Chapter 1

Curves in the plane and in space

In this chapter, we discuss two mathematical formulations of the intuitive
notion of a curve. 1 The precise relation between them turns out to be quite
subtle, so we begin by giving some examples of curves of each type and
practical ways of passing between them.

1.1 What is a curve?

If asked to give an example of a curve, you might give a straight line, say
y − 2x = 1 (even though this is not curved!), or a parabola, say y − x2 = 0.
x2 y 2
or perhaps circle, say x2 + y 2 = 1 or an ellipse + = 1.
9 4
More generally, we have also seen curves given by the general second
degree equation

ax2 + 2hxy + by 2 + 2gx + 2f + c = 0

represents either a pair of lines, an ellipse, a hyperbola or a parabola. All of

these curves are described by means of their Cartesian equation

f (x, y) = c,

where f is a function of x and y and c is a constant. From this point of view,

a curve is a set of points, namely

C = {(x, y) ∈ R2 |f (x, y) = c}. (1.1)

These examples are all curves in the plane R2 , but we can also consider curves
in R3 for example, the x-axis in R3 is the straight line given by y = 0, z = 0,
1
These are lecture notes based on TYBSc syllabus of Pune university delivered by
Dr. V. V. Acharya in the academic year 2010-2011, 2011-2012 and 2015-16 in Fergusson
College, Pune. These lecture notes were also used in the Annual Foundation schools held
at Bhaskaracharya Pratishthana, Pune.November 1, 2019

1
 f

2 Dr. V. V. Acharya

4.
2.
3.
1.
2.

−2. −1. 0 1. 2. c
1.
−1.

−2. −2. −1. 0 1. 2. 3.

y − 2x = 1 y − x2 = 0

2.
e
1.

1.
d −3. −2. −1. 0 1. 2. 3.
−1.
−1. 0 1.
−1. −2.
x2 y 2
x2 + y 2 = 1 + = 1.
9 4

Figure 1.1: Some level Curves

and more generally a line in R3 might be defined by a pair of equations

x − x1 y − y1 z − z1
= = ,
l m n
where l, m, n are directed cosines of the line. We have also studied that
intersection of two planes is a line while the intersection of a sphere with a
plane or another sphere is a circle. More generally, a curve in R3 might be
defined by a pair of equations

f1 (x, y, z) = c1 , f2 (x, y, z) = c2 .

Curves of this kind are called level curves, the idea being that the curve in
Differential Geometry 3

Eq. 1.1, for example, is the set of points (x, y) in the plane at which the
quantity f (x, y) reaches the level c.
But there is another way to think about curves which turns out to be more
useful in many situations. For this, a curve is viewed as the path traced out
by a moving point. Thus, if γ(t) is the position of the point at time t, the
curve is described by a function γ of a scalar parameter t with vector values
(in R2 for a plane curve, in R3 for a curve in space). We use this idea to give
our first formal definition of a curve in Rn (we shall be interested only in the
cases n = 2 or 3, but it is convenient to treat both cases simultaneously).

Definition 1.1 A parametrized curve in Rn is a map γ : (α, β) → Rn , for

some α, β with −∞ ≤ α < β ≤ ∞.

The symbol (α, β) denotes the open interval

(α, β) = {t ∈ R|α < t < β} .

A parametrized curve, whose image is contained in a level curve C, is called

a parametrization of (part of) C. The following examples illustrate how to
pass from level curves to parametrized curves and back again in practice.
Example 1.1 Let us find a parametrization γ(t) of the parabola y = x2 . If
γ(t) = (γ 1 (t), γ 2 (t)), the components γ 1 (t) and γ 2 (t)of γ must satisfy

γ 2 (t) = γ 1 (t)2 (1.2)

for all values of t in the interval (α, β) where γ is defined (yet to be decided),
and ideally every point on the parabola should be equal to (γ 1 (t), γ 2 (t)) for
some value of t ∈ (α, β).
Of course, there is an obvious solution to Eq. 1.2: take γ 1 (t) = t, γ 2 (t) =
t2 . To get every point on the parabola we must allow t to take every real
number value (since the x-coordinate of γ(t) is just t, and the x-coordinate
of a point on the parabola can be any real number), so we must take (α, β)
to be (−∞, ∞). Thus, the desired parametrization is

γ : (−∞, ∞) → R2 , γ(t) = (t, t2 ).

But this is not the only parametrization of the parabola. Another choice is
γ(t) = (t3 , t6 ) (with (α, β) = (−∞, ∞). ). Yet another is (2t, 4t2 ), and of
course there are (infinitely many) others. So the parametrization of a given
level curve is not unique.
4 Dr. V. V. Acharya

Example 1.2 Now we try the circle x2 + y 2√= 1. It is tempting to take

2
x = t as √ in the previous example, so that y = 1 − t (we√could have taken
y = − 1 − t2 . So we get the parametrization γ(t) = (t, 1 − t2 ).√But this
is only a parametrization of the upper half of √ the circle because 1 − t2 is
always ≥ 0. Similarly, if we had taken y = − 1 − t2 , we would only have
covered the lower half of the circle. If we want a parametrization of the
whole circle, we must try again. We need functions γ 1 (t) and γ 2 (t) such that
γ 1 (t)2 + γ 2 (t)2 = 1 for all t ∈ (α, β), and such that every point on the circle
is equal to (γ 1 (t), γ 2 (t)) for some t ∈ (α, β). There is an obvious solution to
Eq. 1.3: γ 1 (t) = cos t and γ 2 (t) = sin t (since cos2 t + sin2 t = 1 for all values
of t). We can take (α, β) = (−∞, ∞), although this is overkill: any open
interval (α, β) whose length is greater than 2π will suffice.
The next example shows how to pass from parametrized curves to level
curves.
Example 1.3
2
Take the parametrized curve (called
an astroid)
1.5

γ(t) = (cos3 t, sin3 t), t ∈ R.

1
2 2
Since cos t + sin t = 1 for all t, the
coordinates x = cos3 t, y = sin3 t of 0.5
the point γ(t) satisfy x2/3 +y 2/3 = 1.
This level curve coincides with the
-1 -0.5 0.5 1
image of the map.

Example 1.4 A cycloid is the plane curve traced out by a point on the
circumference of a circle as it rolls without slipping along a straight line.
Show that, if the straight line is the x-axis and the circle has radius a > 0,
the cycloid can be parametrised as γ(t) = a(t − sin t, 1 − cos t).
Solution. When the circle has rotated through an angle t, its centre has
moved to (at, a), so the point on the circle initially at the origin is now at
the point (a(t − sin t), a(1 − cos t)).
Generalise this by finding parametrisations of an epicycloid (resp.
hypocycloid), the curve traced out by a point on the circumference
of a circle as it rolls without slipping around the outside (resp.
inside) of a fixed circle.
Differential Geometry 5

Let the fixed circle have radius a, and the moving circle radius b (so that
b < a in the case of the hypocycloid), and let the point P of the moving
circle be initially in contact with the fixed circle at (a, 0). When the moving
circle has rotated through an angle ϕ the line joining the origin to the point
of contact of the circles makes an angle θ with the positive x-axis, where
aθ = bϕ. The point P is then at the point

γ(θ) = ((a + b) cos θ) − b cos(θ + ϕ), (a + b) sin θ − b sin(θ + ϕ))

= ((a + b) cos θ − b cos((a + b)θ/b), (a + b) sin θ − b sin((a + b)θ/b))

in the case of the epicycloid, and

γ(θ) = ((a − b) cos θ) − b cos(ϕ − θ), (a − b) sin θ − b sin(ϕ − θ))

= ((a − b) cos θ − b cos((a − b)θ/b), (a − b) sin θ − b sin((a − b)θ/b))

in the case of the hypocycloid.

 
2 1
Example 1.5 Show that γ(t) = cos t − , sin t cos t, sin t is a parametri-
2
sation of the curve of intersection of the circular cylinder
 ofradius 21 and axis
1
the z-axis with the sphere of radius 1 and centre − , 0, 0 . (This is called
2
Viviani’s Curve).

Figure 1.2: Viviani’s curve

We shall be studying parametrized curves (and later, surfaces) using

methods of calculus. Such curves and surfaces will be described almost ex-
clusively in terms of smooth functions: a function f : (α, β) → R is said to
dn f
be smooth if the derivative n exists for all n ≥ 1 and all t ∈ (α, β). If f (t)
dt
6 Dr. V. V. Acharya

and g(t) are smooth functions, it follows from standard results of calculus
that the sum f (t) + g(t), product f (t)g(t), quotient f (t)/g(t), and composite
f (g(t)) are smooth functions, where they are defined.
To differentiate a vector-valued function such as γ(t), we differentiate
componentwise: if γ(t) = (γ 1 (t), γ 2 (t), . . . , γ n (t)), then
 2  2
d γ 1 d2 γ 2 d2 γ n
 
dγ dγ 1 dγ 2 dγ n dγ
= , ,..., , 2 = , ,..., , etc.
dt dt dt dt dt dt2 dt2 dt2
dγ d2 γ
To save space, we often denote by γ̇(t), 2 by γ̈(t), etc. We say that
dt dt
dn γ
γ is smooth if the derivatives n exist for all n ≥ 1 and all t ∈ (α, β); this
dt
is equivalent to requiring that each of the components γ 1 , γ 2 , . . . , γ n of γ is
smooth. From now on, all parametrized curves studied will be assumed to
be smooth.
Definition 1.2 If γ is a parametrized curve, its first derivative γ̇(t) is called
the tangent vector of γ at the point γ(t).2
γ(t + δt) − γ(t)
To see the reason for this terminology, note that the vector
δt
is parallel to the chord joining the points γ(t) and γ(t + δt) of the image
C of γ. As δt tends to zero the length of the chord also tends to zero, but
we expect that the direction of the chord becomes parallel to that of the
tangent to C at γ(t). But the direction of the chord is the same as that of the
γ(t + δt) − γ(t)
vector which tends to dγ/dt as δt tends to zero. Of course,
δt
this only determines a well defined direction tangent to the curve if dγ/dt is
non-zero. If that condition holds, we define the tangent line to C at a point
p of C to be the straight line passing through p and parallel to the vector
dγ/dt.
Theorem 1.1 If the tangent vector of a parametrized curve is constant, the
image of the curve is (part of) a straight line.
Proof. If γ̇(t) = a for all t, where a is a constant vector, we have, integrating
component-wise,
Z Z
dγ
γ(t) = dt = adt = ta + b,
dt
2
at the parameter value t.
Differential Geometry 7

where b is another constant vector. If a 6= 0, this is the parametric equation

of the straight line parallel to a and passing through the point b. If a = 0,
the image of γ is a single point (namely, b).

Example 1.6 For a logarithmic spiral γ(t) = (et cos t, et sin t), show that
the angle between γ(t) and the tangent vector at γ(t) is independent of t.
Solution: Note that γ̇(t) = (et (cos t − sin t), et (sin t + cos t)), so the angle
between γ(t) and γ̇(t) is given by

γ(t) · γ̇(t) e2t (cos2 t − sin t cos t + sin2 t + sin t cos t) 1

cos θ = = p =√ ,
kγk kγ̇k e2t (cos t − sin t)2 + (cos t + sin t)2 2
π
and so θ = .
4

Example 1.7 Show that the curve C with Cartesian equation y 2 = 2

√x(1−x )
is not connected. For what range of values of t is γ(t) = (t, t − t3 ) a
parametrization of C? What is the image of this parametrization?

Figure 1.3: y 2 = x(1 − x2 )

8 Dr. V. V. Acharya

Example 1.8 (Witch of Agnesi) The witch of Agnesi (pronounced

‘Anyesi’), sometimes called the witch of Maria Agnesi (named for Maria
Agnesi) is the curve defined as follows.
Starting with a fixed circle, a point O on the circle is chosen. For any
other point A on the circle, the secant line OA is drawn. The point M is
diametrically opposite O. The line OA intersects the tangent of M at the
point N. The line parallel to OM through N, and the line perpendicular to
OM through A intersect at P. As the point A is varied, the path of P is the
witch.

Figure 1.4: Witch of Agnesi

Suppose the point O is the origin, and that M is on the positive y-axis.
Suppose the radius of the circle is a. Then the curve has Cartesian equation
8a3
y= .
x2 + 4a2
Note that if a = 1/2, then this equation becomes rather simple:
1
y= .
x2 +1
Parametrically, if θ is the angle between OM and OA, measured clock-
wise, then the curve is defined by the equations

x = 2a tan θ, y = 2a cos2 θ.

Another parametrization, with being the angle between OA and the x-axis,
increasing anti-clockwise is

x = 2a cot θ, y = 2a sin2 θ.
Differential Geometry 9

1.5

0.5

0.5 1 1.5 2 2.5 3

-0.5

-1

-1.5

Figure 1.5: The limacon

Example 1.9 The limacon is the parametrized curve

γ(t) = ((1 + 2 cos t) cos t, (1 + 2 cos t) sin t), t ∈ R

(see the diagram below). Note that γ has a self-intersection at the origin in
the sense that γ(t) = 0 for t = 2π/3 and for t = 4π/3. The tangent vector is

γ̇(t) = (− sin t − 2 sin 2t, cos t + 2 cos 2t).

p p
In particular, γ̇(2π/3) = ( 3/2, −3/2), γ̇(4π/3) = (− 3/2, −3/2). So what
is the tangent vector of this curve at the origin? Although γ̇(t) is well defined
for all values of t, it takes different values at t = 2π/3 and t = 4π/3, both of
which correspond to the point 0 on the curve.

This example shows that we must be careful while talking about a ‘point’
of a parametrized curve γ : strictly speaking, this should be the same thing as
a value of the curve parameter t,and not the corresponding geometric point
γ(t) ∈ Rn .

1.2 Arc-length
We recall that, if v = (v1 , . . . , vn ) is a vector in Rn , its length is
q
kvk = v12 + · · · + vn2 .
10 Dr. V. V. Acharya

If u is another vector in Rn , ku − vk is the length of the straight line segment

joining the points u and v in Rn .
To find a formula for the length of a parametrized curve γ, note that, if
δt is very small, the part of the image C of γ between γ(t) and γ(t + δt)is
nearly a straight line, so its length is approximately

kγ(t + δt) − γ(t)k .

γ(t + δt) − γ(t)
Again, since δt is small, is nearly equal to γ̇(t), so the length
t
is approximately
kγ̇(t)k δt. (1.3)
If we want to calculate the length of a (not necessarily small) part of C, we
can divide it into segments, each of which corresponds to a small increment
δt in t, calculate the length of each segment using (1.3), and add up the
results. Letting δt tend to zero should then give the exact length.

This motivates the following definition:

Definition 1.3 The arc-length of a curve γ starting at the point γ(t0 ) is

Differential Geometry 11

the function s(t) given by

Z t
s(t) = kγ̇(u)k du (1.4)
t0

Thus, s(t0 ) = 0 and s(t) is positive or negative according to whether t is

larger or smaller than t0 . If we choose a different starting point γ(t˜0 ), the
R t˜
resulting arc-length s̃ differs from s by the constant t00 kγ̇(u)k du

Example 1.10 For a logarithmic spiral γ(t) = (et cos t, et sin t) we have

γ(t) = (et (cos t − sin t), et (sin t − cos t)).

Hence,
kγ̇(t)k2 = e2t (cos t − sin t)2 + (sin t − cos t)2 = 2e2t .
 

Hence, the arc-length of γ starting at γ(0) = (1, 0) (for example) is

Z t√ √
2eu du = 2 et − 1 .


s=
0

If s is the arc-length of a curve γ starting at γ(t0 ), we have

d t
Z
ds
= kγ̇(u)k du = kγ̇(t)k (1.5)
dt dt t0
ds
Thinking of γ(t) as the position of a moving point at time t, is the
dt
speed of the point (rate of change of distance along the curve). For this
reason, we make

Definition 1.4 If γ : (α, β) → Rn is a parametrised curve, it’s speed at the

point γ(t) is kγ̇(t)k , and γ is said to be a unit-speed curve if γ̇(t) is a unit
vector for all t ∈ (α, β).

Proposition 1.1 Let n(t) be a unit vector that is a smooth function of a

parameter t. Then, the dot product

n(t) · ṅ(t) = 0

for all t, i.e. ṅ(t) is zero or perpendicular to n(t) for all t.

In particular, if γ is a unit speed curve, then γ̈ is zero or perpendicular to γ̇.
12 Dr. V. V. Acharya

Proof. We use the ’product formula’ for differentiating dot products of

vector-valued functions a(t) and b(t):

d db da
(a(t) · b(t)) = a(t) · + · b(t).
dt dt dt

Using this to differentiate both sides of the equation n(t) · n(t) = 1 with
respect to t, gives

ṅ(t) · n(t) + n(t) · ṅ(t) = 0 i. e. 2ṅ(t) · n(t) = 0.

The last part follows by taking n(t) = γ̇(t).

1.3 Reparametrisation
Definition 1.5 A parametrised curve γ̃ : (α̃, β̃) → Rn is a reparametrisation
of a parametrised curve γ : (α, β) → Rn if there is a smooth bijective map
φ : (α̃, β̃) → (α, β) (the reparametrisation map) such that the inverse map
φ−1 : (α, β) → (α̃, β̃) is also smooth and

γ̃(t̃) = γ(φ(t̃) for all t̃ ∈ (α̃, β̃).

Note that, since φ has a smooth inverse, γ is a reparametrisation of γ̃ :

γ̃(φ−1 (t)) = γ(φ(φ−1 (t))) = γ(t) for all t ∈ (α, β).

Remark 1.1 Note that reparametrization is an equivalence relation.

Example 1.11 Note that γ(t) = (cos t, sin t) and γ̃(t) = (sin t, cos t) are
parametrizations of the unit circle x2 + y 2 = 1.
To see that γ is a reparametrisation of γ̃, we have to find a reparametri-
sation map φ such that (cos φ(t), sin φ(t)) = (sin t, cos t). One solution is
φ(t) = π2 − t.

Definition 1.6 A point γ(t) of a parametrized curve γ is called a regular

point if γ̇(t) 6= 0; otherwise γ(t) is a singular point of γ. A curve is regular
if all of its points are regular.
Differential Geometry 13

Proposition 1.2 Any reparametrisation of a regular curve is regular.

Proof. Suppose that γ and γ̃ are related as in definition, let t = φ(t̃),
and let ψ = φ−1 so that t̃ = ψ(t). Differentiating both sides of the equation
φ(ψ(t)) = t with respect to t and using the chain rule gives
dφ dψ
= 1.
dt̃ dt
dφ
This shows that is never zero. Since γ̃(t̃) = γ(φ(t̃)), another applica-
dt̃
tion of the chain rule gives
dγ̃ dγ dφ
= ,
dt̃ dt dt̃
dγ̃ dγ
which shows that is never zero if is never zero.
dt̃ dt
Proposition 1.3 If γ(t) is a regular curve, its arc-length s, starting at any
point of γ, is a smooth function of t.
Proof. We have already seen that (whether or not γ is regular) s is a
ds
differentiable function of t and = kγ̇(t)k .
dt
To simplify the notation, assume from now on that γ is a plane curve,
say γ(t) = (u(t), v(t)), where
√ u and v are smooth functions of t. Define
2
f : R → R by f (u, v) = u + v 2 , so that
2

ds
= f (u̇2 + v̇ 2 ). (1.6)
dt
The crucial point is that f is smooth on R2 \ {(0, 0}, which means that all
the partial derivatives of f of all orders exist and are continuous functions
except at the origin (0, 0). For example,
∂f u ∂f v
=√ , =√ ,
∂u u + v ∂v
2 2 u + v2
2

are well defined and continuous except where u = v = 0, and similarly for
higher derivatives. Since γ is regular, u̇ and v̇ are never both zero, so the
chain rule and Eq. (1.6) shows that ds/dt is smooth.

Proposition 1.4 A parametrised curve has a unit-speed reparametrisation

if and only if it is regular.
14 Dr. V. V. Acharya

Proof. Suppose first that a parametrised curve γ : (α, β) → Rn has a unit-

speed reparametrisation γ̃, with reparametrisation map φ. Letting t = φ(t̃),
we have
dγ̃ dγ dt
γ̃(t̃) = γ(t). Hence, = .
dt̃ dt dt̃
dγ̃ dγ dt
Thus, = .
dt̃ dt dt̃

dγ̃ dγ
Since γ̃ is unit-speed, = 1, so clearly cannot be zero.
dt̃ dt
dγ
Conversely, suppose that the tangent vector is never zero. Note that
dt
ds dγ
= > 0 for all t, where s is the arc·length of γ starting at any point of
dt dt
the curve, and by Proposition 1.4 s is a smooth function of t. It follows from
the inverse function theorem of multivariable calculus that s : (α, β) → R
is injective, that its image is an open interval (α̃, β̃), and that the inverse
map s−1 : (α̃, β̃) → (α, β) is smooth. We take φ = s−1 and let γ̃ be the
corresponding reparametrization of γ, so that γ̃(s) = γ(t). Then,

dγ̃ ds dγ dγ̃ ds dγ ds
= . Hence, = = .
ds dt dt ds dt dt dt
dγ̃
Thus, = 1.
ds

Corollary 1.1 Let γ be a regular curve. Suppose γ̃ is a unit-speed repara-

metrisation of γ:
γ̃(u(t)) = γ(t) for all t,
where u is a smooth function of t. Then, if s is the arc-length of γ (starting
at any point), we have
u = ±s + c, (1.7)
where c is a constant. Conversely, if u is given by Eq. (1.7) for some value
of c and with either sign, then γ̃ is a unit-speed reparametrisation of γ.

Example 1.12 Show that for the parabola y = x2 , γ 1 (t) = (t, t2 ) and
γ 2 (t) = (t3 , t6 ) are parametrisations. Further, show that γ 2 (t) is not a
reparametrisation of γ 1 (t).
Differential Geometry 15

Solution: For the parametrization γ 1 (t)(t) = (t, t2 ) of the parabola y = x2 ,

γ̇(t) = (1, 2t) is obviously never zero, so γ is regular. But γ 2 (t) = (t3 , t6 ) is
also a parametrization of the same parabola. This time, γ̇ 2 (t) = (3t2 , 6t5 ),
and this is zero when t = 0, so γ 2 is not regular. Any reparametrization of
a regular curve is regular. Hence, the result.
Theorem 1.2 Let f (x, y) be a smooth function of two variables (which
means that all the partial derivatives of f, of all orders, exist and are contin-
uous functions). Assume that, at every point of the level curve
C = (x, y) ∈ R2 |f (x, y) = 0 ,


fx and fy are not both zero. If P is a point of C, with coordinates (x0 , y0 ),

say, there is a regular parametrized curve γ(t), defined on an open interval
containing 0, such that γ passes through P when t = 0 and γ(t) is contained
in C for all t.
Exercise Set
1. Show that γ(t) = ( 54 cos t, 1 − sin t, − 35 cos t) is a unit speed curve.
2. Show that γ(t) = (t, cosh t) is a regular curve.
3. Show that γ(t) = (et cos t, et sin t) is a regular curve.
4. Show that γ(t) = (a cos t, b sin t) is a simple closed curve.
(1 + t)3/2 (1 − t)3/2 t
 
5. Find the curvature of the curve γ(t) = , ,√ .
3 3 2
6. State the necessary and sufficient conditions so that a unit speed curve
γ(t) in R3 is part of a circle.
7. A map γ : I → R3 is called a curve of class C k if each of the coordi-
nate functions in the expression γ(t) = (x(t), y(t), z(t)) has continuous
derivatives up to order k. If γ is merely continuous, we say that γ is of
class C 0 . A curve γ is calIed simple if the map γ is one-to-one.
Let γ : I → R3 be a simpIe curve of class C 0 . We say that a has a weak
tangent at t = t0 ∈ I if the line determined by γ(t0 + h) and γ(t0 ) as
a limit position when h → 0. We say that γ has a strong tangent at
t = t0 if the line determined by γ(t0 + h) and γ(t0 + k) has a limit
position when h, k → 0. Show that
16 Dr. V. V. Acharya

(a) γ(t) = (t3 , t2 ), t ∈ R, has a weak tangent but not a strong tangent
at t = 0.
(b) If γ : I → R3 is of class C 1 and regular at t = t0 , then it has a
strong tangent at t = t0 .
(c) The curve given by γ(t) = (t2 , t|t|), t ∈ R is of class C 1 but not of
class C 2 . Draw a sketch of the curve and its tangent vectors.
8. Let γ : [a, b] → R3 be a differentiable curve, where [a, b] is a closed and
bounded interval. For every partition
a = t0 < t1 < · · · < tn = b
of [a, b], consider the sum ni=1 kγ(ti ) − γ(ti−1 )k = l(γ, P ), where P
P
stands for the given partition. The norm kP k ( of a partition P is
defined as
kP k = max(ti − ti−1 ), i = 1, . . . , n.
Geometrically, l(γ, P ) is the length of a polygon inscribed in γ([a, b])
with vertices in γ(ti) (see Fig. Ex1).
The point of the exercise is to show that the arc length of γ([a, b]) is,
in some sense, a limit of lengths of inscribed poIygons.

Figure Ex 1

Prove that given  >0 there exists δ > 0 such that if kP k < δ then
Z b
kγ̇(t)k − l(γ, P ) < .
a
 Chapter 2
How Much Does a Curve Curve?
In this chapter, we associate to any curve in R3 two scalar functions,
called its curvature and torsion. The curvature measures the extent to which
a curve is not contained in a straight line (so that straight lines have zero
curvature), and the torsion measures the extent to which a curve is not
contained in a plane (so that plane curves have zero torsion). It turns out
that the curvature and torsion together determine the shape of a curve.
2.1 Curvature
We want to find a measure of how ‘curved’ a curve is. Since this ‘curvature’
should depend only on the ‘shape’ of the curve:

(i) the curvature should be unchanged when the curve is reparametrised.

Further, the measure of curvature should agree with our intuition in
simple special cases, for example:

(ii) the curvature of a straight line should be zero, and large circles should
have smaller curvature than small circles.

Bearing (ii) in mind, we get a clue as to what the definition of curvature

should be from Proposition 1.1: this tells us that, if γ is a plane curve with
γ̇ = 0 everywhere, the curve γ is part of a straight line, and hence should
have zero curvature. So we might be tempted to define the curvature of γ to
be kγ̈k (we take the norm because we want the curvature to be a scalar, not
a vector). Unfortunately, however, this depends (in a fairly complicated way)
on the parametrisation of γ. So let us remove this freedom to reparametrise
by insisting that γ is unit-speed, so that kγ̇k = 1 everywhere. (Actually,
this does not quite rule out the possibility of reparametrising - see Corollary
1.1.) So we make

Definition 2.1 If γ is a unit-speed curve with parameter s, its curvature

κ(s) at the point γ(s) is defined to be kγ̈k .

17
18 Dr. V. V. Acharya

The first part of condition (ii) will certainly be satisfied. As to the second
part, consider the circle centred at (x0 , y0 ) and of radius R. This has a unit-
speed parametrisation
s s
γ(s) = (x0 + R cos , y0 + R sin ).
R R
We have γ̇(s) = (− sin Rs , cos Rs ), Hence,



 kγ̇(s)k
 s  = 1, showing s  that γ
is indeed unit-speed, and hence γ̈(s) = − R1 cos , − R1 sin . so the
R R
curvature of the circle is inversely proportional to its radius. As to condition
(i), recall from Corollary 1.1 that, if u = ±s + c, and c is a constant. Then,
by the chain rule,
dγ dγ du dγ
= =± .
ds du ds du
Hence,
d2 γ d2 γ
 
d dγ
= ± = .
ds2 du du du2
This shows that the curvature of the curve computed using the unit-speed
parameter s is the same as that computed using the unit-speed parameter u.
But what if we are given a curve γ(t) that is not unit-speed? If γ is regular
then it has a unit-speed reparametrisation γ̃. We define the curvature of γ
to be that of the unit-speed curve γ̃. But since it is not always possible to
find the unit-speed reparametrisation explicitly, we really need a formula for
the curvature in terms of γ and t only.

Proposition 2.1 Let γ(t) be a regular curve in R3 . Then, its curvature is

kγ̈ × γ̇k
κ= (2.1)
kγ̇k3

where the × indicates the vector (or cross) product and the dot denotes d/dt.
Proof. Let γ̃ (with parameter s) be a unit-speed reparametrisation of γ,
and let us denote d/ds by a dash. Then, by the chain rule,

dγ̃ ds dγ
= = γ̇, so
ds dt dt
! 2
d2 γ̃ d γ̇ γ̈ ds
dt
− γ̇ ddt2s
= = .
ds2 ds ds ds 3


dt dt
Differential Geometry 19
 2
ds
Now, = kγ̇k2 = γ̇ · γ̇, and differentiating with respect to t gives
dt
ds d2 s
· = γ̇ · γ̈. Thus,
dt dt2
ds 2 d2 s
− γ̇ ds


d2 γ̃ γ̈ dt dt dt2 γ̈(γ̇ · γ̇) − γ̇(γ̇ · γ̈)
= = .
ds2 ds 4 kγ̇k4


dt

d2 γ̃ γ̈(γ̇ · γ̇) − γ̇(γ̇ · γ̈)

κ= = .
ds 2
kγ̇k4
Using the vector triple product identity

a × (b × c) = b(a · c) − c(a · b),

(where a, b, c ∈ R3 ), we get

γ̈(γ̇ · γ̇) − γ̇(γ̇ · γ̈) = γ̇ × (γ̈ × γ̇) .

Further, γ̇ and γ̈ × γ̇ are perpendicular vectors, so

kγ̇ × (γ̈ × γ̇)k = kγ̇k kγ̈ × γ̇k .

γ̈(γ̇ · γ̇) − γ̇(γ̇ · γ̈)

Hence, κ =
kγ̇k4
kγ̇ × (γ̈ × γ̇)k kγ̇k kγ̈ × γ̇k kγ̈ × γ̇k
= 4 = 4 = .
kγ̇k kγ̇k kγ̇k3

2.2 Plane Curves

For plane curves, it is possible to refine the definition of curvature slightly
and give it an appealing geometric interpretation.
Suppose that γ(s) is a unit-speed curve in R2 . Denoting d/ds by a dot, let
t = γ̇ be the tangent vector of γ; note that t is a unit vector. There are two
unit vectors perpendicular to t; we make a choice by defining ns , the signed
unit normal of γ, to be the unit vector obtained by rotating t anti-clockwise
by π/2.
20 Dr. V. V. Acharya

Note that γ̈ is perpendicular to t, and hence parallel to ns . Thus, there

is a number κs such that
γ̈ = κs ns . (2.2)
The scalar κs is called the signed curvature of γ (it can be positive, negative
or zero). Note that, since kns k = 1, we have

κ = kγ̈k = kκs ns k = |κs | , (2.3)

so the curvature of γ is the absolute value of its signed curvature.

The signed curvature has a simple geometric interpretation:

Proposition 2.2 Let γ(s) be a unit-speed plane curve, and let ϕ(s)1 be the
angle through which a fixed unit vector must be rotated anti-clockwise to
bring it into coincidence with the unit tangent vector t of γ.2 Then,
dϕ
κs = . (2.4)
ds
Proof. Let a be the fixed unit vector and let b be the unit vector obtained
by rotating a anti-clockwise by π/2. Then, t = a cos ϕ + b sin ϕ.
dϕ dϕ
Hence, ṫ = (−a sin ϕ + b cos ϕ) . Hence, ṫ · a = − sin ϕ . Note that
ds ds
ṫ = κs ns . Hence,
dϕ
ṫ · a = κs (ns · a) = − sin ϕ . (2.5)
ds
But the angle between ns and a is ϕ + π/2, since t must be rotated anti-
clockwise by π/2 to bring it into coincidence with ns
Hence, ns · a = cos(ϕ + π/2) = − sin ϕ. Inserting this into (2.5) gives the
required result.

The next result shows that a unit-speed plane curve is essentially deter-
mined once we know its signed curvature at each point of the curve. The
meaning of ‘essentially’ here is ‘up to a rigid motion of R2 . Recall that a rigid
motion of R2 is a map M : R2 → R2 of the form

M = Ta ◦ Rθ ,
1
Note that, even though the angle ϕ is only determined up to adding an integer multiple
of 2π, the derivative dϕ
ds is well defined.
2
Thus, the signed curvature is the rate at which the tangent vector of the curve rotates.
Differential Geometry 21

where Rθ is an anti·clockwise rotation by an angle θ about the origin,

Rθ (x, y) = (x cos θ − y sin θ, x sin θ + ycosθ),

and TA is the translation by the vector a, TA (v) = v + a, for any vectors

(x, y) and v ∈ R2 . .

Theorem 2.1 Let κ : (α, β) → R be any smooth function. Then, there is

a unit-speed curve γ : (α, β) → R2 whose signed curvature is κ. Further, if
γ̃ : (α, β) → R2 is any other unit-speed curve whose signed curvature is κ,
there is a rigid motion M of R2 such that

g̃(s) = M (γ(s)) for all s ∈ (α, β).

Proof. For the first part, fix s0 ∈ (α, β) and define, for any s ∈ (α, β),
Z s
ϕ(s) = κ(u)du, (cf.P roposition2.2),
s0
Z s Z s 
γ(s) = cos ϕ(t)dt, sin ϕ(t)dt .
s0 s0

Then, the tangent vector of γ(s) is γ̇(s) = (cos ϕ(s), sin ϕ(s)) which is a
unit vector making an angle ϕ(s) with the x-axis. Thus, γ is unit-speed and,
by Proposition 2.2, its signed curvature is

d s
Z
dϕ
= κ(u)du = κ(s).
ds ds s0

For the second part, let ϕ̃(s) be the angle between the x·axis and the unit
tangent vector γ̃˙ of γ̃. Thus,

˙
γ̃(s) = (cos ϕ̃(s), sin ϕ̃(s)),
Z s Z s 
γ̃(s) = cos ϕ̃(t)dt, sin ϕ̃(t)dt + γ̃(s0 ). (2.6)
s0 s0

By Proposition 2.2,
dϕ̃
= κ(s).
ds
22 Dr. V. V. Acharya
Rs
ϕ̃(s) = s0 κ(u)du + ϕ̃(s0 ). Inserting this into Eq. (6), and writing a for the
constant vector γ̃(s0 ) and θ for the constant scalar ϕ̃(s0 ), we get
Z s Z s
γ̃(s) = Ta ( cos(ϕ̃(t) + θ)dt, sin(ϕ̃(t) + θ)dt)
s0 s0
Z s Z s
= Ta (cos θ cos ϕ̃(t)dt − sin θ sin ϕ̃(t)dt,
s0 s0
Z s Z s
sin θ cos ϕ̃(t)dt + cos θ sin ϕ̃(t)dt))
s0 s0
Z s Z s
= Ta Rθ ( cos ϕ̃(t)dt, sin ϕ̃(t)dt)
s0 s0
= Ta Rθ (γ(s)) .

Proposition 2.3 Any regular plane curve whose curvature is a positive con-
stant is part of a circle.
Proof. Let κ be the (constant) curvature of the curve γ, and let κs be its
signed curvature. Then, κs = ±κ.
A priori, we could have κs = κ at some points of the curve and κs = −κ.
at others, but in fact this cannot happen since κs is a continuous function of
s, so the Intermediate Value Theorem tells us that, if κs takes both the value
κ and the value −κ, it must take all values between. Thus, either κs = κ at
all points of the curve, or κs = −κ at all points of the curve. In particular,
κs is constant.
The idea now is to show that, whatever the value of κs , we can find a
parametrised circle whose signed curvature is κs . The theorem then tells us
that any curve whose signed curvature is κs can be obtained by applying a
rigid motion to this circle. Since rotations and translations obviously take
circles to circles, it follows that any curve whose signed curvature is constant
is (part of) a circle.

2.3 Curves in R3
Let γ be a unit speed curve. Then t = γ̇. We assume that ṫ = γ̈ 6= 0 for all
1 1
t. Then we define κ = ṫ = kγ̈k and n = ṫ = γ̈. Thus,
κ κ
dt
= κn. (2.7)
ds
Differential Geometry 23

Also, we define b = t × n. Note that

b = t × n, n = b × t, t=n × b.
db dt dn dn
Hence, = ×n+t× = t× . Note that b is a unit vector.
ds ds ds ds
db db
Hence, is orthogonal to b as well as t. Hence, is parallel to n. We
ds ds
define
db
= −τ n. (2.8)
ds
Note that n = b × t. Differentiating w.r.t. s, we get
dn db dt
= ×t+b× = −τ n × t + b × κn = −κt + τ b. (2.9)
ds ds ds
Theorem 2.2 Let γ(t) be a regular curve in R3 with nowhere vanishing
d
curvature. Then, denoting by a dot, its torsion is given by
dt
...
(γ̇ × γ̈) · γ
τ= (2.10)
kγ̇ × γ̈k2

We first treat the case in which γ is unit-speed. In this
case, we have ḃ = τ n.

Hence, τ = −n · ḃ = −n · t ×˙ n = −n · ṫ × n + t × ṅ = −n · (t × ṅ) . Now
1 1
n = t = γ̈. Hence,
κ κ
  
1 d 1
τ = − γ̈ · γ̇ × γ̈
κ dt κ
  
1 1 1 ...
= − γ̈ · γ̇ × − 2 κ̇γ̈ + γ
κ κ κ
  
1 1 ...
= − γ̈ · γ̇ × γ
κ κ
1 ...
= 2 (γ̇ × γ̈) · γ ,
κ
... ...
as γ̈ · (γ̇ × γ̈) = 0 and −γ̈ · (γ̇ × γ ) = (γ̇ × γ̈) · γ . This agrres with (2.10)
as γ is unit-speed, γ̇ and γ̇ are perpendicular, so

kγ̇ × γ̈k = kγ̇k kγ̈k = κ.

24 Dr. V. V. Acharya

In the general case, let s be arc-length along γ and denote d/ds by a dash.
Then
ds 0
γ̇ = γ
dt
 2
ds 00 d2 s 0
γ̈ = γ + 2γ
dt dt
 3
... ds ds d2 s 00 d3 s 0
γ = γ 000 + 3 γ + 3γ .
dt dt dt2 dt
 3
ds
Hence, γ 0 × γ 00 = (γ̇ × γ̈)
dt
 6
... 0 ds
γ (γ × γ 00 ) = γ 000 × (γ̇ × γ̈)
dt
and so ...
(γ̇ × γ̈) · γ (γ 0 × γ 00 ) · γ 000
= .
kγ̇ × γ̈k2 kγ 0 × γ 00 k2
Theorem 2.3 Let γ(s) and γ̃(s) be two unit-speed curves in R3 with the
same curvature κ(s) > 0 and the same torsion τ (s) for all s. Then, there is
a rigid motion M of R3 such that
γ̃(s) = M (γ(s)) for all s.
Further, if κ and τ are smooth functions with κ > 0 everywhere, there is a
unit-speed curve in R3 whose curvature is κ and whose torsion is τ.
Proof. Let t, n and b be the tangent vector, principal normal and bi-
normal of γ and let t̃, ñ and b̃ be those of γ̃. Let s0 be a fixed value of
the parameter s. Since {t(s0 ), n(s0 ), b(s0 )} and {t̃(s0 ), ñ(s0 ), b̃(s0 )} are both
right-handed orthonormal bases of R3 , there is a rotation about the origin
of R3 that takes t(s0 ), n(s0 ) and b(s0 ) to t̃(s0 ), ñ(s0 ) and b̃(s0 ) respectively.
Further, there is a translation that takes γ(s0 ) to γ̃(s0 ) (and this has no
effect on t, n and b). By applying the rotation followed by the translation,
we can therefore assume that
γ(s0 ) = γ̃(s0 ), t(s0 ) = t̃(s0 ), n(s0 ) = ñ(s0 ), b(s0 ) = b̃(s0 ) (2.11)
Consider the expression
A(s) = t(s)t̃(s) + n(s)ñ(s) + b(s)b̃(s).
Differential Geometry 25

Using (2.11), we get A(s0 ) = 3. On the other hand, since t and t̃ are unit
vectors, t · t̃ ≤ 1, with equality holding if and only if t = t̃ and similarly
n · ñ ≤ 1, and b · b̃ ≤ 1. It follows that A(s) ≤ 3, with equality holding if and
only if t(s) = t̃(s), n(s) = ñ(s) and b(s) = b̃(s). Thus, if we can prove that
A is constant, it will follow in particular that t(s) = t̃(s), i.e. that γ̇ = γ̃,˙
and hence that γ(s) − γ̃(s) is a constant. But by (2.11) again, this constant
vector must be zero, so γ = γ̃.
We are therefore reduced to proving that A is constant. But, using the
dA
Frenet-Serret equations, it follows that = 0.
ds
26 Dr. V. V. Acharya
 Chapter 3
Global properties of curves

All the properties of curves that we have discussed so far are ‘local’: they
depend only on the behaviour of a curve near a given point and not on the
‘global’ shape of the curve. Proving global results about curves often requires
concepts from topology, in addition to the calculus techniques we have used
in the first two chapters of this book. Since we are not assuming that readers
of this book have extensive familiarity with topological ideas, we will not be
able to give complete proofs of some of the global results about curves that
we discuss in this chapter.

3.1 Simple closed curves

In this chapter, we shall consider plane curves of the following type.

Definition 3.1 A simple closed curve in R2 is a closed curve in R2 that has

no self-intersections.

It is a standard, but highly non-trivial, result of the topology of R2 , called

the Jordan Curve Theorem, that any simple closed curve in the plane has an
‘interior’ and an ‘exterior’: more precisely, the complement of the image of
γ (i.e., the set of points of R2 that are not in the image of γ) is the disjoint
union of two subsets of R2 , denoted by int(γ) and ext(γ), with the following
properties:
(i) int(γ) is bounded, i.e., it is contained inside a circle of sufficiently
large radius.
(ii) ext(γ) is unbounded.
(iii) Both of the regions int(γ) and ext(γ) are connected, i.e., they have
the property that any two points in the same region can be joined by a curve
contained entirely in the region (but any curve joining a point of int(γ) to a
point of ext(γ) must cross the curve γ).

27
28 Dr. V. V. Acharya

Example 3.1 The ellipse γ(t) = (p cos t, q sin t), where p and q are non-zero
constants, is a simple closed curve with period 2π. The interior and exterior
of γ are, of course, given by
2
y2 2
y2
   
2 x 2 x
(x, y) ∈ R | 2 + 2 < 1 and (x, y) ∈ R | 2 + 2 > 1 .
p q p q
respectively.
Not all examples of simple closed curves have such an obvious interior
and exterior, however. Is the point p in the interior or the exterior of the
simple closed curve shown below?
Example 3.2 The limacon in Example 1.1.7 is closed but is not a simple
closed curve as it has a self-intersection . See Exercise 3.1.1.
The fact that a simple closed curve has an interior and an exterior enables
us to distinguish between the two possible orientations of γ. We shall say
that γ is positively-oriented if the signed unit normal ns of γ. (see Section
2.2) points into int(γ) at every point of γ. This can always be achieved by
replacing the parameter t of γ by −t, if necessary. In the diagrams below, the
arrow indicates the direction of increasing parameter. Is the simple closed
curve shown above positively-oriented?

3.2 The isoperimetric inequality

The area contained by a simple closed curve γ is
ZZ
A( int (γ)) = dxdy. (3.1)
int γ
This can be computed by using the following theorem.
Theorem 3.1 (Green’s Theorem)Let f (x, y), g(x, y) be smooth functions
(i.e., functions with continuous partial derivatives of all orders), and let γ be
a positively oriented simple closed curve. Then,
ZZ   Z
∂g ∂f
− dxdy = f (x, y)dx + g(x, y)dy, (3.2)
∂x ∂y
int γ γ

if γ is a positively-oriented simple closed curve in R2 .

Differential Geometry 29

Theorem 3.2 If γ(t) = (x(t), y(t)) is a positively-oriented simple closed

curve in R2 with period a then
Za
1
A(int (γ)) = (xẏ − ẋy)dt. (3.3)
2
0

1 1
Proof. Taking f (x, y) = − y, g(x, y) = x in Green’s theorem, we get
2 2
Z Z ZZ   
1 1 1
f (x, y)dx + g(x, y)dy = xdy − ydx = − − dxdy
2 2 2
γ γ int γ
Z a ZZ
1
Thus, (xẏ − ẋy)dt = dxdy = A(int (γ)).
2
0 int γ

Theorem 3.3 (Isoperimetric Inequality) Let γ be a simple closed curve,

let `(γ) be its length and let A(intγ) be the area of its interior. Then,
1
A(intγ) ≤ `(γ)2
4π
and equality holds if and only if γ is a circle.

Lemma 3.1 (Wirtinger’s Inequality) Let F : [0, π] → R be a smooth

function such that F (0) = F (π) = 0. Then, prove that
Z π Z π
dF 2
( ) dt = F (t)2 dt.
0 dt 0

Further, equality holds if and only if F (t) = A sin t for all t ∈ [0, π], where A
is a constant.
Proof. Let G(t) = F (t)/ sin t. Then, denoting d/dt by a dot as usual,
Z π
dF
( )2 dt
dt
Z0 π
= (Ġ sin t + G cos t)2 dt
Z0 π Z π Z π
2 2
= Ġ sin t + 2 GĠ sin t cos t + G2 cos2 tdt
0 0 0
30 Dr. V. V. Acharya

Integrating by parts:
Z π Z π
2 π
2 GĠ sin t cos tdt = G sin t cos t|0 − G2 (cos2 t − sin2 t)dt
0 0
Z π
= G2 (sin2 t − cos2 t)dt
0

Hence,
Z π 2 Z π Z π Z π
dF 2 2 2 2 2
dt = Ġ sin t + G (sin t − cos t)dt + G2 (cos2 t)dt
0 dt
Z0 π 0 0

= (G2 + Ġ2 ) sin2 tdt

Z0 π Z π
2
= F dt + Ġ2 sin2 tdt
0 0

and so Z π Z π Z π
dF
( )2 dt − 2
F dt = Ġ2 sin2 tdt
0 dt 0 0

The integral on the right-hand side is obviously ≥ 0, and it is zero if and

only if Ġ = 0 for all t, i.e., if and only if G(t) is equal to a constant, say A,
for all t, which means that F (t) = A sin t.

3.3 Four Vertex Theorem

Definition 3.2 A vertex of a curve γ(t) in R2 is a point where its signed
curvature κs has a stationary point, Le. where dκs /dt = 0.

Theorem 3.4 (Four Vertex Theorem) Every convex simple closed curve
in R2 has at least four vertices. 1
Proof. Let γ be a parametrization of a convex simple closed curve in R2 ,
and let ` be its length. Assume for a contradiction that γ has fewer than
four vertices. We show first that there is a straight line L that divides γ into
two segments, in one of which κ̇s > 0 and in the other κ̇s ≤ 0 (or possibly
κ̇s ≥ 0 on one and κ̇s < 0 on the other). Indeed, κs attains all of its values on
1
The conclusion of this theorem actually remains true without the assumption of con-
vexity, but the proof is then more difficult than the one we are about to give.
Differential Geometry 31

the closed interval [0, l], so κs must attain its maximum and minimum values
at some points P and Q of γ. We can assume that P 6= Q, since otherwise
κ̇s would be constant, γ would be a circle,and every point of γ would be a
vertex.
If P and Q were the only vertices of γ, we would have κ̇s > 0 on one of
the segments into which the line through P and Q divides γ and κ̇s < 0 on
the other.
Let a be a unit vector perpendicular to L, so that γ · a > 0 on one side
of L and γ · a < 0 on the other. Then, the quantity κ̇s (γ · a) is either always
> 0 or always < 0, except at the two points in which L intersects the curve.
It follows that Suppose now that there is just one more vertex, say R. Then,
P, Q and R divide γ into three segments, on each of which either κ̇s > 0 or
κ̇s < 0. It follows that there are two adjacent segments on which κ̇s > 0 or
two on which κ̇s < 0 (except at the point at which the two segments meet).
This proves our assertion.
Let a be a unit vector perpendicular to L, so that γ · a > 0 on one side
of L and γ · a < 0 on the other. Then, the quantity κ̇s (γ · a) is either always
> 0 or always < 0, except at the two points in which L intersects the curve.
It follows that Z a
κ̇s (γ · a) ≥ 0, (3.4)
0

as this integral is definitely > 0 in the first case and < 0 in the second. But,
using the equation ṅs = −κs t, we get

d d
κ̇s γ = (κs γ) − κs γ̇ = (κs γ + ns ),
dt dt

so the integrand on the left-hand side of (3.4) is the derivative of (κs γ + ns ) ·

a = λ, say. Since γ is `-periodic,

γ(t + `) = γ(t) for all t,

differentiating with respect to t shows that the tangent vector t of γ is also

`-periodic:
t(t + `) = γ̇(t + `) = γ̇(t) = t(t).
π
Rotating by 2
gives
ns (t + `) = ns (t),
32 Dr. V. V. Acharya

and hence κs (t + `) = κs (t). It follows that λ(t + `) = λ(t) for all t, so the
integral in (3.4) is equal to
Z `
λ̇(t)dt = λ(`) − λ(0) = 0. (3.5)
0

This contradiction proves that γ must have at least four vertices.

 Chapter 4
Surfaces
4.1 History
1
The Greek astronomer Claudius Ptolemy (100−168 A.D.) was the first
known to produce the data for creating a map showing the inhabited world
as it was known to the Greeks and Romans of about 100− 150 A.D. In his
work Geography he explains how to project a sphere onto a flat piece of paper
using a system of gridlines - longitude and latitude.
As we know from peeling oranges and trying to flatten the peels on a
table, the sphere cannot be projected onto the plane without distortions and
therefore certain compromises must be made.
The orthographic projection (a), which was known to the Egyptians and
Greeks more than 2000 years ago, modifies both angles and areas, but the
directions from the centre of projection are true. Probably the most widely
used projection is the stereographic projection (b) usually attributed to Hip-
parchus (190−120 B.C.). It is a conformal projection, i.e., it preserves angles
(at the expense of areas). It also maps circles to circles, no matter how large
(great circles are mapped into straight lines), but a loxodrome is plotted
as a spiral. A loxodrome is a line of constant bearing and of vital impor-
tance in navigation. In 1569, the Flemish cartographer Gerardus Mercator
(1512 − 1594), whose goal was to produce a map which sailors could use
to determine courses, overcame this drawback with his conformal cylindrical
Mercator projection (c) which draws every loxodrome as a straight line. Nei-
ther the stereographic nor the Mercator projections preserve areas however.
Johann Heinrich Lambert (1728−1777) found the first equiareal projection
(d) in 1772, at the cost of giving up the preservation of angles.
All these projections can be seen as functions that map a part of the
surface of the sphere to a planar domain and the inverse of this mapping
1
These are lecture notes based on TYBSc syllabus of Pune university delivered by Dr.
V. V. Acharya in the academic year 2010-2011,2011-12 and 2015-16 in Fergusson College,
Pune. November 1, 2019

33
34 Dr. V. V. Acharya

is usually called a parameterization. Many of the principles of parametric

surfaces and differential geometry were developed by Carl Friedrich Gauss
(1777−1855).

4.2 Homeomorphism

Definition 4.1 A subset U of Rn is called open if, whenever a is a point

in U, there is a positive number  such that every point u ∈ Rn within a
Differential Geometry 35

distance  of a is also in U :

a ∈ U and ku − ak <  =⇒ u ∈ U.

For example, the whole of Rn is an open set, as is

Dr (a) = {u ∈ Rn | ku − ak < r},

the open ball with centre a and radius r > 0. 2 (If n = 1, an open ball is
called an open interval; if n = 2 it is called an open disc.) However,

Dr (a) = {u ∈ Rn | ku − ak ≤ r},

is not open because however small the positive number  is, there is a point
within a distance of the point (a1 + r, a2 , ..., an ) ∈ Dr (a) (say) that is not in
Dr (a) (for example, the point (a1 + r + 2 , a2 , ..., an )).
Next, if X and Y are subsets of Rm and Rn , respectively, a map f : X →
Y is said to be continuous at a point a ∈ X if points in X near a are mapped
by f to points in Y near f (a). More precisely, f is continuous at a if, given
any number  > 0, there is a number δ > 0 such that

u ∈ X and ku − ak < δ =⇒ kf (u) − f (a)k < .

Then f is said to be continuous if it is continuous at every point of X.

Composites of continuous maps are continuous.
In view of the definition of an open set, this is equivalent to the following:
f is continuous if and only if, for any open set V of Rn , there is an open set
U of Rm such that U ∩ X = {x ∈ X|f (x) ∈ V }.
If f : X → Y is continuous and bijective, and if its inverse map f −1 :
Y → X is also continuous, then f is called a homeomorphism and X and Y
are said to be homeomorphic.

4.3 Surfaces, An Introduction

Definition 4.2 A subset S of R3 is a surface if, for every point p ∈ S, there
is an open set U in R2 and an open set W in R3 containing p such that S ∩W
is homeomorphic to U. A subset of a surface S of the form S ∩ W, where W
is an open subset of R3 , is called an open subset of S.
2
Show that Dr (a) is an open set in Rn ..
36 Dr. V. V. Acharya

A homeomorphism s : U → S ∩ W as in this definition is called a surface

patch or parametrization of the open subset S ∩ W of S. A collection of such
surface patches whose images cover the whole of S is called an atlas of S.

Example 4.1 Every plane in R3 is a surface with an atlas consisting of a

single surface patch. In fact, let a be a point on the plane, and let p and q
be two unit vectors that are parallel to the plane and perpendicular to each
other.3 If r is any point of the plane, r − a is parallel to the plane, and so
r − a = up + vq for some scalars u and v. Thus, the desired surface patch is
σ(u, v) = a + up + vq. By the definition, σ is onto. It is easy to see that σ
is one-one. Hence, the inverse map exists. Further, its inverse map is given
by σ −1 (r) = ((r − a) · p, (r − a) · q). Using these formulas, we get that σ and
σ −1 are continuous, and hence that σ is a homeomorphism.

Example 4.2 A circular cylinder is the set of points of R3 that are at a fixed
distance (the radius of the cylinder) from a fixed straight line (its axis). For
example, the circular cylinder of radius 1 and axis the z-axis, which we shall
call the unit cylinder, is

S = {(x, y, z) ∈ R3 |x2 + y 2 = 1}.

Figure 4.1: Right circular cylinder

The simplest parametrization of S is σ(u, v) = (cos u, sin u, v). We note

that σ(u, v) ∈ S for all (u, v) ∈ R2 , and every point of S is of this form.
3
We take p and q to be unit vectors which are perpendicular to each other so that
the inverse can be computed easily.
Differential Geometry 37

Moreover, σ is continuous. However, σ is not injective, and so is not a

homeomorphism, because σ(u, v) = σ(u + 2π, v) for all (u, v).
To get an injective map we can restrict u to lie in an interval of length 2π,
say 0 ≤ u < 2π. However, although the restriction σ|V of σ to V = {(u, v) ∈
R2 |0 ≤ u < 2π} is injective, V is not an open subset of R2 and so σ|V is not
a surface patch. The largest open subset of R2 contained in V is

U = {(u, v) ∈ R2 |0 < u < 2π},

and the restriction σ|U of σ to U is a surface patch. However, σ|U does not
cover the whole of S, but only the open subset obtained by removing the line
x = 1, y = 0 from S.
To get an atlas for S we therefore need at least one more surface patch.
We can take σ|U1 , where U1 = {(u, v) ∈ R2 | − π < u < π}; this covers the
open subset of S obtained by removing the line x = 1, y = 0. Every point of
S is in the image of at least one of the surface patches σ|U , σ|U1 , so σ|U , σ|U1
is an atlas for S, and S is a surface.

Example 4.3 A sphere is the set of points of R3 that are at a fixed distance
(the radius of the sphere) from a fixed point (its centre). For example, the
sphere of radius 1 and centre the origin, called the unit sphere, is

S 2 = {(x, y, z) ∈ R3 |x2 + y 2 + z 2 = 1}.

The most popular parametrization of S 2 is that given by latitude θ and

longitude ϕ: if p is a point of the sphere, the line through p parallel to the
z-axis intersects the xy-plane at a point q, say; then, θ is the angle between
q and p and ϕ is the angle between q and the positive x-axis. The circles on
the sphere corresponding to a constant value of θ are called parallels; those
corresponding to a constant value of ϕ are called meridians.
To obtain an explicit formula for this parametrization, we must express p
in terms of the angles θ and ϕ. From the right-angled triangle with vertices 0,
p and q, we see that the z-component of p is sin θ. The x- and y-components
can be found from the right-angled triangle in the xy-plane with vertices 0,
q and r, where r is the foot of the perpendicular from q to the x-axis.
The length of the hypotenuse of this triangle is kqk = cos θ, so the x- and
y-components of p are kqk cosϕ = cos θ cos ϕ and kqk sin ϕ = cos θ sin ϕ,
respectively. Putting all these together gives

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

38 Dr. V. V. Acharya

1
0.5
0
-0.5
-1
1

0.5

-0.5

-1
-1
-0.5
0
0.5
1

Figure 4.2: Sphere

Denote the right-hand side of this equation by σ(θ, ϕ); this is the latitude-
longitude parametrization of S 2 . See Fig. 4.3.

Figure 4.3: Latitude-Longitude Parametrization

As in the case of the cylinder, σ is not injective since (for example)

σ(θ, ϕ) = σ(θ, ϕ + 2π). In fact, a little thought shows that to cover the
whole sphere, it is sufficient to take
π π
− ≤ θ ≤ , 0 ≤ ϕ ≤ 2π.
2 2
Differential Geometry 39

(i) (ii)

Figure 4.4: An atlas for sphere

However, the set of points (θ, ϕ) satisfying these inequalities is not an open
subset of R2 . The largest open set consistent with the above inequalities is
π π
U = {(θ, ϕ)| < θ < , 0 < ϕ < 2π}.
2 2
however, the image of σ|U is not the whole of the sphere, but the open subset
obtained by removing the semicircle C consisting of the points of the sphere
of the form (x, 0, z) with x ≥ 0. (See Fig. 4.4 (i))
To show that the sphere is a surface, we must therefore produce at least
one more surface patch covering the part of the sphere omitted by σ. One
possibility is the patch σ
b obtained by first rotating σ by π about the z-axis
π
and then by about the x-axis. Explicitly, σ b : U → R3 is given by
2
b (θ, ϕ) = (− cos θ cos ϕ, − sin θ, − cos θ sin ϕ).
σ

(the open set U is the same as for σ). The image of σ b is the open subset
of S 2 obtained by removing the semicircle C̃ consisting of the points of the
sphere of the form (x, y, 0) with x ≤ 0. (Fig. 4.4 (ii))
It is clear that C and C̃ do not intersect, so the union of the images of
σ|U and σ b |U is the whole sphere.

Example 4.4 Show that the unit cylinder can be covered by a single surface
patch, but that the unit sphere cannot.
Solution.
√ For the first part, let U = {(u, v) ∈ R2 |0 < u2 + v 2 < π 2 }, let
r = u2 + v 2 , and define σ : U → R3 by σ(u, v) = ( ur , vr , tan(r − π2 )). It
40 Dr. V. V. Acharya

is easy to see that σ is a homeomorphism. Hence, the unit cylinder can be

covered by a single surface patch
If S 2 could be covered by a single surface patch s : U → R3 , then S 2
would be homeomorphic to the open subset U of R2 . As S 2 is a closed and
bounded subset of R3 , it is compact. Hence, U would be compact, and hence
closed. But, since R2 is connected, the only non-empty subset of R2 that
is both open and closed is R2 itself, and this is not compact as it is not
bounded.

Example 4.5 The circular cone with vertex a point v, axis a straight line l
passing through v, and angle α, where 0 < α < π2 , is the set of points p in
R3 such that the straight line through v and p makes an angle α with the
line l. For example, if v is the origin, l is the z-axis and α = π4 , the circular
cone is
S = {(x, y, z) ∈ R3 |x2 + y 2 = z 2 }.
To see that this is not a surface, suppose that σ : U → S ∩ W is a surface
patch containing the vertex (0, 0, 0) of the cone, and let a ∈ U correspond to
the vertex. We can assume that U is an open ball with centre a, since any
open set U containing a must contain such an open ball. The open set W
must obviously contain a point p in the lower half S− of S where z < 0 and
a point q in the upper half S+ where z > 0; let b and c be the corresponding
points in U.

Figure 4.5: Vertex of Cone

It is clear that there is a curve π in U passing through b and c, but not

passing through a. This is mapped by σ into the curve γ = σ ◦ π lying
Differential Geometry 41

entirely in S, passing through p and q, and not passing through the vertex.
(It is true that γ will, in general, only be continuous, and not smooth, but this
does not affect the argument.) This is clearly impossible for if we take the
projection map Π : R3 → R defined by Π(x, y, z) = z then Π is a continuous
map and Π(p) < 0 and Π(q) > 0. Hence, there must exist a point r on γ
such that Π(r) = 0. But the only possible such point on the double cone is
(0, 0, 0) which does not lie on γ.
If we remove the vertex, however, we do get a surface S− ∪ S+ . It has
an atlas consisting of the two surface patches σ ± : U → R3 , where U =
R2 − {(0, 0)}, given by the inverse of projection onto the xy plane:
√
σ ± (u, v) = (u, v, ± u2 + v 2 ).
As the example of the sphere shows, a point a of a surface S will generally
lie in the image of more than one surface patch. In general, suppose then that
σ : U → S ∩W and σ e :Ue → S ∩W f are two patches such that a ∈ S ∩W ∩ W f.

Figure 4.6: Transition Maps

e are homeomorphisms, σ −1 (S ∩ W ∩ W
Since σ and σ e −1 (S ∩ W ∩ W
f ) and σ f)
are open sets V ⊆ U and Ve ⊆ U e , respectively. The composite homeomor-
−1
phism σ ◦ σ e : Ve → V is called the transition map from σ to σ e . If we
denote this map by Φ, we have
σ
e (e
u, ve) = σ(F (e u, ve) ∈ Ve .
u, ve)) for all (e
Example 4.6 Define surface patches σ x± : U → R3 for S 2 by solving the
equation x2 + y 2 + z 2 = 1 for x in terms of y and z:
√
σ x± (u, v) = (± 1 − u2 − v 2 , u, v),
42 Dr. V. V. Acharya

defined on the open set U = {(u, v) ∈ R2 |u2 + v 2 < 1}. Define σ y± and σ z± .
similarly (with the same U ) by solving for y and z, respectively. Show that
these six patches give S 2 the structure of a surface.

4.4 Smooth surfaces

In Differential Geometry we use calculus to analyse surfaces (and other ge-
ometric objects). We must be able to make sense of the statement that a
function on a surface is differentiable, for example. For this, we have to
consider surfaces with some extra structure.
First, if U is an open subset of Rm , we say that a map f : U → Rn is
smooth if each of the n components of f, which are functions U → R, have
continuous partial derivatives of all orders. The partial derivatives of f are
then computed componentwise. For example, if m = 2 and n = 3, and
f(u, v) = (f1 (u, v), f2 (u, v), f3 (u, v)), then
   
∂f ∂f1 ∂f2 ∂f3 ∂f ∂f1 ∂f2 ∂f3
= , , , = , ,
∂u ∂u ∂u ∂u ∂v ∂v ∂v ∂v
and similarly for higher derivatives. We often use the following abbreviations:
∂f ∂f ∂ 2f ∂ 2f ∂ 2f ∂ 2f
= fu , = fv , 2 = fuu , = fuv , = fvu , = fvv
∂u ∂v ∂u ∂u∂v ∂v∂u ∂v∂v
and so on. From advanced calculus 4 we know that fuv = fvu if f is smooth.
It now makes sense to say that a surface patch σ : U → R3 is smooth. But
we shall require one further condition.
Definition 4.3 A surface patch σ : U → R3 is called regular if it is smooth
and the vectors σ u and σ v are linearly independent at all points (u, v) ∈ U.
Equivalently, σ should be smooth and the vector product σ u × σ v should be
non-zero at every point of U.5
We can finally define the class of surfaces which we want to study.
Definition 4.4 If S is a surface, an allowable surface patch for S is a regular
surface patch σ : U → R3 such that σ is a homeomorphism from U to an
4
Note that since second order partial derivatives are continuous, mixed partial deriva-
tives are equal.
5
In other words, we are saying that the Jacobian matrix of σ is of rank 2.
Differential Geometry 43

open subset of S. A smooth surface is a surface S such that, for any point
p ∈ S, there is an allowable surface patch σ as above such that p ∈ σ(U ).
A collection A of allowable surface patches for a surface S such that every
point of S is in the image of at least one patch in A is called an atlas for the
smooth surface S.
Example 4.7 The plane in Example 1 is a smooth surface. For
σ(u, v) = a + up + vq
is clearly smooth and σ u = p and σ v = q are linearly independent because
p and q were chosen to be perpendicular unit vectors.
Example 4.8 The unit cylinder in Example 2 is a smooth surface. Indeed,
σ(u, v) = (cos u, sin u, v)
is clearly smooth and σ u = (− sin u, cos u, 0), σ v = (0, 0, 1) are mutually
orthonormal vectors and hence are linearly independent for all (u, v), so σ|U
and σ̃|Ũ are regular surface patches.
Example 4.9 For the unit sphere S 2 in Example 3, it is again clear that σ
and σ̃ are smooth. As for regularity,
σ θ = (− sin θ cos ϕ, sin θ sin ϕ, cos θ), σ ϕ = (− cos θ sin ϕ, cos θ cos ϕ, 0),
σ θ × σ ϕ = (− cos2 θ cos ϕ, − cos2 θ sin ϕ, − sin θ cos θ) and hence
kσ θ × σ ϕ k = | cos θ|.
But if (θ, ϕ) ∈ U, then − π2 < θ < π
2
so cos θ 6= 0. Similarly, one checks that
σ̃ is regular.
In Example 4.6 we gave another family of allowable surface patches cov-
ering the unit sphere S 2 (it is easy to check that they are regular.) An
obvious question is: which of these two atlases should we use to study the
sphere? The answer is that we can use either, or both. The eight patches
in Example 3 and Example 6 together form a third atlas. In most situations
(although not in all.), one might as well use the maximal atlas for a given
surface S consisting of all of its allowable surface patches. The maximal atlas
is independent of any arbitrary choices.
Although not at first sight very interesting, the next two results are very
important for what is to follow.
44 Dr. V. V. Acharya

Proposition 4.1 The transition maps of a smooth surface are smooth.

The proof of this will be given later. The next result is a kind of converse.

Proposition 4.2 Let U and Ũ be open subsets of R2 and let σ : U → R3

be a regular surface patch. Let Φ : Ũ → U be a bijective smooth map with
smooth inverse map Φ−1 : U → Ũ . Then, σ̃ = σ ◦ Φ : Ũ → R3 is a regular
surface patch.
Proof. The patch σ̃ is smooth because any composite of smooth maps is
smooth. As for regularity, let (u, v) = Φ(ũ, ṽ). By the chain rule,
∂u ∂v ∂u ∂v
σ̃ ũ = σu + σ v , σ̃ ṽ = σu + σv ,
∂ ũ ∂ ũ ∂ṽ ∂ṽ
so  
∂u ∂v ∂u ∂v
σ̃ ũ × σ̃ ṽ = − σu × σv (4.1)
∂ ũ ∂ṽ ∂ṽ ∂ ũ
The scalar on the right-hand side of this equation is the determinant of
the Jacobian matrix  ∂u ∂v 
J(Φ) = ∂u∂ ũ ∂ ũ of Φ.
∂v
∂ṽ ∂ṽ

We recall from calculus that, if Ψ and Ψ̃ are two smooth maps between
open sets of R2 ,
J(Ψ̃ ◦ Ψ) = J(Ψ̃)J(Ψ).
(In fact, this is equivalent to the chain rule that expresses the first partial
derivatives of Ψ̃ ◦ Ψ in terms of those of Ψ̃ and Ψ.) Taking Ψ = Φ and
Ψ̃ = Φ−1 , we see that J(Φ−1 ) = J(Φ)−1 . In particular, J(Φ) is invertible, so
its determinant is non-zero and Eq. (4.1) shows that σ̃ is regular.

If regular surface patches σ and σ̃ are related as in this proposition, we

say that σ̃ is a reparametrization of σ, and that Φ is a reparametrization
map. Note that σ is then a reparametrization of σ̃, since σ = σ̃ ◦ Φ−1 .
Note also that, if σ : U → S ∩ W and σ̃ : Ũ → S ∩ W̃ are two allowable
surface patches of a smooth surface S, and if V ⊂ U and Ṽ ⊂ Ũ are the open
subsets such that σ(V ) = σ̃(Ṽ ) = S ∩ W ∩ W̃ , then Φ = σ −1 ◦ σ̃ : Ṽ → V
is bijective, smooth and has a smooth inverse by Proposition 1. Thus, σ̃ is
a reparametrization of σ where they are both defined.
These observations give rise to a very important principle that we shall
use. The principle is that we can define a property of any smooth surface
Differential Geometry 45

provided we can define it for any regular surface patch in such a way that it
is unchanged when the patch is reparametrized. We shall give an important
application of this principle in the next section.
For the rest of this course, by a surface we shall mean a smooth
surface S, and by a surface patch for S we shall mean an allowable
surface patch for S.
Unless we indicate otherwise, we shall also assume that all surfaces
we consider are connected, which means that any two points of S can
be joined by a curve lying entirely in S. This is not a serious restriction,
for it is not difficult to prove that any surface S is a disjoint union
of connected surfaces, each of which is an open subset of S, so S can
be studied by studying each of its connected parts separately. All the
surfaces we have encountered so far are connected except the double
cone of Example 5, which breaks into the union of two disjoint half
cones S± when the vertex is removed, as it must be to have a surface.

4.5 Smooth maps

We want to define the notion of a smooth map f : S1 → S2 , where S1 and S2
are smooth surfaces. It is not obvious how to do this, because so far we only
know how to define smooth maps between open subsets of Euclidean spaces.
In view of the principle stated at the end of the preceding section, we can
assume that S1 and S2 are covered by single surface patches σ 1 : U1 → R3
and σ 2 : U2 → R3 provided we verify that the definition we give is unaffected
by a reparametrization of σ 1 and σ 2 . Since σ 1 and σ 2 are bijective, any
map f : S1 → S2 gives rise to the map σ −1 2 ◦ f ◦ σ : U1 → U2 , and we
say that f is smooth if this map is smooth (we already know what it means
for a map between open subsets of R2 to be smooth). Now suppose that
σ̃ 1 : Ũ1 → R3 and σ̃ 2 : Ũ2 → R3 are reparametrizations of σ 1 and σ 2 , with
reparametrization maps Φ1 : Ũ1 → U1 and Φ2 : Ũ2 → U2 , respectively. We
have to show that the corresponding map σ̃ −1 −1
2 ◦f ◦ σ̃ 1 is smooth if σ 2 ◦f ◦σ 1
is smooth. But this is true, since

σ̃ −1 −1 −1 −1
2 ◦ f ◦ σ̃ 1 = σ̃ 2 ◦ (σ 2 ◦ σ 2 ) ◦ f ◦ (σ 1 ◦ σ 1 )σ̃ 1
= (σ̃ −1 −1 −1
2 ◦ σ 2 ) ◦ σ 2 ) ◦ f ◦ σ 1 ◦ (σ 1 σ̃ 1 )
= Φ−1 −1
2 ◦ (σ 2 ◦ f ◦ σ 1 ) ◦ Φ1

and Φ1 and Φ−1 2

2 are smooth maps (between open subsets of R ).
46 Dr. V. V. Acharya

Note that composites of smooth maps between surfaces are smooth. We

shall be especially interested in smooth maps f : S1 → S2 , which are bijective
and whose inverse map f −1 : S2 → S1 is smooth. Such maps are called
diffeomorphisms, and S1 and S2 are said to be diffeomorphic if there is a
diffeomorphism between them. The following observation will be useful.

Proposition 4.3 Let f : S1 → S2 be a diffeomorphism. If σ 1 is an allowable

surface patch on S1 , then f ◦ σ 1 is an allowable surface patch on S2 .
Proof We can assume that S1 and S2 are covered by single allowable patches
σ 1 : U1 → R3 and σ 2 : U2 → R3 , respectively. Since f is a diffeomorphism,
f (σ 1 (u, v)) = σ 2 (u1 , v1 ) for some (u1 , v1 ) ∈ U2 . This defines a map from U1 to
U2 where F (u, v) = (u1 , v1 ). Note that F = σ −1 2 ◦ f ◦ σ 1 . Hence, F : U1 → U2
−1
is bijective, smooth and F is smooth. Hence, σ 2 ◦ F : U1 → R3 is smooth
map (follows from Proposition 4.2). But σ 2 ◦ F = f ◦ σ 1 . Hence, f ◦ σ 1 is a
regular (an allowable) surface patch on S2 .

A smooth map f : S1 → S2 between smooth surfaces is called a local

diffeomorphism if, for any point p ∈ S1 , there is an open subset O of S1 such
that f (O) is an open subset of S2 and f |O : O → f (O) is a diffeomorphism
(note that open subsets of surfaces are surfaces). It is obvious that every
diffeomorphism is a local diffeomorphism (take O = S1 ). Moreover, Propo-
sition 3 holds if f is a local diffeomorphism, provided that the restriction of
f to the image of σ 1 is injective.

4.6 Level surfaces

As we have already seen, surfaces are often given to us as level surfaces

{(x, y, z) ∈ R3 |f (x, y, z) = 0},

where f is a smooth function. In those examples, we constructed atlases by

fairly ad hoc methods. The following result gives general conditions under
which a level surface is a smooth surface. In fact, it deals with a slightly
more general situation in which different regions of a surface may be defined
by different functions.

Theorem 4.1 Let S be a subset of R3 with the following property: for each
point p ∈ S, there is an open subset W of R3 containing p and a smooth
function f : W → R such that
Differential Geometry 47

(i) S ∩ W = {(x, y, z) ∈ W |f (x, y, z) = 0};

(ii) The gradient ∇f = (fx , fy , fz ) of f does not vanish at p.
Then, S is a smooth surface.
Example 4.10 For the unit sphere S 2 , we can take W = R3 and use the
single function f (x, y, z) = x2 + y 2 + z 2 − 1. Then, ∇f = (2x, 2y, 2z) so
k∇f k = 2 at all points of S 2 . In particular, ∇f is non-zero everywhere on
S 2 . Hence, Theorem 1 tells us that S 2 is a smooth surface.
Example 4.11 For the circular cone, f (x, y, z) = x2 + y 2 − z 2 . Hence, ∇f =
(2x, 2y, −2z), and this vanishes only at the vertex (0, 0, 0). Theorem 1 applies
with W = {(x, y, z) ∈ R3 |z 6= 0}, so the circular cone with the vertex removed
is a smooth surface, as we have already seen.
EXERCISES
Show that the following are smooth surfaces:
(i) x2 + y 2 + z 4 = 1.
(ii) (x2 + y 2 + z 2 + a2 − b2 )2 = 4a2 (x2 + y 2 ), where a > b > 0 are constants.
Example 4.12 Consider the quadric x2 + 2y 2 + 6x − 4y + 3z = 7. Setting
x0 = x + 3, y 0 = y − 1 (a translation), we get
x02 + 2y 02 + 3z = 18.
Setting z 0 = z − 6 (another translation) gives
x02 + 2y 02 + 3z 0 = 0.
Finally, setting x00 = x, y 00 = −y 0 , z 00 = −z 0 (a rotation by π about the x-axis)
gives
1 002 2 002
x + y = z 00 ,
3 3
which is an elliptic paraboloid. It can be parametrized by setting
1 2
x00 = u, y 00 = v, z 00 = u2 + v 2 .
3 3
This corresponds to
1 2
x = u − 3, y = 1 − v, z = 6 − u2 − v 2 ,
3 3
and shows that the given quadric is a smooth surface with an atlas consisting
of the single surface patch
1 2
σ(u, v) = (u − 3, 1 − v, 6 − u2 − v 2 ).
3 3
48 Dr. V. V. Acharya

4.7 Tangent Space

A natural way to study a surface S is via the (smooth) curves γ that lie in
S. This enables us to define the notion of a tangent vector to a surface.

Definition 4.5 A tangent vector to a surface S at a point p ∈ S is the

tangent vector at p of a curve in S passing through p. The tangent space
Tp S of S at p is the set of all tangent vectors to S at p.

To understand the tangent space Tp S, choose a surface patch σ : U → R3

of S such that p is in the image of σ, say σ(u0 , v0 ) = p. If a curve γ lies in
S and passes through p when t = t0 , say, there are functions u(t) and v(t)
such that
γ(t) = σ(u(t), v(t)) (4.2)
for all values of t close to t0 , and u(t0 ) = u0 , v(t0 ) = v0 . The functions u and
v are necessarily smooth; conversely, it is obvious that if t 7→ (u(t), v(t)) is
smooth, then Equation (4.2) defines a curve lying in S.
Proposition 4.4 Let σ : U → R3 be a patch of a surface S containing a
point p ∈ S, and let (u, v) be coordinates in U. The tangent space to S at p is
the vector subspace of R3 spanned by the vectors σ u and σ v (the derivatives
are evaluated at the point (u0 , v0 ) ∈ U such that σ(u0 , v0 ) = p).
Proof Let γ be a smooth curve in S, say γ(t) = σ(u(t), v(t)).
Denoting d/dt by a dot, we have, by the chain rule, γ̇ = σ u u̇+σ v v̇. Thus,
γ̇ is a linear combination of σ u and σ v . (Perhaps it is worth spelling out what
this last equation means: the vectors σ u and σ v are smooth functions of u
and v; in these formulas one replaces u and v by the functions u(t) and v(t);
then the right-hand side of the equation becomes a function of t only, and
the equation says that this function is equal to dγ/dt.)
Conversely, any vector in the vector subspace of R3 spanned by σ u and
σ v is of the form λσ u + µσ v for some scalars λ and µ. Define

γ(t) = σ(u0 + λt, v0 + µt).

Then, γ is a smooth curve in S and at t = 0, i.e., at the point p ∈ S, we

have
γ̇ = λσ u + µσ v .
This shows that every vector in the span of σ u and σ v is the tangent
vector at p of some curve in S.
Differential Geometry 49

Since σ is assumed to be regular, σ u and σ v are linearly independent so

the tangent space is two-dimensional, and will be called the tangent plane
from now on. Note that Definition 5 shows that the tangent plane is indepen-
dent of the choice of patch containing p, even though this is not immediately
obvious from Proposition 4.
Note also that the vectors σ u and σ v that form a basis of the tangent plane
at some point σ(u0 , v0 ) of the surface are the tangent vectors of the parameter
curves on the surface, i.e., the curves u 7→ σ(u, v0 ) and v 7→ σ(u0 , v) (we
shall sometimes describe these curves as ‘the parameter curves u = u0 and
v = v0 ’).
1. Find the equation of the tangent space of each of the following surface
patches at the indicated points:
(i) σ(u, v) = (u, v, u2 − v 2 ), (1, 1, 0).
We note that σ u = (1, 1, 2u) and σ v = (1, 1, −2v). Hence, at (1, 1, 0),
σ u = (1, 0, 2), σ v = (0, 1, −2) so σ u × σ v = (−2, 2, 1) and the equation
of tangent space is −2x + 2y + z = 0.
(ii) σ(r, θ) = (r cosh θ, r sinh θ, r2 ), (1, 0, 1).
We note that σ r = (cosh θ, r sinh θ, 2r) and σ θ = (r sinh θ, r cosh θ, 0)
and at (1, 0, 1), r = 1, θ = 0. Hence, σ r = (1, 0, 2), σ θ = (0, 1, 2)
so σ r × σ θ = (−2, −2, 1) and the equation of the tangent space is
−2x − 2y + z = 0.
2. Show that, if σ(u, v) is a surface patch, the set of linear combinations
of σ u and σ v is unchanged when σ is reparametrized.
Let σ̃(ũ, ṽ) be a reparametrization of σ. Then,
∂ ũ ∂ṽ ∂ ũ ∂ṽ
σu = σ̃ ũ + σ̃ ṽ and σ v = σ̃ ũ + σ̃ ṽ ,
∂u ∂u ∂v ∂v
so σ u and σ v are linear combinations of σ̃ ũ and σ̃ ṽ . Hence, any linear
combination of σ u and σ v is a linear combination of σ̃ ũ and σ̃ ṽ .
The converse is also true since σ is a reparametrization of σ̃.

4.8 Normals and orientability

Since the tangent plane Tp S of a surface S at a point p ∈ S passes through
the origin of R3 , it is completely determined by giving a unit vector perpen-
dicular to it, called a unit normal to S at p. There are, of course, two such
50 Dr. V. V. Acharya

vectors, but Proposition 4 shows that choosing a surface patch σ : U → R3

containing p leads to a definite choice, namely
σu × σv
Nσ = (4.3)
kσ u × σ v k

(with the derivatives evaluated at the point of U corresponding to p), for

this is clearly a unit vector perpendicular to every linear combination of σ u
and σ v . This is called the standard unit normal of the surface patch σ at p.
Unlike the tangent plane, however, Nσ is not quite independent of the
choice of patch σ containing p. In fact, if σ̃ : Ũ → R3 is another surface
patch in the atlas of S containing p, we showed in the proof of Proposition
4.2) that
(σ˜u × σ̃ v ) = det(J(Φ)) (σ u × σ v ) ,
where J(Φ) is the Jacobian matrix of the transition map Φ from σ to σ̃. So
the standard unit normal of σ̃ is
σ̃ u × σ̃ v σu × σv
Nσ̃ = =± = ±Nσ
kσ̃ u × σ̃ v k kσ u × σ v k

where the sign is that of the determinant of J(Φ). This leads to the following
definition.

Definition 4.6 A surface S is orientable if there exists an atlas A for S

with the property that, if Φ is the transition map between any two surface
patches in A, then det(J(Φ)) > 0 where Φ is defined.

The preceding discussion gives the following proposition.

Proposition 4.5 Let S be an orientable surface equipped with an atlas a as

in Definition 6. Then, there is a smooth choice of unit normal at any point
of S: take the standard unit normal of any surface patch in a.

An oriented surface is a surface S together with a smooth choice of unit

normal N at each point, i.e., a smooth map N : S → R3 (meaning that each
of the three components of N is a smooth function S → R) such that, for all
p ∈ S, N (p) is a unit vector perpendicular to Tp S.
Any oriented surface is orientable! To see this, start with the maximal
atlas of S and retain a patch σ(u, v) if σ u × σ v is a positive multiple of N at
Differential Geometry 51

all points in the image of σ, otherwise discard it. The patches that remain
form an atlas a satisfying the condition in Definition 6.
From now onwards, whenever we are dealing with an oriented surface S,
we shall only use surface patches for S whose standard unit normal is the
same as the chosen normal of S.

4.9 Inverse Function Theorem

Suppose f : U → Rn is a smooth map, where U is an open subset of Rm . If
we write (ũ1 , . . . , ũn ) = f (u1 , . . . , um ), the Jacobian matrix of f is
∂ ũ1 ∂ ũ1
 
 ∂u1 · · · ∂um 
 . .. .. 
J(f ) =   . . . . .
 ∂ ũn ∂ ũn 
···
∂u1 ∂um
Theorem 4.2 (Inverse Function Theorem) Let f : U → Rn be a smooth
map defined on an open subset U of Rn (n ≥ 1). Assume that, at some point
x0 ∈ U, the Jacobian matrix J(f ) is invertible. Then, there is an open subset
V of Rn and a smooth map g : V → Rn such that
(i) y0 = f (x0 ) ∈ V,
(ii) g(y0 ) = x0 ,
(iii) g(V ) ⊆ U,
(iv) g(V ) is an open subset of Rn ,
(v) f (g(y)) = y for all y ∈ V.
In particular, g : V → g(V ) and f : g(V ) → V are inverse bijections.
Thus, the inverse function theorem says that, if J(f ) is invertible at some
point, then f is bijective near that point and its inverse map is smooth.

Theorem 4.3 Let f (x, y) be a smooth function of two variables (which

means that all the partial derivatives of f, of all orders, exist and are contin-
uous functions). Assume that, at every point of the level curve
C = {(x, y) ∈ R2 |f (x, y) = 0},
52 Dr. V. V. Acharya

∂f ∂f
and are not both zero. If P is a point of C, with coordinates (x0 , y0 ),
∂x ∂y
say, there is a regular parametrized curve γ(t), defined on an open interval
containing 0, such that γ passes through P when t = 0 and γ(t) is contained
in C for all t.  
∂f ∂f
Proof. We note that n = , is non-zero at every point of C. We
∂x ∂y
∂f
may assume that 6= 0 at P. Then, n is not parallel to the x-axis, so the
∂y
tangent to C at P is not parallel to the y-axis.
This implies that vertical lines x = constant near x = x0 all intersect C
in a unique point (x, y) near P. In other words, the equation

f (x, y) = 0 (4.4)

has a unique solution y near y0 for every x near x0 . 6 Thus, there is a function
g(x), defined for x near x0 , such that y = g(x) is the unique solution of Eq.
4.4 near y0 . We can now define a parametrization γ of the part of C near P
by
γ(t) = (t, g(t)).
Note that g is smooth (which follows from the inverse function theorem),
then γ is regular since ġ = (1, ġ) is obviously never zero.

Theorem 4.4 Let γ be a regular parametrized plane curve, and let γ(t0 ) =
(x0 , y0 ) be a point in the image of γ. Then, there is a smooth real-valued
function f (x, y), defined for x and y in open intervals containing x0 and y0 ,
respectively, and satisfying the conditions in Theorem 4.3, such that γ(t)
is contained in the level curve f (x, y) = 0 for all values of t in some open
interval containing t0 .
Proof. Let γ(t) = (u(t), v(t)), where u and v are smooth functions. Since γ
is regular, at least one of u0 (t0 ) and v 0 (t0 ) is non-zero, say u0 (t0 ). This means
that the graph of u as a function of t is not parallel to the t-axis at t0 .
This implies that any line parallel to the t-axis close to u = x0 intersects
the graph of u at a unique point u(t) with t close to t0 . This gives a function
h(x), defined for x in an open interval containing x0 , such that t = h(x) is
6
Note that this may fail to be the case if the tangent to C at P is parallel to the y-axis
∂f
(i.e., if = 0.)
∂y
Differential Geometry 53

the unique solution of u(t) = x if x is near x0 and t is near t0 . The inverse

function theorem tells us that h is smooth. The function

f (x, y) = y − v(h(x))

has the properties we want.

It is not in general possible to find a single function f (x, y) satisfying

the conditions in Theorem 2 such that the image of γ is contained in the
level curve f (x, y) = 0, for γ may have self-intersections.It follows from the
inverse function theorem that no single function f satisfying the conditions in
Theorem 4.4 can be found that describes a curve near such a self-intersection.

Theorem 4.5 The transition maps of a smooth surface are smooth.

Proof. We want to show that, if σ : U → R3 and σ̃ : Ũ → R3 are two
regular patches in the atlas of a surface S, the transition map from σ to σ̃
is smooth where it is defined. Suppose that a point P lies in both patches,
say σ(u0 , v0 ) = σ̃(ũ0 , ṽ0 ) = P. Write

σ(u, v) = (f (u, v), g(u, v), h(u, v)).

Since σ u and σ v are linearly independent, the Jacobian matrix

 
fu fv
 gu gv 
hu hv

 2 ×
of σ has rank 2 everywhere. Hence, at least one of its three 2 submatrices
f f
is invertible at each point. Suppose that the submatrix u v is invertible
gu gv
at P. (The proof is similar in the other two cases.) By the inverse function
theorem applied to the map F : U → R2 given by

F (u, v) = (f (u, v), g(u, v)),

there is an open subset V of R2 containing F (u0 , v0 ) and an open subset W

of U containing (u0 , v0 ) such that F : W → V is bijective with a smooth
inverse F −1 : V → W.
Since σ : W → σ(W ) is bijective, the projection π : σ(W ) → V given
by π(x, y, z) = (x, y) is also bijective, since π = F ◦ σ −1 on σ(W ). It follows
54 Dr. V. V. Acharya

that W̃ = σ̃ −1 (σ(W )) is an open subset of Ũ and that σ̃ −1 ◦ σ̃ = F −1 ◦ F̃ on

W̃ , where F̃ = π ◦ σ̃. Since F −1 and F̃ are smooth on W̃ , so is the transition
map σ −1 ◦ σ̃. Since σ −1 ◦ σ̃ is smooth on an open set containing any point
(u0 , v0 ) where it is defined, it is smooth.

Theorem 4.6 Let S be a subset of R3 with the following property: for each
point P ∈ S, there is an open subset W of R3 containing P and a smooth
function f : W → R such that
(i) S ∩ W = {(x, y, z) ∈ W |f (x, y, z) = 0};
(ii) The gradient ∇f = (fx , fy , fz ) of f does not vanish at P.
Then, S is a smooth surface.
Proof. Suppose that P = (x0 , y0 , z0 ) and that fz 6= 0 at P. (The proof is
similar in the other two cases.) Consider the map F : W → R3 defined by

F (x, y, z) = (x, y, f (x, y, z)).

 
1 0 0
The Jacobian matrix of F is  0 1 0  , and is clearly invertible at P
fx fy fz
since fz 6= 0. By the inverse function theorem, there is an open subset V
of R3 containing F (x0 , y0 , z0 ) = (x0 , y0 , 0) and a smooth map G : V → W
such that W̃ = G(V ) is open and F : W̃ → V and G : V → W̃ are inverse
bijections.
Since V is open, there are open subsets U1 of R2 containing (x0 , y0 ) and
U2 of R containing 0 such that V contains the open set U1 × U2 of all points
(x, y, w) with (x, y) ∈ U1 and w ∈ U2 . Hence, we might as well assume that
V = U1 × U2 . The fact that F and G are inverse bijections means that
G(x, y, w) = (x, y, g(x, y, w)) for some smooth map g : U1 × U2 → R, and

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

for all (x, y) ∈ U1 , w ∈ U2 .

Define σ : U1 → R3 by σ(x, y) = (x, y, g(x, y, 0)). Then σ is a homeo-
morphism from U1 to S ∩ W̃ (whose inverse is the restriction to S ∩ W̃ of
the projection π(x, y, z) = (x, y)). It is obvious that σ is smooth, and it is
regular because σ x × σ y = (−gx , −gy , 1) is nowhere zero. So σ is a regular
surface patch on S containing the given point P. Since P was an arbitrary
point of S, we have constructed an atlas for S making it into a (smooth)
surface.
 Chapter 5
First and second fundamental forms
Suppose that σ is a surface patch in R3 with standard unit normal N.
we define
E = kσ u k2 , F = σ u · σ v , G = kσ v k2 . (5.1)
Then, the expression

Edu2 + 2F dudv + Gdv 2 (5.2)

is called the first fundamental form of the surface patch σ(u, v). Note that
the coefficients E, F, G and the linear maps du, dv depend on the choice of
surface patch for S, but the first fundamental form itself depends only on S
and p. Let
L = σ uu · N, M = σ uv · N, N = σ vv · N. (5.3)
The expression
Ldu2 + 2M dudv + N dv 2 (5.4)
is called the second fundamental form of the surface patch σ.
Example 5.1 For the plane σ(u, v) = a + up + vq with p and q being
perpendicular unit vectors, we have σ u = p, σ v = q, so E = kσ u k2 =
kpk2 = 1, F = σ u · σ v = p · q = 0, G = kσ v k2 = kqk2 = 1, and the first
fundamental form is simply du2 + dv 2 . Since σ u = p and σ v = q are constant
vectors, we have σ uu = σ uv = σ vv = 0. Hence, the second fundamental form
of a plane is zero.1

Example 5.2 Consider the Mercator’s parametrization of the sphere

σ(u, v) = (sechu cos v, sechu sin v, tanh u).

u −u u −u
Note that sinh u = e −e
2
, cosh u = e +e
2
sinh u
, tanh u = cosh u
1
, sechu = cosh u
.
2 2 2 2
Hence, cosh u − sinh u = 1 and 1 − tanh u = sech u. Further, derivative
1
Note that in this case, we did not compute N.

55
56 Dr. V. V. Acharya

of sinh u, cosh u are cosh u and sinh u respectively. Hence, the derivative of
tanh u and sechu are sech2 u and −sechu tanh u resp.

σ u = (−sechu tanh u cos v, −sechu tanh u sin v, sech2 u)

and σ v = (−sechu sin v, sechu cos v, 0).

Hence,

E = σ u · σ u = sech2 u tanh2 u cos2 v + sech2 u tanh2 u sin2 v + (sech2 u)2

= sech2 u(tanh2 u + sech2 u) = sech2 u
F = −sech2 u tanh u cos v sin v + sech2 u tanh u cos v sin v = 0
G = sech2 u.

Hence, the first fundamental form is

Edu2 + 2F dudv + Gdv 2 = sech2 u(du2 + dv 2 ).

σu × σv
Now, N = = (−sechu cos v, −sechu sin v, − tanh u) = −σ(u, v).
kσ u × σ v k
Note that N 6= 0. Hence, σ is regular,

σ uu = sechu((2 tanh2 u − 1) cos v, (2 tanh2 u − 1) sin v, −2sechu tanh u),

σ uv = (sechu tanh u sin v, −sechu tanh u cos v, 0),
σ vv = (−sechu cos v, −sechu sin v, 0). Hence,
L = N · σ uu = −sech2 u(2 tanh2 u − 1) + 2sech2 u tanh2 u = sech2 u
M = 0 and N = sech2 u

and the second fundamental form is

Ldu2 + 2M dudv + N dv 2 = sech2 u(du2 + dv 2 ).

Note that the first and the second fundamental form of S 2 are same.

Example 5.3 A surface of revolution is the surface obtained by rotating

a plane curve, called the profile curve, around a straight line in the plane.
The circles obtained by rotating a fixed point on the profile curve around
the axis of rotation are called the parallels of the surface, and the curves on
the surface obtained by rotating the profile curve through a fixed angle are
Differential Geometry 57

called its meridians. (This agrees with the use of these terms in geography,
if we think of the earth as the surface obtained by rotating a circle passing
through the poles about the polar axis and we take u and v to be latitude
and longitude, respectively.)
Let us take the axis of rotation to be the z-axis and the plane to be the
xz-plane. Any point p of the surface is obtained by rotating some point q of
the profile curve through an angle v (say) around the z-axis. If
γ(u) = (f (u), 0, g(u))
is a parametrization of the profile curve containing q, p is of the form
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)).
To check regularity, we compute (with a dot denoting d/du):
σ u = (f˙ cos v, f˙ sin v, ġ), σ v = (−f sin v, f cos v, 0),
σ u × σ v = (f ġ cos v, −f ġ sin v, f f˙),
kσ u × σ v k2 = f 2 (f˙2 + ġ 2 ).
Thus, σ u × σ v will be non-vanishing if f (u) is never zero, i.e., if γ does not
intersect the z-axis, and if f˙ and ġ are never zero simultaneously, i.e., if γ is
regular. In this case, we might as well assume that f (u) > 0, so that f (u) is
the distance of σ(u, v) from the axis of rotation. Then, σ is injective provided
that γ does not self-intersect and the angle of rotation v is restricted to lie
in an open interval of length ≤ 2π. Under these conditions, surface patches
of the form σ give the surface of revolution the structure of a surface.
We now calculate the first fundamental form for surface of revolution.
Note that f (u) > 0 for all values of u and that the profile curve u 7→
(f (u), 0, g(u)) is unit-speed, i.e., f˙2 + ġ 2 = 1 (a dot denoting d/du). Then:
σ u = (f˙ cos v, f˙ sin v, ġ), σ v = (−f sin v, f cos v, 0). Hence, E = kσ u k2 =
f˙2 + ġ 2 = 1, F = σ u · σ v = 0, G = kσ v k2 = f 2 . So the first fundamental form
is du2 + f (u)2 dv 2 .
Note that
σu × σv
N= = (ġ cos v, −ġ sin v, f˙),
kσ u × σ v k
σ uu = (f¨ cos v, f¨ sin v, g̈), σ uv = (−f˙ sin v, f˙ cos v, 0), and
σ vv = (−f cos v, −f sin v, 0). Hence,
L = σ uu · N = f˙g̈ − f¨f˙, M = σ uv · N = 0, N = σ vv · N = f ġ,
58 Dr. V. V. Acharya

so the second fundamental form is (f˙g̈ − f¨ġ)du2 + f ġdv 2 .

A special case is the unit sphere S 2 in latitude-longitude coordinates. We
take u = θ, v = ϕ, f (θ) = cos θ, g(θ) = sin θ, giving the first fundamental
form of S 2 as dθ2 + cos2 θdϕ2 , and the second fundamental form of S 2 as
dθ2 + cos2 θdϕ2 . 2
Note that the first fundamental form of S 2 and the second fun-
damental form of S 2 are same.
Yet another special case.
If the surface is the unit cylinder, we can take f (u) = 1, g(u) = u (again, the
conditions f > 0 and f˙2 + ġ 2 = 1 are satisfied). Thus,

σ(u, v) = (cos v, sin v, u).

So the first fundamental form is du2 + f (u)2 dv 2 = du2 + dv 2 .

Further, L = M = 0, N = 1, so the second fundamental form of the
cylinder is dv 2 .

Example 5.4 A ruled surface is a surface that is a union of straight lines,

called the rulings (or sometimes the generators) of the surface.
Suppose that C is a curve in R3 that meets each of these lines. Any point
p of the surface lies on one of the given straight lines, which intersects C at
q, say. If γ : (α, β) → R3 is a parametrization of C with γ(u) = q, and if
δ(u) is a non-zero vector in the direction of the line passing through γ(u),
then
p = γ(u) + δ(u),
for some scalar v. Denoting the right-hand side by σ(u, v), it is clear that
σ : U → R3 is a smooth map, where U = (u, v) ∈ R2 |α < u < β. Moreover,
denoting d/du by a dot,

σ u = ġ + v δ̇, σ v = δ.

Thus, σ is regular if ġ + v δ̇ and δ are linearly independent. This will be true,

for example, if ġ and δ are linearly independent and v is sufficiently small.
Thus, to get a surface, the curve C must never be tangent to the rulings.
An important special case is that in which the rulings are all parallel to
each other; the ruled surface S is then called a generalized cylinder. In the
2
You should compute the first and the second fundamental form using the surface patch
σ(θ, ϕ) = (cos θ cos ϕ, cos θ sin ϕ, sin θ) for the sphere, unit cylinder x2 + y 2 = 1 and the
double cone x2 + y 2 = z 2 . (without vertex).
Differential Geometry 59

above notation, we can take δ to be a constant unit vector, say a, parallel to

the rulings, and the parametrization becomes

σ(u, v) = γ(u) + va.

Sinceσ(u, v) = σ(u0 , v 0 ) ⇐⇒ γ(u) = γ(u0 ) = (v 0 − v)a, for σ to be a injective

(and hence a surface patch), no straight line parallel to a should meet γ in
more than one point. Finally, σ u = ġ, σ v = a, so σ is regular if and only if
γ is never tangent to the rulings.
The parametrization is simplest when γ lies in a plane perpendicular to a
(in fact, this can always be achieved by replacing γ by its perpendicular pro-
jection onto such a plane.). The regularity condition is then clearly satisfied
provided γ̇ is never zero, i.e., provided γ is regular. We might as well take
the plane to be the xy-plane and a = (0, 0, 1) to be parallel to the z-axis.
Then, γ(u) = (f (u), g(u), 0) for some smooth functions f and g, and the
parametrization becomes

σ(u, v) = (f (u), g(u), v).

For example, starting with a circle, we get a circular cylinder. Taking

the circle to have centre the origin, radius 1 and to lie in the xy-plane, it
can be parametrized by γ(u) = (cos u, sin u, 0), defined for 0 < u < 2π and
−π < u < π, say. This gives the atlas for the unit cylinder.
We now calculate the fundamental forms for the generalized cylinder

σ(u, v) = γ(u) + va.

We will assume that γ is unit-speed, a is a unit vector, and γ is contained

in a plane perpendicular to a. Then, denoting d/du by a dot,

σ u = γ̇, σ v = a,

so E = kσ u k2 = kġk2 = 1, F = σ u · σ v = ġ · a = 0, G = kσ v k2 = kak2 = 1,
and the first fundamental form of σ is du2 + dv 2 . Note that this is the same
as the first fundamental form of the plane.
Further, N = γ̇ × a and σ uu = γ̈, σ uv = 0 and σ vv = 0. Hence, L =
σ uu · N = γ̈ · (γ̇ × a), M = 0, N = 0 and the second fundamental form is
σ uu · N = γ̈ · (γ̇ × a)du2 .
In particular, if the surface is the unit cylinder, γ(u) = (cos u, sin u, 0) and
a = (0, 0, 1). Then the first fundamental form of σ is du2 + dv 2 , where
60 Dr. V. V. Acharya

N = (− sin u, cos u, 0) × (0, 0, 1) = (cos u, sin u, 0) and L = σ uu · N =

(− cos u, − sin u, 0)·(cos u, sin u, 0) and the second fundamental form is −du2 .
We note that there are two different second fundamental forms for the
unit cylinder. When we considered the unit cylinder as surface of revolution
and as a special case of a generalized cylinder. (If you compute N then you
can see that the normals are in opposite direction.)

Example 5.5 The second special case we shall consider is that in which
the rulings all pass through a certain fixed point, say v; then S is called a
generalized cone with vertex v.
We can take δ(u) = γ(u) − v, giving σ(u, v) = (1 + v)γ(u) − vv. Now,
σ(u, v) = σ(u0 , v 0 ) ⇐⇒ (1 + v)γ(u) − (1 + v 0 )γ 0 (u0 ) + (v 0 − v)v = 0; since
(1 + v) − (1 + v 0 ) + (v 0 − v) = 0, the equation on the right-hand side means
that the points v, σ(u) and σ(u0 ) are collinear. So, for σ to be a surface
patch, no straight line passing through v should pass through more than one
point of γ (in particular, γ should not pass through v). Finally, we have
σ u = (1 + v)γ̇, σ v = γ − v, so σ is regular provided v 6= −1, i.e., the vertex
of the cone is omitted (cf. Example 4.1.5), and none of the straight lines
forming the cone is tangent to γ.
The parametrization is simplest when γ lies in a plane. If this plane
contains v, the cone is simply part of that plane. Otherwise, we can take
v to be the origin and the plane to be z = 1. Then, σ(u) = (f (u), g(u), 1)
for some smooth functions f and g, and the parametrization takes the form
σ(u, v) = v(f (u), g(u), 1), after making the reparametrization v 7→ v − 1.
Before computing its first fundamental form, we make some simplifica-
tions to σ.
First, translating the surface by v (which does not change its first funda-
mental form), we get the surface patch σ 1 = σ − v = (1 + v)(γ − v), so if
we replace γ − v by γ 1 we get σ 1 = (1 + v)γ 1 . This means that we might as
well assume that v = 0 to begin with.
Next, we have already seen that for σ to be a regular surface patch, γ
must not pass through the origin, so we can define a new curve γ̃ by γ̃(u) =
γ(u)/ kγ(u)k . Setting ũ = u, ṽ = (1+v)/ kγ(u)k , we get a reparametrization
σ̃(ũ, ṽ) = ṽ γ̃(ũ) of σ with kγ̃k = 1. We can therefore assume to begin with
that σ(u, v) = vγ(u) with kγ(u)k = 1 for all values of u (geometrically,
this means that we can replace γ by the intersection of the cone with S 2 ).
Finally, reparametrizing again, we can assume that γ is unit-speed, for we
have already seen that for σ to be regular, γ must be regular.
Differential Geometry 61

With these assumptions, and with a dot denoting d/du, we have σ u =

v γ̇, σ v = γ, so E = kv γ̇k2 = v 2 kγ̇k2 = v 2 , F = v γ̇ · γ = 0 (since kγk = 1),
G = kγk2 = 1, and the first fundamental form is v 2 du2 + dv 2 .
Note that, as for the generalized cylinder, the first fundamental form of
the generalized cone does not depend on the curve γ.

5.1 Why first fundamental form?

If γ is a curve lying in the image of a surface patch σ, we have γ(t) =
σ(u(t), v(t)) for some smooth functions u(t) and v(t). Then, denoting d/dt
by a dot, we have γ̇ = u̇σ u + v̇σ v by the chain rule, so

hγ̇, γ̇i = E u̇2 + 2F u̇v̇ + Gv̇ 2 ,

and the length of γ is given by

Z
(E u̇2 + 2F u̇v̇ + Gv̇ 2 )1/2 dt.

Definition 5.1 If S1 and S2 are surfaces, a diffeomorphism f : S1 → S2 is

called an isometry if it takes any curve in S1 to a curve of the same length
in S2 . If an isometry f : S1 → S2 exists, we say that S1 and S2 are isometric.

Proposition 5.1 A diffeomorphism f : S1 → S2 is an isometry if and only

if, for any surface patch σ 1 of S1 , the patches σ 1 and f ◦ σ 1 of S1 and S2 ,
respectively, have the same first fundamental form.
Proof. Since the length of any curve can be computed as the sum of the
lengths of curves each lying in a single surface patch, we can assume that
S1 and S2 are covered by single surface patches. Moreover, since f is a
diffeomorphism, Proposition 4.3 shows that we can assume that these patches
are of the form σ 1 : U → R3 (for S1 ) and f ◦ σ 1 = σ 2 (for S2 ). We have
to show that f is an isometry if and only if σ 1 and σ 2 have the same first
fundamental form.
Suppose first that σ 1 and σ 2 have the same first fundamental form. If
t 7→ (u(t), v(t)) is any curve in U, and γ 1 (t) = σ 1 (u(t), v(t)) and γ 2 (t) =
σ 2 (u(t), v(t)) are the corresponding curves in S1 and S2 , then f takes γ 1 to
γ 2 . since

f (γ 1 (t)) = f (σ 1 (u(t), v(t))) = σ 2 (u(t), v(t)) = γ 2 (t).

62 Dr. V. V. Acharya

It is clear that γ 1 and γ 2 have the same length, since both lengths are found
by integrating the expression (E u̇2 +2F u̇v̇ +Gv̇ 2 )1/2 , where Edu2 +2F dudv +
Gdv 2 is the (common) first fundamental form of σ 1 and σ 2 .
Conversely, suppose that f is an isometry. If t 7→ (u(t), v(t)) is any
curve in U, defined for t ∈ (α, β), say, the curves γ 1 (t) = σ 1 (u(t), v(t)) and
γ 2 (t) = σ 2 (u(t), v(t)) have the same length. Hence,
Zt1 Zt1
(E1 u̇2 + 2F1 u̇v̇ + G1 v̇ 2 )1/2 dt = (E2 u̇2 + 2F2 u̇v̇ + G2 v̇ 2 )1/2 dt,
t0 t0

for all t0 , t1 ∈ (α, β), where E1 , F1 and G1 are the coefficients of the first
fundamental form of σ 1 , and E2 , F2 and G2 those of σ 2 . This implies that
the two integrands are the same, and hence that
E1 u̇2 + 2F1 u̇v̇ + G1 v̇ 2 = E2 u̇2 + 2F2 u̇v̇ + G2 v̇ 2 (5.5)
Fix t0 ∈ (α, β), and let u0 = u(t0 ), v0 = v(t0 ). We now apply Eq. (5.5)
for the following three choices of the curvet 7→ (u(t), v(t)) in U : (i) u =
u0 + t − t0 , v = v0 : this gives E1 = E2 ; (ii) u = u0 , v = v0 + t − t0 : this gives
G1 = G2 ; (iii) u = u0 + t − t0 , v = v0 + t − t0 : this gives
E1 + 2F1 + G1 = E2 + 2F2 + G2 ,
and hence (in view of (i) and (ii)) F1 = F2 .

5.2 Conformal mapping

Suppose that two curves γ and γ̃ on a surface S intersect at a point p. The
angle θ of intersection of γ and γ̃ at p is defined to be the angle between
the tangent vectors γ̇ and γ̃˙ (evaluated at t = t0 and t = t̃0 , respectively).
Using the dot product formula for the angle between vectors, we see that θ
is given by
˜
γ̇ · γ̇
cos θ = (5.6)
kγ̇k γ̇ ˜
Suppose that γ and γ̃ lie in a surface patch σ of S, so that γ(t) =
σ(u(t), v(t)) and γ̃(t) = σ(ũ(t), ṽ(t)) for some smooth functions u, v, ũ and
ṽ. If Edu2 + 2F dudv + Gdv 2 is the first fundamental form of σ, then
γ̇ = σ u u̇ + σ v v̇, γ̃˙ = σ u ũ˙ + σ v ṽ˙
Differential Geometry 63

and by (5.6) we have

E u̇ũ˙ + F (u̇ṽ˙ + v̇ ũ)

˙ + Gv̇ ṽ˙
cos θ = . (5.7)
(E u̇2 + 2F u̇v̇ + Gv̇ 2 )1/2 (E ũ˙ 2 + 2F ũ˙ ṽ˙ + Gṽ˙ 2 )1/2

Example 5.6 The parameter curves on a surface patch σ(u, v) can be

parametrized by γ(t) = σ(u0 , t), γ̃(t) = σ(t, v0 ), respectively, where u0 is
the constant value of u and v0 is the constant value of v in the two cases.
Thus,
u(t) = u0 , v(t) = t, ũ(t) = t, ṽ(t) = v0 ,
Hence, u̇ = 0, v̇ = 1, ũ˙ = 1, ṽ˙ = 0.
These parameter curves intersect at the point σ(u0 , v0 ) of the surface. By
Eq. (5.7), their angle of intersection θ is given by

F
cos θ = √ ,
EG

where E, F and G are evaluated at (u0 , v0 ). In particular, the parameter

curves are orthogonal if and only if F = 0.
If we consider the sphere given by σ(θ, ϕ) = (cos θ cos ϕ, cos θ sin ϕ, sin θ)
then the first fundamental form is dθ2 + cos2 θdϕ2 . Thus, F = 0. The circles
on the sphere corresponding to a constant value of θ are called parallels; those
corresponding to a constant value of ϕ are called meridians.(Refer Example
3). This tells us that parallels and meridians are orthogonal to each other.
Note that a similar comment can be made for the unit cylinder and the
double cone (without vertex) x2 + y 2 = z 2 as F = 0 in the first fundamental
form.

Definition 5.2 If S1 and S2 are surfaces, a diffeomorphism f : S1 → S2 is

said to be conformal if, whenever f takes two intersecting curves γ 1 and γ˜1
in S1 to curves γ 2 and γ˜2 on S2 , the angle of intersection of γ 1 and γ˜1 is
equal to the angle of intersection of γ 2 and γ˜2 .

Thus, in short, f is conformal if and only if it preserves angles.

Proposition 5.2 A diffeomorphism f : S1 → S2 is conformal if and only if,

for any surface patch σ 1 of S1 , the first fundamental forms of the patches σ 1
and f ◦ σ 1 of S1 and S2 , respectively, are proportional.
64 Dr. V. V. Acharya

Proof. We can assume that S1 and S2 are covered by single surface patches.
Moreover, since f is a diffeomorphism, Proposition 4.3 shows that we can
assume that these patches are of the form σ 1 : U → R3 (for S1 ) and f ◦ σ 1 =
σ 2 (for S2 ). We have to show that f is a conformal map if and only if the
first fundamental form σ 1 and σ 2 are proportional.
Suppose that their first fundamental forms E1 du2 + 2F1 dudv + G1 dv 2 and
E2 du2 + 2F2 dudv + G2 dv 2 are proportional, say

E2 du2 + 2F2 dudv + G2 dv 2 = λ(E1 du2 + 2F1 dudv + G1 dv 2 )

for some smooth function λ(u, v), where (u, v) are coordinates on U. Note
that λ > 0 everywhere, since (for example) E1 and E2 are both > 0.
Suppose that γ and γ̃ lie in a surface patch σ of S1 , so that γ(t) =
σ 1 (u(t), v(t)) and γ̃(t) = σ 1 (ũ(t), ṽ(t)) for some smooth functions u, v, ũ and
ṽ, then f takes γ and γ̃ to the curves σ 2 (u(t), v(t)) and σ 2 (ũ(t), ṽ(t)) in S2
resp. Using (5.7), the angle θ) of intersection of the latter curves on S2 is
given by

E2 u̇ũ˙ + F2 (u̇ṽ˙ + v̇ ũ)

˙ + G2 v̇ ṽ˙
cos θ =
(E2 u̇2 + 2F2 u̇v̇ + G2 v̇ 2 )1/2 (E2 ũ˙ 2 + 2F2 ũ˙ ṽ˙ + G2 ṽ˙ 2 )1/2
λE1 u̇ũ˙ + λF1 (u̇ṽ˙ + v̇ ũ)
˙ + λG1 v̇ ṽ˙
=
(λE1 u̇2 + 2λF1 u̇v̇ + λG1 v̇ 2 )1/2 (λE1 ũ˙ 2 + 2λF1 ũ˙ ṽ˙ + λG1 ṽ˙ 2 )1/2
E1 u̇ũ˙ + F1 (u̇ṽ˙ + v̇ ũ)
˙ + G1 v̇ ṽ˙
=
(E1 u̇2 + 2F1 u̇v̇ + G1 v̇ 2 )1/2 (E1 ũ˙ 2 + 2F1 ũ˙ ṽ˙ + G1 ṽ˙ 2 )1/2

since the λ’s cancel. But, using (5.7) again, we see that the right-hand side
is the cosine of the angle of intersection of the curves γ and on γ̃. Hence, f
is conformal.
For the converse, we must show that if

E1 u̇ũ˙ + F1 (u̇ṽ˙ + v̇ ũ)

˙ + G1 v̇ ṽ˙
(E1 u̇2 + 2F1 u̇v̇ + G1 v̇ 2 )1/2 (E1 ũ˙ 2 + 2F1 ũ˙ ṽ˙ + G1 ṽ˙ 2 )1/2
E2 u̇ũ˙ + F2 (u̇ṽ˙ + v̇ ũ)
˙ + G2 v̇ ṽ˙
= . (5.8)
(E2 u̇2 + 2F2 u̇v̇ + G2 v̇ 2 )1/2 (E2 ũ˙ 2 + 2F2 ũ˙ ṽ˙ + G2 ṽ˙ 2 )1/2

for all pairs of intersecting curves γ(t) = σ 1 (u(t), v(t)) and γ̃(t) = σ 1 (ũ(t), ṽ(t))
in S1 , then the first fundamental forms of σ 1 and σ 2 are proportional. Fix
Differential Geometry 65

(a, b) ∈ U and consider the curves

γ(t) = σ 1 (a + t, b), γ̃(t) = σ 1 (a + t cos φ, b + t sin φ),

where φ is a constant, for which u̇ = 1, v̇ = 0, ũ˙ = cos φ and ṽ˙ = sin φ.

Substituting in (5.8) gives

E1 cos φ + F1 sin φ
(E1 )1/2 (E1 φ + 2F1 cos φ sin φ + G1 sin2 φ)1/2
cos2
E2 cos φ + F2 sin φ
= . (5.9)
(E2 )1/2 (E2 cos2 φ + 2F2 cos φ sin φ + G2 sin2 φ)1/2

Squaring both sides of (5.9) and writing

(E1 cos φ + F1 sin φ)2 = E1 E1 cos2 φ + 2F1 cos φ sin φ + G1 sin2 φ

− E1 G1 − F12 sin2 φ


(5.10)

we get, after cacellation of terms,

E1 G1 − F12 (E2 ) E2 cos2 φ + 2F2 cos φ sin φ + G2 sin2 φ

= E2 G2 − F22 (E1 )(E1 cos2 φ + 2F1 cos φ sin φ + G1 sin2 φ). (5.11)

Setting λ = (E2 G2 − F22 )E1 /(E1 G1 − F12 )E2 , we get

E2 cos2 φ+2F2 cos φ sin φ+G2 sin2 φ = λ(E1 cos2 φ+2F1 cos φ sin φ+G1 sin2 φ).

which can be rewritten as

(E2 − λE1 ) cos2 φ + 2 (F2 − λF1 ) cos φ sin φ + (G2 − λG1 ) sin2 φ = 0. (5.12)
π π π
Taking φ = we get G2 = λG1 and then φ = and φ = gives E2 = λE1 ,
2 4 6
and F2 = λF1 .

Example 5.7 We consider the unit sphere S 2 . If Q is any point of S 2 other

than the north pole N = (0, 0, 1), the straight line joining N and Q intersects
the xy-plane at some point P, say. The map that takes Q to P is called
stereographic projection from S 2 to the plane, and we denote it by Π. We
are going to show that Π is conformal.
66 Dr. V. V. Acharya

Let P = (u, v, 0), Q = (x, y, z). Since P, Q, N lie on a straight line, there
is a scalar ρ such that
Q − N = ρ(P − N ),
and hence

(x, y, z) = (0, 0, 1) + ρ((u, v, 0) − (0, 0, 1)) = (ρu, ρv, 1 − ρ). (5.13)

Hence, ρ = 1 − z, u = x/(1 − z), v = y/(1 − z) and we have

 
x y
Π(x, y, z) = , ,0 .
1−z 1−z

On the other hand, from Eq. 5.13 and x2 + y 2 + z 2 = 1 we get ρ = 2/(u2 +

v 2 + 1) and hence

u2 + v 2 − 1
 
2u 2v
Q= , , .
u2 + v 2 + 1 u2 + v 2 + 1 u2 + v 2 + 1

If we denote the right-hand side by σ 1 (u, v), then σ 1 is a parametrization of

S 2 with the north pole removed. Parametrizing the xy-plane by σ 2 (u, v) =
(u, v, 0), we then have

Π(σ 1 (u, v)) = σ 2 (u, v).

To show that Π is conformal we have to show that the first fundamental

forms of σ 1 and σ 2 are proportional. The first fundamental form of σ 2 is
du2 + dv 2 . As to σ 1 , we get
 2
2(v − u2 + 1)

4uv 4u
(σ 1 )u = ,− 2 ,
(u2 + v 2 + 1)2 (u + v 2 + 1)2 (u2 + v 2 + 1)2
2(u2 − v 2 + 1)
 
4uv 4v
(σ 1 )v = − 2 , ,
(u + v 2 + 1)2 (u2 + v 2 + 1)2 (u2 + v 2 + 1)2
This gives
4(v 2 − u2 + 1)2 + 16u2 v 2 + 16u2 4
E1 = (σ 1 )u · (σ 1 )u = = .
(u2 + v 2 + 1)4 (u2 + v 2 + 1)2

Similarly, F1 = 0, G1 = 4/(u2 + v 2 + 1)2 . Thus, the first fundamental form of

1
σ 2 is λ times that of σ 1 , where λ = (u2 + v 2 + 1)2 .
4
Differential Geometry 67

It is often useful to think of Π as a map to the complex numbers C rather

than to the xy-plane, by identifying u + iv ∈ C with (u, v, 0). Moreover, we
can parametrize S 2 itself in a partly complex way by identifying (x, y, z) ∈ S 2
with (x + iy, z). Then, S 2 becomes the set of pairs (w, z) where w ∈ C, z ∈
R and |w|2 + z 2 = 1. Stereographic projection then takes the simple form
w
Π(w, z) = , and the surface patch σ 1 is given by
1−z

|w|2 − 1
 
2w
σ 1 (w) = , .
|w|2 + 1 |w|2 + 1

The inconvenience of having to exclude the north pole from the domain of
definition of Π can be overcome by introducing a ‘point at infinity’ ∞ and
defining the ‘extended complex plane’ C∞ = C ∪ {∞}. If we agree that Π
maps the north pole to ∞, it defines a bijection Π : S 2 → C∞ .

5.3 Surface Area

Suppose that σ : U → R3 is a surface patch on a surface S. The image
of σ is covered by the two families of parameter curves obtained by setting
u = constant and v = constant, respectively. Fix (u0 , v0 ) ∈ U ; since the
change in σ(u, v) corresponding to a small change ∆u in u is approximately
σ u ∆u and that corresponding to a small change ∆v in v is approximately
σ v ∆v, the part of the surface contained by the parameter curves on the
surface corresponding to u = u0 , u = u0 + ∆u, v = v0 and v = v0 + ∆v is
approximately a parallelogram in the plane with sides given by the vectors
σ u ∆u and σ v ∆v (the derivatives being evaluated at (u0 , v0 )):
Recalling that the area of a parallelogram in the plane with sides a and
b is ka × bk , we see that the area of the parallelogram on the surface is
approximately
σ u ∆u × σ v ∆v = kσ u × σ v k ∆u∆v.
This suggests the following definition.

Definition 5.3 The area Aσ (R) of the part σ(R) of a surface patch σ :
U → R3 corresponding to a region R ⊆ U is
ZZ
Aσ (R) = kσ u × σ v k dudv.
R
68 Dr. V. V. Acharya

Of course, this integral may be infinite - think of the area of a whole

plane, for example. However, the integral will be finite if, say, R is contained
in a rectangle that is entirely contained, along with its boundary, in U. The
quantity kσ u × σ v k that appears in the definition of area is easily computed
in terms of the first fundamental form Edu2 + 2F dudv + Gdv 2 of σ:
1
Proposition 5.3 kσ u × σ v k = (EG − F 2 ) 2 .
Proof. We use a result from vector algebra: if a, b, c and d are vectors in
R3 , then
(a × b) · (c × d) = (a · c)(b · d) − (a · d)(b · c).
Applying this to kσ u × σ v k2 = (σ u × σ v ) · (σ u × σ v ), we get

kσ u × σ v k2 = (σ u · σ u )(σ v · σ v ) − (σ u · σ v )2 = EG − F 2 .

Note that, for a regular surface, EG − F 2 > 0 everywhere, since for a

regular surface σ u × σ v is never zero. Thus, our definition of area is
ZZ
AS (R) = (EG − F 2 )1/2 dudv. (5.14)
R

Note that this is a double integration.

We sometimes denote (EG − F 2 )1/2 dudv by dAσ .
But we have still to check that this definition is sensible, i.e., it is un-
changed if σ is reparametrized. This is certainly not obvious, since E, F and
G change under reparametrization.

Proposition 5.4 The area of a surface patch is unchanged by reparametriza-

tion.
Proof Let σ : U → R3 be a surface patch and let σ̃ : Ũ → R3 be a
reparametrization of σ, with reparametrization map Φ : Ũ → U. Thus, if
Φ(ũ, ṽ) = (u, v), we have σ̃(ũ, ṽ) = σ(u, v). Let R̃ ⊆ Ũ be a region, and let
R = Φ(R̃) ⊆ U. We have to prove that
ZZ ZZ
kσ u × σ v k dudv = kσ̃ ũ × σ̃ ṽ k dũdṽ.
R R̃

We showed in the proof of Proposition 4.2) that

   
∂u ∂v ∂u ∂v ∂(u, v)
(σ̃ ũ × σ̃ ṽ ) = − (σ u × σ v ) = det (σ u × σ v ) .
∂ ũ ∂ṽ ∂ṽ ∂ ũ ∂(ũ, ṽ)
Differential Geometry 69

Hence,
ZZ ZZ  
∂(u, v)
kσ̃ ũ × σ̃ ṽ k dũdṽ = det kσ ũ × σ ṽ k dũdṽ
∂(ũ, ṽ)
R̃ R̃

By the change of variables formula for double integrals, the right-hand side
of this equation is exactly
ZZ
kσ u × σ v k dudv.
R

Example 5.8 Determine the area of the part of the paraboloid z = x2 + y 2

with z ≤ 1 and compare with the area of the hemisphere x2 + y 2 + z 2 = 1,
z ≤ 0.
Solution. Parametrize the paraboloid by σ(u, v) = (u, v, u2 + v 2 ). Note
that σ u = (1, 0, 2u), σ v = (0, 1, 2v) and hence its first fundamental form is
(1 + 4u2 )du2 + 8uvdudv + (1 + 4v 2 )dv 2 . Hence, the required area is
Z p
1 + 4(u2 + v 2 )dudv,

taken over the disc u2 + v 2 < 1. To calculate the area, we use change of
variables. Let u = r sin θ, v = r cos θ then the area is
Z 1√ Z 5√
2π π
2π 1 + 4r2 rdr = tdt = (53/2 − 1).
0 8 0 6
This is less than the area 2π of the hemisphere.3

Example 5.9 Determine the surface area of the portion of the plane y+2z =
2 within the cylinder x2 + y 2 = 3.
v
Solution. Note that σ(u, v) = (u, v, 1 − ). Hence, σ u = (1, 0, 0) , σ v =
  √ 2 √
1 5 3 5
0, 1, − . Hence, kσ u × σ v k = . Thus, A = π.
2 2 2
xy
Example 5.10 Determine the area of the surface z = intercepted by the
3
cylinder x2 + y 2 ≤ 16.
3
Put 1 + 4r2 = t. Hence, 8rdr = dt.
70 Dr. V. V. Acharya

uv  v
Solution. Note that σ(u, v) = (u, v, ). Hence, σ u = 1, 0, , σv =
3 3
2
 u  v 2 uv u2 2
0, 1, . Thus, the first fundamental form is du + 2 dudv + dv .
3 9 9 9
1 RR √
Hence, the required area is u + v + 9dudv, taken over the disc u2 +
2 2
3
v 2 < 16. 4 To calculate the area, we use change of variables. Let u =
r sin θ, v = r cos θ then the area is

π
Z 4 √ π 25 √
Z
π 2  25 2π 196π
2
9 + r 2rdr = tdt = × t3/2 9 = [125 − 27] = .
3 0 3 9 3 3 9 9

Now that we have a good definition of area, we can ask which maps between
surfaces are area-preserving.

Definition 5.4 Let S1 and S2 be two surfaces. A diffeomorphism f : S1 →

S2 is said to be equiareal if it takes any region in S1 to a region of the same
area in S2 (we assume that each of the regions is sufficiently small, so that
it is contained in the image of some surface patch).

Theorem 5.1 A diffeomorphism f : S1 → S2 is equiareal if and only if, for

any surface patch σ(u, v) on S1 , the first fundamental forms

E1 du2 + 2F1 dudv + G1 dv 2 and E2 du2 + 2F2 dudv + G2 dv 2

of the patches σ on S1 and f ◦ σ on S2 satisfy

E1 G1 − F12 = E2 G2 − F22 . (5.15)

Proof. Let σ : U → R3 ; f is equiareal if and only if

ZZ ZZ
2
(E1 G1 − F1 )dudv = (E2 G2 − F22 )dudv
R R

for all regions R ⊆ U. This holds if and only if the two integrands are equal
everywhere, i.e., if and only if E1 G1 − F12 = E2 G2 − F22 .
 v u 
4
Alternatively, we can find σ u × σ v = − , − , 1 . Hence, kσ u × σ v k =
3 3
1√ 2
u + v 2 + 9.
3
Differential Geometry 71

Example 5.11 We use Archimedes’s theorem to compute the area of a

‘Lune,’ i.e. the area enclosed between two great circles.
We can assume that the great circles intersect at the poles, since this can
be achieved by applying a rotation of S 2 , and this does not change areas.
If θ is the angle between them, the image of the lune under the map f is a
curved rectangle on the cylinder of width θ and height 2. If we now apply
the isometry which unwraps the cylinder on the plane, this curved rectangle
on the cylinder will map to a genuine rectangle on the plane, with width θ
and height 2. By Archimedes’ theorem, the lune has the same area as the
curved rectangle on the cylinder, and since every isometry is an equiareal
map, this has the same area as the genuine rectangle in the plane, namely
2θ. Note that this correctly gives the area of the whole sphere to be 4π.

5.4 Normal and geodesic curvatures

It is obvious that the shape of a surface influences the curvature of curves
on the surface. For example, a curve on a plane or a cylinder can have zero
curvature everywhere, but this is not possible for curves on a sphere since
no segment of a straight line can lie on a sphere. Thus, another natural
way to investigate how much a surface curves is to look at the curvature of
curves on the surface. We shall see that this leads, once again, to the second
fundamental form of the surface.
If γ is a unit-speed curve on an oriented surface S, then γ̇ is a unit vector
and is, by definition, a tangent vector to S. Hence, γ̇ is perpendicular to the
unit normal N of S, so γ̇ , N and N × γ̇ are mutually perpendicular unit
vectors.
Again since γ is unit-speed, γ̈ is perpendicular to γ̇, and hence is a linear
combination of N and N × ġ :
γ̈ = κn N + κg N × γ̇ (5.16)
Definition 5.5 The scalars κn and κg in Eq. (5.16) are called the normal
curvature and the geodesic curvature of γ, respectively.
Proposition 5.5 With the above notation, we have
κn = γ̈ · N, κg = γ̈ · (N × γ̇), (5.17)
κ2 = κ2n + κ2g , (5.18)
κn = κ cos ψ, κg = ±κ sin ψ (5.19)
72 Dr. V. V. Acharya

where κ is the curvature of γ and ψ is the angle between N and the principal
normal n of γ.
Proof. Equations (5.17) and (5.18) follow from (5.16) and the fact that N
and N× γ̇ are perpendicular unit vectors. The first equation in (5.19) follows
from γ̈ = κn, and the second then follows from Eq. (5.18)

Note that κn and κg both change sign when N is replaced by −N, so on a

general (not necessarily orientable) surface only the magnitudes of κn and κg
are well defined.
If γ is regular, but not necessarily unit-speed, we define the geodesic and
normal curvatures of γ to be those of a unit-speed reparametrization of γ.
When a unit-speed parameter t is changed to another such parameter ±t + c,
where c is a constant, it is clear that κn 7→ κn and κg 7→ ±κg , so κn is well
defined for any regular curve, while κg is well defined up to sign. Equations
(5.18) and (5.19) continue to hold if γ is any regular curve.
Suppose where γ is a normal section of the surface, i. e. γ is the intersec-
tion of the surface with a plane Π that is perpendicular to the tangent plane
of the surface at every point of γ. We want to find the normal and geodesic
curvature of γ.
Since γ lies in Π, the principal normal N is parallel to Π, and since Π is
perpendicular to the tangent plane, N is also parallel to Π. Since n and N
are both perpendicular to γ̇, and since γ̇, is parallel to Π, n and N must be
parallel to each other, i.e. ψ = 0 or π. Hence, κn = κ and κg = 0.
Proposition 5.6 Let σ(u, v) be a surface patch with standard unit normal
N(u, v). Then, Nu · σ u = −L, Nu · σ v = Nv · σ u = −M, Nv · σ v = −N, where
L, M and N are as defined in Eq. (5.1).
Proof. Since σ u and σ v are tangent vectors to the surface patch,

N · σ u = N · σ v = 0.

Differentiating these equations with respect to u and v gives

Nu · σ u = −N · σ uu = −L, Nv · σ u = −N · σ uv = −M,
Nu · σ v = −N · σ uv = −M, Nv · σ v = −N · σ vv = −N.

Proposition 5.7 If γ(t) = σ(u(t), v(t)) is a unit-speed curve on a surface

patch σ, its normal curvature is given by

κn = Lu̇2 + 2M u̇v̇ + N v̇ 2 ,
Differential Geometry 73

where Ldu2 + 2M dudv + N dv 2 is the second fundamental form of σ.

Proof. Let n denote the standard unit normal of σ. Note that
d d
κn = n · γ̈ = n · (γ̇) = n · (σ u u̇ + σ v v̇)
dt dt
= n · (σ u ü + σ v v̈ + (σ uu u̇ + σ uv v̇)u̇ + (σ uv u̇ + σ vv v̇)v̇)
= Lu̇2 + 2M u̇v̇ + N v̇ 2

This result means that two curves which touch each other at a point p of
a surface (i.e., which intersect at p and have parallel tangent vectors at p)
have the same normal curvature at p.

Theorem 5.2 (Meusnier’s Theorem) Let p be a point of a surface S and

let v be a unit tangent vector to S at p. Let Πθ be the plane containing the
line through p parallel to v and making an angle θ with the tangent plane
to S at p and assume that Πθ is not parallel to the tangent plane to S at p
Suppose that Πθ intersects S in a curve with curvature κg . Then, κθ sin θ is
independent of θ.
Proof. Assume that γ θ is a unit-speed parametrization of the curve of
intersection of Πθ and S. Then, at p, γ˙θ = ±v, so γ̈ θ is perpendicular to v
and is parallel to Πθ . Thus, in the notation of Proposition (5.5), ψ = π/2 − θ
and so Eq. (5.19) gives
κθ sin θ = κn .
But κn depends only on p and v, and not on θ.

An important special case is that in which γ is a normal section of the

surface, i.e., γ is the intersection of the surface with a plane Π that is per-
pendicular to the tangent plane of the surface at every point of γ

Proposition 5.8 The curvature κ, normal curvature κn and geodesic cur-

vature κg of a normal section of a surface are related by

κn = ±κ, κg = 0.

Proof. As in the proof of Proposition 10, κn = κ sin θ, where θ = ±π/2 for

a normal section. This gives the first equation; the second follows from it
and Eq.(5.19).
74 Dr. V. V. Acharya

To analyze κn further, we use the matrix notation. Let Edu2 + 2F dudv +

Gdv 2 and Ldu2 + 2M dudv + N dv 2 are the first and the second fundamental
forms of a surface σ. We introduce the following 2 × 2 matrices:
   
E F L M
FI = and FII =
F G M N

Let t1 = ξ1 σ u + η1 σ v and t2 = ξ2 σ u + η2 σ v be two tangent vectors at some

point of σ. Then

t1 · t2 = (ξ1 σ u + η1 σ v ) · (ξ2 σ u + η2 σ v )
= Eξ1 ξ2 + F (ξ1 η2 + ξ2 η1 ) + Gη1 η2
  
 E f ξ2
= T1t FI T2 ,

= ξ1 η1
F G η2
     
ξ1 ξ2 u̇
where, T1 = and T2 = . Similarly, if γ̇ = σ u u̇ + σ v v̇ and T =
η1 η2 v̇
t
then using proposition 5.7, we see that κn = T FII T as
  
 L M u̇
= Lu̇2 + 2M u̇v̇ + N v̇ 2 .

u̇ v̇
M N v̇

Definition 5.6 The principal curvatures of a surface patch are the roots of
the equation
det(FII − κFI ) = 0, (5.20)
i.e
L − κE M − κF
= 0. (5.21)
M − κF N − κG

Since, (5.21) is a quadratic equation for κ, there are two roots. It turns out
that both the roots are real! If κ is one of the principal curvatures, equation
(5.20) says that FII − κFI is not invertible, so, assuming that κ is real, there
is a non-zero 2 × 1 column matrix T having real number entries such that

(FII − κFI )T = 0. (5.22)

 
ξ
Definition 5.7 If T = satisfies equation (5.22), the corresponding tan-
η
gent vector
Differential Geometry 75

Example 5.12 Consider the unit sphere S 2 and use the latitude longitude
parametrization as usual.We found that E = 1, F = 0, G = cos2 θ and that
L = 1, M = 0, N = cos2 θ. So the principal curvatures are the roots of

1−κ 0
= 0,
0 cos2 θ − κ cos2 θ

i.e., κ = 1 (repeated root). Every tangent vector is a principal vector.

Example 5.13 Consider the unit cylinder parametrized in the usual way:

σ(u, v) = (cos v, sin v, u).

We found that E = 1, F = 0, G = 1 and L = 0, M = 0, N = 1. So the

principal curvatures are the roots of

0−κ 0
= 0,
0 1−κ

i.e., κ = 0 or 1. Any principal vector t1 corresponding to (κ = 1) satisfies

  
−1 0 ξ1
= 0,
0 0 η1
so ξ1 = 0 and t1 is a multiple of σ v = (− sin v, cos v, 0). Similarly, one finds
that any principal vector corresponding to (κ = 0) is a multiple of σ u =
(0, 0, 1).

Proposition 5.9 Let κ1 and κ2 be the principal curvatures at a point P of

a surface patch σ. Then,
(i) κ1 and κ2 are real numbers;

(ii) if κ1 = κ2 = κ, say, then FII = κFI and (hence) every tangent vector
to σ at P is a principal vector;

(iii) if κ1 6= κ2 then any two (non-zero) principal vectors t1 and t2 corre-

sponding to κ1 and κ2 respectively, are perpendicular.
In case (ii), P is called an umbilic.
Proof. For (i), let t1 and t2 be any two perpendicular unit tangent vectors
to the surface at P (not yet known to be principal vectors).
76 Dr. V. V. Acharya

Example 5.14 Compute the Geodesic curvature of any circle on a sphere

(not necessarily a great circle).
Solution. If the sphere has radius R, the parallel with latitude θ has radius
r = R cos θ; if p is a point of this circle, its principal normal at p is parallel
to the line through p perpendicular to the z-axis, while the unit normal to
the sphere is parallel to the line through p and the centre of the sphere.
The angle ψ in Eq. κg = ±κ sin ψ is therefore equal to θ or π − θ so κg =
± 1r sin θ = ± R1 tan θ. Note that this is zero if and only if the parallel is a
great circle.
Example 5.15 Consider the surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u))
where u 7→ (f (u), 0, g(u)) is a unit-speed curve in R3 . Compute the geodesic
curvature of
(i) a meridian v = constant;
(ii) a parallel u = constant.
Solution. Note that σu = (f˙(u) cos v, f˙(u) sin v, ġ(u)), where a dot denotes
d/du. and σv = (−f (u) sin v, f (u) cos v, 0). Thus,
σu × σv
N= = (−ġ cos v, −ġ sin v, f˙).
kσu × σv k
On a meridian v = constant, we can use u as the parameter; since σu is a
unit vector, u is a unit-speed parameter on the meridian and
f¨(u) cos v f¨(u) sin v g̈(u)
κg = σuu · (N × σu ) = −ġ cos v −ġ sin v f˙ = 0,
f˙(u) cos v f˙(u) sin v ġ(u)
as first two columns are identical after taking common cos v and sin v.
On a parallel u = constant, we can use v as a parameter, but σv is not
ds
a unit vector. The arc-length s is given by = kσv k = f (u), so s = f (u)v
dv
(we can take the arbitrary constant to be zero). Then,
−f (u) cos v −f (u) sin v 0
1 1 1 f˙(u)
κg = 2
σvv · (N × σv ) = −ġ cos v −ġ sin v f˙ = .
f (u) f (u) f (u)3 f (u)
−f (u) sin v f (u) cos v 0
Differential Geometry 77

5.5 The Gaussian and Mean Curvatures

We start by defining two new measures of the curvature of a surface.

Definition 5.8 Let κ1 and κ2 be the principal curvatures of a surface patch.

Then, the gaussian curvature of the surface patch is

K = κ1 κ2 (5.23)

and its mean curvature is

1
H= (κ1 + κ2 ) . (5.24)
2
Note that the gaussian curvature stays the same when a surface patch is
reparametrised, while the mean curvature either stays the same or changes
sign. It follows that the gaussian curvature is well defined for any surface S.
It is easy to get explicit formulas for H and K :

Proposition 5.10 Let σ(u, v) be a surface patch with first and second fun-
damental forms

Edu2 + 2F dudv + Gdv 2 and Ldu2 + 2M dudv + N dv 2 ,

respectively. Then,
LN − M 2
(i) K = ,
EG − F 2
LG − 2M F + N E
(ii) M = ,
2 (EG − F 2 )
√
(iii) the principal curvatures are H ± H 2 − K.

Proof. Note that by definition, the principal curvatures are the roots of the
equation
L − κE M − κF
= 0.
M − κF N − κG
Hence, (L − κE) (N − κG) − (M − κF )2 = 0, i.e.

2
EG − F 2 κ2 − (LG − 2M F + N E) + LN − M 2 = 0.


78 Dr. V. V. Acharya

Using the relation between roots and coefficients of the quadratic equation,
we get (i) and (ii). By the definition of H and K, κ1 and κ2 are the roots of

κ2 − 2Hκ + K = 0,
√
Hence, κ = H ± H 2 − K.

Example 5.16 Find the Gaussian and mean curvature of the unit sphere
and right circular cylider of radius 1.