Lecture Notes ni Mathematics Edited yb .A Dold and .B Eckmann 698 limE Grosswald Bessel slaimonyloP galreV-regnirpS Berlin Heidelberg New kroY 8791 Author Emil Grosswald Department of Mathematics Temple University Philadelphia, PA 19122/USA 1 .6 .NAJ 9791 AMS Subject Classifications (1970): primary: 33 A 70 secondary: 33 A 65, 33 A 40, 33 75, A 33 A 45, 33-01,33-02, 33-03, 35 J 05, 41A10, 44A10, 30A22, 30A80, 30A84, 10F35, 12A20, 12D10, 60E05 ISBN 3-540-09104-1 Springer-Verlag Berlin Heidelberg New York ISBN 0-387-09104-1 Springer-Verlag New York Heidelberg Berlin This work is subject ot All copyright. rights era whether reserved, eht whole or specifically is concerned, of the material part those reprinting. of translation, reproduction of broadcasting, illustrations, re-use yb or photocopying machine means, similar dna storage ni data banks. Under § 45 of Copyright German the waL where copies era for than private other made .esu to the fee is a payable to fee the of amount the publisher, eb determined yb publisher. the with agreement :-c( yb Berlin Heidelberg Springer-Verlag 8791 Printed ni Germany Printing dna binding: Offsetdruck, Beltz Hemsbach/Bergstr. 012345-0413/1412 To HTEBAZILE EHCNALB and IV IV NA FOREWORD The present book consists of an Introduction, 15 Chapters, an Appendix, two Bibliographies and two Indexes.The chapters are numbered consecutively, from 1 to iS and are grouped into four parts, as follows: Part I - A short historic sketch i( Chapter) followed by the basic theory (5 Chapters); Part II - Analytic properties 5( Chapters); Part III- Algebraic properties 4( Chapters); Part IV - Applications and miscellanea (2 Chapters). According to its subject matter, the chapter on asymptotic properties would fit better into Part II; however, some of the proofs require results obtained only in Chapter 10 properties of zeros) and, for that reason, the chapter has been incorpo- rated into Part Ill. The Appendix contains a list of some 12 open problems. In the first bibliography are listed all papers, monographs, etc., that could be located and that discuss Bessel Polynomials. It is quite likely that, despite all efforts made, absolute completeness has not been achieved. The present writer takes this opportunity to apologize to all authors, whose work has been overlooked. A second, separate bibliography lists books and papers quoted in the text, but not directly related to Bessel Polynomials. References to the bibliographies are enclosed in square brackets. Those refer- ing to the second bibliography are distinguished by heavy print. So i refers to: W.H. Abdi - A basic analog of the Bessel Polynomials; while ~ refers to: M. Abramo- witz and l.E. Segun - Handbook of Mathematical Functions. Within each chapter, the sections, theorems, lemmata, corollaries, drawings, and formulae are numbered consecutively. If quoted, or referred to within the same chapter, only their own number is mentioned. If, e.g., in Chapter 10 a reference is made to formula (12), or to Section 2, this means formula (12), or Section 2 of Chapter 10. The same formula, or section quoted in another chapter, would be refer- red to as formula (I0.12), or Section (10.2), respectively. The same holds, mutatis mutandis, for theorems, drawings, etc. ~%ile writing this book, the author has received invaluable help from many col- leagues; to all of them he owes a great debt of gratitude. Of particular importance was the great moral support received from Professors H.L. Krall and O. Frink, as well as A.M. Krall. Professors Krall also read most of the manuscript and made valu- able suggestions for improvements. As already mentioned, there is no hope for an absolutely complete bibliography; however, many more omissions would have occurred, were it not for the help received, in addition to the mentioned colleagues, also from Professors R.P. Agarwal, W.A. AI- Salam, H.W. Gould, M.E.H. Ismail, C. Underhill, and A. Wragg. Last, but not least, thanks are due to Ms. Gerry Sizemore-Ballard, for her skill IV and infinite patience in typing the manuscript and to my daughter Vivian for her help with the Indexes. Part of the work on this book was done during the summer 1976, under a Summer Research Grant offered by Temple University and for which the author herewith ex- presses his gratitude. July 1978 .E Grosswald TABLE FO STNETNOC INTRODUCTION ...................................................... IX PART I CHAPTER 1 Historic Sketch ................................................ CHAPTER 2 Bessel Pol~nomials and Bessel Functions ........................ Differential equations, their d-forms and their S-forms. Polynomial solutions. Their relations to Bessel functions. Generalized Bessel Polynomials. RETPAHC 3 Recurrence Relations ........................................... ,~ Recurrence relations for yn,gn,~n. Representation of BP by determinants. Recurrence relations for the generalized polynomials. RETPAHC 4 Moments and Orthogonality on the Unit Circle .................... ~5 Moment problems and solutions by Stieltjes, Tchebycheff, Hamburger; the Bessel alternative. Weight function of the generalized BP. Moments of the simple BP. Orthogonality on the unit circle. PART I I CHAPTER S Relations of the BP to the classical orthonormal pol~nomials and to other functions ~l~ ................................ ..,,..°.. ., BP as generalized hypergeometric functions, as limits of Jacobi Polynomials, as Laguerre Polynomials; their representation by ~ittaker functions and by Lommel Polynomials. CHAPTER 6 - Generating Functions ........................................... h" Generating functions and pseudogenerating functions. Results of Krall and Frink, Burchnall, Ai-Salam, Brafman, Carlitz, and others. The theory of Lie groups and generating functions. Results of Weisner, Chatterjea, Das, McBride, Chen and Feng, and others. Different types of generating functions. CHAPTER 7 - Formulas of Rodrigues Type ..................................... ~ Methods of differential operators, of moments and of generating functions. Combinatorial Lemmas. CHAPTER 8 - The BP and Continued Fractions ................................. ~9 The BP as partial quotients. Approximation of the exponential function by ratios of BP. CHAPTER 9 - Expansions of functions in series of BP ........................ 6a Formal expansions in series of the polynomials ynZ;a,b), or 8n(Z;a,b .) The Boas-Buck theory of generalized Appell Polynomials. Convergence and summab~lity of expansions in BP. Applications to expansions of powers and of exponentials. PART llI CHAPTER i0 - Properties of the zeros of PB .................................. 75 Location of zeros. Results of Burchnall, Grosswald, Dickinson, Agarwal, Barnes, van Rossum, Nasif, Parodi, McCarthy, DoPey, Wragg and Underhill, Saff and Varga. Olver's theorem. Laguerre's Theorem. Results of Ismail and Kelker. Sums of powers of the zeros. IIIV CHAPTER ii - On the algebraic irreducibility of the BP ...................... 9)" Theorems of Dumas, Eisenstein, and Breusch. Newton Polygon. Degrees of possible factors. Cases of irreducibility. Schemes of factorization. Two conjectures. CHAPTER 12 - The Galois Groul! of BP ......................................... 416 Theorems of Schur, Dedekind, Jordan, Cauchy, and Burnside. Resolvent and Discriminant. The Galois Group of the irreducible BP is the symmetric group. Details of the case n = 8. C~PTER 13 - Asymptotic properties of the BP ................................ ~24 Case of n constant, z ~- O. Case of constant z, n + =. Results of Grosswald, Obreshkov, Do£ev. PART IV CHAPTER 14 - Applications ................................ i .................. d 3~ The irrati~-ality of e r (r rational) and of ~ . Solution of the wave equation. The infinite divisibility of the Student t-distribution. Bernstein's theorem. Electrical networks with maximally flat delay. The inversion of the Laplace transform. Salzer's theorem. CHAPTER iS - Miscellanea .................................................... q 50 Mention of the work by many authors, not discussed in the preceding chapters. APPENDIX - Some open problems related to BP ............................... ~62 BIBLIOCRAPI~ of books and papers related to BP .............................. J6/4 BIBLIOGRAPHY of literature not directly related to BP ....................... 17 ~ SUPdECT INDEX ................................................................ < 75 79 NAME INDEX ................................................................... ,I" PARTIAL LIST OF SYMBOLS ...................................................... N84 INTRODUCTION Let us look at a few problems that, at first view, have little in common. PROBLEM :i To prove that if r = a/b is rational, then e r is irrational; also that is irrational. Following C.L. Siegel $$ (who strea~dined an idea due to Hermite), one first deter- mines two polynomials An(X ) and Bn(X), both of degree n, such that eX+An(X)/Bn(X) has a zero of order (at least) 2n + 1 at x = .0 This means, in particular, that the power series expansion of Rn(X) = Bn(X)e x + An(X) starts with the term of degree 2n+l 2n+2 2n + ,i Rn(X) = ClX + c2x + ... say. By counting the number of conditions and the number of available coefficients, it turns out that An(X ) and Bn(X) are uni- quely defined, up to a multiplicative constant. By proper choice of this constant one can obtain that An(X ) and Bn(X ) should have integer coefficients. By simple manipulations one shows that An(-X ) = -Bn(X ) and that %(x) = (n!)-ix 2n+l I 0 1 tn(l_t)netXdt The last assertions are proved by effective construction of the polynomials involved (see Sections 14.2 and 14.3 for details). tI follows from the integral representation that )X(nR.' 'i ~ (nl)-lixl ' '2n+leXl and that Rn(X) > 0 for x ~ .0 If now e r = e a/b were rational, also e a would be rational; let q > 0 be its denominator. As already observed, Bn(a) and An(a ) are integers, $o that m = qRn(a ) = q(Bn~a)e a + An(a)) is a positive integer. Using the bound on Rn(a), 0 < m < q.(n~)-la2n+le a and, by Stirling's formula, 0 < m < q(a2n+lea/nn+i/2e-n(2z)i/2)(l+E), where e ÷ 0 as n + =. For sufficiently large n, 0 < m < ,I which is absurd, because m is an integer. Hence, e r cannot be rational. Next setting x = ~i, Rn(~i ) = -An(-~i)(-i ) + An(~i ) = (-l)n+ 1 2n+I I: tn(l-t)nsin zt dt n! ~0 (the last equality depends on some computations and will be justified in Chapter 14). The integrand is positive, so that Rn(ni) # .0 Let k = ~ where x stands for the greatest integer function; then A(x) + An(-X) is a polynomial in x 2 of degree k and 2 with integer coefficients. Hence, if ~ is rational, with denominator q > O, then qkRn(~i) = qk{AnC~i ) + An(-~i)} = m, an integer, possibly negative, but certainly X # .0 Also, by using the integral representation of Rn(X), 0 < lmI = 2/lq( n)2 2n+l qk!Rn(wi) l < (n)-lqk ' - ~ ~ < nn+I/2e-n(2v)I/2 (l+e) a( ÷ 0 for n ~ =), or 0 < !nI < i for sufficiently large n. This is, of course impossible for integral m. Hence 2, and afortiori w are irrational. A (highly nontrivial) modification of this proof permits one to show much more, namely that e r is actually transcendental for real, rational .r In particular, for r = ,i this implies the transcendency of e itself. PROBLEM .2 To prove that the Student t-distribution of 2n+l degrees of freedom is infinitely divisible. We do not have to enter here into the probabilistic relevance, or even into the exact meaning of this important problem. Suffice it to say that, based on the pioneering work of Paul Le~ .44 and of Gnedenko and Kolmogorov ~9, Kelker 65 and then Ismail and Kelker 60 proved that the property holds if, and only if the Kn_i/2 ( xx~ ) function ~(x) = ~x Kn+i/2(~x) is completely monotonic on 0,~), which means that (-l)k¢(k)(x) > 0 for 0 < x < = and all integral k ~ .0 Now, it is well-knowm (see, e.g. ~, 10.2.17) that if the index of K )z( (the so called modified Hankel func- tion) is of the form n+i/2 (n an integer), then (2z/~)I/2eZKn+i/2(z) = Pn(I/z), with Pn(U) a polynomial of exact degree .n Previous relation can now be written as Pn_l(X -I/2) Pn_l(X I/2) n ¢(x - xl/2pn(X_i/2 ) , or, with Pn(U) = u Pn(i/u), ¢(x) - (xl/2) We now use Pn Bernstein's theorem (see 68); this asserts that ¢(x) is completely monotonic if, and only if it is the Laplace transfarm of a function G(t), non-negative on 0 < t < ~. In the present case it is possible to study the pol)~omials Pn(X) and compute G(t). tI turns out that G(t) ~ 0 for small t > 0 and also for t sufficiently large. This alone is not quite sufficient to settle the problem, but if we also knew that G(t) is monotonic, then the conclusion immediately follows. In fact, by playing around with G(t), one soon suspects that it is not only monotonic, but actually completely monotonic. In order to prove this, one appeals once more to Bernstein's theorem and finds that G(t) is the Laplace transform of ¢(x) = ~( 2 )x -1/2{i ÷ n X ~j(x+ 2- j=l aj) i}, where ~l,a2 ..... a n are the zeros of the polynomial Pn(U). A detailed study of these zeros permits one to reduce the large bracket to the form xn/q(x), where q(x) is a polynomial with real coefficients and such that q(x) > 0 at least for x > - min lj~l .2 This shows that ¢(x) > 0 for l<jsn IX 0 < x < ~; hence, G(t) >_ 0 on 0 < t < - and ¢(x) is indeed completely monotonic, as we wanted to show. PROBLEM .3 To solve the equation with partial derivatives 1 32V )1( AV = -~-- where A is the Laplacian, c ~t 2 32 2 3 i 32 3 1 32 A = ~3r + r Dr + -2-r (~-~ + cot 0 ~) + r2sin2 e 302 with the boundary conditions )i( and (ii): )i( V = V(r,e,¢,t) is symmetric with respect to a "polar axis" through the origin, so that, in fact, V = V(r,O,t) only; (ii) V is monochromatic, i.e., all "waves" have the same frequency ~; and with the initial condition (iii) the values of V are prescribed along the polar axis at t = 0, say V(r,0,0) = f(r), a given function of r. Here r, ,0 and ¢ are the customary spherical coordinates, t stands for the time and c represents dimensionwise a velocity. Conditions )i( and (ii) are imposed only in order to simplify the problem and can be omitted, but the added complexity can easily be handled by classical methods and has nothing to do with the problem on hand. Following the lead of Krall and Frink 68 we look, in particular, at solutions of )1( of the form (obtained by separation of variables) u = r-ly(i/ikr) L(cos 0)e ik(ct-r) Here y and L are, so far, undetermined functions and we shall determine them precise- ly hy the condition that u be a solution of (I), while k is a parameter related to the frequency ~ by k = ~/c. On account of (ii), k is a well defined constant. The (artificially looking) device of introducing complex elements into this physical problem is useful for obtaining propagating, rather then stationary waves. The real components v, w of u = v + iw will be real solutions of )I( and represent waves traveling with the velocity c. For c = 0 and with x = i/ikr, one obtains, of course, directly a real, stationary solution of (i). We now substitute u into (i), by taking into account that 8u/3# = O, and obtain, with x = I/ikr and z = cos 8, that L(z)(x2y"(x) + (2+2x)y'(x)) + y(x)((l-z2)L"(z) - 2z L'(z))= O, or, equivalently, x2~'"(x)+(2+2x))' '(x) = _ (l-zZ}n"(z)-2zn '(z) y(x) L(z)