Journal of Inequalities in Pure and Applied Mathematics http://jipam.vu.edu.au/ Volume 3, Issue 3, Article 43, 2002 INEQUALITIES ON LINEAR FUNCTIONS AND CIRCULAR POWERS PANTELIMON STĂNICĂ AUBURN U NIVERSITY M ONTGOMERY, D EPARTMENT OF M ATHEMATICS , M ONTGOMERY, AL 36124-4023, USA. stanica@strudel.aum.edu URL: http://sciences.aum.edu/˜ stanpan Received 25 February, 2002; accepted 10 April, 2002 Communicated by C.P. Niculescu A BSTRACT. We prove some inequalities such as x x F (x1 σ(1) , . . . , xnσ(n) ) ≤ F (xx1 1 , . . . , xxnn ), where F is a linear function or a linear function in logarithms and σ is a permutation, which is a product of disjoint translation cycles. Stronger inequalities are proved for second-order recurrence sequences, generalizing those of Diaz. Key words and phrases: Binary Sequences, Fibonacci Numbers, Linear Forms, Inequalities. 2000 Mathematics Subject Classification. 11B37, 11B39, 26D15. 1. I NTRODUCTION Define the second-order recurrent sequence by (1.1) xn+1 = axn + bxn−1 , x0 ≥ 0, x1 ≥ 1, with a, b ≥ 1. If a = b = 1 and x0 = 0, x1 = 1 (or x0 = 2, x1 = 1), then xn is the Fibonacci sequence, Fn (or Pell sequence, Pn ). Inequalities on Fibonacci numbers were used recently by Bar-Noy et.al [1], to study a 9/8− approximation for a variant of the problem that models the Broadcast Disks application (model for efficient caching of web pages). In [2], J.L. Diaz proposed the following two inequalities: Fn+2 Fn+2 Fn+1 Fn (a) FnFn+1 + Fn+1 + Fn+2 < FnFn + Fn+1 + Fn+2 , Fn+1 Fn+2 Fn Fn Fn+1 Fn+2 (b) Fn Fn+1 Fn+2 < Fn Fn+1 Fn+2 . In this note we show that the inequalities proposed by Diaz are not specific to the Fibonacci sequence, holding for any strictly increasing sequence. Moreover, we prove that stronger inequalities hold for any second-order recurrent sequence as in (1.1). Furthermore, we pose a problem for future research. ISSN (electronic): 1443-5756 c 2002 Victoria University. All rights reserved. The author would like to thank Professor C.P. Niculescu for his helpful comments, which improved significantly the presentation. 015-02 2 PANTELIMON S TĂNIC Ă 2. T HE R ESULTS We wondered if the inequalities (a), (b) were dependent on the Fibonacci sequence or if they can be extended to binary recurrent sequences. From here on, we assume that all sequences have positive terms. Without too great a difficulty, we prove, for a binary sequence, that Theorem 2.1. For any positive integer n, x x x n+2 n+1 n+2 n , + xn+2 + xxn+2 < xxnn + xn+1 xxnn+1 + xn+1 x x x n+2 n+1 n+2 xn . xn+2 xn+2 < xxnn xn+1 xxnn+1 xn+1 (2.1) (2.2) Proof. We shall prove (2.3) xy + y ax+by + (ax + by)x < xx + y y + (ax + by)ax+by , if 0 < x < y, which will imply our theorem. For easy writing, we denote z = ax + by. Then (2.3) is equivalent to (2.4) xx xy−x − 1 + y y y z−y − 1 < z y z z−y − z −(y−x) . Now, xx + y y < xy + y y < (x + y)y ≤ z y , since a, b ≥ 1. Moreover, xy−x − 1 + y z−y − 1 = xy−x + y z−y − 2 < xz−y + y z−y − 1 < (x + y)z−y − 1 < z z−y − z −(y−x) . Taking A = xx , B = y y , C = xy−x − 1, D = y z−y − 1 and using the inequality for positive numbers AC + BD ≤ (A + B)(C + D), we obtain (2.4). The inequality (2.2) is implied by xy y z z x < xx y y z z ⇐⇒ xy−x y z−y < z z−x . (2.5) But z z−x = z (z−y)+(y−x) ≥ (x + y)z−y (x + y)(y−x) > y z−y xy−x . The theorem is proved. Remark 2.2. We preferred to give this proof since it can be seen that the two inequalities are far from being tight. We remark that inequality (2.2) can be also shown by using Theorem 2.7. With a little effort, while not attempting to have the best bound, we can improve it, and also prove that the gaps are approaching infinity. Theorem 2.3. We have xn+1 xn+2 xn+2 n + xn+2 − xn+2 xn+1 −xn < xxnn + xn+1 + xxn+2 xxnn+1 + xn+1 x x x x n+1 xn n+1 n+2 n+2 xn xn+2 . xn+2 − 3xxnn xn+1 xn+2 < xxnn xn+1 xxnn+1 xn+1 In particular, xn+2 xn+1 xn+2 n ]=∞ + xxn+2 + xn+2 − xxnn+1 + xn+1 lim [ xxnn + xn+1 n→∞ x x x n+2 xn n+1 n+2 xn+2 ] = ∞. xn+2 − xxnn+1 xn+1 lim [xxnn xn+1 n→∞ In fact, the inequalities (2.1), (2.2) are not dependent on binary sequences, at all. A much more general statement is true. Take σ a permutation, which is a product of disjoint cyclic (translations by a fixed number, c(i) = i + t) permutations. Theorem 2.4. Let n ≥ 2 and 1 ≤ x1 < x2 < . . . < xn a strictly increasing sequence. Then n n X X xσ(i) (2.6) xi ≤ xxi i , i=1 J. Inequal. Pure and Appl. Math., 3(3) Art. 43, 2002 i=1 http://jipam.vu.edu.au/ I NEQUALITIES ON L INEAR F UNCTIONS AND C IRCULAR P OWERS 3 and n Y (2.7) x xi σ(i) ≤ i=1 n Y xxi i , i=1 with strict inequality if σ is not the identity. Proof. If σ is the identity permutation, the equality is obvious. Now, assume that σ(i) = i + t. We take the case of t = 1 (the others are similar). We prove (2.6) by induction on n. If n = 2, we need xx1 2 + xx2 1 < xx1 1 + xx2 2 , which is equivalent to xx1 1 xx1 2 −x−1 − 1 < xx2 1 xx2 2 −x−1 − 1 . The last inequality is certainly valid, since xx1 1 < xx2 1 and xx1 2 −x1 − 1 < xx2 2 −x1 − 1. Assuming the statement holds true for n, we prove it for n + 1. We need n+1 X (2.8) x xi i+1 < i=1 n+1 X xxi i , i=1 where xn+2 := x1 . We re-write (2.8) as x n+1 1 (xx1 2 + xx2 3 + · · · + xxn1 ) + xnxn+1 + xxn+1 − xxn1 < xx1 1 + xx2 2 + · · · + xxnn + xn+1 , and using induction, it suffices to prove that x n+1 1 xnxn+1 + xxn+1 − xxn1 < xn+1 . The previous inequality is equivalent to xn+1 −x1 1 xxn1 xxnn+1 −x1 − 1 < xxn+1 xn+1 −1 , which is obviously true, since xn < xn+1 . The inequality (2.7) (when σ(i) = i + t) can be proved by induction, as well. If n = 2, then xx1 2 xx2 1 < xx1 1 xx2 2 ⇐⇒ x1x2 −x1 < x2x2 −x1 , which is true since x1 < x2 . Assuming the inequality holds true for n, we prove it for n + 1. We need xn+1 n 1 xx1 2 · · · xxnn+1 xxn1 = xx1 2 · · · xxn−1 xxn1 xxnn+1 −x1 xxn+1 < xx1 1 · · · xn+1 . Using the induction step, it suffices to prove x x n+1 n+1 1 < xn+1 ⇐⇒ xxnn+1 −x1 < xn+1 xxnn+1 −x1 xxn+1 −x1 , which is valid since xn < xn+1 . Now, take the general permutation σ 6=identity, which is a product of disjoint cyclic permutations. Thus, σ can be written as a product of disjoint cycles as σ = C1 × C2 × · · · × Cm . Recall that σ was taken such that all of its cycles Ck are translations by a fixed number, say tk . Take Ck a cycle of length ek and choose an index in Ck , say ik . Since σ is not the identity, then there is an index k such that ek 6= 1. We write the inequalities (2.6) and (2.7) as m eX k −1 X xσj+1 (i k) xσj (ik ) < k=1 j=0 m eY k −1 Y k=1 j=0 J. Inequal. Pure and Appl. Math., 3(3) Art. 43, 2002 m eX k −1 X xσj (i ) xσj (ikk) , k=1 j=0 xσj+1 (i xσj (ik ) k) < m eY k −1 Y xσj (i ) xσj (ikk) , k=1 j=0 http://jipam.vu.edu.au/ 4 PANTELIMON S TĂNIC Ă Therefore, it suffices to prove that, for any k, with ek 6= 1, we have eX k −1 xσj+1 (i xσj (ik ) k) < eX k −1 j=0 j=0 eY k −1 eY k −1 xσj+1 (i xσj (ik ) k) < j=0 xσj (i ) xσj (i ) xσj (ikk) , xσj (ikk) , j=0 that is, eX k −1 xσj (i xσj (ikk) )+tk < eX k −1 j=0 j=0 eY k −1 eY k −1 xσj (i xσj (ikk) )+tk < j=0 xσj (i ) xσj (i ) xσj (ikk) , xσj (ikk) . j=0 For k fixed, the above inequalities are just applications of the previous step (of σ(i) = i + t), by taking a sequence yl to be xσj (ik ) in increasing order (we could take from the beginning ik to be the minimum index in each cycle Ck ). We can slightly extend the previous result (for a similar permutation σ) in the following (we omit the proof). Theorem 2.5. For any increasing sequence 0 < x1 < · · · < xn , we have n X (2.9) x n X i=1 i=1 n Y x n Y ai xi σ(i) ≤ ai xi σ(i) ≤ ai xxi i , and ai xxi i , i=1 i=1 where ai ≥ 0. A parallel result involving logarithms is also true (σ is a permutation as before). Theorem 2.6. For any finite increasing sequence, 0 < x1 < x2 < · · · < xn , and any positive real numbers ai , we have n X i=1 n Y ai xσ(i) log(xi ) ≤ ai xσ(i) log(xi ) = i=1 n X ai xi log(xi ), and i=1 n Y ai xi log(xi ). i=1 The second identity is easily true since every ai , xi and log xi occurs in both sides. We omit the proof of the first inequality, since it can be deduced easily (as the referee observed) from the known fact (see [3, p. 261]) Theorem 2.7. Given two increasing sequences u1 ≤ u2 ≤ · · · ≤ un and w1 ≤ w2 ≤ · · · ≤ wn , then n n n X X X ui wn+1−i ≤ uτ (i) wσ(i) ≤ ui wi , i=1 i=1 i=1 for any permutations σ, τ . J. Inequal. Pure and Appl. Math., 3(3) Art. 43, 2002 http://jipam.vu.edu.au/ I NEQUALITIES ON L INEAR F UNCTIONS AND C IRCULAR P OWERS 5 3. F URTHER C OMMENTS We believe that other inequalities of the type occurring in our theorems can also be constructed. Let F : Rn → R be a function, with the properties (3.1) If xi ≤ yi , i = 1, . . . , n, then F (x1 , . . . , xn ) ≤ F (y1 , . . . , yn ), with strict inequality if there is an index i such that xi < yi . and For 0 < x1 < x2 < · · · < xn , then, F (xx1 2 , xx2 3 , . . . , xxn1 ) ≤ F (xx1 1 , xx2 2 , . . . , xxnn ). Pn As examples, we have the linear i=1 ai xi , the linear form in Pn polynomial F (x1 , . . . , xn ) = logarithms F (x1 , . . . , xn ) = i=1 ai log(xi ), and the corresponding products. We ask for more examples of functions satisfying (3.1) and (3.2), which cannot be derived trivially from the previous examples (by raising each variable to the same power, for instance). Is it true that any symmetric polynomial satisfies (3.1) and (3.2)? In addition to more examples, it might be worth investigating the general form of polynomial functions that satisfy these properties. This looks like a mathematical version of the philosophy saying: Going one step at the time it is far better than jumping too fast and then at the end falling to the bottom. (3.2) R EFERENCES [1] A. BAR-NOY, R. BHATIA, J. NAOR AND B. SCHIEBER, Minimizing Service and Operating Costs of Periodic Scheduling, Symposium on Discrete Algorithms (SODA), p. 36, 1998. [2] J.L. DIAZ, Problem H-581, Fibonacci Quart., 40(1) (2002). [3] G. HARDY, J.E. LITTLEWOOD, G. PÒLYA, Inequalities, Cambridge Univ. Press, 2001. J. Inequal. Pure and Appl. Math., 3(3) Art. 43, 2002 http://jipam.vu.edu.au/