Chapter 2

Error Analysis Of Corrected

3 th
Simpson’s 8 Rule

2.1 Introduction

In the previous chapter we have listed number of quadrature rules used to evaluate the
1 rd
integral given in equation (1.0.1). Among these rules, Trapezoidal rule, Simpson’s 3

3 th
rule and Simpson’s 8 rule, which are Newton-Cote’s type, are most commonly used

quadrature rules. Newton-Cote’s type quadrature rules are studied and generalized

by many researchers. In [57], Ujević considered an averaged midpoint - Trapezoidal

rule of the form

b    
(b − a)
Z
a+b
f (x)dx = f (a) + 2f + f (b) .
a 4 2

The rule is obtained by using convex combination of midpoint rule and Trapezoidal

rule. Ujević and Erceg [55] constructed corrected averaged midpoint - Trapezoidal

rule of the form

b
(b − a)2 0
   
(b − a)
Z
a+b
f (x)dx = f (a) + 2f + f (b) + (f (b) − f 0 (a)),
a 4 2 48
2.1 Introduction 14

which involves first order derivative of function at end points. Generalization of this

rule is studied by Liu [90]. Some error inequalities are also given. In [79], Ramachan-

dran and Parimala derived some higher order open Newton-Cote type quadrature

rules by adding the derivative of function at midpoint to the basic open Newton-

Cote type quadrature rules. Degree of precision of this new family of rules is two

more than that of the basic quadratures. Zafar and others [25], constructed a family

of corrected open type Newton - Cote quadrature rules. These rules includes first

derivative of function : (i) at all grid points, (ii) only at interior points and (iii) only

at end points. Dehghan and others [44], presented a family of Newton - Cote type

quadrature rules. In [79], semi open quadrature rules are derived by excluding the

lower limit of integration. In [45], closed type rules are constructed. Quadrature

rules with respect to Chebyshev weight function √ 1 of semi-open type and closed
1−x2

type are given in [73] and [51] respectively. Burg [16] constructed some corrected

Newton-Cote quadrature rules of closed type that involve : (a) first derivatives at all

mesh points, (b) first derivatives only at end points and (c) higher order derivatives

only at end points. These rules have degree of precision respectively :(a) 2n + 2, (b)

n + 3 for odd values of n and n + 4 when n is even and (c) 2k + 2 where k is the number

of derivatives used in that rule. Burg and Degny [17], derived a family of corrected

mid point quadrature rules that includes value of function at mid point and value of

odd derivatives of function derived at end points.

Zhao and Hongxing [88], proposed corrected Trapezoidal rule, corrected Simpson’s
1 rd 3 th
3 as well as 8 rule and corrected Bool’s rule. These quadrature rules contains

certain ordered derivatives of function evaluated at mid point of the interval.

In [31, 32, 33] Euler Maclaurin’s formula [35] is used to correct the dual Simp-
3 th 3 th
son’s 8 rule, dual Euler-Simpson rule and Euler-Simpson’s 8 rule. For corrected
2.2 Error Inequality 15

quadratures various error bounds are obtained for different class of functions. In [58]

corrected Simpson’s formula of order six is derived. Further it is proved that the error

bounds obtained for this method are better than usual Peano-bounds. In [56] upper

and lower bounds for the error term in classical Simpson’s rule are obtained. These

error bounds are attained for function f (x) = x4 .

3 th
In this chapter optimal error bounds for corrected Simpson’s 8 rule are obtained.

To demonstrate the sharpness of error bounds, error bounds for function f (x) = x8
3 th
are evaluated. In section 2.2, we have derived corrected Simpson’s 8 formula and the

error in corrected quadrature rule by using the method of undetermined coefficients.

It is observed that error obtained here is much better than that obtained in [86]

and the references therein. Piecewise continuous functions are constructed to obtain

upper and lower bounds for the error term in the corrected quadrature. In section

2.3, we proved that the error bounds obtained in section 2.2 are optimal.

2.2 Error Inequality

3 th
The corrected Simpson’s 8 formula given in [86] is
Z 1     
31 81 1 2 19 0
f (x)dx = [f (0) + f (1)] + f +f + [f (0) − f 0 (1)]
0 224 224 3 3 3360
81 1 8 (8)
      
9 0 2 0 1
+ f −f − f (ξ), (2.2.1)
1120 3 3 4480 3

where f ∈ C 8 [0, 1].

Let
Z 1    
1 2
f (x)dx = a0 f (0) + a1 f + a2 f + a3 f (1)
0 3 3
   
1 2
+b0 f 0 (0) + b1 f 0 + b2 f 0 + b3 f 0 (1).
3 3
2.2 Error Inequality 16

Since there are eight parameters, the rule can be made exact for polynomials of

degree less than or equal to seven. By evaluating the system of equations obtained

for f (x) = xi , i = 0, 1, 2, ..., 7, we get

31 81 19 9
a0 = a3 = , a1 = a2 = , b0 = b3 = and b1 = b2 = .
224 224 3360 1120

3 th
Thus the corrected Simpson’s 8 rule derived by using method of undetermined

coefficients turns out to be

Z 1     
31 81 1 2 19 0
f (x)dx = [f (0) + f (1)] + f +f + [f (0) − f 0 (1)]
0 224 224 3 3 3360
f (8) (ξ)
    
9 0 2 0 1
+ f −f +C , (2.2.2)
1120 3 3 8!

where the error constant C [48] is

"   8 #
1
1 8
Z
31 8 8 8 81 2 19
C = x dx − [0 + 1 ] − + − [8 × 07 − 8 × 17 ]
0 224 224 3 3 3360
"  7  7 #
9 2 1
− 8× −8×
1120 3 3
1
= .
17010

The quadrature rule obtained by approximating function by Hermite interpolation

3 th
coincides with the corrected Simpson’s 8 formula given in equation (2.2.2). Define

error E as
Z 1
E= f (x)dx − Qc , (2.2.3)
0

where
    
31 81 1 2 19 0
Q c
= [f (0) + f (1)] + f +f + [f (0) − f 0 (1)]
224 224 3 3 3360
    
9 0 2 0 1
+ f −f .
1120 3 3

The error expression in (2.2.2) is sharper than the error term given in equation (2.2.1).
3 th
Since the error term in corrected Simpson’s 8 rule is of the form M f (8) (ξ), where M
2.2 Error Inequality 17

is constant, consider
Z 1 Z 1 Z 2 Z 1
(8) 3 (8) 3 (8)
P8 (x)f (x)dx = P8 (x)f (x)dx + P8 (x)f (x)dx + P8 (x)f (8) (x)dx.
0 0 1 2
3 3

On integrating this term by parts recursively we get,

Z 1 h i1 h i2 h i1
3 3
P8 (x)f (8) (x)dx = P8 (x) f (7) (x) + P8 (x) f (7) (x) 1 + P8 (x) f (7) (x) 2
0 0 3 3
h i1 h i2 h i1
3 3
− P7 (x) f (6) (x) − P7 (x) f (6) (x) 1 − P7 (x) f (6) (x) 2
0 3 3
h i1 h i2 h i1
3 3
+ P6 (x) f (x) + P6 (x) f (x) 1 + P6 (x) f (5) (x) 2
(5) (5)
0 3 3
h i1 h i2 h i1
3 3
− P5 (x) f (4) (x) − P5 (x) f (4) (x) 1 − P5 (x) f (4) (x) 2
0 3 3
h i1 h i2 h i1
3 3
+ P4 (x) f (3) (x) + P4 (x) f (3) (x) 1 + P4 (x) f (3) (x) 2
0 3 3
h i1 h i2 h i1
3 3
− P3 (x) f (2) (x) − P3 (x) f (2) (x) 1 − P3 (x) f (2) (x) 2
0 3 3
h i1 h i2 h i1
3 3
+ P2 (x) f (1) (x) + P2 (x) f (1) (x) 1 + P2 (x) f (1) (x) 2
0 3 3
1 2
Z 1
−[P1 (x) f (x)]03 − [P1 (x) f (x)] 31 − [P1 (x) f (x)]12 + f (x)dx,
3 3 0

(2.2.4)

where Pi (x) is a polynomial of degree i, such that

0
Pi+1 (x) = Pi (x); i = 1, 2, 3, ..., 7. (2.2.5)

On comparing the equations (2.2.3) and (2.2.4), we get

   
31 1 2 81
P1 (0) = P1 (1) = , P1 = P1 = ,
224 3 3 224
   
−19 1 2 9
P2 (0) = P2 (1) = , P2 = −P2 = ,
3360 3 3 1120
   
1 2
Pi (0) = Pi (1) = Pi = Pi = 0; for i = 3, 4, 5, 6, 7, 8.
3 3

Thus the polynomials that satisfy above conditions and equation (2.2.5) are as follows:

31 1
P1 (x) = x − , 0≤x≤ ;
224 3
2.2 Error Inequality 18

1 1 2
= x− , <x≤ ;
2 3 3
193 2
= x− , < x ≤ 1.
224 3

x2 31x 19 1
P2 (x) = − + , 0≤x≤ ;
2 224 3360 3
x2 x 397 1 2
= − + , <x≤ ;
2 2 3360 3 3
x2 193x 617 2
= − + , < x ≤ 1.
2 224 1680 3

 
x 2 93x 19 1
P3 (x) = x − + , 0≤x≤ ;
3! 224 560 3
x3 x2 397x 39 1 2
= − + − , <x≤ ;
3! 4  3360 2240  3 3
(x − 1) 2 355x 693 2
= x − + , < x ≤ 1.
3! 224 1120 3

x2 2 31x
 
19 1
P4 (x) = x − + , 0≤x≤ ;
4! 56 280 3
x4 x3 397x2 39x 1 1 2
= − + − + , <x≤ ;
4! 12  6720 2240 560 3 3
(x − 1)2 2 81x 18 2
= x − + , < x ≤ 1.
4! 56 35 3

x3 2 155x
 
19 1
P5 (x) = x − + , 0≤x≤ ;
5! 224 168 3
x5 x 4 397x 3 39x 2 x 11 1 2
= − + − + − , <x≤ ;
5! 48  20160 4480 560 80640 3 3
(x − 1)3 2 293x 283 2
= x − + , < x ≤ 1.
5! 224 672 3
x4 2 93x
 
19 1
P6 (x) = x − + , 0≤x≤ ;
6! 112 112 3
x 6 x 5 397x 4 13x 3 x 2 11x 1 1 2
= − + − + − + , <x≤ ;
6! 240 80640 4480 1120 80640 120960 3 3
(x − 1)4 2 131x 19 2
= x − + , < x ≤ 1.
6! 112 56 3
2.2 Error Inequality 19

x5 2 31x 19
 
1
P7 (x) = x − + , 0≤x≤ ;
7! 32 80 3
x 7 x 6 397x 5 13x 4 x 3 11x2 x 1 1 2
= − + − + − + − , <x≤ ;
7! 1440 403200 17920 3360 161280 120960 2419200 3 3
(x − 1)5 2 33x
 
43 2
= x − + , < x ≤ 1.
7! 32 160 3

x6 2 31x 19
 
1
P8 (x) = x − + , 0≤x≤ ;
8! 28 60 3
x8 x7 397x6 13x5 x4
= − + − +
8! 10080 2419200 89600 13440
11x3 x2 x 1 1 2
− + − + , <x≤ ;
483840 241920 2419200 57153600 3 3
(x − 1)6 2 25x

22 2
= x − + , < x ≤ 1.
8! 28 105 3

Thus equation (2.2.4) becomes

Z 1 Z 1
(8)
P8 (x)f (x)dx = f (x)dx − Qc . (2.2.6)
0 0

Observe that
Z 1
27
P8 (x)dx = (2.2.7)
0 10! × 5103

and
37
max P8 (x) = . (2.2.8)
x[0,1] 10! × 5103

To construct upper and lower bounds for error function we define two real numbers,

γ8 = min f (8) (x) and Γ8 = max f (8) (x). (2.2.9)

x[0,1] x[0,1]

Then γ8 ≤ f (8) (x) ≤ Γ8 for x  [0, 1].

From equations (2.2.6) and (2.2.7) we have

Z 1h i Z 1 Z 1
(8) (8)
f (x) − γ8 P8 (x)dx = f (x)P8 (x)dx − γ8 P8 (x)dx
0 0 0
Z 1
27γ8
= f (x)dx − Qc − . (2.2.10)
0 10! × 5103
2.2 Error Inequality 20

Simillarly
Z 1h i Z 1 Z 1
Γ8 − f (8) (x) P8 (x)dx = Γ8 P8 (x)dx − f (8) (x)P8 (x)dx
0 0
Z 1 0 
27Γ8 c
= − f (x)dx − Q . (2.2.11)
10! × 5103 0

Observe that
Z 1h i Z 1
f (8) (x) − γ8 P8 (x)dx ≤ max |P8 (x)| |f (8) (x) − γ8 |dx (2.2.12)
0 x[0,1] 0

and
Z 1h i Z 1
(8)
Γ8 − f (x) P8 (x)dx ≤ max |P8 (x)| |Γ8 − f (8) (x)|dx. (2.2.13)
0 x[0,1] 0

Since γ8 ≤ f (8) (x) for all x  [0, 1], |f (8) (x) − γ8 | = f (8) (x) − γ8

and from equations (2.2.10), (2.2.12) and (2.2.8) we get

Z 1 Z 1h
27γ8 c
i
f (x)dx − Q − = f (8) (x) − γ8 P8 (x)dx
0 10! × 5103 0
Z 1
37
≤ |f (8) (x) − γ8 |dx
10! × 5103 0
37 h i
≤ f (7) (1) − f (7) (0) − γ8
10! × 5103
37
≤ [S7 − γ8 ] ,
10! × 5103
where S7 = f (7) (1) − f (7) (0). (2.2.14)

Thus we have
Z 1
1
f (x)dx − Qc ≤ [37S7 − 10γ8 ] . (2.2.15)
0 10! × 5103

Simillarly, from equations (2.2.11), (2.2.13) and (2.2.8) we have

Z 1
1
[37S7 − 10Γ8 ] ≤ f (x)dx − Qc . (2.2.16)
10! × 5103 0

Equations (2.2.15) and (2.2.16) together with equation (2.2.3) gives

1 1
[37S7 − 10Γ8 ] ≤ E ≤ [37S7 − 10γ8 ] , (2.2.17)
10! × 5103 10! × 5103

where Γ8 , γ8 and S7 are defined in the equations (2.2.9) and (2.2.14).

2.3 Optimal Error Bounds 21

2.3 Optimal Error Bounds

This section is devoted to prove the sharpness of error bounds. Both upper and lower

bounds listed in equation (2.2.17) are calculated for f (x) = x8 .

For the function f (x) = x8 ; γ8 = Γ8 = 8! and S7 = f (7) (1) − f (7) (0) = 8!.

1 1
Lower bound = [37S7 − 10Γ8 ] = , (2.3.1)
10! × 5103 17010

1 1
Upper bound = [37S7 − 10γ8 ] = (2.3.2)
10! × 5103 17010

and
"   8 #
1 1
1 8
Z Z
c 8 31 8 81 2
f (x)dx − Q = x dx − [0 + 18 ] − +
0 0 224 224
3 3
"    7 #
19 7 7 9 2 7 1
− [8 × 0 − 8 × 1 ] − 8 −8
3360 1120 3 3
1
= . (2.3.3)
17010

Thus from equations (2.3.1) and (2.3.2) we observe that for f (x) = x8 , upper and
3 th
lower bound coincides with the error term E in corrected Simpson’s 8 formula.

From above example it is seen that, the bounds given in equation (2.2.17) are optimal.
2.4 Conclusions 22

2.4 Conclusions
3 th
The corrected Simpson’s 8 formula given by equation (2.2.1) and equation (2.2.2) is
3 th
of order 7. The error E in corrected Simpson’s 8 formula is obtained by constructing

polynomial functions. Error bounds for corrected quadrature rule Qc are obtained.

Optimality of error bounds is demonstrated by constructing upper and lower bounds

for function f (x) = x8 .

3 th
To evaluate integral of function f (x), Simpsons 8 formula is corrected by includ-

ing the values of derivatives of f (x) at mesh points. The quadrature rule is derived by

using the method of undetermined coefficients. The approximate error in the integra-
1 f (8) (ξ)
tion formula is 17010 8! . The double integral inequalities are used to obtain better

error bounds. By suitable example it is observed that the error bounds obtained by

double integral inequalities are the best possible error bounds.