March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 1

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Chapter I

Curves
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Students at the college and university level are familiar with graphs of
functions and parametrized curves. In this chapter, we define the cur-
vature of plane curves and space curves, which measures the “amount
of bending” of the curve, and introduce its geometric meanings. As ap-
plications, the rotation index for closed plane curves and properties of
spirals are introduced. While space curves are closely related to surface
theory, which is treated in Chapter II, the rotation index for plane curves
is also used in the proof of the Gauss-Bonnet theorem for 2-dimensional
manifolds in Chapter III.

1. What exactly is a “curve”?

With pencil and paper, one can draw a variety of curves.

(i) (ii) (iii) (iv)

Fig. 1.1 Various curves.

The first picture (i) in Fig. 1.1 is an example of a smooth curve. The
second picture (ii) is also of a smooth curve, but now without endpoints,
since traveling along the curve will bring you back to where you started.
This type of curve is called a closed curve. Like in the picture (iii), we

1
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 2

2 Differential Geometry of Curves and Surfaces

allow crossing points, which we can also call self-intersections, and we still
regard it as a smooth closed curve. The picture (iv) is a closed curve,
but as it has sharp angles at particular points, it is not smooth at those
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

points. This type of curve is called a piecewise smooth curve (cf. page 225
in Appendix B.3). In this case, we are dealing with a curve that is still
smooth almost everywhere. We now introduce mathematical methods for
representing curves in the plane R2 .
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Graphs of functions. In this text, we use the word smooth to refer to

maps that are of class C ∞ (see Appendix A.1 for details). The graph of a
smooth function y = f (x) in the coordinate plane R2 is an example of a
smooth curve. However, general curves are not necessarily √ representable as
graphs of functions. For example, the graph of y = 1 − x2 is the upper
half of a circle of radius 1 with center at the origin, but the full circle√cannot
be expressed as the graph of any single function, because y = ± 1 − x2
cannot be called a single function.

The implicit function representation for curves. The points in the

plane satisfying the equation x2 + y 2 = 1 are those lying on the curve
that is the circle of radius 1 with center (0, 0). In general, if we take a
function F (x, y) of two variables and collect together all points in the plane
satisfying F (x, y) = 0 to produce a curve, we call this the implicit function
representation for curves by F (x, y). As a second example, taking a function
f (x) and defining F (x, y) = y − f (x), the resulting curve will be the graph
of y = f (x).
Example 1.1. Here we give examples of curves, represented implicitly by
the solution sets of the following four equations (a and b are positive reals),
see Fig. 1.2:
x2 y2
Ellipse: + − 1 = 0,
a2 b2
x2 y2
Hyperbola: 2
− 2 − 1 = 0,
a b
Lemniscate1 : (x2 + y 2 )2 − a2 (x2 − y 2 ) = 0,
Astroid2 : (a2 − x2 − y 2 )3 − 27a2 x2 y 2 = 0.

1 Each point on the lemniscate satisfies that the product of its distances to the two
√
points (±a/ 2, 0) is a2 /2. This curve was studied by Jacob Bernoulli (1654–1705).
2 The astroid is a locus of one point of a circle with radius a/4 rotating around the

inside of a circle of radius a. See Problem 1 of Appendix B.1.

March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 3

Curves 3
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

An ellipse. A hyperbola.

The lemniscate. The astroid.

Fig. 1.2

For a smooth function F (x, y) of two variables, we consider the figure

produced by the solutions of F (x, y) = 0. For a point (x0 , y0 ) in this figure
(that is, F (x0 , y0 ) = 0 is satisfied), if
∂F
Fy (x0 , y0 ) = (x0 , y0 ) 6= 0
∂y
holds, points in the figure near (x0 , y0 ) can be given as the graph y = f (x) of
a function f (x), and in particular the figure has no self-intersections near
(x0 , y0 ) (cf. Theorem A.1.6 (the implicit function theorem) in Appendix
A.1). Similarly, if Fx (x0 , y0 ) 6= 0, then, switching the roles of x and y, we
find that the figure is the graph of some function x = g(y) near (x0 , y0 ).
On the other hand, if Fx (x0 , y0 ) = Fy (x0 , y0 ) = 0, then the point is
called a singular point. For example, four singular points appear in the
astroid in Example 1.1, where the curve is cusped, or pointed. Another
singularity is seen at (0, 0) in the lemniscate, where the curve crosses itself.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 4

4 Differential Geometry of Curves and Surfaces

There are many varieties of shapes that singular points can appear as.

Parametrizations of curves. Thinking of a point moving through the

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

plane with respect to a time variable t, we denote by (x(t), y(t)) the point
depending on t. Then the trajectory of these points can be imagined as a
curve. Here, t is called a parameter of the curve. This way of representing
a curve is called a parametrization. Pairing up the two functions x(t), y(t)
and naming the pair γ(t), we have the curve
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

(1.1) γ(t) = (x(t), y(t)).

For the curve obtained as the graph of y = f (x), we can choose the special
form

γ(t) = (t, f (t))

for its parametrization.

Thinking of the figure for the curve as a road and γ(t) as a car traversing
it, the various possible choices for the coordinate t give us the various times
and speeds by which the car can traverse the road.
Example 1.2. In terms of the parameter t, setting3
1 − t2 2t
(1.2) x(t) := , y(t) := (t ∈ R)
1 + t2 1 + t2
gives us a curve that is the radius 1 circle with center at the origin and
with the point (−1, 0) removed. When both coordinates of a point in the
plane are rational numbers, we say it is a rational point. A point on the
circle of radius 1 centered at the origin different from (−1, 0) is a rational
point if and only if it is represented as in (1.2) with a rational number t.
In terms of another parameter s, we set

(1.3) x(s) := cos s, y(s) := sin s (−π ≤ s ≤ π).

If we restrict s to the open interval (−π, π), we again get the same circle
as in (1.2) with one point removed. When (x(t), y(t)) and (x(s), y(s)) are
the same point, the parameters s and t are related by t = tan(s/2).
In general, a curve is assumed to be represented as a smooth parametr-
ization, with parameter t, by

γ(t) = (x(t), y(t)) (a ≤ t ≤ b).

3 The symbol “:=” means “the object on the left-hand side is defined to be the object

on the right-hand side”.

March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 5

Curves 5

Here, we say that a function is smooth over a closed interval [a, b] if it

extends to a function defined and smooth on an open interval containing
[a, b].
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

If we have a strictly increasing smooth function t : [c, d] → [a, b] such

that t(c) = a, t(d) = b and dt/du > 0 on the interval [c, d], then the curve
defined by γ̃(u) = γ(t(u)) (c ≤ u ≤ d) will give the same set of points as
the curve γ(t). In this sense the two curves are the same, and we say that
γ̃ is obtained from γ by a change of parametrization. When there is no risk
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

of misinterpretation, we may write γ̃(u) simply as γ(u).

Example 1.3. The curves in (1.1) can be parametrized as follows:

Ellipse: x(t) := a cos t, y(t) := b sin t (0 ≤ t ≤ 2π),

Hyperbola: x(t) := a cosh t, y(t) := b sinh t (t ∈ R),
a cos t a sin t cos t
Lemniscate: x(t) := , y(t) := (0 ≤ t ≤ 2π),
1 + sin2 t 1 + sin2 t
Astroid: x(t) := a cos3 t, y(t) := a sin3 t (−π ≤ t ≤ π),

where cosh t, sinh t are hyperbolic functions (see page 195 in Appendix A.1).
Given a parametrization γ(t) = (x(t), y(t)) for a curve, and viewing the
curve as the trajectory of the point γ(t) that is moving with respect to
t, and in turn regarding t as representing time, we then have the velocity
vector
 
dx dy
γ̇(t) := (ẋ(t), ẏ(t)) ẋ = , ẏ = ,
dt dt
where an overhead dot “ ˙ ” denotes differentiation with respect to t. If the
velocity vector γ̇(c) vanishes, the value t = c of the parameter or the image
γ(c) is called a singular point (cf. Fig. 1.3). The singular point appearing
in Fig. 1.3 is called a cusp. See Appendix B.9 for a criterion for cusps.

Fig. 1.3 The singular point of the map t 7→ (t2 , t3 ) and its image.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 6

6 Differential Geometry of Curves and Surfaces

For a given curve, the word “singular point” may have different mean-
ings, depending on whether it is represented by an implicit function or as a
parametrized curve. For example, the origin (0, 0) is a singular point in the
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

implicit function representation of the lemniscate in Example 1.1, which

corresponds to the points t = ±π/2 of the parametrization of the same
lemniscate in Example 1.3, which are not singular points of the parametri-
zation.
On the other hand, a point where the velocity vector γ̇(t) is not the zero
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

vector is called a regular point. A curve γ(t) defined on an interval [a, b] is

called a regular curve if all points are regular points. If γ(t) is a regular
curve, γ̇(t) points in the direction tangent to the curve at the point γ(t). In
particular, at a value t = t0 where ẋ(t0 ) 6= 0, the inverse function theorem
(Theorem A.1.5 in Appendix A.1) tells us that for some neighborhood of
x(t0 ), the function x = x(t) has an inverse function t = g(x). In that case,
writing f (x) = y(g(x)), some portion of the curve γ(t) sufficiently close to
the point (x(t0 ), y(t0 )) can be described as the graph of a smooth function
y = f (x). In the same way, by switching the roles of x and y, if ẏ(t0 ) 6= 0,
there exists a function h(y) depending on y so that a part of the curve is
the same as the graph of x = h(y).
Recall that a point where the velocity vector (i.e. the tangent vector) of
the curve γ(t) = (x(t), y(t)) vanishes, that is, ẋ(t) = ẏ(t) = 0, is a singular
point. It is possible that the curve looks like it is not smooth near that
point, for example, it can be sharply pointed there, even though it actually
is a smooth mapping. For example, differentiating the parametrization of
the astroid in Example 1.3
(1.4) γ(t) := a(cos3 t, sin3 t) (t ∈ R, a > 0),
the velocity vector
γ̇(t) = 3a cos t sin t(−cos t, sin t)
vanishes if and only if t is an integral multiple of π/2, because the vector
(− cos t, sin t) does not vanish for any t. (By Theorem B.9.1 in Appendix
B.9, one can show that these are singular points called cusps.)

Length of curves. For a parametrized curve γ(t) = (x(t), y(t)) (a ≤ t ≤

b), take two values t and t + ∆t for the parameter, with ∆t close to zero, to
get two points γ(t) and γ(t + ∆t). The distance between these two points
is
|γ(t + ∆t) − γ(t)|.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 7

Curves 7

Writing the change in the two coordinate functions x(t), y(t) of these two
points as
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

∆x := x(t + ∆t) − x(t), ∆y := y(t + ∆t) − y(t),

we have
s 2  2
p ∆x ∆y
|γ(t + ∆t) − γ(t)| = (∆x)2 + (∆y)2 = + ∆t.
∆t ∆t
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Adding up many such short distances and taking the limit of such sums as
those distances approach zero, we arrive at the length
s
Z b  2  2 Z bp
dx dy
(1.5) L(γ) = + dt = ẋ2 + ẏ 2 dt
a dt dt a

of the curve.4 This length does not depend on the choice of the parametr-
ization for the regular curve (Problem 2 at the end of this section asks for
a proof of this). p
The integrand in (1.5) is the norm |γ̇| := ẋ(t)2 + ẏ(t)2 of the velocity
vector γ̇, so we can write
Z b
(1.6) L(γ) = |γ̇(t)| dt.
a

In particular, recalling that the graph of a function y = f (x) (a ≤ x ≤ b)

can be regarded as a parametrized curve γ(t) = (t, f (t)), the length in (1.6)
becomes
s  2
Z b
dy
L(γ) = 1+ dx.
a dx
We could also consider curves coming from polar graphs r = r(θ) of func-
tions r depending on θ, given in terms of the polar coordinates (r, θ) of the
plane. Such a curve can be parametrized as γ(θ) = (r(θ) cos θ, r(θ) sin θ),
(a ≤ θ ≤ b). The length of this curve is then
s  2
Z b
dr
(1.7) L(γ) = r2 + dθ.
a dθ
In this way, we can compute the length of curves using integrals, but
those integrals will not necessarily be expressible in terms of elementary
functions even when the curve can be parametrized by elementary functions.
Here, elementary functions are those that can be described, using a finite
4 We do not give a rigorous proof here. See, for example, [37, Problems in Chap. 13].
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 8

8 Differential Geometry of Curves and Surfaces

number of arithmetic operations and compositions, in terms of polynomials,

rational functions, n’th root functions, exponential functions, logarithmic
functions, trigonometric functions and inverse trigonometric functions.
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

For example, even for the graph y = 1/x, it is known that the length
cannot be represented in terms of elementary functions. But by knowing
the equation above, we can approximate the length by approximating this
integral (using a computer, for example). Next we give an example where
we can compute the length explicitly:
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Example 1.4. The graph of the function (c > 0)

1
(1.8) y= cosh ax (−c ≤ x ≤ c)
a
gives a regular curve called a catenary, and it is exactly the curve that
one would get by hanging a string of uniform density between two fixed
points (see Fig. 1.4). Here a is a fixed positive real number, and x = ±c

Fig. 1.4 The catenary.

give the fixed endpoints of the hanging string. By equation (1.8), we have
dy/dx = sinh ax, and so the length l of the curve is
Z cp Z c
2
(1.9) l= 1 + sinh2 ax dx = 2 cosh ax dx = sinh ac.
−c 0 a
Using this, one can observe that for a given length l (> 0) of the string and
a distance 2c (2c < l) between the two endpoints, there is only one positive
constant a that satisfies (1.9).
A closed curve without self-intersection points is called a simple closed
curve, and if it is regular, it is called a simple closed regular curve. It is
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 9

Curves 9

known as the Jordan closed curve theorem 5 that a simple closed curve on
the plane separates the plane into two connected open subsets, one bounded
and the other not.
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Definition 1.5. For a given simple closed curve in R2 , the bounded (resp.
unbounded) region6 whose boundary is a given curve is called the interior
(resp. exterior ) of the curve.

The Jordan closed curve theorem can be proven in a rather simple way
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

for polygons (cf. [4]). A proof for general continuous curves can be found
in the textbook [8, Sec. 18] on algebraic topology.
For a simple closed curve with length l, the area A of the interior of the
curve satisfies the inequality

(1.10) 4πA ≤ l2 ,

which is called the isoperimetric inequality, for which equality holds if and
only if the curve is a circle. See Appendix B.3 for the proof, noting that
this appendix is dependent on material in the next section.

Exercises 1
1 Find the length of the portion of the parabola y = x2 where |x| ≤ 1.
2 Show that the length of a regular curve as given in (1.5) does not depend on
the choice of parametrization.
3 (1) For 0 < b < a, show that the length of the ellipse (a cos t, b sin t), 0 ≤ t ≤
2π), is given by
r !
Z 2π p a2 − b2
a 1 − ε2 cos2 t dt ε := .
0 a2

Here ε (0 ≤ ε < 1) is the eccentricity of the ellipse. When ε 6= 0, this

integral cannot be computed in terms of elementary functions. When
ε = 0 we have a circle, and when ε = 1 we have a line segment.
(2) A longitude line of the earth is the intersection of the earth and a plane
containing the axis of the earth (the line joining the north pole and the
south pole), which is an almost circular ellipse whose minor-axis is con-
tained in the axis of the earth and whose major-axis is contained in the
5 Jordan,
Camille (1838–1922).
6A connected open subset of R2 or R3 is called a region or domain (cf. page 194 in
Appendix A.1).
April 24, 2017 8:31 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 10

10 Differential Geometry of Curves and Surfaces

equatorial plane. Using the approximation7

√ x
1−x≈1−
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

for sufficiently small x (which is obtained by Taylor’s formula, cf. equation

(A.1.1) in Appendix A.1), approximate the length of this ellipse, and
estimate the error using the remainder term of Taylor’s formula. Here,
the semi-major-axis a and the semi-minor-axis b of this ellipse are a =
6377.397 km and b = 6356.079 km, respectively.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

4 Taking a circle of radius a (> 0) and rolling it without slippage along the
horizontal x-axis, any given point on the circle traces out an image curve
called a cycloid (see the next figure). We choose that given point so that it
starts at the origin (0, 0) in the x-axis, and we let t denote the oriented angle
between the spoke from the center of the circle to the given point and the
vertical direction. We choose t so that it is zero when the given point is at
(0, 0) and is continuously increasing as the circle rolls to the right. Then the
resulting cycloid can be parametrized as

γ(t) = a(t − sin t, 1 − cos t).

As the circle rotates completely around once, t increases from 0 to 2π, giving
us one period of the cycloid. Show that the length of one period of the cycloid
is eight times the radius of the circle.

2. Curvature and the Frenet formula

Here we introduce the notion of curvature, which is a measure of the
“amount of bending”, of a curve. From now on, we consider only pa-
rametrized regular curves, that is, the velocity vector of the curve never
vanishes. First we define arc-length parameters, the most standard way of
giving parameters.
7 The symbol “≈” denotes “approximately equal”.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 11

Curves 11

Arc-length parameters. Looking at Example 1.2, we can see that there

is more than one way to parametrize a regular curve. This causes us to
wonder what the most natural way to choose a parametrization would be.
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Using the parametrization given in the previous section, we could take a

starting point on the regular curve, and take the distance from that point
along the regular curve as one natural choice of parametrization.

For a regular curve γ(t) (a ≤ t ≤ b),

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Z t
(2.1) s(t) = |γ̇(u)| du
a

is the length of the image under γ(t) of the interval [a, t]. Taking the
derivative of this equation and noting that γ̇ 6= 0, we have
ds
= |γ̇(t)| > 0.
dt
Hence, if l is the length of the regular curve γ(t) over the closed interval
[a, b], then s(t) is a strictly increasing function from the interval [a, b] to the
interval [0, l], and thus the inverse correspondence [0, l] 3 s 7→ t(s) ∈ [a, b]
exists. By the inverse function theorem (Theorem A.1.5 in Appendix A.1),
t(s) is also smooth. Then we can parametrize γ by s as well:
γ(s) := γ(t(s)) (0 ≤ s ≤ l).
We call this s an arc-length parameter of the regular curve.
Up to now we have been using the dot “ ˙ ” to represent the derivative
with respect to t, but when we have an arc-length parameter s, we will
denote the derivative with respect to s by using a “ 0”. The purpose of this
change of notation is to help clarify when we are using an arc-length pa-
rameter. By the chain rule for the derivative of a composition of functions,
we have
dγ dγ dt γ̇(t)
(2.2) γ 0 (s) = = = .
ds dt ds |γ̇(t)|
Therefore we have |γ 0 (s)| = 1. In other words, a curve parametrized by
arc-length always has unit speed.
Conversely, suppose we have a curve γ(t) whose speed is always 1, that
is, |γ̇| = 1 holds. Then, by (2.1), the distance along the curve from some
starting point differs from t only by an additive constant. Thus, an arc-
length parameter means a parameter of unit speed, that is, the norm of the
velocity vector is 1.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 12

12 Differential Geometry of Curves and Surfaces

Thinking of the curve as a road, with a car moving along it with respect
to time t, the car is moving at constant speed 1 when t is an arc-length
parameter, and a general parametrization corresponds to travel where speed
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

can vary.
For a curve γ(s) = (x(s), y(s)) parametrized by the arc-length s,
(2.3) e(s) := γ 0 (s) = (x0 (s), y 0 (s))
is a unit tangent vector of γ(s). Furthermore,
(2.4) n(s) := (−y 0 (s), x0 (s))
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

is a unit vector perpendicular to e(s), on the left side of e(s) (see Fig. 2.1).
This is called the leftward unit normal vector of γ(s), at each point of γ(s),
with respect to the direction of travel given by s.

κ(s) > 0 κ(s) < 0

Fig. 2.1 Sign of the curvature and the shape of the curve.

Definition of curvature. Now, parametrizing the curve γ(s) by an arc-

length parameter s, the velocity vector will have length 1. In other words,
γ 0 (s) · γ 0 (s) = 1
holds. Here the “ · ” means the inner product (also called the scalar product)
for vectors in the plane. Differentiating both sides of this equation with
respect to s using Corollary A.3.10 in Appendix A.3, we have
2γ 00 (s) · γ 0 (s) = 0,
so the acceleration vector γ 00 (s) is perpendicular to γ 0 (s). This means that
γ 00 (s) is proportional to n(s). The ratio κ(s) of γ 00 (s) to n(s) is called the
curvature of the curve at γ(s). In other words, we have
(2.5) γ 00 (s) = κ(s)n(s).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 13

Curves 13

For each given choice of s, the curvature becomes a real number, and so
the curvature κ(s) is a function of s; hence, we can call it the curvature
function. When κ(s) is positive (negative), the curve is turning to the left
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(right) near γ(s). For example, while driving a car, if the steering is not
turned at all, the car is moving straight forward on a path with curvature 0,
and when the steering wheel is turned toward the left or right, the curvature
of the path is positive or negative, respectively (see Fig. 2.1).
Considering e and n as column vectors, together they form a 2 × 2
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

orthogonal matrix (e, n). Since this matrix has determinant 1, we have
that (see Problem 2 at the end of this section)

(2.6) κ(s) = det(γ 0 , γ 00 ),

where “det” denotes the determinant of the matrix.

Example 2.1. For a line parallel to a unit vector v, we can take a point
on the line with position vector c. Then any other point on the line can
be given as γ(s) = c + sv, and s becomes an arc-length parameter for the
line. In particular, γ 00 (s) = 0, so the curvature is identically zero.
Let us compute the curvature of a circle. For a positive constant a,
 
s s
γ(s) = a cos , a sin
a a

is a leftward-turning parametrization of a circle of radius a, and its velocity

vector
 
s s
γ 0 (s) = − sin , cos
a a

has norm 1, so s is an arc-length parameter. By (2.4), the unit normal

vector is
 
s s
n(s) = − cos , − sin
a a

and we have
 
00 1 s s 1
γ (s) = − cos , − sin = n(s).
a a a a
Thus a leftward-turning circle of radius a (> 0) has constant curvature
κ(s) = 1/a. When the circle is rightward-turning and of radius a (> 0),
the curvature would be −1/a.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 14

14 Differential Geometry of Curves and Surfaces

When the parameter is not an arc-length parameter, we can find the

curvature function by changing to an arc-length parameter. However, an
arc-length parameter is not written in terms of elementary functions gener-
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

ally, so we will also need a formula for the curvature that applies when the
parameter is not necessarily arc-length. If t is any parameter for a regular
curve γ(t) = (x(t), y(t)), the curvature function κ(t) is
ẋÿ − ẏẍ det(γ̇, γ̈)
(2.7) κ(t) = = .
2
(ẋ + ẏ )2 3/2 |γ̇|3
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Proof of equation (2.7). Using (2.2) and the chain rule for differentiating
compositions of functions, we have
 0
0 γ̇ 00 γ̈ 1
γ = , γ = + γ̇.
|γ̇| |γ̇|2 |γ̇|
Substituting these into (2.6) and using the properties of determinants, we
arrive at the desired equation.

In particular, the regular curve given by the graph of a function y = f (x)

has curvature
d2 y
 
ÿ dy
(2.8) κ(x) = ẏ = , ÿ = .
(1 + ẏ 2 )3/2 dx dx2
Example 2.2. By (2.7), the ellipse γ(t) = (a cos t, b sin t) (a, b > 0) has
curvature
ab
(2.9) κ(t) = .
(a2 sin2 t + b2 cos2 t)3/2
When a > b, the curvature has a maximum value a/b2 at the points (±a, 0)
where the ellipse intersects the x-axis, and a minimum value b/a2 at the
points (0, ±b) where the ellipse intersects the y-axis.

The osculating circle. At a point on a regular curve, the line that comes
closest to “approximating” the curve near that point is the tangent line at
that point. Now, instead of a line, let us consider what circle would best
“approximate” the curve near a given point on the curve.
Let us take a point γ(s) on the curve where the curvature κ(s) is not
zero. We can then place a circle of radius 1/|κ(s)| so that it is tangent to
the curve at γ(s), and we can give the circle a parametrization so that its
parameter is increasing in the same direction as the s along γ(s). When
κ(s) > 0, we place the circle on the left-hand side of the curve, and when
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 15

Curves 15

κ(s) < 0, we place it on the right-hand side. We call this circle the oscu-
lating circle of the curve γ at the point γ(s); see Fig. 2.2. The osculating
circle has radius 1/|κ(s)| and center
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

1
γ(s) + n(s).
κ(s)
The locus of the center of osculating circles is called the evolute (or focal
curve or caustic) of γ(s). See Appendix B.1 for the construction of the
evolutes and relationships with the original curves.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 2.2 The osculating circles (lighter shaded) of curves (darker shaded).

At a point of γ(s) where the curvature is 0, we say that the tangent line
is the osculating circle (a circle with infinite radius).
We give the osculating circle a pa-
rametrization in the same direction as
the original curve. Then by Example
2.1, we see that the original curve and
the osculating circle will both have the
same curvature at the point where they
touch tangentially. If one is driving a
car, and fixes the steering wheel in the
position it is in at one moment, the car
will start driving along a circle. That
circle is the osculating circle of the path
of the car at the moment the steering Fig. 2.3
wheel was held fixed.
Theorem 2.4 on page 16 will show us that the osculating circle really
is the circle that best approximates the original curve at each point. But
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 16

16 Differential Geometry of Curves and Surfaces

first we must clearly explain what we mean by “the best approximating

circle”. As we are considering regular curves, γ(t) = (x(t), y(t)) has the
non-zero tangent vector γ̇(t0 ) when t takes the value t0 (cf. page 6). Let
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

us further suppose that γ(t0 ) is the origin and γ̇(t0 ) points in the direction
of the x-axis, Then the normal vector at t = t0 on the left-hand side of
the curve will point in the direction of the y-axis. We can thus write the
tangent vector as γ̇(t0 ) = (a, 0) (a := ẋ(t0 ) > 0), see Fig. 2.3. Therefore,
by what we stated in Section 1 (on page 6), for values of t close to t0 (i.e.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

for a portion of the curve near the origin), the curve can be written as a
graph y = f (x). We use these facts in the next definition.
Definition 2.3. If two curves γ1 (t) and γ2 (u) intersect at a point P so
that their derivatives point in the same direction, then we say that the two
curves have first order contact at P. In this case, taking P to be the origin,
and the curves to be moving in the direction of the positive x-axis, and
taking the normal vector to be on the left-hand side so that it points in the
direction of the positive y-axis at P, the two curves γ1 (t) and γ2 (u) given
above can each be written as graphs of functions y = f (x) and y = g(x) so
that
df dg
f (0) = g(0) = 0, (0) = (0) = 0.
dx dx
Then, if we further have
d2 f d2 g
(0) = (0),
dx2 dx2
we say that the two curves have second order contact (see the left-hand side
of Fig. 2.4).
We can similarly define the notion of higher order contact. For example,
if the f (x) and g(x) in Definition 2.3 have the same first, second and third
order derivatives at 0, we say that the two corresponding curves γ1 and
γ2 have third order contact. The parabola y = x2 /2 and the catenary
y = cosh x − 1 have third order contact at the origin. (See Fig. 2.4.)
Since a tangent line of a curve makes first order contact with that curve,
we say it is a first order approximation to the curve. Using the same
phrasing, the next theorem tells us that the osculating circle is a second
order approximation to the curve.
Theorem 2.4. The osculating circle at a point γ(t0 ) of a regular curve
γ(t) makes second order contact with the curve at that point. Conversely,
any circle that makes second order contact with γ(t) at γ(t0 ) must be the
osculating circle at that point.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 17

Curves 17
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Third order contact

Second order contact of two curves.
of a parabola and the catenary.

Fig. 2.4

Proof. Take a coordinate system (x, y) on the plane as in Definition 2.3 at

γ(t0 ), and then the curve can be represented as a graph y = f (x), where
df
f (0) = dx (0) = 0. So by (2.8), the Taylor expansion of f (x) at x = 0 is
κ(t0 ) 2
(2.10) f (x) = x + o(x2 ).
2
Here, o(·) is Landau’s symbol (cf. Appendix A.1).
The graph y = f (x) has second order contact with the x-axis (i.e., the
graph of y = 0) at the origin if and only if κ(t0 ) = 0. So, we consider only
the case κ(t0 ) 6= 0 from now on. By reflecting the curve across the x-axis
if necessary, we may assume κ(t0 ) > 0 without loss of generality. Here, the
circle of radius a turning to the left and
√ tangent to the x-axis at the origin
can be represented as a graph y = a − a2 − x2 , which can be expanded as
1 2
(2.11) y= x + o(x2 )
2a
by (2.10) and the fact that the curvature of the circle is 1/a. Comparing
(2.10) and (2.11), we can conclude that the curve and the circle have second
order contact at the origin if and only if a = 1/κ(t0 ).

The following two propositions give important properties, which will be

used in Section 4.

To test for the contact order of regular curves from the definition, one
has to express the curves as graphs. To avoid this procedure, the following
criteria is useful:
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 18

18 Differential Geometry of Curves and Surfaces

Proposition 2.5. Two regular curves γ1 (t) and γ2 (u) will make first order
(resp. second order) contact at a point P if and only if one can choose a
parameter u for γ2 so that at t = t0 and u = u0 ,
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

dγ1 dγ2
(2.12) γ1 (t0 ) = γ2 (u0 ) = P, (t0 ) = (u0 )
dt du
d2 γ1 d2 γ2
 
resp. additionally (t 0 ) = (u0 ) .
dt2 du2
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Proof. The ‘only if’ part is evident because a graph is a special case of a
parametrization.
We shall now show the ‘if’ part: We denote by “ ˙ ” differentiation with
respect to the parameter t, and “ 0 ” with respect to u (though the parameter
u is not necessarily arc-length, we use this notation to distinguish the two
variables t and u). For the first order contact, the conclusion is evident by
Definition 2.3. We show when the two curves have second order contact.
Set γ1 (t) = (x(t), y(t)), γ2 (u) = (x̃(u), ỹ(u)) and express the two curves
as graphs y = f (x), ỹ = g(x̃), respectively, as in Definition 2.3. Then
y(t) = f (x(t)) and ỹ(u) = g(x̃(u)) hold, and thus we have
df ẏ dg ỹ 0 d2 f ÿ ẋ − ẏẍ d2 g ỹ 00 x̃0 − ỹ 0 x̃00
= , = 0, = , = ,
dx ẋ dx x̃ dx2 ẋ3 dx 2 x̃03
by the chain rule. Hence
d2 f d2 g
(0) = (0)
dx2 dx2
if γ̇1 (t0 ) = γ20 (u0 ) and γ̈1 (t0 ) = γ200 (u0 ), that is, the two curves have second
order contact.

To test for higher contact order, it is sufficient to find certain

parametrizations and compare higher order derivatives (cf. (1) in Prob-
lem 9). The notion of contact order is invariant under (general) change of
coordinates, seen as follows:
For a planar region D ⊂ R2 , we consider a smooth map ϕ : D → R2 ,
written as ϕ(x, y) = (ξ(x, y), η(x, y)). We call this map a diffeomor-
phism, or a coordinate change, if ϕ is a one-to-one map whose inverse map
ϕ−1 : ϕ(D) → D is also smooth. When the Jacobian, that is, the determi-
nant of the Jacobi matrix ηξxx ηξyy is positive, ϕ is said to be orientation

preserving, or a positive coordinate change. In this case, the left-side re-

gion of the curve on the xy-plane is mapped to the left-side region in the
ξη-plane of the corresponding curve. On the other hand, if the Jacobian is
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 19

Curves 19

negative, ϕ is called orientation reversing, or a negative coordinate change.

For example, if we set x = r cos θ, y = r sin θ for r > 0 and θ, then (x, y)
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

is a point on the plane with polar coordinate (r, θ). This correspondence is
regarded as a map from the rθ-plane to the xy-plane as
ϕ : (r, θ) 7−→ (x, y) = (r cos θ, r sin θ).
If one takes a region D := {(r, θ) | r > 0, −π < θ < π} of this mapping, then
ϕ(D) is the xy-plane excluding the negative part of the x-axis, and ϕ gives
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

a bijection of D to ϕ(D). Hence there is an inverse map ψ : ϕ(D) → D.

Moreover, since the Jacobian of ϕ is
   
x x cos θ −r sin θ
det r θ = det = r > 0,
yr yθ sin θ r cos θ
ψ is smooth because of the inverse function theorem (Theorem A.1.5 in
Appendix A.1). Hence ϕ : (r, θ) 7→ (x, y) is an orientation preserving dif-
feomorphism.
Given a diffeomorphism ϕ from one region D to another region ϕ(D) ⊂
2
R , a curve γ(t) in D becomes the curve γ̃(t) = ϕ ◦ γ(t) in ϕ(D), where
ϕ ◦ γ denotes the composition of ϕ and γ, that is,
ϕ ◦ γ(t) = ϕ(γ(t)).
We show in the next proposition that the property of tangency of regular
curves is preserved under such a diffeomorphism ϕ. (See Fig. 2.5.)
Proposition 2.6. Let ϕ : D → ϕ(D) be a diffeomorphism defined on a
planar region D. Then if two regular curves γ1 (t) and γ2 (u) in the region
D have first order (resp. second order) contact at a point P in D, the image
curves γ̃1 (t) = ϕ◦γ1 (t) and γ̃2 (u) = ϕ◦γ2 (u) will also have first order (resp.
second order) contact, now at the point ϕ(P).

Proof. Writing ϕ as ϕ(x, y) = (ξ(x, y), η(x, y)), for the curve γ(t) =
(x(t), y(t)) in the region D we have the corresponding curve
γ̃(t) = ϕ ◦ γ(t) = (ξ(x(t), y(t)), η(x(t), y(t))).
Applying the chain rule,
dξ dη
= ξx ẋ + ξy ẏ, = ηx ẋ + ηy ẏ,
dt dt
d2 ξ
(2.13) = ξx ẍ + ξy ÿ + ξxx ẋ2 + 2ξxy ẋẏ + ξyy ẏ 2 ,
dt2
d2 η
= ηx ẍ + ηy ÿ + ηxx ẋ2 + 2ηxy ẋẏ + ηyy ẏ 2 .
dt2
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 20

20 Differential Geometry of Curves and Surfaces

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 2.5 Diffeomorphism and contact of curves: Two regular curves in the xy-plane
having second order contact are mapped to curves in the ξη-plane having second order
contact, under the diffeomorphism (x, y) 7→ ϕ(x, y) = (ξ, η). Here, the dotted lines in
the right-hand figure are the images under ϕ of dotted lines parallel to coordinate axes
in the left-hand figure.

Assume γ1 (t) := (x1 (t), y1 (t)) and γ2 (u) := (x2 (u), y2 (u)) have first order
contact at P = γ1 (t0 ) = γ2 (u0 ) so that the parameters t and u satisfy
(2.12) in Proposition 2.5. We write γ̃1 (t) = (ξ1 (t), η1 (t)) and γ̃2 (u) =
(ξ2 (u), η2 (u)), that is,

ξ1 (t) = ξ(x1 (t), y1 (t)), η1 (t) = η(x1 (t), y1 (t)),

ξ2 (u) = ξ(x2 (u), y2 (u)), η2 (u) = η(x2 (u), y2 (u)).

Then we have

ξ1 (t0 ) = ξ(x1 (t0 ), y1 (t0 )) = ξ(x2 (u0 ), y2 (u0 )) = ξ2 (u0 ),

η1 (t0 ) = η(x1 (t0 ), y1 (t0 )) = η(x2 (u0 ), y2 (u0 )) = η2 (u0 ),


by the first equality of (2.12), and hence γ̃1 (t0 ) = γ̃2 (u0 ) = ϕ(P) holds.
Moreover, using (2.13) and (2.12),
dξ1 dx1 dy1
(t0 ) = ξx (x1 (t0 ), y1 (t0 )) (t0 ) + ξy (x1 (t0 ), y1 (t0 )) (t0 )
dt dt dt
dx2 dy2
= ξx (x2 (u0 ), y2 (u0 )) (u0 ) + ξy (x2 (u0 ), y2 (u0 )) (u0 )
du du
dξ2 dη1 dη2
= (u0 ), and (t0 ) = (u0 )
du dt du
hold. Thus, we have dγ̃ dγ̃2
dt (t0 ) = du (u0 ). Hence by Proposition 2.5, γ̃1 and
1

γ̃2 have first order contact at ϕ(P). The case of second order contact can
be obtained in a similar way.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 21

Curves 21

A similar result can be stated for third or higher order contact as well
(see (2) in Problem 9 at the end of this section).
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Corollary 2.7. The curvature of a regular curve is invariant under rota-

tions and translations.
In addition to rotations and translations, a reflection about a line is a
congruence of the plane (cf. Appendix A.3). By a reflection, the absolute
value of the curvature is invariant, but the sign is reversed. See Problem 4.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Proof. By a rotation or a translation, a circle turning right (resp. left) is

mapped to a circle turning right (resp. left). Since rotations and transla-
tions are diffeomorphisms of R2 , the order of contact of a curve and a circle
is preserved by these transformations, because of Proposition 2.6. Hence
by Theorem 2.4, applying a rotation and a translation to a curve, the oscu-
lating circle is likewise mapped to the osculating circle of the image curve.
Since the radius of a circle is preserved by rotations and translations, the
conclusion follows.

Corollary 2.7 can be also proved by direct calculations (cf. Problem 6

of this section).

The Frenet formula. Given a regular curve γ(s) = (x(s), y(s)) param-
etrized by an arc-length parameter s, equation (2.5) can be rewritten as:
(2.14) x00 (s) = −κ(s) y 0 (s), y 00 (s) = κ(s) x0 (s).
From this and (2.3) and (2.4), we know that n0 (s) = −κ(s)e(s), so, together
with (2.5), we have that
(2.15) e0 (s) = κ(s)n(s), n0 (s) = −κ(s)e(s).
These two equations are called the Frenet formula.1
It is natural to ask, for a given function, whether there always exists a
regular curve whose curvature function is equal to the given function, and
whether it is unique if it exists. The following theorem is the answer to this
question.
Theorem 2.8 (The fundamental theorem for plane curves). For a smooth
function κ(s) defined for s in the interval 0 ≤ s ≤ l, there exists a regular
curve γ(s) with arc-length parameter s ∈ [0, l] and curvature κ(s). Further-
more, this curve is uniquely determined up to rotations and translations of
the plane.
1 Frenet, Jean Frédéric (1816–1900).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 22

22 Differential Geometry of Curves and Surfaces

Proof. Existence of the curve seems natural when one imagines driving a
car, steering in accordance with the curvature function, while driving at
unit speed.
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

We can give a precise proof as follows: Let γ(s) be a curve parametrized

by an arc-length parameter s, and e(s), n(s) the unit tangent vector and
the leftward unit normal vector, respectively. Then one can take a 2 × 2-
matrix-valued function F(s) := (e(s), n(s)), where we consider e(s) and
n(s) to be column vectors. Then the Frenet formula (2.15) is expressed as
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

 
0 −κ
(2.16) F 0 = FΩ, Ω := .
κ 0
This equation can be regarded as a linear ordinary differential equation
of the unknown matrix-valued function F(s), where κ(s) is a given data
(cf. Appendix A.2). Then there exists a unique solution F(s) of (2.16)
satisfying the initial condition
 
1 0
F(0) := (e(0), n(0)) = I, I := ,
0 1
because of Theorem A.2.2 in Appendix A.2. For each s, F(s) is an or-
thogonal matrix with determinant 1. In fact, since Ω is a skew-symmetric
matrix, Proposition A.3.9 in Appendix A.3 yields that
T
(2.17) (FF T )0 = F 0 F T + FF 0 = FΩF T + FΩ T F T = O,
where F T is the transposition of F, and O denotes the 2 × 2 zero matrix.
T
Hence F(s)F(s) does not depend on s, i.e., is a constant matrix:
T T
F(s)F(s) = F(0)F(0) = I.
This implies that F(s) is an orthogonal matrix for each s. Then the de-
terminant of F(s) is ±1 (cf. (A.3.11) in Appendix A.3), and by the fact
F(0) = I, we can conclude that the determinant of F(s) is 1 by continuity
of det F(s). Thus F(s) is an orthogonal matrix of determinant 1, and hence
n(s) points leftward with respect to e(s) (cf. (3) of Problem 2).
Let
Z s
γ(s) := e(t) dt.
0
Then γ(0) = 0 holds, and γ(s) is the desired curve. In fact, e(s) is the
unit tangent vector of γ(s) since γ 0 (s) = e(s). Moreover, since F(s) is
an orthogonal matrix of determinant 1, n(s) is the left-hand unit normal
vector. Then equation (2.16) for F(s) = (e(s), n(s)) is equivalent to the
Frenet formula for γ(s), and then κ(s) is a curvature function.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 23

Curves 23

Next we prove uniqueness: Let Γ (s) (0 ≤ s ≤ l) be a curve parametrized

by arc-length whose curvature function coincides with the curvature κ(s)
of γ(s). By Corollary 2.7, we may assume that
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

     
0 1 0
(2.18) γ(0) = Γ (0) = , e(0) = E(0) = , n(0) = N (0) =
0 0 1
by applying rotations and translations, without loss of generality. Here,
E(s) and N (s) are the unit tangent vector and the left-hand unit normal
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

vector of Γ (s), respectively. Then G := (E, N ) satisfies (2.16) as well as

F, and the initial values of F and G coincide. Thus by the uniqueness of
solutions of ordinary differential equations, G(s) = F(s) holds for each s.
Moreover,
Z s Z s
γ(s) = e(t) dt = E(t)dt = Γ (s),
0 0
that is, the two curves are the same.

Remark 2.9. For a given curvature function κ(s), the desired regular curve
γ(s) can be obtained explicitly as follows:
Z s Z t  Z t 
(2.19) γ(s) = cos κ(u) du , sin κ(u) du dt.
0 0 0

In fact, |dγ/ds| = 1 holds, that is, s is an arc-length parameter, and com-

puting d2 γ/ds2 , we have that κ(s) is the curvature of γ(s) (cf. Problem
7).
By Theorem 2.8 and Example 2.1 (page 13), we have the following
corollary.
Corollary 2.10. The curvature function of a regular curve is constant if
and only if the curve is a circle or a line.

The four-vertex theorem for convex curves. At the end of this sec-
tion we give an example of an application of the Frenet formula.
In Section 1, a curve was said to be closed when traveling along the
curve would bring you back to where you started. More precisely, closed
curves are defined as follows: If a curve γ(t) defined on the entire real line
R has a period l (> 0), that is,
(2.20) γ(t + l) = γ(t) (t ∈ R),
then the restriction of γ to the closed interval [0, l] is said to be a closed
curve. In other words, a curve γ : [a, b] → R2 is said to be a closed curve
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 24

24 Differential Geometry of Curves and Surfaces

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

A convex curve. A non-convex simple closed curve.

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 2.6 Convex and non-convex closed curves.

if γ can be extended to a periodic map with period l := b − a satisfying

(2.20). In this case, smooth periodicity
γ (n) (0) = lim+ γ (n) (t) = lim− γ (n) (t) = γ (n) (l)
t→0 t→l
for all non-negative integers n, holds, and the curvature function κ(t) is
also a periodic function on R with period l.
As mentioned in the last part of Section 1, a closed regular curve that
does not intersect itself is called a simple closed regular curve.
We say that a simple closed regular curve is convex (resp. strictly con-
vex ) if, for any two distinct points chosen on the curve, the line segment
between them is contained in the closure of the interior region (resp. the
interior region itself) (cf. Definition 1.5 in Section 1) of the curve (Fig. 2.6).
For example, circles and ellipses are strictly convex. Also, it is known that
any simple closed regular curve for which the curvature function does not
change sign (resp. does not have any zeros) is convex (resp. strictly convex).
See Theorem B.2.1 and Corollary B.2.2 in Appendix B.2.
A vertex is a point on a curve where the curvature κ(s) attains a local
maximum or minimum.2 On a closed regular curve which is not a circle, the
curvature function can be considered as a periodic function defined on the
entire real line, and it must have at least one point of maximum value and
one point of minimum value within one period. Furthermore, if, within one
period, there are a finite number of points where the maximum or minimum
is obtained, then the number of maximum points will equal the number of
minimum points. So if the closed regular curve has only a finite number of
vertices, then it will have an even number of vertices (see Fig. 2.7).
We have the following theorem about the vertices of convex curves.
2 In this text, we say that a function κ(s) takes a local maximum (resp. local minimum)

at s = s0 if there exists an interval [a, b] such that s0 ∈ (a, b), and κ(s0 ) is the maximum
(resp. minimum) of κ(s) on [a, b], and κ(a), κ(b) < κ(s0 ) (resp. κ(a), κ(b) > κ(s0 )).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 25

Curves 25
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

2 vertices 4 vertices
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 2.7 Vertices of closed curves.

Theorem 2.11. There exist at least four vertices on any strictly convex
curve that is not a round circle.

This theorem is called the four-vertex theorem, and was discovered by

Syamadas Mukhopadhyaya3 [27] in 1909. Later, in 1912, Adolph Kneser4
[18] showed that the theorem still holds even when the simple closed regular
curve is not assumed to be convex, and in Section 4 (Theorem 4.4) we
introduce the proof of that result.
Using the example of driving a car, the four-vertex theorem says this:
Driving once around a non-circular circuit with no intersections, one must
change the direction that one is turning the steering wheel at least four
times. As seen from Example 2.2, the ellipse has exactly four vertices, so
we know that the value 4 in Theorem 2.11 is the best value that can be
obtained.
On the other hand, there are closed regular curves (with self-intersec-
tions) whose number of vertices is 2. For example, the lemniscate (Example
1.3 of Section 1) has two vertices. See the left side of Fig. 2.7 and (3) in
Problem 5 in Section 4.

Proof of Theorem 2.11. We argue by way of contradiction. Suppose that

the number of vertices is less than 4. Since the number of vertices is even,
the number is less than or equal to 2. Here, since there are maximum and
minimum values of the curvature function, the number of vertices must be
2. We represent the closed curve by γ(s), 0 ≤ s ≤ l, for an arc-length
parameter s. Letting γ(0) and γ(v), 0 < v < l, denote the two vertex
points, and we may assume without loss of generality that the curvature
3 Mukhopadhyaya, Syamadas (1866–1937).
4 Kneser, Adolf (1862–1930).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 26

26 Differential Geometry of Curves and Surfaces

function κ(s) attains a maximum at s = 0 and a minimum at s = v. Then

κ0 (s) ≤ 0 for s in the interval [0, v], and κ0 (s) ≥ 0 for s ∈ [v, l]. In the
case that κ0 (s) is everywhere 0, the function κ(s) would then be a constant.
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Then the closed curve would be a circle (see Corollary 2.10), contradicting
our assumption.
Let
(2.21) ax + by + c = 0 (a, b, c are constants)
be the equation for the straight line from γ(0) to γ(v). As the curve is con-
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

vex, the line segment between these two points must lie in the interior region
of the curve (cf. Definition 1.5), and then the curve γ(s) = (x(s), y(s)) is
split into two parts by the line segment, one part where 0 ≤ s ≤ v lying to
one side, and the other part where v ≤ s ≤ l lying to the other side. The
function ax(s) + by(s) + c is non-negative on one of the intervals [0, v] and
[v, l], and is non-positive on the other. Hence κ0 (s)(ax(s) + by(s) + c) does
not change sign on 0 ≤ s ≤ l, and is not identically zero. Then the integral
Z l
J := κ0 (s) (ax(s) + by(s) + c) ds
0
will not be zero. On the other hand, integrating by parts and using (2.14)
and (2.20), we have
Z l
J =− κ(s) (ax0 (s) + by 0 (s)) ds
0
Z l
s=l
= (−ay 00 (s) + bx00 (s))ds = [−ay 0 (s) + bx0 (s)]s=0 = 0.
0
The last equality above holds because the curve is closed. This gives us a
contradiction, proving the result.

Exercises 2
1 An n × n matrix A is called an orthogonal matrix if AT is the inverse matrix
of A, where AT denotes the transposition of A (cf. Appendix A.3). When
setting n = 2, show that a 2 × 2 orthogonal matrix can be written in the form
 
cos α − sin α
(when the determinant is +1),
sin α cos α
 
cos α sin α
(when the determinant is −1),
sin α − cos α

where α is a real number.

March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 27

Curves 27

2 (1) For a unit vector e in the plane, let n be the unit vector perpendicular
to e. Show that the matrix (e, n) (the 2 × 2 matrix whose first column
is e and second is n) is an orthogonal matrix (cf. Appendix A.3).
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(2) Let A = (e, n) be an orthogonal matrix, where e and n are column

vectors. Show that {e, n} is a pair of unit vectors that are perpendicular,
i.e., form an orthonormal basis.
(3) Continuing from part (2), show that the determinant of A is +1 if and
only if n is obtained by counterclockwise 90 degree rotation of e.
3 Consider a car whose axles are pictured on
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

the right, with axle length ε, and with dis-

tance ∆ between axle midpoints, and with
front axle turned by an angle θ from the
forward direction. Compute the curvature
of the path that the tires will trace out as
the car moves.
4 Show that reflecting a regular curve across a straight line results in changing
the sign of its curvature.
5 For the cycloid γ(t) = a(t − sin t, 1 − cos t) (0 ≤ t ≤ 2π) as in Problem 4 of
Section 1, do the following:
(1) Compute the curvature function using the formula (2.7).
(2) For the interval [0, t], find an arc-length parameter s(t) and its inverse
function t(s).
(3) Writing γ(t) as γ(s), in terms of its arc-length parameter, compute the
curvature. And verify that it coincides with the result of (1).
6 Show by direct computation that the curvature of a regular curve is not
changed by translations and rotations.
7 For the curve γ(s) given in (2.19) in the proof of Theorem 2.8, confirm that
s is an arc-length parameter and that κ(s) is the curvature.
8 Let e(s) be the unit tangent vector to a curve γ(s) with arc-length parameter
s. Let n(s) be the unit normal vector to the left-hand side of e. Notating the
curvature by κ(s), show that for s near 0, we have

s2
γ(s) =γ(0) + se(0) + κ(0)n(0)
2
s3
+ (−κ(0)2 e(0) + κ0 (0)n(0)) + o(s3 ).
6
Here o is Landau’s symbol as in Appendix A.1.
∗
9 (1) Show that two regular curves γ1 (t), γ2 (u) have third order contact at P
if and only if one can choose a parametrization of γ2 such that γ1 (t0 ) =
γ2 (u0 ) = P and

dγ1 dγ2 d2 γ 1 d2 γ 2 d3 γ 1 d3 γ 2
(t0 ) = (u0 ), 2
(t0 ) = 2
(u0 ), 3
(t0 ) = (u0 )
dt du dt du dt du3
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 28

28 Differential Geometry of Curves and Surfaces

hold.
(2) Let ϕ : D → ϕ(D) be a diffeomorphism from a region D to ϕ(D) in the
plane. Show that for two curves γ1 and γ2 in D having third order contact
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

at P ∈ D, ϕ ◦ γ1 and ϕ ◦ γ2 have third order contact at ϕ(P).

(3) Show that a curve γ(t) has a third order contact at γ(t0 ) with its os-
culating circle at t0 if and only if the derivative κ̇(t0 ) of the curvature
vanishes.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

3. Closed curves*
Here we investigate the properties of closed regular curves that do not
change under continuous deformations.
Given two closed regular curves, if there exists a smooth deformation
from one to the other so that, at each moment of the deformation, the curve
at that moment has a nowhere vanishing velocity vector, then we say that
the two regular curves are equivalent or regular homotopic to each other.
More precisely, we define two closed regular curves to be regular homotopic
as follows: Changing parameter if necessary, two closed regular curves γ1 (s)
and γ2 (s) can be parametrized by the same parameter s (0 ≤ s ≤ l). Then,
γ1 and γ2 are regular homotopic if and only if there exists a family of closed
curves σt : [0, l] → R2 (0 ≤ t ≤ 1) so that:
(i) For each t, σt (s) (0 ≤ s ≤ l) is a smooth closed curve whose velocity
vector dσds (s) never vanishes.
t

(ii) σ0 (s) = γ1 (s) and σ1 (s) = γ2 (s). In other words, when t moves from
0 to 1, the curves σt move from γ1 to γ2 .
(iii) σt (s) is continuous with respect to both t and s.
Here we do not assume that s is an arc-length parameter.
This deformation gives an equivalence relation for closed regular curves,
and every closed regular curve lies in one of the classes shown in Fig. 3.1.
In order to check that this is so, we need to introduce the notion of rotation
index for closed regular curves.

The rotation index. For an arc-length parametrized closed curve γ(s)

(0 ≤ s ≤ l) with curvature function κ(s), we call
Z l
Tγ
Tγ := κ(s) ds, iγ :=
0 2π
the total curvature and the rotation index of the curve γ, respectively.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 29

Curves 29
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

iγ = 0 iγ = ±1 iγ = ±2 iγ = ±3 ...

Fig. 3.1 Representatives of regular homotopy classes of closed regular curves.

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Proposition 3.1. The rotation index iγ of a closed regular curve γ is

always an integer.

Proof. Parametrizing the closed regular curve as γ(s), (0 ≤ s ≤ l), it can

be represented as in (2.19) in Remark 2.9:
Z s
γ(s) = (cos θ(t), sin θ(t)) dt,
0

where
Z s
(3.1) θ(s) := κ(u) du.
0

This gives γ 0 (s) = (cos θ(s), sin θ(s)), and so θ(s) gives us the angle between
γ 0 (s) and γ 0 (0) = (1, 0). Here, because we have the condition (2.20) for
closed curves, which implies γ 0 (l) = γ 0 (0), it follows that θ(l) − θ(0) is an
integer multiple of 2π. Therefore iγ is an integer.

Because the rotation index is an integer, it cannot change under con-

tinuous deformations of a closed regular curve described in the beginning
of this section. So if two closed regular curves are regular homotopic, then
they have the same rotation index. However, if we allow deformations un-
der which the velocity vector can become zero at some point (that is, a
singular point) of some curve during the deformation, the curvature cannot
be defined at that point, and it is then possible for the winding number to
change.
For example, Fig. 3.2 shows a deformation, where the velocity vector
becomes zero at one point in the curve (iii), called the cardioid. The initial
curve (i) of this deformation is a counterclockwise wrapping circle with
rotation index 1, but the final curve (v), which is called the limaçon, has
rotation index 2 and so has a different rotation index than the initial curve.
Denote by S 1 the unit circle in the plane, that is, the circle centered at
the origin with radius 1. For a given regular curve γ(s) parametrized by
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 30

30 Differential Geometry of Curves and Surfaces

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(i) σ0 (ii) (iii) σ1/2 (iv) (v) σ1

Fig. 3.2 A deformation of closed curves admitting a singular point:
σt (s) := (1 − 2t sin s)(cos s, sin s) (0 ≤ s ≤ 2π, 0 ≤ t ≤ 1).
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

arc-length, the velocity γ 0 (s) is a unit vector. Therefore, if we translate the

velocity vector so that its tail is at the origin, then we have γ 0 (s) ∈ S 1 . In
this way, we have the mapping
(3.2) [0, l] 3 s 7−→ γ 0 (s) ∈ S 1 ,
which we call the Gauss map 1 of the curve, where S 1 denotes the unit
circle centered at the origin of R2 . By the proof of Lemma 3.1, when we
travel once around a closed regular curve, the number of times that the
Gauss map will travel around the unit circle is equal to the rotation index.
Imagine that the closed regular curve γ(t) is a closed circuit, and a car
is driving along the curve such that the position of the car at the time t is
γ(t). Then the Gauss map is a meter which measures the direction in the
xy-plane at time t. The rotation index is the number of times this meter
loops around when the car traverses the entire γ(t) once. If the number
iγ is positive (resp. negative), the meter turns iγ times (resp. |iγ | times)
counterclockwise (clockwise).
In particular, for the case of simple closed curves, we have the following
theorem.
Theorem 3.2. Every simple closed regular curve has rotation index either
1 or −1.

Proof. Although this appears to be a rather obvious assertion, the proof is

not so simple. The following proof is due to [11]. Translate a line parallel
to the x-axis from far below in the plane in the upward direction, and let d
be such a line which touches the curve for the first time, and let P be one
of the tangent points in the curve (see Fig. 3.3).
Without loss of generality, we can take an arc-length parameter s (0 ≤
s ≤ l) for γ(s) so that γ(0) = γ(l) = P. Changing the direction of the curve
if necessary, we may assume that γ 0 (0) = (1, 0). Defining the map w on a
closed domain D := {(s, t) | 0 ≤ s ≤ t ≤ l} in the st-plane by
1 Gauss, Carl Friedrich (1777–1855).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 31

Curves 31
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 3.3



 γ 0 (s) (when s = t),


 0

−γ (0) (when s = 0 and t = l),
w(s, t) :=
γ(t) − γ(s)


(otherwise),



 |γ(t) − γ(s)|

this w is a continuous map from D to the unit circle S 1 . Then there exists
a unique continuous function Θ on D such that
(3.3) w(s, t) = (cos Θ(s, t), sin Θ(s, t)), Θ(0, 0) = 0.
The function Θ(s, t) represents the angle of the vector w(s, t) with (1, 0).
Since the angle has the ambiguity of integer multiples of 2π, it cannot
be determined uniquely, in general. But if we allow the angle Θ(s, t) to
take arbitrary real number values beyond the interval [0, 2π], and choose
values Θ(s, t) which vary continuously in (s, t), then the function Θ(s, t)
satisfies (3.3). (The existence of this Θ is usually shown using the notion
of covering spaces, and the fact that D is simply connected (cf. [8, 35]).
However, noticing that w(s, t) is of class C 1 , one can show the existence of
the required function Θ without using the notion of covering spaces. See
Problem 4.)
Since γ is a closed curve, γ(l) = γ(0) holds. So we have w(s, l) =
−w(0, s), and we have that Θ(s, l) − Θ(0, s) = π × (an odd integer). Since
the function Θ(s, l) − Θ(0, s) of s is continuous, there exists an integer n
independent of s so that we can write
(3.4) Θ(s, l) − Θ(0, s) = (2n + 1)π (0 ≤ s ≤ l).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 32

32 Differential Geometry of Curves and Surfaces

On the other hand, because w(s, s) = γ 0 (s), and because Θ(0, 0) = 0

(cf. (3.3)), we have that
Z s
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(3.5) Θ(s, s) = κ(u) du

0
holds (see (3.1) on page 29). Therefore, by (3.5), the definition of the
rotation index and (3.4), we have
(3.6) 2πiγ = Θ(l, l) − Θ(0, 0) = (Θ(0, l) + (2n + 1)π) − Θ(0, 0)
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

= (Θ(0, l) + (2n + 1)π)

+ (Θ(0, l) − Θ(0, 0) − (2n + 1)π) − Θ(0, 0)
= 2(Θ(0, l) − Θ(0, 0)).
Now, because of the way we have chosen P, we can take a half-line extending
out from P in the direction of the negative y-axis, and the closed curve will
never intersect that half-line except at P itself. Hence w(0, s) will never
become (0, −1) (see Fig. 3.3). Hence we have
π
Θ(0, s) + 6= 2πm (m = 0, ±1, ±2, . . . ).
2
Then, noticing that Θ(0, 0) = 0, the function Θ(0, s) in s satisfies
π 3π
−
< Θ(0, s) < (0 ≤ s ≤ l) ,
2 2
because it is continuous. Moreover, by (3.4), Θ(0, l) − Θ(0, 0) = π holds.
Then by (3.6), we have iγ = 1.

Theorem 3.3 (Whitney’s Theorem2 [42]). A necessary and sufficient con-

dition for two closed regular curves γ1 (s) and γ2 (s) to be regular homotopic
is that they have the same rotation index.

Proof. The necessity of the equality of the rotation indices is clear. Con-
versely, suppose that the two closed regular curves γ1 and γ2 have the same
rotation index:
iγ1 = iγ2 = m.
We will construct a family σt (s) of curves satisfying the conditions (i)–(iii)
for regular homotopy. Because homotheties do not change the rotation
index, we may assume without loss of generality that both curves have
length 1 and have arc-length parametrizations γ1 (s), γ2 (s) (0 ≤ s ≤ 1).
Letting κ1 (s) and κ2 (s) be the respective curvatures of γ1 and γ2 , from
2 Whitney, Hassler (1907–1989).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 33

Curves 33

equation (2.19) in the proof of the fundamental theorem for plane curves
(Theorem 2.8), we see that we may assume γj (0) = 0 and
Z s Z s
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

γj (s) = (cos θj (u), sin θj (u)) du, θj (s) := κj (u) du (j = 1, 2).

0 0
Since κ1 and κ2 are periodic functions with period 1, from the definition of
the rotation index, we have
Z 1 Z s+1
(3.7) θj (s + 1) = κj (u) du + κj (u) du = 2πm + θj (s) (j = 1, 2).
0 1
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Here, we set
(3.8) ϕt (s) := (1 − t)θ1 (s) + tθ2 (s) (0 ≤ t ≤ 1).
By (3.7), it holds that ϕt (s + 1) = ϕt (s) + 2πm. Then defining
v t (s) := (cos ϕt (s), sin ϕt (s)),
Z 1
ṽ t (s) := v t (s) − v t (u) du,
0
Z s
σt (s) := ṽ t (u) du,
0
one can show that, for each t, v t (s), ṽ t (s) and σt (s) are periodic in s with
period 1, and then σt is a closed curve. Because γj (0) = 0 (j = 1, 2), we
have ṽ t = v t for t = 0, 1, and it follows that σ0 = γ1 , σ1 = γ2 . That is,
σt is a deformation from γ1 to γ2 . In other words, σt (s) satisfies the first
part of the condition (i) and the conditions (ii) and (iii) of the definition of
regular homotopy. So it is sufficient to show that σt (s) satisfies the second
part of the condition (i), that is, ṽ t = dσt /ds does not vanish anywhere in
the deformation. By definition, we have
Z 1
dσt (s)
= ṽ t (s) = v t (s) − v t (u) du,
ds 0
and the triangle inequality for integrals (cf. Theorem A.1.4 in Appendix
A.1) gives
Z 1 Z 1
(3.9) v t (u) du ≤ |v t (u)| du = 1.
0 0
In order for equality to hold, v t (s) (0 ≤ s ≤ 1) must be a constant vector.
(Here, we used the fact that |v t | = 1.)
When the rotation index m is not 0, (3.8) implies that ϕt moves from
0 to 2πm when s moves 0 to 1, and then v t cannot be a constant vector.
Consequently, equality does not hold in (3.9), implying
Z 1
dσt
≥ |v t | − v t (u) du > |v t | − 1 = 0,
ds 0
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 34

34 Differential Geometry of Curves and Surfaces

and so we know dσt /ds does not vanish for all s.

Next, we consider the case that the rotation index is m = 0. If v t is a
constant vector, then the fact ϕt (0) = 0 implies that ϕt (s) = (1 − t)θ1 (s) +
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

tθ2 (s) is zero for all s. In particular, two functions θ1 and θ2 differ by only
constant multiplication. If the curvatures of γ1 , γ2 are identically zero, they
cannot be closed curves. Hence θ1 and θ2 are not constant functions. If θ1
is proportional to θ2 , translating the parameter s of γ2 if necessary, one can
change θ2 with an additional constant, and then cause θ1 and θ2 not to be
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

proportional. Thus, the vector v t is not constant in t. Hence in a similar

way to the case above, one can show dσt /ds 6= 0.

By Theorem 3.3, we can find the rotation index of a given curve without
computation. In particular, for the curves in Fig. 3.1, the leftmost curve
is regular homotopic to the lemniscate and so has rotation index 0, by
Problem 1. Also, the other curves in that figure are regular homotopic to
curves that wrap finitely many times about a circle, so their rotation indices
can be found as well. Also, for closed regular curves γ(s) (0 ≤ s ≤ l) that
are C 1 , we have existence of a continuous function θ : [0, l] → R so that
γ 0 (s) = (cos θ(s), sin θ(s)) holds, and so we can define the rotation index by
2πiγ := θ(l) − θ(0). Then, by adjusting the proofs of Theorems 3.2 and 3.3
a bit, we can reprove those results for the case of C 1 -regular curves.

Corollary 3.4. A simple closed curve is regular homotopic to the circle.

Proof. A circle turning right (resp. left) has the rotation index 1 (resp. −1).
On the other hand, a given simple closed curve γ has the rotation index 1
or −1 by Theorem 3.2. Hence by Theorem 3.3, it is regular homotopic to
a circle.

Advanced Topic: Whitney’s formula. Though it still looks difficult

to find the rotation index for a given closed curve, the following Whitney’s
formula (Theorem 3.5) provides a way to calculate the rotation index by
counting self-intersections of the curve.

A point where a closed curve crosses itself is called a self-intersection.

A closed regular curve is said to be generic if it has a finite number of
self-intersections, and for each self-intersection, two pieces of the curve are
crossing transversely (are not tangent each other) (Fig. 3.4). An arbitrary
closed regular curve can be deformed to a generic curve by a small defor-
mation.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 35

Curves 35
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Generic curves. Non-generic curves.

Fig. 3.4
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Take a point P on an oriented generic curve. When the curve passes

through a self-intersection Q for the first time after starting at P, if another
piece of the curve crosses from left to right (resp. right to left) with respect
to the direction of travel, the self-intersection Q is called a positive self-
intersection (resp. negative self-intersection) with respect to the base point
P (see Fig. 3.5).

A positive self-intersection. A negative self-intersection.

Fig. 3.5

Then we denote the sign of a point Q on γ by


if Q is a positive self-intersection
 
 +1 ,
with respect to the base point P






if Q is a negative self-intersection
 
sgnγ (Q) = sgnP,γ (Q) := −1 ,

 with respect to the base point P




0 (if Q is not a self-intersection) .
The sign of a self-intersection depends on the choice of base point P, and
the sign is reversed when the orientation of the curve is reversed.
The following is a counting formula for the rotation index:
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 36

36 Differential Geometry of Curves and Surfaces

Theorem 3.5 (Whitney’s formula [42]). Take a point P on a generic closed

curve γ such that the image of the curve lies to the left side of the tangent
line of γ at P. (Changing the orientation of the curve, if necessary, one
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

can always find such a point.) Then the rotation index iγ of γ is obtained
by
X
iγ = 1 + sgnP,γ (Q).
Q∈γ
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

To prove the theorem, we prepare the following lemma:

Lemma 3.6. Assume an oriented generic closed curve γ and an oriented
simple closed regular curve σ cross transversely (are not tangent) at a finite
number of points, and all self-intersections of γ do not lie on σ. At an
intersection Q of these two curves, if γ crosses σ from left to right (resp.
right to left) with respect to the direction of σ, we set sgnγ,σ (Q) = +1
(resp. sgnγ,σ (Q) = −1). Then
X
sgnγ,σ (Q) = 0.
Q∈γ∩σ

Proof. By the Jordan closed curve theorem (cf. page 9), the simple closed
curve σ divides the plane into two regions, the interior and the exterior. For
each adjacent pair of self-intersections, if the curve γ travels from the inte-
rior (resp. exterior) to the exterior (resp. interior) at the first intersection,
then at the next intersection, it travels from the exterior (resp. interior) to
the interior (resp. exterior). Noticing that the number of intersections is
even, the conclusion follows.

Proof of Theorem 3.5. When γ is a simple closed curve, it can be oriented

so that the left side of the curve is the interior of the curve, according to the
way the point P is chosen. Then we have iγ = 1, which is the conclusion.
We prove the conclusion by induction with respect to the number of
self-intersections. Assume the conclusion holds for curves with n − 1 self-
intersections, and take a closed curve γ having n self-intersections. We can
find a self-intersection Q1 on the curve γ such that the part of the curve
starting at Q1 and returning again to Q1 is without self-intersection. By
rounding the angle smoothly (see Proposition B.5.5 of Appendix B.5 for
the way to round the angle) at Q1 , this closed subarc of γ turns out to
be a smooth simple closed regular curve, which is denoted by γ1 . We take
the first Q1 among such self-intersections starting from the base point P
(Fig. 3.6). Then γ1 does not pass through self-intersections between P and
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 37

Curves 37
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 3.6

Q1 . In fact, if there exists such a self-intersection, the point has the same
property as Q1 , which contradicts the way we chose Q1 . Moreover, γ1 does
not pass through the base point P, because the curve passes through each
self-intersection twice before returning to P.
On the other hand, let γ2 be a closed regular curve which is on the
opposite side of γ1 where we rounded the vertical angle at Q1 . The integrals
of the curvature on the rounded angles are equal to the increment of the
angles at the rounded parts (cf. (3.1)). Since these angles at the rounded
parts are equal, these increments on γ1 and γ2 cancel, which implies iγ =
iγ1 + iγ2 . Since the rotation index of the simple closed regular curve γ1 is
the sign of the self-intersection Q1 , by the inductive assumption, we have
iγ = iγ1 + iγ2 = sgnP,γ (Q1 ) + iγ2
X
= sgnP,γ (Q1 ) + 1 + sgnP,γ2 (Q).
Q∈γ2
By Lemma 3.6, the sum of the signs of intersections of γ2 with γ1 vanishes.
So we have X
sgnγ2 ,γ1 (R) = 0.
R∈γ1 ∩γ2
Here, γ1 does not pass through self-intersections between P and Q1 . Thus
for each intersection R of γ2 with γ1 , it holds that sgnP,γ (Q) = sgnγ2 ,γ1 (R).
Hence
X X X
sgnP,γ (Q) = sgnP,γ (Q1 ) + sgnγ2 ,γ1 (R) + sgnP,γ2 (S)
Q∈γ R∈γ1 ∩γ2 S∈γ2
X
= sgnP,γ (Q1 ) + sgnP,γ2 (S) = iγ − 1,
S∈γ2
which proves the theorem.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 38

38 Differential Geometry of Curves and Surfaces

Example 3.7. We compute the rotation index for a concrete example.

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 3.7

The left side of Fig. 3.7 has 3 self-intersections. Taking P as a base point,
the signs of these self-intersections are +, −, +, in order of appearance.
Thus by Theorem 3.5, the rotation index is 1 + (1 − 1 + 1) = 2. On the
other hand, the rotation index of the right side figure is 2, because the
sign of the self-intersection is +. Therefore these two curves are regular
homotopic, by Theorem 3.3. One can verify this fact by drawing pictures
of a smooth deformation.

Exercises 3
1 Compute the rotation indices of the ellipse and the lemniscate from the defi-
nition of the rotation index.
2∗ Assume a C ∞ -function ϕ(t) defined on R satisfies ϕ(0) = 0. If we set
(
ϕ(t)/t (when t 6= 0),
ψ(t) :=
ϕ̇(0) (when t = 0),

then show that ψ(t) is a C ∞ -function. Here, ϕ̇ := dϕ/dt.

3∗ (1) For a C ∞ -function f (t) defined on R, we set

 f (t) − f (s) (when s 6= t),
F (s, t) := t−s
f˙(s) (when s = t).

Show that F is a C ∞ -function on R2 using Problem 2.

(2) Let γ(s) (0 ≤ s ≤ l) be a simple closed curve and consider it as an
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 39

Curves 39

l-periodic curve defined on R. If we set

γ(t) − γ(s) |t − s|

(s − l < t < s + l, s 6= t),
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.


t−s |γ(t) − γ(s)|






0
  
γ 0 (s) = γ (s)


(s = t),

w̃(s, t) = |γ 0 (s)|

 γ(t) − γ(s) |t − (s + l)|



− (s + l < t < s + 2l, s 6= t),
 t − (s + l) |γ(t) − γ(s)|


Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com



 0
−γ (s) (s + l = t),

show that w̃ is C ∞ on the domain {(s, t) | s − l < t < s + 2l}, and the
restriction of w̃ to D = {(s, t) | 0 ≤ s ≤ t ≤ l} coincides with the w in the
proof of Theorem 3.2.
4∗ For a positive constant l, let w : D → R2 be a C ∞ -map w : D → R2 defined
on a closed domain D = {(s, t) | 0 ≤ s ≤ t ≤ l} of the st-plane satisfying
|w(s, t)| = 1 and w(0, 0) = (1, 0). Show, in the following way, that there exists
a continuous function Θ on D satisfying w(s, t) = (cos Θ(s, t), sin Θ(s, t)) and
Θ(0, 0) = 0.

(1) Let e : I 3 s 7−→ e(s) ∈ R2 be a C ∞ -map defined on an interval I

containing the origin and satisfying |e(s)| = 1 for all s ∈ I. If e(0) =
(cos α, sin α) for a constant α ∈ R, show that the function θ(s) defined
by Z s
θ(s) := det(e(u), e0 (u)) du + α
0

satisfies e(s) = (cos θ(s), sin θ(s)).

(2) Show that the function
Z t  
∂w
θ(t) := det(w(0, v), w2 (0, v)) dv w2 :=
0 ∂t

defined on the interval 0 ≤ t ≤ l satisfies

w(0, t) = (cos θ(t), sin θ(t)).

(3) Let

Z s Z t
Θ(s, t) = det(w(u, t), w1 (u, t)) du + det(w(0, v), w2 (0, v)) dv
0
0 
∂w ∂w
w1 := , w2 :=
∂s ∂t

for (s, t) ∈ D. Show that Θ(s, t) is the desired function.

March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 40

40 Differential Geometry of Curves and Surfaces

4. Geometry of spirals*
We first introduce two representative examples of spiral curves, the
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

Archimedean spiral and the logarithmic spiral.

The Archimedean spiral. Taking polar coordinates (r, θ) for the plane,
the curve satisfying

r = aθ (a > 0 is a constant, and θ > 0)

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

is called the Archimedean spiral (see Fig. 4.1, left side). If one were to
imagine a horizontal plane rotating about one fixed point at a constant
rate, and if one were to take a nail and press it gently onto that plane by
starting at the fixed point and moving at constant speed out in one un-
changing direction, one would scratch out such a spiral. In ancient Greece,
Archimedes1 investigated the properties of this spiral curve (see Problem 1
at the end of this section).

Fig. 4.1

The logarithmic spiral. In polar coordinates (r, θ), the curve given by

r = aθ (a > 1 is a constant, θ ∈ R)

is called the logarithmic spiral. When drawing a half-line out from the
origin, the half-line will always intersect the spiral at the same angle, re-
gardless of choice of direction of the half-line (see the right-hand side of
Fig. 4.1). One can find the logarithmic spiral in nature, for example, in
spiral shells or in the shape of galaxies. Similarity transformation of the
1 Archimedes (B.C. 287–212).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 41

Curves 41

logarithmic spiral is congruent to the original spiral (this property is called

self-similarity; cf. Problem 2 at the end of this section).
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

In this text, a regular curve is called a spiral if the derivative of the

curvature function never vanishes. A spiral is called a positive spiral (resp.
negative spiral ) if the curvature function is increasing (resp. decreasing),
see Fig. 4.2.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

A positive spiral. A negative spiral.

Fig. 4.2

The mirror image of a positive spiral is a negative spiral. On the other

hand, a positive spiral cannot be superimposed on a negative spiral by a
congruent transformation of R2 unless reflection is allowed, because ro-
tations and translations preserve the curvature function (Corollary 2.7 in
Section 2).
Imagine driving on a spiral road. One can say that
a positive spiral: keep turning the wheel to the left,
a negative spiral: keep turning the wheel to the right,
as a demonstration of what the spirals are. In fact, when one stops turning
the wheel, the car moves in a straight line or a circle, and to drive along a
spiral, one must keep turning the wheel. For example, look at the positive
spiral on the left side of Fig. 4.2: Starting from the state where the wheel is
turned to the right side, we could turn the wheel to the left little by little.
Then the curvature is negative at first, and it increases to 0, the inflection
point. When we continue to turn the wheel to the left, the curvature
becomes positive.
It should be remarked that the positivity and negativity of spirals does
not depend on the direction of the curves (Fig. 4.2). In fact, let γ(s)
(0 ≤ s ≤ l) be a positive spiral parametrized by the arc-length s. Then
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 42

42 Differential Geometry of Curves and Surfaces

the curvature function κ(s) is increasing. Here, consider the curve γ̃(s) :=
γ(l − s) (0 ≤ s ≤ l) obtained by changing the direction of γ(s). Then by
(2.6) on page 13, the curvature function κ̃(s) of γ̃(s) is
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

κ̃(s) = det(γ̃ 0 (s), γ̃ 00 (s)) = − det(γ 0 (l − s), γ 00 (l − s)) = −κ(l − s),

which is an increasing function, because κ(l − s) is decreasing. That is, the

curve obtained by changing direction of a positive spiral is also a positive
spiral (cf. Problem 3).
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

The following theorem holds (Fig. 4.3).

Theorem 4.1. Let γ(s) be a positive (resp. negative) spiral, and let the
osculating circle at each point of the spiral have the same orientation as the
spiral. Then γ(s) tangentially crosses the osculating circle from the circle’s
right side to its left (resp. left to right). In particular, if γ is positively curved
at s, then γ moves from outside (resp. inside) to inside (resp. outside) of
the circle.

A positive spiral. A negative spiral.

Fig. 4.3 Spirals and osculating circles.

Proof. A driver traveling along a positive (resp. negative) spiral keeps turn-
ing the wheel to the left (resp. right). If he would stop turning the wheel,
then the car follows the osculating circle at the point. Then one can see
intuitively that the trajectory of the car crosses the circle from right to left
(resp. left to right).
We give a rigorous proof: For a given positive spiral γ(s) and a point
γ(s0 ) of the spiral, choose an xy-coordinate system such that the point γ(s0 )
is the origin, the unit tangent vector e(s0 ) := γ 0 (s0 ) is the direction of the
x-axis, and the direction of the unit normal vector n(s0 ) is the direction
of the y-axis. Then the curve can be expressed as a graph y = f (x) (cf.
page 19). In particular, since f (0) = 0 and f˙(0) := dx df
(0) = 0 hold, (2.8)
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 43

Curves 43

implies that the curvature and the derivative of the curvature at the origin
are expressed as
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

f¨(0) ...
κ(0) = = f¨(0), κ̇(0) = f (0),
(1 + f˙(0)2 )3/2
where the dot means the derivative with respect to x. Then by Taylor’s
formula (Theorem A.1.1 in Appendix A.1),
1 1
κ(0)x2 + κ̇(0)x3 + o(x3 )
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

(4.1) f (x) =
2 6
holds, where o(x3 ) is Landau’s symbol. On the other hand, the osculating
circle of the curve can be expressed as a graph y = g(x) near the origin,
and we have
1
g(x) = κ(0)x2 + o(x3 )
2
by replacing f in (4.1) with g, since the curvature of the circle y = g(x) is
constant. Hence we have
1
f (x) − g(x) = κ̇(0)x3 + o(x3 ).
6
Here, κ̇(0) > 0, because the graph f (x) is a positive spiral. Then, as x
increases, f (x) − g(x) changes sign from negative to positive at the origin.
This implies that γ(s) tangentially crosses the osculating circle from right
to left.

Transformations mapping spirals to spirals. We identify the com-

plex plane C with the plane R2 by the correspondence
√

R2 3 (x, y) ←→ x + iy ∈ C i := −1 .

For complex numbers a, b, c, d satisfying ad − bc 6= 0, a transformation T

which maps z ∈ C to w = T (z) via
az + b
(4.2) z 7−→ w = T (z) =
cz + d
is called a Möbius 2 transformation, or a linear fractional transformation.
When c 6= 0, though the Möbius transformation (4.2) cannot be defined
at z = −d/c, it gives a diffeomorphism of the domain in C excluding the
point −d/c into C. Both the composition of Möbius transformations, and
the inverse of a Möbius transformation, are also Möbius transformations.
2 Möbius, August Ferdinand (1790–1868).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 44

44 Differential Geometry of Curves and Surfaces

Any Möbius transformation can be expressed as a composition of the

following four basic transformations (cf. Problem 6 at the end of this sec-
tion):
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.



 A translation z 7→ z + d (d ∈ C),
A rotation z 7→ eiθ z (θ ∈ R),

(4.3)

 A similarity transformation z 7→ rz (r ∈ R \ {0}),

Conjugation composed with inversion z 7→ 1/z.
Here we explain inversion: Let C be a circle of radius ρ in the plane centered
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

at the origin O. Take a map which maps each point P on the plane different
from O to the point Q on the half-line starting at O passing through P
such that OP · OQ = ρ2 . Such a map is called inversion with respect to the
circle C. The inversion on the complex plane C with respect to the unit
circle centered at the origin is the map z 7→ 1/z̄, where z̄ is the complex
conjugation of z.
Theorem 4.2. A Möbius transformation maps each positive spiral to a
positive spiral, and maps each negative spiral to a negative spiral. Further-
more, the osculating circles of a regular curve are mapped to the osculating
circles at the corresponding points of the image curve.
For example, by the transformation z 7→ 1/z, the logarithmic spiral is
mapped to a curve congruent to the original spiral. To prove Theorem 4.2,
we prepare the following lemma.
Lemma 4.3. A Möbius transformation maps circles to circles. Moreover,
the region to the left-hand side of a circle will be mapped to the left-hand
side of the image circle. Here the term “circles” includes straight lines,
which can be regarded as circles with infinite radius.
We used the terms “left-hand side” and “right-hand side” of the circle,
instead of the interior and exterior, since a Möbius transformation may
map the interior of the circle to the exterior of the image circle.

Proof. The conclusion is evident for the first three transformations of (4.3).
So we shall prove the lemma for the transformation z 7→ 1/z. A circle on
the xy-plane is expressed as
(4.4) a(x2 + y 2 ) + 2bx + 2cy + d = 0 (a, b, c, d ∈ R, b2 + c2 − ad > 0)
in general. In particular, when a = 0, (4.4) represents a straight line. It is
sufficient to prove that this circle is mapped to a circle by the transformation
z 7→ 1/z. Letting z = x + iy, (4.4) is rewritten as
a|z|2 + (b − ic)z + (b + ic)z̄ + d = 0 (z̄ := x − iy).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 45

Curves 45

Hence if one sets w = 1/z, it becomes

A|w|2 + (B − iC)w + (B + iC)w̄ + D = 0,

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(A := d, B := b, C := −c, D := a).
In particular, since B 2 + C 2 − AD = b2 + c2 − ad > 0, the image is also
a circle. Note that the center of the original circle may not be mapped to
the center of the image circle.
Now we prove the latter conclusion. Assume a point P is mapped to a
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

point Q by a given Möbius transformation T . Then it is sufficient to show

that the determinant of the Jacobi matrix (cf. page 18) is positive. In fact,
if this is true, T is a positive coordinate change of a neighborhood of P
to a neighborhood of Q, and then the left side of the circle is mapped to
the left side of the image circle. Since translations, rotations, and similarity
expansions and reductions have this property, one can accomplish the proof
by checking this property for the transformation z 7→ 1/z. In fact, the
determinant of the Jacobi matrix of the coordinate change z 7→ 1/z, i.e.,
ξ(x, y) = x/(x2 + y 2 ), η(x, y) = −y/(x2 + y 2 ), is positive.

Proof of Theorem 4.2. Let γ(s) be a positive spiral parametrized by the

arc-length s. Then the curvature function κ satisfies κ0 (s) > 0. By a
Möbius transformation z 7→ T (z), the osculating circle Cs of γ at γ(s) is
mapped to a circle T (Cs ) by Lemma 4.3.
By Theorem 2.4, Cs and γ have second order contact at γ(s). Here,
T is a diffeomorphism in a neighborhood of the point. Then the order of
contact is preserved by T , because of Lemma 2.6. Hence the circle T (Cs )
and the curve T ◦ γ(s) obtained by a composition of T and γ have second
order contact. In particular, T (Cs ) is the osculating circle of T ◦ γ(s).
A point where κ0 (s) = 0 is where Cs and γ have third order contact
(cf. Problem 9-(3) in Section 2). The property of third order contact is
preserved by Möbius transformations, as well as the first and second order
cases (cf. Problem 9-(2) in Section 2).
Hence if there exists a point on the curve T ◦ γ where the derivative
of the curvature function vanishes, the circle T (Cs ) and T ◦ γ have third
order contact at s. Then, by composing with the inverse transformation of
T , one can see that Cs and γ have third order contact at s, which implies
κ0 (s) = 0, a contradiction to the assumption that κ0 (s) 6= 0. Therefore T ◦γ
is a spiral.
Moreover, as we assumed that γ is a positive spiral, γ crosses tangen-
tially to Cs from the right-hand side to the left-hand side at s, because of
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 46

46 Differential Geometry of Curves and Surfaces

Theorem 4.1. This implies that T ◦ γ is also a positive spiral.

Using these properties of spirals, we can prove the four-vertex theorem

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

for simple closed curves (not necessarily convex).

Theorem 4.4. There exist at least four vertices on any simple closed reg-
ular curve that is not a circle.

Proof. Let γ(s) be a simple closed regular curve parametrized by arc-length.

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Similar to the proof of Theorem 2.11 in Section 2, we show a contradiction

assuming the number of vertices to be 2.
Let P and Q be the vertices. Without loss of generality, we may assume
that the curvature function of γ attains its minimum at P. We denote by
γ1 (resp. γ2 ) the subarc from P to Q (resp. Q to P). Then the curvature
function is non-decreasing (resp. non-increasing) on γ1 (resp. γ2 ).

Fig. 4.4

If the curvature function κ of γ is positive at P, then the curve is strictly

convex (cf. Corollary B.2.2 in Appendix B.2), and the conclusion follows by
Theorem 2.11. So we consider the case that the curvature is non-positive
at P. If κ < 0 at P, we let C be the osculating circle of γ at P (cf. Fig.
4.4, left). On the other hand, if κ vanishes at P, we choose C to be a circle
tangent to the osculating line lying in the exterior domain of the curve γ.
In this situation, if we give C the orientation that is compatible with the
orientation of the curve, then C is clockwise oriented (cf. Fig. 4.4, right).
Let O be the center of the circle C. By a homothetic transformation with
respect to O, we may assume that C is a unit circle.
Applying the Möbius transformation T (z) = 1/z, the image T (C)(= C)
is the unit circle with anti-clockwise orientation. If κ < 0 at P, T (C) itself
is the osculating circle of T ◦ γ. On the other hand, if κ = 0 at P, then the
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 47

Curves 47

osculating circle lies in the interior of C. Thus, the curvature of T ◦ γ at

T (P) is positive.
Even when the curve has a part of a straight line or a circular arc, where
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

the curvature function is constant (Corollary 2.10 in Section 2), this remains
so under Möbius transformations. By Theorem 4.2 and this fact, the prop-
erty of having a non-decreasing and non-increasing curvature function is
preserved by Möbius transformations. In particular, T (P) is the minimum
of the curvature function of the curve T ◦ γ. Therefore the curvature of
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

T ◦ γ is positive. Then T ◦ γ is a strictly convex curve (cf. Corollary B.2.2

in Appendix B.2), contradicting the four-vertex theorem for strictly convex
curves (Theorem 2.11 in Section 2).

A global property of spirals. Consider a car driving along a positive

spiral. If the driver would stop turning the wheel, the orbit of the car
would become a circle, which is the osculating circle at that point on the
spiral. Then by turning the wheel to the left, the car travels to the inside
of the osculating circle. So, one can see that the future osculating circles lie
inside the present osculating circle. The following theorem describes this
fact rigorously:

Theorem 4.5. Let γ(s) (a ≤ s ≤ b) be a positive (resp. negative) spiral.

For each value of s, let the osculating circle C(s) of γ at s have the same
orientation as γ(s). Denote by Ds the open domain which lies to the left
(resp. right) of C(s). Then, when t < s (resp. t > s), it holds that

Ds ⊂ Dt ,

where Ds is the closure of Ds , that is, the union of Ds and its boundary
∂Ds .

Remark 4.6. When the osculating circle is counterclockwise (resp. clock-

wise) oriented, the interior (resp. exterior) of the circle lies on the left.
In particular, if the curvature of γ is positive, Ds is the interior of the
osculating circle of γ at γ(s) (see Fig. 4.5).

Proof of Theorem 4.5. We shall prove the case that γ(s) (a ≤ s ≤ b) is a

positive spiral, and t = a, s = b. (By reflecting the curve across a line,
the conclusion for negative spirals is obtained.) Without loss of generality,
the parameter is arc-length and κ(a) > 0. In fact, when κ(a) ≤ 0, the
osculating circle Ca of γ(s) at s = a is a clockwise-turning circle. In this
case, choose the point O in the right-hand domain of Ca and apply the
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 48

48 Differential Geometry of Curves and Surfaces

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

Fig. 4.5 Theorem 4.5.

Möbius transformation T (z) = 1/z with respect to the origin O. Then

T (Ca ) is a counterclockwise-turning circle and the curvature of T ◦ γ(s) at
s = a is positive.
Thus, we may assume κ(a) > 0 and prove the case of a positive spiral.
Denote the evolute of γ (i.e., the locus of the centers of the osculating circles
of γ(s), see Appendix B.1) by
1
σ(s) := γ(s) + n(s)
κ(s)
(cf. Theorem B.1.1 of Appendix B.1). Differentiating this, we have σ 0 (s) =
(1/κ(s))0 n(s) by the Frenet formula (2.15).
On the other hand, let ra and rb be the radii of the osculating circles Ca
and Cb at s = a and s = b, respectively. Then, by the triangle inequality
for integrals (Theorem A.1.4 in Appendix A.1), we have
Z b Z b 0
0 1
|σ(b) − σ(a)| = σ (s) ds = n(s) ds
a a κ(s)
Z b  0 Z b 0
1 1
≤ n(s) ds = − ds
a κ(s) a κ(s)
1 1
= − = ra − rb .
κ(a) κ(b)
The condition for equality in the above inequality is that n(s) must be a
constant vector, which contradicts that the curvature of γ(s) is increasing.
Hence we have |σ(b) − σ(a)| < ra − rb , which implies that the distance
|σ(b) − σ(a)| between the center of the osculating circles is less than the
difference of the radii ra − rb . Thus Cb lies to the inside of Ca , proving
Db ⊂ Da .
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 49

Curves 49

Corollary 4.7. A spiral does not have self-intersections.

Proof. Without loss of generality, we may assume γ(s) (a ≤ s ≤ b) is a

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

positive spiral. If there would be a self-intersection, there exist t, s with

a ≤ t < s ≤ b such that γ(s) = γ(t). Then by Theorem 4.5, it holds that
γ(s) ∈ Ds ⊂ Dt , a contradiction to γ(t) 6∈ Dt .

An application of Theorem 4.5 to a generalization of the four-vertex

theorem is given by Pinkall [30]. One of the deepest results is about the
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

arrangement of vertices, stated as follows: For each simple closed curve

γ : [a, b] → R2 , there exist four points (a ≤)t1 < t2 < t3 < t4 (< b) such
that the osculating circles at t1 and t3 (resp. t2 and t4 ) are inscribed (resp.
circumscribed), see [38], and also [20, 39].

Exercises 4
1 Prove that the area bounded by the Archimedean spiral r = aθ and the half-
lines θ = α, θ = β (0 < β − α < 2π) is a2 (β 3 − α3 )/6. This fact was discovered
by Archimedes himself.
2 Show that a similarity expansion of the logarithmic spiral with respect to the
origin coincides with the original spiral up to the rotation about the origin.
Moreover, any half-line starting at the origin will make a single constant angle
with the logarithmic spiral, independent of choice of direction of the half-line
and also of choice of intersection point (see Fig. 4.1).
3 Explain, using the analogy of driving a car, that the positivity and negativity
of spirals do not depend on the orientation of curves.
4 Find the curvature functions of the Archimedean spirals and the logarithmic
spirals.
5 (1) Show that the number of vertices of the hyperbola x2 − y 2 = 1/a2 is 2.
Here a is a positive constant.
(2) Applying the Möbius transformation z 7→ 1/z (z = x + iy), show that the
lemniscate in Examples 1.1 and 1.3 is mapped to the hyperbola x2 − y 2 =
1/a2 . (In this case, the origin of the lemniscate is mapped to infinity.)
(3) Show that the lemniscate has exactly two vertices.
6 Show that any Möbius transformation can be expressed as a composition of
the four transformations in (4.3).
7∗ Show that the locus of the centers of the osculating circles of the logarithmic
spiral (i.e., the evolute, cf. Appendix B.1) is again the logarithmic spiral.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 50

50 Differential Geometry of Curves and Surfaces

5. Space curves
In this section, we consider curves in 3-dimensional Euclidean space, or
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

more briefly “space curves”. To do this, we will need the vector product
(or cross product) of two vectors in space (cf. Appendix A.3).

Space curves. Using a parameter t, we can describe a curve in space as

γ(t) = (x(t), y(t), z(t)).
Just like in the case of planar curves, a curve satisfying γ̇(t) 6= 0 for all
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

t is called a regular curve. In this section, we consider only regular space

curves.
Similar to the case of planar curves, the length of the part of the curve
corresponding to an interval [a, t] is given by
Z t Z tp
s(t) = |γ̇(u)| du = ẋ(u)2 + ẏ(u)2 + ż(u)2 du.
a a
Like the case of planar curves, we can use this to get an arc-length parameter
s.
Now let us consider the space curve γ(s) with arc-length parameter s.
We will denote the derivative with respect to s with a “ 0 ”. Then the same
equation as (2.2) in the case of planar curves holds, and
γ̇
e(s) := γ 0 (s) =
|γ̇|
is a unit vector. We call this the unit tangent vector of the curve. Since
e(s) is a unit vector, the inner product e(s) · e(s) is identically equal to 1.
Differentiating both sides of the equation e(s) · e(s) = 1 with respect to s,
we have (cf. Corollary A.3.10 in Appendix A.3)
e0 (s) · e(s) = 0,
that is, e0 (s) is perpendicular to the unit tangent vector e(s). In the case of
planar curves, the direction perpendicular to the curve (more precisely, to
the tangent direction of the curve) is uniquely determined. However, this
is not the case for space curves. Now, when e0 (s) 6= 0, we can define
e0 (s) γ 00 (s)
 
n(s) := 0 = 00 ,
|e (s)| |γ (s)|
called the principal normal vector to the curve γ. The vector n(s) is a unit
vector perpendicular to e(s).
We say that
κ(s) := |e0 (s)| (when e0 (s) = 0, we set κ(s) = 0),
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 51

Curves 51

is the curvature of the space curve γ(s). When γ 00 (s) = e0 (s) = 0, we

cannot define the principal normal vector, so we now consider the case that
γ 00 (s) never vanishes. Then we have
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(γ 00 (s) =)e0 (s) = κ(s)n(s) (κ(s) > 0).

Since e and n are a pair of perpendicular unit vectors, using the properties
of vector products (see Appendix A.3, page 205), the vector product
b(s) := e(s) × n(s)
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

is a unit vector perpendicular to both e(s) and n(s), and is called the
binormal vector. For each value of s,
{e(s), n(s), b(s)}
is a positively oriented orthonormal basis for R3 , that is, the determinant
of the 3 × 3-matrix (e(s), n(s), b(s)) is equal to 1 (cf. Appendix A.3).
For each s, the plane containing the
point γ(s) and spanned by e(s) and n(s)
is called the osculating plane of the reg-
ular curve γ at s (Fig. 5.1). In addi-
tion, the plane containing the point γ(s)
and spanned by n(s) and b(s) (resp.
e(s) and b(s)) is called the normal plane
(resp. the rectifying plane) of the curve
γ at s (Fig. 5.2).
Here, we define the torsion of the
curve as
Fig. 5.1 The osculating plane.
(5.1) τ (s) := −b0 (s) · n(s),
which measures how quickly the curve pulls away from the osculating plane
(cf. Problem 6 at the end of this section).
If two regular curves differ only by orientation preserving congruences
(cf. page 203 in Appendix A.3), then their curvatures and torsions coincide
(cf. Problem 1).
Example 5.1. The curve parametrized by
γ(t) = (a cos t, a sin t, bt)
is called a helix √
(Fig. 5.3). Here a and b are both non-zero constants.
Because |γ̇(t)| = a2 + b2 , this curve has arc-length parameter
p
s = t a 2 + b2 ,
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 52

52 Differential Geometry of Curves and Surfaces

by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

The normal plane. The rectifying plane.

Fig. 5.2 The normal and rectifying planes.

Fig. 5.3 The helix.

and reparametrized as
 
s s bs p

γ(s) = a cos , a sin , c := a2 + b2
c c c
by the arc-length parameter s. Then we have
 
1 s s
e(s) = γ 0 (s) = −a sin , a cos , b ,
c c c
 
1 s s
e0 (s) = γ 00 (s) = 2 −a cos , −a sin , 0
c c c
which imply that the curvature κ and normal vector n are
e0 (s)
 
|a| |a| a s s
κ = |γ 00 (s)| = 2 = 2 , n = = − cos , − sin , 0 ,
c a + b2 |e0 (s)| |a| c c
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 53

Curves 53

respectively. We then have that the binormal vector b and torsion τ are
a 1 s s 
b=e×n= b sin , −b cos , a ,
|a| c
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

c c
b b
τ = −b0 · n = 2 = 2 .
c a + b2
In particular, both the curvature and torsion of a helix are constant.

The Frenet-Serret formula. Let γ(s) be a regular curve in R3 param-

Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

etrized by arc-length such that κ(s) > 0 for each s. Since the unit tangent
vector e(s), the principal normal vector n(s), and the binormal vector b(s)
form an orthonormal basis for R3 , any vector in R3 can be written as a
linear combination of them. Therefore n0 and b0 can be written as
dn
(5.2) n0 = = α1 e + α2 n + α3 b,
ds
db
(5.3) b0 = = β1 e + β2 n + β3 b,
ds
where αj , βj (j = 1, 2, 3) are real-valued functions depending on s.
Taking inner products of both sides of equation (5.2) with e, n and b,
we find that
α1 = n0 · e = (n · e)0 − n · e0 = −κn · n = −κ,
1
α2 = n0 · n = (n · n)0 = 0,
2
α3 = n0 · b = (n · b)0 − n · b0 = τ.

Again, taking inner products of both sides of equation (5.3) with e, n and
b, we have

β1 = b0 · e = (b · e)0 − b · e0 = −κb · n = 0,
β2 = b0 · n = −τ,
1
β3 = b0 · b = (b · b)0 = 0.
2
Substituting these into (5.2), (5.3) (together with e0 = κn), we have
 0
 e (s) = κ(s) n(s),
(5.4) n0 (s) = −κ(s) e(s) + τ (s) b(s),
 0
b (s) = − τ (s) n(s).
We call these the Frenet-Serret 1 formula.
1 Serret, Joseph Alfred (1819–1885). See page 21 for Frenet.
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 54

54 Differential Geometry of Curves and Surfaces

Take the 3 × 3-matrix F(s) := (e(s), n(s), b(s)), where e(s), n(s) and
b(s) are considered as column vectors. Then F(s) is an orthogonal matrix
with determinant 1, since {e(s), n(s), b(s)} forms a positive orthonormal
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

basis (a right-handed system, see Appendix A.3) for each s. Using this, the
Frenet-Serret formula (5.4) is reformulated as
 
0 −κ 0
dF
(5.5) = FΩ, Ω := κ 0 −τ  .
ds
0 τ 0
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

In Example 5.1, we found the curvature and the torsion by reparametriz-

ing the curve by an arc-length parameter. However, to compute curvatures
and torsions for various examples, it is convenient to prepare formulas to
compute these quantities for general parameters. Let γ(t) be a regular
curve parametrized by (not necessarily arc-length) parameter t. Then its
unit tangent vector e(t), the principal normal vector n(t) and the binormal
vector b(t) are given as
γ̇ (γ̇ × γ̈) × γ̇ γ̇ × γ̈
(5.6) e= , n= , b= .
|γ̇| |(γ̇ × γ̈) × γ̇| |γ̇ × γ̈|
Moreover, the curvature and the torsion are expressed as
...
|γ̇(t) × γ̈(t)| det(γ̇(t), γ̈(t), γ (t))
(5.7) κ(t) = , τ (t) =
|γ̇(t)|3 |γ̇(t) × γ̈(t)|2
(cf. Problem 2). In particular, when γ is parametrized by arc-length s,
(5.7) is rewritten as
det(γ 0 (s), γ 00 (s), γ 000 (s))
κ(s) = |γ 00 (s)|, τ (s) = .
κ(s)2
The fundamental theorem for space curves. We have already seen
in Theorem 2.8 in Section 2 that a plane curve is determined uniquely by
its curvature function. A similar result is true for space curves.
Theorem 5.2 (The fundamental theorem for space curves). Let κ(s) and
τ (s) be two smooth functions defined on an interval [a, b] such that κ(s) > 0
holds on [a, b]. Then there exists a regular curve γ(s) (a ≤ s ≤ b) param-
etrized by arc-length s whose curvature and torsion are κ(s) and τ (s), re-
spectively. Moreover, such a curve is unique up to orientation preserving
congruences.

Proof. The theorem can be proven in a similar way as we proved Theo-

rem 2.8 for plane curves. Consider the initial value problem of the ordinary
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 55

Curves 55

differential equation
 
0 −κ 0
F 0 = FΩ,
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

(5.8) F(a) = I, Ω = κ 0 −τ 
0 τ 0
whose unknown is the function F(s) taking values in the set of 3 × 3-
matrices, where I is the 3 × 3-identity matrix. Since this equation is linear
with respect to the nine components of F, by the fundamental theorem
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

for linear ordinary differential equations (Theorem A.2.2 in Appendix A.1),

there exists a unique solution F(s) of (5.8) defined on the interval [a, b].
Here, since Ω(s) is a skew-symmetric matrix, in a similar way to the case
for plane curves (cf. (2.17)), we know that F(s) is an orthogonal matrix
with determinant 1. Thus, we write F(s) = (e(s), n(s), b(s)), and let
Z s
γ(s) = e(u) du.
s0

Then γ(s) is the desired curve. The uniqueness part and final part of the
proof are left as the exercises for readers.

A simple closed space curve is

called a knot. In contrast to the
planar curve case that any sim-
ple closed planar curve can be de-
formed continuously to a circle, the
knot as in Fig. 5.4 cannot be de-
formed to a circle. One of the im-
portant problems for space curves
is to investigate properties of knots.
Relationships between knots and Fig. 5.4 The trefoil knot.
their total curvatures (defined in a
similar way as that for planar curves, see page 28) are discussed in [26].

Exercises 5
1 Show that the curvature and torsion of a space curve are invariant under ori-
entation preserving congruences. Namely, let Φ be an orientation preserving
congruence (cf. page 203 in Appendix A.3). Then the curvature and tor-
sion for a curve γ(s) and those for Φ ◦ γ(s) coincide. Note that whereas the
curvature is invariant under general congruence, the torsion may change sign.
2 Verify (5.6) and (5.7).
March 29, 2017 9:45 ws-book9x6 World Scientific Book - 9in x 6in 9901-main page 56

56 Differential Geometry of Curves and Surfaces

3 Let γ(s) = (x(s), y(s)) be a planar curve parametrized by arc-length s, and

assume that the curvature of γ does not vanish. Prove that the curvature of
the space curve γ̃(s) := (x(s), y(s), 0) coincides with the absolute value of the
by 2409:40d2:a:8f73:10c2:8aff:fe61:88b7 on 03/10/25. Re-use and distribution is strictly not permitted, except for Open Access articles.

curvature of γ(s), and the torsion of γ̃ is identically zero.

4 Let γ(s) be a space curve parametrized by arc-length s whose curvature and
torsion are κ(s) and τ (s), respectively. Show that, near s = 0, it holds that

s2
γ(s) = γ(0) + se(0) + κ(0)n(0)
2
Differential Geometry of Curves and Surfaces Downloaded from www.worldscientific.com

s3
+ (−κ(0)2 e(0) + κ0 (0)n(0) + κ(0)τ (0)b(0)) + o(s3 ),
6

where o(s3 ) represents a higher order term (cf. Appendix A.1). This formula
is known as Bouquet’s formula 2 . Compare with Problem 8 in Section 2.
5 Show that a space regular curve whose curvature is a positive constant, and
whose torsion is a non-zero constant, coincides with a helix by an orientation
preserving congruence.
6∗ Show that a space regular curve whose torsion vanishes identically (and having
positive curvature) coincides with a curve obtained as in Problem 3 from a
planar curve, up to an orientation preserving congruence.
7∗ Show that a regular curve that lies on a sphere and has constant curvature is
a part of a circle.
8∗ Let γ1 (t) be a planar regular curve with curvature κ1 (t), and γ2 (t) a curve
obtained by reflecting γ1 (t) across a line. Show that the curvature of γ2 (t)
is −κ1 (t) (cf. Problem 4 in Section 2). Let γ̃1 (t) and γ̃2 (t) be space curves
obtained from γ1 and γ2 , respectively, as in Problem 3. Then show that the
curvature functions of these curves are |κ1 |, and the torsion functions vanish
identically.
Here, since γ1 and γ2 have different curvature functions as planar curves, they
cannot coincide by rotations and translations. In contrast, the space curves γ̃1
and γ̃2 have the same curvature and torsion, and hence by the fundamental
theorem for space curves (Theorem 5.2), they coincide by a rotation and
translation (an orientation preserving congruence). Explain the reason for
this difference between planar curves and space curves.

2 Bouquet, Jean Claude (1819–1885).