Photocopyingpermittedby license only the GordonandBreachScience Publishersimprint.
Printed inSingapore.
On Simpson’s Inequality and Applications
S.S. DRAGOMIR*, R.P.
AGARWALt
andP.CERONESchool of CommunicationsInformatics, Victoria Universityof Technology, RO. Box14428,McMelbourne City, 8001Victoria, Australia
(Received10March1999; Revised 26 July1999)
Newinequalities of Simpson type andtheirapplicationtoquadrature formulaeinNumer- icalAnalysis are given.
Keywords: Simpson’s inequality; Quadrature formulae
1991 MathematicsSubjectClassification: Primary26D15,26D20;
Secondary 41A55, 41A99
1.
INTRODUCTION
The following inequality is well known in the literature as Simpson’s inequality:
fab f (x)
dx----.b
a[f (a) + f (b) + 2f (a. + 2 .b)
where the mapping
f: [a, b]
I is assumed to be four times contin- uouslydifferentiableontheinterval(a, b)
and for the fourth derivative* Corresponding author.E-mail:sever@matilda.vu.edu.au.
URL:http://matilda.vu.edu.au/~rgrnia/dragomirweb.html.
Current address: Department of Mathematics, National University of Singapore, 10KentRidgeCrescent,Singapore 119260. E-mail: matravip@leonis.nus.sg.
URL: http://matilda.vu.edu.au/~rgrnia.
533
tobeboundedon
(a, b),
thatisItf
(4)1[o
:=xE(a,b)supIf (4)(x)[ <
cx3.Now,
ifwe assume thatIn"
a x0<
X1<’’"<
Xn-1<
Xn b is aparti- tion of the interval[a, b]
andf
isasabove, then we have the classical Simpson’squadratureformula:
bf(x)
dxAs(f In) + Rs(f In), (1.2)
where
As(f, In)
isthe Simpson rule"-1
2f(xi +Xi+l)h
As(f, In) =:-[f(xi) + f(xi+l)]hi+-
/=0 2
i=0
(1.3)
and theremainderterm
Rs(f, In)
satisfiestheestimaten--1
IRs(f, In)[ <
2880
Ilf(4) [1 Z
i=0h, (1.4)
where
hi
:-- Xi+l Xifor 0,...,n 1.Whenwehaveanequidistantpartitioningof[a,
b]
givenbyIn
xi := a/"b-a i, i=O,...,n;then we have the formula:
bf (x)
dxAS,n(f) + Rs,n(f), (1.6)
where
-[ (
b-a)(
b-ab-a
f
a+.i/f a+.(i+l)
As,n(f)
:= 6ni=0 n n
b-an "2i+2 1) (1.7)
andthe remainder satisfies the estimation
(b-a)
5IRs,,(f)l < 2880"
n4I[f(4)l[" (1.8) For
some otherintegralinequalities seetherecentbook[1]
and the papers[2-4]
and[5-37].
The main purpose of this survey paper is to point out some very recentdevelopmentsonSimpson’s inequality forwhichthe remainderis expressedintermsof lower derivatives than thefourth.
Itiswell known that ifthe mapping
f
is neitherfourtimesdifferen-tiablenoristhefourth
derivativef
4)bounded on(a, b),
thenwe cannotapplythe classicalSimpsonquadratureformula,which, actually,is one of themostusedquadratureformulaeinpractical applications.
The first section ofour paper deals with an upper bound for the remainder in Simpson’s inequality for the class of functions of bounded variation.
Thesecond sectionprovidessomeestimates fortheremainderwhen
f
is a Lipschitzian mapping while the third section is concerned with the sameproblemfor absolutelycontinuousmappingswhose derivatives areintheLebesgue spacesLp[a, b].
The fourthsectionisdevotedtotheapplicationofacelebrated result dueto Griiss toestimatethe remainder inthe Simpson quadrature rule intermsof the supremum and infimum of thefirstderivative. Thefifth sectiondealswith ageneralconvexcombination oftrapezoidand inte- riorpointquadratureformulafrom which,inparticular,wecanobtain theclassicalSimpsonrule.
The last section containssomeresults relatedto Simpson,trapezoid andmidpointformulae for monotonicmappingsandsomeapplications forprobabilitydistribution functions.
Last,
butnotleast,wewouldliketomentionthateverysectioncon- tainsaspecial subsection in which the theoretical resultsareappliedfor the specialmeans oftwo positivenumbers: identric mean,logarithmic mean,p-logarithmicmean etc.and providesimprovementsand related resultstothe classical sequence of inequalitiesH<G<L<I<A, where
H, G, L,
IandAaredefined in thesequel.2. SlMPSON’SINEQUALITY
FOR MAPPINGS
OF BOUNDED VARIATION2.1. Simpson’sInequality Thefollowingresult holds
[2].
THEOREM Let
f:
[a,b] -
beamappingof
boundedvariation on[a, b].
Thenwehavethe inequality:
fab f (x) dx----.
b-alf(a)+f(b) ( +b)l
2 /2f a
2< (b- a)V(f),
b(2.1)
where
/6a(f)
denotes the total variationoff
on the interval [a,b]. Theconstant
1/2
isthe best possible.Proof
Using the integrationbyparts formulafor Reimann-Stieltjes integralwehavefab s(x) df(x)
b-a3If(a)+f(b
2+ 2f (a-k-b)] 2 fab f(x)
dx,(2.2)
where
Indeed,
x- x E a,
s(x)
6 2a+5b
Ia+bb]
x 6 xE 2
b
s(x)
5a+b df(x) +
x-6 +6)/2
a+
65b) df(x)
5a+
f (x)] + I (x a+ f (x)]
b--[(x___b)
6 a(a+b)/2 6 (a+6)/=f(x)dx
b a
[f (a) + f (b) + 2f (a +,b) ] ja
"6----"
2 2f(x)
dxandtheidentityisproved.
Now,
assume thatA,’a--x")< xln)< < ,_
"()< x
n) =b is asequence of divisions with
v(A.)-O
as nc, wherev(An):=
(n) (n) (n) (n) (n)
maxge{0
n_l,(X/+
--X)andi
EIx ,Xi+l].Ifp:[a,b] ,scon-
tinuouson
[a, b]
andv[a, b] --
Ii isof 15oundedvariationon[a,b],
thenbp(x)
dv(x)
n-1
,4zx)o/=o
plim((n)) [v(xl))
vn-1
lim
Xi+l)
(x)o .=
n-1
max
[p(x)[
supx[a,b] A .=
b
max
[p(x)[ V(v).
xE[a,b] a
(2.3)
Applyingthe inequality
(2.3)
forp(x)s(x)
andv(x) =f(x)
weget bs(x)df(x)
b
<
xE[a,b]maxIs(x) V(f). (2.4)
a
Taking into account the fact that the mapping s is monotonic non- decreasingontheintervals
[a, (a + b)/2)
and[(a + b)/2, b]
ands(a)
b-a 6s( a + O) =l(b-a), s(a ,+ 2 b) l(b-a)
and
s(b)=
ba6wededuce that
max
Is(x)[ l(b- a)
xe[a,b]
-
Now,
using the inequality(2.4)
and the identity(2.2)
we deduce the desired result(2.1).
Now,
for the bestconstant.Assume
thatthefollowinginequality holdsb b-a
f (x)
dx--- f (a) + f (b) + 2f (
a+
2< C(b a) V(f)
bawith a constantC
>
0.Letuschoose themapping
f: [a, b]
given bya+b)
ifxE a,
f(x)
2-1 if x a+b
----.
tA
( a +
,b1,
Thenwehave
bf (x)
dxIf(a) +f(b)2 + 2f(ab)l
4(b-a)
and
b
(b- a) V (f) 4(b a).
a
Now,
using the above inequality, we get4C(b-a)> ](b-a)
whichimpliesthatC
> 1/2
andthen1/2
isthebest possibleconstant in(2.1).
Itis naturalto considerthe followingcorollarywhichfollows from identity
(2.2).
COROLLARY Suppose that
f:
[a,b] ll is adifferentiable
mappingwhosederivativeiscontinuous on
(a, b)
andb
IIf’lla If’(x)l
dx<
Thenwehave the inequality:
IL
Thef (x)
following corollarydx b-a--- [f(a)+f(b)
for2 Simpson’s composite+ 2f
2<
formula5 IIf’ll, (b_a)2
holds:(2.5)
COROLLARY 2 Letf: [a, b]
--+IRbea mappingof
boundedvariationon[a,
b]
andIh
apartitionof[a, b].
Thenwehave the Simpson’squadratureformula (1.2)
and the remaindertermRs(f, Ih) satisfies
theestimate:b
Is(f, Ih)l _< (h)V(f), (2.6)
a
where
7(h)
:--max{hl
0,...,n}.
The case of equidistant partitioning is embodied in the following corollary:
COROLLARY 3 Let
In
beanequidistantpartitioningof[a, b]
andf
beasinTheorem 1. Then wehave the
formula (1.6)
and the remaindersatisfies
theestimate:
b
(b- a) V(f) (2.7)
IRs,.(f) -<
a
Remark 1 If we want to approximate the integral
fabf(x)dx
bySimpson’s formula
As, n(f)
with an accuracy less thans>
0, weneedatleast
n
EN pointsfor the divisionIn,
wherene .(b-a) (f) +1
and
[r]
denotes the integer part ofrEIR.Comments Ifthe mapping
f: [a, b]
--+IRisneitherfourtimesdifferenti- ablenorthe fourth derivative isboundedon(a, b),
then wecannotapply the classical estimation inSimpson’sformulausingthefourthderivative.But
ifweassumethatfis
ofbounded variation, thenwe canuse instead the formula(2.6).
We givehereaclass ofmappingswhichareof bounded variation but whichhave the fourth derivative unboundedonthegiveninterval.
Letfp :[a,
b],fp(X)
:--(x a)
pwhere pE(3, 4).
Then obviouslyfA (x)
:=p(x a) p-l,
x(a, b)
and
fp(4) (X) =p(p- 1)(p- 2)(p- 3)
(x- a)
4-p x(a,b).
Itisclear
thatfp
isof bounded variation andb
V(f) (b a)
p<
o,a
but limx_+
fp(4)(x)
2.2. Applications for Special
Means Let
usrecallthefollowingmeans:(1)
Thearithmetic meanA A(a,b):= a+b
a,b>0;(2)
Thegeometric meanG
G(a, b):= v/,
a,b_>
O;(3)
Theharmonic mean HH(a, b)
21/a + 1/b’
a,b>0;(4)
The logarithmicmean LL(a,b)
"= b-alnb- lna a,b>O,
ab;
(5)
Theidentric meanl(bb)
1I-
I(a,b)
:---e
Y
a,b>O,aCb;
(6)
The p-logarithmicmean[
bp+l -ap+lll/P
Lp Lp(a,b)’: C(f li( a)_]
pII\.[-1,O},
a,b> O, a:/:
b.Itis wellknown that
Lp
ismonotonicnondecreasingoverpEIRwith L_ :=LandL0
:=LIn particular,wehave thefollowinginequalitiesH<_G<L<I<A.
Using Theorem 1, some new inequalities arederived for the above means.
1.
Letf: [a,
b] R(0 <
a< b),f(x)
x’,
p]R\{- 1,0}.
Thenlab
b a
f(x)
dxLp(a, b),
f (a) + f (b)
A(aP bp),
2f(a,,,b) :AP(a,b)
and
IIf’[l --[pl(b a)Lfl,
p/R\{-1,0, 1).
Usingthe inequality
(2.5)
wegetL (a, b) - A(a p,
b1 2-
Ap(a, b) < I-l!LpP5(b a) 2.
2.Let
f: [a, b]
--*N(0 <
a< b),f(x) 1Ix.
Thenlab
b a
f(x)
dx L-1(a, b), f f (a) (a + +2 f (b) b)
HA-l - (a’ b) (a, b),
and
IIf’ll G2(a, b)
bnaUsing the inequality
(2.5)
wegetI3AH-
AL-2HL[ <
623.Let
f: [a, b]
I(0 <
a< b), f (x) In
x.Thenf.
bb a
f(x)
dxIn I(a, b), f (a) +f (b) In G(a, b),
f (a +"b)
and
Ilf’l[ =L(a,b)"
b-aUsing the inequality
(2.5)
we obtainIn G1/UA2/3.
(b-a)
2-<
3L3.
SIMPSON’S
INEQUALITYFORLIPSCHITZIAN MAPPINGS
3.1. Simpson’sInequalityThe following result holds
[3]"
THEOREM 2 Let
f: [a,b]--
be an L-Lipschitzian mapping on[a,b].
Thenwehave the inquality:
f (x)
dx---
2+ 2f
2< L(b a) (3.1) .
Proof
Using the integration byparts formula for Riemann-Stieltjes integralwehave(see
also theproofof Theorem1)
thatb b a
[f(a) +f(b)
s(x) df(x)
3 2
+ 2f(a+b)
2l fa bf(x)dx’
(3.2)
where
x 6 x a,
s(x)
:=a/5b__ [a+b b]"
x 6
x
2
v(n)
X(n
n)NOW, assume that
An’a= xn) < xln) < <
n-1<
b is asequence of divisions with
u(A,)
0 as n c, whereu(An)
:=[’.(n) .(n)’ ,..,.,,.1 c(n)
Iv(n)
v(n)].
If p[a, b]
-*IR maxis{0 n-i}Ai+I
-xi ) axt qi |’iis Riemann infegrable ori
[a, b]
and v:[ab]
--+ i L-Lipschitzian on[a,
b],thenp(x) dv(x)
lim
n-,
((}n))l(xl_) xl
n)Vt,,Xi..t._I) (.X’In))
<L
b
L
Ip(x)[
dx.(3.3)
Applyingthe inequality
(3.3)
forp(x)=s(x)
andv(x)--f(x)
wegetdf(x)
b
<
LIs(x)l
dx.(3.4)
Letuscompute
fab Is(x)
dxf(a+b)/915aq-b
va x 6I(5a+b)/6(Sa-Fb
.a 6
f(a+5b)/6(a-+-Sb
+
J(a+b)/2 65(b- a)2.
36
a+
6
5b[
dxx)
dxq_f(a+b)/2(
l(5a+b)/6 X5a+b)
6 dXx)
dx-F(ib+5b)/6 (
x a+
65b.
dxNow,
using the inequality(3.4)
and the identity(3.2)
we deduce the desired result(3.1).
COgOLLAg 4
Suppose
thatf: [a,b]
is adifferentiable
mappingwhose derivativeis continuouson
(a, b).
Thenwehave the inequality:fab
The followingf (x)
dx b-a---
corollary[f(a)+f(b)+2f(a+b)l
for Simpson’s composite formula holds:2 2-< IIf’lloo(b (3.5) a)
2 COROLLARY 5 Letf: [a, b]
/RbeanL-Lipschitzianmapping on [a,b]
and
Ih
a partitionof [a,b].
Then we have the Simpson’s quadratureformula (1.2)
and theremaindertermRs(f, Ih) satisfies
theestimation:n-1
[Rs(f, Ih)l <
Lh. (3.6)
i=o
The case of equidistant partitioning is embodied in the following corollary:
COROLLARY 6 Let
In
beanequidistantpartitioningof[a, b]
andf
beasin Theorem 2. Thenwehave the
formula (1.6)
andthe remaindersatisfies
theestimation:
5 L
(b a)
2[Rs,n(f)[ <_
--
b(3.7)
Remark 2 If we want to approximate the integral
fa f(x)
dx bySimpson’s formula
As, n(f)
with an accuracy less that e>
0, we needatleast
n
EN pointsfor the divisionIn,
wherens:= .-(b-a)
/1and
[r]
denotes the integer part ofrE.
Comments If the mapping
f:
[a,b]
Iisneither four timedifferen- tiable nor the fourth derivative is bounded on(a,
b), then we cannot apply the classical estimation in Simpson’s formula using the fourth derivative. But if we assume thatf
is Lipschitzian, then we can use insteadtheformula(3.6).
We
give hereaclassofmappingswhichareLipschitzian but having the fourth derivative unboundedonthegiveninterval.Letfp [a, b] IR, fp(x)
:=(x a)
pwhere p(3, 4).
Then obviouslyfA(x)
:=p(x a) p-l,
x(a, b)
and
fp(4) (X) =p(p- 1)(p- 2)(p- 3)
(x- a)
4-px (a,b).
Itisclear
thatf
isLipschitzianwiththeconstantL p(b a)
p-1<
o, butlimx--,a+fp(4)(x) --oo.
3.2. Applicationsfor SpecialMeans
Using Theorem2,we nowpointoutsome newinequalities for the special means defined in the previous section.
1.Let
f:
[a,b] IR(0 <
a< b),f(x)=
xp,
pEN\{-1, 0}.
ThenIlf’l[
o6p(a,b)
:={pb- [plaP-
ifif pp>
E(-o, 1)\{-1, 0).
1,Using the inequality
(3.5)
weget[L(a,b) -1/2A(aP,
bp) -AP(a,b)[ <_ Sp(a,b)(b a).
2.Let
f: [a, b]
R(0 <
a< b), f (x) 1Ix.
ThenUsing theinequality
(3.5)
weget5 b-a
I3HA
L.4 2LH< 1-
--g-a LAH"3.
Let f:
[a,b] (0 <
a< b),f(x) In
x.ThenUsing the inequality
(3.5)
wegetIn
IG/3A2/3
4.
SIMPSON’S
INEQUALITYIN TERMS OFTHE p-NORM
4.1. Simpson’sInequality The following resultholds
[4]:
THEOREM 3 Let
f: [a, b]--+ IR
bean absolutely continuousmapping on [a,b]
whose derivativebelongstoLp[a, b].
Thenwehave the inequality:f (x)
dx---
2+ 2f "2
[2
q+l+ 11
1/q< [3(q + ’1) (b a)l+l/qllf’llp, (4.1)
where(1/p)
+
1/q 1,p>
1.Proof
Using the integrationby parts formula for absolutelycontin- uousmappings,wehave"b b a
[.f(a) +f(b)
s(x) f’ (x)
dx 2+ 2f (a. +
2b) 1 fa
bf(x)dx’
(4.2)
where
5a+b [ a+b)
x 6 x a, 2
S(X)
:=a+5b [a+bb]"
x 6
x
2 Indeed
abs(x)f’(x)dx
/.(a+b)/2(
,Aa
5a+b f,(x) dx+
x-6 +b)/2
a
5b) f’(x)
dxbf(x)
dxb a
[f (a) +f (b) + 2f (a +
b)
3 2 2
a+5b)
6f(x) I
b(a+6)/zfbf(x)dx,
and the identity is proved.
ApplyingH61der’s integral inequalityweobtain
6
s(x) f’ (x)
dx<_ (fb Is(x)l
qdx)l/q II/’llp. (4.3)
Letuscompute
b
Is(x)l
qdx(a+b)/21
x 5a+
ba 6
----[(5a+b)/6( "5a-l-b,a
6-lqdx+f(f [+b)/2
X a+ 5b[q
x)
qdx -t-[(a+b)/2(x
1(5a+b)/65a+b)
6 qdx)q f(( a5b)
qx dx
+
x- dx+56)/6 6
)
q+l (5a+b)/6(
5a+ b )q+l
(a+b)/2-x
+
x,a 6 (5a+b)/6
(a+5b)/6(ant-5b) ]--1-
X q+lb(a+b)/2 6 (a+5b)/6
q+l (a+5b)/2
(a +
5b+
d(a+b)/2 6
q/l 6
(a +
65bx)
q+l1
[(5ab )q+ (_+. 5a+
q+
--a.a+b
2 6
(.a-k-5b Iq+l ( a+Sbl q+l]
+
6 a+b2+
b 6(2
q+lif-1)(b a)
q+l3(q + 1)6q
Now,
using the inequaltiy(4.3)
and the identity(4.2)
we deduce the desiredresult(4.1).
The following corollary forSimpson’scomposite formula holds:
COROLLARY7 Let
f
andIn
beasabove. Thewehave Simpson’s rule(1.2)
and theremainderRs(f, Ih) satisfies
theestimate:[2q+l+lql/q (
Igs(f, Ih)l <
- [3(//+ i)J Ilf’llp
\i=0hi
+q(4.4)
Proof
Apply Theorem 3 on the interval [xi, xi+l] (i=0,...,n-1)
to obtaini
x‘+’f(x) dx--hi [.f(xi)+f(xi+l)2
dXi
,rq+,/ll
,iq l+l/q(i
xi+’ip )
lip< g [3(q+ 1)] IIf’lleh, If’(t)
dt\dxi
Summingthe above inequalities over from0 to n-1, usingthe gen- eralizedtriangleinequality and H61der’s discreteinequality,weget
IRs(/, S)I
n-1
ix,+, hi [f (xi)
nt-f (Xi+l
<_ f(x)
dx-
2i=0 xi
,pq/’/,}’/q-’
l+l/q(i
xi+’< /3(q + 1) hi If’(t)l"
dti=0 \,sxi
1
[2q+l _l_|]l/q[ ]l/q
x
(i
x’+’If’(t)l’dt
\i=0 \,xxi
lrq+’/l]’iq ( )
liq=g 13(q + 1) IIf’ll, hi
+q\i=0
and thecorollaryisproved.
The case of equidistant partitioning is embodied in the following corollary:
COROLL,R 8 Let
f
beasaboveandif In
&anequidistantpartitioningof[a, b],
thenwehavetheestimate:[2
q+l+ 1]
1/qIRs’(f)l <-n L3( + 1) (b- a)+l/qllf’l[
p.Remark 3 If we want to approximate the integral
ff(x)dx
bySimpson’s formula
As, n(f)
with an accuracy less that e>
0, weneed atleastn
ENpoints for thedivisionIn,
where(2
q+l-+- 1)1/q
n
:=\- _ (b a)l+l/qllf’[[p] +
and
[r]
denotes the integer part ofr E.
Comments If the mapping
f: [a, b]
Iis neitherfourtimedifferenti- ablenorthe fourth derivativeisboundedon(a, b),
thenwe cannotapply the classical estimationinSimpson’sformulausingthe fourth derivative.Butifwe assume that
fPE Lp(a, b),
thenwe canuse the formula(4.4)
instead.
We give hereaclass of mappings whose first derivatives belong to
Lp(a,b)
but having the fourth derivatives unbounded on the given interval.Letf: [a, b] I,f(x)
:=(x a)
wheres(3, 4).
Then obviouslyfs(X)
:’-s(x- a) s-l,
x E(a, b)
and
fs(4) (X) S(S- 1)(s 2)(s- 3) (x_a)
4-s Itisclearthatlimx-a+ fs
(4)(X)
-[-O,butx
(a,b).
(b a)S-+(I/p) [[fsl[p
s.((s-1)p + 1)
1/p<
c.4.2. Applications for Special
Means
(See
Section2.2 forthedefinition of themeans.)
1.Let
f [a, b] --- I (O <
a< b), f (x) x,
s EN\ {-1, O}.
Thenf(x)
dxLSs(a,b),
b-a
f( a+2 b) As(a’ b),
f (a) + f (b) A(aS,
bs)
and
I[ftl[p -IslZ(_p
s-1(b a)
lipUsingthe inequality
(4.1)
wegetIL(a, b) 1/2 A(a s,
bs) AS(a, -< L3(q +
11/q
s-1
(a,b)(b- a),
IslZ(_p
where(1/p)/1/q 1, p
>
1.2.Let
f:
[a,b]
11(0 <
a< b), f (x) 1Ix.
Thenf
bb a
f(x)
dx Z-1(a, b),
f (a + b) A-l (a’
f (a) + f (b)
H
- (a, b)
2 and
IIf’llp L-p(a,b)(b a) 1/p.
Using the inequality
(4.1)
weget[2q+ + 11
1/qI3HA
LA2LH[ < -AHL[3(q + 1)1 L-p(b a) 1/p.
3.
Let f:
[a,b]
N(0 <
a< b),f(x) In
x.Thenlab
b a
f(x)
dxIn I(a, b),
f (a +
2b)
lnA(a, b)
f (a) + f (b) In A(a, b)
and
IIf’l[p L--lp(a,b)(
ba) 1/p.
Using the inequality
(4.1)
we obtain< g 1)jl L3(q- [2q+l + 1] 1/qL__lp(a, b)(b a).
5.
GRUSS
INEQUALITYFORTHE SIMPSON FORMULA
5.1. SomePreliminary Results
The following integral inequality which establishesaconnection between the integral of theproduct oftwofunctionsand theproductof theinte- grals of the two functions is well known in the literature as Griiss’
inequality[5,p.
296]:
THEOREMq
<_f(x) <_
4g9and’7
Let f, g’[a,<_
g(x)<_ F b] -- for
all xbe twoE(a, b);
integrableq,,
’7functions
andFareconstants.such thatThenwehavethe inequality:
Ib l_a fabf(x)g(x)
dx-b af(x)
dx.b a
g(x)
dx<_ 1/4 v),
and the inequality is sharp in the sense that theconstant
1/4
cannot be replaced byasmallerone.In
1938, Ostrowski(cf.,
for example.J1, p.468])
proved the fol- lowing inequality which gives an approximation of the integral1/(b a) fab f(t)
dtasfollows:THEOREM 5 Let
f: [a,b]--.
be adifferentiable
mapping on(a,b)
whose derivative
f’ :(a,b)--.
is bounded on(a,b),
i.e.,[[f’[l:=
supt(a,b)l f’(t) dt[ <
cxz. Thenfabf(t)
dt<I+
f(x)
b a(x (a + b)/2) 2]
i (b a)1[ f’ [l,
for
all xE(a, b).
In
the recentpaper[6],
Dragomir andWang
proved thefollowing versionof Ostrowski’s inequalitybyusing theGriissinequality(5.1).
THEOREM 6 Let
f:
IC_ -, beadifferentiable
mapping intheinteriorof
Iand let a, bint(I)
witha<
b.If f’
Ll[a,b]
and" <_f’(x) < r
for
all x[a,
b],thenwehavethefollowing inequality:fa
bf(b)-f(a), (a+b.)
xf (x)
b af (t)
dt b a 2_< 1/4 (b a)(r -), (5.3)
for
all x[a, b].
Theyalsoappliedthisresult forspecialmeansand inNumericalInte- gration obtaining somequadratureformulaegeneralizingthemid-point
quadratureruleandthetrapezoidrule.
Note
that theerrorboundsthey obtained are intermsof the first derivativewhich areparticularlyuseful in thecasewhenf"
doesnotexistorisvery largeatsome points in[a, b].Forotherrelated resultsseethepapers
[7-37].
In this section ofour paper we give a generalization of the above inequalitywhichcontainsas aparticularcasethe classical Simpson
for-
mula. Applicationforspecial meansandin NumericalIntegration are alsogive.
5.2. AnIntegralInequalityof Gr(issType
For
any real numbersA, B,
letusconsiderthe function[21]
t-a+A
ifa<t<x,p(t) px(t)
t-b+B ifx<t<_b.
Itisclear thatPxhas thefollowing properties.
(a)
Ithas the jump[P]x (B- A) (b a)
atpoint xand
dpx(t)
dt
1+ [P]x 6(t- x).
(b)
LetMx
:=supte(a,b)Px(t)
andmx
:=infte(a,b)Px(t).
Then the differ- enceMx -mx
canbeevaluatedasfollows"(1)
For B- A<_ O,
wehaveMx mx -[P]x.
(2)
For B- A>
0,the following three casesarepossible (i) If 0<
B A< 1/2(b a),
then-x+b
x--a
fora
_<
x_<
a+ (B- A);
fora
+ (B- A) <
x_<
b(B- A);
for b-
(B-A) <
x_<
b.(ii)
If1/2(a b) <
B- A_< (b a),
then-x+b
Mx mx
B- Axma
for a
_<
x<
b-(B-A);
for b-
(B-A) _<
x< a+ (B- A);
for
q+ (B-A) <_
x<_
b.(iii) IfB-A
>
b a, thenMx mx [P]x"
The following inequality of Ostrowski type holds
[21]
THEOREM 7 Let
f:
[a,b]
Nbeadifferentiable
mapping on(a, b)
whosederivative
satisfies
theassumption"7
< f’(t) <
Pfor
all E(a,b), (5.4)
where"7,Fare givenreal numbers. Thenwehave the inequality."
(C A)f(a) + (b
a B+ A)f(x) + (B C)f(b) fabf(t)dt]
<_ 1/4 (F "7)(Mx mx)(b a), (5.5)
where
Cx 2(b a)[(x a)(x
a+ 2A) (x b)(x
b+ 2B)],
and
A, B, Mx
andmx
areasabove,x[a, b].
Proof
Using the Grfiss inequality(5.1),
we canstatethatf (b) f (a)
px(t)f’(t)dt-
b-a b-a
<_ 1/4 (I’ "7)(Mx mx),
px(t)dt
b-a(5.6)
for all x
(a, b).
Integrating the firsttermbypartsweobtain:
ja
px(t)f’(t)
dtBf(b) Af(a) f(t)
dt+ [p]xf(X).
Also,as
(5.7)
bpx(t)dt
1/2 [(x a)(x
a+ 2A) (x b)(x
b+ 2B)],
then
(5.6)
gives theinequality:Bf(b) Af(a) f(t)
dt+ [p] f(x) Cx .f(b)
b-a x b-a
< 1/4 (r 7)(Mx mx),
whichisclearlyequivalent with the desired result
(5.5).
Remark 4 Setting in
(5.5),
A B 0 and takinginto account,bythe property(b),
thatMx-mx=b-
a, we obtain the inequality(5.3)
by Dragomir andWang.
The followingcorollaryisinteresting:
COROLLARY 9 Let
A,
Bbe real numbersso that0<_
BA <_ (b a)/2.
Iff
isasabove,thenwehave the inequality:2
Af(a)/[b-a (B-A)]f(a- b)
/B-2 Af(b) fabf(t)dt
<_ 1/4 (F "y)(b
a B+ A)(b a). (5.8)
Proof
Considerx(a
/b)/2.
Then,from(5.5)
xua x-b
and
c= A+B
2 xE[a + (B- A),
b(B- A)].
By
property(b)
wehaveMx mx (b a) (B- A).
ApplyingTheorem 7 for x
(a + b)/2,
weget easily(5.8).
Remark 5 Ifwechooseinthe abovecorollary B- A
(b a)/2,
thenweget
(a) + f (b)
+f
a+
b(b-a)-
dt<
-
2 2
(5.9)
which isthe arithmetic mean of themid-pointandtrapezoidformulae.
Remark 6 inequality:
Ifwe choose in
(5.8)
B=A,
then we get the mid-point(b-a)f( a+b)2 fbf(t)dtl <
l(5.10)
discoveredby Dragomirand
Wang
in thepaper[6] (see
Corollary2).
Remark 7 Ifwechoose in
(5.8)
B- A(b- a)/3,
thenweobtainthe celebrated Simpson’s formula:[f(a)+4f(a- b) +f(b) 1 --fabf(t)
dt< -1 (, 7)(b a) :,
(5.11)
forwhichwehaveanestimation intermsof the first derivativenot asin theclassicalcase inwhichthefourth derivativeisrequiredasfollows:
b-a f(a) + 4f(a +
2b) +f(b)]- bf(t)dtl < llf(4)[[o
2880(b- a)
5(5.12)
The method ofevaluation of theerrorfor the Simpsonruleconsidered abovecanbeappliedforany quadratureformula ofNewton-Cotestype.
For example, to get the analogousevaluation of the error for the Newton-Cotesrule oforder 3it issufficienttoreplacethefunctionpx(t) in
(2.3)
bythe functiont-a-A
px(t)
:=a+b
2 t-b-B
ifa
< <
a+h;A+B
ifa+h< <
b-h;2
ifb-h
< <
b;whereB A
(b a)/4,
h(b a)/3.
5.3. Applications forSpecialMeans
(See
Section2.2for the definition of themeans.)
1. Consider themappingf(x)
xP(p > 1),
x>
0. Then F "),(a b)(p 1)LpP_
for a, bE with0
<
a<
b.Consequently,wehave theinequality:AP(a,b) +-A(aP,
bp) L(a,b) _<l(b-a)(p-1)LpP_.
2. Consider themappingf(x)
1/x,
x>
0. Thenb a2
(b a)A(a, b)
/=
ab---
T- 2.G4(a, b)
for 0
<
a<
b.Consequentlywehavethe inequality:A- (a, b) +
-
H(a,b)-L-(a,b) <- (b_ a)2 A(a,b)
- G4(a,b)
which isequivalentto
+
AL AH(b a)
9A2HL
<-
-
G43. Consider themappingf(x)
In
x, x>
0. Thenwehavefor a, bE11 with 0
<
a<
b. Consequently,we have theinequality:lnA
+lnG-
lnI 1(b a)
2
<-
6 G2 whichisequivalentto
(b a)
2-6 G2
5.4. Estimation ofErrorBounds inSimpson’s Rule Thefollowingtheorem holds.
THEOREM 8 Let
f: [a, b]
beadifferentiable
mapping(a, b)
whosederivative
satisfies
thecondition",/<_f’(t) <_
I"for
all(a,b);
where%I"are givenrealnumbers. Thenwehave
bf (t)
dtSn(In,f) + Rn(In,f),
where
n--1
Sn(In,f
- Z
i=ohi[ f (xi) -+- 4f (xi -+- hi) -+- f (xi+l )], (5.14)
Ih
isthepartition givenbyIn:
axo <
X1< <
Xn-1<
Xn bhi
:=1/2(Xi+l
xi), 0,...,n and the remaindertermRn(I,,f satisfies
theestimation:
2 n-I
(I 7) Z h/ (5.15)
[R,(I,,f)[ <
-
i=oProof
Letus setin(5.11)
a xi, b Xi+l,
2hi
Xi+l xi and xi+ hi 1/2 (xi-+-
Xi+l),
where 0,...,n 1.
Thenwehave the estimation:
hi [f (xi) -+- 4f (xi
nt-hi) -+- f (xi+l )] f
Xxi+f (t)
dtfor all i=O, ,n 1.
Aftersummingand using the triangleinequality,weobtain
hi lab f(
i=0
" [f(xi) "+" 4f(xi "+" hi) + f(xi+l)] t)
dt2 n-1
(r "7) Z hi’
<-5
i=0whichproves the required estimation.
COROLLARY 10 Under the aboveassumptions and
if
weputIIf’llo
:--supt(a,b)
lf’(t)[ <
0, then we have the following estimationof
theremainderterminSimpson’s
formula
4 n--1
IR.(I,f)l <_ IIf’llo h/. (5.16)
i=0
The classical error estimates based on the Taylor expansion for Simpson’s ruleinvolvethe fourth derivative
]]f(4)[]o.
Inthe casethatf
(4)doesnotexistorisverylargeatsome pointsin
[a, b],
the classicalesti- mates cannotbe applied, and thus(5.15)
and(5.16)
providealternative error estimates for the Simpson’s rule.6.
A CONVEX COMBINATION
ThefollowinggeneralizationofOstrowski’s inequality holds
[19]:
THEOREM 9 Let
f: [a,b]
R be absolutely continuous on[a,b],
andwhose derivative
f"[a,b]--R
is bounded on [a,b]. DenoteIIf’llo’--
ess
suptta,b]lf’(x)l <
o. Thenf(t) at f(x). (1 ) + f(a) +f(b). (b a)
2
< [(b-a)[+(-l)]+(x-a+b)]llf’ll
2(6.1)
for
all6E[0, 1]
anda+ (b a)/2 <
x<
b 6.(b a)/2.
Proof
Letusdefinethemappingp"[a, b]
2 R given byp(x,t)
:=t-[ b-6"b-2 a]’ E(x,b].
Integratingbyparts,wehave
p(x,t)f’(t)dt
(f(a) + f(b))
6.
(b-a)
2
dt
+ j’xb(t lb
,5 b a ]f’(t)
dt/
(1 6).f(x) f(t)
dt.(6.2)
Onthe other hand
bp(x,t)f’(t)dt
<_ [p(x, t)[ [f’(t)l
dt_< [[f’llo fa Ip(x, t)l
dtdt
+
b
at]
Now,
let usobserve thatfpr It ql
dtfpq (q t)dt
/q.r (t q)
dt(q_ r+p)
2[(q p)2 + (r- q)2] (p r)2 +
2for all r, p, q such that p
<
q<
r.Using the previousidentity,wehave that
l(x_a) 2+ (a+6.
dt= a+x) 22
and
b dt=
l(b_x) / (b-.
x+ )2 b)
Thenweget
1 (x-a)+(b-x)
+
b-ax-a+-
L=..
2 2 2 2(b-a)----2. e+( 1) +
x4 2
x 6
"b-a)
and the theoremisthusproved.
Remark 8
(a)
Ifwechoose in(6.1),
6 0,weget Ostrowski’s inequality.(b)
Ifwe choose in(6.1),
6 and x(a + b)/2
we get the trapezoid inequality:fa
bf(t)dt- f (a) +
2f (b) (b a) <--1 (b a)2llf’[I
o.(6.3)
COROLLARY 11
fa < f
Under the above(t) [(b-a)+(x
dt- 1[ f (x)
assumptions, we+
have theinequality:f (a)
- f (b).] (b a)
a
+ b
2
for
all xE[(b + 3a)/4, (a + 3b)/4],
and,inparticular, thefollowingmixtureof
the trapezoid inequality and mid-point inequality:b
f t)
dt- f (a + + f (a) + f (b-a) (b a)2llf (6.4)
Finally, we also have the following generalization of Simpson’s inequality:
COROLLARY 12 Underthe aboveassumptions,wehave
b
f (t)
dt- If(a) + 4f(x) + f (b)](b a)
-< I----(b-a)2+ (x-a+b)2]
for
all x[(b + 5a)/4, (a + 5b)/4],
and, in particular, the Simpson’s inequality:_1 If(a)+4f(
ab)+f(b)l(b-a
f(t
dt- _< (b a)2llf’[[. (6.5)
6.1. Applicationsin NumericalIntegration
Thefollowing approximationofthe integral
fabf(x)
dx holds[19].
THEOREM 10 Let
f: [a, b]
Rbeanabsolutelycontinuousmapping on [a,b]
whose derivative is bounded on[a, b]. If I
:a Xo< x
<...<
Xn-
<
Xn b is apartitionof
[a,b]
andhi:
Xi+l Xi, 0,...,n 1, thenwehavebf (x)
dxAa(In, , 6,f) + Re(In, , 6,f), (6.6)
where
n-1 n-1
A6(In,,5,f) (1 ) -f(i)hi + SZ f(xi) +
2f(xi+l)
i=0 i=0
hi,
(6.7)
6E[0, 1],xi
+
5.hi < i <
xi+" hi
0,...,n 1;and theremain-derterm
satisfies
theestimation:IR6(In,,6,f)l
< IIf’lloo
62+ (6- 1)
2Z h2i + Z i-
xiq-Xi+l (6.8)
i=o i=o 2
Proof
Applying Theorem 9 on the interval [Xi,Xi+l] i=0,... ,n- wegethi[(1 6)"f({i) + f (xi) +f (xi+,
fXi+l
6
f(x)
dxXi
_< [52-+ (6-I) 2] --+ @i
xi"ql-Xi+’)2 ]lf/I]
for all E
[0, 1]
and(i [xi,xi+], 0,...,n 1.Summingover from 0to n- and using thetriangleinequalitywe get theestimation
(6.8).
Remark 9
(a)
Ifwechoose6 0,thenweget thequadratureformula:bf(x)dx
AT(In, (,f) + RT(In, ,f), (6.9)
whereA
r(In, ,f)
isthe Riemann sum, i.e.,n-1
AT(In, ,f)
:=Z f (i)hi,
i=0
{i
[xi, xi+l],
i=O,...,n-
1;and theremainder term satisfiestheestimate
(see
also[8]):
IRT(I’’f)l < llf’lloo i.... + - (6.10)
(b)
Ifwechoose6 1,thenweget thetrapezoidformula"bf (x)
dxAT(In,f) + RT(In,f) (6.11)
whereA
r(In,f)
isthetrapezoidalrulen-1
AT(In,f) Z f(xi) +
2f(xi+l) hi
i=o
and the remaindertermssatisfiestheestimation:
n-1
[R(I=,f)l < [If’[[
-- h.
4 i=0
(6.12)
COROLLARY 13 Under the aboveassumptions wehave
bf (x)
dxBr(In, C,f)+ Qr(In, (,f), (6.13)
where
1
f((i)hi + y f(xi) +f(xi+l). hi Br(In, (,f)
ki=0 i=o 2
Xi+l
+ 3xi
xiq-3Xi+l.l
i
E 4 4and theremainderterm
satisfies
theestimation:n-1
(
IQr(ln,,/)l < II/’11 ’h
/i-
i=0 i=0
xi
+ x+)
d ) ] (6.14)
Inparticular,wehave
’bf(x)dx
BT(In,f) + QT(In,f), (6.15)
where
f
Xint-Xi+l
hi + Z f(xi) +f(xi+l). hi
Br(In,f)
li=0 2 i=0 2
and
QT(In,f) satisfies
theestimation"n-1
IQ(l,f)[ < Ilf’ll h/ (6 16)
i=0
Finally,wehave the followinggeneralizationof Simpson’s inequality whose remaindertermisestimatedbytheuseof the first derivativeonly.
COROLLARY 14 Under the aboveassumptions wehave
bf(x)dx
ST(In, ,f) + Wr(In, ,f), (6.17)
where
2 n-I n-1
ST(In, (,f)
- y f(i)hi
nt-Z [f(xi) + f(xi+l)]hi,
i=0 i=0
Xi+l
+
5Xi Xi+ 5Xi+l 7
6 6
j
and the remainderterm
Wr(In, ,f) satisfies
the bound:Iwr(I,,f)l <_ IIf’ll _h "- Z i-xi
i+1.i=0 i=0
(6.18)
and,inparticular, the Simpson’s rule:
bf (x)
dxS’(In,f) + Wr(In,f), (6.19)
where
1 ( ) ln-1
2
f X +
xi+lhi + E [f(xi) +f(xi+l)]hi
ST(In,f) =
i=0 2 i=0and the remainderterm
satisfies
theestimation:n-1
Iw(I,f)l <_ IIf’ll
’
i=0h. (6.20)
6.2. Applicationsfor SpecialMeans
Now,
letusreconsidertheinequality(6.1)
in the following equivalent form:f (a) + f (b) (1 6).f(x) +
2
"
b-afabf(t)dt
(x- (a + 6)/2) 2]
] IIS’il
for all 6E
[0, 1]
and xE[a, b]
such thata+6"(b-a) <2
x<_b-5. 21. Consider the mapping
f: (0, cx)
--*(0, oe), f(x)
xp,
pe \ {-
1,0}.
Then,for 0
<
a<
b,wehave fIplbP-1
IIf’ll
’
o t
lPlaP-i
ifp> 1,
if p
(-o, 1]\{-1,0},
andthen, by
(6.21),
wededucethat1(1 6).
xp/6.A(aP,
bp) Zff(a,b)[
< (b a) + )2
4
+ (b-a--- 5p(a,b),
where
( Iplb
p-1Sp(a,b)
:=[plaP-
ifp> 1,
if pE
(-c, 1]\{-1, 0}
and 6E
[0,
1],x[a
/6.(b a)/2,
b 6.(b a)/2].
2. Consider the mapping
f: (0, o) (0, o), f(x), 1Ix
and 0<
a
<
b.Wehave:and thenby
(6.21),
we deduce, for all 6[0, 1],
and a+ . (b a)/2 <
x
<_
b. (b a)/2
that](1-6)6L+Lx6-x6 <_ (b-a)
: + (6-
1):] (x-
4
+
b E-- ) ]
3. Consider the mapping
f: (0, cx) , f(x) In
x and 0<
a<
b.Wehave
1 andthen, by
(6.21),
wededucethat<- (b-a) + )2 (x..A)21
-a
4+
(b a) .]’
for all
[0, 1],
and xe [a +
6.(b a)/2,
b 5.(b a)/2].
7. A
GENERALIZATION FOR MONOTONIC MAPPINGS
In [20],
Dragomir established the followingOstrowski typeinequality formonotonicmappings.THEOREM 11 Let
f: [a, b]
]beamonotonicnondecreasingmapping on[a, hi.
Thenfor
all xE[a, b],
wehavetheinequality:f (x)
b a
f (x)
dx[2x (a + b)]f(x) + sgn(t x)f(t)dt
<-b-a
[(x a)(f(x) -f(a)) + (b x)(f(b) -f(x))]
Ix (a + b)/2[ (f(b) -f(a)).
< +
b-aAllthe inequalitiesaresharpand theconstant
1/2
isthe best possibleone.Inthissection weshallobtainageneralizationof this result which also contains thetrapezoidand Simpson type inequalities.
Thefollowingresult holds
[38]:
THEOREM 12 Let
f: [a, b]
Ibeamonotonicnondecreasingmapping on[a, b]
andtl,t2,t3(a, b)
be such that tl<
t2<
t3. Thenbf(x)dx
[(tl a)f(a) + (b t3)f(b) + (t3 tl)f(t)]
<_ (b t3)f(b) + (2t.
tlt3)f(t2) (tl a)f(a) + T(x)f(x)dx
<_ (b ta)(f(b) -f(t3))+ (t3 t2)(f(t3) f(t2))
+ (t2 tl)(f(t2) -f(tl)) + (tl a)(f(tl) --f(t2))
< max{t1-a,
t2- tl,t3 t2,b-t3}(f(b)-f(a)), (7.1)
where
j" sgn(tl x), for
xe [a, t2],
T(x) sgn(t3- x), for
x[t2,b].
Proof
Using integration bypartsformulaforRiemann-Stieltjesinte- gralwehaveabS(X)
df(x) (tl a)f(a) + (b t3)f(b) + (t3 tl)f(t2)
a’bf(x)d(x),
where
fx-tl, xE[a, t2],
X--t3, XE
[t2, b].
Indeed
s(x) df (x) (x df (x) + (x t3) af (x)
(x- tl)f(x)l
/(x- t3)f(t)lb= f(x) d(x)
(tl a)f(a) + (b- t3)f(b)
/
(t3 tl)f(t2)- f(x)
dx.Assume
thatAn’a xn)< xln)< < x < x(")=
b isasequence of divisions withu(An)O
as noe, whereu(An)
:=( )
maxi=0
,,_ x}"- x}
n} and(}n} xi. ’Xi+lJ
If p:[a, b]---,
11 is a continuous mapping on[a,b]
and v s monotonic nondecreasing on [a,b], thenbp(x)
dv(x)
n-1
lim v
u(An)’-’,o
/=oP@ }n)) v{ Xi+I)-
(n)n-1
v v
n-1
u(A.)--, .=
Xi+I)
fa IP(X)I dv(x). (7.2)
Applyingtheinequality
(7.2)
forp(x) s(x)
andv(x) --f(x),
xE[a, b]
wecanstate:
bs(x)
df(x)
which isthe firstinequalityin
(7.1).
If
f:
[a,b]
11is monotonicnondecreasingin[a,
b],wecanalsostate:atlf
(x)
dx<_f(t )(t a),
tt2f(x)
dx
>_f(t2)(t2- t),
tf(x)dx
<_f(t3)(t3 --t2),
and
bf(x)dx