DIFFERENTIAL GEOMETRY OF CURVES AND SURFACES

1. Curves in the Plane

1.1. Points, Vectors, and Their Coordinates. Points and vectors are fundamental
objects in Geometry. The notion of point is intuitive and clear to everyone. The notion
of vector is a bit more delicate. In fact, rather than saying what a vector is, we prefer
to say what a vector has, namely: direction, sense, and length (or magnitude). It can be
represented by an arrow, and the main idea is that two arrows represent the same vector if
they have the same direction, sense, and length. An arrow representing a vector has a tail
and a tip. From the (rough) definition above, we deduce that in order to represent (if you
want, to draw) a given vector as an arrow, it is necessary and sufficient to prescribe its tail.

b b

a a

Figure 1. We see four copies of the vector a, three of

the vector b, and two of the vector c. We also see a point
P.

An important instrument in handling points, vectors, and (consequently) many other

geometric objects is the Cartesian coordinate system in the plane. This consists of a point
O, called the origin, and two perpendicular lines going through O, called coordinate axes.
Each line has a positive direction, indicated by an arrow (see Figure 2). We denote by R the

O O

Figure 2. The point P has coordinates x, y. The vector

a has also coordinates x, y.

set of all real numbers and by R2 the set of all pairs of numbers, of the form (x, y), where
x, y are in R. Points are identified with elements R2 , as follows: to each point P corresponds
the pair (x, y) consisting of the coordinates of the projections of P on the two axes. We say
that P has coordinates (x, y). Also vectors are identified with elements of R2 , as follows: if
1
2

a is a vector, we move its tail to the origin O, and we take the coordinates of its tip. We say
that a has coordinates (x, y). One of the first advantages of the coordinate system is that
we can use it to compute lengths, as follows. The distance between the points P1 = (x1 , y1 )
and P2 = (x2 , y2 ) is
p
kP1 P2 k = (x2 − x1 )2 + (y2 − y1 )2 .
The length of the vector a = (x, y) is
p
kak = x2 + y 2.

In our course, R2 will denote both the set of all points and the set of all vectors (in the
plane). It will always be possible to understand from the context if a certain object in R2 is
a point or a vector.

1.2. Parametrized Curves. A good way of thinking of a curve is as the object which
describes the motion of a particle in the plane: at the time t, the particle is at the point in
the plane whose coordinates are (x(t), y(t)). We stress from the very beginning that what
we are interested in is not simply what the trajectory of the particle is, but rather how the
trajectory is run. Now comes the exact definition.
Definition 1.2.1. A parametrized curve in the plane is a differentiable function1
α(t) = (x(t), y(t)),
where t satisfies a < t < b (possibly a and/or b can be ∞).
When we say that α(t) = (x(t), y(t)) is “differentiable” we mean that both x(t) and y(t)
have derivative of any order (we also say that they are C ∞ differentiable).
The standard notation for such an object is
α : (a, b) → R2 ,
where (a, b) is the open interval between a and b. As we already mentioned, it is important
to distinguish between the curve α (the assignment which associates to any “time” t the
point (x(t), y(t)) on the “trajectory”) and the image of the function α (the “trajectory”).
The latter is called the trace of the curve α.
Remark. Very often it is possible to describe the trace of a curve α : (a, b) → R2 with
α(t) = (x(t), y(t)), by an equation of the form f (x, y) = 0, where f is a function of variables
x, y. If so, we say that
x = x(t), y = y(t)
are the parametric equations (or explicit equations) of the curve and
f (x, y) = 0
is the implicit equation of the curve. For example, the trace of the curve
x = 1 − 2t, y = 5t
is a straight line (see below). The same line can be described by the implicit equation
5x + 2y = 5.
Also, the circle of radius 1 and centre 0 can be described as the trace of
x = cos t, y = sin t
1It is worth mentioning that the domain of the function α is the interval a < t < b and the range is R2 .
 3

but also by the equation

x2 + y 2 = 1.
Examples. 1. The straight line determined by two points P and Q is the trace of the curve
α(t) = tP + (1 − t)Q,
where t ∈ R (see Figure 3). Note that there are several other curves whose traces are the
same straight line, like for instance
β(t) = tQ + (1 − t)P
or
γ(t) = 2tP + (1 − 2t)Q.

3.5
3
2.5
2
1.5
1
0.5
-2 -1 1 2 3 4 5

Figure 3. The straight line determined by the two in-

dicated points.

2. The circle of center C = (x0 , y0 ) and radius r is the trace of the curve
α(t) = (x0 + r cos(t), y0 + r sin(t)),
for all t in R (see Figure 4). It can be described implicitly as
(x − x0 )2 + (y − y0 )2 = r 2

C
P

Figure 4. A circle of center C .

3. The ellipse is the trace of the curve

α(t) = (a cos(t), b sin(t)),
4

for all t in R (see Figure 5). Here a, b are positive numbers. It can be described implicitly as
x2 y 2
+ 2 = 1.
a2 b

H0,bL
P

H-a,0L F- F+ Ha,0L

H0,-bL

Figure 5. The ellipse with its two foci F− and F+ .

We have the following geometric (coordinate free) characterization of the ellipse. There
exists two points, let’s denote them F− and F+ , which are called the foci of the ellipse, with
the property that for any point P on the ellipse we have
kP F− k + kP F+ k = constant.
√ √
More specifically, F− has coordinates (− a2 − b2 , 0), and F+ has coordinates ( a2 − b2 , 0)
(you are encouraged to calculate the distances kP F− k, kP F+ k, add them up, and check that
the result is constant, that is, independent of t).
4. The parabola is the trace of the curve
α(t) = (t, at2 ),
where a is a number. It can be described implicitly as
y = ax2 .
Again we have a geometric (coordinate free) characterization. Namely, there exists a point

F
d

Figure 6. The parabola with its focus F and directrix

line d.
 5

F , called the focus, and a straight line d, called the directrix, such that for any point P on
the parabola, the distance between P and F is equal to the distance between P and d. For
instance, for α(t) = (t, t2 ), the focus is (0, 1/4) and the directrix is the horizontal line of
equation y = −1/4 (check this!).
1.3. Velocity, reparametrizations, and the length of a curve. We start with the
following definition.
Definition 1.3.1. If α(t) = (x(t), y(t)), with a < t < b is a parametric curve, then the
velocity of α at a given t is the vector
α′ (t) = (x′ (t), y ′(t)).
The speed of α at t is the number kα′(t)k.

Α'HtL

ΑHtL

Figure 7. The vector α′ (t) is tangent to the curve at

the point α(t).

The velocity vector is useful because it is tangent to the curve, so it can be used to describe
the tangent line to the curve (we will get back to this later on). Another reason why the
tangent vector is important is that it can be used to compute the length of a piece of a curve.
Namely, we have the following definition.
Definition 1.3.2. Let α(t), a < t < b be a parametrized curve and let [c, d] be a closed
interval with a < c < d < b. The length of α over the interval [c, d] (or from α(c) to α(d)) is
Z d
ℓ(α|[c,d]) = kα′(t)kdt.
c

To get the idea of this formula, we just say that if we divide the interval [c, d] into n equal
pieces given by c = t0 < t1 < . . . < tn−1 < tn = d, then
n
X
ℓ(α|[c,d]) ≈ kα(ti ) − α(ti−1 ))k,
i=1

where the right hand side represents the length of the polygonal line in Figure 10. For each
i we can approximate
α(ti ) − α(ti−1 )
≈ α′ (ti−1 ).
ti − ti−1
6

α(t ) α(t )
i 1

Figure 8. The length of the line segment joining α(ti−1 )

and α(ti ) is kα(ti ) − α(ti−1 )k.

This is because, by definition, we have

α(t) − α(ti−1 )
α′ (ti−1 ) = lim
t→ti−1 t − ti−1
and ti is “close” to ti−1 . We obtain
n
X
ℓ(α|[c,d]) ≈ kα′ (ti−1 )k · |ti − ti−1 |.
i=1

In fact the actual length of α over [c, d] is the limit as n → ∞ of the right hand side. But the
latter expression is a Riemann sum of the function kα′(t)k. The formula given by Definition
1.3.2 follows.
Now we get back to the idea mentioned in section 1.2, example 1. Namely, we said that
there are several curves whose trace is a given straight line. In other words, it is possible to
parametrize a straight line in several ways. This is a general phenomenon: any curve can be
reparametrized, in a sense made precise by the following definition.
Definition 1.3.3. Let α1 : (a1 , b1 ) → R2 be a parametrized curve. We say that the curve
α2 : (a2 , b2 ) → R2 is a reparametrization of α1 if there exists h : (a1 , b1 ) → (a2 , b2 ) a smooth
function which is differentiable and strictly monotone2 such that α1 (t) = α2 (h(t)), for all t
in (a1 , b1 ).
Note that in the situation described by the definition, the curves α1 and α2 have the same
trace.
Now the problem is if the length of the curve as described by Definition 1.3.1 depends on
the parametrization (of course, it shouldn’t, because the length should depend only on the
trace of the curve). For example, we can parametrize the half circle in the figure as
α1 (t) = (cos(t), sin(t)), 0 ≤ t ≤ π
but also as3
π
α2 (u) = (cos(2u), sin(2u)), 0 ≤ u ≤ .
2
Is ℓ(α1 |[0,π]) the same as ℓ(α2 |[0, π2 ] ) (and actually equal to π)?
The following theorem shows that the length does not depend on the parametrization.
2That is, h is strictly increasing or strictly decreasing.
3What is h?
 7

a b b

2 2

α 1

α 2

Figure 9. The diagram indicates the reparametrization

given by h : (a1 , b1 ) → (a2 , b2 ). The little circles indicate
that the points indicated do not belong to the curve (line
segment).

H-1,0L H1,0L

Figure 10. The half circle which is at the same time

the trace of α1 and α2 .

Theorem 1.3.4. Assume that α2 : (a2 , b2 ) → R2 is a reparametrization of α1 : (a1 , b1 ) → R2

given by h : (a1 , b1 ) → (a2 , b2 ). Take c1 , d1, c2 , d2 such that a1 < c1 < d1 < b1 and a2 < c2 <
d2 < b2 such that h maps [c1 , d1] to [c2 , d2 ]. Then we have
ℓ(α1 |[c1 ,d1 ] ) = ℓ(α2 |[c2 ,d2 ] ).

a b b

c d

a c

1 1 2 2 2 2

α 1

α 2

Figure 11. The thickened piece of the curve is

parametrized in two ways. Is Definition 1.3.1 applied
to α1 and α2 giving the same result for the length?
8

Proof. The situation described in the theorem is represented in Figure 11. We have to
check that
Z d1 Z d2
′
(1) kα1 (t)kdt = kα2′ (u)kdu.
c1 c2

Let us assume that h(t) is increasing (do the case h(t) strictly decreasing!). We start with
the right hand side of (1), where we make the change of variable u = h(t) ⇒ du = h′ (t)dt.
Because h(c1 ) = c2 and h(d1 ) = d2 , we have
Z d2 Z d1 Z d1 Z d1
′ ′ ′ ′ ′
kα2 (u)kdu = kα2 (h(t))kh (t)dt = kα2 (h(t))h (t)kdt = k(α2 (h(t)))′ kdt,
c2 c1 c1 c1

which is just the left hand side of the desired equation (1) (note that on the last step we
have used the chain rule for derivatives). 
From now on we will restrict ourselves to parametric curves with nonzero velocity vectors,
which are called regular.
Definition 1.3.5. A curve α : (a, b) → R2 is called regular if α′ (t) 6= 0, for all t in the
interval (a, b).
Such curves have a convenient parametrization, where the parameter is just the length
of the curve. Let us be more precise. We fix a number c with a < c < b. We define the
arc length function s(t) given by
Z t (
−ℓ(α|[t,c] ), if a < t ≤ c
s(t) = kα′(u)kdu =
c ℓ(α|[c,t] ), if c ≤ t < b
From the fundamental theorem of calculus, we have
d
(2) s(t) = kα′ (t)k,
dt
for all t in (a, b) (as α′ (t) 6= 0), so the function s(t) is strictly increasing. Denote by t(s) the
inverse of this function. Consider the curve
β(s) = α(t(s)),
which is a reparametrization of α. We call it a parametrization by arc length. After a curve
has been reparametrized in the way described above, we say that it is parametrized by arc length.
We have just shown that
Theorem 1.3.6. Any regular curve can be parametrized by arc length.
The advantages of the parametrization by arc length are as follows.
Proposition 1.3.7. Assume that the curve β = β(s) is parametrized by arc length. Then
we have the following two properties.
(a) the length of the velocity vector4 β ′ (s) is
kβ ′ (s)k = 1
(b) ℓ(β|[s1,s2] ) = s2 − s1 , for all s1 < s2 .
Proof. (a) We use the fact that β(s(t)) = α(t). Using the chain rule we obtain
d d d d
α′ (t) = α(t) = β(s) · s(t) = β(s) · kα′(t)k,
dt ds dt ds
4By β ′ we mean d
ds β.
 9

where we have used equation (2). We deduce

d d d
kα′ (t)k = kα(t)k = k β(s) · kα′(t)kk = k β(s)k · kα′ (t)k,
dt ds ds
which implies the desired relation.
(b) We have
Z s2 Z s2
′
ℓ(β|s1 ,s2 ) = kβ (s)kds = 1ds = s2 − s1 .
s1 s1

Note. The following converse of point (a) above is true. Assume that α = α(t) is a curve,
a < t < b, such that kα′ (t)k = 1, for all t (such curves are called unit-speed curves). To
parametrize it by arc length, pick c with a < c < b and calculate
Z t Z t
′
s= kα (u)kdu = 1du = t − c.
c c

This gives t = s + c. So the curve α is already parametrized by arc length, up to the

additive constant c. Modulo this ambiguity, we say that a curve is unit-speed if and only if
it is parametrized by arc length.
Example. Let’s parametrize by arc length the logarithmic spiral
α(t) = (aebt cos t, aebt sin t),
where a, b are positive numbers (see Figure 12). To simplify things we make a = 1. We have

-4 -2 2 4

-2

-4

Figure 12. The logarithmic spiral is characterized by

the fact that at any point on it the angle between the
position vector α(t) and the tangent vector α′ (t) is the
same.

α′ (t) = (bebt cos(t) − ebt sin(t), bebt sin(t) + ebt cos(t))

which implies
p √
kα′ (t)k = e2bt [(b cos t − sin t)2 + (b sin t + cos t)2 ] = ebt 1 + b2 .
10

Then √
Z t
bu
√ 1 + b2 bt
s(t) = e 1 + b2 du = (e − 1).
0 b
To determine t = t(s) we solve √
1 + b2 bt
s= (e − 1)
b
with respect to t. We obtain
1 bs
ln[ √ t= + 1].
b 1 + b2
The parametrization by arc length is β(s) = α(t(s)). To simplify things even more, we make
b = 1. We obtain
s s s s
β(s) = (( √ + 1) cos ln( √ + 1), ( √ + 1) sin ln( √ + 1)).
2 2 2 2
1.4. Curvature; the fundamental theorem of plane curves. We would like to measure
how “curved” is a curve at a certain point (see Figure 13). In other words, we need an
object, called “curvature” of a curve, which measures this. For instance, one would expect
the “curvature” of a straight line to be zero and also that the curvature of a circle at any of
its points to be the same (the circle is equally “curved” at any of its points).

Figure 13. The curve in the figure is more “curved” at

the point B than at A.

Definition 1.4.1. Let α(t) = (x(t), y(t)), a < t < b be a regular curve. The curvature of α
at t0 is5
x′ (t0 )y ′′ (t0 ) − x′′ (t0 )y ′(t0 )
(3) κp (t0 ) = 3 .
(x′ (t0 )2 + y ′ (t0 )2 ) 2
It is at this moment not clear at all why κp given by the strange formula above measures
how curved is the curve α. We will address this point later (at the end of this section). An-
other important question is whether the curvature at a point depends on the parametrization.
For instance, if we consider the parametrizations of the unit circle given by
α1 (t) = (cos(t), sin(t)), α2 (u) = (cos(2u), sin(2u))
is the curvature of α1 at, say, t = π/4 the same as the curvature of α2 at t = π/8? The
answer is given by the following proposition.
5The
subscript p stands for “plane”. We will define the curvature κ for curves in the three dimensional
space, and that will be slightly different from κp . The latter is also called the signed curvature, because it
can be positive or negative.
 11

Theorem 1.4.2. Assume that α2 : (a2 , b2 ) → R2 is a reparametrization of α1 : (a1 , b1 ) → R2

given by h : (a1 , b1 ) → (a2 , b2 ). Then the curvature of α1 at t equals the curvature of α2 at
h(t) up to a sign.
Proof. If we denote α2 (u) = (x(u), y(u)), then
α1 (t) = α2 (h(t)) = (x(h(t)), y(h(t))).
We have
d d2
x(h(t)) = x′ (h(t))h′ (t), x(h(t)) = x′′ (h(t))(h′ (t))2 + x′ (h(t))h′′ (t),
dt dt2
and similar formulas for y. The curvature of α1 at t is
x′ (h(t))h′ (t)(y ′′ (h(t))(h′ (t))2 + y ′(h(t))h′′ (t)) − y ′(h(t))h′ (t)(x′′ (h(t))(h′ (t))2 + x′ (h(t))h′′ (t))
3 .
(x′ (h(t))2 h′ (t)2 + y ′ (h(t))2 h′ (t)2 ) 2
This gives
x′ (h(t))y ′′ (h(t)) − x′′ (h(t))y ′ (h(t))
± 3
(x′ (h(t))2 + y ′(h(t))2 ) 2
with a plus sign if h′ (t) > 0 and minus sign if h′ (t) < 0.
Now we do a few simple examples and see if κp given by (3) responds to our expectations.
Examples. 1. We consider the straight line given by α(t) = tP + (1 − t)Q, where P, Q are
points (see section 1.2). Take P = (x1 , y1) and Q = (x2 , y2 ), then α(t) = (x(t), y(t)), where
x(t) = tx1 + (1 − t)x2 , y(t) = ty1 + (1 − t)y2 .
We can easily see that the formula (3) gives κp (t) = 0 for any t (because x′′ (t) = y ′′ (t) = 0).
2. Let us consider now the circle described by (see again section 1.2)
x(t) = x0 + r cos(t), y(t) = y0 + r sin(t),
where t is in R. We have
x′ (t) = −r sin(t), x′′ (t) = −r cos(t), y ′(t) = r cos(t), y”(t) = −r sin(t)
which gives
1
κp (t) = .
r
This is independent of t, that is, the circle is indeed equally curved at any of its points.
Also note that this also corresponds to the following intuitive fact: circles with larger radius
are less curved than circles with smaller one (for instance, if you go around the Earth along
the equator, you don’t notice that you travel along a circle, but rather along a straight line,
simply because the radius of the Earth at the equator is huge).
3. Exercise. Find the curvature of the ellipse at an arbitrary point (see section 1.2,
Example 3) and check that if a > b then the ellipse is more curved at (a, 0) than at (0, b).
We get a simpler and more natural description of κp if the curve is parametrized by arc
length. So let us assume that α is a curve which is (if you want, has been) parametrized by
arc length (see Theorem 1.3.6). We denote
T (s) = α′ (s),
which, by Proposition 1.3.7, satisfies
kT (s)k = 1.
For each s we choose the vector N(s) which is uniquely determined by the following three
properties:
12

• N(s) is perpendicular to T (s)

• kN(s)k = 1
• we can bring the pair (T (s), N(s)) to (e1 , e2 ) by a rotation, where e1 and e2 are the
vectors of length one pointing in the positive direction on the coordinate axes (see
Figure 14).

Figure 14. In the left-hand side figure, the choice of N

is correct, whereas in the right-hand side it’s wrong.

In fact we can define N(s) simpler, by saying that it is obtained from T (s) by a 90o
rotation in the counterclockwise direction. That is,
 ′
−y ′ (s)
   
0 −1 x (s)
N(s) = = .
1 0 y ′(s) x′ (s)

Theorem 1.4.3. (The Frenet Formulas for Plane Curves) If the curve α is parametrized
by arc length and T , N are as above, then we have6
T ′ (s) = κp (s)N(s),
N ′ (s) = −κp (s)T (s)
for all s.
Proof. We can see that
T (s) = (x′ (s), y ′(s)) ⇒ T ′ (s) = (x′′ (s), y ′′(s)).
Also, from the definition of N(s), we see that this vector is actually obtained from T (s) by
a 90o rotation in the counterclockwise direction. This gives
N(s) = (−y ′ (s), x′ (s)).
From kT (s)k = 1 we deduce
kT (s)k2 = T (s) · T (s) = 1,
where · indicates the scalar product of two vectors. We differentiate and obtain
T ′ (s) · T (s) = 0,
which says that T ′ (s) is perpendicular to T (s). Consequently, T ′ (s) is collinear with N(s),
so there exists a number λ with
T ′ (s) = λN(s).
This equation gives
x′′ (s) = −λy ′ (s), y ′′(s) = λx′ (s).
6The derivative is with respect to s.
 13

We multiply the first equation by −y ′ (s) and the second one by x′ (s) and obtain
−x′′ (s)y ′(s) + x′ (s)y ′′(s) = λ(x′ (s)2 + y ′ (s)2 ).
Because x′ (s)2 + y ′ (s)2 = 1, we deduce that λ = κp (s), QED. 
A straightforward consequence of this theorem is as follows (note that it’s possible for κp
to be negative).
Corollary 1.4.4. If the curve α is parametrized by arc length, then the absolute value of the
curvature is given by
|κp (s)| = kα′′ (s)k.

The following proposition gives a geometric characterization of the curvature.

Proposition 1.4.5. If the curve α is parametrized by arc length, then the curvature is given
by
d
κp (s) = θ(s),
ds
where θ(s) is the angle between Ox and α′ (s).
Proof. Since kα′ (s)k = 1, we deduce (see also Figure 15) that the coordinates of α′ (s) are
cos θ(s) = x′ (s), sin θ(s) = y ′ (s).
This implies that
κp (s) = x′ (s)y ′′ (s) − x′′ (s)y ′ (s) = (cos θ(s))2 θ′ (s) + (sin θ(s))2 θ′ (s) = θ′ (s),
which is exactly the desired equation. 

α ' ( s )

( s )

α ( s )

Figure 15. The angle between α′ (s) and the Ox axis is θ(s).

Note. The angle θ is called the turning angle of the curve α : (a, b) → R2 . It is actually a
differentiable function θ : (a, b) → R. Its exact definition is given in Section 1.5 of [Gr-Ab-Sa].
14

The following theorem shows that a curve is uniquely determined (up to rigid motions)
by its curvature.
Theorem 1.4.6. (Fundamental Theorem of Plane Curves) If κ : (a, b) → R is an
arbitrary differentiable function, then there exists a curve α parametrized by arc length such
that such that the curvature of α is
κp (s) = κ(s).
If β : (a, b) → R2 is another curve parametrized by arc length with curvature equal to κ, then
β differs from α by a translation followed by a rotation. By this we mean that we have
β(s) = A(α(s)) + X,
where A is a matrix of the form
 
cos θ0 − sin θ0
A=
sin θ0 cos θ0
which gives the rotation with angle θ0 and X is a vector of the form
 
x0
X=
y0
which gives the translation.
Proof. For simplicity, assume that 0 is in the interval (a, b). From Proposition 1.4.5. we
deduce that if a curve has curvature κ then the angle θ(s) with the Ox axis is given by
θ(s) = θ0 + θ̄(s),
where we have denoted Z s
θ̄(s) := κ(s)ds.
0
Because x′ (s) = cos θ(s), we deduce that
Z s Z s Z s
x(s) = x0 + cos(θ0 + θ̄(u))du = x0 + cos(θ0 ) cos θ̄(u)du − sin(θ0 ) sin θ̄(u)du,
0 0 0
and similarly,
Z s Z s Z s
y(s) = y0 + sin(θ0 + θ̄(u))du = x0 + sin(θ0 ) cos θ̄(u)du + cos(θ0 ) sin θ̄(u)du.
0 0 0
We deduce immediately that
Z s Z s
α(s) = ( cos θ̄(u)ds, sin θ̄(u)du)
0 0
is a curve of curvature κ and any curve β with this property differs from α by the rotation
of angle θ0 and the translation given by (x0 , y0). 
1.5. Tangent line, normal line, osculating circle, and evolutes.
We begin with the following definition (see also Figure 16).
Definition 1.5.1. The tangent line to the regular curve α : (a, b) → R2 at α(t) is the straight
line which goes through the point α(t) and is parallel to the vector α′ (t). The normal line
to α at α(t) is the straight line which goes through the point α(t) and is perpendicular to
the tangent line.
The tangent line is the one that “approximates the best” a curve at a given point. For
instance, if we take again the example of someone who travels around the Earth along the
equator, we said already that she has at any moment the impression that she goes along a
 15

Figure 16. The two straight lines are the tangent, re-
spectively normal lines at α(t) to the curve.

straight line, namely the tangent line. Now instead of straight lines, we will be trying to
approximate a curve by a circle.
Definition 1.5.2. The osculating circle to the regular curve α : (a, b) → R2 with α(t) =
(x(t), y(t)) at the point α(t) is the circle of radius7 |κp1(t)| and center of coordinates

1
(4) (x̄(t), ȳ(t)) := (x(t), y(t)) + (−y ′(t), x′ (t)).
κp (t)kα′ (t)k
The latter point is called the curvature centre of α at α(t).
The formulas look messy, but they give the answer to the following approximation question.
Question. Given a point α(t) on the curve, which is the circle going through it and giving
“the best” approximation of the curve at α(t)?
By “best approximation” of α by a circle we mean the following three things:
(i) the circle goes through α(t)
(ii) the tangent lines to the circle and the curve α at α(t) are the same
(iii) the curvature of α at t equals the curvature of the circle, possibly up to a negative
sign.
Exercise. Check that the osculating circle given by Definition 1.5.2 satisfies the conditions
(i),(ii), and (iii).
It is worth noticing that the conditions (i),(ii), and (iii) do not determine the osculating
circle uniquely: there are two possibilities, as one can see in Figure 17. The osculating circle
is the one on the concave side of the curve, namely the side where the acceleration vector α′′ (t)
1 ′ ′
points. To prove this fact, we only need to notice that the vector κp (t)kα ′ (t)k (−y (t), x (t)) is

perpendicular to the tangent line at α(t) and its angle with α′′ (t) = (x′′ (t), y ′′(t)) is acute
(as the dot product with the latter vector is positive, in view of Definition 1.4.1. of κp (t)).

7If κp (t) = 0, the radius becomes ∞ and the circle becomes a straight line.
16

( t )

α '

( x ( t ) , y ( t ) )

( t )

1 / k p

( x ( t ) , y ( t ) )

) )

( t )

α ' '

Figure 17. Both circles in the figure approximate the

curve at α(t) = (x(t), y(t)) (in the sense of (i), (ii), and
(iii)), but only one of them is the osculating circle. The
vector with tail at α(t) and tip at the center of the oscu-
1 ′ ′
lating circle is κp (t)kα′ (t)k (−y (t), x (t)).

Let us consider the locus in the plane described by all curvature centres (x̄(t), ȳ(t)) (see
equation (4)). We obtain a curve ᾱ : (a, b) → R2 given by

(5) ᾱ(t) = (x̄(t), ȳ(t)).

The curve ᾱ is called the evolute of α.

Exercise. Show that the evolute of the ellipse

α(t) = (a cos t, b sin(t))

is the curve of equation

(a2 − b2 ) cos3 t (b2 − a2 ) sin3 t
 
ᾱ(t) = , .
a b
The latter curve is an astroid, see Figure 19.
The evolute has the following interesting property.
Proposition 1.5.3. Assume that α is a curve parametrized by arc length and let ᾱ be its
evolute. Then the tangent line to ᾱ at ᾱ(s) coincides with the normal line to α at α(s), for
any s.
Proof. We use the notation T (s) = α′ (s) = (x′ (s), y ′(s)), N(s) = (−y ′ (s), x′ (s)), like in
section 1.4. Then by (5) and (4) we can write
1
ᾱ(s) = α(s) + N(s).
κp (s)
 17

Figure 18. The ellipse and its evolute.

This implies
 
′ d ′ 1 1
(6) ᾱ (s) = α (s) − N(s) + N ′ (s).
ds κp (s) κp (s)
Because N ′ (s) = −κp (s)T (s) (see Theorem 1.4.3.), we deduce that
 
′ d 1
ᾱ (s) = − N(s),
ds κp (s)
which implies that the vectors ᾱ′ (s) and N(s) are collinear. The only thing we still need to
check is that the straight line through ᾱ(s) (see (6)) of direction vector N(s) goes through
α(s). But this is obvious. See also Figure 19. 

( s )

α ( s )

( s )

α ( s )

α ' ( s )

Figure 19. The dotted curve is the trace of the evolute

of α.

We wanted now to raise the following question: given a curve ᾱ, is there any simple way
to find a curve α whose evolute is ᾱ? The answer is yes, namely α is the involute of ᾱ (see
the definition below).
Definition 1.5.4. Let β : (a, b) → R2 be a curve parametrized by arc length. Pick a number
c with a < c < b. The involute of β starting at β(c) is the curve α : (a, b) → R2 given by
α(s) = β(s) + (c − s)β ′(s).
One can show that if α is the involute of β, then the evolute of α is β. For the details of
the proof, one can see [Gr-Abb-Sa], section 4.3.
18

β ( s ) + ( c ? s ) β ' ( s )

β ' ( s )

β ( s ) β ( c )

Figure 20. The dotted curve is the trace of the involute

of the continuous curve.

We also mention the notion of pedal curve.

Definition 1.5.5. Let α : (a, b) → R2 be an arbitrary curve (not necessarily parametrized
by arc length). Pick a point P in the plane. The pedal curve of α with respect to P is the
curve β, where β(t) is the orthogonal projection of P on the tangent line to α at α(t), for
all values of t.
Some more details about this can be found for instance in [Gr-Abb-Sa], section 4.6.
1.6. Examples of Curves. Here are a few examples of important plane curves (see also
sections 1.2 and 1.3).
1. The cycloid is the trajectory described by the motion of a point on a circle which
rolls along a straight line.
Exercise. Assuming that the radius of the circle is 1, find a parametrization of the cycloid
(as α(t) = (x(t), y(t))). Find all points t with α′ (t) = 0 (these are called singular points).
You may want to assume that the circle rolls along the x axis and the initial position of the
point on the circle is the origin O. Determine the length of the piece of the cycloid which
corresponds to a complete (that is, of 3600 ) rotation of the circle.
2. The cardioid is the trajectory described by the motion of a point on a circle which
rolls along a (fixed) circle of the same radius.
Exercise. Assuming that the radius of the circles is 1, find a parametrization of the cardioid
(as α(t) = (x(t), y(t))). Then find the curvature of the cardioid.

Figure 21. The initial position: after that, the dotted

circle is rotated around the other one, and the point runs
along the cardioid.

3. The lemniscate of Bernoulli. By definition, this is the locus of all points in the
plane with the property that the product of distances to two given points (called foci) is
 19

constant, equal to the squared half-distance between the foci8. In other words, if F1 and F2
are the foci, then the lemniscate consists of all points P with
 2
kF1 F2 k
kP F1k · kP F2 k = .
2
There are two ways of describing this curve: implicitly and explicitly (see the remark fol-
lowing Definition 1.2.1). Let’s start with the implicit equation. Choose the two foci of
coordinates (−a, 0) and (a, 0). Then a point P (x, y) is on the lemniscate if and only if
(x − a)2 + y 2 (x + a)2 + y 2 = a4 .



(7)
One can see9 that this is equivalent to
√ √
a 2 cos t a 2 sin t cos t
(8) x= ,y = ,
1 + sin2 t 1 + sin2 t
where t is in R (these give a parametrization of the lemniscate).

Figure 22. The lemniscate of Bernoulli and its foci.

4. The catenary is the curve assumed by a hanging flexible wire supported at its ends.
Considerations from Physics show that the catenary can be described implicitly by10
x
y = a cosh ,
a
where a is a positive number. The obvious parametrization of the catenary is
t
α(t) = (t, a cosh ).
a
Exercise. Find a parametrization by arc length of the catenary, then find the curvature in
terms of s.
5. The tractrix is the trace of the curve
t
α(t) = (sin t, cos t + log(tan )),
2
where 0 < t < π.
Exercise. Show that for any point P on the tractrix the length of the line segment between
P and the y axis is constant (independent of t).
One can show that the evolute of the tractrix is a catenary (see for instance [Gr-Abb-Sa],
section 4.2.).

References
[Gr-Abb-Sa] A. Gray, E. Abbena, and S. Salamon, Modern Differential Geometry with MATHEMATICA,
3rd edition
8Because we want the midpoint of the segment determined by the foci to be on the curve.
9Check at least that x, y given by (8) satisfy (7).
10Recall that the hyperbolic cosine is cosh(u) := eu +e−u .
2
20

Figure 23. The catenary: think that you pick two ar-
bitrary points on it; the piece of the curve between them
is what a hanging wire attached at those two points looks
like.

Figure 24. The tractrix: the trajectory of a point on

the end of an inextensible rod as the other end of the rod
is pulled vertically along the vertical line.