P. 1
blaga_geometry

blaga_geometry

|Views: 3,911|Likes:
Published by Paul Blaga

More info:

Published by: Paul Blaga on Apr 14, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

12/27/2012

pdf

text

original

Sections

  • 1.1 Introduction
  • 1.2 Parameterized curves (paths)
  • 1.3 The definition of the curve
  • 1.4 Analytical representations of curves
  • 1.4.1 Plane curves
  • 1.4.2 Space curves
  • 1.6 The osculating plane
  • 1.7 The curvature of a curve
  • 1.7.1 The geometrical meaning of curvature
  • 1.8.1 The behaviour of the Frenet frame at a parameter change
  • 1.9 Oriented curves. The Frenet frame of an oriented curve
  • 1.10 The Frenet formulae. The torsion
  • 1.10.1 The geometrical meaning of the torsion
  • 1.10.2 Some further applications of the Frenet formulae
  • 1.10.3 General helices. Lancret’s theorem
  • 1.10.4 Bertrand curves
  • 1.11 The local behaviour of a parameterized curve around a biregular point
  • 1.12 The contact between a space curve and a plane
  • 1.13 The contact between a space curve and a sphere. The osculating sphere
  • 1.14 Existence and uniqueness theorems for parameterized curves
  • 1.14.1 The behaviour of the Frenet frame under a rigid motion
  • 1.14.2 The uniqueness theorem
  • 1.14.3 The existence theorem
  • Plane curves
  • 2.1 Introduction
  • 2.2 Envelopes of plane curves
  • 2.2.1 Curves given through an implicit equation
  • 2.2.2 Families of curves depending on two parameters
  • 2.2.3 Applications: the evolute of a plane curve
  • 2.3 The curvature of a plane curve
  • 2.3.1 The geometrical interpretation of the signed curvature
  • 2.4 The curvature center. The evolute and the involute of a plane curve
  • 2.5 The osculating circle of a curve
  • 2.6 The existence and uniqueness theorem for plane curves
  • The integration of the natural equations of a curve
  • 3.1 The Riccati equation associated to the natural equations of a curve
  • 3.2 Examples for the integration of the natural equation of a plane curve
  • General theory of surfaces
  • 4.1 Parameterized surfaces (patches)
  • 4.2 Surfaces
  • 4.2.1 Representations of surfaces
  • 4.3 The equivalence of local parameterizations
  • 4.4 Curves on a surface
  • 4.5 The tangent vector space, the tangent plane and the nor- mal to a surface
  • 4.6 The orientation of surfaces
  • 4.7 Differentiable maps on a surface
  • 4.8 The differential of a smooth map between surfaces
  • 4.9 The spherical map and the shape operator of an oriented surface
  • 4.10 The first fundamental form of a surface
  • 4.10.1 First applications
  • The length of a segment of curve on a surface
  • The angle of two curves on a surface
  • The area of a parameterized surface
  • 4.12 The second fundamental form of an oriented surface
  • 4.13 The normal curvature. The Meusnier’s theorem
  • 4.14 Asymptotic directions and asymptotic lines on a surface
  • 4.15 The classification of points on a surface
  • 4.16.1 The determination of the lines of curvature
  • 4.16.2 The computation of the curvatures of a surface
  • 4.17 The fundamental equations of a surface
  • 4.17.1 Introduction
  • 4.17.2 The differentiation rules. Christoffel’s coefficients
  • Christoffel’s and Weingarten’s coefficients in curvature coordinates
  • 4.17.3 The Gauss’ and Codazzi-Mainardi’s equations for a surface
  • 4.17.4 The fundamental theorem of surface theory
  • 4.18 The Gauss’ egregium theorem
  • 4.19 Geodesics
  • 4.19.1 Introduction
  • 4.19.2 The Darboux frame. The geodesic curvature and geodesic torsion
  • 4.19.3 Geodesic lines
  • Examples of geodesics
  • 4.19.4 Liouville surfaces
  • Special classes of surfaces
  • 5.1 Ruled surfaces
  • 5.1.1 General ruled surfaces
  • The parameterization of a ruled surface
  • The tangent plane and the first fundamental form of a ruled surface
  • 5.1.2 The Gaussian curvature of a ruled surface
  • 5.1.3 Envelope of a family of surfaces
  • 5.1.4 Developable surfaces
  • The envelope of the family of osculating planes
  • The envelope of the family of normal planes (the polar surface)
  • The envelope of the family of rectifying planes of a space curves
  • 5.2 Minimal surfaces
  • 5.2.1 Definition and general properties
  • 5.2.2 Minimal surfaces of revolution
  • 5.2.3 Ruled minimal surfaces
  • 5.3 Surfaces of constant curvature
  • Bibliography

Lectures on the Differential Geometry

of Curves and Surfaces
Paul A. Blaga
To Cristina, with love
Contents
Foreword 9
I Curves 11
1 Space curves 13
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Parameterized curves (paths) . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 The definition of the curve . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4 Analytical representations of curves . . . . . . . . . . . . . . . . . . . 26
1.4.1 Plane curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.2 Space curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5 The tangent and the normal plane . . . . . . . . . . . . . . . . . . . . . 32
1.5.1 The equations of the tangent line and normal plane (line) for
different representations of curves . . . . . . . . . . . . . . . . 35
1.6 The osculating plane . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.7 The curvature of a curve . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.7.1 The geometrical meaning of curvature . . . . . . . . . . . . . . 44
1.8 The Frenet frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.8.1 The behaviour of the Frenet frame at a parameter change . . . . 47
1.9 Oriented curves. The Frenet frame of an oriented curve . . . . . . . . . 48
1.10 The Frenet formulae. The torsion . . . . . . . . . . . . . . . . . . . . . 50
6 Contents
1.10.1 The geometrical meaning of the torsion . . . . . . . . . . . . . 53
1.10.2 Some further applications of the Frenet formulae . . . . . . . . 54
1.10.3 General helices. Lancret’s theorem . . . . . . . . . . . . . . . 56
1.10.4 Bertrand curves . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.11 The local behaviour of a parameterized curve around a biregular point . 62
1.12 The contact between a space curve and a plane . . . . . . . . . . . . . 64
1.13 The contact between a space curve and a sphere. The osculating sphere 66
1.14 Existence and uniqueness theorems for parameterized curves . . . . . . 68
1.14.1 The behaviour of the Frenet frame under a rigid motion . . . . . 68
1.14.2 The uniqueness theorem . . . . . . . . . . . . . . . . . . . . . 70
1.14.3 The existence theorem . . . . . . . . . . . . . . . . . . . . . . 72
2 Plane curves 75
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.2 Envelopes of plane curves . . . . . . . . . . . . . . . . . . . . . . . . 75
2.2.1 Curves given through an implicit equation . . . . . . . . . . . . 77
2.2.2 Families of curves depending on two parameters . . . . . . . . 79
2.2.3 Applications: the evolute of a plane curve . . . . . . . . . . . . 79
2.3 The curvature of a plane curve . . . . . . . . . . . . . . . . . . . . . . 81
2.3.1 The geometrical interpretation of the signed curvature . . . . . 84
2.4 The curvature center. The evolute and the involute of a plane curve . . . 86
2.5 The osculating circle of a curve . . . . . . . . . . . . . . . . . . . . . . 91
2.6 The existence and uniqueness theorem for plane curves . . . . . . . . . 92
3 The integration of the natural equations of a curve 95
3.1 The Riccati equation associated to the natural equations of a curve . . . 95
3.2 Examples for the integration of the natural equation of a plane curve . . 96
Problems 103
II Surfaces 113
4 General theory of surfaces 115
4.1 Parameterized surfaces (patches) . . . . . . . . . . . . . . . . . . . . . 115
4.2 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.2.1 Representations of surfaces . . . . . . . . . . . . . . . . . . . . 116
4.3 The equivalence of local parameterizations . . . . . . . . . . . . . . . . 119
4.4 Curves on a surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Contents 7
4.5 The tangent vector space, the tangent plane and the normal to a surface . 123
4.6 The orientation of surfaces . . . . . . . . . . . . . . . . . . . . . . . . 127
4.7 Differentiable maps on a surface . . . . . . . . . . . . . . . . . . . . . 130
4.8 The differential of a smooth map between surfaces . . . . . . . . . . . 134
4.9 The spherical map and the shape operator of an oriented surface . . . . 136
4.10 The first fundamental form of a surface . . . . . . . . . . . . . . . . . 139
4.10.1 First applications . . . . . . . . . . . . . . . . . . . . . . . . . 140
The length of a segment of curve on a surface . . . . . . . . . . 140
The angle of two curves on a surface . . . . . . . . . . . . . . . 141
The area of a parameterized surface . . . . . . . . . . . . . . . 142
4.11 The matrix of the shape operator . . . . . . . . . . . . . . . . . . . . . 144
4.12 The second fundamental form of an oriented surface . . . . . . . . . . 146
4.13 The normal curvature. The Meusnier’s theorem . . . . . . . . . . . . . 148
4.14 Asymptotic directions and asymptotic lines on a surface . . . . . . . . . 150
4.15 The classification of points on a surface . . . . . . . . . . . . . . . . . 152
4.16 Principal directions and curvatures . . . . . . . . . . . . . . . . . . . . 156
4.16.1 The determination of the lines of curvature . . . . . . . . . . . 159
4.16.2 The computation of the curvatures of a surface . . . . . . . . . 161
4.17 The fundamental equations of a surface . . . . . . . . . . . . . . . . . 162
4.17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.17.2 The differentiation rules. Christoffel’s coefficients . . . . . . . . 162
Christoffel’s and Weingarten’s coefficients in curvature coordinates164
4.17.3 The Gauss’ and Codazzi-Mainardi’s equations for a surface . . 165
4.17.4 The fundamental theorem of surface theory . . . . . . . . . . . 167
4.18 The Gauss’ egregium theorem . . . . . . . . . . . . . . . . . . . . . . 172
4.19 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.19.2 The Darboux frame. The geodesic curvature and geodesic torsion 175
4.19.3 Geodesic lines . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Examples of geodesics . . . . . . . . . . . . . . . . . . . . . . 181
4.19.4 Liouville surfaces . . . . . . . . . . . . . . . . . . . . . . . . . 182
5 Special classes of surfaces 185
5.1 Ruled surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
5.1.1 General ruled surfaces . . . . . . . . . . . . . . . . . . . . . . 185
The parameterization of a ruled surface . . . . . . . . . . . . . 186
The tangent plane and the first fundamental form of a ruled surface187
5.1.2 The Gaussian curvature of a ruled surface . . . . . . . . . . . . 189
8 Contents
5.1.3 Envelope of a family of surfaces . . . . . . . . . . . . . . . . . 190
5.1.4 Developable surfaces . . . . . . . . . . . . . . . . . . . . . . . 191
Developable surfaces as envelopes of a one-parameter family of
planes. The regression edge of a developable surface . 193
5.1.5 Developable surfaces associated to the Frenet frame of a space
curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
The envelope of the family of osculating planes . . . . . . . . . 197
The envelope of the family of normal planes (the polar surface) 198
The envelope of the family of rectifying planes of a space curves 199
5.2 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
5.2.1 Definition and general properties . . . . . . . . . . . . . . . . . 200
5.2.2 Minimal surfaces of revolution . . . . . . . . . . . . . . . . . . 206
5.2.3 Ruled minimal surfaces . . . . . . . . . . . . . . . . . . . . . . 207
5.3 Surfaces of constant curvature . . . . . . . . . . . . . . . . . . . . . . 213
Problems 217
Bibliography 229
Index 233
Foreword
This book is based on the lecture notes of several courses on the differential geometry of
curves and surfaces that I gave during the last eight years. These courses were addressed
to different audience and, as such, the lecture notes have been revised again and again
and once almost entirely rewritten. I tried to expose the material in a modern language,
without loosing the contact with the intuition. There is a common believe that mathe-
matics is eternal and it never changes, unlike other sciences. I think that nothing could
be farther from the truth. Mathematics does change continuously, either for the need of
more rigor, either for the need of more structure. The mathematics we are doing and
teaching today is essentially different (and not only as language!) from the mathemat-
ics of the nineteenth century. This variation also applies, of course, to the teaching of
mathematics.
There are several reason why we need to renew the textbooks from time to time:
• the mathematics evolves and we need to introduce new results and notions;
• some parts of specific courses become obsolete and should be removed;
• the curriculumas a whole evolves and we have to keep pace with the developments
of the neighboring fields;
• some courses move from the undergraduate level to the graduate level or (more
often) the other way around and we have to modify the contents and the language
accordingly.
These are only some of the reasons I decided to write this book.
10 Foreword
The material included is fairly classical. I preferred to discuss in more details the
foundations rather than introduce more and more topics. In particular, unlike most of
books of this kind, I decided not to discuss at all problems of global differential geom-
etry. I believe it is not very useful to discuss such topics, if there is not enough room to
give them all the attention they deserve. Also, the Levi-Civita parallelism and the covari-
ant differentiation are not discussed, because I think they would feel a lot better in the
course of Riemannian geometry. The notion of differentiable manifold is not introduced
but it is only one step away. In fact, I decided to adapt the language of the course in such
a way to qualify for a prerequisite for a course in smooth manifolds.
Is not easy to be original when you teach something that have been taught again and
again for more than two centuries and, on the other hand, should we be always original
at any cost?! As such, it goes without saying that this book owes much to many excellent
classical or more modern textbooks. All of them are mentioned in the bibliography. I
have to mention, however, the books of Gray, do Carmo and Fedenko.
I have to confess that I am not a self-educated person (not entirely, anyway), and my
ideas regarding differential geometry have been influenced by my teachers (the late Prof.
Marian ¸ Tarin˘ a and Prof. Dorin Andrica), as well by the colleagues with whom I had the
pleasure to work and to share the ideas, Prof. Pavel Enghi¸ s, Csaba Varga, Cornel Pintea,
Daniel V˘ ac˘ are¸tu and Liana ¸ Topan.
This book would have never existed without my students and their enthusiasm (or
lack of enthusiasm) determined many changes during the years.
A book is not something that you leave at the office and this one was no exception
and it would never have been finished without the infinite patience of my wife who
tolerated the (undeterministic) chaos created by books and manuscripts.
Cluj-Napoca, April 2005.
Part I
Curves
CHAPTER 1
Space curves
1.1 Introduction
Intuitively, the curves are just deformations of straight lines. They can be thought of,
therefore, as “one-dimensional” objects. We are familiar with some of themalready from
elementary mathematics, since, obviously, the graphs of functions can be considered as
curves, from this point of view. On the other hand, clearly, usually the curves are not
graphs of functions (well, at least not globally); it is enough to think of a conical section
(for instance an ellipse or a circle, in particular). Thus, in general we cannot represent
a curve by an equation of the form y = f (x), as it would be the case for a graph. On
the other hand, we can represent a conical section also by an implicit equation of the
form f (x, y) = 0, where, in this particular case, of course, as it is known, f is a second
degree polynomial in x and y. Finally, we can represent the coordinates of each point
of the curve as functions of one parameter. As we shall see, this is, usually, the most
convenient way to represent, locally, an arbitrary curve.
An important issue that has to be dealt with is related to the smoothness of the func-
tions used to describe a curve. Of course, we are interested to apply the tools of dif-
ferential calculus. We shall assume, therefore, that all the functions are at least once
continuous differentiable and, if higher order derivatives are involved, we shall assume,
without stating it in clear, that all the derivatives exists and are continuous. We shall use
for the functions which are satisfying these conditions the generic term “smooth”. Apart
14 Chapter 1. Space curves
from the computational aspects, there are other, deeper reasons for considering functions
at least once continuously differentiable. Suppose, for instance, that a curve is described
by a set of equations of the form

¸
¸
¸
¸
¸
¸
x = f (t),
y = g(t)
.
Now, it can be shown that if the functions f and g are only continuous, the curve can
fill an entire square (or the entire plane). The first example of such an anomalous
curve (which, obviously, contradicts the common sense image of a curve as an one-
dimensional object) has been constructed by the Italian mathematician Giuseppe Peano,
at the end of the XIX-th century. Moreover, this strange phenomenon does not disappear
not even if the functions f and g are differentiable, but not continuously differentiable. In
the figure 1.1 we indicate an iterative process that defines a square-filling (Peano) curve.
The curve itself is the limit of the curves obtained by this iterative process. It is possible,
actually, to describe analytically this curve (i.e. we can find an expression for each iter-
ation), but, as these “curves” are not the subject of our investigation, we prefer to let the
reader satisfy the curiosity by himself. We have to say, however, that the functions we
use to describe a curve do not have to be, necessarily, continuous differentiable to avoid
aforementioned anomalies. What it is required is a little bit less, namely the functions
have to be of bounded variation. It is a well-known fact that the continuous differentiable
functions do verify this condition and, moreover, as we already emphasized, they also
endow us with the necessary computational tools, which are not available for arbitrary
functions with bounded variation.
1.2 Parameterized curves (paths)
Let I be an interval on the real axis R. We shall not assume always that the interval
is open. Sometimes it is even important that the interval be closed. In particular, the
interval can be unbounded and can coincide with the entire real axis.
Definition 1.2.1. A C
k
(k > 0) parameterized curve ( or path) in the Euclidean space R
3
is a C
k
mapping
r : I → R
3
: t → (x(t), y(t), z(t)). (1.2.1)
A parameterized curve is, typically, denoted by (I, r), (I, r = r(t)) or, when the interval
is implicit, just r = r(t). We note that, in fact, a path is C
k
iff the (real valued) functions
x, y, z are C
k
. If the interval is not open, we shall assume, first of all, that the functions
we are considering are of class C
k
in the interior of the interval and all the derivatives, up
1.2. Parameterized curves (paths) 15
(a) First iteration (b) Second iteration
(c) Third iteration (d) Fourth iteration
Figure 1.1: The Peano’s curve (first four iterations)
to the k-th order, have a finite lateral (left, or right) limit at the extremities of the interval,
if these extremities belong to the interval.
The path is called compact, half-open or open, if the interval I is, respectively, com-
pact, half-open or open.
If the interval I is bounded from below, from above or from both parts, then the
imagine of any extremity of I is called an endpoint of the path. If, in particular, the
curve is compact and the two endpoints coincide, the path is called closed. An alternative
denomination used for a closed path is that of a loop.
Occasionally (for instance in the theory of the line integral) we might need to con-
sider paths which are of class C
k
at all the points of an interval, with the exception of a
finite number of points. The following definition will make this more precise.
16 Chapter 1. Space curves
Definition 1.2.2. We shall say that a compact parameterized curve r : [a, b] → R
3
is
piecewise C
k
if there exists a finite subdivision (a = a
0
, a
1
, . . . , a
n
= b) of the interval
[a, b] such that the restriction of r to each compact interval [a
i−1
, a
i
] is of class C
k
, where
i ∈ ¦1, . . . , n¦.
Remark. It is not difficult to show that a parameterized curve r : [a, b] → R
3
is piecewise
C
k
iff the following conditions are simultaneously fulfilled:
(i) The set
S =
¸
t ∈ [a, b] | f
(k)
does not exist
¸
is finite.
(ii) f
(k)
is continuous on [a, b] \ S .
(iii) f
(k)
has a finite left and right lateral limit at each point of S .
Hereafter we shall suppose, all the time, that the order of smoothness k is high
enough and we will not mention it any longer (with some exceptions, however), us-
ing the generic term of smooth parameterized curve (meaning, thus, at least C
1
and, in
any particular case, if kth order derivatives are involved, at least C
k
).
The image r(I) ⊂ R
3
of the interval I through the mapping (1.2.1) is termed the
support of the path (I, r).
If r(t
0
) = a, we shall say that the parameterized curve passes through the point
a for t = t
0
. Sometimes, for short, we shall refer to this point as the point t
0
of the
parameterized curve.
Examples
1. Let r
0
∈ R
3
be an arbitrary point, and a ∈ R
3
a vector, a 0, while I = R. The
parameterized curve R → R
3
, t → r
0
+ ta is called a straight line. Its support is
the straight line passing through r
0
(for t = 0) and having the direction given by
the vector a.
2. I = R, r(t) = r
0
+ t
3
a. The support of this path is the same straight line.
3. I = R, r(t) = (a cos t, a sin t, bt), a, b ∈ R. The support of this parameterized curve
is called a circular cylindrical helix (see figure 1.2).
4. I = [0, 2π], r(t) = (cos t, sin t, 0). The support of the path is the unit circle, lying
in the xOy-plane, having the center at the origin of the coordinates.
1.2. Parameterized curves (paths) 17
5. I = [0, 2π], r(t) = (cos 2t, sin 2t, 0). The support is the same from the previous
example.
6. I = R, r(t) = (t
2
, t
3
, 0). This curve has a cusp at the point t = 0.
Definition 1.2.3. A parameterized curve (1.2.1) is called regular for t = t
0
if r
·
(t
0
) 0
and regular if it is regular for each t ∈ I.
As we shall see a little bit later, the notion of regularity at a point of a parameterized
curve is related to the existence of a well defined tangent to the curve in that point.
The curves from the previous example are regular, with the exception of those from
the points 2 and 6, which are not regular from t = 0.
Remark. The fact that the same support can correspond both to a regular and a nonreg-
ular parameterized curve suggests that the absence of the regularity at a point doesn’t
necessarily mean that the corresponding point of the support has some geometric pe-
culiarities. It’s only that the regularity guarantees the absence of these peculiarities.
Indeed, if we look, again, at the curves 2 and 6 from the previous example, we notice
immediately that, although they are, both, nonregular for t = 0, only for the second curve
this analytic singularity implies a geometric singularity (a cusp), while for the first one
the support is just a straight line, without any special points.
To each path corresponds a subset of R
3
, its support. Nevertheless, as the examples 1
and 2 show, different parameterized curves may have the same support. A parameterized
curve can be thought of as a subset of R
3
, with a parameterization.
1
The support of a
parameterized curve corresponds to our intuitive image of a curve as a one dimensional
geometrical object. As we shall see later, the support of a parameterized curve may have
self-intersections or cusps, which, for many reasons, are not desirable in applications.
The regularity conditions rules out the cusps, but not the self-intersections. To eliminate
them, we have to impose some further conditions.
We have seen that different parameterized curves may have the same support. In the
end, it is the support, as a set of points, we are interested in. It is, therefore, necessary to
identify some relations between the parameterized curves that define the same support.
For reasons that will become clear later, for the moment, at least, we are interested only
in regular curves. Therefore, for instance, passing from a parameterized representation
of the support to another should not change the regularity of the curve.
Definition 1.2.4. Let (I, r = r(t)), (J, ρ = ρ(s)) two parameterized curves. A diffeomor-
phism λ : I → J : t → s = λ(t) such that r = ρ ◦ λ, i.e. r(t) ≡ ρ(λ(t)), is called a
1
Of course, the information is redundant, since the parametric representation defines the support.
18 Chapter 1. Space curves
Figure 1.2: The circular helix
parameter change or a reparameterization. Two parameterized curves for which there
is a parameter change are called equivalent, while the points t and s = λ(t) are called
correspondent.
Remarks. 1) The relation just defined is an equivalence relation on the set of all parame-
terized curves.
2) The reparameterization has a simple kinematical interpretation. If we interpret
the parametric equations of the path as being the equations of motion of a particle, then
the support is just the trajectory, while the vector r
·
(t) is the velocity of the particle. The
effect of a reparameterization is the modification of the speed with which the trajectory
is traversed. Also, if λ
·
(t) < 0, then the trajectory is traversed in the opposite sense. Note
that the two velocity vectors of equivalent parameterized curves in correspondent points
have the same direction. They may have different lengths (the speed) and sense.
Example. The parameterized curves from the examples 1 and 2 are not equivalent, al-
1.2. Parameterized curves (paths) 19
though, as mentioned, they have the same support.
Remark. Sometimes, the equivalence classes determined by the relation defined before
between parameterized curves are called curves. We shall not follow this line here,
because we would like the curves to be objects slightly more general that just supports
of parameterized curves. In particular, as we shall see in a moment, usually the curves
cannot be represented globally through the same parametric equations. It is enough to
think about the circle. One of the most common parametric representation of the circle
is

¸
¸
¸
¸
¸
¸
x = cos θ,
y = sin θ
.
Now, if we let the parameter vary only in the interval (0, 2π), then one of the points of
the circle is not represented. Of course, we can extend the interval, but then the same
point corresponds to more than a value of the parameter, which is, again, not acceptable.
Among all parameterized curves equivalent to a given parameterized curve, there is
one which has a special theoretical meaning and which simplifies many proofs in the
theory of curve, although, in most cases, it is very difficult to find it analytically and,
thus, its practical value is limited.
Definition 1.2.5. We shall say that a parameterized curve is naturally parameterized if
|r
·
(s)| = 1 for any s ∈ I. Usually, the natural parameter is denoted by s.
Remark. One can see immediately that any smooth, naturally parameterized curve (I, r =
r(s)) is regular, since, clearly, r
·
(s) cannot vanish, as its norm is nowhere vanishing.
It is by no means obvious that for any parameterized curve there is another one,
equivalent to it and naturally parameterized. To construct such a curve, we need first
another notion.
The arc length of a path (I, r = r(t)) between the points t
1
and t
2
is the number
2
l
t
1
,t
2
=

t
2
t
1
|r
·
(t)|dt

.
Remark. There is a good motivation for defining in this way the length of an arc. We
can consider an arbitrary division t
1
= a
0
< a
1
< < a
n
= t
2
of the interval [t
1
, t
2
] and
examine the polygonal line γ
n
= r(a
0
)r(a
1
) r(a
n
). The length of this polygonal line is
the sum of the lengths of its segments. It can be shown that if the parameterized curve
2
Although the integrand is positive we did not assume that t
1
< t
2
, therefore the integral might be
negative and the absolute value is necessary, if we want to obtain something positive.
20 Chapter 1. Space curves
(I, r) is “good enough” (for instance at least once continuously differentiable), then the
limit of length of the polygonal line γ
n
, when the norm of the division goes to zero, exists
and it is equal to the arc length of the path. It is to be mentioned, also, that the definition
of the arc length makes sense also for piecewise smooth curves, because in this case the
tangent vector has only a finite number of discontinuity points and, therefore, its norm
is integrable.
We are going to show that the arc lengths of two equivalent paths between corre-
spondents points are equal, therefore the arc length is, in a way, a characteristic of the
support
3
.
Indeed, let r(t) = ρ(λ(t)), then r
·
(t) = λ
·
(t)ρ

(λ(t)). Therefore,

t
2
t
1
|r
·
(t)|dt

=

t
2
t
1

(λ(t))| |λ
·
|dt

=
=

t
2
t
1

(λ)| λ
·
(t)dt
....

=

λ
2
λ
1

(λ)|dλ

.
For naturally parameterized curves, (I, r = r(s)),
l
s
1
,s
2
= |s
2
− s
1
|.
In particular, if 0 ∈ I (which can always be assumed, since the translation is a diffeomor-
phism), then l
0,s
= |s|, i.e., up to sign, the natural parameter is the arc length.
Proposition 1. For any regular parameterized curve there is a naturally parameterized
curve equivalent to it.
Proof. Let (I, r = r(t)) be a regular parameterized curve, t
0
∈ I, and
λ : I → R, t →

t
t
0
|r
·
(τ)|dτ.
The function λ is smooth and strictly increasing, since λ
·
(t) = |r
·
(t)| > 0. Therefore, its
image will be an open interval J, while the function λ : I → J will be a diffeomorphism.
The parameterized curve (J, ρ(s) = r(λ
−1
(s)) is equivalent to (I, r) and it is naturally
parameterized, since ρ

(s) = r
·

−1
(s))(λ
−1
)
·
(s), while

−1
)
·
(s) =
1
λ
·

−1
(s))
=
1
|r
·

−1
(s))|
3
In a way, because we may represent the same set of points as support of another parameterized curve,
which is not equivalent to the initial one. The new path might have a different arc length between the same
points of the support.
1.2. Parameterized curves (paths) 21
and, hence,

(s)| = |r
·

−1
(s))| |(λ
−1
)
·
(s)| = 1.

Remark. In the proof of the previous proposition we used, in an essential way, the fact
that all the points of the curve are regular. On an interval on which the curve has singular
points, there is no naturally parameterized curve equivalent to it.
Example. For the circular helix

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a cos t
y = a sin t
z = bt,
we get, through a parameter change,
s(t) =

t
0
|r
·
(τ)|dτ =

t
0
|¦−a sin τ, a cos τ, b¦|dτ =

a
2
+ b
2
,
therefore,
t =
s

a
2
+ b
2
.
Thus, the natural parameterization of the helix is given by the equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a cos
s

a
2
+b
2
y = a sin
s

a
2
+b
2
z = b
s

a
2
+b
2
.
Exercise 1.2.1. Find a natural parameterization of the curve
x = e
t
cos t, , y = e
t
sin t, z = e
t
.
Exercise 1.2.2. Show that the parameter along the curve
x =
s
2
cos

ln
s
2

, y =
s
2
sin

ln
s
2

, z =
s

2
is a natural parameter.
22 Chapter 1. Space curves
Remark. It should be noticed that, usually, the natural parameter along a parameterized
curve cannot be expressed in finite terms (i.e. using only elementary functions) with
respect to the parameter along the curve. This is, actually, impossible even for very
simple curves, such that the ellipse

¸
¸
¸
¸
¸
¸
x = a cos t
y = b sin t,
with a b, for which the arc length can be expressed only in terms of elliptic functions
(this is, actually, the origin of their name!). Therefore, although the natural parameter is
very important for theoretical consideration and for performing the proofs, as the reader
will have more than once the opportunity to see in this book, for concrete examples of
parameterized curves we will hardly ever use it.
1.3 The definition of the curve
As we mentioned before, we can imagine, intuitively, a curve as being just a deformation
of a straight line, without thinking, necessary, at an analytical representation. We expect
the curve to have a well defined tangent at each point. In particular, this condition should
rule out both cusps and self intersections.
Definition 1.3.1. A subset M ⊂ R
3
is called a regular curve (or a 1-dimensional smooth
submanifold of R
3
) if, for each point a ∈ M there is a regular parameterized curve (I, r),
whose support, r(I), is an open neighbourhood in M of the point a (i.e. is a set of the
form M ∩ U, where U is an open neighborhood of a in R
3
), while the map r : I → r(I)
is a homeomorphism, with respect to the topology of subspace of r(I). A parameterized
curve with these properties is called a local parameterization of the curve M around the
point a. If for a curve M there is a local parameterization (I, r) which is global, i.e. for
which r(I) = M, the curve is called simple.
Remark. In some books in the definition it is required that the map r : I → r(I) be
smooth, which is not completely rigorous, as r(I) is not an open subset of R
3
. What is
meant, is, however, exactly the same, i.e. the map r : I → R
3
is smooth.
Examples. 1. Any straight line in R
3
is a simple curve, because it has a global para-
meterization, given by a function of the form r : R → R
3
, r(t) = a + b t, where a
and b are constant vectors, b 0.
2. The circular helix is a simple regular curve, with the global parameterization r :
R → R
3
, given by r(t) = (a cos t, b sin t, bt).
1.3. The definition of the curve 23
3. A circle in R
3
is a curve, but it is not simple, since no open interval can be home-
omorphic to the circle, which is a compact subset of R
3
.
Thus, a regular curve is just a subset of R
3
obtained “gluing together smoothly”
supports of parameterized curves. If we look carefully at the definition of the curve, we
see that not any regular parameterized curve can be used as a local parameterization of
a curve. For an arbitrary parameterized curve (I, r), the map r : I → R
3
is not injective
and, thus, it cannot be a local parameterization. Let us, also, mention that even if the
function is injective, r : I → r(I) might fail to be a homeomorphism (even if the map is
continuous and bijective, the inverse might not be continuous).
If, for instance, we consider the parameterized curve (I, r), with I = R and r : R →
R
3
,
r(t) = (cos t, sin t, 0),
then the support of this parameterized curve is the unit circle in the coordinate plane
xOy, centred at the origin. We should not conclude, however, that the circle is a simple
curve, since r is not a homeomorphism onto (in fact, it is not even injective, because it
is periodic).
Now let us assume that (I, r = r(t)) and (J, ρ = ρ(τ)) are two local parameterizations
of a regular curve M around the same point a ∈ M. Then, as one could expect, the two
parameterized curves are equivalent, if we restrict the definition intervals such that they
have the same support. More specifically, the following theorem holds:
Theorem 1.3.1. Let M ⊂ R
3
be a regular curve and (I, r = r(t)), (J, ρ = ρ(τ)) – two
local parameterizations of M such that W ≡ r(I) ∩ρ(J) ∅. Then (r
−1
(W), r|
r
−1
(W)
) and

−1
(W), ρ|
ρ
−1
(W)
) are equivalent parameterized curves.
Proof.
(I, r(t) = ( ˜ x(t), ˜ y(t), ˜ z(t))
and
(J, ρ(τ) = (x(τ), y(τ), z(τ))
be two local parameterizations of the curve M. To simplify the notations, we shall
assume, from the very beginning, that r(I) = ρ(J). Clearly, the generality is not reduced
by this assumption. We claim that the map λ : I → J, λ = ρ
−1
◦ r, is a diffeomorphism,
providing, thus, a parameter change between the two parameterized curves.
λ is, clearly, a homeomorphism, as a composition of the homeomorphisms
r : I → r(I)
24 Chapter 1. Space curves
and
ρ
−1
: ρ(J) → J.
Moreover r = ρ ◦ λ. Therefore, all we have to prove is that the maps λ and λ
−1
are both
smooth. One might be tempted, at this point, to write λ as
λ = ρ
−1
◦ r
and conclude that λ is smooth, as composition of two smooth functions. While this rep-
resentation is legitimate, as both r and ρ are homeomorphism onto, ρ
−1
is not a differen-
tiable function and, for the time being, at least, it doesn’t make sense to speak about its
differentiability, as its domain of definition is not an open subset of the ambient Euclid-
ean space. We will prove, instead that, locally, ρ
−1
is the restriction of a differentiable
mapping, defined, this time, on an open subset of R
3
.
Since the notion of differentiability is a local notion, it is enough to prove that λ
is smooth in a neighbourhood of each point of the interval I. This could be achieved,
for instance, by representing λ, locally, as a composition of smooth functions. Let t
0

I, τ
0
= λ(t
0
). Due to the regularity of the map ρ, we have ρ


0
) 0. We can assume,
without restricting the generality, that the first component of this vector is different from
zero, i.e. x
·

0
) 0. From the inverse function theorem, applied to the function x,
there is an inverse function τ = f (x), defined and smooth in an open neighbourhood
V ⊂ R of the point x
0
= x(τ
0
). Then, in the open neighbourhood ρ( f (V)) of the point
(x(τ
0
), y(τ
0
), z(τ
0
)) from M we will have ρ
−1
(x, y, z) = f (x), which means that, in fact,
we have
ρ
−1

ρ( f (V))
= f ◦ pr
1
|
ρ( f (V))
,
where pr
1
: R
3
→ R is the projection of R
3
on the first factor.
With this local expression of ρ
−1
in hand, we can write λ on the open neighbourhood
r
−1
(ρ( f (V))) of t
0
as
λ|
r
−1
(ρ( f (V)))
= ρ
−1

ρ( f (V))
◦ r|
r
−1
(ρ( f (V)))
= f ◦ pr
1
|
ρ( f (V))
◦ r|
r
−1
(ρ( f (V)))
.
As the functions f , pr
1
and r are, all of them, smooth on the indicated domains, it follows
that λ is smooth on the open neighbourhood r
−1
(ρ( f (V))) of t
0
. As t
0
was arbitrary, λ is
smooth on the entire I. The smoothness of λ
−1
is proven in the same way, changing the
roles of r and ρ.
It follows from the definition that any regular curve is, locally, the support of a
parameterized curve. Globally, this is not true, unless the curve is simple. Also, in
1.3. The definition of the curve 25
general, the support of an arbitrary parameterized curve is not a regular curve. Take, for
instance, the lemniscate of Bernoulli (R, r(t) = (x(t), y(t), z(t))), where

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x(t) =
t(1+t
2
)
1+t
4
y(t) =
t(1−t
2
)
1+t
4
z = 0
.
r is continuous, even bijective, but the inverse is not continuous. In fact, the support has
a self intersection, because lim
t→−∞
r = lim
t→∞
r = r(0) (see the figure 1.3). However,
we can always restrict the domain of definition of a parameterized curve such that the
support of the restriction is a regular curve.
Figure 1.3: The Bernoulli’s lemniscate
Theorem 1.3.2. Let (I, r = r(t)) a regular parameterized curve. Then each point t
0
∈ I
has a neighbourhood W ⊂ I such that r(W) is a simple regular curve.
Proof. The regularity of r at each point means, in particular, that r
·
(t
0
) 0. Without
restricting the generality, we may assume that x
·
(t
0
) 0. Let us consider the mapping
ψ : I × R
2
→ R
3
, given by
ψ(t, u, v) = r(t) + (0, u, v),
where (u, v) ∈ R
2
. ψ is, clearly, smooth and its Jacobi matrix at the point (t
0
, 0, 0) is
26 Chapter 1. Space curves
given by
J(ψ)(t
0
, 0, 0) =
,
¸
¸
¸
¸
¸
¸
¸
¸
¸
x
·
(t
0
) 0 0
y
·
(t
0
) 1 0
z
·
(t
0
) 0 1
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
Its determinant is
det J(ψ)(t
0
, 0, 0) = x
·
(t
0
),
therefore ψ is a local diffeomorphism around the point (t
0
, 0, 0). Accordingly, there exist
open neighborhoods U ⊂ R
3
of (t
0
, 0, 0) and V ⊂ R
3
of ψ(t
0
, 0, 0) such that ψ|
V
is a
diffeomorphism from U to V. Let us denote by ϕ : V → U its inverse (which, of course,
is, equally, a diffeomorphism from V to U, this time). If we put
W := ¦t ∈ I : (t, 0, 0) ∈ U¦ ,
then, clearly, W is an open neighborhood of t
0
in I such that
ϕ(V ∩ r(W)) = ϕ(ψ(W × ¦(0, 0)¦) = W × ¦(0, 0)¦.

Remark. The previous theorem plays a very important conceptual role. It just tells us
that any local property of regular parameterized curves is valid, also, for regular curves,
if it is invariant under parameter changes, without the assumptions of the parameterized
curves being homeomorphisms onto. Of course, all the precautions should be taken
when we investigate the global properties of regular curves.
1.4 Analytical representations of curves
1.4.1 Plane curves
A regular curve M ⊂ R
3
is called plane if it is contained into a plane π. We shall,
usually, assume that the plane π coincides with the coordinate plane xOy and we shall
use, therefore, only the coordinates x and y to describe such curves.
Parametric representation. We choose an arbitrary local parameterization (I, r(t) =
(x(t), y(t), z(t))) of the curve. Then the support r(I) of this local parameterization is an
open subset of the curve. For a global parameterization of a simple curve, r(I) is the
1.4. Analytical representations of curves 27
entire curve. Thus, any point a of the curve has an open neighbourhood which is the
support of the parameterized curve

¸
¸
¸
¸
¸
¸
x = x(t)
y = y(t)
. (1.4.1)
The equations (1.4.1) are called the parametric equations of the curve in the neighbour-
hood of the point a. Usually, unless the curve is simple, we cannot use the same set of
equations to describe the points of an entire curve.
Explicit representation. Let f : I → R be a smooth function, defined on an open
interval from the real axis. Then its graph
C = ¦(x, f (x)) | x ∈ I¦, (1.4.2)
is a simple curve, which has the global parameterization

¸
¸
¸
¸
¸
¸
x = t
y = f (t)
. (1.4.3)
The equation
y = f (x) (1.4.4)
is called the explicit equation of the curve (1.4.2).
In the literature for the explicit representation of a curve it is, also, used the term non-
parametric form, which we do not find particularly appealing since, in fact, an explicit
representation can be thought of as a particular case of parameter representation (the
parameter being the coordinate x).
Implicit representation. Let F : D → R be a smooth function, defined on a domain
D ⊂ R
2
, and let
C = ¦(x, y) ∈ D | F(x, y) = 0¦ (1.4.5)
be the 0-level set of the function F. Generally speaking, C is not a regular curve (we can
only say that it is a closed subset of the plane). Nevertheless, if at the point (x
0
, y
0
) ∈ C
the vector grad F = ¦∂
x
F, ∂
y
F¦ is non vanishing, for instance ∂
y
F(x
0
, y
0
) 0, then, by
the implicit functions theorem there exist:
• an open neighborhood U of the point (x
0
, y
0
) in R
2
;
28 Chapter 1. Space curves
• a smooth function y = f (x), defined on an open neighborhood I ⊂ R of the point
x
0
,
such that
C ∩ U = ¦(x, f (x)|x ∈ I¦.
If grad F 0 in all the points of C, then C is a regular curve (although, in general, not a
simple one).
Figure 1.4: The bisectors of the coordinate axes
Examples. 1. F : R
2
→ R, F(x, y) = x
2
+ y
2
− 1. Let
(x
0
, y
0
) ∈ C = ¦(x, y) ∈ R
2
| x
2
+ y
2
− 1 = 0¦.
Then we have
grad F(x
0
, y
0
) = ¦2x
0
, 2y
0
¦.
Obviously, since x
2
0
+ y
2
0
= 1, the vector grad F cannot vanish on C and, thus, C is
a curve (the unit circle, centred at the origin).
2. F : R
2
→ R, F(x, y) = x
2
− y
2
. C is not a curve in this case (the gradient is
vanishing at the origin). In fact, the set C has a self intersection at the origin (C
is just the union of the two bissectors of the coordinate axes, see figure 1.4). It
might not be obvious why we encounter problems in the neighbourhood of the
origin for this “curve”. The point is that there is no neighbourhood of the origin
(on C), homeomorphic to an open interval on the real axis. An neighbourhood of
the origin of C is the intersection of an open neighbourhood of the origin in the
plane and the set C. Now, if we restrict the neighbourhood of the origin in the
plane, its intersection with the set C will be a cross. If we remove the origin from
the cross, the remaining set will have four connected components. On the other
hand, suppose there is a homeomorphism f from the cross to an open interval
from the real axis. If we remove from the interval the image of the origin through
the homeomorphism f , we will get, clearly, only two connected components, or it
can be proved that the number of connected components resulted by removing a
point is homeomorphisms-invariant.
Remark. It should be clear that the condition of nonsingularity of the gradient of F
is only a sufficient condition for the equation F(x, y) = 0 to represent a curve. If the
gradient of F is zero at a point, we cannot claim that the equation represent a curve in
1.4. Analytical representations of curves 29
the neighborhood of that point, but we cannot claim the opposite, either. Consider, as a
trivial example, the equation
F(x, y) ≡ (x − y)
2
= 0.
Then we have
grad F(x, y) = 2¦x − y, −(x − y)¦
and if we denote
C = ¦(x, y) ∈ R
2
| F(x, y) = 0¦,
then grad F = 0 at all the points of C. But, clearly, C is a curve (it is easy to see that it is
the first bissector of the coordinate axes, i.e. a straight line).
1.4.2 Space curves
Parametric representation. As in the case of plane curves, with a local parameteriza-
tion

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = x(t)
y = y(t)
z = z(t)
(1.4.6)
we can represent either the entire curve, or only a neighbourhood of one of its points.
Explicit representation. If f , g : I → R are two smooth functions, defined on an open
interval from the real axis, then the set
C = ¦(x, f (x), g(x)) ∈ R
3
| x ∈ I¦ (1.4.7)
is a simple curve, with a global parameterization given by

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = t
y = f (t)
z = g(t)
. (1.4.8)
The equations

¸
¸
¸
¸
¸
¸
y = f (x)
z = g(x)
(1.4.9)
are called the explicit equations of the curve. Let us note that, in fact, each equation of
the system (1.4.9) is the equation of a cylindrical surface, with the generators parallel
to one of the coordinate axis. Therefore, representing explicitly a curve actually means
representing it as an intersection of two cylindrical surfaces, with the two families of
generators having orthogonal directions.
30 Chapter 1. Space curves
Implicit representation. Let F, G : D → R, defined on a domain D ⊂ R
3
. We consider
the set
C = ¦(x, y, z) ∈ D | F(x, y, z) = 0, G(x, y, z) = 0¦,
in other words, the solutions’ set for the system

¸
¸
¸
¸
¸
¸
F(x, y, z) = 0,
G(x, y, z) = 0.
(1.4.10)
In the general case, the set C is not a regular curve. Nevertheless, if at the point a =
(x
0
, y
0
, z
0
) ∈ C the rank of the Jacobi matrix
¸

x
F ∂
y
F ∂
z
F

x
G ∂
y
G ∂
z
G

(1.4.11)
is equal to two, then there is an open neighborhood U ⊂ D of the point (x
0
, y
0
, z
0
) such
that C ∩ U — the set of solutions of the system (1.4.10) in U — is a curve. Indeed,
suppose, for instance, that
det
¸

y
F(a) ∂
z
F(a)

y
G(a) ∂
z
G(a)

0.
Then, from the implicit functions theorem, there is an open neighborhood U ⊂ D such
that the set C ∩ U can be written as
C ∩ U = ¦(x, f (x), g(x))|x ∈ W¦,
where W is an open neighborhood in R of the point x
0
, while y = f (x), z = g(x) are
smooth functions, defined on W. Clearly, C ∩ U is a simple curve, while the pair
(W, r(t) = (t, f (t), g(t)) is a global parameterization of it.
If the rank of the matrix (1.4.11) is equal to two everywhere, then C is a curve
(although, generally, not a simple one).
Example (The Viviani’s temple). An important example of space curve given by implicit
equations is the so-called temple of Viviani
4
. This curve is obtained as the intersection
between the sphere with the center at the origin and radius 2a and the circular cylinder
4
Vincenzo Viviani (1622–1703) was a Italian mathematician and architect, who was in contact with
Galileo Galilei in the last years of the great scientist, and who liked to recommend himself as “Galileo’s
last student”.
1.4. Analytical representations of curves 31
with radius a and axis parallel to the z-axis, situated at a distance a from this axis. In
other words, the equations of the Viviani’s temple are

¸
¸
¸
¸
¸
¸
x
2
+ y
2
+ z
2
= 4a
2
,
(x − a)
2
+ y
2
= a
2
.
It is instructive to make some computations for the case of the Viviani’s temple. As we
shall see, it is not, globally, a curve. We will have to eliminate a point to get, indeed, a
regular curve. In fact, the shape of Viviani’s temple is easy to understand. The curve has
a shape which is similar to a Bernoulli’s lemniscate, lying on the surface of a sphere. So,
let

¸
¸
¸
¸
¸
¸
F(x, y, z) = x
2
+ y
2
+ z
2
− 4a
2
,
G(x, y, z) = (x − a)
2
+ y
2
− a
2
.
Then the equations of the curve reads

¸
¸
¸
¸
¸
¸
F(x, y, z) = 0,
G(x, y, z) = 0.
We have, now,
¸
F
·
x
F
·
y
F
·
z
G
·
x
G
·
y
G
·
z

=
¸
2x 2y 2z
2(x − a) 2y 0

= 2
¸
x y z
x − a y 0

.
To get a singular point, the following system of equations has to be fulfilled

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
y = 0
yz = 0
(x − a)z = 0
.
Clearly, the only solution of the system that verifies, also, the equations of the curve
is x = 2a, y = z = 0. Thus, the Viviani’s temple (see figure 1.5) is a regular curve
everywhere except the point which has these coordinates. It is not difficult to show that
Viviani’s temple is, in fact, the support of the parameterized curve
r(t) = (a(1 + cos t), a sin t, 2a sin
t
2
),
with t ∈ (−2π, 2π). If we compute r
·
(t) we get
r
·
(t) = ¦−a sin t, a cos t, a cos
t
2
¦,
32 Chapter 1. Space curves
Figure 1.5: Viviani’s temple
which means that this parameterized curve is regular. In particular, the existence of
this regular parametric representation of the Viviani’s temple shows that the point of
coordinates x = 2a, y = z = 0 is, in fact, a point of self-intersection, rather then a
singular point.
1.5 The tangent and the normal plane. The normal at a plane
curve
Definition 1.5.1. For a parameterized curve r = r(t) the vector r
·
(t
0
) is called the tangent
vector or the velocity vector of the curve at the point t
0
. If the point t
0
is regular, then the
straight line passing through r(t
0
) and having the direction of the vector r
·
(t
0
) is called
the tangent to the curve at the point r(t
0
) (or at the point t
0
).
The vectorial equation of the tangent line reads, thus:
R(τ) = r(t
0
) + τr
·
(t
0
). (1.5.1)
1.5. The tangent and the normal plane 33
Example. The cylindrical helix has the parameterization
r(t) = (a cos t, a sin t, bt),
therefore, for a point t
0
,
r
·
(t
0
) = ¦−a cos t
0
, a sin t
0
, b¦.
Thus, the equation of the tangent is
R(τ) =(a cos t
0
− τa sin t
0
, a sin t
0
+ τa cos t
0
, bt
0
+ τb) =
=(a(cost
0
− τ sin t
0
), a(sint
0
+ τ cos t
0
), b(t
0
+ τ)).
Proposition 1.5.1. The tangent vectors of two equivalent parameterized curves at cor-
responding points are colinear, while the tangent lines coincide.
Proof. Let (I, r = r(t)) and (J, ρ = ρ(s)) the two equivalent parameterized curves and
λ : I → J the parameter change, i.e. r = ρ(λ(t)). Then, according to the chain rule,
r
·
(t) = ρ

(λ(t)) λ
·
(t),
with λ
·
(t) 0.
Remarks. 1. Clearly, r
·
and ρ

have the same sense when λ
·
> 0 (the parameter
change does not modify the sense in which the support of the parameterized curve
is traversed) and they have opposite sense when λ
·
< 0.
2. Since the parameter change modifies the tangent vector, it doesn’t make sense to
define the tangent vector at a point of a regular curve, through a local parameter-
ization. Nevertheless, as we have seen, only the orientation and the length of the
tangent vector can vary, but not the direction. Thus, it makes sense to speak about
the tangent line at a point of a regular curve, defined through any local parameter-
ization of the curve around that point.
We can use a more “geometric” way to define the tangent to a parameterized curve.
Let r(t
0
+ ∆t) be a point of the curve close to the point r(t
0
). Then, according to the
Taylor’s formula,
r(t
0
+ ∆t) = r(t
0
) + ∆t r
·
(t
0
) + ∆t , (1.5.2)
with lim
∆t→0
= 0. We consider an arbitrary straight line π, passing through r(t
0
) and
having the direction given by the unit vector m. Let
d(∆t)
def
= d((r(t
0
+ ∆t), π).
34 Chapter 1. Space curves
Theorem1.5.1. The straight line π is the tangent line to the parameterized curve r = r(t)
at the point t
0
iff
lim
∆t→0
d(∆t)
|∆t|
= 0. (1.5.3)
Proof. From the Taylor’s formula (1.5.2), we have
∆r ≡ r(t
0
+ ∆t) − r(t
0
) = ∆t r
·
(t
0
) + ∆t .
The distance d(∆t) is equal to
|∆r × m| = |∆t||r
·
(t
0
) × m+ × m|.
Thus,
lim
∆t→0
d(∆t)
|∆t|
= lim
∆t→0
|r
·
(t
0
) × m+ × m
....
→0
| = |r
·
(t
0
) × m|.
Now, if the straight line π is the tangent line at t
0
, then the vectors r
·
(t
0
) and m are
colinear, therefore r
·
(t
0
) × m = 0.
Conversely, if the condition (1.5.3) is fulfilled, then |r
·
(t
0
) ×m| = 0, therefore either
r
·
(t
0
) = 0 (which cannot happen, since the parameterized curve is regular), either the
vectors r
·
(t
0
) and m are colinear, i.e. π is the tangent line at t
0
.
Remark. The condition (1.5.3) is expressed by saying that the tangent line and the curve
have a first order contact (or tangency contact). Another way of interpreting this formula
is that the tangent line is the limit position of the straight line determined by the chosen
point and a neighbouring point on the curve, when the neighbouring point is approaching
indefinitely to the given one.
Hereafter, if not specified otherwise, all the parameterized curves considered will be
regular.
Definition 1.5.2. Let r = r(t) be a parameterized curve and t
0
∈ I. The normal plane
at the point r(t
0
) of the curve r = r(t) is the plane which passes through r(t
0
) and it is
perpendicular to the tangent line to the curve at r(t
0
).
If r = r(t) is a plane parameterized curve (i.e. its support is contained into a plane,
which we will assume to be identical to the coordinate plane xOy), then the normal to
the curve at the point r(t
0
) will be the straight line through r(t
0
), which is perpendicular
to the tangent line to the curve at the point r(t
0
).
Remark. Because it makes sense to define the tangent line at a point of a regular curve,
by using an arbitrary local parameterization around that point, the same is true for the
normal plane (or the normal line, in the case of plane curves ).
1.5. The tangent and the normal plane 35
The vectorial equation of the normal plane (line) follows immediately from the def-
inition:
(R − r(t
0
)) r
·
(t
0
) = 0. (1.5.4)
1.5.1 The equations of the tangent line and normal plane (line) for different
representations of curves
Parametric representation
If we start from the vectorial equation (1.5.1) of the tangent line and project it on the
coordinate axes, we obtain the parametric equations of the tangent line, i.e., for space
curves,

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
X(τ) = x(t
0
) + τx
·
(t
0
),
Y(τ) = y(t
0
) + τy
·
(t
0
),
Z(τ) = z(t
0
) + τz
·
(t
0
),
(1.5.5)
and, for plane curves,

¸
¸
¸
¸
¸
¸
X(τ) = x(t
0
) + τx
·
(t
0
),
Y(τ) = y(t
0
) + τy
·
(t
0
).
(1.5.6)
If we eliminate the parameter τ, we get the canonical equations:
X − x
x
·
=
Y − y
y
·
=
Z − z
z
·
, (1.5.7)
for space curves, respectively
X − x
x
·
=
Y − y
y
·
, (1.5.8)
for plane curves.
As for the equation of the normal plane (line), we can obtain it from (1.5.4), express-
ing it as
¦X − x, Y − y, Z − z¦ ¦x
·
, y
·
, z
·
¦ = 0,
for space curves and
¦X − x, Y − y¦ ¦x
·
, y
·
¦ = 0,
for plane curves. Expanding the scalar products, we get:
(X − x)x
·
+ (Y − y)y
·
+ (Z − z)z
·
= 0, (1.5.9)
for the equation of the normal plane to a space curve and, for the normal line to a plane
curve,
(X − x)x
·
+ (Y − y)y
·
= 0. (1.5.10)
36 Chapter 1. Space curves
Explicit representation
If we have a space curve given by the equations

¸
¸
¸
¸
¸
¸
y = f (x)
z = g(x)
,
then we can construct a parameterization

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = t
y = f (t)
z = g(t).
For the derivatives we obtain immediately the expressions

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x
·
= 1
y
·
= f
·
z = g
·
,
which, when substituted into the equations (1.5.7), give
X − x =
Y − f (x)
f
·
(x)
=
Z − g(x)
g
·
(x)
, (1.5.11)
while for the equation of the normal plane, after substituting the derivatives into the
equation (1.5.9), we obtain
X − x + (Y − f (x)) f
·
(x) + (Z − g(x))g
·
(x) = 0. (1.5.12)
For a plane curve given explicitly
y = f (x),
we have the parametric representation

¸
¸
¸
¸
¸
¸
x = t
y = f (t),
and, thus, the equation of the tangent line is
X − x =
Y − f (x)
f
·
(x)
(1.5.13)
1.5. The tangent and the normal plane 37
or, in a more familiar form,
Y − f (x) = f
·
(x)(X − x), (1.5.14)
while for the normal we get
X − x + (Y − f (x)) f
·
(x) = 0 (1.5.15)
ore
Y − f (x) = −
1
f
·
(x)
(X − x). (1.5.16)
Implicit representation
Let us consider a curve given by the implicit equations

¸
¸
¸
¸
¸
¸
F(x, y) = 0
G(x, y) = 0.
(1.5.17)
Let us suppose that, at a point (x
0
, y
0
, z
0
)
det
¸
F
·
y
F
·
z
G
·
y
G
·
z

0
Then, as we saw before, around this point, the curve can be represented as

¸
¸
¸
¸
¸
¸
y = f (x)
z = g(x),
(1.5.18)
i.e. the system (1.5.17) can be written as

¸
¸
¸
¸
¸
¸
F(x, f (x), g(x)) = 0
G(x, f (x), g(x)) = 0.
By computing the derivatives with respect to x of F and G, we get the system

¸
¸
¸
¸
¸
¸
F
·
x
+ f
·
(x)F
·
y
+ g
·
(x)F
·
z
= 0
G
·
x
+ f
·
(x)G
·
y
+ g
·
(x)G
·
z
= 0,
therefore

¸
¸
¸
¸
¸
¸
f
·
F
·
y
+ g
·
F
·
z
= −F
·
x
f
·
G
·
y
+ g
·
G
·
z
= −G
·
x
.
38 Chapter 1. Space curves
From this system, one can obtain f
·
and g
·
, through the Cramer method:
∆ =

F
·
y
F
·
z
G
·
y
G
·
z

not
=
D(F, G)
D(y, z)
hyp
0,

f
· =

−F
·
x
F
·
z
−G
·
x
G
·
z

=

F
·
z
F
·
x
G
·
z
G
·
x

not
=
D(F, G)
D(z, x)

g
· =

F
·
y
−F
·
x
G
·
y
−G
·
x

=

F
·
x
F
·
y
G
·
x
G
·
y

not
=
D(F, G)
D(x, y)
,
therefore

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
f
·
=
D(F,G)
D(z,x)
D(F,G)
D(y,z)
g
·
=
D(F,G)
D(x,y)
D(F,G)
D(y,z)
(1.5.19)
As we saw before, for the curve (1.5.18) the equations of the tangent are
X − x
0
=
Y − f (x
0
)
f
·
(x
0
)
=
Z − g(x
0
)
g
·
(x
0
)
or, using (1.5.19),
X − x
0
=
Y − f (x
0
)
D(F,G)
D(z,x)
D(F,G)
D(y,z)
=
Z − g(x
0
)
D(F,G)
D(x,y)
D(F,G)
D(y,z)
,
from where, having in mind that f (x
0
) = y
0
and g(x
0
) = z
0
, we have
X − x
0
D(F,G)
D(y,z)
=
Y − y
0
D(F,G)
D(z,x)
=
Z − z
0
D(F,G)
D(x,y)
.
For a plane curve
F(x, y) = 0,
if at the point (x
0
, y
0
) there is fulfilled the condition F
·
y
0, then, from the implicit
functions theorem, locally, y = f (x), hence the equation of the curve can be written as
F(x, f (x)) = 0.
By differentiating this relation with respect to x, one obtains
F
·
x
+ f
·
F
·
y
= 0 =⇒ f
·
= −
F
·
x
F
·
y
.
1.6. The osculating plane 39
Thus, from the equation of the tangent:
Y − y
0
= f
·
(x
0
)(X − x
0
),
we deduce
Y − y
0
= −
F
·
x
F
·
y
(X − x
0
)
or
(X − x
0
)F
·
x
+ (Y − y
0
)F
·
y
= 0,
while for the normal we obtain the equation
(X − x
0
)F
·
y
− (Y − y
0
)F
·
x
= 0.
1.6 The osculating plane
Definition 1.6.1. A parameterized curve r = r(t) is called biregular (or in general posi-
tion) at the point t
0
if the vectors r
·
(t
0
) and r
··
(t
0
) are not colinear, i.e.
r
·
(t
0
) × r
··
(t
0
) 0.
The parameterized curve is called biregular if it is biregular at each point
5
.
Remark. It is not difficult to check that the notion of a biregular point is independent
of the parameterization: if a point is biregular for a given parameterized curve, then its
corresponding point through any parameter change is, also, a biregular point.
Definition 1.6.2. Let (I, r) be a parameterized curve and t
0
∈ I – a biregular point. The
osculating plane of the curve at r(t
0
) is the plane through r(t
0
), parallel to the vectors
r
·
(t
0
) and r
··
(t
0
), i.e. the equation of the plane is
(R − r(t
0
), r
·
(t
0
), r
··
(t
0
)) = 0, (1.6.1)
or, expanding the mixed product,

X − x
0
Y − y
0
Z − z
0
x
·
y
·
z
·
x
··
y
··
z
··

= 0. (1.6.2)
5
Abiregular curve is also called in some books a complete curve. We find this terma little bit misleading,
since this terms has, usually, other meaning in the global theory of curves (and, especially, surfaces).
40 Chapter 1. Space curves
Theorem 1.6.1. The osculating planes to two equivalent parameterized curves at cor-
responding biregular points coincides.
Proof. Let (I, r = r(t)) and (J, ρ = ρ(s)) be two equivalent parameterized curves and
λ : I → J – the parameter change. Then
r(t) = ρ(λ(t)),
r
·
(t) = ρ

(λ(t)) λ
·
(t),
r
··
(t) = ρ

(λ(t)) (λ
·
(t))
2
+ ρ

(λ(t)) λ
··
(t).
Since λ
·
(t) 0, from these relations it follows that the system of vectors ¦r
·
(t), r
··
(t)¦
and ¦ρ

(λ(t)), ρ

(λ(t))¦ are equivalent, i.e. they generate the same vector subspace R
3
,
hence the osculating planes to the two parameterized curves at the corresponding points
t
0
and λ(t
0
) have the same directing subspace, therefore, they are parallel. As they have
a common point (since r(t
0
) = ρ(λ(t
0
)), they have to coincide.
Remark. From the previous theorem it follows that the notion of an osculating plane
makes sense also for regular curves.
As in the case of the tangent line, there is a more geometric way of defining the
osculating plane which is, at the same time, more general, since it can be applied also
for the case of the points which are not biregular.
Let r(t
0
) ¸ si r(t
0
+ ∆t) two neighboring points on a parameterized curve, with r(t
0
)
biregular. We consider a plane α, of normal versor e, passing through r(t
0
). and denote
d(∆t) = d(r(t
0
+ ∆t), α).
Theorem 1.6.2. α is the osculating plane to the parameterized curve r = r(t) at the
biregular point r(t
0
) iff
lim
∆t→0
d(∆t)
|∆t|
2
= 0,
i.e. the curve and the plane have a second order contact.
Proof. From the Taylor formula we have
r(t
0
+ ∆t) = r(t
0
) + ∆t r
·
(t
0
) +
1
2
(∆t)
2
r
··
(t
0
) + (∆t)
2
,
with lim
∆t→0
= 0.
On the other hand,
d(∆t) = |e (r(t
0
+ ∆t) − r(t
0
))| =
= |(e r
·
(t
0
)) ∆t +
1
2
(e r
··
(t
0
)) (∆t)
2
+ (e ) (∆t)
2
|.
1.6. The osculating plane 41
Thus,
lim
∆t→0
d(∆t)
|∆t|
2
= lim
∆t→0

e r
·
(t
0
)
∆t
+
1
2
(e r
··
(t
0
)) + e
....
→0

=
= lim
∆t→0

e r
·
(t
0
)
∆t
+
1
2
(e r
··
(t
0
))

.
If lim
∆t→0
d(∆t)
|∆t|
2
= 0, then e r
·
(t
0
) = 0 and e r
··
(t
0
) = 0, i.e. e | r
·
(t
0
) ×r
··
(t
0
), meaning that
α is the osculating plane.
The converse is obvious.
Remarks. (i) The previous theorem justifies the name of the osculating plane. Actu-
ally, the name (coined by Johann Bernoulli), comes from the Latin verb osculare,
which means to kiss end emphasizes the fact that, among all the planes that are
passing through a given point of a curve, the osculating plane has the higher order
(“the closer”) contact.
(ii) Defining the osculating plane through the contact, one can define the notion of an
osculating plane also for the points which are not biregular, but in that case any
plane passing through the tangent is an osculating plane, in the sense that it has a
second order contact with the curve. Saying that the osculating plane at a biregular
point is the only plane which has at that point a second order contact with the curve
is the same with saying that the osculating plane is the limit position of a plane
determined by the considered point and two neighbouring points when this ones
are approaching indefinitely to the given one. Also, on can define the osculating
plane at biregular point of a parameterized curve as being the limit position of a
plane passing through the tangent at the given point and a neighboring point from
the curve, when this point is approaching indefinitely the given one.
A natural question that one may ask is what happens with the osculating plane in
the particular case of a plane parameterized curve. The answer is given by the following
proposition, whose prove is left to the reader:
Proposition 1.6.1. If a biregular parameterized curve is plane, i.e. its support is con-
tained into a plane π, then the osculating plane to this curve at each point coincides
to the plane of the curve. Conversely, if a given biregular parameterized curve has the
same osculating plane at each point, then the curve is plane and its support is contained
into the osculating plane.
42 Chapter 1. Space curves
1.7 The curvature of a curve
Let (I, r = r(t)) be a regular parameterized curve. Let (J, ρ = ρ(s)) be a naturally para-
meterized curve, equivalent to it. Then |ρ

(s)| = 1, while the vector ρ

(s) is orthogonal
to ρ

(s)
6
.
One can show that ρ

(s) does not depend on the choice of the naturally parameter-
ized curve equivalent to the given curve r = r(t). Indeed, if (J
1
, ρ
1
= ρ
1
( ˜ s)) is another
equivalent naturally parameterized curve with the parameter change ˜ s = λ(s), then, from
the condition

(s)| = |ρ
·
1
(λ(s))| = 1
we get |λ
·
(s)| = 1 for any s ∈ J. Thus, λ
·
= ±1 and, therefore,
˜ s = ±s + s
0
, where s
0
is a constant. It follows that
ρ

(s) = ρ

1
( ˜ s) (λ
·
(s))
2
....
=1

1
λ
··
(s)
....
=0
= ρ

1
( ˜ s).
Definition 1.7.1. The vector k = ρ

(s(t)) is called the curvature vector of the parame-
terized curve r = r(t) at the point t, while its norm, k(t) = |ρ

(s(t))| – the curvature of
the parameterized curve at the point t.
We shall express nowthe curvature vector k(t) as a function of r(t) and its derivatives.
We choose as natural parameter the arc length of the curve. Then we have
r(t) = ρ(s(t)) ⇒
r
·
(t) = ρ

(s(t)) s
·
(t)
r
··
(t) = ρ

(s(t)) (s
·
(t))
2
+ ρ

(s(t)) s
··
(t),
where s
·
(t) = |r
·
(t)|, s
··
(t) = |r
·
(t)|
·
=
d
dt


r
·
(t) r
·
(t)

=
r
·
(t) r
··
(t)
|r
·
(t)|
. Thus, we have
k(t) = ρ

(s(t)) =
r
··
|r
·
|
2

r
·
r
··
|r
·
|
4
r
·
. (1.7.1)
Now, since the vectors ρ

and ρ

are orthogonal, while ρ

has unit length, we have
k(t) = |k(t)| = |ρ

| = |ρ

× ρ

|.
6
Indeed, since ρ

is a unit vector, we have ρ
2
= 1. Differentiating this relation, we obtain that ρ

ρ

= 0,
which expresses the fact that the two vectors are orthogonal.
1.7. The curvature of a curve 43
Substituting ρ

=
r
·

|
, and ρ

by the formula (1.7.1) we obtain
k(t) =
|r
·
× r
··
|
|r
·
|
3
. (1.7.2)
Remarks. 1. From the formula (1.7.2) it follows that a parameterized curve r = r(t)
is biregular at a point t
0
iff k(t
0
) 0.
2. Since for equivalent parameterized curve the naturally parameterized curve to each
of them are equivalent between them, the notion of curvature makes sense also for
regular curves.
Examples. 1. For the straight line r = r
0
+ ta the curvature vector (and, therefore,
also the curvature) is identically zero.
2. For the circle S
1
R
= ¦(x, y, z) ∈ R
3
|x
2
+ y
2
= R
2
, z = 0¦ we choose the parameteri-
zation

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = Rcos t
y = Rsin t
z = 0
0 < t < 2π.
Then
r
·
(t) = ¦−Rsin t, Rcos t, 0¦, r
··
(t) = ¦−Rcos t, Rsin t, 0¦
and, thus, |r
·
(t)| = R, r
·
(t) r
··
(t) = 0. Therefore, for the curvature vector we get
k(t) =


1
R
cos t, −
1
R
sin t, 0

= −
1
R
¦x, y, z¦, k(t) =
1
R
.
Remarks. 1. The computation we just made explains why the inverse of the curvature
is called the curvature radius of the curve.
2. We saw that the curvature of a straight line is identically zero. The converse is
also true, in some sense, as the following proposition shows.
Proposition 1.7.1. If the curvature of a regular parameterized curve is identically zero,
then the support of the curve lies on a straight line.
Proof. Suppose, for the very beginning, that we are dealing with a naturally parameter-
ized curve (I, ρ = ρ(s)). From the hypothesis, ρ

(s) = 0, therefore ρ

(s) = a = const,
ρ(s) = ρ
0
+ sa, i.e. the support ρ(I) lies on a straight line. As two equivalent pa-
rameterized curves have the same support, the result still holds also for non-naturally
parameterized curves.
44 Chapter 1. Space curves
Remark. Be careful, the fact that a parameterized curve has zero curvature simply means
that the support of the curve lies on a straight line, but it doesn’t necessarily means that
the parameterized curve is (the restriction of) an affine map from R to R
3
, nor that it is
equivalent to such a particular parameterized curve. We can consider, as we did before,
the parameterized curve r : R → R
3
, r = r
0
+ at
3
, where a is a constant, nonvanishing,
vector from R
3
. Then we have, immediately, r
·
(t) = 2at
2
and r
··
(t) = 3at, which means
that the velocity and the acceleration of the curve are parallel, hence the curve has zero
curvature but, as we also saw earlier, this parameterized curve is not equivalent to an
affine parameterized curve.
1.7.1 The geometrical meaning of curvature
Let us consider a naturally parameterized curve (I, r = r(s)). We denote by ∆ϕ(s) the
measure of the angle between the versors r(s) and r(s + ∆s). Then
|r(s + ∆s) − r(s)| = 2

sin
∆ϕ(s)
2

.
Therefore,
k(s) = |r
··
(s)| =
¸
¸
¸
¸
¸
lim
∆s→0
r(s + ∆s) − r(s)
∆s
¸
¸
¸
¸
¸
= lim
∆s→0
2

sin
∆ϕ(s)
2

|∆s|
=
= lim
∆s→0

∆ϕ(s)
∆s

sin
∆ϕ(s)
2

∆ϕ(s)
2

= lim
∆s→0

∆ϕ(s)
∆s

=


ds

.
Thus, if we have in mind that ∆ϕ(s) is the measure of the angle between the tangents to
the curve at s and s + ∆s, the last formula gives us:
Proposition 1.7.2. The curvature of a curve is the speed of rotation of the tangent line
to the curve, when the tangency point is moving along the curve with unit speed.
1.8 The Frenet frame (the moving frame) of a parameterized
curve
At each point of the support of a biregular parameterized curve (I, r = r(t)) one can
construct a frame of the space R
3
. The idea is that, if we want to investigate the local
properties of a parameterized curve around a given point of the curve, that it might be
easier to do that if we don’t use the standard coordinate system of R
3
, but a coordinate
1.8. The Frenet frame 45
system with the origin at the given point of the curve, while the coordinate axes have
some connection with the local properties of the curve. Such a coordinate system was
constructed, independently, at the middle of the XIX-th century, by the French mathe-
maticians Frenet and Serret.
Definition 1.8.1. The Frenet frame (or the moving frame) of a biregular parameterized
curve (I, r = r(t)) at the point t
0
∈ I is an orthonormal frame of the space R
3
, with
the origin at the point r(t
0
), the coordinate versors being the vectors ¦τ(t
0
), ν(t
0
), β(t
0
)¦,
where:
• τ(t
0
) is the versor of the tangent to the curve at t
0
, i.e.
τ(t
0
) =
r
·
(t
0
)
|r
·
(t
0
)|
.
τ(t
0
) is also called the unit tangent at the point t
0
;
• ν(t
0
) = k(t
0
)/k(t
0
) is the versor of the curvature vector:
ν(t
0
) =
k(t
0
)
k(t
0
)
and it is called the unit principal normal at the point t
0
.
• β(t
0
) = τ(t
0
) × ν(t
0
) it is called the unit binormal at the point t
0
.
• The axis of direction τ(t
0
) is, obviously, the tangent to the curve at t
0
.
• The axis of direction ν(t
0
) is called the principal normal. In fact, this straight line
is contained into the normal plane (since it is perpendicular on the tangent), but
it is also contained into the osculating plane. Thus, the principal normal is the
normal contained into the osculating plane.
• The axis of direction β(t
0
) is called the binormal. The binormal is the normal
perpendicular on the osculating plane.
• The plane determined by the vectors ¦τ(t
0
), ν(t
0
¦ is the osculating plane (O in the
figure 1.8).
• The plane determined by the vectors ¦ν(t
0
), β(t
0
)¦ is the normal plane (N in the
figure 1.8).
46 Chapter 1. Space curves
Figure 1.6: The Frenet frame of a parameterized curve
• The plane determined by the vectors ¦τ(t
0
), β(t
0
)¦ is called the rectifying plane ,
for reasons that will become clear later (R in the figure 1.8).
For a naturally parameterized curve (J, ρ = ρ(s)), the expressions of the vectors of the
Frenet frame are quite simple:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ(s) = ρ

(s)
ν(s) =
ρ

(s)

(s)|
β(s) ≡ τ(s) × ν(s) =
ρ

(s) × ρ

(s)

(s)|
. (1.8.1)
For an arbitrary biregular parameterized curve (I, r = r(t)) the situation is a little bit
more complicated. Thus, obviously, from the definition, at an arbitrary point t ∈ I,
τ(t) =
r
·
(t)
|r
·
(t)|
. (1.8.2)
1.8. The Frenet frame 47
Then, having in mind that
k(t) =
r
··
(t)
|r
·
(t)|
2

r
·
(t) r
··
(t)
|r
·
(t)|
4
r
·
(t)
and
k(t) =
|r
·
(t) × r
··
(t)|
|r
·
(t)|
3
,
we get
ν(t) ≡
k(t)
k(t)
=
|r
·
(t)|
|r
·
(t) × r
··
(t)|
r
··
(t) −
r
·
(t) r
··
(t)
|r
·
(t)| |r
·
(t) × r
··
(t)|
r
·
(t), (1.8.3)
while
β(t) ≡ τ(t) × ν(t) =
r
·
(t) × r
··
(t)
|r
·
(t) × r
··
(t)|
. (1.8.4)
Remark. The above computations show that, in practice, for an arbitrary parameterized
curve (I, r = r(t)) it is easier to compute directly τ and β and then compute ν by the
formula
ν = β × τ.
1.8.1 The behaviour of the Frenet frame at a parameter change
A notion defined for parameterized curves makes sense for regular curves iff it is in-
variant at a parameter change, in other words if it doesn’t change when we replace a
parameterized curve by another one, equivalent to it. The Frenet frame is “almost” in-
variant, i.e. we have
Theorem 1.8.1. Let (I, r = r(t)) and ρ = ρ(u) be two equivalent parameterized curves
with the parameter change λ : I → J, u = λ(t). Then, at the corresponding points t
and u = λ(t), their Frenet frame coincide if λ
·
(t) > 0 . If λ
·
(t) < 0, then the origins and
the unit principal normals coincide, while the other two pairs of versors have the same
direction, but opposite senses.
Proof. Since r(t) = ρ(λ(t)), the origins of the Frenet frames coincide in any case. As
seen before, the curvature vectors of two equivalent curves coincide and, thus, the same
is true for the unit principal normals. From r
·
(t) = ρ

(λ(t)) λ
·
(t) it follows the coinci-
dence of the Frenet frames for λ
·
(t) > 0. If λ
·
(t) < 0, the tangent vectors ρ

(u) and r
·
(t)
have opposite senses, and the same is true for their versors. From β = τ × ν it follows
that, in this case, the unit binormals also have opposite senses.
48 Chapter 1. Space curves
1.9 Oriented curves. The Frenet frame of an oriented curve
As we saw before, the Frenet frame of a parameterized curve is not invariant at a para-
meter change (well, at least not at any parameter change). Therefore, in order to be able
to use this apparatus also for regular curve, we need to make it invariant. The idea is to
modify a little bit the definition of the regular curve, imposing some further condition on
the local parameterization, to make sure that the parameter changes will not modify the
Frenet frames.
Definition 1.9.1. Two parameterized curves are (I, r = r(t)) and (J, ρ = ρ(u)) are called
positively equivalent if there is a parameter change λ : I → J, u = λ(t), with λ
·
(t) > 0,
∀t ∈ I.
Definition 1.9.2. An orientation of a regular curve C ⊂ R
3
is a family of local parame-
terizations ¦(I
α
, r
α
= r
α
(t))¦
α∈A
such that
a) C =

α∈A
r
α
(I
α
),
b) For any connected component C
b
αβ
of the intersection C
αβ
= r
α
(I
α
) ∩ r
β
(I
β
) with
α, β ∈ A the parameterized curves (I
b
α
, r
b
α
) and (I
b
β
, r
b
β
) with I
b
α
= r
−1
α
(C
b
αβ
), r
b
α
= r
α
|
I
b
α
,
I
b
β
= r
−1
β
(C
0
αβ
), r
b
β
= r
β
|
I
b
β
are positively equivalent.
Example. For the unit circle S
1
the following parameterizations:
(I
1
= (0, 2π), r
1
(t) = (cos t, sin t, 0))
and
(I
2
= (−π, π), r
2
(t) = (cos t, sin t, 0))
give an orientation of S
1
. C
12
= r
1
(I
1
)∩r
2
(I
2
) has two connected components (the upper
and the lower half circles).
Starting with the upper component, C
1
12
, we have
I
1
1
= r
−1
1
(C
1
12
) = (0, π),
I
1
2
= r
−1
2
(C
1
12
) = (0, π)
and the parameter change is the identity, λ : (0, π) → (0, π), λ(t) = t, ∀t ∈ (0, π),
therefore the two parameterized curves are, clearly, positively equivalent.
1.9. Oriented curves. The Frenet frame of an oriented curve 49
As for the lower connected component, C
2
12
, we get
I
2
1
= r
−1
1
(C
2
12
) = (π, 2π),
I
2
2
= r
−1
2
(C
2
12
) = (−π, 0)
and the parameter change is λ : I
2
1
→ I
2
2
, λ(t) = t − 2π, therefore, since λ
·
(t) = 1 > 0,
also this time the two local parameterizations are positively equivalent.
Definition 1.9.3. A regular curve C ⊂ R
3
, with an orientation, is called an oriented
regular curve.
Example. If C is a simple regular curve, it can be turned into an oriented curve by using
an orientation given by any global parameterization (I, r).
Remark. If C is a connected regular curve, it has only two distinct orientations, corre-
sponding to the two possible senses of moving along the curve.
Definition 1.9.4. A local parameterization (I, r) of an oriented regular curve C is called
compatible with the orientation defined by the family ¦(I
α
, r
α

α∈A
if on the intersections
r(I) ∩ r
α
(I
α
) the parameterized curves (I, r) and (I
α
, r
α
) are positively equivalent.
I
1
1
= r
−1
1
(C
1
12
) = (0, π),
I
1
2
= r
−1
2
(C
1
12
) = (0, π)
Remark. For an oriented regular curve C, with the orientation given by the family of
local parameterizations ¦(I
α
, r
α

α∈A
, one can define, by using the vectors r
·
α
(t), a sense
on each tangent line, since, passing to another local parameterization r
β
(t), the vectors
r
·
α
and r
·
β
have the same direction and the same sense (only their norms may differ).
The orientation itself can be given through the choice of a sense on each tangent
line. Thus, if the direction of the tangent vector at x ∈ C is given by the vector a(x), then
wave to impose the continuity of the map C → R
3
, x → a(x). For this definition of the
orientation, a local parameterization (I, r) is compatible with the orientation if, for each
point x ∈ C, x = r(t), the vectors a(x) and r
·
(t) have the same sense.
Definition 1.9.5. The Frenet frame of an oriented biregular curve C at a point x ∈ C
is the Frenet frame of a biregular parameterized curve r = r(t) at t
0
, where r = r(t)
is a local parameterization of the curve C, compatible with the orientation, such that
r(t
0
) = x.
Remark. Clearly, this definition does not depend on the choice of the local parameteri-
zation, compatible with the orientation of the curve.
50 Chapter 1. Space curves
1.10 The Frenet formulae. The torsion
Let (I, r = r(t)) be a biregular parameterized curve. Then the vectors τ(t), ν(t), β(t) are,
in fact, smooth vector functions with respect to the parameter t. We want to find their
derivatives with respect to t, more precisely, the decomposition of these derivatives with
respect to the vectors ¦τ, ν, β¦. These derivatives show, in fact, the way the vectors of the
Frenet frame vary along the curve. From the definition, we have,
τ =
r
·
|r
·
|
=
r
·

r
·2
.
Therefore,
τ
·
=
r
··
|r
·
| − r
·

r
·
r
··
|r
·
|
|r
·
|
2
=
r
··
|r
·
|
2
− r
·
(r
·
r
··
)
|r
·
|
3
=
= |r
·
|
|r
·
× r
··
|
|r
·
|
3
....
k
,
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
|r
·
|
|r
·
× r
··
|
r
··

r
·
r
··
|r
·
| |r
·
× r
··
|
r
·
....
ν
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
.
Thus,
τ
·
= |r
·
|k ν. (1.10.1)
Further,
β = τ × ν ⇒ β

= τ

× +τ × ν

= k( ν × ν
....
=0
) + τ × ν

⇒ β

= τ × ν

,
hence β

⊥ τ. On the other hand,
β β = 1 ⇒ β

β = 0 ⇒ β

⊥ β.
Thus, the vector β

is colinear to the vector ν = β × τ and we can write
β

= −|r
·
| χν,
where χ is a proportionality factor, the meaning of which will be made clear later.
We differentiate now the equality
ν = β × τ.
We have
ν

= β

× τ + β × τ

= −|r
·
| χ(ν × τ) + |r
·
| k(β × ν),
1.10. The Frenet formulae. The torsion 51
therefore,
ν

= |r
·
|(−kτ + χβ). (1.10.2)
We obtained, thus, the equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ

(t) = |r
·
|k(t)ν(t)
ν

(t) = |r
·
|(−k(t)τ(t) + χ(t)β(t))
β

(t) = −|r
·
|χ(t)ν(t).
(1.10.3)
These equations are called the Frenet formulae for the parameterized curve r = r(t). If
we are dealing win a naturally parameterized curve ρ = ρ(s), then the Frenet equations
are a little bit simpler:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ

(t) = k(t)ν(t)
ν

(t) = −k(t)τ(t) + χ(t)β(t)
β

(t) = −χ(t)ν(t).
(1.10.3
·
)
Definition 1.10.1. The quantity χ(t) is called the torsion (or second curvature) of the
biregular parameterized curve r = r(t) at the point t.
We shall compute first the torsion for a naturally parameterized curve ρ = ρ(s). For
such a curve, the versors of the Frenet frame are given by the expressions:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ = ρ

ν =
1
k
ρ

β =
1
k
ρ

× ρ

.
From the third Frenet formula, we have:
β

ν = −χ(s) ( ν × ν
....
=1
) = −χ(t).
But, on the other hand, from the definition,
β

=
¸
1
k

·
+
1
k
ρ

× ρ

....
=0
+
1
k
ρ

× ρ

,
hence
χ = −β

ν = −
¸
1
k

·

ρ

× ρ

1
k
ρ

....
=0

1
k

ρ

× ρ

1
k
ρ

,
52 Chapter 1. Space curves
therefore,
χ =
1
k
2

ρ

, ρ

, ρ

. (1.10.4)
Now, the following theorem gives us the way to compute the torsion for an arbitrary
(biregular) parameterized curve:
Theorem. If (I, r = r(t)) and (J, ρ = ρ(u)) are two positively equivalent parameterized
curves, with the parameter change λ : I → J, λ
·
> 0, then they have the same torsion at
the corresponding points t and u = λ(t).
Proof. Let ¦τ, ν, β¦, respectively ¦τ
1
, ν
1
, β
1
¦ be the Frenet frames of the two parameter-
ized curves at the corresponding points t and uλ(t), respectively, and χ and χ
1
their –
torsions at these points. Then
β
1
(λ(t)) = β(t)
ν
1
(λ(t)) = ν(t),
r
·
(t) = ρ

(λ(t)) λ
·
(t) ⇒ r
·
(t) =
d
du

1
(u))λ
·
(t).
From the last Frenet equation for the curve r, we get
β

(t) ν(t) = −|r
·
(t)| χ(t),
i.e.
χ(t) = −
1
|r
·
(t)|
β

(t) ν(t) = −
1

(λ(t))| λ
·
(t)
β
1

(λ(t)) λ
·
(t) ν
1
(λ(t)) =
= −
1

(λ(t))|

−|ρ

(λ(t))| χ
1
(λ(t))

= χ
1
(λ(t)),
where we used once more the last Frenet equation, but this time for the curve ρ, as well
as the fact that the vector ν
1
(λ(t)) has unit length.
Let now ρ = ρ(s) be a naturally parameterized curve, positively equivalent to the
parameterized curve r = r(t), where s = λ(t) is the parameter change. Then r and its
derivatives up to the third order can be expressed as functions of ρ and its derivatives as:
r(t) = ρ(λ(t))
r
·
(t) = ρ

(λ(t)) λ
·
(t)
r
··
(t) = ρ

(λ(t)) λ
·2
(t) + ρ

(λ(t)) λ
··
(t)
r
···
(t) = ρ

(λ(t)) λ
·3
(t) + 3ρ

(λ(t)) λ
·
(t) λ
··
(t) + ρ

(λ(t)) λ
···
(t),
1.10. The Frenet formulae. The torsion 53
therefore the mixed product of the first three derivatives of r reads

r
·
(t), r
··
(t), r
···
(t)

=

ρ

(λ(t)) λ
·
(t), ρ

(λ(t)) λ
·2
(t) + ρ

(λ(t)) λ
··
(t),
ρ

(λ(t)) λ
·3
(t) + 3ρ

(λ(t)) λ
·
(t) λ
··
(t) + ρ

(λ(t)) λ
···
(t)

=
= λ
·6
(t)

ρ

(λ(t)), ρ

(λ(t)), ρ

(λ(t))

,
all the other mixed products from the right hand side vanishing, because two of the
factors are colinear. Hence

ρ

(λ(t)), ρ

(λ(t)), ρ

(λ(t))

=
1
λ
·6
(t)

r
·
(t), r
··
(t), r
···
(t)

.
But, since ρ is naturally parameterized and the two curves are positively equivalent, we
have λ
·
= |r
·
|, therefore the previous formula becomes

ρ

, ρ

, ρ

=
(r
·
, r
··
, r
···
)
|r
·
|
6
.
Moreover (see (1.7.2)), the curvature can be expressed with respect to the derivatives of
r through the formula
k =
|r
·
× r
··
|
|r
·
|
3
,
hence, from the expression of the torsion (1.10.4) and the previous relation, we get
χ(t) =
(r
·
, r
··
, r
···
)
|r
·
× r
··
|
2
. (1.10.5)
Exercise 1.10.1. Let ω = χτ + kβ. Show that Frenet’s formulae can be written as

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ
·
= ω × τ
ν
·
= ω × ν
β
·
= ω × β
(1.10.6)
The vector ω is called the Darboux vector.
1.10.1 The geometrical meaning of the torsion
The torsion is, in a way, an analogue of the curvature (this is the reason why in the old-
fashioned books the torsion is called the second curvature). What we mean is that the
torsion can be also interpreted as being the speed of rotation of a straight line, this time
the binormal. In other words, we have
54 Chapter 1. Space curves
Proposition 1.10.1. If (I, r = r(s)) is a naturally parameterized curve and ∆α is the
angle of the osculating planes of the curve at r(s) and r(s + ∆s) (in other words, the
angle of the binormals of the curve at that points), then we have
χ(s) = lim
∆s→0
∆α
∆s
Notice that, this time, unlike the case of the curvature, the torsion is the algebraic
value of the limit, not the absolute value. We have to say, however, that the curvature
of space curve is defined to be positive because one could find no geometrical meaning
of a signed curvature. As we shall see in the sequel, for plane curve we can define a
signed curvature, whose absolute value is the curvature and which will help us to get
some more information about the curve.
As we said before, the torsion is analogue to the curvature. Thus, the curvature is a
measure of the deviation of a curve from a straight line. On the other hand, the torsion
is a measure of the deviation of the curve from a plane curve. More precisely, we have
Theorem 1.10.1. The support of a biregular parameterized curve lies in a plane iff the
torsion of the curve vanishes identically.
Proof. Let (I, r = r(t)) be a biregular parameterized curve such that r(I) ⊂ π, where π is
a plane. Then, obviously, the vectors r
·
(t) and r
··
(t) are parallel to this plane, which, as
we know, is the osculating plane of the curve. Therefore, β(t) = const, hence we have
0 = β

(t) = − |r
·
|
....
0
χ(t) ν(t)
....
0
⇒ χ(t) ≡ 0.
Conversely, if χ(t) ≡ 0, then the unit binormal, β(t), is always equal to a constant vector,
β
0
. But β(t) = r
·
(t) × r
··
(t), therefore the vector r
·
(t) is always perpendicular to the
constant vector β
0
. Thus, we have the series of implications
r
·
(t) β
0
= (r β
0
)
·
= 0 ⇒ r β
0
= const = r
0
β
0
⇒ (r − r
0
) × β
0
= 0,
i.e. the support of r(I) the curve is contained in a plane perpendicular to the constant
vector β
0
, and passing through the point r
0
.
1.10.2 Some further applications of the Frenet formulae
We saw that the curvature of a parameterized curve vanishes identically if and only if the
support of the curve lies on a straight line. On the other hand, we know that the curvature
of a circle is constant and equal to the inverse of the radius of the circle. We could expect
1.10. The Frenet formulae. The torsion 55
that the converse is also through, in other words that if the curvature of a parameterized
curve is constant, then its support lies on a circle. Unfortunately, this claim is not true.
It is enough to think about the circular cylindrical helix, which has constant curvature
(and, also, constant torsion). We have, however, the weaker result:
Proposition 1.10.2. If (I, r = r(s)) is a naturally parameterized curve with a curvature
k equal to a positive constant k
0
, while the torsion vanishes identically, then the support
of the curve lies on a circle of radius 1/k
0
.
Proof. Since the torsion is identically zero, the curve is plane. Let us consider the para-
meterized curve (I, r
1
= r
1
(s)), where
r
1
= r +
1
k
0
ν. (*)
We differentiate with respect to s and we get, using the second Frenet formula for r,
r
·
1
= r
·
+
1
k
0
ν

= τ +
1
k
0
(−k
0
τ) = τ − τ = 0.
Thus, the curve r
1
reduces to a point, say r
1
(s) ≡ c = const. But, from (*), we get
|r − c| =
¸
¸
¸
¸
¸
1
k
0
ν ν
¸
¸
¸
¸
¸
=
1
k
0
,
which means that any point of the support of the curve r is at a (constant) distance 1/k
0
from the fixed point c, i.e. the support r(I) lies on the circle of radius 1/k
0
, centred at
c.
Another interesting situation is when the support of the curve lies not in a plane, but
on a sphere. In this case we have
Proposition 1.10.3. If a naturally parameterized curve (I, r = r(s)) has the support on
a sphere centred at the origin and radius equal to a, then its curvature is subject to
k ≥
1
a
.
Proof. The distance from a point of the curve to the origin is equal to |r|, i.e. we have
r
2
= a
2
. Differentiating, we get r r
·
= 0 or r τ. Differentiating once more, we obtain
r
·
τ + r τ

= 0,
56 Chapter 1. Space curves
or
1 + r τ

= 0 ⇐⇒ 1 + kr ν = 0 =⇒ kr ν = −1.
From the properties of the scalar product, we have
|r ν| ≤ |r| |ν| = |r| = a,
therefore,
k = |k| =
1
|r ν|

1
|r| |ν|
=
1
a
.

In the figure ?? we give an example of a curve which lies on a sphere, it is a so-called
spherical helix, because it is both a general helix (see the next section) and a spherical
curve.
Figure 1.7: A spherical helix
1.10.3 General helices. Lancret’s theorem
Definition 1.10.2. A parameterized curve curve (I, r) is called a general helix if its
tangents make a constant angle with a fixed direction in space.
1.10. The Frenet formulae. The torsion 57
The following theorem was formulated in 1802 by the French mathematician Paul
Lancret, but the first known proof belongs to another celebrated French mathematician,
known especially for his contributions to mechanics, A. de Saint Venant (1845).
Theorem 1.10.2 (Lancret, 1802). A space curve with the curvature k > 0 is a general
helix if and only if the ratio between its torsion and its curvature is constant.
Proof. Let us assume, to begin with, that the curve is parameterized by the arc length.
To prove the first implication, let us suppose that r is a general helix and let c be the
versor of the fixed direction:
τ c = cos α
0
= const.
By differentiating, we get
τ

c = 0,
hence
k c ν = 0.
As, by hypothesis, k > 0, it follows that
c ν = 0,
i.e., at each point of the curve, c ⊥ ν. This means that c is in the rectifying plane and
therefore
β c = sin α
0
.
By differentiating the relation ν c = 0 we get, having in mind that c is constant and
using the second formula of Frenet:
(−kτ + χβ) c = 0
which leads to
−k cos α
0
+ χ sin α
0
= 0,
i.e.
χ
k
= cot α
0
= const.
Conversely, let us assume that
χ(s)
k(s)
= c
0
= const,
58 Chapter 1. Space curves
or
c
0
k − χ = 0.
On the other hand, from the first and the third of the Frenet’s equations, we get,
(c
0
k − χ)ν = c
0
τ
·
+ β
·
= 0.
We integrate once and we obtain
c
0
τ + β = c

,
where c

0 is a constant vector.
We define
c :=
c

|c

|
=
c
0
τ + β
|c
0
τ + β|
=
c
0
τ + β

1 + c
2
0
,
whence
c τ =
c
0

1 + c
2
0
= const ≤ 1.
Hence the vectors c ¸ si τ make a constant angle and the curve is a general helix.
We end this paragraph by noting, as a historical curiosity, that, although in most of
the books this theorem is credited to Lancret, in fact he (and, with him, Saint-Venant),
only gave one implication, the first one: on a general helix, the ratio between the torsion
and the curvature is constant. The other implication was given and proved later by Joseph
Bertrand (see, for instance, the book of Eisenhart [16]).
1.10.4 Bertrand curves
An interesting problem in the theory of curves is whether it is possible for several curves
to share the same family of tangents, principal normal or binormals. For the tangents,
the answer is easily seen to be negative: the family of tangents uniquely determines the
curve. For the principal normals, the problem, raised by the same Saint-Venant, was
answered by Joseph Bertrand, who discovered that, for an arbitrary curve, the answer is
negative, however, there are special curves for which there might be, also, other curves
with the same family of principal normals. These curves are called Bertrand curves.
Usually, for a Bertrand curve, there is only one curve having the same principal normals.
We will say that the two curves are Bertrand mates, or that they are associated, or
conjugated Bertrand curves. It turns out that if a Bertrand curve has more than one
1.10. The Frenet formulae. The torsion 59
Bertrand mate, then it has an infinity and the curve (and all of its mates) is a circular
cylindrical helix.
We shall prove, in what follows, a series of interesting results related to Bertrand
curves, without respecting the historical order. Let r and r

be two Bertrand mates. We
assume that the first curve is naturally parameterized. Then the second curve is, in fact,
also dependent on the arc length s of the first and we assume that r

(s) is the point on
the Bertrand mate having the same principal normal as the first one at s. The two points
are called corresponding. We have the following result:
Theorem 1.10.3 (Schell). The angle of the tangents of two associated Bertrand curves
at corresponding points is constant.
Proof. Clearly, if ν(s) is the versor of the principal normal of the first curve, then we
have
r

(s) = r(s) + a(s)ν(s). (1.10.7)
As the two curves have the same principal normals, the second curve ought to have the
versor of the principal normal
ν

(s) = ±ν(s), (1.10.8)
We differentiate the relation (1.10.7) with respect to s and we get
dr

ds
=
dr
ds
+ a

ds
+
da
ds
r (1.10.9)
or, using the second Frenet formula,
dr

ds
= (1 − ak) τ +
da
ds
ν + aχβ. (1.10.10)
The vector
dr

ds
is tangent to the second curve, therefore it is perpendicular both on ν

and
on ν. We deduce, by multiplying both sides of (1.10.10) by ν, that
da
ds
= 0, i.e. a is a
constant. As such, the relation (1.10.10) turns into
dr

ds
= (1 − ak) τ + aχβ. (1.10.11)
We denote by s

the arc length of the second curve. Then
τ

=
dr

ds

=
dr

ds
ds
ds

(1.10.12)
or, using (1.10.11)
τ

= (1 − ak)
ds
ds

τ + aχ
ds
ds

β. (1.10.13)
60 Chapter 1. Space curves
Let ω be the angle of the tangents of the two curvs at the corresponding points. Then
this angle is given by
cos ω = ττ

, (1.10.14)
where, of course, τ

is the versor of the second curve at the point r

(s). In terms of ω,
τ

and the versor of the binormal of the second curve„ β

, can be written as
τ

= cos ωτ + sin ωβ, (1.10.15)
β

= ε (−sin ωτ + cos ωβ) , (1.10.16)
where ε = ±1.
We differentiate the relation (1.10.15) with respect to s and we get:


ds
= −sin ω

ds
τ + k cos ων + cos ω

ds
β − χ sin ων. (1.10.17)
If we multiply scalarly both sides of the relation (1.10.17) by τ and then by β and use
the fact that, on the ground of the first Frenet formula,


ds
is colinear to ν

, and, thsu,
also to ν, we obtain the relations:
sin ω

ds
= 0, cos ω

ds
= 0, (1.10.18)
i.e.

ds
= 0, hence ω is constant.
The following theorem was proved by Joseph Bertrand and it is considered to be the
central result of the entire theory of Bertrand curves.
Theorem 1.10.4 (Bertrand). A curve r is a Bertrand curve if and only if its torsion and
curvature verify a relation of the form
a k + b χ = 1, (1.10.19)
with constant a and b.
Proof. Let us assume that r is a Bertrand curve. Comparing the relations (1.10.13) and
(1.10.15), we obtain that
cos ω = (1 − ak)
ds
ds

, (1.10.20)
sin ω = aχ
ds
ds

. (1.10.21)
1.10. The Frenet formulae. The torsion 61
Dividing side by side these two equalities, we obtain
ctg ω =
1 − ak

(1.10.22)
or
1 − ak = aχ ctg ω. (1.10.23)
If we denote
b = a ctg ω, (1.10.24)
we obtain the relation (1.10.19).
Let us suppose now, conversely, that the relation (1.10.19) holds. Let r

be the curve
given by (1.10.7), where a is the constant fromthe equation (1.10.19). Differentiating the
relation (1.10.7), we get, as we saw earlier, the relation (1.10.11). As the basis ¦τ, ν, β¦
is orthonormal we conclude from (1.10.11) that
¸
ds

ds

2
= (1 − ak)
2
+ a
2
χ
2
(1.10.25)
or, using (1.10.25),
¸
ds

ds

2
= (a
2
+ b
2

2
, (1.10.26)
whence
χ
ds
ds

= const. (1.10.27)
In the same way follows that
(1 − ak)
ds
ds

= const. (1.10.28)
Taking into account these two relations, we differntiate the relation (1.10.13) and we
obtain, by using the first and the third Frenet formulae for the curve r:


ds
=
,
k(1 − ak) − aχ
2
¸
ds
ds

ν (1.10.29)
or, using the first Frenet formula for the curve r

:
k

ν

=
,
k(1 − ak) − aχ
2
¸
¸
ds
ds

2
ν, (1.10.30)
whence it follows that ν

= ±ν, i.e., indeed, r is a Bertrand curve.
62 Chapter 1. Space curves
Corollary 1.10.1. The circular cylindrical helices, the curves of constant torsion (in
particular the plane curves), and the curves of constant curvature are Bertrand curves.
Remark. For a plane curve, the principal normals are, in fact, the normals, therefore,
as one can see easily, any plane curve has an infinity of Bertrand mates, all of them
congruent (they are, all of them, parallel to the given one).
Corollary 1.10.2. If a Bertrand curve has more than one mate, then it has an infinity
and it is a circular cylindrical helix.
Proof. As we saw from the proof of the theorem of Bertrand, the constant a from the
relation (1.10.19) is the one that identifies the Bertrand mate of our curve. Thus, if a
curve r has more then one Bertrand mate, this means that there are (at least) two distinct
pairs of real numbers (a, b), (a
1
, b
1
) such that
a k + b χ = 1,
a
1
k + b
1
χ = 1.
Subtracting side by side the two relations, we obtain
(a − a
1
) k = (b
1
− b) χ.
At least one of the coefficients (a − a
1
) and (b
1
− b) is different from zero, therefore
the ratio between the curvature and torsion is a constant. This means, via the Lancret’s
theorem, that the curve is a general helix. However, as, say, χ = c k, with c – constant,
from (1.10.19) follows that
(a + b c) k = 1,
which means that k (and, hence, also χ) is a constant, therefore the curve is a circular
cylindrical helix. Clearly, for the case of a cylindrical helix, when both the curvature
and the torsion are constants, we can get an infinity of pairs of real numbers that ver-
ify (1.10.19), so our curve has an infinity of Bertrand mates. They lie on circular cylin-
ders which have the same axis of symmetry with the cylinder associated to the given
curve.
1.11 The local behaviour of a parameterized curve around a
biregular point
Let (I, r) be a naturally parameterized curve. We shall assume that 0 ∈ I, as an interior
point and M
0
≡ r(0) is a biregular point of the curve. We shall make use of the following
notations: τ
0
= τ(0), ν
0
= ν(0), β
0
= β(0).
1.11. The local behaviour of a parameterized curve around a biregular point 63
We expand r in a Taylor series around the origin. Up to the third order, the Taylor
expansion is
r(s) = r(0) + sr
·
(0) +
1
2
s
2
r
··
(0) +
1
6
s
3
r
···
(0) + o(s
3
). (1.11.1)
We want to express the derivatives of r as functions of the Frenet vectors τ
0
, ν
0
, β
0
. We
have, obviously, since r is naturally parameterized,
r
·
(0) = τ
0
. (1.11.2)
On the other hand, from the definition of the curvature vector, we have
r
··
(0) = k(0) = k(0) ν
0
. (1.11.3)
Moreover, if we differentiate the relation
r
··
(s) = k(s) = k(s) ν (1.11.4)
we get
r
···
(s) = k
·
(s)ν + k(s)ν
·
= k
·
(s)ν + k(s)(−k(s)τ + χ(s)β) =
= −k
2
(s)τ + k
·
(s)ν + k(s)χ(s)β,
(1.11.5)
where we have made use of the second Frenet equation.
Substituting s = 0 in (1.11.5), we get
r
···
(0) = −k
2
(0)τ
0
+ k
·
(0)ν
0
+ k(0)χ(0)β
0
. (1.11.6)
Thus, the relation (1.11.1) becomes
r(s) −r(0) = sτ
0
+
1
2
s
2
k(0)ν
0
+
1
6
s
3

−k
2
(0)τ
0
+ k
·
(0)ν
0
+ k(0)χ(0)β
0

+o(s
3
) (1.11.7)
or
r(s) − r(0) =
¸
s −
1
6
k
2
(0)s
3
+ o(s
3
)

τ
0
+
¸
1
2
k(0)s
2
+
1
6
k
·
(0)s
3
+ o(s
3
)

ν
0
+
+
¸
1
6
k(0)χ(0)s
3
+ o(s
3
)

β
0
.
(1.11.8)
We consider now a coordinate frame with the origin at M
0
and having as axes the Frenet
axes. The position vector of a point of the curve with respect to this frame is exactly the
64 Chapter 1. Space curves
vector r(s) − r(0) ≡
−−−−→
M
0
M, where M = r(s). Therefore, projecting (1.11.8) on the axes,
we get

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x(s) = s −
1
6
k
2
(0)s
3
+ o(s
3
)
y(s) =
1
2
k(0)s
2
+
1
6
k
·
(0)s
3
+ o(s
3
)
z(s) =
1
6
k(0)χ(0)s
3
+ o(s
3
)
. (1.11.9)
It is not at all difficult to see that locally, close enough to 0, we have the following
relations between coordinates, giving, in fact, the equations of the projections of the
curve on the coordinate planes of the Frenet frame:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
y =
1
2
k(0)x
2
,
z =
1
6
k(0)χ(0)x
3
,
z
2
=
2
9
χ
2
(0)
k(0)
y
3
.
(1.11.10)
Thus, the projection of the curve on the xOy-plane (the osculating plane) is a parabola,
the projection on the xOz-plane (the rectifying plane) is a cubic, while the projection on
the yOz-plane (the normal plane) is a semicubic parabola.
1.12 The contact between a space curve and a plane
We consider a naturally parameterized space curve (I, r). We assume, as we did before,
that 0 is an interior point of the interval I and that the point M
0
= r(0) of the curve is
biregular. We consider a plane Πpassing through the point M
0
. We choose as coordinate
frame the Frenet frame of the curve r at the point M
0
, F
0
= ¦M
0
; τ
0
, ν
0
, β
0
¦. With respect
to this coordinate frame, he cartesian equation of the plane Π will be of the form
F(x, y, z) ≡ ax + by + cz = 0. (1.12.1)
Now, if x = x(s), y = y(s), z = z(s) are the local equations of the curve with respect to
the Frenet frame (see (1.11.9)), the intersection condition between the the plane and the
curve is F(x(s), y(s), z(s)) = 0, i.e.
a
¸
s −
1
6
k
2
(0)s
3
+ o(s
3
)

+ b
¸
1
2
k(0)s
2
+
1
6
k
·
(0)s
3
+ o(s
3
)

+ c
¸
1
6
k(0)χ(0)s
3
+ o(s
3
)

= 0
(1.12.2)
or
as +
1
2
bk(0)s
2
+
1
6

−ak
2
(0) + bk
·
(0) + ck(0)χ(0)

s
3
+ o(s
3
) = 0. (1.12.3)
We have, now, several possibilities,
1.12. The contact between a space curve and a plane 65
(a) The osculating plane (b) The normal plane
(c) The rectifying plane
Figure 1.8: The projection of a space curve on the planes of the Frenet frame
66 Chapter 1. Space curves
a) If a 0 (i.e. the plane does not contain the tangent), that the plane has a contact of
order zero with the curve (intersection, they have a single point in common).
b) If a = 0, then the intersection equation has s = 0 as a double solution, which means
that the curve and the plane have a contact of order 1 (tangency contact). It can be
observed immediately that in this case the plane Π passes through the tangent at the
curve at the point M
0
(which is, in fact, the straight line of equations y = 0, z = 0 in
the coordinate frame considered by us).
c) If we want a contact of order at least 2 (osculation contact), then the coefficient of
s
2
in the intersection equation should also vanish. This may happen only if b = 0
(as the point M
0
is biregular, the curvature is not zero at this point). Thus, we have
osculation contact if we impose a = 0 and b = 0. But this choice leads us to the
equation z = 0 for the plane Π, i.e. the plane is already completely determined (the
osculating plane).
d) As the plane Π is completely determined by the condition of having a second order
contact with the curve, we cannot specialize more, in order to have higher order
contact (superosculation). Anyway, looking at the intersection equation, we conclude
that the osculating plane has a contact or superosculation with the curve at the planar
points of the curve, where the torsion vanishes. At all other points, the contact is just
of second order (osculation).
1.13 The contact between a space curve and a sphere. The
osculating sphere
As in the previous paragraph, we consider a naturally parameterized curve (I, r), we
assume that 0 is an interior point of the interval I and that r(0) is a biregular point of the
curve. We choose as coordinate frame the Frenet frame of the curve r at the point M
0
,
F
0
= ¦M
0
; τ
0
, ν
0
, β
0
¦. Then an arbitrary sphere passing through the point M
0
will have,
with respect to this coordinate system, the equation
F(x, y, z) ≡ x
2
+ y
2
+ z
2
− 2ax − 2by − 2cz = 0, (1.13.1)
where Ω(a, b, c) is the center of the sphere. Then the condition of the intersection will be,
again, F(x(s), y(s), z(s)) = 0, where x = x(s), y = y(s), z = z(s) are the local equations of
1.13. The contact between a space curve and a sphere. The osculating sphere 67
the curve with respect to the Frenet frame at M
0
. Thus, in our case, we have
F(x(s), y(s), z(s)) =
¸
s −
1
6
k
2
(0)s
3
+ o(s
3
)

2
+
¸
1
2
k(0)s
2
+
1
6
k
·
(0)s
3
+ o(s
3
)

2
+
+
¸
1
6
k(0)χ(0)s
3
+ o(s
3
)

2
− 2a
¸
s −
1
6
k
2
(0)s
3
+ o(s
3
)


− 2b
¸
1
2
k(0)s
2
+
1
6
k
·
(0)s
3
+ o(s
3
)

− 2c
¸
1
6
k(0)χ(0)s
3
+ o(s
3
)

= 0.
(1.13.2)
or
−2as + (1 − bk(0))s
2
+
1
3
(ak
2
(0) − bk
·
(0) − ck(0)χ(0))s
3
+ o(s
3
). (1.13.3)
Now, the discussion that follows is, also, similar to that from the case of the contact
between a curve and a plane. Still, here we have more cases to take into consideration.
a) If a 0, then s = 0 is a simple solution of the intersection equation. Thus, in this
case the sphere and the curve have a contact of order zero (intersection contact).
b) The sphere and the curve have a first order contact (tangency contact) if and only if
the intersection equation has a double zero at the origin. This happens, obviously, iff
a = 0. This means that the first coordinate of the center of the sphere is zero, i.e. this
center lies in the normal plane of the curve at M
0
. It is easy to see that in this case
the tangent line at M
0
of the curve lies in the tangent plane of the sphere at the same
point, which justifies once more the denomination of tangency contact.
c) For an osculation (second order) contact, the coefficient of s
2
in the intersection equa-
tion also has to vanish and we get
b =
1
k(0)
= R(0), (1.13.4)
where R(0) is the curvature radius of the curve at the point M
0
. Thus, in this case, the
center of the sphere lies on the intersection line of the planes x = 0 (the normal plane)
and y = R(0). This line, which, as one can easily see, is parallel to the binormal of
the curve at M
0
, is called curvature axis or polar axis of the curve. It intersects the
osculating plane at M
0
of the curve at the center of curvature at M
0
of the curve.
Thus, any sphere with the center on the curvature axis of the curve has an osculation
contact with this one.
68 Chapter 1. Space curves
d) We shall say that the sphere has a contact of superosculation with the curve if they
have a contact of order at least three, which means that in the intersection equation
the coefficients of the of the powers up to three of s have to vanish simultaneously. It
is usually claimed that there is a single sphere which has a contact of superosculation
with the curve and this sphere is called the osculating sphere. In reality, the things
are a little bit more subtle, as we shall see in a moment.
1) Let us assume, first, that the torsion of the curve at M
0
does not vanish, i.e. χ(0)
0. In this case, as one sees immediately, between the sphere and the curve there is
a superosculation contact if and only if we have

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
a = 0,
b =
1
k(0)
,
c =
k
·
(0)
k
2
(0)χ(0)
,
(1.13.5)
i.e. in this case the sphere is uniquely determined and we will call it osculating
sphere
7
.
2) If χ(0) = 0, while k
·
(0) 0, as one can see easily, the coefficient of s
3
is never
zero, therefore there is no sphere which has a superosculation contact with the
curve. However, as we have seen in the previous paragraph, in this case the oscu-
lating plane has a superosculation contact with the curve and we can think of this
plane as playing also the role of the osculating sphere, of infinite radius, though.
3) If both χ(0) and k
·
(0) vanish, then the coefficient of s
3
in the intersection equation
vanishes for any c, in other words in this case any sphere which has an osculation
contact with the curve also has a superosculation contact. Thus, now we have an
infinity of osculating spheres.
1.14 Existence and uniqueness theorems for parameterized
curves
1.14.1 The behaviour of the Frenet frame under a rigid motion
Definition 1.14.1. A rigid motion of R
3
is a map D : R
3
→ R
3
, D(x) = 7 x +b, where
7 ∈ M
3×3
(R) is an orthogonal matrix, 7
t
7 = I
3
, with the determinant equal to one:
det 7 = 1, while b ∈ R
3
is a constant vector. The linear map A : R
3
→ R
3
, A(x) = 7 x
is called the homogeneous part of the rigid motion.
7
In some books all the spheres which have an osculation contact with the curve are called osculating
spheres, while the one which has a superosculation contact is called, accordingly, superosculating sphere
1.14. Existence and uniqueness theorems for parameterized curves 69
Remarks. (i) A rigid motion in R
3
is just a rotation around an arbitrary axis followed
by a translation.
(ii) If we don’t ask det 7 = 1, we get, equally, an isometry of R
3
. However, in this case
(if det 7 = −1), a transformation D(x) = 7 x + b does not reduce anymore to a
rotation and a translation, we have to add, also, a reflection with respect to a plane.
In many books the term “motion” implies only that the matrix 7is orthogonal and
for what we call “rigid motion” it is used the term “proper motion”.
Definition 1.14.2. Let D : R
3
→ R
3
be a rigid motion, with the homogeneous part
A : R
3
→ R
3
. The image of a parameterized curve (I, r = r(t)) through D is, by
definition, the parameterized curve (I, r
1
= (D ◦ r)(t)).
Remark. Since A is a nondegenerate linear map, the image of a regular parameterized
curve is, also, a regular parameterized curve.
Theorem 1.14.1. Let D be a rigid motion of R
3
, with the homogeneous part A, (I, r =
r(t)) – a biregular parameterized curve, (I, r
1
= D ◦ r) – its image through D and
¦r(t); τ(t), ν(t), β(t)¦ – the Frenet frame of the parameterized curve r at t. Then the frame
¦r
1
(t); A(τ(t)), A(ν(t)), A(β(t))¦ is the Frenet frame of r
1
at t.
Proof. Let r(t) = (x(t), y(t), z(t)), D(x, y, z) = (x
1
, y
1
, z
1
), where

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x
1
= α
11
x + α
12
y + α
13
z + b
1
y
1
= α
21
x + α
22
y + α
23
z + b
2
z
1
= α
31
x + α
32
y + α
33
z + b
3
(1.14.1)
Then
r
1
(t) = (x
1
(t), y
1
(t), z
1
(t)),
hence
r
·
1
(t) = A(r
·
(t)), r
··
1
(t) = A(r
··
(t)). (1.14.2)
Since A is a linear isometry, preserving the orientation of R
3
, we have
|r
·
1
| = |A(r
·
)| = |r
·
|, r
·
1
r
··
1
= r
·
r
··
,
r
·
1
× r
··
1
= A(r
·
× r
··
), |r
·
1
× r
··
1
| = |r
·
× r
··
|.
70 Chapter 1. Space curves
Then:
τ
1
=
r
·
1
|r
·
|
=
A(r
·
)
|A(r
·
)|
=
A(r
·
)
|r
·
|
= A
¸
r
·
|r
·
|

= A(τ)
ν
1
=
|r
·
1
|
|r
·
1
× r
··
1
|
r
··
1

r
·
1
r
··
1
|r
·
1
| |r
·
1
× r
··
1
|
r
·
1
=
|r
·
|
r
·
× r
··
A(r
··
)−

r
·
r
··
|r
·
| |r
·
× r
··
|
A(r
·
) = A(ν)
β
1
=
r
·
1
× r
··
1
|r
·
1
× r
··
1
|
= A(β).

Consequence. The parameterized curves (I, r) and (I, r
1
= D◦ r) have the same curva-
ture and torsion.
Proof. We have
k
1
=
|r
·
1
× r
··
1
|
|r
·
1
|
=
|r
·
× r
··
|
|r
·
|
= k.
For the torsion, the situation is slightly more complicated. We have, from the theorem,
β
1
(t) = A(β(t)) and ν
1
(t) = A(ν(t)).
A being a linear operator, a similar result holds for the derivatives of the two Frenet
vectors, i.e. we have
β
·
1
(t) = A(β

(t)) and ν
·
1
(t) = A(ν

(t)).
Using the last Frenet equations for the two curves, we have the equalities
−χ
1
(t)ν
1
(t) = A(−χ
(
t)ν(t)) = −χ(t)A(ν(t)) = −χ(t)ν
1
(t),
and, thus, the two torsions are equal, as claimed.
1.14.2 The uniqueness theorem
Theorem 1.14.2. Let (I, r = r(t)) and (I, r
1
= r
1
(t)) be two biregular parameterized
curves. If k(t) = k
1
(t), χ(t) = χ
1
(t) and |r
·
(t)| = |r
·
1
(t)| ∀t ∈ I, then there is a single
rigid motion D of R
3
such that r
1
= D ◦ r.
1.14. Existence and uniqueness theorems for parameterized curves 71
Proof. Let t
0
∈ I be an arbitrary point and D – the rigid motion of R
3
sending the Frenet
frame ¦r(t
0
); τ
0
, ν
0
, β
0
¦ of the curve r at t
0
into the Frenet frame ¦r
1
(t
0
); τ
10
, ν
10
, β
10
¦
of the curve r
1
at the same point. Obviously, there is a single rigid motion with this
property. Let (I, r
2
(t
=
D ◦ r(t)) – the image of the curve r through D, and k
2
, χ
2
– the
curvature and the torsion, respectively, of the parameterized curve r
2
. Then
k
2
(t) ≡ k(t) ≡ k
1
(t)
χ
2
(t) ≡ χ(t) ≡ χ
1
(t)
and, moreover,
|r
·
2
(t)| ≡ |r
·
1
(t)|.
Therefore, the vector functions τ
1
(t), ν
1
(t), β
1
(t) and τ
2
(t), ν
2
(t), β
2
(t) giving the Frenet
frame are solutions of the same system of Frenet equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ

= |r
·
1
|k
1
ν
ν

= −|r
·
1
|k
1
τ + |r
·
1

1
β
β

= −|r
·
1

1
ν.
Since for t = t
0
the solutions coincide, due to the uniqueness theorem for the solution of
a Cauchy problem they have to coincide globally. In particular, we have
τ
1
(t) ≡ τ
2
(t) or
r
·
1
(t)
|r
·
1
(t)|
=
r
·
2
(t)
|r
·
2
(t)|
,
hence
r
·
1
(t) − r
·
2
(t) = 0 ⇒ r
1
(t) − r
2
(t) = const.
But for t = t
0
, r
1
(t
0
) − r
2
(t
0
) = 0, hence the two functions coincide for all t, thus
r
1
(t) ≡ r
2
(t) = D ◦ r(t).
As for the uniqueness of D, we notice that, for any other point t
1
∈ I, since r
1
≡ r
2
,
D sends the Frenet frame of the curve r at t
1
into the Frenet frame of the curve r
1
at
t
1
.
Remark. For naturally parameterized curves the condition |r
·
(t)| = |r
·
1
(t)| is always
fulfilled.
72 Chapter 1. Space curves
1.14.3 The existence theorem
Theorem 1.14.3. Let f (s) and g(s) be two smooth functions, defined on an interval
I, such that f (s) > 0, ∀t ∈ I. Then there is a single naturally parameterized curve
(I, r = r(s)) for which f (s) = k(s) ∀s ∈ I and g(s) = χ(s) ∀s ∈ I. This curve is uniquely
defined, up to a rigid motion of R
3
.
Proof. Let ¦r
0
; T
0
, N
0
, B
0
¦ be a direct orthonormal frame in
mathbbR
3
. We consider the system of linear ordinary differential equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
T
·
(s) = f (s)N(s)
N
·
(s) = −f (s)T(s) + g(s)B(s)
B
·
(s) = −g(s)N(s)
(1.14.3)
with respect to the vector functions T(s), N(s), B(s).
If we denote
X(s) = (T(s), N(s), B(s)), (1.14.4)
then the system (1.14.3) can be written as
X
·
(s) = A(s) X(s), (1.14.5)
with
A(s) =

¸
¸
¸
¸
¸
¸
¸
¸
¸
0 f (s) 0
−f (s) 0 g(s)
0 −g(s) 0

¸
¸
¸
¸
¸
¸
¸
¸
¸
.
In the theory of ordinary differential equations one proves that the system (1.14.5) has a
single solution subject to
X(s
0
) = (T
0
, N
0
, B
0
),
where s
0
∈ I, while the columns of the matrix X(s
0
) are the vectors T
0
, N
0
, B
0
of the
initial orthonormal basis.
We are going to show first that, for any s ∈ I the vectors din (T(s), N(s), B(s))
form an orthonormal basis. It is enough to show for any s ∈ I, X(s) is orthogonal, i.e.
X
t
(s) X(s) = I
3
, for any s ∈ I. We have
d
dt

X
t
X

=
d
dt

X
t
(s)

X + X
t

d
dt
(X(s)) = X
t
(A
t
X + AX) = X
t
(A
t
+ A)X.
but, as A is skew symmetrical, A
t
+ A = 0, therefore
d
dt

X
t
X

= 0 ⇒ X
t
X = const.
1.14. Existence and uniqueness theorems for parameterized curves 73
On the other hand, from the initial condition,

X
t
X

(s
0
) = I
3
, hence X
t
(s) X(s) = I
3
for any s ∈ I.
Let us define now
r = r
0
+
s

s
0
T(s)ds, (sol)
where r
0
is the origin of the original frame, while T(s) is the first column of X(s). We
are going to show that (I, r(s)) is the curved searched for. We have, clearly:
r
·
(s) = T(s),
|r
·
(s)| = |T(s)| = 1,
r
··
(s) = T
·
(s) = f (s)N(s).
We note immediately that r
·
(s) × r
··
(s) 0, therefore r(s) is a biregular naturally para-
meterized curve. On the other hand,
r
···
(s) = ( f (s)N)
·
= f
·
N + f N
·
= f
·
N + f (−f T + gB) = −f
2
T + f
·
N + f gB,
hence
(r
·
, r
··
, r
···
) = (T, f N, −f
2
T + f
·
N + f gB) = (T, f N, f gB) = f
2
g (T, N, B)
....
=1
= f
2
g.
Now we have all we need to compute the curvature and the torsion:
k(s) =
|r
·
× r
··
|
|r
·
|
3
= | f (s)| = f (s),
χ(s) =
f
2
(s)g(s)
f
2
(s)
= g(s),
hence the curve r fulfils the conditions of the theorem.
The uniqueness of r, up to rigid motions, follows from the previous theorem.
74 Chapter 1. Space curves
CHAPTER 2
Plane curves
2.1 Introduction
After the exposition of the theory of space curves in all the generality, we shall focus in
this chapter on topics which are specific to the theory of plane curves. In particular, we
shall discuss here notions which cannot be defined for space curves (such as the signed
curvature), or which are easier to investigate and richer in contents for plane curves.
2.2 Envelopes of plane curves
In this section, if not mentioned otherwise, all the curves are parameterized curves, if
not specified otherwise. Let
r = r(t, λ) (2.2.1)
be a family of plane parameterized curves, depending smoothly on a real parameter λ.
Definition. The envelope of the family (2.2.1) is a parameterized curve (J, Γ) which, at
each point, is tangent to a curve from the family.
Theorem. The points of the envelope of the family r(t, λ) are subject to
r = r(t, λ) (2.2.2)
r
·
λ
× r
·
t
= 0. (2.2.3)
76 Chapter 2. Plane curves
Proof. If Γ is the envelope of the family (γ
λ
) and P is a point of Γ, then P is a tan-
gency point between Γ and a curve from the family, corresponding to some value of the
parameter λ. Thus, the equation of Γ will be of the form
r
1
= r
1
(λ).
On the other hand, P is on a curve γ
λ
and, therefore, it verifies
r
1
= r(t(λ), λ).
The tangency condition between Γ and γ
λ
reads
r
·

| r
·
t
or
r
·

× r
·
t
= 0
or, also, since r
·

= r
·
t
t
·
λ
+ r
·
λ
,
(r
·
t
t
·
λ
+ r
·
λ
) × r
·
t
= 0,
and the theorem is proven, because r
·
t
× r
·
t
= 0.
Remarks. 1. The set of points described by the equations (2.2.2) and (2.2.3) is
called the discriminant set of the family γ
λ
. It includes not only the support of the
envelope, but, also, the singular points of the curves from the family, for which
r
·
t
= 0.
2. The equation r
·
λ
× r
·
t
= 0 can be written also as

i j k
x
·
λ
y
·
λ
0
x
·
t
y
·
t
0

= 0 ⇔ x
·
λ
y
·
t
− x
·
t
y
·
λ
= 0 ⇔

x
·
λ
x
·
t
=
y
·
λ
y
·
t
. (2.2.4)
Example. Let us consider the family of curves
r(t, λ) = (λ + a cos t, λ + a sin t), λ, t ∈ R, a > 0.
2.2. Envelopes of plane curves 77
Clearly, we are dealing with a family of circles of radius a, with the centres on the first
bisector of the coordinate axes. Then we have
r
·
λ
= ¦1, 1¦,
r
·
t
= ¦−a sin t, a cos t¦,
therefore, the points of the envelope (and only them, since the circles have no singular
points) verify

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = λ + a cos t
y = λ + a sin t
x
·
λ
y
·
t
= x
·
t
y
·
λ
or

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = λ + a cos t
y = λ + a sin t
a cos t = −a sin t.
Eliminating t, we get the parametric equations of the envelope:

¸
¸
¸
¸
¸
¸
¸
¸
x(λ) = λ ±
a

2
y(λ) = λ ∓
a

2
,
i.e. the envelope is, in fact, a pair of straight lines, parallel to the first bissector of the
coordinate axes.
2.2.1 Curves given through an implicit equation
Proposition 2.2.1. The points of the envelope of a family of plane curves given through
the implicit equation
F(x, y, λ) = 0 (2.2.5)
verify the system of equations

¸
¸
¸
¸
¸
¸
F(x, y, λ) = 0
F
·
λ
(x, y, λ) = 0
(2.2.6)
Proof. Locally, around each point of a the curve of the family, we can parameterize the
curve, i.e. we can represent it as

¸
¸
¸
¸
¸
¸
x = x(t, λ)
y = y(t, λ)
.
78 Chapter 2. Plane curves
By substituting into the equation of the family, we get
F(x(t, λ), y(t, λ), λ) = 0,
whence, differentiating with respect to t and λ, respectively, we obtain the system:

¸
¸
¸
¸
¸
¸
F
·
x
x
·
t
+ F
·
y
y
·
t
= 0
F
·
x
x
·
λ
+ F
·
y
y
·
λ
+ F
·
λ
= 0
.
But, from (2.2.4),
x
·
λ
= Kx
·
t
, y
·
λ
= Ky
·
t
,
with K = const., therefore, the second equation from above becomes
K(F
·
x
x
·
t
+ F
·
y
y
·
t
....
=0
) + F
·
λ
= 0
or
F
·
λ
= 0.

Example. We consider again the family of circles from the previous paragraph, this time
given through their implicit equation
F(x, y, λ) ≡ (x − λ)
2
+ (y − λ)
2
− a
2
= 0.
Then the second equation of the discriminant set will be
F
·
λ
(x, y, λ) = −2(x + y − 2λ) = 0,
whence we get
λ =
x + y
2
,
which, when substituted into the equation of the family, gives
(x − y)
2
= 2a
2
,
i.e. we get, again, the same equations of the envelope as before, namely
y = x ± a

2.
2.2. Envelopes of plane curves 79
2.2.2 Families of curves depending on two parameters
Proposition 2.2.2. Suppose we are given a family of curves depending smoothly on two
parameters, λ and µ
F(x, y, λ, µ) = 0, (2.2.7)
where the parameters λ and µ are connected through a relation
ϕ(λ, µ) = 0, (2.2.8)
then the points of the envelope verify the system

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
F(x, y, λ, µ) = 0
ϕ(λ, µ) = 0
D(F, ϕ)
D(λ, µ)
= 0
. (2.2.9)
Proof. From the equation
ϕ(λ, µ) = 0,
we can assume, if, for instance, that
µ = µ(λ),
therefore, substituting in F, and ϕ,

¸
¸
¸
¸
¸
¸
F(x, y, λ, µ(λ)) = 0,
ϕ(λ, µ(λ)) = 0.
Differentiating with respect to λ these two equations, we get:

¸
¸
¸
¸
¸
¸
F
·
λ
+ F
·
µ
µ
·
λ
= 0
ϕ
·
λ
+ ϕ
·
µ
µ
·
λ
= 0.
Eliminating the derivative µ
·
λ
between the two equations, we get the third equation
from (2.2.9), as required.
2.2.3 Applications: the evolute of a plane curve
Definition. Let (I, r = r(t)) be a plane parameterized curve. The evolute of r is, by
definition, the envelope of the family of normals of the curve.
80 Chapter 2. Plane curves
The following result holds:
Proposition 2.2.3. The parametric equations of the evolute of the curve r = r(t) =
(x(t), y(t)) are

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
X = x −
y
·
(x
·2
+ y
·2
)
x
·
y
··
− x
··
y
·
Y = y +
x
·
(x
·2
+ y
·2
)
x
·
y
··
− x
··
y
·
(2.2.10)
Proof. As we know, the equation of the normal of a plane curve is
F(X, Y, t) = (X − x(t)) x
·
(t) + (Y − y(t)) y
·
(t) = 0.
The relations verified by the points of the envelope of the family of normals (and only
by them, since in this case the curves of the family are straight line and they have no
singular points) are (see (2.2.6)):

¸
¸
¸
¸
¸
¸
F(X, Y, t) = 0
F
·
t
(X, Y, t) = 0
,
i.e.

¸
¸
¸
¸
¸
¸
x
·
(t)X + y
·
(t)Y = x(t) x
·
(t) + y(t) y
·
(t)
x
··
(t)X + y
··
(t)Y = x
·2
(t) + x
··
(t) x(t) + y
·2
(t) + y(t) y
··
(t)
.
The equations (2.2.10) follow now instantly, after solving this system of linear equations
with respect to X and Y.
Example. For the ellipse

¸
¸
¸
¸
¸
¸
x = a cos t
y = b sin t
one gets, after the computations,

¸
¸
¸
¸
¸
¸
X =
a
2
−b
2
a
cos
3
t
Y =
b
2
−a
2
b
sin
3
t
or, eliminating the parameter t,
a
2
3
X
2
3
+ b
2
3
Y
2
3
= (a
2
− b
2
)
2
3
.
The curved described by this equation is called a lengthened astroid (see figure 2.2.3).
2.3. The curvature of a plane curve 81
Figure 2.1: The evolute of an ellipse
2.3 The curvature of a plane curve
As we saw, in the case of an arbitrary space curve, the curvature is always a positive
scalar. Of course, this concept of curvature can be equally applied for plane curves. It
turns out, however that, in this particular case, we can obtain more information about
the curve if we use a slightly different concept, allowing the curvature to have a sign.
To define the curvature of a plane curve we shall use a little technical trick, which will
allow us to give the definition in a coordinate independent way.
Definition. The complex structure on R
2
is the map J : R
2
→ R
2
, defined by
J(x, y) = (−y, x).
Remark. Applying J simply means to rotate the vector ¦x, y¦ by
π
2
or multiplying the
complex number x + iy by the complex unit i (this is, actually, the origin of the name).
Some obvious properties of the complex structure are collected into the following
proposition:
Proposition 2.3.1. a) Jv Jw = v w.
b) (Jv) v = 0.
82 Chapter 2. Plane curves
c) J(Jv) = −v (i.e. J
2
= −id).
All these properties follow immediately from the geometrical interpretation of the
complex structure.
Anticipating a little bit, we would like to say a few words about the kind of curvature
we are going to define. We remember that the curvature of an arbitrary parameterized
space curve r = r(t) can be computed by
k(t) =
|r
·
(t) × r
··
(t)|
|r
·
(t)|
3
.
Now, if r is a plane curve, with the support lying into the coordinate plane xOy, then the
vectors r
·
(t) and r
··
(t) are situated, also, into this plane. Therefore, their vector product
is a vector which is directed along the z axis, and, therefore, the norm of this vector is
just the absolute value of the z-component. Now, the idea of the definition of the signed
curvature is just to replace this absolute value with the component itself. To do that, the
following characterization of the vector product of two vectors in the plane will prove
very useful.
Proposition 2.3.2. Let u(x
1
, y
1
), v(x
2
, y
2
) ∈ R
2
. Then
u × v = [v Ju] k.
Proof. As it is known, the cross product of the vectors u and v can be computed by
u × v =

i j k
x
1
y
1
0
x
2
y
2
0

= (x
1
y
2
− x
2
y
1
) k.
On the other hand,
[v Ju] = ¦x
2
, y
2
¦ ¦−y
1
, x
1
¦ = −x
2
y
1
+ x
1
y
2
,
whence the equality announced.
We are now ready to define the curvature of a plane curve.
Definition. Let r = r(t) be a plane parameterized curve. The signed curvature of r is,
by definition, the quantity
k
±
=
r
··
Jr
·
|r
·
|
3
. (2.3.1)
2.3. The curvature of a plane curve 83
Remark. According to the proposition 2.3.2, the signed curvature is just the projection of
the curvature vector on the z-axis. Now, as the curvature vector is parallel to the z-axis,
we have, thus
|k
±
| =
|r
·
× r
··
|
|r
·
|
3
= k.
Another immediate, but important, for computational reasons, result, is the follow-
ing:
Proposition 2.3.3. Let (I, r = r(t)) be a plane parameterized curve. If r(t) = (x(t), y(t)),
then the signed curvature of r can be expressed as
k
±
(t) =
x
·
(t)y
··
(t) − x
··
(t)y
·
(t)

x
·2
(t) + y
·2
(t)

3/2
.
Consequence. If y = f (x) is the explicit equation of a plane curve, then its signed
curvature is given by
k
±
(x) =
f
··
(x)
(1 + f
·
2
)
3/2
.
Remark. The previous consequence show that, for an explicitly given curve (i.e. the
graph of a real function of a single real variable) the sign of the signed curvature is, in
fact, the sign of the second derivative of the function f , i.e., as it known from calculus,
the signed of the signed curvature is a n indication of the convexity or concavity of the
function.
Exactly as happens for the torsion of the space curves, the signed curvature of a plane
curve is “almost” invariant at a parameter change, i.e. we have
Theorem. Let (I, r = r(t)) be a plane parameterized curve (J, ρ = ρ(u)) an equivalent
parameterized curve, with the parameter change λ : I → J, u = λ(t). Then
k
±
[ρ](u) = sgn(λ
·
) k
±
[r](t).
Proof. We have
r(t) = ρ(λ(t))
r
·
(t) = ρ

(λ(t)) λ
·
(t)
r
··
(t) = ρ

(λ(t)) λ
·2
(t) + ρ

(λ(t)) λ
··
(t)
r
··

= (ρ

λ
·2
+ ρ

λ
··

J(λ
·
ρ

) =
= λ
·3
ρ

+ λ
·
λ
··
ρ

....
= 0 = λ
·3
ρ

|r
·
|
3
= |λ
·3
||ρ

|
3
84 Chapter 2. Plane curves
therefore
k
±
[r](t) =
r
··
Jr
·
|r
·
|
3
=
λ
·3

·3
|

ρ

(u) Jρ

(u)

(u)|
3
= sgn(λ
·
) k
±
[ρ](u),
whence
k
±
[ρ](u) = sgn(λ
·
) k
±
[r](t).

Remark. The previous theorem shows that the signed curvature is invariant under any
positive parameter change, therefore it makes sense to define it also for regular oriented
plane curves.
The curvature vector of a plane, naturally parameterized curve can be expressed
easily as a function of the signed curvature:
Lemma. Let (I, r = r(s)) be a plane naturally parameterized curve. Then
r
··
(s) = k
±
(s) Jr
·
(s).
Proof. We have r
·2
(s) = 1 (the curve is naturally parameterized), therefore r
·
r
··
= 0,
whence it follows that r
··
⊥ r
·
or, which is the same, r
··
| Jr
·
.
On the other hand, from the definition of the signed curvature, k
±
(s) = r
··
(s) Jr
·
(s).
If we put r
··
(s) = α(s) Jr
·
(s), then we should have r
··
(s) Jr
·
(s) = α(s) [Jr
·
(s)]
2
= α(s),
whence α(s) = k
±
(s).
2.3.1 The geometrical interpretation of the signed curvature
For the signed curvature of plane curve we have a geometrical interpretation which is
similar to the geometrical interpretation of the curvature of a space curve, except that,
this time the sign is also taken into account. We will need first the following definition.
Definition. Let (I, r = r(t)) be a plane parameterized curve. The rotation angle of r is
the function θ[r] : I → R, defined through:
τ(t) = ¦cos θ[r](t), sin θ[r](t)¦ = exp(iθ[r](t)), (2.3.2)
where τ(t) is the unit tangent, i.e. θ[r] is the angle made by the unit tangent with the
positive direction of the x axis.
2.3. The curvature of a plane curve 85
Remark. This definition is very innocent looking and natural. After all, is should be
clear for anyone that θ is just the angle made by the unit tangent vector with the positive
direction of the x-axis. In reality, however, for an arbitrary plane curve, it is by no mean
obvious that we can find a continuous angle function, let alone a smooth one. Such
functions do exist (see, for instance, for a modern proof, the book of Bär, [1]) and any
two such functions differ by an integer multiple of 2π.
The following lemma provides the connection between the signed curvature and
the variation of the rotation angle. The contents of this lemma is quite similar to the
geometrical interpretation of the curvature of a space curve. In fact, when the plane
curve is regarded as a particular case of space curve, then the variation of the rotation
angle is equal (as absolute value) with the variation of the contingency angle, therefore,
in fact, the geometrical interpretation of the absolute curvature of a plane curve (regarded
as a curve in space) is an immediate consequence of this lemma.
Lemma. If (I, r = r(t)) is a plane regular parameterized curve, θ is its rotation angle
and k
±
– its signed curvature, then:

dt
= |r
·
(t)|k
±
(t).
Proof. From the definition of the unit tangent vector we have τ(t) =
r
·
(t)
|r
·
(t)|
, whence

dt
=
r
··
(t)
|r
·
(t)|
+ r
·
(t)
d
dt
¸
1
|r
·
(t)|

.
On the other hand, if we use the expression of τ(t) as a function of the angle θ, given by
(2.3.2), we get for

dt
the formula

dt
=

dt
¦−sin θ(t), cos θ(t)¦ =

dt
Jτ(t).
Combining the two relations, we get the equality
r
··
(t)
|r
·
(t)|
+ r
·
(t)
d
dt
¸
1
|r
·
(t)|

=

dt
Jτ(t) ≡

dt

Jr
··
(t)
|Jr
·
(t)|
.
Multiplying both sides by Jr
·
(t) and having in mind that Jr
·
(t) r
·
(t) = 0 and Jr
·
(t)
Jr
·
(t) = |r
·
(t)|
2
, we obtain:
r
··
(t) Jr
·
(t)
|r
·
(t)|
=

dt
|r
·
(t)|,
86 Chapter 2. Plane curves
whence, using the definition of the signed curvature, it follows that

dt
|r
·
(t)| = k
±
(t) |r
·
(t)|
2
or, after simplification,

dt
= k
±
(t) |r
·
(t)|,
which is exactly what we had to prove.
Corollary. For a naturally parameterized curve, (I, r = r(s)), we have
k
±
(s) =

ds
.
Remark. From the previous formula we get
k ≡ |k
±
| =


ds

,
which is exactly the formula for the curvature of arbitrary space curves, which, thus,
remains valid, as expected, for the particular case of plane curves.
2.4 The curvature center. The evolute and the involute of a
plane curve
Definition. A point Ω ∈ R
2
is called the curvature center at r
0
= r(t
0
) of a plane
parameterized curve r : I → R
2
if there is a circle (γ), centred at Ω, which is tangent to
the curve at r
0
= r(t
0
), with t
0
∈ I, such that the signed curvatures r and γ at r
0
coincide,
whence the position of the point Ω for an arbitrary t ∈ I:
Ω(t) = r(t) +
1
k
±
(t)
Jr
·
(t)
|r
·
(t)|
.
Remark. The notion of curvature center is invariant at a parameter change: if (J, ρ =
ρ(u)) is equivalent to r, with the parameter change λ : I → J, then r
·
(t) = ρ

(λ(t))λ
·
(t)
and k
±
[r](t) = sgn(λ
·
)k
±
[ρ](λ(t)). Obviously, we could have problems only when λ
·
< 0,
but in this case Jr
·
changes the sense and k
±
changes the sign and, overall, the situation
remains unchanged.
We defined the evolute of a plane curve as being the envelope of the normals of the
curve. The following proposition provides a different approach.
2.4. The curvature center. The evolute and the involute of a plane curve 87
Proposition 2.4.1. The evolute of a plane curve is the geometrical locus of the curvature
centers of the curve.
Proof. The curvature center of the curve at an arbitrary value of the parameter is
Ω(t) = r(t) +
|r
·
(t)|
2
r
··
(t) Jr
·
(t)
Jr
·
(t) = (x(t), y(t)) +
x
·2
+ y
·2
x
·
y
··
− x
··
y
·
¦−y
·
, x
·
¦.
Thus, if Ω(t) = (X(t), Y(t)), projecting the previous equation on the coordinate axes, we
get the parametric equations of the locus of curvature centers:

¸
¸
¸
¸
¸
¸
¸
¸
X(t) = x(t) −
y
·
(x
·2
+y
·2
)
x
·
y
··
−x
··
y
·
Y(t) = y(t) +
x
·
(x
·2
+y
·2
)
x
·
y
··
−x
··
y
·
,
which are exactly the equations of the evolute.
From the previous remark, we get immediately:
Corollary. The definition of the evolute makes sense also for regular curves (i.e. two
equivalent parameterized curves have the same evolute).
Exercise 2.4.1. Find the evolute of the astroid

¸
¸
¸
¸
¸
¸
x(t) = a cos
3
t,
y = a sin
3
t
.
Show that the evolute is also an astroid (see the figure ??.
Exercise 2.4.2. Find the evolute of the cycloid

¸
¸
¸
¸
¸
¸
x(t) = a(t − sin t),
y = a(1 − cos t)
.
Show that the evolute is also an cycloid (see the figure ??.
Another interesting plane curve associated to a given one is the so-called involute,
which,a s we shall see immediately, is, in a sense, the inverse of the evolute.
Definition. Let (I, r = r(s)) be a naturally parameterized curve c ∈ I. The involute of r
with the origin at r(c) (or at c, for short) is the parameterized curve (I, ρ[r, c] = ρ[r, c](s))
where
ρ[r, c](s) = r(s) + (c − s)r
·
(s).
88 Chapter 2. Plane curves
Figure 2.2: The evolute of an astroid
Figure 2.3: The evolute of an cycloid
Remark. Generally speaking, s is not a natural parameter along ρ.
If (I, r = r(t)) is an arbitrary parameterized curve, the we can replace the parameter
t by the arc length s =
t

0
|r
·
(τ)|dτ and define the involute of r as being the involute of
the naturally parameterized curve equivalent to it, the natural parameter being the arc
length. Is is easy to see that the following proposition holds true:
Proposition 2.4.2. Let (I, r = r(t)) be a parameterized curve. The the involute of r with
2.4. The curvature center. The evolute and the involute of a plane curve 89
the origin at c ∈ I is given by
ρ(t) = r(t) + (c − s(t))
r
·
(t)
|r
·
(t)|
,
where s = s(t) is the arc length of r.
Example. Let r(t) = (a cos t, a sin t) be a circle. Then

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
·
(t) = ¦−a sin t, a cos t¦
x
·2
+ y
·2
= a
2
s(t) =
t

0
adt = at,
hence the equation of the involute is
ρ(t) = (a cos t, a sin t) +
(c − at)
a
¦−a sin t, a cos t¦
or, in projection,

¸
¸
¸
¸
¸
¸
X(t) = a cos t − (c − at) sin t
Y(t) = a sin t + (c − at) cos t
.
We represented in the figure 2.4 an involute of a circle of radius 1.5, with the origin at
the point of parameter 0.
The following lemma establishes the connection between the signed curvature of a
parameterized curve and that of one of its involutes and will serve as a tool to establish
the relation between evolute and involute.
Lemma. Let (I, r = r(s)) be a naturally parameterized curve and ρ the involute of r
with the origin at c ∈ I. Then the signed curvature of ρ is given by
k
±
[ρ](s) =
sgn(k
±
[r](s))
|c − s|
.
Proof. We have
ρ

(s) = r
·
(s) + (c − s)r
··
(s) − r
·
(s) = (c − s)r
··
(s) = (c − s)k
±
[r](s) Jr
·
(s)
ρ

(s) = −k
±
[r](s

Jr
·
(s) + (c − s)(k
±
[r](s))
·
Jr
·
(s) + (c − s)k
±
[r](s) Jr
··
(s)
= [−k
±
[r](s) + (c − s)(k
±
[r](s))
·
] Jr
·
(s) − (c − s)(k
±
[r](s))
2
r
·
(s),
90 Chapter 2. Plane curves
Figure 2.4: An involute of a circle
whence

= −(c − s)k
±
[r](s) r
·
(s),
while
ρ

(s) Jρ

(s) = (c − s)
2
(k
±
[r](s))
3
.
The conclusion follows now from the definition of the signed curvature.
The following theorem provides a connection between the involute and the evolute.
In many textbooks this connection is taken, in fact, as the definition of the involute.
Theorem. Let (I, r = r(s)) be a naturally parameterized curve and ρ – its involute with
the origin at c ∈ I. The the evolute of ρ is r.
Proof. The evolute of ρ is given, as known, by the equation
ρ
1
(s) = ρ(s) +
1
k
±
[ρ](s)

(s)

(s)|
.
Using the previous lemma to express the signed curvature of ρ as a function of the signed
curvature of r, we get
ρ
1
(s) = r(s) + (c − s)r
·
(s) +
|c − s|
sgn(k
±
[r](s))

(c − s)k
±
[r](s) J
2
r
·
(s)
|(c − s)k
±
[r](s) Jr
·
(s)|
= r(s).

2.5. The osculating circle of a curve 91
2.5 The osculating circle of a curve
Definition. Let (I, r = r(t)) be a parameterized curve. The osculating circle of r at a
point t ∈ I is the circle centered at the curvature center Ω(t), with the radius equal to the
curvature radius
1
k(t)
of the curve at that point.
Exactly as the osculating plane at a point of a space curve can be regarded as a limit
position of a plane determined by three neighboring points, when these points approach
indefinitely the given one, the osculating circle is the limit position of a circle determined
by three neighboring points, when these points approach indefinitely the given one. More
precisely, we have:
Theorem. Let (I, r = r(t)) be a plane parameterized curve and t
1
< t
2
< t
3
∈ I Let
C(t
1
, t
2
, t
3
) be the circle passing through r(t
1
), r(t)2), r(t
3
). We assume that, for a value
t ∈ I of the parameter we have k
±
(t) 0. Then the osculating circle of r at the point r(t)
is the circle
C = lim
t
1
→t
t
2
→t
t
3
→t
C(t
1
, t
2
, t
3
).
Proof. Let A(t
1
, t
2
, t
3
) be the centre of the circle C(t
1
, t
2
, t
3
) and f : I → R – the function
defined through f (t) = |r(t) − A|
2
. Then, clearly, f is smooth and we have:

¸
¸
¸
¸
¸
¸
f
·
(t) = 2r
·
(r(t) − A)
f
··
(t) = 2r
··
(t) (r(t) − A) + 2|r
·
(t)|
2
.
Since f is differentiable and f (t
1
) = f (t
2
) = f (t
3
), from the mean value theorem it
follows that there exist two points u
1
, u
2
∈ I, with t
1
< u
1
< t
2
< u
2
< t
3
such that
f
·
(u
1
) = f (u
2
) = 0.
On the other hand, applying once more the mean value theorem, this time to the deriva-
tive f
·
, which is also differentiable , it follows that there is a v ∈ (u
1
, u
2
) such that
f
··
(v) = 0.
Now, if we let t
1
, t
2
, t
3
→ t, then we will have, equally, u
1
, u
2
, v → t, therefore, at the
limit, we should get:

¸
¸
¸
¸
¸
¸
r
·
(t) (r(t) − A(t)) = 0
r
··
(t) (r(t) − A(t)) = −|r
·
(t)|
2
, (*)
92 Chapter 2. Plane curves
where
A(t) = lim
t
1
→t
t
2
→t
t
3
→t
A(t
1
, t
2
, t
3
).
From (*) and the definition of the signed curvature, it follows that
r(t) − A(t) =
−1
k
±
(t)

Jr
·
(t)
|r
·
(t)|
,
hence C is the osculating circle of the curve r at the point r(t).
2.6 The existence and uniqueness theorem for plane curves
The existence and uniqueness theorem for plane parameterized curves is similar to the
analogue theorem for space curves and can be proved in the same manner, therefore we
shall skip the proof here.
Theorem 2.6.1. Let f : I → R be a continuous function. Then there is a regular,
naturally parameterized curve (I, r = r(s)) such that ∀s ∈ I, k
±
(s) = f (s). r is unique,
up to a proper motion of R
2
.
To give an example, we will find the curve r for the particular case when the function
f is a constant α, for any real value of the parameter s.
Starting from the geometrical interpretation of the signed curvature, we get
α = k
±
(s) =

ds
,
therefore θ (the rotation angle) will be a linear function of s:
θ = αs + θ
0
,
where θ
0
is a constant. On the other hand, from the definition of the rotation angle, we
obtain:
τ(s) ≡

dx
ds
,
dy
ds

= ¦cos θ(s), sin θ(s)¦ = ¦cos(αs + θ
0
), sin(αs + θ
0

whence the system of differential equations:

¸
¸
¸
¸
¸
¸
dx
ds
= cos(αs + θ
0
)
dy
ds
= sin(αs + θ
0
)
.
2.6. The existence and uniqueness theorem for plane curves 93
Since the equations are separated, the system can be integrated very easy and we get the
solution:

¸
¸
¸
¸
¸
¸
x =
1
α
[sin αs sin θ
0
+ cos αs sin θ
0
] + x
0
,
y =
1
α
[−cos αs cos θ
0
+ sin αs sin θ
0
] + y
0
, (*)
where x
0
and y
0
are two integration constant. The solution (*) can be written in the
matrix form
,
x
y
¸
=
,
¸
¸
¸
¸
¸
1
α
sin αs sin θ
0
+
1
α
cos αs sin θ
0

1
α
cos αs cos θ
0
+
1
α
sin αs sin θ
0
¸
¸
¸
¸
¸
¸
+
,
x
0
y
0
¸
or, also,
,
x
y
¸
=
,
¸
¸
¸
¸
¸
¸
cos

π
2
− θ
0

sin

π
2
− θ
0

−sin

π
2
− θ
0

cos

π
2
− θ
0

¸
¸
¸
¸
¸
¸
¸
,
¸
¸
¸
¸
¸
1
α
cos αs
1
α
sin αs
¸
¸
¸
¸
¸
¸
+
,
x
0
y
0
¸
,
which shows that any plane curve of constant signed curvature, equal to α, can be ob-
tained from the curve

¸
¸
¸
¸
¸
¸
x =
1
α
cos αs
y =
1
α
sin αs
by applying a rotation followed by a translation, i.e. a rigid motion. But this is the circle
of radius 1/α, centered at the origin. The conclusion is that the only plane curve of
constant positive curvature α is the circle of radius 1/α.
94 Chapter 2. Plane curves
CHAPTER 3
The integration of the natural equations of a curve
3.1 The Riccati equation associated to the natural equations
of a curve
The general theory of differential equations guarantees the existence and uniqueness (up
to a rigid motion) for the Frenet equations. However, finding analytically a curve when
the curvature and torsion are given is quite another matter. This, of course, amounts
to finding an analytical solution for the Frenet equations. Although they look quite
innocently, it turns out that, in the general situation, it is not possible to integrate the
system.
Clearly, if we manage to solve the Frenet system, the, in particular, we can find the
tangent versor and, by another quadrature, we can find the curve. Apparently, the Frenet
system should be equivalent to a third order vectorial differential equation or to a system
of three third order scalar equations. Nevertheless, it turns out that the system (made of
none scalar equations) actually contains three identical sets of three equations, therefore
it should be equivalent to a single third order scalar equations. Moreover, the three
vectors of the solution are subject to the orthonormality conditions, which should allow
further reductions. In fact, we shall prove this indirectly, by showing that the Frenet
system is equivalent to a second order Ricatti differential equation. As it is known from
the theory of ordinary differential equations, we can find the general solution of a Ricatti
equation if and only if a particular solution is already available (and there is no general
96 Chapter 3. The integration of the natural equations of a curve
procedure for finding such a solution).
Notice, first of all, that the Frenet system contains three copies of the scalar system:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
X
·
= k(s)Y,
Y
·
= −k(s)X + χ(s)Z
Z
·
= −χ(s)Y
(3.1.1)
and the solution of the system should verify, also, the condition
X
2
+ Y
2
+ Z
2
= 1. (3.1.2)
The idea of the method we are going to describe (and which goes back to Sophus Lie
and Gaston Darboux), relies exactly on the supplementary condition (3.1.2). Namely, it
was observed that this equation can be decomposed (over the complex numbers) as
(X + iY)(X − iY) = (1 − Z)(1 + Z).
We introduce now the complex functions u and v by letting
u =
X + iY
1 − Z
; −
1
v
=
X − iY
1 − Z
. (3.1.3)
Clearly, w and −1/v are conjugated to each other. It is now possible to express X, Y, Z in
terms of u and v. One can check easily that
X =
1 − uv
u − v
; Y = i
1 + uv
u − v
; Z =
u + v
u − v
. (3.1.4)
Thus, the solution of the Frenet system amounts to the finding of the complex functions
u and v. Now, an easy manipulation of the formulas show that both u and v are solution
of the Riccati equation
dw
ds
= −
i
2
χ(s) − ik(s)w +
iχ(s)
2
w
2
. (3.1.5)
This is, as we mentioned earlier, an indirect prove of the fact that the natural equations
of a space curve cannot be integrated through quadratures, in the general situation.
3.2 Examples for the integration of the natural equation of a
plane curve
We saw, in the previous section, that the plane curves of constant curvature are the cir-
cles. We shall give, here, other interesting examples of natural equation for plane curves
3.2. Examples for the integration of the natural equation of a plane curve 97
that can be integrated. We remind that in the case of plane curves the problem is not the
integrability of the natural equation, but, rather, the possibility of expressing the solution
in terms of elementary functions. Usually this is not the case, and this makes even more
interesting the (few) cases in which this possibility does exist. For convenience, we shall
use, for the rest of this paragraph, the curvature radius R = 1/k instead of the curvature.
The logarithmic spiral. In this case the curvature radius is given by
R = a s, (3.2.1)
where a is a constant. Let α be the contingency angle. Then, as we know, we have

ds
=
1
R(s)
=
1
as
,
therefore, by integrating and neglecting the integration constant, we get
α =
1
a
ln s, whence s = e

.
To get the coordinates, we have to integrate the system of differential equations
dx
ds
= cos α,
dy
ds
= sin α.
We have (neglecting, again, the integration constants)
x =

cos αds = a

cos αe

dα =
ae

a
2
+ 1
(a cos α + sin α)
and, analogously,
y =

sin αds =
ae

a
2
+ 1
(a sin α − cos α).
Hence, the parametric equations of the logarithmic spiral are

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x =
ae

a
2
+ 1
(a cos α + sin α)
y =
ae

a
2
+ 1
(a sin α − cos α)
. (3.2.2)
The logarithmic spiral (see figure 3.1) is most conveniently described through its polar
98 Chapter 3. The integration of the natural equations of a curve
Figure 3.1: The logarithmic spiral
equation, r = r(ϕ). We shall indicate now how can one obtain that equation starting from
the parametric equations. First of all, we notice that
x
2
+ y
2
=
a
2
e
2aα
a
2
+ 1
,
whence
r =

x
2
+ y
2
=
ae


a
2
+ 1
.
We put a = tan ψ. Then, for the polar angle we get
tan ϕ =
y
x
=
a sin α − cos α
sin α + a cos α
= −cot(α + ψ),
therefore ϕ = α + ψ +
π
2
, hence α = ϕ − ψ −
π
2
. Thus
r = sin ψe
ϕ−ψ−
π
2
,
i.e. the polar equation of the logarithmic spiral is of the form
r = C e
ϕ
, (3.2.3)
where C is a constant.
Cycloidal curves. They correspond to the natural equation
s
2
a
2
+
R
2
b
2
= 1, (3.2.4)
3.2. Examples for the integration of the natural equation of a plane curve 99
where a and b are nonvanishing constants. A possibility would be to express R in terms
of s and then integrate. However, this would lead to complications, due to the the pres-
ence of the square root and the sign ambiguity. We prefer, instead, to introduce a new
parameter t, through the relations
s = a sin t, R = b cos t.
It is, then, very easy to find the contingency angle in terms of this new parameter. Indeed,
we have
α =

1
R
ds =

1
b cos t
a cos t dt =
at
b
.
We can proceed now with the determination of the coordinates x and y, in terms of the
parameter t:
x =

cos αds =

cos
at
b
a cos t dt =
a
2
¸
b
a − b
sin
(a − b)t
b
+
b
a + b
sin
(a + b)t
b

,
y =

sin αds =

sin
at
b
a cos t dt = −
a
2
¸
b
a − b
cos
(a − b)t
b
+
b
a + b
cos
(a + b)t
b

.
The clothoid. This is the curve whose natural equation is
R =
a
2
s
. (3.2.5)
Thus, for the clothoid (also known as the Cornu’s spiral), the radius of curvature is
proportional to the inverse of the arc length. From this point of view, it is, to some
extent, similar to the logarithmic spiral, where the radius of curvature was proportional
to the arc length, rather than to its inverse. We include this curve here to show that even
in the case of a very simple expression of the curvature in terms of the arc length (in this
case the curvature is proportional to the arc length, i.e. it is a linear function), we might
not be able to find a parametric representation in terms of elementary functions.
The contingency angle is easily found:
α =
s
2
2a
, (3.2.6)
therefore the coordinates are
x =

cos
s
2
2a
ds, x =

sin
s
2
2a
ds. (3.2.7)
100 Chapter 3. The integration of the natural equations of a curve
Unfortunately, it is a well known fact from analysis that the integrals from (3.2.7) cannot
be expressed in terms of elementary functions. They carry the name of Fresnel integrals,
after the French physicists who used them first in his works on optics (the theory of
diffraction, to be more specific)
1
.
The catenary. For the catenary, the natural equation is
R = a +
s
2
a
, (3.2.8)
where a is a non-vanishing constant. Again, instead of integrating directly, to get the
contingency angle, we introduce, first, the new parameter t through the relations
s = a tan t, R =
a
cos
2
t
. (3.2.9)
Then the contingency angle will be
α ≡

1
R(s)
ds =

cos
2
t a
a cos
2
t
dt = t.
The coordinates are, now, easy to find in terms on the new parameter:
x =

cos αds =

a cos t
cos
2
t
dt = a

1
cos t
dt = ln
¸
1 + sin t
cos t

,
and, analogously,
y = a
1
cos t
.
It is not difficult to check that one can eliminate the parameter t from the previous two
equation and one gets the usual explicit equation of the catenary, i.e.
y = a cosh
x
a
. (3.2.10)
1
According to Gino Loria, in fact, these integrals were first considered by Euler, in 1781.
3.2. Examples for the integration of the natural equation of a plane curve 101
Figure 3.2: The catenary
The involute of the circle. Let us start with the natural equation
R
2
= 2as. (3.2.11)
We introduce a new parameter, t, such that
s =
at
2
2
, R = at.
Then the contingency angle is
α =

1
R
ds =

1
at
at dt = t,
therefore
x =

cos αds =

at cos t dt = a(cos t + t sin t)
and
y =

sin αds =

at sin t dt = a(sin t − t cos t),
which are, indeed, the parametric equations of the involute of the circle.
The tractrix. Finally, we start with the natural equation
R
2
+ a
2
= a
2
e
−2s/a
, (3.2.12)
where a is a non-vanishing constant. We introduce the parameter t such that
e

s
a
=
1
cos t
, R = a tan t.
102 Chapter 3. The integration of the natural equations of a curve
Figure 3.3: The tractrix
Then
α =

1
R
ds =

1
a tan t
(−a tan t) dt = −t,
therefore
x =

cos αds =

cos t(−a tan t) dt = a cos t
and
y =

sin αds =

sin t a tan t dt = a
¸
ln
1 + sin t
cos t
− sin t

.
Problems
1. Write the equations of the geometric locus of the points for which the product of their
distances to two fixed points F
1
and F
2
, with d(F
1
, F
2
) = 2b, is a constant quantity,
equal to a
2
(the curve is called a Cassini oval).
2. We consider a circle of diameter OA = 2a and its tangent at the point A. Through the
point O passes a half-line OC, where C is the intersection point between the half-line
and the circle. On this half-line one takes a segment OM, of length BC, where B is the
intersection point between the half-line and the tangent to the circle at A. If we let the
half-line OC rotate around the point O, then the point M describe a curve, called the
Diocles’ cissoid. Find the equations of this curve.
3. An arbitrary half-line OE, starting from the origin O, intersects at the points D and E
the circle
x
2
+

y −
a
2

2
=
a
2
4
and its tangent at C, the opposite of O on the circle. We draw, through the points D and
E, two straight line, parallel to the axes Ox and Oy, respectively. The two lines intersect
at M. Find the equations of the geometrical locus of M ( the witch of Maria Agnesi).
4. A point M is moving uniformly on a straight line ON, which is rotating with con-
stant angular velocity around the point O. Find the equations of the trajectory of M
(Archimede’s spiral).
5. A straight line OL rotates around the point O with a constant angular velocity ω. A
point M moves along the line OL with a speed proportional to the distance OM. Find
the equation of the curve described by the point M (the logarithmical spiral).
104 Problems
6. A straight line segment, of constant length, equal to 2a, moves on the plane, keeping
its ends, A and B, on the axes of a rectangular. coordinate system. We construct the
perpendicular OM from the origin of coordinates on this (M ∈ AB). Write the equation
of the geometrical locus of the point M (the four petals rosette).
7. On a fixed circle of radius a, one takes an invariable point O. A ray with the origin at
O rotates around O and intersects the circle in a variable point A. On this ray one takes
two points M
1
and M
2
, symmetrical with respect to A such that AM
1
= AM
2
= 2a. Find
the equations of the curve described by the points M
1
and M
2
(the cardioid).
8. We take a fixed point O on the circle r = 2a cos ϕ and a variable ray OM, with the
origin at O. Let A be the point of intersection between the ray and the circle. We take on
the ray two points M
1
and M
2
, symmetrical with respect to A, such that AM
1
= AM
2
=
2b. Write down the equations of the geometrical locus described by M
1
and M
2
when
the ray rotates around O (the Pascal’s limaçon). What curve do we get for a = b?
9. The straight line x = a cuts the Ox-axis at the point A and an arbitrary ray OB at the
point B. On the ray we take two points M
1
and M
2
, symmetrical with respect to B, such
that BM
1
− BM
2
= AB. Write down the equations of the geometrical locus described by
the points M
1
and M
2
when the ray rotates around the origin (the strophoid).
10. Through the point E

a,
π
2

, given through its polar coordinates, one constructs a
straight line parallel to the polar axis. A variable ray OK intersects this line at the point
K. On the ray one takes two points M
1
and M
2
, symmetrical with respect to the point
K, such that KM
1
= KM
2
= l. Write down the equations of the geometrical locus of the
points M
1
and M
2
(the Nicomede’s conchoid).
11. The ends of a straight line segment of constant length AB = a slide on the axes
of a rectangular coordinate system. The straight lines AC and BC are parallel to the
coordinate axes and they intersect each other at the point C. Find the equations of the
geometrical locus of the foot M of the perpendicular drawn from C on the straight line
AB ( the astroid).
12. A circle of radius a rules without sliding on a straight line. Find the parametrical
equations of the curve described by a point invariably linked to the circumference of the
circle (the cycloid).
13. A circular disk of radius a rules without sliding on a straight line. Find the paramet-
rical equations of the curve described by a point M, invariably linked to the disk, lying
at a distance d from the centre of the circle (for d < a the curve is called the shortened
cycloid, while for d > a — the lengthened cycloid).
Problems 105
14. A circular disk of radius r rules without sliding on a circle of radius R, at its exterior.
Find the parametrical equations of a point M, invariably linked to the circumference of
the mobile disk (the epicycloid).
15. A circular disk of radius r rules without sliding on a circle of radius R, at its interior.
Find the parametrical equations of a point M, invariably linked to the circumference of
the mobile disk (the hypocycloid).
16. Find the equation of the tangent and the equation of the normal for the following
curves:
a) r = a cos ti + b sin tj (the ellipse);
b) r =
a
2
¸
t +
1
t

i +
b
2
¸
t −
1
t

j (the hyperbola);
c) r = a cos
3
ti + a sin
3
tj (the astroid);
d) r = a(t − sin t)i + a(1 − cos t)j (the cycloid);
e) r = aϕcos ϕi + aϕsin ϕj (the Archimede’s spiral);
f) r =
at
1 + t
3
i +
at
2
1 + t
3
j (Descartes’ folium);
g) r =
a
1 + t
2
i +
a
t(1 + t
2
)
j (Diocles’ cissoid).
17. Find the equation of the tangent and the equation of the normal to the curves:
a) x
2
(x + y) − a
2
(x − y) = 0 at the point (0, 0);
b) 2x
2
− x
2
y
2
− 3x + y + 7 = 0 at the point (1, −2);
18. The sides of a right angle are tangent to the astroid

¸
¸
¸
¸
¸
¸
x =
3
4
Rcos
t
4
+
1
4
Rcos
3t
4
y =
3
4
Rsin
t
4

1
4
Rsin
3t
4
.
Find the geometrical locus of the vertex of this angle when its sides occupy all the pos-
sible positions, remaining perpendicular to each other and tangent to the astroid.
106 Problems
19. Show that the Bernoulli’s lemniscate
(x
2
+ y
2
)
2
− 2a
2
(x
2
− y
2
) = 0
is the set of the symmetrical points of the centre of an equilateral hyperbola with respect
to the tangents to this hyperbola.
20. At which point the tangent to the parabola y = x
2
makes an angle of 45

with the
x-axis?
21. Show that the inclination angle ϕ of the tangent to the curve
y = x
5
+ 2x
3
+ x − 1
on the x-axis belongs to the interval
π
4
≤ ϕ ≤
π
2
.
22. Find the equations of the tangents to the curve
y =
1
1 + x
2
at the points where the curve intersects the hyperbola
y =
1
1 + x
23. Show that the curves
y = a sin
x
a
, y = a tan
x
a
, y = a ln
x
a
, a ∈ R,
intersect the x-axis under the same angle, for any value of the parameter a, provided all
the functions involved are defined for the respective value.
24. Find the tangents to the astroid
x
2
3
+ y
2
3
= a
2
3
which are the most distant from the origin of the coordinates.
25. Find the intersection angle between two coaxial parabolas if the vertex of each one
coincides with the focus of the other one.
Problems 107
26. Show that all the normals to the curve

¸
¸
¸
¸
¸
¸
x = a(cos t + sin t)
y = a(sin t − cos t)
lie at the same distance with respect to the origin.
27. Show that the segment from the normal to the curve

¸
¸
¸
¸
¸
¸
x = 2a sin t + a sin t cos
2
t
y = −a cos
3
t
intercepted between the coordinate axes has constant length, equal to 2a.
28. It is possible to draw two pairs of tangents to the Bernoulli’s lemniscate (x
2
+y
2
)
2
=
a
2
(x
2
− y
2
), parallel to a given direction. Show that the straight lines connecting the
tangency points of each pair intercept an angle of 120

.
29. Prove that the circles
x
2
+ y
2
+ a
1
x + b
1
y + c
1
= 0
x
2
+ y
2
+ a
2
x + b
2
y + c
2
= 0
intersect each other under a right angle iff
a
1
a
2
+ b
1
b
2
= 2(c
1
+ c
2
).
30. Show that if all the normals to a given curve are passing through the same point,
then the curve is an arc of circle.
31. Find the equation of the curve from whose points a given parabola is seen under a
constant angle, α. Which curve do we get for α =
π
2
?.
32. Find the envelope of a family of straight lines passing through the extremities of a
pair of conjugates diameters of an ellipse.
33. Find the envelope of a family of straight lines which determine on a right angle a a
triangle of constant area.
34. Find the envelope of a family of straight lines which determine on a right angle a a
triangle of constant perimeter.
108 Problems
35. Find the envelopes of the families of curves
a) r = (t
2
+ λ) i + (t
3
+ λ) j;
b) r = (t
2
+ λ) i + t
3
j.
36. Let O be the centre of an equilateral hyperbola M – an arbitrary point on the hy-
perbola. The segment OM is projected on the assymptotes of the hyperbola and on the
projections, as semiaxes, one constructs ellipses. Find the envelope of the family of
ellipses obtained this way.
37. There are given two concentrical circles, of radii a abd b, respectively. One con-
structs a family of circles, with the centres on the circle of radius b, their radii being,
all of them, equal to a, and a second family of circles, this time with the centres on the
circle of radius a and having the radii equal to b. Show that the two families of circles
have the same envelope and find this envelope.
38. Find the envelope of the family of circles having as diameters the chords of a
parabola, passing through the focus of the parabola.
39. A radius vector of an ellipse projects on two conjugate diameters of the ellipse and
on the projections, as semiaxes, one builds ellipses. Find the envelope of the family of
ellipses obtained this way.
40. Find the envelope of a family of circles with the centers on a given ellipse and
passing through the same focus of the ellipse.
41. Find the envelope of the family of curves
x
2
λ
+
y
2
µ
= 1,
where λ + µ = 1.
42. A vertex of a plane curve is a point of the curve at which the radius of the osculating
circle is stationary. Find the vertices of the curve y = ln x.
43. Find the osculating circle of minimum radius of the equilateral hyperbola xy = 1.
44. Find the radius of the osculating circle at an arbitrary point of the conical section
a
11
x
2
+ 2a
12
xy + a
22
y
2
+ 2a
1
x + 2a
2
y + a = 0.
Problems 109
45. Let M be an arbitrary point of the cycloidcycloid and P – the corresponding point of
the generating circle. We extend the segment MP with a distance PC = MP. Show that
C is the center of the osculating circle of the cycloid at M.
46. Compute the radius of the osculating circle of the hyperbola
x
2
a
2

y
2
b
2
= 1 at one of
its vertices and give a geometrical method for the construction of this osculating circle.
Consider the particular case of the equilateral hyperbola x
2
− y
2
= a
2
.
47. We consider a curve (γ) given through the vectorial equation r = r(t), such that the
function r is defined and at least three times continuously differentiable on an interval
[a, b], while r
·
× r
··
0 on this interval. The vector
ν
a
= 3r
··
× (r
·
× r
··
) − r
·
× (r
·
× r
···
)
is called the directing vector of the affine normal of the curve (γ) at the considered point.
Show that if at a point M
0
of the curve (γ) this vector is not perpendicular on the tangent
to the curve at this point, then the osculating circle of the curve at the point M
0
intersects
the curve.
48. Find the directing vector of the affine normal for the curves:
a) r = a cos ti + b sin tj (the ellipse);
b) r =
a
2
¸
t +
1
t

i +
b
2
¸
t −
1
t

j (the hyperbola);
c) y = ax
2
(the parabola);
d) y = ln x
and find the position of the osculating circle with respect to the curve in each case.
49. Show that the circle cercul (x−a)
2
+(y−a)
2

8a
2
9
= 0 and the parabola

3x+

3y =
2

a have a third order contact at the point

a
3
,
a
3

.
50. Find the curvature radii of the following curves:
a) y = sin x at the vertex;
b)

¸
¸
¸
¸
¸
¸
x = (1 + m) cos mt − amcos(1 + m)t
y = (1 + m) sin mt − amsin(1 + m)t
(the epicycloid);
c) y = cosh
x
a
(the catenary);
110 Problems
d) x
2
y
2
= (a
2
− y
2
)(b + y)
2
(the conchoid);
e) r = a cos
3
t i + a sin
3
t j (the astroid).
51. Find the evolutes of the following curves:
a) r = a cos t i + b sin t j (the ellipse);
b) r = x i + ax
2
j (the parabola);
c) r =
a
2
¸
t +
1
t

i +
b
2
¸
t −
1
t

j (the hyperbola);
d) r = a(t − sin t)i + a(1 − cos t)j (the cycloid);
e) r = a cos
3
ti + a sin
3
tj (the astroid).
52. Find the equation of the evolute of the curve
R = r −
r
·
|r
·
|
t

0
|r
·
|dt.
53. Find the involute of a circle.
54. Find the involute of the parabola x
2
= 2py.
55. Find the parametrical equations of the curves having the following natural equations:
a) R = as;
b)
s
2
a
2
+
R
2
b
2
= 1;
c) RS = a
2
;
d) R = a +
s
2
a
;
56. Show that the curve

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = sin 2ϕ,
y = 1 − cos 2ϕ,
z = 2 cos ϕ
lies on a sphere.
Problems 111
57. Consider the curve obtained by intersecting the sphere x
2
+ y
2
+ z
2
= a
2
with the
cylinder x
2
+ y
2
= ay (the Viviani’s window).
a) Find a set of parametrical equations of the curve.
b) Find the curvature and the torsion at an arbitrary point of the curve.
c) Find the projection of the curve on the coordinate plane yOz.
58. Consider the curve

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = p

p
2
− q
2
cos t
y = q

p
2
− q
2
(1 + sin t)
z =

p
2
− q
2
(1 + sin t)
where p, q are constant, while p > q.
1. Show that this curve lies on the elliptical paraboloid
x
2
p
2
+
y
2
q
2
= 2z.
2. Show that this curve is a circle, by computing its curvature and torsion.
3. Check that the osculating plane to the curve is one of the planes of circular section
in the paraboloid.
59. Find the projection of the coordinate plane xOy of the intersection curve between
the paraboloid z = x
2
+ y
2
and the plane x + y + z = 1.
60. Find the equation of the tangent to the curve

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x =
t
4
4
,
y =
t
3
3
,
z =
t
2
2
parallel to the plane x + 3y + 2z = 0.
61. Write the equation of the tangent to the curve

¸
¸
¸
¸
¸
¸
x
2
+ y
2
= 10,
y
2
+ z
2
= 25
at the point (1, 3, 4).
112 Chapter 3. The integration of the natural equations of a curve
62. Show that the tangents to the curve
x = a(sin t + cos t), y = a(sin t − cos t), z = be
−t
intersects the coordinate plane xOy after the circle
x
2
+ y
2
= 4a
2
63. Find the equation of the normal plane at an arbitrary point of the curve

¸
¸
¸
¸
¸
¸
z = x
2
+ y
2
,
y = x.
64. Show that the normal planes to the curve

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a cos t,
y = a sin αsin t,
z = a cos αsin t
are passing through the straight line

¸
¸
¸
¸
¸
¸
x = 0,
z + y tan α = 0.
65. Show that (τ, β,
˙
β) = χ.
66. Show that (
˙
β,
¨
β,
...
β) = χ
5
¸
k
χ

.
67. Show that (˙ τ, ¨ τ,
...
τ) = k
5

χ
k

.
68. Show that if the principal normals to a curve form a constant angle with the direction
of the vector e, then
d
ds
=
k
2
+ χ
2
k
d
ds
χ
k
+ χ = 0
and conversely: if the previous relation is verified, then the principal normals to the
curve form a constant angle with the direction of a vector. Find this vector.
69. Show that if all the osculating planes to a given curve (which is not a straight line)
contain the same vector, then the curve is a plane curve.
Part II
Surfaces
CHAPTER 4
General theory of surfaces
4.1 Parameterized surfaces (patches)
Definition. A regular parameterized surface (patch) in R
3
is a smooth map r : U → R
3
,
(u, v) → r(u, v), where U is a domain (an open, connected subset) in R
2
, while r is
subject to
r
·
u
× r
·
v
0. (4.1.1)
The condition (4.1.1) is called the regularity condition.
A parameterized surface is usually denoted by (U, r), (U, r = r(u, v)) or just r =
r(u, v).
Definition. The set r(U) ⊂ R
3
is called the support of the parameterized surface (U, r).
Remark. Usually, one and the same point of the support of a parameterized surface
(U, r) may correspond to several distinct pairs (u, v), since the map r is not assumed to
be injective.
Definition. Two parameterized surfaces (U, r) and (V, r
1
) are called equivalent if there
is a diffeomorphism λ : U → V such that r = r
1
◦ λ.
Remark. The supports of two equivalent parameterized surfaces always coincide.
116 Chapter 4. General theory of surfaces
Examples. 1. If U = R
2
, while r = r
0
+ ua + vb, a × b 0, the the support of the
parameterized surface is the plane passing through r
0
, perpendicular to the vector
a × b. This parameterized surface is called a plane.
2. Let U = ¦(u, v) ∈ R
2
|π/2 < u < π/2, 0 < v < 2π¦ and
r(u, v) = (Rcos u cos v, Rcos u sin v, Rsin u).
The support of this parameterized surface is the sphere of radius R, centred at the
origin of R
3
, with a meridian removed. The parameters u and v are analogues of
the geographical coordinates.
3. Let U = R
2
, r = r
0
+ ua + v
3
b, a × b 0. The support of this parameterized
surface is identical to the support of the parameterized surface from the example
1), although the two surfaces are not equivalent, since the map (u, v) → (u, v
3
) is
not a diffeomorphism.
4.2 Surfaces
Definition. A subset S ⊂ R
3
is called a regular surface if each point a ∈ S has an open
neighbourhood W in S such that there is a parameterized surface (U, r) with r(U) = W,
while the map r : U → W is a homeomorphism. The pair is called a local parameter-
ization of the surface S around the point a, while the support r(U) is called the domain
of the parameterization. A surface S which has a global parameterization (i.e. a local
parameterization (U, r) for which r(U) = S ) is called a simple surface.
4.2.1 Representations of surfaces
The same kind of representations we used for the case of curves are, equally, available
for surfaces.
Parametrical representation. If S is a surface and (U, r) is a local parameterization
of S , then, if r(u, v) = (x(u, v), y(u, v), z(u, v)), then the equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = x(u, v)
y = y(u, v)
z = z(u, v)
, (u, v) ∈ U,
are called the parametric equations of the surface. We would like to emphasize, once
more, that these are only local equations, they cannot be used to describe all the points
of the surface, unless we are dealing with a global parameterization of a simple surface.
4.2. Surfaces 117
Explicit representation. If f : U → R is a smooth function, where U ⊂ R
2
is a
domain, then its graph, S = ¦(x, y, z) ∈ R
3
| z = f (x, y)¦ is a simple surface. Indeed, we
have the global parameterization r : U → R
3
, r(u, v) = (u, v, f (u, v)).
Implicit representation. Let F : V → R, with V ⊂ R
3
an open set, be a smooth
function. We denote by S = ¦(x, y, z) ∈ R
3
| F(x, y, z) = 0¦ the 0-level set of F. If, at
each point of S , the vector
grad F =

∂F
∂x
,
∂F
∂y
,
∂F
∂z

is different from zero, then S is a surface. Indeed, for instance, at (x
0
, y
0
, z
0
) ∈ S , we
have F
·
z
(x
0
, y
0
, z
0
) 0, then, from the implicit function theorem, there is an open (in the
topology of R
3
) neighbourhood M of (x
0
, y
0
, z
0
) such that the set M∩S (which is an open
neighbourhood of (x
0
, y
0
, z
0
), this time in the topology of S ) is the graph of a smooth
function z = f (x, y), therefore, according to the previous paragraph, there is a local
parameterization of S around the point (x
0
, y
0
, z
0
). We note that this parameterization is
global for M ∩ S but, generally, it is not so for the entire S . Even if F
·
z
is different from
zero on the entire S , it is still not sure that the function f can be defined globally. Thus,
in general, unlike that case of the surfaces given explicitly, the implicitly given surfaces
are not simple.
Examples. 1. The plane Π passing through r
0
and having the direction given by the
vectors a and b, with a×b 0 is a simple surface with the global parameterization
r(u, v) = r
0
+ua+vb. In projection on the coordinate axes, the parametric equations
of the plane will be

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = x
0
+ ua
x
+ vb
x
y = y
0
+ ua
y
+ vb
y
z = z
0
+ ua
z
+ vb
z
, (u, v) ∈ R
2
.
2. Revolution surfaces
Let C be a curve in the plane xOz, which does not intersect the z-axis, and S – the
subset of R
3
obtained by rotating C around the z-axis. Let v be the rotation angle
of the plane xOz and a
·
– the point of S obtained by rotating the point a ∈ C by an
angle v
0
. Let (I, ρ = ρ(t)) be a local parameterization of the curve C around the
point a, ρ(t) = (x(t), z(t)). The we get the local parameterization os S around a
·
,
r(t, v) = (x(t) cos(v + v
0
), x(t) sin(v + v
0
), z(t)),
defined on the domain U = I ×


π
2
,
π
2

.
118 Chapter 4. General theory of surfaces
3. The sphere. Let F : R
3
→ R, F(x, y, z) = x
2
+ y
2
+ z
2
− R
2
. F is, obviously, a
smooth function and its 0-level set
S
2
R
= ¦(x, y, z) ∈ R
3
| F(x, y, z) = 0¦
is the sphere of radius R, with the centre at the origin. The gradient of F is
grad F = ¦2x, 2y, 2z¦
and, obviously, it does not vanish on the sphere S
2
R
, therefore this sphere is a reg-
ular surface. Note that S
2
R
is not simple since it is compact subset of R
3
, therefore
it cannot be homeomorphic to an open subset of R
2
, which is not compact.
4. The torus. We choose now F : R
3
→ R,
F(x, y, z) = (

x
2
+ y
2
− a)
2
+ z
2
− b
2
, 0 < b < a.
Its 0-level set,
T
2
= ¦(x, y, z) ∈ R
3
|F(x, y, z) = 0¦
is called the 2-dimensional torus. If we compute the derivatives of F with respect
to the coordinates, we get

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
F
·
x
= 2

x
2
+ y
2
− a

x

x
2
+y
2
F
·
y
= 2

x
2
+ y
2
− a

y

x
2
+y
2
F
·
z
= 2z
,
therefore the gradient of F vanishes iff

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = y = z = 0 or
x
2
+ y
2
= 0, y = 0, z = 0 or
x = 0, x
2
+ y
2
= a
2
, z = 0 or
x
2
+ y
2
= a
2
, z = 0
.
It is easy to check that grad F is different from zero on T
2
, therefore the torus is a
surface (again, it is compact, therefore it cannot be simple).
The torus can also be obtained by rotating the circle (x − a)
2
+ z
2
= b
2
(lying in
the plane xOz) around the axis Oz.
4.3. The equivalence of local parameterizations 119
Figure 4.1: The torus
4.3 The equivalence of local parameterizations
Definition. Let S be a surface, (U, r) – a local parameterization and W = r(U). Then the
map r
−1
: W → U is a bijection, associating to each point of W a pair of real numbers
(u, v) ∈ U. This correspondence is called a curvilinear coordinate system on S or a
chart on S .
Remark. We ought to mention here that in many books the fundamental objects used
to describe a surface are not the local parameterizations, but the charts. Of course,
the two approaches are completely equivalent, but they came from different directions.
The description of surfaces by using local parameterizations has, probably, the origin in
mathematical analysis, where the surfaces are thought off either as images of functions
or as graphs of functions, in any case, the central objects is the function. The “charts’s
approach” has the origin in chartography. In fact, assigning a local chart to a surface
simply means to apply that surface on a portion of a plane, in other words, what is meant
is the construction of a “map” of the surface and, actually, in the advanced differential
geometry, a collection of charts that cover the entire surface is called an atlas, exactly as
happens in chartography.
Theorem4.3.1. (of the parameters’ change) Let (U, r) and (U
1
, r
1
) be two local parame-
terizations of a surface S and r(U) = r(U
1
). Then there is a diffeomorphism λ : U → U
1
such that r = r
1
◦ λ. The diffeomorphism λ is called a parameters’ change.
Before proving the theorem, some remarks are in order. If a change of parameters
λ does exist, then, from the relation r = r
1
◦ λ, we should have, of course, λ = r
−1
1
◦ r.
The real difficulty is to show that λ and its inverse are smooth. Although r is smooth,
r
−1
is not
1
, because its domain, r
1
(U
1
), is not an open subset of the Euclidean space R
3
.
1
not in the classical sense, at least
120 Chapter 4. General theory of surfaces
We will show, however, in the following lemma, that, locally, r
−1
1
is the restriction of a
smooth function. We ought to emphasize, however, that this representation of r
−1
1
is only
local: in general, r
−1
1
cannot be written, globally, as the restriction of a single smooth
function, defined on an open set (with respect to the topology of R
3
), which contains the
set r
1
(U
1
).
Lemma. Let (U, r) be a local parameterization of the surface S , r(U) = W and r
−1
:
W → U – the inverse map. Then, for each point a ∈ W there is an open set (in the
topology of R
3
) B ÷ a and a smooth map G : B → U such that r
−1
|
W∩B
= G|
W∩B
.
Proof of the lemma. Let r(u, v) = ( f
1
(u, v), f
2
(u, v). f
3
(u, v)) and a = r(u
0
, v
0
). Due to the
regularity of r, the Jacobi matrix

¸
¸
¸
¸
¸
¸
¸
¸
¸
f
·
1u
f
·
1v
f
·
2u
f
·
2v
f
·
3u
f
·
3v

¸
¸
¸
¸
¸
¸
¸
¸
¸
has the rank two. Without restricting the generality, we may assume that

f
·
1u
f
·
1v
f
·
2u
f
·
2v

0.
Then, from the inverse function’s theorem for the map
f : (u, v) −→ (x = f
1
(u, v), y = f
2
(u, v)),
there is an open neighbourhood V of the point (u
0
, v
0
) in U and an open neighbourhood
¯
V of the point (x
0
= f
1
(u
0
, v
0
), y
0
= f
2
(u
0
, v
0
)) in the plane xOy such that f : V →
¯
V
is a diffeomorphism. Since the map r : U → W is a homeomorphism, r(V) is an
open neighbourhood in S of the point a = r(u
0
, v
0
), therefore, in R
3
there is an open
neighbourhood B of the point a such that r(V) = B ∩ S = B ∩ W. Let p : R
3
→ R
2
:
(x, y, z) → (x, y) be the orthogonal projection on the coordinate plane xOy. We are going
to show that the map G = ( f
−1
◦ p)|
B
: B → U is the map searched for. Indeed, G is
smooth, as a composition of smooth maps. Moreover, to each point (x, y, z) from B ∩ W
corresponds a single point (u, v) = r
−1
(x, y, z) from V, and to each point (x, y) ∈
¯
V – the
point (u, v) = f
−1
(x, y) from V. Thus, for the points (x, y, z) ∈ B ∩ W, we have
r
−1
(x, y, z) = f
−1
(x, y) = f
−1
(p(x, y, z)) = G(x, y, z).

4.3. The equivalence of local parameterizations 121
Proof of the theorem. Let (U, r) and (U
1
, r
1
) – two local parameterizations of the surface
S such that r(U) = r
1
(U
1
) = W. We consider the map λ = r
−1
1
◦ r : U → U
1
. Since
r
1
: U
1
→ W is a homeomorphism, the same is true for r
−1
1
: W → U
1
and, thus, λ is
a homeomorphism, as a composition of two homeomorphisms. We have to prove only
that λ and λ
−1
are smooth. To prove the smoothness of λ este it is enough to prove that
each point (u
0
, v
0
) ∈ U has an open neighbourhood V ⊂ U such that λ|
V
is smooth. We
apply the previous lemma to the parameterization (U
1
, r
1
) at a point a = r
1
(λ(u
0
, v
0
)).
Let G : B → U
1
be the smooth map for which r
−1
1
|
B∩W
= G|
B∩W
and V = r
−1
(B ∩ W).
Then λ|
V
= r
−1
1
◦r|
V
= (G◦r)|
V
and, therefore, λ|
V
is smooth, as a restriction of a smooth
map. The smoothness of λ
−1
is proved in the same way, replacing the parameterization
r
1
by the parameterization r.
Locally, each surface is the support of a parameterized surface. The converse is not
true, i.e. the support of an arbitrary parameterized surface is not a surface. Still, if we
choose an arbitrary parameterized surface, by restricting the domain, we can obtain a
parameterized surface whose support is a regular surface. Thus, we have:
Theorem4.3.2. Let (U, r) be a regular parameterized surface. Then each point (u
0
, v
0
) ∈
U has an open neighbourhood V ⊂ U such that the r(V) is a simple surface in R
3
, for
which the pair (V, r|
V
) is a global parameterization.
Proof. The only extra condition we have to impose on V is that the map r|
V
: V → r(V)
to be a homeomorphism. Let r(u, v) = ( f
1
(u, v), f
2
(u, v), f
3
(u, v)). Without restricting the
generality, we may assume that the Jacobian of the map f : (u, v) → (x = f
1
(u, v), y =
f
2
(u, v)) is different from zero at (u
0
, v
0
). Then, from the inverse function’s theorem,
there is an open neighbourhood V ⊂ U of the point (u
0
, v
0
) and an open neighbourhood
¯
V of the point (x
0
, y
0
) = f (u
0
, v
0
) such that the map F : V →
¯
V is a diffeomorphism. We
shall prove first the injectivity of the map r|
V
: V → r(V). Let (u
1
, v
1
), (u
2
, v
2
) ∈ V, such
that r(u
1
, v
1
) = r(u
2
, v
2
). Then, in particular,
f
1
(u
1
, v
1
) = f
1
(u
2
, v
2
) ¸ si f
2
(u
1
, v
1
) = f
2
(u
2
, v
2
),
i.e. f (u
1
, v
1
) = f (u
2
, v
2
). Or, f is diffeomorphism, so it is, in particular, an injective
map, therefore (u
1
, v
1
) = (u
2
, v
2
). The map r : U → R
3
is continuous, so it is continuous
also its restriction r|
V
: V → R
3
. Since on r(V) we take the subspace topology, the map
r|
V
: V → r(V) is, also, continuous. To prove the continuity of the inverse map, we note
that its is a composition of the following continuous maps: (x, y, z) ∈ r(V) → (x, y) ∈
¯
V → (u, v) = f
−1
(x, y) ∈ V, as seen in the proof of the lemma.
122 Chapter 4. General theory of surfaces
4.4 Curves on a surface
We shall say that a smooth parameterized curve (I, ρ = ρ(t)) lies on a surface S if its
support ρ(I) is included in S . It is easy to describe the parameterized curves whose
support is contained into the domain of a parameterization (U, r) of the surface S .
Theorem 4.4.1. Let (U, r) be a parameterization of the surface S and (I, ρ = ρ(t)) a
smooth parameterized curve whose support is included in r(U). Then there is a single
smooth parameterized curve (I, ˜ ρ) on U such that
ρ(t) ≡ ρ( ˜ ρ(t)). (4.4.1)
Conversely, any smooth parameterized curve ˜ ρ on U defines, through the formula (4.4.1),
a smooth parameterized curve on r(U). The regularity of ρ at t is equivalent to the
regularity of ˜ ρ at t.
Proof. Since the map r : U → r(U) is a homeomorphism, while ρ(I) ⊂ r(U), from the
formula (4.4.1), we can obtain ˜ ρ, by putting
˜ ρ = r
−1
◦ ρ.
Clearly, ˜ ρ is continuous, as a composition of two continuous maps. We shall check now
that ˜ ρ is, actually, smooth. Let t ∈ I, then ρ(t) ∈ r(U). According to the lemma from
the previous paragraph there is an open neighbourhood B of the point ρ(t) in R
3
and a
smooth map G : B → U such that r
−1
|
B∩r(U)
= G|
B∩r(U)
. Therefore, the map ˜ ρ, can be
represented, in the neighbourhood of the point t, as a composition G◦ ρ of smooth maps
and, thus, it is smooth. The converse affirmation can be proved even simpler, because
we have
ρ = r ◦ ˜ ρ
and, as r and ˜ ρ are smooth, so is ρ.
To verify the equivalence of the regularity conditions for ρ and ˜ ρ, we consider the
components of the path ˜ ρ:
˜ ρ = (u(t), v(t)).
Then the equality (4.4.1) becomes
ρ(t) = r(u(t), v(t)).
Differentiating this relation, we get
ρ

(t) = r
·
u
u
·
(t) + r
·
v
v
·
(t).
4.5. The tangent vector space, the tangent plane and the normal to a surface 123
Since the vectors r
·
u
and r
·
v
are not colinear (because the surface, as always, is supposed
to be regular), from the previous relation it follows that ρ

(t) = 0 iff u
·
(t) = 0 and
v
·
(t) = 0, i.e. iff ˜ ρ
·
(t) = 0.
Definition. The parameterized curve ˜ ρ(t) on the domain U is called the local repre-
sentation of the parameterized curve ρ(t) in the local parameterization (U, r), while the
equations

¸
¸
¸
¸
¸
¸
u = u(t)
v = v(t)
are called the local equations of ρ(t) in the considered parameterization.
Example. Let (U, r = r(u, v)) a local parameterization S and (u
0
, v
0
) ∈ U. We consider,
in r(U) ⊂ S the paths defined by the local equations

¸
¸
¸
¸
¸
¸
u = u
0
+ t
v = v
0
(4.4.2)
and

¸
¸
¸
¸
¸
¸
u = u
0
v = v
0
+ t
(4.4.3)
It is easy to see that the supports of these paths lie, indeed, on S . Through each r(u
0
, v
0
) ∈
r(U) pass exactly two such curves, one of each kind. The parameterized curves are called
coordinate lines or coordinate curves on the surface S , in the local parameterization
(U, r).
4.5 The tangent vector space, the tangent plane and the nor-
mal to a surface
Let us denote, for any a ∈ R
3
, by R
3
a
the space of bound vectors with the origin at a. This
is, obviously, a 3-dimensional vector space, naturally isomorphic to R
3
.
Definition 4.5.1. A vector h ∈ R
3
a
is called a tangent vector to the surface S at the point
a if there is a parameterized curve (I, ρ(t)) on S and a t
0
∈ I such that ρ(t
0
) = a and
ρ

(t
0
) = h. Thus, a tangent vector to a surface is just a tangent vector to a curve on the
surface.
We shall denote by T
a
S the set of all the tangent vectors to the surface S at a ∈ S .
The following lemma is trivial, but it plays an important role in the sequel.
124 Chapter 4. General theory of surfaces
Lemma. Let ρ = ρ(t) be a parameterized curve on S , given by the local equations
u = u(t), v = v(t), with respect to a local parameterization (U, r) of S . Then we have the
relation
ρ

(t) = u
·
(t)r
·
u
(u(t), v(t)) + v
·
(t)r
·
v
(u(t, v(t)). (4.5.1)
Proof. We just differentiate the relation ρ(t) = r(u(t), v(t)) with respect to t.
Theorem 4.5.1. The set T
a
S is a 2-dimensional vector subspace of R
3
. If (U, r) is a local
parameterization of S , while a = r(u
0
, v
0
), then the vectors r
·
u
(u
0
, v
0
) and r
·
v
(u
0
, v
0
) make
up a basis of this subspace, called the natural basis or the coordinate basis of the tangent
space.
Proof. Let (U, r) be a local parameterization of S , with a = r(u
0
, v
0
). if the parame-
terized curve (I, ρ(t)) is on the surface and ρ(t
0
) = a, then, restricting, if necessary, the
interval I, we may assume that ρ(I) ⊂ r(U), while its local equations in this parameter-
ization of the surface are u = u(t), v = v(t). Then, from the formula (4.5.1), it follows
that
ρ

(t
0
) = u
·
(t
0
)r
·
u
(u
0
, v
0
) + v
·
(t
0
)r
·
v
(u
0
, v
0
).
Conversely, any vector of the form
h = αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
)
is tangent to the parameterization curve given by the local equations

¸
¸
¸
¸
¸
¸
u = u
0
+ αt
v = v
0
+ βt
,
which is a curve on S , passing through a for t
0
, therefore h ∈ T
a
S .
The vector space T
a
S is called the tangent space to S at a. As we mentioned before,
R
3
a
is naturally
2
isomorphic to R
3
. Based on this isomorphism, we may consider, when
is convenient, that T
a
S is, in fact, a subspace of R
3
rather than of R
3
a
. In this case, T
a
S
is a vectorial plane, in the sense that it passes through the origin of R
3
a
; then the plane
passing through a and having T
a
S as directing plane (i.e. the plane parallel to T
a
S ,
passing through a), is called the tangent plane to S at the point a and it is denoted by
Π
a
S .
2
In this context, naturally means that there is an isomorphism which is independent of the choice of
bases in the two vector spaces.
4.5. The tangent vector space, the tangent plane and the normal to a surface 125
If r(u, v) = (x(u, v), y(u, v), z(u, v)) is a local parameterization of the surface S and
a = r(u
0
, v
0
) = (x
0
, y
0
, z
0
) ∈ S , then, clearly, the equation of the tangent plane of S at a
should be

X − x
0
Y − y
0
Z − z
0
x
·
u
y
·
u
z
·
u
x
·
v
y
·
v
z
·
v

= 0.
Let, now, (U, r = r(u, v)) be a parameterization of S and (u
0
, v
0
) ∈ U. If we modify
the arguments by ∆u = α∆t, ∆v = β∆t, with α, β ∈ R fixed, the Taylor’s formula gives:
r(u
0
+ ∆u, v
0
+ ∆v) = r(u
0
, v
0
) + ∆t (αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
)) + ∆t ε,
with lim
∆t→0
ε = 0. Using this formula, we shall give another characterization of the tangent
plane. Let Π be a plane in R
3
, passing through a = r(u
0
, v
0
), d – the distance from the
point ∆a = r(u
0
+ ∆u, v
0
+ ∆v) to the plane Π, and h – the distance between the points a
and ∆a.
Theorem 4.5.2. The plane Π is the tangent plane to the surface S at the point a iff for
any modification of the arguments of the form ∆u = α∆t, ∆v = β∆t, with α, β ∈ R,
α
2
+ β
2
0, we have
lim
∆t→0
d
h
= 0, (4.5.2)
i.e. the plane and the surface have a first order contact at a.
Proof. Let n be the versor of the normal to the plane Π, ∆r = r(u
0
+ ∆u, v
0
+ ∆v) −
r(u
0
, v
0
). Then d = ∆r n, h = |∆r|. Replacing ∆r by its expression, we get:
lim
∆t→0
d
h
= lim
∆t→0
∆t (αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
) + ε) n
|αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
) + ε|
= ±
(αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
)) n
|αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
)|
,
therefore
lim
∆t→0
= 0 ⇐⇒ (αr
·
u
(u
0
, v
0
) + βr
·
v
(u
0
, v
0
)) n = 0. (4.5.3)
Necessity. If Π is the tangent plane, then the vectors r
·
u
and r
·
v
, as directing vectors of Π,
are orthogonal to n, therefore the relation (4.5.3) is fulfilled.
Sufficiency. Let us assume, now, that (4.5.3) is fulfilled. Choosing α = 1, β = 0, we
get r
·
u
n = 0. In the same way, for α = 0, β = 1, one obtains r
·
v
× n = 0. Thus, n is
orthogonal to the tangent plane, i.e. Π is the tangent plane.
Definition 4.5.2. The straight line passing through a point of a surface, perpendicular to
the tangent plane to the surface at that point, is called the normal to the surface at the
considered point.
126 Chapter 4. General theory of surfaces
Thus, if (U, r) is a parameterization of the surface around a point a = r(u
0
, v
0
) =
(x
0
, y
0
, z
0
) ∈ S , then the a directing vector of the normal will be r
·
u
(u
0
, v
0
) × r
·
v
(u
0
, v
0
),
which means that the equations of the normal at the point a will be given by:
X − x
0

y
·
u
(u
0
, v
0
) z
·
u
(u
0
, v
0
)
y
·
v
(u
0
, v
0
) z
·
v
(u
0
, v
0
)

=
Y − y
0

z
·
u
(u
0
, v
0
) x
·
u
(u
0
, v
0
)
z
·
v
(u
0
, v
0
) x
·
v
(u
0
, v
0
)

=
Z − z
0

x
·
u
(u
0
, v
0
) y
·
u
(u
0
, v
0
)
x
·
v
(u
0
, v
0
) y
·
v
(u
0
, v
0
)

.
(4.5.4)
To construct the tangent plane and the normal to a surface given in an implicit rep-
resentation, the following result is very useful.
Theorem 4.5.3. At the point (x
0
, y
0
, z
0
) of the surface given through the equation
F(x, y, z) = 0
the vector grad F
0
= ¦F
·
x
(x
0
, y
0
, z
0
), F
·
y
(x
0
, y
0
, z
0
), F
·
z
(x
0
, y
0
, z
0
)¦ is perpendicular to the
tangent plane to the surface at this point.
Proof. Let r(u, v) = (x(u, v), y(u, v), z(u, v)) be a local parameterization of the surface
around the point (x
0
, y
0
, z
0
) = r(u
0
, v
0
). Then we have the identity
F(x(u, v), y(u, v), z(u, v)) = 0,
whence, by differentiation, we get

¸
¸
¸
¸
¸
¸
0 = F
·
u
= F
·
x
x
·
u
+ F
·
y
y
·
u
+ F
·
z
z
·
u
≡ grad F r
·
u
0 = F
·
v
= F
·
x
x
·
v
+ F
·
y
y
·
v
+ F
·
z
z
·
v
≡ grad F r
·
v
,
i.e. grad F ⊥ L(r
·
u
, r
·
v
) ≡ T
(x
0
,y
0
,z
0
)
S .
Consequence 1. The equation of the tangent plane to the surface given by the implicit
equation F(x, y, z) = 0 at the point (x
0
, y
0
, z
0
) has the form
(X − x
0
)F
·
x
(x
0
, y
0
, z
0
) + (Y − y
0
)F
·
y
(x
0
, y
0
, z
0
) + (Z − z
0
)F
·
z
(x
0
, y
0
, z
0
) = 0,
while the equations of the normal to the surface at the same point are
X − x
0
F
·
x
(x
0
, y
0
, z
0
)
=
Y − y
0
F
·
y
(x
0
, y
0
, z
0
)
=
Z − z
0
F
·
z
(x
0
, y
0
, z
0
.
Consequence 2. For any point a of the sphere S
2
R
the tangent space T
a
S
2
R
is orthogonal
to the radius vector of the point a.
4.6. The orientation of surfaces 127
Proof. The sphere S
2
R
can be described by the equation
F(x, y, z) ≡ x
2
+ y
2
+ z
2
− R
2
= 0,
whence it follows
grad F = 2¦x, y, z¦ = 2a,
therefore the radius vector is parallel to the gradient of the function F, hence it is per-
pendicular to the tangent space.
4.6 The orientation of surfaces
Definition 4.6.1. An orientation of a surface S is a choice of an orientation in each
tangent space T
a
S , i.e. a choice of the unit normal vector of T
a
S , n(a). It is assumed,
in this context, that the map n : S → R
3
, a → n(a) is continuous. The surfaces on
which is is possible to define an orientation are called orientable, while those on which
an orientation has been already chosen – oriented.
Examples. a) We can define an orientation on the sphere S
2
R
using the versor of the
exterior normal. It is not difficult to see that, if a is the radius point of the point
a ∈ S
2
R
, then n(a) =
1
R
a. Therefore, the map
n : S
2
R
→ R
3
,
defining the orientation of the sphere can be represented as a composition of contin-
uous maps:
S
2
R
i
− → R
3
1
R
−→ R
3
: a −→ a −→
1
R
a.
b) Let S be a simple surface, with the global parameterization (U, r). This surface can
be oriented by using the vector field
n(u, v) =
r
·
u
× r
·
v
|r
·
u
× r
·
v
|
.
c) Let S be a surface given by the implicit equation F(x, y, z) = 0. Then the surface can
be oriented through the gradient vector field:
n(x, y, z) =
grad F
| grad F|
.
128 Chapter 4. General theory of surfaces
If the orientation of a surface S is given by the vector function (vector field) n(a),
then the vector field −n(a) also defines an orientation on S , called the opposite orien-
tation of the orientation given by n. If the orientable surface S is connected, then each
orientation of S should coincide to one of the two orientations just mentioned. Indeed,
if N(a) is an orientation of the surface S , then we must have N(a) = λ n(a), where λ
is a continuous function on S , taking values into the finite set ¦−1, 1¦, therefore, if S
is connected, λ has to be a constant function. Thus, a connected orientable surface has
only two distinct orientations. Of course, if the surface is not connected, there are more
orientations, corresponding to different combinations of the two possible orientations on
each connected component of the surface.
Remark. Not any surface is orientable. We consider, for instance, the support of the
parameterized surface
r(u, v) = (cos u + v cos
u
2
cos u, sin u + v cos
u
2
sin u, v sin
u
2
),
with u, v ∈ R (the Möbius’s band, see the next figure). It will be shown bellow that S
is not orientable. (It has a single side: it is possible to move continuously the origin of
the unit normal along a close path on S such that, after the “trip”, the unit normal will
change into its opposite.) Notice that S is not simple, as it might seem, because r is not
a parameterization, since it is not a homeomorphism on the image.
The non-orientability of the Möbius’s band. We consider two local parameteriza-
tions to describe the Möbius’s band:
r : T =

(s, t) | −
1
2
< s <
1
2
, 0 < t < 2π

,
r(s, t) =

cos t

1 + s cos
t
2

, sin t

1 + s cos
t
2

, s sin
t
2

and
ρ : V =

(u, v) | −
1
2
< u <
1
2
, −π < v < π

,
ρ(u, v) =

cos v

1 + u cos
v
2

, sin v

1 + u cos
v
2

, u sin
v
2

It is not difficult to see that the domain of the diffeomorphism (parameters’ change)
Φ = ρ
−1
◦ r is the set T

, equal to T, with the segment t = π removed. We can write Φ
explicitly as Φ(s, t) ≡ (u, v) = (ϕ
1
(s, t), ϕ
2
(s, t)), where
u = ϕ
1
(s, t) = s, ∀(s, t) ∈ T

4.6. The orientation of surfaces 129
and
v = ϕ
2
(s, t) =

¸
¸
¸
¸
¸
¸
t if (s, t) ∈ T

, 0 < t < π
π − t if (s, t) ∈ T

, π < t < 2π
.
The Jacobi matrix of the map Φ is, as one can readily see,
J(Φ)(s, t) ≡

¸
¸
¸
¸
¸
¸
¸
∂u
∂s
∂u
∂t
∂u
∂s
∂v
∂t

¸
¸
¸
¸
¸
¸
¸
=

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸

¸
¸
¸
¸
¸
¸
1 0
0 1

¸
¸
¸
¸
¸
¸
if 0 < t < π

¸
¸
¸
¸
¸
¸
1 0
0 −1

¸
¸
¸
¸
¸
¸
if π < t < 2π
The supports of r and ρ are, clearly, orientable, as simple surfaces. However, admitting
that the Möbius’s band is orientable, the two local parameterizations do not define the
same orientation on it, since, as one can see from the computation we made before, the
determinant of the Jacobi matrix of the parameters’ change is not always positive.
On the other hand, as our surface is connected, if it is orientable, it can have only two
distinct orientations, in other words, any local parameterization of S should be positively
equivalent either to r or to ρ.
Let us suppose, now, that S is orientable. This means that there is a family of local
parameterizations r
1
, r
2
, . . . , which are pairwise positively equivalent and their supports
cover S . We may assume, without restricting the generality, that on the support of r all
this parameterizations are positively equivalent to r. There should be a local parameter-
Figure 4.2: The Möbius’s band
ization in the family such that its support intersects the segment t = 0. We shall suppose
130 Chapter 4. General theory of surfaces
that this parameterization is r
1
. We may assume that the support of r
1
is included into
the support of ρ (otherwise, if necessary, we may shrink the domain of r
1
). It follows
then that the Jacobi determinant of the map ρ
−1
◦ r
1
should be either always positive or
always negative on the domain of r
1
. On the other hand, we have, obviously, from the
chain rule, that
J(ρ
−1
◦ r) = J(ρ
−1
◦ r
1
)J(r
−1
1
◦ r),
whence
det J(ρ
−1
◦ r) = det J(ρ
−1
◦ r
1
) det J(r
−1
1
◦ r).
Now, in the right hand side, the last determinant is always positive, since the two para-
meterizations are assumed to be positively equivalent. The first determinant, from the
hypothesis, is always positive or always negative. Thus, the right hand side has constant
sign. On the other hand, as we saw previously, the left hand side has opposed signs
on the two sides of the segment t = 0, whence the contradiction which shows that the
Möbius’s band is not orientable.
In the figure 4.6 we indicate how one can construct a Möbius band, from a strip of
paper. Another example of a non-orientable surface is the so-called Klein’s bottle (see
figure ??)
Definition. Let S be an oriented surface with the orientation n(a). A local parameter-
ization (U, r) of S is said to be compatible with the orientation n(a) if for any point
a = r(u, v) we have
n(a) =
r
·
u
× r
·
v
|r
·
u
× r
·
v
|
or, which is the same, if the frame ¦r
·
u
, r
·
v
, n(a)¦ is right-handed.
4.7 Differentiable maps on a surface
Definition 4.7.1. Let S be a surface in R
3
. A map f : S → R
k
is called differentiable or
smooth if for any parameterization (U, r) of S the map f ◦ r : U → R
k
is smooth. The
map f
r
≡ f ◦ r is called the expression of f în in the curvilinear coordinates (u, v) or the
local representation of f with respect to the parameterization (U, r).
Remarks. 1. On can define similarly the differentiability of maps defined on any
open subset of a surface S .
2. Any differentiable map f : S → R
k
is continuous, since, locally, it can be written
as a composition of continuous maps: f = f ◦ (r ◦ r
−1
) = ( f ◦ r) ◦ r
−1
= f
r
◦ r
−1
.
4.7. Differentiable maps on a surface 131
Figure 4.3: Construct your own Möbius band!
Examples. a) Any constant map f : S → R
k
: a → A
0
, A
0
∈ R
k
is smooth, because
its local representation with respect to any local parameterization of S is, equally, a
constant map, hence it is differentiable.
b) If F : R
3
→ R
k
is a smooth map, then, for any surface S the map f = F|
S
: S →
R
k
is a smooth map. Indeed, for any local parameterization (U, r) of S the local
representation of f is f
r
= F ◦ r, where F and r are smooth maps, in the usual sense.
In particular, the orthogonal projections of a surface S on the coordinate axes and
planes are, all of them, smooth maps.
c) The inclusion i : S → R
3
: a → a is smooth, since for any local parameterization
(U, r) of S , the local representation of i is i
r
= i ◦r = r. (In fact, i is just the restriction
of the identity map, 1
R
3 and, therefore, we can also apply the previous example).
Apparently, it is quite difficult to check whether a map defined on a surface is smooth,
132 Chapter 4. General theory of surfaces
Figure 4.4: Klein’s bottle
because we have to check the smoothness of its local representations with respect to all
the local parameterizations of the surface, which are, of course, infinitely many. For-
tunately, as the following theorem shows, it is enough to take only some of the local
parameterizations, such that their domains cover the surface. In particular, if the surface
is simple, it is enough to check for the global parameterization.
Theorem 4.7.1. A map f : S → R
k
is smooth iff for any point a ∈ S there is a local
parameterization (U, r) of the surface S with a ∈ r(U), such that the local representation
f
r
= f ◦ r : U → R
k
is smooth.
Proof. The necessity is obvious, since, if F is smooth, then it local representation f
r
is
smooth for any local parameterization (U, r) of S .
Conversely, let’s suppose that that a ∈ S is an arbitrary point of the surface and (U, r)
4.7. Differentiable maps on a surface 133
is a local parameterization of S around a such that the local representation f
r
= f ◦ r
is smooth. Obviously, it is enough to show that then the local representation of f in
any other parameterization of S around a is, also, smooth. We choose, thus, another
parameterization, (U
1
, r
1
), around a and let W = r(U) ◦ r
1
(U
1
). Then in r
−1
(W) ⊂ U
1
,
f
r
1
can be represented as
f
r
1
= f ◦ r
1
= f ◦ (r ◦ r
−1
) ◦ r
1
= ( f ◦ r) ◦ (r
−1
◦ r
1
) = f
r
◦ (r
−1
◦ r
1
).
Since both f
r
(from the hypothesis) and r
−1
◦ r
1
(from the theorem 4.3.1) are smooth, it
follows that f
r
1
is, equally smooth.
Example. Let S be a surface and (U, r) a local parameterization of S . As we explained
earlier, the map r
−1
: r(U) → R
2
is nor smooth in the classical sense. The reason is that
its domain is not an open set of an Euclidean space, so the notion itself doesn’t make
sense for it. We also showed that, however, locally, r
−1
is the restriction of a smooth map
defined on an open set of R
3
. The notion we just defined is, actually, the natural frame of
discussing this important map, which, in fact, assigns to each point of the surface (lying
in r(U), of course), a pair of coordinates. Indeed, as a map defined on an open set of
S , r
−1
is smooth, as we can see easily, because the local representation of r
−1
in the
parameterization (U, r) is

r
−1

r
≡ r
−1
◦ r = 1
U
.
The next natural step will be to define the notion of a smooth map between two
surface rather then from a surface to an Euclidean space. The idea is the following. Let
S
1
, S
2
be two surfaces In R
3
. Then any map F : S
1
→ S
2
can be regarded as a map
F : S
1
→ R
3
. More specifically, one can associate to F the map i ◦ F : S → R
3
, where
i : S
2
→ R
3
is the inclusion.
Definition 4.7.2. Let S
1
, S
2
⊂ R
3
be two surfaces. A map F : S
1
→ S
2
is called smooth
if the map F
1
= i ◦ F : S
1
→ R
3
is smooth.
Remarks. 1. It easy to see that any smooth between surfaces is continuous.
2. Let S
1
⊂ R
3
a surface and G : R
3
→ R
3
a diffeomorphism. Then S
2
= G(S
1
) is,
also, a surface, while the map G|
S
1
: S
1
→ S
2
is smooth.
3. Let S
1
, S
2
⊂ R
3
two surfaces and F : S
1
→ S
2
a map. Then F is smooth iff
for any a ∈ S
1
, any local parameterization (U
1
, r
1
) of S
1
around a and any local
parameterization (U
2
, r
2
) of S
2
around F(a), the map
F
r
1
,r
2
≡ r
−1
2
◦ F ◦ r
1
: U
1
→ U
2
134 Chapter 4. General theory of surfaces
is smooth (in the usual sense). F
r
1
,r
2
is called the local representation of F with
respect to the local parameterizations (U
1
, r
1
) and (U
2
, r
2
).
Definition 4.7.3. A map F : S
1
→ S
2
is called a diffeomorphism if F is bijective and
both F and F
−1
are smooth maps.
4.8 The differential of a smooth map between surfaces
The notion of a smooth map between surfaces is a natural generalization of the notion
of smooth map between open sets in Euclidean spaces. The same should hold true, of
course, also for the notion of differential. We shall emphasize, therefore, a property
of the differential of map between Euclidean spaces that will be used to construct the
generalization. Let G : B → R
3
be a smooth map, with B ⊂ R
3
– an open set, while
G(x, y, z) = (g
1
(x, y, z), g
2
(x, y, z), g
2
(x, y, z)). Then, for any point a = (x, y, z) ∈ B, the
differential of G at a,
d
a
G : R
3
a
→ R
3
G(a)
is a linear map, whose matrix is the Jacobi matrix
D(g
1
, g
2
, g
3
)
D(x, y, z)

(x
0
,y
0
,z
0
)
= (α
i j
), 1 ≤ i, j ≤ 3,
where
α
i j
= ∂
i
g
i
(x
0
, y
0
, z
0
).
For any vector h ∈ R
3
a
, of components ¦h
1
, h
2
, h
3
¦, the vector d
a
G(h) has the components

¸
¸
¸
¸
¸
¸
¸
¸
3
¸
j=1
α
1j
h
j
,
3
¸
j=1
α
2 j
h
j
,
3
¸
j=1
α
3 j
h
j
¸
¸
¸
¸
¸
¸
¸
¸
¸
.
Let us suppose now that the vector h is tangent to the parameterized curve ρ(t) =
(x(t), y(t), z(t)) at point t = t
0
, i.e. h = ρ

(t
0
). We are going to show that the vector
d
a
G(h) is the tangent vector to the parameterized curve (G ◦ ρ)(t) at t = t
0
. To this end,
we differentiate the relation
(G ◦ ρ)(t) = (g
1
(x(t), y(t), z(t)), g
2
(x(t), y(t), z(t)), g
3
(x(t), y(t), z(t)))
and we obtain:
(
−−−−→
G ◦ ρ)
·
(t
0
) =

¸
¸
¸
¸
¸
¸
3
¸
k=1
∂g
1
∂x
k
(x
0
, y
0
, z
0
)h
k
,
3
¸
k=1
∂g
2
∂x
k
(x
0
, y
0
, z
0
)h
k
,
3
¸
k=1
∂g
3
∂x
k
(x
0
, y
0
, z
0
)h
k
¸
¸
¸
¸
¸
¸
¸
=
=

¸
¸
¸
¸
¸
¸
3
¸
k=1
α
1k
h
k
,
3
¸
k=1
α
2k
h
k
,
3
¸
k=1
α
3k
h
k
¸
¸
¸
¸
¸
¸
¸
,
4.8. The differential of a smooth map between surfaces 135
where
¦h
1
, h
2
, h
3
¦ = ¦x
·
(t
0
), y
·
(t
0
), z
·
(t
0
)¦ = ρ

(t
0
).
Thus, the differential d
a
G assigns to a tangent vector to the path ρ(t) at t = t
0
the
tangent vector to the path G(ρ(t)) at t = t
0
.
Let now F : S
1
→ S
2
be a smooth map between the surfaces S
1
and S
2
and a ∈ S
1
.
Then to any smooth path (I, ρ) on S
1
corresponds a smooth path (I, F ◦ ρ) on S
2
. If ρ(t)
passes through a for t = t
0
, then the path F ◦ ρ(t) will pass through F(a) for t = t
0
.
Definition 4.8.1. The map T
a
S
1
→ T
F(a)
S
2
, assigning to each tangent vector ρ

(t
0
)
to a parameterized curve ρ(t) on S
1
, with ρ(t
0
) = a, the tangent vector (
−−−−→
F ◦ ρ)
·
(t
0
) to
the parameterized curve F ◦ ρ at t = t
0
is called the differential of the smooth map
F : S
1
→ S
2
at the point a and it is denoted by d
a
F.
Now, there might be a little difficulty here. Namely, we could have on S
1
two dif-
ferent curves, which have the same tangent vector at a contact point. As the images of
the two parameterized curves through F are, generally, distinct, it may happen that the
images do not have the same tangent vector at the contact point. Well, as we shall see in
a moment, this not actually the case. Indeed, let a ∈ S
1
and (I, ρ = ρ(t)), (I
1
, ρ
1
= ρ
1
(s))
– two parameterized curves on S
1
such that ρ(t
0
) = ρ
1
(s
0
) = a and ρ

(t
0
) = ρ

1
(s
0
). We
choose an arbitrary local parameterization (U, r) on S
1
, around a. As we are interested
only on the local phenomena that happen around a, we may assume, without restricting
the generality, that ρ(I) ⊂ r(U) and ρ
1
(I
1
) ⊂ r(U). Let’s suppose that the local equation
of the curves in the parameterization (U, r) are
(ρ)

¸
¸
¸
¸
¸
¸
u = u(t)
v = v(t)
,
and,

1
)

¸
¸
¸
¸
¸
¸
u = u
1
(s)
v = v
1
(s)
,
respectively. Then the vectors ρ

(t
0
) and ρ
1

(s
0
) have, in the natural basis ¦r
·
u
, r
·
v
¦ the
expressions
ρ

(t
0
) = ¦u
·
(t
0
), v
·
(t
0

ρ
1

(s
0
) = ¦u
·
1
(s
0
), v
·
1
(s
0
)¦.
Moreover, in the chosen parameterization,
(F ◦ ρ)(t) = F
r
(u(t), v(t))
(F ◦ ρ
1
)(s) = F
r
(u
1
(s), v
1
(s)),
136 Chapter 4. General theory of surfaces
with F
r
= F ◦ r. Thus,
(
−−−−→
F ◦ ρ)(t
0
) =
d
dt
(F
r
(u(t), v(t)))(u
0
, v
0
) =
−−−−→
(F
r
)
·
u
(u
0
, v
0
)u
·
(t
0
) +
−−−−→
(F
r
)
·
v
(u
0
, v
0
)v
·
(t
0
)
(
−−−−−→
F ◦ ρ
1
)(s
0
) =
d
ds
(F
r
(u(s), v(s)))(u
0
, v
0
) =
−−−−→
(F
r
)
·
u
(u
0
, v
0
)u
·
1
(s
0
) +
−−−−→
(F
r
)
·
v
(u
0
, v
0
)v
·
1
(s
0
).
Now, since ρ

(t
0
) = ρ
1

(s
0
), it follows that
(
−−−−→
F ◦ ρ)
·
(t
0
) = (
−−−−−→
F ◦ ρ
1
)
·
(s
0
),
i.e. the definition of the differential of F makes sense.
Theorem. The map dF : T
a
S
1
→ T
F(a)
S
2
is linear.
Proof. From the computations made above, it follows immediately that if a vector h ∈
T
a
S
1
has in the natural basis ¦r
·
u
, r
·
v
¦ of the linear space T
a
S
1
, the components ¦h
1
, h
2
¦,
then
d
a
F(h) =
−→
F
·
r
u
(u
0
, v
0
)h
1
+
−→
F
·
r
v
(u
0
, v
0
)h
2
, (4.8.1)
whence the linearity.
Example. Let S
1
, S
2
be two surfaces in R
3
, D : R
3
→ R
3
a diffeomorphism such that
D(S
1
) = S
2
, F = D
S
1
: S
1
→ S
2
, a ∈ S ⊂ R
3
. Then we have
d
a
F = d
a
D|
T
a
S
1
, (4.8.2)
where d
a
D : R
3
a
→ R
3
D(a)
is the differential of the map D at a. In particular, let S
2
R
and
S
2
r
be the spheres of radii R and r, respectively, centred at the origin and D : R
3

R
3
: (x, y, z) →
r
R
(x, y, z) – a homothety, which is, obviously, a diffeomorphism such that
D(S
2
R
) = S
2
r
. Then, for the map F = D|
S
2
R
: S
2
R
→ S
2
r
, we have d
a
F(h) =
r
R
h.
4.9 The spherical map and the shape operator of an oriented
surface
Let S ⊂ R
3
be an oriented surface and S
2
– the unit sphere centred at the origin. If
the orientation of S is given by the unit normal n(a), a ∈ S , then we can build a map
Γ : S → S
2
, associating to each a ∈ S the point on S
2
has as radius vector Γ(a) = n(a).
The map Γ is called the spherical map of the surface S . This map place a central role in
the theory of surfaces. We are going to show, first of all, that Γ is smooth:
4.9. The spherical map and the shape operator of an oriented surface 137
Theorem 4.9.1. The spherical map Γ : S → S
2
of an oriented surface S into the unit
sphere S
2
is a smooth map between surfaces.
Proof. Let a ∈ S . We choose a local parameterization (U, r) of the surface S around a,
compatible with the orientation. Clearly, since S is orientable, such a parameterization
always exists. Indeed, if we choose a parameterization (U
1
, r
1
) which is not compatible
with the orientation, i.e. we have
r
·
1u
× r
·
1v
|r
·
1u
× r
·
1v
|
= −n(u, v),
then we replace the domain U
1
by U

1
, the symmetric of U
1
with respect to the Ov-axis,
and the map r
1
(u, v) by r

1
(u, v) = r
1
(−u, v). It is easy to see that the pair (U

1
, r

1
) is a
parameterization of the surface, compatible with the orientation.
Now we have
(Γ ◦ r)(u, v) = Γ(r(u, v)) = Γ
r
(u, v) = n(u, v) =
r
·
u
× r
·
v
|r
·
u
× r
·
v
|
.
Thus, the local representation of Γ is smooth, therefore Γ itself is smooth.
Examples. (i) For a plane Π the spherical map is constant.
(ii) For the sphere S
2
R
the map Γ : S
2
R
→ S has the expression Γ(x, y, z) =
1
R
(x, y, z),
with x
2
+ y
2
+ z
2
= R
2
.
As we saw, the tangent space to the sphere, T
Γ(a)
S
2
, is orthogonal to the radius
vector n(a) of the point Γ(a). On the other hand, n(a) is orthogonal to T
a
S . Thus, if we
identify R
3
a
and R
3
Γ(a)
to R
3
, then the subspaces, T
a
S and T
Γ(a)
S
2
coincide. Therefore,
we may think of the differential d
a
Γ : T
a
S → T
Γ(a)
S
2
as being, in fact, a linear operator
T
a
S → T
a
S .
Definition 4.9.1. The linear operator d
a
Γ : T
a
S → T
a
S is called the shape operator of
the oriented surface S at the point a and it is denoted by A or A
a
.
Remark. There is no general agreement regarding the definition, nor the name of the
shape operator. In some books, the shape operator carries an extra minus sign. Some-
times it is called the Weingarten mapping, or, also the principal, or fundamental op-
erator. Historically, it is true, indeed, that Weingarten was the first to write down the
formulae for the differentiation of the spherical map (in other words, he was the one to
find the partial derivatives of the unit normal vectors in terms of the derivatives of the
138 Chapter 4. General theory of surfaces
radius vector). Nevertheless, the shape operator, as a linear map, was introduced in dif-
ferential geometry by the Italian mathematician Cesare Burali-Forti ([6]), in 1912, under
the name omografia fondamentale, i.e. fundamental homography.
Example. (i) For a plane, the shape operator vanishes.
(ii) For a sphere, the shape operator is a homothety.
Let now (U, r) be a local parameterization of S , with the orientation. Then the local
representation of the spherical map Γ of S will be given by
n(u, v) =
r
·
u
× r
·
v
|r
·
u
× r
·
v
|
,
therefore, for the shape operator we will have:
A(h) = n
·
u
h
1
+ n
·
v
h
2
, (4.9.1)
where h ∈ T
r(u,v)
S ; (h
1
, h
2
) are the components of h with respect to the natural basis
¦r
·
u
, r
·
v
¦. In particular, we have
A(r
·
u
) = n
·
u
, A(r
·
v
) = n
·
v
. (4.9.2)
Theorem. The shape operator A is self-adjoint, i.e. ∀h, p ∈ T
a
S , we have
A(h) p = h A(p). (4.9.3)
Proof. It is enough to make the proof for the vectors, r
·
u
and r
·
v
. The case h = p is trivial,
therefore it is enough to check that
A(r
·
u
) r
·
v
= r
·
u
A(r
·
v
),
i.e.
n
·
u
r
·
v
= r
·
u
n
·
v
. (4.9.4)
To prove this, we start from the obvious equalities:
r
·
v
n = 0 (*)
¸ si
r
·
u
n = 0 (**)
We differentiate (*) with respect to u and (**) with respect to v and we get

¸
¸
¸
¸
¸
¸
r
··
uv
n + r
·
v
n
·
u
= 0
r
··
uv
n + r
·
u
n
·
v
= 0
whence, subtracting the two equalities, we get the conclusion.
4.10. The first fundamental form of a surface 139
Consequence. In each tangent space T
a
S there is an orthonormal basis, made up of
eigenvectors of the shape operator A.
Proof. Since A is self-adjoint, the two eigenvalues λ
1
, λ
2
are real. If λ
1
λ
2
, then the
eigenvectors corresponding to λ
1
and λ
2
are orthogonal, hence it is enough to choose a
unit vector in each eigenspace. If λ
1
= λ
2
, then A is just a homothety and any orthonor-
mal basis will do (since, in this case, any tangent space is an eigenvector of A).
4.10 The first fundamental form of a surface
Let S be a surface in R
3
. Then the scalar product in R
3
induces a scalar product in each
R
3
a
and, hence, it also induces a scalar product in each tangent space T
a
S , a ∈ S .
Definition 4.10.1. The first fundamental form of a surface S is, by definition, the func-
tion ϕ
1
, associating to each a ∈ S the restriction of the scalar product of R
3
a
to T
a
S . We
shall say usually, informally, that the first fundamental form is the restriction itself, but
it should be kept in mind, however, what really happens. Thus, for any a ∈ S and any
p, q ∈ T
a
S , we will have
ϕ
1
(p, q) = p q. (4.10.1)
Remark. In many textbooks, especially the older ones, the first fundamental form is not
defined as the restriction of the scalar product to T
a
S , but rather as the quadratic form
associated to this restriction.
If (U, r) is a local parameterization of S , then for any (u, v) ∈ U the tangent space
R
2
(u,v)
of the domain U at the point (u, v) can be identified with the space T
r(u,v)
S , asso-
ciating to the vectors ¦1, 0¦, ¦0, 1¦, making up a basis of R
2
(u,v)
, the vectors r
·
u
(u, v) and
r
·
v
(u, v). It is easy to see that, in fact, this identification is just the linear isomorphism
dr
(u,v)
: R
2
(u,v)
→ T
r(u,v)
S . Using this identification, one can transport the first funda-
mental form ϕ
1
of S to the domain S (which can be seen, as a matter of fact, as being a
simple surface, with the global parameterization given by the identical map of U). Thus,
for any (u, v) ∈ U, in the tangent space T
r(u,v)
U ≡ R
2
(u,v)
at the domain U, the scalar
product of two vectors is defined by the rule
¯ϕ
1
(ξ, η) = ϕ
1
(dr
(u,v)
(ξ), dr
(u,v)
(η)) = dr
(u,v)
(ξ) dr
(u,v)
(η).
It easy to see that, by construction, the map dr
(u,v)
: R
2
(u,v)
→ T
r(u,v)
S is isometrical
140 Chapter 4. General theory of surfaces
with respect to the scalar products ¯ϕ
1
and ϕ
1
respectively. We introduce the notations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
E(u, v) = r
·
u
(u, v) r
·
u
(u, v)
F(u, v) = r
·
u
(u, v) r
·
v
(u, v)
G(u, v) = r
·
v
(u, v) r
·
v
(u, v)
.
Then the functions E, F, G are smooth on U, while the matrix G =

E F
F G

is the ma-
trix of the scalar product ϕ
1
on the tangent space T
r(u,v)
S with respect to the basis
¦r
·
u
(u, v), r
·
v
(u, v)¦, but it is also the matrix of the scalar product ¯ϕ
1
on the the tangent
space R
2
(u,v)
= T
(u,v)
U with respect to the basis ¦¦1, 0¦, ¦0, 1¦¦.
Examples. 1. For the plane Π given by the global parameterization r = r
0
+ua +vb,
a × b 0, we have:

¸
¸
¸
¸
¸
¸
r
·
u
= a
r
·
v
= b
hence

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
E = a
2
F = a b
G = b
2
.
If Π is the coordinate plane xOy, then we may set r
0
= 0, a = ı, b = , therefore
the first fundamental form has the matrix G =

1 0
0 1

.
2. For the sphere S
2
R
, we choose the local parameterization (U, r), with
r(u, v) = (Rcos u cos v, Rcos u sin v, Rsin u),
and U =


π
2
,
π
2

×


π
2
,
π
2

. We get immediately E = R
2
, F = 0, G = R
2
cos
2
u,
hence the matrix of the first fundamental form is given by
G = R
2
¸
1 0
0 cos
2
u

4.10.1 First applications
The length of a segment of curve on a surface
Let S be a surface, (U, r) – a local parameterization of S and (I, ρ) – a parameterized
curve with ρ(I) ⊆ r(U), given by the local equations u = u(t), v = v(t). Then, in the
natural basis, the tangent vector of ρ, ρ

(t) has the components ¦u
·
(t), v
·
(t)¦ and we can
4.10. The first fundamental form of a surface 141
compute its length using the matrix G. Therefore, we have, for the length of the segment
of ρ between t
1
and t
2
:
l
t
1
,t
2
=
t
2

t
1

(t)|dt =
t
2

t
1

E(t)u
·2
+ 2F(t)u
·
v
·
+ G(t)v
·2
dt,
where

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
E(t) = E(u(t), v(t))
F(t) = F(u(t), v(t))
G(t) = G(u(t), v(t))
.
Example. We take, on the sphere S
2
R
, the curve given by the local equations (in the
parameterization described in the previous example) u = 0, v = t, where t ∈ (0, 2π)
(the equator with a point removed). As we saw above, the first fundamental form of the
sphere has the matrix G = R
2

1 0
0 cos
2
u

. As along the curve we have u = 0, it follows
that u
·
(t) = 0, v
·
(t) = 1. On the other hand, cos
2
u = cos
2
0 = 1, hence, along the curve,
the matrix G will be the identity matrix, multiplied by R
2
. If we want to compute, for
instance, the length of the segment of curve between t
1
=
π
2
and t
2
= π we will get
l
π
2

=
π

π
2

R
2
0 + 2 0 + R
2
1dt = R t

π
π
2
=
πR
2
,
which was to be expected (the segment of curve is a quarter of a great circle on the
sphere).
The angle of two curves on a surface
Let (U, r) be a parameterization of a surface S , (I, ρ = ρ(t)), (I
1
, ρ
1
= ρ
1
(s)) – two
curves on S such that ρ(I) ⊂ r(U), ρ
1
(I
1
) ⊂ r(U). We assume that the supports of the
two curves intersect at r(u
0
, v
0
), i.e. there are t
0
∈ I, s
0
∈ I
1
such that:
ρ(t
0
) = ρ
1
(s
0
) = r(u
0
, v
0
).
If the local equations of the two curves are
(ρ)

¸
¸
¸
¸
¸
¸
u = u
1
(t)
v = v
1
(t)
,
142 Chapter 4. General theory of surfaces
and

1
)

¸
¸
¸
¸
¸
¸
u = u
2
(s)
v = v
2
(s)
,
respectively, then the decomposition of the tangents vector at the intersection point with
respect to the natural basis will be:

¸
¸
¸
¸
¸
¸
ρ

(t
0
) = ¦u
·
1
(t
0
), v
·
1
(t
0

ρ
1

(s
0
) = ¦u
·
2
(s
0
), v
·
2
(s
0

,
therefore the cosine of the angle of the curves
3
at the contact point is, as it is well known,
cos θ =
ρ

(t
0
) ρ
1

(s
0
)

(t
0
)| |ρ
1

(s
0
)|
=
Eu
·
1
u
·
2
+ F(u
·
1
v
·
2
+ u
·
2
v
·
1
) + Gv
·
1
v
·
2

Eu
·
1
2
+ 2Fu
·
1
v
·
1
+ Gv
·
1
2

Eu
·
2
2
+ 2Fu
·
2
v
·
2
+ Gv
·
2
2
,
where

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
E = E(u
0
, v
0
)
F = F(u
0
, v
0
)
G = G(u
0
, v
0
)
and

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
u
·
1
= u
·
1
(t
0
)
v
·
1
= v
·
1
(t
0
)
u
·
2
= u
·
2
(s
0
)
v
·
2
= v
·
2
(s
0
)
.
The area of a parameterized surface
Let (U, r) be an oriented parameterized surface. There are many ways of introducing
the notion of area. All of them are more or less connected to integral calculus, so we
are not going to enter into any details here. Basically, as in the case of plane geomet-
ric figures, the area should be a function associating to each oriented patch a positive
number, subject to some restrictions. We choose, following Stoker, the following three
restrictions:
a) The area should be given by an integral of the form
A =

U
f dudv,
where f should depend only upon u, v, r, r
·
u
, r
·
v
(no higher derivatives of r should be
involved!).
3
We are speaking, of course, about the angle of the tangent vectors.
4.10. The first fundamental form of a surface 143
b) It is invariant with respect to rigid motions of the space and to parameter transforma-
tions that preserve the orientation.
c) A square of side length 1 has area 1.
It can be proved that the only formula for area that verifies these three axioms is
A =

U

EG − F
2
dudv ≡

U
|r
·
u
× r
·
v
|dudv. (4.10.2)
We will give a heuristic motivation for the formula (4.10.2). It should not be taken,
however, as being a “proof”, because no such claim is being currently made.
The “classical” approach. Let (U, r) be a parameterized surface and D ⊂ U – a com-
pact subset of U such that r(∂U) is a piecewise smooth curve in R
3
. We want to define
the area of r(D) ⊂ r(U). The basic idea is that we already have a notion of area for
plane figures, in particular for parallelograms. Thus, let (u, v) ∈ D and M = r(u, v).
Through M pass two coordinate lines, one from each family. Let M
1
= r(u + ∆u, v),
M
2
= r(u, v + ∆v) be two points on these lines, shifted from M with the parameter shifts
∆u and ∆v, respectively, and M
·
= r(u+∆u, v+∆v). If ∆u and ∆v are small enough, then
the projection of the curvilinear parallelogram MM
1
M
·
M
2
on the tangent plane to the
surface at the point M is (approximately, of course), a plane parallelogram in the tangent
plane. The sides of this parallelogram are r
·
u
∆u and r
·
v
∆v and its area will be, then
∆σ = |r
·
u
∆u × r
·
v
∆v| = |r
·
u
× r
·
v
|∆u∆v =

EG − F
2
∆u∆v,
where, of course, the coefficients of the first fundamental form have been computed at
the point of M. It is only natural, then, to define the area of r(D) as
A = lim
∆u→0
∆v→0
Σ∆σ =

D

EG − F
2
dudv,
where the sum in the middle term is taken after all the small curvilinear parallelograms
that cover r(U).
Remark. We might expect to obtain for the area of domain on a surface an interpretation
similar to the one we got for the length of a segment of curve. Namely, we may discretize
the domain of the parameterization and consider the images of the selected points. They
will determine a polygonal surface inscribed into the surface. Then we may consider that
the area of the polygons that make up this polygonal surface goes to zero and define the
144 Chapter 4. General theory of surfaces
area of our surface as being the limit of the area of the polygonal surface. Unfortunately,
as a celebrated example of H.A. Schwartz shows, it just doesn’t work, because the limit
is not independent on the type of polygons we consider and, in particular, for some
“polygonizations” of the surface, the area may be infinite and for other ones finite. Of
course, the things can be fixed, with a little care, but this is a subject that belongs to
theory of integration rather than to differential geometry, therefore we shall not insist.
4.11 The matrix of the shape operator of a surface in the nat-
ural basis
Let (U, r) be a local parameterization of the oriented surface S , compatible with the
orientation. We shall denote by 7the matrix of the shape operator A with respect to the
natural basis ¦r
·
u
, r
·
v
¦. Since, as we saw earlier,
A(r
·
u
) = n
·
u
, A(r
·
v
) = n
·
v
,
we have
( n
u
n
·
v
) = (r
·
u
r
·
v
) 7. (4.11.1)
We multiply from the left with the matrix

r
·
u
r
·
v

and we get:
¸
r
·
u
r
·
v

n
·
u
n
·
v

=
¸
r
·
u
n
·
u
r
·
u
n
·
v
r
·
v
n
·
u
r
·
v
n
·
v

=
=
¸
r
·
u
r
·
v

r
·
u
r
·
v

7 =
¸
r
·
u
r
·
u
r
·
u
r
·
v
r
·
v
r
·
u
r
·
v
r
·
v

7 = G 7.
We introduce the functions

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
L(u, v) = r
·
u
n
·
u
M(u, v) = r
·
u
n
·
v
N(u, v) = r
·
v
n
·
v
, (4.11.2)
and the matrix 1, defined by
1 =
¸
L M
M N

Then the last equality reads
1 = G 7.
4.11. The matrix of the shape operator 145
Since the scalar product on R
3
is nondegenerate, the same is true for its restriction to any
subspace and, as a consequence, the matrix G is invertible. If G
−1
be its inverse, then for
the matrix of the shape operator we get
7 = G
−1
1, (4.11.3)
where, as one can see very easily,
G
−1
=
1
EG − F
2
¸
G −F
−F E

.
If we perform the computation, we obtain:
7 =
1
EG − F
2
¸
GL − FM GM − FN
−FL + EM −FM + EN

(4.11.4)
All we have to do now is to express the quantities L, M, N in terms of the derivatives of
the function r. To this end, we differentiate the relations r
·
u
n = 0 and r
·
v
n = 0 with
respect to u and v and we get:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
··
u
2
n + r
·
u
n
·
u
= 0
r
··
uv
n + r
·
u
n
·
v
= 0
r
··
uv
n + r
·
v
n
·
u
= 0
r
··
v
2
n + r
·
v
n
·
v
= 0
,
whence we obtain from L, M, N the expressions:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
L = r
·
u
n
·
u
= −n r
··
u
2
M = r
·
u
n
·
v
= −n r
··
uv
N = r
·
v
n
·
v
= −n r
··
v
2
(4.11.5)
or, having in mind that
n =
r
·
u
× r
·
v
|r
·
u
× r
·
v
|
,
while
|r
·
u
× r
·
v
| = H(=

EG − F
2
),

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
L = −
1
H
(r
·
u
, r
·
v
, r
··
u
2
)
M = −
1
H
(r
·
u
, r
·
v
, r
··
uv
)
N = −
1
H
(r
·
u
, r
·
v
, r
··
v
2
)
. (4.11.6)
146 Chapter 4. General theory of surfaces
Example. For the helicoid

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v
y = u sin v
z = bv
, (u, v) ∈ R
2
, b > 0,
we can define the orientation by putting
n(u, v) =

b sin v

b
2
+ u
2
, −
b cos v

b
2
+ u
2
,
u

b
2
+ u
2

.
On gets, after a straightforward computation:
G =
¸
1 0
0 b
2
+ u
2

, 1 =

¸
¸
¸
¸
¸
¸
0
b

b
2
+u
2
b

b
2
+u
2
0

¸
¸
¸
¸
¸
¸
, 7 =

¸
¸
¸
¸
¸
¸
0
b
(b
2
+u
2
)
3/2
b
(b
2
+u
2
)
3/2
0

¸
¸
¸
¸
¸
¸
.
4.12 The second fundamental form of an oriented surface
Definition 4.12.1. The second fundamental form of an oriented surface S is a map as-
sociating to each a ∈ S the application ϕ
2
(a) : T
a
S × T
a
S → R given by
ϕ
2
(ξ, η) = −ϕ
1
(A(ξ), η), ∀ξ, η ∈ T
a
S. (4.12.1)
Remark. The minus sign in the previous definition is a consequence of our particular
choice of sign in the definition of the shape operator. We found natural to choose the
shape operator to be the differential of the spherical map rather then the opposite of the
differential, but then in the definition of the second fundamental formwe had to introduce
an extra minus sign, in order to be consistent with the generally accepted definition of
the second fundamental form.
Proposition 4.12.1. For each a ∈ S , ϕ
2
(a) is a symmetrical bilinear form.
Proof. We take two arbitrary tangent vectors ξ, η ∈ T
a
S and two arbitrary real numbers
α, β ∈ R. Then we have, first of all:
ϕ
2
(η, ξ) = −ϕ
1
(A(η), ξ)
A
=
self-adjoint
−ϕ
1
(η, A(ξ))
ϕ
1
=
symmetrical
−ϕ
1
(A(ξ), η) = ϕ
2
(ξ, η),
which means that ϕ
2
is symmetrical. Due to the symmetry, it is enough to prove the
linearity in the first variable only. We have
ϕ
2
(αξ
1
+ βξ
2
, η) = −ϕ
1
(A(αξ
1
+ βξ
2
), η)
A
=
linear
−ϕ
1
(αA(ξ
1
) + βA(ξ
2
), η)
ϕ
1
=
bilinear
ϕ
1
=
bilinear
−αϕ
1
(A(ξ
1
), η) − βϕ
1
(A(ξ
2
), η) = αϕ
2

1
, η) + βϕ
2

2
, η),
4.12. The second fundamental form of an oriented surface 147
which shows the linearity in the first variable and concludes the proof.
Let (U, r) be a local parameterization of the oriented surface S , compatible with the
orientation. Then the matrix [ϕ
2
] of the second fundamental form with respect to the
canonical basis ¦r
·
u
, r
·
v
¦ has the form:
,
ϕ
2
¸
=
¸
ϕ
2
(r
·
u
, r
·
u
) ϕ
2
(r
·
u
, r
·
v
)
ϕ
2
(r
·
v
, r
·
u
) ϕ
2
(r
·
v
, r
·
v
)

.
But

¸
¸
¸
¸
¸
¸
ϕ
2
(r
·
u
, r
·
u
) = −ϕ
1
(A(r
·
u
), r
·
u
) = −ϕ
1
( n
·
u
, r
·
u
) = −n
·
u
r
·
u
ϕ
2
(r
·
u
, r
·
v
) = ϕ
2
(r
·
v
, r
·
u
) = −n
·
u
r
·
v
= −n
·
v
r
·
u
ϕ
2
(r
·
v
, r
·
v
) = −n
·
v
r
·
v
,
and, thus, we get the matrix
,
ϕ
2
¸
= −
¸
n
·
u
r
·
u
n
·
u
r
·
v
n
·
v
r
·
u
n
·
v
r
·
v

= −
¸
L M
M N

not.
=
¸
D D
·
D
·
D
··

.
It follows, this way, that the matrix of the second fundamental form in the canonical
basis is just −1. Thus, for the coefficients of the second fundamental form with respect
to the natural basis, we have

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
D = n r
··
u
2
D
·
= n r
··
uv
D
··
= n r
··
v
2
(4.12.2)
Remark. The reader should be careful with the notations for the coefficients of the second
fundamental form. For them there is, also, used the notation e, f , g. Also, in some books,
the letters L, M, N are used to denote the coefficients of the second fundamental form
themselves. The notations D, D
·
, D
··
are usually credited to Gauss (in Disquisitiones). It
should be noted, however, that for Gauss, the meaning of the symbols is a little different:
they are not the coefficients of the second fundamental form as we know it, but rather
these coefficients multiplied by

EG − F
2
.
Example. For the sphere S = S
2
R
, we have, as we saw before, n =
1
R
¦x, y, z¦, therefore,
as we mentioned earlier, the shape operator A is a homothety of ratio 1/R, i.e.
A(p) =
1
R
p, , ∀p ∈ T
a
S
2
R.
148 Chapter 4. General theory of surfaces
Thus,
ϕ
2
(p, q) = −ϕ
1
¸
1
R
p, q

= −
1
R
ϕ
1
(p, q) = −
1
R
p q.
therefore, for the sphere, the first two fundamental forms are proportional. Clearly,
the same is true for the plane, when the second fundamental form vanishes identically.
Surprisingly as it might seem, actually, it can be shown that these are the only two
surfaces with this property.
4.13 The normal curvature. The Meusnier’s theorem
Let S be an oriented surface and n – the unit normal. We take a regular parameterized
curve, ρ = ρ(t), lying on S .
Definition 4.13.1. The projection of the curvature vector k(t) of ρ ( as a signed scalar)
on n(ρ(t)) is called the normal curvature of the curve ρ(t) at t and it is denoted by k
n
(t).
Let θ(t) be the angle between the osculating plane of ρ(t) and n(ρ(t)). Then, clearly,
k
n
(t) = k(t) cos θ(t), (4.13.1)
where k(t) is the curvature of the curve ρ(t).
Examples. 1. The normal curvature of any plane curve is zero (in this case the angle
θ(t) is always
π
2
, therefore cos θ(t) ≡ 0).
2. If the support of a parameterized curve lies on a straight line, then its normal
curvature is always zero, no matter on what surface is the curve situated, because,
this time, the curvature of the curve vanishes identically.
Remark. The relation (4.13.1) has a simple geometrical interpretation (the Meusnier’s
theorem): the center of curvature of a curve ρ, lying on a surface S , is the orthogonal
projection on its osculating plane of the center of curvature of the normal section tangent
to ρ at that point.
The normal curvature of a curve on a surface can be expressed easily if we know the
first two fundamental forms of the surface. Indeed, we have:
Theorem 4.13.1. The normal curvature of a parameterized curve ρ(t), lying on an ori-
ented surface S , is given by the formula
k
n
(t) =
ϕ
2

(t), ρ

(t))
ϕ
1

(t), ρ

(t))
. (4.13.2)
4.13. The normal curvature. The Meusnier’s theorem 149
Proof. As is many times the case with parameterized curves, the proof is simpler in the
case of naturally parameterized curves. Since the curvature of any regular parameterized
curve is invariant to a parameter change, we can replace the curve ρ(t) by a n equivalent,
naturally parameterized curve, ρ
1
(s), the natural parameter being the arc length. The
curvature vector of the curve ρ
1
(s) will be ρ
1

(s). We choose a local parameterization
(U, r) of the surface S and we assume that ρ
1
(s) has in this parameterization the local
equations u = u(s), v = v(s), i.e. ρ
1
(s) = r(u(s), v(s)). Then
ρ
1

(s) = r
··
u
2
(u
·
)
2
+ 2r
··
uv
u
·
v
·
+ r
··
v
2
(v
·
)
2
+ r
·
u
u
··
+ r
·
v
v
··
.
Thus, for the normal curvature of the curve ρ
1
(s) we get the expression:
k
n
(s) = k(s) n(ρ
1
(s)) = ρ
1

(s) n(ρ
1
(s)) =
= r
··
u
2
n (u
·
)
2
+ 2r
··
uv
n u
·
v
·
+ r
··
v
2
n (v
·
)
2
=
= −L (u
·
)
2
− 2M u
·
v
·
− N (v
·
)
2
= ϕ
2

1

(s), ρ
1

(s)).
Now, we come back to the initial parameterized curve. We have
ρ

(s) = ρ
1

(s(t)) s
·
(t) where s
·
(t) ≡ |ρ

(t)|.
Thus,
ρ
1

(s(t)) =
ρ

(t)

(t)|
,
therefore,
k
n
(t) = ϕ
2
¸
ρ

(t)

(t)|
,
ρ

(t)

(t)|

=
1
ρ

(t) ρ

(t)
ϕ
2

(t), ρ

(t)) =
ϕ
2

(t), ρ

(t))
ϕ
1

(t), ρ

(t))
.

Consequence. If two curves on an oriented surface have a common point and they have
the same tangent line at this point, then the two curves have the same normal curvature
at the contact point.
Proof. Let p and q be the tangent vectors to the two curves at their common point. From
the hypothesis, p = αq, hence, from the theorem,
k
n
=
ϕ
2
(p, p)
ϕ
1
(p, p)
=
ϕ
2
(αq, αq)
ϕ
1
(αq, αq)
=
α
2
ϕ
2
(q, q)
α
2
ϕ
1
(q, q)
=
ϕ
2
(q, q)
ϕ
1
(q, q)

150 Chapter 4. General theory of surfaces
Remark. The previous consequence can be interpreted in another way. We take a family
of biregular parameterized curves on the surface, ¦ρ
α
(t)¦
α∈A
, passing through the same
point and having the same tangent line at the contact point. We denote by k
α
the cur-
vature of the curve ρ
α
and by θ
α
the angle between the normal to the surface and the
osculating plane of the curve ρ
α
. The consequence is equivalent with the affirmation
that the product k
n
= k
α
cos θ
α
does not depend on the choice of the curve from the fam-
ily. It makes sense, thus, to choose an arbitrary straight line in the tangent plane, passing
through the tangency plane, and to speak about the normal curvature of the surface in
the direction of this line or, in other words, we can define a map k
n
on the set of all
non-vanishing tangent vectors to the surface with real values, by putting
k
n
(h) =
ϕ
2
(h, h)
ϕ
1
(h, h)
. (4.13.3)
The quantity k
n
(h) is called the normal curvature of the surface in the direction of the
vector h (since, clearly, it depends only on the direction of the vector h, but not on its
length or sense.) Thus, the normal curvature of an oriented surface in the direction of
a vector h is the normal curvature of an arbitrary curve, lying on the surface, passing
through the origin of h and whose tangent line is parallel to h.
4.14 Asymptotic directions and asymptotic lines on a surface
We saw previously that the normal curvature of a surface at a given point and in a given
direction can be expressed in terms of the first two fundamental forms of the surface and
that, as stressed, although it is initially defined in terms of curves on surface, it actually
depends only on the direction of the tangent vector of the curve. It is interesting for us
to identify those directions for which the normal curvature vanish.
Definition 4.14.1. Let S be an oriented surface and p ∈ S . A non-vanishing vector
h ∈ T
p
S is said to have asymptotic direction if the normal curvature in its direction van-
ishes. Alternatively, based on the previous section, we can define a vector of asymptotic
direction as being one for which
ϕ
2
(h, h) = 0.
Accordingly, an asymptotic line or curve on a surface is a curve on the surface for which
all the tangent vectors have asymptotic direction.
Theorem 4.1. Let p ∈ S a point of an oriented surface. Then at this point we have
asymptotic directions if and only if the quadratic form associated to the second funda-
mental form of S at p is negatively semi-defined. If we choose a local parameterization
4.14. Asymptotic directions and asymptotic lines on a surface 151
(U, r) of S such that pr(u
0
, v
0
) for some (u
0
, v
0
) ∈ U, then this condition simply means
that
D(u
0
, v
0
) D
··
(u
0
, v
0
) − D
·2
(u
0
, v
0
) ≤ 0. (4.14.1)
Proof. Let h = ¦h
1
, h
2
¦ ∈ T
p
S be a non-vanishing tangent vector to the surface at the
point p. Then, with respect to the local parameterization chosen, h has asymptotic di-
rection if and only if
D(u
0
, v
0
)h
2
1
+ 2D
·
(u
0
, v
0
)h
1
h
2
+ D
··
(u
0
, v
0
)h
2
2
= 0.
As h 0, we may assume, for instance, that h
2
0. Then the previous equation can be
written as
D(u
0
, v
0
)
¸
h
1
h
2

2
+ 2D
·
(u
0
, v
0
)
h
1
h
2
+ D
··
(u
0
, v
0
) = 0.
This equation, obviously has real solutions if and only if discriminant vanishes, but this
is exactly the condition 4.14.1.
Remark. From the proof of the previous theorem it follows that when the second funda-
mental form is negatively defined we have two asymptotic directions, while at the points
when it is degenerate we have only one asymptotic direction (or two confounded ones).
From the definition of the normal curvature of a curve on a surface, it follows imme-
diately that
Proposition 4.14.1. Any straight line lying on a surface is an asymptotic line.
Proof. Indeed, the straight lines have zero curvature, therefore their normal curvature at
each point also vanishes.
The differential equation of the asymptotic lines on a surface can be obtained directly
from the definition.
Theorem 4.2. Let S be an oriented surface and ρ : I → S – curve on the surface.
We assume that there is a local parameterization of S , (U, r = r(u, v)) such that ρ(I) ⊂
r(U) and the local equations of the curve with respect to this parameterization are u =
u(t), v = v(t). Then ρ is an asymptotic line on S if and only if
D(u(t), v(t)) u
·2
(t) + 2 D
·
(u(t), v(t)) u
·
(t) v
·
(t) + D
··
(u(t), v(t)) v
·2
(t) = 0. (4.14.2)
Proof. The equation (4.14.2) is just that condition for the tangent vector of the curve
(which has, with respect to the natural basis of the tangent space, the components
¦u
·
(t), v
·
(t)¦) to have asymptotic direction.
152 Chapter 4. General theory of surfaces
Let us assume, now, that the curve ρ from the previous theorem is biregular, which
means, in particular, that is curvature is always strictly positive. From the definition of
the normal curvature it follows immediately that, if ν is the unit principal normal vector
of the curve, then the curve is anasymptotic line if and only if
ν(t) n(u(t), v(t)) = 0,
where n is the unit normal vector to the surface. This actually means that the principal
normal of the curve lies on the tangent plane of the surface, at each point of the curve.
Thus, we obtain the following characterization of the asymptotic lines:
Theorem 4.3. Let S be an oriented surface and ρ a biregular parameterized curve on
S . Then ρ is an asymptotic line if and only if at each point its osculating plane coincide
with the tangent plane to the surface at that point.
Another immediate result concerning the asymptotic lines is the following:
Proposition 4.14.2. Let S be an oriented surface and (U, r = r(u, v)) – a local parame-
terization of S . Then the coordinate lines u = const and v = const are asymptotic lines
on r(U) if and only if D = D
··
= 0.
Example. Let us consider the helicoid, given by the parameterization

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v,
y = u sin v
z = b v
where b is a constant. A straightforward computation leads to D = D
··
= 0, D
·
(u, v) =
−b/

b
2
+ u
2
. This means that, in this particular case, we have, at each point of the
surface, two asymptotic lines and these are nothing but the coordinate lines, u = const
and v = const. We notice that the helicoid is a ruled surface (see the last chapter) and at
each point, one of the coordinate lines is a straight line (or a line segment). This is, of
course, the line v = const (see the figure 4.14).
4.15 The classification of points on a surface
The first fundamental form of a surface is positively defined. The second one is not. This
is fortunate, as it allows us to give a classification of the points of the surface, according
to the sign of the second fundamental form or, more specifically, according to the sign
of its discriminant DD
··
− D
·2
.
4.15. The classification of points on a surface 153
Figure 4.5: Asymptotic lines on the helicoid
Definition 4.15.1. A point a ∈ S of an oriented surface is called
(i) elliptic if the second fundamental form is positively defined at a;
(ii) parabolic if the second fundamental form is zero, but at least one of the coefficients
is different from zero;
(iii) hyperbolic, if the second fundamental form is negatively defined at a;
(iv) flat, or planar, if all the coefficients of the second fundamental form vanish at a.
Figure 4.6: Parabolic points on a surface
It is not difficult to see that this definition do not depend on the choice of the local
parameterization. Let us discuss, now separately, what happens in each particular case
and, also, give some examples.
154 Chapter 4. General theory of surfaces
Elliptic points. At an elliptic point, the normal curvature has the same signs in all
directions
4
. Applying the Meusnier’s theorem, this means that the centers of curvature
of all normal sections lie on the same side of the surface. An example of surface that has
only elliptic points is the ellipsoid, given, for instance, by the parameterization
r(u, v) = (a cos u cos v, b sin u cos v, c sin v). (4.15.1)
At an elliptic point there is no real asymptotic direction, therefore no asymptotic line
passes through an elliptic point.
Figure 4.7: Elliptic points on a surface
Parabolic points. In this case the normal curvature does not change the sign, but there
is exactly one direction where it vanishes. This is, clearly, an asymptotic direction. Thus,
through a parabolic point of a surface passes only one asymptotic line. The cylinders and
cones (with the apex removed) have only parabolic points.
Hyperbolic points. In the case of hyperbolic points, it is possible for k
n
to change sign
and there are exactly two direction where it vanishes. Thus, through a hyperbolic point
of a surface pass two asymptotic lines. The hyperbolic points are also called saddle
points. Such points we can find, for instance, on a hyperbolic paraboloid.
There are, of course, many surfaces on which we can meet all three kind of points
(for instance on the torus).
4
It is not necessarily positive.
4.15. The classification of points on a surface 155
Figure 4.8: The monkey saddle
Figure 4.9: Hyperbolic points on a surface
Flat points. The shape of the surface around a flat point might be quite complicated
and it is difficult to study it. In fact, in many cases, when proving a theorem in surface
theory one explicitly assumes that the surface has no flat points. The surface from the
figure 4.15 (the monkey saddle has a flat point at the origin of coordinates.
156 Chapter 4. General theory of surfaces
4.16 Principal directions, principal curvatures, Gauss curva-
ture and mean curvature
Definition 4.16.1. The directions on the tangent plane to an oriented surface S at a point
a ∈ S , T
a
S , corresponding to the eigenvectors of the shape operators A are called the
principal directions of the surface at the point a.
Remark. At each point, an oriented surface either has two orthogonal principal directions
(if the eigenvalues of A are distinct), either all the directions are orthogonal (if the two
eigenvalues coincide).
Definition 4.16.2. A curve (Γ) on a surface S is called a principal line or a curvature
line if its tangent, at each point, has a principal direction.
Definition 4.16.3. A principal curvature of an oriented surface S at a point a ∈ S is the
normal curvature of S at a in a principal direction.
Proposition 4.16.1. The principal curvatures of a surface are the eigenvalues of the
shape operator, taken with opposite sign.
Proof. If e is an eigenvector of A, then A(e) = λ e, where λ is the eigenvalue corre-
sponding to e, therefore
k
n
(e) =
ϕ
2
(e, e)
ϕ
1
(e, e)
=
−A(e) e
e e
=
−λ e e
e e
= −λ.

Hereafter we shall denote by k
1
and k
2
the principal curvatures and we will always
assume that k
1
≥ k
2
.
Definition 4.16.4. An orthonormal basis ¦e
1
, e
2
¦ of the tangent space at a point of a
surface is called a basis of principal directions of the tangent space if the vectors of the
basis have principal directions.
Thus, the vectors of a basis of principal directions are subject to
A(e
i
) = −k
i
e
i
, i = 1, 2.
We shall fix now a point of the surface and ask the following question: find the
normal curvature in the direction of a vector e, such that ∠(e, e
1
) = θ.
4.16. Principal directions and curvatures 157
As the length of e is not important, we shall assume that e is a unit vector: |e| = 1.
Then e = e
1
cos θ + e
2
sin θ, therefore
k
n
(e) =
ϕ
2
(e, e)
ϕ
1
(e, e)
=
−A(e) e
e e
....
=1
= −A(e
1
cos θ + e
2
sin θ) (e
1
cos θ + e
2
sin θ) =
= (k
1
cos θ e
1
+ k
2
sin θ e
2
) (e
1
cos θ + e
2
sin θ) = k
1
cos
2
θ + k
2
sin
2
θ.
Thus, we obtained:
Theorem 4.16.1. Let S be an oriented surface. Then the normal curvature at a point of
the surface, in the direction of a vector e, is given by the Euler’s formula:
k
n
(e) = k
1
cos
2
θ + k
2
sin
2
θ, (4.16.1)
where k
1
and k
2
are the principal curvature of the surface, while θ = ∠(e, e
1
).
An immediate consequence of the Euler’s formula is:
Theorem 4.16.2. The principal curvatures of a surface at a point are extreme values
of the normal curvature of the surface in the direction of a vector e, when the vector e
rotates around the origin of the tangent space of the surface at that point.
Proof. From the Euler’s formula, we have
k
n
(e) = k
1
cos
2
θ + k
2
(1 − cos
2
θ) = k
2
+ (k
1
− k
2
) cos
2
θ.
It is clear now that the maximum value of the normal curvature is reached for θ = 0 (we
assumed that k
1
≥ k
2
!) and, in this case, we get k
n
= k
1
, while the minimal value – for
θ =
π
2
, obtaining k
n
= k
2
.
Definition 4.16.5. The quantities K
t
= k
1
k
2
and K
m
=
1
2
(k
1
+k
2
) are called, respectively,
the total (Gaussian) and mean curvature of the surface.
The total and mean curvatures of a surface can be computed easily if the matrix of
the shape operator in an arbitrary basis is known. Indeed, we have:
Proposition 4.16.2.
K
m
= −
1
2
Tr 7 (4.16.2)
K
t
= det 7. (4.16.3)
158 Chapter 4. General theory of surfaces
Proof. As it is well known from linear algebra, the determinant and the trace are invari-
ants of any linear operator, which means that they are the same in any basis, although
the matrix of the operator does change, generally, if we modify the basis. In a basis of
principal directions, since the eigenvalues of the shape operator are just the opposite of
the principal curvature, the matrix of the shape operator will be:
7 =
¸
−k
1
0
0 −k
2

therefore
det 7 = k
1
k
2
= K
t

1
2
Tr 7 = −
1
2
(−k
1
− k
2
) =
1
2
(k
1
+ k
2
) = K
m
.

Joachimstahl’s theorem
We will see below how can one find the lines of curvature of a surface by integrating a
differential equation. Anyway, in some special situations it is possible to find these lines
using other methods. For instance, sometimes, if we knowthe curvature lines of a surface
it is possible to find such lines on another surface. Such an instance is exemplified by
the following theorem, belonging to the German mathematician Joachimstahl.
Theorem. Let γ be a curve lying at the intersection of two regular oriented surfaces
S
1
and S
2
from R
3
. Let n
i
be the unit normals of the two surfaces (i = 1, 2). Let us
assume that S
1
and S
2
intersects under a constant angle, i.e. along the curve γ we have
n
1
n
2
= const.. Then γ is a curvature line on S
1
iff it is a curvature line on S
2
.
Proof. Let r = r(t) be a local parameterization of the curve γ. Then, since n
1
n
2
=
const, we have
0 =
d
dt
( n
1
n
2
) = n
·
1
n
2
+ n
1
n
·
2
.
If γ is a principal line on S
1
, then
n
·
1
= −k
1
r
·
,
where k
1
is one of the principal curvatures of the surface S
1
. On the other hand, since
the curve γ lies also on S
2
, we have r
·
⊥ n
2
. From here and from the previous formula
we obtain that n
·
1
n
2
= 0, therefore
n
1
n
·
2
= 0.
4.16. Principal directions and curvatures 159
Since n
·
2
⊥ n
2
(since n
2
has constant length), it follows from here and from the previous
equation that n
·
2
⊥ r
·
or, in other words, that there is a k
2
∈ R such that
n
·
2
= −k
2
r
·
,
i.e. γ is a line of curvature on the surface S
2
, too.
Consequence. The meridians and the parallels on a revolution surface are lines of cur-
vature.
Proof. Let γ be the rotating curve and S the resulted revolution surface. A meridian
is obtained by intersecting a plane Π
m
, passing through the rotation axis and S . If p ∈
Π
m
∩ S , then the unit normal of the surface S , n(p), lies on the plane Π
m
, therefore the
normal of the surface and the normal of the plane Π
m
make up a constant angle, equal
to
π
2
. As for the plane the second fundamental form vanishes identically, the same is
true for the shape operator, which means that all the plane curves are lines of curvature.
It follows, that, in particular, the meridian is, also, a line of curvature in the plane Π
m
,
which implies, using the Joachimstahl theorem, that it is a line of curvature also for the
surface S .
A parallel is an intersection curve between the surface S and a plane Π
p
, passing
through a point of the curve γ, perpendicular to the rotation axis. It is obvious, from
symmetry reasons, that along a parallel we should have ∠( n, Π
p
) = const and we apply
again the reasoning we used before.
4.16.1 The determination of the lines of curvature
As we saw earlier, the lines of curvature are curves on a surfaces whose tangent vectors
are eigenvectors of the shape operator. Therefore, before showing how one can find the
lines of curvature, we will indicate a way to find the eigenvectors of the shape operator.
Lemma. Let r : U → R
3
be a local parameterization of an oriented surface S . A
tangent vector v = v
1
r
·
u
+ v
2
r
·
v
has principal direction iff

v
2
2
−v
1
v
2
v
2
1
E F G
D D
·
D
··

= 0. (4.16.4)
Proof. Since v has principal direction iff it is an eigenvector of the shape operator A, i.e.
A(v) = λ v, it follows that v has principal direction iff A(v) × v = 0. But, from the
160 Chapter 4. General theory of surfaces
definition,
A(v) = 7 v =

G
−1
1

v =
1
H
2
¸
GL − FM GM − FN
−FL + EM −FM + EN
¸
v
1
v
2

=
=
1
H
2
¸
(GL − FM)v
1
+ (GM − FN)v
2
(−FL + EM)v
1
+ (−FM + EN)v
2

or
A(v) =
1
H
2
α
u
....
[(GL − FM)v
1
+ (GM − FN)v
2
] r
·
u
+
1
H
2
[(−FL + EM)v
1
+ (−FM + EN)v
2
]
....
α
v
r
·
v
.
Therefore,
A(v)×v = 0 ⇐⇒ (α
u
r
·
u

v
r
·
v
)×(v
1
r
·
u
+v
2
r
·
v
) = 0 ⇐⇒ (α
u
v
2
−α
v
v
1
)(r
·
u
×r
·
v
) = 0.
As r
·
u
× r
·
v
0, since the surface is regular, it follows that
A(v) × v = 0 ⇐⇒ α
u
v
2
− α
v
v
1
= 0
or, having in mind the notations we made,
(FL − EM)v
2
1
+ (GL − EN)v
1
v
2
+ (GM − FN)v
2
2
= 0.
Using now the coefficients of the second fundamental form instead of the components
of the matrix 1, the previous relation becomes
(ED
·
− FD)v
2
1
+ (ED
··
− GD)v
1
v
2
+ (FD
··
− GD
·
)v
2
2
= 0. (4.16.5)
which is just another form of the relation (4.16.4)
Consequence (the differential equation of the lines of curvature). Let γ be a curve lying
in the domain r(U) of a local parameterization(r, U) of a surface S , with the local
equation ρ(t) = r(u(t), v(t)). Then γ is a line of curvature on S iff
(ED
·
− FD)u
·2
(t) + (ED
··
− GD)u
·
(t)v
·
(t) + (FD
··
− GD
·
)v
·2
(t) = 0 (4.16.6)
or
(ED
·
− FD) + (ED
··
− GD)
dv
du
+ (FD
··
− GD
·
)
¸
dv
du

2
= 0. (4.16.7)
Proof. The relation (4.16.6) express, clearly, the condition for the vector ρ

= u
·
(t)r
·
u
+
v
·
(t)r
·
v
to have principal direction, while (4.16.7) follows immediately from (4.16.6), by
eliminating the parameter t.
4.16. Principal directions and curvatures 161
4.16.2 The computation of the curvatures of a surface
Theorem. Let r : U → R
3
be a local parameterization of an oriented surface S . Then
the total and the mean curvatures of S are given, respectively, by the formulas:
K
t
=
DD
··
− D
·2
H
2
(4.16.1)
K
m
=
DG − 2D
·
F + D
··
E
2H
2
. (4.16.2)
Proof. As we saw earlier, the matrix of the shape operator of a surface is given by
7 =
1
H
2
¸
GL − FM GM − FN
−FL + EM −FM + EN

or 7 =
1
H
2
¸
FD
·
− GD FD
··
− GD
·
FD − ED
·
FD
·
− ED
··

,
therefore,
K
t
= det 7 =
1
H
4
,
F
2
D
·2
− EFD
·
D
··
− FGDD
·
+ EGDD
··
− F
2
DD
··
+ EFD
·
D
··
+
+FGDD
·
− EGD
·2
¸
=
1
H
4
,
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
(EG − F
2
)
....
H
2
(DD
··
− D
·2
)
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
=
DD
··
− D
·2
H
2
,
K
m
= −
1
2
Tr7 = −
1
2H
2

2FD
·
− GD − ED
··

=
DG − 2D
·
F + D
··
E
2H
2
.

Corollary. The principal curvatures k
1
and k
2
are the roots of the equation
k
2
− 2K
m
k + K
t
= 0, (4.16.3)
i.e.
k
1
= K
m
+

K
2
m
− K
t
, (4.16.4)
k
2
= K
m

K
2
m
− K
t
. (4.16.5)
Corollary. A non-flat point of a surface is
1. elliptic iff K
t
> 0;
2. parabolic iff K
t
= 0;
3. hyperbolic iff K
t
< 0.
162 Chapter 4. General theory of surfaces
4.17 The fundamental equations of a surface
4.17.1 Introduction
We saw that, in the case of curves, the curvature and the torsion completely determine
a space curve (up to a rigid motion of the space). We may ask whether a similar result
holds for the case of surfaces. It is not completely obvious what are the entities that
should replace the curvature and the torsion, but, clearly, the first two fundamental forms
could be good candidates. Thus, we can formulate the question in the following way: If
we are given a domain U ⊂ R
2
and two families of symmetric bilinear forms, with the
coefficients depending smoothly on the coordinates on U, such that, at each point of U,
the first form is positively defined, is there any regular parameterized surface r : U → R
3
such that the two families of bilinear forms be the two first fundamental forms of this
surface? The answer is not affirmative, because, as we shall see, the coefficients of
the first two fundamental forms of a surface are not independent, therefore, our initial
data should satisfy some compatibility conditions (which are, in fact, the integrability
conditions for a system of partial differential equation). If, however, this conditions
are fulfilled, then the negative answer turns into an affirmative one. It is the aim of
this section to establish the compatibility conditions and to formulate the existence and
uniqueness theorem for parameterized surfaces.
4.17.2 The differentiation rules. Christoffel’s coefficients
If (U, r) is a regular parameterized surface, then, for each (u, v) ∈ U, the vectors r
·
u
, r
·
v
, n
for a basis of the vector space R
3
r(u,v)
. Therefore, in particular, the derivatives of these
vectors can be expressed in terms of the vectors themselves. We already saw how to
express the derivatives of the normal versor. They define, essentially, the shape operator
of the surface. We shall obtain now similar formulae for the second order derivatives of
the radius vector. These formulae should be of the form
r
··
u
2
= Γ
1
11
r
·
u
+ Γ
2
11
r
·
v
+ An
r
··
uv
= Γ
1
12
r
·
u
+ Γ
2
12
r
·
v
+ Bn (4.17.1)
r
··
v
2
= Γ
1
22
r
·
u
+ Γ
2
22
r
·
v
+ C n
The coefficients Γ from these equations are called the Christoffel’s coefficients (of the
second kind). The coefficients A, B, C are easily identifiable as the coefficients of the
second fundamental form. To get the others, first of all, we shall express the scalar prod-
ucts r
··
u
2
r
·
u
, r
··
u
2
r
·
v
and the analogues in terms of the coefficients of the first fundamental
4.17. The fundamental equations of a surface 163
form and its derivatives. We have, first of all, r
·
u
2
= E, whence, differentiating with
respect to u, we get
r
··
u
2
r
·
u
=
1
2
E
·
u
. (4.17.2)
Differentiating the same equality with respect to v, we get
r
··
uv
r
·
u
=
1
2
E
·
v
(4.17.3)
Exactly in the same manner, starting from the definition of the other two coefficients of
the first fundamental form, we will obtain
r
··
uv
r
·
v
=
1
2
G
·
u
(4.17.4)
r
··
u
2
r
·
v
= F
·
u

1
2
E
·
v
(4.17.5)
r
··
v
2
r
·
u
= F
·
v

1
2
G
·
u
(4.17.6)
r
··
v
2
r
·
v
=
1
2
G
·
v
(4.17.7)
Returning to our problem, we multiply the first equation of (4.17.1) successively by r
·
u
and by r
·
v
. Using the formulae (4.17.2) and (4.17.5), as well as the definitions of the
coefficients of the first fundamental form, we get the system

¸
¸
¸
¸
¸
¸

1
11
+ FΓ
2
11
=
1
2
E
·
u

1
11
+ GΓ
2
11
= F
·
u

1
2
E
·
u
.
It is very easy to solve this system and we get

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
Γ
1
11
=
GE
·
u
− 2FF
·
u
+ FE
·
v
2(EG − F
2
)
Γ
2
11
=
2EF
·
u
− EE
·
v
− FE
·
u
2(EG − F
2
)
(4.17.8)
and, exactly in the manner, starting from the other two equations from (4.17.1), we get

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
Γ
1
12
=
GE
·
v
− FG
·
u
2(EG − F
2
)
Γ
2
12
=
EG
·
u
− FE
·
v
2(EG − F
2
)
(4.17.9)
164 Chapter 4. General theory of surfaces
and

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
Γ
1
22
=
2GF
·
v
− GG
·
u
− FG
·
v
2(EG − F
2
)
Γ
2
22
=
EG
·
v
− 2FF
·
v
− FG
·
u
2(EG − F
2
)
. (4.17.10)
As for the derivatives of the normal versor, looking back at the expression of the shape
operator in terms of the coefficients of the first two fundamental forms, we shall get
immediately the formulae

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
n
·
u
=
FD
·
− GD
EG − F
2
r
·
u
+
FD − ED
·
EG − F
2
r
·
v
n
·
v
=
FD
··
− GD
·
EG − F
2
r
·
u
+
FD
·
− ED
··
EG − F
2
r
·
v
(4.17.11)
Remark. The formulae (4.17.11) were obtained by the German mathematician Julius
Weingarten, therefore they are called the Weingartens’s formulae. Clearly, these formu-
lae uniquely define the shape operator. This is the reason why in many books the shape
operator (or its opposite) is called the Weingarten’s operator or Weingarten’s mapping.
Christoffel’s and Weingarten’s coefficients in curvature coordinates
Let us assume that in our parameterization the coordinate lines are lines of curvature.
Then, as we saw earlier, we should have F = 0 and D
·
= 0 on the entire support of the
parameterization. Then, as one can convince oneself easy, the Christoffel’s coefficients
become:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
Γ
1
11
=
1
2
E
·
u
E
=

∂u
ln E, Γ
2
11
= −
E
·
v
2G
,
Γ
1
12
=

∂v
ln E, Γ
2
12
=

∂u
lnG,
Γ
1
22
= −
G
·
u
E
, Γ
2
22
=

∂v
lnG,
(4.17.12)
while the Weingarten’s equation become:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
n
·
u
= −
D
E
r
·
u
n
·
v
= −
D
··
G
r
·
v
(4.17.13)
One should not be surprised that the partial derivatives of the unit normal vector are, each
of them, colinear to one of the partial derivative of the radius vector. This is, actually,
nothing but the definition of the lines of curvatures.
4.17. The fundamental equations of a surface 165
4.17.3 The Gauss’ and Codazzi-Mainardi’s equations for a surface
We shall prove now that between the coefficients of the first two fundamental forms of a
parameterized surface exists some relations, that we will call the Gauss’ equations and
the Codazzi-Mainardi’s equations, respectively. We summarize them in the following
theorem.
Theorem4.17.1 (Gauss, Codazzi, Mainardi). In any local parameterization of a surface,
the following systems of equations are fulfilled,
∂Γ
2
11
∂v

∂Γ
2
12
∂u
+ Γ
1
11
Γ
2
11
+ Γ
2
11
Γ
2
21
− Γ
1
12
Γ
2
11
− Γ
2
12
Γ
2
12
= EK
t
∂Γ
1
12
∂u

∂Γ
1
11
∂v
+ Γ
2
12
Γ
1
12
− Γ
2
11
Γ
1
22
= FK
t
∂Γ
1
22
∂u

∂Γ
1
12
∂v
+ Γ
1
22
Γ
1
11
+ Γ
2
22
Γ
1
12
− Γ
1
12
Γ
1
12
− Γ
2
12
Γ
1
22
= GK
t
∂Γ
2
12
∂v

∂Γ
2
22
∂u
+ Γ
1
12
Γ
2
12
− Γ
1
22
Γ
2
11
= FK
t
,
(4.17.14)
called the Gauss’s equations, and the relations
∂D
∂v

∂D
·
∂u
= DΓ
1
12
+ D
·

2
12
− Γ
1
11
) − D
··
Γ
2
11
∂D
·
∂v

∂D
··
∂u
= DΓ
1
22
+ D
·

2
22
− Γ
1
12
) − D
··
Γ
2
12
,
(4.17.15)
called the Codazzi-Mainardi’s equations. Here E, F, G are the coefficients of the first
fundamental form of the surface, while D, D
·
, D
··
are the coefficients of the second fun-
damental form.
Proof. To simplify a little bit the notations, let us rewrite the Weingarten’s equations as

¸
¸
¸
¸
¸
¸
n
·
u
= a
11
r
·
u
+ a
12
r
·
v
n
·
v
= a
21
r
·
u
+ a
22
r
·
v
. (4.17.16)
We have, obviously, the relation
r
···
u
2
v
− r
···
uvu
= 0.
But
r
··
u
2
= Γ
1
11
r
·
u
+ Γ
2
11
b
·
v
+ Dn,
166 Chapter 4. General theory of surfaces
hence
r
···
u
2
v
=
∂Γ
1
11
∂v
r
·
u
+ Γ
1
11
r
··
uv
+
∂Γ
2
11
∂v
r
·
v
+ Γ
2
11
r
··
v
2
+
∂D
∂v
n + Dn
·
v
=
=
∂Γ
1
11
∂v
r
·
u
+ Γ
1
11

1
12
r
·
u
+ Γ
2
12
r
·
v
+ D
·
n)+
+
∂Γ
2
11
∂v
r
·
v
+ Γ
2
11

1
22
r
·
u
+ Γ
2
22
r
·
v
+ D
··
n) +
∂D
∂v
n + D(a
12
r
·
u
+ a
22
r
·
v
) =
= r
·
u

¸
¸
¸
¸
¸
∂Γ
1
11
∂v
+ Γ
1
11
Γ
1
12
+ Γ
2
11
Γ
1
22
+ a
12
D

¸
¸
¸
¸
¸
+
+ r
·
v

¸
¸
¸
¸
¸
∂Γ
2
11
∂v
+ Γ
1
11
Γ
2
12
+ Γ
2
11
Γ
2
22
+ a
22
D

¸
¸
¸
¸
¸
+
+ n
¸
∂D
∂v
+ Γ
1
11
D
·
+ Γ
2
11
D
··

.
Analogously,
r
··
uv
= Γ
1
12
r
·
u
+ Γ
2
12
r
·
v
+ D
·
n,
hence
r
···
uvu
=
∂Γ
1
12
∂u
r
·
u
+ Γ
1
12
r
··
u
2
+
∂Γ
2
12
∂u
r
·
v
+ Γ
2
12
r
··
uv
+
∂D
·
∂u
n + D
·
n
·
u
=
=
∂Γ
1
12
∂u
r
·
u
+ Γ
1
12

1
11
r
·
u
+ Γ
2
11
r
·
v
+ Dn)+
+
∂Γ
2
12
∂u
r
·
v
+ Γ
2
12

1
12
r
·
u
+ Γ
2
12
r
·
v
+ D
·
n) +
∂D
·
∂u
n + D
·
(a
11
r
·
u
+ a
21
r
·
v
) =
= r
·
u

¸
¸
¸
¸
¸
∂Γ
1
12
∂u
+ Γ
1
12
Γ
1
11
+ Γ
2
12
Γ
1
12
+ a
11
D
·

¸
¸
¸
¸
¸
+
+ r
·
v

¸
¸
¸
¸
¸
∂Γ
2
12
∂u
+ Γ
1
12
Γ
2
11
+ Γ
2
12
Γ
2
12
+ a
21
D
·

¸
¸
¸
¸
¸
+
+ n
¸
∂D
·
∂u
+ Γ
1
12
D + Γ
2
12
D
·

.
4.17. The fundamental equations of a surface 167
If we subtract the previous relations, we obtain
0 = r
···
u
2
v
− r
···
uvu
= r
·
u

¸
¸
¸
¸
¸
∂Γ
1
11
∂v

∂Γ
1
12
∂u
+ Γ
2
11
Γ
1
22
− Γ
2
12
Γ
1
12
+ a
12
D − a
11
D
·

¸
¸
¸
¸
¸
+
+ r
·
v

¸
¸
¸
¸
¸
∂Γ
2
11
∂v

∂Γ
2
12
∂u
+ Γ
1
11
Γ
2
12
+ Γ
2
11
Γ
2
22
+ a
22
D − a
21
D
·

¸
¸
¸
¸
¸
+
+ n
¸
∂D
∂v

∂D
·
∂u
+ Γ
1
12
D − D
·

2
12
− Γ
1
11
) + Γ
2
12
D
··

.
As the vectors r
·
u
, r
·
v
and n are linearly independent, a linear combination of them can
vanish only if each coefficient vanishes. If we equate to zero the coefficient of n, one
sees that one gets the first of the Codazzi-Mainardi’s equations. If we equate to zero
the coefficient of r
·
u
follows, using the expressions of a
21
and a
22
, the first of the Gauss’
equations, while from the coefficient of r
·
v
follow the second of the Gauss’s equations.
The other three equations can be obtained in the same way, using, this time, the relation
r
···
uv
2
= r
···
vuv
.

Corollary 4.17.1. If the coordinate lines are lines of curvature, then the Codazzi-Mainardi
equations can be written a lot simpler:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂D
∂v
= D
∂ ln E
∂v
+
D
··
2G
∂E
∂v
∂D
··
∂u
=
D
2E
∂G
∂u
+ D
··
∂ lnG
∂u
.
(4.17.17)
4.17.4 The fundamental theorem of surface theory
The theorem we are going to prove in this section is the analogue of the existence and
uniqueness theorem for space curves. It was first established by the French mathemati-
cian Ossian Bonnet, in 1860.
To shorten the formulas, in the following we shall denote the coordinates by u
1
and
u
2
instead of u and v and we shall use the index notation to denote the components of
the matrices of the first two fundamental forms of the surface. More precisely, for the
coefficients of the first fundamental form we shall write
g
11
= E, g
12
= g
21
= F, g
22
= G, (4.17.18)
168 Chapter 4. General theory of surfaces
while for the coefficients of the second fundamental form we shall write
h
11
= D, h
12
= h
21
= D
·
, h
22
= D
··
. (4.17.19)
Also, we shall denote by g the determinant of the matrix of the first fundamental form.
Finally, we shall denote by r
·
i
the derivative of r with respect to the coordinate u
i
and by
r
··
ij
– the second order derivative of r with respect to the coordinates u
i
and u
j
, where i
and j can take the values 1 and 2. Clearly, as we always assume that the surfaces are as as
smooth as one expects to be (i.e., in this context, at least C
2
), the order of differentiation
is not relevant, therefore we shall always have r
··
ij
= r
··
ji
.
Theorem (Ossian Bonnet, 1860). Let U ⊂ R
2
be an open set. On U there are given the
symmetric matrix functions
g
i j
= g
i j
(u
1
, u
2
), h
i j
= h
i j
(u
1
, u
2
), i, j = 1, 2 (4.17.20)
of classes C
2
and C
1
, respectively, such that for each (u
1
, u
2
) ∈ U the quadratic for asso-
ciated to the bilinear form whose matrix is (g
i j
) is positively defined and, moreover, the
components of the two functions verify the Gauss and Codazzi-Mainardi compatibility
conditions. We choose u
0
= (u
1
0
, u
2
0
) ∈ U, p
0
∈ R and the vectors
r
·
1
(0)
, r
·
2
(0)
, n
(0)
∈ T
p
0
R
3
, (4.17.21)
such that r
·
i
(0)
r
·
j
(0)
= g
i j
(u
0
), n
(0)
r
·
i
(0)
= 0, n
(0)
n
(0)
= 1, while the triple ¦r
·
1
(0)
, r
·
2
(0)
, n
(0)
¦
is a right-handed basis of the vector space T
p
0
R
3
. Then there exists a single regular pa-
rameterized surface of class C
3
, r : V → R
3
, with V ⊂ U – an open set, such that the
following conditions are fulfilled:
(i) r(u
0
) = p
0
(the surface “passes” through p
0
for u = u
0
).
(ii)
∂r
∂u
i
(u
0
) = r
·
i
(0)
, i = 1, 2.
(iii) n(u
0
) = n
(0)
.
(iv) g
i j
and h
i j
are the coefficients of the first two fundamental forms of the parameter-
ized surface r (with respect to the orientation of r defined by the unit normal vector
n).
4.17. The fundamental equations of a surface 169
Proof. We consider the system of partial differential equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂r
·
i
∂u
j
=
2
¸
k=1
Γ
k
i j
r
·
k
+ h
i j
n,
∂n
∂u
i
= −
2
¸
j=1
2
¸
k=1
h
i j
g
jk
r
·
k
(4.17.22)
with respect to the unknown functions r
·
1
, r
·
2
, n, where the coefficients Γ
k
i j
are computed
with the formula (4.17.8)–(4.17.10). This is a linear and homogeneous system which is
completely integrable, because the compatibility conditions

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸

2
r
·
i
∂u
j
∂u
k
=

2
r
·
i
∂u
k
∂u
j

2
n
∂u
j
∂u
k
=

2
n
∂u
k
∂u
j
(4.17.23)
are equivalent, as we saw, to the Gauss-Weingarten equations, which are satisfied, by
hypothesis. Therefore, from standard results from the theory of partial differential equa-
tions of first order, it follows that there exists an open neighborhood W ⊂ U of the point
u
0
and a set of three C
2
vector functions
5
r
·
1
, r
·
2
, n : W → R
3
, which are solutions of the
system (4.17.22), with the given initial condition at the point u
0
.
Let us notice, also, that a set of initial conditions as in the theorem always exists,
due, mainly, to the fact that the quadratic form associated to g
i j
is positively defined.
Indeed, we can make, for instance, the following choice:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
·
1
(0)
=
¸

g
11
(u
0
), 0, 0
¸
r
·
2
(0)
=

¸
¸
¸
¸
¸
¸
g
12
(u
0
)

g
11
(u
0
)
,

g
11
(u
0
)g
22
(u
0
) − (g
12
(u
0
))
2

g
11
(u
0
)
, 0
¸
¸
¸
¸
¸
¸
¸
n
(0)
= ¦0, 0, 1¦
. (4.17.24)
We leave to the reader to check that, indeed, these vectors verify the conditions from the
hypothesis.
We consider, now, the system of partial differential equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂r
∂u
1
= r
·
1
,
∂r
∂u
2
= r
·
2
. (4.17.25)
5
The order of smoothness is with one unit greater then the smallest order of smoothness of the coeffi-
cients, as the system is of first order.
170 Chapter 4. General theory of surfaces
This system is, again, completely integrable, because the integrability condition

2
r
∂u
i
∂u
j
=

2
r
∂u
j
∂u
i
(4.17.26)
is equivalent to the condition
∂r
·
i
∂u
j
=
∂r
·
j
∂u
i
(4.17.27)
which is true, as one can convince oneself looking at the first equation (4.17.22), because
of the symmetry of the second matrix (h
ji
= h
i j
) and because of the symmetry of the
Christoffel’s coefficients in the lower indices. Therefore, applying again the existence
and uniqueness theorem, it follows that there exists an open neighborhood V ⊂ W ⊂ U
of u
0
and a single C
3
function r : V → R
3
such that r(u
0
) = p
0
.
We are not done yet, because we still have to show that g
i j
and h
i j
are the first two
fundamental forms of the parameterized surface defined by r. Apparently, we also have
to show that r is regular. But this follows immediately if we prove that g is the first
fundamental form, because
g
i j
=
∂r
∂u
i

∂r
∂u
j
and then
∂r
∂u
i
×
∂r
∂u
j
0,
because the square of the norm of this vector is not zero (as it is equal to the determinant
of the first fundamental form, which is strictly greater then zero, since the form is pos-
itively defined). It is, actually, enough to show that, all over V, the following relations
are fulfilled:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
·
i
r
·
j
= g
i j
,
r
·
i
n = 0,
n n = 1
. (4.17.28)
To this end, we shall compute the derivatives with respect to the coordinates of the scalar
products, taking into account the Gauss-Weingarten equations and we get the system of
first order partial differential equations:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂(r
·
i
r
·
j
)
∂u
k
=
2
¸
l=1
Γ
l
ik
(r
·
l
r
·
j
) +
2
¸
l=1
Γ
l
jk
(r
·
l
r
·
i
) + h
ik
(r
·
j
n) + h
jk
(r
·
i
n)
∂(r
·
j
n)
∂u
i
= −
2
¸
l,k=1
h
il
g
lk
(r
·
k
r
·
j
) +
2
¸
l=1
Γ
l
i j
(r
·
l
n) + h
i j
( n n)
∂( n n)
∂u
i
= −2
2
¸
l,k=1
h
il
g
lk
(r
·
k
n)
. (4.17.29)
4.17. The fundamental equations of a surface 171
As probably the reader suspects already (and it is asked to check for himself), this system
is, again, completely integrable, which means that it has a single solution for prescribed
initial values. We “guess” that this solution is, exactly, (4.17.28) and we shall check
this in the following. For the last equation, there is nothing to check: if we substi-
tute (4.17.28), we get, simply, 0 = 0. For the second equation, after the substitution, the
left-hand side is zero, while the right-hand side becomes:

2
¸
l,k=1
h
il
g
lk
g
k j
+ h
i j
= −
2
¸
l=1
h
il
δ
l
j
+ h
i j
= −h
i j
+ h
i j
= 0,
so we are done. Finally, the first equation becomes
∂g
i j
∂u
k
=
2
¸
l=1
Γ
l
ik
g
k j
+
2
¸
l=1
Γ
l
jk
g
li
which is very easy to prove, using the definition of the Christoffel’s coefficients. Thus,
the functions (4.17.28) give a solution of the system (4.17.29), which, obviously, verify
the initial conditions of the theorem. As the solution is unique, for the given initial
conditions we have

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
·
i
r
·
j
= g
i j
,
r
·
i
n = 0,
n n = 1
(r
·
1
, r
·
2
, n) > 0
, (4.17.30)
which means that we proved already that
- r
·
i
are the derivatives of r;
- n is the unit normal of the parameterized surface given by r;
- g
i j
are the coefficients of the first fundamental form of r.
We are left with the proof of the fact that h
i j
are the coefficients of the second fundamen-
tal form of r. Let us denote, for the moment, by b
i j
these coefficients. As we know, they
are given by
b
il
= −n
·
i
r
·
l
=
2
¸
j,k=1
h
i j
g
jk
r
·
k
r
·
l
=
2
¸
j,k=1
h
i j
g
jk
g
kl
=
2
¸
j=1
h
i j
δ
j
l
= h
il
,
which concludes the proof of the Bonnet’s theorem.
172 Chapter 4. General theory of surfaces
4.18 The Gauss’ egregium theorem
The theorem that we are going to prove in this section (and which is implicit in the
equations of Gauss above) is one of the most important theorem in classical differential
geometry. Not by accident Gauss called it, in his famous Disquisitiones circa superfi-
cies curvas “theorema egregium”, i.e. the remarkable theorem. As we just saw in the
previous section, the total curvature of a surface in R
3
can be expressed in terms of the
determinants of the first two fundamental forms of the surface. This would mean, in
principle, that the total curvature depends both on the intrinsic data (the first fundamen-
tal form) and extrinsic data (i.e. the second fundamental form). It turns out, however,
that the situation is different, i.e. we have:
Theorem4.4 (Gauss, 1827). The total curvature of a surface of class at least C
3
depends
only on the coefficients of the first fundamental form of the surface and their first order
derivatives with respect to the coordinates.
Proof. There exists several proofs of this theorem (which is called, in many books, us-
ing the Latin term used by Gauss, namely theorema egregium. The original proof, given
by Gauss in 1827, is quite complicated. The proof we are going to give here is the first
different proof, belonging to the German mathematician Richard Baltzer (1867)
6
, al-
though Struik credits the formula which will be established to the Italian mathematician
Brioschi. We start with the formula
K
t
=
DD
··
− D
·2
EG − F
2
,
which we rewrite in the form
K
t
(EG − F
2
) = DD
··
− D
·2
, (4.18.1)
or, having in mind the expressions of the coefficients of the second fundamental form,
K
t
(EG − F
2
)
2
= (r
··
u
2
, r
·
u
, r
·
v
) (r
··
v
2
, r
·
u
, r
·
v
) − (r
··
uv
, r
·
u
, r
·
v
)
2
. (4.18.2)
The right hand side of the equation (4.18.2) can be written in a more convenient form if
we notice that each term is, in fact, the product of two determinants. We shall use, now,
6
The name of Richard Baltzer (1818–1887) is not well known today, however, in the second half of
the nineteenth century he was considered an important geometer. For the history of mathematics, there are
important his “Elementhe der Mathematik”, published in several editions, where was mentioned, for the
first time, the Non-Euclidean geometry.
4.18. The Gauss’ egregium theorem 173
the following formula for the product of two determinants, known from vector algebra:
(a, b, c) (d, e, f) =

a d a e a f
b d b e b f
c d c e c f

. (4.18.3)
Using the formula (4.18.3), the formula (4.18.2) becomes
K
t
(EG − F
2
)
2
= (r
··
u
2
r
··
v
2
− r
··
uv
2
)(EG − F
2
)+
+

0 r
··
u
2
r
·
u
r
··
u
2
r
·
v
r
·
u
r
··
v
2
E F
r
·
v
r
··
v
2
F G


0 r
··
uv
r
·
u
r
··
uv
r
·
v
r
··
uv
r
·
u
E F
r
··
uv
r
·
v
F G

.
(4.18.4)
Thus, already a part of the terms involved in the computation of the total curvature are
expressed in terms of the coefficients of the first fundamental form. As one can see
immediately, the remaining terms are of two kinds: either a product of a second order
derivative of r and a first order one, either a product of two second order derivatives. The
first kind of terms are easier to be taken care of. Indeed, starting from the definition of
the coefficients of the first fundamental form, E = r
·
u
r
·
u
, F = r
·
u
r
·
v
, G = r
·
v
r
·
v
one
obtains, differentiating with respect to the coordinates, the following expressions:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
r
··
u
2
r
·
u
=
1
2
E
·
u
r
··
uv
r
·
u
=
1
2
E
·
v
r
··
v
2
r
·
v
=
1
2
G
·
v
r
··
uv
r
·
v
=
1
2
G
·
u
r
··
u
2
r
·
v
= F
·
u

1
2
E
·
v
r
··
v
2
r
·
u
= F
·
v

1
2
G
·
u
. (4.18.5)
Differentiating once more the fourth equation above with respect to u and the fifth with
respect to v and subtracting them side by side, we get also the expression for the products
of second order derivatives of r:
r
··
u
2
r
··
v
2
− r
··
uv
2
= −
1
2
G
··
u
2
+ F
··
uv

1
2
E
··
v
2
. (4.18.6)
174 Chapter 4. General theory of surfaces
Combining everything we obtained, we get the following expression (due, as we said
before, to Baltzer) for the total curvature
K
t
=
1
(EG − F
2
)
2


1
2
G
··
u
2
+ F
··
uv

1
2
E
··
v
2
1
2
E
·
u
F
·
u

1
2
E
·
v
F
·
v

1
2
G
·
u
E F
1
2
G
·
v
F G



1
(EG − F
2
)
2

0
1
2
E
·
v
1
2
G
·
u
1
2
E
·
v
E F
1
2
G
·
u
F G

,
(4.18.7)
which concludes the proof, as we got an expression of K
t
which, indeed, depends only on
the coefficients of the first fundamental form and their derivatives up to second order.
Exercise 4.18.1 (Frobenius). Show that the total curvature of a surface can be written,
also, in the following form, easier to remember:
K
t
= −
1
4(EG − F
2
)
2

E E
·
u
E
·
v
F F
·
u
F
·
v
G G
·
u
G
·
v

+
+
1
2

EG − F
2


∂u
¸
F
·
v
− G
·
u

EG − F
2

+

∂v
¸
F
·
u
− E
·
v

EG − F
2

.
(4.18.8)
In the particular case of an orthogonal coordinate system (F ≡ 0), we get the following
nice “divergence” expression for the total curvature:
K
t
= −
1
2

EG


∂u
¸
G
·
u

EG

+

∂v
¸
E
·
v

EG

, (4.18.9)
which will be used later for some integral formulae.
Exercise 4.18.2 (Liouville). Prove the following (slightly asymmetric) formula for the
total curvature of a surface:
K
t
= −
1
2

EG − F
2

¸
¸
¸
¸
¸
¸

∂u

¸
¸
¸
¸
¸
¸
G
·
u
+
F
G
G
·
v
− 2F
·
v

EG − F
2

¸
¸
¸
¸
¸
¸
+

∂v

¸
¸
¸
¸
¸
¸
E
·
v

F
G
G
·
u

EG − F
2

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
. (4.18.10)
4.19. Geodesics 175
4.19 Geodesics
4.19.1 Introduction
The curves we are going to study in this chapter are direct generalizations of the straight
lines. More specifically, they are curves that projects on the tangent planes of the surface
as straight lines. There are many different approaches to geodesics. We chose here the
one we find more elementary.
4.19.2 The Darboux frame. The geodesic curvature and geodesic torsion
Let (I, ρ) be a parameterized curve whose support lies on a surface S and let M = ρ(t
0
)
be a point of the curve, with t
0
∈ I. As ρ is, in particular, a space curve, we can attach
to it the Frenet frame at the point M, ¦M; τ, ν, β¦. As we saw in the first part of the book,
this frame is good enough if we want to investigate the curve ρ as an independent object,
but it is not very helpful if one wishes to study the connections between the curve and
the surface. To this end, we shall introduce another orthonormal frame, which involves
both vectors related to the curve and to the surface. The first vector of the new frame
will be still the unit tangent vector of the curve, τ. The second, related, this time to the
surface, is the unit normal of the surface, n. The third one, let it be denoted by N, will
be chosen in such a way that the basis ¦τ, N, n¦ be direct, or in other words, such that
(τ, N, n) = 1. This means, of course, that
N = n × τ.
Clearly, N lies in the normal plane of the curve at M, therefore it will be called the unit
tangential normal vector of the curve. The name tangential comes, of course from the
fact that N lies, also, in the tangent plane of the surface at M.
The frame ¦M; τ, N, n¦ is called the Darboux frame or the Ribaucour-Darboux frame
of the surface S along the curve ρ.
The next step we are going to make is to compute the derivatives of the vectors of
the Darboux frame and to obtain a set of linear differential equation which is similar to
the Frenet frame and which will play an important role in the following sections. To get
them, the intention is exactly to use the Frenet equations. Therefore, we shall start by
expressing the vectors N and n in terms of the vectors of the Frenet frame. We denote
by θ the angle between the vectors ν and n. Then, as one can see immediately that

¸
¸
¸
¸
¸
¸
ν = cos(N, ν) N + sin(N, ν)n
β = cos(N, β) N + sin(N, β)n
176 Chapter 4. General theory of surfaces
As (N, ν) =
π
2
and (N, β) = π − θ, we get

¸
¸
¸
¸
¸
¸
ν = sin θN + cos θn
β = −cos θN + sin θn
.
Conversely, we get

¸
¸
¸
¸
¸
¸
N = sin θ ν − cos θ β
n = cos θ ν + sin θ β
.
Now, the derivatives of the vectors of the Darboux frame with respect to the arclength of
the curve can be expressed in terms of these vectors as

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ = a(s) N + b(s)n
N
·
= c(s)τ + d(s) n
n
·
= e(s) τ + f (s) N,
(4.19.1)
where a, b, c, d, e, f are smooth functions of the arclength. We remind that the derivative
of a vector of the Darboux frame is perpendicular on that vector, because the frame is
orthonormal. For the same reason, we deduce immediately that the six coefficients are
not independent and, in fact, we have the relations: c = −a, e = −b, f = −d, therefore
the system (4.19.1) becomes

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
τ = a(s) N + b(s)n
N
·
= −a(s)τ + d(s) n
n
·
= −b(s) τ − d(s) N,
(4.19.2)
We shall express now the quantities a, b, d in terms of the characteristics of the curve and
the angle θ. We notice, first of all, that, from the first of the Frenet’s formula, we get
τ
·
= kν = k(sin θ N + cos θ n),
hence, identifying the coefficients with those of the first equation of the system (4.19.2),
we obtain
a = k sin θ; b = k cos θ. (4.19.3)
The quantity k cos θ is already known to us: it is nothing but the normal curvature
k
n
of the surface that was studied previously. The function k sin θ, instead, is new. We
4.19. Geodesics 177
will denote it by k
g
and we will call it geodesic or tangential curvature of the curve
7
. To
find the quantity d, we start from the relation
N = sin θ ν − cos θ β.
By differentiating with respect to the arclength of the curve, we get, using the last two
of the Frenet’s formulae:
N
·
= θ
·
sin θ ν − sin θ(−k τ + χ β) + θ
·
sin θ β + χ cos θ ν =
= −k sin θ τ + (θ
·
+ χ) (cos θ ν + sin θ β) =
= −k sin θ τ + (θ
·
+ χ) n,
and, comparing with the second equation from (4.19.2), we get
d = θ
·
+ χ.
This function is denoted by χ
g
and it s called the geodesic torsion. His geometric mean-
ing will be made clear later. Clearly, we cannot claim, as we did with the geodesic
curvature, that the geodesic torsion is the torsion of the projection of the curve on the
tangent plane, as the torsion of that curve is always zero, while the geodesic torsion of
the given curve is not, usually.
The geodesic curvature plays a much more important role in the differential geometry
of surfaces than the geodesic torsion does, so we start by focusing on it. We notice, to
begin with, that
k
g
= τ
·
N = −τ N
·
. (4.19.4)
Since, as we saw earlier, N = n × τ, one obtains for k
g
the expression
k
g
= τ
·
N = τ
·
(n × τ),
i.e.
k
g
= (τ, τ
·
, n). (4.19.5)
This formula holds for naturally parameterized curves. Let us consider, now, an arbitrary
regular parameterized curve on S , given by the local equations u = u(t), v = v(t). We
have, therefore, r = r(u(t), v(t)). If we denote by a dot the differentiation with respect to
the parameter t along the curve, we get
τ ≡
dr
ds
=
dr
dt
dt
ds
=
1
˙ s
˙ r,
7
Here the term “tangential” refers to the tangent plane of the surface, not to the tangent line of the curve.
In fact, it can be shown that the geodesic curvature at a point of a curve lying on a surface is the signed
curvature of the projection of the curve on the tangent plane to the surface at that particular point.
178 Chapter 4. General theory of surfaces
therefore
τ
·
=

ds
=
1
˙ s
3
(¨ r ˙ s − ˙ r¨ s).
Thus, for the geodesic curvature one gets
k
g
= (τ, τ
·
, n) =
¸
1
˙ s
˙ r,
1
˙ s
3
(¨ r ˙ s − ˙ r¨ s), n

=
1
˙ s
4
(˙ r, ¨ r ˙ s − ˙ r¨ s, n) =
1
˙ s
4
(˙ r, ˙ s ¨ r, n),
i.e.
k
g
=
1
˙ s
3
(˙ r, ¨ r, n). (4.19.6)
The formulas we got so far for the geodesic curvature use mixture of information
about the curve and about the surface, but they don’t use explicitly the things we usually
compute when we are given a local parameterization of a surface or, more generally,
a regular parameterized surface, namely the coefficients of the first two fundamental
forms. So, the next step will be to obtain a formula for the geodesic curvature in terms of
the coefficients of the fundamental forms, when the curve is given by its local parametric
equations with respect to a local parameterization of the surface. We shall discover, in
fact, that the geodesic curvature can be expressed in terms of the coefficients of the first
fundamental form and they first order partial derivatives with respect to the coordinates.
To simplify the notations, we shall use, for a while, the index notations. In other
words, we shall denote the coordinates with u
1
and u
2
, instead of u and v, while the
first partial derivatives of the radius vector r with respect to the coordinates will be
denoted by r
·
i
and those of second order by r
··
i j
, i, j = 1, 2. Moreover, we shall use the
Einstein’s summation convention: every time an index enter twice in a monomial, once
in an inferior position and once in a superior one, then one makes summation after all
the allowed values of the index (in our case, 1 and 2). Thus, an expression of the form
a
i
u
i
should be read as
a
i
u
i
= a
1
u
1
+ a
2
u
2
.
Going back to our problem, we have, with the newly introduced notations:

¸
¸
¸
¸
¸
¸
˙ r ≡
dr
dt
= r
·
i
˙ u
i
¨ r = r
··
i j
˙ u
i
˙ u
j
+ r
·
k
¨ u
k
.
The decomposition of the second order partial derivatives of the radius vectors in terms
of the basis ¦r
·
1
, r
·
2
, n¦ is already known to us: it can be described using the Christoffel’s
coefficients and the second fundamental form as
r
··
i j
= Γ
k
i j
r
·
k
+ h
i j
n,
4.19. Geodesics 179
where, as we know, h
i j
are the coefficients of the second fundamental formof the surface.
Thus, we have
¨ r =

Γ
k
i j
r
·
k
+ h
i j
n

˙ u
i
˙ u
j
+ r
·
k
¨ u
k
=

¨ u
k
+ Γ
k
i j
˙ u
i
˙ u
j

r
·
k
+ h
i j
˙ u
i
˙ u
j
n
or
¨ r =

¨ u
k
+ Γ
k
i j
˙ u
i
˙ u
j

r
·
k
+ ϕ
2
(˙ r, ˙ r) n.
With these in hand, we can write now
k
g
=
1
˙ s
3
(˙ r, ¨ r, n) =
1
˙ s
3

˙ u
m
r
·
m
,

¨ u
k
+ Γ
k
i j
˙ u
i
˙ u
j

r
·
k
+ ϕ
2
(˙ r, ˙ r), n

=
=
1
˙ s
3
,
˙ u
1

¨ u
2
+ Γ
2
i j
˙ u
i
˙ u
j

− ˙ u
2

¨ u
1
+ Γ
1
i j
˙ u
i
˙ u
j
¸
r
·
1
, r
·
2
, n

.
But

r
·
1
, r
·
2
, n

=

r
·
1
× r
·
2

n =

r
·
1
× r
·
2

r
·
1
× r
·
2
|r
·
1
× r
·
2
|
= |r
·
1
× r
·
2
| =

g,
where g is the determinant of the first fundamental form, whence
k
g
=

g
˙ s
3
,
˙ u
1

¨ u
2
+ Γ
2
i j
˙ u
i
˙ u
j

− ˙ u
2

¨ u
1
+ Γ
1
i j
˙ u
i
˙ u
j
¸
. (4.19.7)
If, in particular, the curve is naturally parameterized, we get
k
g
=

g
,
(u
1
)
·

(u
2
)
··
+ Γ
2
i j
(u
i
)
·
(u
j
)
·

− (u
2
)
·

(u
1
)
·
+ Γ
1
i j
(u
i
)
·
(u
j
)
·
¸
. (4.19.8)
In the traditional notations, the formula for the geodesic curvature of a naturally para-
meterized curve is
k
g
=

EG − F
2
,
Γ
2
11
u
·3
+


2
12
− Γ
1
11

u
·2
v
·
+

Γ
2
22
− 2Γ
1
12

u
·
v
·2

−Γ
1
22
v
·3
+ u
·
v
··
− u
··
v
·
¸
.
(4.19.9)
This equation can also be written in a more elegant form as
k
g
=

EG − F
2
det
¸
u
·
u
··
+ u
·2
Γ
1
11
+ 2u
·
v
·
Γ
1
12
+ v
·2
Γ
1
22
v
·
v
··
+ u
·2
Γ
2
11
+ 2u
·
v
·
Γ
2
12
+ v
·2
Γ
2
22

. (4.19.10)
If, in particular, the surface S is the plane, with the Cartesian coordinates on it, then
the determinant of the first fundamental form is, of course, one, as the matrix of the
fundamental form is the unit matrix, while all the Christoffel’s coefficients vanish. As a
consequence, in this situation the geodesic curvature of the curve (which is, in this case,
a plane curve), is nothing but its signed curvature:
k
g
= u
·
v
··
− u
··
v
·
≡ k
±
.
Remark. It can be shown that, in fact, the geodesic curvature of a curve on a surface is
nothing but the signed curvature of the projection of the curve on the tangent plane.
180 Chapter 4. General theory of surfaces
4.19.3 Geodesic lines
Definition 4.19.1. Let S be a surface. A parameterized curve ρ : I → S is called a
geodesic line, or simply, a geodesic if, at each point, its geodesic curvature vanishes.
Remark. According to the formula (4.19.12), the geodesic curvature of a curve can be
written as
k
g
= (τ, τ
·
, n).
On the other hand, from the first formula of Frenet, τ ×τ
·
= kβ, therefore, k
g
vanishes at
a point of a curve on the surface if and only if the binormal of the curve is perpendicular
on the normal of the surface at that point or, in other words, the geodesic curvature
vanishes if and only if the normal of the surface is contained in the osculating plane.
Thus, the geodesic lines are exactly those lines on the surface for which the osculating
plane at each of their points contain the normal of the surface at that point. In fact, in
many books, this property is taken as the definition of the geodesics.
Another remark that should be made is that since, as we noticed earlier, the geodesic
curvature is the signed curvature of the projection of the curve on the tangent plane,
we can say that the geodesics are those curves which project on each tangent plane of
the surface after a straight line. In this sense, we might say that the geodesics are the
straightest lines on the surfaces.
The formula 4.19.9 leads to
Theorem 4.5. The differential equation of the geodesic lines is
Γ
2
11
u
·3
+


2
12
− Γ
1
11

u
·2
v
·
+

Γ
2
22
− 2Γ
1
12

u
·
v
·2
− Γ
1
22
v
·3
+ u
·
v
··
− u
··
v
·
= 0. (4.19.11)
Also, from (4.19.10) we can deduce that
Theorem 4.6. The geodesic lines verify the system of differential equations

¸
¸
¸
¸
¸
¸
u
··
+ u
·2
Γ
1
11
+ 2u
·
v
·
Γ
1
12
+ v
·2
Γ
1
22
= 0
v
··
+ u
·2
Γ
2
11
+ 2u
·
v
·
Γ
2
12
+ v
·2
Γ
2
22
= 0
. (4.19.12)
There is, apparently, a contradiction between the previous two theorems, as the first
claim that the geodesics are solutions of a single equations, while the second – that they
are a solution of a system of different equation. In fact, however, the two equations of the
system (4.19.12) are not independent, because we impose the geodesics to be naturally
parameterized, therefore, between the functions u and v there is an extra relation.
The existence (locally, at least) of geodesics passing through a point of a surface and
having at that point a given tangent vector is a consequence of standard results in the
theory of ordinary differential equations. Finding geodesics, is, usually, a very delicate
business and can be done explicitly only in special situations.
4.19. Geodesics 181
Examples of geodesics
The geodesics of the plane. By the geometrical interpretation of the geodesics, as
the curves that project on the tangent plane on straight lines, we deduce immediately
that the geodesics of the plane are the straight lines and only them. On the other hand,
by using cartesian coordinates, we fined immediately that the Christoffel’s coefficients
vanish identically, therefore the equations of geodesics become

¸
¸
¸
¸
¸
¸
u
··
= 0
v
··
= 0
,
which lead to u(s) = a
1
s +b
1
, v(s) = a
2
s +b
2
, i.e., again, the geodesics are straight lines.
The geodesics of the sphere. The geodesics of the sphere can be found very easily
from their interpretation as being those curves for which the osculating plane contains
the normal to the surface. As we know, in the case of the sphere all the normals are
passing through the center of the sphere, therefore the osculating planes should pass, all
of them through the center, which leads immediately to the idea that the geodesics are
arcs of great circles of the sphere.
Also, using a standard parameterization of the sphere (with spherical coordinates),
we can find (do that!) the equations of the geodesics as:

¸
¸
¸
¸
¸
¸
¨
θ − sin θ cos θ ˙ ϕ
2
= 0
¨ ϕ + 2 cot θ
˙
θ ˙ ϕ = 0
.
We assume that the explicit equation of the geodesic is θ = θ(ϕ). Then
˙
θ =

ds
=


˙ ϕ,
¨
θ =
d
ds
¸

˙ ϕ +


¨ ϕ =


¨ ϕ +
d
2
θ

2
˙ ϕ
2
.
Therefore, the first equation of the system becomes:
¨ ϕ


+
d
2
θ

2
˙ ϕ
2
− sin θ cos θ ˙ ϕ
2
= 0.
On the other hand, from the first equation,
¨ ϕ = −2 cot θ
˙
θ ˙ ϕ = −2 cot θ


˙ ϕ
2
,
182 Chapter 4. General theory of surfaces
hence:
˙ ϕ
2
¸
d
2
θ

2
− 2 cot θ


− sin θ cos θ

= 0.
We have two possibilities: either ˙ ϕ = 0, i.e. ϕ = const, and, in this case, the curve is,
obviously, a great circle of the sphere (a meridian), either
d
2
θ

2
− 2 cot θ


− sin θ cos θ = 0.
In this case, we make the substitution z = cot θ and, after a straightforward computation,
we get
d
2
z

2
+ z = 0,
and this equation has the general solution
z = cot θ = Acos ϕ + Bsin ϕ
or
Asin θ cos ϕ + Bsin θ sin ϕ − cos θ = 0,
which is the equation of a great circle, lying in the plane passing through the origin and
having as normal vector the vector (A, B, −1).
4.19.4 Liouville surfaces
Definition 4.19.2. A surface is called a Liouville surface if it can be parameterized
locally in such a way that the first fundamental form can be written as
ds
2
= (U(u) + V(v))(du
2
+ dv
2
), (4.19.13)
where U and V are smooth functions of a single variable.
These surfaces have been introduced by the French mathematician Joseph Liouville
in the Note III of the 5th edition of the book of Gaspard Monge, Application. In partic-
ular, the surfaces of revolution are examples of Liouville surfaces. The most important
characteristic of Liouville surfaces is the fact that their geodesics an be found through
quadratures. This is exactly what we are going to prove in the rest of this section.
First of all, it is an easy matter to prove that the equation of the geodesics for the
metric (4.19.13) can be written as
(U
·
v
·
− V
·
u
·
)(u
·2
+ v
·2
) + 2(U + V)(u
·
v
··
− v
·
u
··
) = 0. (4.19.14)
4.19. Geodesics 183
Here, for the functions U and V the prime denotes the derivative with respect to u and v,
respectively, while for the coordinate functions u and v the prime denotes the derivative
with respect to the parameter t along the geodesic. Of course, U and V are, both, func-
tions of t, through the coordinate functions. Therefore, the equation of geodesics can be
rewritten as
u
·
v
·
dU
dt

v
·
u
·
dV
dt
+ 2(U + V)
u
·
v
··
− v
·
u
··
u
·2
+ v
·2
= 0. (4.19.15)
We see that this equation actually does not involve the coordinate functions u and v, but
only their derivatives. We substitute them with other two functions of t, ρ and α, defined
by

¸
¸
¸
¸
¸
¸
u
·
= ρ cos α
v
·
= ρ sin α
. (4.19.16)
In the sequel, the equation of the geodesics becomes
sin
2
α
dU
dt
− cos
2
α
dV
dt
+ 2(U + V) sin αcos α

dt
= 0.
This equation can be also written as
d
dt

U sin
2
α − V cos
2
α

= 0,
whence
U sin
2
α − V cos
2
α = a,
where a is a constant. Returning to the old coordinate functions, we get
v
·2
U − u
·2
V = a(u
·2
+ v
·2
),
whence

du

U − a
= ±

dv

V + a
+ b, (4.19.17)
where b is another integration constant.
184 Chapter 4. General theory of surfaces
CHAPTER 5
Special classes of surfaces
5.1 Ruled surfaces
Intuitively speaking, the ruled surfaces are the surfaces generated by a moving straight
line, which is subject to a further condition, usually this condition being that the moving
line remains tangent to a given surface or intersects a given regular space curve. More
precisely
5.1.1 General ruled surfaces
After the plane, the ruled surfaces are, no doubt, the simplest surfaces. We recall that a
surface is called ruled if it is generated by a straight line (the ruling, moving in space,
lying all the time on a given curve, called the directrix. Clearly, once the ruled surface
was defined, the directrix can be replaced by another curve , obtained, for instance, by
sectioning the surface by a plane or a sphere. In particular, it is, usually, convenient to
choose a directrix which is, at each point, orthogonal to the ruling passing through that
point (in other words, the directrix is an orthogonal trajectory of the rulings. As we
shall see shortly, this particular choice of the ruling has as effect the diagonalization of
the first fundamental form of the surface.
The simplest ruled surfaces are the cylindrical surfaces, for which the rulings are
always parallel to a fixed direction and the conical surfaces, in the case of which the
rulings are passing through a fixed point. The ruled surfaces have been studied in details
186 Chapter 5. Special classes of surfaces
Figure 5.1: The hyperboloid with one sheet and its rulings
for the first time by the French geometer Gaspard Monge, at the end of the XVIIIth
century and most of the important results in the field belong to him or to some of his
students. In the figure 5.1.1 we represented the hyperboloid of rotation with one sheet,
together some of its rulings.
Another way of getting interesting ruled surfaces is to take as rulings the axes of
the Frenet frame of a space curve. For instance, in the figure 5.1.1 we represented the
surface described by the binormals of the Viviani’s temple.
The parameterization of a ruled surface
We assume that the directrix of the surface has a parameterization of the form
ρ = ρ(u), u ∈ I,
where I is an interval on the real axis. For each u ∈ I, we denote by b(u) the versor of the
ruling passing through the point ρ(u). Then, if M is a point on this ruling, its coordinates
will be determined by the relation
r = r(u, v) = ρ(u) + vb(u), (5.1.1)
where v is the parameter along the ruling, i.e. if M
0
= ρ(u), then we have
−−−−→
M
0
M = vb(u).
5.1. Ruled surfaces 187
Figure 5.2: The surface of binormals of the Viviani’s temple
The relation 5.1.1 provides a parameterization of the ruled surface,
r : I × R → S.
This parameterization is not, usually, global, as the parameterization of the directrix is
not global.
With respect to this parameterization, the rulings will be coordinate lines (u = const),
and the directrix is, equally, a coordinate line (v = 0). Generally, the coordinate lines
v = const have the property that they are “parallel” to the directrix, in the sense that all
the points from such a coordinate curve lie at the same distance (equal to |v|) from the
directrix, when we measure the distance along the ruling passing through each point.
The tangent plane and the first fundamental form of a ruled surface
To compute the coefficients of the first fundamental form of a ruled surface we need,
first of all, the partial derivatives of the radius vector of a point of the surface. We have,
obviously,
r
·
u
= ρ
·
+ bb
·
; r
·
v
= b
·
u
. (5.1.2)
188 Chapter 5. Special classes of surfaces
Thus, the coefficients of the first fundamental form of the surface will be
E ≡ r
·
u
r
·
u
= ρ
·2
+ 2vρ
·
b
·
+ v
2
b
·2
;
F ≡ r
·
u
r
·
v
= ρ
·
b;
G ≡ r
·
v
r
·
u
= 1.
(5.1.3)
It follows that the first fundamental form of a ruled surface can be written as:
ds
2
=

ρ
·2
+ 2vρ
·
b
·
+ v
2
b
·2

du
2
+ 2(ρ
·
b) dudv + dv
2
. (5.1.4)
To find the tangent plane at a point of a ruled surface, we notice, first of all, that the
direction of the normal to the plane (and, hence, to the surface) at a given point is given
by the vector r
·
u
× r
·
v
, i.e. by the vector
N ≡ r
·
u
× r
·
v
= ρ
·
× b + v

b
·
× b

. (5.1.5)
Therefore, if R is the position vector of a point from the tangent plane to the surface at
a point corresponding to the pair of parameters(u, v), then the equation of the tangent
plane can be written under the form
(R − r) N = 0,
i.e.
(R − ρ − vb)

ρ
·
× b + v

b
·
× b

= 0
or

R, ρ
·
, b

+ v

R, b
·
, b

ρ, ρ
·
, b

− v

ρ, b
·
, b

= 0
or, also,
,
R × ρ
·
+ v

R × b
·

− ρ × ρ
·
− v

ρ × b
·
¸
b = 0. (5.1.6)
A characteristic property of the ruled surfaces is described by the following proposi-
tion:
Proposition 5.1.1. The tangent planes to a ruled surface in point located along the same
ruling, belong to the pencil of planes determined by that ruling, or, to put it another way,
the tangent plane at a point of a ruled surface contains the ruling passing through that
point.
5.1. Ruled surfaces 189
Proof. Since the ruling and the tangent plane already have a point in common (the very
point of tangency), it is enough to prove that the ruling is parallel to the tangent plane
or, which is the same, that it is perpendicular to the normal to the surface at the tangency
point. We have, indeed,
N b =
,
ρ
·
× b + v

b
·
× b
¸
b =

ρ
·
, b, b
·

+ v

b
·
, b, b
·

= 0.

The proposition we just proved indicates how varies the tangent plane to a ruled
surface along a ruling: it just rotates around the ruling. In other words, its normal vector
stays all the time parallel to a fixed plane, which is perpendicular to the considered
ruling.
5.1.2 The Gaussian curvature of a ruled surface
We start, of course, from the parameterization (5.1.1). Then we have
r
·
u
= ρ
·
+ vb
·
, r
·
v
= b.
Therefore, for the first fundamental form, we will get
E = r
·
u
r
·
u
= (ρ
·
+ vb
·
) (ρ
·
+ vb
·
) = ρ
·2
+ 2vρ
·
b
·
+ v
2
b
·2
,
F = r
·
u
r
·
v
= (ρ
·
+ vb
·
) b = ρ
·
b,
while
G = r
·
v
r
·
v
= 1,
since b is a versor. Let H be the determinant of the first fundamental form. Then, for the
coefficients of the second fundamental form, we will get
D =
1
H

r
·
u
, r
·
v
, r
··
u
2

=
1
H

ρ
·
+ vb
·
, b, ρ
··
+ vb
··

,
D
·
=
1
H

r
·
u
, r
·
v
, r
··
uv

=
1
H

ρ
·
+ vb
·
, b, b
·

=
1
H

ρ
·
, b, b
·

,
D
··
=
1
H

r
·
u
, r
·
v
, r
··
v
2

= 0.
Thus,
190 Chapter 5. Special classes of surfaces
Proposition 5.1.2. The total curvature of a ruled surface given by the local parameter-
ization (5.1.1) is given by
K
t
=
DD
··
− D
·2
EG − F
2
= −

·
, b, b
·
)
2
|(ρ
·
+ vb
·
) × b|
2
. (5.1.7)
The curvature is always negative and it vanishes if and only if

ρ
·
, b, b
·

= 0. (5.1.8)
5.1.3 Envelope of a family of surfaces
In this section, a surface will always be a parameterized surface. The theory of envelopes
of surfaces is analogue to the theory of envelopes of plane curves, therefore there will
be no proofs, as the proofs are completely analogue to the ones we provided for plane
curves.
Definition 5.1.1. Let
r = r(u, v, λ) (5.1.9)
be a family of surfaces. The envelope of the family (if there is any) is a surface which is
tangent to all the surfaces from the family. The contact between the envelope and each
surface is made along a curve which is called a characteristic. Thus, the envelope is the
geometrical locus of the chracteristics of the family of surfaces.
Proposition 5.1.3. The points of the envelope of the family (5.1.9) verify, beside this
equation, the equation
(r
·
u
, r
·
v
, r
·
λ
) = 0. (5.1.10)
Of course, as in the case of plane curves, these equations are also verified by the
singular points of the surfaces and to get the envelope, we have to eliminate them first.
If the family of surfaces is given implicitly, i.e. through the equation
F(x, y, z, λ) = 0, (5.1.11)
then to get the envelope, we have to add to this equation the equation
F
·
λ
(x, y, z, λ) = 0, (5.1.12)
as we did in the case of plane curves. Something that is specific to the theory of envelopes
of surfaces is the so-called regression edge. Let us assume that we have a family of
5.1. Ruled surfaces 191
surfaces, given, say, implicitly. Then we can form the system:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
F(x, y, z, λ) = 0,
F
·
λ
(x, y, z, λ) = 0
F
··
λ
2
(x, y, z, λ) = 0
. (5.1.13)
The first two equations of this system are the equations of the family of characteristic
lines of the family of surfaces. If the system is compatible, then what we find solving it
is a curve which, at each point, is tangent to one of the characteristics, in other words,
it is the envelope of the characteristics. This envelope (when exists), it is called the
regression edge of the envelope of the family of surfaces.
An example of envelope. We consider the family of spheres of constant radius, with
the center of a given circle, verifying the hypothesis that the radius of the circle is greater
than the radius of the spheres. Then the envelope of this family is, as one can see imme-
diately, the torus (see the figure 5.1.3).
Figure 5.3: The torus as envelope of spheres
5.1.4 Developable surfaces
There is a subclass of ruled playing an important role in differential geometry, the so-
called developable surfaces:
Definition 5.1.2. A ruled surface is called developable if the tangent plane to the surface
is the same at all the points of a ruling.
We would like to find conditions for a ruled surface to be developable. We have the
following result:
192 Chapter 5. Special classes of surfaces
Proposition 5.1.4. A ruled surface given by the equation
r(u, v) = ρ(u) + v b(u)
is developable if and only if

ρ
·
, b, b
·

= 0. (5.1.14)
Proof. The invariance condition for the tangent plane along a ruling is equivalent to the
condition of invariance of the normal line to the surface along the same ruling. As we
saw previously, a normal vector to the surface is the vector
N(u, v) = ρ
·
(u) × b(u) + v

b
·
(u) × b(u)

.
Thus, the condition for the surface to be developable is that the direction N be indepen-
dent on the coordinate v. This may happen, obviously,in three situations:
1. The first component of the vector N vanishes, i.e.
ρ
·
(u) × b(u) = 0.
In this case, the vector N is constant only as direction, but its norm and orientation
might vary. As b is a versor, it cannot vanish, therefore the condition mentioned
above can be fulfilled in two situations:
(a) ρ
·
= 0. In this case, the directrix degenerates to a point, therefore the surface
is conical.
(b) ρ
·
| b. In this case, the rulings are nothing but the tangents to the directrix,
i.e. the surface is generated by the tangents to a space curve.
2. The second component of the normal vector vanishes, i.e.
b
·
(u) × b(u) = 0.
From this relation follows, in fact, that the direction of b is fixed, i.e. the surface
is cylindrical.
3. The two components of the normal vector are parallel, i.e. we have
ρ
·
(u) × b(u) | b
·
(u) × b(u).
5.1. Ruled surfaces 193
The three conditions we just listed are equivalent to the unique condition

ρ
·
(u) × b(u)

×

b
·
(u) × b(u)

= 0.
On the other hand,

ρ
·
× b

×

b
·
× b

=

ρ
·
, b
·
, b

b −

b, b
·
, b

....
=0
ρ
·
= −

ρ
·
, b, b
·

b,
which concludes the proof of the proposition.
Corollary 5.1.1. A ruled surface is developable if and only if its Gaussian curvature
vanishes identically.
Remark. The ruled surfaces which are not developable are also called scrolls.
Developable surfaces as envelopes of a one-parameter family of planes. The regres-
sion edge of a developable surface
We saw previously that that there are three classes of developable surfaces, namely:
1. cylindrical surfaces;
2. conical surfaces;
3. surfaces generated by the tangents to a space curve.
We shall see, in this paragraph, that, in fact, these three classes exhaust all the devel-
opable surfaces.
In the figure 5.1.4 we give an example of developable surface with a regression edge
(the tangent developable of the Viviani’s temple). One can notice immediately, from the
very definition of the developable ruled surfaces, that a ruled surface is developable if
and only if it is the envelope of a family of planes, depending on a single parameter
1
.
We consider a one-parameter family of planes
N(λ) r + D(λ) = 0. (5.1.15)
1
These planes are, of course, the tangent planes of the surface. Clearly, any surface is the envelope of
a family of planes, namely the family of tangent planes to the surface. Nevertheless, this family depends,
usually, on two parameters. In the case of developable surfaces, it depends on a single parameter, namely
the parameter along the directrix of the surface.
194 Chapter 5. Special classes of surfaces
Figure 5.4: The tangent developable of Viviani’s temple
We shall assume, without restricting the generality, that the normal vector N is a versor.
By differentiating the equation (5.1.15) with respect to λ, we obtain
N
·
(λ) r + D
·
(λ) = 0. (5.1.16)
According to the theory of envelopes of the families of surfaces, the equations (5.1.15)
and (5.1.16) determine the envelope of the family of planes, if such an envelope exists
2
.
The characteristic curves of the family of planes are straight lines, obtained as inter-
sections between the planes of equations (5.1.15) and (5.1.16), respectively. Clearly,
these characteristics exist only if the planes of the family are not parallel, which we
admit to be true in our case.
The points of the envelope have to verify, also, the equation obtained from (5.1.16),
by differentiating once more with respect to λ, i.e.
N
··
(λ) r + D
··
(λ) = 0. (5.1.17)
We have, thus, the following system of equations:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
N(λ) r + D(λ) = 0
N
·
(λ) r + D
·
(λ) = 0
N
··
(λ) r + D
··
(λ) = 0.
(5.1.18)
2
As a matter of fact, these equations describe the discriminant set, including, also, the singular points
of the surfaces from the family. However, in this particular case, the surfaces are planes and they have no
singular points whatsoever.
5.1. Ruled surfaces 195
We can regard this system as being a system of three equations with three unknowns (the
components of the radius vector r).
Let us assume, to begin with, that the system is compatible and determined, i.e. the
three planes intersect at a single point. We admit, at the first stage, that the solution of
the system depends in a nontrivial manner, on the parameter λ, in other words, if
r = ˜ r(λ) (5.1.19)
is the solution of the system, then ˜ r(λ) 0. This means, in fact, that the equation
(5.1.19) describes a space curve, called the edge of regression of the developable surface.
This is, of course, the edge of regression of the envelope of the family of planes, in the
sense of the theory of envelopes of surfaces. We know from this theory that all the
characteristic curves have to be tangent to the edge of regression. Therefore, we must
have
N ˜ r
·
= 0 (5.1.20)
ži
N
·
˜ r
·
= 0, (5.1.21)
because the characteristic is obtained by the intersections of the planes of normal vectors
N and N
·
, respectively.
By differentiating the relation (5.1.20), we obtain
N
·
˜ r
·
+ N ˜ r
··
= 0 (5.1.22)
whence, using (5.1.21),
N ˜ r
··
= 0. (5.1.23)
The relations (5.1.20) and (5.1.23) tell us that the plane of the normal vector N contains
both the vector ˜ r
·
, and the vector ˜ r
··
, i.e. is nothing but the osculating plane of the
regression edge at the point associated to the value λ of the parameter..
As such, a developable surface admitting an edge of regression can be described in
two different ways:
a) as the surface generated by the tangents to the edge of regression;
b) as the envelope of the family of osculating planes of the edge of regression.
In the degenerate case, when the edge of regression reduces to a point, all the character-
istic curves (i.e. the generators of the surface) are passing through that point, hence the
surface is a conical surface.
196 Chapter 5. Special classes of surfaces
We shall focus now on the case when the system (5.1.18) has a vanishing determi-
nant. In principle, two situations are possible: either the system is compatible and unde-
termined, either it is incompatible. We shall see that the first possibility is, in practice,
excluded.
We notice, first of all, that the determinant of the linear system (5.1.18) is nothing
but (N, N
·
, N
··
), therefore, the condition ∆ = 0 is equivalent to the condition

N, N
·
, N
··

= 0. (5.1.24)
This means, of course, that the vectors N, N
·
, N
··
lie in the same plane. We denote by
n = n(λ) the normal versor of this plane, in other words, we have
n N = n N
·
= n N
··
= 0. (5.1.25)
By differentiating the first two equations from the previous system, we obtain
n
·
N + n N
·
= n
·
N + n N
··
= 0,
whence, using (5.1.25),
n
·
N = n
·
N
·
= 0,
i.e. either the vector n
·
vanishes, either it is perpendicular on the vectors N ži N
·
.
Assuming that we would be in the second situation, as the vectors N and N
·
are not
colinear, we would obtain, as one can see easily, that n
·
is parallel to n, which is not
possible since, as, n is a versor, we have n n
·
= 0.
We must have, therefore, n
·
= 0, i.e. the vector n is constant. It follows from here
that for any λ the vectors N(λ) and N
·
(λ) are perpendiculars on the constant vector n,
i.e. all the rulings of the family are parallel to a constant direction, and the surface is a
cylindrical surface. It is easy to see that in this case the system of equations (5.1.18) is
incompatible (the third plane is parallel to the straight line determined by the first two).
The considerations from this paragraph prove that, in fact, we have the following
result::
Proposition 5.1.5. If S is a developable ruled surface, then it is either a cylindrical
surface, either a conical one, either a surface generated by the tangents to a space
curve.
Thus, the classes of surfaces mentioned in the previous paragraph do exhaust all the
types of developable ruled surfaces. We mention that the plane can be considered as a
limit case for any of the three classes of surfaces listed in the proposition.
5.1. Ruled surfaces 197
5.1.5 Developable surfaces associated to the Frenet frame of a space curve
In all of this section r = r(s) will be a smooth naturally parameterized biregular space
curve. We assume, moreover, that the curve is skew, in other words, its torsion never
vanishes.
The envelope of the family of osculating planes
The equation of osculating planes at a point of the curve is
(R − r(s)) β(s) = 0. (5.1.26)
The parameter along the family of planes will be exactly the natural parameter s along
the curve.
By differentiating the relation (5.1.26) with respect to s, we get
−r
·
(s) β(s) + (R − r(s)) β
·
(s) = 0
or, having in mind that r
·
β = 0 and using the third formula of Frenet,
(R − r(s)) ν(s) = 0, (5.1.27)
where we used the fact that, according to the hypotheses we made about the curve, the
torsion of the curve never vanishes.
From (5.1.26) and (5.1.27) one obtains that the characteristic straight line is perpen-
dicular both on β and on ν, in other words, it is nothing but the tangent to the curve
.
To get the edge of regression, we differentiate once more (5.1.27) and we get
−r
·
(s) ν(s) + (R − r(s)) ν
·
(s) = 0
or, using the fact that r
·
(s) ν(s) = 0,a s well as the second Frenet formula,
(R − r(s))
,
χ(s)β(s) − kν(s)
¸
= 0.
As χ(s)β(s) − kν(s) 0 if χ and k don’t vanish, we obtain that
R = r(s), (5.1.28)
i.e. the edge of regression is nothing but the given curve.
198 Chapter 5. Special classes of surfaces
The envelope of the family of normal planes (the polar surface)
The equation of the normal plane at an arbitrary point of a space curve is, as we know,
(R − r(s)) τ(s) = 0. (5.1.29)
We differentiate this relation once, to complete the system of equations for the charac-
teristics and we get
−r
·
(s) τ(s) + (R − r(s)) τ
·
(s) = 0
or, having in mind that we have a curve parameterized by the arc length and using the
first of the Frenet formulae,
−1 + (R − r(s)) k(s) ν(s) = 0. (5.1.30)
It follows, then, that
(R − r(s)) ν(s) =
1
k(s)
. (5.1.31)
The equation (5.1.31) is the equation of a plane which is parallel to the rectifying plane
(because it has as normal vector the versor of the principal normal). Thus, the charac-
teristic of the envelope of the normal planes passing through the point of position vector
r(s) is obtained by intersecting the normal plane to the curve at this point with a plane
which is parallel to the rectifying plane of the curve at the same point, i.e. it is a straight
line which is parallel to the binormal of the curve at the point r(s).
On the other hand, it is clear that this characteristic passes through the point of
position vector r(s)+
1
k(s)
ν(s), which is exactly the center of curvature of the curve. Thus,
the characteristics of the family of normal planes are nothing but the axes of curvature
(or polar axes) of the curve. This is the reason why the envelope of the family of normal
planes is called, sometimes, the polar surface of the curve.
To get the edge of regression of the surface, we differentiate once more the relation
(5.1.30) and we get
−r
·
(s) k(s) ν(s)
....
=0
+(R − r(s)) k
·
(s) ν(s) + (R − r(s)) k(s) ν
·
(s) = 0
or, using (5.1.31) and the second Frenet formula,
k
·
(s)
k(s)
+ (R − r(s)) k(s)
,
χ(s) β(s) − k(s)τ(s)
¸
= 0
or, also, from (5.1.29),
(R − r(s)) β(s) =
k
·
(s)
k
2
(s) χ(s)
. (5.1.32)
Comparing (5.1.29) to (5.1.32), we find out the following things:
5.1. Ruled surfaces 199
• The straight line of intersection between the planes (5.1.29) and (5.1.32) is parallel
to the osculating plane to the rectifying plane and the normal plane, i.e. is parallel
to the binormal.
• This straight line passes through the point
r(s) +
k
·
(s)
k
2
(s) χ(s)
β(s).
Thus, the current point from the edge of regression is obtained by intersecting the lines
R(λ) = r(s) +
1
k(s)
ν(s) + λβ(s) (5.1.33)
and
R(µ) = r(s) +
k
·
(s)
k
2
(s)χ(s)
β(s) + µν(s). (5.1.34)
This point will have, thus, the position vector
R(s) = r(s) +
1
k(s)
ν(s) +
k
·
(s)
k
2
(s)χ(s)
β(s). (5.1.35)
But the right hand side of the relation (5.1.35) is nothing but the position vector of the
osculating sphere of the given curve. Thus, the edge of regression of the polar surface of
a space curve is the locus of the centers of the osculating spheres of the given curve.
The envelope of the family of rectifying planes of a space curves
The equation of the rectifying plane is
(R − r(s)) ν(s) = 0. (5.1.36)
By differentiation, we get
−r
·
(s) ν(s)
....
=0
+(R − r(s)) ν
·
(s) = 0
or, using the second Frenet equation,
(R − r(s))
,
χ(s)β(s) − k(s)τ(s)
¸
= 0. (5.1.37)
The points of the curve verify the equations (5.1.36) and (5.1.37), hence the curve lies
on the rectifying surface and can be takes as the directrix of the surface, regarded as
200 Chapter 5. Special classes of surfaces
a ruled surface. The relations (5.1.36) and (5.1.37) indicate that the rulings (i.e. the
characteristic lines of the envelope) are perpendicular both on ν, and on ν
·
, hence their
direction is given by the vector
ν × ν
·
= ν × (χβ − kτ) = χ(ν × β) − k(ν × τ) = χτ + kβ = δ,
where δ is the Darboux vector of the curve. Thus, the characteristics of the rectifying
surface of a space curve are the Darboux axes of the curve or the instantaneous axes of
rotation of the Frenet frame.
The tangent plane to the rectifying surface at a point lying on the given curve contain
both the tangent vector of the curve, r
·
= τ, and the Darboux vector of the curve, δ.
Therefore, the normal to the surface at such a point is colinear to the vector
τ × δ = τ × (χτ + kβ) = k(τ × β) = −kν.
Thus, the normal to the rectifying surface of a space curve at a point of the curve coin-
cides to the principal normal of the curve. This actually means that the osculating plane
of the curve is perpendicular on the tangent space of the surface, i.e. the curve is a geo-
desic of its rectifying surface. This is, actually, the origin of the name of the rectifying
plane and, implicitly, of the rectifying surface: the curve, when regarded as lying on its
rectifying surface, is a geodesic, which is the surface analogue of a straight line. In other
words, the curve is rectified (straightened) by its rectifying planes.
As we saw, the total of a developable surface is always zero. In many books of
differential geometry it is claimed that the converse is, also, true, in other words, any
surface of zero Gaussian curvature is a developable surface or a part of a developable
surface. However, this is true only if we assume that the surface doesn’t have any planar
points, otherwise, first of all, it might not be not even a ruled surface. An example of such
surface is given in the book of Klingenberg [31]. The surface is not, actually, described
by using a parameterization or an implicit equation, but rather by prescribing the first
two fundamental forms and showing that they fulfill the compatibility conditions.
5.2 Minimal surfaces
5.2.1 Definition and general properties
Definition 5.2.1. A minimal surface is, by definition, a surface whose mean curvature
vanishes identically.
This is not the original definition of a minimal surface. As a matter of fact, Lagrange
extended to the two-dimensional case the method of Euler for finding the extrema of a
functional and he proposed the following problem:
5.2. Minimal surfaces 201
Given a closed curve and a connected surface patch S bounded by the curve, to finde
the surface such that the inclosed area shall be minimum.
We consider, thus, the area functional,
A =

U
Hdudv.
As it is known from the variational calculus, a necessary condition for the minimum of
this functional is furnished by the Lagrange equations, which, in this case, are:

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
∂H
∂x


∂u

∂H
∂x
·
u



∂v

∂H
∂x
·
v

= 0
∂H
∂y


∂u

∂H
∂y
·
u



∂v

∂H
∂y
·
v

= 0
∂H
∂z


∂u

∂H
∂z
·
u



∂v

∂H
∂z
·
v

= 0
.
After long, but otherwise straightforward computations, the Lagrange equations be-
come

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸

y
·
u
z
·
u
y
·
v
z
·
v

K
m
= 0

x
·
u
z
·
u
x
·
v
z
·
v

K
m
= 0

x
·
u
y
·
u
x
·
v
y
·
v

K
m
= 0
,
and, thus, they are satisfied iff K
m
= 0.
Let us mention, also, that, initially, Lagrange used the explicit representation z =
f (x, y) for the surface, in which case the area functional is

U

1 + p
2
+ q
2
dxdy,
while the stationarity condition finally becomes

∂x
p

1 + p
2
+ q
2
+

∂y
p

1 + p
2
+ q
2
= 0
or
(1 + q
2
)r − 2pqs + (1 + p
2
)t = 0.
It was Meusnier who showed, sixteen years later, that the minimality condition induces
a relation among the coefficients of the first two fundamental forms of a parameterized
202 Chapter 5. Special classes of surfaces
surface. Eisenhart claims that Meusnier actually showed that for minimal surfaces the
mean curvature vanishes. This is not quite correct: Meusnier was not aware of the
true meaning of the the relation he found. In fact, the notion of mean curvature was
introduced only fifty years later, by Sophie Germain.
An alternative proof of the fact that the minimal surfaces are the solution of the
variational problem was found by Darboux. A particularly clear exposition of this proof
can be found in the book of Hsiung.
For further convenience, we shall need the following definition:
Definition 5.2.2. A local parameterization (U, r) of a surface S is called isothermic if,
with respect to this parameterization, the coordinate lines are orthogonal (i.e. F = 0)
and E = G. This means, actually, that the matrix of the first fundamental form is just the
unit matrix multiplied by a positive function.
On any surface which is at least C
2
there exist isothermic parameterizations. The
proof of this claim is based on the theory of partial differential equations and it is not
interesting for our exposition.
Proposition 5.2.1. A surface is minimal iff its asymptotic directions are orthogonal.
Proof. A surface is minimal iff the coefficients of the first two fundamental forms verify
the equation
ED
··
− 2FD
·
+ GD = 0.
Let us choose, for the surface, an isothermic parameterization, with respect to which we
have
E = G = λ
2
(u, v), F = 0.
Then the minimality condition for the surface becomes
E(D
··
+ D) = 0
and, as E 0, the surface being regular, we have
D
·
+ D
··
= 0.
On the other hand, the equation for the slopes of the asymptotic directions is
Dt
2
+ 2D
·
t + D
··
= 0
and one can see immediately that the asymptotic directions are orthogonal iff the roots
of this equation verify t
1
t
2
= −1, which is equivalent to D = −D
··
.
5.2. Minimal surfaces 203
The following technical lemma will allow us to obtain a very nice characterization
of minimal surfaces
Lemma 1. Let r = r(u, v) be an isothermic parameterization of a surface S . Then
r
··
u
2
+ r
··
v
2
= 2λ
2
K
m
n, (5.2.1)
where λ
2
= E = G.
Proof. Since r is isothermic, we have
r
·
u
2
= r
·
v
2
= λ
2
, r
·
u
r
·
v
= 0, (5.2.2)
therefore the mean curvature is just
K
m
=
D + D
··
λ
2
. (5.2.3)
By differentiating (5.2.2), we get
r
··
u
2
r
·
u
= r
··
uv
r
·
v
= −r
·
u
r
··
v
2
,
whence
(r
··
u
2
+ r
··
v
2
) r
·
u
= 0,
and, similarly,
(r
··
u
2
+ r
··
v
2
) r
·
v
= 0.
Thus, the vector r
··
u
2
+r
··
v
2
is perpendicular both to r
·
u
and r
·
v
, i.e. is colinear with the unit
normal:
r
··
u
2
+ r
··
v
2
= a n. (5.2.4)
If we multiply (5.2.4) by n and use the equation (5.2.2), as well as the definition of the
second fundamental form, we obtain
a = D + D
··
= 2λ
2
K
m
, (5.2.5)
which proves the claim.
Corollary 1. if r = r(u, v) is an isothermic parameterization of a surface S , then S is
minimal iff the components of r are harmonical functions, in other words iff we have
∆x = ∆y = ∆z = 0.
204 Chapter 5. Special classes of surfaces
It is a classical (but nontrivial) result that on any smooth (at least C
2
) surface there
exists isothermic parameterization. On a minimal surface we have a very special isother-
mic parameterization:
Proposition 5.2.2. The lines of curvature on a minimal surface form an isothermic sys-
tem.
Proof. We choose as parametric lines the lines of curvature of the surface. Then, as we
already know, we should have F = D
·
= 0 (both first fundamental forms are diagonal in
such a kind of parameterization). Thus, the minimality condition for the surface becomes
ED
··
+ GD = 0
In this parameterization the Codazzi-Mainardi equations become
∂D
∂v

1
2
¸
D
E
+
D
··
G

= 0,
∂D
··
∂u

1
2
¸
D
E
+
D
··
G

= 0,
which, for a minimal surface, reduce to
∂D
∂v
= 0,
∂D
··
∂u
= 0,
i.e. D depends only on u and D
··
– only on v. Since
E
D
= −
G
D
··
,
we have

2
log E
∂u∂v
=

2
logG
∂u∂v
,
which is exactly the condition that the parametric lines are isothermic.
Proposition 5.2.3. The spherical image of a minimal surface is conformal (i.e. it pre-
serves the angles).
5.2. Minimal surfaces 205
Proof. Let e, f , g be the coefficients of the first fundamental form of the spherical image.
We have, then,
e = −EK
t
+ DK
m
, f = −FK
t
+ D
·
K
m
, g = −GK
t
+ D
··
K
m
.
Thus, for a minimal surface, we have
e = −EK
t
, f = −FK
t
, g = −GK
t
,
i.e. the two surfaces are, indeed, conformally equivalent.
The reciprocal of this proposition is almost true, i.e. we have
Proposition 5.2.4. If the spherical map of a surface is conformal, then the surface is
either minimal, either a sphere.
Proof. If the spherical map is conformal, then we have
−EK
t
+ DK
m
= µE, −FK
t
+ D
·
K
m
= µF, −GK
t
+ D
··
K
m
= µG. (5.2.6)
Thus,
DK
m
= E(K
t
+ µ), D
·
K
m
= F(K
t
+ µ), D
··
K
m
= G(K
t
+ µ). (5.2.7)
We multiply the three relations by D
··
, −2D
·
, D, respectively and we add the results, we
get
2K
t
K
m
= K
m
(K
t
+ µ)
or
K
m
(µ − K
t
) = 0. (5.2.8)
On the other hand, if we multiply the relations (5.2.7) by G, −2F, E, respectively, we get
K
2
m
= 2(K
t
+ µ). (5.2.9)
If we put K
m
, the (5.2.8) is satisfied, while (5.2.9) is satisfied iff µ = −K
t
and we redis-
cover the result from the proof of the previous proposition, i.e. the spherical image of a
minimal surface is conformal and the conformal factor is

−K
t
. The second possibility
to satisfy (5.2.8) is K
t
= µ. Notice that, as µ > 0, we cannot have at the same time K
t
= µ
and K
m
= 0, because then we would get, also, K
t
= −µ, i.e. K
t
= µ = 0.
If K
t
= µ, then, from (5.2.9) we get
K
2
m
= 4µ = 4K
t
,
i.e. the principal curvatures are equal at each point (all the points are umbilical). Or, as
we know, the only surface with only umbilical points is the sphere (and its limit case,
the plane).
206 Chapter 5. Special classes of surfaces
Proposition 5.2.5. The spherical image of isothermic lines on a minimal surface are
isothermic lines on the sphere.
Proof. This is an immediate consequence of the fact that the spherical map of a minimal
surface is conformal.
The first two minimal surfaces (apart from the plane) were discovered in the eigh-
teenth century by the French mathematician Meusnier. These surface were the catenoid
and the right helicoid. As we know, the first surface is a revolution surface, while the
second is a ruled surface. For a couple of decades, nobody was able to find other ex-
amples. The surfaces found by Meusnier are very simple and they belong classes of
surfaces which are relatively easy to study, but, as we shall see in the following, these
classes do not contain other minimal surfaces.
5.2.2 Minimal surfaces of revolution
Proposition 5.2.6. The catenoid is the only revolution surface which is, also, a minimal
surface.
Proof. An arbitrary surface of revolution (around the z-axis, in our case), can be para-
meterized as

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v
y = u sin v
z = f (u)
.
Therefore, the coefficients of the first two fundamental forms of the surface can be writ-
ten as
E = 1 + f
·2
, F = 0, G = u
2
,
D =
f
··

1 + f
·
2
, , D
·
= 0, D
··
=
uf
·

1 + f
·
2
.
Then the minimality condition for the surface reads
(1 + f
·2
)uf
·
+ f
··
u
2
= 0
which can be integrated and gives for f an expression of the form
f (u) = a cosh
−1
u
a
+ c,
where a and c are constants, which corresponds to the catenoid.
5.2. Minimal surfaces 207
Figure 5.5: The catenoid
5.2.3 Ruled minimal surfaces
We recall that the right conoid is the surface generated by s straight line moving parallel
to a plane (z = 0 in our case) and intersecting a line which is perpendicular to this plane
(in our case – the z-axis). As we saw, the equations of a right conoid can be written as

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v
y = u sin v
z = f (v)
.
Then the coefficients of the first two fundamental forms of the surface will be
E = 1, F = 0, G = u
2
+ f
·2
,
D = 0, D
·
= −
f
·

u
2
+ f
·
2
, D
··
=
u f
··

u
2
+ f
·
2
.
We are interested to see which right conoids are minimal surfaces. Well, as we will
show, there is only one.
Proposition 5.2.7. The only right conoid which is a minimal surface is the right helicoid.
Proof. As we know, the condition for a surface to be minimal is that the coefficients of
the first two fundamental forms verify the equation
ED
··
− 2FD
·
+ GD = 0,
208 Chapter 5. Special classes of surfaces
Figure 5.6: The minimal helicoid
which, in our case, reduces to
uf
··

u
2
+ f
·
2
= 0,
or f
··
= 0. But this leads to f (v) = av + b, where a and b are constants, i.e. the surface
is a right helicoid.
One might think that there are other ruled minimal surfaces, which are not right
conoids. For such more general ruled surfaces we do not have such a nice parameteriza-
tion, therefore this direct approach cannot be applied. Still, as we shall see next, there is
another method to prove that, in fact, there are no other ruled minimal surfaces.
Proposition 5.2.8 (Catalan, 1842). The only minimal ruled surface is the right helicoid.
Proof. Let S be a ruled surface which is not a plane. Then, as we saw previously,
the surface is minimal iff the asymptotic lines are orthogonal. Through each point of
the surface pass exactly two asymptotic lines. On the other hand, through each point
passes one straight line (the generator). To prove that the surface is a right helicoid, it
is enough to prove that it has as directrix a circular cylindrical helix. Since the surface
is minimal, the second asymptotic line through each point is orthogonal to the generator
passing through that point. The asymptotic lines have the property that their principal
normals are contained into the tangent plane. This means that for an asymptotic line
from the second family, the generators that intersect it are principal normals. But they
are principal normals also for any other asymptotic line that intersects them. Thus, any
5.2. Minimal surfaces 209
asymptotic line from the second family has an infinity of Bertrand mates. But then, as
we know, they have to be cylindrical helices.
We shall give now, following Barbosa and Colares, a more direct prove of the Cata-
lan’s theorem, using a local parameterization of the ruled surface.
Another proof of the Catalan’s theorem. Let S ⊂ R
3
be a ruled surface. Then from the
definition, S has local parameterizations of the form
r(u, v) = α(u) + vγ.
We may choose the directrix α of the ruled surface to be an orthogonal trajectory of the
family of rulings, and γ, the directing vector of the rulings, to be a unit vector, at each
point of the directrix. We shall assume, moreover, that u is the natural parameter of the
directrix. A straightforward computation provides, for the coefficients of the first two
fundamental forms of the surface, the expressions:
E = 1 + 2vα
·
γ
·
+ v
2
γ
·2
,
F = 0,
G = 1,
D =
1

1 + 2vα
·
γ
·
+ v
2
γ
·2

·
+ vγ
·
, γ, α
··
+ vγ
··
),
D
·
=
1

1 + 2vα
·
γ
·
+ v
2
γ
·2

·
+ vγ
·
, γ, γ
·
),
D
··
= 0.
Thus, the minimality condition reads

·
+ vγ
·
, γ, α
··
+ vγ
··
) = 0,
or, which is the same,

·
× γ) α
··
+ v
,

·
× γ) γ
··
+ (γ
·
× γ) α
··
¸
+ v
2

·
× γ) γ
··
= 0.
As the left hand side of the previous equation is a polynomial function in v, the equality
holds for any v iff the coefficients of the polynomial function vanish simultaneously:

·
× γ) α
··
= 0 (a)

·
× γ) γ
··
+ (γ
·
× γ) α
··
= 0 (b)

·
× γ) γ
··
= 0. (c)
210 Chapter 5. Special classes of surfaces
From the relation (a) it follows that the vector α
··
must be contained in the plane gener-
ated by the vectors α
·
and γ. But, as the curve α is naturally parameterized, the vector
α
·
has unit length, which means that α
··
is perpendicular to α
·
. As, from construction,
γ is also perpendicular to α
·
, it follows that α
··
is colinear to γ. But the, the relation (b)
reduces to

·
× γ) γ
··
= 0. (b’)
Now, from (b’) and (c) follows that the vector γ
··
should belong to the intersection of the
planes determined by the vectors α
·
, γ and γ
·
, γ, respectively. Clearly, this two planes
are not parallel, since they contain at least the vector γ. Therefore, only two situations
are possible: either their intersection is a straight line, either they coincide. Suppose,
first, that there exists a point where γ
··
is not parallel to γ. Then, at this point (and,
by continuity, in an entire neighborhood of it), these planes have to coincide. Indeed,
otherwise γ
··
cannot belong simultaneously to both planes. Moreover, in this situation
we have, also α
·
| γ
·
. Indeed, both γ
·
and α
·
are perpendicular on γ. As the three
vectors are coplanar, it follows the parallelism just mentioned. Now, as we assumed
that α and γ are not just smooth, but analytic functions, the two vectors are parallel for
any value of the parameter, and, also, the two planes will also coincide. The plane that
contains all our vectors could, in principle, vary from point to point. In reality, instead,
this not happens. Indeed, we have
(γ × α
·
)
·
= γ
·
× α
·
+ γ × α
··
= 0.
But this means that the plane of the vectors is invariant, therefore the curve α is a plane
curve and the surface is a plane.
Suppose, now, that γ
··
is always parallel to γ. If α
·
is always parallel to γ
·
, this
means, actually, that the two planes always coincide and we are, in fact, in the previous
situation. Let us assume, therefore, that there exists a point (hence an entire open set)
where the two vectors are not parallel. We are going to show that, in this case, α is a
circular helix. As we know, to this end it is enough to prove that both the curvature and
the torsion of the curve are constants.
Let ¦τ, ν, β¦ the Frenet frame of the curve α. We intend to use the Frenet equation
to find the curvature and the torsion, instead of using the explicit formulas we obtained
when we studied the space curves. First of all, as α is naturally parameterized, we have
α
··
(u) = k(u)ν(u).
On the other hand, again, because α is naturally parameterized, we have α
··
(u) ⊥ α
·
(u).
But α
·
is also perpendicular on γ, and the three vectors are coplanar, which means that,
5.2. Minimal surfaces 211
as γ is a unit vector, we have γ
·
(u) = ±ν, hence, using the previous equation,
±k(u) = α
··
(u) γ(u).
But, since α
·
⊥ γ, we have
0 = (α
·
× γ)
·
= α
·
× γ
·
+ α
··
× γ,
therefore, we have for the curvature the expression
±k(u) = −α
·
× γ
·
.
Thus, the derivative of the curvature will be given by
±
dk
du
= −(α
·
× γ
·
)
·
= −α
··
× γ
·
− α
·
× γ
··
= 0,
i.e. the curvature of α is constant. The torsion, on the other hand, can be found using the
third Frenet’s formula. Indeed, this formula is
β
·
= −χν.
As we saw, ν = ±γ. On the other hand, we have τ = α
·
and, thus,
β ≡ τ × ν = ±α
·
× γ,
whence
β
·
= ±(α
·
× γ)
·
= ±(α
··
× γ + α
·
× γ
·
) = ±α
·
× γ
·
.
Hence the torsion will be given by
χ = −β ν = ±(α
·
× γ
·
) γ
and it is easy to see that the derivative of the torsion is, also, identically zero, i.e. the
torsion is a constant function.
Now, from the existence and uniqueness theorem for space curves, we know that
there exists a curve which has the prescribed curvature and torsion and this curve is
unique, up to a motion in space. But, on the other hand, we know a curve which has
constant curvature and torsion, namely the circular helix. Thus, α should be a circular
helix and has a parameterization of the form
α(u) = (Acos au, Asin au, bu),
212 Chapter 5. Special classes of surfaces
where, since α is supposed to be naturally parameterized, the constants A, a, b are subject
to A
2
a
2
+ b
2
= 1. On the other hand, as γ is parallel to α
··
, we have, also,
γ = ±¦cos au, sin au, 0¦.
If, now, we put s = A ± u, t = v, the equation of the surface, with α and γ just found,
becomes
r(s, t) = (s cos at, s sin at, bt),
which means that the surface is a part of a right helicoid, as claimed.
Consequence. The only developable minimal surface is the plane.
As we already mentioned, Meusnier was the one who discovered that the right heli-
coid is a minimal surface. It is both interested and instructive to see how Meusnier got
to this result. He searched a minimal surface in an explicit representation, z = f (x, y).
As we saw, in this case, the minimality condition can be written as
(1 + f
·2
y
) f
··
x
2
− 2 f
·
x
f
·
y f
··
xy
+ (1 + f
·2
x
) f
··
y
2
= 0.
Meusnier was interested in solutions of the previous equation for which the level curves
f (x, y) = const are straight lines in the plane. Obviously, such surfaces would be ruled.
He started by noticing that, for a plane curve given in the implicit form f (x, y) = c, the
signed curvature can be written as
k
±
=
−f
··
x
2
f
·2
y
+ 2 f
·
x
f
·
y
f
··
xy
− f
··
y
2
f
·2
x
| grad f |
3
.
Therefore, the minimality condition becomes
∆f = k
±
| grad f |
3
.
If the level curves are straight lines, then their signed curvature vanishes, which means
that f is a harmonical function, i.e.
∆f ≡

2
f
∂x
2
+

2
f
∂y
2
= 0.
Now, it can be shown that the only solution of this equation for which the level curves
are straight lines is
f (x, y) = a arctg
y − y
0
z − z
0
+ b,
5.3. Surfaces of constant curvature 213
with a, b, x
0
, y
0
constants.
The graph of the function z = f (x, y) is, then, either a a plane (if a = 0), either a
portion of the right helicoid

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = x
0
+ u cos v
y = y
0
+ u sin v
z = av + b.
As one can see, even the original deduction of Meusnier could make one suspect that the
only ruled minimal surface is the right helicoid.
5.3 Surfaces of constant curvature
Definition 5.3.1. We say that a surface S has constant curvature if its Gaussian curvature
is independent of the point, i.e. is a constant.
We already met surfaces which have zero curvature, namely the plane and the de-
velopable surfaces. We also know an example of a surface that has constant positive
curvature (the sphere). We shall give, now, an example of a (parameterized) surface
that has constant negative curvature, namely the pseudosphere, given by the parametric
equations

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a sin u cos v
y = a sin u sin v
z = a

ln tan
u
2
+ cos u

. (5.3.1)
A straightforward computation shows that the pseudosphere has the Gaussian curvature
−1/a
2
. Loosely, we can define a surface to be complete if it is unextendible (like the
sphere, or the plane). It turns out that any surface of constant negative curvature, if
extended enough, becomes singular. In other words, there are no non-singular complete
surfaces of constant negative curvature in R
3
. The pseudosphere (see the figure 5.3) is
no exception. A classical theorem of Minding (1839) claims that all the surfaces that
have the same constant curvature are locally isometric. This means that they have local
parameterizations, defined on the same domain in the plane, in such a way that, for a
given pair of parameters, they have the same coefficients for the first fundamental form.
This result does not hold, however, if the surfaces don’t have constant curvature, as the
following counterexample shows.
Counterexample. Let r
1
: D
1
≡ (0, ∞) × (0, 2π) → R
3
, given by
r
1
(u
1
, v
1
) = (u
1
cos v
1
, u
1
sin v
1
, ln u
1
)
214 Chapter 5. Special classes of surfaces
Figure 5.7: The pseudosphere
and r
2
: D
2
≡ (0, ∞) × (0, 2π) → R
3
, given by
r
2
(u
2
, v
2
) = (u
2
cos v
2
, u
2
sin v
2
, v
2
) .
As one can see, the first surface is a surface of revolution, generated by the graph of
the logarithmic function, while the second one is part of the right helicoid. We consider
the map Λ : D
1
→ D
2
, Λ(u
1
, v
1
) = (u
2
, v
2
). As D
1
= D
2
, clearly Λ is, in fact, the
identity and it is, therefore, a diffeomorphism. An easy computation shows that the first
fundamental forms for the two parameterized surfaces are, respectively:
E
1
= 1 +
1
u
2
1
, F
1
= 0, G
1
= u
2
1
and
E
2
= 1, F
2
= 0, G
2
= 1 + u
2
2
,
therefore they are not isometric (because, for instance, E
1
E
2
). However, as one can
convince oneself, for their Gaussian curvatures we get
K
1
= −
1
(1 + u
2
1
)
2
, K
2
= −
1
(1 + u
2
2
)
2
,
i.e. they are equal.
On the other hand, even if, for instance, the surfaces of the same constant negative
curvature are locally isometric with the pseudosphere, they might look very different, as
the surfaces from the figure 5.3, belonging to Kuen and Dini show.
5.3. Surfaces of constant curvature 215
(a) Surface of revolution (b) Helicoid
Figure 5.8: Non-isometric parameterized surfaces
(a) Kuen’s surface (b) Dini’s surface
Figure 5.9: Surfaces of constant negative curvature
216 Chapter 5. Special classes of surfaces
Problems
1. We consider, in the coordinate plane xOz, the curve given by

¸
¸
¸
¸
¸
¸
x = f (u),
z = ϕ(u).
Find the parametrical equation of the surface obtained rotating this curve around the
z-axis.
2. Find the equations of the torus, obtained by rotating the circle

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a + cos u,
y = 0,
z = b sin u, b < a,
around the z-axis.
3. Find the equations of the catenoid, obtained by rotating the catenary

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a cosh
u
a
,
y = 0,
z = u
around the z-axis.
218 Problems
4. Find the equations of the pseudosphere, obtained by rotating the tractrix

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a sin u,
y = 0,
z = a

ln tan
u
2
+ cos u

around the z-axis.
5. Find the equation of the tangent plane at an arbitrary point of the right helicoid
r = ¦u cos v, u sin v, hv¦.
6. Show that if a smooth surface Φ and a plane α have a single common point P, then
the plane α is the tangent plane to the surface at the point P.
7. Write the equation of the tangent plane to the sphere of radius a, centred at the origin,
at the point of coordinates (0, 0, a) (“north pole”).
8. Write the equations of the normal to the pseudosphere

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a sin u cos v,
y = a sin u sin v,
z = a ln tan
u
2
+ a cos u
at an arbitrary point and find the unit normal vector.
9. Find the unit normal vector to the right helicoid

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v
y = u sin v
z = hv
.
10. Determine the equation of the tangent plane to the surface xy
2
+ z
2
= 8 at the point
(1, 2, 2) and find the unit normal vector at this point.
11. Show that the surfaces
x
2
+ y
2
+ z
2
= αx,
x
2
+ y
2
+ z
2
= βy
and
x
2
+ y
2
+ z
2
= γz
are pairwise orthogonal to each other.
Problems 219
12. Show that the normal to a surface of rotation at an arbitrary point passes through the
rotation axis.
13. Show that the tangent plane at an arbitrary point of the surface z = xϕ

y
x

, where ϕ
is a smooth function of a single real variable, passes through the origin of coordinates.
14. Show that the tangent plane to a surface generated by the tangents to a space curve
is the same at all the points of a given generator.
15. Write the equation of the tangent plane and the equations of the normal at the surface
generated by the binormals of a curve ρ = ρ(s).
16. The surface Φ is generated by the binormals of a curve γ. Show that at the points
of the curve γ the tangent plane to Φ coincide with the osculating plane to γ, while the
normal to the surface is the principal normal to the curve γ.
17. A surface is generated by the principal normals of a curve γ. Establish the equation
of the surface. Show that at the points of the curve γ the tangent plane to the surface
coincide with the osculating plane to the curve γ.
18. On the normals to a surface Φ, in the same direction, there are taken segments
of the same length, measured from the surface. The ends of these segments describe
another surface, Φ

, “parallel” to the surface Φ. Show that the surfaces Φ and Φ

have,
at corresponding points, common normals and parallel tangent planes.
19. Study the variation of the sign of the function r n for the torus, where r is the
radius vector of a point of the torus, while n is the normal to the torus at that point. Find
the geometrical locus of the points from the torus at which r n = 0 (the origin of the
coordinates is taken at the centre of the torus).
20. Find a close surface on which the function r n is a constant.
21. Find the points of the torus
ρ = ¦(R + r cos u) cos v, (R + r cos u) sin v, r sin u¦
at which its normal is parallel to the vector a = ¦l, m, n¦
22. Write the equation of the tangent plane to the torus

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = (3 + 2 cos u) cos v,
y = x = (3 + 2 cos u) sin v,
z = 2 sin u,
parallel to the plane x + y +

2z + 5 = 0.
220 Problems
23. Find the tangent plane to the surface z = x
4
− 2xy
3
, perpendicular to the vector
a = ¦−2, 6, 1¦. Find, also, the tangency point.
24. Find the envelope and the regression edge of the family of spheres of constant radius,
equal to a, with the centers on the circle

¸
¸
¸
¸
¸
¸
x
2
+ y
2
= b
2
,
z = 0.
25. Find the envelope of the family of spheres
x
2
+ y
2
+ (z − C)
2
= 1.
26. Find the envelope, the characteristics and the regression edge for a family of spheres
of constant radius a, with centers on a given curve ρ = ρ(s) (tubular surface).
27. Find the envelopes and the regression edge for a family of spheres passing through
the origin of the coordinates and with the centers on the curve

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = t
3
,
y = t
2
,
z = t.
28. Find the envelope of a family of planes forming with the coordinate solid angle
x > 0, y > 0, z > 0 a triangular pyramid of constant volume V.
29. Find the envelope and the regression edge for the family of normal planes to the
helix

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = a cos t,
y = a sin t,
z = bt.
30. Find the envelopes, the characteristics and the regression edge for the family of
normal planes to a space curve ρ = ρ(s) (the polar surface of the curve).
31. Find the envelope, the characteristics and the regression edge for the family of rec-
tifying planes of a space curve ρ = ρ(s).
32. Find the envelope, the characteristics and the regression edge of the family of oscu-
lating planes of a space curve ρ = ρ(s).
Problems 221
33. Find the envelope, the characteristics and the regression edge of the family of planes
(r, n(α)) + D(α) = 0,
where |n| = 1, |n
·
(α)| 0, |n
··
(0)| 0.
34. Find the envelope and the characteristics of the family of circular cylinders of con-
stant radius r:
a) (x − C)
2
+ y
2
= r
2
;
b) x
2
+ (y − C)
2
= r
2
;
c) (x − C)
2
+ (y − C)
2
= r
2
.
35. Write the equation of the tangent plane at an arbitrary point of the surface generated
by the tangents to the space curve r = r(s). Study its position when the tangency point
moves along a generators of the surface.
36. Show that the osculating plane to the regression edge of a developable surface coin-
cides with the tangent plane to the surface at that point.
37. Show that the intersection curve between a developable surface and the osculating
plane at its regression edge has, for arbitrary point of this edge, the curvature equal to
3
4
from the curvature of the regression edge at that point.
38. A Catalan surface is a skew ruled surface whose generators are parallel to a given
plane, the directing plane. Show that a ruled surface
r = r
1
(u) + vb(u)
is a Catalan surface iff
(b, b
·
, b
··
) = 0, b
··
0.
39. Find the first fundamental form of the second degree surfaces
1. x
2
+ y
2
+ z
2
= r
2
(the sphere);
2.
x
2
a
2
+
y
2
b
2
+
z
2
c
2
= 1 (the ellipsoid);
3.
x
2
a
2
+
y
2
b
2

z
2
c
2
= 1 (the hyperboloid with one sheet);
4.
x
2
a
2
+
y
2
b
2

z
2
c
2
= −1 (the hyperboloid with two sheets);
222 Problems
5.
x
2
p
+
y
2
q
= 2z (the elliptical paraboloid);
6.
x
2
p

y
2
q
= 2z (the hyperbolical paraboloid);
7.
x
2
a
2
+
y
2
b
2

z
2
c
2
= 0 (the cone);
8.
x
2
a
2
+
y
2
b
2
= 1 (the elliptical cylinder);
9.
x
2
a
2

y
2
b
2
= 1 (the hyperbolical cylinder);
10. y
2
= 2px (the parabolical cylinder).
40. Specify which of the following quadratic forms can play the role of the first funda-
mental form for a surface:
1. ds
2
= du
2
+ 4dudv + dv
2
;
2. ds
2
= du
2
+ 4dudv + 4dv
2
;
3. ds
2
= du
2
− 4dudv + 6dv
2
;
4. ds
2
= du
2
+ 4dudv − 2dv
2
.
41. Find the transformation formulas for the coefficients of the first fundamental form
and of the quantity
H =

EG − F
2
for a parameters’ change.
42. Show that any revolution surface has a parameterization for which the first funda-
mental form can be written under the form
ds
2
= du
2
+ G(u)dv
2
.
43. Acoordinate net on a surface is called a Tchebysheff net if the segments of coordinate
lines from a family intercepted between two coordinate lines from the other family have
constant length. Show that a coordinate net on a surface is a Tchebysheff net iff E
·
v
=
0, G
·
u
= 0.
44. Find a coordinate transformation such that the first fundamental formof the pseudosphere,
ds
2
= a
2
ctg
2
udu
2
+ a
2
sin
2
udv
2
becomes
ds
2
= du
2
+ G(u)dv
2
.
Problems 223
45. Find the equation of a curve which intersect the meridians of a revolution surface
under a constant angle α (the loxodrom).
46. Find the loxodroms of the sphere.
47. Find the orthogonal trajectories of the rectilinear generators of a conical surface.
48. Find the differential equation of the curve which intersects the rectilinear generators
of a tangent developable under a constant angle α.
49. Find the orthogonal trajectories of the family of curves
u + v = const,
lying on the sphere

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = Rcos u cos v,
y = Rcos u sin v,
z = Rsin u.
50. On the circular cone

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v,
y = u sin v,
z = u,
one considers the family of curves
v = u
2
+ α,
where α is a parameter. Find the orthogonal trajectories of this family.
51. Find the equations of the curves from the sphere

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = Rcos u sin v,
y = Rsin u sin v,
z = Rcos v
which bissect the angle between the meridians and the parallels.
52. Find the intersection angle between the curves
u + v = 0, u − v = 0
224 Problems
from the right helicoid

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v,
y = u sin v,
z = av.
53. Find the perimeter and the interior angles of the curvilinear triangle
u = ±
1
2
av
2
, v = 1,
from the surface with the first fundamental form
ds
2
= du
2
+ (u
2
+ a
2
)dv
2
.
54. On the surface with the first fundamental form
ds
2
= du
2
+ sinh
2
udv
2
find the length of the arc of the curve u = v between the points M
1
(u
1
, v
1
) and M
2
(u
2
, v
2
).
55. Find the angle between the curves
v = u + 1 ¸ si v = 3 − u
from the surface

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos c,
y = u sin v,
z = u
2
.
56. On a sphere there is given a right angle triangle, whose sides are arcs of great circles
of the sphere. Find:
1. the relation between the sides of the triangle;
2. the area of the triangle.
57. Find the area of the quadrilater from the right helicoid

¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
x = u cos v,
y = u sin v,
z = av.
bounded by the curves
u = 0, u = a, v = 0, v = 1.
Problems 225
58. Find the second fundamental form of the following surfaces of rotation:
1. x = f (u) cos v, , y = f (u) sin v, z = ϕ(u) – the surface of rotation of rotation axis
Oz;
2. x = Rcos u cos v, y = Rcos u sin v, z = Rsin u – the sphere;
3. x = a cos u cos v, y = a cos u sin v, z = c sin u – the ellipsoid of rotation;
4. x = a cosh u cos v, y = a cosh u sin v, z = c sinh u – the hyperboloid of rotation
with one sheet;
5. x = a sinh u cos v, y = a sinh u sin v, z = c cosh u – the hyperboloid of rotation
with two sheets;
6. x = u cos v, y = u sin v, z = u
2
– the paraboloid of rotation;
7. x = Rcos v, y = Rsin v, z = u – the circular cylinder;
8. x = u cos v, y = u sin v, z = ku – the circular cone;
9. x = (a + b cos u) cos v, (a + b cos u) sin v, z = b sin u – the torus;
10. x = a cosh
u
a
cos v, y = a cosh
u
a
sin v, z = u – the catenoid;
11. x = a sin u cos v, y = a sin u sin v, z = a

ln tg
u
2
+ cos u

– the pseudosphere.
59. Find the principal curvatures and the principal directions of the right helicoid
x = u cos v, y = u sin v, z = av.
60. Find the principal curvatures at the vertices of the hyperboloid with two sheets
x
2
a
2

y
2
b
2

z
2
c
2
= 1.
61. Compute the principal curvatures of the surface
x
2
p
+
y
2
q
= 2z
at the point M(0, 0, 0).
226 Problems
62. Show that at each point of the surface
x = u cos v, y = u sin v, z = λu
one of the normal principal sections is a straight line.
63. Find the curvatures of the normal sections of the surface y =
1
2
x
2
:
1. at an arbitrary point and in an arbitrary direction;
2. at the points of the lines obtained by sectioning the surface with planes z = k and
in directions tangents to this curves;
3. at the point M(2, 2, 4), in the direction of the tangent to the curve
y =
1
2
x
2
, z = x
2
.
64. On the surface
x = u
2
+ v
2
, y = u
2
− v
2
, z = uv
we take the point P(u = 1, v = 1).
1. Compute the principal curvatures of the surface at P.
2. Find the equations of the tangents PT
1
, PT
2
to the principal normal sections at the
indicated point.
3. Find the curvature of the normal section passing through the tangent to the curve
v = u
2
.
65. We are given the surface
z = 2x
2
+
9
2
y
2
.
1. Find the equation of the Dupin’s indicatrix at the origin of coordinates.
2. Compute, at the origin of coordinate, the curvature radius of the normal section
curve whose angle with the x-axis is 45

.
66. In the tangent plane at the point M to a surface on draws n straight lines passing
through the origin and intercepting between them angle equal to
π
n
. Show that
K
m
=
1
n
¸
1
r
1
+
1
r
2
+ +
1
r
n

,
where
1
r
i
are the normal curvatures of the curves from the surface which are tangent to
the given straight lines, while K
m
is the mean curvature of the surface at the point M.
Problems 227
67. Through a point M of an ellipsoid of rotation one draws all the possible curves, lying
on the ellipsoid. Find the geometrical locus of the curvature centers of these curves at
M.
68. Find the surfaces for which the first fundamental form is a perfect square.
69. Through a point M of a surface one considers all the possible plane sections. Find
the equation of the surface which contains the centers of the osculating circles of these
sections.
70. Aparabolical cylinder intersects with a plane which is perpendicular to its generators
after a parabola C.Let M be the vertex of this parabola and MT – the tangent to the
parabola at the vertex M. On what curve lie the foci of the parabolas obtained sectioning
the cylinder with planes passing through the straight line MT?
71. Starting from the fact that an ellipse can be projected onto a circle, find the curvature
radius of an ellipse at one of its vertices, using Meusnier’s theorem.
228 Problems
Bibliography
[1] Bär, C. – Elementare Differentialgeometrie, Walter de Gruyter, 2001
[2] Barbosa, J.L.M., Colares, A.G. – Minimal Surfaces in R
3
, Springer Verlag, Berlin,
1986 (Lecture Notes in Mathematics, 1195)
[3] Beltrami, E. – Ricerche di analisi applicata alla geometria, Giorn. di Mat., 2-3,
1864–1865
[4] Bianchi, L. – Lezioni di geometria differenziale, Spoerri, Pisa, 1894
[5] Blaschke, W. – Vorlesungen über Differentialgeometrie. I. Elementare Differential-
geometrie, Springer Verlag, Berlin, 1921
[6] Burali-Forti, C. – Fondamenti per la geometria differenziale di una superficie col
metodo vettoriale generale, Rend. Palermo, 33(1912)
[7] do Carmo, M.P. – Differential Geometry of Curves and Surfaces, Prentice Hall,
1976
[8] Catalan, E. – Journal de Mathém., 7 (1842), p.203
[9] Codazzi, D. – Sulle coordinate curvilinee d’una superficie e dello spazio, Ann. di
Mat. (2), 1, 1867–1868; 2, 1868–1869; 4, 1870–1871.
[10] Darboux, G. – Leçons sur la théorie générale des surfaces, Volumes I to IV,
Gauthier-Villars, Paris, 1887-1896
230 Bibliography
[11] Dupin, Ch. – Développement de géométrie, Paris, 1813
[12] Dupin, Ch. – Applications degéométrie et de m’ecanique, Paris, 1822
[13] Enneper, A. – Zeitschrift für Mathem. und Physik, 9 (1864), p.108
[14] Euler, L. – Introduction in analysin infinitorum, I–II, Lausannae, 1748
[15] Euler, L. – De projectione geographica superficiei sphaerici, Acta Ac. Petrop., I,
1777
[16] Eisenhart, L.P. – A Treatise on the Differential Geometry of Curves and Surfaces,
Ginn & Co., Boston, 1909
[17] Eisenhart, L.P. – An Introduction to Differential Geometry, Princeton University
Press, 1947
[18] Favard, J. – Cours de géométrie differentielle locale, Gauthier-Villars, Paris, 1957
[19] Fedenko, A.S. (ed.) – Differencial’naja geometrija, Minsk, 1982
[20] Finikov, S.P. – Kurs differencial’noi geometrii, Moscow, 1952
[21] Forsyth, A.R. – Lectures on the Differential Geometry of Curves and Surfaces,
Cambridge University Press, 1912
[22] Frenet, F. – Sur les courbes à double courbure, Journ. de Math. 1, 1852
[23] Gauss, C.F. – Disquisitiones generales circa superficies curvas, Comm. Soc. Göt-
tingen, Bd. 6, 1823-1827
[24] Gray, A. – Modern Differential Geometry of Curves and Surfaces, CRCPress, 1993
[25] Hilbert, D., Cohn-Vossen, S. – Geometry and the Imagination, Chelsea, New York,
1952
[26] Hopf, H. – Selected Topics in the Differential Geometry in the Large, Notes by T.
Klotz, New York, Institute of Mathematical Sciences, New York University, 1955
[27] Hopf, H. – Lectures on Differential Geometry in the Large, Notes by J. W. Gray,
Standford, Applied Mathematics and Statistics Laboratory, Standford University,
1955
[28] Hsing, C.C. – A First Course in Differential Geometry, John Wiley, 1981
Bibliography 231
[29] Jellet, J.H. – On the properties of inextensible surfaces, Trans. Irish Ac. Dublin,
textbf22 [1855], 1856
[30] Joachimstahl, F. – Demonstrationes theorematum ad superficies curvas spectan-
tum, Journ. reine angew. Math. 30, 1846
[31] Klingenberg, W. – A Course in Differential Geometry, Springer (Graduate Texts in
Mathematics, 51), 1983
[32] Kreyszig, E. – Differential Geometry, University of Toronto Press, Toronto, 1959
[33] Lagrange, J.L. – Essai d’une nouvelle méthode pour déterminer les maxima et les
minima des formules intégrales indéfinies, Miscellanea Taurinesnea, II, 1760–1761
[34] Lagrange, J.L. – Sur la construction des cartes géographiques, Nouv. Mém. Ac.
Berlin, 1779
[35] Lamé, G. – Leçons sur les coordonnées curvilignes et leurs diverses applications,
Paris, 1859
[36] Lelong-Ferrand, J., Arnaudiès, J.M. – Cours de mathématiques, Tome 3:
Géométrie et cinématique, 2
e
édition, Dunod, Paris, 1977
[37] Meusnier, J.B. – Mémoire sur la courbure des surfaces, Mémoires des savants
étrangers, 10 (lu 1776), 1785, 477–510
[38] Mainardi, G. – Sulla teoria generale delle superficie, Giorn. Ist. Lombardo, 9, 1856
[39] Minding, F. – Ueber die Biegung krummer Flächen, Journ. reine angew. Math., 18,
1838
[40] Minding, F. – Wie sich entscheiden lässt, ob zwei gegebene krumme Flächen auf
einander abwickelbar sind oder nicht; näbst Bemerkungen über die Flächen von
unveränderlichem Krümmungsmasse, Journ. reine angew. Math., 19, 1839
[41] Monge, G. – Mémoire sur l’intégration de quelques équations aux derivées par-
tielles, Mém. Ac. sci., 1787
[42] Monge, G. – Applications de l’analyse à la géométrie, Paris, 1807
[43] Nitsche, J.C.C. – Vorlesungen über Minimalflächen, Springer, 1975
[44] O’Neill, B. – Elementary Differential Geometry, Academic Press, New York, 1966
232 Bibliography
[45] Osserman, R. – A Survey of Minimal Surfaces, second edition, Dover, 1986
[46] Pogorelov, A.V. – Differential Geometry, Noordhoff, Groningen, 1966
[47] Raševskii, P.K. – Kurs differencial’noi geometrii, 4
th
edition, Moscow, 1956
[48] Riemann, B. – Über die Hypothesen, welche der Geometrie zu Grunde liegen, Gött.
Abh., 13, 1868
[49] Rodrigues, O. – Recherches sur la théorie analytique des lignes et des rayons de
courbure des surfaces, Corrésp. École polyt., textbf3, 1815
[50] Germain, S. – Mémoire sur la courbure des surfaces, Journ. reine angew. Math., 7,
1830
[51] Scheffers, G. – Anwendung der Differential- und Integralrechnung auf Geometrie.
I. Einführung in die Theorie der Kurven in der Ebene und imRaume; II. Einführung
in die Theorie der Flächen, Teubner, Berlin-Leipzig, 1900, 1902
[52] Scherk, P. – Crelle’s Journal f. Mathem., 13 (1835)
[53] Serret, A. – Mémoire sur quelques formules relatives á la théorie des courbes à
double courbure, Journ. de Math., 16, 1851
[54] Steiner, J. – Ueber parallele Flächen, Mon. Ber. Ak. Wiss., Berlin, 1840, pp. 114–
118
[55] Stoker, J.J. – Differential Geometry, John Wiley and Sons, 1969
[56] Struik, D.J. – Lectures on Classical Differential Geometry, second edition, Dover,
1988
[57] Thorpe, J.A. – Elementary Topics in Differential Geometry, Springer (Undergrad-
uate Texts in Mathematics), 1979
[58] Vygodskii, M. Ya. – Differential’naja geometriya, Moscow-Leningrad, 1949
[59] Willmore, T. – An Introduction to Differential Geometry, Clarendon Press, Oxford,
1959
Index
affine normal, 109
arc length, 19, 20, 42, 57, 59, 88, 89, 99,
149, 198, 224
area, 142–144
astroid, 104–106, 110
lengthened, 80
asymptotic
direction, 150, 151, 154, 202, 208
line, 150–154, 208, 209
Bertrand
curve, 58–62
mate, 58, 59, 62, 209
binormal, 45, 53, 60, 67, 180, 198, 199
biregular curve, 39, 41, 43–46, 49–52,
54, 64, 69, 70, 73, 150, 152,
197
biregular curve , 49
biregular point, 39–41, 62, 66
Bonnet’s theorem, 168
catenary, 100, 109, 217
catenoid, 206, 217, 225
characteristic, 190, 191, 194, 195, 197,
198, 200, 220, 221
Christoffel’s coefficients, 162, 164, 170,
171, 178, 179, 181
circle, 13, 16, 19, 23, 28, 43, 48, 54,
55, 77, 78, 86, 89–93, 96, 101,
103–105, 107–112, 118, 141,
181, 182, 191, 217, 220, 224,
227
clothoid, 99
Codazzi-Mainardi equations, 165, 167,
168, 204
conchoid, 110
conchoid of Nicomede, 104
cone, 154, 222, 223, 225
conformal mapping, 204–206
conoid, 207, 208
contact, 34, 40, 41, 64, 66–68, 109, 125
Cornu’s spiral, 99
curvature, 150
absolute, 85
axis, 67, 198
center, 67, 86, 87, 91, 148, 154, 198,
227
coordinates, 164
curvature, 210
Gaussian, 156, 157, 189, 190, 193,
234 Index
200, 213, 214
geodesic, 175, 177–180
line, 156, 158–160, 164, 167, 204
mean, 156, 157, 161, 200, 202, 203,
226
normal, 148–152, 154, 156, 157, 176,
226
of a curve, 42–44, 53–55, 57, 58,
60, 62, 66, 70, 71, 73, 81, 82,
84–86, 95, 97, 99, 111, 148–
152, 162, 210, 211, 221
geometrical meaning, 44
of a plane curve, 81, 82, 96
of a regular curve, 43
principal, 156–158, 161, 205, 225,
226
radius, 43, 67, 91, 97, 99, 109, 226,
227
signed, 54, 75, 82–86, 89, 90, 92,
93, 177, 179, 180, 212
tangential, 177
total, 157, 161, 172–174, 190, 213–
215
vector, 42, 43, 45, 47, 63, 83, 84,
148, 149
curvatures
of a surfaces, 161
cycloid, 104, 105, 109, 110
lengthened, 104
shortened, 104
cycloidal curves, 98
cylinder, 111, 154, 227
circular, 30, 62, 221, 225
elliptical, 222
hyperbolical, 222
parabolical, 222, 227
Darboux axis, 200
Darboux frame, 175, 176
Darboux vector, 53, 200
developable surface, 191–193, 195–197,
200, 212, 213, 221, 223
directrix, 185–187, 192, 193, 199, 208,
209
edge of regression, 190, 191, 193, 195,
197–199, 220, 221
ellipse, 13, 22, 80, 81, 105, 107–110,
227
ellipsoid, 154, 221, 225, 227
elliptic point of a surface, 153, 154, 161
envelope
of plane curves, 75–80, 86, 107, 108,
190
of planes, 193–195, 197–200, 220,
221
of space curves, 191
of surfaces, 190, 191, 194, 195, 220,
221
epicycloid, 105, 109
Euler’s formula, 157
evolute, 79–81, 86, 87, 89, 90, 110
Frenet
axes, 63
equations, 51, 52, 58, 63, 70, 71,
95, 175, 199, 210
formulae, 50, 51, 54, 55, 57, 59–61,
176, 177, 180, 197, 198, 211
frame, 44–52, 64, 66–69, 71, 175,
197, 200, 210
system, 95, 96
vectors, 63, 70
fundamental equations of a surface, 162
fundamental form, 209
first, 139–141, 143, 152, 163, 165,
167, 168, 170–174, 178, 179,
Index 235
182, 185, 187–189, 202, 205,
213, 214, 221, 222, 224, 227
second, 146–148, 150–153, 159, 160,
162, 165, 168, 171, 172, 178,
179, 189, 203, 225
fundamental forms, 148, 150, 162, 164,
165, 167, 168, 170, 172, 178,
200–202, 204, 206, 207
fundamental homography, 138
fundamental operator, 137
fundamental theorem of surface theory,
167
Gauss egregium theorem, 172
Gauss equations, 165, 167, 168, 172
geodesic lines, 180
geodesics, 175, 180–183, 200
helicoid, 146, 152, 153, 206–208, 212–
214, 218, 224, 225
helix
circular, 16, 21, 22, 33, 55, 59, 62,
208–211, 220
general, 56–58, 62
hyperbola, 105, 106, 108–110
equilateral, 106, 108, 109
hyperbolic point of a surface, 153, 154,
161
hyperboloid
of rotation with one sheet, 225
of rotation with two sheets, 225
with one sheet, 221
with two sheets, 221, 225
hypocycloid, 105
involute, 86–90, 101, 110
isothermic lines, 204, 206
isothermic parameterization, 202–204
isothermic system, 204
Joachimstahl’s theorem, 158, 159
Klein’s bottle, 130, 132
Liouville surfaces, 182
Möbius band, 128–131
Meusnier’s theorem, 148, 154, 227
minimal surfaces, 200, 202–208, 212,
213, 229, 232
monkey saddle, 155
moving frame, 44, 45
natural basis, 124, 135, 136, 138, 140,
142, 144, 147, 151
natural equations of a curve, 95–101, 110
natural parameter, 19–22, 42, 88, 149,
197, 209
natural parameterization, 21
naturally parameterized curve, 19–21, 42–
44, 46, 51–55, 59, 62–64, 66,
71–73, 84, 86–90, 92, 149, 177,
179, 180, 197, 210, 212
normal line, 32, 34, 35, 37, 39, 45, 80,
105, 107
normal line , 35, 105
normal plane, 32, 34–36, 45, 64, 67, 112,
175
normal section, 148, 154
normal to a plane, 159
normal to a surface, 123, 125, 126, 150,
159, 180, 181, 189, 192
normal to asurface, 126
normal vector, 189
normal versor, 162, 164
normal versor of a plane, 40, 125
orientable surface, 127–130, 137
orientation
236 Index
of a curve, 48, 49
of a surface, 127–130, 136–138, 143,
144, 146, 147, 168
oriented curve, 48, 49, 84
oriented surface, 127, 130, 136, 137, 142,
144, 146–153, 156–159, 161
osculating sphere, 68, 199
osculating circle, 91, 92, 108, 109, 227
osculating plane, 39–41, 45, 54, 64, 66–
68, 91, 111, 112, 148, 150, 152,
180, 181, 195, 197, 199, 200,
219–221
osculating sphere, 68
osculation contact, 67
parabola, 64, 106–110, 227
semicubic, 64
parabolic point of a surface, 153, 154,
161
paraboloid
elliptical, 111, 222
hyperbolical, 154, 222
of rotation, 225
polar axis, 67, 104, 198
polar surface, 198, 199, 220
principal direction, 156, 158–160
principal line, 156, 158
principal normal, 45, 47, 58, 59, 62, 112,
152, 198, 200, 208, 219, 226
principal operator, 137
pseudosphere, 213, 214, 218, 222, 225
rectifying
plane, 46, 57, 64, 198–200, 220
surface, 199, 200
Riccati equation, 95, 96
ruled surface, 152, 185–193, 196, 200,
206–209, 212, 213, 221
ruling, 185–189, 191, 192
saddle point of a surface, 154
second curvature (torsion), 51, 53
shape operator, 137–139, 144–147, 156–
159, 161, 162, 164
sphere, 30, 31, 55, 66–68, 110, 111, 116,
118, 126, 127, 136–138, 140,
141, 147, 148, 181, 182, 185,
199, 205, 206, 213, 218, 221,
223–225, 236
spherical map, 136–138, 146, 204–206
superosculation contact, 68
surfaces of constant curvature, 213–215
surfaces of revolution, 117, 159, 182,
206, 214, 222, 223
tangent, 17, 22, 32–36, 38–41, 44, 45,
49, 59, 66, 67, 75, 103, 105,
106, 111, 124, 134, 148–150,
156, 177, 185, 197, 226, 227
tangent developable surface, 223
tangent plane, 67, 123–126, 143, 150,
152, 156, 175, 177, 179–181,
187–189, 192, 193, 200, 208,
218–221, 226
tangent plane , 125, 177, 189, 191
tangent space to a surface, 123, 124, 126,
127, 137, 139, 140, 151, 156,
157
tangent vector, 20, 32, 33, 47, 49, 123,
134, 135, 140, 142, 146, 149–
151, 159, 175, 180, 200
torsion, 50–55, 57, 58, 60, 62, 66, 68,
70, 71, 73, 83, 95, 111, 162,
177, 197, 210, 211, 236
geodesic, 175, 177
torsion, 177
Index 237
torus, 118, 119, 154, 191, 217, 219, 225
unit binormal, 45, 54
unit principal normal, 45, 47, 152
unit tangent vector, 45, 84, 85, 95
Weingarten
coefficients, 164
formulae, 164, 165
mapping, 137, 164
operator, 164

To Cristina, with love

. .10 The Frenet formulae. .2 Space curves .3 The definition of the curve . . . . . . . . . . . . . . . The torsion . . . . . . . . 1. . . . . . . . . .1 The behaviour of the Frenet frame at a parameter change . . . . . . . .1 Plane curves . . . . . . . . .9 Oriented curves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . .1 The equations of the tangent line and normal plane (line) for different representations of curves . . . . . . . . . . . . . . . 1. . . . . 1. . . . 1. . . . . . . . . . . . . . . . . . . . . .7 The curvature of a curve . . . . The Frenet frame of an oriented curve . . . . . . . . 1. . . . . . . . . . . . .Contents Foreword 9 I Curves 1 Space curves 1. . .4. 1. . . . . . . . . . . .5.8.1 Introduction . . . .8 The Frenet frame . . . . . . . . . . . . . . 1.2 Parameterized curves (paths) . . . . .7. . . . . . .4. . . . . . . . . . . . . . . . 1. . . . . . .1 The geometrical meaning of curvature . . . . . . . . . 1. .6 The osculating plane . . . . . . . . . . 1.5 The tangent and the normal plane . . . . . . .4 Analytical representations of curves . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . 1. . 11 13 13 14 22 26 26 29 32 35 39 42 44 44 47 48 50 . . . 1. . . . . . . . . . . . . . . . . . . .

. The local behaviour of a parameterized curve around a biregular point . . . .3 General helices. . . . .5 The osculating circle of a curve . . . .10. . . . . . The contact between a space curve and a plane . . .1 The geometrical meaning of the torsion . . . . . . . 2. . . .4 Bertrand curves . . . .4 Curves on a surface . . . . . .2 The uniqueness theorem . . 2. .1 The geometrical interpretation of the signed curvature . . . . . .3 The existence theorem . . . . . . . . . . . . . . . . . . . . . . . . . .14. . . . . . . . . .1 The Riccati equation associated to the natural equations of a curve . . . 4. . . 2. . . . . . . The evolute and the involute of a plane curve 2.1 Introduction .6 Contents 1. . . . .14. . The osculating sphere Existence and uniqueness theorems for parameterized curves . . . . . . . 2. . . . .2 Examples for the integration of the natural equation of a plane curve . 53 54 56 58 62 64 66 68 68 70 72 75 75 75 77 79 79 81 84 86 91 92 95 95 96 103 1. . . . . . . .6 The existence and uniqueness theorem for plane curves . . . .3 Applications: the evolute of a plane curve . . . . . . . . 3 The integration of the natural equations of a curve 3. . . . . . . . . Lancret’s theorem . . . . . . . . . . . . . 3. . . . . . . . . . .14. 113 115 115 116 116 119 122 4 General theory of surfaces 4. . .2. 4. . . . . . . . .11 1. . . . . . . . . 1. . . . . . 1. .3 The equivalence of local parameterizations . . . . .2 Envelopes of plane curves . . . . .2 Surfaces . . . . . .14 2 Plane curves 2. . . . . .10. 2.4 The curvature center. . . .1 The behaviour of the Frenet frame under a rigid motion . . . . . . .2 Families of curves depending on two parameters . . . . . . . . . . . . . . . . . . . . . .10. . . . . . . . . 4.1 Parameterized surfaces (patches) .10. . . . . . . .1 Representations of surfaces .3 The curvature of a plane curve . . . . . . . The contact between a space curve and a sphere. . . . . 2. .2 Some further applications of the Frenet formulae . . . . . . 1. . . .2. . . . . . . . . . Problems II Surfaces . . . . . . . . . . . . . . 1. .1 Curves given through an implicit equation . . . . . . 1. . . . . . .13 1. . . . .3. . . . . . . . .2. . . . . 4. . . . . . . . . . . . 1. . . 2. . . . .12 1. . . . . . . . . . . . . . .2. . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19. . . . . . . . . 152 Principal directions and curvatures . . . . . . . . . . . . 161 The fundamental equations of a surface . . . . . . . . . . .2 The Gaussian curvature of a ruled surface . . . . . 123 The orientation of surfaces . . . . . . . . . . . . 180 Examples of geodesics . . . . . 165 4. . . . . . . . 162 4. . .1 Introduction . . . . . .17. 148 Asymptotic directions and asymptotic lines on a surface . . . . . 136 The first fundamental form of a surface . . . 172 Geodesics . . . . . . . . . . . . . 150 The classification of points on a surface . .4 The fundamental theorem of surface theory . . . . 156 4. . . . . . . . . . .10 7 4.17.1 General ruled surfaces . . . . . . . . . 162 Christoffel’s and Weingarten’s coefficients in curvature coordinates164 4. 140 The length of a segment of curve on a surface .11 4. . 139 4.17. . . . . . . . .12 4. . . . . . . . . . . . . . . . . . .16. 181 4. . . . . . . . . . . . . . . . . . . . . . . .5 4. . . . . . . . . . . .13 4. . . .19 The tangent vector space. . . . . . . . . . . . . . . . . . . . . The Meusnier’s theorem . . . . . . . . . . . . . . . . . . . .19. . . . . .7 4. . . . . . . . . . . . . . . . 130 The differential of a smooth map between surfaces . . 182 5 Special classes of surfaces 185 5. . . . . 141 The area of a parameterized surface . . 140 The angle of two curves on a surface . . . . . . . . . .4 Liouville surfaces . . . . . . . . . . . . . . . . . The geodesic curvature and geodesic torsion 175 4. . . . .3 The Gauss’ and Codazzi-Mainardi’s equations for a surface . . . . . 162 4. . . .9 4. . . . . .1 The determination of the lines of curvature . . 185 5.10. . . . . . . . . . . . . . . .1 Introduction . . . . 159 4. . . . . . . . .2 The computation of the curvatures of a surface . . . . . . . . . Christoffel’s coefficients . . . . the tangent plane and the normal to a surface . . . .17. .1 First applications . . . . . . .2 The differentiation rules. .1. . . . . .1 Ruled surfaces . . .16. . . . . . .14 4. . . 185 The parameterization of a ruled surface . .17 4. . . . . . . .8 4. . .16 4. . . . . . 189 . . . . 167 The Gauss’ egregium theorem .3 Geodesic lines .2 The Darboux frame. .19. . . . . 142 The matrix of the shape operator . . . . . . . . . .19. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 The spherical map and the shape operator of an oriented surface . . . . . . . . . . . . . . . . .Contents 4. . . . . . . . . . . . . 186 The tangent plane and the first fundamental form of a ruled surface187 5. . . . . . . . . . . . . . . . . . 175 4. . . . .1. . . . . . . . . . . . . . . .18 4. . 127 Differentiable maps on a surface . . . . . . 146 The normal curvature. . . . . . . . . . . . . . . . . . . . . . . . . .15 4.6 4. . . . . 144 The second fundamental form of an oriented surface . . . . . . . . . . . . . . . . . . 175 4.

8 5.1.3 5.1.4

Contents Envelope of a family of surfaces . . . . . . . . . . . . . . . . . Developable surfaces . . . . . . . . . . . . . . . . . . . . . . . Developable surfaces as envelopes of a one-parameter family of planes. The regression edge of a developable surface . 5.1.5 Developable surfaces associated to the Frenet frame of a space curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The envelope of the family of osculating planes . . . . . . . . . The envelope of the family of normal planes (the polar surface) The envelope of the family of rectifying planes of a space curves Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Definition and general properties . . . . . . . . . . . . . . . . . 5.2.2 Minimal surfaces of revolution . . . . . . . . . . . . . . . . . . 5.2.3 Ruled minimal surfaces . . . . . . . . . . . . . . . . . . . . . . Surfaces of constant curvature . . . . . . . . . . . . . . . . . . . . . . 190 191 193 197 197 198 199 200 200 206 207 213 217 229 233

5.2

5.3

Problems Bibliography Index

Foreword

This book is based on the lecture notes of several courses on the differential geometry of curves and surfaces that I gave during the last eight years. These courses were addressed to different audience and, as such, the lecture notes have been revised again and again and once almost entirely rewritten. I tried to expose the material in a modern language, without loosing the contact with the intuition. There is a common believe that mathematics is eternal and it never changes, unlike other sciences. I think that nothing could be farther from the truth. Mathematics does change continuously, either for the need of more rigor, either for the need of more structure. The mathematics we are doing and teaching today is essentially different (and not only as language!) from the mathematics of the nineteenth century. This variation also applies, of course, to the teaching of mathematics. There are several reason why we need to renew the textbooks from time to time: • the mathematics evolves and we need to introduce new results and notions; • some parts of specific courses become obsolete and should be removed; • the curriculum as a whole evolves and we have to keep pace with the developments of the neighboring fields; • some courses move from the undergraduate level to the graduate level or (more often) the other way around and we have to modify the contents and the language accordingly. These are only some of the reasons I decided to write this book.

10

Foreword

The material included is fairly classical. I preferred to discuss in more details the foundations rather than introduce more and more topics. In particular, unlike most of books of this kind, I decided not to discuss at all problems of global differential geometry. I believe it is not very useful to discuss such topics, if there is not enough room to give them all the attention they deserve. Also, the Levi-Civita parallelism and the covariant differentiation are not discussed, because I think they would feel a lot better in the course of Riemannian geometry. The notion of differentiable manifold is not introduced but it is only one step away. In fact, I decided to adapt the language of the course in such a way to qualify for a prerequisite for a course in smooth manifolds. Is not easy to be original when you teach something that have been taught again and again for more than two centuries and, on the other hand, should we be always original at any cost?! As such, it goes without saying that this book owes much to many excellent classical or more modern textbooks. All of them are mentioned in the bibliography. I have to mention, however, the books of Gray, do Carmo and Fedenko. I have to confess that I am not a self-educated person (not entirely, anyway), and my ideas regarding differential geometry have been influenced by my teachers (the late Prof. Marian Tarin˘ and Prof. Dorin Andrica), as well by the colleagues with whom I had the ¸ a pleasure to work and to share the ideas, Prof. Pavel Enghi¸, Csaba Varga, Cornel Pintea, s Daniel V˘ c˘ re¸u and Liana Topan. a a t ¸ This book would have never existed without my students and their enthusiasm (or lack of enthusiasm) determined many changes during the years. A book is not something that you leave at the office and this one was no exception and it would never have been finished without the infinite patience of my wife who tolerated the (undeterministic) chaos created by books and manuscripts. Cluj-Napoca, April 2005.

Part I

Curves

.

clearly. in this particular case. Finally. On the other hand. We are familiar with some of them already from elementary mathematics. locally. that all the functions are at least once continuous differentiable and. it is enough to think of a conical section (for instance an ellipse or a circle. as it is known. Apart . obviously. the curves are just deformations of straight lines. As we shall see. usually the curves are not graphs of functions (well. we are interested to apply the tools of differential calculus.CHAPTER 1 Space curves 1. as it would be the case for a graph. They can be thought of. where. We shall assume. On the other hand. therefore. that all the derivatives exists and are continuous. we shall assume. as “one-dimensional” objects. in particular). An important issue that has to be dealt with is related to the smoothness of the functions used to describe a curve. since.1 Introduction Intuitively. if higher order derivatives are involved. f is a second degree polynomial in x and y. from this point of view. this is. an arbitrary curve. the most convenient way to represent. without stating it in clear. y) = 0. the graphs of functions can be considered as curves. Thus. in general we cannot represent a curve by an equation of the form y = f (x). we can represent a conical section also by an implicit equation of the form f (x. Of course. usually. therefore. We shall use for the functions which are satisfying these conditions the generic term “smooth”. at least not globally). we can represent the coordinates of each point of the curve as functions of one parameter. of course.

we prefer to let the reader satisfy the curiosity by himself. A C k (k > 0) parameterized curve ( or path) in the Euclidean space R3 is a C k mapping r : I → R3 : t → (x(t). that the functions we are considering are of class C k in the interior of the interval and all the derivatives. 1.1. deeper reasons for considering functions at least once continuously differentiable. up . Space curves from the computational aspects. but. z are C k . We note that. namely the functions have to be of bounded variation.2.14 Chapter 1.2 Parameterized curves (paths) Let I be an interval on the real axis R. Definition 1. r = r(t)) or.   . as we already emphasized. z(t)). In the figure 1. we shall assume. It is a well-known fact that the continuous differentiable functions do verify this condition and. necessarily. denoted by (I. contradicts the common sense image of a curve as an onedimensional object) has been constructed by the Italian mathematician Giuseppe Peano.2. obviously.1 we indicate an iterative process that defines a square-filling (Peano) curve. a path is C k iff the (real valued) functions x. r). Moreover. when the interval is implicit. the interval can be unbounded and can coincide with the entire real axis. moreover. In particular. (1. y(t).e. they also endow us with the necessary computational tools. that the functions we use to describe a curve do not have to be. there are other. actually. What it is required is a little bit less. It is possible. typically. If the interval is not open. the curve can fill an entire square (or the entire plane). at the end of the XIX-th century. we can find an expression for each iteration). however. this strange phenomenon does not disappear not even if the functions f and g are differentiable. for instance.  y = g(t)  Now. (I. first of all. y. The curve itself is the limit of the curves obtained by this iterative process. it can be shown that if the functions f and g are only continuous. continuous differentiable to avoid aforementioned anomalies. Suppose. We have to say. in fact. that a curve is described by a set of equations of the form   x = f (t).1) A parameterized curve is. as these “curves” are not the subject of our investigation. which are not available for arbitrary functions with bounded variation. Sometimes it is even important that the interval be closed. but not continuously differentiable. to describe analytically this curve (i. just r = r(t). The first example of such an anomalous curve (which. We shall not assume always that the interval is open.

if the interval I is. The path is called compact. half-open or open.1: The Peano’s curve (first four iterations) to the k-th order. An alternative denomination used for a closed path is that of a loop. or right) limit at the extremities of the interval. from above or from both parts. .2. half-open or open. Occasionally (for instance in the theory of the line integral) we might need to consider paths which are of class C k at all the points of an interval. respectively. in particular. Parameterized curves (paths) 15 (a) First iteration (b) Second iteration (c) Third iteration (d) Fourth iteration Figure 1. the curve is compact and the two endpoints coincide. have a finite lateral (left. then the imagine of any extremity of I is called an endpoint of the path. the path is called closed. if these extremities belong to the interval. If. with the exception of a finite number of points. If the interval I is bounded from below. compact.1. The following definition will make this more precise.

lying in the xOy-plane. The image r(I) ⊂ R3 of the interval I through the mapping (1. having the center at the origin of the coordinates. 2. r). however). using the generic term of smooth parameterized curve (meaning. in any particular case. I = [0.2). n}. We shall say that a compact parameterized curve r : [a.2. . 2π]. 4. at least C k ). I = R. for short. Remark. r(t) = (a cos t. an = b) of the interval [a.2. . b] → R3 is piecewise C k if there exists a finite subdivision (a = a0 .1) is termed the support of the path (I. . The support of this path is the same straight line.2. at least C 1 and. we shall say that the parameterized curve passes through the point a for t = t0 . ai ] is of class C k . I = R. 0). The parameterized curve R → R3 . Examples 1. . that the order of smoothness k is high enough and we will not mention it any longer (with some exceptions. Let r0 ∈ R3 be an arbitrary point. Space curves Definition 1. all the time. If r(t0 ) = a. while I = R. thus.16 Chapter 1. Its support is the straight line passing through r0 (for t = 0) and having the direction given by the vector a. r(t) = r0 + t3 a. r(t) = (cos t. (iii) f (k) has a finite left and right lateral limit at each point of S . t → r0 + ta is called a straight line. (ii) f (k) is continuous on [a. It is not difficult to show that a parameterized curve r : [a. b] | f (k) does not exist is finite. b] such that the restriction of r to each compact interval [ai−1 . b ∈ R. The support of the path is the unit circle. a. Sometimes. if kth order derivatives are involved. a sin t. Hereafter we shall suppose. sin t. . . . bt). we shall refer to this point as the point t0 of the parameterized curve. b] \ S . and a ∈ R3 a vector. a1 . The support of this parameterized curve is called a circular cylindrical helix (see figure 1. where i ∈ {1. a 0. 3. . . b] → R3 is piecewise C k iff the following conditions are simultaneously fulfilled: (i) The set S = t ∈ [a.

which. 0). To each path corresponds a subset of R3 . ρ = ρ(s)) two parameterized curves. The regularity conditions rules out the cusps. its support. The fact that the same support can correspond both to a regular and a nonregular parameterized curve suggests that the absence of the regularity at a point doesn’t necessarily mean that the corresponding point of the support has some geometric peculiarities. as a set of points. while for the first one the support is just a straight line. again. 0). passing from a parameterized representation of the support to another should not change the regularity of the curve. which are not regular from t = 0. but not the self-intersections. For reasons that will become clear later. we are interested only in regular curves. Therefore. at the curves 2 and 6 from the previous example. 0 As we shall see a little bit later. Indeed. (J. In the end. t3 . The support is the same from the previous example. we have to impose some further conditions. sin 2t. are not desirable in applications. Parameterized curves (paths) 17 5.2. therefore. 6. As we shall see later. it is the support. is called a 1 Of course. the notion of regularity at a point of a parameterized curve is related to the existence of a well defined tangent to the curve in that point. This curve has a cusp at the point t = 0.3. Definition 1. different parameterized curves may have the same support. with the exception of those from the points 2 and 6. I = R. necessary to identify some relations between the parameterized curves that define the same support. r(t) = (cos 2t. r(t) ≡ ρ(λ(t)). since the parametric representation defines the support. 2π]. A diffeomorphism λ : I → J : t → s = λ(t) such that r = ρ ◦ λ. without any special points. for instance. I = [0. A parameterized curve (1. Nevertheless. r = r(t)). for many reasons. only for the second curve this analytic singularity implies a geometric singularity (a cusp). Let (I. with a parameterization. for the moment. we are interested in. the information is redundant. at least.2.1.1) is called regular for t = t0 if r (t0 ) and regular if it is regular for each t ∈ I. nonregular for t = 0. A parameterized curve can be thought of as a subset of R3 . r(t) = (t2 . It’s only that the regularity guarantees the absence of these peculiarities. To eliminate them. Definition 1. . if we look. as the examples 1 and 2 show.1 The support of a parameterized curve corresponds to our intuitive image of a curve as a one dimensional geometrical object. We have seen that different parameterized curves may have the same support. Remark. we notice immediately that. although they are. It is.2.2.e. the support of a parameterized curve may have self-intersections or cusps. The curves from the previous example are regular. i. both.4.

The parameterized curves from the examples 1 and 2 are not equivalent. Note that the two velocity vectors of equivalent parameterized curves in correspondent points have the same direction. then the trajectory is traversed in the opposite sense. while the points t and s = λ(t) are called correspondent. if λ (t) < 0. The effect of a reparameterization is the modification of the speed with which the trajectory is traversed. Remarks. 2) The reparameterization has a simple kinematical interpretation.2: The circular helix parameter change or a reparameterization. Example. al- . Two parameterized curves for which there is a parameter change are called equivalent. 1) The relation just defined is an equivalence relation on the set of all parameterized curves. If we interpret the parametric equations of the path as being the equations of motion of a particle. They may have different lengths (the speed) and sense. Also. Space curves Figure 1. while the vector r (t) is the velocity of the particle.18 Chapter 1. then the support is just the trajectory.

Among all parameterized curves equivalent to a given parameterized curve.2. there is one which has a special theoretical meaning and which simplifies many proofs in the theory of curve. in most cases.2. which is. we can extend the interval. Usually. its practical value is limited. as its norm is nowhere vanishing. because we would like the curves to be objects slightly more general that just supports of parameterized curves. usually the curves cannot be represented globally through the same parametric equations. then one of the points of the circle is not represented. Parameterized curves (paths) though.   . Definition 1. We can consider an arbitrary division t1 = a0 < a1 < · · · < an = t2 of the interval [t1 . we need first another notion. r = r(t)) between the points t1 and t2 is the number2 lt1 . Remark. equivalent to it and naturally parameterized. not acceptable. We shall not follow this line here. 2π). although. as mentioned. clearly. t2 ] and examine the polygonal line γn = r(a0 )r(a1 ) · · · r(an ). since. thus. therefore the integral might be negative and the absolute value is necessary. The arc length of a path (I. the equivalence classes determined by the relation defined before between parameterized curves are called curves.  y = sin θ  Now. It is by no means obvious that for any parameterized curve there is another one. One can see immediately that any smooth. r = r(s)) is regular. In particular. We shall say that a parameterized curve is naturally parameterized if r (s) = 1 for any s ∈ I. they have the same support. Of course. r (s) cannot vanish. . if we let the parameter vary only in the interval (0. it is very difficult to find it analytically and.5. It is enough to think about the circle.t2 = t2 t1 r (t) dt . It can be shown that if the parameterized curve 2 Although the integrand is positive we did not assume that t1 < t2 . naturally parameterized curve (I.1. as we shall see in a moment. 19 Remark. Sometimes. To construct such a curve. One of the most common parametric representation of the circle is   x = cos θ. There is a good motivation for defining in this way the length of an arc. again. if we want to obtain something positive. but then the same point corresponds to more than a value of the parameter. Remark. the natural parameter is denoted by s. The length of this polygonal line is the sum of the lengths of its segments.

since the translation is a diffeomorphism). Therefore. The parameterized curve (J. while the function λ : I → J will be a diffeomorphism. because in this case the tangent vector has only a finite number of discontinuity points and. then r (t) = λ (t)ρ (λ(t)). It is to be mentioned. In particular. in a way. because we may represent the same set of points as support of another parameterized curve. For any regular parameterized curve there is a naturally parameterized curve equivalent to it. r) and it is naturally parameterized. Let (I. then the limit of length of the polygonal line γn . i.. We are going to show that the arc lengths of two equivalent paths between correspondents points are equal. that the definition of the arc length makes sense also for piecewise smooth curves. its image will be an open interval J. t2 t1 r (t) dt = = t2 t1 t2 t1 ρ (λ(t)) · |λ |dt = ρ (λ) λ (t)dt = dλ λ2 λ1 ρ (λ) dλ . Therefore. which is not equivalent to the initial one. let r(t) = ρ(λ(t)). when the norm of the division goes to zero. therefore.e. r = r(t)) be a regular parameterized curve. since λ (t) = r (t) > 0. therefore the arc length is. The new path might have a different arc length between the same points of the support. the natural parameter is the arc length. Proof. and λ : I → R.s2 = |s2 − s1 |. For naturally parameterized curves. r) is “good enough” (for instance at least once continuously differentiable). its norm is integrable. ρ(s) = r(λ−1 (s)) is equivalent to (I. while (λ−1 ) (s) = 3 1 1 = λ (λ−1 (s)) r (λ−1 (s)) In a way. then l0.s = |s|. . Indeed. r = r(s)). l s1 . Proposition 1. t0 The function λ is smooth and strictly increasing. since ρ (s) = r (λ−1 (s))(λ−1 ) (s). Space curves (I. up to sign. exists and it is equal to the arc length of the path. also. t0 ∈ I. if 0 ∈ I (which can always be assumed. (I.20 Chapter 1. a characteristic of the support3 . t → t r (τ) dτ.

Find a natural parameterization of the curve x = et cos t. the fact that all the points of the curve are regular. y = et sin t. s(t) = 0 t r (τ) dτ = 0 t {−a sin τ. s t= √ . For the circular helix   x = a cos t      y = a sin t     z = bt.2. b} dτ = a2 + b2 . ρ (s) = r (λ−1 (s)) · |(λ−1 ) (s)| = 1. hence. Exercise 1.  2 2 a +b Exercise 1. in an essential way. 21 Remark. z = et .2. a cos τ. therefore. In the proof of the previous proposition we used.1. Example. . we get. the natural parameterization of the helix is given by the equations   x = a cos √ s    a2 +b2    y = a sin √ 2s 2   a +b    z = b √ s . through a parameter change.2.2.1. Show that the parameter along the curve x= is a natural parameter. there is no naturally parameterized curve equivalent to it. s s cos ln . Parameterized curves (paths) and. On an interval on which the curve has singular points. a2 + b2 Thus. 2 2 s z= √ 2 . 2 2 y= s s sin ln .

for concrete examples of parameterized curves we will hardly ever use it. the map r : I → R3 is smooth. i. as the reader will have more than once the opportunity to see in this book. we can imagine. as r(I) is not an open subset of R3 . the curve is called simple. where a and b are constant vectors.e. although the natural parameter is very important for theoretical consideration and for performing the proofs. A parameterized curve with these properties is called a local parameterization of the curve M around the point a. This is. such that the ellipse   x = a cos t    y = b sin t. intuitively. r(t) = a + b · t. What is meant. this condition should rule out both cusps and self intersections.3. Therefore. necessary. using only elementary functions) with respect to the parameter along the curve. which is not completely rigorous.1. however. is. Space curves Remark. If for a curve M there is a local parameterization (I. 1. i. given by r(t) = (a cos t. actually. because it has a global parameterization. r(I). In particular. is a set of the form M ∩ U. the origin of their name!). bt). a curve as being just a deformation of a straight line. actually.  with a b. with respect to the topology of subspace of r(I). without thinking.22 Chapter 1. Examples. usually.e. We expect the curve to have a well defined tangent at each point.e. 2. is an open neighbourhood in M of the point a (i. Definition 1. The circular helix is a simple regular curve. for each point a ∈ M there is a regular parameterized curve (I. Remark.3 The definition of the curve As we mentioned before. r) which is global. exactly the same. . b sin t. with the global parameterization r : R → R3 . the natural parameter along a parameterized curve cannot be expressed in finite terms (i. A subset M ⊂ R3 is called a regular curve (or a 1-dimensional smooth submanifold of R3 ) if. whose support. for which r(I) = M. 1. Any straight line in R3 is a simple curve.e. while the map r : I → r(I) is a homeomorphism. given by a function of the form r : R → R3 . for which the arc length can be expressed only in terms of elliptic functions (this is. b 0. It should be noticed that. impossible even for very simple curves. at an analytical representation. In some books in the definition it is required that the map r : I → r(I) be smooth. where U is an open neighborhood of a in R3 ). r).

0). a regular curve is just a subset of R3 obtained “gluing together smoothly” supports of parameterized curves. sin t. r = r(t)) and (J. also. Then. the two parameterized curves are equivalent. however. ρ(τ) = (x(τ). thus. λ = ρ−1 ◦ r. which is a compact subset of R3 . mention that even if the function is injective. the inverse might not be continuous). but it is not simple. If. We claim that the map λ : I → J. ρ = ρ(τ)) are two local parameterizations of a regular curve M around the same point a ∈ M. r|r−1 (W) ) and (ρ−1 (W). providing. with I = R and r : R → R3 . we consider the parameterized curve (I. Proof. y(τ). r(t) = (cos t. More specifically. For an arbitrary parameterized curve (I. if we restrict the definition intervals such that they have the same support. ρ = ρ(τ)) – two local parameterizations of M such that W ≡ r(I) ∩ ρ(J) ∅. y(t). We should not conclude. a parameter change between the two parameterized curves. Let M ⊂ R3 be a regular curve and (I. (J. as one could expect. since r is not a homeomorphism onto (in fact. thus. the following theorem holds: Theorem 1.3. Clearly. λ is. Let us. because it is periodic). from the very beginning. Thus. r(t) = ( x(t). r : I → r(I) might fail to be a homeomorphism (even if the map is continuous and bijective. z(τ)) be two local parameterizations of the curve M. that the circle is a simple curve. for instance.1. The definition of the curve 23 3. z(t)) ˜ ˜ ˜ and (J. To simplify the notations. that r(I) = ρ(J). clearly. r). since no open interval can be homeomorphic to the circle.3. Then (r−1 (W). r = r(t)). it cannot be a local parameterization. Now let us assume that (I. ρ|ρ−1 (W) ) are equivalent parameterized curves. the generality is not reduced by this assumption. If we look carefully at the definition of the curve.1. a homeomorphism. it is not even injective. A circle in R3 is a curve. we shall assume. centred at the origin. r). the map r : I → R3 is not injective and. as a composition of the homeomorphisms r : I → r(I) . then the support of this parameterized curve is the unit circle in the coordinate plane xOy. we see that not any regular parameterized curve can be used as a local parameterization of a curve. is a diffeomorphism. (I.

locally. Also. for instance. at least. Let t0 ∈ I. we can write λ on the open neighbourhood r−1 (ρ( f (V))) of t0 as λ|r−1 (ρ( f (V))) = ρ−1 ρ( f (V)) ◦ r|r−1 (ρ( f (V))) = f ◦ pr1 |ρ( f (V)) ◦ r|r−1 (ρ( f (V))) . z) = f (x). Due to the regularity of the map ρ. the support of a parameterized curve. ρ−1 is the restriction of a differentiable mapping. it doesn’t make sense to speak about its differentiability. We can assume. it is enough to prove that λ is smooth in a neighbourhood of each point of the interval I. locally. this is not true. locally. unless the curve is simple. it follows that λ is smooth on the open neighbourhood r−1 (ρ( f (V))) of t0 .e. there is an inverse function τ = f (x). τ0 = λ(t0 ). that the first component of this vector is different from zero. Moreover r = ρ ◦ λ. It follows from the definition that any regular curve is. We will prove. λ is smooth on the entire I. As the functions f. defined. as composition of two smooth functions. this time. on an open subset of R3 . we have ρ (τ0 ) 0. From the inverse function theorem. Then. One might be tempted. by representing λ. changing the roles of r and ρ. defined and smooth in an open neighbourhood V ⊂ R of the point x0 = x(τ0 ). in the open neighbourhood ρ( f (V)) of the point (x(τ0 ). Since the notion of differentiability is a local notion. With this local expression of ρ−1 in hand. which means that. in . y. all of them. smooth on the indicated domains. as its domain of definition is not an open subset of the ambient Euclidean space. where pr1 : R3 → R is the projection of R3 on the first factor.24 and Chapter 1. without restricting the generality. pr1 and r are. Globally. to write λ as λ = ρ−1 ◦ r and conclude that λ is smooth. While this representation is legitimate. y(τ0 ). instead that. i. z(τ0 )) from M we will have ρ−1 (x. x (τ0 ) 0. applied to the function x. all we have to prove is that the maps λ and λ−1 are both smooth. Space curves ρ−1 : ρ(J) → J. in fact. we have ρ−1 ρ( f (V)) = f ◦ pr1 |ρ( f (V)) . at this point. as both r and ρ are homeomorphism onto. This could be achieved. The smoothness of λ−1 is proven in the same way. ρ−1 is not a differentiable function and. as a composition of smooth functions. As t0 was arbitrary. Therefore. for the time being.

clearly. Let us consider the mapping ψ : I × R2 → R3 . ψ is.3.3). r = r(t)) a regular parameterized curve.2. r is continuous. given by ψ(t. we can always restrict the domain of definition of a parameterized curve such that the support of the restriction is a regular curve. The regularity of r at each point means. even bijective. However. Take. r(t) = (x(t). z(t))). because limt→−∞ r = limt→∞ r = r(0) (see the figure 1.1. Figure 1. where (u. smooth and its Jacobi matrix at the point (t0 . v) = r(t) + (0. v) ∈ R2 . In fact. Let (I. Then each point t0 ∈ I has a neighbourhood W ⊂ I such that r(W) is a simple regular curve. y(t). The definition of the curve 25 general. for instance. but the inverse is not continuous. we may assume that x (t0 ) 0. the support of an arbitrary parameterized curve is not a regular curve. Without restricting the generality. v). 0) is . the support has a self intersection. Proof. where   x(t) =      y(t) =     z = 0 t(1+t2 ) 1+t4 t(1−t2 ) 1+t4 . the lemniscate of Bernoulli (R. u. 0.3: The Bernoulli’s lemniscate Theorem 1. in particular. that r (t0 ) 0.3. u.

0. 0). 0. We choose an arbitrary local parameterization (I. all the precautions should be taken when we investigate the global properties of regular curves. assume that the plane π coincides with the coordinate plane xOy and we shall use. 0. 0) such that ψ|V is a diffeomorphism from U to V. We shall. if it is invariant under parameter changes. for regular curves. It just tells us that any local property of regular parameterized curves is valid. If we put W := {t ∈ I : (t. 0)}. is. The previous theorem plays a very important conceptual role. z(t))) of the curve. clearly. therefore ψ is a local diffeomorphism around the point (t0 . Of course. Accordingly. y(t). r(t) = (x(t).4 1. usually. W is an open neighborhood of t0 in I such that ϕ(V ∩ r(W)) = ϕ(ψ(W × {(0. 0) = y (t0 ) 1 0       z (t0 ) 0 1 Its determinant is det J(ψ)(t0 . then. 0. r(I) is the . 0. 1. equally. Space curves    x (t0 ) 0 0         J(ψ)(t0 . Then the support r(I) of this local parameterization is an open subset of the curve.1 Analytical representations of curves Plane curves A regular curve M ⊂ R3 is called plane if it is contained into a plane π. 0. Remark. Parametric representation. 0) ∈ U} . 0) = x (t0 ).26 given by Chapter 1. For a global parameterization of a simple curve.4. this time). of course. 0) and V ⊂ R3 of ψ(t0 . a diffeomorphism from V to U. Let us denote by ϕ : V → U its inverse (which. only the coordinates x and y to describe such curves. without the assumptions of the parameterized curves being homeomorphisms onto. therefore. there exist open neighborhoods U ⊂ R3 of (t0 . also. 0)}) = W × {(0.

Nevertheless. Generally speaking.  The equation y = f (x) (1. we cannot use the same set of equations to describe the points of an entire curve.1) are called the parametric equations of the curve in the neighbourhood of the point a. also.1)  y = y(t) .4.4. . is a simple curve. Thus.4. if at the point (x0 .1. Then its graph C = {(x. y0 ) in R2 . C is not a regular curve (we can only say that it is a closed subset of the plane). Let f : I → R be a smooth function.3) be the 0-level set of the function F.2) (1. D ⊂ R2 .2). In the literature for the explicit representation of a curve it is.  The equations (1.4) is called the explicit equation of the curve (1. y0 ) 0. defined on an open interval from the real axis.4. then. unless the curve is simple. an explicit representation can be thought of as a particular case of parameter representation (the parameter being the coordinate x). defined on a domain C = {(x. used the term nonparametric form. which we do not find particularly appealing since. ∂y F} is non vanishing. by the implicit functions theorem there exist: • an open neighborhood U of the point (x0 . y0 ) ∈ C the vector grad F = {∂ x F. and let Let F : D → R be a smooth function. y) ∈ D | F(x.4. f (x)) | x ∈ I}. Explicit representation.4.4. Implicit representation. Usually. in fact. y) = 0} (1. any point a of the curve has an open neighbourhood which is the support of the parameterized curve   x = x(t)   (1. Analytical representations of curves 27 entire curve.5) (1. which has the global parameterization  x = t    y = f (t) .4. for instance ∂y F(x0 .

1. or it can be proved that the number of connected components resulted by removing a point is homeomorphisms-invariant. we will get. On the other hand. we cannot claim that the equation represent a curve in . An neighbourhood of the origin of C is the intersection of an open neighbourhood of the origin in the plane and the set C.4: The bisectors of the coordinate axes Examples. y0 ) = {2x0 . such that C ∩ U = {(x. the vector grad F cannot vanish on C and. only two connected components. y) = x2 + y2 − 1. then C is a regular curve (although. since x0 + y2 = 1.4). Figure 1. If the gradient of F is zero at a point. clearly. y0 ) ∈ C = {(x. Now. In fact. the set C has a self intersection at the origin (C is just the union of the two bissectors of the coordinate axes. see figure 1. If we remove from the interval the image of the origin through the homeomorphism f . The point is that there is no neighbourhood of the origin (on C). It might not be obvious why we encounter problems in the neighbourhood of the origin for this “curve”. Let (x0 . F : R2 → R. F(x. defined on an open neighborhood I ⊂ R of the point x0 .28 Chapter 1. y) = x2 − y2 . F(x. It should be clear that the condition of nonsingularity of the gradient of F is only a sufficient condition for the equation F(x. Space curves • a smooth function y = f (x). if we restrict the neighbourhood of the origin in the plane. f (x)|x ∈ I}. If grad F 0 in all the points of C. 2y0 }. suppose there is a homeomorphism f from the cross to an open interval from the real axis. the remaining set will have four connected components. Then we have grad F(x0 . centred at the origin). If we remove the origin from the cross. 2 Obviously. y) = 0 to represent a curve. in general. F : R2 → R. homeomorphic to an open interval on the real axis. C is 0 a curve (the unit circle. Remark. y) ∈ R2 | x2 + y2 − 1 = 0}. its intersection with the set C will be a cross. thus. not a simple one). 2. C is not a curve in this case (the gradient is vanishing at the origin).

4. the equation F(x. or only a neighbourhood of one of its points. g : I → R are two smooth functions. either. representing explicitly a curve actually means representing it as an intersection of two cylindrical surfaces. y) = 0}. with the two families of generators having orthogonal directions. a straight line).9) is the equation of a cylindrical surface. −(x − y)} and if we denote C = {(x.4. Consider. in fact.1. Then we have grad F(x.4. with a global parameterization given by  x = t       y = f (t) . Therefore. y) = 2{x − y. as a trivial example.4.9) (1. then the set C = {(x. Explicit representation. clearly.7) (1. g(x)) ∈ R3 | x ∈ I} is a simple curve. . with a local parameterization   x = x(t)      (1.e. but we cannot claim the opposite.8) are called the explicit equations of the curve. i.4. then grad F = 0 at all the points of C.2 Space curves Parametric representation. Let us note that. defined on an open interval from the real axis. If f. 1. As in the case of plane curves. y) ∈ R2 | F(x. Analytical representations of curves 29 the neighborhood of that point. But. each equation of the system (1.4.6) y = y(t)     z = z(t) we can represent either the entire curve.4. with the generators parallel to one of the coordinate axis. C is a curve (it is easy to see that it is the first bissector of the coordinate axes. y) ≡ (x − y)2 = 0. f (x).    z = g(t) The equations  y = f (x)    z = g(x)  (1.

Clearly. C ∩ U is a simple curve. y. while the pair (W. If the rank of the matrix (1.10) In the general case. z0 ) ∈ C the rank of the Jacobi matrix ∂ x F ∂y F ∂z F ∂ xG ∂yG ∂zG (1. y. G(x. Indeed. Space curves Implicit representation. This curve is obtained as the intersection between the sphere with the center at the origin and radius 2a and the circular cylinder Vincenzo Viviani (1622–1703) was a Italian mathematician and architect. then C is a curve (although. y.4. y0 . z) ∈ D | F(x.    G(x. Then. z) = 0.11) is equal to two everywhere. Let F.  (1. not a simple one). Nevertheless. who was in contact with Galileo Galilei in the last years of the great scientist. y. We consider the set C = {(x. z = g(x) are smooth functions. r(t) = (t. Example (The Viviani’s temple). in other words. if at the point a = (x0 .10) in U — is a curve. defined on a domain D ⊂ R3 .30 Chapter 1. g(x))|x ∈ W}.4. the solutions’ set for the system  F(x. then there is an open neighborhood U ⊂ D of the point (x0 . 4 . while y = f (x). suppose. z) = 0. and who liked to recommend himself as “Galileo’s last student”. there is an open neighborhood U ⊂ D such that the set C ∩ U can be written as C ∩ U = {(x.11) is equal to two. y0 . An important example of space curve given by implicit equations is the so-called temple of Viviani4 . the set C is not a regular curve. f (x). from the implicit functions theorem. where W is an open neighborhood in R of the point x0 .4. g(t)) is a global parameterization of it. G : D → R. for instance. z) = 0}. generally. that det ∂y F(a) ∂z F(a) ∂yG(a) ∂zG(a) 0. f (t). defined on W. y. z0 ) such that C ∩ U — the set of solutions of the system (1. z) = 0.4.

=2 = x−a y 0 2(x − a) 2y 0 G x Gy Gz To get a singular point.1. z) = 0. Analytical representations of curves 31 with radius a and axis parallel to the z-axis. The curve has a shape which is similar to a Bernoulli’s lemniscate. the Viviani’s temple (see figure 1. situated at a distance a from this axis. y. the shape of Viviani’s temple is easy to understand. z) = x2 + y2 + z2 − 4a2 . the equations of the curve is x = 2a. 2a sin ).4.  It is instructive to make some computations for the case of the Viviani’s temple. y. We will have to eliminate a point to get. 2 with t ∈ (−2π. the support of the parameterized curve t r(t) = (a(1 + cos t). y.    G(x. globally.    (x − a)2 + y2 = a2 . it is not. in fact. also. a sin t. indeed. lying on the surface of a sphere. F x Fy Fz x y z 2x 2y 2z . 2 . 2π). If we compute r (t) we get t r (t) = {−a sin t. yz = 0     (x − a)z = 0 Clearly. now. So.  We have. the only solution of the system that verifies. y. the following system of equations has to be fulfilled  y = 0      . let  F(x. a curve. a regular curve. It is not difficult to show that Viviani’s temple is. z) = (x − a)2 + y2 − a2 . a cos }.5) is a regular curve everywhere except the point which has these coordinates. As we shall see. the equations of the Viviani’s temple are   x2 + y2 + z2 = 4a2 . Thus. In other words. a cos t. In fact. z) = 0. y = z = 0.  Then the equations of the curve reads  F(x.    G(x.

in fact. then the straight line passing through r(t0 ) and having the direction of the vector r (t0 ) is called the tangent to the curve at the point r(t0 ) (or at the point t0 ).5.5 The tangent and the normal plane. y = z = 0 is.5: Viviani’s temple which means that this parameterized curve is regular. 1. thus: R(τ) = r(t0 ) + τr (t0 ). the existence of this regular parametric representation of the Viviani’s temple shows that the point of coordinates x = 2a. The vectorial equation of the tangent line reads. If the point t0 is regular. rather then a singular point. a point of self-intersection. The normal at a plane curve Definition 1. In particular. Space curves Figure 1.5.1. (1.32 Chapter 1.1) . For a parameterized curve r = r(t) the vector r (t0 ) is called the tangent vector or the velocity vector of the curve at the point t0 .

defined through any local parameterization of the curve around that point. Let d(∆t) = d((r(t0 + ∆t).5. (1. Let (I. The tangent vectors of two equivalent parameterized curves at corresponding points are colinear. π). Then. a sin t0 . r = r(t)) and (J. while the tangent lines coincide.5. Nevertheless. as we have seen. r(t0 + ∆t) = r(t0 ) + ∆t · r (t0 ) + ∆t · . Then. Remarks. according to the chain rule. 33 Thus. bt0 + τb) = =(a(cost0 − τ sin t0 ). but not the direction. The cylindrical helix has the parameterization r(t) = (a cos t. We can use a more “geometric” way to define the tangent to a parameterized curve.e. according to the Taylor’s formula. r = ρ(λ(t)). only the orientation and the length of the tangent vector can vary. a sin t0 + τa cos t0 . We consider an arbitrary straight line π. r and ρ have the same sense when λ > 0 (the parameter change does not modify the sense in which the support of the parameterized curve is traversed) and they have opposite sense when λ < 0. 1. a(sint0 + τ cos t0 ). r (t0 ) = {−a cos t0 . i. Clearly. passing through r(t0 ) and . b(t0 + τ)). therefore. a sin t.1.1. 2. with λ (t) 0. bt). b}. Thus. through a local parameterization. ρ = ρ(s)) the two equivalent parameterized curves and λ : I → J the parameter change. for a point t0 . Proposition 1.2) with lim having the direction given by the unit vector m. The tangent and the normal plane Example.5. Let r(t0 + ∆t) be a point of the curve close to the point r(t0 ). it doesn’t make sense to define the tangent vector at a point of a regular curve. it makes sense to speak about the tangent line at a point of a regular curve. the equation of the tangent is R(τ) =(a cos t0 − τa sin t0 . Since the parameter change modifies the tangent vector. r (t) = ρ (λ(t)) · λ (t). def ∆t→0 = 0. Proof.

Remark. The distance d(∆t) is equal to ∆r × m = |∆t| r (t0 ) × m + × m . Another way of interpreting this formula is that the tangent line is the limit position of the straight line determined by the chosen point and a neighbouring point on the curve. therefore r (t0 ) × m = 0. if not specified otherwise.2.5. Conversely. then the normal to the curve at the point r(t0 ) will be the straight line through r(t0 ).e. Remark. The condition (1. all the parameterized curves considered will be regular.5.3) is expressed by saying that the tangent line and the curve have a first order contact (or tangency contact). ∆t→0 |∆t| →0 Now.34 Chapter 1. π is the tangent line at t0 . either the vectors r (t0 ) and m are colinear. Let r = r(t) be a parameterized curve and t0 ∈ I.5. . ∆t→0 lim d(∆t) = lim r (t0 ) × m + × m = r (t0 ) × m . Definition 1. From the Taylor’s formula (1. If r = r(t) is a plane parameterized curve (i. by using an arbitrary local parameterization around that point. since the parameterized curve is regular). The straight line π is the tangent line to the parameterized curve r = r(t) at the point t0 iff d(∆t) lim = 0.e. (1. Space curves Theorem 1. the same is true for the normal plane (or the normal line. Hereafter. therefore either r (t0 ) = 0 (which cannot happen. i.2).3) ∆t→0 |∆t| Proof.5. then the vectors r (t0 ) and m are colinear. if the straight line π is the tangent line at t0 . which is perpendicular to the tangent line to the curve at the point r(t0 ). when the neighbouring point is approaching indefinitely to the given one. in the case of plane curves ).3) is fulfilled. Thus. if the condition (1. then r (t0 ) × m = 0.1.5. we have ∆r ≡ r(t0 + ∆t) − r(t0 ) = ∆t · r (t0 ) + ∆t · . Because it makes sense to define the tangent line at a point of a regular curve.5. which we will assume to be identical to the coordinate plane xOy). The normal plane at the point r(t0 ) of the curve r = r(t) is the plane which passes through r(t0 ) and it is perpendicular to the tangent line to the curve at r(t0 ). its support is contained into a plane.

5.5.      (1. Expanding the scalar products.    Y(τ) = y(t ) + τy (t ).9) for the equation of the normal plane to a space curve and. for the normal line to a plane curve. for space curves and {X − x. respectively X−x Y −y = .1.1 The equations of the tangent line and normal plane (line) for different representations of curves Parametric representation If we start from the vectorial equation (1. (1. The tangent and the normal plane 35 The vectorial equation of the normal plane (line) follows immediately from the definition: (R − r(t0 )) · r (t0 ) = 0. expressing it as {X − x.4) 1. x y z for space curves. (X − x)x + (Y − y)y = 0. for plane curves. Z − z} · {x .5.5.10) .5.5.5.  X(τ) = x(t0 ) + τx (t0 ).7) (1. z } = 0. we can obtain it from (1.. Y − y.e.4).5. i. (1.     Z(τ) = z(t0 ) + τz (t0 ). and.5. (1. As for the equation of the normal plane (line).5. for plane curves. we get: (X − x)x + (Y − y)y + (Z − z)z = 0. for space curves.  X(τ) = x(t0 ) + τx (t0 ).  0 0 If we eliminate the parameter τ.8) (1.1) of the tangent line and project it on the coordinate axes. y } = 0. y .5) Y(τ) = y(t0 ) + τy (t0 ).6) for plane curves. we get the canonical equations: X−x Y −y Z−z = = . we obtain the parametric equations of the tangent line.5. x y (1. Y − y} · {x .

For the derivatives we obtain immediately the expressions  x = 1       y = f    z = g . we have the parametric representation  x = t    y = f (t).  and. f (x) g (x) (1.5.5. when substituted into the equations (1.5.5.11) while for the equation of the normal plane. Space curves If we have a space curve given by the equations  y = f (x)    z = g(x) . the equation of the tangent line is X−x= Y − f (x) f (x) (1.9). we obtain X − x + (Y − f (x)) f (x) + (Z − g(x))g (x) = 0.  then we can construct a parameterization  x = t      y = f (t)     z = g(t).36 Explicit representation Chapter 1. give X−x= Y − f (x) Z − g(x) = . thus. which.12) .5. after substituting the derivatives into the equation (1. For a plane curve given explicitly y = f (x).7).13) (1.

The tangent and the normal plane or. f (x).  y z x . z0 ) det Fy Fz Gy Gz 0 1 (X − x).e. as we saw before.5. y) = 0    G(x.5. f (x) 37 (1. the curve can be represented as  y = f (x)    z = g(x). the system (1.  (1.5.  Let us suppose that.  x y z therefore   f F + g F = −F   y z x   f G + g G = −G . in a more familiar form.1.  i. g(x)) = 0    G(x.17) Then.17) can be written as  F(x. Y − f (x) = f (x)(X − x). y) = 0. f (x).5.15) (1.5.18) By computing the derivatives with respect to x of F and G. y0 .5. g(x)) = 0.14) (1.5.16) (1. while for the normal we get X − x + (Y − f (x)) f (x) = 0 ore Y − f (x) = − Implicit representation Let us consider a curve given by the implicit equations  F(x. around this point. at a point (x0 . we get the system  F + f (x)F + g (x)F = 0  x  y z  G + f (x)G + g (x)G = 0.

one obtains F x + f Fy = 0 =⇒ f = − Fx . then.5. F F x not D(F.x) D(F. G) −F x Fz = = z Gz G x −G x Gz D(z. from where. y) = 0.z) D(F.G) D(y. y = f (x). G) Fy −F x = = x .y) D(F. y)      f =            g =      D(F. Fy . locally. Gy −G x G x Gy D(x.G) D(z.x) = Z − z0 D(F.G) D(y.18) the equations of the tangent are X − x0 = or.G) D(y.x) D(F.G) D(z. f (x)) = 0.z) .z) (1.G) D(z.19).G) D(y. z) hyp 0. Space curves From this system.y) D(F.G) D(x.5.y) .z) Y − f (x0 ) Z − g(x0 ) = f (x0 ) g (x0 ) = Z − g(x0 ) D(F. having in mind that f (x0 ) = y0 and g(x0 ) = z0 . y0 ) there is fulfilled the condition Fy 0. through the Cramer method: ∆= ∆f = ∆g = therefore Fy Fz not D(F. G) = Gy Gz D(y. x) F Fy not D(F.z) = Y − y0 D(F. for the curve (1. For a plane curve F(x.38 Chapter 1. hence the equation of the curve can be written as F(x. one can obtain f and g . from the implicit functions theorem. if at the point (x0 . X − x0 = Y − f (x0 ) D(F. using (1.G) D(x. By differentiating this relation with respect to x.19) As we saw before. we have X − x0 D(F.G) D(y.G) D(x.5.

6.1. while for the normal we obtain the equation (X − x0 )Fy − (Y − y0 )F x = 0. r) be a parameterized curve and t0 ∈ I – a biregular point.1. usually. We find this term a little bit misleading. The osculating plane of the curve at r(t0 ) is the plane through r(t0 ).e. It is not difficult to check that the notion of a biregular point is independent of the parameterization: if a point is biregular for a given parameterized curve.6.6. The osculating plane Thus. i. Definition 1. X − x0 Y − y0 Z − z0 x y z = 0. x y z (1. Let (I. from the equation of the tangent: Y − y0 = f (x0 )(X − x0 ). Fx (X − x0 ) Fy 39 1.6.1) (1. or. The parameterized curve is called biregular if it is biregular at each point5 . Remark. r (t0 ) × r (t0 ) 0. especially. then its corresponding point through any parameter change is. we deduce Y − y0 = − or (X − x0 )F x + (Y − y0 )Fy = 0. a biregular point. the equation of the plane is (R − r(t0 ).e. also.6. A parameterized curve r = r(t) is called biregular (or in general position) at the point t0 if the vectors r (t0 ) and r (t0 ) are not colinear. parallel to the vectors r (t0 ) and r (t0 ). r (t0 )) = 0.2) 5 A biregular curve is also called in some books a complete curve. . i. other meaning in the global theory of curves (and.2. r (t0 ).6 The osculating plane Definition 1. expanding the mixed product. surfaces). since this terms has.

2 = 0. Remark. passing through r(t0 ). more general. therefore. Proof.1. ∆t→0 . As they have a common point (since r(t0 ) = ρ(λ(t0 )). at the same time. From the previous theorem it follows that the notion of an osculating plane makes sense also for regular curves. Let r(t0 ) si r(t0 + ∆t) two neighboring points on a parameterized curve. Let (I. they generate the same vector subspace R3 . hence the osculating planes to the two parameterized curves at the corresponding points t0 and λ(t0 ) have the same directing subspace. and denote d(∆t) = d(r(t0 + ∆t). ρ (λ(t))} are equivalent. The osculating planes to two equivalent parameterized curves at corresponding biregular points coincides. there is a more geometric way of defining the osculating plane which is. r (t) = ρ (λ(t)) · λ (t). they are parallel. Theorem 1. d(∆t) = |e · (r(t0 + ∆t) − r(t0 ))| = 1 = |(e · r (t0 )) · ∆t + (e · r (t0 )) · (∆t)2 + (e · ) · (∆t)2 |. As in the case of the tangent line.40 Chapter 1.e. ∆t→0 |∆t|2 i. From the Taylor formula we have 1 r(t0 + ∆t) = r(t0 ) + ∆t · r (t0 ) + (∆t)2 · r (t0 ) + (∆t)2 · . r (t)} and {ρ (λ(t)). r (t) = ρ (λ(t)) · (λ (t))2 + ρ (λ(t)) · λ (t).6. i. from these relations it follows that the system of vectors {r (t). α is the osculating plane to the parameterized curve r = r(t) at the biregular point r(t0 ) iff d(∆t) lim = 0. r = r(t)) and (J. Then r(t) = ρ(λ(t)). Space curves Theorem 1. since it can be applied also for the case of the points which are not biregular. of normal versor e. with r(t0 ) ¸ biregular.6. We consider a plane α. ρ = ρ(s)) be two equivalent parameterized curves and λ : I → J – the parameter change. Since λ (t) 0. the curve and the plane have a second order contact.e. they have to coincide.2. α). Proof. 2 with lim On the other hand.

in the sense that it has a second order contact with the curve. comes from the Latin verb osculare. i. The osculating plane Thus. among all the planes that are passing through a given point of a curve. but in that case any plane passing through the tangent is an osculating plane. If a biregular parameterized curve is plane. e r (t0 ) × r (t0 ).e. on can define the osculating plane at biregular point of a parameterized curve as being the limit position of a plane passing through the tangent at the given point and a neighboring point from the curve. i. the osculating plane has the higher order (“the closer”) contact. meaning that α is the osculating plane. A natural question that one may ask is what happens with the osculating plane in the particular case of a plane parameterized curve. when this point is approaching indefinitely the given one. then e · r (t0 ) = 0 and e · r (t0 ) = 0. Saying that the osculating plane at a biregular point is the only plane which has at that point a second order contact with the curve is the same with saying that the osculating plane is the limit position of a plane determined by the considered point and two neighbouring points when this ones are approaching indefinitely to the given one. ∆t 2 = 0. .e.6. The answer is given by the following proposition. which means to kiss end emphasizes the fact that. The converse is obvious. Also. Remarks. the name (coined by Johann Bernoulli). then the curve is plane and its support is contained into the osculating plane. (i) The previous theorem justifies the name of the osculating plane. if a given biregular parameterized curve has the same osculating plane at each point. Actually. whose prove is left to the reader: Proposition 1. then the osculating plane to this curve at each point coincides to the plane of the curve. Conversely.6. one can define the notion of an osculating plane also for the points which are not biregular.1.1. e · r (t0 ) 1 d(∆t) = lim + · (e · r (t0 )) + e · 2 ∆t→0 |∆t| ∆t→0 ∆t 2 lim = lim If lim d(∆t) 2 ∆t→0 |∆t| ∆t→0 41 = →0 e · r (t0 ) 1 + · (e · r (t0 )) . its support is contained into a plane π. (ii) Defining the osculating plane through the contact.

We choose as natural parameter the arc length of the curve. r = r(t)) be a regular parameterized curve. Thus. equivalent to it. we have k(t) = k(t) = ρ = ρ ×ρ .7. then. ρ1 = ρ1 ( s)) is another ˜ equivalent naturally parameterized curve with the parameter change s = λ(s).1) Now. if (J1 . Indeed. which expresses the fact that the two vectors are orthogonal. λ we get |λ (s)| = 1 for any s ∈ J. One can show that ρ (s) does not depend on the choice of the naturally parameterized curve equivalent to the given curve r = r(t). ρ = ρ(s)) be a naturally parameterized curve.1. s = ±s + s0 .42 Chapter 1. Then we have r(t) = ρ(s(t)) ⇒ r (t) = ρ (s(t)) · s (t) r (t) = ρ (s(t)) · (s (t))2 + ρ (s(t)) · s (t). The vector k = ρ (s(t)) is called the curvature vector of the parameterized curve r = r(t) at the point t. where s (t) = r (t) . r 4 (1. We shall express now the curvature vector k(t) as a function of r(t) and its derivatives. Then ρ (s) = 1. while ρ has unit length. we have ρ 2 = 1.7. ˜ 1 ( s). we have dt r (t) r r 2 k(t) = ρ (s(t)) = − r ·r ·r . Differentiating this relation. while the vector ρ (s) is orthogonal to ρ (s)6 . s (t) = r (t) = d √ r (t) · r (t) r (t) · r (t) = . where s0 is a constant. 6 Indeed. therefore. we obtain that ρ ·ρ = 0. since the vectors ρ and ρ are orthogonal. Definition 1. k(t) = ρ (s(t)) – the curvature of the parameterized curve at the point t. from ˜ the condition ρ (s) = ρ1 (λ(s)) = 1 Thus.7 The curvature of a curve Let (I. Space curves 1. since ρ is a unit vector. . It follows that ˜ ρ (s) = ρ ˜ 1 ( s) (λ (s))2 +ρ 1 · λ (s) = ρ =1 =0 = ±1 and. while its norm. Let (J.

7. r (t) = R. r (t) · r (t) = 0. R sin t. Proof. If the curvature of a regular parameterized curve is identically zero. 1. 0 = − {x. as the following proposition shows. r 3 43 (1. also the curvature) is identically zero.1. z = 0} we choose the parameterization   x = R cos t      0 < t < 2π. the result still holds also for non-naturally parameterized curves. i. Proposition 1. thus.1) we obtain ρ k(t) = r ×r .7. 2. 1. 0}. therefore ρ (s) = a = const. y. .7. in some sense. the notion of curvature makes sense also for regular curves. r (t) = {−R cos t. z) ∈ R3 |x2 + y2 = R2 . and ρ by the formula (1. As two equivalent parameterized curves have the same support.7. ρ = ρ(s)). Therefore. For the straight line r = r0 + ta the curvature vector (and. Suppose. 1. Examples. − sin t.1. 1 2. Since for equivalent parameterized curve the naturally parameterized curve to each of them are equivalent between them.2) it follows that a parameterized curve r = r(t) is biregular at a point t0 iff k(t0 ) 0. ρ(s) = ρ0 + sa. R Remarks. From the formula (1. then the support of the curve lies on a straight line. for the curvature vector we get 1 1 1 k(t) = − cos t.  y = R sin t    z = 0 Then r (t) = {−R sin t. The computation we just made explains why the inverse of the curvature is called the curvature radius of the curve.e. We saw that the curvature of a straight line is identically zero. z}. that we are dealing with a naturally parameterized curve (I.2) Remarks. R cos t. The converse is also true. therefore. for the very beginning. 0} and. The curvature of a curve Substituting ρ = r . From the hypothesis.7. For the circle S R = {(x. 2. ρ (s) = 0. the support ρ(I) lies on a straight line. y. R R R k(t) = 1 .

as we also saw earlier. 2 sin ∆ϕ(s) 2 r(s + ∆s) − r(s) = k(s) = r (s) = lim = lim ∆s→0 ∆s→0 ∆s |∆s| ∆ϕ(s) ∆ϕ(s) sin 2 ∆ϕ(s) dϕ = lim = lim · = . r = r(t)) one can construct a frame of the space R3 . 1. which means that the velocity and the acceleration of the curve are parallel. the fact that a parameterized curve has zero curvature simply means that the support of the curve lies on a straight line. ∆ϕ(s) ∆s→0 ∆s→0 ∆s ∆s ds 2 ∆ϕ(s) .1 The geometrical meaning of curvature Let us consider a naturally parameterized curve (I. nor that it is equivalent to such a particular parameterized curve. The curvature of a curve is the speed of rotation of the tangent line to the curve. r = r(s)).7. vector from R3 . the parameterized curve r : R → R3 . this parameterized curve is not equivalent to an affine parameterized curve. as we did before. that it might be easier to do that if we don’t use the standard coordinate system of R3 . when the tangency point is moving along the curve with unit speed.2. Then we have. hence the curve has zero curvature but.8 The Frenet frame (the moving frame) of a parameterized curve At each point of the support of a biregular parameterized curve (I. Then r(s + ∆s) − r(s) = 2 sin Therefore. the last formula gives us: Proposition 1. We denote by ∆ϕ(s) the measure of the angle between the versors r(s) and r(s + ∆s). immediately. nonvanishing. We can consider. where a is a constant. r = r0 + at3 . 1. but it doesn’t necessarily means that the parameterized curve is (the restriction of) an affine map from R to R3 . 2 Thus. The idea is that.7. Be careful. if we have in mind that ∆ϕ(s) is the measure of the angle between the tangents to the curve at s and s + ∆s. but a coordinate . if we want to investigate the local properties of a parameterized curve around a given point of the curve. Space curves Remark. r (t) = 2at2 and r (t) = 3at.44 Chapter 1.

• β(t0 ) = τ(t0 ) × ν(t0 ) it is called the unit binormal at the point t0 . r (t0 ) τ(t0 ) is also called the unit tangent at the point t0 . this straight line is contained into the normal plane (since it is perpendicular on the tangent). The binormal is the normal perpendicular on the osculating plane.1.e. • ν(t0 ) = k(t0 )/k(t0 ) is the versor of the curvature vector: ν(t0 ) = k(t0 ) k(t0 ) and it is called the unit principal normal at the point t0 . β(t0 )} is the normal plane (N in the figure 1. In fact. • The axis of direction ν(t0 ) is called the principal normal. with the origin at the point r(t0 ). while the coordinate axes have some connection with the local properties of the curve. • The plane determined by the vectors {ν(t0 ).8. • The axis of direction β(t0 ) is called the binormal. β(t0 )}.8. r = r(t)) at the point t0 ∈ I is an orthonormal frame of the space R3 . Definition 1. The Frenet frame (or the moving frame) of a biregular parameterized curve (I.8). The Frenet frame 45 system with the origin at the given point of the curve. but it is also contained into the osculating plane. independently.8). obviously. at the middle of the XIX-th century. Thus. • The plane determined by the vectors {τ(t0 ). i.1. where: • τ(t0 ) is the versor of the tangent to the curve at t0 . the tangent to the curve at t0 . by the French mathematicians Frenet and Serret. ν(t0 ). . the principal normal is the normal contained into the osculating plane. Such a coordinate system was constructed. the coordinate versors being the vectors {τ(t0 ). τ(t0 ) = r (t0 ) . • The axis of direction τ(t0 ) is. ν(t0 } is the osculating plane (O in the figure 1.

1)  ρ (s)      β(s) ≡ τ(s) × ν(s) = ρ (s) × ρ (s)   ρ (s) For an arbitrary biregular parameterized curve (I. for reasons that will become clear later (R in the figure 1. τ(t) = r (t) .8.8. Thus. ρ = ρ(s)). the expressions of the vectors of the Frenet frame are quite simple:  τ(s) = ρ (s)       ν(s) = ρ (s)  . For a naturally parameterized curve (J. from the definition. r = r(t)) the situation is a little bit more complicated.46 Chapter 1. (1. Space curves Figure 1. at an arbitrary point t ∈ I. r (t) (1.8). obviously.6: The Frenet frame of a parameterized curve • The plane determined by the vectors {τ(t0 ).2) . β(t0 )} is called the rectifying plane .

1. then the origins and the unit principal normals coincide. Since r(t) = ρ(λ(t)). the tangent vectors ρ (u) and r (t) have opposite senses. but opposite senses. From β = τ × ν it follows that. the curvature vectors of two equivalent curves coincide and.8. r = r(t)) and ρ = ρ(u) be two equivalent parameterized curves with the parameter change λ : I → J.1. If λ (t) < 0. thus. for an arbitrary parameterized curve (I. the same is true for the unit principal normals. their Frenet frame coincide if λ (t) > 0 .1 The behaviour of the Frenet frame at a parameter change A notion defined for parameterized curves makes sense for regular curves iff it is invariant at a parameter change. The Frenet frame Then. the origins of the Frenet frames coincide in any case.e. r (t) × r (t) r (t) r (t) · r (t) − · r (t) 2 r (t) r (t) 4 r (t) × r (t) . having in mind that k(t) = and k(t) = we get ν(t) ≡ while β(t) ≡ τ(t) × ν(t) = k(t) r (t) r (t) · r (t) = · r (t) − · r (t). r (t) 3 47 (1. equivalent to it. while the other two pairs of versors have the same direction. the unit binormals also have opposite senses.8. Let (I. r = r(t)) it is easier to compute directly τ and β and then compute ν by the formula ν = β × τ. in this case. Then. The above computations show that. The Frenet frame is “almost” invariant. we have Theorem 1. in practice. k(t) r (t) × r (t) r (t) · r (t) × r (t) r (t) × r (t) .8.8. in other words if it doesn’t change when we replace a parameterized curve by another one.1. As seen before. From r (t) = ρ (λ(t)) · λ (t) it follows the coincidence of the Frenet frames for λ (t) > 0.8. u = λ(t). . Proof. If λ (t) < 0. i. at the corresponding points t and u = λ(t).4) Remark. and the same is true for their versors.3) (1.

Definition 1. rb ) and (Iβ . therefore the two parameterized curves are. b α α α β b = r−1 (C 0 ). π). 2π). π) 2 and the parameter change is the identity.48 Chapter 1. C12 . rα = rα (t))}α∈A such that a) C = α∈A rα (Iα ). Therefore. to make sure that the parameter changes will not modify the Frenet frames. with λ (t) > 0. r = r(t)) and (J. 0)) and (I2 = (−π. at least not at any parameter change). rb = rα |Iα . Space curves 1.2. sin t. λ(t) = t. C12 = r1 (I1 )∩r2 (I2 ) has two connected components (the upper and the lower half circles). π). ∀t ∈ I.9 Oriented curves. imposing some further condition on the local parameterization. π). clearly. 1 Starting with the upper component.1. b b) For any connected component Cαβ of the intersection Cαβ = rα (Iα ) ∩ rβ (Iβ ) with b b b b α. positively equivalent.9. rb ) with Iα = r−1 (Cαβ ). The idea is to modify a little bit the definition of the regular curve. Definition 1.9. Two parameterized curves are (I. rb = r | are positively equivalent. in order to be able to use this apparatus also for regular curve. β ∈ A the parameterized curves (Iα . ∀t ∈ (0. we need to make it invariant. sin t. For the unit circle S 1 the following parameterizations: (I1 = (0. r2 (t) = (cos t. ρ = ρ(u)) are called positively equivalent if there is a parameter change λ : I → J. we have 1 1 I1 = r−1 (C12 ) = (0. 0)) give an orientation of S 1 . u = λ(t). the Frenet frame of a parameterized curve is not invariant at a parameter change (well. . λ : (0. π). An orientation of a regular curve C ⊂ R3 is a family of local parameterizations {(Iα . π) → (0. r1 (t) = (cos t. 1 1 1 I2 = r−1 (C12 ) = (0. Iβ β Ib β β αβ β Example. The Frenet frame of an oriented curve As we saw before.

0) 2 49 2 2 and the parameter change is λ : I1 → I2 . π). C12 . . r) of an oriented regular curve C is called compatible with the orientation defined by the family {(Iα .9. if the direction of the tangent vector at x ∈ C is given by the vector a(x). this definition does not depend on the choice of the local parameterization. also this time the two local parameterizations are positively equivalent. is called an oriented regular curve. Clearly. The Frenet frame of an oriented curve 2 As for the lower connected component. Oriented curves. then wave to impose the continuity of the map C → R3 . it has only two distinct orientations. with the orientation given by the family of local parameterizations {(Iα . where r = r(t) is a local parameterization of the curve C. The Frenet frame of an oriented biregular curve C at a point x ∈ C is the Frenet frame of a biregular parameterized curve r = r(t) at t0 . Example. such that r(t0 ) = x.9. Definition 1. since λ (t) = 1 > 0. A regular curve C ⊂ R3 . it can be turned into an oriented curve by using an orientation given by any global parameterization (I. For this definition of the orientation. Remark. by using the vectors r α (t).9. x = r(t). x → a(x). with an orientation. a local parameterization (I. rα )}α∈A if on the intersections r(I) ∩ rα (Iα ) the parameterized curves (I. 1 1 I1 = r−1 (C12 ) = (0. compatible with the orientation of the curve. since. 2π). The orientation itself can be given through the choice of a sense on each tangent line. 1 1 1 I2 = r−1 (C12 ) = (0. Remark. passing to another local parameterization rβ (t). A local parameterization (I. r).5. For an oriented regular curve C. π) 2 Remark.4. compatible with the orientation. λ(t) = t − 2π. rα ) are positively equivalent. Thus.3. rα )}α∈A .1.9. r) is compatible with the orientation if. therefore. a sense on each tangent line. corresponding to the two possible senses of moving along the curve. If C is a simple regular curve. the vectors a(x) and r (t) have the same sense. we get 2 2 I1 = r−1 (C12 ) = (π. Definition 1. r) and (Iα . If C is a connected regular curve. one can define. 1 2 2 I2 = r−1 (C12 ) = (−π. the vectors r α and r β have the same direction and the same sense (only their norms may differ). Definition 1. for each point x ∈ C.

Space curves 1. where χ is a proportionality factor.       Thus. the vector β is colinear to the vector ν = β × τ and we can write β = − r · χν.10 The Frenet formulae. β = τ × ν ⇒ β = τ × +τ × ν = k( ν × ν ) + τ × ν ⇒ β = τ × ν . r = r(t)) be a biregular parameterized curve. β · β = 1 ⇒ β · β = 0 ⇒ β ⊥ β. ν. the meaning of which will be made clear later. β(t) are. more precisely. τ= Therefore. in fact. ν(t). On the other hand. We have ν = β × τ + β × τ = − r · χ(ν × τ) + r · k(β × ν). The torsion Let (I. Further. smooth vector functions with respect to the parameter t. These derivatives show. =0 (1.1) hence β ⊥ τ. the way the vectors of the Frenet frame vary along the curve. We want to find their derivatives with respect to t. . We differentiate now the equality ν = β × τ. Thus.50 Chapter 1. in fact. we have. Then the vectors τ(t). From the definition. the decomposition of these derivatives with respect to the vectors {τ.10. τ = r k · ν. β}. τ = r · r −r · r 2 r ·r r r r r = √ . r2 = r · r 2 − r (r · r ) = r 3 r ·r r · r ×r ν = r · r ×r r 3 k       r       r ×r    ·r −         · r .

10. k .3 ) ν (t) = −k(t)τ(t) + χ(t)β(t)     β (t) = −χ(t)ν(t). For such a curve.10. the versors of the Frenet frame are given by the expressions:  τ = ρ      ν = 1 ρ  k    β = 1 ρ × ρ . from the definition.3) These equations are called the Frenet formulae for the parameterized curve r = r(t).1. The quantity χ(t) is called the torsion (or second curvature) of the biregular parameterized curve r = r(t) at the point t.1.2) (1. we have: β · ν = −χ(s) · ( ν × ν ) = −χ(t). The Frenet formulae. We obtained. Definition 1. =1 But. 51 (1. β = hence χ = −β · ν = − 1 k ρ ×ρ =0 1 k + 1 1 ρ ×ρ + ρ ×ρ .10. then the Frenet equations are a little bit simpler:  τ (t) = k(t)ν(t)      (1. ν = r (−kτ + χβ). If we are dealing win a naturally parameterized curve ρ = ρ(s). on the other hand. the equations  τ (t) = r k(t)ν(t)       ν (t) = r (−k(t)τ(t) + χ(t)β(t))    β (t) = − r χ(t)ν(t). thus. We shall compute first the torsion for a naturally parameterized curve ρ = ρ(s).10. The torsion therefore. k From the third Frenet formula. k k =0 1 1 · ρ − ρ ×ρ k k 1 · ρ .10.

as well as the fact that the vector ν1 (λ(t)) has unit length. ν1 . r (t) = ρ (λ(t)) · λ (t) ⇒ r (t) = From the last Frenet equation for the curve r.52 therefore. we get β (t) · ν(t) = − r (t) · χ(t). Space curves 1 ρ . Then β1 (λ(t)) = β(t) ν1 (λ(t)) = ν(t).10. χ(t) = − 1 1 · β (t) · ν(t) = − · β (λ(t)) · λ (t) · ν1 (λ(t)) = r (t) ρ (λ(t)) · λ (t) 1 1 =− · − ρ (λ(t)) · χ1 (λ(t)) = χ1 (λ(t)). ρ (λ(t)) d (β (u))λ (t). β}. but this time for the curve ρ. β1 } be the Frenet frames of the two parameterized curves at the corresponding points t and uλ(t). r = r(t)) and (J. Let now ρ = ρ(s) be a naturally parameterized curve. . respectively {τ1 . respectively. the following theorem gives us the way to compute the torsion for an arbitrary (biregular) parameterized curve: χ= Theorem.e. Chapter 1.ρ .ρ . and χ and χ1 their – torsions at these points. du 1 where we used once more the last Frenet equation. then they have the same torsion at the corresponding points t and u = λ(t). Let {τ. If (I. i. ρ = ρ(u)) are two positively equivalent parameterized curves. ν. where s = λ(t) is the parameter change. Then r and its derivatives up to the third order can be expressed as functions of ρ and its derivatives as: r(t) = ρ(λ(t)) r (t) = ρ (λ(t)) · λ (t) r (t) = ρ (λ(t)) · λ 2 (t) + ρ (λ(t)) · λ (t) r (t) = ρ (λ(t)) · λ 3 (t) + 3ρ (λ(t)) · λ (t) · λ (t) + ρ (λ(t)) · λ (t).4) k2 Now. positively equivalent to the parameterized curve r = r(t). with the parameter change λ : I → J. (1. Proof. λ > 0.

In other words. The Frenet formulae. ρ (λ(t)) · λ 2 (t) + ρ (λ(t)) · λ (t). The torsion therefore the mixed product of the first three derivatives of r reads r (t). r 3 hence.10.1. r ) . from the expression of the torsion (1. 1. because two of the factors are colinear. we have .10. r . an analogue of the curvature (this is the reason why in the oldfashioned books the torsion is called the second curvature). ρ (λ(t)) · λ 3 (t) + 3ρ (λ(t)) · λ (t) · λ (t) + ρ (λ(t)) · λ (t) = = λ 6 (t) ρ (λ(t)).5) Exercise 1. r 6 Moreover (see (1. ρ (λ(t)).1 The geometrical meaning of the torsion The torsion is.1. 53 all the other mixed products from the right hand side vanishing. since ρ is naturally parameterized and the two curves are positively equivalent. ρ (λ(t)) . r ) . we have λ = r .10. we get χ(t) = (r . ρ (λ(t)). r (t) = ρ (λ(t)) · λ (t). Let ω = χτ + kβ. in a way. r . r ×r 2 (1.7. this time the binormal.10. therefore the previous formula becomes ρ . Show that Frenet’s formulae can be written as  τ = ω × τ      (1. Hence ρ (λ(t)). λ (t) 6 But. What we mean is that the torsion can be also interpreted as being the speed of rotation of a straight line.10. r (t). the curvature can be expressed with respect to the derivatives of r through the formula r ×r k= .ρ .10.2)). r (t) . ρ (λ(t)) = 1 r (t).4) and the previous relation. r (t).6) ν = ω × ν     β = ω × β The vector ω is called the Darboux vector.ρ = (r .

β(t) = const.10. which.54 Chapter 1. unlike the case of the curvature. i. whose absolute value is the curvature and which will help us to get some more information about the curve. We have to say. The support of a biregular parameterized curve lies in a plane iff the torsion of the curve vanishes identically. that the curvature of space curve is defined to be positive because one could find no geometrical meaning of a signed curvature. then the unit binormal. therefore the vector r (t) is always perpendicular to the constant vector β0 . 1.1. β0 . But β(t) = r (t) × r (t). β(t).10. We could expect . If (I. Therefore. As we said before. obviously. Then. is the osculating plane of the curve. hence we have 0 = β (t) = − r 0 ·χ(t) · ν(t) ⇒ χ(t) ≡ 0. the vectors r (t) and r (t) are parallel to this plane. Proof. 0 Conversely. Space curves Proposition 1. the torsion is a measure of the deviation of the curve from a plane curve. is always equal to a constant vector. and passing through the point r0 . then we have χ(s) = lim ∆s→0 ∆α ∆s Notice that. r = r(s)) is a naturally parameterized curve and ∆α is the angle of the osculating planes of the curve at r(s) and r(s + ∆s) (in other words. however. this time. On the other hand. the torsion is the algebraic value of the limit. r = r(t)) be a biregular parameterized curve such that r(I) ⊂ π.1. the curvature is a measure of the deviation of a curve from a straight line. Let (I.e. where π is a plane. we know that the curvature of a circle is constant and equal to the inverse of the radius of the circle. the torsion is analogue to the curvature. not the absolute value. the support of r(I) the curve is contained in a plane perpendicular to the constant vector β0 . As we shall see in the sequel. the angle of the binormals of the curve at that points). we have Theorem 1. On the other hand.2 Some further applications of the Frenet formulae We saw that the curvature of a parameterized curve vanishes identically if and only if the support of the curve lies on a straight line. as we know. for plane curve we can define a signed curvature.10. Thus. we have the series of implications r (t) · β0 = (r · β0 ) = 0 ⇒ r · β0 = const = r0 · β0 ⇒ (r − r0 ) × β0 = 0. More precisely. Thus. if χ(t) ≡ 0.

then its support lies on a circle. say r1 (s) ≡ c = const.10.10. where r1 = r + 1 ν.10.3.2. the weaker result: Proposition 1. while the torsion vanishes identically. r1 = r1 (s)). also. we have r2 = a2 . using the second Frenet formula for r. centred at c. we get r−c = 1 1 ν·ν = . If (I. Differentiating once more. Another interesting situation is when the support of the curve lies not in a plane. from (*). then the support of the curve lies on a circle of radius 1/k0 . the curve is plane. constant torsion). this claim is not true. Let us consider the parameterized curve (I. We have. Since the torsion is identically zero. which has constant curvature (and. the curve r1 reduces to a point. i.e. Unfortunately. The Frenet formulae. . r = r(s)) has the support on a sphere centred at the origin and radius equal to a. k0 k0 which means that any point of the support of the curve r is at a (constant) distance 1/k0 from the fixed point c. Proof. k0 (*) We differentiate with respect to s and we get. But. in other words that if the curvature of a parameterized curve is constant. In this case we have Proposition 1. i. however. we get r · r = 0 or r · τ. the support r(I) lies on the circle of radius 1/k0 . The torsion 55 that the converse is also through.e. r1 = r + 1 1 ν = τ + (−k0 τ) = τ − τ = 0. we obtain r · τ + r · τ = 0.1. but on a sphere. If a naturally parameterized curve (I. It is enough to think about the circular cylindrical helix. then its curvature is subject to 1 k≥ . r = r(s)) is a naturally parameterized curve with a curvature k equal to a positive constant k0 . k0 k0 Thus. The distance from a point of the curve to the origin is equal to r . Differentiating. a Proof.

. |r · ν| r · ν a In the figure ?? we give an example of a curve which lies on a sphere. r) is called a general helix if its tangents make a constant angle with a fixed direction in space.3 General helices. because it is both a general helix (see the next section) and a spherical curve. we have |r · ν| ≤ r · ν = r = a. Figure 1. it is a so-called spherical helix.10.56 or 1+r·τ =0 Chapter 1. Lancret’s theorem Definition 1.10. A parameterized curve curve (I. therefore.2. Space curves ⇐⇒ 1 + kr · ν = 0 =⇒ kr · ν = −1. k = |k| = 1 1 1 ≥ = . From the properties of the scalar product.7: A spherical helix 1.

let us assume that χ(s) = c0 = const. having in mind that c is constant and using the second formula of Frenet: (−kτ + χβ) · c = 0 which leads to −k · cos α0 + χ · sin α0 = 0. k > 0. A. at each point of the curve. As. it follows that c · ν = 0. The torsion 57 The following theorem was formulated in 1802 by the French mathematician Paul Lancret. Proof. de Saint Venant (1845). to begin with. by hypothesis. To prove the first implication. By differentiating the relation ν · c = 0 we get. i. c ⊥ ν.e. k Conversely. This means that c is in the rectifying plane and therefore β · c = sin α0 . Let us assume. known especially for his contributions to mechanics. hence k · c · ν = 0. we get τ · c = 0. By differentiating.2 (Lancret.10.10.e.. let us suppose that r is a general helix and let c be the versor of the fixed direction: τ · c = cos α0 = const. Theorem 1. but the first known proof belongs to another celebrated French mathematician. i. χ = cot α0 = const. that the curve is parameterized by the arc length. 1802). The Frenet formulae. k(s) . A space curve with the curvature k > 0 is a general helix if and only if the ratio between its torsion and its curvature is constant.1.

We integrate once and we obtain c0 τ + β = c∗ . other curves with the same family of principal normals. for a Bertrand curve. we get. the first one: on a general helix. Hence the vectors c si τ make a constant angle and the curve is a general helix. Saint-Venant). It turns out that if a Bertrand curve has more than one . principal normal or binormals. The other implication was given and proved later by Joseph Bertrand (see. the answer is easily seen to be negative: the family of tangents uniquely determines the curve. On the other hand. was answered by Joseph Bertrand. with him. for an arbitrary curve. For the principal normals.4 Bertrand curves An interesting problem in the theory of curves is whether it is possible for several curves to share the same family of tangents. or that they are associated. the problem. also. These curves are called Bertrand curves. 1. For the tangents. from the first and the third of the Frenet’s equations.58 or Chapter 1. raised by the same Saint-Venant. as a historical curiosity.10. that. (c0 · k − χ)ν = c0 τ + β = 0. although in most of the books this theorem is credited to Lancret. c c0 τ + β 1 + c2 0 whence c·τ= c0 1+ c2 0 = const ≤ 1. who discovered that. where c∗ 0 is a constant vector. Space curves c0 · k − χ = 0. Usually. or conjugated Bertrand curves. the book of Eisenhart [16]). We will say that the two curves are Bertrand mates. however. the answer is negative. for instance. ¸ We end this paragraph by noting. there are special curves for which there might be. in fact he (and. only gave one implication. We define c∗ c0 τ + β c0 τ + β = c := ∗ = . there is only one curve having the same principal normals. the ratio between the torsion and the curvature is constant.

10. Proof. Clearly. if ν(s) is the versor of the principal normal of the first curve.10.10) The vector dr is tangent to the second curve. (1.e. The two points are called corresponding. We deduce.10. Then the second curve is. in what follows. ds ds ∗ (1.10) by ν. The torsion 59 Bertrand mate.11) . Then τ∗ = or.12) (1. also dependent on the arc length s of the first and we assume that r∗ (s) is the point on the Bertrand mate having the same principal normal as the first one at s.10.10.10.7) As the two curves have the same principal normals.1. We shall prove. the relation (1. a series of interesting results related to Bertrand curves.10. using the second Frenet formula.8) We differentiate the relation (1. therefore it is perpendicular both on ν∗ and ds on ν.10.10. We have the following result: Theorem 1.11) τ∗ = (1 − ak) ds ds τ + aχ ∗ β. The angle of the tangents of two associated Bertrand curves at corresponding points is constant. The Frenet formulae. dr∗ da = (1 − ak) τ + ν + aχβ. then we have r∗ (s) = r(s) + a(s)ν(s). in fact.10. a is a ds constant. We assume that the first curve is naturally parameterized. then it has an infinity and the curve (and all of its mates) is a circular cylindrical helix.13) dr∗ dr∗ ds = ds∗ ds ds∗ (1. ∗ ds ds (1. without respecting the historical order. that da = 0. As such.10. ds We denote by s∗ the arc length of the second curve. Let r and r∗ be two Bertrand mates.9) (1.10.7) with respect to s and we get dr∗ dr dν da = +a + r ds ds ds ds or. using (1. (1. by multiplying both sides of (1.3 (Schell). i. the second curve ought to have the versor of the principal normal ν∗ (s) = ±ν(s).10) turns into dr∗ = (1 − ak) τ + aχβ.10.

10. We differentiate the relation (1.4 (Bertrand).21) (1. dτ is colinear to ν∗ .18) = 0. ds∗ (1. we obtain that cos ω = (1 − ak) sin ω = aχ ds .10. (1. A curve r is a Bertrand curve if and only if its torsion and curvature verify a relation of the form a · k + b · χ = 1. hence ω is constant. Let us assume that r is a Bertrand curve. τ∗ and the versor of the binormal of the second curve„ β∗ .17) If we multiply scalarly both sides of the relation (1. on the ground of the first Frenet formula.10. The following theorem was proved by Joseph Bertrand and it is considered to be the central result of the entire theory of Bertrand curves. and.15) (1.14) where. Space curves Let ω be the angle of the tangents of the two curvs at the corresponding points. ds also to ν. dω ds dω = 0. thsu. can be written as τ∗ = cos ωτ + sin ωβ.19) .17) by τ and then by β and use ∗ the fact that. Then this angle is given by cos ω = ττ∗ .10.10. of course. In terms of ω.20) (1. ∗ (1.10. Proof.13) and (1.15).10. ds (1.10. ds∗ ds . Theorem 1.10.10. τ∗ is the versor of the second curve at the point r∗ (s). ds cos ω dω = 0.10. Comparing the relations (1.60 Chapter 1.10. β = ε (− sin ωτ + cos ωβ) .e.16) where ε = ±1.15) with respect to s and we get: dτ∗ dω dω = − sin ω τ + k cos ων + cos ω β − χ sin ων. we obtain the relations: sin ω i. with constant a and b. ds ds ds (1.10.

10. ds∗ (1.10. the relation (1.10.13) and we obtain.19).11). Let us suppose now.27) Taking into account these two relations.11) that ds∗ ds or. we differntiate the relation (1. β} is orthonormal we conclude from (1.10. indeed. Differentiating the relation (1. ds∗ (1.10. ν. that the relation (1. 1 − ak aχ 61 (1.7).30) whence it follows that ν∗ = ±ν.10.23) (1.19).10.7). .10.10..24) we obtain the relation (1. where a is the constant from the equation (1.10. Let r∗ be the curve given by (1.10. ds∗ ds whence χ In the same way follows that (1 − ak) ds = const.10.22) (1.10. as we saw earlier.29) ν.10.25) = (a2 + b2 )χ2 . by using the first and the third Frenet formulae for the curve r: dτ∗ ds = k(1 − ak) − aχ2 ν ds ds∗ or.10. using (1.26) (1. The Frenet formulae.19) holds. If we denote b = a ctg ω. The torsion Dividing side by side these two equalities.28) 2 2 = (1 − ak)2 + a2 χ2 (1. conversely. As the basis {τ. i. we get. r is a Bertrand curve.10.1.10.25).e.10. (1.10. using the first Frenet formula for the curve r∗ : k∗ ν∗ = k(1 − ak) − aχ2 ds ds∗ 2 (1. we obtain ctg ω = or 1 − ak = aχ ctg ω. ds = const.

and the curves of constant curvature are Bertrand curves. say.2.19) follows that (a + b · c) · k = 1. 1. that the curve is a general helix. . as. the curves of constant torsion (in particular the plane curves). then it has an infinity and it is a circular cylindrical helix. we obtain (a − a1 ) · k = (b1 − b) · χ. χ = c · k. We shall assume that 0 ∈ I. from (1. For a plane curve.11 The local behaviour of a parameterized curve around a biregular point Let (I. all of them congruent (they are.10. (a1 . if a curve r has more then one Bertrand mate. parallel to the given one).10. hence. β0 = β(0). Clearly. also χ) is a constant. Proof. any plane curve has an infinity of Bertrand mates. the constant a from the relation (1. this means that there are (at least) two distinct pairs of real numbers (a. Remark. r) be a naturally parameterized curve. ν0 = ν(0). when both the curvature and the torsion are constants. as one can see easily. which means that k (and. We shall make use of the following notations: τ0 = τ(0). Corollary 1. we can get an infinity of pairs of real numbers that verify (1.19) is the one that identifies the Bertrand mate of our curve. therefore. as an interior point and M0 ≡ r(0) is a biregular point of the curve. via the Lancret’s theorem. Subtracting side by side the two relations. They lie on circular cylinders which have the same axis of symmetry with the cylinder associated to the given curve.10. in fact. At least one of the coefficients (a − a1 ) and (b1 − b) is different from zero. for the case of a cylindrical helix. the principal normals are. Space curves Corollary 1. therefore the curve is a circular cylindrical helix. If a Bertrand curve has more than one mate. so our curve has an infinity of Bertrand mates. Thus. therefore the ratio between the curvature and torsion is a constant. b). with c – constant. This means. the normals. b1 ) such that a · k + b · χ = 1. all of them. a1 · k + b1 · χ = 1.1.10.19). The circular cylindrical helices.62 Chapter 1. As we saw from the proof of the theorem of Bertrand. However.10.

since r is naturally parameterized.11.11.1.11.5) (1. where we have made use of the second Frenet equation.11.11. Substituting s = 0 in (1.11.5). the relation (1. the Taylor expansion is 1 1 r(s) = r(0) + sr (0) + s2 r (0) + s3 r (0) + o(s3 ). Thus. we have r (0) = k(0) = k(0) · ν0 .6) (1.3) (1. from the definition of the curvature vector. Up to the third order.1) 2 6 We want to express the derivatives of r as functions of the Frenet vectors τ0 .11.2) (1.11. ν0 . We have.11. if we differentiate the relation r (s) = k(s) = k(s) · ν we get r (s) = k (s)ν + k(s)ν = k (s)ν + k(s)(−k(s)τ + χ(s)β) = = −k2 (s)τ + k (s)ν + k(s)χ(s)β. Moreover.11. r (0) = τ0 . On the other hand. β0 .7) 2 6 or 1 1 1 r(s) − r(0) = s − k2 (0)s3 + o(s3 ) τ0 + k(0)s2 + k (0)s3 + o(s3 ) ν0 + 6 2 6 1 + k(0)χ(0)s3 + o(s3 ) β0 . The position vector of a point of the curve with respect to this frame is exactly the . The local behaviour of a parameterized curve around a biregular point 63 We expand r in a Taylor series around the origin. we get r (0) = −k2 (0)τ0 + k (0)ν0 + k(0)χ(0)β0 .4) (1. (1. 6 (1. obviously.8) We consider now a coordinate frame with the origin at M0 and having as axes the Frenet axes.1) becomes 1 1 r(s) − r(0) = sτ0 + s2 k(0)ν0 + s3 −k2 (0)τ0 + k (0)ν0 + k(0)χ(0)β0 + o(s3 ) (1.11.

   2 2 χ2 (0) 3  z = 9 k(0) y . the intersection condition between the the plane and the curve is F(x(s). y(s). z(s)) = 0. close enough to 0.e. We choose as coordinate frame the Frenet frame of the curve r at the point M0 . (1.3) 2 6 We have.9)). the projection on the xOz-plane (the rectifying plane) is a cubic. z = z(s) are the local equations of the curve with respect to the Frenet frame (see (1.  2 6    1 3 + o(s3 ) z(s) = k(0)χ(0)s 6 It is not at all difficult to see that locally. if x = x(s).2) or 1 1 as + bk(0)s2 + −ak2 (0) + bk (0) + ck(0)χ(0) s3 + o(s3 ) = 0.12 The contact between a space curve and a plane We consider a naturally parameterized space curve (I.8) on the axes. (1. projecting (1. while the projection on the yOz-plane (the normal plane) is a semicubic parabola. τ0 . y = y(s).12.64 Chapter 1. 1.11. where M = r(s).12.10) z = 6 k(0)χ(0)x3 . With respect to this coordinate frame.9) y(s) = 1 k(0)s2 + 1 k (0)s3 + o(s3 ) . . i.12. r). z) ≡ ax + by + cz = 0.11.11. giving. ν0 . We consider a plane Π passing through the point M0 .11. we get   x(s) = s − 1 k2 (0)s3 + o(s3 )   6    (1. 1 1 1 1 a s − k2 (0)s3 + o(s3 ) + b k(0)s2 + k (0)s3 + o(s3 ) + c k(0)χ(0)s3 + o(s3 ) = 0 6 2 6 6 (1. several possibilities.   2    1 (1. in fact. that 0 is an interior point of the interval I and that the point M0 = r(0) of the curve is biregular. Thus. y. we have the following relations between coordinates. he cartesian equation of the plane Π will be of the form F(x.1) Now. as we did before. the projection of the curve on the xOy-plane (the osculating plane) is a parabola. We assume. the equations of the projections of the curve on the coordinate planes of the Frenet frame:  y = 1 k(0)x2 . now. Space curves −−→ −− vector r(s) − r(0) ≡ M0 M. R0 = {M0 . β0 }. Therefore.

12. The contact between a space curve and a plane 65 (a) The osculating plane (b) The normal plane (c) The rectifying plane Figure 1.8: The projection of a space curve on the planes of the Frenet frame .1.

1. we cannot specialize more. F(x(s). again. ν0 . we assume that 0 is an interior point of the interval I and that r(0) is a biregular point of the curve. then the coefficient of s2 in the intersection equation should also vanish. we conclude that the osculating plane has a contact or superosculation with the curve at the planar points of the curve. c) If we want a contact of order at least 2 (osculation contact). the straight line of equations y = 0.e. y = y(s). b) If a = 0. Thus. But this choice leads us to the equation z = 0 for the plane Π. where the torsion vanishes.1) where Ω(a. z) ≡ x2 + y2 + z2 − 2ax − 2by − 2cz = 0. i. looking at the intersection equation. Then the condition of the intersection will be. z = 0 in the coordinate frame considered by us). Then an arbitrary sphere passing through the point M0 will have. the plane is already completely determined (the osculating plane). the equation F(x. that the plane has a contact of order zero with the curve (intersection. in fact. in order to have higher order contact (superosculation). they have a single point in common). d) As the plane Π is completely determined by the condition of having a second order contact with the curve. The osculating sphere As in the previous paragraph.13 The contact between a space curve and a sphere. z(s)) = 0.13. τ0 . the contact is just of second order (osculation). with respect to this coordinate system. Space curves a) If a 0 (i. c) is the center of the sphere. we have osculation contact if we impose a = 0 and b = 0.e. We choose as coordinate frame the Frenet frame of the curve r at the point M0 . b. then the intersection equation has s = 0 as a double solution. (1. z = z(s) are the local equations of . where x = x(s). At all other points. we consider a naturally parameterized curve (I. Anyway. It can be observed immediately that in this case the plane Π passes through the tangent at the curve at the point M0 (which is. This may happen only if b = 0 (as the point M0 is biregular. r). y. the plane does not contain the tangent).66 Chapter 1. R0 = {M0 . which means that the curve and the plane have a contact of order 1 (tangency contact). β0 }. y(s). the curvature is not zero at this point).

a) If a 0.2) or 1 −2as + (1 − bk(0))s2 + (ak2 (0) − bk (0) − ck(0)χ(0))s3 + o(s3 ). i. z(s)) = s − k2 (0)s3 + o(s3 ) 6 2 2 67 + 1 1 k(0)s2 + k (0)s3 + o(s3 ) + 2 6 2 1 1 k(0)χ(0)s3 + o(s3 ) − 2a s − k2 (0)s3 + o(s3 ) − 6 6 1 1 1 − 2b k(0)s2 + k (0)s3 + o(s3 ) − 2c k(0)χ(0)s3 + o(s3 ) = 0. 2 6 6 + (1. as one can easily see. It intersects the osculating plane at M0 of the curve at the center of curvature at M0 of the curve. The osculating sphere the curve with respect to the Frenet frame at M0 . in this case. 3 (1. the discussion that follows is.e. here we have more cases to take into consideration. any sphere with the center on the curvature axis of the curve has an osculation contact with this one. then s = 0 is a simple solution of the intersection equation.13.13. Thus. is called curvature axis or polar axis of the curve. Thus. also. which justifies once more the denomination of tangency contact.1. c) For an osculation (second order) contact. similar to that from the case of the contact between a curve and a plane. This line. the center of the sphere lies on the intersection line of the planes x = 0 (the normal plane) and y = R(0).13. in our case.3) Now. Thus. which. k(0) (1. It is easy to see that in this case the tangent line at M0 of the curve lies in the tangent plane of the sphere at the same point. the coefficient of s2 in the intersection equation also has to vanish and we get b= 1 = R(0). we have 1 F(x(s). this center lies in the normal plane of the curve at M0 . iff a = 0. The contact between a space curve and a sphere. . b) The sphere and the curve have a first order contact (tangency contact) if and only if the intersection equation has a double zero at the origin. obviously.4) where R(0) is the curvature radius of the curve at the point M0 . This happens. Thus. Still.13. in this case the sphere and the curve have a contact of order zero (intersection contact). y(s). is parallel to the binormal of the curve at M0 . This means that the first coordinate of the center of the sphere is zero.

68

Chapter 1. Space curves

d) We shall say that the sphere has a contact of superosculation with the curve if they have a contact of order at least three, which means that in the intersection equation the coefficients of the of the powers up to three of s have to vanish simultaneously. It is usually claimed that there is a single sphere which has a contact of superosculation with the curve and this sphere is called the osculating sphere. In reality, the things are a little bit more subtle, as we shall see in a moment. 1) Let us assume, first, that the torsion of the curve at M0 does not vanish, i.e. χ(0) 0. In this case, as one sees immediately, between the sphere and the curve there is a superosculation contact if and only if we have  a = 0,      1 (1.13.5) b = k(0) ,    c = k (0) ,  k2 (0)χ(0) i.e. in this case the sphere is uniquely determined and we will call it osculating sphere 7 . 2) If χ(0) = 0, while k (0) 0, as one can see easily, the coefficient of s3 is never zero, therefore there is no sphere which has a superosculation contact with the curve. However, as we have seen in the previous paragraph, in this case the osculating plane has a superosculation contact with the curve and we can think of this plane as playing also the role of the osculating sphere, of infinite radius, though. 3) If both χ(0) and k (0) vanish, then the coefficient of s3 in the intersection equation vanishes for any c, in other words in this case any sphere which has an osculation contact with the curve also has a superosculation contact. Thus, now we have an infinity of osculating spheres.

1.14
1.14.1

Existence and uniqueness theorems for parameterized curves
The behaviour of the Frenet frame under a rigid motion

Definition 1.14.1. A rigid motion of R3 is a map D : R3 → R3 , D(x) = A · x + b, where A ∈ M3×3 (R) is an orthogonal matrix, At · A = I3 , with the determinant equal to one: det A = 1, while b ∈ R3 is a constant vector. The linear map A : R3 → R3 , A(x) = A · x is called the homogeneous part of the rigid motion.
7 In some books all the spheres which have an osculation contact with the curve are called osculating spheres, while the one which has a superosculation contact is called, accordingly, superosculating sphere

1.14. Existence and uniqueness theorems for parameterized curves

69

Remarks. (i) A rigid motion in R3 is just a rotation around an arbitrary axis followed by a translation. (ii) If we don’t ask det A = 1, we get, equally, an isometry of R3 . However, in this case (if det A = −1), a transformation D(x) = A · x + b does not reduce anymore to a rotation and a translation, we have to add, also, a reflection with respect to a plane. In many books the term “motion” implies only that the matrix A is orthogonal and for what we call “rigid motion” it is used the term “proper motion”. Definition 1.14.2. Let D : R3 → R3 be a rigid motion, with the homogeneous part A : R3 → R3 . The image of a parameterized curve (I, r = r(t)) through D is, by definition, the parameterized curve (I, r1 = (D ◦ r)(t)). Remark. Since A is a nondegenerate linear map, the image of a regular parameterized curve is, also, a regular parameterized curve. Theorem 1.14.1. Let D be a rigid motion of R3 , with the homogeneous part A, (I, r = r(t)) – a biregular parameterized curve, (I, r1 = D ◦ r) – its image through D and {r(t); τ(t), ν(t), β(t)} – the Frenet frame of the parameterized curve r at t. Then the frame {r1 (t); A(τ(t)), A(ν(t)), A(β(t))} is the Frenet frame of r1 at t. Proof. Let r(t) = (x(t), y(t), z(t)), D(x, y, z) = (x1 , y1 , z1 ), where   x1 = α11 x + α12 y + α13 z + b1      y1 = α21 x + α22 y + α23 z + b2     z1 = α31 x + α32 y + α33 z + b3 Then r1 (t) = (x1 (t), y1 (t), z1 (t)), hence r 1 (t) = A(r (t)), r 1 (t) = A(r (t)). (1.14.2)

(1.14.1)

Since A is a linear isometry, preserving the orientation of R3 , we have r 1 = A(r ) = r , r 1 · r r 1×r
1 1

=r ·r ,
1

= A(r × r ), r 1 × r

= r ×r .

70 Then:

Chapter 1. Space curves

r1 r A(r ) A(r ) = = =A = A(τ) r A(r ) r r r1 · r1 r1 r r1 = r1 − · A(r )− ν1 = r1 × r1 r1 · r1 × r1 r ×r r ·r − · A(r ) = A(ν) r · r ×r r × r1 = A(β). β1 = 1 r1 × r1 τ1 =

Consequence. The parameterized curves (I, r) and (I, r1 = D ◦ r) have the same curvature and torsion. Proof. We have k1 = r1 × r1 r ×r = r1 r = k.

For the torsion, the situation is slightly more complicated. We have, from the theorem, β1 (t) = A(β(t)) and ν1 (t) = A(ν(t)).

A being a linear operator, a similar result holds for the derivatives of the two Frenet vectors, i.e. we have β1 (t) = A(β (t)) and ν1 (t) = A(ν (t)).

Using the last Frenet equations for the two curves, we have the equalities −χ1 (t)ν1 (t) = A(−χ( t)ν(t)) = −χ(t)A(ν(t)) = −χ(t)ν1 (t), and, thus, the two torsions are equal, as claimed.

1.14.2

The uniqueness theorem

Theorem 1.14.2. Let (I, r = r(t)) and (I, r1 = r1 (t)) be two biregular parameterized curves. If k(t) = k1 (t), χ(t) = χ1 (t) and r (t) = r1 (t) ∀t ∈ I, then there is a single rigid motion D of R3 such that r1 = D ◦ r.

1.14. Existence and uniqueness theorems for parameterized curves

71

Proof. Let t0 ∈ I be an arbitrary point and D – the rigid motion of R3 sending the Frenet frame {r(t0 ); τ0 , ν0 , β0 } of the curve r at t0 into the Frenet frame {r1 (t0 ); τ10 , ν10 , β10 } of the curve r1 at the same point. Obviously, there is a single rigid motion with this property. Let (I, r2 (t= D ◦ r(t)) – the image of the curve r through D, and k2 , χ2 – the curvature and the torsion, respectively, of the parameterized curve r2 . Then k2 (t) ≡ k(t) ≡ k1 (t) χ2 (t) ≡ χ(t) ≡ χ1 (t) and, moreover, r2 (t) ≡ r1 (t) . Therefore, the vector functions τ1 (t), ν1 (t), β1 (t) and τ2 (t), ν2 (t), β2 (t) giving the Frenet frame are solutions of the same system of Frenet equations  τ = r k 1 ν   1    ν = − r1 k1 τ + r1 χ1 β      β = − r χ1 ν. 1 Since for t = t0 the solutions coincide, due to the uniqueness theorem for the solution of a Cauchy problem they have to coincide globally. In particular, we have τ1 (t) ≡ τ2 (t) or hence r1 (t) − r2 (t) = 0 ⇒ r1 (t) − r2 (t) = const. But for t = t0 , r1 (t0 ) − r2 (t0 ) = 0, hence the two functions coincide for all t, thus r1 (t) ≡ r2 (t) = D ◦ r(t). As for the uniqueness of D, we notice that, for any other point t1 ∈ I, since r1 ≡ r2 , D sends the Frenet frame of the curve r at t1 into the Frenet frame of the curve r1 at t1 . Remark. For naturally parameterized curves the condition r (t) = r1 (t) is always fulfilled. r1 (t) r1 (t) = r2 (t) r2 (t) ,

72

Chapter 1. Space curves

1.14.3

The existence theorem

Theorem 1.14.3. Let f (s) and g(s) be two smooth functions, defined on an interval I, such that f (s) > 0, ∀t ∈ I. Then there is a single naturally parameterized curve (I, r = r(s)) for which f (s) = k(s) ∀s ∈ I and g(s) = χ(s) ∀s ∈ I. This curve is uniquely defined, up to a rigid motion of R3 . Proof. Let {r0 ; T0 , N0 , B0 } be a direct orthonormal frame in mathbbR3 . We consider the system of linear ordinary differential equations  T (s) = f (s)N(s)      N (s) = − f (s)T(s) + g(s)B(s)     B (s) = −g(s)N(s) with respect to the vector functions T(s), N(s), B(s). If we denote X(s) = (T(s), N(s), B(s)), then the system (1.14.3) can be written as X (s) = A(s) · X(s), with   f (s) 0   0         0 g(s) . A(s) = − f (s)       0 −g(s) 0 In the theory of ordinary differential equations one proves that the system (1.14.5) has a single solution subject to X(s0 ) = (T0 , N0 , B0 ), where s0 ∈ I, while the columns of the matrix X(s0 ) are the vectors T0 , N0 , B0 of the initial orthonormal basis. We are going to show first that, for any s ∈ I the vectors din (T(s), N(s), B(s)) form an orthonormal basis. It is enough to show for any s ∈ I, X(s) is orthogonal, i.e. X t (s) · X(s) = I3 , for any s ∈ I. We have d t d d t X ·X = X (s) · X + X t · (X(s)) = X t (At X + AX) = X t (At + A)X. dt dt dt but, as A is skew symmetrical, At + A = 0, therefore d t X · X = 0 ⇒ X t · X = const. dt (1.14.5)

(1.14.3)

(1.14.4)

1.14. Existence and uniqueness theorems for parameterized curves

73

On the other hand, from the initial condition, X t · X (s0 ) = I3 , hence X t (s) · X(s) = I3 for any s ∈ I. Let us define now
s

r = r0 +
s0

T(s)ds,

(sol)

where r0 is the origin of the original frame, while T(s) is the first column of X(s). We are going to show that (I, r(s)) is the curved searched for. We have, clearly: r (s) = T(s), r (s) = T(s) = 1, r (s) = T (s) = f (s)N(s). We note immediately that r (s) × r (s) meterized curve. On the other hand, 0, therefore r(s) is a biregular naturally para-

r (s) = ( f (s)N) = f N + f N = f N + f (− f T + gB) = − f 2 T + f N + f gB, hence (r , r , r ) = (T, f N, − f 2 T + f N + f gB) = (T, f N, f gB) = f 2 g (T, N, B) = f 2 g.
=1

Now we have all we need to compute the curvature and the torsion: r ×r = | f (s)| = f (s), r 3 f 2 (s)g(s) χ(s) = = g(s), f 2 (s) k(s) = hence the curve r fulfils the conditions of the theorem. The uniqueness of r, up to rigid motions, follows from the previous theorem.

74

Chapter 1. Space curves

we shall discuss here notions which cannot be defined for space curves (such as the signed curvature). Γ) which. Theorem.2.2.1) be a family of plane parameterized curves. or which are easier to investigate and richer in contents for plane curves. all the curves are parameterized curves.CHAPTER 2 Plane curves 2. if not specified otherwise. λ) (2. 2. depending smoothly on a real parameter λ. Definition.2 Envelopes of plane curves In this section. In particular.1) is a parameterized curve (J. is tangent to a curve from the family.1 Introduction After the exposition of the theory of space curves in all the generality. (2. Let r = r(t. λ) rλ × r t = 0.2) (2. if not mentioned otherwise. we shall focus in this chapter on topics which are specific to the theory of plane curves.3) . The envelope of the family (2.2.2. The points of the envelope of the family r(t. λ) are subject to r = r(t. at each point.

for which r t = 0. corresponding to some value of the parameter λ.76 Chapter 2. If Γ is the envelope of the family (γλ ) and P is a point of Γ. then P is a tangency point between Γ and a curve from the family. the singular points of the curves from the family. 1. since r1λ = r t · tλ + rλ . (r t · tλ + rλ ) × r t = 0. Plane curves Proof. The set of points described by the equations (2. Thus. The tangency condition between Γ and γλ reads r1λ or r1λ × r t = 0 or.2.2. a > 0. λ). λ) = (λ + a cos t. λ + a sin t). λ.4) rt Example. Let us consider the family of curves r(t. and the theorem is proven. The equation rλ × r t = 0 can be written also as i j k xλ yλ 0 = 0 ⇔ xλ yt − xt yλ = 0 ⇔ xt yt 0 ⇔ xλ yλ = . it verifies r1 = r(t(λ). therefore. because r t × r t = 0.2) and (2. but.3) is called the discriminant set of the family γλ . P is on a curve γλ and.2. xt yt (2. the equation of Γ will be of the form r1 = r1 (λ). also. On the other hand. . 2. It includes not only the support of the envelope. t ∈ R. Remarks. also.

Envelopes of plane curves 77 Clearly. y. Then we have r λ = {1. we are dealing with a family of circles of radius a. λ) = 0 (2. we get the parametric equations of the envelope:   x(λ) = λ ± √ a    2  y(λ) = λ √ . i. we can represent it as   x = x(t.5) verify the system of equations  F(x. in fact. Eliminating t. The points of the envelope of a family of plane curves given through the implicit equation F(x. parallel to the first bissector of the coordinate axes.2. therefore. Locally.2. λ)    y = y(t.e. 1}.1.2.  . λ) = 0    F (x.1 Curves given through an implicit equation Proposition 2.2.2.2. y. a cos t}. 2. λ) = 0  λ (2. λ) . we can parameterize the curve. since the circles have no singular points) verify   x = λ + a cos t      y = λ + a sin t     x · y = x · y t t λ λ or   x = λ + a cos t      y = λ + a sin t     a cos t = −a sin t.6) Proof. a   2 i. the envelope is. the points of the envelope (and only them.e. y. r t = {−a sin t. a pair of straight lines. around each point of a the curve of the family. with the centres on the first bisector of the coordinate axes.

λ) ≡ (x − λ)2 + (y − λ)2 − a2 = 0. i. we get. y(t. λ). this time given through their implicit equation F(x. namely √ y = x ± a 2. λ).  x y λ λ λ But. from (2. therefore. when substituted into the equation of the family. Then the second equation of the discriminant set will be Fλ (x. gives λ= (x − y)2 = 2a2 . xλ = K xt . again. Plane curves By substituting into the equation of the family.2. λ) = 0. y. y.4). we obtain the system:  F x + F y = 0  x t  y t  F x + F y + F = 0 . x+y . We consider again the family of circles from the previous paragraph. we get F(x(t. yλ = Kyt . whence we get . differentiating with respect to t and λ.. 2 which. the second equation from above becomes K(F x xt + Fy yt ) + Fλ = 0 =0 or Fλ = 0.78 Chapter 2. whence. Example. with K = const. respectively.e. the same equations of the envelope as before. λ) = −2(x + y − 2λ) = 0.

(2. The evolute of r is. substituting in F. Suppose we are given a family of curves depending smoothly on two parameters. that µ = µ(λ). ϕ)       D(λ.3 Applications: the evolute of a plane curve Definition.9) 2. µ) = 0 Proof. as required. (2. r = r(t)) be a plane parameterized curve. Envelopes of plane curves 79 2. if.2. µ(λ)) = 0.2. λ and µ F(x. y. µ) = 0      ϕ(λ. by definition. and ϕ. µ(λ)) = 0. we get the third equation from (2. λ.  µ λ λ Eliminating the derivative µλ between the two equations. then the points of the envelope verify the system  F(x. From the equation ϕ(λ.2. µ) = 0    D(F. λ.8) .2.2.9). we can assume.7) where the parameters λ and µ are connected through a relation ϕ(λ.2.2. therefore. λ.2. the envelope of the family of normals of the curve. µ) = 0.  F(x.    ϕ(λ.2 Families of curves depending on two parameters Proposition 2. Let (I.2. .2.  Differentiating with respect to λ these two equations. (2. µ) = 0. µ) = 0. for instance. we get:  F + F µ = 0  λ  µ λ  ϕ + ϕ µ = 0. y. y.

6)):  F(X.3. The equations (2. Plane curves Proposition 2. Example. eliminating the parameter t. t) = (X − x(t)) · x (t) + (Y − y(t)) · y (t) = 0. For the ellipse   x = a cos t    y = b sin t  one gets. As we know. y(t)) are   y (x 2 + y 2 )   X = x −    xy −x y (2. the equation of the normal of a plane curve is F(X.  X =    Y =  or. since in this case the curves of the family are straight line and they have no singular points) are (see (2.10)    x (x 2 + y 2 )   Y = y +  xy −x y Proof.e. The curved described by this equation is called a lengthened astroid (see figure 2. a 3 X 3 + b 3 Y 3 = (a2 − b2 ) 3 . 2 2 2 2 2 a2 −b2 a b2 −a2 b cos3 t sin3 t . Y.3). after the computations.80 The following result holds: Chapter 2. t) = 0 .2. Y. The relations verified by the points of the envelope of the family of normals (and only by them. Y.2.10) follow now instantly.  t i.   x (t)X + y (t)Y = x(t) · x (t) + y(t) · y (t)     x (t)X + y (t)Y = x 2 (t) + x (t) · x(t) + y 2 (t) + y(t) · y (t)  . after solving this system of linear equations with respect to X and Y.2. t) = 0    F (X.2. The parametric equations of the evolute of the curve r = r(t) = (x(t).2.

.3. It turns out. Definition. in this particular case. defined by J(x. x).3.1. y} by π or multiplying the 2 complex number x + iy by the complex unit i (this is. we can obtain more information about the curve if we use a slightly different concept. the origin of the name). allowing the curvature to have a sign.3 The curvature of a plane curve As we saw. which will allow us to give the definition in a coordinate independent way.1: The evolute of an ellipse 2. actually. a) Jv · Jw = v · w. the curvature is always a positive scalar. y) = (−y. Remark.2. however that. The curvature of a plane curve 81 Figure 2. b) (Jv) · v = 0. Some obvious properties of the complex structure are collected into the following proposition: Proposition 2. The complex structure on R2 is the map J : R2 → R2 . To define the curvature of a plane curve we shall use a little technical trick. this concept of curvature can be equally applied for plane curves. Of course. Applying J simply means to rotate the vector {x. in the case of an arbitrary space curve.

Then u × v = [v · Ju] · k.1) r 3 . Plane curves All these properties follow immediately from the geometrical interpretation of the complex structure. Proof. Now.82 c) J(Jv) = −v (i. into this plane.e. also. if r is a plane curve. Anticipating a little bit.3. then the vectors r (t) and r (t) are situated. Definition. therefore. the quantity r · Jr k± = . v(x2 . x2 y2 0 On the other hand. We remember that the curvature of an arbitrary parameterized space curve r = r(t) can be computed by k(t) = r (t) × r (t) . x1 } = −x2 y1 + x1 y2 .3. whence the equality announced. the idea of the definition of the signed curvature is just to replace this absolute value with the component itself. Let u(x1 . we would like to say a few words about the kind of curvature we are going to define. We are now ready to define the curvature of a plane curve. y2 } · {−y1 . the cross product of the vectors u and v can be computed by i j k u × v = x1 y1 0 = (x1 y2 − x2 y1 ) · k. Therefore.2. J 2 = −id). Chapter 2. r (t) 3 Now. by definition. y1 ). The signed curvature of r is. Proposition 2. Let r = r(t) be a plane parameterized curve. (2. with the support lying into the coordinate plane xOy. their vector product is a vector which is directed along the z axis. To do that. [v · Ju] = {x2 . and. the following characterization of the vector product of two vectors in the plane will prove very useful. y2 ) ∈ R2 . As it is known. the norm of this vector is just the absolute value of the z-component.

r = r(t)) be a plane parameterized curve (J. then the signed curvature of r can be expressed as k± (t) = x (t)y (t) − x (t)y (t) x 2 (t) + y 2 (t) 3/2 .e. Let (I. Now. y(t)).3. According to the proposition 2. the sign of the second derivative of the function f . as it known from calculus. result. If y = f (x) is the explicit equation of a plane curve. the signed curvature is just the projection of the curvature vector on the z-axis. Proof. i. in fact. thus r ×r |k± | = = k. as the curvature vector is parallel to the z-axis. the graph of a real function of a single real variable) the sign of the signed curvature is. but important. the signed curvature of a plane curve is “almost” invariant at a parameter change. If r(t) = (x(t). Exactly as happens for the torsion of the space curves. Consequence.2. i. u = λ(t). we have.3. (1 + f 2 )3/2 Remark. Let (I. The previous consequence show that. for computational reasons. r 3 Another immediate. The curvature of a plane curve 83 Remark. with the parameter change λ : I → J..e. we have Theorem. ρ = ρ(u)) an equivalent parameterized curve.e. We have r(t) = ρ(λ(t)) r (t) = ρ (λ(t)) · λ (t) r (t) = ρ (λ(t)) · λ 2 (t) + ρ (λ(t)) · λ (t) r · Jρ = (ρ λ 2 + ρ λ· J(λ ρ ) = = λ 3 ρ · Jρ + λ λ ρ · Jρ = 0 = λ 3 ρ · Jρ r 3 = |λ 3 | ρ 3 .2. then its signed curvature is given by f (x) k± (x) = . the signed of the signed curvature is a n indication of the convexity or concavity of the function. is the following: Proposition 2. r = r(t)) be a plane parameterized curve. for an explicitly given curve (i.3. Then k± [ρ](u) = sgn(λ ) · k± [r](t).3.

naturally parameterized curve can be expressed easily as a function of the signed curvature: Lemma. except that. therefore it makes sense to define it also for regular oriented plane curves. r = r(t)) be a plane parameterized curve. Definition.e. Proof.1 The geometrical interpretation of the signed curvature For the signed curvature of plane curve we have a geometrical interpretation which is similar to the geometrical interpretation of the curvature of a space curve. whence α(s) = k± (s). Let (I. The curvature vector of a plane. On the other hand. this time the sign is also taken into account. i. r Jr .3. whence it follows that r ⊥ r or. We will need first the following definition. defined through: τ(t) = {cos θ[r](t). . The rotation angle of r is the function θ[r] : I → R. which is the same. θ[r] is the angle made by the unit tangent with the positive direction of the x axis. 2. then we should have r (s)· Jr (s) = α(s)·[Jr (s)]2 = α(s). Plane curves λ 3 ρ (u) · Jρ (u) r · Jr = 3 · = sgn(λ ) · k± [ρ](u). Let (I. from the definition of the signed curvature. r 3 ρ (u) 3 |λ | k± [ρ](u) = sgn(λ ) · k± [r](t). Then r (s) = k± (s) · Jr (s).3. k± (s) = r (s) · Jr (s). If we put r (s) = α(s)· Jr (s). The previous theorem shows that the signed curvature is invariant under any positive parameter change. r = r(s)) be a plane naturally parameterized curve. Remark. therefore r · r = 0. (2.84 therefore k± [r](t) = whence Chapter 2.2) where τ(t) is the unit tangent. sin θ[r](t)} = exp(iθ[r](t)). We have r 2 (s) = 1 (the curve is naturally parameterized).

for instance.3. is should be clear for anyone that θ is just the angle made by the unit tangent vector with the positive direction of the x-axis. θ is its rotation angle and k± – its signed curvature. in fact. In fact. After all. however. This definition is very innocent looking and natural. then the variation of the rotation angle is equal (as absolute value) with the variation of the contingency angle. In reality. r = r(t)) is a plane regular parameterized curve. If (I. if we use the expression of τ(t) as a function of the angle θ. we get for dτ the formula dt dτ dθ dθ = {− sin θ(t). cos θ(t)} = Jτ(t).2). The contents of this lemma is quite similar to the geometrical interpretation of the curvature of a space curve. we get the equality r (t) d + r (t) r (t) dt 1 r (t) = dθ dθ Jr (t) Jτ(t) ≡ · . dt dt Jr (t) Multiplying both sides by Jr (t) and having in mind that Jr (t) · r (t) = 0 and Jr (t) · Jr (t) = r (t) 2 . let alone a smooth one.2. r (t) dt . From the definition of the unit tangent vector we have τ(t) = dτ r (t) d = + r (t) dt r (t) dt 1 r (t) . we obtain: r (t) · Jr (t) dθ = · r (t) . dt dt dt Combining the two relations. the geometrical interpretation of the absolute curvature of a plane curve (regarded as a curve in space) is an immediate consequence of this lemma. dt Proof. Lemma. The following lemma provides the connection between the signed curvature and the variation of the rotation angle. for a modern proof. given by (2. r (t) r (t) . [1]) and any two such functions differ by an integer multiple of 2π. then: dθ = r (t) k± (t). The curvature of a plane curve 85 Remark. therefore.3. whence On the other hand. the book of Bär. when the plane curve is regarded as a particular case of space curve. it is by no mean obvious that we can find a continuous angle function. Such functions do exist (see. for an arbitrary plane curve.

after simplification. r = r(s)).4 The curvature center. k± (t) r (t) Remark. For a naturally parameterized curve. which is tangent to the curve at r0 = r(t0 ). Corollary. We defined the evolute of a plane curve as being the envelope of the normals of the curve. with t0 ∈ I. Plane curves whence. ds 2 which is exactly the formula for the curvature of arbitrary space curves. which. whence the position of the point Ω for an arbitrary t ∈ I: Ω(t) = r(t) + 1 Jr (t) . From the previous formula we get k ≡ k± = dθ . overall. the situation remains unchanged.86 Chapter 2. we could have problems only when λ < 0. thus. as expected. centred at Ω. it follows that dθ · r (t) = k± (t) · r (t) dt or. The following proposition provides a different approach. . such that the signed curvatures r and γ at r0 coincide. remains valid. with the parameter change λ : I → J. The evolute and the involute of a plane curve Definition. using the definition of the signed curvature. but in this case Jr changes the sense and k± changes the sign and. A point Ω ∈ R2 is called the curvature center at r0 = r(t0 ) of a plane parameterized curve r : I → R2 if there is a circle (γ). Obviously. then r (t) = ρ (λ(t))λ (t) and k± [r](t) = sgn(λ )k± [ρ](λ(t)). dθ = k± (t) · r (t) . 2. for the particular case of plane curves. ds dθ . The notion of curvature center is invariant at a parameter change: if (J. ρ = ρ(u)) is equivalent to r. we have k± (s) = Remark. dt which is exactly what we had to prove. (I.

if Ω(t) = (X(t). r = r(s)) be a naturally parameterized curve c ∈ I. ρ[r. we get immediately: Corollary. From the previous remark. Let (I.4. c](s) = r(s) + (c − s)r (s). The curvature center of the curve at an arbitrary value of the parameter is Ω(t) = r(t) + r (t) 2 x 2+y 2 · Jr (t) = (x(t). Definition.   Y(t) = y(t) + x (x 2 +y 2 )  x y −x y which are exactly the equations of the evolute.1. Another interesting plane curve associated to a given one is the so-called involute. The evolute of a plane curve is the geometrical locus of the curvature centers of the curve. Find the evolute of the cycloid   x(t) = a(t − sin t). the inverse of the evolute. r (t) · Jr (t) xy −x y Thus. Exercise 2. c] = ρ[r. The curvature center. Y(t)).a s we shall see immediately. Proof. projecting the previous equation on the coordinate axes. x }.e.4. The evolute and the involute of a plane curve 87 Proposition 2. The involute of r with the origin at r(c) (or at c.2.4. The definition of the evolute makes sense also for regular curves (i. for short) is the parameterized curve (I.1. which.4. we get the parametric equations of the locus of curvature centers:  2 2  X(t) = x(t) − y (x +y )   x y −x y . Exercise 2.2. Find the evolute of the astroid   x(t) = a cos3 t. y(t)) + {−y . c](s)) where ρ[r. is. in a sense.    y = a(1 − cos t)  . two equivalent parameterized curves have the same evolute). Show that the evolute is also an astroid (see the figure ??. Show that the evolute is also an cycloid (see the figure ??.    y = a sin3 t  . .

s is not a natural parameter along ρ.3: The evolute of an cycloid Remark. the natural parameter being the arc length. Generally speaking.2: The evolute of an astroid Figure 2. the we can replace the parameter t by the arc length s = t r (τ) dτ and define the involute of r as being the involute of 0 the naturally parameterized curve equivalent to it. r = r(t)) be a parameterized curve. Is is easy to see that the following proposition holds true: Proposition 2.4.2.88 Chapter 2. If (I. Let (I. The the involute of r with . Plane curves Figure 2. r = r(t)) is an arbitrary parameterized curve.

2.5. (c − at) {−a sin t. Example.4. r = r(s)) be a naturally parameterized curve and ρ the involute of r with the origin at c ∈ I. a cos t} a We represented in the figure 2. We have ρ (s) = r (s) + (c − s)r (s) − r (s) = (c − s)r (s) = (c − s)k± [r](s) · Jr (s) ρ (s) = −k± [r](s· Jr (s) + (c − s)(k± [r](s)) · Jr (s) + (c − s)k± [r](s) · Jr (s) = [−k± [r](s) + (c − s)(k± [r](s)) ] · Jr (s) − (c − s)(k± [r](s))2 · r (s). Let r(t) = (a cos t. sgn(k± [r](s)) . in projection. Then the signed curvature of ρ is given by k± [ρ](s) = Proof.  X(t) = a cos t − (c − at) sin t    Y(t) = a sin t + (c − at) cos t  . Then  r (t) = {−a sin t. |c − s| .4 an involute of a circle of radius 1. Lemma. a sin t) + or. The following lemma establishes the connection between the signed curvature of a parameterized curve and that of one of its involutes and will serve as a tool to establish the relation between evolute and involute. with the origin at the point of parameter 0. a sin t) be a circle. The curvature center. Let (I.     0 89 r (t) . r (t) hence the equation of the involute is ρ(t) = (a cos t. The evolute and the involute of a plane curve the origin at c ∈ I is given by ρ(t) = r(t) + (c − s(t)) where s = s(t) is the arc length of r. a cos t}     2   x + y 2 = a2     t    s(t) = adt = at.

Plane curves Figure 2. k± [ρ](s) ρ (s) Using the previous lemma to express the signed curvature of ρ as a function of the signed curvature of r. The conclusion follows now from the definition of the signed curvature. we get ρ1 (s) = ρ(s) + ρ1 (s) = r(s) + (c − s)r (s) + |c − s| (c − s)k± [r](s) · J 2 r (s) · = r(s).90 Chapter 2. The the evolute of ρ is r. sgn(k± [r](s)) (c − s)k± [r](s) · Jr (s) . r = r(s)) be a naturally parameterized curve and ρ – its involute with the origin at c ∈ I. In many textbooks this connection is taken. Proof. as known. by the equation Jρ (s) 1 · . in fact. The following theorem provides a connection between the involute and the evolute. as the definition of the involute. The evolute of ρ is given.4: An involute of a circle whence Jρ = −(c − s)k± [r](s) · r (s). Let (I. while ρ (s) · Jρ (s) = (c − s)2 · (k± [r](s))3 . Theorem.

Now. t2 . The osculating circle of a curve 91 2. with t1 < u1 < t2 < u2 < t3 such that f (u1 ) = f (u2 ) = 0. Let (I. when these points approach indefinitely the given one. then we will have. Then.  Since f is differentiable and f (t1 ) = f (t2 ) = f (t3 ). We assume that. t2 . t3 ) and f : I → R – the function defined through f (t) = r(t) − A 2 .5. t2 .2. we have: Theorem. f is smooth and we have:   f (t) = 2r · (r(t) − A)     f (t) = 2r (t) · (r(t) − A) + 2 r (t) 2 . r(t)2). v → t. t3 → t. Let A(t1 . u2 ) such that f (v) = 0. t3 ) be the circle passing through r(t1 ). On the other hand. t3 ). the osculating circle is the limit position of a circle determined by three neighboring points. at the limit. u2 . applying once more the mean value theorem. r(t3 ). if we let t1 . therefore. u1 . t2 . Let (I. when these points approach indefinitely the given one. with the radius equal to the 1 curvature radius k(t) of the curve at that point. this time to the derivative f . r = r(t)) be a plane parameterized curve and t1 < t2 < t3 ∈ I Let C(t1 . t3 ) be the centre of the circle C(t1 . t2 .5 The osculating circle of a curve Definition. from the mean value theorem it follows that there exist two points u1 . r = r(t)) be a parameterized curve. t1 →t t2 →t t3 →t Proof. clearly.  . More precisely. The osculating circle of r at a point t ∈ I is the circle centered at the curvature center Ω(t). Then the osculating circle of r at the point r(t) is the circle C = lim C(t1 . Exactly as the osculating plane at a point of a space curve can be regarded as a limit position of a plane determined by three neighboring points. equally. which is also differentiable . u2 ∈ I. it follows that there is a v ∈ (u1 . we should get:  r (t) · (r(t) − A(t)) = 0   (*)  r (t) · (r(t) − A(t)) = − r (t) 2 . for a value t ∈ I of the parameter we have k± (t) 0.

Let f : I → R be a continuous function. 2. it follows that r(t) − A(t) = −1 Jr (t) · .6 The existence and uniqueness theorem for plane curves The existence and uniqueness theorem for plane parameterized curves is similar to the analogue theorem for space curves and can be proved in the same manner. therefore we shall skip the proof here. k± (s) = f (s). Plane curves A(t) = lim A(t1 . . sin(αs + θ0 )} ds ds whence the system of differential equations:   dx = cos(αs + θ0 )  ds   dy  = sin(αs + θ )  0 ds . To give an example.92 where Chapter 2. for any real value of the parameter s.6. = {cos θ(s). t3 ). Then there is a regular.1. sin θ(s)} = {cos(αs + θ0 ). r is unique. k± (t) r (t) hence C is the osculating circle of the curve r at the point r(t). t2 . Starting from the geometrical interpretation of the signed curvature. up to a proper motion of R2 . from the definition of the rotation angle. we get α = k± (s) = dθ . t1 →t t2 →t t3 →t From (*) and the definition of the signed curvature. ds therefore θ (the rotation angle) will be a linear function of s: θ = αs + θ0 . where θ0 is a constant. we obtain: τ(s) ≡ dx dy . On the other hand. naturally parameterized curve (I. we will find the curve r for the particular case when the function f is a constant α. Theorem 2. r = r(s)) such that ∀s ∈ I.

(*)  y = 1 [− cos αs cos θ + sin αs sin θ ] + y  0 0 0 α where x0 and y0 are two integration constant.6. a rigid motion.2. The conclusion is that the only plane curve of constant positive curvature α is the circle of radius 1/α. centered at the origin. also.  0 sin αs α which shows that any plane curve of constant signed curvature. equal to α. But this is the circle of radius 1/α. i.   α . The solution (*) can be written in the matrix form   1 1  α sin αs sin θ0 + α cos αs sin θ0  x0 x   +   = 1  y  1 y 0 − α cos αs cos θ0 + α sin αs sin θ0 or. . can be obtained from the curve   x = 1 cos αs   α  y = 1 sin αs  α by applying a rotation followed by a translation. The existence and uniqueness theorem for plane curves 93 Since the equations are separated.e. the system can be integrated very easy and we get the solution:   x = 1 [sin αs sin θ0 + cos αs sin θ0 ] + x0 .   cos  x  =  − sin y π 2 − θ0 π 2 − θ0 sin cos π 2 π 2 − θ0 − θ0 1    α cos αs x0    +    1  y .

Plane curves .94 Chapter 2.

finding analytically a curve when the curvature and torsion are given is quite another matter. Moreover. which should allow further reductions. we can find the tangent versor and.CHAPTER 3 The integration of the natural equations of a curve 3. Clearly. In fact. if we manage to solve the Frenet system. This. we can find the general solution of a Ricatti equation if and only if a particular solution is already available (and there is no general . we shall prove this indirectly. As it is known from the theory of ordinary differential equations. Apparently. the Frenet system should be equivalent to a third order vectorial differential equation or to a system of three third order scalar equations. in particular. of course. in the general situation.1 The Riccati equation associated to the natural equations of a curve The general theory of differential equations guarantees the existence and uniqueness (up to a rigid motion) for the Frenet equations. However. amounts to finding an analytical solution for the Frenet equations. we can find the curve. it turns out that. therefore it should be equivalent to a single third order scalar equations. Nevertheless. the. by showing that the Frenet system is equivalent to a second order Ricatti differential equation. the three vectors of the solution are subject to the orthonormality conditions. it is not possible to integrate the system. by another quadrature. Although they look quite innocently. it turns out that the system (made of none scalar equations) actually contains three identical sets of three equations.

1.1. (3. ds 2 2 (3. u−v Z= u+v . relies exactly on the supplementary condition (3. as we mentioned earlier.3) Clearly.2). the condition X 2 + Y 2 + Z 2 = 1.1.2) The idea of the method we are going to describe (and which goes back to Sophus Lie and Gaston Darboux). Z in terms of u and v. Y. in the general situation.5) This is. in the previous section. that the Frenet system contains three copies of the scalar system:  X = k(s)Y.4) Thus. also.1. the solution of the Frenet system amounts to the finding of the complex functions u and v. first of all.1. u−v Y=i 1 + uv .96 Chapter 3. Notice.1) Y = −k(s)X + χ(s)Z     Z = −χ(s)Y and the solution of the system should verify.1. u−v (3. One can check easily that X= 1 − uv .      (3. It is now possible to express X. We introduce now the complex functions u and v by letting u= X + iY . Namely. it was observed that this equation can be decomposed (over the complex numbers) as (X + iY)(X − iY) = (1 − Z)(1 + Z). here. an indirect prove of the fact that the natural equations of a space curve cannot be integrated through quadratures. We shall give.2 Examples for the integration of the natural equation of a plane curve We saw. an easy manipulation of the formulas show that both u and v are solution of the Riccati equation dw i iχ(s) 2 = − χ(s) − ik(s)w + w . w and −1/v are conjugated to each other. that the plane curves of constant curvature are the circles. The integration of the natural equations of a curve procedure for finding such a solution). v 1−Z (3. 1−Z 1 X − iY − = . Now. 3. other interesting examples of natural equation for plane curves .

by integrating and neglecting the integration constant.3. ds dy = sin α. We remind that in the case of plane curves the problem is not the integrability of the natural equation. we shall use. In this case the curvature radius is given by R = a · s. the possibility of expressing the solution in terms of elementary functions. ds We have (neglecting. as we know. (3. rather. Let α be the contingency angle.1) where a is a constant. again. the parametric equations of the logarithmic spiral are  aα   x = ae (a cos α + sin α)     a2 + 1    aα   y = ae (a sin α − cos α)   a2 + 1 . we have dα 1 1 = = .2.2. analogously. Examples for the integration of the natural equation of a plane curve 97 that can be integrated. Then.1) is most conveniently described through its polar . (3. we get α= 1 ln s. y= sin α ds = aeaα (a sin α − cos α). Usually this is not the case. a2 + 1 cos αds = a cos αeaα dα = aeaα (a cos α + sin α) a2 + 1 Hence. For convenience. The logarithmic spiral. ds R(s) as therefore.2) The logarithmic spiral (see figure 3. and this makes even more interesting the (few) cases in which this possibility does exist. we have to integrate the system of differential equations dx = cos α. To get the coordinates. but. for the rest of this paragraph. the integration constants) x= and. the curvature radius R = 1/k instead of the curvature.2. a whence s = eaα .

i. the polar equation of the logarithmic spiral is of the form r = C · eϕ . They correspond to the natural equation s2 R 2 + = 1.2.98 Chapter 3. Cycloidal curves. Thus 2 2 r = sin ψeϕ−ψ− 2 . we notice that x 2 + y2 = whence a2 e2aα . The integration of the natural equations of a curve Figure 3. for the polar angle we get r= tan ϕ = y a sin α − cos α = = − cot(α + ψ).1: The logarithmic spiral equation. hence α = ϕ − ψ − π . x sin α + a cos α therefore ϕ = α + ψ + π . r = r(ϕ). Then. a2 + 1 aeaα . where C is a constant. We shall indicate now how can one obtain that equation starting from the parametric equations.2. a2 b2 (3.4) (3. x 2 + y2 = √ a2 + 1 We put a = tan ψ. First of all.3) π .e.

3. it is. for the clothoid (also known as the Cornu’s spiral). The contingency angle is easily found: α= therefore the coordinates are x= cos s2 ds. to introduce a new parameter t.2. where the radius of curvature was proportional to the arc length.e. We prefer. R = b cos t.2. s (3. b 2 a−b b a+b b The clothoid. in terms of the parameter t: x= y= cos α ds = sin α ds = cos a b (a − b)t b (a + b)t at · a · cos t dt = sin + sin . 2a (3.7) s2 . instead. then. due to the the presence of the square root and the sign ambiguity. this would lead to complications. very easy to find the contingency angle in terms of this new parameter. similar to the logarithmic spiral. From this point of view. We include this curve here to show that even in the case of a very simple expression of the curvature in terms of the arc length (in this case the curvature is proportional to the arc length. Examples for the integration of the natural equation of a plane curve 99 where a and b are nonvanishing constants. i. to some extent. It is. However. 2a (3. This is the curve whose natural equation is R= a2 . R b cos t b We can proceed now with the determination of the coordinates x and y. rather than to its inverse. Indeed.2. through the relations s = a sin t. the radius of curvature is proportional to the inverse of the arc length. b 2 a−b b a+b b sin at a b (a − b)t b (a + b)t · a · cos t dt = − cos + cos .5) Thus. we might not be able to find a parametric representation in terms of elementary functions.6) . A possibility would be to express R in terms of s and then integrate. we have 1 1 at α= ds = a cos t dt = .2. it is a linear function). 2a x= sin s2 ds.

Again. in fact. a (3.7) cannot be expressed in terms of elementary functions.10) 1 According to Gino Loria.2.2. the natural equation is R=a+ s2 . For the catenary.9) The coordinates are. i. analogously. in 1781. a · cos2 t R= a . instead of integrating directly.8) where a is a non-vanishing constant. They carry the name of Fresnel integrals. after the French physicists who used them first in his works on optics (the theory of diffraction. cos2 t (3.100 Chapter 3. Then the contingency angle will be α≡ 1 ds = R(s) cos2 t · a dt = t. first. these integrals were first considered by Euler. dt = ln cos t cos t It is not difficult to check that one can eliminate the parameter t from the previous two equation and one gets the usual explicit equation of the catenary. easy to find in terms on the new parameter: x= and. a (3. to be more specific)1 . The catenary. The integration of the natural equations of a curve Unfortunately. we introduce.2.e. . it is a well known fact from analysis that the integrals from (3. cos t cos αds = a cos t dt = a cos2 t 1 1 + sin t . y=a 1 . now. x y = a cosh .2. the new parameter t through the relations s = a tan t. to get the contingency angle.

cos α ds = at cos t dt = a(cos t + t sin t) 1 ds = R 1 · at dt = t. Finally.2. at at2 . . The tractrix.2. (3. Examples for the integration of the natural equation of a plane curve 101 Figure 3. where a is a non-vanishing constant. we start with the natural equation R2 + a2 = a2 e−2s/a . 2 R = at.2. such that s= Then the contingency angle is α= therefore x= and y= sin α ds = at sin t dt = a(sin t − t cos t). t.12) 1 .3. the parametric equations of the involute of the circle. cos t R = a tan t.11) which are. We introduce a new parameter.2: The catenary The involute of the circle. indeed. We introduce the parameter t such that e− a = s (3. Let us start with the natural equation R2 = 2as.

3: The tractrix Then α= therefore x= and y= sin α ds = sin t · a · tan t dt = a ln cos α ds = cos t(−a tan t) dt = a cos t 1 + sin t − sin t .102 Chapter 3. cos t 1 ds = R 1 (−a tan t) dt = −t. a tan t . The integration of the natural equations of a curve Figure 3.

is a constant quantity. An arbitrary half-line OE. We draw. Through the point O passes a half-line OC. 3. two straight line. On this half-line one takes a segment OM. starting from the origin O. the opposite of O on the circle. Find the equations of the geometrical locus of M ( the witch of Maria Agnesi). 2. with d(F1 . through the points D and E. A straight line OL rotates around the point O with a constant angular velocity ω. Find the equations of the trajectory of M (Archimede’s spiral). respectively. A point M is moving uniformly on a straight line ON. called the Diocles’ cissoid. where B is the intersection point between the half-line and the tangent to the circle at A. of length BC. . 4. equal to a2 (the curve is called a Cassini oval). which is rotating with constant angular velocity around the point O. F2 ) = 2b. Find the equations of this curve. intersects at the points D and E the circle a 2 a2 x2 + y − = 2 4 and its tangent at C. Write the equations of the geometric locus of the points for which the product of their distances to two fixed points F1 and F2 . Find the equation of the curve described by the point M (the logarithmical spiral). If we let the half-line OC rotate around the point O. We consider a circle of diameter OA = 2a and its tangent at the point A. 5. parallel to the axes Ox and Oy.Problems 1. then the point M describe a curve. The two lines intersect at M. where C is the intersection point between the half-line and the circle. A point M moves along the line OL with a speed proportional to the distance OM.

of constant length. The straight lines AC and BC are parallel to the coordinate axes and they intersect each other at the point C. such that AM1 = AM2 = 2b. Find the equations of the geometrical locus of the foot M of the perpendicular drawn from C on the straight line AB ( the astroid). On this ray one takes two points M1 and M2 . symmetrical with respect to the point K. symmetrical with respect to B. 7. 10. such that BM1 − BM2 = AB. On the ray we take two points M1 and M2 . equal to 2a. 13. The ends of a straight line segment of constant length AB = a slide on the axes of a rectangular coordinate system. Write down the equations of the geometrical locus described by the points M1 and M2 when the ray rotates around the origin (the strophoid). on the axes of a rectangular. with the origin at O. while for d > a — the lengthened cycloid). coordinate system. 12. A variable ray OK intersects this line at the point K. symmetrical with respect to A. π . A circle of radius a rules without sliding on a straight line. keeping its ends. one takes an invariable point O. moves on the plane. one constructs a 2 straight line parallel to the polar axis. A ray with the origin at O rotates around O and intersects the circle in a variable point A. We take a fixed point O on the circle r = 2a cos ϕ and a variable ray OM. A straight line segment. The straight line x = a cuts the Ox-axis at the point A and an arbitrary ray OB at the point B. A and B. Find the parametrical equations of the curve described by a point M. . A circular disk of radius a rules without sliding on a straight line. given through its polar coordinates. symmetrical with respect to A such that AM1 = AM2 = 2a. Find the parametrical equations of the curve described by a point invariably linked to the circumference of the circle (the cycloid). What curve do we get for a = b? 9. Through the point E a. Let A be the point of intersection between the ray and the circle. We construct the perpendicular OM from the origin of coordinates on this (M ∈ AB). invariably linked to the disk. Write down the equations of the geometrical locus of the points M1 and M2 (the Nicomede’s conchoid).104 Problems 6. Write down the equations of the geometrical locus described by M1 and M2 when the ray rotates around O (the Pascal’s limaçon). 8. lying at a distance d from the centre of the circle (for d < a the curve is called the shortened cycloid. Find the equations of the curve described by the points M1 and M2 (the cardioid). On a fixed circle of radius a. such that K M1 = K M2 = l. On the ray one takes two points M1 and M2 . We take on the ray two points M1 and M2 . Write the equation of the geometrical locus of the point M (the four petals rosette). 11.

Find the parametrical equations of a point M. 0). b) r = a 1 b 1 t + i + t − j (the hyperbola). at its exterior. invariably linked to the circumference of the mobile disk (the hypocycloid). e) r = aϕ cos ϕi + aϕ sin ϕj (the Archimede’s spiral). −2). Find the equation of the tangent and the equation of the normal to the curves: a) x2 (x + y) − a2 (x − y) = 0 at the point (0. A circular disk of radius r rules without sliding on a circle of radius R.Problems 105 14. Find the parametrical equations of a point M. at its interior. invariably linked to the circumference of the mobile disk (the epicycloid). A circular disk of radius r rules without sliding on a circle of radius R. The sides of a right angle are tangent to the astroid   x = 3 R cos t + 1 R cos 3t   4 4 4 4  y = 3 R sin t − 1 R sin 3t  4 4 4 4 . d) r = a(t − sin t)i + a(1 − cos t)j (the cycloid). 15. . 2 t 2 t c) r = a cos3 ti + a sin3 tj (the astroid). 16. b) 2x2 − x2 y2 − 3x + y + 7 = 0 at the point (1. Find the geometrical locus of the vertex of this angle when its sides occupy all the possible positions. remaining perpendicular to each other and tangent to the astroid. i+ 2 1+t t(1 + t2 ) 17. Find the equation of the tangent and the equation of the normal for the following curves: a) r = a cos ti + b sin tj (the ellipse). f) r = g) r = at at2 i+ j (Descartes’ folium). 1 + t3 1 + t3 a a j (Diocles’ cissoid). 18.

20. Show that the Bernoulli’s lemniscate (x2 + y2 )2 − 2a2 (x2 − y2 ) = 0 is the set of the symmetrical points of the centre of an equilateral hyperbola with respect to the tangents to this hyperbola. 2 22. Find the intersection angle between two coaxial parabolas if the vertex of each one coincides with the focus of the other one. a a ∈ R. Find the equations of the tangents to the curve y= 1 1 + x2 at the points where the curve intersects the hyperbola y= 23. Show that the inclination angle ϕ of the tangent to the curve y = x5 + 2x3 + x − 1 on the x-axis belongs to the interval π 4 ≤ ϕ ≤ π. Find the tangents to the astroid x3 + y3 = a3 2 2 2 which are the most distant from the origin of the coordinates. At which point the tangent to the parabola y = x2 makes an angle of 45◦ with the x-axis? 21. . a x y = a tan . 24. 25. for any value of the parameter a. 1 1+x intersect the x-axis under the same angle. provided all the functions involved are defined for the respective value. a x y = a ln . Show that the curves x y = a sin .106 Problems 19.

31. 29. 27. Show that the straight lines connecting the tangency points of each pair intercept an angle of 120◦ . α. 30. 2 32. parallel to a given direction. 34. Prove that the circles x2 + y2 + a1 x + b1 y + c1 = 0 x2 + y2 + a2 x + b2 y + c2 = 0 intersect each other under a right angle iff a1 a2 + b1 b2 = 2(c1 + c2 ).Problems 26. 107 28. Find the envelope of a family of straight lines passing through the extremities of a pair of conjugates diameters of an ellipse. 33. Find the envelope of a family of straight lines which determine on a right angle a a triangle of constant perimeter. Find the equation of the curve from whose points a given parabola is seen under a constant angle. Which curve do we get for α = π ?. It is possible to draw two pairs of tangents to the Bernoulli’s lemniscate (x2 + y2 )2 = a2 (x2 − y2 ). Find the envelope of a family of straight lines which determine on a right angle a a triangle of constant area. equal to 2a. Show that the segment from the normal to the curve   x = 2a sin t + a sin t cos2 t    y = −a cos3 t  intercepted between the coordinate axes has constant length. Show that if all the normals to a given curve are passing through the same point. then the curve is an arc of circle. . Show that all the normals to the curve   x = a(cos t + sin t)    y = a(sin t − cos t)  lie at the same distance with respect to the origin.

41. A radius vector of an ellipse projects on two conjugate diameters of the ellipse and on the projections. one builds ellipses. The segment OM is projected on the assymptotes of the hyperbola and on the projections. 44. One constructs a family of circles. Find the envelope of a family of circles with the centers on a given ellipse and passing through the same focus of the ellipse. with the centres on the circle of radius b.108 Problems 35. this time with the centres on the circle of radius a and having the radii equal to b. Find the envelope of the family of ellipses obtained this way. their radii being. 39. 36. Find the envelope of the family of curves x 2 y2 + = 1. as semiaxes. 37. Find the envelopes of the families of curves a) r = (t2 + λ) i + (t3 + λ) j. . respectively. all of them. b) r = (t2 + λ) i + t3 j. Find the radius of the osculating circle at an arbitrary point of the conical section a11 x2 + 2a12 xy + a22 y2 + 2a1 x + 2a2 y + a = 0. Find the vertices of the curve y = ln x. 38. one constructs ellipses. passing through the focus of the parabola. λ µ where λ + µ = 1. equal to a. Show that the two families of circles have the same envelope and find this envelope. Find the osculating circle of minimum radius of the equilateral hyperbola xy = 1. and a second family of circles. Find the envelope of the family of ellipses obtained this way. Find the envelope of the family of circles having as diameters the chords of a parabola. of radii a abd b. A vertex of a plane curve is a point of the curve at which the radius of the osculating circle is stationary. There are given two concentrical circles. Let O be the centre of an equilateral hyperbola M – an arbitrary point on the hyperbola. as semiaxes. 40. 42. 43.

  x = (1 + m) cos mt − am cos(1 + m)t   b)  y = (1 + m) sin mt − am sin(1 + m)t  x c) y = cosh a (the catenary). We extend the segment MP with a distance PC = MP. 3 3 50.Problems 109 45. (the epicycloid). b]. √ 2 49. Compute the radius of the osculating circle of the hyperbola a2 − b2 = 1 at one of its vertices and give a geometrical method for the construction of this osculating circle. Find the curvature radii of the following curves: a) y = sin x at the vertex. such that the function r is defined and at least three times continuously differentiable on an interval [a. while r × r 0 on this interval. d) y = ln x and find the position of the osculating circle with respect to the curve in each case. Let M be an arbitrary point of the cycloidcycloid and P – the corresponding point of the generating circle. . then the osculating circle of the curve at the point M0 intersects the curve. Show that if at a point M0 of the curve (γ) this vector is not perpendicular on the tangent to the curve at this point. Show that the circle cercul (x−a)2 +(y−a)2 − 8a = 0 and the parabola 3x+ 3y = 9 √ 2 a have a third order contact at the point a . Consider the particular case of the equilateral hyperbola x2 − y2 = a2 . a . Show that C is the center of the osculating circle of the cycloid at M. 48. 2 2 47. 2 t 2 t c) y = ax2 (the parabola). b) r = a 1 b 1 t + i + t − j (the hyperbola). Find the directing vector of the affine normal for the curves: a) r = a cos ti + b sin tj (the ellipse). The vector νa = 3r × (r × r ) − r × (r × r ) is called the directing vector of the affine normal of the curve (γ) at the considered point. We consider a curve (γ) given through the vectorial equation r = r(t). y x 46.

52. e) r = a cos3 t i + a sin3 t j (the astroid).     z = 2 cos ϕ 56. Find the evolutes of the following curves: a) r = a cos t i + b sin t j (the ellipse). Show that the curve lies on a sphere. b) r = x i + ax2 j (the parabola). Find the involute of a circle. a2 b2 c) RS = a2 . 2 t 2 t d) r = a(t − sin t)i + a(1 − cos t)j (the cycloid). Find the equation of the evolute of the curve t r R=r− |r | 0 |r |dt. 51. d) R = a + s2 . Find the involute of the parabola x2 = 2py. e) r = a cos3 ti + a sin3 tj (the astroid). 53. a   x = sin 2ϕ. 54. 55. .      y = 1 − cos 2ϕ. Find the parametrical equations of the curves having the following natural equations: a) R = as. b) s2 R 2 + = 1. c) r = 1 b 1 a t + i + t − j (the hyperbola).110 Problems d) x2 y2 = (a2 − y2 )(b + y)2 (the conchoid).

61. Find the projection of the coordinate plane xOy of the intersection curve between the paraboloid z = x2 + y2 and the plane x + y + z = 1.    2 y + z2 = 25  at the point (1. 3. q are constant. a) Find a set of parametrical equations of the curve. Find the equation of the tangent to the curve   x = t4 . while p > q.Problems 111 57. p2 q2 2. 1. . Write the equation of the tangent to the curve   x2 + y2 = 10.     z = t 2 2 parallel to the plane x + 3y + 2z = 0. c) Find the projection of the curve on the coordinate plane yOz.   4   3  y = t3 . Consider the curve obtained by intersecting the sphere x2 + y2 + z2 = a2 with the cylinder x2 + y2 = ay (the Viviani’s window). Consider the curve   x = p p2 − q2 cos t      y = q p2 − q2 (1 + sin t)     z = p2 − q2 (1 + sin t) where p. 59. 58. 4). by computing its curvature and torsion. Show that this curve lies on the elliptical paraboloid x2 y2 + = 2z. 3. Check that the osculating plane to the curve is one of the planes of circular section in the paraboloid. b) Find the curvature and the torsion at an arbitrary point of the curve. Show that this curve is a circle. 60.

y = a(sin t − cos t). β) = χ. β. 67. β. Show that if all the osculating planes to a given curve (which is not a straight line) contain the same vector. Find this vector. Show that the normal planes to the curve   x = a cos t.112 Chapter 3. .. Show that (˙ .. k ˙ ¨ . k 68..  64. Show that the tangents to the curve x = a(sin t + cos t). z = be−t intersects the coordinate plane xOy after the circle x2 + y2 = 4a2 63.    z + y tan α = 0. then d k2 + χ2 = +χ=0 d χ ds k ds k and conversely: if the previous relation is verified. then the principal normals to the curve form a constant angle with the direction of a vector. Find the equation of the normal plane at an arbitrary point of the curve  z = x2 + y2 . .  ˙ 65. 69. then the curve is a plane curve. β) = χ5 χ χ . Show that (β. Show that if the principal normals to a curve form a constant angle with the direction of the vector e. τ ) = k5 τ ¨ .     z = a cos α sin t are passing through the straight line   x = 0.    y = x. τ. The integration of the natural equations of a curve 62. Show that (τ..      y = a sin α sin t. 66.

Part II Surfaces .

.

Definition. Usually. r1 ) are called equivalent if there is a diffeomorphism λ : U → V such that r = r1 ◦ λ.1) The condition (4.1. Remark.1. . while r is subject to ru × rv 0. r = r(u. r) and (V. The set r(U) ⊂ R3 is called the support of the parameterized surface (U. since the map r is not assumed to be injective. r) may correspond to several distinct pairs (u. r). r). The supports of two equivalent parameterized surfaces always coincide. (4.1) is called the regularity condition. (U. v). A regular parameterized surface (patch) in R3 is a smooth map r : U → R3 .1 Parameterized surfaces (patches) Definition. A parameterized surface is usually denoted by (U. Definition. Two parameterized surfaces (U. v) → r(u. v). where U is a domain (an open. Remark. v)) or just r = r(u. one and the same point of the support of a parameterized surface (U. connected subset) in R2 . v). (u.CHAPTER 4 General theory of surfaces 4.

r = r0 + ua + v3 b. 4. v) are called the parametric equations of the surface. z(u. (u. v) ∈ R2 |π/2 < u < π/2. although the two surfaces are not equivalent. 2. R sin u). 3. The pair is called a local parameterization of the surface S around the point a. if r(u. This parameterized surface is called a plane. the the support of the parameterized surface is the plane passing through r0 . while the support r(U) is called the domain of the parameterization. r) with r(U) = W. then. while the map r : U → W is a homeomorphism. v)). v) ∈ U. Let U = {(u. once more. while r = r0 + ua + vb. since the map (u. 4. then the equations   x = x(u. v)      y = y(u. a × b 0.e. perpendicular to the vector a × b. We would like to emphasize. they cannot be used to describe all the points of the surface. The support of this parameterized surface is the sphere of radius R. a local parameterization (U. Let U = R2 . General theory of surfaces Examples. with a meridian removed. equally. The parameters u and v are analogues of the geographical coordinates. centred at the origin of R3 . available for surfaces. unless we are dealing with a global parameterization of a simple surface.2.116 Chapter 4. y(u. The support of this parameterized surface is identical to the support of the parameterized surface from the example 1). r) for which r(U) = S ) is called a simple surface. Parametrical representation. 0 < v < 2π} and r(u. v3 ) is not a diffeomorphism. v) → (u. If U = R2 . v) = (x(u. R cos u sin v. . v) = (R cos u cos v. A surface S which has a global parameterization (i. If S is a surface and (U. that these are only local equations. v). A subset S ⊂ R3 is called a regular surface if each point a ∈ S has an open neighbourhood W in S such that there is a parameterized surface (U.1 Representations of surfaces The same kind of representations we used for the case of curves are. r) is a local parameterization of S . v) .     z = z(u. a × b 0. v).2 Surfaces Definition. 1.

y0 . z) ∈ R3 | z = f (x. z0 ). The we get the local parameterization os S around a . 2 2 .2. therefore. Revolution surfaces Let C be a curve in the plane xOz. x(t) sin(v + v0 ). Thus. z) ∈ R3 | F(x. v) = (u. Implicit representation. according to the previous paragraph. (u. y0 . z0 ). unlike that case of the surfaces given explicitly. in general. In projection on the coordinate axes. y = y0 + uay + vby     z = z0 + uaz + vbz 2. z(t)). z0 ) ∈ S . then its graph. the parametric equations of the plane will be   x = x0 + ua x + vb x      . then. 1. If f : U → R is a smooth function. the vector grad F = ∂F ∂F ∂F . defined on the domain U = I × − π . at (x0 . it is still not sure that the function f can be defined globally. ρ(t) = (x(t). be a smooth function. Surfaces 117 Explicit representation. the implicitly given surfaces are not simple. We denote by S = {(x. where U ⊂ R2 is a domain. Let (I. r(t. y. with V ⊂ R3 an open set. Even if Fz is different from zero on the entire S . z0 ) 0. r(u. z(t)). Let F : V → R. v. Let v be the rotation angle of the plane xOz and a – the point of S obtained by rotating the point a ∈ C by an angle v0 . v) = r0 +ua+vb. y. ∂x ∂y ∂z is different from zero. Examples. generally. from the implicit function theorem. v) = (x(t) cos(v + v0 ). for instance. Indeed. y)} is a simple surface. ρ = ρ(t)) be a local parameterization of the curve C around the point a. v) ∈ R2 . S = {(x. If.4. π . at each point of S . it is not so for the entire S . which does not intersect the z-axis. there is a local parameterization of S around the point (x0 . We note that this parameterization is global for M ∩ S but. . f (u. z) = 0} the 0-level set of F. y0 . and S – the subset of R3 obtained by rotating C around the z-axis. there is an open (in the topology of R3 ) neighbourhood M of (x0 . y). with a×b 0 is a simple surface with the global parameterization r(u. The plane Π passing through r0 and having the direction given by the vectors a and b. v)). then S is a surface. z0 ) such that the set M∩S (which is an open neighbourhood of (x0 . we have Fz (x0 . we have the global parameterization r : U → R3 . y0 . y. y0 . this time in the topology of S ) is the graph of a smooth function z = f (x. Indeed.

y = 0. therefore this sphere is a reg2 is not simple since it is compact subset of R3 . z) = ( Its 0-level set. it is compact. Note that S R it cannot be homeomorphic to an open subset of R2 . y. 4. y. z) ∈ R3 |F(x. F(x. If we compute the derivatives of F with respect to the coordinates. . The sphere. z) = 0} is called the 2-dimensional torus. F(x. z) = 0} is the sphere of radius R.118 Chapter 4. 0 < b < a. it does not vanish on the sphere S R . General theory of surfaces 3. obviously. F is. y. Let F : R3 → R. therefore the torus is a surface (again. z = 0 . The torus. y. z = 0 or      2   x + y2 = a2 . The gradient of F is grad F = {2x. z) ∈ R3 | F(x. We choose now F : R3 → R. therefore it cannot be simple). It is easy to check that grad F is different from zero on T 2 . z) = x2 + y2 + z2 − R2 . 2z} 2 and. T 2 = {(x. with the centre at the origin. Fy = 2 x   x2 +y2    F = 2z  z x2 + y2 − a)2 + z2 − b2 . therefore ular surface. y. z = 0 or     x = 0. x2 + y2 = a2 . 2y. a smooth function and its 0-level set 2 S R = {(x. The torus can also be obtained by rotating the circle (x − a)2 + z2 = b2 (lying in the plane xOz) around the axis Oz. which is not compact. we get   F = 2 x 2 + y2 − a √ x  x    x2 +y2    2 + y2 − a √ y . therefore the gradient of F vanishes iff   x = y = z = 0 or     2   x + y2 = 0. y. obviously.

Before proving the theorem. 1 not in the classical sense. r) – a local parameterization and W = r(U). (of the parameters’ change) Let (U. associating to each point of W a pair of real numbers (u. the central objects is the function. then. v) ∈ U. from the relation r = r1 ◦ λ. of course. assigning a local chart to a surface simply means to apply that surface on a portion of a plane. actually. 1 The real difficulty is to show that λ and its inverse are smooth. the two approaches are completely equivalent. Although r is smooth. in other words.1. Of course. λ = r−1 ◦ r.1: The torus 4. The equivalence of local parameterizations 119 Figure 4. r1 ) be two local parameterizations of a surface S and r(U) = r(U1 ). We ought to mention here that in many books the fundamental objects used to describe a surface are not the local parameterizations. at least . This correspondence is called a curvilinear coordinate system on S or a chart on S . If a change of parameters λ does exist. but the charts. r1 (U1 ). The “charts’s approach” has the origin in chartography. (U. is not an open subset of the Euclidean space R3 . we should have. Then the map r−1 : W → U is a bijection.3. but they came from different directions. where the surfaces are thought off either as images of functions or as graphs of functions. exactly as happens in chartography. what is meant is the construction of a “map” of the surface and. in any case.4. in the advanced differential geometry. a collection of charts that cover the entire surface is called an atlas. In fact.3. Then there is a diffeomorphism λ : U → U1 such that r = r1 ◦ λ. Remark. r) and (U1 . because its domain. The description of surfaces by using local parameterizations has. Let S be a surface.3 The equivalence of local parameterizations Definition. r−1 is not1 . Theorem 4. the origin in mathematical analysis. some remarks are in order. probably. The diffeomorphism λ is called a parameters’ change.

Proof of the lemma. v)) and a = r(u0 . Due to the regularity of r. f2 (u. that. however. y) ∈ V – the point (u. v)). Let p : R3 → R2 : (x. Since the map r : U → W is a homeomorphism. z) from B ∩ W corresponds a single point (u. locally. v0 ). v0 ) in U and an open neighbourhood V of the point (x0 = f1 (u0 . from the inverse function’s theorem for the map f : (u. defined on an open set (with respect to the topology of R3 ). for each point a ∈ W there is an open set (in the topology of R3 ) B a and a smooth map G : B → U such that r−1 |W∩B = G|W∩B . y. we have r−1 (x. Indeed. in R3 there is an open neighbourhood B of the point a such that r(V) = B ∩ S = B ∩ W. to each point (x. the Jacobi matrix    f1u f1v       f   2u f2v         f3u f3v has the rank two. y = f2 (u. y. y) be the orthogonal projection on the coordinate plane xOy. as a composition of smooth maps. there is an open neighbourhood V of the point (u0 . v). We ought to emphasize. in the following lemma. Without restricting the generality. r(V) is an open neighbourhood in S of the point a = r(u0 . v) = ( f1 (u. z)) = G(x. v) = f −1 (x. v0 )) in the plane xOy such that f : V → V is a diffeomorphism. Then. z). v0 ). G is smooth. General theory of surfaces We will show. Then. therefore. r(U) = W and r−1 : W → U – the inverse map.120 Chapter 4. f3 (u. y0 = f2 (u0 . v). v0 ). Moreover. however. as the restriction of a single smooth 1 function. y. which contains the set r1 (U1 ). for the points (x. y. Let (U. and to each point (x. r−1 is the restriction of a 1 smooth function. . Lemma. that this representation of r−1 is only 1 local: in general. Thus. y. z) → (x. y. r) be a local parameterization of the surface S . We are going to show that the map G = ( f −1 ◦ p)|B : B → U is the map searched for. globally. we may assume that f1u f2u f1v f2v 0. v) −→ (x = f1 (u. z) from V. y. v) = r−1 (x. v). z) = f −1 (x. z) ∈ B ∩ W. r−1 cannot be written. y) from V. y) = f −1 (p(x. Let r(u.

The converse is not true. λ|V is smooth. from the inverse function’s theorem. so it is continuous also its restriction r|V : V → R3 . an injective map. y. r1 ) – two local parameterizations of the surface S such that r(U) = r1 (U1 ) = W. The only extra condition we have to impose on V is that the map r|V : V → r(V) to be a homeomorphism. Let (u1 . The equivalence of local parameterizations 121 Proof of the theorem. v1 ) = f (u2 . To prove the continuity of the inverse map. To prove the smoothness of λ este it is enough to prove that each point (u0 . v2 ). v0 ). the support of an arbitrary parameterized surface is not a surface.3. Let (U. Let (U. continuous. y0 ) = f (u0 . Without restricting the generality. v)). We consider the map λ = r−1 ◦ r : U → U1 . v1 ) = f2 (u2 . y) ∈ V. Then. v1 ). Since on r(V) we take the subspace topology. v2 ). so it is. for which the pair (V. y) ∈ V → (u. v2 ). r) be a regular parameterized surface. (u2 . thus. We shall prove first the injectivity of the map r|V : V → r(V). v). f3 (u. y = f2 (u.e. such that r(u1 . The map r : U → R3 is continuous. each surface is the support of a parameterized surface. v1 ) = r(u2 . v2 ) ∈ V. Then each point (u0 . r1 ) at a point a = r1 (λ(u0 . v0 ) and an open neighbourhood V of the point (x0 . f2 (u. in particular. i. we can obtain a parameterized surface whose support is a regular surface. Or. v0 ) ∈ U has an open neighbourhood V ⊂ U such that λ|V is smooth. v1 ) = (u2 . v) → (x = f1 (u. i.3. as a restriction of a smooth 1 map. as a composition of two homeomorphisms. Still. v2 ) si ¸ f2 (u1 . therefore (u1 . r|V ) is a global parameterization.4. therefore. v0 )). the map r|V : V → r(V) is. in particular. Let r(u. We have to prove only that λ and λ−1 are smooth. v0 ) such that the map F : V → V is a diffeomorphism. z) ∈ r(V) → (x.e. by restricting the domain. v2 ). replacing the parameterization r1 by the parameterization r. v). Since 1 r1 : U1 → W is a homeomorphism. Thus. r) and (U1 . if we choose an arbitrary parameterized surface. v) = f −1 (x. Let G : B → U1 be the smooth map for which r−1 |B∩W = G|B∩W and V = r−1 (B ∩ W). We apply the previous lemma to the parameterization (U1 . we have: Theorem 4. v0 ) ∈ U has an open neighbourhood V ⊂ U such that the r(V) is a simple surface in R3 . the same is true for r−1 : W → U1 and. we note that its is a composition of the following continuous maps: (x. v) = ( f1 (u. there is an open neighbourhood V ⊂ U of the point (u0 . . The smoothness of λ−1 is proved in the same way. as seen in the proof of the lemma. Locally. Then. we may assume that the Jacobian of the map f : (u. v)) is different from zero at (u0 . Proof. v). f (u1 . 1 Then λ|V = r−1 ◦ r|V = (G ◦ r)|V and. v1 ) = f1 (u2 . also. f is diffeomorphism.2. f1 (u1 . λ is 1 a homeomorphism.

The converse affirmation can be proved even simpler. thus.4. from the ˜ formula (4. as a composition of two continuous maps. ˜ To verify the equivalence of the regularity conditions for ρ and ρ. through the formula (4.1) ˜ Conversely. (4. Therefore. Differentiating this relation. it is smooth. a smooth parameterized curve on r(U).1). as a composition G ◦ ρ of smooth maps and.4. Theorem 4. as r and ρ are smooth. any smooth parameterized curve ρ on U defines. ρ) on U such that ˜ ρ(t) ≡ ρ(ρ(t)).4. The regularity of ρ at t is equivalent to the ˜ regularity of ρ at t. because we have ˜ ρ=r◦ρ ˜ and. the map ρ. we consider the ˜ components of the path ρ: ˜ ρ = (u(t).4.1.1). smooth. We shall check now ˜ that ρ is. while ρ(I) ⊂ r(U). ρ = ρ(t)) lies on a surface S if its support ρ(I) is included in S . in the neighbourhood of the point t.4 Curves on a surface We shall say that a smooth parameterized curve (I. we can obtain ρ. Since the map r : U → r(U) is a homeomorphism. ρ is continuous. r) of the surface S . ρ = ρ(t)) a smooth parameterized curve whose support is included in r(U). we get ρ (t) = ru · u (t) + rv · v (t). Proof. so is ρ. According to the lemma from the previous paragraph there is an open neighbourhood B of the point ρ(t) in R3 and a ˜ smooth map G : B → U such that r−1 |B∩r(U) = G|B∩r(U) . Let t ∈ I. ˜ Clearly. General theory of surfaces 4. v(t)).1) becomes ρ(t) = r(u(t). Then the equality (4. . r) be a parameterization of the surface S and (I. by putting ˜ ρ = r−1 ◦ ρ. v(t)). can be represented. then ρ(t) ∈ r(U). Then there is a single ˜ smooth parameterized curve (I.122 Chapter 4.4. It is easy to describe the parameterized curves whose support is contained into the domain of a parameterization (U. Let (U. actually.

in the local parameterization (U.3) It is easy to see that the supports of these paths lie. r). in r(U) ⊂ S the paths defined by the local equations  u = u0 + t   (4. indeed.5. Definition 4. is supposed to be regular). but it plays an important role in the sequel.5 The tangent vector space.4. as always. naturally isomorphic to R3 . a tangent vector to a surface is just a tangent vector to a curve on the surface. Let (U. obviously. v0 ) ∈ r(U) pass exactly two such curves. Example.e. r = r(u. by R3 the space of bound vectors with the origin at a. one of each kind. 4.5. the tangent plane and the normal to a surface Let us denote. iff ρ (t) = 0.2)  v = v  0 and  u = u0    v = v + t  0 (4. This a is. v0 ) ∈ U.1. v)) a local parameterization S and (u0 . while the equations  u = u(t)    v = v(t)  are called the local equations of ρ(t) in the considered parameterization. The parameterized curves are called coordinate lines or coordinate curves on the surface S . Thus. The parameterized curve ρ(t) on the domain U is called the local representation of the parameterized curve ρ(t) in the local parameterization (U. from the previous relation it follows that ρ (t) = 0 iff u (t) = 0 and ˜ v (t) = 0. The tangent vector space. a 3-dimensional vector space.4.4. We consider. ρ(t)) on S and a t0 ∈ I such that ρ(t0 ) = a and ρ (t0 ) = h. r). i. the tangent plane and the normal to a surface 123 Since the vectors ru and rv are not colinear (because the surface. on S . The following lemma is trivial. ˜ Definition. A vector h ∈ R3 is called a tangent vector to the surface S at the point a a if there is a parameterized curve (I. Through each r(u0 . . for any a ∈ R3 . We shall denote by T a S the set of all the tangent vectors to the surface S at a ∈ S .

Let (U. 2 In this context. r) is a local parameterization of S . then. then the plane a passing through a and having T a S as directing plane (i. called the natural basis or the coordinate basis of the tangent space. If (U. passing through a for t0 . v0 ) + βrv (u0 . General theory of surfaces Lemma. from the formula (4. while its local equations in this parameterization of the surface are u = u(t). v(t)) with respect to t. v(t)). with a = r(u0 . Then. v = v(t). Based on this isomorphism. v(t)) + v (t)rv (u(t. v0 ). In this case. Let ρ = ρ(t) be a parameterized curve on S . if necessary. v0 ) and rv (u0 .1) ρ (t) = u (t)ru (u(t). therefore h ∈ T a S . v0 ) is tangent to the parameterization curve given by the local equations  u = u0 + αt    v = v + βt . passing through a). we may consider. v = v(t). naturally means that there is an isomorphism which is independent of the choice of bases in the two vector spaces. a subspace of R3 rather than of R3 . Then we have the relation (4. then the vectors ru (u0 . is called the tangent plane to S at the point a and it is denoted by Πa S . r) be a local parameterization of S .5. the plane parallel to T a S .5. R3 a The vector space T a S is called the tangent space to S at a. restricting. when is convenient. given by the local equations u = u(t).1. Proof. in the sense that it passes through the origin of R3 . we may assume that ρ(I) ⊂ r(U). The set T a S is a 2-dimensional vector subspace of R3 . if the parameterized curve (I. ρ(t)) is on the surface and ρ(t0 ) = a.1). that T a S is. is naturally2 isomorphic to R3 .e. We just differentiate the relation ρ(t) = r(u(t). .  0 which is a curve on S . Proof. v0 ) + v (t0 )rv (u0 .124 Chapter 4. Conversely. T a S a is a vectorial plane. r) of S . v0 ). in fact. Theorem 4. the interval I. v0 ) make up a basis of this subspace. v0 ). it follows that ρ (t0 ) = u (t0 )ru (u0 . As we mentioned before. with respect to a local parameterization (U. any vector of the form h = αru (u0 .5. while a = r(u0 .

(4. ∆t→0 h ∆t→0 αru (u0 . Sufficiency. Thus. If Π is the tangent plane.e. v0 ) + βrv (u0 . v0 ). r = r(u. α2 + β2 0.5.5. h = ∆r . The straight line passing through a point of a surface. is called the normal to the surface at the considered point. v0 ) + ε αru (u0 .5. β ∈ R fixed. (4. therefore the relation (4. Theorem 4. now. v)) is a local parameterization of the surface S and a = r(u0 . Let us assume. ∆v = β∆t. Then d = ∆r · n. z(u. clearly. Using this formula. Definition 4. v0 )) · n d = lim =± u . Π is the tangent plane.5. β = 1. with α. we get ru · n = 0. the plane and the surface have a first order contact at a. and h – the distance between the points a and ∆a. The tangent vector space. yu xu zv yv xv Let. y0 .3) is fulfilled. the equation of the tangent plane of S at a should be X − x0 Y − y0 Z − z0 zu = 0. with lim ε = 0. Proof. passing through a = r(u0 . v). ∆r = r(u0 + ∆u. v0 ) + βrv (u0 .e. d – the distance from the point ∆a = r(u0 + ∆u.5. v0 )) + ∆t · ε.3) is fulfilled. are orthogonal to n. v0 + ∆v) − r(u0 . In the same way. v0 ). The plane Π is the tangent plane to the surface S at the point a iff for any modification of the arguments of the form ∆u = α∆t. z0 ) ∈ S . β ∈ R. we have d lim = 0.3) Necessity. the Taylor’s formula gives: r(u0 + ∆u.5. v0 ) lim therefore ∆t→0 ∆t→0 lim = 0 ⇐⇒ (αru (u0 . Let n be the versor of the normal to the plane Π. v0 + ∆v) = r(u0 . If we modify the arguments by ∆u = α∆t. n is orthogonal to the tangent plane.2. one obtains rv × n = 0. for α = 0. Replacing ∆r by its expression. (U.2. perpendicular to the tangent plane to the surface at that point. then. v)) be a parameterization of S and (u0 . v0 ) + βrv (u0 . v0 + ∆v) to the plane Π. with α. v0 )) · n = 0. then the vectors ru and rv . β = 0. v0 ) + βrv (u0 . v). v0 ) + βrv (u0 . ∆v = β∆t. y(u.5. Let Π be a plane in R3 . we shall give another characterization of the tangent plane. v) = (x(u. . that (4. v0 ) + ε) · n (αr (u0 . the tangent plane and the normal to a surface 125 If r(u. we get: ∆t · (αru (u0 .2) ∆t→0 h i. v0 ) + βrv (u0 .4. v0 ) = (x0 . i. v0 ) + ∆t · (αru (u0 . now. v0 ) ∈ U. as directing vectors of Π. Choosing α = 1.

For any point a of the sphere S R the tangent space T a S R is orthogonal to the radius vector of the point a. rv ) ≡ T (x0 . v0 ). y0 . v0 ) zv (u0 . v0 ) xv (u0 . z0 ) + (Y − y0 )Fy (x0 . v) = (x(u. v0 ). by differentiation. y0 . v0 ) (4. z) = 0 the vector grad F0 = {F x (x0 . z0 ) has the form (X − x0 )F x (x0 . whence. y0 . v0 ) . v). y0 . v). y. z0 ) ∈ S . z) = 0 at the point (x0 . v0 ) xu (u0 . y(u. z0 2 2 Consequence 2. z(u. grad F ⊥ L(ru .e. z0 ) = 0. Theorem 4. v)) be a local parameterization of the surface around the point (x0 . v)) = 0. y(u. z0 ) Fz (x0 . Fz (x0 . v0 ) zv (u0 .4) To construct the tangent plane and the normal to a surface given in an implicit representation. v0 ) xu (u0 .126 Chapter 4. v0 ) yv (u0 . y0 . v). Let r(u.z0 ) S . y0 . z0 )} is perpendicular to the tangent plane to the surface at this point. the following result is very useful. y.5. which means that the equations of the normal at the point a will be given by: Y − y0 Z − z0 X − x0 = = yu (u0 . v0 ) xv (u0 . z0 ). y0 . v0 ) yu (u0 . z0 ) = r(u0 . z0 ). . then the a directing vector of the normal will be ru (u0 . if (U. z0 ) of the surface given through the equation F(x. y0 . i. v).5.y0 . Proof. The equation of the tangent plane to the surface given by the implicit equation F(x. Consequence 1. Then we have the identity F(x(u. z0 ) + (Z − z0 )Fz (x0 . z(u. y0 . r) is a parameterization of the surface around a point a = r(u0 .3. v0 ) = (x0 . z0 ) Fy (x0 . General theory of surfaces Thus. y0 . v0 ) zu (u0 . v0 ) zu (u0 . v0 ) × rv (u0 . F x (x0 . we get  0 = Fu = F x · xu + Fy · yu + Fz · zu ≡ grad F · ru    0 = F = F · x + F · y + F · z ≡ grad F · r  v x v y v z v v . y0 . Fy (x0 . while the equations of the normal to the surface at the same point are X − x0 Y − y0 Z − z0 = = . y0 . At the point (x0 . y0 . yv (u0 .

6.1. 4. that the map n : S → R3 . An orientation of a surface S is a choice of an orientation in each tangent space T a S . v) = ru × rv . The orientation of surfaces 2 Proof. y. y.6. n(a). the map 2 n : S R → R3 . 2 Examples. i. z) ≡ x2 + y2 + z2 − R2 = 0. while those on which an orientation has been already chosen – oriented. a) We can define an orientation on the sphere S R using the versor of the exterior normal.6 The orientation of surfaces Definition 4. y. Therefore. with the global parameterization (U. This surface can be oriented by using the vector field n(u. then n(a) = R a. z) = 0. hence it is perpendicular to the tangent space. It is not difficult to see that. Then the surface can be oriented through the gradient vector field: n(x. It is assumed. ru × rv c) Let S be a surface given by the implicit equation F(x. therefore the radius vector is parallel to the gradient of the function F. z} = 2a. grad F . if a is the radius point of the point 1 2 a ∈ S R . The sphere S R can be described by the equation 127 F(x.4. − → R b) Let S be a simple surface. in this context. a choice of the unit normal vector of T a S . The surfaces on which is is possible to define an orientation are called orientable. whence it follows grad F = 2{x.e. y. r). defining the orientation of the sphere can be represented as a composition of continuous maps: 1 1 R 2 i S R → R3 − R3 : a −→ a −→ a. z) = grad F . a → n(a) is continuous.

The non-orientability of the Möbius’s band. 1}. sin t 1 + s cos . We consider. as it might seem. then we must have N(a) = λ n(a).) Notice that S is not simple. v sin ). ϕ2 (s. after the “trip”. taking values into the finite set {−1. t) ∈ T ∗ . (It has a single side: it is possible to move continuously the origin of the unit normal along a close path on S such that. called the opposite orientation of the orientation given by n. 2 2 v v v ρ(u. 2 2 t t t r(s. if S is connected. We can write Φ explicitly as Φ(s. s sin 2 2 2 and 1 1 < u < . t) ≡ (u. v) | − It is not difficult to see that the domain of the diffeomorphism (parameters’ change) Φ = ρ−1 ◦ r is the set T ∗ . t) = cos t 1 + s cos .128 Chapter 4. sin u + v cos sin u. then each orientation of S should coincide to one of the two orientations just mentioned. −π < v < π . t). General theory of surfaces If the orientation of a surface S is given by the vector function (vector field) n(a). where λ is a continuous function on S . If the orientable surface S is connected. Remark. therefore. t)). the unit normal will change into its opposite. Not any surface is orientable. where u = ϕ1 (s. It will be shown bellow that S is not orientable. because r is not a parameterization. Thus. sin v 1 + u cos . ∀(s. v) = cos v 1 + u cos . t) | − We consider two local parameteriza- 1 1 < s < . if the surface is not connected. equal to T . see the next figure). u sin 2 2 2 ρ : V = (u. with the segment t = π removed. v) = (cos u + v cos u u u cos u. t) = s. corresponding to different combinations of the two possible orientations on each connected component of the surface. Of course. then the vector field − n(a) also defines an orientation on S . there are more orientations. Indeed. since it is not a homeomorphism on the image. the support of the parameterized surface r(u. 0 < t < 2π . v) = (ϕ1 (s. tions to describe the Möbius’s band: r : T = (s. for instance. 2 2 2 with u. λ has to be a constant function. if N(a) is an orientation of the surface S . v ∈ R (the Möbius’s band. a connected orientable surface has only two distinct orientations.

t) =  π − t  129 if (s. as one can see from the computation we made before.4. admitting that the Möbius’s band is orientable. We shall suppose . that on the support of r all this parameterizations are positively equivalent to r. the determinant of the Jacobi matrix of the parameters’ change is not always positive. if it is orientable. any local parameterization of S should be positively equivalent either to r or to ρ. which are pairwise positively equivalent and their supports cover S . t) ≡      ∂u ∂v  =        1 0  if π < t < 2π ∂s ∂t     0 −1  The supports of r and ρ are. Let us suppose. .2: The Möbius’s band ization in the family such that its support intersects the segment t = 0. . . r2 . clearly. that S is orientable. On the other hand. without restricting the generality. π < t < 2π The Jacobi matrix of the map Φ is. in other words. There should be a local parameter- Figure 4. However. orientable. the two local parameterizations do not define the same orientation on it.6. 0 < t < π . it can have only two distinct orientations. We may assume. This means that there is a family of local parameterizations r1 . if (s.   1 0        ∂u ∂u   if 0 < t < π     ∂s ∂t  0 1          J(Φ)(s. t) ∈ T ∗ . as one can readily see. The orientation of surfaces and  t   v = ϕ2 (s. . as simple surfaces. t) ∈ T ∗ . now. as our surface is connected. since.

from a strip of paper. if necessary. 2. r) of S the map f ◦ r : U → Rk is smooth. whence the contradiction which shows that the Möbius’s band is not orientable. A local parameterization (U. rv . Let S be an oriented surface with the orientation n(a). 4. the left hand side has opposed signs on the two sides of the segment t = 0. On the other hand. 1. Let S be a surface in R3 . The map fr ≡ f ◦ r is called the expression of f în in the curvilinear coordinates (u. in the right hand side. that J(ρ−1 ◦ r) = J(ρ−1 ◦ r1 )J(r−1 ◦ r). we have. General theory of surfaces that this parameterization is r1 . from the hypothesis. we may shrink the domain of r1 ). A map f : S → Rk is called differentiable or smooth if for any parameterization (U.7 Differentiable maps on a surface Definition 4. r) of S is said to be compatible with the orientation n(a) if for any point a = r(u. locally. Thus. n(a)} is right-handed. The first determinant. since the two parameterizations are assumed to be positively equivalent. as we saw previously. v) or the local representation of f with respect to the parameterization (U. is always positive or always negative.130 Chapter 4. Another example of a non-orientable surface is the so-called Klein’s bottle (see figure ??) Definition. 1 Now. from the chain rule. since. Remarks. Any differentiable map f : S → Rk is continuous. We may assume that the support of r1 is included into the support of ρ (otherwise. if the frame {ru . it can be written as a composition of continuous maps: f = f ◦ (r ◦ r−1 ) = ( f ◦ r) ◦ r−1 = fr ◦ r−1 . r). On the other hand. obviously. .7.6 we indicate how one can construct a Möbius band. v) we have r × rv n(a) = u ru × rv or. In the figure 4. 1 whence det J(ρ−1 ◦ r) = det J(ρ−1 ◦ r1 ) det J(r−1 ◦ r). the last determinant is always positive. On can define similarly the differentiability of maps defined on any open subset of a surface S . which is the same. It follows then that the Jacobi determinant of the map ρ−1 ◦ r1 should be either always positive or always negative on the domain of r1 .1. the right hand side has constant sign.

we can also apply the previous example).4. then. for any surface S the map f = F|S : S → Rk is a smooth map. A0 ∈ Rk is smooth. since for any local parameterization (U. In particular. it is quite difficult to check whether a map defined on a surface is smooth.7. b) If F : R3 → Rk is a smooth map. . a constant map. Indeed. c) The inclusion i : S → R3 : a → a is smooth. the orthogonal projections of a surface S on the coordinate axes and planes are. all of them. the local representation of i is ir = i◦r = r. 1R3 and. (In fact. therefore. equally. Apparently. smooth maps. i is just the restriction of the identity map. Differentiable maps on a surface 131 Figure 4. hence it is differentiable. because its local representation with respect to any local parameterization of S is. where F and r are smooth maps. in the usual sense. r) of S .3: Construct your own Möbius band! Examples. for any local parameterization (U. a) Any constant map f : S → Rk : a → A0 . r) of S the local representation of f is fr = F ◦ r.

132 Chapter 4.4: Klein’s bottle because we have to check the smoothness of its local representations with respect to all the local parameterizations of the surface. if the surface is simple.1. A map f : S → Rk is smooth iff for any point a ∈ S there is a local parameterization (U. it is enough to take only some of the local parameterizations. which are. Fortunately. such that their domains cover the surface. let’s suppose that that a ∈ S is an arbitrary point of the surface and (U. it is enough to check for the global parameterization. if F is smooth. r) of the surface S with a ∈ r(U). then it local representation fr is smooth for any local parameterization (U. In particular.7. since. r) . of course. General theory of surfaces Figure 4. Theorem 4. Proof. The necessity is obvious. Conversely. infinitely many. such that the local representation fr = f ◦ r : U → Rk is smooth. as the following theorem shows. r) of S .

which. Let S 1 .3. r) a local parameterization of S . also. because the local representation of r−1 in the parameterization (U. (U1 . Differentiable maps on a surface 133 is a local parameterization of S around a such that the local representation fr = f ◦ r is smooth. Then in r−1 (W) ⊂ U1 . actually. assigns to each point of the surface (lying in r(U). another parameterization.r2 ≡ r−1 ◦ F ◦ r1 : U1 → U2 2 . r The next natural step will be to define the notion of a smooth map between two surface rather then from a surface to an Euclidean space. We choose. r) is r−1 ≡ r−1 ◦ r = 1U . A map F : S 1 → S 2 is called smooth if the map F1 = i ◦ F : S 1 → R3 is smooth. a pair of coordinates. it is enough to show that then the local representation of f in any other parameterization of S around a is. in fact. We also showed that. equally smooth. Let S be a surface and (U. More specifically. Remarks. Since both fr (from the hypothesis) and r−1 ◦ r1 (from the theorem 4. S 2 be two surfaces In R3 . Let S 1 . Then F is smooth iff for any a ∈ S 1 . as we can see easily. Obviously. so the notion itself doesn’t make sense for it. Indeed. The idea is the following. fr1 can be represented as fr1 = f ◦ r1 = f ◦ (r ◦ r−1 ) ◦ r1 = ( f ◦ r) ◦ (r−1 ◦ r1 ) = fr ◦ (r−1 ◦ r1 ). r−1 is smooth.2. as a map defined on an open set of S . also. locally. Let S 1 . thus. r−1 is the restriction of a smooth map defined on an open set of R3 . Definition 4. Example. r2 ) of S 2 around F(a). it follows that fr1 is. As we explained earlier. 3. r1 ).7.1) are smooth. Then any map F : S 1 → S 2 can be regarded as a map F : S 1 → R3 . of course). S 2 ⊂ R3 be two surfaces. 2. The notion we just defined is. while the map G|S 1 : S 1 → S 2 is smooth.4. 1. where i : S 2 → R3 is the inclusion. one can associate to F the map i ◦ F : S → R3 .7. S 2 ⊂ R3 two surfaces and F : S 1 → S 2 a map. a surface. It easy to see that any smooth between surfaces is continuous. Then S 2 = G(S 1 ) is. the natural frame of discussing this important map. however. around a and let W = r(U) ◦ r1 (U1 ). r1 ) of S 1 around a and any local parameterization (U2 . the map Fr1 . The reason is that its domain is not an open set of an Euclidean space. the map r−1 : r(U) → R2 is nor smooth in the classical sense. any local parameterization (U1 . smooth. Let S 1 ⊂ R3 a surface and G : R3 → R3 a diffeomorphism.

A map F : S 1 → S 2 is called a diffeomorphism if F is bijective and both F and F −1 are smooth maps. y. j ≤ 3. (x0 .     k=1 k=1 k=1 . the vector daG(h) has the components    3  3 3       α2 j h j .134 Chapter 4. g3 (x(t). z(t))) and we obtain:   3 3 3     ∂g2 ∂g3 ∂g1 −−→ −−   (x0 . y(t). y(t). y(t). α3 j h j  . General theory of surfaces is smooth (in the usual sense). we differentiate the relation (G ◦ ρ)(t) = (g1 (x(t). h3 }. z). y0 .7. For any vector h ∈ R3 . z) = (g1 (x. y.y0 . z(t)). To this end. 4. We shall emphasize.      j=1  j=1 j=1 Let us suppose now that the vector h is tangent to the parameterized curve ρ(t) = (x(t). y0 .e. for any point a = (x.z0 ) where αi j = ∂i gi (x0 . We are going to show that the vector daG(h) is the tangent vector to the parameterized curve (G ◦ ρ)(t) at t = t0 . while G(x. z(t)). 3 daG : R3 → RG(a) a is a linear map. also for the notion of differential. Definition 4. (x0 . with B ⊂ R3 – an open set. a property of the differential of map between Euclidean spaces that will be used to construct the generalization. The same should hold true. y. Fr1 . z)). g2 (x(t). z0 )hk . D(x. h = ρ (t0 ). g2 (x. g3 ) = (αi j ). α2k hk . y0 . of course. the differential of G at a. y(t). y. h2 .3. y0 . z) ∈ B. z0 )hk  = (G ◦ ρ) (t0 ) =    k k k   ∂x ∂x ∂x k=1 k=1 k=1  3  3 3       =  α1k hk . a of components {h1 . Let G : B → R3 be a smooth map. g2 . g2 (x.8 The differential of a smooth map between surfaces The notion of a smooth map between surfaces is a natural generalization of the notion of smooth map between open sets in Euclidean spaces. z0 )hk . z). Then. y. y. 1 ≤ i. i. r2 ). therefore. z) (x0 .r2 is called the local representation of F with respect to the local parameterizations (U1 . r1 ) and (U2 . α3k hk  .  α1 j h j . z(t)) at point t = t0 . z0 ). whose matrix is the Jacobi matrix D(g1 .

this not actually the case. 135 Thus. Moreover. Then to any smooth path (I. Let’s suppose that the local equation of the curves in the parameterization (U. (I1 . z (t0 )} = ρ (t0 ). we may assume. Then the vectors ρ (t0 ) and ρ1 (s0 ) have. If ρ(t) passes through a for t = t0 . v (t0 )} ρ1 (s0 ) = {u1 (s0 ). v1 (s0 )}. Now. with ρ(t0 ) = a. there might be a little difficulty here. Well. in the natural basis {ru . respectively. ρ) on S 1 corresponds a smooth path (I. (F ◦ ρ)(t) = Fr (u(t). As we are interested only on the local phenomena that happen around a. r) are  u = u(t)   (ρ)  v = v(t) . We choose an arbitrary local parameterization (U. r) on S 1 . that ρ(I) ⊂ r(U) and ρ1 (I1 ) ⊂ r(U). Indeed. Let now F : S 1 → S 2 be a smooth map between the surfaces S 1 and S 2 and a ∈ S 1 . which have the same tangent vector at a contact point. then the path F ◦ ρ(t) will pass through F(a) for t = t0 . h2 . The differential of a smooth map between surfaces where {h1 . Namely.  and. as we shall see in a moment. the differential daG assigns to a tangent vector to the path ρ(t) at t = t0 the tangent vector to the path G(ρ(t)) at t = t0 . v1 (s)).4. . we could have on S 1 two different curves. the tangent vector (F ◦ ρ) (t0 ) to the parameterized curve F ◦ ρ at t = t0 is called the differential of the smooth map F : S 1 → S 2 at the point a and it is denoted by da F. F ◦ ρ) on S 2 .1. ρ1 = ρ1 (s)) – two parameterized curves on S 1 such that ρ(t0 ) = ρ1 (s0 ) = a and ρ (t0 ) = ρ 1 (s0 ). distinct. let a ∈ S 1 and (I.8.  u = u1 (s)   (ρ1 )  v = v (s)  1 . y (t0 ). rv } the expressions ρ (t0 ) = {u (t0 ). v(t)) (F ◦ ρ1 )(s) = Fr (u1 (s). assigning to each tangent vector ρ (t0 ) −−→ −− to a parameterized curve ρ(t) on S 1 . As the images of the two parameterized curves through F are. around a. h3 } = {x (t0 ). without restricting the generality. generally. Definition 4. in the chosen parameterization.8. it may happen that the images do not have the same tangent vector at the contact point. ρ = ρ(t)). The map T a S 1 → T F(a) S 2 .

y. D : R3 → R3 a diffeomorphism such that D(S 1 ) = S 2 . v0 )u1 (s0 ) + (Fr )v (u0 . respectively. v0 )v1 (s0 ). R 4. v0 )u (t0 ) + (Fr )v (u0 . then we can build a map Γ : S → S 2 .e. a diffeomorphism such that r 2 ) = S 2 . (4. Then. In particular. it follows immediately that if a vector h ∈ T a S 1 has in the natural basis {ru . obviously. it follows that −−→ −− −− − − −→ (F ◦ ρ) (t0 ) = (F ◦ ρ1 ) (s0 ). If the orientation of S is given by the unit normal n(a). the components {h1 . centred at the origin and D : R3 → r R3 : (x. z) – a homothety. v0 )h2 . that Γ is smooth: . z) → R (x. h2 }. y.9 The spherical map and the shape operator of an oriented surface Let S ⊂ R3 be an oriented surface and S 2 – the unit sphere centred at the origin. Let S 1 . Example. S 2 be two surfaces in R3 .1) whence the linearity.8.136 with Fr = F ◦ r. v(t)))(u0 . a ∈ S . then − → − → da F(h) = Fr u (u0 . v0 ) = (Fr )u (u0 . v0 ) = (Fr )u (u0 . F = DS 1 : S 1 → S 2 . i. ds Now. Thus.2) 2 where da D : R3 → R3 is the differential of the map D at a. v0 )v (t0 ) dt −−→ −− −−→ −− d −− − − −→ (F ◦ ρ1 )(s0 ) = (Fr (u(s). We are going to show. we have da F(h) = R h. first of all. This map place a central role in the theory of surfaces. rv } of the linear space T a S 1 . which is. From the computations made above.8. v(s)))(u0 . (4. for the map F = D| 2 2 D(S R r S 2 : S R → S r . The map dF : T a S 1 → T F(a) S 2 is linear. General theory of surfaces −−→ −− −−→ −− d −−→ −− (F ◦ ρ)(t0 ) = (Fr (u(t). Proof. v0 )h1 + Fr v (u0 . The map Γ is called the spherical map of the surface S . Theorem. the definition of the differential of F makes sense. a ∈ S ⊂ R3 . Then we have da F = da D|Ta S 1 . since ρ (t0 ) = ρ1 (s0 ). Chapter 4. let S R and a D(a) 2 S r be the spheres of radii R and r. associating to each a ∈ S the point on S 2 has as radius vector Γ(a) = n(a).

− (u. It is easy to see that the pair (U − . v) = r (−u. Therefore. therefore Γ itself is smooth. z). Examples. r1 ) which is not compatible with the orientation. Remark. the shape operator carries an extra minus sign. such a parameterization always exists. v)) = Γr (u. n(a) is orthogonal to T a S . or fundamental operator.e. in fact. compatible with the orientation. ru × rv Thus. In some books. On the other hand. a Γ(a) we may think of the differential da Γ : T a S → T Γ(a) S 2 as being. − then we replace the domain U1 by U1 . Proof. v) = n(u. Clearly. Let a ∈ S . The spherical map Γ : S → S 2 of an oriented surface S into the unit sphere S 2 is a smooth map between surfaces.9. or. nor the name of the shape operator. Definition 4. r− ) is a and the map r1 (u. v) by r1 1 1 1 parameterization of the surface. (i) For a plane Π the spherical map is constant. The linear operator da Γ : T a S → T a S is called the shape operator of the oriented surface S at the point a and it is denoted by A or Aa . i.1. with x As we saw. Thus. The spherical map and the shape operator of an oriented surface 137 Theorem 4. then the subspaces. v). it is true. Historically. r) of the surface S around a. the symmetric of U1 with respect to the Ov-axis. a linear operator TaS → TaS .9. v) = Γ(r(u. since S is orientable. y. that Weingarten was the first to write down the formulae for the differentiation of the spherical map (in other words. T Γ(a) S 2 . v) = ru × rv . Sometimes it is called the Weingarten mapping. y. we have r1u × r1v r1u × r1v = − n(u. v). if we choose a parameterization (U1 . z) = 2 + y2 + z2 = R2 . 1 R (x.9. compatible with the orientation. Indeed.1. the local representation of Γ is smooth. if we identify R3 and R3 to R3 . T a S and T Γ(a) S 2 coincide. is orthogonal to the radius vector n(a) of the point Γ(a). There is no general agreement regarding the definition. Now we have (Γ ◦ r)(u. he was the one to find the partial derivatives of the unit normal vectors in terms of the derivatives of the . the tangent space to the sphere. also the principal.4. 2 2 (ii) For the sphere S R the map Γ : S R → S has the expression Γ(x. indeed. We choose a local parameterization (U.

h2 ) are the components of h with respect to the natural basis {ru . The case h = p is trivial.2) Proof. v) = ru × rv . It is enough to make the proof for the vectors. (ii) For a sphere. To prove this. Then the local representation of the spherical map Γ of S will be given by n(u.4) . we start from the obvious equalities: rv · n = 0 si ¸ ru · n = 0 We differentiate (*) with respect to u and (**) with respect to v and we get  r uv · n + r · n = 0   v u  r  uv · n + ru · nv = 0 whence. for the shape operator we will have: A(h) = nu h1 + nv h2 . in 1912. was introduced in differential geometry by the Italian mathematician Cesare Burali-Forti ([6]). ∀h. r) be a local parameterization of S . i. where h ∈ T r(u. i.e. the shape operator vanishes.v) S . Theorem. (4. therefore it is enough to check that A(ru ) · rv = ru · A(rv ). with the orientation. General theory of surfaces radius vector).9.1) therefore. under the name omografia fondamentale. we get the conclusion. fundamental homography.138 Chapter 4.9.9. subtracting the two equalities. as a linear map. the shape operator is a homothety. the shape operator. In particular. ru and rv . Let now (U. ru × rv (4. we have A(h) · p = h · A(p). i.e. (i) For a plane. p ∈ T a S . rv }. Example. A(rv ) = nv .e.9. (**) (*) (4. (h1 . we have A(ru ) = nu . Nevertheless.3) (4. nu · rv = ru · nv . The shape operator A is self-adjoint.

with the global parameterization given by the identical map of U).4.v) S is isometrical (u. the scalar (u. especially the older ones. If λ1 λ2 . the map dr(u. what really happens. we will have ϕ1 (p. however. by construction. q) = p · q. for any (u.v) . as a matter of fact. Thus. {0. If λ1 = λ2 . The first fundamental form of a surface 139 Consequence. the function ϕ1 . hence it is enough to choose a unit vector in each eigenspace. v) can be identified with the space T r(u. then A is just a homothety and any orthonormal basis will do (since. but it should be kept in mind. associating to the vectors {1. any tangent space is an eigenvector of A). Since A is self-adjoint. but rather as the quadratic form associated to this restriction. If (U. then for any (u. this identification is just the linear isomorphism dr(u. associating to each a ∈ S the restriction of the scalar product of R3 to T a S . for any a ∈ S and any p.v) rv (u. v) ∈ U the tangent space of the domain U at the point (u. making up a basis of R2 . We a shall say usually. 0}. (4.1. r) is a local parameterization of S . λ2 are real.10.v) ϕ1 (ξ.v) product of two vectors is defined by the rule R2 (u.10 The first fundamental form of a surface Let S be a surface in R3 . that the first fundamental form is the restriction itself.v) (η). as being a simple surface. The first fundamental form of a surface S is. in this case. the vectors ru (u. Thus. It easy to see that. a ∈ S . v) and (u.v) mental form ϕ1 of S to the domain S (which can be seen. the two eigenvalues λ1 . v) ∈ U. In many textbooks.v) U ≡ R2 at the domain U.v) (ξ) · dr(u.v) (η)) = dr(u.10. made up of eigenvectors of the shape operator A.v) (ξ). it also induces a scalar product in each tangent space T a S . in fact. 1}. It is easy to see that.v) : R2 → T r(u. 4. Using this identification. by definition. hence.v) : R2 → T r(u.v) S . Then the scalar product in R3 induces a scalar product in each R3 and. v). Proof. η) = ϕ1 (dr(u. one can transport the first funda(u. the first fundamental form is not defined as the restriction of the scalar product to T a S .1) Remark. in the tangent space T r(u. dr(u. informally. a Definition 4. In each tangent space T a S there is an orthonormal basis.v) S .10. q ∈ T a S . then the eigenvectors corresponding to λ1 and λ2 are orthogonal.

If Π is the coordinate plane xOy. π . (u. R cos u sin v. We introduce the notations  E(u. v = v(t). π × − π . 2 2 2 2 hence the matrix of the first fundamental form is given by G = R2 1 0 0 cos2 u 4. v)   u u    F(u. rv (u. v) · rv (u. R sin u). 0. in the natural basis. 01 2 2. we choose the local parameterization (U.v) S with respect to the basis {ru (u. b = . while the matrix G = F G is the matrix of the scalar product ϕ1 on the tangent space T r(u. the tangent vector of ρ. {0. a = ı. v). v) · r (u. v) · r (u. r) – a local parameterization of S and (I. given by the local equations u = u(t). v) v v . G are smooth on U. General theory of surfaces with respect to the scalar products ϕ1 and ϕ1 respectively. ρ (t) has the components {u (t). and U = − π . then we may set r0 = 0. v)     G(u.10. 0}. For the plane Π given by the global parameterization r = r0 + ua + vb. G = R2 cos2 u. E F Then the functions E. F. r). v) = ru (u. (U. v) = (R cos u cos v. 1}}. with r(u. We get immediately E = R2 . ρ) – a parameterized curve with ρ(I) ⊆ r(U). Then. but it is also the matrix of the scalar product ϕ1 on the the tangent space R2 = T (u. v)}. v (t)} and we can . therefore the first fundamental form has the matrix G = 1 0 .v) U with respect to the basis {{1. we have:  r = a  u   r = b  v   E = a2      F = a · b     G = b2 hence .v) Examples. v) = r (u. For the sphere S R .1 First applications The length of a segment of curve on a surface Let S be a surface. a×b 1. F = 0. v) = r (u.140 Chapter 4.

4. On the other hand. The first fundamental form of a surface 141 compute its length using the matrix G. on the sphere S R . s0 ∈ I1 such that: ρ(t0 ) = ρ1 (s0 ) = r(u0 . for the length of the segment of ρ between t1 and t2 : t2 t2 lt1 . r) be a parameterization of a surface S . (I1 . where t ∈ (0. the length of the segment of curve between t1 = π and t2 = π we will get 2 π π l π . ρ = ρ(t)). We assume that the supports of the two curves intersect at r(u0 .10. v = t. v (t) = 1. hence. the first fundamental form of the 0 sphere has the matrix G = R2 1 cos2 u . The angle of two curves on a surface Let (U. 2 which was to be expected (the segment of curve is a quarter of a great circle on the sphere). v(t)) . Therefore. where  E(t) = E(u(t). (I. it follows 0 that u (t) = 0. If we want to compute. v(t))     G(t) = G(u(t). multiplied by R2 . for instance. ρ1 = ρ1 (s)) – two curves on S such that ρ(I) ⊂ r(U). we have. v0 ). v(t))      F(t) = F(u(t). .π = 2 π 2 R2 · 0 + 2 · 0 + R2 · 1dt = R · t π 2 = πR . there are t0 ∈ I. i. cos2 u = cos2 0 = 1. As along the curve we have u = 0. the matrix G will be the identity matrix. 2 Example.t2 = t1 ρ (t) dt = t1 E(t)u 2 + 2F(t)u v + G(t)v 2 dt. the curve given by the local equations (in the parameterization described in the previous example) u = 0. If the local equations of the two curves are  u = u1 (t)   (ρ)  v = v (t)  1 . along the curve. As we saw above. 2π) (the equator with a point removed). We take.e. ρ1 (I1 ) ⊂ r(U). v0 ).

v (s )} . Basically. following Stoker. cos θ = ρ (t0 ) · ρ1 (s0 ) = ρ (t0 ) · ρ1 (s0 ) Eu1 u2 + F(u1 v2 + u2 v1 ) + Gv1 v2 Eu1 2 + 2Fu1 v1 + Gv1 2 · Eu2 2 + 2Fu2 v2 + Gv2 2 . v0 )      F = F(u0 . where  E = E(u0 . v0 ) The area of a parameterized surface and  u = u (t0 )  1  1    v = v (t0 )  1  1  u = u (s )   2  2 0    v = v (s0 ) 2 2 . about the angle of the tangent vectors. the area should be a function associating to each oriented patch a positive number. General theory of surfaces  u = u2 (s)   (ρ1 )  v = v (s)  2 . All of them are more or less connected to integral calculus. so we are not going to enter into any details here. as in the case of plane geometric figures. Let (U. We choose. . v. v0 )     G = G(u0 . of course. the following three restrictions: a) The area should be given by an integral of the form A= U f dudv. rv (no higher derivatives of r should be involved!). where f should depend only upon u.142 and Chapter 4. ru . then the decomposition of the tangents vector at the intersection point with respect to the natural basis will be:  ρ (t0 ) = {u (t0 ). There are many ways of introducing the notion of area. respectively. v (t0 )}   1 1  ρ (s ) = {u (s ). r. subject to some restrictions. 3 We are speaking. as it is well known.  1 0 0 2 0 2 therefore the cosine of the angle of the curves3 at the contact point is. r) be an oriented parameterized surface.

however. of course). They will determine a polygonal surface inscribed into the surface. Let M1 = r(u + ∆u. respectively. v) ∈ D and M = r(u.10. where. The “classical” approach. Through M pass two coordinate lines. v + ∆v) be two points on these lines. It can be proved that the only formula for area that verifies these three axioms is A= U EG − F 2 dudv ≡ U ru × rv dudv. If ∆u and ∆v are small enough. Thus. The basic idea is that we already have a notion of area for plane figures. Let (U.10. Then we may consider that the area of the polygons that make up this polygonal surface goes to zero and define the . in particular for parallelograms. one from each family. where the sum in the middle term is taken after all the small curvilinear parallelograms that cover r(U).4. then the projection of the curvilinear parallelogram MM1 M M2 on the tangent plane to the surface at the point M is (approximately. The sides of this parallelogram are ru ∆u and rv ∆v and its area will be.2) We will give a heuristic motivation for the formula (4. we may discretize the domain of the parameterization and consider the images of the selected points. because no such claim is being currently made. shifted from M with the parameter shifts ∆u and ∆v. v). We want to define the area of r(D) ⊂ r(U). as being a “proof”.10.2). It should not be taken. the coefficients of the first fundamental form have been computed at the point of M. Namely. to define the area of r(D) as A = lim Σ∆σ = ∆u→0 ∆v→0 D EG − F 2 dudv. of course. v + ∆v). a plane parallelogram in the tangent plane. then ∆σ = ru ∆u × rv ∆v = ru × rv ∆u∆v = EG − F 2 ∆u∆v. let (u. Remark. M2 = r(u. v). It is only natural. c) A square of side length 1 has area 1. (4. We might expect to obtain for the area of domain on a surface an interpretation similar to the one we got for the length of a segment of curve. then. r) be a parameterized surface and D ⊂ U – a compact subset of U such that r(∂U) is a piecewise smooth curve in R3 . The first fundamental form of a surface 143 b) It is invariant with respect to rigid motions of the space and to parameter transformations that preserve the orientation. and M = r(u + ∆u.

General theory of surfaces area of our surface as being the limit of the area of the polygonal surface. 4. A(rv ) = nv . it just doesn’t work. the things can be fixed. A(ru ) = nu .11. the area may be infinite and for other ones finite.11 The matrix of the shape operator of a surface in the natural basis Let (U. v) = ru · nv     N(u. defined by H= Then the last equality reads H = G · A. rv rv · ru rv · rv .11. in particular. v) = r · n   u u     M(u. We multiply from the left with the matrix ru · nu rv = = We introduce the functions  L(u. compatible with the orientation. but this is a subject that belongs to theory of integration rather than to differential geometry.A. Schwartz shows. L M M N .1) and we get: nv ru · nu ru · nv = rv · nu rv · nv r ·r r ·r ru · ru rv · A = u u u v · A = G · A. because the limit is not independent on the type of polygons we consider and.144 Chapter 4. r) be a local parameterization of the oriented surface S . (4. as we saw earlier. we have ( nu nv ) = (ru rv ) · A. for some “polygonizations” of the surface. Unfortunately. as a celebrated example of H. We shall denote by A the matrix of the shape operator A with respect to the natural basis {ru . therefore we shall not insist. Of course. v) = r · n v v and the matrix H. with a little care. Since.2) ru rv (4. rv }.

11. r . M.11. The matrix of the shape operator 145 Since the scalar product on R3 is nondegenerate. N the expressions:  L = r · n = − n · r   u u  u2    M = ru · nv = − n · ruv     N = r · n = − n · r v v v2 or. r .6) . r )   H u v u2    1  M = − H (ru . then for the matrix of the shape operator we get A = G−1 · H. N in terms of the derivatives of the function r.  L = − 1 (r . M. 2 −F E EG − F (4.4) All we have to do now is to express the quantities L. the matrix G is invertible.11. If G−1 be its inverse.11.11. having in mind that n= while ru × rv = H(= EG − F 2 ). G−1 = 1 G −F . where. as a consequence.4.  uv   v u    r · n + r · n = 0 v v v2 whence we obtain from L. r ) H u v v2 ru × rv .3) If we perform the computation. we obtain: A= 1 GL − F M GM − FN 2 −FL + EM −F M + EN EG − F (4. we differentiate the relations ru · n = 0 and rv · n = 0 with respect to u and v and we get:  r 2 · n + ru · nu = 0  u     r · n + r · n = 0  uv  u v  r · n + r · n = 0 . To this end. ruv ) . as one can see very easily. the same is true for its restriction to any subspace and.5) (4. ru × rv (4.     N = − 1 (r . rv .

√ b2 + u2 b2 + u2 b2 + u2 On gets.  4. (4. η) = ϕ2 (ξ. which means that ϕ2 is symmetrical. v) = b +u  b  (b2 +u2 )3/2    0 . Proof. η ∈ T a S and two arbitrary real numbers α.1) Remark. η) linear ϕ1 bilinear A ϕ1 bilinear = = −αϕ1 (A(ξ1 ). η). β ∈ R. η) = −ϕ1 (A(ξ). General theory of surfaces Example. η). Proposition 4. We have ϕ2 (αξ1 + βξ2 . The minus sign in the previous definition is a consequence of our particular choice of sign in the definition of the shape operator.1. η ∈ T a S . b > 0. but then in the definition of the second fundamental form we had to introduce an extra minus sign.1. Then we have.12. ξ) self-adjoint = A −ϕ1 (η. A =  0  b    H = b G=    b 2 + u2 . η) + βϕ2 (ξ2 .12. η) = −ϕ1 (A(αξ1 + βξ2 ). after a straightforward computation:    √ b  0    1 0 2 +u2  . ∀ξ. Due to the symmetry. in order to be consistent with the generally accepted definition of the second fundamental form. For each a ∈ S . it is enough to prove the linearity in the first variable only. η). For the helicoid   x = u cos v      y = u sin v     z = bv .− √ . √  0 b 0  3/2 2 2 (b2 +u2 ) n(u. ξ) = −ϕ1 (A(η). ϕ2 (a) is a symmetrical bilinear form. We found natural to choose the shape operator to be the differential of the spherical map rather then the opposite of the differential.12. The second fundamental form of an oriented surface S is a map associating to each a ∈ S the application ϕ2 (a) : T a S × T a S → R given by ϕ2 (ξ. .12 The second fundamental form of an oriented surface Definition 4. (u. We take two arbitrary tangent vectors ξ. A(ξ)) ϕ1 symmetrical = −ϕ1 (A(ξ). we can define the orientation by putting b cos v u b sin v . η) = −ϕ1 (αA(ξ1 ) + βA(ξ2 ). √ .146 Chapter 4. v) ∈ R2 . η) − βϕ1 (A(ξ2 ). first of all: ϕ2 (η. η) = αϕ2 (ξ1 .

It should be noted. f. this way. D . as we saw before. we have  D = n · r 2   u      (4. as we mentioned earlier. in some books. D . r ) = − n · r   u u u u u u u u  ϕ (r . thus. ru ) ϕ2 (rv . D are usually credited to Gauss (in Disquisitiones). r ) = −ϕ1 ( n . The notations D. r) be a local parameterization of the oriented surface S . the letters L. The second fundamental form of an oriented surface which shows the linearity in the first variable and concludes the proof. 1 2 Example. ϕ2 (rv . r ) = −ϕ1 (A(r ). 147 Let (U. we get the matrix ϕ2 = − nu · ru nv · ru nu · rv L =− M nv · rv M not. The reader should be careful with the notations for the coefficients of the second fundamental form. we have. the meaning of the symbols is a little different: they are not the coefficients of the second fundamental form as we know it. D = N D D .2) D = n · ruv       D = n · r v2 Remark. ∀p ∈ T a S 2 R. the shape operator A is a homothety of ratio 1/R. for the coefficients of the second fundamental form with respect to the natural basis. used the notation e. compatible with the orientation. ϕ2 (ru . M.4.12. For them there is. . rv ) . y. r ) = ϕ (r . also. but rather √ these coefficients multiplied by EG − F 2 . therefore. A(p) = 1 p. ru ) ϕ2 (ru .12. R . that the matrix of the second fundamental form in the canonical basis is just −H. r ) = − n · r  2 u v 2 v u 2 v v u v v u v v and. For the sphere S = S R . Thus.e. g. that for Gauss. z}. i. N are used to denote the coefficients of the second fundamental form themselves. n = R {x. Also. Then the matrix [ϕ2 ] of the second fundamental form with respect to the canonical basis {ru . rv ) It follows. r ) = − n · r = − n · r ϕ (r . rv } has the form: ϕ2 = But  ϕ2 (r . however.

the first two fundamental forms are proportional.1. Then. The normal curvature of any plane curve is zero (in this case the angle θ(t) is always π .148 Thus.1) . Indeed. the same is true for the plane. ρ = ρ(t). is given by the formula kn (t) = ϕ2 (ρ (t).13. If the support of a parameterized curve lies on a straight line.13. therefore cos θ(t) ≡ 0). Examples. this time. The normal curvature of a parameterized curve ρ(t). The projection of the curvature vector k(t) of ρ ( as a signed scalar) on n(ρ(t)) is called the normal curvature of the curve ρ(t) at t and it is denoted by kn (t). lying on a surface S .13 The normal curvature. the curvature of the curve vanishes identically. The normal curvature of a curve on a surface can be expressed easily if we know the first two fundamental forms of the surface. R R R therefore. we have: Theorem 4.13. lying on S . kn (t) = k(t) · cos θ(t). Remark. for the sphere. Definition 4.13. clearly.2) (4. is the orthogonal projection on its osculating plane of the center of curvature of the normal section tangent to ρ at that point. then its normal curvature is always zero. The Meusnier’s theorem Let S be an oriented surface and n – the unit normal. q) = − p · q. where k(t) is the curvature of the curve ρ(t).1) has a simple geometrical interpretation (the Meusnier’s theorem): the center of curvature of a curve ρ.1. 4. 1. 2 2. q) = −ϕ1 1 1 1 p. it can be shown that these are the only two surfaces with this property. Let θ(t) be the angle between the osculating plane of ρ(t) and n(ρ(t)). Surprisingly as it might seem. q = − ϕ1 (p. We take a regular parameterized curve. because. actually. ϕ1 (ρ (t). The relation (4. Chapter 4. when the second fundamental form vanishes identically. General theory of surfaces ϕ2 (p.13. ρ (t)) . ρ (t)) (4. no matter on what surface is the curve situated. lying on an oriented surface S . Clearly.

From the hypothesis. We have ρ (s) = ρ1 (s(t)) · s (t) Thus. q) = = = ϕ1 (p. hence. We choose a local parameterization (U. i. p) ϕ2 (αq. Let p and q be the tangent vectors to the two curves at their common point. ρ1 (s). kn (t) = ϕ2 ρ (t) ρ (t) . p) ϕ1 (αq. q) ϕ2 (q. ρ (t)) 1 · ϕ2 (ρ (t). ρ (t)) = . v(s)). αq) α2 ϕ1 (q. Consequence. r) of the surface S and we assume that ρ1 (s) has in this parameterization the local equations u = u(s).e. Since the curvature of any regular parameterized curve is invariant to a parameter change. the proof is simpler in the case of naturally parameterized curves. ρ (t) ρ (t) = ϕ2 (ρ (t). from the theorem. ρ1 (s) = r(u(s). we can replace the curve ρ(t) by a n equivalent. ρ (t)) where ρ (t) . then the two curves have the same normal curvature at the contact point. naturally parameterized curve. Now. Proof. αq) α2 ϕ2 (q. q) . we come back to the initial parameterized curve. Thus. Then ρ1 (s) = ru2 · (u )2 + 2ruv · u v + rv2 · (v )2 + ru · u + rv · v . for the normal curvature of the curve ρ1 (s) we get the expression: kn (s) = k(s) · n(ρ1 (s)) = ρ1 (s) · n(ρ1 (s)) = = ru2 · n · (u )2 + 2ruv · n · u v + rv2 · n · (v )2 = = −L · (u )2 − 2M · u v − N · (v )2 = ϕ2 (ρ1 (s). If two curves on an oriented surface have a common point and they have the same tangent line at this point.13. As is many times the case with parameterized curves.4. The normal curvature. The Meusnier’s theorem 149 Proof. ρ1 (s(t)) = therefore. ρ (t) s (t) ≡ ρ (t) . v = v(s). kn = ϕ2 (p. q) ϕ1 (q. ρ (t) · ρ (t) ϕ1 (ρ (t). The curvature vector of the curve ρ1 (s) will be ρ1 (s). the natural parameter being the arc length. p = αq. ρ1 (s)).

we can define a vector of asymptotic direction as being one for which ϕ2 (h. General theory of surfaces Remark. The consequence is equivalent with the affirmation that the product kn = kα cos θα does not depend on the choice of the curve from the family. Definition 4. h) = 0. 4.1. It makes sense. Let p ∈ S a point of an oriented surface. We denote by kα the curvature of the curve ρα and by θα the angle between the normal to the surface and the osculating plane of the curve ρα . it actually depends only on the direction of the tangent vector of the curve. passing through the origin of h and whose tangent line is parallel to h. h) (4. The previous consequence can be interpreted in another way. clearly. A non-vanishing vector h ∈ T p S is said to have asymptotic direction if the normal curvature in its direction vanishes. Let S be an oriented surface and p ∈ S . lying on the surface. ϕ1 (h.13. we can define a map kn on the set of all non-vanishing tangent vectors to the surface with real values.150 Chapter 4. It is interesting for us to identify those directions for which the normal curvature vanish. Then at this point we have asymptotic directions if and only if the quadratic form associated to the second fundamental form of S at p is negatively semi-defined. and to speak about the normal curvature of the surface in the direction of this line or. Theorem 4. based on the previous section. passing through the same point and having the same tangent line at the contact point. passing through the tangency plane.3) The quantity kn (h) is called the normal curvature of the surface in the direction of the vector h (since. Accordingly. to choose an arbitrary straight line in the tangent plane. as stressed. We take a family of biregular parameterized curves on the surface. thus.1. h) . it depends only on the direction of the vector h.) Thus. If we choose a local parameterization . in other words. Alternatively. an asymptotic line or curve on a surface is a curve on the surface for which all the tangent vectors have asymptotic direction. but not on its length or sense. {ρα (t)}α∈A .14 Asymptotic directions and asymptotic lines on a surface We saw previously that the normal curvature of a surface at a given point and in a given direction can be expressed in terms of the first two fundamental forms of the surface and that.14. the normal curvature of an oriented surface in the direction of a vector h is the normal curvature of an arbitrary curve. although it is initially defined in terms of curves on surface. by putting kn (h) = ϕ2 (h.

v0 )h2 + 2D (u0 . v0 ) for some (u0 .14. v (t)}) to have asymptotic direction. v0 ) ∈ U.14.1. Let S be an oriented surface and ρ : I → S – curve on the surface. Let h = {h1 . v0 ) + D (u0 .4. We assume that there is a local parameterization of S . v0 ) + 2D (u0 . v0 ) ≤ 0. v(t)) · u (t) · v (t) + D (u(t). 1 2 As h 0. Theorem 4.2. then this condition simply means that D(u0 . v0 ) = 0. h2 h2 This equation. Then. the components {u (t). v(t)) · u 2 (t) + 2 · D (u(t). Indeed.1) Proof. From the proof of the previous theorem it follows that when the second fundamental form is negatively defined we have two asymptotic directions. v)) such that ρ(I) ⊂ r(U) and the local equations of the curve with respect to this parameterization are u = u(t). h has asymptotic direction if and only if D(u0 . h2 } ∈ T p S be a non-vanishing tangent vector to the surface at the point p. Any straight line lying on a surface is an asymptotic line. for instance. From the definition of the normal curvature of a curve on a surface. v0 )h2 = 0. that h2 0. with respect to the natural basis of the tangent space.14. v = v(t). Then ρ is an asymptotic line on S if and only if D(u(t). v0 ) · D (u0 . v0 )h1 h2 + D (u0 . while at the points when it is degenerate we have only one asymptotic direction (or two confounded ones).1. (4. r) of S such that pr(u0 . The differential equation of the asymptotic lines on a surface can be obtained directly from the definition. Then the previous equation can be written as 2 h1 h1 D(u0 . obviously has real solutions if and only if discriminant vanishes. the straight lines have zero curvature. Remark. it follows immediately that Proposition 4. (4. .14. but this is exactly the condition 4. v(t)) · v 2 (t) = 0. Proof.2) Proof. therefore their normal curvature at each point also vanishes.14. v0 ) − D 2 (u0 . r = r(u. Asymptotic directions and asymptotic lines on a surface 151 (U.14. we may assume. with respect to the local parameterization chosen. (U. The equation (4.2) is just that condition for the tangent vector of the curve (which has.

This is fortunate. General theory of surfaces Let us assume. Thus. given by the parameterization   x = u cos v. This is. From the definition of the normal curvature it follows immediately that. . at each point of the surface. we have. as it allows us to give a classification of the points of the surface. Let S be an oriented surface and (U. Let us consider the helicoid. v)) – a local parameterization of S .14.152 Chapter 4. Let S be an oriented surface and ρ a biregular parameterized curve on S . that the curve ρ from the previous theorem is biregular.14). u = const and v = const. which means.2. of course. 4. more specifically.3. The second one is not. Example. the line v = const (see the figure 4. Another immediate result concerning the asymptotic lines is the following: Proposition 4. at each point of the curve. v(t)) = 0. Then ρ is an asymptotic line if and only if at each point its osculating plane coincide with the tangent plane to the surface at that point. according to the sign of the second fundamental form or. in this particular case. according to the sign of its discriminant DD − D 2 .      y = u sin v     z = b · v where b is a constant.15 The classification of points on a surface The first fundamental form of a surface is positively defined. where n is the unit normal vector to the surface. v) = √ −b/ b2 + u2 . if ν is the unit principal normal vector of the curve. Then the coordinate lines u = const and v = const are asymptotic lines on r(U) if and only if D = D = 0. This means that. we obtain the following characterization of the asymptotic lines: Theorem 4. one of the coordinate lines is a straight line (or a line segment). that is curvature is always strictly positive. r = r(u. We notice that the helicoid is a ruled surface (see the last chapter) and at each point. D (u. A straightforward computation leads to D = D = 0. This actually means that the principal normal of the curve lies on the tangent plane of the surface. now. two asymptotic lines and these are nothing but the coordinate lines. then the curve is anasymptotic line if and only if ν(t) · n(u(t). in particular.

A point a ∈ S of an oriented surface is called (i) elliptic if the second fundamental form is positively defined at a. if the second fundamental form is negatively defined at a.6: Parabolic points on a surface It is not difficult to see that this definition do not depend on the choice of the local parameterization. also. if all the coefficients of the second fundamental form vanish at a. now separately. what happens in each particular case and.15. give some examples. (iv) flat. Let us discuss.5: Asymptotic lines on the helicoid Definition 4. The classification of points on a surface 153 Figure 4.15. . (ii) parabolic if the second fundamental form is zero. Figure 4. (iii) hyperbolic.1. but at least one of the coefficients is different from zero. or planar.4.

154 Chapter 4. Such points we can find. An example of surface that has only elliptic points is the ellipsoid.7: Elliptic points on a surface Parabolic points. . an asymptotic direction. Thus. Applying the Meusnier’s theorem. clearly. given. it is possible for kn to change sign and there are exactly two direction where it vanishes. many surfaces on which we can meet all three kind of points (for instance on the torus). (4. c sin v). therefore no asymptotic line passes through an elliptic point. on a hyperbolic paraboloid. v) = (a cos u cos v. Hyperbolic points. 4 It is not necessarily positive. of course. General theory of surfaces Elliptic points. There are. At an elliptic point. In this case the normal curvature does not change the sign. Thus.15. the normal curvature has the same signs in all directions4 . The hyperbolic points are also called saddle points. This is. The cylinders and cones (with the apex removed) have only parabolic points. this means that the centers of curvature of all normal sections lie on the same side of the surface.1) At an elliptic point there is no real asymptotic direction. through a parabolic point of a surface passes only one asymptotic line. In the case of hyperbolic points. for instance. for instance. but there is exactly one direction where it vanishes. Figure 4. by the parameterization r(u. b sin u cos v. through a hyperbolic point of a surface pass two asymptotic lines.

In fact.15. The shape of the surface around a flat point might be quite complicated and it is difficult to study it. in many cases. .4. The surface from the figure 4.9: Hyperbolic points on a surface Flat points. The classification of points on a surface 155 Figure 4.8: The monkey saddle Figure 4. when proving a theorem in surface theory one explicitly assumes that the surface has no flat points.15 (the monkey saddle has a flat point at the origin of coordinates.

16. at each point. taken with opposite sign. 2. The directions on the tangent plane to an oriented surface S at a point a ∈ S .1. an oriented surface either has two orthogonal principal directions (if the eigenvalues of A are distinct). Definition 4. T a S .16. principal curvatures. At each point. i = 1. therefore kn (e) = ϕ2 (e. General theory of surfaces 4. Proposition 4. either all the directions are orthogonal (if the two eigenvalues coincide). e1 ) = θ. The principal curvatures of a surface are the eigenvalues of the shape operator. Definition 4. Gauss curvature and mean curvature Definition 4. e) −A(e) · e −λ · e · e = = = −λ. We shall fix now a point of the surface and ask the following question: find the normal curvature in the direction of a vector e. Proof. where λ is the eigenvalue corresponding to e. corresponding to the eigenvectors of the shape operators A are called the principal directions of the surface at the point a. ϕ1 (e. Thus.2.16. An orthonormal basis {e1 . has a principal direction. the vectors of a basis of principal directions are subject to A(ei ) = −ki ei . e2 } of the tangent space at a point of a surface is called a basis of principal directions of the tangent space if the vectors of the basis have principal directions. A principal curvature of an oriented surface S at a point a ∈ S is the normal curvature of S at a in a principal direction. Definition 4. such that ∠(e.16 Principal directions. e) e·e e·e Hereafter we shall denote by k1 and k2 the principal curvatures and we will always assume that k1 ≥ k2 .16.4.3. . A curve (Γ) on a surface S is called a principal line or a curvature line if its tangent. If e is an eigenvector of A. then A(e) = λ · e.156 Chapter 4.16. Remark.1.

From the Euler’s formula. where k1 and k2 are the principal curvature of the surface. e) −A(e) · e = = −A(e1 cos θ + e2 sin θ) · (e1 cos θ + e2 sin θ) = ϕ1 (e. therefore kn (e) = ϕ2 (e. It is clear now that the maximum value of the normal curvature is reached for θ = 0 (we assumed that k1 ≥ k2 !) and. Thus. Proof. is given by the Euler’s formula: kn (e) = k1 cos2 θ + k2 sin2 θ. e) e·e =1 = (k1 cos θ · e1 + k2 sin θ · e2 ) · (e1 cos θ + e2 sin θ) = k1 cos2 θ + k2 sin2 θ.16. we shall assume that e is a unit vector: e = 1. (4. Then the normal curvature at a point of the surface. we have: Proposition 4. in the direction of a vector e. we get kn = k1 .16. Principal directions and curvatures 157 As the length of e is not important. 2 Definition 4. obtaining kn = k2 . An immediate consequence of the Euler’s formula is: Theorem 4.16. in this case. The quantities Kt = k1 ·k2 and Km = 1 (k1 +k2 ) are called.16. The total and mean curvatures of a surface can be computed easily if the matrix of the shape operator in an arbitrary basis is known.16. we obtained: Theorem 4.16.1. Indeed. Then e = e1 · cos θ + e2 · sin θ. when the vector e rotates around the origin of the tangent space of the surface at that point. while θ = ∠(e.16. Let S be an oriented surface.16. we have kn (e) = k1 cos2 θ + k2 (1 − cos2 θ) = k2 + (k1 − k2 ) cos2 θ.3) (4.5. The principal curvatures of a surface at a point are extreme values of the normal curvature of the surface in the direction of a vector e.4.2. 1 Km = − Tr A 2 Kt = det A. e1 ). 2 the total (Gaussian) and mean curvature of the surface.1) . respectively. while the minimal value – for θ = π .2) (4.2.

since the curve γ lies also on S 2 . sometimes. . which means that they are the same in any basis. in some special situations it is possible to find these lines using other methods. On the other hand. For instance.e. the matrix of the shape operator will be: A= therefore det A = k1 · k2 = Kt 1 1 1 − Tr A = − (−k1 − k2 ) = (k1 + k2 ) = Km . since the eigenvalues of the shape operator are just the opposite of the principal curvature. although the matrix of the operator does change. belonging to the German mathematician Joachimstahl. Let ni be the unit normals of the two surfaces (i = 1. In a basis of principal directions. then n1 = −k1 · r .. if we modify the basis. we have r ⊥ n2 . Let γ be a curve lying at the intersection of two regular oriented surfaces S 1 and S 2 from R3 . General theory of surfaces Proof. Theorem. Anyway. Then γ is a curvature line on S 1 iff it is a curvature line on S 2 . i. we have d 0 = ( n1 · n2 ) = n1 · n2 + n1 · n2 . As it is well known from linear algebra. generally. 2). Such an instance is exemplified by the following theorem. dt If γ is a principal line on S 1 . along the curve γ we have n1 · n2 = const. Then. 2 2 2 −k1 0 0 −k2 Joachimstahl’s theorem We will see below how can one find the lines of curvature of a surface by integrating a differential equation. Proof.158 Chapter 4. Let us assume that S 1 and S 2 intersects under a constant angle. Let r = r(t) be a local parameterization of the curve γ. the determinant and the trace are invariants of any linear operator. From here and from the previous formula we obtain that n1 · n2 = 0. if we know the curvature lines of a surface it is possible to find such lines on another surface. therefore n1 · n2 = 0. where k1 is one of the principal curvatures of the surface S 1 . since n1 · n2 = const.

that. that along a parallel we should have ∠( n. It is obvious. too.16. A(v) = λ · v. Lemma. which implies. Consequence. passing through the rotation axis and S . equal to π . Principal directions and curvatures 159 Since n2 ⊥ n2 (since n2 has constant length). But. that it is a line of curvature also for the surface S . it follows from here and from the previous equation that n2 ⊥ r or. The meridians and the parallels on a revolution surface are lines of curvature. Since v has principal direction iff it is an eigenvector of the shape operator A. in other words. that there is a k2 ∈ R such that n2 = −k2 r . it follows that v has principal direction iff A(v) × v = 0. we will indicate a way to find the eigenvectors of the shape operator.16. the lines of curvature are curves on a surfaces whose tangent vectors are eigenvectors of the shape operator. lies on the plane Πm . Therefore. before showing how one can find the lines of curvature. the meridian is. A parallel is an intersection curve between the surface S and a plane Π p . using the Joachimstahl theorem. γ is a line of curvature on the surface S 2 . from the .4. D D D (4.e.16.4) Proof. 4. If p ∈ Πm ∩ S . in particular. It follows. a line of curvature in the plane Πm . i. which means that all the plane curves are lines of curvature. A tangent vector v = v1 ru + v2 rv has principal direction iff v2 −v1 v2 v2 2 1 E F G = 0.1 The determination of the lines of curvature As we saw earlier. from symmetry reasons. A meridian is obtained by intersecting a plane Πm . Let γ be the rotating curve and S the resulted revolution surface. Π p ) = const and we apply again the reasoning we used before. As for the plane the second fundamental form vanishes identically. then the unit normal of the surface S . i. perpendicular to the rotation axis. n(p). also.e. passing through a point of the curve γ. the same is 2 true for the shape operator. therefore the normal of the surface and the normal of the plane Πm make up a constant angle. Let r : U → R3 be a local parameterization of an oriented surface S . Proof.

The relation (4. General theory of surfaces A(v) = A · v = G−1 · H · v = = or 1 GL − F M GM − FN v1 = H 2 −FL + EM −F M + EN v2 1 (GL − F M)v1 + (GM − FN)v2 H 2 (−FL + EM)v1 + (−F M + EN)v2 αu 1 1 A(v) = 2 [(GL − F M)v1 + (GM − FN)v2 ] ru + 2 [(−FL + EM)v1 + (−F M + EN)v2 ] rv .16.16. . with the local equation ρ(t) = r(u(t). A(v)×v = 0 ⇐⇒ (αu ru +αv rv )×(v1 ru +v2 rv ) = 0 As ru × rv ⇐⇒ (αu ·v2 −αv ·v1 )·(ru ×rv ) = 0. while (4. v(t)). by eliminating the parameter t.5) (4.6) = 0.16. since the surface is regular.160 definition. having in mind the notations we made.16. 1 2 Using now the coefficients of the second fundamental form instead of the components of the matrix H.6). Chapter 4. 1 2 which is just another form of the relation (4. (4. H H αv Therefore. the previous relation becomes (ED − FD)v2 + (ED − GD)v1 v2 + (FD − GD )v2 = 0.6) express. (FL − EM)v2 + (GL − EN)v1 v2 + (GM − FN)v2 = 0.16.7) Proof. it follows that A(v) × v = 0 ⇐⇒ αu · v2 − αv · v1 = 0 or. 0. U) of a surface S . Then γ is a line of curvature on S iff (ED − FD)u 2 (t) + (ED − GD)u (t)v (t) + (FD − GD )v 2 (t) = 0 or (ED − FD) + (ED − GD) dv dv + (FD − GD ) du du 2 (4.16. clearly.7) follows immediately from (4.16.4) Consequence (the differential equation of the lines of curvature). the condition for the vector ρ = u (t)ru + v (t)rv to have principal direction. Let γ be a curve lying in the domain r(U) of a local parameterization(r.

      H  H2  H2 1 1 Km = − TrA = − 2 2FD − GD − ED 2 2H = DG − 2D F + D E .16. A non-flat point of a surface is 1.5) .16. hyperbolic iff Kt < 0. elliptic iff Kt > 0. 2 Km − Kt . Let r : U → R3 be a local parameterization of an oriented surface S .3) (4.1) (4.2) Proof.e. by the formulas: DD − D 2 H2 DG − 2D F + D E Km = .2 The computation of the curvatures of a surface Theorem.16. k1 = K m + k2 = K m − Corollary. (4. The principal curvatures k1 and k2 are the roots of the equation k2 − 2Km · k + Kt = 0.16. respectively. As we saw earlier. 3. 2. H 2 FD − ED FD − ED therefore. Then the total and the mean curvatures of S are given.4.16. parabolic iff Kt = 0. the matrix of the shape operator of a surface is given by A= 1 GL − F M GM − FN 2 −FL + EM −F M + EN H or A= 1 FD − GD FD − GD . i. Kt = det A = 1 F 2 D 2 − EFD D − FGDD + EGDD − F 2 DD + EFD D + H4      DD − D 2    1     +FGDD − EGD 2 = 4 (EG − F 2 ) ·(DD − D 2 ) = . 2 Km − Kt . 2H 2 Kt = (4.16.4) (4. Principal directions and curvatures 161 4.16. 2H 2 Corollary.

the first form is positively defined. with the coefficients depending smoothly on the coordinates on U. We shall obtain now similar formulae for the second order derivatives of the radius vector. They define. n for a basis of the vector space R3 .17 4. in the case of curves. It is the aim of this section to establish the compatibility conditions and to formulate the existence and uniqueness theorem for parameterized surfaces. in particular. we can formulate the question in the following way: If we are given a domain U ⊂ R2 and two families of symmetric bilinear forms.17. the curvature and the torsion completely determine a space curve (up to a rigid motion of the space). the shape operator of the surface. therefore.v) vectors can be expressed in terms of the vectors themselves. the integrability conditions for a system of partial differential equation). first of all. r) is a regular parameterized surface. Thus. We may ask whether a similar result holds for the case of surfaces. These formulae should be of the form ru2 = Γ1 ru + Γ2 rv + A n 11 11 ruv = Γ1 ru + Γ2 rv + B n 12 12 rv2 = Γ1 ru + Γ2 rv + C n 22 22 The coefficients Γ from these equations are called the Christoffel’s coefficients (of the second kind). General theory of surfaces 4. ru2 · rv and the analogues in terms of the coefficients of the first fundamental (4. the first two fundamental forms could be good candidates. for each (u. is there any regular parameterized surface r : U → R3 such that the two families of bilinear forms be the two first fundamental forms of this surface? The answer is not affirmative. B. we shall express the scalar products ru2 · ru . the coefficients of the first two fundamental forms of a surface are not independent. however. then.2 The differentiation rules.1 The fundamental equations of a surface Introduction We saw that. C are easily identifiable as the coefficients of the second fundamental form. v) ∈ U.17. We already saw how to express the derivatives of the normal versor.162 Chapter 4. the vectors ru . our initial data should satisfy some compatibility conditions (which are. Therefore. Christoffel’s coefficients If (U. but. 4. such that. at each point of U. If. in fact. It is not completely obvious what are the entities that should replace the curvature and the torsion. clearly. rv . this conditions are fulfilled. because.17.1) . essentially. the derivatives of these r(u. as we shall see. The coefficients A. then the negative answer turns into an affirmative one. To get the others.

Using the formulae (4. first of all.1) successively by ru and by rv .9) .7) Returning to our problem.17.4) (4. whence.17.3) Exactly in the same manner. The fundamental equations of a surface 163 form and its derivatives.17. starting from the definition of the other two coefficients of the first fundamental form.17.17.4. exactly in the manner.17. we get the system  EΓ1 + FΓ2 = 1 E  11  2 u 11 . ru 2 = E. starting from the other two equations from (4.17.5) (4.8) and.17. (4.17.5).17.17. we get ruv · ru = 1 E 2 v (4.6) (4. we multiply the first equation of (4. We have.17.1). differentiating with respect to u. we will obtain 1 ruv · rv = Gu 2 1 ru2 · rv = Fu − Ev 2 1 rv2 · ru = F v − G u 2 1 rv2 · rv = Gv 2 (4.2) and (4.17. as well as the definitions of the coefficients of the first fundamental form. we get 1 ru2 · ru = Eu .2) 2 Differentiating the same equality with respect to v.  1 FΓ + GΓ2 = F − 1 E  u 2 u 11 11 It is very easy to solve this system and we get   1 Γ = GEu − 2FFu + FEv  11    2(EG − F 2 )      2 Γ = 2EFu − EEv − FEu   11 2(EG − F 2 ) (4. we get   1 Γ = GEv − FGu  12    2(EG − F 2 )      Γ2 = EGu − FEv   12 2(EG − F 2 ) (4.

13)     n = −D r   v G v One should not be surprised that the partial derivatives of the unit normal vector are. This is the reason why in many books the shape operator (or its opposite) is called the Weingarten’s operator or Weingarten’s mapping. we shall get immediately the formulae  FD − ED FD − GD    nu = r + r   2 u  EG − F EG − F 2 v  (4.11)      n = FD − GD r + FD − ED r   v EG − F 2 u EG − F 2 v Remark. Christoffel’s and Weingarten’s coefficients in curvature coordinates Let us assume that in our parameterization the coordinate lines are lines of curvature.17. Γ12 =  12  ∂v ∂u        1  Γ = − Gu . General theory of surfaces   1 Γ = 2GFv − GGu − FGv  22    2(EG − F 2 )  .  22 22 E ∂v while the Weingarten’s equation become:  D    n u = − ru    E  (4. the Christoffel’s coefficients become:   1 Γ = 1 Eu = ∂ ln E.17. nothing but the definition of the lines of curvatures.17.17. Clearly. Γ2 = ln G. as we saw earlier. we should have F = 0 and D = 0 on the entire support of the parameterization. Then.12) ln E. therefore they are called the Weingartens’s formulae.  11   11  2 E ∂u 2G       ∂ ∂  1 (4. each of them. Then. . colinear to one of the partial derivative of the radius vector. This is.164 and Chapter 4. actually. looking back at the expression of the shape operator in terms of the coefficients of the first two fundamental forms. Γ2 = − Ev . Γ2 = ∂ ln G.11) were obtained by the German mathematician Julius Weingarten. (4.10)     2 Γ = EGv − 2FFv − FGu   22 2(EG − F 2 ) As for the derivatives of the normal versor. as one can convince oneself easy. these formulae uniquely define the shape operator. The formulae (4.17.

(4. In any local parameterization of a surface.4. and the relations ∂D ∂D − = DΓ1 + D (Γ2 − Γ1 ) − D Γ2 12 12 11 11 ∂v ∂u ∂D ∂D − = DΓ1 + D (Γ2 − Γ1 ) − D Γ2 . ∂Γ2 11 ∂v ∂Γ1 12 ∂u ∂Γ1 22 ∂u ∂Γ2 12 ∂v ∂Γ2 12 ∂u ∂Γ1 − 11 ∂v ∂Γ1 − 12 ∂v ∂Γ2 − 22 ∂u − + Γ1 Γ2 + Γ2 Γ2 − Γ1 Γ2 − Γ2 Γ2 = EKt 11 11 11 21 12 11 12 12 + Γ2 Γ1 − Γ2 Γ1 = FKt 12 12 11 22 (4. while D.17.14) + Γ 1 Γ1 22 11 + Γ 2 Γ1 22 12 − Γ1 Γ1 12 12 − Γ2 Γ1 12 22 = GKt + Γ1 Γ2 − Γ1 Γ2 = FKt . obviously. We summarize them in the following theorem. the relation ru2 v − ruvu = 0. that we will call the Gauss’ equations and the Codazzi-Mainardi’s equations. let us rewrite the Weingarten’s equations as   n = a11 r + a12 r  u  u v . the following systems of equations are fulfilled.17. Here E.15) called the Codazzi-Mainardi’s equations. 11 11 . D . But ru2 = Γ1 ru + Γ2 bv + D n.1 (Gauss.17. 22 22 12 12 ∂v ∂u (4. Mainardi). Proof. respectively.17.17. Theorem 4. G are the coefficients of the first fundamental form of the surface. The fundamental equations of a surface 165 4. F. 12 12 22 11 called the Gauss’s equations.17. D are the coefficients of the second fundamental form. Codazzi. To simplify a little bit the notations.3 The Gauss’ and Codazzi-Mainardi’s equations for a surface We shall prove now that between the coefficients of the first two fundamental forms of a parameterized surface exists some relations.16)  n = a r + a r  v 21 u 22 v We have.

12 12 ∂u ruvu = .166 hence Chapter 4. 11 11 ∂v Analogously. 12 12 hence ∂Γ1 ∂Γ2 ∂D 12 ru + Γ1 ru2 + 12 rv + Γ2 ruv + n + D nu = 12 12 ∂u ∂u ∂u ∂Γ1 = 12 ru + Γ1 (Γ1 ru + Γ2 rv + D n)+ 12 11 11 ∂u ∂Γ2 ∂D + 12 rv + Γ2 (Γ1 ru + Γ2 rv + D n) + n + D (a11 ru + a21 rv ) = 12 12 12 ∂u ∂u   1   ∂Γ     = ru  12 + Γ1 Γ1 + Γ2 Γ1 + a11 D  +   12 11 12 12 ∂u  2   ∂Γ      + rv  12 + Γ1 Γ2 + Γ2 Γ2 + a21 D  +   12 11 12 12 ∂u ∂D +n + Γ1 D + Γ2 D . ruv = Γ1 ru + Γ2 rv + D n. General theory of surfaces ∂Γ1 ∂Γ2 ∂D 11 1 ru2 v = r + Γ11 ruv + 11 rv + Γ2 rv2 + n + D nv = 11 ∂v u ∂v ∂v ∂Γ1 = 11 ru + Γ1 (Γ1 ru + Γ2 rv + D n)+ 11 12 12 ∂v 2 ∂Γ ∂D n + D(a12 ru + a22 rv ) = + 11 rv + Γ2 (Γ1 ru + Γ2 rv + D n) + 11 22 22 ∂v ∂v   1   ∂Γ11   1 1 2 1   + Γ11 Γ12 + Γ11 Γ22 + a12 D + = ru    ∂v   2   ∂Γ11   1 2 2 2    + Γ11 Γ12 + Γ11 Γ22 + a22 D + + rv   ∂v ∂D +n + Γ1 D + Γ2 D .

∂u 2E ∂u ∂u 4.17. (4.4. If the coordinate lines are lines of curvature. To shorten the formulas. If we equate to zero the coefficient of n. using. using the expressions of a21 and a22 . one sees that one gets the first of the Codazzi-Mainardi’s equations. The fundamental equations of a surface If we subtract the previous relations. in the following we shall denote the coordinates by u1 and 2 instead of u and v and we shall use the index notation to denote the components of u the matrices of the first two fundamental forms of the surface. the first of the Gauss’ equations. If we equate to zero the coefficient of ru follows. we obtain   1   ∂Γ11 ∂Γ1   12 2 1 2 1    0 = ru2 v − ruvu = ru  − + Γ11 Γ22 − Γ12 Γ12 + a12 D − a11 D  +  ∂v ∂u   2   ∂Γ11 ∂Γ2   12 1 2 2 2   − + Γ11 Γ12 + Γ11 Γ22 + a22 D − a21 D  + + rv    ∂v ∂u ∂D ∂D − + Γ1 D − D (Γ2 − Γ1 ) + Γ2 D . then the Codazzi-Mainardi equations can be written a lot simpler:  ∂ ln E D ∂E  ∂D     ∂v = D ∂v + 2G ∂v   (4. It was first established by the French mathematician Ossian Bonnet. this time. The other three equations can be obtained in the same way.18) . for the coefficients of the first fundamental form we shall write g11 = E. g22 = G.17.17. the relation ruv2 = rvuv .17. Corollary 4. rv and n are linearly independent. while from the coefficient of rv follow the second of the Gauss’s equations. More precisely.1. g12 = g21 = F.17)   ∂D   D ∂G ∂ ln G    = +D . in 1860.4 The fundamental theorem of surface theory The theorem we are going to prove in this section is the analogue of the existence and uniqueness theorem for space curves. +n 12 12 11 12 ∂v ∂u 167 As the vectors ru .17. a linear combination of them can vanish only if each coefficient vanishes.

∂ui (iii) n(u0 ) = n(0) . General theory of surfaces while for the coefficients of the second fundamental form we shall write h11 = D. On U there are given the symmetric matrix functions gi j = gi j (u1 .. we shall denote by g the determinant of the matrix of the first fundamental form. r2 (0) . 1860). where i and j can take the values 1 and 2. (iv) gi j and hi j are the coefficients of the first two fundamental forms of the parameterized surface r (with respect to the orientation of r defined by the unit normal vector n).19) Also. therefore we shall always have rij = rji . hi j = hi j (u1 . such that the following conditions are fulfilled: (i) r(u0 ) = p0 (the surface “passes” through p0 for u = u0 ). Finally.168 Chapter 4. i = 1.e. u2 ). . i. 2 (4. in this context. n(0) ∈ T p0 R3 . u2 ).17. (4. n(0) · n(0) = 1. Then there exists a single regular parameterized surface of class C 3 . h22 = D . h12 = h21 = D . the components of the two functions verify the Gauss and Codazzi-Mainardi compatibility conditions.20) of classes C 2 and C 1 . with V ⊂ U – an open set. Clearly. r2 (0) . Let U ⊂ R2 be an open set. n(0) } is a right-handed basis of the vector space T p0 R3 . Theorem (Ossian Bonnet. j = 1. u2 ) ∈ U. while the triple {r1 (0) . 2. we shall denote by ri the derivative of r with respect to the coordinate ui and by rij – the second order derivative of r with respect to the coordinates ui and u j . (ii) ∂r (u0 ) = ri (0) .17.17. r : V → R3 . respectively. p0 ∈ R and the vectors 0 0 r1 (0) . at least C 2 ). (4. We choose u0 = (u1 . as we always assume that the surfaces are as as smooth as one expects to be (i. the order of differentiation is not relevant.21) such that ri (0) ·rj (0) = gi j (u0 ). such that for each (u1 . n(0) ·ri (0) = 0. moreover. u2 ) ∈ U the quadratic for associated to the bilinear form whose matrix is (gi j ) is positively defined and.

0. the system of partial differential equations   ∂r    1 = r1 . which are satisfied.17. (4.22) with respect to the unknown functions r1 .   ∂u  .23)   2  ∂ n  ∂2 n     j k = k j ∂u ∂u ∂u ∂u are equivalent. due. it follows that there exists an open neighborhood W ⊂ U of the point u0 and a set of three C 2 vector functions5 r1 . 1} We leave to the reader to check that. The fundamental equations of a surface Proof. Therefore.8)–(4. which are solutions of the system (4.17. indeed. for instance. where the coefficients Γkj are computed i with the formula (4. mainly.17.17. n. r2 .24) . n : W → R3 . we can make. as the system is of first order.22). these vectors verify the conditions from the hypothesis. r 2 =       g11 (u0 )   g11 (u0 )    (0)  n = {0.17. We consider. to the Gauss-Weingarten equations.4. r2 . . the following choice:  (0) r  1 = g (u ). 0    11 0      (0)  g12 (u0 )   g11 (u0 )g22 (u0 ) − (g12 (u0 ))2     (4. 0. 0 . as we saw. also. Indeed.10). to the fact that the quadratic form associated to gi j is positively defined. because the compatibility conditions  2  ∂ ri ∂2 ri     j k = k j  ∂u ∂u   ∂u ∂u (4. . with the given initial condition at the point u0 . by hypothesis. This is a linear and homogeneous system which is completely integrable. now.  ∂u i   k=1    ∂n 2 2     hi j g jk rk  i =−  ∂u j=1 k=1 169 (4. that a set of initial conditions as in the theorem always exists. from standard results from the theory of partial differential equations of first order.25)    ∂r     2 = r2 ∂u 5 The order of smoothness is with one unit greater then the smallest order of smoothness of the coefficients.17.17. We consider the system of partial differential equations  ∂r 2  i    j = Γkj rk + hi j n. Let us notice.17.

completely integrable.28)  ri · n = 0.27) ∂ui which is true.26) (4.k=1 hil g (rk · rj ) + l=1 Γi j (rl · n) + hi j ( n · n)       ∂( n · n)  2    hil glk (rk · n)   ∂ui = −2 l. We are not done yet. But this follows immediately if we prove that g is the first fundamental form. It is.17. actually. because we still have to show that gi j and hi j are the first two fundamental forms of the parameterized surface defined by r. because ∂r ∂r gi j = i · j ∂u ∂u and then ∂r ∂r × j 0.  i j     . Therefore. we shall compute the derivatives with respect to the coordinates of the scalar products. i ∂u ∂u because the square of the norm of this vector is not zero (as it is equal to the determinant of the first fundamental form. General theory of surfaces This system is. applying again the existence and uniqueness theorem. as one can convince oneself looking at the first equation (4. enough to show that. all over V.22). it follows that there exists an open neighborhood V ⊂ W ⊂ U of u0 and a single C 3 function r : V → R3 such that r(u0 ) = p0 .170 Chapter 4. (4. the following relations are fulfilled:  r · r = g i j . since the form is positively defined). which is strictly greater then zero. because of the symmetry of the second matrix (h ji = hi j ) and because of the symmetry of the Christoffel’s coefficients in the lower indices.17.k=1 .17. again. (4.    n · n = 1 ∂u j To this end.17. Apparently. because the integrability condition ∂2 r ∂2 r = ∂ui ∂u j ∂u j ∂ui is equivalent to the condition ∂ri = ∂rj (4. taking into account the Gauss-Weingarten equations and we get the system of first order partial differential equations:  ∂(r · r ) 2 2  i j   l l    ∂uk = Γik (rl · rj ) + Γ jk (rl · ri ) + hik (rj · n) + h jk (ri · n)   l=1 l=1     ∂(r · n)   j 2 2  lk l .29)    ∂ui = − l. we also have to show that r is regular.17.

29).17. As the solution is unique. (4. the functions (4. exactly. for the moment. We are left with the proof of the fact that hi j are the coefficients of the second fundamental form of r. For the second equation. simply. after the substitution.ri are the derivatives of r.17. n) > 0 1 2 which means that we proved already that . by bi j these coefficients. they are given by 2 2 2 bil = −ni · rl = j.17.17. .28). which means that it has a single solution for prescribed initial values.   i . We “guess” that this solution is. the left-hand side is zero.28) and we shall check this in the following.k=1 hi j g jk rk · rl = j.gi j are the coefficients of the first fundamental form of r. the first equation becomes ∂gi j = ∂uk 2 2 Γl gk j + ik l=1 l=1 Γljk gli which is very easy to prove. The fundamental equations of a surface 171 As probably the reader suspects already (and it is asked to check for himself). (4.  i j     r · n = 0.k=1 hil glk gk j + hi j = − l=1 hil δlj + hi j = −hi j + hi j = 0. For the last equation.17. Finally. r . Let us denote. there is nothing to check: if we substitute (4.17.28) give a solution of the system (4. while the right-hand side becomes: 2 2 − l. this system is. As we know. completely integrable. j which concludes the proof of the Bonnet’s theorem.k=1 hi j g jk gkl = j=1 hi j δl = hil . again. using the definition of the Christoffel’s coefficients.30)   n · n = 1      (r . we get. verify the initial conditions of the theorem. . 0 = 0. for the given initial conditions we have  r · r = g i j . which.4. Thus. obviously. so we are done.n is the unit normal of the parameterized surface given by r. .

18. in many books.e. in the second half of the nineteenth century he was considered an important geometer. As we just saw in the previous section. We start with the formula Kt = which we rewrite in the form Kt (EG − F 2 ) = DD − D 2 . given by Gauss in 1827.4 (Gauss. using the Latin term used by Gauss. rv ) · (rv2 . the Non-Euclidean geometry. although Struik credits the formula which will be established to the Italian mathematician Brioschi. in principle. in his famous Disquisitiones circa superficies curvas “theorema egregium”. The original proof. It turns out. where was mentioned. the remarkable theorem.18. ru . This would mean. now.18 The Gauss’ egregium theorem The theorem that we are going to prove in this section (and which is implicit in the equations of Gauss above) is one of the most important theorem in classical differential geometry. having in mind the expressions of the coefficients of the second fundamental form. For the history of mathematics. The total curvature of a surface of class at least C 3 depends only on the coefficients of the first fundamental form of the surface and their first order derivatives with respect to the coordinates. however.e. The proof we are going to give here is the first different proof.e. 1827). Kt (EG − F 2 )2 = (ru2 .1) DD − D 2 . Proof. (4. however. the product of two determinants. rv )2 . that the situation is different.18. belonging to the German mathematician Richard Baltzer (1867)6 . the total curvature of a surface in R3 can be expressed in terms of the determinants of the first two fundamental forms of the surface. General theory of surfaces 4. rv ) − (ruv . ru .172 Chapter 4. is quite complicated. Not by accident Gauss called it. i.2) The right hand side of the equation (4. that the total curvature depends both on the intrinsic data (the first fundamental form) and extrinsic data (i. there are important his “Elementhe der Mathematik”. for the first time. ru . published in several editions. we have: Theorem 4. There exists several proofs of this theorem (which is called. in fact. i. The name of Richard Baltzer (1818–1887) is not well known today. We shall use.2) can be written in a more convenient form if we notice that each term is. the second fundamental form). namely theorema egregium. EG − F 2 or. 6 . (4.

(4.6) .3) (4. c) · (d. e. known from vector algebra: a·d a·e a·f b·d b·e b·f . f) = c·d c·e c·f Using the formula (4. F = ru · rv . The Gauss’ egregium theorem 173 the following formula for the product of two determinants.4. Indeed. F G ruv · rv Thus. the formula (4. the remaining terms are of two kinds: either a product of a second order derivative of r and a first order one.2) becomes Kt (EG − F 2 )2 = (ru2 · rv2 − ruv )(EG − F 2 )+ 0 ru · rv2 + rv · rv2 ru2 · ru ru2 · rv E F − F G 2 (4.18. (a.18. E = ru · ru . b.18. 2 2 (4. already a part of the terms involved in the computation of the total curvature are expressed in terms of the coefficients of the first fundamental form. starting from the definition of the coefficients of the first fundamental form.4) 0 ruv · ru ruv · rv E F − ruv · ru . G = rv · rv one obtains.18.18. As one can see immediately.18. differentiating with respect to the coordinates. the following expressions:  r · r  2 u  u    r · r  uv u      r · r  2 v  v  r · r  uv v       r 2 · r v  u    r · r  2 u v = 1 Eu 2 = 1 Ev 2 = 1 Gv 2 = 1 Gu 2 = Fu − 1 Ev 2 1 = Fv − 2 Gu . either a product of two second order derivatives.3).5) Differentiating once more the fourth equation above with respect to u and the fifth with respect to v and subtracting them side by side. we get also the expression for the products of second order derivatives of r: 1 1 2 ru2 · rv2 − ruv = − Gu2 + Fuv − Ev2 .18. The first kind of terms are easier to be taken care of.

174

Chapter 4. General theory of surfaces

Combining everything we obtained, we get the following expression (due, as we said before, to Baltzer) for the total curvature
1 − 1 Gu2 + Fuv − 2 Ev2 2 1 Fv − 2 Gu 1 2 Gv 1 2 Eu 1 Fu − 2 Ev

Kt =

1 (EG − F 2 )2

E F

F G

− (4.18.7)

0 − 1 (EG − F 2 )2
1 2 Ev 1 2 Gu

1 2 Ev

1 2 Gu

E F

F , G

which concludes the proof, as we got an expression of Kt which, indeed, depends only on the coefficients of the first fundamental form and their derivatives up to second order. Exercise 4.18.1 (Frobenius). Show that the total curvature of a surface can be written, also, in the following form, easier to remember: E Eu Ev 1 F Fu Fv + Kt = − 4(EG − F 2 )2 G Gu Gv + √ 2 EG − F 2 1 Fv − Gu Fu − Ev ∂ ∂ + √ √ ∂u ∂v EG − F 2 EG − F 2 .

(4.18.8)

In the particular case of an orthogonal coordinate system (F ≡ 0), we get the following nice “divergence” expression for the total curvature: Ev Gu 1 ∂ ∂ Kt = − √ + √ √ ∂u ∂v EG EG 2 EG which will be used later for some integral formulae. Exercise 4.18.2 (Liouville). Prove the following (slightly asymmetric) formula for the total curvature of a surface:       ∂  Gu + F Gv − 2Fv  ∂  Ev − F Gu    1      G G + √  .   (4.18.10) Kt = − √      √   ∂v     ∂u  2  2 2  2 EG − F EG − F EG − F  , (4.18.9)

4.19. Geodesics

175

4.19

Geodesics

4.19.1 Introduction
The curves we are going to study in this chapter are direct generalizations of the straight lines. More specifically, they are curves that projects on the tangent planes of the surface as straight lines. There are many different approaches to geodesics. We chose here the one we find more elementary.

4.19.2 The Darboux frame. The geodesic curvature and geodesic torsion
Let (I, ρ) be a parameterized curve whose support lies on a surface S and let M = ρ(t0 ) be a point of the curve, with t0 ∈ I. As ρ is, in particular, a space curve, we can attach to it the Frenet frame at the point M, {M; τ, ν, β}. As we saw in the first part of the book, this frame is good enough if we want to investigate the curve ρ as an independent object, but it is not very helpful if one wishes to study the connections between the curve and the surface. To this end, we shall introduce another orthonormal frame, which involves both vectors related to the curve and to the surface. The first vector of the new frame will be still the unit tangent vector of the curve, τ. The second, related, this time to the surface, is the unit normal of the surface, n. The third one, let it be denoted by N, will be chosen in such a way that the basis {τ, N, n} be direct, or in other words, such that (τ, N, n) = 1. This means, of course, that N = n × τ. Clearly, N lies in the normal plane of the curve at M, therefore it will be called the unit tangential normal vector of the curve. The name tangential comes, of course from the fact that N lies, also, in the tangent plane of the surface at M. The frame {M; τ, N, n} is called the Darboux frame or the Ribaucour-Darboux frame of the surface S along the curve ρ. The next step we are going to make is to compute the derivatives of the vectors of the Darboux frame and to obtain a set of linear differential equation which is similar to the Frenet frame and which will play an important role in the following sections. To get them, the intention is exactly to use the Frenet equations. Therefore, we shall start by expressing the vectors N and n in terms of the vectors of the Frenet frame. We denote by θ the angle between the vectors ν and n. Then, as one can see immediately that  ν = cos(N, ν) · N + sin(N, ν)n    β = cos(N, β) · N + sin(N, β)n 

176 As (N, ν) =
π 2

Chapter 4. General theory of surfaces and (N, β) = π − θ, we get  ν = sin θN + cos θn    β = − cos θN + sin θn  .

Conversely, we get  N = sin θ · ν − cos θ · β    n = cos θ · ν + sin θ · β  .

Now, the derivatives of the vectors of the Darboux frame with respect to the arclength of the curve can be expressed in terms of these vectors as  τ = a(s) · N + b(s)n      N = c(s)τ + d(s) · n     n = e(s) · τ + f (s) · N,

(4.19.1)

where a, b, c, d, e, f are smooth functions of the arclength. We remind that the derivative of a vector of the Darboux frame is perpendicular on that vector, because the frame is orthonormal. For the same reason, we deduce immediately that the six coefficients are not independent and, in fact, we have the relations: c = −a, e = −b, f = −d, therefore the system (4.19.1) becomes  τ = a(s) · N + b(s)n       N = −a(s)τ + d(s) · n    n = −b(s) · τ − d(s) · N,

(4.19.2)

We shall express now the quantities a, b, d in terms of the characteristics of the curve and the angle θ. We notice, first of all, that, from the first of the Frenet’s formula, we get τ = kν = k(sin θ · N + cos θ · n), hence, identifying the coefficients with those of the first equation of the system (4.19.2), we obtain a = k · sin θ; b = k · cos θ. (4.19.3) The quantity k · cos θ is already known to us: it is nothing but the normal curvature kn of the surface that was studied previously. The function k · sin θ, instead, is new. We

4.19. Geodesics

177

will denote it by kg and we will call it geodesic or tangential curvature of the curve7 . To find the quantity d, we start from the relation N = sin θ · ν − cos θ · β. By differentiating with respect to the arclength of the curve, we get, using the last two of the Frenet’s formulae: N = θ · sin θ · ν − sin θ(−k · τ + χ · β) + θ sin θ · β + χ · cos θ · ν = = −k · sin θ · τ + (θ + χ) · (cos θ · ν + sin θ · β) = = −k · sin θ · τ + (θ + χ) · n, and, comparing with the second equation from (4.19.2), we get d = θ + χ. This function is denoted by χg and it s called the geodesic torsion. His geometric meaning will be made clear later. Clearly, we cannot claim, as we did with the geodesic curvature, that the geodesic torsion is the torsion of the projection of the curve on the tangent plane, as the torsion of that curve is always zero, while the geodesic torsion of the given curve is not, usually. The geodesic curvature plays a much more important role in the differential geometry of surfaces than the geodesic torsion does, so we start by focusing on it. We notice, to begin with, that kg = τ · N = −τ · N . (4.19.4) Since, as we saw earlier, N = n × τ, one obtains for kg the expression kg = τ · N = τ · (n × τ), i.e. kg = (τ, τ , n). (4.19.5) This formula holds for naturally parameterized curves. Let us consider, now, an arbitrary regular parameterized curve on S , given by the local equations u = u(t), v = v(t). We have, therefore, r = r(u(t), v(t)). If we denote by a dot the differentiation with respect to the parameter t along the curve, we get τ≡
7

dr dr dt 1 ˙ = = · r, ds dt ds s ˙

Here the term “tangential” refers to the tangent plane of the surface, not to the tangent line of the curve. In fact, it can be shown that the geodesic curvature at a point of a curve lying on a surface is the signed curvature of the projection of the curve on the tangent plane to the surface at that particular point.

2. with the newly introduced notations:  r ≡ dr = r ui ˙  i˙ dt  r = r ui u j + r uk . we shall use. when the curve is given by its local parametric equations with respect to a local parameterization of the surface. We shall discover. in fact. n = 4 (˙ . ¨ ij˙ ˙ k¨ The decomposition of the second order partial derivatives of the radius vectors in terms of the basis {r1 . 3 (¨ · s − r s). that the geodesic curvature can be expressed in terms of the coefficients of the first fundamental form and they first order partial derivatives with respect to the coordinates. r2 . Moreover. ds s ˙ Thus. a regular parameterized surface. for the geodesic curvature one gets τ = kg = (τ. then one makes summation after all the allowed values of the index (in our case. for a while. i. once in an inferior position and once in a superior one. n) = i. the next step will be to obtain a formula for the geodesic curvature in terms of the coefficients of the fundamental forms. In other words. more generally. To simplify the notations.e. n} is already known to us: it can be described using the Christoffel’s coefficients and the second fundamental form as ri j = Γkj rk + hi j · n. n). we shall denote the coordinates with u1 and u2 . r.178 therefore Chapter 4.6) s3 ˙ The formulas we got so far for the geodesic curvature use mixture of information about the curve and about the surface. i . while the first partial derivatives of the radius vector r with respect to the coordinates will be denoted by ri and those of second order by ri j . τ . Thus. Going back to our problem. we shall use the Einstein’s summation convention: every time an index enter twice in a monomial. 1 1 1 1 ˙ r ˙ ˙¨ r ¨ ˙ ˙¨ r ˙ ¨ r. but they don’t use explicitly the things we usually compute when we are given a local parameterization of a surface or. r ¨ (4.19. s · r. the index notations. r · s − r s. n). s s ˙ ˙ s ˙ s ˙ kg = 1 (˙ . instead of u and v. namely the coefficients of the first two fundamental forms. an expression of the form ai ui should be read as ai ui = a1 u1 + a2 u2 . we have. General theory of surfaces dτ 1 r ˙ ˙¨ = 3 (¨ · s − r s). n) = 4 (˙ . So. 1 and 2). j = 1.

4. r2 . we get √ kg = g (u1 ) (u2 ) + Γ2j (ui ) (u j ) − (u2 ) (u1 ) + Γ1j (ui ) (u j ) . ˙ ¨ ˙ ¨ i ˙ ˙ i ˙ ˙ s ˙ But r × r2 √ = r1 × r2 = g. r) · n. in particular. i i (4. Remark. is nothing but its signed curvature: kg = u v − u v ≡ k± . in fact. whence √ g ˙ ¨ ˙ ¨ kg = 3 u1 u2 + Γ2j ui u j − u2 u1 + Γ1j ui u j . a plane curve). as we know. It can be shown that. the formula for the geodesic curvature of a naturally parameterized curve is kg = EG − F 2 Γ2 u 3 + 2Γ2 − Γ1 u 2 v + Γ2 − 2Γ1 u v 2 − 11 12 11 22 12 −Γ1 v 3 + u v − u v .9) If. r). the surface S is the plane. . then the determinant of the first fundamental form is. r2 .10) (4.19. we can write now 1 1 r ¨ kg = 3 (˙ . n = ˙ ¨ r ˙ i ˙ ˙ s ˙ s ˙ 1 = 3 u1 u2 + Γ2j ui u j − u2 u1 + Γ1j ui u j r1 .19.8) In the traditional notations.7) (4. r1 . Thus. we have ¨ r = Γkj rk + hi j · n ui u j + rk uk = uk + Γkj ui u j rk + hi j ui u j n ˙ ˙ ¨ ¨ ˙ ˙ i i ˙ ˙ or ¨ r = uk + Γkj ui u j rk + ϕ2 (˙ . in this situation the geodesic curvature of the curve (which is.19.19. the curve is naturally parameterized. with the Cartesian coordinates on it. ¨ r ˙ i ˙ ˙ With these in hand. uk + Γkj ui u j rk + ϕ2 (˙ . while all the Christoffel’s coefficients vanish. i ˙ ˙ i ˙ ˙ s ˙ If. As a consequence. of course. n) = 3 um rm . one. Geodesics 179 where. as the matrix of the fundamental form is the unit matrix. 22 This equation can also be written in a more elegant form as kg = EG − F 2 det u v u + u 2 Γ1 + 2u v Γ1 + v 2 Γ1 11 12 22 .19. r. in this case. in particular. n = r1 × r2 · n = r1 × r2 · 1 r1 × r2 where g is the determinant of the first fundamental form. the geodesic curvature of a curve on a surface is nothing but the signed curvature of the projection of the curve on the tangent plane. n . hi j are the coefficients of the second fundamental form of the surface. v + u 2 Γ2 + 2u v Γ2 + v 2 Γ2 11 12 22 (4.

therefore. in other words.12).12) There is. from the first formula of Frenet. a very delicate business and can be done explicitly only in special situations. as the first claim that the geodesics are solutions of a single equations.11) 11 12 11 22 12 22 Also. or simply. at each point. as we noticed earlier. however. a contradiction between the previous two theorems.19. General theory of surfaces 4. between the functions u and v there is an extra relation. n).19. (4.19.180 Chapter 4. while the second – that they are a solution of a system of different equation. the geodesic curvature of a curve can be written as kg = (τ. we might say that the geodesics are the straightest lines on the surfaces. . apparently. therefore. this property is taken as the definition of the geodesics. Let S be a surface. The existence (locally. The formula 4. kg vanishes at a point of a curve on the surface if and only if the binormal of the curve is perpendicular on the normal of the surface at that point or. because we impose the geodesics to be naturally parameterized. usually.19.19. we can say that the geodesics are those curves which project on each tangent plane of the surface after a straight line. at least) of geodesics passing through a point of a surface and having at that point a given tangent vector is a consequence of standard results in the theory of ordinary differential equations.19.9 leads to Theorem 4. in many books. On the other hand. the geodesic curvature vanishes if and only if the normal of the surface is contained in the osculating plane. The geodesic lines verify the system of differential equations  u + u 2 Γ1 + 2u v Γ1 + v 2 Γ1 = 0   11 12 22  v + u 2 Γ2 + 2u v Γ2 + v 2 Γ2 = 0 .3 Geodesic lines Definition 4. τ × τ = kβ. the geodesic curvature is the signed curvature of the projection of the curve on the tangent plane. τ .5. the two equations of the system (4. is. from (4. a geodesic if.12) are not independent.19. Finding geodesics. Another remark that should be made is that since. According to the formula (4. In this sense. the geodesic lines are exactly those lines on the surface for which the osculating plane at each of their points contain the normal of the surface at that point.1. The differential equation of the geodesic lines is Γ2 u 3 + 2Γ2 − Γ1 u 2 v + Γ2 − 2Γ1 u v 2 − Γ1 v 3 + u v − u v = 0.  11 12 22 (4. In fact. A parameterized curve ρ : I → S is called a geodesic line. Thus.19. In fact.6. its geodesic curvature vanishes.10) we can deduce that Theorem 4. Remark.

we deduce immediately that the geodesics of the plane are the straight lines and only them.19. Geodesics Examples of geodesics 181 The geodesics of the plane. ¨ ˙ ds dϕ dϕ dϕ dϕ Therefore.. by using cartesian coordinates.  ϕ + 2 cot θθϕ = 0 ¨ ˙˙ We assume that the explicit equation of the geodesic is θ = θ(ϕ).  which lead to u(s) = a1 s + b1 . θ= ˙ ds dϕ d dθ dθ dθ d2 θ ¨ θ= ·ϕ+ ˙ ϕ= ¨ · ϕ + 2 · ϕ2 . Also. Then ˙ dθ = dθ · ϕ. therefore the osculating planes should pass. By the geometrical interpretation of the geodesics. ˙ dϕ . The geodesics of the sphere can be found very easily from their interpretation as being those curves for which the osculating plane contains the normal to the surface. ˙ ˙ dϕ dϕ2 On the other hand. in the case of the sphere all the normals are passing through the center of the sphere. using a standard parameterization of the sphere (with spherical coordinates). ˙ ˙ ϕ = −2 cot θθ · ϕ = −2 cot θ ¨ dθ 2 ·ϕ . we can find (do that!) the equations of the geodesics as:  θ − sin θ cos θϕ2 = 0 ¨ ˙  . As we know.4. the first equation of the system becomes: ϕ· ¨ dθ d2 θ 2 + · ϕ − sin θ cos θ · ϕ2 = 0. i. all of them through the center. again. we fined immediately that the Christoffel’s coefficients vanish identically.e. v(s) = a2 s + b2 . The geodesics of the sphere. On the other hand. as the curves that project on the tangent plane on straight lines. from the first equation. therefore the equations of geodesics become  u = 0    v = 0 . the geodesics are straight lines. which leads immediately to the idea that the geodesics are arcs of great circles of the sphere.

182 hence: Chapter 4.19. it is an easy matter to prove that the equation of the geodesics for the metric (4. and. ˙ obviously. ϕ = const. a great circle of the sphere (a meridian). which is the equation of a great circle. The most important characteristic of Liouville surfaces is the fact that their geodesics an be found through quadratures. General theory of surfaces d2 θ dθ − sin θ cos θ = 0. Application.13) can be written as (U v − V u )(u 2 + v 2 ) + 2(U + V)(u v − v u ) = 0.13) . This is exactly what we are going to prove in the rest of this section. (4. 2 dϕ dϕ In this case. A surface is called a Liouville surface if it can be parameterized locally in such a way that the first fundamental form can be written as ds2 = (U(u) + V(v))(du2 + dv2 ). we get d2 z + z = 0. dϕ2 and this equation has the general solution z = cot θ = A cos ϕ + B sin ϕ or A sin θ cos ϕ + B sin θ sin ϕ − cos θ = 0. the curve is.4 Liouville surfaces Definition 4. First of all. the surfaces of revolution are examples of Liouville surfaces. either d2 θ dθ − 2 cot θ · − sin θ cos θ = 0.e.19. i. where U and V are smooth functions of a single variable. −1).14) (4. we make the substitution z = cot θ and. In particular. in this case. lying in the plane passing through the origin and having as normal vector the vector (A. These surfaces have been introduced by the French mathematician Joseph Liouville in the Note III of the 5th edition of the book of Gaspard Monge.2.19.19. B. − 2 cot θ · 2 dϕ dϕ ϕ2 ˙ We have two possibilities: either ϕ = 0. 4. after a straightforward computation.19.

ρ and α. Therefore. dt whence U sin2 α − V cos2 α = a. the equation of geodesics can be rewritten as u dU v dV u v −v u − + 2(U + V) 2 = 0.19.19.19.  In the sequel. whence du =± √ U −a where b is another integration constant.19. defined by  u = ρ cos α   (4. dv + b. through the coordinate functions. the equation of the geodesics becomes sin2 α dU dV dα − cos2 α + 2(U + V) sin α cos α = 0. for the functions U and V the prime denotes the derivative with respect to u and v. where a is a constant. while for the coordinate functions u and v the prime denotes the derivative with respect to the parameter t along the geodesic.4.15) v dt u dt u +v 2 We see that this equation actually does not involve the coordinate functions u and v.16)  v = ρ sin α . (4. we get v 2 U − u 2 V = a(u 2 + v 2 ).17) . We substitute them with other two functions of t. dt dt dt This equation can be also written as d U sin2 α − V cos2 α = 0. functions of t. Returning to the old coordinate functions. both. Geodesics 183 Here. Of course. but only their derivatives. √ V +a (4. respectively. U and V are.

184 Chapter 4. General theory of surfaces .

As we shall see shortly. We recall that a surface is called ruled if it is generated by a straight line (the ruling. the directrix can be replaced by another curve . Clearly. called the directrix. the ruled surfaces are the surfaces generated by a moving straight line. the directrix is an orthogonal trajectory of the rulings. orthogonal to the ruling passing through that point (in other words. which is subject to a further condition. the simplest surfaces.1 General ruled surfaces After the plane. in the case of which the rulings are passing through a fixed point.1. usually this condition being that the moving line remains tangent to a given surface or intersects a given regular space curve. by sectioning the surface by a plane or a sphere. at each point. the ruled surfaces are. no doubt. it is. once the ruled surface was defined. More precisely 5. obtained. usually.CHAPTER 5 Special classes of surfaces 5.1 Ruled surfaces Intuitively speaking. moving in space. The simplest ruled surfaces are the cylindrical surfaces. this particular choice of the ruling has as effect the diagonalization of the first fundamental form of the surface. convenient to choose a directrix which is. for which the rulings are always parallel to a fixed direction and the conical surfaces. for instance. lying all the time on a given curve. In particular. The ruled surfaces have been studied in details .

1) .1. where I is an interval on the real axis. we denote by b(u) the versor of the ruling passing through the point ρ(u). u ∈ I.1. Special classes of surfaces Figure 5. its coordinates will be determined by the relation r = r(u. (5.1: The hyperboloid with one sheet and its rulings for the first time by the French geometer Gaspard Monge. Then.e.1 we represented the surface described by the binormals of the Viviani’s temple. For instance. Another way of getting interesting ruled surfaces is to take as rulings the axes of the Frenet frame of a space curve.1 we represented the hyperboloid of rotation with one sheet. in the figure 5. In the figure 5. at the end of the XVIIIth century and most of the important results in the field belong to him or to some of his students. v) = ρ(u) + vb(u).186 Chapter 5. i. if M is a point on this ruling. together some of its rulings. then we have −−→ −− M0 M = vb(u). For each u ∈ I. The parameterization of a ruled surface We assume that the directrix of the surface has a parameterization of the form ρ = ρ(u). where v is the parameter along the ruling.1. if M0 = ρ(u).

The tangent plane and the first fundamental form of a ruled surface To compute the coefficients of the first fundamental form of a ruled surface we need.2: The surface of binormals of the Viviani’s temple The relation 5. and the directrix is. global. obviously. usually.2) . rv = b u . We have. when we measure the distance along the ruling passing through each point. Ruled surfaces 187 Figure 5. first of all.5. Generally.1 provides a parameterization of the ruled surface. as the parameterization of the directrix is not global. (5. This parameterization is not. in the sense that all the points from such a coordinate curve lie at the same distance (equal to |v|) from the directrix.1. equally. the rulings will be coordinate lines (u = const).1. With respect to this parameterization. r : I × R → S. the partial derivatives of the radius vector of a point of the surface. a coordinate line (v = 0).1. the coordinate lines v = const have the property that they are “parallel” to the directrix. ru = ρ + bb .

(5.6) A characteristic property of the ruled surfaces is described by the following proposition: Proposition 5. It follows that the first fundamental form of a ruled surface can be written as: ds2 = ρ 2 + 2vρ · b + v2 b 2 du2 + 2(ρ · b) dudv + dv2 . i. R×ρ +v R×b −ρ×ρ −v ρ×b · b = 0. also. belong to the pencil of planes determined by that ruling.4) (5. the tangent plane at a point of a ruled surface contains the ruling passing through that point.3) To find the tangent plane at a point of a ruled surface.1. v). ρ .1. to the surface) at a given point is given by the vector ru × rv . if R is the position vector of a point from the tangent plane to the surface at a point corresponding to the pair of parameters(u.e. first of all. by the vector N ≡ ru × rv = ρ × b + v b × b . or. we notice. b − ρ. b . b + v R. (5. Special classes of surfaces Thus.5) Therefore.1. then the equation of the tangent plane can be written under the form (R − r) · N = 0. to put it another way. b . hence. that the direction of the normal to the plane (and. b = 0 or.188 Chapter 5. F ≡ ru · rv = ρ · b. i. The tangent planes to a ruled surface in point located along the same ruling. .e. the coefficients of the first fundamental form of the surface will be E ≡ ru · ru = ρ 2 + 2vρ · b + v2 b 2 . (R − ρ − vb) · ρ × b + v b × b = 0 or R.1.1. (5. ρ . G ≡ rv · ru = 1. b − v ρ.1.

which is the same. rv .1.5. ruv = ρ + vb . ρ + vb H H . from the parameterization (5. while G = rv · rv = 1. its normal vector stays all the time parallel to a fixed plane. Since the ruling and the tangent plane already have a point in common (the very point of tangency). Ruled surfaces 189 Proof. H u v v Thus. rv . r 2 = 0.2 The Gaussian curvature of a ruled surface We start. b + v b . Therefore. rv = b.1. indeed. N · b = ρ × b + v b × b · b = ρ . we will get E = ru · ru = (ρ + vb ) · (ρ + vb ) = ρ 2 + 2vρ · b + v2 b 2 . Then we have ru = ρ + vb . Let H be the determinant of the first fundamental form. r . In other words.1. that it is perpendicular to the normal to the surface at the tangency point. Then. which is perpendicular to the considered ruling. of course. b. The proposition we just proved indicates how varies the tangent plane to a ruled surface along a ruling: it just rotates around the ruling. 5. for the coefficients of the second fundamental form.1). b. b. b = ρ . . ru2 = ρ + vb . we will get D= D = 1 1 ru . b. for the first fundamental form. b. 1 1 1 ru . b = 0. since b is a versor. We have. F = ru · rv = (ρ + vb ) · b = ρ · b. b . it is enough to prove that the ruling is parallel to the tangent plane or. H H H 1 D = r .

e. (5. Special classes of surfaces Proposition 5.3 Envelope of a family of surfaces In this section.3. EG − F 2 (ρ + vb ) × b 2 (5. The contact between the envelope and each surface is made along a curve which is called a characteristic. then to get the envelope. Let r = r(u. b.7) The curvature is always negative and it vanishes if and only if ρ . λ) = 0.1. v.11) as we did in the case of plane curves. we have to add to this equation the equation Fλ (x. y. beside this equation. The theory of envelopes of surfaces is analogue to the theory of envelopes of plane curves.1. (5. therefore there will be no proofs. Something that is specific to the theory of envelopes of surfaces is the so-called regression edge. Let us assume that we have a family of . b )2 DD − D 2 =− .1.1. The envelope of the family (if there is any) is a surface which is tangent to all the surfaces from the family. i.1. Thus. these equations are also verified by the singular points of the surfaces and to get the envelope.1. as the proofs are completely analogue to the ones we provided for plane curves.1. b.1.9) be a family of surfaces. The total curvature of a ruled surface given by the local parameterization (5.9) verify.1. as in the case of plane curves.10) Of course. the equation (ru . we have to eliminate them first.190 Chapter 5. b = 0.1.1. a surface will always be a parameterized surface. y.2. the envelope is the geometrical locus of the chracteristics of the family of surfaces. λ) (5.1. If the family of surfaces is given implicitly. Definition 5. Proposition 5. The points of the envelope of the family (5. (5. through the equation F(x. rλ ) = 0.12) (5.1) is given by Kt = (ρ . rv .8) 5. λ) = 0. z.1. z.

λ) = 0. given. Figure 5. Ruled surfaces surfaces. it is called the regression edge of the envelope of the family of surfaces. verifying the hypothesis that the radius of the circle is greater than the radius of the spheres. as one can see immediately. the socalled developable surfaces: Definition 5.1. the torus (see the figure 5. λ) = 0 λ2 191 . Then the envelope of this family is. at each point. in other words.5. We have the following result: .4 Developable surfaces There is a subclass of ruled playing an important role in differential geometry. An example of envelope.1.      Fλ (x. y.1. If the system is compatible. implicitly. then what we find solving it is a curve which. with the center of a given circle.3: The torus as envelope of spheres 5. We would like to find conditions for a ruled surface to be developable. say.2. (5.13) The first two equations of this system are the equations of the family of characteristic lines of the family of surfaces. z. Then we can form the system:  F(x.1. y. y.1. z.3). is tangent to one of the characteristics. λ) = 0     F (x. A ruled surface is called developable if the tangent plane to the surface is the same at all the points of a ruling. it is the envelope of the characteristics. This envelope (when exists). We consider the family of spheres of constant radius. z.

ρ (u) × b(u) = 0. Special classes of surfaces Proposition 5. the rulings are nothing but the tangents to the directrix.192 Chapter 5. obviously. In this case. (5. A ruled surface given by the equation r(u. the vector N is constant only as direction. The two components of the normal vector are parallel. From this relation follows.e. (b) ρ b. b. therefore the surface is conical. the condition for the surface to be developable is that the direction N be independent on the coordinate v. This may happen. the surface is generated by the tangents to a space curve.14) Proof. i. i. the directrix degenerates to a point. The second component of the normal vector vanishes. In this case.1. but its norm and orientation might vary. the surface is cylindrical.e. As b is a versor.e. it cannot vanish.4.e. v) = ρ(u) + v · b(u) is developable if and only if ρ . i.1. .e. we have ρ (u) × b(u) b (u) × b(u). therefore the condition mentioned above can be fulfilled in two situations: (a) ρ = 0. The first component of the vector N vanishes. that the direction of b is fixed. i. v) = ρ (u) × b(u) + v b (u) × b(u) . 3. In this case. As we saw previously. 2. Thus. a normal vector to the surface is the vector N(u.in three situations: 1. b (u) × b(u) = 0. in fact. The invariance condition for the tangent plane along a ruling is equivalent to the condition of invariance of the normal line to the surface along the same ruling. i. b = 0.

it depends on a single parameter. on two parameters. that a ruled surface is developable if and only if it is the envelope of a family of planes. 2. In the figure 5. A ruled surface is developable if and only if its Gaussian curvature vanishes identically.15) These planes are. .4 we give an example of developable surface with a regression edge (the tangent developable of the Viviani’s temple). Developable surfaces as envelopes of a one-parameter family of planes. Clearly. this family depends. ρ × b × b × b = ρ . Nevertheless. 3. depending on a single parameter1 . that.1. surfaces generated by the tangents to a space curve. in fact. namely the parameter along the directrix of the surface. any surface is the envelope of a family of planes. b b. b. in this paragraph. conical surfaces. the tangent planes of the surface. We shall see. =0 193 which concludes the proof of the proposition. We consider a one-parameter family of planes N(λ) · r + D(λ) = 0.1.1. b ρ = − ρ . One can notice immediately. 1 (5. b .1. Ruled surfaces The three conditions we just listed are equivalent to the unique condition ρ (u) × b(u) × b (u) × b(u) = 0.5. In the case of developable surfaces. namely the family of tangent planes to the surface. On the other hand.1. Corollary 5. The ruled surfaces which are not developable are also called scrolls. usually. Remark. b . these three classes exhaust all the developable surfaces. of course. The regression edge of a developable surface We saw previously that that there are three classes of developable surfaces. cylindrical surfaces. from the very definition of the developable ruled surfaces. namely: 1. b b − b.

the equation obtained from (5. that the normal vector N is a versor. respectively. However. thus.1. N (λ) · r + D (λ) = 0.1. these equations describe the discriminant set.15) and (5.194 Chapter 5. we obtain N (λ) · r + D (λ) = 0.1. the surfaces are planes and they have no singular points whatsoever. also.16) According to the theory of envelopes of the families of surfaces. (5. the equations (5. which we admit to be true in our case.16) determine the envelope of the family of planes. if such an envelope exists2 . these characteristics exist only if the planes of the family are not parallel. in this particular case.1.1. The characteristic curves of the family of planes are straight lines. including. Clearly. The points of the envelope have to verify. Special classes of surfaces Figure 5. i. by differentiating once more with respect to λ. obtained as intersections between the planes of equations (5. By differentiating the equation (5.17) (5. We have. .1. 2 (5.15) and (5.4: The tangent developable of Viviani’s temple We shall assume.15) with respect to λ.e.1.1. the singular points of the surfaces from the family.16).18) As a matter of fact. without restricting the generality. also.1. the following system of equations:  N(λ) · r + D(λ) = 0      N (λ) · r + D (λ) = 0     N (λ) · r + D (λ) = 0.16).

Ruled surfaces 195 We can regard this system as being a system of three equations with three unknowns (the components of the radius vector r). we must have ˜ N·r =0 (5. we obtain ˜ ˜ N ·r +N·r =0 whence. is nothing but the osculating plane of the regression edge at the point associated to the value λ of the parameter. if ˜ r = r(λ) (5.5. Let us assume. This means. to begin with. then r(λ) 0. Therefore.1. i.21) because the characteristic is obtained by the intersections of the planes of normal vectors N and N . i.22) .e.20) ži ˜ N · r = 0. (5. the edge of regression of the envelope of the family of planes. using (5. that the equation (5. of course. As such. in the sense of the theory of envelopes of surfaces.19) describes a space curve.19) ˜ is the solution of the system. b) as the envelope of the family of osculating planes of the edge of regression. We admit. that the system is compatible and determined.23) The relations (5.1.1. This is. called the edge of regression of the developable surface.e.21). a developable surface admitting an edge of regression can be described in two different ways: a) as the surface generated by the tangents to the edge of regression. the three planes intersect at a single point. that the solution of the system depends in a nontrivial manner.1.. In the degenerate case.20) and (5. hence the surface is a conical surface. ˜ N · r = 0. the generators of the surface) are passing through that point. By differentiating the relation (5.1. respectively.23) tell us that the plane of the normal vector N contains ˜ ˜ both the vector r .1. and the vector r . (5. at the first stage.1.1. on the parameter λ. all the characteristic curves (i.1. when the edge of regression reduces to a point. We know from this theory that all the characteristic curves have to be tangent to the edge of regression.20). in other words.e. in fact. (5.1.1.

The considerations from this paragraph prove that.18) has a vanishing determinant. (5. We mention that the plane can be considered as a limit case for any of the three classes of surfaces listed in the proposition. N ). It follows from here that for any λ the vectors N(λ) and N (λ) are perpendiculars on the constant vector n. i.1. in practice. n · N = n · N = 0.1. either it is incompatible. In principle.1. of course.1. excluded. i. that the determinant of the linear system (5. using (5. N = 0.e.1. we would obtain.e. either it is perpendicular on the vectors N ži N . we have n · N = n · N = n · N = 0. (5. that the vectors N. we have n · n = 0. N lie in the same plane. the vector n is constant.1. in other words. as one can see easily.24) This means. n is a versor. It is easy to see that in this case the system of equations (5. as the vectors N and N are not colinear. n = 0.e. either a conical one. as. If S is a developable ruled surface. the classes of surfaces mentioned in the previous paragraph do exhaust all the types of developable ruled surfaces. i. N . we have the following result:: Proposition 5. Assuming that we would be in the second situation. By differentiating the first two equations from the previous system.5. which is not possible since. the condition ∆ = 0 is equivalent to the condition N. We shall see that the first possibility is. We must have. all the rulings of the family are parallel to a constant direction. therefore.18) is nothing but (N. whence.1. either the vector n vanishes. we obtain n · N + n · N = n · N + n · N = 0. either a surface generated by the tangents to a space curve. Thus.18) is incompatible (the third plane is parallel to the straight line determined by the first two). two situations are possible: either the system is compatible and undetermined. We denote by n = n(λ) the normal versor of this plane.25) . that n is parallel to n.196 Chapter 5. first of all. Special classes of surfaces We shall focus now on the case when the system (5. in fact. and the surface is a cylindrical surface. N .25). therefore. N . then it is either a cylindrical surface. We notice.

1. it is nothing but the tangent to the curve . the edge of regression is nothing but the given curve.1. its torsion never vanishes. To get the edge of regression.28) . the torsion of the curve never vanishes.1. (R − r(s)) · ν(s) = 0. we get −r (s) · β(s) + (R − r(s)) · β (s) = 0 or. in other words.26) with respect to s. in other words.a s well as the second Frenet formula.27) where we used the fact that. according to the hypotheses we made about the curve. As χ(s)β(s) − kν(s) 0 if χ and k don’t vanish. By differentiating the relation (5. we differentiate once more (5.27) one obtains that the characteristic straight line is perpendicular both on β and on ν.5.1.1. We assume.e. Ruled surfaces 197 5.1. (5.26) and (5. The envelope of the family of osculating planes The equation of osculating planes at a point of the curve is (R − r(s)) · β(s) = 0.1. moreover. (5. we obtain that R = r(s). using the fact that r (s) · ν(s) = 0.1.26) The parameter along the family of planes will be exactly the natural parameter s along the curve. i.5 Developable surfaces associated to the Frenet frame of a space curve In all of this section r = r(s) will be a smooth naturally parameterized biregular space curve.1.27) and we get −r (s) · ν(s) + (R − r(s)) · ν (s) = 0 or. having in mind that r · β = 0 and using the third formula of Frenet. (R − r(s)) · χ(s)β(s) − kν(s) = 0. that the curve is skew. From (5. (5.

(R − r(s)) · τ(s) = 0.31) and the second Frenet formula.32). −1 + (R − r(s)) · k(s) · ν(s) = 0. On the other hand.1. we find out the following things: (R − r(s)) · β(s) = k2 (s) (5.30) and we get (R − r(s)) · ν(s) = − r (s) · k(s) · ν(s) + (R − r(s)) · k (s) · ν(s) + (R − r(s)) · k(s) · ν (s) = 0 =0 (5.29) to (5. that 1 .29). k (s) + (R − r(s)) · k(s) · χ(s) · β(s) − k(s)τ(s) = 0 k(s) or.1. having in mind that we have a curve parameterized by the arc length and using the first of the Frenet formulae.1.1.e. Special classes of surfaces The envelope of the family of normal planes (the polar surface) The equation of the normal plane at an arbitrary point of a space curve is. the polar surface of the curve. Thus.29) We differentiate this relation once. using (5. (5.31) k(s) The equation (5. (5. also. i.32) . the characteristic of the envelope of the normal planes passing through the point of position vector r(s) is obtained by intersecting the normal plane to the curve at this point with a plane which is parallel to the rectifying plane of the curve at the same point.1. k (s) . the characteristics of the family of normal planes are nothing but the axes of curvature (or polar axes) of the curve. To get the edge of regression of the surface.1. as we know. It follows.1.31) is the equation of a plane which is parallel to the rectifying plane (because it has as normal vector the versor of the principal normal). which is exactly the center of curvature of the curve.1.1. then. This is the reason why the envelope of the family of normal planes is called.30) or. it is a straight line which is parallel to the binormal of the curve at the point r(s). we differentiate once more the relation (5. Thus.198 Chapter 5. · χ(s) Comparing (5.1. it is clear that this characteristic passes through the point of 1 position vector r(s)+ k(s) ·ν(s). sometimes. from (5. to complete the system of equations for the characteristics and we get −r (s) · τ(s) + (R − r(s)) · τ (s) = 0 or.

e.34) This point will have.1. (5. i.1. thus. we get − r (s) · ν(s) + (R − r(s)) · ν (s) = 0 =0 (5.1. is parallel to the binormal.33) (5. using the second Frenet equation.1. the edge of regression of the polar surface of a space curve is the locus of the centers of the osculating spheres of the given curve. regarded as . (5.1.1.36) or.5. (R − r(s)) · χ(s)β(s) − k(s)τ(s) = 0. Ruled surfaces 199 • The straight line of intersection between the planes (5.1. Thus.37) The points of the curve verify the equations (5. the position vector R(s) = r(s) + 1 k (s) · β(s).35) is nothing but the position vector of the osculating sphere of the given curve. · χ(s) Thus.1. the current point from the edge of regression is obtained by intersecting the lines R(λ) = r(s) + and R(µ) = r(s) + 1 · ν(s) + λβ(s) k(s) k (s) k2 (s)χ(s) · β(s) + µν(s). · ν(s) + 2 k(s) k (s)χ(s) (5.35) But the right hand side of the relation (5.29) and (5. By differentiation.1.36) and (5. hence the curve lies on the rectifying surface and can be takes as the directrix of the surface.37). The envelope of the family of rectifying planes of a space curves The equation of the rectifying plane is (R − r(s)) · ν(s) = 0. • This straight line passes through the point r(s) + k2 (s) k (s) · β(s).32) is parallel to the osculating plane to the rectifying plane and the normal plane.1.1.

r = τ. when regarded as lying on its rectifying surface. However. As a matter of fact. the curve is rectified (straightened) by its rectifying planes. any surface of zero Gaussian curvature is a developable surface or a part of a developable surface. The relations (5.e. otherwise.37) indicate that the rulings (i. and the Darboux vector of the curve. The surface is not. true. it might not be not even a ruled surface. This is. Thus.2. δ. and on ν . Thus. the characteristics of the rectifying surface of a space curve are the Darboux axes of the curve or the instantaneous axes of rotation of the Frenet frame.36) and (5. the curve is a geodesic of its rectifying surface. also.1. where δ is the Darboux vector of the curve. which is the surface analogue of a straight line.e. Lagrange extended to the two-dimensional case the method of Euler for finding the extrema of a functional and he proposed the following problem: . The tangent plane to the rectifying surface at a point lying on the given curve contain both the tangent vector of the curve. An example of such surface is given in the book of Klingenberg [31]. a surface whose mean curvature vanishes identically.1. In many books of differential geometry it is claimed that the converse is. implicitly. of the rectifying surface: the curve. the origin of the name of the rectifying plane and. hence their direction is given by the vector ν × ν = ν × (χβ − kτ) = χ(ν × β) − k(ν × τ) = χτ + kβ = δ. This is not the original definition of a minimal surface. by definition. This actually means that the osculating plane of the curve is perpendicular on the tangent space of the surface. described by using a parameterization or an implicit equation. In other words. 5.1 Minimal surfaces Definition and general properties Definition 5. first of all. A minimal surface is. i. actually. As we saw. Special classes of surfaces a ruled surface. the total of a developable surface is always zero. the characteristic lines of the envelope) are perpendicular both on ν.2.2 5. but rather by prescribing the first two fundamental forms and showing that they fulfill the compatibility conditions. this is true only if we assume that the surface doesn’t have any planar points. actually. the normal to the rectifying surface of a space curve at a point of the curve coincides to the principal normal of the curve.1. Therefore. is a geodesic. in other words. the normal to the surface at such a point is colinear to the vector τ × δ = τ × (χτ + kβ) = k(τ × β) = −kν.200 Chapter 5.

Let us mention. to finde the surface such that the inclosed area shall be minimum. a necessary condition for the minimum of this functional is furnished by the Lagrange equations. thus. the area functional. also. in which case the area functional is 1 + p2 + q2 dxdy. the Lagrange equations become  y z  u u      y z Km = 0  v v      x z  u u    x z Km = 0 . U while the stationarity condition finally becomes ∂ ∂x or (1 + q2 )r − 2pqs + (1 + p2 )t = 0. As it is known from the variational calculus.     ∂H  − ∂ ∂H − ∂ ∂H = 0 ∂z ∂u ∂zu ∂v ∂zv After long. are:  ∂H  − ∂ ∂H − ∂ ∂H = 0  ∂x ∂u ∂xu  ∂v ∂xv   ∂H  ∂ ∂H ∂ ∂H  ∂y − ∂u ∂yu − ∂v ∂yv = 0 . Minimal surfaces 201 Given a closed curve and a connected surface patch S bounded by the curve. that the minimality condition induces a relation among the coefficients of the first two fundamental forms of a parameterized p 1 + p2 + q2 + ∂ ∂y p 1 + p2 + q2 =0 . initially.2. y) for the surface.   v v     x y  u u      x y Km = 0  v v and. Lagrange used the explicit representation z = f (x. they are satisfied iff Km = 0. It was Meusnier who showed. which. that. A= U Hdudv. in this case. We consider. but otherwise straightforward computations. thus.5. sixteen years later.

F = 0) and E = G.202 Chapter 5. with respect to which we have E = G = λ2 (u.2. A local parameterization (U. the coordinate lines are orthogonal (i. r) of a surface S is called isothermic if. the equation for the slopes of the asymptotic directions is Dt2 + 2D t + D = 0 and one can see immediately that the asymptotic directions are orthogonal iff the roots of this equation verify t1 t2 = −1. actually.e. An alternative proof of the fact that the minimal surfaces are the solution of the variational problem was found by Darboux. which is equivalent to D = −D . the surface being regular. On any surface which is at least C 2 there exist isothermic parameterizations. by Sophie Germain. On the other hand. For further convenience. This is not quite correct: Meusnier was not aware of the true meaning of the the relation he found. that the matrix of the first fundamental form is just the unit matrix multiplied by a positive function. Let us choose. Proof. we shall need the following definition: Definition 5. F = 0. v). with respect to this parameterization. the notion of mean curvature was introduced only fifty years later. A surface is minimal iff the coefficients of the first two fundamental forms verify the equation ED − 2FD + GD = 0. an isothermic parameterization. The proof of this claim is based on the theory of partial differential equations and it is not interesting for our exposition.2. A particularly clear exposition of this proof can be found in the book of Hsiung. Special classes of surfaces surface. as E 0. for the surface. Eisenhart claims that Meusnier actually showed that for minimal surfaces the mean curvature vanishes. A surface is minimal iff its asymptotic directions are orthogonal. we have D + D = 0.2. This means.1. Proposition 5. Then the minimality condition for the surface becomes E(D + D) = 0 and. . In fact.

Minimal surfaces 203 The following technical lemma will allow us to obtain a very nice characterization of minimal surfaces Lemma 1. if r = r(u. (5. similarly. then S is minimal iff the components of r are harmonical functions. i. λ2 (5. the vector ru2 + rv2 is perpendicular both to ru and rv . (5.2.2.2.4) If we multiply (5. where λ2 = E = G. we have ru = rv = λ2 . Proof. as well as the definition of the second fundamental form.1) ru · rv = 0. is colinear with the unit normal: ru2 + rv2 = a n. we obtain a = D + D = 2λ2 Km . (ru2 + rv2 ) · rv = 0.2. and.2) . Then ru2 + rv2 = 2λ2 Km n.e. whence (ru2 + rv2 ) · ru = 0. which proves the claim.2.4) by n and use the equation (5. we get ru2 · ru = ruv · rv = −ru · rv2 .2. Thus. therefore the mean curvature is just Km = By differentiating (5.3) 2 2 (5.2.2). Let r = r(u.5) D+D . v) be an isothermic parameterization of a surface S .2). v) is an isothermic parameterization of a surface S . Corollary 1.2.5. (5. Since r is isothermic.2. in other words iff we have ∆x = ∆y = ∆z = 0.

We choose as parametric lines the lines of curvature of the surface. ∂u i.e. = 0. D D we have ∂2 log E ∂2 log G = . ∂u∂v ∂u∂v = 0.3. as we already know. Proof.e.2.2.2. ∂v ∂D = 0. the minimality condition for the surface becomes ED + GD = 0 In this parameterization the Codazzi-Mainardi equations become ∂D 1 D D − + ∂v 2 E G ∂D 1 D D − + ∂u 2 E G which. . reduce to ∂D = 0. The spherical image of a minimal surface is conformal (i. Then. we should have F = D = 0 (both first fundamental forms are diagonal in such a kind of parameterization). D depends only on u and D – only on v. Thus. Proposition 5. On a minimal surface we have a very special isothermic parameterization: Proposition 5.204 Chapter 5. it preserves the angles). Since G E =− . The lines of curvature on a minimal surface form an isothermic system. which is exactly the condition that the parametric lines are isothermic. Special classes of surfaces It is a classical (but nontrivial) result that on any smooth (at least C 2 ) surface there exists isothermic parameterization. for a minimal surface.

−2F. we get (5. If the spherical map of a surface is conformal. while (5.e.5.e. f = −FKt + D Km . for a minimal surface. we cannot have at the same time Kt = µ and Km = 0.2. i. . (5. If Kt = µ.8) is Kt = µ. (5. indeed. we have Proposition 5.7) −FKt + D Km = µF.2. from (5.6) We multiply the three relations by D . the two surfaces are. as we know. D Km = G(Kt + µ). Kt = µ = 0. respectively and we add the results. Notice that. then.2.2. D Km = F(Kt + µ). 2 Km = 2(Kt + µ).2. because then we would get.2. Or. then the surface is either minimal.4. −2D . the principal curvatures are equal at each point (all the points are umbilical). as µ > 0.e.2. i. f. D.9) If we put Km . the (5. Thus. Let e. we have e = −EKt . conformally equivalent.9) we get 2 Km = 4µ = 4Kt .e.e. either a sphere. g = −GKt . i. i.8) is satisfied. Thus.9) is satisfied iff µ = −Kt and we rediscover the result from the proof of the previous proposition. the only surface with only umbilical points is the sphere (and its limit case. then.2. g = −GKt + D Km . (5.2. f = −FKt . Minimal surfaces 205 Proof.2.7) by G. E. the plane).8) On the other hand. i. also. The second possibility to satisfy (5. DKm = E(Kt + µ). Proof. then we have −EKt + DKm = µE. g be the coefficients of the first fundamental form of the spherical image. e = −EKt + DKm .2. the spherical image of a √ minimal surface is conformal and the conformal factor is −Kt . respectively. if we multiply the relations (5. If the spherical map is conformal. −GKt + D Km = µG. we get 2Kt Km = Km (Kt + µ) or Km (µ − Kt ) = 0. We have. The reciprocal of this proposition is almost true. Kt = −µ.

but. For a couple of decades. Proof. can be parameterized as   x = u cos v      y = u sin v . also.2. while the second is a ruled surface. uf 1+ f 2 . f D= . D = 1+ f 2 Then the minimality condition for the surface reads (1 + f 2 )u f + f u2 = 0 which can be integrated and gives for f an expression of the form u f (u) = a cosh−1 + c. F = 0. nobody was able to find other examples. The first two minimal surfaces (apart from the plane) were discovered in the eighteenth century by the French mathematician Meusnier. As we know.     z = f (u) Therefore. These surface were the catenoid and the right helicoid. a where a and c are constants. in our case). G = u2 . The catenoid is the only revolution surface which is.206 Chapter 5. Special classes of surfaces Proposition 5.5. as we shall see in the following. D = 0.2 Minimal surfaces of revolution Proposition 5. An arbitrary surface of revolution (around the z-axis. 5. The surfaces found by Meusnier are very simple and they belong classes of surfaces which are relatively easy to study. Proof. The spherical image of isothermic lines on a minimal surface are isothermic lines on the sphere. . .6. This is an immediate consequence of the fact that the spherical map of a minimal surface is conformal.2. which corresponds to the catenoid. these classes do not contain other minimal surfaces. the first surface is a revolution surface. a minimal surface.2. the coefficients of the first two fundamental forms of the surface can be written as E = 1 + f 2 .

G = u2 + f 2 . uf u2 + f 2 . as we will show.7. D = u2 + f 2 F = 0. f D =− . D = 0. As we know. Minimal surfaces 207 Figure 5.2. there is only one. the equations of a right conoid can be written as  x      y     z = u cos v = u sin v . The only right conoid which is a minimal surface is the right helicoid. the condition for a surface to be minimal is that the coefficients of the first two fundamental forms verify the equation ED − 2FD + GD = 0. = f (v) Then the coefficients of the first two fundamental forms of the surface will be E = 1. As we saw.2.2. Proposition 5.5. Well. . Proof. We are interested to see which right conoids are minimal surfaces.5: The catenoid 5.3 Ruled minimal surfaces We recall that the right conoid is the surface generated by s straight line moving parallel to a plane (z = 0 in our case) and intersecting a line which is perpendicular to this plane (in our case – the z-axis).

as we saw previously. there is another method to prove that. the second asymptotic line through each point is orthogonal to the generator passing through that point. i. reduces to uf u2 +f 2 = 0. For such more general ruled surfaces we do not have such a nice parameterization. the surface is a right helicoid. the generators that intersect it are principal normals. Through each point of the surface pass exactly two asymptotic lines. This means that for an asymptotic line from the second family. The only minimal ruled surface is the right helicoid. through each point passes one straight line (the generator). Thus. Let S be a ruled surface which is not a plane. where a and b are constants. any . Then. The asymptotic lines have the property that their principal normals are contained into the tangent plane. One might think that there are other ruled minimal surfaces. On the other hand. To prove that the surface is a right helicoid. therefore this direct approach cannot be applied. as we shall see next. which are not right conoids. in our case.8 (Catalan.e. But this leads to f (v) = av + b. the surface is minimal iff the asymptotic lines are orthogonal.208 Chapter 5. Still. 1842). Special classes of surfaces Figure 5. in fact. Since the surface is minimal. there are no other ruled minimal surfaces. it is enough to prove that it has as directrix a circular cylindrical helix. But they are principal normals also for any other asymptotic line that intersects them. Proof.2. or f = 0. Proposition 5.6: The minimal helicoid which.

the expressions: E = 1 + 2vα · γ + v2 γ 2 .5. γ. that u is the natural parameter of the directrix. (α × γ) · α + v (α × γ) · γ + (γ × γ) · α + v2 (γ × γ) · γ = 0. and γ. A straightforward computation provides. D= D = D = 0. We shall give now. We may choose the directrix α of the ruled surface to be an orthogonal trajectory of the family of rulings. (α + vγ . the directing vector of the rulings. Thus. as we know. at each point of the directrix. which is the same. using a local parameterization of the ruled surface. Then from the definition. α + vγ ) = 0. the equality holds for any v iff the coefficients of the polynomial function vanish simultaneously: (α × γ) · α = 0 (α × γ) · γ + (γ × γ) · α = 0 (γ × γ) · γ = 0. a more direct prove of the Catalan’s theorem. they have to be cylindrical helices. Another proof of the Catalan’s theorem. We shall assume. the minimality condition reads (α + vγ . following Barbosa and Colares. (a) (b) (c) . α + vγ ). 1 1 + 2vα · γ + v2 γ 2 1 1 + 2vα · γ + v2 γ 2 (α + vγ . As the left hand side of the previous equation is a polynomial function in v. γ.2. γ ). for the coefficients of the first two fundamental forms of the surface. v) = α(u) + vγ. Let S ⊂ R3 be a ruled surface. But then. or. S has local parameterizations of the form r(u. G = 1. γ. to be a unit vector. F = 0. moreover. Minimal surfaces 209 asymptotic line from the second family has an infinity of Bertrand mates.

we have α (u) = k(u)ν(u). both γ and α are perpendicular on γ. Now. at this point (and. these planes have to coincide. since they contain at least the vector γ. (b’) Now. instead. in this situation we have. actually. Moreover. it follows the parallelism just mentioned. as α is naturally parameterized. in this case. As. ν. we have α (u) ⊥ α (u). γ and γ . this means. γ. only two situations are possible: either their intersection is a straight line. In reality. The plane that contains all our vectors could. On the other hand. . either they coincide. Therefore. now. But this means that the plane of the vectors is invariant. and the three vectors are coplanar. therefore. First of all. the vector α has unit length. Clearly. But the. Then. in principle. also. to this end it is enough to prove that both the curvature and the torsion of the curve are constants. vary from point to point. As the three vectors are coplanar. in an entire neighborhood of it). β} the Frenet frame of the curve α. Suppose. which means that. therefore the curve α is a plane curve and the surface is a plane. by continuity. that there exists a point (hence an entire open set) where the two vectors are not parallel. and. Indeed. in fact. Let us assume.210 Chapter 5. Suppose. we have (γ × α ) = γ × α + γ × α = 0. this two planes are not parallel. As we know. respectively. it follows that α is colinear to γ. the two vectors are parallel for any value of the parameter. that γ is always parallel to γ. But α is also perpendicular on γ. which means that α is perpendicular to α . because α is naturally parameterized. as we assumed that α and γ are not just smooth. But. first. Indeed. that the two planes always coincide and we are. otherwise γ cannot belong simultaneously to both planes. We are going to show that. the two planes will also coincide. as the curve α is naturally parameterized. α is a circular helix. We intend to use the Frenet equation to find the curvature and the torsion. Special classes of surfaces From the relation (a) it follows that the vector α must be contained in the plane generated by the vectors α and γ. that there exists a point where γ is not parallel to γ. again. Indeed. instead of using the explicit formulas we obtained when we studied the space curves. this not happens. Let {τ. also α γ . but analytic functions. the relation (b) reduces to (α × γ) · γ = 0. in the previous situation. from (b’) and (c) follows that the vector γ should belong to the intersection of the planes determined by the vectors α . from construction. If α is always parallel to γ . γ is also perpendicular to α .

i. α should be a circular helix and has a parameterization of the form α(u) = (A cos au. also. But. the curvature of α is constant. Thus. this formula is β = −χν. Minimal surfaces as γ is a unit vector. A sin au. from the existence and uniqueness theorem for space curves. the derivative of the curvature will be given by ± dk = −(α × γ ) = −α × γ − α × γ = 0. we have γ (u) = ±ν. on the other hand. we have τ = α and. β ≡ τ × ν = ±α × γ.e. Hence the torsion will be given by χ = −β · ν = ±(α × γ ) · γ and it is easy to see that the derivative of the torsion is.5. ±k(u) = α (u) · γ(u). the torsion is a constant function. hence. Now. whence β = ±(α × γ) = ±(α × γ + α × γ ) = ±α × γ . Thus. du 211 i. ν = ±γ.2. identically zero. On the other hand. we know a curve which has constant curvature and torsion. therefore. thus. we have 0 = (α × γ) = α × γ + α × γ. But.e. can be found using the third Frenet’s formula. The torsion. on the other hand. up to a motion in space. we have for the curvature the expression ±k(u) = −α × γ . using the previous equation. As we saw. Indeed. we know that there exists a curve which has the prescribed curvature and torsion and this curve is unique. . bu). namely the circular helix. since α ⊥ γ.

He searched a minimal surface in an explicit representation. y). t) = (s cos at. Special classes of surfaces where. Obviously. becomes r(s. As we already mentioned. the constants A. He started by noticing that. Therefore. with α and γ just found. ∂x2 ∂y2 Now. On the other hand. the minimality condition becomes ∆ f = k± grad f 3 .e. we have. It is both interested and instructive to see how Meusnier got to this result. As we saw. since α is supposed to be naturally parameterized. the equation of the surface. which means that the surface is a part of a right helicoid. Meusnier was the one who discovered that the right helicoid is a minimal surface. If the level curves are straight lines. γ = ±{cos au. such surfaces would be ruled. y x Meusnier was interested in solutions of the previous equation for which the level curves f (x. as γ is parallel to α . the minimality condition can be written as (1 + f 2 ) f x2 − 2 f x f y f xy + (1 + f 2 ) fy2 = 0. we put s = A ± u.212 Chapter 5. The only developable minimal surface is the plane. for a plane curve given in the implicit form f (x. ∆f ≡ ∂2 f ∂2 f + = 0. in this case. then their signed curvature vanishes. Consequence. y) = c. z = f (x. y) = const are straight lines in the plane. the signed curvature can be written as k± = − f x2 f 2 y + 2 f x fy f xy − fy2 f grad f 3 2 x . b are subject to A2 a2 + b2 = 1. i. as claimed. y) = a arctg + b. which means that f is a harmonical function. now. 0}. z − z0 . also. it can be shown that the only solution of this equation for which the level curves are straight lines is y − y0 f (x. bt). a. sin au. s sin at. t = v. If.

y) is. ∞) × (0. i. (5. in such a way that. as the following counterexample shows. ln u1 ) . even the original deduction of Meusnier could make one suspect that the only ruled minimal surface is the right helicoid. We say that a surface S has constant curvature if its Gaussian curvature is independent of the point. We already met surfaces which have zero curvature. x0 . v1 ) = (u1 cos v1 . Counterexample. We shall give. The graph of the function z = f (x.e. given by r1 (u1 . given by the parametric equations   x = a sin u cos v      . Surfaces of constant curvature 213 with a.3) is no exception. This result does not hold.5. if the surfaces don’t have constant curvature. either a a plane (if a = 0). It turns out that any surface of constant negative curvature. In other words. however. As one can see. We also know an example of a surface that has constant positive curvature (the sphere). is a constant. y0 constants.3.1) y = a sin u sin v    z = a ln tan u + cos u  2 A straightforward computation shows that the pseudosphere has the Gaussian curvature −1/a2 . defined on the same domain in the plane. Let r1 : D1 ≡ (0. A classical theorem of Minding (1839) claims that all the surfaces that have the same constant curvature are locally isometric. 5. we can define a surface to be complete if it is unextendible (like the sphere. if extended enough. namely the pseudosphere. now. u1 sin v1 . they have the same coefficients for the first fundamental form. becomes singular.1. b. The pseudosphere (see the figure 5. an example of a (parameterized) surface that has constant negative curvature.3.3. either a portion of the right helicoid   x = x0 + u cos v      y = y0 + u sin v     z = av + b. This means that they have local parameterizations. there are no non-singular complete surfaces of constant negative curvature in R3 . for a given pair of parameters. then. 2π) → R3 .3 Surfaces of constant curvature Definition 5. namely the plane and the developable surfaces. or the plane). Loosely.

1 . (1 + u2 )2 2 . v1 ) = (u2 . as one can therefore they are not isometric (because. 2 E2 ). they are equal. they might look very different. (1 + u2 )2 1 K2 = − 1 . On the other hand. for their Gaussian curvatures we get K1 = − i. as the surfaces from the figure 5. therefore. E1 convince oneself. a diffeomorphism. As one can see. even if.3. given by r2 (u2 . An easy computation shows that the first fundamental forms for the two parameterized surfaces are. G 2 = 1 + u2 . generated by the graph of the logarithmic function. v2 ) = (u2 cos v2 . u2 1 F1 = 0. F2 = 0. the identity and it is. for instance.7: The pseudosphere and r2 : D2 ≡ (0. in fact. v2 ). v2 ) . Λ(u1 . while the second one is part of the right helicoid. the first surface is a surface of revolution. 2π) → R3 . G 1 = u2 1 1 . for instance. Special classes of surfaces Figure 5. the surfaces of the same constant negative curvature are locally isometric with the pseudosphere. u2 sin v2 . clearly Λ is.e. ∞) × (0. respectively: E1 = 1 + and E2 = 1. We consider the map Λ : D1 → D2 . However. belonging to Kuen and Dini show. As D1 = D2 .214 Chapter 5.

8: Non-isometric parameterized surfaces (a) Kuen’s surface (b) Dini’s surface Figure 5.9: Surfaces of constant negative curvature .3. Surfaces of constant curvature 215 (a) Surface of revolution (b) Helicoid Figure 5.5.

Special classes of surfaces .216 Chapter 5.

2. b < a.      y = 0.    z = ϕ(u).    z = u  around the z-axis. . Find the equations of the torus.     a    y = 0. in the coordinate plane xOz. Find the equations of the catenoid. the curve given by   x = f (u).  Find the parametrical equation of the surface obtained rotating this curve around the z-axis. obtained by rotating the circle   x = a + cos u. We consider. around the z-axis. 3.     z = b sin u. obtained by rotating the catenary   x = a cosh u .Problems 1.

0.    z = a ln tan u + cos u  2 around the z-axis. . 8. u sin v. 9. Write the equations of the normal to the pseudosphere   x = a sin u cos v.     z = hv 10.      y = a sin u sin v. at the point of coordinates (0. 2. 6.     z = a ln tan u + a cos u 2 at an arbitrary point and find the unit normal vector. a) (“north pole”). Write the equation of the tangent plane to the sphere of radius a. Show that the surfaces x2 + y2 + z2 = αx. hv}. 7. 5. then the plane α is the tangent plane to the surface at the point P. Show that if a smooth surface Φ and a plane α have a single common point P. 2) and find the unit normal vector at this point. obtained by rotating the tractrix   x = a sin u.218 Problems 4. 11. Find the equation of the tangent plane at an arbitrary point of the right helicoid r = {u cos v.      y = 0. x2 + y2 + z2 = βy and x2 + y2 + z2 = γz are pairwise orthogonal to each other. Find the equations of the pseudosphere. Determine the equation of the tangent plane to the surface xy2 + z2 = 8 at the point (1. Find the unit normal vector to the right helicoid   x = u cos v      y = u sin v . centred at the origin.

Write the equation of the tangent plane and the equations of the normal at the surface generated by the binormals of a curve ρ = ρ(s). n} 22. 13. Establish the equation of the surface. while n is the normal to the torus at that point. A surface is generated by the principal normals of a curve γ. where r is the radius vector of a point of the torus. Find a close surface on which the function r · n is a constant. measured from the surface. while the normal to the surface is the principal normal to the curve γ. Find the geometrical locus of the points from the torus at which r · n = 0 (the origin of the coordinates is taken at the centre of the torus). 15. Φ∗ . 18. Show that the tangent plane at an arbitrary point of the surface z = xϕ y . Show that at the points of the curve γ the tangent plane to the surface coincide with the osculating plane to the curve γ. passes through the origin of coordinates. The ends of these segments describe another surface. 16. Write the equation of the tangent plane to the torus   x = (3 + 2 cos u) cos v. 14. On the normals to a surface Φ. 20. Show that the tangent plane to a surface generated by the tangents to a space curve is the same at all the points of a given generator. (R + r cos u) sin v. Show that at the points of the curve γ the tangent plane to Φ coincide with the osculating plane to γ. . 19. Show that the surfaces Φ and Φ∗ have. there are taken segments of the same length. 17.     z = 2 sin u. Study the variation of the sign of the function r · n for the torus. 21. Show that the normal to a surface of rotation at an arbitrary point passes through the rotation axis. common normals and parallel tangent planes. r sin u} at which its normal is parallel to the vector a = {l. The surface Φ is generated by the binormals of a curve γ. m. in the same direction. Find the points of the torus ρ = {(R + r cos u) cos v. where ϕ x is a smooth function of a single real variable. √ parallel to the plane x + y + 2z + 5 = 0.      y = x = (3 + 2 cos u) sin v.Problems 219 12. “parallel” to the surface Φ. at corresponding points.

y > 0.     z = t. also. Find the envelopes and the regression edge for a family of spheres passing through the origin of the coordinates and with the centers on the curve   x = t3 . equal to a. with the centers on the circle   x2 + y2 = b2 . 28. 6. the tangency point. the characteristics and the regression edge for the family of rectifying planes of a space curve ρ = ρ(s). Find the envelope of a family of planes forming with the coordinate solid angle x > 0. Find the envelope of the family of spheres x2 + y2 + (z − C)2 = 1. Find the envelopes. z > 0 a triangular pyramid of constant volume V. Find the envelope. 29. 30. the characteristics and the regression edge for the family of normal planes to a space curve ρ = ρ(s) (the polar surface of the curve). the characteristics and the regression edge of the family of osculating planes of a space curve ρ = ρ(s). the characteristics and the regression edge for a family of spheres of constant radius a.220 Problems 23.    z = 0. 1}.  25. Find the envelope.     z = bt. Find. Find the envelope and the regression edge of the family of spheres of constant radius. . perpendicular to the vector a = {−2.      y = a sin t. 27. 24. 32. 31. Find the envelope and the regression edge for the family of normal planes to the helix   x = a cos t. Find the envelope. 26.      y = t 2 . Find the tangent plane to the surface z = x4 − 2xy3 . with centers on a given curve ρ = ρ(s) (tubular surface).

x2 + y2 + z2 = r2 (the sphere). the curvature equal to 3 4 from the curvature of the regression edge at that point. c) (x − C)2 + (y − C)2 = r2 . . 37. b) x2 + (y − C)2 = r2 . 35. n (α) 0. Find the first fundamental form of the second degree surfaces 1. 4. Show that the osculating plane to the regression edge of a developable surface coincides with the tangent plane to the surface at that point. where n = 1. A Catalan surface is a skew ruled surface whose generators are parallel to a given plane. Write the equation of the tangent plane at an arbitrary point of the surface generated by the tangents to the space curve r = r(s). 2. the characteristics and the regression edge of the family of planes (r. = −1 (the hyperboloid with two sheets). = 1 (the hyperboloid with one sheet). 34. for arbitrary point of this edge. Find the envelope and the characteristics of the family of circular cylinders of constant radius r: a) (x − C)2 + y2 = r2 . b ) = 0. n(α)) + D(α) = 0. n (0) 0.Problems 221 33. Find the envelope. the directing plane. b 0. Study its position when the tangency point moves along a generators of the surface. 3. 39. x2 a2 x2 a2 x2 a2 + + + y2 b2 y2 b2 y2 b2 + − − z2 c2 z2 c2 z2 c2 = 1 (the ellipsoid). 38. 36. Show that a ruled surface r = r1 (u) + vb(u) is a Catalan surface iff (b. Show that the intersection curve between a developable surface and the osculating plane at its regression edge has. b .

Gu = 0. 40. ds2 = du2 + 4dudv + 4dv2 . x2 p x2 p x2 a2 x2 a2 x2 a2 Problems + − + + − y2 q y2 q y2 b2 y2 b2 y2 b2 = 2z (the elliptical paraboloid). 2. Find the transformation formulas for the coefficients of the first fundamental form and of the quantity H = EG − F 2 for a parameters’ change. 7. 3. Show that any revolution surface has a parameterization for which the first fundamental form can be written under the form ds2 = du2 + G(u)dv2 . 10. ds2 = du2 − 4dudv + 6dv2 . 9. 4. Find a coordinate transformation such that the first fundamental form of the pseudosphere. − z2 c2 = 0 (the cone). = 2z (the hyperbolical paraboloid). Specify which of the following quadratic forms can play the role of the first fundamental form for a surface: 1. ds2 = du2 + 4dudv − 2dv2 . = 1 (the hyperbolical cylinder). 42. Show that a coordinate net on a surface is a Tchebysheff net iff Ev = 0. 43. A coordinate net on a surface is called a Tchebysheff net if the segments of coordinate lines from a family intercepted between two coordinate lines from the other family have constant length. 6. = 1 (the elliptical cylinder). y2 = 2px (the parabolical cylinder). . 8. ds2 = du2 + 4dudv + dv2 . ds2 = a2 ctg2 udu2 + a2 sin2 udv2 becomes ds2 = du2 + G(u)dv2 .222 5. 41. 44.

On the circular cone   x = u cos v.     z = R cos v which bissect the angle between the meridians and the parallels. Find the orthogonal trajectories of this family. Find the orthogonal trajectories of the rectilinear generators of a conical surface. 49.      y = u sin v.     z = u.     z = R sin u. Find the equations of the curves from the sphere   x = R cos u sin v. lying on the sphere   x = R cos u cos v. 51. u − v = 0 .      y = R sin u sin v. 52. where α is a parameter. 46.Problems 223 45. 48. Find the loxodroms of the sphere. Find the orthogonal trajectories of the family of curves u + v = const. 47. Find the equation of a curve which intersect the meridians of a revolution surface under a constant angle α (the loxodrom). one considers the family of curves v = u2 + α.      y = R cos u sin v. 50. Find the intersection angle between the curves u + v = 0. Find the differential equation of the curve which intersects the rectilinear generators of a tangent developable under a constant angle α.

v = 1. 2 v = 1. v2 ). Find the area of the quadrilater from the right helicoid   x = u cos v. 56. u = a. bounded by the curves u = 0. Find the angle between the curves v = u + 1 si ¸ from the surface v=3−u   x = u cos c.224 from the right helicoid Problems   x = u cos v. whose sides are arcs of great circles of the sphere. 55. On a sphere there is given a right angle triangle.      y = u sin v. 57. from the surface with the first fundamental form ds2 = du2 + (u2 + a2 )dv2 . Find: 1. v1 ) and M2 (u2 . 2. 54. On the surface with the first fundamental form ds2 = du2 + sinh2 udv2 find the length of the arc of the curve u = v between the points M1 (u1 .       y = u sin v. the area of the triangle. the relation between the sides of the triangle. Find the perimeter and the interior angles of the curvilinear triangle 1 u = ± av2 .      y = u sin v. 53.     z = av.     z = av. v = 0. .    z = u 2 .

3. 8. Find the principal curvatures and the principal directions of the right helicoid x = u cos v. y = u sin v. x = f (u) cos v. x = a sin u cos v. 59. 2. x = a cosh u cos v. y = R cos u sin v. y = a cos u sin v. Find the principal curvatures at the vertices of the hyperboloid with two sheets x 2 y2 z2 − − = 1. x = a sinh u cos v. y = a cosh u sin v. z = u2 – the paraboloid of rotation. 4. 60. y = u sin v. a2 b2 c2 61. z = R sin u – the sphere. 9. z = av. x = u cos v. x = a cosh u cos v. y = a cosh u sin v. a a u 11. y = u sin v. 7. 0). z = c sin u – the ellipsoid of rotation. . z = c cosh u – the hyperboloid of rotation with two sheets. x = R cos v. Compute the principal curvatures of the surface x2 y2 + = 2z p q at the point M(0. 0. 6. y = R sin v. Find the second fundamental form of the following surfaces of rotation: 225 1. y = f (u) sin v. z = ku – the circular cone. z = ϕ(u) – the surface of rotation of rotation axis Oz. z = b sin u – the torus. x = u cos v. x = (a + b cos u) cos v. 5. (a + b cos u) sin v. x = a cos u cos v. z = a ln tg 2 + cos u – the pseudosphere. z = u – the circular cylinder.Problems 58. . x = R cos u cos v. y = a sinh u sin v. 10. z = u – the catenoid. z = c sinh u – the hyperboloid of rotation with one sheet. y = a sin u sin v.

On the surface x = u2 + v2 . In the tangent plane at the point M to a surface on draws n straight lines passing through the origin and intercepting between them angle equal to π . 2. Show that n Km = 1 1 1 1 + + ··· + . 65. z = uv we take the point P(u = 1. 2. in the direction of the tangent to the curve y= 64. Find the curvatures of the normal sections of the surface y = 2 x2 : 1. Compute the principal curvatures of the surface at P. Find the equations of the tangents PT 1 . 1. 4). y = u sin v. n r1 r2 rn 1 2 x . at the origin of coordinate. z = λu one of the normal principal sections is a straight line. 2 1 where ri are the normal curvatures of the curves from the surface which are tangent to the given straight lines. the curvature radius of the normal section curve whose angle with the x-axis is 45◦ . 1 63. Compute.226 Problems 62. 3. y = u2 − v2 . 3. while Km is the mean curvature of the surface at the point M. Find the equation of the Dupin’s indicatrix at the origin of coordinates. 2 1. 2. 2. at the point M(2. at the points of the lines obtained by sectioning the surface with planes z = k and in directions tangents to this curves. v = 1). z = x2 . . at an arbitrary point and in an arbitrary direction. Show that at each point of the surface x = u cos v. PT 2 to the principal normal sections at the indicated point. We are given the surface 9 z = 2x2 + y2 . 66. Find the curvature of the normal section passing through the tangent to the curve v = u2 .

69. using Meusnier’s theorem. . Through a point M of a surface one considers all the possible plane sections. 70. On what curve lie the foci of the parabolas obtained sectioning the cylinder with planes passing through the straight line MT ? 71.Let M be the vertex of this parabola and MT – the tangent to the parabola at the vertex M. find the curvature radius of an ellipse at one of its vertices. 68. lying on the ellipsoid.Problems 227 67. Find the geometrical locus of the curvature centers of these curves at M. Find the surfaces for which the first fundamental form is a perfect square. Through a point M of an ellipsoid of rotation one draws all the possible curves. Starting from the fact that an ellipse can be projected onto a circle. Find the equation of the surface which contains the centers of the osculating circles of these sections. A parabolical cylinder intersects with a plane which is perpendicular to its generators after a parabola C.

228 Problems .

A. C. 1868–1869. Springer Verlag. Berlin. M. Springer Verlag. E. 33(1912) [7] do Carmo. J. Giorn. 1976 [8] Catalan. Colares.. 1195) [3] Beltrami.M.P. p. – Sulle coordinate curvilinee d’una superficie e dello spazio. 1867–1868. (2). W.L. – Minimal Surfaces in R3 . 2. Rend. Gauthier-Villars. di Mat. Pisa. Berlin. 1887-1896 . 2001 [2] Barbosa. – Journal de Mathém. 1986 (Lecture Notes in Mathematics.Bibliography [1] Bär. 4. Elementare Differentialgeometrie..G. 2-3. Paris. 1864–1865 [4] Bianchi. 1870–1871. 1. – Leçons sur la théorie générale des surfaces. Prentice Hall. G.203 [9] Codazzi. C. [10] Darboux. – Lezioni di geometria differenziale. Volumes I to IV. – Vorlesungen über Differentialgeometrie. E. di Mat. Walter de Gruyter. – Differential Geometry of Curves and Surfaces. I. – Fondamenti per la geometria differenziale di una superficie col metodo vettoriale generale. 7 (1842). Spoerri. – Elementare Differentialgeometrie. L. Ann.. 1894 [5] Blaschke. 1921 [6] Burali-Forti. D. – Ricerche di analisi applicata alla geometria. Palermo.

Ch. Standford. 1952 [26] Hopf. 1852 [23] Gauss.P. C. – Geometry and the Imagination. – De projectione geographica superficiei sphaerici.P.. – Développement de géométrie. D. 1813 [12] Dupin. 1982 [20] Finikov.C. Cohn-Vossen. Bd. Comm. 1912 [22] Frenet.S. A. H.) – Differencial’naja geometrija. de Math. Applied Mathematics and Statistics Laboratory. L. Gauthier-Villars. 1748 [15] Euler. C. – Lectures on Differential Geometry in the Large. Petrop. Paris. John Wiley. I. – Sur les courbes à double courbure. S. – Lectures on the Differential Geometry of Curves and Surfaces. Institute of Mathematical Sciences.108 [14] Euler. – Selected Topics in the Differential Geometry in the Large. F. 1823-1827 [24] Gray. New York. – Cours de géométrie differentielle locale. – Applications degéométrie et de m’ecanique. L. Gray. Paris. Ginn & Co. Boston. p.P. 1955 [28] Hsing. Notes by J. I–II. 1981 .. S.F. H. A. W.230 Bibliography [11] Dupin. – Disquisitiones generales circa superficies curvas. Göttingen. Klotz. Soc. 1957 [19] Fedenko. A. (ed. 1909 [17] Eisenhart. 1777 [16] Eisenhart. – An Introduction to Differential Geometry. Minsk. Moscow. – Zeitschrift für Mathem. 1993 [25] Hilbert. New York. – Modern Differential Geometry of Curves and Surfaces. Acta Ac. Standford University. Notes by T. J. Lausannae. und Physik. – A First Course in Differential Geometry. 1. Cambridge University Press. Princeton University Press. 9 (1864). 1955 [27] Hopf. Paris.. 6.R. – Introduction in analysin infinitorum. L. 1952 [21] Forsyth. Journ. A. – A Treatise on the Differential Geometry of Curves and Surfaces. Ch. 1947 [18] Favard. 1822 [13] Enneper. New York University. – Kurs differencial’noi geometrii. Chelsea. CRC Press. L.

Toronto. Nouv. 1787 [42] Monge. 1839 [41] Monge. reine angew. – A Course in Differential Geometry. 51). J. J. 1975 [44] O’Neill. New York. Journ. Math.B. Paris.L. G. – Differential Geometry. 1838 [40] Minding. näbst Bemerkungen über die Flächen von unveränderlichem Krümmungsmasse. Ac. 1959 [33] Lagrange. 18. E. J.. 1760–1761 [34] Lagrange. – Applications de l’analyse à la géométrie. Journ. W. Mém. Ac. 2e édition. University of Toronto Press. Math.. – Mémoire sur la courbure des surfaces. B. Springer (Graduate Texts in Mathematics. – Ueber die Biegung krummer Flächen. 1846 [31] Klingenberg. Mém. – On the properties of inextensible surfaces. F. – Wie sich entscheiden lässt. 1785. – Cours de mathématiques. Academic Press. Paris. – Sur la construction des cartes géographiques. 10 (lu 1776). Giorn. 19. F. – Leçons sur les coordonnées curvilignes et leurs diverses applications. J. 1779 [35] Lamé. 30. 1856 [30] Joachimstahl. F. – Vorlesungen über Minimalflächen. G. Math.H.. Paris. – Essai d’une nouvelle méthode pour déterminer les maxima et les minima des formules intégrales indéfinies. Tome 3: Géométrie et cinématique. textbf22 [1855]. Irish Ac.C. – Mémoire sur l’intégration de quelques équations aux derivées partielles. II.C. reine angew. J.M. – Sulla teoria generale delle superficie. Trans. Ist. – Demonstrationes theorematum ad superficies curvas spectantum. J. ob zwei gegebene krumme Flächen auf einander abwickelbar sind oder nicht. Springer. 1977 [37] Meusnier. 1983 [32] Kreyszig. G. G. Mémoires des savants étrangers. reine angew. Berlin. – Elementary Differential Geometry. 1966 .. Journ. 1859 [36] Lelong-Ferrand. 9. J. Dublin.L. 1807 [43] Nitsche. 477–510 [38] Mainardi. 1856 [39] Minding. Lombardo.Bibliography 231 [29] Jellet. Dunod. sci. Arnaudiès. Miscellanea Taurinesnea.

16. Math. Einführung in die Theorie der Flächen. reine angew. D. Dover. J. Journ.V. A. Moscow. Journ. second edition. – Crelle’s Journal f. 1956 [48] Riemann. 1902 [52] Scherk. – Anwendung der Differential. Dover. – Differential Geometry. 4th edition. O. Clarendon Press. A. Teubner. Mon.K. 1986 [46] Pogorelov. – Elementary Topics in Differential Geometry. Ya. 1900.232 Bibliography [45] Osserman. T. textbf3. 1868 [49] Rodrigues. Einführung in die Theorie der Kurven in der Ebene und im Raume. G. 1979 [58] Vygodskii. – Mémoire sur quelques formules relatives á la théorie des courbes à double courbure. – Kurs differencial’noi geometrii.J. 1815 [50] Germain. J..und Integralrechnung auf Geometrie. Noordhoff. – Lectures on Classical Differential Geometry. – An Introduction to Differential Geometry. 114– 118 [55] Stoker. Berlin-Leipzig. 1851 [54] Steiner. 1969 [56] Struik. John Wiley and Sons. de Math. – Mémoire sur la courbure des surfaces. École polyt. – Recherches sur la théorie analytique des lignes et des rayons de courbure des surfaces.. I. S. – Über die Hypothesen. 1949 [59] Willmore. Groningen. Berlin. Gött. Oxford. M. Ber. Corrésp. 1966 [47] Raševskii... Abh. P. B. II. 1830 [51] Scheffers. 13 (1835) [53] Serret. Mathem. 1959 . 7. – A Survey of Minimal Surfaces. P. – Differential Geometry. 1840.A. – Differential’naja geometriya. second edition. – Ueber parallele Flächen. R.. Springer (Undergraduate Texts in Mathematics). Ak. welche der Geometrie zu Grunde liegen.. 13. Wiss.J. 1988 [57] Thorpe. Moscow-Leningrad. J. pp.

28. 154. 107–112. . 149. 180. 73. 198. 55. 43–46. 220. 191. 77. 223. 13. 150. 67. 170. 194. 167. 59. 43. 224. 224 area. 209 Bertrand curve. 42. 89–93. 207. 208 line. 150–154. 54. 148. 109 arc length. 227 coordinates. 89. 67. 154. 80 asymptotic direction. 200. 208. 209 binormal. 110 conchoid of Nicomede. 16. 154. 227 clothoid. 190. 217 catenoid. 100. 220. 20. 193. 99 Codazzi-Mainardi equations. 164 curvature. 156. 179. 217. 49–52. 66–68. 104 cone. 162. 91. 150 absolute. 60. 96. 109. 109. 57. 142–144 astroid. 198.Index affine normal. 181. 86. 198 center. 195. 34. 151. 99. 64. 217. 110 lengthened. 181 circle. 141. 67. 39. 222. 48. 62. 157. 165. 125 Cornu’s spiral. 191. 225 characteristic. 150. 58–62 mate. 168. 49 biregular point. 39–41. 87. 199 biregular curve. 210 Gaussian. 69. 54. 19. 198. 204–206 conoid. 103–105. 164. 189. 85 axis. 41. 40. 190. 45. 225 conformal mapping. 204 conchoid. 70. 23. 197 biregular curve . 152. 58. 118. 41. 53. 62. 221 Christoffel’s coefficients. 182. 78. 178. 206. 99 curvature. 64. 198. 208 contact. 66 Bonnet’s theorem. 88. 168 catenary. 101. 104–106. 86. 19. 59. 197. 202. 171.

89. 197. 220. 57. 175. 154. 96 of a regular curve. 53. 179. 227 Darboux axis. 157. 99. 226 normal. 107–110. 83. 109. 143. 54. 50. 213. 195. 158–160. 191. 92. 221 of space curves. 161 envelope of plane curves. 95. 66–69. 91. 109. 209 edge of regression. 95. 156–158. 81. 211 frame. 84. 55. 208. 161. 148. 221. 221 epicycloid. 220. 221. 89. 227 signed. 64. 42–44. 109 Euler’s formula. 148–152. 179. 63. 154. 105. 161 cycloid. 212. 203. 139–141. 71. 191–193. 107. 161. 177. 75–80. 13. 81. 148– 152. 200 . 80. 175. 165. 225 elliptical. 97. 154. 200. 170–174. 197–200. 193. 200 developable surface. 205. 47. 225. 195–197. 90. 51. 90. 108. 177–180 line. 221 ellipse. 105. 81. 190 of planes. 199. 156. 63. 176. 190. 163. 157 evolute. 43 principal. 154. 110 lengthened. 73. 63. 226 radius. 30. 70. 192. 164. 70. 71. 190. 86. 110 Frenet axes. 204 mean. 79–81. 105. 42. 190. 175. 180. 59–61. 167. 54. 178. 87. 193–195. 63 equations. 149 curvatures of a surfaces. 104 shortened. 43. 199. 162. 176 Darboux vector. 156. 58. 82. 200. 202. 67. 152. 71. 209 first. 197–199. 52. 177. 104. 200. 86. 226 of a curve. 111. 96 vectors. 62. 225. 222 hyperbolical. 44 of a plane curve. 213. 162 fundamental form. 210 system. 95. 44–52. 75. 51. 193. 60. 221 geometrical meaning. 227 elliptic point of a surface. 175. 157. 227 circular. 82–86. 221. 98 cylinder. 167.234 Index Darboux frame. 161. 227 ellipsoid. 156. 157. 222 parabolical. 111. 194. 176. 211. 177 total. 45. 195. 172–174. 212 tangential. 210 formulae. 22. 180. 93. 226. 82. 70 fundamental equations of a surface. 220. 53–55. 57. 198. 153. 213– 215 vector. 191 of surfaces. 104 cycloidal curves. 66. 223 directrix. 222. 214 geodesic. 210. 99. 43. 200. 97. 84–86. 58. 62. 191. 197. 185–187. 168.

19–22. 180. 213. 128–131 Meusnier’s theorem. 105. 137 orientation . 225 helix circular. 124. 168. 212– 214. 146. 206 isothermic parameterization. 35. 110 isothermic lines. 59. 146–148. 147. 189. 206. 42– 44. 62–64. 225 fundamental forms. 222. 182 235 Möbius band. 165. 221. 138 fundamental operator. 179. 159 Klein’s bottle. 204 Joachimstahl’s theorem. 108–110 equilateral. 200–202. 33. 106. 205. 181. 218. 204. 88. 135. 185. 159. 208–211. 125 orientable surface. 165. 152. 159 normal to a surface. 227 second. 206–208. 101. 37. 212 normal line. 45. 148. 148. 39. 34. 56–58. 232 monkey saddle. 160. 154 normal to a plane. 229. 202–208. 175 normal section. 95–101. 200 helicoid. 154. 71–73. 62. 153. 32. 34–36. 180 geodesics. 44. 46. 197. 19–21. 187–189. 149. 21. 35. 189 normal versor. 167 Gauss egregium theorem. 142. 109 hyperbolic point of a surface. 105 involute. 202. 138. 221. 130. 144. 168. 162. 164. 189. 59. 180–183. 180. 207 fundamental homography. 125. 172. 175. 162. 106. 200. 172 Gauss equations. 225 hypocycloid. 178. 225 of rotation with two sheets. 64. 164 normal versor of a plane. 170. 151 natural equations of a curve. 32. 158. 150–153. 108. 161 hyperboloid of rotation with one sheet. 16. 204. 126 normal vector. 224. 209 natural parameterization. 148. 213. 110 natural parameter. 21 naturally parameterized curve. 126. 67. 178. 177. 210. 159. 225 with one sheet. 212. 150. 84. 137 fundamental theorem of surface theory. 55. 22. 42. 227 minimal surfaces. 112. 153. 80. 105 normal plane. 45 natural basis. 197. 154. 62 hyperbola. 224. 167. 155 moving frame. 132 Liouville surfaces. 165. 167. 86–90. 162. 149. 172 geodesic lines. 105. 86–90. 150. 192 normal to asurface. 66. 123. 221 with two sheets. 40. 220 general.Index 182. 203. 45. 168. 214. 171. 136. 140. 127–130. 179. 107 normal line . 92. 202–204 isothermic system. 51–55. 172.

161 osculating sphere. 139. 44. 95. 136. 112. 127. 59. 204–206 superosculation contact. 134. 213–215 surfaces of revolution. 75. 219. 162. 68. 159. 124. 185–193. 180. 140. 20. 103. 195. 182. 177. 205. 116. 91. 83. 67. 136–138. 108. 225 rectifying plane. 191 tangent space to a surface. 118. 111. 147. 156. 134. 146–153. 158 principal normal. 59. 112. 213. 73. 31. 152. 148–150. 58. 127–130. 49. 221 . 66. 214. 175. 156. 55. 177 torsion. 48. 64 parabolic point of a surface. 64. 47. 197. 177. 208. 123. 49. 146. 198–200. 206. 181. 200. 213. 110. 137. 68 surfaces of constant curvature. 168 oriented curve. 175. 126. 22. 140. 48. 32–36. 67 parabola. 111. 126. 197. 57. 64. 222 of rotation. 152. 60. 46. 50–55. 123–126. 33. 111. 70. 154 second curvature (torsion). 143. 137. 192 saddle point of a surface. 17. 149– 151. 123. 211. 30. 68. 162. 206. 177. 154. 156. 150. 222. 164 sphere. 222. 147. 141. 226 tangent plane . 227 semicubic. 153. 49 of a surface. 157 tangent vector. 39–41. 109. 66–68. 223 tangent plane. 130. 117. 96 ruled surface. 210. 127. 227 osculating plane. 152. 221. 197. 91. 179–181. 45. 198. 49. 208. 199. 136–138. 158–160 principal line. 154. 159. 192. 199 osculating circle. 38–41. 144. 54. 136–138. 143. 189. 185. 64. 144. 226. 148. 148. 125. 199. 193. 68 osculation contact. 92. 177. 200. 111. 198. 45. 124. 223 tangent. 180. 156. 111. 156– 159. 66. 66– 68. 199. 214. 62. 137 pseudosphere. 156–159. 191. 106. 150. 218. 213. 200 torsion. 142. 199. 185–189. 144–147. 200 Riccati equation. 152. 51. 236 geodesic. 84 oriented surface. 45. 135. 67. 219–221 osculating sphere. 146. 226 principal operator. 177 of a curve.236 Index ruling. 142. 220 surface. 161. 206–209. 218–221. 62. 225 polar axis. 212. 106–110. 182. 185. 105. 57. 156. 95. 196. 223–225. 47. 181. 198 polar surface. 137–139. 71. 67. 218. 146. 58. 127. 236 spherical map. 32. 187–189. 175. 220 principal direction. 222 hyperbolical. 161 paraboloid elliptical. 151. 140. 200. 104. 227 tangent developable surface. 53 shape operator. 200.

217. 225 unit binormal. 164 formulae. 47. 165 mapping. 164 operator. 54 unit principal normal. 45. 95 Weingarten coefficients. 164 237 . 152 unit tangent vector.Index torus. 45. 191. 118. 154. 219. 119. 84. 45. 137. 164. 85.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->