SOLVING DIFFERENTIAL GEOMETRY

Solutions to Problems Suggested

by Dr. Ghomi

Huy Bui
BS in Mathematics, Georgia Institute of Technology
This is a work of math problems. Hints, solutions and answers are the products
of the author’s efforts. Typos and mistakes are totally unavoidable.

Copyright © 2021 Huy Bui

This work is in copyright. It is subject to statutory exceptions and to the

provisions of relevant licensing agreements; with the exception of the Creative
Commons version the link, which is provided below, no reproduction of any part
of this work may take place without the written permission of the author at
[email protected]

An online version of this work is published at https://n2t.net/ark:/13960/s2q2d3v8gwr

under a Creative Commons Open Access license CC-BY-NC-ND 4.0, which per-
mits reuse, distribution, and reproduction in any medium for noncommercial pur-
poses providing appropriate credit to the original work is given. You may not
distribute derivative works without permission. To view a copy of this license,
visit https://creativecommons.org/licenses/by-nc-nd/4.0

All versions of this work may contain content reproduced under license from third
parties. Permission to reproduce this third-party content must be obtained from
these third-parties directly. When citing this work, please include a reference to
the ARK https://n2t.net/ark:/13960/s2q2d3v8gwr

First published in June 2020/Edition: November 2021

A catalogue record for this publication is available from Internet Archive.

ISBN-13: 979-8657643800 KDP Paperback

Cover image: “An Indochine soldier in late 1800s” by @phamthimeo | Twitter

Cover design by Huy Bui

Printed and sold by Amazon.com Services LLC

 Preface

The present volume contains hints or full solutions to many of the exer-
cises in two volumes of Lecture Notes on Differential Geometry [Gho07] by
Mohammad Ghomi, Professor of Mathematics, Georgia Institute of Tech-
nology, and in the first eight chapters of the text Riemannian Geometry
[Car92] by Manfredo do Carmo, Emeritus Researcher at the IMPA. Solving
problems being an essential part of the learning process, my goal is to pro-
vide those learning and teaching differential geometry with a large number
of worked out exercises. Ghomi’s notes and do Carmo’s textbook cover most
“superb” topics in “classical” differential geometry that are usually taught
at the undergraduate level: the four vertex theorem, Fairy-Milnor theorem,
Gauss’s Theorema Egregium, and the Gauss-Bonnet theorem; and topics
in “modern” differential geometry that are usually taught at the graduate
level: geodesics, the Riemann curvature tensor, Jacobi fields, Hopf-Rinow
and Hadamard theorems, and spaces of constant curvature. Therefore this
solutions manual can be helpful to anyone learning or teaching differential
geometry at both levels.
As Ghomi’s notes and do Carmo’s textbook contain exercises to test the
student’s understanding and extend knowledge and insight into the subject,
I encourage the reader to work through all of the exercises. To make the
solutions concise, I have included only the necessary arguments; the reader
may have to fill in the details to get complete proofs.
Comments and questions on possibly erroneous solutions, as well as
suggestions for more elegant or more complete solutions will be greatly
appreciated.

Huy Bui
Georgia Tech, 2019

3
 Part 1

Curves and Surfaces

 Exercise 1 (The simplest proof of the Cauchy-Schwartz inequality).
For all p and q in Rn . Prove that the equality holds if and only if p = λq
for some λ ∈ R without using either the quadratic formula or the Lagrange
multipliers.
Solution. Assume, as in the second proof above, that kpk = kqk = 1.
Then
0 6 kp − hp, qi qk2
⇐⇒ 0 6 hp − hp, qi q, p − hp, qi qi
⇐⇒ 0 6 hp, pi − (hp, qi)2 − (hp, qi)2 + (hp, qi)2 kqk2
⇐⇒ 0 6 kpk2 − 2(hp, qi)2 + (hp, qi)2 kqk2
⇐⇒ 0 6 1 − 2(hp, qi)2 + (hp, qi)2
⇐⇒ (hp, qi)2 6 1,
giving that
|hp, qi| ≤ 1
= kpk kqk .
Let
p q
p0 = , q0 = .
kpk kqk
Then
kpk0 = kqk0 = 1.
Apply the result above to obtain
 
p q
, 6 1,
kpk kqk
so that
|hp, qi| 6 kpk kqk .
The case p = 0 or q = 0 is trivial.
Exercise 2 (The triangle inequality). Show that

dist(p, q) + dist(q, r) > dist(p, r)

for all p, q in Rn .
 7

Solution (1). Let

p = (p1 , p2 , . . . , pn ),
q = (q1 , q2 , . . . , qn ),
r = (r1 , r2 , . . . , rn ).
Then

kp − rk2
n
X
= [(pi − qi ) + (qi − pi )]2
i=1
Xn n
X n
X
= (pi − qi )2 + (qi − pi )2 + 2 (pi − qi )(qi − ri )
i=1 i=1 i=1

= kp − qk2 + kq − rk2 + 2 hp − q, q − ri
6 kp − qk2 + kq − rk2 + 2 |hp − q, q − ri|
= kp − qk2 + kq − rk2 + 2 kp − qk kq − rk by Cauchy-Schwartz inequality
2
= (kp − qk + kq − rk) .

Therefore

kp − rk 6 kp − qk + kq − rk ,
so that

dist(p; r) 6 dist(p; q) + dist(q; r),

as required.
Solution (2). We have that
kp − rk2
= hp − r, p − ri
= h(p − q) + (q − r), (p − q) + (q − r)i
= hp − q, p − qi + 2 hp − q, q − ri + hq − r, q − ri
6 kp − qk2 + 2 |hp − q, q − ri| + kq − rk2
6 kp − qk2 + 2 kp − qk kq − rk + kq − rk2 by Cauchy-Schwartz inequality
2
= (kp − qk + kq − rk)
Therefore
kp − rk 6 kp − qk + kq − rk ,
8

so that
dist(p; r) 6 dist(p; q) + dist(q; r),
as required.
Exercise 3. Suppose that p, o, q lie on a line and o lies between p and
q. Show that then ∠poq = π.
Solution. o lies between p and q so there exists the real number t < 0
such that p − o = t(q − o). Then

hp − o, q − oi
∠poq := cos−1
kp − ok kq − 0k
ht(q − o), q − oi
= cos−1
kt(q − o)k kq − 0k
t kq − 0k2
= cos−1
|t| kq − 0k2
= cos−1 (−1) since t < 0
=π
Exercise 4 (Sum of the angles in a triangle). Show that the sum of
the angles in a triangle is π (Hint: through one of the vertices draw a line
parallel to the opposite side).
Solution. Let have the triangle in the figure. Since lines a and b are
parallel, it follows that ∠BAC = ∠B 0 CA and ∠ABC = ∠BCA0 . Clearly,
∠B 0 CA + ∠ACB + ∠BCA0 = 180◦ since ∠B 0 CA, ∠ACB, ∠BCA0 does a
complete rotation from one side of the straight line to the other. Hence
∠ABC + ∠BCA + ∠CAB = 180◦ , giving that the sum of the angles in the
triangle ABC is 180◦ or π.
Exercise 5. Find a formula for the curve which is traced by the motion
of a fixed point on a wheel of radius r rolling with constant speed on a
flat surface (Hint: Add the formula for a circle to the formula for a line
generated by the motion of the center of the wheel. You only need to make
sure that the speed of the line correctly matches the speed of the circle)
Solution (Adapted from George F. Simmons, Calculus Gems: Brief
Lives and Memorable Mathematics, McGraw-Hill, 1992.). We assume that
the rolling circle of radius r rolls along the x-axis, starting from a position
in which the center of circle is on the positive y-axis. The curve which is
 9

traced by the rolling circle is the locus of the point P on the circle which
was located at the origin O when the center C is on the y-axis. Let θ be
the angle through which the radius CP turns as the circle rolls to a new
position. If x and y are the coordinates of P , then the rolling of the circle
implies that OB = arclength BP = rθ, so x = OB − AB = OB − P Q =
rθ − r sin θ = r(θ − sin θ). Also, y = BC − QC = r − r cos θ = r(1 − cos θ).
The desired curve therefore has the reparametrization
α
[0, 2π] 3 t 7−→ α(t) = (r(t − sin t), r(1 − cos t)).
This is a cycloid.
Exercise 6. Show that if the position vector and velocity of a planar
curve α : I → R2 are always perpendicular, i.e., hα(t), α0 (t)i, for all t ∈ I,
then α(I) lies on a circle centered at the origin of R2 .
Solution. Let (x(t), y(t)) be the position vector, then the velocity
vector α0 (t) = (x0 (t), y 0 (t)).
α(t), α0 (t) = 0
⇐⇒ x(t)x0 (t) + y(t)y 0 (t) = 0
d 2
⇐⇒ (x (t) + y 2 (t)) = 0
dt
⇐⇒ x2 (t) + y 2 (t) = C,
for some constant C. Thus α(I) lies on a circle centered at the origin of
R2 .
Exercise 7. Compute the length of a circle of radius r, and the length
of one cycle of the curve traced by a point on a circle of radius r rolling on
a straight line.
Solution. The reparametrization of a circle centered at the origin
having radius r is

α
[0, 2π] 3 t 7−→ α(t) = (r cos t, r sin t).
Then

α0 (t) = (−r sin t, r cos t).

The length of this circle is
10

Z 2π
length = α0 (t) dt
0
Z 2π p
= r2 sin2 t + r2 cos2 t dt
0
Z 2π
=r dt since sin2 t + cos2 t = 1
0
= 2πr.

The reparametrization of the first cycle of a cycloid through the origin,

with a horizontal base given by the line y = 0, is

α
[0, 2π] 3 t 7−→ α(t) = (r(t − sin t), r(1 − cos t)).

Then

α0 (t) = (r − r cos t, r sin t).

The length of one cycle of this cycloid is

 11

Z 2π
length = α0 (t) dt
0
Z 2π p
== (r − r cos t)2 + (r sin t)2 dt
0
Z 2π q
=r (1 − cos t)2 + sin2 t dt
0
Z 2π p
=r 1 − 2 cos t + cos2 t + sin2 t dt.
0
Z
p 2π
=r 2(1 − cos t) dt
0
Z 2π r
t t
=r 4 sin2 dt since 1 − cos t = 2 sin2
0 2 2
Z 2π
t
=r 2 sin dt
0 2
t 2π
= −4r cos |0
2
= 4r − (−4r)
= 8r.

Exercise 8. Show that if α : [a, b] → R2 is a closed curve with width

w and length L, then

L
w≤ .
π
Solution. We prove this by contradiction. Assume that L < πw. Let
H be the convex hull of the set of points bounded by the closed curve α.
Let length[M N ] be the width of H , where M and N are points in the
boundary β of H . Stretch β into a circle Cα . Then length[Cα ] ≤ L since
concave arcs are replaced by straight segments. The radius of Cα is

length[Cα ] L πw w
rα = ≤ < = .
2π 2π 2π 2
12

This implies that length[M N ] ≤ 2rα < w, contradicting the the minimality
of w as the width of α. Therefore
L
w≤ .
π
Exercise 9. Show that if the equality in the exercise above holds then
is a curve of constant width.
Solution. Suppose the equality in the exercise above holds, that is,

L
,
w=
π
which is a constant. Therefore the curve has a constant width, for example,
the circle of radius w/2. The reverse is also true. It has been known since
Barbier in 18601, and follows from the Cauchy-Crofton formula, that for
closed planar curves L/w ≥ π, where equality holds only for curves of
constant width. (see Ghomi, Mohammad. (2016). The length, width, and
inradius of space curves. Geometriae Dedicata. 10.1007/s10711-017-0312-
3).
Exercise 10. Show that the curvature of a circle of radius r is 1r ,
and the curvature of the line is zero (First you need to find arclength
parametrizations for these curves).
Solution. We saw earlier in Section 1.1 of Lecture Notes 1 that the
parametrization of a circle of radius r with respect to arc length was

α
[0, 2π] 3 t 7−→ (r cos(t), r sin(t)) ∈ R2 .
First, we need to compute T (t). By definition,

α0 (t)
T (t) = .
kα0 (t)k
So, we must first compute α0 (t).

1 t 1 t
α0 (t) = (r(− sin( )), r( cos( )))
r r r r
t t
= (− sin( ), cos( )).
r r
1E. Barbier. Note sur le probl‘eme de l’aiguille et le jeu du joint couvert. Journal
de math´ematiques pures et appliqu´ees, pages 273–286, 1860.
 13

We can see that

r
0 t t
α (t) = [− sin( )]2 + [cos( )]2
r r
r
t t
= sin2 ( ) + cos2 ( ) = 1.
r r
Thus

t t
T (t) = (− sin( ), cos( )).
r r
It follows that

1 t 1 t
T 0 (t) = (− cos( ), − sin( )),
r r r r
and therefore

κ(t) = T 0 (t)
r
1 t 1 t
= (− cos( ))2 + (− sin( ))2
r r r r
r
1 t t
= (cos2 ( ) + sin2 ( ))
r2 r r
1
= .
r
In other words, the curvature of a circle is the inverse of its radius. This
agrees with our intuition of curvature. Curvature is supposed to measure
how sharply a curve bends. The larger the radius of a circle, the less it will
bend, that is the less its curvature should be. This is indeed the case. The
larger the radius, the smaller its inverse.
Earlier also in Section 1.1 of Lecture Notes 1, we found that the parametriza-
tion of a line which passes through a point p ∈ Rn and is parallel to the
vector v ∈ Rn , with respect to arc length was

α
[0, 2π] 3 t 7−→ p + tv ∈ Rn .
As before, we need to compute T (t) which can be obtained from α0 (t).
14

α0 (t) = v ∈ Rn .
We can see that

α0 (t) = kvk ∈ R.
Thus

v
T (t) = ∈ Rn .
kvk
Notice that each component of T (t) is a scalar, so differentiating T (t), we
get

T 0 (t) = (0, 0, . . . , 0) ∈ Rn ,
and therefore

κ(t) = T 0 (t)
= 0.
Exercise 11. Show that the curvature of a planar curve which satisfies
the equation y = f (x) is given by

|f 00 (x)|
κ(x) = p .
( 1 + (f 0 (x))2 )3
(Hint: Use the parametrization α(t) = (t, f (t), 0), and use the formula in
previous exercise.) Compute the curvatures of y = x, x2 , x3 , and x4 .
Solution. First, let us remark that it is easy to parametrize the curve
given by y = f (x) as a parametrized curve. We can simply use

x = x

y = f (x)

z = 0.


Using t as the name if the parameter. Thus, the position vector of our
curve α(t) = (t, f (t), 0). It follows that
 15

α0 (t) = (1, f 0 (x), 0),

and

α00 (t) = (0, f 00 (x), 0).

Thus

α0 (t) × α00 (t) = (0, 0, f 00 (x))

Hence

α0 (t) × α00 (t) = f 00 (x)

and

α0 (t) =
p
1 + (f 0 (x))2 .
Therefore, by the formula

kα0 (t) × α00 (t)k

κ(t) =
kα0 (t)k3
in Exercise 9 of Lecture Notes 2, we obtain

|f 00 (x)|
κ(x) = p .
( 1 + (f 0 (x))2 )3
We illustrate these formulas with some examples. If y = x, then

|f 00 (x)|
κ(x) = p
( 1 + (f 0 (x))2 )3
|0|
= √
( 1 + 1 2 )3
= 0.
If y = x2 , then
16

|f 00 (x)|
κ(x) = p
( 1 + (f 0 (x))2 )3
|2|
= p
( 1 + (2x)2 )3
2
= √ .
( 1 + 4x2 )3
If y = x3 , then

|f 00 (x)|
κ(x) = p
( 1 + (f 0 (x))2 )3
|6x|
= p
( 1 + (3x2 )2 )3
|6x|
= √ .
( 1 + 9x4 )3
If y = x4 , then

|f 00 (x)|
κ(x) = p
( 1 + (f 0 (x))2 )3
12x2
= p
( 1 + (4x3 )2 )3
12x2
= √ .
( 1 + 16x6 )3
Exercise 12. Let α, β : (−1, 1) → R2 be a pair of C 2 curves with
α(0) = β(0) = (0, 0). Further suppose that α and β both lie on or above
the x-axis, and β lies higher than or at the same height as α. Show that
the curvature of β at t = 0 is not smaller than that of α at t = 0.
Solution. Let v(t) be the tangent at t = 0 of α. Consider the rotation

cos(− tan−1 v(0)) − sin(− tan−1 v(0))

 
R1 = .
sin(− tan−1 v(0)) cos(− tan−1 v(0))
Then R1 ◦ α has the tangent at t = 0 be the x-axis. Hence
 17

(R1 ◦ α)0 (0) = 0.

Let κα (t) be the curvature of α. Then

|f 00 (t)|
Kα (0) |(R1 ◦ α)(0)| by the formula κ(t) = p .
( 1 + (f 0 (x))2 )3
Similarly, let R2 be the rotation for β, then since β ≥ α, so consider an
open neighborhood V sufficiently small and use Taylor expansion to obtain

1
0 ≤ (R2 ◦ β − R1 ◦ α)(t) = [(R2 ◦ β − R1 ◦ α)(0)]00 t2 + o(t2 ) since β ≥ α,
2
implying

(R2 ◦ β)00 (0) ≥ (R1 ◦ α)00 (0) ≥ 0

since R1 ◦ α lies above the x-axis, so it is convex in some neighborhood U
of 0, so (R1 ◦ α)00 ≥ 0. Moreover,

κβ (0) = (R2 ◦ β)00 (0)

= (R2 ◦ β)00 (0)
≥ (R1 ◦ α)00 (0)
= (R1 ◦ α)00 (0)
= κα (0).
Therefore κβ (0) ≥ κα (0).
Exercise 13. Show that if α : I → R2 is a C 2 closed curve which is
contained in a circle of radius r, then the curvature of α has to be bigger
than 1/r at some point. In particular, closed curves have a point of nonzero
curvature. (Hint: Shrink the circle until it contacts the curve, and use the
exercise above).
Solution. Consider an arbitrary pointα(t0 ) on the closed curve α.
Shrink the circle C until it contacts the curve α at the point α(t0 ). Let C1
be the shrunk circle. Then the radius r1 of the circle C1 is smaller or equal
to the radius r of the circle C. Define the rectangular coordinate system
that originate at the point α(t0 ). Exercise 11 of Lecture Note 1 gives that
κ(t0 ) ≥ r11 ≥ 1r where κ(t) is the curvature of the closed curve α.
18

Exercise 14. Let α : I → R2 be a closed planar curve, show that

2π
length[α] ≥
max κ
(Hint: Recall that the width w of α is smaller than or equal to its length
divided by π to show that a piece of α should lie inside a circle of diameter
at least w).
Solution. Recall that the width w of α is smaller than or equal to its
length L divided by π, that is,

L
w≤ ,
π
implying

L ≥ πw. (1)
The definition of the width of a curve gives that a piece of α should lie
inside a circle of diameter at least w. For any point α(t) on this piece, the
exercise above gives the following inequality

1 2
κ(t) ≥ = ,
w/2 w
implying

2 2
w≥ ≥ . (2)
κ(t) max κ(t)
It follows from (1) and (2) that

2π
L≥ .
max κ(t)
Exercise 15. Show that the only curves with constant zero curvature
in Rn are straight lines. (Hint: We may assume that our curve, α : I → Rn
has unit speed. Then κ = kα00 k. So zero curvature implies that α = 0.
Integrating the last expression twice yields the desired result.)
Solution. We may assume that our curve, α : I → Rn has unit speed.
If the curve α is parametrized with respect to arc length then by definition
κ = kα00 k = 0. This is equivalent to α00 = 0. Integrating the last expression
 19

to obtain α0 = v ∈ Rn where kvk = 1. Integrating the last expression to

obtain

α(t) = p + vt,
where kvk = 1, p ∈ Rn . Thus the curve α is a straight line.
Exercise 16. Show that T (t) and N (t) are orthogonal. (Hint: Differ-
entiate both sides of the expression hT (t), T (t)i = 1).
Solution. We have

α0 (t)
T (t) = .
kα0 (t)k
Thus

1
kT (t)k = · α0 (t)
kα0 (t)k
= 1.
Hence

hT (t), T (t)i = kT (t)k2

= 1.
Differentiate both sides of the last expression to obtain

2 T (t), T 0 (t) = 0
since
T 0 (t), T 0 (t) = T (t), T 0 (t)
and
(hu(t), v(t)i)0 = u0 (t), v(t) + u(t), v 0 (t) .
This implies that

T (t), T 0 (t) N (t) = 0

T 0 (t)
since N (t) = kT 0 (t)k . In other words,
20

T 0 (t) hT (t), N (t)i = 0.

Thus hT (t), N (t)i = 0, that is, T (t) and N (t) are orthogonal.
Exercise 17. Check that the osculating circle of α is tangent to α at
α(t) and has the same curvature as α at time t.
Solution. The osculating circle equation of α is

1 1
s(u) − α(t) − N (t) = ,
κ(t) κ(t)
where u ∈ [0, 2π]. The point α(t) satifies this equation, so α(t) ∈ s(u)
1 1
(note that kN (t)k = 1, so κ(t) N (t) = κ(t) ). From the osculating circle
equation, we have
 
1 1 1
s(u) − α(t) − N (t), s(u) − α(t) − N (t) = .
κ(t) κ(t) κ2 (t)
Differentiate both sides of the last expression with respect to u to obtain
 
0 1
s (u), s(u) − α(t) − N (t) = 0,
κ(t)
for all u. Since α(t) ∈ s(u), there exists u0 ∈ [0, 2π] such that s(u0 ) = α(t)
and the expression above at time u0 is
 
0 1
s (u0 ), s(u) − α(t) − N (t) = 0
κ(t)
or
 
0 1
s (u0 ), − N (t) = 0.
κ(t)
This implies

1
s0 (u0 ), N (t) = 0,
κ(t)
so that
 21

s0 (u0 ), N (t) = 0.

Hence the tangent of s(u) at the point α(t) coincides with T (t), so s(u) is
tangent to α at the point α(t).
The curvature of the osculating circle of α is

1
κ(t) = = κ(t)
1/κ(t)

(since the osculating circle of α has radius 1/κ(t)). Thus the osculating
circle of α and α have the same curvature at time t.

Exercise 18. Show that if α has constant curvature c, then (i) p(t)
is a fixed point, and (ii) α(t) − p(t) = 1/c (Hint: For part (i) differentiate
p(t); part (ii) follows immediately from the definition of p(t)).

Solution. (i) Suppose that α has constant curvature c. We can assume

further that the curve α is parametrized with respect to arc length then by
definition

1
p(t) = α(t) + N (t).
c
Differentiate both sides of the last expression to obtain

1
p0 (t) = α0 (t) + N 0 (t).
c
1 0 α0 (t)
= T (t) + N (t) since T (t) = = α0 (t)
c kα0 (t)k
1
= T (t) + (−cT (t)) since N 0 (t) = −cT (t)
c
=0

for all t. Hence p(t) is a fixed point.

(ii) From the definition of p(t), we have
22

1
kα(t) − p(t)k = − N (t)
c
1
= kN (t)k
c
1
=
c
(since kN (t)k = 1). Thus

1
kα(t) − p(t)k = .
c
Exercise 19. Show that

hγ 0 (t) × γ 00 (t), (0, 0, 1)i

κ(t) := .
kγ 0 (t)k3
Solution. Since γ(t) is a planar curve, it may be seen as a curve in
R3 as γ(t) = (x(t), y(t), 0) ∈ R3 . Then

γ 0 (t) = (x0 (t), y 0 (t), 0).

Thus

γ 0 (t)
T (t) =
kγ 0 (t)k
(x0 (t), y 0 (t), 0)
=p .
(x0 (t))2 + (y 0 (t))2
Hence

1 x0 (t)[x0 (t) + y 0 (t)] 00 y 0 (t)[x0 (t) + y 0 (t)]

T 0 (t) = 2 (x00 (t) γ 0 (t) − 0
, y (t) γ 0 (t) − , 0).
kγ 0 (t)k kγ (t)k kγ 0 (t)k
So

1
iT (x) = (−y 0 (t), x0 (t), 0).
kγ 0 (t)k
Therefore
 23

hT 0 (t), iT i
κ(t) =
kγ(t)k
1 00 0 0 0 00 0
= 4 (−x (t)y (t) γ (t) + x (t)y (t) γ (t) )
0
kγ (t)k
x0 y 00 − x00 y 0
= . (1)
kγ 0 (t)k3
Note that

γ 0 (t) = (x0 (t), y 0 (t), 0),

giving

γ 00 (t) = (x00 (t), y 00 (t), 0).

Thus

y 0 (t) 0 0 x0 (t) x0 (t) y 0 (t)

γ 0 (t) × γ 00 (t) = ( 00 , 00 , )
y (t) 0 0 x (t) x00 (t) y 00 (t)
= (0, 0, x0 y 00 − x00 y 0 ).
Hence

hγ 0 (t), γ 00 (t), (0, 0, 1)i x0 y 00 − x00 y 0

= . (2)
kγ 0 (t)k3 kγ 0 (t)k3
Combine (1) and (2) to obtain

hγ 0 (t) × γ 00 (t), (0, 0, 1)i

κ(t) := .
kγ 0 (t)k3
Exercise 20. (i) Compute the total curvature and rotation index of
a circle which has been oriented clockwise, and a circle which is oriented
counterclockwise. Sketch the figure eight curve (cos t, sin 2t), 0 ≤ t ≤ 2π,
and compute its total signed curvature and rotation index.
Solution. The parametrization of a circle, which has been oriented
clockwise, centered at the origin having radius r is
24

α
[0, 2π] 3 t 7−→ α(t) = (r cos(−t), r sin(−t)).
The arclength from time 0 to time t is
Z t
s(t) = α0 (u) du
0
Z t
= k(r sin(−t), −r cos(−t)k du
0
Z tq
= r2 [sin2 (−t) + cos2 (−t)] du
0
Z t
= r du
0
= rt,
giving

s(t)
t=
.
r
The reparametrization by arclength of a circle, which has been oriented
clockwise, centered at the origin having radius r is

β s s s s
[0, 2πr] 3 s 7−→ β(s) = (r cos(− ), r sin(− )) = β(s) = (r cos( ), −r sin( )),
r r r r
so

s s
β 0 (s) = (− sin( ), − cos( ))
r r
1 s 1 s
β 00 (s) = (− cos( ), sin( )).
r r r r
Hence

κ(s) = β 00 (s)
r
1 s s
= [cos2 ( ) + sin2 ( )]
r2 r r
1
= .
r
 25

The total curvature of a clockwise circle is

Z 2πr
totalκ[α] = κ(s) ds
0
Z 2πr
1
= ds
0 r
2πr
= = 2π.
r

Note that the curvatures of circles of the same radius are the same, so
the total curvature of a clockwise circle equals to the total curvature of a
counterclockwise circle, that is, 2π.
The parametrization of a circle, which has been oriented counterclock-
wise, centered at the origin having radius r is

α
[0, 2π] 3 t 7−→ α(t) = (r cos t, r sin t),

so

α0 (t) = (−r sin t, r cos t)

α00 (t) = (−r cos t, −r sin t).

The signed curvature of α at time t is

x0 y 00 − x00 y 0
κ(t) =
kγ 0 (t)k3
r2 sin2 t + r2 cos2 t
= p
( r2 (sin2 t + cos2 t)3
1
= .
r

The total signed curvature of a counterclockwise circle is

26

Z 2πr
totalκ[α] = κ(s) ds
0
Z 2πr
1
= ds
0 r
2πr
= = 2π.
r

The rotation index of a counterclockwise circle is

totalκ[α] 2π
rot[α] = = = 1.
2π 2π

The parametrization of a circle, which has been oriented clockwise,

centered at the origin having radius r is

α
[0, 2π] 3 t 7−→ α(t) = (r cos t, −r sin t),

so

α0 (t) = (−r sin t, −r cos t)

α00 (t) = (−r cos t, r sin t).

The signed curvature of α at time t is

x0 y 00 − x00 y 0
κ(t) =
kγ 0 (t)k3
−r2 sin2 t − r2 cos2 t
= p
( r2 (sin2 t + cos2 t)3
1
=− .
r

The total signed curvature of a clockwise circle is

 27

Z 2πr
totalκ[α] = κ(s) ds
0
Z 2πr
1
= − ds
0 r
2πr
=− = −2π.
r
The rotation index of a clockwise circle is

totalκ[α] −2π
rot[α] = = = −1.
2π 2π
We spit the figure eight circle into two parts. Then, by Theorem 13
(Hopf) of Lecture Note 4, the rotation index of the figure eight curve is +2,
giving the total signed curvature of the figure eight curve is 2 · (2π) = 4π.
Exercise 21. Use the above formula to show that the only closed
curves of constant curvature in the plane are circles.
Solution. If the closed curve has constant curvature, then κ(t) = c
where c is some constant and c 6= 0. This implies that θ0 (t) = c, so

Z t
θ(t) = c ds + θ(0)
0
= ct + θ(0).
We have that
Z t Z t
α(t) = ( cos θ(s) ds, sin θ(s) ds) + α(0)
0 0
Z t Z t
=( cos(θ0 + cs) ds, sin(θ0 + cs) ds) + α(0) where θ0 = θ(0)
0 0
1 1
= ( sin ct, − cos ct) + α(0) choose θ0 = 0
c c
1 1
= (x0 + sin ct, y0 + cos ct) where α(0) = (x0 , y0 ).
c c
Thus α(t) is a circle with the center at (x0 , y0 ) and radius r = 1/ |c|.
The converse is trivial.
28

Exercise 22. Show that if α : I → R2 is a C 4 curve with monotone

nonvanishing curvature, then its evolute is a regular curve which also has
nonvanishing curvature. In particular contains no line segments.
Solution. For the first part, we may asume that kα0 (t)k = 1. We have
that

β(t) = α(t) + r(t)N (t).

Differentiate both sides of the last expression to obtain

β 0 (t) = α0 (t) + r0 (t)N (t) + r(t)N 0 (t)

1
= α0 (t) + r0 (t)N (t) − T (t) since N 0 (t) = − T (t)
r(t)
= α0 (t) + r0 (t)N (t) − α0 (t) since T (t) = α0 (t)
= r0 (t)N (t).
Thus

β 0 (t) = r0 (t)N (t)

= r0 (t) kN (t)k
= r0 (t) since kN (t)k = 1
|κ0 (t)|
=
[κ(t)]2
6 0
= for all t.
Hence β 0 (t) 6= 0 for all t. Therefore β is a regular curve.
For the last part, suppose, towards a contradiction, that β has vanishing
curvature. Then β 0 (t) = const, which contradicts the fact that

|κ0 (t)|
β 0 (t) =
[κ(t)]2
is not constant. Thus β has nonvanishing curvature, so it does not contain
any line segments.
Exercise 23. Show that a curve with monotone curvature cannot have
any bitangent lines.
 29

Solution. Suppose, towards a contradiction, that a curve α with

monotone curvature has a bitangent line. Then then there are two osu-
lating circles of α have the same radius, which contradicts the fact that
osculating circles of α are pairwise disjoint by Kneser’s Nesting Theorem.
Exercise 24. Show that an ellipse has exactly 4 vertices, unless it is a
circle.
Solution. The parametrization of an ellipse having the radius a and
b on the x and y axes respectively, is

α
[0, 2π] 3 t 7−→ α(t) = (a cos t, b sin t).
The signed curvature of this ellipse is

x0 (t)y 00 (t) − x00 (t)y 0 (t)

κ(t) =
[(x0 (t))2 + (y 0 (t))2 ]3/2
−a sin t(−b sin t) + a cos t(b cos t)
=
[(−a sin t)2 + (b cos t)2 ]3/2
ab
= 2
(a sin t + b2 cos2 t)3/2
2

ab
=
[a (1 − cos t) + b2 cos2 t]3/2
2 2

ab
= .
[a + (b − a2 ) cos2 t]3/2
2 2

Thus
(
a
b2
if b > a (cos t = ±1),
min κ(t) = b
a2
if b ≤ a (cos t = 0),
and
(
b
a2
if b > a (cos t = 0),
max κ(t) = a
b2
if b ≤ a (cos t ± 1).
Since the signed curvature of the ellipse has local max and local min at 4
points, it follows that the ellipse has 4 vertices.
30

Exercise 25. Verify the following sentence: A simple closed convex

curve has at least four vertices.
Solution. This is The Four-Vertex Theorem. The following proof for
this theorem is from [Car16].
Before starting the proof, we need a Lemma:
Let α : [0, l] → R2 be a plane closed curve parametrized by arc length
and let A, B, and C be arbitrary real numbers. Then
Z l
dκ
(Ax + By + C) ds = 0, (1)
0 ds
where the functions x = x(s), y = y(s) are given by α(s) = (x(s), y(s)),
and κ is the curvature of α.
Proof of the Lemma. Recall that there exists a differentiable function
θ : [0, l] → R such that x0(s) = cos θ, y0(s) = sin θ. Thus, κ(s) = θ0 (s) and

x00 = −ky 0 , y 00 = kx0 .

Therefore, since the functions involved agree at 0 and l,
Z l
k 0 ds = 0,
0
Z l Z l Z l
0 0
xk ds = − kx ds = − y 00 ds = 0,
0 0 0
Z l Z l Z l
0 0
yk ds = − ky ds = x00 ds = 0.
0 0 0
Proof of the Theorem. Parametrize the curve by arc length, α : [0, l] →
2
R . Since κ = κ(s) is a continuous function on the closed interval [0, l],
it reaches a maximum and a minimum on [0, l]. Thus, α has at least two
vertices, α(s1 ) = p and α(s2 ) = q. Let L be the straight line passing
through p and q, and let β and γ be the two arcs of α which are determined
by the points p and q.
We claim that each of these arcs lies on a definite side of L. Otherwise,
it meets L in a point r distinct from p and q (Fig. 1(a)). By convexity, and
since p, q, r are distinct points on C, the tangent line at the intermediate
point, say p, has to agree with L. Again, by convexity, this implies that L
is tangent to C at the three points p, q, and r. But then the tangent to a
point near p (the intermediate point) will have q and r on distinct sides,
 31

unless the whole segment rq of L belongs to C (Fig. 1(b)). This implies

that κ = 0 at p and q. Since these are points of maximum and minimum
for κ, κ ≡ 0 on C, a contradiction.
Let Ax + By + C = 0 be the equation of L. If there are no further
vertices, κ0 (s) keeps a constant sign on each of the arcs β and γ.We can
then arrange the signs of all the coefficients A, B, C so that the integral in
Eq. (1) is positive. This contradiction shows that there is a third vertex
and that κ0 (s) changes sign on β or γ, say, on β. Since p and q are points
of maximum and minimum, κ0 (s) changes sign twice on β. Thus, there is
a fourth vertex.
Exercise 26. Prove the four vertex theorem for convex curves using
the Schur’s arm lemma.
Solution. Let L be the length of the curve α. Let α1 be the curve
given by

α1 = α|[0, L ] ,
2

i.e., the parametrization of α1 is

L α1
] 3 t 7−→
[0, α1 (t) = α(t),
2
and α2 be the curve given by

α2 = α|[ L ,L] ,
2

i.e., the reparametrization of α2 is

L α2 L
[0, ] 3 t 7−→ α2 (t) = α( + t).
2 2
Since κ1 : [0, 2 ] → R, the curvature of the curve α1 , is continuous and [0, L2 ]
L

is a compact set, so κ1 (t) attains its maximum and mininum on [0. L2 ]. Thus
α1 has at least two vertices. Without loss of generality, we can assume that
κ1 attains its maximum and minimum at t1 and t2 , respectively, such that
0 ≤ t1 < t2 ≤ L2 . We can assume that t1 = 0 by choosing the curve
α1 so that α1 starts at t1 = 0. If α has only two vertices above, then
κ02 (t) > 0. This implies that κ2 (t) ≥ κ2 ( L2 ) = κ1 (0) = κ1 (t1 ) ≥ κ1 (t) for
all 0 ≤ t ≤ L2 . Thus κ2 (t) ≥ κ1 (t). Apply the Schur’s arm lemma to
32

obtain dist(α2 (0), α2 ( L2 )) < dist(α1 (0), α1 ( L2 )). This contradicts the fact
that dist(α2 (0), α2 ( L2 )) = dist(α1 (0), α1 ( L2 )) since α1 (0) = α2 ( L2 ), α1 ( L2 ) =
α2 (0). Hence α2 has at least one vertices. But κ02 (t) changes its sign two
times, so α has at least 4 vertices.

Exercise 27. Verify the inequlaity

Length[α] < Length[α].

Rbq 0
Hint: It is enough to check that a 1 + f (x)2 dx is strictly smaller
than either of the integrals in the above formula for the length of α.

Solution. We first show that

r
f 0 (x) − g 0 (x) 2 p p
1+( ) < 1 + (f 0 (x))2 + 1 + (g 0 (x))2 , (1)
2
which is equivalent to show that

f 0 (x) − g 0 (x) 2
1+( ) < 1 + (f 0 (x))2 + 1 + (g 0 (x))2
2
q
+ 2 [1 + (f 0 (x))2 ][1 + (g 0 (x))2 ]
⇔ 4 + (f 0 (x))2 + (g 0 (x))2 − 2f (x)g 0 (x) < 8 + 4(f 0 (x))2 + 4(g 0 (x))2
q
+ 8 [1 + (f 0 (x))2 ][1 + (g 0 (x))2 ]
⇔ 0 < 3(f 0 (x))2 + 3(g 0 (x))2 + 2f 0 (x)g 0 (x)
q
+ 4 + 8 [1 + (f 0 (x))2 ][1 + (g 0 (x))2 ]
⇔ 0 < 2[(f 0 (x))2 + (g 0 (x))2 ] + [f 0 (x) + g 0 (x)]2 .
q
+ 4 + 8 [1 + (f 0 (x))2 ][1 + (g 0 (x))2 ]

Since the last inequality is obvious, so (1) has been proved.

Integrate both sides of (1) to obtain

Z br Z bp Z bp
f 0 (x) − g 0 (x) 2 0
1+( ) dx < 2
1 + (f (x)) dx + 1 + (g 0 (x))2 dx.
a 2 a a

Moreover, note that f (a) − g(a) ≥ 0 and f (b) − g(b) ≥ 0. So

 33

r
b Z bp
f 0 (x) − g 0 (x) 2
Z
1+( ) dx < [f (a) − g(a)] + 1 + (f 0 (x))2 dx
a 2 a
Z bp
+ 1 + (g 0 (x))2 dx + [f (b) − g(b)].
a

Thus

Length[α] < Length[α].

Exercise 28. Computer the curvature and torsion of the circular helix

(r cos t, r sin t, ht)

where r and h are constants. How does changing the values of r and h
effect the curvature and torsion.

Solution. For the first part, the curvature of α at t is

|T 0 (t)|
κ(t) =
|α0 (t)|
Note that

α(t) = (r cos t, r sin t, ht),

α0 (t) = (−r sin t, r cos t, h),
p
α0 (t) = r2 + h2 .

so

α0 (t) 1
T (t) = 0
=√ (−r sin t, r cos t, h),
|α (t)| r + h2
2
1
T 0 (t) = √ (−r cos t, −r sin t, 0).
r + h2
2

Hence
34

|T 0 (t)|
κ(t) =
|α0 (t)|
1
q
= 2 r2 (cos2 t + sin2 t)
r + h2
r
= 2 .
r + h2
The torsion of α is

hB 0 (t), N (t)i
τ (t) = − .
|α0 (t)|
Note that

1
N (t) = T 0 (t)
|T 0 (t)|
√
r2 + h2 1
= √ (−r cos t, −r sin t, 0)
r r + h2
2

= (− cos t, − sin t, 0)
B(t) = T (t) × N (t)
1 r cos t h h −r sin t −r sin t r cos t
=√ ( , , )
+hr2
2 − sin t 0 0 − cos t − cos t − sin t
1
=√ (h sin t, −h cos t, r)
r2 + h2
1
B 0 (t) = √ (h cos t, h sin t, 0).
r + h2
2

Hence

hB 0 (t), N (t)i
τ (t) = −
|α0 (t)|
h
= 2 (cos2 t + sin2 t)
r + h2
h
= 2 .
r + h2
For the last part, we first consider effects of changing the values of r
and h on the curvature. Let g(r) be the function defined by
 35

r
g(r) =
r2 + h2
h2 − r2
g 0 (r) = 2 .
(r + h2 )2
Thus
r 0 h +∞
g 0 (r) + 0 −
1
κ(r) % 2h &
Let f (h) be the function defined by

r
f (h) =
r2+ h2
2hr
f 0 (h) = − 2 <0 if h, r > 0.
(r + h2 )2
Thus if h is increasing, then κ(t) is decreasing, and if h is decreasing, then
κ(t) is increasing .
We now consider effects of changing the values of r and h on the torsion.
Let m(h) be the function defined by

h
m(h) =
r2
+ h2
r 2 − h2
m0 (h) = 2 .
(r + h2 )2
Thus
h 0 r +∞
m0 (h) + 0 −
1
τ (h) % 2h &
Let n(r) be the function defined by

h
n(r) =
r 2 + h2
2hr
n0 (r) = − 2 <0 if h, r > 0.
(r + h2 )2
Thus if r is increasing, then τ (t) is decreasing, and if r is decreasing, then
τ (t) is increasing.
36

Exercise 29. By setting v 0 = 0 show that

κ
v=T +
B,
τ
and check that v is the desired vector, i.e. hT, vi = const and v 0 = 0.
Solution. For the first part, by setting v 0 = 0, we have that v is a
fixed point. Then

hT, vi = a(t)
=a
since hT, Bi = hT, N i = 0. By setting a = 1, we have that a0 = 0. Thus

v 0 = T 0 (t) + b0 (t)N (t + b(t)N 0 (t) + c0 (t)B(t) + c(t)B 0 (t)

= κ(t)N (t) + b0 (t)N (t) + b(t)[−κ(t)T (t) + τ (t)B(t)] + c0 (t)B(t) − c(t)τ (t)N (t)
since T 0 (t) = κ(t)N (t) and N 0 (t) = −κ(t)T (t) + τ (t)B(t). Hence

v 0 = −κ(t)b(t)T (t) + [κ(t) + b0 (t) − c(t)τ (t)]N (t) + [b(t)τ (t) + c0 (t)]B(t).
Since v 0 = (0, 0, 0), so

−κ(t)b(t) = 0,
0
κ(t) + b (t) − c(t)τ (t) = 0,
b(t)τ (t) + c0 (t) = 0,
κ(t)
giving b(t) = 0 since κ(t) 6= 0 and giving c(t) = τ (t) . Therefore,
κ
v = T + B.
τ
For the last part, it is clear that
D κ E
hT, vi = T, T + B
τ
= hT, T i
=1
= const
and
 37

κ 0
v0 = T 0 + B
τ
κ
= κN + (−τ N )
τ
=0
since κ
τ = const, T 0 = κN and B 0 = −τ N . Thus v is the desired vector.
Exercise 30. Check the converse, that is supposing that the curvature
and torsion of some curve satises the above expression, verify whether the
curve has to lie on a sphere of radius r.
Solution. Let

1
ρ(t) = .
κ(t)
Then

κ0 (t)
ρ0 (t) = − .
κ2 (t)
We have that
s
κ0 (t) 1
− 2 + κ(t)τ (t) r2 − 0 = 0.
κ (t) κ (t)
This implies that

[ρ0 (t)]2
ρ2 (t) + = r2 .
τ 2 (t)
Differentiate both sides of the last expression to obtain

d 2 [ρ0 (t)]2
(ρ (t) + 2 )=0
dt τ (t)
ρ0 (t) 0 ρ0 (t)
⇐⇒ 2ρ(t)ρ0 (t) + 2( ) =0
τ (t) τ (t)
ρ0 (t) ρ0 (t) 0
⇐⇒ 2 [ρ(t)τ (t) + ( ) ] = 0,
τ (t) τ (t)
38

implying

ρ0 (t) 0
ρ(t)τ (t) + ( ) =0 since ρ0 (t) 6= 0 (1)
τ (t)
On the other hand,

d ρ0 (t)
(α(t) + ρ(t)N (t) + B(t))
dt τ (t)
ρ0 (t) 0 ρ0 (t) 0
=α0 (t) + ρ0 (t)N (t) + ρ(t)N 0 (t) + ( ) B(t) + B (t)
τ (t) τ (t)
ρ0 (t) 0
=T (t) + ρ0 (t)N (t) + ρ(t)[−κ(t)T (t) + T (t)B(t)] + ( ) B(t)
τ (t)
0
ρ (t) 0
=[ρ(t)T (t) + ( ) ]B(t) since ρ(t)κ(t) = 1
τ (t)
=0. by (1)
Thus

ρ0 (t)
α(t) + ρ(t)N (t) +
B(t) = p,
τ (t)
which is a fixed point. This implies that

ρ2 (t)
kα(t) − pk = ρ(t)N (t) + B(t)
τ (t)
since N (t) and B(t)
ρ2 (t) 2
= ρ2 (t) + [ ] satisfy kN (t)k = kB(t)k = 1
τ (t) and hN (t), B(t)i = 0.
= r2 .
Therefore α(t) lies on the sphere centered at ρ having radius r.
Exercise 31 (Monge Patch). Let f : U ⊂ R2 → R be a differen-
tiable map. Show that the mapping X : U → R3 , defined by X(u1 , u2 ) :=
(u1 , u2 , f (u1 , u2 )) is regular (the pair (X, U ) is called a Monge Patch).
Solution. The Jacobian of X at p is an 3 × 2 matrix defined by
 
1 0
Jp (X) =  0 1 .
∂f ∂f
∂u1 ∂u2
 39

1 0
But = 1. So the rank of Jp (X) is 2. Thus X(u1 , u2 ) := (u1 , u2 , f (u1 , u2 ))
0 1
is regular.
Exercise 32. Show that f : U ⊂ R2 → R3 is regular at p if and only
if

kD1 f (p) × D2 f (p)k =

6 0.
Solution. Assume that f (u, v) = (f1 (u, v), f2 (u, v), f3 (u, v)) ∈ R3 .
Then

∂f1 (u, v) ∂f2 (u, v) ∂f3 (u, v)

D1 f (p) = ( , , )
∂u ∂u ∂u
∂f1 (u, v) ∂f2 (u, v) ∂f3 (u, v)
D2 f (p) = ( , , )
∂v ∂v ∂v
and
 ∂f1 ∂f1 
∂u ∂v
Jp (f ) ==  ∂f2 ∂f2 
.
∂u ∂v
∂f3 ∂f3
∂u ∂v
If f is regular, then rank of Jp (f ) is 1. This implies that there are two
rows of Jp (f ) are linealy independent, for example, the 2th row and the 3rd
row. Then we have that

∂f2 ∂f2
∂u ∂v 6= 0.
∂f3 ∂f3
∂u ∂v
Thus

∂f2 ∂f3
∂u
∂f2
∂u
∂f3 6= 0 since det(A) = det(AT ).
∂v ∂v
This implies that
v
∂f3 2 ∂f1 2 ∂f2 2
u
u ∂f2 ∂f3 ∂f1
kD1 f (p) × D2 f (p)k = t ∂u ∂u + ∂u ∂u + ∂u ∂u > 0.
∂f2 ∂f3 ∂f3 ∂f1 ∂f1 ∂f2
∂v ∂v ∂v ∂v ∂v ∂v
40

Conversely, if kD1 f (p) × D2 f (p)k =

6 0, then one of the three terms
∂f2 ∂f3 ∂f3 ∂f1 ∂f1 ∂f2
∂u ∂u , ∂u ∂u , and ∂u ∂u has to be different from 0, for exam-
∂f2 ∂f3 ∂f3 ∂f1 ∂f1 ∂f2
∂v ∂v ∂v ∂v ∂v ∂v
∂f3 ∂f1 ∂f3 ∂f3
∂u ∂u ∂u ∂v
ple, ∂f3 ∂f1 6= 0. This implies that ∂f1 ∂f1 6= 0. Thus the rank of
∂v ∂v ∂u ∂v
Jp (f ) is 2, and hence f is regular. Therefore f : U ⊂ R2 → R3 is regular
at p if and only if
kD1 f (p) × D2 f (p)k =
6 0.
Exercise 33 (Surfaces of Revolution). Let α : I → R2 , α(t) =
(x(t), y(t)), be a regular simple closed curve. Show that the image of X : I ×
R → R3 given by

X(t, θ) := (x(t) cos θ, x(t) sin θ, y(t)),

is a regular embedded surface.
Solution. (x(t), y(t)) is a parametrization of α, given z and x2 + y 2 =
[(x(t) cos θ)2 + (x(t) sin θ)2 ] = x2 (t), we can determine t uniquely. Thus
X(t, θ) is one-to-one. Since (x(t), y(t)) is apparametrization of α, it follows
that t is a continuous function of z and of x2 + y 2 , and so is a continious
function of (x, y, z). We now need to prove that X −1 is continuous. Note
that X : U = {(θ, t) : 0 < θ < 2π, a < t < b} → S where S is the image of
X.
If θ 6= π, then

θ sin 2θ
tan =
2 cos 2θ
2 sin 2θ cos 2θ
=
2 cos2 2θ
sin θ
=
1 + cos 2θ
y
x(t)
= x
1 + x(t)
y
= p .
x + x2 + y 2
 41

Thus

y
θ = 2 tan−1 p .
x + x2 + y 2
Hence if θ 6= π, then θ is a continuous function of (x, y, z).
If θ = π, then using the fact that

θ cos 2θ
cot =
2 sin 2θ
and calculations similar to the one above to obtain

y
θ = 2 cot−1 p .
−x + x2 + y 2
Thus θ is a continuous function of (x, y, z). This shows that X −1 is a
continuous function, and hence the image of X : I × R → R3 given by
X(t, θ) := (x(t) cos θ, x(t) sin θ, y(t)),
is a regular embedded surface.
Exercise 34. Show that n : S2 → S2 , given by n(p) := p is a Gauss
map (Hint: Define f : R3 → R by f (p) := kpk2 and compute its gradient.
Note that S2 is a level set of f . Thus the gradient of f at p must be
orthogonal to S2 ).
Solution. Define f : R3 → R by f (p) := kpk2 . Note that S2 is a level
set of f . If α(t) = (x(t), y(t), z(t)) is a parametrized curve in S2 , then

2xx0 + 2yy 0+= + 2zz 0 = 0,

which shows that the vector (x, y, z) is normal to the sphere at the point
(x, y, z). Restricted to the curve α(t), the normal vector

N (t) = (x(t), y(t), z(t))

is a vector function of t, and therefore

dN (t) = (x0 (t), y 0 (t), z 0 (t)) = T (t).

42

Since p ∈ S2 , which is equivalent to hp, pi = 1, it follows that hp0 , pi = 0.

Thus n(p) = p is normal to Tp (S 2 ). and hence n(p) = p is a Gauss map.
Exercise 35. Compute the curvature of a sphere of radius r (Hint:
Use exercise 9).
Solution. The parametrization of the sphere of radius R is

x(θ, ϕ) = (R sin θ cos ϕ, R sin θ sin ϕ, R cos θ).

We have that

xθ = (R cos θ cos ϕ, R cos θ sin ϕ, −R sin θ)

xϕ = (−R sin θ sin ϕ, R sin θ cos ϕ, 0)
xθ × xϕ = (R2 sin2 θ cos ϕ, R2 sin2 θ sin ϕ, R2 sin θ cos θ)
q
kxθ × xϕ k = R4 sin4 θ(cos2 ϕ + sin2 ϕ) + R4 sin2 θ cos2 θ
q
= R4 sin2 θ(sin2 θ + cos2 θ)
= R2 sin θ.
Therefore,

n(θ, ϕ) = (sin θ cos ϕ, sin θ sin ϕ, cos θ).

Let S be the shape operator. Then

Sxθ = −∂θ n(θ, ϕ)

x(θ, ϕ)
= −∂θ ( )
R
1
= − xθ
R
Sxϕ = −∂ϕ n(θ, ϕ)
x(θ, ϕ)
= −∂ϕ ( )
R
1
= − xϕ .
R
 43

Hence the shape operator is equal to

1
S =− ·I
R
where I is the identity operator. The matrix of S in the basis xθ , xϕ is
equal to
1 
R 0
− 1 .
0 R
So, the curvature of the sphere of radius R is

− R1 0
K=
0 − R1
1
= 2.
R
Exercise 36. Compute the Gaussian curvature of a surface of revolu-
tion, i.e., the surface covered by the patch

X(t, θ) = (x(t) cos θ, x(t) sin θ, y(t)).

Next, letting

(x(t), y(t)) = (R + r cos t, r sin t),

i.e., a circle of radius r centered at (R, 0), compute the curvature of a torus
of revolution. Sketch the torus and indicate the regions where the curvature
is postive, negative, or zero.
Solution. We have that

Xt = (x0 (t) cos θ, x0 (t) sin θ, y 0 (t))

Xθ = (−x(t) sin θ, x(t) cos θ, 0)

Xt × Xθ = (−x(t)y 0 (t) cos θ, −x(t)y 0 (t) sin θ, x(t)x0 (t))

p p
kXt × Xθ k = [x(t)y 0 (t)]2 + [x(t)x0 (t))]2 = |x(t)| [x0 (t)]2 + [y 0 (t)]2
44

1
n(t, θ) = p (−x(t)y 0 (t) cos θ, −x(t)y 0 (t) sin θ, x(t)x0 (t))
|x(t)| [x (t)]2 + [y 0 (t)]2
0

E = g11 = hXt , Xt i = [x0 (t)]2 + [y 0 (t)]2

F = g12 = hXt , Xθ i = 0

G = g22 = hXθ , Xθ i = [x(t)]2

Xtt = (x00 (t) cos θ, x00 (t) sin θ, y 00 (t))

Xtθ = (−x0 (t) sin θ, x0 (t) cos θ, 0)

Xθθ = (−x(t) cos θ, −x(t) sin θ, 0)

−x(t)x00 (t)y 0 (t) + x(t)x0 (t)y 00 (t)

L = hXtt , n(t, θ)i = p
|x(t)| [x0 (t)]2 + [y 0 (t)]2
M = hXtθ , n(t, θ)i = 0

[X(t)]2 y 0 (t)
N= p .
|x(t)| [x0 (t)]2 + [y 0 (t)]2
The Gaussian curvature of a surface of revolution

L M
M N
K(t) =
E F
F G
LN − M 2
=
EG − F 2
−[x(t)]2 [y 0 (t)]2 x00 (t) + [x(t)]3 x0 (t)y 0 (t)y 00 (t)
=
[x(t)]4 {[x0 (t)]2 + [y 0 (t)]2 }2
−[y 0 (t)]2 x00 (t) + x0 (t)y 0 (t)y 00 (t)
= .
x(t){[x0 (t)]2 + [y 0 (t)]2 }2
 45

Applying the calculations above for (x(t), y(t)) = (R + r cos t, r sin t) to

obtain

x(t) = R + r cos t
y(t) = r sin t
x0 (t) = −r sin t
x00 (t) = −r cos t
y 0 (t) = r cos t
y 00 (t) = −r sin t
[x0 (t)]2 + [y 0 (t)]2 = r2 (sin2 t + cos2 t) = r2 .

The curvature of a torus of revolution is

r3 cos3 t + r3 sin3 t cos t

K(t) =
r4 (R + r cos t)
cos t(cos2 t + sin2 t)
=
r(R + r cos t)
cos t
= .
r(R + r cos t)

From this expression, it follows that K = 0 along the parallels t = π/2 and
t = 3π/2; the points of such parallels are therefore parabolic points. In the
region of the torus given by π/2 < t < 3π/2, K is negative (notice that
r > 0 and R > r); the points in this region are therefore hyperbolic points.
In the region given by 0 < t < π/2 or 3π/2 < u < 2π, the curvature is
positive and the points are elliptic points (Fig. 3-15).

Exercise 37. Show that hSp (ei (p)), ej (p)i = lij (0, 0) (Hints: First
note that hn(p), ej (p)i = 0 for all p ∈ V . Let γ : (−, ) → M be a
curve with γ(0) = p and γ(0) = ei (p). Define f : (−, ) → M by f (t) :=
hn(γ(t)), ej (γ(t))i, and compute f 0 (0).)

Solution. First note that hn(p), ej (p)i = 0 for all p ∈ V . Let γ : (−, ) →
M be a curve with γ(0) = p and γ(0) = ei (p). Define f : (−, ) → M by
f (t) := n(γ(t)), ej (γ(t)). Since hn(p), ej (p)i = 0 for all p ∈ V and γ(t) ∈ V
for t ∈ (−, ), it follows that
46

f (t) = hn(γ(t)), ej (γ(t))i = 0.

Differentiate both sides of the last expression to obtain

hd n(γ(t)), ej (γ(t))i + hn(γ(t)), d ej (γ(t))i = 0.

This implies that
D E
d nγ(t) (γ 0 (t)), ej (γ(t) = − n(γ(t)), d ejγ(t) (γ 0 (t)) .
For t = 0, then

hSp (ei (p)), ej (p)i = − hn(p), Dij X(0, 0)i

or
hSp (ei (p)), ej (p)i = lij (0, 0),
as required
Exercise 38. Compute the curvature of the graph of z = ax2 + by 2 ,
where a and b are constants. Note how the signs of a and b effect the
curvature and shape of the surface. Also note the values of a and b for
which the curvature is zero.
Solution. The surface z = ax2 + by 2 is a Monge patch, i.e,

X(u1 , u2 ) = (u1 , u2 , au21 + bu22 ),

where f (u1 , u2 ) = au21 + bu22 . The Hessian matrix of f is
 
2a 0
Hess (u1 , u2 ) = ,
0 2b
so det(Hess (u1 , u2 )) = 4ab. The gradient of f is

grad f = (2au1 , 2bu2 ).

Thus

grad f (0, 0) = (0, 0).

 47

Hence

det(Hess f (0, 0))

K(p) =
(1 + kgrad f (0, 0)k2 )2
= 4ab.
If a and b have same signs, then K(p) > 0. If a and b have opposite signs,
then K(p) < 0. If a = 0 or b = 0, then K(p) = 0.
Exercise 39. Let M be the Monkey saddle, i.e., the graph of the
equation z = y 3 − 3yx2 , and p := (0, 0, 0). Show that K(p) = 0, but M is
not locally convex at p.
Solution. For the first part, the surface z = ax2 + by 2 is a Monge
patch, i.e,

X(u1 , u2 ) = (u1 , u2 , u32 − 3u2 u21 ),

where f (u1 , u2 ) = u32 − 3u2 u21 . The Hessian matrix of f is
 
−6u2 −6u1
Hess (u1 , u2 ) = ,
−6u2 6u2
so det(Hess (u1 , u2 )) = −36(u21 +u22 ) and det(Hess (0, 0)) = 0. The gradient
of f is

grad f = (−6u1 u2 , −3u22 − 3u21 ).

Thus

grad f (0, 0) = (0, 0).

Hence

det(Hess f (0, 0))

K(p) =
(1 + kgrad f (0, 0)k2 )2
= 0,
as required.
For the last part, we have that
48

Xu1 (u1 , u2 ) = (1, 0, −6u2 u1 )

Xu2 (u1 , u2 ) = (0, 1, 3u22 − 3u21 ).
Thus

Xu1 (0, 0) = (1, 0, 0)

Xu2 (0, 0) = (0, 1, 0).
The equation of the tangent surface at (0, 0, 0) is

x y z
1 0 0 =0
0 1 0
or

y z
= 0,
1 0
that is,

z = 0.
In a neigborhood of (0, 0, 0) for z = u2 (u22 − 3u21 ), then z > 0 as u2 >
0 and u22 > 3u21 ; and z < 0 as u2 > 0 and u22 < 3u21 . therefore, in a
neighborhood of (0, 0, 0), then M does not lie on one side of the tangent
surface, so M M is not locally convex at (0, 0, 0)
Exercise 40. Show that if ac − b2 > 0, then Q is definite, and if
ac − b2 < 0, then Q is not definite. (Hints: For the first part, suppose
that x 6= 0, but Q(x, y) = 0. Then ax2 + 2bxy + cy 2 = 0, which yields
a + 2b(x/y) + c(x/y)2 = 0. Thus the discriminant of this equation must be
positive, which will yield a contradiction. The proof of the second part is
similar).
Solution. For the first part, let ac − b2 > 0. Suppose, towards a
contradiction, that Q is not definite, i.e, there exists x 6= 0, but Q(x, y) = 0.
Then ax2 + 2bxy + cy 2 = 0, which yields a + 2b(y/x) + c(y/x)2 = 0. Let
t = y/x. Then the last equation becomes ct2 + 2bt + a = 0. This equation
 49

has solutions so its discriminant must be positive, that is, b2 − ac > 0 which
contradicts the fact that ac − b2 > 0. Thus Q is definite.
For the second part, let ac − b2 < 0. Similarly, suppose, towards a
contradiction, that Q is definite, i.e, Q(x, y) 6= 0 whenver x 6= 0. Then
ax2 + 2bxy + cy 2 6= 0 for any x 6= 0. This is equivalent to c(y/x)2 +
2b(y/x) + a = 0 has no solution, so the discriminant of this equation must
be negative, that is, b2 − ac < 0 which contradicts the fact that ac − b2 < 0.
Thus Q is not definite.
Exercise 41. Verify the middle step in the above formula, i.e., show
that dn(Di X) = Di (n ◦ X).
Solution. Let f (u1 , u2 ) = n ◦ X(u1 , u2 ). We have that

Di f (u1 , u2 ) = Di (n ◦ X(u1 , u2 ))
= dn(Di X(u1 , u2 )).
Therefore
dn(Di X) = Di (n ◦ X).

Exercise 42. Let F, G : U ⊂ R2 → R3 be a pair of mappings such

that hF, Gi = 0. Prove that hDi F, Gi = − hF, Di Gi.
Solution. Since hF, Gi = 0 so Di (hF, Gi) = 0. But Di (hF, Gi) =
hDi F, Gi+hF, Di Gi. Thus hDi F, Gi+hF, Di Gi = 0. Therefore, hDi F, Gi =
− hF, Di Gi.
Exercise 43. Show that there exist a patch (U, X) centered at p such
that [gij (0, 0)] is the identity matrix. (Hint: Start with a Monge patch with
respect to Tp M ).
Solution. Consider the following maps (assume that p = (p1 , p2 , f (p1 , p2 ))):
u1 (t) : (−, ) → U given by u2 (t) = (t, p2 ) ,
c1 : (−, ) → R3 given by c1 (t) = √ 1 0 2 X◦u1 (t) = √ 1 0 2 (t, p2 , f (t, p2 )),
1+(fu ) 1+(fu )
u2 (t) : (−, ) → U given by u2 (t) = (p1 , t + p2 ) ,
c2 : (−, ) → R3 given by c2 (t) = √ 1 0 2 X ◦ u2 (t) = √ 1
(p1 , t +
1+(fv ) 1+(fv0 )2
p2 , f (p1 , t + p2 )) (
c1 (t) if t < 0,
γ(t) : (−, ) → R3 given by γ(t) =
c2 (t) if t ≥ 0.
50

It is clear that γ(0) = (p1 , p2 , f (p1 , p2 )) = p. If t < 0, then D1 X ◦γ(t) =

√ 1
0 2
(1, 0, fu0 ) and D2 X ◦ γ(t) = (0, 0, 0). If t ≥ 0, then D1 X ◦ γ(t) =
1+(fu )
(0, 0, 0) and D2 X ◦ γ(t) = √ 1
(0, 1, fv0 ). Thus
1+(fv0 )2
(
0 if i 6= j,
hDi X ◦ γ(t), Dj X ◦ γ(t)i =
1 if i = j.
Therefore [gij (0, 0)] is the identity matrix.
Exercise 44. Show that kv (p) does not depend on γ.
Solution. We have that

kv (p) = n(p), γ 00 (0)

= (n ◦ γ)(0), γ 00 (0)
= − (n ◦ γ)0 (0), γ 0 (0)
= − hd np (v), vi
= hSp (v), vi
= IIp (v).
Therefore kv (p) does not depend on γ.
Exercise 45. Show that IIp is symmetric, i.e., IIp (v, w) = IIp (w, v) for
all v, w ∈ Tp M .
Solution. Let γ(t) = x(u(t), v(t)), where x(u, v) is a parametrization
of S at p and {xu , xv } is the associated basis of Tp (S). Note that γ(0) = p.
We have that

d Np (γ 0 (0)) = d Np (xu u0 (0) + xv v 0 (0))

dN
= (u(t), v(t))|t=0
dt
= Nu u0 (0) + Nv v 0 (0).
In particular, d Np (xu ) = Nu , d Np (xv ) = Nv so to prove that d Np is a self-
adjoint, we need only show that hNu , xv i = hNv , xu i. Take derivative both
sides of hN, xu i = 0 and hN, xv i = 0 with respect to v and u, respectively,
to obtain
 51

hNv , xu i + hN, xuv i = 0

hNu , xv i + hN, xvu i = 0

Thus hNu , xv i = − hN, xvu i = − hN, xuv i = hNv , xu i (since xuv = xvu ).
Hence d Np : Tp (S) → Tp (S) is a self-adjoint linear map, so the shape op-
erator Sp = −d Np is also a self-adjoint linear map. Therefore hSp (u), vi =
hu, Sp (v)i = hSp (v), ui. This implies that IIp (u, v) = IIp (v, u) (Note that
IIp (u, v) = hSp (u), vi).

Exercise 46. Verify the above claim, and show that minimum and
maximum values of IIp are λ1 and λ1 respectively. Thus k1 (p) = λ1 , and
k2 (p) = λ2 .

Solution. We have that

v(θ) = cos θ · e1 + sin θ · e2 .

Then

Sp (v(θ)) = Sp (cos θ · e1 + sin θ · e2 )

= Sp (cos θ · e1 ) + Sp (sin θ · e2 ) since Sp is linear
= cos θ · Sp (e1 ) + sin θ · Sp (e2 ) since Sp is linear
= λ1 cos θ · e1 + λ2 sin θ · e2 since Sp (e1 ) = λ1 e1 and Sp (e2 ) = λ2 e2 .

Thus

IIp (v(θ), v(θ)) = hSp (v(θ)), v(θ)i

= hλ1 cos θ · e1 + λ2 sin θ · e2 , cos θ · e1 + sin θ · e2 i
= λ1 cos2 θ + λ2 sin2 θ

since
(
0 if i 6= j
hei , ej i =
1 if i = j
52

By the hypothesis that λ1 ≤ λ2 , we have that

IIp (v(θ), v(θ)) = cos2 θ · λ1 + sin2 θ · λ2
≤ cos2 θ · λ2 + sin2 θ · λ2
= (cos2 θ + sin2 θ) · λ2
= λ2
and
IIp (v(θ), v(θ)) = cos2 θ · λ1 + sin2 θ · λ2
≥ cos2 θ · λ1 + sin2 θ · λ1
= (cos2 θ + sin2 θ) · λ2
= λ1 .
Thus λ1 ≤ IIp (v(θ), v(θ)) ≤ λ2 for all v(θ) ∈ U Tp (M ), and it is clear
that IIp (v(0), v(0)) = λ1 and IIp (v( π2 ), v( π2 )) = λ2 . Hence minimum and
maximum values of IIp are λ1 and λ1 respectively.
Moreover, we have that

hSp (v), vi = − hd np (v), vi

= − (n ◦ γ)0 (0), γ 0 (0)
= (n ◦ γ)(0), γ 00 (0)
= n(p), γ 00 (0)
= kv (p).
Thus kv (p) = hSp (v), vi = IIp (v, v). Hence k1 = minv kv (p) = minv IIp (v, v) =
λ1 and k2 = maxv kv (p) = maxv IIp (v, v) = λ2 .

Exercise 47. What is det(lij ) equal to?

Solution. Since (lij ) = (lij )(gij )−1 , it follows that

det(lij ) = det(lij ) det[(gij )−1 ]

det(lij ) 1
= since det[(gij )−1 ] =
det(gij ) det(gij )
det(hN, Xij i)
= .
det(hXi , Xj i)
 53

det(hN,Xij i)
Thus det(lij ) = det(hXi ,Xj i) =?.

Exercise 48. Show that Ni = −dn(Xi ) = S(Xi ).

Solution. The matrix of dn(Xi ) is
    −1
a11 a21 e f E F
=−
a12 a22 f g F G


−1
= − lij gij
= − lij .

Thus Ni = −li1 X1 − li2 X2 = dn(Xi ) = −S(Xi ).

Exercise 49. Compute the Christoffel symbols of a surface of revolu-
tion.
Solution. Consider a surface of revolution parametrized by

x(u, v) = (ϕ(v) cos u, ϕ(v) sin u, ψ(v))

0 < u < 2π, a < v < b, ψ(v) 6= 0.
We have that

X1 = (−ϕ(v) sin u, ϕ(v) cos u, 0)

X2 = (ϕ0 (v) cos u, ϕ0 (v) sin u, ψ 0 (v))

X11 = (−ϕ(v) cos u, −ϕ(v) sin u, 0)

X12 = (−ϕ0 (v) sin u, ϕ0 (v) cos u, 0)

X21 = (−ϕ0 (v) sin u, ϕ0 (v) cos u, 0)

X22 = (ϕ00 (v) cos u, ϕ00 (v) sin u, ψ 00 (v))

g11 = hX1 , X1 i = ϕ2 (v) sin2 u + ϕ2 (v) cos2 u = ϕ2 (v)

g12 = hX1 , X2 i = 0 = g21

54

and

g22 = [ϕ0 (v)]2 cos2 u + [ϕ0 (v)]2 sin2 u + [ψ 0 (v)]2

= [ϕ0 (v)]2 + [ψ 0 (v)]2 = 1
since assume that the rotating curve is parametrized by arc-length.Thus
 2 
ϕ 0
(gij ) = ,
0 1
 
ij −1 1 1 0
(g ) = (gij ) = 2 .
ϕ 0 ϕ2
• Γ111 , Γ211 : We have that

hX11 , X1 i = 0,
hX11 , X2 i = −ϕ(v)ϕ0 (v) = −ϕϕ0 .
 1      
Γ11 1 0 0 0
Thus = ϕ12 = . Hence Γ111 = 0, Γ211 =
Γ211 0 ϕ2 −ϕϕ0 −ϕϕ0
−ϕϕ0 .
• Γ112 , Γ212 , Γ121 , Γ221 : We have that

hX12 , X1 i = ϕϕ0 ,
hX12 , X2 i = 0.
!
ϕ0
 1     0
Γ12 1 0 ϕϕ ϕ0
Thus = ϕ12 = ϕ . Hence Γ1 = = Γ121 ,
Γ212 0 ϕ2 0 0 12 ϕ

Γ212 = 0 = Γ221 (Note that Γkij = Γkji ).

• Γ122 , Γ222 : We have that

hX22 , X1 i = 0,
hX22 , X2 i = ϕ0 ϕ00 + ψ 0 ψ 00 .
 1      
Γ22 1 1 0 0 0
Thus = ϕ2 = . Hence
Γ222 0 ϕ2 ϕ0 ϕ00 + ψ 0 ψ 00 ϕ0 ϕ00 + ψ 0 ψ 00
Γ122 = 0, Γ222 = ϕ0 ϕ00 + ψ 0 ψ 00 .
 55

Exercise 50. Show that if M = R2 , then Rijk

l = 0 for all 1 ≤ i, l, j, k ≤
2 both intrinsically and extrinsically.
Solution. We first show the claim extrinsically. If M = R2 , then
X(u, v) = (u, v, 0). Thus Xij = (0, 0, 0) for i, j ∈ {1, 2}. Hence lij =
l
hN, Xij i = 0 for i, j ∈ {1, 2}. Therefore Rijk = lik ljl − lij lkl = 0 for all
1 ≤ i, l, j, k ≤ 2.
We next show the claim intrinsically. We have that
(
0 if i 6= j,
gij =
1 if i = j.
Thus
2
X 1
Γkij = [(gli )j + (gjl )i − (gij )l ]g lk
2
l=1
=0 for all 1 ≤ i, l, j, k ≤ 2.
Hence

2
(Γpik Γlpj − Γpij Γlpk )
X
l
Rijk = (Γlik )j − (Γlij )k +
p=1
=0 for all 1 ≤ i, l, j, k ≤ 2.

Exercise 51. Compute the Riemann curvature tensor for S 2 both in-
trinsically and extrinsically.
Solution. It is convenient to relate parametrizations to the geograph-
ical coordinates on S 2 . Let V = {(θ, ϕ); 0 < θ < π, 0 < ϕ < 2π} and let
x : V → R3 be given by

x(θ, ϕ) = (sin θ cos ϕ, sin θ sin ϕ, cos θ).

By the convention that ϕ(θ) = sin θ, ψ(θ) = cos θ, we have that

X(ϕ, θ) = (ϕ(θ) cos ϕ, ϕ(θ) sin ϕ, ψ(θ))

where 0 < ϕ < 2π, 0 < θ < π, ϕ(θ) > 0. Since S 2 is a surface of revolution
and X(ϕ, θ) is a parametrization of S 2 .
56

We first show the claim extrinsically. Using Exercise 3 of Lecture Notes

12 to obtain

g11 = sin2 θ,
g12 = g21 = 0,
g22 = 1.
Thus
 2 
sin θ 0
(gij ) = ,
0 1
 1 
0
(g ij ) = (gij )−1 = sin2 θ .
0 1
We have that

X1 × X2
N= = (− sin θ cos ϕ, − sin θ sin ϕ, − cos θ),
kX1 × X2 k
l11 = hN, X11 i = sin2 θ,
l12 = l21 = hN, X12 i = 0,
l22 = hN, X22 i = 1,
 2  1   
j ij sin θ 0 sin2 θ
0 1 0
(li ) = (lij )(g ) = = .
0 1 0 1 0 1
Hence
(
1 if i = j,
lij =
0 if i 6= j.
Thus

l
Riii =0 for i = 1, 2, l 6= i,

l
R111 = sin2 θ − sin2 θ = 0,

2
R222 = 1 · 1 − 1 · 1 = 0,
 57

1
R112 = l12 l11 − l11 l21 = 0,

1
R121 = l11 l21 − l12 l11 = 0,

1
R211 = l21 l11 − l21 l11 = 0,

2
R112 = l12 l12 − l11 l22 = − sin2 θ,

2
R121 = l11 l22 − l12 l12 = sin2 θ,

2
R211 = l21 l12 − l21 l12 = 0,

1
R122 = l12 l21 − l12 l21 = 0,

1
R212 = l22 l11 − l21 l21 = 1,

1
R221 = l21 l21 − l22 l11 = −1,

2
R122 = l12 l22 − l12 l22 = 0,

2
R212 = l22 l12 − l21 l22 = 0,

2
R221 = l21 l22 − l22 l12 = 0.
We next show the claim intrinsically. Using Exercise 3 of Lecture Notes 12
to obtain

Γ111 = 0,
Γ211 = − sin θ cos θ,
cos θ
Γ112 = Γ121 = ,
sin θ
Γ212 = Γ221 = 0,
Γ122 = 0,
Γ222 = ϕ0 ϕ00 + ψ 0 ψ 00 = cos θ(− sin θ) + (− sin θ)(− cos θ) = 0.
58

Thus

l
Riii = (Γlii )i − (Γlii )i + Γ1ii Γl1i − Γ1ii Γl1i
+ Γ2ii Γl2i − Γ2ii Γl2i
= 0, for i, l ∈ {1, 2}.

1
R112 = (Γ112 )1 − (Γ111 )2 + Γ112 Γ111 − Γ111 Γ112
+ Γ212 Γ121 − Γ211 Γ122
= 0,

1
R121 = (Γ111 )2 − (Γ112 )1 + Γ111 Γ112 − Γ112 Γ111
+ Γ211 Γ122 − Γ212 Γ121
= 0,

1
R211 = (Γ121 )1 − (Γ121 )1 + Γ121 Γ111 − Γ121 Γ111
+ Γ221 Γ121 − Γ221 Γ121
= 0,

2
R112 = (Γ212 )1 − (Γ211 )2 + Γ112 Γ211 − Γ111 Γ212
+ Γ212 Γ221 − Γ211 Γ222
= cos2 θ − sin2 θ − cos2 θ = − sin2 θ,

2
R121 = (Γ211 )2 − (Γ212 )1 + Γ111 Γ212 − Γ112 Γ211
+ Γ211 Γ222 − Γ212 Γ221
d cos θ
= (− sin θ cos θ) − (− sin θ cos θ)
dθ sin θ
= − cos2 θ + sin2 θ + cos2 θ = sin2 θ
 59

2
R211 = (Γ221 )1 − (Γ221 )1 + Γ121 Γ211 − Γ121 Γ211
+ Γ221 Γ221 − Γ221 Γ221
= 0,

1
R122 = (Γ112 )2 − (Γ112 )2 + Γ112 Γ112 − Γ112 Γ112
+ Γ212 Γ122 − Γ212 Γ122
= 0,

1
R212 = (Γ122 )1 − (Γ121 )2 + Γ122 Γ111 − Γ121 Γ112
+ Γ222 Γ121 − Γ221 Γ122
d cos θ cos2 θ
=− ( )−
dθ sin θ sin2 θ
− sin2 θ − cos2 θ cos2 θ
=− − = 1,
sin2 θ sin2 θ
1
R221 = (Γ121 )2 − (Γ122 )1 + Γ121 Γ112 − Γ122 Γ111
+ Γ221 Γ122 − Γ222 Γ121
d cos θ cos2 θ
= ( )+
dθ sin θ sin2 θ
2
− sin θ − cos2 θ cos2 θ
= + = −1,
sin2 θ sin2 θ

2
R122 = (Γ212 )2 − (Γ212 )2 + Γ112 Γ212 − Γ112 Γ212
+ Γ212 Γ222 − Γ212 Γ222
= 0,

2
R212 = (Γ222 )1 − (Γ221 )2 + Γ122 Γ211 − Γ121 Γ212
+ Γ222 Γ221 − Γ221 Γ222
= 0,

2
R221 = (Γ221 )2 − (Γ222 )1 + Γ121 Γ212 − Γ122 Γ211
+ Γ221 Γ222 − Γ222 Γ221
= 0.
60

Exercise 52. Show that if Z is a tangent vectorfield of A and f : A →

R is a function, then

∇W +Z V = ∇W V + ∇Z V, and ∇f W V = f ∇W V.
Further if Z : A → Rn is any vectorfield, then

∇W (V + Z) = ∇W V + ∇Z Z, and ∇W (f V ) = (W f )V + f ∇W V.
Solution. Assume that the curve γ : (−, ) → A ⊂ Rn with γ(0) = p,
γ 0 (0) = W .Then

∇W V = (V ◦ γ)0 (0)
= ((V 1 ◦ γ 1 )(0), (V 2 ◦ γ 2 )(0), . . . , (V n ◦ γ n )(0)),
where γ = (γ 1 , γ 2 , . . . , γ n ) and γ k are the component functions of γ and
V = (V 1 , V 2 , . . . , V n ) with V i are the component functions of V . Moreover,

((V 1 ◦ γ 1 )0 (0), (V 2 ◦ γ 2 )0 (0), . . . , (V n ◦ γ n )0 (0)) = (W V 1 , W V 2 , . . . W V n ).

Thus
∇W V = (W V 1 , W V 2 , . . . , W V n ). (1)
Using (1) to obtain

∇W +Z V
= ((W + Z)V 1 , (W + Z)V 2 , . . . , (W + Z)V n )
(W + Z)f = hW + Z, gradf i
= (W V 1 + ZV 1 , W V 2 + ZV 2 , . . . .W V n + ZV n ) = hW, gradf i + hZ, gradf i
= W f + Zf for all f : A → R
= (W V 1 , W V 2 , . . . , W V n ) + (ZV 1 , ZV 2 , . . . , ZV n )
= ∇W V + ∇Z V,

as required. Again, applying (1) to obtain

∇f W V = (f W V 1 , f W V 2 , . . . f W V n )
= f (W V 1 , W V 2 , . . . , W V n )
= f ∇W V,
as required. Further, we have that
 61

∇W (V + Z) = [(V + Z) ◦ γ]0 (0)

= (V ◦ γ)0 (0) + (Z ◦ γ)0 (0)
= ∇W V + ∇W Z,
as required. Moreover, we have that

∇W (f V ) = (W f V 1 , W f V 2 , . . . , W f V n )
= ((f V 1 ◦ γ)0 (0), (f V 2 ◦ γ)0 (0), . . . , (f V n ◦ γ)0 (0)), (2)
and

(W f )V = (W f V 1 , W f V 2 , . . . , W f V n ),
and

f ∇W W = (f W V 1 , f W V 2 , . . . , f W V n ).
Therefore,

(W f )V + f ∇W V = (W f V 1 + f W V 1 , W f V 2 + f W V 2 , . . . , W f V n + f W V n ). (3)
But by Leibnitz rule,

(f V i ◦ γ)0 (0) = W f V i + f W V i for i = 1, 2, . . . , n. (4)

It follows from (2), (3) and (4) that

∇W (f V ) = (W f )V + f ∇W V,
as required.
Exercise 53. Note that if V and W are a pair of vectorfields on A
then hV, W i : A → R defined by hV, W ip := hVp , Wp i is a function on A,
and show that

Z hV, W i = ∇Z V, W + V, ∇Z W .
62

Solution. Assume that the curve γ : (−, ) → A ⊂ Rn with γ(0) = p,

γ 0 (0)= Z.We have that

Z hV, W i
= (hV, W i ◦ γ)0 (0)
= (hV ◦ γ, W ◦ γi)0 (0)
= (V ◦ γ)0 (0), W ◦ γ(0) + (V ◦ γ)(0), (W ◦ γ)0 (0) since (hu, vi)0 = u0 , v + u, v 0
= ∇Z V, W + V, ∇Z W ,

as required.

Exercise 54. Show that R ≡ 0.

Solution. Pn For every Pnf : A → R, iwe have that W f = (f ◦

function
γ)0 (0) i 0
= i=1 Di f (p)(γ ) (0) = i=1 Di f (p)W = hW, gradf i where γ =
(γ 1 , γ 2 , . . . , γ n ), γ(0) = p, γ 0 (0) = W and W = (W 1 , W 2 , . . . , W n ).
Next, we define Hessf (V, W ) = V, ∇W gradf , and we shall show that
Hessf (V, W ) = Hessf (W, V ). Indeed,

Hessf (V, W ) = V, ∇W gradf

= hV, (W (D1 f ), W (D2 f ), . . . , W (Dn f ))i
= hV, (hW, gradD1 f i , hW, gradD2 f i , . . . , hW, gradDn f i)i
n
X
= V i hW, gradDi f i
i=1
Xn Xn
= V i( W j Dji f )
i=1 j=1
X n
= V i W j Dij f
i,j=1
= hW, (hV, gradD1 f i , hV, gradD2 f i , . . . , hV, gradDn f i)i
= W, ∇V gradf
= Hessf (W, V ).

Using all facts above to obtain that

 63

V (W f ) − W (V f )
= V hW, gradf i − W hV, gradf i
= ∇V W, gradf + W, ∇V gradf by Exercise 4 of Lecture Notes 13
− ∇W V, gradf − V, ∇W gradf solved above
= ∇V W − ∇W V, gradf
+ Hessf (W, V ) − Hessf (V, W )
by the fact that
= h[V, W ], gradf i
Hessf (V, W ) = Hessf (W, V ) proved above
= [W, W ]f.

Thus
V (W f ) − W (V f ) = [W, W ]f.
Let Z = (Z 1 .Z 2 , . . . , Z n ). Then

V (W Z i ) − W (V Z i ) = [V, W ]Z i
for i = 1, 2, . . . n. Thus
(V (W Z 1 ), V (W Z 2 ), . . . , V (W Z n )) − (W (V Z 1 ), W (V Z 2 ), . . . , W (V Z n ))
=([V, W ]Z 1 , [V, W ]Z 2 , . . . , [V, W ]Z n ).

Using the fact that ∇W Z = (W Z 1 , W Z 2 , . . . , W Z n ) to obtain

∇V ∇W Z − ∇W ∇V Z = ∇[V,W ] Z.

Therefore,

R(V, W )Z = ∇V ∇W Z − ∇W ∇V Z − ∇[V,W ] Z
= 0,

and so R ≡ 0 as required.

Exercise 55. Show that, with respect to tangent vectorfields on M , ∇

satisfies all the properties which were listed for ∇ in Exercises 52 and 53.

Solution. Let n be a unit normal vector to Tp M at p.

Firstly, we have that
64

∇W +Z V
= (∇W +Z V )T
= (∇W V + ∇Z V ) − ∇W V + ∇Z V, n n since ∇W +Z V = ∇W V + ∇Z V
= [∇W V − ∇W V, n n] + [∇Z V − ∇Z V, n n]
= [∇W V − (∇W )⊥ ] + [∇Z V − (∇Z )⊥ ]
= (∇W V )T + (∇Z V )T
= ∇W V + ∇Z V.

Thus ∇W +Z V = ∇W V + ∇Z V .
Secondly,

∇f M V = (∇f M V )T
= ∇f M V − ∇f M V, n n
= f ∇M V − f ∇M V, n n since ∇f M V = f ∇M V
= f [∇M V − ∇M V, n n]
= f [∇M V − (∇M V )⊥ ]
= f (∇M V )T
= f ∇M V.

Thus ∇f M V = f ∇M V .
Thirdly,

∇W (V + Z)
= ∇W (V + Z) − ∇W (V + Z), n n
= ∇W V + ∇W Z − ∇W V, n n − ∇W Z, n n since ∇W (V + Z) = ∇W V + ∇W Z
= (∇W V − ∇W V, n n) + (∇W Z − ∇W Z, n n)
= [∇W V − (∇W V )⊥ ] + [∇W Z − (∇W Z)⊥ ]
= (∇W V )T + (∇W Z)T
= ∇W V + ∇W Z.

Thus ∇W (V + Z) = ∇W V + ∇W Z.
Finally,
 65

D E D E
h∇Z V, W i + hV, ∇Z W i = ∇Z V − (∇Z V )⊥ , W + V, ∇Z W − (∇Z W )⊥
D E D E
= ∇Z V, W + V, ∇Z W − (∇Z V )⊥ , W − V, (∇Z W )⊥
= Z hV, W i

since (∇Z V )⊥ , W = V, (∇Z W )⊥ = 0 as V and W belong to the tangent

vector field on M . Thus h∇Z V, W i + hV, ∇Z W i = Z hV, W i.
Exercise 56. Verify the following sentence:

∇W V = ∇W V + hV, S(W )i n.
Solution. Since V ∈ Tp M and n is a local Gauss map, it follows that
hV, n(p)i = 0. Thus W hV, n(p)i = 0. But W hV, n(p)i = ∇W V, n(p) +
V, ∇W n(p) by Exercise 4 of Lecture Notes 13 solved above. Therefore,
∇W V, n = − V, ∇W n . This implies that

∇W V, n = − V, ∇W n
= − hV, dn(W )i
= hV, −dn(W )i
= hV, S(W )i .
Thus

(∇W V )⊥ = ∇W V, n(p) n(p)

= hV, S(Wp )i n(p) since ∇W V, n n = hV, S(W )i
which in turn yields

∇W V = (∇W V )T + (∇W V )⊥
= ∇W V + hV, S(W )i n.

Therefore ∇W V = ∇W V + hV, S(W )i n as required.

Exercise 57. Verify the following sentence:

∇V S(W ) − ∇W S(V ) = S([V, W ]).

66

Solution (1). We shall show that

∇V S(W ) − ∇W S(V ) = S([V, W ]).

Since R ≡ 0, so the normal component of R(W, W )Z equals to 0. This
implies that

h∇V S(W ), Zi − h∇W S(V ), Z)i = hS([V, W ]), Zi = 0.

This is equivalent to

h∇V S(W ) − ∇W S(V ) − S([V, W ]), Zi = 0.

In particular, Z = ∇V S(W ) − ∇W S(V ) − S([V, W ]), then

h∇V S(W ) − ∇W S(V ) − S([V, W ]), ∇V S(W ) − ∇W S(V ) − S([V, W ])i = 0.

This implies that

k∇V S(W ) − ∇W S(V ) − S([V, W ])k = 0.

Therefore,

∇V S(W ) − ∇W S(V ) − S([V, W ]) = 0.

Thus

∇V S(W ) − ∇W S(V ) = S([V, W ]).

Solution (2). We shall verify that in local coordinates, the equations

R(V, W )Z = hS(W ), Zi S(V ) − hS(V ), Zi S(W )

and

∇V S(W ) − ∇W S(V ) = S([V, W ]).

take on the forms which we had derived earlier.
• The Gauss equations:
In local coordinates, we calculate
 67

R(Xi , Xj )Xk = hS(Xj ), Xk i S(Xi ) − hS(Xi ), Xk i S(Xj )

= −lkj Ni + lki Nj , (1)

since lkj = hS(Xj ), Xk i and S(Xi ) = Ni

∇Xj Xk = (∇Xj Xk )T
= (Xj Xk )T
= (Xkj )T
2
X
= Γikj Xi ,
i=1

∇Xj Xk = Xkj ,

∇Xi Xj , Xk = hXij , Xk i
2
X
= Γlij hXl , Xk i
l=1
2
X
= Γlij gkl
l=1

since hXl , Xk i = gkl = glk . Using the calculations above we have

R(Xi , Xj )Xk = ∇Xi ∇Xj Xk − ∇Xj ∇Xi Xk − ∇[Xi .Xj ] Xk

2 2
X X since [Xi , Xj ] = 0,
= ∇ Xi ( Γlkj Xl ) − ∇Xj ( Γlki Xl )
so ∇[Xi .Xj ] Xk = 0
l=1 l=1
2
X 2
X
= [(Xi Γlkj )Xl + Γlkj ∇Xi X l ] − [(Xj Γlki )Xl + Γlki ∇Xj Xl ]
l=1 l=1
2
X 2
X
= [(Γlkj )i − (Γlki )j + (Γrkj Γlir − Γrik Γljr )]Xl . (2)
l=1 r=1

Since Ni = −li1 X1 − li2 X2 , Nj = −lj1 X1 − lj2 X2 , so

68

−lkj Ni + lki Nj = −lkj (−li1 X1 − li2 X2 ) + lki (−lj2 X1 − lj2 X2 ).

= (lkj li1 − lki lj1 )X1 + (lkj li2 − lki lj2 )X2 .
Combine with (1) we have

R(Xi , Xj )Xk = (lkj li1 − lki lj1 )X1 + (lkj li2 − lki lj2 )X2 . (3)
Since X1 and X2 are linearly independent, so comparing coefficients of X1
and X2 in (2) and (3) to obtain the classical Gauss equation:

2
X
(Γrkj )i − (Γrki )j + (Γljk Γril − Γlji Γrkl ) = lkj lir − lki ljr .
l=1
• The Codazzi-Mainardi equations:
In local coordinates, we have that

X hS(Y ), Zi − Y hS(X), Zi = h∇X S(Y ), Zi + hS(Y ), ∇X Zi − h∇Y S(X), Zi − hS(X), ∇Y Zi

= hS(Y ), ∇X Zi − hS(X), ∇Y Zi + h∇X S(Y ) − ∇Y S(X), Zi
= hS(Y ), ∇X Zi − hS(X), ∇Y Zi + hS[X, Y ], Zi

since ∇X S(Y ) − ∇Y S(X) = S[X, Y ]. Thus

X hS(Y ), Zi − Y hS(X), Zi = hS(Y ), ∇X Zi − hS(X), ∇Y Zi + hS[X, Y ], Zi .

Let X = Xk , Y = Xj , Z = Xi . Then the above equation becomes

D E
Xk hS(Xj ), Xi i − Xj hS(Xk ), Xi i = S(Xj ), ∇Xk Xi − S(Xk ), ∇Xj Xi + hS[Xk , Xj ], Xi i .

Hence
* 2
+ * 2
+
X X
(lij )k − (lik )j = S(Xj ), Γlki Xl − S(Xk ), Γlji Xl
l=1 l=1
2
X
= (Γlki ljl − Γlji lkl ). (Codazzi-Mainardi)
l=1

Therefore the classical Codazzi-Mainardi equations have been proved.

Exercise 58. Show that if V and W are general vectorfields (not nec-
essarily orthonormal), then
 69

R(V, W, W, V )
K= .
kV × W k
Solution. Firstly, we shall prove that if V and W are orthonormal
in the tangent vectofield on M , then hR(V, W )W.V i = K, where K is the
Gaussian curvature. Indeed, since V and W belong to Tp (M ), so taking
{V, W } be a orthonormal basis of Tp M . Since S : Tp (M ) → TM (M ), so we
can assume S(V ) = aV + bW and S(W ) = cV + dW . We have that

hS(V ), V i hS(W ), W i − hS(W ), V i hS(V ), W i

= haV + bW, V i hcV + dW, W i − hcV + dW, V i haV + bW, W i
since hV, V i = 1, hW, W i = 1,
= ad − bc
hV, W i = hW, V i = 0
= det(S)
= K.

But note that by Gauss’s equation

hR(V, W )W, V i = hS(V ), V i hS(W ), W i − hS(W ), V i hS(V ), W i .

Therefore

hR(V, W )W, V i = K.
Now, we consider the case that V and W are general vectorfields (not
necessarily orthonormal).
Let

V V
V0 = and W0 = x + yW where x, y ∈ R.
kV k kV k
We find x and y such that V 0 and W 0 are orthonormal. This is equivalent
to

W 0 , V 0 = 0, (1)
W 0 , W 0 = 1. (2)
(Note that hV 0 , V 0 i = 1). We have that
70

 
0 0 V V hW, V i
W ,V = x + yW, =x+y ,
kV k kV k kV k
 
V V hW, V i
W 0, W 0 = x + yW, x + yW = x2 + 2xy + y 2 kW k2 .
kV k kV k kV k

Thus (1) and (2), respectively, become

hW, V i
x+y = 0, (3)
kV k
hW, V i
x2 + 2xy + y 2 kW k2 = 1. (4)
kV k

It follows from (3) that x = −y hW,V i

kV k and substitute it into (4) to obtain

hW, V i2 2 hW, V i
2
y2 − 2y + y 2 kW k2 = 1
kV k2 kV k2
or

kV k2 kW k2 − hW, V i2
y2( ) = 1.
kV k2

Thus

kV × W k2
y2 =1 by Lagrange’s identity.
kV k2
kV k
Therefore y = kV ×W k and x = − kVhW,V i
×W k . Now vectors

V hW, V i V kV k
V0 = and W0 = · + ·W
kV k kV × W k kV k kV × W k

are orthonormal. Apply the above result for V 0 and W 0 to obtain

 71

K = R(V 0 , W 0 )W 0 , V 0
 
V V V V
= R( ,x + yW )(x + yW ),
kV k kV k kV k kV k
 
1 V V
= 2
R(V, x + yW )(x + yW ), V since R is linear in every variable
kV k kV k kV k
y2 since hR(V, W )V, V i = 0,
= hR(V, W )W, V i
kV k2 hR(V, V )V, V i = 0, hR(V, V )W, V i = 0
kV k2
= hR(V, W )W, V i
kV × W k2 kV k2
R(V, W, W, V )
= ,
kV × W k2

as required.
Exercise 59. Show that the absolute geodesic curvature of great circles
in a sphere and helices on a cylinder are everywhere zero.
Solution. Firstly, by choosing a suitable coordinates system, we can
assume that the great circle in a sphere is

α
[0, 2π] 3 t 7−→ α(t) = (r cos t, r sin t, 0).
We have that
α0 (t) = (−R sin t, R cos t, 0),
α00 (t) = (−R cos t, −R sin t, 0).
1
Since n(p) = Rp is a Gauss map in a neighborhood of α(t), it follows that

(α00 (t))T = α00 (t) − α00 (t), (cos t, sin t, 0) (cos t, sin t, 0)
= (−R cos t, −R sin t, 0) + (R cos t, R sin t, 0)
= (0, 0, 0).
Thus

|κg | = (α00 (t))T = 0.

Therefore, the absolute geodesic curvature of great circles in a sphere is
everywhere zero.
The equation of the helix is
72

α
R 3 t 7−→ α(t) = (a cos t, a sin t, bt).
We have that

α0 (t) = (−a sin t, a cos t, 0),

α00 (t) = (−a cos t, −a sin t, 0).
The equation of the cyclinder surface is

X(u, v) = (a cos u, a sin u, v).

We have that

X1 = (−a sin u, a cos u, 0),

X2 = (0, 0, 1),
X1 × X2 1
N (u, v) = = (a cos u, a sin u, 0)
kX1 × X2 k a
= (cos u, sin u, 0),
p1 p2
n(p) = N ◦ X −1 (p) = ( , , 0), p = (p1 , p2 , p3 ).
a a
This implies that
where np (t) = n(p(t)),
np (t) = (cos t, sin t, 0)
p(t) = (a cos t, a sin t, bt).
Thus

(α00 (t))T = α00 (t) − α00 (t), np (t) np (t)

= (−a cos t, −a sin t, 0) + a(cos t, sin t, 0)
= (0, 0, 0).
Hence
|κg | = (α00 (t))T = 0.
Therefore, the absolute geodesic curvature of helices on a cylinder are ev-
erywhere zero.
 73

Exercise 60. Let S2 be oriented by its outward unit normal, i.e.,

n(p) = p, and compute the geodesic curvature of the circles in S2 which lie
in planes z = h, −1 < h < 1. Assume that all these circles are oriented
consistenly with respect to the rotation about the z-axis.

Solution (1). The equation for the circle in S2 which lie in planes
z = h, −1 < h < 1 is

p p
α(t) = ( 1 − h2 cos t, 1 − h2 sin t, h).

√
Let r = 1 − h2 to obtain

α(t) = (r cos t, r sin t, h).

Thus

α0 (t) = (−r sin t, r cos t, 0),

α00 (t) = (−r cos t, −r sin t, 0),
n(p) = p.

This implies that

n(p(t)) = (r cos t, r sin t, h).

Hence

Jα0 = n × α0 = (−hr cos t, −hr sin t, r2 ).

Therefore
74

hα00 , Jα0 i
κg =
kα0 k3
hr2 cos2 t + hr2 sin2 t
=
r3
hr (cos t + sin2 t)
2 2
=
r3
hr 2
= since cos2 t + sin2 t = 1
r3
h
=
r
h
= √ .
1 − h2
Solution (2). The equation parametrized by arclength for the circle
in S2 which lie in planes z = h, −1 < h < 1 is

p t p t
α(t) = ( 1 − h2 cos √ , 1 − h2 sin √ , h).
1−h2 1 − h2
√
Let r = 1 − h2 to obtain

t t
α(t) = (r cos , r sin , h).
r r
Thus

t t
α0 (t) = (− sin , cos , 0),
r r
1 t 1 t
α00 (t) = (− cos , − sin , 0),
r r r r
n(p) = p.
This implies that

t t
n(p(t)) = (r cos , r sin , h).
r r
Hence
 75

t t
Jα0 = n × α0 = (−h cos , −h sin , r).
r r
Therefore

κg = α00 , Jα0
h t h t
= cos2 + sin2
r r r r
h t t
= (cos2 + sin2 )
r r r
h t t
= since cos2 + sin2 = 1
r r r
h
=√ .
1 − h2
Exercise 61. Show that if α is a geodesic, then it must have constant
speed.
e0 : α(I) → R3 given by α
Solution. Define α e0 (α(t)) = α0 (t). We have
that

e0 = (α0 (t)α10 (t), α0 (t)α20 (t), α0 (t)α30 (t))

∇α0 (t) α
= (α100 (t), α200 (t), α300 (t))
= α00 (t),
where α(t) = (α1 (t), α2 (t), α3 (t)). Thus

e0 = α00 (t).
∇α0 (t) α
Hence

α00 , Jα0 = (α00 )T , Jα0

e0 , Jα0 .
= ∇ α0 α
This implies that

e0 , Jαi
h∇α0 α
κg = .
kα0 k3
76

Since α is a geodesic, then κg = 0, so h∇α0 αe0 , Jα0 i = 0. Since Jα0 ∈ Tp M

so from h∇α0 α 0 0
e , Jα i = 0, we deduce that ∇α0 α e0 belongs to the orthogonal
complement of Tp M , but by the definition of ∇α0 α e0 , we have that ∇α0 α
e0 ∈
Tp (M ). Therefore ∇α0 α 0 ⊥
e ∈ Tp (M ) ∩ (Tp M ) = {0}. This implies that
0
e = 0. Thus ∇α0 α
∇ α0 α 0 e0 )⊥ . Therefore
e = (∇α0 α

0
α0 , α0 = α00 , α0 + α0 , α00
= 2 α0 , α00
e 0 , α0
= 2 ∇ α0 α
D E
= 2 (∇α0 αe 0 )⊥ , α 0
= 0,

giving that kα0 k = const.

Exercise 62. Show that if M = R2 , and n = (0, 0, 1), then J is

clockwise rotation about the origin by π/2.

Solution. Since M = R2 , we have Tp M = M = R2 . For n = (0, 0, 1),

any V ∈ Tp M is of the form V = (V 1 , V 2 , 0). We have that

JV = n × V
0 1 1 0 0 0
=( , , )
V2 0 0 V1 V1 V2
= (−V 2 , V 1 , 0)

and

hV, JV i = −V 1 V 2 + V 2 V 1 = 0.

This implies that V ⊥ JV . Let

W = V + JV
= (V 1 − V 2 , V 1 + V 2 , 0).

We have that
 77

hW, V i = V 1 (V 1 − V 2 ) + V 2 (V 1 + V 2 )
= (V 1 )2 + (V 2 )2 ≥ 0.
Thus

π
(V, JV ) = − .
2
It follows from the facts that V ⊥ JV and (V, JV ) = − π2 proved above
that J is the clockwise rotation about the origin by π = π/2.
Exercise 63. Verify the following two equations:

1 − α00 (s−1 ), α0 (s−1 )

α00 =α00 (s−1 ) · + α0 (s−1 ) · ,
kα0 (s−1 )k2 kα0 (s−1 )k4

hα00 , Ja0 i
κg = .
kα0 k3
Solution. Firstly, we shall verify that

1 − α00 (s−1 ), α0 (s−1 )

α00 =α00 (s−1 ) · + α0 (s−1 ) · .
kα0 (s−1 )k2 kα0 (s−1 )k4
Since (s−1 )0 = 1
kα0 k , be the chain rule

α0 (t) = α0 (s−1 (t))[s−1 (t)]0

1
= α0 (s−1 (t)) 0 −1 .
kα (s (t))k
Differentiate both sides of the above equation to obtain

1 1
α00 = α00 (s−1 )(s−1 )0 · +[ ]0 · α0 (s−1 ).
kα0 (s−1 )k kα0 (s−1 )k
1 1
= α00 (s−1 ) · + α0 (s−1 )[ ]0 . (1)
kα0 (s−1 )k2 kα0 (s−1 )k
Since
78

2
α0 (s−1 ) = α0 (s−1 ), α0 (s−1 ) ,
so

2
( α0 (s−1 ) )0 = 2 α00 (s−1 ), α0 (s−1 ) (s−1 )0 .
Thus

0 2 α00 (s−1 ), α0 (s−1 )

2 α0 (s−1 ) α0 (s−1 ) = .
kα0 (s−1 )k
Hence

0 α00 (s−1 ), α0 (s−1 )

α0 (s−1 ) = ,
kα0 (s−1 )k2
and so

1 −1 0
( )0 = ( α0 (s−1 ) )
kα0 (s−1 )k
−2 0
= − α0 (s−1 ) α0 (s−1 )
α00 (s−1 ), α0 (s−1 )
=− . (2)
kα0 (s−1 )k4
Substitute (2) into (1) to obtain

1 − α00 (s−1 ), α0 (s−1 )

α00 =α00 (s−1 ) · + α0 (s−1 ) · ,
kα0 (s−1 )k2 kα0 (s−1 )k4
as required.
Secondly, we shall verify that

hα00 , Ja0 i
κg = .
kα0 k3
Indeed,
 79

κg
= κg
= α00 , Ja0
* +
1 − α00 (s−1 ), α0 (s−1 )
= α00 (s−1 ) · + α0 (s−1 ) · , Ja0
kα0 (s−1 )k2 kα0 (s−1 )k4
* +
1 1 since α0 (s−1 ), Ja0 (s−1 )
= α00 (s−1 ) · , Ja0 1
kα0 (s−1 )k2 kα0 (s−1 )k = kα0 k
hα0 , Jα0 i = 0

hα00 , Ja0 i
= , since J(tV ) = tJV, t ∈ R,
kα0 k3

as required.
Exercise 64. Show that if α is parametrized by arclength, then

e0 .
|κg | = ∇α0 α
Solution. It is clear that

e0 = α00 (t)
∇α0 (t) α
e0 = [e
since ∇α0 (t) α α0 ◦ α(t)]0 = [α0 (t)]0 = α00 (t). Since α is parametrized by
arclength, it follows that

|κg | = (α00 )T
e0 )T
= (∇α0 (t) α
e0
= ∇α0 (t) α
e0 ,
= ∇ α0 α
as required.
e0 ≡ 0.
Exercise 65. Show that is α a geodesic if and only if ∇α0 α
e0 ≡ 0. Then
Solution. Suppose first that ∇α0 α

e0 , Jα0 i
h∇α0 α
κg = = 0.
kα0 k3
Thus α is a geodesic.
80

Conversely, if α a geodesic, i.e., κg = 0 , then α = α ◦ s−1 is a

parametrization by arclength, so by Exercise 6 of Lecture Notes 15 proved
above, we have that

e0 = 0
∇α0 α
⇐⇒ e0 = 0
∇(α◦s−1 )0 α
0
⇐⇒ ∇α0 (s−1 )(s−1 )0 α
e =0
0
⇐⇒ (s−1 )0 ∇α0 α
e =0 since ∇f U V = f ∇U V
1
=⇒ ∇α0 [α ◦ (s−1 )]0 = 0 since (s−1 )0 = 6= 0
kα0 k
=⇒ ∇α0 α0 (s−1 )(s−1 )0 = 0
=⇒ α0 (s−1 )0 α0 + (s−1 )0 ∇α0 α
e0 = 0 use ∇U f V = U f V + f ∇U V
since (s−1 )0 = const because α is a geodesic,
=⇒ e0 = 0
∇α0 α
and so has constant speed, and thus α0 (s−1 )0 = 0.
e0 ≡ 0.
Therefore, α a geodesic if and only if ∇α0 α
Exercise 66. Write down the equations of the geodesic in a surface
of revolution. In particular, verify that the great circles in a sphere are
geodesics.
Solution. We first compute the Christoffel symbols of a surface of
revolution. Consider a surface of revolution parametrized by

x(u, v) = (ϕ(v) cos u, ϕ(v) sin u, ψ(v))

0 < u < 2π, a < v < b, ψ(v) 6= 0.
We have that

X1 = (−ϕ(v) sin u, ϕ(v) cos u, 0)

X2 = (ϕ0 (v) cos u, ϕ0 (v) sin u, ψ 0 (v))

X11 = (−ϕ(v) cos u, −ϕ(v) sin u, 0)

X12 = (−ϕ0 (v) sin u, ϕ0 (v) cos u, 0)

X21 = (−ϕ0 (v) sin u, ϕ0 (v) cos u, 0)

 81

X22 = (ϕ00 (v) cos u, ϕ00 (v) sin u, ψ 00 (v))

g11 = hX1 , X1 i = ϕ2 (v) sin2 u + ϕ2 (v) cos2 u = ϕ2 (v)

g12 = hX1 , X2 i = 0 = g21

g22 = [ϕ0 (v)]2 cos2 u + [ϕ0 (v)]2 sin2 u + [ψ 0 (v)]2

= [ϕ0 (v)]2 + [ψ 0 (v)]2 = 1
since assume that the rotating curve is parametrized by arc-length.Thus

 2 
ϕ 0
(gij ) = ,
0 1
 
1 1 0
(g ij ) = (gij )−1 = .
ϕ2 0 ϕ2

• Γ111 , Γ211 : We have that

hX11 , X1 i = 0,
hX11 , X2 i = −ϕ(v)ϕ0 (v) = −ϕϕ0 .
 1      
Γ11 1 0 0 0
Thus = ϕ12 = . Hence Γ111 = 0, Γ211 =
Γ211 0 ϕ2 −ϕϕ0 −ϕϕ0
−ϕϕ0 .
• Γ112 , Γ212 , Γ121 , Γ221 : We have that

hX12 , X1 i = ϕϕ0 ,
hX12 , X2 i = 0.
!
ϕ0
 1     0
Γ12 1 1 0 ϕϕ ϕ0
Thus = = ϕ . Hence Γ112 = = Γ121 ,
Γ212 ϕ2 0 ϕ2 0 0 ϕ

Γ212 = 0 = Γ221 (Note that Γkij = Γkji ).

• Γ122 , Γ222 : We have that
82

hX22 , X1 i = 0,
hX22 , X2 i = ϕ0 ϕ00 + ψ 0 ψ 00 .
 1      
Γ22 1 1 0 0 0
Thus = ϕ2 = . Hence
Γ222 0 ϕ2 ϕ0 ϕ00 + ψ 0 ψ 00 ϕ0 ϕ00 + ψ 0 ψ 00
Γ122 = 0, Γ222 = ϕ0 ϕ00 + ψ 0 ψ 00 .
The two equations

u001 + Γ111 (u01 )2 + 2Γ112 u01 u02 + Γ122 (u02 )2 = 0,

u002 + Γ211 (u01 )2 + 2Γ212 u01 u02 + Γ222 (u02 )2 = 0,
become

ϕ0 0 0
u001 + 2 u u = 0, (1)
ϕ 1 2
u002 − ϕϕ0 (u01 )2 + (ϕ0 ϕ00 + ψ 0 ψ 00 )(u02 )2 = 0. (2)
The equation of the sphere centered at 0 of radius r is

x(u, v) = (r sin v cos u, r sin v sin u, r cos v).

In this case, we have that ϕ(t) = r sin t and ψ(t) = r cos t, and so the two
equations (1) and (2) become

cos t 0 0
u001 + 2 u u = 0, (3)
sin t 1 2
u002 − r2 sin t cos t(u01 )2 = 0. (4)
Without loss of generality, we assume that the great circle α(t) lies in the
xz-plane and so its equation is

α(t) = (r sin t, 0, r cos t).

The function u(t) satifies x(u(t)) = α(t) be u(t) = (0, t), i.e, u1 (t) = 0 and
u2 (t) = t. The functions u1 (t) and u2 (t) satisfy the two equations (3) and
(4). Thus the great circles in a sphere are geodesics.
Exercise 67. Show that the sum of the angles in a triangle is π.
 83

Solution. Every
P line in a plane is a geodesic curve and has Gaussian
curvature κ = 0, 3i=1 θi = π. By Gauss-Bonnet Theorem,

2 Z
X si+1 ZZ 2
X
κg (S) ds + κ dσ + θi = 2π.
i=0 si R i=0
Since κ = 0 , so the equation above becomes

2
X
θi = 2π.
i=0
But

2
X 2
X
θi = 3π − αi ,
i=0 i=0
where αi , i = 0, 1, 2, is the internal angles of the triangle. Thus

2
X
3π − αi = 2π,
i=0
and so

2
X
αi = π.
i=0
Therefore the sum of the angles in a triangle is π.
Exercise 68. Show that the total geodesic curvature of a simple closed
planar curve is 2π.
Solution. Since a simple closed planar curve has κ = 0 and 3i=1 θi =
P
0, it follows from Gauss-Bonnet Theorem that

k Z
X si+1
κg (S) ds = 2π.
i=0 si

Thus by Corollary 12 of Lecture Notes 15, the total geodesic curvature of

a simple closed planar curve is 2π.
84

Exercise 69. Show that the Gaussian curvature of a surface which is

homeomorphic to the torus must alwasy be equal to zero at some point.
Solution. Let S be the surface of the torus and M a surface which is
homeomorphic to S. Let f be the homeomorphism from M onto S. The
Euler-Poincare characteristic of the torus S is χ(S) = 0 and since M is
homeomorphic to S, so χ(M ) = χ(S) = 0. Since S is a compact surface
and M is homeomorphic to S, so M is also a compact surface (since the
image of a compact set under a continuous map is also a compact set).
Thus, by Corollary 2 of the Gauss-Bonnet Theorem (see p. 280 of [Car92]),
(or since the surface M is homeomorphic to the torus, so its boundary is
empty, and so the following expression is followed from the Gauss-Bonnet
theorem)
ZZ
κ dσ = 2πχ(M ) = 0,
M
where κ is the Gaussian curvature over M . Therefore, there exists some
point p of M such that κ(p) = 0 because κ is a continuous function on M ,
if there are p1 , p2 ∈ M such that κ(p1 ) > 0 and κ(p2 ) < 0, then by the
intermediate value theorem, there is q ∈ M satisfying κ(q) = 0 and it is
clear that if either κ > 0 for all pRR∈ M or κ < 0 for all p ∈ M occur, then
this contradicts to the fact that M κ dσ = 0.
Exercise 70. Show that a simple closed curve with total geodesic cur-
vature zero on a sphere bisects the area of the sphere.
Solution. The parametrization of the sphere of radius r is

x(θ, ϕ) = (r sin θ cos ϕ, r sin θ sin ϕ, r cos θ).

We have that

xθ = (r cos θ cos ϕ, r cos θ sin ϕ, −r sin θ),

xϕ = (−r sin θ sin ϕ, r sin θ cos ϕ, 0),

xθθ = (−r sin θ cos ϕ, −r sin θ sin ϕ, −r cos θ),

xθϕ = (r cos θ sin ϕ, r cos θ cos ϕ, 0),

 85

xϕϕ = (−r sin θ cos ϕ, −r sin θ sin ϕ, 0),

xθ × xϕ = (r2 sin2 θ cos ϕ, r2 sin2 θ sin ϕ, r2 sin θ cos θ),

q
kxθ × xϕ k = r4 sin4 θ(cos2 ϕ + sin2 ϕ) + r4 sin2 θ cos2 θ
q
= r4 sin2 θ(sin2 θ + cos2 θ) = r2 sin θ,

xθ × xϕ
n=
kxθ × xϕ k
= (sin θ sin ϕ, sin θ sin ϕ, cos θ),

E = hxθ , xθ i = r2 ,

F = hxθ , xϕ i = 0,

G = hxϕ , xϕ i = r2 sin2 θ,

e = hn, xθθ i = −R,

f = hn, xθϕ i = 0,

g = hn, xϕϕ i = −r sin2 ϕ,

eg − f 2
κ=
EG − F 2
r2 sin2 θ
= 4 2
r sin θ
1
= 2.
r
By Gauss-Bonnet Theorem,
86

k Z
X si+1 ZZ X
κg (S) ds + κ dσ + = 2π
i=0 si R i=0
where R is the rehion bounded by the simple closed curve. Since the curve
is simple closed, it follows that ki=0 θi = 0. We also have that κ = R12 and
P
Pk R si+1
i=0 si κg (s) ds = 0 because the total geodesic curvature of the curve
is zero, and so the above equation becomes
ZZ
1
dσ = 2π.
R r2
1
RR
Thus r2
Area(R) = 2π since R dσ = Area(R) where Area(R) is the area
of the region R. Hence Area(R) = 2πr2 = 12 (Area of the sphere surface).
Exercise 71. Show that there exists at most one closed geodesic on a
cylinder with negative curvature.
Solution. If there are two geodesics γ1 and γ2 which start from p ∈ S
(the cylinder) and they meet at again at a point q ∈ S in such a way that
the traces of γ1 and γ2 constituteRRthe boundary of a simple region R of S,
then by Gauss-Bonnet Theorem R κ dσ + θ1 + θ2 = 2π where θ1 and θ2
are the external angles of the region R. Since geodesics γ2 and γ2 cannot
be mutually tangent. we have θi < π, i = 1, 2. On the other hand, κ < 0,
whence the contradiction.
Thus if S contains one closed geodesic Γ and another closed geodesic Γ0 ,
then Γ and Γ0 does not intersect Γ0 . Otherwise, the arcs of Γ and Γ0 between
two consecutive intersectionpoints r1 and r2 , would be the boundary of a
simple region, contradicting above.
Applying the Gauss-Bonnet Theorem to the region R bounded by two
simple non-intersecting geodesics Γ and Γ0 of S to obtain
ZZ
κ dσ = 2πχ(R) = 0 since χ(R) = 0
R
which is a contradictiobn, since κ < 0.
Therefore there exists at most one closed geodesic on a cylinder with
negative curvature.
Exercise 72. Show that the area of a geodesic polygon with k vertices
on a sphere of radius 1 is equal to the sum of its angles minus (k − 2)π.
 87

Solution. We have that F = k − 2, E = k, V = k. Thus χ =

F − E + V = k − 2. By Gauss-Bonnet Theorem,

k Z
X ZZ k
X
κg (s) ds + κ dσ + θi = 2(k − 2)π.
i=1 ci R i=1

But since κ = 1, ki=1 ci κg (s) ds = 0 (since the sides of a geodesic polygon

P R

are geodesics), R 1 dσ = Area(R), ki=1 θi = 3π(k − 2) − ki=1 αi (since a

RR P P
geodesic polygon is divided into k − 2 triangles), where αi are the internal
angles of a geodesic polygon so the equation above becomes

k
X
Area(R) = αi − 3(k − 2)π + 2(k − 2)π
i=1
Xk
= αi − (k − 2)π.
i=1

Exercise 73. Let p be a point of a surface M , T be a geodesic triangle

which contains p, and α, β, γ be the angles of T . Show that

α+β+γ−π
κ(p) = lim κ(ξ) = lim .
T →p T →p Area(T )
In particular, note that the above proves Gauss’s Theorema Egregium.
P3 R
Solution. Since T is a geodesic triangle, so i=1 Ci κg (S) ds = 0
and F = 1, V = E = 3. Thus χ = F − E + V = 1. Further, 3i=1 θi =
P
3π − (α + β + γ), where θi , i = 1, 2, 3, are the external angles of the triangle
T . Applying the Gauss-Bonnet Theorem to obtain
ZZ
κ dσ + 3π − (α + β + γ) = 2π.
T
Thus
ZZ
κ dσ = α + β + γ − π. (1)
T
By the mean value theorem,
88

ZZ
κ dσ = κ(ξ)Area(T )
T
where ξ is some point in T . Therefore, the equation (1) becomes

α+β+γ−π
κ(ξ) = .
Area(T )
Hence

α+β+γ−π
κ(p) = lim κ(ξ) = lim ,
T →p T →p Area(T )
as required.
Exercise 74. Show that the sum of the angles of a geodesic triangle on
a surface of positive curvature is more than π, and on a surface of negative
curvature is less than π.
P3 R
Solution. Since T is a geodesic triangle, so i=1 Ci κg (S) ds = 0
and F = 1, V = E = 3. Thus χ = F − E + V = 1. Further, 3i=1 θi =
P
3π − (α + β + γ), where θi , i = 1, 2, 3, are the external angles of the triangle
T . Applying the Gauss-Bonnet Theorem to obtain
ZZ
κ dσ + 3π − (α + β + γ) = 2π.
T
Thus
ZZ
κ dσ = α + β + γ − π. (1)
T
RR
If κ > 0, then RRT κ dσ > 0 and so we deduce from (1) that α + β + γ > π.
If κ < 0, then T κ dσ < 0 and so we deduce from (1) that α + β + γ < π.
In particular, if κ = 0 then we deduce from (1) that α + β + γ = π.
Exercise 75. Show that on a simply connected surface of negative
curvature two geodesics emanating from the same point will never meet.
Solution. Suppose, towards a contradiction, that a simply connected
surface of negative curvature two geodesics emanating from the same point
will meet. By the Gauss-Bonnet Theorem,
 89

ZZ
κ dσ + θ1 + θ2 = 2π
R
where R is the simple region bounded by two acts of two consecutive inter-
section points, and θ1 , θ2 are the external angles of R. Since the geodesics γ1
and γ2 cannot be mutually tangent, so γi < π, i = 1, 2. OnRRthe other hand,
κ < 0, whence the contradiction (note that if κ < 0 then R κ dσ < 0).

Exercise 76. Let M be a surface homeomorphic to a sphere in R3 ,

and let Γ ⊂ M be a closed geodesic. Show that each of the two regions
bounded by Γ have equal areas under the Gauss map.
Solution. We may assume that the curve α is parametrized by ar-
clength. Let s denote the arclengthD of the curve n = n(s) on S 2 . The
·· ·
E
geodesic curvature og n(s) is κg = n, n × n where the dots denote the
differentiate with respect to s. Since

· dn ds ds
n= · = (−κt − τ b)
ds ds ds
·· d2 s ds ds
n = (−κt − τ b) 2 + (−κ0 t − τ 0 b)( )2 − (κ2 + τ 2 )n( )2
ds ds ds
ds 2 1
( ) = 2 ,
ds κ + τ2
so we obtain
D ·· ·
E
κg = n, n × n
ds D ··
E
= κt − τ b, n
ds
ds
= ( )3 (−κt0 + κ0 τ )
ds
τ 0 κ − κ0 τ ds
=− 2 ·
κ + τ 2 ds
d τ ds
= − tan−1 ( ) · .
ds κ ds
Applying the Gauss-Bonnet Theorem to one of the regions bounded by n(I)
and using the fact that κ = 1 to obtain
90

ZZ Z
2π = κ dσ + κg ds = Area(R).
R ∂R

Since the area of S 2 is 4π, so each of the two regions bounded by Γ have
equal areas under the Gauss map.

Exercise 77. Compute the area of the pseudo-sphere, i.e. the surface
of revolution obtained by rotating a tractrix.

Solution (1). The equation of the tractrix α : (0, π) → R2 is given by

t
α(t) = (sin t, cos t + log(tan )).
2

The area of the surface of revolution around the y-axis is

Z π p
S= 2πx [x0 (t)]2 + [y 0 (t)]2 dt.
0

Note that

x0 (t) = cos t
1 1
2 cos2 t
0 2
y (t) = − sin t + t
tan 2
1 1 cos 2t
= − sin t + · t ·
2 cos2 2 sin 2t
1
= − sin t +
sin t

Thus
 91

1
[x0 (t)]2 + [y 0 (t)]2 = cos2 t + sin2 t + −2
sin2 t
1
= −1 since cos2 t + sin2 t = 1
sin2 t
1 − sin2 t
=
sin2 t
cos2 t
=
sin2 t
= cot2 t.
Hence

Z π
S = 2π sin t · | cot t| dt
0
Z π/2 Z π
cos t cos t
= 2π sin t 2 dt − 2π sin t dt
0 sin t π/2 sin t
Z π/2 Z π
= 2π cos t dt − 2π cos t dt
0 π/2
π/2
= 2π sin t|0 − 2π sin t|ππ/2
= 2π − 2π(0 − 1)
= 4π.

Solution (2). The parametrization equation of the surafce of revolu-

tion obtained by rotating a tractrix is

t
x(u, v) = (sin v, cos u, sin v sin u, cos v + log tan ).
2
We have that

xu = (− sin v sin u, sin v cos u, 0)

1
xv = (cos v cos u, cos v sin u, − sin v + ).
sin v
Thus
92

E = hxu , xu i
= sin2 v(sin2 u + cos2 u)
= sin2 v, since sin2 v + cos2 v = 1,

F = hxu , xv i = 0,

G = hxv , xv i
1
= cos2 v(cos2 u + sin2 u) + sin2 v + −2
sin2 v
1
= cos2 v + sin2 v + −2 since sin2 v + cos2 v = 1
sin2 v
1 cos2 v
= 2 −1= .
sin v sin2 v
Thus the area of the surface of revolution obtained by rotating by a tractrix
is
Z π Z 2π p
S= EG − F 2 du dv
v=0 u=0
Z π Z 2π
= | cos v| du dv
v=0 u=0
Z π/2 Z 2π Z π Z 2π
= cos v du dv − cos v du dv
v=0 u=0 v=π/2 u=0
π/2
= 2π sin v|0 − 2π sin v|ππ/2
= 4π.
 Part 2

Manifolds
 Exercise 78 (Product Manifolds). If M and N are manifolds of di-
mension m and n respectively, show that M × N is a manifold of dimension
m + n, with respect to its product topology. In particular, the torus T n is
an n-dimensional manifold.

Solution. Consider two distinct points (u, v) and (r, s) in M × N .

Without loss of generality, we assume that u 6= r. Since M is Hausdorff,
it follows that there exists two disjoint open subsets Uu and Ur in M con-
taining u and r, respectively. Thus Uu ∩ Ur = ∅. Let Vv and Vs be the
open neighborhoods in N of v and s, respectively. The subsets Uu × Vv
and Ur × Vs are disjoint open subsets in M × N containing (u, v) and (r, s),
respectively. Hence, M × N is Hausdorff.
Since M and N are manifolds, so they have countable bases BM and
BN , respectively. Therefore, M × N have countable bases BM and BN =
{U × V | U ∈ BM , V ∈ BN }.
Since M is a manifold, so for every x ∈ M , there is a neighborhood
Ux which is homeomorphic to Rn (The dimension of M is n) under the
mapping fx .
Similarly, for every y ∈ N , there is a neighborhood Vy in N which is
homeomorphic to Rm (the dimension of N is m) under the mapping gy .
The map hx,y : Ux × Vy → Rm+n is defined by hx,y (u, v) = (fx (u), gy (v))
for (u, v) ∈ Ux × Vy . Then Ux × Vy is homeomorphic to Rm+n under the
mapping hx,y since fx and gy are homeomorphisms and the dimension of
the manifold M × N is m + n.

Exercise 79 (Group Actions). Show that if a group G acts properly

discontinuosly on a manifold M , then M/G is a manifold. (Hints: Openness
of π ensures that M/G has a countable basis. Condition (i) in the defintion
of proper discontinuity ensures that π is locally one-to-one, which together
with openness, yields that M/G is locally homeomorphic to Rn . Finally,
condition (ii) implies that M/G is hausdorf.)

Solution. Assume that [p] 6= [q] for [p] and [q] in M/G. Since [p] 6= [q],
so p 6= hg (q) for any g ∈ G (hg is a homeomorphism on M ). Since G acts
properly discontinuous on M , so there exists open neighborhoods U and
V , respectively, such that U ∩ hg (V ) = ∅ for all g ∈ G (condition
S (ii) in the
defintion of the properly discontinuous). Therefore, U ∩ ( g∈G hg (V )) = ∅.
This implies that
 95

[
π(U ) ∩ ( hg (V )) = ∅.
g∈G

S Since π is open and hg (g ∈ G) are homeomorphisms so π(U ) and

g∈G h (V ) are open subsets in M/G. It is clear that [p] ∈ π(U ) and
Sg S
[q] ∈ g∈G hg (V ). Therefore, π(U ) and g∈G hg (V ) are open neighbor-
hoods of [p] and [q], respectively. Thus M/G is Hausdorff.
Since M is a manifold, so M has a countable basis BM . We shall prove
that π(BM ) = {π(U ) | U ∈ BM } is a countable basis of M/G. Indeed,
for any open subset V in M/G, then π −1 (V ) is open in M (since π is
continuous). Since BM is a countable basis of M , so there exists U ∈ BM
such that U ⊂ π −1 (V ). Thus π(U ) ⊂ V . Therefore, π(BM ) is a countable
basis in M/G.
For each p ∈ M , there is an open neighborhood V of p which is home-
omorphic to Rn .

π
V → M/G
↓ .
f
Rn
Let g be the homeomorphism from V to Rn . Then f : π(V ) ⊂ M/G →
Rn is defined by f (π(v)) = g(v) where v = π(v)\{hg (v) | g ∈ G\{e}}.
Since π is an open mapping, so π(V ) is open in M/G and it contains [p].
Thus π(V ) is an open neighborhood of [p] in M/G. Since G acts properly
discontinuous on M , so π is locally one-to-one (condition (i)). Thus f
defined above is a homeomorphism from π(V ) onto Rn . Thus M/G is a
manifold.
Exercise 80. Show that RPn is homeomorphic to Sn /{±1}, so it is a
manifold.
Solution. We have

RPn = {(x, −x) | x ∈ Sn },

Sn /{±1} = {(f1 (x), f−1 (x)) | x ∈ Sn , fi (i = ±1) is the homeomorphism on Sn }.

The mapping g : RPn → Sn /{±1} is defined by

96

g((x, −x)) = (f1 (x), f−1 (x))

Since f1 and f−1 are homeomorphisms on Sn , so g is a homeomorphism from
Sn onto Sn /{±1}. Since RPn is a manifold and RPn is homeomorphic to
Sn /{±1}, so Sn /{±1} is a manifold.
Exercise 81 (Hopf Fibration). Note that, if C denotes the complex
plane, then S1 = {z ∈ C | kzk = 1}. Thus, since kzwk = kzk kwk,
S1 admits a natural group structure. Further, note that S3 = {(z1 , z2 ) |
kz1 k2 + kz2 k2 = 1. Thus, for every w ∈ S1 , we may define a mapping
fw : S3 → S3 by fw (z1 , z2 ) := (wz1 , wz2 ). Show that this defines a group
action on S3 , and S3 /S1 is homeomorphic to S2 .
Solution. The unit element of the group S1 is 1. We have

f1 (z1 , z2 ) = {1z1 , 1z2 }

= {z1 , z2 } for (z1 , z2 ) ∈ S1 .
Thus f1 is the identity function on S1 . For w1 , w2 ∈ S1 and (z1 , z2 ) ∈ S3 ,
we have

fw1 ◦ fw2 (z1 , z2 ) = fw1 (w2 z1 , w2 z2 )

= ((w1 w2 )z1 , (w1 w2 )z2 )
= fw1 w2 ((z1 , z2 )).
Hence fw1 ◦ fw2 ((z1 , z2 )) = fw1 w2 ((z1 , z2 )). Therefore the group S1 acts on
S3 .
Consider the mapping f : S3 /S1 → S2 is given by

f ([z1 , z2 ]) = (2z1 z 2 , kz1 k2 − kz2 k2 ).

Assume that f ([z1 , z2 ]) = f ([u1 , u2 ]). Then this is equivalence to
(
z1 z 2 = u1 u2 (1)
kz1 k2 − kz2 k2 = ku1 k2 − ku2 k2 (2)
It follows from (1) that
 97

z1 u2
= =v for u1 6= 0, z2 6= 0.
u1 z2
Hence
z1 = vu1 , u2 = vz2 . (3)
It follows from (2) that
kz1 k2 − ku1 k2 = kz2 k2 − ku2 k2 . (4)
Substituting (3) into (4) gives that

kvk2 ku1 k2 − ku1 k2 = kz2 k2 − kvk2 kz2 k2 since kvk = kvk .

Thus

ku1 k2 (kvk2 − 1) = kz2 k2 (1 − kvk2 ).

This only happens as kvk = 1 or v = w ∈ S1 . Therefore, z1 = wu1 , and
substituting this into (2) gives

wu1 z 2 = u1 u2 .
This implies z2 = wu2 , so z2 = wu2 .
If u1 = 0, then it follows from (1) and (2) that z1 = 0 and kz2 k = ku2 k,
that is z2 = w0 u2 , w0 ∈ S1 .
If z2 = 0, then it follows from (1) and (2) that u2 = 0 and kz1 k = ku1 k,
that is z1 = w0 u1 , w0 ∈ S1 .
Hence [z1 , z2 ] = [u1 , u2 ]. Thus f is an injective.
We shall prove that f is well-defined. Indeed, if (u1 , u2 ) ∼ (s1 , s2 ), then
u1 = ws1 and u2 = ws2 for some w ∈ S1 . We have

f ([u1 , u2 ]) = (2u1 u2 , ku1 k2 − ku2 k2 )

= (2ws1 ws2 , kwk2 ks1 k2 − kwk2 ku2 k2 )
= (2s1 s2 , ks1 k2 − ks2 k2 ) since ww = kwk2 = 1
= f ([s1 , s2 ]).
Hence f is well-defined.
98

2
√ t) ∈ S ,
We next prove that f is surjective. Indeed, for arbitrary (z,
iw
(z ∈ C, t ∈ R), we have z = re 0 for w0 = Arg(z), where r = 1 − t = 2
2 2
kzk (since kzk + t = 1). Consider
r
1+t
z1 = w,
2
r
1 − t iw0
z2 = we ,
2
where w ∈ S1 . Then (z1 , z2 ) ∈ S3 /S1 and

f ([z1 , z2 ]) = (2z1 z 2 , kz1 k2 − kz2 k2 )

p 1+t 1−t
= ( 1 − t2 eiw0 , − ) since kwk = 1, eiw = 1.
2 2
= (z, t).
Thus f is surjective. So f is bijective.
It is clear that f is continuous and S3 /S1 is compact and f is a bijective
from a compact set onto a Hausdorff space S2 , therefore f is a homeomor-
phism. That is to say S3 /S1 is homeomorphic to S2 .
Exercise 82. Show that Sn is a smooth (C ∞ ) manifold.
Solution. Let
Ui = {(x1 , x2 , . . . , xn+1 ) ∈ Rn+1 | xi = 0, x21 + x22 + · · · + x2i−1 + x2i+1 + · · · x2n+1 < 1},

ϕi ((x1 , x2 , . . . , xi−1 , xi+1 , . . . , xn+1 )) = (x1 , x2 , . . . , xi−1 , Di , xi+1 , . . . , xn+1 ) ∈ Rn+1 ,

q
where Di = 1 − (x21 + · · · x2i−1 + x2i+1 + · · · x2n+1 ), and

ψi ((x1 , x2 , . . . , xi−1 , xi+1 , . . . xn+1 )) = (x1 , . . . , xi−1 , −Di , xi+1 , . . . xn+1 ).

It is clear that ϕi ◦ ϕ−1 −1 −1 −1 ∞
j , ψi ◦ ψj , ϕi ◦ ψj , ψi ◦ ϕj are the smooth (C )
functions.
We also have
n+1
[
U i = Sn .
i=1
 99

The family {(Ui , ϕi ), (Ui , ψi ), i = 1, n + 1} is an atlas of the manifold

Sn . The functions ϕi and ψi are homeomorphisms. Therefore Sn is a
smooth (C ∞ ) manifold.
Exercise 83. Show that the notion of smoothness of a function f : M →
N is well-defined (i.e., it is independent of the choice of local charts).
Solution. Assume that for evry p ∈ M there exists other local charts
(U 0 , ϕ0 ) of M and (V 0 , ψ 0 ) of N , centered at p and f (p) respectively. Since M
has a differential structure, so let U1 = U ∩U 0 , V1 = V ∩V 0 , then (U1 , ϕ0 |U1 )
and (V1 , ψ 0 |V1 ) are local charts, centered at p and f (p), respectively.
We have ψ 0 |V1 ◦f ◦ϕ0 |U1 = ψ◦f ◦ϕ−1 is smooth (since (U1 , ϕ0 |U1 ) ⊂ (U, ϕ)
and (V1 , ψ 0 |V1 ) ⊂ (V, ψ)). Therefore the notion of smoothness of a function
f : M → N is well-defined.
Exercise 84 (Exercise 0.2 from [Car92].). Prove that the tangent bun-
dle of a differential manifold M is orientable (even though M may not be).

Solution. Let {(Uα , xα )} be any atlas of M , and  be the atlas of T M

given by {(Vα = π −1 (Uα ), x̂α = (xα × id) ◦ ϕα )}, where ϕα : Vα → Uα × Rn
are the trivializing maps. We shall prove that this is an oriented atlas.
Define φ̂α (u, v) = x̂β ◦ x̂−1 α .
For a vector (u, v) ∈ R × Rn , then φ̂(u, v) = (φ(u), A(u) · v), where
n
∂φi
φ = xβ ◦ x−1
α , and Aij (u) = ∂uj |u .
Let (u1 , . . . , un , v 1 , . . . , v n ) be coordinates around (u, v) ∈ Rn × Rn .
Given i, j = 1, 2, . . . , n, one has

∂φi ∂φi
|(u,v) = = Aij (u),
∂uj ∂xj
∂φi
| = 0,
∂v j (u,v)
∂φn+i ∂
j
|(u,v) = j |(u,v) (A(u) · v) = Aij (u).
∂v ∂v
Therefore, the Jacobian matrix of φ̂ has the block form
 
A ∗
0 A
and its determinant equals det(A)2 > 0. Thus T M is orientable.
100

Exercise 85 (Exercise 0.5 (Embedding of P 2 (R) in R4 ) from [Car92].).

Let F : R3 → R4 be given by

F (x, y, z) = (x2 − y 2 , xy, xz, yz), (x, y, z) = p ∈ R3 .

Let S 2 ⊂ R3 be the unit sphere with the origin 0 ∈ R3 . Observe
that the restriction ϕ = F |S 2 í such that ϕ(p) = ϕ(−p), and consider the
mapping ϕ̃ : P 2 (R) → R4 given by

ϕ̃([p]) = ϕ(p), [p] = equiv. class of p = {p, −p}.

Prove that
(a) ϕ
e is an immersion.
e is injective; together with (a) and the compactness of of P 2 (R),
(b) ϕ
this implies that ϕe is an embedding.
Solution. (b) We shall prove that ϕ e is an embedding of P 2 (R) into
R4 . We first prove ϕe is injective. Indeed, assume that ϕ([p])
e = ϕ([q])
e for
3
p = (x, y, z) and q = (u, v, w) in R . This is equivelent to

 2 2
x − y
 = u2 − v 2 , (1)

xy = uv, (2)

xz = uw, (3)

yz = vw.


Since P 2 (R) is the quotient of S2 by the equivalent relation that identi-

fies p ∈ S2 with its anipodal point, so kpk = kqk = 1, that is x2 + y 2 + z 2 =
u2 + v 2 + w2 = 1. Combine this together with (1) and (3) to obtain
(
x2 + x2 = u2 + w2
xz = uw
This system has solutions
(
x =u
z =w
or
 101

(
x = −u
z = −w
Combine these with (2) to obtain

x = u

y =v

z =w

or

x = −u

y = −v

z = −w


That is p ∼ q. Then [p] = [q], and hence ϕ

e is injective.
2
Since P (R) is a compact set and ϕ e is injective from P 2 (R) into the
4
Hausdorff space R , so ϕe is an embedding of P 2 (R) in R4 .
(a) But an embedding is also an emersion, hence ϕ e is an immersion.
Exercise 86 (Exercise 0.8 from [Car92].). Let M1 and M2 ve differen-
tial manifolds. Let ϕ : M1 → M2 be a local diffeomorphism. Prove that if
M2 is orientable, then M1 is orientable.
Solution. Assume that M2 is orientable and it admits a differentiable
structure {(Uα , xα )}. At each point p ∈ M2 , there exists a neighborhood
V of p and W of ϕ(p) such that V and W are diffeomeomorphic under the
mapping ϕ.
Let yα : Rn → M1 be given by yα = ϕ−1 ◦ xα .
The differential structure {(Vα , yα )} of M1 satisfies

yα−1 ◦ yβ = yα−1 ◦ ϕ−1 ◦ xp

= x−1
α ◦ϕ◦ϕ
−1
◦β
= x−1
α ◦ xβ since ϕ ◦ ϕ−1 = id.
But M2 is orientable, so the differential of x−1
α ◦xβ has positive determinant.
This implies that the differential of yα−1 ◦ yβ has positive determinant. Thus
M1 is orientable.
102

Exercise 87 (Exercise 0.9 from [Car92].). Let G × M → M be a

properly discontinuous action of a group G on a differentiable manifold M .
(a) Prove that the manifold M/G is orientable if and only if there exists
an orientation of M that is preserved by all the diffeomorphisms of G.
(b) Use (a) to show that the projective plane P 2 (R), the Klein bottle
and the Mobius band are non-orientable.
(c) Prove that P n (R) is orientable if and only if n is odd.

Solution. (a) Let π : M → M/G ve given by

π(p) = [p] for p ∈ M.

M is orientable if and only if there exist a chart {(Uα , xα )} such that for
every pair α, β with xα (Uα ) ∩ xβ (Uβ ) 6= ∅, the differential of x−1
α ◦ xβ has
positive determinant.
Note that the family {(π(Uα ), π ◦ xα )} is a differntial structure of M/G.
(Note that xα : Uα ⊂ Rn → M and π is an open mapping).
We have that

(π ◦ xα )−1 ◦ (π ◦ xβ ) = x−1
α ◦π
−1
◦ π ◦ xβ
= x−1
α ◦ xβ .

The differential of (π ◦ xα )−1 ◦ (π ◦ xβ ) has positive determinant if and

only if the differential of x−1
α ◦ xβ has positive determinant. Thus M/G is
orientable if and only if M is orientable.
(b) We first show that Mobius strip is not orientable. Consider M is the
right circular cylinder given by C = {(x, y, z) ∈ R3 | x2 + y 2 = 1, −1 < z <
1}. The diffeomorphisms of M are formed by {A, id}, where A(p) = −p
for p ∈ M . Then M/G is a Mobius strip. Since M is non-orientable, so by
applying (a) we obtain that the Mobius strip is non-orientable.
Since P 2 (R) contains a Mobius strip and Mobius strip is non-orientable,
so P 2 (R) is non-orientable since it admits an atlas such that there exists a
pair of charts (Uα , xα ), (Uβ , xβ ) of M , x−1
α ◦xβ is not orientation preserving.
When M is the torus of revolution T 2 , then M/G is is a Klein bottle
and we know that M is non-orientable, so M/G is non-orientable by (a).
(c) Recall that we have a natural mapping π : Sn → RPn which is
a local diffeomorphism, and is given simply by π(p) = {p, −p}. Now let
r : S n → Sn be the reflection through the origin. Then
 103

π ◦ r = π.
n
If RP is orientable then we may assume that π preserves orientation.
Then the above inequality implies that π ◦ r preserves orientation as well.
This is not possible only if r preserves orientation which is the case only
when n is odd. Thus RPn is not orientable when n is even.
It remains to show that RPn is orientable when n is odd. In this we
may orient each tangent space T[p] RPn is as follows: take a representative
from q ∈ [p] = {p, −p}. Choose a basis of Tq Sn which is in its orientation
class, and let the image of this basis under dπ determine the orientation
class of T[p] RPn . This orientation is well defined because it not effected by
whether q = p or q = −p. Indeed, in (b1 , . . . , bn ) is a basis in the orientation
class of Tp Sn and (b01 , . . . , b0n ) is a basis in the orientation class of T−p Sn
then

(dπp (b1 ), . . . , dπ(bn )) and (dπ−p (b01 ), . . . , dπ(b0n ))

belong in the same orientation class of T[p] RPn ; because

dπp (bi ) = d(π ◦ r)p (bi ) = dπr(p) ◦ drp (bi ) = dπ−p ◦ drp (bi )
and r preserves orientation, i.e., (drp (b1 ), . . . , drp (bn )) belongs in the same
orientation class as (b01 , . . . , b0n ).
Exercise 88. Show that On is a smooth n(n − 1)/2-dimensional sub-
manifold of GLn . (Hint: Define f : GLn → GLn by f (A) := AT A. Then
show that TA GLn is given by the equivalence class of curves of the form
A + tB where B is any n × n matrix. Finally, show that dfA (TA GLn ) is
isomorphic to the space of symmetric n × n matrices).
Solution. Consider the mapping f : GLn → GLn given by

f (A) = AT A
for any A ∈ GLn . We first note that A ∈ On is equivalence to AT A = I
where I is the identity matrix. But det(AT A) = A2 and det I = 1, so
det A = ±1. This implies that On = f −1 (I). f is the composition of
C ∞ -maps, so f is C ∞ .
Since [f (A)]T = (AT A)T = AT A = f (A), so f (A) is a symmetrix
matrix.
104

We have

(A + tB)T (A + tB) − AT A
dfA (B) = lim
t→0 t
(AT + tB T )(A + tB) − AT A
= lim
t→0 t
= AT B + B T A
= AT B + (AT B)T .

dfA is a composition of two maps

ϕ : GLn → GLn
B 7→ AT B,
ψ : GLn → Sym(n)
B 7→ B + B T ,

where Sym(n) is the set of invertible symmetric matrices. ϕ is onto since

AT is invertible and ψ is onto since for any C ∈ Sym(n), then ψ( 21 C) = C.
Thus dfA = ψ ◦ ϕ is surjective and has rank n(n+1)
2 .
Applying Theorem 2 of Lecture Note 7 to obtain that f −1 (I) = On is a
smooth submanifold of GLn and it has dimensional n2 − n(n+1)2 = n(n−1)
2 .
n(n−1)
Thus On is a smooth 2 dimensional submanifold of GLn .

Exercise 89 (Exercise 1.1 from [Car92].). Prove that the antipodal

mapping A : S n → S n given by A(p) = −p í an ísometry of S n . Use this
fact to introduce a Riemannian metric on the real projective space P n (R)
such that the natural projection π : S n → P n (R) is a local isometry

Solution. Since A(p) = −p, so

A(p + tq) − A(p)

dAp (q) = − lim
t→0 t
−p − dq − (−p)
= lim
t→0 t
= −q.
 105

Thus aAp (q) = −q. Thereofore

hu, vip = h−u, −vi−p since the inner product is bilinear
= hdAp (u), dAp (v)iA(v) .
Hence A is an isometry of Sn . Define an inner product on P n (R) by

h[p] , [q]i = |hp.qi|

for p.q ∈ Sn .
This inner product is well-defined because it is not depend
on the representation of the class.
At each point p ∈ Sn , there is a neighborhood U such that for any
u, v ∈ U , then hu, vi ≥ 0 since hp, pi ≥ 0 and the inner product on Sn is
continuous. Thus for u, v ∈ U , we have
hu, vi[p] = h−u, −vip

= hdAp (u), dAp (v)iA(p)

= hdπp (u), dπp (v)iA(p)

= hdπp (u), dπp (v)iπ(p) since dπp (u) = π(u) and Ap (u) ∈ [u]
Thus this inner product is a Riemann metric on P n (R) and π is a local
isometry.
Exercise 90 (Exercise 1.3 from [Car92].). Obtain an isometric immer-
sion of the flat torus T n into R2n
Solution. We have T n = S 1 × S 1 × · · · × S 1 (The Cartesian product
of n coppies of S 1 ).
Consider the mapping

ψ : T n → R2n
is defined by

ψ(eix1 , eix2 , . . . , eixn ) = (cos x1 , sin x1 , cos x2 , sin x2 , . . . , cos xn , sin xn )

where xi ∈ R, ∀i ∈ {1, 2, . . . , n}.
For p = (eix1 , . . . , eixn ), q = (eiu1 , . . . , eiun ) ∈ T n , then

p + tq = (eix1 + teiu1 , . . . , eixn + teiun ).

106

So

ψ(p + tq) − ψ(p)

dψp (q) = lim = q.
t→0 t
Thus dψp is injective, and so ψ is an immersion.
We have

hu, vip = hdψp (u), dψp (v)iψ(p)

since dψp (q) = q. Hence ψ is an isometric immersion.
Exercise 91. Show that the the Poincare half-plane and the half-disk
are isometric (Hint: identify the Poincare half-plane with the region y > 1
in R2 and do an inversion).
Solution. Let k·kp be the norm generated by the inner product gp (X, Y )=
hX,Y i p
2 2 . That is kXkp =
(1−kpk )
gp (X, X), X ∈ B n = {X ∈ Rn | kXkp < 1}.
Let H n = {(x1 , . . . , xn ) ∈ Rn | xn > 0}. Define the mapping f : B n → H n
by

p − p0
f (p) = 2 − (0, 0, . . . , 0, 1)
kp − p0 k2
where p0 = (0, 0, . . . , 0, −1) ∈ Rn , p ∈ B n . For v is a vector at p and h·, ·i
denotes the inner product in the Euclidean metric,

hv, vi
hdfp (v), dfp (v)i = .
kp − p0 k4
Let f (p) = (f1 (p), f2 (p), . . . , fn (p)). Then

p0 + 1
fn (p) = 2 −1
kp − p0 k2
2pn + 2 − ni=1 p2i − 2pn − 1
P
=
kp − p0 k2
1 − kpk2
= >0
kp − p0 k2
for p = (p1 , p2 , . . . , pn ) ∈ B n .
 107

hX,Xi
Note that X ∈ B n . Then kXk2p < 1. so (1−kpk2 )2
< 1. Thus kXk <
2
1 − kpk < 1. We have

hdfp (v), dfp (v)i

hp (v, v) =
[fn (p)]2
kp − p0 k4 hv, vi
=
(1 − kpk2 )2 kp − p0 k4
= gp (v, v)

where hp (X, Y ) = hX,Y

(pn )2
i
is the inner product in H n . Since f is injective, so
we conclude that f is an isometry of B n onto H n .
Exercise 92. Compute the metric of the surface given by the graph of
a function f : Ω ⊂ R2 → R.
Solution. Let z = f (x, y) be the function representing the surface in
R3 . Define the function g : R2 → R3 by

g(x, y) = (x, y, f (x, y)).

We have

∂f
D1 g(x, y) = (1, 0, ),
∂x
∂f
D2 g(x, y) = (0, 1, ).
∂y
Therefore, gij (f (x, y)) is given by
!
1 + ( ∂f
∂x )
2 ∂f ∂f
∂x ∂y
∂f ∂f .
∂x ∂y 1 + ( ∂f
∂y )
2

Exercise 93. Compute the length of the radius of the Poincare-disk

(with respect to the Poincare metric).
p
Solution. Let B n = {p ∈ Rn | kpkB < 1} where kpkB = gp (p, p)
is the norm generated by the inner product gp . For any X ∈ B n , then
kXk2B < 1. This is equivalnce to
108

gp (X, X) < 1
hX, Xi
⇐⇒ <1
(1 − kpk2 )2
⇐⇒ kXk < 1 − kpk2 .

Therefore the radius of B n is r = 1 − kpk2 .

Exercise 94 (Exercise 2.1 from [Car92].). Let M be a Riemannian
manifold. Consider mapping
P = Pc,to ,t : Tc(to ) M → Tc(t) M
defined by: Pc,to t (v), v ∈ Tc(to ) M , is the vector obtained by parallel
transporting the vector v along the curve c. Show that P is an isometry
and that, if M is orientel, P preserves the orientation.
Solution. The vector field V in the parallel transporting along the
curve c. Since DV
dt = 0, so

   
d DV (t1 ) DV (t2 )
hV (t1 ), V (t2 )i = , V (t2 ) + V (t1 ),
dt dt dt
=0
for all t1 , t2 ∈ I and c : I → M . Therofore hV (t1 ), V (t2 )i = const. In
particular, hV (t), V (t)i = const, for t ∈ I. Thus

hPc,t0 ,t (V (t0 )), Pc,t0 ,t (V (t0 ))i = hV (t), V (t)i

= hV (t0 ), V (t0 )i
or

kPc,t0 ,t (u)k = kuk ,

for u ∈ Tc(t0 ) M .
Since Pc,t0 ,t (v), v ∈ Tc(t0 ) M is linear, so for u, v ∈ Tc(t0 ) M ,

kPc,t0 ,t (u − v)k = ku − vk ,
which is equivalent to
 109

kPc,t0 ,t (u) − Pc,t0 ,t (v)k = ku − vk .

Thus Pc,t0 ,t is a linear isometry.
Let {E1 , E2 , . . . , En } be an oriented basis at Tp M where p = c(t0 ). De-
note Pt = Pc,t0 ,t be the parallel transporting. Since the parallel transporting
preserves angles and lengths, so {Pt E1 , Pt E2 , . . . , Pt En } is an orthogonal
frame along c. We have

dp Pc,t0 ,t = (Pt E1 , Pt E2 , . . . , Pt En ).
Let f (t) = det(Pt E1 , . . . , Pt En ). If f (b) < 0, where b ∈ [0, 1] (and c : [0, 1] →
M is a smooth curve), then by Picard-Lindeloft, f (t) is snooth in t, and so
by the intermediate value theorem, there is a t ∈ [0, a] such that f (t) = 0.
This contradicts to the fact that {Pt E1 , . . . , Pt En } is an orthogonal basis.
Thus, the parallel transporting preserves orientation.
Exercise 95 (Exercise 2.2 from [Car92].). Let X and Y be differ-
entiable vector fields on a Riemannian manifold M . Let p ∈ Mand let c:
I → M be an integral curve of Xthrough p, i.e. c(to ) = p and dc
dt = X(c(t)).
Prove that the Riemannian connection of M is
d −1
(∇X Y )(p) = (P (Y (c(t))))t=to
dt c,t0 ,t
where Pc,to ,t : Tc(to ) M → Tc(t) M is the parallel transport along c, from to to
t (this shows how the connection can be reobtained from the concept of
parallelism).
Solution. By Proposition 2.2 from [Car92],

DV
∇ dc Y (p) = ∇X(c(t)) Y (p) = ∇X Y (p) = |t (t) (1)
dt dt 0
for V (t) = Y (c(t)). Moreover, since

Pc,t0 ,t (V (t0 )) = V (t),

so

−1
V (t0 ) = Pc,t 0 ,t
(V (t))
−1
= Pc,t 0 ,t
(Y (c(t))).
110

But
d −1 dV (t0 )
Pc,t0 ,t (Y (c(t)))|t0 = |t0 = 0 (2)
dt dt
It follows from (1) and (2) that

d −1
∇ dc Y (p) = P (Y (c(t)))|t0
dt dt c,t0 ,t
n+k
Exercise 96 (Exercise 2.3 from [Car92].). Let f : M n → M be
an immersion of differentiable mani-fold M into a Riemannian manifold M .
Assume that M has the Riemannian metric inducel by f (cf. Example 2.5
of Chap. 1). Let p ∈ M and let U ⊂ M be a neighborhood of p such
that f (U ) ⊂ M is a submanifold of M . Futher, suppose that X, Y are
differentiable vector fields on f (U ) which extend to differentiable vector
fields on f (U ) which extend to differentiable vector fields X, Y on an open
set of M . Define (∇X Y )(p) =tangential component of ∇X Y (p), where ∇
is the Riemannian connection of M
Solution. Denote

⊥
∇X Y (p) = (∇X Y )(p) + ∇X Y (p)
⊥
where (∇X Y )(p), (∇X Y )(p) are the tangential component and the normal
component. respectively. We have
⊥ ⊥
∇X Y − ∇Y X = ∇X Y (p) − ∇X Y (p) − ∇Y X(p) + ∇Y X(p)
⊥ ⊥
= [∇X Y (p) − ∇Y X(p)] + [∇Y X(p) − ∇X Y (p)]
where pr1 is the projection
= [X, Y ] − pr1 [∇X Y (p) − ∇Y X(p)]
on the normal at p
= [X, Y ] − pr1 [[X, Y ]]
= [X, Y ].

Thus ∇X Y is symmetric.
For any pair of parallel vector fields P and P 0 along C, we have

DP 0
   
d DP 0
P, P 0 = , P + P,
dt dt dt
=0
 111

0
since DP DP 0
dt = dt = 0. Thus hP, P i = const. Therefore ∇ is compatible
with the Riemann metric and so ∇ is the Riemann connection of M .
Exercise 97 (Exercise 2.4 from [Car92].). Let M 2 ⊂ R3 be a surface in
R3 with the induced Riemannian metric. Let c : I → M be a differentiable
curve on M and let V be vector field tangent to M along c; V can be thoughr
of as a smooth function V : I → R3 , with V (t) ∈ Tc(t) M .
(a) Show that V is parallel if and only if dV
dt is perpendicular to Tc(t) M ⊂
dV
R where dt is the usual derivative of V : I → R3 .
3

(b) If S 2 ⊂ R3 is the unit sphere of R3 , show that the velocity field

along great circles, parametrized by arc length, is a parallel field. A similar
argument holds for S n ⊂ Rn+1 .
Solution. (a) Note that on the Euclidean space Rn , Γkij = 0, so DV dt =
dV
dt where V : I → Tp M . That is, the covariant derivative concides with
the usual derivative of V : I → Tp M .
We first suppose that V is parallel. We have DV dV
dt = 0, and so dt = 0.
Thus dVdt is perpendicular to Tc(t) M .
Conversely, suppose that dV dV
dt is perpendicular to Tc(t) M . Then dt , P =
0 for any P ∈ Tc(t) M . This implies that DVdt , P = 0 for any P ∈ Tc(t) M
DV dV DV
(since dt = dt ). In particular, if P = dt , then we obtain
 
DV DV
, = 0.
dt dt
Thus DV
dt = 0. Therefore, V is parallel.
(b) WLOG, we consider the great circle

c(θ, 0) = (sin θ, 0, cos θ).

The velocity vector at p = (sin θ, 0, cos θ) is

·
c(θ, 0) = (cos θ, 0, − sin θ).
·
D E
dc
We have c(θ, 0), c(θ, 0) = sin θ cos θ − cos θ sin θ = 0. Thus dθ is perpen-
dicular to Tp S2, so the velocity field is a parallel field by (a).
For S n = {(x1 , . . . , xn+1 ) | n+1 2
P
i=1 xi = P1}, then as xn+1 = 0 we obtain
n n 2
the great circle F = {(x1 , . . . , xn , 0) | i=1 xi = 1}. Consider V =
n 2 2 2
(x1 , . . . , xn , 0) ∈ F . Then x1 + x2 + · · · + xn = 1. Thus
112

· ·
x1 x1 + · · · + xn xn = 0. (1)
The tangent plane surface at p = (ξ1 , . . . , ξn ) is

n
X
(xi − ξi )ξi = 0.
i=1

For any (x1 , . . . , xn ) ∈ Tp S n , it follows from (1) that

 
dV
(x1 , . . . , xn ), = 0.
dt
dV
Thus dt is perpendicular to Tp S n . Therefore the velocity field is parallel.
Exercise 98 (Exercise 2.5 from [Car92].). In Euclidean space, the par-
allel transport of a vector between two points does not depend on the curve
joining the two points.Show, by example, that this fact may not be true on
an arbitrary Riemannian manifold
Solution. Note that in the Euclidean space, DV dV
dt = dt , where V : I →
DV dV
Tp M . If V is a parallel transport, then dt = dt = 0. Thus it is not depend
on the curve joining the two points.
Consider the upper half plane

R2+ = {(x, y) | y > 0}

with the metric given by g11 = g22 = y12 , g12 = 0. Hence we obtain
Γ111 = Γ212 = Γ122 = 0, Γ211 = y1 , Γ112 = Γ222 = − y1 .
Since Γ212 6= 0 and Γ222 = Γ112 6= 0, so the fact

DV X dv k X dxi
= { + Γkij v j }Xk
dt dt dt
k i,j

implies DV dV
dt 6= dt .
Further, consider v0 = (0, 1) be a tangent vector at point (0, 1) of R2+ .
Let v(t) be the parallel tranport of v0 along the curve x = t, y = 1. The
field v(t) = (a(t), b(t)) satisfies
 113

(
da
dt + Γ112 b = 0,
db
dt + Γ211 a = 0.
Taking a = cos θ(t), b = sin θ(t) and along the given curve y = 1, we obtain
from the equation above that dθ π
dt = −1. Since v(0) = v0 , so θ(t) = 2 − t.
This shows that the parallel transport depends on the curve joining the two
points.
Exercise 99. Show that the bracket satisfies the following properti

[X, Y ] = −[Y, X] and [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
Solution. We first prove that
[X, Y ] = −[Y, X].
We have that

[X, Y ]p f = Xp (Y f ) − Yp (Xf )
= −[Yp (Xf ) − Xp (Y f )]
= −[Y, X]p f
for any function f on M that is differentiable at p. Thus

[X, Y ] = −[Y, X],

as required.
We now prove that
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
We have that

[X, [Y, Z]]p f = Xp ([Y, Z]p f ) − [Y, Z]p (Xf )

= Xp (Yp (Zf )) − Xp (Zp (Y f )) − Yp (Zp (Xf )) + Zp (Yp (Xf ))
= Xp Yp (Zf ) − Xp Zp (Y f ) − Yp Zp (Xf ) + Zp Yp (Xf ). (1)
Similarly,

[Y, [Z, X]]p f = Yp Zp (Xf ) − Yp Xp (Zf ) − Zp Xp (Y f ) + Xp Zp (Y f ) (2)

114

and

[Z, [X, Y ]]p f = Zp Xp (Y f ) − Zp Yp (Xf ) − Xp Yp (Zf ) + Yp Xp (Zf ). (2)

Add (1), (2) and (3) to obtain

[X, [Y, Z]]p f + [Y, [Z, X]]p f + [Z, [X, Y ]]p f = 0

for any function f that is differentiable at p. Therefore

[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.

Exercise 100. Show that a connection is symmetric if and only the
correspding Christoffel symbold satisfy

Γkij = Γkji
in every local chart.
Solution. By Corollary 0.3 of Lecture Notes 15,

[Xi , Xj ] = 0.
Thus

∇Xi Xj − ∇Xj Xi = [Xi , Xj ]

= 0.
So

∇Xi Xj = ∇Xj Xi .
But

n
X n
X
∇Xi Xj = Γkij Xk and ∇Xj Xi = Γkji Xk .
k=1 k=1

Therefore
 115

n
X n
X
Γkij Xk = Γkji Xk .
k=1 k=1
This implies that

n
X
(Γkij − Γkji )Xk = 0.
k=1
Since X1 , X2 , . . . , Xk are linear independent, it follows that

Γkij − Γkji = 0 for k ∈ {1, 2, . . . , n}.

Therefore
Γkij = Γkji for k ∈ {1, 2, . . . , n} and i, j ∈ {1, 2, . . . , n}
in every local chart.
Exercise 101 (Exercise 2.7 from [Car92].). Let S 2 ⊂ R3 be the unit
sphere, c an arbitrary parallel of latitude on S 2 and Vo a tangent vector to
S 2 ai a point of c. Describe geometrically the parallel transport of Vo along
c.
Hint: Consider the cone C tengent to S 2 along c and show that the
parallel transport of Vo along c is the same, whether taken relative to S 2 or
to C.
Solution. Consider the cone C tangent to S 2 along c. Then the
latitude c(t) lies on the cone C. For any tange vector V0 at the point p
of c(t) and q is the arbitrary point of c(t). Let Rpqt be the rotation that
has the rotation axis be the axis of the cone C and it maps p to q where
p = c(t0 ), q = c(t). Set V (t) = Rpqt (V0 ) where Rpqt V0 is a parallel transport
the vector V0 along c. Thus the parallel transport of V0 along c is the same,
whether taken relative to S 2 or to C.\
Exercise 102 (Exercise 2.8 from [Car92].). Consider the upper half-
plane

R2+ = {(x, y) ∈ R2 ; y > 0}

1
with the metric given by g11 = g22 ≈ ,g
y 2 12
= 0 (metric of Lobatchevski’s
non-euclidean geometry).
116

(a) Show that the Christoffel symbols of the Riemannian connection

are: Γ111 = Γ212 = Γ122 = 0, Γ211 = y1 , Γ112 = Γ222 = − y1
(b) Let vo = (0, 1) be a tangent vector at point (0, 1) or R2+ (vo = (0, 1)
is a unit vector on the y-axis with origin at (0, 1)).
Let v(t) be the parallel transport of vo along the curve x = t, y = 1.
Show that v(t) makes an angle t with the direction of the y-axis, measured
in the clockwise sense.
Hint : The field v(t) = (a(t), b(t)) satisfies the system (2) which defines
a parallel and which, in this case, simplifies to
(
da 1
dt + Γ12 b = 0,
db 2
dt + Γ11 a = 0.
Talking a ≈ cos θ(t), b = sin θ(t) and nothing that along the given carve
we have y = 1, we obtain from the equations above that dθ dt = −1. Since
v(0) = vo , this implies that θ(t) = π/2 − t.
Solution. (a) We have
!
1
y2
0
gij = 1 .
0 y2

This imples that

y2 0
 
ij −1
g = (gij ) = .
0 y2
Thus

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ111 = ( g11 + g11 − g11 )g 11 + ( g11 + g11 − g11 )g 21
2 ∂x ∂x ∂x 2 ∂x ∂x ∂x
=0
1 ∂ 1
since g11 = y2
, so ∂x y 2 = 0 and g 21 = 0. Moreover,

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ112 = ( g21 + g11 − g11 )g 12 + ( g22 + g21 − g12 )g 22
2 ∂x ∂y ∂x 2 ∂x ∂y ∂y
1 ∂ 1 2
= ( )y = 0
2 ∂x y 2
1
since g 12 = 0, g21 = 0 and g22 = y2
. Further,
 117

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ122 = ( g21 + g12 − g22 )g 11 + ( g22 + g22 − g22 )g 22
2 ∂y ∂y ∂x 2 ∂y ∂y ∂y
1 −2 1 1 −2 1
= − ( 2 ) 2 + ( 2 ) 2 = 0.
2 y y 2 y y
Hence Γ111 = Γ212 = Γ122 = 0. We have

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ211 = ( g11 + g11 − g11 )g 12 + ( g12 + g21 − g11 )g 22
2 ∂x ∂x ∂x 2 ∂x ∂x ∂y
1 ∂ 1 1 1 2
= ( 2
− )y
2 ∂x y ∂y y 2
1
= − (−2y −3 )y 2
2
y2 1
= 3 =
y y
since g 12 = 0. Moreover,

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ222 = ( g21 + g12 − g11 )g 12 + ( g22 + g22 − g22 )g 22
2 ∂y ∂y ∂y 2 ∂y ∂y ∂y
1 ∂ 1 2
= ( )y
2 ∂y y 2
1 2
= (− 3 )y 2
2 y
−1
=
y
since g 12 = 0. Further,

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ112 = ( g21 + g11 − g12 )g 11 + ( g22 + g21 − g12 )g 21
2 ∂x ∂y ∂y 2 ∂x ∂y ∂y
1 ∂ 1 2
= ( )y
2 ∂y y 2
1 2
= (− 3 )y 2
2 y
−1
=
y
118

since g 21 = 0, g21 = g12 = 0, g11 = y12 , g 11 = y 2 .

Thus Γ211 = y1 and Γ112 = Γ222 = − y1 .
(b) The field v(t) = (a(t), b(t)) satisfies the system

dv k X k j dxj
+ Γij v = 0, for k = 1, n.
dt dt
i,j

In this case, this simplies to

(
da
dt + Γ112 b = 0,
db (1)
dt + Γ211 a = 0.
Taking a = cos θ(t), b = sin θ(t). Then

da dθ db dθ
= − sin θ(t) and = cos θ(t) .
dt dt dt dt
The system (1) becomes
(
− sin θ(t) dθ 1
dt − y sin θ(t) = 0,
cos θ(t) dθ 1
dt + y cos θ(t) = 0.
Note that Γ112 = − y1 , Γ211 = y1 by part (a). We obtain from the system
of equations that dθ 1
dt = − y . But note that along the given curve, we have
y = 1, so dθdt = −1. This implies that θ(t) = −t + C for C = const.
Subsitute θ(t) = −t + C into v(t) to obtain

v(t) = (cos(−t + C), sin(−t + C)).

Since v(t) is a parallel transport of v0 = (0, 1) along the curve x = t, y = 1,
so v(0) = v0 = (0, 1). Therefore, (cos c, sin c) = (0, 1), which implies that
C = π2 . Hence θ(t) = −t + π2 .
Exercise 103 (Exercise 2.9 from [Car92].). (Pseudo-Riemannian Met-
rics). A pseudo-Riemannian metric on a smooth manifold M is a choice,
at every point p ∈ M , of a non-degenerate symmetric bilinear form h, i (not
necessarily positive definite) on Tp M which varies differentiably with p. Ex-
cept for the fact that h, ineed not be postive definite, all of the definitions
that have been presendted up to now make sense for a pseudo-Riemannian
 119

metric. For example, an affine connection on M conpatible with a pseudo-

Riamannian metric on M satisfies equation (4); if, in addition, () holds, the
affine connection is said to be symmetric.
(a) Show that the theorem of Levi-Civita extends to pseudo-Riemannian
metrics. The connection so abtained is called the pseudo-Riemannian con-
nection
(b) Introduce a pseudo-Riemannian metric on Rn+1 by using the qua-
dratic form:

Q(x0 , ..., xn ) = −(xo )2 + (x1 )2 + ... + (xn )2 , (xo , ..., xn ) ∈ Rn+1 .

Show that the parallel transport corresponding to the Levi_Civita connec-
tion of the metric coincides with the usual parallel transport of Rn+1 (this
pseudo-Riemannian metric is called the Lorentz metric; for n = 3, it ap-
pears naturally in relativity).
Solution. (a) The demonstration og the theorem of Levi-Civita dies
not use positive definite and the identities which do not depend on positive
definite. Therefore, for the pseudo-Riemannian connection, the theorem of
Levi-Civita also holds.
(b) Let g(x, y) be the pseudo-inner product with respect to the pseudo-
Riemannian connection. The quadratic form Q(x), x ∈ Rn+1 satisfies
Q(x) = g(x, x), where

Q(x + y) = g(x + y, x + y) for x, y ∈ Rn+1

= g(x, x) + 2g(x, y) + g(y, y)
Q(x) + 2g(x, y) + Q(y).
Therefore, g(x, y) = 21 [Q(x + y) − Q(x) − Q(y)]. For x = (x0 , . . . , xn ),
y = (y0 , . . . , yn ), then

Q(x) = g(x, x) = −x20 + x21 + x22 + · · · + x2n ,

Q(y) = g(y, y) = −y02 + y12 + y22 + · · · + yn2 ,

Q(x + y)
= − (x0 + y0 )2 + (x1 + y1 )2 + (x2 + y2 )2 + · · · + −(xn + yn )2 ,
=(−x20 + x21 + x22 + · · · + x2n ) + (−y02 + y12 + y22 + · · · + yn
2
) + 2(−x0 y0 + x1 y1 + x2 y2 + · · · + xn yn )
=Q(x) + Q(y) + 2(−x0 y0 + x1 y1 + x2 y2 + · · · + xn yn ).
120

This implies that

1
g(x, y) = [Q(x + y) − Q(x) − Q(y)]
2
= −x0 y0 + x1 y1 + x2 y2 + · · · + xn yn .
Let gij = g(ei , ej ), i, j = 0, n + 1 (ei = (0, . . . , 0, 1
|{z} , 0, . . . , 0)).
ith position
Then

0,
 if i 6= j,
gij = 1, if i = j ≥ 1,

−1, if i = j = 0.


Use the formula

1X ∂ ∂ ∂
Γm
ij = { gjk + gki − gij }g km
2 ∂xi ∂xj ∂xk
k
to obtain

Γk00 = 0 for all k.

For i, j ≥ 1, then gij = hei , ej i = δij , so as in the Euclidean space, we have
Γkij = 0 for all i, j ≥ 1. thus Γkij = 0 for all k and all i, j. Hence

DV X dv k X dxi dV
= { + Γkij v j }Xk = .
Dt dt dt dt
k i,j

Thus the parallel transport corresponding to the pseudo-Levi-Civita con-

nection of this pseudo-metric coincides with the usual parallel transport of
Rn+1 .
Exercise 104 (Exercise 3.1 from [Car92].). (Geodesics of a surface of
revolution). Denote by (u, v) the cartesian coordinates of R2 . Show that
the function ϕ : U ⊂ R2 → R3 given by ϕ(u, v) = (f (v) cos u, f (v) sin u, g(v)),

U = {(u, v) ∈ R2 : uo < u < u1 ; vo < v < v1 },

where f and g are differentiable functions, with f 0 (v)2 + g 0 (v)2 6= 0 and
f (v) 6= 0, is an immersion. The image ϕ(U ) is the surface generated by
the rotation of the curve (f (u), g(v)) around the axis 0z and is called a
 121

surface of revolution S. The image by ϕ(U ) of the curves u =constant and

v =constant are called meridians and parallels, respectively, of S.
(a) Show that the induced matric in the coordinates (u, v) is given by

g11 = f 2 , g12 = 0, g22 = (f 0 )2 + (g 0 )2 .

(b) Show that local equations of a geodesic γ are

d2 u 2f f 0 du dv
+ 2 = 0,
dt2 f dt dt

d2 v ff0 du 2 f 0 f 00 + g 0 g 00 dv 2
− ( ) + 02 ( ) = 0.
dt2 (f 0 )2 + (g 0 )2 dt (f + (g 0 )2 dt
(c) Obtain the following geometric meaning of the equations above: the
second equation is, except for meridians and parallels, equivalent to the
fact that the “energy” | γ 0 (t) |2 of a geodesic is constant along γ; the first
equation signifies that if β(t) is the oriented angle, β(t) < π, of γ with a
parallel P intersecting γ at γ(t), them

r cos β =const.,
where r is the radius of the parallel P (the equation above is called (Clairaut’s
relation).
(d) Use Clairaut’s relation to show that a geodisic of the paraboloid

(f (v) = v, g(v) = v 2 , 0 < v < ∞, −ε < u < 2π + ε),

which is not a meridian, intersects itself an infinite number of times (Fig.
6).

Solution. The matrix of dϕ is

−f (v) sin u f 0 (v) cos u

 

(dϕ(u, v)) =  f (v) cos u f 0 (v) sin u  .

0 g 0 (v)

For p = (u, v), then

122

−f (v) sin u f 0 (v) cos u  

 
ξ
dϕp (ξ, η) =  f (v) cos u f 0 (v) sin u 
η
0 g 0 (v)
−f (v) sin(u)ξ + f 0 (v) cos(u)η
 

=  f (v) cos(u)ξ + f 0 (v) sin(u)η  .

g 0 (v)η
We have that


0
−f (v) sin(u)ξ + f (v) cos(u)η = 0 (1)

(ξ, η) ∈ ker dϕp ⇐⇒ f (v) cos(u)ξ + f 0 (v) sin(u)η = 0 (2)
 0

g (v)η = 0 (3)

* If g 0 (v) 6= 0, then it follows from (3) that η = 0. Since (f (v) sin u)2 +
(f (v) cos u)2 = f 2 (v) > 0, so it follows from (1), (2) and η = 0 that ξ = 0.
* If g 0 (v) = 0, then f 0 (v) = 0 (since [f 0 (v)]2 + [g 0 (v)]2 6= 0), whence
the system consisting of (1) and (2) has the determination 2f (v)f 0 (v) 6= 0.
Therefore it has a unique trivial solution (0, 0). Thus ker dϕp = {(0, 0)}, so
dϕp is an injective. Hence ϕ is an immersion.
(a) We have

X1 = (−f (v) sin u, f (v) cos u, 0),

X2 = (f 0 (v) cos u), f 0 (v) sin u, g 0 (v)).
Thus

g11 = hX1 , X1 i = [f (v)]2 (sin2 u + cos2 u) = [f (v)]2 ,

g12 = hX1 , X2 i = 0,
g22 = hX2 , X2 i = [f 0 (v)]2 (sin2 u + cos2 u) + [g 0 (v)]2 = [f 0 (v)]2 + [g 0 (v)]2 .

Therefore the induced metric in the coordinates (u, v) is

g11 = f 2 , g12 = 0, g22 = (f 0 )2 + (g 0 )2 .

(b) We have
 123

f2
 
0
(gij ) = .
0 (f 0 )2 + (g 0 )2
Therefore
!
1
f2
0
(g ij ) = 1 .
0 (f 0 )2 +(g 0 )2

Next, we compute Γkij ,

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ111 = [ g11 + g11 − g11 ]g 11 + [ g12 + g21 − g11 ]g 21
2 ∂x1 ∂x1 ∂x1 2 ∂x1 ∂x1 ∂x2
1 ∂ 1 ∂ 1 −f · f 0
= 2[ (f (v))2 ] − (f (v))2 0 2 =
2f ∂x1 2 ∂x2 (f ) + (g 0 )2 (f 0 )2 + (g 0 )2

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ112 = [ g21 + g11 − g12 ]g 11 + [ g22 + g21 − g12 ]g 21
2 ∂x1 ∂x2 ∂x1 2 ∂x1 ∂x2 ∂x2
ff0
= 2,
f

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ122 = [ g21 + g12 − g22 ]g 11 + [ g22 + g22 − g22 ]g 21
2 ∂x2 ∂x2 ∂x1 2 ∂x2 ∂x2 ∂x2
= 0,

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ121 = [ g11 + g12 − g21 ]g 11 + [ g12 + g22 − g21 ]g 21
2 ∂x2 ∂x1 ∂x1 2 ∂x2 ∂x1 ∂x2
ff0
= 2,
f

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ211 = [ g11 + g11 − g11 ]g 12 + [ g12 + g21 − g11 ]g 22
2 ∂x1 ∂x1 ∂x1 2 ∂x1 ∂x1 ∂x2
1 ∂ 1 ff0
=− (f )2 0 2 = −
2 ∂x2 (f ) + (g 0 )2 (f 0 )2 + (g 0 )2

1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ212 = [ g21 + g11 − g12 ]g 12 + [ g22 + g21 − g12 ]g 22
2 ∂x1 ∂x2 ∂x1 2 ∂x1 ∂x2 ∂x2
= 0,
124

Γ221 = 0,
1 ∂ ∂ ∂ 1 ∂ ∂ ∂
Γ222 = [ g21 + g12 − g22 ]g 12 + [ g22 + g22 − g22 ]g 22
2 ∂x2 ∂x2 ∂x1 2 ∂x2 ∂x2 ∂x2
1 ∂ 1 f 0 f 00 + g 0 g 00
= [(f 0 )2 + (g 0 )2 ] 0 2 = .
2 ∂x2 (f ) + (g 0 )2 (f 0 )2 + (g 0 )2
Thus the local equations of a geodesic γ are

d2 x1 2f f 0 dx1 dx2
+ 2 = 0,
dt f dt dt
d2 x2 dx1 2 dx2 2
+ Γ211 ( ) + Γ212 ( ) = 0,
dt2 dt dt
or

d2 u 2f f 0 du dv
+ 2 = 0, (4)
dt f dt dt
d2 v ff0 du 2 f 0 f 00 + g 0 g 00 dv 2
− ( ) + ( ) = 0, (5)
dt2 (f 0 )2 + (g 0 )2 dt (f 0 )2 + (g 0 )2 dt
(c) Now we consider geodesics except for meridians and parallels. We
have

γ 0 (t) = (f 0 v 0 cos u − f u0 sin u, f 0 v 0 sin u + f u0 cos u, g 0 v 0 ).

So

|γ 0 |2
= (f 0 v 0 )2 cos2 u + (f u0 ) sin2 u − 2f f 0 u0 v 0 cos u sin u
+ (f 0 v 0 )2 sin2 u + (f u0 )2 cos2 u + 2f f 0 u0 v 0 sin u cos u + (g 0 v 0 )2
= (f 0 v 0 )2 (cos2 u + sin2 u) + (f u0 )2 (sin2 u + cos2 u) + (g 0 v 0 )2
= (f 0 v 0 )2 + (f u0 )2 + (g 0 v 0 )2 since sin2 u + cos2 u = 1.

Thus

|γ 0 |2 = [(f 0 )2 + (g 0 )2 ](v 0 )2 + f 2 · (u0 )2 .

Taking derivative the equation above to obtain
 125

d
(|γ 0 |2 ) = 2(f 0 f 00 + g 0 g 00 )(v 0 )3 + 2[(f 0 )2 + (g 0 )2 ]v 0 v 00 + 2f f 0 (u0 )2 v 0 + 2f 2 u0 u00 . (6)
dt
Therefore

1 (|γ 0 |2 )0 f 0 f 00 + g 0 g 00 0 3 0 00 ff0 0 2 0 f 2 u0 u00

= (v ) + v v + (u ) v + .
2 (f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2
ff0 0 00
f f +g g 0 00
Substitute v 00 = (f 0 )2 +(g 0 )2
(u0 )2 − (f 0 2
0 )2 +(g 0 )2 (v ) from (5) into (6) to obtain

1 (|γ 0 |2 )0
2 (f 0 )2 + (g 0 )2
f 0 f 00 + g 0 g 00 0 2 ff0 f 0 f 00 + g 0 g 00 0 2 ff0 f 2 u0 u00
= 0 0
(v ) + v 0 [ 0 2 0
(u0 )2 − 0 2 0
(v ) ] + 0 2 0
(u0 )2 v 0 + 0 2
2
(f ) + (g ) 2 (f ) + (g )2 (f ) + (g ) 2 (f ) + (g ) 2 (f ) + (g 0 )2
2f f 0 f2
= (u0 )2 v 0 + 0 2 u0 u00 . (7)
(f 0 )2
+ (g )0 2 (f ) + (g 0 )2
0
Substitute u00 = − 2ff 2f u0 v 0 from (4) into (7) to obtain

1 (|γ 0 |2 )0 2f f 0 f2 2f f 0 0 2 0
0 2 0 2
= 0 2 0 2
(u0 )2 v 0 − 0 2 0
(u ) v
2 (f ) + (g ) (f ) + (g ) (f ) + (g ) f 2
2

= 0.
This implies that (|γ 0 |2 )0 = 0, so |γ 0 |2 = const.
Conversely, assume that |γ 0 |2 = const. Then (|γ 0 |2 )0 = 0, so

f 0 f 00 + g 0 g 00 0 3 0 00 ff0 0 2 0 f2
(v ) + v v + (u ) v + u0 u00 = 0. (8)
(f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2
0
Substitute u00 = − 2ff 2f u0 v 0 from (4) into (8) to obtain

f 0 f 00 + g 0 g 00 0 2 ff0 f2 ff0 0 2
v0[ (v ) + v 00
+ (u0 2
) − 2 (u ) ] = 0.
(f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2 f 2
Since v 0 6= 0, so the equation above is equivalent to

ff0 f 0 f 00 + g 0 g 00 0 2
v 00 − (u0 2
) + (v ) = 0.
(f 0 )2 + (g 0 )2 (f 0 )2 + (g 0 )2
Hence the equation (5) is equivalent to |γ 0 |2 = const.
We have
126

(f 2 u0 )0 = 2f f 0 u0 v 0 + f 2 u00
ff0 0 0
= f 2 (u00 + 2 uv)
f2
=0 by (4).
This implies that f 2 u0 = const. Since

| hxu , xu u0 + xv v 0 i |
cos β = = |f u0 |
|xu |
and f = r, so

r cos β = |f 2 u0 | = const.
(d) Let p0 be a point of the paraboloid and let P0 be the parallel of
radius r0 passing through p0 . Let γ be a parametrized geodesic passing
through p0 and making an angle θ0 with P0 . Since, by Clairaut’s relation,

π
r cos θ = const. = |c|, 0≤θ≤ ,
2
we conclude that θ increases with r.
Therefore, if we follow in the direction of the increasing parallels, θ
increases. It may happen that in some revolution surfaces γ approaches
asymptotically a meridian. We shall show in a while that such is not the
case with a paraboloid of revolution. That is, the geodesic γ intersects all
the meridians, and therefore it makes an infinite number of turns around
the paraboloid.
On the other hand, if we follow the direction of decreasing parallels,
the angle θ decreases and approaches the value 0, which corresponds to a
parallel of radius |c| (observe that if θ0 6= 0, |c| < r). Since no parallel
of the paraboloid is a geodesic, the geodesic γ is actually tangent to the
parallel of radius |c| at the point p1 . Because 1 is a maximum for cos θ,
the value of r will increase starting from p1 .We are, therefore, in the same
situation as before. The geodesic will go around the paraboloid an infinite
number of turns, in the direction of the increasing r’s, and it will clearly
intersect the other branch infinitely often.
Observe that if θ0 = 0, the initial situation is that of the point p1 .
 127

It remains to show that when r increases, the geodesic γ meets all the
meridians of the paraboloid. Observe initially that the geodesic cannot be
tangent to a meridian. Otherwise, it would coincide with the meridian by
the uniqueness part of Prop. 5 of Section 4.4 from [Car16]. Since the angle
θ increases with r, if γ did not cut all the meridians, it would approach
asymptotically a meridian, say M .
Let us assume that this is the case and let us choose a system of local
coordinates for the paraboloid z = x2 + y 2 , given by

x = v cos u, y = v sin u, z = v2,

0 < v < +∞, 0 < u < 2π,

in such a way that the corresponding coordinate neighborhood contains M
as u = u0 . By hypothesis u → u0 when v → ∞. On the other hand, the
equation of the geodesic y in this coordinate system is given by (cf. Eq.
(6)), Example 5 from [Car16] and choose an orientation on γ such that
c > 0)

r
1 + 4v 2
Z Z
1 dv
u=c dv + const. > c + const.,
v v 2 − c2 d
since

1 + 4v 2
> 1.
v 2 − c2
It follows from the above inequality that as v → ∞, u increases beyond
any value, which contradicts the fact that γ approaches M asymptotically.
Therefore, γ intersects all the meridians and this completes the proof of the
assertion made in part (d).
Exercise 105 (Exercise 3.7 (Geodisic frame) from [Car92].). Let M
be a Riemannian manifold of dimension n and let p ∈ M . Show that there
exists a neighborhood U ⊂ M of p and n vector fields E1 , ..., En ∈ χ(U ),
orthonormal at each point of U , such that, at p, ∇Ei Ej (p) = 0.
Such a family Ei , i = 1, ..., n, of vector fields is called a (local) geodesic
frame at p.
128

Solution. Let U = Br (p) ⊂ M n be a normal neighborhood. For each

q ∈ U , there is a nomalized geodesic γq joint p and q. Let {v1 , v2 , . . . , vn }
be an orthogonal basis of Tp M and let {V1 , V2 , . . . , Vn } be their respective
parallel transpots along γq . For each j = 1, n, define the field Ej by

Ej (q) = Vf (d(p, q))

where d is the Riemann distance. We have Ej is a field C ∞ , because

curves γq vary C ∞ with q in the sense that EDO’s geodesics γq have their
coefficients depending on C ∞ of q.
Now consider σi (s) be the normalized geodesics such that σi (0) = p and
0
σi (0) = vi = Vi (0) = Ei (p). We have

D(Ej ◦ σi )
∇Ei Ej (p) = ∇Ei (p) Ej = ∇σi0 (0) Ej = |s=0 .
ds
Since (Ej ◦ σi )(s) = Vj (d(p, σi (s))) = Vj (s) is a parallel field along γσi (s) =
σi |[0,s] , we have that

D(Ej ◦ σi )
∇Ei Ej (p) = |s=0
ds
DVj
= (0)
ds
= 0.

Exercise 106 (Exercise 3.8 from [Car92].). Let M be a Riemannian

manifold. Let X ∈ χ(M ) and f ∈ D(M ). Define the divergence of X as
a function divX:M → Rgiven by divX(p) = trace of the linear mapping
Y (p) → ∇Y X(p), p ∈ M,and the gradient of f as a vector field grad f on
M defined by

hgradf (p), vi = dfp (v), p ∈ M, v ∈ Tp M.

(a) Let Ei , i = 1, ..., n = dim M , be a geodesic frame at p ∈ M (see
Exercise Exercise 3.7). Show that:

Xn
gradf (p) = (Ei (f ))Ei (p),
i=1
 129

X n X
divX(p) = Ei (fi )(p), where X = fi Ei .
i=1 i
∂
(b) Suppose that M = Rn , with coordinates (x1 , ..., xn ) and ∂xi =
(0, ..., 1, ..., 0) = ei . show that:
n
X ∂f
gradf = ei ,
∂xi
i=1

X ∂fi X
divX = , where X = fi ei .
∂xi
i i

Solution. Define f (Y ) = ∇Y (X). We have

f (Ei ) = ∇Ei X
X X
= (Ei (fi ) + Ei fj Γkij )Ek . (1)
k i,j

But
X
Γkji Ek = ∇Ej Ei
k
=0
since {E1 , E2 , . . . , En } is a local geodesic frame at p. Since E1 , E2 , . . . , En
are linear independent, so Γkji = 0, ∀k. This implies that Γkij = 0, ∀k, i, j ∈
{1, 2, . . . , n}. Therefore the equation (1) becomes
X
f (Ei ) = Ei (fk )Ek .
k
The trace of the mapping f is

n
X
trace(f ) = Ei (fi ).
i=1
Thus
130

n
X
divX(p) = trace(f ) == Ei (fi )(p),
i=1
P
where X = i fi Ei .
Represent gradf (p) on the basis {E1 , . . . , En } by

n
X
gradf (p) = αi Ei (p).
i=1
Since {E1 , . . . , En } is orthonormal, so

hgradf (p), Ei i = αi .
Thus

n
X
gradf (p) = hgradf (p), Ei i Ei
i=1
Xn
= (Ei (f ))Ei (p).
i=1
∂f
(b) In the case M = Rn , the Ei = ei and Ei (f ) = hgradf, ei i = ∂xi .
Thus

n
X ∂f
gradf = ei
∂xi
i=1
and

n
X
divf = (Ei (f ))
i=1
n
X ∂f
= .
∂xi
i=1

Exercise 107 (Exercise 3.10 from [Car92].). Let f : [0, 1] × [0, a] → M

be a parametrized surface such that for all to ∈ [0, a], at the point f (0, to ),
s ∈ [0, 1], is a geodesic parametrized by arc length, which is orthogonal to
 131

the curve t → f (0, t), t ∈ [0, a], at the point f (0, to ). Prove that, for all
(so , to ) ∈ [0, 1] × [0, a], the curves s → f (s, to ), t → f (so , t) are orthogonal
Hint: Differentiate h ∂f ∂f
∂s , ∂t i with respect to s, obtaining

d ∂f ∂f D ∂f ∂f ∂f D ∂f
h , i=h , i+h , i
ds ∂s ∂t ds ∂s ∂t ∂s ∂t ∂s
1 d ∂f ∂f
= h , i = 0,
2 dt ∂s ∂s
D ∂f
where we used the symmetry of the onnection and the fact that ds ∂s = 0.
D E
Solution. Differentiate ∂f ,
∂s ∂t
∂f
with respect to s, obtaining
     
d ∂f ∂f D ∂f ∂f ∂f D ∂f
, = , + , by the symmetry of the connection
ds ∂s ∂t ds ∂s ∂t ∂s dt ∂s
 
∂f D ∂f D ∂d
= , since =0
∂s dt ∂s ds ∂s
We have
   
d ∂f ∂f D ∂f ∂f
, =2 , since the inner product is symmetry.
dt ∂s ∂t dt ∂s ∂s
Thus
   
∂f D ∂f 1d ∂f ∂f
, = , .
∂s dt ∂s 2 dt ∂s ∂t
Hence
   
d ∂f ∂f 1d ∂f ∂f
, = ,
ds ∂s ∂t 2 dt ∂s ∂s
= 0.
Therefore
 
∂f ∂f
, = const
∂s ∂t
∀s ∈ [0, 1] and t ∈ [0, a]. In particular, we have
132

   
∂f ∂f ∂f ∂f
(s0 ), (t0 ) = (0), (t0 ) = 0.
∂s ∂t ∂s ∂t
Thus the curves s → f (s, t0 ), t → s(s0 , t) are orthogonal.
Exercise 108 (Exercise 4.4 from [Car92].). Let M be a Riemannian
manifold with the following property: given any two points p, q ∈ M , the
parallel transport from p to q does not depend on the curve that joins p
to q. Prove that the curvature of M is identically zero, that is, for all
X, Y, Z ∈ ℵ(M ),R(X, Y )Z = 0.
Hint: Consider a parametrized surface f : U ⊂ R2 → M , where

U = {(s, t) ∈ R2 ; −ε < t < 1 + ε, −ε < s < 1 + ε, ε > 0}

and f (s, 0) = f (0, 0), for all s. Let Vo ∈ Tf (0,0) (M ) and define a field V
along f by: V (s, 0) = Vo and, if t 6= 0, V (s, t) is the parallel transport of
Vo along the curve t → f (s, t). Then, from Lemma 4.1,
DD DD ∂f ∂f
V =0= V + R( )V.
∂s ∂t ∂t ∂s ∂t ∂s
Since parallel transport does not depend on the curve chosen V (s, 1) is the
D
parallel transport of V (0, 1) along the curve s → f (s, 1), hence ∂s V (s, 1) =
0. Thus,
∂f ∂f
(0, 1),
Rf (0,1) ( (0, 1))V (0, 1) = 0.
∂t ∂s
Use the arbitrariness of f and Vo to conclude what is required.
Solution. Consider a parametrized surface f : U ⊂ R2 → M , where

U = {(s, t) ∈ R2 ; − < t < 1 + , − < s < 1 + ,  > 0}

and f (s, 0) = f (0, 0), for all s. Let V0 ∈ Tf (0,0) (M ) abd define a field V
along f by: V (s, 0) = V0 and, if t 6= 0, V (s, t) is the parallel tranport of V0
along the curve t → f (s, t). Then, from Lemma 4.1,

DD DD ∂f ∂f
V =0= V + R( , )V.
∂s ∂t ∂t ∂s ∂t ∂s
Since parallel transport does not depend on the curve chosen, V (s, 1) is the
D
parallel transport of V (0, 1) along the curve s → f (s, 1), hence ∂s V (s, 1) =
0. Thus
 133

∂f ∂f
Rf (0,1) (
(0, 1), (0, 1))V (0, 1) = 0.
∂t ∂s
By the arbitrariness of f and V0 , and by the existence and uniqueness de-
pendent on initial conditions theorem of the ordinary differential equations,
we conclude that R(X, Y )Z = 0.
Exercise 109 (Exercise 4.7 from [Car92].). Prove the 2nd Bianchi
Identity:

∇R(X, Y, Z, W, T ) + ∇R(X, Y, W, T, Z) + ∇R(X, Y, T, Z, W ) = 0

for all X, Y, Z, W, T ∈ ℵ(M ).
Hint: Since the objects involved are all tensors, it suffices to prove the
equality at a point p ∈ M . Choose a geodesic frame {ei } based at p (See
Exercise 7 of Chap. 3). In the frame ∇ei ej (p) = 0, hence

∇R(ei , ej , ek , el, eh ) = eh hR(R(ei ej ))ek , el i = eh hR(ek , el )ei , ej i

= h∇eh ∇el ∇ek ei − ∇eh ∇ek ∇el ei + ∇eh ∇[ek ,el ] ei , ej i.
Therefore, using the Jacobi identity for the bracket, we find

∇R(ei , ej , ek , el , eh ) + ∇R(ei , ej , el , eh , ek )
+∇R(ei , ej , eh .ek , el ) = R(el , eh , ∇ek ei , ej )
+R(eh , ek , ∇el ei , ej ) + R(ek , el , ∇eh ei , ej ) = 0,
since each one of the summands vanishes at p. The general case follows by
linearity.
Solution. Choose a geodesic frame {ei } based at p. By Corollary 3.3
in Chapter 2 from [Car92],

eh hR(ek , el )ei , ej i = h∇eh R(ek , el )ei , ej i + hR(ek , el )ei , ∇eh ej i .

For the geodesic frame then ∇eh ej (p) = 0, so

eh hR(ek , el )ei , ej i = h∇eh R(ek , el )ei , ej i . (1)

Since ∇eh ej (p) = 0, and R is a mutilinear mapping, so at p, we have
134

∇R(ei , ej , ek , el , eh ) = eh R(ei , ej , ek , el ) − R(∇eh ei , . . . el ) − · · · − R(ei , ej , . . . , ∇eh el )

= eh R(ei , ej , ek , el )
= eh hR(ei , ej )ek , el i . (2)
By Proposition 2.5 in Chapter 4 from [Car92], then

hR(ei , ej )ek , el i = (ei , ej , ek , el )

= (ek , el , ei , ej )
= hR(ek , el )ei , ej i . (3)
It follows from (1), (2) and (3) that

∇R(ei , ej , ek , el , eh ) = eh hR(ek , el )ei , ej i

= ∇eh ∇el ∇ek ei − ∇eh ∇ek ∇el ei + ∇eh ∇[ek ,el ] ei , ej . (4)
Similarly,

∇R(ei , ej , el , eh , ek ) = ∇ek ∇eh ∇el ei − ∇ek ∇el ∇eh ei + ∇ek ∇[el ,eh ] ei , ej , (5)
∇R(ei , ej , eh , ek , el ) = ∇el ∇ek ∇eh ei − ∇el ∇eh ∇ek ei + ∇el ∇[eh ,ek ] ei , ej . (6)
Add (4), (5) and (6) to obtain

∇R(ei , ej , ek , el , eh ) + ∇R(ei , ej , el , eh , ek )
+∇R(ei , ej , eh , ek , el ) = ∇eh ∇el ∇ek ei − ∇eh ∇ek ∇el ei + ∇eh ∇[ek ,el ] ei
+ ∇ek ∇eh ∇el ei − ∇ek ∇el ∇eh ei + ∇ek ∇[el ,eh ] ei
+∇el ∇ek ∇eh ei − ∇el ∇eh ∇ek ei + ∇el ∇[eh ,ek ] ei , ej . (7)

But

R(el , eh , ∇ek ei , ej ) = hR(el , eh )∇ek ei , ej i

= ∇eh ∇el ∇ek ei − ∇el ∇eh ∇ek ei + ∇[el ,eh ] ∇ek ei , ej
= ∇eh ∇el ∇ek ei − ∇el ∇eh ∇ek ei + ∇[el ,eh ] ∇ek ei , ej . (8)
Similarly,

R(eh , ek , ∇el ei , ej ) = ∇ek ∇eh ∇el ei − ∇eh ∇ek ∇el ei + ∇[eh ,ek ] ∇el ei , ej , (9)
R(ek , el , ∇eh ei , ej ) = ∇el ∇ek ∇eh ei − ∇ek ∇el ∇eh ei + ∇[ek ,el ] ∇eh ei , ej . (10)
We have
 135

∇ek ∇[el ,eh ] − ∇[el ,eh ] ∇ek = [ek , [el , eh ]], (11)
∇el ∇[eh ,ek ] − ∇[eh ,ek ] ∇el = [el , [eh , ek ]], (12)
∇eh ∇[ek ,el ] − ∇[ek ,el ] ∇eh = [eh , [ek , el ]]. (13)
Add (11), (12) and (13) to obtain

∇ek ∇[el ,eh ] + ∇el ∇[eh ,ek ] + ∇eh ∇[ek ,el ]

−(∇[el ,eh ] ∇ek + ∇[eh ,ek ] ∇el + ∇[ek ,el ] ∇eh ) = [ek , [el , eh ]] + [el , [eh , ek ]] + [eh , [ek , el ]]
= 0,
by the Jacobi identity. Therefore,

∇ek ∇[el ,eh ] + ∇el ∇[eh ,ek ] + ∇eh ∇[ek ,el ] = ∇[el ,eh ] ∇ek + ∇[eh ,ek ] ∇el + ∇[ek ,el ] ∇eh . (14)

It follows from (7), (8), (9), (10) and (14) that

∇R(ei , ej , ek , el , eh ) + ∇R(ei , ej , el , eh , ek )
+∇R(ei , ej , eh , ek , el ) = R(el , eh , ∇ek ei , ej ) + R(eh , ek , ∇el ei , ej )
+ R(ek , el , ∇eh ei , ej ).

Since each one of the summands vanishes at p (∇ek ei (p) = ∇el ei (p) =
∇eh ei (p) = 0), so

∇R(ei , ej , ek , el , eh ) + ∇R(ei , ej , el , eh , ek ) + ∇R(ei , ej , eh , ek , el ) = 0.

The general case follows from by linearity.
Exercise 110 (Exercise 4.8 from [Car92].). (Schur’s Theorem). Let
M n be a connected Riemannian man-ifold with n ≥ 3. Suppose that M is
isotropic, that is, for each p ∈ M,the sectional curvature K(p, σ) does not
depend on σ ⊂ Tp M . Prove that M has constant sectional curvature, that
is, K(p, σ) also does not depend on p.
Hint: Define a tensor R0 of order 4 by

R0 (W, Z, X, Y ) = hW, XihZ, Y i − hZ, XihW, Y i.

If K(p, σ) = K does not depend on σ, by Lemma 3.4, R = KR0 . Therefore,
for all U ∈ ℵ(M ), ∇U R = (U K)R0 . Using the 2nd Bianchi identity (see
Exercise 7):
136

∇R(W, Z, X, Y, U ) + ∇R(W, Z, Y, U, X) + ∇R(W, Z, U, X, Y ) = 0,

we obtain, for all X, Y, W, Z, U ∈ ℵ(M ),

0 = (U K)(hW, XihZ, Y i − hZ, XihW, Y i) + (XK)(hW, Y ihZ, U i − hZ, Y ihW, U i)

+ (Y K)(hW, U ihZ, Xi − hZ, U ihW, Xi).

Fix p ∈ M . Because n ≥ 3, it is possible, fixing X at p, to choose Y and Z

at p such that hX, Y i = hY, Zi = hZ, Xi = 0, hZ, Zi = 1. Put U = Z at p.
the relation above yields, for all W ,

h(X, K)Y − (Y K)X, W i = 0.

Since X and Y are linearly independent at p, we conclude that XK = 0
for alll X ∈ Tp M . Thus K = const.
Solution. As in Lemma 3.4 of Chapter 4 from [Car92], we define a
tensor R0 of order 4 by

R0 (W, Z, X, Y ) = hW, Xi hZ, Y i − hZ, Xi hW, Y i .

If K(p, σ) = Kp does not depend on σ, by Lemma 3.4, R = Kp R0 . There-
fore, for all U ∈ X(M ),

∇U R = ∇U Kp · R0
= Kp ∇U R0 + (U K)R0
= (U K)R0 .
Using the 2nd Bianchi identity (see Exercise 7):

∇R(W, Z, X, Y, U ) + ∇R(W, Z, Y, U, X) + ∇R(W, Z, U, X, Y ) = 0,

we obtain, for all X, Y, W, Z, U ∈ X(M ),

(U KP )(hW, Xi hZ, Y i − hZ, Xi hW, Y i)

+(XKp )(hW, Y i hZ, U i − hZ, Y i hW, U i)
+(Y Kp )(hW, U i hZ, Xi − hZ, U i hW, Xi) = 0.
 137

Fix p ∈ M . Because n ≥ 3, we can fix X at p, to choose Y and Z at p

such that hX, Y i = hY, Zi = hZ, Xi = 0, hZ, Zi = 1. Put U = Z at p. The
relation above yields, for all W ,

(XKp )(hW, Y i) − (Y Kp )(hW, Xi)

= (XKp )(hY, W i) − (Y Kp )(hX, W i) since the inner product is symmetric
= h(XKp )Y − (Y Kp )X, W i
= 0.

Thus

h(XKp )Y − (Y Kp )X, W i = 0.
Since W is arbitrary, so take W = (XKp )Y − (Y Kp )X to obtain

h(XKp )Y − (Y Kp )X, (XKp )Y − (Y Kp )Xi = 0.

This implies that

(XKp )Y − (Y Kp )X = 0. (1)
Since X and Y are independent at p (taking X 6= 0, Y 6= 0 and hX, Y i = 0
above), so it follows from (1) that XKp = 0 for all X ∈ Tp M . Thus
Kp = const.
Exercise 111 (Exercise 4.9 from [Car92].). Prove that the scalar cur-
vature K(p) at p ∈ M is given by
Z
1
K(p) = Ricp (x)dS n−1 ,
ωn−1 S n−1
where ωn−1 is the area of the sphere S n−1 in Tp M and dS n−1 is the area
elements on S n−1 .
Hint: Use the following general argument on quadratic
P forms. Consider
an orthonormal basis e1 , ..., en in Tp M such that if x ni=1 xi ei ,
X
Ricp (x) = hi x2i , hi
real. Because | x |= 1, the vector (x1 , ..., xn ) = v is a unit normal vector
on S n−1 . Denoting V = (h1 x1 , ..., hn xn ), and using Stokes Theorem, we
obtain
138

Z Z
1 X 1
( hi x2i )dS n−1 = hV, vidS n−1
ωn−1 S n−1 ωn−1 sn−1
Z
1
= divV dB n ,
ωn−1 B n
where B n is the unit ball whose boundary is S n−1 = ∂B n . Noting that vol
B n /ωn = 1/n, we conclude that
Z P
1 1 n−1 hi
Ricp (x)dS = divV =
ωn−1 S n−1 n n
P
Ricp (ei )
= = K(p).
n
Solution. We know that a quadratic form can be tranformed to a
standard from (linear algebra). Therefore, P there exists an orthonomrla
basis e1 , e2 , . . . , en in Tp (M ) such that if x = ni=1 xi ei ,

n
X
Ricp (x) = λi x2i , λi real.
i=1

Because ||x|| = 1, the vector ν = (x1 , . . . , xn ) is a unit normal vector on

S n−1 . Denoting V = (λ1 x1 , . . . , λn xn ), and using Stokes Theorem, we
obtain

Z n Z
1 X 1
( λi x2i ) dS n−1 = hV, νi dS n−1
ωn−1 S n−1 ωn−1 S n−1
i=1
Z
1
= divV dB n ,
ωn−1 Bn

where B n is the unit

Pn ball Pnboundary is S
whose n−1 = ∂B n .
∂Vi
But divV = i=1 ∂xi = i=1 λi , Ricp (ei ) = λi , B n dB n = vol(B n ),
R
vol(Bn )
and ωn−1 = n1 , we obtain
Z Pn
1 n−1 1 λi
Ricp (x) dS = divV = i=1 since divV = const
ωn−1 S n−1 n n
Pn
Ricp (ei )
= i=1 = K(p).
n
 139

Exercise 112 (Exercise 5.1 from [Car92].). Let M be a Riemannian

manifold with sectional curvature identically zero. Show that, for every
p ∈ M , the mapping expp : Bε (0) ⊂ Tp M → Bε (p) is an isometry, where
Bε (p) is a normal ball at p.
Solution. The mapping expp : B (0) ⊂ Tp (M ) → B (p) is a diffeo-
morphism and applying Lemma 3.5 of Chapter 3 from [Car92], we have

(dexpp )v v, (dexpp )v w = hv, wi .

Thus the mapping expp is an isometry.
Exercise 113 (Exercise 5.2 from [Car92].). Let M be a Riemannian
manifold, γ : [0, 1] → M a geodesic, and J a Jacobi field along γ. Prove
that there exists a parametrized surface f (t, s), where f (t, 0) = γ(t) and
the curves t → f (t, s) are geodesics, such that J(t) = ∂f
∂s (t, 0).
Hint: Choose a curve h(s), s ∈ (−ε, ε) in M such that h(0) = γ(0),
h0 (0) = J(0). Along h choose a vector field W (s) with W (0) = γ 0 (0),
DW DJ ∂f
ds (0) = dt (0). Define f (s, t) = exph(s) tW (s) and verify that ∂s (0, 0) =
dh
ds (0) = J(0) and

D ∂F D ∂f DW DJ
(0, 0) = (0, 0) = (0) = (0).
dt ∂s ds ∂t ds dt
Solution. We choose a curve λ(s), s ∈ (−, ) in M such that λ(0) =
γ(0), λ0 (0) = J(0). Along λ choose a vector field W (s) with W (0) = γ 0 (0),
DW DJ
ds (0) = dt (0).
Define f (s, t) = expλ(s) tW (s). We have

∂f
(0, 0) = J(0)
∂s
= λ0 (0)
dλ
= (0),
ds
D ∂f D ∂f
(0, 0) = (0, 0) by Lemma 3.4
dt ∂s ds ∂t
DW
= (0)
ds
DJ
= (0).
dt
140

Thus

∂f
J(t) =
(t, 0) by Proposition 2.4.
∂s
Exercise 114 (Exercise 5.3 from [Car92].). Let M be a Riemannian
manifold with non-positive sectional curvature. Prove that, for all p, the
conjugate locus C(p) is empty.
Hint: Assume the existence of a non-trivial Jacobi field along the ge-
odesic γ : [0, a] → M , with γ(0) = p, J(0) = J(a) = 0. Use the Jacobi
d DJ
equation to show that dt h dt , Ji ≥ 0. Conclude that h DJ
dt , Ji ≡ 0. Since
d DJ 2
dt hJ, Ji = 2h dt , Ji ≡ 0, we have || J || = const. = 0, a contradiction

Solution. Asume the existence of a non-trivial Jacobi field along the

geodesic γ : [0, a] → M , with γ(0) = p, J(0) = J(a) = 0. We have

D2 J
     
d DJ DJ DJ
,J = ,J + ,
dt dt dt2 dt dt
 
DJ DJ
= −R(γ 0 (t)), J(t))γ 0 (t), J + , by the Jacobi equation.
dt dt
DJ DJ
Since non-position section curvature and dt , dt ≥ 0, so the equation
above implies
 
d DJ
,J ≥ 0.
dt dt
Define
 
DJ
g(t) = (t), J(t) .
dt
dg d DJ
Since dt = dt dt , J ≥ 0, so g(t) is increasing, therefore
 
DJ
0 = g(0) ≤ g(t) ≤ g(a) = (a), J(a) = 0.
dt
Thus

g(t) ≡ 0 for all t ∈ [0, a].

or
 141

 
DJ
,J ≡ 0.
dt
d
Since dt hJ, Ji = 2 DJ 2
dt , J = 0, so ||J|| = const = 0 (since J(0) = 0).
This contradicts to non-trivial Jacobi field.
Exercise 115 (Exercise 5.6 from [Car92].). Let M be a Riemann-
ian manifold of dimension two (in this case we say that M is a surface).
Let B∂ (p) be a normail ball around the point p ∈ M and consider the
parametrized surface

f (ρ, θ) = expp ρv(θ), 0 < ρ < δ, −π < θ < π,

where v(θ) is a circle of radius δ in Tp M parametrized by the central angle
θ.
(a) Show that (ρ, θ) are coordinates in an open set U ⊂ M formed by
the open ball Bδ (p) minus the ray expp (−ρv(0))0 < ρ < δ. Such coordinates
are called polar coordinates at p.
(b) Show that the coeffcients gij if the Riemannian metrix in these
coordinates are:

∂f 2 ∂f 2
g12 = 0, g11 =| | =| v(θ) |2 = 1, g22 ≈| | .
∂ρ ∂θ
(c) Show that, along the geodesic f (ρ, 0), we have
√
( g22 )ρρ = −K(p)ρ + R(ρ),
where lim R(ρ)
ρ = 0 and K(p) is the sectional curvature of M at p.
ρ→0
d) Prove that
√
( g22 )ρρ
lim √ = −K(p).
ρ→0 g22
This last expression is the value of the Gaussian curvature of M at p given
in polar coordinates (Cf., for example [Car76] p. 288). This fact from
the theory of surfaces, and (d) shows that, in dimension two, the sectional
curvature coincides with the Gaussian curvature. In the next chapter, we
shall give a more direct proof of this fact.
142

Solution. (a) Let expp (l) = L, where L is the ray expp (−ρv(θ)),
0 < ρ < δ. Let x : U \l → Bδ (p)\L be a system of geodesic polar coordinates
(ρ, θ). Every point in Bδ (ρ)\L is defined uniquely by a some pair (ρ, θ) and
converse since expp is an isomorphism. Thus (p, θ) are coordinates in an
open set U ⊂ M formed by the open ball Bδ (p) minus the ray expp (−ρv(θ)),
0 < ρ < δ. D E
(b) Since g12 = ∂x , ∂x
∂ρ ∂θ , so

∂ 2 x ∂x ∂x ∂ 2 x
   
(g12 )p = , + , .
∂ρ2 ∂θ ∂ρ ∂θ∂ρ
Since θ = const, is a geodesic, we have
 
∂x ∂ ∂x
(g12 )p = , ( )
∂ρ ∂θ ∂ρ
 
1 ∂ ∂x ∂x
= , = 0.
2 ∂θ ∂ρ ∂ρ
For each q ∈ Bδ (p), we denote α(σ) as the geodesic circle that passes
through q, where σ ∈ (−π, π). (Note that if q = p then α(σ) is the constant
curve α(σ) = p.) We denote γ(s), where s is the arc length of γ, as the
radical geodesic that passes through q. With this notation we may write
 
dα dγ
g12 (ρ, θ) = , .
dσ ds
The coefficient g12 (ρ, θ) does not defined at p. However, if we fix the radical
geodesic θ = const, the second number of the above equation is defined for
every point of this geodesic. Since at p, α(σ) = p, so dα
dσ = 0, we obtain

 
dα dγ
lim g12 = lim , = 0.
ρ→0 ρ→0 dσ ds
Since (g12 )p = 0, so g12 does not depend on p and the fact above implies
that g12 = 0.
By the definition of the exponential map, ρ measures the arc length
along the curve θ = const. This implies that
∂f 2
g11 = | | = |v(θ)|2 = 1
∂ρ
 143

and we also have

 
∂f ∂f ∂f 2
g22 = , =| | .
∂θ ∂θ ∂θ
(c) By Corollary 2.10 of Chapter 5 from [Car92], then

1 R(ρ)
|J(ρ)| = ρ − K(p)ρ3 + R(ρ), where lim 3 = 0.
6 ρ→0 ρ
√ √
Since g22 = | ∂f ∂θ | = |J(p)|, so g22 = ρ − 16 K(p)ρ3 + R(ρ). Taking
derivative twice with respect to ρ to obtain

√
( g12 )ρρ = −K(p)ρ + (R)ρρ .

Put R = (R)ρρ , then

√
( g22 )ρρ = −K(p)ρ + R(ρ). (1)
Using l’Hopital rule to obtain

R (R)ρρ
0 = lim 3
= lim .
ρ→0 ρ ρ→0 6ρ

Hence

(R)ρρ
lim = 0.
ρ→0 ρ
Thus

R(ρ)
lim = 0. (2)
ρ→0 ρ
Equations (1) and (2) satisfy the requires of the problem.
(d) Firstly, we shall prove that
√
g22
lim = 1.
ρ→0 ρ

Indeed,
144

√
g22 |J(p)|
lim = lim
ρ→0 ρ ρ→0 ρ
ρ − 16 K(p)ρ3 + R(ρ)
= lim =1
ρ→0 ρ
R(ρ)
since limρ→0 ρ = limρ→0 ρ2 R(ρ)
ρ3
= 0. Next, we have

√ √
( g22 )ρρ ( g22 )ρρ ρ
lim √ = lim · lim √
ρ→0 g22 ρ→0 ρ ρ→0 g22
√
( g22 )ρρ ρ
= lim since lim √ =1
ρ→0 ρ ρ→0 g22
−K(p)ρ + R(ρ)
= lim
ρ→0 ρ
= −K(p)
R(ρ) √
since limρ→0 ρ = 0 and g22 = −K(p)ρ + R(ρ).
Therefore

√
( g22 )ρρ
lim √ = −K(p).
ρ→0 g22
Exercise 116 (Exercise 5.7 from [Car92].). Let M be a Riemannian
manifold of dimension two. Let p ∈ M and let V ⊂ Tp M be a neighborhood
of the origin where expp is a diffeomorphism. Let Sr (0) ⊂ V be a circle
of radius r centered at the origin, and let Lr be the length of the curve
expp (Sr ) in M . Prove that the sectional curvatire at p ∈ M is given by

3 2πr − Lr
K(p) = lim .
r→0 π r3
Hint: Use Exercise 6.
Solution. (a) By Exercise 6, we have

√ 1
g22 = ρ − K(p)ρ3 + R(ρ).
6
We also have
 145

π−
√
Z
π 3
Lr = lim g22 dθ = 2πr − r K(ρ) + R1
→0 −π+ 3
R1
where limr→0 r3
= 0. This implies that

3 2πr − Lr + R1
K(p) = lim ( )
r→0 π r3
3 2πr − Lr R1
= lim · since lim = 0.
r→0 π r3 r→0 r 3

Exercise 117 (Exercise 6.1 from [Car92].). Let M1 and M2 be Rie-

mannian manifolds, and consider the product M1 × M2 , with the product
matric. Let ∇1 be the Riemannian connection of M1 and let ∇2 be the
Riemannian connection of M2
(a) Show that the Riemannian connection ∇ of M1 × M2 is given by
∇Y1 +Y2 (X1 + X2 ) = ∇1Y1 X1 + ∇2Y2 X2 , X1 , Y1 ∈ X(M1 ), X2 , Y2 ∈ X(M2 ).
(b) For every p ∈ M1 , the set (M2 )p = {(p, q) ∈ M1 × M2 ; q ∈ M2 } is a
submanifold of M1 × M2 , naturally diffeomorphic to M2 . Prove that (M2 )p
is a totally geodesic submanifold of M1 × M2 .
(c) Let σ(x, y) ⊂ T(p,q) (M1 × M2 ) be a plane such that x ∈ Tp M1 and
y ∈ Tq M2 . Show that K(σ) = 0.
Solution. (a) Let p ∈ M1 , q ∈ M2 . Consider the carmonical maps
M1 → M1 × {q} and M2 → {p} × M2 be immersions of M1 and M2 into
M1 × M2 . For X1 , Y1 ∈ M1 × {q}, X2 , Y2 ∈ {p} × M2 ,

∇1Y1 X1 = (∇Y X)tan M1 ,

where X, Y are extensions of X1 and Y1 . We get a similar formula for
∇2Y2 X2 . Define X, Y by

X(a, b) = X1 (a, q) + X2 (p, b),

Y (a, b) = Y1 (a, q) + Y2 (p, b),
for all (a, b) ∈ M1 × M2 . This gives

∇1Y1 X1 + ∇2Y2 X2 = (∇Y X)tan M1 + (∇Y X)tan M2

= ∇Y X,
146

as required.
(b) Since M2 is natural diffeomorphic to (M2 )p and by part (a), so

2
∇Y X = ∇2Y X for Y, X ∈ M2 ,
2
where ∇Y X = ∇Y X. Thus B(X, Y ) = 0 for all X, Y ∈ (M2 )p . Therefore
B(x, x) = 0 for all x ∈ (M2 )p . For every y ∈ M1 × M2 , y = (y1 , y2 ), where
y1 ∈ M1 , y2 ∈ M2 . So y ∈ (M2 )y1 . This implies that B(y, y) = 0 for all
y ∈ M1 × M2 . Hence (M2 )p is totally geodesic on M1 × M2 .
(c) Let (X1 , X2 , . . . , Xn ) and (Y1 , Y2 , . . . , Ym ) be basis in Tp M1 and
∂
Tq M2 , respectively, where Xi = ∂x i
, Yj = ∂y∂ j . Whence ((Xi , Yj )) is a
basis in T(p,q) (M1 × M2 ). Since ∇1Xi Xj = 0 and ∇2Yi Yj = 0 for all i, j,
so ∇(Xi ,Yj ) (Xk Yl ) = 0 and [(Xi , Yj ), (Xk , Yl )] = 0 for all i, j, k, l. Thus
(x, y, x, y) = 0 for all x, y ∈ σ (since (x, y, u, v) is multilinear). Therefore
K(σ) = 0.
Exercise 118 (Exercise 6.2 from [Car92].). Show that x : R2 → R4 given
by

1
x(θ, ϕ) = √ (cos θ, sin θ, cos ϕ, sin ϕ), (θ, ϕ) ∈ R2
2
is an immersion of R2 into the unit sphere S 3 (1) ⊂ R4 , whose image x(R2 )
is a torusT 2 with sectional curvature zero in the induced matric
Solution. We have

− √12 sin θ 0
 
1
 − √2 cos θ 0 
dx = 
 
1
0 − 2 sin θ
√

 
0 − √12 cos θ
which is certainly of rank 2, so x is an immersion. Moreover,

1
||x(θ, ϕ)|| = (cos2 θ + sin2 θ + cos2 ϕ + sin2 ϕ)
2
= 1,
so the image of x is contained in S 3 (1). Since (cos θ, sin θ) and (cos ϕ, sin ϕ)
parametrize the unit circle, so the image of x is S 1 × S 1 = T 2 .
 147

∂ ∂ √1 (− sin θX1 +cos θX2 ) and ∂ = √1 (− sin ϕ, X3 +

Define Xi = ∂xi , ∂θ = 2 ∂ϕ 2
∂ ∂
cos ϕ, X4 ). We note that K( ∂θ , ∂ϕ ) = 0 since all sectional curvature of R4
are zero. Hence, by Gauss Theorem,

∂ ∂ ∂ ∂ ∂ ∂
K( , ) = K( , ) − K( , )
∂θ ∂ϕ ∂θ ∂ϕ ∂θ ∂ϕ
 
∂ ∂ ∂ ∂ ∂ ∂ 2
= B( , ), B( , ) − |B( , )|
∂θ ∂θ ∂ϕ ∂ϕ ∂θ ∂ϕ
 
∂ ∂ ∂ ∂ ∂ ∂ 2
= ∇∂/∂θ − ∇∂/∂θ , ∇∂/∂ϕ − ∇∂/∂ϕ − |∇∂/∂θ − ∇∂/∂θ |
∂θ ∂θ ∂ϕ ∂ϕ ∂ϕ ∂ϕ
 
∂ ∂ ∂ 2
= ∇∂/∂θ , ∇∂/∂ϕ − |∇∂/∂θ |
∂θ ∂ϕ ∂ϕ

∂ ∂ ∂
since ∇∂/∂θ ∂θ = 0, ∇∂/∂ϕ ∂ϕ = 0 and ∇∂/∂θ ∂ϕ = 0 (because θ and ϕ give
2
coordinates on R ). Consider the vector field V where V (x1 , x2 , x3 , x4 ) =
√1 (−x2 , x1 , −x4 , x3 ). Then
2

∂
∇∂/∂θ = ∇V V
∂θ
1
= ∇−x2 X1 +x1 X2 (−x2 X1 + x1 X2 )
2
1
= [−x2 ∇X1 (−x2 X1 + x1 X2 ) + x1 ∇X2 (−x2 X1 + x1 X2 )]
2
1
= [−x2 X1 (x1 )X2 + x1 X2 (−x2 )X1 ]
2
1
= [−x1 X1 − x2 X2 ]
2

since all the other terms in the expression of the third line are zero. A
similar argument shows that

∂ 1
∇∂/∂ϕ = [−x3 X3 − x4 X4 ]
∂ϕ 2

∂
and ∇∂/∂θ ∂ϕ = 0. Therefore
148

 
∂ ∂ ∂ ∂ ∂ 2
K( , )= ∇∂/∂θ , ∇∂/∂ϕ − |∇∂/∂θ |
∂θ ∂ϕ ∂θ ∂ϕ ∂ϕ
1
= (−x1 X1 − x2 X2 − x3 X3 − x4 X4 ) − 0
4
= 0.

Exercise 119 (Exercise 6.3 from [Car92].). Let M be a Riemannian

manifold and let N ⊂ K ⊂ M be submanifolds of M . Suppose that N is
totally geodesic in K and that K is totally geodesic in M . Prove that N
is totally geodesic in M .

Solution. For every γ is geodesic on N , then γ is also geodesic on K

because N is totally geodesic in K. Since K is totally geodesic in M , so
γ is also geodesic on M . Thus every geodesic on K is also geodesic on M .
Therefore N is totally geodesic on M .

Exercise 120 (Exercise 6.6 from [Car92].). Let G be a Lie group with
a bi-invariant metric. Let H be a Lie group and let h : H → G be an
immersion that is also a homomorphism of groups (that is, H is a Lie
subgroup of G). Show that h is a totally geodesic immersion.

Solution. Since the metric H inherits from G is bi-invariant on H, we

know that the geodesics through the identity in H are one-parameter sub-
groups of H. But a one-parameter subgroup of H is also a one-parameter
subgroup of G. Since the one-parameter subgroups of G are just geodesics
through the identiy, this implies that the geodesics through the identity in
H are geodesics in G, which demonstrates that H is a geodesics immersion
at the identity. Since lift-translation is an isometry of H and istrometries
take geodesics to geodesics, this in turn means that H is a geodesic immer-
sion at every p ∈ H, so H is totally geodesic immerson.
The metric inherited from G on H is simply

hu, viH = hdhg (u), dhg (v)iG

where u, v ∈ Tg H. We shall prove that the metric H inherited from G is

bi-invariant on H. Suppose u, v ∈ Tg H. Then
 149

hu, viH = hdhg (u), dhg (v)iG

= (dLh(g0 ) )g ◦ (dhg )u, (dLh(g0 ) )g ◦ (dhg )v G
= d(Lh(g0 ) ◦ h)g u, d(Lh(g0 ) ◦ h)g v G
−1
= d(h ◦ h | ◦ Lh(g0 ) ◦ h)g u, d(h ◦ h−1 |g0 ◦ Lh(g0 ) ◦ h)g v
g0 G
= d(h ◦ Lg0 )g u, d(h ◦ Lg0 )g v G
= (dh) ◦ (dL )g u, (dh) ◦ (dLg0 )g v
g0 g0 g0 G
= (dLg0 )g u, (dLg0 )g v H
.
A similar calculation shows that this inherited metric is righ-invariant, so
it is a bi-invariant metric.
Exercise 121 (Exercise 6.8 (The Clifford torus) from [Car92].). . Con-
sider the immersion x : R2 → R4 given in Exercise 2.
(a) Show that the vectors

e1 = (− sin θ, cos θ, 0, 0), e2 = (0, 0, − sin ϕ, cos ϕ)

form an orthonormal basis of the tangent space, and that the vectors n1 =
√1 (cos θ, sin θ, cos ϕ, sin ϕ), n2 = √1 (− cos θ, − sin θ, cos ϕ, sin ϕ) form an
2 2
orthonor-mal basis of the normal space.
b) Use the fact that

hSnk (ei ), ej i = −h∇ei nk , ej i = h∇ei ej , nk i,

where ∇ is the covariant derivative (that os, the usual derivative) of R4 ,
and i, j, k = 1, 2, to establish that the matrices of Sn1 and Sn2 with respect
to the basic {e1 , e2 } are
 
−1 0
Sn1 =
0 −1
 
1 0
Sn2 =
0 −1
(c) From Exercise 2, x is an immersion of the torus T 2 into S 3 (1) (the
Clifford torus). Show that x is a minimal immersion.
150

Solution. (a) By the solution of Exercise 3, we have rank(dx) = 2.

So the dimension of the tangent space are 2. We have

he1 , e2 i = − sin θ · 0 + cos θ · 0 + 0 · (− sin θ) + 0 cos θ = 0,

he1 , e1 i = (− sin θ)2 · 0 + (cos θ)2 = 1,
he2 , e2 i = 02 + 02 + (− sin θ)2 + (cos θ)2 = 1.
Therefore e1 and e2 form an orthogonal basis of the tangent space. More-
over, for all x in the tangent space, x = α1 e1 +α2 e2 where α1 and α2 belong
to R and

hn1 , xi = α1 hn1 , e1 i + α2 hn1 , e2 i

α1 α2
= √ (− sin θ cos θ + cos θ sin θ) + √ (− sin ϕ cos ϕ + cos ϕ sin ϕ),
2 2
hn2 , xi = α1 hn2 , e1 i + α2 hn2 , e2 i
α1 α2
= √ (sin θ cos θ − cos θ sin θ) + √ (− sin ϕ cos ϕ + cos ϕ sin ϕ) = 0.
2 2
So n1 and n2 are orthogonal to the tangent space, that is, n1 and n2 belong
to the orthogonal complement of the tangent space, and

1
hn1 , n2 i = (− cos2 θ − sin2 θ + cos2 ϕ + sin2 ϕ) = 0,
2
1
hn1 , n1 i = (cos2 θ + sin2 θ + cos2 ϕ + sin2 ϕ) = 1,
2
1
hn2 , n2 i = (cos2 θ + sin2 θ + cos2 ϕ + sin2 ϕ) = 1.
2
Therefore n1 and n2 form an orthogonal basis of the normal space.
(b) We have

√ 1 √ ∂
e1 = (− sin θ, cos θ, 0, 0) = 2 + √ (− sin θ, cos θ, 0, 0) = 2 .
2 ∂θ
Similarly,

√ ∂
e2 = 2 .
∂ϕ
Since
 151

hSnk (ei ), ej i = − ∇ei nk , ej = ∇ei ej , nk ,

so

√ ∂
 
hSn1 (e1 ), e1 i = ∇e1 e1 , n1 = ∇√2 ∂ 2 , n1
∂θ ∂θ
 
1
= − cos θX1 − sin θX2 , √ (cos θX1 + sin θX2 + cos ϕX3 + sin ϕX4 )
2
1 2 2 −1
= √ (− cos θ − sin θ) = √ ,
2 2
and similarly,

√ ∂
 
hSn1 (e1 ), e2 i = ∇√2 ∂ 2 , n1
∂θ ∂ϕ
√ ∂
=0 since ∇√2 ∂ 2 = (0, 0, 0, 0)
∂θ ∂ϕ
and

hSn1 (e2 ), e1 i = 0,
and

√ ∂
 
hSn1 (e2 ), e2 i =∇√2 ∂ 2 , n1
∂ϕ ∂ϕ
 
1
= − cos ϕX3 − sin ϕX4 , √ (cos θX1 + sin θX2 + cos ϕX3 + sin ϕX4 )
2
1 2 2 −1
= √ (− cos ϕ − sin ϕ) = √ .
2 2
Moreover,

√ ∂
 
hSn2 (e1 ), e1 i = ∇e1 e1 , n2 = ∇√2 ∂ 2 , n2
∂θ ∂θ
 
1
= − cos θX1 − sin θX2 , √ (− cos θX1 − sin θX2 + cos ϕX3 + sin ϕX4 )
2
1 2 2 −1
= √ (cos θ sin θ) = √ ,
2 2
and similarly,
152

√ ∂
 
hSn2 (e1 ), e2 i = ∇√2 ∂ 2 , n2
∂θ ∂ϕ
√ ∂
=0 since ∇√2 ∂ 2 = (0, 0, 0, 0)
∂θ ∂ϕ
and

hSn2 (e2 ), e1 i = 0,
and

√ ∂
 
hSn2 (e2 ), e2 i = ∇√ ∂
2 ∂ϕ 2 , n2
∂ϕ
 
1
= − cos ϕX3 − sin ϕX4 , √ (− cos θX1 − sin θX2 + cos ϕX3 + sin ϕX4 )
2
1 2 2 −1
= √ (− cos ϕ − sin ϕ) = √ .
2 2
The metrices of Sn1 and Sn2 with respect to the basis {e1 , e2 } are
 
1 −1 0
Sn1 = √
2 0 −1
 
1 1 0
Sn2 = √
2 0 −1
(c) Since

2
1 1
√ (cos θ, sin θ, cos ϕ, sin ϕ) = (cos2 θ + sin2 θ + cos2 ϕ + sin2 ϕ) = 1,
2 2
so n1 is thought of as a point which lies on the sphere S 3 (1), so n1 is thought
of as a vector which is orthogonal to S 3 (1). Therefore, the subspace of Tp S 3
is normal to Tp T 2 , where T 2 is the image of x spanned by n2 . Therefore
the condition of minimal that trace Sn2 = 0, which  is clearly
 does, given
1 0
the computation in part (b) above (Sn2 = √12 ). Thus x is a
0 −1
minimal immersion.
n+k
Exercise 122 (Exercise 6.10 from [Car92].). Let f : M n → M be
an isometric immersion and let Sη : T M → T M be the operator associated
to the second fundamental form of f along the normal field η. Consider
 153

Sη as a tensor of order 2 given by Sη (X, Y ) = hSη (X), Y i, X, Y ∈ X(M ).

Observe that saying the operator Sη is selfadjoint is equivalent to saying
that the tensor Sη os symmetric, that is, Sη (X, Y ) = Sη (Y, X). Prove that
for all V ∈ X(M ), the tensor ∇V Sη is symmetric.
Hint: Differentiating hSη X, Y i = hX, Sη Y i with respect to V , we obtain

h∇V (Sη X), Y ) + hSη X, ∇V Y i = h∇V X, Sη Y i + hX, ∇V (Sη Y )i.

Using the fact that

h(∇V Sη )X), Y i = h∇V (Sη X), Y i − hSη (∇V X), Y i

and the previous expression, we obtain easily that

h(∇V Sη )S, Y i = hX, (∇V Sη )Y i.

Solution. Differentiating hSη X, Y i = hX, Sη Y i with respect to V to
obtain

h∇V (Sη X), Y i + hSη X, ∇V Y i = h∇V (X), Sη Y i + hX, ∇V (Sη Y )i . (1)

We know that

h(∇V Sη )X, Y i = h∇V (Sη X), Y i − hSη (∇V X), Y i . (2)

Combining (1) and (2) to obtain

h(∇V Sη )X, Y i
= h∇V (Sη X), Y i − hSη (∇V X), Y i by (2)
= h∇V X, Sη Y i + hX, Sη (∇V Y )i by (1)
− hSη X, ∇V Y i − hSη (∇V X), Y i
since h∇V X, Sη Y i = hSη (∇V X), Y i
= hX, ∇V (Sη Y )i − hSη X, ∇V Y i
by the operator Sη is self-adjoint
= hX, ∇V (Sη Y )i − hX, Sη (∇V Y )i since Sη is self-adjoint
= hX, (∇V Sη Y )i by (2).

Thus

h(∇V Sη )X, Y i = hX, (∇V Sη )Y i .

Therefore for all V ∈ X(M ), the tensor ∇V Sη is symmetric.
154

n+1
Exercise 123 (Exercise 6.11 from [Car92].). Let f : M → R be a
differentiable function. Define the Hessian, Hess f of f at p ∈ M as the
linear operator
Hess f : Tp M → Tp M , (Hess f )Y = ∇Y grad f , Y ∈ Tp M , where ∇
is the Riemannian connection of M . Let a be a regular value of f and let
n+1
Mn ⊂ M be the hypersuperface in M defined by M = {p ∈ M ; f (p) =
a}. Prove that:
(a) The Laplacian ∆f is given by

∆f = trace Hess f.
(b) If X, Y ∈ X(M ),then

h(Hessf )Y, Xi = hY, (Hessf )Xi.

Conclude that Hess f is self-adjoint, hence determines a symmetric bilinear
form on Tp M , p ∈ M , given by (Hessf )(X, Y ) = h(Hessf )X, Y ), X, Y ∈
Tp M .
(c) The mean curvature H of M ⊂ M is given by

gradf
nH = −div( ).
| gradf |
gradf
Hint: Take an orthonormal frame E1 , ..., En , En+1 = |gradf | = η in a neight-
borhood of p ∈ M in M and use the definition of divergence in Exercise 8,
Chapter 3, to obtain

Xn
nH = traceSη = hSη (Ei ), Ei i
i=1
n
X n+1
X
=− h∇Ei η, Ei i − h∇η η, ηi = h∇Ei η, Ei i
i=1 i=1
gradf
= −divM η = −div( ).
| gradf |
n+1
(d) Observe that every embedded hypersurface M n ⊂ M is locally
the inverse image of a regular value. Conclude from (c) that the mean
curvature H of such a hypersuperface is given by
 155

1
H = − divN,
n

where N is an appropriate local extension of the unit normal vector field

n+1
on M n ⊂ M .

Solution. (a) By defintion, for every Y ∈ Tp M ,

∆f (Y ) = div gradf (Y )
= trace∇Y gradf
= trace Hessf (Y )

since Y was arbitrary, so we have ∆f = Trace Hessf .

(b) Let p ∈ M and E1 , E2 , . . . , En+1 be a geodesic fram at the point p.
The, for i, j ∈ {1, 2, . . . , n + 1},

h(Hessf )Ei , Ej i = ∇Ei gradf , Ej

n+1
* +
X
= ∇Ei (Ek (f ))Ek , Ej
k=1
*n+1 +
X
= [(Ek (f ))∇Ei Ek + Ei (Ek (f ))Ek ], Ej
k=1
*n+1 +
X
= Ei (Ek (f ))Ek , Ej
k=1
= Ei (Ej (f )).

Thus

h(Hessf )Ei , Ej i = Ei (Ej (f )). (1)

On the other hand,

156

hEi , (Hessf )Ej i = Ei , ∇Ej gradf

n+1
* +
X
= Ei , ∇Ej (Ek (f ))Ek
k=1
n+1
* +
X
= Ei , [Ek (f )∇Ej Ek + Ej (Ek (f ))Ek ]
k=1
= Ej (Ei (f )).
Hence

hEi , (Hessf )Ej i = Ej (Ei (f )). (2)

But

0 = ∇Ei Ej − ∇Ej Ei
= [Ei , Ej ]
= Ei Ej − Ej Ei ,
so

Ei (Ej f ) = Ef (Ei f ).
Therefore, combining with (1) and (2) to obtain

h(Hessf )Ei , Ej i = hEi , (Hessf )Ej i .

Since this holds for all i, j ∈ {1, 2, . . . , n + 1}, we conclude that Hessf is
self-adjoint.
gradf
(c) Let p ∈ M ⊂ M . let E1 , E2 , . . . , En , En+1 = |gradf | = η be a an
orthogonal frame about p. We have

1
∇η η, η = η hη, ηi
2
1
= η(1)
2
= 0.
Thus by definition of divergent and of the mean curvature,
 157

n
X
nH = TraceSη = hSη (Ei ), Ei i
i=1
n
X n+1
X
=− (∇Ei η, η) − ∇η η, η = ∇Ei η, Ei
i=1 i=1
= −divM η
gradf
= −div( ).
|gradf |
(d) Let M be an embedded hyppersurface. Then, locally, M = f −1 (r)
for some regular value r. hence, for each point p ∈ M , gradf (p) 6= 0, so the
expression derived in (c) is well-defined on all of M . Hence H = − n1 divN
where N is an appropriate local extension of the unit normal vector field
n+1
on M n ⊂ M .
Exercise 124 (Exercise 6.12 (Singularities of a Killing field) from
[Car92].). Let X be a Killing vector field on a Riemannian manifold M .
Let N = {p ∈ M ; X(p) = 0}. Prove that:
(a) If p ∈ N , and V ⊂ M is a normal neighborhood of p, with q ∈
N ∩ V , then the radial geodesic segment γ joining p to q is contained in N .
Conclude that γ ∩ V ⊂ N .
(b) If p ∈ N, there exists a neighborhood V ⊂ M of p such that V ∩ N
is a submanifold of M (this implies that every connected component of N
is a submanifold of M ).
Hint: Proceed by induction, using (a). If p is isolated, noth_ing has
to be dome. In the contrary case, let V ⊂ M be a nor-mal neighborhood
of p such that there exists q1 ∈ V ∩ N and consider the radial geodesic γ1
joining p to q1 . If V ∩ N = γ1 , by (a), the proof is complete. Otherwise,
let q2 ∈ V ∩ N − {γ1 } and let γ2 be the radial geodesic joining p to q2 .
Let Q ⊂ Tp M be the subspace generated by the vectors exp−1 p (q1 ) and
exp−1
p (q2) and let N 2 = exp p (Q ∩ exp −1 (V )). Show that for all t ∈ R, the
p
restriction of the differential (dXt )p of the flow Xt : M → M , to Q, is the
identity; conclude now that N2 ⊂ V ∩ N . Proceed in this way until the
dimension of Tp M is exhausted.
(c) The codimension, as a submanifold of M , of a con-nected component
Nk of N is even. Assume the fol-lowing fact: if a sphere has a non-vanishing
differen-tiable vector field on it then its dimension must be odd (for a proof,
see Armstring, [Ar], p. 198).
158

Hint: Let Ep = (Tp Nk )⊥ and let V ⊂ M be a normal neigh-borhood of

p. Set Nk⊥ = expp (Ep ∩ exp−1
p (V )). Since, for all t, (dXt )p : Ep → Ep „ we
have that X is tangent to N k1 . On the other hand, X 6= 0 is tangent to the
⊥ with cinterp. From the theorem mentioned above,
geodesic spheres of NK
the dimension of such a phere os odd. Hence dim NK ⊥ = dim E is even.
p

Solution. (a) Let ϕt denote the flow of X. If p ∈ N and V ⊂ M

is a normal neighborhood of p, with q ∈ N ∩ V , then let γ be the radial
geodesic joining p to q. Since q ∈ N , ϕt (q) = q for all t, where the flow is
defined. Now, since ϕt is an isometry on V , ϕt maps geodesics to geodesics,
specifically, ϕt (γ) is a geodesic. Since ϕt fixes p and q, ϕt (γ) is a geodesic
passing through p and q in V ; by uniqueness of geodesics, then ϕt (γ) = γ.
Therefore, for any q 0 ∈ γ ∩ V , ϕt (q 0 ) = q 0 , so X(q 0 ) = 0. Thus, γ ∩ V ⊂ N .
(b) We shall prove this by induction. If p is an isolated point, then
{p} is a neighborhood of p and N ∩ {p} = {p} is a 0-submanifold of M .
Otherwise, let V ve a normal neighborhood of p such that there exists
q1 ∈ V ∩ N not equal to p. Let q1 be the radial geodesic from p to q1 . By
part (a), γ1 ∩ V ⊂ N . if V ∩ N = γ1 ∩ V , then we are done, since γ1 ∩ V is
a 1-submanifold of M .
If V ∩ N 6= γ1 ∩ V , then let q2 ∈ V ∩ N \{γ1 }. Let γ2 be the radial
geodesic joining p to q2 . By the same argument as in (a), γ2 ∩ V ⊂ N .
Now let Q ⊂ Tp M be the subspace generated by exp−1 −1
p (q1 ) and expp (q2 ),
−1
let N2 = expp (Q ∩ expp (V )). Suppose v ∈ N2 . Then v = a expp (q1 ) + −1

b exp−1
p (q2 ) for some a, b ∈ R. Hence

(dϕt )p (v) = (dϕt )p (a exp−1 −1

p (q1 ) + b expp (q2 ))
= a(dϕt )p (exp−1 −1
p (q1 )) + b(dϕt )p (expp (q2 ))
= a exp−1 −1
p (q1 ) + b expp (q2 )
= v.
Since q1 and q2 are fixed by ϕt . Therefore, we have (dϕt )p is the identity
on Q. Therefore, if q ∈ N2 ,

ϕt (q) = expp ((dϕt )p (exp−1

p (q)))
= expp (exp−1
p (q1 ))
= q,
so N2 ⊂ V ∩ N .
 159

If V ∩ N = N2 , then we are done, since N2 is a 2-submanifold of M .

Otherwise if V ∩ N 6= N2 , we pick q3 ∈ (V ∩ N )\N2 and iterating the
above procedure. At each stage, either this alogrithm terminates with Ni ,
an i-submanifold of M , or we proceed to the next stage. Since we have only
n-dimentions to work with, we see that this procedure must terminate, and
so V ∩ N is a submanifold of M .
(c) Let p ∈ Nk ⊂ N . Let Ep = (Tp Nk )⊥ and let V ⊂ M be a normal
neighborhood of p. Let Nk⊥ = expp (Ep ∩ exp−1 p (V )). Now, suppose v ∈
Ep . Then (dϕt )p (v) ∈ Tp Nk and so expp ((dϕt )p v) ∈ Nk ∩ V , meaning
that ϕt (expp ((dϕt )p v)) = expp ((dϕt )p v) = ϕt (expp (v)). However, since
expp v 6∈ Nk , it can never flow to a point in Nk . Thus we conclude that
(dϕt )p : Ep → Ep , so X is tangent to Nk⊥ . On the other hand, since p is
unique point on Nk⊥ , where X vanishes, we know that X 6= 0 is tangent to
the geodesic spheres of Nk⊥ centered at p. Thus X is a non-vanishing vector
field on the geodesic spheres of Nk⊥ , which are homeomorphic to plain odd
spheres. Therefore, by the given fact, these geodesics spheres must be odd-
dimensional, so Nk⊥ must be even-dimensional. Since dim Nk⊥ = codimNk ,
we conclude that Nk has even codimention in M .
Exercise 125 (Exercise 7.2 from [Car92].). Let M f be a covering space
of a Riemannian manifold M. Show that it is possible to give M f a Riemann-
ian structure such that the covering map π : M → M is a local isometry
f
(this matric is called the covering metric). Show that M f is complete in the
covering metric if and only if M is complere.
Solution. Let π : M f → M ve a covering map. Since π(M f) = M , so
for every me ∈ M , π(m)
f e belongs to M . Assume that {(Uα , xα )} is a maximal
differntiable structure on M . Whence (Uαm , ταm ) is a parametrization of
M at m. Since the differentiable structure is maximal, so Uαm concides
with the distinguished neighborhood Um of m such that π −1 (Um ) = Uα Vα
where the Vα ’s are pariwise disjoint open sets and π|Vα is a homeomorphism
of Vα onto Um . The family {(Vα , π −1 |Vα ◦ xα )} is a differentiable structure
on Mf. The inner product induced on M f defined by

hx, yiq = hdπq (x), dπq (y)iπ(q)

for q ∈ Mf, x, y ∈ Tq (Mf) which defined a Riemannian structure on M f.

Thus M f has a Riemannian structure. For the inner product above, π is a
local isometry. Since π is a local isometry, so M is a complete metric space
160

if and only if M
f is a complete metric space. Using the Hoft and Rinow
Theorem, Mf is complete if and only if M is complete.

Exercise 126 (Exercise 7.8 from [Car92].). Let M be a complete Rie-

mannian manifold, M a connected Riemannian manifold, and f : M → M
a differentiable mapping that is locally an isometry. Assume that any two
points of M can be joined by a unique geodesic of M . Prove that f is
injective and surjective (and, therefore, f is a global isometry).
Solution. Let p 6= q anf γ : I → M be a geodesic, γ(0) = p and
γ(1) = q. Cover γ(I) by open sets Uα where f |Uα : Uα → f (Uα ) is an
isometry, by caompactness, there is a finite covering of Ui ’s or equivalently,
there is a partition of I where 0 = t0 < t1 < · · · < tn = 1,

D d(f ◦ γ) D dγ
( |[ti ,ti+1 ] ) = df ( ( |[ti ,ti+1 ] ))
dt dt dt dt
= df (0) = 0.
D d(f ◦γ)
Therefore dt ( dt ) = 0 on I implying that f ◦ γ is a geodesic in M
joining f (p) to f (q). If f (p) = f (q), then f ◦ γ would be a closed geodesic,
contradicting the uniqueness assumption. Thus f is injective.
Uniqueness of geodesic joining any two points of M f implies that M is
complete, then expp : Tp M → M is surjective for any q ∈ M .
Let p ∈ M is fixed. There is an open neighborhood U ⊂ M containing p
such that f |U → f (U ) is an isometry. We have f (expp (v)) = expf (p) (dfp (v))
for all v ∈ Tp M . Let q ∈ N be arbitrary. There is q ∈ Tf (p) M such that
expf (p) (w) = q. Since dfp is an isometry of vector spaces, there is a u ∈ Tp M
such that dfp (u) = w. Hence f (expp (v)) = expf (p) (dfp (v)) = q. Thus f is
surjective.
Exercise 127 (Exercise 7.9 from [Car92].). Consider the upper half-
plane

R2+ = {(x, y) ∈ R2 ; y > 0}

with the Riemannian metric given by

1
g11 = 1, g12 = 0, g22 = .
y
Show that the length of the vertical segment
 161

x = 0,  ≤ y ≤ 1, with ε > 0,
tends to 2 as  → 0. Conclude from this that such a metric is not complete.
(Observe, nevertheless, that when y → 0 the length of vector, in this metric,
becomes arbitrarily large.)
Solution. Parametrize the vertical sugement by

γ(t) : [t, 1] → Rn+

t 7→ γ(t) = (0, t).
For a given Riemannian metric,

1
ds2 = dx2 + dy 2 .
2
So the length of the vertical segment is

Z 1
1
l= √ dt
t t
√ 1
= 2 t|t
√
= 2 − 2 .
Therefore,

√
lim l = lim (2 − 2 )
→0 →0
= 2.
Since when y → 0, the length of vectors in this metric becomes arbitrary
large, so this metric is not complete.
Exercise 128 (Exercise 7.10 from [Car92].). Prove that the upper half-
plane R2+ with the Lobatchevski metric:

1
g11 = g22 = , g12 = 0
y 20
is complete.
162

Solution. The metric given by

1
ds2 = (dx2 + dy 2 ).
y2
The mapping Ta (x, y) = (x +a, y) is an isometry. This implies we can move
any point to other point of the form (0, y). Next, the mapping Dλ (x, y) =
(λx, λy) is an isometry for λ > 0. Putting these together shows that the
hyperbolic plane is homogenous. Use T−x to move (x, y) to (x0 , y 0 ), move
(x, y) to (0, y), then use Dy0 /y to move (0, y) to (0, y 0 ), then use Tx0 to mocw
(0, y 0 ) to (x0 , y 0 ). Thus, the upper half-plane R2+ with the Lobatchevski
metric is homogeneous. By Exercise 12 below, it is complete.
Exercise 129 (Exercise 7.12 from [Car92].). A Riemannian manifold
is said to be homogeneous if given p, q ∈ M there exists an isometry of M
takes p into q. Prove that any homegeneous manifold is complete.
Solution. The homogeneity of the manifold implies that there is  > 0
such that expp is defined on B(0, ) ⊂ Tp (M ), ∀p ∈ M . Indeed, suppose
that for a given p ∈ M , the exponential map expp is defined on B(0, ) ⊂
Tp (M ). Then given q ∈ M and f ∈ Iso(M, g) with f (p) = q, the diagram

f
M → M
expp ↑ ↑ expp
B(0, ) ⊂ Tp (M ) → B(0, ) ⊂ Tp (M )
commutes by uniqueness of geodesics and the fact that isometries preserve
geodesic. Therefore, any geodesics defined on a closed interval [a, b] can be
extended to (a − , b − ). That is, (M, g) is geodesically complete. Using
Hoft-Rinow theorem to obtain M is complete.
 Bibliography

[Car76] M. P. D. Carmo, Differential Geometry of Curves and Surfaces, Prentice Hall,

New Jersey, 1976.
[Car92] M. P. D. Carmo, Riemannian Geometry, Birkhäuser, Boston, 1992.
[Car16] M. P. D. Carmo, Differential Geometry of Curves and Surfaces, Revised and
Updated Second Edition, Dover Publications, New York, 2016.
[Gho07] M. Ghomi, Lecture Notes on Differential Geometry, Georgia Tech, Atlanta,
2007.

163