Calculus on Rn

Seminar 6 2020

Differential calculus for vector functions

1 Vector functions of real variable

We start by considering derivatives of vector functions, of real variable.

The general context is the following:

A⊆R
a ∈ A ∩ A0
f : A → Rm


If
f (x) − f (a)
∃ lim ∈ Rm ,
x→a x−a
then this limit is called the derivative of f at a and is usually denoted
by:
df
f 0 (a) = (a).
dx

Since f is a vector function, of real variable, it actually means that it

can be represented through m real functions of real variable

∀i ∈ {1, ..., m}, ∃fi : A → R, and f = (f1 , ..., fm ).

The function f has a derivative at a if and only if, for all i ∈ {1, ..., m},
each function fi has a derivative at a. Moreover, the following equality
holds:
f 0 (a) = (f10 (a), ..., fm
0
(a)).

1
2 Vector Functions of Vector Variable

The general context is the following:

n

A⊆R
a ∈ intA
f : A → Rm


If there exists a linear function ϕ : Rn → Rm such that

 
1
lim · f (x) − f (a) − ϕ(x − a) = 0m , (1)
x→a kx − ak

then the function f is said to be differentiable at a, and the linear

function ϕ is said to be the differential of f at a, and is denoted by

df (a).

Notice that the differential is a function, which has the following pro-
perties:
• : Rn → R m ;
• is linear;
• it satisfeis (1)
• the differential is a notion attached to a function at a single
point. So if we change the point, we may expect, if f is
still differentiable at the new point, to even get a different
differential function.
Theorem: If the function f is differentiable at a, then f is
continuous at a.
In practice, when computing effectively the differential of a function f
at a point a, instead of computing a limit such as (1) we may study instead
 
1
lim · f (a + h) − f (a) − ϕ(h) = 0m , (2)
h→0n khk

this is simply done because we replace x − a by h.

2
 Remark: The reason behind searching for something different from
the partial derivatives lies in the fact that a function may have partial
derivatives with respect to all variables at a point, and in the same time,
it may not be continuous at that point. We had at least two such examples
in Seminar 5. In contrast, if a functionis differentiable at a point, then it is
continuous at that point which is a property inherited from the derivatives
of real functions of real variable.

3 The Determination of the Differential Function -

mostly theoretic
3.1 Once Again The Case of Vector Functions of real variable
To begin with we go back to the case when n = 1, thus
A ⊆ R.
Then
f has a derivative at a ⇐⇒ f is differentiable at a
and the following equality holds
df (a)(x) = xf 0 (a), ∀x ∈ R .

3.2 Once Again The Case of Vector Functions of Vector Vari-

able
Let now n > 1. Because the codomain of the function is Rm , f is repre-
sented as
f = (f1 , ...fm ), with fi : A → R ∀i ∈ {1, ..., m}.
Since f is a vector function, of real variable, it actually means that
∀i ∈ {1, ..., m}, ∃fi : A → R, and f = (f1 , ..., fm ).
Then
f is differentiable a ⇐⇒ ∀i ∈ {1, ..., m}fi is differentiable at a
and the following equality holds
df (a)(x) = (df1 (a), ..., dfm (a)) . ∀x ∈ Rn .

3
4 Actual Algorithm for Determining the Differential
4.1 Real Functions of Vector Variables
n

A⊆R
a ∈ intA
f :A→R


Step 1: Study if f has partial derivatives at a with respect to all variables.


Y ES → go to Step 2
N O → STOP the function does not have a differential at a
In fact, we have to determine
∂f
(a) ∀j ∈ {1, ..., n}.
∂xj

Step 2: Study the limit

 
1
l = lim f (a + h) − f (a) − hh, ∇f (a)i . (3)
h→0n khk

Recall that a = (a1 , ..., an ) and h = (h1 , ..., hn ) ∈ Rn

q
a + h = (a1 + h1 , ..., an + hn ) and khk = h21 + ... + h2n .
Moreover, the gradient of a function at a point is the vector composed of
all of its partial derivatives at that point. Thus
n
∂f ∂f X ∂f
hh, ∇f (a)i = h1 · (a) + ... + hn · (a) = hj · (a).
∂x1 ∂xn j=1
∂x j

This means that the limit (3) is actually

Pn ∂f
f (a1 + h1 , ..., a2 + h2 ) − f (a1 , ..., an ) − j=1 hj · ∂xj (a)
l = lim p . (4)
h→0n h21 + ... + h2n
We encounter the following cases

l=0 → go to Step 3
otherwise → STOP the function does not have a differential at a

4
Step 3: We are in the case when l = 0. This means that the function f is
differentiable at the point a. Its differential is the linear function

df (a) : Rn → R

defined by
n
X ∂f
df (a)(h) = hh, ∇f (a)i = hj · (a), ∀h ∈ Rn .
j=1
∂xj

4.2 Vector Function of Vector Variable

n

A⊆R
a ∈ intA
f : A → Rm


Recall that f is represented as

f = (f1 , ...fm ), with fi : A → R ∀i ∈ {1, ..., m}.

For each i ∈ {1, ..., m}, the function fi must be analysed as described in
section 4.1, since it is a real function of vector variable. At the lecture,
there is a theorem that prove that
f is differentiable at a ⇐⇒ ∀i ∈ {1, ..., m}, fi is differentiable
at a. Moreover df (a) : Rn → Rm is equal to

df (a) = (df1 (a), df2 (a), ..., dfm (a)).

5 The Jacobi Matrix

If the function f : A → Rm , with A ⊆ Rn is partially derivable at a, then
one can construct the so-called Jacobi matrix of the function f at a
as  
∂f1 ∂f1 ∂f1
(a) ∂x2 (a) ... ∂xn (a)
 ∂x1
 ... ... ... ... 

J(f )(a) = 
 ... ... ... ... 

∂fm ∂fm ∂fm
∂x1 (a) ∂x2 (a) ... ∂xn (a)

5
 This matrix is used to simplify the writing of the differential function
of a vector function of a vector variable.
So, for a differentiable function, with A ⊆ Rn , with f : A → Rm , the
differential function at the point a ∈ intA is df (a) : Rn → Rm , such that

∀h = (h1 , ..., hn ) ∈ Rn , df (a)(h) = J(f )(a) · h.

This means that for a random h = (h1 , ..., hn ) ∈ Rn , df (a)(h) is a vector

in Rm , and, as we know from algebra, a vector may be represented as a
column matrix. Therefore,
  
∂f1 ∂f1 ∂f1 
(a) (a) ... (a) h 1
 ∂x1 ∂x2 ∂xn
 ... ... ... ...   ... 

df (a)(h) =  ·  .
 ... ... ... ...  ... 
 
∂fm ∂fm ∂fm
∂x1 (a) ∂x2 (a) ... ∂xn (a)
hn

Once me compute the matrix multiplication we reach the result

 Pn 
∂f1
i=1 ∂xi (a) · hi
...
 
df (a)(h) = 
 
...

P 
n ∂f1
i=1 ∂xi (a) · hi

which is the matrix form of the vector df (a)(h).

For the particular case m = 1, the Jacobi matrix is the Gradient

6
6 Exercise/applications
In practice, we often use the following theorem:
Let A ⊆ Rn is an open set, and f : A → Rm be a function which
has all partial derivatives. If all partial derivatives are continuous
on A, then f is differentiable on A.
This means that is all the partial derivatives of a function are conti-
nuous, we may skip Step 2 in the Algorithm from Section 4, and jump
directly to Step 3, which is actually the construction of the differential
function, by means of :
• the Gradient of f , if f is a real function of vector variable
• the Jacobi matrix of f , if f is a vector function of vector variable.

Recall Exercise 4 from Seminar 5:

Determine all first-order and all second-order partial derivati-

ves of the function
f : R3 → R, f (x, y, z) = z 2 cos(x − y).

Determine now its differential at the point (1, 1, 1).

Solution: Let (x, y, z) ∈ R3 be randomly chosen. Then

∂f ∂f
(x, y, z) = −z 2 sin(x − y), (x, y, z) = −z 2 sin(x − y)(x − y)0y = z 2 sin(x − y) and
∂x ∂y

∂f
(x, y, z) = 2z cos(x − y).
∂z
Hence, the Gradient of the function f at (x, y, z) is
 
∂f ∂f ∂f
∇f (x, y, z) = (x, y, z), (x, y, z), (x, y, z) =
∂x ∂y ∂z

= −z 2 sin(x − y), z 2 sin(x − y), 2z cos(x − y) .

Since all of these partial derivatives are continuous functions on R3 , we conclude that
the function f is differentiable at each a ∈ R3 .
We will compute this differential for two different points.
Case 1: Consider now a = (1, 1, 1) ∈ R3 . The differential function associated to f at
the point (1, 1, 1) is
df (1, 1, 1) : R3 → R

7
defined for all h = (h1 , h2 , h3 ) ∈ Rn , by

df (a)(h) = df (1, 1, 1)(h1 , h2 , h3 ) = hh, ∇f (a)i =

∂f ∂f ∂f
= h1 (1, 1, 1) + h2 (1, 1, 1) + h1 (1, 1, 1) =
∂x ∂y ∂z
= h1 · (−1 sin 0) + h2 · (1 sin 0) + h3 · (2 · 1 cos 0) = h1 · 0 + h2 · 0 + h3 · 2 =
= 2h3 .
now a = π2 , 0, 1 ∈ R3 . The differential function associated to f at


Case 2: Consider
the point π2 , 0, 1 is


π 
df , 0, 1 : R3 → R
2
n
defined for all h = (h1 , h2 , h3 ) ∈ R , by
π 
df (a)(h) = df , 0, 1 (h1 , h2 , h3 ) = hh, ∇f (a)i =
2
∂f  π  ∂f  π  ∂f  π 
= h1 , 0, 1 + h2 , 0, 1 + h1 , 0, 1 =
∂x 2 ∂y 2 ∂z 2
π π π
= h1 · (−1 sin ) + h2 · (1 sin ) + h3 · (2 · 1 cos ) = h1 · 1 + h2 · 1 + h3 · 2 · 0 =
2 2 2
= −h1 + h2 .
Conlusion: for the two distinct particular cases considered, we have determined two
different differential functions, namely

df (1, 1, 1) : R→ R df (1, 1, 1)(h1 , h2 , h3 ) = 2h3 , ∀(h1 , h2 , h3 ) ∈ R3 .

and π  π 
df , 0, 1 : R3 → R, df , 0, 1 = −h1 + h2 , ∀(h1 , h2 , h3 ) ∈ R3 .
2 2

Recall Exercise 5 from Seminar 5:

Determine the gradient of the function f at the point a in the
following cases:
a) f : R2 → R, f (x, y) = e−x sin(x + 2y) and a = 0, π2 ;

b) f : R3 → R, f (x, y, z) = (x − y) cos(πz) and a = 1, 0, 12 .

Determine now the differential function of f at the given a.

Solution:
∂f
a) Let (x, y) ∈ R2 be randomly chosen. Then ∂x
(x, y) = −e−x sin(x+2y)+e−x cos(x+
2y) and ∂f
∂y
(x, y) = 2e−x cos(x + 2y).

8
 We notice that the partial derivatives with respect to both variables are continuous
functions. This means that
f has a differential function at all the points in R2 .
For the given a = 0, π4 , the gradient is
 π   ∂f  π  ∂f  π 
∇f 0, = 0, , 0,
4 ∂x 4 ∂y 4
  π   π  π 
= −e−0 sin 0 + + e−0 cos 0 + , 2e−0 cos 0 + = (−1, 0).
2 2 2
The differential is the function df 0, π4 : R2 → R, defined by

 π  π
df 0, (h1 , h2 ) = hh, ∇f 0, i = h1 · (−1) + h2 · 0 = −h1 , ∀h = (h1 , h2 ) ∈ R2 .
4 4

b) Let (x, y, z) ∈ R3 be randomly chosen. Then ∂f ∂x

(x, y, z) = cos πz, ∂f∂x
(x, y, z) =
∂f
− cos πz and ∂x (x, y, z) = −(x − y)π sin πz.
We notice that the partial derivatives with respect to both variables are continuous
3
functions. This means that
f has a differential function at all the points in R .
1
For the given 1, 0, 2 , the Gradient of f is
  
1 π π π
∇f 1, 0, = cos , − cos , −(1 − 0)π sin = (0, 0, −π).
2 2 2 2

The differential function associated to f at the point 1, 0, 21 is df 1, 0, 12 : R3 → R,

given by
   
1 1
df 1, 0, (h1 , h2 , h3 ) = hh, ∇f 1, 0, i = h1 · 0 + h2 cot 0 + h3 · (−π).
2 2
Thus  
1
df 1, 0, (h1 , h2 , h3 ) = −π · h3 ∀h = (h1 , h2 , h3 ) ∈ R3 .
2

Exercise: Determine the differential of the following function,

f : R2 → R3 , defined by

f (x, y) = (cos x + sin y, sin x + cos y, ex−y ), ∀(x, y) ∈ R2 ,

at a random point (x, y) ∈ R2 .
Solution:

Let (x, y) ∈ R2 be randomly chosen. To begin with, we notice that our function is a
vector one, of vector variables, so it takes vectors of dimension two and maps them into
vectors of dimension three. In order to determine the differential function, we need (if
they exist and are continuous functions) all the partial derivatives. We will consider

f = (f1 , f2 , f3 ), with f1 , f2 , f3 : R2 → R

9
 f1 (x, y) = cos x + sin y, f2 (x, y) = sin x + cos y, f3 (x, y) = ex−y , ∀(x, y) ∈ R2 .
We start computing the partial derivatives of f1 .
∂f1
(x, y) = (cos x + sin y)0x = − sin x + 0 = − sin x,
∂x
and
∂f1
(x, y) = (cos x + sin y)0y = 0 + cos y = cos y.
∂y
We continue with
∂f2
(x, y) = (sin x + cos y)0x = cos x + 0 = cos x,
∂x
and
∂f2
(x, y) = (sin x + cos y)0y = 0 − sin y = − sin y.
∂y
And conclude with
∂f3
(x, y) = (ex−y )0x = ex−y · (x − y)0x = ex−y · (1 − 0) = ex−y
∂x
and
∂f3
(x, y) = (ex−y )0y = ex−y · (x − y)0y = ex−y · (0 − 1) = −ex−y .
∂y
We construct the Jacobi matrix associated to the function f at the point (x, y),
∂f1 ∂f1
 
∂x (x, y) ∂y (x, y)

− sin x

cos y
   
  
∂f2 ∂f2

J(f )(x, y) =  ∂x (x, y) ∂y (x, y)  = cos x − sin y 
  
   
   
∂f3 ∂f3
∂x (x, y) ∂y (x, y)
ex−y −ex−y

We notice that all the partial derivatives are continuous functions, thus f is differen-
tiable at each random point (x, y) ∈ R2 . The differentiable function is df (x, y) : R2 → R3
given by  
h1
df (x, y)(h1 , h2 ) = J(f )(x, y) · =
h2

− sin x

cos y  
    − sin x · h1 + cos y · h2
  h1
=
 cos x − sin y 
 · h2 =
 cos x · h1 − sin y · h2 
  ex−y · h1 − ex−y · h2
ex−y −ex−y

10
This is the matrix form of the linear map df (x, y). It vector form is

ex−y · h1 − ex−y · h2 ,


d(f )(x, y)(h1 , h2 ) = − sin x · h1 + cos y · h2 , cos x · h1 − sin y · h2 ,

for all (h1 , h2 ) ∈ R2 .

π π


For the particular case when (x, y) = ,
2 2
, the Jacobi matrix is
 
−1 0
π π   


J(f ) , = 0 −1 
2 2 



1 −1

And the differential at this point is df π2 , π2 : R2 → R3 , defined by

π π 
df , (h1 , h2 ) = (−h1 , −h2 , h1 − h2 ) , ∀(h1 , h2 ) ∈ R2 .
2 2

Let us recall the fact that a vector in Rn may be written in its matrix form, as a
matrix with n rows and 1 column, thus
 
x1
 x2 
x = (x1 , ..., xn ) =  
 ... 
x3

As Homework, please solve completely exercises 1, 2, 3 and 4 from Seminar5-ttrif

11