Professional Documents
Culture Documents
Functional Equations,
Inequalities and
Applications
Edited by
Themistocles M. Rassias
Department of Mathematics,
National Technical University of Athens,
Zografou Campus, Athens, Greece
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
Themistocles M. Rassias
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
Department of Mathematics, Chungnam National University
Taejon 305-764, Korea
wgpark@math.cnu.ac.kr
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
00 1
■?(!, y, z) := £ SSTP(3"‘i. 3mv, 3”“z) < oo (i)
m=0
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
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),
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),
||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)||
. z x a2f(3nx} „ f(3nax) x
a2Q(rr) = hm ------ ------ = hm —-- ------= Q(ax)
’ n—>oo 9" n-+oo 9n ’
(
|a|Ax ) = |a|2Q ( Ax) = |a|2 Q(x) = a2Q(x),
|a| / \lal / lal
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
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)
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. □
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),
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) ||
< p(3n~1x.l3n~1x.l3n~lx)
+ K (^(3"-7 3n-1x, 3n~lx) + 3||/(0) 10 + 31|/(0) ||
Corollary 2.3 Let p / 2, 0 > 0 be real numbers and f: pBi —> bB2 a
mapping such that
Corollary 2.4 Let Ei and E% be complex Banach spaces and f: E\ —> E<i a
mapping such that
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: 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 <p satisfies (i). Replacing x, y and z by 3n-1x in (3.1),
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
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)\\
8 Functional Equations, Inequalities and Applications
A r bf(3nx) f(3nax)
bQ(x) = hm —- ----- = hm —-- ------ = Q(ax)
’ n—foo 9n n—>oo 9"
Q(ax) = Q
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).
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 3"-1x in (3.5),
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),
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)||
<?(3n-xx,3n~lx,3n~lx')
+ K^p^x, 3n^x, 3-1x) + 3||/(0)||) +3||/(0)||
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.
10 Functional Equations, Inequalities and Applications
[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
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.
1. Introduction
Let (X, +) and (V, +) be arbitrary groups. For f, g, h: X —> Y we define
the differences
11
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 11-19.
© 2003 Kluwer Academic Publishers.
12 Functional Equations, Inequalities and Applications
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
Then we have
Lemma 2.1 The space X\ with norm (2.1) is a linear normed space.
Now denote
||5(T,y)|| x,ytX,
Lemma 2.2 The space X^ with norm (2.4) is a linear normed space.
HOU = 3. (2.6)
Moreover,
Ar
{ 1,
0,
X = 2xn,
otherwise,
Suppose that ||C|| <3. Then there exists £ > 0 such that
||C(/)H(3-s)||/|| forall/eXA.
which is imposible. □
3. Pexider operator
Define
and
Theorem 3.1 The Pexider operator P: —> X^, given by the formula (3.1),
is a linear bounded operator and
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)
Corollary 4.2 The quadratic operator Q: Xx —> X^, given by formula (4.3),
is a linear operator satisfying the inequality
IIW)K6||/|| forfexx.
16 Functional Equations, Inequalities and Applications
4.3 A-operator
Let’s denote
Corollary 4.3 The operator A: X\ —> X^, given by formula (4.4), is a linear
bounded operator satisfying the property
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
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
if and only if
Set
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
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
18 Functional Equations, Inequalities and Applications
□ The simple verification of the other part of the theorem is left to the reader.
Theorem 6.2 Let the assumptions of Theorem 6.1 be fulfilled. Then the equa
tion
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.
Cauchy and Pexider operators in some function spaces 19
[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.
[8] Th.M. Rassias: ‘On the stability of the linear mapping in Banach spaces’,
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
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.
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
21
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 21-37.
© 2003 Kluwer Academic Publishers.
22 Functional Equations, Inequalities and Applications
of the interval [a, fe] and Rn(Im is the remainder, usually expressed in the
integral form
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
where
/(r)|l := ess sup (t)
ll[a,6],oo t6[o>6] I
0 & Ja
provided f is absolutely continuous with f' G Loo [a, b], or Cebysev’s inequal
ity
provided f and g are absolutely continuous with f' G Loo [a,b], are other
natural examples.
The median principle for inequalities and applications 23
If6 1 rb If6
/ f G) 9 (*) dt - —— / f (t) dt ■ —— / g (t) dt
b a Ja ba J a b a Ja
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.
The following result, which will be called the “Median Principle", holds.
—oo < 7 < g(n~^ (z) < T < oo for a.e. x E [a, b], (2.2)
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
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.
Z(/l;a,6)<p(||/r||M)OO;a,6), (3.1)
The median principle for inequalities and applications 25
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
•6 b
h (t) du (t) (3.3)
a
b
h (a?) I (a;) du (a;) < IMm.U'IIm.oo V«- (3.4)
a
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)
f I (x) dx = 0. (3.6)
Ja
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)
< 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,
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).
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)
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
0 if x = 0 or x = 1.
u (a;) =
1 if are (0,1).
We have
and
b
V (”) =2
a
Then, by (3.15) we deduce 2D > 1 giving D > 1/2, and the theorem is thus
proved. □
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
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) |
1 fb
iW = sW~
Then, obviously
f I (x) du (rr) = 0,
Ja
and by (3.17) we deduce (3.18).
The median principle for inequalities and applications 29
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
1 rb
h (rr) — ------- / h (i) dt
b ~ a Ja
1 (x-
-+ ----- — llh/||[a,6],oo(6-°)’ ®€[a,6]; (4.1)
4 \ b—a
b—a
1 (x-9^
- + --------— IMIm.oo - °) > (4.2)
4 \ b—a
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.
The median principle for inequalities and applications 31
<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.
— oo < 7 < f' (rr) < T < oo for a.e. x e [a, b]. (4.5)
i rb
f 0) - 7—- / f (0 dt -
b — a Ja
1
(r-7)(b-a) (4.6)
~ 2 4 \ b—a
PROOF: Consider the function h(x) = f (x) — 2^x, x e [a, 6]. Applying
the “Median Principle” for Ostrowski’s inequality, we get
Theorem 4.2 Let f be as in Theorem 4.1. Then one has the inequality
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).
— oo < m < g (x) < M < oq for a.e. x e [a, b], (4.8)
< L (6 _ a) (M - m) (r - 7) (4.9)
1
g (z) dx
b—a
(4.10)
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)
n—1
l + (-l)fc (b - a)fc+1
f wdt - 52
(k + 1)! 2fc+i
A:=0
34 Functional Equations, Inequalities and Applications
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)
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)!
4" In Fn Tn
for a.e. t G [a, b]
2 2
and
Corollary 5.1 With the assumptions in Theorem 5.1, one has the inequality:
The median principle for inequalities and applications 35
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>
[ f CO dt
Ja
= £ [(I ~0)1+1 /W <o) + <-1)‘ l'b ~ o)‘+1 fti>
rb
f (t) dt
1
< M [( *
-< +1 + (6-
r
* +1l , (5-10)
(n + 1;! II lloo L J
/ f (0 * - n
£~^ (fcTI)!
1 (“) [r/W (0) + (-1)‘f<t> (6)]-I
Theorem 5.2 With the assumptions in Theorem 5.1, we have the inequality
36 Functional Equations, Inequalities and Applications
(5.12)
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]
The median principle for inequalities and applications 37
[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
Department of Mathematics, Chungnam National University
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.
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
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 39-65.
© 2003 Kluwer Academic Publishers.
40 Functional Equations, Inequalities and Applications
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
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
Then there exists exactly one quadratic function g: V —> X such that
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) — limn^oo 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
°° 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’)
5x 3x\ (3x bx
+ft (2x, —2x) + ft(2x, 2x) + 2ft (2z, —x) + 2ft(x, 2x)
f3x 3x\ f3x 3x\
42 Functional Equations, Inequalities and Applications
\ ~ (x ~ ( —x y\ „ rx y\
M2(<p,x,y) = <pl-,— )+<p —)+<4_ n
~x —y ~x $y\ , ~ (^x -y 3x 3y
( 2 ’ 2 )+ ( 2 ’ 2
T’T
for all x, y 6 V \ {0}.
Then there exist exactly one quadratic Junction Q: V —> X and two unique
additive functions T,T': V —> X such that
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^(2^i’^T) <o° (d>)
1=0
for all x, y e V.
Let
M3((p, x) = | ^(x, x) + 2<^(0, x) + 2<^(x, 0)
for all x, y e V.
The following theorem is also shown in [13].
Theorem 1.3 Let <p satisfy one of the conditions (c) or (c’) and one of the
conditions (d) or (d’). Suppose that the functions f,g,h)k: V —> 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))
k^-k^-Q^-^T^-r(x^
li™ g^x)-g^-z-x)
linin_>oo Qn+l ? ifp satisfies (d)
limn_+0O2n(y(2-na:) - 5(0)), if(p satisfies (d’)
or
oo
or
00
^(x, y, z) = 22 3 V (3Z+1 > 3Z+1 3Z+1) < 00 (f )
l=o
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)
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
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
/(^) = fs(xfi
g(x) = fe(-x) + /7(0),
2h(x) = f2(x) + fi(-x)
Hyers-Ulam-Rassias stability of a Pexiderized quadratic equation 47
and
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
f'(x) = f5(x),
g'W = y6(o) + /7(z),
2/r'(x) = /i(t) + /2(x)
and
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
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 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
\\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)
Then there exist exactly one quadratic function Q: V —> X and four addi
tive functions V —> X satisfying
(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))
(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, ~*
for i = 1,2,3,4, 5,6,7, k = 1,2,3 and for all x e V and the functions Tk
(k = 1,2,3), T4 are given by
um )-fi(-2 ”x)-f5(2n*
fl(2n* )+f 5(-2nx)
iiiiin_>oo 2n+1
if (p satisfies (h)
lim^oo 2-1 (A (^) - A (^) - f5 (^) + f5 (^))
if ip satisfies (h’)
(2.21)
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
and
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)
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)||
Hyers-Ulam-Rassias stability of a Pexiderized quadratic equation 53
for all x e V. We get (2.17) from (2.13), (2.25) and the inequality
for all x eV. We get (2.18) from (2.16), (2.22) and the inequality
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')
||/5(s)-/5(0)-Q(2:)-7Hz)||
\\h(x)-feW-Q(x)-T2(xfi\
' ] 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.
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
||/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)
2’ 2/ ’
56 Functional Equations, Inequalities and Applications
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)
limn^oo if <p satisfies (f’)
PROOF: First assume that <p satisfies (e) and (f). Replacing y and z by y/2 and
y/2 in (3.1), we obtain
, / ’y2’y\
*
^( 2)
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
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).
Hyers-Ulam-Rassias stability of a Pexiderized quadratic equation 57
-4/5g)-4/6(y)-4/7g)||
(X x\
M2’y’2J
/'(
*
) = A(x),
g'W = /2O),
2/i'(a;)=4/5n+477n
\ Zi / X /
and
. /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
-4A(^) -4
* (I) -4/7(y)ll
(X X \
<^(2’2’^
/"(x) = A(x),
/(x) = /3(x),
58 Functional Equations, Inequalities and Applications
2/l"(I)=4/s(^)+4Ag)
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
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 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) =
limn^oo^ ifp>t
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
||/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)||
||/4(a;)-/4(0)-Q(T)-74^)||
x) + M3^, -x) + M4(V», x, —x) + ^(V’, -x, x)
- /5(o) - Q& -
- /6(0) - Q(x) - r*
i(
)~ )+W)
T*
2( + 74(rr)
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)
1- /i(2nx)
limn_-x ifcp satisfies (g)
Q(®) =
lim^oo 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
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)
Hyers-Ulam-Rassias stability of a Pexiderized quadratic equation 61
and
2fc'(x)=4/6(x)+4/7(0)
|| 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) -
= lim
n—>00 4n
= lim h^nx) + f^n^
471
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
(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) +
( 2 1 4 + 2? 1\ .. .._ 35 ,,n.
“ \J4 - 2?| + 2? + 2|2 - 2?| + 4/ + 6
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
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),
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.
64 Functional Equations, Inequalities and Applications
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,^i2/2 — 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.
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:
*This work was financially supported by KOSEF (2000); project no. R02-2000-00005.
67
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 67-71.
© 2003 Kluwer Academic Publishers.
68 Functional Equations, Inequalities and Applications
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).
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)
2xy = u,
(2.5)
x 2 — 2y^ — v
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
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
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
for all xi, X2, yi, y2 G X, then f(x, 0) is either bounded or multiplicative and
further it satisfies
for xi, x2 G X. If we substitute m(x) for /(x, 0) in the above inequality, then
we have
for all x, y G X. □
On the Hyers-Ulam-Rassias stability of a functional equation 71
References
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.
1. Introduction
In the present paper we deal with the simultaneous pair of functional in
equalities
73
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 73-89.
© 2003 Kluwer Academic Publishers.
74 Functional Equations, Inequalities and Applications
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
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])
A = {kd : ke Z}.
Lemma 2.2 Let a,b G R, 0 < a < 1 < b, be fixed. Let us denote M =
{ambn : m,n € N}.
additively-multiplicative type:
multiplicatively-additive type:
/ 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 (x) = px + f (0), x e R,
76 Functional Equations, Inequalities and Applications
where p :=
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),
and, consequently,
Setting
a (3
P-= ~ = V,
a b
we can write this inequality in the form
tn G A (n G N), lim tn = xq — x.
n—>oo
A pair of functional inequalities of iterative type 77
Letting here n —> oo, and making use of the continuity of f at xq, we obtain
/(rr) <ptn + n G N.
Letting here n —> oo, and making again use of the continuity of f at xq, leads
to the inequality
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(kd + t) = — kd + /(t), k G Z, x G R.
f(d + x) = —d + /(t), x G R.
78 Functional Equations, Inequalities and Applications
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.
satisfies system (3.2), is continuous, and / |[o,d]— fo- Since the ‘moreover’
part is obvious, the proof is complete. □
if and only if
where p
r := a
A pair of functional inequalities of iterative type 79
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,
Setting
log a log (3
P-=-------=
a b
we can write this inequality in the form
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
Letting here n —> oo, and making use of continuity of f at xq, we obtain
/(x0) < x g R.
tn G A (n G N), lim tn = x — xq
n—>oo
then
where p:=
exists, is greater than 1, and, for every continuous function fo: [1, d] —>
R satisfying the condition
Proof: To prove 1), suppose that I = (0, oo). Similarly as in the previous
proofs, from (3.8) by induction, we get
Hence, putting
a f3
V' log a log b ’
M={ambn:
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
tn G M (n 6 N), lim tn = —,
n—>oo Xq
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
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).
Theorem 3.4 Let a,b, a, (3 be fixed positive real numbers such that
1 , logo: _ log/3
’ log a log b
then
/(
*
) = /(-l) z<0,
where p:=
exists, is greater than 1, and, for every continuous function fo: [1, d] —>
R satisfying the condition
Hence, putting
_ logo: log/3
’ log a log b ’
we obtain
tn G M (n G N), lim tn = —.
n—>oo x
From (3.13) we obtain
Letting here n —> oo, and making use of the continuity of f at xq, we get
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 < b and Note that if I = R or I = [0, oo) or I = (—oo, 0]
and f: I —> R satisfies system (3.11), then /(0) = 0.
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).
w/iere p :=
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.
Theorem 4.1 Let (Q, S, p) be a measure space with two sets A,B € S such
that
Suppose (j)\ (0, oo) —> (0, oo) is a bijection such that </>-1 is continuous at
least at one point and
log^^A)^!)) = log^"1^^)^!))
(4.1)
log^(A) log/z(B)
Iflpep satisfies the condition
(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 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
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
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)
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.
88 Functional Equations, Inequalities and Applications
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-T^ioga+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.
A pair of functional inequalities of iterative type 89
ON APPROXIMATE
ALGEBRA HOMOMORPHISMS
Chun-Gil Park
Department of Mathematics, Chungnam National University
DaeJeon 305-764, South Korea
cgpark@math.cnu.ac.kr
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
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
91
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 91-104.
© 2003 Kluwer Academic Publishers.
92 Functional Equations, Inequalities and Applications
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.
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.
3
On approximate algebra homomorphisms 93
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. □
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, x)
r \ a J
for all x e a^- So one can obtain
- ^-f 1 /
or o
and hence
(2.1)
On approximate algebra homomorphisms 95
||£>iT(t,?/,2)||
/ 7 \ n
= lim f—
n—>oo \ or /
/3r\n /3r\n \
< —
- I — I
2\^r) — JI X.'\d
I\\d I --- J X 7 ,’
— J X,' \d
96 Functional Equations, Inequalities and Applications
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),
d /3r \
DuTfjC) x, x) = -uT I —x ) + 3uT(x) — &I\ux) = 0
r \ d J
uT(x) = T(ux)
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
for all x e a&- Obviously, T(0x) = OT(rc) for all x G a&- Hence
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. □
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
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
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
2r \
)
/(
* - 7T-f d xJ
— < -<p(x,x,xf
Zr
and hence
/(*)
I (2.3)
2r\n
T(x) = lim ~d) X (2.4)
n—»oc
= lim (■£-
n—>oo \ 2r
r ( d
< lim —
n—»oo \ 2r
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),
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 )
100 Functional Equations, Inequalities and Applications
d ( 2t \
DuT(x, x, x) = 3uT(x) + 3uT(x) — 3-T [ —ux I = 0
r \ d J
uT(x) = T(ux)
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.
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.
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)
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 /
/ 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
102 Functional Equations, Inequalities and Applications
— T(x)f(y) (3.1)
/ \i \ ( // V \\
(
/ /3r\3 \ \
\
//TrV \
((dj y)J
/
( -
\ \ aJ
V
x y =T
j 1
-
\ \ dJ
x /(y)
1
= -7 rW/W
\ a j
/ d V / /3rV \
d f v) y = Ti-x'>Ky'> <3-2>
T(xyT(y) = T(x)f(y)
T(x-y) = T(x)T(y)
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).
The rest of the proof is the same as the proof of Theorem 3.1. □
On approximate algebra homomorphisms 103
for all x € A.
The rest of the proof is the same as the proof of Theorem 3.1. □
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.
104 Functional Equations, Inequalities and Applications
References
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.
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
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 105-137.
© 2003 Kluwer Academic Publishers.
106 Functional Equations, Inequalities and Applications
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
(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.
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
+ E Sf [/’‘-’’(I) - 7l‘-,)
k=2
(2r)! Jo
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.
108 Functional Equations, Inequalities and Applications
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)
fl fl t2
/ tP2r(t)dt = -b2r - / P2r+l(t)dt + fe2r— Io
Jo Jo
If1 1
~ 2^r J(j dt = — — 62r.
o
/•l
f0\-iyp2r(t)dt
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)
/•t , । / yl \
|#r| < yo |P2r(i)| • |/(2r)(i)| dt < ( / |P2r(f)| dt\ X
110 Functional Equations, Inequalities and Applications
)2r+1^/(2'> - H'(a’b>
Hadamard and Dragomir-Agarwal inequalities 111
(2r)! 2
91 V9
< (6-«)2r+1^ |/<2r>(< + |/M(t)|
2
b) If |/(2r> | is concave,
< (t>-«)2r+1^JT
where
To obtain appropriate results for numerical integration from the Euler trapezoid
formula, we will apply the results above to each interval of the subdivision
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 '
h2T^M
112 Functional Equations, Inequalities and Applications
(2.10)
[/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
rnh
/ Wdt-Ttf-h)
Jo
Hadamard and Dragomir-Agarwal inequalities 113
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
h2 fnfl h2
-M (/"; ft) < T(f- K) - 2 /(t) dt < —T (/"; ft).
f(t)dt-T(f-,h)
h2
f(t)dt-T(f-,h)
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
so that
j\l - i)|p2(i)|di = = |(1 - 21-2r)|B2r|.
/Q) = p(o
*+EpfMD -/--»(»)]
We can parallel the development of the previous section with the following
two theorems.
To obtain appropriate results for integration from the Euler midpoint formula,
we apply the above results to each interval of the subdivision
First we introduce
nh
U
- - /(2"1)(0)] '■ ■
1 /"(0) +/"(!)
24 2
< _1_
“ 24 2
1
- 24 \2J
h2 rnh h2
Jo
118 Functional Equations, Inequalities and Applications
< - V
- 24 2
m=l
Zt:
h2
dt-M(f-K)
/(0)+4/Q)+/(l)
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
-2—<X
< h _ o2-2rx/(2r)(l)+/(2r)(0)
- 3(2r)!1 ’ 2
b) is concave, then
lx'sW(1-2"2r /'2r>
To obtain appropriate results for integration via the Euler-Simpson formula,
we apply the above results to each interval of the subdivision
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)
l^l<(2fc)2r+1^4(l-22-2’')
x max {|/(2r)(0)|,|/2r\2n/l)|}.
+ +/(!)]
£f(t)dt-± /(0)+4/Q)+/(l)
Hadamard and Dragomir-Agarwal inequalities 121
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 \"/
dt — X(f;li)
“ 2880
m=l
n(2h)5
2880
where
= Bl, (1 - t) + Bl, - <) - B2, (1) - B2r (|) .
We note that this does not change sign on the interval [0,1] and
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
b) is concave, then
To obtain appropriate results for integration via the Euler two-point formula,
we apply the above results to each interval of the subdivision
First we introduce
2 fc=i L '
and
nh
U
r-1 h2k r 1'
k^l1 JJ
-ft2r2M(1-31’2r)T (/(2M-
Iff is (2r + 2)-concave, the inequalities are reversed.
(1 - 31“2r)
b) is concave, 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.
'|/W(0)|g + |/W(i)|a]I/’
< 13
- 19440 2
b) is concave, then
©
Corollary 5.3 If f: [0, nh] —> R is 6-convex, then
Jo
Hadamard and Dragomir-Agarwal inequalities 125
< 13/l5 y-
~ 19440 2
m=l _
b) is concave, then
rnh
/ f(t) dt-p(f'h) < 13/i4
Jo “ 19440
where
1 \ /1\ /q \
(-~i
41 / -b;,l;-t +2s;r --t
J
— 2B2r
\ 41 /
+ B2r — 2B2r
We note that this does not change sign on the interval [0,1] and
--2(l-2-2r)B2r.
We can parallel the development of the second section with the following
two theorems. For the sequel we shall utilize
l^2r| /r>3—4r q *
rjl—2r । i\/
(2r) (
3^(23~4r"3’21~2r + 1)
To obtain appropriate results for numerical integration from the dual Euler-
Simpson formula, we apply the results above to each interval of the subdivision
Let us denote
= (kh-^h\ - f f kh - J + 2/ (kh -
6 t=i L '
Hadamard and Dragomir-Agarwal inequalities 127
and
rnh
D* ■■= (-1) b f0
dt - D(f; ft)
*
|£)
| < ^2r+ljgd(23-4r _ 3.2l-2r +
x max||/(2r)(0)|, |/(2r)(n/i)|} .
7 yw(o)i8+i/W(i)n1/5
- 23040 2
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)
Hadamard and Dragomir-Agarwal inequalities 129
where
1 \ /o \
(o
t j + 3B2r
J
— 2jE?2r ~
\o
3jE?2r
J
“ 3jE?2r ( q ) •
We note that this does not change sign on the interval [0,1] and
= (1 — 32-2r)|B2r|.
J0
Also, F£(l - t) = F^(t), so
We can parallel the development of the second section with the following
two theorems. Define
- E Unit1 - [f^m - 1.
f' o(2/c)l L J
130 Functional Equations, Inequalities and Applications
<71 1/9
IQ I < l^2rl fl — o2—2r
8(2r)!(1 3
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
First we introduce
S(/;h)
and
(1 - 32-2r)
b) is concave, then
*
I
IQ |-®2r| /-. o2—2r
8(2r)r ~ 3
1
6480J \2j
sH/(o)+3/G)+3/©+/(i
< 1 /(4)(0) + /(4)(l)
- 6480 2
Iff is 6-concave, the reversed inequalities apply.
o
i ri/(4|(°)r+i/|4)(Drr/J
~ 6480 2
b) is concave, then
132 Functional Equations, Inequalities and Applications
< 1 f (4) (
- 6480 J \2J
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)
where
(i \ /1 \ /k \
j - t) + 2B2‘r (2 - + 3B5r \6 “ 9
-3B2r(|)-2B2rQ)-3B2r(|).
We note that this does not change sign on the interval [0,1] and
We can parallel the development of the second section with the following
two theorems. Define
First we introduce
fc=i
and
nh
Lt:
r-1
gw(1-21-2‘,(1-32‘“
b) is concave, then
^4o/W G)s Z1 /w dt -
< 7 /W(o) + /W(i)
G)+2/ G)+3/ G)
S 51840 2
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
dt — L(f;h)
7h5
~ 51840
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).
Hadamard and Dragomir-Agarwal inequalities 137
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
Themistocles M. Rassias
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.
139
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 139-147.
© 2003 Kluwer Academic Publishers.
140 Functional Equations, Inequalities and Applications
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
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).
On Ulam stability in the geometry of PDE’s 141
following:
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.
G(H (gi (x1, ..., x^)),..., H(gp(x\ ..., x’)), x1,..., x’) = 0 (2.1)
On Ulam stability in the geometry of PDE’s 143
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).
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].
||G(/,(g1,---,gn))||<^1,---,9n), (2.4)
with <p: Xn —> [0,00) fixed, there exists a solution f of (2.3) such that
VqeX,
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:
144 Functional Equations, Inequalities and Applications
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.
(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.
On Ulam stability in the geometry of PDE’s 145
Lemma 3.1 The linear Navier-Stokes equation (NS)s C JZ>2(E), at the so
lution s, has the following form:
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.
(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)\
3The vertical tangent space vTW is calculated with respect to the fiber structure tt: W —► M.
146 Functional Equations, Inequalities and Applications
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:
Corollary 3.1 The stability of a global laminar solution of (NS) can be char
acterized by the following conditions:
References
[1] D.H. Hyers: ‘On the stability of the linear functional equation’, Proc. Nat.
Acad. Sci. USA 27 (1941), 222-224.
Themistocles M. Rassias
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.
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
&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
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 149-158.
© 2003 Kluwer Academic Publishers.
150 Functional Equations, Inequalities and Applications
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).
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).
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
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)
F(M(a,b),AT(a,b)) = F(a,b),
Fa(M(a,b),N(a,b)) ■ Ma(a,b)
+ Fb(M(a,bfN(a,b^ • Na(a,bfia,b) = Fa(a,b).
(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
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)+
On certain functional equations and mean value theorems 153
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). □
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,
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) =
?/(?) + (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
□
154 Functional Equations, Inequalities and Applications
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
Corollary 2.3 If the function f defined by (2.1) verifies the functional equation
p(X) = aX^+2s+1^k-n + (3
Corollary 2.4 If the function f defined by (2.1) verifies the functional equation
where
and
ro,fc(0) = lim r„,fc(0) = akcos2ebksia2e (k ± 0),
n—>0
On certain functional equations and mean value theorems 155
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
f(a,b-,p,rnff) = p([£g(a,b)](fc-1)lV(a,b)) .
/a(a,fe;p, rn>fc)
= p'tco - !)(£,(«, X
fab(a,b;p,rn^ = p'/(C1)—C1-C1+p,(C1)-~C1.
oa ob oaob
Setting a = b = c yields
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). □
NabM =
k nr k\ (—^kq + 4q — 2k + 2(1 — s) A // «
cV(c + ------------ 7?------------ -------- - + 1 + n P (c ) = 0-
y fcz /
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.
158 Functional Equations, Inequalities and Applications
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.
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
159
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 159-165.
© 2003 Kluwer Academic Publishers.
160 Functional Equations, Inequalities and Applications
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
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:
and
*
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
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}
in (2.5).
162 Functional Equations, Inequalities and Applications
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,
We thus obtain
i 4iIgII«k(x)- (2.10)
Ilf ll£2(x)-" 21
Corollary 2.1 For any member G € 77k (X) in (2.5), we obtain the estimate
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
3. Example
As a typical example, we shall consider the integral transform
K{x^x2} = ITU(x2,T-^U^,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
2) F'(x)GL2(0,n
3) F(T) = 0,
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
164 Functional Equations, Inequalities and Applications
2) F"(x) e L2(0,T),
3) F(T) = F'(0) = 0,
F"(x) = -/(x),
[T F F"(xfdx. (3.5)
Jo Jo
Then, from Corollary 2.1, we obtain
- ^IIsIIl2(o,t)-
From Corollary 2.3, we have
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.
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
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 167-181.
© 2003 Kluwer Academic Publishers.
168 Functional Equations, Inequalities and Applications
)
*
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
(H3) 6X * 8e = 6e * 8X = <5X;
(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]).
= 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
Functional equations on hypergroups 169
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.
(D3) 6X * 6e = 6e * 8X = 8X\
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
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
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
J-l PkPnPm dp
c(n, m, fc) =
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|
Functional equations on hypergroups 171
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|
xPn^ = 2^nPn+i(T)+2^uPn-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.
Tym(x) = m{x)m{y)
172 Functional Equations, Inequalities and Applications
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
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.
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)
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
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. □
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)
n+l
On the other hand, we have for any positive integer n and for all real x that
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
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
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)
p(f>n — p(n) + 1,
hence
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. □
< Po(n) = 1,
< Pl(n) = aP^f),
^n^a^Pf + bP^,
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
and
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
wW = a2P"(l) + 6P'(1)
for all n in N, and the theorem is proved. □
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 * (Sy *
5 2)].
Jk
The assumption of the theorem means that
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
[ [ 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
= 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. □
Mx / d(5x * 6y)
Jk
for any bounded function : K —> C and for any y in K.
(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(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
180 Functional Equations, Inequalities and Applications
for all x, z in K, then the above equation can be written in the form
and the right hand side is 0 by the invariance of M. It means that a is additive.
On the other hand,
, y)
(* / fd(dx * Sy) - f(x) - /(y).
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).
Functional equations on hypergroups 181
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.
1. Introduction
The Cauchy functional equation [14, 15]
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
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 183-189.
© 2003 Kluwer Academic Publishers.
184 Functional Equations, Inequalities and Applications
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
^akfk\z) = 0 (2.2)
k=0
which denotes the left hand side of (2.1) with y replaced by c — x. Then (2.1)
can be written as
= 0. (2.5)
OX
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)
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].
186 Functional Equations, Inequalities and Applications
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
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
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,
n
£ Ff™ (x)F
* n~V (y) = F?(n~r'> (x + y) (3.7)
k=0
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.
188 Functional Equations, Inequalities and Applications
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.
191
Th.M. Rassias (ed.), Functional Equations, Inequalities and Applications, 191-221.
© 2003 Kluwer Academic Publishers.
192 Functional Equations, Inequalities and Applications
(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
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
On the Aleksandrov-Rassias problem for isometric mappings 193
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.
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.
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.
Figure 1.2.
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
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:
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.
|>|7(r)-7(«-2)|
On the Aleksandrov-Rassias problem for isometric mappings 197
>|/(n-l)-/(n-2)|-|/(r)-/(n-l)|
>1 —|r —(n —1)|
- 1
~ 2
/(-)=
and
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.
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:
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.
On the Aleksandrov-Rassias problem for isometric mappings 199
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 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^
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)
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
linear isometry up to translation.
200 Functional Equations, Inequalities and Applications
Example 2.1 ([4]) Let p > 0 be a fixed real number. Consider the function
V’: R —> R defined by
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.
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.
On the Aleksandrov-Rassias problem for isometric mappings 201
and
(z—x—y,z—x—y) = (z, z)-2(z, x)-2(z, y)+(x, x)+2(x, y)+(y, y) = 0.
Pl
Figure 2.3.
202 Functional Equations, Inequalities and Applications
Figure 2.4-
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,
On the Aleksandrov-Rassias problem for isometric mappings 203
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:
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
Hence
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
Figure 2.6.
204 Functional Equations, Inequalities and Applications
and
n
||T(9n)-T(p)|| < l|r(Q1)-T(p)|| + £||T(%)-T(9fc_1)|| < nVs.
k=2
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
V^a2 + 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.
On the Aleksandrov-Rassias problem for isometric mappings 205
1 + (w, w) - (x — y, x — y) + (w, w)
= (x + y, x + y) + (w, w) - 4(x, y)
> 2(w,x + y) - 4(x,y). (2.1)
Hence
Therefore
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.
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.
/ 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
P3 P3
Figure 2.8.
Theorem 2.5 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
f: X —> y satisfies the property that
ii) f preserves the distances a, b with the ratio of a and b being for
some positive integer n with n > 1.
Proof:
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.
208 Functional Equations, Inequalities and Applications
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
and
(t + y, x + y) = 1.
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
On the Aleksandrov-Rassias problem for isometric mappings 209
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.
Theorem 2.7 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
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.
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.
210 Functional Equations, Inequalities and Applications
Corollary 2.3 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
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.
Corollary 2.4 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
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.
Corollary 2.5 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
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
isometry up to translation.
Theorem 2.8 Let X and Y be real Hilbert spaces with dim X > 2. Suppose
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
isometry up to translation.
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:
and
Hence
Ikll2 = k,
)
* = (^,^)+(z,y) = (z,x+y) < ikiHk+yll = x/4-u2lkl|.
Hence
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,
R<?i)-T(p)||<^4-a2,
||T(g2)-T(gi)||<y4-a2,
and
< ny 4 — a2.
Since Y is a Hilbert space and T preserves the distance nV4 — a2, then
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
Theorem 2.9 Let X and Y be real Hilbert spaces with dim X > 3. Suppose
that T: X —> Y preserves the two distances 1 and y/2. Then T is a linear
isometry up to translation.
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 .
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.
On the Aleksandrov-Rassias problem for isometric mappings 215
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. □
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.
Then f is a linear isometry up to translation.
216 Functional Equations, Inequalities and Applications
Proof:
1 , x 1 + a, x
a:i =x + ——(y-a;), x2 = x + —— (y - x).
2 +a 2+a
Then
It follows that
||/(x1)-/(x)|| = l,
- /(®2)|| = a, \\f(y) - /(®1)II = 1 + a,
ll/M - /(z)|| = 1 + a, *
Wfiy) - /(
2)|| = 1,
and
*
/(
1) = f (x) + - fix)),
1+a 2+a . .
II = a:+2T^<!'_l)’ X2 = X+2T2^y~x}-
Then
ll/toWWII=2 + 2o.
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.
Then f is a linear isometry up to translation.
Proof:
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).
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.
and
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.
On the Aleksandrov-Rassias problem for isometric mappings 219
References
[19] Th.M. Rassias: ‘On the stability of the linear mapping in Banach spaces’,
Proc. Amer. Math. Soc. 72 (1978), 297-300.
[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.
[34] Th.M. Rassias: ‘On the stability of functional equations and a problem of
Ulam’, Acta. Appl. Math. 61 (2000), 23-130.
[35] Th.M. Rassias (Ed.): 'Functional Equations and Inequalities’, Kluwer
Academic Publishers, 2000.
On the Aleksandrov-Rassias problem for isometric mappings 221
*-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
224 Functional Equations, Inequalities and Applications