Curves and Surfaces

Lecture Notes for Geometry 1

Henrik Schlichtkrull
Department of Mathematics
University of Copenhagen
i
ii
Preface
The topic of these notes is dierential geometry. Dierential geometry
is the study of geometrical objects using techniques of dierential calculus,
in particular dierentiation of functions. The objects that will be studied
here are curves and surfaces in two- and three-dimensional space, and they
are primarily studied by means of parametrization. The main properties
of these objects, which will be studied, are notions related to the shape.
We will study tangents of curves and tangent spaces of surfaces, and the
notion of curvature will be introduced. These notions are dened through
dierentiation of the parametrization, and they are related to rst and second
derivatives, respectively.
The notion of curvature is quite complicated for surfaces, and the study
of this notion will take up a large part of the notes. The culmination is a
famous theorem of Gauss, which shows that the so-called Gauss curvature of
a surface can be calculated directly from quantities which can be measured on
the surface itself, without any reference to the surrounding three dimensional
space. This theorem has played a profound role in the development of more
advanced dierential geometry, which was initiated by Riemann.
The theory developed in these notes originates from mathematicians of
the 18th and 19th centuries. Principal contributors were Euler (1707-1783),
Monge (1746-1818) and Gauss (1777-1855), but the topic has much deeper
roots, since it builds on the foundations laid by Euclid (325-265 BC).
In these notes a signicant emphasis is placed on the interplay between
intuitive geometry and exact mathematics. Ideas are explained by numerous
illustrations, but they are also given rigorous proofs. It is my hope that the
student of the text will perceive the importance of both viewpoints. The
notes are adapted to an intensive course which runs over 7 weeks, so that
each chapter corresponds approximately to one week of teaching.
The notes were written and used for the rst time in 2005. The present
version, intended for 2011, has been improved and corrected thanks to the
suggestions of many students. Undoubtedly there are places where further
revision would be desirable, and I will appreciate all comments and correc-
tions. The drawings are made with Anders Thorups program SPLINES,
downloadable from http://www.math.ku.dk/thorup/splines/.
Henrik Schlichtkrull
March 2011
iii
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . -2
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . -3
1. Parametrized curves and surfaces. . . . . . . . . . . . . . . 1
1.1 Curves . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Graphs . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Level sets . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 The implicit function theorem, two variables . . . . . . . . 7
1.6 The implicit function theorem, more variables . . . . . . . 11
1.7 The implicit function theorem, more equations . . . . . . . 12
1.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . 16
2. Tangents . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Regular curves and tangent lines . . . . . . . . . . . . . 19
2.2 The tangent line of a level set . . . . . . . . . . . . . . 21
2.3 The tangent plane of a regular surface . . . . . . . . . . 22
2.4 Curves on a surface . . . . . . . . . . . . . . . . . . . 25
2.5 Reparametrization of curves . . . . . . . . . . . . . . . 26
2.6 Reparametrization of surfaces . . . . . . . . . . . . . . 27
2.7 Invariance under reparametrization . . . . . . . . . . . . 28
2.8 The unit normal, orientation . . . . . . . . . . . . . . . 29
2.9 Regular curves as graphs . . . . . . . . . . . . . . . . 30
2.10 The inverse function theorem . . . . . . . . . . . . . . 31
2.11 Regular surfaces as graphs . . . . . . . . . . . . . . . . 33
2.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . 34
3. The rst fundamental form . . . . . . . . . . . . . . . . . 39
3.1 Arc length . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Lines as shortest curves . . . . . . . . . . . . . . . . . 41
3.3 Unit speed parametrization . . . . . . . . . . . . . . . 42
3.4 The rst fundamental form . . . . . . . . . . . . . . . 43
3.5 Introduction to areas and plane integrals . . . . . . . . . 46
3.6 Null sets . . . . . . . . . . . . . . . . . . . . . . . 55
3.7 Double integrals . . . . . . . . . . . . . . . . . . . . 50
3.8 Transformation of integrals . . . . . . . . . . . . . . . 52
3.9 Surface area . . . . . . . . . . . . . . . . . . . . . . 53
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . 55
4. Curvature . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1 Curvature of plane curves . . . . . . . . . . . . . . . . 59
4.2 Curvature of unit speed curves . . . . . . . . . . . . . . 61
4.3 The tangent angle . . . . . . . . . . . . . . . . . . . 63
4.4 Curvature of space curves . . . . . . . . . . . . . . . . 64
4.5 Torsion . . . . . . . . . . . . . . . . . . . . . . . . 66
iv
4.6 The osculating plane and the binormal vector . . . . . . . 67
4.7 The Frenet formulas . . . . . . . . . . . . . . . . . . 68
4.8 Curvature of curves on a surface . . . . . . . . . . . . . 69
4.9 Interpretation of normal curvature . . . . . . . . . . . . 72
4.10 Geodesics . . . . . . . . . . . . . . . . . . . . . . . 73
4.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . 74
5. The second fundamental form . . . . . . . . . . . . . . . . 77
5.1 The shape operator . . . . . . . . . . . . . . . . . . . 77
5.2 The second fundamental form . . . . . . . . . . . . . . 80
5.3 Coordinate expressions for the second fundamental form. . . 80
5.4 A formula for the shape operator . . . . . . . . . . . . . 82
5.5 Diagonalization of the second fundamental form . . . . . . 83
5.6 The graph of a quadratic form . . . . . . . . . . . . . . 85
5.7 The type of a surface . . . . . . . . . . . . . . . . . . 88
5.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . 89
6. Teorema egregium . . . . . . . . . . . . . . . . . . . . . 93
6.1 The Gaussian curvature . . . . . . . . . . . . . . . . . 93
6.2 Intrinsic geometry . . . . . . . . . . . . . . . . . . . 94
6.3 Christoel symbols . . . . . . . . . . . . . . . . . . . 96
6.4 The remarkable theorem of Gauss . . . . . . . . . . . . 99
6.5 Isometries . . . . . . . . . . . . . . . . . . . . . . 100
6.6 Exercises . . . . . . . . . . . . . . . . . . . . . . 106
7. Geodesics . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1 The geodesic equations . . . . . . . . . . . . . . . . 109
7.2 Existence of geodesics . . . . . . . . . . . . . . . . . 112
7.3 Geodesic coordinates . . . . . . . . . . . . . . . . . 112
7.4 The rst fundamental form of a geodesic coordinate system 113
7.5 Interpretation of the Gauss theorem . . . . . . . . . . 115
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . 117
Appendices . . . . . . . . . . . . . . . . . . . . . . . . 119
A Euclidean spaces . . . . . . . . . . . . . . . . . . . 119
B Dierentiable functions of several variables . . . . . . . 120
C Normal vectors and cross products . . . . . . . . . . . 125
D Diagonalization of symmetric matrices . . . . . . . . . 126
E Hyperbolic functions . . . . . . . . . . . . . . . . . 130
Notational index . . . . . . . . . . . . . . . . . . . . . . 132
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Chapter 1
Parametrized curves and surfaces
In this chapter the basic concepts of curves and surfaces are introduced,
and examples are given. These concepts will be described as subsets of R
2
or
R
3
with a given parametrization, but also as subsets dened by equations.
The connection from equations to parametrizations is drawn by means of the
implicit function theorems (Theorems 1.5, 1.6 and 1.7).
1.1 Curves
It is well known from elementary geometry that a line in R
2
or R
3
can be
described by means of a parametrization t p + tq where q ,= 0 and p are
xed vectors, and the parameter t runs over the real numbers. Likewise, a
circle in R
2
(say with center 0) can be parametrized by t (r cos t, r sin t)
where t R. The common nature of these examples is expressed in the
following denition.
Denition 1.1. A parametrized continuous curve in R
n
(n = 2, 3, . . . ) is a
continuous map : I R
n
, where I R is an open interval (of end points
a < b ).
a b

x
y
The image set ( = (I) R
n
is called the trace of the curve. It is
important to notice that we distinguish the curve and its trace. Physically,
a curve describes the motion of a particle in n-space, and the trace is the
trajectory of the particle. If the particle follows the same trajectory, but
with dierent speed or direction, the curve is considered to be dierent.
For example, the positive x-axis is the trace of the parametrized curve
(t) = (t, 0) where t I =]0, [, but it is also the trace of (t) = (e
t
, 0) with
t R.
Notice also that we do not require the parametrization to be injective. A
point in the trace, which corresponds to more than one parameter value t, is
2 Chapter 1
called a self-intersection of the curve. For example, in the above parametriza-
tion of the circle, all points are self-intersections because values t +2k cor-
respond to the same point for all k Z.
In these notes we will mainly be concerned with plane curves (n = 2) and
space curves (n = 3), but in order to treat both cases simultaneously it is
convenient not to specify n. We do not assume n 3 for the time being,
since it does not lead to any simplications.
A parametrized continuous curve, for which the map : I R
n
is dif-
ferentiable up to all orders, is called a parametrized smooth curve. Recall
that a map f into R
n
is dierentiable if each of its components f
1
, . . . , f
n
is dierentiable. The class of continuous curves is wide and the requirement
of smoothness is a strong limitation. For example, the bizarre Peano curve,
which is dened on [0, 1] and has the entire unit square as trace, is continu-
ous but not smooth. In these notes we will only study smooth curves, and
we therefore adopt the convention that from now on a parametrized curve is
smooth, unless otherwise mentioned.
We have already seen that lines and circles can be parametrized as smooth
curves. Here are some further examples.
Example 1.1.1. The constant curve given by (t) = p, t I, where p R
n
is xed and I some open interval, is a parametrized curve.
Example 1.1.2. The map (t) = (a cos t, b sint), where a, b > 0 are con-
stants, parametrizes the ellipse ( = (x, y) [
x
2
a
2
+
y
2
b
2
= 1.
x
y
Example 1.1.3. Let (t) = (a cosht, b sinht) where a, b > 0 and (see
Appendix E)
cosh t =
e
t
+e
t
2
, sinh t =
e
t
e
t
2
.
Using the equation cosh
2
t sinh
2
t = 1 we see that is a parametrization
of the hyperbola (branch) ( = (x, y) [
x
2
a
2

y
2
b
2
= 1, x > 0.
x
y
Parametrized curves and surfaces 3
Example 1.1.4. The space curve (t) = (t, r cos(t), r sin(t)), where
r > 0 and , ,= 0 are constants, is called a helix. It is the spiraling motion
of a point which moves along the x-axis with velocity while at the same
time rotating around this axis with radius r and angular velocity .
z
y
x
1.2 Surfaces
Denition 1.2. A parametrized continuous surface in R
3
is a continuous
map : U R
3
, where U R
2
is an open, non-empty set.
u
v
U
p

y
z
x
(p)
(U)
It will often be convenient to consider the pair (u, v) U as a set of
coordinates of the point (u, v) in the image o = (U). However, since is
not assumed to be injective, the same point in o may have several pairs of
coordinates.
We call a parametrized continuous surface smooth if the map : U R
3
is smooth, that is, if the components
i
, i = 1, 2, 3, of
(u, v) = (
1
(u, v),
2
(u, v),
3
(u, v))
have continuous partial derivatives with respect to u and v, up to all or-
ders. We adopt the convention that a parametrized surface is smooth, unless
otherwise mentioned.
Example 1.2.1. Plane. Let p, q
1
, q
2
R
3
be xed vectors and let
(u, v) = p +uq
1
+vq
2
for (u, v) U = R
2
. Then is a parametrized surface. If q
1
, q
2
are linearly
independent, the image (U) is a plane. Otherwise it is a line or a point.
4 Chapter 1
Example 1.2.2. Sphere. Let
(u, v) = (cos ucos v, cos usin v, sinu)
where (u, v) R
2
. This is a standard parametrization of the unit sphere
S
2
= (x, y, z) R
3
[ x
2
+y
2
+z
2
= 1.
The parameters u and v are called latitude and longitude, and together they
are called spherical coordinates.
y
z
x
u
v
(u, v)
This parametrization covers the total sphere, but it is not injective. On
the other hand, if we request, for example, that u ]

2
,

2
[ and v ] , [,
then is injective, but it is not surjective, since a half-circle on the back of
the sphere will be outside the image of .
Example 1.2.3. Cylinder. Let r > 0 and put
(u, v) = (r cos v, r sin v, u)
where (u, v) R
2
. The image o of is the cylinder (x, y, z) [ x
2
+y
2
= r
2

of radius r.
y
z
x
As before we have to restrict to a smaller set U if we want to be injective,
for example by requiring v to belong in a xed open interval of length 2.
Parametrized curves and surfaces 5
Example 1.2.4. Cone. Let > 0 and
(u, v) = (ucos v, usinv, u)
where (u, v) R
2
, then the image of is the cone (x, y, z) [ x
2
+y
2
=
2
z
2
.
y
z
x
1.3 Graphs
By denition, the graph of a map h: A B, where A and B are arbitrary
sets, is the set of all pairs (x, h(x)) AB, where x A.
Let h: I R be a smooth function, where I R is an open interval. The
map t (t, h(t)) from I to R
2
parametrizes the graph and makes it into a
parametrized plane curve. We shall always regard the graph as being this
parametrized curve.
x
y
I
(t) = (t, h(t))
Likewise we shall regard the graph of a smooth function h: I R
2
as the
parametrized curve t (t, h(t)) = (t, h
1
(t), h
2
(t)) in R
3
.
Example 1.3.1. The graph of an ane linear function h(t) = at+b, R R
(where a, b R), is the line in R
2
parametrized by (t, at +b). All lines which
are not perpendicular to the x-axis can be parametrized in this fashion.
Similarly the graph of an ane linear function h(t) = at + b, R R
2
(where a = (a
1
, a
2
), b = (b
1
, b
2
) R
2
), is the line in R
3
parametrized by
(t, a
1
t + b
1
, a
2
t + b
2
). All lines of direction not perpendicular to the x-axis
can be parametrized in this fashion.
Example 1.3.2. The helix in Example 1.1.4 with = 1 is the 3-dimensional
graph of the map h: R R
2
dened by h(t) = (r cos(t), r sin(t)).
6 Chapter 1
We shall also consider surfaces, which are graphs. If h: U R is a smooth
function dened on an open set U R
2
, then the graph of h is the set
(u, v, h(u, v)) [ (u, v) U R
3
.
Equipped with the map
(u, v) = (u, v, h(u, v)), (u, v) U,
the graph is clearly a parametrized smooth surface.
Example 1.3.3. The graph of an ane linear function R
2
R is a plane in
R
3
. Say h(u, v) = au+bv+c where a, b, c R, then (u, v) = (u, v, au+bv+c).
All planes, except those which are perpendicular to the xy-plane, can be
parametrized in this fashion.
Example 1.3.6. The graph of the function h(u, v) =

1 u
2
v
2
, dened
on the unit disk U = (u, v) R
2
[ u
2
+v
2
< 1 is a half-sphere.
1.4 Level sets
Very often a plane curve is described, not by means of a parametrization,
but by an equation. For example, a line is represented by an equation of the
form ax+by = c with a, b, c R and (a, b) ,= (0, 0), and a circle is represented
by an equation of the form (x x
0
)
2
+ (y y
0
)
2
= r
2
with r > 0.
Similarly a surface can be described by an equation. For example, a plane
in R
3
is the set of solutions to an equation ax+by +cz = d, where (a, b, c) ,=
(0, 0, 0), and a sphere is represented by (xx
0
)
2
+(y y
0
)
2
+(z z
0
)
2
= r
2
.
We shall now give a general denition which covers both situations.
Denition 1.4.1. Let R
n
be open and let f: R be a continuous
function. The level sets for f are the sets
( = x [ f(x) = c
of solutions in to the equation f(x) = c, where c R is a xed constant.
In this course the function f will be assumed to be smooth. However, the
smoothness alone does not ensure that the level sets for f can be parametrized
as smooth curves or surfaces (in case n = 2 or 3). For example, a level set for
the trivial function f = 0 on R
2
, that is, the set of solutions to an equation
0 = c, is either the empty set or the full set R
2
. Some extra condition will
be needed on f in order that the set is a curve.
Denition 1.4.2. Let f: R be smooth, where R
n
is open. A point
p is called critical if
f
x
1
(p) = =
f
x
n
(p) = 0.
Parametrized curves and surfaces 7
Let us consider some examples in the plane case n = 2. It will be seen
in all the examples that if we exclude critical points, the level sets can be
parametrized as curves. A precise statement to this eect is given in the
corollary in Section 1.5.
Example 1.4.1. Consider the linear equation ax + by = c whose solutions
comprise a level set for f(x, y) = ax + by. If (a, b) ,= (0, 0) then there
are no critical points. In this case the set of solutions form a line, hence
can be parametrized as a curve. On the other hand, if (a, b) = (0, 0) then
f(x, y) = ax +by is the trivial function and all points are critical.
Example 1.4.2. Let f(x, y) = x
2
+ y
2
, then
f
x
= 2x and
f
y
= 2y, so (0, 0)
is the only critical point. The level sets for c > 0 contain no critical points.
They are circles, hence can be parametrized as smooth curves. The level set
for c = 0 consists only of the critical point (0, 0) and it is exactly in this case
the circles degenerate to a point.
Example 1.4.3. Consider the equation f(x, y) = xy = 0. Here f/x = y
and f/y = x, and hence the origin (0, 0) is the only critical point. In fact,
the level set given by f(x, y) = 0 is the union of the two axes, which exactly
fails to be a reasonable curve at the origin.
1.5 The implicit function theorem, two variables
The implicit function theorem describes conditions under which a given
equation in two variables can be solved to obtain one of the variables as
a function of the other variable. For some simple equations, for example
y
2
2xy + 1 = 0, explicit solutions are easily obtained by simple algebra,
here y = x +

x
2
1 and y = x

x
2
1, but for other equations such
explicit solutions cannot be derived. The reason for this need not be lack of
algebraic skill on our side, since a solution may not exist at all. The implicit
function theorem expresses a simple condition which guarantees the existence
of a function h, such that the solution is y = h(x).
Theorem 1.5. Let f: R be a smooth function, where R
2
is open.
Let
( = (x, y) [ f(x, y) = c
be the set of solutions to the equation f(x, y) = c. Let p = (x
0
, y
0
) ( be
given, and assume that
f
y
,= 0 at p.
Then there exist open intervals I and J around x
0
and y
0
, respectively,
such that the rectangle W = I J is contained in , and a smooth map
h: I J such that
( W = (x, h(x)) [ x I, (1)
that is, in the neighborhood W of p, ( is the graph of h.
8 Chapter 1
x
0
y
0
(
p
W
Proof. Assume for simplicity that c = 0, and that the value of
f
y
at p is
positive. These properties can be arranged by a simple replacement of f
which does not aect the set (. Choose > 0 such that the neighborhood
(x, y) [ [xx
0
[ , [y y
0
[ of p lies inside , and such that
f
y
a on
this neighborhood, for some constant a > 0 (continuity of
f
y
is used). Then
y f(x, y) is strictly increasing on the interval [y
0
, y
0
+], for each xed
x with [x x
0
[ < .
In particular, since p ( we have f(p) = f(x
0
, y
0
) = 0, and hence
f(x
0
, y
0
) < 0 and f(x
0
, y
0
+) > 0.
By continuity in x
0
of each of the maps x f(x, y
0
), there exists a
positive number such that f(x, y
0
) < 0 and f(x, y
0
+) > 0 for all
x with [x x
0
[ < .
x
0
y
0
x
0
+
y
0
+
x
0

y
0

p
f>0
f<0
f=0
Let I = x [ [x x
0
[ < , and let x I. Since y f(x, y) is strictly
increasing and continuous, and since f(x, y
0
) < 0 and f(x, y
0
+ ) > 0,
there exists a unique y between y
0
and y
0
+ with f(x, y) = 0. This value
of y is denoted h(x). Then h maps I into J =]y
0
, y
0
+ [ and satises
f(x, h(x)) = 0. The identity of the sets in (1) follows from the uniqueness of
y. We will complete the proof of the theorem by showing that h is smooth.
We rst prove that h is continuous. Fix x I and let y = h(x), then
f(x, y) = 0. Let x be suciently small so that x+x I. Associated to x
we dene y such that y+y = h(x+x), then also f(x+x, y+y) = 0.
The asserted continuity amounts to the statement that y 0 when
x 0. The function
t (t) = f(x +tx, y +ty)
Parametrized curves and surfaces 9
is zero both for t = 0 and t = 1. By the mean value theorem (Rolles theorem)
there exists a number (0, 1) (depending on x) such that

() = 0.
Dierentiating by means of the chain rule we thus obtain
f
x
(x +x, y +y)x +
f
y
(x +x, y +y)y = 0.
Hence
y =
f
x
(x +x, y +y)
f
y
(x +x, y +y)
x,
and since [
f
x
[ is bounded, and
f
y
a > 0, it follows that y 0 when
x 0, as claimed.
Next we prove that h is dierentiable, which with the notation from above
amounts to the convergence of y/x as x 0. In fact, since
f
x
and
f
y
are continuous, this follows immediately from the equation above. Moreover,
the limit is given by
lim
x0
y
x
=
f
x
(x, y)
f
y
(x, y)
.
Hence h is dierentiable and satises
h

(x) =
f
x
(x, h(x))
f
y
(x, h(x))
. (2)
Finally, we prove by induction that h is smooth. Assuming that h is r
times dierentiable for some natural number r, we see from equation (2) that
so is h

. Hence h is r + 1 times dierentiable.

Corollary 1.5. Let f: R be a smooth function, where R
2
is open.
Let
( = (x, y) [ f(x, y) = c
and let p = (x
0
, y
0
) (. Assume that p is not a critical point.
Then there exists an open rectangle W around p, such that ( W
is the graph of a smooth function h, considered either as y = h(x) or as
x = h(y).
In particular, it follows that the level set can be parametrized as a smooth
curve in a neighborhood of each non-critical point.
Proof. By assumption
f
x
and
f
y
are not both zero at p. If
f
y
(p) ,= 0 the
conclusion is already in the previous theorem. Otherwise, we interchange x
and y.
10 Chapter 1
Example 1.5.1. Let f: R
2
R be given by f(x, y) = y
2
2xy + 1, and
consider the level set ( = (x, y) [ f(x, y) = 0. Then
f
y
= 2y 2x, which is
zero if and only if x = y. Inserting y = x in the equation y
2
2xy+1 = 0, we
see that the only points in ( where
f
y
= 0 are p = (1, 1) and q = (1, 1).
We can then conclude from the theorem that the level set ( can be attained
as a graph of the form y = h(x) in a neighborhood of each of its points, except
possibly p and q (the theorem gives no information in case
f
y
= 0).
On the other hand, the partial derivative
f
x
= 2y is never zero on (
(since y = 0 in y
2
2xy + 1 = 0 leads to a contradiction), and thus ( has
the form of a graph x = h(y) in a neighborhood of all its points.
In fact, the equation can be easily solved with respect to both x and y:
y = x
_
x
2
1, x =
1
2
(y +
1
y
).
The formula on the left gives two expressions, each with y a function of x.
Only one of these is relevant in a neighborhood of a given point (x, y)
(, provided [x[ > 1. However, at the points where x = 1, these two
expressions for y collapse, and neither of them gives a well-dened function
in a neighborhood, because [x[ < 1 is not allowed in the square root. Notice
that these are exactly the two points where
f
y
= 0.
The expression for x, on the other hand, is dened and smooth for all
y ,= 0 (and y = 0 never occurs in ().
x
y
p = (1, 1)
q = (1, 1)
Example 1.5.2. Let f(x, y) = x
4
x
2
+y
2
, and ( = (x, y) [ f(x, y) = 0.
The derivatives
f
x
= 4x
3
2x and
f
y
= 2y
are both zero if and only if (x, y) is one of the three points
(0, 0), (
_
1
2
, 0), (
_
1
2
, 0).
Only the rst one of these belongs to (, and this point is therefore the only
critical point in (.
Parametrized curves and surfaces 11
The set ( is shown in the following gure. It can be shown that ( is
the trace of the parametrized curve (t) = (cos t, cos t sint), which has a
self-intersection exactly in the critical point (0, 0).
x
y
1.6 The implicit function theorem, more variables
We will now consider the analogue for surfaces of the theory of the pre-
ceding section. Where the solution set for an equation in two variables was
described as a parametrized curve, the analogous theorem describes the so-
lution set for an equation in three variables as a parametrized surface.
In fact, it is convenient to state the theorem as a theorem treating an
equation in n variables, with arbitrary n 2. In this fashion the theorem
becomes a generalization rather than an analogue.
In order to compare easily with the previous theorem we denote our vari-
ables in R
n
by (x, y) = (x
1
, . . . , x
n1
, y) where x R
n1
and y R. By
denition, an interval in R
n
is a product I
1
I
n
of intervals in R.
Theorem 1.6. Let f: R be a smooth function, where R
n
is open.
Let
o = (x, y) [ f(x, y) = c
and let p = (x
0
, y
0
) o. Assume that
f
y
,= 0 at p.
Then there exist open intervals I R
n1
and J R around x
0
and y
0
,
respectively, such that the interval W = I J is contained in , and a smooth
map h: I J such that
o W = (x, h(x)) [ x I,
that is, in the neighborhood W of p, o is the graph of h.
Proof. Notice the similarity with Theorem 1.5, the only dierence being that
x R is replaced by (x
1
, . . . , x
n1
) R
n1
. In fact, the proof is a rather
straightforward generalization of the proof of Theorem 1.5, and it is therefore
omitted.
Let us take n = 3, and replace the notation (x
1
, x
2
, y) by (x, y, z). We
obtain the following result which is analogous to the corollary in Section 1.5.
12 Chapter 1
Corollary 1.6. Let f: R be a smooth function, where R
3
is open.
Let
o = (x, y, z) [ f(x, y, z) = c
and let p = (x
0
, y
0
, z
0
) o. Assume that p is not a critical point.
Then there exist an open interval W around p, such that o W is the
graph of a smooth function h, considered either as z = h(x, y), as y = h(x, z)
or as x = h(y, z).
In particular, it follows that the level set can be parametrized as a smooth
surface in a neighborhood of each non-critical point.
Proof. By assumption at least one of the partial derivatives
f
x
,
f
y
and
f
z
is not zero at p. Interchanging z with x or y if necessary, we may assume
that it is
f
z
. The conclusion then follows from the previous theorem.
Example 1.6.1. The equation for a plane in R
3
, ax + by + cz = d, where
(a, b, c) ,= (0, 0, 0), satises the assumption of the preceding corollary. If c ,=
0, the plane is the graph z = h(x, y) of the function h(x, y) = (daxby)/c.
If c = 0 the plane is vertical, and we cannot exhibit it as a graph of the
form z = h(x, y), but then we can exhibit it as a graph over one of the other
coordinate planes.
Example 1.6.2. The surfaces introduced in Examples 1.2.2-1.2.4, sphere,
cylinder and cone, are level sets for, respectively, f(x, y, z) = x
2
+ y
2
+ z
2
,
f(x, y, z) = x
2
+y
2
and f(x, y, z) = x
2
+y
2
z
2
. These functions all satisfy
the assumption of the corollary above, except for the vertex (0, 0, 0) of the
cone.
Example 1.6.3. Let f(x, y, z) = ze
x
+ yz and consider the equation
f(x, y, z) = 1 in a neighborhood of the point p = (0, 0, 1) (which solves
the equation). The partial derivative
f
z
= e
x
+ y is 1 at this point, so by
the implicit function theorem the solution z exists as a function of (x, y)
in a neighborhood of (0, 0, 1). In fact, the equation has the solution z =
1/(e
x
+ y). However, if the equation is replaced by for example f(x, y, z) =
ze
x
+ sin(yz) = 1, then it is impossible to write down a solution to the
equation in terms of known functions, but the conclusion from the implicit
function theorem remains the same since we still have
f
z
= 1 at p.
1.7 The implicit function theorem, more equations
We have seen that an equation f(x, y) = c in R
2
denes a plane curve,
and that an equation f(x, y, z) = c in R
3
denes a surface (under suitable
circumstances). In order to dene a curve in R
3
we need two equations. For
example, the x-axis is the set dened by equations y = 0 and z = 0. Consider
two equations of the form
f
1
(x, y, z) = c
1
,
f
2
(x, y, z) = c
2
,
Parametrized curves and surfaces 13
where f
1
, f
2
maps an open set R
3
into R. We say that we have two
equations in three variables. It is actually more convenient to write the
equations in the form
f(x, y, z) = c
where f = (f
1
, f
2
) maps into R
2
, and where c = (c
1
, c
2
). We want to
generalize Theorem 1.6 in order to deal with this situation.
In fact, we will generalize even further, to functions f: R
n
R
m
, that
is, to the case of m equations in n variables, the only condition being that
n > m. For the application to space curves, only n = 3 and m = 2 is needed,
and the reader is encouraged to specialize to this case at rst reading.
It is convenient for the comparison with Theorem 1.6 to write elements in
R
n
as (x, y) where x = (x
1
, . . . , x
nm
) R
nm
and y = (y
1
, . . . , y
m
) R
m
.
The goal is to obtain y as a function of x.
Theorem 1.7. Let f: R
m
be a smooth function, where R
n
is
open. Let c R
m
be xed. Let
( = (x, y) [ f(x, y) = c
and let p = (x
0
, y
0
) (. Assume that the determinant of the mm matrix
A =
f
i
y
j
(p),
consisting of the last m columns of the Jacobian Df(p), is non-zero.
Then there exist open intervals I R
nm
and J R
m
around x
0
and y
0
,
respectively, such that W = I J , and a smooth map h: I J such
that
( W = (x, h(x)) [ x I,
that is, in the neighborhood W of p, ( is the graph of h: R
nm
R
m
.
Proof. The most common technique for solving several equations in several
variables is elimination of variables. That is, we use one of the equations to
express a particular variable in terms of the others, and insert this expression
in the remaining equations. The chosen variable has then been eliminated,
and the number of equations is reduced by one. This will also be our strategy
in the present proof.
We prove the theorem by induction on m. The case m = 1 was already
treated in Theorem 1.6. Thus, we assume that m 2 and that the conclusion
of the theorem is valid for functions into R
m1
. We can safely assume that
c = 0, since this can be arranged by subtraction of the constant from f.
Since det A is non-zero, A is invertible. We want to replace f: R
n
R
m
by the function A
1
f: R
n
R
m
, obtained by multiplying all image vectors
f(x, y) R
m
with the constant matrix A
1
. Since multiplication by A
1
14 Chapter 1
is a bijection, the equations f(x, y) = 0 and A
1
f(x, y) = 0 are equivalent.
The Jacobian of the linear map, multiplication by A
1
, is the matrix A
1
itself (see Example B.1), and hence it follows from the chain rule that
D(A
1
f)(p) = A
1
Df(p).
We see that the last m columns of D(A
1
f)(p) comprise a unit matrix

kj
. By the replacement of f with A
1
f we thus obtain a function whose
Jacobian matrix at p has a unit matrix in its last m columns, and for which
the solution set ( is unaltered. From now on we assume this replacement has
been carried out, that is, we assume f
k
/y
j
=
kj
.
In particular, for the function f
m
whose derivatives are in the last row
of Df, we have that f
m
/y
m
(p) = 1. We will apply Theorem 1.6 to the
equation f
m
(x, y) = 0. The eect of the theorem is that the last variable,
y
m
, can be written as a smooth function of the remaining variables. We
write the remaining variables as (x, y

) R
n1
where y

= (y
1
, . . . , y
m1
).
More precisely, it then follows that there exists an interval neighborhood
W
1
= I
1
J
1
around p, where I
1
R
n1
and J
1
R, and a smooth function
h: I
1
J
1
such that for (x, y) = (x, y

, y
m
) W
1
we have f
m
(x, y) = 0 if
and only if
y
m
= h(x, y

).
Let the function F: I
1
R
n1
R
m1
be dened by
F
k
(x, y

) = f
k
(x, y

, h(x, y

)) (3)
for k = 1, . . . , m 1, where as before y

= (y
1
, . . . , y
m1
). The partial
derivatives of F
k
are obtained by applying the chain rule to (3):
F
k
y
j
=
f
k
y
j
+
f
k
y
m
h
y
j
(j = 1, . . . , m1),
and at p we thus have
F
k
y
j
=
f
k
y
j
=
kj
(because
f
k
y
m
= 0). The determinant
of this matrix being non-zero, we can apply our induction hypothesis to
F, and we obtain the existence of an interval neighborhood W
2
= I
2
J
2
around (x
0
, y
0

), where I
2
R
nm
and J
2
R
m1
, and a smooth function
g: I
2
J
2
such that the solution set for the equation F(x, y

) = 0 in W
2
is
the graph of g, that is, F(x, y

) = 0 if and only if y

= g(x).
Let the interval W R
n
be dened by
W = W
1
(x, y) [ (x, y

) W
2
.
We now see that for (x, y) in this set we have
(x, y) (
Parametrized curves and surfaces 15
if and only if
f
k
(x, y) = 0, k = 1, . . . , m
if and only if
f
k
(x, y) = 0, k = 1, . . . , m1 and y
m
= h(x, y

)
if and only if
F(x, y

) = 0 and y
m
= h(x, y

)
if and only if
y

= g(x) and y
m
= h(x, y

)
if and only if
y = (g(x), h(x, g(x))).
The function x (g(x), h(x, g(x))) is thus seen to be the desired function
whose graph is ( in a neighborhood of p.
Corollary 1.7. Let f: R
2
be smooth, where R
3
is open. Let c R
2
and
p ( = (x, y, z) [ f(x, y, z) = c.
Assume the rows of Df(p) (a 2 3 matrix) are linearly independent.
Then there exist an open interval W around p, such that ( W can
be parametrized as a smooth curve in the form of a graph, considered either
as (y, z) = h(x), as (x, z) = h(y) or as (x, y) = h(z).
Proof. At least one of the three 2 2 submatrices of Df(p) has non-zero
determinant. With suitable reorganization of variables the theorem can be
applied.
Example 1.7. The set of equations
x +xy +z
2
= 3 x
3
+ 2xz y
2
z
2
= 2 (4)
has the form f(x, y, z) = c with
f(x, y, z) =
_
x +xy +z
2
x
3
+ 2xz y
2
z
2
_
, c =
_
3
2
_
.
The Jacobian matrix is
Df(x, y, z) =
_
1 +y x 2z
3x
2
+ 2z 2yz
2
2x 2y
2
z
_
.
In the point (1, 1, 1) the equations (4) are satised and the Jacobian is
_
2 1 2
5 2 0
_
.
16 Chapter 1
The determinant of the last two columns is det A = 4. Since this determinant
is not zero, the implicit function theorem assures that the equations can
be solved for (y, z) as function of x, in a neighborhood of (1,1,1). In this
neighborhood the set of solutions can thus be parametrized as a smooth curve
in the form of a 3-dimensional graph (t, h(t)) where h(t) = (y(t), z(t)) R
2
.
This example demonstrates the theoretical power of the implicit function
theorem, since the explicit solving of (4) for y and z as functions of x is
clearly a dicult task.
1.8 Exercises
1 Let : I R
n
be a parametrized curve with

(t) = 0 for all t. Show that

it is a line or a constant curve.
2 The following parametrized curve is called the cycloid
(t) = (t sint, 1 cos t), (t R).
It is constructed by a circle of radius 1 rolling without slipping along the
positive x-axis. The curve is the path of a point on the circumference of
the circle. Explain the formula above from this construction.
x
y
3 Determine a parametrized surface with image (x, y, z) [ x+2y 2z = 1.
4 Consider the equation x
3
+xy
2
2ay
2
= 0 in R
2
, where a > 0 is a constant.
Show that the parametrized curve
(t) = (
2at
2
1 +t
2
,
2at
3
1 +t
2
),
where t R, is bijective onto the set of solutions. This curve is called the
cissoid of Diocles. Draw a sketch of it (say for a = 1).
5 Let o R
3
denote the set of solutions to the equation x
2
+y z
2
= 1.
a. Show that the map : R
2
R
3
given by
(u, v) = (u +v, 1 4uv, u v)
is an injective parametrized surface with image o.
b. Show that o contains no critical points for the function f(x, y, z) =
x
2
+y z
2
, and determine a parametrization of o as a graph of a smooth
function h.
Parametrized curves and surfaces 17
6 Consider the equation x
3
y
3
3x + y = 1, which is satised by (x, y) =
(1, 1). Show that it is possible to describe the level set as a graph y = h(x)
in a neighborhood of this point.
7 Denote by ( the level set in R
2
for the equation 4x
4
5x
2
y
2
+y
4
= 0.
a. Show by means of the implicit function theorem that for each point
(x
0
, y
0
) ( (0, 0) there exists a neighborhood in which ( can be
described as a graph y = h(x).
b. Solve the equation and determine ( explicitly.
8 The condition f

y
(x
0
, y
0
) ,= 0 in Theorem 1.5 is sucient but not necessary.
That is, if f

y
(x
0
, y
0
) = 0 it may still be possible to describe the level set
as a graph y = h(x) of a smooth function h in a neighborhood of (x
0
, y
0
).
Give an example.
9 Let f be as in Theorem 1.5, but assume instead that f

y
(x
0
, y
0
) = 0. Show
that if f

x
(x
0
, y
0
) ,= 0, then it is not possible to describe the level set as a
graph y = h(x), where h is smooth, in any neighborhood of (x
0
, y
0
).
10 The equation xy + xz + sin z = 0 is solved by (x, y, z) = (0, 0, 0). Show
that the solution set in a neighborhood of this point allows a description
as a graph. Show that this is the case in a neighborhood of all solutions.
11 Let o = (x, y, h(x, y)) R
3
be the graph of the function h(x, y) =
yxy
3
. Show that in a neighborhood of each point of o in which 3xy
2
,= 1,
it is possible to write o as a graph of the form y = g(x, z) with g smooth.
12 Consider the system of equations in R
3
2x
2
x
2
z
2
y
2
= 0 xyz = 1
to which (1, 1, 1) is a solution. Show that there exists a neighborhood in
which the solution set can be described as a graph of the form (x, y) =
h(z), where x and y both are functions of z.
13 Let f: R
3
R
2
and c R
2
be given by
f(x, y, z) = (x
2
+y
2
+z
2
, (x
1
2
)
2
+y
2
), c = (1, a
2
)
where a 0. Let L R
3
denote the set of solutions to the system
f(x, y, z) = c.
a. Explain why L is the intersection of a sphere and a cylinder, and
determine their radii.
b. Determine, in each of the following 6 cases, the set of points in L
for which the rank of Df(p) is < 2, that is, where the rows are linearly
dependent.
a = 0, 0 < a <
1
2
, a =
1
2
,
1
2
< a <
3
2
, a =
3
2
,
3
2
< a.
c. What does the implicit function theorem tell about L in each case.
Explain by means of the observation in a.
18 Chapter 1
Chapter 2
Tangents
We have equipped parametrized curves and surfaces with the standing
assumption that the parametrization is smooth. However, smoothness alone
is not enough to ensure a simple geometric appearance. For example, the
plane curve (t) = (t
3
, t
2
) is perfectly smooth, but in (0) = (0, 0) the
trace of the curve has a sharp fold (a so-called cusp), which conicts with
the intuitive notion of smooth. Another striking example will be given in
Example 2.1.4 below.
x
y
(t) = (t
3
, t
2
)
In this chapter we will dene a notion of regularity for parametrized curves
and surfaces, which is motivated by the desire to exclude anomalies as the
one just mentioned. The geometric signicance of the regularity condition
will be that it allows us to dene notions of tangent lines and tangent planes.
2.1 Regular curves and tangent lines
Let : I R
n
be a parametrized curve and let t
0
I be given.
Denition 2.1.1. The curve is called regular in t
0
if

(t
0
) ,= 0. Otherwise
it is called singular. If is regular in all points of I we call it a regular
parametrized curve or just a regular curve.
For example, the plane curve mentioned above is regular for t ,= 0 but it
is singular at t = 0. The standard parametrizations of line and sphere (see
Section 1.1), and the curves described in Examples 1.1.2, 1.1.3 and 1.1.4 are
all regular curves.
Example 2.1.1 A constant curve (Example 1.1.1) is everywhere singular.
Conversely, an everywhere singular parametrized curve is constant, since

= 0 implies that is constant.

20 Chapter 2
Example 2.1.2 A graph (t) = (t, h(t)) (Section 1.3) is a regular curve in
R
2
, since

(t) = (1, h

(t)) ,= (0, 0) (also if h

(t) = 0). Hence, by Corollary

1.5, a level set f(x, y) = c can be parametrized as a regular curve in a
neighborhood of each point which is not critical.
Denition 2.1.2. The vector

(t
0
) is called the tangent vector to at t
0
.
If is regular at t
0
, the line through p = (t
0
) with direction

(t
0
) is called
the tangent line of the curve.
The latter denition is motivated by the following result, which describes
the tangent vector geometrically. The notation |v| for vectors v R
n
is
dened in Appendix A.
Theorem 2.1. Assume that is regular at t
0
, and let v =

(t
0
)/|

(t
0
)|
be the unit vector in the direction of the tangent vector. Then
v = lim
tt
+
0
(t) (t
0
)
|(t) (t
0
)|
= lim
tt

0
(t
0
) (t)
|(t
0
) (t)|
. (1)
In other words, the unit tangent vector v is the limit position of the direction
from (t
0
) to (t), as t approaches t
0
from above, and the limit position of
the direction from (t) to (t
0
), as t approaches t
0
from below.
(t
0
)
(t)
v
Proof. By denition

(t
0
) = lim
tt
0
(t) (t
0
)
t t
0
.
In particular, since

(t
0
) ,= 0 it follows that (t) ,= (t
0
) for all t I
suciently close to (but dierent from) t
0
. Thus the denominator of the
fraction in (1) is not zero. Moreover for t > t
0
,
(t) (t
0
)
|(t) (t
0
)|
=
1
|
(t)(t
0
)
tt
0
|
(t) (t
0
)
t t
0

1
|

(t
0
)|

(t
0
),
and similarly for t < t
0
.
Example 2.1.3 According to the theorem, regularity of is a sucient
condition for (1) to hold. It is not a necessary condition. For example, the
Tangents 21
curve (t) = (t
3
, 0) which has the x-axis as its trace, is singular at t
0
= 0,
but nevertheless both limits in (1) exist and are equal to the unit vector
v = (1, 0).
Example 2.1.4 A sophisticated example of a non-regular point on a smooth
curve can be constructed as follows. Let : R R be the function dened
by
(t) =
_
exp(
1
t
) if t > 0
0 otherwise
then it can be shown that is smooth (the derivatives up to all orders vanish
at 0). The graph of is shown to the left in the gure below.
x
y = (x) (x, y) = (t)
x
y y
1
1
Dene (t) = ((t), (t)) for t R. This is a smooth curve whose trace
consists of the line segment from 1 to 0 on the y-axis followed by the line
segment from 0 to 1 on the x-axes. It is not regular at the origin, which is
in accordance with the sharp turn of the curve in that point.
2.2 The tangent line of a level set
We have seen in Example 2.1.2 that the level set given by f(x, y) = c can
be parametrized as a regular curve (t) in a neighborhood of each non-critical
point p. We will determine the tangent line of such a parametrization.
Theorem 2.2. Let ( R
2
be a level set of a smooth function f, and let
p = (x
0
, y
0
) ( be non-critical. Let : I R
2
be any parametrized curve
with trace (I) ( and with (t
0
) = p for some t
0
I, in which is regular.
Then the tangent line of at t
0
is characterized by the equation
f
x
(p)(x x
0
) +
f
y
(p)(y y
0
) = 0.
Proof. We shall be using the following simple fact from plane geometry. The
line with normal vector (a, b) ,= (0, 0) through (x
0
, y
0
) is given by the equation
a(x x
0
) +b(y y
0
) = 0.
We thus have to prove that the tangent line has (
f
x
(p),
f
y
(p)), which is
non-zero by assumption, as a normal vector.
22 Chapter 2
Write (t) in coordinates as (t) = (x(t), y(t)). Since maps into the
level set we have f(x(t), y(t)) = c for all t. By dierentiation with the chain
rule we obtain
x

(t
0
)
f
x
(p) +y

(t
0
)
f
y
(p) = 0,
which exactly shows that (
f
x
(p),
f
y
(p)) is perpendicular to the tangent vec-
tor

(t
0
) = (x

(t
0
), y

(t
0
)).
Notice that it follows from the theorem that the tangent line depends on
the level set through the function f, but it is independent of the chosen
parametrization .
2.3 The tangent plane of a regular surface
Let : U R
3
be a parametrized surface with image o = (U), and let a
point p = (u
0
, v
0
) U be given.
The notion of regularity for a parametrized surface is somewhat more
complicated than that for a curve, because of the fact that we can dierentiate
with respect to both u and v. Let = (
1
,
2
,
3
) and put

u
=
_
_
_

1
u

2
u

3
u
_
_
_
and

v
=
_
_
_

1
v

2
v

3
v
_
_
_
.
These vectors are the columns in the Jacobi matrix
D =
_
_
_

1
u

1
v

2
u

2
v

3
u

3
v
_
_
_
.
Notice that

u
(p) and

v
(p) are the tangent vectors at t = 0 to the curves
t (u
0
+t, v
0
) and t (u
0
, v
0
+t), respectively.
u
v
U
p

y
z
x

v
(U)
Denition 2.3.1. A parametrized surface is called regular at p = (u
0
, v
0
)
if the partial derivatives

u
and

v
, evaluated at p, are linearly independent.
Tangents 23
Otherwise it is called singular. If is regular in all points of U we call it a
regular parametrized surface or just a regular surface.
Recall (see Appendix C) that for two vectors a = (a
1
, a
2
, a
3
) and b =
(b
1
, b
2
, b
3
) in R
3
we dene the cross product by
a b =
_

a
2
b
2
a
3
b
3

a
1
b
1
a
3
b
3

a
1
b
1
a
2
b
2

_
.
Since a and b are linearly independent if and only if a b ,= 0, the regularity
condition above is equivalent to

v
,= 0.
Example 2.3.1 For the standard spherical coordinates
(u, v) = (cos ucos v, cos usin v, sinu)
we derive

u
=
_
_
sin ucos v
sin usin v
cos u
_
_
,

v
=
_
_
cos usin v
cos ucos v
0
_
_
and hence

v
= (cos
2
ucos v, cos
2
usinv, cos usinu) = cos u(u, v). (2)
In particular, since (u, v) ,= 0 (it has length 1), we see that

v
= 0
if and only if cos u = 0, that is u =

2
(up to multiples of 2). The
points (p) on the sphere, where is singular at p, are thus the two poles
(0, 0, 1). Notice however that by choosing a dierent parametrization of
the sphere, we can arrange that these points are in the regular range (at the
cost of some other points becoming singular). For example with (u, v)
(cos ucos v, sinu, cos usin v), which diers from by an interchange of y and
z, the points (p) with p singular are (0, 1, 0).
Example 2.3.2 The graph (u, v) = (u, v, h(u, v)) (Section 1.3) of a func-
tion of two variables is a regular surface in R
3
, since

u
= (1, 0,
h
u
) and

v
= (0, 1,
h
v
)
are linearly independent (also if the partial derivatives of h are 0). Hence, by
Corollary 1.6, a level surface f(x, y, z) = c can be parametrized as a regular
surface in a neighborhood of each point which is not critical.
Denition 2.3.2. The linear subspace of R
3
spanned by the partial deriva-
tives

u
(p) and

v
(p) is called the tangent space of at p. It is denoted T
p
.
24 Chapter 2
If is regular at p = (u
0
, v
0
), then the plane through (p) and parallel to
T
p
is called the tangent plane at p.

v
Notice that the tangent space T
p
is a two-dimensional linear subspace of
R
3
if and only if is regular. In this case the pair of vectors

u
(p) and

v
(p)
form a basis for T
p
, and the use of the word plane for the tangent plane is
justied.
Example 2.3.3 Let be the standard parametrization of the unit sphere,
as in the Example 2.3.1. At (u
0
, v
0
) = (0, 0) we have (0, 0) = (1, 0, 0) and

u
= e
3
and

v
= e
2
(where e
1
, e
2
, e
3
are the standard basis vectors in R
3
).
The tangent space at (0, 0) is therefore the span of e
2
and e
3
(the yz-plane),
and the tangent plane is the plane through (1, 0, 0) parallel to this plane. On
the other hand, if u
0
=

2
(and v
0
is arbitrary) so that (u
0
, v
0
) = (0, 0, 1),
then

u
= (cos v
0
, sinv
0
, 0) and

v
= 0, so in this singular case the
tangent space at (u
0
, v
0
) is one-dimensional. However, the degeneracy of the
tangent space at this point is caused by the singularity of the parametrization,
and it has no geometric signicance for the sphere.
It is convenient to have the notion of tangent space because of its structure
as a linear space. On the other hand the tangent plane is more easy to
visualize, because it passes through the given point on o = (U).
For a level set we have the following analogue of Theorem 2.2.
Theorem 2.3. Let o R
3
be a level set of a smooth function f, and let
p = (x
0
, y
0
, z
0
) o be non-critical. Let : U R
3
be any parametrized
surface with image (U) o and with (u
0
, v
0
) = p for some (u
0
, v
0
) U,
in which is regular. Then the tangent plane of at (u
0
, v
0
) is characterized
by the equation
f
x
(p)(x x
0
) +
f
y
(p)(y y
0
) +
f
z
(p)(z z
0
) = 0.
Proof. Entirely analogous to that of Theorem 2.2.
As in Section 2.2 we observe that the tangent plane in p of the level set o
depends on f but is independent of the particular parametrization .
Tangents 25
2.4 Curves on a surface
We shall now give a geometric characterization of the tangent space based
on the following denition. Let : U R
3
be a parametrized surface.
Denition 2.4. A parametrized curve on is a parametrized curve written
in the form = : I R
3
where : I U is a parametrized plane smooth
curve.
u
v
U

I
R

(p)

(U)
Notice that is not uniquely determined by since we have not assumed
to be injective. Furthermore, even if is injective, so that is uniquely
determined by , then smoothness of is not ensured just by the smoothness
of . It is therefore important to emphasize that in the above denition the
smooth curve is assumed to be given together with .
Formally, a parametrized curve on is a pair of smooth curves and
satisfying = . The plane curve is said to be the coordinate curve of
the pair.
Example 2.4.1 The helix (t) = (t, r cos(t), r sin(t)) in Example 1.1.4
is realized as a curve on the cylinder (u, v) = (u, r cos v, r sin v) with
coordinate curve (t) = (t, t).
Lemma 2.4. Let = be a parametrized curve on . Then

(t) = u

(t)

u
((t)) +v

(t)

v
((t)) (3)
where u(t) and v(t) are the coordinates of (t) = (u(t), v(t)) U.
Proof. This follows from the chain rule for D(GF) with G = : U R
3
and
F = : I U, see Appendix B, equation (B.2). The 3 2 matrix DG = D
has columns

u
,

v
, and the derivative F

(t) =

(t) has the elements u

(t)
and v

(t). Their product DG(F(t))F

(t) is exactly the linear combination

u

u
+v

v
, as in (3).
26 Chapter 2
Theorem 2.4. The tangent space T
p
is equal to the set of tangent vectors

(t
0
) of all parametrized curves = on with (t
0
) = p for some
t
0
I.
Proof. It follows from (3) that

(t
0
) belongs to the span T
p
of

u
and

v
for all parametrized curves on with (t
0
) = p.
Conversely, let a linear combination a

u
+ b

v
T
p
be given. Let p =
(u
0
, v
0
) and dene (t) = (u(t), v(t)) = (u
0
+ at, v
0
+ bt) for t suciently
close to 0, so that (t) U. Let = . Then u

(t) = a and v

(t) = b,
hence it follows from the expression (3) for

(t) that

(0) = a

u
+b

v
.
2.5 Reparametrization of curves
It can often be useful to change the way a given curve is parametrized. For
example, one may prefer to parametrize the unit circle not by (cos t, sint), but
by (cos(t), sin(t)) for some angular velocity . This concept is formalised
in the following denition.
x
y

I
R

J
R

Denition 2.5.1. Let : I R

n
be a parametrized curve, and let : J
I be a smooth bijective map with a smooth inverse (I and J being open
intervals in R). The curve = : J R
n
is called a reparametrization
of .
For the justication of the condition on we recall the following result
from the calculus of functions of one variable.
Theorem 2.5. Let J R be an open interval and : J R a smooth map.
Let I = (J), then the following conditions are equivalent:
(i)

(u) ,= 0 for all u J,

(ii) : J I is bijective, I is an open interval, and
1
: I J is smooth.
Moreover, if these conditions hold, then
(
1
)

(t) =
1

(u)
(4)
Tangents 27
for all t I, where u =
1
(t).
Notice that if

(u) ,= 0 for all u J then, by continuity, either

(u) > 0
for all u or

(u) < 0 for all u. Thus is either monotonically increasing or

monotonically decreasing.
Let = be a reparametrization. It follows from the chain rule that
the tangent vectors of are related to those of through

(u) =

(u)

((u)). (5)
Since

(u) ,= 0 we see that is regular if and only if is regular. Moreover,

if

> 0 the tangent vectors of and have mutual directions, and if

< 0 they have opposite directions. We say in the former case, where is
increasing, that the reparametrization preserves direction and in the latter
case, where is decreasing, that the reparametrization reverses direction.
Example 2.5.1 Let p, q R
n
be xed, q ,= 0. The curve (u) = p+tan uq,
u ]

2
,

2
[, is a reparametrization of the line (t) = p + tq, t R. The
transformation between t and u is given by t = (u) = tan u. On the other
hand, the curve (v) = p+v
3
q is not a reparametrization, since v v
3
does
not have a dierentiable inverse (and in fact, is not regular).
2.6 Reparametrization of surfaces
We will now generalize some of these concepts to surfaces. The situation
is considerably more complicated, because the higher dimensional Euclidean
spaces R
n
present some subtleties which do not show up in case n = 1. In
particular, the theorem given above does not generalize directly to R
n
, as
will be explained thoroughly later in this chapter (in Section 2.10).
Denition 2.6.1. Let U, W R
n
be open sets. A map : W U which is
smooth, bijective and has a smooth inverse is called a dieomorphism.
For example, a linear map L: R
n
R
n
is a dieomorphism if and only if
the n n matrix A that represents it (with respect to some basis for R
n
) is
invertible. If A is invertible, then L is bijective and its inverse is the linear
map represented by A
1
, hence this is a smooth map. If A is not invertible,
then L is not bijective.
The expression (4) for the derivative of the inverse of a map J I, where
I, J R, has the following generalization for a dieomorphism : W U:
D(
1
)(p) = (D(q))
1
where q =
1
(p). Here D is the Jacobi matrix of , and the inverse on
the right side is that of a matrix. This formula follows by application of the
28 Chapter 2
chain rule to the identity
1
= I. In particular, the Jacobi matrix of a
dieomorphism is invertible, that is
det(D(q)) ,= 0
for all q W.
Denition 2.6.2. Let : U R
3
be a parametrized surface, and let : W
U be a dieomorphism (U and W being open sets in R
2
)). The surface
= : W R
3
is called a reparametrization of .
U
(q)
W
q
=

(U) = (W)
2.7 Invariance under reparametrization
A reparametrization of a curve is considered geometrically insignicant
(at least if it is direction-preserving), and geometric properties of curves are
required to be unchanged by such a reparametrization; otherwise they do
not qualify for being geometric. For example, it follows from (5) that the
tangent vector in u of the reparametrized curve diers by a multiple from
that of in t = (u), hence the tangent vector is not geometric. However,
it also follows from (5) that the tangent line is unchanged, hence qualies
better as a geometric object related to the curve. The corresponding result
for surfaces is as follows.
Theorem 2.7. Let = be a reparametrization of . Then the tangent
spaces are identical:
T
q
= T
(q)
, q W.
Moreover, is regular at q, if and only if is regular at (q).
We say that the tangent space is invariant under reparametrization. It
therefore qualies as a proper geometric object related to the surface.
Tangents 29
Proof. It follows from the chain rule that the partial derivatives of are
related to those of through
D(q) = D((q)) D(q),
where the dot denotes matrix multiplication.
Let (u, v) denote the coordinates in U and let (s, t) denote the coordinates
in W. Let
D(q) =
_
a c
b d
_
.
Writing out the above matrix product in terms of the columns

s
and

t
of
D(q) and the columns

u
and

v
of D((q)), it becomes

s
= a

u
+b

v
,

t
= c

u
+d

v
. (6)
These identities show that

s
and

t
are linear combinations of

u
and

v
,
hence they belong to the tangent space T
(q)
. It follows that T
q
T
(q)
.
Since =
1
is also a reparametrization, the same argument with
reversed roles of and shows that T
(q)
T
q
. Thus the equality of the
tangent spaces follows.
Now is regular in q if and only if T
q
is two-dimensional, and is
regular in (q) if and only if T
(q)
is two-dimensional. The equivalence of
the regularity of and follows.
2.8 The unit normal, orientation
Denition 2.8. If is regular at p = (u, v), the vector
N = N(p) =

v
|

v
|
is called the unit normal of the parametrization in p.
(p)
N
Example 2.8.1 It follows from (2) that the unit normal for the unit sphere
with spherical coordinates is
N(u, v) = (u, v),
30 Chapter 2
which is the unit vector pointing from(u, v) towards the center of the sphere.
The unit normal is perpendicular to the tangent plane in p and has unit
length. These properties determine it uniquely up to multiplication with 1.
Let = be a reparametrization as in Theorem 2.7. Since

u
=

v
= 0 and

u
=

v
, it follows from (6) that

t
= (a

u
+b

v
) (c

u
+d

v
)
= (ad bc)

v
,
(7)
where ad bc = det(D(q)) ,= 0. This equation shows that under reparame-
trisation the unit normal is multiplied with the sign of adbc. If adbc > 0
we say that the reparametrization preserves the orientation at (q), otherwise
it reverses orientation. This notion is analogous to the notion of direction of
a parametrized curve.
Example 2.8.2 Let : U R
3
be a parametrized surface, and put W =
(v, u) R
2
[ (u, v) U. The map : W U given by (v, u) = (u, v)
is a dieomorphism, and thus = is a reparametrization. The eect
of this reparametrization is just that it reverses the order of u and v. The
Jacobian of is D =
_
0 1
1 0
_
, which has determinant 1. Therefore
reverses orientation.
2.9 Regular curves as graphs
We have given three general descriptions of plane curves, namely as para-
metrized curves, as graphs of real functions, and as level sets of two-variable
functions. In Section 1.5 it was seen that away from critical points, a level
set is a graph. Conversely, the graph of a function y = h(x) can be realized
as the level set f(x, y) = 0 of the function f(x, y) = h(x) y.
As remarked in Example 2.1.2 it is clear that all graphs are regular
parametrized curves. We shall now establish the converse, that a regular
parametrized curve can be reparametrized as a graph in a neighborhood of
each of its points. This will complete the description of interconnections
between these various types of curves, the conclusion being essentially that
they are all the same.
For simplicity we limit our considerations to plane curves, although a
completely similar result holds for curves in R
3
.
Theorem 2.9. Assume that is a plane curve, regular at t
0
I. Then
there exists a neighborhood of t
0
in which allows a reparametrization as the
graph of a smooth function h, considered either as y = h(x) or as x = h(y).
That is, there exists an open interval I

such that t
0
I

I, an open
interval J and a smooth bijective map : J I

with smooth inverse, such

that
((u)) = (u, h(u))
Tangents 31
for all u J, or
((u)) = (h(u), u)
for all u J.
a
b
t
0
I

x
y
(t
0
)
J
Proof. Write (t) = (x(t), y(t)). The assumption is that (x

(t
0
), y

(t
0
)) ,=
(0, 0). We are going to prove that if x

(t
0
) ,= 0, so that the tangent vector
is not vertical, then the curve allows a reparametrization as a graph of the
form y = h(x). An exchange of x and y then implies that if y

(t
0
) ,= 0, then
the curve allows a reparametrization as a graph of the form x = h(y).
Assume x

(t
0
) ,= 0. By continuity, there exists an open interval I

around
t
0
in which x

(t) ,= 0. Let J = x(t) [ t I

. It follows from Theorem 2.5

that the function t x(t) from I

to J is bijective with a smooth inverse.

When we use this inverse function : J I

for reparametrization we obtain

(u) = ((u)) = (x((u)), y((u))) = (u, h(u)) where h(u) = y((u)).
Example 2.9.1 Let (t) = (cos t, sint) with t R be the standard para-
metrisation of the circle, then

(t) = (sin t, cos t). On the upper half

circle, where t ]0, [, we have x

(t) ,= 0. Then t x(t) = cos t is bi-

jective I

=]0, [] 1, 1[ and has a smooth inverse cos

1
: J I

. The
reparametrization of is then
((u)) = (u, sin(cos
1
u)) = (u,
_
1 u
2
), u J =] 1, 1[.
The following corollary is readily obtained, because the parametrization
t (t, h(t)) of a graph is injective.
Corollary 2.9. A regular parametrized curve is locally injective, that is,
there exist around each t
0
I a neighborhood such that the restriction of
to this neighborhood is injective.
2.10 The inverse function theorem
The following fundamental result from multivariable calculus plays a very
prominent role in dierential geometry. We need it to obtain the analog of
Theorem 2.9 for surfaces.
32 Chapter 2
Theorem 2.10. Let F: U R
m
be smooth, where U R
m
is open, and
let q U be given. Suppose that det(DF(q)) ,= 0. Then there exist an open
set W U containing q and an open set V R
m
containing F(q) such that
V = F(W) and such that the restriction of F is a dieomorphism of W onto
V (see Denition 2.6.1).
y
1
y
2
U
W
q
x
1
x
2
V=F(W)
F(q)
F
Proof. It is convenient to distinguish the variables in the source space and
the target space (both being R
m
) in the way that we view x = F(y) R
m
as a function of y U. The inverse function that we are seeking will then
give y W as a function of x V .
We shall apply the implicit function theorem with n = 2m to the map
f: R
m
U R
m
given by f(x, y) = x + F(y) where x R
m
, y U.
Notice that f(x, y) = 0 if and only if F(y) = x. Therefore, if we can exhibit
the solution set to the equation f(x, y) = 0 as the graph y = h(x) of a
function h, then F(y) = x if and only if y = h(x). This means exactly that
h is inverse to F.
Let y
0
R
m
denote the given point q, and let x
0
= F(y
0
). The matrix
A =
f
y
of Theorem 1.7 is exactly DF(q), hence it has a non-vanishing
determinant. Thus, according to the theorem there exist open intervals I
and J around x
0
and y
0
, respectively, and a smooth map h: I J such that
f(x, y) = 0 if and only if y = h(x), for all (x, y) I J. Let W = J F
1
(I),
then W is open (since F is continuous). It is now seen, as remarked above,
that F: W I and h: I W are the inverse maps of each other. Hence F
is a dieomorphism of W onto V = I.
Remark The present theorem represents an analogue for functions of sev-
eral variables of Theorem 2.5. There is, however, a fundamental dierence
between the two theorems. The theorem we have proved is local, as it only
asserts the existence of an inverse to F in some neighborhood of F(q). Even
if the condition det(DF(q)) ,= 0 holds for all q U, an inverse of F need
not exist on all of F(U). This is illustrated in the example below, and it
contrasts the situation for n = 1: If F

(x) ,= 0 on an interval, then F is

monotone on that interval, hence bijective, as also stated in Theorem 2.5.
Example 2.10.1 Let F: U R
2
be given by
F(x, y) = (x
2
y
2
, 2xy),
Tangents 33
where U = R
2
(0, 0). The Jacobian of F,
DF(x, y) =
_
2x 2y
2y 2x
_
has non-zero determinant for all q = (x, y) ,= (0, 0), hence the inverse func-
tion theorem implies that for each q U, the restriction of F to a suitable
neighborhood of q is invertible. However, since F(q) = F(q) for all q, F
itself is not injective.
Corollary 2.10. Let F: U R
m
be smooth, where U R
m
is open, and
suppose that det(DF(q)) ,= 0 for each q U. Then F(U) is open. If in
addition F is injective, then F is a dieomorphism of U onto F(U).
Proof. Let p F(U) be given, and write p = F(q). According to the theorem
above there exists an open set W U around q such that F(W) open. This
open set F(W) is then an open neighborhood of p in F(U), hence F(U) is
open.
If F is injective, it has an inverse map F
1
: F(U) U. According to the
theorem, F
1
is smooth in a neighborhood p. Since p was arbitrary, F
1
is
smooth.
2.11 Regular surfaces as graphs
In this section we prove the following analogue for surfaces of Theorem 2.9.
Let : U R
3
be a parametrized surface.
Theorem 2.11. Assume that is regular at p U. Then there exists a
neighborhood of p in which allows a reparametrization such that it becomes
the graph of a smooth function , considered either as z = (x, y), y =
(x, z) or as x = (y, z).
As a consequence of this, together with Theorem 1.6, we see, as we saw
for curves in Section 2.9, that there are simple connections between reg-
ular parametrized surfaces, graphs of two-variable functions and level sets
of three-variable functions, away from critical points. Essentially these are
dierent descriptions of the same kind of objects.
Proof. Write
(u, v) = (f(u, v), g(u, v), h(u, v)).
Since is regular at p the columns of the Jacobian
D =
_
_
f

u
f

v
g

u
g

v
h

u
h

v
_
_
34 Chapter 2
are linearly independent at p. By changing the order of the coordinates on R
3
if necessary, we may arrange that the two rst rows of D(p) are independent.
Let : R
3
R
2
denote the projection (x, y, z) (x, y) and put
F = : U R
2
.
Then F(u, v) = (f(u, v), g(u, v)) and
DF =
_
f

u
f

v
g

u
g

v
_
.
It follows that det DF(p) ,= 0. By the inverse function theorem there exists
an open neighborhood W of p in U such that F is a dieomorphism of W
onto the open set V = F(W) = ((W)) R
2
.
Let = F
1
: V W, then = F is the identity map on V ,
that is, the rst two coordinates of ((s, t)) are exactly s and t. We dene
the function (s, t) as the third coordinate of ((s, t)), then ((s, t)) =
(s, t, (s, t)) as desired.
Corollary 2.11. A regular parametrized surface is locally injective, that is,
there exist around each point p U a neighborhood such that the restriction
of to this neighborhood is injective.
2.12 Exercises
1 The following curve is called the cardioid (because of its heart-like shape):
(t) = (2 cos t + cos 2t, 2 sint + sin 2t).
For which values of t is it regular? Find the point where t is singular in
the gure below. The curve is constructed by a circle of radius 1 rolling
without slipping on the outside of a xed circle also of radius 1. The curve
is the trace of a point on the circumference of the rolling circle.
x
y
Tangents 35
2 Let (t) be a parametrized curve which does not pass through the origin,
and let (t
0
) be a point of the trace which is closest to the origin. Show
that the position vector (t
0
) is orthogonal to the tangent vector

(t
0
).
3 Let : I R
3
be a parametrized curve in the xz-plane, that is (u) =
(f(u), 0, g(u)), and assume that f(u) > 0 for all u I. This curve, called
the prole curve, is rotated around the z-axis. The result is a so-called
surface of revolution :
(u, v) = (f(u) cos v, f(u) sinv, g(u)),
y
z
x
a. Explain how the parameter v describes the rotation around the z-axis.
b. Examples: (u) = (1, 0, u) and (u) = (u, 0, u) (the last case requires
u > 0). Describe the corresponding surfaces of revolution.
c. Describe a sphere, minus two poles, as a surface of revolution. Which
is the prole curve, and which coordinates on the sphere are obtained?
d. Assume that is a regular parametrized curve. Show that is a
regular parametrized surface.
4 Let (u, v) = (cos ucos v, cos usin v, sinu), the standard spherical coordi-
nates.
a. Show that the tangent space of at p = (0, ) is T
p
= Span(e
2
, e
3
).
Determine also the tangent space at p = (

4
, 0).
c. Let w = (1, 1, 1). Show that w T
p
where p = (

4
, 0), and determine
a curve on through p and with w as its tangent vector.
5 Consider the surface (u, v) = (u
3
, v
3
, uv), (u, v) R
2
.
a. For which points p = (u, v) is it regular?
b. Determine the tangent space T
p
R
3
for each of the points p
1
=
(1, 0), p
2
= (1, 1) and p
3
= (0, 0). Determine also the tangent plane in p
1
and p
2
. Why not in p
3
?
c. Show that is a bijection of R
2
onto o = (x, y, z) [ xy z
3
= 0.
d. Use Theorem 2.3 to determine the tangent plane in p
1
.
e. The vector v = (3, 6, 3) belongs to T
p
where p = p
2
. Find a curve
on with (t
0
) = (p) and

(t
0
) = v (it exists by Theorem 2.4).
36 Chapter 2
6 Let (u, v) be as in Exercise 5. In each of the following cases, determine
whether can be realized as a parametrized curve on .
a) (t) = (t
3
, t
3
, t
2
), b) (t) = (t
3
, t
3
, t
3
), c) (t) = (t, t
2
, t).
7 Let
(t) =
_
1 t
2
1 +t
2
,
2t
1 +t
2
_
, t R.
Which curve is obtained through the reparametrization = , where
(u) = tan
u
2
for u ] , [?
8 Let (u, v) = (u, v, h(u, v)) be the graph of a function h: R
2
R. Show
that the unit normal is given by
N(u, v) =
(h

u
, h

v
, 1)
_
1 + (h

u
)
2
+ (h

v
)
2
.
9 Let (u, v), (u, v) R
2
be a smooth surface and put (s, t) = (t, s).
Show that is obtained from by a reparametrization. Does it preserve
or reverse orientations?
10 Let (u, v) = (u, uv,
1
2
v
2
), (u, v) R
2
. Determine

u
,

v
and

v
. For
which (u, v) is regular? Determine the unit normal N at (u, v) = (4, 2).
11 Let (u, v) = (u, uv,
1
2
v
2
) for (u, v) U = (u, v) R
2
[ u ,= 0. Show
that (u, v) (u, uv) is a dieomorphism U U, and determine the
inverse map : U U. Show that the reparametrization : U R
3
of
is a graph of the form z = h(x, y), (x, y) U.
12 Let again (u, v) = (u, uv,
1
2
v
2
). Find two open sets U, W R
2
(non-
empty), and a dieomorphism : W U, such that the reparametrization
of [
U
is a graph of the form x = h(y, z), where (y, z) W.
13 Let U = (u, v) R
2
[ u > v and (u, v) = (
1
2
(u +v),
1
2
(u
2
+v
2
), uv) for
(u, v) U. Let p = (2, 0).
a. Show that is regular at p, and determine T
p
.
b. Let W = (s, t) R
2
[ s
2
> t and dene : W R
2
by
(s, t) = (s +
_
s
2
t, s
_
s
2
t).
Show that is a dieomorphism of W onto U, and determine whether it
preserves or reverses orientation.
c. The surface = is the graph of a function. Which?
d. Find q W such that (q) = p, and determine then (q) and T
q
.
Tangents 37
14 In this exercise we identify the set M
2,2
of 2 2 real matrices with R
4
by
numbering the entries in some (arbitrary) fashion. Let F: M
2,2
= R
4

M
2,2
= R
4
denote the map A A
2
where the square is computed by
matrix multiplication. Determine the 4 4 matrix DF(I), where I is the
identity matrix in M
2,2
. Show that every matrix suciently close to I has
a square root, which is unique if it is required to be suciently close to I.
15 Let : U R
3
be an injective and regular parametrized surface, and
assume that its image (U) is contained in the xy-plane. Show that the
set V = (s, t) [ (s, t, 0) (U) is open in R
2
, and that the plane surface
(s, t) = (s, t, 0), (s, t) V , can be achieved as a reparametrization of
(hint: apply Corollary 2.10 to F = in the proof of Theorem 2.11).
38 Chapter 2
Chapter 3
The rst fundamental form
We shall introduce notions that allow us to treat metric questions on
curves and surfaces, for example the determination of the length of a curve
and the area of a subset of a surface. The notion of distance along a curve
will be closely associated with the standard notion of the length of a vector in
Euclidean space R
2
or R
3
. The Euclidean notion of length is used on tangent
vectors, and it can be viewed as a means to dene the distance of very close
points (innitesimal distances). The (global) distance between two points
along the curve is then obtained by integration of these local distances.
Areas are dened similarly by multiple integrals. Certain topics in con-
nection with the latter will be dealt with on a more intuitive level, because
they are most eciently treated by means of the Lebesgue measure, which
we do not assume the reader to be acquainted with. The notion of area will
only be used sporadically in the following chapters, but it is an important
concept in the geometry of surfaces.
3.1 Arc length
Let : I R
n
be a smooth curve. The speed of at t I is dened to
be the length |

(t)| of the derivative

(t) R
n
, in accordance with the
physical interpretation of as describing the motion of a particle in n-space.
In this interpretation

(t) is the velocity vector for the particle at time t.

The vector from (t) to (t + t) is approximately

(t)t, according
to the rst order (linear) approximation of , hence the distance between
these points on the curve is approximately |

(t)|t. Adding up all these

distances and taking the limit t 0, we are lead to the following formula
for the distance along between (t
1
) and (t
2
):
_
t
2
t
1
|

(t)| dt. (1)

The derivation we gave for this formula is not a rigorous proof. Rather than
carrying out such a proof we will take the formula as a denition, and regard
the derivation as motivation.
Denition 3.1. Let : I R
n
be a smooth curve. The arc-length of from
t
1
I to t
2
I is
_
t
2
t
1
|

(t)| dt.
40 Chapter 3
An arc-length function for is a primitive of t |

(t)|, that is, a dier-

entiable function : I R with

(t) = |

(t)|. The arc-length from t

1
to t
2
is then (t
2
) (t
1
). Notice that we do not require t
1
t
2
. If t
2
< t
1
, then
the arc-length is negative.
Example 3.1.1 Let (t) = p +tq be a straight line (where q ,= 0). The arc
length along from p +t
1
q to p +t
2
q is
_
t
2
t
1
|q| dt = |q|(t
2
t
1
) = |(t
2
) (t
1
)|
if t
1
< t
2
.
Example 3.1.2 A circle of radius r is parametrized by (t) = (r cos t, r sin t),
for which the speed |

(t)| = |(r sin t, r cos t)| = r is constant. Hence the

arc-length from 0 to t is
_
t
0
|

(t)| dt = rt.
Example 3.1.3 For the helix given by (t) = (t, r cos(t), r sin(t)) (see
Example 1.1.4) we have

(t) = (, r sin(t), rcos(t)) and the speed

|

(t)| =

2
+r
2

2
is again constant. Hence the arc-length measured
from 0 is this constant times t.
As explained in Section 2.7, reasonable geometric notions are invariant un-
der reparametrizations that do not reverse orientation. The following result
shows that arc length has this property.
Theorem 3.1. Let : I R
n
be a parametrized curve, and let = : J
R
n
be a reparamatrization. Let u
1
, u
2
J and let t
i
= (u
i
) for i = 1, 2.
If preserves the direction then the arc-length of from u
1
to u
2
equals
the arc-length of from t
1
to t
2
. If reverses direction the arc-lengths are
of the same absolute size but have opposite signs.
x
y

t
1
t
2
R

u
1
u
2
R

The rst fundamental form 41

Proof. By the chain rule

(u) =

((u))

(u). Hence
_
u
2
u
1
|

(u)| du =
_
u
2
u
1
|

((u))|[

(u)[ du =
_
t
2
t
1
|

(t)| dt
where in the last step we have used the substitution t = (u). The sign in
front is positive if

is positive, and otherwise negative.

3.2 Lines as shortest curves
Let P
1
, P
2
R
n
. The linear curve from P
1
to P
2
is parametrized by
t P
1
+ t(P
2
P
1
) where t [0; 1]. It has length |P
2
P
1
| (see Example
3.1.1). Its trace, the line segment from P
1
to P
2
, is denoted by [P
1
, P
2
].
The geometric interpretation of the following theorem is that the linear
curve is shortest from P
1
to P
2
. Notice however that because of the possibility
of reparametrization, the linear curve is not unique in this respect.
Theorem 3.2. Let : I R
n
be a parametrized curve. Let t
1
< t
2
in I and
let L denote the arc length of from t
1
to t
2
. Then
L |(t
2
) (t
1
)|. (2)
x
y
(t
1
)
(t
2
)
L
|(t
2
) (t
1
)|
Proof. Let P
1
= (t
1
), P
2
= (t
2
) and w = P
2
P
1
, and consider the function
(t) = (t) w.
We have (t
2
) (t
1
) = (P
2
P
1
) w = |w|
2
, hence by the fundamental
theorem of calculus
|w|
2
=
_
t
2
t
1

(t) dt.
It is easily seen that

(t) =

(t) w, hence

(t) [

(t)[ |

(t)| |w|.
We conclude
|w|
2

_
t
2
t
1
|

(t)| |w| dt = L|w|

from which (2) follows.
42 Chapter 3
3.3 Unit speed parametrization
A parametrized curve is said to have unit speed if |

(t)| = 1 at all
points. It is common practice to replace the symbol for the variable by s in
this case. For a curve with unit speed, the determination of arc-lengths is
particularly simple, because by (1) the arc-length from s
1
to s
2
is equal to the
dierence of the parameters s
2
s
1
. We say that the curve is parametrized
by arc-length.
Theorem 3.3. A regular parametrized curve allows a direction-preserving
reparametrization with unit speed.
Proof. Let (t) be an arbitrary arc-length function for , that is, a primitive
of the speed function t |

(t)|. The speed function is smooth since

(t) is
smooth and never zero. Hence is smooth. Notice that

(t) = |

(t)| > 0.
We apply Theorem 2.5 to the function . It follows that is bijective
onto its image. Furthermore, the inverse function =
1
is smooth, and its
derivative is given by

(s) =
1

(t)
=
1
|

(t)|
> 0
where s = (t). We use the function for the reparametrization. Then
( )

(s) =

((s))

(s) =

(t)
|

(t)|
where t = (s). Hence has unit speed.
Example 3.3.1 For a curve with constant speed c ,= 0, the function
(t) = ct is a primitive of the speed function. The inverse of the map t ct
is (s) =
s
c
, hence a unit speed reparametrization is obtained by inserting
t =
s
c
in the expression for . For example a unit speed reparametrization of
the circle (t) = (r cos t, r sin t) (see Example 3.1.2) is
(s) = (
s
r
) = (r cos
s
r
, r sin
s
r
),
and the helix in Example 3.1.3 is reparametrized with unit speed in
(s) = (
s
c
, r cos(
s
c
), r sin(
s
c
))
where c =

2
+r
2

2
.
The rst fundamental form 43
3.4 The rst fundamental form
Let : U R
3
be a parametrized surface. We dene the following three
functions on U, associated with :
E(p) = |

u
(p)|
2
, F(p) =

u
(p)

v
(p), G(p) = |

v
(p)|
2
,
where p U. These functions together should be seen as the analogue for
surfaces of the speed |

(t)| of a curve (or rather, of |

(t)|
2
).
The functions E, F and G are useful for the computation of lengths of
tangent vectors. If a vector w T
p
has coordinates a, b with respect to the
basis

u
(p),

v
(p), that is w = a

u
+b

v
, then its length is given by
|w|
2
= (a

u
+b

v
) (a

u
+b

v
) = Ea
2
+ 2Fab +Gb
2
.

v
w =
a

u
+b

v
Denition 3.4. The map I
p
: T
p
R that associates to a tangent vector
at p the square of its length,
w I
p
(w) = |w|
2
= E(p)a
2
+ 2F(p)ab +G(p)b
2
,
is called the rst fundamental form of in p. The coecients E, F and G
are called the component functions.
The component functions E, F and G are conveniently arranged as the
entries of a symmetric matrix
_
E F
F G
_
.
By noting that

u
and

v
are the columns of the Jacobian matrix D, we
see that the denition of E, F and G amounts to the matrix identity
_
E F
F G
_
= (D)
t
D (3)
where t denotes transposition. The formula for the rst fundamental form
can also be put in matrix form
I
p
(w) =
_
a
b
_
t
_
E(p) F(p)
F(p) G(p)
__
a
b
_
.
44 Chapter 3
By denition, a quadratic form on a two dimensional real vector space V
with basis vectors v
1
, v
2
is a map Q: V R, which has the form
w = av
1
+bv
2
Q(w) = ea
2
+ 2fab +gb
2
for some numbers e, f, g R. The rst fundamental form I
p
is a quadratic
form on T
p
, for each p U.
Example 3.4.1 For the plane parametrized by (u, v) = p + uq
1
+ vq
2
,
where q
1
, q
2
are linearly independent vectors in R
3
, we have

u
= q
1
and

v
= q
2
. It follows that the component functions are constant:
E = |q
1
|
2
, F = q
1
q
2
, G = |q
2
|
2
.
In particular, if q
1
, q
2
is an orthonormal pair, we have E = G = 1, F = 0.
Example 3.4.2 For the parametrization (u, v) = (r cos v, r sin v, u) of the
cylinder, we obtain

u
= (0, 0, 1) and

v
= (r sin v, r cos v, 0), so that
E = 1, F = 0, G = r
2
.
As before, the component functions are constant.
Example 3.4.3 For the unit sphere with spherical coordinates we deter-
mined

u
and

v
in Example 2.3.1. An easy computation shows that
E = 1, F = 0, G = cos
2
u.
Notice that in this case the component function G(p) is not constant.
The following theorem illustrates how the rst fundamental form enters in
the computation of arc lengths on surfaces. Recall from Section 2.4 that (t)
is called a parametrized curve on if it has the form (t) = (u(t), v(t)) for
a pair of smooth functions with (u(t), v(t)) U, and that in this case (see
Lemma 2.4)

= u

u
+v

v
. (4)
Theorem 3.4. The arc length of a parametrized curve (t) = (u(t), v(t))
on is given with respect to the coordinates (u(t), v(t)) as follows:
_
t
2
t
1
(Eu
2
+ 2Fu

+Gv
2
)
1/2
dt
where the component functions E, F, G are evaluated in (u(t), v(t)) and the
derivatives u

, v

are evaluated in t.
Proof. This is immediate from Denition 3.1, since by (4)
|

(t)|
2
= I
p
(

(t)) = Eu

(t)
2
+ 2Fu

(t)v

(t) +Gv

(t)
2
.
The rst fundamental form 45
Example 3.4.4 On the unit sphere consider the circle (t) = (u, t) with a
xed latitude u. Since u is constant, we have u

= 0, and since v(t) = t, we

have v

= 1.
y
z
x

u
t
(u, t)
With the values of E, F and G from Example 3.4.3 we obtain the total length
of :
_
2
0
(Eu
2
+ 2Fu

+Gv
2
)
1/2
dt =
_
2
0
cos udt = 2 cos u.
The rst fundamental form can also be used to determine the angle be-
tween (non-zero) tangent vectors, say between
w = a

u
(p) +b

v
(p) and w = a

u
(p) +

v
(p)
in T
p
. If the angle is [0, ], then it is well-known from Euclidean
geometry (see Appendix A) that cos =
w w
w w
, from which we obtain
cos =
Ea a +F(a

b +b a) +Gb

b
(Ea
2
+ 2Fab +Gb
2
)
1/2
(E a
2
+ 2F a

b +G

b
2
)
1/2
. (5)
Although not particularly simple this formula allows the computation of
from knowledge of the coordinates a, b, a and

b. In particular, the angle
between

u
and

v
is given by
cos =
F

EG
.
A parametrized surface (u, v) is called orthogonal, if F(p) = 0 for all
p U, or equivalently, if

u
(p) and

v
(p) are perpendicular for all p.
Example 3.4.5 Let (t) be a curve on the unit sphere, which in spherical
coordinates is described by (t) = (u(t), v(t)). We will determine the angle
between the tangent vector

(t) and the direction (North) of the meridians.

46 Chapter 3
y
z
x

The coordinates of

(t) with respect to (

u
,

v
) are determined from (4).
They are a = u

(t) and b = v

(t). The meridians are characterized by having

a xed longitude v, hence the tangent vector of a meridian has direction

u
(with coordinates a = 1,

b = 0). With the values of E, F and G from
Example 3.4.3 inserted in (5) we obtain
cos =
u

((u

)
2
+ cos
2
u(v

)
2
)
1/2
.
3.5 Introduction to areas and plane integrals
In this section we will give a short introduction to the theory of plane
integrals of continuous functions. Not all proofs will be given.
Consider a plane set D R
2
. If D = [a, b] [c, d], where a b, c d, we
call it a rectangle, and we dene that it has the area A(D) = (b a)(d c).
Moreover, in this case if f: D R is continuous we dene the integral of f
over D by
_
D
f dA =
_
b
a
_
d
c
f(u, v) dv du. (6)
It can be shown that the inner integral,
_
d
c
f(u, v) dv, depends continuously
on u, so that the outer integral makes sense. One can also prove that we
have as well
_
D
f dA =
_
d
c
_
b
a
f(u, v) dudv,
that is, the order of the integrations can be interchanged.
If D is not a rectangle, it is more complicated to dene its area, and to
dene integrals over it. By a block-set we will understand a set K which is
a nite union of closed rectangles. Notice that by decomposing further the
rectangles used, such a set K can always be written as a nite union of closed
rectangles, which only overlap on the boundaries. Such a decomposition will
The rst fundamental form 47
be called a partition of the block set. In general, the same block-set may
have several dierent partitions, as in the following gure.
The area A(K) is dened as the sum of the areas of the rectangles in a
chosen partition, and the integral
_
K
f dA of a continuous function f over K
is dened as the sum of the integrals over these rectangles. Since the partition
of K is not unique, a proper treatment would require that it is veried that
these notions are independent of the choice of partition. Intuitively this is
quite clear, and we are not going to verify it here. Notice that it follows from
these denitions that the area of K is the integral over K of the constant
function 1, and that in general
[
_
K
f dA[ A(K) sup
pK
[f(p)[,
since this inequality holds for each of the subrectangles in K. Moreover, if
K
1
, K
2
are block sets which only overlap on their boundaries, then
_
K
1
K
2
f dA =
_
K
1
f dA+
_
K
2
f dA.
We will now consider more general sets D R
2
. In the following de-
nition, we consider smooth curves dened on closed intervals. That is,
: [a, b] R
2
, where < a < b < . This means that is smooth
on (a, b) and that and all its derivatives have continuous extensions to
[a, b] (that is, they have limits for t a from the right and for t b from
the left).
Denition 3.5.1. A set D R
2
is called an elementary domain if it is
closed and bounded, and if its boundary D is a nite union of (the trace
of) smooth curves dened on closed intervals, as above.
x
y
D
D
An elementary domain
48 Chapter 3
In particular a block-set is an elementary domain, since its boundary is a
union of line segments.
Denition 3.5.2. Let D R
2
be an elementary domain. The area of D is
dened by
A(D) = sup
KD
A(K),
where the supremum is taken over all block-sets K D. The integral of a
continuous function f: D R with f(p) 0 for all p, is dened by
_
D
f dA = sup
KD
_
K
f dA.
It should be noticed that the supremums are nite. Since D is bounded,
it is contained in a square of suciently large side length, say N. Hence the
area A(K) of any block-set K inside D is bounded above by the area N
2
of the square, and hence the same bound is valid for the supremum of the
A(K). The integral
_
K
f dA is bounded by A(K) sup
pK
f(p), which in turn
is bounded by A(D) sup
pD
f(p), which is nite since f is continuous. The
same bound is then valid for the supremum in the denition of the integral.
The assumption f 0 is now removed. Let f: U R be continuous, and
put
f
+
(x) = max0, f(x) and f

(x) = max0, f(x),

so that f
+
0, f

0, and f = f
+
f

. We dene
_
D
f dA =
_
D
f
+
dA
_
D
f

dA.
It is easily seen that if D is already a block set, these denitions of area
and integral amount to the same as was already dened. Moreover, plane
integrals share the following familiar properties of ordinary integrals (with
obvious notation), of which we shall give no proof:
_
D
f +g dA =
_
D
f dA+
_
D
g dA
_
D
cf dA = c
_
D
f dA
[
_
D
f dA[
_
D
[f[ dA
_
D
1
D
2
f dA =
_
D
1
f dA+
_
D
2
f dA,
where in the last line D
1
and D
2
are assumed to intersect only with their
boundaries.
The rst fundamental form 49
3.6 Null sets
In this section we will prove a theorem which serves as motivation for the
preceding denitions of area and integrals over an elementary domain D. In
that denition we only considered block sets which were inside D, and the
legitimate question is whether we miss a substantial part of D by this. The
theorem below shows that this is not the case, and thus the denitions are
reasonable.
We say that a closed bounded set D is a null set if for each > 0 there
exists a block-set K of area < such that D K.
As an example, consider a smooth curve : [a, b] R
2
, where < a <
b < . This means that is smooth on (a, b) and that all derivatives have
a continuous extension to [a, b].
Lemma 3.6. Let : [a, b] R
2
be smooth. The trace ([a, b]) is a null set.
Proof. Using the continuous arc-length function s(t), we can divide in N
pieces of equal length /N, where is the total length. Each piece is contained
in the disk of radius /2N centered in the mid-point of the piece (this follows
from Theorem 3.2). Hence the piece is also contained in the square of side
length /N with the same center.
The union of these N squares has area at most N(/N)
2
=
2
/N, which is
for N suciently large.
Since a nite union of null sets is a null set, it follows from the preceding
lemma that the boundary of an elementary domain is a null set.
Theorem 3.6. Let U R
2
be an open set, and let f: U [0, [ be a
continuous function. Let D U be an elementary domain. Then
_
D
f dA = inf
DKU
_
K
f dA (7)
where the inmum is taken over block-sets K.
Proof. We rst observe that there exist block sets K such that D K U.
The proof of this depends on the fact, that D is closed and bounded and U
is open (details are omitted). Thus the inmum on the right is not vacuous.
For later use, we choose a xed block set K
0
with D K
0
U.
50 Chapter 3
k D D K U
If k D and K D are block-sets, then k K and hence
_
k
f dA
_
K
f dA. It then follows from Denition 3.5.2 that
_
D
f dA
_
K
f dA, and
hence the inequality holds in (7).
Let > 0 be given. The boundary D is a null set, according to Lemma
3.6. Hence there exists a block-set L around D with area A(L) .
L
We may assume that L K
0
(otherwise we replace L by its intersection
with K
0
). Let K denote the union D L and let k be the dierence D L
together with its boundary. Then k and K are block-sets with k D K
K
0
, and since
_
k
f dA
_
D
f dA
_
K
f dA we obtain
0
_
K
f dA
_
D
f dA
_
K
f dA
_
k
f dA =
_
L
f dA M
where M = sup
K
0
f. Since was arbitrary, (7) follows.
Thus for functions f as above the integral over D, which was dened by
an approximation from the inside of D, can be approximated as well from
the outside.
3.7 Double integrals
In the preceding section we have dened the notion of a plane integral
over an elementary domain. In the simplest case when the elementary do-
main happens to be a rectangle, the integral was dened by two consecutive
integrals (see equation (6)). In fact a similar formula can be given for a much
larger class of elementary domains.
Let , : [a, b] R be smooth functions with (u) < (u) for u (a, b).
The set
D = (u, v) [ a u b, (u) v (u)
The rst fundamental form 51
of points between the graphs of and ,
u
v
a b
D
v = (u)
v = (u)
is an elementary domain.
Theorem 3.7. The set D has the area
A(D) =
_
b
a
[(u) (u)] du,
and the plane integral of a continuous function f over D is
_
D
f dA =
_
b
a
_
(u)
(u)
f(u, v) dv du
We will not prove this. The formula for the area is well known from
elementary calculus. When it comes to computation of plane integrals in
practice, it is this formula which is used (not the denition given earlier).
More complicated sets are treated by means of a disjoint division into subsets
of this form (possibly with u and v interchanged).
Example 3.7.1 The triangle D = (u, v) [ 0 u, 0 v, 2u + v 2, has
the form as above with inequalities
0 u 1, 0 v 2 2u.
u
v
v = 2 2u
1
2
D
The set D is bounded above and below by the graphs of (u) = 2 2u and
(u) = 0. The area is then
A(D) =
_
1
0
(2 2u) du = 1.
52 Chapter 3
Furthermore, with f(u, v) = v, then
_
D
v dA =
_
1
0
_
22u
0
v dv du =
_
1
0
1
2
(2 2u)
2
du =
2
3
.
Notice that D can also be regarded as a set of the form as before, but with
the inequalities
0 v 2, 0 u 1
1
2
v
(that is, with interchanged roles of u and v).
v
u
u = 1
1
2
v
2
1
D
Of course, the corresponding formulas for the area and the integral lead
to the same results as above,
A(D) =
_
2
0
(1
1
2
v) dv = 1
and
_
D
v dA =
_
2
0
_
1
1
2
v
0
v dudv =
_
2
0
v(1
1
2
v) dv =
2
3
.
3.8 Transformation of integrals
We shall need the important theorem of transformation of plane integrals,
which is a generalization of the formula for substitution of variables in or-
dinary integrals. Let : W U be a dieomorphism (see Denition 2.6.1),
where U, W R
2
are open.
Theorem 3.8. Assume that D R
2
is closed and bounded and contained
in W. If D is an elementary domain, then so is its image (D) U.
Moreover,
_
(D)
f dA =
_
D
(f ) [ det(D)[ dA
for f: U R continuous.
We shall not prove this theorem here. In particular, with f = 1 we obtain
the following formula for the area
A((D)) =
_
D
[ det(D)[ dA.
The rst fundamental form 53
3.9 Surface area
Let : U R
3
be a parametrized surface, and let D R
2
be an elemen-
tary domain, which is contained in U.
u
v
U
D

o = (U)
(D)
Denition 3.9. The area of the surface over D is
A(, D) =
_
D
|

v
| dA. (8)
Recall that

v
is a normal vector to the tangent plane. Its length
can be expressed by means of the rst fundamental form as follows
|

v
| = (EGF
2
)
1/2
. (9)
This identity is an immediate consequence of the following general rule of
vector calculus:
|a b|
2
= |a|
2
|b|
2
(a b)
2
,
(see Appendix C).
We often denote the area by A((D)), although this is not quite legitimate,
because in general the area depends on both and D, and not just their
image (D), unless some injectivity is assumed of .
Notice that if we consider the (x, y)-plane as the surface parametrized by
(u, v) = (u, v, 0), then E = G = 1 and F = 0 (see Example 3.4.1) and hence
(8) reads
A(, D) =
_
D
1 dA,
by which we see that the new notion of area coincides with the previous one
for plane sets.
The denition of area can be motivated by the following geometrical con-
sideration, which is analogous to the motivation that was given for the de-
nition of arc length. Consider a small rectangle in D with (u, v) as its lower
left corner and with sides of length u and v. This rectangle is mapped
approximately to the parallelogram in R
3
placed at (u, v) and with the
54 Chapter 3
vectors u

u
and v

v
as its sides, according to the rst order (linear)
approximation of .
u
v
U
D
v
u

(U)
(D)

v
The area of this parallelogram is
|u

u
v

v
| = |

v
|uv.
Adding up all these areas and taking the limit (u, v) (0, 0) leads to
the formula (8).
Further justication that our denition of surface area is reasonable can
be found in the following theorem, which is analogous to Theorem 3.1.
Theorem 3.9. Surface area is invariant under reparametrization.
Proof. Let = : W R
3
be a reparametrization (see Section 2.5),
and let E U be an elementary domain. Then D =
1
(E) W is an
elementary domain. The statement of the theorem amounts to the identity
A(, D) = A(, E).
Since = we have (D) = (E), and the statement that these
sets have the same area thus appears to be a tautology. However, as we
noted earlier, in the denition (8) of the area, reference is made to both the
parametrization and the domain, not just their image. For the area of (D),
we have
A(, D) =
_
D
|

t
| dA.
The claim is that this equals
A(, E) =
_
E
|

v
| dA.
We have from equation (7) in Section 2.8 that for q W

s
(q)

t
(q) = det(D)(q)

u
((q))

v
((q)).
Inserting this expression in the formula for A((D)) and using the substitu-
tion of variables in Theorem 3.8, we see that A(, D) = A(, E).
The rst fundamental form 55
Example 3.9.1 As an illustration, let us compute the surface area of the
sphere with radius 1. It is, as usual, parametrized by the spherical coordi-
nates (u, v) (see Example 1.2.2), where (u, v) U = R
2
. Let D be the
rectangle where

2
u

2
and v , then is injective on the
interior of D. We found in Example 3.4.3 that the rst fundamental form is
given by E = 1, F = 0 and G = cos
2
u, so that (EG F
2
)
1/2
= cos u. We
therefore obtain the area
_

_
2

2
cos ududv = 4.
3.10 Exercises
1 Let (t) = (3t, 3t
2
, 2t
3
). Show that the speed of the curve is |

(t)| =
3(1 + 2t
2
), and determine the arc length of from t = 0 to t.
2 Let (t) = (t,
4
3
t
3/2
, t
2
) for t > 0. Determine that value t
0
for which the
length of from t = t
0
to t = 1 is equal to the length from t = 1 to t =
3
2
.
3 Let (t) = (t cos t, t sint), t R. The section of the curve where t 0
(drawn below) is called the spiral of Archimedes, because it was described
in a book by Archimedes. Determine the arc length of the curve, measured
from t = 0. The following formula can be used
_
_
1 +x
2
dx =
x
2
_
1 +x
2
+
1
2
ln(x +
_
1 +x
2
) +c.
4 The parametrized curve (t) = (e
ct
cos t, e
ct
sin t), t R, where c > 0 is a
constant, is called a logarithmic spiral. Determine an arc length function
s(t) for , and show that s(t) has a limit s
0
for t . Show that
s(t) s
0
, which can be interpreted as the arc length from () =
(0, 0) to (t), is proportional to |(t)|. This curve appears in nature, for
example in the shape of snail shells. The natural appearance is explained
by proportionality in the growth of the diameter of the shell and the length
of the snail.
x
y
spiral of Archimedes
y
logarithmic spiral
56 Chapter 3
5 Determine a unit speed parametrization of the line through (0, 1, 3) and
(3, 3, 3).
6 Show that the curve (t) = (cos t sin t, sin
2
t,
3
4
t) has constant speed, and
determine a constant k for which the reparametrization t (kt) has unit
speed.
7 Let (t) = (
2
3
cos
3
t,
2
3
sin
3
t) for t R. For which values of t is regular?
Determine a direction preserving reparametrization with unit speed of the
segment where 0 < t <

2
. (Use the formula
_
cos x sin x dx =
1
2
sin
2
x+c.)
8 Let (t) = (e
t
cos t, e
t
sin t), t R, be the logarithmic spiral with c = 1
(see exercise 4). Determine a reparametrization (s), s > 0, with unit
speed such that (s) (0, 0) for s 0 (the solution explains the name
of the curve).
9 Let = : J R
n
be a direction preserving reparametrization of
: I R
n
, where I and J are open intervals, and assume that both curves
and have unit speed. Show that there exists a constant c such that
(s) = s + c for all s J. If I =]a, b[, then what is J? State and prove
similar statements for a direction reversing reparametrization.
10 The surface (u, v) = (ucos v, usinv, av), (u, v) R
2
, with a ,= 0 con-
stant, is called a helicoid. It resembles a (double) spiral staircase. The
following gure shows one winding of the surface (u from 1 to 1, v from
0 to 2)
Show that is regular at all (u, v) R
2
, and that the coecients of the
rst fundamental form are E = 1, F = 0 and G = a
2
+u
2
.
11 Consider a parametrized surface : R
2
R
3
for which E = 1, F = 0,
G = 1+u
2
. Determine the arc length of the curve t (
3
4
,
4
5
t) from t = 0
to t = 1. Determine also the angle between the tangent vector at t = 0 of
this curve and the tangent vector at t = 0 of the curve t (
3
4
+t,
4
5
t).
The rst fundamental form 57
12 Show that the coecients E, F and G for the surface of revolution (see
page 35) (u, v) = (f(u) cos v, f(u) sinv, g(u)) are given by
E = f

(u)
2
+g

(u)
2
, F = 0, G = f(u)
2
a. The curve t (u
0
, t) on , where u
0
is constant, is called a parallel
curve. Show that it has constant speed.
b. The curve t (t, v
0
) on , where v
0
is constant, is called a meridian.
Show that it has unit speed if the prole curve has unit speed.
13 Let (u, v) = (

1 u
2
cos v,

1 u
2
sin v, u) for (u, v) U =] 1, 1[R.
Show that is a regular parametrization of the unit sphere, minus the
two poles. A map of the Earth based on this parametrization is called a
cylinder projection. Explain! Determine E, F and G, and show that
is area preserving, that is, the area A(, D) equals the area of D for all
elementary domains D ] 1, 1[[, ].
14 Draw the following sets in R
2
and verify that they are elementary domains:
a. D = [1, 2] [0, 1].
b. D = (u, v) [ 1 u 2, 0 v u 1.
c. D = (u, v) [ 0 v 1, v + 1 u 2.
d. D = (u, v) [ 1 u 2, 0 v 2 u.
Compute in each case the integral
_
D
udA. Repeat the computations but
with the opposite order of the integrations with respect to u and v.
15 Let (t) = (3t, 4t, 5

1 t
2
) for t ] 1, 1[.
a. Determine a reparametrization of with unit speed. (Use the formula
_
(1t
2
)
1/2
dt = sin
1
t +c, where sin
1
: ] 1, 1[]

2
,

2
[ is the inverse
function of sin: ]

2
,

2
[] 1, 1[.)
b. Let
(u, v) = (3u, 4u, 5
_
1 v
2
),
for u R and 1 < v < 1. Verify that can be realized as a parametrized
curve on , and determine the coecients of the tangent vector

(t) with
respect to the basis (

u
,

v
) for T
(t)
when v ,= 0.
c. Determine E, F and G for , and write down a formula for the area
A(, D) where D is the rectangle D = [0, 1] [0,
1
2
].
16 Let be a surface of revolution (see Exercise 12). Let
D = (u, v) [ a u b, v
and assume that [a, b] is contained in the interval on which the prole
curve is dened.
58 Chapter 3
Verify that the area of over D is given by
2
_
b
a
(f

(u)
2
+g

(u)
2
)
1/2
f(u) du.
Determine the area of the belt on a sphere of radius 1, where the latitude
satises [u[

4
. Determine also the area of the cap, where

4
u

2
.
17 Let denote the graph of a smooth function z = h(x, y), and let D R
2
be an elementary domain. Verify the formula
A(, D) =
_
D
_
1 + (h

x
)
2
+ (h

y
)
2
dA
(assuming that D is contained in the open set where h is dened).
Write down an integral formula for the area of that part of a sphere of
radius 1 and centered at the origin, where [x[ and [y[ both are
1

2
(dis-
regard the assumption above about D). The computation of the integral
is not quite simple. Instead the area can be determined from area of the
cap (see Exercise 16) by a simple geometric consideration. How?
18 The torus is the surface of revolution whose prole curve is the circle
in the xz-plane with radius r and center (R, 0, 0), where R > r). It is
parametrized by
(u, v) = ((R+r cos u) cos v, (R+r cos u) sin v, r sin u).
Determine its total area.
z
x
19 Let (u, v) = (ucos v, usinv, v) (the helicoid, see Exercise 10). Determine
the area of over D = (u, v) [ 0 v 1, v u 1.
Chapter 4
Curvature
In this chapter we introduce and study a quantity, called curvature, which
describes the shape of a curve in a given point. More precisely, it is a mea-
sure of the rate at which the curve is turning in the point. The number is
associated with the second derivative

(t) of a parametrization.
We shall also discuss the curvature of curves on a given surface. In par-
ticular, we introduce the so-called geodesic curvature, which describes the
turning of a curve relative to the given surface containing the curve.
4.1 Curvature of plane curves
Let : I R
2
be a regular parametrized curve.
Denition 4.1. The real number
(t) =
det[

(t)

(t)]
|

(t)|
3
(1)
is called the curvature of at t. Here [

(t)

(t)] denotes the 2 2 matrix

with columns

(t) and

(t).

(t)

(t)
The idea behind the denition is that the turning at t is described by the
position and size of the vector

(t) relative to

(t). This relative position of

the two vectors is described through their determinant, which measures the
area of the parallelogram that they span. For example, if

(t) has the same

direction as

(t), then the curve is not turning at all, and the determinant is
zero. The power 3 in the denominator will be explained shortly by our desire
to have a quantity independent of reparametrization (see Theorem 4.1).
Example 4.1.1 For a straight line with arbitrary parametrization, the vec-
tors

and

will both have the same direction as the line, hence their
determinant is zero. Thus = 0 for a line.
60 Chapter 4
Example 4.1.2 For a circle of radius r with counter clockwise parametriza-
tion (t) = (r cos t, r sin t) we have

(t) = (r sin t, r cos t),

(t) = (r cos t, r sin t)

and
(t) =
det[

(t)

(t)]
|

(t)|
3
=
1
r
.
Similar computations show that the circle with the clockwise parametrization
(t) = (r cos t, r sin t) has curvature =
1
r
.
Example 4.1.3 For an ellipse
(t) = (a cos t, b sint)
we have

(t) = (a sint, b cos t),

(t) = (a cos t, b sint)

and
(t) =
ab
(a
2
sin
2
t +b
2
cos
2
t)
3/2
.
Say for example that a > b. Then attains its maximal value
a
b
2
when
sin t = 0 (where the denominator is minimal), and it attains its minimal
value
b
a
2
when cos t = 0 (where the denominator is maximal).
x
y
minimal
maximal
Example 4.1.4 Let (t) = (t, h(t)) be the graph of a smooth function h,
dened on an open interval I R. Then

(t) = (1, h

(t)) and

(t) =
(0, h

(t)), and we obtain

(t) =
h

(t)
(1 +h

(t)
2
)
3/2
.
In particular if h

(t) = 0 then (t) = h

(t).
Curvature 61
Theorem 4.1. The curvature of a plane curve is unchanged under a direc-
tion-preserving reparametrisation, and it is multiplied by 1 under a direc-
tion-reversing reparametrization.
Proof. Let (u) = ((u)) denote the reparametrization, and let = 1
denote the sign of

. Then

(u) =

(u)

((u)) (2)
and

(u) =

(u)

((u)) +

(u)
2

((u)). (3)
Hence
det[

(u)

(u)] =

(u)
3
det[

((u))

((u))]
and
|

(u)| = [

(u)[ |

((u))|.
By insertion in the denition (1), applied to the curve , we see that the
curvature of at u is ((u)).
Notice that the power 3 in the denominator of (1) was crucial in the
preceding proof.
4.2 Curvature of unit speed curves
For a unit speed curve the expression (1) for the curvature becomes sim-
pler. Notice that unit speed is not a serious limitation because of Theorems
3.3 and 4.1.
Let : I R
2
be a unit speed curve. As usual, the variable is then denoted
by s. Let

(s) denote the normal vector of

(s) (see Appendix C), which is

the unit vector perpendicular to

(s) and pointing to the left.

Theorem 4.2. For a curve with unit speed

. (4)
In particular, it follows that = |

|, where the sign is + if

and

have the same direction, and if they have opposite directions.

Proof. According to the lemma below

(s) is perpendicular to

(s), hence
a scalar multiple of

(s). The scalar is given by

= det[

] = .
This proves (4).
62 Chapter 4
Notice that if > 0 then

and

have the same direction and the curve

turns towards the left, and if < 0 they have opposite direction and the
curve turns to the right.

(s)

(s)

(s)

(s)
positive negative
Lemma 4.2. Let F(t) R
n
be a smooth function of t I R with |F(t)| =
1 for all t. Then F(t) F

(t) = 0 for all t.

Proof. We shall dierentiate the expression
F(t) F(t) = 1.
Observe that the ordinary rule for dierentiation of products also holds for
the dierentiation of a dot product, that is, if f and g are dierentiable maps
I R
n
, then
(f g)

= f

g +f g

.
Applying this rule we obtain F

F + F F

= 0 and hence F F

= 0 as
claimed.
Notice that Theorem 4.2 suggests a way to determine a plane unit speed
curve from its curvature function (s). With (s) = (x(s), y(s)), equation
(4) is equivalent with the system of dierential equations x

= y

and
y

= x

. By solving this system we can determine x

and y

(up to some
constants), and after an integration we obtain x and y (up to further con-
stants). A simple example of this procedure is given in the following proof.
Corollary 4.2. A regular parametrized curve is part of a line if and only if
its curvature is zero everywhere.
Proof. We may assume that the curve has unit speed. Assume that (s) = 0
for all s, then

(s) = 0 by (4). Integrating twice we obtain (s) = p + sq

where p and q are constant vectors.
The statement only if was seen in Example 4.1.1.
Curvature 63
4.3 The tangent angle
Any unit vector w R
2
can be written in the form w = (cos , sin),
where the angle R is determined up to addition of integral multiples of
2. In particular, if w is the tangent direction

(t)/|

(t)| of a regular plane

curve, we call a tangent angle at t. Viewed as a function of t, we call a
tangent angle function.
Example 4.3.1 The parametrized circle (t) = (r cos t, r sin t) has tangent
angle function (t) = t +

2
, because

(t)/|

(t)| = (sint, cos t) = (cos(t +

2
), sin(t +

2
)).
Example 4.3.2 Consider the curve (t) = (t, t
2
), where t R. It has the
tangent vector

(t) = (1, 2t). Since the rst coordinate is positive, we can

determine a tangent angle as (t) = tan
1
(2t).
Because of the ambiguity in the choice of , it is not obvious that a tangent
angle can be chosen which depends smoothly on t. The following lemma,
when applied to w(t) =

(t)/|

(t)|, shows that this is the case.

Lemma 4.3. Let w(t) be a unit vector in R
2
depending smoothly on a pa-
rameter t in an open interval I R. There exists a smooth map : I R
such that
w(t) = (cos (t), sin(t)) (5)
for all t I.
Proof. Write w(t) = (u(t), v(t)) and notice for motivation that if (5) is valid
for some function , then (u

, v

) = (

sin ,

cos ) and hence uv

vu

(cos
2
+ sin
2
) =

.
Choose an arbitrary initial value t
0
I, and an angle
0
such that w(t
0
) =
(cos
0
, sin
0
). Dene a smooth function by
(t) =
0
+
_
t
t
0
uv

vu

dt,
then (t
0
) =
0
and

= uv

vu

. We claim that this function satises

(5). In order to show this identity of unit vectors in R
2
, it suces to show
that w (cos , sin) = 1, since otherwise the dot product would be strictly
smaller (see (A.1)).
From the identity u
2
+ v
2
= 1 we obtain uu

+ vv

= 0. By simple
computations we then derive
(ucos )

= u

cos usin (uv

vu

) = u

cos v

sin
(v sin)

= v

sin +v cos (uv

vu

) = v

sin u

cos .
It follows that (ucos + v sin )

= 0, hence the expression in the bracket is

constant. At t = t
0
its value is 1. Hence w (cos , sin) = 1 as desired.
64 Chapter 4
Theorem 4.3. Assume that (s) is a smooth tangent angle for a plane curve
(s) with unit speed. Then the curvature of at s is given by
(s) =

(s).
Proof. From

(s) = (cos (s), sin(s)) we derive

(s) = (

(s) sin(s),

(s) cos (s))

and (s) = det[

(s)

(s)] =

(s).
Thus the curvature is the rate of change of the tangent angle. In Example
4.3.1 with r = 1 (so that there is unit speed), we have

(t) = 1, which
matches with the curvature 1 of the circle.
4.4 Curvature of space curves
Let : I R
3
be a regular parametrized curve.
Denition 4.4. The non-negative number
(t) =
|

(t)

(t)|
|

(t)|
3
is called the curvature of at t. For a unit speed curve it is
(s) = |

(s)|. (6)
The simpler expression for a curve with unit speed is derived from the fact
that in this case

(s) is perpendicular to the unit vector

(s) (by Lemma

4.2) and hence
|

| = |

|.
Note that |

(t)

(t)| is easily computed by means of Appendix C (iii).

The motivation is similar to the one given in Section 4.1 for plane curves.
It will be shown below that the curvature is unchanged by reparametri-
sation. For a unit speed curve (6) shows that describes the rate of change
of the direction of the curve. Notice that the conclusions of Example 4.1.1
and Corollary 4.2 are valid for space curves as well, with similar proofs.
Notice however that in contrast with the situation for plane curves in
Section 4.1, the curvature of a space curve is always 0. This is related to
the fact that the curvature for a space curve does not contain information
about the direction to which the curve is turning. For a plane curve there
are only two possibilities, left and right, which can be determined by the sign
Curvature 65
of the curvature, but for a space curve there are innitely many possibilities,
and it would be impossible to distinguish them just by a sign.
In connection with this, it should be remarked that if we apply the present
denition to a plane curve, viewed as a space curve in the xy-plane, we obtain
the absolute value of the previous denition. Indeed, if (t) = (x(t), y(t), 0)
then

= (x

, y

, 0) (x

, y

, 0) = (0, 0, det
_
x

_
)
and hence |

| =

det
_
x

.
The notion of curvature for space curves is thus more primitive than that
for plane curves. This is also reected when the following theorem is com-
pared with Theorem 4.1.
Theorem 4.4. The curvature of a space curve is unchanged under repara-
metrisation.
Proof. We use the notation in the proof of Theorem 4.1 (but now applied to
a space curve). It follows from (2) and (3) that

(u)

(u) =

(u)
3

((u))

((u)). (7)
This equation together with (2) implies the theorem.
Example 4.4.1 Let
(t) = (t, r cos(t), r sin(t))
be a helix, as in Example 3.1.3. We nd

(t) = (, r sin(t), rcos(t))

(t) = (0, r
2
cos(t), r
2
sin(t))
with |

(t)| =

2
+r
2

2
. Furthermore

(t)

(t) = (r
2

3
, r
2
sin(t), r
2
cos(t))
with |

(t)

(t)| = r
2

r
2

2
+
2
. Hence
(t) =
r
2
r
2

2
+
2
.
Notice that the curvature is constant, which is reasonable from a geometric
point of view, because the helix has the same shape everywhere.
66 Chapter 4
4.5 Torsion
For space curves with non-zero curvature it is possible to dene a higher
curvature called torsion, which is associated with the third derivative

. It
describes the twisting of the curve. For a plane curve, regarded as a curve
in R
3
, the torsion is 0.
Let : I R
3
be a regular parametrized curve with curvature (t).
Denition 4.5. Let t I and assume that (t) ,= 0. The number
(t) =
det[

(t)

(t)

(t)]
|

(t)

(t)|
2
is called the torsion of at t.
Notice the resemblance of this formula with (1). The denominator is
(t)
2
|

(t)|
6
, which is non-zero by assumption. Motivation for the denition
will be given in the following section.
Example 4.5.1 For a curve which is contained in a xed plane in R
3
, the
three vectors

(t),

(t) and

(t) will all be parallel to this plane. Hence

they are linearly dependent and their determinant is zero. Therefore = 0
(if it is dened).
Example 4.5.2 For the helix of Example 4.4.1 we nd

(t) = (0, r
3
sin(t), r
3
cos(t))
and hence the determinant det[

] is
det
_
_
0 0
r sin(t) r
2
cos(t) r
3
sin(t)
r cos(t) r
2
sin(t) r
3
cos(t)
_
_
= r
2

5
.
Hence
=

r
2

2
+
2
.
Again we obtain a constant, which is reasonable by the same argument as in
Example 4.4.1.
Theorem 4.5. The torsion of a space curve is unchanged under a repara-
metrisation.
Proof. It follows by dierentiation of equation (3) in the proof of Theorem
4.1 that

(u) =

(u)

((u)) + 3

(u)

(u)

((u)) +

(u)
3

((u)). (8)
Curvature 67
From (2), (3) and (8) we see that
det[

(u)

(u)

(u)] =

(u)
6
det[

((u))

((u))

((u))].
The theorem follows from this, combined with (7).
Notice that the torsion of a curve is unchanged also when the direction
of the curve is reversed. The sign of the torsion allows us to separate space
curves with non-zero curvature and torsion in two kinds, right and left.
For example, a helix for which and have the same sign is called a right
helix (compare the thread of a conventional screw) and a helix for which they
have opposite signs is called a left helix.
4.6 The osculating plane and the binormal vector
The geometric signicance of the torsion will now be explained. As be-
fore, let : I R
3
be a regular parametrized curve with non-zero curvature
(t). Then

(t) and

(t) are linearly independent vectors in R

3
. The plane
through (t) with directions spanned by these two vectors is called the oscu-
lating plane. It can be viewed as the plane in R
3
to which the curve comes
closest in the vicinity of (t) (osculare in Latin means to kiss), because of
the Taylor approximation of order two
(t + t) (t) + t

(t) +
1
2
(t)
2

(t),
where the right hand side belongs to the osculating plane for all t. We will
show that the torsion describes the rate of change of the osculating plane.
It follows from equations (2) and (3) that the osculating plane is unchanged
if the curve is reparametrized. Because of Theorem 4.5, we may therefore
assume that the given curve has unit speed. We introduce the notation t(s) =

(s) for the unit tangent vector. Keeping the assumption that (s) ,= 0, let
n(s) =

(s)
|

(s)|
.
This unit vector, called the principal normal, is orthogonal to t(s) by Lemma
4.2. The unit vector
b(s) = t(s) n(s),
which is called the binormal of the curve, is orthogonal to t(s) as well.
Notice that the vectors t(s) and n(s) span the directions of the osculating
plane, and that the binormal b(s) is normal to the osculating plane. It
follows that the rate of change of the osculating plane is expressed by the
size of the derivative b

(s). The following result shows that this is exactly

what the torsion measures.
68 Chapter 4
Theorem 4.6. For a curve in R
3
with unit speed and non-zero curvature
we have
b

= n.
In particular, = |b

|.
Proof. We rst show that b

is proportional to n. For this it suces to show

that it is orthogonal to t and b. That b

b is immediate from Lemma 4.2.

By dierentiation of the equation b t = 0 we obtain that b

t +b t

= 0.
Hence b

t if and only if b t

. By the denition of n we have t

= n,
hence b t

follows from b n.
We thus conclude that b

= cn for some constant c, which we now claim

is . Since

= n we have

= (n)

n +n

. Then
det[

] = (

= (t n) (

n +n

) =
2
(t n) n

.
It follows that
= (t n) n

= b n

. (9)
From b n = 0 we obtain by dierentiation that b

n + b n

= 0, hence
b n

= b

n = cn n = c, and the proof is nished.

We read from Theorem 4.6 that the absolute size of (s) measures the
rate of change of the osculating plane. Moreover, the sign determines the
direction to which the osculating plane is turning, according to the following
rule. Follow the curve with your right hand such that the index nger is in
the tangent direction t and the thumb is in the normal direction n, then if
> 0, the hand will be turning as a right screw (the middle nger, pointing
in direction b, turns towards the thumb n).
Corollary 4.6. A regular space curve with ,= 0 is contained in a xed
plane if and only if = 0 everywhere.
Proof. Assume = 0. From the preceding theorem we have b

(s) = 0, hence
b is a constant vector. Since t(s) b we have

(s) b = 0 for all s. Since

( b)

b we conclude that b is a constant c. Hence (s) belongs to

the plane R
3
[ b = c for all s. The converse implication was seen in
Example 4.5.1.
4.7 The Frenet formulas
The three vectors t(s), n(s), b(s) constitute a positively ordered orthonor-
mal basis for R
3
(depending on s), which is called the moving frame of Frenet
for the curve. We have seen that t

= n and b

= n. It is of interest also
to determine n

. We collect all three formulas in a single theorem.

Curvature 69
Theorem 4.7. For a curve with unit speed and non-zero curvature
t

= n
n

= t +b
b

= n
Proof. Since t, n, b is an orthonormal basis for R
3
we have
n

= (n

t) t + (n

n) n + (n

b) b.
By Lemma 4.2, n

n = 0, and in (9) we saw that b n

= . Finally, from
n t = 0 we derive by dierentiation that
n

t = n t

= n n = .
The formulas in this theorem are called the formulas of Frenet (or of
Frenet-Serret). Since each of the functions t, n and b have three coordinates,
this is essentially a linear system of 9 rst order dierential equations in these
coordinates. By solving this system one can (at least in principle) determine
a curve from its curvature (s) and torsion (s), up to integration constants.
4.8 Curvature of curves on a surface
We will now study some rened notions of curvature for curves which are
contained in a given surface. Let : U R
3
be a parametrized surface, and
let : I R
3
be a parametrized curve on . This means (see Section 2.4)
that = where : I U is a plane curve. Assume that is regular
and that is regular at (t) for all t I. We denote by
N =

v
|

v
|
the unit normal vector of (see Section 2.8), and put
m(t) = N((t)),
the unit normal vector of along .
Denition 4.8. The numbers

g
(t) =
det[

(t)

(t) m(t)]
|

(t)|
3
and
n
(t) =

(t) m(t)
|

(t)|
2
70 Chapter 4
are called, respectively, the geodesic (or tangential) curvature and the normal
curvature of at t with respect to . For a unit speed curve, they are

g
(s) =

(s) u(s) and

n
(s) =

(s) m(s), (10)

where u(s) = m(s) t(s) with t(s) =

(s). The vector u(s) is called the

tangent normal of with respect to .
The formulas (10) for unit speed curves are easily obtained from the gen-
eral denitions. For a unit speed curve on , the three vectors
t(s), u(s), m(s)
again consitute a positively ordered orthonormal basis for R
3
, this is called
the moving frame of Darboux. The rst two basis vectors span the tangent
space T
(s)
. Since in this case

(s) is orthogonal to t(s) =

(s), it follows
from (10) that the decomposition of

(s) according to this basis reads

(s) =
g
(s) u(s) +
n
(s)m(s). (11)
m = N
t
u

(U)

=
g
u +
n
m
Since
g
(s) u(s) T
(s)
and
n
(s)m(s) T
(s)
, this explains the reason
behind the terms tangential and normal curvature for
g
and
n
.
Theorem 4.8. The geodesic curvature
g
is unchanged under a direction-
preserving reparametrisation of , and multiplied by 1 under a direction-
reversing reparametrization. The normal curvature
n
is unchanged in both
cases.
Both
g
and
n
are unchanged under orientation-preserving reparametri-
sations of , and multiplied by 1 under orientation-reversing reparametriza-
tions.
Proof. The statements concerning reparametrization of are easily seen
from (2) and (3), and the statements concerning reparametrization of are
straightforward, since is only represented in the denitions through the
presence of N (see Section 2.8 for the notion of orientation).
Curvature 71
Corollary 4.8. The curvature, the geodesic curvature and the normal cur-
vature of satisfy

2
=
2
g
+
2
n
. (12)
Proof. We may assume the curve has unit speed by the preceding theorem.
The equation then follows from (11) by the theorem of Pythagoras.
Example 4.8.1 A plane curve regarded as a space curve (t) = (x(t), y(t), 0)
(as in Section 4.3) is a curve on the surface (u, v) = (u, v, 0) (the xy-plane).
The normal vector of this surface is N = (0, 0, 1). It is easily seen that
the denition of
g
in this case is identical with the original denition of
curvature of plane curves in Section 4.1, and
n
= 0.
Example 4.8.2 We will compute the curvatures
g
and
n
of a circle on
a sphere of radius 1. Such a curve is the intersection of the sphere and a
plane. It is called a great circle if the plane passes through the center of the
sphere, otherwise a small circle. For simplicity we assume that the plane is
horizontal (this is not a serious restriction, as it can be arranged by a spatial
rotation around the center of the sphere).
y
z
x

u
t
(u, t)
With standard spherical coordinates the circle can be parametrized by
t (t) = (u, t) = (cos ucos t, cos usint, sinu)
with a xed latitude u. The radius of the circle is cos u, hence it has curvature
=
1
cos u
(see Example 4.1.2). We nd

(t) = (cos usin t, cos ucos t, 0),

(t) = (cos ucos t, cos usin t, 0),

and from Example 2.8.1 we have
m(t) = (u, t).
72 Chapter 4
Hence
det[

(t)

(t) m(t)]
= det
_
_
cos usint cos ucos t cos ucos t
cos ucos t cos usin t cos usint
0 0 sin u
_
_
= cos
2
usin u
and

(t) m(t) = cos

2
u.
We conclude that

g
(t) = tan u and
n
(t) = 1.
We can verify
2
=
2
g
+
2
n
by the formula
1
cos
2
u
= tan
2
u + 1.
4.9 Interpretation of normal curvature
A curve which is required to be on a given surface has to follow the shape
of the surface, and is therefore forced to some amount of curvature. The
interpretation of the normal curvature
n
is exactly, that it is this part of
(in the decomposition (12)) which the curve is forced to have by being on .
This interpretation is supported by the following theorem.
Theorem 4.9. Given a point p = (u
0
, v
0
) U and a non-zero vector w
0

T
p
. All parametrized curves = on with (t
0
) = p and

(t
0
) = w
0
have the same normal curvature
n
(t
0
).
Notice that by (12) this common value of
n
is then a lower bound for the
curvature for all such curves.
Part of the proof of the theorem is separated in the following lemma.
Lemma 4.9. Let = be a parametrized curve on and let m(t) =
N((t)) for t I. Let t I be given, and let p = (t) U and (a, b) =

(t) R
2
. Then

(t) = a

u
(p) +b

v
(p) (13)
m

(t) = aN

u
(p) +bN

v
(p). (14)
Proof of the lemma. Equation (13) was established by means of the chain rule
in Lemma 2.4, and Equation (14) is obtained in exactly the same manner.
Proof of the theorem. Since

(t) belongs to the tangent space at (t), we

have

(t) m(t) = 0 for all t. By dierentiation

(t) m(t)+

(t) m

(t) = 0,
from which it follows that

n
(t) =

(t) m(t)
|

(t)|
2
=

(t) m

(t)
|

(t)|
2
. (15)
Curvature 73
From (13) we see that the coordinates a and b are the same for all curves
with tangent vector

(t
0
) = w
0
, and from (14) we then see that m

(t
0
) is the
same for all such curves. It then follows from (15) that
n
(t
0
) is the same
for all such curves.
Example 4.9.1 For circles on the unit sphere we found in Example 4.8.2
that
n
= 1. Since every tangent direction w
0
in every point is the tangent
direction of some circle on the sphere (in fact, of a unique great circle), we
conclude from the preceding theorem that
n
= 1 at all points on all curves
on the sphere.
It follows from Theorem 4.9 that the normal curvature is a property of the
surface rather than of the curve , and the following denition makes sense.
Denition 4.9. Let p and w
0
be as in Theorem 4.9. The normal curvature
of in p with direction w
0
is the normal curvature
n
(t
0
) of any parametrized
curve = on with (t
0
) = p and

(t
0
) = w
0
.
It follows from Theorem 4.8 that the normal curvature of is unchanged
under reparametrizations, except for the sign which changes if orientation is
reversed.
4.10 Geodesics
Denition 4.10. A geodesic on a surface is a regular parametrized curve on
the surface whose geodesic curvature is identically zero.
Thus by (12) together with Theorem 4.9, a geodesic on a surface is a
curve which in each point is as straight as possible, in the sense that it has
the least possible curvature of a curve on the surface in that point and with
that direction. For this reason (among others) we regard geodesics on a
surface as the analogues of straight lines on a plane.
The property of being a geodesic is unchanged under reparametrizations
of as well as , also those which revert direction or orientation (since
g
= 0
if and only if
g
= 0).
Example 4.10.1 It follows from Examples 4.8.1 and Corollary 4.2 that the
geodesics on the xy-plane are the straight lines contained in the plane.
Example 4.10.2 It follows from Example 4.8.2 that great circles on the
unit sphere o = S
2
are geodesics, and that small circles are not. In fact, the
great circles are the only geodesics on the sphere (up to reparametrization
and restriction to subsets). This can be veried as follows. Assume (s) is a
unit speed geodesic on o. From Example 2.8.1 we have m(s) = (s). Since

g
= 0 for a geodesic and
n
= 1 for all curves on a sphere (see Example
4.9.1) we conclude from (11) that

(s) = (s) for all s. Hence

and it follows that det[

] = 0. Hence the torsion is zero, and by

74 Chapter 4
Corollary 4.6 the curve is contained in a xed plane. Being in the intersection
of a plane and the sphere, the curve is contained in a great circle or a small
circle. However, the latter possibility was already excluded.
Theorem 4.10. Let = be a regular parametrized curve on . Then
is a geodesic and has constant speed if and only if

(t) is normal to T
(t)

for all t.
Proof. If has constant speed, a unit speed reparametrization is obtained
by multiplying t with a constant. The second derivative of is proportional
to the second derivative of the reparametrized curve. Hence if the curve is a
geodesic, it follows from (11) that

(t) is normal to the surface.

Conversely, if

(t) is normal to T
(t)
for all t, then

(t)

(t) = 0, and
hence
d
dt
|

(t)|
2
= 0, from which we conclude there is constant speed. After
reparametrization to unit speed we conclude from (11) that
g
= 0.
According to this theorem, a particle which moves on the surface with
no acceleration in the tangent directions of the surface, follows a geodesic.
The only acceleration is that which is necessary to keep the particle on the
surface, and it is normal to the surface.
4.11 Exercises
1 Determine the curvature of the following curves in R
2
:
a. (t) = (2t, t
2
), b. (t) = (e
t
cos t, e
t
sin t), (see page 55).
2 Let (s) be a unit speed curve in R
2
, about which it is assumed that the
curvature is a non-zero constant. Prove that the curve dened by
(s) = (s) +
1

(s)
is a constant curve, that is, a point p. Conclude that the trace of is
contained in a circle centered in p.
3 Let (s) be a unit speed curve in R
2
, and assume that the curvature is
non-zero at s = 0. Let k = (0) and put
C = (0) +
1
k

(0).
Prove that the circle parametrized by
(s) = C +
1
k
(cos(ks)

(0) + sin(ks)

(0)),
satises (0) = (0),

(0) =

(0),

(0) =

(0). Its trace, which has

radius 1/[k[, is called the osculating circle. The center C is called the
center of curvature of at t = 0. See the following gure.
Curvature 75

(0)
(0)
C
4 Let : I R
2
be a regular parametrized curve, and assume that |(t)| has
a local maximum in a given value t
0
I. Prove that [(t
0
)[ 1/|(t
0
)|.
Hint: Assume unit speed. The condition on implies that the second
derivative of t |(t)|
2
is 0 at t
0
. Conclude that (t
0
)

(t
0
) 1
and employ the Cauchy-Schwarz inequality (see Appendix A)
5 Let (s) = (sinh
1
(s),

1 +s
2
). Determine

(s) and show that the curve

has unit speed. Determine

(s) and the curvature (s). Determine a

tangent angle (s), and verify Theorem 4.3 for this curve.
The following formula for the inverse function sinh
1
: R R can be used
d
dy
sinh
1
y =
1
_
1 +y
2
.
6 Let : I R
2
and : I R
2
be two unit speed curves with a common
interval of denition I, and with smooth tangent angles : I R and
: I R. Assume that they have equal curvature (s) for all s I,
and that there exists some value s
0
I for which (s
0
) = (s
0
) and

(s
0
) =

(s
0
). Prove that then (s) = (s) for all s I.
7 Determine the arc length s(t), the curvature (t) and the torsion (t) for
the curve (t) = (3t, 3t
2
, 2t
3
).
8 The curve (t) = (t, cosh t, sinht) is called a hyperbolic helix. Determine
its curvature and torsion.
9 Let (s) = (3 sin
s
5
, 4 sin
s
5
, 5 cos
s
5
). Find t, n and b for this curve. Find
also the curvature and torsion, and show that the curve is contained in a
xed plane. Give a normal vector for this plane.
10 Let : I R
3
be a regular parametrized curve on a regular parametrized
surface : U R
3
. Assume the image of is a (segment of) a straight
line. Prove that is a geodesic on .
11 Let denote the cylinder (u, v) = (cos v, sinv, u) where (u, v) R
2
.
a. Let (t) = (a cos t, t) for t R, where a R is constant. Determine

n
and
g
for . For which value of a is this a geodesic?
b. Instead, let (t) = (at + b, t) for t R, where a, b and are
constants. Describe the curve and show it is a geodesic on .
76 Chapter 4
c. Determine two geodesic curves on which both have end points (1, 0, 0)
and (1, 0, 1), but which have dierent trace in between these two points.
Are there other geodesics between the same two points?
12 Let (u, v) = (f(u) cos v, f(u) sinv, g(u)) be a surface of revolution (see
pages 35 and 57).
a. Show that the meridians t (t, v) are geodesics.
b. Verify that for the parallel curve t (u, t)

g
(t) =
f

(u)
f(u)(f

(u)
2
+g

(u)
2
)
1/2
,
n
(t) =
g

(u)
f(u)(f

(u)
2
+g

(u)
2
)
1/2
.
Give a necessary and sucient condition for it to be a geodesic.
13 Let = : I R
3
be a regular parametrized curve on a regular
parametrized surface . Assume that there exists a xed plane in R
3
containing the image of . If for some t
0
I the plane is orthogonal
to the tangent plane T
(t
0
)
of at (t
0
), we call a normal section of
at this point (two planes in R
3
are orthogonal if their normal vectors are
orthogonal). For example, a great circle on a sphere is a normal section at
each of its points, because it belongs to a plane that intersects orthogonally
with the tangent spaces.
a. Show that a normal section at t
0
has
g
(t
0
) = 0.
b. Use part a to verify Exercise 12a, that the meridians of a surface of
revolution are geodesics. Verify also the geodesics found in Exercise 12b.
14 Let : I R
3
be a unit speed curve with curvature (t) ,= 0 for all t. Let
b(t) be the binormal of the curve at t. Put (u, v) = (v) + ub(v) for
(u, v) R I.
a. Show that is a regular parametrized surface. It is called the binormal
surface of the curve.
b. Show that is a geodesic on the binormal surface.
15 Consider a cone (u, v) = (ucos v, usinv, au), with u > 0 and with a > 0
a xed number. A sphere of radius 1 is inserted in the cone such that it
exactly touches (like a scoop of ice cream in a cone).
a. Determine the center of the sphere, and parametrize the intersection
of the surfaces as a smooth curve.
b. Give an argument, showing that this curve has the same geodesic
curvature
g
and the same normal curvature
n
with respect to the two
surfaces (the sphere is assumed oriented with its normal pointing towards
the center). Determine [
g
[ and
n
.
Chapter 5
The second fundamental form
We shall now extend the notion of curvature from curves to surfaces.
However the description is considerably more complicated, and the curvature
of a surface in a given point p will not be completely described by a single
number. The description of curvature will be based on the concept of normal
curvature (see Denition 4.9), which associates innitely many numbers to
each point p, namely one for each unit tangent vector at p, describing the
curvature of the surface in that direction. One of the central goals of this
chapter will be to organize these numbers in an ecient way.
5.1 The shape operator
In order to treat the notion of curvature eciently, we will use some con-
cepts from linear algebra. The main object that describes the curvature at
p will be a linear map W from the tangent space at p to itself. The map W
is called the shape operator, or the Weingarten map. It will be explained in
Section 5.2 how W relates to the normal curvature of Denition 4.9.
For a plane unit speed curve the description of its curvature is embodied
in the formula t

t (see Theorem 4.2), which expresses that the curvature

is given by the rate of change of the direction t of the tangent line. For
surfaces we will take a similar view, and our denition of W at p will reect
the rate of change of the tangent space at p.
Let : U R
3
be a regular parametrized surface, and let p = (u
0
, v
0
)
U be given. The position of the tangent space T
p
in R
3
is completely
determined by the unit normal vector
N =

v
|

v
|
at p. We will regard the derivative of N as a measure for the curvature of
the surface.
In fact, it will be more convenient to use the negative of the derivative of
N. That this is actually in accordance with the description of the curvature
of a plane curve will be explained below in Example 5.1.3.
However, since N is a function of the two variables u and v, both partial
derivatives N

u
and N

v
have to be taken into account. This could be
done by considering the Jacobian matrix for N: U R
3
, whose columns
78 Chapter 5
are exactly the two vectors N

u
and N

v
, but for reasons to be explained,
we prefer to proceed somewhat dierently.
Observe that it follows from Lemma 4.2 that N

u
and N

v
are perpen-
dicular to N, hence they both belong to the tangent space T
p
.
Denition 5.1. Let p = (u
0
, v
0
) U. The linear map
W = W
p
: T
p
T
p
,
dened by W(

u
) = N

u
and W(

v
) = N

v
, and hence
W(a

u
+b

v
) = aN

u
bN

v
(1)
for all a, b R, is called the shape operator of the surface at p (the derivatives

u
,

v
, N

u
and N

v
are all evaluated in p).
It follows from the fact that the pair (

u
,

v
) is a basis for T
p
, that W
is a well dened linear map T
p
T
p
. The motivation for studying this
map rather than just the vectors N

u
, N

v
is to obtain a notion which behaves
nicely under reparametrizations. The idea is that a reparametrization will
change N

u
and N

v
, but also

u
and

v
, and it turns out that these changes
are always so related that the map remains essentially the same. This will be
seen in the theorem below, and the conclusion is that the shape operator is
more directly related to a geometric property of the surface than the vectors
N

u
and N

v
.
Example 5.1.1 Let (u, v) = p+uq
1
+vq
2
be the plane through p spanned
by two linearly independent vectors q
1
, q
2
R
3
. Then N =
q
1
q
2
q
1
q
2

is con-
stant, and the shape operator W is the zero operator.
Example 5.1.2 For the unit sphere with standard spherical coordinates
(u, v) we have seen in Example 2.8.1 that N(u, v) = (u, v). Hence N

u
=

u
and N

v
=

v
, and it follows that the shape operator W
p
is the identity
operator on T
p
for all p.
Example 5.1.3 Let : I R
2
be a plane curve with unit speed and tangent
vector t(s) =

(s). Since (

t)

=

t

it follows from Theorem 4.2 that

(

t)

=

t

t =

t = t,
where we used that

t = t. Thus it is the negative of the derivative of the

normal vector

t which describes the curvature .
If we view N and

t as analogues of each other, the derivatives of N are
analogous to (

t)

. The analogue of the map W dened by (1) is therefore the

linear map a

a(

t)

(where a R is arbitrary) of the 1-dimensional tan-

gent space to itself. By the equation found above this map is multiplication
by . In this sense W is a higher dimensional version of .
Recall from Section 2.4 that a parametrized curve on by denition is
a curve of the form = where : I U is a parametrized plane curve.
The second fundamental form 79
Lemma 5.1. Let = be a parametrized curve on with (t
0
) = p.
Then
W(

(t
0
)) = m

(t
0
) (2)
where m = N is the surface normal along the curve.
Proof. Immediate from (1) and the two expressions in Lemma 4.9.
In other words, the shape operator associates to a tangent vector w the
derivative of N along any curve on the surface which has w as its tangent.
Theorem 5.1. The shape operator W is unchanged under reparametriza-
tions which preserve orientation, and it changes to W under reparametriza-
tions which reverse orientation.
Proof. Let = : V R
3
denote a reparametrization with dieomor-
phism : V U. We denote the shape operator of by W

, and the
corresponding map for by W

. The claim is that W

= W

.
According to Theorem 2.4 each tangent vector w T
p
is of the form
w =

(t
0
) for some parametrized curve = on . We can then use the
formula (2) to determine W

:
W

(t
0
)) = (N

(t
0
).
The curve = can also be written as = , where =
1
: I
V (see the gure below). Hence may be considered as a parametrized curve
on as well.
U

Hence we can use (2) to determine also W

:
W

(t
0
)) = (N

(t
0
).
The unit normals for and are related by N

= N

: V R
3
with
sign + if and only if preserves orientation (see Section 2.8). Hence
(N

(t
0
) = ((N

) (
1
))

(t
0
) = (N

(t
0
).
The theorem follows immediately.
80 Chapter 5
5.2 The second fundamental form
We shall now introduce another fundamental object through which we
wish to describe the curvature of a surface in a given point. It is closely
related to the shape operator W, and it serves to relate this map to the
normal curvature which was dened in Section 4.9. Let : U R
3
be a
regular parametrized surface, and let p U be given.
Denition 5.2. The map w T
p
II
p
(w) = w W(w) R is called the
second fundamental form of in p.
It follows from Theorem 5.1 that the second fundamental form does not
change under reparametrizations, except by a sign if the orientation is re-
versed.
Let a tangent vector w
0
T
p
be given. Recall from Denition 4.9 that
the normal curvature
n
of in p with direction w
0
is the normal curvature
of any curve on through p with that tangent vector.
Theorem 5.2. The normal curvature in direction w
0
, is

n
=
II
p
(w
0
)
|w
0
|
2
(3)
Proof. Let = be a curve on with (t
0
) = p and

(t
0
) = w
0
. It
follows from Section 4.9, equation (15), that
n
=

(t
0
) m

(t
0
)/|

(t
0
)|
2
.
Hence
n
= w
0
W
p
(w
0
)/|w
0
|
2
follows from (2).
Thus, if we assume |w
0
| = 1, then II
p
(w
0
) is the normal curvature at p of
any curve on , which has the tangent vector w
0
in this point. The relation
(3) describes the geometric content of the second fundamental form.
5.3 Coordinate expressions for the second fundamental form.
In the following theorem we give an explicit expression by which the second
fundamental form can be computed for a given parametrization.
Theorem 5.3. The second fundamental form is given by
II
p
(a

u
+b

v
) = La
2
+ 2Mab +Nb
2
, a, b R, (4)
with respect to the basis

u
,

v
. Here
L = N

uu
=
det[

uu
]
|

v
|
M = N

uv
=
det[

uv
]
|

v
|
N = N

vv
=
det[

vv
]
|

v
|
,
(5)
The second fundamental form 81
where all the terms are evaluated in the given point p U.
Recall the analogous expression I
p
(a

u
+ b

v
) = Ea
2
+ 2Fab + Gb
2
for
the rst fundamental form. The two fundamental forms are quadratic forms
on the tangent space (see page 44)
Proof. We shall derive the following expression for all a, b, a,

b R,
(a

u
+b

v
) W( a

u
+

v
) = La a +M(a

b +b a) +Nb

b. (6)
Taking a = a, b =

b we then obtain (4).
By linearity of W the left side of (6) equals
a a

u
W(

u
) +a

u
W(

v
) +b a

v
W(

u
) +b

v
W(

v
).
The expression (6) follows if we prove

u
W(

u
) = L,

u
W(

v
) = M,

v
W(

u
) = M,

v
W(

v
) = N,
(7)
with L, M and N dened by (5). By denition of W (see (1)),

u
W(

u
) =

u
N

u
,

u
W(

v
) =

u
N

v
,

v
W(

u
) =

v
N

u
,

v
W(

v
) =

v
N

v
.
(8)
From

u
N = 0 we derive by dierentiation with respect to u and v that

uu
N+

u
N

u
= 0 and

uv
N+

u
N

v
= 0, (9)
and from

v
N = 0 we derive similarly that

vu
N+

v
N

u
= 0 and

vv
N+

v
N

v
= 0. (10)
The expressions in (7) then follow from (8) and (5) (since

uv
=

vu
).
The coecients L, M and N are conveniently arranged as the entries of
a symmetric matrix
_
L M
M N
_
,
so that the formula for the second fundamental form can be put in matrix
form
II
p
(a

u
+b

v
) =
_
a
b
_
t
_
L M
M N
__
a
b
_
.
Example 5.3 Let (u, v) = (r cos ucos v, r cos usin v, r sin u) be the stan-
dard parametrization of a sphere with radius r > 0. A straightforward
computation shows that the rst fundamental form has coecients E = r
2
,
F = 0 and G = r
2
cos
2
u. Moreover, N = (cos ucos v, cos usinv, sinu) and
L = N

uu
= r, M = N

uv
= 0, N = N

vv
= r cos
2
u.
Hence the second fundamental form at p = (u, v) is the map
a

u
+b

v
r(a
2
+b
2
cos
2
u).
82 Chapter 5
5.4 A formula for the shape operator
With the aid of the coecients L, M and N we can establish a formula
by which the shape operator can be computed in a given parametrization.
Theorem 5.4. The matrix for the shape operator W
p
: T
p
T
p
with re-
spect to the basis (

u
,

v
) is
_
E F
F G
_
1
_
L M
M N
_
.
Recall that for a 2 2 matrix,
_
a b
c d
_
1
=
1
ad bc
_
d b
c a
_
.
Proof. We express W(

u
) and W(

v
) as linear combinations
W(

u
) = h

u
+j

v
, W(

v
) = i

u
+k

v
with coecients i, j, k, l to be determined. The matrix for W
p
will then be
_
h i
j k
_
.
From (7) we obtain
_
L M
M N
_
=
_

u
(h

u
+j

v
)

u
(i

u
+k

v
)

v
(h

u
+j

v
)

v
(i

u
+k

v
)
_
=
_
Eh +Fj Ei +Fk
Fh +Gj Fi +Gk
_
=
_
E F
F G
__
h i
j k
_
.
Hence
_
h i
j k
_
=
_
E F
F G
_
1
_
L M
M N
_
.
Notice that it does not follow that W
p
is represented by a symmetric
matrix (the product of two symmetric matrices need not be symmetric). In
fact, this will not be the case in general.
Example 5.4 For the sphere with radius r we obtain from Example 5.3
that the matrix for W with respect to

u
and

v
is
_
r
2
0
0 r
2
cos
2
u
_
1
_
r 0
0 r cos
2
u
_
=
_
1/r 0
0 1/r
_
.
The second fundamental form 83
5.5 Diagonalization of the second fundamental form
We shall now introduce a general method by which one can deduce from
the operator W (and its matrix expression) some detailed information about
the shape of a surface. The information is obtained through diagonalization
of W (see Appendix D).
Denition 5.5. An eigenvector for the shape operator W
p
is called a princi-
pal vector in T
p
, and the corresponding eigenvalue is called the correspond-
ing principal curvature at p.
Notice that if w T
p
is a principal vector with unit length and corre-
sponding principal curvature , then by Theorem 5.2 the normal curvature
at p in direction w is

n
= II
p
(w) = w W
p
(w) = w w = .
This explains why is called a curvature.
It follows from Theorem 5.1 that the notion of a principal vector is un-
changed under a reparametrization, and that the corresponding principal
curvatures are unchanged in absolute value, but with the opposite sign if the
orientation is reversed.
We see from (6) that the shape operator W: T
p
T
p
is symmetric with
respect to the dot product, that is
w
1
W(w
2
) = W(w
1
) w
2
(11)
for all w
1
, w
2
T
p
.
Theorem 5.5. There exists for each p U an orthonormal basis w
1
, w
2
for
T
p
consisting of principal vectors with corresponding principal curvatures

1
,
2
R.
With respect to this basis the second fundamental form is given by
II
p
(aw
1
+bw
2
) =
1
a
2
+
2
b
2
(12)
for all a, b R.
Proof. The rst statement follows immediately from Corollary D.1 in Ap-
pendix D with U = T
p
R
3
.
The expression (12) now follows by evaluation of w W(w) with w =
aw
1
+bw
2
.
Corollary 5.5.1. Let w
1
, w
2
and
1
,
2
be as above, and let R. The
normal curvature in direction
w
0
= cos w
1
+ sin w
2
84 Chapter 5
is

n
=
1
cos
2
+
2
sin
2
. (13)
In particular,
n
belongs to the interval between
1
and
2
, which are the
extremal values of
n
.
Proof. It follows from Theorem 5.2 that
n
= II
p
(w
0
). Then (13) is obtained
from (12). Furthermore, if for example
1

2
, then from (13)

n
=
1
cos
2
+
2
sin
2

2
cos
2
+
2
sin
2
=
2
and similarly
n

1
.
The principal curvatures and directions can be explicitly determined by
means of the matrix for W. We summarize the conclusion:
Corollary 5.5.2. The principal curvatures
1
,
2
are the roots of the equa-
tion
det
_
_
E F
F G
_
1
_
L M
M N
_

_
1 0
0 1
_
_
= 0.
The corresponding principal vectors are a

u
+b

v
where
_
a
b
_
is non-zero and
solves
_
E F
F G
_
1
_
L M
M N
__
a
b
_
=
i
_
a
b
_
.
Proof. This follows from the fact that the shape operator is represented by
the matrix
_
E F
F G
_
1
_
L M
M N
_
according to Theorem 5.4.
Example 5.5.1 Let (u, v) = (cos v, sinv, u), then parametrizes a cylinder
(Example 1.2.3). We will determine the principal curvatures and principal
vectors at the point (u, v). We nd

u
= (0, 0, 1),

v
= (sinv, cos v, 0)
and hence E = G = 1, F = 0, and N = (cos v, sinv, 0). Furthermore

uu
=

uv
= 0,

vv
= (cos v, sin v, 0)
and hence
L = M = 0, N = 1.
The matrix of the shape operator with respect to

u
,

v
is therefore
_
1 0
0 1
_
1
_
0 0
0 1
_
=
_
0 0
0 1
_
.
The principal curvatures are the eigenvalues of this matrix,
1
= 0 and

2
= 1. Principal vectors are

u
and

v
since the matrix is already diago-
nal. The normal curvature in direction

u
(vertical) is zero, and the normal
curvature in direction

v
(horizontal) is 1.
The second fundamental form 85
5.6 The graph of a quadratic form
In order to illustrate the theory of the previous section, we will study the
surface formed by the graph of a particularly simple function.
A quadratic form on R
2
is a function q: R
2
R of the form
q(x, y) = ax
2
+ 2bxy +cy
2
(14)
for some constants a, b, c R. It is convenient to write the formula for q in
matrix form
q(x, y) =
_
x
y
_
t
_
a b
b c
__
x
y
_
, (15)
where t denotes transposition. As in the previous section, the key to the
analysis is the diagonalization known from linear algebra. Recall that every
symmetric matrix A is orthogonally diagonalizable, that is, there exists an
orthogonal 2 2 matrix C such that
D = C
1
AC
is a diagonal matrix with real entries (see Appendix D).
We apply the diagonalization to the matrix A of our quadratic form (15).
As explained in Appendix D, the columns of C are chosen as an orthonormal
basis of eigenvectors for A. Let w =
_
x
y
_
R
2
be given. The coordinates of
w, with respect to the basis given by the columns of C, are denoted
_
x

_
.
Then
w = C
_
x

_
.
Write w

=
_
x

_
, then w = Cw

and we obtain from (15) that

q(w) = w
t
Aw = (Cw

)
t
A(Cw

) = w

t
C
t
ACw

= w

t
Dw

since C
t
= C
1
and C
1
AC = D. Let
1
,
2
be the eigenvalues in the
diagonal of D. It follows from the preceding calculation that
q(w) =
1
x

2
+
2
y

2
.
Thus the change of variables from (x, y) to (x

, y

) results in a simplication
of the expression for q, where the product term xy disappears.
Notice that det C = 1, and by changing the sign on one of the columns,
if necessary, we can arrange that det C = 1 (the columns will still be an
orthonormal set of eigenvectors). Then C has the form
C =
_
cos sin
sin cos
_
86 Chapter 5
for some R, and it corresponds to a counterclockwise rotation by the angle
. The basis vectors in the columns of C are obtained from the standard basis
vectors e
i
exactly by this rotation, and the new coordinates x

and y

are the
coordinates of w with respect to the rotated basis.
x
y
x

w
_
x

_
=
_
cos sin
sin cos
__
x
y
_
We have established the following theorem.
Theorem 5.6. Let q(w) = w
t
Aw be a quadratic form on R
2
with symmetric
2 2 matrix A. There exists a rotation of R
2
such that in the rotated x

-
coordinates
q(w) =
1
x

2
+
2
y

2
,
where
1
,
2
are the eigenvalues of A.
In these rotated coordinates we can easily describe the graph of q. Notice
that the vertical cross section of the graph, obtained by taking the intersec-
tion with one of the two vertical coordinate planes (x

z-plane and y

z-plane
respectively), is a parabola (z =
1
x

2
and z =
2
y

2
, respectively). There-
fore the surface is called a paraboloid.
The shape of the horizontal cross sections of the graph depend very much
on the eigenvalues
1
and
2
. If the eigenvalues are both positive or both
negative, then each horizontal cross section of the graph is an ellipse, and
the graph is called an elliptic paraboloid. The graph is shown below in the
positive case (the negative case is similar, but upside down).
elliptic paraboloid
If
1
and
2
are both non-zero but have dierent signs, then the graph
is called a hyperbolic paraboloid, because each horizontal cross section of the
The second fundamental form 87
graph is a hyperbola. In this case the graph has the shape of a saddle, see
below.
hyperbolic paraboloid
If one of the eigenvalues is zero, but not the other one, then the graph
is called a parabolic cylinder (it is a cylinder in which the cross section is
a parabola instead of a circle). Finally, if
1
=
2
= 0 then q is the zero
function and the graph is a plane.
parabolic cylinder
The relation to the theory in Section 5.5 is as follows. In the rotated
coordinates we obtain a graph of the form (u, v) = (u, v,
1
u
2
+
2
v
2
). A
simple calculation shows that at (u, v) = (0, 0) we have
_
E F
F G
_
=
_
1 0
0 1
_
,
_
L M
M N
_
=
_
2
1
0
0 2
2
_
.
We see that the rotation of coordinates exactly has the eect that the shape
operator is diagonalized. The principal curvatures are 2
1
and 2
2
, and
principal vectors are along the two horizontal axes.
Example 5.6.1 To the quadratic form q(x, y) = x
2
+ xy + y
2
corresponds
the symmetric matrix
_
1
1
2
1
2
1
_
which is diagonalized in Example D.1. The diagonalized matrix is
D = C
1
AC =
_
1
2
0
0
3
2
_
where
C =
_
1

2
1

2
1

2
_
.
88 Chapter 5
The quadratic form x
2
+xy +y
2
thus becomes
1
2
x

2
+
3
2
y

2
in rotated coor-
dinates. The graph is an elliptic paraboloid. Its axes are rotated from the x
and y axes by the angle determined from cos =
1

2
, sin =
1

2
, that is,
clockwise by 45 degrees.
5.7 The type of a surface
The principal curvatures and vectors can be explained geometrically as fol-
lows. For simplicity we assume that the given point (p) on the surface is the
origin, and that the tangent plane in this point is exactly the xy-coordinate
plane. This can always be arranged by a suitable translation followed by a
suitable rotation of R
3
, and it can be shown that such a transformation does
not alter
1
and
2
. Furthermore, it follows from Theorem 2.11 (and its
proof) that allows an orientation preserving reparametrization as a graph
over the xy-plane. Observe that the principal curvatures are unchanged also
by such a reparametrization. We therefore assume that is already of this
form, that is
(u, v) = (u, v, h(u, v))
where h(u, v) is smooth.
Since (p) = (0, 0, 0) we have p = (0, 0) and h(0, 0) = 0. Now

u
= (1, 0, h

u
),

v
= (0, 1, h

v
)
and since T
p
is the xy-plane we conclude that h

u
(0, 0) = h

v
(0, 0) = 0. In
particular, we see that the rst fundamental form has
E = G = 1, F = 0
in p. The unit normal vector is N = (0, 0, 1), and since

uu
= (0, 0, h

uu
),

uv
= (0, 0, h

uv
),

vv
= (0, 0, h

vv
)
we obtain from Theorem 5.3 that at p
L = h

uu
(0, 0), M = h

uv
(0, 0), N = h

vv
(0, 0).
The Taylor expansion to order two of now reads (see Appendix B)
(u, v) (0, 0) +u

u
(0, 0) +v

v
(0, 0)
+
1
2
(u
2

uu
(0, 0) + 2uv

uv
+v
2

vv
(0, 0))
= (u, v,
1
2
(u
2
L + 2uvM +v
2
N)) = (u, v,
1
2
II
p
(u

u
+v

v
)).
We thus see that is approximated near p by the graph of
1
2
II
p
, and we can
read o the shape of from the shape of this graph. Since II
p
is a quadratic
The second fundamental form 89
form, its shape was described in Section 5.6. The conclusion is that after a
suitable rotation of the xy-plane, which brings the principal vectors in the
direction of the axes, the surface will have an appearance like one of the
gures in Section 5.6, depending on the signatures of the numbers
1
,
2
.
Denition 5.7. The type of a point p U is dened as follows. It is called
an elliptic point of the surface if the principal curvatures
1
,
2
at p are non-
zero and have the same sign, and a hyperbolic point if they are non-zero with
opposite signs. If one of the principal curvatures is zero, but the other not,
the point is called parabolic, and nally if
1
=
2
= 0 the point is called
planar.
Notice that the type of a point does not change by reparametrization,
since the principal curvatures are either unchanged or both change sign.
5.8 Exercises
1 Let denote the helicoid (u, v) = (ucos v, usinv, v).
a. Determine
g
and
n
for the helix (t) = (a cos t, a sint, t) on . Here
a R is a constant (in the degenerate case a = 0, the helix is a line).
b. Determine W(

(t)), where W is the shape operator for at p = (a, t).

c. Answer the same questions for the curve (t) = (t cos b, t sinb, b) on ,
with b R a constant.
d. Which of the mentioned curves are geodesics on the helicoid?
2 For the helicoid (u, v) = (ucos v, usinv, av), where a ,= 0 is a constant,
the rst fundamental form was determined in Exercise 10, page 56. De-
termine the coecients L, M and N of the second fundamental form.
3 For a surface of revolution (u, v) = (f(u) cos v, f(u) sinv, g(u)) the rst
fundamental form was determined in Exercise 12, page 57. Verify the
following expressions for the second fundamental form:
L =
f

_
(f

)
2
+ (g

)
2
, M = 0, N =
fg

_
(f

)
2
+ (g

)
2
4 Let : U R
3
be a regular parametrized surface. Show that if (U) is
contained in a xed plane x R
3
[ n x = c, where n R
3
is a unit
vector and c R, then L = M = N = 0.
Prove also the following converse. Assume that U is a rectangle ]a, b[]c, d[
and that the second fundamental form is identically 0. Then (U) is
contained in a plane. (Hint: Use (9)-(10) to prove that N is constant.
Prove next that N is constant).
90 Chapter 5
5 Let be a regular parametrized surface for which the image is contained
in a xed sphere x R
3
[ |x a| = r where a R
3
and r > 0.
Show that then rN(u, v) = (u, v) a for all (u, v), and prove that the
fundamental forms are proportional: (rL, rM, rN) = (E, F, G).
Prove also the following converse. Assume that U is a rectangle ]a, b[]c, d[
and that there exists a constant r ,= 0 such that (rL, rM, rN) = (E, F, G).
Then a = + rN is constant and (U) is contained in the sphere with
this center and radius [r[.
6 Consider the parametrized surface (u, v) = (u v, u + v, u
2
+ v
2
) for
(u, v) R
2
.
a. Determine the coecients E, F and G.
b. Let p = (
1
2
,
1
2
). Show that the vectors e
1
= (1, 0, 0) and e
2
+ e
3
=
(0, 1, 1) belong to T
p
, and determine their coordinates with respect to

u
(p),

v
(p).
c. Determine L, M and N at p = (
1
2
,
1
2
).
d. Show that e
1
and e
2
+ e
3
are principal vectors at p = (
1
2
,
1
2
), and
determine the corresponding principal curvatures
1
and
2
.
e. Let (t) = (
1

2
(cos t sin t),
1

2
(cos t + sin t),
1
2
) for t R. Show that
can be realized as a curve on , and determine the curvatures
n
and

g
at t =

4
. One of them coincides with
1
. Explain why.
7 Let (u, v) = (u, v, uv) for (u, v) R
2
and consider the point p = (1, 0).
Compute E, F, G, L, M and N for at p, and determine the normal
curvature of in the direction w
0
= (
1
3
,
2
3
,
2
3
) T
p
.
Determine the principal curvatures and principal vectors for at p.
8 Let (u, v) be a regular parametrized surface. Assume at a given point
(u
0
, v
0
) that F(u
0
, v
0
) = M(u
0
, v
0
) = 0. Show that then

u
and

v
are
principal vectors at this point, with corresponding principal curvatures

1
=
L
E
and
2
=
N
G
.
In the converse direction, show also that if

u
(u
0
, v
0
) and

v
(u
0
, v
0
) are
principal vectors with corresponding curvatures
1
,
2
, which are dierent,
then F = M = 0 at this point. Give nally an example which shows that
this converse conclusion cannot be reached if
1
=
2
.
9 Let F: R
3
R
3
be a map of the form F(x) = Ax + b, where A is an
orthogonal 3 3-matrix with det A = 1, and b R
3
a constant vector
(such a map is called a rigid motion).
Prove that if : U R
3
is a regular parametrized surface, then so is
= F (use Exercise C2 in Appendix C). Verify furthermore that the
coecients E, F, G, L, M, N are equal for and . Verify that if w R
3
is a principal vector for , then Aw is a principal vector for with the
same principal curvature .
The second fundamental form 91
10 Let q(x, y) = 2x
2
+ 4xy + 5y
2
. Determine a rotation of R
2
which brings
q in the form of Theorem 5.6. Of which type is the graph of q? Describe
the level set q(x, y) = 1?
Answer the same questions for q(x, y) = ax
2
+ 24xy + (a + 7)y
2
, for all
possible values of a R.
11 Suppose a quadratic form q(x, y) = ax
2
+3xy +by
2
can be brought to the
form 4(x

)
2
(y

)
2
by a rotation of R
2
. Determine the possible values of
a and b.
12 Consider the graph of h(u, v) = uv cos u cos v, where u, v ] , [.
Show that each point (u, v) ,= (0, 0) is hyperbolic, and that (u, v) = (0, 0)
is parabolic.
13 The graph of h(u, v) = u
3
3uv
2
is called the monkey saddle because the
point (0, 0, 0) is a saddle point with slopes for both two legs and a tail.
Determine E, F, G and L, M, N at (u, v) = (0, 0). Determine also the
principal curvatures
1
,
2
in this point. Which is the type of the point
(0, 0, 0) on the monkey saddle?
14 Let (u, v) = (u + v, v,
1
2
u
2
+ uv + 2v
2
). Compute E, F, G, L, M, N and
the principal curvatures
1
,
2
at (u, v) = (0, 0). Determine also the cor-
responding principal vectors and the type of the point.
15 Find a function h(u, v) of the form h(u, v) = au + bv + cu
2
+ duv + ev
2
,
for which the graph has
E = 5/4, F = 1/2, G = 2, L = 3/4, M = 3/2, N = 3
at (u, v) = (0, 0). Determine the principal curvatures, corresponding prin-
cipal vectors, and the type of the point (0, 0, 0) on the graph.
92 Chapter 5
16 Let : U R
3
be a regular parametrized surface. Let = : I R
3
be
a regular parametrized curve on , and assume the image of is contained
in a straight line. Let
1
and
2
be the principal curvatures for at some
point on the curve, say (t
0
) U where t
0
I. Prove that
1
0
2
or
2
0
1
.
Chapter 6
Teorema egregium
In the investigation of the geometry of surfaces one of the central issues
is to determine which geometric quantities of the surface can be determined
solely on the basis of computations involving measurements of arc lengths on
the surface. Such a quantity is called intrinsic. The point of the notion is
that an intrinsic quantity is an internal property of the surface, independent
of the surrounding space. For example, the distance between two opposite
poles on a sphere of radius 1 is 2, but the shortest distance that can be
measured on the surface is , along a great circle. The distance measured
through the surrounding space is not intrinsic.
In this chapter we will investigate some of the geometric notions we have
introduced from this perspective. Most importantly, we shall prove a famous
theorem of Gauss, which asserts that a particular measure for the curvature,
called the Gaussian curvature, is intrinsic.
6.1 The Gaussian curvature
In the preceding chapter we have described the curvature of a surface in
a given point either by means of a linear map or by means of a quadratic
form, both being rather complicated objects. It would be tempting to try to
reduce to a description by means of a single number. One such number is
the following measure of curvature, which was introduced by Gauss.
Recall, that if U R
n
is an m-dimensional linear space and L: U U
a linear map, the determinant of L, denoted by det L, is dened as the
determinant of the mm matrix that represents L in some basis for U. It is
a theorem of linear algebra that the determinant is independent of the chosen
basis (the matrix will be dierent in another basis, but the determinant will
remain the same).
Denition 6.1. The Gaussian curvature (or total curvature) K(p) of at
p is the determinant of the map W. That is (compare Theorem 5.4)
K(p) = det
__
E F
F G
_
1
_
L M
M N
__
=
LN M
2
EGF
2
.
Notice that the determinant K(p) does not depend on the use of the basis
(

u
,

v
) for T
p
, which is used in the above expression. It follows that K(p) is
94 Chapter 6
unchanged by reparametrizations, since by Theorem 5.1 the shape operator
W is unchanged or changes to W (the latter change does not alter the
determinant).
It will be seen in the examples below that there exist surfaces with quite
dierent shapes, which have the same Gaussian curvature everywhere. There-
fore, the Gaussian curvature does not hold complete information about the
shape of the surface.
Example 6.1.1 For the plane we saw in Example 5.1.1 that W is the zero
operator. Hence its Gaussian curvature is K = 0. For the unit sphere
we determined W to be the identity operator (see Example 5.1.2), and we
conclude that the Gaussian curvature is K = 1. More generally, it follows
from Example 5.4 that the Gaussian curvature of a sphere of radius r is
K = 1/r
2
.
Example 6.1.2 Consider again the cylinder (u, v) = (cos v, sinv, u) from
Example 5.5.1. We will determine the Gauss curvature in the point (u, v).
We saw that E = G = 1, F = 0, and L = M = 0, N = 1. It follows that the
Gaussian curvature is K = 0. Notice that the cylinder and the plane thus
have the same Gaussian curvature, although they have dierent shapes.
The sign of the Gaussian curvature has a particular geometric signicance,
which is explained in the following result.
Theorem 6.1. The Gauss curvature of at p is the product
K(p) =
1

2
.
In particular, is elliptic at p if and only if K(p) > 0, it is hyperbolic at
p if and only if K(p) < 0, and it is parabolic or planar at p if and only if
K(p) = 0.
Proof. With respect to a basis of eigenvectors, the matrix of W is diagonal
with
1
,
2
in the diagonal. The determinant is then the product of these
entries.
We see that although the Gauss curvature K(p) does not give the com-
plete picture, it holds sucient information to determine the type of the
surface, except that it does not permit distinction between parabolic and
planar points.
6.2 Intrinsic geometry
We shall now make the considerations in the introduction to this chapter
more precise.
We can determine lengths of tangent vectors as follows. Let a tangent
vector w T
p
be given. Choose a curve (t) on with w as tangent vector
Teorema egregium 95

(t
0
) = w. Let (t) denote the arc length of from t
0
to t, then this function
is determined by measurements of arc lengths. Since
|w| =

(t
0
)
we conclude that the length of w is intrinsic.
In particular, the coecients E = |

u
|
2
and G = |

v
|
2
of the rst
fundamental form can thus be determined by measuring the arc lengths of
the curves t (t, v) and t (u, t), to which

u
and

v
are the tangent
vectors. By measuring arc lengths along t (t, t), whose tangent vector is

u
+

v
, we can determine |

u
+

v
|, and since |

u
+

v
|
2
= E+G+2F we
can thus determine F as well. Therefore, any quantity that can be expressed
in terms of E, F and G, can also be expressed in terms of lengths of curves.
Conversely, the arc length of a parametrized curve on was expressed by
means of E, F and G, in Theorem 3.4. The property of being expressible
in terms of arc lengths is therefore equivalent with the property of being
expressible in terms of the rst fundamental form.
The following denition is a more concise version of what was explained
above.
Denition 6.2. A quantity or property of a parametrized surface , which
can be expressed purely in terms of the coecient functions E, F and G
of the rst fundamental form for , is called intrinsic. If in addition it is
invariant under reparametrizations of , it is called intrinsic invariant.
As discussed above, the arc length of a parametrized curve on is intrinsic
invariant. Other examples are the angle between tangent vectors (see Section
3.4, eq. (5)) and the area of a subset (see Denition 3.9).
The coecients E, F and G are intrinsic but not invariant, because they
change when the surface is reparametrized. On the other hand, the coordi-
nates in R
3
of (u, v) are not intrinsic since they cannot be determined from
E, F and G alone. To see this, it suces to notice that a translation of the
surface will change these coordinates without changing E, F and G.
The coecients L, M and N of the second fundamental form are not
intrinsic either. For example, we have seen that the plane and the cylinder
can both be parametrized such that E = G = 1 and F = 0, but the second
fundamental forms do not agree.
The shape operator W and the principal curvatures
1
and
2
are invariant
under reparametrization (up to ), but the same example of the plane and
the cylinder shows that they are not intrinsic.
We thus see that being intrinsic invariant is a quite rare property for the
quantities we have introduced to describe surfaces. This is not surprising,
if we compare with the analogue for curves. The corresponding denition
of intrinsic invariant for a quantity related to a curve, say in R
2
, requires
that the quantity can be determined only from the measurement of lengths
96 Chapter 6
along the curve. However, we know from Theorem 2.5 that all curves can
be reparametrized to unit arc length, and hence no curves at all can be
distinguished from each other by means of intrinsic invariants. Remarkably,
we shall see in the following sections that the situation is less hopeless for
surfaces.
6.3 Christoel symbols
We have earlier mentioned that the coecient functions E, F and G are
the analogs for a parametrized surface of the function t |

(t)|
2
for a
parametrized curve. From the latter function one can easily determine the
dot product

(t)

(t), since

(t)

(t) =
1
2
d
dt

(t)

(t) =
1
2
d
dt
|

(t)|
2
. (1)
We will now derive the analog for surfaces of this observation.
In order to express coecients in an ecient way, it is convenient to change
notation and use indices. We number the coordinates u and v by 1 and 2,
thus

1
=

u
,

2
=

v
and

11
=

uu
,

12
=

uv
, etc.
The matrices of components of the two fundamental forms are denoted
_
g
11
g
12
g
21
g
22
_
=
_
E F
F G
_
,
_
b
11
b
12
b
21
b
22
_
=
_
L M
M N
_
,
that is,
g
ij
=

j
, b
ij
=

ij
N. (2)
The analog of (1) is
Lemma 6.3. The expression

ij

k
is intrinsic. It can be determined from
the coecients of the rst fundamental form by means of the following for-
mulas

ij

k
=
1
2
_
g
ik
u
j
+
g
jk
u
i

g
ij
u
k
_
, (i, j, k = 1, 2). (3)
Proof. By dierentiation of g
ik
=

k
we obtain
g
ik
u
j
=

ij

k
+

kj

i
.
Teorema egregium 97
We insert this expression in the right side of (3), with proper permutations
of the symbols. The equality with the left side of (3) is obtained by simpli-
cation with the symmetry rule

ij
=

ji
.
In the following it will be convenient to work with some quantities which
are closely related to the

ij

k
. These are the so-called Christoel symbols.
Denition 6.3. The Christoel symbols associated with are the functions

k
ij
: U R dened for i, j, k = 1, 2 by
_

1
ij

2
ij
_
=
_
g
11
g
12
g
21
g
22
_
1
_

ij

ij

2
_
. (4)
At any given point p U the three vectors

u
,

v
, N (5)
constitute a basis for R
3
, which can be seen as analogous to the moving frame
(t, n, b) of a curve (see Section 4.7), although in general (5) is not orthonor-
mal. The motivation for the symbols
k
ij
is that together with the coecients
b
ij
of the second fundamental form they appear in the representation of

ij
with respect to the basis (5).
Theorem 6.3. Let coecients
k
ij
for i, j, k = 1, 2 be dened as above. Then

ij
=
1
ij

1
+
2
ij

2
+b
ij
N. (6)
Proof. It follows from denition (4) that
g
11

1
ij
+g
12

2
ij
=

ij

1
g
21

1
ij
+g
22

2
ij
=

ij

2
.
Since

k
= g
lk
we then obtain
_

1
ij

1
+
2
ij

2
+b
ij
N
_

k
= g
k1

1
ij
+g
k2

2
ij
=

ij

k
.
On the other hand since N is a unit vector
_

1
ij

1
+
2
ij

2
+b
ij
N
_
N = b
ij
=

ij
N
Thus the vectors on each side of (6) have equal dot products with all elements
of a basis. This implies that they are equal.
The following corollary expresses that the Christoel symbols can be de-
termined from E, F and G. However, they are not intrinsic invariants, since
in general they change when the surface is reparametrized (see Example
6.3.2).
98 Chapter 6
Corollary 6.3. The Christoel symbols
k
ij
are intrinsic. They can be ex-
pressed by a formula which involves only the coecients of the rst funda-
mental form and their (rst order) derivatives with respect to u and v.
Proof. Immediate from (3) and (4).
The actual formula for
k
ij
is somewhat complicated, and the fact that
it exists is more important than its detailed appearance. Let the inverse
matrix of g
ij
be denoted by g
ij
, with superscript indices, then it follows from
equations (3) and (4) that

k
ij
=
1
2

l
g
kl
_
g
il
u
j
+
g
jl
u
i

g
ij
u
l
_
. (7)
If we insert this formula (7) into (6), we obtain an expression for

ij
which
is called the formula of Gauss.
Consider in particular the case where we have an orthogonal parametriza-
tion, that is, where F = 0. In this case the formulas (3) and (7) become
considerably simpler and can be expressed in our original notation of E, F
and G as follows:

11

1
=
1
2
E

u
,

12

1
=
1
2
E

v
,

22

1
=
1
2
G

u
,

11

2
=
1
2
E

v
,

12

2
=
1
2
G

u
,

22

2
=
1
2
G

v
,
and

1
11
=
1
2E
E

u
,
1
12
=
1
21
=
1
2E
E

v
,
1
22
=
1
2E
G

u
,

2
11
=
1
2G
E

v
,
2
12
=
2
21
=
1
2G
G

u
,
2
22
=
1
2G
G

v
.
Example 6.3.1 It follows from the denition in (4) that the Christoel
symbols for a plane (u, v) = p +uq
1
+vq
2
are all zero, since all the second
derivatives

ij
vanish. This can be seen as well from the formulas above,
since E = G = 1 and F = 0 in this case.
Example 6.3.2 Consider the xy-plane with polar coordinates (u, v) =
(ucos v, usinv, 0). Here

u
= (cos v, sinv, 0) and

v
= (usin v, ucos v, 0),
and hence E = 1, F = 0 and G = u
2
. By insertion in the formulas above we
see that the Christoel symbols are
1
11
=
1
12
=
2
11
=
2
22
= 0,
1
22
= u
and
2
12
=
1
u
. In particular, they dier from those of the preceding example.
Teorema egregium 99
6.4 The remarkable theorem of Gauss
The following theorem was found by Gauss in 1827, who described it (in
latin) as egregium, most remarkable. Since then it has become customary
to call it teorema egregium.
Theorem 6.4. The Gauss curvature K is intrinsic, that is, there exists a
general formula expressing K by means of the component functions E, F and
G of the rst fundamental form.
Proof. More precisely, we will show that a formula can be given, which ex-
presses the value of K in a given point by means of the values of E, F and
G and their derivatives (with respect to u and v) up to order 2 in this point.
Since we have already seen (below Denition 6.1) that K is invariant under
reparametrization, the theorem then follows.
We use the notation from the preceding section. Since
K =
det(b
ij
)
det(g
ij
)
(8)
it suces to show that the determinant of the matrix (b
ij
) can be expressed
in terms of the component functions g
ij
and their derivatives.
From the expression (see Theorem 6.3)

ij
=
2

m=1

m
ij

m
+b
ij
N
we obtain by dierentiation with respect to u
k

ijk
=
2

m=1
(

m
ij
u
k

m
+
m
ij

mk
) +
b
ij
u
k
N+b
ij
N

k
.
It follows that

ijk

l
=
2

m=1
(

m
ij
u
k
g
ml
+
m
ij

mk

l
) +b
ij
N

l
,
and since
N

l
= N

lk
= b
lk
(see Section 5.3, (9)-(10)) we obtain

ijk

l
=
2

m=1
(

m
ij
u
k
g
ml
+
m
ij

mk

l
) b
ij
b
lk
.
100 Chapter 6
We introduce the abbreviation R
jkil
, called the Riemann symbol, for the
dierence
R
jkil
=
2

m=1
(

m
ij
u
k
g
ml
+
m
ij

mk

l
)
2

m=1
(

m
ik
u
j
g
ml
+
m
ik

mj

l
), (9)
where the two sums only dier by j and k being interchanged.
Then since

ijk
=

ikj
we conclude that
R
jkil
b
ij
b
lk
+b
ik
b
lj
=

ijk

ikj

l
= 0,
hence
R
jkil
= b
ij
b
lk
b
ik
b
lj
.
In particular,
R
1212
= det(b
ij
). (10)
The Riemann symbol R was introduced as an abbreviation for an ex-
pression involving the quantities
m
ij
, g
ij
and

ij

k
(with various indices
i, j, k, m). Hence it follows from Lemma 6.3 and Corollary 6.3 that R can
be expressed by means of the g
ij
. An inspection shows that derivatives up
to order 2 are involved. According to (10) this implies the statement of the
theorem.
From the equation (10) one can derive an explicit, but quite complicated,
expression for the Gauss curvature in terms of the coecients of the rst
fundamental form. If F = 0 it becomes considerably simpler, and reads
K =
1
2

EG
_
_
G

EG
_

u
+
_
E

EG
_

v
_
. (11)
The verication of this formula is a long but straightforward computation
based on (8), (9), (10) and the formulas given in the end of Section 6.3.
6.5 Isometries
A useful interpretation of the notion of intrinsic geometry is obtained
from the concept of isometries of surfaces. Basically, an isometry from one
surface to another is a distance-preserving map. The denition is simplest
for parametrized surfaces that have a common domain U, so we shall start
by considering this situation.
Denition 6.5.1. Let : U R
3
and : U R
3
be parametrized surfaces
dened on a common open set U R
2
. Then and are said to be isometric
if their rst fundamental forms are equal, that is if
E

= E

, F

= F

, G

= G

.
Teorema egregium 101
In order to provide some intuition assume temporarily that is injective.
In this case we can dene a map
: (U) (V ), ((p)) = (p), (p U).
When the surfaces are isometric, this map is said to be a bending of one sur-
face to the other, because the deformation (without stretching) for example
of a piece of paper, provides an example.
Example 6.5.1 Let (u, v) = (u, v, 0) and (u, v) = (cos v, sinv, u) be the
plane and the cylinder, both dened on U = R
2
. Then and both have
E = G = 1, F = 0, hence they are isometric. In this case the bending
: (U) (U) corresponds to the folding of a cylinder from a plane piece
of paper.
We now turn to the general situation of parametrized surfaces dened
on dierent domains, say : U R
3
and : V R
3
. We assume that a
dieomorphism : U V is given. Then : U R
3
is a reparametrization
of (see Section 2.6) with the same domain U as .
u
v
U V

Denition 6.5.2. The dieomorphism : U V is said to induce an isom-

etry from to , if and are isometric, that is, if
E

= E

, F

= F

, G

= G

. (12)
It is important to stress that the condition expressed in (12) is that
should have the same rst fundamental form as , but after the reparametri-
zation by .
102 Chapter 6
Note that a reparametrization of a surface is a trivial case of an isometry.
More precisely, if = is a reparametrization of , then induces an
isometry from to , since obviously is isometric to itself. The purpose of
the more involved Denition 6.5.2, compared to the previous one, is exactly to
get a notion of isometry that takes possible reparametrizations into account.
If in Denition 6.5.2 we assume that is injective, then
: (U) (V ), ((p)) = ((p)), (p U)
is well-dened. Intuitively it is this bending, called the lift of , which is the
isometry induced by . It takes place between the images of the parameter
sets. However, if is not injective, then the construction of may not
be possible. Dierent elements p and q in U may have (p) = (q) but
((p)) ,= ((q)), so that is not well dened. For example, the unfolding
from cylinder to plane, which is inverse to the bending in Example 6.5.1, is
really only a well-dened map on the level of the parameter sets, since the
same point on the cylinder corresponds to more than one point in the plane.
Notice that by Corollary 2.11, a regular parametrized surface is injective
in some neighborhood of each parameter point p U, so that the lifting
can be done in that neighborhood.
Example 6.5.2 Let and both denote the sphere of radius 1, both
parametrized by spherical coordinates as in Example 1.2.2, with domains
U = (u, v) [ /2 < u < /2, < v <
for and
V = (s, t) [ /2 < s < /2, + < t < +
for . Here R is some constant. The map U V dened by (u, v) =
(u, v + ) induces an isometry from to . This follows from the fact that
E, F and G are independent of v (see Example 3.4.3). The corresponding
lift is the rotation of the sphere around the z-axis by the angle .
It can be shown (see Exercise 8), that if induces an isometry from to ,
then
1
induces an isometry from to . Moreover, if in addition a third
parametrized surface : W R
3
is given, together with a dieomorphism
: V W inducing an isometry from to , then induces an isometry
from to .
The most important observation in connection with the concept of isom-
etry is that the agreement of the rst fundamental forms, as expressed by
(12), ensures that all intrinsic quantities are preserved. In particular, this
explains the term isometry, since length is intrinsic. The fact that length
is preserved is expressed more precisely in the following lemma.
Let = : I R
3
be a parametrized curve on , and assume
induces isometry from to , as above. By = : I R
3
we dene
a parametrized curve on , said to be the image of by (see the gure
below).
Teorema egregium 103
Lemma 6.5. When induces an isometry the arc lengths of and are
equal. That is, let t
1
, t
2
I then the arc length of from t
1
to t
2
is equal to
the arc length of from t
1
to t
2
.
Proof. Let (t) = (u(t), v(t)) denote the coordinates of (t) in the parametri-
zation = by means of . The arc length of is expressed in Theorem
3.4 by means of the functions u(t) and v(t) together with E

, F

, G

.
u
v
U

R
I

Writing = ( ) we can regard as a parametrized curve on .

When we regard in this fashion, its coordinates (u(t), v(t)) are those of
(t), that is, they are the same as before. Applying Theorem 3.4 once more,
but this time to on , we obtain an expression for the arc length of
by means of u(t) and v(t) together with the coecients E

, F

, G

of the rst fundamental form of . Hence if induces an isometry, the

expression is exactly the same as before, and the arc lengths on and
agree.
Theorem 6.5. Assume that : U V induces an isometry from to .
Then the Gauss curvature of in p is equal to the Gauss curvature of in
(p), for all p U.
We say that the Gauss curvature is invariant under isometries.
Proof. By Theorem 6.4 the Gauss curvature in p can be expressed by means
of the functions E, F and G and their derivatives in p. Hence the Gauss
104 Chapter 6
curvatures K

and K

for and are identical functions on U,

K

(p) = K

(p), p U.
It was observed in Section 6.1 that the Gauss curvature is unchanged by
reparametrizations, hence
K

(p) = K

((p)).
Example 6.5.3 Let : U = (u, v) [ u > 0 R
3
be the parametrization
(u, v) = (ucos v, usinv, u)
of a cone (see Example 1.2.4) and let : V = (r, ) [ r > 0 R
3
be the
parametrization by polar coordinates
(r, ) = (r cos , r sin , 0)
of the xy-plane (without (0, 0, 0)).
For each constant k > 0 the map (u, v) = (ku, v/k) is a dieomorphism
of U to V , since it is smooth and bijective with the smooth inverse (r, )
(u, v) = (r/k, k).
The component functions of the rst fundamental form for are E =
1 +
2
, F = 0 and G = u
2
. The reparametrization
(u, v) = (kucos(v/k), kusin(v/k), 0)
of has components

E = k
2
,

F = 0 and

G = u
2
. Therefore, induces an
isometry of to if and only if k
2
= 1 +
2
.
The conclusion from the theorem above is then that if k
2
= 1+
2
then the
cone and the plane have the same Gaussian curvature in points (u, v) and
(ku, v/k) (in fact, both Gaussian curvatures are zero, as we knew already).

2/k

u
v

r
/k
/k

Teorema egregium 105

Notice that the comparison of component functions took place between
those of and those of , whereas those of itself played no role.
The map : (u, v) (r, ) = (ku, v/k) between parameter values is lifted
to the map
: (u, v) (ku, v/k)
from cone to plane. However, is well dened only if we restrict v to an
open interval of length 2, since we have (u, v + 2) = (u, v) but in
general ((u, v + 2)) ,= ((u, v)). The map can be described as the
unfolding of the cone.
Example 6.5.4 Since the sphere has Gauss curvature dierent from zero
in all points, we can conclude from the Gauss theorem that no portion of
a sphere can be mapped isometrically into a plane. In other words, it is
impossible to draw a map of a portion of the globe, which preserves all
lengths (in appropriate units). Such a map is called an ideal map, and its
non-existence is a theorem originally due to Euler.
Example 6.5.5 Let (u, v) = (a cosh ucos v, a coshusin v, au) for (u, v)
U = R
2
, where a > 0 is a constant. This surface is called a catenoid (it is a
surface of revolution, see page 35).
y
z
x
For the second surface let (s, t) = (s cos t, s sint, at) where (s, t) V = R
2
.
This surface is called a helicoid (see page 56). We shall verify that the map
(u, v) = (a sinhu, v) induces an isometry from to . It is a dieomorphism
since sinh: R R is bijective with a smooth inverse (by Theorem 2.5). An
elementary computation shows that the rst fundamental forms for and
for are given by E = G = a
2
cosh
2
u and F = 0 in both cases. Hence
induces an isometry.
Notice that the catenoid is not injective. If we restrict to the subset
(u, v) [ < v < of U, then is injective. The image by of this set
corresponds to one winding of the helicoid.
106 Chapter 6
6.6 Exercises
1 Verify the following formula for the Gauss curvature of a surface of revo-
lution
K =
(f

)g

(f
2
+g
2
)
2
f
.
Show that if the prole curve has unit speed, then K =
f

f
.
2 The plane curve (t) = (sint, cos t + lntan
t
2
), where 0 < t < is called
the tractrix.
x
y
t = 0
t =
t =

2
Show that the curve is regular for t ,=

2
. The surface of revolution
y
z
x
(u, v) = (sin ucos v, sinusinv, cos u + ln tan
u
2
), 0 < u < , v R
is called a pseudosphere. Verify that K = 1 everywhere, except at u =

2
,
(so that resembles a sphere of radius 1, which has constant K = 1).
3 Compute the coecients L, M and N for the surfaces and in Example
6.5.5 (see also Exercises 2 and 3, page 89), and use these to determine
their Gauss curvatures. Verify the Teorema Egregium for these surfaces.
Teorema egregium 107
4 Show that the surface of revolution (s, t) = (s cos t, s sint, a lnt), where
(s, t) U = (s, t) [ t > 0 and a > 0 is constant, has the same Gauss
curvature K(s, t) as the helicoid in Exercise 3, restricted to U. Never-
theless, the rst fundamental form of is dierent. Does this contradict
the Teorema Egregium?
5 a. Let three numbers e, f, g R with eg > f
2
and e, g > 0 be given.
Prove that there exists a regular parametrized surface : R
2
R
3
, the
image of which is the xy-plane, such that E(u, v) = e, F(u, v) = f and
G(u, v) = g for all (u, v) R
2
. Hint: Try a linear map : R
2
R
3
.
b. Let next : U R
3
be an arbitrary regular parametrized surface for
which E, F and G are constant. Prove that there exists a dieomorphism
which induces an isometry from to a parametrized surface of which the
image is contained in the xy-plane.
6 a. Let 0 < a < 1 and let
f(s) = a cos s and g(s) =
_
s
0
_
1 a
2
sin
2
r dr.
Verify that the curve (s) = (f(s), g(s)) has unit speed.
b. Let (s, t) = (f(s) cos t, f(s) sint, g(s)), be the surface of revolution
with prole curve , where
(s, t) V = (s, t) [

2
< s <

2
, < t < .
Furthermore, let (u, v) denote the part of a unit sphere with standard
spherical coordinates (Example 1.2.2), for which the domain of denition
is reduced to U = (u, v) [

2
< u <

2
, a < v < a, that is, a
segment on the back has been removed.
y
z
x
(u, v)
y
z
x
(s, t)
Show that the map : U V given by (u, v) = (u, v/a) induces an
isometry (for a =
1
2
one can visualize by the bending of a half sphere,
for example the peel of half an orange). What can one conclude about the
curvature of ?
108 Chapter 6
7 Let (u, v) = (u, v +
c
u
) for (u, v) U = (u, v) R
2
[ u > 0, where
c R is a constant.
a. Let : U R
3
be a parametrized smooth surface, and let = .
Verify

u
(u, v) =

u
((u, v))
c
u
2

v
((u, v)),
and determine a similar expression for

v
(u, v).
b. Assume that the rst fundamental form for is given by
E = 1 +v
2
, F = uv, G = u
2
for (u, v) U. Show that induces an isometry from to itself (that is,
take V = U and = in Denition 6.5.2).
c. Without explicitly computing the Gauss curvature K(u, v) of , show
that it does not depend on v (hint: use that c was arbitrary).
8 Prove the following statements (see page 102) by applying the chain rule
and the identity (3) in Section 3.4:
a. If induces an isometry from to , then
1
induces an isometry
from to .
b. If in addition a third parametrized surface : W R
3
is given, together
with a dieomorphism : V W inducing an isometry from to , then
induces an isometry from to .
Chapter 7
Geodesics
In this nal chapter we investigate some properties of geodesics. Recall
from Denition 4.10, that a geodesic on a surface is a curve with zero geodesic
curvature. We shall see that the property of a curve, that it is geodesic, is
intrinsic. Furthermore we introduce the notion of geodesic coordinates on a
surface, and we use these to give a geometric interpretation of the theorem of
Gauss. Some of the results presented in this chapter require more advanced
analytic tools than we have presupposed in the rest of the notes, and we shall
be content with stating them without proof.
7.1 The geodesic equations
We aim to show that the absolute value [
g
(t)[ of geodesic curvature is an
intrinsic property of a curve on a surface. It is invariant under reparametriza-
tions by Theorem 4.8 (but notice the necessity of taking the absolute value).
Theorem 7.1. Let = be a regular parametrized curve on . The
geodesic curvature
g
(t) satises

g
= |

|
3
det(g
ij
)
1/2
_
(u
1
)

2
(u
2
)

1
_
where g
ij
is the rst fundamental form of at (t), u
1
, u
2
are the coordinates
of (t) and (u
1
)

, (u
2
)

are their derivatives with respect to t, and where

i
denotes the function

i
(t) = (u
i
)

(t) +
2

j,k=1

i
jk
((t)) (u
j
)

(t)(u
k
)

(t), i = 1, 2,
for i = 1, 2, in terms of the Christoel symbols
i
jk
.
In view of Corollary 6.3, we see that the expressions
i
can be determined
from E, F and G. Hence it follows from the equation above for
g
, that it too
can be determined. Hence the absolute value [
g
[ is intrinsic. In particular,
it follows that the property of being a geodesic curve is intrinsic.
The proof invokes two lemmas, which are stated and proved on the fol-
lowing page.
110 Chapter 7
Proof. By denition
g
= |

(t)|
3
det[

m]. Recall from Lemma 2.4

that

(t) = u

(t)

u
+v

(t)

v
. (1)
where the tangent vectors

u
and

v
are evaluated in (u(t), v(t)). The second
derivative

is determined in Lemma 7.1.1 below, and the determinant can

then be computed by means of Lemma 7.1.2, where we take w

= u

1
+u

2
and w

=
1

1
+
2

2
(since a multiple of m in

does not contribute to

the determinant). The equation for
g
follows.
Lemma 7.1.1. Let and
1
,
2
be as above. Then

equals
1

1
+
2

2
plus a multiple cm of m.
The factor is c =

b
jk
(u
j
)

(u
k
)

but we do not need this formula.

Lemma 7.1.2. Let be as above and let two vectors w

, w

T
(t)
be given.
If w

= a
1

1
+a
2

2
and w

= b
1

1
+b
2

2
then
det[w

m] = det(g
ij
)
1/2
(a
1
b
2
b
1
a
2
)
Proof of Lemma 7.1.1. In order to determine

(t) we dierentiate (1). For

this we need to dierentiate

u
and

v
with respect to t.
We apply the chain rule to the function t

u
(u(t), v(t)). It follows that
d
dt

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

(t)

uu
+v

(t)

uv
.
Similarly
d
dt

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

(t)

vu
+v

(t)

vv
.
Hence

(t) = u

(t)

u
+u

(t)
d
dt

u
+v

(t)

v
+v

(t)
d
dt

v
= u

(t)

u
+u

(t)(u

(t)

uu
+v

(t)

uv
)
+v

(t)

v
+v

(t)(u

(t)

vu
+v

(t)

vv
)
=

i
u

i
+

jk
u

j
u

jk
.
We use the expression (6) from Theorem 6.3 and insert it for

jk
. It follows
that

i

i

i
+cm for the number c mentioned below the lemma.
Proof of Lemma 7.1.2. By a straightforward computation
det[w

m] = (w

) m = (a
1
b
2
b
1
a
2
) (

2
) m.
The lemma follows since by (9) page 53

2
= |

2
| m = (EGF
2
)
1/2
m.
Geodesics 111
Corollary 7.1. Let (s) = (u
1
(s), u
2
(s)) be a regular parametrized smooth
curve on . Then is a geodesic and has constant speed if and only if the
coordinate functions u
1
and u
2
satisfy the following system of second order
dierential equations
(u
i
)

+
2

j,k=1

i
jk
(u
j
)

(u
k
)

= 0, i = 1, 2. (2)
with coecients
i
jk
evaluated at (t).
Proof. The system of equations (2) is written
1
=
2
= 0 in the notation
of the preceding theorem. It follows from Lemma 7.1.1 that this condition
holds exactly when

(t) is proportional to m for all t. The corollary now

follows from Theorem 4.10.
The dierential equations (2) are called the geodesic equations. By Corol-
lary 7.1 the determination of the geodesics on a given surface is a matter
of solving these equations. However, for a general surface they are quite
complicated non-linear dierential equations which are not easy to solve.
As mentioned, the property of a curve of being a geodesic is intrinsic. It
follows that an isometry will carry geodesic curves to geodesic curves.
Example 7.1.1 Consider again the isometry in Example 6.5.3 from cone
to plane. The geodesics on the plane are the straight line segments, hence
we conclude that a curve on the cone is a geodesic if and only if its image by
is a line segment in V .
For example, the plane unit speed line (s) = (1, s, 0) is in polar coordi-
nates (r, ) = (r cos , r sin , 0) given by
(s) = (r(s), (s)) = (
_
1 +s
2
, tan
1
s)).
The image by
1
is then
(s) = (
1
(r(s), (s))) = (k
1
_
1 +s
2
, k tan
1
s).
Recall that (u, v) = (ucos v, usinv, u). We obtain that
(s)
= (k
1
_
1 +s
2
cos(k tan
1
s), k
1
_
1 +s
2
sin(k tan
1
s), k
1
_
1 +s
2
)
is a geodesic on the cone when k
2
= 1+
2
. An idea of the shape of the curve
can be obtained by folding a cone out of a piece of paper with a straight line
drawn on it (see the following gure).
112 Chapter 7

7.2 Existence of geodesics

A further analysis of geodesics on a surface can be based on the dierential
equations (2). This requires the use of the fundamental theorem of existence
and uniqueness of solutions of ordinary dierential equations. Without going
into details, we cite the following important consequence.
Theorem 7.2. Through every point of a regular parametrized surface passes
a unique geodesic curve in each direction.
More precisely, let p U and w T
p
0 be given. There exists a
geodesic curve = : I R
3
on with
p = (t
0
) and w =

(t
0
) (3)
for some t
0
I. Moreover, if two unit speed geodesics dened on intervals
I, J both satisfy (3) for some common t
0
I J, then they agree on I J.
Proof. Omitted.
This property is of course well known for lines on a plane.
Example 7.2.1 Through every point on a sphere passes a unique great
circle in each direction, namely the great circle obtained as the intersection
of the sphere with the unique plane through the center of the sphere which
contains the given point and the given direction vector.
7.3 Geodesic coordinates
We shall now describe a particularly useful type of parametrization of a
surface.
Denition 7.3. Let : J R
3
be a unit speed curve. A regular parame-
trized surface : U R
3
is called a geodesic coordinate system transversal to
if U = I J for some interval I and
Geodesics 113
(i) there exists u
0
I such that (v) = (u
0
, v) for all v, and this curve
is a geodesic on ,
(ii) all the coordinate curves I u (u, v) are unit speed geodesics on
, which intersect orthogonally with (that is, the tangent vector

u
(u
0
, v)
is orthogonal to

(v) =

v
(u
0
, v) for all v J).
Notice that while we are requiring (u, v) to be geodesic as a function of
u for all xed v, we are only requiring it to be geodesic as a function of v for
the xed value u
0
of u, where it produces the original curve .
u
v
I
u
0
J
U = I J

Example 7.3.1 The standard coordinates (x, y) on the xy-plane are geodesic
coordinates. Perhaps more interestingly, the spherical coordinates (u, v) on
the unit sphere is a geodesic coordinate system. Indeed, the curve (v) =
(0, v), the equator, is geodesic, and the meridians u (u, v) are geodesics
that intersect orthogonally with . Notice that in this case the curves v
(u, v) are small circles if u ,= 0, hence not geodesics.
Theorem 7.3(Existence of geodesic coordinates). Let : U R
3
be a reg-
ular parametrized surface, and let a point p U and a unit speed geodesic
= on be given with (0) = p. There exists an open rectangle
W = I J around (0, 0) in R
2
and a dieomorphism of W onto an
open neighborhood U

U of p such that (0, 0) = p and such that the

reparametrization
(s, t) = ((s, t))
of [
U
is a geodesic coordinate system transversal to [
J
.
Proof. The proof which relies on Theorem 7.2 is omitted.
7.4 The rst fundamental form of a geodesic coordinate system
Let : U R
3
be a regular surface, dened on a set U R
2
of the form
U = I J with open intervals I, J R. Let u
0
I be xed, and let
: J R
3
denote the curve t (u
0
, t) on .
114 Chapter 7
Theorem 7.4. The surface is a geodesic coordinate system transversal to
if and only if the following condition hold.
The coecients of the rst fundamental form satisfy
E(u, v) = 1, F(u, v) = 0
for all (u, v) U and
G(u
0
, v) = 1, G

u
(u
0
, v) = 0
for all v J.
Proof. The proof is based on the lemma below, from which we conclude that
u (u, v) is geodesic if and only if
E(u, v) = 1 and E

v
(u, v) 2F

u
(u, v) = 0 (4)
for all u, and (by interchanging u and v in the lemma) v (u
0
, v) is
geodesic if and only if
G(u
0
, v) = 1 and G

u
(u
0
, v) 2F

v
(u
0
, v) = 0 (5)
for all v.
Assume is a geodesic coordinate system. Then (4) and (5) hold for all
(u, v). In particular, E(u, v) = 1 and G(u
0
, v) = 1.
From E = 1 we conclude that E

u
= E

v
= 0, hence (4) implies that
F

u
(u, v) = 0, from which we infer that u F(u, v) is constant for each v.
In fact this constant is 0 because the assumption that the coordinate curves
intersect orthogonally with implies that F(u
0
, v) = 0. Finally, since F = 0
the second condition in (5) implies G

u
(u
0
, v) = 0.
The statement if is seen similarly.
Lemma 7.4. Let : U R
3
be a regular parametrization. The coordinate
curve u (u, v
0
) is a unit speed geodesic if and only if E = 1 and E

v

2F

u
= 0 in all points of the curve.
Proof. Unit speed is equivalent with E = 1. The second derivative of u
(u, v
0
) is

11
=

uu
, hence it follows from Theorem 4.10 that the curve is a
geodesic if and only if

11

k
= 0 for k = 1, 2.
By (3) in Lemma 6.3 this condition is equivalent with
2
g
1k
u
1

g
11
u
k
= 0 for k = 1, 2.
For k = 1 this equation reads
g
11
u
1
= 0, which is already a consequence of the
unit speed condition E = 1, and for k = 2 it reads 2
g
12
u
1

g
11
u
2
= 0, which
is exactly the last condition of the lemma.
Geodesics 115
7.5 Interpretation of the Gauss theorem
Let : U R
3
be a geodesic coordinate system transversal to = .
For simplicity we assume that u
0
= 0 so that (v) = (0, v). It follows from
Theorem 7.4 and the formula (11) in Chapter 6, that Gauss formula for K
in terms of the rst fundamental form is
K =
1
2

G
_
G

G
_

u
.
Since (

G)

u
=
G

u
2

G
we can rewrite the formula as
K =
1

G
(

G)

uu
. (6)
We shall now give a geometric interpretation of this formula.
Let p = (0, 0) = (0) U. For > 0 let D

denote the square

D

= [, ] [, ]
about (0, 0) in R
2
. It has area A(D

) = (2)
2
. In the following we assume
that is suciently small so that D

U. The set
(D

)
is called a square about p on . Its area is denoted A(, D

) (see Section 3.9).

u
v
U
D

y
z
x
(D

)
(U)
Theorem 7.5. Let : U R
3
be a geodesic coordinate system around p =
(0, 0) U. The Gauss curvature K of in p is given by
K =
3
2
lim
0

4
(A(, D

) A(D

)). (7)
The interpretation of K(p) is thus that it is a measure for the dierence
between the area of a small square about p and the corresponding area of a
plane square. Since areas are intrinsic properties, and since the properties
116 Chapter 7
that went into the denition of a square (geodesics and right angles) are
also intrinsic, Gauss Teorema Egregium is certainly a consequence of this
theorem. However, this serves as a geometric explanation rather than a new
proof of the theorem, since the proof given below of (7) uses Gauss formula
for K, of which the Teorema is already an immediate consequence.
In particular we notice the minus in the limit formula for K. Thus, in
an elliptic point, the area of (D

) will be smaller than that of D

, for
suciently small, and in a hyperbolic point it will be larger.
Proof. We shall use the Taylor approximation formula for the smooth func-
tion f(u, v) =
_
G(u, v), see Appendix B. With (u
0
, v
0
) = (0, 0) it reads
f(u, v) = f(0, 0) +f

u
(0, 0)u +f

v
(0, 0)v
+
1
2
(f

uu
(0, 0)u
2
+ 2f

uv
(0, 0)uv +f

vv
(0, 0)v
2
) +R(u, v)
where the remainder R(u, v) satises [R(u, v)[ C|(u, v)|
3
in a neighbor-
hood of (0, 0) for a constant C.
By Theorem 7.4 we have G(0, v) = 1 and G

u
(0, v) = G

v
(0, v) = 0. Hence
f(0, v) = 1 and f

u
(0, v) = f

v
(0, v) = 0,
and by dierentiation with respect to v,
f

uv
(0, v) = f

vv
(0, v) = 0.
Finally, by the Gauss formula (6), f

uu
(0, 0) = K. The Taylor formula is
thus
_
G(u, v) = 1
1
2
Ku
2
+R(u, v).
Since EGF
2
= G, the area of (D

) is by denition
A(, D

) =
_
D

GdA
and hence
A(, D

) A(D

) =
_
D

_
G(u, v) 1 dA
=
_

1
2
Ku
2
+R(u, v) dudv
=
2
3

4
K +
_

R(u, v) dudv.
Since [R(u, v)[ is bounded by a constant times
3
, its integral over D

is
bounded by a constant times
5
. The limit formula for K follows immedi-
ately.
Geodesics 117
7.6 Exercises
1 Let U = (u, v) [ v > 1 and suppose : U R
3
is a regular parametrized
surface with E = G = v
2
and F = 0.
a. Determine the Gauss curvature K, as a function of (u, v).
b. Compute the Christoel symbols for .
c. Verify that the curve , where
(s) = (a, e
s
) or (s) = (a +r tanhs, r
1
coshs
),
has unit speed, and show that it is a geodesic. Here a R and r > 0 are
constants, and s is assumed to belong in an interval for which (s) U.
Make a sketch of each curve in the (u, v)-plane, say with a = r = 1
(Hint: Notice that tanh
2
s + (
1
cosh s
)
2
= 1).
d. Suppose in addition that the mentioned surface has coecients M = 0
and N = v
2
(v
2
1)
1
2
in the second fundamental form. Determine L and
the principal curvatures
1
,
2
.
2 Let U = R
2
, and let : U R
3
be a regular parametrized surface for
which E = 1, F = 0 and G = 1 +u
2
(see for example Exercise 3.10).
a. Determine the Christoel symbols.
b. Show that t (t, v) is a geodesic for all v.
c. Find the geodesic curvature of the curve t (u, t) for u R.
d. Verify that is a geodesic coordinate system, and determine the Gauss
curvature by means of equation (6).
3 Let : U R
3
be a geodesic coordinate system for which the Gaussian
curvature is constant, K = 0. Show that G = 1 and that is isometric
to a part of a plane (Hint: Conclude from (6) that G = (au +b)
2
where a
and b are functions of v. Determine a and b from Theorem 7.4).
4 Let : U = I J R
3
be a geodesic coordinate system transversal to
the curve (t) = (0, t). Assume that the Gaussian curvature is constant,
K = 1. Show that G = cos
2
u and that is isometric to a part of the unit
sphere (Hint: Conclude from (6) that G = (a cos u+b sin u)
2
where a and
b are functions of v. Determine a and b from Theorem 7.4).
118 Chapter 7
Appendices
Appendix A. Euclidean spaces
The set R
n
is called Euclidean n-space. It is a vector space with the
standard addition and scalar multiplication. In this appendix we recall some
elementary notions for this space. The dot product of two vectors v, w R
n
is the number dened by
v w = v
1
w
1
+ +v
n
w
n
R.
The norm of v R
n
is given by
|v| = (v v)
1/2
= (v
2
1
+ +v
2
n
)
1/2
,
and the Euclidean distance between v, w R
n
is then dened as the norm
|v w| of their dierence. The angle between v and w is dened to be the
number [0, ] for which
cos =
v w
|v| |w|
(A.1)
provided the vectors are non-zero. It follows from the Cauchy-Schwarz in-
equality
[v w[ |v| |w|
that the right hand side of (A.1) belongs to [1, 1], so that the angle is
well dened.
The vectors v and w are said to be orthogonal if v w = 0, or equivalently,
if the angle between them is

2
, and they are said to be orthonormal if in
addition they both have length 1. An orthonormal basis for R
n
(or for a
subspace) is a basis whose members are pairwise orthonormal, as for example
the standard basis e
1
, e
2
, e
3
for R
3
.
For r > 0 and p R
n
the set
B
r
(p) = x [ |x p| < r
is called the open ball around p of radius r. A neighborhood of p is a set
U R
n
which contains the open ball B
r
(p) for some r > 0. A set U R
n
is called open if it is a neighborhood of each of its points, that is, if for every
p U there exists r > 0 such that all x R
n
with |x p| < r belong to U.
120 Appendix B
For instance, a set in R
2
of the form U =]a, b[]c, d[, with open intervals
]a, b[ and ]c, d[, is open.
The interior of an arbitrary set A R
n
is the set of points p A for
which A is a neighborhood. This set is often denoted A

, and it is an open
set. It is the largest open set contained in A. In particular, the interior of
an open set is the set itself.
The boundary of A R
n
is the set of points p R
n
(not necessarily in
A) for which every open ball around p contains at least one point of A and
at least one point of the complement R
n
A. It is often denoted A. A set
A R
n
is called closed if A A.
For example, the boundary of U =]a, b[]c, d[ consists of the four line
segments that connect the corners of U.
A set A R
n
is called bounded if there exists R > 0 such that |x| R
for all x A.
We recall that a function f: A R, where A R
n
, is called continuous if
for each p A and each > 0 there exists > 0 such that if |xp| < then
[f(x) f(p)[ < . A function f: A R
m
is continuous if the components
f
1
, . . . , f
m
: A R dened by f(x) = (f
1
(x), . . . , f
m
(x)), are continuous.
Exercises
A.1 Determine the angle between (1, 1, 1, 1) and (1, 1, 1, 0) in R
4
.
A.2 Let (t) = (3t, 3t
2
, 2t
3
). Show that the angle between the tangent vector
of and the line given by y = 0, z = x, is a constant.
A.3 Verify that u = (
2
3
,
2
3
,
1
3
) and v = (
1
3
,
2
3
,
2
3
) are orthonormal vectors.
Find a third vector w R
3
, such that u, v, w is orthonormal basis.
Determine the coordinates for a = (1, 1, 1) with respect to this basis.
A.4 Prove that the set (u, v) R
2
[ u, v > 0 is open in R
2
, and that
(u, v) R
2
[ u, v 0 is not open.
A.5 Assume that f: R
n
R is continuous. Prove that x R
n
[ f(x) < c
is an open set for every constant c R.
A.6 Verify that the open ball B
r
(p) really is open.
Appendix B. Dierentiable functions of several variables
Dierentiability
Let R
n
be open, and let f: R. A partial derivative of f is dened
as the derivative of f with respect to one of the variables x
1
, . . . , x
n
, the
others being treated as constants. For example the rst partial derivative
f

x
1
=
f
x
1
at a is the derivative at a
1
of
t f(t, a
2
, . . . , a
n
).
Appendices 121
The partial derivative at a is dened when this function of t is dierentiable at
a
1
. If this is the case for all i = 1, . . . , n, we say that f has partial derivatives
at a. If f has partial derivatives at all a , and if these partial derivatives
are continuous functions of a, then we say that f is continuously dierentiable
or a C
1
-function on . The set of such functions on is denoted C
1
().
Let F: R
m
be a vector function, and let F
1
, . . . , F
m
: R denote
the components. The partial derivatives (if they exist) of these components
functions are conveniently arranged in the Jacobi matrix
DF(a) =
_
_
_
F
1
x
1
(a) . . .
F
1
x
n
(a)
.
.
.
.
.
.
F
m
x
1
(a) . . .
F
m
x
n
(a)
_
_
_
.
Notice that DF is a map that associates a matrix to each point a . If
n = 1 we identify the single column matrix DF(a) with a vector in R
m
.
The vector function DF: R R
m
is in this case denoted F

and called the

derivative of F. The Jacobi matrix is the analogue for functions of several
variables of this derivative.
Example B.1 A linear map L: R
n
R
m
is represented by an mn matrix
A = (a
ij
) as follows:
L(x) = Ax =
_
_
a
11
. . . a
1n
.
.
.
.
.
.
a
m1
. . . a
mn
_
_
_
_
x
1
.
.
.
x
n
_
_
=
_
_
a
11
x
1
+ +a
1n
x
n
.
.
.
a
m1
x
1
+ +a
mn
x
n
_
_
.
It is easily seen that the Jacobian of this map is exactly the matrix A, that
is, DL(x) = A for all x R
n
.
We call F continuously dierentiable if each coordinate function F
j
is con-
tinuously dierentiable, or in other words, if each entry in the Jacobi matrix
exists and depends continuously on a. Recall the following fundamental the-
orem, which asserts that x F(a) +DF(a)(x a) approximates F near a.
Theorem B.1. Let F: R
m
be continuously dierentiable and let a .
Then
|F(x) [F(a) +DF(a)(x a)]|
|x a|
0 for x a, (B.1)
that is, the vector dierence F(x) [F(a) + DF(a)(x a)] tends to 0 even
after division by |x a|.
A function F which satises (B.1) is called dierentiable at a, and the the-
orem simply asserts that continuously dierentiable implies dierentiable.
122 Appendix B
Composition
The dierentiation of composed maps is governed by the chain rule. For
functions of one variable it is the well known rule
(g f)

(a) = g

(f(a))f

(a),
and for functions of several variables it takes the following form.
Theorem B.2 (Chain rule). Let R
n
and

R
m
be open, and let
F:

and G:

R
l
,
be continuously dierentiable. Then
G F: R
l
is continuously dierentiable and has the Jacobi matrix
D(G F)(a) = DG(F(a)) DF(a)
for all a , where the product on the right is given by ordinary matrix
multiplication.
In particular, if n = 1 we can write the chain rule in the following form
(G F)

(a) = DG(F(a)) F

(a). (B.2)
Example B.2 Let F: R R
2
be given by F(t) = (t
2
, t+1) and let G: R
2

R be given by G(y
1
, y
2
) = y
1
y
2
2
y
2
1
. Then
F

(t) =
_
2t
1
_
, and DG(y) = ( y
2
2
2y
1
2y
1
y
2
) .
Hence G F: R R has the derivative
(G F)

(t) = DG(F(t))F

(t)
= ( (t + 1)
2
2t
2
2t
2
(t + 1) )
_
2t
1
_
= (t
2
+ 2t + 1)2t + (2t
3
+ 2t
2
)1 = 6t
2
+ 2t.
Notice that we could also rst have determined the expression G F(t) =
t
2
(t + 1)
2
t
4
= 2t
3
+ t
2
and then dierentiated (G F)

= 6t
2
+ 2t. For
the purpose of computing (G F)

, this would clearly be much faster. The

importance of the chain rule is more theoretical, it gives a general expression
for the derivative.
Symmetry of mixed partials
A function f: R
n
R is called a C
2
-function if it is C
1
and all the
rst order partial derivatives are also C
1
-functions. The partial derivatives
of the partial derivatives, that is, the functions

2
f
x
i
x
j
=

f
x
j
x
i
are called higher or mixed partial derivatives.
Appendices 123
Theorem B.3. Let f: R be C
2
. Then

2
f
x
i
x
j
=

2
f
x
j
x
i
for all i and j.
Similarly, we can consider derivatives of order higher than 2. If f has
partial derivatives up to order k, and if these are continuous, then f is called
a C
k
-function. From the theorem above we can derive similar statements
about symmetry of these higher derivatives, for example

3
f
x
2
1
x
2
=

3
f
x
1
x
2
x
1
=

3
f
x
2
x
2
1
when f is a C
3
-function. In short, the conclusion is that dierentiations with
respect to x
1
, . . . , x
n
commute with each other (when applied to functions
which are continuously dierentiable up to sucient order).
A function which is C
k
for all k is called C

or smooth. The set of such

functions on is denoted C

(). This is the class of functions that is mainly

used in dierential geometry.
Taylors theorem
Taylors theorem allows us to approximate a smooth function by a poly-
nomial of any given order, in the vicinity of a given point.
For a smooth function f: R, where R
2
, it reads to the order one
f(u
0
+h, v
0
+k) f(u
0
, v
0
) +f

u
(u
0
, v
0
)h +f

v
(u
0
, v
0
)k
and to the order two
f(u
0
+h, v
0
+k) f(u
0
, v
0
) +f

u
(u
0
, v
0
)h +f

v
(u
0
, v
0
)k
+
1
2
f

uu
(u
0
, v
0
)h
2
+f

uv
(u
0
, v
0
)hk +
1
2
f

vv
(u
0
, v
0
)k
2
.
These statements are qualitative, because the approximation is not a well
dened relation.
There are more precise versions, where the remainder, which by denition
is the dierence between the two sides of , is estimated. To the order one
f(u
0
+h, v
0
+k) = f(u
0
, v
0
) +f

u
(u
0
, v
0
)h +f

v
(u
0
, v
0
)k +R
1
(h, k),
and the estimate, which is valid for a C
2
-function f: R, is as follows.
For a given point (u
0
, v
0
) there exist constants > 0 and C > 0 such
that
[R
1
(h, k)[ C|(h, k)|
2
124 Appendix C
for all (h, k) R
2
with |(h, k)| < .
Likewise, to the order two,
f(u
0
+h,v
0
+k) = f(u
0
, v
0
) +f

u
(u
0
, v
0
)h +f

v
(u
0
, v
0
)k
+
1
2
f

uu
(u
0
, v
0
)h
2
+f

uv
(u
0
, v
0
)hk +
1
2
f

vv
(u
0
, v
0
)k
2
+R
2
(h, k).
with the following estimate valid for a C
3
-function f. For a given point
(u
0
, v
0
) there exist constants > 0 and C > 0 such that
[R
2
(h, k)[ C|(h, k)|
3
for all (h, k) R
2
with |(h, k)| < .
Exercises
B.1 Find f

u
and f

v
for each of the functions f: R
2
R:
1) f(u, v) = u
2
+v
2
+ 3uv u 4v,
2) f(u, v) = e
2uv+1
B.2 Determine the Jacobi matrix at (1, 1) for f: (u, v) [ u, v > 0 R
2
,
given by f(u, v) = (u
2
v, 2

uv).
B.3 Let f be a dierentiable map R
3
R. Determine the derivative of
t f(t, t
2
, e
t
).
B.4 Let g: R
2
R be given by g(x, y) = xy, and let : R
2
R
2
be given
by (x, y) = (x+y, xy). Determine the Jacobi matrices for g, , and
for the inverse map f(u, v) =
1
(u, v). Determine the Jacobi matrix
for g f in each of the following two ways:
1) By using the chain rule.
2) Through explicit computation of g f(u, v).
B.5 Let be an arbitrary dierentiable function R R, and let F(x, y) =
xy (y/x) for (x, y) R
2
with x ,= 0. Show that x
F
x
+y
F
y
= 2xy.
B.6 Prove the rule
(f g)

= f

g +f g

for f, g: R R
n
(see page 62).
B.7 Let : I R
n
be smooth with (t) ,= 0 for all t. Show that t |(t)|
is dierentiable and has the derivative

(t) (t)
|(t)|
.
Appendices 125
Appendix C. Normal vectors and cross products
In this appendix the construction of normal vectors in R
2
and cross prod-
ucts in R
3
is briey presented. These notions appear naturally in many
geometrical constructions. For example the geometry of planes in R
3
is often
expressed by means of cross products. Cross products also play a prominent
role in mechanics and electromagnetic theory.
The common background for the denitions in this appendix for R
2
and
R
3
is a choice of orientation, which we will rst explain generally for R
n
.
For an ordered set of n vectors a
1
, a
2
, . . . , a
n
in R
n
we denote by [a
1
a
2
. . . a
n
]
the n n matrix which has a
1
, a
2
etc as its columns (in that order). We
divide the bases for R
n
in two classes, depending on whether the determinant
of [a
1
a
2
. . . a
n
] is positive or negative (the determinant is non-zero since the
vectors are linearly independent). An orientation of R
n
is a choice of one of
the two classes. The standard choice is the class of bases which have positive
determinant. Such a basis is then called positively ordered.
Having made this standard choice we thus say that two basis vectors a
and b in R
2
are positively ordered if det[ab] > 0 and we say that three basis
vectors a, b and c in R
3
are positively ordered if det[abc] > 0. For example,
the standard basis vectors e
1
= (1, 0) and e
2
= (0, 1) for R
2
and e
1
= (1, 0, 0),
e
2
= (0, 1, 0) and e
3
= (0, 0, 1) for R
3
are positively ordered.
In R
2
this choice of orientation means that a, b is a positively ordered basis
if and only if the direction of b can be reached from the direction of a by a
counter clockwise rotation of an angle between 0 and , and in R
3
it means
that a, b, c is positively ordered if and only if the vectors form a right-handed
triple.
Let a = (a
1
, a
2
) R
2
. We dene the normal vector by a = (a
2
, a
1
). It is
the vector obtained by rotating a the angle

2
in counter clockwise direction
(which is the positive direction according to our chosen orientation). Notice
that
det[ab] = a
1
b
2
a
2
b
1
= a b (C.1)
for all b R
2
, where the dot denotes the standard dot product (see Appendix
A). The map a a is linear.
The construction of a is particular for R
2
. In R
3
there is no way to
distinguish a normal vector to a given vector, since there are innitely many
normal vectors. Instead, the analog of the construction points out a normal
vector to two given vectors.
For two vectors a = (a
1
, a
2
, a
3
) and b = (b
1
, b
2
, b
3
) in R
3
we dene the
cross product, which is again a vector in R
3
, by
a b =
_

a
2
b
2
a
3
b
3

a
1
b
1
a
3
b
3

a
1
b
1
a
2
b
2

_
.
126 Appendix D
It follows from the determinant identity

a
1
b
1
c
1
a
2
b
2
c
2
a
3
b
3
c
3

= c
1

a
2
b
2
a
3
b
3

c
2

a
1
b
1
a
3
b
3

+c
3

a
1
b
1
a
2
b
2

that a b is uniquely characterized by the property that

det[abc] = (a b) c (C.2)
for all c R
3
. Equation (C.2) is the 3-dimensional analogue of (C.1), the
expression on the right is called the triple product of a, b and c. It follows
from this characterization that a b depends linearly on both a and b (since
the determinant depends linearly on each of its columns).
The following properties of a b are easily veried
(i) a b is perpendicular to both a and b
(ii) b a = a b
(iii) |a b|
2
= |a|
2
|b|
2
(a b)
2
(iv) |a b| is the area of the parallelogram spanned by a and b
(v) a and b are linearly independent if and only if a b ,= 0
(vi) if a b ,= 0 then a, b, and a b is a right handed triple.
The last property in this list reects our chosen orientation of R
3
. No-
tice that properties (i), (iv) and (vi) together determine a b uniquely in
geometric terms.
Exercises
C.1 Prove that if three vectors u, v, w R
3
satisfy u + v + w = 0, then
u v = v w = w u.
C.2 Let A be an orthogonal 3 3-matrix with det A = 1. Prove
A(w
1
w
2
) = Aw
1
Aw
2
for all w
1
, w
2
R
3
(Hint: Use the characterization by (C.2) of the cross
product, together with the equation (D.1)).
Appendix D. Diagonalization of symmetric matrices
Let A be an n n matrix. By denition, a diagonalization of A is ac-
complished by an invertible matrix C, if the matrix D = C
1
AC is diagonal
(that is, all entries outside the diagonal are 0). Diagonalization plays a very
important role in linear algebra, basically because it is a means to simplify
expressions involving A. It is not possible to diagonalize all matrices A, cer-
tain conditions have to be imposed. The main result of this appendix, which
Appendices 127
is stated in the theorem below, gives one such condition (but not the most
general one).
The theory of diagonalization is closely connected with the theory of eigen-
vectors and eigenvalues. Recall that an eigenvalue for the matrix A is a
number for which there exists a non-zero vector w R
n
such that
Aw = w.
The vector w is called a corresponding eigenvector.
It is a fact known from linear algebra that a number is an eigenvalue if
and only if it is a root in the characteristic polynomial
det(AI).
The corresponding eigenvectors are the nonzero solutions w to (AI)w = 0.
Recall that the matrix A is called symmetric if it equals its transposed
matrix A
t
, that is, if its elements satisfy a
ij
= a
ji
for all i, j. Recall also
that an orthogonal matrix is a square matrix C with real entries for which
the transposed matrix C
t
is inverse to C, that is, C
t
C = I. Equivalently, it
is a matrix whose columns form an orthonormal set.
Theorem D.1. Let A be a symmetric n n matrix. There exists an or-
thogonal n n matrix C such that D = C
1
AC is a diagonal matrix with
these roots as its entries. The columns of C are eigenvectors for A, and the
eigenvalues are the diagonal elements of D (in the corresponding order).
Proof. For simplicity we shall assume n = 2 in the proof. This is not a serious
restriction for the present notes, where all the applications have n = 2. In
order to pinpoint where the assumption is critically used, we will keep n
arbitrary in the beginning of the proof.
We regard vectors in R
n
as matrices with a single column, and note that
the dot product v w = v
1
w
1
+. . . v
n
w
n
can be written by means of matrix
multiplication as v w = v
t
w. For any n n matrix A we have
Av w = (Av)
t
w = v
t
A
t
w = v A
t
w, (D.1)
and if A is symmetric we thus obtain
Av w = v Aw (D.2)
for all v, w R
n
.
We rst prove that eigenvectors corresponding to dierent eigenvalues are
orthogonal, that is, if v, w R
n
0 and Av = v, Aw = w with ,= ,
then v w = 0. By (D.2) we have
v w = (Av) w = v (Aw) = v w,
128 Appendix D
and since ,= we must indeed have v w = 0.
We next observe that if w
1
, . . . , w
n
R
n
is an orthonormal set of eigenvec-
tors for A, then the matrix C = [w
1
. . . w
n
] with these columns is orthogonal.
Since the columns of the matrix product AC are obtained by multiplication
of A with the columns of C, we see that
AC = [Aw
1
. . . Aw
n
] = [
1
w
1
. . .
n
w
n
] = CD,
where D is the diagonal matrix with the eigenvalues
1
, . . . ,
n
as entries. It
follows that C
1
AC = D, so that C diagonalizes A.
It remains to be shown that there exist such an orthonormal set of eigen-
vectors. For this last step we assume n = 2. Let
A =
_
a b
b d
_
We may assume b ,= 0, since otherwise A is already diagonal and we can take
w
1
= e
1
, w
2
= e
2
. The characteristic polynomial is
2
(a +d) + ad b
2
,
and its roots are given by
=
a +d
_
(a +d)
2
4(ad b
2
)
2
=
a +d
_
(a d)
2
+ 4b
2
2
.
Since b ,= 0 the expression inside the square root is positive, and thus there
are two dierent real roots
1
,
2
. As mentioned above each root in the
characteristic polynomial is an eigenvalue, let w
1
, w
2
R
2
be normalised
eigenvectors. Then w
1
w
2
by what was shown above, and hence they form
an orthonormal set.
Example D.1 The eigenvalues of the symmetric matrix
_
1
1
2
1
2
1
_
are determined as the roots
1
2
and
3
2
of
det
_
1
1
2
1
2
1
_
=
2
2 +
3
4
.
An eigenvector corresponding to =
1
2
is determined by solution of
_
1
1
2
1
2
1
__
x
y
_
=
1
2
_
x
y
_
which leads to, for example
_
x
y
_
=
_
1
1
_
.
Appendices 129
An eigenvector for =
3
2
is then found as
_
x
y
_
=

_
1
1
_
=
_
1
1
_
.
The corresponding matrix C has normalisations of these eigenvectors as its
columns
C =
_
1

2
1

2
1

2
_
,
and the diagonalized matrix is
D =
_
1
2
0
0
3
2
_
.
It is useful to express the preceding theorem as a result about linear maps,
rather than matrices. This is done in the following corollary.
Corollary D.1. Let U R
m
be a linear subspace, and let L: U U be a
linear map which is symmetric, that is
L(u
1
) u
2
= u
1
L(u
2
), u
1
, u
2
U. (D.3)
Then there exists an orthonormal basis for U consisting of eigenvectors for L.
Proof. Let
1
, . . . ,
n
be an arbitrary orthonormal basis for U, and let A
denote the n n matrix which represents L with respect to this basis, that
is,
L
j
=
n

i=1
a
ij

i
, (j = 1, . . . , n). (D.4)
The proof is based on the following two observations, which are well known
from linear algebra. A vector u = x
1

1
+ + x
n

n
U is an eigenvector
for L with eigenvalue ,
Lu = u,
if and only if the column of its coordinates
x =
_
_
x
1
.
.
.
x
n
_
_
R
n
is an eigenvector for A,
Ax = x
130 Appendix E
with the same eigenvalue. The second observation is that the dot product of
two vectors u, v U is computed as the dot product of their coordinates:
(x
1

1
+ +x
n

n
) (y
1

1
+ +y
n

n
) = x
1
y
1
+ +x
n
y
n
.
Since the basis is orthonormal, it follows from (D.4) that
a
ij
= L(
j
)
i
,
and hence it follows from the symmetry of L, that A is a symmetric matrix.
By Theorem D.1 there exists an orthogonal n n matrix C whose columns
are eigenvectors for A. Let v
1
, . . . , v
n
be the vectors in U whose coordinates
with respect to
1
, . . . ,
n
are the columns of C, that is
v
i
= c
1i

1
+ +c
ni

n
U, (i = 1, . . . , n).
The rst observation made above implies that v
1
, . . . , v
n
are eigenvectors for
L, and the second observation implies that they form an orthonormal set.
Since the dimension of U is n, they form a basis for U.
Exercises
D.1 Let w = (3, 4) R
2
, and let
A =
_
a 12
12 a + 7
_
.
Show that w and w are eigenvectors for A, and determine their eigen-
values.
D.2 Let
A =
_
2 2
2 5
_
Determine an orthogonal matrix C and a diagonal matrix D, such that
D = C
1
AC. How many pairs of 2 2-matrices (C, D) satisfy these
requirements?
D.3 Let A be an n n-matrix. Show that if there exists an orthogonal
matrix diagonalizing A, then A is symmetric.
Appendix E. Hyperbolic functions
The hyperbolic cosine and hyperbolic sine functions are dened by
cosh t =
e
t
+e
t
2
, sinh t =
e
t
e
t
2
, (E.1)
Appendices 131
for t R. The equation
cosh
2
t sinh
2
t = 1,
which resembles the well-known cos
2
t+sin
2
t = 1, is easily derived from (E.1).
It follows that the point (cosht, sinh t) lies on a hyperbola (see Example
1.1.3), and because of this the functions are viewed as hyperbolic counterparts
to the trigonometric functions cos and sin. The graphs of the two functions
are shown below.
x
y = cosh x y = sinh x
x
y y
1
Notice that cosh is even and sinh is odd. In analogy with the denitions of
the tangent and cotangent one denes
tanh x =
sinh x
cosh x
, coth x =
cosh x
sinh x
.
It is easily seen that the derivatives of these functions are
d
dx
cosh x = sinh x,
d
dx
sinh x = cosh x,
d
dx
tanh x =
1
cosh
2
x
,
d
dx
coth x =
1
sinh
2
x
.
132 Notation
Notational Index
A: area, 53
b(s): binormal, 67
b
ij
: component matrix of II
p
, 96
(: level set, 6
D: Jacobi matrix, 121
E: component of I
p
, 43
F: component of I
p
, 43
G: component of I
p
, 43
g
ij
: component matrix of I
p
, 96
I
p
: rst fundamental form, 43
II
p
: second fundamental form, 80
K: Gaussian curvature, 93
L: component of II
p
, 80
M: component of II
p
, 80
m(s): normal along curve, 69
N: component of II
p
, 80
N: unit normal, 29
n(s): principal normal, 67
R
jkil
: Riemann symbol, 100
o: level set, 11
T
p
: tangent space, 23
t(s): unit tangent, 67
u(s): tangent normal, 70
W
p
: shape operator, 78, 82

k
ij
: Christoel symbol, 97
: a parametrized curve, 1
(t): tangent angle function, 63
: curvature, 59, 64

g
: geodesic curvature, 70

n
: normal curvature, 70

1
,
2
: principal curvatures, 83
(t): plane coordinates of curve, 25
: a parametrized surface, 3

u
,

v
: partial derivatives, 22
: torsion, 66
: a dieomorphism, 27
: lift of isometry, 101
_
dA: plane integral, 48
: boundary, 120
: cross product, 125
a: normal vector of a, 125
Index
angle,
in Euclidean space, 119
on surface, 45
Archimedes spiral, 55
arc-length, 39
function, 40
area,
of block-set, 47
of elementary domain, 48
of rectangle, 46
of surface, 53
basis, positively ordered, 125
bending, 101
binormal, 67
block-set, 46
boundary, 120
C
1
-function, 121
cardioid, 34
catenoid, 105
Cauchy-Schwarz inequality, 119
center of curvature, 74
chain rule, 122
characteristic polynomial, 127
Christoel symbols, 97
circle, 1
great, on sphere 71
osculating, 74
small, on sphere 71
cissoid, 16
components, 120
component functions, 43
cone, 5
continuous, 120
coordinate curve, 25
coordinates, 3
spherical, 4
cosh, 130
coth, 131
critical point, 6
cross product, 23, 125
curvature,
center of, 74
Gaussian, 93
geodesic, 70
normal, 70, 72
normal, of surface, 73
of plane curve, 59
of space curve, 64
tangential, 70
curve,
constant, 2, 19
on surface, 25
parametrized, 1
by arc-length, 42
regular, 19
singular, 19
smooth, 2
cycloid, 16
cylinder, 4
derivative, partial 120
diagonalization, 126
dieomorphism, 27
direction, 27
distance, Euclidean, 119
domain, elementary, 47
dot product, 119
eigenvalue, 127
eigenvector, 127
elementary domain, 47
ellipse, 2
elliptic point, 89
Euclidean space, 119
rst fundamental form, 43
Frenet formula, 69
form,
rst fundamental, 43
quadratic, 44
second fundamental, 80, 80
134 Index
Gauss
curvature, 93, 94
formula, 98
theorem, 99
geodesic, 73
coordinates, 112
equations, 111
graph, 5, 20
helicoid, 56
rst fundamental form, 56
isometry with catenoid, 105
second fundamental form, 89
helix, 3,
arclength, 40
curvature, 65
hyperbolic, 75
torsion, 66
hyperbola, 2
hyperbolic
functions, 130
helix, 75
point, 89
ideal map, 105
induced isometry, 101
integral,
double, 51
over rectangle, 46
plane, 48
transformation of, 52
interior, 120
intrinsic, 95
invariant, 28
isometric, 100
isometry, 101
Jacobi matrix, 121
latitude, 4
level set, 6
lift, 102
line, 1
segment, 41
logarithmic spiral, 55
longitude, 4
matrix,
orthogonal, 127
symmetric, 127
meridian, 57
motion, rigid, 90
moving frame, 68
neighborhood, 119
norm, 119
normal,
principal, 67
unit, 29
vector, 125
null set, 49
orientation, 30, 125
orthogonal, 119
parametrization, 45
orthonormal, 119
osculating circle, 74
plane, 67
parabolic
cylinder, 87
point, 89
paraboloid, 86
elliptic, 86
hyperbolic, 86
parallel curve, 57
partial derivative, 120
partition, of block-set, 47
Peanos curve, 2
planar point, 89
plane, 3
osculating, 67
preserve
direction, 27
orientation, 30
principal
curvature, 83, 84
normal, 67
vector, 83, 84
pseudosphere, 106
rectangle, 46
reparametrization, 26, 28
Index 135
reverse
direction, 27
orientation, 30
revolution, surface of, 35
rst fundamental form, 57
Gauss curvature, 106
geodesics, 76
second fundamental form, 89
Riemann symbol, 100
rigid motion, 90
second fundamental form, 80
coordinate expression, 80
self-intersection, 2
set
bounded, 120
closed, 120
open, 119
shape operator, 78, 82
sinh, 130
smooth function, 123
speed, 39
sphere, 4
spherical coordinates, 4
spiral,
logarithmic, 55
of Archimedes, 55
surface,
of revolution, 35
parametrized, 3
regular, 22
singular, 23
tangent
angle, 63
line, 20
plane, 24
space, 23
vector, 20
tanh, 131
Taylors theorem, 123
teorema egregium, 99
theorem,
Gauss, 99
implicit function, 7, 11, 13
inverse function, 26, 32
Taylor, 123
torsion, 66
torus, 58
trace, 1
tractrix, 106
type, of surface, 89
unit
normal, 29
speed, 42
Weingarten map, 77