TOPICS IN POLYNOMIALS: EXTREMAL PROBLEMS, INEQUALITIES, ZEROSTOPICS IN POLYNOMIALS: EXTREMAL PROBLEMS, INEQUALITIES, ZEROSPublished by ‘World Scientific Publishing Co, Pte. Ltd PO Box 128, Farrer Road, Singapore 912805 USA office: Suite 1B, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE British Library Cataloguing-in-Publication Data ‘A catalogue record for this book is available from the British Library, First published 1994 Reprinted 1999 ‘TOPICS IN POLYNOMIALS: EXTREMAL PROBLEMS, INEQUALITIES, ZEROS ‘Copyright © 1994 by World Scientific Publishing Co. Pre. Lid. All rights reserved. This book, or parts thereof. may not be reproduced in any form or by any means, electronic or mechanical. including photocopying. recording or any information storage and retrieval system ners known or to be invented, without written permission frarn the Publisher For photocopying of material in this volume, please pay a copying fee through the Copyright ‘Clearance Center, inc., 222 Rosewood Drive, Danvers. MA 01923, USA. In this case permission to photocopy is not required from the publisher. ISBN. 981-02-0499-K Printed in SingaporeTOPICS IN POLYNOMIALS: EXTREMAL PROBLEMS, INEQUALITIES, ZEROS G. V. Milovanovié Department of Mathematics University of Nis, Yugoslavia D. S. Mitrinovié Department of Mathematics University of Belgrade, Yugoslavia Th. M. Rassias Department of Mathematics University of La Verne, Greece World Scientific Singapore «Mew 2ersey *!.ondon+ HongKongContents Preface xi CHAPTER 1 1 General Concept of Algebraic Polynomials 1.1. POLYNOMIALS AND EQUATIONS ............. - I 1.1.1. Preliminaries 1 1.1.2. Polynomial equations 5 1.1.3. The quadratic equation 9 1.1.4. The cubic equation . 9 1.1.5. The quartic equation . 1 1.1.5. The quintic equation 15, 1.2. SOME CLASSES OF POLYNOMIALS 16 1.2.1. Self-inversive polynomials 16 1.2.2, Algebraic and trigonometric polynomials 19 1.2.3. Fejér-Riesz’s representation of non-negative trigonometric polynomials 22 1.2.4. Non-negative algebraic polynomials . 25 1.2.5. Lorentz representation of polynomials . 28 1.2.6. Basic properties of orthogonal polynomials 31 1.2.7. The classical orthogonal polynomials 37 1.2.8. Some Appell polynomial systems... 47 1.3. POLYNOMIALS IN SEVERAL VARIABLES . 52 1.3.1. Symmetric polynomials 52 1.3.2. Properties of symmetric polynomials in two and three variables .. 58 1.3.3. Resultants and discriminants of polynomials .............. 59 1.3.4. Discriminants of the classical orthogonal polynomials and related problems ........6.2.000c0cccceeseeeseeseeeteeeeee 65 1.3.5. Homogeneous polynomials . 70 1.3.6. Some inequalities for symmet n 1.3.7. Properties of zeros 4 1.4. REFERENCESvt CHAPTER 2 CONTENTS & Selected Polynomial Inequalities 2.1. 2.1.1. 2.1.2. 2.13. 2.1.4, 2.16, 2.1.7. 2.2. 221 2.2.2. 2.2.3. 224. 2.25. 2.3. INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS ... Inequalities for polynomi Some important trinomials . Inequalities for polynomials with real zeros. Inequalities involving zeros of polynomials - Inequalities for non-negative polynomials Inequalities for the coefficients of strane Integral inequalities ........... INEQUALITIES WITH ‘TRIGONOMETRIC POLYNOMIALS Inequalities for trigonometric sums Inequalities in L norm Inequalities for Dirichlet’s kernel Extremal problems for non-negative polynomials . Inequalities for moments and coefficients of non-negative cosine polynomials REFERENCES CHAPTER 3 173 Zeros of Polynomials 3.1. 3.11 3.1.2. 3.1.3. 3.14, 3.1.5. 3.1.6. 3.2. 3.2.1. 3.2.2. DISTRIBUTION OF ZEROS OF ALGEBRAIC POLYNOMIALS Some basic results .... Ganss-Lucas theorem anal related inequalities Zeros of polar derivative ef Grace theorem and some applications... Zeros of the Wronskian of a polynomial . Distribution of zeros of real polynomials THE SENDOV-ILIEFF CONJECTURE AND RELATED TOPICS .................-.......0.0.. 7 216 Introduction ...........-..-.0- 000 cece eee ee eee . 216 fnost five) fererieetettetsette sett C eet eeceteeererercecen irr sist 2183.2.3. 3.3. 3.3.1. 3.3.2. 3.3.3. 3.3.4. 3.4. CONTENTS vii The Sendov-Ilieff conjecture for polynomials of any degree 230 BOUNDS FOR THE ZEROS AND THEIR NUMBER IN A GIVEN DOMAIN ..........-.-..--- 243 Bounds for the moduli of the zeros . Zeros in astrip ........... Enestre-Kakeya theorem and its generalizations Number of zeros in a given domai REFERENCES 277 CHAPTER 4 299 Inequalities Connected With Trigonometric Sums 41. 411. 4.1.2, 4.1.3. 4.2. 421. 4.2.2. 423. 424, 4.25. 426. 4.2.7, 4.3. CLASSICAL RESULTS . Preliminaries Fejér-Gronwall-Jackson’s, Young’s and related inequalities Inequalities of Rogosinski and Szeg6 and their extensions .. 323 POSITIVITY AND MONOTONICITY OF CERTAIN SUMS Turan’s inequalities .... ry of some classes of trigonometric sums ry of some orthogonal polynomial sums Completely monotonic functions . » Absolutely monotonic functions , Monotenicity of some trigonometric sums .. Positivity of some Jacobi polynomial sums .. REFERENCES ................-.-..0006+ CHAPTER 5 383 Extremal Problems for Polynomials 5.1. 5.1.1. 5.1.2. 5.1.3. 5.14. POLYNOMIALS WITH MINIMAL NORM AND ESTIMATES FOR COEFFICIENTS ................ 383 Preliminaries ... Polynomials with minimal uniform norm . Polynomials with minimal L” norm ..... Further generalizations on polynomials with minimal norm 422viii 5.1.5. 5.1.6. 5.17. 5.1.8. 5.1.9. 5.1.10. 5.2. 5.2.1. 5.2.2. 5.2.3. 5.3. 5.3.1. 5.3.2. 5.3.3, 5.4. CONTENTS Estimates for coefficients of ealzmome | in L? norm with prescribed zeros Remez inequality . On the maximum modulus of polynomials . Extremal problems in mixed norms . Extremal problems for polynomials with ee zeros .. 455 Szegé’s and related extremal problems 5 INCOMPLETE POLYNOMIALS AND WEIGHTED POLYNOMIAL INEQUALITIES Incomplete polynomials and extremal problems Extremal problems with exponential weights . L” inequalities for Freud weights INEQUALITIES OF NIKOL'SKIl TYPE Inequalities of Nikol'skil type .. . Further generalizations and extensions Sharp Nikol’skif inequalities with exponent IREFERENGES) 2 .2:2....seceessseererseccsesseemerce=s~ CHAPTER 6 527 Extremal Problems of Markov-Bernstein Type 6.1. 6.1.1. 6.1.2. 6.1.3. 6.1.4. 6.1.5. 6.1.6. 6.1.7. 6.1.8. 6.1.9. 6.1.10. 6.2. 6.2.1. INEQUALITIES OF MARKOV AND BERNSTEIN TYPE Preliminaries Classical results of Markov and Bernstein . Markov-Duffin-Schaeffer inequalities .... ; Inequalities of Markov type for curved majorants ......... 546 Markov and Sobolev type inequalities on compact sets MRM oe eceeeceeeeeeeee Extremal problems in L? norm . Generalizations in L” norm ... Extremal problems in different norms . Inequalities of Markov-Bernstein type for Freud weights ... 607 Markov-Bernstein type inequalities for differential operators 612 EXTREMAL PROBLEMS FOR RESTRICTED POLYNOMIAL CLASSES ... Extremal problems in uniform norm6.3.3. 6.4. CONTENTS ix Further generalizations and extensions -..... eeeee 637 Extremal problems in L? metric for non- negative polynomials 5 é Extremal problems for Lorentz classes of {plone Generalizations in L" norm Extremal problems of Turén type - EXTREMAL PROBLEMS IN A CIRCLE . Extremal problems of Bernstein’s type in uniform norm Further generalizations and extensions . Extremal problems in L™ norm REB ERENCES porter cecteer ee cescier sec eecece es CHAPTER 7 725 Some Applications of Polynomials TL. TAAL talk? 7.1.3. TAA 7.2. 7.2.1. 7.2.2. 7.23 7.3. 73.1. 7.3.2. 7.3.3. TAL 7A2. 74.3. TAA. LEAST SQUARES APPROXIMATION WITH CONSTRAINTS ..............0.0ececeeeec cence essences 725 Introduction ...... ° Approximations with simple constraints Computing the inner products with high accuracy - Approximations with more general constraints SIMULTANEOUS APPROXIMATION ... Preliminaries Problems of simultaneous approximation Vectorial approximation ........ THE BERNSTEIN CONJECTURE IN APPROXIMATION THEORY The Bernstein conjecture ..............- 749 Computing the bounds for the Bernstein constant . 753 Computing the numbers {2nEn(|z|)} with high accuracy. 755 APPLICATIONS IN COMPUTER AIDED GEOMETRIC DESIGN ... . Bernstein polynomials and Bézier curves .. Properties of Bernstein basis Properties of Bézier curves .. Some extremal properties of Bernstein polynomials ro] Bezier] cat vies eq aoa miaractmerat oto ote loool elorrortet eee 769 737 74L . 749 .. 759 - 7627A.5. Concluding remarks ...........0.c00cc0eseceeeseeeeeeeees m7 7.5. REFERENCES ....00..0...0.0000000000° cecceseees THB Symbol Index 783 Name Index 787 Subject Index e0sPreface The theory of best approximation with respect to the supremum norm was established as a bona fide branch of mathematica] anal- ysis mainly by the work of P.L. Chebyshev (1821-1894), who in the 1850s studied some of the properties of polynomials with least deviation from a given continuous function. Since that time, the work of the celebrated Petersburg school of Mathematics ~ also called the Chebyshev school — has had a lasting impact in theoretical and applied mathematics and produced students such as A.A. Markov (1856-1922). The present book contains some of the most important results on the analysis of polynomials and their derivatives. Besides the fundamental results which are treated with their proofs, the book also provides an account of the most recent developments concerning extrema] properties of polynomials and their derivatives, as well as properties of their zeros. An attempt has been made to present the material in an integrated and a self-contained fashion. The book is intended, not only for the specialist mathematician, but also for those researchers in the applied and computational sciences who use polynomials as a tool. The subject of polynomial inequalities is of course vast. We chose to restrict ourselves here only to a few directions. We present some striking results (to us at least), but also novel aspects as well as old and new results not normally found in book form. On the other hand, subjects such as orthogonal] polynomials are not included here as such (for that subject see the excellent books of G. Szegé, G. Freud, T.S. Chihara, and P.K. Suetin). Some 1200 references have been cited here, including preprints. As a rule, we have studied the original sources and in some cases have tetrieved some forgotten but useful results. The references appear at the end of each chapter. At the end of the book we include a symbol index, as well as a name and subject index. The first chapter reviews some of the classical results on polyno-xii PREFACE. mials of one and several variables. The second chapter provides an account of some selected inequalities involving algebraic polynomials as well as inequalities with trigonometric polynomials. The third chapter studies zeros of polynomials, with emphasis on the distribution of zeros of the algebraic polynomials, the Sendov- Tlieff conjecture, as well as bounds for the zeros and their number in certain domains. We also consider the Enestrom-Kakeya theorem and its various generalizations. Chapter 4 treats inequalities connected with trigonometric sums. Besides of the classical results of L. Fejér, T.H. Gronwall, D. Jack- son, W.H. Young, W. W. Rogosinski and G. Szegé, we give a special emphasis to the analysis of positivity and monotonicity of certain trigonometric sums and some related orthogonal polynomial] sums. In particular, we point out an inequality of R. Askey and G. Gasper which was the final step in L. de Branges’s remarkable proof (1984) of the Bieberbach conjecture (1916). The fifth and sixth chapters are devoted to extremal problems for polynomials. In Chapter 5 we investigate the extremal problems for polyno- mials and their coefficients, (which, as is known, are suggestive of results in a much more general context!) starting by the classical results of P.L. Chebyshev, A.A. Markov, E.J. Remez, $.N. Bern- stein, A.N. Korkin and E.I. Zolotarev, which are basic to approxi- mation theory. In particular, we study polynomials with minimal L* norm, many generaliz: s of such polynomials, estimates for coef- ficients of polynomials in L? norm with prescribed zeros, extremal problems in mixed norms, as well as G. Szeg6's and related extremal problems. Section 5.2 treats incomplete polynomials introduced in 1976 by G.G. Lorentz and weighted polynomial inequalities, includ- ing extremal problems with exponential weights and L" inequalities for Freud weights. Section 5.3 is devoted to extremal problems and inequalities of Nikol’skii type. Inequalities of Markov and Bernstein-type are fundamental for the proof of many inverse theorems in polynomia] approximation the- ory. Frequently further progress in inverse theorems has depended on first obtaining a corresponding generalization or analog of Markov’s and Bernstein’s inequalities. There are many results on Markov’sPREFACE xiii and Bernstein’s theorems and their generalizations. In Chapter 6 we consider such problems involving additional other classical results of Markov and Bernstein. In Section 6.2 we study the corresponding extremal problems on restricted classes of polynomials, and finally, in Section 6.3, we conclude with the extremal problems in a circle on the complex plane. The final chapter provides some selected applications of polynomi- als. There are applications to least squares approximation with con- straints, to vectorial and simultaneous approximations and to com- puter aided geometric design (CAGD) so much in vogue today. We also study the Bernstein conjecture of 1913, which was settled nega- tively in 1985 by R.S. Varga and A.J. Carpenter. Finally, we wish to thank Professors R. Askey, B. Bojanov, R. Z. Djordjevié, G. Gasper, A. Guessab, Lj.M. Koci¢, I.Z. Milovanovié, P. Nevai, M. Tomié, R.S. Varga, who read parts of the manuscript of the book and provided some very useful comments. Thanks also go to Professor A. W. Goodman who kindly donated to us his collection of papers on the theory of polynomials and to Professor Lj. M. Kocié who wrote Section 7.4 dealing with computer aided geometric design. We also wish to thank Professor D. Carmocolias for his assistance in proofreading the manuscript in English. Last but not least our very special thanks are due to our families for their ever-lasting support and encouragement. It is a pleasure to acknowledge the superb assistance that the staff of World Scientific Publishing Co. has provided. Ni8/Belgrade/ Athens Gradimir V. Milovanovié March 1994 Dragoslav S. Mitrinovié Themistocles M. RassiasCHAPTER 1 General Concepts of Algebraic Polynomials 1.1. POLYNOMIALS AND EQUATIONS 1.1.1. Preliminaries Let F be a field. The ring of polynomials in x over F, denoted by Fiz], is defined to be the set of all formal expressions P(z) = ao + az+--- +42", where the aj, the coefficients of the polynomial P in z, are elements in F. Ifa, # @, then the degree of P, denoted by dg P, isn. Therefore, the degree of a polynomial P is the highest power of z that occurs in the expression for P(z) with a non-zero coefficient. In F[z] we define equality, sum, and product of two polynomials as follows. Given two polynomials P(z) = Go + az +--+ +0,2" and Q(z) = bo + biz t + +bmz™, then the polynomials are equal if, and only if, aj = 6; for all i > 0, that is when their corresponding coefficients are equal. We define the sum of P and Q by (P+ Q)(z) = Pz) + Q(z) = cot ert t+ +e", a2 GENERAL CONCEPTS OF POLYNOMIALS where for each i, c; = a; + b;, and the product of P and Q by (PQ)(z) = P(z)Q(z) = cot cz +-- +052", where 6; = jbo + a;-4by +--+ + aybi-1 + Gobi, for every i. Then, one can show that F{z] is a commutative ring with unit. It is easy to prove that if P(z) and Q(z) are non-zero elements of F[z], then dg(PQ) = dg P + dgQ. Also, if P(x), Q(z) € F[z] and P(z) + Q(z) # 0, then dg(P +.Q) < max{dg P, dg Q}. We do not assign a degree to O(z) = Oz" +02" 4 --- +02 +0. The polynomials of degree 0 are called the constants (these are the elements of the field F). Let ky,...,km be non-negative integers and k= (hy... skim), Ukl=ki tes thm, ok = 2h... zte, As in the preceding work, we can define the ring of polynomials in m independent variables 2, ..., I, over F, denoted by Flz1,.-. ,Zm], as the set of all expressions (1.1.1) P(z) = P(z1,.-. 52m) = Yaz*, where the ay, the coefficients of the polynomial P in x1, ..., 2m, are elements in F. The product 2* = {1 ..-zkm is called the primitive monomial of degree |k] = ky +---+ km. A polynomial ort! --2tr (a F)POLYNOMIALS AND EQUATIONS 3 is called a monomial (not necessarily primitive). If P(x) is a poly- nomial in F[z1,... ,2m] given by (1.1.1) and P # 0, then we define the degree of P to be the maximum of the degrees of the monomi- als which occur in the polynomial. (Their coefficients a, are dif- ferent from zero.) Notice that the degree of P is 0 if and only if P(x) = az? ---2°, =a(a# 0). A polynomial P(2),... ,2m) in m variables can be considered as a polynomial in z; with coefficients in F[zz,... Zn]. The degree d, of such a polynomial is called the degree of P in zy. It is clear that dy is the largest integer occurring as an exponent of z; in a monomial o,2* with a, # 0. Similarly, we can define the degree of P in any variable z, (k = 1,...,m). Of course, these degrees dy are different from the degree of P, which is sometimes called the total degree. Example 1.1.1, The polynomial P(z1,22,23) = z}z3z3 + 2z4z2 has total degree 6, and has degree 4 in x}, 2 in z2, and 1 in x3. A polynomial P in the independent variables z1,... ,2m is called a homogeneous polynomial if and only if all its non-zero monomials are of the same degree. The degree of the homogeneous polynomial is called degree of homogeneity of the polynomial. If P(z1,... ,Zm) is a homogeneous polynomial of degree of homo- geneity d then P(tz,... ,t%m) = t?P(x,... +2m)- Example 1.1.2. 1° The polynomial P(z1,22,23) = 22123 - Sei2229 +23 is homogeneous of degree 3. 2° Wf P(x4, 22, 23) = (21 — 22)*(z1 — 23)*(z2 — x3)? we have that P(tz,, t22, tra) = #°P(21, 22,73). ‘Thus, P is a homogeneous polynomial of degree 6. Given the polynomials P(x), Q(z) € F{z], where Q(z) #0, then P(z) = Q(z)S(z) + R(z),4 GENERAL CONCEPTS OF POLYNOMIALS where $(z), R(z) € F{z] and R(z) = 0 or dg R < dg S. This property which is something like Euclidean Algorithm is called the Division Algorithm. If the coefficient of the highest power of a polynomial P(z) € F{z] is one, the polynomial is called monic. Thus, a monic polynomial of degree n can be written as P(2) = a9 +ayz +--+ +O,12"' +2". If P(z), Q(z) € Fiz] and Q(z) #0, then Q(z) divides P(z), writ- ten as Q(z)|P(z), if P(x) = $(z)Q(z) for some S(z) € F[z]. The polynomials P(z),Q(z) € F[z] are relatively prime if their greatest common divisor is one. The polynomial P(z) € F[z] of positive de- gree is irreducible in F{z] if, given any polynomial Q(z) in F[z], then either P(z)|Q(z) or P(z) is relatively prime to Q(z). Let K be an extension field of F. If r € K and P(z) = a9 +ayz+-++ tan2", then P(r) = a9 + ayn +-+- + an7” is an element in K. The element r € K is a zero of the polynomial P(z) € F[z] if P(r) = 0. Let T be a non-empty set and Fa field, and let F7 denote the set of maps t+ f(t) of T into F. As usual, f = g for all t and addition and multiplication of functions are defined by (fF + 9) = f) + 9(t) and (f9)(t) = S(O9(0)- Ifa € F then a defines the constant function a such that a(t) = a for all t € T. In particular, we have the constant functions 0 and 1. Taking that (—f)(t) = —f(t), one can show that FT is a commutative ting with unit (cf. Jacobson [1, p. 130]). Also, the map of F into FT which maps any @ € F into the corresponding constant function is an injective homomorphism (monomorphism). Identifying F with its image, F? becomes an extension of the field F.POLYNOMIALS AND EQUATIONS 5 Setting T = F we can consider the ring of maps of F into itself. Besides the constant functions, we mention here an important map — the identity function t + i(t) = t. Consider now the subring R{t] generated by the field of constant functions F and by the identity function t), The elements of this ring are called polynomial functions in one variable over F. Since the ting F{t] is generated by F and t there is the surjective homomorphism (epimorphism) of F[z] onto F[é], which is the identity map on F and sends z+ t. Here P(z) + P(t) and P(t) is the function tr ag t+ ait +--+ a_t™ if P(z) = ao + 4,2 +--- +a,2". The homomorphism P(z) > P(t) is an isomorphism if and only if F is infinite (cf. Jacobson [1, pp. 130-131]). The definition of polynomial functions in severa) variables is an immediate generalization of the preceding. In this book we deal only with polynomials over the field of com- plex numbers C (or the real numbers R), i.e., with the polynomial functions t ++ P(t), which map C to C. The next subsections are devoted to general concepts of algebraic polynomials on the field C and to the solution of algebraic equations of degree less than five. Certain solvable quintic equations will be considered in Section 1.1.6. Some useful references in this respect include the classical books of Dickson [1], Kurosh [1], Lang [2], Uspensky [1], van der Waerden [1], Herstein [1], and Jacobson [1] (see also Dummit and Foote [1]). 1.1.2. Polynomial Equations It is a main problem of algebra to find the solution of algebraic equations. A quadratic equation can be solved algebraically for all possible values of its coefficients. The difficulty of solving equations increases with their degree, if not for other reasons. The solution of the general cubic equation is credited to the six- teenth-century Italian algebraists Scipione del Ferro, Niccolé Tartag- lia, and Gerolamo Cardano. Their work is of great importance for the history of algebra. Scipione del Ferro never published his solu- tion; he only told a few friends. Among these was his pupil Antonio 1) We use the simple notation ¢ for the identity function t — t.6 GENERAL CONCEPTS OF POLYNOMIALS Maria Fiore, from Venice. Cardano was a famous medical doctor, astrologer, philosopher, and mathematician, who lived in Milan. All attempts during the following two centuries to find an alge- braic solution of equations of degree greater than the fourth without specializing coefficients in any way failed. The most important con- tributions of Carl Friedrich Gauss to the theory of algebraic equations are (ef. van der Waerden [1]) as follows: 1° The complete solution of the cyclotomic equation, 2"-1=0, by means of radicals. The equation is called cyclotomic because its solution is closely connected with the construction of a regular poly- gon of n sides inscribed in a given circle. The Pythagoreans knew how to construct regular polygons of 3, 4, 5 and 6 sides. Their constructions can be found in Book 4 of the Elements of Euclid. Gauss was still in the University when he discovered, in 1796, that a regular 17-sided polygon could be inscribed in acircle, using ruler and compass only?). During that period Gauss began to study at the University of Gottingen, where he had access to the works of Pierre de Fermat, Leonhard Euler, Joseph Lagrange and Adrien Legendre. He wrote a book on the theory of numbers, which appeared in 1801 as Disquisitiones arithmeticae, generally regarded as his most important accomplishment. 2° The proof that every polynomial in one variable with real co- efficients is a product of linear and quadratic factors. This theorem implies what we now call the fundamental theorem of algebra*): 2)Gauss considered it one of his master achievements and wanted a replica of his construction placed on his tombstone. Later, he proved that a regular n-sided polygon can be constructed with ruler and compass if, and only if, either (1) n is a prime number of the form 2" +1, or a product of distinct primes of this form, (2) nis a power of 2, or (3) n is a product of numbers satisfying conditions (1) and (2). On the basis of these notions, it has been shown that regular polygons with 257 and 65, 537 sides can also be constructed. In 1832, F. J. Richelot published a study of the regular polygon of 257 sides, and, a professor from Lingen, Germany, named Hermes spent 10 years of his life trying to construct a regular polygon of 65, 537 sides. Gersten and Stallings [1] observed that Gauss's original proof of the fun-POLYNOMIALS AND EQUATIONS 7 ‘THEOREM 1.2.1. Every polynomial z ++ P(z) with complex coef- ficients is a product of linear factors. Gauss showed in his doctoral thesis that numbers of the form a+ib, where a and 6 are real numbers and 2 is the square root of —1, can be regarded as analogous to points on a plane. If an algebraic equation of degree n of the form 4,2" + Op az”! +--+ +412 + ao = 0, with complex coefficients ao,@1,.-. ,a, and a, # 0 has n roots the numbers 21, 22, --., Zn, it is well known since F. Viéte that Gn-1 Hatite +i, =-—, Gn — Gn-2 aay tam te ttt, = SF, in ao 222° Z, = (-1)"—. Gp The above elementary symmetric functions (see Subsection 1.3.1) are denoted by 01, 02, -.-, On, respectively. It follows, very easily, that nz” + Gn 12" | + an2Z”? + --- +042 +0 =a,(z — 21)(z — z2)---(z— Zn) =a,(2" — 02") 4.022"? —--- + (-1)"on). damental theorem of algebra (see Gauss [1, pp. 1-31]) appeared to bear upon criteria for words in a free group to represent the identity element (see Struik [1]. Gersten and Stallings noted the following: It is remarkable that Gauss him- self was not able to fill in the details, except to give a plausibility argument at the point in the proof where the free group enters; the deep point in this argument involves the geometry of the 2-cell, especially a version of the Jordan curve the- orem which 18 used in the geometric theory of free groups. Smale {1] has given a beautiful account of the old and new developments on the fundamental theorem of algebra, as well as its relation to economics and numerical analysis.8 GENERAL CONCEPTS OF POLYNOMIALS In his treatise Waring [1] has proved that all rational symmetric functions of the roots can be expressed as rational functions of the coefficients of the equation. He has also derived expressions for the power sums ay t2zy+--- +20, and also for arbitrary symmetric polynomials. P. Ruffini published several papers claiming to prove that the general equation of the fifth degree or higher cannot be solved by radicals. His first treatise was published in 1898 in Bologna in a paper with title Teoria generale delle equazioni, in cut si dimostra impossibile la soluzione algebraica delle equazioni generale di grado superiore al quarto. Ruffini’s meth- ods were essentially those of J. L. Lagrange. Ruffini was a student and admirer of Lagrange. N. H. Abel in the spring of 1824 succeeded in proving that a solution by radicals of the general quintic equation is impossible. The proof can be found in Abel [1-2]. Abel has made use of the results obtained by Lagrange and Cauchy concerning the number of values a function of n variables can attain if the variables are permuted. In 1829, Galois had finished a very good part of his investigations on the solution of algebraic equations. In 1830, Galois [1] announced the following very significant theorem: THEOREM 1.2.2. In order that an equation of prime degree be solvable by radicals, it is necessary and sufficient that, if two of its roots are known, the others can be expressed rationally. This beautiful result implies that the general equation of degree 5 cannot be solved by radicals. However, for such equations whose co- efficients are numerically given, one can use numerical methods (e.g. Newton’s method, Bernoulli’s method, methods for simultaneous de- termination of zeros, etc.) to estimate their zeros. Smale [1] showed that a classical algorithm, Newton’s method with a standard modifi- cation, is a tractable method for finding a zero of a complex polyno- mia]. By tractable, Smale means that the cost of finding a zero does not grow exponentially with the degree, in a certain statistical sense. Shub and Smale [1] proved that the number of iterations required for a certain fast algorithm to find a zero of a complex polynomial of degree n is linear in n, provided that an arbitrarily small set of problems is excluded. Smale [2-3] gave an account of a global theoryPOLYNOMIALS AND EQUATIONS 9 concerning the efficiency of algorithms for solving systems of equa- tions. This approach introduces geometrical and topological ideas into numerical analysis. Therefore, our goal now is to give a method for solving quadratic, cubic and quartic equations following Ungar [1-2] (see also Uspensky fi). 1.1.3. The Quadratic Equation Let the two roots x1 and 2 of the complex quadratic equation (1.3.1) z? +a,z +a =0 be represented by the complex numbers a and , and let (1.3.2) r=atf, m=a-f. Tt follows that (1.3.3) Ya=-a, a -f? =a. From equations (1.3.3) we obtain @ = —a;/2 as well as a first degree equation for 6? from which +/ can be obtained. Thus _ ai 1 fy _ 4 1 fy (1.3.4) n= 9 tga — 420, m=-5 5a ap, which is very well known. Remark 1.3.1. The square root branch in (1.3.4) can be chosen ar- bitrarily. The choice affects only the order of the solutions z; and z2. Interestingly, a similar remark is applicable to the cubic and to the quartic equations, in 1.1.4 and 1.1.5, as well. 1.1.4. The Cubic Equation Let the three roots 21,22 and z3 of the complex cubic equation (1.4.1) 2° 422" 4+)2 +09 =0lo GENERAL CONCEPTS OF POLYNOMIALS be represented by the complex numbers a, fj and +, and let 2 =a+t+Goh +07, (1.4.2) zy =atnbt ny, z3=a0+@8+H7, where q,, k = 0,1,2, are the three cubic roots of unity: -14 iv3 ~iv3 are . mal, n= It follows that 2) +22+23 = -@2, (1.4.3) 2t24+21%3+%273= 1, 212274 = —ag. Thus 3a = —a, (1.4.4) 30? — 367 = 44, oF + f° +7° - 307 = —a9. From equations (1.4.4) one can find the sum 39(@° + 73) and the product 3°(4*7°), which are the coefficients of the quadratic equation (1.4.5) below, the two solutions of which are (38)° and (37)°, (1.4.5) (30) + (203 — 9a,a; + 27a9)(3v)? + (a3 - 3a,)* = 0. Let =a - 3a, (1.46) Ciels, a R= -203 + 9aya2 ~ 2709. From the quadratic equation (1.4.5) it follows that 3 Rt VR (sep =, R- RAG = (1.4.7) (3)? =POLYNOMIALS AND EQUATIONS n Thus a2 a 1 3f Rt R? - 4Q3 AB) = > 4fJ — (1.48) a5 — af R— y= v| 3 » IR ul 3 The cubic root branch in the second equation in (1.4.8) can be chosen arbitrarily while, by the second equation in (1.4.4), the cubic root branch in the third equation in (1.4.8) is such that By = Q/9. 1.1.5. The Quartic Equation Let the four roots 21, 22,23 and x4 of the complex quartic equation (1.5.1) 2! 4a32° +22" +012 +09 =0 be represented by m=atft7ts m2 =atp-7-4, a3 =a-fAty-6, tq =a-f-y+, (1.5.2) where a, f,7,6 are complex numbers. Then we have 4a = -a3, 60? - 6? +7°+6)= a2, (1.5.3) 4a° — 4a(6? + 7? + 8) +8676 = a, af + (0? +77 +8) — 20°(8? +7? +67) —4(674? + 676? + 7767) + Bay = ap. From equations (1.5.3) we obtain 2B +46), — 44(6% 7? + G0? + 776), 48 24262,12 GENERAL CONCEPTS OF POLYNOMIALS which are the coefficients of the cubic equation (1.5.4) below, the three solutions of which are (49)?, (47)? and (48)?, (16v?)® — (33 ~ 8a2)(160)? (1.5.4) +(3a4 ~ 162203 + 160,43 + 1603 — 64a9)( 160") — (a3 — 4azas + 801)’ = 0. Therefore equations (1.5.3) determine the value of a, a = —a3/4, and imply a third degree equation (1.5.4), for J, 7? and 6”. If we set P = a3 ~ 4aza3 + 8a, (1.5.5) Q = 12a + a3 — 30,03, R = 2agad — 9a,a703 + 20} — 72ag02 + 27a} and 8 a = 03 ~ 3c 2 3 (15.6) B= 4 Reve a ve _ 4 of R-YRPAG w=3— > then ee 7 1 B= 7 V 00 + Bo +40 (1.5.7) 7 1= 7Ve0 + 10 + G10» 1 5 = FV 00 + oho + 1170 where g; and q2 are the two non-real cubic roots of unity, the roles of which are interchangeable.POLYNOMIALS AND EQUATIONS 13, Any choice of the branches of the cubic root in (1.5.6) for which (1.5.8) Poo = 82 is allowed and, similarly, any choice of the signs of the square roots in (1.5.7) for which (153) pe=-& is allowed. The restriction (1.5.9) for the choice of the signs of 6,7 and 6 in (1.5.7) follows from the first three equations in (1.5.3). Ungar [2] has also obtained and proved the following characteriza tions: TweoreM 1.5.1. Let (1.5.10) a! 4432" + a92? +.a,2 +09 =0 be a real quartic equation and let Q = 12a9 + a3 — 3aja3, R = 27a9a4 — 9a,0203 + 2a} — 72agaz + 27a}, and 2 3 (1.5.11) 1 =e 4 8to( f EVE v=). With multiplicities counted, (1) if R? — 4Q3 > 0 then two and only two roots of (1.5.10) are real; (2) if R? -4Q® = 0 then two roots of (1.5.10) are real and the remaining two roots of (1.5.10) are real if and only if T > 0 for all the three possible cubic roots in (1.5.11); (3) if R? —4Q% < 0 then (a) the four roots of (1.5.10) are real if and only if T > 0 for all the three possible cubic roots in (1.5.11); and (b) the four roots of (1.5.10) are non-real if and only if T < 0 for at least one of the three possible cubic roots in (1.5.11).14 GENERAL CONCEPTS OF POLYNOMIALS THEOREM 1.5.2. 1° The cubic equation (1.4.1) has two equal roots if and only if RP - 49% =0, and has three equal roots if and only if R=Q=0, where R=—-2a3+9a203- 274, Q = 43-302; 2° The quartic equation (1.5.1) has two equal roots if and only if R498 =0 and has three equal roots if and only if R=Q=0, where R = 27agay — 9aya203 + 2a3 — 72apaz + 2707, Q = 12a9 +43 — 3a,03. Remark 1.5.1. The results in the above theorem can be found in the book of Dickson [1]. Their proof is immediate if based on the algebraic solutions of cubic and quartic equations as presented above. THEOREM 1.5.3. 1° The quartic equation (1.5.1) has two pairs of equal roots if and only if R’-4Q° =0 and = 32R = 27a}. 2° The quartic equation (1.5.1) has four equal roots if and only if R=Q=a=0.POLYNOMIALS AND EQUATIONS 15 1.1.6. The Quintic Equations Let z1,22,23,24,25 be the roots of the general quintic equation oy ayz* + ag? + az? + az +a = 0. By making a translation, we can assume aq = 0, i.e., that we can consider the equation (1.6.1) P(z) = 25 + p25 + qx? +rz+s=0. It is well-known that an irreducible quintic with coefficients in the rational numbers Q if solvable by radicals if and only if its Galois group is contained in the Frobenius group Fo of the order 20 in the symmetric group Ss. Recently, Dummit [1] has obtained a criterion for the solvability of such a general quintic (1.6.1) in terms of the ex- istence of a rational zero of the following explicit associated resolvent sextic polynomial (1.6.2) Pao(z) = 2© + Brz® + byt + bgz + bya? + biz + bo, where by =g® — 13pq®r + p'g’r? + 65p"q'r” — 4p°r® — 128p°q?r* + 17q¢r? + 48p'r! — 16pq’r* — 192p7r> + 256r° — 4p%q°s — 12p*q>s + 18p°grs + 12p%q°rs — 124q°rs + 196p'gr?s + 590pg°r?s — 160pgr°s — 1600gr4s — 27p's” — 150p'q?s? — 125pq*s? — 99p° rs? — 725p°q?rs* + 1200p*r?s? + 3250q?r?s* —2000pr°s? — 1250pgrs® + 3125p"s* — 9375rs*, by = — 2pg® + 19p?q'r — Stp'g?r? + 3q'r? + 32p'r? + 76pq?r? = 256p?r! + 512r® — 31p%q°s — 58g°s + 117p'grs + 105pg°rs + 260p" r?s — 2400gr°s — 108p"s? — 325p"q"s” + 525p'rs? +2750q"rs? — 500pr*s? + 625pqs® — 31258", by =p?q! — Gp'g’r — 8g'r + Sp*r? + T6pq?r? — 136p"r° + 400r4 — 50pq°s + 90p’grs — 1400gr”s + 625q7s” + 500prs”, bg = — 2g! + 2Ipq?r — 40p?r? + 160r° — 15p*qs ~ 400grs + 125ps?, bg =2pq" — 6p’r + 40r? — 50gs.16 GENERAL CONCEPTS OF POLYNOMIALS For the particular case when P(x) = 2° + az +6, this polynomial is simply Pyo(x) = 2° + 8az5 + 400?x* + 160032? + 400a*2? + (512a° — 312564)z + (256a° — 937546"). THEOREM 1.6.1. The irreducible quintic P(x) = 2° + px? +2? +rx+5 € Q[z] is solvable by radicals if and only if the polynomial Pzo(z) in (1.6.2) has a rational zero. If this is the case, the sextic Pzo(z) factors into the product of a linear polynomial and an irreducible quintic. When P(z) is solvable by radicals, Dummit [1] has also obtained for the roots 21,22,23,24,Z5 in terms of p,q,7,s which produce the roots in a cyclic order. 1.2. SOME CLASSES OF POLYNOMIALS 1.2.1. Self-Inversive Polynomials Let (2.1.1) ze P(z)= ane" =a, [] (z~ 2) val #=0 be a polynomial of degree n, with zeros z,22,.-.,2n- Define the polynomial z+ P*(z) by (2.1.2) P*(z) = 2" P(I/z) = Soa =a% Te -z) k=0 vl whose zeros z{ = 1/2, are the inverses of the zeros z, with respect to the unit circle |z] = 1. Suppose that the polynomial P has m {m < n) zeros in |2| < 1. Then any zero of P on the unit circle is also a zero of P*. However, if P has no zeros on the unit circle, then P* has also no zeros on |z| = 1 and has n — m zeros in the interior of this circle. For polynomial P* we say that it is the “inverse” of P.SOME CLASSES OF POLYNOMIALS: 17 DEFINITION 2.1.1. A polynomial z++ P(z) of degree n with zeros Ztye+. yn is self-inversive if (2.1.3) {21s225006 en} = {1/215 1/225-+- Mf Fn} THEOREM 2.1.1. Let z ++ P(z) be a polynomial of degree n. If there exists u € C, |u| = 1, such that P*(z) = uP(z), then the polynomial P is self-inversive. PRroor. From (2.1.1) and (2.1.2) it follows P*(2) = 2"P(1/2) = 2"PU/E) = 2am TT] é -a), val Pr(2)= a [I~ s5)= rs Ile ~ +), because of (2.1.3). Using the Viéte formula 222 ---2n = (—1)"ao/an we obtain that _ Gn 21.4 P*(z) = = P(z). (2.1.4) (= 7 Pe) Since for points on the unit circle |z| = 1 we have [P*(@)| = 2" PG/2)I = PL, we conclude from (2.1.4) that it must be |a,| = |ao]. Therefore (2.1.4) reduces to P*(z) = uP(z), where u=G,/ap and |u| =1. O n THEOREM 2.1.2. If P(z) = 3° agz*, an # 0, then the following k=0 statements are equivalent: 1° P is self-inversive. 2° @,P(z) = agz"P(1/z) for each complex number z. 3° a, = ue,_,, k= 0,1,...,n, where |u| = 1. The following result can be found in the paper of O’Hara and Rodriguez [1]:18 GENERAL CONCEPTS OF POLYNOMIALS n TweEoREM 2.1.3. If P(z) = ¥ axz*, an # 0, és self-inversive, k=0 then (2.1.5) a, [nP(z) — zP"(z)] = a92z""1 P'(1/z) and (2.1.6) ae -1f-1 for each z on |2| Proor. We obtain (2.1.5) by differentiating the equality 2° in Theorem 2.1.2. Taking z on the unit circle and using the equality |an| = lao], (2-1.6) follows immediately from (2.1.5). O Theorem 2.1.3 implies the following statement on zeros of P’ (see also Marden [1, p. 205]): TweoreM 2.1.4. If P is a self-inversive polynomial, then P' has no zeros on the unit circle |z| = 1 except at the multiple zeros of P. Using this result we can prove (cf. Marden [1, p. 204]): THEOREM 2.1.5. If P(z) = ¥ aaz*, an # 0, is a self-inversive k=0 polynomial, then P has as many zeros on the disk |z| < 1 as the polynomial n-1 Pi(z) = [Pa = So(n- Wine" k=0 That is, P and P! have the same number of zeros for |z| > 1. It is interesting to mention the condition for polynomials with all zeros on the unit circle (see Cohn [1}). THEOREM 2.1.6. A necessary and sufficient condition for all the zeros of P to lie on the unit circle is that P satisfy the conditions Gp = ud, — (K=0,1,-..42), where |u| = 1, and that all the zeros of P' lie in or on this circle.SOME CLASSES OF POLYNOMIALS 19 1.2.2. Algebraic and Trigonometric Polynomials A trigonometric polynomial in @ of degree n has the form (2.2.1) T(0) = ao + a; cos6 + by sin@ + ---+ a, cosné + 6, sinnd, where a, and b, are arbitrary complex coefficients. Sometimes, we put ao/2 instead of ao. If jan] + |b,| > 0 we say that n is the precise degree of the polynomial (2.2.1). ‘There are two particular cases of (2.2.1). Namely, if 6, = --- = 6, = 0 we have a cosine polynomial (8) = ag + a) cos@ + --- +a, cosnO. For ag = @ = +++ = a, = Oit is a sine polynomial $(6) = by sind + --- 45, sinnd, Consider now an algebraic polynomial z+ P(z) of degree n. Tak- ing z on the circumference |z| = 1, ie, z =e”, P(z) becomes a trigonometric polynomial T(8) of degree n, tn (8) = Ple*) = J (ax cosk# + by sin 8), k=0 with complex coefficients in the general case. An arbitrary trigonometric polynomial T(@) of the form (2.2.1), which is not identically zero, has the Haar property, i.e., it cannot have more than 2n distinct zeros in (-7, 7]. Indeed, putting z =e and using Euler’s formulae 1 = 5 Li =i cos v6 = ze" +e), sin vO = ale vi _ emiv), we obtain , T(6) = > engue™, where ‘ }(a_p+ib,), v <0, Cnty = 4 Go, v=0, }(a,- ib), > 0.20 GENERAL CONCEPTS OF POLYNOMIALS Thus we have 7 e*T(8) = Q(z), where Q is an algebraic polynomial of degree 2n. If T # 0, then T cannot have more than 2n distinct zeros in (—1, 7]. Consider now only real trigonometric polynomials‘). If we put =a, c=a,—ibh (v=1,.-.,2) where a,,b, € R, then we can represent a real trigonometric poly- nomial as the real part of an algebraic polynomial on the |z| = 1. Namely, n (6) = Re{ J 2"} = Re{ (a, - ib, )eiv*} 0 aa v=0 8 = a9 + )>(a, cos v6 +b, sin v8). val On the other hand, since ne{ oes") =) (0 + yer) we have n TI =n ty yg now (6) Liles +e2"")| ys Le, 7(0) = See where (2.2.2) Q(z) = tq t---+ G2" $ 2eoz2" eyz™* 4 penz™™. Notice that Q(z) = Q*(z) = 27"Q(1/z), i-e., that Q is a self-inversive polynomial of degree 2n. According to the above we conclude that (223) IF(@)| = 5106). ‘)Polynomials with real coefficients.SOME CLASSES OF POLYNOMIALS 21 Remark 2.2.1. If z, is a zero of the polynomial Q with multiplicity k, then 1/2 is also a zero of Q with the same multiplicity. We note that the inverses of 0 and 00 are 00 and 0, respectively, where co as a zero of the order k means that the coefficients of the k highest powers in Q vanish Remark 2.2.2. For a real trigonometric polynomial of exact degree n, the corresponding polynomial Q must be different from zero at the origin. Indeed, Q(0) = én # 0. If'dgT =m -1. The Chebyshev polynomials of the ae and second kind correspond to pa- rameters a = § = —1/2 and a = f = 1/2, respectively. Subsections 1.2.6 and 1.2.7 will be devoted to en polynomials. 1.2.3. Fejér-Riesz’s Representation of Non-negative Trigonometric Polynomials n Let R(z) = } 1,2” be an algebraic polynomial of degree n with =0 complex coefficients r, (v = 0,1,... ,n). Consider the square of the absolute value of P(z) on the unit circumference |z| = 1. Thus, let z=e (-1 << 7m). Since [R(e* )P = Rle*)R(e) = R(e")R(e~**) = (Sone) (Sore) = PP retelle oe v=0 n=O v=0 u=0 n=, me = =) nit > (o roeaFu)ei”? A (© Feral}, v=0 v=) u=0 B=0 we have n [R(e*)|? = a + Y>(a, cos v8 + b, sin v8), valSOME CLASSES OF POLYNOMIALS. 23 where 7 aoe (2.3.1) a= ort, ai = 2) reuFy (Y= 1-0). 10 u=0 Thus, |R(e'*)|? is a real trigonometric polynomial of degree at most n. Since ay, — ib, = 27,79, we see that this polynomial is precisely of degree n if ro #0 and ry # 0. If the coefficients of R are real, then by = --- = bn = 0 and the polynomial |R(e'*)|? reduces to a cosine polynomial. Starting from the previous consideration, in 1915 L. Fejér and F. Riesz (see Fejér [1] and Szegé [2]) gave a parametric representation of non-negative trigonometric polynomials. THEOREM 2.3.1. Let 6 ++ T(6) be an arbitrary non-negative tri- gonometric polynomial of degree n. Then there exists a polynomial z+ R(z) of the same degree as T(8) such that (2.3.2) (8) = |R(e)/?. Conversely, the expression |R(e**)|? always represents a non-negative trigonometric polynomial in @ of the same degree as the polynomial R. Proor. Let T be an arbitrary non-negative trigonometric poly- nomial of degree n. Then, according to (2.2.3), we have (233) 7(6) = ITO) = 510, where Q) is a self-inversive algebraic polynomial given by (2.2.2). Sup- pose that the non-negative polynomial T(@) has real zeros. Then, each real zero 8, (€ {—1,7]) of T(8) is of the even multiplicity, and etfs = (, is a zero of Q(z) of the same multiplicity. Regarding to the zero distribution of the polynomial Q (see Remarks 2.2.1 and 2.2.2), we can take the following factorization P q (2.3.4) Q(2) = enz* [] @- 22-22) Te - 6, vl ul24 GENERAL CONCEPTS OF POLYNOMIALS where kt p+q =n, k,p,q> 0, 22 = 1/2,0< |z,| < land [¢,| = 1.9 For p > 0,0 < |z,[ < 1, and z = e', we have i | z—-2al lz-1/% . z— s = 1, | Wat |r lz}. Thus, using (2.3.3) and (2.3.4), we obtain that pe a 7(0) = 2 Il Je 2)? I le — CP, 1 pst where pre lead i 211 lI vs This gives (2.3.2). The second part of the statement is obvious. O For cosine polynomials we have the following corollary of the pre- vious theorem: CoroLiary 2.3.2. Let 0 ++ C(8) be an arbitrary non-negative cosine polynomial of degree n. Then there exists a polynomial z + R(z) with real coefficients of the same degree as C(8) such that C(8) = [R(e!*)P. The representation (2.3.2) is not unique. Namely, if € is an arbi- trary zero of z+ R(z), then the polynomial z— =, _ 1-éz i zr Bz) = TER) furnishes another representation. Supposing that T(6) # 0, the using this method we can remove all the zeros from |z| < 1 and obtain the following result (cf. Szegé [2, p. 4]): S)If p = 0 or g = 0, the corresponding product in (2.3.4) should be equal to 1.SOME CLASSES OF POLYNOMIALS 25 THEOREM 2.3.3. Let @ ++ T(8) be an arbitrary non-negative trigo- nometric polynomial of degree n and T(@) #0. Then a representation T(6) = |H(e'®)[? exists such that z+ H(z) is a polynomial of the same degree as T, with 1° H(z) #0 in |2| <1, and 2 H(0)>0. The polynomial H is uniquely determined. If T is a cosine poly- nomial, then H is a polynomial with real coefficients. Example 2.3.1. Let C@)= Fn +1) +ncosé +(n— 1)eos28+-- + cos6, for which ( /2)\2 _ Ipsin(n +198, cw =5(- sin(0/2) yo Here, we have co = (n + 1)/2, c1 =n, -.., cn = I, and therefore Qe) = 14 224-421 4 (nt Aye” en2™th gp ae2mol yy Pm ie., QAz)=(l4z4---42"7. ‘The zeros of the polynomial Q are zy = exp(i2mu/(n + 1)), w= 1,.-. ,n, with multiplicity two. We note that |z,] = 1, # = 1,... ,n, so that H(z)= +zte-$2") 1 a" and €(6) = |H(e%*)/? 1.2.4. Non-negative Algebraic Polynomials Consider a real algebraic polynomial of degree n zu P(z)= > a,z*, 2026 GENERAL CONCEPTS OF POLYNOMIALS which is non-negative for every z € R. Then its degree n must be even number, i.e., 2 = 2m, and zeros 2_ = Ek +%Mky Impk = Ek ky k=1,...,m,so that Plz) = an] ((2-&) +f] (an > 0). k=1 Using the identity (a? + 8)(c? +d?) = (ac + bd)? + (ad — be)’, we can prove the following representation: THEOREM 2.4.1. Every non-negative polynomial P on the real line R can be expressed in the form P(z) = A(z)? + Biz)’ = (dgP =n = 2m), where z ++ A(z) and z + B(z) are real algebraic polynomials of degree at most m. If P(z) > 0 when z € [0,+00), the following theorem is valid (cf. Karlin and Studden [1, Ch. 5] and Timan [1, Sect. 4.11]. THEOREM 2.4.2. Every algebraic polynomial P of degree n, which is non-negative on [0,+00) can be uniquely expressed in the form P(z) = A(z) + 2B(z)’, with [n/2] {41/2} Az)=a [[ (2-6), Ble)=8 J] (z-m), v=1 va? a,6>0, m=0, w B(z) are algebraic polynomials of degree at most n and n—1, respectively, we obtain the following representation of a non-negative polynomial on [—1, 1] P(z) = A(z)’ + (1 — 2?) B(z)*. Also, an arbitrary polynomial of degree n which is non-negative on [-1, 1] can be expressed in the form (cf. Pélya and Szegé [1, Ch. 6]) P(x) = A(z)? + (1 + 2)C(2)? + (1 - 2) D(z? + (1 - 27) Biz)’, where A, B,C, D are algebraic polynomials. If dg P = n = 2m, then there exists a representation for which B(z) = C(z) = 0, dg A = m, and dg D=m-1. More generally, Videnskii [1] proved (see also Bernstein [2]): THEOREM 2.4.3. Let z ++ P(z) be an arbitrary polynomial of degree n which is non-negative on [-1, 1]. Then the following repre- sentations uniquely ezist: 1° For each m > nf2, P(z) = A(z)’ + (1 - 27)B(z)’, where x ++ A(z) and z + B(z) are real polynomials of degree m and m— 1, respectively, their zeros lie in [-1,1] and mutually separate each other, and A(1) > 0, B(1) > 0;2B GENERAL CONCEPTS OF POLYNOMIALS 2° For each m > (n—1)/2, Pla) =(142)C(2) + (1- 2)D(2)*, where z ++ C(z) and z+ D(z) are real polynomials of degree m. All the zeros of z+ V1 +4 2C(z) and z+ /1 —ZD(z) lie in [—1,1] and mutually separate each other, and C(1) > 0, D(1) > 0. 1.2.5. Lorentz Representation of Polynomials Let z + P(z) be a real algebraic polynomials of degree n. A Lorentz representation of the polynomial P on the interval [a, }] is a Tepresentation of the form t (2.5.1) P(z) = Yo eu(b- 2)*(z - a)". v=0 It is easy to see that (2.5.1) is not unique in general; for example, on [0,1] for every € > 0 we have that é 12 (‘a 2). v=0 However, if we insist in (2.5.1) that € be equal to the degree of P, then such representation becomes unique. In 1915 Bernstein [1] observed that every polynomial z + P(z), having no zeros in (a,b)® for sufficiently large ¢, can be expressed in the form (2.5.1), with all c, > 0 or all c, < 0. The smallest natural number ¢ for which such a representation holds is called the Lorentz degree of P and it is denoted by €,,4(P) or simpler €(P). The Lorentz degree €(P) can be much larger than the ordinary degree dgP. It is interesting to mention that every polynomial of degree n with integer coefficients has the Lorentz representation (2.5.1), with all integer coefficients c,, and in which € = dgP =n. Recently Lubinsky and Ziegler [1] have investigated the relation- ship between the uniform norm of P on (0, 1] and |c,|,v = 0,1,...,€. ®)Bernstein stated this result for (a,b) = (0,1).SOME CLASSES OF POLYNOMIALS 29 In approximation theory, Lorentz [1] introduced the classes of poly- nomials £(a,b) = {P | the polynomial P(z) has the form (2.5.1), with all c, > 0 or all c, < 0} and obtained several interesting results. The classes &¢(—1,1) were studied by Erdélyi and Szabados [1]. THEOREM 2.5.1. A polynomial not identically zero belongs to the class £¢(—1,1) if and only if it has no zeros in the interval (—1, 1). The “only if” part of the statement is trivial: a polynomial from L (1,1) not identically zero cannot have zeros in (-1,1). The “if” part is more difficult and it will follow from the next theorem, which can be considered as a quantitative version of Theorem 2.5.1. Let P,, be the set of all real algebraic polynomials of degree at most nand let z + y(z) be a positive continuous function on (—1, 1). Also, let Diy) = {z=2+iyEC [ ly < lz), zl <1} and Waly) ={PEPa| P(z) #0 in D(y)}, &n(y)=_ sup &(P). Pew, (vy) Erdélyi and Szabados [1] proved the following results: THEOREM 2.5.2. If (28.2) 2 1 con >) << <> wegen} Ww) Ss where a (lal < 1) is a point where the infimum in (2.5.2) is attained,30 GENERAL CONCEPTS OF POLYNOMIALS THEOREM 2.5.3. Let C be the open unit disk of the complez plane. 1° If D(y) 2 C, then (2.5.3) Gly)=n forall n=1,2,... 2° If (2.5.3) holds for some n > 2, then D(y) 2 C. Theorem 2.5.3 was improved by Erdélyi [1]: THEOREM 2.5.4. Under the condition (2.5.2) we have SF 1, then &(P) = dg P= k + 2m if m is large enough. The Lorentz degree of trigonometric polynomials was also intro- duced by Erdélyi and Szabados [2]. To end this subsection we mention a result of Laguerre: If z > P(zx) is a positive polynomial for x > 0, then it can be expressed in the form P(x) = Q(z)/S(x), where Q and S are polynomials with non-negative coefficients. Bernstein (1] proved that one can always put S(z) = (1 +z), providing s to be sufficiently large. In other words, if P(z) is positive when z > 0, then we can always take s to be large enough, such that all coefficients of the polynomial z + Q(z) = P(z)(1+z)* be non-negative. Indeed, if we put z = y/(1—-y), then _ Ry) ) _ RY Pe) = (ayy where R(y) > 0 for 0 < y < 1. Therefore, (dg P = n), e Rly) = Yeu = y)Py, v=0SOME CLASSES OF POLYNOMIALS 31 with all non-negative c, and sufficiently large é. However, z ie, 1 _ i Tear’ Y= Ty y= gives fon =n, yo = fm Plz) = Yeu wy” Yoon (=) ' Thus, t Qz)= Vea’ and S(z)=(1 +2)" v=0 are polynomials with non-negative coefficients. 1.2.6. Basic Properties of Orthogonal Polynomials In next two subsections we give a short account of the basic prop- erties of orthogonal polynomials. In general, we refer to the books of Szegé [2], Freud [1], Chihara (1], Suetin [1], as well as to the papers of Nevai [1] and {3}. In the sequel let ,, be the set of all algebraic polynomials P (# 0) of degree at most n and let o:R — R be a fixed non-decreasing function with infinitely many points of increase for which all moments He = fat* do(t), k = 0,1,..., exist and are finite. Then the improper Stieltjes integral f, P(t)do(t) exists for ev- ery polynomial P. By the application of the Lebesgue-Stieltjes in- tegral fy f(t) do(t) to characteristic functions of sets, the function ¢ engenders a Lebesgue-Stieltjes measure do(t), which is known also as m-distribution (cf. Freud [I]). Moreover, if t + o(t) is an ab- solutely continuous function, then we say that o’(t) = w(t) is a weight function. In that case, the measure do can be expressed as do(t) = w(t)dt, where the weight function t + w(t) is a non-negative and measurable in Lebesgue’s sense for which all moments exists and Ho = fy w(t)dt > 0. Remark 2.6.1. In the general case the function o can be written in the form o = Gac + 0s + 9), where Cac is absolutely continuous, ¢s is singular, and g; is a jump function.32 GENERAL CONCEPTS OF POLYNOMIALS The support of the measure, i.e., the set of points of increase of t > a(t) we denote by supp(do). It is always an infinite and closed set. If supp(do) is bounded, then the smallest closed interval containing supp(do) we will denote by A(do). For example, if A(de) = [a,6] then we say that do(2) lies in [a,b]. In that case we have o(t) = o(—00) for t < a and o(t) = a(00) for t > a. In addition, if t > a(t) is absolutely continuous, then the weight function t+ w(t) vanishes outside of (a, 6], or more generally, outside of supp(da). For any m-distribution do(t) there exists a unique system of or- thonormal polynomials m4(-) = (-;do), k = 0,1,---, defined by T(t) = byt* + cyt*-! + lower degree terms, bx > 0, 2.6.1 ae) (ky Tm) = bkms km 20, where b;, = (do), cx = c,(do) and the inner product is given by (ha)= f 1att) ate (fg € P°(R) = L?(R;do)). Ifwe have an absolutely continuous function ¢ + a(t), then instead of m4(-3a), 6,(do), supp(do), ..., we usually write m(-;w), b.(w), supp(w), ..., respectively, where o'(t) = w(t). If we have supp(w) = [a,], where —0o < a < b < +00, we say that {1,} is a system of orthonormal polynomials in a finite interval {a, 6]. For (a,6) we say that it is an interval of orthogonality. Now we give a few basic properties of orthogonal polynomials: THEOREM 2.6.1. The system of orthonormal polynomials {n,}, associated with the distribution do(t), satisfy a three-term recurrence relation (2.6.2) tmy(t) = Unga mega(t) + vam (t) + unme1(t) (k > 0), where x(t) = 0 and the coefficients up = ug(do) and vy = v,(d0) are given by mat and ye [ tra(t)? do((t). by rRSOME CLASSES OF POLYNOMIALS. 33 Since mo(t) = by = 1//fo and by = ub, we conclude that by = bo /(urug --- U4). In some considerations we use the monic orthogonal polynomials f(t) = m,(t)/b, = t*+ lower degree terms. Such polynomials satisfy the following three-term recurrence relation fear (t) = (t~ on)ta(t) — Bettea(t), = 0,1,2,..., #a(t)=0, fo(t)=1, where a, = vz and By = uz > 0. Since (Fet1s Fx) = (tits Fe) — On (iE, He) — Bale Fe), (Fears Fea) = (tite, Fe—1) — OK (es Te—1) — Ba(Fe1, 7-1), and (they ear) = (testa) = (ies tte) + Ona (H es He—1) + Baa (ths Fe-2)s because of orthogonality (2.6.1), we obtain that _ (titky x) — (Fert) on = GEEH (E20, B= GUE (b> 0). The coefficient Ho, which multiplies #_, = 0 in three-term recurrence relation may be arbitrary. Sometimes, it is convenient to define it by Bo = wo = J, do(t). Then the norm of i, can be express in the form (2.6.3) l#all = V(x, 4) = VB 0A --- Br - THEOREM 2.6.2. All zeros of t+ m,(t;do), n > 1, are real and distinct and are located in the interior of the interval A(do). Let a, k=1,...,n, denote the zeros of ,(t;do) in increasing order 1h) gh crc oO),4 GENERAL CONCEPTS OF POLYNOMIALS THEOREM 2.6.3. For the zeros of ,(t;do) and mn4i(t;do) we have TY cr? (a1. as EN). Define the function (t,z) + K,,(t,z) by n=l (2.6.4) K,(t2)= Do m()r(z) (n> 1), v=0 which plays a fundamental role in the integral representation of par- tial sums of the orthogonal expansions. Namely, A’,(t,z) represents the kernel of the n-th partial sum. Notice that K,,(t,z) = K,(z,!). Using the three-term recurrence relation from Theorem 2.6.1 we can prove: THEOREM 2.6.4. Let K,,(t,z) be defined by (2.6.4). Then we have (2.6.5) K,(t,2) = ug Tel tdeat(2) teal thrafe) , where u,, is defined in Theorem 2.6.1. Letting t — x we find K (2,2) = un(™,(2)tn—1(2) — 7,-1(2)R(2))- Formula (2.6.5) is known as the Christoffel-Darbouz identity. The Christoffel function \,,(do) corresponding to a is defined by nd -1 An(z;do) = (x in(zidoy) (n= 1,2,...). k=0 It is easy to see that A,(z;do)~} = K,(z,z). This function is con- nected to the Christoffel numbers AS), k = 1,...,n, which appear in the Gauss-Christoffel quadrature formula (see Gautschi (1]) [14000 = DAL) + BaD k=l which is exact for any polynomial f of degree at most 2n — 1, i.e., Ry(Pan—1) = 0. Here 7!" = 7 (do) are zeros of the polynomial tr n,(t;do) and A) = M(do) = An(7;do) (k= Ay... sn). The following extremal problem is related to the Christoffel func- tion.SOME CLASSES OF POLYNOMIALS Ky THEOREM 2.6.5. For every P € Py_, such that P(z) = 1, we have (266) [PCP date > raCaido), f with equality only for — peggy — Kaltez) (2.6.7) P= P= Rey Proor. Let x be a fixed number and P € P,,_;. In order to find the minimum of the integral on the left hand side in (2.6.6) under the constraint P(z) = 1, we represent P(t) as a linear combina- tion of the orthonormal polynomials 1,(t) = 7(t;de), i-e., P(t) = nol DX cxm,(t), and consider the associated function (co, ¢1,--- ,€n—1) > k= F = F(co, 015+. Cn—1)s viz. r= [ (= ard) do(t) - o( nl a n—1 i 1(Sente) - i), i=0 50 oA vem ~- 1) I te where pis a Lagrange multiplier, whose value we have to determine. We must have ar Fag = te bma(2)=0, = O41,....m— 1, which yields nl LB n-1 n-l nl Vearde)= FY real and Yeh = EY con(2). k=0 k=0 k=0 k=0 Using the constraint P(z) = 1 we find 1 _ _m(z) _ Kates)’ 7 Ki(a,z) Ob 1) B 2Rid GENERAL CONCEPTS OF POLYNOMIALS where Ky is defined by (2.6.4). Since A,(z;do)"! = Kp(z,2) and <1 "Sek = y/2, we conclude that F attains its minimum Fria = k=0 An(x; do) only for P(t) = P*(t), where P* is given by (2.6.7). O According to this theorem, the Christoffel function can be express also in the form An(zido) = min fi IP(O)F? do(t). Pizyet® Many results on Christoffel functions can be found in the paper of Nevai (2, pp. 57-123]. Nevai also introduced and studied the generalized Christoffel functions, Ag(zirde) = mi [ |P()['do(t) (0 0. These polynomials were introduced and studied by Szegé [1]. The monic polynomials {¢;} on the unit circle satisfy the recurrence re- lation (but not three-term relation as before) desa(z) = 2alz) + d4(0)z*4a(1/z), k= 0,1,2,.... For more details consult Szegé [2], Freud [1], Nevai [2-3], Maté, Nevai, and Totik (1-2], and Jones, Njastad, and Thron (1}. Similarly, orthog. onal polynomials on a rectifiable curve or arc lying in the complexSOME CLASSES OF POLYNOMIALS 7 plane can be considered (cf. Geronimus [1], Szegé [2]). Also, complex orthogonal polynomials may be constructed with double integrals. Namely, introducing the inner product by (f.9)= | f f2)\o(2)w(z) dzdy, / for a suitable positive weight function z ++ w(z), where B is a bounded region of the complex plane, a system of orthogonal poly- nomials can be generated (see Carleman [1] and Bochner [1]). Recently, Gautschi and Milovanovi¢ [2-3] introduced a type of or- thogonality, the so-called orthogonality on the semicircle The inner product is defined by (2.68) (h)= i Hle)a(e\izy de, where I" is the semicircle F = {z € C : z = e”,0 < 6 < ny}. Alternatively, (2.6.8) can be expressed in the form thad= [se ae'*)a8 Note that the second factor g is not conjugated, so that the inner product is not Hermitian. However, the corresponding (monic) or- thogonal polynomials exist uniquely and satisfy a three-term recur- rence relation. The general case of orthogonality with a complex “weight function” was considered by Gautschi, Landau, and Milo- vanovié [1] (see also Milovanovi¢ {1-3]). 1.2.7. The Classical Orthogonal Polynomials A very important class of orthogonal polynomials on an interval of orthogonality (a,b) € R is constituted by so-called the classical orthogonal polynomials. They are distinguished by several particular properties. The classical orthogonal polynomials {Q,.} on (a,b) can be specifi- cated as the Jacobi polynomials P(t) (a, 8 > -1) on (-1, 1), the3 GENERAL CONCEPTS OF POLYNOMIALS generalized Laguerre polynomials L(t) (8 > —1) on (0, +00), and fi- nally as the Hermite polynomials Hy(t) on (—00, +00). Their weight functions ¢ ++ w(t) on (a,6) satisfy the differential equation F(ACiw(t) = BCe)u(t), where the functions t+ A(t) and t + B(t) are defined as in Table 2.7.1. For such classical weights we will write vw € CW. ‘TABLE 2.7.1 The Classification of the Classical Orthogonal Polynomials (a,b) w(t) A) BY) At Ci) G-9(+)? 1-@ Boa-l(a+p+ayt Ree eos) (0, +00) tren" t s+i-t (~00, +00) et 1 =2t fe The classical orthogonal polynomial t ++ Q;(t) is a particular so- lution of the differential equation of the second order (27.1) Ly] = A(t)y" + B(t)y' + Axy = 0, Le, (2.7.2) i (cnn?) + Agu(t)y = 0, where (2.73) de = —k (3(b- 1)A"(0) + BY(O)) The m-th derivative of Q; is also the classical orthogonal polynomial, with respect to the weight t+ w,,(t) = A(t)"w(t), and satisfies the differential equation (2.7.4) 4 (Aerentn$t) + Ak.mUm(t)y = 0, where (2.7.5) kim = —(k — m) (3(k + m — 1)A"(0) + BY(0)) .SOME CLASSES OF POLYNOMIALS 39 We note that (2.7.5) reduces to (2.7.3) for m = 0, i.e., Axo = Ak- Remark 2.7.1. The characterization of the classical orthogonal poly- nomials by differential equation (2.7.1), i.e. (2.7.2), was proved by Lesky [1], and conjectured by Aczél [1] (eee also Bochner [2]). Special cases of the Jacobi polynomials are: 1° The Legendre polynomials P(t) (for a = 6 = 0); 2° The Chebyshev polynomials of the first kind T(t) (for @ = B= -1/2); 3° The Chebyshev polynomials of the second kind $,(t) (for = 1/2); 4° The Gegenbauer or ultraspherical polynomials CA(t) (for a= 6 =- 1/2). If s = 0, the generalized Laguerre polynomials reduces to the stan- dard Laguerre polynomials L;(t). The classical orthogonal polynomials possess a Rodrigues’ type formula (cf. Tricomi [1] and Suetin [1]) (2.7.6) Qk(t) = = (aqoytu(o), wl ati ae where C, are constants different from zero. Using the Cauchy formula for k-th derivative of a regular function, (2.7.6) can be represented in the following integral form A(z)Fu(z) 4 2.7.7 thet. g AP (2.7.7) Qx(t) = a x P(e oen @ where I’ is a closed contour such that ¢ € int I’. The constants Cy, in (2.7.6) and (2.7.7) can be chosen in different way (for example, Q, to be monic, orthonormal, etc.). A historical teason leads to or for POM), 1 for L(t), (-1)* — for Hy,(t).40 GENERAL CONCEPTS OF POLYNOMIALS In addition, the Gegenbauer and the Chebyshev polynomials need che = art, Pee) (a= A= 1/2), ni = Gy em, ask sig = Te mM, k where (5), is the standard notation for Pochhammer’s symbol T(s +k) T(s) For such defined polynomials Qx(t) = ax(t* + rat*-! +--+), we give the leading coefficient a;, the coefficient r,, and the norm ||Qxll- 1° Jacobi case: _(kKtatBt Ie _ Ka-f) aE > TRS OR Fat Bp (s)k = 8(8+1)---(stk-1) = (I is the gamma function). ypteOye = Zeke + DMEF A+) H@k+at+h+IK(kta+pFi)’ 2° Gegenbauer case: = Oy eH 0, CNP = ve EOLA) (e+ ETO) ? 3° Legendre case: — (2h)! 7 2 oe eee? =0, IPI’ = a ii 4° Chebyshev case of the first kind: =P, 4 =0, (al? =", ITeIP= 5 (& #0)SOME CLASSES OF POLYNOMIALS Al 5° Chebyshev case of the second kind: a=2, %=0, ISP = z 5° Generalized Laguerre case: a, =(-1)', re =—k(k+5), [LEIP = KI(R +941); 6° Hermite case: an =2*, re = 0, Hall? = 2*kIV The Jacobi polynomials can be expressed in the following explicit form k PAN) = ry Gx) (: ee yee. We note that Plt) = (—1)#P(?-")() and ply = Ce) _@ a ; Using the integra] representation (2.7.7) one can determine a gen- erating function for the classical orthogonal polynomials, +00 = ,2)-> oH) ot, k=0 where Q;(t) = Qx(t)/Cy. For example, for the Gegenbauer, the generalized Laguerre, and the Hermite polynomials we have (2.7.8) (1-2t2 $2?) = Sct = a- a)“ ent (0-2) = a va 5 e(2", = (2.7.9) etre = x alate, k=042 GENERAL CONCEPTS OF POLYNOMIALS respectively. Remark 2.7.2. Milovanovié and Djordjevié [1-2] introduced and stud- ied the generalized Gegenbauer polynomials {p},,} defined by the gener- ating function too (1-2t 2") = Vo ph mite", k=0 where m € N and A > -1/2. The explicit form of these polynomials is [kfm] AVk—(m =I) poy k—me Phat = Ye Cu Sete antm The cases m = 1 and m = 3 were considered by Horadam [1-2] and Horadam and Pethe [1]. Similar results for the generalized Hermite poly- nomials have been obtained by Djordjevié [1] and Milovanovic [4] Using the generating function (2.7.8) and some manipulations we may prove the following recurrence relations for the Gegenbauer poly- nomials (e+ NCR ACO) = 2k + ACR ~ (+24 ~ CRA. G1) “ee nOaW = Achy — ache, 2d(1 = PCM = (24 + ACRE) — (+ CAA) (2.7.11 ee = (E41 + DYCH A) — (b+ CR 20), Starting values are given by C2(t) = 1, CA(t) = 2. The Gegenbauer polynomials can be expressed in the following explicit form AL ayy cio EGER v=0 Notice that ~1)F(X CR) = FM and Chys(0) =0..SOME CLASSES OF POLYNOMIALS 43 We also need the formulae C}(—t) = (-1)*C(t) and k+ 24-1 (2d (2.7.12) CR) = ( Z ) = Be. The three-term recurrence relation for the Legendre polynomials P, is known as the Bonnet’s relation: Peat) = Shy Eo rat, PAN = 1, AUD= 1, The corresponding relations for the Chebyshev polynomials T;, and S, are the same, i.e., Trar(t) = 24Te(t)—Tha(t), Tot} = 1, Tilt) = 2, Shar(t) = 2tSk(t)— Ska(t), — So(t) = 1, Sa(t) = 2. Remark 2.7.3. At the end of Subsection 1.2.2 we listed expressions for T,(t) and S,(t) for k < 5. Using the norm of C}, ae (+ A)AQ) T(A+1) Vat(x + 1/2)" where I is the gamma function, we can obtain the coefficients u, in the corresponding recurrence relation (2.6.2), for the orthonor- ma] Gegenbauer polynomials, as well as the coefficients b, in (2.6.1). Thus, _1 ] Ak+2~1) ag OO + De “k= OV (e+ QE TA—1)’ B20) Ke In this case, the coefficients vu, in (2.6.2) are, evidently, equal to zero. Of interest is the limit behavior of C(t) when ’ — 0, given by (see, e.g., Szegd [2, p. 81]) lim CH) _ a0 hy = |ICAIP = CRO), AQ) = (2.7.13) emi, k=1,2,...,a4 GENERAL CONCEPTS OF POLYNOMIALS where T(t) denotes the Chebyshev polynomial of the first kind. Using the hypergeometric function +00 a” (2.7.14) 2Filabie2) = > ore oo v=0 - . one can give the following representations of the Gegenbauer polyno- mials (cf. Bateman and Erdélyi (1, p. 176]) (2.7.15) R(t) = C2.) 2F(—k, K+ GA + 1/2;1- 0), (2.7.16) Chega(t) = Coeyr(l)taFi(-k, k++ 1544 1/2;1-0). Remark 2.7.4. Notice that expansion (2.7.14) for a = —k reduces to a polynomial. For the generalized Laguerre polynomials L(t), given by k = Deu (Yore ene we have in) =k #14) Het MLO, with Lg(t) = 1 and L(t) = s+ 1-1. The corresponding orthonor- mal polynomials 74(t) = bgt +--+ satisfy the three-term recurrence relation (2.6.2) with uy = Vek +5) and vy, =2k+s4l. The coefficient by is given by b. = 1//kIT(k+ 5+ 1). Finally, for the Hermite polynomials H(t), given by (ef) Hy(t) =k! x ERs ort (2th,SOME CLASSES OF POLYNOMIALS 48 we have Haqa(t) = 2tHe(t)—2kHea(t), Ho(t)=1, H(t) = 2t. The corresponding coefficients for the orthonormal Hermite polyno- mials are given by k [2 w= u = 0, b= 6 Now we give the coefficients ay (k > 0) and By (k > 1) in the three- term recurrence relation for the monic classical orthogona) polyno- mials Q;(t) on (a,6), Qnar(t) = (t- an)Qe(t) — BeQur(t) (k 20), Gilt) =0, Qolt) = 1. (2.7.17) Also, we give the moment pio = f° w(t)dt (w € CW). 1° Jacobi case: P{)(t) = 2k! /((k +0 + B + 1x) P(t), _ tT 0 + TG +1) (> T(a +f +2) y Be -a? Ok = mh? (2k +04 f)(2k+a+h +2) = Ak(k + a)(k + Bk +0 +B) © (2k +0 + B)*{(2k + a + BY? - 1)" 2° Gegenbauer case: CR(t) = kY/(24(A),)CAt), _VMQ+12) | gg _Me+2A~1) = Fa) 7 TO B= GEE Aa 3° Legendre case: P,{t) = 2*(k!)?/ ((2k)!) P(t), R Ho=2, o =0, f= ae-:46 GENERAL CONCEPTS OF POLYNOMIALS 4° Chebyshev case of the first kind: To(t) = 1, T(t) = 2'-*Ti(t), 1 1 Ho =m, a =0, A=5, Pr=y (k > 2). 5° Chebyshev case of the second kind: §;(t) = 2-*Si(t), (k> 1). aie mo =, a, = 0, Pe = 5° Generalized Laguerre case: £z(t) = (1) Lj(t), Bo =T(s+1), oe =2k4+s4+1, Pe=k(k+5). 6° Hermite case: H,(t) = 2-* H(t), k wo=Vvr, an=0, Be=a- The norm of monic orthogonal polynomials can be calculated using (2.6.3). At the end of this subsection we mention the following formula for the monic orthogonal polynomials (2.7.18) Ay Gal) = (pat + an)Ou(t) + we BOnal?)s where A(t) is given in Table 2.7.1 and the coefficient i, is the same as in the three-term recurrence relation (2.7.17). In the Jacobi case we have _ a-8) Pe=—k, %- eat Bp’ wo, = 2ktatH+1, In the generalized Laguerre case and the Hermite case, formula (2.7.18) reduces to Sing =kE() + kk +s)Ei_,(t) and Say =ki alt), respectively.SOME CLASSES OF POLYNOMIALS 47 1.2.8. Some Appell Polynomial Systems Suppose that the function z ++ g(z) can be expanded in a potential series +00 (2.8.1) Az)= Part, a =1. 0 Appell polynomial systems are defined by $00 (2.8.2) o(z)e* = S> pa(t)z*. k=0 In fact, it follows from (2.8.1) and (2.8.2) that a polynomial system {px} is Appell if and only if for k = 0,1,.-. k Y (28.3) m= Daw or = peal v=0 holds. Taking g(z) = 1 wesee that the simplest Appell polynomial system is the system of the powers {t*/k!}. Also, the Hermite polynomials {H,(t)/2*k!} form an Appell system. Indeed, this follows from (2.7.9) setting g(z) = exp(—z?/4). In this subsection we will consider two particular polynomial sys- tems: 1° The Bernoulli polynomial system {By(t)/K!}, defined by ret SS Belt) (2.8.4) = ra (lz| < 22); k=0 2° The Euler polynomial system {Ey(t)/k!}, defined by 2etz FR Bt) ea17 2X rd (|z| < x). (2.8.5)48 GENERAL CONCEPTS OF POLYNOMIALS The coefficient B, in the expansion = Bk ak 7 k=0 (2.8.6) az) = = is called the nth Bernoulli numbers. The numbers B, are rational. In particular, Boal, B=-3, Bee, 1 > and Borys = 0 for k > 1. Remark 2.8.1, The Bernoulli numbers can be expressed by Euler’s formula 1 2(2k)! (2.8.7) By, =(-1) (ene (2k), where +00 1 = Rez>) font is the Riemann zeta function. From (2.8.7) we can conclude that for all k > 1 the following inequality (—1)**" Bo, > 0 holds. Using (2.8.3) the Bernoulli polynomials t + B,(t) can be expressed in the form E(k H(k Bilt) = if Baar => ( aa. io v=o In particular, Bolt) = 1, BQ) =t-3, B)=#-t45, 5 32 1 -7 3 2 1 B(t)=P-5t +58 Bata +e 30° Sy os pal =e 2ty2p_l B;(t) = gh tae otSOME CLASSES OF POLYNOMIALS 49 Now, we give some basic properties of the Bernoulli polynomials and numbers (cf. Nérlund [1], Abramowitz and Stegun [1], Gel’fond [1], Luke [1], and Dilcher {1]). The Bernoulli polynomials satisfy the functional equations By(t-+1)— Be(t) = kt) and B(1—t) = (-1) x(t), and the derivatives satisfy By(t) = kKBy_-i(t). Some special values of these polynomials are By (0) = (-1)*Bx(1) = Be, Be (1/2) = -(1 - 2'-*) Be, Bas( 1/3) = Baa(2/3) = -3(1-3°-**) Bay, For 0 <¢ < 1 and k > 1 the following Fourier expansions Bu(t) = (1) eee py = 22k — 1)! 5 sin(2mvt) Box-a(t) = (-1)* (ony pik-t hold. Also, we mention the following interesting integrals Besi(z) — Brsi(e) it B; 4 (t) dt = +1 ai +1 ; if By(t)Bmn(t) dt = (—1)*-'———_ Bypm (ky € N). as = Using the functional equation f(t/2) + f((L = £)/2) = 2-H FQ) (k= 0,1,...), Haruki and Rassias [1] proved the following integral representations Box(a) = (— yp AED (log 2 yrt-alog(a? —Pecos(ra) +1) a,50, GENERAL CONCEPTS OF POLYNOMIALS 2k 2: Beogsi(a) = coe (log)? eS ron where a is a real number such that 0 1. Because of (2.8.5) and (2.8.8) we have S Ex(t) eS 2e* 2 welt 12) “E +l eh te? , k=O ie, SS Ext) ake S Ey zt Ese (t- ray" ae y 7 Et Oe k=0 ia =o Ft z) wherefrom k kev K\ Ey 1 Ex(t) = x oo ae (: - 3) , Setting t = 1/2 we obtain the connection between the Euler num- bers and polynomials, Ey = 2°E,(1/2).SOME CLASSES OF POLYNOMIALS 51 The Euler polynomials satisfy the following relations: E,(t+1) + Ex(t) = 2t", E,(1—t) = (-1)* Fx (0), y(t) = KEx-1(2)- Also, we have that [ By(t)at = Bere ot if Eu()E (tat = (P44? = 1) pO Bs Remark 2.8.2. Haruki and Rassias [1] also found the integral repre- sentations for the Euler polynomials. Remark 2.8.3. The numbers By and £;, are appeared in the following series expansions _1_z 528 (CIT PEB og ant cotr= > - $F ae --- (kI<), 2 6126 -1)*E; wernt FH Cte? pata elise Remark 2.8.4. The following summation formulas — Brta(n +1) - Bepr a k+l . Sayer = CPOs 1+ alo vel hold for every k,n € N. Remark 2.8.5. Gautschi and Milovanovié [1] considered the construc- tion of Gaussian quadrature formulas on (0, +00) with respect to the weight functions which appear in (2.8.6), (2.8.4) and (2.8.5), and showed that such formulas can be applied to sum slowly convergent series whose general term is expressible in terms of the derivative of a Laplace transform, or in terms of the Laplace transform itself.52 GENERAL CONCEPTS OF POLYNOMIALS For some other classes of polynomials defined by generating func- tions see Rainville [1] and Sheffer [1]. Rassias and Srivastava [2] de- veloped some general families of generating functions for the Laguerre polynomials (see also Rassias and Srivastava [1]). 1.3. POLYNOMIALS IN SEVERAL VARIABLES 1.3.1. Symmetric polynomials A polynomia] P in the independent variables z1,72,.-.,2, is called @ symmetric polynomial or a symmetric function of these vari- ables if it is unaltered by changing the order of the variables in any manner whatsoever. For this it is enough that the interchange of any two variables leaves the polynomial unaltered. For example, P, = Pi(a,b,c) = a +B 4c? — 3abe, P, = P,(a,b,c) = a? +b? 4c? — ab— be — ca (3.1.1) are symmetric polynomials, or symmetric functions, of the variables a,b,c. If one takes the sum of all distinct terms obtained from a typical term of the form ka ky Kon ay ry? Sats where ky ,k2,-.. , km are positive integers, and replacing the indices 1, 2, ..., m by all possible arrangements of m (< n) numbers taken out of 1, 2,..-, n, a symmetric function of the variables 21,22,... ,2n is obtained and it is denoted by Dare: or in short as [ky,k2,-.- -Kml- In general, if t is a product of powers of 2),22,...,2,, whose exponents are positive integers, then )>t denotes the sum of this term ¢ and all of the distinct terms obtained from it by permutations of the variables. For example, if we have only three variables a,b,c, then we write Yate = [2,21] = abe + ea + c2a"b.POLYNOMIALS IN SEVERAL VARIABLES 53 Such symmetric polynomials are called the L-polynomials. In n variables 21, 22,... ,Z, we mention two classes of E-polynom- ials: 1° Central symmetric functions (polynomials) se= ata lMaattett---+2n (k 20); 2° Elementary (basic) symmetric functions oe= oaze--te= [11.51 (1S <2). times For k = 0 we have so = n and op = 1. Also, we put o, = 0 for k>n. In the L-notation, polynomials in (3.1.1) can be expressed in the following simple forms P,= oe - 3) abe = [3] - 31,1, 1], Pp= Soa? ~ Sab = [2]- [1,1]. Using the functions s,, and o,, we have Py = 83-303, = Py = 52-02. The elementary symmetric functions of z1,22,...,Zn were used in Subsection 1.1.2 to derive the formulas of F. Viéte for the roots of an algebraic equation of degree n in one variable. Namely, we had Pe) =[]te-2) i=l (3.1.2) = 2" —042"" 4 +--+ (-1)"*on-12 + (-1)"on- Consider now the polynomial z + Q;(z), defined by Plz) =F z @s)= [Ite- #4) = agisn) = vat vet54 GENERAL CONCEPTS OF POLYNOMIALS and the elementary symmetric functions of 21, ..-, 24-1, Zits -+5 Z,. Putting MP =f... Osksn-1), of = ke times: we have Qala) = 24 = ofa? $e (10. This expression and (3.1.2) give (3.1.3) =o 42400, . Using (3.1.3) we find of) =1, oY =0,-2;, of) =0,-2501 +25. In general, we have (3.1.4) of) = =Ye 1)’r¥o,-» (kK =0,1,...,2-1). A connection between the functions s, and oy is given by the following Newton theorem: TueoREM 3.1.1. For every k (1 n, putting, of course, o, = 0 for v > n. For k = 1,2,..., (3.1.5) gives the following triangular system $1 = lo, 51-2 = 202, 028)— 9182 + 83 = 303, n~151 — Oh-282 + 04-383 — +--+ (-1)*' sy = kon,56 GENERAL CONCEPTS OF POLYNOMIALS wherefrom we obtain the following explicit formula 1 loy oO 1 209 4) % oO 1 303 Sk = (-1)* Tk-2 Th-3 ThA 1 (k-1)ona Ok-1 Tk? Th-3 “1 kon For example, for k = 1,2,3,4, we have 2 S1=01, $2 = 07-202, 53 = 0} — 30,07 + 303, 84 = 08 — dota, + doyos + 203 — 404. Contrary, the following formula 1 st $1 2 $2 SQ Ss) 3 83 Sk-2 Sk-3 Sk—-4 k-1 Sea Sk-1 Sk-2 Sk-3 3 Sk holds. For example, for k = 1,2,3,4, we have 1 1 1=5, = 3st — 52), 03 = gst — 35152 + 283), 1 m= rig — 6s}52 + 85153 + 353 — 654). Now we formulate the Fundamental Theorem on Symmetric Func- tions: THEOREM 3.1.2. Every symmetric polynomial P of the variables 21, 22, --+, Z_ can be expressed as a polynomial p in the elementary symmetric functions 01, 02, ..+) On, 4-€., (31.6) PCL, 2250+ 42m) = (01, 925-+- 9)POLYNOMIALS IN SEVERAL VARIABLES 57 Furthermore, coefficients of this polynomial are built up by additions and subtractions of the coefficients of the symmetric function. In particular, if the latter are integers, the former will be integers also. Several proofs of this theorem have been given. The reader is referred to the book of Uspensky {1, pp. 264-266], where Cauchy proof is presented. Let 01,02,...,0n be variables. We define the weight of a mono- mial a CA ae to be a, +202 +---+na,,. Also, we define the weight of a polynomial P01, 02,.+. ,On) to be the maximum of the weights of the monomials occurring in p. If each of these monomials has the same weight, we say that p is an isobaric polynomial (see Kurepa [1, p. 698}). THEOREM 3.1.3. Let P(z1,22,...,n) be @ symmetric polyno- mial and let p(o1,02,...,0,) be determined by (3.1.6). The total degree of the polynomial p is equal to the degree of P in any vari- able zp. THEOREM 3.1.4. Let P(z1,22,...,Zn) be @ homogeneous sym- metric polynomial of degree d and let p(o1,02,-.. ,0n) be determined by (3.1.6). Then p is an isobaric polynomial of the weight d. The proofs of these theorems can be found in the book of Kurepa {1, p. 697-698}. Example 3.1.1. (a) Let P(z1, 22,23) = (21 + 22)(z2 + 23)(z3 + 21). This is a symmetric homogeneous polynomial of degree 3. Its degree in the variable x} is dy = 2. Also, dy = d3 = 2. The corresponding polynomial Pin the variables 01, 02,03 will be an isobaric polynomial of the weight 3 and of total degree 2. According to the previous theorems, it must be of the form (3.1.7) P(x, 22,23) = p(o1,02,03) = Aoio2 + Bos, where A and B are constants. Setting x = 0 and zz = xg = 1 we have oy = 2, 02 = 1,03 = 0, P=2. Then, from (3.1.7) we get A = 1. Similarly, for 2; 2 = 23 = 1 we find that 01 = 02 = 3, 03 = 1, P = 8, and therefore B = —1. Thus, (01, 02,03) = 0102 — 93.58 GENERAL CONCEPTS OF POLYNOMIALS (b) Let P(21,z2,...,2n) = SDztzoz3. This is a symmetric homoge- neous polynomial of degree 4. Since its degree in z1 is 2 we conclude that the corresponding polynomial p in 01, 02, ..., on has the following form (3.1.8) P(21,22,..-.2n) =p(01,02,--- .0n) = Acres + Bod + Cos, where A, B, C are constants. For n = 3 and 2; = 0, 72 = 23 = 1 we have 0} = 2, 02 = 1, 03 = 0, o4 =0, and P = 0. Then (3.1.8) gives B = 0. Similarly, for 2) = ~1, 2 = 23 = | we get 0) = 1, 02 = —1, 03 = -1, and P= ajzp23 + 232123 + 232122 becomes P= -1. Thus, (3.1.8) reduces to -1 = A-1-(-1), ie, A=1 Forn=V4jendieii=[es=is[=[eal=)! welobtain o1=4, 02=6, 03=4, o=1, P=12 From (3.1.8) we get C= —4. Thus, for every n we have Ytieozs = P(o1,02,... ,on) = 0103 — 404. (c) Let n > 4 and P(z1,22,... ,2n) = 0 2}z3z3. Then P(z1,22,--- ,2n) = p(01,02,--. ,¢n) = 0203 — 30104 + Sos. 1.3.2. Properties of Symmetric Polynomials in Two and Three Variables In this subsection we give a few properties of the symmetric poly- nomials of two and three variables. 1° If a polynomial P(z1,22,23) is symmetric with respect to the variables z1,z2,23 and the polynomial is divisible by 2; + 22 then it is also divisible by zz + 23 and also by 73 + 2). 2° The general form of a symmetric polynomial of second degree in three variables 21, 72,23 is given by Py = P2(21,22, 23) = A(z} + 23 +23) + B(2yz2 + 2923 + 2321) +C(z1 + 22 +23) + D,POLYNOMIALS IN SEVERAL VARIABLES 59 where A, B,C, D are constants. However, the general form of a symmetric polynomial of third degree in three variables x1, 22,23 is Ps = Ps(21,22,23) = A(z} + 23 + 23) + B(zize + 2}z9 + 2321 + 2325 + 2321 + 2322) + Cxy29r3 + D(x? + 23 + 23) + Ezz + 22x39 + 321) + F(t; +224 23)+G, where A, B,C, D, E, F,G are constants. In terms of o, functions these polynomials can be expressed in the forms P, = Ao? +(B-2A)o.+Co,+ D, P; = Ao} +(B — 3A)o,02 + (3A — 3B + Choy + Do} +(E-2D)or + For + G. It is clear that the genera] form for a symmetric polynomial of first degree in three variables is P, = Pi(21, 22,23) = A(z: + 22 + 23) + B= Aoi + B, where A and B are constants. 3° If a symmetric polynomial P(21,22) is divisible by 21 — 22, then it must also be divisible by (2, — z2)?. 1.3.3. Resultants and Discriminants of Polynomials Let P(z) = Srava"’ and Q(z) = She" v=0 v=0 be two polynomials of degrees n and m, respectively, where agby # 0. Let 21,22,...,2n be the roots of P(z) = 0. The two equations P(z) = 0 and Q(z) = 0 have a root in common if and only if Q(x1)Q(22)-+-Qlzn) = 0.60 GENERAL CONCEPTS OF POLYNOMIALS This symmetric function of the roots of the equation P(z) = 0 is of degree m in any one root and therefore it can be expressed as a polynomial of degree m in the elementary symmetric functions. This is the fundamental theorem on symmetric functions. We define (3.3.1) RP,Q) = af Q(21)Q(22)---Q(zn) to be the resultant (or eliminant) of P and Q. It is equal to a rational integral function of the coefficients of P and Q and it can be expressed as a determinant of the order n +m (cf. Lang [2, p. 135] and Kurosh [1, p. 348) a a)... Oy Q a «... Ay 5 k rows RPQO=|, 4. bo by we bm a n Tows. by by wee bn The blank spaces are supposed to be filled with zeros. Example 3.3.1. For two quadratic polynomials P(z)=aoz? +a1z4+az and Q(z) 0), the discriminant of P can be expressed by the following determinant of the order n $9 81 eee Sned 31 Sn (3.3.4) D=ar-?| * . Sn-1 Sn Sin-2 Indeed, if V, = Vn(21,22,.-+ Zn) is the Vandermonde’s determinant of the order 1, i.e., Arai) ‘xi. ete 1 2 2 ae Va=|. , TV (za tz, gel then we have pad TT (2-2) =v, ISjéign ie, -1 1 ee sy... ony zy In 1 2 a i , nt 1} |) = % a 1 zy ant wherefrom we conclude that (3.3.4) holds. In certain cases we can find an explicit formula for the discrimi- nant (see the next subsection for the case of the classical orthogonal polynomials). In the following examples we derive the explicit for- mulas for the case of quadratic and cubic polynomials as well as for ‘a general trinomial, Example 3.3.2. Let n = 2, P(z) = az* + br +c, and x1 and x2 be the zeros of P. Since sp = 2 and b P-2 srsntm=-7, se=atte}= (a1 +2)’ - 222 =—S,POLYNOMIALS IN SEVERAL VARIABLES 63 formula (3.3.4) gives the discriminat of the quadratic polynomial, D =a" (ss; — st) =b° — 4ac. Example 3.3.3. Let n = 3, P(z) = apz* + a2” + agz +43, and z1, z2, and z3 be the zeros of P. We will calculate the discriminant of this polynomial using an expansion of the homogeneous symmetric function F(21,22,23) = (21 ~ 22)" (1 - 23)*(#2 - 23) in terms of the elementary symmetric functions 0), 02, 03. Since F is of degree 4 in any zero and also is a homogeneous polynomial of degree of homogeneity d = 6, this expansion must be (3.3.5) F = Ao} + Bo} + Co10203 + Dojos + Eoto3, where A,B,C,D,E are constants, whose values we have to determine. Notice that each term in (3.3.5) is of the form co??09?023, where c is a constant and (a1, 02,a3) € Nj such that 0 0, n> 2, n-2 n-k-1 a= TT Goa, r= TT (9) I<} 1 for all posi- tive t. This shows that (3.4.4) is an improvement of the classical AG inequality, provided the numbers z},...,Zp are different. Introducing the abbreviations 1 $= (tite ttn) PpHty---Zy, yy ttn R(t) = Pa(t)Qn(t)" = [] a poet I RR(t 4+ k— 1A} and eliminating p from (3.4.3) and (3.4.4), we obtain sr) > Ry(uA. Since k-1 Fog Ralt) = een 44) ee ee ~ Li (t¢n-Mt¢k— 1)? for all positive t, it follows that R,(u) > R,(0) and (3.4.5) sll) > RA(O)A. The latter inequality was established by Schur [1]. Obviously, (3.4.4) is a refinement of (3.4.5). There are several extremal problems which are closely connected with the discriminants of the classical orthogonal polynomials. Such problems were treated by Stieltjes [1-3], Schur {1], Fejér (2], Egervary and Turdn [1-2], Kiefer and Wolfowitz [1], Schoenberg [1], and others (cf. Szegé [2, pp. 140-142] and Karlin and Studden [1, Ch. 10]).70 GENERAL CONCEPTS OF POLYNOMIALS 1.3.5. Homogeneous Polynomials In Subsection 1.1.1 we defined homogeneous polynomials in several variables. Here we give some properties of such polynomials: 1° If P(z,,22,-.. ,Zn) is a homogeneous polynomial of degree of homogeneity d then P(tzy,tz2,... ,tZ,) = t*P(21,22,-.- Zn)» 2° Every polynomial P(z;,22,... ,Z») can be expressed uniquely as a sum of homogeneous polynomials P(21,22,---,2n) With re- spective mutually distinct degrees of homogeneity. The polynomials P,(21,22,-+- ,2n) are called homogeneous components of the poly- nomial P(z1,22,--- »2n)- 3° If the product of two polynomials P(21,22)---y2n) and Q(21,22)--- ,2n) is a homogeneous polynomial, then both polynomials are homoge- neous. Remark 3.5.1. Polynomials which are simultaneously homogeneous and symmetric are very important for both pure and applied mathematics. The general form of such polynomials in three variables of degrees at most first, second and third is given by P(21, 22,23) = A(z + 22 +23), P(z1, 22,23) = A(z} +2} +23) + B(ziz2 + 2223 + 2321), P(z1,z2,23) = A(z} +23 +23) + Blztze + 2123 +2325 +2223 + 2521 + 2324) + C21 + 22 +23), respectively, where A, B,C are arbitrary constants. We mention a result (cf. Hardy, Littlewood, and Pélya [1, pp. 104-105}, Beckenbach and Bellman (1, p. 11]), which is an immediate consequence of Rolle’s theorem: Lemma 3.5.1. If the polynomial P(z,y) = coz" +z" 1 y +++ tony”POLYNOMIALS IN SEVERAL VARIABLES n has all its zeros x/y real, then the same is true of all polynomials obtained from it by partial differentiation with respect to z and y. Consider now the polynomial 2 P(z,y) = [](z + 2:9), fs where 2; are real. Putting —(x/y) instead of z in (3.1.2) and px = o/ (7), we find that a n 3 kk a ke, P(z,y) = done" y= > (;) me" yf. k=0 By repeated differentiation with respect to z and y we obtain a P(z,y) XS fn Kt (n- ax'oy™ mz (oa my (n-k If we take m =r—landl=n—r—-1(1 ( ona) )"s ( vale ya on-r(@ +) On-r(2) on-r(y)POLYNOMIALS IN SEVERAL VARIABLES 73 with equality if and only if x ~ y or k=1,r=1. The proof of this theorem can be found in Mitrinovi¢ [3]. In a special case when r = k, inequality (3.6.2) reduces to one ty) > or(2)/* + oy) (kn), with equality only when z ~ y or k = 1. If we introduce the k-th symmetric mean of the sequence z by o; (363) p= pala)= on (0 0 and (3.6.5) Ph Phas > 0 hold, with equality in (3.6.5) if and only if 2) = 22 =---=2n. Proor. Inequality (3.6.5) was proved in Subsection 1.3.5 (see in- equality (3.5.1)). Inequality (3.6.4) is similar but weaker than (3.6.5). Indeed, using (3.6.3), inequality (3.6.5) reduces to 1 1 (i + a (t + OK => OK> Oy (1S KS n-1). Defining Amz = Y-w(5)sn Mitrinovié [1] has obtained some interesting extensions of (3.6.4) and (3.6.7):14 GENERAL CONCEPTS OF POLYNOMIALS THEOREM 3.6.4. [f1 0 for 1 < p< y, then (-1)’A%ox-y > 0. Applying (3.6.4) to the polynomial z ++ (x — 1)” Il (z +24), we 1 obtain (3.6.8). This idea was used to obtain further results (see Mitri- novié (1-2]). For example, using the polynomial 2+ (2-a)] [le +2;) il we find that A(on—10441 — OF)(Ob-20% — Tht) > (Oh-10k — h-27K41)- An improvement of (3.6.6) was given by Dougall [1]. Namely, if k 0. To end this subsection, we mention that there are many results for central symmetric functions s,(z) of the variables 21, 72, ..., Zn- 1.3.7. Properties of Zeros David Hilbert [2], in his lecture delivered before the international congress of mathematicians at Paris in 1900, posed 23 problems; the 16th problem concerns the topology of algebraic curves and surfaces. In this subsection we present localization properties of zeros of poly- nomials in two (or three) variables following results of Rassias [1-4]. Proofs of theorems will be omitted. We start with the following geometric property: PROPOSITION 3.7.1. The number of regions into which m straight lines lh, la, ts, ---, lm can divide the plane, R?, is at most 1 + [m(m + 1)/2]. Remark 3.7.1. One can always achieve the number 1 +(m(m + 1)/2].REFERENCES iy THEOREM 3.7.2. Let p(z,y) be a polynomial of degree m. Then the set Zp = {(z,y) € R? | p(z,y) = 0} divides the plane R? into at most 1 + [m(m + 1)/2] regions. In the following we mention the analogous problem for a homoge- neous irreducible polynomial of degree k in three real variables, which defines a non-singular algebraic curve in RP? ~ $?/{+1} (cf. Lang [1] or Seidenberg [1] for an introduction to Algebraic Geometry). Rassias [3-4] proved the following result: THEOREM 3.7.3. Let p be a homogeneous irreducible polynomial of degree k in three real variables z1, 22, 23 which defines a non- singular algebraic curve in X = RP*. Let Z={2eERP*|p(z)=0}, 2 =([21, 22, 23] Define x:S? + X the universal covering map and Z = x~1(Z) C S?. Let n be the number of connected components of Y = S?— Z. Then (a) If k is even, n< k -3k45; (b) Ifk is odd, n "ogken ** where the best constants A, and B,, are given by Sear, me()' Gl: 8s86 SELECTED POLYNOMIAL INEQUALITIES If z+ az? +z +c is a polynomial with non-zero complex coeffi- cients, then its zeros lie in the closed disc (a) rs| bya fé| . Indeed, since [ f,\4 | VD [=| Ve = aac = 1M ley 4+ (], | vB I< i (1+ ||) = wie | |, we obtain 1 wld Pk zllal which implies (1.1.1). Let p,q, be real numbers. If all three roots of P(z) = 2° + pz? + qz+r = Oare real, then they are all less or equal to (2,/p? — 3q—p)/3. This result is due to Rao [1]. The following solution was given by Losey [1]. Let Q(x) = P(z—p/3). Then Q(z) = 2945244, where s = qt+p?/3 and 1 = 7 — p®/27 and s and ¢ are real if p,q and r are real. Note that the zeros of P(z) are real and < (2\/p? — 3q — p)/3 exactly if the zeros of Q(z) are real and < 2,/—s/3. Let s and ¢ be real and let k be a real zero of Q(z). Then k° + sk+t=0 and so Q(z) = 29 + sc —B - sk = (2- h(x? 4 he +P +5). The other two zeros of Q(z) are real exactly if k? — 4(k? + s) > 0, i.e., exactly if k? < —48/3. Tf Q(z) has real zeros z1 > z2 > 23, then the above argument with k = 2 gives the required result. If only one zero of Q(z) is real, say Q(k) = 0, k real, then k? > —48/3. Thus, in the given problem oneINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 87 can say that either all the roots are real and < (2\/p? - 3q— p)/3 or only one, k, is real and & > (2\/—p? — 39 — p)?/9. A necessary and sufficient condition that otpe+qr+1>0 for 2>0 is given by Verblunsky [1] in the form Pat 6(p +9) +9+Ap+q+3)¥? >0, where p,q are real numbers and p+q+3> 0. For real polynomials of the fourth degree Maiik [2] proved: THEOREM 1.1.1. Let P(z) = apz* + 402° + 60277 + 4a3z t a4 be a real polynomial with a9 > 0, |a3| + |aa| > 0. Put E = aga + 303 — 4aj03, $= aga, + 2a,03, F = aga} + ajay + a3 ~ agagay — 20,0203, D= E°-21F*, R= S? —9aga3a4. Then P is non-negative if and only if D > 0 and either R-< 0 and a2 > 0, or R> 0 and —aga4 < 2a;43 < 2apa,. Verblunsky [2] was also concerned with necessary and sufficient conditions which must be satisfied by the coefficients of a real quartic polynomial x ++ P(x) = agz* + a,2° + azz? + a3z + ay in order that P(x) > 0 if x > 0. Since it must have ap > 0 and ay > 0, putting z = et where c is the positive number which satisfies ape = ay, it is enough to consider the quadratic polynomial t + Q(ct)/a4 for t > 0. ‘Therefore, Verblunsky [2] considered a quartic polynomial of the form (1.1.2) Q(2) = 24 +ac* + bx? te 41. TueoreM 1.1.2. Let p= 2- (b- V12+ 8 — 3ac)/3, L= L(b) =(p—-4) Vb + 2p- 2-(a+c), 32(b-2), if b>6, m= { Seaaye if b<6.88 SELECTED POLYNOMIAL INEQUALITIES The polynomial (1.1.2) satisfies the condition P(x) > 0 for x > 0 in one of the following mutually eaclusive cases and only in such cases: 1° a>0,c>0,b+2+4 2fac > 0; 2 a>0,c>0,b4+24+2fac <0, L <0; 3° a <0,c<0, L <0, M(b) > Bac, b+2>0; 4° ac<0, Lb <0. Let 21,22,23 be zeros of the real polynomial of the third degree ze P(z)= 29 + prt and let 2 d= min{|z1 — 22), |22~ z3h)|z3—- 21/}, D= a + Le Wrona [1] proved the following statements: 1° If D=0, then d = 0; 2 If D <0, then 2V-3D/\|p| < d < 3V-3D/|p| < Viel; 3° If D> O and p > 0, then Visas yp Ves; 4° If D> 0 and p< 0, then V3—q]2- Vp 5 with equality if € = +1/3. Thus, we conclude that P(e) > Per 2 3° On the other hand, for all quadratic polynomials P we have 1 P(1/3) = i P(x) de — FP(-0), whence 2 max P(1/3)< 3 which completes the proof. Remark 1.1.1. Schoenberg's problem can be solved using the following representation of non-negative polynomials on [-1, 1], P(z) = (az+b)'+(1-27)c? (a,b,c ER). Let (1.3) P)=T](1-2), i=l J where the points w; lie on the unit circle C. Cohen [1] has proved that on some path [° which joins the origin to C, the inequality |P(z)| <1 holds everywhere except at z = 0. C. Loewner sometime later established the existence of a polynomial (1.1.3) for which every radius of the unit disc passes through a point at which |P(z)| > 1.90 SELECTED POLYNOMIAL INEQUALITIES Erdés, Herzog, and Piranian [1] described a very simple example of a polynomial (1.1.3) with the property that on each radius of the unit disc there exist two points z’ and z” such that [P(z’)| < ] and |P(z")| > 1. Erdés, Herzog, and Piranian [1] discussed the following question: Does there ezist a universal constant L such that for every polynomial (1.1.3) the inequality |P(z)| < 1 holds on a path which connects the origin to C and has length at most L? This question has been answered in the negative by MacLane [1]. Erdés, Herzog, and Piranian [1] dealt with the polynomials (1.1.3) in the case n < 4 and obtained the following result. n Tueorem 1.1.3. Let P(z) = [] (z—2), with |z,] =1. Ifn<4, there ezist two values 6 and 0" such that (114) [P(re™@)<=r"] and [P(re") > 140" for0 @ > B > 7 >O0anda+f+4+¥7= 22. To prove the existence of 6’, we set 2) = 1, 22 = e, and zs = e~"7, and we prove that P< ll-] for O 0 Al We have (1.1.5) 0< (B~7)/2< H/2 1—cos[( — 7)/2] 20. It follows that A’ > 1°(1— 1)? [9 - 16sin?(G/2) sin? (7/2)] - Using (1.1.5), we obtain 0 < 4sin(4/2)sin(7/2) = 2cos [(6 — 7)/2] — 2cos (6 + )/2] < 3, and therefore the value 6 = 0 has the required property. To show the existence of 6" in the case n = 3, let a, 6, 7 be the same as above. Set 2 = e@/?, 2p = e*9/?, and 23 = ef(2/2+0)_ We will show that [P(r] 21478 for O xr, then [P(r)} > (14+ 7? 3? > 1473. In the rest of the proof we restrict a to the interval 2t/3 (1 = 2rt + 2) [1 — 2r (40 — 3) 4]? It is sufficient to prove that the quantity AY = (1—2rt+ 0°) [1 - 2r(40 — 3t) +7] -(14 8) is non-negative for 0 0, Ag = (1 — 407)(3 — 81?) > 0, As = -2+4t+ 88° — 3285. Then A" > (242 + As) 2 0. In the case n = 4, let a, f, 7, 6 denote, in cyclical order, the four non-negative angles formed by the radii 0z,, with a+f+7+6 = 2n. In order to establish the existence of 6’, we assume that the notation has been chosen in such a way that 746 <1. Set 2 = 1, 22 =e", 24 = eil5t0) = @ilO4), 2 = e-i7. We will show that IPOs ll-r'] for Onr>7+6, atb>r>Bt+y, Pr2é. Set z = e7#9/?, zy = el@?, 23 = efl@/2+9), 2, = e-il2/248). From (1.1.6) we get (LL7) w-a/2 1/2, we will prove that |P(r)| > 1+r4 for 0 < r < 00. Fora > 7, the result follows. For 7/2 [r— et = Ir tal, and therefore [P(r)| > |r — Pr + zal? = Je? — FP = 1-27? cosatrt > 14%. If a < 1/2, we shall prove that [P(ir)| > 1444 for 0 ir - e+) = fir + zal, and therefore (ir — z5)(ér — 24)] > |r? + 22] > (14 2r? cose +r)? . Similarly, [Cir — 4 )(r — z2)| = (14 2r? cosa + 47? , so that. |P(ir)| > 142r? cosa tet > 14r4. The proof of the theorem is now completed. O Remark 1.1.2. In the case n = 4 the inequality [P(z)| > 1 does not necessarily hold everywhere on the bisector of the greatest of the four angles involved. For this, let z1 = e'/3, z2 = -1, 23 = v4 = e i783 Then |P(1/2)| = (9/16)V3 < 1. Even if we allow continuity we see that if a > B > 7 > 6 > 0, the inequality |P(z)| > 1 need not be satisfied everywhere on the bisector of a. The problem of determining the greatest degree n for which a polynomial (1.1.3) always satisfies the inequalities (1.1.4) on two appropriate radii of the unit disc or on two half-lines from the origin seems to be an interesting problem.94 SELECTED POLYNOMIAL INEQUALITIES 2.1.2. Some Important Trinomials We start this subsection with a remarkable result (cf. Beckenbach and Bellman [1, p. 12], Hardy, Littlewood, and Pélya [1, p. 61], Mitri- novié [1, p. 226]): For all z € R and any even number n € N, (1.2.1) Q(z) =2"-nz$n-120, with equality if and only if z = 1. Evidently, if 2 <0 we see that Q(z) > 0. According to Descartes’ tule of signs, the polynomial Q cannot have more than two positive zeros. Since Q(1) = Q‘(1) = 0, ie., Q has a double zero at x = 1, we conclude that z = 1 is the unique real zero of Q. This implies inequality (1.2.1). This inequality has been used in a number of cases as a starting point in the process of finding other inequalities (cf. Mitrinovié [1, p. 126-129]). In this direction, Dilcher, Nulton, and Stolarsky [1] mention the trinomial (1.2.2) zHT(z)=mz"-nz™+n—-m, n>m>0. and the corresponding inequality T(z) > 0 for 0 < z <1. For example, replacing z by (z/y)'/", with y > z, and m/n by 4, we obtain the weighted arithmetic-geometric (AG) mean inequality (1.2.3) Ar+(1-Ajy> 22y!. The zero distribution of z + T(z) recently has been investigated by Dilcher, Nulton, and Stolarsky [1]. In the sequel we present their results. Let C = {z € C | J2| = 1} and let ged(m,n) be the greatest common divisor of the integers m and n. THEOREM 1.2.1. Leta > b > 0 be real numbers and n > m>0 be integers. Then the number of zeros of ze Plz) = b2"—a2™4a—b strictly inside C is m—ged(m,n) ifa/b > n/m, and m if a/b < n/m. As a consequence of this result we have:INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 95 CoROLLARY 1.2.2. Ifn > m > 0 are two relatively prime integers, then the trinomial z+ T(z) in (1.2.2) has m—1 zeros strictly inside C,n—-—m-—1 zeros strictly outside C, and a double zero at z = 1. For n > 3 the zeros of z+ T(z)/(z—1)* lie in the following annuli: (a) Form=1 T+ (n= 2) < [2] < [an VO; (b) For 2 Pn(z) has a zero inside the unit circle if and only if d is not the reciprocal of a positive integer. Regarding univalent polynomials in the unit disk, Brannan {1-2} considered trinomials of the form z+» P3(z) = z+ aj2” + tz3, where t is real and positive. In fact, he proved: THEOREM 1.2.5. For 0 < t < 1/5, the trinomial z + P3(z) is univalent in |z| < 1 if and only if az lies in the ellipse t= {erivec| (a) +(e :)" <3}. whereas, for 1/5 < t < 1/3, z+ Py(z) is univalent in |z| <1 if and only if az lies in the intersection [\(1 on tga¢3 Eo. Of the family of ellipses i toca {+i eC | (Jt (4) < am} . In order to prove this result, Brannan used the following general criterion of Dieudonné [1]: THEOREM 1.2.6. The polynomial z+ P,(z) = z +0227 +---+ a,2" is univalent in |2| <1 if and only if for every 8 in [0,x/2} the associated polynomial |, sin26 sion nt (z,8) = 1+ and Q2zZ+°- = fenlet me” does not vanish in |z| < 1. For @ = 0, $(z,@) can be interpreted as Pa(z). Theorem 1.2.5 was also obtained by Cowling and Royster [1] fol- lowing a completely different method. Ruscheweyh and Wirths [I] and Rahman and Szynal [1] considered a more general trinomial z ++ P(z) = z + Qm2™ + Bmz?™-1, where m € N\ {1}, Gms8m € C. Rahman and Szynal [1] proved:INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS. 97 ‘THEOREM 1.2.7. Suppose P(z) = z+am2z™+tz?"-1, where t is real and positive, and a, € C. If _ 14 tS2m—2(u) _ 1=#S2m-2(u) Am(u) = aes 0h, and B,,(u) = Sacmer(a)y where u ++ S,(u) is the Chebyshev polynomial of the second kind of degree k, then z+ P(z) is univalent in |z| < 1 if and only if an, lies in the intersection Dmt =), Emus of the ellipses ‘m= 2 y ae ed ee as}. O (Zn-1 + 2n)/2- For real polynomials of the form - nk (1.3.2) P(z)= Lega (n> 3; aoa, #0), tO Maik {1] proved the following result:INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 99 THEOREM 1.3.3. If the polynomial (1.3.2) has only real zeros, then af > apa (k= 1,....9-1) and 4 (ah 41 — an042) (ab42 — on¢10K+3) 2 (Oxt19K42 — OkGKES)"s fork =0,1,...,n-3. If the roots of the equation (13.3) age” - (iyae + (Sener =e (-1)"an = 0 are positive and distinct, then yg > 4,0, for all p,q,r,s such that p+q = r+ and |p—q| < |r—s]. In particular, 4Gn-p > 9d, (p=1,...,n—1). This was given by Olds and Starke (1]. Jolliffe [1] considered the equation (1.3.3) with ag = 1, and proved that a? > a,_1a,41 when no condition is imposed on the roots except that they are to be real. A simple inequality for a real polynomial zee P(t) = aot” +a,2"" +---+0n424+0n (apa, #0), with all positive zeros z1,...,2,, has been obtained by Fransén and Lohne [1]: Q1Gn-1 4g 0, Indeed, by the arithmetic-harmonic mean inequality, we have >? Ont Or 1 “oo an Set a 2 Constantinescu [1] proved:100 SELECTED POLYNOMIAL INEQUALITIES 1 THEOREM 1.3.4, Let z + P(x) = ¥) agz* and z + Q(z) = 0 F byz* be real polynomials all of whose roots are real and an > 0, ‘k=0 bm > 0. If these roots separate each other, then for m= n, ay-1b > andbe-y— (K=1,.-- 7), and form=n-1, Oy 1b, 0. Let z, (v = 1,...,m) be zeros of a polynomial P of degree n. Using the logarithmic derivative of the polynomial P, we find that ~ o ()? — an Oh) P'(z)? ~ P(z)P"(z) = P(zy L Grn If P is a polynomial with real zeros, then the following inequality of Laguerre {1, p. 171 f.] holds Pz) - P(z)P"(z)20 (ER). A detailed study of the zeros of this particular polynomial z > Q(z) = P"(z)? — P(z)P"(z) has been made by Dilcher [1] and Dilcher and Stolarsky [1-2] (see Chapter 3). Let z ++ P(z) be a real polynomial of degree n with real zeros x, sss, 2n and let 1, «+5 Yn—1 be zeros of the derivative P’ such that BySeS ay, and oy S++ Syne Putting (n- 1), = (21 +-+°4+2n) — 2kINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 101 and (n= 2)ne = (m1 +++ + n-1) — Ye Popoviciu [2] (see also Bray [1] and Toda [1]) proved the following inequalities n n-1 (13.4) 1 ae2 yew v=l wal 1 n-1 = nO) Sy ds where z ++ $(z) is a convex function. If the zeros of the polynomial P are non-negative and Aa hayteten G2 (2-29), then CK (GG) SA and 2 es Aas hte ia =o-n( 8 -z) - (Se-n)\ Igi 0 we obtain that (1.3.7) fee-al< 1G —DA, ic., all the zeros of the polynomial P lie in the interval I= [: . 1 Vea iA, 2+ 2G = 7] ; If y1,---) Yn—1 are zeros of the derivative P’, Lupag [2] stated that n n-1 (138) f(@)+ +s mly pene ped Mud 7 laX n 2n{n—1)? vl vat with equality for f(z) = z?. Thus, the first inequality in (1.3.9) is proved. Similarly, setting ¢(z) = —f(z) + }Mz?, we prove the second inequalities in (1.3.8) and (1.3.9). Taking x) > 22 > --- > tn, Lupag [3] gave sharper estimates than (1.3.7).104 SELECTED POLYNOMIAL INEQUALITIES Tueore 1.3.5. If P € Pa(ar,a2) then ifs R, defined by o(P)=2-, and m(P)=, min. (24 — 23), respectively, where P(z;) = 0, 21 > 22 > ++ > Zn. Sometimes, the number o(P) is called the span of the polynomial P. THEOREM 1.3.6. For every P € P,(a1, a2) we have (1.3.11) [ses 2. where p = [n/2], q = [(n + 1)/2] and [t] denotes the integral part of the real number t. The left inequality becomes an equality if and only 7 P resents (nen fl) (28) while the upper bound is attained if and only if P(2) = P*(2) = ((2-2)- aye ~ayr?,INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 105 The upper bound in (1.3.11) was found by Sz.-Nagy [1] and redis- covered by Popoviciu [1], who also conjectured the existence of the lower bound. We now give the proof of Lupas [3]. Proor OF THEOREM 1.3.6. It is easy to conclude that the poly- nomials P, and P* belong to the class P,(a1,42). In order to prove that (1.3.12) pene ae(h) =0(P*)= 3. we consider A in the form nol (1.3.13) A =n \0(2, - 2) + (zi - za)? + 2n(Z — 21)(2 — 20). vs? Since z z 1 2 1 \2 (2— m1 )(@— tn) > —G(@1 — ta)" = — 9 0( PY’, with equality for Z = (z; + 2,)/2, from (1.3.13) it follows that A> tory, ie., o(P) < /2A/n. If P € Pn(a1, a2) is of the form P(z) = (2 - 11)(2- 2n)(z- 2)", then the equality holds. Taking into account that P,(a1,a2) contains only one element of this type, namely P*, the proof of (1.3.12) is complete. In order to find the lower bound, we consider A as a function in n variables 21, ..., Zy- At first we have a. =2 -z). az nz, — Z). Supposing that 2, >--- > 2, > Z > 241 > ++ > Zp, One finds A< Kn— k(x — 2).106 SELECTED POLYNOMIAL INEQUALITIES On the other hand, k(n — k) < pg with equality if and only if k= p=[n/2] or k = q = [(n + 1)/2]. This enables us to assert that A < pgo(P)*. Finally, in the set P,(a1, a2) the elements of the form (z-2)"(z-2,)"*, k=p or k=9, are those denoted by P.. Therefore, i =o(P.)= [a reminves ed) let Ee) i=) Pq Qo For the functional m Lupas [3] proved (see also Lupag [1]) the following result: THEOREM 1.3.7. If P € Pn(a1, a2) then m(P) <2 with equality holding for Pla) = Pre) = J] (2-2 - ae). =i Robinson [1] has considered the problem of determining the max- imum span of the k-th derivative of polynomials of degree exactly n having 7 real zeros, when the span of the polynomial itself was given. Here, it is sufficient to consider monic polynomials with zeros in the interval [-1,1]. Then all of the zeros of P(*) lie in the same interval. The problem is trivial if n > 2k +4 2. For in this case, we can put k +1 zeros of P at each end point z = +1. Then P(*) will have a zero at each end point, and will therefore have span 2. If n > 2k +2, some of the zeros of P are arbitrary, whereas for n = 2k + 2 the span o( P(*)) is maximized only for P(z) = (2? — 1)*+!. Robinson [1] considered the non-trivial cases of this problem, i.e., those with k42 3, Namely, applying it to PU-)) (j = . »k) we have o( PY) > o( PO- ne a he Gah... 5k) whence n-k n-k-1y1/? ocr) oc({] Sa) OS Sa)" with equality holding only if o(Pt) = of P)(*= ay 7 In order to prove (1.3.15) for k = 1, without loss of generality, we can assume that s = 1 and that the zeros of P € P,,1 are located in [-1,1]. Then P(z)=e(z*-1) [] (2-2), -1< m1 <-++< a2 <1 Let yw <-+- < yn—a be zeros of P’. It is clear that the minimum of. o( P') cannot be attained if z} = —1 and z,_2 = 1. From the Rolle’s theorem it follows that either (i) both y, and y,—; are the extreme solutions of PO). wt Py) yl L5 (1.3.16) vel or (ii) one of them is fixed, say y,- = 1, and the other y; is the smallest root of (1.3.16).INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 109 Tt can be showed that the minimum of o(P’) cannot be attained in the case when 11 < t,_2. Namely, in that case differentiating y partially with respect to z, on the hyperplane 2; +Z,_2 = const, we have from (1.3.16), auf 1 AR+y)\ rd Bay (x Galt a 7) "aay Gael ou Since y, < 21 < Zn-2, we conclude that —— > 0. A similar argument br shows that in case (i) as well as in case (ii) one has that tet <0, so Go(P!) _ Aun _ OM Ox, 0m, Oy This shows that the minimum span can occur only if Bet. SRE. IE According to P(x) = (x? — 1)(x — a)"~?, we have My a? + n(n nv o(P') =2 The minimum of this function is attained fora = 0. O Supposing some zero distributions of polynomials, Meir and Sha- rma {1} proved the following results: THEOREM 1.3.9. Let P' be the subclass of all polynomials of de- gree exactly n, with real zeros z, (v = 1,...,n) satisfying the condi- tions n 20 V=l..m), Yoasa, and let k 0, and z isa real number, Let ||P|| and z be defined by ~I 1 nf? PIE 5. {o(o/m,4/2)} holds, where the equality holds if and only if r = 0. THEOREM 1.3.12. If Z > 03, where 0 < ap < 1, then 1 nf2 IPI) > 55 fie(00,0)}", and the equality holds if and only if ag = 1, i.e., either p = n or gen. THEOREM 1.3.13. Given p,q, |2| = 03 and let p an?112 SELECTED POLYNOMIAL INEQUALITIES and the equality holds if and only if ag = 1—2p/n. In the case q

0) shows that F3; = 2((a — b)/n)°. Thus, F; may be zero, positive, or negative. Consider now a general expression Day = A (1) where 2!) (v = 1,...,n—k) are zeros of the k-th derivative P(*) and p< n, Evidently, Dy..2 = Fm. The following result was proved by Aihara [1]; THEOREM 1.3.14. Dm,p = 0 for m= 1,2,... ,p. At the end of this subsection, we mention a result, of Erdés [2]: THEOREM 1.3.15. If P(x) = cg +12 +---+en2", Coch # 0, has integral coefficients only and has all its zeros in (-1,1), then len! > 277. Recently Reddy [1] obtained the following generalization:INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 113 THEOREM 1.3.16. If P(z) = en(z—21)(t-Z2) ...(2-Z) (n > 1) has all its zeros in [—1, 1] and satisfies, for k = 0,1,...,m(m > 1), [P(n)| 24 >0, me = cos(ke/m), then len| > "Va. Equality holds for P(z) = +aTm(z)"/™ (a > 0) whenever m divides n, where Tm is the m-th degree Chebyshev polynomial of the first kind, defined by T,,(2) = cos(marccos z) {(-1 < z < 1). Notice that for a = 1 and m = 2 this result reduces to Theorem 1.3.15. 2.1.4. Inequalities Involving Zeros of Polynomials f2a=& Kt in, (v = 1,...,n) are the zeros of the polynomial ze P(z)= B a,z”, Ostrowski [3] proved the following inequalities for all real values of z: y (2) >Re P(z)PM(z)— P(x? Qe P(zy , 1)? ae Pilia)? = Plix)PMiz) »D (z) 2 Re P(iz)?? In addition, if m # 0 or & # 0 (v =1,... ,n), then we have n 2 2 " (2) > BRe Piz) fe (z) Sn (2) and “712 P(iz)P"(iz) — P'(ix)? y (G) 2 8R PG , Tespectively. These a are equivalent, respectively to 71 =2aa, sl 2agay — —) >8Re ' =) >8Re———. D(z) 2am, YE) a De Bruijn [1] proved for some special lasses of polynomials the following result:na SELECTED POLYNOMIAL INEQUALITIES THEOREM 1.4.1. Let the polynomial z > P(z) of degree n > 1, have the zeros 21,... ,Zn, and let wy,...,Wn—1 be those of z+ Pz). Then we have ou Ie (1.4.1) A Yiimu,| < = Yo Im... v=1 vl The equality holds if and only if no two zeros of z+ P(z) are sepa- rated by the real azis. In the following we will prove the generalized form of the above theorem (due to de Bruijn and Springer [1]), namely for polynomials with arbitrary real or complex coefficients. For the proof de Bruijn and Springer [1] introduced an auxiliary function z+ P*(z) obtained from P(z) by replacing the zeros of z+ P(z) in the lower half-plane by their complex conjugates. Theorem 1.4.1 follows very easily when all zeros of z + P(z) lie in Imz > 0. Because in that case, by the Gauss-Lucas theorem (see Theorem 1.2.1 in Chapter 3), the same is true for the zeros of z+ P'(z), so that the imaginary parts of 2,... Zn, Wiy-++)Wn-1 all have the same sign. The theorem then follows from the relation Therefore in (1.4.1) the sign of equality holds in this case. The general case is reduced to this one after applying the following lemma of de Bruijn and Springer [1]. Lema 1.4.2. Let ik n P(z)=e]] (2-2) J @-2) ( 0 (v =1,...,k), Imz, <0 (v=k+1,...,n). Ifwe sel k n (142) P*()=a][ (2-2) [] @-2) val vektlINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 115 then IP(@) < IP"@) for all real values of z. The equality holds for all real x if and only if no two zeros of z++ P(z) are separated by the real azis. Proor. Setting n 1 1 y =QtiR, DY 5 2, vokt) ud =S4+iT where z, Q, R, 5, T are real, we obtain P'(2) | (2) | Play (2+ S)+HRET) and Pez) (2 FS) +HR-7). It follows from Imz, 20 (v=1,....k) mz, <0 (v=k+1,...,0) that R>0,T <0. Thus |R+7|<|R—-T|, which implies Fe Piz) P(z) If z is real then |P(z)| = |P*(2)|, and therefore IP(2) < lP"@)I- The equality holds for all rea] z only if either R = 0 or T = 0, in other words if all zeros of z+ P(z) lie either inImz0. O Applying the above lemma, the proof of Theorem 1.4.1 follows. Proor of THEOREM 1.4.1. Let the polynomial z 4 P*(z) be defined by (1.4.2). Because of the fact all zeros of z+ P*(z) lie in Imz > 0, it follows that n~1 1 nor hee 1.4.3 Imw}] = — (1.43) Y leat alma,née SELECTED POLYNOMIAL INEQUALITIES where wj,..., w%_, denote the zeros of z++ P*’(z). Also, A A aa ff roglPteide sf teele elds (A> 0) -A -A or nol n-1 (1.4.5) ¥ [steele eS [lew We have A (1.4.6) il log |z — al dz = 2(Alog A — A) + x[Ima]+ O(1/A). -A From (1.4.5) and (1.4.6), and taking the limit as A ++ +00 we obtain n-1 n-1 (1.4.7) Yitmu,| < Ym uy]. tat oa From (1.4.3) and (1.4.7) we obtain aol LS |Imw, [<1 hima) ws ls vt vs1 which proves (1.4.1). The equality in (1.4.7) holds if and only if there is equality in (1.4.4), in other words if |P(z)| = |P*'(z)| for all real values of z. Then using Lemma 1.4.2 one has that all zeros of z ++ P(z) lie either in Imz > O or in Imz < 0. This completes the proof of Theorem 14.1. 0 Remark 1.4.1. In geometric terms Theorem 1.4.1 means that the zeros of z+ P"(z) lie, in the mean, closer to the real axis, than the zeros of z+ P(z). The same applies about any line when we replace [Im z] by |Im(az + A)|, where o and f are complex numbers. This can be proved if one applies Theorem 1.4.1 to the function z ~ P((z — B)/a). In an analogous manner de Bruijn and Springer [1] proved the following theorems.INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS M7 THEOREM 1.4.3. Let 2,...,2n be the zeros of the polynomial 2 P(2), wy... ,Un-1 the zeros of P'(z), and let Im 2 < Im z < oS Imzy. If z+ Y(z) is a conver real-valued function for x € [Im 2, Imz,], and if 12 1 oad DWP) = = vm 2.) - —> Yo W(Imw,), 1 val then D(y,P) > 0, with equality only if is a linear function for x € [Imz,,Imz,]. This implies the case Im 2, = --- = Im 2p- One obtains a special case when #(z) = |z|" (r > 1), for which the following theorem holds: THEOREM 1.4.4. Using the notation of Theorem 1.4.1, for every r>1, we have 1 12 . ao Dla $y dlimat" = vel The equality holds in the following two cases: a) if r = 1 and all zeros of P lie in the same half-plane Imz > 0 or Imz <0, and b) ifr > 1 and Imz =---=Imzy. THEOREM 1.4.5. Under the hypotheses of Theorem 1.4.1 we have, forr>1, 1 n-1 1 n (1.48) ales. Dial a m The equality holds in the following two cases: a) if r = 1 and all zeros of P lie in the same half-line with endpoint 0, and b) if 21 = --- = Zn. Proor. The distance of the point z € C to the line through the point 0 making an angle a with the positive rea] axis, is [cos alm z ~ sina Rez|. By Theorem 1.4.4 we have n-1 n i, 1 > \cosalm z—sinaRez|" < = > |cosa Im z— sina Rez|", v=l v1 n-1us SELECTED POLYNOMIAL INEQUALITIES wherefrom, by integrating we obtain 12 an al | cosa Im z — sin a Re z|"da nah ilo PS pr sip | cosa Im z — sin a Re 2|"da, ny=i/0 or (1.4.8). O Some similar results can be found in the papers of Bray [1], Erdés and Niven [1], and de Bruijn and Springer (2]. Let a, b, and ¢ be three points in the complex plane, and let A(a, b,c) denote the area of the triangle determined by those points. Kramer [1] proved the following result: THEOREM 1.4.6. Let z ++ P(z) be a polynomial of degree n with Z€TOS 21, -.-, Zn, and let wy, ..., Wa be the zeros of the derivative P!. Then we have ar > Alstps Wyte) $ > Al Zi, 25s 2k). Agp 2, : [AP oa Papo [Poel Pe (14 oles) ek 2 vel where equality occurs if and only if \ = k = 2 or, forX =2 ifk-2 of the left-hand terms are equal to 1. In the book of Ostrowski [1, p. 32] the following result is mentioned as a problem: THEOREM 1.4.8. If 2),... ,2n are zeros of the complex polynomial R z+ P(z) = } a,2” (a, € C; a, #0) then v0 (1.4.9) Shall” < £ len] Te + 2-4). v=0 We give the proof of Kalajdaié [1]. Proor or THEOREM 1.4.8. Let Qe) = Soh = Jol [Te +a) (bn = lanl) vat and let o,(z1,-..,2n) be the elementary symmetric function of 71, s+ ®n (cf. Subsection 1.3.1). Since —|z,| (v = 1,... ,n) are zeros of the polynomial Q, according to Viete’s formulae we conclude that (-1ytev(lah- be) = (D8,120 SELECTED POLYNOMIAL INEQUALITIES baw = lanlor(I2il,--- »Lznl) 2 lenllo.(215--- zn) On-y Gy, = {onl = lan] (v=... yn). Therefore, for v = 0,1,... ,n we have 6, — |a,| > 0, which implies the inequality " YL. - leet’ 20. v=0 This inequality is equivalent to (1.4.9), foreveryz€C. O At the end of this subsection we mention a result of Colucci [1]: Tueorem 1.4.9. If the moduli of each of the zeros of a complez polynomial z ++ P(z) = bo az” (a, € C; a, # 0) do not exceed a positive number M, then’ [PM(2)| < u(t 1) lel +m)", for every k = 0,1,...,n. 2.1.5. Inequalities for Non-negative Polynomials For polynomials with non-negative coefficients, Milovanovié [2] proved the following result: THEOREM 1.5.1. Let P(z) = S az” with a, >0(0 0 the inequality, (15.1) 2(P(2) ~ Ple)P"(2)) < P@)P(2) holds. Proor. Using the Cauchy-Schwarz-Buniakowsky inequality | Yew Ps (o I.) (5 Iwl?),INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 121 for z, = af/*z4/? and y, = val/*z"/? (2 > 0) we obtain (Deen) < (La)(Sa-’) which is equivalent to (1.5.1). O For the Lorentz’s class of non-negative polynomials on [—1, I], i-e., for polynomials of the form (1.5.2) Q(t)= ya —yr(en’, by 20 (OS 0) we obtain Pz) P"(z) < 1 Piz) Pz) ~ 2° Of course, for z > 0 the following inequalities PHY z) _ PIED ( 2) PUD (2) PUN(z) (15.4) st usesn) also hold. Since P("+))(z) = 0, using (1.5.4) we find that (z>0), which was proved by Soble [1]. The case P(x) = 2” shows that this is the best possible result. If, moreover, the constant term in P(z) is zero, then Piz) P(z) The case P(x) = z shows that this is the best possible result. Soble [1] also proved the following three theorems: 1 <> (e>0). ‘Tneonem 1.5.3. Let P(z) = Ss eyz"-", with c, >0. If01(v=1,...,n), then i P\(z) P(z) (2 -¢/2? a(z—cy ls n (z >).INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS — 123 n THeoreM 1.5.5. Let P(z) = } cya"-” (c, > 0). Ife, > eva =0 (v=1,...,n), then Pz P(z ntl sz (0e). 2.1.6. Inequalities for the Coefficients of Polynomials In 1945 Visser [1] proved: Theorem 1.6.1. Ifz++ P(z) = 3> a,2” is a polynomial of degree w=0 n and |P(z)| <1 for |2| <1, then leo] + len] < 1. There are several results for self-inversive polynomials (see Chapter 1, Subsection 1.2.1). Using the previous theorem, Visser [1] stated the following result: n THEOREM 1.6.2. If z++ P(z) = ¥> a,z” is a self-inversive poly- =0 nomial of degree n and |P(2)| <1 for |z| <1, then mie leo] < An improvement of Theorem 1.6.2 was obtained and proved by Malik [2] in the following theorem: n THEOREM 1.6.3. Let z +> P(z) = ¥ a,2" be a self-inversive =0 polynomial of degree n and |P(2)| <1 for [zl < 1, then a9 (1.6.1) tat + (7) lac ( 0 we have $(n,0) = 1. Suppose now that k < 1. mS O62) sone S7()() = (1) +0 (62) +72") so Feo V()-C0 10 “Ee(S Ci) = S(n-1,k-1). Iterating this identity we prove Lemma 1.6.4. O In order to prove Lemma 1.6.5, Malik [2] used the Visser’s theorem 1.6.1, the previous lemma and the following result of de Bruijn [2] (see also Chapter 3, Subsection 3.1.3):INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 125 THEOREM 1.6.6. Let z++ P(z) be a polynomial of degree n such that |P(z)| < 1 for |z| < 1, and let z ++ Pe(z) be @ polynomial of degree n — 1, defined by 1 Pez) = E- z)P"(z) + P(z). Then, |Pe(z)| < 1 for |z| < 1 and |€| < 1. Proor oF THEOREM 1.6.3. Since P is given to be a self-inversive polynomial of degree n, we have an_, = a, (k = 0,1,...,n). Then Lemma 1.6.5 implies -1 n 1 6. <= (1.6.2) |eo+a(?) a |< 5 for |A| = 1. Choose arg A = — arg(ax/ao) a priori, then (1.6.2) implies (1.6.1). O An immediate corollary of Theorem 1.6.3 is the following result: Coroutary 1.6.7. Let z > P(z) = s a,z” be a self-inversive =0 polynomial of degree n, such that |P(2)|'< 1 for |z| < 1 and |ao| = 1/2. Then P(z) = Fast +&) for |al = 1. Using the method of Visser [1], van der Corput and Visser [1] proved the following result: n THEOREM 1.6.8. Let z + P(z) = > a,z” be a polynomial of =0 degree n such that |P(z)| < M for |2| <1. If ap, a, (p< 9) ore two coefficients such that for no other coefficient a, # 0 we have 5 =pmod(q-— p), then (1.6.3) lopl+ lagl < Mf. Also On (1.6.4) lol + lol < 3 if [P(e a.126 SELECTED POLYNOMIAL INEQUALITIES Also, they considered real trigonometric polynomials of the form . (1.6.5) T(0)= Y> ae*, ven where a, = @,, and proved that 2 1 Br fol + 2a] < maxI7(O), lool + Floal <¢ f”17(0108 for k > n/2. The constant 1/4 in the last inequality was improved by Boas [1] to (3 + V2)/(6r) = 0.23418. Later Boas [2] found the best constant C, for any positive y and k > n/2 in the following inequality 2m lal +2rlae] < Cy f” Ire). , Namely, if y is the smallest positive root of the equation : 1 sing = gue — 2y), then C, = (21 — 4y)-!. For 7 = 1/3 the corresponding constant becomes 0.213653... . Also, Boas [2] proved that 1 7 2m lal < jos [ ercoi, where p = 2[(n — k)/(2k)] + 1 is the largest odd integer with pk n. If on the other hand, z ++ Q(z) has all its zeros in the closed interior of the unit disk, then lao! + lon! < [bol men; max{[ao|, |x|} < lbm| — bol, m > 15 and (1.6.6) hold.128 SELECTED POLYNOMIAL INEQUALITIES THEOREM 1.6.11. If z 4 P(z) = x a,z” is a polynomial of =0 degree n and ay, ay (p < q) are two coefficients such that for no other coefficient a, # 0 we have s = pmod(q — p), then for every r>1, eres lal + lol < 208s (= [Veter a0)", where T(ir+l 062) Mt=c,= 2% = HD) Rs erdy BT (Rr + 3) The result is best possible. Remark 1.6.1. The constant Cy given by (1.6.7) appears in several extremal problems with polynomials (cf. Theorem 1.7.13 in Chapter 5). Rahman [1] also proved: n THEOREM 1.6.12. If z ++ P(z) = }> a,2” is a polynomial of =0 degree n andO 1, <2c(y( 2 [ |rre*yra0)” lol + log < 20) (5— f“1Pte ras) where if" = (r-1)/r -(1 ja r/(r-1) cte)= (ge f° Joos Hanya? a)" The result is best possible. THEOREM 1.6.13. If z+ Plz) = 35 a,2” is @ polynomial of v=0 degree n such that |P(z)| < M for |z| < 1, then 5M la,| + lag] < TrINEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 129 if either 0 <2p Q such that 4c An Soa where n € N. It is a theorem of Rahman [1] that 2 T (1.6.9) An(P,9) $ Pa where 0 < p (m—2)(q— p). Inequality (1.6.9) cannot be sharp in every case, since A 4 elim, An(p.9) = for p,q fixed, which contradicts (1.6.8). To find the value of A,(p,q) in the cases where (1.6.9) is not sharp seems to be a more difficult problem. Brickman, Rahman and Ruscheweyh [1] asked the following interesting question: Narrow the gap in (1.6.10) and (1.6.11) and to decide the question of whether or not alim,, B,, ezists. At the end of this subsection we mention some inequalities of Mabler [1] and Duncan [1]. n Let z+ P(z) = } a,2” be an arbitrary polynomial with complex =0 coefficients. Set H(P) = max{ ao], |ail,--- ,Jan]}, Le(P) = (Jaal” + lal! +--+ laal")"”" (7 > 0), and m(P)=exn( [oe Pte*)),INEQUALITIES INVOLVING ALGEBRAIC POLYNOMIALS 131 with M(P) = 0, when P(z) = 0. Then, for any two polynomials P and Q, we have the following inequalities for their lengths Ty(PQ)S L(P)Li(Q), — n(P #Q) < Li(P) + L1(Q). The height H, the length £,, and the measure M of P are connected by the inequalities (see Mahler (1]) = (era) H(P) < M(P) < H(P)VaFT, 2 Ly(P) < M(P) & Li(P)- If both P and Q are at most of degree n, then M(P¥Q) $ Li(P ¥Q) < La(P) + Li(Q) < 2"(M(P) + M(Q)). Duncan [1] proved that the factor 2” in the inequality M(P+Q) < 2"(M(P) + M(Q)) can be replaced by (°*)'/”. He also proved the following inequalities ~1f2 (*) LAP) < M(P) < Li(P) and Lx(PYLAQ) < y C2) CR) La PQ), where dg P = n and dgQ =m. Remark 1.6.2. These inequalities have applications in the theory of transcendental numbers. Some extremal problems for coefficients of polynomials will be treated in Chapter 5.132 SELECTED POLYNOMIAL INEQUALITIES 2.1.7. Integral Inequalities Lukacs [1] proved the following results: THEOREM 1.7.1. If P is an algebraic polynomial of degree at most 2n~1, which is non-negative on [a,b], then (1.7.1) P(a)< met) if ” pa)de and b (1.7.2) P(t) < wet i P(2) dz with equality in (1.7.1) if and only if P(z) = C(b- 2) {r (a 4 74) y i. and in (1.7.2) if and only if _ 2 bta\)* rerecte-afra(sta- Hs) ), where C is a positive constant and P,, is the Legendre polynomial of degree n defined by") Le Pal) = Tow a (-1)". Under the same conditions the following inequality n n b P(2)< ee f P(z)dz (a 2) with only real zeros, and such that P(-1) = P(1) = 0, P(z) # 0 for -l 2) having only real zeros, and let P(z) > 0 for z € (-1,1). Then -I “a max P(z). If, in addition, P(—1) = P(1) = 0, then io P(z)dz > 7 Hes +— a)” amas P(z). Dimovski and Cakalov i" provided the following generalization: TueoreM 1.7.7. Let P and Q be real polynomials of degree at most m and n, respectively having no zeros in the open disk |z| <1, and let P(x) > 0 and Q(z) > 0 for z € (-1,1). Then i. 2 men \ ie [Pees ( 5 ) _pm, PCE)_ max, Oe), with equality if and only if P(z) = a(1 tz)" and Q(z) = B(1 Fz)", where a and 3 are arbitrary positive constants. They also proved:INEQUALITIES WITH TRIGONOMETRIC POLYNOMIALS 135, THEOREM 1.7.8. Let P be a real polynomial of degree m having no zeros in the open disk |2| <1 and let_max P(z)=1. IQ is a -1 PE ("t ar 2.2, INEQUALITIES WITH TRIGONOMETRIC POLYNOMIALS 2.2.1. Inequalities for Trigonometric Sums ‘We start this subsection with some simple estimates of trigono- metric sums. Let m and m + 1 be natural numbers and 0 < @ < 2z. Since | by itt |= le torn Ea | [=i sin(o/2) we obtain fa a simple inequality ntm (2.1.1) | yet? 1 naa sin(0/2) Taking a monotonically decreasing sequence {r,}#°¢ such that (0<@< 2). lim r,=0 k+00 and applying the Abel’s inequality (cf. Mitrinovic [1, pp. 32-33]) to inequality (2.1.1) we find ntm ike Tntl Tn | Do ne | sin(6/2) ~ sin(0/2)° kant Finally, letting m — +00, we obtain an inequality of Blankfield and Zeitlin [1], namely (s ryei*® |< —™ (0<@<2n). Ree sin(8/2) Using a geometric method Kuz’min (1] proved the following result:136 SELECTED POLYNOMIAL INEQUALITIES THEOREM 2.2.1. Let n be a natural number > 2 and let ax (k = 1,...,) be real numbers such that 0 (bie 7) Me Assuming that (2.1.2) 2x (lends bal) =1