Differential Geometry @ UCI

Math 162AB

Author: Zhiqin Lu
Institute: University of California, Irvine
Date: September 15, 2020
Version: 2.0
Bio: Zhiqin Lu is a professor at UCI

Lecture Notes based on Barrett O’Neill’s book Elementary Differential Geometry.

Partially supported by a NSF grant DMS-19-08513.
 Contents

0 Preliminaries 1
0.1 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.2 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.3 Linear Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.4 Lines, planes, and spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.5 Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.6 Einstein Convention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1 Calculus on Euclidean Space 15

1.1 Euclidean Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3 Directional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Curves in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.7 Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2 Frame Fields 31
2.1 Dot Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 The Frenet Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Arbitrary-Speed Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.5 Covariant Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6 Frame Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7 Connection forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4 Calculus on Surface 18
Chapter 4 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.6 Integration of Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.7 Topological properties of surfaces . . . . . . . . . . . . . . . . . . . . . . . . 22
4.8 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
 CONTENTS

5 Shape Operators 26
5.1 The shape operator of M ⊂ R3 . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Normal Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.3 Gauss Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.4 Computational Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.5 The implicit Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.6 Surface of revolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6 Geometry of Surfaces in R3 36
6.1 The fundamental equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2 Isometries and local isometries . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3 Intrinsic geometry of surfaces in . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.4 Orthogonal Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.5 Integration and Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.6 Total Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.7 Congruence of surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

ii
 Chapter 0 Preliminaries

Introduction
h Vector space h Lines, planes, and spheres
h Inner product on vector space h Einstein Convention
h Linear transformation h Vector Calculus

0.1 Vector Spaces

Definition 0.1. Vector Space

A vector space V is a nonempty set with two binary operations “+” and scalar multi-
plication “·” satisfying the following eight axioms: let u, v, w ∈ V and r, s ∈ R, we
have
u + v = v + u;
u + (v + w) = (u + v) + w;
0 + u = u + 0 = u for a vector called 0;
(rs) · u = r · (s · u);
(r + s) · u = r · u + s · u;
r · (u + v) = r · u + r · v;
0 · u = 0;
1 · u = u.
♣

As is well-known, the set V satisfying the first three axioms form an Abel semi-group.
The existence of the inverse of a vector u can be verified by using the scalar multiplication.
Let u be a vector, we claim that (−1) · u is the inverse of u because

u + (−1) · u = (1 + (−1)) · u = 0.

Similarly, we can define the subtraction by

u − v = u + (−1) · v.

Note We sometimes omit the · and write, for example, ru for r · u.
External Link. Here is the video explanation of vector space (linear indepedence).
Useful!
External Link. Here is the Math 162A pre-requisite videos.

In linear algebra, we restrict ourselves to finite dimensional vector space. But, a lot of
results in finite dimensional case can be extended to infinite dimensional case as well as abstract
vector space cases.
 0.1 Vector Spaces

In the following, we give some examples of vector spaces.

Example 0.1 Rn , the n-dimensional Euclidean space, is the set of all n-vectors
   

 x 1 

n
  .  
R = x= ..  | xi ∈ R .
   
 
xn
 

Example 0.2 The set of all m × n matrices for an (mn)-dimensional vector space.
Example 0.3 The space of polynomials of degree no more than n, where n is a nonnegative
integer, is a vector space.
A single-variable polynomial of degree no more than n can be expressed as

p(t) = a0 + a1 t + · · · + an tn .

Let
q(t) = b0 + b1 t + · · · + bn tn

be another polynomial. Define the addition to be

(p + q)(t) = (a0 + b0 ) + (a1 + b1 )t + · · · + (an + bn )tn ,

and the scalar multipliction to be: let λ ∈ R

(λp)(t) = (λa0 ) + (λa1 )t + · · · + (λan )tn .

With respect to the addition and scalar multiplication, the space is a vector space.
In the above two examples, the dimensions of the vector spaces are finite. Let’s show some
examples of infinite dimensional vector spaces.

Note The set of all polynomials of degree equal to n is not a vector space.
Example 0.4 Moreover, the space of all real-valued functions on a set is a (infinite dimensional)
vector space.
Let’s recall the definition of a function.
Definition 0.2
A function f : X → Y is a triple (f, X, Y ), where X, Y are sets, and f is an assignment,
or a rule, that for any element x in X, there is a unique y = f (x) in Y attached to it.
X is called the domain of f , and Y is called the codomain of f . The range is the subset
of the codomain Y consists of all f (x) when X is running through X. The assignment
sometimes is written as x 7→ f (x). Thus a complete description of a function can be given
as
f : X → Y, x 7→ f (x).

v
♣

2
 0.2 Inner Product

As above, we can define the addition and scalar multiplication as

(f + g)(x) = f (x) + g(x), (λf )(x) = λf (x).

Remark The vector space we study in differential geometry are the “abstract” vector space,
which is on the contrary to the vector spaces we studied in Math 3A.
The following concepts are defined in abstract vector spaces similar to those in Rn .
1. linear combination, span;
2. linear dependence and independence;
3. basis and dimension.
Remark In infinite dimensional space, a basis is defined by a set of vectors

I = {v1 , v2 , · · · , vn , · · · }

satisfying the following

1. any finite subset of I is linearly independent;
2. any element can be expressed as a (finite) linear combination of element in I.
Let (V, +, ·) be a vector space. We can endow geometric structure onto it by defining the
concept of inner product.
Remark Let
S = {v1 , v2 , · · · , vn }

be a finite set. Let

V = Span S = Span {v1 , v2 , · · · , vn }.

Then we say S spans V , and S is a spanning set of V .

0.2 Inner Product

The addition and scalar multiplication define the algebraic structure of a vector space. In
order to introduce geometry to linear algebra, we can endow geometric structure onto it by
defining the concept called inner product.

Definition 0.3
An inner product on a vector space V is a function h , i

V ×V →R

satisfying the following properties: let u, v, w ∈ V and r, s ∈ R, we have

(a). hu, vi = hv, ui Symmetry
(b). hu, rv + swi = rhu, vi + shu, wi Linearity
(c).hu, ui ≥ 0 and equality is true if and only if u = 0 P ositivity
♣

Once we introduce the inner product, we introduce geometry into vector space. For example

3
 0.2 Inner Product

Definition 0.4
(Length of a vector) We can define the length, or the norm, of a vector to be
p
kuk = hu, ui.

(Distance) Let u, v be two vectors. Then their distance is defined to be

p
dist(u, v) = ku − vk = hu − v, u − vi.
♣

Note in the definition of distance, we used both the geometric structure (inner product) and
algebraic structrue (subtraction is defined by u − v = u + (−1) · v).
Example 0.5 In R3 (and in Rn ), the ordinary dot product

h(a1 , a2 , a3 ), (b1 , b2 , b3 )i = a1 b1 + a2 b2 + a3 b3

is an inner product.
However, there are many other ways one can define inner product on R3 . For example, we
can define
h(a1 , a2 , a3 ), (b1 , b2 , b3 )i = 2a1 b1 + 3a2 b2 + 4a3 b3 .

It is an inner product. In general, let

 
r11 r12 r13
 
R=
r21 r22 r23 


r31 r32 r33

be a positive definite matrix. Then
3
X
h(a1 , a2 , a3 ), (b1 , b2 , b3 )i = rij ai bj
i,j=1
is an inner product.
Example 0.6 We let R[t] be the vector space of all polynomials. Define the inner product
Z 1
hp, qi = p(x)q(x) dx
−1
More general, if ρ(x) > 0 be a positive continuous function, then
Z 1
hp, qi = p(x)q(x)ρ(x) dx
−1
is an inner product.
One of the most important inequality in mathematics is called the Cauchy-Schwarz Inequal-
ity.

Theorem 0.1
Let u, v be two vectors. Then we have

|hu, vi| ≤ kuk · kvk,

and the equality holds if and only if u, v are linearly dependent.

♥

4
 0.3 Linear Transformation

Proof. This is the standard proof. Let t be a real number. Then by the positivity of
the inner product, we have

hu + tv, u + tvi ≥ 0.

Using the linearity, we get

t2 hv, vi + 2thu, vi + hu, ui ≥ 0.

Since the above inequality is true for any real number t, the discriminant

∆ = 4|hu, vi|2 − 4hu, ui · hv, vi ≤ 0,

which proves the inequality. 

Proof. [Second Proof] Assume that v 6= 0. Then we have

hu, vi hu, vi
hu − v, u − vi ≥ 0.
hv, vi hv, vi
Expanding the above inequality, we obtain the Cauchy-Schwarz Inequality. 

Remark The Cauchy-Schwarz inequality is also called the Cauchy-Bunyakovsky-Schwarz in-

equality.
The inequality for sums was published by Augustin-Louis Cauchy (1821), while the corre-
sponding inequality for integrals was first proved by Viktor Bunyakovsky (1859). The modern
proof of the integral version was given by Hermann Schwarz (1888).

Definition 0.5
Two vectors u, v are called orthogonal, if hu, vi = 0.
♣

Example 0.7 In R3 , the vector (1, 2, 3) is orthogonal to (4, −5, 2) with respect to the dot product,
because
(1, 2, 3) · (4, −5, 2) = 1 · 4 + 2 · (−5) + 3 · 2 = 0.

Example 0.8 Under the inner product

Z 1
hp, qi = p(x)q(x) dx.
−1
The function x and x2 + 1 are orthogonal because p(x) is an odd function.
An orthonormal basis of an n-dimensional vector space is a set {v1 , · · · , vn } such that

 1, if i = j
hvi , vj i = .
 0, if i 6= j

We shall introduce the Kronecker symbol δij as follows


 1, if i = j
δij = .
 0, if i 6= j

Under this notation, we have

hvi , vj i = δij .

5
 0.3 Linear Transformation

0.3 Linear Transformation

Definition 0.6
Given two vector spaces, V and W , a linear transformation T from V to W is a mapping

T :V →W

such that
T (αu + βv) = αT (u) + βT (v)

for all u, v ∈ V and α, β ∈ R.

♣

A linear transformation of a vector space to itself is called an endomorphism. One of the

most important concept of an endomorphism is its eigenvalue and eigenvector.

Definition 0.7
Let T : V → V be an endomorphism. Assume that there is a v ∈ V and v 6= 0 such that

T (v) = λv

for some complex number λ. Then λ is called an eigenvalue of T and v is an eigenvector

of λ.
♣

Note In the following, we need to prove that the definition of eigenvalue is equivalent to the
definition of an eigenvalue of a matrix.
Let V be a finite dimensional space and let

B = {v1 , · · · , vn }

be a basis of V . We use the notation [x]B to represent the coordinates of x ∈ V , that is, when
we write x in terms of the linear combination of the basis,

x = c1 v1 + · · · + cn vn ,

we have  
c1
.
[x]B =  .
 . .
cn

Now let T : V → W be a linear transformation. As above, we assume that B is the basis

of V , and let C = {w1 , · · · , wm } be a basis of W . Then the matrix representation M of T is

M = [[T (v1 )]C , · · · , [T (vn )]C ]

in the sense that

[T (x)]C = M · [x]B .

6
 0.3 Linear Transformation

Now we specialize the above result to the following case. Let

T :V →V

be a linear transformation from V to itself. Let

B = {v1 , · · · , vn }, C = {w1 , · · · , wn }

be two bases. Let M1 , M2 be the matrix representatives with respect to the two bases respectively.
We thus have

[T (x)]B = M1 · [x]B ;
[T (x)]C = M2 · [x]C .

Let A be the invertible matrix such that

[x]B = A · [x]C .

Such a matrix is called a transition matrix. Then

[T (x)]B = A · [T (x)]C .

As a result,

A · M2 · [x]C = A · [T (x)]C = [T (x)]B = M1 · A · [x]C .

Thus we have
AM2 = M1 A,

or
M2 = A−1 M1 A.

Thus M1 , M2 are similar, having the same eigenvalue set.

Now we talk about orientation and cross product.
Let B and C be two bases. We say that B and C are having the same orientation, if when
we write
n
X
vi = aij wj
j=1

for i = 1, · · · , n, then we have det(A) = det(aij ) > 0. They give the opposite orientation if
det(aij ) < 0.
Example 0.9 The left hand and the right hand define two opposite orientations of R3 .
Example 0.10 Let
B = {(1, 0, 0), (0, 1, 0), (0, 0, 1)}

and
C = {(1, 1, 0), (1, 0, −1), (2, 1, 3)}

are of the opposite orientation.

7
 0.4 Lines, planes, and spheres

Definition 0.8
Let {e1 , e2 , e3 } be the standard basis of R3 . If
3
X 3
X
u= ai ei , v= bj ej
i=1 j=1

are two vectors in R3 , the cross product of u, v is given by

u × v = (a2 b3 − a3 b2 )e1 + (a3 b1 − a1 b3 )e2 + (a1 b2 − a2 b1 )e3 ,

or we can write  
e1 e2 e3
 
u × v = det 
a 1 a2 a 3
.

b1 b2 b3
♣

The cross product satisfies the following properties:

Lemma 0.1
Let u, v, w ∈ R3 and r ∈ R. Then
u × v = −v × u
(ru) × v = r(u × v)
u × v = 0 if and only if u and v are linearly dependent
(u + v) × w = u × w + v × w
u × v is perpendicular to both u and v under the usual dot product
ku × vk = kuk · kvk sin θ, where θ is the angle between u and v.
{u, v, u × v} gives a right hand orientation to R3 if {u, v} is linearly independent
♥

As a result, we have the relationship between the inner product and outer product (cross
product)
|hu, vi|2 + ku × vk2 = kuk2 · kvk2 ,

which implies the Cauchy-Schwarz Inequality in the three dimensional space.

Definition 0.9
The mixed (or triple) product of u, v, w is

[u, v, w] = hu × v, wi.
♣

A geometric interpretation of the norm of the cross product is that it is the area of the
parallelogram spanned by u, v. A geometric interpretation of the mixed scalar product is that it
is the volume of the parallelepiped spanned by u, v, w.

8
 0.4 Lines, planes, and spheres

0.4 Lines, planes, and spheres

In this section, we use vector notations to express some basic objects in analytic geometry.

Definition 0.10
The line through x0 ∈ R3 and parallel to a vector v 6= 0 has the equation

α(t) = x0 + tv.
♣

Remark This is a vector notation of parametrization of a line.

Example 0.11 Let x1 , x2 ∈ R3 be two points. Then the line through x1 and x2 in R3 has the
equation.
α(t) = x1 + t(x2 − x1 ).

Definition 0.11
The plane through x0 perpendicular to n 6= 0 has the equation

hx − x0 , ni = 0.
♣

Lemma 0.2
Let {u, v} be two linearly independent vectors. Then the place through x0 and parallel
to the subspace spanned by {u, v} has the equation

[u, v, x − x0 ] = hx − x0 , u × vi = 0.
♥

Definition 0.12
The sphere in R3 with center m and radius r > 0 has equation

hx − m, x − mi = kx − mk2 = r2 . (1)
♣

Remark Let    
x1 m1
   
x2  ,
x=  m=
m2  .


x3 m3

Then we get the usual equation of a sphere

(x1 − m1 )2 + (x2 − m2 )2 + (x3 − m3 )2 = r2 .

Example 0.12 (Kelvin Transformation)

We consider the equation of a sphere (1). Let x0 be a point on the sphere, that is, we have

hx0 − m, x0 − mi = r2 .

The Kelvin Transformation is a map

x − x0
K : R3 → R3 , x 7→ x0 +
kx − xo k2

9
 0.5 Vector Calculus

By a straightforward computation, we have K 2 = id. Assume

hK(x) − m, K(x) − mi = r2 .

We get
1 + 2hx − x0 , x0 − mi = 0.

So the Kelvin transformation maps a sphere to a plane.

External Link. The detailed computation can be found here.
Example 0.13 (Ptolemy Inequality) Let u, v, w, x be four vectors in the Euclidean plane. Then
we have
ku − wk · kv − xk ≤ ku − vk · kx − wk + ku − xk · kv − wk.

The equality is valid if and only if these four vectors are concyclic.
External Link. The Ptolemy Inequality is closely related to the Ptolemy Theorem. For details
of the Ptolemy and his theorem, see Wikepedia of Ptolemy Theorem

0.5 Vector Calculus

In differential geometry, in addition to study functions of several variables. We also need

to study vector-valued functions.
We can define derivatives, indefinite integral and definite integrable in similar ways to those
of multi-variable functions.
Let V, W be finite dimensional vector spaces. Let

F :V →W

be a differentiable function, with definition as follows.

Definition 0.13
We fix a basis of V and using that basis, we identify V to Rn . Similarly, and we fix a basis
of W and identify it to Rm . Then we can identify F : V → W by F : Rn → Rm . So F is
differentiable if and only if F is a differentiable as a mapping of Rn → Rm .
♣

External Link. Here is a video clip for the details of the above definition.

Let f : R → Rn be a single variable vector-valued function. We can define

df1
 
 dt 
df  . 
 
=  .. 
dt  
 df 
n
dt

10
 0.5 Vector Calculus

if  
f1
.
f = .
.
fn

Likewise, we can define Z Z b

f (t) dt, f (t) dt
a

in similar ways.
If f : R → V be an abstract vector-valued function, we can identity V with Rn under a
fixed basis, and define the derivative, integral, etc, similarly.

Lemma 0.3
Let f : R → V, g : R → V and let h , i be an inner product on V . Then if f and g are
differentiable, so is hf, gi, which is a function of one variable. Moreover, we have
d df dg
hf, gi = h , gi + hf, i.
dt dt dt
♥

Similarly, we have

Lemma 0.4
Using the notations as in the above lemma, we have
d df dg
(f × g) = ×g+f × .
dt dt dt
♥

Both of the above two lemmas can be proved directly. Moreover, we can generalize the
above results into the following.
Let V, W, S be vector spaces (probably of infinite dimensional) and let

K :V ×W →S

be a map. We say K is bilinear, if K is linear with respect to each component.

Both the inner product and cross product are bilinear mappings.
Let f : R → V , g : R → W be differentiable functions. Then the function

h(t) = K(f (t), g(t))

is differentiable, and
dh df dg
= K( , g(t)) + K(f (t), ).
dt dt dt
Definition 0.14
Let f : R → R be a continuous function. We say f is of class C k , if all derivatives up
through order k exist and are continuous.
f : Rn → R is of class C k if all its (mixed) partial derivatives of up through order k exist

11
 0.6 Einstein Convention

and are continuous. A vector-valued function is of class C k if all of its components with
respect to a given basis are of class C k .
If f is of class C k for any k, we say f is of C ∞ , or we say f is smooth.
♣

We assume most of the functions we shall study in this course are smooth, or at least of C 3 .
Finally, we review the chain rule. Let x be a function of (u1 , · · · , un ), and if each ui are
functions of (v1 , · · · , vm ), say,
ui = ui (v1 , · · · , vm ).

for i = 1, · · · , n. Then we have the chain rule

n
∂x X ∂x ∂ui
= · (2)
∂vα ∂ui ∂vα
i=1
for α = 1, · · · , m.

0.6 Einstein Convention

Definition 0.15. Einstein Convention

When an index variable appears twice in a single term and is not otherwise defined, it
implies summation of that term over all the values of the index. When an index variable
appears only once, it implies that the equation is valid for every value of such an index.
♣

For example, Equation (2) can be written as

∂x ∂x ∂ui
= · .
∂vα ∂ui ∂vα
In the above equation, the index i in the right appears twice, so we assume the expression is
summing over all possible i. On the other hand, the index α appears only once, so we assume
the equation is valid for all range of α.
Example 0.14 Let A = (aij ), B = (bij ), and C = (cij ) be matrices. Then the matrix
multiplication,
C = AB,

can be written using the Einstein Convention as

cij = aik bkj .

The Einstein Convention gives another way to express and generalize linear algebra.

Note Let’s discuss the representation of a matrix. In linear algebra, there are three ways to
represent a matrix
 
a11 · · · a1n
 . .. 
A = [a1 , · · · , an ] =  ..
 ,
. 
am1 · · · amn

12
 0.6 Einstein Convention

where a1 , · · · , an are column vectors. The easiest way to represent a matrix is to a capital letter,
say A. But this would contain the least amount of information about the matrix. On the other
extreme, if we represent a matrix by providing all the details, it would be too clumsy to write
down.
Here we give the fourth method of representing a matrix, by writing it as (aij ), which takes
care of both simplicity and information.
Example 0.15 Prove the associativity of matrix multiplication.
Proof. Let A = (aij ), B = (bij ) be matrices. Let D = (dij ) be the matrix D = AB.
In terms of the Einstein Convention, D = AB is equivalent to

dij = aik bkj . (3)

Here the index k is called a dummy index in the sense that we can replace it with other
indices without changing the equations:

dij = aik bkj = ait btj = aiα bαj . (4)

Now let C = (cij ), E = BC = (eij ), F = (AB)C = (fij ) and G = A(BC) = (gij ).

Then the entries for (AB)C = DC would be

fij = dik ckj = dit ctj .

From (2.1), we know that dit = aik bkt . Thus

fij = aik bkt ctj .

The reason we use t as the dummy index in (2.1) is because k has been used in (3) so
we need to use a different one, keeping indices repeated at most twice.
Similarly, we have
gij = ait btk ckj .

Thus fij = gij and hence

(AB)C = A(BC),

proving the associativity. 

Example 0.16 Using the Einstein Convention to prove the following version of the Cauchy
inequality. Let x = (x1 , · · · , xn ) and y = (y1 , · · · , yn ). Then

|x · y|2 ≤ kxk2 · kyk2 .

Proof. Using the Einstein Convention, we can write
Xn
x·y = xi yi = xi yi .
i=1
Thus
n
!2 n
!  n 
X X X
|x · y|2 = x i yi = x i yi ·  xj yj  = xi yi xj yj .
i=1 i=1 j=1

13
 0.6 Einstein Convention

On the other hands, we can write

n
!  n

X X
kxk2 · kyk2 = x2i · yj2  = x2i yj2 .
i=1 j=1

Thus the Cauchy inequality, written under the Einstein Convention, is

1 1
x2i yj2 − xi yi xj yj = (x2i yj2 + x2j yi2 ) − xi yi xj yj = (xi yj − xj yi )2 ≥ 0.
2 2
This completes the proof.


Note If n = 3, then we can define

(x, y) 7→ (x2 y3 − x2 y2 , x3 y1 − x1 y3 , x1 y2 − x2 y1 ) = x × y

which is the cross product. If n 6= 3, then the vector (xi yj − xj yi ) for i < j is of dimension
n(n − 1)/2 6= 3. This explains why we can only define the cross product in 3 dimensional vector
space. A more general algebraic product, called wedge product, will be used in any dimensional
vector spaces to catch in the excess of the Cauchy inequality.
External Link. As fun reading, you can find the Shoelace Formula in the Wikipedia, which is
related to both the cross product and wedge product.

14
 Chapter 1 Calculus on Euclidean Space

Introduction
h Calculus on Euclidean space h The wedge product
h Tangent Vectors and tangent space h The differential operator d
h The space of differential forms h The differential of a mapping

1.1 Euclidean Space

Definition 1.1
Euclidean 3-space R3 is the set of all ordered triples of real numbers. Such a triple
p = (p1 , p2 , p3 ) is called a point of R3 .
♣

By last chapter, R3 is a vector space.

Definition 1.2
On R3 , there are three natural real-valued functions x, y, z, defined by

x(p) = p1 , y(p) = p2 , z(p) = p3 .

These functions are called natural coordinate functions of R3 .

♣

Remark We shall also use index notation for these functions, writing

x1 = x, x2 = y, x3 = z.

Definition 1.3
A real-valued function f on R3 is differentiable (or infinitely differentiable, or smooth, or
of class C ∞ ) provided all partial derivatives of f , of all orders, exist and are continuous.
♣

As we know from the previous chapter, the space of smooth functions forms a vector space,
that is, let f, g be two smooth functions of R3 and let λ ∈ R, we have

(f + g)(p) = f (p) + g(p), (λf )(p) = λf (p).

In addition, we have
(f g)(p) = f (p)g(p).

The space of smooth functions, with respect to the three operations: addition, scalar multiplica-
tion, and the multiplication forms an algebra.
 1.2 Tangent Vectors

1.2 Tangent Vectors

Definition 1.4
A tangent vector, or a vector vp to R3 consists of two points of R3 : its vector part v and
its point of application p.
♣

Definition 1.5
Let p be a point of R3 . The set Tp (R3 ) consisting of all tangent vectors that have p as
point of application is called the tangent space of R3 at p.
♣

Note Tangent space is a vector space.

Definition 1.6
A tangent vector field, or a vector field V on R3 is a function that assigns to each point p
of R3 a tangent vector V (p) to R3 at p.
♣

Note Vector field is one of the most important concepts in differential geometry. By the above
definition, a vector field is just a vector valued function. This is because R3 is a flat space, and
hence there are global basis under which all tangent spaces can be identified as R3 . In general,
a vector field defines a different type of “functions” comparing to the traditional one.
Remark By definition, a vector field doesn’t have to be smooth. However, in this course, we
always assume it is smooth (or at least of C 3 ) when regarding it as a vector-valued function.
The domain of a vector field doesn’t have to be on the whole R3 : it could be an open set of
R3 , or a curve or a surface in R3 . In the latter to cases, we say that the vector field is along the
curve or surface.
The space of vector fields is obviously a vector space. However, it has finer structure than
that. It is a module over the algebra of smooth functions.
There are two operations on the space of vector fields: addition and scalar multiplication.
Let V, W be two vector fields such that V = vi Ui , W = wi Ui . Let λ ∈ R. Then we can define
X
V +W = (vi + wi )Ui
i
X
λV = (λvi )Ui .

Moreover, let f be a smooth function of R3 . Then we can define

(f V )(p) = f (p)V (p)

for all p. Of course, such kind of multiplication can be localized to the case when V and f are
only defined on a subset of R3 .

Note The scalar multiplication coincides with the above multiplication by regarding a scalar as
a constant function.

16
 1.3 Directional Derivatives

Note We can also say that the space of vector fields is a module over the ring of smooth
functions. But the algebra of smooth functions carries more structure than the ring structure of
smooth functions: it has the additional scalar multiple structure. Therefore it is better to say the
module over the algebra of smooth functions than that over the ring of smooth functions.

Definition 1.7
Let U1 , U2 and U3 be the vector fields on R3 such that

U1 (p) = (1, 0, 0)p

U2 (p) = (0, 1, 0)p
U3 (p) = (0, 0, 1)p

for each p of R3 . We call {U1 , U2 , U3 } the natural frame field of R3 .

♣

Remark For fixed point, {U1 , U2 , U3 } provides the standard basis of R3 , usually expressed as
{e1 , e2 , e3 }.
The following result is a generalization of what we have learned in linear algebra.

Lemma 1.1
If V is a vector field of R3 , then there are three uniquely determined real-valued functions
v1 , v2 , v3 on R3 such that
V = v1 U1 + v2 U2 + v3 U3 .

These three functions are called Euclidean coordinate functions of V .

♥

Proof. For fixed p ∈ R3 , V (p) defines a vector in R3 , therefore there is unique

numbers v1 (p), v2 (p), and v3 (p) such that

V (p) = v1 (p)U1 (p) + v2 (p)U2 (p) + v3 (p)U3 (p).

Thus
V = vi Ui

by definition. 

1.3 Directional Derivatives

Definition 1.8
Let f be a differentiable real-valued function on R3 , and let vp ∈ Tp (R3 ) be a tangent
vector to R3 . Then the number
d
vp [f ] = (f (p + tv))|t=0
dt

17
 1.3 Directional Derivatives

is called the derivative of f with respect to vp

♣

Note It is called the directional derivative because p + tv for non-negative real number t
represents a ray starting from p in the direction v. We have encountered directional derivative
in Calculus. Here the emphasis is that “vector” (which is an algebraic concept) can be identified
as a “derivative” (which is a calculus concept).

Lemma 1.2
If vp = (v1 , v2 , v3 ) is a tangent vector to R3 , then
X ∂f
vp [f ] = vi (p).
∂xi
♥

Proof. Let p = (p1 , p2 , p3 ); then

p + tv = (p1 + tv1 , p2 + tv2 , p3 + tv3 ).

We then
d X ∂f
vp [f ] = (f (p + tv))|t=0 = (p)vi .
dt ∂xi
i

Example 1.1 Let f (x, y, z) = x2 yz. Let p = (1, 1, 0) and v = (1, 0, −3). Then
∂f ∂f ∂f
= 2xyz, = x2 z, = x2 y.
∂x ∂y ∂z
Thus
∂f ∂f ∂f
(p) = 0, (p) = 0, (p) = 1.
∂x ∂y ∂z
Thenfore
vp [f ] = 0 + 0 + 1 · (−3) = −3.

Theorem 1.1
Let f and g be functions on R3 , vp and wp tangent vectors, a and b numbers. Then
(avp + bwp )[f ] = avp [f ] + bwp [f ];
vp (af + bg) = avp [f ] + bvp [g];
vp [f g] = vp [f ] · g(p) + f (p) · vp [g].
♥
Proof. Only the 3rd equation is new, which can be proved using Lemma 1.2: We
have
∂(f g) ∂g ∂f
vp [f g] = vi (p) = vi f (p) (p) + vi g(p) (p).
∂xi ∂xi ∂xi
By definition, we have

vp [f g] = vp [f ] · g(p) + f (p) · vp [g].

18
 1.4 Curves in R3

1.4 Curves in R3

One of the fundamental questions in curve theory is: how to define a curve? In Euclidean
geometry, only two kinds of curves are studied: straight line and circle. In analytic geometry,
we study parabola, ellipse, and hyperbola. These curves have quite explicit geometric meanings.
For example, an ellipse is the the set of all points in a plane such that the sum of the distances
from two fixed points (foci) is constant. If we want to study more general curves, we should not
expect them have clear geometric meanings.
In differential geometry, we define a curve in R3 by a function

α : I → R3 , α(t) = (α1 (t), α2 (t), α3 (t)),

where I is an open interval. In order to use calculus, we usually assume that such a function is
smooth.
Definition 1.9
A curve in R3 is a differentiable function α : I → R3 from an open interval into R3 .
♣

We shall give a couple of examples of curves.

Example 1.2 (Straight Line) A straight line can be expressed best using the vector notations. Let
p, q be two vectors and let q 6= 0. Then we can use

α(t) = p + tq

to represent a curve with direction q.

Example 1.3 (Helix) The parameter equations for a circle (in R3 ) can be expressed by

t 7→ (a cos t, a sin t, 0).

If we allow this curve to rise , then we obtain a helix α : R → R3 , given by the formula

α(t) = (a cos t, a sin t, bt),

where a > 0 and b 6= 0.

19
 1.4 Curves in R3

Definition 1.10
Let α : I → R3 be a curve in R3 with α = (α1 , α2 , α3 ). For each number t ∈ I, the
velocity vector of α at t is the tangent vector
 
dα1 dα2 dα3
α0 (t) = (t), (t), (t)
dt dt dt α(t)

at the point α(t) in R3 .

;
♣

Example 1.4 For the helix,

α(t) = (a cos t, a sin t, bt),

the velocity vector is

α0 (t) = (−a sin t, a cos t, b)α(t) .

Definition 1.11
Let α : I → R3 be a curve. If h : J → I is a differentiable function on an open interval
J, then the composition function

β = α(h) : J → R3

is a curve called a reparametrization of α by h.

♣

Note The above definition is a key concept. See the next lemma.

Lemma 1.3
If β is the reparametrization of α by h, then
df
β 0 (s) = · α0 (h(s)).
ds
♥
Proof. This is a straightforward application of the chain rule.


Lemma 1.4
Let α be a curve in R3 and let f be a differentiable function on R3 . Then
d(f (α))
α0 (t)[f ] = (t).
dt
♥

20
 1.4 Curves in R3

We end up this section by a discussion of dual space.

Dual Space

Definition 1.12
Given any vector space V , the dual space V ∗ is defined as the set of all linear transforma-
tions ϕ : V → R. The dual space is a vector space by the following definition of addition
and scalar multiplication.

(ϕ + ψ)(x) = ϕ(x) + ψ(x)

(aϕ)(x) = a(ϕ(x))

for all ϕ, ψ ∈ V ∗ and a ∈ R, x ∈ V . Elements of V ∗ is called a covector, or linear

functional, or a one-form.
♣

Example 1.5 On Rn , any linear function

`(x) = c1 x1 + · · · + cn xn ,

where x = (x1 , · · · , xn ) and c1 , · · · , cn being real numbers, is a liner functional.

Example 1.6 Let C([0, 1]) be the vector space of continuous functions over [0, 1]. Then
Z 1
f 7→ f (x)dx
0
defines a linear functional.
Example 1.7 Let p ∈ R3 . Let vp be a vector on Tp (R3 ). Then the directional derivative

f 7→ vp [f ]

is a linear functional on the vector space of differentiable functions.

External Link. Here is a good video of the dual space. The first 8 minutes is useful, and the
last part is beyond the scope of this course.

Theorem 1.2. (The Riesz Representation Theorem)

Let V be a finite dimensional vector space and let h , i be an inner product of V . Let x ∈ V .
Then it defines a linear functional `x such that `x (y) = hx, yi for any y ∈ V ; conversely,
let ` be a linear functional, then there is a unique x ∈ V such that `(y) = hx, yi for any
y ∈V.
♥

Remark In order words, every linear functional can be represented, through a fixed inner product,
as an element of the vector space.
Remark We elaborate the Riesz Representation Theorem in the context of the vector space Rn
with the dot product. Let x ∈ Rn , and let ` be a linear functional. By the above theorem, there
is a vector c such that
`(x) = c · x = c1 x1 + · · · + cn xn .

21
 1.5 1-forms

In this way, the dual space of Rn can be identified to Rn .


Note There is an infinite dimensional version of the Riesz Representation Theorem on normed
vector space, but the linear functional in question needs to be replaced by bounded linear
functional.
External Link. The linear algebra over infinite dimensional vector spaces is called Functional
Analysis.

1.5 1-forms

Definition 1.13
A 1-form φ on R3 is a real-valued function on the set of all tangent vectors to R3 such that
φ is linear at each point, that is,

φ(av + bw) = aφ(v) + bφ(w)

for any number a, b and tangent vectors v, w at the same point of R3 .a

a
Recall in Definition 1.4, a vector on R3 is a pair (p, v), where p ∈ R3 and v is the vector part.
♣

As before, the space of 1-forms is a module over the algebra of smooth functions. Let ϕ, ψ
be two 1-forms; let v be a vector on R3 ; let λ ∈ R. Then we can define

(ϕ + ψ)(v) = ϕ(v) + ψ(v); (λϕ)(v) = λϕ(v).

Note ϕ(v) is a smooth function on R3 . Let f be a smooth function. Let vp be the vector part of
v.
(f ϕ)(vp ) = f (p)ϕ(vp ).

In fact, there is a natural way to extend a 1-from as a function over vector fields. A 1-form
is a linear functional in two ways: first, it is a linear functional over the vector space of vector
fields, that is, if ϕ is a 1-form, for any vector field v, (ϕ(v))(p) = ϕ(vp ) is a smooth function;
second, for any fixed point p, ϕ is a linear functional over Tp (R3 ).

Definition 1.14
If f is a differentiable function on R3 . Then df is a 1-form defined by

df (vp ) = vp [f ].
♣

Example 1.8 1-forms on R3 : by the above definition, we can define 1-froms dx1 , dx2 , dx3 . Let

vp = vi Ui .

Then by definition,
dxi [vp ] = vp [xi ] = vi .

22
 1.5 1-forms

Let’s consider the 1-form

ψ = fi dxi ,

where fi are functions. Then

ψ[vp ] = fi dxi [vp ] = fi (p)vi .

Definition 1.15. Dual Basis

Let V be a vector space and let (e1 , · · · , en ) be a basis of V . (f1 , · · · , fn ) is called the
dual basis of (e1 , · · · , en ), if fi ∈ V ∗ , and if

fi (ej ) = δij .
♣

Theorem 1.3. Dual Basis Theorem

Using the above notations, then (dx1 , dx2 , dx3 ) is the dual basis of (U1 , U2 , U3 ).
♥
Proof. We have
∂xi
dxi (Uj ) = Uj [xi ] = = δij .
∂xj


Corollary 1.1

If f is a differentiable function on R3 , then

∂f
df = dxi = fi dxi .
∂xi
♥
 ∂f
Note As we have seen, we use fi to represent ∂xi . In differential geometry, this would greatly
∂f
simplify complicated computations. In general, whether fi is ∂xi or an arbitrary function
depends on the context.

Proof. By definition, df [vp ] = vp [f ]. But vp [f ] = vi fi (p) = (fi dxi )[vp ].



From the definition of df , we observed that we can regard d as an operator, that would send
a function f to a 1-form df . Such an operator is called a differential operator, which plays one
of the center role in differential geometry.

Lemma 1.5
Let f g be the product of differentiable functions f and g on R3 . Then

d(f g) = gdf + f dg.

♥

23
 1.6 Differential Forms

Proof. We have

d(f g) = (f g)i dxi = (gfi + f gi )dxi = gfi dxi + f gi dxi = g df + f dg.

Lemma 1.6
Let f be a function on R3 and let h : R → R be a function of single variable. Then

d(h(f )) = h0 (f )df.
♥
Proof.
d(h(f )) = (h(f ))i dxi = h0 (f )fi dxi = h0 (f )df.

Example 1.9 Let f be the function

f (x, y) = (x2 − 1)y + (y 2 + 2)z.

Then
df = 2xydx + (x2 + 2yz − 1)dy + (y 2 + 2)dz.

As a result, we have

df [v] = 2xyv1 + (x2 + 2yz − 1)v2 + (y 2 + 2)v3 .

Thus
df [vp ] = 2p1 p2 v1 + (p21 + 2p2 p3 − 1)v2 + (p23 + 2)v3 .

We also have
vp [f ] = 2p1 p2 v1 + (p21 + 2p2 p3 − 1)v2 + (p23 + 2)v3 .

This verifies df [vp ] = vp [f ].

1.6 Differential Forms

The space of differential 1-forms is a vector space, or more precisely, it is a module over
the algebra of smooth functions. To get more information from tangent spaces where the space
of differential 1-forms are dual spaces of them at each point, we shall define multiplication of
differential 1-forms. Since all differential 1-forms are generated by dx1 , dx2 , dx3 , we just need
to define their multiplications.
What is dxi dxj , or we called the wedge product dxi ∧ dxj of them? We don’t know at this
moment. But we shall assume that

dxi dxj = −dxj dxi

24
 1.6 Differential Forms

for 1 ≤ i, j, ≤ 3. Obviously, this would create a new kind of algebra. For multiplication of real
numbers, we have commutativity, which means, for any two real numbers a, b, we have ab = ba.
On the other hand, for two n × n matrics A, B, in general, we have AB 6= BA. The property
for the multiplication of 1-forms are different from both of the above two. It is called skew
commutativity. The algebra defined by the skew commutativity leads to the so-called exterior
algebra.
A first observation on the definition of the wedge product reveals that, since dxi ∧ dxi =
−dxi ∧ dxi , we must have dxi ∧ dxi = 0. A quick counting shows that the only non-zero
independent products would be dx1 dx2 , dx1 dx3 and dx2 dx3 .
In general, we can define the whole system of p-forms. we have already encountered 0-
forms, which are smooth functions, and 1-forms. Taking multiplication of dxi with dxj , we
can define the space of two forms to be generated by dx1 dx2 , dx1 dx3 and dx2 dx3 over smooth
functions, that is, all two forms can be expressed by

f dx1 dx2 + gdx1 dx3 + hdx2 dx3 ,

where f, g, h are functions.

We can define the 3-forms in an obvious way: all three forms have the expressions

f dx1 dx2 dx3 ,

where f is a function.
In the high dimensional case, we can define the p-forms for p > 3. However, on R3 ,
all higher differential forms would be zero: consider, for example, a 4-form dxi dxj dxk dxl .
Since the space is of 3 dimensional, at least two of the indices must be the same. By skew
commutativity, all 4-forms must be zero.
Example 1.10 Compute the Wedge products
(1). Let
φ = xdx − ydy, ψ = zdx + xdz.

Then

φ ∧ ψ = (xdx − ydy) ∧ (zdx + xdz)

= xzdxdx + x2 dxdz − yzdydx − xydydz
= yzdxdy + x2 dxdz − xydydz.

(2). Let θ = zdy. Then

θ ∧ φ ∧ ψ = −x2 z dxdydz.

(3). Let η = ydxdz + xdydz. Then

φ ∧ η = (xdx − ydy) ∧ (ydxdz + xdydz)

= (x2 + y 2 ) dxdydz.

Note It should be clear from these examples that the wedge product of a p-form and a q-form is

25
 1.6 Differential Forms

a (p + q)-form. Thus such a product is automatically zero whenever p + q > 3.

Lemma 1.7
Let φ, ψ be 1-forms. Then
φ ∧ ψ = −ψ ∧ φ.
♥
Proof. Let φ = fi dxi , ψ = gi dxi . Then

φ ∧ ψ = fi dxi gj dxj = fi gj dxi dxj = −fi gj dxj dxi = −ψ ∧ φ.

Remark The space of any p-forms forms a module over smooth functions1. However, given that
a p-form wedge a q-form to be a (p + q)-form, we can take the direct sum of the modules of all
p-forms. Obviously, this would give us a module over functions where the wedge product is well
defined.
In what follows we will define arguably the most important concept in differential geometry.

Definition 1.16
If φ = fi dxi is a 1-form on R3 . The exterior derivative, or differential, of φ is the 2-form

dφ = dfi ∧ dxi .
♣

Example 1.11 If
φ = fi dxi = f1 dx1 + f2 dx2 + f3 dx3 ,

then we have
     
∂f2 ∂f1 ∂f3 ∂f1 ∂f3 ∂f2
dφ = − dx1 dx2 + − dx1 dx3 + − dx2 dx3 .
∂x1 ∂x2 ∂x1 ∂x3 ∂x2 ∂x3
Thus if we identify φ to a vector-valued function E = (f, f2 , f3 ), then dφ can be identified as
curl(E). In this sense, dφ generalize the curl operator.

Theorem 1.4
Let f, g be functions and φ, ψ be 1-forms. Then
1. d(f g) = df g + f dg;
2. d(f φ) = df ∧ φ + f dφ;
3. d(φ ∧ ψ) = dφ ∧ ψ − φ ∧ dψ a .
a
This is more a definition than a property of the differential operator.
♥
Proof. Property (1) is just the product rule We proved in Lemma 1.5. To prove (2),
we let φ = fi dxi . Then

d(f φ) = d(f fi ) ∧ dxi = df ∧ fi dxi + f dfi ∧ dxi = df ∧ φ + f dφ.

1For any p, even if p > 3 or p < 0, where we defined the module to be 0.

26
 1.7 Mappings

Property (3) is, straightly speaking, a definition rather than a property, since we have
never defined the differential of 2-forms before. Nevertheless, let’s work on it. First,

d(φ ∧ ψ) = d(fi gj dxi dxj ).

As in the case of 1-forms, we define

d(fi gj dxi dxj ) = d(fi gj ) ∧ dxi ∧ xj .

We then have

d(φ ∧ ψ) = dfi gj ∧ dxi ∧ dxj + fi dgj ∧ dxi ∧ dxj

= dfi ∧ dxi ∧ gj dxj − fi ∧ dxi ∧ dgj ∧ dxj
= dφ ∧ ψ − φ ∧ dψ.

Example 1.12 Let

φ = f1 dx2 dx3 + f2 dx3 dx1 + f3 dx1 dx2

be a 2-form. Then  
∂f1 ∂f2 ∂f3
dφ = + + dx1 dx2 dx3 .
∂x1 ∂x2 ∂x3
Thus if we identify E = (f1 , f2 , f3 ). Then dφ can be identify to div(E).
Example 1.13 Let f be a function, then
∂f ∂f ∂f
df = dx1 + dx2 + dx3
∂x1 ∂x2 ∂x3
can be identified as ∇f .
Example 1.14 If we identify φ = fi dxi to u and ψ = gi dxi to v, then φ ∧ ψ can be identified to
u × v.
 Exercise 1.1 Can you use exterior algebra to define the dot product?

1.7 Mappings

In this section we discuss functions from Rn to Rm . If n = 3 and m = 1, this is just a

function on R3 . In the other extreme, if n = 1 and m = 3, then this is a single variable R3 -valued
function, and by the previous sections, they can be used to represent curves in R3 . All of these
functions have been studied in Calculus, but in this section, we shall study them using the idea
of linearization.
Recall that a linear transformation from Rn to Rm is a linear function from Rn to Rm .
Definition 1.17
Given a function F : Rn → Rm , let f1 , · · · , fm denote the real-valued function on Rn

27
 1.7 Mappings

such that
F (p) = (f1 (p), f2 (p), · · · , fm (p))

for all points p ∈ Rn . These functions are called the Euclidean coordinate functions of
F , and we can write F = (f1 , · · · , fm ).
The functions F is differentiable provided its coordinate functions are differentiable in the
usual sense. A differentiable function F : Rn → Rm is called a mapping from Rn to Rm .
♣

Definition 1.18
If α : I → Rn is a curve in Rn and F : Rn → Rm is a mapping, then the composition
function β = F (α) : I → Rm is a curve in Rm called the image of α under F .
♣

Note Let B be the set of all curves in Rn and let C be the set of all curves in Rm . Then a
mapping F : Rn → Rm induces a map B → C .

Definition 1.19
Let F : Rn → Rm be a mapping. If va is tangent vector to Rn at p. Let F∗ (v) be the
initial velocity of the curve t 7→ F (p + tv). The resulting function F∗ sends a tangent
vectors to Rn to tangent vectors to Rm , and is called the tangent map of F .
a
If v = 0, the straight line p + tv is degenerated to a point p. But the definition is still valid.
♣

Proposition 1.1
Let F = (f1 , · · · , fm ) be a mapping from Rn to Rm . If v is a tangent vector to Rn at p,
then
F∗ (v) = (v[f1 ], · · · , v[fm ])

at F (p).
♠
Proof. We take m = 3 for simplicity. By definition, the curve t 7→ F (p + tv) can
be written as

β(t) = F (p + tv) = (f1 (p + tv), f2 (p + tv), f3 (p + tv)).

By definition, we have F∗ (v) = β 0 (0). To get β 0 (0), we take the derivatives, at t = 0,

of the coordinate functions of β. But
d
(fi (p + tv))|t=0 = v[fi ].
dt
Thus
F∗ (v) = (v[f1 ], v[f2 ], v[f3 ])|β(0) ,

where β(0) = F (p).

28
 1.7 Mappings

Let p ∈ Rn . Then we have the linear transformation

F∗p : Tp (Rn ) → TF (p) (Rm )

called the tangent map of F at p.

Corollary 1.2
If F : Rn → Rm is a mapping, then at each point p of Rn , the tangent map F∗p :
Tp (Rn ) → TF (p) (Rm ) is a linear transformation.
♥

Note For any nonlinear function F , we can define a semi-linear function F∗p , where for fixed p,
the function is a linear transformation. But the function F∗ : Rn × Rn → Rm , (p, v) 7→ F∗p (v)
is nonlinear with respect to p.

Note Let f (t) be a function of single variable. Then f∗ : R × R → R, (t, s) 7→ sf 0 (t) is the
tangent map of f . Such a tangent map can be identified to the derivative f 0 (t).

Corollary 1.3
Let F : Rn → Rm be a mapping. If β = F (α) is the image of a curve α in Rn , then
β 0 = F∗ (α0 ).
♥

Corollary 1.4
If F = (f1 , · · · , fm ) is a mapping from Rn to Rm , then
m
X ∂fi
F∗ (Uj (p)) = (p)Ūi (F (p)),
∂xj
i=1
where {Ūi }, for i = 1, · · · , m are natural frame fields of Rm .
♥

Definition 1.20
The matrix  
∂fi
J= (p)
∂xj 1≤i≤m,1≤j≤n

is called the Jacobian matrix of F at p.

♣

In terms of matrix notations, we have

F∗ [U1 (p), · · · , Un (p)] = [Ū1 (F (p)), · · · , Ūm (F (p))] · J.

Definition 1.21
A mapping F : Rn → Rm is regular provided that at every point p of Rn the tangent map
F∗p is one-to-one.
♣

Remark By linear algebra, the following are equivalent

29
 1.7 Mappings

(1) F∗p is one-to-one.

(2) F∗p (vp ) = 0 implies vp = 0.
(3) The Jacobian matrix of F at p has rank n, the dimension of the domain Rn of F .
Remark If m = n, then we know that, by the Invertible Matrix Theorem, that F∗p is one-to-one
if and only if it is onto.

Definition 1.22
Let U , V be two open sets of Rn . We say that U and V are diffeomorphic, if there is
a differentiable map F : U → V which is one-to-one and onto. Moreover, the inverse
mapping: F −1 : U → V is also differentiable. We also say that F is a diffeomorphism
of U to V .
♣

Theorem 1.5. (Inverse Function Theorem)

Let F : Rn → Rm be a mapping between Euclidean spaces of the same dimension. If
F∗p is one-to-one at a point p, there is an open set U containing p such that F restricted
to U is a diffeomorphism of U onto an open set V .
♥

External Link. In the video here, I further elaborate the Chain Rule using the tangent map.

30
 Chapter 2 Frame Fields

Introduction
h The dot product revisited h The Frenet formulas

2.1 Dot Product

We have discussed inner product in Chapter 0. The dot product is a special case of inner
product. So we shall very quick go through it.

Definition 2.1
Let p = (p1 , p2 , p3 ) and q = (q1 , q2 , q3 ) in R3 . The dot product is defined by

p · q = p 1 q1 + p 2 q2 + p 3 q3 .

The norm of a point p is defined by

√
kpk = p · p.

The Euclidean distance from p to q is the number

d(p, q) = kp − qk.

Vectors are called orthogonal , if

p · q = 0.

More generally, the angle θ between vectors p, q is defined by the equation

p · q = kpk · kqk · cos θ.

A vector p is called a unit vector, if kpk = 1.

♣

Definition 2.2
A set {e1 , e2 , e3 } of three mutually orthogonal unit vectors tangent to R3 at p is called a
framea at the point p.
a
it is also called an orthonormal basis.
♣

Thus {e1 , e2 , e3 } is a frame if and only if

e1 · e1 = e2 · e2 = e3 · e3 = 1;
e1 · e2 = e1 · e3 = e2 · e3 = 0.

Using the Einstein Convention, we have ei · ej = δij .

 2.2 Curves

Definition 2.3
Let e1 , e2 , e3 be a frame at a point p of R3 . The 3 × 3 matrix A whose rows are the
Euclidean coordinates of these three vectors is called the attitude matrix of the frame.
♣

Explicitly, if

e1 = (a11 , a12 , a13 )p ,

e2 = (a21 , a22 , a23 )p ,
e3 = (a31 , a32 , a33 )p ,

then  
a11 a12 a13
 
A=
a 21 a22 a 23


a31 a32 a33

We shall prove that A is an orthogonal matrix: we have

δij = ei · ej = aik ajk .

The matrix expression of the above equation is

AAT = I,

where AT is the transpose of A.


Note If AAT = I, then AT = A−1 . Therefore

AT A = A−1 A = I.

Note that this is not a trivial result. Take a 2 × 2 matrix, for example, Let
 
a b
A= .
c d
Then AAT = I is equivalent to

a2 + b2 = 1, ac + bd = 0, c2 + d2 = 1,

and
AT A = I

is equivalent to
a2 + c2 = 1, ab + cd = 0, b2 + d2 = 1.

Can you find an elementary proof of the above fact? Hint: we can prove the identity

(a2 + c2 − 1)2 + (b2 + d2 − 1)2 + 2(ab + cd)2

= (a2 + b2 − 1)2 + (c2 + d2 − 1)2 + 2(ac + bd)2 .

32
 2.2 Curves

2.2 Curves

Let α(t) = (α1 (t), α2 (t), α3 (t)) be a curve. It is well known that

α0 (t) = (α10 (t), α20 (t), α30 (t))

is the velocity of the curve. The velocity is a vector field. Its norm, kα0 (t)k, is called the speed
of the curve. The speed of a curve is a function along the curve.
In terms of the components, we can write
 2  2  2 !1/2
dα 1 dα 2 dα 3
ν = kα0 k = + + .
dt dt dt
The length of the curve, from t = a to t = b, can be expressed by
Z b
kα0 (t)k dt.
a

Definition 2.4
A curve α : I → R3 is called regular, if the velocity vector field α0 (t) 6= 0 for any t ∈ I.
This is equivalently to say that the speed function ν = kα0 (t)k is not zero at any point
t ∈ I. If kα0 (t)k = 1, then the curve is called a unit speed curve.
♣

Definition 2.5
A reparametrization is a map t : [c, d] → [a, b] such that the function is one-to-one and
onto, and that t0 (s) 6= 0 for any s ∈ [c, d].
♣

Proposition 2.1
Let t(s) be a reparametrization. Then we have the following two cases:
(1) t0 (s) > 0 and t(c) = a, t(d) = b;
(2) t0 (s) < 0 and t(c) = b, t(d) = a.
♠

Proof. Since t0 (s) 6= 0, we must have either t0 (s) > 0 or t0 (s) < 0. In the first case,
t(s) is monotonically increasing. Thus t(c) must be minimal and hence equal to a,
and t(d) must be maximum, hence equal to b. This proves the first case.


Theorem 2.1
The length of a curve is an invariant.
♥

Proof. Let t = t(s) be a reparametrization, that is t : [c, d] → [a, b] such that

the function t is one-to-one and onto and we assume that t0 (s) 6= 0. Let s = s(t) :
[a, b] → [c, d] be the inverse function. Let β(s) = α(t(s)). Then the length of the

33
 2.2 Curves

curve β, from c to d, is
Z d
kβ 0 (s)kds.
c

By the chain rule, β 0 (s)

= t0 (s)α0 (t(s)). We then have
Z d Z d Z b
0 0 0
kβ (s)kds = |t (s)| · kα (t(s))kds = kα0 (t)k dt.
c c a


Theorem 2.2
If α is a regular curve, then there is a parametrization β such that β has unit speed.
♥
Proof. Let t = t(s) be a reparametrization and let β(s) = α(t(s)). The requirement
is that
1 = kb0 (s)k = |t0 (s)| · kα0 (t(s))k

for any s. Thus in order to find the unit speed parametrization, we need to solve the
differential equation
1
t0 (s) =
kα0 (t(s))k
with the initial value t(0) = a. This is a separable equation. We write s0 (t) = ds/dt.
Let s = s(t) be the inverse function of t(s). Then
Z t
s(t) = kα0 (u)kdu
a
defines the unit speed reparametrization.



Note Note that s(a) = 0, and s(b) is the length of the curve from a to b.
Example 2.1 Consider the helix

α(t) = (a cos t, a sin t, bt).

The velocity of α is
α0 (t) = (−a sin t, a cos t, b).

The speed of α is
p p
kα0 (t)k = a2 sin2 t + a2 cos t2 + b2 = a2 + b2 .
√
Therefore α has constant speed c = a2 + b2 .
Let Z t
s(t) = c du = ct.
0

Then t = s/c. We thus have

 
s s bs
β(s) = α(s/c) = a cos , a sin ,
c c c
is the unit speed reparametrization.

34
 2.3 The Frenet Formulas

Definition 2.6
Let Y = yi Ui be a vector field along a curve α. Thus each function yi can be expressed
as a function of t via α. The differentiation of Y is simply differentiation on its Euclidean
coordinate functions.
♣

Example 2.2 Let

Y (t) = t2 U1 − tU3 .

Then
Y 0 (t) = 2tU1 − U3 .

In particular, we can define the acceleration α00 (t) of the curve α(t).

Lemma 2.1
(1) A curve α is constant if and only if its velocity is zero, α0 = 0.
(2) A non-constant curve α is a straight line if and only if its acceleration is zero,
α00 = 0.
(3) A vector field Y on a curve is parallel if and only if its derivative is zero, Y 0 = 0a .
a
This is the definition of the parallelism rather than a statment.
♥

2.3 The Frenet Formulas

Let β : I → R3 be a unit-speed curve. Let T = β 0 (s) be the velocity vector field. Then we
have kT k = 1. We consider T 0 = β 00 (s). Since kT k = 1, we have T · T = 1. Taking derivative1
on both sides, we have T 0 · T + T · T 0 = 0. Thus T · T 0 = 0, and T 0 is always orthogonal to T .

Definition 2.7
The curvature κ(s) = kT 0 (s)k = kβ 00 (s)k.

By the above definition, we know that k(s) ≥ 0. In order to introduce the Frenet Formulas,
we further assume that κ > 0.

1This technique will be used repeatedly throughout the rest of the book.

35
 2.3 The Frenet Formulas

When κ > 0, we have N = T 0 /κ. By definition, we have kN k = 1. By the similar

argument as above, we have N · N 0 = 0.

Definition 2.8
Assume that κ(s) > 0. Define N = T 0 /κ, and B = T × N . Then (T, N, B) is an
orthonormal basis of the tangent space at point β(s). We call (T, N, B) the Frenet
frame field, or TNB frame field on β. The collection {T, N, B, κ, τ } is called the Frenet
Apparatus.
♣

Remark If κ ≡ 0, then the curve is a straight line. On the other hand, if κ is nowhere zero, then
we are able to define the Frenet frame field. It is beyond the scope of this book to discuss curves
with vanishing curvature at isolated points.
By definition, we have kN k = 1. By the similar argument as above, we have N · N 0 = 0.
Thus
Definition 2.9
We can define
τ = N 0 · B,

where τ is called the torsion of the curve.

♣

Theorem 2.3. Frenet Formulas

If β : I → R3 is a unit-speed curve with curvature κ > 0 and torsion τ , then we have the
following system of ordinary differential equations:
T0 = κN
N 0 = −κT +τ B .
B0 = −τ N
In matrix notation, we have
 
0 −κ 0
(T, N, B)0 = (T, N, B) 
 
κ 0 .
−τ  (2.1)
0 τ 0
♥
Proof. The proof is an application of the orthogonality of the frame. The first
equation follows by definition. Since N · N 0 = 0, we can write

N 0 = aT + bB.

Thus a = (N 0 − τ B) · T = N 0 · T . Since N · T = 0, we have N 0 · T + N · T 0 = 0.

Thus
N 0 · T = −N · T 0 = −N · κT = −κ.

36
 2.3 The Frenet Formulas

By definition, we have
b = N 0 · B = τ.

This proves the second equation.

In order to obtain the third equation, we used the similar method. Write

B 0 = pT + qN + rB.

We have
p = B 0 · T = −B · T 0 = −B · κN = 0,

and
q = B 0 · N = −B · N 0 = −B · (−κT + τ B) = −τ,

and
r = B 0 · B = 0.

This proves the third formula.



Example 2.3 We compute the Frenet frame T, N, B and the curvature and torsion functions of
the unit-speed helix  
s s bs
β(s) = a cos , a sin , ,
c c c

where c = (a2 + b2 )1/2 and a > 0. Now

 
0 a s a s b
T (s) = β (s) = − sin , cos , .
c c c c c
Hence
 a s a s 
T 0 (s) = − 2 cos , − 2 sin , 0 .
c c c c
Thus
a a
κ(s) = kT 0 (s)k = 2
= 2 > 0.
c a + b2
Since T 0 = κN , we get
 s s 
N (s) = − cos , − sin , 0 .
c c
Therefore, we have  
b s b s a
B =T ×N = sin , − cos , .
c c c c c
The torsion is given by
b
τ = N0 · B = .
c2
In what follows, we use Frenet formulas to study the properties of curves.

37
 2.3 The Frenet Formulas

Definition 2.10
A plane curve in R3 is a curve that lies in a single plane of R3 .
♣

Theorem 2.4
Let β be a unit-speed curve in R3 with κ > 0. Then β is a plane curve if and only if τ = 0.
♥

Proof. If a curve β(s) = (β1 (s), β2 (s), β3 (s)) lies in a plane, then there are constants
a, b, c, d such that
aβ1 + bβ2 + cβ3 = d.

Using the vector notation, if v = (a, b, c), then v · β = d. As a result, we have

v · β 0 (s) = v · β 00 (s) = v · β 000 (s) = 0.

Since β 0 ⊥ β 00 , they are linearly independent. Thus we can write β 000 as a linear
combination of β 0 and β 00 .
β 000 = pβ 0 + qβ 00 ,

where p, q are constants. Since

B = T × N = κ−1 β 0 × β 00 ,

we have
τ = N 0 · B = (κ−1 β 00 )0 · B = 0.

On the other hand, if τ ≡ 0, then B 0 = 0. Thus B is a constant vector. Let

f (s) = β(s) · B − β(0) · B.

Then f 0 = 0 and since f (0) = 0, we know that f ≡ 0. Thus β(s) is on the plane

x · B − β(0) · B = 0.

In the following, we shall use the Taylor’s formula to study the curve β. We have
1 1
β(s) = β(0) + β 0 (0)s + β 00 (0)s2 + β 000 (0)s3 + o(s3 ).
2 6
Apparently, we have

β 0 (0) = T (0),
β 00 (0) = κ(0)N (0),
β 000 (0) = (κN )0 = κ0 (0)N + κ(0)(−κ(0)T (0) + τ (0)B(0)).

Therefore, we have
1
β(s) − β(0) = sT (0) + s2 κ(0)N (0)
2 (2.2)
1 3 0
+ s (κ (0)N + κ(0)(−κ(0)T (0) + τ (0)B(0))) + o(s3 ).
6

38
 2.4 Arbitrary-Speed Curves

From the above formula, we know that the Frenet Apparatus completely determined the curve,
at least for the first three terms in the Taylor’s expansion.

Remark We have emphasized all along the distinction between a tangent vector and a point of
R3 . However, Euclidean space has, as we have seen, the remarkable property that given a point
p, there is a natural one-to-one correspondence between points (v1 , v2 , v3 ) and tangent vectors
(v1 , v2 , v3 )p at p. Thus one can transform points into tangent vectors (and vice versa) by means
of this canonical isomorphism. In the next two sections particularly, it will often be convenient
to switch quietly from one to the other without change of notation. Since corresponding objects
have the same Euclidean coordinates, this switching can have no effect on scalar multiplication,
addition, dot products, differentiation, or any other operation defined in terms of Euclidean
coordinates.
So in (2.2), the left side β(s) − β(0) is a vector on R3 , and the right hand side, which is a
linear combination of T (0), N (0) and B(0), is a vector in the tangent space of β(0). We identify
Tβ(0) R3 with R3 so the equality makes sense.

2.4 Arbitrary-Speed Curves

It is a simple matter to adapt the results of the previous section to the study of a regular
curve α : I → R3 that does not necessarily have unit speed. We merely transfer to α the Frenet
apparatus of a unit-speed reparametrization ᾱ of α. Explicitly, if s is an arc length function for
α, then
α(t) = ᾱ(s(t))

for all t. Or, we can write α(t) = ᾱ(s).

Assume that T̄ , N̄ , B̄, κ̄ > 0, τ̄ are the Frenet Apparatus with respect to the unit speed curve
ᾱ(s). Then we define
curvature function: κ = κ̄(s),
torsion function: τ = τ̄ (s),
unit tangent vector field: T = T̄ (s),
principal vector field: N = N̄ (s),
binomial vector field: B = B̄(s).

The speed of the curve is defined by ν = kα0 (t)k. We can regard ν as a function of t but
throught t = t(s), it can be regarded as a function of s as well. Since α(t) = ᾱ(s), we have
α0 (t) = ᾱ0 (s) ds
dt . Thus we have
ds
= kα0 (t)k = ν.
dt

39
 2.4 Arbitrary-Speed Curves

Lemma 2.2
Assume that ds/dt > 0a . If α is a regular curve in R3 with κ > 0, then we have
T0 = κνN
N 0 = −κνT +τ νB .
B0 = −τ νN

a
We shall always assume this for the rest of the lecture notes. Thus we have ν = ds/dt.
♥

Proof. We shall use the Frenet formulas for unit speed curves. Since T (t) =
T̄ (s), N (t) = N̄ (s) and B(t) = B̄(s), we have T 0 (t) = T̄ 0 (s)ν, N 0 (t) = N̄ 0 (s)ν,
and B 0 (t) = B̄(s)ν. The lemma then follows from Theorem 2.3.



Note There is a commonly used notation for the calculus that completely ignores change of
parametrization. For example, the same letter would designate both a curve α and its unit-speed
parametrization ᾱ, and similarly with the Frenet apparatus of these two curves. Differences in
derivatives are handled by writing, say, dT /dt for T 0 (t) and dT /ds for T̄ 0 (s).

Lemma 2.3
If α is a regular curve with speed function ν, then the velocity and acceleration of α are
given by
dν
α0 = νT, α00 = T + κν 2 N.
dt
♥
Proof. The proof is not difficult but the notations are confusing. By the context,
α0 = α0 (t) and α00 = α00 (t). We thus have
ds
α0 = T = T ν,
dt
and
dν dν
α00 = T + ν T0 = T + κν 2 N.
dt dt


Remark The formula α0 = ν T is to be expected since α0 and T are each tangent to the curve
and T has a unit length, while kα0 k = ν.

40
 2.4 Arbitrary-Speed Curves

The formula for acceleration is more interesting. By definition, α00 is the rate of change of the
and in general both the length and the direction of α0 are changing. The tangential component
dν
dt T of α00 measures the rate of change of the length of α0 (that is, of the speed of α). The normal
component κν 2 N measures the rate of change of the direction of α0 . Newton’s laws of motion
show that these components may be experienced as forces. For example, in a car that is speeding
dν
up or slowing down on a straight road, the only force one feels is due to dt T . If one takes an
unbanked curve at speed ν, the resulting sideways force is due to κν 2 N . Here κ measures how
sharply the road turns; the effect of speed is given by ν 2 , so 60 miles per hour is four times as
unsettling as 30.

Note Assume that we live in a 1-dimensional space defined by the above curve α. We then can
only measure the tangential component of the acceleration α00 . As a 1-dimensional creature, it
is not possible for the creature to understand κ, the curvature of the curve. It would think it lives
in a straight line.

Theorem 2.5
Let α be a regular curve in R3 . Then

T = α0 /kα0 k,
N = B × T, κ = kα0 × α00 k/kα0 k3 ,
B = α0 × α00 /kα0 × α00 k, τ = (α0 × α00 ) · α000 /kα0 × α00 k2 .
♥
Proof. The proof is just a matter of applications of the chain rule and the Frenet
formulas. But it contains several basic techniques in differential geometry.
First, we have T = α0 /kα0 k. Next, using the above Lemma 2.3, we have

α0 × α00 = kα0 k T × (ν 0 T + κν 2 N ) = κν 3 B. (2.3)

Taking the norm of the above equation, we prove the formula for κ is proved.
That N = B × T follows from the definition. From Lemma 2.3, we have

α000 = (ν 0 T )0 + (κν 2 N )0 .

From the above, we know that the B components of α000 is κτ ν 3 . Thus we have

(α0 × α00 ) · α000 = κ2 τ ν 6 .

Since kα0 × α00 k2 = κ2 ν 6 , the formula for τ is proved.



External Link. As we know, the definition of the curvature is κ = ν −1 kT 0 (t)k. Can we obtain
the above formula by a straightforward computation. Yes, it is complicated, but such kind of
computation contains useful techniques in differential geometry. See here for the method.
Example 2.4 We compute the Frenet apparatus of the 3-curve

α(t) = (3t − t3 , 3t2 , 3t + t3 ).

41
 2.4 Arbitrary-Speed Curves

We have

α0 (t) = 3 (1 − t2 , 2t, 1 + t2 ),
α00 (t) = 6 (−t, 1, t),
α000 (t) = 6 (−1, 0, 1).

First, we have
√ √
ν = kα0 k = α0 · α0 = 3 2(1 + t2 ).

We also have
U1 U2 U3
0 00
α × α = 18 1 − t2 2t 1 + t2 = 18 (−1 + t2 , −2t, 1 + t2 ).
−t 1 t
Hence
√
kα0 × α00 k = 18 2(1 + t2 ).

We compute
(α0 × α00 ) · α000 = 6 · 18 · 2.

It remains only to substitute this data into the formulas in Theorem 2.5 with N being computed
by another cross product. The final results are
(1 − t2 , 2t, 1 + t2 )
T = √ ,
2(1 + t2 )
(−2t, 1 − t2 , 0)
N= ,
1 + t2
(−1 + t2 , −2t, 1 + t2 )
B= √ ,
2(1 + t2 )
1
κ=τ = .
3(1 + t2 )2

For the rest of the section, we shall do some applications of the above formulas.

Definition 2.11
The spherical image of a unit-speed curve β(s) is the curve σ(s) = T (s) = β 0 (s).
♣

Let σ be the spherical image of β. Then σ 0 = β 00 = κN , where κ is the curvature of β.

That κ > 0 ensures that σ is a regular curve. In order to compute the curvature κσ of σ, we
compute2

σ 0 = β 00 = κ N,
σ 00 = β 000 = κ0 N + κ N 0 = κ0 N + κ(−κ T + τ B).

2If κ > 0, then σ 00 6= 0. Thus the Frenet Apparatus always exists for σ.

42
 2.4 Arbitrary-Speed Curves

Thus
σ 0 × σ 00 = κ2 (κB + τ T ).

Using the formula in Theorem 2.5, we have

kσ 0 × σ 00 k
κσ = = (1 + (τ /κ)2 )1/2 ≥ 1.
ν3
Definition 2.12
A regular curve α in R3 is a cylindrical helix provided the unit tangent vector field T of
α has constant angle θ with some fixed unit vector u; that is, T (t) · u = cos θ for all t.
♣

Theorem 2.6
A regular curve α with κ > 0 is a cylindrical helix if and only if the ratio τ /κ is constant.
♥
Proof. First assume that a unit speed curve is a cylindrical helix curve. Then there
is a constant vector u such that T · u = c, a constant. Taking derivative on both sides,
we get κ N · u = 0. Since κ > 0, we get N · u = 0. Therefore, if we write u as a
linear combination of T, N, B, there would be no N component.
By the assumption, we would get

u = cos θ T + sin θ B.

Taking derivative of the above equation again, we get

0 = cos θ κ N − sin θ τ N.

We thus have κ cos θ − τ sin θ = 0. Hence τ /κ = cot θ is a constant.

Conversely, if τ /κ is a constant, we write τ /κ = cot θ. Let

u = cos θ T + sin θ B.

Then
u0 = cos θ κ N − sin θ τ N = 0.

Therefore u is a constant vector field. Obviously, kuk = 1. Then T · u = cos θ which

means that the curve is a cylindrical helix.


Can we prove the above result by solving the Frenet differential equations? Yes, in the
following, we give another proof of the fact that if τ /κ = c be a constant, then there must be a
unit vector u such that T · u is a constant.

43
 2.5 Covariant Derivatives

We use (2.1) to obtain

   
0 −κ 0 0 −1 0
(T, N, B)0 = (T, N, B) 
   
κ 0 −τ 
 = κ (T, N, B) 1 0 −c .
 
0 τ 0 0 c 0
Thus we have (See the remark below)
Rt
(T, N, B) = (T (0), N (0), B(0))eA 0 κ(u) du
,

where  
0 −1 0
 
A=
1 0 −c.

0 c 0

Since A is a skew-symmetric 3 × 3 matrix, it is singular3. Let v be a unit eigenvector of A with

respect to the zero eigenvalue. Let u be the vector such that

u · (T (0), N (0), B(0)) = v.

Then  
1
Rt
T · u = u · T = u · (T (0), N (0), B(0))eA 0 κ(u) du · 
 
 .
0 

Note that v, A = 0, we thus have

∞ Z t k
A
Rt
κ(u) du
X 1 k
ve 0 =v A κ(u) du = v.
k! 0
k=0
Thus  
1
 
T ·u=v·
0


is a constant.

Remark Let A(t) be a matrix-valued function. We consider the system of differential equations

A0 (t) = A(t)B(t), A(0) = A0 ,

where B(t) is a matrix-valued function. Then in general, we don’t have a formula for the (unique)
solution. If for any t, t0 , we have

B(t)B(t0 ) = B(t0 )B(t),

then we have the solution

Rt
B(u)du
A(t) = A0 e 0 .

For further explanation, see here.

3One can prove this fact by a straightforward computation. Alternatively, we observe that for skew-symmetric matrices,
all eigenvalues must be purely imaginary unless they are zero. Since all purely imaginary eigenvalues must be in
pairs (conjugate eigenvalues), there must be at least one zero eigenvalue for odd dimensional matrices.

44
 2.5 Covariant Derivatives

2.5 Covariant Derivatives

In this section, we shall define probably one of the most important concepts in differential
geometry, or may be even one of the most important concepts in mathematics: covariant
derivatives on vector fields.
Definition 2.13
Let W be a vector field on R3 , and let v be a tangent vector to R3 at the point p. Then the
covariant derivative of W with respect to v is the tangent vector

∇v W = W (p + tv)0 (0)

at the point p.
♣

On surface, the covariant derivative is just another version of directional derivative, es-
pecially we have already defined such kind of derivative along a curve before. But we shall
systematically use this kind of derivative under the frame of differential operators.
Example 2.5 Let W = x2 U1 + yzU3 , and let

v = (−1, 0, 2),

at p = (2, 1, 0). Then

p + tv = (2 − t, 1, 2t).

So
W (p + tv) = (2 − t)2 U1 + 2tU3 .

Hence
∇v W = W (p + tv)0 (0) = −4 U1 (p) + 2 U3 (p).

Lemma 2.4
If W = wi Ui is a vector field on R3 , and v is a tangent vector at p, then

∇v W = v[wi ]Ui (p).

♥
Proof. We havea

W (p + tv) = wi (p + tv)Ui (p + tv).

Differentiation of the above at t = 0, we obtain

∇v W = W (p + tv)0 (0) = v[wi ]Ui (p).

a
This equation is from the book which is, of course, correct. But in all previous formulas, we use Ui
instead of Ui (p + tv).

In short, to apply ∇v to a vector field, apply v to its Euclidean coordinates.

45
 2.6 Frame Fields

Theorem 2.7
Let v and w be tangent vectors R3 at p, and let Y and Z be vector fields on R3 . Then for
numbers a, b and function f
(1) ∇av+bw Y = a∇v Y + b∇w Y ,
(2) ∇v (aY + bZ) = a∇v Y + b∇v Z,
(3) ∇v (f Y ) = v[f ]Y (p) + f (p)∇v Y ,
(4) v[Y · Z] = ∇v Y · Z(p) + Y (p) · ∇v Z.
♥
Proof. The proof is straightforward. For example, to prove (3), let Y = yi Ui and
then f Y = (f yi )Ui . Thus

∇v (f Y ) = v[f yi ]Ui = v[f ]yi Ui + f v[yi ]Ui = v[f ]Y (p) + f (p)∇v Y.

If we regard the set of vector fields as a module over the algebra of smooth functions, then
the operator ∇v is linear with respect to the addition and scalar multiplication. It is a derivative
on the multiplication of smooth functions to vector fields.

Corollary 2.1

Let V, W, Y and Z be vector fields on R3 ; f, g be smooth functions. Then

(1) ∇f V +gW Y = f ∇V Y + g∇W Y ,
(2) ∇V (aY + bZ) = a∇V Y + b∇V Z,
(3) ∇V (f Y ) = V [f ]Y + f ∇V Y ,
(4) V [Y · Z] = ∇V Y · Z + Y · ∇V Z.
♥

2.6 Frame Fields

When the Frenet formulas were discovered (by Frenet in 1847, and independently by Serret
in 1851), the theory of surfaces in R3 was already a richly developed branch of geometry. The
success of the Frenet approach to curves led Darboux (around 1880) to adapt this “method of
moving frames” to the study of surfaces. Then it was Cartan who brought the method to full
generality. His essential idea was very simple: To each point of the object under study (a curve,
a surface, Euclidean space itself, . . .) assign a frame; then using orthonormal expansion express
the rate of change of the frame in terms of the frame itself. This, of course, is just what the Frenet
formulas do in the case of a curve.
In the next three sections we shall carry out this scheme for the Euclidean space R3 . We
shall see that geometry of curves and surfaces in R3 is not merely an analogue, but actually a
corollary, of these basic results.

46
 2.6 Frame Fields

Definition 2.14
Vector fields E1 , E2 , E3 on R3 constitute a frame field on R3 provided

Ei · Ej = δij .
♣

Apparently, U1 , U2 , U3 is a frame field. We call such a frame field Euclidean frame field.
In the following, we shall introduce two more important frame fields.
Example 2.6 (The cylindrical frame field) Let r, θ, z be the usual cylindrical coordinate functions
on R3 . We shall pick a unit vector field in the direction in which each coordinate increases (when
the other two are held constant). For r, this is evidently

E1 = cos θ U1 + sin θ U2 .

Then
E2 = − sin θ U1 + cos θ U2

points in the direction of increasing θ as in Fig. 2.19. Finally, the direction of increase of z is,
of course, straight up, so
E3 = U3 .

Remark I don’t think in the above example, deducing the cylindrical frame field from the
cylindrical coordinates by the picture, is mathematically rigid, or even correct. In fact, as long
as the above {E1 , E2 , E3 } is an orthonormal basis at each point, it defines a frame field. We can
call it the cylindrical frame field. There is no proof needed here.
A better way to show the relationship between the cylindrical coordinates and the cylindrical
frame field is to use the chain rule. We have x = r cos θ, y = r sin θ, z = z. Then by the chain
rule, for any function f , we have
∂f ∂x ∂f ∂y ∂f ∂z ∂f
= · + · + · .
∂r ∂r ∂x ∂r ∂y ∂r ∂z
In terms of vector notation, this implies that
∂ ∂x ∂y ∂z
= U1 + U2 + U3 = cos θ U1 + sin θ U2 .
∂r ∂r ∂r ∂r
Using the same method, we have
∂
= U3 .
∂z

47
 2.7 Connection forms

However,
∂ ∂x ∂y ∂z
= U1 + U2 + U3 = −r sin θ U1 + r cos θ U2
∂θ ∂θ ∂θ ∂θ
which is not E2 but rE2 .
In general, if {x1 , x2 , x3 } are another set of coordinate functions, then the vector fields
 
∂ ∂ ∂
, ,
∂x1 ∂x2 ∂x3
do not define a frame field (they may not be orthonormal).
Example 2.7 (The spherical frame field) Now we define the spherical frame field {F1 , F2 , F3 }.
Let ρ, θ, ϕ be the spherical coordinates. We then have

x = ρ cos ϕ cos θ,
y = ρ cos ϕ sin θ,
z = ρ sin ϕ.

The spherical frame field is defined by

F1 = cos ϕ (cos θ U1 + sin θ U2 ) + sin ϕ U3

F2 = − sin θ U1 + cos θ U2 ,
F3 = − sin ϕ (cos θ U1 + sin θ U2 ) + cos ϕ U3 .

By a straightforward computation, we have

∂
= E1 ,
∂ρ
∂
= −ρ cos ϕ sin θ U1 + ρ cos ϕ cos θ U2 = ρ cos ϕ F2 ,
∂θ
∂
= −ρ sin ϕ cos θ U1 − ρ sin ϕ sin θ U2 + ρ cos ϕ U3 = ρ F3 .
∂ϕ

Lemma 2.5
Let {E1 , E2 , E3 } be a frame field on R3 .
(1). If V is a vector field on R3 , then V = fi Ei , where the functions fi = V · Ei are
called coordinate functions of V with respect to {E1 , E2 , E3 }.
(2). If V = fi Ei and W = gi Ei , then V · W = fi gi . In particular, kV k = ( fi2 )1/2 .
P
♥

48
 2.7 Connection forms

2.7 Connection forms

Let {E1 , E2 , E3 } be a frame field, and let v be a vector at p ∈ R3 . Let p ∈ R3 . Then we

are able to write

∇v E1 = c11 E1 (p) + c12 E2 (p) + c13 E3 (p),

∇v E2 = c21 E1 (p) + c22 E2 (p) + c23 E3 (p),
∇v E3 = c31 E1 (p) + c32 E2 (p) + c33 E3 (p).

Using the Einstein Convention, we have

∇v Ei = cij Ej (p).

Since {E1 , E2 , E3 } is a frame, we then have4

cij = ∇v Ei · Ej (p).

We then can define a one form ωij such that

ωij = cij = ∇v Ei · Ej (p).

Lemma 2.6
Using the above notations, then ωij are 1-forms satisfying

ωij = −ωji .

These 1-forms are called connection forms of the frame field {E1 , E2 , E3 }.
♥
Proof. In order to prove ωij is a 1-form, we just need to prove that

ωij (av + bw) = a ωij (v) + b ωij (w),

where v, w are vectors and a, b are real numbers. But this follows from

∇av+bw Ei = a ∇v Ei + b ∇w Ei .

To prove ωij = −ωji , we observe that since Ei · Ej = δij . Then

0 = v[Ei · Ej ] = ∇v Ei · Ej + Ei · ∇v Ej = ωij (v) + ωji (v).

Theorem 2.8
Let ωij be the connection 1-froms of a frame field {E1 , E2 , E3 } on R3 . Then for any
vector field V on R3 , we have

∇V Ei = ωij (V )Ej .
♥

4We know that 1 ≤ i, j ≤ 3. But this is implied by the Einstein convention so can be omitted.

49
 2.7 Connection forms

By ωij = −ωji , we have ωii = 0. We write

   
ω11 ω12 ω13 0 ω12 ω13
   
ω= ω21  = −ω12
ω22 ω23   0 .
ω23 
ω31 ω32 ω33 −ω13 −ω23 0
If we write
E = (E1 , E2 , E3 ),

then in matrix notations, we have

∇v E = E · ω T .

This can be used to compare with the Frenet formula

 
0 −κ 0
(T, N, B)0 = (T, N, B) 
 
κ 0 .
−τ 
0 τ 0

50
 Bibliography

[1] Lu, Z. (2019). Lecture Notes of UCI’s Math 162AB. Hand-written.

[2] O’Neill, B. (2006). Elementary differential geometry. Elsevier/Academic Press, Amsterdam, second edition.