Chapter 1

Solutions of Algebraic and

Transcendental Equations

1.1 Introduction
In engineering and sciences, a frequently occurring problem is that of determining
the solutions of equations of the form

f (x) = 0 (1.1)

where f (x) can be a linear, quadratic, cubic or quartic expression, in which case
algebraic methods of solution are available. However, when f (x) is a polynomial
of higher degrees, or a transcendental function, algebraic methods of solution are
not available, hence we have to look for approximate methods of solution. In this
section, we shall consider approximation methods such as the bisection method,
fixed-point iterative method, method of false position and the Newton-Raphson
method.

1
1.2 The Bisection Method
Recall from MTH103 the intermediate value theorem thus:

Theorem 1 If f (x) is a continuous function on [a, b], and if f (a) and f (b) are of
opposite signs, then there exists (a, b) such that f ()=0.

In particular, suppose that f (a) < 0 and f (b) > 0, then there is a root lying
between a and b. The approximate value of this root is given by

a+b
x0 = .
2

If f (x0 ) = 0, we conclude that x0 is the root of the equation (1.1). Otherwise, the
root lies between a and x0 or between x0 and b according as f (x0 ) is negative or
positive. Continuing, we bisect the interval and repeat the process until a root is
known to the desired accuracy. This process is shown in the diagram below:

Example 1
Find a real root of the equation x3 x 1 = 0.
Solution
Note that f (1) < 0 and f (2) > 0, thus a root lies between 1 and 2 and we thus
take
1+2 3
x0 = = .
2 2
Hence  
3 27 3 7
f (x0 ) = f = 1 = > 0.
2 8 2 8
Thus the root lies between 1 and 32 . We then obtain x1 = 1+1.5
2
= 1.25, and
f (x1 ) = 19
64
< 0. We therefore conclude that the root lies between 1.25 and 1.5.
Continuing, one obtains

1.25 + 1.5
x2 = = 1.375.
2

2
The procedure is repeated and we obtain the successive approximations

x3 = 1.3125, x4 = 1.34375, x5 = 1.328125, . . .

Stop the iteration as soon as the required accuracy is attained.

1.3 Fixed Point Iteration Method

We transform the equation (1.1) algebraically into the form

x = (x) (1.2)

The solution of (1.2) is called a fixed point of , hence the name of the method.
From (1.1), we may get several forms of (1.2). For example, the equation

x3 + x2 1 = 0

can be expressed as

(i) x = 1
1+x

(ii) x = 1 x2

3
(iii) x = 1 x2 and so on.

Suppose x0 is an approximate value of the desired root . Substituting it for x in

(1.3), we obtain the first and and successive approximations as

x1 = (x0 )

x2 = (x1 )

x3 = (x2 )

3
 ..
.

xn = (xn1 )

and in general,
xn+1 = (xn ).

However, the sequence of successive approximations x0 , x1 , x2 , , xn do not

always converge to some value . For example, take the equation

x = 10x + 1.

If we take x0 = 0, x1 = 2, x2 = 102 +1 = 101, x3 = 10101 +1, . This sequence

obviously diverges as n increases. When does the iteration converge?

Theorem 2 (Convergence of Fixed Point Iteration Method)

Let x = s be a solution of x = g(x) and suppose that g has a continuous derivative
in some interval I containing s. Then if |g (x)| K 1 in I, the iteration process
defined by x = g(x) converges for any s in I.

Example 2
Use the fixed point iteration method to find the solution of the equation
x3 + x 1 = 0.
Solution:
1
The equation may be written in the form x = 1+x2
, while the iteration can be
expressed as
1
xn+1 = .
1 + x2n
A rough sketch indicates that there is a solution near x = 1, and taking x0 = 1,
we obtain the iterations

x1 = 0.5, x2 = 0.800, x3 = 0.610, x4 = 0.729, x5 = 0.653, x6 = 0.701,

4
The solution to 6 places of decimals is 0.682328.
Exercise
Use the fixed point iteration method to determine the solution of the equation
x2 x = 0. Compare your solution with the exact solution.

1.4 Newton-Raphson Method

This is another iteration method for solving equation (1.1), where f is assumed
to have a continuous first derivative f . Suppose x = x0 + h is the correct root of
(1.1), then expanding f (x0 + h) in Taylor series leads to

f (x0 + h) = f (x0 ) + hf (x0 ) + = 0.

Neglecting the second and higher order derivatives, on obtains

f (x0 )
h= .
f (x0 )
Therefore,
f (x0 )
x1 = x0 + h x1 = x0
f (x0 )
f (x1 )
x2 = x1
f (x1 )
f (x2 )
x3 = x2
f (x2 )
..
.
We then have the iteration scheme
f (xn )
xn+1 = xn (1.3)
f (xn )

If f (xn ) = 0, then use a different starting value. The Newton-Raphson method is

used to improve the results obtained by previous methods.

5
 Example 3
Determine the solution of the equation x3 + x 1 = 0 using the Newton-Raphson
method.
Solution:

f (x) = x3 + x 1, f (x) = 3x2 + 1,

f (xn ) x3 + xn 1
xn+1 = xn = xn n 2 .
f (xn ) 3xn + 1
taking x0 = 1, one obtains
x30 + x0 1
x1 = x0 = 0.75,
3x20 + 1
x31 + x1 1
x2 = x1 = 0.68604651,
3x21 + 1
and iterate to convergence to the required degree of accuracy.
Example 4
Find the root of the equation x sin x + cos x = 0
using the Newton-Raphson method.
Solution:

f (x) = x sin x + cos x, f (x) = x cos x,

The iteration formula then becomes

xn sin xn + cos xn
xn+1 = xn .
xn cos xn
Take x0 = and then obtain
x1 = 2.823

x2 = 2.7986

x3 = 2.7984

x4 = 2.7984,

Note the convergence of the solution above to 4 places of decimals.

6
1.4.1 Convergence of Newton-Raphson Method
f (x) f (x)f (x)
Consider (x) = x f (x)
, thus (x) = (f (x))2
. We assume f (x), f (x) and
f (x) are continuous and bounded on any interval containing the root x = of
equation (1.1). If is a simple root, then f () 6= 0, and since f (x) is continu-
ous, |f (x)| , for some > 0 in a suitable neighbourhood of . Within this
neighbourhood, we can select an interval such that |f (x)f (x)| < 2 and this is
possible since f () = 0 and since f (x) is continuously twice differentiable. Thus,
in this interval we have | (x)| < 1.
Hence by theorem 1, Newton-Raphson method converges, provided the initial
approximation is sufficiently close to . Newton-Raphson method still converges,
but slowly when is a multiple root. To determine the rate of convergence of the
method, we note that f () = 0 so that Taylors expansion gives

1
f (xn ) + ( xn )f (xn ) + ( xn )2 f (xn ) + = 0
2

from which we get

1
2
( xn )2 f (xn ) f (xn )
( xn ) + = .
f (xn )
f (xn )

Using Newton-Raphson iteration formula, we obtain

1
2
( xn )2 f (xn )
xn+1 =
f (xn )

and setting n = xn , we have

1 2
f ()
2 n
n+1 = (1.4)
f ()

Hence Newton-Raphson method has a second order or quadratic convergence.

Assignment
Obtain a root of each of the equations
1. 2x = 3 + cos x

7
 3
2. ex tan1 x 2
=0
3. x2 + cos x xex = 0
correct to 3 places of decimals using

(a) the bisection method

(b) fixed point iteration method

(c) Newton-Raphson method.

Compare the three results in tabular form.

1.5 Numerical Evaluation of Multiple Roots of

Algebraic Equations
We say that the function f (x) has a root of multiplicity p > 1 if

f (x) = (x )p h(x) (1.5)

with h() 6= 0 and h(x) continuous at x = . We restrict p to be a positive

integer, although the result could be valid for non-integral values for some cases.

1.5.1 Newton-Raphson Method and Multiple Roots

Most of the rootfinding methods considered so far perform poorly when the root
being sought is multiple. We now investigate this for Newton-Raphson method as
a fixed point method with f (x) satisfying (1.5). Thus
f (x)
xn+1 = g(xn ), g(x) = x , x 6= .
f (x)
The expression for g(x) on using (1.5) is obtained as

f (x) = (x )p h (x) + p(x )p1h(x)

8
 (x )p h(x) (x )h(x)
g(x) = x = x .
(x )p1 [(x )h (x)] ph(x) + (x )h (x)
One then obtains
 
h(x) d h(x)
g (x) = 1 (x )
ph(x) + (x )h (x) dx ph(x) + (x )h (x)

and g () = 1 p1 6= 0 for p > 1. You can then find p for any given . This shows
p1
that Newton-Raphson Method is a linear method with rate of convergence p
.
Example 5
Determine the multiplicity of the root of the equation

x4 5.4x3 + 10.56x2 8.954x + 2.7951 = 0

lying in the interval 1 x 2.

Solution:
f (x) = x4 5.4x3 + 10.56x2 8.954x + 2.7951,
f (x) = 4x3 16.2x2 + 21.12x 8.954
x1 = 1.032352941, x2 = 1.054435334, x3 = 1.069406530,
x4 = 1.079501623, x5 = 1.086286231,
Thus
f (x)
g(x) = x .
f (x)
Now taking x 1.086 as the approximate root, we obtain
g (1.086) = 0.6698244,
p1
thus we solve the equation p
= 0.6698244 which yields p = 3.02869144.
Hence p 3. Therefore the root has multiplicity 3.

1.6 Solutions of Systems of Nonlinear Equations

We consider in this section the numerical solution of simultaneous equations by
two methods:

9
 (a) Fixed Point Iteration method and

(b) Newton-Raphson Method

We shall consider the case of two equations in two unknowns only.

1.6.1 Fixed Point Iteration Method

Consider the nonlinear system of equations

f (x, y) = 0 (1.6)

g(x, y) = 0

Re-write the system in the form

x = F (x, y) (1.7)

y = G(x, y)

where F and G satisfy the conditions of convergence

   
 F   G 
 x   x  < 1
 + 

and    
 F   G 
 y  +  y  < 1
   

in the neighbourhood of the root. Suppose (x0 , y0 ) is the initial approximation to

the root. Then the sequence of successive approximations becomes

x1 = F (x0 , y0 ), y1 = G(x1 , y0 ),

x2 = F (x1 , y1 ), y1 = G(x2 , y1 ),

x3 = F (x2 , y2 ), y3 = G(x3 , y2 ),

10
 ..
.

xn+1 = F (xn , yn ), yn+1 = G(xn+1 , yn ).

Iterate to convergence.
Example 6
Find the root of the system of equations

x2 + y 2 1 = 0 (1.8)
2
x
+ 4y 2 1 = 0 (1.9)
2
correct to 3 places of decimals using the initial value (x0 , y0 ) = (1, 1). Trans-
forming the equations lead to the iteration scheme
r
p 1 x2n+1
xn+1 = 1 yn2 , yn+1 = .
4 8
Thus we obtain the iterations
x1 = 0, y1 = 0.5
x2 = 0.866025404, y2 = 0.3952847075
x3 = 0.9185586535, y2 = 0.3801726581
x4 = 0.9249155367, y4 = 0.3782412011
x5 = 0.9257070778, y5 = 0.3779990751
x6 = 0.9258059728, y6 = 0.3779687984
We have then achieved 3 places of decimals accuracy.
That is x 0.926, y 0.378.

1.6.2 Newton-Raphson Method

Suppose the initial approximation to the root of the system is (x0 , y0 ), and if
(x0 + h, y0 + k) is the exact root of the system, then we must have that

f (x0 + h, y0 + k) = 0, g(x0 + h, y0 + k) = 0.

11
If f and g are sufficiently differentiable, we can expand them in Taylors series as

f f
f0 + h +k + = 0 (1.10)
x0 y0
g g
g0 + h +k + = 0
x0 y0

where
   
f f f f
f0 = f (x0 , y0), g0 = g(x0 , y0), = , = .
x0 x x=x0 y0 y y=y0

Neglecting the second and higher order terms, we obtain the system of linear
equations

f f
h +k = f0 (1.11)
x0 y0
g g
h +k = g0
x0 y0

The system (1.9) can be solved for h and k to obtain the new solution as

x1 = x0 + h, y1 = y0 + k.

The process is repeated until the required accuracy is attained.

Example 6
Find the real roots of the equations

x2 y 2 = 4

x2 + y 2 = 16

Solution:

This system has the solutions x = 10 = 3.162 and y = 6 = 2.449. To apply
the Newton-Raphson method, we replace the hyperbola by its asymptote y = x.

This produces the initial values x0 = y0 = 2 2 = 2.828. However, in general,
this is obtained from a rough graph.

12
Taking f = x2 y 2 4, g = x2 + y 2 16
f
x
= 2x, f
y
g
= 2x, x = 2x, g
y
= 2y,
and f0 = x20 y02 4 = 4, g0 = x20 + y02 16 = 0.
f
f g g
x 0
= 2x 0 = 4 2, y 0
= 2x0 = 4 2, x 0
= 4 2, y 0
= 4 2. Thus one
obtains the equations
h k = 0.707

h+k =0

which have the solutions h = 0.354, k = 0.354. The second approximation is

then given by
x1 = x0 + h = 3.182, y1 = y0 + k = 2.474.

Using (x1 , y1 ) and repeating the process, we obtain h = 0.0204, k = 0.0242

and thus x2 = x1 + h = 3.162, y2 = y1 + k = 2.450.

For a system of equations, the procedure is modified accordingly. Thus to

obtain the solution to the system

f (x, y) = x2 + y 2 1 = 0

x2
g(x, y) = + 4y 2 1 = 0
2
subject to the initial approximation [x0 , y0 ] = [1, 1], we replace f (x) in the
Newton-Raphson iteration scheme with the Jacobian J(x) of the system. Thus
the iteration scheme becomes

f (xn )
xn+1 = xn = xn f (xn )J 1 (xn ) (1.12)
J(xn )

f f
xn yn
where Xn = (xn , yn ) and J(xn ) = and J 1 (xn ) is the inverse of
g g
xn yn

13
J(xn ). Thus the iteration scheme becomes

(f (xn , yn ), g(xn , yn ))
(xn+1 , xn+1 ) = (xn , yn ) (1.13)
f f
xn yn
g g
xn yn

Hence, for the above system, we obtain

 2

x2 + y 2 1, x2 + 4y 2 1
(xn+1 , yn+1 ) = (xn , yn )
2x 2yn
n
xn 8yn

Taking x0 = (1, 1) leads to the following sequence of approximations:

(1, 3.5) 0.57142857 0.14285714
(x1 , y1 ) = (1, 1) = (1, 1) (1, 3.5)
2 2 0.07142857 0.14285714

1 8
= (0.67857143, 0.64285714).

(x1 y1 ) = (0.67857143 0.64285714)

(x2 y2 ) = (0.88305212 0.41999860)
(x3 y3 ) = (0.92764766 0.38043036) (1.14)
(x4 y4 ) = (0.92613160 0.37779498)
(x5 y5 ) = (0.92581236 0.37794889)

Hence we have achieved convergence to three places of decimals, that is

x 0.926, y 0.378.
Class Exercise:
Use the fixed point iteration method to obtain the solution of example 6 above.
General Exercises
1. Obtain a root for each of the following equations using the bisection method
(a) x3 + x2 + x + 7 = 0 (b) x3 x2 1 = 0

14
2. Use the fixed point iteration method to find a root of each of the following
equations correct to 4 s.f.
1
(a) cos x = 3x 1 (b) x = (1+x)2

3. Use Newton-Raphson method to solve the equation

(a) x4 x3 2x 34 = 0 (b) tan x = 4x.

15
Chapter 2

Interpolation

Given a function f (x) defined on the interval x0 < x < xn , and assuming that
f (x) is single-valued and continuous and that it is known explicitly, then the val-
ues of f (x) corresponding to certain given values of x, say x0 , x1 , x2 , , xn , can
easily be computed and tabulated. However, the major problem of numerical anal-
ysis is the converse, that is, given the set of tabular values
(x0 , y0 ), (x1 , y1 ), (x2 , y2 ), , (xn , yn ) satisfying the relation y = f (x) where the
explicit nature of f (x) is unknown, it is required to find a simpler function (x)
such that f (x) and (x) agree at the set of tabulated points. Such a process is
known as interpolation. If (x) is a polynomial, then the process is called poly-
nomial interpolation and (x) is called the interpolating polynomial.

2.1 Finite Differences

Assuming we have a table of values (xi , yi ), i = 0, 1, 2, , n of any function
y = f (x), the values of x being equally spaced, i.e. xi = x0 +ih, i = 0, 1, 2, , n.
Suppose that we are required to recover the values of f (x) for some intermediate
values of x or to obtain the derivative of f (x) for some x in the range x0 < x < xn .

16
The methods for the solution of these problems are based on the concept of the
differences of a function which we now proceed to define.

2.1.1 Forward Differences

If y0 , y1, y2 , , yn denote a set of values of y, then

y1 y0 , y2 y1 , y3 y2 , , yn yn1

are called the differences of y. Denoting the differences by

y0 , y1 , y2 , , yn1 respectively, we have

y0 = y1 y0 , y1 = y2 y1 , y2 = y3 y2 , , yn1 = yn yn1 .

where is called the forward difference operator and y0 , y1 , y2 , are

called first forward differences. The differences of the first forward differences
are called second forward differences and are denoted by 2 y0 , 2 y1 , 2 y2 , .
Similarly, one can define the third and higher differences. Thus,

2 y0 = (y1 y0 ) = y1 y0 = y2 2y1 + y0

3 y0 = 2 y1 2 y0 = y2 y1 (y1 y0 )

= y3 y2 (y1 y0 ) [y2 y1 (y1 y0 )]

= y3 3y2 + 3y1 y0 .

Similarly,

4 y0 = 3 y1 3 y0 = y4 4y3 + 6y2 4y1 + y0 .

Note that the coefficients in the above expansions are binomial coefficients.

17
 x y 2 3 4 5 6
x0 y0 y0
x1 y1 y1 2 y0 3 y0
x2 y2 y2 2 y1 3 y1 4 y0
x3 y3 y3 2 y2 3 y2 4 y1 5 y0
x4 y4 y4 2 y3 3 y3 4 y2 5 y1 6 y0
x5 y5 y5 2 y4 3 y4 4 y3
x6 y6 y6 2 y5

Table 2.1: The table of forward differences.

2.1.2 Backward differences

The differences y1 y0 , y2 y1 , y3 y2 , , yn yn1 are called first back-

ward differences if the are denoted by y1 , y2 , y3, , yn respectively so
that y1 = y1 y0 , y2 = y2 y1 , , yn = yn yn1 , where is called the
backward difference operator. In a similar way, one can define backward differ-
ences of higher orders. Thus we have

2 y2 = y2 y1 = y2 2y1 + y0 ,

3 y3 = 2 y3 2 y2 = y3 3y2 + 3y1 y0 ,

and so forth.

2.1.3 Central Differences

The central difference operator is defined by the relations

y 1 = y1 y0 , y 3 = y2 y1 , , yn 1 = yn yn1 .
2 2 2

Higher order central differences can be obtained in a similar manner. Using the
values from the two tables above, the central difference table can be formed thus

18
 x y 2 3 4 5 6
x0 y0
x1 y1 y1
x2 y2 y2 2 y 2
x3 y3 y3 2 y 3 3 y 3
x4 y4 y4 2 y 4 3 y 4 4 y 4
x5 y5 y5 2 y 5 3 y 5 4 y 5 5 y 5
x6 y6 y6 2 y 6 3 y 6 4 y 6 5 y 6 6 y 6

Table 2.2: The table of backward differences.

x y 2 3 4 5 6
x0 y0 y 1
2

x1 y1 y 3 2 y1 3y 3
2 2
2 3
x2 y2 y 5 y2 y5 4 y2 5y 5
2 2 2
2 3 4 5
x3 y3 y 7 y3 y7 y3 y7 6 y3
2 2 2

x4 y4 y 9 2 y4 3y 9 4 y2
2 2

x5 y5 y 11 2 y5
2

x6 y6

Table 2.3: The table of central differences.

Note from the three tables that

y0 = y1 = y 1 ,
2

3 y2 = 3 y5 = 3 y 7 ,
2

and so on.

19
2.1.4 symbolic Relations

We can establish difference formulae by using the shift operator E in addition to

the operators , and defined above. The shift operator E is defined by

Eyi = yi+1

E 2 yi = E(Eyi ) = E(yi+1) = yi+2

and in general
E n (yi ) = yi+n (2.1)

Now, a relationship between and E can be derived as follows:

y0 = y1 y0 = Ey0 y0 = (E 1)y0 . Hence = E 1 or E = 1 + . This
implies that the effect of E on y0 is the same as that of the operator (1 + ) on y0 .
The higher order forward differences in terms of the given function values are

2 y0 = (E 1)2 y0 = (E 2 2E + 1)y0 = E 2 y0 2Ey0 + E 0 y0 = y2 2y1 + y0 ,

3 y0 = (E 1)3 y0 = (E 3 3E 2 + 3E 1)y0 = y3 3y2 + 3y1 y0 .

The following relations can also be established from the definitions

y0 = y0 E 1 y0 = (1 E 1 )y0

1 E 1 .

Also,
1 1 1 1
y0 = E 2 y0 E 2 = (E 2 E 2 )y0
1 1
E 2 E 2 ,

Hence
1
E E 2 .

20
2.1.5 Detecting Errors by Use of Difference Tables

Errors in tabulated values can easily be detected using difference tables. For ex-
ample, if we introduce an error e in a tabulated values of zeroes, we can notice
how the error propagates fanwise at higher differences and also gets magnified.
The following characteristics of error propagation can be deduced from the table
that follows:

(i) Error increases with the order of the differences

(ii) The errors in any one column are the binomial coefficients with alternating
signs

(iii) The algebraic sum of the errors in any difference column is zero

(iv) The maximum error occurs opposite the function value containing the error.

Suppose an error e is introduced into the table of values of a given function, then
these indicators can be used to detect errors in difference tables as can be seen
from table (2.4) below.

2.2 Differences of a Polynomial

Suppose y(x) is a polynomial of degree n so that

y(x) = a0 xn + a1 xn1 + a2 xn2 + + an (2.2)

We then obtain the difference of y(x) as

y(x) = y(x + h) y(x)

= a0 [(x + h)n xn ] + a1 (x + h)n1 xn1 +

 
(2.3)

= a0 nhxn1 + a1 xn2 + + an

21
 y 2 3 4 5
0
0 0 0 0
0 0 0 0 0 e
0 0 0 e e -5e
0 e e -3e -4e 10e
e -e -2e 3e 6e -10e
0 0 e -e -4e 5e
0 0 0 0 e -e
0 0 0 0 0
0 0 0
0 0

Table 2.4: Error spread in a difference table

which shows that the difference of a polynomial of degree n is a polynomial of

degree n1. Similarly, the second difference will be a polynomial of degree n2
and n y(x) will then be a constant.
For example, consider the polynomial y(x) = ax2 + bx + c.
The first and second forward differences become

y(x) = y(x + h) y(x)

= a(x + h)2 + b(x + h) + c ax2 bx c

= (2ax + ah + b)h

2 y(x) = y(x + h) y(x)

= 2ah2

22
2.2.1 Newtons Interpolation Formulae

Given the set of n + 1 values, (x0 , y0), (x1 , y1 ), (x2 , y2 ), , (xn , yn ), of x and
y, it is required to find yn (x), a polynomial of degree n such that y and yn (x)
agree at the tabulated points. Suppose the values of x are equidistant, and let
xi = x0 + ih, i = 0, 1, 2, , n. Since yn (x) is a polynomial of degree n, it may
be written as

yn (x) = a0 + a1 (x x0 ) + a2 (x x0 )(x x1 ) + + an (x x0 )(x x1 ) (x xn1 )

(2.4)
Now, we impose the condition that y and yn (x) should agree at the set of tabulated
points, i.e y(xi ) = yi (xi ), we obtain

y1 y0 y0
y0 = a0 , y1 = a1 (x1 x0 ) + y0 a1 = = .
x1 x0 h

y2 = a0 + a1 (x2 x1 ) + a2 (x2 x0 )(x2 x1 )

y1 y0
= y0 + (x2 x0 ) + a2 h(x2 x0 )
x1 x0
y2 y0 y0
= + a2 h
x2 x0 h
 
1 y2 y0 y1 y0
a2 =
h 2h h
y2 2y1 + y0 2 y0
= = .
2h2 2h2

Similarly,
3 y0 4 y0 n y0
a3 = , a4 = , , an = .
3!h3 4!h4 n!hn
Setting x = x0 + ph and substituting for a0 , a1 , a2 , , an in (2.4) above gives

p(p 1) 2 p(p 1)(p 2) 3

yn (x) = y0 + py0 + y0 + y0
2! 3! (2.5)
p(p 1)(p 2) ()p n + 1 n
++ y0
n!

23
This is called Newtons forward interpolation formula and it is useful for interpo-
lation near the beginning of a set of tabular values.
The error committed in replacing the function y(x) by the polynomial yn (x)
can be obtained as

(x x0 )(x x1 ) (x xn ) (n+1)
y(x) yn (x) = y () (2.6)
(n + 1)!

for x0 < < xn . However, (2.6) is not useful in practice, since we have no
information concerning y (n) (x). In any case, if y (n) (x) does not vary rapidly in
the interval, a useful estimate of the derivative can be obtained thus:
We obtain the Taylors series expansion of y(x + h) as

h2
y(x + h) = y(x) + hy (x) + y (x) +
2!

and neglecting the term containing h2 and higher powers of h, produces

1 1
y (x) = [y(x + h) y(x)] = y(x).
h h
1
Thus y (n+1) (x) = hn+1
n+1 y(x). Equation (2.6) can then be written as

p(p 1)(p 2) (p n) n+1

y(x) yn (x) = y() (2.7)
(n + 1)!

a form suitable for computation.

If we assume (2.4) in the form

yn (x) = a0 +a1 (xxn )+a2 (xxn )(xxn1 )+ +an (xxn )(xxn1 ) (xx1 )

and then impose the condition that y(x) and yn (x) should agree at the tabulated
points xn , xn1 , xn2 , , x2 , x1 , x0 , we obtain in a similar manner as above

p(p + 1) 2 p(p + 1) (p + n 1) n
yn (x) = yn + pyn + yn + + yn
2! n!
(2.8)

24
 xxn
where p = h
. This is called Newtons backward difference interpolating for-
mula and it uses tabular values to the left of yn . This formula is useful for inter-
polation near the end of tabulated values. The error is estimated as

p(p + 1)(p + 2) (p + n) n+1 n+1

y(x) yn (x) = h y () (2.9)
(n + 1)!

where x0 < < xn and x = xn + ph.

Example 7
Find the cubic polynomial which takes the following values:
x 0 1 2 3
y 1 0 1 10

Hence or otherwise obtain y(4).

Solution:
We form the forward difference table thus:

x y 2 3
0 1 -1
1 0 1 2
2 1 9 8 6
3 10
Here, h = 1 (from the forward difference table). Using x = x0 + ph and choosing
x0 = 0, one obtains x = 0 + p = p. Thus substituting for p in (2.5), one obtains
x(x 1) x(x 1)(x 2)
y(x) = 1 + x(1) + (2) + (6) = x3 2x2 + 1,
2! 3!
which is the polynomial from which we obtained the tabulated values above. y(4)
can be obtained by substituting x = 4 in the polynomial obtained above. Note that
this process of obtaining the value of y for some value of x outside the given range
is called extrapolation and this demonstrates the fact that if a tabulated function is
a polynomial, then interpolation and extrapolation would yield exact values.

25
 Example 8
The table below gives the values of tan x for 0.10 x 0.30.

x 0.10 0.15 0.20 0.25 0.30

y = tan x 0.1003 0.1511 0.2027 0.2553 0.3093

Find: (a) tan(0.12) (b) tan(0.26) (c) tan(0.40) (d) tan(0.50).

Solution:

The table of froward differences is

x y 2 3 4
0.10 0.1003
0.15 0.1511 0.0508 0.0008
0.20 0.2027 0.0516 0.0010 0.0002
0.25 0.2553 0.0526 0.0014 0.0004 0.0002
0.30 0.3093 0.0540

(i) Taking x = x0 + ph, where x0 = 0.10, h = 0.05, thus to find the value of
tan(0.12), we have that 0.12 = 0.10 + 0.05p p = 0.4, and using (2.5) yields

0.4(0.4 1)
tan(0.12) = 0.1003 + 0.4(0.0508) + (0.0008)
2
0.4(0.4 1)(0.4 2) 0.4(0.4 1)(0.4 2)(0.4 3)
+ (0.0002) + (0.0002) = 0.1205.
6 24

(i) To obtain tan(0.26), we have 0.26 = 0.30 + p(0.05) p = 0.8. We now use
Newtons backward interpolation formula (2.8), since we are estimating a value at

26
the end of the data. Hence one obtains

0.8(0.8 + 1)
tan(0.26) = 0.3093 0.8(0.0540) (0.0014)
2
0.8(0.8 + 1)(0.8 + 2) 0.8(0.8 + 1)(0.8 + 2)(0.8 + 3)
(0.0004) (0.0002)
6 24
= 0.2662.

Using formula (2.5) and proceeding as in (i), one gets

(iii) tan(0.40) = 0.4241 and
(iv) tan(0.50) = 0.5543.
Note that the actual values of tan(0.12), tan(0.26), tan(0.40), tan(0.50) are re-
spectively 0.1206, 0.2660, 0.4228 and 0.5463. Thus if the tabulated function is
not a polynomial, then extrapolation very far from the table limits would be dan-
gerous, although interpolation can be carried out very accurately.

2.3 Interpolation with unevenly spaced points

In the previous section, we derived interpolation formulae and discussed their
practical use in some detail. However, they possess a major disadvantage of re-
quiring the values of the independent variable to be equally spaced (i.e uniform
grids). It is therefore desirable to have interpolation formulae with unequally
spaced values of the argument (non-uniform grids). We discuss in this section
Lagranges interpolation formula which uses only the function values.

2.3.1 Lagranges interpolation formula

Let y(x) be continuous and differentiable (n+1) times in the interval (a, b). Given
the n+1 points (x0 , y0 ), (x1 , y1 ), (x2 , y2 ), , (xn , yn ), where the values of x need
not necessarily be equally spaced, we wish to find a polynomial of degree n, say

27
n (x) such that
i (xi ) = y(xi ) = yi , i = 0, 1, 2, , n (2.10)

Let
n (x) = a0 + a1 x + a2 x2 + + an xn (2.11)

be the desired polynomial. Substituting the condition (2.10) into (2.11), we obtain
the system of equations

y0 = a0 + a1 x0 + a2 x20 + + an xn0

y1 = a0 + a1 x1 + a2 x21 + + an xn1

y2 = a0 + a1 x2 + a2 x22 + + an xn2 (2.12)

.. .. .. .. ..
. . . . .

yn = a0 + a1 xn + a2 x2n + + an xnn

The system (2.12) will have a solution if the determinant

 
2
 1 x0 x0 x0 
 n 
 
 1 x1 x21 xn1 
 
 
 1 x2 x22 xn2  6= 0 (2.13)
 
 .. .. .. .. 
 
 . . . . 
 
 1 xn x2n xnn 
 

The value of this determinant called Vandermondes determinant is

(x0 x1 )(x0 x2 ) (x0 xn )(x1 x2 ) (x1 xn ) (xn1 xn ).

28
Eliminating a0 , a1 , a2 , , an from equations (2.11) and (2.12), we obtain
 
2
 n (x) 1 x x x 
 n 
 
2 n
 y0 1 x0 x0 x0 
 
 
2 n  = 0 (2.14)
 y1 1 x1 x1 x1

 .. .. .. .. . 
 
 . . . . .. 
 
1 xn x2n xnn 
 
 yn

which shows that n (x) is a linear combination of y0 , y1 , y2, , yn . Hence we

write
n
X
n (x) = Li (x)yi (2.15)
i=0
where Li (x) are polynomials in x of degree n. Since j (xj ) = yj for
j = 0, 1, 2, , n, equation (2.15) gives

Li (xj ) = 0,

if i 6= j
ij =
Li (xj ) = 1,

if i = j

where ij is the Kronecker delta. Hence Li (x) may be written as

(x x0 )(x x1 ) (x xi1 )(x xi+1 ) (x xn )

Li (x) = (2.16)
(xi x0 )(xi x1 ) (xi xi1 )(xi xi+1 ) (xi xn )
which satisfies the condition above. The function

(x) = (x x0 )(x x1 ) (x xi1 )(x xi+1 ) (x xn )

Y n
= (x xi ) (2.17)
i=0

may be used to express the Lagrange multiplier function in the more compact
form
(x)
Li (x) = .
(x xi ) (xi )
Hence (2.15) yields
n
X (x)
n (x) = yi , (2.18)
i=0
(x xi ) (xi )

29
which is known as Lagranges interpolation formula. The Li (x) are called
Lagranges interpolation coefficients. Interchanging x and y in (2.18) yields the
formula
n
X (y)
n (y) = xi , (2.19)
i=0
(y yi ) (yi )

which is useful for inverse interpolation. It is important to note that the Lagranges
interpolation polynomial is unique. From (2.18), we note that

1 (x) = L0 (x)y0 + L1 (x)y1 (2.20)

where
(x) (x x0 )(x x1 ) x x1
L0 (x) = = =
(x x0 ) (x)
(x x0 )(x0 x1 ) x0 x1
(x) (x x0 )(x x1 ) x x0
L1 (x) = = =
(x x1 )(x0 x1 ) (x x1 )(x0 x1 ) x1 x0
x x1 x x0
(x) = y0 + y1 .
x0 x1 x1 x0
This is a linear interpolation formula. For a quadratic interpolation, we have

2 (y) = L0 (x)y0 + L1 (x)y1 + L2 (x)y2 (2.21)

where

(x x0 )(x x1 )(x x2 ) (x x1 )(x x2 )

L0 (x) = = ,
(x x0 )(x0 x1 )(x0 x2 ) (x0 x1 )(x0 x2 )
(x x0 )(x x1 )(x x2 ) (x x0 )(x x2 )
L1 (x) = =
(x x1 )(x1 x0 )(x1 x2 ) (x1 x0 )(x1 x2 )
and
(x x0 )(x x1 )(x x2 ) (x x0 )(x x1 )
L2 (x) = = .
(x x2 )(x2 x0 )(x2 x1 ) (x2 x0 )(x2 x1 )
(x x1 )(x x2 ) (x x0 )(x x2 ) (x x0 )(x x1 )
2 (x) = y0 + y1 + y2 .
(x0 x1 )(x0 x2 ) (x1 x0 )(x1 x2 ) (x2 x0 )(x2 x1 )
Example 9
Obtain a linear Lagranges polynomial for the tabulated values for y = ln x.

30
Hence find the value of y(9.2) correct to 4 places of decimals, and determine
the error in the estimate.

x 9.0 9.5
y = ln x 2.1972 2.2513

Solution:
Here, x0 = 9.0, x1 = 9.5, y0 = 2.1972, y1 = 2.2513.

x x1 x x0
1 (x) = L0 (x)y0 + L1 (x)y1 = y0 + y1
x0 x1 x0 x1
x 9.5 x 9.0 x 9.5 x 9.0
= ln(9.0) + ln(9.5) = (2.1972) + (2.2513)
9.0 9.5 9.5 9.0 0.5 0.5
9.2 9.5 9.2 9.0
1 (9.2) = (2.1972) + (2.2513) = 2.2188.
0.5 0.5
error = 2.2192 2.2188 = 0.0004.

Thus linear interpolation is not sufficient to get 4 d.p. accuracy, it would suffice
for 3 d.p. accuracy.
Example 10
Obtain the value of ln(9.2) using the quadratic interpolation polynomial using the
data below:
x 9.0 9.5 11.0
y = ln x 2.1972 2.2513 2.3979

Solution:

(x 9.5)(x 11.0)
L0 (x) = = x2 20.5x 104.5, L0(9.2) = 0.5400
(9.0 9.5)(9.0 11.0)

(x 9.0)(x 11.0) 4
L1 (x) = = (x2 20x + 99), L1 (9.2) = 0.4800
(9.5 9.0)(9.5 11.0) 3
(x 9.0)(x 9.5) 1
L2 (x) = = (x2 18.5x + 85.5), L2 (9.2) = 0.0200.
(11.0 9.0)(11.0 9.5) 3

31
 2 (9.2) = ln(9.2) = 0.5400(2.1972)+0.4800(2.2513)0.0200(2.3979) = 2.2192

which is exact to 4 d.p.

Example 11
Fit a cubic Lagranges interpolation polynomial to the data below:

x 0 1 2 3
y 1 0 1 10

Solution:

(x x1 )(x x2 )(x x3 ) 1
L0 (x) = y0 = (x 1)(x 2)(x 3)
(x0 x1 )(x0 x2 )(x0 x3 ) 6

(x x0 )(x x2 )(x x3 )
L1 (x) = y1 = 0
(x1 x0 )(x1 x2 )(x1 x3 )
(x x0 )(x x2 )(x x3 ) 1
L2 (x) = y2 = x(x 1)(x 3)
(x2 x0 )(x2 x1 )(x2 x3 ) 2
(x x0 )(x x1 )(x x2 ) 5
L3 (x) = y3 = x(x 1)(x 2).
(x3 x0 )(x3 x1 )(x3 x2 ) 3
3 (x) = L0 (x)y0 + L1 (x)y1 + L2 (x)y2 + L3 (x)y3
1 1 5
= (x1)(x2)(x3)+0 x(x1)(x3)+ x(x1)(x2) = (x1)(x2 x1)
6 2 3
3 (x) = x3 2x + 1.

This is the same polynomial we have earlier obtained using Newtons forward
interpolation.

2.4 Least Squares Method for Curve Fitting

The least squares method is one of the most systematic procedures for fitting a
unique curve through given data points and is widely used in practical computa-
tions.

32
Let the set of data points be (xi , yi ), i = 0, 1, 2, , m, and let the curve given by
y = f (x) be fitted to this data. At x = xi , the experimental (or observed) value of
the ordinate is yi and the corresponding value on the fitting curve is f (xi ). If ei is
the error of approximation at x = xi , then we have

ei = yi f (xi ) (2.22)

Now, if we let

S = [y1 f (x1 )]2 + [y2 f (x2 )]2 + + [ym f (xm )]2 (2.23)

= e21 + e22 + + e2m ,

then the method of least squares consists of minimising S, that is the sum of the
squares of the errors.

2.4.1 Fitting a straight line

Let y = a0 + a1 x be the straight line to be fitted to the given data. Then, corre-
sponding to equation (2.23), we have

S = [y1 (a0 + a1 x1 )]2 + [y2 (a0 + a1 x2 )]2 + + [ym (a0 + a1 xm )]2

(2.24)
For S to be minimum, we have

S
= 2[y1 (a0 + a1 x1 )] 2[y2 (a0 + a1 x2 )] 2[ym (a0 + a1 xm )] = 0
a0

and

S
= 2x1 [y1 (a0 +a1 x1 )]2x2 [y2 (a0 +a1 x2 )] 2xm [ym (a0 +a1 xm )] = 0
a1

Simplifying the above equations yield

ma0 + a1 (x1 + x2 + + xm ) = y0 + y1 + y2 + + ym

33
and

a0 (x1 + x2 + + xm ) + a1 (x21 + x22 + + x2m ) = x1 y1 + x2 y2 + + xm ym .

We then have the equations

m
X m
X
ma0 + a1 xi = yi (2.25)
i=1 i=1
m
X m
X m
X
a0 xi + a1 x2i = xi yi
i=1 i=1 i=1

where
yi )( x2i ) ( xi )( xi yi )
P P P P
(
a0 =
m x2i ( xi )2
P P

and P P P
m xi yi ( xi )( yi )
a1 = .
m x2i ( xi )2
P P

Since the xi and yi are known quantities, the equations (2.25) are called the normal
2S
equations, and can be solved for the unknowns a0 and a1 . We also note that a20
2S
and a21
are both positive at the points a0 and a1 . Thus these values provide a
minimum of S.
Now, dividing the first of equations (2.25) by m yields

a0 + a1 x = y (2.26)

where (x, y) is the centroid of the given data points. Hence the fitted straight line
passes through the centroid of the data points.
Example 12
The table below gives the temperatures T (in C) and lengths l (in mm) of heated
rod. If l = a + bT , find the best values for a and b that approximate the data of the
table below.
T 20 30 40 50 60 70
l 800.3 800.4 800.6 800.7 800.9 801.0

34
 Solution:

T l T2 lT
20 800.3 400 16006
30 800.4 900 24012
40 800.6 1600 32024
50 800.7 2500 40035
60 800.9 3600 48054
70 801.0 4900 56070
270 4803.9 13900 216201

Using the normal equations, one obtains

6a + 270b = 4803.9

270a + 13900b = 216201

which lead to a = 800 and b = 0.0146.

l = 800 + 0.0146b.

2.4.2 Nonlinear curve fitting

A nonlinear function y = xc can be fitted to the given set of data points. Taking
the logarithms of both sides yields

Y = a0 + a1 X (2.27)

where Y = log y, a0 = log , a1 = c and X = log x. Thus the procedure outlined

above can be followed to obtain the values of a0 and a1 , hence and c can be
calculated.
Now consider the parabola

y = a0 + a1 x + a2 x2 (2.28)

35
The least squares parabola approximating the set of points (xi , yi ), i = 0, 1, 2, , n
has the equation (2.28). The constants a0 , a1 and a2 can be determined by solving
simultaneously the equations.
X X X
y = na0 + a1 x + a2 x2
X X X X
xy = a0 x + a1 x2 + a2 x3 (2.29)
X X X X
x2 y = a0 x2 + a1 x3 + a2 x4

The equations (2.29) are called the normal equations for the least squares parabola
(2.28). This parabola is said to be the regression curve of y on x, since an estimate
of y is obtained for any given value of x. Similarly, we can have a regression
equation of x on y.
Example 13
Fit a curve of the form Y = ex to the data below

x 1 2 3 4
Y 7 11 17 27

Solution:
Taking the natural logarithms of both sides leads to

ln y = ln + x.

This is of the form y = a + bx, where y = ln Y, a = ln and b = . Thus we fit

a least squares line to the data

x 1 2 3 3
y 1.95 2.40 2.83 3.30

and one obtains = 4.48 and = 0.45. Thus the curve becomes

Y = 4.48e0.45x .

36
Exercise
Obtain a quadratic curve which fits the data in the table below, using the least
squares method.

x 1.2 1.8 3.1 4.9 5.7 7.1 8.6 9.8

y 4.5 5.9 7.0 7.8 7.2 6.8 4.5 2.7

37
Chapter 3

Numerical Differentiation and

Integration

3.1 Introduction
In chapter 2, we considered the problem of interpolation, that is, given a set of tab-
ulated values (x0 , y0 ), (x1 , y1 ), , (xn , yn ) of x and y, can we find a polynomial
yn (x) of the lowest degree such that y(x) and yn (x) agree at the set of tabulated
points? In this chapter, we are concerned with numerical differentiation and inte-
gration. This involves deriving formulae for derivatives and integrals from given
set of tabulated values of x and y in the interval [x0 , xn ].

3.2 Numerical Differentiation

A straightforward method for deriving numerical differentiation formulae is to
differentiate the interpolation formulae obtained in the previous chapter. Differ-
entiating the Newtons forward interpolating formula

yn (x) = y0 + (p1 ) y0 + (p2 ) 2 y0 + (p3 ) 3 y0 + + (pn ) n y0 (3.1)

38
 p!
where (pr ) = (pr)!r!
, leads to the following results:

3p2 6p + 2 3 2p3 9p2 + 11p 3 4

   
1 1 2
yn (x) = y0 + p y0 + y0 + y0 + (3.2)
h 2 6 12
6p2 18p + 11 4
 
1 2 3
yn (x) = 2 y0 + (p 1) y0 + y0 + (3.3)
h 12
 
1 3 2p 3 4
yn (x) = 3 y0 + y0 + (3.4)
h 2
1
yn(iv) (x) = 4 [4 y0 + ](3.5)
h

For tabulated value x0 , p = 0 and hence equations (3.2) - (3.4) simplify to

 
1 1 2 1 3 1 4
yn (x) = y0 y0 + y0 y0 + (3.6)
h 2 3 4
 
1 2 3 11 4
yn (x) = 2 y0 y0 + y0 + (3.7)
h 12
 
1 3 3 4
yn (x) = 3 y0 y0 + (3.8)
h 2

Using only the first terms of the formulae above, one obtains the forward finite
difference approximations to the following derivatives:

1 y(x + h) y(x)
y (x) = y0 =
h h
1 2 y(x + 2h) 2y(x + h) + y(x)
y (x) = y 0 =
h2 h2
1 y(x + 3h) 3y(x + 2h) + 3y(x + h) y(x)
y (x) = 3 3 y0 =
h h3
and so on.
In a similar vein, we obtain other formulae using Newtons backward differ-

39
ence interpolating formula (2.8) as

3p2 + 6p + 2 3 2p3 + 9p2 + 11p + 3 4

   
1 1 2
yn (x) = yn + p + yn + yn + yn +
h 2 6 12
6p2 + 18p + 11 4
 
1 2 3
yn (x) = 2 yn + (p + 1) yn + yn +
h 12
 
1 3 2p + 3 4
yn (x) = 3 yn + yn +
h 2
1
yn(iv) (x) = 4 [4 yn + ]
h
(3.9)

For tabulated value x = xn , p = 0 and the formulae listed in (3.9) simplify to

 
1 1 2 1 3 1 4
yn (x) = yn + y0 + yn + yn + (3.10)
h 2 3 4
 
1 2 3 11 4
yn (x) = 2 yn + yn + yn + (3.11)
h 12
 
1 3 3 4
yn (x) = 3 yn + yn + (3.12)
h 2

Similar finite difference approximations to the derivatives can be obtained from

the first terms of the above formulae as in the case of forward differences.
Example 14
From the table of values below, obtain the value of (i) y (x) and (ii) y (x) at the
point x = 1.2.

x 1.0 1.2 1.4 1.6 1.8 2.0 2.2

y 2.7183 3.3201 4.0552 4.9530 6.0496 7.3891 9.0250

Solution:
We construct the difference table for the tabulated values as follows

40
 x y 2 3 4 5 6
1.0 2.7183
1.2 3.3201 0.6018
1.4 4.0552 0.7351 0.1333
1.6 4.9530 0.8978 0.1627 0.0294
1.8 6.0496 1.0966 0.1988 0.0361 0.0067
2.0 7.3891 1.3395 0.2429 0.0441 0.0080 0.0013
2.2 9.0250 1.6359 0.2964 0.0535 0.0096 0.0014 0.0001

From the difference table, we obtain

1
0.7351 12 (0.1627) + 13 (0.0361) 14 (0.008) + 15 (0.0014)
 
(i) y (1.2) = 0.2
3.3203
1 11 10
 
(ii) y (1.2) = (0.2)2
0.1627 0.0361 + 12
(0.008) 12
(0.0014)) 3.3192.

3.3 Numerical Integration

Numerical integration refers to the evaluation of the integral
Z b
I= y(x)dx (3.13)
a

where y(x) is a function given analytically by a formula or empirically by a table

of values. Geometrically, the integral is the area under the graph of y(x) between
a and b. To determine this area, we divide the interval [a, b] into n subintervals
such that
a = x0 < x1 < x2 < < xn = b

where xn = x0 + nh. Thus the integral becomes

Z xn
I= y(x)dx.
x0

41
Approximating y(x) by Newtons forward difference interpolating formula leads
to the integral
Z xn  
p(p 1) 2 p(p 1)(p 2) 3
I= y0 + py0 + y0 + y0 + dx
x0 2! 3!
(3.14)
Since x = x0 + ph, dx = hdp and so the integral (3.14) becomes
Z n 
p(p 1) 2 p(p 1)(p 2) 3
I=h y0 + py0 + y0 + y0 + dp
0 2! 3!
(3.15)
which on simplification yields
Z xn
n(n 2)2 3
 
n n(2n 3) 2
y(x)dx = nh y0 + y0 + y0 + y0 +
x0 2 12 24
(3.16)
Equation (3.16) is a general formula from which one can obtain different integra-
tion formulae by putting n = 1, 2, 3, .
We derive from this general formula a few important numerical integration
formulae such as the trapezoidal and Simpsons rules.

3.3.1 The Trapezoidal Rule

Putting n = 1 into (3.16), all differences higher than the first become zero and
one obtains
x1  
1
Z
h
y(x)dx = h y0 + y0 = [y0 + y1 ]
x0 2 2
Similarly, we deduce for the remaining intervals [x1 , x2 ], , [xn1 , xn ] as
Z x2
h
y(x)dx = [y1 + y2 ]
x1 2
..
.

Z xn
h
y(x)dx = [yn1 + yn ]
xn1 2

42
Combining these integrals, we obtain the formula
Z xn
h
y(x)dx = [y0 + 2y1 + 2y2 + 2y3 + + yn ] (3.17)
x0 2

which is normally referred to as the trapezoidal rule. The geometrical interpreta-

tion of this rule is that the curve y = f (x) is replaced by n straight lines joining
the points (x0 , y0 ) and (x1 , y1 ),(x1 , y1 ) and (x2 , y2 ), ,(xn1 , yn1 ) and (xn , yn ).
The area bounded by the curve y = f (x), the ordinates x = x0 and x = xn
and the xaxis is then approximately equivalent to the sum of the areas of the n
trapeziums obtained [see fig.], hence the name of the formula.

3.3.2 Error bounds and estimate for the Trapezoidal Rule

The error of the trapezoidal formula can be estimated as follows: Let y = f (x)
be continuous and many times differentiable in [x0 , xn ]. Thus expanding f (x) in
Taylor series about x = x0 , we get
Z xn Z xn 
(x x0 )2


f (x)dx = y0 + (x x0 )y0 + y0 + dx
x0 x0 2!
h2 h3
= hy0 + y0 + y0 + (3.18)
2 6
In a similar vein,
h2 h3
 
h h
[y0 + y1 ] = y0 + y0 + hy0 + y0 + y0 +
2 2 2 6
2
h h3
= hy0 + y0 + y0 + (3.19)
2 4
From (3.18) and (3.19) we obtain
Z xn
h h3
f (x)dx [y0 + y1 ] y0 + (3.20)
x0 2 12

which is the error in the interval [x0 , x1 ]. Continuing in a similar way, we obtain
the errors in the remaining subintervals [x1 , x2 ], [x2 , x3 ], , [xn1 , xn ].

43
Summing all these, one obtains the error as
h3
[y + y1 + y2 + + yn1

] (3.21)
12 0
Taking y (x) as the largest value of the n quantities on the right hand side of
(3.21), we get
h3 (b a) 2
ny (x) h y (x) (3.22)
12 12
where nh = b a.
Error bounds can now be obtained by taking the largest value for y (x), say
MU and the smallest value ML , in the interval of integration, hence

KML KMU ,

where K = ba
12
h2 .

3.3.3 Simpsons Rule

To obtain Simpsons rule, otherwise called Simpsons 13 -rule, we need to divide the
interval of integration a x b into an even number of subintervals n = 2m.
Thus, Simpsons 13 -rule can easily be obtained from (3.16) by substituting n = 2.
This leads to the relation
Z x2  
1 2 h
y(x)dx = 2h y0 + y0 + y0 = [y0 + 4y1 + y2 ] (3.23)
x0 6 3
Similarly, for the remaining intervals [x2 , x4 ], [x4 , x6 ] , [xn2 , xn ], one obtains
Z x4
h
y(x)dx = [y2 + 4y3 + y4 ]
x2 3
Z x6
h
y(x)dx = [y4 + 4y5 + y6 ]
x4 3
..
.

Z xn
h
y(x)dx = [yn2 + 4yn1 + yn ].
xn2 3

44
Summing the above integrals, we obtain the composite Simpsons 13 -rule as
Z xn
h
y(x)dx = [y0 + 4y1 + 2y2 + 4y3 + 2y4 + + 2yn2 + 4yn1 + yn ]
x0 3
(3.24)
Simpsons 13 -rule has the error term

(b a) 4 (4)
h f (x)
180

and it is highly accurate for most problems.

3.3.4 Simpsons 38 -Rule

The scheme is obtained from (3.16) by substituting n = 3, thus yielding

Z x3  
3 3 2 1 3
y(x)dx = 3h y0 + y0 + y0 + y0
x0 2 4 8
(3.25)
3h
= [y0 + 3y1 + 3y2 + y3 ]
8
Computing the integrals for the remaining subintervals [x3 , x6 ], [x6 , x9 ], , [xn3 , xn ],
the composite Simpsons 38 -rule as
xn
3h
Z
y(x)dx = [y0 + 3y1 + 3y2 + 2y3 + 3y4 + 3y5 + 2y6 + + 2yn3 + 3yn2 + 3yn1 + yn ]
x0 8
(3.26)
Simpsons 83 -rule has the error term

3 5 (4)
h f (x)
80

and it is much less accurate than the Simpsons 13 -rule, hence it is rarely used for
computations.
Example 15
R1 dx
Evaluate I = 0 1+x2
, correct to 4 places of decimals.
Using the Trapezoidal, Simpsons 13 - and 38 -rules for h = 0.5, 0.25 and 0.125, we

45
have the following results:
(a) h = 0.5. The values are tabulated below
x 0 0.5 1.0
y 1.00000 0.80000 0.50000

(i) Trapezoidal rule yields

0.5
I= (1.00000 + 2 0.80000 + 0.50000) = 0.77500
2
(ii) Simpsons 13 -rule yields
0.5
I= (1.00000 + 4 0.80000 + 0.50000) = 0.78333
3
(b) h = 0.25. The values are tabulated thus
x 0 0.25 0.50 0.75 1.00
y 1.00000 0.94118 0.80000 0.64000 0.50000

(i) Trapezoidal rule yields

0.25
I= [1.00000 + 2(0.94118 + 0.80000 + 0.64000) + 0.50000] = 0.78280
2
(ii) Simpsons 13 -rule gives
0.25
I= [1.00000 + 4(0.94118 + 0.64000) + 2(0.80000) + 0.50000] = 0.78539
3
(c) h = 0.125. The tabulated data become
x 0 0.125 0.250 0.375 0.500 0.625 0.750 0.875 1.000
y 1.00000 0.98452 0.94118 0.87671 0.80000 0.71910 0.64000 0.56637 0.50000

(i) Using Trapezoidal rule, one obtains

0.125
I= [1.500+2(0.98452+0.94118+0.87671+0.8000+0.71910+0.6400+0.56637)] = 0.78474
2
(ii) Simpsons 13 -rule yields

46
 0.125
I= 3
[1.50000 + 4(0.98452 + 0.87671 + 0.71910 + 0.56637) + 2(0.94118 +
0.80000 + 0.64000)] = 0.78538
Simpsons 83 -rule can now be applied to obtain
0.375
(iii) I = 8
[1.50000+3(0.98452+0.80000+0.56637)+3(0.94118+0.71910)+
2(0.87671 + 0.64000)] = 0.77657.
Note that the exact solution correct to 5 places of decimals is I 0.78540.
Example 16
R1 2
Evaluate the integral I = 0
ex dx, using
(i) Trapezoidal rule with n = 5
(ii) Trapezoidal rule with n = 10
(iii) Simpsons 13 -rule with n = 10
(iv) Simpsons 38 -rule with n = 10.

Solution:

2
n xn x2n f (xn ) = exn
0 0 0 1.000000
1 0.2 0.04 0.960789
2 0.4 0.16 0.852144
3 0.6 0.36 0.697676
4 0.8 0.64 0.527292
5 1.0 1.00 0.367879
1.367879 3.037901

(i) Note that h = 0.2, thus

0.2
I= (1.367879 + 2 3.037901) = 0.744368.
2

47
 2
n xn x2n f (xn ) = exn
0 0 0 1.000000
1 0.1 0.01 0.99050
2 0.2 0.04 0.960789
3 0.3 0.09 0.913931
4 0.4 0.16 0.852144
5 0.5 0.25 0.778801
6 0.6 0.36 0.697676
7 0.7 0.49 0.612626
8 0.8 0,64 0.527292
9 0.9 0.81 0.444858
10 1.0 1.00 0.367879
1.367879 6.778167

(ii) Here, h = 0.1, thus

0.1
I= (1.367879 + 2 6.778167) = 0.746211.
2

(iii) For Simpsons 31 -rule,

0.1
I= (1.367879 + 4 3.740266 + 2 3.037901) = 0.746825.
3

(iv) Simpsons 83 -rule produces

0.3
I= (1.367879 + 3 2.45482 + 3 2.266882 + 2 2.056465) = 0.736722.
8

48