Journal of Inequalities in Pure and Applied Mathematics http://jipam.vu.edu.au/ Volume 7, Issue 2, Article 76, 2006 FOUR INEQUALITIES SIMILAR TO HARDY-HILBERT’S INTEGRAL INEQUALITY W.T. SULAIMAN C OLLEGE OF C OMPUTER S CIENCES AND M ATHEMATICS U NIVERSITY OF M OSUL , I RAQ . waadsulaiman@hotmail.com Received 22 September, 2005; accepted 05 January, 2006 Communicated by B. Yang A BSTRACT. Four new different types of inequalities similar to Hardy-Hilbert’s inequality are given. Key words and phrases: Hardy-Hilbert integral inequality. 2000 Mathematics Subject Classification. 26D15. 1. I NTRODUCTION R∞ R∞ Suppose that f and g are real functions, such that 0 < 0 f 2 (t)dt < ∞ and 0 < 0 g 2 (t)dt < ∞, then Z ∞ 12 Z ∞Z ∞ Z ∞ f (x)g(y) 2 2 (1.1) <π f (t)dt g (t)dt , x+y 0 0 0 0 where n ) and (bn ) are sequences of real numbers such that 0 < P∞ π2 is best possible. P∞If (a 2 n=1 an < ∞ and 0 < n=1 bn < ∞, then ! 12 ∞ X ∞ ∞ ∞ X X X am b n (1.2) <π a2n b2n . m+n n=1 m=1 n=1 n=1 The inequalities (1.1) and (1.2) are called Hilbert’s inequalities. These inequalities play an important role in analysis (cf. [1, Chap. 9]). In their recent papers Hu [5] and Gao [3] gave two distinct improvements of (1.1) and Gao [4] gave a strengthened version of (1.2). The following definitions are given: r+λ−2 ϕλ (r) = (r = p, q), kλ (p) = B (ϕλ (p), ϕλ (q)) , r and B is the beta function. ISSN (electronic): 1443-5756 c 2006 Victoria University. All rights reserved. The author is grateful to the referee who read through the paper very carefully and correct many typing mistakes. 285-05 2 W.T. S ULAIMAN Recently, by introducing some parameters, Yang and Debnath [2] gave the following extensions: Theorem A. If f, g ≥ 0, p > 1, p1 + 1q = 1, λ > 2 − min {p, q} , such that Z ∞ Z ∞ 1−λ p 0< t f (t)dt < ∞ and 0 < t1−λ g q (t)dt < ∞, 0 0 then ∞ Z ∞ Z (1.3) 0 f (x)g(y) (Ax + By)λ 0 dxdy ∞ Z kλ (p) < ϕ (p) ϕ (q) A λ B λ x 1−λ p p1 Z ∞ y f (x)dx 1−λ q g (y)dy 1q , 0 0 where the constant factor [kλ (p)/Aϕλ (p) B ϕλ (q) ] is the best possible. Theorem B. If f ≥ 0, p > 1, R ∞ 1−λ p t f (t)dt < ∞, then 0 Z ∞ y (1.4) (λ−1)(p−1) Z 0 ∞ 1 p + 1 q = 1, λ > 2 − min {p, q} , A, B > 0 such that 0 < !p f (x) dx (Ax + By)λ 0 dy < kλ (p) ϕ λ A (p) B ϕλ (q) p Z ∞ x1−λ f p (x)dx, 0 p where the constant factor kλ (p)/Aϕλ (p) B ϕλ (q) is the best possible. The inequalities (1.3) and (1.4) are equivalent. Theorem C. If an , bn > 0 (n ∈ N), p > 1, p1 + 1q = 1, 2 − min {p, q} < λ < 2, A, B > 0 such that ∞ ∞ X X 1−λ p 0< n an < ∞ and 0 < n1−λ bqn < ∞, n=1 n=1 then ! p1 ∞ ! 1q ∞ X X kλ (p) (1.5) < ϕ (p) ϕ (q) n1−λ apn n1−λ bqn , λ λ λ A B (Am + Bn) n=1 m=1 n=1 n=1 where the constant factor kλ (p)/Aϕλ (p) B ϕλ (q) is the best possible. ∞ X ∞ X am b n Theorem D. If an ≥ 0 (n ∈ N), p > 1, p1 + P 1−λ p that 0 < ∞ an < ∞, then n=1 n (1.6) ∞ X n=1 n (λ−1)(p−1) ∞ X 1 q !p am λ m=1 = 1, 2 − min {p, q} < λ ≤ 2, A, B > 0 such (Am + Bn) < kλ (p) ϕ A λ (p) B ϕλ (q) p X ∞ n1−λ apn , n=1 p where the constant factor kλ (p)/Aϕλ (p) B ϕλ (q) is the best possible. The inequalities (1.5) and (1.6) are equivalent. 2. N EW I NEQUALITIES The aim of this paper is to give the following results: J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 http://jipam.vu.edu.au/ F OUR I NEQUALITIES S IMILAR T O H ARDY-H ILBERT ’ S I NTEGRAL I NEQUALITY 3 Theorem 2.1. Let ln f , ln g be convex for nonnegative functions f and g such that f (0) = g(0) = 0, f (∞) = g(∞) = ∞, f 0 (s) ≥ 0, g 0 (s) ≥ 0, s ∈ {xp , y q } . Let λ > max {p, q} , p > 1, p1 + 1q = 1. Let Z ∞ −p2 /q2 Z ∞ −q2 /p2 [f (tp )]2−λ+p/q t t [g(tq )]2−λ+q/p 0< dt < ∞, 0 < dt < ∞, p q [f 0 (t)] q [g 0 (t)] p 0 0 then we have Z ∞Z ∞ f (xy)g(xy) dxdy (2.1) (f (xp ) + g(y q ))λ 0 0 1 1 1 p (p, λ − p) B q (q, λ − q) ≤ √ B √ p p q q ! p1 Z ! 1q Z ∞ −p2 /q2 ∞ −q 2 /p2 t [f (tp )]2−λ+p/q t [g(tq )]2−λ+q/p × dt dt . p q [f 0 (t)] q [g 0 (t)] p 0 0 xp p p Proof. Since ln f is convex and xy ≤ ln f x p f (xy) = eln f (xy) ≤ e Therefore, we have Z ∞Z ∞ f (xy)g(xy) 0 0 (f (xp ) + g(y q ))λ ∞ yq , q q + yq then ln f (xp ) ln f (y q ) + p q ≤e 0 1 f p (xp )g q (y q ) [g (y 1 q )] p q−1 p p−1 x q 1 ∞ Z ∞ ≤ 0 × λ q 1 dxdy f (xp )g p/q (y q )g 0 (y q )y q−1 0 = M pN q, p−1 q q−1 y p x (f (xp ) + g(y q )) q ! p1 dxdy p x(p−1)p/q [f 0 (xp )] q (f (xp ) + g(y q ))λ Z ∞Z ∞ f (y q )g q/p (xp )f 0 (xp )xp−1 0 1 1 p )] q 1 [g 0 (y q )] p λ Z 0 1 (f (xp ) + g(y q )) p 0 1 f q (y q )g p (xp ) [f (x y 1 [f 0 (xp )] q ≤ 0 1 = f p (xp )f q (y q ). dxdy 1 ∞Z Z + 0 ! 1q y (q−1)q/p [g 0 (y q )] p (f (xp ) + g(y q ))λ dxdy say. Then 1 M= q Z 1 = q Z ∞ 2−λ+p/q x(1−p)p/q [f (xp )] dx p [f 0 (x)] q 0 ∞ 2 /q 2 x−p ∞ Z g(y q ) f (xp ) 0 [f (xp )]2−λ+p/q Z ∞ pq 1+ q−1 g 0 (y q ) qy f (xp ) dy λ q g(y ) f (xp ) up/q dx du p (1 + u)λ [f 0 (x)] q 0 Z ∞ −p2 /q2 1 x [f (xp )]2−λ+p/q = B (p, λ − p) dx. p q [f 0 (x)] q 0 0 Similarly, 1 N = B (q, λ − q) p J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 Z ∞ y −q 2 /p2 [g(y q )]2−λ+q/p q 0 [g 0 (y)] p dy. http://jipam.vu.edu.au/ 4 W.T. S ULAIMAN Therefore Z ∞Z ∞ 0 f (xy)g(xy) dxdy (f (xp ) + g(y q ))λ 1 1 1 p (p, λ − p) B q (q, λ − q) ≤ √ B √ q p p q ! p1 Z ∞ −p2 /q2 p 2−λ+p/q t [f (t )] × t dt p [f 0 (t)] q 0 0 Z ∞ −q 2 /p2 t q 2−λ+q/p [g(t )] q 0 [g 0 (t)] p ! 1q dt . Theorem 2.2. Let f, g, h be nonnegative functions, h(x, y) is homogeneous of order n such that h(0, 1) = h(1, 0) = 0, h(∞, 1) = h(1, ∞) = ∞. Let p > 1, p1 + 1q = 1, n o 0 < 1 + µ − r < λ, r ∈ pq , pq , hx (x, y) ≥ 0, hy (x, y) ≥ 0, where hx = dh/dx, 0 < Z (2.2) 0 ∞ Z ∞ R∞ tp/q f p (t)dt < ∞, 0 < 0 R∞ 0 tq/p g q (t)dt < ∞, then f (x)g(y)hµ (x, y) dxdy (1 + h(x, y))λ 1 p q p q 1 p1 q B ≤ B 1 + µ− , λ − 1 − µ+ 1 + µ− , λ − 1 − µ− n q q p p Z ∞ p1 Z ∞ 1q p q p q q−1 p−1 × t f (t)dt t g (t)dt . 0 0 Proof. Observe that Z ∞Z 0 ∞ 0 f (x)g(y)hµ (x, y) dxdy (1 + h(x, y))λ ! p1 Z ∞Z ∞ p f (x)hµ (x, y)hy (x, y) ≤ dxdy p/q hx (x, y) (1 + h(x, y))λ 0 0 ! 1q Z ∞Z ∞ q g (y)hµ (x, y)hx (x, y) × dxdy q/p hy (x, y) (1 + h(x, y))λ 0 0 0 1 1 = M pN q, say. Then Z M= ∞ p Z f (x)dx 0 0 ∞ hµ (x, y)hy (x, y) p/q hx (x, y) (1 + h(x, y))λ dy. Let y = xv, dy = xdv and hence dh(x, xv) dh(1, v) dh(1, v) = xn = xn−1 = xn−1 hv (1, v), dy dy dv dh(x, xv) d n hx (x, y) = = x h(1, v) = nxn−1 h(1, v), dx dx hy (x, y) = J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 http://jipam.vu.edu.au/ F OUR I NEQUALITIES S IMILAR T O H ARDY-H ILBERT ’ S I NTEGRAL I NEQUALITY 5 therefore M= 1 ∞ Z x Z p/q−1 p ∞ f (x)dx [xn h(1, v)]µ−p/q xn hv (1, v) (1 + xn h(1, v))λ Z ∞ 1 p p = p/q B 1 + µ− , λ − 1 − µ+ xp/q−1 f p (x)dx. n q q 0 np/q 0 dv 0 Similarly, 1 q q N = q/p B 1 + µ − , λ − 1 − µ + n p p Z ∞ y q/p−1 g q (y)dy. 0 This implies Z ∞Z ∞ f (x)g(y)hµ (x, y) 0 0 dxdy (1 + h(x, y))λ 1 1 p1 p p q q ≤ B 1 + µ − , λ − 1 − µ+ B q 1 + µ − , λ − 1 − µ+ n q q p p Z ∞ p1 Z ∞ 1q p/q−1 p q/p−1 q × t f (t)dt t g (t)dt . 0 0 The following lemma is needed for the coming result. Lemma 2.3. Let s ≥ 1, 0 < 1 + µ ≤ min {α, λ} and define Z s tµ −α f (s) = s dt, λ 0 (1 + t) then f (s) ≤ f (1). Proof. We have 0 f (s) = s ≤ sµ −α Z λ (1 + s) s µ−α + s tµ λ dt (−α) s−α−1 0 (1 + t) −α−1 Z s αs tµ dt (1 + s)λ 0 − (1 + s)λ α sµ−α 1− = ≤ 0. 1+µ (1 + s)λ This shows that f is nonincreasing and hence f (s) ≤ f (1). Rs Theorem 2.4. Let f, g, F, G be nonnegative functions such that Fn(s) = o Rs p q 1 1 + = 1, (1 − λ/2)r ≤ λ/2 + α ≤ 2α, r ∈ , , g(t)dt, let, p > 1, p q q p 0 Z f (t)dt, G(s) = x 0< Z0 x 0< 0 (x − t)t(1−λ/2)p/q−λ/2−α F p−1 (t)f (t)dt < ∞, (x − t)t(1−λ/2)q/p−λ/2−α Gq−1 (t)g(t)dt < ∞, 0 J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 http://jipam.vu.edu.au/ 6 W.T. S ULAIMAN then Z x x Z F (s)G(t) dsdt (s + t)λ Z x p1 √ √ p p q q ≤ xα B λ2 , λ2 (x − t)t(1−λ/2)p/q−λ/2−α F p−1 (t)f (t)dt 2 0 Z x 1q (1−λ/2)q/p−λ/2−α q−1 × (x − t)t G (t)g(t)dt . (2.3) 0 0 0 Proof. Observe that Z xZ x F (s)G(t) 0 0 (s + t)λ x Z dsdt F (s) x Z = 0 x x Z ≤ 0 1 t1/p s1/q λ/2−1 G(t) · (s + t)λ/p 0 Z F p (s)tλ/2−1 1 = M pN q, s1/q t1/p λ/2−1 dsdt (s + t)λ/q ! p1 Z (s + t)λ s(λ/2−1)p/q 0 x Z x dsdt 0 0 Gq (t)sλ/2−1 (s + t)λ t(λ/2−1)q/p ! 1q dsdt say. Then x Z M= s (1−λ/2)p/q−λ/2 F (s)ds 0 0 x Z = s (1−λ/2)p/q−λ/2 p F (s)ds 0 Z x Z p x ≤ s 0 α x = B 2 (1−λ/2)p/q−λ/2 λ λ , 2 2 Z p F (s)ds t λ/2−1 1 s s t λ 1+ s dt x α x −α Z s s x α Z s 0 1 x/s 0 uλ/2−1 (1 + u)λ uλ/2−1 (1 + u)λ du du x s(1−λ/2)p/q−λ/2−α F p (s)ds, 0 by virtue of the lemma. As Z p F (s) = p s F p−1 (u)f (u)du, 0 then Z x Z s p α λ λ (1−λ/2)p/q−λ/2−α M= x B , s ds F p−1 (u)f (u)du 2 2 2 0 Z0 x Z s p α λ λ = x B , u(1−λ/2)p/q−λ/2−α F p−1 (u)f (u)dsdu 2 2 2 Z0 x 0 p λ λ = xα B , (x − s)s(1−λ/2)p/q−λ/2−α F p−1 (s)f (s)ds. 2 2 2 0 Similarly, q N = xα B 2 λ λ , 2 2 Z J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 x (x − t)t(1−λ/2)q/p−λ/2−α Gq−1 (t)g(t)dt. 0 http://jipam.vu.edu.au/ F OUR I NEQUALITIES S IMILAR T O H ARDY-H ILBERT ’ S I NTEGRAL I NEQUALITY 7 Therefore, we have Z 0 x x Z F (s)G(t) dsdt (s + t)λ Z x p1 √ √ p p q q λ λ α (1−λ/2)p/q−λ/2−α p−1 ≤ x B , (x − t)t F (t)f (t)dt 2 2 2 0 Z x 1q (1−λ/2)q/p−λ/2−α q−1 × (x − t)t G (t)g(t)dt . 0 0 Theorem 2.5. Let f, g be nonnegative functions, f is submultiplicative and g is concave nonincreasing, f 0 (x), g 0 (y) ≥ 0, f (0) = g(0) = 0, f (∞) = g(∞) = ∞, p > 1, p1 + 1q = 1, 0 < a + 1 < λ, 0 < b + 1 < λ, Z ∞ Z ∞ [f (x)]µp−bp/q [g(x)]1+a−λ [g(y)]µq−aq/p [f (y)]1+b−λ 0< dx < ∞, 0 < dy < ∞, p q [f 0 (x)] q [g 0 (y)] p 0 0 then ∞ Z ∞ Z (2.4) 0 0 Proof. Since 1 1 f µ (xy) dxdy ≤ 2λ/2 B p (a + 1, λ − a − 1) B q (b + 1, λ − b − 1) λ/2 g (xy) ! p1 Z ! 1q Z ∞ ∞ [f (t)]µp [g(t)]1+a−λ−bp/q [f (t)]µq [g(t)]1+b−λ−aq/p × dt dt . p q [g 0 (t)] q [g 0 (t)] p 0 0 √ xy ≤ x+y , 2 then 2 2 √ x+y g(x) + g(y) √ 2 2 ≥ , g(xy) = g ( xy) ≥ (g ( xy)) ≥ g 2 2 and hence Z 0 ∞ Z 0 ∞ f µ (xy) dxdy g λ/2 (xy) Z ∞Z λ/2 ≤2 0 ∞ f µ (x)f µ (y) (g(x) + g(y))λ 0 a/p = 2λ/2 ∞ Z ∞ 1/p 0 [g (y)] f µ (x) [g(y)] [g(x)]b/q [g 0 (x)]1/q λ b/q · [g(x)] f µ (y) [g(y)] a/p · [g 0 (x)]1/q [g 0 (y)]1/p λ (g(x) + g(y)) q ! p1 Z ∞Z ∞ f µp (x)g a (y)g 0 (y) dxdy p [g(x)]bp/q [g 0 (x)] q (g(x) + g(y))λ 0 0 ! 1q Z ∞Z ∞ f µq (y)g b (x)g 0 (x) × dxdy q [g(y)]aq/p [g 0 (y)] p (g(x) + g(y))λ 0 0 0 ≤ 2λ/2 Z dxdy 0 1 1 = 2λ/2 M p N q , J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 (g(x) + g(y)) p say. http://jipam.vu.edu.au/ 8 W.T. S ULAIMAN Then Z ∞ M= µp 1+a−λ−bp/q [f (x)] [g(x)] dx p 0 ∞ Z [g 0 (x)] q 0 ∞ Z = B (a + 1, λ − a − 1) g(y) g(x) a 1+ g 0 (y) g(x) g(y) g(x) λ dy [f (x)]ηp [g(x)]1+a−λ−bp/q p [g 0 (x)] q 0 dx. Similarly, we can show that Z N = B (b + 1, λ − b − 1) ∞ [f (y)]µq [g(y)]1+b−λ−aq/p q 0 [g 0 (y)] p dy. The result follows. R EFERENCES [1] G.H. HARDY, J.E. LITTLEWOOD bridge, 1952. AND G. POLYA, Inequalities, Cambridge Univ. Press. Cam- [2] B. YANG AND L. DEBNATH, On the extended Hardy-Hilbert’s inequality, J. Math. Anal. Appl., 272 (2002), 187–199. [3] M. GAO, TAN LI AND L. DEBNATH, Some improvements on Hilbert’s integral inequality, J. Math. Anal. Appl., 229 (1999), 682–689. [4] M. GAO, A note on Hilbert double series theorem, Hunan Ann. Math., 12(1-2) (1992), 142–147. [5] HU KE, On Hilbert’s inequality, Chinese Ann. of Math., 13B(1) (1992), 35–39. [6] W. RUDIN, Principles of Mathematical Analysis, Third Edition, McGraw-Hill Book Co. 1976. J. Inequal. Pure and Appl. Math., 7(2) Art. 76, 2006 http://jipam.vu.edu.au/