Analysis of Several Variables

Jaydeb Sarkar
Statistics and Mathematics Unit
Indian Statistical Institute, Bangalore
LATEXed by

Trishan Mondal
Soumya Dasgupta
Aaratrick Basu
Priyatosh Jana
Bikram Halder
Saraswata Sensarma
Deepta Basak

B. Math (Hons.) (2021-’24)

Indian Statistical Institute, Bangalore

ii
Contents

Credits ii

Contents iii

Notation vii

Lecture 1 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Review: Rn as a vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Linear Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.4 Analytic ideas in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Lecture 2 5
2.1 Distance and Topology in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Limits and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Lecture 3 9
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Properties of Continuous functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Lecture 4 13
4.1 More examples of Continuous maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Lecture 5 17
5.1 Matrix representation of the derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Further properties of differentiable functions . . . . . . . . . . . . . . . . . . . . . . . . 18

Lecture 6 21
6.1 Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.2 Geometric Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.4 Higher Order Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.5 Clairaut’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

Lecture 7 25
7.1 Schwarz Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.2 Partial and Total Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

Lecture 8 27
8.1 A kind of converse of Theorem 7.2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

iii
Lecture 9 31
9.1 Directional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
9.2 Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
9.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Lecture 10 35
10.1 Extension of MVT to Several Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
10.2 More Partials and Chain Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

Lecture 11 39
11.1 Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
11.2 Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
11.3 Extrema of a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Lecture 12 45
12.1 Hessian Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
12.2 Positive Definite, Negative Definite, Semi Definite Matrices . . . . . . . . . . . . . . . . 46

Lecture 13 49
13.1 Taylor’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

Lecture 14 53
14.1 Compact subsets of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
14.2 Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

Lecture 15 59
15.1 Inverse function theorem: Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
15.2 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
15.3 Solving systems of equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Lecture 16 63
16.1 Riemann-Darboux Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Lecture 17 65
17.1 Properties of Riemann-Darboux Integration . . . . . . . . . . . . . . . . . . . . . . . . 65
17.2 Iterated Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Lecture 18 69
18.1 Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Lecture 19 73
19.1 Integration over Bounded Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
19.2 Two Elementary Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Lecture 20 77
20.1 Fubini’s Theorem on Elementary Regions . . . . . . . . . . . . . . . . . . . . . . . . . . 77
20.2 Change of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Lecture 21 81
21.1 Change of Variables (Continued) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Lecture 22 85
22.1 Curves and Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

Lecture 23 89
23.1 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

iv
Lecture 24 93
24.1 Planes and Normals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
24.2 Surface and Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
24.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

Lecture 25 97
25.1 Tangent Plane Of G(f ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
25.2 Surface Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

Lecture 26 101
26.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
26.2 Surface Integral over Scalar fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
26.3 Surface Integral over a Vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Lecture 27 105
27.1 Conservative Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
27.2 Green’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

Lecture 28 111
28.1 Green’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
28.2 Gauss Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
28.3 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Index 117

v
vi
List of Notation

ℓ(γ) Arclength of a rectifiable curve γ

Br (a) Open ball centered at a ∈ Rn of radius r

∂S Boundary of a set S ⊆ Rn
B(S) Space of all bounded functions defined over S
n
Y
n
B, B = [ai , bi ] Closed box in Rn
i=1
C k (S) Class of all functions defined on S for which all partial
derivatives on S upto order k exist and are continuous
S Closure of S ⊆ Rn
∇ × F⃗ Curl of F⃗

Dr (a) Deleted ball of radius r around a ∈ Rn

d(A) Diameter of the set A ⊆ Rn
X m!
(∇ · h)m = hα ∂ α Differential order of order m and coefficients h
α!
|α|=m
(Du f )(a) Directional derivative of f along u at a
∇ · F⃗ Divergence of F⃗

Ext(S) Exterior of a set S ⊆ Rn

∇f Gradient of f
∇f (x) Gradient of f at x
G(f ) Graph of a f : R2 → R

Hf (x) Hessian of f at x
∂ ∂f ∂2f
fxi xj (a), , Partial derivative of f with respect to the ith and j th coordi-
∂xj ∂xi ∂xi ∂xj
nates at a

χS Indicator function of S
⟨x, y⟩ Euclidean inner product of x and y
R(S) Space of all Riemann-Darboux integrable functions over S
Int(S),
Z S◦ Interior of a set S ⊆ Rn
f dV Riemann-Darboux integral of f over S
S

Jf (a) Jacobian matrix of f at a

∆f (x) Laplacian of f at x

vii
lim f Limit of f as x tends to a
x→a
′
S
Z Z Set of limit points of S ⊆ Rn
f, f Line integral of f over C = ran(γ)
Zγ C Z
F⃗ · d⃗r, F⃗ · d⃗r Line integral of F⃗ over C = ran(γ)
Zγ C

f Lower Riemann integral of f over B

B
L(f, P ) Lower Riemann sum of f with respect to the partition P

∥P ∥ Mesh of the partition P

∂ |α|
∂α = αth derivative, for α ∈ Zn≥0
∂x1 · · · ∂xn αn
α 1
n
Y
α! = αi ! Factorial of α ∈ Zn≥0
i=1
Xn
|α| = αi Modulus of α ∈ Zn≥0
i=1

∥x∥ Euclidean norm of x ∈ Rn

On , O Open set in Rn

P(B n ) Set of all partitions of the box B n

∂f
, fxi (a) Partial derivative of f with respect to the ith coordinate at a
∂xi
R[x1 , . . . , xn ] Space of polynomials with real coefficients in n variables
P(S) Power set of the set S
Πi Projection onto ith coordinate

QA (x) = xt Ax Quadratic form associated to the matrix A evaluated at x

Rn n−dimensional Euclidean space

Z
f dS Surface integral of f over S
ZS Z
F⃗ · dS
⃗= F⃗ · ⃗n dS Surface integral of F⃗ over S
S S

TP S Tangent plane of S through P

X 1
pa,k (f, h) = (∂ α f )(a)hα Taylor polynomial of f at a, of order k and difference h
α!
|α|≤k
Df (a) Derivative of f at a
Z
f Upper Riemann integral of f over B
B
U (f, P ) Upper Riemann sum of f with respect to the partition P

Vol(S) Volume of the (measurable) set S

viii
Lecture 1

1.1 Introduction
We will talk about n-variable calculus, that is, calculus on Rn . Recall the following:

• The setting is,

Rn = R × R · · · × R = {x = (x1 , . . . , xn ) | xi ∈ R ∀ i = 1, 2, . . . , n}
| {z }
n times

• Analysis on R consisted of ideas like open sets, compact sets, convergence, limits, differentiability,
integrability etc.

• Rn is an n-dimensional inner product space over R, with the standard orthonormal basis
{ej }nj=1

Extending the analytic ideas to Rn exploiting the algebraic structure is the matter of this course,
which further gives way to differential geometry.

1.2 Review: Rn as a vector space

(i) The standard orthonormal basis of Rn is {ei }ni=1 .

(ii) For all x ∈ Rn , there is a unique representation

n
X
x= xi ei , x i ∈ R
i=1

Thus we identify x with the coordinates (x1 , x2 , . . . , xn ).

(iii) Euclidean inner product on Rn :

For x = (x1 , . . . , xn ), y = (y1 , . . . , yn ) ∈ Rn , we define
n
X
⟨x, y⟩ = xi yi
i=1

1.3 Linear Functions

In doing analysis on R, the main motive was to study functions f : R → R and their properties,
namely continuity, differentiability, integrability etc. We now wish to do the same for functions
f : Rn → Rm for arbitrary natural numbers n, m.
Two easy examples of such functions are:

1
1.4 Analytic ideas in Rn

(i) Constant maps

f : Rn → R m
∀ x ∈ Rn , f (x) = a, a ∈ Rm

(ii) Linear maps

A function L : Rn → Rm is linear if for all α ∈ R, x, y ∈ Rn , L(αx + y) = αL(x) + L(y).

It turns out that linear maps are useful in understanding most other ‘nice’ functions, and so we now
look at these in more detail.
Let L be a linear map from Rn to Rm . Consider the domain {αx + y | α ∈ R}, that is, the line
through y in the direction of x. The image under L is,

{αLx + Ly | α ∈ R}

which is the line through Ly in the direction of Lx. Hence, L maps lines to lines.
Exercise. Is the converse also true?

Matrix representation of a linear map

Let L : R → R be a linear map. Then,

L(x) = xL(1) ∀x ∈ R

Therefore,
L(R, R) := {set of all linear maps from R to R} ↔ R
Now consider the general case; let L : Rn → Rm be a linear map. If we fix the bases {ej }nj=1 of Rn
and {ei }m m
i=1 of R , L is determined uniquely by the equations

m
X
Lej = aij ei
i=1

and hence,
L ↔ (aij )m×n ∈ Mm,n (R)

1.4 Analytic ideas in Rn

We have the Euclidean norm on Rn defined by,

n
! 12
X
∥x∥ = x2i ∀ x ∈ Rn
i=1

This induces the metric given as,

n
! 21
X
d(x, y) = ∥x − y∥ = (xi − yi )2 ∀ x, y ∈ Rn
i=1

Theorem 1.4.1 (Cauchy-Schwarz Inequality)

For all x, y ∈ Rn ,
⟨x, y⟩ ≤ ∥x∥∥y∥

2
 Lecture 1

Proof. Consider x, y ∈ Rn . We have,

n X
X n
(xi yj − xj yi )2 ≥ 0
i=1 j=1

But the left hand side is, after expanding,

n
X n
X n
X 2 2 2
x2i yj2 + x2j yi2 −2 xi xj yi yj = 2∥x∥ ∥y∥ − 2 ⟨x, y⟩
i,j=1 i,j=1 i,j=1

which gives the desired inequality.

Note: The proof shows that equality holds only if there is λ ∈ R such that for all i, either xi = λyi
or yi = λxi .
Recall the triangle inequality for R, for all x, y ∈ R

|x + y| ≤ |x| + |y|

Theorem 1.4.2 (Triangle inequality for Rn )

For all x, y ∈ Rn ,
∥x + y∥ ≤ ∥x∥ + ∥y∥

Proof. We have,
2
∥x + y∥ = ⟨x + y, x + y⟩
2 2
= ∥x∥ + 2 ⟨x, y⟩ + ∥y∥
2 2
≤ ∥x∥ + ∥y∥ + 2∥x∥∥y∥ (Cauchy Schwarz inequality)
=⇒ ∥x + y∥ ≤ ∥x∥ + ∥y∥

which is the desired inequality.

The following is a technical result, which can be thought of as an analogue of the Lipschitz condition
for linear maps on Rn . It hints towards continuity of linear maps, and we will see that it is indeed so
later.

Theorem 1.4.3
Let L : Rn → Rm be a linear map. There is M > 0 such that

∥Lx∥ ≤ M ∥x∥ ∀ x ∈ Rn

Pn
Proof. We have, for x = i=1 xi ei ,
n
X
∥Lx∥ = xi Lei
i=1
n
X
≤ |xi |∥Lei ∥ (Triangle inequality)
i=1
n
! 12
X 2
=⇒ ∥Lx∥ ≤ ∥x∥ ∥Lei ∥ (Cauchy Schwarz inequality)
i=1
P  12
n 2
Taking M = i=1 ∥Lei ∥ , we get the result.

3
1.4 Analytic ideas in Rn

4
Lecture 2

2.1 Distance and Topology in Rn

Using the inner product on Rn , we get the Euclidean distance
v
u n
p uX
d(x, y) = ∥x − y∥ = ⟨x − y, x − y⟩ = t (xi − yi )2
i=1

We wish to extend the notions of (R, |·|) to (Rn , ∥·∥). We already know from Lecture 1 that the
triangle inequality holds,
∥x + y∥ ≤ ∥x∥ + ∥y∥

Definition 2.1.1 ▶ Open balls

The open ball centered at a ∈ Rn of radius r is,

Br (a) = {x ∈ Rn | ∥x − a∥ < r}

Exercise. Show that open balls are convex sets.

Definition 2.1.2 ▶ Open sets

A set O ⊆ Rn is open if O = ϕ or for all x ∈ O, there is r > 0 such that Br (x) ⊆ O.

Example 2.1.1

(i) Any open ball is open.

Qn
(ii) We define open boxes in Rn to be the subsets of the form i=1 (ai , bi ). Any open box is
open.

Definition 2.1.3 ▶ Convergence of Sequences

Let {xm }m∈N ⊆ Rn and x ∈ Rn . We say xm −→ x if for all ε > 0, there is N ∈ N such that
∥xm − x∥ < ε ∀ m ≥ N
⇐⇒ d(xm , x) < ε ∀ m ≥ N
⇐⇒ xm ∈ Bε (x) ∀m ≥ N

Exercise. Show that the limit of a sequence in Rn is unique whenever it exists.

5
2.1 Distance and Topology in Rn

Definition 2.1.4 ▶ Limit points

We define the deleted ε-neighbourhood of a ∈ Rn to be Dε (a) = Bε (a) \ {a}. The point a is a
limit point of S ⊆ Rn if for all ε > 0, Dε (a) ∩ S ̸= ϕ. If we do not delete a, we get isolated
points.

Definition 2.1.5 ▶ Projections

For all i ∈ {1, 2, . . . , n} we define the maps

Π i : Rn → R
x = (x1 , . . . , xn ) 7→ xi

Πi is called the projection onto the ith coordinate.

Theorem 2.1.1
Let {xm }m∈N ∪ {x} ⊆ Rn . Then,

xm −→ x
⇐⇒ Πi (xm ) −→ Πi (x) ∀ i ∈ {1, 2, . . . , n}

Proof. Assume xm −→ x. Now, for all j ∈ {1, 2, . . . , n},

n
X
2 2 2
∥xm − x∥ = |Πi (xm ) − Πi (x)| ≥ |Πj (xm ) − Πj (x)|
i=1
=⇒ |Πj (xm ) − Πj (x)| −→ 0
=⇒ Πj (xm ) −→ Πj (x)
Now assume Πj (xm ) −→ Πj (x) for all j ∈ {1, 2, . . . , n}. Then,
|Πj (xm ) − Πj (x)| −→ 0, ∀ j ∈ {1, 2, . . . , n}
n
X 2
=⇒ |Πj (xm ) − Πj (x)| −→ 0
i=1
2
=⇒ ∥xm − x∥ −→ 0
=⇒ xm −→ x

Definition 2.1.6 ▶ Closed sets

A set C ⊆ Rn is closed if Rn \ C is open.

Exercise. Show that a set C ⊆ Rn is closed iff ∀ {xm }m∈N ⊆ C with xm −→ x for some x ∈ Rn , we
have x ∈ C.
Exercise. Show that:
(1) Arbitrary union of open sets is open.
(2) Finite intersection of open sets is open.
(3) Arbitrary intersection of closed sets is closed.
(4) Finite union of closed sets is closed.
(5) Any finite subset of Rn is closed.

6
 Lecture 2

Definition 2.1.7 ▶ Interior of a set

Let ϕ ̸= S ⊆ Rn . The interior of S is,

Int(S) = {a ∈ S | ∃ r > 0, Br (a) ⊆ S}

Exercise. Show that:

(1) For any nonempty set S ⊆ Rn , Int(S) is open.

(2) A set S is open iff Int(S) = S.

Definition 2.1.8 ▶ Exterior of a set

Let ϕ ̸= S ⊆ Rn . The exterior of S is,

Ext(S) = {a ∈ Rn | ∃ r > 0, Br (a) ∩ S = ϕ}

Exercise. Show that Ext(S) = Int(Rn \ S).

Example 2.1.2

For S = [0, 2] \ {1} = [0, 1) ∪ (1, 2], 1 ∈

/ Ext(S).

Definition 2.1.9 ▶ Boundary of a set

Let ϕ ̸= S ⊆ Rn . The boundary of S is,

∂S = {a ∈ Rn | ∀ r > 0, Br (a) ∩ S ̸= ϕ and Br (a) ∩ (Rn \ S) ̸= ϕ}

Example 2.1.3

For S = [0, 1) ∪ (1, 2] ∪ {5}, ∂S = {0, 1, 2, 5} but the set of limit points is {0, 1, 2}.

Exercise. Show that:

(1) S is open iff S ∩ ∂S = ϕ.

(2) S is closed iff S ⊇ ∂S.

(3) S is closed iff S = S =: S ∪ ∂S = S ∪ {Limit points of S}

(4) S = Int(S) ⊔ ∂S. This gives the partition Rn = Int(S) ⊔ ∂S ⊔ Ext(S)

(5) ∂S is closed.

Qn
(6) Let {Oi }ni=1 ⊆ P(R) and define O = i=1 Oi . If Oi ’s are open (closed), O is open (closed).

7
2.2 Limits and Continuity

2.2 Limits and Continuity

Recall the notion of limit in R:
Suppose f : (a, b) \ {c} → R is a function. We say lim f exists if there is α ∈ R such that
x→c
∀ ε > 0, ∃ δ > 0 such that x ∈ Dδ (c) =⇒ |f (x) − α| < ε, that is, f (Dδ (c)) ⊆ Bε (α); in such a case
we say that lim f = α.
x→c
We now extend this to Rn .

Definition 2.2.1 ▶ Limits in Rn

Let S ⊆ Rn , a ∈ {Limit points of S} and, f : S \ {a} → Rm . We say that lim f = b if for all
x→a
ε > 0, there is δ > 0 such that f (x) ∈ Bε (b) for all x ∈ Dδ (a) ∩ S.
In other words, lim f = b if for all ε > 0, there is δ > 0 such that
x→a

∥f (x) − b∥ < ε ∀ x ∈ S, 0 < ∥x − a∥ < δ

Definition 2.2.2 ▶ Continuity in Rn

Let S ⊆ Rn , a ∈ S and, f : S → Rm . We say that f is continuous at a if for all ε > 0, there
is δ > 0 such that f (x) ∈ Bε (f (a)) for all x ∈ Bδ (a) ∩ S, that is, ∥f (x) − f (a)∥ < ε for all
x ∈ S with ∥x − a∥ < δ.

Note: Any function defined on S is vacuously continuous at an isolated point a by our definition.

8
Lecture 3

3.1 Introduction
We denote the set of limit points of S ⊆ Rn by S ′ . Let f : S → Rm and a ∈ S. f is continuous at a
iff for all ε > 0, there is δ > 0 such that

∥f (x) − f (a)∥ < ε ∀ ∥x − a∥ < δ, x ∈ S

⇐⇒
f (Bδ (a) ∩ S) ⊆ Bε (f (a))

For a ∈ S ′ , we then get that f is continuous at a iff

lim ∥f (x) − f (a)∥ = 0 ⇐⇒ lim ∥f (a + h) − f (a)∥ = 0

∥x−a∥→0 ∥h∥→0

But ∥h∥ → 0 ⇐⇒ h → 0, and so we have f is continuous at a ∈ S ′ iff

lim ∥f (a + h) − f (a)∥ = 0
h→0

The proof of the following theorem is left as an exercise.

Theorem 3.1.1
Let S ⊆ Rn , a ∈ S ′ , b ∈ Rm , f : S → Rm . The following are equivalent:
(i) lim f = b
x→a

(ii) ∀ {xp } ⊆ S \ {a} with xp −→ a, we have f (xp ) −→ b

(iii) lim ∥f (x) − b∥ = 0
x→a

Note: If a ∈ S, we can take b = f (a) and get analogous results for continuity at a.

3.2 Properties of Continuous functions

Definition 3.2.1 ▶ Continuity on sets

Let S ⊆ Rn and f : S → Rm . f is continuous on S if it is continuous at all a ∈ S.

Note: It is convenient to take S to be open, as f is continuous at any isolated points of S vacuously.

9
3.2 Properties of Continuous functions

Theorem 3.2.1
Let S ⊆ Rn , f : S → Rm . The following are equivalent:

(1) f is continuous on S.
(2) ∀ {xp } ⊆ S with xp −→ a ∈ S, we have f (xp ) −→ f (a)
(3) (Assuming S is open) f −1 (O) is open for all O ⊆ Rm open.
(4) (Assuming S is open) f −1 (C) is closed for all C ⊆ Rm closed.

Proof. We have the following cases.

• (3) ⇐⇒ (4)
This is true as if g : X → Y then for all A ⊆ Y , g −1 (Y \ A) = X \ g −1 (A).

• (1) ⇐⇒ (2)
True by 3.2.1

• (1) =⇒ (3)
Let O ⊆ Rm be open, and without loss of generality, f −1 (O) ̸= ϕ.

Let a ∈ f −1 (O) so that f (a) ∈ O. Hence, there is r > 0 such that Br (f (a)) ⊆ O. By continuity,
there is δ > 0 such that

f (Bδ (a)) ⊆ Br (f (a)) ⊆ O

=⇒ Bδ (a) ⊆ f −1 (O)

Hence, O is open.

• (3) =⇒ (1)
Fix a ∈ S and let ε > 0. As Bε (f (a)) is open in Rm , we have f −1 (Bε (f (a))) is open in Rn .
But a ∈ f −1 (Bε (f (a))), and so, there is δ > 0 such that

Bδ (a) ⊆ f −1 (Bε (f (a)))

=⇒ f (Bδ (a)) ⊆ Bε (f (a))

Hence, f is continuous at a for all a ∈ S.

This theorem gives us a huge simplification. Recall that x −→ y iff Πi (x) −→ Πi (y) for all i.
Now consider some {xp } with xp −→ a. We have

f (xp ) −→ f (a) ⇐⇒ Πi (f (xp )) −→ Πi (f (a)) ∀ i

That is, f is continuous iff f is continuous coordinate wise! Hence, for talking about continuity, it is
enough to discuss real valued functions rather than f : Rn → Rm for arbitrary m.

10
 Lecture 3

3.3 Examples

Example 3.3.1

Consider the function,

f : R2 \ {(0, 0)} → R
2xy
f (x, y) = 2
x + y2

Consider the line L1 defined by y = 0 and approach (0, 0) from the right (x −→ 0+ ). We
have,  
1
f L ≡ 0 =⇒ lim f = lim f 0, =0
1 (x,y)→0 along L1 n→∞ n
Now consider the line L2 defined by x = y. We have,
 
1 1
f L2
≡ 1 =⇒ lim f = lim f , =1
(x,y)→0 along L2 n→∞ n n

Hence, lim f does not exist.

(x,y)→(0,0)

Note: The approach in the above example is often useful for showing non-existence of limits, or that
a function is not continuous.

Example 3.3.2

x3
We wish to compute lim . We have, for all (x, y) ̸= (0, 0),
(x,y)→(0,0) x2 + y 2

x3 x3
≤ = |x| ≤ ∥(x, y)∥
x2 + y 2 x2

and because ∥(x, y)∥ goes to 0 as (x, y) −→ (0, 0), we get

x3
lim =0
(x,y)→(0,0) x2 + y 2

Hence, the function (

x3
(x, y) ̸= (0, 0)
x2 +y 2 ,
f (x, y) =
0, (x, y) = (0, 0)
is continuous at (0, 0).

Example 3.3.3


sin x2 + y 2
Consider lim . We have (x2 + y 2 ) −→ 0 as (x, y) → (0, 0), and hence, we
(x,y)→(0,0) x2 + y 2
get

sin x2 + y 2 sin t
lim 2 2
= lim =1
(x,y)→(0,0) x +y t→0 t

11
3.3 Examples

Exercise. Let S ⊆ Rn , a ∈ S ′ and f, g : S → R. Suppose lim f = α, lim g = β exist. Show that:

x→a x→a

(i) lim (rf + g) = rα + β, ∀ r ∈ R

x→a

(ii) limx→a f g = αβ
f α
(iii) If β ̸= 0, lim =
x→a g β
(iv) If f ≤ h ≤ g for some h : S → R and α = β, then lim h = α
x→a

Note: Similar results hold for continuity as well, using which we get the next examples.

Example 3.3.4 (Some classes of continuous functions)

(1) The projection maps Πi : Rn → R are continuous.

(2) xi ∈ R[x1 , . . . , xn ] is continuous for all i.
(3) x2i ∈ R[x1 , . . . , xn ] is continuous for all i.

(4) All monomials in R[x1 , . . . , xn ] are continuous.

(5) Any p ∈ R[x1 , . . . , xn ] is continuous.
p
(6) q is continuous at a ∈ Rn , where p, q ∈ R[x1 , . . . , xn ] and q(a) ̸= 0.

12
Lecture 4

4.1 More examples of Continuous maps

Example 4.1.1

Consider the function

√ xy
(
, (x, y) ̸= (0, 0)
f (x, y) = x +y 2
2

0, (x, y) = (0, 0)
f is clearly continuous on R2 \ {(0, 0)} so we need only check the limit at (0, 0). We have,

xy 1 x2 + y 2 1
0≤ p ≤ p = ∥(x, y)∥
x2 + y 2 2 x2 + y 2 2

By the squeeze theorem, we get lim f (x, y) = 0, and hence, f is continuous on R2 .

(x,y)→(0,0)

Exercise. Show that any linear map is continuous. (Hint: Use that the norm is continuous.)

Example 4.1.2

Let D = {(x, y) ∈ R2 | y ̸= 0} = Π−1

2 (R \ {0}). Consider the function

f :D→R
1
f (x, y) = x sin
y
Clearly f is continuous on D. We also have,

1
0 ≤ x sin ≤ ∥(x, y)∥
y

and hence, by the Squeeze theorem, lim f (x, y) = 0

(x,y)→(0,0)
So, we can extend f to (0, 0) continuously by defining it to be 0.

13
4.2 Uniform Continuity

4.2 Uniform Continuity

Definition 4.2.1 ▶ Uniform Continuity

Let f : On → Rm , where On is open in Rn . f is uniformly continuous if for all ε > 0, there is
δ > 0 such that

f (x) ∈ Bε (f (a)) ∀ x ∈ Bδ (a) ∩ On , ∀ a ∈ On

~
w

∥f (x) − f (a)∥ < ε ∀ ∥x − a∥ < δ, a, x ∈ On

Exercise. Show that uniform continuity implies continuity.

4.3 Derivatives
We will use the following notation for the sake of brevity:
(1) On denotes an open set in Rn , and we omit the n if it is clear from context.
(2) A function f : Rn → Rm has components f = (f1 , f2 , . . . , fm )

Recall the notion of derivative in R:

Let f : O1 → R be a function and a ∈ O1 . Then f is differentiable at a if there is a real number
α(= f ′ (a)) such that
f (a + h) − f (a)
lim =α (1)
h→0 h
This clearly cannot be carried over verbatim to functions of several variables; we can’t divide by a
vector! To get a reasonable definition, we note that (1) is equivalent to,
f (a + h) − f (a) − αh
lim =0
h→0 h
Now, h 7→ αh is a linear map and we already know {linear maps on R} ←→ R. So the derivative
f ′ (a) is really a linear map! This leads to the following definition.

Definition 4.3.1 ▶ Derivative in several variables

Let f : On → Rm and a ∈ On . f is differentiable at a if there is a linear map L : Rn → Rm
such that
f (a + h) − f (a) − Lh
lim =0
h→0 ∥h∥
~
w

∥f (a + h) − f (a) − Lh∥
lim =0
h→0 ∥h∥
~
w

∥f (x) − f (a) − L(x − a)∥
lim =0
x→a ∥x − a∥

If f is differentiable at a ∈ On , its derivative is denoted as Df (a). We say f is differentiable

on On if it is differentiable at all a ∈ On .

14
 Lecture 4

Theorem 4.3.1
Let f : On → Rm be differentiable at a ∈ On . The derivative Df (a) is unique.

Proof. Let L = Df (a) and L1 be any linear map from Rn to Rm such that

∥f (a + h) − f (a) − L1 h∥
lim =0
h→0 ∥h∥

̸ L. Then, there is h0 ∈ Rn such that ∥h0 ∥ = 1 and Lh0 ̸= L1 h0 . Consider the map
Suppose L1 =
n
h : R → R given by h(t) = th0 . Then, by the triangle inequality,

∥L(h(t)) − L1 (h(t))∥ ∥f (a + h) − f (a) − L(h(t))∥ ∥f (a + h) − f (a) − L1 (h(t))∥

≤ +
|t| ∥h(t)∥ ∥h(t)∥

Taking the limit as t → 0, both terms on the right go to 0 by definition, and hence

∥L(h(t)) − L1 (h(t))∥ |t|∥Lh0 − L1 h0 ∥

lim = 0 =⇒ lim = 0 =⇒ Lh0 = L1 h1
t→0 |t| t→0 |t|

which clearly contradicts the assumption. So, the derivative Df (a) is unique.

4.4 Examples
Example 4.4.1

Consider f : On → Rm defined by f (x) = c. For any a ∈ On ,

f (a + h) − f (a) − Oh
lim =0
h→0 ∥h∥

where O denotes the zero linear map. Hence, f is differentiable on On and Df (a) = O for
any a ∈ On .

Example 4.4.2

Consider a linear map L : Rn → Rm . For all a ∈ Rn ,

L(a + h) − La − Lh
lim =0
h→0 ∥h∥

Hence, L is differentiable everywhere and DL(a) = L for all a ∈ Rn . This is as expected, as

the best linear approximation of a linear map is itself.

At this point, we are faced with the problem of actually computing derivatives of non-trivial maps. A
priori, it is not even clear if functions that are made of various differentiable functions of one variable,
say f (x, y, z) = (x2 ey z, y 3 sin(xy) cos z), are differentiable! We will perform a series of reductions
that will answer such basic questions about differentiability of functions and even provide techniques
to compute the derivatives.

15
4.4 Examples

16
Lecture 5

We have figured out that the reasonable definition of the derivative of a map f : On → Rm at a point
a ∈ On is Df (a) = L where
f (a + h) − f (a) − Lh
lim =0
h→0 ∥h∥
We now proceed to perform a series of reductions that will actually answer the question of
differentiability of functions and also provide techniques to compute the derivative.

5.1 Matrix representation of the derivative

Theorem 5.1.1
Let f : On → Rm and a ∈ On . Then, f is differentiable at a iff fi is differentiable at a for all
i. In that case, we have  
Df1 (a)
 Df2 (a) 
Df (a) = 
 
.. 
 . 
Dfm (a)

Proof. Assume that f is differentiable at a and let L = Df (a). For all i = 1, 2, . . . , m we set
Li = Πi ◦ L. Further, let f˜i (h) = fi (a + h) − fi (a) − Li h, so that

f (a + h) − f (a) − Lh = (f˜1 (h), f˜2 (h), . . . , f˜m (h))

But then, f˜i (h) ≤ ∥f (a + h) − f (a) − Lh∥ which proves that fi is differentiable at a and has
derivative Dfi (a) = Li , for all i.
Assume that fi is differentiable at a for all i and define
 
Df1 (a)
 Df2 (a) 
L=
 
.. 
 . 
Dfm (a)

Let Πi (f (a + h) − f (a) − Lh) = fi (a + h) − fi (a) − Dfi (a)h = f˜i (h), so that

f˜i (h)
lim = 0 ∀i
h→0 ∥h∥

f (a + h) − f (a) − Lh
=⇒ lim =0
h→0 ∥h∥

which proves that f is differentiable at a and has derivative Df (a) = L.

17
5.2 Further properties of differentiable functions

Note the following:

• Because of this theorem, it is enough to study differentiability of maps f : On → R.
• Let γ : O1 → Rn . Then, γ is differentiable at a ∈ O1 iff γi is differentiable at a for all i and in
that case,  ′ 
γ1 (a)
 γ2′ (a) 
Dγ(a) = γ ′ (a) =  . 
 
 .. 
γn′ (a)

This is the notion of velocity vector of a curve in Rn from elementary calculus.

5.2 Further properties of differentiable functions

Theorem 5.2.1 (Differentiability implies continuity)

Let f : On → Rm . If f is differentiable at a ∈ On , f is continuous at a.

Proof. Let f be differentiable at a. Then, for all x ∈ On ,

0 ≤ ∥f (x) − f (a)∥ ≤ ∥f (x) − f (a) − Df (a)(x − a)∥ + ∥Df (a)(x − a)∥

Taking the limit as x → a, the first term on the right goes to 0 by the definition of Df (a) and the
second term goes to zero as any linear map is continuous. Hence, by the squeeze theorem,

lim ∥f (x) − f (a)∥ = 0

x→a

which proves that f is continuous at a.

Theorem 5.2.2 (Chain rule)

Consider maps f, g such that
f g
On Om Rp

g◦f

Assume that f is differentiable at a ∈ On and g is differentiable at f (a) ∈ Om . Then g ◦ f is

differentiable at a and further

D(g ◦ f )(a) = Dg(f (a)) · Df (a)

| {z } | {z } | {z }
Rn →Rp Rm →Rp Rn →Rm

Proof. Let A = Df (a) and B = Dg(b) where b = f (a). For x, y in sufficiently small neighbourhoods
of a, b respectively, we consider the maps

rf (x) = f (x) − f (a) − A(x − a)

rg (y) = g(y) − g(b) − B(y − b)
r(x) = g(f (x)) − g(b) − BA(x − a)

By definition of the derivative,

rf (x) rg (y)
lim =0 lim =0
x→a ∥x − a∥ y→b ∥y − b∥

18
 Lecture 5

We wish to prove that

r(x)
lim =0
x→a ∥x − a∥
We have,

r(x) = g(f (x)) − g(b) − BA(x − a)

= g(f (x)) − g(b) + B(rf (x) − f (x) + f (a))
= Brf (x) + g(f (x)) − g(b) − B(f (x) − f (a))
=⇒ r(x) = Brf (x) + rg (f (x))

Now,  
Brf (x) Brf (x)
lim =B lim =0
x→a ∥x − a∥ x→a ∥x − a∥

The other term requires some more analysis. Fix ε > 0. There is δ > 0 such that ∥rg (y)∥ < ε∥y − b∥
for all y with 0 < ∥y − b∥ < δ. By continuity of f at a, there is δ1 > 0 such that ∥f (x) − f (a)∥ < δ
for all x with 0 < ∥x − a∥ < δ1 . Hence,

0 < ∥x − a∥ < δ1 =⇒ ∥rg (f (x))∥ < ε∥f (x) − f (a)∥

and so
rg (f (x))
lim =0
x→a ∥f (x) − f (a)∥
We now note,
∥f (x) − f (a)∥ = ∥rf (x) + A(x − a)∥ ≤ ∥rf (x)∥ + MA ∥x − a∥
where we get MA > 0 by Theorem 1.4.3. Hence, we finally have

r(x) rg (f (x)) ∥f (x) − f (a)∥

lim = lim =0
x→a ∥x − a∥ x→a ∥f (x) − f (a)∥ ∥x − a∥

which completes the proof.

19
5.2 Further properties of differentiable functions

20
Lecture 6

6.1 Partial Derivatives

Now, we discuss the notion of partial derivatives. This tries to treat differentiation of functions with
multiple arguments in the 1-variable setting. This construction ends up being an indispensable tool
in the computation of the Total derivative in the standard basis.

Consider a function f : On → R and a point a ∈ On , and fix 1 ≤ i ≤ n. Now, we define functions

ηi : (−ϵ, ϵ) → R such that η(t) = f (a + tei ) for all t ∈ (−ϵ, ϵ).

Definition 6.1.1 ▶ Partial Derivatives

For the given function f : On → R, the partial derivatives of f with respect to the co-ordinate
xi is given by:

∂f dηi f (a + tei ) − f (a)

fxi (a) ≡ (a) := (a) = lim if it exists.
∂xi dt t→0 t

Considering the maps hi : R → Rn such that hi (t) = a + tei , we have ηi = f ◦ hi . Thus, by Chain
Rule, if f is differentiable, all its partial derivatives exist.

6.2 Geometric Meaning

The partial derivative measures the change of a func-
tion at a point due to a particular variable, keeping all
others constant. The geometry of partial derivatives
is best visualized in 3 dimensions. Taking f : O2 → R,
we consider the surface S ⊂ R3 defined by z = f (x, y).
Let P0 = (x0 , y0 , f (x0 , y0 )) be a point on S. Then the
value fx (x0 , y0 ) (if it exists) is the slope of the tangent
to S at (x0 , y0 , f (x0 , y0 )) pointing in the positive x
direction.
Another interpretation is to consider the plane
P = {(x, y, z) | y = y0 }, and the curve C on surface
S given by C = S ∩ P. Then, fx (x0 , y0 ) is the slope
of the tangent to the curve C, in the direction of
increasing x co-ordinate.
For f : On → R with n > 2, although it becomes
harder to visualize, the interpretation remains the
same.

Figure 6.1: Geometric Meaning

21
6.3 Examples

6.3 Examples
We now lay out some interesting and instructive examples, which would illustrate some general
results about partial derivatives.

Example 6.3.1

Consider f : R2 → R given by f (x, y) = x3 + y 4 + sin(xy). Then,

∂f ∂f
= 3x2 + y cos xy and = 4x3 + x cos xy
∂x ∂y

Example 6.3.2

We know that Differentiable functions f : On → R are continuous. However, even existence of

all partial derivatives is too weak to ensure continuity.
Consider f : R2 → R given by
xy
f (x, y) = 2
x + y2
Evidently, all the partial derivatives exist, but as shown previously, this function is discontin-
uous at (0, 0).

Definition 6.3.1
A function f : O → R is called C k (O) if all the k th order partial derivatives exist and are
continuous.

6.4 Higher Order Partial Derivatives

Assume that partial derivatives of f : On → R exist in a neighbourhood of a ∈ On . Then we can
∂f
talk about the partial derivatives of ∂x i
: On → R at a. We denote:
 
∂fxi ∂ ∂f 1 ∂f ∂f
fxi xj (a) ≡ := = lim (a + hei ) − (a)
∂xj ∂xj ∂xi h→0 h ∂xi ∂xi

We can define higher order partial derivatives similarly. Please note that the order of differentiation
matters in general. For starters, of fxi xj and fxj xi , one may exist while the other may not. Also,
even if both exist, they may not necessarily be equal over the entire domain. We leave it to the
reader to find an example of the former case, and provide an example for the latter.

Example 6.4.1

Consider f : R2 → R given by
xy(x2 −y 2 )
(
x2 +y 2 (x, y) ̸= (0, 0)
f (x, y) =
0 (x, y) = (0, 0)

In this case, it is easy to show (Exercise!) that both fxy and fyx exist, but

fxy (0, 0) = 1 ̸= −1 = fyx (0, 0)

22
 Lecture 6

Example 6.4.2

However, in many well-behaved cases, we will find fxy = fyx . For instance, consider f : R2 → R
given by f (x, y) = sin(x) + ey + xy. Show that fxx = fyy = 1 over R2 .

Often, the dependence of the partial derivative on the order of differentiation is the exception rather
than the rule. We now develop a sufficient condition for fxy = fyx to hold.

6.5 Clairaut’s Theorem

Theorem 6.5.1 (Clairaut)

Let (a, b) ∈ O2 and f : O2 → R. Suppose fx , fy , fxy , and fyx all exist on O2 . If fxy and fyx
are continuous at (a, b), then fxy (a, b) = fyx (a, b).

Proof. Without loss of generality, we take (a, b) = (0, 0) ∈ O2 . As O2 is open, we choose a box
[h, 0] × [0, k] ⊂ O2 . Now, we have

∂2f fx (x, y + k) − fx (x, y)

fxy (x, y) = (x, y) = lim
∂y∂x k→0 k
1
= lim lim (f (x + h, y + k) − f (x, y + k) − f (x + h, y) + f (x, y))
k→0 h→0 hk

We define
1
F (h, k) = (f (h, k) − f (0, k) − f (h, 0) + f (0, 0))
hk
Thus, by the above result, we have

fxy (x, y) = lim lim F (h, k), and similarly

k→0 h→0
fyx (x, y) = lim lim F (h, k)
h→0 k→0

Now we proceed for the proof in earnest. Define f1 (x) = f (x, k) − f (x, 0), which is continuous on
[0, h] and differentiable on (0, h). Thus, by Lagrange’s Mean Value Theorem, there exists c1 ∈ (0, h)
(depending upon both h and k), such that, h(f1′ (c1 )) = (f1 (h) − f1 (0)), i.e.
1
∴ fx (c1 , k) − fx (c1 , 0) =
(f (h, k) − f (0, k) − f (h, 0) + f (0, 0)) = kF (h, k)
h
1
=⇒ F (h, k) = (fx (c1 , k) − fx (c1 , 0))
k
Again, define f2 (y) = fx (c1 , y), which again satisfies all conditions for the Mean Value Theorem.
Thus, there exists c2 ∈ (0, k) such that k(f2′ (c2 )) = (f2 (k) − f1 (0)), which gives, F (h, k) = fxy (c1 , c2 ).
Repeating this entire construction, we can find (c′1 , c′2 ) ∈ [0, h] × [0, k] such that F (h, k) = fyx (c′1 , c′2 ).
Thus, fxy (c1 , c2 ) = fyx (c′1 , c′2 ). But, 0 < c1 , c′1 < h and 0 < c2 , c′2 < k. Thus, as (h, k) can be made
arbitrarily small, taking (h, k) → 0, we have

lim fxy (c1 , c2 ) = lim fyx (c′1 , c′2 )

(c1 ,c2 )→0 (c′1 ,c′2 )→0

By the continuity of fxy and fyx , we have fxy (0, 0) = fyx (0, 0) .

In particular, fxy = fyx for C 2 functions over a given domain. In the next lecture, we sharpen this
result slightly, and relate the partial derivatives to the total derivative.

23
6.5 Clairaut’s Theorem

24
Lecture 7

7.1 Schwarz Theorem

In the previous lecture, we discussed the notion of partial derivatives. In general, the partial
derivatives depend on the order of differentiation. However, using Clairaut’s Theorem, we found a
necessary when fxy = fyx . Now, we conclude that discussion by the following result.

Theorem 7.1.1 (Schwarz)

Let (a, b) ∈ O2 and f : O2 → R. Suppose fx , fy , and fxy exist on O2 . If fxy is continuous at
(a, b), then fyx exists in a neighbourbood of (a, b) and fxy (a, b) = fyx (a, b).

Proof. Just as before, we take (a, b) = (0, 0). From the proof of Clairaut’s Theorem, we have
F (h, k) = fxy (c1 , c2 ) for some 0 < c1 < h and 0 < c2 < k. By continuity of fxy at (0, 0), for any
ϵ > 0 there exists hϵ , kϵ > 0, such that

| fxy (u, v) − fxy (0, 0) |< ϵ ∀ (u, v) ∈ [0, hϵ ] × [0, kϵ ]

But then | F (h, k) − fxy (0, 0) |< ϵ ∀ (u, v) ∈ [0, hϵ ] × [0, kϵ ], that is, F is continuous at (0, 0) with
limit fxy (0, 0) at (0, 0).
As fy (h, 0) exists for h sufficiently small, limk→0 F (h, k) exists for h sufficiently small. Thus, by
continuity of F at (0, 0),

lim lim F (h, k) exists and is equal to lim F (h, k)

h→0 k→0 (h,k)→(0,0)

Thus, fyx (0, 0) exists and is equal to fxy (0, 0)

This is a slightly more useful version of Clairaut’s Theorem. However, in many applications (say,
partial differential equations), we work with C 2 or even C ∞ functions, in which case both of these
hold automatically.
Exercise: Formulate and prove a similar result for higher order derivatives. In particular, provide a
sufficient condition for f : On → R so that
∂nf ∂nf
=
∂xi1 ∂xi2 . . . ∂xim ∂xσ(i1 ) ∂xσ(i2 ) . . . ∂xσ(im )

over On for any permutation σ of the elements {i1 , i2 , . . . , in }.

7.2 Partial and Total Derivatives

We will now see that the partial derivatives provide an effective way of proving the existence and
computing the total derivatives of a function f : On → Rm . In this lecture and the next, we will
develop the relations between partial and total derivatives by a series of results.

25
7.2 Partial and Total Derivatives

Definition 7.2.1 ▶ Jacobian Matrix

∂fi
For a function f = (f1 , f2 , . . . , fm ) : On → Rm , if all the partial derivatives ∂xj at a ∈ On ,
we define the Jacobian of the function at a by the m × n matrix,
 
∂fi
Jf (a) = (a)
∂xj m×n

Theorem 7.2.1
Consider a function f = (f1 , f2 , . . . , fm ) : On → Rm differentiable at a ∈ On . Then all the
∂fi
partial derivatives ∂xj
exist at a. In particular, for f differentiable at a, we have,
 
∂fi
(Df )(a) = Jf (a) = (a)
∂xj m×n

Proof. Without loss of generality, we take m = 1, and let a = (a1 , a2 , . . . , an ). Fix an arbitrary index
i ∈ {1, 2, . . . , n}. We define ηi : [ai − ϵ, ai + ϵ] → Rn , defined by

ηi (t) = (a1 , . . . , ai−1 , t, at+1 , . . . , an ) = a + (t − ai )ei

As On is open and ηi is continuous, we can find ϵ small such that f ([[ai − ϵ, ai + ϵ]]) ⊆ On ⊆ Rn .
Evidently, ηi is differentiable and (Dηi ) = [0, . . . , 1, . . . , 0]t = eti over [ai − ϵ, ai + ϵ]. Now, by the
definition of partial derivatives, D(f ◦ ηi )(ai ) = fxi (a).
Again, by chain rule, as f is differentiable at a, D(f ◦ ηi )(ai ) = fxi (a) exists, and

D(f ◦ ηi )(ai ) = Df (ηi (ai )) · Dηi (ai )

=⇒ fxi (a) = Df (a) · eti = [Df (a)]i

As the index i was arbitrary to begin with, this completes the proof.

This theorem proves that differentiability of a function implies the existence of its partial derivatives,
and gives the form of the derivative in the standard basis. But it is often quite elaborate and laborious
to prove that a function is differentiable, whereas computation of the partial derivatives is much
more straightforward. In the next lecture, we formulate a sufficient condition for differentiability of a
function based on its partial derivatives.

26
Lecture 8

8.1 A kind of converse of Theorem 7.2.1

As we have seen in previous lecture, the differentiability of a function gives an explicit expression
for derivative with the existence of partials. In this lecture, we will prove a sufficient condition for
differentiability based on its partials, which will be our final reduction for derivatives.

Theorem 8.1.1 (Final Reduction)

∂fi
Let f : On → Rm and a ∈ On . Suppose, all partial derivatives ∂xj exists on On and
continuous at a ∈ On . Then,
 
∂fi
Df (a) = Jf (a) = (a)
∂xj m×n

Proof. Without loss of generality, we take a = (0, . . . , 0) ∈ On and m = 1.

Let’s do some back calculation: We already “know”,

n
X ∂f
(0) ∀ h ∈ Rn


L = Jf (a) = fx1 (0) · · · fxn (0) and Lh = hi
i=1
∂xi

which we use in the following claim,

Claim

1
|f (h) − f (0) − Lh| → 0 as h → 0
∥h∥

Proof. Simply calculating,

n
1 1 X ∂f
|f (h) − f (0) − Lh| = f (h) − f (0) − hi (0)
∥h∥ ∥h∥ i=1
∂xi

For every i, we define, ĥi = (h1 , . . . , hi , 0, . . . , 0) and ĥ0 = 0. Then,

| {z }
n−r
    
f (h) − f (0) = f (ĥ1 ) − f (ĥ0 ) + f (ĥ2 ) − f (ĥ1 ) + · · · + f (ĥn ) − f (ĥn−1 )
n 
X 
= f (ĥi ) − f (ĥi−1 )
i=1

27
8.2 Examples

which implies,
n n  
X ∂f X ∂f
f (h) − f (0) − hi (0) = f (ĥi ) − f (ĥi−1 ) − hi (0)
i=1
∂xi i=1
∂xi

n
X ∂f   ∂f
= hi ĥi−1 + ci ei − hi (0)
i=1
∂xi ∂xi
| {z }
By MVT, as explained below

For a fixed h, we fix i ∈ [n]. And we consider the map, ηi : (hi − ϵ, hi + ϵ) → R

ηi

t ĥi−1 + tei f (ĥi−1 + tei )

defined by, ηi (t) = f (ĥi−1 + tei ). Clearly, ηi is differentiable on (0, hi ) and continuous on [0, hi ].
Then, by Mean Value Theorem,

ηi (hi ) − ηi (0) = ηi′ (ci )hi = fxi (ĥi−1 + ci ei )hi (for some ci ∈ (0, hi ))
| {z } | {z }
f (ĥi ) f (ĥi−1 )

Now, observe that, as h → 0, ĥi−1 + ci ei → 0 which in turn implies, fxi (ĥi−1 + ci ei ) → fxi (0).
Therefore,
n
1 1 X ∂f   ∂f
|f (h) − f (0) − Lh| = hi ĥi−1 + ci ei − hi (0)
∥h∥ ∥h∥ i=1 ∂xi ∂xi
n
1 X ∂f   ∂f
≤ |hi | ĥi−1 + ci ei − (0) (Triangle inequality)
∥h∥ i=1 ∂xi ∂xi
n
X ∂f   ∂f
≤ ĥi−1 + ci ei − (0) (as ∥h∥ ≥ |hi | ∀ i)
i=1
∂xi ∂xi
−→ 0 as h → 0

And, this completes the proof.

With Theorem 8.1.1, computation of derivative is much easier when we are in favorable situation.
Note that,
∂fi
(i) If f is differentiable at a then ∂xj (a) exists for all i, j and Df (a) = Jf (a).

∂fi
(ii) If ∂xj is continuous at a then f is differentiable and Df (a) = Jf (a).

The gap between (i) and (ii) is the continuity of partials, which is removable.

8.2 Examples
We conclude the lecture with some instructive examples.

28
 Lecture 8

Example 8.2.1 (Differentiable but discontinuous)

Take,
  
 x2 + y 2
sin √ 1 , (x, y) ̸= 0
f (x, y) = x2 +y 2
0, (x, y) = 0


Then,
!
2 2 1
|f (x, y) − f (0, 0)| = x + y sin p
x2 + y 2
2
≤ (x2 + y 2 ) = ∥(x, y)∥

implies that f is continuous at (0, 0). For all (x, y) ̸= (0, 0), the partials of f ,
! !
1 x 1
fx (x, y) = 2x sin p −p cos p
x2 + y 2 x2 + y 2 x2 + y 2
! !
1 y 1
fy (x, y) = 2y sin p −p cos p
x2 + y 2 x2 + y 2 x2 + y 2

And at (0, 0),

f (t, 0) − f (0, 0)
fx (0, 0) = lim =0
t→0 t
f (0, t) − f (0, 0)
fy (0, 0) = lim =0
t→0 t
Also,
   
1
h p 1
√ f (h, k) − f (0, 0) − 0 0 = h2 + k 2 sin √ ≤ ∥(h, k)∥
h2 + k 2 k h2 + h2


which shows that, f is differentiable at (0, 0) and Df (0, 0) = 0 0 . But, fx and fy are not
continuous at (0, 0)!

So, even if a function is differentiable at some point, its partials may still not be continuous there!

Example 8.2.2 (Exercise)

Take,
( 4
y


x 3 sin x , x ̸= 0
f (x, y) =
0, x=0

• Show that,
1. f is differentiable on R2 .

2. fx and fy exist and continuous on O2 = (x, y) ∈ R2 : x ̸= 0 .
3. fx is not continuous at (0, y) for all y ̸= 0.

• Discuss the nature of continuity of f at the origin.

29
8.2 Examples

Example 8.2.3

Let, f : R3 → R4 be defined by,

f (x, y) = (x + 2y + 3z, xyz, cos x, sin x)

Then, the at (x, y, z)

 
1 2 3
 yz zx xy 
Jf (x, y, z) = 
 − sin x

0 0
cos x 0 0

which has every entry continuous, thus

Jf (x, y, z) = Df (x, y, z)

30
Lecture 9

9.1 Directional Derivatives

We now introduce another extension of one-dimensional derivative, called the Directional Derivative.

Definition 9.1.1 ▶ Directional Derivative

Let u ∈ Rn (indicates a direction) be a unit vector and f : On → R be a scalar-valued function.
Take, a ∈ On , the directional derivative of f at a in the direction of u is defined as

f (a + tu) − f (a)
(Du f ) (a) := lim
t→0 t
the rate of change of f at a in the direction of u, provided the limit exists.

∂f
Note that (Dei f ) (a) = ∂xi (a), i.e., the partial derivatives are directional derivatives along the
standard basis vectors.

Now, consider the following

ϕ f
t a + tu f (a + tu)

η(t)=f (a+tu)

Take, η : (−ε, ε) → R as η(t) := f (a + tu). If f is differentiable at Figure 9.1: t 7→ a + tu

a then η is differentiable at 0 (by chain rule) iff f has directional
derivative at a along u and,

η ′ (0) = (Du f ) (a)

= Df (ϕ(0)) · (Dϕ)(0)
| {z }
u
 
u1
(a)  ... 
 
∂f ∂f
= (a) · · ·
 
∂xi ∂xn
un

So,
n
X ∂f
η ′ (0) = ui (a) = (Du f ) (a)
i=1
∂xi

31
9.2 Gradient

9.2 Gradient

Definition 9.2.1 ▶ Gradient

Given a scalar-valued function (a.k.a., scalar field) f : On → R and x ∈ On , the Gradient of f
at x is defined as
∇f (x) := ⟨fx1 (x), . . . , fxn (x)⟩
the dual of the total derivative, i.e., Df (x)t , provided that all the partials fxi exists at x.

Observe that, for a differentiable function f : On → R at a

(Du f ) (a) = (∇f ) (a) · u

= ∥(∇f ) (a)∥ cos θu (where, θu is angle between (∇f ) (a) and u)

which tells us, The steepest slope is achieved when θu ∈ {0, π}, i.e., when u points along or opposite
(∇f ) (a)
to the direction of (∇f ) (a) that means max (Du f ) (a) is attained at u = (direction
∥(∇f ) (a)∥
| {z }
provided ∥(∇f )(a)∦=0
of the steepest slope). Hence, we get the following theorem,

Theorem 9.2.1
Let f : On → R be a differentiable function at a ∈ On and suppose (∇f ) (a) ̸= 0 then the
vector (∇f ) (a) points in the direction of the greatest increment of f at a with the greatest
rate ∥(∇f ) (a)∥

9.3 Examples
Example 9.3.1

Find the directional derivative of f (x, y, z) = x2 yz along

D ⟨1, 1, −1⟩E at a = (1, 1, 0).
⟨1,1,−1⟩ 1
Solution. We have the unit vector u = ∥⟨1,1,−1⟩∥ = √3 , √13 , − √13 . So,

(Du f ) (a) = (∇f ) (a) · u

√1
 
3
x2 z x2 y  √1 


= 2xyz 3 
− √13

Therefore, (Du f ) (1, 1, 0) = − √13 and hence maximum value of (Du f ) (1, 1, 0) is
∥(∇f ) (1, 1, 0)∥ along the unit vector ⟨0, 0, 1⟩.

Example 9.3.2

Take,
x2 y
(
x2 +y 2 (x, y) = (0, 0)
f (x, y) =
0 (x, y) ̸= (0, 0)
x2 y
then |f (x, y) − f (0, 0)| = x2 +y 2 ≤ |y| ≤ ∥(x, y)∥ implies that f is continuous at (0, 0). Now,

32
 Lecture 9

fix u = ⟨u1 , u2 ⟩ with ∥u∥ = 1. We get,

f (tu) − 0
(Du f ) (0, 0) = lim
t→0 t
1 tu21 u2
= lim ·
t→0 t 1
2
= u1 u2 ̸= 0 (Because, u is an unit vector)
 
u1
If we assume f to be differentiable then (∇f ) (0, 0)· = 0 ̸= u21 u2 which is a contradiction!
u2

Example 9.3.2 shows that, the existence of all partial and directional derivatives at a point fails to
imply differentiability at that point.

33
9.3 Examples

34
Lecture 10

Example 10.0.3 (Exercise)

Take the function f : R2 → R defined as,

x3 y
(
x2 +y 2 (x, y) ̸= (0, 0)
f (x, y) =
0 (x, y) = (0, 0)

(i) Prove that, f is continuous at (0, 0).

(ii) Find (Du f )(0, 0) ∀ u.

(iii) Prove that, f is not differentiable.

10.1 Extension of MVT to Several Variables

In Analysis I, you learned about the Mean Value Theorem (MVT) for functions of a single variable.
Now, we extend this concept to several variables in the context of multivariable calculus.

Theorem 10.1.1 (Multivariate MVT)

Let On ⊆ Rn be an open and convex set, and let f : On → R be a differentiable function. For
any two points a, b ∈ On define the line segment

La,b := {tb + (1 − t)a : t ∈ [0, 1]}

Then, there exists a point c ∈ La,b such that,

f (b) − f (a) = (∇f )(c) · (b − a) = ⟨fx1 (c), . . . , fxn (c)⟩ · ⟨(b1 − a1 ), . . . , (bn − an )⟩

Proof. We consider the function η : [0, 1] → On

η f
[0, 1] On R

f ◦η

defined by η(t) = (1 − t)a + tb. This function is differentiable, and its derivative is
 
b1 − a1
η ′ (t) =  ... 
 

bn − an

35
10.2 More Partials and Chain Rules

By applying the standard Mean Value Theorem to the composition f ◦ η, there exists t0 ∈ (0, 1) such
that,
f (η(1)) − f (η(0)) = (f ◦ η)′ (t0 ) = Df (η(t0 )) · Dη(t0 )
Expanding the dot product, we have,
 
b1 − a 1
 b2 − a 2 

 
f (b) − f (a) = fx1 (η(t0 )) fx2 (η(t0 )) · · · fxn (η(t0 ))  . 
 .. 
bn − a n

Simplifying further, we obtain,

f (b) − f (a) = ⟨fx1 (η(t0 )), · · · , fxn (η(t0 ))⟩ · ⟨(b1 − a1 ), · · · , (bn − an )⟩

This expression can be rewritten as,

f (b) − f (a) = (∇f )(η(t0 )) · (b − a)

Hence, there exists c = η(t0 ) ∈ La,b such that f (b) − f (a) = (∇f )(c) · (b − a).

10.2 More Partials and Chain Rules

In this section, we further explore the chain rule for differentiable functions of several variables.
Consider two functions f and g as following,

f g
On Om Rp

Assuming that f is differentiable at a ∈ On and g is differentiable at b = f (a) ∈ Om the chain rule

states that the derivative of the composite function g ◦ f is given by,

D(g ◦ f )(a) = (Dg)(f (a)) · (Df )(a)

| {z }
Rn →Rp
= Dg(b) · Df (a)
| {z } | {z }
Rm →Rp Rn →Rm

This can be expressed in matrix form as,

Jg◦f (a)p×n = Jg (f (a))p×m · Jf (a)m×n (10.1)

Moreover, if we consider the function components in each individual coordinates as

• g ◦ f = ((g ◦ f )1 , (g ◦ f )2 , . . . , (g ◦ f )p )

• g = (g1 , g2 , . . . , gp )

• f = (f1 , f2 , . . . , fm ).

Then, the (i, j)th entry of both sides of (10.1) would become,
 ∂f1 
∂xj(a)
∂f2
∂(g ◦ f )i   ∂xj(a) 
∂gi ∂gi ∂gi
 
(a) = ∂y (f (a)) ∂y2 (f (a)) ··· ∂ym (f (a))  . 

∂xj  .. 
1 
∂fm
∂xj (a)

36
 Lecture 10

which is our familiar chain rule for partial derivatives,

m
∂ (g ◦ f )i X ∂gi ∂fk
(a) = (b) · (a)
∂xj ∂yk ∂xj
k=1

where, b = f (a).

Furthermore, if we define yk = fk (x1 , x2 , . . . , xn ) and zi = gi (y1 , y2 , . . . , ym ) we can express the

chain rule for partial derivatives as,

m
∂zi X ∂zi ∂yk
(x) = (y) · (x)
∂xj ∂yk ∂xj
k=1

Remark. In addition, the chain rule can be applied in the context of a composition map with respect
to a parameter t. For functions f : On → R, η : O1 → On and z = f ◦ η as shown below,

η f
O1 On R

η f
t (x1 (t), . . . , xn (t)) f (x1 (t), . . . , xn (t))

z=f ◦η

the chain rule states,

n
dz X ∂f dxk
= ·
dt ∂xk dt
k=1

If we treat f as a function of t, the same can be written as,

n
df X ∂f dxk
= ·
dt ∂xk dt
k=1

37
10.2 More Partials and Chain Rules

38
Lecture 11

11.1 Chain Rule

We will begin by recalling some results from the previous lecture.

g◦f

On f
Om g Rp

If we are given two differentiable function f : On → Om and g : Om → Rp , then g ◦ f is also

differentiable. We also derived how to compute Dg◦f by chain rule as following,

Dg◦f (a) = Dg (f (a)).Df (a)

Now, comparing the (i, j)th element, we get,
m
∂(g ◦ f )i (a) X ∂gi (b) ∂fk (a)
= ·
∂xj ∂yk ∂xj
k=1

where b = f (a). This can be rewritten in a slightly more suggestive form by introducing new variables,

yk = fk (x1 , . . . , xn )
zi = gi (y1 , . . . , ym )

Then, since (g ◦ f )i = gi ◦ f , the equation above can be written as,

m
∂zi X ∂zi ∂yk
= ·
∂xj ∂yk ∂xj
k=1

This form of the chain rule is reminiscent of the one-variable chain rule.

Example 11.1.1
df
Let, f (x, y, z) = xy 2 z and x = t, y = et , z = 1 + t, we want to calculate dt in two ways.

First, we can write f as a function of t,

f (x, y, z) = t(et )2 (1 + t)
= (t + t2 )e2t
Hence, we have,
df d
t + t2 e2t


=
dt dt
= (1 + 2t)e2t + 2(t + t2 )e2t

39
11.2 Laplacian

= (2t2 + 4t + 1)e2t

Alternatively, if we apply the chain rule, we obtain,

df ∂f dx ∂f dy ∂f dz
= + +
dt ∂x dt ∂y dt ∂z dt
= y z · 1 + 2xyz · e + xy 2 · 1
2 t

= e2t (1 + t) + 2t(1 + t)et .et + te2t

= e2t (1 + t + 2t + 2t2 + t)
= (2t2 + 4t + 1)e2t

As we can see, both methods yield the same result!

11.2 Laplacian
The Laplacian operator plays a fundamental role in analyzing the behavior of functions and fields in
multidimensional spaces. It quantifies the overall rate of change and spatial variations of a function,
providing valuable insights into its properties and behavior.

Definition 11.2.1 ▶ Laplacian

f : On → R be a function. Then the Laplacian of f is defined as,
n
X ∂2f
∆f =
i=1
∂x2i

Observe that,
n
X ∂2f
∆f =
i=1
∂x2i
   
∂ ∂ ∂f ∂f
= ,..., . ,...,
∂x1 ∂xn ∂x1 ∂xn
= ∇ · ∇f

Hence, Laplacian can be written as, ∆f = ∇ · ∇f = ∇2 f .

Laplacian in Polar Coordinate

Let, f be a twice differentiable function f : R2 → R. We can express f (x, y) in polar coordinates as
a function of (r, θ) by substituting, x = r cos θ and y = r sin θ. Now, observe the following partial
derivatives,

∂x ∂x
= cos θ, = −r sin θ
∂r ∂θ
∂y ∂y
= sin θ, = r cos θ
∂r ∂θ
We want to express fxx and fyy in terms of partial derivatives of f in polar coordinates. Notice that,

∂f ∂f ∂x ∂f ∂y
= +
∂r ∂x ∂r ∂y ∂r

40
 Lecture 11

∂f ∂f ∂f
i.e., = cos θ + sin θ
∂r ∂x ∂y
Differentiating once more with respect to r, we have,
∂2f
 
∂ ∂f ∂f
= cos θ + sin θ
∂r2 ∂r ∂x ∂y
   
∂ ∂f ∂ ∂f
= cos θ + sin θ
∂r ∂x ∂r ∂y
 2
∂ 2 f ∂y
 2
∂ f ∂x ∂ 2 f ∂y
 
∂ f ∂x
= cos θ + + sin θ +
∂x2 ∂r ∂y∂x ∂r ∂x∂y ∂r ∂y 2 ∂r
 2
∂2f ∂2f ∂2f
  
∂ f
= cos θ cos θ + sin θ + sin θ cos θ + sin θ
∂x2 ∂y∂x ∂x∂y ∂y 2
= cos θ [cos θfxx + sin θfxy ] + sin θ [cos θfxy + sin θfy ]

Hence, we get,
∂2f
= cos2 θfxx + sin2 θfy + sin 2θfxy
∂r2
∂2f
Similarly, we can find the expression for ∂θ 2 ,

∂2f
= −r (cos θfx + sin θfy ) + r2 sin2 θfxx + r2 cos2 θfy − r2 sin 2θfxy


∂θ2
Combining the above two result we can write,

∂2f 1 ∂f 1 ∂2f
∆f = fxx + fyy = + · + ·
∂r2 r ∂r r2 ∂θ2

Example 11.2.1 (Writing Laplacian in New coordinate)

Let, z = z(u, v) where,

u(x, y) = x2 y and v(x, y) = 3x + 2y
We want to express the Laplacian with respect to u and v. Starting with the given coordinates,
∂u ∂u
= 2xy, = x2
∂x ∂y
∂v ∂v
= 3, =2
∂x ∂y
∂z
We can find ∂x using the chain rule,
∂z ∂z ∂u ∂z ∂v
= +
∂x ∂u ∂x ∂v ∂x
∂z ∂z ∂z
=⇒ = 2xy +3
∂x ∂u ∂v

Differentiating once more with respect to x, we have,

∂2z
 
∂ ∂z ∂z
= 2xy + 3
∂x2 ∂x ∂u ∂v
   
∂z ∂ ∂z ∂ ∂z
= 2y + 2xy +3
∂u ∂x ∂u ∂x ∂v

41
11.3 Extrema of a function

           
∂z ∂ ∂z ∂u ∂ ∂z ∂v ∂ ∂z ∂u ∂ ∂z ∂v
= 2y + 2xy + +3 +
∂u ∂u ∂u ∂x ∂v ∂u ∂x ∂u ∂v ∂x ∂v ∂v ∂x
 2
∂ 2 z ∂v
 2
∂ z ∂u ∂ 2 z ∂v
 
∂z ∂ z ∂u
= 2y + 2xy + + 3 +
∂u ∂u2 ∂x ∂v∂u ∂x ∂u∂v ∂x ∂v 2 ∂x
∂2z ∂2z ∂2z ∂2z
   
∂z 1
= 2y + 2xy 2xy 2 + 3 +3 · +3 2
∂u ∂u ∂v∂u 2xy ∂u∂v ∂v

Hence, we get,
∂2z
= 2yzu + 4x2 y 2 zuu + 6xyzuv + 6xyzvu + 9zvv
∂x2

Exercise. Find zyy , zyx , zxy and check if zxy = zyx .

11.3 Extrema of a function

Finding the extrema of a function is crucial in calculus, allowing us to identify maximum and minimum
points and to relate the structure of functions. We will now extend this concept to functions of
several variables.

Definition 11.3.1 ▶ Extrema

Let, a is an interior point of S ⊆ Rn and f : S → R be a function.

• f attains a local maximum at a if there exists an open neighborhood On of a such

that, f (a) ≥ f (x) ∀ x ∈ On .
• Similarly, f attains a local minimum at a if there exists an open neighborhood On of
a such that, f (a) ≤ f (x) ∀ x ∈ On .

Any point at which f attains a local(global) maxima (or minima) is called extremum point of
that function. In plural, it is called Extrema.

Definition 11.3.2 ▶ Critical Point or Stationary Point

Let, f : S(⊆ Rn ) → R be a function and a ∈ On ⊆ S. We say that a is a critical point or
stationary point. If
(∇f ) (a) = 0
∂f
Or, equivalently all the partial derivatives ∂xi are zero.

Theorem 11.3.1
Let, f : On → R is differentiable at a ∈ On . If a is a local extremum, then

(∇f ) (a) = 0

∂f
Proof. Fix i ∈ {1, 2, . . . , n}. We want to show ∂x i
= 0. For this set, ϕi : (ai − ϵ, ai + ϵ) → R defined
by
ϕi (t) = f (a1 , . . . , ai−1 , t, ai+1 , . . . , an )

Notice that, dϕ
dt = fxi (a). Since a is local extremum of f , we can say that ai is a local extremum of
i

∂f (a)
ϕi . So, dϕ
dt = 0, which means, ∂xi = 0. We can do this for all i and hence, (∇f )(a) = 0.
i

42
 Lecture 11

Question. When we did calculation for local extremum for the functions f with one variable, we
used to evaluate the stationary points by calculating, f ′ (x) = 0. Then we used to check the second
derivative in order to know whether the stationary point is local minima or maxima or saddle point.
For multivariate case also, we need 2nd order derivative to know the behavior of the stationary point.
Now what could be 2nd order total derivative?
Answer. For this purpose we will introduce Hessian Matrix in next class.

43
11.3 Extrema of a function

44
Lecture 12

12.1 Hessian Matrix

We start by defining Hessian matrix, which is a natural extension of the concept of the second
derivative in higher dimensions, allowing us to analyze the rate of change and curvature of a function
in multiple directions simultaneously.

Definition 12.1.1 ▶ Hessian

Suppose f : On → R is C 2 at a ∈ On . The Hessian of f at a is defined as the matrix,
 2 
∂ f
Hf (a) = (a)
∂xi ∂xj n×n

In explicit notation, it has the following form,

 ∂2f ∂2f ∂2f

∂x21 ∂x1 ∂x2 ... ∂x1 ∂xn
2
∂2f ∂2f 
 ∂ f 

 ∂x ∂x ∂x22
... ∂x2 ∂xn 
Hf =  2. 1 .. .. .. 
 . .
 . . .


∂2f ∂2f ∂2f
∂xn ∂x1 ∂xn ∂x2 ... ∂x2n

It is important to note that for any function f that is twice continuously differentiable (f ∈ C 2 ), its
Hessian matrix Hf is symmetric, meaning that Hf = Hf t .

Example 12.1.1

Let, f : R2 → R be a function defined by f (x, y) = sin2 x + x2 y + y 2 . Then,

Df = sin 2x + 2xy x2 + 2y : R2 → R is linear.

The gradient is given by,

∇f = sin 2x + 2xy, x2 + 2y ∈ R2

And the Hessian matrix Hf is,

   
fxx fxy 2(cos 2x + y) 2x
∵ f ∈ C2
 
Hf = =
fxy fyy 2x 2

Now let’s introduce some notation. Given A = (aij )n×n ∈ Mn (R) and x ∈ Rn we denote QA (x)
by,

QA (x) = xt Ax = ⟨Ax, x⟩Rn

45
12.2 Positive Definite, Negative Definite, Semi Definite Matrices

  
a11 a12 ··· a1n x1
a21

 a22 ··· a2n   x2 
= x1 x2 ··· xn  .
 
.. .. ..   .. 
 .. . . .  . 
a31 a32 ··· a3n xn
n X
X n
= aij xi xj
i=1 j=1

Definition 12.1.2 ▶ Quadratic Form

A function f : Rn → R is called a Quadratic Form if it can be expressed as f (x) = QA (x)
for all x and for some symmetric A ∈ Mn (R)

It is important to note that a Quadratic Form represents a homogeneous polynomial of degree 2.

For instance, in the case of a bivariate polynomial p(x, y) = a11 x2 + a22 y 2 + a12 xy, the matrix
a11 21 a12
 
A= 1
2 a12 a22
corresponds to the quadratic form p(x, y) = QA (x, y), capturing the essential quadratic behavior of p.

12.2 Positive Definite, Negative Definite, Semi Definite Matrices

Definition 12.2.1 ▶ Positive Definite, Negative Definite, Semi Definite

• A symmetric matrix A ∈ Mn (R) is called Positive Definite if

⟨Ax, x⟩ > 0 ∀ x ∈ Rn \ {0}

• A symmetric matrix A ∈ Mn (R) is called Negative Definite if

⟨Ax, x⟩ < 0 ∀ x ∈ Rn \ {0}

• A symmetric matrix A ∈ Mn (R) is called Semi Definite if

⟨Ax, x⟩ ≥ 0 ∀ x ∈ Rn \ {0}

Example 12.2.1

1. In is positive definite because for any vector x ∈ Rn \ {0} the inner product
2
⟨In x, x⟩ = ∥x∥ > 0

is strictly positive.
2. For any matrix A ∈ Mn (R), if there exists a matrix B ∈ Mn (R) such that A = B t B,
we can examine the inner product ⟨Ax, x⟩ for any x ∈ Rn \ {0}.
Let’s compute this inner product

⟨Ax, x⟩ = ⟨B t Bx, x⟩
= xt B t Bx
= (Bx)t (Bx)
2
= ∥Bx∥ for all x ∈ Rn \ {0}.

46
 Lecture 12

Therefore, we conclude that ⟨Ax, x⟩ ≥ 0 for all x ∈ Rn \ {0}.

Moreover, if ⟨Ax, x⟩ = 0, then Bx = 0, which implies that x is in the kernel of B.
Conversely, if A is positive definite, there is no non-zero vector x such that ⟨Ax, x⟩ = 0.
This implies that the columns of B are linearly independent.
In summary, when A can be written as A = B t B for some matrix B, we can conclude
that A is positive semi-definite (and the converse also holds).

3. Consider the matrix  

1 0
A=
0 −1
We can compute the quadratic form associated with A as QA = x21 − x22 .
By examining this expression, we observe that it is the difference between the squares
of two variables. This indicates that the sign of QA can change depending on the values
of x1 and x2 . Consequently, the matrix A is considered indefinite.

4. Consider the matrix  

1 0
0 0
To determine the definiteness of A, we can

compute the quadratic form associated with
A as QA (x) = x21 ≥ 0 for all x = x1 x2 . This indicates that the matrix A is positive
semi-definite.

Consider a positive definite matrix A. For any non-zero vector h, we have,

⟨Ah, h⟩ = ∥Ah∥∥h∥ cos θ > 0
where θ is the angle between vectors Ah and h. Since the cosine of any angle θ in the interval [0, π/2)
is positive, we can conclude that,
π
cos θ > 0 =⇒ 0 ≤ θ <
2
π
Thus, for any positive definite matrix A, the angle θ between Ah and h satisfies 0 ≤ θ < 2.

It is worth noting that classifying positive definite matrices becomes more challenging for higher
dimensions (n > 2). However, for 2 × 2 matrices, we can easily determine it from the next theorem.

Theorem 12.2.1
 
a b
Let A = ∈ M2 (R) be symmetric. Then,
b c

(i) A is Positive Definite ⇐⇒ a > 0 and ac − b2 > 0

(ii) A is Negative Definite ⇐⇒ a < 0 and ac − b2 > 0

(iii) A is Indefinite ⇐⇒ ac − b2 < 0

Proof. We have ⟨Ah, h⟩ = ht Ah for any vector h. Now, consider a non-zero vector x = (x1 , x2 ) ∈
R2 \{(0, 0)} with x2 ̸= 0. Without loss of generality, we can scale x as x = (x, 1) for some x ∈ R.
Then, we have,
⟨Ax, x⟩ = ax2 + 2bx + c > 0 ∀ x ∈ R


If x2 = 0, we can choose x = 1 0 (after scaling). Then, ⟨Ax, x⟩ = a. Therefore, we can summarize
the conditions as follows,
A is Positive Definite

47
12.2 Positive Definite, Negative Definite, Semi Definite Matrices

⇐⇒ a > 0 and ax2 + bx + c > 0 ∀ x ∈ R

⇐⇒ a > 0 and (2b)2 − 4ac < 0
⇐⇒ a > 0 and ac − b2 > 0

Similarly, we can derive the conditions for negative definite and indefinite matrices as,

A is Negative Definite
⇐⇒ a < 0 and ax2 + bx + c < 0 ∀ x ∈ R
⇐⇒ a < 0 and (2b)2 − 4ac < 0
⇐⇒ a < 0 and ac − b2 > 0

And finally for indefinite ones,

A is Indefinite
⇐⇒ ax2 + bx + c < 0 for some x ∈ R
and ax2 + bx + c > 0 for some x ∈ R
⇐⇒ (2b)2 − 4ac > 0
⇐⇒ ac − b2 < 0

Lemma 12.2.1
 
a1 (x) a2 (x)
Let, a ∈ On , A(x) = . Suppose, A is continuous at a (i.e., ai ’s are continuous
a2 (x) a3 (x)
at a). Then, A is Positive Definite at a would imply that A is Positive Definite in a
neighborhood of a.

Proof. A(a) is Positive Definite, i.e., a1 (a) > 0 and a1 (a)a3 (a)−a2 2 (a) > 0. As a1 (x) and a1 (x)a3 (x)−
a22 (x) are polynomial of continuous functions, we can find an ϵ > 0 such that both are positive in
Bϵ (a).

48
Lecture 13

13.1 Taylor’s Theorem

Recall, Taylor’s theorem for one variable.

Definition 13.1.1 ▶ Taylor’s Polynomial

Let, f : O1 → R be C k (k ∈ N). Then for all h such that a + h ∈ O1,
k
X f (n) (a) n
pa,k (a + h) = h
n=0
n!

is called the Taylor’s Polynomial of f around a.

Question. “Is f (x) ≈ pa,k (x), for x close to a”?

We have,
k
X f (n) (a)
pa,k (x) = (x − a)n
n=0
n!
Take, f (x) − pa,k (x) = ra,k (x)

Theorem 13.1.1 (Taylor’s Theorem)

Let f : O1 → R be C k+1 . Then, f (x) = pa,k (x) + ra,k (x) where,

f k+1 (c)
ra,k = (x − a)k+1
(k + 1)!

for some c in between a and x ∈ O1 .

We introduce the following notation for the sake of clarity in the multivariate Taylor expansion. Let
Z
α = (α1 , α2 , · · · , αn ) ∈ n≥0 , and define
Pn
• |α| = i=1 αi
• α! = α1 !α2 ! · · · αn ! (product of coordinate-wise factorials)
∂ |α|
• ∂α = α1
(αth derivative)
∂x1 · · · ∂xn αn
n
X ∂
• ∇·h= hi
i=1
∂xi
The last definition when iterated gives,
X m!
(∇ · h)m = hα ∂ α
α!
|α|=m

49
13.1 Taylor’s Theorem

Theorem 13.1.2 (Taylor’s Theorem in Multivariate Case)

Let, f : On → R be a C k+1 function and assume On is convex. If h, a + h ∈ On , then
X 1
f (a + h) = (∂ α f )(a)hα + ra,k (h)
α!
|α|≤k

where,
X 1 α
ra,k (h) = (∂ f )(a + ch)hα for some c ∈ (0, 1)
α!
|α|=k+1

Proof. Define, η : [0, 1] → R as η(t) = f (a + th)

t a + th f (a + th)
η

which implies η is a C k+1 function around 0.

∴ η ′ (t) = ∇f (a + th) · h = (∇ · h)f (a + th)

Claim

η (m) (t) = (∇ · h)m f (a + th) ∀ m ∈ {0, 1, . . . , k + 1}

Proof. The first derivative of η,

n
X
η ′ (t) = ∇f (a + th) · h = fxi (a + th)hi
i=1

which we use to compute the second derivative,

n
!
d X
η ′′ (t) = fxi (a + th)hi
dt i=1
n
X d
= (fxi (a + th)hi )
i=1
dt
n
X n
X
= hi fxi xj (a + th)hj (Chain rule of partials)
i=1 j=1
Xn
= hi hj fxi xj (a + th)
i,j=1

= (∇ · h)2 f (a + th)

Proceeding with induction on the order of the derivative, we get η (m) (t) = (∇ · h)m f (a + th) for all
0 ≤ m ≤ k + 1 which is our claim.

By one-variable Taylor’s Theorem,

η(1) = p0,k (1) + r0,k (c) for some c ∈ (0, 1) (13.1)

with
η ′ (0) η (k) (0)
p0,k (1) = η(0) + + ··· + (13.2)
1! k!

50
 Lecture 13

and
η (k+1) (c)
r0,k (c) = (13.3)
(k + 1)!
Substituting η (m) (t) in (13.1) we have,
X 1
f (a + h) = (∂ α f )(a)hα + ra,k (h)
α!
|α|≤k

We note that, in particular, if f : O2 → R is a C 2 function then we have,

1
f (a + h) = f (a) + ∇f (a) · h + ht Hf (a + ch)h (13.4)
2
where,  
fxx (a) fxy (a)
Hf (a) =
fxy (a) fyy (a)
and c ∈ (0, 1).

Theorem 13.1.3 (Extremum)

Let f : O2 → R be a C 2 function such that Df (a) = 0. We write
 
fxx (a) fxy (a)
Hf (a) =
fxy (a) fyy (a)

Then,
(i) f (a) is a local maximum if fxx (a) < 0 and det(Hf (a)) > 0
(ii) f (a) is a local minimum if fxx (a) > 0 and det(Hf (a)) > 0

(iii) a is a saddle point if det(Hf (a)) < 0

Proof. As a is an interior point of O2 we can get an r > 0 such that a, a + h ∈ Br (a) ⊆ O2 . By

(13.4),
1
f (a + h) − f (a) = ∇f (a) · h + ht Hf (a + ch)h
2
We will prove (ii), other statements can be proved similarly. Our assumptions tell that Hf (a) is
positive definite. Hence, by the Lemma 12.2.1, there exist ϵ > 0 such that Hf (x) is positive definite
∀ x ∈ Bϵ (a). So for every x ∈ Bϵ (a) the quadratic form ht Hf (x)h > 0 with h ̸= 0 which implies
f (x) − f (a) > 0 in Bϵ (a) that means a is a point of local minimum.

Example 13.1.1 (Finding Critical points of a function and their nature)

Find the critical points and discuss the nature of the function
f (x, y) = x3 − 6x2 − 8y 2
Solution. Setting ∇f (x, y) = 0, i.e., (fx , fy )(x, y) = 0, we get the system of equations
3x2 − 12x = 0 and − 16y = 0
whose solution set is {(0, 0), (4, 0)} implying that (0, 0), (4, 0) are critical points.
The 2nd derivatives are,
fxx (x, y) = 6x − 12, fyy (x, y) = −16 and fxy (x, y) = 0

51
13.1 Taylor’s Theorem

Now, we compute the determinant of the hessian at these points to tell their nature. For
(0, 0),
−12 0
det(Hf (0, 0)) = > 0 and fxx (0, 0) = −12 < 0
0 −16
So, f has a local maximum at (0, 0). And at (4, 0),

12 0
det(Hf (4, 0)) = <0
0 −16

which shows that (4, 0) is a saddle point. ■

52
Lecture 14

14.1 Compact subsets of Rn

We start with the definition of Compactness which refers to a property of sets that captures the
notion of being finite or having no “holes”.

Definition 14.1.1 ▶ Compact Subset

A subset K ⊆ Rn is said to be compact if every sequence {xn } ⊆ K has a subsequence {xnk }
that is convergent to some x ∈ K.

This is known as the Bolzano-Weierstrass Property.

Observe that a compact subset of Rn is always closed. To see this, note that every sequence
{xn } ⊆ K, where K is a compact subset of Rn that converges to some x ∈ Rn has a convergent
subsequence {xnk } that converges to the same x. Since K is compact, we can say that x ∈ K. So,
the convergent sequence {xn } converges to a point in K. Hence, K is closed.

More is true. A compact subset of Rn is bounded too. Assume that a compact subset K ⊆ Rn is
not bounded. Note that, a subset of Rn is bounded iff it is contained inside an open ball. Since K
is unbounded, we can get a sequence {xm } ⊆ K with ∥xm ∥ > m, which doesn’t have a convergent
subsequence. This shows that K is not compact that contradicts our assumption.

Therefore, a compact subset of Rn is closed and bounded. What about the converse?

Theorem 14.1.1
A closed and bounded box in Rn is compact.

n
Y
Proof. We take a closed and bounded box K := [ai , bi ] ⊆ Rn . Fix i ∈ [n]. Consider a sequence
i=1
{xm } ⊆ K. We take its projection on the ith coordinate, i.e., {Πi (xm )} ⊆ [ai , bi ]. By Bolzano-
Weierstrass Theorem, it has a convergent subsequence {Πi (xm )t } ⊆ [ai , bi ] converging to αi ∈ [ai , bi ].
\n
Consider the intersection of inverse images of all such convergent subsequences, Π−1
i {Πi (xm )t },
i=1
which is a convergent subsequence of {xm } converging to α = (α1 , . . . , αn ) ∈ K.

Theorem 14.1.2 (Heine-Borel Theorem)

A subset K ⊆ Rn is compact iff it is closed and bounded.

Proof. =⇒ Done!

53
14.2 Inverse Function Theorem

⇐= Since, K is bounded, it is contained in a closed box, i.e., there exists r > 0 such that K ⊆ [−r, r]n .
So, Theorem 14.1.1 implies that all sequences in K has a convergent subsequence, which must
converge in K because K is closed. Hence, K is compact.

Theorem 14.1.3
Let f : On → Rm be a continuous map. Then f sends compact sets to compact sets.

In other words, continuous image of a compact set is compact.

Proof. Let K ∈ On be compact. Take a sequence {xk } ⊆ K with a convergent subsequence {xkt } ⊆ K
converging to x ∈ K. Then, {f (xk )} is sequence in f (K) with convergent subsequence {f (xmt )}
converging to f (x). The last statement about convergence follows from the continuity of f . This
shows that f (K) is compact.

Theorem 14.1.4 (Extreme Value Theorem)

Let K ⊆ Rn be compact and f : K → R a continuous map. Then ∃ a, b ∈ K such that
f (a) ≤ f (x) ≤ f (b) for all x ∈ K.

Proof. By Theorem 14.1.3, f (K) is compact. So f is bounded which implies sup f, inf f exist! Since
K K
f (K) is closed, they must exist inside f (K).

14.2 Inverse Function Theorem

We are now going to study Inverse Function Theorem which relates the differentiability of a function
to the differentiability of its inverse, enabling the study of local behavior and solving equations in
higher dimensions. But before that we prove a lemma that is essential to prove the theorem.

Lemma 14.2.1
Let On ⊆ Rn be open and convex. Suppose f : On → Rn be a C 1 function. If ∃ M > 0 such
that
∂fi
sup (x) ≤ M for all i, j
x∈On ∂xj

Then ∥f (x) − f (y)∥ ≤ n2 M ∥x − y∥ for every x, y ∈ On .

Proof. Pick x, y ∈ On and i ∈ [n]. Then using Mean Value Theorem, we can get ci ∈ Lx,y such that

fi (x) − fi (y) = ∇fi (ci ) · (x − y)

n
X ∂fi
=⇒ |fi (x) − fi (y)| = (ci ) · (xi − yi )
j=1
∂xj
n
X ∂fi
≤ (ci ) |xi − yi | (Triangle inequality)
j=1
∂xj
n
X
≤M |xi − yi | ≤ nM ∥x − y∥
j=1

The last inequality follows from the inequality |xi − yi | ≤ ∥x − y∥ which holds for all i.

54
 Lecture 14

Using the above,

v v
u n u n
uX 2
uX 2
∥f (x) − f (y)∥ = t |fi (x) − fi (y)| ≤ t n2 M 2 ∥x − y∥ ≤ n2 M ∥x − y∥
i=1 i=1

we obtain the result.

Theorem 14.2.1 (Inverse Function Theorem)

Let f : On → Rn be a C 1 function and a ∈ On . Suppose Df (a) is invertible. Then, there exist
open sets V and W containing a and f (a) respectively, such that f : V → W is invertible.
−1
Moreover, the local inverse f −1 ≡ (f |V ) : W → V is differentiable and for all y ∈ W ,


−1
Df −1 (y) = (Df ) f −1 (y)

i.e., locally, the derivative of the inverse is the matrix inverse of the derivative.

Proof .
We call L = Df (a) which is given to be invertible and take g(x) :=
L−1 f (x). Then,

Dg(a) = L−1 (f (a)) · (Df )(a)

= L−1 · Df (a) = I
 

As this transformation can be made, without loss of generality, we may

assume that Df (a) = In which would imply that there exists a closed Figure 14.1
box U containing a such that for all x ∈ U \ {a}, f (a) ̸= f (x). To see
this, let f (a) = f (a + h) with arbitrarily small ∥h∥. But then,
1 h
(f (a + h) − f (a) − Ih) = ̸= 0
∥h∥ ∥h∥

which contradicts the definition of derivative. Note that det Jf (a) ̸= 0. So, by continuity det Jf (x) ̸= 0
for all x ∈ U (we may shrink U if necessary). Hence, Df (x) is invertible for all x ∈ U . Again by
continuity, for all x ∈ U (we may shrink U if necessary),

∂f ∂f 1
(x) − (a) ≤ 2
∂xj ∂xj 2n
| {z }
δij

Now we claim the following,

Claim
For all x, y ∈ U ,
1
∥f (x) − f (y)∥ ≥ ∥x − y∥
2

Proof. We take g(x) = f (x) − x for all x ∈ U . Taking derivative, we get

Dg(x) = Df (x) − I = Df (x) − Df (a)

∂gi ∂fi ∂fi
=⇒ (x) = (x) − (a)
∂xj ∂xj ∂xj
∂gi 1
=⇒ (x) ≤ 2 ∀ x ∈ U
∂xj 2n

55
14.2 Inverse Function Theorem

Then by Lemma 14.2.1, for all x, y ∈ U ,

1
∥g(x) − g(y)∥ ≤ n2 · ∥x − y∥
2n2
1
=⇒ ∥(f (x) − f (y)) − (x − y)∥ ≤ ∥x − y∥
2
1
=⇒ ∥f (x) − f (y)∥ ≥ ∥x − y∥ (14.1)
2

Where (14.1) follows from the Triangle inequality. So, we get the claim! It shows that f is injective.
Next we look at the compact set ∂U ⊆ U . Since a ̸∈ ∂U , we can say f (x) ̸= f (a) for all x ∈ ∂U . So,
by continuity of f and compactness of ∂U , we can find a d ∈ R≥0 such that

∥f (x) − f (a)∥ ≥ d ∀ x ∈ ∂U

Figure 14.2

Now, we take W = B d (f (a)). Then, for every y ∈ W and x ∈ ∂U ,

2

∥y − f (a)∥ < ∥y − f (x)∥ (14.2)

| {z } | {z }
d atleast d
atmost 2

Claim

For a fixed y ∈ W, ∃ a unique x0 ∈ U ◦ such that f (x0 ) = y

Proof. We define a continuous function g : U → R with

n
X
2 2
g(x) = ∥y − f (x)∥ = (yi − fi (x))
i=1

Since, inf U g cannot occur at the boundary ∂U , but must occur in U , there exists x0 ∈ U ◦ such that
∂g
∇g(x0 ) = 0 i.e., (x0 ) = 0 ∀j
∂xj

Now, the partials of g,

n
∂g ∂ X 2
(x) = (yi − fi (x))
∂xj ∂xj i=1
n
X ∂fi
= −2 (yi − fi (x)) (x)
i=1
∂xj

56
 Lecture 14

At x0 , we get,
n
X ∂fi
(yi − fi (x0 )) (x0 ) = 0 ∀j
i=1
∂xj
 t
∂fi
=⇒ ∂xj (x0 ) (y − fi (x0 )) = 0
| {z }
Df (x0 )t

As Df is invertible in U and x0 ∈ U ◦ we obtain y = f (x0 ), which shows the existence of x0 . The

uniqueness follows from (14.1).

We now set V := U ∩ f −1 (W ). Since U is closed, V = U ◦ ∩ f −1 (W ). Hence, f |V : V → W is

invertible!

Claim
−1
f −1 ≡ (f |V ) : W → V is continuous.

Proof. (14.1) gives,

1
∥f (x1 ) − f (x2 )∥ ≥ ∥x1 − x2 ∥ ∀ x1 , x2 ∈ V ⊆ U
2
equivalently,
2∥y1 − y2 ∥ ≥ f −1 (y1 ) − f −1 (y2 ) (where, yi = f (xi ))
which shows that f −1 is Lipschitz, hence continuous.

Claim

f −1 is differentiable.

Proof. We fix y0 = f (x0 ) ∈ W for some x0 ∈ V and take A = Df (x0 ). As we know,

1
f −1 (y0 + h) − f −1 (y0 ) − A−1 h = 0


lim
h→0 ∥h∥
1
f −1 (y) − f −1 (y0 ) − A−1 (y − y0 ) = 0


⇐⇒ lim (14.3)
y→y0 ∥y − y0 ∥

We set ϕ(h) = f (x0 + h) − f (x0 ) − Ah for h in a neighborhood of 0. Now,

A−1 (f (x0 + h) − f (x0 )) = h + A−1 ϕ(h)

= ((x0 + h) − x0 ) + A−1 (ϕ((x0 + h) − x0 ))

Also set y = f (x0 + h). Then,

A−1 (y − y0 ) = f −1 (y) − f −1 (y0 ) + A−1 ϕ f −1 (y) − f −1 (y0 )

=⇒ − A−1 ϕ f −1 (y) − f −1 (y0 ) = f −1 (y) − f −1 (y0 ) − A−1 (y − y0 )

(14.4)

So, it is now enough to prove that,

1 −1
ϕ f −1 (y) − f −1 (y0 ) = 0



lim A
h→0 ∥h∥
1
ϕ f −1 (y) − f −1 (y0 ) = 0



⇐⇒ lim
y→y0 ∥y − y0 ∥

57
14.2 Inverse Function Theorem

But,
ϕ f −1 (y) − f −1 (y0 ) ϕ f −1 (y) − f −1 (y0 ) f −1 (y) − f −1 (y0 )



= · =0
∥y − y0 ∥ ∥f −1 (y) − f −1 (y0 )∥ ∥y − y0 ∥
| {z } | {z }
0 ≤2

Hence, the limit (14.3) is true.

To show that f −1 is C 1 , observe that all the partials of f −1 are rational polynomial functions (with
non-zero denominators) of those of f . This completes the proof.

58
Lecture 15

15.1 Inverse function theorem: Example

Recall that the inverse function theorem (14.2.1) states that if f : Rn → Rn is a C 1 function and there
is a ∈ On such that Df (a) is invertible, then there exists O fn and f −1 : O
fn such that f (a) ∈ O fn → On
1 −1 −1
exists and is a C function and further, D(f ) = (Df ) at each point. Thus, given that f satisfies
the conditions, the theorem guarantees that f is locally invertible with a differentiable inverse. We
discuss an important example where the theorem is used.
Consider the polar coordinate transformation,

x = r cos θ y = r sin θ

We can rephrase this with the function,

F : R>0 × R → R2
F (r, θ) = (x, y)

F is defined on an open set, and it is C 1 because we have,

∂x ∂y
= cos θ = sin θ
∂r ∂r
∂x ∂y
= −r sin θ = r cos θ
∂θ ∂θ

 
cos θ −r sin θ
=⇒ JF (r, θ) = =⇒ det JF (r, θ) = r
sin θ r cos θ

and hence, det JF (r, θ) is non-zero in the domain we chose. The inverse function theorem then
guarantees that we can express (r, θ) as a C 1 function of (x, y), locally. Further, we also have
 
−1 −1

−1 1 r cos θ r sin θ
DF (x, y) = DF (F (x, y)) =
r − sin θ cos θ

In other words,

∂r ∂r
= cos θ = sin θ
∂x ∂y
∂θ 1 ∂θ 1
= − sin θ = cos θ
∂x r ∂y r

which can also be verified directly.

59
15.2 Implicit Function Theorem

15.2 Implicit Function Theorem

In one variable, we could differentiate functions given as y = f (x). We have developed several variable
calculus far enough that we can now do the same for functions given as xn = f (x1 , . . . , xn−1 ). But
what about functions that are given as f (x, y) = 0, that is, implicitly in both variables? We will now
discuss an important theorem that will allow us to take derivatives of such functions as well without
needing to solve the equation for the dependent variable.

Example 15.2.1

Let F (x, y) = ax + by + c. We have,

a c
F (x, y) = 0 ⇐⇒ ax + by + c = 0 ⇐⇒ y = − x −
b b
for all b ̸= 0. Hence, given that F (x, y) = 0, we get

∂F
y = f (x) for a differentiable f ⇐⇒ b ̸= 0 ⇐⇒ ̸= 0
∂y
∂F
In other words, ∂y ̸= 0 ⇐⇒ F (x, f (x)) = 0 for some differentiable f .

Example 15.2.2

Let F (x, y) = x2 + y 2 − 1. We have,

p
F (x, y) = 0 ⇐⇒ x2 + y 2 = 1 ⇐⇒ y 2 = 1 − x2 ⇐⇒ y = ± 1 − x2

but the last expression is not a function! More precisely,

p
y ≥ 0 =⇒ y = f1 (x) = 1 − x2
p
y ≤ 0 =⇒ y = f2 (x) = − 1 − x2

Note that ∂F ∂F 1
∂y = 2y ̸= 0 for all y ̸= 0. Hence, for F (x0 , y0 ) ̸= 0, ∂y ̸= 0, there exists a C f
defined in a neighbourhood of x0 such that F (x, f (x)) = 0 for all x in that neighbourhood.

Note

Consider F (x, y) = 0 and y = f (x) for some C 1 f such that F (x, f (x)) = 0. We have,

F (x, f (x)) = 0
dF
=⇒ =0
dx
∂F ∂F dy
=⇒ + =0 (Chain rule)
∂x ∂y dx
∂F
dy ∂x
=⇒ = − ∂F
dx ∂y

Hence, the condition ∂F∂y ̸= 0 is necessary for differentiating functions defined implicitly.
Keeping the feeling of the above examples in mind, we now discuss the full theorem that
shows that it is also sufficient.

60
 Lecture 15

15.2.1 Proof of the theorem

We first introduce some notation. Throughout this section, X ∈ Rn , Y ∈ Rm so that (X, Y ) ∈ Rn+m .
The point (a, b) ∈ Rn+m is defined with a ∈ Rn , b ∈ Rm . O denotes an open set in Rn+m , and
F : O → Rm has the coordinate functions fi (X, Y ). Assuming F ∈ C 1 (O), its jacobian is
 
∂fi ∂fi
DF =
∂xj ∂yk

where i, k ∈ {1, . . . , m} and j ∈ {1, . . . , n}.

Theorem 15.2.1 (Implicit Function Theorem)

 
∂fi
Let F ∈ C 1 (O) and F (a, b) = 0. If det ∂y k
̸= 0, then there exists an open
m×m
neighbourhood U ⊂ Rn of a and a C 1 function f : U → Rm such that f (a) = b and
F (x, f (x)) = 0 for all x ∈ U .

Proof. We define the function F̃ : O → Rn × Rm as

Fe(X, Y ) = (X, F (X, Y ))

As F is C 1 , so is Fe, and we have

 
In 0
JFe =  ∂fi
 
∂fi 
∂xj ∂yk
 
∂fi
Hence, det JFe (a, b) ̸= 0 as det ∂yk ̸= 0, and so we can use the inverse function theorem!
By Inverse Function Theorem (14.2.1), there exists a neighbourhood U0 ⊆ O of (a, b) and a
neighbourhood V0 ⊆ Rn+m of (a, 0) such that Fe : U0 → V0 has a C 1 inverse.
We shrink U0 (also shrinking V0 accordingly) so that U0 = A×B where a ∈ U ⊆ Rn and b ∈ U ′ ⊆ Rm ,
for open sets A, B. We also have,

Fe−1 (X, Y ) = (X, g(X, Y ))

for some C 1 function g, from the definition of Fe. Now, consider the map

Π 2 : Rn × Rm → Rm
(X, Y ) 7→ Y

Then, Π2 ◦ Fe = F . So we get,

(X, g(X, Y )) = Fe−1 (X, Y )

=⇒ F (X, g(X, Y )) = Π2 (X, Y ) = Y

where (X, Y ) ∈ V0 , (X, g(X, Y )) ∈ U0 . Hence, for Y = 0 and X ∈ U ,

F (X, g(X, 0)) = 0

Therefore, f (X) = g(X, 0), X ∈ U, works.

61
15.3 Solving systems of equations

15.3 Solving systems of equations

Example 15.3.1

Consider the system of equations,

x2 + 2y 2 + z 2 + w = 6
2x3 + 4y 2 + z + w2 = 9

We wish to know whether (z, w) can be expressed as a function of (x, y) locally. Construct
the function F : R4 → R2 as F = (f1 , f2 ) where

f1 (x, y, z, w) = x2 + 2y 2 + z 2 + w − 6
f2 (x, y, z, w) = 2x3 + 4y 2 + z + w2 − 9

Consider α = (1, −1, −1, 2) ∈ R4 . We have,

∂f1 ∂f1
!  
∂z ∂w 2z 1
J= =
∂f2 ∂f2 1 2w
∂z ∂w
=⇒ det J(α) = −9 ̸= 0

Hence, by the implicit function theorem, there is a neighbourhood of (1, −1) such that on
that neighbourhood, (z, w) = f (x, y) for some C 1 f .

62
Lecture 16

16.1 Riemann-Darboux Integration

Definition 16.1.1 ▶ Volume of a closed/open ball

n n
We define the volume of a closed ball B n = [ai , bi ] as Vol(B n ) =
Q Q
(bi − ai ). We also define
i=1 i=1
n
the volume of the open set On = (ai , bi ) to be equal to that of B n , i.e., Vol(On ) = Vol(B n ).
Q
i=1

We now introduce some notation. Fix i ∈ {1, 2, . . . , n}. We define a partition of the ith interval
[ai , bi ] as
Pi : ai = ai,0 < ai,1 < · · · < ai,ni = bi
and the intervals of its partition as

Ii,t = [xi,t−1 , xi,t ], ∀ 1 ≤ t ≤ ni

and set
n
B(t 1 ,t2 ,...,tn )
= Bα = [x1,t1 −1 , x1,t1 ] × · · · × [xn,tn −1 , xn,tn ] = I1,t1 × · · · × In,tn

where α is chosen from the indexing set Λ(P ) = {α = (t1 , . . . , tn ) | 1 ≤ ti ≤ ni , i = 1, . . . , n}.

Note

1. B n = Bαn
S
α∈Λ(P )

2. Vol(B n ) = Vol(Bαn )
P
α∈Λ(P )

We call P(B) = {P1 × · · · × Pn | Pi ∈ P[ai , bi ]} the set of all partitions of B n .

Definition 16.1.2 ▶ Refinement of Partitions

n
Y n
Y
Given P = Pi and Pe = Pei with P, Pe ∈ P[a, b], then Pe is called a refinement of P if
i=1 i=1
Pei ⊃ Pi ∀i = 1, 2, . . . , n.

Theorem 16.1.1
Let f be a bounded function over B n . Let P, Pe ∈ P(B n ) and Pe ⊃ P . Then

L(f, P ) ≤ L(f, Pe) ≤ U (f, Pe) ≤ U (f, P )

63
16.1 Riemann-Darboux Integration

Proof. Note that L(f, Pe) ≤ U (f, Pe) follows directly from the fact that mα (Pe) ≤ Mα (Pe) ∀Pe ∈ P[a, b],
where mα = inf Bαn f and Mα = supBαn f .

Corollary (Inequality of upper and lower sums)

For all P, Pe ∈ P(B n ), the following inequality holds.

m × Vol(B n ) ≤ L(f, P ) ≤ U (f, Pe) ≤ M × Vol(B n )

We denote B(A) = {f : A → R | supA |f | < ∞} as the set of all bounded functions over A for any
A ⊆ Rn .

Definition 16.1.3 ▶ Upper and Lower Darboux Integrals

For f ∈ B(B n ), we define
Z Z
f= inf U (f, P ) and f= sup L(f, P )
Bn P ∈P(B n ) P ∈P(B n )
Bn

as the Upper and Lower Darboux Integrals, respectively.

We have L(f, P ) ≤ U (f, P ′ ) for all P, P ′ ∈ P(B n ) by taking the common refinement Pb = P ∪ P ′ .
Hence, Z Z
f≤ f
Bn Bn

Definition 16.1.4 ▶ Darboux Integral

Let f ∈ B(B n ). f is said to be Riemann-Darboux Integrable or Riemann Integrable or just
Integrable if Z Z
f= f
Bn Bn

In this case, we introduce the notation,

Z Z Z Z
f dV = f (x1 , . . . , xn ) dx1 · · · dxn = f= f
Bn Bn Bn Bn

Z
At this point, the notation f (x1 , . . . , xn ) dx1 · · · dxn does not indicate repeated integration, but
Bn
we will see that it represents repeated integration for “nice” functions.

64
Lecture 17

17.1 Properties of Riemann-Darboux Integration

In the previous lecture, we introduced the Riemann-Darboux Integral. In this lecture, we will explore
some important properties of this integral, starting with a characterization.

Theorem 17.1.1 (Classification of Riemann Integrable Functions)

Let f ∈ B(B n ). Then f ∈ R(B n ) if and only if for every ϵ > 0, there exists a partition
P ∈ P(B n ) of B n such that

(0 ≤) U (f, P ) − L(f, P ) < ϵ

Proof. =⇒ Suppose f ∈ R(B n ). Then we have

Z Z
f− f =0

Thus,
Z Z
0= f− f= inf U (f, P ) − sup L(f, P )
P ∈P(B n ) P ∈P(B n )

= inf (U (f, P ) − L(f, P ))

P ∈P(B n )

Hence for all ϵ > 0, there exists a partition P ∈ P(B n ) such that U (f, P ) − L(f, P ) < ϵ.

⇐= Conversely, assume that for every ϵ > 0, there exists a partition P ∈ P(B n ) such that
Z Z Z
U (f, P ) − L(f, P ) < ϵ. We want to show that f = f . Since U (f, P ) ≥ f and L(f, P ) ≤
Z
f , it follows that for all ϵ > 0,
Z Z
0≤ f− f <ϵ

Z Z
This implies f= f , showing that f ∈ R(B n ).

Exercise. Let f, g ∈ R(B n ). Then show that,

Z Z
• |f | ∈ R(B ) and
n
f ≤ |f |
Bn Bn

65
17.1 Properties of Riemann-Darboux Integration

• R(B n ) is a R−algebra by showing the following,

(i) For any α, β ∈ R, αf + βg ∈ R(B n )
(ii) f g ∈ R(B n )
Next, we demonstrate the Riemann integrability of a class of “nice” functions (continuous). However,
before proceeding, let’s introduce the concepts of the diameter of a set and the mesh of a partition.

Definition 17.1.1 ▶ Diameter of a set

For a set A ⊆ Rn , the diameter d(A) is defined as d(A) = sup{|x − y| | x, y ∈ A}.

Exercise. Show that d(B n ) = max {∥vi − vj ∥ | vi , vj are vertices of B n }

Definition 17.1.2 ▶ Mesh of a Partition

For a partition P ∈ P(B n ), the mesh ∥P ∥ is defined as ∥P ∥ = max{d(Bαn ) | α ∈ Λ(P )}.

Theorem 17.1.2 (Continuous Functions are Riemann Integrable)

The set of all continuous functions over B n is contained in R(B n ), i.e.,

C(B n ) ⊂ R(B n )

Proof. Let f ∈ C(B n ). Since f is uniformly continuous, for any ϵ > 0, there exists δ > 0 such that
for all x, y ∈ B n with ∥x − y∥ < δ, we have,
ϵ
|f (x) − f (y)| < (17.1)
2 Vol(B n )
| {z }
Call it ϵ̃

Let P ∈ P(B ) be a partition such that ∥P ∥ < δ. For each α ∈ Λ(P ), let aα ∈ Bαn . Then ∥x−aα ∥ < δ
n

for all x ∈ Bαn . It follows from the uniform continuity condition (17.1) that,

|f (x) − f (aα )| < ϵ̃

i.e., f (aα ) − ϵ̃ < f (x) < f (aα ) + ϵ̃ (17.2)

Since, (17.2) holds for all α ∈ Λ(P ), aα ∈ Bαn and for all x ∈ Bαn we have,

f (aα ) − ϵ̃ ≤ mα ≤ Mα ≤ f (aα ) + ϵ̃

Multiplying the volumes of Bαn and summing over Λ(P ), we obtain,

X ϵ X ϵ
f (aα ) Vol(Bαn ) − ≤ L(f, P ) ≤ U (f, P ) ≤ f (aα ) Vol(Bαn ) +
2 2
α∈Λ(P ) α∈Λ(P )

Thus, U (f, P ) − L(f, P ) < ϵ, and since ϵ is arbitrary, we conclude that f ∈ R(B n ).

Now, let’s consider an important question: Does an analogue of the Fundamental Theorem of Calculus
exist in higher dimensions? Z
In one dimension (n = 1), we have the useful relationship df = f , which aids in computa-
[a,b] ∂[a,b]
tion.
However, this relationship becomes less practical in higher dimensions. For
Z instance, in n = 1, the
X
continuous counterpart to a sum an is the one-dimensional integral f . Similarly, in n = 2,
B1 Z
X
the continuous analogue to a double sum amn is the two-dimensional integral f.
B2

66
 Lecture 17

17.2 Iterated Integrals

Before delving deeper into the concept of integrability, let’s take a brief detour to discuss the idea of
a double sum.

Definition 17.2.1 ▶ Convergence of Double Sequence

A double sequence {amn } converges to a if for every ϵ > 0 there exists a δ > 0 such that
|amn − a| < ϵ for all m, n ≥ N

Consider the following examples,

Example 17.2.1

• Let’s take the sequence {amn } defined by amn = m+n

1
N
for all m, n ∈ . This sequence
1
is bounded, and for N > 2ϵ , we have |amn − 0| = amn < ϵ for all m, n ≥ N .
• Now consider the sequence {amn } defined as follows,
(
n if m = 1
amn = 1
m+n otherwise

This sequence is also convergent but not bounded.

Recall the relation between total limit and iterated limits in double sequence,

Theorem
For a double sequence {amn } if limm,n→∞ amn exists and limm→∞ amn exists for all n, then
 
lim lim amn = lim amn
n→∞ m→∞ m,n→∞

An important analogue of the above result is Fubini’s Theorem. Computation of the Darboux integral
is typically a challenging task. However, Fubini’s Theorem offers a valuable approach that simplifies
the computation by utilizing iterated integrals.

Visualization
We look at slice functions along each axis, which enables us to simplify computations and apply
Fubini’s Theorem for efficient evaluation of multivariable integrals.
Consider a function f : B 2 → R. For each x ∈ [a1 , b1 ] we define a slice function fx : [a2 , b2 ] → R
given by fx (y) = f (x, y) for all y ∈ [a2 , b2 ]. This function is obtained by fixing x and slicing along
the y-axis at that x-coordinate. Then an iterated integral becomes
Z Z !
fx (y) dy dx
[a1 ,b1 ] [a2 ,b2 ]

The question arise whether this quantity is invariant under the interchange of x and y, i.e., we may
slice f along x−axis at y to obtain fy : [a1 , b1 ] → R given by fy (x) = f (x, y) for every x ∈ [a1 , b1 ]
and want to investigate the equality of
Z Z ! Z Z ! Z
? ?
fx (y) dy dx = fy (x) dx dy = f (17.3)
[a1 ,b1 ] [a2 ,b2 ] [a2 ,b2 ] [a1 ,b1 ] B2

67
17.2 Iterated Integrals

In this context, we observe that a partition P ∈ P(B 2 ) can be decomposed into the partitions of the
individual coordinates. Specifically, we have P = P1 × P2 for the two coordinate intervals [a1 , b1 ] and
[a2 , b2 ], and the corresponding indexing sets satisfy Λ(P ) = Λ(P1 ) × Λ(P2 ).

Now consider the following example,

Example 17.2.2 (A discrepancy: Function integrable, slices not)

Let I = [0, 1] and B 2 = I × I. Consider the function f : B 2 → R given by,

Q
(
1 if x = 12 , y ∈ ∩ [0, 1]
f (x, y) =
0 otherwise

So the x−slice becomes,

Qc ∩ [0, 1]
(
1 1 if y ∈
fx ≡ 0 for all x ̸=
2
and f 12 (y) =
0 if y ∈ Q ∩ [0, 1]
Dirichlet Function

and y−slice,
(
For y ∈ Q ∩ [0, 1], fy (x) =
1 if x = 21
0 otherwise
For y ∈ Qc ∩ [0, 1], fy ≡ 0

Clearly, fx ∈ R(I) for all x ∈ 1

2 but f 21 ̸∈ R(I) and fy ∈ R(I) for every y. So,
Z Z Z 
fy = 0 =⇒ fy (x) dx dy = 0
I I I
Z Z
But, fx doesn’t exist for x = 12 which means x 7−→ fx is not a well-defined function on
I Z Z  I

[0, 1]. Hence, fx (y) dy dx doesn’t exist. Yet f ∈ R(B 2 ). To see this, we fix ϵ > 0
I I
and consider the partition P = P1 × P2 where,
(
P1 : 0 < 12 − ϵ < 1
2 +ϵ<1
P2 : 0 < 1
(      )
1 1 1 1
So, P = 0, − ϵ × I , − ϵ, + ϵ × I , − ϵ, 1 × I .
2 2 2 2
| {z } | {z } | {z }
Bα1 Bα2 Bα3
Then mα1 = mα2 = mα3 = 0, Mα1 = Mα3 = 0 and Mα2 = 1, which implies U (f, P ) −
L(f, P ) = 2ϵ < 3ϵ. This shows that f ∈ R(B 2 ). Again L(f, P ) = 0 for all P ∈ P(B 2 ) and
hence, Z
f =0
B2

Question. Under which conditions does (17.3) hold?

Answer. Fubini’s Theorem. The conditions for (17.3) to hold will be discussed in the next lecture.

68
Lecture 18

18.1 Fubini’s Theorem

In this lecture, we explore Fubini’s Theorem. Let’s begin by setting up the necessary framework.
Consider the (m + n)-dimensional space, where m and n are positive integers. We can decompose a
box B m+n ⊆ Rm+n as the Cartesian product of two boxes, B m+n = B m × B n . Here, B m represents
a box in Rm , and B n represents a box in Rn .
Now, suppose we have a partition P ∈ P(Rm+n ). We can express this partition as the Cartesian
product of two partitions, P = P m × P n , where P m ∈ P(Rm ) and P n ∈ P(Rn ). The corresponding
indexing set for this partition becomes Λ(P ) = Λ(P m ) × Λ(P n ). Consequently, the elements of Λ(P )
can be written as α(P ) = (α(P m ), α(P n )).
By extending this decomposition, we can also break down the elements of the boxes, Bα(P ) =
Bα(P m ) × Bα(P n ) .

Throughout this section, we take x ∈ B m and y ∈ B n to represent the point (x, y) ∈ B m+n . For a
bounded function f ∈ B(B m+n ), we define the slice functions

• fx : B n → R as y 7→ f (x, y) for all x ∈ B m .

• fy : B m → R as x 7→ f (x, y) for all y ∈ B n .

It is worth noting that fx ∈ B(B n ) and fy ∈ B(B m ). For a fixed x ∈ B m , we can compute the
lower and upper integrals of fx over B n , denoted as f (x) and f (x) respectively. Similarly, we can
compute the lower and upper integrals of fy over B m for fixed y ∈ B m . These are given by,
Z Z
f (x) = fx (y) dV (y) and f (x) = fx (y) dV (y)
Bn Bn

with similar expressions for y. Now, let’s state Fubini’s Theorem.

Theorem 18.1.1 (Fubini’s Theorem)

Let f ∈ R(B m+n ). Then f , f ∈ R(B n ) and,

Z Z Z
f= f= f
Bm Bm B m+n

Consequently, we have the following corollaries,

69
18.1 Fubini’s Theorem

Corollary

For any f ∈ R(B m+n ), the following equalities hold,

Z Z ! Z Z 
f (x, y) dV (y) dV (x) = f (x, y) dV (x) dV (y)
Bm Bn Bm Bn
Z
= f (x, y) dV (x, y)
B m+n

Furthermore, if fx ∈ R(B n ) for all x, then f = f and

Z Z  Z
f (x, y) dV (y) dV (x) = f (x, y) dV (x, y)
Bm Bn B m+n

Corollary

If f ∈ C(B n ), then all possible slice functions are continuous and hence Riemann Integrable.
Thus, multidimensional integral becomes the iterated one-dimensional integrals,
Z Z Z Z Z  
f= ··· f dx1 dx2 · · · dxn−1 dxn
Bn

where, xi ’s can be in any order.

(n = 2) Hence, if f ∈ C(B 2 ), then (17.3) holds,

! !
Z Z Z b1 b2 Z b2 Z b1
f= f (x, y) dy dx = f (x, y) dx dy
B2 a1 a2 a2 a1

Proof of the Fubini’s Theorem. Let P = P m × P n be a partition of B m+n . Then, we can express
the lower sum L(f, P ) as follows,
X  
m+n
L(f, P ) = mα(P ) Vol Bα(P )
α(P )∈Λ(P )
X X    
n m
= m(α(P m ),α(P n )) Vol Bα(P n) Vol Bα(P ) m

α(P m )∈Λ(P m ) α(P n )∈Λ(P n )

| {z }
lP m

For each x ∈ B m and α(P n ) ∈ Λ(P n ), let mα(P n ) (x) = inf y∈Bα(P
n
n)
fx (y). It follows that mα(P n ) (x) ≥
m
m(α(P m ),α(P n )) for every x ∈ Bα(P m ) . Consequently, we have,
X  
n
lP m ≤ mα(P n ) Vol Bα(P n)

α(P n )∈Λ(P n )
Z
= L(fx , P n ) ≤ fx
Bn

m
Taking infimum over all x ∈ Bα(P m ) , we obtain,

Z
lP m ≤ inf
m
fx
x∈Bα(P m )
Bn

70
 Lecture 18

= m
inf f (x) = mα(P m )
x∈Bα(P m)

Thus, the lower sum becomes,

X  
m
L(f, P ) ≤ mα(P m ) Vol Bα(P m) = L(f , P m )
α(P m )∈Λ(P m )

Similarly, we can show that U (f, P ) ≥ U (f , P m ). Consequently, f ∈ R(B m ), and we have,

Z Z
f dV (x) = f
Bm B m+n

By following analogous arguments, we can show that f ∈ R(B m ) and,

Z Z Z
f dV (x) = f= f dV (x)
Bm B m+n Bm

Question: Will the function be Riemann integrable if all the slices are Riemann integrable?
We will address this question later. In the meantime, let’s conclude this lecture with an example.

Example 18.1.1

Consider the integral Z

xy dx dy
[0,1]2 | {z }
dv

We can evaluate this integral by iterated integration as follows,

Z 1 Z 1  Z 1 Z 1 
xy dx dy = y x dx dy
0 0 0 0
Z 1
y 1
dy =
=
0 2 4
Z 1 Z 1  Z 1  Z 1 
Alternatively, we can also express it as, y x dx dy = x dx y dy , which
0 0 0 0
yields the same result.

71
18.1 Fubini’s Theorem

72
Lecture 19

19.1 Integration over Bounded Domain

Now that we know how to do integration over boxes, in this lecture we will discuss how to integrate
a bounded function over an arbitrary bounded set.
Qn
Let Ω ⊆ Rn , and f ∈ B(Ω) and Ω bounded, then there exists B n ⊇ Ω where B n = i=1 [ai , bi ].

Definition 19.1.1
Given Ω bounded, let B n ⊇ Ω. For f ∈ B(Ω), define
(
f (x) if x ∈ Ω
f˜B n (x) =
0 if x ∈ B n \ Ω

An immediate question that arises now is: If B n ⊇ Ω and Bˆn ⊇ Ω then will it be true that

(I) f˜B n ∈ R(B n )

Z Z
(II) And if (I) holds, is it necessarily true that f˜B n = f˜Bˆn .
Bn Bˆn

It is in fact true that (I) =⇒ (II), but we won’t cover the proof here.

Definition 19.1.2
Z
Let f ∈ B(Ω). We say that f ∈ R(Ω) if f˜B n exists for some B n ⊇ Ω, and in this case we
Bn
define Z Z
f := f˜B n
Ω Bn

Definition 19.1.3 ▶ Content Zero Sets

Let S ⊆ Rn , we say that S is of content zero if for all ε > 0, there exists boxes {Bjn }pj=1 (for
N
some p ∈ ) such that
p n
[ X
S⊆ Bjn and Vol(Bjn ) < ε
j=1 j=1

For example a line segment in Rn is of content zero, provided n > 1. We then have the following
theorems:

73
19.2 Two Elementary Regions

Theorem 19.1.1
(i) Let f ∈ B(B n ) and let D = {x ∈ B n | f is not continuous at x} be the set of disconti-
nuities of f , if D is of content zero, then f ∈ R(B n ).
(ii) If S is a content zero set then int (S) = ∅.

(iii) Let Ω ⊆ Rn and On ⊆ Ω is bounded. Let f ∈ B(Ω) and f |On ∈ C(On ), if Ω̄ \ On is

content zero then f ∈ R(Ω) and
Z Z
f= f
Ω On

Particularly, (i) and (iii) of Theorem 19.1.1 are very important.

Now that we know how to integrate on arbitrary domains, the next question that comes to our mind
is, does there exist a Fubini’s theorem for integration over arbitrary sets? Before that we define
elementary regions.

19.2 Two Elementary Regions

Definition 19.2.1 ▶ Elementary Regions

A set Ω ⊆ R2 is y-simple/type I if there exists functions φ1 , φ2 ∈ B([a, b]) such that

Ω = {(x, y) | x ∈ [a, b], y ∈ [φ1 (x), φ2 (x)]}

Similarly a set Ω ⊆ R2 is x-simple/type II if there exists functions ψ1 , ψ2 ∈ B([c, d]) such that

Ω = {(x, y) | y ∈ [c, d], x ∈ [ψ1 (x), ψ2 (x)]}

0
x

Figure 19.1: Example of a y-simple region.

Example 19.2.1 (Examples of elementary regions)

The region H given by

H = {(x, y) | 0 ≤ x ≤ 1, and x2 ≤ y ≤ x}

74
 Lecture 19

is a y-simple region. (see Figure 19.1)

Figure 19.2: Plot of the region H.

Exercise. Show that the region bounded by x2 + y 2 ≤ 1 and y ≥ 0 in R2 is an x-simple as well as a

y-simple region.

75
19.2 Two Elementary Regions

76
Lecture 20

In the previous lecture we extended integration over boxes to over what we called elementary regions.
This lecture explores the extention of Fubini’s theorem to integration over such sets, and talk about
some applications thereof. Towards the end, we discuss the celebrated change of variables formula of
multivariable calculus.

20.1 Fubini’s Theorem on Elementary Regions

For integration over elementary regions, Fubini’s theorem takes the following form.

Theorem 20.1.1
Let f ∈ R(Ω) where Ω ⊆ R2 is a bounded elementary domain.
Z φ2 (x)
(1) If Ω = {(x, y) | a ≤ x ≤ b, and φ1 (x) ≤ y ≤ φ2 (x)} and if f (x, y) dy exists for
φ1 (x)
all x ∈ [a, b] then !
Z Z b Z φ2 (x)
f dA = f (x, y) dy dx
Ω a φ1 (x)

(2) Similarly we have !

Z Z d Z ψ2 (y)
f dA = f (x, y) dx dy
Ω c ψ1 (y)

when Ω is x-simple.

Proof. There exists c, d ∈ R such that Ω ⊆ [a, b] × [c, d] = B 2 . We know f˜ ∈ R(B 2 ) where
(
˜ f (x, y) if (x, y) ∈ Ω
f (x, y) =
0 if (x, y) ∈ B 2 \ Ω
Z φ2 (x)
Since f˜ ∈ R(B 2 ), and since f (x, y) dy exists for fixed x, hence
φ1 (x)

f˜(x, ·) [φ1 (x),φ2 (x)]

and f˜(x, ·) [c,d]\[φ1 (x),φ2 (x)]
≡0
Z d
are both Riemann integrable. Thus, we get that f˜(x, ·)|[c,d] ∈ R([c, d]) and hence f˜(x, y) dy exists
c
for all x ∈ [a, b] and further we have
Z d Z φ2 (x)
f˜(x, y) dy = f (x, y) dy ∀ x ∈ [a, b]
c φ1 (x)

77
20.1 Fubini’s Theorem on Elementary Regions

Then we get
! !
Z b Z d Z b Z φ2 (x)
=⇒ f˜(x, y) dy dx = f (x, y) dy dx
a c a φ1 (x)
!
Z Z b Z φ2 (x)
(∗)
=⇒ f˜ dA = f (x, y) dy dx
B2 a φ1 (x)
!
Z Z b Z φ2 (x)
=⇒ f dA = f (x, y) dy dx
Ω a φ1 (x)

where (∗) follows from Theorem 18.1.1. An analogous argument works for the x-simple regions.

Example 20.1.1

Let f ∈ C(Ω) where Ω = (x, y) | 0 ≤ x ≤ 1 − y2 , and 0 ≤ y ≤ 2 . Then we can write Ω as a



y-simple region as follows:

Ω = {(x, y) | 0 ≤ x ≤ 1, and 0 ≤ y ≤ 2(1 − x)} .

Now, using Theorem 20.1.1 we get the required result.

Z Z 2 Z 1− y2 ! Z 1 Z 2(1−x)
!
f dA = f (x, y) dx dy = f (x, y) dy dx
Ω 0 0 0 0

Example 20.1.2
Z
2
Let B = [0, π] × [− π2 , π2 ], and we want to evaluate the integral sin(x + y) dA.
B2
Z Z Z
sin(x + y) dA = sin x cos y dA + sin y cos x dA
B2 B2 B2
Z π ! Z π
! Z
2 π  Z 2 π 
= cos y dy sin x dx + sin y dy cos x dx
−π
2 0 −π
2 0

=4

Example 20.1.3
Z
Let Ω be the region bounded by y = 1 and y = x2 , and we want to find x2 y dV . We can
Ω
write Ω as a y-simple region as follows:

Ω = {(x, y) | −1 ≤ x ≤ 1, and x2 ≤ y ≤ 1}

Then using Theorem 20.1.1 we get that

Z Z 1 Z 1 
x2 y dA = x2 y dy dx
Ω −1 x2
1 1
y2
Z  
2 2
= x dx =
−1 2 x2 15

78
 Lecture 20

Example 20.1.4
Z
Compute f dA where f : [0, 1]2 → R is given by
[0,1]2
(
x if y ≤ x2
f (x, y) = ∀ (x, y) ∈ [0, 1]2
y if y > x2

We define the regions

Ω1 = {(x, y) | 0 ≤ x ≤ 1, 0 ≤ y ≤ x2 } and Ω2 = {(x, y) | 0 ≤ x ≤ 1, x2 ≤ y ≤ 1}.

Then f |Ω1 and f |Ω2 are both Riemann integrable, and while f |y=x2 is not continuous, the
set {(x, x2 ) | 0 ≤ x ≤ 1} is of content zero. Hence f is integrable, and we can make sense of
writing the given integral as a sum
Z Z Z
f dA = f dA + f dA
[0,1]2 Ω1 Ω2

Using Theorem 20.1.1, we can simplify each of these parts:

Z Z Z 2 !
1 x
1
f dA = x dy dx =
Ω1 0 0 4
Z Z 1 Z 1 
2
f dA = y dy dx =
Ω2 0 x2 5
Thus, we have the result Z
13
f dA =
[0,1]2 20

Example 20.1.5
Z π Z π
2 sin y
2
Compute dy dx using Fubini’s theorem.
0 x y
We consider the region Ω = (x, y) | 0 ≤ x ≤ π2 , x ≤ y ≤ π2 ⊆ R2 can be written as a x-simple


region as follows: n π o
Ω = (x, y) | 0 ≤ y ≤ , 0 ≤ x ≤ y
2
This shows that
Z Z π2 Z π2
sin y sin y
dA = dy dx
Ω y 0 x y
Z π2 Z y
sin y
= dx dy
0 0 y
Z π2
= sin y dy
0
=1

79
20.2 Change of Variables

20.2 Change of Variables

Before discussing the theorem in the multivariable case, we recall the change of variable rule for real
valued functions on the real line.

Theorem 20.2.1 (Change of Variable on R)

Let φ : O1 → R be a C 1 function where φ′ (x) ̸= 0 for all x ∈ O1 . Then, for [a, b] ⊆ O1 and
f ∈ C(φ[a, b]), we have
Z φ(b) Z b
f= (f ◦ φ)φ′ .
φ(a) a

Here we effectively compensate for the change of variable by introducing the scale change factor of
φ′ . As we have seen, the scale change factor at a point for a transformation on Rn is given by the
detereminant of the Jacobian matrix at that point. Thus, this theorem has the following natural
extension to Rn :

Theorem 20.2.2 (Change of Variable on Rn )

Let φ : On → Rn be an injective and C 1 function, where det(Jφ (x)) ̸= 0 for all x ∈ On . Let
Ω ⊆ On , then for f ∈ R(φ(Ω))
Z Z
f dV = (f ◦ φ)| det Jφ |
φ(Ω) Ω

Although it is not too hard to get a feel for the theorem from its applications, the proof is quite long
and technical, and thus omitted. We recommend the interested and daring readers to have a look
at page 67 of Calculus on Manifolds by Michael Spivak. In the next lecture, we will discuss some
applications of this result.

80
Lecture 21

In the previous lecture, we extended our notion of Riemann integration over boxes to elementary
regions. Additionally, we had discussed the change of variable formula for multivariable calculus
(Theorem 20.2.2). We now try to motivate the use of the same with some rather important applications.

21.1 Change of Variables (Continued)

We start off with a particularly useful example. When dealing with functions on R2 , particularly if
the situation is radially symmetric, it is often useful to work in polar coordinates. Here we analyse
how that change of coordinates transforms the integrals over a given region.

Example 21.1.1 (Polar coordinates)

This example illustrates how we can compute in-

tegrals when converting to polar coordinates from
Cartesian coordinates.
Consider φ : R2 → R2 given by φ(r, θ) =
(r cos θ, r sin θ). Then, the Jacobian matrix of
φ is given by
 
cos θ −r sin θ
Jφ (r, θ) =
sin θ r cos θ
Thus, det(Jφ (r, θ)) = r ̸= 0 for all r > 0. Taking
the domain to be O2 := (0, ∞) × (0, 2π), the
function φ|O2 : O2 → R2 is C 1 and injective, with
det(Jφ (r, θ)) = ̸ 0. For some fixed 0 < r1 < r2
and 0 ≤ θ1 < θ2 < 2π, consider the set Ω =
(r1 , r2 ) × (θ1 , θ2 ). Clearly the boundary ∂Ω is of
content zero (since it is union of finitely many Figure 21.1: Transforming into polar coor-
line segments) and thus using Theorem 20.2.2 we dinates
see that for f ∈ R(φ(Ω)),
Z Z
f = (f ◦ φ)| det(Jφ (r, θ))|
φ(Ω) Ω
Z
= rf (r cos θ, r sin θ)
Ω
Zr2 Z θ2
= rf (r cos θ, r sin θ) dθ dr
r1 θ1

As a simple example, consider the following integral:

Z
2 2
e−(x +y ) dA
x2 +y 2 <1

81
21.1 Change of Variables (Continued)

The domain of integration can be written as

{(x, y) | x2 + y 2 < 1} = φ((0, 1) × [0, 2π))

2
+y 2 )
Clearly f (x, y) = e−(x ∈ C 1 (φ((0, 1) × [0, 2π))) which shows that:
Z Z 1 Z 2π 
−(x2 +y 2 ) −r 2
e dA = re dθ dr
x2 +y 2 <1 0 0
Z 1
2
= 2π re−r dr
0 
1
=π 1−
e

Definition 21.1.1 ▶ Area or volume of a region

For Ω ⊆ Rn , the volume of the region Ω is defined by the integral
Z
χΩ
Rn

if it exists, where χΩ is the indicator function of the region Ω, given by

(
1 if x ∈ Ω,
χΩ =
0 otherwise.

Example 21.1.2
n 2 2
o
Compute the area of Ω = (x, y) | x 3 + y 3 < 1 . Consider the function

φ : [0, 1] × [0, 2π] → R2

(r, θ) 7→ (r cos3 θ, r sin3 θ)

then clearly φ([0, 1] × [0, 2π]) = Ω. Also φ is injective and C 1 , but we have

det(Jφ (r, θ)) = 3r sin2 θ cos2 θ

and thus det(Jφ (r, θ)) = 0 if r = 0 or θ ∈ 0, π2 , π, 3π


2 , 2π , but set of all such points are of
content zero, hence we can safely ignore them while doing our integration. We get

Area of Ω = Area of φ ([0, 1] × [0, 2π])

Z
= 1 dA
φ([0,1]×[0,2π])
Z 1 Z 2π
3 3π
= r sin2 2θ dθ dr =
0 0 4 8

Next up, consider a change of coordinates in R3 , from Cartesian system to the spherical co-ordinate
system. Just like the previous case, this comes in handy when dealing with functions and regions
which are spherically symmetric. One canonical example may be its use in the theory of central
forces in physics.

82
 Lecture 21

Example 21.1.3 (Spherical coordinates)

The spherical co-ordinate system gives a unique

representation to all points in R3 not lying on the
z-axis. For all (x, y, z) ∈ R3 \ {(0, 0, α) | α ∈ R},
set

(x, y, z) = (r sin ϕ cos θ, r sin ϕ sin θ, r cos ϕ)

where r > 0, 0 < θ < 2π and 0 < ϕ < π. Define

the set

O3 := {(r, ϕ, θ) | r > 0, 0 < ϕ < π and 0 < θ < 2π}

and the map

φ : O3 → R3 , φ(r, ϕ, θ) = (x, y, z)

Then, the Jacobian matrix of the map φ is:

Figure 21.2: Transforming into spherical  
coordinates sin ϕ cos θ r cos ϕ cos θ −r sin ϕ sin θ
Jφ (r, ϕ, θ) =  sin ϕ sin θ r cos ϕ sin θ r sin ϕ cos θ 
cos ϕ −r sin ϕ 0

This gives det(Jφ (r, ϕ, θ)) = r2 sin ϕ, which is non-vanishing in the given domain. As φ
is injective and C 1 , we can use Theorem 20.2.2 to transform from Cartesian to spherical
coordinates. We provide a simple example for some clarity.

Consider the solid sphere of radius a, given by Ω = {(x, y, z) | x2 + y 2 + z 2 ≤ a2 }. Then, the

volume is:
Z 2π Z π Z a
Vol(Ω) = r2 sin ϕ dr dϕ dθ
0 0 0
4
= πa3
3

We leave it to the reader to do a similar analysis for the cylindrical co-ordinate system. This formula
also finds extensive use in probability theory, where it is commonly referred to as the change of
density formula (see, for instance A First Course in Probability by Sheldon Ross, or any introductory
probability book for that matter). Hopefully, we have demostrated to the reader the central role
Theorem 20.2.2 plays in analysis of several variables, enough to convince him to actually read the
proof! We will now depart from the general study of functions and integrals in Rn , and delve into
the theory of curves and surfaces, dealing primarily with R3 .

83
21.1 Change of Variables (Continued)

84
Lecture 22

22.1 Curves and Surfaces

We now study the basic concepts of curves and surfaces as subsets of R2 or R3 (mainly) with a
given parametrization, but also as subsets defined by equations. The connection from equations to
parametrization is drawn by means of the Implicit function theorem.

Definition 22.1.1 ▶ Parametrized Curve

A parametrized curve is a continuous function γ : [a, b] → Rn . We say that parametrized
curve is C 1 if t 7→ γi (t) is C 1 for all i = 1, . . . , n. A parametrized curve γ : I → Rn is smooth
if γ ′ (t) ̸= 0 for all t ∈ I. The path of a parametrized curve γ is the set

{γ(t) | t ∈ [a, b]}

Let us consider some examples.

Example 22.1.1

1. Let γ : [0, 1] → R2 be γ(t) = (1 − 2t, 2 + t). Clearly γ is C 1 and γ ′ (t) = (−2, 1) ̸= 0 and
thus γ is smooth.
2. γ : [0, 2π] → R2 given by γ : t 7→ (r cos t, r sin t) where r > 0 is constant. Then
γ ′ (t) = (−r sin t, r cos t) ̸= 0 ∀ t ∈ [0, 2π], thus γ is smooth.

3. Fix r > 0 and c ̸= 0, and define

γ : [0, nπ] → R3
t 7→ (r cos t, r sin t, ct)

Then γ ′ (t) = (−r sin t, r cos t, c) ̸= 0 and hence γ is smooth. The path of γ is called a
helix.

4. γ : [−1, 1] → R2 given by γ(t) = (|t|, t), then γ is not C 1 , hence it is not smooth.
5. γ : [0, 1] → R2 given by γ(t) = (0, t2 ), then even though γ is C 1 , but it is not smooth as
γ ′ (0) = 0.
6. γ : [0, 2π] → R2 given by γ : t 7→ (r cos t, r sin t), then path of γ is given by

{(r cos t, r sin t) | t ∈ [0, 2π]} = {(x, y) | x2 + y 2 = r2 }

= path of γ̃

where γ̃(t) = (r cos 2t, r sin 2t).

85
22.1 Curves and Surfaces

Definition 22.1.2 ▶ Piecewise Smooth Curve

A parametrized curve γ : [a, b] → Rn is called piecewise smooth if there exists a partition
a = t0 < t1 < · · · < tm = b such that

γ|[ti−1 ,ti ] is smooth ∀ i ∈ [m]

Definition 22.1.3 ▶ Equivalent Curves

Two parametrized curves γ : [a, b] → Rn and γ̃ : [ã, b̃] → Rn are equivalent, denoted by
γ ∼ γ̃ if there exists a strictly increasing surjective function which is differentiable (even C 1 ),
φ : [ã, b̃] → [a, b] such that γ̃ = γ ◦ φ.
γ̃
[ã, b̃] Rn
φ γ

[a, b]

Definition 22.1.4
Let γ : [a, b] → Rn be a C 1 curve, then
(i) ∥γ ′ (t)∥ := speed of γ at time t.
Z b
(ii) ∥γ ′ (t)∥dt := arc length of γ.
a

Let’s try to look at more natural how equation (ii) in the previous definition gives us the arc length
of a curve γ.

Let γ : [a, b] → Rn and let P := a = t0 < t1 < · · · < tm = b be a partition of the interval [a, b]. Now
define
Xm
ℓ(γ, P) = ∥γ(ti−1 ) − γ(ti )∥
i=1

Definition 22.1.5
A curve γ : [a, b] → Rn is rectifiable or said to have arc length if

lim ℓ(γ, P) = ℓ(γ) exists

∥P∥→0
P∈P[a,b]

which is equivalent to saying that for all ε > 0, there exists δ > 0, such that

∥ℓ(γ, P) − ℓ(γ)∥ < ε ∀ P ∈ P([a, b]) such that ∥P∥ < δ

Theorem 22.1.1
Z b
For a piecewise smooth curve γ : [a, b] → Rn it is rectifiable and ℓ(γ) = ∥γ ′ (t)∥dt.
a

Remark. Rectifiable curve ̸⇒ piecewise smooth, counter example: Cantor’s function (popularly
called the Devil’s staircase).

86
 Lecture 22

Theorem 22.1.2
Let γ : [a, b] → Rn be a rectifiable parametrized curve and let γ̃ = γ ◦ φ, where φ is a strictly
increasing surjective and continuous function, then γ̃ is rectifiable and ℓ(γ) = ℓ(γ̃).

Theorem 22.1.3
Let γ : [a, b] → Rn be a smooth curve, then there exists a parametrization φ such that
∥γ̃ ′ (s)∥ = 1 for all s ∈ [c, d].

87
22.1 Curves and Surfaces

88
Lecture 23

We will begin this lecture with few examples.

Example 23.0.2 (Path Of a Projectile)

γ(t) = (αt, βt − 16t2 )

=⇒ γ ′ (t) = (α, β − 32t)
Z
path length = ∥γ ′ (t)∥ dt
Z p
= α2 + (β − 32t)2 dt

Example 23.0.3 (Perimeter of a Circle)

Parametrization of a circle of radius r is given by γ(t) = (r cos(t), r sin(t)), t ∈ [0, 2π).

γ ′ (t) = (−r sin(t), r cos(t))

Z 2π
ℓ(γ) = ∥γ ′ (t)∥dt
0
Z 2π
= r dt
0
= 2πr

Example 23.0.4 (Arc Length of graph of functions)

Let, f : [a, b] → R be a C 1 function. Consider γ(t) = (t, f (t)). It is a smooth curve.

γ ′ (t) = (1, f ′ (t))

Z b
ℓ(γ) = ∥γ ′ (t)∥dt
a
Z b p
= 1 + f ′ (t)2 dt
a

23.1 Line Integrals

To integrate a function over a curve we use Line integral. The function we should integrate maybe a
Scalar Field or a Vector Field.(A quick example of a Vector Field: f : On → R be a differentiable
function, then ∇f is a vector field.)

89
23.1 Line Integrals

Z
Question. Given a scalar field f : On → R and γ ≡ C be a curve, we want to define f . But
C
exactly how we can do this?
Answer. C is a curve, so it is bounded subset of Rn . How about thinking of Riemann Integration?
For n ≥ 2, C is content zero in Rn . This does not make any sense! The right way is as following.

Let, γ : [a, b] → Rn be a smooth curve (or piecewise smooth) and C := ran(γ) (on other words path
of γ). Let, f ∈ B(C). Given P ∈ P[a, b], P : a = t0 < t1 < · · · < tm = b.

Let, Ii = [ti−1 , ti ] be the sub-intervals and

Ci = γ(Ii ). Since, γ is smooth there is nice corre-
spondence between Ii and Ci . Also denote si by
∥γ(ti ) − γ(ti )∥. As previous, define mi = inf Ci f
and Mi = supCi f .

m
X
U (f, P) = Mi · si
i=1
Xm
L(f, P) = mi · si
i=1
Figure 23.1: Curve C
The above expressions are same as upper and
lower Riemann sum respectively. This opens up
“The Pandora’s box!”.

We can now use all the theory we used for the

standard Riemann Integrals. We say f is line
integrable over γ if,
inf U (f, P) = sup L(f, P)
P∈P[a,b] P∈P[a,b]
Z
More over we will write the common value of the above equality as f and call this “The Line
C
Integral over curve C”.
Now we can invoke all theory we derived for 1 variable integration! We call R(C) the set of all
Riemann integrable functions over C.

Theorem 23.1.1
Let, γ be a “Rectifiable” smooth(or piecewise smooth) and C = ran(γ) and f ∈ B(C). Then,
1. f ∈ C 0 (C) =⇒ f ∈ R(C)

2.
m
X Z
f ∈ R(C) ⇐⇒ lim f (ζi )si exist and equal to f.
||P||→0 C
i=1

Here, ζi is tag of the interval Ii .

3. (This requires smoothness) If γ is C 1 and smooth, f ∈ R(C), then
Z Z b
f= f (γ(t))∥γ ′ (t)∥ dt
C a

Proof. Exercise.

90
 Lecture 23

In the above theorem equation in 3 is also independent of choice of Parametrization of γ. Because

any other smooth parametrized curve γ̃ = γ ◦ φ where, φ is an onto continuous function.

Facts: C be a piecewise smooth, parametrized curve γ. f, g ∈ R(C) and r ∈ R,then

Z Z Z
• f + rg = f + r g
Z Z
• f ≥ g over C then f≥ g
Z Z
• |f | ≥ f

• If a < d < b, if γ1 := γ|[a,d] and γ2 := γ|[d,b] then

Z Z Z
f= f+ f
C γ1 γ2

We have resolved the problems for Scalar field. What about vector fields?

Figure 23.2: Work done by a constant Force

Suppose a particle moves a distance d under a constant force F , then work done by the force is
⃗
F d cos θ = F⃗ · d.
If the force was not constant throughout the path
then how can we calculate work done by that
force? Consider the case where F is the vector
field (Force in this case) defined over a curve
(path)γ. Here, γ : [a, b] → Rn and C = ran(γ).
So, work done throughout the whole path will
be, Z
F⃗ · d⃗r
C
Z b
Which is equal to, F⃗ (γ(t)) · ∇γ(t) dt
a Figure 23.3: F⃗ is the vector field over the curve γ
Now we will look into some examples.

Example 23.1.1

Find work done by the field F (x, y, z) =

(xy, xz, yz) along the curve γ(t) =
(t2 , −t3 , t4 ), t ∈ [0, 1].
Answer.

γ ′ (t) = (2t, −3t2 , 4t3 )

91
23.1 Line Integrals

Z 1
=⇒ Work done = (−t5 , t6 , −t7 ) · (2t, −3t2 , 4t3 ) dt
0
31
=−
88

23.1.1 Line Integration of a Vector Field

F : On → Rn be a vector field and γ : [a, b] → On be a curve. We consider a partition P : a = t0 <
t1 < · · · < tm = b, Let Ci = γ|[ti−1 ,ti ] and γi = γ(ti ), ∆ri = γi − γi−1 .
m
X
R(F ; P) = F (γi ) · ∆ri
i=1

Z
F⃗ · d⃗r = lim R(F, P) (if the limit exists)
C ∥P∥→0

Just like the scalar field, if γ is C 1 and smooth, then

Z Z b
F⃗ · d⃗r = F (γ(t)) · γ ′ (t) dt (23.1)
C a

92
Lecture 24

Theorem 24.0.2
Let, f : On → R be a C 1 function and γ be a piecewise smooth C 1 curve on On , joining two
points A and B. Then Z
∇f · d⃗r = f (B) − f (A)
c
Which means the above line integral is “independent of parametrization”.

Proof .
Let, r be a parametrization of the curve γ and r : [a, b] → On such that
r(a) = A and r(b) = B.
Evaluating the LHS gives,
Z Z b
(23.1)
∇f · d⃗r = ∇f (r(t)) · r′ (t) dt
c a
n
!
Z b X ∂f ′
= × ri (t) dt
a i=1
∂ri
Z b Figure 24.1: Paths from
d
= (f (r(t))) dt A to B
a dt
= f (r(b)) − f (r(a))
= f (B) − f (A)

This result has an important consequence that we often use in Physics.

“Work done by a conservative force always depends on the starting point and the end
point, not on the path followed by the particle.”

We know if the force is conservative then we can define potential energy U as F⃗ = −∇U . Work done
by the force is simply the change of potential.

Now we will recall the basics of the Planes and Normals.

24.1 Planes and Normals

Let, P⃗0 = ⟨x0 , y0 , z0 ⟩ be a fixed vector in R3 . N ⃗ = ⟨a, b, c⟩ =
̸ ⃗0. The plane through P⃗0 with N
⃗ as
normal to this plane is,
n o n o
P⃗0 + P⃗ | P⃗ · N
⃗ = 0 = ⃗r | (⃗r − P⃗0 ) · N
⃗ =0

93
24.2 Surface and Surface Integrals

Equation of the plane: Consider an arbitrary point P⃗ = ⟨x, y, z⟩ on the plane then,

⟨x − x0 , y − y0 , z − z0 ⟩ · ⟨a, b, c⟩ = 0

So, equation of the plane is

a(x − x0 ) + b(y − y0 ) + c(z − z0 ) = 0 (24.1)

Let Q⃗1 , Q⃗2 be independent in R3 and also satisfying Q ⃗i · N

⃗ = 0. Clearly, Q⃗1 × Q⃗2 =
̸ ⃗0. We can see
(Q⃗1 × Q⃗2 ) · Q
⃗ i = 0, i.e., {Q⃗1 , Q⃗2 , Q⃗1 × Q⃗2 } form a basis of R3 .

∴ Q⃗1 × Q⃗2 = cN
⃗
n o
So, P⃗0 + r1 Q⃗1 + r2 Q⃗2 | r1 , r2 ∈ R describes the same plane as (24.1).

24.2 Surface and Surface Integrals

Definition 24.2.1 ▶ Region

A subset R ⊆ R2 is called a “Region” if R is Open ans R has an area (i.e. ∂R is content
zero)

Definition 24.2.2 ▶ Parametrized Surface

Let R ⊆ R2 be a region. A C 1 function r : R → R3 said to be a “Parametrized Surface” if :

• The component functions ri have bounded partials

• r is 1 − 1 function
• ∂⃗
r
∂u × ∂⃗
r
∂v ̸= ⃗0 for all (u, v) ∈ R2 . This means total derivative of r has rank 2.
We will call range of r as a Surface, S = ran(r).

η
Let, η be a map defined on (−ε, ε) that maps t 7−→ r(uo = 0 + t, v0 ). Clearly, η defines smooth curve
η̃
on S. Similarly, we can define η̃ on (−ε, ε) that maps t 7−→ r(u0 , v0 + t).
Thus, η̃ also defines a smooth curve on S.
From the Figure 24.2 we can see η and η̃ are
the curves. ru (u0 , v0 ) = dη
dt t=0 , which gives the
tangent of the curve η at point (u0 , v0 ) along x
axis.
Similarly, for η̃, rv (u0 , v0 ) gives the tangent of
η̃ along y axis. The vectors ru (u0 , v0 ), rv (u0 , v0 )
spanned together to form a plane. This plane is
known as Tangent Plane.
Since r is C 1 , both r⃗u and r⃗v are continuous
and hence r⃗u × r⃗v is continuous. Also, r⃗u × r⃗v is
along the normal vector of S at r(u0 , v0 ) which
follows from the previous section.

Now we will move towards a very important Figure 24.2

definition.

94
 Lecture 24

Definition 24.2.3 ▶ Tangent Plane

For a Parametrized Surface r, let ran(r) = S and r(u0 , v0 ) = P , then the plane generated by
ru (u0 , v0 ) and rv (u0 , v0 ) through r(u0 , v0 ) is called the Tangent Plane of S through P .
We denote it by TP S.
Every element of TP S is called Tangent Vectors at P on S.

It can be shown that TP S is independent of parametrization of S, i.e., TP S is independent of r

(Exercise). Actually, for different r̃ of S the basis for TP S will be changed. But they still generate
the same plane.

24.3 Examples

We will go through some examples.

Example 24.3.1 (Graph of Function)

Let f : O2 → R be a C 1 function then the graph of f is G(f ) = {(x, y, f (x, y)) : (x, y) ∈ O2 }.
Under the conditions O2 is bounded and partial derivatives of f is bounded, we want to find
a parametrization of this Surface G(f ).
Answer. Here, we use the trivial parametrization r : O2 → R3 that is r(x, y) = (x, y, f (x, y)).
Clearly, r is one - one. Now, ru (u, v) = (1, 0, fu ) and rv (u, v) = (0, 1, fv ).
So, ru × rv = (−fu , −fv , 1) ̸= ⃗0. So, it is a parametrization of the surface.

Example 24.3.2 (Torus)

Figure 24.3: Torus.

Parametrization is given by, (0 < r < R)

r(θ, φ) = ((R + r cos θ) cos φ, (R + r cos θ) sin φ, r sin θ); 0 ≤ θ, φ ≤ 2π (24.2)

95
24.3 Examples

Example 24.3.3 (Surface of Revolution)

Let f, g be C 1 functions on [0, b]. Consider the curve t 7−→ (0, f (t), g(t)) is a C 1 curve. If we
rotate the curve with respect to z axis, we must get a surface. Parametrization of the surface
is given by,
r(u, v) = (f (u) cos v, f (u) sin v, g(u)) ; u ∈ [0, b], v ∈ [0, 2π]

Exercise. Find the Parametrization of a sphere of radius R.

96
Lecture 25

25.1 Tangent Plane Of G(f )

Let, f : O2 → R be a C 1 function and r(u, v) = (u, v, f (u, v)). Here ran(r) defines a surface G(f )
as we have showed in Example 24.3.1. We also have calculated rv × ru = (−fu , −fv , 1). Now using
(24.1) we can write down the equation of Tangent space at a point P = (a, b, f (a, b)) on G(f ).
Equation of the tangent space TP S,

fu (a, b)(x − a) + fv (a, b)(y − b) − (z − f (a, b)) = 0

=⇒ z = f (a, b) + fu (a, b)(x − a) + fv (a, b)(y − b) (25.1)

Equation of Normal at the point P on the surface G(f ) is,

x−a y−b z − f (a, b)

= = (25.2)
−fu (a, b) −fv (a, b) 1

2x
Example 25.1.1 (Equation of Tangent and Normal to z = f (x, y) = y − y 2 at (1, 1, 1))

Solution. This is a graph function so obviously a surface fx (x, y) = y2 and fy = − 2x

y 2 − 2y.
x−1 y−1 z−1
So,⟨−fx , −fy , 1⟩ = ⟨−2, 4, 1⟩. So equation of Normal is −2 = 4 = 1 and the equation of
Tangent Plane is, 2(x − 1) − 4(y − 1) − (z − 1) = 0. ■

1.99
Example 25.1.2 (Use Tangent Plane to approximate (1.99)2 − 1.01 )

Solution. Consider z = x2 − xy = f (x, y). This describes a surface. Now consider P = (2, 1, 2)
be the point on the surface. Here,⟨−fx , −fy , 1⟩ = ⟨−3, −2, 1⟩. So, equation of Tangent plane
at P is,
z = 2 + 3(x − 2) + 2(y − 1)
The given expression can be approximated as, (by putting value of x, y in the above equation
of tangent plane) z(1.99, 1.01) ≈ 1.99. ■

Our next goal is to calculate area of different surfaces. We will start with very basic example, that is,
area of a plane.

97
25.2 Surface Area

25.2 Surface Area

Suppose P0 , P1 , P2 be the points on R3 and coordinate
vector of the points is given by,

−−→
OP0 = ⟨a0 , b0 , c0 ⟩
−−→
OP1 = ⟨a1 , b1 , c1 ⟩
−−→
OP2 = ⟨a2 , b2 , c2 ⟩

We will actually look at the parallelogram generated

by,
−−−→
v⃗1 = P0 P1 Figure 25.1: Plane S
−−−→
v⃗2 = P0 P2

Any point inside the parallelogram must look like v⃗0 +

t1 v⃗1 +t2 v⃗2 for some t1 , t2 ∈ [0, 1]. So the parallelogram
can be explicitly written as,

S = {v⃗0 + t1 v⃗1 + t2 v⃗2 | 0 ≤ t1 , t2 ≤ 1}

We know area of S is ∥v⃗1 × v⃗2 ∥. We can describe this plane differently. If the equation of the plane was
z = ax+by +c, √ then the surface of the plane can be described by S = {(x, y, ax+by +c) | (x, y) ∈ B 2 }.
2 2 2
Area of S = 1 + a + b × Area(B ). Now we should move forward to the general case.
Let, r : B 2 → R3 be a function defined as
r(x, y) = (x, y, f (x, y)) (Here f is C 1 function).
Let, ran(r) be the surface S.
As we have done in the case of Riemann Integra-
tion. We should make partition of B 2 into tiny
boxes. Let, P ∈ P(B 2 ). Then,
[
B= Bα2
α∈Λ(P)

For any α ∈ Λ(P) fix (xα , yα ) ∈ Bα2 . Consider

the tangent plane of S at r(xα , yα ) over Bα2 . Now
the Tangent Plane at r(xα , yα ) is given by, Figure 25.2: Tr(xα ,yα ) S

z − f ((xα , yα )) = (Df )(xα , yα ) · (xα , yα )

=⇒ z = (Df )(xα , yα ) + f (xα , yα )
=⇒ z = fx (xα , yα ) · (x − xα ) + fy (xα , yα ) · (y − yα ) + f (xα , yα )

So, Area of Tr(xα ,yα ) S over Bα2 is

q
1 + fx2 (xα , yα ) + fy2 (xα , yα ) × Area(Bα )2

= ∥rx × ry ∥ × Area(Bα )2

98
 Lecture 25

Since f is C 1 function so is r. So the total area of the surface S is given by,

Area(S) = lim ∥rx × ry ∥ × Area(Bα )2

∥P∥→0
Z
= ∥rx × ry ∥ dA
2
ZB q
= 1 + fx2 + fy2 dA (In this case)
B2

Will will do the above integration over any bounded set Ω as we have done in Riemann integration
chapter. Over a bounded set Ω area of S is given by,
Z q
Area(S) = 1 + fx2 + fy2 dA (25.3)
Ω
The General method for finding surface area is described by the following theorem.

Theorem 25.2.1
Let R ⊆ R2 be a region. r : R → R3 be the parametrization of the surface S. Then,
Z
Area(S) = ∥ru × rv ∥ dA
R

99
25.2 Surface Area

100
Lecture 26

26.1 Examples
Recall in previous lecture, we discussed how to find area of a surface over a bounded region. Now we
will look towards some examples.

Example 26.1.1 (Truncated Cylinder)

Let R ⊆ R2 be a region. r : R → R3 be a map defined

by r(x, y) = (cos(x), sin(x), y). Here the region is
given by,
π h πi
R = {(x, y) | 0 ≤ x ≤ , 0 ≤ y ≤ 1} = 0, × [0, 1]
2 2
It can be seen easily that r is parametrization of a
surface ran(r) = S. We want to calculate the surface
area of it.
For this case, rx = (− sin x, cos x, 0) and ry = (0, 0, 1).
So, rx × ry = (cos x, sin y, 0) and hence, ∥rx × ry ∥ = 1.
Therefore,
Z Z π2 Z 1
π
Area(S) = 1dA = 1dA = Figure 26.1: Truncated Cylinder
R 0 0 2

Example 26.1.2 (Hemisphere)


S = (x, y, z) | x2 + y 2 + z 2 = 1, z > 0 is a sur-
face. This surface actually surface of unit Hemi-
sphere.
Parametrization of the surface S is given by,
r(θ, φ) = (cos θ cos φ, sin θ cos φ, sin φ) where,
0 ≤ φ ≤ π2 , 0 ≤ θ ≤ 2π. So,

rθ = (− sin θ cos φ, cos θ cos φ, 0)

rφ = (− cos θ sin φ, − sin θ sin φ, cos φ)
∴ rθ × rφ = (cos θ cos2 φ, sin θ cos2 φ, cos φ sin φ)
=⇒ ∥rθ × rφ ∥ = cos φ
Z π
2
Z 2π Figure 26.2: Hemisphere
Hence, Area(S) = cos φ dθdφ = 2π
0 0

101
26.2 Surface Integral over Scalar fields

26.2 Surface Integral over Scalar fields

Let, R ∈ R2 be a region. r : R → R3 be parametrization of a surface S. f be a scalar function
defined on the surface S. Let f ∈ C(S) then we can define integration of f over the surface as
following,

Z X
f dS := lim f (r(xα ))∥ru (xα ) × rv (xα )∥ Area(Bα2 )
S ∥P∥→0
α∈Λ(P)
Z
= (f ◦ r)(u, v)∥ru × rv ∥ dA (26.1)
R

As we have noticed in the case of Line Integral of over a scalar field the integration do not
depend on the parametrization of the path. In this case also if we have different two parametrization
r, r̃ for the surface S then there exist a one-one continuous function φ so that the following diagram
commutes, (in other words r = r̃ ◦ φ̃)

S
r r̃

R φ R̃

We can show that the integration in (26.1) is also same if we replace r by r̃. In other words the
surface Integral over a surface is independent of its parametrization (See Figure 26.3).

Example 26.2.1 (Surface Integral over a Cone)

We want to evaluate Z
(x2 + y 2 + z 2 )dS
S
n p o
where, S = (x, y, z) | z = x2 + y 2 , 0 ≤ z ≤ 1
  p 
Solution. Let,R = (x, y)|x2 + y 2 ≤ 1 be the region and r(x, y) = x, y, x2 + y 2 is the
   
x y
parametrization of the surface S. Here, rx = 1, 0, √ 2 2 and ry = 0, 1, √ 2 2 .
x +y x +y

Z Z
∴ (x2 + y 2 + z 2 ) dS = (f ◦ r) · ∥rx × ry ∥ dA
S R
s
x2 y2
Z
= (f ◦ r) 1 + 2 2
+ 2 dA
R x +y x + y2
√ Z
= 2 2 (x2 + y 2 ) dA
R
√ Z 1 Z 2π 3
=2 2 r dθ dr (It’s the polar substitution)
0 0
√
= 2π

26.3 Surface Integral over a Vector field

Let F⃗ : S → R3 be a vector field defined on a surface S. We want to calculate the flux of F⃗ over the
surface S.

102
 Lecture 26

Let S be a surface and F⃗ (This function is

C 1 over S) be the Vector field defined over
it. We can assume F⃗ to be a force. We
want to compute the extent to which F⃗ is
pushing the surface along normal to S. In
other words, we want to calculate Flux /
Flow of F⃗ through S.

Let dS be a tiny part of this surface. Flux

through dS is given by F⃗ · ⃗n dS where, ⃗n is
the unit normal vector through the surface
at a point x ∈ dS. Since dS is small enough
we can assume ⃗n is almost constant over dS. Figure 26.3: A surface with Vector field
So, the total flux must be
Z
Flux = F⃗ · ⃗n dS
S
But, there is a problem! If the surface S has a point where two normal vectors with different
direction are the at same point then the concept of flux does not make any sense. That’s why
we have to come up with some conditions. Namely, Orientation of Surface.

So we formalize the notion of “Orientation” with the following definition. Thought it is only for the
surfaces in R3 . Later in Differential Geometry, you will see this notion in a more general way (for
manifolds).

Definition 26.3.1 ▶ Oriented Surface

A surface is oriented if there exists a continuous function ⃗n : S → R3 such that ⃗n(x) is
normal to S at the point x and ∥⃗n(x)∥ = 1.

Surfaces like Möbius strip or Klein Bottle are not oriented.

Definition 26.3.2 ▶ Surface Integral over Vector field

Let, S be an oriented surface with normal vector ⃗n and F⃗ : S → R3 be a C 1 vector field
defined over S. Then the “Surface Integral” of F⃗ over S is defined as following,
Z Z
⃗ ⃗
F · dS = F⃗ · ⃗n dS
S S

103
26.3 Surface Integral over a Vector field

104
Lecture 27

27.1 Conservative Vector Fields

In the previous lecture we introduced the notion of an oriented surface. For an oriented surface
S ⊆ R3 , we call the orientation vector field ⃗n : S → R3 the normal vector field. Now we give an
example of such a vector field.

Example 27.1.1

Take Sn−1 := {x ∈ Rn | ∥x∥ = 1}.

Then
⃗n1 (x) = x ∀ x ∈ Sn−1
and
⃗n2 (x) = −x ∀ x ∈ Sn−1
are the two normal vector fields on the sphere.

Generally we consider the outward normal vector, i.e.,

the normal vector field given by ⃗n1 as the standard Figure 27.1: Standard normal vector
normal vector field on the sphere. field on a sphere

Formula for Normal Vector Field.

Let G(f ) = {(x, y, f (x, y)) | (x, y) ∈ O2 } where f :
O2 → R is a C 1 function. Then a parametrization of
the surface G(f ) is given by the function

⃗r : O2 → R3
(x, y) 7→ (x, y, f (x, y))

Then we have

⃗rx × ⃗ry = (−fx , −fy , 1) Figure 27.2: Normal vector fields on

a graph surface
Then a normal vector field is given by

⃗rx × ⃗ry
⃗n(x, y) =
∥⃗rx × ⃗ry ∥

Unless otherwise mentioned this will be our standard orientation of the normal vector field.
Z
Usually computation of F⃗ · dS
⃗ is complicated, let us look at some examples to gain more familiarity.
S

105
27.1 Conservative Vector Fields

Example 27.1.2

Consider the vector field F⃗ (x, y, z) = (x, y, z) on S = ran(r), where

π
⃗r(x, y) = (cos x, sin x, y) 0≤x≤ ,0≤y≤1
2
Then ⃗rx × ⃗ry = (cos x, sin x, 0), so ⃗n(x, y) = (cos x, sin x, 0) is a normal vector field.
Z Z
F⃗ · dS
⃗= F⃗ · ⃗nds
S S
π
Z1 Z 2
= F⃗ (⃗r(x, y)) · (⃗rx × ⃗ry ) dA
0 0
π
Z 1 Z 2
= (cos x, sin x, y) · (cos x, sin x, 0)dA
0 0
π
Z 1Z 2
= dA
0 0
π
=
2

Z
We already know that ∇f · dr = f (B) − f (A), now a natural question that arises is
C

Question: Given F⃗ , does there exist f a scalar field such that ∇f = F⃗ ?

Definition 27.1.1 ▶ Conservative Vector Field

A vector field F⃗ on On is called conservative if there exists a scalar field f ∈ C 1 (On ) such
that ∇f = F⃗ , then f is called the potential function.

Theorem 27.1.1
Let F⃗ be a vector field over On , the following are equivalent:

1. F⃗ is conservative.
Z
2. F⃗ · dr = 0, for all closed and piecewise smooth curve C.
C
Z Z
3. F⃗ · dr = F⃗ · dr, for all curves C1 and C2 with same initial and end points.
C1 C2

Question: Given a vector field F⃗ , can we conclude F⃗ is conservative? (NO!)

We will give a general picture for the most common case, when n = 3. Let F⃗ = (P, Q, R) where
P, Q, R are scalar fields. Now if F⃗ = ∇f for some scalar field f , then we would have

∂f
fx ≡ =P
∂x
∂f
fy ≡ =Q (27.1)
∂y
∂f
fz ≡ =R
∂z

106
 Lecture 27

Then we can define curl of a vector field

î ĵ k̂
∇ × F⃗ := ∂
∂x
∂
∂y
∂
∂z
P Q R

Then expanding this out and using the relations (27.1) and others we get that ∇ × F⃗ = 0. So, we
have proved that if F⃗ is conservative then ∇ × F⃗ = 0.

Remark. Thus, a necessary condition for a vector field to be conservative is that, its curl should be
the zero vector field.

Example 27.1.3

Let F⃗ (x, y) = (y − 3, x + 2) = (P, Q) (say), then ∂P

∂y = ∂Q
∂x = 1. Let f be a possible potential
function, then
∂f ∂f
= y − 3 and =x+2
∂x ∂y
Then by Fundamental Theorem of Calculus (assuming domain is convex) we get

f (x, y) = xy − 3x + g(y)
∂f
But then using ∂y = x + 2 we get

∂f
x + g ′ (y) = = x + 2 ⇒ g ′ (y) = 2
∂y

Therefore taking f (x, y) = xy − 3x + 2y gives us a potential function for the vector field F⃗ .

Remark. This approach works for all F⃗ such that ∇ × F⃗ = 0 and the domain is convex.

Example 27.1.4
 
−y
Let F⃗ (x, y) = x2 +y 2
x
, x2 +y 2 = (P, Q) (say) on R2 \ {0}. Then we have ∂P
∂y = ∂Q
∂x , but we
will show that F⃗ is not conservative. Consider the curve

C : γ(t) = (cos t, sin t), 0 ≤ t ≤ 2π

then
Z Z 2π
F⃗ · dr = F⃗ (γ(t)) · γ ′ (t) dt
C 0
Z 2π
= (− sin t, cos t) · (− sin t, cos t) dt
0
Z 2π
= dt
0
= 2π
Z
But C is clearly a closed curve, hence by Theorem 27.1.1 we must have F⃗ · dr = 0.
C
(Contradiction!)

107
27.2 Green’s Theorem

27.2 Green’s Theorem

Definition 27.2.1 ▶ Simply Connected Domain

Let D be an open and connected set. Let C be a simple and closed curve if C can be shrunk
continuously to a point inside D, then we say D is simply connected.

Figure 27.3: Examples of simply connected and not simply connected region

Theorem 27.2.1 (Green’s Theorem)

Let R ⊆ R2 be a simply connected domain with boundary curve C where parametrization is
taken in anti-clockwise direction. Let F⃗ = (P, Q) be a C 1 vector field on R, then
Z Z Z  
∂Q ∂P
F⃗ · dr := P dx + Q dy = − dA
C C R ∂x ∂y

What happens when R is not simply connected?

Figure 27.4: You break up the region C with the hole into two regions without holes C1 and C2 .

108
 Lecture 27

Z Z Z
P dx + Q dy = P dx + Q dy + P dx + Q dy
C C̃1 C̃2
Z Z
= (Qx − Py ) dA + (Qx − Py ) dA
R1 R2
Z
= (Qx − Py ) dA
R

109
27.2 Green’s Theorem

110
Lecture 28

28.1 Green’s Theorem

Theorem 28.1.1 (R2 version of Green’s Theorem)

Let R ⊆ R2 be a simply connected domain with boundary curve C where parametrization is
taken in anti-clockwise direction. Let F⃗ = (P, Q) be a C 1 vector field on R, then
Z Z Z  
∂Q ∂P
F⃗ · dr := P dx + Q dy = − dA
C C R ∂x ∂y

Proof .
(for Simple region)
Let R = {(x, y) | a ≤ x ≤ b, φ1 (x) ≤ y ≤ φ2 (x)}
be a simple region. Here C = C1 ∪ V2 ∪ C2 ∪ V1
is the curve bounding the region along anti-
clockwise direction (as shown in Figure 28.1).

Now,
Z Z b Z φ2 (x)
∂P ∂P
− dA = − dy dx
R ∂y a φ1 (x) ∂y
Z b
=− (P (x, φ2 (x)) − P (x, φ1 (x))) dx
a
Figure 28.1: A simple region

The curves C1 , C2 , V1 , V2 can be explicitly written as,

V1 = {(a, t) | φ1 (a) ≤ t ≤ φ2 (a)}

C1 = {(x, φ1 (x)) | a ≤ x ≤ b}
V2 = {(a, t) | φ1 (b) ≤ t ≤ φ2 (b)}
C2 = {(x, φ2 (x)) | a ≤ x ≤ b}

We compute the integrals for P over these curves and obtain,

Z Z
dx(t)
P dx = P (x(t), y(t)) dt = 0
V1 V1 dt
Z Z b
P dx = P (t, φ1 (t)) dt
C1 a
Z Z b
P dx = P (t, φ2 (t)) dt
C2 a

111
28.1 Green’s Theorem

Z Z
∂P
=⇒ P dx = − dA
C R ∂y
Z Z
∂Q
By similar mechanism we can show Q dy = dA. The rest follows from here.
C R ∂y

Example 28.1.1

Let C be the boundary of [0, 1]2 , i.e., ∂[0, 1] × [0, 1] = C. Evaluate

Z
⟨x2 − y 2 , 2xy⟩
C

Solution. We can decompose C = C1 ∪ C2 ∪ C3 ∪ C4 (as in the following picture)

Let P (x, y) = x2 − y 2 , Q(x, y) = 2xy. Then the integral,

Z ZZ
P dx + Q dy = (2y + 2y) dA (Green’s Theorem)
C [0,1]2
Z 1Z 1
= 4y dy dx
0 0
=2
Figure 28.2: ∂([0, 1]2 )
If we try to calculate the integral directly, we will end up getting
same result.

Area of a closed Region. Let R (simply connected) be a closed region and C = ∂R be the
curve enclosing the region. Using Green’s Theorem we get,
xdy − ydx
Z Z Z Z
Area(R) = dA = xdy = −ydx =
R C C C 2

x2 y2
Example 28.1.2 (Area inside the ellipse: a2 + b2 = 1)

Solution. Parametrization of ellipse x = a cos t, y = b sin t where t ∈ [0, 2π). Using the above
application of Green’s Theorem we can write,
Z Z 2π
Area = xdy = ab cos2 tdt = πab
C 0

Theorem 28.1.2 (Independence of path)

Z
Let F⃗ be a C 1 vector field on R2 such that F⃗ · d⃗r is independent of path. Then F⃗ is
C
conservative over an open and simply connected domain.

Proof. Let D be an open and connected domain. F⃗ = ⟨P, Q⟩ is defined over D. Also let P0 = ⟨x0 , y0 ⟩
be a fixed point in the domain D and P1 = ⟨x, y⟩ ∈ D be a variable point. C be a smooth curve
joining P0 and P1 . Define
Z
φ(x, y) = F⃗ · d⃗r
C

112
 Lecture 28

Since, D is open set, so we must get an open ball centered at P1 contained in D. Take a point
P1′ = ⟨x1 , y⟩ inside that open ball such that x1 < x. Let C1 be a smooth curve from P0 to P1 and C2
be a line segment from P1′ to P1 . So, C1 ∪ C2 defines a smooth curve from P0 to P1 .

C2
P1′ (x1 , y) P1 (x, y)
C1

C
P0 (x0 , y0 )
Z
As F⃗ · d⃗r is path independent We can write,
C
Z Z
φ(x, y) = F⃗ · d⃗r + F⃗ · d⃗r
C1 C2

Now we take the partial derivative of both sides of this equation with respect to x. The first integral
does not depend on the variable x since C1 is the path from P0 (x0 , y0 , z0 ) to P1′ (x1 , y, z) and so
partial differentiating this line integral with respect to x is zero.

Z Z 
∂φ ∂ ⃗ ⃗
= F · d⃗r + F · d⃗r
∂x ∂x C C
Z 1  2 Z 
∂ ⃗ ∂ ⃗
= F · d⃗r + F · d⃗r
∂x C1 ∂x C2
| {z }
=0

Also, C2 can be parametrized as r(t) = ⟨t, y⟩ where t ∈ [x1 , x]. So,

Z  Z x 
∂ ∂
F⃗ · d⃗r = ⟨P (t, y), Q(t, y)⟩ · ⟨1, 0⟩ dt
∂x C2 ∂x x
Z x1 
∂
= P (t, y) dt
∂x x1
= P (x, y) [Fundamental Theorem of calculus]

Similarly, we can show that, ∂φ

∂y = Q(x, y). And hence, ∇φ = F⃗ (x, y). We can define φ as the
potential of F⃗ .

Theorem 28.1.3
Let D be a simply connected domain in R2 and F⃗ is a C 1 vector field on D. Then F⃗ is
conservative iff ∇ × F⃗ = 0 on D.

Proof. (⇒) This direction is trivial.

Z
(⇐) From Green’s Theorem we can say that F⃗ · d⃗r = 0 over all closed curve C. For any two point
C
p0 , p1 ∈ D if γ1 , γ2 : [0, 1] → D are two smooth curves joining p0Z and p1 . (i.e.,
Z γ1 (0) = γ2 (0) = p0 and
γ1 (1) = γ2 (1) = p1 ) then γ1 ∪ γ2 (1 − t) is a closed curve. So, F⃗ · d⃗r = F⃗ · d⃗r. Which means
γ1 γ2
the integral is path independent. Using the previous theorem we can say, F⃗ is conservative on D.

113
28.2 Gauss Divergence Theorem

28.2 Gauss Divergence Theorem

Definition 28.2.1 ▶ Divergence of a vector field

⃗ = (f1 , · · · , fn ) : Rn → Rn , the “Divergence” of F
Given a vector field F ⃗ is,
n
X ∂fi ⃗
div(F ) = ≡∇·F
i=1
∂x i

Theorem 28.2.1 (Gauss Divergence Theorem)

Let D ⊆ R3 a solid domain, ∂D be an oriented surface. Let F⃗ = ⟨P, Q, R⟩ be a C 1 vector

field on an open surface containing D ∪ ∂D. Then,
Z Z
F⃗ · dS
⃗= ∇ · F⃗ dV
| ∂D=S{z } |D {z }
surface integral volume integral

Just like FTC, the behavior over a volume is fully determined by the behavior at the boundary.
Proof of this theorem is beyond our reach. But we can see the proof for simple cases.

Proof. (For a simple case) Consider D = {(x, y, z) | φ1 (x, y) ≤ z ≤ φ2 (x, y), (x, y) ∈ [a, b] × [c, d]}.
(Exercise.) Complete the proof!

Example 28.2.1

F (x, y, z) = x + y, z 2 , x2 and S be the hemisphere x2 + y 2 + z 2 = 1, z > 0. Compute,

Z
F⃗ dS
⃗
S

Solution. Notice that S is open surface. We want to use Gauss Theorem 28.2.1. So we need a
close surface. Let S1 be the surface x2 + y 2 ≤ 1. Then S ⊔ S1 is a closed surface.

Z Z
F⃗ · dS
⃗= ∇ · F⃗ dV
S⊔S1 x2 +y 2 ,z 2 ≤1,z≥0
Z
= dV
x2 +y 2 ,z 2 ≤1,z≥0
2π
=
3

Parametrization of the surface S1 = (x, y, 0) | x2 + y 2 = 1 . So, rx × ry = ⟨1, 0, 0⟩ × ⟨0, 1, 0⟩.

Z Z Z
F⃗ · dS
⃗= x + y, z , x 2 2
· ⟨0, 0, 1⟩ dA = x2 dA
S1 x2 +y 2 ≤1 x2 +y 2 ≤1
Z 2π Z 1
π
= r3 cos2 θ dr dθ =
0 0 4
Z
⇒ ⃗ = 11π
F⃗ · dS
S 12

114
 Lecture 28

28.3 Stokes’ Theorem

Theorem 28.3.1 (Stokes’ Theorem)

Let C be a C 1 curve enclosing an oriented surface S in R3 . Let, F⃗ = ⟨P, Q, R⟩ be a C 1 vector

field on an open set containing S. Then,
Z Z  
⃗ · d⃗r =
F ⃗ · dS
∇×F ⃗
C S
Here orientation of S and direction of C is same.

Example 28.3.1
Z
Compute F⃗ · d⃗r, where C : x2 + y 2 = 9, z = 4 and F⃗ = ⟨−y, x, xyz⟩.
C

Solution. ∇ × F⃗ = ⟨xz, −yz,, 2⟩. By convention, we should assume direction of C is along

counter-clockwise direction. So, The normal vector of S is along negative z axis. So, required
integral,

Z Z
F⃗ · d⃗r = (∇ × F⃗ ) · dS
⃗
C S
Z
= (∇ × F⃗ ) · ⟨0, 0, −1⟩ dA
x2 +y 2 ≤1,z=4
Z
= −2 dA
x2 +y 2 ≤1,z=4
= −18π

Stoke’s Theorem is the R3 −analogue of Green’s Theorem 28.1.1. If we take the third component of
F⃗ to be zero, i.e., R = 0, then Stoke’s Theorem 28.3.1 gives us back Green’s Theorem 28.1.1.

There is a generalized version of Stokes’ theorem. Just for information the theorem is stated
below.
• If Ω is an oriented n-manifold (with boundary) and ω is a differential form ((n − 1) form).
Then integral of ω over the boundary ∂Ω of the manifold Ω is given by,
Z Z
ω= dω
∂Ω Ω

115
28.3 Stokes’ Theorem

116
Index

L Lipschitz 3
Linear map 2

117