Orthogonal polynomials

Transcription

1 Korteweg-de Vries Institute, University of Amsterdam Notes of two lectures given at the LHCPHENOnet School Integration, Summation and Special Functions in Quantum Field Theory, RISC, Schloss Hagenberg, Austria, 9 13 July 2012 last modified: February 25, 2013

2 Contents 1 Definition of orthogonal polynomials 2 Three-term recurrence relation 3 Christoffel-Darboux formula 4 Zeros of orthogonal polynomials 5 Jacobi, Laguerre and Hermite polynomials 6 Very classical orthogonal polynomials 7 Electrostatic interpretation of zeros 8 Quadratic transformations 9 Kernel polynomials 10 Criteria for bounded support and for uniqueness of orthogonality measure 11 and continued fractions 12 Measures in case of non-uniqueness 13 Gauss quadrature 14 Finite systems of orthogonal polynomials 15 Classical orthogonal polynomials; Askey scheme 16 The q-case

3 Some books on orthogonal polynomials G. Szegő,, Amer. Math. Soc., Fourth ed., T. S. Chihara, An introduction to orthogonal polynomials, Gordon and Breach, 1978; reprinted, Dover, G. E. Andrews, R. Askey and R. Roy, Special Functions, Cambridge University Press, 1999.

4 R. Koekoek, P. A. Lesky and R. F. Swarttouw, Hypergeometric orthogonal polynomials and their q-analogues, Springer-Verlag, 2010; in particular Chapters 9, 14, based on the Koekoek-Swarttouw report NIST Handbook of Mathematical Functions, Cambridge University Press, 2010; ; in particular Ch. 18 on Orthogonal polynomials.

5 Definition of orthogonal polynomials P is the space of all polynomials in one variable with real coefficients. This is a real vector space. Assume a (positive definite) inner product f, g (f, g P) on P. Orthogonalize the sequence 1, x, x 2,... with respect to the inner product (Gram-Schmidt), resulting into p 0, p 1, p 2,.... So p 0 (x) = 1 and, if p 0 (x), p 1 (x),..., p n 1 (x) are already produced and mutually orthogonal, then n 1 p n (x) := x n x n, p k p k, p k p k(x). k=0 Indeed, p n (x) is a linear combination of 1, x, x 2,..., x n, and n 1 p n, p j = x n x n, p k, p j p k, p k p k, p j k=0 = x n, p j x n, p j p j, p j p j, p j = 0 (j = 0, 1,..., n 1).

6 Definition of orthogonal polynomials (cntd.) Constants h n and k n : p n, p n = h n, p n (x) = k n x n + polynomial of lower degree. The p n are unique up to a nonzero constant real factor. We may take them, for instance, orthonormal (h n = 1, if also k n > 0 then unique) or monic (k n = 1). In general we want x f, g = f, x g. This is true, for instance, if for a weight function w(x) 0: f, g := or if for weights w j > 0: b a f, g := f (x) g(x) w(x) dx, f (x j ) g(x j ) w j. j=0

7 Intermezzo about measures The cases with the weight function and with the weights are special cases of a (positive) measure µ on R: dµ(x) = w(x) dx on (a, b) and = 0 outside (a, b); resp. µ = j=1 w j δ xj, where δ xj is a unit mass at x j. A measure µ on R can also be thought as a non-decreasing M function µ on R. Then f (x) dµ(x) = lim f (x) dµ(x) M R M can be considered as a Riemann-Stieltjes integral. µ has in x a mass point of mass c > 0 if µ has a jump c at x, i.e., if lim δ 0 ( µ(x + δ) µ(x δ) ) = c > 0. The number of mass points is countable. More generally, the support of µ consists of all x R such that µ(x + δ) µ(x δ) > 0 for all δ > 0. This set supp(µ) is always closed.

8 Definition of orthogonal polynomials (cntd.) In the most general case let µ be a (positive) measure on R (of infinite support, i.e., not µ = N j=1 w j δ xj ) such that for all n = 0, 1, 2,... x n dµ(x) <, and take R f, g := R f (x) g(x) dµ(x). A system {p 0, p 1, p 2,...} obtained by orthogonalization of {1, x, x 2,...} with respect to such an inner product is called a system of orthogonal polynomials (OP s) with respect to the orthogonality measure µ. Typical cases are: (weight function) dµ(x) = w(x) dx on an interval I. (weights) µ = j=1 w j δ xj.

9 First examples of orthogonal polynomials 1 Legendre polynomials P n (x), orthogonal on [ 1, 1] with respect to the weight function 1. Normalized by P n (1) = 1. 2 Hermite polynomials H n (x), orthogonal on (, ) with respect to the weight function e x 2. Normalized by k n = 2 n. 3 Charlier polynomials c n (x, a), orthogonal on the points x = 0, 1, 2,... with respect to the weights a x /x! (a > 0). Normalized by c n (0; a) = 1. The h n can be computed: 1 2 π 1 2 e a x=0 1 1 P m (x) P n (x) dx = 1 2n + 1 δ m,n, H m (x) H n (x) e x 2 dx = 2 n n! δ m,n, c m (x, a) c n (x, a) ax x! = a n n! δ m,n.

10 Three-term recurrence relation Theorem p n (x) satisfy xp n (x) = a n p n+1 (x) + b n p n (x) + c n p n 1 (x) (n > 0), xp 0 (x) = a 0 p 1 (x) + b 0 p 0 (x) with a n, b n, c n real constants and a n c n+1 > 0. Also a n = k n c n+1, = a n. k n+1 h n+1 h n Indeed, xp n (x) = n+1 k=0 α kp k (x), and if k n 2 then Furthermore, c n+1 = xp n+1, p n p n, p n xp n, p k = p n, xp k = 0, hence α k = 0. = xp n, p n+1 = xp n, p n+1 h n+1 h n+1 = a n. h n p n+1, p n+1 h n h n Hence a n c n+1 = a 2 n h n+1 /h n > 0. Hence c n+1 /h n+1 = a n /h n.

11 Three-term recurrence relation (cntd.) Theorem (Favard) If polynomials p n (x) of degree n (n = 0, 1, 2,...) satisfy xp n (x) = a n p n+1 (x) + b n p n (x) + c n p n 1 (x) (n > 0), xp 0 (x) = a 0 p 1 (x) + b 0 p 0 (x) with a n, b n, c n real constants and a n c n+1 > 0 then there exists a (positive) measure µ on R such that the polynomials p n (x) are orthogonal with respect to µ. Remarks 1 The measure µ may not be unique (up to constant factor). 2 If µ unique then the polynomials are dense in L 2 (µ). 3 If there is a µ with bounded support then µ is unique.

12 Favard Il a depuis longtemps une belle notoriété dans le monde mathématique lorsqu il est mobilisé en septembre 1939 comme officier d artillerie. Fait prisonnier en juin 1940, il est envoyé à l oflag XVIII, à Lienz (Autriche), où il crée une Faculté des Sciences dont il est le doyen. Des mathématiciens autrichiens veulent le faire libérer s il consent à enseigner à Vienne; il refuse. Dès 1941, il a été nommé professeur à la faculté des Sciences de Paris, mais il ne prend ses fonctions à la Sorbonne qu à sa libération en 1945.

13 Three-term recurrence relation (cntd.) For orthonormal polynomials: xp n (x) = a n p n+1 (x) + b n p n (x) + a n 1 p n 1 (x) (n > 0), xp 0 (x) = a 0 p 1 (x) + b 0 p 0 (x). For monic orthogonal polynomials: xp n (x) = p n+1 (x) + b n p n (x) + c n p n 1 (x) (n > 0), xp 0 (x) = p 1 (x) + b 0 p 0 (x) with c n = h n /h n 1 > 0. If the orthogonality measure is even (µ(e) = µ( E)) then p n ( x) = ( 1) n p n (x), hence b n = 0, so xp n (x) = a n p n+1 (x) + c n p n 1 (x). Examples: Legendre and Hermite polynomials.

14 Moment functional The recurrence relation (with a n c n+1 > 0) determines the orthogonal polynomials p n (x) (up to constant factor because of the choice of the constant p 0 ). The p n determine (up to constant factor) the moment functional π π, 1 on P by the rule p n, 1 = 0 for n > 0. Thus the inner product f, g = fg, 1 on P is determined by the recurrence relation, independent of the choice of the orthogonality measure µ. The moment functional π π, 1 on P is determined by the moments µ n := x n, 1. The condition a n c n+1 > 0 is equivalent to positive definiteness of the moments, stated as n := det(µ i+j ) n i,j=0 > 0 (n = 0, 1, 2,...).

15 Christoffel-Darboux kernel P: space of all polynomials; P n : space of polynomials of degree n ; p n (x): orthogonal polynomials with respect to measure µ. Christoffel-Darboux kernel: Then (Π n f )(x) := K n (x, y) := R n j=0 p j (x)p j (y) h j K n (x, y) f (y) dµ(y) defines an orthogonal projection Π n : P P n. Indeed, if f (y) = k=0 α kp k (y) (finite sum) then n α k (Π n f )(x) = p j (x) p j (y) p k (y) dµ(y) = h j j=0 k=0 R n α j p j (x). j=0

16 Christoffel-Darboux formula n j=0 p j (x)p j (y) = k n p n+1 (x)p n (y) p n (x)p n+1 (y) h j h n k n+1 x y (x y), = k n h n k n+1 (p n+1 (x)p n(x) p n(x)p n+1 (x)) (x = y). Indeed, Hence xp j (x) = a j p j+1 (x) + b j p j (x) + c j p j 1 (x), yp j (y) = a j p j+1 (y) + b j p j (y) + c j p j 1 (y). (x y)p j (x)p j (y) h j = a j h j (p j+1 (x)p j (y) p j (x)p j+1 (y)) c j h j (p j (x)p j 1 (y) p j 1 (x)p j (y)). Use c j /h j = a j 1 /h j 1. Sum from j = 0 to n. Use that a n = k n /k n+1. We have the C-D formula for x y.

17 Zeros of orthogonal polynomials Theorem Let p n (x) be an orthogonal polynomial of degree n. Let µ have support within the closure of the interval (a, b). Then p n has n distinct zeros on (a, b). Proof (for (a, b) = (, )) Assume k n > 0 (no loss of generality). Suppose p n has k < n sign changes on R at x 1, x 2,..., x k. Hence p n (x)(x x 1 )... (x x k ) 0 on R. Hence R p n(x)(x x 1 )... (x x k ) dµ(x) > 0. But by orthogonality R p n(x)(x x 1 )... (x x k ) dµ(x) = 0. Contradiction.

18 Zeros of orthogonal polynomials (cntd.) By the Christoffel-Darboux formula: If k n, k n+1 > 0 then p n+1 (x)p n(x) p n(x)p n+1 (x) = h nk n+1 k n n p j (x) 2 > 0. h j j=0 Hence, if y, z are two successive zeros of p n+1 then p n+1 (y)p n(y) > 0, p n+1 (z)p n(z) > 0. Since p n+1 (y) and p n+1 (z) will have opposite signs, p n (y) and p n (z) will have opposite signs. Hence p n must have a zero in (y, z). Theorem The zeros of p n and p n+1 alternate.

19 Graphs of Legendre polynomials Alternating zeros of Legendre polynomials P 8 (x) (blue graph) and P 9 (x) (red graph):

20 Graphs of Legendre polynomials (cntd.)

21 Jacobi polynomials Definition of Jacobi polynomials p n (x) = P (α,β) n (x), dµ(x) = w(x) dx on [ 1, 1], w(x) = (1 x) α (1 + x) β (α, β > 1), P (α,β) n (1) = (α + 1) n. n! Explicit expression (see Paule for hypergeometric functions) P (α,β) n (x) = (α + 1) n n! Symmetry P (α,β) n 2F 1 ( n,n+α+β+1 α+1 ; z ( n, n + α + β + 1 2F 1 ; 1 α ). (1 x) ( x) = ( 1) n P (β,α) n (x). Hence ) = ( 1)n (β+1) n (α+1) n 2F 1 ( n,n+α+β+1 β+1 ; 1 z ).

22 Jacobi polynomials (cntd.) Second order differential equation for p n (x) = P (α,β) n (x) (1 x 2 )p n(x) + ( β α (α + β + 2)x ) p n(x) = n(n + α + β + 1) p n (x). Shift operator relations d n (x) = 1 2 (n + α + β + 1)P(α+1,β+1) n 1 (x), dx P(α,β) (1 x 2 ) d dx P(α+1,β+1) n 1 (x) + ( β α (α + β + 2)x ) P (α+1,β+1) n 1 (x) = (1 x) α (1 + x) β d ( ) (1 x) α+1 (1 + x) β+1 P (α+1,β+1) dx n 1 (x) = 2n P (α,β) n (x). Rodrigues formula P (α,β) n (x) = ( 1)n 2 n n! (1 x) α (1 + x) β ( ) d n ( (1 x) α+n (1 + x) β+n). dx

23 Jacobi polynomials (special cases) Gegenbauer or ultraspherical polynomials (α = β = λ 1 2 ) Cn λ (x) := (2λ) n (λ ) n P (λ 1 2,λ 1 2 ) n (x). Legendre polynomials (α = β = 0) P n (x) := P (0,0) n (x). Chebyshev polynomials (α = β = ± 1 2 ) T n (cos θ) := cos(n θ) = n! ( 1 2 ) n U n (cos θ) := sin(n + 1)θ sin θ = (2) n ( 3 2 ) n P ( 1 2, 1 2 ) n (cos θ), P ( 1 2, 1 2 ) n (cos θ).

24 Laguerre polynomials Definition of Laguerre polynomials p n (x) = L α n (x), dµ(x) = w(x) dx on [0, ), w(x) = x α e x L α n (0) = (α + 1) n. n! Explicit expression L α n (x) = (α + 1) n n! (α > 1), 1F 1 ( n α + 1 ; x ).

25 Laguerre polynomials (cntd.) Second order differential equation for p n (x) = L α n (x) x p n(x) + (α + 1 x) p n(x) = n p n (x). Shift operator relations d dx Lα n (x) = L α+1 n 1 (x), x d dx Lα+1 n 1 (x) + (α + 1 x)lα+1 n 1 (x) = x α e x d dx Rodrigues formula L α n (x) = x α e x ( ) d n ( x n+α e x). n! dx ( x α+1 e x L α+1 n 1 (x) ) = n L α n (x).

26 Hermite polynomials Definition of Hermite polynomials p n (x) = H n (x), dµ(x) = e x 2 dx, k n = 2 n. Explicit expression H n (x) = n! [n/2] j=0 ( 1) j (2x) n 2j j! (n 2j)! Second order differential equation H n (x) 2xH n(x) = 2nH n (x). Shift operator relations H n(x) = 2n H n 1 (x), Rodrigues formula H n (x) = ( 1) n e x 2 ( d dx ) n ( e x 2).. H n 1 (x) 2xH n 1(x) = H n (x)

27 Derivation of previous formulas (a, b) open interval; w, w 1 > 0 on (a, b) and C 1. On (a, b) monic OP s p n (x), q m (x) with respect to w resp. w 1. Then under suitable boundary assumptions for w and w 1 : b a p n(x) q m 1 (x) w 1 (x) dx b = p n (x) w(x) 1 d ( w1 (x) q m 1 (x) ) w(x) dx. a dx Suppose that for certain a n 0 : w(x) 1 d ( w 1 (x) x n 1) = a n x n + polynomial of degree < n. dx Then p n(x) = n q n 1 (x), w(x) 1 d dx (w 1(x) q n 1 (x)) = a n p n (x), w(x) 1 d dx ( w1 (x) p n(x) ) = na n p n (x), n b a b q n 1 (x) 2 w 1 (x) dx = a n p n (x) 2 w(x) dx. a

28 Derivation of previous formulas (cntd.) Work with monic Jacobi polynomials p (α,β) n (x). Then (a, b) = ( 1, 1), w(x) = (1 x) α (1 + x) β, p n (x) = p (α,β) n (x), w 1 (x) = (1 x) α+1 (1 + x) β+1, q m (x) = p (α+1,β+1) m (x). Then a n = (n + α + β + 1), ( (1 x 2 ) d dx + ( β α (α + β + 2)x )) p (α+1,β+1) n 1 For x = 1: Then iterate: p (α,β) n (1) = p (α,β) n (1) = (x) = (n + α + β + 1) p (α,β) n (x). 2(α + 1) n + α + β + 1 p(α+1,β+1) n 1 (1). 2 n (α + 1) n (n + α + β + 1) n. So we know p n (1)/k n, which is independent of the normalization.

29 Derivation of previous formulas (cntd.) Hypergeometric series representation of Jacobi polynomials obtained by Taylor expansion: p (α,β) n (x) = = n (x 1) k k=0 k! n (x 1) k k=0 k! Quadratic norm h n obtained by iteration: = 1 1 p (α,β) n (x) 2 (1 x) α (1 + x) β dx n n + α + β ( ) d k p (α,β) n (x) dx x=1 n! (n k)! p(α+k,β+k) n k (1). p (α+1,β+1) n 1 (x) 2 (1 x) α+1 (1 + x) β+1 dx. So we know h n /k 2 n, which is independent of the normalization.

30 Very classical orthogonal polynomials Jacobi, Laguerre and Hermite polynomials together, for the given parameter ranges, are called very classical orthogonal polynomials. Up to constant factors and up to transformations x ax + b of the argument they are uniquely determined as OP s p n (x) satisfying any of the following three criteria: (Bochner s theorem) The p n are eigenfunctions of a second order differential operator. The polynomlals p n+1 (x) are again orthogonal polynomials. The polynomials are orthogonal with respect to a positive C weight function w(x) on an open interval I and there is a polynomial X(x) such that the Rodrigues formula holds on I: p n (x) = const. w(x) 1 ( d dx ) n ( X(x) n w(x) ).

31 Rodrigues Benjamin Olinde Rodrigues ( ) lived in Paris. He had in his thesis the Rodrigues formula for the Legendre polynomials. Afterwards he became a banker and became a relatively wealthy man as he supported the development of the French railway system. Rodrigues was an early socialist. He argued that working men were kept poor by lending at interest and by inheritance. He also argued in favour of mutual aid societies and profit-sharing for workers. Rodrigues joined the Paris Ethnological Society. He argued strongly that all races had equal aptitude for civilization in suitable circumstances and that women will one day conquer equality without any restriction. These views were much criticised by other members: Rodrigues was sentimental and science proved that he was wrong.

32 Limits for very classical OP s Monic versions: Jacobi: p (α,β) n (x), w(x) = (1 x) α (1 + x) β on ( 1, 1) Laguerre: l α n (x), w(x) = e x x α on (0, ) Hermite: h n (x), w(x) = e x 2 on (, ) α n/2 p (α,α) n (x/α 1/2 ) h n (x), (1 x 2 /α) α e x 2, α ( β/2) n p (α,β) n (1 2x/β) l α n (x), x α (1 x/β) β x α e x, β (2α) n/2 l α n ((2α) 1/2 x + α) h n (x), (1 + (2/α) 1/2 x) α e (2α)1/2x e x 2, Laguerre Jacobi Hermite α

33 Electrostatic interpretation of zeros Let p n (x) = P (2p 1,2q 1) n (x)/k n = (x x 1 )(x x 2 )... (x x n ) be monic Jacobi polynomials (p, q > 0). We know that (1 x 2 )p n(x) + 2(q p (p + q)x)p n(x) = n(n + 2p + 2q 1)p n (x). Hence i.e., 1 2 i.e., (1 xk 2 )p n(x k ) + 2(q p (p + q)x k )p n(x k ) = 0, p n(x k ) p n(x k ) + j, j k 1 x k x j + p x k 1 + p x k 1 + q x k + 1 = 0, q x k + 1 = 0, i.e., ( V )(x 1,..., x n ) = 0, where V (y 1,..., y n ) = i<j log(y j y i ) p j log(1 y j ) q j log(1 + y j ). Logarithmic potential from charges q, 1,..., 1, p at 1 < y 1 <... < y n < 1 achieves minimum at the zeros of P (2p 1,2q 1) (x). n

34 Stieltjes Thomas Stieltjes, assistant at Leiden astronomical observatory. Was corresponding with Hermite honorary doctorate of Leiden University professor in Toulouse.

35 Quadratic transformations P (α,α) 2n 1 0 = (x) is polynomial p n (2x 2 1) of degree n in x 2. For m n 0 Hence Similarly Theorem p m (2y 2 1)p n (2y 2 1) (1 y 2 ) α dy = const. P(α,α) 2n (x) P (α,α) 2n 1 1 p m (x)p n (x) (1 x) α (1 + x) 1 2 dx. (α, 1 2 ) (1) = P n (2x 2 1). P (α, 1 2 ) n (1) P (α,α) 2n+1 (x) (α, xp 1 2 ) n (2x 2 1) P (α,α) =. 2n+1 (1) P (α, 1 2 ) n (1) Let p n (x) be monic orthogonal polynomial with respect to even weight function w(x) on R. Then p 2n (x) = q n (x 2 ) and p 2n+1 (x) = x r n (x 2 ) with q n (x) and r n (x) OP s on [0, ) with respect to weight functions x 1 2 w(x 1 2 ) resp. x 1 2 w(x 1 2 ).

36 Kernel polynomials Christoffel-Darboux formula: n p j (x)p j (y) = k n p n+1 (x)p n (y) p n (x)p n+1 (y) h j h n k n+1 x y j=0 Suppose the orthogonality measure µ has support within (, b] and fix y b. Then for k n 1: b K n (x, y) x k (y x) dµ(x) = y k (y y) = 0. (x y). Hence x q n (x) = K n (x, y) is an OP of degree n on (, b] with respect to the measure (y x) dµ(x). Hence q n (x) q n 1 (x) = p n(y) h n p n (x), p n (y)p n+1 (x) p n+1 (y)p n (x) = h nk n+1 k n (x y)q n (x).

37 True interval of orthogonality p n (x). Let p n (x) have zeros x n,1 < x n,2 <... < x n,n. Then x i,i > x i+1,i >... > x n,i ξ i, and x j,1 < x j+1,2 <... < x n,n j+1 η j. Definition The closure of the interval (ξ 1, η 1 ) is called the true interval of orthogonality of the OP s p n (x). Remarks The true interval of orthogonality I has the following properties. 1 I is the smallest closed interval containing all zeros x n,i. 2 There is an orthogonality measure µ for the p n (x) such that I is the smallest closed interval containing the support of µ. 3 If µ is any orthogonality measure for the p n (x) and J is a closed interval containing the support of µ then I J.

38 Criteria for bounded support of orthogonality measure xp n (x) = p n+1 (x) + b n p n (x) + c n p n 1 (x) (c n > 0). Theorem 1 {b n } bounded, {c n } unbounded = (ξ 1, η 1 ) = (, ). 2 {b n }, {c n } bounded [ξ 1, η 1 ] bounded. 3 b n b, c n c (b, c finite) = supp(µ) bounded with at most countably many points outside [b 2 c, b + 2 c ] and b ± 2 c limit points of supp(µ). Example Monic Jacobi polynomials kn 1 P (α,β) n (x) : β 2 α 2 b n = (2n + α + β)(2n + α + β + 2) 0. c n = 4n(n + α)(n + β)(n + α + β) (2n + α + β 1)(2n + α + β) 2 (2n + α + β + 1) 1 4. Hence [b 2 c, b + 2 c ] = [ 1, 1].

39 Criteria for uniqueness of orthogonality measure (See Shohat & Tamarkin, The problem of moments, AMS, 1943.) Let p n (x) be orthonormal polynomials, i.e., solutions of xp n (x) = a n p n+1 (x)+b n p n (x)+a n 1 p n 1 (x) ( Put ρ(z) := p n (z) 2) 1 (z C). Theorem n=0 (a n > 0, b n R). The orthogonality measure is not unique iff ρ(z) > 0 for all z C. Hence it is unique iff ρ(z) = 0 for some z C. In fact, if there is a unique orthogonality measure µ then ρ(x) = µ({x}) if µ has a mass point at x, and ρ(z) = 0 for z C outside the mass points of µ. In case of non-uniqueness, for each x R the largest possible jump of a measure µ at x is ρ(x) and there is a measure realizing this jump.

40 Criteria for uniqueness of orthog. measure (cntd.) Orthonormal polynomials p n (x). xp n (x) = a n p n+1 (x) + b n p n (x) + a n 1 p n 1 (x) x p n(x) k n = p n+1(x) k n+1 + b n p n (x) k n µ n := x n, 1 (moments). + a 2 n 1 p n 1 (x) k n 1 (a n > 0, b n R), (monic version), Theorem (Carleman) There is a unique orthogonality measure for the p n if one of the following two conditions is satisfied. 1 2 n=1 n=1 µ 1/(2n) 2n =. a 1 n =.

41 Criteria for uniqueness of orthog. measure: Examples Hermite: µ 2n = x 2n e x 2 dx = Γ(n ). log Γ(n ) = n log(n ) + O(n) as n, so µ 1/(2n) 2n (n ) 1 2. Hence µ 1/(2n) 2n = : unique orthogonality measure. n=1 Monic Laguerre p n (x) = kn 1 L α n (x) : xp n (x) = p n+1 (x) + (2n + α + 1)p n (x) + n(n + α)p n 1 (x). n=0 1 (n(n + α)) 1/2 = : unique orthogonality measure. Also Lα n (0) 2 Γ(n + α + 1) = h n Γ(n + 1)Γ(α + 1) nα. n=1 nα = (α > 1): unique orthogonality measure.

42 Example of non-unique orthogonality measure e u2 (1 + C sin(2πu)) du = π 1/2. Substitute u = log x 1 2 (n + 1) and take 1 < C < 1. π x n (1 + C sin(2π log x)) e log2 x dx = e (n+1)2 /4. The moments are independent of C. The corresponding orthogonal polynomials are the Stieltjes-Wigert polynomials.

43 and continued fractions Let p n (x) be monic OP s given by p 0 (x) = 1, p 1 (x) = x b 0, xp n (x) = p n+1 (x) + b n p n (x) + c n p n 1 (x) (n 1, c n > 0). The monic first associated OP s or numerator polynomials p (1) n (x) are defined by p (1) 0 (x) = 1, p(1) 1 (x) = x b 1, xp (1) n (x) = p (1) n+1 (x) + b n+1p (1) n (x) + c n+1 p (1) n 1 (x) (n 1). Recursively define F 1 (x) := 1 1, F 2 (x) := x b 0 x b 0 c, 1 x b 1 1 F 3 (x) := x b 0 c 1 x b 1 c 2 x b 2 by replacing b n 1 by b n 1 + x b n Theorem (essentially Stieltjes), and F n+1 (x) obtained from F n (x) c n (continued fraction). F n (x) = p(1) n 1 (x), p (1) p n (x) n 1 (y) = 1 p n (y) p n (x) dµ(x). µ 0 R y x

44 OP s and continued fractions (cntd.) F n (z) := 1 z b 0 c 1 z b 1 c n 2 z b n 2 c n 1 = p (1) n 1 (z) z b n 1 p n (z) Suppose that there is a (unique) orthogonality measure µ of bounded support for the p n. Let [ξ 1, η 1 ] be the true interval of orthogonality. Theorem (Markov). lim F n(z) = 1 η1 dµ(x) n µ 0 ξ 1 z x on compact subsets of C\[ξ 1, η 1 ]. uniformly

45 Measures in case of non-uniqueness Take p n and p (1) n orthonormal: p 0 (x) = 1, p 1 (x) = (x b 0 )/a 0, xp n (x) = a n p n+1 (x) + b n p n (x) + a n 1 p n 1 (x) (n 1, a n > 0), p (1) 0 (x) = 1, p(1) 1 (x) = (x b 1)/a 1, xp (1) n (x) = a n+1 p (1) n+1 (x) + b n+1p (1) n (x) + a n p (1) n 1 (x). (n 1). Let µ 0 = 1, µ 1, µ 2,... be the moments for the p n. Assume non-uniquness of µ satisfying R x n dµ(x) = µ n. The set of these µ is convex and weakly compact. Then the following functions are entire. A(z) := z n=0 C(z) := 1 + z p (1) n (0) p (1) n (z), B(z) := 1 + z n=1 p n (0) p (1) n 1 (z), n=1 p (1) n 1 (0) p n(z), D(z) = z p n (0) p n (z). n=0

46 Measures in case of non-uniqueness (cntd.) Theorem (Nevanlinna, M. Riesz) The identity dµ φ (t) t z R A(z)φ(z) C(z) = B(z)φ(z) D(z) (Im z > 0) gives a one-to-one correspondence φ µ φ between the set of functions φ being either identically or a holomorphic function mapping the open upper half plane into the closed upper half plane (Pick function) and the set of measures solving the moment problem. Furthermore the measures µ t (t R { }) are precisely the extremal elements of the convex set, and also precisely the measures µ solving the moment problem for which the the polynomials are dense in L 2 (µ). All measures µ t are discrete.

47 Gauss quadrature Let be given n real points x 1 < x 2 <... < x n. Put p n (x) := (x x 1 )... (x x n ). Let l k (x) be the unique polynomial of degree < n such that l k (x j ) = δ k,j (j = 1,..., n). Then (Lagrange interpolation polynomial) l k (x) = j; j k (x x j) j; j k (x k x j ) = p n (x) (x x k ) p n(x k ) for all polynomials r of degree < n: Theorem (Gauss quadrature) Let p n be an OP with respect to µ. Put λ k := R l k(x) dµ(x). Then λ k = R l k(x) 2 dµ(x) > 0 and for all polynomials of degree 2n 1: R f (x) dµ(x) = n k=1 λ k f (x k ). and r(x) = n k=1 r(x k) l k (x).

48 Gauss quadrature: Proof Let f (x) be polynomial of degree 2n 1. Then for certain polynomials q(x) and r(x) of degree n 1: f (x) = q(x)p n (x) + r(x). Hence f (x k ) = r(x k ) and n f (x) dµ(z) = r(x) dµ(x) = r(x k ) l k (x) dµ(x) = R n λ k r(x k ) = k=1 Also λ k = R n λ k f (x k ). k=1 n λ j l k (x j ) 2 = j=1 R k=1 R l k (x) 2 dµ(x) > 0.

49 Finite systems of orthogonal polynomials We saw: R f (x) dµ(x) = n λ k f (x k ) (f P 2n 1 ). k=1 In particular, for i, j n 1, h j δ i,j = p i (x)p j (x) dµ(x) = R n λ k p i (x k ) p j (x k ). k=1 Thus the finite system p 0, p 1,..., p n 1 forms a set of orthogonal polynomials on the finite set {x 1,..., x n } of the n zeros of p n with respect to the weights λ k and with quadaratic norms h j. All information about this system is already contained in the finite system of recurrence relations xp j (x) = a j p j+1 (x) + b j p j (x) + c j p j 1 (x) (j = 0, 1,..., n 1) with a j c j+1 > 0 (j = 0, 1,..., n 2). In particular, the λ k are obtained up to constant factor by solving the system n λ k p j (x k ) = 0 (j = 1,..., n 1). k=1

50 Finite systems of orthogonal polynomials (cntd.) For exampe, consider orthogonal polynomials p 0, p 1,..., p N on the zeros 0, 1,..., N of the polynomial p N+1 (x) := x(x 1)... (x N) with respect to nice explicit weights w x (x = 0, 1,..., N) like: ( ) 1 n w x := p x (1 p) N x (0 < p < 1). x Then the p n (x) are the Krawtchouk polynomials ( n, x K n (x; p, N) := 2 F 1 N ; 1 p ) = n k=0 2 w x := (α + 1) x (β + 1) N x (α, β > 1). x! (N x)! Then the p n (x) are the Hahn polynomials ( n) k ( x) k ( N) k k! ( n, n + α + β + 1, x Q n (x; α, β, N) := 3 F 2 ; 1 α + 1, N ). 1 p k.

51 Hahn and Krawtchouk polynomials (cntd.) Hahn polynomials are discrete versions of Jacobi polynomials: ( ) n, n + α + β + 1, Nx Q n (Nx; α, β, N) = 3 F 2 ; 1 α + 1, N ( ) n, n + α + β + 1 2F 1 ; x = const. P (α,β) n (1 2x) α + 1 and N 1 Q m (Nx; α, β, N)Q n (Nx; α, β, N) w Nx x {0, 1 N, 2 N,...,1} const. 1 0 P (α,β) m (1 2x)P (α,β) (1 2x) x α (1 x) β dx. Jacobi and Krawtchouk polynomials are different ways of looking at the matrix elements of the irreps of SU(2). The 3j coefficients or Clebsch-Gordan coefficients for SU(2) can be expressed as Hahn polynomials. n

52 Classical orthogonal polynomials of Hahn class Hahn and Krawtchouk polynomials are orthogonal polynomials p n (x) on 0, 1,..., N which are eigenfunctions of a second order difference operator: A(x)p n (x 1) + B(x)p n (x) + C(x)p n (x + 1) = λ n p n (x). Moreover, the polynomials q n (x) := p n+1 (x + 1) p n+1 (x) are orthogonal polynomials on 0, 1,..., N 1. If we also allow orthogonal polynomials on 0, 1, 2,... then Meixner polynomials M n (x; β, c) and Charlier polynomials C n (x; a) also have these properties. Here ( n, x M n (x; β, c) := 2 F 1 ; 1 1 β c C n (x; a) := 2 F 0 ( n, x; ; a 1 ), ), w x := (β x) c x, x! w x := a x /x!.

53 Classical orthogonal polynomials More generally we can ask for orthogonal polynomials which are eigenfunctions of a second order operator L of the form (Lf )(x) := A(x)f (x + i) + B(x)f (x) + C(x)f (x i) or (the so-called quadratic lattice) (Lf )(q(x)) := A(x)f (q(x + 1)) + B(x)f (q(x)) + C(x)f (q(x 1)), where q(x) is a fixed polynomial of second degree. All such orthogonal polynomials have been classified. There are only 13 families, all but the Hermite depending on parameters, at most four, and all expressible as hypergeometric functions, the most complicated as 4 F 3. They can be arranged hierarchically according to limit transitions denoted by arrows.

54 Askey scheme Wilson Racah Cont. Hahn Dual Hahn Hahn Jacobi Meixner Krawtchouk Cont. dual Hahn Meixner- Pollaczek Laguerre Hermite Charlier Dick Askey discrete OP quadratic lattice Hahn class very classical OP

55 The q-case On top of the Askey-scheme is lying the q-askey scheme, from which there are also arrows to the Askey scheme as q 1. We take always 0 < q < 1 and let q 1 to the classical case. Some typical examples of q-analogues of classical concepts are (see Gasper & Rahman, Basic hypergeometric series): q-number: [a] q := 1 qa 1 q a n 1 q-shifted factorial: (a; q) n := (1 aq k ) (also for n = ). (q a ; q) k (1 q) a (a) k. q-hypergeometric series: ( a1,..., a s+1 s+1φ s ; q, z b 1,..., b s ( q a 1,..., q a s+1 s+1φ s q b 1,..., q b s ; q, z k=0 ) := ) k=0 (a 1 ; q) k... (a s+1 ; q) k z k (b 1 ; q) k... (b s ; q) k ( a1,..., a s+1 s+1 F s ; z b 1,..., b s ). (q; q) k.

56 The q-case (cntd.) q-derivative: (D q f )(x) := q-integral: 1 0 f (x) d q x := (1 q) f (x) f (qx) (1 q)x f (q k ) q k k=0 f (x). 1 0 f (x) dx. The q-case allows more symmetry which may be broken when taking limits for q to 1. In the elliptic case lying above the q-case there is even more symmetry. Askey-Wilson polynomials (up to constant factor): ( q n, q n 1 abcd, ae iθ, ae iθ ) p n (cos θ; a, b, c, d q) := 4 φ 3 ; q, q. ab, ac, ad Orthogonal with respect to a weight function on ( 1, 1). A special case are the continuous q-ultraspherical polynomials (a = c = β 1 2, b = d = (qβ) 1 2 ).

57 Continuous q-ultraspherical polynomials For m n: π 0 C m (cos θ; β q) C n (cos θ; β q) (e 2iθ ; q) (βe 2iθ ; q) 2 dθ = 0. Generating function: (βe iθ t; q) (e iθ t; q) 2 = C n (x; β q)t n. n=0 Limit formula to ultraspherical polynomials: C n (x; q λ q) C λ n (x). These have generating function (1 2xt + t 2 ) λ = Cn λ (x)t n. n=0

58 SIAM Activity Group OPSF The SIAM Activity Group on Orthogonal Polynomials and Special Functions Sends out a free bimonthly electronic newsletter; Organizes minisymposia on SIAM conferences; Awards the biennial Gábor Szegö Prize to an early-career researcher (at most 10 years after PhD) for outstanding research contributions in the area of orthogonal polynomials and special functions. Nominations before September 15, See