Professional Documents
Culture Documents
New Error Bounds For The Simpson's Quadrature Rule and Applications
New Error Bounds For The Simpson's Quadrature Rule and Applications
www.elsevier.com/locate/camwa
Received 17 June 2005; received in revised form 26 October 2006; accepted 12 December 2006
Abstract
New error bounds for the well-known Simpson’s quadrature rule are derived. If we use these bounds then we can apply the
Simpson’s rule to functions whose first, second or third derivatives are unbounded below or above. Furthermore, these error bounds
can be (much) better than some recently obtained bounds. Applications in numerical integration are also given.
c 2007 Elsevier Ltd. All rights reserved.
Keywords: Simpson’s rule; New error bounds; Improvements; Applications; Numerical integration
1. Introduction
In recent years a number of authors have written about Simpson’s inequality. For example, the mentioned inequality
is considered in [1–9]. Simpson’s inequality gives an error bound for the well-known Simpson’s quadrature rule:
Z b
b−a a+b
f (t)dt = f (a) + 4 f + f (b) + E( f ), (1)
a 6 2
where a, b ∈ R, a < b. Here we suppose that f, f 0 , f 00 : [a, b] → R are absolutely continuous functions.
There are few known ways to express the term E( f ). The different variants of E( f ) give different estimations of
the error. The main aim of this paper is to point out some new estimations of E( f ). Similar type of inequalities are
given in [10] where a modified Simpson’s rule is considered.
Furthermore, we show that these new estimations can be (much) better than the following estimations (obtained
in [7]).
If f : [a, b] → R is such that f (n−1) is an absolutely continuous function and
γn ≤ f (n) (t) ≤ Γn , (a.e.) on [a, b] ,
for some real constants γn and Γn , then for n = 1, 2, 3, we have
Z b
b−a a+b
(t)dt (a) (b) ≤ Cn (Γn − γn )(b − a)n+1 ,
f − f + 4 f + f (2)
a 6 2
where
5 1 1
C1 = , C2 = , C3 = . (3)
72 162 1152
Applications of the above mentioned results in numerical integration are also given.
2. Main results
Theorem 1. Let f : [a, b] → R be an absolutely continuous function with γ1 ≤ f 0 (t), (a.e.) on [a, b], for some
constant γ1 ∈ R. Then
Z b
b−a a+b 1
(t)dt (a) (b) ≤ (S − γ1 )(b − a)2 ,
f − f + 4 f + f (4)
a 6 2 3
where S = f (b)−
b−a
f (a)
. If there exists a constant Γ1 ∈ R such that f 0 (t) ≤ Γ1 , (a.e.) on [a, b] then
Z b
b−a a+b 1
f (t)dt − f (a) + 4 f + f (b) ≤ (Γ1 − S)(b − a)2 .
(5)
a 6 2 3
Proof. We define
5a + b a+b
t −
, t ∈ a,
6 2
S1 (t) = (6)
a + 5b a+b
, ,b .
t −
t∈
6 2
Integrating by parts, we obtain
Z b Z b
b−a a+b
S1 (t) f (t)dt =
0
f (a) + 4 f + f (b) − f (t)dt. (7)
a 6 2 a
Let C be a constant. Then we have
Z b Z b
S1 (t) f (t)dt =
0
S1 (t) f 0 (t) − C dt,
(8)
a a
Rb
since a S1 (t)dt = 0.
From (8) it follows
Z b Z b
S1 (t) f (t)dt =
0
S1 (t) f (t) − C dt
0
a a
Z b
≤ max |S1 (t)| f (t) − C dt.
0
(9)
t∈[a,b] a
If we choose C = γ1 in (9) then we have
Z b
f (t) − γ1 dt = f (b) − f (a) − (b − a)γ1
0
a
f (b) − f (a)
= − γ1 (b − a)
b−a
= (S − γ1 )(b − a), (10)
since f 0 (t)
≥ γ1 , t ∈ [a, b].
We also have
b−a
max |S1 (t)| = . (11)
t∈[a,b] 3
From (7)–(11) we easily get (4).
66 N. Ujević / Computers and Mathematics with Applications 53 (2007) 64–72
(see Section 1) with the inequalities (4) or (5). For that purpose, we can use the monomials 1, t, t 2 , . . . or t λk , λk ≥ 0.
These monomials play a very important role in approximation theory ([11, pp. 139–144]). For example, they have
dense linear span in the space C [a,
b]. If f
is continuously differentiable on [a, b] and ε > 0 then there is a polynomial
p such that k f − pk∞ < ε and
f 0 − p 0
∞ < ε ([11, p. 37]). Besides, it is easy to work with them.
Further, it is known that E( f ) = E( f + t k ) (E is defined by (1)), for k = 0, 1, 2, 3, since the Simpson’s rule is
exact for t k , k = 0, 1, 2, 3. Thus, we also choose k > 3.
Theorem 5. Let f : [a, b] → R be a function such that f 0 is absolutely continuous with γ2 ≤ f 00 (t) (a.e.) on [a, b]
for some constant γ2 ∈ R. Then
Z b
b−a a+b 1
f (t)dt − f (a) + 4 f + f (b) ≤ (S − γ2 )(b − a)3 ,
(17)
a 6 2 24
f 0 (b)− f 0 (a)
where S = b−a . If there exists a constant Γ2 ∈ R such that f 00 (t) ≤ Γ2 , (a.e.) on [a, b] then
b
Z
b−a a+b 1
f (t)dt − f (a) + 4 f + f (b) ≤ (Γ2 − S)(b − a)3 .
(18)
a 6 2 24
Proof. We define
1 2a + b a+b
(t − a) t − , t ∈ a,
2 3 2
S2 (t) = (19)
1 a + 2b a + b
(t − b) t − , t∈ ,b .
2 3 2
Integrating by parts, we obtain
Z b Z b
00
S2 (t) f (t)dt = − S1 (t) f 0 (t)dt
a a
b−a
a+b
Z b
=− f (a) + 4 f + f (b) + f (t)dt, (20)
6 2 a
Remark 7. We have
Z b Z b
S2 (t)dt = t S2 (t)dt = 0.
a a
Thus,
b (b − a)3
Z
b−a a+b
f (t)dt − f (a) + 4 f + f (b) ≤ G 1 ( f 00 ) ,
a 6 2 162
where G 1 ( f ) is defined by (16). Note also that, in this case, p0∗ = Γ2 +γ
2 .
2
Theorem 8. Let f : [a, b] → R be a function such that f 00 is absolutely continuous with γ3 ≤ f 000 (t) (a.e.) on [a, b].
Then
Z b
b−a a+b 1
f (t)dt − f (a) + 4 f + f (b) ≤ (S − γ3 )(b − a)4 ,
(28)
a 6 2 324
Proof. We define
1 a+b a+b
(t − a) t − , t ∈ a,
2
6 2 2
S3 (t) = (30)
1 a + b a + b
(t − b)2 t − , t∈ ,b .
6 2 2
Integrating by parts, we have
Z b Z b Z b
00
S3 (t) f 000 (t)dt = − S2 (t) f (t)dt = S1 (t) f 0 (t)dt
a a a
b−a
a+b
Z b
= f (a) + 4 f + f (b) − f (t)dt. (31)
6 2 a
(b − a)3
max |S3 (t)| = . (35)
t∈[a,b] 324
From (31)–(35) we easily get (28).
In a similar way we get (29).
3. Applications
Theorem 10. Let f : [a, b] → R be a function such that f (n) is absolutely continuous with γn ≤ f (n) (t) ≤ Γn , (a.e.)
on [a, b], for some constants γn , Γn ∈ R, n ∈ {1, 2, 3}. Then
Z b
f (t)dt = T (π, f ) + Rn (π, f ),
a
where π represents the subdivision {a = x0 < x1 < · · · < xm = b} of the interval [a, b], h i = xi+1 − xi and
1 m−1
X
xi + xi+1
T (π, f ) = h i f (xi ) + 4 f + f (xi+1 ) .
6 i=0 2
The error Rn (π, f ) satisfies
m−1
X
|Rn (π, f )| ≤ Cn (Γn − γn ) h in+1 , (39)
i=0
5 1 1
where C1 = 72 , C2 = 162 , C3 = 1152 .
Proof. Apply (13), (25) and (36) to the intervals xi , xi+1 and sum. Then the triangle inequality gives the proof.
Theorem 11. Let f : [a, b] → R be a function such that f (n) is absolutely continuous with γn ≤ f (n) (t), (a.e.) on
[a, b], for some constant γn ∈ R, n ∈ {1, 2, 3}. Then
Z b
f (t)dt = T (π, f ) + Q n (π, f ) (40)
a
and
m−1
X
|Q n (π, f )| ≤ K n (Sni − γn )h in+1 , (41)
i=0
(xi+1 )− f(n−1)
(xi ) (n−1)
where K 1 = 31 , K 2 = 24 1 1
, K 3 = 324 , Sni = f hi , i = 0, 1, . . . , m − 1 and π represents the subdivision
{a = x0 < x1 < · · · < xm = b} of the interval [a, b], h i = xi+1 − xi . If there exist a constant Γn ∈ R such that
f (n) (t) ≤ Γn , (a.e.) on [a, b], then
Z b
f (t)dt = T (π, f ) + Pn (π, f ), (42)
a
where
m−1
X
|Pn (π, f )| ≤ K n (Γn − Sni )h in+1 . (43)
i=0
xi+1
Z
hi xi + xi+1
f (t)dt − f (xi ) + 4 f + f (xi+1 ) ≤ K n (Sni − γn )h in+1 ,
(44)
xi 6 2
for i = 0, 1, 2, . . . , m − 1, n = 1, 2, 3.
On the other hand, we have
Z xi+1 Z xi+1
hi xi + xi+1 h i
f (t)dt − f (xi ) + 4 f + f (xi+1 ) = Sn (t) f (n) (t) − γn dt,
xi 6 2 xi
for i = 0, 1, 2, . . . , m − 1, n = 1, 2, 3.
If we now sum the above relation over i from 0 to m − 1 then we get (40). We use the triangle inequality and (44)
to get (41).
In a similar way we can get (42) and (43).
Remark 12. If we choose h i = h = (b − a)/m then the inequalities (41) and (43) become
(b − a)n+1
|Q n (π, f )| ≤ K n (S − γn ) ,
mn
(b − a)n+1
|Pn (π, f )| ≤ K n (Γn − S) ,
mn
where S = ( f (b) − f (a))/(b − a), n = 1, 2, 3.
Remark 13. In practice the estimations (41) and (43) can be (much) better than (39), since in most cases we have
m−1
X m−1
X
(Sni − γn )h in+1 ≤ (Γn − γn )h in+1 ,
i=o i=o
m−1
X m−1
X
(Γn − Sni )h in+1 ≤ (Γn − γn )h in+1 .
i=o i=o
We now show that the above remark is valid.
Example 14. Let us choose f (t) = t k k > 3, a = 0, b > 0, h i = h = b−a m , i = 0, 1, . . . , m − 1. Then we have
f 0 (t) = kt k−1 and γ1 = 0, Γ1 = kbk−1 . From (39), for n = 1, we get the right-hand side:
5k k+1
R.H.S. (39) = b . (45)
72m
From (41), for n = 1, we get the right-hand side:
1 k+1
R.H.S. (41) = b . (46)
3m
From (45) and (46) we see that the estimation (39) is better than the estimation (41) if 3 < k < 24
5 . If k > 24
5 then
(41) is better than (39). In fact, if k 24
5 then (41) is much better than (39).
Finally, we consider one “real” example, i.e. we consider the function f (t) = exp(t 2 − 16) which cannot be
integrated analytically. We choose [a, b] = [0, 4] and, for the sake of simplicity, we choose h i = h = (b − a)/m.
From the inequality (39) we get
5
|R1 (π, f )| ≤ (Γ1 − γ1 )(b − a)2 . (47)
72m
From the inequality (41) we get
1 f (b) − f (a) (b − a)2
|Q 1 (π, f )| ≤ − mγ1
3 h m2
(b − a)2 f (b) − f (a)
= − γ1 . (48)
3m b−a
72 N. Ujević / Computers and Mathematics with Applications 53 (2007) 64–72
Example 15. Let f (t) = exp(t 2 − 16) such that f 0 (t) = 2t exp(t 2 − 16) and γ1 = 0, Γ2 = 8 on [a, b] = [0, 4],
f (0) = exp(−16) ≈ 0, f (4) = 1. Then from (47) it follows
80
|R1 (π, f )| ≤ (49)
9m
and from (48) it follows that
4
|Q 1 (π, f )| ≤ .
3m
Suppose that we have to calculate the integral
Z 4
exp(t 2 − 16)dt (50)
0
with the tolerance of error ε = 10−3 . Then we easily find that it has to be m > 8888, if we use (47), and m > 1333, if
we use (48). (The last results are obtained from (49) and (50).) We see that we have to apply the elementary Simpson’s
rule (at least) 8889 times if we use the first estimation and we have to apply the same rule (at least) 1334 times if we
use the second estimation. From (49) and (50) we also see that for any m we get, at least, 6 times better results if we
use the second inequality.
References
[1] P. Cerone, Three points rules in numerical integration, Nonlinear Anal. TMA 47 (4) (2001) 2341–2352.
[2] D. Cruz-Uribe, C.J. Neugebauer, Sharp error bounds for the trapezoidal rule and Simpson’s rule, J. Inequal. Pure Appl. Math. 3 (4) (2002)
1–22. Article 49.
[3] S.S. Dragomir, R.P. Agarwal, P. Cerone, On Simpson’s inequality and applications, J. Inequal. Appl. 5 (2000) 533–579.
[4] S.S. Dragomir, P. Cerone, J. Roumeliotis, A new generalization of Ostrowski’s integral inequality for mappings whose derivatives are bounded
and applications in numerical integration and for special means, Appl. Math. Lett. 13 (2000) 19–25.
[5] S.S. Dragomir, J. Pečarić, S. Wang, The unified treatment of trapezoid, Simpson and Ostrowski type inequalities for monotonic mappings and
applications, Math. Comput. Modelling 31 (2000) 61–70.
[6] I. Fedotov, S.S. Dragomir, An inequality of Ostrowski type and its applications for Simpson’s rule and special means, Math. Inequal. Appl. 2
(4) (1999) 491–499.
[7] M. Matić, Improvement of some inequalities of Euler–Grüss type, Comput. Math. Appl. 46 (2003) 1325–1336.
[8] C.E.M. Pearce, J. Pečarić, N. Ujević, S. Varošanec, Generalizations of some inequalities of Ostrowski–Grüss type, Math. Inequal. Appl. 3 (1)
(2000) 25–34.
[9] N. Ujević, Sharp inequalities of Simpson type and Ostrowski type, Comput. Math. Appl. 48 (2004) 145–151.
[10] N. Ujević, A.J. Roberts, A corrected quadrature formula and applications, ANZIAM J. 45(E) (2004) E41–E56.
[11] N.L. Carothers, Approximation Theory, Bowling Green State University, Ohio, 1998.