ally lowers the Euler characteristic by 1, so χ(RP2 ) = 2 − 1 = 1. Figure 27.4
shows a sphere with one cross cap and one handle attached; the result is a sur-
face S with χ(S) = 2 − 1 − 2 = −1. Since S incorporates at least one cross cap
it must be nonorientable, and Theorem 27.23 tells us that S is homeomorphic
to a sphere with 3 cross caps attached. Visualizing this equivalence is however
best done with planar models of the sort discussed in [FiGa].
The last two results of this section will enable us to globalize the discussion
of the Gauss–Bonnet theorem.
918 CHAPTER 27. THE GAUSS–BONNET THEOREM

Proposition 27.28. Let M be an oriented abstract surface, and suppose that

A = {(xα , Uα )} is a family of parametrizations covering M, compatible with the
orientation of M. Then there is an oriented triangulation T of M such that
every triangular region T ∈ T is contained in some coordinate neighborhood
UT for some (xT , UT ) ∈ A. Furthermore, if the boundary of every triangular
region in T is positively oriented, then adjacent triangles determine opposite
orientations in common edges.

Proposition 27.29. Assume that M, A and T are as in Proposition 27.28, and

let ds2α = Eα du2α + 2Fα duα dvα + Gα dvα2 be the expression of the metric of M
in terms of the coordinates given by xα . Let Q be a regular region, and let
f : Q → R be a continuous function. Then
X ZZ p
(f ◦ xα ) Eα Gα − Fα2 duα dvα
T ∈T x−1
α (T )

is the same for any triangulation T of Q such that every region T ∈ T is

contained in some neighborhood Uα with (xα , Uα ) ∈ A.

27.6 The Global Gauss–Bonnet Theorem

The Gauss–Bonnet Theorem is concerned with subsets of abstract surfaces
whose boundaries are piecewise-regular curves. More precisely:

Theorem 27.30. (Gauss–Bonnet, global version) Let Q be a regular region of

an oriented abstract surface M, and let C1 , . . . , Cn be piecewise-regular curves
which form the boundary of Q. Suppose that each Ci is positively oriented. Let
ε1 , . . . , εp be the external angles of C1 , . . . , Cn . Then
ZZ Z p
X
(27.24) K dA + κg ds + εj = 2π χ(Q).
Q ∂Q j=1

Proof. Let T be a triangulation of Q such that every triangular region in T is

contained in a coordinate neighborhood of a family of orthogonal patches. We
give the positive orientation to each triangular region in T ∈ T. In this way,
adjacent triangular regions give opposite orientations to their common edge.
We apply Corollary 27.20 to every triangular region in T and add up the
results. The integrals of the geodesic curvature along the interior edges (that
is, those edges that are not contained in one of the boundary curves) cancel one
another. We obtain
ZZ Xn Z X
(27.25) K dA + κg ds = −π F + ιij ,
R i=1 Ci
27.6. GLOBAL GAUSS–BONNET THEOREM 919

P
where F is the number of triangular regions in T and ιij is the sum of all
interior angles. Let
Ve = number of external vertices of T,
Vi = number of internal vertices of T,
Ee = number of external edges of T,
Ei = number of internal edges of T.
The angles around each internal vertex add up to 2π; hence the sum of all
internal interior angles is 2πVi . A similar calculation computes the sum of the
external interior angles, and so
X p
X
ιij = 2πVi + πVe − εj .
j=1

Thus (27.25) can be rewritten as

ZZ Xn Z p
X
(27.26) K dA + κg ds + εj = π(−F + 2Vi + Ve ).
R i=1 Ci j=1

We have the relations

(27.27) Ve = Ee , V = Ve + Vi , E = Ee + Ei .
Furthermore, 3F = 2Ei + Ee , which can be written as
(27.28) −F = 2F − 2Ei − Ee .
From (27.26)–(27.28), we compute
ZZ Xn Z p
X
K dA + κg ds + εj = π(2F − 2Ei − Ee + 2Vi + Ve )
R i=1 Ci j=1
 
= π 2F − 2(E − Ee ) − Ee + 2(V − Ve ) + Ve

= π(2F − 2E + 2V)

= 2π χ(Q),
proving (27.24).
In the special case of a compact surface we obtain:
Corollary 27.31. Let M be a compact orientable abstract surface. Then
ZZ
(27.29) K dA = 2π χ(M),
M

where K is the Gaussian curvature with respect to any metric on M.

Proof. A compact surface has no boundary. Hence (27.24) reduces to (27.29).
920 CHAPTER 27. THE GAUSS–BONNET THEOREM

In 1828, Gauss proved Theorem 27.30 for the special case of polygonal re-
gions whose sides are geodesics, and Bonnet generalized it to the form above
in 1848 (see [Gauss2] and [Bonn1]). The use of Green’s Theorem in proving
Theorem 27.13 was clarified by Darboux ([Darb2, volume 3, pages 122–128]).

27.7 Applications of the Gauss–Bonnet Theorem

A simple but important application of Corollary 27.31 is the determination of
the diffeomorphism type of compact orientable surfaces with positive Gaussian
curvature.

Theorem 27.32. A compact orientable surface M with positive Gaussian cur-

vature is diffeomorphic to a sphere.

Proof. If M has positive Gaussian curvature, Corollary 27.31 implies that

M has positive Euler characteristic. Since M is orientable, it follows from
Theorem 27.23 that M is diffeomorphic to a sphere.

Next, we prove a theorem of Jacobi (see [Jacobi]).

Theorem 27.33. (Jacobi, 1843) Let α : [a, b] → R3 be a closed regular curve

with nonzero curvature. Assume that the curve described by the unit normal N
(considered as a curve in the unit sphere S 2 (1) ⊂ R3 ) is a simple closed curve.
Then N([a, b]) divides S 2 (1) into two regions of equal areas.

Proof. Without loss of generality, we can assume that α has unit speed. Let
s denote the arc length function of α, and let s̃ denote the arc length function
of N, considered as a curve on the unit sphere. The second Frenet formula of
(7.12), page 197, implies that

dN
= −κ T + τ B,
ds
and so
d2 N dκ dτ
2
=− T+ B − (κ2 + τ 2 )N.
ds ds ds
From these two equations we compute the vector triple product

dN d2 N
 
dτ dκ
N =κ −τ ,
ds ds2 ds ds

and we also deduce that

 2
dN dN ds
1= · = (κ2 + τ 2 ).
ds̃ ds̃ ds̃
27.8. EXERCISES 921

We then obtain
dτ dκ
dN −3

dN d2
 κ −τ
κg (N) = N
N
= ds2 ds ds
ds ds ds2 κ + τ 2 ds̃
d τ  ds
= arctan
ds κ ds̃
d τ
= arctan .
ds̃ κ
Let C denote the curve N([a, b]). Since this is closed,
d
Z Z
τ
(27.30) κg (N)ds̃ = arctan ds̃ = 0.
C C ds̃ κ
Let R be the region enclosed on one side by C ; then χ(R) = 1 because C is a
simple curve. Therefore, from (27.24) and (27.30) we obtain
Z
2π = 2π χ(R) = K dA = area(R).
R

Since the total area of the sphere is 4π, the theorem follows.
Finally, we state without proof an important generalization of Corollary 27.31
due to Cohn-Vossen [CV].

Theorem 27.34. Let M be a complete orientable abstract surface. Then

ZZ
(27.31) K dA 6 2π χ(M),
M

provided both sides of (27.31) are finite.

27.8 Exercises
1. Prove Lemma 27.11 and Corollary 27.14.

2. Prove Proposition 27.25.

3. Let M ⊂ R3 be a compact orientable surface not homeomorphic to a

sphere. Show that there are points where the Gaussian curvature is posi-
tive, negative and zero.

4. Let H denote mean curvature. Show that

Z
H 2n dA.
S 2 (a)

is independent of the radius of the sphere S 2 (a) if and only if n = 1.

 922 CHAPTER 27. THE GAUSS–BONNET THEOREM

M 5. Let p
2
(x, y, z) ∈ R3 | z 2 + x2 + y 2 − a = b2

T (a, b) =
be a torus of wheel radius a and tube radius b (see Section 10.4). Use the
parametrization of T (a, b) on page 406 to compute
Z Z
K dA and H 2 dA.
T (a,b) T (a,b)

Do the calculations both by hand and using techniques from Notebook 27.
What does the Gauss–Bonnet Theorem tell us about the integral of the
Gaussian curvature over T (a, b)? Is the integral over a surface of the square
of the mean curvature a topological invariant?

6. Let M be an abstract surface homeomorphic to a cylinder with Gaussian

curvature K < 0. Show that M admits at most one simple closed geodesic.