Geometry of curves and surfaces

B. H. Bowditch
University of Warwick
[October 2012]

0. Introduction.

These are informal notes intended to accompany the course MA3D9. These focus on
the theory. We leave pictures and many of the examples to lectures and support classes.
These notes might not coincide exactly with what is done in lectures. Where there is
any difference, it is the material presented in lectures that will determine the examinable
content of the course.

Sources :
I have made much use of the lecture notes by Nigel Hitchin [H], combined with material
from McClearys book [M], as well as some examples from do Carmos Book [C]. I would
probably also have made use of [T] if Id been aware of its existence earlier.

Lecture notes :

[H] Nigel Hitchin, The geometry of surfaces, University of Oxford, available at:
http://people.maths.ox.ac.uk/hitchin/hitchinnotes/hitchinnotes.html.
[T] George Terizakis, MA3D9 Geometry of curves and surfaces, University of Warwick,
available via MathStuff.

Books :

[M] John McCleary, Geometry from a differentiable viewpont, CUP 1994.

[S] Dirk J. Struik, Lectures on classical differential geometry : Addison-Wesley 1950
[C] Manfredo P. do Carmo, Differential geometry of curves and surfaces : Prentice-Hall
1976
[O] Barrett ONeill, Elementary differential geometry : Academic Press 1966
[MR] Sebastian Montiel, Antonio Ros, Curves and surfaces : American Mathematical
Society 1998
[G] Alfred Gray, Modern differential geometry of curves and surfaces : CRC Press 1993

Web page :
A web page for this course will be maintained at:
http://www.warwick.ac.uk/masgak/cas/course.html

1
 curves and surfaces

Contents :
0. Introduction ... 1
1. Curves in Rn ... 4
2. Planar curves ... 6
3. Curves in R3 ... 12
4. Parameterised surfaces in R3 ... 17
5. Embedded surfaces in R3 ... 28
6. The second fundamental form ... 37
7. Covariant derivatives and parallel transport ... 44
8. Gauss curvature ... 52
9. The local Gauss-Bonnet Theorem ... 55
10. Some global properties of surfaces ... 60
11. 2-manifolds ... 64
The end ... 68

History :
Most of the theory described here was developed in the first half of the 19th century,
though of course, we adopt more modern terminology and notation.
The notions of curvature and torsion of space curves were developed in independent
work of Frenet and Serret around 1850. The differential geometry of surfaces was explored
by Euler in the 18th century and later by Monge. Much of what we describe here originates
in work of Gau around 1825.
Around 1854, Riemann developed a much more general theory of manifolds. This
allows one to describe surfaces abstractly, without describing any embedding in euclidean
space. If there is time, we will say something about this at the end of the course.

Background and notation.

(a) Linear algebra.
Rn will denote n-dimensional euclidean space. An element (x1 , x2 , . . . , xn ) Rn can
be thought of either as locating a point is space, or as a vector indicating direction. In its
latter role, it will frequently be denoted as a column vector:

x1

.
x = .. .
xn

Formally however, there is no distinction between these a point is the same as its
position vector from the origin.
Given x, y Rn we write x.y, ||x|| = x.x and xy respectively for the dot product,
norm, and wedge (or cross) product. The angle between x and y is cos1 (x.y/||x|| ||y||).
The distance between x and y is ||x y||.

2
 curves and surfaces

A linear map : Rn Rm has the form [x 7 Ax] for an m n matrix A. If

m = n, or A is orthogonal if AT A = AAT = I (where AT denotes transpose).

Exercise : is orthogonal (x, y)((x).(y) = x.y) (x)(||(x)|| = ||x||).

A dilation is a linear map of the the form x 7 x, for some > 0, i.e. given by the
matrix I.

If is invertible, is orientation preserving (respectively reversing) if det A > 0

(respectively det A < 0).

Exercise : If is orthogonal, then (x y) = (x) (y), where the sign depends on

whether it is orientation preserving or reversing.

Definition : An affine map from Rn to Rm is a map of the form x 7 Ax + b where A

is an m n matrix, and b Rm .

Definition : A rigid motion is an affine map where A is orthogonal.

(A rigid motion is commonly called an isometry, though we shall reserve that term here
for isometric maps between surfaces, as described in Section 4.)

Exercises :
(1) A rigid motion preserves distances.
(2) (Harder) A map from Rn to Rn that preserves distances is a rigid motion. (The hard
part is showing that such a map must be affine.)
We make some use of bilinear and quadratic forms. The relevant background will be
discussed in Section 6.

(b) Topology.
We will use some basic notions from topology, namely open and closed sets, continuity
etc. In practice these will only be applied to subsets of Rn .

(c) Differentiation.

Definition : We say that a map f : Rn Rm is smooth if it has derivatives of all

orders.

In practice, we will only be taking derivatives up to third order here.

If p Rn , the derivative of f at p is a linear map f,p : Rn Rm . We will
abbreviate this to f when the point p is implicitly understood. It is represented by the
m n jacobian matrix (fi /xj )ij . We frequently abbreviate fi /xj to (fi )xj . Note
that f is injective if and only if it has rank n (which implies n m).

3
 curves and surfaces

Theorem 0.1 : If f has rank n, then f is locally injective at p. (That is, there is an
open set, U Rn with p U such that f |U is injective.)

Theorem 0.2 : (Inverse function theorem) If f : Rn Rn has rank n (i.e. is

invertible) then there are open subsets U, V Rn with p U such that f |U is bijective,
and (f |U )1 : V U is also smooth.

We refer to f |U as a diffeomorphism from U to V .

A proof of Theorem 0.2 can be found in [MA225 ???]. Theorem 0.1 can be deduced
from Theorem 0.2 (exercise).
We will also be using a result on the existence and uniqueness of solutions of systems
of first order linear differential equations (Theorem 3.4)).

1. Curves in Rn .

Let I R be an interval, i.e. I has the form [a, b], (a, b], [a, b) or (a, b) with a
[, ), b (, ].
Let : I Rn be smooth. We write

(t) =
t

Definition : (t) is the tangent to at t.

Definition : || (t)|| is the speed .

Example : The curve : R2 given by (t) = (t2 , t3 ) has (t) = (2t, 3t2 ). In
particular, (0) = (0, 0).

Definition : is regular at t if || (t)|| =

6 0.
is regular if it is regular at t for all t I.

Examples :
(1) Straight line : t 7 at + b.
(2) Circle : t 7 (a cos t, a sin t)
(3) Helix : t 7 (a cos t, a sin t, bt)
Straight line, circle, helix.

Note that the curve [t 7 (t2 , t3 ) described above not regular.

Exercise : In fact there is no regular curve with the same image.

4
 curves and surfaces

Definition : The length of is defined as

Z
length() = || (t)||dt.
I

(This is finite if I = [a, b] and with a, b R.)

Exercise : d((a), (b)) length ).

Reparameterisation.

Definition : Suppose I, J are intervals. A diffeomorphism from I to J is a map : I J

which is bijective and such that and 1 are smooth.

Note : : For 1 to be smooth, its enough that be smooth and (t) 6= 0 for all t.

Definition : If : I Rn and : J Rn are curves, then is a reparameterisation

of of there is a diffeomorphism : J I such that = .

(Note that this is reflexive symmetric and transitive.)

The chain rule tells us that

(u) = ((u)) (u).

Lemma 1.1 : Length is invariant under reparameterisation.

Proof : Let = . Then

Z
length() = || (u)|| du
ZJ
= || ((u))|| | (u)| dt
ZI
= || (t)|| dt
I
= length().

Suppose : I Rn is smooth where I = [a, b]. Given t [a, b], let

Z t
s(t) = || (u)|| du = length(|[a, t]).
a

5
 curves and surfaces

Then s (t) = || (t)||. We have s : I J, where J = [0, length()]. If is regular, then

s is a diffeomorphism. Let : J I be the inverse function, and let = . Thus,
: J Rn is a reparameterisation of . Moreover, || (t)|| = || ((t))|| | (t)| = 1, i.e.
has unit speed.

Definition : We refer to s as arc length and to as the arc-length reparameterisation of

.

Example : Circle

Piecewise smooth paths.

Suppose a = t0 < t1 < < tn = b. A curve : [a, b] Rn is piecewise smooth if is

continuous and |[ti , ti+1 ] is smooth for all i. Let
X n
length() = length(|[ti1 , ti ]).
i=1

Definition : is polygonal if each segment |[ti , ti+1 ] is a straight line. We refer to the
points (ti ) as its vertices.

Clearly is determined (up to reparameterisation) by its vertices.

Exercise : If is polygonal, then

n
X
length() = ||(ti ) (ti1 )||.
i=1

Fact : If is a smooth curve, given > 0, there is some > 0 such that if a = t0 < t1 <
< tn = b and |ti+1 ti | for all i, then if is the polygonal curve with vertices
(t0 ), (t1 ), . . . , (tn ), then length() length() . (Note that by the from the earlier
exercise, we always have length length .)

2. Planar curves.

Let : I R2 be a curve parameterised by arc-length. Given s I, let T(s) = (s) be

the unit tangent.
Let N(s) be the unit normal obtained by rotating T(s) through /2 anticlockwise.
Now T.T = 1 and so T.T = 0. Thus we can write
T (s) = (s)N(s),
where (s) R. Here is a smooth function of s. (Question: why is it smooth?)

6
 curves and surfaces

Definition : is the (signed) curvature of .

(We refer to || as the unsigned curvature.)

Note that
|(s)| = ||T (s)|| = || (s)||.

Example : Circle of radius r. This has constant curvature 1/r.

Definition : If 6= 0, then (s) = 1/(s) is the (signed) radius of curvature.

The point c(s) = (s) + (s)N(s) is the centre of curvature.
The circle of centre c(s) and radius (s) is the osculating circle.

Note that for a circle, the osculating circle is constant and equal to the original curve.

General formula.
Suppose (t) = (x(t), y(t)) is regular (not necessarily parameterised by arc-length).

Proposition 2.1 : The unsigned curvature of is given by:

 
 x y y x 
|(t)| =  2 .
((x ) + (y )2 )3/2 

Proof : Exercise.

Example : The ellipse: (t) = (a cos t, b sin t):

ab
(t) = p .
(a + (b a2 ) cos2 t)3 )
2 2

What are the maximum and minimum curvatures?

Remarks :
(1) Unsigned curvature is invariant under (postcompositon with) rigid motion of R2 . The
sign remains the same if and only if the rigid motion is orientation preserving. What are
the maximum and minimum curvatures?
(2) Under dilation by a factor of > 0, it gets multiplied by 1/.

7
 curves and surfaces

The tractrix.
The tractrix is defined as a curve : (0, 1] R2 with (0) = (1, 0), such that the tangent
at any point (t) intercepts the y-axis at a point a unit distance away from (t). (As usual
we write (t) = (x(t), y(t)).) In other words:

dy 1 x2
= .
dx x

Using the formula for curvature we get:

x
|| = .
1 x2

Let s 7 (s) s 0 be the arc length paramerisation, with (0) = (1, 0). Since (x )2 +
 2
dy dy
(y )2 = 1, we get x = x (for example, note that dx = y /x and express 1 + dx in
terms of x and in terms of x ). Solving this differential equation with x(0) = 0, we get
x = es .

Also
dy 1 x2 dx p p
= = 1 x2 = 1 e2s .
ds x ds
Integrating: p
y = cosh1 (es ) 1 e2s
so p
(s) = (es , cosh1 (es ) 1 e2s )
p
(s) = es / 1 e2s .

Another parametrisation, used in Section 8, is derived by taking x = sin(). Integrating

for y (as in lectures) we get:

() = (sin(), log(tan(/2)) cos()).

Note that () = tan().

Other examples : Ellipse, cycloid etc.

Closed curves.

Definition : A curve : R R2 is periodic with period t0 > 0 if (t + t0 ) = (t) for

all t R.

8
 curves and surfaces

Example : The circle (t) = (r cos t, r sin t) has period 2.

Definition : A fundamental domain is any interval of the form [a, b] R, where b = a+t0 .

Thus (a) = (b), (a) = (b), (a) = (b) etc.

Remark : We can more intuitively think of as a map of the circle into R2 gluing
the endpoints a and b together to form circle. In these terms, we can refer to as a closed
curve. To be more specific we can identify the circle with the unit circle in the plane, with
the parameter t corresponding to the point (cos(2t/t0 ), sin(2t/t0 )) in the circle.

R Rb
If f (t) is any continuous periodic function in t, we can define f (t)dt = a f (t)dt,
where [a, b] is any fundamental domain. It is independent of the choice of fundamental
domain. Again, we can think of it more informally as an integral over the circle.

Total turning
Suppose that is a regular closed curve. We write T(t) for the unit tangent. Thus
T(t) = (cos (t), sin (t)) for some (t) R well defined up to some integer multiple of 2.
R
Definition : The total turning of is equal to
(t)dt.

Since (a) = (b) up to some multiple of 2, we have that the total turning of is equal
to 2n for some n Z. Here n is called the turning number of .

Examples : n = 1, 1, 0, 2 etc.

Remark : We can define total turning for any regular closed curve: set (t) = r(t)(cos (t), sin (t)).
Note that this is invariant under reparameterisation.

Jordan curves.
We state some facts which are intuitively clear, though they are more difficult to prove
formally. They more properly belong in a course on topology.

Definition : A (smooth) Jordan curve is a closed curve that is injective (as a map from
the circle into R2 ).

Examples : Circle, ellipse.

The following, though intuitively reasonable, will not be proven in this course.

9
 curves and surfaces

Theorem 2.2 : (Smooth Schoenflies) If is a smooth Jordan curve, then there is a

diffeomorphism f : R2 R2 sending the unit circle to .

Here a diffeomorphism f is a bijective map f : R2 R2 such that f and f 1 are both

smooth.
If we think of as a map of the unit circle into R2 then this means that f is simply
an extention of : S 1 R2 .
If we think of as a periodic map of period t0 , then we get

(t) = f ((cos(2t/t0 ), sin(2t/t0 )).

Let D = {x R2 | ||x|| 1} be the unit disc. Its boundary is the unit circle, and so
mapping under f we see that the image of is the boundary of , where = f (D).

Definition : A (smooth) Jordan domain is the image of a smooth disc under some
diffeomorphism of R2 .

Thus, the image any smooth Jordan curve, , is the boundary of a smooth Jordan
domain, . We can think of as the inside of . We shall sometimes write informally:
= . (In fact, is uniquely determined by .)

Note that may be positively or negatively oriented depending on whether the unit normal
(as defined above) points in or out of . In the former case, it has total turing 2 and in
the latter case, it has total turning 2 (though we wont prove that here).

We can generalise this to a piecewise smooth closed curve. This has a sequence of exterior
angles, 1 , . . . , n , in [, ] atR the vertices
P where f is not differentiable. In this case,
the total turning is defined as dt + i i .

For example, for a polygonal

P curve, we have = 0 on all the straight segments, and so
the total turning is just i i . Thus, for a positively oriented polygonal
P Jordan curve (i.e.
a polygon in the traditional sense) we get the familiar formula: i i = 2.

Area.
(We will assume the notion of integration of smooth functions over a nice subset, of
R2 . Thus, for example, the area of can be defined by intergating the constant function
1 over . The formal definitions of these concepts requires some measure theory, beyond
the scope of this course. See [Measure theory course ???].)

Suppose that is a Jordan domain with boundary the smooth Jordan curve, . We write
(t) = (x(t), y(t)). Let A be the area of .

10
 curves and surfaces

Lemma 2.3 :
1
Z Z Z

A = y(t)x (t)dt = x(t)y (t)dt = xy yx dt.
2

Proof : See lectures.

Example : Circle.

Remark : We have the following:

Theorem : Let be a Jordan domain, with boundary = . Let l = length() and

A = area(). Then A l2 /4, with equality if and only if is a circle.

A relatively simple proof can be found for example in [C, p33].

Greens Theorem.

Suppose that is a Jordan domain, and = is positively oriented. Suppose

that P, Q : R2 R are smooth functions. (We only really need them defined in a
neighbournood of .) Then:

Theorem 2.4 : (Green):

Q P
Z Z

P x + Qy dt = dA.
x y

Here (t) = (x(t), y(t)) R2 , and dA = dx dy is the area element.

Proof : See vector analysis course ???.

Remark : This is sometimes expressed as follows. We can view w = (P, Q) as a vector

field on R2 , that is w(x, y) = (P (x, y), Q(x, y)) is the vector based at (x, y). Embedding
R2 R3 by (x, y) 7 (x, y, 0), its curl is:
 
Q P
curl w = 0, 0, .
x y
 
Thus, curl w.dA = curl w.k dA = Q x P
y dt. Also (t) = (x (t), y (t)) so P x +Qy =
w((t)). (t). Thus, Greens Theorem reduces to:
Z Z

w((t)). (t)dt = curl w.dA

for any planar vector field w. (This is Stokess Theorem in the plane.)

11
 curves and surfaces

Area again.
Q P
Set P = y and Q = x. Then x y = 2. So
Z
LHS = yx + xy dt

Z
RHS = 2 dA = 2A.

Thus
1
Z
A= xy yx dt
2
as before.

3. Curves in R3 .

Frenet frames.
Let : I R3 be a smooth curve parameterised by arc length. Let T(s) = (s)
be the unit tangent vector. As in Section 2, T.T = 0, so we can write T = N, where N
is a unit normal to and (s) = (s) = |T (s)| = | (s)| 0. Here 0 there is no
preferred sign. We refer to as the curvature of .
If 6= 0, then N is well defined. Let
B = T N.
Thus, T, N, B is an orthonormal frame (based at (s)).

Definition : We refer to T, N and B as the unit tangent, normal and binormal respec-
tively, to .
We refer to (T, N, B) as the Frenet frame.

(If there may be any confusion, we refer to N as the principal normal to though
this terminology is not standard.)
Note that, by definition of B, the Frenet frame is positively oriented (so that there
is an orientation preserving orthogonal linear map taking T, N, B to the standard basis,
(1, 0, 0), (0, 1, 0), (0, 0, 1) of R3 ).
Note that
N .N = B .B = 0.
Also:
B = (T N) = T N + T N = T N .
Thus B .T = 0. Since also B .B = 0, we have
B = N
for some R.

12
 curves and surfaces

Definition : (s) = (s) is the torsion of at (s).

This is defined when (s) 6= 0.

Remark : In some texts, is used instead of .

The Frenet matrix.

Sometimes the following notation is convenient.
Let T = (T1 , T2 , T3 ), N = (N1 , N2 , N3 ), B = (B1 , B2 , B3 ), and let P be the 3 3 matrix:

T1 T2 T3
P = N1 N2 N3 .
B1 B2 B3

Since T, N, B is orthonormal, P P T = I and so P T P = I. Note that P = P (s) depends on

the parameter s. (Formally we can think of P as a curve I R9 , where the co-ordinates
are the entries of the matrix.)
Now T , N and B can each be expressed as a linear combination of T, N and B.
In other words, we can find a matrix A(s) such that dPds (s) = A(s)P (s), which we simply
write as
P = AP.
Now P P T = I, so
0 = (P P T )
= P P T + P (P T )
= (AP )P T + P (P )T
= AP P T + P (AP )T
= A + AT .
(We have used the general matrix formula: (XY ) = X Y + XY .) Thus, AT = A, i.e.
A is antisymmetric.
By the definition of and , we know that A has the form:

0 0
.
0 0

Since A + AT = 0, we must have:

0 0
A =
0 .
0 0

From the middle row, we deduce the Frenet-Serret (or Serret-Frenet) formula:

13
 curves and surfaces

Theorem 3.1 :
dN
= T + B.
ds

We also note that and are invariant under rigid motion. More precisely:

Proposition 3.2 : If : R3 R3 is a rigid motion, and : I R3 is a curve, then

for all s, (s) = (s) and (s) = (s), where = , and the sign depends on
whether is orientation preserving or reversing.

Proof : Exercise.

Examples :
(1) Straight line: 0. The normal, binormal and torsion are undefined.
(2) Planar curve: That is there is a fixed vector, a, with (t).a constant. Then T.a = 0.
We get T = N with N parallel to the plane and so B = T N is constant and ortogonal
to the plane. We get 0.
(3) Helix: Let
(s) = (a cos s, a sin s, bs).

With a > 0 and b constant. We verify that || (s)|| = a2 + b2 , so we set = 1/ a2 + b2
to get arc length.
Exercise:
= a 2 = b 2 .

In particular, the curvature and torsion are constant.

Exercise: We can realise any constant positive curvature and any torsion with some helix.

Proposition 3.3 : Let be a curve in R3 .

(1) If 0, then is a straight line.
(2) If (s) 6= 0 for all s, and 0, then is planar.

Proof : (1) T = is constant.

d
(2) B = N, so B is constant. Thus, ds (.B) = .B = T.B = 0, so .B is constant.

We shall see later that if 6= 0 and are constant, then is a helix (up to rigid
motion).

14
 curves and surfaces

Linear systems of differential equations.

Before proceeding, we quote the following result regarding solutions of linear differen-
tial equations.
2 2
In what follows, W : I Rn is formally a smooth curve in Rn , though we think
of it as a smooth family of matrices. That is, for any t, we think of W (t) as the n n
2
matrix whose entries, (wij (t))ij , are the coordinates in Rn .

Theorem 3.4 : Suppose I R is an interval, and t0 I is any element of I. Suppose

2
that W : I Rn is a smooth family of n n matrices. Suppose p Rn is fixed. Then
there is unique smooth path, x : I Rn , such that x(t0 ) = p, and such that for all t we
have
x (t) = W (t)x(t).

Here we are writing x(t) as a collumn vector:

x1 (t)
..
. .
xn (t)
In other words we have a system of n linear first order differential equations, and the
theorem tells us that it has a unique solution.
For a proof, see ???.
We shall apply this to matrices in the following form:

Corollary 3.5 : Suppose that t 7 A(t) is a smooth family of n n matrices, for t I.

Suppose that t0 I and that P0 is a fixed n n matrix. Then there is a unique family of
n n matrices, t 7 P (t), with P (t0 ) = P0 and with

P (t) = A(t)P (t)

for all t.

Proof : This follows by simply observing that P (t) = A(t)P (t) is a system of n2 linear
differential equations in the matrix entries. More formally, to relate this to the above,
write P (t) as a column vector p(t) with entries

p11 (t), . . . , p1n (t), p21 (t), . . . , pnn (t)

where pij (t) are the entries in P (t). Let W (t) be the n2 n2 matrix, with sequence of n
copies of A(t) down the diagonal, and 0 everywhere else. The equation P (t) = A(t)P (t)
is then the same as p (t) = W (t)p(t).

Another variation on this will be used in the proof of Theorem 3.6 below.

15
 curves and surfaces

Existence and uniqueness of curves with prescribed curvature and torsion.

Up to rigid motion in R3 there always exists a unique curve with specified (positive)
curvature and torsion. More precisely:

Theorem 3.6 : Let I R be any interval, and let : I (0, ) and : I R

be smooth functions. Then there is a smooth curve : I R3 parameterised by arc
length, such that for all s I, (s) = (s) and (s) = (s). Moreover, if : I R3 is
another such curve, then there is an (orientation preserving) rigid motion, : R3 R3
such that = .

Proof : Let s0 I be any point. Recall that curvature and torsion are invariant under
orientation preserving rigid motion. Moreover, up to rigid motion, we can assume that
(s0 ) = (0, 0, 0) and the Frenet frame T(s0 ), N(s0 ), B(s0 ) is just (1, 0, 0), (0, 1, 0), (0, 0, 1).
Thus we claim there is a unique curve is with the specifed curvature and torsion and
with this initial condition.
Given s I we set
0 (s) 0
A(s) = (s) 0 (s) .
0 (s) 0
Thus A : I A(s) is smooth and A(s) + A(s)T = 0. Now by Corollary 3.5, there is a
unique family of matrices, s 7 P (s) such that P (s0 ) = I and such that P (s) = A(s)P (s)
for all s.
We claim that, for all s, P (s) is an orthogonal matrix, i.e. P (s)P (s)T = I. To see
this, set Q(s) = P (s)P (s)T . Thus, Q(s0 ) = I. Moreover,

Q(s) = P (s) P (s)T + P (s)(P (s)T )

= A(s)P (s)P (s)T + P (s)(A(s)P (s))T .
= A(s)Q(s) + Q(s)A(s)T

I.e.
Q(s) = A(s)Q(s) + Q(s)A(s)T ().
Note that Q(s) = I for all s is one solution to () (since A(s) + A(s)T = 0). But () is
a system of n2 first order linear differential equations in the matrix entries. Thus () has
a unique solution (cf. the proof of Corollary 3.5). It follows that we must indeed have
Q(s) = I for all s. That is P (s)P (s)T = I, and so also P (s)T P (s) = I.
Now, let T(s), N(s) and B(s) be respectively the first second and third rows of P (s),
thought of as vectors in R3 . Since P (s) is orthogonal, {T(s), N(s), B(s)} is an orthogonal
frame. Rs
Now, let (s) = s0 T(u)du. Thus : I R3 is a smooth curve, with (s0 ) = 0,
and tangent (s) = T(s). In particular, s is an arc length parameter.
Using P (s) = A(s)P (s), we get T (s) = (s)N(s) and B (s) = (s)N(s). Directly
from their definitions, we see that N(s) and B(s) are the normal and binormal respectively,
and that (s) and (s) are its curvature and torsion respectively.

16
 curves and surfaces

This proves the existence of .

For uniqueness, suppose that is another curve with the same curvature and torsion.
As observed above, we can assume that (s0 ) = 0 and that and have the same Frenet
frame at s0 . Let T (s), N (s), B (s) be the Frenet frame to at s. Let P (s) be the
3 3 matrix formed by arranging T (s), N (s), B (s) in three rows (as in the definition of
Frenet matrix). Thus, P (s) = A(s)P (s) for all s. By the uniqueness of such a solution
(Corollary 3.5), we get P (s) = P (s) for all s. In particular, the first row is T (s) = T(s).
Thus, Z s Z s
(s) = T (u)du = T(u)du = (s).
s0 s0

Thus = as required.

One immediate consequence is:

Proposition 3.7 : Suppose : R R3 is a unit speed curve with constant non-zero

curvature and torsion. Then is a helix (up to rigid motion of R3 ).

Proof : Let 0 > 0 and 0 6= 0 be the curvature and torsion. When discussing helices
(lectures) we saw that we can always find a helix with 0 and = 0 . Thus, by
Theorem 3.6, there is a rigid motion, , of R3 with = .

What if is not of unit speed?

Proposition 3.8 : Suppose that : I R3 is a regular curve. Then the curvature

and torsion are given respectively by

| |
=
| |3

and
( ).
= .
| |2

Proof : Exercise. Or see [M p.87].

4. Parameterised surfaces in R3 .

We now begin the study of 2-dimensional surfaces in 3-space.

Let U R2 be open, and let r : U R3 be smooth, i.e. it has derivatives of all
orders. We write ru = r/u and rv = r/v, where u, v are the coordinates of R2 . Thus,
ru : U R3 and rv : U R3 are smooth.

17
 curves and surfaces

Definition : We say that r : U R3 is a parameterised surface if, for all (u, v) U ,

ru (u, v) and rv (u, v) are linearly independent.

Recall that the linear map [(, ) 7 ru + rv ] : R2 R3 is the derivative of r.

We denote it by r . That is, r (, ) = ru + rv . Thus, the definition says that the
derivative has rank 2 everywhere. With respect to the standard co-ordinates (u, v) on R2
and (x, y, z) on R3 , the derivative has matrix

xu xv
yu yv
zu zv

where xu = x/u etc. This is the jacobian matrix.

Suppose, for the moment, that r is injective. We write S = r(U ) R3 . Given
(u, v) U , write Tp (S) R3 for the vector space spanned by ru (u, v) and rv (u, v), where
p = r(u, v). This is the same as the image of the linear map r (u, v). By assumption it is
2-dimensional.

Definition : Tp (S) is the tangent space to S at p.

The plane p + Tp (S) R3 is the tangent plane to S at p.
The unit vector
ru rv
n = n(u, v) =
||ru rv ||
is the normal to S at p.
Less formally we can think of the tangent plane as touching S at the point p. Note
that the parameterisation determines a direction of n. If we were to replace v by v, for
example, we would replace n by n. We can think of this as changing the orientation on
S.
We can generalise the above to the case where r is not assumed injective. But if there
are two or more points mapped to p, then we need to specify which of the preimages is
being used to define the tangent space.
We remark that, by Theorem 0.1, r will always be locally injective. That is, if (u, v)
U , then there is always an open neighbourhood, V , of (u, v) in U such that r|V is injective.
Thus, for the purpose of studying local properties, there is no loss in assuming injectivity.
We say more about this in Section 5.

Examples :

(1) Plane. U = R2 .
r(u, v) = a + bu + cv
where a, b, c are fixed and b c 6= 0.

n = (b c)/|b c|.

18
 curves and surfaces

(2) Cylinder. U = R2 .
r(u, v) = (a cos v, a sin v, u)
n = (cos v, sin v, 0)

(3) More generally, we have a product surface: Let : I R2 be a curve. Write

(v) = (x(v), y(v)). Set U = I R and set
r(u, v) = (x(v), y(v), u)
.
(4) Cone. U = (0, ) R.
r(u, v) = (au cos v, au sin v, u)

(5) Sphere (minus poles) in spherical polar co-ordinates: U = (/2, /2) R.

r(u, v) = (a cos v cos u, a sin v cos u, a sin u)
Here, n = a1 r.
(It is customary to use the notation = v, = u.)
(6) Helicoid. U = R2 .
r(u, v) = (au cos v, au sin v, v)

(7) Torus. U = R2 , a > b are constants.

r(u, v) = ((a + b cos u) cos v, (a + b cos u) sin v, b sin u).
(Note r is doubly periodic: the plane wraps around the torus in both the u and v
direcstions.)
(8) Surface of revolution. Suppose I R is open and U = I R, and that : I R2
is a planar curve. Let (u) = ((u), (u)) and suppose (u) > 0 for all u. Set
r(u, v) = ((u) cos v, (u) sin v, (u)).
Examples are the cylinder, cone, sphere and torus above.
Another, the pseudosphere obtained by spinning the tractrix will be discussed later in
Section 8.
A third example, the catenoid is obtained from the catenery: (u) = a cosh u, (u) = u.
(9) Graph of a function. Suppose f : U R is smooth. Let r(u, v) = (u, v, f (u, v)).
e.g. paraboloid f (u, v) = au2 + bv 2 .
(10) Developable surface
Let be a unit speed curve in R3 . Let
r(u, v) = (u) + vT(u)
where T(u) = (u). If (u) 6= 0, write N(u), B(u) for the normal and binormal to
respectively. Note that ru = T + vT = T + vN and rv = T. If 6= 0, then
ru rv = vN T = vB, so we get a parameterised surface for v 6= 0.

19
 curves and surfaces

Reparameterisation.

Suppose U, V R2 are open.

Definition : A diffeomorphism between V and U is a map : V U such that both

and 1 are smooth.

Writing (, ) = (u(, ), v(, )) for , V , we have Jacobian

 
u u
v v

invertible for all , .

Suppose r : U R3 is a parameterised surface, and write S = r(U ).

Definition : A reparameterisation of S is a composition s = r : V R3 , where

: V U is a diffeomorphism.

Note that S = s(V ) = r(U ). We have

s = u ru + v rv
.
s = u ru + v rv

Thus s and s are linearly independent. (So s : V R3 is indeed a parametrised

surface.) If (u, v) = (, ), we see that s (, ) and s (, ) span the same subspace of R3
as ru (u, v) and rv (u, v). Thus, the tangent space, Tp (S) for p = r(u, v) = s(, ) is well
defined, independently of such reparameterisation. (We say more about this in Section 5.)
We say that preserves or reverses orientation, depending on whether
 
u u
det
v v

is positive or negative. The normal, n, to S at p is then fixed or reversed accordingly.

Area.
Suppose that D S = r(U ) and let = r1 D U . (We will assume that r is injective
here.) We define the area of D by
Z
area(D) = ||ru rv || dudv.

For this, we need to assume that is sufficiently nice for this integral to be defined. The
technical term is measurable see the measure theory course. However, it is sufficent
that should be open or closed, or equivalently that D be open or closed in S. This
should be enough for our purposes.

20
 curves and surfaces

Lemma 4.1 : Area is invariant under reparameterisation.

Proof : In other words, suppose that : V U is a diffeomorphism, and let s = r :

V R2 . Let = 1 = s1 D. Then

s s = (u ru + v rv ) (u ru + v rv )
= (u v v u )ru rv ,

so Z Z
||s s || dd = ||ru rv ||(u v v u ) dd

Z ,
= ||ru rv || dudv

since  
u u
u v v u = det .
v v

The first fundamental form.

We introduce the first fundamental form of a parameterised surface. The idea behind it
to describe the intrinsic geometry of the surface. First we recall some linear algebra.

Suppose that W is a vector space over R. A bilinear form on W is a map of the form
x, y 7 xT P y, where x, y are elements of W written as column vectors with respect to some
basis of W , and P is a square matrix. (One should check that this notion is independent
of the basis we have chosen for W .) The corresponding map x 7 xT P x is a quadratic
form on W . Such a form is positive definite if xT P x > 0 for all x 6= 0.

Example : If W = Rn , then the inner product, x.y = xT y = xT Iy is a positive definite

bilinear form. The square of the norm, ||x||2 = x.x is the corresponding quadratic from.
If X W is any subspace, then this induces a form on X, simply by restriction.

Suppose that U R3 is a parameterised surface, and p = r(u, v) r(U ) = S.

Definition : The first fundamental form on the tangent space Tp (S) is the quadratic form
induced by the inner product on R3 .

Clearly this is positive definite. We want to express this in terms of the intrinsic
coordinates u, v of S.
Suppose (1 , 1 ), (2 , 2 ) R2 . Then x = 1 ru + 1 rv and y = 2 ru + 2 rv lie in
Tp (S). We see that:   
E F 2
x.y = ( 1 1 )
F G 2

21
 curves and surfaces

where
E = ru .ru
F = ru .rv
G = rv .rv .
Note that x = r (1 , 1 ) and y = r (2 , 2 ). Thus, under the identification of R2 with
Tp (S) via the derivative map, r , the first fundamental form is the bilinear form with
matrix:  
E F
.
F G
Note that E, G > 0 and that
 
E F
det = EG F 2 = ||ru ||2 ||rv ||2 |ru .rv |2 = ||ru rv ||2 > 0.
F G

(From basic linear algebra, this gives another proof that the first fundamental form is
positive definite.)

The first fundamental form is commonly written in the notation:

Edu2 + 2F dudv + Gdv 2 .

One justification for this is as follows. Suppose : V U is a diffeomorphism, and

s = r . So s = u ru + v rv and s = u ru + v rv . We check that in the new
coordinates, the first fundamental form is given by the matrix
 
E F
F G

where:
E = s .s = Eu2 + 2F u v + Gv2
F = s .s = Eu u + F (u v + u v ) + Gv v .
G = s .s = Eu2 + 2F u v + Gv2
A more convenient way of carrying out the above calculation is to write du = u d+u d
and dv = v d + v d and then expand:

Edu2 + 2F dudv + Gdv 2

= E(u d + v d)2 + 2F (u d + u d)(v d + v d) + G(u d + v d)2

= Ed 2 + 2F dd + Gd 2 .
The above should be viewed as a formal manipulation, with no particular intrinsic meaning
attached to du or dv etc. Luckily for us, it gives the correct answer.

22
 curves and surfaces

Definition : We say that a property of S is intrinsic if it can be expressed in terms of

the first fundamental form.

(That is, the first fundamental form as a function on S we are allowed to take derivatives
of the form with respect to u and v etc.)
For example area. Suppose that D S is nice. Then
Z Z p
area(D) = ||ru rv || dudv = EG F 2 dudv
D D

by the formula for ||ru rv || given earlier.

Another example is length. Suppose that : I U is a smooth curve in U R2 . Then

= r : I S R3 . Writing (t) = (u(t), v(t)) its length is given by
Z Z Z p

|| (t)|| dt =
||ru u + rv v || dt =
E(u )2 + 2F u v + G(v )2 dt.
I I I

(This also gives another justification of the notation Edu2 + 2F dudv + Gdv 2 .)

A third example are angles: Suppose that , : I U R2 are regular curves, and
(t0 ) = (t1 ) = q say. Let = r and = r . The tangent vectors are (t0 ) and
(t1 ), so they meet at q at an angle
 
1 (t0 ). (t1 )
cos
|| (t0 )|| || (t1 )||

which can again be expressed in terms of the first fundamental form of S at p.

Examples :
(1) Cartesian coordinates (x, y) in the plane: r(x, y) = (x, y, 0). Then rx = i, ry = j. So
E = i.i = 1, F = i.j = 0, G = j.j = 1. We get the first fundamental form to be:

dx2 + dy 2 .

(2) Polar coordinates (r, ) in the plane (minus the origin). Let r(r, ) = (r cos , r sin , 0).
Then rr = (cos , sin , 0), r = (r sin , r cos , 0), so E = 1, F = 0, G = r 2 . We get f.f.f.
equal to
dr 2 + r 2 d 2 .

Note that we can view this as a reparameterisation of the plane minus the origin. The
formal manipulation discussed above, with x = r cos and y = r sin gives: dx2 + dy 2 =
(cos dr r sin d)2 + (sin dr + r cos d)2 = dr 2 + r 2 d 2 .
(3) The cylinder r(u, v) = (a cos v, a sin v, u).

23
 curves and surfaces

So ru = (0, 0, 1), rv = (a sin v, a cos v, 0), and E = ru .ru = 1, F = ru .rv = 0, G =

rv .rv = a2 . Thus the f.f.f. is:
du2 + a2 dv 2 .

(4) The cone: The f.f.f. is:

(1 + a2 )du2 + a2 u2 dv 2
.
(5) The sphere of radius a in spherical polars (u, v). We get

a2 du2 + a2 cos2 udv 2 .

(6) Surface of revolution: r(u, v) = ((u) cos v, (u) sin v, (u)).

We get:
( (u)2 + (u)2 )du2 + (u)2 dv 2 .

(7) Helicoid. We get

a2 du2 + (1 + a2 u2 )dv 2 .

(8) Graphs. We get ru = (1, 0, fu), rv = (0, 1, fv ), so the f.f.f. is:

(1 + fu2 )du2 + 2fu fv dudv + (1 + fv2 )du2 .

(8) Developable surfaces. ru = T + vN, rv = T. The f.f.f. is:

(1 + v 2 2 )du2 + 2dudv + dv 2 .

Example : As an example, we can compute the area of a surface of revolution. Suppose

that I and J = (0 , 1 ) R are intervals. Let U = I J, and let r : U R3 be given
by r(u, v) = ((u) cos v, (u) sin v, (u)). From the formula for the first fundamental form
above, we calculate
EG F 2 = (u)2 ( (u)2 + (u)2 )
and so the total area is
Z p
(1 0 ) (u) ( (u))2 + (u)2 du.
I

If 0 = 0 and 1 = 2, we get
Z p
2 (u) ( (u))2 + (u)2 du.
I

24
 curves and surfaces

Isometries.
Suppose that r : U R3 and s : U R3 are two parameterised surfaces with the
same domain, U . Let S = r(U ) and = s(U ). If (u, v) U , we will refer to p r(u, v) S
and q = s(u, v) as corresponding points.
We also have invertible linear maps r : R2 Tp (S) and s : R2 Tq (). Thus
s r1
: Tp (S) Tq () is linear and invertible. We shall refer to the image of a vector
in Tp (S) under this map as the corresponding vector in Tp (). If : I U is a smooth
path, then we get paths = r : I S and = s : I . Again we refer to
these as corresponding paths. For any t, (t) and (t) are corresponding points, and (t)
and (t) are corresponding vectors.

Definition : We say r : U R2 and s : U R2 are isometric parameterised surfaces

if the first fundamental forms on U are equal.

Note that, by the definition of first fundamental form, this is the same as saying that
at any pair of corresponding points, p S and q , the natural correspondence between
Tp (S) and Tq () respects the dot products induced from R3 .

Lemma 4.2 : The parametrised surfaces, S = r(U ) and = s(U ) are isometric if and
only if all pairs of corresponding curves have the same length.

Proof : We write the first fundamental forms of S and respectively as

Edu2 + 2F dudv + Gdu2

Edu2 + 2F dudv + Gdu2 .

If and are corresponding curves, their lengths are respectively:
Z p
E(u )2 + 2F u v + G(v )2 dt
I
Z q
E(u )2 + 2F u v + G(v )2 dt
I

If the first fundamental forms are equal, these lengths are equal.
Conversely, suppose these are equal for all and . In particular, given (u, v) U
and (, ) R2 , can find a curve : [a, a] U with (0) = (u, v) and (0) = (, )
(e.g. a short straight line segment.) Set = r and = s . Now length(|[a, t]) =
length(|[a, t]), and so the integrals over [a, t] are equal. Differentiating at t = 0, we
get: q
p
E(u ) + 2F u v + G(v ) = E(u )2 + 2F u v + G(v )2 .
2 2

with (u , v ) = (, ). Thus

E2 + 2F + G2 = E2 + 2F + G2 .

25
 curves and surfaces

Since this holds for all (, ) R2 we deduce that E = E, F = F and G = G. Moreover

this holds for any pair of corresponding points.

If r and s are injective and isometric, then the map s r1 : S is referred to as

an isometry.
Note that an intrinsic property can equivalently be viewed as one that is invariant
under isometry.

Examples :
(1) Rigid motion:
If r : U R3 is a parameterised surface, and : R3 R3 is a rigid motion, the
r and r are isometric.
(2) Cylinder and plane.
Consider:
(u, v) 7 (a cos v, a sin v, u)

(u, v) 7 (u, av, 0)

We get first fundamental forms du2 + a2 dv 2 in the case of the cylinder, and du2 + (adv)2 =
du2 + a2 dv 2 in the case of the plane. These are equal, so the cylinder is isometric to the
plane (locally).
This accords with the experience that one can bend a flat piece of paper into a cylinder
without tearing it their intrinsic geometries are the same.
(3) Cone and plane. Similar.
(4) Catenoid and helicoid. Consider the catenoid:

r(u, v) = (cosh u cos v, cosh u sin v, u)

and helicoid:
(u, v) 7 (u cos v, u sin v, v).
We begin by reparametrising the helicoid, by setting u = sinh u, and leaving v unchanged,
so that it becomes:
s(u, v) = (sinh u cos v, sinh u sin v, v).
We see that the f.f.f.s of both are given by:

cosh2 u(du2 + dv 2 )

so they are isometric.

Note we are thinking of each as a map from R2 into R3 . The helicoid is injective,
whereas the catenoid is periodic in the v direction, so the maps into R3 are quite different.
In fact, its possible to find a smooth family of isometric surfaces which morph a
catenoid to a helicoid. See example sheet. There are several animations of this on the web.

26
 curves and surfaces

In dealing with isometries, it is often convenient to allow for reparameterisation at the

same time. (In the above example, it would be more sensible to parameterise the plane
with coordinates x = u and y = av.)
Suppose that V, U are open, : V U is a diffeomorphism, and r : U R, and
s : V R are surfaces. Then r and s are isometric if sends the first fundamental
form of s on V to the first fundamental form of r on U .

Examples :
(1) In the above example (2), we could take

(u, v) 7 (a cos v, a sin v, u)

(x, y) 7 (x, y, 0)

and let (u, v) = (u, av), i.e. x = u, y = av. This sends dx2 + dy 2 to du2 + a2 dv 2 , and so
gives us an isometry.
(2) The cone and the plane.
These are respectively:
(u, v) 7 (au cos v, au sin v, u)

(r, ) 7 (r cos , r sin , 0)

(in polar coordinates). The f.f.f.s are:

(1 + a2 )du2 + a2 u2 dv 2

dr 2 + r 2 d 2 .

Define (u, v) = (r, ) where r = u 1 + a2 and = av/ 1 + a2 . This gives an isometry
of a subset of the cone to a subset of the plane.
(Exercise: describe suitable domains U and V so that is a diffeomorphism.)
(3) The catenoid and helicoid.
If we use the original parameterisation of the helicoid as (u, v) 7 (u cos v, u sin v, v), then
we can write (u, v) = (sinh u, v). This describes an isometry from the catenoid to the
helicoid.

Conformal maps.
These are more general than isometries.

27
 curves and surfaces

Definition : Suppose r : U R3 and s : U R3 are parameterised surfaces. Then r

and s are conformally equivalent if there is a smooth function : U (0, ) such that
for all (u, v) U , the first fundamental form of s at (u, v) is (u, v) times that of r at
(u, v).

In other words, it respect the first fundamental form pointwise up to scale. Note that
angles depend on the first fundamental form up to scale. Thus, conformally equivalent
surfaces respect angles between corresponding curves. (In fact, the converse is also true,
so conformal is the same as angle preserving.)
As with isometries, we can allow for reparametisation in dealing with conformal maps.

Examples :
(1) Any postcomposition with a dilation of R3 will be conformal.
(2) The plane and helicold or catenoid. Local scaling is cosh u (see above).
(3) Complex analytic maps are conformal. To intepret this in terms of surfaces in R3 , we
can identify the complex plane C with R2 and embed it in R3 by (x, y) 7 (x, y, 0).
As an explicit example, set V = {(u, v) | u > 0} and U = R2 \ ((, 0] {0}). Let
: V U be the diffeomorphism that sends (u, v) to (x, y) = (u2 v 2 , 2uv). To
compare first fundamental forms, we make the following formal manipulation:

dx = 2udu 2vdv

dy = 2vdu + 2udv.
We get:
dx2 + dy 2 = 4(u2 + v 2 )(du2 + dv 2 )
and so we set (u, v) = 4(u2 + v 2 ). We see that the map is conformal.
In terms of complex analysis, we have x + iy = (u + iv)2 , and so setting w = u + iv,
the map is simply [w 7 w2 ].
(4) We will see other examples in Section 5 (stereographic projection, Mercator projection).

5. Embedded surfaces in R3 .
In this section, we give a slightly different formulation of the notion of a surface, that
does not require a particular parameterisation.
Recall that a parameterised surface was defined in terms of a map r : U R3 .
This has the advantage that it allows the surface to cross itself, or to wrap around
itself. (The latter is particularly useful in dealing with surfaces of revolution, for example.)
It has a number of disadvantages. It refers to a particular parametrisation. Also, many
surfaces are not parametrised surfaces in this sense. With the sphere, for example, we
saw that we had to remove the poles to make it conform naturally to this definition.
For this reason, we make the following definitions:

28
 curves and surfaces

Definition : Suppose S is a subset of R3 . By a chart for S, we mean a parameterised

surface r : U R3 where U R2 is open, such that there exists an open set W R3
with r(U ) = S W , and such that the inverse map r1 : S W U is continuous.
We say that S is a regular surface if for each p S, there is some chart, r : U S with
p r(U ).

We will refer to the image of a chart as a coordinate patch.

Thus, a regular surface is a subset of R3 that can be covered by such coordinate
patches. These patches may overlap (and we will need to consider how the different
coordinates behave on these overlaps). The collection of all coordinate charts is called an
atlas for S.
The technical requirement of a chart that r1 be continuous is required to rule out
certain unwanted phenomena [picture]. One consequence of it will be that if V U is
open, then r|V is also a chart (Lemma 5.1).

Examples :
(1) Let r : U R3 be a parameterised surface. Suppose r is injective and write = r(U ).
Suppose also that the inverse map r1 : U is continuous. Then is a regular surface
we only need one chart, namely r itself.
Note: As a consequence, if r : U S is a chart of a surface S, then then r(U ) is itself a
regular surface.
(2) The unit sphere (using steriographic projection).
Let S2 = {x | ||x|| = 1} R3 . (This is standard notation for the 2-sphere. It does not
mean S S!) Let n = (0, 0, 1) S2 be the north pole and s = (0, 0, 1) S2 be the
south pole
Define r : R2 R3 by the formula:

x2 + y 2 1
 
2x 2y
r(x, y) = , , .
x2 + y 2 + 1 x2 + y 2 + 1 x2 + y 2 + 1

One can interpret this geometrically as follows. One can check that P = {(x, y, 1) | x, y
R} is the tangent plane to the south pole (exercise). identify this with the plane R2 by
(x, y) (x, y, 1). The above map then sends a point p P to the point q S2 , where q
is the intersection of S2 with the straight line in R3 connecting p to n. (Exercise). (The
inverse of this map is called steriographic projection.)
One can now verify (exercise) that r us a chart of S2 , with image S2 \ {n}.
Now swapping the roles of n and s, we get a chart with image S2 \ {s}. These cover
2
S , and so it is a regular surface. We have constructed an atlas with just two charts.

Exercise : Describe a simpler altas for the 2-sphere using 6 charts.

We begin with a couple of technical observations:

29
 curves and surfaces

Lemma 5.1 : If r : U S is a chart, and V U is open, then r|V : V S is a

chart.

Proof : We have r(U ) = S W for some open W R3 . The fact that r1 : S W U

is continuous implies that r(V ) is open in S W . This means that there is an open set
W W with r(V ) = S W . (These just use the definitions of subspace topology and
continuity.) Now W is also open in R3 .
The properties of a chart follow easily.

Lemma 5.2 : Suppose S R3 is a regular surface, and r : U S is a chart (with

r(U ) = W S, with W R3 open). Suppose that (u0 , v0 ) U and let p = r(u0 , v0 ) S.
Then there are open sets U0 U , W0 W , with (u0 , v0 ) U0 , p W0 and a smooth map
f : W0 R2 with f (W0 S) = U0 , and with f r = 1U0 (the identity).

In other words, we can find a smooth left inverse locally that is defined on a neighbourhood
of S in R3 .

Proof : Write r(u, v) = (x(u, v), y(u, v), z(u, v)). The jacobian

xu xv
yu yv
zu zv

has rank 2 at (u0 , v0 ). Thus (swapping the roles of y and z if necessary) we can assume
that (xu , xv ) and (yu , yv ) are linearly independent, i.e.
 
xu xv
det 6= 0
yu yv

Define a map a : U R R3 by

a(u, v, w) = r(u, v) + (0, 0, w).

This has jacobian

xu xv 0
yu yv 0
zu zv 1
which is non-singular at (u0 , v0 , 0). Thus, by the inverse function theorem (Theorem 0.2),
there are neighbourhoods V0 of (u0 , v0 , 0) and W0 of p in R3 , with a(V0 ) = W0 , and
with a smooth inverse map, a1 : W0 V0 . Let f be the composition of a1 with the
projection map (u, v, w) 7 (u, v) : R3 R2 . Since these are both smooth, f is smooth.
Let U0 = {(u, v) R2 | (u, v, 0) V0 }. This is open in R2 (exercise). If (u, v) U0 , then
by construction, a1 r(u, v) = a1 a(u, v, 0) = (u, v, 0), so f r(u, v) = (u, v), i.e. f r
is the identity on U0 as required.

30
 curves and surfaces

Transition maps.
Suppose that r : U S and s : V S are charts. Let U0 = r1 s(V ) and
V0 = s1 r(U ). Thus, U0 and V0 are open subsets of U and V respectively. Moreover, by
Lemma 5.1, r|U0 and s|V0 are charts. These subcharts have the same image in S (namely
r(U ) s(V ), and so = r1 s : V0 U0 is bijective. In fact it is a homeomorphism
(i.e. it is continuous with continuous inverse).

Definition : The map : V0 U0 is the transition map between these two charts.

(Note that we allow for the possibility that U0 = V0 = , though there is not much to be
said in that case.)

Lemma 5.3 : A transition map is a diffeomorphism.

Proof : In other words, we claim that : V0 U0 and 1 : U0 V0 are both

smooth. Swapping the roles of r and s, its enough to prove the former statement.
Suppose that (u1 , v1 ) V0 . Let p = s(u1 , v1 ) and let (u0 , v0 ) = r1 (p) U0 . Now
Lemma 5.2 gives us a smooth inverse f of r defined on a neighbourhood of p in R3 . Thus,
for all (u, v) in a neighbourhood of (u1 , v1 ), we have (u, v) = r1 s(u, v) = f s(u, v). In
particular, = f s is smooth at (u1 , v1 ) V0 . Since (u1 , v1 ) was arbitrary, is smooth
everywhere in V0 .

Thus, we have s = r : V0 R3 where is a diffeomorphism. In other words

s|V0 is a reparameterisation of r|U0 . We have shown that different charts differ by repa-
rameterisation where they overlap. We can therefore bring the results of Section 4 into
play.

Tangent spaces.
Suppose that p S. Let r : U S be a chart with p r(U ). This defines a tangent
space Tp (S). By Lemma 5.3, and the discussion in Section 4, we see that Tp (S) is well
defined, independently of the choice of chart.

Definition : Tp (S) is the tangent space to S at p. We refer to p + Tp (S) as the tangent

plane.

Curves in S.

Definition : A smooth curve in S is a smooth curve, : I R3 with (I) S.

Lemma 5.4 : Suppose that is a smooth curve in S, and r : U S is a chart with

(I) r(U ). Then there is a smooth curve, : I U with = r .

31
 curves and surfaces

Proof : Let = r1 : I U . To see that is smooth, note that it locally has the
form = f , where f is a local inverse to r from an open subset of R3 to R2 (cf. the
proof of Lemma 5.3).

Lemma 5.4 gives us the same notion of a path in a parameterised surface that we used in
Section 4, so we can bring the results there into play.
If the image of does not lie in a single coordinate patch, we need to say a bit more. The
idea to deal with this is to cut S can be cut into pieces, each of which lies in a coordinate
patch, and then apply Lemma 5.4 to each piece. The argument uses the fact (Heine-Borel)
that an interval of the form [a, b] is compact. (Exercise for those who know the relevant
topology.)

Lemma 5.5 : If : I S is smooth, and t I, then (t) Tp (S), where p = (t).

Proof : Exercise: use a chart r : U S with p r(U ), and apply Lemma 5.4.

In fact (exercise) each element of Tp (S) is the derivative (t), for some curve in S
with (t) = p. This allows us to make sense of tangent space without explicit reference
to charts. Thus, Tp (S) is the set of (0) such that is a smooth curve with (0) = p. This
makes it clear that Tp (S) is well defined as a subspace of R3 . However, it is not imediate
from this definition that Tp (S) is a subspace of R3 .

Smooth maps and diffeomorphisms.

Definition : A map f : S Rn is smooth if, for every chart, r : U S, the

composition f r : U Rn is smooth.

Definition : We say that f is smooth at a point p if for any chart, r : U S with

p r(U ), f r is smooth at r1 (p). (Note that by Lemma 5.3, it is enough to show that
there is some such chart with p r(U ) for which this is true.)

Thus f is smooth if and only if it is smooth at every point.

Example : If r : U S is a chart, then r1 : r(U ) R2 is smooth.

Definition : Suppose that S and are surfaces. A diffeomorphism from S to is a

bijective map, f : S such that f and f 1 are both smooth (as maps into R3 ).

2 2 2
Example : Let S = S2 be the unit sphere, and let = {(x, y, z) | xa2 + yb2 + zc2 = 1} be an
ellipsoid (where a, b, c > 0 are constant). Then (x, y, z) 7 (ax, by, cz) is a diffeomorphism
from S to .

32
 curves and surfaces

Lemma 5.6 : Suppose that f : S is a bijective map. Then f is a diffeomorphism

if and only if, for all charts, r : U S and s : V , there is a diffeomorphism
: U0 V0 such that
f r|U0 = s

where U0 = U (f r)1 s(V ) and V0 = V s1 (f r)(U ).

Proof : Exercise, using Lemma 5.3.

This allows us to deal with diffeomorphisms using charts. As an example, we consider

tangent spaces.
Suppose p S and let q = f (p) . There is a bijective linear map from Tp (S)
to Tq () defined as follows. Let r : U S and s : V be charts with p r(U ),
q s(). Let : U0 V0 be defined by Lemma 5.6. Let p = r(u, v), so q = f r(u, v) =
s (u, v). Note that (u, v) U0 . Taking derivatives, we have bijective linear maps

r : R2 Tp (S)

s : R2 Tq ()

: R2 R2 .

Taking a composition s r1
, we get a bijective linear map

Tp (S) Tq ()

which we denote by f . This is the derivative of f at p.

Exercise : Let f : I be a diffeomorphism. If : I S is a smooth curve with

(0) = p, then f : I is a smooth curve, and f ( (0)) = (f ) (0).
Explain how this gives another equivalent definition of the derivative f . What advantages
and disadvantages does it have over that given above?

Isometries.

Definition : An isometry from S to is a diffeomorphism f : S such that for all

p S, the map f : Tp (S) Tf (p) () sends the first fundamental form on Tp (S) to the
first fundamental form on Tf (p) ().

Lemma 5.7 : A diffeomorphism f : S is an isometry if and only if for any curve

: I S we have length(f ) = length().

33
 curves and surfaces

Proof : We saw in Section 4 that the first fundamental form of a surface determines and
is determined by the lengths of curves there.
To be more precise, that result applied to parameterised surfaces, so we need the
additional information that a curve in S can be split into subpaths, such that each path
lies in a coordinate patch in S and its image under f also lies in a coordinate patch in .
This is an exercise in compactness arguments, for those who have the relevant background
in topology.

Note that an isometry preserves all the intrinsic properties of a surface, such as area
and angles.
We also note the following more general definition.

Definition : A diffeomorphism f : S is conformal if there is a smooth function,

: S (0, ) such that for all p S, the map f : Tp (S) Tf (p) () multiplies the
first fundamental form by a factor of (p).

Thus a diffeomorphism is conformal if and only if it preserves angles.

Exercise : A diffeomorphism that is conformal and preserves area is an isometry.

Level sets.
Suppose that f : Rn R is smooth. For each x Rn , we have a derivative map
with jacobian ( fx1 fx2 fx3 ), where fxi = f /xi.

Definition : We say that f is critical at x if fxi = 0 for all i.

Otherwise it is regular

Definition : A level set of f is a set of the form f 1 (t) for some t R.

Definition : A level set, f 1 (t) is regular if f is regular at each point of f 1 (t).

In this case, we refer to t as a regular value.

As an example think of a smooth map f : R2 R. We can imagine f (x) as measuring

the height at the location x. The level sets are then the contours of this function. Note
that typically on a contour map most contours are either circles or lines. One dimension
up, one might expect most contours to be 2-dimensional surfaces. Here is a precise
formulation:

Lemma 5.8 : Suppose that f : R3 R is smooth and that S = f 1 (t) is regular.

Then S is an regular surface in R3 .

34
 curves and surfaces

Proof : Let p S (so f (p) = t). We can suppose (after permuting coordinates) that
fz 6= 0 at p. Define F : R3 R3 by

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

Thus F is smooth, and its jacobian is:

1 0 0
0 1 0
fx fy fz

which is invertable at p.
Thus, by the inverse function theorem, there is an inverse F 1 : V R3 to F defined
in an open neighbourhood, V , of F (p) in R3 . Let U = {(x, y) R2 | (x, y, t) V }. We
see that U is open in R2 . Define r : U R3 by r(x, y) = F 1 (x, y, t). Then r is smooth.
Moreover, F (r(x, y)) = F F 1 (x, y, t) = (x, y, t), so by the definition of F , f (r(x, y)) = t,
and so r(x, y) S. In other words, r(U ) S. In fact, r(U ) = S W , where W = F 1 (V )
is an open subset of R3 . Moreover, r1 : r(U ) R2 is continuous (since F is). It follows
that r : U S is a chart with p r(U ). Since p S was arbitrary, it follows that S is a
surface.

Examples :
(1) f (x, y, z) = x2 + y 2 + z 2 .
fx = 2x, fy = 2y, fz = 2z.

Thus, f 1 (t) is an embedded surface for all t > 0. (A sphere of radius t.)
x2 y2 z2
More generally, f (x, y, z) = a2 + b2 + c2 .
The level sets for t > 0 are ellipsoids.
(2) f (x, y, z) = x2 + y 2 z 2 .
f 1 (0) is a cone which is singular at the origin.
f 1 (t) is non-singular for all t 6= 0. It is a hyperboloid 1-sheeted for t > 0 and 2-sheeted
for t < 0.

Cartography.
We can apply some of these ideas to describing maps of the earths surface. Let the
earth be:
S = S2 = {x R3 | ||x|| = 1}.

35
 curves and surfaces

Definition : A projection of S is the inverse of a chart, i.e.: a map w : V R2 where

V S and r = w1 : U S is a chart with r(U ) = V .

(This is geographers terminology: its not completely standard mathematically.)

We will see (in Section 8) that such a map cannot be an isometry to R2 so it cannot
preserve distances (lengths). We need to do the best we can. We can try to preserve angles
or area, etc. (Note that by the above fact, and earlier exercise we cannot simultanously
preserve angles and area.)

Recall spherical polar coordinates:

(, ) 7 (cos cos , sin cos , sin )

from R (/2, /2) S. If we restrict to U = (, ) (/2, /2), we get a chart

with image S \ L where L is the idealised date line, i.e. the 180 degree longitude. The first
fundamental form is:
cos2 d 2 + d2 .
Its inverse is a projection (see the map of The Living Earth outside room B1.28). Here
are some other examples:

(1) Stereographic projection.

Suppose that (, ) is projected to the point (x, y) in the plane. An argument using similar
2 cos
triangles (see lectures) shows that the distance of (x, y) from the origin is 1sin
, and so
 
2 cos cos 2 sin cos
(x, y) = , .
1 sin 1 sin

from which one calculates

4
dx2 + dy 2 = (cos2 d 2 + d2 ).
(1 sin )2

It follows that stereographic projection is conformal.

(2) [See M, p.126] We consider projections that send latitudes and longitudes to straight
lines. To this end, define a projection by: (x, y) (, v()). Thus, x = 1, x = 0, y = 0,
y = v . We get the euclidean first fundamental form:

dx2 + dy 2 d 2 + (v )2 d2

which we want to compare with the spherical first fundamental form:

cos2 d 2 + d2 .

(2a) Lamberts cylindrical projection.

36
 curves and surfaces

If we want our projection to be area preserving, we need (v )2 = cos2 (so that

dudv dd). So v = cos and we integrate to give v = sin . That is,

(x, y) (, sin ).

Geometrically, this is the same as projecting the sphere to a cylinder radially outward from
the north-south axis, and then unwrapping the cylinder into the plane.
This is sometimes seen as a politically correct projection, since countries far from the
equator dont appear bigger than they deserve to be. But they end up getting distorted
quite dramatically.
It was described in detail by Lambert in 1772, though was probably known in some form to
Archimedes. A variation where the vertical direction is stretched linearly so as to distribute
distortion more evenly was advocated by Gall in the mid nineteenth century, and more
recently publicised by some bloke called Peters.
(2b) Mercators projection.
2 2
If we want our projection to be conformal
 we need (v ) = 1/ cos . So v = sec
and we integrate to give v = log tan 4 + 2 . that is:
  

(x, y) , log tan + .
4 2

Exercise: This sends loxodromes (paths of constant compass bearing) to straight lines.
The Mercator projection gives an accurate picture locally, but it grossly misrepresents area
on a large scale. It makes Greenland look about the same size as Africa.
This projection was introduced by Gheert Cremer (a.k.a. Gerardus Mercator) in 1569. Its
still used today.

(3) Other projections.

There are loads of other projections in use. For example, the sinusoidal projection
is another example of an area-preserving map.
Exercise : Check these out on the web, and verify they have the properties claimed of
them.

6. The second fundamental form.

Unlike the first fundamental form, the second fundamental form is an extrinsic prop-
erty of the surface. It describes how the surface is bent into 3-space. It can be viewed in a
number of different ways. For example, it describes how the first fundamental form varies
if we move the surface in the normal direction. We will introduce it from this viewpoint.
Let r : U R3 be a parameterised surface. We write S = r(U ). Given t R, let

f (u, v, t) = r(t)(u, v) = r(u, v) tn(u, v).

37
 curves and surfaces

(Recall that n(u, v) is the unit normal.) We assume that r(t) : U R3 is itself a
parameterised surface for small t. (There is no essential loss in making this assumption.
Exercise: given any (u, v) U there is a neighbourhood U0 of (u, v) in U and some
t0 > 0 such that r(t)|U0 is a parameterised surface for all t [t0 , t0 ]. Since the second
fundamental form is a local property, so this is good enough.) We will refer to r(t) as the
normal displacement of r. We get
fu = ru tnu
fv = rv tnv .
The first fundamental form of r(t) is
E(t)du2 + 2F (t)dudv + G(t)dv 2 ,
where
E(t) = fu .fu = ru .ru 2tru .nu + t2 nu .nu ,
and so 
1 d 
E(t) = ru .nu .
2 dt t=0
We similarly get: 
1 d  1
F (t) = (ru .nv + rv .nu )
2 dt t=0 2

1 d 
G(t) = rv .nv .
2 dt t=0
Thus, half the first variation of the first fundamental form under normal displacement is
given by:
Ldu2 + 2M dudv + N dv 2 ,
where
L = ru .nu
1
M = (ru .nv + rv .nu ) .
2
N = rv .nv

Definition : The expression

Ldu2 + 2M dudv + N dv 2
is the second fundamental form of the parameterised surface at (u, v).

More formally, it should be viewed as a quadratic form with matrix:

 
L M
M N
with respect to the standard basis of R2 .

38
 curves and surfaces

Remark : : Some books use the letters e, f, g in place of L, M, N .

Unlike the first fundamental form, it need not be positive definite.

Here is another expression. Note that ru .n = 0, so

(ru .n) = ruu .n + ru .nu = 0.
u
Similarly,
rvu .n + rv .nu = 0
ruv .n + ru .nv = 0
rvv .n + rv .nv = 0.
Note that ruv = rvu and so rv .nu = ru .nv . Thus:

L = ru .nu = ruu .n
M = ru .nv = rv .nu = ruv .n .
N = rv .nv = rvv .n

Examples :
(1) The plane. r(u, v) = a+ub+vc. Here ruu = ruv = rvv = 0, so the second fundamental
form is identically 0.
(2) The sphere of radius a:
Here r = an, so ru = anu , rv = anv , and it follows that L = ru .nu = a1 E.
Similarly M = a1 F and N = a1 G. and so

Ldu2 + 2M dudv + N dv 2 = a1 (Edu2 + 2F dudv + Gdv 2 ).

So, for example, in spherical polars (u, v) = (, ), we get the s.f.f. as

a1 (a2 cos2 d 2 + a2 d2 ) = a cos2 d 2 + ad2 .

(3) Graphs:
Let r(u, v) = (u, v, f (u, v)). Then ruu = (0, 0, fuu), ruv = (0, 0, fuv ), rvv = (0, 0, fvv ).
Suppose fu = fv = 0 at (u, v). Then n = (0, 0, 1) and so:
   
L M fuu fuv
=
M N fuv fvv

In other words, at a critical point, the s.f.f. is just the hessian of the function f .
For example f (u, v) = au2 + bv 2 gives 2adu2 + 2bdv 2 at the origin.

39
 curves and surfaces

Exercise : If the second fundamental form is identically zero, then the surface is planar.

The shape operator.

Another interpretation allows us to view the second fundamental form in terms of
variation of normals.
Note that nu and nv are both orthogonal to n, and so lie in the tangent space, that
is the subspace, Tp (S), of R3 spanned by ru and rv . There is therefore a linear map,
: Tp (S) Tp (S), which sends ru to nu and rv to nv . With respect to the basis
ru , rv for Tp (S) we write its matrix as
 
a b
.
c d

In other words (ru ) = aru + crv and (rv ) = bru + drv . This is the same as the matrix
for with respect the standard basis of R2 under 2
 the identification r : R Tp (S).

1 0
(Recall that r sends the standard basis, , of R2 to the basis ru , rv of Tp (S).)
0 1
The map is sometimes called the shape operator .
We claim that    1  
a b E F L M
= .
c d F G M N

(This is a form of the Weingarten equations.)

By definition we have:
nu = aru + crv

nv = bru + drv .

Taking dot products with ru and rv , we get the four equalities:

L = aE + cF
M = bE + dF
M = aF + cG
N = bF + dG

which can be written as:

    
E F a b L M
= ,
F G c d M N

so the above statement follows.

40
 curves and surfaces

Interpretation in terms of linear algebra.

First we recall some abstract linear algebra. The following makes sense in Rn , though
we restrict here to R2 .
Suppose that h., .i is a symmetric bilinear form on R2 given by the symmetric 2 2
matrix A. That is:
hx, yi = xT Ay = yT AT x = yT Ax = hy, xi
for all x, y R2 .

Suppose that h., .i is another symmetric bilinear form given by a matrix B. That is

hx, yi = xT By.

Suppose that A is positive definite. Then there is a (unique) linear map : R2 R2

such that for all x, y

hx, yi = hy, xi = h(x), yi = hx, (y)i.

In fact, we set to be the transformation given by the matrix P = A1 B. Thus

hx, (y)i = xT AP y = xT AA1 By = xT By = hx, yi

h(x), yi = (P x)T Ay = xT BA1 Ay = xT By = hx, yi .

The map is diagonalisable. (This is because it is self-adjoint with respect to h., .i.
If we take a basis for R2 that is orthornormal with respect to h., .i then will be repre-
sented by a symmetric matrix.) Let w1 , w2 be eigenvectors, and 1 , 2 the corresponding
eigenvalues. That is, (w1 ) = 1 w1 and (w2 ) = 2 w2 . We can normalise so that
hw1 , w1 i = hw2 , w2 i = 1. If 1 6= 2 then necessarily hw1 , w2 i = 0. If 1 = 2 , we can
always choose w1 , w2 so that hw1 , w2 i = 0. In other words, we can assume that w1 , w2
are orthonormal with respect to h., .i.

In the case where h., .i and h., .i are respectively the first and second fundamental
forms, then we get:    
E F L M
A= B=
F G M N
and so  
1 a b
P =A B= .
c d

In other words, corresponds under the derivative r : R2 Tp (S) to the map

described above.
In summary, we have a map

: Tp (S) Tp (S).

41
 curves and surfaces

We have shown that, with respect to the basis ru , rv of Tp (S) it has matrix:
 1  
E F L M
.
F G M N
Equivalently, the map induced from R2 to R2 under the derivative identification r :
R2 Tp (S) is represented by the above matrix with respect to the standard basis for
R2 .
As far as we know, for the moment, the map might depend on the parameterisation,
r, of S, since it was defined in these terms. In Section 7, we will see that it is in fact
independent of parameterisation (see Corollary 7.2).
Now is diagonalisable. As discussed above, we can find w1 , w2 Tp (S) which are
othonormal with respect to the first fundamental form (that is ||w1 || = ||w2 || = 1 and
w1 .w2 = 0 with respect to the standard dot product on R3 ) and 1 , 2 R such that
(w1 ) = 1 w1 and (w2 ) = 2 w2 .

Definition : 1 and 2 are the principal curvatures of S = r(U ) at the point p = r(u, v).

Definition : The product, = 1 2 , is the Gauss curvature of S = r(U ) at the point

p = r(u, v).

(Again, we will see in Section 7 that these notions are invariant under reparameteri-
sation of S.)
Note that
= 1 2
 
1 0
= det
0 2
 
a b
= det
c d
 1  
E F L M
= det
F G M N
LN M 2
= .
EG F 2
(To get between lines 2 and 3 above, we have used the fact that these two matrices represent
the same linear map, , with respect to different bases.)
We have shown:

Theorem 6.1 : The Gauss curvature is given by:

LN M 2
= .
EG F 2

We will return to Gauss curvature in Section 8.

42
 curves and surfaces

Examples :
(1) Plane.
Clearly 1 = 2 = = 0.
(2) Sphere of radius a.
We saw that the s.f.f. is a1 times the f.f.f. Thus
 1  
E F L M
= a1 I.
F G M N
and so
1 = 2 = a1 = a2 .
(3) Graphs.
(u, v) 7 (u, v, f (u, v)), with critical point at the origin. The f.f.f. is the identity and the
s.f.f. is the hessian. Thus, 1 , 2 are the eigenvalues of the hessian matrix, and
 
fuu fuv 2
= 1 2 = det = fuv fuu fvv .
fuv fvv
Note that if f has a maximum or minimum at the critical point, then 0. If it has a
saddle, then 0.
(4) Developable surface.
r = (u) + vT(u)
ru = T + vN
rv = T.
Here N is the (principal) normal to . (Recall that T = N.) The normal to the surface
is B = T N, i.e. the binormal to .
ruu = N + v N + v(T + B)
since N = T + B, and
ruv = Tu = N rvv = 0.
Thus
L = ruu .B = v M = ruv .B = 0 N = 0.
And so the second fundamental form is
v du2 .
Also
1  1 
1 + v 2 2
     
E F L M 1 v 0 /v 0
= =
F G M N 1 1 0 0 /v 0
so the principal curvatures are 0, /v, with the 0 eigenvalue corresponding to the rv
direction. In particular, we see that = 0.

43
 curves and surfaces

Exercise : The principal (hence Gauss) curvatures are invariant under postcomposition
with a rigid motion of R3 .

Mean curvature.
Let H = 12 (1 + 2 ), where 1 , 2 are the principal curvatures.

Definition : H is the mean curvature of S at p.

Like Gauss curvature, this is a smooth function on the surface.

Note that this is the same as half the trace of the shape operator.

Exercise :
GL + EN 2F M
H= .
2(EG F 2 )

(Recall that the trace of a matrix is invariant under conjugation.)

The mean curvature measures the first variation in area under normal displacement
of S (see exercise sheet).
Definition : A minimal surface is one for which the mean curvature vanishes.

In other words, GL + EN = 2F M everywhere.

For example a surface that locally minimises area is a minimal surface. Hence soap
films are minimal surfaces (provided the air pressures on both sides are equal).
Examples : The plane, helicoid and catenoid are all minimal surfaces. (Exercise by
tedious calculation, or see exercise sheet for a simpler way.)

7. Covariant derivatives and parallel transport.

We shall formulate this notion in terms of embedded surfaces in R3 .

Let S be a smooth surface. Recall that a map f : S Rn is smooth if and only if
f r : U Rn is smooth for all charts r : U S.

Definition : An ambient vector field on S is a smooth map a : S R3 .

Here we imagine a(p) as a vector based at the point p. We can also speak of ambient
tangent fields restricted to subsets of S.

Example : The normal vector field p 7 n(p).

44
 curves and surfaces

Definition : A tangential vector field is an ambient vector field, a, such that a(p) Tp (S)
for all p S.

Here we can imagine a(p) as lying on the surface S.

Example : If r : U S is a chart, then ru and rv are tangential vector fields on r(U ).

Suppose that a is an ambient vector field on S. If r : U S is a chart, then the derivatives

au , av are also vector fields (on r(U )). (More precisely, we define au = u a r(u, v) and
av similarly.) Suppose that e Tp (S) is a vector (which we imagine to be based at p). We
write e = ru + rv , and set
De (a) = au + av .

[This is not standard notation, though the notation e derived from it below, is fairly
standard.]

Lemma 7.1 : De a is independent of the chart r.

Proof : Suppose we have another chart, say s, with coordinates , . By Lemma 5.3,
transition maps are smooth, so we can write:

s = u ru + v rv

s = u ru + v rv .

In terms of these new coordinates, write

e = s + s .

We verify that = u + u and = v + v .

Now, in terms of the new coordinates we have:

De a = a + a
= (u au + v av ) + (u au + v av )
= (u + u )au + (v + v )av
= au + av

which agrees with the original formula.

Note also that the map e 7 De a is linear in e.

45
 curves and surfaces

Application to the second fundamental form.

Before continuing, we relate this to the shape operator, as described in Section 6.
Recall that : Tp (S) Tp (S) was the linear map defined by (ru ) = nu and (rv ) =
nv . In the above notation, this becomes (ru ) = Dru n and (rv ) = Drv n. By
linearity of both sides we see that

(e) = De n

for all tangent vectors in e Tp (S). In particular, using Lemma 7.1, we deduce the result
promised in Section 6, namely:

Corollary 7.2 : The map : Tp (S) Tp (S) is invariant under reparametisation.

Note that in these terms, we can write the second fundamental form as:

(e, f ) = e.(f ) = e.Df n,

where e, f Tp (S). It follows that the second fundamental form on Tp (S) is also well
defined.

Exercise : Verify directly that (e, f ) = (f , e).

Remark : We can also view the shape operator as (minus) the derivative of the Gauss
map. This is just a shift of perspective. Instead of imagining n as an ambient vector field
on S, we can think of it as a map n : S S2 to the unit 2-sphere, S2 . This is called the
Gauss map. We check that the tangent spaces Tp (S) and Tn(p) (S2 ) are equal, and that
is just the derivative, n , of this map.

Continuing where we left off :

Suppose that e is now a tangent vector field on S. We define the ambient vector field
De a by setting De a(p) to be De(p) a evaluated at p. Note that, by definition, Dru a = au .
(Strictly speaking, ru is not a vector field on r(U ), since its domain is U , not r(U ).
Formally we should write it as ru r1 : r(U ) R3 . When precomposed with the chart
r, we recover (ru r1 ) r = ru : U R3 . Informally when the chart is clear, it is often
omitted from the notation. We informally think of u and v as co-ordinates locally in the
surface. Thus we can think of (u, v) as determining the point p = r(u, v) without explicitly
mentioning r.)
If f : S R is a smooth function, then we can similarly define Dru f = fu = f /u
and Drv f = fv = f /v. As with vector fields, we can extend linearly to define De f ,
which is another smooth function on S. As before, we can verify that this is independent
of parametisation. Moreover, a simple calculation gives the product formula:

De (f a) = (De f )a + f De a.

46
 curves and surfaces

(There are also obvious formulae for differentiating products of functions, and dot products
of vector fields.)
Note that De a need not be a tangential field, even if a is. To rectify this, we define:

e a = De a ((De a).n)n.

In other words, e a is the orthogonal projection of De a to Tp (S). By definition, this is a

tangential field. (Of course, one needs to verify that all these constructions are smooth.)
We dont need a itself to be tangential in the above.
If a is tangential we refer to e a as the covariant derivative of a in the direction of
e. Viewed as a function with two arguments, e and a, is sometimes referred to as a
connection. (More precisely, it is the Levi-Cevita connection in this case.)
If f is a smooth function, we will simply write: e f = De f .

Lemma 7.3 : Suppose that is a smooth function on S, and that a is a tangential

vector field. Then
e (a) = (e )a + e a.

Proof : By linearity in e, its enough to show that:

u (a) = u a + u a.

Using a.n = 0, we have:

u (a) = (a)u ((a)u .n)n

= u a + au u (a.n)n (au .n).n
= u a + (au (au .n)n)
= u a + u a

as required.

By a similar argument, if n is the unit normal to S, then

e (n) = e n.

Intuitively, the contribution to the derivative due to varying is in the normal direction,
and so this term vanishes on projecting to the tangent space.
We next aim to prove:

47
 curves and surfaces

Theorem 7.4 : Suppose that a, e are tangential vector fields. Then e a depends only
on the first fundamental form of S.

In other words, it is an intrinsic construction. In fact, if r : U S is a chart, then

we can express e a in terms of the coordinates of e and a with respect to ru and rv , the
derivatives of these coordinates with respect to u and v, the functions E, F, G, featuring
in the first fundamental form, and their derivatives, Eu , Fu , Gu , Ev , Fv , Gv . We shall not
derive an explicit expression here, though this is in principle possible by working carefully
through the argument given below. (See [???] for a formula.)
For the proof we make the following observation:

Lemma 7.5 : Suppose that x, y, z R3 are vectors and that x and y are linearly
independent. Then the orthogonal projection of z to the plane spanned by x and y is
given by
((y.y)(x.z) (x.y)(y.z))x + ((x.x)(y.z) (x.y)(x.z))y
(x.x)(y.y) (x.y)2

Proof : Exercise.

All we really need to note from this is that the coordinates are expressible in terms of
x.x, y.y, x.y, x.z and y.z.

Proof of Theorem 7.4 : By linearity in e, it is enough to express u a and v a in

terms of the first fundamental form. We will just deal with u a. Now a = ru + rv
where , are smooth functions on S. Also, by Lemma 7.3,

u (ru ) = u ru + u ru

u (rv ) = u rv + u rv .
Thus, its enough to express u ru and u rv in terms of the first fundamental form.
Consider u ru . This is the projection of ruu to the plane spanned by ru and rv . Thus,
by Lemma 7.5, its coefficients are expressible in terms of ru .ru = E, ru .rv = F , rv .rv = G
together with ruu .ru and ruu .rv . But ruu .ru = 12 (ru .ru )u = 21 Eu , and one checks that
ruu .rv = Fu 12 Ev .
Similarly for u rv , its enough to note that ruv .ru = rvu .ru = 21 Ev etc.

Lemma 7.6 : Suppose that e, a, b are tangent vector fields on S. Then

e (a.b) = a.e b + b.e a.

48
 curves and surfaces

Proof : Exercise.

Suppose that f : S is a diffeomorphism. For each p S, we have an induced

map f : Tp (S) Tf (p) () (as described in Section 5). A tangential vector field, a, on
S thus gives rise to a tangential vector field on , which we denote by f a. If f is an
isometry, then it respects the first fundamental forms of S and . If e and a are tangential
vector fields, then applying Lemma 7.5, it follows that f e f a = f (e a). This can be
expressed informally by saying that commutes with f .

Let be a smooth curve in S. If a is a tangent vector field on S we define:

D
a((t)) = (t) a.
dt
D
We will frequently abbreviate this to dt a when the curve is implicitly assumed.

Remark : We only really need a to be defined on the image of for this to make sense.
More precisely, if a, b are tangential vector fields, and a((t)) = b((t)) for all t, then
D D
dt a = dt b (exercise).

In this notation, Lemma 7.6 implies that, for any vector fields a, b on S, we have
d D D
(a.b) = a. b + b. a
dt dt dt
d d
(where dt (a.b) is an abbreviation for dt (a((t).b((t))).

d D
From the definition, it follows easily that D (t) a = dt (a (t)) and so dt a is the
d
othogonal projection of dt (a (t)) to S, or more precisely, to the tangent space of S at
(t).
Suppose now that s is an arc-length parameterisation of . That is (s). (s) = 1,
D
and so, by Lemma 7.6, (s). ds (s) = 0. Therefore we can write
D
(s) = S NS
ds
where NS is a unit tangent vector in S normal to (s).

Definition : S is the geodesic curvature of in S.

(In fact, it can be given a sign, depending on where NS is to the left or right of in the
orientation of S.)
D
Here is another description of geodesic curvature. We have observed that ds (s) is the
d
projection of ds (s) = (s) to T(s) S. But (s) = N , where N is the (principal)
normal to and is the curvature of (as defined in Section 2). Thus, S NS is the
othogonal projection of N to T(p) S.

49
 curves and surfaces

Examples :
(1) If is a curve in the plane, S = R2 R3 , then (s) lies in R2 so S = and
NS = N .
(2) The unit 2-sphere.
The geodesic curvature of the latitude of angle is tan .
Remark: one can view this in terms of the cone that touches the latitude tangentially.

Parallel transport.
Suppose that a is a tangential vector field on S, and that is a curve in S.

D
Definition : We say that a is parallelly transported along if dt
a = 0, for all t.

Remark : We only need a to be defined on the image of for this to make sense.

D D
Note that by lemma 7.6, dt (a.a) = 2a. dt a = 0, and so ||a((t))|| is constant along .

The following give the existence and uniqueness of parallel transport of a vector along a
given curve:

Theorem 7.7 : Suppose that : [t0 , t1 ] S is a smooth path and a0 T(t0 ) (S).
Then there is a unique smooth vector field, a, along , tangential to S, with a(t0 ) = a0
D
and dt a((t)) = 0 for all t.

Proof : (Sketch) Let r : U S be a chart. Assume first that ([t0 , t1 ]) r(U ).

Write (t) = (t)ru + (t)rv . It follows from Lemma 5.4 that and are smooth. Let
a((t)) = a1 (t)ru ((t)) + a2 (t)rv ((t)). The functions a1 , a2 are smooth if and only if a is
smooth (directly from the definition of vector fields). We saw (as in the proof of Theorem
D
7.4) that dt a(t) = (t) a((t)) can be expressed as a function of , , E, F, G, a1, a2 , and
their derivatives with respect to u and v. (Following that argument as given there, we
need to assume that a is actually defined on a neighbourhood of the image of , but this
is not essential to the proof.) Looking closely at the argument we see that the derivatives
of ai with respect to u, v only appear as part of expessions of the form (ai )u ru + (ai )v rv
d
and so these can be replaced by dt (ai (t)) = ai (t). Moreover, the resulting expression is
linear in the terms a1 (t), a2 (t), a1 (t), a2 (t). In summary, we get a system of first order

linear differential equations for a1 (t) and a2 (t). The coefficients are expressible in terms of
(t), (t), E((t)), Eu((t)) etc. and are hence smooth functions of t. Thus, by Theorem
3.4, they have a unique solution with the given initial condition.
To deal with the general case, we cut into a finite number of subpaths, such that the
image of each subpath lies in one coordinate patch. (This uses compactness of the closed
interval.) We then apply the above inductively to each subpath in turn.

Suppose that is a curve parameterised by arc length, s, and that a is a parallelly

50
 curves and surfaces

transported vector field along . Then

D D
( (s) a((s))) = (s) = S NS .
ds ds

Thus we can think of the geodesic curvature, S as measuring the rate at which the curve
turns in relation to parallel transport.

Total turning.
Suppose that is a closed curve in a surface S. Then
R (applying the same definition
as in Section 2), we can define the total turning of as S (s)ds.

Examples :
(1) For a curve in the plane S = R2 R3 , this agrees with the definition given in Section
2.
As
R 1an example,1 we can compute the total turning of a round circle, , of radius r as
r
ds = (2r) r = 2.
Indeed, we remarked there that the total turning of any Jordan curve was 2, depending
on orientation.
(2) The latitude on the unit 2-sphere has length 2 cos , and hence total turning

2 cos tan = 2 sin .

Foucaults pendulum
The direction of oscillation of a pendulum carried along the earths surface will move
by parallel transport. If the suspension of the pendulum is stationary relative to the earth,
then in absolute terms it is being carried once around a latitude every day (or every 23
hours and 56 minutes, if one takes into account the earths motion relative to the sun in
that time). The observed angle it turns through is therefore equal to the total turning of
the latitude. That is 2 sin at latitude . As expected this is 0 at the equator and 2
at the poles.
Foucault hung his pendulum from the dome of the Pantheon in Paris in 1851. It was
the first dynamical demonstration of the Earths rotation. The original pendulum was
reinstalled there in 1995. Foucault also suggested the use of a vertical gyroscope to similar
effect, though this was harder to implement in practice.

Geodesics.

D
Definition : A path in S is geodesic if dt
(t) = 0.

This can equivalenty be expressed by saying any one of the following:

(a) The tangent vector is parallelly transported.

51
 curves and surfaces

(b) The geodesic curvature is identically 0, or

(c) (t) is always orthogonal to T(t) (S).
Note that || (t)|| is constant, that is, has constant speed.

Examples :
It is generally complicated to write out explicit formulae for geodesics. However, certain
examples can be seen by symmetry: where the surface admits an (orientation reversing)
isometry fixing a given curve, thereby showing that its geodesic curvature must be 0. For
example:
Straight lines in the plane.
Great circles of the sphere.
Generators of surfaces of revolution.

Remarks : (We wont prove these here.)

(1) Constant speed length minimising paths are geodesics. In fact any geodesic is locally
length minimising.
(2) Given any p S and any x Tp (S) there is a geodesic path : [a, a] R with
(0) = 0 and (0) = x, where a > 0. Any two such paths agree on the intersection of
their domains.
(3) We say that S is complete if any geodesics can be extended indefinitely (that is, with
domain R). If S is complete, then any two points can be connected by a geodesic (the
Hopf-Rinow Theorem). One can show that if S R3 is closed (as a subset of R3 ) then
S is complete.

8. Gauss curvature.

Let us recall some definitions.

Let S be a regular surface and p S. We have defined the linear map:

: Tp (S) Tp (S)

by
e 7 e n

where n the unit normal ambient vector field. (Note that e n = De n since n.De n = 0.)
This has two real eigenvalues, 1 and 2 . The Gauss curvature is defined as = 1 2 .
(Our sign convention regarding the shape operator is not going to matter here, since
= 1 2 = (1 )(2 ).) We have seen that it can be expressed in terms of the first and
second fundamental forms as:
LN M 2
= .
EG F 2

52
 curves and surfaces

In fact, it was shown by Gauss to be intrinsic to S, that is, it depends only on the first
fundamenal form (and its derivatives) and is therefore preserved by isometry. He called
this a Whopping Great Theorem or Theorema Egregium in Latin:

Theorem 8.1 : (Theorema Egregium) The Gauss curvature, , depends only on the
first fundamental form.

In fact, if r : U S is a coordinate chart, then can be expressed in terms of the

functions E, F, G and their first and second derivatives with respect to u and v. Recall
that n = (ru rv )/||ru rv ||.
The proof we present here is taken from Hitchins notes [H] (with some elaboration).

Proof : We will consider how the operator

v u u v

behaves when applied to smooth tangential vector fields on S.

Let a be a tangential vector field. Recall that

u a = au (au .n)n
= au + (a.nu)n.

So
v u a = v (au + (a.nu)n)
= v au + v ((a.nu )n)
= auv (auv .n)n + v ((a.nu)n).
As observed after Lemma 7.3, we have v ((a.nu)n) = (a.nu )v n = (a.nu )nv . (Since
v n = nv (nv .n)n = nv . Thus,

v u a = auv (auv .n)n + (a.nu )nv .

Swapping u and v, and using auv = avu , we get:

v u a u v a = (a.nu )nv (a.nv )nu

= (nu nv ) a.

Now nu , nv Tp (S) and so nu nv is normal to S. Thus, we can write

nu nv = n

, where is a smooth real valued function on S. We have:

(v u u v )a = n a.

53
 curves and surfaces

(In other words, u v v u rotates a though /2 in each tangent space, and then scales
it by a variable factor of .) By Theorem 7.4, the action of v u u v is intrinsic to
S, and so therefore is the function .
Now
n.ru rv = (nu nv ).(ru rv )
= (nu .ru )(nv .rv ) (nu .rv )(nv .ru )
= LN M 2
and p
n.ru rv = ||ru rv || = EG F 2 .
Thus,
LN M 2
= .
EG F 2
It follows that p
LN M 2 = EG F 2
is intrinsic, and so
LN M 2
=
EG F 2
is intrinsic also.

Going through the above argument carefully one can derive an explicit formula for ,
though we wont attempt that here. A formula can be found, for example in [M p.213].
Apparently this is known as Brioschis formula.
Note that it follows that if f : S is an isometry, then the Gauss curvature of S
at p is equal to the Gauss curvature of at f (p).
For example, we have seen that the Gauss curvature of the sphere is non-zero. Thus,
there can be no isometry from the sphere to the plane, even locally. This is why any planar
map of the earths surface must inevitably distort distances.

Surfaces of revolution.
Let (t) = ((t), (t)) be a curve in the plane with (t) > 0 for all t. Recall that the
associated surface of revolution can be defined by:

r(u, v) = ((u) cos v, (u) sin v, (u)).

From the formula = (LN M 2 )/(EF G2 ) one can derive:

( )2
= .
(( )2 + ( )2 )2

Exercise: or see or [M p.141].

54
 curves and surfaces

If t is arc-length, then ( )2 + ( )2 = 1 and so = . In this case, the above

formula reduces to:

= .

As an example, the unit sphere (minus the poles) can be viewed as the surface of
revolution of the semicircle (t) = (cos t, sin t), for t (/2, /2). In this case, we get
(t) = cos t = (t), = 1 as before.

The pseudosphere.
The pseudosphere is the surface of revolution of the tractrix, discussed in Section 2.
Recall that a unit speed parameterisation of the tractix has the form

(s) = (es , y(s))

and so
(s) = es
giving
= 1.
In other words, the pseudosphere has constant Gauss curvature 1.
One can also get the same result by a more laborious calculation using the parame-
terisation
(t) = (sin t, log tan(t/2) cos t)
(Exercise, or see [M p142].)

Remark : Note that the pseudosphere is not complete. In fact, it cannot be extended
beyond the boundary circle (obtained by spinning (0)) in such a way as to have curvature
1.
In fact, it is a theorem of Hilbert that there is no complete embedded surface in R3 of
constant curvature 1. (See [M p.203].)
This is one reason to develop a theory of intrinsic geometry of surfaces without reference
to embeddings in 3-dimensional space. This is the subject of riemannian geometry. It is
discussed in Section 11.

9. The local Gauss-Bonnet Theorem.

Here we give a local form of the Gauss-Bonnet theorem. This relates the Gauss
curvature of a surface to the geodesic curvature of a curve. In the next section we will give
a global form for closed surfaces.
Let S be a regular surface, and S be a smooth disc in S. Its boundary, ,
is a smooth closed curve, which we parameterise by arc-length, s. We assume that is
positively oriented, that is is to the left of when proceeding in a positive direction.

55
 curves and surfaces

Recall the definitions of Gauss curvature, , of S, and geodesic curvature, S , of , from

earlier sections.

Theorem 9.1 : (Gauss-Bonnet)

Z Z
S ds = 2 dA.

Here dA denotes the area element (as discussed in Section 4). Note that the LHS is
the total turning of in S as defined in Section 7.
By a smooth disc in S, we mean a closed subset, S that is the image of a Jordan
domain in the plane under an injective smooth regular map. In fact, it will be convenient
to insist that the map is defined on a neighbourhood of the Jordan domain. More precisely,
there is a chart, r : U S such that r(U ) and = r1 is a Jordan domain. The
boundary curve, , of gets sent to the boundary curve, , of under r. Note that, by
the smooth Schoenflies Theorem (2.2), there is no loss in assuming that is, in fact, the
unit disc.
The argument is again based on Hitchins notes [H].
Let r : U S, with r(U ). Let = r1 be a Jordan domain as above. We
write for its boundary, viewed as a smooth curve.
We recall Greens theorem from Section 2. Namely that if P, Q are smooth functions
defined on , then Z Z

(P u + Qv ) dt = (Qu Pv ) dudv ()

In what follows, to simplify notation, we will generally suppress mention of the map,
r, and just work with (u, v) coordinates on r(U ) S. We can then apply Greens theorem
() directly to r(U ).

Proof of Theorem 9.1 : Set e = ru / E. Thus e is a tangential vector field on r(U )
with ||e|| = 1. Let f = n e. Thus, f is another unit tangential field on r(U ), orthogonal
to e. Note that e, f , n is an orthonormal frame in R3 . Now since e.e = 1, we have
e.u e = e.v e = 0 so we can write

u e = P f v e = Qf ,

where P and Q are smooth functions, namely P = (u e).f and Q = (v e).f .

We parameterise by arc length, s, and write t = (s) for the unit tangent. Given
D
any vector field, a, along , we write We write a = ds a = t a.
Now the left hand side of () is:
Z Z
(P u + Qv ) ds = (u u e + v v e).f ds


Z .

= e .f ds

56
 curves and surfaces

Since ||t|| = 1, we can write t = cos e + sin f , where is a smooth function on the
image of . In fact, itself is only defined up to addition by an integer multiple of 2,
though the derivative with respect to t, is well defined.
We claim that
e .f = S .
To see this, note that

n t = cos n e + sin n f
= cos f sin e

is the unit vector in Tp (S) obtained by rotating t through /2. Now, differentiating t using
Lemma 7.3, we get

t = cos e sin e + sin f + cos f .

From the definition of geodesic curvature, S , and using e.f = f .f = e .f + e.f = 0, we

get
S = t .(n t)
= (cos2 e .f sin2 e.f ) + (cos2 + sin2 )
= e .f + .
verifying the claim.
Thus, the left hand side of () is:
Z Z
e .f ds = (S ) ds.


R
Now ds measures the turning of the curve relative to the field e, which must be
equal to is equal to 2. (A formal statement of this is given as Lemma 9.2 below.) Thus,
the left hand side of () becomes:
Z
S ds 2.

For the right hand side, first note that f .u f = 0, and so u f = He for some real
valued function, H. But (e.f )u = f .u e + e.u f = 0, and so P + H = 0, that is H = P
and we get:
u f = P e.
Similarly
v f = Qe.
Thus:
v u e = v (P f )
= P v f + P v f
= Pv f P Qe.

57
 curves and surfaces

Similarly
u v e = Qu f P Qe
and so
(v u u v )e = (Pv Qu )f .
But as in the proof of Theorem 8.1 we have:

(v u u v )e = f

where p
= EG F 2 .
In other words, p
Pv Qu = EG F 2
and so the right hand side of () becomes:
Z p
EG F 2 dudv.


But EG F 2 dudv is precisely the area element in S, and so
Z p Z
2
EG F dudv = dA.

Equating the two sides, we have shown:

Z Z 
dA = S ds 2

Z
= 2 S ds.

Here is a more formal statement regarding the total turning of a curve relative to a
vector field, needed in the proof.
Let S is a smooth disc with boundary curve : S1 S where S1 is the unit
circle. Suppose that e is a unit tangent vector field on . Given S1 , let () be the
angle () makes with e(()). More formally, we can write:

()/|| ()|| = cos e + sin (n e).

Note that this determines cos and sin completely, and so () is well defined up to
adding some element of 2Z. It follows that () is completely well defined. (We can
assume, for any Rparticular , that is chosen to be continuous in a small neighbourhood of
.) Let T () = () d. (One can check that this is invariant under reparameterisation
of .) We can think of it as the turning of relative to the vector field e.

58
 curves and surfaces

Lemma 9.2 : T () = 2.

Proof : (Sketch) [There must be a better way of doing this...]

We begin by observing that T () 2Z. To see this, we parameterise S1 by radial
angle , varying between 0 and 2. We fix a particular value of (0). We define () =
R 2
(0) + 0 () d. This gives a particular value of () for all [0, 2]. By definition
R 2
of the closed curve integral, we have T () = 0 () d = (2) (0). Now, from
the definition of we see that ()/|| ()|| = cos () e + sin () (n e) holds for all
[0, 2]. In particular, cos (0) = cos (2) and sin (0) = sin (2), and so (2)
(0) 2Z as claimed.
Let D = {x R | ||x|| 1}. Let , be polar co-ordinates. Then there is a smooth
injective map h : D S which is a diffeomorphism to h(D). Given (0, 1], define
: S1 S by () = h(, ).R This is regular curve. Let () be the angle between
(). and e( ()). Let T () = () d. As for , we see that T ( ) 2Z.
Note that [ T ( )] is continuous, and always lies in 2Z. Thus, T ( ) = T ()
for all (0, 1].
Now let c = h(0) S, and let h : R2 Tc (S) be the derivative at this point. As
0, () converges uniformly to 0 (), where 0 () is the angle h (t()) makes with
the fixed tangent vector, e(c), where t() = ( sin , cos ). (Note that t() is the unit
tangent to S1 at . For given , the unit tangent vector ()/|| ()|| tends to h (t()).)
As goes once around S 1 , so does 0 (). (This can be written out explicity, given that
h is just a linear map. It is Rthe statement that the total turning of an ellipse in the plane
is equal to 2.) So T ( ) S1 0 () d = 2, and so T ( ) = 2 for all .

Examples : (of the local Gauss-Bonnet theorem).

(1) A Jordan curve in the plane.
R
In this case = 0 and we get S ds = 2 which is just the formula for total turning.
(2) Latitude of the unit sphere.
Let be the -latitude. This bounds the northerly spherical cap, . We have already
seen that the total turning of is:
Z
S ds = 2 sin .

We also have
area() = 2(1 sin ).

(This can be conveniently seen using radial projection to a cylinder, namely Lamberts
projection discussed in Section 5. We saw this was area preserving, and so the area of the
cap is the area of the portion of the cylinder lying above the horizontal plane of through
the latitude, i.e. of height sin .)

59
 curves and surfaces

Now = 1 and so: Z

dA = area()

= 2 2 sin
Z
= 2 S ds

which agrees with the Gauss-Bonnet formula.

Piecewise smooth curves.

We finish this section by remarking that the main result can be generalised to a
piecewise smooth Jordan curve. If is such a curve, we can define its exterior angles
1 , 2 , . . . , n [, ], at the vertices, cf. the discussion of plane curves in Section 2.
(The angles are defined with respect R to the first
Pn fundamental form on S.) We can define
the total turning, (), of as S ds + i=1 i . Suppose that bounds a disc .
R
The Gauss-Bonnet theorem holds in the form () = 2 dA. It can be proven,
for example, by approximating by smooth curves, though the details are a bit technical.
(Here the term disc should be interpreted to be the image of a disc under an injective
continuous map. Such a map can be taken to be smooth away from the vertices. We will
not worry too much about these technicalities here.)

10. Some global properties of surfaces.

Much of the discussion here will be less than rigorous. The necessary formal back-
ground could be found in courses on measure theory and differential geometry.
The following bit about orientations could have come earlier, but weve not needed it
until now.

Orientations.
Let U, V R2 be open. Suppose : U V is a diffeomorphism. Given x U ,
write D(x) = det (x), i.e. the determinant of the jacobian. Note that D(x) 6= 0 for all x.

Definition : We say that is orientation preserving (resp. orientation reversing) if

D(x) > 0 for all x U (resp. D(x) < 0 for all x U ).

We note that U is connected if and only if any two points of U can be connected by a
smooth path. (This is an exercise if you know about the general notion of connectedness,
otherwise it can serve as a definition here.)

Exercise : If U is connected, then V is connected, and any such diffeomorphism is

either orientation preserving or reversing.

60
 curves and surfaces

Definition : Suppose r : V R3 is a parameterised surface, and s = r is a

reparameterisation. We say they are consistently oriented if is orientation preserving.
An atlas is oriented if all pairs charts are consistently oriented on their overlap.
S is orientable if it admits an oriented atlas.

Lemma 10.1 : S is orientable if and only if there is a smooth ambient normal field, n.
(i.e. n(p) is perpendicular to Tp (S) for all p S).

Note that we can always assume n to be a unit normal field.

We have noted that any surface has a locally defined normal field, i.e. defined in the
neighbourhood of any point. The point here is that it can be defined globally, i.e. on all
of S.

Proof : Suppose S is orientable, and choose an oriented atlas. For each chart r set
n(r((u, v))) = (ru rv )/||ru rv ||. This is well defined on the overlaps.
We leave the converse as an exercise.

By an orientation on S, we mean a unit normal field.

Remark : Orientation can be expressed purely in terms of the atlas of S.

Definition : We say that S is smoothly path connected if and only if every pair of points
are connected by a smooth path.

Exercise : (for topologists) S is smoothly path connected if any only if it is path con-
nected. (If you dont know what this means, dont worry.)

Exercise : If S is smoothly path connected and orientable then it has exactly two orien-
tations.

The following terminology is confusing, but standard.

Definition : A closed surface in R3 is surface S R3 that is closed (in the usual

topological sense), bounded, and smoothly path connected.

Examples : Sphere, torus etc.

Remark : (for topologists) A subset of R3 is (topologically) closed and bounded if and

only if it is compact. In this case, S is smoothly path connected if and only if it is path
connected and if and only if it is connected.

61
 curves and surfaces

Fact : A any closed surface in R3 is orientable.

If we dont assume closed, this is false: Mobius band.

In fact, if S is a closed surface in R3 , then R3 \S has exactly two components one bounded
inside and another unbounded outside. This requires some topology to prove.

Euler characteristic
This can be defined for a topological space as an alternating sum of Betti numbers
the dimensions of the homology groups with real coefficients. For a closed surface S, it
can be described as follows:
A graph embedded in S consists of a finite set, V S, of vertices, and a finite
set, E, of edges, such that each edge is a smooth regular injective path connecting two
distinct points of V and meeting V only at its endpoints, and such that and any two
distinct edges meet at most at their endpoints. We write for the union of the edges.
A face is a component of S \ . We write F for the set of faces. Let v = |V |, e = |E|
and f = |F |. We say that is a cellulation if the closure of each face is homeomorphic
to a disc. A triangulation is a cellulation in which each face has exactly three edges in its
boundary.

Fact : A closed surface admits a triangulation. (In fact, we can assume that each triangle
lies in a co-ordinate patch.)

Fact : The number

=ve+f

is same for any cellulation.

If you believe these facts, then this gives us a well defined number, = (S) called
the Euler characteristic

Examples : The Euler characteristic of the sphere is 2 (Eulers formula) and that of
the torus is 0.
In fact we get an infinite sequence of surfaces with Euler characteristic 2, 0, 2, 4....

Theorem 10.2 : (Classification of closed (orientable) surfaces) The Euler characteristic

of any closed surface is one of the above numbers (i.e. 2 2g for some g N. (i.e. 2 2g
for some g N. Any two surfaces with the same Euler characteristic are diffeomorphic.

In other words any closed surface is diffeomorphic to exacty one of the above examples:
i.e. a surface with g holes, where g is called the genus and = 2 2g.

62
 curves and surfaces

Proof : For the general idea, see a book on topology (though these dont always discuss
diffeomorphisms).

Remarks :
(1) We say two surfaces, S and are homeomorphic if there is a continuous bijective
map f : S such that f 1 is also continuous. It follows that they have the same euler
characterictic and are hence diffeomorphic. (i.e. we can replace f by another map h such
that h and h1 are both smooth.

(2) If S is diffeomorphic to a 2-sphere then there is a diffeomorphism of R3 to itself sending

S to the round sphere. (This is the 3-dimensional Schoenflies theorem.) The analogous
statement is also true an (n 1)-sphere in n space for n 5 but unresolved for n = 4.
This fails for higher genus sufaces: we can knot tori etc.

Note that if we have a triangulation, then 2e = 3f and so

3 f
=v f +f =v .
2 2

Integration.
Suppose f : S R is a smooth function, and Q S is a nice subset with
Q r(U ) for some chart r : U S. We define:
Z Z
f dA = f r(u, v) ||ru rv || dudv.
Q r1 Q

Here dA is the area element. (Recall that ||ru rv || = EG F 2 , where E, F, G are
the coefficients of the first fundamental form.) In the above, nice means measurable
in the sense of measure theory. It is sufficient that Q be open or closed, for example.

Exercise : This is independent of the chart: If Q s(V ), where s : V S is another

chart, then
Z Z
f r(u, v) ||ru rv || dudv = f s(, ) ||s s || dd.
r1 Q s1 Q

Now given a triangulation of S, we can define:

Z XZ
f dA = f dA
S

as ranges over all faces of the triangulation.

63
 curves and surfaces

Fact : This is independent of the choice of triangulation.

One would expect this at least to be invariant under subdivision of the triangulation.
If it were true that any two triangulations had a common subdivision, the fact would then
follow. Unfortunately, life is more complicated. To do this properly it would be better to
define integration using partitions of unity instead of triangulations see the differential
geometry course.

Theorem 10.3 : (Gauss-Bonnet) Let S be a closed surface. Let : S R be the

Gauss curvature. Then Z
dA = 2(S).
S

Proof : Let V, E, F be the sets of vertices, edges and faces of some triangulation of S.
Suppose that F . Let 1 , 2 , 3 be the interior angles. Then the total turning of is
XZ
3 1 2 3 + S ds,

where ranges over the three edges of , positively oriented, S is the geodesic curvature,
and s is arc length. Thus, the local version of the Gauss-Bonnet theorem (as discussed at
the end of Section 9) gives:
Z XZ
dA = 1 + 2 + 3 S ds.

R
We now sum over all F . The LHS becomes S dA. For the RHS, summing the angles
gives a contribution of 2 for each vertex of V , that is, giving 2v in total. Summing the
s gives f . The path integrals cancel out: for each edge of E, we get two contributions,
one for each of the two incident faces. But these are oriented in opposite directions, and
so the signs of the geodesic curvatures are opposite. Equating, we get:
Z
dA = 2v f = 2(v f /2) = 2
S

as required.

Examples :
(1) For example for the round 2-sphere, 1, and so we recover the formula:
area(S2 ) = (2)2 = 4.

(2) Suppose that (u) = ((u), (u)) is a closed embedded curve in the plane, with (u) > 0
for all u. The surface of revolution, S, is diffeomorphic to a torus, so (S) = 0. If u is arc
length, recall that (u, v) = (u) /(u) and the f.f.f. is
du2 + (u)2 dv 2

64
 curves and surfaces

so the area element is dA = du dv. Thus

Z Z Z
dA = (u) du dv = 2 (u)du = 0

S S
since we are intergrating around a closed curve.

11. 2-manifolds.
This section may or may not be examinable. Well decide depending on how much
time we have at the end of the course.
Heres how to define a surface (or 2-manifold) without reference to embeddings in
R3 .

Let be a metric space. (Were only really interested in it as a topological space.)

Definition : A (topological) chart is a bijective map r : U S, where U R2 and

r(U ) are open, and r and r 1 are continuous. (i.e r is a homeomorphism from U to r(U ).)
The image of a chart is a patch.
An (topological) atlas is a collection of charts such that each point lies in a patch.
is a topological surface if it admits an atlas.
If r : U , and s : V are two charts, we refer to the map
s r 1 : V s1 (r(U )) U r 1 (s(V ))
as a transition map.
(Note that a transition map is a homeomorphism.)

Definition : An atlas is smooth if every transition function is smooth.

A smooth surface is a surface together with a smooth atlas.

Fact : Every topological surface admits a smooth atlas.

Examples :
(1) The plane, R2 , is a smooth surface. We only need one chart: the identity map. Any
open subset is also a surface.
(2) Any regular surface is a smooth surface. We can take the atlas to consist of the set of
all smooth charts (as defined earlier). We use Lemma 5.3.
(3) Define the torus as the quotient of R2 under the action of Z2 (i.e. (x, y) (z, w) if
x z, y w Z). Take an atlas to consist of all inclusions of open discs of radius less
than 1/4 composed with the quotient map. All transition functions are translations of
subsets of the plane, and are in particular smooth. This is a smooth atlas. (We refer to
this construction as the square torus.)

65
 curves and surfaces

Definition : A map f : S between smooth surfaces is smooth if s1 f r is smooth

for all charts r and s of S and respectively, wherever and whenever this is defined.
We say f is a diffeomorphism if it is bijective and both f and f 1 are smooth. We say
that S and are diffeomorphic if there is a diffeomorphism between them.

Fact : If two surfaces are homeomorphic, then they are diffeomorphic. (The homeomor-
phism we first thought of need not be smooth of course, so this is highly non-trivial.)

Let : I be a continuous path.

Definition : is smooth if r 1 is smooth for all charts r, whenever and wherever its
defined.
Note, for regular surfaces, this agrees with the definition already given, using Lemma 5.4.

Tangent spaces.
Suppose we have two smooth paths, : I R2 and : J R2 , with 0
int I int J.
Fix some a R2 , and let C(a) be the set of curves : I R2 with 0 int I and
with (0) = R2 . Given , C(a), write if (0) = (0).
Its easy to see that this an equivalence relation with C(a)/ identified with R2 via
(0).
Suppose a U R2 is open and : U R2 and : U V is a diffeomorphism.
Let b = (a) V . We identify the tangent spaces at a and b with R2 . Let : R2 R2
be the derivative at a.

Lemma 11.1 : If C(a) then C(b). Moreover, under the

idenfications C(a)/ C(a)/ R2 , sends the equivalence class [] to [ ].

Proof : Exercise.

Now suppose that is a smooth surface and p . Let C(p) be the set of smooth
curves : I S with (0) = p. Define a relation on C(p) as follows. Let r : U
be a chart with p r(U ). Let a = r 1 p U . If , C(p), we write if
r 1 r 1 where these are defined.

Exercise : This is well defined, i.e. independent of the choice of chart.

We write Tp () = C(p)/.
Note that we can identify Tp () with C(a)/ R2 . If s is another chart, with
s1 p = b, we can also identify Tp () with C(b)/ R2 . The compositions of these
idenfications, as a map R2 R2 , is precisely , where is the transition map. Since
is linear, we see that the vector spaces stuctures agree, and so Tp () has a natural
structure as a vector space.

66
 curves and surfaces

Definition : Tp () is the tangent space of at p.

Note that if is a smooth path in , then we have a well defined tangent vector
(t) T(t) .

Exercise : If is a regular surface in R3 , we can naturally identify Tp () with the

tangent space already defined.

Exercise : If f : S is smooth, then there is natural linear derivative map f :

Tp () Tf (p) (S).

We can now go on to redefine many of notions used for regular sufaces in R2 by

interpreting them in terms of charts. For example:

Definition : A riemannian metric on associates to each tangent space a positive definite

symmetric bilinear form such that it varies smoothly over .
A riemannian surface is a smooth surface with a riemannian metric. [Note, this is not the
same thing as a Riemann surface!]

(To define smoothness, here, we refer back to coordinate charts and coordinates, and
observe that it is independent of choices of charts.)

Examples :
(1) If is a regular surface, then the first fundamental form is a riemannian metric.
(2) Suppose that U R2 is open, and : U (0, ) is smooth, then we can define
a riemannian metric as (u, v)2 (du2 + dv 2 ). (i.e., the form is given by the matix I.) In
other words, we rescale the euclidan metric locally by .
(3) The square torus described above has a metric defined by du2 + dv 2 , where u, v are
the euclidean coordinates (Check this is well-defined.) This is locally isometric to (or
modelled on) the euclidian plane every point as a neigbourhood isometric to a subset
of R2 .

A riemannian metric on allows us to define lengths of smooth curves, angles, area, etc,
using the same formulae as before, and reinterpreting the f.f.f. as the riemannian metric.
One needs to check that they are well defined, independently of the charts, but that follows
by essentially the same argument.

Example : In example (2) above, the length of a curve, : I U is given by

Z
((t))|| (t)|| dt
I

67
 curves and surfaces

Definition : An isometry between two riemannian surfaces and S is a smooth map

f : S such that for all p , f sends the riemannian form on Tp () to that on
Tf (p) (S).

An isometry respects lengths, angles and area. (In fact, its equivalent to length-preserving.)

We can now go on to define parallel transport and Gauss curvature. One (very inefficient)
way to do this would be to go carefully through the proofs that these were intrinsic prop-
erties of regular surfaces, derive explicit formulae for these in terms of the f.f.f. use the
same formulae as definitions in the riemannian case. One needs to check that they are
independent of charts. In principle, this will work, but there are better ways. This is the
subject of a course in riemannian geometry.

One has (the same) classification of compact orientable surfaces. The Gauss-Bonnet the-
orem remains true. (For example, for the euclidean square torus above, we get 0 on both
sides.) It also works in the non-orientable case.

Remark : For almost everything we said we can replace 2 by n, and talk about smooth
and riemannian n-manifolds. These are not classified, and curvature is more subtle. There
is a generalisation of the Gauss-Bonnet theorem when n is even.

Example : (Hyperbolic plane). Let D = {x R2 | ||x|| < 1}. Rescale the metric by a
factor of
2
(x) =
1 ||x||2
as above. This is the Poincare model of the hyperbolic plane. We denote it H2 .

Exercise : All rotations of the disc are ismetries of H2 .

Facts :
(1) H2 is homogeneous, that is, given any x, y H2 , there is an isometry taking x to y.
(2) H2 has constant Gauss curvature 1.
(3) H2 cannot be realised as a regular surface in R3 . (Though bits of it can the
pseudosphere is locally isometric to H2 .)

Waffle : Any surface of genus at least 2 admits a metric of constant curvature 1, locally
modelled on H2 . (cf. flat torus described above).
These surfaces cannot be realised as regular surfaces in R3 .

THE END

68