Rassias T. Functional Equations, Inequalities and Applications 2013

Functional Equations, Inequalities and Applications
Functional Equations,
Inequalities and
Applications
Edited by
Themistocles M. Rassias
Department of Mathematics,
National Technical University of Athens,
Zografou Campus, Athens, Greece
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A C.LP. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481 -6406-6 ISBN 978-94-017-0225-6 (eBook)

DOI 10.1007/978-94-017-0225-6
Printed on acid-free paper
All Rights Reserved

© 2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint of the hardcover 1st edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
Contents
Preface vii
1
Hyers-Ulam stability of a quadratic functional equation in Banach 1
modules
Jae-Hyeong Bae and Won-Gil Park
2
Cauchy and Pexider operators in some function spaces 11
Stefan Czerwik and Krzysztof Dlutek
3
The median principle for inequalities and applications 21
Sever S. Dragomir
4
On the Hyers-Ulam-Rassias stability of a Pexiderized quadratic 39
equation II
Kil- Woung Jun and Yang-Hi Lee
5
On the Hyers-Ulam-Rassias stability of a functional equation 67
Soon-Mo Jung
6
A pair of functional inequalities of iterative type related to a Cauchy 73
functional equation
Dorota Krassowska and Janusz Matkowski
7
On approximate algebra homomorphisms 91
Chun-Gil Park
8
Hadamard and Dragomir-Agarwal inequalities, the Euler formulae 105
and convex functions
Josip Pecaric and Ana Vukelic
v
vi Functional Equations, Inequalities and Applications
9
On Ulam stability in the geometry of PDE’s 139
Agostino Prdstaro and Themistocles M. Rassias
10
On certain functional equations and mean value theorems 149
Themistocles M. Rassias and Young-Ho Kim
11
Some general approximation error and convergence rate estimates 159
in statistical learning theory
Saburou Saitoh
12
Functional equations on hypergroups 167
Ldszlo Szekelyhidi
13
The generalized Cauchy functional equation 183
Abraham A. Ungar
14
On the Aleksandrov-Rassias problem for isometric mappings 191
Shuhuang Xiang
Index 223
Preface
Functional equations, inequalities and applications provides an extensive

study of several important equations and inequalities useful in a number of
problems in mathematical analysis. Subjects dealt with include: The general
ized Cauchy functional equation, the Ulam stability theory in geometric partial
differential equations, stability of a quadratic functional equation in Banach
modules, functional equations and mean value theorems, isometric mappings,
functional inequalities of iterative type related to a Cauchy functional equa
tion, stability of a Pexiderized quadratic equation, functional equations on hy
pergroups, some general approximation error and convergence rate estimates
in statistical learning theory, the median principle for inequalities and applica
tions, Hadamard and Dragomir-Agarwal inequalities, the Euler formulae and
convex functions, approximate algebra homomorphisms, stability of the ho
mogeneous functions and applications.
In addition to these, applications to certain problems in pure and applied
mathematics are included.
I wish to express my appreciation to the distinguished mathematicians who
contributed to this volume. We also wish to acknowledge the superb assistance
provided by the staff of Kluwer Academic Publishers.
I would also like to take this opportunity to thank my High School Professor
Elias Karakitsos, who inspired in me the love for mathematics early in life.
Professor of Mathematics
National Technical University of Athens,
May 2003
vii
Chapter 1
HYERS-ULAM STABILITY OF A
QUADRATIC FUNCTIONAL EQUATION
IN BANACH MODULES
Jae-Hyeong Bae
Department of Mathematics, Chungnam National University
Taejon 305-764, Korea
jhbae@math.cnu.ac.kr
Won-Gil Park
Taejon 305-764, Korea
wgpark@math.cnu.ac.kr
Abstract We extend the Hyers-Ulam-Rassias stability of a quadratic functional equation

f(x + y + z) + f(x - y) + f(y - z) + f(x - z) = 3>f(x) + 3f(y) + 3/(z)
to Banach modules over a Banach algebra.
2000 MSC: 39B72, 39B32 (Primary).
1. Introduction
In 1940, S.M. Ulam gave a wide ranging talk before the Mathematics Club
of the University of Wisconsin in which he discussed a number of important
unsolved problems ([16]). Among those was the question concerning the sta
bility of homomorphisms:
Let Gi be a group and let G<2 be a metric group with metric </(-,•)•
Given s > 0, does there exist a 6 > 0 such that if a function
h: Gi —» G2 satisfies the inequality d(h(xy), hfxjhfyf) < 5for
all x^y e Gi then there is a homomorphism H: Gi —» G2 with
d(h(x), H (#)) < s for all x € G\?
1
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 1-10.
© 2003 Kluwer Academic Publishers.
2 Functional Equations, Inequalities and Applications
The case of approximately additive mappings was solved by D.H. Hyers

[7] under the assumption that Gi and G? are Banach spaces. In 1978, Th.M.
Rassias [10] gave a generalization of the Hyers’s result. Recently, Gavruta [6]
also obtained a further generalization of the Hyers-Ulam-Rassias theorem.
The quadratic functional equation
f(x + y) + f(x ~y)~ 2/(ar) - 2/(y) = 0 (A)
clearly has f(x) = ex2 as a solution with c an arbitrary constant when f is a

real function of a real variable. We define any solution of (A) to be a quadratic
function. A Hyers-Ulam stability theorem for the equation (A) was proved
by F. Skof for functions f: V —> X where V is a normed space and X a
Banach space ([15]). In the paper [5], S. Czerwik proved the Hyers-Ulam-
Rassias stability of the quadratic functional equation (A) and this result was
generalized by a number of mathematicians ([3], [4], [8], [9], [11], [12], [13],
[14]).
Consider the following functional equation:
f(x + y + 2) + f(x - y) + /(y - z) + f(x - z)

= 3f(x) + 3f(y) + 3f(z). (B)
Recently, the first author investigated the generalized Hyers-Ulam stability

problem of the equation (B) and proved the Hyers-Ulam stability of the equa
tion on restricted domains ([1], [2]). In sections 2 and 3 of this paper, we extend
the generalized Hyers-Ulam stability of the quadratic functional equation (B)
to Banach modules over a Banach algebra, and prove the stability.
2. The Results on B-Quadratic Mappings

In this paper, let B be a unital Banach algebra with norm | • |, and Bi = {a G
B : |a| = 1}, let bBi and b®2 be left Banach B-modules with norms || • ||
and || • ||, respectively, and let : b®i \ {0} x b®i \ {0} x #Bi \ {0} —> [0,00)
be a function satisfying
00 1
■?(!, y, z) := £ SSTP(3"‘i. 3mv, 3”“z) < oo (i)
m=0
for all x, y, z G b®i \ {0}, or

00 z \
S(x,y, z) := -——r, o -——r | < 00 (ii)
m=0 ' '
for all x, y, z G b®i \ {0}.

Hyers-Ulam stability of a quadratic functional equation in Banach modules 3
Definition 2.1 A quadratic mapping Q: pBi is called B-quadratic if

Q(ax) = a2Q(x) for all a e B and all x G b®i-
Theorem 2.1 Let f: #Bi —> b®2 be a mapping such that
||a2/(x + y + z) + a2f(x - y) + a2f(y - z) + a2f(x - z)

- 3f(ax) - 3f(ay) - 3/(a2)|| < p(x,y,z) (2.1)
for all a e Bi and all x,y,z € b®i \ {0}, and let /(0) = Qfor the case (ii). If
f(tx) is continuous int e R for each fixed x e b®i> then there exists a unique
B-quadratic mapping Q : b®i —> b®2 such that
3
Q(z) - f(x) + -/(0) < p(x, x, x) (2.2)
o
for all x € \ {0}. The mapping Q: bBi —> b®2 is given by
limn_>Oo if V satisfies (i)

Q(z) = <
limnôo 9"/(^r) if p satisfies (ii)
for all x G bBi.
PROOF: By [2], it follows from the inequality of the statement for a — 1 that
there exists a unique quadratic mapping Q: pBi —> b®2 satisfying (2.2). The
quadratic mapping Q is similar to the additive mapping T given in the proof
of [10]. Under the assumption that f(tx) is continuous in t G R for each fixed
x G bBi, by the same reasoning as the proof of [10], the quadratic mapping
Q : bBi —> b®2 satisfies Q(tx) = t2Q(x) for all x 6 b®i and all t 6 R.
Suppose that p satisfies (i). Replacing x, y and z by 3n~1x in (2.1),
||a2/(3na;) + 3a2/(0) - Qf^âx)|| < p^^x, 3n~1x) (2.3)
for all a g Bi and all x G gBi \ {0}. Putting a = 1 and replacing x, y and z
by x in (2.1),
||/(3a;) + 3/(0) - 9/(x) || < p(x, x, x) (2.4)
for all x e b®i \ {0}. Replacing x by 3n rax in (2.4),
||/(3nax) + 3/(0)-9/(3n“1ax)|| < p^âx, 3n~1ax, 3n~iax>

) (2.5)
for all a e Bi and all x, y, z e b®i \ {0}.

Note that for each a € B and each w e b®2,
< K\a\ ■ Ml (2.6)

for some K > 0. Using (2.3), (2.5) and (2.6),
||a2/(3"x) —/(3narr)||
< ||a2/(3nrr) + 3a2/(0) - 9f(3n~lax)||
+ \\9f(3n~lax) - f(3nax) - 3a2/(0)||
< y(3n-xx,3n-lx,3n-xx}
+ \\f(3nax) + 3/(0) - 9/(3"-1ax)|| + || - 3/(0) + 3a2/(0)||
+ p^âx, 3n~1ax, Sâx) + 3K\a2 - 1| • ||/(0)||
for all a € Bi and all x G #Bi \ {0}.

So 9“n ||a?f(3nx) — /(3naa;)|| —> 0 as n —> oo for all a G Bi and all
z G b®i \ {0}. By the definition of Q, for each element a G Bi,
. z x a2f(3nx} „ f(3nax) x
a2Q(rr) = hm ------ ------ = hm —-- ------= Q(ax)
’ n—>oo 9" n-+oo 9n ’
for all x G b®i \ {0}- For a G Bi \ {0},

\ / \ 2
(
|a|Ax ) = |a|2Q ( Ax) = |a|2 Q(x) = a2Q(x),
|a| / \lal / lal
for all x G b®i \ {0}. So Q is B-quadratic.

By the same reasoning as above, for the case (ii), one can show that the
R-quadratic mapping Q is B-quadratic, as desired. □
Notice that, for the case (ii), Q(x) = limn_,oo 9n(/(3-nar) — /(0)) in [9].
To get Q(ax) = a2Q(x) for all a G Bi and for all x G gBi, we need the
assumption /(0) — 0. In this paper, assume that /(0) = 0 for the case (ii).
Corollary 2.1 Let p f 2, 0 > 0 be real numbers and f: #Bi —> b®2 «
mapping such that
||a2/O + y + z) + a2f(x - y) + a2f(y - z) + a2f(x - z)

- 3f(ax) - 3f(ay) - 3/(az)|| < 0(||x||^ + ||< + ||<)
for all a E Bi and all x,y,z e b®i \ {0}- If f(fx) is continuous in t e

R/or each fixed x € b®i, then there exists a unique B-quadratic mapping
Q: b®i —> b®2 such that
3 30
Q(t) - /(a;) + -/(0) ll<
o |3p — 3|
for all x 6 b®i \ {0}.
Corollary 2.2 Let Ei and be complex Banach spaces and f: Ei —» E^ a

mapping such that
|| A2/(a? + y + z) + A2/(z -y) + X2f(y - z) + A2/(t - z)

- 3/(Aa?) - 3/(Ay) - 3/(Az) || < <^(x, y, 2)
for all A E C with | A| = 1 and all x^z € Ei\ {0}. If f(tx) is continuous in
t e Rfor each fixed x e Ei, then there exists a unique C-quadratic mapping
Q: Ei —» £?2 such that
3
+ -/(0) < <p(x, X, x)
for all x e Ei \ {0}.
PROOF: Since C is a Banach algebra, the Banach spaces Ei and E% are con
sidered as Banach modules over C. By Theorem 2.1, there exists a unique
C-quadratic mapping Q: Ei —» E^ satisfying the conditions given in the state
ment. □
Theorem 2.2 Let f: b®i —> b®2 be a mapping such that
|| f(ax + ay + az) + f(ax — ay) + f(ay — az) + f(ax — az)

- 3a2/(x) - 3a2/(y) - 3a2/(z) || < y>(x, y, z) (2.7)
for all a E Bi and all x,y,z e b®i \ {0}. If f(tx) is continuous in t e

for each fixed x E B®b then there exists a unique B-quadratic mapping
Q: b®i —> B®2 such that (2.2). The mapping Q: b®i —> B®2 is given by the
same formula as the statement of Theorem 2.1.
PROOF: By the same reasoning as the proof of Theorem 2.1, there exists a
unique R-quadratic mapping Q: b®i —> b®2 satisfying (2.2).
Suppose that <p satisfies (i). Replacing x, y and z by 3n-1z in (2.7),
\\f(3nax) + 3/(0) - 9a2/(3n-1x)|| < ^(3"-^, 3"-^, S71"1^) (2.8)
for all a e Bi and all x e \ {0}. Putting a = 1 and replacing x, y and z

by x in (2.7),
||/(3a?) + 3/(0) - 9/(x) || < <p(x, x, x) (2.9)
for all x e bBi \ {0}. Replacing x by 3n-1or in (2.9),
||/(3™a?) + 3/(0) - 9/(3n-1ir)|| < ^(T^x, S^x, T^x) (2.10)

for all a e Bi and all x, y, z e gBi \ {0}. Using (2.8), (2.10) and (2.6),
\\f(3nax) — a2f(3nx)\\
< ||/(3”ax) + 3/(0) - 9a2/(3"-1x)||
+ - a2/(3nx) - 3/(0) ||
+ W^f^x) - a2/(3^)|| + || - 3/(0)||

< <-^3n-\3n-M
+ 7<||9/(3-M-/(3n7ll+ 3||/(0)||
<p(3n~xx, 3n~1x,3n-1x)
+ K'^||9/(3n-1x) - /(3"x) - 3/(0)|| + ||3/(0)|0 + 3||/(0)||
< p(3n~1x.l3n~1x.l3n~lx)
+ K (^(3"-7 3n-1x, 3n~lx) + 3||/(0) 10 + 31|/(0) ||
for all a e Bi and all x e bBi \ {0}.

The rest of the proof is the same as the proof of Theorem 2.1. Hence, the
unique R-quadratic mapping Q: #Bi —> 582 is a B-quadratic mapping, as
desired. □
Corollary 2.3 Let p / 2, 0 > 0 be real numbers and f: pBi —> bB2 a
mapping such that
|| / (ax + ay + az) + f(ax — ay) + f(ay — az) + f(ax — az)

- 3a2/M - 3a2/(w) - 3«2/(z)|| < 6(||< + ||< + |MIP)
for all a e Bi and all x^y^z e b®i \ {0}- If f(fx) Is continuous in t e

R for each fixed x e B®b then there exists a unique B-quadratic mapping
Q: b®i —> B®2 such that
3 30
Q(x)-/7) + -/(0) ll<
o |3?-3|
for all x G bBi \ {0}.
Corollary 2.4 Let Ei and E% be complex Banach spaces and f: E\ —> E<i a
mapping such that
||/(Ax + Xy + Xz) + f(Xx - Xy) + /(Ay - Xz) + f(Xx - Xz)

- 3X2f(x) - 3A2/(y) - 3A2/(£) || < ^(x, y, z)
for all A G C with |A| = 1 and all x,y,z E Ei\ {0}. If f(tx) is continuous in
t G R for each fixed x G Ei, then there exists a unique C-quadratic mapping
Q -. Ei —> E<2 satisfying the condition given in the statement of Corollary 2.2.
PROOF: The proof is similar to the proof of Corollary 2.2. □
3. The Results on Bsa-Quadratic Mappings

In this section, let B be a complex unital Banach *
-algebra with norm | ■ |.
Definition 3.1 For a e B, let b = aa

,
* a* a, or aa‘+a‘a, A quadratic mapping
Q: —> bB>2 is called Bsa-quadratic if Q(ax) = bQ(x) for all a G B, and
all x G bBi.
Theorem 3.1 Let f: #Bi —> 582 be a mapping such that
\\bf(x + y + z) + bf(x — y) + bf(y - 2) + bf(x - z)

if (ay) - “
- 3f(ax) - “ if (az) || < <p(x, y, z) (3.1)
for all a e Bi and all x,y,z G bBi \ {0}. If f(tx) is continuous in t G R

for each fixed x G bBi, then there exists a unique Bsa-quadratic mapping
Q: bBi -> bB2 satisfying (2.2).
Proof: By the same reasoning as the proof of Theorem 2.1, there exists a
unique R-quadratic mapping Q: #Bi —> 582 satisfying (2.2).
Suppose that (3n~1x, 3n-1x, 3n-1x) (3.2)
for all a G Bi and all x G bBi \ {0}. Putting a = 1 and replacing x, y and z
by x in (3.1), we get
||/(3x) + 3/(0) - 9f(x)|| < <p(x, x, x) (3.3)
for all x G bBi \ {0}. Replacing x by 3n-1ax in (3.3), we get
\\f(3nax) + 3f(0)-9f(3n~1ax)\\ < <p(3n~1ax,3n-1ax,3n~1ax) (3.4)
for all a G Bi and all x, y, z G bBi \ {0}. Using (3.2), (3.4) and (2.6), we get
\\bf(3nx)-f(3nax)\\
< \\bf(3nx) + 36/(0) - 9/(3"-M||
+ || - 36/(0) + 9f(3n~1 ax) - f(3nax)\\
+ \\f(3nax) + 3/(0) - 9/(3n-1^)|| + || - 3/(0) + 36/(0)||
+ <p(3n~lax, 3^ ax, 3n~1ax) + 3K|6 - 1| • ]|/(□) ||
for all a e Bi and all x G b®i \ {0}.

So 9_"||6/(3"a:) — f(3nax)|| —> 0 as n —> oo for all a G Bi and all
x G bBi \ {0}. By the definition of Q, for each element a G Bi,
A r bf(3nx) f(3nax)
bQ(x) = hm —- ----- = hm —-- ------ = Q(ax)
’ n—foo 9n n—>oo 9"
for all x G b®i \ {0}. For a G B \ {0},
Q(ax) = Q
for all x G b®i \ {0}. So Q is Bsa-quadratic.

By the same reasoning as above, for the case (ii), one can show that the
R-quadratic mapping Q is Bsa-quadratic, as desired. □
Theorem 3.2 Let f: b®i —* b®2 be a mapping such that
||/(ax + ay + az) + f(ax - ay) + f(ay - az) + f(ax - az)

- 3bf (x) - 36/(1/) - 36/(2) || < <p(x,y,z) (3.5)
for all a E Bi and all x,y,z G b®i \ {0}- If ffac) is continuous intE'R
for each fixed x e B®b then there exists a unique Bsa-quadratic mapping
Q -. bBi -> b®2 satisfying (2.2).
unique R-quadratic mapping Q: b®i —> b®2 satisfying (2.2).
Suppose that <p satisfies (i). Replacing x, y and z by 3"-1x in (3.5),
||/(3"ax) + 3/(0) - 96/(3"-1x) || < ^^x, 3n~xx, 3n~xx) (3.6)
for all a G Bi and all x G b®i \ {0}. Putting a = 1 and replacing x, y and z
by x in (3.5),
11/(3x) + 3/(0) - 9f(x) || < ?(x, x, x) (3.7)
for all x G b®i \ {0}- Replacing x by 3"-1x in (3.7),
||/(3nx) + 3/(0) - 9/(3n-1x) || < ^(3"-^, 3n~1x, 3n~lx) (3.8)

for all a € Bi and all x, y, z 6 \ {0}. Using (3.6), (3.8) and (2.6),
||/(3^)-b/(3+r)||
< ||/(3naa;) + 3/(0) — 9b/(3"-1x)||
+ ||96/(3n-1x) — 6/(3"x) — 3/(0)||
<^3n-1x,3n-1x,3n~1x')
+ ||96/(3"-1x)-6/(3^)|| +||-3/(0)||
<^3>n-1x,3n-1xX~1x')
+ JC||9/(3n-1a:) - /(3^)|| + 3||/(0)||
+ K^\\9f (3n~lx) - f(3nx) - 3/(0)|| + ||3/(0)||) + 3||/(0)|(
<?(3n-xx,3n~lx,3n~lx')
+ K^p^x, 3n^x, 3-1x) + 3||/(0)||) +3||/(0)||
for all a € Bi and all x G sBi \ {0}.

The rest of the proof is the same as the proof of Theorem 3.1. Hence, the
unique R-quadratic mapping Q: #Bi —> b®2 is a Bsa-quadratic mapping, as
desired. □
References
[1] J.-H. Bae and I.-S. Chang: ‘On the Ulam stability problem of a quadratic
functional equation’, Korean J. Comput. & Appl. Math. (Series A) 8
(2001), 561-567.
[2] J.-H. Bae and H.-M. Kim: ‘On the generalized Hyers-Ulam stability of a
quadratic mapping’, Far East J. Math. Sci. 3 (2001), 599-608.
[3] C. Borelli and G. Forti: ‘On a general Hyers-Ulam stability result’, In-
temat. J. Math. Math. Sci. 18 (1995), 229-236.
[4] P.W. Cholewa: ‘Remarks on the stability of functional equations’, Aequa-
tiones Math. 27 (1984), 76-86.
[5] S. Czerwik: ‘On the stability of the quadratic mapping in the normed
space’, Abh. Math. Sem. Hamburg 62 (1992), 59-64.
[6] P. Gavruta: ‘A generalization of the Hyers-Ulam-Rassias stability of ap
proximately additive mappings’, J. Math. Anal, and Appl. 184 (1994),
431-436.
[7] D.H. Hyers: ‘On the stability of the linear functional equation’, Proc. Nat.
Acad. Sci. USA 27 (1941), 222-224.
[8] D.H. Hyers, G. Isac and Th.M. Rassias: ‘Stability of Functional Equa
tions in Several Variables', Birkhauser, Berlin, Basel and Boston, 1998.
[9] K.-W. Jun and Y.-H Lee: ‘On the Hyers-Ulam-Rassias stability of a Pex-
iderized quadratic inequality’, Math. Ineq. Appl. 4 (2001), 93-118.
[10] Th.M. Rassias: ‘On the stability of the linear mapping in Banach spaces’,
Proc. Amer. Math. Soc. 72 (1978), 297-300.
[11] Th.M. Rassias: ‘On the stability of the quadratic functional equation’,
Mathematica (to appear),.
[12] Th.M. Rassias: ‘On the stability of functional equations and a problem of
Ulam’, Acta Applicandae Mathematicae 62 (2000), 23-130.
[13] Th.M. Rassias and P. Semrl: ‘On the Hyers-Ulam stability of linear map
pings’, J. Math. Anal. Appl. 173 (1993), 325-338.
[14] Th.M. Rassias and J. Tabor: ‘What is left of Hyers-Ulam stability?’, J.
Nat. Geom. 1 (1992), 65-69.
[15] F. Skof: ‘Proprieta locali e approssimazione di operatori’, Rend. Sem.
Mat. Fis. Milano 53 (1983), 113-129.
[16] S.M. Ulam: ‘Problems in Modem Mathematics', Wiley, New York, 1960.
Chapter 2
CAUCHY AND PEXIDER OPERATORS

IN SOME FUNCTION SPACES
Stefan Czerwik
Institute of Mathematics, Silesian University of Technology,
Kaszubska 23, 44-100 Gliwice, Poland
steczer@zeus.polsl.gliwice.pl
Krzysztof Dhitek
Institute of Mathematics, Silesian University of Technology,
Kaszubska 23, 44-100 Gliwice, Poland
kdlutek@zeus.polsl.gliwice.pl
Abstract In this paper we define Cauchy and Pexider operators by applying the well-
known Cauchy and Pexider differences and then consider some properties of
such operators. Other type of operators (A-operators, quadratic operators, Jensen
operators) that are also very important in the theory of functional equations are
investigated. Moreover, some new functional equations closely related to this
problem, are discussed as well.
2000 MSC: 39B99 (Primary); 47A05 (Secondary).
Keywords: Cauchy and Pexider differences, linear operators.
1. Introduction
Let (X, +) and (V, +) be arbitrary groups. For f, g, h: X —> Y we define
the differences
y) := f(x + y) — /(a:) - /(y),

P(/, g, h)(x, y) := f(x + y) - y(x) - /(y),
y) := f(x + y) + f(x -y)~ 2/(z) - 2/(y)
11
for all x, y G X, which are well known as the Cauchy, Pexider and quadratic
differences respectively.
If
C(/)(z,y) = 0,
P(/,p,h)(x,y) = 0,
=0,
then one obtains well-known and interesting results for the Cauchy, Pexider
and quadratic equations. There are also very many generalizations of these
equations. For some information concerning Cauchy and Pexider equations
see [1]; for the quadratic equation see the paper [7].
Recently several mathematicians investigate the problem of Hyers-Ulam-
Rassias stability of functional equations (e.g. Cauchy, Pexider, quadratic,...)
(cf. [2], [3], [4], [5], [6], [8]).
In the present paper we consider a new problem: The Cauchy, Pexider,
quadratic,... differences generate interesting operators from function spaces
into other function spaces. We study this problem on a special space of func
tions. Similar problems have been studied in [9].
2. Cauchy operator
Let X and Y be normed spaces. Fix A > 0 and consider a function f: X
Y satisfying the condition
||/(x)|| M(/kAW, xeX,
where M(/) is a constant depending on f. Let’s denote by Xx the space of all

such functions.
Define
||/|| := sup {e-A|M||/^)||} , feXx. (2.1)
xex 1 J
Then we have
Lemma 2.1 The space X\ with norm (2.1) is a linear normed space.
Now denote
C(f)(x,y) := f(x + y) - /(x) - f(y), f e Xx, x,y e X. (2.2)
Then C(/): X x X -> Y and
||C(/)(x,y)|| ^3M(/)eA(l|l||+M), f € Xx, x,y G X. (2.3)

The space of all g: X x X —> Y satisfying the condition
||5(T,y)|| x,ytX,
where M (g) is a constant depending on g, is denoted by X%.

We define
Ill'll := sup |e-A(l|:r||+ll3/ll)||5(x,y)||| . (2.4)
Lemma 2.2 The space X^ with norm (2.4) is a linear normed space.
We shall prove the following
Theorem 2. 1 The Cauchy operator C : Xx —> X% defined by (2.2) is a linear

bounded operator and
\\Cm 3||/|| for all f G. Xx- (2.5)
PROOF: We have the inequalities
||C(/)|| := sup{e-A(M+M)||/(rc + j/)-/(rC)-/(y)||}

x,y€X
sup {e-WII+M)(|)/(l + </)ll + ll/WII + ll/fo)II)}
x,y^X
sup \\f(x + y)\\}

x,y€X
+ sup{e-AM ||/(z)||} + sup{e-AM ||/(y)||} = 3||/||.
xex yex
□
In the following, R+ denotes the set of nonnegative real numbers.
Theorem 2. 2 Let R+ C X, R+ C Y and ||x|| — \x]for x € R+. Then
HOU = 3. (2.6)
PROOF: Let {a?n} C R+ be a decreasing sequence such that xn —> 0 as

n —> oo. Define
_g2A®. x = xn,
fn(x) := < e2Aa;", X — 2xn,
otherwise.
Then ||/n(#)|| eA^neA||2?|| x x, and therefore fn G X\ for all n G N.

Moreover,
Ar
{ 1,
0,
X = 2xn,
otherwise,
and hence ||/n|| = eXxn, n € N.

Now we obtain
||C(/n)|| := sUp{e-A(M+M)||/n(^ + j/)-/n(x)-/n(y)||}
x,yeX
-e2Xx^\fn(2xn)-2fn(xn)\
_ _ ^2Xxn |g2Arrn j rjgÂxn । __
Suppose that ||C|| <3. Then there exists £ > 0 such that
||C(/)H(3-s)||/|| forall/eXA.
But for fn G Xx one has
3 ||C(/„)|| < (3 - s)||/n|| = (3 - £)eXxn 3-£ as n oo,
which is imposible. □
3. Pexider operator
Define
P(f, g, h)(x, y) := f(x + y) - g(x) - h(y), (3.1)

f,g,heXx, x,yeX,
*h= {(f,9,hy. f,g,heXx},
and
||(/, g,h)\\ := max{||/l|, ||p|| >||}, f,g,heXx. (3.2)
Then is a linear normed space.
Theorem 3.1 The Pexider operator P: —> X^, given by the formula (3.1),
is a linear bounded operator and
||P(U)||^3|H forueX3x. (3.3)
PROOF: We get for f,g,h E Xx,

■= sup {e-A(l|:r|l+l|y|l)||/(a: + y) - g(x) - %)||}
sup {e A(M+lh/ll) \\f(x + y)||}

x,yeX
+ sup{e-A||x|l||y(x)||} + sup{e-A|l2/ll||/i(y)||}
xfzX yfzX
= 11/11 +IMI + H
3max{||/||, ||< R|}=3||(/,g,h)\\.
□
Corollary 3.1 If^-+ C X, R+ C Y and ||x|| = |x| for x G R+, then

||P|| = 3. (3.4)
Proof: Suppose for the contrary that ||P\\ < 3. Then for f = g = h we have
\\P(f,g,h-)\\ = ||C(/)K l|P|HI(/,/,/)ll = IIPIIII/II,
whence
IM/)KI|P||||/|| forall/eXA
and ||P|| < 3, which is imposible in view of Theorem 2.1. □
4. Others operators
4.1 Jensen operator
Denote
y) := f[sx + (1 - s)y] - sf(x) - (1 - s)/(y), (4.1)
for x,y e X, f e X\, where 0 < s < 1.
Then we have (simple proof is omitted here)
Corollary 4.1 The Jensen operator Js: X\ —> X^, given by formula (4.1), is
a linear operator and
U(/)K 2||/|| for f E Xx. (4.2)
4.2 Quadratic operator

Define
y) := /(x + y) + /(x - y) - 2f(x) - 2f{y\ (4.3)
for x, y G X, f G Xx.
Corollary 4.2 The quadratic operator Q: Xx —> X^, given by formula (4.3),
is a linear operator satisfying the inequality
IIW)K6||/|| forfexx.
4.3 A-operator
Let’s denote
A(/)(z,y) := f(x + y) - f(x), x,y e X, f e X\. (4.4)
Then one has
Corollary 4.3 The operator A: X\ —> X^, given by formula (4.4), is a linear
bounded operator satisfying the property
||A(/)||^ 2||/|| forfeXx.
Now let’s consider
Xhmxy.^f{x + h}-f{x\ xeX, feXx,
where h € X is fixed.
Then for A/j: —► Xx and X and Y equal to R or C (the set of complex
numbers) and 0 < h e JR, we obtain ||Zk/i|| = 2. This is easily seen if one
takes the function / defined as follows:
x e (0, h\
/(*) x e (fo, oo),
otherwise.
Hence, we get
IIA^/)|| = sup\f(x + h) - f(x)\ = \f(h) ~ /(0)| = 2

xex
and because of the fact ||A/l(/)|| < 2||/||, for all f e X\, it follows that
INM
5. Pexider equation
The problem of existence of inverse operators for the operators that we have
already introduced leads to the solution of certain functional equations.
In the following we shall discuss some of these equations.
Theorem 5.1 Let X be a semigroup with zero, Y an abelian group and <&,
: X —> Y given functions. The functions f,g,h: X —> Y satisfy the equa
tion
f(x + y) - g(x) - hy) = <&(z) + ’J'(y), x,yeX, (5.1)
if and only if
/(x) = A(x) + a + b + S(0) + #(0),

y(x) = ?l(x) - $(x) + a + $(0), (5.2)

h(x) = A(x) — + b + ^(0), x E X,
where A: X —> Y is an additive function and a,b eY are arbitrary constants.
PROOF: Set x = 0. Then from (4.1) we get
fc(y) = /(y)-*(y)-y(0)-<fr(0) (5.3)
and by substitution in (5.1), one gets
f(x + y) - g(x) - f(y) = $(*) - 0(0) - $(0). (5.4)
Putting (5.3) and (5.4) into (5.1), we get
f(x + y) - /(x) - f(y) + /(0) = 0 for all x, y e X. (5.5)
Set
A(x):=/(x)-/(0), xeX, (5.6)
then (5.5) shows that A is an additive function.

From (5.1) for x = y = 0, we get
/(0)=y(0) + /i(0) + $(0) + ^(0)
and setting a = y(0), b = h(0) from (5.3), (5.4), (5.6) we obtain the formulas
(5.2).
Conversly, any functions /, g, h given by the formulas (5.2), where A is
an additive mapping and a, b e X are arbitrary constants, are solutions of the
equation (5.1). This completes the proof. □
6. Generalized A-equation
Theorem 6.1 Let X be a semigroup with zero and Y an abelian group. As
sume F : X x X —> y is a given function. The equation
f(x + y) - g(x) = F(x,y), x,y e X, (6.1)
has a solution the functions f^giX—^Y if and only if a function F satisfies

the following equation
F(x, y) + F(0, x) = F(0, x + y) + F(x, 0), x, y e X. (6.2)
If F satisfies the equation (6.2), then f,g:X —> Y form a solution of the
equation (6.1) if and only if
f(x) = F(0, x) + c
g(x) =F(O,x)-F(x,O) + c, (6.3)
where ceY is an arbitrary constant.
Proof: Assume that the equation (6.1) has a solution. Substituting x = 0, we

obtain
/(y) = F(0,y) +j?(0)
and setting y = 0 into (6.1), we get
g(x) = f(x)~ F(x, 0) = F(0, x) - F(x, 0) + g(0).
Hence if we denote <?(0) = c, we obtain the formulas (6.3).

Since (/, g) is a solution of the equation (6.1), we have
F(0, x + y) + 5(0) - F(0, x) + F(x, 0) - #(0) = F(x, y),
i.e. F satisfies the equation (6.2).
□ The simple verification of the other part of the theorem is left to the reader.
In the special case f — g, we get the following result:
Theorem 6.2 Let the assumptions of Theorem 6.1 be fulfilled. Then the equa
tion
f(x + y) - f(x) = F(x,y), x,y € X, (6.4)
has a solution f:X—>Y if and only if a function F satisfies the equation
F(x,y) +F(0,x) = F(0,x + y), x,y e X. (6.5)
If F satisfies the equation (6.5), then f: X —> Y is a solution of the equation

(6.4) if and only if
f(x) = F(0, x) + c, x € X, (6.6)
where ccY is an arbitrary constant.
References
[1] J. Aczel: 'Lectures on Functional Equations and their Applications', Aca
demic Press, New York and London, 1966.
[2] S. Czerwik: ‘On the stability of the homogeneous mapping’, C. T. Math.
Acad. Sci. Canada XIV (6) (1994), 268-272.
[3] S. Czerwik: ‘The stability of the quadratic functional equation’, in: ‘Sta
bility of mappings of Hyers-Ulam type’ (eds: Th.M. Rassias and J. Tabor),
Hadronic Press, Palm Harbor, Florida, 1994, pp. 81-91.
[4] G.L. Forti: ‘Hyers-Ulam stability of functional equations in several vari
ables’, Aequat. Math. 50(1995), 143-190.
[5] D.H. Hyers and Th.M. Rassias: ‘Approximate homomorphisms’, Aequat.
Math. 44 (1992), 125-153.
[6] M. Kuczma: ‘An Introduction to the Theory of Functional Equations and
Inequalities', Panstwowe Wydawnictwo Naukowe, Warszawa-Krakow-
Katowice, 1985.
[7] S. Kurepa: ‘On the quadratic functional’, Publ. Inst. Math. Acad. Serbe
Sci. Beograd 13 (1959), 57-72.
Proc. Amer. Math. Soc. ’ll (1978), 297-300.
[9] S. Siudut: ‘Cauchy difference operator in Lp spaces', Preprint Pedagogical
University in Rzeszow, 1996.
Chapter 3
THE MEDIAN PRINCIPLE

FOR INEQUALITIES
AND APPLICATIONS
Sever S. Dragomir
School of Communications and Informatics,
Victoria University of Technology
PO Box 14428, MCMC 8001
Victoria, Australia
sever@matilda.vu.edu.au
http://rgmia.vu.edu.au/SSDragomirWeb.html
Abstract The “Median Principle” for different integral inequalities of Gruss and Ostrowski
type is applied.
2000 MSC: 26D15 (Primary); 26D10 (Secondary).
Keywords: median principle, Gruss type inequality, Ostrowski’s inequality
1. Introduction
There are many mathematical inequalities whose right hand side may be
expressed in terms of the sup-norm of a certain derivative for the involved
functions.
For instance, in Numerical Analysis, the integral of a function f: [a, 6] —>
R may be represented by
J" f(t)dt = An(In,f) + Rn(<In,f^, (1.1)
where An (In, f) is the quadrature rule defined on a given division
In • Ob — *^0 < 1 X/i —
21
of the interval [a, fe] and Rn(Im is the remainder, usually expressed in the
integral form
-Rn Kn(In,t)f^ (1-2)
where Kn (In, •) : [a, 6] —> R is an appropriate Peano kernel and f-r> is the
r-th derivative of f assumed to be essentially bounded on [a, b].
If the integral
l-Kn (Ai, I) | dt
can be exactly computed or bounded above by different techniques, then we

have the error estimate
< |/(r)|l [b\Kn(In,t)\dt, (1.3)

I \ /II II[a,6],oo Ja
where
/(r)|l := ess sup (t)
ll[a,6],oo t6[o>6] I
that provides a large class of examples of inequalities mentioned above.

In Analytic Inequalities Theory, results such as Ostrowski’s inequality
0 & Ja
re G [a, b], (1.4)
provided f is absolutely continuous with f' G Loo [a, b], or Cebysev’s inequal
ity
7-^— [ f(t)g (t) dt - —f f (i) dt ■ —[ g (i) dt

0 & Ja & Ja & Ja
<^('--“)2||/'ll|.,1,1.»l|/|l1.,tl.oc. (1-5)
provided f and g are absolutely continuous with f' G Loo [a,b], are other
natural examples.
Since, in order to estimate ||/^^||[a,b],oo> in practice it is usually necessary

to find the quantities
Mr := sup (t) and mr := inf (t)

te[a,6]
(as ||/(r)||[a,b],oo = max{|Mr|, |mr|}), the knowledge of ||/(r)||[o>b]>oo may be

as difficult as the knowledge of Mr and mr.
Consequently, it is a natural problem in trying to establish inequalities where
instead of Wf^ || [O,b],oo»one would have the positive quantity Mr—mr. We also
must note that for functions whose derivatives f(r^ have a “modest variation”,
the quantity Mr — mr may usually be a lot smaller than ||/^||[0,b],oo-
We note that there are many examples of inequalities where the bounds are
expressed in terms of Mr — mr, from which we would like to mention only the
celebrated result due to Gruss:
If6 1 rb If6
/ f G) 9 (*) dt - —— / f (t) dt ■ —— / g (t) dt
b a Ja ba J a b a Ja
< - (M — m) (N — n), (1.6)
provided /, g e L [a, b] with
m < f < M, n < g < N a.e. on [a, b]. (1.7)
It is the main purpose of this paper to point out a general strategy for trans
forming an inequality whose right hand side is expressed in terms of \\f^ ||[a,6],oo
into an inequality for which the same side will be expressed in terms of the
quantity Mr — mr> 0. We call this method the “Median Principle”. A formal
statement of this method is provided in the next section. Applications for some
well-known inequalities are given as well.
2. The Median Principle

Consider the class of polynomials
P® := {Pn : Pn (x) = xn + aixn~1 + a2xn~2 H------- F an_]X + an, e R} .
The following result, which will be called the “Median Principle", holds.
Theorem 2.1 Let f: [a, b] C R —> R be a function so that f^n~^ is abso

lutely continuous and f^ e [a, b\. Assume that the following inequality
holds:
L (f, /(1), • • •, /(n); a, b)<R (||/W ||[o &] ; a, bj , (2.1)
where ; a, b) : R(n+1) —> R is a general function, R : [0, oo) —> R

and R is monotonic nondecreasing on [0, oo).
If g: [a, b] —> R is such that g^n~^ is absolutely continuous and
—oo < 7 < g(n~^ (z) < T < oo for a.e. x E [a, b], (2.2)
then one has the inequality
sup L (Q ~ + rP +r
bap ±7 i y rniy rn ,
PnGpo \ 2 2
. . . , 5(n-l) _ 1+Tp^n-l) (n) _ 1±I ■ a f\
' J Q 71 'J C) > > I
Proof: LetP„ € Pn^ and define/: [a, 6] —> R,/(x) = g(x) — (x),
where g satisfies (2.2).
Obviously, e Zoo [a, and
(x) = 9I”> (x) - (x) = 9<"> (x) -
Also,
giving
l|f(n)|l < r~7
II ll[a,b],oo 2
Using the inequality (2.1) and the monotonicity of R, we deduce the desired
inequality (2.3). □
Remark 2.1 Similar results may be obtained if the function L or/and R de
pend on other functions h, I, etc.
The example provided in the next sections will show how the “Median Prin
ciple” works in practice.
3. Inequalities of the Oth-Degree

An inequality of the form
Z(/l;a,6)<p(||/r||M)OO;a,6), (3.1)
i.e., no derivatives of the function f are involved, is said to be of the Qth degree.
For example, the following inequality:
rb rb
I h(x)l(x)dx <\\h\\[a<b]oo \l(x)\dx, (3.2)
a Ja
provided h e [a, b] and I e L± [a, 6], and
•6 b
h (t) du (t) (3.3)
a
provided h G C [a, 6] (the class of continuous functions) and u G BV [a, 6]

(the class of functions of bounded variation), are inequalities of Ot/l-degree. As
a generalisation of (3.3), if h, I G C [a, 6] and u G BV [a, 6], then also
b
h (a?) I (a;) du (a;) < IMm.U'IIm.oo V«- (3.4)
a
Here and in (3.3), \/b

a (u) denotes the total variation of u in [a, b\.
The following result holds.
Theorem 3.1 Let f, I: [a, b] —> R be such that there exists the constants
m,M E R with the property
—oo < m < f (x) < M < oo for a.e. x e [a, b], (3.5)
and I e Li [a, b], such that
f I (x) dx = 0. (3.6)
Ja
Then we have the inequality
f f(x)l(x)dx — f |Z(x)|dx. (3.7)

Ja Ja
The constant 1/2 is sharp.
PROOF: Using the “Median Principle" for the inequality (3.2) we have
fh ( m + M\ 1 rb
J (x)------- 2—) l(x)dx < -(M -m) J \l (a?) | dx (3.8)
and since fb I (a:) dx = 0, we deduce (3.7).

Now, assume that (3.7) holds with a constant C > 0, i.e.,
< C (M — m) f |Z(x)|dx.
/ (z) I (x) dx (3.9)
Ja
If we choose f = I and f: [a, b] —> R, where
-1, x e [a,
1, xe (^,b
Then
m = —1, M = 1,
and thus, by (3.9) we deduce C > 1/2, proving the theorem. □

Corollary 3.1 Let f, g: [a, 6] —> R be such that f satisfies (3.5) and g €
Li [a, 6]. Then we have the inequalities
1 1 1
b—a b—a b—a
b 1 fb
9 (z) - 7----- / g (y) dy dx. (3.10)
°-aJa
The constant 1/2 is sharp in (3.10).
PROOF: Follows by (3.7) on choosing I (x) = g (x} - g (y) dy. □
Remark 3.1 The inequality (3.6) was proved in a different, more complicated,
way in [4]. Generalisations for abstract Lebesgue integrals, the weighted and
discrete cases were obtained in [1].
The following result also holds.
Theorem 3.2 Let f: [a, 6] —> R be a continuous function with the property
that
—oo < m < f (x) < M < oo for a.e. x € [a,fe], (3-11)
and u 6 BV [a, 6] with the property that
u(a) = u(b). (3.12)

Then \\'e have the inequality:
1 6
f (ar) du (x) (3.13)
a
Proof: Using the “Median Principle” for the inequality (3.3) we have
m + M\ 1 6
—2— ) du (x) < - (M - m) \/ (u). (3.14)
a
* du (x) = u(b) — u (a) = 0, from (3.14) we deduce (3.13).

Since jj
Now, assume that the inequality (3.13) holds with a constant D > 0. That
is,
b
x) du (a:) < D (M - m) \/ (u). (3.15)
a
Consider a = 0, b = 1, f: [0,1] —> R, / (a;) = x, and u: [0,1] —> R given by
0 if x = 0 or x = 1.
u (a;) =
1 if are (0,1).
We have
[ f(x)du(x) = u(x)f(x) |J - [ u(x)df(x)

do J0
= — I u (z) dx = —1,
Jo
M = 1, m = 0
and
b
V (”) =2
a
Then, by (3.15) we deduce 2D > 1 giving D > 1/2, and the theorem is thus
proved. □
Another result generalizing the above ones also holds.

Theorem 3.3 Let f,l: [a, b] —> R be continuous and f such that the condition
(3.11) holds. Ifu e BV ([a, 6]) is such that
(z) du (z) (3.16)
then we have the inequality:
1 b
f (z) I (x) du (x) (3.17)
a
The constant 1/2 in (3AT) is sharp.
PROOF: Follows by the “Median Principle” applied for the inequality (3.4).
The sharpness of the constant follows by Theorem 3.2 on choosing I = 1. □
As a corollary of the above result, we may state the following Gruss type
inequality.
Corollary 3.2 Let f,g G C [a, 6] and f such that (3.11) holds. Ifu G BV [a, 6]
and u(b') u (a), then one has the inequality
1 fb
—7^------- / f (x)g(x)du(x)
u(b) — u (a) Ja
If6 1 fb
------ 77X------- TX / f du (x) • —TTT-------- / g (x) du (x)
u(b) — u(a) Ja J ' v 7 u(b) — u(a) Ja 7 V7
1,., x 1
< - (M — m) 7—-r------- —
2 ' |u (b) — u (a) |
The constant 1/2 is sharp in (3.18).
PROOF: We choose in Theorem 3.3,1: [a, 6] —> R,
1 fb
iW = sW~
Then, obviously
f I (x) du (rr) = 0,
Ja
and by (3.17) we deduce (3.18).
To prove the sharpness of the constant 1/2 in (3.18), we assume that it holds
with a constant C > 0, i.e.,
V~7 \ f f(x)g(x)du(x)
/.<■ 1*
u(b)-u (a) Ja
------ —rx [ f du ■ 1 fb
U (b) - U (a) Ja u(b)-u (a) Ja
<C(M — m) 1—
|u (6) -«(a)|
b
i rb
V(^- (3-19)
$ u(b) — u (a)
[a,b],oo a
Let us choose f = g, f: [a, 6] —> R, with f (i) = t and u: [a, b] —> R given
by
-1 if t = a
u (i) = < 0 if t 6 (a, b),
1 if t = b.
Then
1 rb i rb
------------- jr- / f (ar) g (x) du (x) = - / u2du (x)
u (b) - u (a) Ja 2 Ja
b
xu(x) dx
b2 + a2
2 ’
1 fb . a+b b-a
9------7k\------ r\ / 9(s)ds = sup x------ — =——,
*1/(b) U (tt) Ja xG[a,6] 2
b
V («) =2’ M = b, m=a
a
Functional Equations, Inequalities and Applications
and thus, by (3.19), we get
giving C > 1/2, and the corollary is proved.
For other results of this type see [6].
4. Inequalities of the lst-Degree

An inequality that contains at most the first derivative of the involved func
tions will be called an inequality of the lst-degree.
For example, Ostrowski’s inequality
1 rb
h (rr) — ------- / h (i) dt
b ~ a Ja
1 (x-
-+ ----- — llh/||[a,6],oo(6-°)’ ®€[a,6]; (4.1)
4 \ b—a
provided h is absolutely continuous and hf e Lqo [a, b], is such an inequality,

Also, the generalised-trapezoid inequality:
b—a
1 (x-9^
- + --------— IMIm.oo - °) > (4.2)
4 \ b—a
provided h is absolutely continuous, h' € [a, b], is another example of such

an inequality.
In both the inequalities above, the constant 1/4 is sharp in the sense that it
cannot be replaced by a smaller constant.
If one would like examples of such inequalities for two functions, the fol
lowing Ostrowski’s inequality obtained in [7] is the most suitable
------- / h (x) I (x) dx — •;------ / h (x) dx • ------- I I (x) dx

b-aja b — aja b-aja
< -m) (4.3)
provided —oo < m < h (rr) < M < oo for a.e. x G [a, 6], and I is absolutely
continuous and such that I' e [a, b]. The constant 1/8 is sharp.
Another example of such an inequality is the Cebysev one
------- I h (x) I (x) dx — ------- I h (x) dx ■ ------- / I (x) dx

b — aja b — aja b-aja
<44)
provided h,l are absolutely continuous and hfl' e Loo [&,&]. The constant
1/12 here is sharp.
The following perturbed version of Ostrowski’s inequality holds.
Theorem 4.1 Let f: [a, b] —> R be an absolutely continuous function on [a, b]

such that
— oo < 7 < f' (rr) < T < oo for a.e. x e [a, b]. (4.5)
Then one has the inequality:
i rb
f 0) - 7—- / f (0 dt -
b — a Ja
1
(r-7)(b-a) (4.6)
~ 2 4 \ b—a
for any x e [a, b].

PROOF: Consider the function h(x) = f (x) — 2^x, x e [a, 6]. Applying
the “Median Principle” for Ostrowski’s inequality, we get
which is clearly equivalent to (4.6).

The sharpness of the constant follows by the sharpness of Ostrowski’s in-
equality on choosing 7 = - ||f'||[O)6]>0O and T = ||/'||[0)b])00. We omit the
details. □
Remark 4.1 For a different proof of this fact, see [5].

Now, we may give a perturbed version of the generalised trapezoid inequal

ity (4.2) as well (see also [5]).
Theorem 4.2 Let f be as in Theorem 4.1. Then one has the inequality
for any x e [a, b\.

PROOF: The proof follows by the inequality (4.2) and we omit the details. □
Now, we are able to point out the following perturbation of the second Os
trowski’s inequality (4.3).
Theorem 4.3 Let f: [a, b] —> R be an absolutely continuous function on [a, b]

such that the derivative /': [a, b] —> R satisfies the condition
— oo < 7 < ff (x) < T < oo for a.e. x e [a, 6] . (4.7)
If g: [a, b] —> R is such that
— oo < m < g (x) < M < oq for a.e. x e [a, b], (4.8)
then we have the inequality:
< L (6 _ a) (M - m) (r - 7) (4.9)
The constant 1/16 is best possible.
PROOF: Consider h(x) = f (z) — Applying the “Median Principle”

for the Ostrowski’s inequality (4.3), we have:
1
g (z) dx
b—a
(4.10)
that after some elementary computations is equivalent to (4.9).

The sharpness of the constant 1/16 follows by the fact that the constant 1/8
is sharp in (4.3) on taking 7 = - ||/'||M>OO and T = ||/'||[O)b])OO. We omit the
details. □
5. Inequalities of the nth-Degree

In [2], the authors proved the following identity:
provided that f(n is absolutely continuous on [a, b] and that the kernel
Kn: [a, b]2 —> R is given by
(t - <
if a <t < x <b,
n!
Kn (x, t) := < (5.2)
(t - b)n
if a < x < t <b.
n\
Using the representation (5.1), they proved the following inequality:
/(fc)(x)
5 Ill’ll (5.3)
for any x G [a, b], and in particular, for x =
n—1
l + (-l)fc (b - a)fc+1
f wdt - 52
(k + 1)! 2fc+i
A:=0
< 7T r1 A', ll/(n)ll (b ~ a)n+1 • (5-4)

“ 2n (n + 1)! Ir lloov ’
The following result holds:
Theorem 5.1 Let f: [a, b] C R —> R be a function such that the derivative
f(n~^ is absolutely continuous on [a, b] and there exists the constants 7„, Tn G
R so that
—oo < 7n < f^ (i) < r„ < oo for a.e. t G [a, b]. (5.5)
Then we have the inequality
Proof: Observe, by (5.1), that we have
1 [ P (i - a)n dt + [b (i - b)n dt
(5.7)
a Jx
(x - q)n+1 -(x- b)n+1
(n + 1)1
(x - a)n+1 + (-l)n+1 (b - x)n+1
(n + 1)!
Taking the modulus in (5.1) and using the fact that
4" In Fn Tn
for a.e. t G [a, b]
2 2
and
we easily deduce (5.6).
Corollary 5.1 With the assumptions in Theorem 5.1, one has the inequality:
n— 1
l + (-l)fc
f wdt - 22
(fc + 1)!
k=0
n rn + 7n [l + (-ir+1l (b-a)n+1
2 (n + 1)! 2n+1
(5-8>
In [3], the authors also obtained the following identity:
[ f CO dt
Ja
= £ [(I ~0)1+1 /W <o) + <-1)‘ l'b ~ o)‘+1 fti>
k=0 ' "T '*

4-1 [\x-t)nf{n)(t)dt, (5.9)
n'. ./„
provided f^n is absolutely continuous on [a, 6].

By the use of this identity, they obtained the inequality
rb
f (t) dt
1
-E (TTTh [(i “ a^+'(o) + D6 (6 -1)‘+1 w]
< M [( *
-< +1 + (6-
r
* +1l , (5-10)
(n + 1;! II lloo L J
for any x e [a, b].

In particular, for x = we get the inequality:
/ f (0 * - n
£~^ (fcTI)!
1 (“) [r/W (0) + (-1)‘f<t> (6)]-I
< » / 1 "7T7 ll/(n)|| (b ~ «)n+1 • (5-11)

- 2n (n + 1)! Ir Ilex? 7
Finally, we may state the following result:
Theorem 5.2 With the assumptions in Theorem 5.1, we have the inequality
n—1 \b - x)k+r + (-l)fc (x-a)M

f (*) dt ~ 22
(fc + 1)!
k=0
1
(n. + 1)!
(5.12)
for any x € [a, &].

In particular, for x — we have the corollary
Corollary 5.2 With the above assumptions, we have
The interested reader may find many other examples which can be treated
in a similar fashion. We omit the details.
References
[1] P. Cerone and S.S. Dragomir: ‘A Refinement of the Gruss Inequality and
Applications’, RGMIA Res. Rep. Coll. 5 No. 2 (2002), Article 14. [online:
http: //rgmia.vu.edu.au/v5n2 .html].
[2] P. Cerone, S.S. Dragomir and J. Roumeliotis: ‘Some Ostrowski type in
equalities for n-time differentiable mappings and applications’, Demon-
stratio Mathematica 32 No. 2 (1999), 697-712.
[3] P. Cerone, S.S. Dragomir, J. Roumeliotis and J. Sunde: ‘A new gener
alistion of the trpezoid formula for n-time differentiable mappings and
applications’, Demonstratio Mathematica 33 No. 4 (2000), 719-736.
[4] X.-L. Cheng and J. Sun: ‘Note on the perturbed trapezoid inequal
ity’, J. Ineq. Pure. & Appl. Math. 3 No. 2 (2002), Article 29. [online:
http: //jipam.vu.edu.au/v3n2/046_01 .html]
[5] S.S. Dragomir: ‘Improvements of Ostrowski and Generalised Trapezoid
Inequality in Terms of the Upper and Lower Bounds of the First Deriva
tive’, RGMIA Res. Rep. Coll. 5 (2002), Supplement, Article 10. [online:
http: / /rgmia. vu. edu. au/v5 (E) . html]
[6] S.S. Dragomir: ‘Sharp bounds of Cebysev functional for Stieltjes inte
grals and application’ (in preparation).
[7] A. Ostrowski: ‘On an integral inequality’, Aequat. Math. 4 (1970), 358-
373.
Chapter 4
ON THE HYERS-ULAM-RASSIAS
STABILITY OF A PEXIDERIZED
QUADRATIC EQUATION II*
Kil-Woung Jun
Taejon 305-764, Republic of Korea
kwjun@math.chungnam.ac.kr
Yang-Hi Lee
Department of Mathematics Education, Kongju National University of Education
Kongju 314-711, Republic of Korea
Iyhmzi@kongjuw2.kongju-e.ac.kr
Abstract In this paper we prove a generalization of the stability of the Pexiderized quadratic
equations fi(x + y + z) + f2(x) + fa(y) + fa(z) - fa(x + y) - fa(y + z)-
f7(x + z) = 0 and fa(x + y +z) + fa(x— y +z) + fa(x+ y — z) + fa(—x +
y + z) — 4/5(x) — 4/e(t/) — 4f?(z) = 0 in the spirit of D.H. Hyers, S.M. Ulam,
Th.M. Rassias and P. Ghvruta.
2000 MSC: 39B72, 47H15 (Primary).
Keywords: Quadratic function, Pexiderized quadratic equation
1. Introduction
In 1940, S.M. Ulam [32] raised the following question: Under what con
ditions does there exist an additive mapping near an approximately additive
mapping?
*This work was supported by grant No. RO1-2000-000-00005-0(2002) from the KOSEF
39
In 1941, D.H. Hyers [5] proved that if f: V —> X is a mapping satisfying
||/(a: + j/) —/(a?) —/(y)|| <6
for all x, y e V, where V and X are Banach spaces and J is a given positive
number, then there exists a unique additive mapping T: V —> X such that
\\f(x)—T(x)\\<6
for all x G V. In 1978, Th.M. Rassias [22] gave a significant generalization

of the Hyers’ result. Th.M. Rassias [28] during the 27th International Sym
posium on Functional Equations, which took place in Bielsko-Biala, Poland,
in 1990, asked the question whether such a theorem can also be proved for
a more general setting. Z. Gadja [3], following Th.M. Rassias’s approach (
[22]) gave an affirmative solution to the question. Recently, P. Gavruta [4]
also obtained a further generalization of the Hyers-Rassias theorem (see also
[6]-[l 1], [15], [17], [20], [21], [23]—[25], [28], [29]). Lee and Jun [18], [19]
also obtained the Hyers-Ulam-Rassias stability of the Pexider equation of
f(x + y) = g(x) + h(y) (see also [12]).
Throughout this paper, let V and X be a normed space and a Banach space,
respectively.
In 1983, the stability theorem for the quadratic functional equation
f(x + y) + f (x -y)~ 2f(x) - 2/(y) = 0
was proved by F. Skof [31] for the function f: V —> X. In 1984, P.W. Cholewa
[1] extended V to the case of an Abelian group G in the Skof’s result.
In 1992, S. Czerwik [2] gave a generalization of the Skof-Cholewa’s result
in the following way:
Theorem 1.1 Let p 2, 0 > 0 be real numbers. Suppose that the function
f: V —> X satisfies
\\f(x + y) + f(x — y) — 2/(x) - 2/(3/)|| < 0(||< + ||<).
Then there exists exactly one quadratic function g: V —> X such that
||/(t) -y(x)|| < c+fc0||z||p
for all x in V ifp > 0 and for all x G V \ {0} ifp < 0, where: when p < 2,
c — ||/(0)||/3, k = 2/(4 — 2P) and g(x) = limn_oo4-n/(2"x) for all x in
V. When p > 2, c = 0, k = 2/(2P - 4) and g(x) — limnôo 4n/(x/2n)/or
all x in V. Also, if the mapping t —> f(tx)from Rte X is continuous for each
fixed x in V, then g(tx) = t^g(x) for all t in R.
Hyers-Ulam-Rassias stability of a Pexiderized quadratic equation 41
Since then, the stability problem of the quadratic equation has been extensively
investigated by a number of mathematicians ([13], [26], [27], [30]).
Jun and Lee [14] obtained the following stability theorems for the Pexider
ized quadratic functional equation
f(x + y) + g(x — y) — 2h(x) - 2k(y) = 0
for the function f,g,h,k: V —> X:

Let <p: V \ {0} x V \ {0} —> X be a function such that
°° 1
p(x, y) = 22 7z+i < 00 (a)
i=o 4
i / x y \
E4 <o° (a’)
and
(b)
i / x y \
E (^+1,^71) <o° (b’)
for all x, y e V \ {0}. Let
5x 3x\ (3x bx
+ft (2x, —2x) + ft(2x, 2x) + 2ft (2z, —x) + 2ft(x, 2x)
f3x 3x\ f3x 3x\
\ ~ (x ~ ( —x y\ „ rx y\
M2(<p,x,y) = <pl-,— )+ X satisfy
\\f(x+y)+g(x — y) — 2h(x) — 2k(y)\\ <<p(x,y) for all x,y € V \ {0}.
Then there exist exactly one quadratic Junction Q: V —> X and two unique
additive functions T,T': V —> X such that
WfW —/(0) — Q(x) — T(x) ||

Mi(<p,x) +-x) + M2(yp,x,x) + M2(<p, -x, -x)
2
||g(x)-g(0)-Q(x)-T'(x)||
Mîp, x) + Mi(y, -x) + M2(p, x, -x) + M2(<p, -x, x)
2
h(x) + fc(x) /(0) + g(0) T(x)
------- 2------------------- 4---------- " —
2x) 4- —2x)
8
( M2(<p, 2x, 2x) + M2(p, 2x, —2x)
*■ 8
M2(ip, —2x, 2x) + M2(<p, —2x, —2x)
8
+ x) + (p(x, —x) + _v(~X') x) + </?( —£, — x)
for all x EV \ {0}. The function Q is given by
' limn_>oo = lim„_oo ^2^1 if satisfies (a),

Q(x) = limôo (/ + f (-^r) - 2/(0))
k = limn_oo^ (p(^) +g (-#■) - 2g(0)) if Oo X^2.3^~3 if T satisfies (b)

T(x) =
limnôo 3" (/ (^-) - /(0)) ifoo 23n
limnôo 3n (5 (^r) - y(0)) if
T(x) + T'(x) = (limôo fc(3n*
)-g~ 3n*
)
if 
T(x) - T'(x) = f limn_).oo fe(3ni)~3fcn(~3ni) ifip satisfies (b)
2 (limôo (k (^-) - k (-^)) if satisfies (b’)
for all x eV.
Throughout the rest of this section, let ip: V x V —> [0,00) be a function such
that
00 1
<p(x, y) = ^+1 2^) < 00 (c)
1=0 4
or
00
p(x> y) =E (2S1’ < 00 (c>)
1=0
and
00 1
2/) = 52 ^(2^, 2zy) < 00 (d)
1=0 2
or
00
^y) = 522^(2î’^T) <o° (d>)
1=0
for all x, y e V.
Let
M3((p, x) = | ^(x, x) + 2<^(0, x) + 2<^(x, 0)
+^(x,-x) + 2<^(0,0) + 99^,^) ’
x, y) = (p[x, 0) + £(0, y) + ^(x, y),
Ms(<p, x) = -[<£(x, -x) + 2<£(x, 0) + <£(2x, -x) + <^(-x, x)
+<^(0, x) + p(—x, 2x) + <^(x, x) + <^(2x, x)

+<p(—x, —x) + (£(0, —x) + <p(-x, -2x)]
for all x, y e V.
The following theorem is also shown in [13].
Theorem 1.3 Let X satisfy
\\f(x + y) + y(rr - y) - 2h(x) - 2fc(y)|| < <p(x,y) for all x,y eV.
Then there exist exactly one quadratic function Q : V X and two unique
additive functions T^T': V —» X such that
||/(z)-/(0)-Q(z)-T(z)||
< M3(<p, x) + M3(<p, -x) + M4(9?, x, x) + M4(<p,-x, -x)
2
\\g(x) — y(0) -Q(z)
< M3(<p, x) + M3(<p, -x) + M4(^, x, -x) + M4(^, -x, x)
2
h(x) - fc(0) - Q(x) - ±(T(x) + T'(x))
< M3(<p, x) + M3(yp, -x) + M5(y>, x) + A45(^, -x)

2
+ _ 0) +
0) + 2y?(0, x, 0) ,
k^-k^-Q^-^T^-r(x^
M3(if>,x) + M3(ip, -x) + M5{<p,x} + M5(<p,-x)

2
। y(0, x) + 2y(0,0) + 9?(0, -x)
for all x e V. The function Q is given by
liriinôo 9{\nX>> if <p satisfies (c)

Q(x) = lim.l_oo^(/(^)+ /(-#)- 2/(0))
= limn_oo f (5 (#) + <?(-#)- 25(0)) ifoo Qn+l ? ifp satisfies (d)
limn_+0O2n(y(2-na:) - 5(0)), if(p satisfies (d’)
In this paper, we prove the stability of the Pexiderized quadratic equations:
fi(x + y + z) + /2(x) + /s(y) +

- fs(x + y) - fe(y + z) - f7(x + z) =0, (1.1)
fi(x + y + z) + /2(x -y + z)+ f3(x + y- z) + fa(-x + y + z)

- 4/5(x) - 4/6(y) - 4/7(z) = 0. (1.2)
2. Stability of the equation (1.1)

In this section, we denote by : V \ {0} x V \ {0} x V \ {0} —> [0, oo) a
function such that
OO |
<p(.x, y,z) = ^2 7th <p(2lx, 2ly, 2lz) < oo (e)

i=o 4
or
oo
rtx,y,z) = <00 (e’)

1=0
and
OO |
9S(z, y, z) = 22 3Z+1 ¥’(3^, 3'y, 3zz) <00 (f)

l=o 6
or
00
^(x, y, z) = 22 3 V (3Z+1 > 3Z+1 3Z+1) < 00 (f )
l=o
for all or, y, z G V \ {0}.
Theorem 2.1 Suppose that the functions fi: V —> X, i = 1,2,..., 7, satisfy
\\fl(x+y+z)+f2(x}+f3(y)+f^z)-f^x+y)-fG{y+z)~ f7{x-\-z)\\
<ip(x,y,z) (2.1)
for all x,y, z G V \ {0}.

Then there exist exactly one quadratic function Q.V—+X and three addi
tive functions : V —> X satisfying
||/5(
)-/
* 5(0)-Q(z)-Ti(z)||
M^,x) + Mi(^,-x) + M2(^,x,x) + M2(fi>,-x,-x)
, (2.2)
2
||/6(a;)-/6(0)-Q(a;)-T2(T)||
Mi(fi>,x) + Mi^,-x) + M2(fi>,x,-x) + M2(fi>,-x,x)
, (Lj)
2
\\f7(x) — f7(O) — Q(x) — T3(x)\\
Mi(r, x) + Mi(r, -x) + M2(t, x, -rr) + M2(t, -x, x)
(2.4)
2
for all x E V \ {0}, where V \ {0} x V \ {0} —> X are defined by
tl?(x,y) = <p(x,y,—x), r(x,y) =<p(x,y,-y) for all x,y € V \ {0}.
Moreover, the function Q is given by
fj(2nz)+M-2nx)
>oo 2.4^ if p satisfies (e),
(2.5)
ifp satisfies (e’)
for i = 1,5,6,7 and for allx eV and the functions Tk (k — 1,2,3) are given
by
T z x = f limn—»oo /fc+4(3nx)2 3^+4( 3"^ ifp> satisfies (f),

k X [limnôo (/fc+4 (^r) - /fc+4 (^r)) ifp> satisfies (f’)
for all x eV.
PROOF: Assume that satisfies (e) and (f). Replacing z by — x in (2.1), we

obtain
||/1(3/) + foW + /s(y) + ~ + y)

- - x) - /7(0)|| < <p(x, y, -x)
for all x, y E V \ {0}. Define f,g,h,k: V —> X by
/(^) = fs(xfi
g(x) = fe(-x) + /7(0),
2h(x) = f2(x) + fi(-x)
and
2fc(x) = falx) + f3(x)
for all x e V \ {0}. Then we get
\\f(x + y) + g(x — y) — 2h(x) - 2fc(y)\\<^(x,y)
for all x, y e V \ {0}. By Theorem 1.2 there exist exactly one quadratic func
tion Q: V —> X and two unique additive functions 7i, : V —> X satisfying
(2.2) and (2.3). The function Q is given by
and 7i, ^2 are given by (2.6). Replacing z by — y in (2.1), we obtain
\\fl(.x)+f2(x)+f3(y)+f4(-y)-f5(x+y)-f6(Q)-f7(x-y)\\ < <p(x,y,-y)
for all x, y 6 V \ {0}. Define g', h',k': V —> X by
f'(x) = f5(x),
g'W = y6(o) + /7(z),
2/r'(x) = /i(t) + /2(x)
and
2k'(x) = f3(x) + f4(-x)
for all x e V \ {0}. Then we get
\\f'(x + y) + g'(x — y) — 2h'(x) - 2fc'(y)|| < r(x, y)
for all x e V \ {0}. From Theorem 1.2, there exist exactly one quadratic func
tion Q: V —> X and two unique additive functions T), T3: V —> X satisfying
(2.2) and (2.4). The quadratic function Q is given by
and 7i, T3 are given by (2.6).

Replacing y and z by x and 2x, respectively in (2.1), we obtain
||/i(4t) + /2(^) + + /4(2x) - - f&{3x) - f7(3x)||

< p(x,x,2x) (2.9)
for all z € V \ {0}. Replacing y and z by x and — 2x, respectively in (2.1), we

obtain
11/1(0) + /2(x) + + /4(—2ar) - /5(2x) - /6(-ar) - /7(—rr)||

< p{x, x, — 2rr) (2.10)
for all x E V \ {0}. From (2.9) and (2.10), we obtain
||/1 (4rc) - /i(0) + /4(2x) - /4(-2x) - /6(3x) + fe(-x)

-f7(3x) + /7(—rr)||
< <p(x, x, 2x) + <p(x, x, —2x)
H/1 (-4x) - /i(0) + /4(—2z) - /4(2x) - /6(-3x) + /6(x)
-/7(-3t) + /7(a:)||
< ip(—x, —x, —2x) + ip(—x, —x, 2x)
for all x e V \ {0}. Hence, we have
||/i(4z) + /i(-4x) - 2/i(0) - /6(3x) - /6(-3x) + ftfix) + fs(-x)

—A(3ar) - /7(-3x) + /7(x) + /7(-x)||
< ip(x, x, 2x) + qfix, x, —2rr) + <p(—x, —x, —2or) + <^(—x, —x, 2x)
From (2.7), (2.8) and the above inequality, we have
lim /,(2^x) + /,(-2^) = 32O(I).

n—>oo 4n
for all x € V. From (2.7), (2.8) and the above equality, we obtain (2.5). By a
similar method, we obtain the rest of the proof. □
Corollary 2.1 Let p / 1,2 and 0 > Obe fixed real numbers. Suppose that the
functions fa: V —> X, i = 1,2,..., 7, satisfy
\\h(x+y+z')+f2(x}+f^y)+f^z)-f5(x+y)-f&(y+z)-f7(x+z)\\
< 0(ll< + IK + HO
for all x,y, z e V \ {0}.

Then there exist exactly one quadratic function Q : V —» X and three addi
tive functions Tk '. V —> X (k = 1,2,3) satisfying
||/fc+4(^) - /fc+4(0) - Q(x) - Tk(x)II

/3-5p + 6-4p + 6-3p + 6-2p + 6 3 6(3 + 3^) \ ... .._
- (---------------- 2P[4--2Pj---------------- + 2? + 2p(3 - 3-)) ’
for all x e V \ {Q}.

~ - £(0)) ifp > 2
for i = 1, 5,6, 7 and for all x eV and the functions (k = 1,2,3) are given
by
A+4(3nx)-A+4(-3nx)
limn_+oo 2-3n ifP < 1,
Tk(x) =
limn_>00 V (/fc+4 (^t) — /fc+4 (3^)) ifp > 1
for all x 6 V.
Throughout the rest of this section, we denote by <p: V x V x V —> [0, oo) a
function such that
00 1
<^(z, y, 2) = _^(2'x, 2ly, 2lz) < 00 (g)
1=0
or
00
\ Al ( x x z \
= <o°
1=0
and
00
y, z) = 22^-<p(2lx, 2ly,2lz) < 00 (h)
1=0 Z
or
00
= 2^12^) <°o (h’)
1=0
for all x^y^z e V.
Theorem 2.2 Suppose that the functions f: V —» X, i = 1,2,..., 7, satisfy
\\fi(x+y+z)+f2(x)+f3(y)+f4(z)-f5(x+y')-f6(y+z)-f7(x+z')\\
<p(x,y,z) (2.11)
for all x,y,zE V.

Then there exist exactly one quadratic function Q: V —> X and four addi
tive functions V —> X satisfying
\\fs(x) — fs(ty — Q(x) — Ti(x)\\

M3(i/>, x) + M3(i/>, -x) + x, x) + -x, -x)
2
(2.12)
||/6(» -/6(0)-Q(x)- T2(x)||
< M3(f>, x) + M3(f>, -x) + Mfy-f, x, -x) + -x, x)
_ 2 ’
(2.13)
\\f7(x) — f7(fy) — Q(x) — T3(x)\\
M3(t,x) + M3(t, -x) + Mfyr, x, -x) + Mfyr, -x,x)
— 2 ’
\\fy(x)-fy^-Q^-T^x)-Tfyx)\\
M3(ji, x) + M3(jj,, —x) + Mfyp,, x, x) + Mfyp,, —x, —x)
< _
x) + M3(f>, —x) + Mfy-ip, x, x) + —x, —x)

+ _ ,
(2.15)
||/2(a') - /2(0) - Q(x) - T3(x) + T4(x)||
< M3(/i, x) + M3(p,, -x) + Mstp,, x) + -rr)
+ |(¥’(a;,0,0) +299(0,0,0) + 9?(-x,0,0))
M3(t, re) + M3(t, —x) + Mfyr, x, -x) + Mfyr, —x, x)

+ _ ,
(2.16)
||/30) - /3(0) - Q(x) - T2(x) + Tfyx) ||
< M3{h, x) + M3(/x, -x) + M5(/x, x) + M5(/x, -x)
+ |(<X°> °) + x, 0))
°> °) + ^(0, ~*
f M3(f>, x) + M3(f>, — x) + x, —x) + Mfyip, —x, x)

+ ,
(2-17)
||/4(x) - /4(0) - Q(x) + n(x) - T2(x) - T3(x) + T4(x)||
< M3{f), x) + M3(f>, -x) + M5(^, x) + M5(^, -x)
+ |<X°> x> °) + ^(0, ~x> °)
+2^(0,0,0) + ^(x, 0, -x) + |</?(-x, 0, x)
+Af3(/z, x) + M3(/l, —x) + Mg(/z, x) + Mstja, —x)

, M3(t, x) + M3(t, -x) + M4(t, x, -x) + M4(t, -x, x)
_|-------------------------------------- -------------------------------------- ( Z. 16 )
2
for all x eV, where 'i/>,T,y,: V \ {0} x V \ {0} —> X are defined by
'4’(x,y) = <p(x,y,-x), r(x,y) = <p(x,y,-y), yfx,y) = <p(x,y,G)

for all x,y eV.
limnôo if oo /fc+4'2 ^>n~4ifc't'4(2 X) tfv satisfies (h),

7fc(x) = <
limnôo 2n-1 (/fc+4 (^) - /fc+4 (=£)) ifoo 2n+1
if (p satisfies (h)
limôo 2-1 (A (^) - A (^) - f5 (^) + f5 (^))
if ip satisfies (h’)
(2.21)
for all x eV.
PROOF: Assume that satisfies (g) and (h). Applying Theorem 1.3, we obtain
the functions Q and satisfying (2.12), (2.13), (2.14) in the same
way as in Theorem 2.1. Also the functions Q, Ti, : V —> X satisfy the
inequalities
||/2(x) + /4(-x) - /2(0) - /4(0) - 2Q(x) - Tx(x) + T2(x)||

< x) + M3{tp, -x) + x) + M5(V>, ~x)
+ |(<^(x, 0, -x) + 2</?(0,0,0) + 9?(-x, 0, x)) (2.22)
for all x € V and Q, are given by (2.19) and (2.20). Replacing z by

0 in (2.11), we obtain
||/i(x+?/)+/2(2:) + /3(y) + /4(0)-/5(^+?/)-/6(y)-/7(^)|| < <f(x,y,0)
for all x, y € V. Define f", g", h", k": V —> X by
f"(x) = /i(x) - f5(x),

g"(x) = A(0),
2h"(x) = -fa(x) + /7(x)
and
2F(x) = -fa(x) + /6(x)
for all x 6 V. Then we get
\\f"(x + y) + g"(x — y) — 2h"(x) - 2k"(y) || < ^x, y)
for all x e V. From Theorem 1.3, there exists one unique additive function
Tn: V —> X such that
||/1(a;)-/5(x)-A(0) + /5(0)-T4(^||
< M3(/z, x) + M3(/z, -x) + M4(/z, x, x) + M4(/z, -x, -x) 23)
2
II - /2(^) + /7(x) + /2(0) - /7(0) - T4(x)||
< A/3(y, a;) + M3(/z, -x) + Af5(/Z, x) + M5(/z, -x)
0,0) + 2<X0,0,0) + 0,0)) (2.24)
|| - f3(x) + /6(x) + /3(0) - /6(0) - Tn{x) ||

< M3(/x, x) + M3(/x,-x) + M5(/x, x) + M5(/z,-x)
+|(<^(0, x, 0) + 2<^(0,0,0) + ¥>(0, -x, 0)) (2.25)
for all x e V and the function T4 is given by (2.21). We get (2.15) from (2.12),
(2.23) and the inequality
(x)-T4(x)\\
< *
)
||/i( - fs(x) - A(0) + /5(0) - T4(z)||
+ ll/5^)-/5(0)-Q(a;)-T1(x)||
for all x eV. We get (2.16) from (2.14), (2.24) and the inequality
||/2(^)-/2(0)-Q(x)-73(;c)+T4(x)||
< 11/2(2;) - f7(x) - /2(0) + /7(0) + 74(a;)||

+||/7(rr)-/7(0)-Q(2;)-713(2;)||
for all x e V. We get (2.17) from (2.13), (2.25) and the inequality
||/3(x) - /3(0) - Q(x) - T2(x) + T4(2r)||

< ||/3(x) - /6(x) - /3(0) + /6(0) + T4(x)||
+ ll/6(rr)-/6(0)-Q(2;)-T2(2;)||
for all x eV. We get (2.18) from (2.16), (2.22) and the inequality
||/4(z) - /4(0) - Q(x) + 7\(x) - T2(x) - T3(x) + T4(x)||

< Wh(-x) + /4(x) - /2(0) - /4(0) - 2Q(-x) - T^-x) - T2(x)||
for all x eV. From the above inequalities, we easily get
for i = 1,2,3,4 and for all x e V. □
Corollary 2.2 Let p f 1,2, 0 > 0 be fixed real numbers. Let ip: V —> [0, oo)
be a mapping such that iptx) = ||2:||p for x 0 and V’(O) = 0 if p > 1.
Suppose that the functions fa: V —> X (i = 1,2,... ,7) satisfy
\\fi(,x+y+z)+f2(x)+f3(y)+ffyz')-f5(x+y>)-f6(y+z)-f7(x+z)\\
< ip(x) + V’(y) + 'fl’(z')
for all x,y, z € V.

tive functions Ti, 7*2, T3: V —> X satisfying
||/5(s)-/5(0)-Q(2:)-7Hz)||
\\h(x)-feW-Q(x)-T2(xfi\
S + + |2^2f!) W|P + 5V,(0)’

NAM -A(0) -QM-T3WII
s (idî + + rdW wmm
IIAM-AM-OM-AM-^MII
~ (|4 - 2?| + 2? + |2 - 2?|) + 3
IIA(x) - /2(0) - Q(x) - T3(x) + 74(s)||
Il/3(x) - /3(0) - Q(x) - T2(x) + T4(x)||

- (|4î + 2? + |2-2P| + 9 l|x||P +
11/4(x) - /4(0) - Q(x) + T^x) - T2(x) - T3(x) + T4(x)||

- (|4î 13 &26
/ 26 + + 5 ^1"'
+ 72 • 2P 109T
+\ ) H + , V'( }
for all xe V\ {0}.

(iim AM
I iAllln__>oo
' ] im 4n
I lHlln_>00 41 2
for € = 1,2,3,4, 5,6, 7 and for all x eV and the functions Tk (k = 1,2,3/
T4 are given by
lim .A+4(2_^)-A+4(-2^) f x
llLLln_>00 2n+l lJ
Tk(x) =
limn_oo 2n-1 (/fc+4 (^-) - fk+4 (=£)) ifp > 1,
i:m /i(2»i)-/i(-2"i)-/s(2"x)+/5(-2’‘t)
llLLln_>QO 2n + 1
Tfyx) = <
limn_oo2—1 (A (^)
ifP> 1
for all x € V.
The following corollary, which is due to S.M. Jung [16], can be easily ob
tained from Theorem 2.2.
Corollary 2.3 Suppose that the functions f: V —> X satisfy
A(x + y + 2)-l-/2(x)-l-/3(y) + /4(2:) = /5(x+y) + /6(y+^) + /7(x + 2:)
for all x^y^z e V. Then there exist exactly one quadratic function Q: V X
and four additive functions Ti, T2, T3, T4: V —> X such that
fy(x) = Q(x) + Ti(x) + r4(x) + A(0),

= Q(x) + T3(x) - T±(r) + /2(0),

/3(x) = Q(x) + T2(x) - T4(x) + /3(0),
fi{x) = Q(x) - Tfx) + T2(x) + T3(x) - T^x) + /4(0),
/5(x) = Q(x)+T1(x) + /5(0),
/6(a?) = Q(x) + T2(x) + /6(0),
f7(x) = Q(x) + T3(x) + f7(0)
for all x eV.
3. Stability of the equation (1.2)

Theorem 3.1 Let tp satisfy the condition (e) or (e’) and the condition (f) or
(f’). Suppose that the Junctions fa: V —> X satisfy
ll/l + y + z) + h(x -y + z) + fafix + y-z) + faf-x + y + z)

~ 4/5^) - 4/6(y) - 4/7(2:)|| < <p(x, y, z) (3.1)
for all x,y, z E V \ {0}.

tive functions TiyT2,T3yTi: V —> X such that
\\fat(x) — fafO) — Q(pc) — Ti(x)\\

< x) + -x) + Mity, x, x) + Mzlip, —x, — x)
(3.2)
||/2(2:) -/2(0)-Q(x)-T2(x)||
Mfr, x) + Mfr, —x) + M2(t, x, — x) + M2(t, —x, x)
< _ , (3-3)
||/3(x)-/3(0)-Q(x)-T3(x)||
Mfp, xj + M-fp, -x) + M2(/x, x, -x) + M2(/z, -x, x)
(3.4)
11/4(2:) _/4(0)-Q(x)-T4(x)||
x) + —x) + x, —x) + —x, x)
< _ (3.5)
for all x eV \ {0}, where ^,T,p: V \ {0} x V \ {0} —> X is defined by
2’ 2/ ’
for all x,y G V \ {0}. The function Q is given by
limnôo fi^nx>> if <p satisfies (e),

Q(x) = <
lininôo (fa (^) + fi (^-) - 2/j(0)) if<p satisfies (e’)
for i = 1,2,3,4 and for all x € V. The Junctions Tfi = 1,2,3,4) are given
by
fi(3nx)—fi(—3nx)
limn—»oo 2-3n if <p satisfies (f),
Tfx) = (3.7)
limnôo if <p satisfies (f’)
for all x eV.
PROOF: First assume that X by
f(x) = 710),
y(®) = Pi(-x),
2h(x) = -pfa) - f3(x) + 4/5(s)
and
2fc(a;)=4/6(^+4/7n
for all x e V \ {0}. Let t/fix, y) = <p(x, y/2, y/2). Then we get
\\f(x + y) + g(x — y) — 2/i(z) - 2fc(y)|| < -fi(x, y)
for all x, y e V \ {0} where ip: V \ {0} x V \ {0} —> X is defined by

ilfix, y) = (p{x, y/2, y/2) for all x,y e V\{0}. From Theorem 1.2, there exist
exactly one quadratic function Q: V —> X and two unique additive functions
71, T4: V —> X satisfying (3.2) and (3.5). Also the function Q is given by
A(2^) f4(2nx)
Q(x) = hm ——- = hm ——- (3.8)
n—>00 4n n—>00 4n
for all x e V and Ti, T4 are given by (3.7).
Replacing x and z by x/2 and x/2 in (3.1), we obtain
||/i0 + y) + fo(x - y) + /3(y) + /<(!/)
-4/5g)-4/6(y)-4/7g)||
(X x\
M2’y’2J
for all x, y € V \ {0}. Define /', g', h', k': V —> X by
/'(
*
) = A(x),
g'W = /2O),
2/i'(a;)=4/5n+477n
\ Zi / X /
and
2k'(x) = -fa(x) - f4(x) + 4/6(x)
for all x G V \ {0}. Then we get
\\f'(x + y) + g'(x — y) — 2h\x) - 2k'(y) || < r(x, y)
for all x,y G V \ {0} where r: V \ {0} x V \ {0} —> X is defined by

r(x,y) = y>(x/2, y, x/2) for all x, y G V\{0}. From Theorem 1.2, there exist
Ti, Tz: V X satisfying (3.3). Also the function Q is given by
. /i(2"x) /2(2^)
Q(x) = hm —i-—- = hm —4-—- (3.9)
n—>00 4n n—>00 4n
for all x e V and 7i, Tz are given by (3.7). Replacing y and z by x/2 and y in
(3.1), we obtain
||/i(x + y) + h(y) + - y) + A(y)
-4A(^) -4
* (I) -4/7(y)ll
(X X \
<^(2’2’^
for all x, y G V \ {0}. Define /", </', h", k": V ~^> X by
/"(x) = A(x),
/(x) = /3(x),
2/l"(I)=4/s(^)+4Ag)
2fc"(I) = -A(I)-A(I) + 4/7(a;)
for all x € V \ {0}. Then we get
||/"0 + y) + g"(x — y) — 2h"(x) - 2fc"(y)|| < M(x, y)
for all x, y € V \ {0} where pt: V \ {0} x V \ {0} —> X is defined by

p.(x, y) = <p(x, x, 2y) for all x, y G V \ {0}. From Theorem 1.2, there exist
71, T3: V —> X satisfying (3.4). Also the function Q is given by
for all x e V and are given by (3.7). From (3.8), (3.9) and (3.10), we
obtain (3.6). The rest of the proof are obtained by the similar method. □
Corollary 3.1 Letp f 1,2, 0 > Obe real numbers. Suppose that the functions
fi: V X satisfy
11/1(2: + y + z) + f2(x -y + z) + /3(rr + y - z) + fi(-x + y + z)

- 4f5(x) - 4f6(y) - 4/7(z)|| < 0(||< + \\y\\? + ||<)
for all x,y,z eV \ {0}.

Then there exist exactly one quadratic function Q : V —» X and four addi
tive functions T{\V —» X (i=l,2,3,4) such that
Wffx^-f^-Q^-T^W
10P + 2- 8P + 2- 6? + 2- 5P + 6- 4P + 4- 3?+ 6- 2?+ 4
4P|4- 2P|
4?|3 - 3p|
for all x eV \ {0}.

The function Q is given by
limn_,oo ifp < 2

Q{x} = <
lim^ f (/i (^) + f - 2/i(0)) if 2 < p,
for i — 1,2, 3,4 and for all x e V. The functions Ti (i = 1,2, 3,4) are given
by
flim^ ifP<^
Ti(x) =
limnôo^ ifp>t
for all x eV.
Theorem 3.2 Let <p satisfies the condition (g) or (g’) and the condition (h) or
(h’). Suppose that the functions f: V —» X satisfy
+ y + z) + f2(x -y + z) + f3(x + y - z) + fi(~x + y + z)

- 4f5(x) - 4f6(y) - 4f7(z)|| < <p(x, y, z) (3.11)
for all x,y,z e V.

Then there exist exactly one quadratic function Q: V X and four addi
tive functions V —> X such that
||/i(^-/i(0)-Q(x)-T1(t)||
M3 (V’, x) + M3 (^>, — x) + M4 (V’, x, x) + M4 ($,—x, —x)
< _ ,
||/2(rr)-/2(0)- Q(z)-T2(z)||
M3(t, x) + M3(t, -x) + M4(t, x, —x) + M4(t, —x, x)
< _ ,
*
||/3(
)-/3(0)-Q( z)-T3(z)||
M3(/j,, x) + M3(y,, —x) + Afaf/i, x, —x) + A/4(/z, — x, x)

< 2 ’
||/4(a;)-/4(0)-Q(T)-74^)||
x) + M3^, -x) + M4(V», x, —x) + ^(V’, -x, x)
- /5(o) - Q& -
71/3(1/, x) + 71/3(1/, -x~) + 71/5(^, x) + 71/5(1/, -x)

4
! ^(^, 0,0) + 2y(0,0,0) + y(~rr, 0,0) 12)
- /6(0) - Q(x) - r*
i(
)~ )+W)
T*
2( + 74(rr)
71/3(1/, x) + 71/3(1/, -x) + 71/5(i/, x) + 71/5(1/, —x)

4
< /p(0, x, 0) + 2y>(0,0,0) + <^(0, —x, 0)

+ 8 ’
f7(x) - /7(0) - Q(x) - fax) + T2(x) - T3(x) + T^x))
M3(7t, x) + M3(tt, —x) + x) + M3(tv, —x)

< _
। y>(0,0, x) + 2<^(0,0,0) + <^(0,0, — x)
for all x € V \ {0}, where ip,r, 11,0, tv: V \ {0} x V \ {0} —> X is defined by
y y\
x, —, — ,
. 2 2/
' 27 ’
F(x>y) = ¥’ |^2’ 2,y) ’
i/(rr,y) = <^(x,y,0),
7t(t,?/) = <p(x,0,y)
for all x,y eV \ {0}.

The function Q is given by
1- /i(2nx)
limn_-x ifcp satisfies (g)
Q(®) =
limôo f (fi (^) + fi (^) - 2/i(0)) iftp satisfies (g’)
(3.15)
for i = 1, 2,..., 7 and for all x eV. The functions Ti(i = 1, 2, 3,4) given
by
< limn_>oo /i(2n^n+i( 2 if satisfies (h),

limn_+oo 2n-1 (fi (^-) - fi (-^r)) if<p satisfies (h’)
for all x eV.
Proof: By the same method as in Theorem 3.1, we obtain the result in this
theorem with the exception of the inequalities (3.12)-(3.14) and equality (3.15)
for i = 5,6,7. Assume that satisfies (g) and (h). Replacing z by 0 in (3.11),
we obtain
Wfl(x+y)+f2(x-y)+f3(x+y)+f4(-x+y)-Af5(x)-4f6(y')-4:f7(Q')\\
< v’OmaO)
for all x, y G V. Define /', g', h',k': V —> X by
f'(x) = ZiW + /j(4

g'(x) = h(x) + fa(-x),
2h!(x) = 4/5(x)
and
2fc'(x)=4/6(x)+4/7(0)
for all x eV. Then we get
|| f'(x + y) + g'(x — y) — 2h'(x) - 2k'(y) || < v(x, y) for all x,y eV.
Applying Theorem 1.3 to the above equality, one can prove that there exist
exactly one quadratic function Q': V —> X and two unique additive functions
: —> X satisfying
2/5(*
) - 2/5(0) - Q'(x) -
< M3(v, x) + M3(v, -x) + M

>(v,x)
* + M5(y,-x)
~ 2
। ip(x, 0,0) + 2<^(0,0,0) + <^(—x, 0,0)
(3.17)
2/6(z) - 2/6(0) - Q'(x) - Ti(x)~T2(x)
Ms(v, x) + 7l/3(z/, -x) + 71/5(1/, x) + 71/5(1/, -x)

2
</?(0, x, 0) + 2(^(0,0,0) + </?(0, —x, 0)
+ 4 ’ (3.18)
for all x e V where the function Qf is given by
= lim
n—>00 4n
= lim h^nx) + f^n^
471
for all x E V and are given by
_ A(2^) + /3(2"x) - /x(-2"x) - /3(-2"x)

1^
1^x>- n* 2^
T'm - + ~ ~ f^nx)
-Z. 2\*^ ) — 11111
2n
for all x eV. From (3.15), (3.16) and the above inequalities, we easily obtain
T[(x)=Ti(x)+T3(x),
T^x) = T2(x)—T4(x),
Q\x} = 2Q(x)
for all x eV. From (3.17), (3.18) and the above, we obtain (3.12) and (3.13).
The inequalities of (3.15) can be verified by a similar method. □
Corollary 3.2 Let p 1,2, 0 > 0 be real numbers. Let : V —> [0, oo) be a
mapping such that ^(x) = ||x||p for x / 0 and ^(O) = 0 ifp > 1. Suppose
that the functions fa: V —» X satisfy
||/i(x + y + z) + f2(x - y + 2) + f3(x + y - z) + f4(-x + y + z)

- - 4fs(y) - 4/7(2)11 < -tKx) + V’(y) +
for all x,y,z e V. Then there exist exactly one quadratic function Q: V —» X
and four additive functions Ti, T2, T3, T4: V —> X satisfying
||/i(^)-/i(0)-Q(x)-T1(x)||
9. OP-1-4 1 9 9.2^ + 4 \
(2^2?i + V + 4?+
11/2 W -MO-QW-TIWII
M” + 5m
< / 2 ■ 2P + 4 12 2 ■ 2- + 4 \
— \2p|4 —2?| 2? 4p 2?|2 —2p|J ’
||/3(rr)-/3(0)-Q(x)-T3(a:)||
/ 2 • 2? + 4 1 2 2 • 2P + 4 \ r ,
- \2?|4 — 2p| 2? 4? 2?|2 — 2?|/
\\f4(x) — f4(0) — Q(x)— T4(x)\\
4
^ + 2; 4
2 + ^) \ + 26
(
IIAM - A(0) - <?« - IffiW + T2(x) + t3(x) - 74(1))!!
/ 2 1 4 + 2p 1\ „ „„ 35
~ \|4 - 2?| + 2? + 2|2 - 2p| + 4) +
\\fe(x) - /6(0) - Q(x) - l(Ti(x) - T2(x) + T3(x) + T4(z))||
“ (|4- 2?| + 2P + 2|2 - 2?| + 4) +
H/7^) - /7(0) - Q(x) - ^(THz) + T2(x) - T3(x) + T4(x))||

( 2 1 4 + 2? 1\ .. .._ 35 ,,n.
“ \J4 - 2?| + 2? + 2|2 - 2?| + 4/ + 6
for all x eV. The function Q is given by
Q(x\ _ < limn—oo ^4n ifp < 2,

y J (#) + fi (-#) " 2-M°)) < P’
for i = 1,2,..., 7 and for all x eV and the functions Ti (i = 1, 2, 3,4) are
given as in Corollary 3.1.
The following Corollary, which is due to S.M. Jung [16], can be easily ob
tained from Theorem 2.1.
Corollary 3.3 Let V be a normed space, X a Banach space. Suppose that the
functions fa: V —» X satisfy
/l(z + y + z) + f2(x — y + z) + f3(x + y-z) + fa(-x + y + z)

= 4/5(ar) + 4/6(y) + 4/7(2)
for all x,y,z e V. Then there exist exactly one quadratic function Q: V —> X
and four additive functions T\, T3, T4: V —> X such that
f\(x} = Q(x)+T\(x) +
f2(x) = Q(x)+T2(x') +
/3(rr) = Q(rr) + T3(a;) +/3(0),
fa(x) = Q(x) + T^x) + /4(0),
/5(x) = Q(x) + l(T!(x) + T2(x) + T3(x)- T^xf) + /5(0),
/6(x) = Q(x) + |(Ti(x) - T2(x) + T3(x)+ T4(x)) + /6(0),
/7(x) = Q(x) + l(Ti(x) + T2(x) - T3(s)+ T^x)) + /7(0)
for all x eV.
References
[1] P.W. Cholewa: ‘Remarks on the stability of functional equations’, Aequat.
Math. 27 (1984), 76-86.
[2] S. Czerwik: ‘On the stability of the quadratic mapping in the normed
space’, Abh. Math. Sem. Hamburg 62 (1992), 59-64.
[3] Z. Gajda: ‘On the stability of additive mappings’, Internat. J. Math, and
Math. Sci. 14 (1991), 431-434.

proximately additive mappings’, J. Math. Anal, and Appl. 184 (1994),
431-436.
Acad. Sci. USA 27 (1941), 222-224.
Math. 44 (1992), 125-153.
[7] D.H. Hyers, G. Isac and Th.M. Rassias: ‘On the asymptoticity aspect of
Hyers-Ulam stability of mappings’, Proc. Amer. Math. Soc. 126 (1998),
425-430.
tions in Several Variables', Birkhauser, 1998.
[9] G. Isac and Th.M. Rassias: ‘On the Hyers-Ulam stability of ^-additive
mappings’, J. Approx. Theory 72 (1993), 131-137.
[10] G. Isac and Th.M. Rassias: ‘Functional inequalities for approximately
additive mappings', pages 117- 125 In: Stability of Mappings of Hyers-
Ulam Type, Hadronic Press, Inc., Florida, 1994.
[11] G. Isac and Th.M. Rassias: ‘Stability of ^-additive mappings: Appli
cations to nonlinear analysis’, Intemat. J. Math, and Math. Sci. 19 (2)
(1996), 219-228.
[12] K.W. Jun, D.S. Shin and B.D. Kim: ‘On the Hyers-Ulam-Rassias stabil
ity of the Pexider equation’, J. Math. Anal. Appl. 239 (1999), 20-29.
[13] K.W. Jun and Y.H. Lee: ‘On the Hyers-Ulam-Rassias stability of a gen
eralized quadratic equation’, Bull. Korean Math. Soc. (to appear).
[14] K. W. Jun and Y.H. Lee: ‘On the Hyers-Ulam-Rassias stability of a Pex-
iderized quadratic inequality’, Math. Ineq. Appl. (to appear)
[15] S.-M. Jung: ‘On the Hyers-Ulam-Rassias stability of approximately ad
ditive mappings’, J. Math. Anal. Appl. 204 (1996), 221-226.
[16] S.-M. Jung: ‘Quadratic functional equations of Pexider type’, Intemat. J.
Math, and Math. Sci. 24 no. 5 (2000), 351-359.
[17] Y.H. Lee and K.W. Jun: ‘A generalization of the Hyers-Ulam-Rassias
stability of Jensen’s equation’, J. Math. Anal. Appl. 238 (1999), 305-315.
[18] Y.H. Lee and K.W. Jun: ‘A generalization of the Hyers-Ulam-Rassias
stability of Pexider equation’, J. Math. Anal. Appl. 246 (2000), 627-638.
[19] Y.H. Lee and K.W. Jun: ‘A note on the Hyers-Ulam-Rassias stability of
Pexider equation’, J. Korean Math. Soc. 37 (2000), 111-124.
[20] Y.H. Lee and K.W. Jun: ‘On the stability of approximately additive map
pings’, Proc. Amer. Math. Soc. 128 (2000), 1361-1369.
[21] Th.M. Rassias: ‘On a modified Hyers-Ulam sequence’, J. Math. Anal.

Appl. 158(1991), 106-113.
[23] Th.M. Rassias: ‘On the stability of functional equations in Banach
spaces’, J. Math. Anal. Appl. 251 (2000), 264-284.
Ulam’, Acta Appl. Math. 62 (2000), 23-130.
[25] Th.M. Rassias: ‘On the stability of functional equations originated by a
problem of Ulam’ Studia Univ. “Babes-Bolyai" (to appear).
[26] Th.M. Rassias: ‘On the stability of the quadratic functional equation’,
Mathematica XLV (2) (2000), 77-114.
[27] Th.M. Rassias: ‘On the stability of the quadratic functional equation and
its applications’, Studia Univ. “Babes-Bolyai” (to appear).
[28] Th.M. Rassias: ‘Report of the 27th Intemat. Symposium on Functional
Equations’, Aeq. Math. 39 (1990), 292-293. Problem 16,2°, Same report,
p. 309
[29] Th.M. Rassias and P. Semrl: ‘On the behavior of mappings which does
not satisfy Hyers-Ulam stability’, Proc. Amer. Math. Soc. 114 (1992),
989-993.
Natural Geometry 1 (1992), 65-69.
[31] F. Skof: ‘Local properties and approximations of operators’, Rend. Sem.
Mat. Fis. Milano 53 (1983), 113-129.
[32] S.M. Ulam: 'Problems in Modem Mathematics. Chap. VF, Science eds.,
Wiley, New York, 1960.
Chapter 5
ON THE HYERS-ULAM-RASSIAS
STABILITY OF A
FUNCTIONAL EQUATION
*
Soon-Mo Jung
Mathematics Section, College of Science and Technology
Hong-Ik University, 339-701 Chochiwon, Korea
smjung@wow.hongik.ac.kr
Abstract In this paper, we will introduce a new functional equation /(rci, )f(x2,2/2) —
f(x\X2 + 7/i?/2,î2/2 — 7/13:2), which is strongly related to a well known ele
mentary formula of number theory, and investigate the solutions of the equation.
Moreover, we will also study the Hyers-Ulam-Rassias stability of that equation.
2000 MSC: 39B82, 39B22 (Primary).
Keywords: Hyers-Ulam-Rassias stability, functional equation.
1. Introduction
In 1940, S.M. Ulam gave a wide ranging talk before the Mathematics Club
of the University of Wisconsin in which he discussed a number of important
unsolved problems ([14]). Among those was the question concerning the sta
bility of homomorphisms:
Let Gi be a group and let be a metric group with a metric

•). Given any s > 0, does there exist a 8 > 0 such that if a
function h: G± —> G% satisfies the inequality d(fi(xy\ h^r^h^y^) <
8for all x,y G Gi, then there exists a homomorphism H: G± —»
G2 with d(h(x), H(z)) < e for all x G G\?
*This work was financially supported by KOSEF (2000); project no. R02-2000-00005.
67
If the answer is affirmative, the functional equation for homomorphisms

is said to be stable in the sense of Hyers and Ulam because the first result
concerning the stability of functional equations was presented by D.H. Hyers.
Indeed, he has answered the question of Ulam for the case where Gi and G?
are assumed to be Banach spaces (see [4]).
In 1978, Th. M. Rassias [9] addressed the Hyers stability theorem and at
tempted to weaken the condition for the bound of the norm of the Cauchy
difference h(xy) — h(x')h(y). He proved a considerably generalized Hyers’
result (see also [2]). Since then, the stability problems of various functional
equations have been investigated by many authors (see [1], [3], [5], [6], [7],
[8], [10], [11], [12]).
Taking into account the large influence of D.H. Hyers, S.M. Ulam and Th.M.
Rassias on the study of stability problems of functional equations, the stability
phenomenon that was proved by Th.M. Rassias is called Hyers-Ulam-Rassias
stability.
According to a well-known theorem in number theory, a positive integer of
the form m2n, where each divisor of n is not the square of an integer, can be
represented as the sum of two squares of integers if and only if every prime
factor of n is not of the form 4k + 3. In the proof of this theorem, we make use
of the following elementary equality
(x2 + yi){x2 + 3/2) = (^2 + 3/U/2)2 + (213/2 - yi x2}2- (i-i)
In this paper, we will introduce a new functional equation:
?/i)/(^2,2/2) = f(xiX2 + 3/13/2, X13/2 - 3/12:2) (1-2)
and investigate the solutions of the equation. Moreover, we will also prove
the Hyers-Ulam-Rassias stability of that equation. We can easily note that the
above functional equation is strongly related to the equality (1.1).
2. Solutions of eq. (1.2)

In this section we will investigate the solutions of the functional equation
(1.2) in the class of functions f: R2 —> R.
Theorem 2.1 If a function f: R2 —> R satisfies the functional equation (1.2)

for all xi,X2,yi,y2 £ R, then there exist a multiplicative function m: R —> R
and a signum function cr: R2 —> {±1} such that
f(x, 3/) = cr(x, 3/) m x2 + 3/2 )
for all real numbers x and y.

PROOF: Put y\ = y2 = 0 in (1.2) to get
/(xi,0)/(x2,0) = /(x1x2,0) (2.1)
for all a?2 G R- Replace the Xi’s by x and the yfs by y in (1.2) to get
f(x,y)f(x,y) = /(z2 + y2,0) (2.2)
for any x, j e R. Using (2.2) twice, we have

f(.y, x)f(y, x) = f(y2 + x2,0) = f(x2 + y2,0) = f(x, y)f(x, y)
and hence we may define a function <ti : R2 —> {±1} by
/(y, x) = <ri(x, y)f(x, y) (2.3)
for all real numbers x and y.

From (1.2), (2.2) and (2.3), it follows that
f(2xy, x2 - y2) = f(x, y)f(y, x)

= y)f^.y)f(x, y)
= <Ti(x,y)/(a:2 + y2,0) (2.4)
for any real numbers x and y.

We note that for any given u, v e R, the following system of equations
2xy = u,
(2.5)
x 2 — 2y^ — v
has solutions (x(u, v), y(u, v)) in R2 as we see in the following:
for u — 0, v > 0,
for u = 0, v < 0,
_|_______ u_______
for u / 0.
2v-]-2\/ u2+v%
It follows from (2.5) that x2 + y2 = y/u2 + v2, According to (2.4) and (2.5),
we obtain
/(u,v) = cri(x(u,r;),?/(u,v))/ ^\/u2 + v2, (2.6)
for any u, v € R.
Taking (2.6) into account, we may introduce another function a: R2 —>
{±1} that satisfies the equality
f(u,v) = cr(u,v)f u2 + v2, o) (2.7)

for all u, v G R.
Finally, define a function m: R —> R by m(x) = /(x,0) for each x € R.
Then, (2.1) and (2.7) ensure that m is a multiplicative function and that
/(x, y) = <r(x, y) m (\A2 +y2)
for all real numbers x and y. □
3. Hyers—Ulam—Rassias stability of Eq. (1.2)

In this section we will allude to the Hyers-Ulam-Rassias stability of the
functional equation (1.2). In view of Theorem 2.1, we can guess that the sta
bility problem of Eq. (1.2) is strongly connected with multiplicative functions.
Theorem 3.1 Let X be afield and Mi, M2, Ni, N2: X —> [0,00) functions.
If a function f: X2 —> C satisfies the following inequality
2/2) - /(®ix2+ yiy2,xxy2 - yix2)l

< min{Mi(xi), M2(x2), Ni(yi), (V2(y2)} (3.1)
for all xi, X2, yi, y2 G X, then f(x, 0) is either bounded or multiplicative and
further it satisfies
\f(x,y)2 - /(x2 + y2,0)| < min{M1(x),M2(x),ATi(y), AT2(y)}
for any x,y G X.
PROOF: With yi = y2 = 0, (3.1) implies
|/(xi,0)/(x2,0) - f(xiX2,0)| < min{Mi(xi), M2(x2), M(0), 7V2(0)}
for xi, x2 G X. If we substitute m(x) for /(x, 0) in the above inequality, then
we have
|m(xi)m(x2) -m(xix2)| < min{Mi(xi), M2(x2), 7Vi(0), 7V2(0)}
for all xi, x2 € X.

Applying a theorem of L. Szekelyhidi [13] (or see [8, Corollary 8.4]), we
conclude that m is either bounded or multiplicative.
Finally, put xi = x2 = x and yi = y2 = y in (3.1) to get
|/(x, y)2 - /(x2 + y2,0)| < min{Mi(x),M2(x),M(y),-M2(y)}
for all x, y G X. □
References
[1] G.L. Forti: ‘Hyers-Ulam stability of functional equations in several vari

ables’ , Aequat. Math. 50(1995), 143-190.
[2] Z. Gajda: ‘On stability of additive mappings’, Intemat. J. Math. Math.
Sci. 14 (1991), 431-434.
proximately additive mappings’, J. Math. Anal. Appl. 184 (1994), 431-
436.
Acad. Sci. USA 27 (1941), 222-224.
tions in Several Variables', Birkhauser, Boston-Basel-Berlin, 1998.
[6] D.H. Hyers and Th. M. Rassias: ‘Approximate homomorphisms’, Ae
quat. Math. 44 (1992), 125-153.
[7] S.-M. Jung: ‘Hyers-Ulam-Rassias stability of functional equations’, Dy
namic Sys. Appl. 6 (1997), 541-566.
[8] S.-M. Jung: ‘Hyers-Ulam-Rassias Stability of Functional Equations in
Mathematical Analysis’, Hadronic Press, Palm Harbor, 2001.
[10] Th.M. Rassias: ‘On the stability of functional equations originated by a
problem of Ulam’, Studia Univ. Babes-Bolyai (to appear).
Ulam’, Acta Appl. Math. 62 (2000), 23-130.
[12] Th.M. Rassias: ‘On the stability of functional equations in Banach
spaces’, J. Math. Anal. Appl. 251 (2000), 264—284.
[13] L. Szekelyhidi: ‘On a theorem of Baker, Lawrence and Zorzitto’, Proc.
Amer. Math. Soc. 84 (1982), 95-96.
[14] S.M. Ulam: ‘Problems in Modem Mathematics, Chap. VI', Wiley, 1964.
Chapter 6
A PAIR OF FUNCTIONAL INEQUALITIES

OF ITERATIVE TYPE RELATED TO A
CAUCHY FUNCTIONAL EQUATION
Dorota Krassowska
Institute of Mathematics, University of Zielona Gora
ul. Podgorna 50, PL 65-246 Zielona Gora, Poland
D. K rassowska ©i m. uz.zgora. pl
Janusz Matkowski
Institute of Mathematics, University of Zielona Gora
ul. Podgorna 50, PL 65-246 Zielona Gora, Poland
J.Matkowski@im.uz.zgora.pl
Abstract It is shown that, under some general algebraic conditions on fixed real numbers
a,b,a,/3, every continuous at a point solution f of the system of functional
inequalities f(x + a) < f(x) 4- a, f(x 4- b) < f(x) 4- (3 (x e R) must be
a polynomial of order 1. Analogous results for three remaining counterparts of
this simultaneous system are presented. An application to characterization of
Lp-norm is given.
1991 MSC: 39B72, 26D15 (Primary); 46E30 (Secondary).
Keywords: functional inequality, functional equation, Kronecker’s theorem, Lp-norm-like

functional, subhomogeneity, characterization of Lp-norm
1. Introduction
In the present paper we deal with the simultaneous pair of functional in
equalities
f (a + x) < a + f (ar) , f(b + ar) < + f (x), x G R,
73
where a, b, a, (3 are fixed real numbers. The main result says that, under some
general algebraic type conditions on a, b, a, (3, the only continuous at a point
solution of this pair must be of the form fix) = px + /(0) (x G R), where
p = a/a and /(0) is an arbitrary constant. If the algebraic conditions fail
to hold, then a continuous solution of the accompanying system of functional
equations
/(a-f-x) = a + /(x), f(b + z) = /3 + f (x), x G R,
depends on an arbitrary function.

The additively-multiplicative, multiplicatively-additive and multiplicative
counterparts of the above pair are also considered.
The results of this paper, concerning the pair of functional inequalities, gen
eralize the results by Matkowski [4], where only the pair of functional equa
tions were considered (cf. also Matkowski, Sablik [7]). They allow, in partic
ular, to derive the classical theorems on the Cauchy type functional equation
(cf. J. Aczel [1] and M. Kuczma [3]).
As an application, in section 3 we characterize the L/’-norm with the aid of
a rather weak subhomogeneity condition of the 1/-norm-like functional p<^.
This is closely related to the results by Matkowski [5], Wnuk [8], Zaanen [9],
where is assumed to be positively homogeneous.
2. Auxiliary results
In this paper N, Z, Q, R, stand, respectively, for the set of natural, integer,
rational, and real numbers.
A well known Kronecker’s theorem says that if a and b are noncommensu-
rable real numbers, then the set {ma + nb : m, n € Z} is dense in R. As an
easy consequence we obtain (cf. [6])
Lemma 2.1 Let a, b G R, a < 0 < b, be fixed. Let us denote A = {ma + nb :

m, n G N}.
1) If Q, then the set A is dense in R.
2) If € Q, then there is d > 0 such that
A = {kd : ke Z}.
The multiplicative version of this Lemma reads as follows:
Lemma 2.2 Let a,b G R, 0 < a < 1 < b, be fixed. Let us denote M =
{ambn : m,n € N}.
7) If Q, then the set M is dense in (0, oo).

A pair of functional inequalities of iterative type 75
2) If e Q, then there is d> 0 such that
M = {dk : fee Z}.
3. Functional inequalities and related equations

In this section we deal (in turn) with the simultaneous pair of functional
inequalities of additive type'.
f(a + x)<a + f(x), f(b + x) < (3 + f (x);
additively-multiplicative type:
f (a +x) < af (x), /(b + x) </?/(x);
multiplicatively-additive type:
f (ax) < a + f (x), /(bx) < /3 + f (x);
and multiplicative type:
f (ax) < af (x), /(bx) < /3f (x),
where a, b, a, (3 are fixed real numbers. Assuming some special conditions on

a, b, a, (3 and the continuity of f at least at one point, we determine the form
of their solutions. Note that if a = f (a), (3 = f(b), then the additive type pair
is a restriction of the classical Cauchy type inequality, called subadditivity of
f,
f(x + y)< f(x) +f(y), x,ytR,
to the set {(x, y) e R2 : y = a V y = b}.

The first result reads as follows:
Theorem 3.1 Let a, b, a, (3 6 R be fixed numbers such that
/ n / a a @
a < 0 < b, - =
a b
and suppose that a junction f: R —> R is continuous at least at one point.
1) If £ Q, then f satisfies the system offunctional inequalities
f (a + x) < a + f (x), f(b + x) < /3 + f (x), x e R, (3.1)
if, and only if,
f (x) = px + f (0), x e R,
where p :=
2) If e Q, then the general continuous solution of the simultaneous sys

tem offunctional equations
f (a +x) = a + f (x), f(b + x) = /3 + f (x), x e R, (3.2)
depends on an arbitrary function; more precisely: the number
d := min{ma + nb > 0 : G N}
exists, is positive, and, for every continuous function fo: [0, d] —> R
satisfying the condition
/o(d) = ^d + /o(O),
there exists a unique continuous solution f: R —> R of system (3.2) such

that f |[o,c/]— fo- Moreover, if fo is (strictly) monotonic, then so is f.
Proof: 1) From (3.1), by induction, we obtain
f (ma + x) < ma+f (x), f(nb+x) < n/3+f (x), m, n e N, x e R.
Replacing x by nb + x in the first of these inequalities we have
f(ma + nb + x) < ma + f(nb + x) < ma + n/3 + f(x),
and, consequently,
f(ma + nb + x) < ma + n/3 + f(x\ m^n e N, x e R.
Setting
a (3
P-= ~ = V,
a b
we can write this inequality in the form
f(t + x) < pt + f(x), teA, x g R, (3.3)
where the set
A = {ma + nb : m,n G N},
according to Lemma 2.1, part 1), is dense in R.

Suppose that xq is a continuity point of /, and take an arbitrary x e R. By
the density of A there exists a sequence (tn) such that
tn G A (n G N), lim tn = xq — x.
n—>oo
From inequality (3.3) we have
f(tn + x) < ptn + f(x), n G N.
Letting here n —> oo, and making use of the continuity of f at xq, we obtain
/(t0) < p(x0 - x) + f(x), X G R.
To prove the converse inequality note first that replacing x by x — t in (3.3)

gives
/(t) < pt + f(x — i), t G A, x e R.
Now taking an arbitrary fixed x G R, and, by the density of A, a sequence (t„)

such that
tn G A (n € N), lim tn = x — xq,

n—>oo
we hence get
/(rr) <ptn + n G N.
Letting here n —> oo, and making again use of the continuity of f at xq, leads
to the inequality
/(x) < p(x - To) + f(xo), X g R.
Thus we have shown that
/(z) =px+ (/(t0) -px0), tgR.
Since the ‘if’ part is obvious the proof of the first part is complete.
To prove 2), note, similarly as in the first part, that system (3.2) implies
f(ma + nb + t) = — (ma + rib) + /(t), m,n6N, x € R. (3.4)
According to Lemma 2.1, part 2), the number d := min{ma + nb > 0 :

m,n 6 N} is well defined and positive. Moreover {ma + nb : m, n G N} =
{kd : k G Z}. It follows that (3.4) has the form
f(kd + t) = — kd + /(t), k G Z, x G R.
It is easy to see that this system of functional equations is equivalent to the

functional equation
f(d + x) = —d + /(t), x G R.
Now we define /i: (d, 2d] —> R by the formula
/i(z) := ~d + fo(x - d), x E (d,2d[.
Assume that fn: {nd, {n + l)d] —> R (n € N) is defined. Then we define

fn+i: {{n + l)d, {n + 2)d] —> R by the recurrence relation
fn+i{x) := ^d + fn{x - d), x G {{n + l)d, {n + 2)d], n G N.
Analogously, let /_i: [—d, 0) —> R be defined by
f-l{x) -~d + f0{x + d), x G [-d, 0).
Suppose that we have already defined /_n: [—nd, (—n + l)d) —> R (n G N).
Then we define : [— (n + l)d, —nd) —> R by
/_(n+i)(rr) := -*
^d + f_n{x + d), x G [-(n + l)d, -nd), n G N.
It is easy to verify (cf. [2]) that f: R —> R defined by

p-n(a?)
for x G [—nd, (—n + l)d),
f^ = < fo(x) for x G [0, d], n G N,
(/n(z)
for x G (nd, (n + l)d],
satisfies system (3.2), is continuous, and / |[o,d]— fo- Since the ‘moreover’
part is obvious, the proof is complete. □
Now we consider an ‘additively-multiplicative’ case.
Theorem 3.2 Let a, b G R and a, (d > 0 be fixed numbers such that

log a log (d
a R is continuous at least at one point.
1) If Q. then f satisfies the system offunctional inequalities
f (a + x) <af {x), f{b + x) < /3f {x), x G R, (3.5)
if and only if
f {x) = f {0) epx, xgR,
where p
r := a
2) If - e Q, then the general continuous solution of the simultaneous sys

tem offunctional equations
f (a + x) = af (x), f(b + x) = (3f (x), x € R, (3.6)
d := min{ma + nb > 0 : E N}
exists, is positive, and, for every continuous function fa: [0, d] —> R
satisfying the condition
fo(d) = fo(O)e^d,
there exists a unique continuous solution f: R —> R of system (3.6) such

that f |[o,c/]— fa- Moreover, if fa is (strictly) monotonic, then so is f.
Proof: 1) Similarly as in the proof of Theorem 3.1, from (3.5) we obtain
f(ma + nb + x) < am/3nf (x), m, n e N, x e R.
Setting
log a log (3
P-=-------=
a b
we can write this inequality in the form
f(ma + nb + x) < (ema+n6)Pf m,n e N, x e R.
Hence, as in the previous case, we have
/(t + x) < eptf(x), t e A, x e R, (3.7)
where
A = {ma + nb : E N}
is dense in R.
Suppose that f is continuous at xq, and take an arbitrary x e R. Hence, by
density of A, there exists a sequence (tn) such that
tn e A (n e N), lim tn = xq — x.
n—>oo
From inequality (3.7) we have
f(tn +x) < eptnf(x), n e N.

Letting here n —> oo, and making use of continuity of f at xq, we obtain
/(x0) < x g R.
On the other hand, replacing x by x — t in (3.7) and then taking a sequence

(tn) such that
tn G A (n G N), lim tn = x — xq
n—>oo
in the same way, we get the inequality
/(x) < e^-IOV(x0), x G R.
Therefore f has the required form.

2) Since the construction of the solution of system (3.6) is similar to that
given in Theorem 3.1, part 2), we omit the proof of this part. □
In the ‘multiplicatively-additive’ case we have the following
Theorem 3.3 Let a, b, (3 be fixed real numbers such that
Suppose that a function f: I —> R is continuous at least at one point.
7) If Q and f satisfies the system offunctional inequalities
f (ax) <& + f(x), f(bx) < (3 + f (x), x e I, (3.8)
then
i) in the case I = (0, oo),
f(x) =plogz + /(l), X > 0,
ii) in the case I = (—oo, 0),
f (x) = plog(-x) + /(-l), x < 0,
where p:=
2) If G Q, then the general continuous solution of the simultaneous

system offunctional equations
f (ax) = a + f (x), f(bx) = (3 + f (x), x G I, (3.9)

d := min{am6n > 1 : m,n E N}
exists, is greater than 1, and, for every continuous function fo: [1, d] —>
R satisfying the condition
fo{d) = log d + /0(l),

log a
there exists a unique continuous solution f: I —> R of system (3.9) such

that f | = /o- Moreover, if fo is (strictly) monotonic, then so is f.
Proof: To prove 1), suppose that I = (0, oo). Similarly as in the previous
proofs, from (3.8) by induction, we get
f(ambnx) < ma + n/3 + /(z), m, n G N, x > 0.
Hence, putting
a f3
V' log a log b ’
we obtain the inequality
f(ambnx) < plog(am6n) + f(x), m,neN, x > 0,
which can be written in the form
f(tx) < plogt + /(z), t G M, x > 0, (3.10)
where the set
M={ambn:
according to Lemma 2.2, part 1), is dense in I.

Suppose that f is continuous at xq > 0, and take an arbitrary x > 0. By the
density of M in I there exists a sequence (tn) such that
tn G M (n G N), lim tn = —.
n—>oo x
From (3.10) we get
/(
*
nZ) <Plogin+ /(>), n e N.
Letting here n —> oo, and using the continuity of f at xq, we get
/(xo) < Plogy+ /(»), x > 0.

Similarly as previously, replacing x by f in (3.10) and taking the sequence

(tn) such that
tn G M (n 6 N), lim tn = —,
n—>oo Xq
we get the inequality
/(t) 0.

^0
Thus we have shown that
f(x) = /(x0) -plogzo +plogrc, x > 0,
which completes the proof of i). To prove ii) it is enough to observe that the
function g: (0, oo) —> R defined by the formula p(rr) = f(—x) (x < 0)
satisfies system (3.8) and apply the already proved part i).
Since an argument for 2) is analogous to that of Theorem 3.1, part 2), we
omit it. □
Corollary 3.1 Let a, b, a, (3 satisfy the assumptions of Theorem 3.3, part 1).
If a function f: [(—oo, 0) U (0, oo)] —> R satisfies the system of inequalities
(3.8) and in each of the intervals (—oo, 0) and (0, oo) there is at least one point
of continuity of f, then
f(x\ = (pi°gz + f(i) f°rx e (0, oo),

(plog(-z) + /(-I) for x e (-oo,0),
where p'.—
Remark 3.1 Suppose that a, b, a, (3 are fixed real numbers such that 0 < a <
1 < b and Note that if 0 e I then there is no function satysfying
system (3.8).
Indeed, putting x = 0 to inequalities (3.8) gives 0 < a, 0 < (3, which

contradicts to the assumptions which imply that a(3 < 0.
Finally we consider the ‘multiplicative’ case.
Theorem 3.4 Let a,b, a, (3 be fixed positive real numbers such that
1 , logo: _ log/3
’ log a log b
Suppose that a function f: I —> R is continuous at least at one point.

7) If Q, and f satisfies the system offunctional inequalities
f (ax) < af (x), f(bx) < /?/ (x), x e f (3.11)
then
i) in the case I = (0, oo),
f(x) = f(l)xp, z > 0,
ii) in the case I = (—oo, 0),
/(
*
) = /(-l) z<0,
where p:=
2) If G Q, then the general continuous solution of the simultaneous

system offunctional equations
f (ax) = af (x), f(bx) = (3f (ar), x e I, (3.12)
d min{ambn > 1 : m,n E N}
exists, is greater than 1, and, for every continuous function fo: [1, d] —>
R satisfying the condition
there exists a unique continuous solution f: I —> R of system (3.12)

such that f |[i d]= fo- Moreover, if fo is (strictly) monotonic, then so is
f-
Proof: To prove part li), suppose that I = (0, oo). System (3.11) implies
that
f(ambnx) < am/3nf(x), m,n^, x > 0.
Hence, putting
_ logo: log/3
’ log a log b ’
we obtain
f(ambnx) < (ambn)pf (x), m,neN, x> 0.

Therefore, for all t G M, where M = {ambn : m, n € N}, we get
/(tor) < tpf(x), te M, x > 0. (3.13)
Suppose that xq > 0 is a continuity point of /, and fix an arbitrary x > 0.

By the density of the set M in I, there exists a sequence (fn) such that
tn G M (n G N), lim tn = —.
n—>oo x
From (3.13) we obtain
/(tn;r) < tpf(x), n G N.
Letting here n —> oo, and making use of the continuity of f at xq, we get
f(x0) < (—\Pf(x), X > 0.

\X /
Now replacing x by y in (3.13) and considering a sequence (fn) such that
tn G M (n G N), lim tn -
n—>oo Xo
we get also the converse inequality
\p
(— ) f(xo),
^0 /
X > 0,
which completes the proof of li). To prove lii) it is enough to note that
g: (0, oo) —» R defined by g(x) = f(—x) (x < 0) is a solution of (3.11)
and then apply the already proved part li).
We omit an easy argument for 2). □
Remark 3.2 Suppose that a, b, a, (3 are fixed positive real numbers such that
a < 1 R satisfies system (3.11), then /(0) = 0.
Indeed, by assumption either a < 1 < /3 or 0 < 1 < a and, moreover,

/(0)(l - a) < 0 and /(0)(l - /?) < 0. Thus /(0) = 0.
Making use of this remark, we get
Remark 3.3 i) Suppose that f: [0, oo) R satisfies system (3.11). If

f l(0,oo) and a, b, a, (3 satisfy all remaining assumptions of Theorem 3.4,
part 1), then
jf(l)xp fora: G (0,oo),

J(x) = < _
where p:=
ii) Suppose that f: (—oo,0] —> R satisfies system (3.11). If f |(-oo,0) and
a, b, a, 0 satisfy all remaining assumptions of Theorem 3.4, part 1), then
for x G (—oo,0),
v ) 0
for x = 0,
where p:=
Corollary 3.2 Let (3 satisfy the assumptions of Theorem 3.4, part 1).
If f •’ [(“°°> 0) U (0? °°)] “> K satisfies the system of inequalities (3.11)

and in each of the intervals (—oo,0) and (0, oo) there is at least one
point of continuity of f, then
f(x) = / fwxp f°r x e (°> °°)’

|/(-l)(-x)P for x G (-00,0),
w/iere p :=
2) If f: R —> R satisfies the system of inequalities (3.11) and in each of the

intervals (—oo, 0) and (0, oo) there is at least one point of continuity of
f, then
f(x)zp for x G (0, oo),

/(
*
) - o for x - 0,
(f(-l)(-x)P forx G (—oo,0),
w/iere p :=
Remark 3.4 For obvious reasons the counterparts of all Theorems 3.1-3.4,
for an unknown function f satisfying the converse inequalities, remain true.
We conclude this section with the following
Remark 3.5 Assuming some additional conditions of the ranges or domains

of the unknown functions /, the main parts of Theorems 3.2, 3.3, 3.4 can be
deduced from Theorem 3.1. For instance, if f: R —> (0, oo) then Theorem 3.2
is an easy consequence of Theorem 3.1.
4. Application to a characterization of Lp-norm

For a measure space (Q, S, /i) denote by S = S(Q, S, /z) the linear space
of all ^-integrable simple functions rr: Q —> R. For an arbitrary bijection
</>: (0, oo) (0, oo) we define p^,: S —> R by
< & 1 (fax 0 lxl'

P^) ■= x 6 S,
X0 p^Six) = o,
where Qx := {u> e Q : rr(w) f 0}. Note that the form of the functional
is similar to the IT5-norm and for </>(£) = <^(l)ip, with p > 1, it coincides with
this norm. In the sequel Xa denotes the characteristic function of the set A.
Applying Theorem 3.4 we prove
Theorem 4.1 Let (Q, S, p) be a measure space with two sets A,B € S such
that
0 < < 1 < M(B) < oo,
Suppose (j)\ (0, oo) —> (0, oo) is a bijection such that </>-1 is continuous at
least at one point and
log^Â)^!)) = log^"1^^)^!))
(4.1)
log^(A) log/z(B)
Iflpep satisfies the condition
p^(ix) < tp^(x), i>0, xE{xa,Xb}, (4.2)
then there exists p € R\{0} such that
< t>(f) = <t>(I)tp, i > 0.
(In particular, ifp > 1 then functional p<^ coincides with the U’-norm.)
PROOF: Let a = p(A) and b = p(Bf From (4.2) we obtain
< ^-1(a<^(i)) < (a<^(l)), f~\b<f>(ty) < £</>-1(6^>(l)), t > 0,
which with a := <^>-1(a</>(l)) and (3 := <^-1(6<^(l)) reduces to the system of

functional ineqalities
< ^-1(a</>(£)) < at, <^>-1(6</>(i)) < /3t, t > 0.
From the bijectivity of <j>, replacing here t by we get the equivalent

system of inequalities
< ^-1(af) < a^>-1(f), t > 0,
which, with f = <f> 1 and I = (0, oo) takes the form (3.11).
Now our result follows from Theorem 3.4. □
Remark 4.1 In view of Remark 3.4, the counterpart of Theorem 4.1 in which
the subhomogeneity condition (4.2) is replaced by the following superhomo
geneity one
p^(tx) > ip^(x), t > 0, x€{xa,Xb},
is also true.
Since in (4.2) only two functions xa,Xb are involved, this condition is a
very weak substitute of the homogeneity of the functional p^.
Finally, let us discuss the necessity of the assumptions in the last theorem.
First note that it is rather exceptionally if the numbers and
are simultaneously rational. Thus, given a measure space (Q, S, /z) such that
/z(S) A (0,1) / 0 /z(S) Fl (1> °°)> in general there are sets A, B e S for
which the conditions 0 < p(A) < 1 < /z(B) < oo and Q are
satisfied.
Now in examples we show that the assumption of the existence of sets A
and B with 0 < p(A) < 1 < /z(B) < oo is essential. Namely, we indicate
wide classes of non-power functions <f for which the functional p<^ satisfies the
condition (4.2), in each of cases
/z(A) <1 or /z(A) = oo for every A G S;

/z(A) >1 or /z(A) = 0 for every A G S.
Example 4.1 Let (Q, S, p) be a measure space such that /z(Q) < 1. Put
6 inf{/z(A) : A G S A p(A) > 0}. Let </>: (0, oo) —> (0, oo) be an in
creasing bijection such that the function (0, oo) 9 t —> is non-increasing
and <^>(5) = 6, </>( 1) = 1. Then </>(£) = t for all t G [5,1], the function
(0, oo) 9 t —> t is non-decreasing, and therefore for each a := p(A) > 0,
A € S, we have
P^Xa) = a^f)
< = ta = ^-1W(i)) = tp<t>(xA), t > o,
which proves that p<f> satisfies the subhomogeneity condition (4.2). Since, of
course, <^> is not a power function, this proves that the existence of a set B G S
with^(B) G (l,oo) is necessary.
Example 4.2 Let (Q, S, pi) be a measure space for which p(A') > 1 for every
set A e S, such that p(A) > 0, and there exists B e S such that 1 <
p(B) < co. Then 8 := inf{/z(A) : A G S A pi(A) > 0} > 1. Let
</>: (0, oo) —> (0, oo) be a bijection such that the function (0, oo) 9 t —>
is non-decreasing and <^>(1) = 1, f(8) = 8. Then f is strictly increasing,
f{t) = t for all t e [1,5], the function (0, oo) 9 t —> is non-increasing,
and therefore, in the same way as in the previous example, for all B G S such
that 0< p(B) < oo, we have
P^Xb) < t > 0.
This shows that p<^ satisfies condition (4.2). Thus we have shown that the
existence of A G S such that 0 < pi(A) < 1 is essential.
We end our discussion with an example showing that the assumption (4.1)
is indispensable.
Example 4.3 Let (Q, S, pi) be an arbitrary measure space, f: R —> R a bijec
tion such that /(0) = 0 and /-1 is subadditive (for instance, for f one can
take the inverse function to x —> x + |sin rc|). Define (j>: (0, oo) —> (0, oo) by
</>(i) = e^logt\ Then, making use of the definition of p<f>, the subadditivity of
/-1, and the monotonicity of the exponential function, for all a := p(A) > 0,
A € X, we have
P^tXA) = fW
*
)) = ef~1(~losa^ = e-Tîoga+fOogt))
< gJ-Viog^gZ-^/Ciogt)) = = tp^(xx), t > 0.
Thus p<f> satisfies the subhomogeneity condition (4.2) for all functions xa,
independently whether the measure p(A) is smaller or greater than 1. In par
ticular, this shows that, in Theorem 4.1, condition (4.1) cannot be omitted.
References
[1] J. Aczel: 'Lectures on Functional Equations and their Applications’, Aca
demic Press, New York-London, 1966.
[2] M. Kuczma: 'Functional Equations in a Single Variable', Monografic
Mat. 46, Polish Scientific Publishers (PWN), Warszawa, 1968.
[3] M. Kuczma: 'An Introduction to the Theory of Functional Equations
and Inequalities, Cauchy’s Equation and Jensen’s Inequality', Polish
Scientific Publishers (PWN), Silesian University, Warszawa-Krakow-
Katowice, 1985.
[4] J. Matkowski: 'Cauchy functional equation on a restricted domain and
commuting functions’, pages 101-106 In: Iteration Theory and its Func
tional Equations, Proc., Schloss Hofen 1984, Springer Verlag, 1985.
[5] J. Matkowski: ‘On a characterization of 17-norm’, Ann. Polon. Math. 50

(1989), 137-144.
[6] J. Matkowski: ‘The converse of the Minkowski’s inequality theorem and
its generalization’, Proc. Amer. Math. Soc. 109 (1990), 663-675.
[7] J. Matkowski and M. Sablik.
[8] W. Wnuk: ‘Orlicz spaces cannot be normed analogously to 17-spaces’,
Indagationes Math. 46 (1984), 357-359.
[9] A. C. Zaanen: ‘Some remarks about the definition of an Orlicz space’,
Lecture Notes 945 (1982), 263-268.
Chapter 7
ON APPROXIMATE
ALGEBRA HOMOMORPHISMS
Chun-Gil Park
DaeJeon 305-764, South Korea
cgpark@math.cnu.ac.kr
Abstract We are going to prove the generalized Hyers-Ulam-Rassias stability of modi

fied Popoviciu functional equations in Banach modules over a unital C
*
-algebra.
It is applied to show the stability of algebra homomorphisms between Banach
algebras associated with modified Popoviciu functional equations in Banach al
gebras.
2000 MSC: 39B52, 47B48, 46L05 (Primary).
Keywords: Banach module over C -algebra,

* stability, unitary group, modified Popoviciu
functional equation, approximate algebra homomorphism
1. Introduction
Let Ei and E% be Banach spaces with norms || • || and || • ||, respectively.
Consider f: Ei E% to be a mapping such that /(irr) is continuous in t € R
for each fixed x e Ei. Assume that there exist constants e > 0 and p G [0,1)
such that
\\f(x + y) - f(x) - /(y)|| < £(||< + IK)
for all x,y e Ei. Th.M. Rassias [7] showed that there exists a unique R-linear
mapping T: Ei —> E^ such that
\\f^-T^\\<^\\x\f
for all x G Ei.
91
In [9, Theorem 2.1], T. Trif proved that, for vector spaces V and W, a
mapping f: V —> W with /(0) =0 satisfies the Popoviciu functional equation
3/ P + + + /(
*
) + /(y) + /(z)
\ O J
^(x-\-y\ „ (V + z\ „ (z + x\\
(/ +/ (1)
for all x, y, z e V if and only if the mapping f: V —> W satisfies the additive
Cauchy equation f(x + y) = f(x) + f(y) for all x^y eV.
We are going to give a modification of the Popoviciu functional equation.
Lemma A Let V and W be vector spaces, and d and r positive integers.
1) A mapping f: V —» W with f (0) = 0 satisfies the functional equation
d(rx + ry + rz\
-f -------- --------- + /(
)
* + f(y) + f(z)
r \ d /
(A)
for all x,y,z e V if and only if the mapping f:V —> W satisfies the
additive Cauchy equation f(x + y) = f(x) + f(y)for all x^y eV.
2) A mapping f: V —> W with /(0) = 0 satisfies the functional equation
for all x,y,z e V if and only if the mapping f: V —» W satisfies the

additive Cauchy equation f(x + y) = f(x) + f(y) for all x^y eV.
PROOF: 1) Assume that a mapping f:V —> W satisfies (A). Put y = z = x

in (A). Then (^) + = 6/(rr) for all x e V. So f (^) = ^/(x)
for all x eV. Hence
3
for all x, y, z € V. So we get the equation (i). By the Trif’s theorem, the
mapping/: V —> W satisfies the additive Cauchy equation f(x+y) = /(z) +
f(y) forallx,y e V.
The converse is obvious.
2) Assume that a mapping f: V —> W satisfies (B). Put y = z = x in (B).
Then 6/(x) = ^/ (^) for all x e V. So f (^) = ^/(x) for all x e V.
Hence
for all x, y, z € V. So we get the equation (i). By the Trif’s theorem, the
mapping f: V —> W satisfies the additive Cauchy equation f(x+y) = f(x) +
/(y) for all x, y e V.
The converse is obvious. □
Throughout this paper, let A be a unital C -algebra

* with norm | • |, U(A)
the unitary group of A. Let a 23 and aC be left Banach A-modules with norms
|| • || and || • ||, respectively. Let d and r be positive integers.
The main purpose of this paper is to prove the generalized Hyers-Ulam-
Rassias stability of the functional equations (A) and (B) in Banach modules
over a unital C -algebra,
* and to prove the generalized Hyers-Ulam-Rassias
stability of algebra homomorphisms between Banach algebras associated with
the functional equations (A) and (B).
2. Stability of modified Popoviciu functional

equations in Banach modules over a
C
-algebra
*
We are going to prove the generalized Hyers-Ulam-Rassias stability of the
functional equations (A) and (B) in Banach modules over a unital C
* -algebra
associated with its unitary group.
For a given mapping /: —> 4C and a given a € A, we set
Daf(x, y, 2) := -af (rx + ry + rz\ + + af(yj + af^

r \ & /
for all x, y, z G A&.
Theorem 2.1 Let f: a& aC be a mapping with /(0) = 0 for which there
exists a function <p: a&3 —> [0, oo) such that
(2.i)
IIDuf(x, y, z) || < y>(x, y, z) (2.ii)
for all u G U(A) and all x,y,z G a&- Then there exists a unique A-linear
mapping T: a& —> aC such that
||/(x) - T(x) || < |^(z, x, x) (2.iii)

o
for all x € a&-
PROOF: Put u — 1 6 U (yl). Let x = y = z in (2.ii). Then we get
< X, x)
r \ a J
for all x e a^- So one can obtain
- ^-f 1 /
or o
and hence
( d\n (("ir\n /3r\n /3r\n \

^3^; x\~d) x,u)x)
for all x e a&- So we get
(2.1)
for all x e a-B.

Let x be an element in a&- For positive integers n and m with n > m,
n m
d \
f f
3r /
which tends to zero as m —> oo by (2.i). So { (^)” f ((^)n j is a Cauchy
sequence for all x e a&- Since aC is complete, the sequence

converges for all x e a&- We can define a mapping T: a& —* aC by
T(x) = lim (2.2)

n—kx
for all x € a&-

By (2.ii) and (2.2), we get
||£>iT(t,?/,2)||
/ 7 \ n
= lim f—
n—>oo \ or /
for all x, y, z e a&- Hence DiT(x,y,z) = 0 for all x, y, z e a&- By

Lemma A, 1), T is additive. Moreover, by passing to the limit in (2.1) as
n —> oo, we get the inequality (2.iii).
Now let L: 4 23 —> aC be another additive mapping satisfying
\\f(x) - £(a:)|| < ^x,x,x)
o
for all x € aI3-
/3r\n /3r\n \
< —
- I — I
2\^r) — JI X.'\d
I\\d I --- J X 7 ,’
— J X,' \d
which tends to zero as n —> oo by (2.i). Thus T(x') — L(x) for all x G a&-
This proves the uniqueness of T.
By the assumption, for each u G U(A),
f d\n r. f3r\n /3rV \

11W x’ W V
i \ n / /o \ n /v \ n /o \ n
I or\ \\
(—3ra J\ (( or \ [ or\
II—r I x. ( —-I x, —
\\ dJ \ dJ \ dJ J
x ]
for all x e and
for all u e U(A) and all x e a&- Hence
d /3r \
DuTfjC) x, x) = -uT I —x ) + 3uT(x) — &I\ux) = 0
r \ d J
for all u e U{A) and all x e So
uT(x) = T(ux)
for all u e U(A) and all x e a&-

Now let a e A (a 0) and M an integer greater than 4 |a |. Then
|A| - J_| | lQl _ 1 i ? _ 1

ImI ” M'a' < 4|a| “ 4 < 3 —3
By [6, Theorem 1], there exist three elements ui,U2,uz e such that
3-^ = ui + rt2 + 1/3. And T(x) = T (3 • = 3T Qa;) for all x G A&- So
T (|t) = |T(x) for all x G a&- Thus
T(ax) =t(^--3-^x\ =M-T (l-3^-x\ = ^-T (3-^-x)

v ’ 1^3 M ) \3 M J 3 \ M /
M M
= —T(uix + U2X + u^x) = —(T(u]x) + T(u2x) + Tû^x}')
o o
= v(wi + ^2 + u3)T(x) = ■ 3^-T(x) = a.T(x}
O O 1V1
for all x e a&- Obviously, T(0x) = OT(rc) for all x G a&- Hence
T(ax + by) = T(ax) + T(by) = aT(x) + bT(y)
for all a, b € A and all x,y G a&- So the unique additive mapping T: a& —>
aC is an A-linear mapping, as desired. □
Applying the unital C'

-algebra
* C to Theorem 2.1, one can obtain the fol
lowing.
Corollary 2.1 Let E\ and be complex Banach spaces with norms || • || and
|| • ||, respectively. Let f: Ei —> E? be a mapping with /(0) = 0 for which
there exists a function <p: Ef —> [0, oo) such that
||.Da/(z,?/,2)|| < <p(x,y,z)
for all A G T1: = {A G C : |A| = 1} and all x,y,z e E±. Then there exists
a unique C-linear mapping T: Ei —» such that
||/(x) -T(x)|| < }-p(x,x,x)

o
for all x e Ei.
Theorem 2.2 Let f: a& —» aC be a mapping with J(0) = Ofor which there
exists a function cp: aZ33 —> [0, oo) such that
for all u e U(A) and all x,y,z e a&- Then there exists a unique A-linear
mapping T: a& —> aC such that
||/(x) -T(a;)|| < ^<p(x,x,x)

(2.vi)
for all xEaB-
Proof: Put u = 1 G U (?1). Let x = y — z in (2,v). Then we get
6/(x) - 3-/ < cp(x, X, x)

r \ a J
for all x e So one can obtain
2r \
)
/(
* - 7T-f d xJ
— < -<p(x,x,xf
Zr
and hence
1/dV //2r\n /2rV /2r

Hw x\~d)
for all x e a&- So we get
/(*)
I (2.3)
for all x 6 aB.

Let x be an element in aB- For positive integers n and m with n > m,
which tends to zero as m —> ooby(2.iv). So {(57)”/ isa Cauchy

sequence for all x e aB. Since aC is complete, the sequence {(^)n f
converges for all x e a&- We can define a mapping T: a& —* aC by
2r\n
T(x) = lim ~d) X (2.4)
n—»oc
for all x e a&-

By (2.v) and (2.4), we get
= lim (■£-
n—>oo \ 2r
r ( d
< lim —
n—»oo \ 2r
for all x,y,z € aB- Hence DiT(x, y, z) = 0 for all x, y, z G aB- By

Lemma A, 2), T is additive. Moreover, by passing to the limit in (2.3) as
n —> oo, we get the inequality (2.vi).
Now let L: aB —* aC be another additive mapping satisfying
\\f(x) - I(x)|| < ^p(x,x,x)
for all x G a&-
which tends to zero as n —> oo by (2.iv). Thus T(x) = L(x) for all x G aB-
This proves the uniqueness of T.
By the assumption, for each u G U(A),
for all x G aB, and
2r /
as n —> oo for all x G aB- So
V/2r\n /2rV (2r\n
DuT\x,x,x) = hm — I Duf — I x, — x, — =0
n—>oo \2r J \\ a / \ a I \ a )
for all u G U(A) and all x € a&- Hence
d ( 2t \
DuT(x, x, x) = 3uT(x) + 3uT(x) — 3-T [ —ux I = 0
r \ d J
for all u e and all x e a&- So
uT(x) = T(ux)
for all u e U(A) and all x E

The rest of the proof is the same as the proof of Theorem 2.1. □
From now on, assume that a&3 —> [0, oo) is a function satisfying (2.i).
We are going to prove the generalized Hyers-Ulam-Rassias stability of the
functional equation (A) in Banach modules over a unital C
*
-algebra.
Theorem 2.3 Assume that is not an integer. Let f: a& aC be a mapping

with /(0) =0 such that
\\Dif(x,y,z) || < <p(x,y,z)
for all x,y,z G ,4$. Then there exists a unique additive mapping T: a& —> aC
satisfying (2.iii). Further, if /(Ax) is continuous in A € R for each fixed
x e a^> then the additive mapping T: a& —> aC is ^.-linear.
Proof: By the same reasoning as the proof of Theorem 2.1, there exists a
unique additive mapping T: a& —> aC satisfying (2.iii).
Assume that f(Xx) is continuous in A G R for each fixed x G aB. By
the assumption, is a rational number which is not an integer. The additive
mapping T given above is similar to the additive mapping T given in the proof
of [7, Theorem]. By the same reasoning as the proof of [7, Theorem], the
additive mapping T: a& —> aC is R-linear. □
Similarly, for the case that tp: AB3 [0, 00) is a function satisfying (2.iv),
one can obtain a result similar to the theorem given above.
3. Stability of modified Popoviciu functional

equations in Banach algebras and
approximate algebra homomorphisms
In this section, let A and B be Banach algebras with norms || • || and || • ||,
respectively.
D.G. Bourgin [2] proved the stability of ring homomorphisms between Ba
nach algebras. In [1], R. Badora generalized Bourgin’s result.
We prove the generalized Hyers-Ulam-Rassias stability of algebra homo

morphisms between Banach algebras associated with the functional equation
(A).
Theorem 3.1 Let A and B be real Banach algebras, and not an integer.
Let f: A —> B be a mapping with /(0) = Qfor which there exists a function
: A x A —> [0, oo) such that
(3.i)
IPi/(^,y, Oil < <p(x,y,z), (3.ii)

\\f<A ■ y) - /(O/(y)II < V’(®, y) (3.iii)
for all x,y, z € A. If ffXx) is continuous in A € R/or each fixed x € A, then

there exists a unique algebra homomorphism T: A —> B satisfying (2.iii).
Further, if A and B are unital, then f itself is an algebra homomorphism.
Proof: Under the assumption (2.i) and (3.ii), in Theorem 2.3 we showed that
there exists a unique R-linear mapping T: A —> B satisfying (2.iii). The R-
linear mapping T: A —> B was given by
/dV
Tfx) = lim f
j—>oo \ or /
for all x e A. Let
R{x,y) = f(x-y) - f(x)f(y)
for all x, y e A. By (3.i), we get
/ d \J ( f 3r V \
lim — R -y z, y = 0
j—+oo yor/ \ y aJ /
for all x, y G A. So
T(x • y) = lim ( — ) f ( — ) (x • y)
j—>oo \or J \\ a J J
r ( d\j f (( (3r\j \ \
= lim q- j Hr x y
j—>oo yor/ \ \ \a / 1 /
/d\J/ //3rAJ \ Z/3rV
= .lim ( n- ) \f ( -T ) Z /(y) + R I ( -T ) X, y
j—>oo \or J \ \\ a / j \\a J
— T(x)f(y) (3.1)
for all x, y € A. Thus
/ \i \ ( // V \\
(
/ /3r\3 \ \
\
//TrV \
((dj y)J
/
( -
\ \ aJ
V
x y =T
j 1
-
\ \ dJ
x /(y)
1
= -7 rW/W
\ a j
for all x, y e A. Hence
/ d V / /3rV \
d f v) y = Ti-x'>Ky'> <3-2>
for all x, y e A. Taking the limit in (3.2) as j —> oo, we obtain
T(xyT(y) = T(x)f(y)
for all x, y G A. Therefore,
T(x-y) = T(x)T(y)
for all x, y g A. So T: A —> B is an algebra homomorphism.

Now assume that A and B are unital. By (3.1),
T(y) = T(l-y) = T(l)/(y) = /(y)
for all y G A. So f: A —> B is an algebra homomorphism, as desired. □
Theorem 3.2 Let A and B be complex Banach algebras. Let f: A —* B be a

mapping with /(0) = Ofor which there exists a function ■ A x A —> [0, oo)
satisfying (3.i) and (3.iii) such that
\\Dxf(x,y,z)\\ <tp(x,y,z) (3.iv)
for all A € T1 and all x,y,z € A. Then there exists a unique algebra homo-
morphism T : A —» 13 satisfying (2.iii). Further, if A and B are unital, then f
itself is an algebra homomorphism.
Proof: Under the assumption (2.i; and (3.iv), in Corollary 2.1 we showed that
there exists a unique C-linear mapping T : A —> B satisfying (2.iii).
Theorem 3.3 Let A and B be complex Banach ^-algebras. Let f: A —» B be

a mapping with /(0) = Qfor which there exists a function ip: A x A —> [0, oo)
satisfying (3.i) and (3.iii) such that
y, z)|| < y, z), (3.iv)
ll/O
)
* ~ /O)
ll
* < <p(x,x,x) (3.v)
for all A 6 T1 and all x.y.z € A. Then there exists a unique * -algebra
homomorphism T : A —> B satisfying (2.iii). Further, if A and B are unital,
then f itself is a *
-algebra homomorphism.
unique C-linear mapping T: A —> B satisfying (2.iii).
Now
for all x e A. Thus
as j —> oo for all x € A. Hence
for all x € A.
Similarly, for the case that p: A3 —> [0, oo) and ip: A x A —> [0, oo) are
functions such that
for all x, y, z e A, one can obtain similar results to the theorems given above.
References
[1] R. Badora: ‘On approximate ring homomorphisms’, J. Math. Anal. Appl.

276 (2002), 589-597.
[2] D.G. Bourgin: ‘Approximately isometric and multiplicative transforma
tions on continuous function rings’, Duke Math. J. 16 (1949), 385-397.
[3] D.H. Hyers, G. Isac and Th.M. Rassias: ‘Stability of Functional Equations
in Several Variables', Birkhauser, Berlin-Basel-Boston, 1998.
[4] G. Isac and Th.M. Rassias: ‘On the Hyers-Ulam stability of ^-additive
mappings’, J. Approx. Theory 72 (1993), 131-137.
[5] B.E. Johnson: ‘An approximately multiplicative map between Banach al
gebras’, J. London Math. Soc. 37 (1988), 294-316.
[6] R. Kadison and G. Pedersen: ‘Means and convex combinations of unitary
operators’, Math. Scand. 57 (1985), 249-266.
Proc. Amer. Math. Soc. 'll (1978), 297-300.
[8] H. Schroder: ‘K-Theory for Real *
-Algebras
C and Applications', Pitman
Research Notes in Mathematics 290, Longman, Essex, 1993.
[9] T. Trif: ‘Hyers-Ulam-Rassias stability of a Jensen type functional equa
tion’, J. Math. Anal. Appl. 250 (2000), 579-588.
Chapter 8
HADAMARD AND
DRAGOMIR-AGARWAL
INEQUALITIES, THE EULER FORMULAE
AND CONVEX FUNCTIONS
Josip Pecaric
Faculty of Textile Technology, University of Zagreb
Pierottijeva 6, 10000 Zagreb, Croatia
pecaric@hazu.hr
Ana Vukelic
Faculty of Food Technology and Biotechnology, University of Zagreb
Pierottijeva 6, 10000 Zagreb, Croatia
avukelic@pbf.hr
Abstract The Euler formula is used with functions possessing various convexity and con
cavity properties to derive inequalities pertinent to numerical integration.
2000 MSC: 26D15, 26D20, 26D99 (Primary).
Keywords: Hadamard inequality, r-convexity, integral inequalities, Euler trapezoid formula,

Euler midpoint formula, Euler-Simpson formula, Euler two-point formula, dual
Euler-Simpson formula, Euler-Simpson 3/8 formula, Euler-Maclaurin formula
1. Introduction
One of the cornerstones of nonlinear analysis is the Hadamard inequality,
which states that if [a, 6] (a < b), is a real interval and f: [a, b] —> R is a
convex function, then
(1.1)
105
Recently, S.S. Dragomir and R.P. Agarwal [8] considered the trapezoid for
mula for numerical integration for functions such that \f'\q is a convex function
for some q > 1. Their approach was based on estimating the difference be
tween the two sides of the right-hand inequality in (1.1). Improvements of their
results were obtained in [11]. In particular, the following tool was established.
Suppose f: 1° C R —> R is differentiable on 1° and that | f |9 is convex on
[a, 6] for some q > 1, where a, b G 1° (a < b). Then
In this paper we consider some generalizations to higher-order convexity and

applications of this results and further related results. We will use the interval
[0,1] because of simplicity and it involves no loss in generality.
The most natural nexus for those developments would appear to be the well-
known Euler formula
=/(t) dt+£ ^Bt w
- 1 f /'”’(«) [BI (»-()- B’ «] dt (1.3)

* «/ 0
(see [9, p. 17]), which holds for every x e [0,1] and every function f: [0,1]
R with n > 2 continuous derivatives. Here Bfc(-), k > 0, is the fcth Bernoulli
polynomial and Bk = Bfc(O) = Bfc(l) (k > 0) the fcth Bernoulli number. We
denote by B£(-) (k > 0) the function of period one with B£(x) — Bk(x) for
0 < x < 1.
2. The main results

First we start with a result of Hadamard’s type (see [2]).
Theorem 2.1 Suppose f: [0,1] —> R is a (2r + 2)-convex function. Then
l-ftfrl ,(2r) MA
(2r)!Z \2J
+E^[/(2fc-1)(i)-/(2fc-1)(°)] >
|B2r|/(2r)(l) + /(2r)(0) ,
~ (2r)! 2 ’ 1
Hadamard and Dragomir-Agarwal inequalities 107
Iff is (2r 4- 2)-concave, the inequality is reversed.
PROOF: For x = 1 and n — 2r, (1.3) becomes
/(1)= f f(t)dt + [fO) — f(ty]Bi

Jo
2r—l
+ E Sf [/’‘-’’(I) - 7l‘-,)
k=2
Since Bi = 1/2 and B2j+i = 0 for j > 1, we have
■f(0) + /(1) = f 7(0 dt + £ ■>(i) - /®-‘i(0)]

2 Jo \ZJ)' L J
(2r)! Jo
Further, B2r(l ~ 0 = B2r(i), so we may write this as the Euler trapezoid

formula
f f^dt = 5 [7(0) + 7(1)] - E [7(2‘~11 (1) - 7(“-1) (0)]

JO Z fc=1 L J
+ r P2r(0/(2r)(0^, (2-2)
Jo
where Pn(t) := [Bn(t) — Bn]/nl (n > 1). See [1].
Note that P2r(t) does not change sign on the interval [0,1] and
Pn(0) = Pn(l) = 0, P4(0=Pn-l(0+&n-b n>l,
where
h -n h - ^2fc - f 1^+1 2 Vs 1
&2fc-t - 0, b2k - - ( 1) (2fc)2fc nk,
and
f1 Pir(t)dt = -b2r = -^-. (2.3)
Jo (2r)!
Since (—l)r+1B2r > 0, we have (—l)rP2r(f) > 0.
If f is a (2r + 2)-convex function, then is convex and the following

inequality is valid
y(2r)(i) = /(2r)((1 _ . q +t. < t/(2r)(1) + (1 _
So, using the discrete Jensen inequality for the convex function f(2r\ we have
from (2.2) that
dt
/(2r)(0) f\l-t)P2r(t)dt .
tP2r(t) dt +
JO
(2.4)
Note that by (2.3)
f tP2r(t)dt+ [ (1 - tjPzrft) dt = [ P2r(t)dt = —b2r- (2.5)

Jo Jo Jo
On the other hand, by partial integration we have
[ tP2r(t) dt = t(P2r+1(t) - fort) |o “ / (Ar+l(f) “ W) dt, (2.6)

Jo Jo
where we have used substitutions u = t, dv = P2r(f) dt so that du = dt and
v = P2r+i(f) — 62rt. Therefore,
fl fl t2
/ tP2r(t)dt = -b2r - / P2r+l(t)dt + fe2r— Io
Jo Jo
If1 1
~ 2^r J(j dt = — — 62r.
With respect of (2.5), we have
/ tP2r(t)dt = [ (1 - i)P2r(t) dt = -^b2r.

Jo Jo 2
Now (2.4) becomes
rr < / 1V+1 B2r /^(l) + (0) _ |P2r| /^(l) + /^(O)
r - 1 ; (2r)! 2 “ (2r)l 2
By Jensen’s integral inequality

1
o
/•l
f0\-iyp2r(t)dt
which proves our assertion. □

Theorem 2.2 Suppose f: [0,1] —> R is (2r)-times differentiable
a) is convex for some q > 1, then
rl 1
y B2k
91 V9
(2.7)
- (2r)! 2
is concave, then
1 1
Jo *
y
< |B2r| (2r) f
~ (2r)! J V (2.8)
PROOF: Let Hr be defined as in (2.4). So, by the power mean inequality we

have
/•t , । / yl \
|#r| < yo |P2r(i)| • |/(2r)(i)| dt < ( / |P2r(f)| dt\ X
Now, by Jensen’s inequality
Moreover, if | f(2r) |9 is concave then
|#r| < |P2rW| ’ |/(2r)W| dt < (J° |P2r(t)| dt^ X
A2r) f fp l^2r(^)| = lg2rl f(2r) (JA

(2r)! J \2j
and the theorem is completely proved. □

Remark 2.1 For (2.8) to be satisfied it is enough to suppose that |/(2r)| is a
concave function. In fact, if |g|9 is concave on [0,1], for some q > 1, then for
x, y, A e [0,1]
|p(Ax + (1 - A)y) I’ > A |p(x) |9 + (1 - A) |<?«

> (A|<7(»| + (1 - A) |5(y)|)9
by the power mean inequality. Therefore, |p| is also concave on [0,1].
Remark 2.2 It is simple to prove that if f: [a, b] —> R is (2r + 2)-convex,

then
)2r+1^/(2'> - H'(a’b>
(2r)! 2
Also, if f: [a, 6] —> R is (2r) -times differentiable then

a) If \f(2r>\g is convex for some q > 1,
91 V9
< (6-«)2r+1^ |/<2r>(< + |/M(t)|
2
b) If |/(2r> | is concave,
< (t>-«)2r+1^JT
where
;= (-If fit) dt - b-^- <J(a) + /(b)]
To obtain appropriate results for numerical integration from the Euler trapezoid
formula, we will apply the results above to each interval of the subdivision
[0, h], [h, 2/i],... , [(n — l)/i, nh\.
Let us denote
n— 1
T(f',h) = h + +
k=l
and
n— 1 z 1
M(f-h') = h^f(kh-^h
k=L '
Theorem 2.3 Iff: [0, nh] —> R is (2r + ^-convex, then
h2T^M
(2.10)
Iff is (2r + 2)-concave, the inequalities are reversed.
PROOF: It follows from Theorem 2.1 and Remark 2.2, since
[/2fc-1)(n/i)-/2fc-1)(0)]
n •mh i
f(t)dt — -[/((m-l)/r) + f(mli)] +
(2.11)
fc=l
□
Theorem 2.4 Suppose f: [0, nh] —> R is (2r)-times differentiable,
a) Z/|/(2rf is convex for some q > 1, then
pnh
B2kh2k
(2/c)!
fc=l
n 91 i/9
(2r)!
v ' m=l L 2
n/i2r+1 max (2.12)

(2r)!
b) (f|/(2r)| is concave, then
rnh
/ Wdt-Ttf-h)
Jo
Proof:
nh
0 fc=i
n ’ t-mh i.
/ /(f) dt - - [/((m - l)/i) + f(mh)]
J (m—l)h
k=l
n
fmh h
I f(t)dt--[f((m-l)h)+f(mh')]
m=l (m— l)h
fc=l
n 91 V9
- (2r)l 2
m=l
by Theorem 2.2 and Remark 2.2 applied to each interval [(m — l)/i, mh].
Hence
B2kh™ [/(2fc-1)(n/z)-/(2fc-1)(0)]
/(i)df-T(/;A) +
(2fc)l
m=l
The result of a) now follows from the convexity of |/(2r) |g.

The proof b) is similar. □
The resultant formulae in Theorems 2.1-2.4 when r = 1 and the sums in

Hr are empty are of the special interest and we isolate them as corollaries.
Corollary 2.1 If f: [0,1] —> R is 4-convex, then
(|) < | [/(0) + /(!)] f(t) dt R is 2-times differentiable,

a) tf\f"\q is convex for some q > 1, then
jG(^~[/(O)+ /(!)] R is 4-convex, then
h2 fnfl h2
-M (/"; ft) < T(f- K) - 2 /(t) dt < —T (/"; ft).
If f is ^-concave, the inequalities are reversed.
Corollary 2.4 Suppose f: [0, nh] —> R is 2-times differentiable,
a) tf\fH\q is convex for some q > 1, then
f(t)dt-T(f-,h)
< — max {|/"(0)|, |/"(nft)|} .

1 Zi
b) If\f"\ is concave, then
h2
f(t)dt-T(f-,h)
3. The Euler midpoint formula

In this section we take a different path from (1.3) (see [3]), one leading to
the Euler midpoint formula instead of the Euler trapezodial formula. In place
of the function P^r of the trapezodial formula, it turns out that we shall have
recourse to
/1 _ 9/ \ /1 \
P2r(«)=B2*r -B2r - .
We note that this does not change sign on the interval [0,1] and that it is sym
metric about t = So, we have
[ (1-f)|p2(f)|<ft = / i|p2(i)|di.
Jo Jo
Further (—l)r-1p2r(i) > Ofori e [0,1]. Then
[ (1-f)|p2(f)|cft + [ i|p2(i)|di = [ |p2r(i)|di

Jo Jo Jo
= (-If-1 [ P2rdt
Jo
= (-l)r-1(l - 21-2r)B2r = (1 - 21-2r)\B2rI,
so that
j\l - i)|p2(i)|di = = |(1 - 21-2r)|B2r|.
Put x = | and n = 2r in (1.3). Since B2j+i(1/2) = 0 for j > 0, we have

obtained the Euler midpoint formula
/Q) = p(o
*+EpfMD -/--»(»)]
Note that B2fc(l/2) = (l - 21-2fc)B2fc.

For the sequel we shall utilize
We can parallel the development of the previous section with the following
two theorems.
Theorem 3.1 Suppose f: [0,1] —> R is (2r + 2)-convex. Then

|B2r| ( _ _J_\ /(2r)(l)+/(2r)(0)

~ (2r)! V 22r-! J 2
Iff is (2r + 2)-concave, the inequality is reversed.
Theorem 3.2 Suppose f: [0,1] —> R is (2r)-times differentiable,

a) Z/|/<2r)|9 is convex for some q > 1, then
b) If\f^\ is concave, then
\M । < IdM A____ __ A f(2r) (

1 r|“(2r)!V 22r-17/ V2/
To obtain appropriate results for integration from the Euler midpoint formula,
we apply the above results to each interval of the subdivision
[0, /i], [/i, 2h],... , [(n — l)/i, nh]
First we introduce
nh
U
- - /(2"1)(0)] '■ ■
Theorem 3.3 If f: [0, nh] —> R is (2r + 2)-convex, then
/l2r-y—-ry A _ 1\ M Zy(2r).^\ < M*

(2r)! \ 22r-1 J V J r
< f1” •
(2r)! \ 22r-1 J V ’ /
If f is (2r + 2)-concave, the inequalities are reversed.
Theorem 3.4 Suppose f: [0, nh] —> R is (2r)-times differentiable.

a) If\f^\q is convex for some q > 1, then
\\M < h2r+1

* । fl------ - —
1 rl~ (2r)! V 22r-1/
b) //|/(2r) | is concave, then
The resultant formulae in Theorems 3.1-3.4 when r = 1 and the sums in Mr

and M* are empty are of the special interest and we isolate them as corollaries.
Corollary 3.1 If f[0,1] —> R is ^-convex, then
1 /"(0) +/"(!)
24 2
Iff is 4-concave, the reversed inequalities apply.

a) If\f"\qis convex for some q > 1, then
< _1_
“ 24 2
b) If\f"\ is concave, then
1
- 24 \2J
Corollary 3.3 Iff: [0, nh] —> R is 4-convex, then
h2 rnh h2
If f is ^-concave, the inequalities are reversed.
Corollary 3.4 Suppose f: [0, nh] —> R is 2-times differentiable.

a) tf\f"\q is convex for some q > 1, then
rnh
Jo
< - V
- 24 2
m=l
Zt:
b) lf\f"\ is concave, then
h2
dt-M(f-K)
4. The Euler—Simpson formula

In this section (see [3]) we explore a path that is associated with the Euler-
Simpson formula.
If f is defined on segment [0,1] and has 2r (r > 2) continuous derivatives
there, then the Euler-Simpson formula (see [4]) states that
/(0)+4/Q)+/(l)
+ E 3(^)711 “ 22’2fc)B2fc - /(2fc-x) (0)]
f1 /(2r)mrm
3(2r)! Jo
where
F2r(t) - B2*r (1 -1) + 2B2*r (±-t}-B2r- 2B2r (.
\Z J \Z J
It was proved in [9] that F2r(l — t) = F2r(t) and that (—l)r~1F2r(t) > 0.
Also,
/’1|F2r(i)| dt = (1 - 22-2r)|B2r|.
Jo
We can parallel the development of the second section with the following
two theorems. Define
x,-(-ir1 ma-1- f(o)+tf Q)+f (i)

- £ [/<2^1)(D - /‘“-“’(O)] - ■
Theorem 4. 1 Suppose f: [0,1] —> R is (2r + 2)-convex. Then
-2—<X
< h _ o2-2rx/(2r)(l)+/(2r)(0)
- 3(2r)!1 ’ 2
Theorem 4. 2 Suppose f: [0,1] —» R is (2r\-times differentiable,

a) |/(2r)|g is convex for some q > 1, then
b) is concave, then
lx'sW(1-2"2r /'2r>
To obtain appropriate results for integration via the Euler-Simpson formula,
[0,2h], [h, 4h],... , [2(n — l)/i, 2nh]
First we introduce
n—1 n—1
/(0) + 7(2^) + 2 f (2z/i) + 4 f ((2z - l)h)
o
i=l i=l
and
f r2nh
*
x ■.= (-if-112 f(t)dt — X(f;h)
r— 1 /Qj \2k "

- E Skd
3'2“
- L
ml J .
Theorem 4.3 If f : [0,2nh] —> R is (2r + 2)-convex, then
- 22’2r)M (fW-,2h) < Xr*

< (2^2^(1 - 22-20T .
Theorem 4.4 Suppose f: [0,2nh] —> R is (2r)-times differentiable.
a) Z/|/(2r)|9 is convex for some q > 1, then
l^l<(2fc)2r+1^4(l-22-2’')
” r|/P'>(2™<+ |/^>(2(m- Wl’l1'’

x 2
m=l L
x max {|/(2r)(0)|,|/2r\2n/l)|}.
b) If\fW\ is concave, then
|x;i < (2/i)2r+1 (i - 22-2om (|/(

-)|
* -2h).
The resultant formulae in Theorems 4.1^1.4 when r = 2 and the sums in Xr

* are empty are of special interest and we isolate them as corollaries.
and X
Corollary 4.1 If f: [0,1] —» R is 6-convex, then
+ +/(!)]
< 1 /(4)(0) + /(4)(l)

- 2880 2
Corollary 4.2 Suppose f: [0,1] —> R is 4-times differentiable.
a) If |/(4) |9 is convex for some q > 1, then
£f(t)dt-± /(0)+4/Q)+/(l)
g"| 1/^
1 |/(4)(0)|g+ |/(4)(1)|
- 2880 2
b) is concave, then
f /(
)
* - I [/(0) 4-4/fi')-f-/(1)1 ■
Jo o |_ \“ / J 2ooU \"/
Corollary 4.3 Iff: [0,2nh\ —> R is 6-convex, then
(f^;2h\ < X(f-, h) - [2nk f(t) dt < (/(4); 2h) .

2880 V ’ 7 “ u ' Jo J ’ “ 2880 V ’ )
If f is 6-concave, the inequalities are reversed.
Corollary 4.4 Suppose f: [0,2nh] —> R is 4-times differentiable,

a) If |/(4)|9 is convex for some q > 1, then
dt — X(f;li)
< (2fr)5 A |/(4)(2m/<+|/(4)(2(m-l)/<~|1/<?
“ 2880
m=l
n(2h)5
2880
b) Zf|/(4)| is concave, then
f(t)dt — X(f’,h) (|/(4)| ;2h)

- 2880
5. The Euler two-point formula

In this section we explore a path that is associated with the Euler two-point
formula.
there, then the Euler two-point formula (see [10]) states that
+E - 31_2/w [/(2fc-x)(i) - /(2fc-x)(o)]

where
= Bl, (1 - t) + Bl, - <) - B2, (1) - B2r (|) .
We note that this does not change sign on the interval [0,1] and
F£ (I) = 2(1 “ 2’2r)(1 " 31’2^B2-
Since (—l)r+1B2r > 0, we have (—l)r-1F^(i) > 0.

It was proved in [10] that
Jo
Also, F£(l - f) = E£(f), so
|f£(
*
)| dt = j\i -1) |F£(i)| dt = |(i - 31-23|B2r|.
We can parallel the development of the second section with the following two
theorems. Define
IB2.I fl 01-2nf<2,) m R is (2r)-times differentiable.

a) If |/(2r)|9 is convex for sonie q > 1, then
ip I < l^2rl (i _ 31-2r

|ri-2(2r)!( 2
b) is concave, then
|-^2r| ol-2r\ /(2r)

\Pr\< 2(2r)! 1 J 7
To obtain appropriate results for integration via the Euler two-point formula,
[0, h], [h, 2h],... , [(n — l)h, nh\.
First we introduce
2 fc=i L '
and
nh
U
r-1 h2k r 1'
k^l1 JJ
Theorem 5.3 If f: [0, nh] —> R is (2r + 2)-convex, then
- 31-2rW (/(2r);/>) < Pr

*
-ft2r2M(1-31’2r)T (/(2M-

a) If |/2rf is convex for some q > 1, then
(1 - 31“2r)
- n/l2r+12M(1 ~ 31’2r) max {|/(2r)(°)| > |/(2r)W|} •

b) is concave, then
|P*1 - h"T2^)!(1 “ 31’2r)M (K’l •h) ■
The resultant formulae in Theorems 5.1-5.4 when r = 2 and the sums in Pr

and P
Corollary 5.1 Iff: [0,1] —> R is 6-convex, then
—/(4) Cf^dt
1944CT V2 / 2l \3/ \3/J Jo
< 13 /W(0) + /W(l)
“ 19440 2
If f is 6-concave, the reversed inequalities apply.

a) If\f(4">\g is convex for some q > 1, then
'|/W(0)|g + |/W(i)|a]I/’
< 13
- 19440 2
b) is concave, then
©
(/«>; /.) < P(f; !■} - f (/<«; K) .

a) Z/|/(4)r is convex for some q > 1, then
rnh
Jo
< 13/l5 y-
~ 19440 2
m=l _
b) is concave, then
rnh
/ f(t) dt-p(f'h) < 13/i4
Jo “ 19440
6. The dual Euler—Simpson formula

In this section we explore a path that is associated with the dual Euler-
Simpson formula.
there, then the dual Euler-Simpson formula (see [5]) states that
1 /q3—4/c _ 3.2l~‘2k + l)B2fe [/(2fc-1)(l) - /(2fc-1)(0)]

3(2fc)P
where
1 \ /1\ /q \
(-~i
41 / -b;,l;-t +2s;r --t
J
— 2B2r
\ 41 /
+ B2r — 2B2r
--2(l-2-2r)B2r.
Since (—l)r+1B2r > 0, we have (—l)rF^(f) > 0.

f |F£(f)| dt = (23-4r - 3 • 2x-2r + 1)|B2r|.

Also, F2^(l - t) = Fg(t), so
]\\Fg(t)\dt = jf(l-t)|j£(i)| dt = l(23-4’--3,2>-2’- + l)|B2r|.
two theorems. For the sequel we shall utilize
A := (-1)' /(i) dt - 1 [2/ (1) - / Q) + 2/ (|)
+ E 3(^1 (23-" - 3'2' 2‘' + [/' “ l I1) - /(“-1|(0)] > ■
Theorem 6. 1 Suppose f: [0,1] —> R is (2r + 2)-convex. Then
l^2r| /r>3—4r q *
rjl—2r । i\/
(2r) (
<D < , ,1-2. , n/(2r)(l)+/<2r)(0)

~ rS3(2rj!1 +1) 2
Theorem 6. 2 Suppose f: [0,1] —» R is (2r)-times differentiable,

a) If\fW\q is convex for some q > 1, then
3^(23~4r"3’21~2r + 1)
|£>r| - 3^j! (23’4r " 3 ’ 21’2r + /(2r) (I)
To obtain appropriate results for numerical integration from the dual Euler-
Simpson formula, we apply the results above to each interval of the subdivision
[0, h], [h, 2h],..., [(n — l)h,nh]
Let us denote
= (kh-^h\ - f f kh - J + 2/ (kh -
6 t=i L '
and
rnh
D* ■■= (-1) b f0
dt - D(f; ft)
+ E ^?(23’4fc " 3 ’ 21“2fc + .

frf 3(2fc)! L J
Theorem 6.3 Iff: [0, n/i] —> R is (2r + 2)-convex, then
^2r l^2rJ,23-4r _ 3.2l-2r + Zy(2r). A 1, then
*
|£)
| < ^2r+ljgd(23-4r _ 3.2l-2r +
< ^2r+1;^y(23_4r - 3 ■ 21-2r + 1)
x max||/(2r)(0)|, |/(2r)(n/i)|} .
b) Z/|/(2r)| is concave, then
|Z)r*| < /l2r^yi(23_4r - 3 • 21-2r + 1)M (|/(2r)| ;/>) •
The resultant formulae in Theorems 6.1-6.4 when r = 2 and the sums in Dr

and D
230407 \2/ “Jo 3 2/G)-JI)+yG)

< 7 /W(O) + /W(1)

S 23040 2

a) If\fW\q is convex for some q > 1, then
7 yw(o)i8+i/W(i)n1/5
- 23040 2
[ f(t)dt — - [2/ -f +2/ f-^1 <—— /(4)f-^

Jo 3[J \4J J \2J \4j J - 23040 J \2 J
(f^;h\ < [nk f(t)dt- D(f;h) < .

23040 V ’J ~ Jq v ~ 23040 V )

a) If\fW\9 is convex for some q > 1, then
f^dt-D(f-h-)
[|/<4>(m< + |/<4>((m-l)<] V’
< 7h5 V
“ 23040 2
m=l
b) is concave, then
f(f)dt-D(J;h)
7. The Euler—Simpson 3/8 formula

In this section we explore a path that is associated with the Euler-Simpson
3/8 formula.
there, then the Euler-Simpson 3/8 formula (see [6]) states that
Jf1 dt = 81 K /2\ + f (1)

/i\ + 3/ \3j
f (0) + 3/ \3j
+ E 8(^)!(1 " 32“2')B2fc [/(M(1) " /(2fc-1)(°)]
where
1 \ /o \
(o
t j + 3B2r
J
— 2jE?2r ~
\o
3jE?2r
J
“ 3jE?2r ( q ) •
F£ = 2(1 - 21-2r)(l - 32-2r)B2r.
Since (— l)r+1B2r > 0, we have (—l)r-1F^(i) > 0.

= (1 — 32-2r)|B2r|.
J0
Also, F£(l - t) = F^(t), so
£ t |F2sr(<)| dt = j\l - t) |F^r (t)| dt = |(1 - 32-2r)|B2r|.
Sr := (-1)’--1 fit) dt - 1 f (0) + 3f (1) + 3/ + f (1)
- E Unit1 - [f^m - 1.
f' o(2/c)l L J
Theorem 7.1 Suppose /: [0,1] —> R is (2r + 2)-convex. Then
J^l,1_o2-2n,(2r) MA , o , JJM<1_,2-2n/(2r’(l) +/(2r’(0)

8(2r)ll 11 - r - 8(2r)! 1 ‘ 2
If f is (2r + 2)-concave, the inequality is reversed.
Theorem 7.2 Suppose f: [0,1] —> R is (2r)-times differentiable,
<71 1/9
IQ I < l^2rl fl — o2—2r
8(2r)!(1 3
2
b) If |/(2r) I is concave, then
|Sr| < 2^(1 - 32~2’')
To obtain appropriate results for integration via the Euler-Simpson 3/8 for
mula, we apply the above results to each interval of the subdivision
[0, h], [h, 2/i],... , [(n — l)h, nh]
First we introduce
S(/;h)
= /((fc-l^+3/ (kh--h +3/ ikh--h + f (kh)

8^L V 3 / V 3
and
r-1 7,2k r i'

- S - 32-2l)fl2t [/‘“-‘’("/.J - /<2fc-‘>(0)] • .
Theorem 7.3 Iff: [0, nh] —> R is (2r + 2)-convex, then
^2r8^j!(1"32’2r)M (/(2,M S* ^^t^1-32’2^ (/(2r)^) •

(1 - 32-2r)
< n/l2r+1^jy(1 - 32-2r)max {|/(2r)(0)|, |/(2r)(n/l)|} .
b) is concave, then
*
I
IQ |-®2r| /-. o2—2r
8(2r)r ~ 3
The resultant formulae in Theorems 7.1-7.4 when r = 2 and the sums in S,

and S
Corollary 7.1 Iff: [0,1] —> IR. is 6-convex, then
1
6480J \2j
sH/(o)+3/G)+3/©+/(i
< 1 /(4)(0) + /(4)(l)
- 6480 2
o
i ri/(4|(°)r+i/|4)(Drr/J
~ 6480 2
b) is concave, then
/'/(«) * -i [/(») + 3/ (1)+ 3/ (|)+/d)]
< 1 f (4) (
- 6480 J \2J
(/«>; k) < SU- k} - f dt < JA-T (/«>; k) .
Iff is 6-concave, the inequalities are reversed.

a) If\f^\q is convex for some q > 1, then
f(t)dt — S(f',h)
h5 n ’ |/W(m<+|/W((m-1)<] 1/9
- 6480 2
nh5 max {|/(4)(0)|,|/(4)(n^|}.
~ 6480
/w dt - S(/; h)
8. The Euler—Maclaurin formula

In this section we explore a path that is associated with the Euler-Maclaurin
formula.
there, then the Euler-Maclaurin formula (see [7]) states that
r1 i r /i\ /i\ /t>\’

/0 f^dt= 8 (e) +yG0 +3/G)J
- Z 8(S)!(1 “ - 32_2,,)B» [/|2t-1)(1) - (0)]

where
(i \ /1 \ /k \
j - t) + 2B2‘r (2 - + 3B5r \6 “ 9
-3B2r(|)-2B2rQ)-3B2r(|).
= —(2 — 21-2r)(l — 32-2r)B2r.
Since (—l)r+1B2r > 0, we have (—l)rF^.(t) > 0.

f1 \F^r(t)\ dt = (1 - ^(l - 32-2r)|B2r|.
JO
Also, F2^(l - i) = so
t |j£(i)| dt = |/^(i)| dt = |(l-21-2r)(l-32-2fc)|B2r|.
+£ - 32~2‘)s» [/“’“(i) - (0)1 >.

■H 8(2fc)! L J
1^1(1 — 21-2r)(l — 32-2r)/(2r^ Q)
<r< 1^1/1 ol-2rVl .2-2n/(2r)(l)+/(2r)(0)

-r~ 8(27)!(1"2 )(1"3 > 2------------ •
Theorem 8.2 Suppose f: [0,1] —» R is (2r)-times differentiable.

a) 7f|/<2^ T is convex for some q > 1, then
iLri - ssi(i ■ 2i~2r,(i ■ 32~2r) |/(2r) G)

To obtain appropriate results for integration via the Euler-Maclaurin formula,
[0, h], [fo, 2h],..., [(n — l)/i, nh].
First we introduce
fc=i
and
nh
Lt:
r-1
gw(1-21-2‘,(1-32‘“
Theorem 8.3 Iff: [0, nh] —> R is (2r + 2)-convex, then
- 21-2r)(l - 32-2r)M (7(2r); h)
< i; < (1 - 21-2r)(l - 32-2r
|i;i < /i2r+1^yr(1 - 21-2r)(i - 32-2r)

n
iXf’
2
m=l L
Fl2r+1^ji(1 - 21-2r)(l - 32-2r
Hadamard and Dmgomir-Agarwal inequalities 135
x max ||/(2r)(0)|, |/(2r)(n/i)11.
b) is concave, then
|Z!| < Zl2r^rr(1 - 21-2r)(l - 32~2r)M (|/(2r)|; h) .

8(2r)! \l I /
The resultant formulae in Theorems 8.1-8.4 when r = 2 and the sums in Lr

and L
^4o/W G)s Z1 /w dt -
< 7 /W(o) + /W(i)
G)+2/ G)+3/ G)
S 51840 2

a) If\f^\g is convex for some q > 1, then
3/G)+2/G)+3/ G)
yo J \2 J \o j
7 'i/wmi’+i/wtijni/s
~ 51840 2
b) is concave, then
yi 1 r /i \ /i \ /k\ I
/„ ^d(-s[3/Gr2/G)+3/G)J
- 51840 J \2J
(/(^ ft) < £ dt - L(/; h) < (/|4>; h) .

Corollary 8.4 Suppose f: [0, nh] —> R is 4-times differentiable.

dt — L(f;h)
7h5
~ 51840
< —I— max (1 (0) I, I (nh) I} .

“ 51840 lr M’r v J\)
b) Z/’|/(4)| is concave, then
References
[1] I.S. Berezin and N.P. Zhidkov: ‘Computing methods, Vol. V, Pergamon
Press, Oxford-London-Edinburgh-New York-Paris-Frankfurt, 1965.
[2] Lj. Dedic, C.E.M. Pearce and J. Pecaric: ‘Hadamard and Dragomir-
Agarwal inequalities, higher-order convexity and the Euler formula’, J.
Korean Math. Soc. 38 (2001), 1235-1243.
[3] Lj. Dedic, C.E.M. Pearce and J. Pecaric: ‘The Euler formulae and convex
functions’, Math. Inequal. Appl. 3 No. 2 (2000), 211-221.
[4] Lj. Dedic, M. Matic and J. PeCaric: ‘On Euler-Simpson formulae’,
PanAmer. Math. J. 11 No. 2 (2001), 47-64.
[5] Lj. Dedic, M. Matic and J. Pecaric: ‘On dual Euler-Simpson formulae’,
Bull. Belg. Math. Soc. 8 (2001), 479-504.
[6] Lj. Dedic, M. Matic, J. Pecaric and A. Vukelic: ‘On Euler-Simpson 3/8
formulae’ (submited).
[7] Lj. Dedic, M. Matic and J. Pecaric: ‘On Euler-Maclaurin formulae’,
Math. Inequal. Appl. (to appear).
[8] S.S. Dragomir and R.P. Agarwal: ‘Two inequalities for differentiable
mappings and applications to special means of real numbers and to trape-
soidal formula’, Appl. Mat. Lett. 11 No. 5 (1998), 91-95.
[9] V.I. Krylov: ‘Approximate calculation of integrals', The MacMillan
Comp., New York, 1962.
[10] M. Matic, C.E.M. Pearce and J. PeCaric: ‘Two-point formulae of Euler
type’, ANZIAMJ. (to appear).
[11] C.E.M. Pearce and J. Pecaric: ‘Inequalities for differentiable mappings

and applications to special means and quadrature formulas’, Appl. Math.
Lett. 13 (2000), 51-55.
Chapter 9
ON ULAM STABILITY
IN THE GEOMETRY OF PDE’S
Agostino Prastaro
Department of Mathematical Methods and Models for Applied Sciences
University of Rome lLa Sapienza’
Via A.Scarpa, 16-00161 Roma, Italy
Prastaro@dmmm.uniromal.it
Department of Mathematics, National Technical University of Athens
Zografou Campus, 15780 Athens, Greece
trassias@math.ntua.gr
Abstract The article is concerned with the problem of the unstability of flows correspond
ing to solutions of the Navier-Stokes equation in relation with the stability of
a new functional equation (functional Navier-Stokes equation), that is stable as
well as superstable in an extended Ulam sense. In such a framework a natural
characterization of stable global laminar flows is given also.
2000 MSC: 39B72, 47H19, 58G20, 58G37, 58H99 (Primary).
Keywords: stability, functional equations, integral (co)bordism groups, Navier-Stokes equa

tion
1. Geometric analysis in the Navier-Stokes

equation
In the geometric theory of PDEs [2], [5] we can describe the non-isothermal
Navier-Stokes equation (NS), for incompressible fluids, on the Galilean space
time M, as a 74-dimensional submanifold of the second order jet-derivative
space JV2(W) [2], where W = xmT$M xmT$M, with =
the first order jet-derivative space for motions, i.e., the first derivative of sec-
139
tions m: T —> M of the Galilean affine fiber bundle t: M —>T, where T rep
resents the time axis. TqM = M x R. A section s: M —> W of tt: W —> M
is a triplet s = (u, p, 0), where v = velocity vector field, p = isotropic pressure
field, 0 = temperature field. More precisely, let us assume the following ther
modynamic constitutive equations: e = internal energy = e(0), x = viscosity
— constant, v = thermal conductivity = constant. Furthermore, for the sake of
simplicity we shall assume that Cp = (dO.e) = constant and the body volume
force is conservative (with respect to an inertial frame ip): B = — grad(/)).
Then, in coordinates, adapted to the frame ip: M: (xa), 0 < a < 3; W:
(xa,xl,p,0), 1 < i < 3; J"D(Wy. (xa,x\p,Q,xlp,pp,()p), 0 < < 3, we
can write (NS) in the following form:
{j10 = xkGj k + = o}
(pj _ ( xsB?s + isilpG^ + xsiip + pxi
I "I +^S3k + iisTkS_+ Pi9ij + p(dxi.f)g^ =0
fp4= f eopCp + xk0keisEispCp n
1 -1 +xkipwkp + + xkxpYkp = o JJ
with-Ri = -2x[Gtik(dxs.gk:,)+GJki(dxs.gkz)+(dxl.dxs.gt:,)],SJks = -2
[-G^
*
Gl^ - 2G3ikgis + 3jk(dxp.gP°)\, T3kis = 2X(/^ + Eis = -vg”,
Wkp = -2xGikgjs(dxp.gsiy W£s = 2X[G^s5* + Gpsk + (dxk.gP
*)g ae],
Ykp = 2x[gksgip + 3kdls], 1 < ?, j, k,p,s < 3. Here G’p are the spatial compo-
nents of the connection coefficients of the canonical connection on the Galilean
space-time and are the components of the vertical metric on M [2]. Then
(NS) is an algebraic submanifold of JD2(W). One has the following
Theorem 1.1 ([3]~[5])
1) The equation (NS) is a not formally integrable and not completely in
tegrable PDE of second order on the fiber bundle 7r: W —» M. In
side (NS) one can distinguish an important sub-PDE (NS) C (NS) C
JD2(W) C J^(W):X (NS) is an involutive formally integrable PDE,
but with zero characteristic distribution. The expression of (NS) is the
1J4 (W) is the second order jet space for 4-dimensional submanifolds of W [2], [3]. JV2(W) is an open
submanifold of (W).
following:
M) If0 = itG,jk + ii,6- = 0}

<S) {f2 = +±
c
* ’jt + = 0}
I . . . J 0<ck<3
(NS) < rn Jpj1 = S ±SRi + ^sxipGjs + xsx:,sp +pxi + i^Sl3
1 [ +xksTks + + p(dxi.f)gz:> = 0
1<7<3
f p4 = (OopCp + xkekeisEispCp + ±kxpwkp 11
( } I “I +ikispWPks + ikispY£ - 0 //
where the symbols of the coefficients are as above. The Cartan distribu
tion Ej<2(NS) of (NS), that is the Cauchy characteristic distribution asso
ciated to the contact ideal C2 G Qe(J| (PK)) characterizing (NS), is a
distribution of dimension 46. The Cartan distribution HtyNS) on (NS) is
of dimension 50.
2) (Existence of local regular solutions for (NS))

For any point q G (NS) C (NS) one can construct a (smooth) regular
solution of (NS) C J%(W).
3) (Cauchy problem for (NS) C J%(W))
Any (regular) integral submanifold N G (NS) G (NS) G J±(W) of

dimension 3 that satisfies the initial conditions and the transversality
conditions with respect to a time-like vector field ( 6 $(£2), where C2 is
the ideal encoding (NS) given in Theorem 1.1, and 5 denotes 'infinitesi
mal symmetry', generates a (regular, smooth) solution of (NS) if N has
no frozen singularities [3].
4) (Integral bordism groups in (NS))

The following summarizes the calculated integral (singular) bordism
groups. of (NS): 0 1 < q < 3;
Z2 = The infinity prolongation (NS)+oo of
(NS) has the same integral (singular) bordism groups with (NS).
5) (Tunnel effects in the Navier-Stokes equation)

In the set of p-dimensional, 2 < p < 4, integral manifolds of (NS) one
has manifolds that change their sectional topology. In particular in the
set &ol(NS), of all solutions of the Navier-Stokes equation, there are
also solutions that present (multi)tunnel effects. These solutions are not,
in general, representable as image of second derivative of sections of the
fiber bundle tt-.W = JV(M) x M TqM xM TqM -> M.
6) (Existence of global (smooth) solutions of (NS))

We define a global (smooth) solution of (NS) as any (smooth) solution
of (NS) that is defined in any space-like region inside the boundary and
for any t > to, where to is an initial time. For a generic (smooth)
boundary value problem contained into (NS) C (NS) there exist global
(smooth) solutions of (NS). Furthermore, the boundary conditions do
not, in general, guarantee the uniqueness of solutions.
Remark 1.1 It is important to emphasize that the existence of global smooth

solutions for any ‘initial condition' in the sub-equation (NS) of the Navier-
Stokes equation as stated in the above theorem, does not imply that the charac
teristic flows of such solutions are free of singularities and such solutions are
stable.
2. Ulam stability in functional equations

Definition 2.1 Let Gi be a group and G2 be a metric group with a metric
The functional equation h(xy) = h(x)h(y) is Ulcun-stable if given
e > 0, there exists a 5 > 0 such that if a function h: Gi —> G^ satisfies
the inequality d(h(xy), h(x)h(yf) < 6, for all x,y G Gi, then there exists a
homomorphism H: Gi —> G2 with d(h(x), H(x)) < e, for all x G Gi- (See
[7] for an extensive contribution to Ulam problem.)
Theorem 2.1 (D.H. Hyers, [1]) If a function f: Ei —> E? defined between

Banach spaces satisfies the inequality \\f(x + y) — f(x) — f (y) || < 5, for
some 6 > 0 and for all x,y € Ei, then the limit A(x) = limn-,00 ^f(2nx)
exists for each x G Ei and A: Ei —> E^ is the unique additive function such
that ||/(x) — A(x) || < 5, for all x G E\. Moreover, if f(tx) is continuous in t
for each fixed x € Ei, then the function A is linear.
Theorem 2.2 (Th.M. Rassias, [6]) If a function f: E\ —> E? between Banach

spaces satisfies the functional inequality ||/(or-|-?/) — f(x) — /(y)|| < 0(||rr||p +
\\y\\p), for some 0 > 0, somepwithO <p< landfor all x,y G Ei, thenthere
exists a unique additive function A: E\ —> E% such that ||/(rr) — A(a;) || <
2z^||ar||p, for all x G Ei. If, in addition, f(tx) is continuous in t for each
fixed x G Ei, then the function A is linear.
Definition 2.2 1) Let E\ and E2 be some appropriate spaces with algebraic

structures. For some p, q G N and for any i G {1,... ,p}, let gi: Ef
Ei and G: E? x Ef —> E^ be functions. Let us consider the following
functional equation:
G(H (gi (x1, ..., x^)),..., H(gp(x\ ..., x’)), x1,..., x’) = 0 (2.1)
for mappings H: E\ —» E2. Then (2.1) is Hyers-Ulam-Rassias stable

if for every function f: E± —> E2 satisfying the inequality
||G(/(51(x1,..., xq)),f(gp(x\ xq»,x\..., x’)||

< <p(x\ ... ,xq) (2.2)
for all x1,..., xq e E\, with ip-. Eq —> [0, oo) fixed, there exists a
solution H of (2.1) such that ||/(rr) — 7J(x)|| < 3>(x), Vx G Ei, for a
suitable 4>: Ei —> [0, oo).
2) If each solution f: E± —>■ E% of the inequality (2.2) is either a solution

of (2.1) or satisfies some stronger regular conditions, then we say that
equation (2.1) is superstable (on (E\, E2)).
Remark 2.1 The Ulam stability is a particular case of this definition. For a
generalized solution of Ulam problem see [5], [6], [8], [9].
Remark 2.2 In order to apply the concept of HUR-stability to the Navier-

Stokes equation, we need to generalize such a concept in the following way.
Definition 2.3 1) Let F be a functional space, i.e., a space of suitable ap

plications f: X —> Y between finite-dimensional Riemannian mani
folds X and Y. Let E be a Banach space and S a subset of Xn. Let us
consider a functional equation:
G(/> 91, • • • , qn) = 0, Vfe1, • • • , qn) G S C Xn, (2.3)
defined by means of a mapping
G: F x Xn^ E, (/, (Q1, • • • , O ~ G(J, (q1, • • • , <f)).
We say that such a functional equation is Ulam-extended stable if for

any function f € F satisfying the inequality
||G(/,(g1,---,gn))||<^1,---,9n), (2.4)
with <p: Xn —> [0,00) fixed, there exists a solution f of (2.3) such that
VqeX,
for suitable $: X —> [0,00).2
2Here dy(.,.) is the metric induced by the Riemannian structure of Y. More precisely, dy(a,b) =
in/ f[o 1] x/b eY, where 7 € C'1([0,1], Y), with 7(0) = a, 7(1) = b and gy
the Riemannian metric on Y. One has the following important propositions:
2) If each solution f G F of the inequality (2.4) is either a solution of the

functional equation (2.3) or satisfies some stronger conditions, then we
say that equation (2.3) is Ulam-extended superstable.
Remark 2.3 Definition 2.2 is a particular case of Definition 2.3 taking
X = #i
Y = E2
S = Ef,
,--- ,xg) = e(li)l9)(G o ((/T o (51 x ••• x gp)),idEg))
G(H,x1(ii)
(iii)
where e(xi ... is the evaluation function at the point (x\ • • • , xq) e Ef.
Furthermore, the metric in Ei is canonically induced by the norm of Ei.
3. Ulam stability in the Navier—Stokes equation

This section is concerned with the problem of characterization of the (un)stability
of flows corresponding to solutions of the Navier-Stokes equation by means of
a new functional equation. The following theorem is the main result of the
article.
Theorem 3.1 Let us consider solutions V C (NS) of the Navier-Stokes equa

tion, for boundary value problems for fluids contained in 3-dimensional do
mains that have the topology of the 3-dimensional disk D3, and with smooth
boundaries that smoothly evolve without change their topology. The flows
(characteristic flows), associated to such solutions, are mappings <t>x-Vt
Vt+\, where Vt C V is the fiber of V over t G T, with respect to the fiber
bundle structure (NS) —» M —» T over the time axis T, that satisfy a func
tional equation (functional Navier-Stokes equation), that is Ulam-extended
(super)stable.
Proof: Let us consider the following lemmas.
(i) If 7 € C1 ([0,1], Y), such that 7(0) = a, 7(1) = b, and y/gy^^nit^dt = dy(a,6),
then 7([0,1]) C Y is a geodesic. If in addition, 7 has constant speed, then 7 is a C°° geodesic.
(ii) (de Rham) When Y is geodesically complete, then given a, b e Y, there exists at least one geodesic
7 in Y connecting a to b with \/py(7(f), 7W)dt = dy (a, b).
(iii) (Hopf, Rinov) The geodesic completeness of Y is equivalent to the completeness of Y as a metric
space, which is equivalent to the statement that a subset of Y is compact iff it is closed and bounded.
(iv) Y is complete whenever it is compact.
Lemma 3.1 The linear Navier-Stokes equation (NS)s C JZ>2(E), at the so
lution s, has the following form:
____ [ (I) ^sS? + ukGj k = O

(NS)s (2) S[s] + ^p + vkB3k[s] + vsAl[s] + =0
[ (3) r^Bf [s] + pMfcH + 4e)5is[S] + vjfpCp + =0
with A£[s] = R3S + 2pG3isxt + x3sp, B£s[s] = pxs63k + S3k, Afc[s] = xp(Wkp +
Wpk)+ek0isEis_PCp, Bf[s] = ikWPs+x^Y^+Y^, Cfc[s] = xkdisEispCp,
D’s[s] = ik6kElspCp. (For any solution s of (NS), identified with a section
s: M —> W, the above coefficients are fixed numerical functions on M.) Here
E = s*vTW =. M x S x R2 and S denotes the 3-dimensional Euclidean space
of space-like vectors of the Galilean space-time [2].3
Lemma 3.2 For the equation (NS), the equation (NS)s is not formally inte
grable and is not completely integrable. However, inside (NS)s one can identify
____ ____
a linear equation (NS)s C (NS)s, that satisfies such integrability conditions.
More precisely, (NS)s is the linearization of (NS) at the solution s.
Let us emphasize, here, that the fiber JP2(Wz)t of JD2(W) = M x I x

IR2 x M0 (S x IR2) x M©M® (S x IR2) over t 6 T is identified with the 78-
dimensional affine Euclidean space E = Mt xIxR2xM®(SxR2)xM©M®
(SxIR2), having E = SxSx1R2xM0(SxIR2) xM©M0(SxlR2)) as space
of free vectors. Furthermore Vt = V A JD2(TV)t is a 3-dimensional (compact)
space-like Riemannian manifold, as well as Vt+A = V A JP2(Hz)t+A- In the
following we shall assume that Vt and Vt+A are compact manifolds. If V is a
regular solution of (NS) C T>2(W), then its characteristic flow fix : Vx —> Vt+x
diffeomorphically projects on a flow : Bt —> Bt+x in M, where Bt =
7t2(Vt) C M. If V is a more general solution, some singularities can arise in
the projected flow <px- Then, by using the general theorem on the tunnel effects
in (NS), (Theorem 1.1, (5)), one can distinguish two cases:
(i) V is a laminar regular solution, hence its characteristic flow has not sin
gular points. In such a case Bt and B(+> are diffeomorphic (by means
of the flow <px)\
(ii) V is a perturbed solution of a global laminar one and the characteristic

flow has some isolated singular points, qi,..., qn € Vt+x- Then Bt+x is
not diffeomorphic to Bt (by means of the flow <px)- In other words, one
has some cell contractions.
3The vertical tangent space vTW is calculated with respect to the fiber structure tt: W —► M.
In case (i) the corresponding equivalent condition on the flow is ker(D<p\(p)) =

0, Vp G Bt. In case (ii) one can, instead, write ker(B<pA(p)) 7^ 0, for some
p € Bf. Therefore the solutions of type (i) generate flows that satisfy the
functional equation (functional Navier-Stokes equation):4
J dim(ker(B<pA(p)))=0, Vp G Bt, VAg[0,A] 1

( )f | Fl(y)=0, V = UAe[o,A]<MM Bt = 7r2(Vt) p
On the other hand taking into account the Euclidean structure of R, one can
also claim that the solutions of type (ii) satisfy the following conditions:
J ||dim(ker(OpA(p)))|| < Va(p), Vp e Bt, VAj [0, A] 1

I F'm =0, Vs UAeM tM B, = p2(H) j lJ-1’
for any fixed function [0,oo). Now, (NS)p is Ulam-extended sta

ble, when for any singular solution </>A, i.e., solution of second type (and hence
satisfying equation (3.1), with ipx fixed, there exists a solution cb\ of type (i),
i.e., laminar solution, thus satisfying (NS)/? with d(<pA(p), <pA(p)) < <hA(p),
Vp € Bt, for suitable 4>A: B4 —> [0, oo). On the other hand we know that,
under the hypotheses of Theorem 3.1, such solutions </>A exist. In fact, for
such boundary value problems we can have solutions V with characteristic
vectorjields without zeros. (For more details see [5].) Then, as the solu
tions <j)\ are, by hypothesis, perturbed solutions of global laminar ones, one
can conclude that (NS)/? is Ulam-extended stable. Furthermore, (NS)/? is also
Ulam-extended superstable because a flow <f> satisfying equation (3.1) either is
a laminar solution of (NS) (i.e. is of type (i)) or is of type (ii) and therefore
it satisfies some stronger conditions, i.e., dim(ker(BpA(p'))) 7^ 0> at some
points p' G Bt.5 □
Corollary 3.1 The stability of a global laminar solution of (NS) can be char
acterized by the following conditions:
< lim <&a(p) = 0, lim V'a(p) = 0 > , Vp G Bt.

>oo A—>oo J
References
Acad. Sci. USA 27 (1941), 222-224.
4FJ are the functions defining (NS) (see Theorem 1.1).

5With reference to the notation of section 2, Definition 2.3, here F = : Bt —> Bt+x} = set of
flows of solutions hording Bt with Bt+x, X = Bt, Y = Bt+x, E = R, n = 1, G(y>x,p) —
dim(ker(r)</?A(p))).
[2] A. Prastaro: ‘‘Geometry of PDEs and Mechanics’, World Scientific Publ.

Co., Singapore, 1996.
[3] A. Prdstaro: ‘Quantum and integral (co)bordism groups in partial differen
tial equations’, ActaAppl. Math. 51 (1998), 243-302.
‘Quantum and integral bordism groups in the Navier-Stokes equation’,
pages 343-359 In: New Developments in Differential Geometry, Budapest
1996, J. Szenthe (ed.), Kluwer Academic Publishers, Dordrecht (1999).
‘(Co)bordism groups in PDEs’, Acta Appl. Math. 59 (2) (1999), 111-201.
[4] A. Prastaro, ‘Local and global solutions of the Navier-Stokes equation’,
Steps in Differential Geometry, L. Kozman, P.T. Nagy and L. Tomassy
(eds.), University of Debrecen, Hungary, 2001, 263-271.
[5] A. Prastaro: ‘Navier-Stokes equation. Global existence and uniqueness’
(to appear)
Ulam’, ActaAppl. Math. 62 (2000), 23-130.
[8] Th.M. Rassias: ‘On a modified Hyers-Ulam sequence’, J. Math. Anal.
Appl. 158(1991), 106-113.
[9] Th.M. Rassias: ‘On a problem of S.M. Ulam and the asymptotic stability
of the Cauchy functional equation with applications’, Intern. Ser. Numer.
Math. 123 (1997), 297-309.
Chapter 10
ON CERTAIN FUNCTIONAL EQUATIONS

AND MEAN VALUE THEOREMS
Department of Mathematics, National Technical University of Athens,
Zografou Campus, 15780 Athens, Greece
trassias@math.ntua.gr
Young-Ho Kim
Department of Applied Mathematics, Changwon National University,
Changwon 641-773, Korea
yhkim@sarim.changwon.ac.kr
Abstract In this paper we prove certain new characterizations of mean values in the spirit
of Gauss type functional equations.
2000 MSC: 26E60, 34A34, 39B22 (Primary).
Keywords: Gauss type functional equation, monotonic function, power mean, Lehmer mean
1. Introduction
In what follows we shall refer to a famous example of mean. Given two
positive numbers a and b, let us define successively the terms
= ^4(®n>bn), ^n+1 ~ G(^dn,bn^, n > 0,
where a$ = a and bo = b. It follows that
&0 — > **’ > > * * * > bn > * * * > bi > bo,
for a > b. It is known (cf. [2]) that the sequences {an} and {bn } are convergent
to a common limit, which is denoted by A • G(a, b). This function A ■ G(a, b)
was first investigated by Gauss. Thus, it is called the arithmetic-geometric
149
mean of Gauss. The following representation formula is also known (cf. [1]):
A-G(a,b) = [/(a, ft)]"1,
where
rz M 1 f2" ™
I(a,b) = — / • . .. — =.
2^ jo \/a2 cos2 0 + b2 sin2 0
The proof of this formula is based on the fact that the function f verifies the
relation
f(A(a,b),G(a,by)=f(a,b)
which can be called Gauss’ junctional equation. These results where general
ized as follows. Let us denote
r„,fc(0) = (afcn cos2 0 + bkn sin2 0)1/n, (fc, M 0)
and
„ (a\ 2 ffiksin2 &
ro,k(0) = 1,'™ {U) - „/ccos
Ilin „rn kZZ)\ a b (fc/0).
If p: R+ —> R is a strictly monotonous function and k, n are real numbers,

then
1 f27r
Mg(a,b;p,rntk) = (I9(a,6))1-fc — / p(rn>k(py) d0 (a,b>0)(l.l)
Jo
defines a symmetric mean value, that is M(a, 6) = M(6, a), where our termi
nology here is taken from [6, p. 232] wherein
r . a’+i + &+1
Lg(a, b) =---------- ;-----
denote the Lehmer means of a, b > 0. A well-known example of mean (1.1)

is given for k = 1, n = 2 by p(x) = x~l. It corresponds to the arithmetic
geometric mean of Gauss. The essential step was done in [13] by the consid
eration of the definition (1.1) for k = l,n = 2 with an arbitrary p(x). The
values n = 1 and n = — 1 were studied in [19] with k — 2, q = — 1 and an
arbitrary p(x). The cases of k = 2 and q = — 1 were studied in [27] with an
arbitrary p(x). The general case of arbitrary q and k(f 0) was studied in [21],
In this paper we shall generalize the results of the integral mean (1.1).
We use in what follows the power means defined by
/ „n i kn \ !/n
for n 0. while for n = 0 it is the geometric mean

P0(M) = G(a,b) = Vab.
2. The functional equations

Let us consider the strictly monotonous function p and the function rn,k($)-
Using them we define the following definite integral
1 f2ir
f(a,b;p,rn<k) = — p(rntk(0)) dd (a,b>0). (2.1)
2?r Jo
It is well known (cf. [7]) that the proof of the fact that Mq(a, b) is the
arithmetic-geometric mean is based on the relation
f (A(a, b), G(a, b);p(z) = 1/x, r2,i) = f (a, b;p(x) = 1/x, r2,i).
In [21], a general functional equation of this type is considered, namely:
F(L,(a,b),Ls(a,b)) = F(a,b). (2.2)
As it is proved in [21, Theorem 2.1], if the function f given by (2.1) is a

solution of the equation (2.2) and p has a continuous second order derivative,
then
p(x) = QZ2(g+s+l)/fc-n +
with a, (3 arbitrary real constants. In [26], Gauss’ functional equation was also
replaced by more general equation
/(M(a,b),lV(a,b);p,rg(6>)) = /(a,b;p,rg(0))
where rq(0') = g-1[g(a) - sin2 f) + q(b) - sin2 0], M and N are two given means.
Now, we consider a more general functional equation than (2.2), namely
F(M(a, b), 7V(a, b)) = F(a, b), (2.3)
where M and N are two given means.

From the paper [21] we can deduce the following result:
Lemma 2.1 Letp\ R+ —> R be twice continuously differentiable in R+. Then

the Junction f defined by (2.1) has the following partial derivatives:
i) fa{c,c\p,rn>k) = fb(c,c,p,rn,k) = {l/2')k^~1p\c/t),
H) fab(.c,c-,p,rntk') = [k2c2k~2p"((ff) + (1 - n)/c2c/c“2p'(cfc)]/8,
iii) faa(c, c,p, rn,k) — [3fc2c2fc-2p"(cfc) + (k2n + 3fc2 - 4fc)cfc-2p/(cfc)]/8,
where c are arbitrary positive real numbers.
Using Lemma 2.1, we shall prove the following theorems.

Theorem 2.1 If the function f defined by (2.1) verifies the functional equation
(2.3), then the function p is a solution of the differential equation
k2ckp"(ck} (Ma(c, c)A(M, N) + Na(c, N) - 1)
+(k2n + 3k2 - 4k)p'(ck>) (Ma(c, c)Mb(c, c) + Na(c, c)Nb(c, c))
+(1 - n)k2p'(ck) (Ma(c, c)Nb(c, c) + Mb(c, c)Na(c, c) - 1)
+4fcq/(cfc) (Mo6(c, c) + Nab(c, c)) = 0, (2.4)
where A(M, N) = 3Mb(c, c) + Nb(c, c), B(M, N) = 3Nb(c, c) + Mb(c, c).
PROOF: Taking in the functional equation
F(M(a,b),AT(a,b)) = F(a,b),
the partial derivatives with respect to a we obtain
Fa(M(a,b),N(a,b)) ■ Ma(a,b)
+ Fb(M(a,bfN(a,b^ • Na(a,bfia,b) = Fa(a,b).
Taking again the derivatives with respect to b it follows that
(Foa(M(a,b),jV(a,b)).Mt(a,b)
+Fab(M(a, b), N(a,b)) ■ Nb(a, b)) • Na(a,b)
+ (Fab(M(a,b),N(a,b)) • Mb(a,b)
+Fbb(M(a, b), N(a, b)) ■ Nb(a, b)) ■ Na(a, b)
+Fa(M(a, b), N(a, b)) • Mab(a, b)
+F6(M(a,b),7V(a,b))-7Va6(a,b)
= Fab(a,b).
Setting a = b = c, yields
(Foo(c, c) • Mb(c, c) + Fab(c, c) • Nb(c, c)) • Na(c, c)

+ (Fat(c, c) • Mb(c, c) + Fbb(c, c) • Nb(c, c)) • 7Va(c, c)
+Fa(c, c) • Mah(c, c) + F{,(c, c) ■ Nab(c, c)
Fi/>(c, c).
Since the function F = f can be defined by (2.3) we apply Lemma 2.1, i)—iii),
and obtain the equation (2.4). □
Corollary 2.1 If the function f defined by (2.1) verifies the functional equa
tion (2.3), where the means M and N are symmetric, then the function p is a
solution of the differential equation
k2ckp'\ck)+
+ (nA:2 + k2 - 2k + 4kc (Mab(c, c) + 7Vab(c, c))) p'(ck) = 0, (2.5)
where c is an arbitrary real number.
PROOF: As the means are symmetric, their partial derivatives of first order are
equal to 1/2 (see. [9]), thus (2.4) becomes (2.5). □
Corollary 2.2 If thefunction f defined by (2.1) verifies the functional equation
f (Pq(a,bfPs(a,by,p,rntk) = f (a,b;p,rn<k),
where the power mean Pq(a, b) is symmetric, then the function p is given by
p(X) = + /3,
where a and (3 are arbitrary constants.
PROOF: We have in (2.5) M — Pq(a, b) and N = Ps(a, b). Thus
Mab(a,b) - dadbpq^,b) - 2 2(2)
and
b) —
d as 1 (1 — s)bs 1 Zas + bs\ 1/5-2
Ps(a, b)
dadb ~2 2 \ 2 J
Setting a = b = c yields
Mab(c, c) = Nab(c, c) =
Replacing them in (2.5) we obtain the differential equation
?/(?) + (n + 1 - p'(cfc) = 0.
Since c is an arbitrary fixed positive real number, we can replace cfc by a posi
tive real variable X in the above equality. Hence, we have
Xp"(X) + (n + 1 - p'(X) = 0.
\ /
with the solution
p(X) = ax(s+’)/fc-n + (3.
□
Remark 2.1 As has been shown in [25], the condition is also sufficient for
q = —s = n, k = 1. That is, if p(X) — aX~n + (3, then the function f given
by (2.3) verifies the functional equation
f {P_n(a,b\Pn{a,b)-p{X) = X~n,rn^ = f (a,b-p(X) = X~n,rn^ .
Since we can prove the following Corollaries by using a similar argument

to that in Corollary 2.2, we omit the proofs.
Corollary 2.3 If the function f defined by (2.1) verifies the functional equation
f (Pq(a,b),Ls(a,by,p,rn<k) = f (a, b;p,rn>k),
then the function p is given by
p(X) = aX^+2s+1^k-n + (3
where a and /? are arbitrary constants.
Corollary 2.4 If the function f defined by (2.1) verifies the functional equation
f (Lq(a, bf Ls(a, bfip, rn>k) = f (a, b;p, rn<k),
then the function p is given by
p(X) = ax2^+s+^/k~n + /3,
Remark 2.2 This Corollary was obtained in [21, Theorem 2.1].
3. The integral means

Recently, several authors have established a theory on the symmetric inte
gral mean (cf. [12]—[14], [17]—[20], [24]—[27]). Subsequently, Kim and Ras-
sias [21] established a theory on symmetric integral means as follows:
Mq(a,b',p,rnk) = (Lg(a,6))1-fc — J p(rn<k(0y)dO (a,6>0),
where
rn,fc(0) = (akn cos2 6 + bkn sin2 (k, n / 0),
and
ro,fc(0) = lim r„,fc(0) = akcos2ebksia2e (k ± 0),
n—>0
which provide an extension of the mean values in [ 12]—[ 13] and [18]. A prob
lem which is studied for the integral means defined in [12]—[13], [18] and [24]
is that of the determination of the cases in which the mean reduces at a given
one, usually a power mean. Similar results can be given also in more general
circumstances. Let us prove the following
Theorem 3.1 Iffor a given mean N(a, b) we have Mq(a, b; p, rnff) — N(a, b),
then the function p is a solution of the differential equation
k — 1 + Na(c, c)^ f + “ fc2

ck 8 p"(cfc)
2
+ [2(fc - 1)C2 + 8cNab(c, c) - (1 - n)A:2] p'(cfc) = 0, (3.1)
where C2 = —2q + k — 2 + 2Na(c, c) + 2Nb(c, c).
PROOF: Under the given hypotheses we have
f(a,b-,p,rnff) = p([£g(a,b)](fc-1)lV(a,b)) .
Taking partial derivatives in respect to a, we have
/a(a,fe;p, rn>fc)
= p'tco - !)(£,(«, X
x7V(a,b) + (Lg(a,b^k-V Na(a,b)} .
where Ci = [Lq(a, ^N(a, b). If we take the derivative with respect to b

we obtain
fab(a,b;p,rn^ = p'/(C1)—C1-C1+p,(C1)-~C1.
oa ob oaob
fab(c, c,p, rnff)

= c2t-2 (Lzl + NJo, c)) + Nile, c)} p"(c‘)
+ ck~2 fff + e)] pV).

\ £ J
where C2 = — 2q + k — 2 + 2Na(c, c) + 2Nb(c, c). Using equality ii) of Lemma

2.1, we obtain the differential equation as follows
c2fc"2 W^ + 7Va(c,c)) f^-l + 7Vb(C,C)^ -fc2}p"(cfc)
+ ck~2 {2(fc - 1)C2 + 8cNab(c, c) - (1 - n)fc2} p'(?) = 0.
Hence we obtain the equation (3.1). □
Corollary 3.1 If we have Mq(a,b',p,rntk) = N(a, b), with the symmetric

mean N(a, b), then the Junction p is a solution of the differential equation
ckk2p'\ck) + (-4fcg + 4q - 2k + 8dVab(c, c) + (1 + n)A:2) p’(ck) = 0.

(3.2)
PROOF: As the means are symmetric, their partial derivatives of first order are
equal to 1/2 (see [9]), thus (3.1) becomes (3.2). □
Corollary 3.2 If we have Mq{a.b\p.rn^ = Ps(a, 6), then the function p is

given by
p(X) = ax^+2k-^+2s-2yk2-n + 0, (3.3)
PROOF: We have in (3.2) N = Ps(a, b). Thus
0 o / as 1 (1 ~ s)bs 1 (as + bs\1/s~2

Nab(,a, b) = dadbPs^ 2 2
\ 2 )
NabM =
Replacing them in (3.2) we obtain the differential equation
k nr k\ (—^kq + 4q — 2k + 2(1 — s) A // «
cV(c + ------------ 7?------------ -------- - + 1 + n P (c ) = 0-
y fcz /
Since c is an arbitrary fixed positive real number, we can replace ck by a posi

tive real variable X in the above equality. Hence, we have
Xp"(X) + ( 4kq + 4q ±..2SL.J1 + 1 + p’(X) = 0,

\ rb /
with the solution (3.3). □

Corollary 3.3 If we have Mq(a,b;p,rn k) = Lq(a,b), then the function p is

given by
p(X) = ax(4fc9+2fc-49+4s)/fc2-n 4.
Proof: Since we can prove Corollary 3.3 by using a similar argument to that
in Corollary 3.2, we omit the proof. □
Remark 3.1 This last result was obtained in [21, Theorem 3.7]. In [21, The
orem 3.8] it is shown that the condition in Corollary 3.3 is also sufficient for
some case as q = s = —1/2. As shown in [21, Theorems 3.1-3.6], the mean
Mq(a, b;p, rnjf) can represent several known means for some special choices
of k, n, q and the function p. For some further results on means one may also
consult [17], [18], [21], [23].
Acknowledgments
We wish to express our thanks to Professor G. Toader for helpful comments.
References
[1] J. Aczel: ‘Lectures on Functional Equations and their Applications',
Academic Press, New York-London, 1966. (pp. 234—244)
[2] J. Aczel: ‘The notion of mean values’, Norske Vid. Selsk. Forh. Trondheim
19 (1946), 83-86.
[3] G. Almkvist and B. Berndt: ‘Gauss, Landen, Ramanujan, the arithmetic
geometric mean, ellipses, tt and the Ladies Diary’, Amer. Math. Monthly
95(1988), 585-607.
[4] J. Arazy, T. Cleasson, S. Janson and J. Peetre: ‘Means and their itera
tions' .pages 191-217 In: Proceedings of the Nineteenth Nordic Congress
of Mathematicians, Reykjavik, 1984.
[5] E.F. Beckenbach: ‘A class of mean value functions’, Amer. Math.
Monthly 57 (1950), 1-6.
[6] J.M. Borwein and P.B. Borwein: ‘Pi and the AGM', Wiley, New York,
1987.
[7] P.S. Bullen, D.S. Mitrinovic and P. M. Vasic: ‘Mean and Their Inequali
ties' , Reidel, Dordrecht, 1988.
[8] D.A. Cox: ‘The arithmetic-geometric mean of Gauss’, L’Enseign. Math.
30 (1984), 275-330.
[9] D.M.E. Foster and G.M. Phillips: ‘A generalization of the archimedean
double sequence’, J. Math. Anal. Appl. 101 (1984), 575-581.
[10] C.F. Gauss: ‘Werke', Gottingen-Leipzig, 1868-1927.

[11] G.H. Hardy: ‘A Course of Pure Mathematics', Cambridge Univ. Press,
1958.
[12] H. Haruki: ‘New characterizations of the arithmetic-geometric mean of
Gauss and other well-known values’, Publ. Math. Debrecen 38 (1991),
323-332.
[13] H. Haruki and Th.M. Rassias: ‘New characterizations of some mean
values’, J. Math. Anal. Appl. 202 (1996), 333-348.
[14] H. Haruki and Th.M. Rassias: ‘A new analogue of Gauss’ functional
equation’, Int. J. Math. Math. Sci. 18 (1995), 749-756.
tions in Several Variables', Birkhauser, Boston, 1998.
[16] D.H. Hyers, G. Isac and Th.M. Rassias: ‘Topics in Nonlinear Analysis
and Applications', World Scientific Publ., New Jersey-London. 1998
[17] P. Kahlig and J. Matkowski: ‘On the composition of homogeneous quasi
arithmetic means’, J. Math. Anal. Appl. 216 (1997), 69-85.
[18] Y.H. Kim: ‘New characterizations of well-known mean-values’, Far East
J. Math. Sci. 6 (6) (1998), 939-947.
[19] Y.H. Kim: ‘On some further extensions of the characterizations of mean
values by H. Haruki and Th. M. Rassias’, J. Math. Anal. Appl. 235 (1999),
598-607.
[20] Y.H. Kim and J.S. Ume: ‘New characterizations of well-known mean
values II’, Far East J. Math. Sci. 2 (3) (2000), 453^61.
[21] Y.H. Kim and Th.M. Rassias: ‘On some mean value theorems and func
tional equations’Appl. Math. Com. (to appear)
[22] D.H. Lehmer: ‘On the compounding of certain means’, J. Math. Anal.
Appl. 36 (1971), 183-200.
[23] P.K. Sahoo and T. Riedel: ‘Mean Value Theorems and Functional Equa
tions', World Scientific Publ., Singapore-New Jersey-London. 1998
[24] Gh. Toader: ‘Some mean values related to the arithmetic-geometric
mean’, J. Math. Anal. Appl. 218 (1998), 358-368.
[25] Gh. Toader and Th.M. Rassias: ‘New properties of some mean values’, J.
Math. Anal. Appl. 232 (1999), 376-383.
[26] S. Toader, Th.M. Rassias and G. Toader: ‘A Gauss type functional equa
tion’, Int. J. Math. Math. Sci. 25 (9) (2001), 565-569.
[27] J.S. Ume and Y.H. Kim: ‘Some mean values related to the quasi
arithmetic mean’, J. Math. Anal. Appl. 252 (2000), 167-176.
[28] E.T. Whittaker and G.N. Watson: ‘A Course of Modem Analysis', Cam
bridge Univ. Press, Cambridge, 1927.
Chapter 11
SOME GENERAL
APPROXIMATION ERROR
AND CONVERGENCE RATE ESTIMATES
IN STATISTICAL LEARNING THEORY
Saburou Saitoh
Department of Mathematics, Faculty of Engineering
Gunma University, Kiryu 376-8515, Japan
ssaitoh@math.sci.gunma-u.ac.jp
Abstract In statistical learning theory, reproducing kernel Hilbert spaces are used basi
cally as the hypothese space in the approximation of the regression function.
In this paper, in connection with a basic formula by S. Smale and D. X. Zhou
which is fundamental in the approximation error estimates, we shall give a gen
eral formula based on the general theory of reproducing kernels combined with
linear mappings in the framework of Hilbert spaces. We shall give a prototype
example.
2000 MSC: 68T05, 30C40, 44A05, 35A22 (Primary).
Keywords: learning theory, convergence rate, approximation, reproducing kernel, Hilbert

space, linear mapping, inequality
1. Introduction
In statistical learning theory, reproducing kernel Hilbert spaces are used ba
sically. See [1] for an excellent survey article and [3] for very interesting recent
results.
In order to state our motivation and in order to state our results compactly,
we shall start with the fundamental
Proposition 1.1 ([ 1 ] and [3]). Let H be a Hilbert space and A a self-adjoint,

strictly positive compact operator on H. Let s, r e R such that s > r > 0.
159
Then, for R> 0 and for a e H,

r
min ||a-b||<(4) ||A-ra|| . (1.1)
||A-sb||<K \RJ
The operator A~r is defined by using the eigenvalues and eigenvectors of

the operator A and the right hand side of (1.1) is understood as infinity, if A~ra
is not defined. It seems that the definition and properties of A~r are formal and
abstract, in a sense. In this paper, we shall show how to realize the operator
A~r by using the general theory of reproducing kernels in [2] combined with
linear mappings in the framework of Hilbert spaces. We shall give a prototype
example.
2. General results
We shall state our results on the line in [1, Chapter III].
Let X be a compact metric space. Let y be a Borel measure on X and
£^(X) be the Hilbert space of square integrable functions on X. Let h: X x
X —> R be a continuous function. Then, the linear mapping defined by
(Lhf)(x) - F(x) = [ f(t)h(t,x)dy(t) (2.1)

Jx
is well defined from into £^(X). For several fundamental properties
for the operator Lh, see [1, Chapter III].
Now, in order to examine, the integral transform (2.1) for £%(X) we form
the function K(x, yjonExE defined by
K(x,y) = [ h(t,y)h(t,x)dy(t). (2.2)
Jx
Then, K{x, y) is a positive matrix on E in the sense: for any finite set {xj}j C
X and for any real numbers
CiCjEC(xj, x^ > 0
i 3
(now we shall consider all on the real number field R, following [1]). There
fore, following the fundamental theorem by Aronszajn-Moore, there exists a
uniquely determined reproducing kernel Hilbert space Hk(X) which is char
acterized by the properties:
Kf,y)eHK(X) for any y€ A"
and
for any F e Hk(X) and for any y G X :

General approximation error and convergence rate estimates 161
For simplicity, we shall assume that both {h(t, x) : x e X} and {h(t, x) : t E

X} are complete in £^(X). Then, we have the isometric identity
ll/lk-sw = IlJ’kxm. <2-3)
and, in some general principle, we can obtain the inversion formula:
*
f
F-
in (2.1). The important point is that the image functions F in (2.1) for £%{X}
are characterized as the members of the reproducing kernel Hilbert space Hk(X).
We gave various isometric identities and the inversion formulas in [2] from this
viewpoint.
We note that the adjoint operator L* h of Lh is realized by the integral trans
form
= [ f(t)h(x,t)di/(t) (2.4)
Jx
on Then, the composition operator L^Lh is realized in the form
F(x) = [ f(t)K*(t,x)di/(t) (2.5)

Jx
for
K*(t,x') = [ h(x,^h(t,^dy^ (2.6)

Jx
as we see, using Fubini’s theorem. We define the associated reproducing ker
nel, again, for (2.5)
K(x,y)= *
[ K\t,y)K
(t,x')dv(t')
Jx
and the associated reproducing kernel Hilbert space -Hk(X). Since the family
(t,x)
*
{K : x e X}
is complete in ZL^(X), we have the isometric identity
ll/llr2(X) = IIÎIhk(X), (2.7)
in (2.5).
Since the operator L^Lh is a self-adjoint, positive compact operator on

£^(X), by Proposition 1.1, for A = L^Lh, s — 2, r = 1 we obtain, for
G 6
min IIG-IILUWIksm < |II~1 Gll^m. (2.8)

III (X)-H 21
Here, we see that G must be a member of the reproducing kernel Hilbert space
Hk(X') for (L^Lh)~lG to have meaning, and furthermore,
||(^^)-1G||£2W = ||G||Hk (X). (2.9)
We thus obtain
Theorem 2.1 For any member G ofH^X) we have the estimate
i 4iIgII«k(x)- (2.10)
Ilf ll£2(x)-" 21
In general, iffc:XxX—>Risa positive matrix, continuous and symmet

ric, then k(x, y) is called a Mercer kernel; it appears in many cases. In our
* (t, x) is a Mercer kernel and when we consider the linear mapping L
line, K
defined by (2.5), then we obtain, using the isometry (2.7),
Corollary 2.1 For any member G € 77k (X) in (2.5), we obtain the estimate
HF,!™11 <JG-LFI^W 5 ^|GII«k(X)- <211>

IkIIhk(x)S^ rv
The estimate (2.11) just meets the approximation form in statistical learning
theory in [3] and approximations in reproducing kernel Hilbert spaces in [2].
In Corollary 2.1, by setting G = Lg for g G we have
Corollary 2.2 For any member g G we obtain the estimate in (2.5)
min \\Lg - LF\\C2W < ^|MI^(X).

IkIIhk(x)S^ ™
By using the isometry in (2.5), we obtain, in Corollary 2.2
Corollary 2.3 For any g e we have the estimate
n,iimin<pll£9-W||r2(x) < 4l|p|lr2(x)-

II-/n,
In connection with the general powers in Propostion 1.1, without loss of

generality, we can consider s = 1, because we can consider As as A.
Meanwhile, for r > 0, consider the relation between the integral transforms
(2.4) and (2.5) for symmetric kernel h(x, t). Then, we can see how to realize
the 1/2 operator from (2.5) to (2.4). In this way, we can see how to consider
the general power l/2n (n > 1, integer) of a positive operator, by repeating
the procedure.
3. Example
As a typical example, we shall consider the integral transform
F(x) = fT ftt)U(x,T-£)d£ (3.1)

Jo
for
which is appeared in the wave equation in one-dimensional space ([2, 143-

146]). Then, the associated reproducing kernel is, on [0, T] x [0, T],
K{x^x2} = ITU(x2,T-Û^,T-C^
Jo
= min(T — xi,T — x2) (3.2)
and the reproducing kernel Hilbert space Hr[0, T] is composed of all functions
F(x) such that
1) F(x) is absolutely continuous on [0, T],
2) F'(x)GL2(0,n
3) F(T) = 0,
and the inner product is given by
4) (F^F^h^'T] = J?F{(x)F!,(x)dx.
Let Lu denote the operator in (3.1) on L2(0, T). Then the operator L^Lu
on L2 (0, T) is realized by the integral transform
T
F(x) = [ f(£)K(x,£)d£. (3.3)
Jo
We shall consider the associated reproducing kernel, on [0, T] x [0, T],

T
K(x1,x2)=/' K{x2,^K{Xl,^. (3.4)
Jo
Then, we see that the reproducing kernel Hilbert space .Hk[0, T] is composed
of all functions F(x) such that
1) F(x), F'(x) are absolutely continuous on [0, T],
2) F"(x) e L2(0,T),
3) F(T) = F'(0) = 0,
and with the inner product

T
(F1,F2)Hk[0>t]= [ F'^F^dx.
Jo
These facts are confirmed directly by
rT rx rT
W=T ftt)d£-x f (£)<%— I
Jo Jo Jx
F'(x) = - [x /(CM
Jo
and
F"(x) = -/(x),
in (3.3). In (3.3), we have the isometric identity
[T F F"(xfdx. (3.5)
Jo Jo
Then, from Corollary 2.1, we obtain
A. For any G E Hk[0, T], we have the estimate
min G(t) - [T F(OK(?,C)dti < II G|I^K |0,r] -

IIÎIhk[o,t] <r Jo l2(o, t)
From Corollary 2.2, we obtain
B. For any g E £2(0, T), we obtain the estimate

min g(^K(x^d^- / F^K{x^d^

IIÎIhk[o,t] <r L2{0,T)
- ÎIsIIl2(o,t)-
From Corollary 2.3, we have
C. For any g € L? (0, T), we have the estimate
min I g(£)K(x,£)d£
llflk2[o,tj -R Jo
f(£)K(ri,£)d£ K(x,rf)dr)
L2(0,T)
IMll2(0,T)-
References
[1] F. Cucker and S. Smale: ‘On the mathematical foundations of learning’,
Bull. Amer. Math. Soc. 39 (2001), 1-49.
[2] S. Saitoh: 'Integral Transforms, Reproducing Kernels and their Applica
tions’, Pitman Res. Notes in Math. Series 369, Addison Wesley Longman,
UK, 1997.
[3] S. Smale and D. X. Zhou: ‘Estimating the approximation error in learning
theory’, Analysis and Applications (to appear).
Chapter 12
FUNCTIONAL EQUATIONS
ON HYPERGROUPS
Laszlo Szekelyhidi
Institute of Mathematics and Informatics, University of Debrecen
Pf. 12. 4010 Debrecen, Hungary
szekely@math.klte.hu
Abstract This paper presents some recent results concerning functional equations on hy
pergroups. The aim is to give some idea for the treatment of classical func
tional equation problems in the hypergroup setting. The general form of addi
tive functions, exponentials and moment functions of second order on discrete
polynomial hypergroups is given. In addition, stability problems for additive and
exponential functions on hypergroups are considered.
2000 MSC: 39B40 (Primary); 20N20 (Secondary).
Keywords: functional equation, hypergroup
1. Introduction
The concept of DJS-hypergroup (according to the initials of C.F. Dunkl,
R.I. Jewett and R. Spector) is due to R. Lasser (see e.g. [4]). One begins with
a locally compact Haussdorff space K, the space A't(K) of all finite complex
regular measures on K, the space of all finitely supported measures in
At(A'), the space of all probability measures in M(K), and the space
A4c(7<) of all compactly supported probability measures in M(K). The point
mass concentrated at x is denoted by 5X. Suppose that we have the following:
'j
*
(H There is a continuous mapping (x, y) i—> 6X * 6y from K x K into
the latter being endowed with the weak
*
-topology with re
spect to the space of compactly supported complex valued continuous
functions on K. This mapping is called convolution.
167
(ffv) There is an involutive homeomorphism x *—> xv from K to K. This

mapping is called involution.
(He) There is a fixed element e in K. This element is called identity.
)
*
Identifying x by 6X the mapping in (H has a unique extension to a con
tinuous bilinear mapping from AA (K) x AA(K) to AA(K). The involution on
K extends to an involution on AA(K). Then a DJS-hypergroup, or simply hy
pergroup is a quadruple (K, *, V, e) satisfying the following axioms: for any
x, y,z in K we have
(Hl) 5X * (6y * 6Z) = (6X * 6y) * 6Z;
(H2) (6X * 6y)^ = 8yv * 8xv,
(H3) 6X * 8e = 6e * 8X = <5X;
(HA) e is in the support of 8X * 8yv if and only if x = y;
(H5) the mapping (x, y) i—> supp(5x * 8y) from K x K into the space of
nonvoid compact subsets of K is continuous, the latter being endowed
with the Michael topology (see [1]).
If 8X * 6y = 6y * 6X holds for all x, y in K, then we call the hypergroup

commutative. If xv = x holds for all x in K then we call the hypergrvup
Hermitian. By (H2) any Hermitian hypergroup is commutative. For instance,
if K = G is a locally compact Haussdorff group, 8X * 8y = 8xy for all x, y in
K, xv is the inverse of x, and e is the identity of G, then we obviously have
a hypergroup (K, *, V, e), which is commutative if and only if the group G
is commutative. The same works if K = S is a locally compact Haussdorff
semigroup with identity, and involution is the identity mapping. However,
not every hypergroup originates in this way, as it is shown by the following
example.
Let 0 < 0 < 1 be arbitrary and let K = {0,1}. We define e as 0 and
involution as the identity map. The products = ô, = <$i *<$o = <^1
are obvious, and we let
= 6>5O 4- (1 — 6>)5X.
It is easy to see that we get a hypergroup for any 0 in ]0,1]. In fact, any two-
element hypergroup arises in this way. For 0 = 1 we get the two-element
group of integers modulo 2.
In any hypergroup K we identify x by 6X and we define the right translation
operator Ty by the element y in K according to the formula:
Tyf(x) = f fd(6x*
6 y),
Jk
for any / integrable with respect to 8X * 6y. In particular, Ty is defined for

any continuous complex valued function on K. Similarly, we can define left
translation operators but at this moment we do not need any extra notation for
them.
Sometimes one uses the suggestive notation
f(xty) = [ fd(8x*
8 y),
JK
for any x,y in K. However, we call the attention to the fact that actually
f(x * y) has no meaning in itself, because x * y is in general not an element
of K, hence f is not defined at x * y. The expression x * y denotes a kind of
“indistinct” product. If B is a Borel subset of K, then 6X * 8y(B), or, what is
the same, x * y(B) expresses the probability of the event that the “product” of
x and y belongs to the set B. In the special case of groups, or semigroups this
probability is 1, if B contains xy, and is 0 otherwise, that is, exactly 8xy(B).
If K is a discrete topological space, then we call the hypergroup a discrete
hypergroup. As in this paper we deal mainly with discrete hypergroups we
summarize here its axioms.
Let K be a set and suppose that the following properties are satisfied.
*)
(£> There is a mapping (x, y) 8xt 8y from K x K into Adj (K), the space
of all finitely supported probability measures on K endowed with the
weak
-topology
* with respect to the space of finitely supported complex
valued functions on K. This mapping is called convolution.
(£>v) There is an involutive bijection x i—> xy from K to K. This mapping is

called involution.
(De) There is a fixed element e in K. This element is called identity.
Identifying x by 6X as above convolution extends to a continuous bilinear

mapping from M(K) x M(K) to M(K), and the involution K extends
to an involution on M.(K). Then a discrete DJS-hypergroup is a quadruple
(K, *, V, e) satisfying the following axioms: for any x, y,z in K we have
(DI) 6X * (6y * 6Z) = (8X * 6y) * 6Z;
(D2) (6X * 6y)v = 6yv * 6xv,
(D3) 6X * 6e = 6e * 8X = 8X\
(Z)4) e is in the support of 6X * 8yv if and only if x = y.

2. Polynomial hypergroups
An important special class of Hermitian hypergroups is closely related to
orthogonal polynomials.
Let (an)neN, and (cn)nepj be real sequences with the following
properties: cn > 0, bn > 0, an+i > 0 for all n in N, moreover ao = 0, and
an + bn + cn = 1 for all n in N. We define the sequence of polynomials
(Pn)neN by P0(x) = 1, -Pi(z) = x, and by the recursive formula
xPn{x) = anPn-i(x) + bnPn(x) + CnPn+i(x)
for all n > 1 and x in R. The following theorem holds.
Theorem 2.1 If the sequence ofpolynomials (Pn)neN satisfies the above con
ditions, then there exist constants 0(71, m, k)for all n, m, k in N such that
n-f-m
pnpm -- 52 c(n> k^Pk

Zc=|n—m|
holds for all n, m in N.
Proof: By the theorem of Favard (see [2], [5]) the conditions on the sequence
of polynomials (Pn)neN imply that there exists a probability measure /z on
[—1,1] such that (Pn)neN forms an orthogonal system on [—1,1] with respect
to p,. As Pn has degree n, we have
n-\-m
PnPm ~ k)Pfc
k=0
for all n, m in N, where
J-l PkPnPm dp
c(n, m, fc) =
holds for all n,m, k in N. The orthogonality of (Pn)neN with respect to p,

implies c(n, m, k) = 0 for k > n + m or n > m + k or m > n + k. Hence
our statement is proved. □
The formula in the theorem is called linearization formula, and the coeffi
cients c(n, m, fc) are called linearization coefficients. The recursive formula
for the sequence (Pn)neN implies Pn(l) = 1 for all n in N, hence we have
n-f-m
^2 c(n> fc) = 1
Zc=|n—m|
for all n in N. It may or may not happen that c(n, m, fc) >0 for all n, m, k
in N. If it happens then we can define a hypergroup structure on N by the
following rule:
b
*
bn m= c(n,m,/c)<5fc
Zc=|n—m|
for all n, m in N, with involution as the identity mapping and with e as 0.

The resulting discrete Hermitian (hence commutative) hypergroup is called the
polynomial hypergroup associated with the sequence (Pn)neN-
As an example we consider the hypergroup associated with the Legendre
polynomials. The corresponding recurrence relation is
xPn^ = 2^nPn+i(T)+2ûPn-1(x)
for all n > 1 and x in R. It can be seen easily that the linearization coefficients
are nonnegative, and the resulting hypergroup associated with the Legendre
polynomials is the Legendre hypergroup.
3. Exponential polynomials on hypergroups

In the case of commutative groups exponential polynomials play a funda
mental role in several problems concerning functional equations. As exponen
tial polynomials are built up from additive and exponential functions, which are
closely related to translation operators, the presence of translation operators on
hypergroups makes it possible to define these basic functions on hypergroups,
too.
Let AT be a hypergroup with convolution *, involution v, and identity e.
For any y in K let Ty denote the right translation operator on the space of all
complex valued functions on K which are integrable with respect to 6X * 5y for
any x, y in K. In particular, any continuous complex valued function belongs
to this class. We call the continuous complex valued function a on K additive,
if it satisfies
Tya(x) = a(x) + a(y)
for all x, y in K. In more details this means that
/ a(t) d(6x * = a(x) + a(y)

JK
holds for any x, y in K. The continuous complex valued function m on K is
called an exponential, if it is not identically zero, and
Tym(x) = m{x)m{y)
holds for all x, y in K. In other words m satisfies the functional equation
/ zn(t) d(6x * Sy)(F) = m(x)zn(y).

JK
It is obvious that any linear combination of additive functions is additive again.
However, in contrast with the case of groups, the product of exponentials is not
necessarily an exponential. The bounded exponential m is called a character
if m(xv) = m(x) holds for any x in K. Obviously a(e) = 0 for any additive
function a, and m(e) = 1 for any exponential m.
An exponential monomial on a locally compact Abelian group G is a prod
uct of additive and exponential functions, that is, a function of the form
x •—> ai(z)a2(a0 • • • an(x)mi(x)m,2(x) ■ • ■ mk(x),
where n, k are nonnegative integers, cq, <22,..., an: G —> C are additive func
tions, and mi, m2,..., m^ are exponentials. As the product of exponentials is
an exponential again, we can write any exponential monomial in the form
re I—> ai(x)a2(x) ■ ■ ■ an(x)m(x},
where n is a nonnegative integer, a1( 02,..., an: G —> C are additive func
tions, and m is an exponential. In the case n = 0 we consider this function to
be identically equal to m. A linear combination of exponential monomials is
called an exponential polynomial. A product of additive functions is called a
monomial, and a product of monomials we call a polynomial.
If we want to introduce these concepts on commutative, or arbitrary, hy
pergroups, we have to remember the fact that a product of exponentials is not
necessarily an exponential. On the other hand in case of commutative groups
exponential polynomials can be characterized by the fact that the linear space
of functions spanned by the translates is finite dimensional. This property is of
fundamental importance from the point of view of spectral synthesis. Hence
it seems to be reasonable to define exponential polynomials by this property,
even on arbitrary - not necessarily commutative - hypergroups. Anyway, ad
ditive and exponential functions on hypergroups obviously have this property
and it seems to be interesting to describe these function classes on different
hypergroups.
4. Exponential functions on polynomial

hypergroups
Let K = (N, *) be the polynomial hypergroup associated with the sequence
of polynomials (Pn)neN- Now we describe all exponential functions defined
on K (see also [1]).
Theorem 4.1 Let K be the polynomial hyperg roup associated with the se
quence ofpolynomials (Pn)neN- The function N —> C is an exponential on
K if and only if there exists a complex number z such that
++) = Pn(z)
holds for all n in N.
Proof: First of all we remark that if a sequence of polynomials (Pn)neN

satisfies a recursion of the form
n+m
Pn{x)Pm(x) = c(n,7n,fc)Pfc(x)
it=o
with some real or complex coefficients c(n, m, fc) for all real x, then the recur
sion holds for all complex x. Let z be a complex number and = Pn(z)
for any n in N. Then by the definition of convolution we have for any m, n in
N
n+m
k—\n—m\
n+m
= 52 c(n,m,k)Pk(z) = Pn(z')Pm(z') -
fc=|n—m\
hence is exponential.
Conversely, let be an exponential on K and we define z = By the
exponential property we have for all positive integer n that
n+1
Z(p(n) = (/?(l)(^(n) = <^(5i * 8n) = c(n, 1, k)cp(k)

k=n—l
= c(n, 1, n — l+(n — 1) + c(zz, 1, n)<^(n) + c(n, 1, n + l)</?(n + 1).
As the same recursion holds for n i-> Pn(z), further <p(0) = 1 = Po(z) and
9?(1) = z = P^z), hence <y?(n) = Pn(z) for all n in N and the theorem is
proved. □
Hence any exponential function on a polynomial hypergroup arises from the

evaluations n i-> Pn(z) associated with the generating polynomials.
5. Additive functions on polynomial

hypergroups
Again we let K = (N, *), the polynomial hypergroup associated with the
sequence of polynomials (Pn)neN- We describe all additive functions defined
on K (see also [1]).
Theorem 5.1 Let K be the polynomial hypergroup associated with the se
quence of polynomials (Pn)n&$- The function a: N —> C is an additive func
tion on K if and only if there exists a complex number c such that
a(n) = cf^(l)
holds for all n in N.
PROOF: Suppose first that a: N —> C is an additive function on K. Then by

the additive property we have for any positive integer n that
n+l
a(n) + a(l) = a(8n * Ji) = c(n, 1, fc)a(fc)

k=n—1
= c(n, 1, n — l)a(n — 1) + c(n, 1, n)a(n) + c(n, 1, n + l)a(n + 1).
On the other hand, we have for any positive integer n and for all real x that
xP(x) — c(n, l,n— l)Pn-i(x) + c(n, l,n)Pn(x)+c(n, 1, n+l)Pn+i(a:).
Differentiating both sides with respect to x and substituting x = 1 we get
1+-F£(1) = c(n,l,n-l)P4-i(l)+c(n,l,n)P4(l) + c(n.,l,n+l)P4+1(l)
for any positive integer n. Multiplying both sides by a(l) we see that the
functions n i—> a(n) and n i—> a(l)P^(l) satisfy the same recursion, further
a(0) = Po(l) and a(l) = a(l)P{(l), hence a(n) = a(l)P^(l) for all n in N.
Conversely, we consider the linearization formula
71+771
Pn(^)Pm(^) = ^2
171—7711
which holds for any n.m in N and for any real x. Differentiating both sides
with respect to x and substituting x = 1 we get
714-771
Pn(l)P^(l) + P4(l)Pm(l)= £ c(n, m, fc)J^(l)

Zc=|71—77l|
for all n, m in N. As P„(l) = Pm(l) = 1 for all n, m in N, this formula shows

that the function n > P^(l) is additive, hence n i—> cPP(]P) is additive for any
complex number c, and the theorem is proved. □
For instance, in the case of the Tchebycheff hypergroup of the first kind one
gets the general form of additive functions as
a(n) = cn2
for all n in N, where c is an arbitrary complex number.
6. Moment functions on polynomial

hypergroups
Let K be a hypergroup and N a positive integer. The continuous function
<p: K —> C is called a momentfunction of order N if there are complex valued
continuous functions p>k- K C for k = 0,1,..., N such that <po = 1,
FN = and
j=o V'
holds for k = 0,1,..., N and for all x, y in K. In this case we say that the
functions <pk (& = 0,1,...,TV) form a moment sequence of order N. Hence
moment functions of order 1 are exactly the additive functions. The study of
moment functions and moment sequences on hypergroups leads to the study of
the above system of functional equations. In case of polynomial hypergroups
we can describe the moment sequences of order N completely. Here we give
the result for TV = 2 only; the general result will be published elsewhere. We
need the following simple theorem.
Theorem 6. 1 Let K be the polynomial hyperg roup associated with the se
quence ofpolynomials (Pn)neN- If the function ip: N —> C satisfies
<M
*
^n = V’W +V’(1)
for all n in N, then ip is additive.

Proof: In the proof we will use the following notation: for any n in N let
pH = Pn<X)-
By differentiating the linearization formula and substituting x = 1 we have
p(f>n — p(n) + 1,
hence
^(l)p(Ai * <51) = ip(T)p(n) + V’(l)

holds for all n in N. As V’(O) = p(0) = 0 and V’(l) = V’(l)p(l)> and the
functions n and n i—> V’(l)p(w) satisfy the same recurrence relation
of order 2, they must be equal: i/i(n) = ^(pP^fX)- The theorem is proved. □
Theorem 6. 2 Let K be the polynomial hypergroup associated with the se

quence of polynomials (Pn)n&i- The complex valued functions <po> <pi, p2 on
K form a moment sequence if and only if
< Po(n) = 1,
< Pl(n) = aP^f),
^nâ^Pf + bP^,
for all n in N, where a, b are arbitrary complex numbers.
Proof: We will use the following notation: for any n in N and for k = 0,1,2
let pk(n) = Pnk\V).
By differentiating k times the linearization formula for the sequence (Pn)neN
(k — 1,2) and substituting x = 1 we have
Pl(5n * 6m) = Pl W + Pl(™),
and
P2(5n * 5m) = P2(n) + 2pi (n)pi (m) + P2(m)
for all n, m in N. Using these relations an easy computation shows that the
functions <pk (k — 0,1,2) given in the theorem form a moment sequence of
order 2 with arbitrary complex constants a, b.
Conversely, suppose now that the functions <pk (k — 0,1,2) given in the
theorem form a moment sequence of order 2. Then we have that <pi is additive
and by the previous theorem <pi(n) = aP^l) for all n in N, with a = <pi(l).
On the other hand for any n in N it follows that
< P2(5n * 5i) = <p2(n) + 2a<pi(n) + <p2(l),
and from the above equation with m = 1
P2(5n * 51) = P2(n) + 2pi(n).
Hence we have the following two equations for all n in N:

< P2(5n * 5J = <p2(n) + 2a2pi(n) + <p2(l),
a2P2(5n * 5i) — a2p2(n) + 2a2pi(n).
By subtraction we have that the function •’ N —> C defined on N by

V >(n) = <p2(n) - a2p2(n)
satisfies the relation
* <51) = V’(w) + V’(l)
for all n in N. By the previous theorem V’ is additive, and by Theorem 5.1 we

get V’(n-) = bPn(F) for all n in N with some complex number h. By substitution
we have
wW = a2P"(l) + 6P'(1)
for all n in N, and the theorem is proved. □
7. Stability problems on hypergroups

Stability theory of functional equations on groups, on semigroups and on
different algebraic structures has attracted the attention of several mathemati
cians recently. In this respect see [8]. According to our knowledge stability
problems for functional equations on hypergroups have not been considered so
far. In this section we show that using similar ideas as in the case of groups,
semigroups, etc., analogous results can be obtained.
First we deal with the stability of exponential functions on hypergroups.
The following result is similar to that of [7].
Theorem7.1 Let K be a hypergroup and let f,g: K —> C be continuous

functions with the property that the function y fKf d(6x * Sy) —
is bounded for all y in K. Then either f is bounded, or g is exponential.
PROOF: It is easy to see ([1]) that for any continuous function p: K —> <C and
for any y in K the functions x w d(8x * 8y) and y fK<pd(5x * 8y)
are continuous. First of all we can check easily that the associativity of the
convolution implies that
[ [ ip(s) d(8t
S
* z)(s) d(Sx*
6 y)(t) = [ [ p(s) d(8x* S z)(t)
S t)(s) d(Sy*
JkJk JkJk
holds for all x, y, z in K. Actually on the left hand side we have
/ pd[(8X * 8y) * 5Z],

JK
and on the right hand side
/ pd[8x * (Sy *
5 2)].
Jk
The assumption of the theorem means that
/(s) d(6t * Sz) - f(t)g(z) <M(z)

holds for all x, y in K with some given continuous function M: K —> C.

Integrating with respect to the measure 6X * 6y and with respect to the variable
t we have for all x, y, z in K:
5 y)(t) <M(z).
[ [ /(s)d(<5t *<5 2)(s)d(<5x *<5j,)(f) - f f(t)g(z)d(6x*
JkJk Jk
Let
A(x,y,z)=[ f f{s)d{6t
8
* z){s)d(8x*8 y)(t')- [ f{t)g{z)d(Sx*
8 y){t')
JKJK Jk
for all x, y, z.
Using again the assumption of the theorem we have
I f(s) d(6x * - f(x)g(t) <M(t)

K
for all x, t in K. Now we integrate both sides with respect to the measure
8y * 8Z and with respect to the variable t to get
[ [ f(s)d(Sx*
S t)(s)d(5y S z)(t)
[ f(x)g(t)d(8y *
JKJK JK
< [ M(t)d(6y * 8z)(t)
JK
for all x, y, z in K. Let
B(x,y,z)=[ [ f(s)d(8x *
5 t)(s)d(5y *
J 2)(i)
JK Jk
- [ f(x)g(t)d(6y *
6 z)(t)
JK
for all rr, y, z in K.
Our next observation is that
6 y)(t)g(z)
f(x)g(y)g(z) - [ fWd(6x*
< g(z)M(y)
JK
holds for any x^ y, z in K. Let
C(rr, y, z) = /(x)p(y)y(z) [ f(t) d(5x * ^(fyg^z)

JK
for all x, y, z in K.
Finally, we consider the identity
A(x, y, z) - B(x, y, z) - C(x, y, z)
= f(x) / d(sy *)
* <M( - g(yW
\JK
which holds for all x, y, z in K. From the above inequalities we can see that
the left hand side is bounded for any fixed y, z in K, hence if g is not an
exponential, then f is bounded, and the theorem is proved. □
The stability of additive functions on hypergroups can be derived easily in

the presence of an invariant mean, which depends on the hypergroup. How
ever, an invariant mean always exists on commutative hypergroups, as it fol
lows from the Markov-Kakutani fixed point theorem (see [3]). We recall that
an invariant mean M on a commutative discrete hypergroup K is a bounded
linear functional on the Banach space of all bounded complex valued functions
defined on K with the property that M (1) = 1, and
Mx / d(5x * 6y)
Jk
for any bounded function : K —> C and for any y in K.
Theorem 7.2 Let K be a commutative discrete hypergroup and f: K —> C a

function with the property that the function
(z,y)^ [ fd(6x*
6y)-f(x)-f(y)
JK
is bounded on K x K. Then there exists an additive function a: K —> C such
that f — a is bounded.
Proof: We can prove this theorem following the lines of [7]. For any fixed y
in K we define
a(y) = Mx I f d(6x * 6y) — f (x)

K
where M is any invariant mean on K, and Mx indicates that we apply M to
the (obviously bounded) function in the brackets, as a function of x. Now we
have
/ a(t)d(5j, *
5 2)(t)-a(y)-a(z)
Jk
= M. 6y)(t)
S t)(s)d(5y *5 x)(t) - f f(t)d(Sx*
[ [ f(s)d(Sz*
.Jk Jk Jk
— Mx f(t) d(fx * 6z)(t) - f(x)

UK
for all t/, z in K. If : K —> C is the function defined by
= / /(<) d($x * W) - f(x)
for all x, z in K, then the above equation can be written in the form
a(f) d(6y * J2)(t) - a(y) - a(z) — Mx d(6x * <^)(t)] -
and the right hand side is 0 by the invariance of M. It means that a is additive.
On the other hand,
|a(y) - /(y)l = Mx [ /(t) d(6x * 5j,)(t) - f(x) - f(y) <
holds for all y in K, where C is a bound for the function
, y)
(* / fd(dx * Sy) - f(x) - /(y).
The theorem is proved.
References
[1] W.R. Bloom and H. Heyer: 'Harmonic Analysis of Probability Measures
on Hypergroups’, Berlin, New York, 1995.
[2] J. Favard: ‘Sur les polynomes de Tchebycheff5, C. R. Acad. Sci. Paris
200(A-B) (1935), 2052-2053.
[3] A.T. Lau: ‘Analysis on a class of Banach algebras with applications to
harmonic analysis on locally compact groups’, Fund. Math. 118 (1983),
161-175.
[4] K.A. Ross: ‘Hypergroups and Signed Hypergroups’, in: ‘Harmonic Anal
ysis on Hypergroups’ (eds: J.M. Anderson, G.L. Litvinov, K.A. Ross, A.I.
Singh, V.S. Sunder, N.J. Wildberger), de Gray ter Studies in Mathematics,
de Grayter-Birkhauser, Boston-Basel-Berlin, 1998, pp. 77-91.
[5] J. Shohat: ‘The relation of the classical orthogonal polynomials to the poly
nomials of Appel’, Amer. J. Math. (1936), 453—464.
[6] L. Szekelyhidi: ‘On a theorem of Baker, Lawrence and Zorzitto’, Proc.
Amer. Math. Soc. 84 (1982), 95-96.
[7] L. Szekelyhidi: ‘Remark 17’, Aequat. Math. 29(1982).
[8] L. Szekelyhidi: Ulam’sproblem, Hyers’s solution — and to where they led,

in: ‘Stability of Functional Equations’ (ed.: Th.M. Rassias), Kluwer Aca
demic Publishers, Boston-Dordrecht-London, 2000, pp. 259-285
Chapter 13
THE GENERALIZED
CAUCHY FUNCTIONAL EQUATION
Abraham A. Ungar
Department of Mathematics
North Dakota State University
Fargo, ND 58105, USA
ungar@gyro.math.ndsu.nodak.edu
Abstract The Cauchy functional equation and the Cauchy-Pexider functional equation are
generalized, and their solutions are determined.
2000 MSC: Primary: 30D05 (Primary).
Keywords: functional equations, Cauchy, Pexider, linear differential equations
1. Introduction
The Cauchy functional equation [14, 15]
F(x)F(y) = F(x + y) (1.1)
and the Cauchy-Pexider functional equation
F(x)G(y)=H(x + y) (1.2)
[2] are well known in the theory of functional equations and inequalities [17].
A complete solution of (1.1) for complex, real and positive valued F was pro
vided by Aczel [1]. A use of (1.2) may be found, for instance, in [8],
The aim of this article is to provide a complete solution to the nth order
generalized Cauchy functional equation
n m
E am+1 ^F<k\x)G(m-k\y) = H(x + y), (1.3)
771=0 k=0
183
n = 0,1,2,..., for entire functions F, G and H, and arbitrarily given n + 1

complex constants ap, p = 1,2,..., n + 1. Clearly, the functional equation
(1.3) reduces to the Cauchy-Pexider functional equation when n = 0 and
<21 = 1.
The importance of the Cauchy functional equation and its various general
izations is evidenced from the literature. Interested readers will be rewarded
by reading about this fascinated functional equation in [2, 3, 4, 5, 6, 7, 12, 13,
14, 15, 16, 17, 18, 19, 20, 21],
2. The Main Result

Theorem 2.1 Two nonzero entire functions of a complex variable, F(z) and
G(z), satisfy the nth order generalized Cauchy functional equation
n m
52 am+1 J; F^(x)G^~k\y) = H(x + y) (2.1)
m=0 k=Q
in ||x||, ||y|| < oo, for some entire function H and given n + 1 complex con
stants dp, p = 1,2,..., n + 1, if and only if F and G are two solutions of the
differential equation
n+l
âkfk\z) = 0 (2.2)
k=0
for some complex constant oq.
Proof: Let I(x, c) be an entire function of two complex variables, given by

the equation
n m
I(x, c) = 22 MI 52 F^k\x)G^-k\c - x) (2.3)
m=0 k=0
which denotes the left hand side of (2.1) with y replaced by c — x. Then (2.1)
can be written as
I(x, c) = H(c) (2.4)
for an unspecified entire function H. Eq. (2.4), in turn, is equivalent to the

equation
= 0. (2.5)
OX
To verify the Theorem it is enough, therefore, to show that F and G satisfy

(2.5) if and only if they satisfy the differential equation (2.2).
Assuming that F and G satisfy the differential equation (2.2), we will es
tablish the validity of (2.5). For the differentiation of the function I(x, c) with
respect to x, in (2.5), we note that
f) m
— ^(fc)(^)G(m-fc)(c - x)
X k=0
m
= ^{F^+1\x)G(m~k\c - x) -F(fc)(z)G(m-fc+1)(c-a;)}
fc=0
= F(m+1)(z)G(c - x) - F(x)G(m+1)(c - x). (2.6)
Hence, from (2.3) and (2.6) we have
= G(c_x) £ am+1F<m+1\x)-F(x) £ am+1G^m+1\c—t).

m=Q m=0
(2.7)
The entire functions F and G satisfy the differential equation (2.2). Hence,
we have
n
£am+1F^+1\x) = -a0F(x) (2.8a)
m=0
and
n
am+1G(m+1)(c - x) = -a0G(c - x). (2.8b)
m=0
Substituting (2.8) in (2.7) we finally establish (2.5).
Conversely, assuming that F and G satisfy (2.5) we will show that F and G
are solutions of the differential equation (2.2). We find from (2.7) that if F and
G satisfy (2.5) then
Fix) G(c-x)
on a region of the complex plane where F(x) and G(c — x) are different from
zero. Identity (2.9) between a function of x and a function of c — x qnqx a
region is valid if and only if each side of the identity is equal to a constant, say,
—a$. This, in turn, implies that F and G satisfy the differential equation (2.2)
and hence the proof. □
The generalized Cauchy functional equation (2.1) exhibits a finite sum de
composition. Finite sum decompositions in analysis are studied in [20].
3. Application
The Theorem can be employed to produce an elegant addition theorem
for solutions to linear, homogeneous, constant coefficient ordinary differential
equations.
Eq. (2.1) implies the equation
n m n m
£ «m+i £ = £ om+1 £ + y)
m=0 k=0 m=0 k=Q
(3.1)
for any two solutions Fr(x) and Fs(rr) of the differential equation (2.2). Eq. (3.1),
in turn, is an addition theorem.
To simplify (3.1) we select Fr(x), r — 0,1,2,..., n, to be the n + 1 lin
early independent solutions of the differential equation (2.2) each of which is
normalized by the initial condition
Frw(0) = 6rk (3.2)
r, k — 0,1,2,..., n, where 6rk is the Kronecker delta (8rk = 0 for r k and

<5rr=l). For these normalized solutions, (3.1) takes the form
n m n
E am+1 ^Fr(k)(^)Fs(m-k\y) = £ am+1F^~r\x + y) (3.3)
m=0 k=0 m=0
r, s = 0,1,2,..., n, thus producing an elegant addition theorem for normal

ized solutions, (3.2), to a linear, homogeneous, constant coefficient differential
equation, (2.2). The right hand side of (3.3) reveals that the variables x and y
are interchangeable, while the left hand side of this equation reveals that the
indices r and s are interchangeable. Hence, we get the identity
n n
E am+1F^~r\x) = £ am+1F^m~s\x), (3.4)
m=0 m=0
r, s = 0,1,2,..., n, for the solutions Fr(z) of the differential equation (2.2)

normalized by the initial conditions in (3.2). A particular case, when r — s =
n, was studied in [22].
A particularly neat addition theorem is associated with the differential equa
tion
/(n+1)(2) = af(z) (3.5)

for any integer n > 0. The differential equation (3.5) is obtained from the
differential equation (2.2) as a special case corresponding to the coefficients
1 if k = n + 1,
ak—\ —oc iffc = 0, (3.6)
0 otherwise,
a being an arbitrary complex constant.

The Addition Law (3.3) for the n +1 linearly independent solutions, F°(z),
r = 0,1,2,..., n, of (3.5), normalized by the initial conditions (3.2), is ac
cordingly
n
£ Ff™ (x)F
* n~V (y) = F?(n~r'> (x + y) (3.7)
k=0
where — (F^(x))(k) = dkFf‘{x)/dxk, etc.

The functions F°{z), called the (n + l)th order a-hyperbolic functions of
rth kind, have been studied in [23] and [11]. The study of two particular cases
corresponding to a = 1 and a — — 1 in (3.5) have a long, checkered history
since they first appeared in a 1757 paper by Vincenzo Riccati (according to
Kaufman [9]) till the recent generalization by A.K. Kwasniewski and B.K.
Kwasniewski [10].
The 2nd order (—l)-hyperbolic functions of first and second kind are the
circular functions cos x and sin x of trigonometry. Similarly, the 2nd order
1-hyperbolic functions of first and second kind are the hyperbolic functions
cosh x and sinh x. Their use in Einstein’s special theory of relativity and its
underlying hyperbolic geometry is fascinating. Readers interested in the hy
perbolic geometry of Bolyai and Lobachevski and its application in Einstein’s
special theory of relativity will be rewarded by reading [24] and [25],
References
[1] J. Aczel: ‘General solutions of a functional equation connected with a
characterization of statistical distributions’, In: Statistical distributions
in scientific work 3 (1975), D. Reidel Publ. Co., Holland, pages 47-55.
[2] Hiroshi Haruki and Themistocles M. Rassias: ‘A new functional equation
of Pexider type related to the complex exponential function’, Trans. Amer.
Math. Soc. 347 (8) (1995), 3111-3119.
[3] D.H. Hyers, G. Isac, and Th. M. Rassias: ‘On the asymptoticity aspect
of Hyers-Ulam stability of mappings’, Proc. Amer. Math. Soc. 126 (2)
(1998), 425^130.
[4] Donald H. Hyers, George Isac, and Themistocles M. Rassias: ‘Stability

of Functional Equations in Several Variables'1, Birkhauser Boston Inc.,
Boston, MA, 1998.
[5] Donald H. Hyers and Themistocles M. Rassias: ‘Approximate homomor
phisms’, Aequat. Math. 44 (2-3) (1992), 125-153.
[6] George Isac and Themistocles M. Rassias: ‘On the Hyers-Ulam stability
of V’-additive mappings’, J. Approx. Theory 72 (2) (1993), 131-137.
[7] George Isac and Themistocles M. Rassias: ‘Stability of ^-additive map
pings: applications to nonlinear analysis’, Intemat. J. Math. Math. Sci.
19 (2) (1996), 219-228.
[8] K.G. Janardan: ‘Characterizations of certain discrete distributions’, In:
Statistical distributions in scientific work 3 (1975), D. Reidel Publ. Co.,
Holland, pages 359-364.
[9] H. Kaufman: ‘A bibliographical note on higher order sine functions’,
Scripta Math. 28 (1967), 29-36.
[10] A.K. Kwasniewski and B.K. Kwasniewski: ‘On trigonometric-like de
compositions of functions with respect to the cyclic group of order n’, J.
Appl. Anal. 8 (1) (2002), in print.
[11] Martin E. Muldoon and Abraham A. Ungar: ‘Beyond sin and cos’, Math.
Mag. 69 (1) (1996), 3-14.
Natur. Geom. 1 (2) (1992), 65-69.
[13] Themistocles M. Rassias: ‘On the stability of the linear mapping in Ba
nach spaces’, Proc. Amer. Math. Soc. 72 (2) (1978), 297-300.
[14] Themistocles M. Rassias (Ed.): ‘Topics in Mathematical Analysis (A vol
ume dedicated to the memory of A.-L. Cauchyf, World Scientific Pub
lishing Co. Inc., Teaneck, NJ, 1989.
[15] Themistocles M. Rassias: ‘On a problem of S. M. Ulam and the asymp
totic stability of the Cauchy functional equation with applications’,
In: General inequalities (Oberwolfach, 1995) 7 (1997), pp. 297-309,
Birkhauser, Basel.
[16] Themistocles M. Rassias. Stability and set-valued functions. In: Analysis
and topology, World Sci. Publishing, River Edge, NJ, 1998, pp. 585-614.
[17] Themistocles M. Rassias (Ed.): ‘Functional Equations and Inequalities',
Kluwer Academic Publishers, Dordrecht, 2000.
[18] Themistocles M. Rassias: ‘On the stability of functional equations in Ba
nach spaces’, J. Math. Anal. Appl. 251 (1) (2000), 264—284.
[19] Themistocles M. Rassias: ‘The problem of S. M. Ulam for approximately
multiplicative mappings’, J. Math. Anal. Appl. 246 (2) (2000), 352-378.
[20] Themistocles M. Rassias and Jaromir Simsa: 'Finite Sums Decomposi

tions in Mathematical Analysis', A Wiley-Interscience Publication, John
Wiley & Sons Ltd., Chichester, 1995.
[21] Themistocles M. Rassias and Jozef Tabor (Eds.): 'Stability of Mappings
ofHyers-Ulam Type’, Hadronic Press Inc., Palm Harbor, FL, 1994.
[22] Abraham A. Ungar: ‘Addition theorems for solutions to linear homo
geneous constant coefficient ordinary differential equations’, Aequati.
Math. 26 (1) (1983), 104-112.
[23] Abraham A. Ungar: ‘Higher order a-hyperbolic functions’, Indian J.
Pure Appl. Math. 15 (3) (1984), 301-304.
[24] Abraham A. Ungar: 'Beyond the Einstein Addition Law and its Gyro
scopic Thomas Precession: The Theory of Gyrogroups and Gyrovec
tor Spaces’, Kluwer Academic Publishers, Dordrecht-Boston-London,
2001.
[25] Abraham A. Ungar: ‘Applications of hyperbolic geometry in relativity
physics’, In: Janos Bolyai Memorial Volume, A. Prekopa, E. Kiss, Gy.
Staar and J. Szenthe (Eds.), Vince Publisher, Budapest, 2002.
Chapter 14
ON THE
ALEKSANDROV-RASSIAS PROBLEM
FOR ISOMETRIC MAPPINGS
Shuhuang Xiang
Department of Applied Mathematics and Software
Central South University
Changsha, Hunan 410083, P. R. China
xiangsh@mail.csu.edu.cn
Abstract Let X and Y be normed real vector spaces. A mapping T : X —> Y is called
preserving the distance r if for all x, y of X with ||x — y\\x = r then ||T(t) —
T(y) || y = r. In this paper, we provide an overall account of the development
of the Aleksandrov problem, especially the Aleksandrov-Rassias problem for
mappings which preserve distances with a noninteger ratio in Hilbert spaces.
2000 MSC: 46B20, 51K06, 54E35 (Primary).
Keywords: isometry, preserving distance, Banach space, Hilbert space
1. Introduction and the Aleksandrov problem

The theory of isometry had its beginning in the important paper by S. Mazur
and S. Ulam (cf. [16]) in 1932. Let X, Y be two metric spaces, di,d2 the
distances on X, Y. A mapping f: X —> Y is defined to be an isometry if
/(y)) = dx(rr, y) for all elements x, y of X. Mazur and Ulam [16]
proved that every isometry of a normed real vector space onto a normed real
vector space is a linear mapping up to translation.
When the target space Y is a strictly convex normed real vector space, for
into mappings J. Baker [2] proved that every isometry of a normed real vector
space into a strictly convex normed real vector space is also a linear isometry
up to translation.
191
However, for into mappings, if Y is not a strictly convex normed space, an

isometry f: X —> Y may not be an affine mapping. For example, f: R —>
(R2, || • Hoo) defined by f(x) = (x, sin x) is an isometry under the normwise
|| • || oo, but f is not an affine mapping.
Given two metric spaces X and Y and a mapping T: X —> Y, what do we
really need to know about T in order to be sure that T is an isometry?
A mapping T: X —> Y is called preserving the distance r if for all x, y of
X with di(x, y) = r then dz(T(x), T(y)) = r. A.D. Aleksandrov (cf. [1]) has
posed the following problem:
Aleksandrov Problem. Examine whether the existence of a single

conservative distance for some mapping T implies that T is an
isometry.
Let X, Y be two normed vector spaces. Consider the following conditions

for T: X —> Y introduced for the first time by Rassias and Semrl [37]: dis
tance one preserving property (DOPP) and strongly distance one preserving
property (SDOPP).
(DOPP) Let x,y e X with ||x — y\\x = 1. Then — T(y)||y = 1.
(SDOPP) Let x,y e X. Then ||T(z) — T(y)||y = 1 if and only if ||x — y\\x =
1.
Mappings satisfying the weaker assumption that they preserve unit distance
in both directions are not very far from being isometries. Rassias and Semrl
[37] proved that
Theorem 1.1 Let X and Y be real normed vector spaces such that one of
them has dimension greater than one. Suppose that f: X —» Y is a surjective
mapping satisfying (SDOPP). Then f is an injective mapping satisfying
I \\f(x) - /(y) II - ||x - y|| I < 1 for all x,y e X.
Moreover, f preserves distance n in both directions for any positive integer n.
In Theorem 1.1, the assumption that one of the spaces has dimension greater
than one cannot be omitted. For example ([37]), let /: R —> R be defined by
x+2 if a; is an integer,
x if x is not an integer,
k
The mapping f is bijective and preserves distance n in both directions for any
positive integer n, but does not satisfy the inequality in Theorem 1.1. The con
dition (SDOPP) cannot be replaced by (DOPP). Let g: [0,1] —> [0,1) x R be a
bijective mapping and define f: R —> R2 by /(t) = g(t — [<]) + ([i], 0), where
[i] denotes the integer part of t. Then f is a bijective mapping preserving dis
tance n for all positive integer numbers n. However, the inequality in Theorem
1.1 is not fulfilled. Furthermore, the inequality in Theorem 1.1 is sharp (cf.
[37]).
A number of authors have discussed Aleksandrov’s problem under certain
additional conditions for a given mapping satisfying (DOPP) in order to be an
isometry and have posed several interesting and new open problems (cf. [3],
[4], [5], [7], [19]—[36], [44], [45]).
Let X and Y be the Euclidean spaces Rn and R"1 respectively. For the real
line R, Beckman and Quarles [3] pointed out that such a mapping T: R —> R
satisfying (DOPP) does not need to be an isometry. For example ([3]), let
T: R —> R be defined by
. x+1 if z is an integer,
T(o:) = <
x otherwise.
T satisfies (DOPP) but T is not an isometry. Rassias [28] gave a counterex

ample even if T: R —> R is continuous, onto and satisfies (SDOPP): Let
T(x) = [rr] + (x — [or])2, Vrr € R, where [rr] is the integer part of x. It is
easy to verify that T is continuous, injective and satisfies (SDOPP). Since R is
a connected set and T is continuous, by the definition of T, it follows that T is
surjective. However, T is not an isometry.
In case n m, up to now the Aleksandrov problem is only partially solved.
In particular, in the most natural case (n = 2, m = 3) the answer is not yet
known (cf. [26]). Rassias in [21] gave some interesting counterexamples for
T: R2 —» R8 and T': R2 —» R6 such that T and T' satisfy (DOPP) but T and
T' are not isometries.
Example 1.1 ([21]). Let us take a unit 8-simplex P in R8. This simplex has
9 vertices. It is enough to find a mapping f: R2 —> P which preserves unit
distance. Partition the plane into squares with unit diagonals as shown in Fig.
1.1. Each square contains the bottom left comer and two edges containing this
comer. Now number the squares using the numbers from 1 to 9 in the way
shown in Fig. 1.1. It is easy to see that if two points lie in squares of equal
numbers, then these points have distance 1. Now, number the vertices of the
simplex P and map all the points of the plane to the vertices denoted by a
suitable number.
If one uses hexagons instead of squares, we may construct such a mapping

from R2 into R6 (cf. [21]). Generally, Rassias [21] proved the following result:
Theorem 1.2 For any integer n > 1, there exists an integer nm such that
N > nm implies that there exists a map T: Rn —> Rm which satisfies the
condition (DOPP) but is not an isometry.
Aleksandrov’s problem has been solved for Euclidean spaces X = Y = Rn.

If 2 < n < oo, T must be an isometry due to the theorem of Beckman and
Quarles [3], Bishop [5] and in a special case Zvengrovski [18, Appendix to
Chapter II] independently:
Theorem 1.3 Let T be a transformation (possibly many-valued) o/Rn (2 <

n < oo) into itself. Let d(p,q) be the distance between points p and q of
R", and let T(p),T(q) be any images ofp and q, respectively. If there is a
length a > 0 such that d(T(p),T(q)) = a whenever d(p, q) = a, then T is a
Euclidean transformation ofW1 onto itself. That is, T is a linear isometry up
to translation.
For X = Y — R°° the Hilbert space of infinite sequences of real numbers

x = (ari, X2, ■ ■.) satisfying x2 < oo, an example of a unit distance
preserving mapping that is not an isometry has been given by Beckman and
Quarles [3]: There is in R°° a denumerable and everywhere dense set of points,
which will be denoted by {yk}^=i. Define a mapping g: X —> such
that the distance d(x,g(xf) < 1/2. Define h: {yk}^Ll —> X by h,(yk) =
ak, where ak — (ak, ak,...) and ak = 8jk/V2 (fijk is the Kronecker delta).
Now let T: X —> X be the mapping T = h o g. It may be readily seen
that T is a transformation of R°° into itself which preserves the unit distance.
For, if d(xx,x2) = 1, then g(xx) 7^ p(x2),7’(x1) T(x2), and therefore
d(T(x1), T(x2)) = 1. However T is not an isometry, for if x1 and x2 are any
two points in X, then d(T(xi),T(x2)) is either 0 or 1. Obviously, from the
above definition of T, the mapping T need not be one-to-one.
A geometric interpretation is that for a given mapping f: Rn —> Rm and
for arbitrary three points p\, P2, P3 in Rn forming an equilateral triangle with
unit length, if /(pi), f(pf) and /(ps) also form an equilateral triangle with
unit length and 1 < m = n < 00,
Figure 1.2.
then arbitrary geometric figures are preserved by f and f must be a linear

isometry up to translation. If m = n = 1 or m = n = oo or m n, for
arbitrary three points pi,P2, P3 in Rn forming an equilateral triangle with unit
length, if /(pi), f(pz) and /(pa) also form an equilateral triangle with unit
length, then f need not be an isometry.
Clearly, the mappings defined in the above counterexamples were not con
tinuous or surjective. What does happen if we impose conditions on T of
continuity and/or surjectivity? The following is the Rassias problem:
Rassias Problem. Let X and Y be normed linear spaces, and

T: X —> Y be a continuous and/or surjective mapping satisfying
(DOPP). Is T necessarily an isometry?
Example 1.2 Let T: (R2, || ■ ||oo) —> (R2, || • ||oo) be a mapping defined by
T(xi, X2) = ([xj] + — [ti])2, X2). It is easy to verify that T is a continuous
surjective mapping satisfying (SDOPP), but T is not an isometry.
Wang in [42] gave some simple examples to show that a distance one pre
serving mapping may be discontinuous or non-injective or non-surjective. Let
</>: R —> R be any periodic function with period 1, for example, ^>(ar) =
asm2irx. Then f: R —> R defined by f(x) = x + </>(x) preserves (DOPP)
but is not an isometry. For an arbitrary normed real vector space E, Wang
[42] constructed a new normed vector space F = 1?QR with the norm
wise ||(x, t)||oo = max{||x||, |/|} and defined a mapping T: F —> F by
Tu = (x,f (t)), Vu = (x,t) e F. If a satisfies that 0 < |a| < 1/2%, then T
is continuous, surjective and satisfies (SDOPP), but T is not an isometry (cf.
[42]).
One positive answer to Rassias problem was provided by Mielnik and Ras
sias [17] as follows
Theorem 1. 4 Consider the real classical Hilbert space denoted by H (H =

Rn, 3 < n < 00), and letT be a homeomorphism of H onto H which preserves
the distance r > 0. Then T is an isometry.
Generally, for real normed vector spaces, Rassias and Semrl [37] proved the
following results based on Theorem 1.1:
Theorem 1. 5 Let X and Y be real normed vector spaces such that one of
them has dimension greater than one. Suppose that T: X —» Y is a Lipschitz
mapping with k = 1:
||T(x) - T(y)|| < ||x - y|| forallx,yeX.
Assume also that T is a surjective mapping satisfying (SDOPP). Then T is a

linear isometry up to translation.
From Theorem 1.5 they got the following corollaries:
Corollary 1.1 Let X and Y be real normed vector spaces, dimX > 1, such
that one of them is strictly convex. Suppose that T : X —» Y is a homeomor
phism satisfying (DOPP). Then T is a linear isometry up to translation.
Corollary 1.2 Let X and Y be real normed vector spaces such that one of
them has dimension greater than one. Assume also that one of them is strictly
convex. Suppose that T : X —» V is a surjective mapping satisfying (SDOPP).
Then T is a linear isometry up to translation.
Based on the following result (cf. [39]), the condition “one of them has
dimension greater than one” in Theorem 1.5 can be omitted.
Theorem 1. 6 Suppose that T: R —> R is a Lipschitz mapping with k = 1:
||T(z) - T(y) || < ||x - y || for all x,y G R.
Assume that T is a mapping satisfying (DOPP). Then T is a linear isometry

up to translation.
Proof: Without loss of generality, assume /(0) = 0. Otherwise, we substitute

f(x) — /(0) for f(x) for all x e R. Since f satisfies (DOPP), then |/(1) —
7(0) | = |1 - 0| = 1. Thus 7(1) = 1 or /(l) = -1.
i) In case 7(1) = 1, firstly, we will prove by induction that f(n) — n for
all nonnegative integers.
Suppose 7(m) = m f°r any integer m with 0 < m < n, where n is a
positive integer greater than one. Since |7(n) — 7(n — 1)1 — |7(n) —
(n — 1)| - 1, then f(n) = n — 2 or f(n) = n. Assume f(n) — n — 2.
Let r = By the contractive property of 7, we obtain
|>|7(r)-7(«-2)|
>|/(n-l)-/(n-2)|-|/(r)-/(n-l)|
>1 —|r —(n —1)|
- 1
~ 2
Therefore |/(r) - /(n — 2)| = 1/2. Applying the same argument, it

follows that |/(r) — f(n — 1)| = 1/2. Since /(n — 2) = n — 2 and
f{n — 1) = n — 1, then
/(-)=
Set s = (n~i)+n. Similarly, we obtain
I/O) - f(n - 1)| = I/O - /(„)| = ~
and
Thus |/(r) — f (s) | = 0 and |r — s| = 1, which contradicts the fact that

f satisfies (DOPP). Hence /(n) = n for any nonnegative integer n.
Similarly, /(—l) must take the value —1. Otherwise /(—l) = 1 and
/(—1/2) = /(1/2) = 1/2, which contradicts the fact that f satisfies
(DOPP). By induction, the proof follows for /(n) = n for any negative
integer n exactly for the case of nonnegative integers we proved before.
For any x G R, there exists an integer no such that x G [no, no+1]. Since
/: R —> R is a Lipschitz mapping with k = 1, by the same argument as
above, it is true that
|/(ar)-/(n0)| = |x-n0| and |/(x)-/(n0 + l)| = |ar-n0-l|.
Therefore together with the fact that /(n) = n for any integer n, f must
satisfy the property f(x) = x for all x G R, and thus f is a linear
isometry.
ii) We will examine the case /(I) = —1.

Let g(x) = —f(x) for all x G R. Then g satisfies (DOPP) and it is
also a Lipschitz mapping with k = 1. By i), g must be a linear isometry
and ^(x) = x for all x G R. Therefore /(ar) is a linear isometry and
/(x) = —x for all x G R. □
Since any real normed vector space with dimension one is linearly isometric
to R, by Theorems 1.5 and 1.6, the following is true:
Corollary 1.3 Let X and Y be real normed vector spaces. Suppose that
T: X —>Y is a Lipschitz mapping with k = 1:
||T(x) - T(y) || < ||ar - y|| for all x,y eX.
Assume also that T is a surjective mapping satisfying (SDOPP). Then T is a

Dolinar [8] introduced the concept of (e,p)-isometries. T : X —> Y is called

an (e,p)-isometry if T satisfies
| ||Tx - Ty\\ - ||x - 3/|| | < ||x - y||p for all x, y e X.
If p > 1, then T must be a Lipschitz mapping with k = 1. In fact, for all x, y

in X with ||z — y\\ 0 and any positive integer n, let xq = x, xn = y and
Xk = x + — x), k = 1,2,..., n, then
||TTfc+1 -Txfe|| < ||xfc+i-xfc|| + ||xfc+1-Tfc||p < -||x-y|| + L||x_y||P

n nP
and
l|Tx - Ty|| < £ ||Txt+1 - 7M < ||z - i,|| + Aj-lk - Jzll”-

fc=O
Let n —> oo, then HTx — Ty\\ < ||rr — y|| (cf. [8]).
Corollary 1.4 Let X and Y be real normed vector spaces such that Y is
strictly convex. Suppose that T: X —> Y is an (s, p)-isometry for p > 1
satisfying (DOPP). Then T is a linear isometry up to translation.
Rassias’ problem is still open for the case that T: R°° —> R°° is a continu
ous mapping satisfying (DOPP).
The Aleksandrov problem was intensively considered by Ciesielski and Ras
sias in [7] for non-standard metrics. In Rn three classical metrics induce
the same topology: d%(x,y) = £”=1 |zi - 3/i|» = max{|xi - :
i = 1,2,..., n} and the Euclidean metric d#, where x — (ari, x%,..., xn),
y = (yi,y2,---,yn)-
Ciesielski and Rassias [7] gave some interesting examples about the Alek
sandrov problem. For (Rn,dm), n > 1, a mapping satisfying (DOPP) need
not be an isometry. For this it is enough to consider the mapping f: Rn —> Rn
defined by /(xi, x^,..., xn) = ([xi], [x^],, [xn]), where [or] denotes the
integer part of x.
For (R2, c?s), a mapping satisfying (DOPP) need not be an isometry either.
For this it is enough to consider the mapping g: (R2, dg) —> (R2, dg) defined
by 5 — (x/2 • ofo (^2 ■ R~/4), where /(xb x2) = (ki], [^2]) and Rn/4
is the rotation:
(x, y) -> ((x + y)/V2,(y- x)/V2) .
The mapping g satisfies (DOPP) but is not an isometry. Furthermore, in [7]

Ciesielski and Rassias proved the following result.
Theorem 1.7 There is no mapping f: (R2,dm) —> (R2,de) satisfying the

condition (DOPP).
The following problems posed by Ciesielski and Rassias (cf. [7]) are still
open:
Problem 1.1 Determine the smallest value ofk(rt)for Euclidean spaces, de
noted by sk(n), such that there exists a (DOPP)-mapping
T:Rn->Wk^
which is not an isometry.
Problem 1.2 Does the condition (DOPP) suffice for a mapping T: Rn —> Rm,
2 < n < m < oo, to be an isometry with respect to these metrics? Especially,
in case n = m, examine whether a mapping
T: (RVs)-+(RVs)
satisfying condition (DOPP) is an isometry for n = 3,4,....
2 . The Aleksandrov-Rassias problem in Hilbert

spaces
What does happen if we require, instead of one conservative distance for
a mapping between normed vector spaces, two conservative distances? An
answer in a generalized form to this question was given by Benz and Berens
[4] who proved the following theorem and pointed out that the condition that
Y be strictly convex cannot be relaxed.
Theorem 2.1 Let X and Y be real normed vector spaces. Assume that dim X >
2 and Y is strictly convex. Suppose T : X —» Y satisfies the property that for
some integer A > 2 andfor all xyy e X with Ik-y|| = p. I|r(x)-T(y)|| < p;
and for all x,y e X with ||x — y\\ — Xp, ||T(x) — T(y) || > Xp. Then T is a
Example 2.1 ([4]) Let p > 0 be a fixed real number. Consider the function
V’: R —> R defined by
x ^(x) = p{[x/p] - (x/p - [x/p])2},
where [a;] denotes the integer part of x. Set X — Y = R2 endowed with the
maximum norm, which is not strictly convex. For each x e R2, x = (a?i, xf),
||rr||oo = max(|rri|, |a?2|) and define /: R2 —> R2 by f(x) = V’(:r2))-
For each x, y e R2 and for each positive integer N, ||a? — y ||oo = Np implies
||/(a;) — /(p)|| = Np, but / is not an isometry.
It is easy to verify that the condition in Theorem 2.1 is equivalent to T

preserving the distances p and Xp.
Aleksandrov-Rassias Problem. If T preserves two distances

with a noninteger ratio, and X and Y are real normed vector
spaces such that Y is strictly convex and dim X > 2, whether or
not T must be an isometry (cf. [23]).
The Aleksandrov-Rassias problem has been considered for Hilbert spaces.

If T: X —> Y preserves two distances with a noninteger ratio, and X and Y
are Hilbert spaces, the following results hold [43]:
Theorem 2.2 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
that T: X —> Y preserves the distances 1, y/3. Then T is a linear isometry up
to translation.
Proof:
i) First, we shall prove that supposing Pi, P2, P3, P4 in X form a rhombus of
unit side with ||pi -p31| = \/3, ||p2-P41| = 1, then T(pi), T(p2), T(p3),
T(pf)\xYY also formarhombus of unit side with ||71(pi)—71(p3)|| = y/3
and ||T(p2)-T(p4) || = 1.
P4
P2
Figure 2.1.
Set x = T(p2) -T(pi), y = T(pi) -T(pi) and z = T(p3) -T(pi). Then

Ikll = Ikll = Ik - yll = Ik - ^11 = Ik - yll =1 and Ikll = >/3. since y is
a real Hilbert space, then
(x - y, x - y) = (x, x) - 2(x, y) + (y, y) = 1,
(z — x, z — x) = (z, z) — 2(z, x) + (x, x) = 1
and
(2 - y, z - y) = (z, z) - 2(z, y) + (y, y) = 1.
Hence
1 3
(or,y) = - and (z,x) = (z,y) =-,
and
(z—x—y,z—x—y) = (z, z)-2(z, x)-2(z, y)+(x, x)+2(x, y)+(y, y) = 0.
So z = x + y. That is to say, 71(pi),T(p2),T’(p3),71(p4) form a rhombus

of unit side with ||T(p3) — T(pi)|| = a/3 and ||T(p4) — T(p2)|| = 1. Espe
cially, T(p3) is in the span generated by T'(pi),T’(p2),T'(p4), 7'(p3)—T(pi) =
(T(p2)-T(pi)) + (T(p4)-T(pi)).
ii) Let p, q e X with ||p — <?|| - 2 and pi - Since dimX > 2, we
can select P2,p3 in X such that ppip2p3 and P1P3P2? form two rhombi
of unit side with ||p2 — p|| = ||p3 — <?|| = as follows:
P3 P2
Pl
Figure 2.3.
By step i), T’(p)T'(pi)7,(p2)71(p3) and T^T^T^T^q) also form

two rhombi of unit side with ||T(p2) — ^(p) || = II^Xps) — ?Xq)|| = \/3.
Seta: = T(pi)-T(p) andy = T(p3)-T(p). Then re-y = T(pi)-T(p3),
T(p2)-T(p) = x+y andT(y)-T(p3) - (T(p1)-T(p3)) + (T(p2)-T(p3)).
Hence T(q) - T(p) = 2x = 2(T(pi) - T(p)) and ||T(q) - T(p)|| = 2. So T
preserves distance 2.
By Theorem 2.1, T is a linear isometry up to translation. □
Figure 2.4-
A geometric interpretation is that for f: X —> Y, X and Y being Hilbert

spaces, and for arbitrary four points Pi,P2,P3,P4 in X forming a rhombus
of unit side with ||pi - p3|| = ||p2 - pJl = 1: if /(pi), /(P2), /(ps)
and f(pi) also form a rhombus of unit side with ||/(pi) — f(ps)|| = a/3,
||/(P2) “ f(pi)II = 1, ên f must be a linear isometry up to translation.
Or otherwise for arbitrary three points pi,P2,P3 in X forming a triangle with
IIpi — P31| = >/3, IIpi — P21| = IIP2 - P3|| = 1: if /(pi), /(P2) and /(p3) also
form a triangle with ||/(pi) - /(p3)|| = V%, ||/(pi) - /(P2)|| = ||/(P2> -
/(p3)|| = 1, then f is an isometry.
Figure 2.5.
Remark 2.1 In Theorem 2.2, the condition on X cannot be relaxed. For ex
ample, let T: R —> R be defined by T(x) — x + <f>{x) where
0, x € A,
= for all x € R,
1, x A,
where A = {a + b>/3 : a,b 6 R are rational numbers}. It is easy to verify that

T preserves the distances 1 and \/3- However, T is not an isometry.
Corollary 2.1 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
that T: X —> Y preserves the two distances 1 and n^/3 for some positive
integer n. Then T is a linear isometry up to translation.
Proof:
i) If n = 1, by Theorem 2.2, it is true.
ii.l) Letp, q € X satisfy ||p — q|| = -\/3. Since dim A" > 2, we can construct
a rhombus ppi<?p2 with ||p2 — pi|| = 1.
Let x = T(pi) — T(p), y = T(pf) — T(p) and z = T(q) — T(p). Then
Ikll = IMI = Ik—3/11 = = Ik—3/11 = 1 and ||x+y|| = a/3.
Hence
(a, y) = - and (z, x) = (z, y) = -(z, z),

& Li
and by the Cauchy-Schwarz inequality,
kll2 = (z,x + y) < Hzllllz + y|| = a/3||z||.
The above inequality turns out to be an equality if and only if there exists
a nonnegative constant a such that z = a(x + V) and ||z|| = ||T(g) -
T(P)|| = V^
ii.2) Suppose that T preserves the distances 1 and m/3 for some positive
integer n > 2. Letp, <71 e X satisfy ||p — Qi || = Set
Qk =P + k(Ql ~P), fc = 2,...,n,
then ||gfc+1 - gfc|| = ||gi - p|| = a/3, k = 1,2,..., n - 1, and ||g - n -

p|| = n\/3. Since dimX > 2, we can construct n rhombi of unit side as
follows:
Figure 2.6.
By step ii.l), we can get
||T(91)-T(p)|| < A ||T(g2)-T(gi)|| < Vs,--- ,||T(Q„)-T(Qn_1)|| < Vs,
and
n
||T(9n)-T(p)|| < l|r(Q1)-T(p)|| + £||T(%)-T(9fc_1)|| < nVs.
k=2
Since Y is an Hilbert space and T preserves distance nVS, then
\\T(q)-T(p)\\ = ||Tte)-T(Q)|| = ••• = ||T(<?„)-r(<?n_1)|| = Vs.
Hence the distance VS is also preserved by T. By Theorem 2.2, T is a

linear isometry up to translation. □
Definition 2.1 Suppose f: X —> Y is a mapping. The distance r is called

contractive by f if and only if for all x, y e X with ||a; — y|| = r it follows
that || /(t) — f(y) || < r. The distance r is called extensive by f if and only if
for all x, y 6 X with ||rr — y|| = r it follows that ||/(rr) — f (y)|| > r.
It is obvious by the triangle inequality that if f: X —> Y preserves the

distance r, then the distance nr is contractive by f for n = 1,2,....
The following results are obtained in [40].
Theorem 2.3 Suppose that ft X —> Y satisfies (DOPP) and that the dis
tances a, b are contractive by f, where a and b are positive numbers with
|a — b| < 1. Then the distance Via2 + 2b2 — 1 is contractive by f. Especially,
if the distance \/2a2 + 2b2 — 1 is extensive by f, then the distances a, b and
Vâ2 + 2b2 — 1 are all preserved by f.
PROOF: Suppose thatp, q E X with ||p— g|| = V2a2 + 2b2 — 1. We will prove
that ||/(p) — f(q)|| < V2a2 + 2b2 — 1. Since the dimension of X is greater
than one, we can select pi and p2 in X and construct a parallelogram with
IIpi-p|| = ||P2-?|| = a, ||p2-p|| = ||q-Pill = b, ||Q-p|| = V2a2 + 2b2 - 1,
||P2 - Pill = 1:
P2 a q
P a Pl
Figure 2.7.
Set x = f(pf) - f(p), y = f(p2) - /(p), z = /(/(?) - /(pi), u =

/(g) ~ /(P2)» v = /(p2) - /(pi) and w = /(g) - /(p). Then v = y - x,
u = w — y and z = w — x. Since / satisfies (DOPP) and the distances a and b
are contractive by /, then ||z|| < a, ||u|| < a, ||y|| < b, ||z|| 2(w,x + y) - 4(x,y). (2.1)
Hence
(w, w) > 2(w, x + y) — 4(x, y) — 1

= 1 + 2(w, x + y) - 2(x — y,x - y) - 4(x, y)
= 1 + 2(w,x + y) - 2(x,x) - 2(y,y). (2.2)
Therefore
(w, w) < 2(w, w) + 2(x, x) + 2(y, y) — 2(w, x + y) — 1

= (x, x) + (y, y) + (w - x, w - x) + (w - y, w - y) - 1
= (®, x) + (y, y) + (z, z) + (u, u) - 1
< 2a2 + 2b2 - 1. (2.3)
Hence, the distance V2a2 + 2b2 — 1 is contractive by /.

According to (2.3), if /: X —> Y satisfies (DOPP), the distances a and b
are contractive by / and the distance yj2c? 4- 2b2 — 1 is extensive by /, then
the distances a, b and \/2a2 + 2b2 — 1 are preserved by /. □
Remark 2.2 For the special case in Theorem 2.3, when |a — b| = 1, the map
ping / must be a linear isometry up to translation due to Theorem 3.1.
Corollary 2.2 Suppose that f: X —> Y satisfies (DOPP) and that the dis
tance a is contractive by f, where a is a positive number. Then the distance
y/4a2 — 1 is contractive by f. Especially, if the distance V4a2 — 1 is extensive
by f, then the distances a and \/4a2 — 1 are preserved by f.
Suppose that /: X —> Y satisfies (DOPP). Then by the triangle inequality,

every positive integer k is contractive by /. By Corollary 2.2, the distances
\/4fc2 — 1, ^/4(4fc2 — 1) — 1,..., yj4mk'2 — are contractive by / where
k — 1,2,..., m = 1,2,....
Together with Theorem 2.1, Corollary 2.2 and Theorem 1.3, the following
result in [40] has been obtained by Rassias and Xiang.
Theorem 2.4 Suppose that T: X —> Y satisfies (DOPP). Assume the distance
nyj4mkf — is preserved by T for some positive integers n, k and m.
Then T must be a linear isometry up to translation.
Proof:
1) In case the distance yjA^k^ — 4m3~1 is extensive by f for some positive
integers k and m: by induction on m and Corollary 2.2, the distances
a/4/c2 — 1 and k are preserved by f. If k > 2, by Theorem 2.1 it follows
that the mapping f is a linear isometry up to translation; if k = 1, then
J3 is preserved by f. By Theorem 1.3, f is a linear isometry up to
translation.
2) In case n > 2, for any p,qi E X with ||p — <?i || = yj^k2 — ^g"1, set
Qj = P + j(p-Qi), J = 2,3,...,n.
Then ||gj+i-Qj|| = ||?i -p|| = yj4mk2 - ^^forj = 1,2,... ,n-l,
and ||qn — p|| = n-^/4mfc2—3~ k Since f satisfies (DOPP) and

yj^k2 — 4"* 3 1 is contractive by f, it follows that
/ 4m — 1
ii/(?i)-/(p)II< iM2-——,
V o
/ Am _ 1
ll/fe+i) - /(?j)|| < t/4mfc2------- - —, j = 1,2,..., n — 1,
V o
and
n— 1
l|/(<?n) - /(p)|| < ||/(91) - /(P)|| + £ ||/(9j+l) - /(Qj)||
J=1
/ 4m — 1
< ny/4mfc2—
3
Since nJA"1 k? — 4m3 1 is extensive by /, we have
ll/(«i) - /(rill = ll/te) - /(®)ll = ■■■

/ — 1
- Il/(9n) - /(9n-l)|| = 1/4-fc2-------- —
V
Hence the distance yj4mk2 — 4"*3 1 is also preserved by /. By step 1),

/ is a linear isometry up to translation. □
However, for arbitrary three points pi, P2, ps in X forming an isosceles

triangle with the ratio between the base with unit length and the height of the
triangle or the ratio between the oblique line with unit length and the height
of the triangle being an even integer, if /(pi), /(P2) and /(ps) also make an
isosceles triangle with the same size, then f must be a linear isometry up to
translation, due to the following theorem of Rassias and Xiang (cf. [39]).
P3 P3
Figure 2.8.
f: X —> y satisfies the property that
i) f preserves the distances a, b with the ratio of a and b being 1 for

some positive integer n; or
ii) f preserves the distances a, b with the ratio of a and b being for
some positive integer n with n > 1.
Then f is a linear isometry up to translation.
Proof:
i) Without loss of generality, suppose that f preserves the two distances

\/4n2 — 1 and n. For the case n = 1, the property follows from Theorem
2.2.
For n > 1, let P2, P4 in X with ||p4 — P2II = 1- Since dimX > 2,
then one can select pi, P3 in X such that pi,P2,P3,P4 in X form a
parallelogram with ||p2~Pi || = ||P3~P4II = ||P3~P2II = ||P4~Pi|| = n,
IIpi - P31| = V^n2 - 1, ||p2 - p4|| = 1.
Because of the fact that f preserves the distances x/4n2 — 1 and n, then
ll/(P2) - /(pi)|| = ||/(p3) - /(p2)|| = ||/(p4) - /(p3)|| = ||/(P4) -
/(pi)ll = and ||/(p3)-/(pi)|| = \/4n2 - 1. Lets = /(pi)-/(p2),
y = /(P3) ~ /(P2) and z = f(p4) - /(p2). Then
Ikll = Ikll = Ik - ®ll = Ik - pH = Ik - pH = vkn2 -1
and
(t + y, x + y) = 1.
Hence (z, x) = {z,y} = ^(z,z). If we use the Cauchy-Schwarz in-

equality, we obtain that
Ikll2 = (z,x + y) < Iklllk + 3/11 = Ikll-
Thus ||/(p4) — /(p2)|| < 1 and therefore, by Theorem 2.1, f is a linear

isometry up to translation.
ii) In a similar way, if f preserves the distances a, b with the ratio of a and
b being ^2n+1 f°r some positive integer n with n > 1, it follows that f
must be a linear isometry up to translation. □
From Beckman and Quarles [3] it follows that if f: R" —> Rn satisfies
(DOPP) for some positive integer n with n > 2, then the distance
is also preserved by f. Generally, if f: X Y satisfies (DOPP), where X
and Y are real Hilbert spaces with dim X > n and n > 2, let p, q in X with
Up - ?ll = then ll/(p) - /(?)ll < In fact, suppose p,
Pi,... ,Pn and q, pi,... ,pn form two equilateral n-simplices with unit side
and ||p - q|| = Let z = f(p) - f(q), Xi = f(p) - f(pi) and
Vi = f(Pi) ~ /(?) fori = 1,2,...,n. Then ||rci|| = ||pi|| = ||xi - xj|| = 1
(i / j) and z = Xi + y^ Thus by (2 - xiy z - Zj) = (z - yi, z - y,) = 1, we

can get
n
2 / 2
(2, 2) = 2(2, Xi) = - I 2,
n \
Xi
n
Xi ini <
\ 2=1 2=1
Therefore ||/(p) - /(?)|| < Hence by Theorem 2.1, we obtain the

following (cf. [39])
Theorem 2.6 Let X and Y be real Hilbert spaces with dim X > n and n> 2.
Suppose f: X —» Y satisfies (DOPP) and preserves the distance
for some positive integer k with k > 2. Then f is a linear isometry up to
translation.
A generalized form of Theorem 2.2 for T preserving three distances was

given in [43] as follows:
T: X —> Y preserves the distances a, b and y/2o? + 62 for some positive
constants a and b satisfying 0 < a < 26. Then T is a linear isometry up to
translation.
Proof: In case 6 = 2a, by Theorem 2.1, T is a linear isometry up to transla

tion.
In case 0 < 6 < 2a, similarly to the proof in Corollary 2.1, i), firstly, it is
easy to verify that if pi, p2, P3, P4 in X form a parallelogram with ||p2 — Pill =
||P3 - P4|| = a, ||p3 — P21| = ||P4 - Pill = b, ||pi - p3|| = V2a^+b2,
||p2 — P4|| = b, then T(pi)T'(p2)T'(p3)T'(p4) is also a parallelogram with
\\T(P2) - T(pi)|| = ||T(p3) - T(p4)|| = a, ||T(p3) - T(p2)|| = ||T(p4) -
T(P1)|| = b, ||T(pi) - T(p3)|| = + ||T(p2) - T(p4)|| = b, and
T(p3) - T(pt) = (T(p2) - T(P1)) + (T(P4) - T(px)).
ii) Secondly, letp, q e X with ||p— y|| = 2a andpi = Ej2. Since dimX >
2, we can select p2,p3 in X such that p,Pi,p2,P3 and pi,P3,p2, q form two
parallelograms with ||pi - p|| = a, ||p2 - pi || = ||p3 - pi || = b.
By step i), T(p),71(pi),T’(p2),T'(p3) and T(pi), T(p3), T(p2), T(q) form

two parallelograms with side length a and b.
Seta; = T(pi)-T(p) andp = T(p3)-T(p). Thenx-y = T(pi)-T(p3),
r(p2)-T(p) = x+yandT(g)-T(p3) = (T(pi)-T(p3))4-(T(p2)-T(p3)).
Hence T(g) - T(p) = 2a; = 2(T(pi) - T(p)) and \\T(q) - T(p)|| = 2. So
T preserves the distance 2a. By Theorem 2.1, T is a linear isometry up to
translation. □
Similarly to the proof of Theorem 2.7, we have
T: X —» Y preserves the distances a, b (b < 2a) and \/a2 + 2b2 for some
positive constants a and b. Then T is a linear isometry up to translation.
Pi
P2 V2a2 + P3 + /(P3)
Figure 2.12.

spaces, and for arbitrary three points pi, p2, p3 in X forming a triangle with
||pi - P21| = ||P1 - P31| = b, ||p2 - p31| = V2a2 + b2 if /(pi), /(p2) and
f(p3) also form a triangle with the same size || f(pi) — f (p2) II = a, || f(pi') —
/(p3) || = b, ||/(p2) — ftps) || = \/2a2 + b2, then f must be an isometry.
Also let X and Y be real Hilbert spaces with dim X > 2, and let T: X —>
Y preserve the two distances a and b. If pi,p2,P3,P4 in X form a parallel
ogram with ||p2 - pill = ||p3 - P4|| = a, ||p3 - p2|| = ||p4 - pill -- b,
||p2 — P41| — b, in the same way as in the proof of Corollary 2.1, we can get
ijl^pa) ~ r(pi)|| < -/2a2 + b2. Since T preserves the distance n\/2a2 + b2,
then following an argument similar to the proof of Corollary 2.1, the distance
\/2a2 + b2 is preserved by T. So by Theorem 2.2, we have the following re

sults (cf. [43]):
T: X —» Y preserves the distances a, b and n\/2a2 + b2 for some positive
constants a and b satisfying 0 < a < 2b, and some positive integer n. Then T
is a linear isometry up to translation.
T: X —» Y preserves the distances a, b (b < 2a) and n\/a2 + 2b2 for some
positive constants a and b, and some positive integer n. Then T is a linear
thatT: X —> Y preserves the three distances 1, a (0 < a < 2) and ny4 — a2
for some nonnegative constant a and positive integer n greater or equal to 2.
Then T is a linear isometry up to translation.
Proof:
i) In case a = 0 or a — 2, by Theorem 2.1 it follows that T is a linear
ii.l) In case 0 < a < 2, first we will prove that supposing that p, q are in X
with \\p — q\\ = \/4 — a2, then ||T(p) — T(q)|| < i/4 — a2.
Let p, q be in X with ||p — q\\ = \/4 — a2. Then we can construct
a parallelogram ppiQp2 with ||pi - p|| = ||p2 - p|| = ||q — Pi || =
\\q — p21| = 1, and ||p2 — pi|| = a as follows:
Since T preserves distances 1 and a, then ||T(pi) — T(p) || = ||T(p2) —

T(p)|| = \\T(q) - T(P1)|| = \\T(q) - T(p2)|| = 1 and ||T(P2) -
T’Cpi)!! = a- Let x = T(pi) - T(p), y = T(p2) - T(p), z =
T(q) —T(p), then
(x - y, x - y) = {x, x) - 2(x, y) + (y, y) = a2,
(z — x, z — x) = (z, z) — 2(z, x) + (x, ir) = 1
and
k - y, z - y) = (2,2) - 2(2, y) + (y, y) = 1.
Hence
Ik + 3/|| = \/4-a2, (2,x) = (2,y) =
So by the Cauchy-Schwarz inequality,
Ikll2 = k,
)
* = (^,^)+(z,y) = (z,x+y) < ikiHk+yll = x/4-u2lkl|.
Hence
lirtaH t(p)|| = W <
ii.2) Secondly, suppose that T preserves the distances 1 and n\/4 — a2 for
some positive integer n > 2. Letp, qi G X satisfy ||p — <7i|| = \/4 — a2.
Set
qk =p + fc(p-Qi), k = 2,
then |kfc+i - qk\\ = ||gi - p|| = v/4 - a2, k = 1,2,..., n - 1 and

Ikn — p|| — n.\/4 — a2. Since dim X > 2, we can construct n rhombi
of unit side as follows:
By step ii.l), we can get
R<?i)-T(p)||<^4-a2,
||T(g2)-T(gi)||<y4-a2,
and
||T(yn) - T(p)|| < ||T(qi) - T(p)|| + £ \\T(qk) - T(qk^\\

k=2
< ny 4 — a2.
Since Y is a Hilbert space and T preserves the distance nV4 — a2, then
- T(p)|| = ||T(q2) - T(q)\\ = • • • = ||T(<m) - T(<?n_1)||

= \/4 — a2.
Hence the distance \/4 — a2 is also preserved by T. By Theorem 2.1, T

is a linear isometry up to translation. □
In Theorem 2.8, set a = 2sin# (0 < 0 < 7r/2) and let 2sin# = 2ncos#,
then tg # = n and a = n\/4 — a2 = • So the following result is true.
Corollary 2.6 Let X and Y be real Hilbert spaces with dim X > 2. Sup
pose that T: X —> Y preserves the two distances 1 and where n
is a positive integer greater or equal to 2. Then T is a linear isometry up to
translation.
Moreover, for dimX > 3 another case for T preserving two distances with
noninteger ratio was given in [43] as follows
that T: X —> Y preserves the two distances 1 and y/2. Then T is a linear
Proof:
i) First we shall prove that if pi, P2, P3, P4 in X form a square of unit side
with ||px - p3|| = ||p2 - p4|| = 'A as then also T(pi), T(p2), T(p3),
T(p4) inF form a square of unit side with || 71(pi)—T'(p3) || = ||T(p2) —
r(p4)|| =
Pi P3
Pl P2
Figure 2.15 .
In fact, set x = T(p2) - T(pi), y = T(p4) - T(pi) and z = T(p3) -

T(pi). Since T preserves distances 1 and \/2, then ||x|| = ||y|| = ||z —
x|| = ||z — 3/1| — 1 and ||^|| = ||x — y || = \/2. Since Y is a real Hilbert
space, then
{x - y, x - y) = (x, x) - 2(x, y) + (y, y) = 2,
(z — x, z — x) — (z, z) — 2(z, x) + (x, x) — 1
and
(z - y, z - y) = (z, z) - 2(z, y) + (y, y) = 1.
Hence
(x, y) = 0 and (z, x) = (z, y) = 1,
and
(z - x - y, z - x - y)
= (z>z) - 2(z>x) - 2(z> y) + (*
, x) + 2(a:> y) + (y, y) = °-
So z = x + y. That is to say, T(pi), T(p2), T(p^) form a square
of unit side with ||T(p3) - T’(pi)|| = ||T’(p4) - T(p2)|| = \/2. Espe
cially, T(p3) isinthespanofT'(pi),71(p2),T'(p4),andT'(p3)—T'(pi) =
(T(p2) - T(pi)) + (T(P4) - ^(pi)) (see Fig. 2.16).
Tipi)--------------- T(p3)
T(P1)I--------------- J T(p2)
Figure 2.16 .
ii) Let p, q € X with ||p — y|| = -\/3. Since dimX > 3, we can construct a
cube of unit side with ||p — y|| = y/3 as follows:
Figure 2.17.
PP1P2P3, P1P29P4, PP1P4P5, P2P3P6?, PP3P6P5 and p4P5P6? form squares

of unit side.
By step i),
T(p)T(p1)T(p2)T(p3), T(pi)T(p2)T(g)T(p4),
T(p)T(p1)T(p4)T(p5), T(p2)T(p3)T(p6)T(Q),
T(p)T(p3)T(p6)T(p5), T(p4)T(p5)T(p6)T(g)
also form squares of unit side, respectively, and Tips') and T(p§) are
in the spans of T(p), T(pi), T'(p4) and by TfpYp Tips), T(q), respec
tively. Hence T(p), T(pi), T(p2), T(p3), T(p4), T(p5), T(p$), T(q)
form a cube of unit side. It is easy to verify that ||T(p) — T(q) || = a/3-
So T preserves the distance a/3- By Theorem 2.2, T is a linear isometry
up to translation. □

spaces, and for arbitrary three points pi, p2, p3 in X forming an isosceles
triangle with ||pi-p2|| = V2, ||pi-p3|| = 1 = ||P2~P3|| = 1, if/(pi),/(P2)
and f (p3) also form an isosceles triangle with the same size 11 / (pi) — / (P2) 11 =
V2, ||/(pi) - /(p3)|| = Il/(P2) - /(P3>|| = 1. then f must be an isometry.
3. The Aleksandrov-Rassias problem in normed

vector spaces
If f preserves two distances with a noninteger ratio, and X and Y are real
normed vector spaces such that Y is strictly convex and dim Y > 2, it is an
open problem whether or not f must be an isometry (Rassias [23]). However,
if f preserves three distances, one can prove the following (cf. [39]).
Theorem 3.1 Let X and Y be real normed vector spaces. Assume that dim X >
2 and that Y is strictly convex. Suppose f: X —> F satisfies the property that
f preserves the three distances 1, a and 1 + a, where a is any positive constant.
Proof:
i) Let x, y € X with ||rr — y|| = 2 + a. Set
1 , x 1 + a, x
a:i =x + ——(y-a;), x2 = x + —— (y - x).
2 +a 2+a
Then
||xi - x|| = 1, ||zi - Z2II = G,

||y - xi|| = 1 + a, ||x2 - x|| = 1 + a, ||y-x2|| = l.
It follows that
||/(x1)-/(x)|| = l,
- /(®2)|| = a, \\f(y) - /(®1)II = 1 + a,
ll/M - /(z)|| = 1 + a, *
Wfiy) - /(
2)|| = 1,
and
ll/(^) - f(x)\\ = || + 11/^2) - /(^l) II = 1 + a,

*
i)
Wfiy) - /G II = II/O2) - /Oi)|| + 11/(1/) - /(
*
2)|| = 1 + a.
Since Y is strictly convex, then
*
/(
1) = f (x) + - fix)),
fix2) = fix.) + - fix,)).

1+a
Hence
fix) = ^-^fix.)--fix2) and /(y) = i^/(x2)-i/(xi).
CLCL cl cl
Thus ||/(2/) — fix) || = 2 + a for all x,y e X with ||x - y|| — 2 + a.

Therefore f also preserves the distance 2 + a.
ii) Let x, y € X with ||x — y|| = 2 + 2a. Set
1+a 2+a . .
II = a:+2T^<!'_l)’ X2 = X+2T2^y~x}-
Then
||xi - x|| = 1 + a, ||xi - x2|| = 1, ||y - X1|| = 1 + a,

||x2 - x|| - 2 + a, ||y - x21| = a.
Since f preserves distances 1, a, 1 + a and 2 + a, in a similar way, we

obtain that
ll/toWWII=2 + 2o.
Hence the mapping f preserves the distance 2 + 2a.
By i), ii) and Theorem 2.1,/ must be an isometry up to translation. □
Remark 3.1 In Theorem 3.1, the result is obviously true if the distances 1, a
and 1 + a are replaced by a, b and a + 6, where a and b are positive constants.
In particular, if the ratio of the two positive constants a and b is an integer, then
/ is a linear isometry up to translation due to a theorem of Benz and Berens
[4].
Corollary 3.1 Let X and Y be real normed vector spaces and assume that
dim X > 2 and Y is strictly convex. Suppose T: X —» V satisfies the property
that T preserves the three distances 1, A and (A), where A > 1 is any positive
constant, (A) = A — [A] and [A] is the greatest integer less than or equal to A.
Proof:
i) If 1 < A < 2, then A = 1 + (A) and by Theorem 3.1 T is a linear

ii) In case A > 2, suppose [A] = n > 2. Let x,y e X with ||rc — y\\ = k.
Set
Az x k \
z = x H—\y — X), Xk = x H—(y — x), k = 1,2,... , n.
n n
Then ||xi - x|| = ||xn_i - y|| = ||xfc - xk^ || = 1 for k = 2,..., n,
||z — z|| = A and ||z — y|| = (A). Since T preserves the distances 1, A
and (A), then
A = ||T(z) — T(z)\\
<\\T^-T^\\
n—1
+ £ \\T(xk+1) - T(xk)\\ + ||T(y) - T(z)\\
k=l
<[A] + (A).
Since Y is strictly convex, then T{xk) — T(x) + — T(x)),

k = 1,2,..., n and ||T(ar) — T(y) || = n.
Hence T also preserves the distance [A]. By Theorem 3.1, T is a linear

isometry up to translation. □
Theorem 3.2 Let X and Y be real normed vector spaces. Assume that Y
is strictly convex. Suppose f: X —» Y satisfies (DOPP) and is a Lipschitz
mapping with k = 1, i.e.
||/k) - /(y)ll < Ik-2/11 forallx,y e X.
PROOF: Let x, y G X with Ik — y|| — 1/2. Set z = x + 2(y — x). Then
lk-2|| = i, 11^ — 3/11 = 1,
and
Ik - 3/11 > ll/k) - /(3/)II > \\f^ - /k)|| - ||/k) - /(y)||

>1-11^-3/11
1
- 2'
Hence \\f(x) - /(y)|| = 1/2. Similarly \\f(z) - /(y)|| = 1/2 and
||/k) - /k)|| = ||/k) - /(3/)|| + 11/(3/) - /k)ll = 1-
Because Y is strictly convex, then
/(») =/w 2f(z) md ll/fo)-/«ll = l
Hence / preserves distances 1 and 1/2. Applying a similar method as in the

proof of Theorem 3.1, the mapping / preserves distance n in both directions
for any positive integer n. Since f is a contractive mapping, it is easy to verify
that f is an isometry due to the proof of Theorem 2.1. Therefore f is a linear
isometry up to translation, due to a theorem of Baker [2]. □
Acknowledgments
I am grateful to Themistocles M. Rassias of the National Technical Univer
sity of Athens, Greece for making me acquainted with this topic and for his
very helpful suggestions and useful information.
This work was supported by Natural Science Foundation of Hunan: 02JJY2006.
References
[1] A.D. Aleksandrov: ‘Mappings of families of sets’, Soviet Math. Dokl. 11

(1970), 116-120.
[2] J.A. Baker: ‘Isometries in normed spaces’, Amer. Math. Monthly 78
(1971), 655-658.
[3] F.S. Beckman and D.A. Quarles: ‘On isometries of Euclidean spaces’,
[4] W. Benz and H. Berens: ‘A contribution to a theorem of Ulam-Mazur’,
Aequat. Math. 34 (1987), 61-63.
[5] B. Bishop: ‘Characterizing motions by unit distance invariance’, Math.
Mag. 46 (1973), 148-151.
[6] D.G. Bourgain: ‘Approximate isometries’, Bull. Amer. Math. Soc. 52
(1946), 704-714.
[7] K. Ciesielski and Th.M. Rassias: ‘On some properties of isometric map
pings’, Facta Univ. Ser. Math. Inform. 7 (1992), 107-115.
[8] G. Dolinar: ‘Generalized stability of isometries’, J. Math. Anal. Appl. 202
(2000), 39-56.
[9] J. Gevirtz: ‘Stability of isometries on Banach spaces’, Proc. Amer. Math.
Soc. 89 (1983), 633-636.
[10] P.M. Gruber: ‘Stability of isometries’, Trans. Amer. Math. Soc. 245
(1978), 263-277.
tions in Several Variables’, Birkhauser Verlag, 1998.
[12] D.H. Hyers, G. Isac and Th.M. Rassias: 'Topics in Nonlinear Analysis
and Applications’, World Scientific Publ. Co., 1997.
[13] D.H. Hyers and S.M. Mazur: ‘On approximate isometries’, Bull. Amer.
Math. Soc. 51 (1945), 288-292.
Math. 44 (1992), 125-153.
[15] J. Lindenstrauss and A. Szankowski: ‘Non linear perturbations of isome
tries’, Asterique 131 (1985), 357-371.
[16] S. Mazur and S. Ulam: ‘Sur les transformations isometriques d’espaces
vectoriels normes’, C. R. Acad. Sci. Paris 194 (1932), 946-948.
[17] B. Mielnik and Th.M. Rassias: ‘On the Aleksandrov problem of conser
vative distances’, Proc. Amer. Math. Soc. 116 (1992), 1115-1118.
[18] P.S. Modenov and A.S. Parkhomenko: ‘Geometric Transformations’, Vol.
1, Academic Press, New York, 1965.
[20] Th.M. Rassias: ‘Is a distance one preserving mapping between metric
spaces always an isometry?’, Amer. Math. Monthly 90 (1983), 200.
[21] Th.M. Rassias: ‘Some remarks on isometric mappings’, Facta Univ., Ser.
Math. Inform. 2 (1987), 49-52.
[22] Th.M. Rassias (Ed.): ‘Nonlinear Analysis’, World Scientific Publ. Co.,
1987.
[23] Th.M. Rassias: ‘On the stability of mappings’, Rendiconti del Seminario
Matematico e Fisico di Milano 58 (1988), 91-99.
[24] Th.M. Rassias: ‘The stability of linear mappings and some problems on
isometries’, pages 175-184 In: Proceedings of the International Confer
ence on Mathematical Analysis and its Applications, Kuwait 1985, Perg-
amon Press, Oxford, 1988.
[25] Th.M. Rassias (Ed.): ‘Topics in Mathematical Analysis. A volume dedi
cated to the memory ofA. L. Cauchy’, World Scientific Publishing Com
pany (Series in Pure Mathematics), 1989.
[26] Th.M. Rassias: ‘Mappings that preserve unit distance’, Indian J. Math.
32 (1990), 275-278.
[27] Th.M. Rassias: ‘Remarks and Problems’, Aequat. Math. 39 (1990), 304.
[28] Th.M. Rassias (Ed.): ‘Inner Product Spaces and Applications’, Pitman
Research Notes in Mathematics Series Vol. 376, Addison Wesley Long
man Ltd, 1997.
[29] Th.M. Rassias: ‘Properties of isometries and approximate isometries’,
In: Recent Progress in Inequalities,MIA, Kluwer Academic Publishers
(1998), 325-345.
[30] Th.M. Rassias: ‘Remarks and Problems’, Aequat. Math. 56 (1998), 304-
306.
[31] Th.M. Rassias (Ed.): ‘Approximation Theory and Applications’,
Hadronic Press, Inc., Florida, 1998.
[32] Th.M. Rassias: ‘On the Mazur-Ulam theorem and the Aleksandrov prob
lem of conservative distances’, Short Communications, Section 8 (Anal
ysis) Intemat. Congress of Mathematicians, Berlin (1998), 151.
[33] Th.M. Rassias: ‘Properties of isometric mappings’, J. Math. Anal. Appl.
235(1997), 108-121.
Ulam’, Acta. Appl. Math. 61 (2000), 23-130.
[35] Th.M. Rassias (Ed.): 'Functional Equations and Inequalities’, Kluwer
Academic Publishers, 2000.
[36] Th.M. Rassias: ‘On the A. D. Aleksandrov problem of conservative dis

tances and the Mazur-Ulam theorem’, Nonlinear Analysis-Theory, Meth
ods & Applications 47 (2001), 2597-2608.
[37] Th.M. Rassias and P. Semrl: ‘On the Mazur-Ulam theorem and the
Aleksandrov problem for unit distance preserving mapping’, Proc. Amer.
Math. Soc. 118 (1993), 919-925.
[38] Th.M. Rassias and C.S. Sharma: ‘Properties of isometries’, J. Natural
Geometry 3 (1993), 1-38.
[39] Th.M. Rassias and Shuhuang Xiang: ‘On mappings with conservative
distances and the Mazur-Ulam theorem’, Publications of the Faculty of
Electrical Engineering. Univ. Belgrade, Series: Mathematics 11 (2000),
1-8.
[40] Th.M. Rassias and Shuhuang Xiang: ‘On Mazur-Ulam theorem and map
pings which preserve distances’, Nonlinear Functional Analysis and Ap
plications 5 (2) (2000), 61-66.
[41] C.S. Sharma and D.F. Almeida: ‘A direct proof of Wigner’s theorem on
maps which preserve probabilities between pure states of Quantum sys
tems’, Ann. Phys. 197 (1990), 300-309.
[42] Risheng Wang: ‘On distance one preserving maps and isometric maps’,
Acta Scientiarum Naturalium Universitatis Nankaiensis 33 (2000), 85-
87.
[43] Shuhuang Xiang: ‘Mappings of conservative distances and Mazur-Ulam
theorem’, J. Math. Anal. Appl. 254 (2001), 262-274.
[44] Shuhuang Xiang: ‘The Aleksandrov problem and Rassias problem for
isometric mappings’, Nonlinear Functional Analysis and Applications 6
(1) (2001), 69-77.
[45] Shuhuang Xiang: ‘Aleksandrov problem for isometries’ (preprint).
Index
absolutely continuous, 163 (s,p)-isometry, 198

addition theorem, 186 Einstein’s special theory of relativity, 187
affine isometry, 192 entire function, 184
Aleksandrov problem, 191-193, 219-221 equilateral triangle, 195
Aleksandrov-Rassias problem, 191, 199, 200, Euclidean structure, 146
215 Euclidean transformation, 194
algebra homomorphism, 102 Euler formula, 105
algebraic manifold, 140 Euler midpoint formula, 114-116
approximately additive mapping, 39 Euler trapezodial formula, 114
approximation error estimates, 159 Euler two-point formula, 121, 123, 136
arithmetic-geometric mean of Gauss, 150 Euler-Maclaurin formula, 132, 134, 136
associativity, 177 Euler-Simpson 3/8 formula, 129, 130, 136
Euler-Simpson formula, 118, 119, 136
B-quadratic mapping, 3 evaluation, 173
Bsa -quadraticmapping, 7 everywhere dense set, 194
Banach algebra, 2, 91,93, 100-102, 180 exponential, 171
Banach module, 1, 93 exponential monomial, 172
bordism group, 141 exponential polynomial, 172
Borel measure, 160 extensive distance, 204
*-algebra, 93,100
C fiber bundle, 141
Cartan distribution, 141 flow, 139
Cauchy characteristic distribution, 141 fluid, 144
Cauchy functional equation, 183, 184 formal integrability, 140
Cauchy problem, 141 Fubini theorem, 161
Cauchy type functional equation, 74 functional Navier-Stokes equation, 139, 146
Cauchy-Pexider functional equation, 183
Cauchy-Schwarz inequality, 205, 212
Caucy operator, 11 Galilean affine fiber bundle, 140
CebySev inequality, 22, 31 Galilean space-time, 139, 140
character, 172 Gauss functional equation, 151
characteristic distribution, 140 Gauss type functional equation, 149
characteristic vector field, 146 Gauss’ functional equation, 150
commutative discrete hypergroup, 179 generalized A-equation, 17
compact manifold, 145 generalized Cauchy functional equation, 183,
conservative distance, 192 185
contractive distance, 204 geodesic completeness, 144
convolution, 167, 169 global laminar solution, 146
global solution, 142
A operator, 11, 16
distance one preserving property, 192 Haussdorff group, 168
DJS-hypergroup, 168 Hermitian hypergroup, 168, 170
Dragomir-Agarwal inequalities, 105 Hilbert space, 159
dual Euler-Simpson formula, 125, 126 Hyers-Rassias theorem, 40
223
Hyers-Ulam sequence, 65, 147 non-isothermal Navier-Stokes equation, 139

Hyers-Ulam-Rassias, 1 normed vector space, 192
Hyers-Ulam-Rassias stability, 9, 10, 12, 39,40,
64, 67, 68, 70,71,91,93, 101, 143 order jet-derivative space, 139
Hyers-Ulam-Rassias theorem, 2 orthogonal polynomials, 170
hyperbolic geometry, 187 Ostrowski inequality, 22, 30-32
hypergroup, 167, 168, 177
parallelogram, 209
identity, 168 Peano kernel, 22
incompressible fluid, 139 Pexider equation, 16, 40
inertial frame, 140 Pexider operator, 11
infinite sequence, 194 Pexiderized quadratic equation, 45
injective mapping, 192 Pexiderized quadratic functional equation, 41
inner product, 164 polynomial hypergroup, 170, 172, 174
integral bordism group, 141 Popoviciu functional equation, 91-93, 100
integral manifold, 141 positive compact operator, 159
integral mean, 150, 154 probability measure, 167, 170
integral transform, 161 projected flow, 145
internal energy, 140
invariant mean, 179 quadratic function, 40
involution, 168 quadratic operator, 11,15
isometric identity, 161,164 quadrature rule, 21
isometry, 162, 191, 198, 199, 203, 206-211,
215, 217-220 Rassias approach, 40
isotropic pressure field, 140 Rassias problem, 195, 198, 221
regular solution, 141
Jensen operator, 15 reproducing kernel, 159, 160
jet-derivative space, 139 Riemannian manifold, 143
Riemannian metric, 143
Kronecker delta, 186, 194 Riemannian structure, 143
Kronecker theorem, 73, 74 right translation operator, 168
ring homomorphism, 100
Lebesgue integral, 26
left translation operator, 169 sectional topology, 141
Legendre hypergroup, 171 Smale-Zhou formula, 159
Legendre polynomials, 171 smooth solution, 142
Lehmer means, 150 square integrable function, 160
linear isometry, 206 strictly convex normed space, 192
linearization coefficients, 170 strongly distance one preserving property, 192
linearization formula, 170 subadditivity, 75
Lipschitz mapping, 196-198 superstable, 139, 143
locally compact Haussdorff semigroup, 168
locally compact Haussdorff space, 167 Tchebycheff hypergroup, 175
temperature field, 140
Markov-Kakutani fixed point theorem, 179 thermal conductivity, 140
Mazur-Ulam theorem, 191, 220, 221 trapezoid inequality, 32
mean value theorem, 149 triangle inequality, 205
measure, 178 Trif theorem, 93
median principle, 21, 23, 25, 27, 28 tunnel effects, 145
Mercer kernel, 162
Michael topology, 168 Ulam stability, 139, 142, 144
Minkowski inequality, 89 Ulam-extended superstability, 144
moment function, 175
moment sequence, 175 viscosity, 140
Navier-Stokes equation, 139, 141, 143-145 Wigner theorem, 221

Rassias T. Functional Equations, Inequalities and Applications 2013

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Rassias T. Functional Equations, Inequalities and Applications 2013

Uploaded by

Copyright:

Available Formats

Functional Equations, Inequalities and Applications

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481 -6406-6 ISBN 978-94-017-0225-6 (eBook)

Printed on acid-free paper

All Rights Reserved

Functional equations, inequalities and applications provides an extensive

Abstract We extend the Hyers-Ulam-Rassias stability of a quadratic functional equation

2000 MSC: 39B72, 39B32 (Primary).

The case of approximately additive mappings was solved by D.H. Hyers

f(x + y) + f(x ~y)~ 2/(ar) - 2/(y) = 0 (A)

clearly has f(x) = ex2 as a solution with c an arbitrary constant when f is a

f(x + y + 2) + f(x - y) + /(y - z) + f(x - z)

Recently, the first author investigated the generalized Hyers-Ulam stability

2. The Results on B-Quadratic Mappings

for all x, y, z G b®i \ {0}, or

for all x, y, z G b®i \ {0}.

Definition 2.1 A quadratic mapping Q: pBi is called B-quadratic if

Theorem 2.1 Let f: #Bi —> b®2 be a mapping such that

||a2/(x + y + z) + a2f(x - y) + a2f(y - z) + a2f(x - z)

for all x € \ {0}. The mapping Q: bBi —> b®2 is given by

limn_>Oo if V satisfies (i)

for all x G bBi.

||a2/(3na;) + 3a2/(0) - Qf^^ax)|| < p^^x, 3n~1x) (2.3)

||/(3a;) + 3/(0) - 9/(x) || < p(x, x, x) (2.4)

for all x e b®i \ {0}. Replacing x by 3n rax in (2.4),

||/(3nax) + 3/(0)-9/(3n“1ax)|| < p^^ax, 3n~1ax, 3n~iax>

for all a e Bi and all x, y, z e b®i \ {0}.

< K\a\ ■ Ml (2.6)

for some K > 0. Using (2.3), (2.5) and (2.6),

+ p^^ax, 3n~1ax, S^ax) + 3K\a2 - 1| • ||/(0)||

for all a € Bi and all x G #Bi \ {0}.

for all x G b®i \ {0}- For a G Bi \ {0},

for all x G b®i \ {0}. So Q is B-quadratic.

||a2/O + y + z) + a2f(x - y) + a2f(y - z) + a2f(x - z)

for all a E Bi and all x,y,z e b®i \ {0}- If f(fx) is continuous in t e

for all x 6 b®i \ {0}.

Corollary 2.2 Let Ei and be complex Banach spaces and f: Ei —» E^ a

|| A2/(a? + y + z) + A2/(z -y) + X2f(y - z) + A2/(t - z)

for all x e Ei \ {0}.

Theorem 2.2 Let f: b®i —> b®2 be a mapping such that

|| f(ax + ay + az) + f(ax — ay) + f(ay — az) + f(ax — az)

for all a E Bi and all x,y,z e b®i \ {0}. If f(tx) is continuous in t e

\\f(3nax) + 3/(0) - 9a2/(3n-1x)|| < ^(3"-^, 3"-^, S71"1^) (2.8)

for all a e Bi and all x e \ {0}. Putting a = 1 and replacing x, y and z

||/(3a?) + 3/(0) - 9/(x) || < <p(x, x, x) (2.9)

for all x e bBi \ {0}. Replacing x by 3n-1or in (2.9),

||/(3™a?) + 3/(0) - 9/(3n-1ir)|| < ^(T^x, S^x, T^x) (2.10)

+ W^f^x) - a2/(3^)|| + || - 3/(0)||

for all a e Bi and all x e bBi \ {0}.

|| / (ax + ay + az) + f(ax — ay) + f(ay — az) + f(ax — az)

for all a e Bi and all x^y^z e b®i \ {0}- If f(fx) Is continuous in t e

for all x G bBi \ {0}.

||/(Ax + Xy + Xz) + f(Xx - Xy) + /(Ay - Xz) + f(Xx - Xz)

- 3X2f(x) - 3A2/(y) - 3A2/(£) || < ^(x, y, z)

PROOF: The proof is similar to the proof of Corollary 2.2. □

3. The Results on Bsa-Quadratic Mappings

Definition 3.1 For a e B, let b = aa

Theorem 3.1 Let f: #Bi —> 582 be a mapping such that

\\bf(x + y + z) + bf(x — y) + bf(y - 2) + bf(x - z)

for all a e Bi and all x,y,z G bBi \ {0}. If f(tx) is continuous in t G R

||6/(3nx) + 36/(0) - 9/(3n-1ax)|| < ^>(3n~1x, 3n-1x, 3n-1x) (3.2)

||/(3x) + 3/(0) - 9f(x)|| < <p(x, x, x) (3.3)

for all x G bBi \ {0}. Replacing x by 3n-1ax in (3.3), we get

\\f(3nax) + 3f(0)-9f(3n~1ax)\\ < <p(3n~1ax,3n-1ax,3n~1ax) (3.4)