Module: Numerical Methods Mme: BENAMARA

Chapter: Solving Non linear Equations

Chapter: Solving Nonlinear Equations

1. Introduction :
Definition 1 :
We say that α ∈ 𝐷𝑓 is a root of the equation 𝑓 (𝑥 ) = 0, if 𝑓(𝛼 ) = 0. Solving the
equation 𝑓 (𝑥 ) = 0is to find all real numbers α such that 𝑓 (𝛼 ) = 0 is verified.
It is not always possible to solve this problem completely for all forms of functions f .
Definition 2:
We say that a root α of an equation f(x) = 0 is separable if we can find an interval [a, b] such
that α is the only root of this equation in [a, b]. The root α is then said to be separated (we also
say isolated root).

2. Root separation
There is no general method for separating the roots of an equation f(x) = 0.
In practice, apart from the direct theoretical study of f if f is given analytically, two types of
methods are used:
2.1 Graphical method
a) We draw the graph of the function f and we look for its intersection with the abscissa
axis (ox) :
b) We decompose f into two functions f1 and f2 which are easy to study, such that
𝑓 = 𝑓1 − 𝑓2 ; and we look for the points of intersection of the graphs of f1 and f2 , whose
abscissas are exactly the roots of the equation f(x) = 0
Note : We often choose f1 and f2 so that their curves are known curves.
Examples:
1
Let's graphically solve the equation:𝑓 (𝑥 ) = log(𝑥 ) − 𝑥 = 0

This equation is written in the form 𝑓1 (𝑥 ) − 𝑓2 (𝑥 ) = 0by posing:

1
𝑓1 (𝑥 ) = log(𝑥 )And 𝑓2 (𝑥 ) = 𝑥

1
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations

The equation 𝑓(𝑥) = 0 admits a root α

2.2 Scanning method:
Consists of dividing the domain of definition of the function f denoted D f at n
equidistant points ( i=0….n ) based on the intermediate value theorem, we seek [𝑥𝑖 , 𝑥𝑖+1 ]such
that𝑓 (𝑥𝑖 ). 𝑓(𝑥𝑖+1 ) < 0
Intermediate Value Theorem:

a- f x is continuous on an interval a , b 

b- f a ×fb 0

According to a and b, there exists at least one value a , b such that f 0

Moreover, if f x is monotone on the interval a , b), then the root is unique on a , b

3. Iterative methods
3.1 Dichotomy Method
We consider a real function f defined on an interval [ a, b ] , with a < b , and continuous
on this interval and we assume that f admits a unique root on I =] a, b [, that is to say that
there exists a unique α ∈ I such as 𝑓(𝛼 ) = 0.
We consider an interval [ a, b ] and a continuous function f from [ a, b ] to R. We assume that
𝑓 (𝑎). 𝑓(𝑏) < 0and that the equation 𝑓 (𝑥 ) = 0has a unique solution α on the interval [ a, b ] .
Dichotomy method consists of constructing a sequence ( x n ) which converges to α in the
following manner:
Initialization: we take x 0 as the middle of [ a, b ] . The root is then in one of the two intervals:
] a, x 0 [ or ] x 0 , b [ or it is equal to x 0 .
• if 𝑓(𝑎). 𝑓(𝑥0 ) < 0, then α ∈ ] a, x 0 [ . We set a 1 = a , b 1 = x 0 .
• if 𝑓(𝑎). 𝑓 (𝑥0 ) = 0, then α = x 0 .
• if 𝑓(𝑎). 𝑓 (𝑥0 ) > 0, then α ∈ ] x 0 , b [ . We set a 1 = x 0 , b 1 = b .

2
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations
We then take for x 1 the middle of [ a 1 , b 1 ] .
We thus construct a sequence:
𝑎+𝑏 𝑎1 +𝑏1 𝑎𝑛 +𝑏𝑛
𝑥0 = , 𝑥1 = , . . . ,𝑥𝑛 =
2 2 2

Proposal :

The sequences (𝑎𝑛 )𝑛∈𝑁 , (𝑏𝑛 )𝑛∈𝑁 and (𝑥𝑛 )𝑛∈𝑁 all converge to . Furthermore, we

have the following estimate:

|𝑏𝑛 −𝑎𝑛 | |𝑏𝑛−1 −𝑎𝑛−1 | |𝑏0 −𝑎0 |
|𝛼 − 𝑥𝑛 | ≤ = =⋯=
2 22 2𝑛+1
|𝑏−𝑎|
From where|𝛼 − 𝑥𝑛 | ≤ 𝑛+1 .
2
Given a precision ε , this method allows to approach α in a predictable number of iterations.
Accuracy of the approximation:
To approximate the root α with a precision lower than using the Dichotomy method, it
is necessary that:
|𝑏0 − 𝑎0 | |𝑏0 − 𝑎0 |
|𝛼 − 𝑥 𝑛 | ≤ < 𝜀 ⇒ < 2𝑛+1
2𝑛+1 𝜀
|𝒃𝟎 −𝒂𝟎 |
𝒍𝒏( )
𝜺 𝑎+𝑏
So 𝒏 + 𝟏 > if we start the iterations from 𝑥0 =
𝒍𝒏𝟐 2

3.2 Fixed point method:

Definition:
Let 𝒈 a function defined on an interval [a,b], the point 𝒙 ̅ which verifies 𝒙 ̅ = 𝒈 ( 𝒙 ̅) with
𝒙 ̅ ∈ [ 𝒂 , 𝒃 ] is said to be a fixed point of the function g .
This method is based on the principle of the fixed point of a function, we write the equation
f(x)=0 in the form x=g(x) , then we look for the fixed point 𝒙 ̅ of the function g.
For this we create the sequence 𝒙 𝒏 + 𝟏 = 𝒈 ( 𝒙 𝒏 ) ( n=0,1,2….) with 𝑥 0 given.
We start from 𝒙 𝟎 for n=0 , we calculate 𝒙 𝟏 = 𝒈 ( 𝒙 𝟎 ) then n=1 , we calculate 𝒙 𝟐 = 𝒈 ( 𝒙 𝟏 )
,..., 𝒙 𝒏 + 𝟏 = 𝒈 ( 𝒙𝒏 ) .
Under certain conditions, the sequence { 𝒙 𝒏 } 𝒏 = 𝟎 ,∞ converges to the solution 𝒙 ̅ fixed point
of 𝒈 and solution of the equation f(x)=0.

3
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations
Example :
Write the equation f(x)=0 in the form x = g(x) for 𝒇 ( 𝒙 ) = 𝒙 𝟐 + 𝟑𝒆 𝒙 − 𝟏𝟐 .
We can write
𝒙 = 𝒈 𝟏 ( 𝒙 ) = 𝒙 𝟐 + 𝟑𝒆 𝒙 − 𝟏𝟐 + 𝒙
𝒙 = 𝒈 𝟐 ( 𝒙 ) =√𝟏𝟐 − 𝟑𝒆𝒙
𝟏𝟐−𝒙𝟐
𝒙 = 𝒈 𝟑 ( 𝒙 ) =𝐥𝐧
𝟑

In order to be able to choose the appropriate form of g for the calculation, a

convergence criterion of this method must be verified.
Convergence and calculation stopping criterion for the fixed point method:
Let g be a differentiable function defined from [a, b] → [a, b] such that
(sufficient condition):
| 𝒈 ′( 𝒙 )| ≤ 𝒌 < 𝟏 ∀ 𝒙 ∈ [ 𝒂 , 𝒃 ]
Then the sequence { 𝒙 𝒏 } 𝒏 = 𝟎 ,∞ defined by 𝒙 𝒏 + 𝟏 =𝒈(𝒙 𝒏 ) ( n=0,1,2….)
converges independently of the value of 𝒙 𝟎 towards the unique fixed point 𝒙 ̅ of g .
If several forms of g satisfy this condition, we will have several values of k . We
choose the one with the minimum value of k .
In practice, we calculate 𝒌 = 𝒎𝒂𝒙 𝒙 ∈[ 𝒂 , 𝒃 ] | 𝒈 ′( 𝒙 )| which must be less than
unity for the method to converge.
We stop the calculations for this method when the absolute difference between
two successive iterations is less than a certain precision 𝜀 given.
| 𝒙 𝒏+𝟏− 𝒙 𝒏| < 𝜺
Local fixed point theorem :
𝒈 x be a real, defined, continuous and once differentiable function on the
interval a, b(we say that 𝒈x is of class C 1 on the interval a , b ). We assume
that 𝒈xadmits a fixed point a , b , such that 𝒈 ' 1 then:
𝒙𝟎 𝒅𝒐𝒏𝒏é
The recurrent sequence: { converges to the fixed point α on I.
𝒙𝒏+𝟏 = 𝒈(𝒙𝒏 ), 𝒏 ∈ 𝑵
Additionally, we have:
𝒌𝒏
|𝒙𝒏 − 𝜶| ≤ |𝒙𝟏 − 𝒙𝟎 | with 𝒌 = 𝒎𝒂𝒙 | 𝒈 ′( 𝒙 )|
𝟏−𝒌

4
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations
Error estimation:
The minimum number of iterations for the solution to be approximated with precision
𝜺 is | 𝒙 𝒏− α | < 𝜺

𝒌𝒏
Knowing that|𝒙𝒏 − 𝜶| ≤ |𝒙𝟏 − 𝒙𝟎 |
𝟏−𝒌

So
(𝟏−𝒌)𝜺
𝒍𝒏( )
|𝒙𝟏 −𝒙𝟎 |
𝒏> , 𝒌 = 𝒎𝒂𝒙 𝒙 ∈ [ 𝒂 , 𝒃 ] | 𝒈 ′( 𝒙 )|
𝐥𝐧(𝒌)

Example :
Find the first root of the equation 𝐥𝐧 ( 𝒙 ) − 𝒙 𝟐 + 𝟐 = 𝟎 which belongs to [0.1, 0.5]
with a precision ε=0.001.
We write this equation in the form x = g(x) and we check the conditions of
convergence.
We can write:
2 −2
𝑥 = 𝑒𝑥 = 𝑔1 (𝑥) And

𝑥 = √ln(𝑥) + 2 = 𝑔2 (𝑥)
Let us check the convergence condition for this method𝒌 = 𝒎𝒂𝒙𝒙∈[𝒂,𝒃] |𝒈′ 𝒙)|
𝟐 −𝟐
𝒌 𝟏 = 𝒎𝒂𝒙 𝒙 ∈ [ 𝟎 . 𝟏 , 𝟎 . 𝟓 ] | 𝒈 𝟏 ′( 𝒙 )| = 𝒎𝒂𝒙 𝒙 ∈ [ 𝟎 . 𝟏 , 𝟎 . 𝟓 ] | 𝟐𝒙𝒆𝒙 |.
we have 𝒈 𝟏 ′( 𝒙 ) strictly increasing therefore
𝟐 −𝟐
𝒌 𝟏 = 𝒎𝒂𝒙 𝒙 = 𝟎 . 𝟓 | 𝟐 ∗ 𝟎. 𝟓𝒆𝟎.𝟓 | = 𝟎 . 𝟏𝟕𝟒 < 𝟏 this form converges.
We then write:
𝟐
𝒙 𝒏 + 𝟏 = 𝒈 𝟏 ( 𝒙 𝒏 )=𝒆𝒙𝒏 −𝟐 (for n=0,1,2,…..)
Let's start x 0 = 0.3 (for example the middle of the given initial interval):
𝟐
𝒏 = 𝟎, 𝒙 𝟏 = 𝒈 𝟏 (𝒙 𝟎 ) = 𝒆𝒙𝟎−𝟐= 0.148
We calculate |𝑥 1 − 𝑥 0 | = 0.152 > 𝜀 ;
𝟐
𝒏 = 𝟏, 𝒙 𝟐 = 𝒈 𝟏 (𝒙 𝟏 ) = 𝒆𝒙𝟏−𝟐= 0.138 .
We calculate |𝑥 2 − 𝑥 1 | = 0.01 > 𝜀
𝟐
𝒏 = 𝟐, 𝒙 𝟑 = 𝒈 𝟏 (𝒙 𝟐 ) = 𝒆𝒙𝟐 −𝟐= 0.138 .
We calculate |𝑥 3 − 𝑥 2 | = 0.00 < 𝜀 , The solution is 𝒙 𝟐 = 𝟎. 𝟏𝟑𝟖

5
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations
3.3 Newton's method:
This is the most efficient and widely used method, it is based on the Taylor
development. If f(x) is continuous and continuously differentiable in the neighborhood
of 𝒙 ̅ solution of f(x)=0 , then the Taylor series development around an estimate 𝒙 𝒏

close to 𝒙 ̅ is written:
̌−𝒙𝒏 )
(𝒙 ̌−𝒙𝒏 )𝟐
(𝒙
𝒇 ( 𝒙 ) = 𝒇 ( 𝒙 𝒏) + 𝒏 𝒇' ( 𝒙 𝒏) + 𝒇'' ( 𝒙 𝒏 ) + ⋯
𝟏! 𝟐!

If 𝒙 𝒏 is an estimate close to 𝒙 ̅ , then the square of the error 𝜺 𝒏 = 𝒙 ̅ − 𝒙 𝒏 and the

terms of higher degrees are negligible. Knowing that 𝒇 ( 𝒙 ̅) = 𝟎 we obtain the
approximate relationship

𝒇 ( 𝒙 𝒏 )+ ( 𝒙 − 𝒙 𝒏 ) 𝒇 '( 𝒙 𝒏 ) ≈ 𝟎
𝒇(𝒙𝒏 )
̌ = 𝒙𝒏 −
SO 𝒙
𝒇′ (𝒙𝒏 )

We can write the ( 𝒏 + 𝟏 ) th iteration approximating 𝑥 ̅ is:

𝒇(𝒙𝒏 )
𝒙𝒏+𝟏 = 𝒙𝒏 − (n=0,1,2,…..)
𝒇′ (𝒙𝒏 )

This sequence, if it converges, must converge to the solution 𝒙 ̅ of f(x)=0 . We note

that f'(x) must be non-zero.
Convergence criterion of Newton's method:
f be a function defined on [a,b] such that:
i. 𝒇 ( 𝒂 ) 𝒇 ( 𝒃 ) < 𝟎
ii. 𝒇 ′( 𝒙 ) 𝒆𝒕 𝒇 ′′( 𝒙 ) are non-zero and keep a constant sign on the given
interval.
Stopping criterion for Newton's method:
If the convergence condition is verified, the iterative process must converge. This
means that each new iteration is better than the previous one, so we can say that if we
have a precision 𝜺 , we stop the calculation when the absolute difference between two
successive approximations is less than the given precision.
That is: | 𝒙 𝒏 + 𝟏 − 𝒙 𝒏 | ≤ 𝜺
If this condition is verified we take 𝒙 𝒏 + 𝟏 as a solution of f(x)=0 .

6
Module: Numerical Methods Mme: BENAMARA
Chapter: Solving Non linear Equations

Newton's theorem :
F ( x ) be a class C 2 function on the interval [ a , b ], such that:
1. 𝒇(𝒂). 𝒇(𝒃) < 0
2. 𝒇’(𝒙) ≠ 𝟎 ( f (monotonic on [ a , b ]) There exists α ]a,b[ such that f (α ) = 0
3. f '' ( x ) keeps a constant sign on the interval [ a , b ] ( f '' ( x ) < 0 where f '' ( x ) > 0 )
4. Starting from a point x 0 which satisfies the inequality 𝒇(𝒙𝟎 ). 𝒇′′(𝒙𝟎 ) > 0(verified by
the choice of x 0 )
If the conditions announced above are satisfied, then Newton's process:
𝒙𝟎 choisi
{ 𝒇(𝒙𝒏 )
𝒙𝒏+𝟏 = 𝒙𝒏 −
𝒇′ (𝒙𝒏 )
Converges for this choice of x 0 to the unique solution a of f ( x )
Additionally, we have the following estimate:
𝑴
| 𝒙𝒏 − 𝜶 | ≤ |𝒙 − 𝜶| , ∀ 𝒏 > 0
𝟐𝒎 𝒏−𝟏
With𝑀 = max |𝑓 ′′ (𝑥 )| , 𝑚 = min |𝑓 ′′ (𝑥 )|
𝑥∈[𝑎,𝑏] 𝑥∈[𝑎,𝑏]

This inequality can be written as a function of x 0

𝑴
| 𝒙𝒏 − 𝜶 | ≤ ( ) |𝒙𝟎 − 𝜶|𝟐𝒏 , ∀𝒏 ∈ 𝑵
𝟐𝒎