Weighted inequalities for monotone and concave functions

Similar documents
ON A CONVEXITY PROPERTY. 1. Introduction Most general class of convex functions is defined by the inequality

INEQUALITIES FOR TWO SPECIFIC CLASSES OF FUNCTIONS USING CHEBYSHEV FUNCTIONAL. Mohammad Masjed-Jamei

Some estimates on the Hermite-Hadamard inequality through quasi-convex functions

SOME INTEGRAL INEQUALITIES OF GRÜSS TYPE

ODE: Existence and Uniqueness of a Solution

SOME HARDY TYPE INEQUALITIES WITH WEIGHTED FUNCTIONS VIA OPIAL TYPE INEQUALITIES

Math 554 Integration

Problem Set 4: Solutions Math 201A: Fall 2016

KRASNOSEL SKII TYPE FIXED POINT THEOREM FOR NONLINEAR EXPANSION

GENERALIZED ABSTRACTED MEAN VALUES

A HELLY THEOREM FOR FUNCTIONS WITH VALUES IN METRIC SPACES. 1. Introduction

Czechoslovak Mathematical Journal, 55 (130) (2005), , Abbotsford. 1. Introduction

The Regulated and Riemann Integrals

NEW INEQUALITIES OF SIMPSON S TYPE FOR s CONVEX FUNCTIONS WITH APPLICATIONS. := f (4) (x) <. The following inequality. 2 b a

A PROOF OF THE FUNDAMENTAL THEOREM OF CALCULUS USING HAUSDORFF MEASURES

New Expansion and Infinite Series

On the Generalized Weighted Quasi-Arithmetic Integral Mean 1

ON THE WEIGHTED OSTROWSKI INEQUALITY

ON THE C-INTEGRAL BENEDETTO BONGIORNO

Advanced Calculus: MATH 410 Notes on Integrals and Integrability Professor David Levermore 17 October 2004

A product convergence theorem for Henstock Kurzweil integrals

The Hadamard s inequality for quasi-convex functions via fractional integrals

Positive Solutions of Operator Equations on Half-Line

Improvement of Ostrowski Integral Type Inequalities with Application

AN INTEGRAL INEQUALITY FOR CONVEX FUNCTIONS AND APPLICATIONS IN NUMERICAL INTEGRATION

S. S. Dragomir. 2, we have the inequality. b a

AN INEQUALITY OF OSTROWSKI TYPE AND ITS APPLICATIONS FOR SIMPSON S RULE AND SPECIAL MEANS. I. Fedotov and S. S. Dragomir

The Bochner Integral and the Weak Property (N)

RIEMANN-LIOUVILLE AND CAPUTO FRACTIONAL APPROXIMATION OF CSISZAR S f DIVERGENCE

Lecture 1. Functional series. Pointwise and uniform convergence.

New Integral Inequalities for n-time Differentiable Functions with Applications for pdfs

ASYMPTOTIC BEHAVIOR OF INTERMEDIATE POINTS IN CERTAIN MEAN VALUE THEOREMS. II

f(x)dx . Show that there 1, 0 < x 1 does not exist a differentiable function g : [ 1, 1] R such that g (x) = f(x) for all

FUNDAMENTALS OF REAL ANALYSIS by. III.1. Measurable functions. f 1 (

The Banach algebra of functions of bounded variation and the pointwise Helly selection theorem

Approximation of functions belonging to the class L p (ω) β by linear operators

arxiv: v1 [math.ca] 11 Jul 2011

SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL (1 + µ(f n )) f(x) =. But we don t need the exact bound.) Set

WENJUN LIU AND QUÔ C ANH NGÔ

Review of Riemann Integral

A basic logarithmic inequality, and the logarithmic mean

A unified generalization of perturbed mid-point and trapezoid inequalities and asymptotic expressions for its error term

UNIFORM CONVERGENCE MA 403: REAL ANALYSIS, INSTRUCTOR: B. V. LIMAYE

W. We shall do so one by one, starting with I 1, and we shall do it greedily, trying

ON AN INTEGRATION-BY-PARTS FORMULA FOR MEASURES

Convex Sets and Functions

Mapping the delta function and other Radon measures

S. S. Dragomir. 1. Introduction. In [1], Guessab and Schmeisser have proved among others, the following companion of Ostrowski s inequality:

Journal of Inequalities in Pure and Applied Mathematics

Regulated functions and the regulated integral

GENERALIZATIONS OF WEIGHTED TRAPEZOIDAL INEQUALITY FOR MONOTONIC MAPPINGS AND ITS APPLICATIONS. (b a)3 [f(a) + f(b)] f x (a,b)

On the Continuous Dependence of Solutions of Boundary Value Problems for Delay Differential Equations

Bounds for the Riemann Stieltjes integral via s-convex integrand or integrator

Journal of Inequalities in Pure and Applied Mathematics

LYAPUNOV-TYPE INEQUALITIES FOR NONLINEAR SYSTEMS INVOLVING THE (p 1, p 2,..., p n )-LAPLACIAN

1. On some properties of definite integrals. We prove

Math 360: A primitive integral and elementary functions

Journal of Inequalities in Pure and Applied Mathematics

Principles of Real Analysis I Fall VI. Riemann Integration

Review of Calculus, cont d

MAT612-REAL ANALYSIS RIEMANN STIELTJES INTEGRAL

arxiv: v1 [math.ca] 7 Mar 2012

International Jour. of Diff. Eq. and Appl., 3, N1, (2001),

INEQUALITIES FOR GENERALIZED WEIGHTED MEAN VALUES OF CONVEX FUNCTION

arxiv:math/ v2 [math.ho] 16 Dec 2003

f (a) + f (b) f (λx + (1 λ)y) max {f (x),f (y)}, x, y [a, b]. (1.1)

Some Hardy Type Inequalities with Weighted Functions via Opial Type Inequalities

DYNAMICAL SYSTEMS SUPPLEMENT 2007 pp Natalija Sergejeva. Department of Mathematics and Natural Sciences Parades 1 LV-5400 Daugavpils, Latvia

Polynomial Approximations for the Natural Logarithm and Arctangent Functions. Math 230

Mathematics. Area under Curve.

Asymptotic behavior of intermediate points in certain mean value theorems. III

ODE: Existence and Uniqueness of a Solution

LYAPUNOV-TYPE INEQUALITIES FOR THIRD-ORDER LINEAR DIFFERENTIAL EQUATIONS

INNER PRODUCT INEQUALITIES FOR TWO EQUIVALENT NORMS AND APPLICATIONS

Multiple Positive Solutions for the System of Higher Order Two-Point Boundary Value Problems on Time Scales

TRAPEZOIDAL TYPE INEQUALITIES FOR n TIME DIFFERENTIABLE FUNCTIONS

THE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS.

New Integral Inequalities of the Type of Hermite-Hadamard Through Quasi Convexity

An inequality related to η-convex functions (II)

MATH 174A: PROBLEM SET 5. Suggested Solution

p n m q m s m. (p q) n

p-adic Egyptian Fractions

Euler, Ioachimescu and the trapezium rule. G.J.O. Jameson (Math. Gazette 96 (2012), )

Some Hermite-Hadamard type inequalities for functions whose exponentials are convex

Parametrized inequality of Hermite Hadamard type for functions whose third derivative absolute values are quasi convex

Chapter 4. Lebesgue Integration

Keywords : Generalized Ostrowski s inequality, generalized midpoint inequality, Taylor s formula.

ACM 105: Applied Real and Functional Analysis. Solutions to Homework # 2.

ON PERTURBED TRAPEZOIDAL AND MIDPOINT RULES. f (t) dt

1 i n x i x i 1. Note that kqk kp k. In addition, if P and Q are partition of [a, b], P Q is finer than both P and Q.

AMATH 731: Applied Functional Analysis Fall Some basics of integral equations

MATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1

The First Fundamental Theorem of Calculus. If f(x) is continuous on [a, b] and F (x) is any antiderivative. f(x) dx = F (b) F (a).

A General Dynamic Inequality of Opial Type

5.7 Improper Integrals

The Riemann-Lebesgue Lemma

Appendix to Notes 8 (a)

A Convergence Theorem for the Improper Riemann Integral of Banach Space-valued Functions

Math Solutions to homework 1

On Error Sum Functions Formed by Convergents of Real Numbers

Transcription:

STUDIA MATHEMATICA 116 (2) (1995) Weighted inequlities for monotone nd concve functions by H A N S H E I N I G (Hmilton, Ont.) nd L E C H M A L I G R A N D A (Luleå) Abstrct. Chrcteriztions of weight functions re given for which integrl inequlities of monotone nd concve functions re stisfied. The constnts in these inequlities re shrp nd in the cse of concve functions, constitute weighted forms of Fvrd Berwld inequlities on finite nd infinite intervls. Relted inequlities, some of Hrdy type, re lso given. 1. Introduction. In 1939 L. Berwld [8] proved, vi generliztion of men vlue inequlity of J. Fvrd [18], tht if f is non-negtive concve continuous function on [, 1] ( 1 ) nd < p q <, then (1.1) f q (p + 1 (q + 1) 1/q f p, where the constnt (p + 1 (q + 1) 1/q is shrp. If p 1 this is clled Fvrd s inequlity. This inequlity my be interpreted s the converse of Hölder s inequlity nd in the limiting cse with q 1, p, it is reverse Jensen inequlity 1 f() d e ( 1 2 ep ) ln f() d, where the constnt e/2 is shrp. Closely relted to the Fvrd Berwld inequlity is n inequlity of Grüss [21]. It sserts tht the L 1 -norms of certin functions f, g re dominted by the L 1 -norm of the product fg. More generlly, one obtins (cf. Brnes [4] with r 1) the inequlity 1991 Mthemtics Subject Clssifiction: Primry 26D15, 26D7. Key words nd phrses: weighted integrl inequlities, weighted Hrdy inequlities, weighted Hrdy inequlities for monotone functions, weighted Fvrd Berwld inequlity, reverse Hölder inequlity, concve functions. ( 1 ) The intervl of integrtion in Berwld s result is [, b] insted of [, 1]. This generliztion my be chieved in this pper with minor modifictions. [133]

134 H. Heinig nd L. Mligrnd f p g q C fg r for certin indices p, q nd r. The object of this pper is to introduce weight functions u, v, w in these inequlities nd provide conditions on the weights which re equivlent to corresponding weighted forms of the Fvrd Berwld inequlity, Grüss inequlity nd the reverse Hölder inequlity. These nd corresponding vrints re given both on finite nd infinite intervls. To prove these results, number of uilliry inequlities re required. One, due to Clderón Scott [14], with specil cses of erlier origin, is proved here with shrp constnt. These inequlities re of the form (1.2) b f() dg() ( b f() γ d[g() γ ]) 1/γ, for < γ 1 nd f decresing, g incresing s well s some other vrints (cf. Th. 2.1). One of the consequences of this inequlity is tht it permits n esy proof of weight chrcteriztions of Hrdy type in the cse 1 p q from the specil cse of p q. Also using inequlity (1.2), we cn give nother proof of the imbeddings f q,u C f p,v for f nd f q,u C f p,v for f, which were proved erlier by Swyer [39], Stepnov [44] nd Heinig Stepnov [24]. All these results re given in Section 2. In Section 3, we study inequlities of the form (1.3) T 1 f q,u C T 2 f p,v, where T 1, T 2 re the positive integrl opertors T i f() k i (, t)f(t) dt, k i (, t), i 1, 2, f monotone nd C shrp constnt. Some of these results re essentilly known (cf. [2], [16], [24], [27], [33], [39], [42] [44]) nd in the min we prove them with shrp constnts. The results re then pplied in Sections 4 nd 5 to obtin shrp weighted inequlities for concve functions, weighted Fvrd Berwld inequlities s well s reverse Hölder inequlities. As specil cses we obtin results of Brnrd nd Wells [3], whose work motivted this study. All functions considered re ssumed mesurble nd non-negtive. We shll write f, respectively f to men tht the function f is incresing non-decresing, respectively decresing non-incresing. For f we define f 1 (t) inf{s : f(s) t}, inf, nd similrly for f, f 1 (t) inf{s : f(s) > t}, inf. Loclly integrble non-negtive weight functions on (, ) re denoted by u, v, w, nd the conjugte inde of p (, ) is denoted by p p/(p 1), even if < p 1, nd similrly for q. We shll sy f L p w, p >, if f p,w ( f() p w() d <. Finlly, χ E denotes the chrcteristic function of the set E. Inequlities (such s in Theorem 2.1) re interpreted to men tht if the right side is finite, so is the left, nd the inequlity holds.

Inequlities for monotone nd concve functions 135 2. Integrl inequlities. In this section we prove inequlities frequently used in the sequel. Theorem 2.1. Let < b nd f on (, b) nd g continuous on (, b). () Suppose f on (, b) nd g on (, b) with lim + g(). Then for ny γ (, 1], (2.1) b f() dg() ( b f() γ d[g() γ ]) 1/γ. If 1 γ <, the inequlity in (2.1) is reversed. (b) Suppose f on (, b) nd g on (, b) with lim b g(). Then for ny γ (, 1], (2.2) b f() d[ g()] ( b If 1 γ <, the inequlity in (2.2) is reversed. We shll give two proofs of the theorem. f() γ d[ g() γ ]) 1/γ. P r o o f 1. It suffices to prove the theorem when the integrls on the right of (2.1) nd (2.2) re finite. Suppose < γ 1. () Since f() dg() 1 γ [f()γ g() γ ] 1/γ 1 f() γ d(g() γ ) 1 γ [ f(t) γ d(g(t) γ )] 1/γ 1f() γ d(g() γ ) integrting from to b yields d [ 1/γ, f(t) γ d(g(t) )] γ d b f() dg() ( ( b ) 1/γ f(t) γ d(g(t) γ b ) f() γ d[g() γ ]) 1/γ.

136 H. Heinig nd L. Mligrnd (b) Also, f() d[ g()] 1 γ [f()γ g() γ ] 1/γ 1 f() γ d[ g() γ ] 1 γ { b d { b d f(t) γ d[ g() γ ]} 1/γ 1f() γ d[ g() γ ] f(t) γ d[ g() γ ]} 1/γ, nd gin integrting from to b one obtins (2.2). It is esily seen tht inequlities in (2.1) nd (2.2) re reversed if 1 γ <. nd P r o o f 2. For t (, b), let h(t) k(t) ( t ( b t ) γ t f() dg() ) γ b f() d[ g()] Then h nd k re continuous nd for < γ 1, ( t h (t) γ f() γ d[g() γ ] t f() γ d[ g() γ ]. f() dg()) γ 1f(t)g (t) γf(t) γ g(t) γ 1 g (t) γ[f(t)g(t)] γ 1 f(t)g (t) γf(t) γ g(t) γ 1 g (t) Therefore h on (, b) nd so h(b) h(), which proves (2.1). Similrly with k(t). R e m r k 2.2. (i) If f nd g re s in Theorem 2.1, then F (γ) ( b ) 1/γ, ( b f() γ d[g() γ ] resp. G(γ).e. ) 1/γ f() γ d[ g() γ ] re decresing on (, ). (ii) Inequlity (2.2) with, b nd constnt 1/γ on the right side ws proved by Clderón Scott [14, Lemm 6.1]. (iii) If g() 1/p, < p 1, nd b, then (2.1) tkes the form ( f() 1/p 1 d p f() p d

Inequlities for monotone nd concve functions 137 for ll f, while g() yields ( b ) p b f() d p ( ) p 1 f() p d for ll f. The lst inequlity hs been obtined by vrious uthors (cf. Hrdy Littlewood Póly [22, p. 1], Lorentz [28, p. 39], Stein Weiss [41, Th. 3.11i], Mz y [31, Lemm 2.2], Bergh Burenkov Persson [7, Lemm 2.1]. (iv) The specil cse of (2.1) when g() p, p 1, nd f is positive incresing on (, b), b <, hs the form b f(b ) d p ( b ) p ( b f(b d f( d) p. This inequlity with some dditionl constnt C p (1, 2) ws proved by Grcí del Amo [2, Th. 5]. We cn lso write the bove inequlity in the form b ( b p, p f() p (b ) p 1 d f() d) where p 1 nd f is positive incresing function. This inequlity ws proved by Bushell Okrsiński [13] for nturl p, nd for ny p 1 quite recently by Wlter Weckesser [45]. (v) If γ, the reversed inequlities (2.1) nd (2.2) re meningless; however, one does hve the following: If f nd g re s in Theorem 2.1(), then b sup f()g() f() dg(). (,b) In fct, sup f()g() (,b) sup f() (,b) sup (,b) dg(t) f(t) dg(t) Similrly, if f nd g re s in Theorem 2.1(b) then sup f()g() (,b) b b f() d[ g()]. f() dg(). The inequlities given in Theorem 2.1 my be pplied in connection with Hrdy s inequlity. It is well known (cf. [36]) tht the weighted Hrdy

138 H. Heinig nd L. Mligrnd inequlity (2.3) ( ( ) q ) 1/q ( f(t) dt d C v()f() p d holds for ll f if nd only if for 1 p q <, A p,q (u, v) sup t> ( t nd for < q < p, p > 1, B p,q (u, v) { [( t ) 1/q ( t d ) 1/q ( t d v() 1 p d <, v() 1 p d) 1/q ] rv(t) 1 p dt} 1/r <, where 1/r 1/q 1/p. Moreover, if C C p,q (u, v) is the best constnt for which (2.3) holds, then A p,q (u, v) C p,q (u, v), respectively, B p,q (u, v) C p,q (u, v). Writing A p,p A p nd u s () ( u(t) dt)s 1, s >, we obtin from Theorem 2.1 the following: Proposition 2.3. () If 1 p q <, then A p,q (u, v) (p/q A p (u p/q, v) nd C p,q (u, v) (p/q C p (u p/q, v). (b) If < q < p < nd p > 1, then A p (u, v) (q/p) 1/q (r/p ) 1/r B p,q (u q/p, v) nd C p (u, v) (p/q) 1/q C p,q (u q/p, v). P r o o f. () First, for ll t >, we hve ( t ( t u p/q () d (q/p( (q/p( t t [ d v() 1 p d ( ) 1/q ( t d ) p/q ] ( t u(s) ds v() 1 p d, v() 1 p d which gives the equlity. Secondly, by Theorem 2.1(b) with γ p/q 1 nd f replced by

Inequlities for monotone nd concve functions 139 ( f(t) dt)q nd g replced by u(t) dt, we hve ( ) q ) q [ f(t) dt d f(t) dt d { ( ( (p/q) q/p{ ( C ) p [ ( f(t) dt d ( v()f() p d) q/p, ] u(s) ds ) p/q ]} q/p u(s) ds pup/q } q/p f(t) dt) () d which mens tht C p,q (u, v) (p/q C p (u p/q, v). (b) For ny α >, B p,q (u q/p, v) { [( ) 1/q ( 1/q ] rv() } 1/r u q/p (t) dt v(t) dt) 1 p 1 p d { [( p d ( ) ) q/p 1/q u(s) ds dt q dt t ( 1/q ] r } 1/r v(t) dt) 1 p v() 1 p d ( ) 1/q p { [( ( 1/q ] rv() } 1/r u(s) ds v(t) dt) 1 p 1 p d q ( ) 1/q p { α [( ( 1/q ] rv() } 1/r u(s) ds v(t) dt) 1 p 1 p d q ( ) 1/q p ( q α ( ) 1/q p ( ) { 1/p α [( ) 1+r/q ] } p 1/r u(s) ds d v(t) 1 p dt q r α ( ) 1/q ( ) p p 1/r ( ( α ) (1+r/q )(1/r) u(s) ds v(t) 1 p dt q r α ( ) 1/q ( ) p p 1/r ( ( α u(s) ds v(t) 1 p dt, q r α { α ( ) r/q ( )} 1/r u(s) ds v(t) 1 p dt d v(t) 1 p dt

14 H. Heinig nd L. Mligrnd which gives ( ( ) 1/q ( ) p p 1/r B p,q (u q/p, v) A p (u, v). q r Similrly, using Theorem 2.1(b) with γ q/p 1 nd f replced by ( f(t) dt)p, nd g replced by u(t) dt, we hve ) p ( ) p [ ] f(t) dt d f(t) dt d u(s) ds { ( ) q [ ( f(t) dt d ) q/p ]} p/q u(s) ds (p/q) p/q{ ( quq/p } p/q ( f(t) dt) () d C which mens tht C p (u, v) (p/q) 1/q C p,q (u q/p, v). ) v()f() p d, R e m r k 2.4. (i) Artol, Tlenti, Tomselli nd Muckenhoupt (cf. [32]) proved tht Hrdy s inequlity (2.3) with p q 1 holds if nd only if A p (u, v) <. Then Brdley [1], Kokilshvili nd Mz y [31] etended this result to the cse 1 p q. The cse 1 q < p ws proved by Mz y [31] nd Swyer [38], nd the cse < q < 1 < p by Sinnmon [4] (see lso [38]). For more informtion we refer to [36]. Our Proposition 2.3() shows tht if the Hrdy inequlity (2.3) is proved in the inde rnge p q 1, then it holds for 1 p < q <. In fct, if A p,q (u, v) <, then by Proposition 2.3(), C p,q (u, v) (p/q C p (u p/q, v) A p (u p/q, v) <. The best reltionship between the constnts C p,q (u, v) nd A p,q (u, v) when 1 < p < q < ws found by Mnkov [3]. (ii) Using Theorem 2.1() one obtins the sme implictions for the dul Hrdy opertor f(t) dt. If, for < p < nd for weight function v, we define the Lorentz spce Λ p (v) s the spce generted by the qusi-norm ( f Λp (v) f () p v() d, where f denotes the decresing rerrngement of f, then the imbedding Λ p (v) Λ q (u) is equivlent to corresponding weighted integrl inequlity for decresing functions. Such results re due to Swyer [39] nd Stepnov [44]. Heinig nd Stepnov [24, Th. 2.1(i)] proved similr result for incresing functions. Theorem 2.1 my lso be pplied in nturl wy in the proof of these results.

Inequlities for monotone nd concve functions 141 Proposition 2.5. Let < p q <. () The inequlity ( ) 1/q ( (2.4) f() q d C v()f() p d holds for ll f if nd only if (2.5) ( t ) 1/q ( t d C (b) The inequlity ( ) 1/q ( (2.6) f() q d D v() d t >. v()f() p d holds for ll f if nd only if ( ) 1/q ( (2.7) d D t t v() d t >. P r o o f. () Tht (2.4) implies (2.5) follows on tking f() χ [,t] (), t >, in (2.4). (2.5) (2.4). It suffices to prove this impliction for those functions f for which supp f (, N] (, ) nd v()f() p d <. Then stndrd limiting rguments give the result. If supp f (, N] (, ) nd v()f() p d <, then integrtion by prts, ssumption (2.4), Theorem 2.1(b) with γ p/q nd gin integrtion by prts yield { } p/q { f() q d [ f() q d ]} p/q u(t) dt {f() q u(t) dt + [ ] } p/q u(t) dt d( f() q ) { [ ] } p/q u(t) dt d( f() q ) { C p C q[ [ q/p } p/q v(t) dt] d( f() q ) ] v(t) dt d( f() p ) C p (b) The necessity follows t once with f() χ [t, ) (), t >. v()f() p d.

142 H. Heinig nd L. Mligrnd Conversely, similrly to () we cn ssume tht supp f [δ, ) (, ) nd v()f() p d <. Using integrtion by prts, ssumption (2.6), Theorem 2.1() with γ p/q nd gin integrtion by prts we obtin { } p/q { [ ]} p/q f() q d f() q d u(t) dt { f() q u(t) dt [ ] } p/q + u(t) dt d[f() q ] { ( ) } p/q u(t) dt d[f() q ] { D q( ( D p nd inequlity (2.6) is proved. q/p } p/q v(t) dt) d[f() q ] ) v(t) dt d[f() p ] D p v()f() p d, 3. Shrp weighted inequlities for monotone functions. We now prove weighted inequlities of the form T 1 f q,u C T 2 f p,v, where T 1 nd T 2 re positive integrl opertors on monotone functions. These estimtes re essentil in proving the weighted Berwld Fvrd inequlities given in the sequel. First we need the following known lemm (cf. Neugebuer [34, Lemms 2.1, 4.2] nd Crro Sori [15, Th. 2.1]): Lemm 3.1. Let k be loclly integrble on (, ) nd < r <. () If f on (, ), then f() r k() d r y r 1(f (b) If f on [, ), <, then f() r k() d r (c) If ϕ on (, ), then [ ] k() d ϕ(y) r dy ϕ(y) f() y r 1( [ 1 (y) f 1 (y) ) k() d dy. ) k() d dy. ] ϕ 1 (s) d(s r ) r ϕ 1 () k() d.

Inequlities for monotone nd concve functions 143 P r o o f. () By the Fubini theorem f() r k() d r r y r 1( y r 1(f {:f()>y} 1 (y) where the lst equlity follows since f is decresing. (b) In similr wy f() r k() d r ϕ(y) r f() y r 1( f() y r 1( ) k() d dy ) k() d dy, { [,):f()>y} f 1 (y) ) k() d dy. ) k() d dy (c) Interchnging the order of integrtion we hve [ ] [ ] k() d ϕ(y) r dy ϕ(y) r dy k() d. ϕ 1 () The following equlity is geometriclly obvious (cf. O Neil [35, p. 13] with r 1): r ϕ 1 (t 1/r ) dt r ϕ 1 () + ϕ 1 () ϕ(y) r dy. Applying these equlities together with r ϕ 1 (t 1/r ) dt ϕ 1 (s) d(s r ) we obtin the proof of (c). The proof of prt (b) of Theorem 3.2 is tken from [16]. We repet it here, with n emphsis on the constnt which in our cse plys n importnt role. (3.1) Theorem 3.2. Suppose k 1 (, t) nd k 2 (, t) re two non-negtive kernels. () Let < p 1 q. Then ( [ ] q ) 1/q k 1 (, t)f(t) dt d ( C [ v() ] p k 2 (, t)f(t) dt d

144 H. Heinig nd L. Mligrnd holds for ll f on [, ) if nd only if ( [ α ] q ) 1/q (3.2) k 1 (, t) dt d ( C [ α ] p v() k 2 (, t) dt d holds for ll α >. (b) Let < p 1 nd p q. Then (3.3) ( [ ] q ) 1/q ( k 1 (, t)f(t) dt d C v()f() p d holds for ll f on [, ) if nd only if ( [ α (3.4) ] q ) 1/q ( α k 1 (, t) dt d C v() d holds for ll α >. (c) Let 1 p q. Then ( ) 1/q ( (3.5) f(t) q d C [ v() ] p k 2 (, t)f(t) dt d holds for ll f on [, ) if nd only if (3.6) ( α holds for ll α >. ) 1/q ( d C [ α v() ] p k 2 (, t) dt d P r o o f. The necessity prts of (), (b) nd (c) follow on tking f(t) χ [,α] (t), α > fied. () To prove sufficiency we pply Lemm 3.1() (with r 1), Minkowski s inequlity twice nd (3.2) to obtin ( [ ] q ) 1/q k 1 (, t)f(t) dt d { [ ( f 1 (y) ( f 1 (y) ) ] q } 1/q k 1 (, t) dt dy d ) k 1 (, t) dt dy q,u f 1 (y) k 1 (, t) dt dy q,u

{ Inequlities for monotone nd concve functions 145 { C [ { C { C { C ( C [ v() ( f 1 (y) ( f 1 (y) [ v() [ v() ( f 1 (y) ( f 1 (y) ) q ] 1/q } k 1 (, t) dt d dy ) p ] 1/p } k 2 (, t) dt d dy ) p 1/p,v} 1/p k 2 (, t) dt d ) p } 1/p k 2 (, t) dt 1/p,v d ( f 1 (y) ) ] p } 1/p k 2 (, t) dt dy d k 2 (, t)f(t) dt] p d. The lst equlity follows from Lemm 3.1() with r 1. (b) By Lemm 3.1() (with r 1), ( ( ) q ) 1/q k 1 (, t)f(t) dt d { [ ( f 1 (y) ) q } 1/q. k 1 (, t) dt dy] d Assume first tht q 1. Since the innermost integrl is decresing function of y, Theorem 2.1() pplies with γ q (cf. Remrk 2.2) so tht the lst integrl is less thn or equl to { q [ y q 1(f { q 1 (y) { C q ) q ] } 1/q k 1 (, t) dt dy d y q 1[ y q 1(f 1 (y) ( f 1 (y) ) q ] } 1/q k 1 (, t) dt d dy v() d) q/p dy } 1/q, where the lst inequlity follows from (3.4). Since p/q 1 we pply gin

146 H. Heinig nd L. Mligrnd Theorem 2.1() with γ p/q to see tht the lst epression is not lrger thn { C p y p 1(f 1 (y) ) } 1/p ( v() d dy C Here the equlity follows gin from Lemm 3.1(). If q > 1, let g f q. Then by Lemm 3.1(), k 1 (, t)f(t) dt 1 q k 1 (, t)g(t) 1/q dt y 1/q 1(g 1 (y) Minkowski s inequlity nd (3.4) now shows tht ( ( ) q ) 1/q k 1 (, t)f(t) dt d 1 q { 1 q 1 q 1 q C q [ y 1/q 1(g y 1/q 1(g 1 (y) 1 (y) ) k 1 (, t) dt dy g 1 (y) y 1/q 1 k 1 (, t) dt dy q,u y 1/q 1[ y 1/q 1(g 1 (y) ( g 1 (y) v() d dy. f() p v() d. ) k 1 (, t) dt dy. ) ] q } 1/q k 1 (, t) dt dy d q,u k 1 (, t) dt) q d ] 1/q dy But p 1, so gin Theorem 2.1() with γ p pplies nd hence the lst epression is not lrger thn { p C q y p/q 1(g 1 (y) { C ) } 1/p v() d dy } 1/p { g() p/q v() d C f() p v() d} 1/p.

Inequlities for monotone nd concve functions 147 (c) Let g f q. Then by Lemm 3.1(), [ ] p/q [ f() q d ( g 1 (y) ) d dy] p/q. Theorem 2.1() with γ p/q 1 nd (3.6) show tht this is not lrger thn p q C p p q C p y p/q 1(g [ p q 1 (y) y p/q 1[ d) p/q dy y p/q 1(g ( g 1 (y) 1 (y) ) pv() ] k 2 (, t) dt d dy ) ] p k 2 (, t) dt dy v() d C p nd gin by Theorem 2.1() with γ 1/p 1 we hve A() p q [ [ y p/q 1(g ( g 1 (y) ( g 1 (y) 1 (y) Thus from the bove nd Lemm 3.1(), [ ] p/q f() q d C p k 2 (, t) dt) p dy k 2 (, t) dt) p d(y p/q ) ) ] p k 2 (, t) dt d(y 1/q ) 1 1 (y) ) p q y1/q 1(g k 2 (, t) dt dy]. C p [ [ A()v() d g 1 1 (y) p q y1/q 1 k 2 (, t) dt dy] v() d g(t) 1/q k 2 (, t) dt] pv() d C p [ f(t)k 2 (, t) dt] pv() d. This proves the theorem.

148 H. Heinig nd L. Mligrnd The net ssertions for incresing functions on [, ) will be proved by similr rguments. Theorem 3.3. Suppose k 1 (, t) nd k 2 (, t) re two non-negtive kernels. () Let < p 1 q. Then ] q ) 1/q (3.7) k 1 (, t)f(t) dt d ( [ ( [ ] p D v() k 2 (, t)f(t) dt d holds for ll f on [, ), <, if nd only if ( [ ] q ) 1/q (3.8) k 1 (, t) dt d α ( [ ] p D v() k 2 (, t) dt d α holds for ll α (, ). (b) Let < p 1 nd p q. Then ( [ ] q ) 1/q ( (3.9) k 1 (, t)f(t) dt d D v()f() p d holds for ll f on [, ), <, if nd only if ( [ ] q ) 1/q ( (3.1) k 1 (, t) dt d D α α v() d holds for ll α (, ). (c) Let 1 p q. Then ( ) 1/q ( (3.11) f() q d D [ ] p v() k 2 (, t)f(t) dt d holds for ll f on [, ), <, if nd only if ( ) 1/q ( [ ] p (3.12) d D v() k 2 (, t) dt d holds for ll α (, ). α P r o o f. The necessity prts of (), (b) nd (c) follow on tking f(t) χ [α,) (t), α > fied. Sufficiency. () The proof is similr to tht of Theorem 3.2(), we need only use Lemm 3.1(b) twice (with r 1): α

Inequlities for monotone nd concve functions 149 { [ ] q } 1/q k i (, t)f(t) dt d { [ f() ( f 1 (y) ) q } 1/q, k i (, t) dt dy] d i 1, 2. (b) Similrly, by Lemm 3.1(b) (with r 1), { [ ] q } 1/q k 1 (, t)f(t) dt d { [ f() ( f 1 (y) ) q } 1/q. k 1 (, t) dt dy] d Assume gin, first, tht q 1. Since the inner integrl is decresing function of y we my pply Lemm 2.1() with γ q so the lst term is not lrger thn { q [ f() y q 1( f 1 (y) { f() q { f() D q y q 1[ ) q ] } 1/q k 1 (, t) dt dy d y q 1[ ( f 1 (y) f 1 (y) ) q ] } 1/q k 1 (, t) dt d dy v() d] q/p dy } 1/q, where the lst inequlity follows from (3.1). But since p/q 1 we pply Theorem 2.1() with γ p/q to see tht the lst epression is not lrger thn { f() D p y p 1[ f 1 (y) If q > 1, let g f q. Then by Lemm 3.1(b), k 1 (, t)f(t) dt ] } 1/p ( 1/p. v() d dy D f() p v() d) 1 q k 1 (, t)g(t) 1/q dt g() y 1/q 1( g 1 (y) ) k 1 (, t) dt dy.

15 H. Heinig nd L. Mligrnd By Minkowski s inequlity nd (3.1), { [ ] q } 1/q k 1 (, t)f(t) dt d 1 q 1 q D q { [ g() y 1/q 1( { g() y 1/q 1[ g() y 1/q 1( g 1 (y) g 1 (y) ( g 1 (y) v() d dy. ) ] q } 1/q k 1 (, t) dt dy d ) q ] 1/q } k 1 (, t) dt d dy But since p 1, Theorem 2.1 pplies gin nd hence by Lemm 3.1(b) the lst term is not lrger thn { g() p D y p/q 1( ) } 1/p v() d dy q D g 1 (y) ( ( 1/p. g() p/q v() d D f() p v() d) (c) The proof is similr to tht of Theorem 3.2(c) nd therefore omitted. R e m r k 3.4. Theorems 3.2(b) nd 3.3(b) were proved in different wy by Stepnov [43], Mysnikov Persson Stepnov [33] nd, in the cse of < p q 1, by Li [27]. Our method of proof is tken from the pper by Crro Sori [15] nd Li [27]. Theorem 3.2(c) ws lso proved by Mysnikov Persson Stepnov [33] nd Li [27]. The choice of k 1 (, t) χ [,] (t) nd k 1 (, t) χ [,] (t), in Theorem 3.2(b), respectively Theorem 3.3(b), gives Corollry 3.5. Let < p 1 nd p q. () The inequlity ( ( ) q ) 1/q ( f(t) dt d C f() p v() d holds for ll f if nd only if for every α >, ( ) 1/q ( α 1/p. (min{, α}) q d C v() d)

Inequlities for monotone nd concve functions 151 (b) For <, the inequlity ( ( ) q ) 1/q ( f(t) dt d D f() p v() d holds for ll f if nd only if for every α (, ), ( ) 1/q ( (min{, α}) q d D α v() d. R e m r k 3.6. If p q 1 nd u v, then the result corresponding to Corollry 3.5() ws proved by Neugebuer [34, Th. 2.2]. In the cse when both the integrl opertors re equl nd of the form f(t) dt, Theorem 3.2() etends to the rnge p q 1. Theorem 3.7. If either < p 1 q < or 1 p q <, then ( ( ) q ) 1/q ( ( ) p (3.13) f(t) dt d C v() f(t) dt d holds for ll f if nd only if ( ) 1/q ( (3.14) (min{, α}) q d C v()(min{, α}) p d holds for every α >. P r o o f. If < p 1 q, then the proof follows immeditely from Theorem 3.2() by tking k 1 (, t) k 2 (, t) χ [,] (t). If 1 p q, then (3.14) with α ϕ(y) mens (min{, ϕ(y)}) p d C p v()(min{, ϕ(y)}) p d. Integrting from to with respect to y nd using Lemm 3.1() nd (c), we obtin ( ) (min{, ϕ(y)}) p d dy [ ϕ(y) ] p d + ϕ(y) p d dy ϕ 1 () p d + [ ϕ(y) [ ] ϕ 1 (s) d(s p ) d. ] ϕ 1 (s) d(s p ) p ϕ 1 () d

152 H. Heinig nd L. Mligrnd Similrly for the epression with weight v we obtin from (3.14), [ ] [ ] ϕ 1 (s) d(s p ) d C p ϕ 1 (s) d(s p ) v() d. Tking ϕ 1 (s) f(s)(s 1 s f(t) dt)p 1 we hve ϕ 1 (s) d(s p ) ( f(t) dt)p nd so [ ] p f(t) dt d C p [ f(t) dt] pv() d. R e m r k 3.8. The sme result with different constnt in the rnge < p q < nd q 1 ws proved by Stepnov [43, Th. 3.3]. 4. Shrp weighted inequlities for concve functions. Consider the Green kernel { (1 t) if t 1, (4.1) K(, t) t(1 ) if t 1. Then the function (4.2) f() 1 K(, t)g(t) dt, where g rnges over ll non-negtive functions in L 1 [, 1], constitute dense subset of non-negtive concve functions on [, 1] (cf. [3], [5]). This fct is used in the proof of the following theorem: (4.3) Theorem 4.1. Suppose either < p 1 q or 1 p q. Then ( 1 ) 1/q ( 1 f() q d C v()f() p d holds for ny concve function f on [, 1] if nd only if (4.4) ( 1 ) 1/q ( 1 K(, α) q d C v()k(, α) p d holds for ll α (, 1), where K(, α) min{, α} min{1, 1 α}. P r o o f. The necessity follows t once on tking f() K(, α), α (, 1) fied, in (4.3). To prove sufficiency in the cse of < p 1 q, it suffices to prove (4.3) for those functions f hving representtion (4.2) nd then stndrd limiting rguments give the generl result. We my lso ssume tht the left side of (4.3) is finite on tking first suitble dense subset. This restriction gin cn be removed by limiting procedures.

Inequlities for monotone nd concve functions 153 Assuming this, we use Hölder s inequlity, (4.4) nd Minkowski s inequlity to obtin f q q,u 1 1 1 C [ 1 f() q 1 [ 1 g(t) [ 1 g(t) 1 g(t) ( 1 C ( 1 C ( 1 C ( 1 C [ 1 ] K(, t)g(t) dt d ] K(, t)f() q 1 d dt ] 1/q [ 1 K(, t) q d ] 1/p K(, t) p v() d dt f q/q q,u ) K(, t) p 1/p f q 1 v() d 1/p,g ] 1/q f() q d dt q,u f K(, t) p q 1 v() 1/p,g d q,u ( 1 ) p f K(, t)v( g(t) dt d q 1 q,u f f() p q 1 v() d q,u nd the result follows on division by f q 1 q,u. C f p,v f q 1 q,u Sufficiency in the cse 1 p q. For p let the differentil opertor on C 2 (, 1) be given by L p [y] (1 ) p d { p+1 (1 ) 1 p d } d d ( p y) (1 )y + (p 1)(2 1)y + p(1 p)y nd consider the Rdon-concve functions of order p, R p Q p (cf. [17], [37]), i.e. the closure in the topology of loclly uniform convergence on (, 1) of Q p {f C 2 [, 1] : L p [f], f() f(1) }. For every continuous function g : [, 1] R the boundry vlue problem L p [f] g, f() f(1),

154 H. Heinig nd L. Mligrnd hs solution where G p (, t) 1 p f() 1 G p (, t)g(t) dt, { p t p if < < t < 1, (1 ) p (1 t) p if < t < 1. Our result now follows from the following two fcts (cf. lso [17] nd [37]): 1. If 1 p q, (4.4) holds nd f R p, then ( 1 ) p/q 1 f() q/p d C In fct (nd similrly to the bove), f q/p 1 q/p,u 1 1 1 p 1 p [ 1 f() q/p 1 [ 1 g(t) [ 1 g(t) 1 1 g(t) [ 1 ] G p (, t)g(t) dt d ] G p (, t)f() q/p 1 d dt ] p/q [ 1 G p (, t) q/p d f()v() d. ] p/q q/p 1 qg q (, t) d dt f q/p,u g(t)t p (1 t) p[ 1 1 1 p Cp g(t)t p (1 t) p[ 1 C p 1 C p 1 C p 1 [ 1 g(t) [ 1 f() q/p d] 1 p/q dt ] p/q K(, t) q q/p 1 d dt f q/p,u ] G p (, t)v() d G p (, t)g(t) dt f()v() d f q/p 1 q/p,u. ] K(, t) p v() d dt f q/p 1 q/p,u dt f q/p 1 q/p,u ] v() d f q/p 1 q/p,u

Inequlities for monotone nd concve functions 155 2. If p 1 nd f R 1, i.e. f is positive concve function on [, 1], then f p R p. It is enough to prove tht if y Q 1 (i.e. y nd y() y(1) ), then y p Q p. Since L p [y p ] (1 )p[(p 1)y p 2 (y ) 2 + y p 1 y ] + (p 1)(2 1)py p 1 y + p(1 p)y p (p 1)y p 1 L p [y] + p(p 1)y p 2 (y y)[(1 )y + y] nd y () y() ty (t) dt, (1 )y () + y() 1 (t 1)y (t) dt it follows tht L p [y p ] nd so y p Q p. R e m r k 4.2. (i) Theorem 4.1 in the cse p 1 ws lso proved by Mligrnd Pečrić Persson [29, Th. 3]. In the cse < p 1 q, Theorem 4.1 is still true for ll functions which hve representtion (4.2) with the non-negtive kernel K(, t). In prticulr, if { 1 if t 1, K(, t) if < t 1, then f() g(t) dt is incresing on [, 1]. Similrly, if { if t 1, K(, t) 1 if < t 1, then f() 1 g(t) dt is decresing on [, 1]. Therefore Theorem 4.1 lso gives the proof of Proposition 2.5 but only in the inde rnge < p 1 q. (ii) If either < p 1 q or 1 p q nd v() 1, then h(t) [ 1 [ t [ t ] 1/q [ 1 K(, t) q d q (1 t) q d + 1 p (1 t) p d + (p + 1 (q + 1) 1/q, t ] 1/p K(, t) p d ] 1/q t q (1 ) q d 1 t ] 1/p t p (1 ) p d which is the quotient, i.e. the constnt C of (4.4), nd hence Theorem 4.1 gives the Fvrd Berwld inequlity f q (p+1 (q+1) 1/q f p for ny positive concve function f on [, 1]. Berwld [8] proved this inequlity in the full rnge < p q. In prticulr, if q 1 then f 1 1 2 (p+1)1/p f p, < p 1. As p this inequlity becomes the reverse Jensen inequlity given in the introduction.

156 H. Heinig nd L. Mligrnd (iii) If v() α, α nd p 1 in Theorem 4.1, then the function h(t) [ 1 ] 1/q /[ 1 K(, t) q α d [(1 t) q t α+1 1 q + α + 1 + (1 t)t α+1 α + 2 t + 1 tα+1 α + 1 ] K(, t) α d (1 ) q α d 1 tα+2 α + 2 ] 1/q cn be shown to be decresing on (, 1). Hence h(t) h() nd therefore the constnt of (4.3) is ( 1 (α + 1)(α + 2) (1 ) q α d) 1/q (α + 1)(α + 2)B(q + 1, α + 1) 1/q, B being the Bet function. This is result of Brnrd Wells [3, Th. 1]. We cn pply Corollry 3.5 nd Theorem 3.7 to prove Fvrd Berwld inequlities for incresing nd decresing concve functions. Note tht if f is concve on [, ) then f is necessrily incresing. R e m r k 4.3. (i) From Theorem 3.7 it follows t once tht if either < p q or 1 p q <, then f q,u C f p,v holds for ny concve incresing function f on [, ) if nd only if (3.14) holds for ll α >. (ii) If 1 p q <, then it follows from Corollry 3.5 nd Theorem 2.1() tht ( 1 ) 1/q 1 f() q d D v()f() d holds for ny concve decresing function f on [, 1] if nd only if ( 1 ) 1/q 1 [min(1, 1 α)] q d D is stisfied for ll α (, 1). We conclude this section with three emples. v() min(1, 1 α) d

Inequlities for monotone nd concve functions 157 Emple 4.4. If f is concve incresing on [, 1], then by Remrk 4.3(i) with p 1, q 2 nd v() α, α >, we obtin (4.5) ( 1 ) 1/2 1 f() 2 α α + 2 d α + 3 f() α d. The constnt (α + 2)/ α + 3 is best possible, s my be seen on tking f() in (4.5). Emple 4.5. If f is concve on [, ), then for t >, ( ) 1/q (4.6) e t f() q d t 1 1/q Γ (q + 1) 1/q e t f() d, 1 < q <. The constnt t 1 1/q Γ (q + 1) 1/q is best possible, s my be seen on tking f() in (4.6). In order to obtin the constnt in (4.6) it clerly suffices to tke t 1. By Remrk 4.3(i) with p 1, we must show tht sup α> {( ) 1/q / } e [min(, α)] q d e min(, α) d {( α sup e q d + α q e α) 1/q/ } (1 e α ) Γ (q + 1) 1/q. α> But if h(α) ( α e q d + α q e α )/(1 e α ) q, then h is incresing on (, ) since h (α) (1 e α ) q{ qα q 1 e α + qe α (1 e α ) q 1[ α nd the bove supremum equls lim α h(α) 1/q Γ (q + 1) 1/q. Emple 4.6. If f is concve on [, ), then ( ) 1/q (4.7) e 2 f() q d e q d + α q e α]} >, ( e q d) 1/q ( q + 1 2 1 1/q Γ 2 ) 1/q e 2 f() d, 1 < q <,

158 H. Heinig nd L. Mligrnd nd the constnt 2 1 1/q Γ ((q + 1)/2) 1/q is shrp, s my be seen on tking f() in (4.7). Agin by Remrk 4.3(i) we only need to show tht {( ) 1/q / } sup e 2 [min(, α)] q d e 2 min(, α) d α> {( α sup e 2 q d + α q α> ( α e 2 d + α For tht it suffices to show tht k(α) ( α e 2 q d + α q α α α ) 1/q e 2 d ) 1 } e 2 d 2 1 1/q Γ )/( α e 2 d e 2 d + α ( ) 1/q q + 1. α 2 ) q e 2 d is n incresing function on (, ), for then the bove supremum is ( ) 1/q / ( ) 1/q q + 1 lim α k(α)1/q e 2 q d e 2 d 2 1 1/q Γ. 2 But ( α k (α) q e 2 d + α α [α q 1 α e 2 d nd hence the function k is incresing s required. ) q 1 ( ) e 2 d e 2 d α α ] e 2 q d > 5. Weighted reverse Hölder inequlities. Closely relted to the Fvrd Berwld inequlity is Grüss inequlity [21] (see lso [23]). It sserts tht the L 1 -norms of f nd g re dominted by the L 1 -norm of the product fg, where f, g re from some clss of functions. This inequlity together with Fvrd s gives reverse Hölder inequlity (cf. [3], [4], [6], [9], [26]). We now discuss weighted versions of Grüss inequlity nd lso give weight chrcteriztions for which such inequlities hold. In our first result we ssume w L 1 (I), where I R is n intervl, nd we let P be the clss of positive mesurble functions on I. Theorem 5.1. () If f 2,w C 2 f 1,w for ny f P, then (5.1) C 2 f 1,w g 1,w fg 1,w for ny f, g P, where C2 2/w(I) C2 2 nd w(i) w() d. I

Inequlities for monotone nd concve functions 159 (b) If f r,w C r f 1,w for every r > 1 nd f P, then (5.2) C p,q f p,w g q,w fg 1,w for 1 p, q <, f, g P nd C p,q C 2 /(C p C q ). P r o o f. () Let f 1,w g 1,w 1. Then f + g 2 2,w f 2 2,w + 2 fg 1,w + g 2 2,w C 2 2 f 2 1,w + 2 fg 1,w + C 2 2 g 2 1,w 2C 2 2 + 2 fg 1,w. On the other hnd, Hölder s inequlity shows tht so tht nd hence 2 f + g 1,w f + g 2,w w(i) 1/2, 2 fg 1,w f + g 2 2,w 2C 2 2 4 w(i) 2C2 2, fg 1,w ( ) 2 w(i) C2 2 f 1,w g 1,w. (b) The proof follows immeditely from () nd the ssumption. For concve functions the constnts given in Theorem 5.1 re not lwys shrp. We illustrte this fct by emples given net. Emple 5.2. () Let P {f : f is concve on [, 1]}. If w() α, α, then Remrk 4.2(iii) shows tht the constnt C r equls (α+1)(α+2) B(r + 1, α + 1) 1/r, where B is the Bet function. Since we hve C 2 [2(α + 1)(α + 2)/(α + 3)] 1/2, it follows tht C2 2(α + 1)/(α + 3) nd C p,q 2[(α + 1)(α + 2) 2 (α + 3)] 1 B(p + 1, α + 1) 1/p B(q + 1, α + 1) 1/q. In prticulr, C 2,2 (α + 2) 1 but it is known tht the best constnt is {(α + 1)/[2(α + 3)]} 1/2 (cf. Brnrd Wells [3]). If α the constnt C p,q 1 6 (p + 1)1/p (q + 1) 1/q is shrp (cf. Brnes [4]). (b) If P {f : f is concve on [, )} nd w() e 2, then Emple 4.6 shows tht C r 2 1 1/r Γ ((r + 1)/2) 1/r. Hence we hve C2 (4 π)π 1/2 > nd ( ) 1/p ( ) 1/q p + 1 q + 1 Cp,q (4 π)π 1/2 2 1/p+1/q 2 Γ Γ. 2 2 To obtin shrp constnts in inequlity (5.2) for specific clss of functions, sy concve functions, we require properties of these functions which re not s generl s those of P.

16 H. Heinig nd L. Mligrnd Theorem 5.3. Suppose f, g re non-negtive concve functions on [, 1] nd < r 1 p, q <. If u, v nd w re weight functions on [, 1], then (5.3) ( 1 holds if nd only if ( 1 f() p d ) 1/q v()g() q d ( 1 C ) 1/r w()[f()g()] r d [ 1 (5.4) C sup K(, t)p d] 1/p [ 1 K(, s)q v() d] 1/q s,t (,1) [ 1 [K(, t)k(, <, s)]r w() d] 1/r where { (1 y) if y 1, K(, y) y(1 ) if y 1. P r o o f. Necessity. Substitute f() K(, t) nd g() K(, s), where s, t (, 1) re rbitrry, into (5.3). Then the necessity follows. Sufficiency. As noted erlier, it suffices to prove (5.3) if f nd g hve the representtion f() 1 K(, t)f 1 (t) dt, g() 1 K(, s)g 1 (s) ds, where f 1, g 1 re some non-negtive functions in L 1 [, 1]. Applying Minkowski s inequlity twice nd (5.3) one obtins f p,u g q,v [ 1 [ 1 1 1 1 C ( 1 [ 1 f 1 (t) 1 1 ( 1 v() ) p ] 1/p K(, t)f 1 (t) dt d ) q ] 1/q K(, s)g 1 (s) ds d ] 1/p 1 K(, t) p d dt [ 1 f 1 (t)g 1 (s) [ 1 g 1 (s) ] 1/p [ 1 K(, t) p d dt { 1 f 1 (t)g 1 (s) [K(, t)k(, s)] r w() d} 1/r ds dt K(, s) q v() d] 1/q ds K(, s) q v() d] 1/qds dt

{ 1 C { 1 C C fg r,w. Inequlities for monotone nd concve functions 161 [ 1 w() [ 1 w() 1 ] r } 1/r f 1 (t)g 1 (s)k(, t)k(, s) ds dt d ] r [ 1 f 1 (t)k(, t) dt ] r } 1/r g 1 (s)k(, s) ds R e m r k 5.4. (i) The method of proof of Theorem 5.3 with the Green function K(, y) is usully clled the Bellmn or Bellmn Weinberger method. Bellmn proved such result in the cse of u v w 1, r 1 nd 1/p + 1/q 1 ( cf. [5], [6], [3], [26], [37], [46], [47]). This method ws lso used erlier by Bückner [11] who proved the cse u v w, r 1 nd p q 2. (ii) Computtions of the supremum (5.4) or (5.6) re often quite tedious. If v() w() 1 in Theorem 5.3, then the supremum over < s t < 1 of H(s, t) ( 1 ( 1 K(, t) p d ) 1/q /( 1 K(, s) q d ) 1/r [K(, t)k(, s)] r d [ (1 s) (p + 1) 1/p (q + 1) 1/q r s r+1 t(1 s) 2r + 1 t + r (1 ) r d + tr (1 t) r+1 ] 1/r 2r + 1 s is not esy to compute (cf. [3], [6], [25], [37], [46] nd [47]). (iii) If f nd g re non-negtive concve incresing functions on [, ), <, then the result corresponding to Theorem 5.3 is the following: If < r 1 p, q <, then ( ( ) 1/q (5.5) f() p d v()g() q d holds if nd only if ( ) 1/r D w()[f()g()] r d [ (5.6) D sup K(, t)p d] 1/p [ K(, s)q v() d] 1/q s,t (,) [ [K(, t)k(, <, s)]r w() d] 1/r

162 H. Heinig nd L. Mligrnd where K(, y) { if y <, y if y <. Finlly, if for p, q 1 nd α, H p,q,α denotes the best constnt H in the reverse Hölder inequlity ( 1 H ( 1 f() p α d ) 1/q 1 g() q α d f()g() α d, f, g concve positive functions on [, 1], then the known results re the following: H 2,2, 1/2 (Frnk Pick [19]), H 1,1, 2/3 (Grüss [21]), H p,p, 1 6 (p + 1)1/p (p + 1 (Bellmn [6]), H p,q, 1 6 (p + 1)1/p (q + 1) 1/q (Brnes [4]), α + 1 H 2,2,α 2(α + 2) (Brnrd Wells [3]), H p,p,α min{v α (p), V α (p ), W α (p)}/[(α + 2)(α + 3)] (Wng Chen [47]), where V α (p) (p + 1 + α B(p + 1, α + 1) 1/p, W α (p) 2/[(α + 1)B(p + 1, α + 1 B(p + 1, α + 1 ]. Also if H + p,q,α is the corresponding constnt for incresing concve functions f, g on [, 1], then nd H + 2,2, 3/2 (Krft, Bückner [11]) H + p,q, min { 1 2 (p+1)1/p, 1 2 (q+1)1/q, 1 3 (p+1)1/p (q+1) 1/q} (Petschke [37]). Other results my lso be obtined from Remrk 5.4(iii); in prticulr, we get (α + 1)(α + 3) H 2,2,α +. α + 2 Acknowledgements. This reserch ws supported in prt by NSERC grnt A-4837 nd Luleå University. The first nmed uthor pprecites the kind hospitlity of the Deprtment of Mthemtics, Luleå University, Luleå, where the reserch ws lmost finished in Jnury Februry 1993.

Inequlities for monotone nd concve functions 163 Both uthors re gretful to the referee for his vluble comments, nd especilly for his observtion tht the proof of Proposition 2.3(b) etends to the inde rnge < q < 1. References [1] K. F. Andersen, Weighted generlized Hrdy inequlities for nonincresing functions, Cnd. J. Mth. 43 (1991), 1121 1135. [2] M. Ariño nd B. Muckenhoupt, Miml functions on clssicl spces nd Hrdy s inequlity with weights for nonincresing functions, Trns. Amer. Mth. Soc. 32 (199), 727 735. [3] R. W. B r n r d nd J. W e l l s, Weighted inverse Hölder inequlities, J. Mth. Anl. Appl. 147 (199), 198 213. [4] D. C. Brnes, Some complements of Hölder s inequlity, ibid. 26 (1969), 82 87. [5] E. F. Beckenbch nd R. Bellmn, Inequlities, Springer, New York, 1983. [6] R. Bellmn, Converses of Schwrz s inequlity, Duke Mth. J. 23 (1956), 429 434. [7] J. Bergh, V. Burenkov nd L. E. Persson, Best constnts in reversed Hrdy s inequlities for qusimonotone functions, Act Sci. Mth. (Szeged) 59 (1994), 221 239. [8] L. Berwld, Verllgemeinerung eines Mittelwertstzes von J. Fvrd, für positive konkve Funktionen, Act Mth. 79 (1947), 17 37. [9] C. Borell, Inverse Hölder inequlities in one nd severl dimensions, J. Mth. Anl. Appl. 41 (1973), 3 312. [1] J. Brdley, Hrdy inequlities with mied norms, Cnd. Mth. Bull. 21 (1978), 45 48. [11] H. Bückner, Untere Schrnken für sklre Produkte von Vektoren und für nloge Integrlusdrücke, Ann. Mt. Pur Appl. 28 (1949), 237 261. [12] V. I. Burenkov, On the best constnt in Hrdy s inequlity for < p < 1, Trudy Mt. Inst. Steklov. 194 (1992), 58 62 (in Russin). [13] R. J. Bushell nd W. Okrsiński, Nonliner Volterr equtions with convolution kernel, J. London Mth. Soc. 41 (199), 53 51. [14] A. P. Clderón nd R. Scott, Sobolev type inequlities for p >, Studi Mth. 62 (1978), 75 92. [15] M. J. Crro nd J. Sori, Weighted Lorentz spces nd the Hrdy opertor, J. Funct. Anl. 112 (1993), 48 494. [16],, Boundedness of some liner opertors, Cnd. J. Mth. 45 (1993), 1155 1166. [17] A. Clusing, Disconjugcy nd integrl inequlities, Trns. Amer. Mth. Soc. 26 (198), 293 37. [18] J. Fvrd, Sur les vleurs moyennes, Bull. Sci. Mth. 57 (1933), 54 64. [19] P. Frnk und G. Pick, Distnzschätzungen im Funktionenrum. I, Mth. Ann. 76 (1915), 354 375. [2] A. Grcí del Amo, On reverse Hrdy s inequlity, Collect. Mth. 44 (1993), 115 123. [21] G. Grüss, Über ds Mimum des bsoluten Betrges von (1/(b )) b f()g() d (1/(b ) 2 ) b f() d b g() d, Mth. Z. 39 (1935), 215 226.

164 H. Heinig nd L. Mligrnd [22] G. H. Hrdy, J. E. Littlewood nd G. Póly, Inequlities, Cmbridge Univ. Press, 1952. [23] H. Heinig nd L. Mligrnd, Chebyshev inequlity in function spces, Rel Anl. Echnge 17 (1991 92), 211 247. [24] H. Heinig nd V. D. Stepnov, Weighted Hrdy inequlities for incresing functions, Cnd. J. Mth. 45 (1993), 14 116. [25] R. A. Hunt, On L(p, q) spces, Enseign. Mth. 12 (1966), 249 276. [26] S. Krlin nd W. J. Studden, Tchebycheff Systems: with Applictions in Anlysis nd Sttistics, Pure Appl. Mth. 15, Wiley, New York, 1966. [27] S. L i, Weighted norm inequlities for generl opertors on monotone functions, Trns. Amer. Mth. Soc. 34 (1993), 811 836. [28] G. G. Lorentz, Some new functionl spces, Ann. of Mth. 51 (195), 37 55. [29] L. Mligrnd, J. Pečrić nd L. E. Persson, Weighted Fvrd nd Berwld inequlities, J. Mth. Anl. Appl. 19 (1995), 248 262. [3] V. M. M n k o v, On the best constnt in weighted inequlities for Riemnn Liouville integrls, Bull. London Mth. Soc. 24 (1991), 442 448. [31] W. G. Mzj [V. G. Mz y], Einbettungssätze für Sobolewsche Räume, Teil 1, Teubner, Leipzig, 1979. [32] B. M u c k e n h o u p t, Hrdy s inequlity with weights, Studi Mth. 44 (1972), 31 38. [33] E. Mysnikov, L. E. Persson nd V. Stepnov, On the best constnts in certin integrl inequlities for monotone functions, Act Sci. Mth. (Szeged) 59 (1994), 613 624. [34] C. J. Neugebuer, Weighted norm inequlities for verging opertors of monotone functions, Publ. Mt. 35 (1991), 429 447. [35] R. O Neil, Convolution opertors nd L(p, q) spces, Duke Mth. J. 3 (1963), 129 142. [36] B. Opic nd A. Kufner, Hrdy-Type Inequlities, Pitmn Res. Notes in Mth. Ser. 219, Longmn, 199. [37] M. P e t s c h k e, Etremlstrhlen konveer Kegel und komplementäre Ungleichungen, Ph.D. Disserttion, Drmstdt, 1989, 16 pp. [38] E. Swyer, Weighted Lebesgue nd Lorentz norm inequlities, Trns. Amer. Mth. Soc. 281 (1984), 329 337. [39], Boundedness of clssicl opertors on clssicl Lorentz spces, Studi Mth. 96 (199), 145 158. [4] G. Sinnmon, Weighted Hrdy nd Opil-type inequlities, J. Mth. Anl. Appl. 16 (1991), 434 445. [41] E. Stein nd G. Weiss, Introduction to Fourier Anlysis on Eucliden Spces, Princeton Univ. Press, Princeton, N.J., 1971. [42] V. D. Stepnov, On boundedness of liner integrl opertors on the clss of monotone functions, Sibirsk. Mt. Zh. 32 (1991), 222 224 (in Russin). [43], Integrl opertors on the cone of monotone functions, J. London Mth. Soc. 48 (1994), 465 487. [44], Weighted Hrdy s inequlity for nonincresing functions, Trns. Amer. Mth. Soc. 338 (1993), 173 186. [45] W. Wlter nd V. Weckesser, An integrl inequlity of convolution type, Aequtiones Mth. 46 (1993), 212 219. [46] C.-L. Wng, An etension of Bellmn inequlity, Utilits Mth. 8 (1975), 251 256.

Inequlities for monotone nd concve functions 165 [47] H.-T. Wng nd S.-Y. Chen, Inverse Hölder inequlities with weight t α, J. Mth. Anl. Appl. 176 (1993), 92 17. DEPARTMENT OF MATHEMATICS MCMASTER UNIVERSITY HAMILTON, ONTARIO L8S 4K1 CANADA E-mil: HEINIG@MCMAIL.CIS.MCMASTER.CA DEPARTMENT OF MATHEMATICS LULEÅ UNIVERSITY S-971 87 LULEÅ, SWEDEN E-mil: LECH@SM.LUTH.SE Received September 27, 1994 (3342) Revised version Mrch 2, 1995