Solutions to Exercises in Curves and Surfaces,

2nd Edition, by Sebastián Montiel and

Antonio Ros

Danilo Gregorin Afonso

[email protected]
Sapienza Università di Roma

Last updated in January 19, 2021

Contents

Preface 2

2 Surfaces in Euclidean Space 3

2.1 Exercises in the text . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Exercises at the end of the chapter . . . . . . . . . . . . . . . . . . . 5

3 The Second Fundamental Form 11

3.1 Exercises in the text . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Exercises at the end of the chapter . . . . . . . . . . . . . . . . . . . 19

4 Separation and Orientability 34

4.1 Exercises in the text . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Exercises at the end of the chapter . . . . . . . . . . . . . . . . . . . 35

7 Intrinsic Geometry of Surfaces 41

7.1 Exercises in the text . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Bibliography 44

Index 45

1
Preface

This text contains solutions for some of the Exercises in Montiel and Ros
[2]. I am studying Differential Geometry by myself and so I am writing these
solutions as means of fixing knowledge. Priority is given to exercises which are
not solved in the book.
I in no way guarantee that the contents of the present text are 100% correct,
and thus I do not assume responsibility for the misuse of this material.
I take the opportunity to acknowledge the MathStackExchange community,
which has been of immense help. I also acknowledge Brazilian science funding
agency CNPq for the financial support.
In the index you will find a list of the solutions present in this text. Unless
otherwise stated, Theorems, Corollaries and so on refer to the ones given in the
book itself.
If you notice any errors or have any questions or comments, I will be the
most glad to hear from you. Reach me via email.
It would make me extremely happy to know that this has been of some use-
fulness for someone, so also write me if this is the case.
I wish you a good study.

2
Chapter 2

Surfaces in Euclidean Space

2.1 Exercises in the text

Exercise 2.20 Note that every surface is locally homeomorphic to an open disc.
To see this, it suffices to consider a parametrization and restrain the domain.
Hence, we can cover the surface with connected open sets. Since the surface is
compact, we can cover it with finitely many connected open sets. If there were
infinitely many connected components, this would not be possible.

Exercise 2.28 Let f : S −! Rm be a differentiable function on the surface S.

Then f ◦ X is differentiable, hence continuous, for every parametrization X .
Now,
f = ( f ◦ X ) ◦ X −1
is continuous, as a composition of continuous maps.
Now, let f : O −! S be a differentiable function defined on an open set O ⊂ Rn
onto a surface S. Then i ◦ f is differentiable, hence continuous. Now, let us take
an open set V ⊂ R3 . Then
³ ´
(i ◦ f )−1 (V ) = f −1 i −1 (V )

is open. Now, every open set in S can be seen to be of the form i −1 (V ), hence f
is continuous.
Finally, let f : S −! S 1 be differentiable. Then i 1 ◦ f ◦X is differentiable for every
parametrization X of S. The argument then goes as the previous one, but now
we have to compose also with X −1 .

Exercise 2.29 Let X : U −! S be a parametrization of S. Then

(f + g)◦ X = f ◦ X + g ◦ X

3
which is differentiable, as a sum of differentiable functions. A similar argument
holds for f − g and λ f .
Now, note that
m
X
〈f ,g〉 = fi gi
i =1
and that
n
X
〈f ,g〉◦ X = ( f i ◦ X )(g i ◦ X )
i =1
which is of course differentiable.
Finally, it suffices to notice that
à !
f f ◦X
◦X =
g g ◦X

which is also differentiable.

Exercise 2.37 Note that

i ◦ X : U −! R3 = X : U −! R3

which is differentiable by the definition of parametrization.

Now, note that
X −1 ◦ X : U −! R2
is the identity, hence is differentiable. Taking into account Lemma 2.36 we get
that X −1 is differentiable.

Exercise 2.43 We have already shown in Exercise 2.37 that X is differentiable,

as well as its inverse. Hence X is a diffeomorphism.

Exercise 2.44 We know from Theorem 2.38 that g ◦ f is differentiable. Now,

note that
(g ◦ f )−1 = f −1 ◦ g −1
which is a composition of differentiable maps and thus is differentiable, by The-
orem 2.38 again.
The other group properties are trivial.

Exercise 2.45 Let X : U −! S be a parametrization of S. Then φ ◦ U is differ-

entiable by the Chain Rule of Calculus. Hence φ is differentiable in the sense
of Definition 2.27 A. On the other hand, let Y : V −! φ(S) be a arametrization
of φ(S). We know that φ−1 : φ(O 1 ) −! O 1 is differentiable btween open sets.
Hence, by the same argument, φ−1 ◦ Y is differentiable, whence φ−1 is differen-
tiable between surfaces.

4
Exercise 2.55 Let
f : R3 −! R
.
p 7−! |p|2 − 〈p, a〉2
Then f is differentiable, r 2 is a regular value and S = f −1 ({r 2 }). We also have
that
(d f )p (v) = 〈p, v〉 − 〈p, a〉〈a, v〉.
To see this, use coordinates:
à !2
p i2 +
X X
f (p) = p i ai
i i

Since for p ∈ S we know that T p S = ker(d f )p , we have that

n o
T p S = v ∈ R3 | 〈p, v〉 − 〈p, a〉〈a, v〉 = 0 .

Now, notice that p − 〈p, a〉a is orthogonal to both the axis of the cylinder and to
T p S.

Exercise 2.62 We know that

φ ◦ φ−1 = i d S 2
is differentiable, since it is a composition of differentiable functions. By the
Chain Rule ³ ´
d (φ ◦ φ−1 ) = (d φ)p ◦ (d φ−1 )φ(p) = i d Tφ(p) S 2 .
φ(p)

With the same argument for φ−1 ◦ φ we complete the proof.

2.2 Exercises at the end of the chapter

(3) Let p ∈ S ∩ P where P is any plane containing the rotation axis R, which is
supposed, without loss of generality, to pass through the origin (which implies
that p 6= 0, once R ∩ S = ;). Let a be a unit vector in the direction of R and
b, c ∈ R3 be such that {a, b, c} is a positively oriented basis. Let α : R −! R3 be
given by

α(t ) = 〈p, a〉a + (〈p, b〉 cos t + 〈p, c〉 sin t )b + (−〈p, c〉 sin t + 〈p, b〉 cos t )c.

Then α(t ) = φt (p) ∈ S, where φt : R3 −! R3 is the rotation of t around R, α(0) =

p and α0 (0) = 〈p, c〉b − 〈p, b〉c = p ∧ a. Therefore, p ∧ a ∈ T p S. Suppose that S
and P do not intersect transversely. Then we would have T p S = T p P = P and

5
hence p ∧ a ∈ P . But both a and p are in P , so this is impossible. Hence, P and
S intersect transversely.
Let Q, P be planes containing R intersecting S. There is a rotation interchanging
them. Hence, if P ∩S = ; then Q ∩S = ; and we would have that S = ;. But this
is impossible, hence S intersects every plane containing R. By Theorem 2.67,
each of these intersections is a simple curve. It is clear that each of these curves
generate S by rotations around R.
Now, let P 0 be the plane containing R perpendicular to c and let p : I −! P 0
be a regular curve defined on an open interval of R whose trace is contained on
the simple curve S ∩ P 0 . We have

p(t ) = x(t )a + y(t )b, t ∈I

for some functions x, y : I −! R such that y(u) 6= 0 for all u and x 0 (u)2 + y 0 (u)2 >
0 (by the regularity of the curve). It then follows that

X (u, v) = x(u)a + y(u) cos vb − y(u) sin vc, (u, v) ∈ I × (θ, θ + 2π)

is a parametrization of S for every θ ∈ R. Let us show that (d X )(u,v) is injective.

We have  
x 0 (u) 0
(d X )(u,v) =  y 0 (u) cos v −y(u) sin v  .
 
−y 0 (u) sin v −y(u) cos v

Let (z, w) ∈ R2 . We will show that we cannot have (d X )(u,v) (z, w) = 0 unless
(z, w) = 0. Indeed, if it was the case, we would have

x 0 (u)z = 0,
y 0 (u) cos v z − y(u) sin v w = 0,
− y 0 (u) sin v z − y(u) cos v w = 0.

If z = 0 then w = 0. Suppose that x 0 (u) = 0 and z 6= 0. Then y 0 (u) 6= 0. If cos v = 0

or sin v = 0 then z = w = 0. If not, then

y 0 (u) 1 y 0 (u)
w= z =− tan v z =⇒ z = − tan2 v z =⇒ z = w = 0,
y(u) tan v y(u)

which concludes the proof.

(4) Suppose that the surface S is not connected. Then each connected compo-
nent is compact, and hence every function f : S −! R attains a local maximum
and a local minimum in each connected component. But then, according to

6
Proposition 2.63, each of these points will be a critical point. Hence, any func-
tion f will have 2n critical points, where n denotes the number of connected
components, that is, will have at least 4 critical points. This is a contradiction
with the fact that there exists a function on S with at most 3 critical points.

(5) Since S is compact, every connected componet is compact. Hence, in ev-

ery connected component the function f attains a local maximum and a local
minimum. These points are critical points of f , by a proposition in the body
of the text. Hence, if n is the number of connected components and c is the
number of critical points, we have that

c ≥ 2n =⇒ n ≤ c/2.

This estimate can be improved by means of Morse theory, but this falls out of
the scope of this text1 .

(8) It is easy to see that the translations τ1 (x, y, z) = (x + 2π, y, z) and τ2 , τ3 de-
fined analogously are linearly independent and keep the surfaces invariant.
Now, note that the surfaces divide the space in two regions, given by R + =
{(x, y, z) ∈ R3 | f (x, y, z) > 0} and R − = {(x, y, z) ∈ R3 | f (x, y, z) < 0} where for the
primitive surface, for example, f (x, y, z) = cos x + cos y + cos z. Now, we want
translations that exchange R + and R − .
For the primitive surface, one such translation is given by τπ (x, y, z) = (x +π, y +
π, z + π). Indeed, f ◦ τπ (x, y, z) = − f (x, y, z).
For the diamond surface, the translation that swaps regions is even simpler:
τ(x, y, z) = (x + π, y, z).
With regard to the gyroid surface, see this answer on MathStackExchange.

(10) We will use exercise (9). Define the following map:

φ : S2 −! R3
.
p 7−! f (p)p

Let v ∈ T p S2 and α : (−ε, ε) −! S2 be a curve such that α(0) = p and α0 (0) = v.

1
If you can improve this estimate in an elementary way, please email me.

7
Then
¯
d ¯¯
(d φ)p (v) = ¯ (φ ◦ α)(t )
dt ¯
¯t =0
d ¯¯
= ¯ f (α(t ))α(t )
dt ¯
t =0
= (d f )α(0) α0 (0)α(0) + f (α(0))α0 (0)
= (d f )p (v)p + f (p)v,

hence the kernel of (d φ)p is trivial.

Also note that φ is also clearly injective. Since S2 is compact, taking Exercise (9)
into account we are done.

(11) First let us show that every surface S( f ) as in Exercise (10) is star-shaped.
It is clear that 0 ∉ S( f ), because f > 0. Now, S( f ) is not star-shaped, that is,
suppose we could draw a line between 0 and f (p)p tangent to S( f ) at f (p)p.
Then we would have that p ∈ Tφ(p) S( f ) = (d φ)p (T p S2 ). Then we have that for
some v ∈ T p S2
p = (d φ)p (v) = (d f )p (v)p + f (p)v.
Taking the scalar product with v, we have that v = 0, which is a contradiction.
Hence S( f ) is indeed star-shaped.
Now, let us show that every star-shaped surface if of the form S( f ) for some
f.

(13) a: Let B 0 = {i 1 , i 2 } be another orthonormal basis of T p S 1 such that

i 1 = a1 e 1 + b1 e 2 , i 2 = a2 e 1 + b2 e 2 .

The transition matrix is " #

a1 a2
b1 b2 ,

whose determinant is a 1 b 2 − a 2 b 1 = 1, since both basis are orthonormal. Then

|(d f )p (i 1 ) ∧ (d f )p (i 2 )| = |(a 1 (d f )p (e 1 ) + b 1 (d f )p (e 2 )) ∧ (a 2 (d f )p (e 1 ) + b 2 (d f )p (e 2 ))|

= |(a 1 b 2 − a 2 b 1 )((d f )p (e 1 ) ∧ (d f )p (e 2 ))|

which concludes the proof.

b: (d f )p is an isomorphism if and only if (d f )p (e 1 ) and (d f )p (e 2 ) are linearly
independent and this happens if and only if | Jac f |(p) 6= 0.

8
(14) We want to show the following lemma:

Lemma 2.2.1. Let S be a surface, I ⊂ R be an open interval and F : S × I −! Rm

be a function. Suppose that for every p ∈ S there exists a parametrization X p :
U p −! S, with p ∈ X p (U p ), such that F ◦ (X p × i d I ) : U p × I −! Rm is differen-
tiable. Then F is differentiable.

Proof. Let X : U −! S be a parametrization of S. Our aim is to show that F ◦(X ×

i d I ) : U × I −! Rm is differentiable at every q ∈ U . To do this, let p = X (q) ∈ S.
Then there is some X p : U p −! S such that F ◦ (X p × i d I ) : U p × I −! Rm is
differentiable. Of course we have that X (U ) ∩ X p (U p ) = O is a nonempty open
set of S with p ∈ O. Now, notice that

(X p−1 ◦ X ) × i d I : X −1 (O) × I −! X p−1 (O) × I

is differentiable, because each term is differentiable (recall that the change of

parameters is a diffeomorphism). Hence
µ ³ ´¶ µ³ ´ ¶
−1
F ◦ (X × i d I ) = F ◦ X p × i d I ◦ X p ◦ X × i d I on X −1 (O) × I

which is a composition of differentiable functions.

(15) Let X : U −! S be a parametrization of S. Then F ◦ (X × i d I ) and G ◦ (X ×

i d I ) are differentiable. But then

(F +G) ◦ (X × i d I ) = F ◦ (X × i d I ) +G ◦ (X × i d I )

is differentiable, as a sum of differentiable functions. An analogous argument

proves that F −G and λF are differentiable.
Now, note that 〈F,G〉 = F 1G 1 + . . . + F m G m . Thus it suffices to show that F i G i is
differentiable for every i = 1, . . . , m. Now, indeed,
¡ ¢ ³ ¡ ¢´ ³ ¡ ¢´
Fi G i ◦ X × i d I = Fi ◦ X × i d I G i ◦ X × i d I ,

which is differentiable as a product of differentiable functions.

Lastly, we want to show that
¡ ¢
F ¡ ¢ F ◦ X × i dI
◦ X × i dI = ¡ ¢
G G ◦ X × i dI

is differentiable. But this is the case since both numerator and denominator are
differentiable.

9
(16) First, we want to show that F ◦ X : U −! Rm is differentiable for every
parametrization X : U −! S of S. Let X : U −! S be a parametrization of S and
let q ∈ U . Let us show that F t is diffeentiable at p = X (q) showing that F t ◦ X is
differentiable at q. Since F is differentiable, for h ∈ U we have:

F t ◦ X (q + h) − F t ◦ X (q) F ◦ (X × i d I )(q + h, t ) − F ◦ (X × i d I )(q, t )

lim = lim
h !0 |h| h !0 |(h, 0)|
¡ ¢
= d (F ◦ (X × i d I )) (q,t ) ((h, 0)),

which can be identified with a linear transformation on U .

The same argument applies to the derivative of αp (t ).

(18) First, note that

j = i 3 × i 1,
where i k is the inclusion on Rk . Hence, if j ◦ G is differentiable, then

(i 3 × i 1 ) ◦ G = (i 3 ◦ p 1 ◦ G) × (i 1 ◦ p 2 ◦ G)

is differentiable, hente both p 1 ◦ G and p 2 ◦ G are differentiable, from what we

conclude that G is differentiable, in the sense of the statement of the exercise.
The converse implication is trivial.

(21) Let {z 1 , z 2 , . . .} be the set of heights. Then for each z i the function X (u, v) =
(u, v, z i ) is a prametrization of the given plane. Hence the union of these planes
is a surface.
Now, suppose that there is some interval I of heights. Then the union of this
horizontal planes would not have empty interior as a subset of R3 , in contra-
diction with exercise (1).

10
Chapter 3

The Second Fundamental Form

3.1 Exercises in the text

Exercise 3.9 We construct a unit normal vector field over S. Let G : S −! S2 ⊂
R3 be the map which associates to every p ∈ S the unit vector at p satisfying P .
By the first hypothesis this map is well defined. Also, it is differentiable, because
it coincides locally with a differentiable vector field. We are done.

Exercise 3.10 Let A be the set of the c ∈ [a, b] such that there exists a map Nc :
[a, c] −! R3 satisfying the properties we want, that is,

Nc (a) = n, |Nc (t )| = 1 ∀t ∈ [a, c], Nc (t ) ⊥ Tα(t ) S ∀t ∈ [a, c].

Since a ∈ A, then A is non-empty. Since [a, b] is connected, if we show that A is

both closed and open then we must have that A = [a, b], as we want.
To show that A is open, let c ∈ A. We will find that there is an open neighbor-
hood of c. We proceed, in some sense, by induction. Key facts are the continu-
ity of α and Lemma 3.2, which will allow us to extend our map Nc . Indeed, by
Lemma 3.2 there is an open neighborhood V of α(c) in S where there is a unit
normal field defined. Note that we can assume that N (α(c)) = Nc (c). If this is
not the case, just take the opposite −N . By the continuity of α, if c < b then
there exists ε > 0 such that

α (c − 2ε, c + 2ε) ⊂ V.
¡ ¢

Then we can define Nc+ε : [a, c + ε] −! R3 as follows:


N (t ), t ∈ [a, c],
c
Nc+ε (t ) =
(N ◦ α)(t ), t ∈ [c, c + ε].

11
The above map satisfies the properties we want. Hence [a, a + ε) ⊂ A is an open
neighborhood of c which proves the claim that A is open.
Let us now show that A is closed. Lemma 3.2 will again be of uttermost im-
portance. Let (c n ) ⊂ A converge to c ∈ [a, b]. We wnat to show that c ∈ A. Let
V be an open neighborhood of α(c) as above. There is some k ∈ Z such that
α([c n , c]) ⊂ V for n ≥ k. Again choosing the sign of the unit normal field N de-
fined in V to our convenience, we can assumed that Nck = N (α(c k )). We now
define Nc : [a, c] −! R3 by

N (t ), t ∈ [a, c ],
ck k
Nc (t ) =
(N ◦ α)(t ), t ∈ [c k , c].

Again, this map satisfies the properties that we require. Hence A is also closed,
and then we conclude that A = [a, b].
We proved that there exists a map N : [a, b] −! R3 with the required properties.
Let us show now that it is unique. Suppose there was another map, say N 0 , as
we required. Then the map 〈N , N 0 〉 : [a, b] −! R assumes only the values −1
and 1 and is continuous. Since it takes the value 1 at a, we can conclude that
N0 = N.
Now, let α be a closed curve (that means that α(a) = α(b) and assume that S is
orientable. Since it is orientable, we can take a unit normal field N on S, and
assume that N (α(a)) = n. Then, by the first part of the exercise we know that
N ◦ α is the only unit normal field on S along α such that N (α(a)) = n. Now, of
course
N ◦ α(b) = N ◦ α(a) = n.

For the last part, we can assume that S is pathwise connected by working on
one of the components. Let p 0 ∈ S be an arbitrary point. Define the map N :
S −! S2 by
N (p) = Nα (1),
where α : [0, 1] −! S is a curve with α(0) = p 0 and α(1) = p and Nα is the unit
normal field along α with initial condition n, whose existence and uniqueness
we proved in the first part of the exercise.
Let us show that the map N is well-defined, that is, does not depend on α. Let
β : [0, 1] −! S be another curve such that β(0) = p 0 and β(1) = p and Nβ be
the corresponding field. We need to show that Nα (1) = Nβ (1). Let ∗ denote
concatenation, that is γ = β−1 ∗ α is the curve that leaves p 0 , goes to p along α
and then comes back to p 0 along β. Notice that γ is a closed curve with γ(0) =

12
p 0 . Suppose that Nβ (1) = −Nα (1), that is, N is not well defined in p = α(1) =
β(1). Then
Nγ = −Nβ−1 ∗ Nα
is the corresponding normal field to the curve γ, the minus sign on Nβ−1 is here
to guarantee that it will "glue well" with Nα . But then

Nγ (0) = Nα (0) = n and Nγ (1) = −Nβ−1 (1) = −Nβ (0) = −n,

which is impossible by what we assumed about closed curves.

It remains only to show that N is differentiable. Let V be an open connected
subset of S where ther is a unit normal field G (which exists by Lemma 3.2).
Then either G = N |V or G = −N |V , and we are done.

Exercise 3.13 Let N1 and N2 be unit normal fields on S 1 and S 2 . By Lemma 3.3,
either they coincide or are opposite on S 1 ∩ S 2 . We can assume the first, since if
it is not the case it suffices to take the opposite of any of them. Then we define
N : S −! S2 by N |S 1 = N1 and N |S 2 = N2 .

Exercise 3.14 Suppose S to be orientable and choose an orientation N . Then

either N = N A , the orientation of S 1 and S 2 on A, or N = −N A . But then we we
walk along a curve in S 1 from A to B , the orientation N is preserved. The same
happens if we walk along a curve in S 2 with the same endpoints, but now the
orientation has changed, which is a contradiction. Hence S is not orientable.
To see that the Möbius strip is non-orientable, let

S 1 = F [0, π/2) × (−1, 1) ∪ (3π/2, 2π] × (−1, 1)

¡ ¢

and ¡ ¢
S 2 = F (π/4, 7π/4) × (−1, 1) .
These neighborhoods satisfy the the properties of the exercise. Indeed, both
open subsets are orientable because they are both covered by a parametrization
as given in page 74. Computing the normals F u ∧ F v as given in the same page,
we see that they coincide in one component and are inverses in the other.

Exercise 3.15 First we show that N1 is well-defined. Let {a, b} and {c, d } be two
bases of T p S 1 with the same orientation. Then

c = αa + βb, d = γa + δb

and hence the matrix of change of basis is

" #
α γ
M= ,
β δ

13
which has positive determinant. Then, by the linearity of the determinant on
the lines,
³ ´ ³ ´
det (d f )p (c), (d f )p (d ), N2 ( f (p)) = det M det (d f )p (a), (d f )p (b), N2 ( f (p)) .

To show that N1 is differentiable, consider a parametrization X : U −! S on a

connected open subset of R2 . Note that {X u , X v } is a basis of the tangent plane
T X S, that
(d f ) X (X u ) = ( f ◦ X )u , (d f ) X (X v ) = ( f ◦ X )v
and that (d f ) X is an isomorphism. Then
¡ ¢
det (d f ) X (X u ), (d f ) X (X v ), N2 ◦ f ◦ X

is a nonvanishing continuous function on U , whose sign we may assume to be

positive. Then
Xu ∧ Xv
N1 ◦ X =
|X u ∧ X v |
and therefore N1 is differentiable.

Exercise 3.16 Let f : S 1 −! S 2 be the diffeomorphism between the two sur-

faces. Suppose one of them, say S 2 , is orientable (if it is S 1 , then just take f −1 as
the diffeomorphism and change the names). By the previous exercise we know
that S 1 must be orientable, since every diffeomorphism is a local diffeomor-
phism.
On the other hand, suppose some of them is not orientable, say S 1 . If S 2 was
orientable, we would arrive at a contradiction, hence the other is not orientable
as well. This concludes the proof.

Exercise 3.17 Since f is a local diffeomorphism, then (d f )p is a linear isomor-

phism for each p ∈ S 1 , by Proposition 2.77. Then | Jac f (p)| 6= 0 for all p ∈ S 1 , by
Exercise (13) of Chapter 2. Since both surfaces are connected, then Jac f never
changes sign. Hence, or f changes orientation or it preserves it.
The result then follows by our definitions.

Exercise 3.22 We know that the hyperbolic paraboloid is the graph of a func-
tion f (x, y) = x 2 − y 2 , hence an orientable surface with unit normal field
1
N=p (− f x , − f y , 1)
1 + |∇ f |2
where the subindex denotes partial differentiation. Then, for our function,
1
N (x, y, z) = p (−2x, 2y, 1).
1 + 4(x 2 + y 2 )

14
A parametrization of this surface is given by X : R2 −! R3 where

X (u, v) = (u, v, f (u, v)).

Then a basis for the tangent space T p S at p is given by {X u (u, v), X v (u, v)}. But

X u (u, v) = (1, 0, f u (u, v)), X v (u, v) = (0, 1, f v (u, v)).

Then, at the point (0, 0, 0) the tangent space has basis {(1, 0, 0), (0, 1, 0)}. Hence
the matrix of (d N )(0,0,0) when seen as a map from T(0,0,0) S to T(0,0,0) S is
" #
−2 0
.
0 2

Now,
K ((0, 0, 0)) = −4, H ((0, 0, 0)) = 0.
Hence, at the origin the hyperbolic paraboloid has negative Gauss curvature
and zero mean curvature.

Exercise 3.23 Since the curvatures are identically zero, the second fundamen-
tal form is identically zero. Hence, (d N )p = 0 for all p ∈ S. Now, since S is
connected, this implies that N is constant. Hence S is contained in some plane
that has N as normal vector.

Exercise 3.24 The surface S is given by

S = {p ∈ R3 | f (p) = 0}

where
f (p) = 〈B p, p〉 + 2〈b, p〉 + c.
Then
(d f )p (v) = 2〈B p, v〉 + 2〈b, v〉.
Then it follows that for v ∈ T p S, B p + b is orthogonal to v, that is, B p + b is
normal to the surface. To get a Gauss map, it suffices divide by the norm. Hence,

Bp +b
N (p) = , p ∈S
|B p + b|

is a Gauss map on S.

15
Now, we use the definition to compute the differential of N at p ∈ S. Let α :
(−ε, ε) −! S be a curve such that α(0) = p, α0 (0) = v for v ∈ T p S. Then
¯
d ¯¯
(d N )p (v) = ¯ N (α(t ))
dt ¯
¯t =0
d ¯¯ B α(t ) + b
= ¯
dt ¯ |B α(t ) + b|
t =0 ¯ ¯
B α0 (t ) ¯¯ (B α(t ) + b)|B α(t ) + b|0 ¯¯
= ¯ +
|B α(t ) + b| ¯ |B α(t ) + b|2
¯
¯
t =0 t =0
Bv 〈B v, B p + b〉
= + (B p + b).
|B p + b| |B p + b|3
Since we are interested only in the tangent component, that is, the restriction
(d N )p |TN (p) S2 , it suffices to take only the first term. Applying the definition of
the second fundamental form gives us the desired expression.

Exercise 3.25 Let p ∈ S and φ(p) ∈ φ(S). We want to show that N 0 = A ◦ N ◦

φ−1 (φ(p)) is orthogonal to Tφ(p) φ(S). Now, note that

Tφ(p) φ(S) = A(T p S).

Hence, if w ∈ Tφ(p) φ(S) then w = Av for some v ∈ T p S. Now,

〈N 0 (φ(p)), w〉 = 〈AN (p), Av〉 = 〈N (p), v〉 = 0.

Hence, N 0 is a normal map. To see that it is a unit field, it suffices to note that A
being in O(3) does not change the norm of vector, hence |N 0 (φ(p))| = |N (p)| = 1.
The expression for the differential of N 0 follows at once from the Chain Rule.
Also,

σ0φ(p) ((d φ)p (v), (d φ)p (w)) = σ0φ(p) (Av, Aw) = −〈A ◦ (d N )p v, Aw〉
= −〈(d N )p v, w〉
= σp (v, w)

because A preserves inner products.

We can then conclude that the Gauss and mean curvatures are preserved.

Exercise 3.32 Let α : (−ε, ε) −! S ∩ P be a regular curve and let a be the di-
rection of the plane P , with |a| = 1. Then N ◦ α(t ) = a or −a for all t ∈ (−ε, ε).
Hence
d
N ◦ α(t ) = (d N )α(t ) α0 (t ) = 0.
dt

16
Hence, for all t ∈ (−ε, ε), that is, for every point α(t ) of the curve we have that

ker(d N )α(t ) 6= ;,

and thus the determinant is null.

Exercise 3.33 If the Gauss curvature is positive then the second fundamental
form is definite (either negative or positive). So in order to show that the Gauss
curvature is not positive it suffices to find a pair v, w ∈ T p S such that σp (v, w) =
0. Let us show that such a pair can be found in every point α(t ) = p 0 + t a ∈ S,
that is, α is the straight line we assume to exist in S.
First, note that for every t ∈ R (N ◦ α)(t ) ⊥ α0 (t ) = a. Hence

〈(N ◦ α)(t ), α0 (t )〉 = 0

and therefore, taking derivatives,

−〈(d N )α(t ) (α0 (t )), α0 (t )〉 = 〈(N ◦ α)(t ), α00 (t )〉.

But since α00 (t ) = 0, the right-hand side above is zero. Hence

σα(t ) (α0 (t ), α0 (t )) = 〈(d N )α(t ) (α0 (t )), α0 (t )〉 = 0

which concludes the proof.

Exercise 3.40 Let f : S −! R be the height function relative to a given plane,

no matter which. By compactness, there is a point p ∈ S which is an minimum
point for f . At this point, by Proposition 3.35 B, the map (d h)p is semi-definite
positive. Then, let v 1 , v 2 be the eigenvectors of σ and λ1 , λ2 be the respective
eigenvalues. We have:

σp (v i , v i ) = λi v iT v i = (d h)p (v i ) ≥ 0 =⇒ λi ≥ 0.

Then K (p) = det σp = λ1 λ2 ≥ 0, that is, the Gauss curvature is non-negative at

p.

Exercise 3.41 By definition, a point p is called elliptic if the Gauss curvature at

p is positive, which is equivalent to the second fundamental form at the point
being definite. Choosing the Gauss map properly, we can make the second
fundamental form be positive definite. But a quadratic form is positive defi-
nite if and only if it is coercive, that is, if there is a constant c > 0 such that
σp (v, v) ≥ c|v|2 for every v ∈ T p S. Indeed, the constant can be taken to be the
smaller eigenvalue of σp .

17
Now, consider
f : S −! R
p 7−! |p − p 0 |2

where p 0 = p + 1c N (p). Then, from Example 3.39 we know that

µ ¶
2 1 2
(d f )p (v) = 2 |v| − σp (v, v) .
c

If σp is coercive, then
(d 2 f )p ≤ 2[|v|2 − |v|2 ] = 0.
On the other hand, if p is at a maximum distance from p 0 , then

σp (v, v) ≥ c|v|2

which concludes the proof.

Exercise 3.42 Let S be a compact surface. Then there is some p ∈ S that max-
imizes the distance to the origin, and hence its square. Then, by Exercise 3.41,
the surface has positive Gauss curvature at p.

Exercise 3.50 Let S be our compact connected surface. Since K > 0 the prin-
cipal curvatures k 1 and k 2 have the same sign and are both nonzero. Then
H = 21 (k 1 + k 2 ) 6= 0 and hence the surface is orientable. Indeed, suppose it is
not orientable. Choose N at p. Then there is a closed curve passing through p
such that N changes sign. But then there is some point at which H = 0, which
is a contradiction.
Let c = H /K , so that H = cK . Then k 1 + k 2 = 2ck 1 k 2 and hence

k2
k1 = .
2ck 2 − 1

But then
dk 1 −1
= ,
dk 2 (2ck 2 − 1)2
what means that k 1 is decreasing as a function of k 2 and thus attains its mini-
mum where k 2 attains its maximum. Let a be this point. By Hilbert’s Theorem,
this point is umbilical: k 1 = k 2 . Then, for all p ∈ S we have that

k 2 (p) ≤ k 2 (a) = k 1 (a) ≤ k 1 (p)

which implies that every point in the surface is umbilical. Then, by Corollary
3.31, S must be a sphere.

18
Exercise 3.51 Let S be our surface with positive Gauss curvature and let k 1 = k
be constant. We have that k and k 2 have the same sign and that H = 21 (k + k 2 ).
Note that we cannot have k = 0, otherwise we would have K = 0. Since k 2 is
continuous and S is compact, k 2 attains its maximum at some point a ∈ S. At
this point, k 1 trivially attains its minimum k. Hence the point a is umbilical.
It then follows that the surface is totally umbilical, and we conclude that it is a
sphere invoking Corollary 3.31 once more.
The case where k 2 is constant is analogous.

3.2 Exercises at the end of the chapter

(1) Let p λ = p + λN (p) and consider the function

f (q) = |p λ − q|2 , q ∈ S.

We know that p is a critical point for f . Moreover,

³ ´
(d 2 f )p (v) = 2 |v|2 − λσp (v, v) , v ∈ T p S.

In order for p to be a local minimum of f (and hence of the distance itself )

we must have (d 2 f )p (v) ≥ 0. Now, let v = ae 1 + be 2 ∈ T p S, where e 1 and e 2
constitute the orthogonal basis of the principal directions. We have

|σp (v, v)| = |〈(d N )p (v), (v)|

= |〈k 1 ae 1 + k 2 be 2 , v〉|
= |k 1 |a 2 + |k 2 |b 2
≤ max |k 1 |, |k 2 | |v|2 .
© ª

But then
³ ´
(d 2 f )p (v) = 2 |v|2 − λσp (v, v) ≥ 2|v|2 (1 − |λ| max |k 1 |, |k 2 | ).
© ª

Hence, p is a minimum point for every p λ such that

© ª
1 − |λ| max |k 1 |, |k 2 | > 0,

which completes the proof.

(2) Let f : S −! R be the square of the distance function to the center a of the
ball of radius r in which S is contained:

f (p) = |a − p|2 .

19
Let p ∈ S be the point at which f attains its maximum (which exists, by com-
pactness). Them, according to proposition 3.35 we have that (d 2 f )p ≤ 0.
For v ∈ T p S, let α : (−ε, ε) −! S be a curve such that α(0) = p and α0 (0) = v.
Then ¯
d2 ¯¯
f ◦ α(t ) = 2〈α00 (0), p − a〉 − 2|v|2 .
dt 2 ¯
¯
t =0
If v is a unit vector, we have that

1 ≤ 〈α00 (0), a − p〉.

But if v is the principal direction e i , then

k i = σp (α0 (0), α0 (0)) = −〈(d N )p (α0 (0)), α0 (0)〉 = −〈(N α)0 (0), α0 (0)〉
= 〈α00 (0), N (p)〉

Using this on the above and noting that N (p) is nothing but ±(a − p) normal-
ized, we get:
1 ≤ 〈α00 (0), N (p)〉|a − p| = ±k i 〈N (p), a − p〉. (3.2.1)

Multiplying the inequalities (3.2.1), we get

1 ≤ k 1 k 2 〈N (p), a − p〉2 = k 1 k 2 r 2 .

Thus
1
K (p) = k 1 k 2 ≥ .
r2
Adding the inequalities (3.2.1), we get:

2
2 ≤ (k 1 + k 2 )〈N (p), a − p〉 = ±(k 1 + k 2 )r =⇒ |k 1 + k 2 | ≥
r
and hence
1 1
|H (p)| = |k 1 + k 2 | ≥ ,
2 r
which completes the proof.

(3) Let p ∈ S. We know that it is a critical point for the function

f λ (q) = |p λ − q|2

where p λ = p + λN (p). By compactness, this function attains both its global

maximum and global minimum, and since there are at most two critical points,
p must be one of these for all λ ∈ R except for maybe a countable number of

20
them. Then, by Proposition 3.35 we have that (d 2 f λ )p is semi-definite, except
maybe forcountably many λ. By Example 3.39, taking v to be a principal direc-
tion, we get that 1−λk i (p), for i = 1, 2 have the same sign, for all these values of
λ. Hence
0 ≤ (1 − λk 1 (p))(1 − λk 2 (p)) = 1 − 2λH (p) + λ2 K (p),
and this holds for all λ ∈ R except maybe for countably many of them. Them
we must have the discriminant of this polynomial to be non-positive (other-
wise we would have that (1 − λk 1 (p))(1 − λk 2 (p)) = 1 − 2λH (p) + λ2 K (p) ≤ 0 for
uncountably many λ). This implies that

H (p)2 ≤ K (p).

But since K (p) ≤ H (p)2 , we see that p is umbilical. We conclude that S is totally
umbilical. Being connected, it is either a piece of a plane or of a sphere. Being
compact, it can only be a sphere.

(4) The function f is clearly continuous. Notice that f (p) = 0 if and only if
p − p 0 ⊥ N (p), that is, if p − p 0 is tangent to S at p. Hence, if S is star-shaped
with respect to p 0 then f cannot change sign. If f does not change sign, there is
no line tangent to S passing through p 0 , and hence S is star-shaped with respect
to p 0 . This concludes the first part.
For the second part, let us compute the differential of f at p. In order to do
this, let α(−ε, ε) −! S be a regular curve such that α(0) = p and α0 (0) = v, for
v ∈ T p S. Then
¯
d f ◦ α(t ) ¯¯
(d f )p (v) = ¯
dt ¯
t =0
= 〈α0 (0), N (p)〉 + 〈p − p 0 , (N ◦ α)0 (0)〉
= 〈p − p 0 , (d N )p (v)〉 = 0.

Choosing v to be a principal direction e i we get that

k i (p)〈p − p 0 , e i 〉 = 0.

Them, either K (p) = 0 or p − p 0 = c N (p). Hence, the points where K does not
vanish are in a sphere of center p 0 and radius |c| > 0. By continuity, for each
point where K (p) 6= 0 we have a neighborhood of p in the said sphere, hence
K (p) = 1/c 2 , and this holds for every point in the neighborhood. Hence, K can
only assume the values 0 and 1/c 2 . But since K is continuous and S is con-
nected, K (p) = 1/c 2 for all p ∈ S, which concludes the proof.

21
(5) Let p ∈ S and let {e 1 , e 2 } be the basis of principal directions in T p S. Then
³ ´
Jac f (p) = det (d N )p (e 1 ), (d N )p (e 2 ), N (p)
¯ ¯
¯1 0 0¯
¯ ¯
= k 1 (p)k 2 (p) ¯0 1 0¯
¯ ¯
¯0 0 1¯
¯ ¯

= K (p),

which concludes the proof.

(6) Let S be our compact oriented surface and N : S −! S2 be the Gauss map.
Suppose K (p) > 0 for all p ∈ S. Then, given p ∈ S, by the previous exercise
we know that (Jac N )(p) = K (p) > 0. Then (d N )p is a linear isomorphism, by
Exercise 2 - (13). Then, by the Inverse Function Theorem (Theorem 2.75), N is
a local diffeomorphism.
On the other hand, assume that N is a local diffeomorphism for every p ∈ S.
Then (d N )p is a linear isomorphism, and hence (Jac N )(p) 6= 0. Then K (p) =
(Jac N )(p) 6= 0 every where. But the Gauss curvature is continuous, and by Ex-
ercise 3.43 there is at least one point where K (p) > 0. Hence K (p) > 0 for all
p ∈ S.

(7) Since everything is preserved by rigid motions, we may assume, up to a

translation, that P is the plane z = 0. Let (e 1 , e 2 , e 3 ) be th canonical orthonormal
basis of R3 .
First, we show that X is a parametrization. By definition it covers S so the first
condition is satisfied. Of course X is continuous. Let p = (p 1 , p 2 , p 3 ) ∈ S. Then

X −1 (p) = (α−1 (p 1 , p 2 ), sgn(a)p 3 )

is the inverse function and is clearly continuous. Finally,

det(d X )(s,t ) = det(α0 (s), a) = α0 (s)a 6= 0,

so (d X )q is injective for all q ∈ I × R. Hence, S is a surface.

A Gauss map in S is given by N (p) = N (X (s, t )) = n(s), where n(s)is the unit
vector orthogonal to α at s. Let e = α0 (s), which we may assume to be a unit
vector. Then {e, a} constitute an orthonormal basis of T p S. If v = λe + δa ∈ T p S,
we have that
(d N ) X (s,t ) (v) = −k(s)λe

22
where k(s) is the curvature of al pha at s, since N does not depend on t . Then,
−k(s) and 0 are the eigenvalues of (d N )p , that is, the principal curvatures. From
this it follows that for p = X (s, t ) ∈ S

K (p) = 0, H (p) = −k(s).

The conclusion is trivial, by the definition of a simple curve.

(8) A parametrization X (u, v) = (X 1 (u, v), X 2 (u, v), X 3 (u, v)) of the thorus is given
by

X 1 (u, v) = (a + r cos v) cos u

X 2 (u, v) = (a + r cos v) sin u
X 3 (u, v) = r sin v

where u ∈ [0, 2π) is the angle around the z axis and v ∈ [0, 2π) is the angular
position of the point in the respective circle (see this page).
Now, recall that

eg − f 2 1 eG + g E − 2 f F
K ◦X = , H ◦X =
EG − F 2 2 EG − F 2
where

E = |X u |2 = (a + r cos v)2 ,
F = 〈X u , X v 〉 = 0,
G = |X v |2 = r 2 ,
1
e= det(X u , X v , X uu ) = −(a + r cos v) cos v,
|X u ∧ X v |
1
f = det(X u , X v , X uv ) = 0,
|X u ∧ X v |
1
g= det(X u , X v , X v v ) = −r.
|X u ∧ X v |

Hence:
cos v a + 2r cos v
K ◦ X (u, v) = , H ◦ X (u, v) = − .
r (a + r cos v) r (+r cos v)

We can then conclude that the parabolic points are the bottom and top circles,
the elliptic points are the "outer" points and the hyperbolic points are the "in-
ternal" points.

23
(9) We know from Exercise 2 - (3) that a parametrization for a surface of revo-
lution is given by

X (u, v) = x(u)a + y(u) cos vb − y(u) sin vc, (u, v) ∈ I × J

where I and J are intervals, J of length 2π, a is a unit vector in the direction of
the axis of rotation and {a, b, c} is a positively oriented orthonormal basis of R3 .
From the same exercise, we also know that α : I −! R3 given by

α(u) = x(u)b + y(u)c, y >0

is a local parametrization of the generating curve.

We have:

X u = x 0 (u), y 0 (u) cos v, −y 0 (u) sin v ,

¡ ¢
¡ ¢
X v = 0, −y(u) sin v, −y(u) cos v ,
X uv = 0, −y 0 (u) sin v, −y 0 (u) cos v ,
¡ ¢

X uu = x 00 (u), y 00 (u) cos v, −y 00 (u) sin v ,

¡ ¢
¡ ¢
X v v = 0, −y(u) cos v, y(u) sin v .

Then

E = x 0 (u)2 + y 0 (u)2 ,
F = 0,
G = y(u)2 .

Also, q
|X u ∧ X v | = y(u) x 0 (u)2 + y 0 (u)2
and therefore
x 0 (u)y 00 (u) − x 00 (u)y 0 (u)
¡ ¢
e= p ,
x 0 (u)2 + y 0 (u)2
f = 0,
−x 0 (u)y(u)
g=p ;
x 0 (u)2 + y 0 (u)2
Then, the mean curvature is given by
1 eG + g E − 2 f F
H ◦X =
2 EG − F 2
2
¡ 0 00 00 0
¢ ³ 0 2 0 2
´
1 y(u) x (u)y (u) − x (u)y (u) − x (u) + y (u) x 0 (u)y(u)
= p ,
2 x 0 (u)2 + y 0 (u)2

24
so it is identically zero if and only if
³ ´
y(u) x 0 (u)y 00 (u) − x 00 (u)y 0 (u) − x 0 (u) x 0 (u)2 + y 0 (u)2 = 0.
¡ ¢

Let us restrict our attention to the open subset of I in which x 0 (u) 6= 0 for a mo-
ment. In this set the function x is invertible, by the inverse function theorem,
and hence we can parametrize y by x. In particular, note that
à !2 d y d2x
d2y dx d2y d u d u2
= + .
d u2 d u d x2 dx
du

Indeed, by the Inverse Function Theorem we have that

du 1
= .
dx dx
du
Taking the derivative with respect to x again,we arrive at

d2 u dx 3 d2 x
µ ¶
= −
dx 2 du du 2

Substituting in the original expression, the term inside the first parenthesis be-
comes à !3
d2y dx d y d 2x d y d 2x
+ − .
d x2 d u d u d u2 d u d u2

Dividing the whole expression by x 0 (u)3 we get

µ ¶2
d2 y dy
y 2
−1− = 0, (3.2.2)
dx dx

d2 y
which implies that > 0 and that, taking derivatives again,
dx 2

dy d3 y d2 y
3 2 d log
d y dy d y 3 d log y dx 2 ,
y 3
− 2
= 0 =⇒ dx = dx
2
=⇒ =
dx dx dx y d y dx dx
dx 2
from which we deduce that
dy
= C y,
dx

25
for some C > 0. The solutions to this equation are of the form
p p
y(x) = λ cosh C x + µ sinh C x, λ, µ ∈ R .

Since such solution must also satisfy (3.2.2), we have that C λ2 − C µ2 = 1 and
hence
1 p
y(x) = p cos( C x + L)
C
for some L ∈ R, what is the equation of a catenary.
Since the generating curve must be connected, either it is indeed a catenary
(and not pieces of one) or x 0 (u) = 0 everywhere, in which case the curve is an
open half-line perpendicular to the rotation axis and the surface is a plane.

(10) Let X and α be like in the previous exercise. We may assume that α is
parametrized by the arc length, so the coefficients of the second fundamental
form become

E = 1,
F = 0,
G = y 2,
e = x 0 y 00 − x 00 y 0 ,
f = 0,
g = −x 0 y.

We also have that x 02 + y 02 = 1, which gives, after derivation, x 0 x 00 + y 0 y 00 = 0.

Then
x 02 y 00 y x 02 y 00 y 00
e =− 0 − = −
x x0 x0
and therefore
eg − f 2 x 0 y y 00 y 00
K= = = .
EG − F 2 x0 y 2 y
We then see that the surfaces of revolution with constant Gauss curvature are
those such that the vertical component of the generating curve satisfies

K y − y 00 = 0,

and this completes the exercise.

(11) Denote the ellipsoid by S. We begin by noting that S is the inverse image
of a regular value of the function

x2 y 2 z2
f (x, y, z) = 2 + 2 + 2 .
a b c

26
Indeed, if f (x, y, z) = 1 then

x y z
µ ¶
∇ f (x, y, z) = 2 2 , 2 , 2 =6 0.
a b c

It is clear that S = f −1 ({1}).

The gradient gives us a Gauss map defined on S:

∇ f (x, y, z)
N (x, y, z) = .
|∇ f (x, y, z)|

Then, there exists a nonvanishing differentiable function h defined on S such

that
x y z
µ ¶
h(x, y, z)N (x, y, z) = 2 , 2 , 2 .
a b c
Using the product rule and the fact that the right-hand side is linear, for all
p = (x, y, z) ∈ S and v = (v 1 , v 2 , v 3 ) ∈ T p S we have that

v1 v2 v3
µ ¶
(d h)p (v)N (p) + h(p)(d N )p (v) = 2 , 2 , 2 .
a b c

The point p is umbilical if and only if (d N )p is a multiple of the identity. We

claim that this happens if and only if the left-hand side vanishes after scalar
multiplication by N (p) ∧ v. It is clear that the first term vanishes, since (N (p) ∧
v) ⊥ N (p). Suppose (d N )p is a multiple of the identity. Then h(p)(d N )p (v) ∈
T p S and the second term vanishes as well. On the other hand, assume that
³ ´
h(p)(d N )p (v) · (N (p) ∧ v) = 0.

Then (d N )p (v) lies in the plane spanned by N (p) and v. But since (d N )p (v) ∈
T p S, if follows that it is a multiple of v, and therefore the claim follows.
From the previous paragraph, we conclude that

v1 v2 v3
µ ¶
, , · (N (p) ∧ v) = 0,
a2 b2 c 2
³ ´
that is, av 12 , bv 22 , vc 23 lies on the plane spanned by N (p) and v. Since these three
vectors are linearly dependent, we have that
¯ v1 v2 v3 ¯
¯ ¯
¯ ¯
¯ a2 b2 c 2 ¯
¯ v 1 v 2 v 3 ¯ = 0, if N (p) · v = 0.
¯ ¯
¯x y z ¯¯
¯
¯ 2
a b2 c 2
¯

27
But the determinant is a quadratic form in the variables v i which vanishes on a
plane:

z 1 x 1 y 1
µ ¶ µ ¶ µ ¶
1 1 1
q(v) = 2 2 − 2 v 1 v 2 + 2 2 − 2 v 2 v 3 + 2 2 − 2 v 1 v 3
c a b a b c b c a

Hence, the matrix of this quadratic form is

 ¶
z 1 y 1
µ ¶ µ
1 1
0 2 2
− 2 −
b 2 µ c 2 a 2 ¶
 
 µ ¶ c a b
1 z 1
 1 x 1 1 
A=  2 2− 2 0 − .

2 c
y 1 a b a2 b2 c 2 
x 1
µ ¶ µ ¶
1 1 
− − 0
 
b2 c 2 a2 a2 b2 c 2

Since q is degenerate, we must have det A = 0, what is true if and only if x y z = 0.

This happens if and only if x = 0 or y = 0 or z = 0. Indeed, note that if two of
them is zero, the fact that q vanishes on a plane forces the third to be zero,
which is impossible.
Now, suppose x = 0. From N (p) · v = 0 we have that

y c2
v3 = − v2.
b2 z
Then
à ¶!
z 1 y c2 1
µ ¶ µ
1 1
q(v) = 2 2 − 2 − 2 − v 1 v 2 = 0 ∀v 1 , v 2 ∈ R.
c a b b z c 2 a2

Then à !
z 1 y2 c2 1 z2 y 2 a2 − c 2
µ ¶ µ ¶
1 1
− = 2 − =⇒ 2 = 2 2 < 0,
c 2 a2 b2 b z c 2 a2 c b b − a2
which is absurd.
If y = 0, by the same argument we arrive at
à !
x2 z2 b2 − a2
= ,
a2 c 2 c 2 − b2

which is feasible.
If in turn z = 0, then à !
x2 t 2 a2 − c 2
= < 0,
a2 b2 c 2 − b2

28
which is also absurd. Hence, the only possible option is y = 0.
We can conclude that the points of coordinates
s s
2
b −a 2 c 2 − b2
x = ±a , y = 0, z = ±c
c 2 − a2 c 2 − a2
are the only four umbilical points of S.

(12) Note that X uu lives in R3 , so we can write it as

X uu = 〈X uu , X u 〉X u + 〈X uu , X v 〉X v + 〈X uu N X 〉N X ,
and analogously for X v v :
X v v = 〈X v v , X u 〉X u + 〈X v v , X v 〉X v + 〈X v v N X 〉N X .
Recall that R3 = T X S ⊕〈N X 〉. Then, taking the inner product 〈X uu , X v v 〉 we have:
〈X uu , N X 〉〈X v v , N X 〉 = 〈X uu , X v v 〉 − 〈X uu
T
,WvTv 〉,
where T represents the components tangent to the surface. Analogously,
〈X uv , N X 〉2 = |X uv |2 − |X uv
T 2
| .
Hence, by the formula for the Gauss curvature we get
(K ◦ X )EG = eg − f 2 = 〈X uu , N X 〉〈X v v , N X 〉 − 〈X uv , N X 〉2
T
= 〈X uu , X v v 〉 − 〈X uu ,WvTv 〉 − |X uv |2 + |X uv
T 2
| .
Now, since F = 0 we have that
〈X uu , X v v 〉 − |X uv |2 = 〈X uu , X v 〉v − 〈X uv , X v 〉u
= −〈X u , X vu 〉v − 〈X uv , X v 〉u
1 1
= − E v v − G uu
2 2
and also that
T 1 Eu 1 Ev
X uu = Xu − Xv ,
2 E 2 G
T 1 Ev 1 Gu
X uv = Xu + Xv ,
2 E 2 G
1 Gu 1 Gv
X vTv = − Xu + Xv ,
2 E 2 G
from what follows that
1 1 1 E v2 1 G u2 1 E u G u 1 E v G v
(K ◦ X )EG = − E v v − G uu + + + + ,
2 2 4 E2 4 G2 4 E2 4 G2
which equals the desired expression.

29
(13) Let p be the elliptic point of S. Then K (p) > 0, by hypothesis. Also, since
the principal curvatures are constant, then the conditions of Hilbert’s Theorem
are satisfied. Then p is an umbilical point, that is, k 1 = k 2 = k. But since they
are constant, then S is totally umbilical, and therefore is either an open subset
of a plane or of a sphere. But planes have null Gauss curvature everywhere, and
K > 0 in all of S. Hence, S is an open subset of a sphere.

(14) A dilatation with center a ∈ R3 and ratio µ ∈ R∗ is given by

φ(p) = a + µ(p − a).

Note that
(d φ)p (v) = µv, v ∈ T p S.
Hence, Tφ(p) S 0 = (d φ)p (T p S) = T p S. That means that if N is a Gauss map for S,
then N 0 = N ◦ φ−1 is a Gauss map for φ(S). Then, by the Chain Rule,
à !
0 q −a 1
N (q) = N + a =⇒ (d N 0 )φ(p) = (d N )p
µ µ

and therefore
1 1 1
σ0φ(p) = σp , K 0 (φ(p)) = K (p), H 0 (φ(p)) = H (p),
µ µ2 µ

which completes the exercise.

(15) Note that for every p ∈ S

1 〈p, v〉
(d φ)p (v) = 2
v −2 p, v ∈ T p S.
|p| |p|4

Let N be a local Gauss map for S. Then

〈N (p), p〉
N 0 (φ(p)) = N (p) − 2 p
|p|2

is a local Gauss map for S 0 . We then have that

p
(d N 0 )φ(p) (d φ)p (v) =(d N )p (v) − 2〈(d N )p (v), p〉
|p|2
p 〈N (p), p〉
+ 4〈N (p), p〉〈v, p〉 2 − 2 v.
|p| |p|2

30
It is then easy to see that if e i is a principal direction with associated principal
curvature k i , then (d φ)p (e i ) is a principal direction of Tφ(p) S 0 :

〈N (p), p〉 p p
(d N 0 )φ(p) (d φ)p (e i ) = −k i e i − 2 e i + 2k i 〈e i , p〉 + 4〈N (p), p〉〈e i , p〉
|p|2 |p|2 |p|4
à ! à !
〈N (p), p〉 〈N (p), p〉 p
= − ki + 2 2
e i + 2 ki + 2 2
〈e i , p〉 2
|p| |p| |p|
à !
〈N (p), p〉
= |p|2 k i + 2 (d φ)p (e i ).
|p|2

From the last line we conclude that

k i0 (φ(p)) = |p|2 k i (p) + 2〈N (p), p〉.

Therefore,

K 0 ◦ φ(p) = k 10 (φ(p))k 20 (φ(p))

= |p|4 k 1 k 2 + |p|2 k 1 2〈N (p), p〉 + |p|2 k 2 2〈N (p), p〉 + 4〈N (p), p〉2
= |p|4 K (p) + 4|p|2 H (p) + 4〈N (p), p〉2 .

and
H 0 ◦ φ(p) = |p|2 H (p) + 2〈N (p), p〉,
which completes the exercise.

(16) We begin noting that

(d F r )p (v) = v + r (d N )p (v), v ∈ T p S.

Then it is clear that

(d F r )p (e i ) = (1 − r k i )e i .
Since F r is a diffeomorphism, its derivative is injective (since it is a linear iso-
morphism), hence 1 − r k i 6= 0. Hence

(1 − r k 1 )(1 − r k 2 ) = 1 − r k 1 − r k 2 + r 2 k 1 k 2 = 1 − 2r H (p) + r 2 K (p) 6= 0,

which completes the solution of the first item.

Let X : U −! S be a parametrization of S and let p = X (q) for some q ∈ U . Then

(F r ◦ X ) : U −! S r

31
is a parametrization of S r at F r (p). Then

TFr (p) S r = (d F r ◦ X )Fr (p) (R2 )

= (i d + r (d N )p )(T p S)
= T p S,

which concludes the second item.

The affine normal line at p is given by the equation

x = p + λN (p).

The affine normal line at F r (p) is given by

y = F r (p) + µN 0 (F r (p)) = p + (r + µ)N (p),

and so they coincide. Thus we have the third item.

We have:

(d N 0 )Fr (p) (i d + r (d N )p )(e i ) = (d N )p (e i ) = −k i e i .

Hence
ki
(d N 0 )Fr (p) (e i ) = − ei ,
1 − rk i
which concludes the fourth item.
A straightforward calculation gives the fifth item.
We know that 1 − 2r H + r 2 K 6= 0, from which follows the first claim. Using the
formula obtained in the fifth item we conclude the exercise.

(17) Assume the common tangent plane P to be given by the equation z = 0.

Consider the function h : P −! R given by

h(u, v) = f 1 (u, v) − f 2 (u, v).

The function h measures the difference of heights relative to P . By hypothesis,

h ≥ 0 and h(0, 0) = 0 is a minimum (we are assuming p = (0, 0, f 1 (0, 0))). Then,
by Proposition 3.35 the Hessian matrix of h is positive definite. But from the
proof of Theorem 3.47 we know that
" #
2 ki 1 0
(d f i )p = ,
0 ki 2

32
which is the matrix of the second fundamental form σi with respect to the basis
of the tangent plane given by the principal directions, chosen to be (1, 0, 0) and
(0, 1, 0). Hence

(d 2 h)p (v) = (d 2 f 1 )p (v) − (d 2 f 2 )p (v) = σ1 (v, v) − σ2 (v, v) ≥ 0,

which concludes the first item.

If σ1 (v, v) > σ2 (v, v) for every v 6= 0 then, by Proposition 3.35 h has a minimum
point at the origin with positive definite Hessian matrix. But h(0, 0) = 0 hence
there is a neighborhood of p where f 1 ≥ f 2 , concluding the exercise.

(18) As the books says, if a ≤ 0 the result is trivial. If a > 0, take the sphere S2R
with radius R < r and constant mean curvature of 1/R with respect to the inner
normal. Approaching this sphere to the surface from above and from below we
obtain at least two points of tangency between them. At least at one of them
the normal to S coincides with the inner normal of the sphere. Now, using the
hypothesis and exercise (17), we have that

1
a ≤ H (p) ≤
R
from what follows that aR ≤ 1 for all R < r , what gives the desired result.

(19) We know that the Gauss curvature is constant, because it is the product
of the principal curvatures. Hence, it is either positive or non-positive. is it is
positive, by Theorem 3.47 every point is umbilical.

33
Chapter 4

Separation and Orientability

4.1 Exercises in the text

Exercise 4.10 Consider the following parametrizations:
α(t ) = p + v t , αn (t ) = p n + v n t , n ∈ N.
If p n ! p and v n ! v, then for all t ∈ R we have that
lim αn (t ) = lim p n + v n t = lim p n + t lim v n = p + v t = α(t ),
which completes the proof.

Exercise 4.11 By definition, there are parametrizations αn (t ) = q n + t w n such

that
αn (t ) ! α(t ) = q + t w
for all t ∈ R. Hence q n ! q and w n ! w. On the other hand, for each n we have
that
p n = q n + t n w n , v n = λn w n ,
from what follows that
〈p n − q n , w n 〉 〈v n , w n 〉
tn = , λn = .
|w n |2 |w n |2
But the right-hand sides of the two equalities above are convergent, so
tn ! t , λn ! λ
for some t , λ ∈ R. Hence
p = q + t w = α(t ) ∈ R
and
v = λw
is a direction of R.

34
Exercise 4.12 Let R n be given by the parametrization αn (t ) = p n + t v n , where
p n belongs to the compact set given and v n ∈ S2 for all n ∈ N. By compactness,
there is a subsequence of the (p n ) which converges to some p in the compact
set. Let v n be the associated subsequence of directions. Again, by compactness,
there is a subsequence such that v n ! v for some v ∈ S2 . Hence, by Exercise
4.10, R n converges to the line R parametrized by

α(t ) = p + t v,

as we wanted.

Exercise 4.13 We can assume that all p n are in a neighborhood of p where we

have a Gauss map at our disposal. By definition, we can find a sequence of
direction vectors v n converging to the direction vector v of R. Now,

0 = lim〈v n , N (p n )〉 = 〈v, N (p)〉,

which completes the proof.

4.2 Exercises at the end of the chapter

(1) Let x ∈ Nδ (S). Then x lies in the segment determined by N (p) for some
unique p ∈ S. Denote
π(x) = p.
Also, x is at a distance r from p. Denote by r (x) this distance if x is on the side
to which N (p) points or minus this distance otherwise. Note that

π = p S ◦ F −1 , r = p (−δ,δ) ◦ F −1

where p S , p (−δ,δ) are the respective projections from S × (−δ, δ), and hence are
differentiable.
As is expected, r −1 ({0}) = S.
By the way we have defined π and r , if we reverse the orientation of the surface
π remains the same and r changes sign.
If 0 < ρ < δ, then Nρ (S) is another tubular neighborhood of S. Indeed, the re-
striction of F remains a diffeomorphism.
Now, let p ∈ S, e 1 , e 2 ∈ T p S be the principal directions of S at p and t ∈ (−δ, δ).
Then 1 − t k i (p) > 0 and
1
(d π)p+t N (p) (e i ) = ei , (d π)p+t N (p) (N (p)) = 0,
1 − t k i (p)

35
 (d r )p+t N (p) (e i ) = 0, (d r )p+t N (p) (N (p)) = 1.
To show the above claims, take a given i ∈ {1, 2}, let α : I ⊂ R −! S, with 0 ∈ I
such that
α(0) = p, α0 (0) = e i .
Define then the curve β in R3 given by

β(s) = F (α(s), t ) = α(s) + t N (α(s)), s ∈ I.

It is clear that β belongs to Nδ (S). We also have that

β(0) = p + t N (p), β0 (0) = (1 − t k i (p))e i = (d F )(p,t ) (e i , 0)

and that
π(β(s)) = α(s), r (β(s)) = t
for all s ∈ I .
Since F |S×(−δ,δ) is a diffeomorphism, we have that (d F )(p,t ) (e i , 0) 6= 0, hence 1 −
t k i (p) 6= 0. But (−δ, δ) is connected and 1 − t k i (p) is positive for t = 0. Hence
1 − t k i (p) > 0 in S × (−δ, δ). Moreover,
¯
0 d ¯
(d π)p+t N (p) (β (0)) = (d π)p+t N (p) ((1 − t k i (p))e i ) = ¯ π(β(s)) = α0 (0) = e i ,
¯
ds ¯
s=0

and therefore
1
(d π)p+t N (p) (e i ) = ei .
1 − t k i (p)
We also have that
¯
d ¯
((d r )p+t N (p) ((1 − t k i (p))e i ) = (d r )p+t N (p) (β0 (0)) = r (β(s)) = 0.
¯
¯
ds ¯
s=0

Now consider the segment

γ(s) = F (p, t + s) = p + (t + s)N (p), s ∈ I.

Then
γ(0) = p + t N (p), γ0 (0) = N (p),
and
π(γ(s)) = p, r (γ(s)) = t + s
for all s ∈ I . Then
¯
d ¯
(d π)p+t N (p) (N (p)) = (d π)p+t N (p) (γ0 (0)) = π(γ(s)) = 0
¯
¯
ds ¯
s=0

36
and similarly
(d r )p+t N (p) (N (p)) = 1,
and the claim is proved.

(2) Let N (p) be the ineer unit normal. For t small enough, p t = p + t N (p) ∈ Ω
and from Exercise (1) of Chapter 3 we know that p is the isolated minimum on
S of the distance function f t (q) = |p t − q|2 . Then there is a neighborhood of V
of p such that V ∩ B t (p t ) = ; and S2t (p t ) ∩ V = {p}, where S2t (p t ) is the sphere
of radius t and center p t . Note that S2t tends to {p} as t ! 0. Since S \ V is a
compact set that does not contain p, we have that S ∩ B t (p t ) = ; if t is small,
because
lim d (S2t (p t ), S \ V ) = d ({p}, S \ V ) > 0
and the distance is continuous. Now, since B t (p t ) is connected, either it is in-
side or outside S, that is, B t (p t ) ⊂ Ω or B t (p t ) ⊂ R3 \Ω. But p t ∈ B t (p t ) ∩ Ω,
hence we can only have the first possibility happening and S2t (p t ) is the sphere
we were looking for.

(3) We want to show that

(d (r − f ◦ π))x 6= 0 if r (x) = f ◦ π(x).

Let v ∈ R3 be given by
v = v 1 e 1 + v 2 e 2 + v 3 N (p).
Then

(d (r − f ◦ π))x (v) =(d r )x (v) + (d f )π(x) (d π)x (v)

v1 v2
µ ¶
=v 3 + (d f )π(x) e1 + e2
1 − r (x)k 1 (π(x)) 1 − r (x)k 2 (π(x))
v1
=v 3 + (d f )π(x) (e 1 )
1 − r (x)k 1 (π(x))
v2
+ (d f )π(x) (e 2 ),
1 − r (x)k 2 (π(x))
which is clearly not always zero.

(4) Fix a ∈ R3 . Since S is compact, there are points in S which maximizes (glob-
ally) the square of the distance from a. At these points, we have that

p + t (p − a) ∉ S ∀t > 0,

that is, points in the half-line starting at a and passing through p cannot belong
to S (otherwise, p would not maximize the distance). Since this set A = {p +
t (p − a) : t > 0} is not bounded but is connected we have that A ⊂ R \Ω.

37
Now, from Example 3.39 we know that

a = p + λN (p)

for some λ ∈ R and also that

|v|2 − λσp (v, v) ≤ 0, v ∈ T p S.

Hence
p + t (p − a) = p − t λN (p) ∈ R3 \Ω, t > 0.
If N (p) is the inner normal, then t λ > 0 and hence

λ = |p − a|.

Therefore,
|v|2
σp (v, v) ≥ > 0, v ∈ T p S,
|p − a|
which concludes the proof.

(5) Fix a ∈ S2 . Let q ∈ S be a minimum of the function h(p) = 〈p, a〉, p ∈ S,

the height function with respect to some plane perpendicular to a. Then, by
Proposition 3.35 and Example 3.36, the second fundamental form of S at q is
semi-definite, with respect to any choice of Gauss map. Therefore K (q) ≥ 0 and
q ∈ P.
Let N be the inner normal map. We claim that N (q) = a. Since q is a critical
point of h we have that

(d h)q (v) = 〈v, a〉 = 0 ∀v ∈ T q S =⇒ N (q) = ±a.

On the other hand, S is over the plane passing through q perpendicularly to a

and therefore Ω is also over this plane, what means that

h(q + t N (q)) = h(q) + t 〈N (q), a〉 ≥ h(q)

for small t . It then follows that 〈N (q), a〉 ≥ 0, which implies that N (q) = a.
Now, look at the elliptic points of the torus of revolution, studied in Exercise (8)
of Chapter 3. The elliptic points are those with v ∈ (−π/2, π/2). Note that none
of them has (0, 0, −1) or (0, 0, 1) as inner normal unit vector. Hence the answer
to the final question is negative.

38
(6) We will assume that S is not compact.
a: Recall that a segment cuts S transversely if it is not tangent to the surface (see
Remark 4.3). If the intersection between a segment and the surface is empty,
then they are trivially transverse. Let p 1 ∈ R3 \S and p 2 ∈ B . Then one of the two
possibilities happen: [p 1 , p 2 ] is already transverse to S or [p 1 , p 2 ] is tangent to
S. In the first case, we are done. If the second holds, then we can move p 2 just
slightly to p 20 such that [p 1 , p 20 ] is transverse to S.
b: Consider the triangle determined by p 1 , p 2 and p 3 . If it does not intersect S
transversely, we can perturb it a little so that it do, without changing the num-
ber of intersections between the lines given and S. The intersection is then a
compact simple curve, of which each connected component is homeomorphic
either to a circle or to a closed interval, whose endpoints will correspond to the
intersections with the lines, which will therefore be twice as many, hence in
even number.
c: Let Ωeven (respectively Ωod d ) be the set of points x ∈ R3 \S from which we
can draw a segment to a point in B which intersects S an even (respectively
odd) number of times. Both these sets are open, in view of Proposition 4.14. By
the previous item these sets are disjoint. Indeed, suppose there is p 1 ∈ Ωeven ∩
Ωod d . It would contradict the statement of the previous item. The set A even is
not empty because for any point in B we can trace a segment to another point
in B , hence the intersection with S is empty. Ωod d is not empty either, because
we can take p ∈ S which minimizes the distance to p 0 and extend this segment a
little to gain a point in Ωod d . Invoking Proposition 4.4 concludes the proof.

(7) Suppose p ∈ Ω. Then, since Ω is open, there is an open neighborhood U of

p in Ω such that (U ∩ R) ⊂ Ω, a contradiction. Hence p ∈ ∂Ω = S.
Now, let δ > 0 be such that Nδ (S) is a tubular neighborhood and let v be a di-
rection vector of R. Since Nδ (S) is open, then p + t v ∈ Nδ (S) for t small. Since
R ∩ Ω = {p}, then r (p + t v) < 0 for all t small, thus attaining a maximum at p.
Then (d r )p (v) = 0. Writing v in terms of e 1 , e 2 and N (p), we see by the results
of Exercise (1) that 〈N (p), v〉 = 0.

(8) If p minimizes the distance to q, then it also minimizes the square of the
distance. By Example 3.39, we know that q lies in the line normal to S passing
through p. Hence
q = p + t N (p)
for some t > 0. Also by Example 3.39, since p is a point of minimum, we have
that
|v|2 ≥ t σp (v, v) ∀v ∈ T p S.

39
In particular, if we take v = e 2 we have, in view of Proposition 3.26, that

t ≤ 1/k 2 (p).

Note that H (p) > 0 implies that k 2 (p) > 0.

(9) Consider the support function of S based on x:

f (p) = 〈p − x, N (p)〉, p ∈ S.

Note that the points through which there is a tangent line passing through x are
precisely those p such that f (p) = 0.
Let p m and p M be the closest and farthest points of S with respect to x, respec-
tively. It is easy to see that f (p m ) > 0 and f (p M ) < 0.
Take the pencil of planes whose axis is the segment joining p m and p M . Since S
is compact, for at least one of these planes the intersection with S is a compact,
connected, closed curve. Now, if we walk along this curve, we must go though
at least two points at which f equals zero. Indeed, if we start at p m , when we
reach p M the function f will have changed its sign. The same happens when
we complete the turn, going from p M to p m . The conclusion then follows.

(10) Let δ > 0 be such that Nδ (S) is a tubular neighborhood and h : R −! R

be a differentiable function such that h(t ) = −δ for t ≤ −δ, h(t ) = δ if t ≥ δ
and which is increasing on the interval (−δ, δ). Moreover, assume that h(0) = 0.
Define f : R3 −! R by

(h ◦ r )(x), x ∈ Nδ (S)


f (x) = δ, x ∈ Ω \ Nδ (S)

−δ, otherwise


where Ω is the inner domain of S and r is the oriented distance in Nδ (S). Then

f −1 ({0}) = r −1 (h −1 ({0})) = r −1 ({0}) = S

and, if p ∈ S, then
(d f )p = h 0 (0)(d r )p 6= 0,
since h is increasing and (d r )p (N (p)) = 1.

(11) Cover the surface with countably many neighborhoods as the ones pro-
vided in Lemma 4.1. Each of them is diffeomorphic to a plane, hence has mea-
sure zero. The claim now follows immediately.

40
Chapter 7

Intrinsic Geometry of Surfaces

7.1 Exercises in the text

Exercise 7.3 Recall the definition of the absolute value of the Jacobian. It {e 1 , e 2 }
is an orthonormal basis of T p S,

| Jac f |(p) = |(d f )p (e 1 ) ∧ (d f )p (e 2 )|

= |〈(d f )p (e 1 ) ∧ (d f )p (e 2 ), (d f )p (e 1 ) ∧ (d f )p (e 2 )〉|.

Now, note that if u, v, x, y are arbitrary vectors in R3 , it holds that

¯ ¯
〈u, x〉 〈v, x〉
¯ ¯
〈u ∧ v, x ∧ y〉 = ¯ ¯. (7.1.1)
¯ ¯
¯〈u, y〉 〈v, y〉¯

Then, since f is a local isometry,

| Jac f |(p) = |(d f )p (e 1 ) ∧ (d f )p (e 2 )|

= |〈(d f )p (e 1 ) ∧ (d f )p (e 2 ), (d f )p (e 1 ) ∧ (d f )p (e 2 )〉|
¯¯ ¯¯¯
¯¯
¯¯〈(d f )p (e 1 ), (d f )p (e 1 )〉 〈(d f )p (e 2 ), (d f )p (e 1 )〉¯¯¯
= ¯¯¯ ¯¯¯
¯¯〈(d f )p (e 1 ), (d f )p (e 2 )〉 〈(d f )p (e 2 ), (d f )p (e 2 )〉¯¯
¯¯ ¯¯¯
¯¯
¯¯〈e 1 , e 1 〉 〈e 2 , e 1 〉¯¯¯
= ¯¯¯ ¯¯¯
¯¯〈e 1 , e 2 〉 〈e 2 , e 2 〉¯¯
=1

for all p ∈ S.
For a proof of (7.1.1) see do Carmo [1].

41
Exercise 7.4 Note that for p = (x, y, 0) ∈ P we have
 
− sin x 0 0
(Jac f )(p) =  cos x 0 0 ,
 
0 1 0

so for u = (u 1 , u 2 ), v = (v 1 , v 2 ) ∈ T p P we have

〈(d f )p (u), (d f )p (v)〉 = 〈(− sin xu 1 , cos xu 1 , u 2 ), (− sin xv 1 , cos xv 1 , v 2 )〉

= u1 v 1 + u2 v 2
= 〈u, v〉,

which completes the proof.

Exercise 7.5 Let S 1 , S 2 , S 3 be surfaces and f : S 1 −! S 2 , g : S 2 −! S 3 be local

isometries. Let p ∈ S 1 and u, v ∈ T p S 1 . Then

〈(d (g ◦ f ))p (u), (d (g ◦ f ))p (v)〉 = 〈(d g ) f (p) ((d f )p (u)), (d g ) f (p) ((d f )p (v))〉
= 〈(d f )p (u), (d f )p (v)〉
= 〈u, v〉

and therefore g ◦ f is a local isometry.

Now let f : S 1 −! S 2 be an isometry (that is, a local isometry and a diffeomor-
phism). Let p ∈ S 1 . Note that the tangent vectors in T f (p) S 2 are of the form
u 0 = (d f )p (u) for u ∈ t p S 1 . For u 0 , v 0 ∈ T f (p) S 2 we have that

〈(d f −1 ) f (p) (u 0 ), (d f −1 ) f (p) (v 0 )〉 = 〈(d f −1 ) f (p) ((d f )p (u)), (d f −1 ) f (p) ((d f )p (v))〉
= 〈(d ( f −1 ◦ f ))p (u), (d ( f −1 ◦ f ))p (v)〉
= 〈u, v〉
= 〈(d f )p (u), (d f )p (v)〉

Consider now the set of all isometries of a surface onto itself. By the first fact
proved above, we have closure of this set with respect to the composition op-
eration. By the second, we have the existence of the inverse element, for every
element of this set. Associativity is trivial, as is the existence of the identity
element. Hence this set forms a group with the operation of composition of
functions.

Exercise 7.6 By hypothesis, for each x ∈ O the map (d F )x : R3 −! R3 is orthog-

onal, that is, preserves the inner product:

〈(d F )x (u), (d F )x (v)〉 = 〈u, v〉 ∀x ∈ O, ∀u, v ∈ R3 .

42
Letting u, v be vectors of the canonical basis of R3 , we have that
* +
∂F ∂F
, = δi j
∂x i ∂x j

for all i , j ∈ {1, 2, 3}. Taking the derivative with respect to x k we get that
* + * +
∂2 F ∂F ∂2 F ∂F
, + , = 0. (7.1.2)
∂x k ∂x i ∂x j ∂x k ∂x j ∂x i

Define * +
∂2 f ∂F
Gi j k = , .
∂x i ∂x j ∂x k
By the equality of mixed partial derivatives and by (7.1.2), we have that G i j k is
symetric in i , j and skew-symmetric in j , k. Hence

G i j k = −G i k j = −G ki j = G k j i = G j ki = −G j i k = −G i j k ,

and therefore G i j k = 0 on O for all i , j , k ∈ {1, 2, 3}. But the partial derivatives
∂F
form a an orthonormal basis of R3 . Hence
∂x i

∂2 F
=0
∂x i ∂x j

on O for i , j ∈ {1, 2, 3}. Then F has the form

F (x) = Ax + b, x ∈O

for some matrix A and some b ∈ R3 , since O is connected. Finally, A = (d F )x

must be orthogonal, which completes the proof.

Exercise 7.39 Take two points in a sphere that are not antipodal. Then there are
two geodesics joining them which togheter form the equator passing through
the points, one greater in length then the other.

43
Bibliography

[1] DO C ARMO, M. P. Differential Geometry of Curves and Surfaces, 2nd ed.

Dover Publications, 2016.

[2] M ONTIEL , S., AND R OS , A. Curves and Surfaces, 2nd ed. No. 69 in Graduate
Studies in Mathematics. American Mathematical Society, 2009.

44
Index

2 - (03), 5 3 - (11), 26
2 - (04), 6 3 - (12), 29
2 - (05), 7 3 - (13), 30
2 - (08), 7 3 - (14), 30
2 - (10), 7 3 - (15), 30
2 - (11), 8 3 - (16), 31
2 - (13), 8 3 - (17), 32
2 - (14), 9 3 - (18), 33
2 - (15), 9 3 - (19), 33
2 - (16), 10 3 - 3.09, 11
2 - (18), 10 3 - 3.10, 11
2 - (21), 10 3 - 3.13, 13
2 - 2.20, 3 3 - 3.14, 13
2 - 2.28, 3 3 - 3.15, 13
2 - 2.29, 3 3 - 3.16, 14
2 - 2.37, 4 3 - 3.17, 14
2 - 2.43, 4 3 - 3.22, 14
2 - 2.44, 4 3 - 3.23, 15
2 - 2.45, 4 3 - 3.24, 15
2 - 2.55, 5 3 - 3.25, 16
2 - 2.62, 5 3 - 3.32, 16
3 - (01), 19 3 - 3.33, 17
3 - (02), 19 3 - 3.40, 17
3 - (03), 20 3 - 3.41, 17
3 - (04), 21 3 - 3.42, 18
3 - (05), 22 3 - 3.50, 18
3 - (06), 22 3 - 3.51, 19
3 - (07), 22 4 - (01), 35
3 - (08), 23 4 - (02), 37
3 - (09), 24 4 - (03), 37
3 - (10), 26 4 - (04), 37

45
4 - (05), 38
4 - (06), 39
4 - (07), 39
4 - (08), 39
4 - (09), 40
4 - (10), 40
4 - (11), 40
4 - 4.10, 34
4 - 4.11, 34
4 - 4.12, 35
4 - 4.13, 35
7 - 7.03, 41
7 - 7.04, 42
7 - 7.05, 42
7 - 7.06, 42
7 - 7.39, 43

46