Real Analysis-II

MULTIVARIABLE CALCULUS

s s
te ysi
Department of Mathematics

March 26, 2026

l
No a
e n
Contents
ur l A
1 Partial Derivatives 2

2 Introduction 4
ct ica
3 The Total Derivative 9

4 Total Derivatives in Terms of Partial Derivatives 12

Le at

5 The Matrix of a Linear Transformation 17

6 Composition of Linear Functions 17

ES em

7 The Jacobian Matrix 19

8 Leibnitz Rule for Differentiation under the Integral sign 25

h

9 Mean Value Theorem for Differentiable functions 26

at

10 Sufficient Condition for Differentiability 29

M

1
 1 Partial Derivatives
Definition
Let S ⊂ Rn be an open set and let f : S → R be a real-valued function.

Let x = (x1 , x2 , . . . , xn ) denote a general point of Rn , and let a = (a1 , a2 , . . . , an ) ∈ S.

For 1 ≤ k ≤ n, suppose that for all sufficiently small h, (a1 , . . . , ak + h, . . . , an ) ∈ S. If

s s
the limit

te ysi
f (a1 , . . . , ak + h, . . . , an ) − f (a1 , . . . , ak , . . . , an )
lim
h→0 h
exists, then this limit is called the partial derivative of f with respect to the variable xk

l
at the point aIt is denoted by

No a
∂f
(a), Dk f (a), fxk (a).

e n
∂xk
ur l A
Remark The existence of the partial derivatives Dk f (a) does not imply that f is con-
tinuous at a, nor does it imply that f is differentiable at a.
Example
Let f (x, y) = x2 y + 3xy 2 .Partial derivative with respect to x
ct ica
By definition,
∂f f (x + h, y) − f (x, y)
(x, y) = lim .
∂x h→0 h
Le at

f (x + h, y) = (x + h)2 y + 3(x + h)y 2

= x2 y + 2xhy + h2 y + 3xy 2 + 3hy 2
ES em

f (x + h, y) − f (x, y) = (2xhy + h2 y + 3hy 2 )

f (x + h, y) − f (x, y)
= 2xy + hy + 3y 2
h
f (x + h, y) − f (x, y)
h

∴ lim = 2xy + 3y 2
h→0 h
∂f
at

(x, y) = 2xy + 3y 2
∂x
By definition,
M

∂f f (x, y + h) − f (x, y)
(x, y) = lim .
∂y h→0 h

2
 f (x, y + h) = x2 (y + h) + 3x(y + h)2
= x2 y + x2 h + 3xy 2 + 6xyh + 3xh2
f (x, y + h) − f (x, y) = x2 h + 6xyh + 3xh2 .
f (x, y + h − f (x, y)
= x2 + 6xy + 3xh
h
f (x, y + h) − f (x, y)
∴ lim = x2 + 6xy.

s s
h→0 h

te ysi
∂f
(x, y) = x2 + 6xy
∂y
Example
Let

l

x + y, if x = 0 or y = 0,

No a

f (x, y) =
1, otherwise.

e n


ur l A
First note that f (0, 0) = 0.

Partial derivative with respect to x:

ct ica
∂f f (h, 0) − f (0, 0)
(0, 0) = lim .
∂x h→0 h

Since y = 0, f (h, 0) = h. Hence ∂f

(0, 0) = limh→0 h
= 1.
Le at

∂x h

Partial derivative with respect to y:

ES em

∂f f (0, h) − f (0, 0)
(0, 0) = lim .
∂y h→0 h
h

Since x = 0, f (0, h) = h. Thus ∂f

∂y
(0, 0) = limh→0 h
h
= 1.
at

Hence both partial derivatives exist.

Now along the path y = x with x ̸= 0, f (x, x) = 1. Thus

M

lim f (x, y) = 1 ̸= 0 = f (0, 0).

(x,y)→(0,0)

Therefore f is not continuous at (0, 0) even though the partial derivatives exist.

Definition 1.1 (Derivative of a Vector–Valued Function). Let

f : (a, b) → Rn

3
 be a vector–valued function. Then

f (x) = (f1 (x), f2 (x), . . . , fn (x)),

where each fk : (a, b) → R is a real–valued function.

We say that f is differentiable at c ∈ (a, b) if each component fk is differentiable at c.

s s
The derivative of f at c is defined by

te ysi
f ′ (c) = (f1′ (c), f2′ (c), . . . , fn′ (c)).

l
Equivalently,

No a
f (c + h) − f (c)
f ′ (c) = lim .

e n
h→0 h

2
ur l A
Introduction
• Partial derivatives of functions f : Rn → R were introduced earlier.
ct ica
• Derivatives of vector-valued functions f : R → Rm were also discussed.

• The present chapter extends derivative theory to functions

Le at

f : Rn → Rm .

• The partial derivative is a generalization of the ordinary derivative.

ES em

• However, the existence of all partial derivatives

D1 f, D2 f, . . . , Dn f
h
at

at a point does not necessarily imply continuity of f at that point.

• A limitation of partial derivatives is that they treat a function of several variables

M

as a function of one variable at a time.

• The partial derivative describes the rate of change of a function in the direction of
each coordinate axis.

• A more general concept is the directional derivative, which measures the rate of
change of a function in an arbitrary direction.

• Directional derivatives apply to both real-valued and vector-valued functions.

Definition (Directional Derivative).

4
 s s
te ysi
l
No a
e n
ur l A Figure 1: Geometric View of Directional Derivatives

Let S ⊂ Rn and let

f : S → Rm
ct ica
be a function. Let c be an interior point of S and let u ∈ Rn be a direction vector.

The directional derivative of f at c in the direction u, denoted by f ′ (c; u), is defined

by
Le at

f (c + hu) − f (c)
f ′ (c; u) = lim ,
h→0 h
ES em

provided the limit exists.

h

Remark. Some authors assume ∥u∥ = 1, but this is not necessary.

at

Important points on Directional Derivatives

M

1. Relation with Partial Derivatives

Let uk = (0, . . . , 0, 1, 0, . . . , 0) be the k-th unit coordinate vector. Then the directional
derivative in this direction is called the partial derivative. Thus

f ′ (c; uk ) = Dk f (c).

Hence a partial derivative is a special case of a directional derivative.

5
 2. Directional Derivative of Vector–Valued Functions

Let
f = (f1 , f2 , . . . , fm ) : S → Rm .

Then f ′ (c; u) exists if and only if each component directional derivative fk′ (c; u) exists for
k = 1, 2, . . . , m. In that case

s s
te ysi
f ′ (c; u) = (f1′ (c; u), f2′ (c; u), . . . , fm
′
(c; u)).

l
3. Special Case: Partial Derivatives

No a
If u = uk , then

e n
ur l A Dk f (c) = (Dk f1 (c), Dk f2 (c), . . . , Dk fm (c)).

Thus partial derivatives of a vector–valued function are obtained by differentiating each

component.
ct ica
4. Relation with Ordinary Derivative

Let
Le at

F (t) = f (c + tu).
ES em

Then

F ′ (0) = f ′ (c; u).

h

More generally, if the derivative exists,

at

F ′ (t) = f ′ (c + tu; u).

M

Thus a directional derivative can be viewed as the ordinary derivative of f along the line
x = c + tu.

5. Example

Let
f (x) = ∥x∥2 .

6
 Define

F (t) = f (c + tu).

Then

s s
F (t) = (c + tu) · (c + tu) = ∥c∥2 + 2t(c · u) + t2 ∥u∥2 .

te ysi
Differentiating with respect to t, we obtain

l
No a
F ′ (t) = 2c · u + 2t∥u∥2 .

e n
Hence
ur l A F ′ (0) = 2c · u.

Therefore the directional derivative is

ct ica

f ′ (c; u) = 2c · u.
Le at

6. Linear Functions
ES em

A function f : Rn → Rm is called linear if

f (ax + by) = af (x) + bf (y)

h

for all x, y ∈ Rn and scalars a, b.

at

If f is linear, then
M

f ′ (c; u) = f (u)

for every c and every direction u.

Directional Derivatives and Continuity

If the directional derivative f ′ (c; u) exists for every direction u, then in particular all
partial derivatives

7
 D1 f (c), D2 f (c), . . . , Dn f (c)

exist. However, the converse is not true.

Example

s s
Consider the function f : R2 → R defined by

te ysi

x + y,

if x = 0 or y = 0,
f (x, y) =
1, otherwise.

l


No a
Then

e n
ur l A D1 f (0, 0) = D2 f (0, 0) = 1.

However, if we consider a direction u = (a1 , a2 ) where a1 ̸= 0 and a2 ̸= 0, then

ct ica
f (0 + hu) − f (0) f (ha1 , ha2 ) 1
= = .
h h h

This expression does not approach a limit as h → 0. Hence the directional derivative in
Le at

such directions does not exist.

ES em

Thus, existence of partial derivatives does not guarantee the existence of directional
derivatives in all directions.
Example
h

Define
at

xy 2

, x ̸= 0,


f (x, y) =  x2 + y 4

M

0,

x = 0.

Let u = (a1 , a2 ) be any direction. Then

f (hu) − f (0)
f ′ (0; u) = lim .
h→0 h

Since

8
 (ha1 )(ha2 )2
f (hu) = f (ha1 , ha2 ) = ,
(ha1 )2 + (ha2 )4

we obtain

f (hu) − f (0) a1 a22

= 2 .

s s
h a1 + h2 a42

te ysi
Taking limit as h → 0,

a22
f ′ (0; u) =

l
.
a1

No a
Thus directional derivatives exist in every direction.

e n
However, along the path x = y 2 ,
ur l A y4 1
f (y 2 , y) = = .
2y 4 2
ct ica
Hence

1
lim f (x, y) = ̸= f (0, 0).
2
Le at

(x,y)→(0,0)

Therefore f is not continuous at (0, 0).

ES em

3 The Total Derivative

In one variable calculus, if f has a derivative at c, then f (c + h) can be approximated by
h

a linear function.
at

Let
f (c + h) − f (c)

− f ′ (c), h ̸= 0,



Ec (h) =
M

h
0, h = 0.



Then
f (c + h) = f (c) + f ′ (c)h + hEc (h).

The term hEc (h) represents the error in approximating f (c+h)−f (c) by f ′ (c)h. Moreover,

Ec (h) → 0 as h → 0.

9
 Thus the derivative provides a linear approximation to the function near c.

Definition (Total Derivative)

Let f : S ⊂ Rn → Rm and let c be an interior point of S. The function f is said to be

differentiable at c if there exists a linear function

s s
te ysi
Tc : Rn → Rm

such that

l
No a
f (c + v) = f (c) + Tc (v) + ∥v∥Ec (v)

e n
where ur l A
Ec (v) → 0 as v → 0.
ct ica
The linear function Tc is called the total derivative of f at c.

Equation above is also written as

Le at

f (c + v) = f (c) + Tc (v) + o(∥v∥) as v → 0.

ES em

This is called the first-order Taylor formula.

Theorem 3.1. Assume f is differentiable at c with total derivative Tc . Then the direc-
h

tional derivative f ′ (c; u) exists for every u ∈ Rn and

at

Tc (u) = f ′ (c; u).

M

Proof. Since f is differentiable at c, there exists a linear map Tc : Rn → Rm such that

f (c + v) = f (c) + Tc (v) + ∥v∥Ec (v),

where Ec (v) → 0 as v → 0. Let v = hu where h is a real number. Then

f (c + hu) = f (c) + Tc (hu) + ∥hu∥Ec (hu).

10
 Since Tc is linear,
Tc (hu) = hTc (u).

Hence
f (c + hu) − f (c) = hTc (u) + |h|∥u∥Ec (hu).

Dividing by h,
f (c + hu) − f (c)

s s
|h|
= Tc (u) + ∥u∥Ec (hu).
h h

te ysi
Taking limit as h → 0, we obtain

f (c + hu) − f (c)
lim = Tc (u).

l
h→0 h

No a
Thus the directional derivative exists and

e n
ur l A f ′ (c; u) = Tc (u).

Theorem 3.2. If f is differentiable at c, then f is continuous at c.

ct ica
Proof. Since f is differentiable at c,
Le at

f (c + v) = f (c) + Tc (v) + ∥v∥Ec (v),

where Ec (v) → 0 as v → 0.
ES em

Let v → 0. Since Tc is linear, we have

h

Tc (v) → 0.
at

Also,
M

∥v∥Ec (v) → 0.

Therefore

f (c + v) → f (c).

Hence

11
 lim f (c + v) = f (c),
v→0

so f is continuous at c.

Remark

es sis
The total derivative Tc is often written as f ′ (c). With this notation the Taylor formula
becomes

ot ly
f (c + v) = f (c) + f ′ (c)(v) + ∥v∥Ec (v).

N a
Here f ′ (c) is a linear transformation, not a number.

e An
Example

If f itself is a linear function, then

ur l
f (c + v) = f (c) + f (v).
ct ca
Hence the derivative exists at every point and
Le ati

f ′ (c) = f.
ES em

Thus the total derivative of a linear function is the function itself.

4 Total Derivatives in Terms of Partial Derivatives

h

Theorem 4.1. Let f : S → Rm be differentiable at an interior point c of S, where

at

S ⊆ Rn .

Let v ∈ Rn be written as
M

v = v1 u1 + v2 u2 + · · · + vn un ,

where
ui = (0, 0, . . . , 1, . . . , 0) ∈ Rn

with 1 in the ith place and 0 elsewhere.

12
 Then the directional derivative of f at c in the direction v is
n
f ′ (c)(v) = vk Dk f (c).
X

k=1

Proof. Since f : S → Rm is differentiable at the interior point c ∈ S, there exists a

linear transformation Tc : Rn → Rm such that

s s
∥f (c + h) − f (c) − Tc (h)∥

te ysi
lim = 0.
h→0 ∥h∥

The linear map Tc is the derivative of f at c. Hence

l
No a
f ′ (c) = Tc .

e n
Let v ∈ Rn . Since {u1 , u2 , . . . , un } is the standard basis of Rn , we can write
ur l A v = v1 u1 + v2 u2 + · · · + vn un .
ct ica
Using the linearity of Tc ,
n n
!
Tc (v) = Tc vk uk = vk Tc (uk ).
X X
Le at

k=1 k=1

But Tc (uk ) is the partial derivative of f with respect to the k th variable at c, i.e.,
ES em

Tc (uk ) = Dk f (c).

Therefore
h

n
f ′ (c)(v) = Tc (v) = vk Dk f (c).
X
at

k=1

Hence the result.

M

Special Case: Real–valued functions

Let f : S ⊆ Rn → R be differentiable at c and let v = (v1 , v2 , . . . , vn ) ∈ Rn .

From the general formula for the derivative we have

n
f ′ (c)(v) = vk Dk f (c),
X

k=1

13
 ∂f
where Dk f (c) = (c).
∂xk
Since f is real–valued, the quantities Dk f (c) are real numbers. Define the vector

∇f (c) = (D1 f (c), D2 f (c), . . . , Dn f (c)) ,

es sis
called the gradient of f at c.

Then
f ′ (c)(v) = v1 D1 f (c) + v2 D2 f (c) + · · · + vn Dn f (c) = ∇f (c) · v.

ot ly
Thus, for real–valued functions the derivative in the direction v is the dot product of

N a
the gradient vector with the direction vector.

e An
Example 4.2. Illustrate the theorem for a function f : R4 → R5 .

Let  

x1 + x 2 
x x
 

 2 3 

f (x1 , x2 , x3 , x4 ) =  x 2
+ x .
ur l
 
 3 4 
ct ca
x 1 x4
 
 
 
x1 + x 2 + x 3 + x 4
Le ati

Let c = (1, 1, 1, 1) and v = (v1 , v2 , v3 , v4 ).

Step 1: Partial derivatives

ES em

   
1
 
1
 
0  x3 
   
   
D1 f (x) =  0 , D2 f (x) =  0
   
   
x4  0
   
h

   
1 1
at

   

0  0
 
 x2  0
M

   
   
D3 f (x) = 2x3  , D4 f (x) =  1 
   
   
 0  x1 
   
   
1 1

Step 2: Evaluate at c = (1, 1, 1, 1)

14
    
1
   
1
0 1
   
   
D1 f (c) =  0 , D2 f (c) = 0
   
   
1 0
   
   
1 1

es sis
   
0
   
0
1 0
   

ot ly
   
D3 f (c) = 
2 ,

D4 f (c) = 1
 
   
0 1
   

N a
   
1 1

e An
Step 3: Apply the theorem

f ′ (c)(v) = v1 D1 f (c) + v2 D2 f (c) + v3 D3 f (c) + v4 D4 f (c)

ur l
ct ca
       
1
 
1
 
0
 
0
 
0 1 1 0
       
       
= v1 0 + v2 0 + v3 2 + v4 1
       
Le ati

       
1 0 0 1
       
       
1 1 1 1
ES em

Step 4: Simplifying

 

v1 + v2 
v2 + v3
 
h

 
 
f ′ (c)(v) = 
 2v3 + v4 .

 
v1 + v4
at

 
 
 
v1 + v2 + v3 + v4
M

Thus the derivative f ′ (c)(v) is the linear combination

4
f (c)(v) =
′
vk Dk f (c),
X

k=1

which illustrates the theorem.

Theorem 4.3. Let u and v be two real–valued functions defined on a subset S of the
complex plane. Assume also that u and v are differentiable at an interior point c of S

15
 and that the partial derivatives satisfy the Cauchy–Riemann equations at c.

Let
f = u + iv.

Then the function f has a derivative at c. Moreover

s s
f ′ (c) = D1 u(c) + iD1 v(c).

te ysi
Proof. We have
f (z) − f (c) = u(z) − u(c) + i{v(z) − v(c)}

l
for each z in S.

No a
Since each of u and v is differentiable at c, for z sufficiently near c we have

e n
ur l A u(z) − u(c) = ∇u(c) · (z − c) + o(∥z − c∥)

and
v(z) − v(c) = ∇v(c) · (z − c) + o(∥z − c∥).
ct ica
Using vector notation and writing z = x + iy and c = a + ib, we find

v(z) − v(c) = {∇u(c) + i∇v(c)} · (z − c) + o(∥z − c∥).

Le at

Thus
ES em

v(z)−v(c) = D1 u(c)(x−a)+D2 u(c)(y −b)+i{D1 v(c)(x−a)+D2 v(c)(y −b)}+o(∥z −c∥).

h

Because of the Cauchy–Riemann equations,

at

D1 u(c) = D2 v(c), D1 v(c) = −D2 u(c),

the above expression becomes

M

f (z) − f (c) = {D1 u(c) + iD1 v(c)}(z − c) + o(∥z − c∥).

Dividing by z − c and letting z → c we see that f ′ (c) exists and

f ′ (c) = D1 u(c) + iD1 v(c).

16
 The Cauchy–Riemann equations are therefore a necessary and sufficient condition for
the differentiability of a complex function f (z) = u(x, y) + iv(x, y) when u and v are
differentiable.

5 The Matrix of a Linear Transformation

s s
Definition 5.1. Let

te ysi
T : Rn → Rm

be a linear function. Let u1 , . . . , un denote the unit coordinate vectors in Rn . If x ∈ Rn ,

then

l
No a
x = x1 u1 + · · · + xn un .

e n
By linearity,
n
T (x) = xk T (uk ).
X
ur l A k=1

Thus the linear function T is completely determined by its action on the coordinate
vectors u1 , . . . , un .Write
m
T (uk ) =
ct ica
X
tik ei
i=1

where e1 , e2 , ..., em are the standard basis vectors. The scalars t1k , . . . , tmk are the coor-
dinates of T (uk ). We display these scalars as the column vector
Le at

 
t
 1k 
 t2k 
ES em

 
 ..  .
 
 . 
 
tmk
h

Placing these column vectors side by side we obtain the rectangular array
at

 
t t12 ··· t1n
 11 
 t21

t22 ··· t2n 
m(T ) = 

 .. .. .. .. 
.
M

 .
 . . . 
tm1 tm2 · · · tmn

This matrix is called the matrix of T .

6 Composition of Linear Functions

Let
T : Rn → Rm

17
 and
S : Rm → Rp

be linear functions. Then we define the composition

(S ◦ T )(x) = S[T (x)]

s s
for all x ∈ Rn . The composition S ◦ T is also linear and maps Rn into Rp .

te ysi
Matrices of S and T
Let

l
No a
u1 , . . . , un

e n
be unit coordinate vectors in Rn ,
ur l A e1 , . . . , em

in Rm , and
w1 , . . . , w p

in Rp . Suppose S and T have matrices (sik ) and (tik ) respectively. Then

ct ica
p
S(ek ) =
X
sik wi
i=1
Le at

for k = 1, 2, . . . , m and
m
T (uj ) =
X
tkj ek
ES em

k=1

for j = 1, 2, . . . , n.
h

Matrix of S ◦ T
Then
at

(S ◦ T )(uj ) = S[T (uj )]

m
!
M

=S
X
tkj ek
k=1
m
= tkj S(ek )
X

k=1
m p !
=
X X
tkj sik wi
k=1 i=1
p m
!
=
X X
sik tkj wi .
i=1 k=1

18
 Thus "m #
m(S ◦ T ) =
X
sik tkj .
k=1

In other words,
m(S ◦ T ) = m(S)m(T ).

s s
7 The Jacobian Matrix

te ysi
Next we show how matrices arise in connection with total derivatives. Let

f : Rn → Rm

l
No a
be a function which is differentiable at c. Let

e n
ur l A T = f ′ (c)

be the total derivative of f at c. To find the matrix of T , consider its action on the unit
coordinate vectors u1 , . . . , un .
ct ica
T (uk ) = f ′ (c)(uk ).

But
f ′ (c)(uk ) = Dk f (c).
Le at

Components
ES em

Since
f = (f1 , . . . , fm ),

we have
h

D f (c)
 
 k 1 
 Dk f2 (c) 
at

 
Dk f (c) = 

.. .
.

 
 
Dk fm (c)
M

Therefore the matrix of T is

D f (c) D2 f1 (c) · · · Dn f1 (c)

 
 1 1 
 D1 f2 (c) D2 f2 (c) · · · Dn f2 (c) 

Df (c) = 

.. .. .. .. .
. . . .
 
 
 
D1 fm (c) D2 fm (c) · · · Dn fm (c)

This matrix is called the Jacobian matrix of f at c.

19
 Special Case
If f : Rn → R then the Jacobian matrix reduces to
h i
Df (c) = D1 f (c) D2 f (c) · · · Dn f (c) .

This vector is called the gradient of f ,

s s
te ysi
∇f (c).

Note:Let

l
f = (f1 , f2 , . . . , fm ) : Rn → Rm

No a
be differentiable at c. Then the derivative of f at c applied to a vector v is given by

e n
ur l A f ′ (c)(v) = (∇f1 (c) · v, ∇f2 (c) · v, . . . , ∇fm (c) · v).

Let {e1 , e2 , . . . , em } denote the standard basis of Rm . Then we may write

ct ica
m
f ′ (c)(v) = (∇fk (c) · v)ek .
X

k=1

Taking norms on both sides, we obtain

Le at

m
∥f ′ (c)(v)∥ = (∇fk (c) · v)ek .
X
ES em

k=1

Using the triangle inequality,

h

m
∥f ′ (c)(v)∥ ≤ |∇fk (c) · v|.
X
at

k=1

By the Cauchy–Schwarz inequality,

M

|∇fk (c) · v| ≤ ∥∇fk (c)∥∥v∥.

Therefore,

m m
∥f ′ (c)(v)∥ ≤ ∥∇fk (c)∥∥v∥ = ∥v∥ ∥∇fk (c)∥.
X X

k=1 k=1

Let

20
 m
M= ∥∇fk (c)∥.
X

k=1

Hence we obtain the estimate

∥f ′ (c)(v)∥ ≤ M ∥v∥.

s s
te ysi
This inequality shows that

f ′ (c)(v) → 0 as v → 0.

l
No a
This estimate will be used in the proof of the chain rule.

e n
ur l A
The Chain Rule
Let f and g be functions such that the composition
ct ica
h=f ◦g

is defined in a neighborhood of a point a.

Le at

The chain rule tells us how to compute the derivative of h in terms of the derivatives of
f and g.
ES em

Theorem 7.1. Assume that g is differentiable at a, with total derivative g ′ (a). Let b =
g(a) and assume that f is differentiable at b, with total derivative f ′ (b).
h

Then the composite function h = f ◦ g is differentiable at a and

at

h′ (a) = f ′ (b) ◦ g ′ (a).

M

Proof
Consider

h(a + y) − h(a)

for small ∥y∥.

21
 Since h(x) = f (g(x)),

h(a + y) − h(a) = f [g(a + y)] − f [g(a)].

Let

s s
b = g(a), v = g(a + y) − b.

te ysi
Then

l
No a
h(a + y) − h(a) = f (b + v) − f (b).

e n
Taylor formula for g
ur l A
Since g is differentiable at a,

v = g ′ (a)(y) + ∥y∥Ea (y),

ct ica
where
Le at

Ea (y) → 0 as y → 0.
ES em

Taylor formula for f

Since f is differentiable at b,
h

f (b + v) − f (b) = f ′ (b)(v) + ∥v∥Eb (v),

at

where
M

Eb (v) → 0 as v → 0.

Substituting v
Substituting v = g ′ (a)(y) + ∥y∥Ea (y),

f (b + v) − f (b) = f ′ (b)[g ′ (a)(y)] + f ′ (b)[∥y∥Ea (y)] + ∥v∥Eb (v).

22
 Define

∥v∥
E(y) = f ′ (b)[Ea (y)] + Eb (v). (1)
∥y∥

Then

s s
te ysi
h(a + y) − h(a) = f ′ (b)[g ′ (a)(y)] + ∥y∥E(y).

Bound for ∥v∥

l
From

No a
e n
v = g ′ (a)(y) + ∥y∥Ea (y),
ur l A
we obtain

∥v∥ ≤ ∥g ′ (a)(y)∥ + ∥y∥∥Ea (y)∥.

ct ica
Using the previous estimate,
Le at

∥g ′ (a)(y)∥ ≤ M ∥y∥.
ES em

Hence

∥v∥
≤ M + ∥Ea (y)∥.
∥y∥
h

Since Ea (y) → 0, ||Ea (y)|| ≤ K for some K > 0 in a neighbourhood of 0 so that

at

∥v∥
≤M +K
∥y∥
M

in a neighbourhood of 0. Thus ∥v∥/∥y∥ remains bounded as y → 0. Further, since

Ea (y) → 0 as y → 0 and since f ′ (a) is linear continuous, we have that

f ′ (b)[Ea (y)] → f ′ (b)[0] = 0

Therefore

23
 lim E(y) = 0
y→0

Hence

es sis
h(a + y) − h(a) = f ′ (b)[g ′ (a)(y)] + ∥y∥E(y).

Thus h is differentiable at a and

ot ly
h′ (a) = f ′ (b) ◦ g ′ (a).

N a
Theorem 7.2 (Matrix Form of the Chain Rule). Let g : Rp → Rn and f : Rn → Rm be

e An
differentiable functions. Let h = f ◦ g and let a ∈ Rp . If b = g(a), then

Dh(a) = Df (b) Dg(a).

Suppose
ur l
ct ca
g = (g1 , . . . , gn ), f = (f1 , . . . , fm ), h = (h1 , . . . , hm ).

Then the above matrix equation is equivalent to the scalar equations

Le ati

n
Dj hi (a) = Dk fi (b) Dj gk (a),
X

k=1
ES em

for i = 1, . . . , m and j = 1, . . . , p.

Equivalently, writing y = f (x) and x = g(t), we obtain

n
h

∂yi ∂yi ∂xk

=
X
.
∂tj k=1 ∂xk ∂tj
at

Special case: If m = 1, then

M

n
h′ (a) = Dk f (b) Dj gk (a) = ∇f (b) · Dg(a).
X

k=1

24
 8 Leibnitz Rule for Differentiation under the Inte-
gral sign
∂f
Theorem 8.1. Let f (x, y) and (x, y) be continuous on the rectangle [a, b] × [c, d]. Let
∂y
p(y), q(y) ∈ [a, b] be differentiable functions.

s s
te ysi
Define
Z q(y)
F (y) = f (x, y) dx, y ∈ [c, d].
p(y)

Then F is differentiable and

l
No a
Z q(y)
∂f
F ′ (y) = (x, y) dx + f (q(y), y) q ′ (y) − f (p(y), y) p′ (y).

e n
p(y) ∂y
ur l A
Proof. Let h → 0. Consider
Z q(y+h) Z q(y)
F (y + h) − F (y) = f (x, y + h) dx − f (x, y) dx.
p(y+h) p(y)
ct ica
Z q(y)
Add and subtract f (x, y + h) dx:
p(y)
Le at

Z q(y) h i
F (y + h) − F (y) = f (x, y + h) − f (x, y) dx
p(y)
Z q(y+h) Z q(y) !
+ f (x, y + h) dx − f (x, y + h) dx .
ES em

p(y+h) p(y)

Step 1: First term

h

1 Z q(y) h i Z q(y)
f (x, y + h) − f (x, y)
f (x, y + h) − f (x, y) dx = dx.
at

h p(y) p(y) h

Taking limit as h → 0 and using continuity,

M

Z q(y)
∂f
lim = (x, y) dx.
h→0 p(y) ∂y

Step 2: Second term

25
 Rewrite:
Z q(y+h) Z q(y)
f (x, y + h) dx − f (x, y + h) dx
p(y+h) p(y)
Z q(y+h) Z p(y+h)
= f (x, y + h) dx − f (x, y + h) dx.
q(y) p(y)

s s
By the Mean Value Theorem for integrals, there exist points ξh and ηh such that

te ysi
Z q(y+h)
f (x, y + h) dx = f (ξh , y + h) [q(y + h) − q(y)],
q(y)
Z p(y+h)
f (x, y + h) dx = f (ηh , y + h) [p(y + h) − p(y)].

l
p(y)

No a
e n
Thus,
ur l A
1
h
(· · · ) = f (ξh , y + h)
q(y + h) − q(y)
h
− f (ηh , y + h)
p(y + h) − p(y)
h
.

Taking limit as h → 0,
ct ica
= f (q(y), y) q ′ (y) − f (p(y), y) p′ (y).
Le at

Step 3: Combine

Z q(y)
∂f
F (y) =
′
(x, y) dx + f (q(y), y) q ′ (y) − f (p(y), y) p′ (y).
ES em

p(y) ∂y
h

9 Mean Value Theorem for Differentiable functions

at

Theorem 9.1. Let S be an open subset of Rn and let f : S → Rm be differentiable at

each point of S.
M

Let x, y ∈ S such that the line segment

L(x, y) = {tx + (1 − t)y : 0 ≤ t ≤ 1} ⊂ S.

Then for every vector a ∈ Rm , there exists a point z ∈ L(x, y) such that
 
a · (f (y) − f (x)) = a · Df (z)(y − x) .

26
 Proof. Let
u = y − x.

Define a real-valued function F : [0, 1] → R by

F (t) = a · f (x + tu).

s s
te ysi
Since f is differentiable, F is differentiable on (0, 1) and continuous on [0, 1].

By the chain rule,

 
F ′ (t) = a · Df (x + tu)(u) .

l
No a
By the Mean Value Theorem for real-valued functions, there exists t0 ∈ (0, 1) such that

e n
F (1) − F (0) = F ′ (t0 ).
ur l A F (1) = a · f (x + u) = a · f (y),

F (0) = a · f (x).
ct ica
Thus,
F (1) − F (0) = a · (f (y) − f (x)).
Le at

Also,
 
F ′ (t0 ) = a · Df (x + t0 u)(u) .
ES em

Let
z = x + t0 u.
h

Then z ∈ L(x, y).

at

 
a · (f (y) − f (x)) = a · Df (z)(y − x) .
M

Result from Mean Value Theorem

By the Mean Value Theorem for vector-valued functions, we have

f (y) − f (x) = Df (z)(y − x),

27
 for some z = x + θ(y − x), where 0 < θ < 1.

Taking norms,
∥f (y) − f (x)∥ ≤ ∥Df (z)∥ · ∥y − x∥.

Let m
M= |∇fk (z)|.
X

s s
k=1

te ysi
Then,
∥f (y) − f (x)∥ ≤ M ∥y − x∥.

l
No a
Special Cases

e n
1. Real-valued case (m = 1)
ur l A
If f : Rn → R, then
f (y) − f (x) = ∇f (z) · (y − x).
ct ica
2. Using Schwarz inequality
If a is a unit vector, then
Le at

|f (y) − f (x)| ≤ ∥∇f (z)∥ · ∥y − x∥.

ES em

3. Lipschitz condition
If all partial derivatives Dj f are bounded on S, then there exists a constant A > 0 such
that
h

∥f (y) − f (x)∥ ≤ A∥y − x∥.

at

Thus f satisfies a Lipschitz condition on S.

M

Theorem 9.2. Let S be an open connected subset of Rn , and let f : S → Rm be differ-

entiable at each point of S.

If
f ′ (c) = 0 for all c ∈ S,

then f is constant on S.

Proof. Since S is open and connected, it is polygonally connected.

28
 Let x, y ∈ S. Then there exist points

p0 = x, p1 , p2 , . . . , pr = y

such that each line segment L(pi−1 , pi ) ⊂ S.

By the Mean Value Theorem,

s s
te ysi
f (pi ) − f (pi−1 ) = Df (ci )(pi − pi−1 ),

for some ci ∈ L(pi−1 , pi ).

l
Since Df (ci ) = 0, we get

No a
f (pi ) − f (pi−1 ) = 0.

e n
Hence, ur l A f (pi ) = f (pi−1 ) for all i.

Adding these equations for i = 1, 2, . . . , r, we get

ct ica
f (y) − f (x) = 0.

Thus,
Le at

f (y) = f (x).

Since x, y are arbitrary, f is constant on S.

ES em

10 Sufficient Condition for Differentiability

h

Theorem 10.1. Assume that one of the partial derivatives D1 f, . . . , Dn f exists at c and
at

that the remaining n − 1 partial derivatives exist in some ball B(c; δ) and are continuous
at c. Then f is differentiable at c.
M

Proof. First, it is enough to prove the result for real-valued functions. So assume f :
Rn → R.

The candidate for the derivative at c is the gradient:

∇f (c) = (D1 f (c), . . . , Dn f (c)) .

29
 We aim to prove:

f (c + v) − f (c) = ∇f (c) · v + o(∥v∥) as v → 0.

Let

s s
v = λy, where ∥y∥ = 1 and λ = ∥v∥.

te ysi
Write
y = y1 u1 + y2 u2 + · · · + yn un ,

l
where uk are standard basis vectors .

No a
Define

e n
v0 = 0, v1 = y1 u1 , v2 = y1 u1 + y2 u2 , ..., vn = y.

Then
ur l A n
f (c + v) − f (c) = f (c + λy) − f (c) = [f (c + λvk ) − f (c + λvk−1 )] .
X

k=1
ct ica
For k = 1, we vary only the first coordinate. So,
Le at

f (c + λy1 u1 ) − f (c) = λy1 D1 f (c) + λy1 E1 (λ),

ES em

where E1 (λ) → 0 as λ → 0.

For k ≥ 2, define
bk = c + λvk−1 .
h

Then
at

f (c + λvk ) − f (c + λvk−1 ) = f (bk + λyk uk ) − f (bk ).

M

By the one-dimensional Mean Value Theorem,

= λyk Dk f (bk ).

Since Dk f is continuous at c,

Dk f (bk ) = Dk f (c) + Ek (λ),

where Ek (λ) → 0 as λ → 0.

30
 Thus,
f (c + λvk ) − f (c + λvk−1 ) = λyk Dk f (c) + λyk Ek (λ).

Summing over k = 1 to n, we get:

n n
f (c + v) − f (c) = λ yk Dk f (c) + λ yk Ek (λ).
X X

s s
k=1 k=1

te ysi
Since
v = λy,

l
we have

No a
n
yk Dk f (c) = ∇f (c) · y.
X

e n
k=1

Thus, ur l A f (c + v) − f (c) = ∇f (c) · v + ∥v∥E(λ),

where n
E(λ) = yk Ek (λ).
X
ct ica
k=1

Since each Ek (λ) → 0, we get

Le at

E(λ) → 0 as λ → 0.
ES em

f (c + v) − f (c) = ∇f (c) · v + o(∥v∥).

h

Hence, f is differentiable at c.
at
M

Example( Mixed Partial Derivatives Need Not Be Equal)

Define
2
− y2)

 xy(x

, (x, y) ̸= (0, 0),
x2 + y 2

f (x, y) =
0, (x, y) = (0, 0).



For (x, y) ̸= (0, 0):

y(x4 + 4x2 y 2 − y 4 )
D1 f (x, y) = ,
(x2 + y 2 )2

31
 x(x4 − 4x2 y 2 − y 4 )
D2 f (x, y) = .
(x2 + y 2 )2

Also,
D1 f (0, 0) = 0, D2 f (0, 0) = 0.

s s
Compute D1,2 f (0, 0):

te ysi
D1 f (0, y) − D1 f (0, 0)
D1,2 f (0, 0) = lim
y→0 y

l
Since D1 f (0, y) = −y, we get:

No a
−y

e n
D1,2 f (0, 0) = lim = −1.
y→0 y
ur l A
Compute D2,1 f (0, 0):

D2 f (x, 0) − D2 f (0, 0)
ct ica
D2,1 f (0, 0) = lim
x→0 x

Since D2 f (x, 0) = x, we get:

Le at

x
D2,1 f (0, 0) = lim = 1.
x→0 x
ES em

Conclusion:
h

D1,2 f (0, 0) ̸= D2,1 f (0, 0).

at

Hence, mixed partial derivatives need not be equal if continuity fails.

M

Theorem 10.2. If both partial derivatives Dj f and Dk f exist in a neighborhood B(c; δ)

and both are differentiable at c, then

Dj,k f (c) = Dk,j f (c).

Proof. It is enough to consider the case n = 2 and c = (0, 0).

Step 1: Define increment

32
 Let h ̸= 0 and define

∆(h) = f (h, h) − f (h, 0) − f (0, h) + f (0, 0).

Step 2: First approach

s s
Let

te ysi
G(x) = f (x, h) − f (x, 0).

Then

l
∆(h) = G(h) − G(0).

No a
By Mean Value Theorem,

e n
∆(h) = hG′ (x1 )
ur l A
for some x1 ∈ (0, h).

Now,
G′ (x1 ) = D1 f (x1 , h) − D1 f (x1 , 0).
ct ica
Since D1 f is differentiable at (0, 0), we use Taylor expansion:
Le at

D1 f (x1 , h) = D1 f (0, 0) + D1,1 f (0, 0)x1 + D2,1 f (0, 0)h + ε1 ,

ES em

D1 f (x1 , 0) = D1 f (0, 0) + D1,1 f (0, 0)x1 + ε2 ,

where ε1 , ε2 → 0 as h → 0.
h
at

Subtracting,
D1 f (x1 , h) − D1 f (x1 , 0) = D2,1 f (0, 0)h + small error.
M

Thus,
∆(h) = h2 D2,1 f (0, 0) + h2 E(h),

where E(h) → 0.

Step 3: Second approach

Similarly, define
H(y) = f (h, y) − f (0, y).

33
 Then
∆(h) = H(h) − H(0).

Repeating the same argument,

∆(h) = h2 D1,2 f (0, 0) + h2 Ẽ(h),

s s
where Ẽ(h) → 0.

te ysi
Step 4: Compare

l
Divide by h2 :

No a
∆(h) ∆(h)
→ D2,1 f (0, 0) and → D1,2 f (0, 0).
h2 h2

e n
Hence, ur l A D1,2 f (0, 0) = D2,1 f (0, 0).
ct ica
Le at
ES emh
at
M

34