4FUZZY OSTROWSKIINEQUALITIES
We present optimal upper bounds for the deviation of a fuzzy continuousfunction from its fuzzy average over [a; b] � R, error is measured in the D-fuzzy metric. The established fuzzy Ostrowski type inequalities are sharp,in fact attained by simple fuzzy real number valued functions. These in-equalities are given for fuzzy Hölder and fuzzy di¤erentiable functions andthese facts are re�ected in their right-hand sides. This chapter relies on[13].
4.1 Introduction
Ostrowski inequality (see [92]) has as follows����� 1
b� a
Z b
a
f(y)dy � f(x)����� �
1
4+
�x� a+b
2
�2(b� a)2
!(b� a)kf 0k1;
where f 2 C1([a; b]), x 2 [a; b]. The last inequality is sharp, see [5].Since 1938 when A. Ostrowski proved his famous inequality, see [92],
many people have been working about and around it, in many di¤erentdirections and with a lot of applications in Numerical Analysis and Prob-ability, etc.One of the most notable works extending Ostrowski�s inequality is the
work of A.M. Fink, see [64]. The author in [5] continued that tradition.
G.A. Anastassiou: Fuzzy Mathematics: Approximation The., STUDFUZZ 251, pp. 65–73.springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
66 4. Fuzzy Ostrowski Inequalities
This chapter is mainly motivated by [5], [64], [92], [103] and extendsOstrowski type inequalities into the fuzzy setting, as fuzziness is a naturalreality genuine feature di¤erent than randomness and determinism.
4.2 Background
We use the Fuzzy Taylor formula.
Theorem 4.1 (Theorem 1 of [11]). Let T := [x0; x0 + �] � R, with � > 0.We assume that f (i) : T ! RF are H-di¤erentiable for all i = 0; 1; : : : ; n�1,for any x 2 T . (I.e., there exist in RF the H-di¤erences f (i)(x + h) �f (i)(x), f (i)(x)� f (i)(x� h), i = 0; 1; : : : ; n� 1 for all small h : 0 < h < �.Furthermore there exist f (i+1)(x) 2 RF such that the limits in D-distanceexist and
f (i+1)(x) = limh!0+
f (i)(x+ h)� f (i)(x)h
= limh!0+
f (i)(x)� f (i)(x� h)h
;
for all i = 0; 1; : : : ; n � 1.) Also we assume that f (n), is fuzzy continuouson T . Then for s � a, s; a 2 T we obtain
f(s) = f(a)� f 0(a)� (s� a)� f 00(a)� (s� a)2
2!
� � � � � f (n�1)(a)� (s� a)n�1
(n� 1)! �Rn(a; s);
where
Rn(a; s) := (FR)
Z s
a
�Z s1
a
� � ��Z sn�1
a
f (n)(sn)dsn
�dsn�1
�� � ��ds1:
Here Rn(a; s) is fuzzy continuous on T as a function of s.We use
Proposition 4.2 (Proposition 1 of [10]). Let F (t) := tn � u, t � 0, n 2 N,and u 2 RF be �xed. Then (the H-derivative)
F 0(t) = ntn�1 � u:
In particular when n = 1 then F 0(t) = u.We mention
Proposition 4.3 (Proposition 6 of [10]). Let I be an open interval of Rand let f : I ! RF be H-fuzzy di¤erentiable, c 2 R. Then
(c� f)0 exists and (c� f)0 = c� f 0(x):
We use the �Fuzzy Mean Value Theorem".
4.3 Results 67
Theorem 4.4 (Theorem 1 of [10]). Let f : [a; b]! RF be a fuzzy di¤eren-tiable function on [a; b] with H-fuzzy derivative f 0 which is assumed to befuzzy continuous. Then
D(f(d); f(c)) � (d� c) supt2[c;d]
D(f 0(t); ~o);
for any c; d 2 [a; b] with d � c.We �nally need the �Univariate Fuzzy Chain Rule".
Theorem 4.5 (Theorem 2 of [10]). Let I be a closed interval in R. Hereg : I ! � := g(I) � R is di¤erentiable, and f : � ! RF is H-fuzzy di¤er-entiable. Assume that g is strictly increasing. Then (f � g)0(x) exists and
(f � g)0(x) = f 0(g(x))� g0(x); 8x 2 I:
4.3 Results
We present the following
Theorem 4.6. Let f 2 C([a; b];RF ), the space of fuzzy continuous func-tions, x 2 [a; b] be �xed. We assume that f ful�lls the Hölder condition
D(f(y); f(z)) � Lf � jy � zj�; 0 < � � 1; 8y; z 2 [a; b];
for some Lf > 0. Then
D
1
b� a � (FR)Z b
a
f(y)dy; f(x)
!� Lf
�(x� a)�+1 + (b� x)�+1
(�+ 1)(b� a)
�:
(4.1)
Proof. We have that
D
1
b� a � (FR)Z b
a
f(y)dy; f(x)
!
= D
1
b� a � (FR)Z b
a
f(y)dy;1
b� a � (FR)Z b
a
f(x)dy
!
=1
b� aD Z b
a
f(y)dy;
Z b
a
f(x)dy
!(by Lemma 1.13)
� 1
b� a
Z b
a
D(f(y); f(x))dy
� Lfb� a
Z b
a
jy � xj�dy =�
Lfb� a
��(x� a)�+1 + (b� x)�+1
�+ 1
�:
68 4. Fuzzy Ostrowski Inequalities
�Optimality of (4.1) comes next.
Proposition 4.7. Inequality (4.1) is sharp, in fact, attained by f�(y) :=jy � xj� � u, 0 < � � 1, with u 2 RF �xed. Here x; y 2 [a; b].Proof. Clearly f� 2 C([a; b];RF ): for letting yn ! y, yn 2 [a; b], then
D(f�(yn); f�(y)) = D(jyn � xj� � u; jy � xj� � u)
(by Lemma 1.2)� jjyn � xj� � jy � xj�jD(u; ~o)! 0; as n! +1:
Furthermore
D(f�(y); f�(z)) = D(jy � xj� � u; jz � xj� � u)(by Lemma 1.2)
� jjy � xj� � jz � xj�jD(u; ~o)� jjy � xj � jz � xjj�D(u; ~o) � jy � zj�D(u; ~o):
That is, for Lf� := D(u; ~o) we get
D(f�(y); f�(z)) � Lf� jy � zj�; 0 < � � 1; any y; z 2 [a; b]:
So that f� is a Hölder function.Finally we have
D
1
b� a � (FR)Z b
a
f�(y)dy; f�(x)
!
= D
1
b� a � (FR)Z b
a
(jy � xj� � u)dy; ~o!
=1
b� a �D (FR)
Z b
a
(jy � xj� � u)dy; ~o!
=1
b� aD Z b
a
jy � xj�dy!� u; ~o
!
=1
b� aD��
(x� a)�+1 + (b� x)�+1�+ 1
�� u; ~o
�=
Lf�
b� a
�(x� a)�+1 + (b� x)�+1
�+ 1
�: �
Next comes the basic Ostrowski type fuzzy result in
Theorem 4.8 let f 2 C1([a; b];RF ), the space of one time continuouslydi¤erentiable functions in the fuzzy sense. Then for x 2 [a; b],
D
1
b� a � (FR)Z b
a
f(y)dy; f(x)
!� supt2[a;b]
D(f 0(t); ~o)
!�(x� a)2 + (b� x)2
2(b� a)
�:
(4.2)
4.3 Results 69
Inequality (4.2) is sharp at x = a, in fact attained by f�(y) := (y � a)(b�a)� u, u 2 RF being �xed.
Proof. We observe that
D
1
b� a � (FR)Z b
a
f(y)dy; f(x)
!
= D
1
b� a � (FR)Z b
a
f(y)dy;1
b� a � (FR)Z b
a
f(x)dy
!
=1
b� aD (FR)
Z b
a
f(y)dy; (FR)
Z b
a
f(x)dy
!(by Lemma 1.13)
� 1
b� a
Z b
a
D(f(y); f(x))dy
(by Theorem 4.4)� 1
b� a
Z b
a
jy � xj supt2[a;b]
D(f 0(t); ~o)
!dy
=
supt2[a;b]
D(f 0(t); ~o)
!b� a
�(x� a)2 + (b� x)2
2
�;
proving (4.2).By Propositions 4.2, 4.3 and Theorem 4.5 we get that f�0(y) = (b�a)�u.
We have that
L:H:S:(4:2) = D
1
b� a � (FR)Z b
a
((y � a)(b� a)� u)dy; ~o!
= D
(FR)
Z b
a
((y � a)� u)dy; ~o!
= D
Z b
a
(y � a)dy!� u; ~o
!
= D
�(b� a)22
� u; ~o�=(b� a)22
D(u; ~o):
And
R:H:S:(4:2) = supt2[a;b]
D((b� a)� u; ~o) (b� a)2
=(b� a)22
D(u; ~o):
That is equality in (4.2) is attained. �We conclude with the following Ostrowski type inequality fuzzy general-
ization in
70 4. Fuzzy Ostrowski Inequalities
Theorem 4.9. Let f 2 Cn+1([a; b];RF ), n 2 N, the space of (n+ 1) timescontinuously di¤erentiable functions on [a; b] in the fuzzy sense. Call
M :=nXi=1
(b� a)i(i+ 1)!
D(f (i)(a); ~o):
Then
D
1
b� a � (FR)Z b
a
f(x)dx; f(a)
!� (4.3)"
M +
supt2[a;b]
D(f (n+1)(t); ~o)
!(b� a)n+1(n+ 2)!
#:
If f (i)(a) = ~o, i = 1; : : : ; n. Then
D
1
b� a � (FR)Z b
a
f(x)dx; f(a)
!� supt2[a;b]
D(f (n+1)(t); ~o)
!(b� a)n+1(n+ 2)!
:
(4.4)Inequalities (4.3) and (4.4) are sharp, in fact attained by
f�(x) := (b� a)(x� a)n+1 � u; u 2 RF being �xed:
Corollary 4.10. Let f 2 C2([a; b];RF ). Then
D
1
b� a � (FR)Z b
a
f(x)dx; f(a)
!(4.5)
�"(b� a)2
D(f 0(a); ~o) +
supt2[a;b]
D(f 00(t); ~o)
!(b� a)26
#:
When f 0(a) = ~o, then
D
1
b� a � (FR)Z b
a
f(x)dx; f(a)
!� supt2[a;b]
D(f 00(t); ~o)
!(b� a)26
:
(4.6)
Proof of Theorem 4.9. Let x 2 [a; b], then by Theorem 4.1 we get
f(x) =
n�1X�
i=1
f (i)(a)� (x� a)i
i!�Rn(a; x);
where
Rn(a; x) := (FR)
Z x
a
�Z x1
a
� � ��Z xn�1
a
f (n)(xn)dxn
�dxn�1
�� � ��dx1
4.3 Results 71
(here we need x � a). We observe that
D
1
b� a � (FR)Z b
a
f(x)dx; f(a)
!
=1
b� aD (FR)
Z b
a
f(x)dx; (FR)
Z b
a
f(a)dx
!
=1
b� aD (FR)
Z b
a
n�1X�
i=0
f (i)(a)� (x� a)i
i!�Rn(a; x)
!dx;
(FR)
Z b
a
f(a)dx
!
=1
b� a �D (FR)
Z b
a
n�1X�
i=1
f (i)(a)� (x� a)i
i!�Rn(a; x)
!dx; ~o
!
=1
b� aD (FR)
Z b
a
nX�
i=1
f (i)(a)� (x� a)i
i!�Rn(a; x)
!dx;
(FR)
Z b
a
f (n)(a)� (x� a)n
n!dx
!
=1
b� aD
nX�
i=1
(FR)
Z b
a
f (i)(a)� (x� a)i
i!dx
� (FR)Z b
a
Rn(a; x)dx; (FR)
Z b
a
f (n)(a)� (x� a)n
n!dx
!
=1
b� aD
nX�
i=1
f (i)(a)� (b� a)i+1
(i+ 1)!� (FR)
Z b
a
Rn(a; x)dx;
(FR)
Z b
a
f (n)(a)� (x� a)n
n!dx
!
� 1
b� a
"nXi=1
(b� a)i+1(i+ 1)!
D(f (i)(a); ~o)
+ D
(FR)
Z b
a
Rn(a; x)dx; (FR)
Z b
a
f (n)(a)� (x� a)n
n!dx
!#
= M +1
b� aD (FR)
Z b
a
Rn(a; x)dx; (FR)
Z b
a
f (n)(a)� (x� a)n
n!dx
!
72 4. Fuzzy Ostrowski Inequalities
= M +1
b� aD�(FR)
Z b
a
�Z x
a
�Z x1
a
� � ��Z xn�1
a
f (n)(xn)dxn
�dxn�1
�� � ��dx1
�dx;
(FR)
Z b
a
�Z x
a
�Z x1
a
� � ��Z xn�1
a
f (n)(a)dxn
�dxn�1
�� � ��dx1
�dx
��
(by Lemmas 1.13, 1.15)� M +
1
b� a
�Z b
a
�Z x
a
�Z x1
a
� � ��Z xn�1
a
D(f (n)(xn); f(n)(a)
�dxn
�dxn�1
�� � ��dx1
�dx
�(by Theorem 4.4)
� M +1
b� a
�Z b
a
�Z x
a
�Z x1
a
� � ��Z xn�1
a
(xn � a)
� supt2[a;b]
D(f (n+1)(t); ~o)
!dxn
�dxn�1
�� � ��dx1
�dx
�
= M +
�supt2[a;b]
D(f (n+1)(t); ~o)�
b� a(b� a)n+2(n+ 2)!
= M +
supt2[a;b]
D(f (n+1)(t); ~o)
!(b� a)n+1(n+ 2)!
:
We have established inequalities (4.3) and (4.4).Consider g(x) := c(x � a)` � u, x 2 [a; b], c > 0, ` 2 Z+, u 2 RF �xed.
We prove that g is fuzzy continuous. Let xn 2 [a; b] such that xn ! x asn! +1. Then
D(g(xn); g(x)) = D(c(xn � a)` � u; c(x� a)` � u)� cj(xn � a)` � (x� a)`jD(u; ~o)! 0:
Hence by the last argument, Propositions 4.2, 4.3 and Theorem 4.5 weobtain that f� 2 Cn+1([a; b];RF ).We see that
f�(i)(a) = ~o; for i = 1; : : : ; n:
That is M = 0. Furthermore it holds
f�(n+1)(x) = (b� a)(n+ 1)!� u:
4.3 Results 73
Finally, we notice that
L:H:S:((4:3); (4:4)) = D
1
b� a � (FR)Z b
a
((b� a)(x� a)n+1 � u)dx; ~o!
= D
u�
Z b
a
(x� a)n+1dx; ~o!= D
�u� (b� a)
n+2
n+ 2; ~o
�=
(b� a)n+2n+ 2
D(u; ~o):
Also we �nd
R:H:S:((4:3); (4:4)) = (b�a)(n+1)!D(u; ~o) (b� a)n+1
(n+ 2)!=(b� a)n+2n+ 2
D(u; ~o):
Proving (4.3) and (4.4) sharp, in fact attained inequalities. �