Arxiv:2107.01747V3 [Math.CA] 14 Jul 2021 3
![Arxiv:2107.01747V3 [Math.CA] 14 Jul 2021 3](http://data.docslib.org/img/13597479/arxiv-2107-01747v3-math-ca-14-jul-2021-3.jpg)
PEARSON EQUATIONS FOR DISCRETE ORTHOGONAL POLYNOMIALS: I. GENERALIZED HYPERGEOMETRIC FUNCTIONS AND TODA EQUATIONS
MANUEL MAÑAS1, ITSASO FERNÁNDEZ-IRISARRI2, AND OMAR F. GONZÁLEZ-HERNÁNDEZ3
Abstract. The Cholesky factorization of the moment matrix is applied to discrete orthogonal polynomials on the homogeneous lattice. In particular, semiclassical discrete orthogonal polynomials, which are built in terms of a discrete Pearson equation, are studied. The Laguerre–Freud structure semi-infinite matrix that models the shifts by ±1 in the independent variable of the set of orthogonal polynomials is introduced. In the semiclassical case it is proven that this Laguerre–Freud matrix is banded. From the well known fact that moments of the semiclassical weights are logarithmic derivatives of generalized hypergeometric functions, it is shown how the contiguous relations for these hypergeometric functions translate as symmetries for the corresponding moment matrix. It is found that the 3D Nijho –Capel discrete Toda lattice describes the corresponding contiguous shifts for the squared norms of the orthogonal polynomials. The continuous Toda for these semiclassical discrete orthogonal polynomials is discussed and the compatibility equations are derived. It also shown that the Kadomtesev– Petvishvilii equation is connected to an adequate deformed semiclassical discrete weight, but in this case the deformation do not satisfy a Pearson equation.
Contents 1. Introduction 2 1.1. Linear functionals and orthogonal polynomials 2 1.2. The shift matrix Λ 3 1.3. The Cholesky factorization of the Gram matrix 3 1.4. The Jacobi matrix 4 1.5. The lower Pascal matrix 5 2. Discrete orthogonal polynomials and discrete Pearson equations 8 2.1. Discrete Pearson equation 8 2.2. Consequences for the Jacobi matrix 9 2.3. Generalized hypergeometric functions and the Pearson equation 10 2.4. The Laguerre–Freud structure matrix and the Cholesky factorization 11 2.5. Contiguous hypergeometric relations 13 2.6. Discrete lattice Toda equation 15 2.7. The Toda flows 18 2.8. Toda–Pearson compatibility 22
arXiv:2107.01747v3 [math.CA] 14 Jul 2021 3. Conclusions and outlook 23 Acknowledgments 24 References 24
1Departamento de Física Teórica, Universidad Complutense de Madrid, Plaza Ciencias 1, 28040-Madrid, Spain & Instituto de Ciencias Matematicas (ICMAT), Campus de Cantoblanco UAM, 28049-Madrid, Spain E-mail addresses: [email protected], [email protected], [email protected]. 2020 Mathematics Subject Classi cation. 42C05,33C45,33C47. Key words and phrases. Discrete orthogonal polynomials, Pearson equations, Cholesky factorization, generalized hypergeometric functions, contigous relations, 3D Nijho –Capel discrete Toda equation, Toda hierarchy. 1Thanks financial support from the Spanish “Agencia Estatal de Investigación" research project [PGC2018-096504-B-C33], Or- togonalidad y Aproximación: Teoría y Aplicaciones en Física Matemática. 1 2 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
1. Introduction In this paper, the Gauss–Borel factorization of the moment matrix approach to orthogonality is applied to study discrete orthogonal polynomials subject to a Pearson equation. Discrete orthogonal polynomials is nowadays a well established subject, the classical case has been treated extensively in [42] and the the Riemann–Hilbert problem has been used to study asymptotics and applica- tions, see [16]. Semiclassical reductions, in where a discrete Pearson equation is fulfilled by the weight has been also treated in the literature. See for example [23, 24] and [21, 22] and references therein for an comprehensive account. For some specific type of weights of generalized Charlier and Meixner types, the corresponding Freud–Laguerre type equations for the coe cients of the three term recurrence has been studied, see for example [19, 26, 27, 28, 43]. For a general account see [31, 32, 17] and for discrete orthogonal polynomials and Painlevé equations see [44]. With this introduction we give a fast briefing, of introductory character, of orthogonal polynomials, its discrete version and the Cholesky factorization of the moment matrix. We also discuss for the first time in this context the so called Pascal matrices and its dressed versions. Then, in §2 we discuss the discrete Pearson equation and give Theorem 1, the first main result of the paper, that describes a new symmetry of the moment matrix, direct consequence of the Pearson equation, in terms of the Pascal matrix. We then deduce the corresponding symmetry for the Jacobi matrix, see Proposition 6. Then, in Theorem 2, the second main result of the paper, we describe a banded semi-infinite matrix, that we called Laguerre–Freud structure matrix, that models the shift in the spectral variable. The name is rooted on the fact that this banded matrix encodes Laguerre–Freud equations for the recurrence coe cients of the orthogonal polynomial sequence, see for example [25]. The number of non-trivial superdiagonals and subdiagonals of this matrix is determined by the Pearson equation. Several properties involving the Jacobi and Laguerre–Freud matrices are derived. Of particular interest are those of compatibility type. As the Pearson equation leads to generalized hypergeometric moments, we study the contiguous relations of these functions and its description as further symmetries of the moment matrix, see Theorem 4. In Theorem 5 the squared norms of these discrete orthogonal polynomials are shown to be solutions of a well known discrete integrable multidimensional lattice, known as the Nijho –Capel lattice, see [30, 41]. We also discuss the Toda hierarchy deformations, with only the first flow preserving the Pearson reduction. The compatibility with this first Toda flow and the Pearson equation is discussed in Proposition 21. We also give the connection of deformations of these weights with solutions of the Kadomtsev–Petviashvilli (KP) equation.
∗ ∗ 1.1. Linear functionals and orthogonal polynomials. Given a linear functional ρz ∈ C [z], here C [z] denotes the dual of the linear space of complex polynomials C[z], the corresponding moment matrix is
n G = (Gn,m),Gn,m = ρn+m, ρn = ρz, z , n, m ∈ N0 := {0, 1, 2,... }, with ρn the n-th moment of the linear functional ρz. If the moment matrix is such that all its truncations, which are Hankel matrices, Gi+1,j = Gi,j+1, ρ0 ρ1 ρ2 ρk−1 ρ1 ρ2 ρk G0,0 G0,k−1 [k] G = = ρ2 G G k−1,0 k−1,k−1 ρk−1 ρk ρ2k−2 DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 3
[k] are nonsingular; i.e. the Hankel determinants ∆k := det G do not cancel, ∆k 6= 0, k ∈ N0, then there exists monic polynomials n 1 n−1 n (1) Pn(z) = z + pnz + ··· + pn, n ∈ N0, 1 with p0 = 0, such that the following orthogonality conditions are fulfilled k n ρ, Pn(z)z = 0, k ∈ {0, . . . , n − 1}, ρ, Pn(z)z = Hn 6= 0.
Moreover, the set {Pn(z)}n∈N0 is an orthogonal set of polynomials ρ, Pn(z)Pm(z) = δn,mHn, n, m ∈ N0. In this case, we have a symmetric bilinear form hF,Giρ := hρ, F Gi, such that the moment matrix is the Gram matrix of this bilinear form and hPn,Pmiρ := δn,mHn. In this paper we will use both denominations, moment matrix or Gram matrix, indistinctly to refer to the matrix G. 1.2. The shift matrix Λ. In terms of the semi-infinite vector of monomials 1 z 2 χ(z) := z the Gram matrix is written as G = ρ, χχ> , and it becomes evident this matrix is symmetric, i.e., G = G>. The vector of monomials χ is an eigenvector of the shift matrix 0 1 0 Λ := i.e., Λχ = xχ. From here it immediately follows that ΛG = GΛ>, i.e., the Gram matrix is a Hankel matrix, as we previously said. The transposed matrix 0 1 > Λ = 0
> > satisfies ΛΛ = I and Λ Λ = I − E0,0, with (Ei,j)k,l = δi,kδj,l, i, j ∈ N0. 1.3. The Cholesky factorization of the Gram matrix. Being the Gram matrix symmetric its Borel–Gauss factorization reduces to a Cholesky factorization (2) G = S−1HS−>, where S is a lower unitriangular matrix that can be written as 1 0 S1,0 1 S = , S S 2,0 2,1 and H = diag(H0,H1,... ) is a diagonal matrix, with Hk 6= 0, for k ∈ N0. The Cholesky factorization hold whenever the principal minors of the moment matrix; i.e., the Hankel determinants ∆k, do not cancel. 4 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
The components Pn(z) of the semi-infinite vector of polynomials (3) P (z) := Sχ(z), are the monic orthogonal polynomials of the functional ρ. Indeed, from the Cholesky factorization we know that D E ρ, χχ> = G = S−1HS−> so that S ρ, χχ> S> = H and, consequently ρ, Sχχ>S> = H and we get ρ, P P > = H, which ∞ recollects the orthogonality of the polynomials {Pn(z)}n=0. 1.4. The Jacobi matrix. Given this construction is natural to introduce the lower Hessenberg semi-infinite matrix (4) J = SΛS−1 that has the vector P (z) as eigenvector with eigenvalue z, i.e. JP (z) = zP (z). The Hankel condition of the Gram matrix ΛG = GΛ> together with the Cholesky factorization leads to ΛS−1HS−> = S−1HS−>Λ>, or, equivalently, SΛS−1H = HS−>Λ>S−>, i.e., (5) JH = (JH)> = HJ >. That is, the Hessenberg matrix JH is symmetric, thus being Hessenberg and symmetric we deduce that is tridiagonal. Hence, the Jacobi matrix J given in (4) reads β0 1 0 γ1 β1 1 J = 0 γ2 β2 1 and the eigenvalue equation JP = zP is a three term recursion relation
zPn(z) = Pn+1(z) + βnPn(z) + γnPn−1(z), that with the initial conditions P−1 = 0 and P0 = 1 completely determines the sequence of orthogonal polynomials {Pn(z)}n∈N0 in terms of the recursion coe cients βn, γn. Proposition 1. The recursion coe cients, in terms of the Hankel determinants, are given by ˜ ˜ 1 1 ∆n ∆n+1 Hn+1 ∆n+1∆n−1 (6) βn = pn − pn+1 = − + , γn+1 = = 2 , n ∈ N0, ∆n ∆n+1 Hn ∆n Proof. For n ∈ := {1, 2,... }, from J > = H−1JH, we get that γ = Hn . On the other hand, as N n Hn−1 −1 1 J = SΛS and recalling that for the coe cients Sn,n−1 of the first subdiagonal of S we have Sn,n−1 = pn, 1 1 the first nontrivial leading coe cient of the monic polynomial Pn, we get βn = pn − pn+1. It can be easily shown that the second kind functions also satisfy the previous three term recursion relation
zQn(z) = Qn+1(z) + βnQn(z) + γnQn−1(z), but now with the initial conditions Q−1 = −H−1 and Q0 = Sρ (the Markov–Stieltjes transform of ρ), with 1 1 βn = qn − qn−1. For future use we introduce the following diagonal matrices
γ := diag(γ1, γ2,... ), β := diag(β0, β1,... ) and > J− := Λ γ, J+ := β + Λ, DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 5 so that we have the splitting > J = Λ γ + β + Λ = J− + J+.
In general, given any semi-infinite matrix A, we will write A = A− + A+, where A− is a strictly lower triangular matrix and A+ an upper triangular matrix. Moreover, A0 will denote the diagonal part of A. 1.5. The lower Pascal matrix. The lower Pascal matrix, built up of binomial numbers, is defined by n , n ≥ m, (7) B = (Bn,m),Bn,m := m 0, n < m. so that (8) χ(z + 1) = Bχ(z). Moreover, n+m n −1 (−1) , n ≥ m, B = (B˜n,m), B˜n,m := m 0, n < m. and χ(z − 1) = B−1χ(z). The lower Pascal matrix and its inverse are explicitly given by 1 0 1 0 1 1 0 −1 1 0 1 2 1 0 1 −2 1 0 1 3 3 1 0 −1 −1 3 −3 1 0 B = ,B = . 1 4 6 4 1 0 1 −4 6 −4 1 0 1 5 10 10 5 1 0 −1 5 −10 10 −5 1 0
Let us introduce the lower unitriangular semi-infinite matrices, lets us refer to them as dressed Pascal matrices, Π := SBS−1, Π−1 := SB−1S−1, that are connection matrices; i.e., (9) P (z + 1) = ΠP (z),P (z − 1) = Π−1P (z). The lower Pascal matrix can be expressed in terms of its subdiagonal structure as follows B±1 = I ± Λ>D + Λ>2D[2] ± Λ>3D[3] + ··· , [k] [1] where the diagonal matrices D,D , with k ∈ N,(D = D ) are given by 1 D = diag(1, 2, 3,... ),D[k] = diag k(k), (k + 1)(k), (k + 2)(k) ··· , k in terms of the falling factorials x(k) = x(x − 1)(x − 2) ··· (x − k + 1). That is, (n + k) ··· (n + 1) D[k] = , k ∈ , n ∈ . n k N N0 The lower unitriangular factor can be also written in terms of its subdiagonals S = I + Λ>S[1] + Λ>2S[2] + ··· 6 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
[k] [k] [k] with S = diag S0 ,S1 ,... diagonal matrices. From (3) is clear the following connection between these subdiagonals coe cients and the coe cients of the orthogonal polynomials, given in (1), holds
[k] k (10) Sk = pn+k.
We will use the shift operators T± acting over the diagonal matrices as follows
T− diag(a0, a1,... ) := diag(a1, a2,... ),T+ diag(a0, a1,... ) := diag(0, a0, a1,... ), where T− is the lowering shift operator and T+ the raising shift operator over the diagonal matrices. These shift operators have the following important properties, for any diagonal matrix A = diag(A0,A1,... ) > > > > (11) ΛA = (T−A)Λ,AΛ = Λ(T+A),AΛ = Λ (T−A), Λ A = (T+A)Λ .
Remark 1. Notice that the standard notation, see [42], for the dierences of a sequence {fn}n∈N0 ,
∆fn := fn+1 − fn, n ∈ N0, ∇fn = fn − fn−1, n ∈ N, and ∇f0 = f0, connects with the shift operators by means of
T− = I + ∆,T+ = I − ∇. In terms of these shift operators we find [2] [3] 2 [2] [2] 2 (12) 2D = (T−D)D, 3D = (T−D)(T−D)D = 2(T−D )D = 2D (T−D) Proposition 2. The inverse matrix S−1 of the matrix S expands in terms of subdiagonals as follows S−1 = I + Λ>S[−1] + Λ>2S[−2] + ··· .
The subdiagonals S[−k] are explicitly given in terms of the subdiagonals of S, the rst few are S[−1] = −S[1], [−2] [2] [1] [1] S = −S + (T−S )S , [−3] [3] [2] [1] 2 [1] [2] 2 [1] [1] [1] S = −S + (T−S )S + (T−S )S − (T−S )(T−S )S , [−4] [4] [3] [1] 2 [2] [2] 2 [2] [1]) [1] 3 [1] [3] S = − S + (T−S )S + (T−S )S − (T−S )(T−S )S + (T−S )S 3 [1] [2]) [1] 3 [1] 2 [1]) [2] 3 [1] 2 [1] [1]) [1] − (T−S )(T−S )S − (T−S )(T−S )S + (T−S )(T−S )(T−S )S . Proposition 3. The following nonlocal expressions for the polynomial coe cients in terms of the recursion coe - cients hold true n n 1 X 2 X X (13) pn+1 = − βk, pn+1 = − γk + βkβl. k=0 k=1 0≤l
Thus, we obtain [1] [1] (15) β = T+S − S , [2] [2] [1] [1] [1] (16) γ = T+S − S + (T−S − S )S , and an infinite set of relations among subdiagonals of S, being the first [3] [−3] [2] [−1] [1] [−2] (17) T+S + S + S S + (T+S )S = 0. The relation (15) component wise is(6). As well, component wise, the relation (16) is [2] [2] [1] [1] [1] 2 2 1 γn+1 = Sn−1 − Sn + (Sn+1 − Sn )Sn = pn+1 − pn+2 − βn+1pn+1. Hence, using telescoping series again, we find (13). Finally, Equation (17) reads
[3] [3] [2] [1] 2 [1] [2] 2 [1] [1] [1] [2] [1] T+S − S + (T−S )S + (T−S )S − (T−S )(T−S )S − S S [1] [2] [1] [1] + (T+S ) − S + (T−S )S = 0, that we can write [3] [3] [2] [2] [1] [1] [1] [1] 2 [1] [1] [2] T+S − S + T− S − T+S − (T−S − S )S S + (T−S − T+S )S [1] [1] [1] [1] + (T+S − T−S )(T−S )S = 0, so that [3] [3] [1] 2 [2] [1] [1] T+S − S = (T−γ)S + (T−β + T−β + β)S − (T−β + β)(T−S )S = 0, and component wise we have (14). Remark 2. In (14) we can sum up on the LHS, observe that is a telescoping series, to get a nonlocal nonlinear 3 k expression, in terms of the recursion coe cients, for pn. A similar statement holds for every pn with k = 4, 5,... . Now, we turn our attention to the dressed Pascal matrix. We also expand the dressed Pascal matrices into subdiagonals Π±1 = I + Λ>π[±1] + (Λ>)2π[±2] + ···
[±n] [±n] [±n] with π = diag(π0 , π1 ,... ). Then, for these subdiagonals we find Proposition 4 (The dressed Pascal matrix coe cients). We have (n + 2)(n + 1) π[±1] = ±(n + 1), π[±2] = ± p1 (n + 1) ∓ (n + 2)p1 n n 2 n+2 n+1 (18) (n + 2)(n + 1) = ∓ (n + 1)β ∓ p1 , 2 n+1 n+1 (19) (n + 3)(n + 2)(n + 1) (n + 2)(n + 1) (n + 3)(n + 2) π[±3] = ± + p1 − p1 n 3 2 n+3 2 n+1 2 2 1 1 1 1 ± (n + 1)pn+3 ∓ (n + 3)pn+2 ± (n + 3)pn+2pn+1 ∓ (n + 2)pn+3pn+1. Moreover, the following relations are ful ll [1] [−1] [2] [−2] [2] [3] [−3] 2 [1] [2] [2] [1] (20) π + π = 0, π + π = 2D , π + π = 2((T−S )D − (T−D )S ), [1] [−1] [2] [−2] [1] [1] π − π = 2D, π − π = 2((T−S )D − (T−D)S ), [3] [−3] [3] [2] 2 [2] 2 [1] [1] 2 [1] [1] π − π = 2 D + (T−S )D − (T−D)S + (T−D)(T−S )S − (T−S )(T−D)S . 8 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
Proof. From (11) we get 2 2 3 2 Π±1 = (I + Λ>S[1] + Λ> S[2] + ··· ) I ± Λ>D + Λ> D[2] ± Λ> D[3] + ··· (I + Λ>S[−1] + Λ> S[−2] + ··· ) 2 2 3 2 = I + (I + Λ>S[1] + Λ> S[2] + ··· ) ± Λ>D + Λ> D[2] ± Λ> D[3] + ··· (I + Λ>S[−1] + Λ> S[−2] + ··· ) so that ±1 > >2 [2] [1] [−1] > 3 [3] 2 [1] [2] [2] [−1] (21) Π = I ± Λ D + Λ D ± (T−S )D ± (T−D)S + (Λ ) ± D + (T−S )D + (T−D )S [2] 2 [−2] 2 [1] [−1] ± (T−S )D ± (T−D)S ± (T−S )(T−D)S + ··· . From (21) and Proposition 2 we obtain (22) π[±1] = ± D, [±2] [2] [1] [1] (23) π =D ± (T−S )D ∓ (T−D)S , [±3] [3] 2 [1] [2] [2] [1] [2] 2 [2] (24) π = ± D + (T−S )D − (T−D )S ± (T−S )D ∓ (T−D)S 2 [1] [1] 2 [1] [1] ± (T−D)(T−S )S ∓ (T−S )(T−D)S . That component wise gives the desired result once we use the expressions for the β’s in (6). Proposition 5. For any polynomial R(z) we have R(Λ)B±1 = B±1R(Λ ± I),B±1R(Λ) = R(Λ ∓ I)B±1, (25) R(J)Π±1 = Π±1R(J ± I), Π±1R(J) = R(J ∓ I)Π±1. Proof. The compatibility condition of ( B±1χ(z) = χ(z ± 1), R(Λ)χ(z) = R(z)χ(z). reads, R(Λ)B±1χ(z) = R(Λ)χ(z ± 1) = R(z ± 1)χ(z ± 1) = R(z ± 1)B±1χ(z) = B±1R(Λ ± I)χ(z), so that R(Λ)B±1 = B±1R(Λ ± I) and, consequently, B±1R(Λ) = R(Λ ∓ I)B±1. Finally, a similarity −1 transformation Λ = S JS gives the result.
2. Discrete orthogonal polynomials and discrete Pearson equations 2.1. Discrete Pearson equation. We now assume that the functional is a sum of Dirac delta functions supported N0, ∞ X ρ = δ(z − k)w(k). k=0
for some weight function w(z) with finite values w(k) for k ∈ N0. Hence, the bilinear form is hF,Gi = P∞ k=0 F (k)G(k)w(k). The moments are ∞ X n (26) ρn = k w(k), k=0 and, in particular, the 0-th moment reads as follows ∞ X (27) ρ0 = w(k). k=0 DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 9
We will study families of weights are those satisfying the following discrete Pearson equation (28) ∇(σw) = τw, that is σ(k)w(k) − σ(k − 1)w(k − 1) = τ(k)w(k), for k ∈ {1, 2,... }, with σ(z), τ(z) ∈ R[z] polynomials. If we write θ := σ − τ, the previous Pearson equation reads
(29) θ(k + 1)w(k + 1) = σ(k)w(k), k ∈ N0. Theorem 1 (Hypergeometric symmetry of the moment matrix). Let the weight w be subject to the discrete Pearson equation (29), where θ, σ are polynomials with θ(0) = 0. Then, the corresponding moment matrix ful lls (30) θ(Λ)G = Bσ(Λ)GB>. Proof. The moment matrix is ∞ X (31) G = χ(k)χ(k)>w(k). k=0 Thus, ∞ X θ(Λ)G = θ(Λ)χ(k)χ(k)>w(k) use (31) k=0 ∞ X = χ(k)χ(k)>θ(k)w(k) use Λχ = zχ and θ(0) = 0 k=1 ∞ X = χ(k + 1)χ(k + 1)>θ(k + 1)w(k + 1) shift the summation variable k=0 ∞ X = χ(k + 1)χ(k + 1)>σ(k)w(k) use Pearson equation (29) k=0 ∞ X = Bχ(k)χ(k)>B>σ(k)w(k) use (8) k=0 ∞ X = Bσ(Λ)χ(k)χ(k)>w(k)B> use Λχ = zχ again k=0 = Bσ(Λ)GB> use (31).
Remark 3. This result extends to the case when θ and σ are entire functions, not necessarily polynomials, and we can ensure some meaning for θ(Λ) and σ(Λ). Later on, see §2.5, we will show that this symmetry of the Gram matrix is a direct consequence of the generalized hypergeometric ODE satis ed by the the 0-th moment. 2.2. Consequences for the Jacobi matrix. We can use the Cholesky factorization (2) and the Jacobi matrix (4) to get Proposition 6 (Symmetry of the Jacobi matrix). Let the weight w be subject to the discrete Pearson equation (29), where the functions θ, σ are entire functions, not necessarily polynomials, with θ(0) = 0. Then, (32) Π−1Hθ(J >) = σ(J)HΠ>
Moreover, Hθ(J >) and σ(J)H are symmetric matrices. Proof. Given the Hankel property, ΛG = GΛ>, we write (30) as Gθ(Λ>) = Bσ(Λ)GB>, and using the Cholesky factorization (2) we get S−1HS−>θ(Λ>) = Bσ(Λ)S−1HS−>B>, so that HS−>θ(Λ>)S> = SBS−1Sσ(Λ)S−1HS−>B>S>, and we get Equation (32). 10 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
Let us prove that the matrices Hθ(J >) and σ(J)H are symmetric. This fact follows from (5), JH = HJ >. Indeed, (Hθ(J >))> = θ(J)H = θ(HJ >H−1)H = Hθ(J >)H−1H = Hθ(J >), (σ(J)H)> = Hσ(J >) = Hσ(H−1JH) = HH−1σ(J)H = σ(J)H.
2.3. Generalized hypergeometric functions and the Pearson equation. Let us assume that θ, σ are polynomials, and denote their respective degrees by N + 1 := deg θ(z) and M := deg σ(z). The roots of N M these polynomials are denoted by {−bi + 1}i=1 and {−ai}i=1. Following [23] we choose
θ(z) = z(z + b1 − 1) ··· (z + bN − 1), σ(z) = η(z + a1) ··· (z + aM ). Notice that we have normalized θ to be a monic polynomial, while σ is not monic, being the coe cient of the leading power denoted by η. This parameter η will be instrumental in what follows. Therefore, see [23], we have that the weight satisfying the Pearson equation (29) is proportional to
(a1)z ··· (aM )z (33) w(z) = ηz, Γ(z + 1)(b1)z ··· (bN )z Γ(α + z) where the Pochhammer symbol is (α) = , that for z ∈ has the standard expression z Γ(α) N
(α)n = α(α + 1)(α + 2) ··· (α + n − 1), (α)0 = 1. The moments of this weight are finite whenever, see [23] and references therein, i) M ≤ N and η ∈ C. ii) M > N, η ∈ C and one or more of the parameters ai is a nonpositive integer. iii) M = N + 1 and |η| < 1. iv) M = N + 1, |η| = 1 and Re(b1 + ··· + bN−1 − a1 − · · · − aM ) > 0. According to (27) the 0-th moment
∞ ∞ k X X (a1)k ··· (aM )k η ρ0 = w(k) = (b1 + 1)k ··· (bN + 1)k k! k=0 k=0 a1 ··· aM = MNF (a1, . . . , aM ; b1, . . . , bN ; η) = M FN ; η . b1 ··· bN is the generalized hypergeometric function, where we are using the two standard notations, see [15]. Then, P∞ n according to (26), for n ∈ N, the corresponding higher moments ρn = k=0 k w(k), are n n a1 ··· aM ∂ (34) ρn = ϑη ρ0 = ϑη M FN ; η , ϑη := η . b1 ··· bN ∂η Given a function f(η), we consider the Wronskian 2 k f ϑηf ϑηf ϑηf 2 k+1 ϑηf ϑηf ϑη f 2 Wn(f) = det ϑηf . k k+1 2k ϑz f ϑη f ϑη f DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 11
[k] Then, we have that the Hankel determinants ∆k = det G determined by the truncations of the corre- sponding Gram matrix are Wronskians of generalized hypergeometric functions a1 ··· aM (35) ∆k = τk, τk := Wk M FN ; η , b1 ··· bN
(36) ∆˜k = ϑητk. We also have τk+1 1 (37) Hk = , pk = −ϑη log τk. τk
Remark 4. The functions τk are knwon in the theory of integrable systems as τ-functions. 2.4. The Laguerre–Freud structure matrix and the Cholesky factorization. Theorem 2 (Laguerre–Freud structure matrix). Let us assume that the weight w is subject to the discrete Pearson equation (29) with θ, σ polynomials such that θ(0) = 0, deg θ(z) = N + 1, deg σ(z) = M. Then, the Laguerre– Freud structure matrix (38) Ψ := Π−1Hθ(J >) = σ(J)HΠ> = Π−1θ(J)H = Hσ(J >)Π> (39) = θ(J + I)Π−1H = HΠ>σ(J > − I), has only N + M + 2 possibly nonzero diagonals (N + 1 superdiagonals, the main diagonal and M subdiagonals) Ψ = (Λ>)M ψ(−M) + ··· + Λ>ψ(−1) + ψ(0) + ψ(1)Λ + ··· + ψ(N+1)ΛN+1, for some diagonal matrices ψ(k). In particular, the lowest subdiagonal and highest superdiagonal are given by M−1 M M+1 > M (−M) M (−M) Y k Y Y (Λ ) ψ = η(J−) H, ψ = ηH T γ = η diag H0 γk,H1 γk,... , − k=0 k=1 k=2 (40) N N+1 N+2 (N+1) N+1 > N+1 (N+1) Y k Y Y ψ Λ = H(J ) , ψ = H T γ = diag H0 γk,H1 γk,... . − − k=0 k=1 k=2 The vector P (z) of orthogonal polynomials ful ll the following structure equations (41) θ(z)P (z − 1) = ΨH−1P (z), σ(z)P (z + 1) = Ψ>H−1P (z). Proof. To get (39) just use (25). Now, notice that the matrices Hθ(J >) and σ(J)H are banded matrices with 2N + 3 diagonals (N + 1 superdiagonals and subdiagonals) and 2M + 1 (M superdiagonals and subdiagonals), respectively. Thus, Π−1Hθ(J >) has at most N + 1 superdiagonals while σ(J)HΠ> has at most M subdiagonals. Consequently, Ψ given by (38) is a banded semi-infinite matrix as described. Then, (40) follow from (38) and the mentioned banded structure. To get the lowest subdiagonal we use Ψ = σ(J)HΠ>, so that the lowest subdiagonal will come from the lowest subdiagonal of σ(J), namely M > η (J−) right multiplied by the diagonal matrix H and then right multiplied the main diagonal of Π , which happens to be the identity. To get the highest superdiagonal we proceed similarly and use Ψ = Π−1Hθ(J >), > N+1 so that the main superdiagonal will come from the product of the three factors I, H and θN (J− ) , in this order. The component wise expressions are direct computations. Finally, we have ΨH−1P (z) = Π−1θ(J)HH−1P (z) Ψ>H−1P (z) = Πσ(J)HH−1P (z) = Π−1θ(J)P (z) = Πσ(J)P (z) = Π−1θ(z)P (z) = Πσ(z)P (z) = θ(z)P (z − 1), = σ(z)P (z + 1), 12 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
Remark 5 (Laguerre–Freud equations). As we will see in the following sections (38) will be instrumental in obtaining non linear recurrences for the recursion coe cients
γn+1 = F1(n, γn, γn−1, . . . , βn, βn−1 ... ), βn+1 = F2(n, γn+1, γn, . . . , βn, βn−1,... ), for some functions F1,F2. These recurrences were named by Alphonse Magnus, attending to [33, 29], as Laguerre–Freud see [34, 35, 36, 37]. This is the reason for the given name to Ψ. Proposition 7. The Laguerre–Freud and Jacobi matrices ful ll σ(J)θ(J + I) = ΨH−1Ψ>H−1, θ(J)σ(J − I) = Ψ>H−1ΨH−1. Proof. We have σ(J)θ(J + I)P (z) = σ(z)θ(z + 1)P (z) = σ(z)ΨH−1P (z + 1) = ΨH−1σ(z)P (z + 1) = ΨH−1Ψ>H−1P (z), θ(J)σ(J − I)P (z) = θ(z)σ(z − 1)P (z) = θ(z)Ψ>H−1P (z − 1) = Ψ>H−1θ(z)P (z − 1) = Ψ>H−1ΨH−1P (z), that hold for all z ∈ C and, consequently, imply the desired result. Theorem 3 (Cholesky factorization). Let us assume that Hθ(J >) and σ(J)H have the following Cholesky factorizations > −1 −> −1 −> (42) Hθ(J ) = Θ HθΘ , σ(J)H = Σ HσΣ , with Θ and Σ lower unitriangular matrices and Hθ and Hσ diagonals matrices. Then, i) Θ−1 and Σ−1 have only rst N + 1 and M subdiagonals possibly nonzero, respectively. ii) We have −1 Π = Θ Σ,Hθ = Hσ =: h. iii) The Laguerre–Freud matrix has the following Gauss–Borel factorization Ψ = Σ−1hΘ−>. Proof. Recall that according to Proposition 6 the matrices Hθ(J >) and σ(J)H are symmetric and, conse- quently, the corresponding Gauss–Borel factorizations, when they exists, will be Cholesky factorizations. i) If follows from the fact that J is tridiagonal and θ and σ polynomials of degrees N + 1 and M, respectively. ii) The symmetry (32) yields −1 −1 −> −1 −> > Π Θ HθΘ = Σ HσΣ Π and given the uniqueness of the Gauss–Borel factorization the result follows. iii) From (38) we have di erent alternatives to show the result, let us see two of them −1 > −1 −1 −> −1 −> > −1 −> > −1 −1 −> Ψ = Π Hθ(J ) = Σ ΘΘ HθΘ = Σ hΘ , Ψ = σ(J)HΠ = Σ HσΣ Σ Θ = Σ hΘ . The compatibility with the recursion relation, i.e. eigenfunctions of the Jacobi matrix, and the recursion matrix leads to some interesting equations. Proposition 8. The following compatibility conditions for the Laguerre–Freud and Jacobi matrices hold (43a) [ΨH−1,J] = ΨH−1, (43b) [J, Ψ>H−1] = Ψ>H−1. DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 13
Proof. To prove (43a), we evaluate the eigenvalue equation JP (z) = zP (z) in z − 1 to get JP (z − 1) = (z −1)P (z −1). Now multiply it by θ(z) to obtain Jθ(z)P (z −1) = (z −1)θ(z)P (z −1). Therefore, recalling (41), JΨH−1P (z) = (z − 1)ΨH−1P (z) = ΨH−1(J − I)P (z) from where the relation follows. Alternatively, observe that (44) ΨH−1 = SB−1θ(Λ)S−1 from where [ΨH−1,J] = S[B−1θ(Λ), Λ]S−1 = S[B−1, Λ]θ(Λ)S−1 = SB−1θ(Λ)S−1 = ΨH−1. To show (43b) we take the transpose of the proved one to get [J >,H−1Ψ>] = H−1Ψ> so that [HJ >H−1,HH−1Ψ>H−1] = HH−1Ψ>H−1 > −1 and recalling that HJ H = J, see (5), we get the desired result. 2.5. Contiguous hypergeometric relations. As we have seen, the polynomial discrete Pearson equation leads to (34), and the Hankel determinants are Wronskians of a generalized hypergeometric function, see (35) and (36). Hence, we expect that some properties of generalized hypergeometric functions may translate to the Gram matrix. To write this dictionary we require of Proposition 11, discussed latter, that ensures, > ϑηG = ΛG = GΛ . We now study three important relations fulfilled by the generalized hypergeometric function, namely: a1···ai ···aM a1···ai + 1 ···aM (45) (ϑη + ai) M FN ; η = ai M FN ; η , b1···bN b1···bN a1···aM a1···aM (46) (ϑη + bj − 1) M FN ; η = (bj − 1) M FN ; η , b1···bj ···bN b1···bj − 1 ···bN QM d a1 ··· aM a1 + 1 ··· aM + 1 i=1 ai (47) M FN ; η = κ M FN ; η , κ := . dη b1 ··· bN b1 + 1 ··· bN + 1 QN j=1 bj From these three equations we also derive M N Y d d Y d a1 ··· aM (48) η η + an u = η η + bn − 1 u, u := M FN ; η . dη dη dη b1 ··· bN n=1 n=1 In (45) and (46) we have basic relations between contiguous generalized hypergeometric functions and its derivatives. M For the analysis of these equations let us introduce the shift operators in the parameters {ai}i=1 and {b }N f a1 ··· aM T T j j=1. Thus, given a function b1 ··· bN of these parameters we introduce the shifts i and j as follows a1···ai ···aM a1···ai + 1 ···aM a1···aM a1···aM iT f = f ,Tjf = f b1···bN b1···bN b1···bj ···bN b1···bj − 1 ···bN −1 −1 and a total shift T = 1T ··· M TT1 ··· TN ; i.e, a ··· a a + 1 ··· a + 1 T f 1 M := f 1 M . b1 ··· bN b1 + 1 ··· bN + 1 Then, we find:
Theorem 4 (Hypergeometric relations). The moment matrix G = (ρn+m)n,n∈N0 of a weight satisfying (29) ful lls the following hypergeometric relations
(49a) (Λ + aiI)G = ai iT G,
(49b) (Λ + (bj − 1)I)G = (bj − 1)TjG, (49c) ΛG = κB(TG)B> 14 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ with B given in (8). G = ϑn+m F a1 ··· aM ; η Proof. We first prove (49a). For that aim we use that n,m η M N b1 ··· bN so that n+m a1 ··· aM (ϑη + ai)Gn,m = (ϑη + ai)ϑη M FN ; η b1 ··· bN n+m a1 ··· aM = ϑη (ϑη + ai) M FN ; η b1 ··· bN n+m a1···ai + 1 ···aM = aiϑ M FN ; η η b1···bN
= ai iTGn,m.
Thus, (ϑη + ai)G = ai iTG, and recalling ϑηG = ΛG we get the result. Relation (49b) is proved similarly. d −1 Finally, (49b) follows from (47), the novelty is that in (47) we have d η = η ϑη, which do not commute with ϑη. Observe that the relation d d η − η = 1 d η d η
−1 −1 n −1 n n can be written η ϑηη = ϑη + 1. Hence, η ϑη η = (η ϑηη) = (ϑη + 1) , that, in turn, implies d d (50) ϑn = (ϑ + 1)n , d η η η d η Thus, d Gn,m d n+m a1 ··· aM n+m d a1 ··· aM = ϑη M FN ; η = (ϑη + 1) M FN ; η d η d η b1 ··· bN d η b1 ··· bN n+m a1 + 1 ··· aM + 1 = (ϑη + 1) κ M FN ; η b1 + 1 ··· bN + 1
n+m X n + m k a1 + 1 ··· aM + 1 = κ ϑηM FN ; η . k b1 + 1 ··· bN + 1 k=0 Using the Chu–Vandermonde identity
k n + m X n m = , k l k − l l=0 we find n+m k d Gn,m X X n m l+k−l a1 + 1 ··· aM + 1 = κ ϑη M FN ; η d η l k − l b1 + 1 ··· bN + 1 k=0 l=0 n+m k X X n m = κ TG l k − l l,k−l k=0 l=0 n m X X n m = κ TG l l,k k l=0 k=0 DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 15 from where the result follows. −1 n n Finally, using ϑη − 1 = ηϑηη , we get (ϑη − 1) η = η(ϑη) and we workout (48) as follows M M Y Y n+m a1 ··· aM η (ϑη + an) Gn,m = η (ϑη + an) ϑη M FN ; η b1 ··· bN n=1 n=1 M n+m Y a1 ··· aM = (ϑη − 1) η (ϑη + an) M FN ; η b1 ··· bN n=1 N Y n+m a1 ··· aM = ϑη (ϑη + bn − 1) (ϑη − 1) M FN ; η b1 ··· bN n=1 N n+m Y X n + m k k a1 ··· aM = ϑη (ϑη + bn − 1) (−1) ϑη M FN ; η k b1 ··· bN n=1 k=0 N n m Y X X n m = ϑ (ϑ + b − 1) (−1)l G (−1)k , η η n l l,k k n=1 l=0 k=0 and, consequently, we finally deduce σ(Λ)G = B−1θ(Λ)GB−>, and result follows at once. Remark 6. From (48) we derive, in an alternative manner, the relation (30). 2.6. Discrete lattice Toda equation. The shifts in the hypergeometric parameters induce corresponding transformations on the discrete orthogonal polynomials. In order to describe them we introduce the follow- ing semi-infinite matrices −1 −1 iΩ := S(i T S) , i ∈ {1,...,M}, Ωk := S (TkS) , k ∈ {1,...,N}, so that the following connection formulas are fulfilled
(51) iΩ i T P (z) = P (z), i ∈ {1,...,M}, (52) Ωj TjP (z) = P (z), j ∈ {1,...,N}. In [39] we proved that 1 0 1 0 1 H1 1 1 H1 1 a TH bj −1 Tj H0 Ω = i i 0 , Ω = , i 1 H j 1 H 0 2 0 2 ai iTH1 bj −1 Tj H1
The connection formulas (51) and (52) for contiguous hypergeometric parameters imply a nonlinear com- patibility equations that leads to a generalized lattice Toda equation found in [41]. We use the following notation u = H a1 ··· aM ; η n n−1 b1 ··· bN We now consider three variables (n, r, s), where r, s are any couple of variables taken from the set of hyper- geometric parameters {a1, . . . , aM , b1, . . . , bN } and denote the corresponding “shifts” in n, r, s as follows
u¯n(r, s) = un+1(r, s), uˆn(r, s) =a ˆuˆn(r ± 1, s), u˜n(r, s) =au ˜ (r, s ± 1), M M where the ± signs is a + if the corresponding variable belongs to {ai}i=1 or a − if it belongs to {bj}j=1, and N N the constants a,ˆ a˜ are taken from {ai}i=1 and {bj − 1}j=1 according to the choice selected for the variables (r) (s) M N r and s. We denote by Ω and Ω , the corresponding matrices Ω’s taken from {iΩ}i=1 and {Ωj}j=1, depending on the variables picked r and s. 16 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
Theorem 5 (Nijho –Capel discrete lattice Toda equations). The squared norms satisfy the following nonlinear equations linking contiguous hypergeometric parameters u¯ˆ − u¯˜ 1 1 (53) = uˆ˜ − . u¯ u˜ uˆ Proof. The connection formulas are Ω(r)Pˆ = P, Ω(s)P˜ = P. Then, we have ˜ Ω˜ (r)Pˆ = P,˜ so that ˜ Ω(s)Ω˜ (r)Pˆ = Ω(s)P˜ = P, ˜ ˆ and the compatibility Pˆ = P˜ leads to the nonlinear condition (54) Ω(s)Ω˜ (r) = Ω(r)Ωˆ (s). Then, as we can write 1 0 1 0
u¯1 u¯1 uˆ 1 u˜ 1 Ω(r) = 1 , Ω(s) = 1 , u¯ u¯ 0 2 0 2 uˆ2 u˜2
Equation (54) is 1 0 1 0 1 0 1 0 u¯˜ u¯ˆ u¯1 1 1 u¯1 1 1 u˜ 1 uˆ˜ uˆ 1 u˜ˆ 1 1 = 1 1 , u¯2 u¯˜2 u¯2 u¯ˆ2 0 u˜ 0 ˜ 0 uˆ 0 ˆ 2 uˆ2 2 u˜2 and, consequently, we deduce u¯ u¯˜ u¯ u¯ˆ + = + , u˜ uˆ˜ uˆ u˜ˆ and the result follows immediately. Remark 7. Equation (53) is the generalized discrete lattice Toda equation described for the rst time by Nijho and Capel [41], that taking adequate continuous limits in the r and s variables (hypergeometric parameters) recovers the 2D Toda equation. This is a canonical equation among the dierence equations in three variables involving up to second order dierences of octahedral type, consistent on the 4D lattice [1], it appears as the type V of the octahedron type integrable discrete equations in §3.9 of the book [30] . Now we study compatibility of the recursion relation and the connection formula for contiguous param- eters; i.e., the compatibility of zP = JP, ω(r) Pˆ =a ˆ P. z + r DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 17
These pair of relations lead to ω(r) ω(r) ω(r) JˆPˆ(z) = zPˆ(z) ⇒ Jˆ P (z) = z P (z) = JP (z), z + r z + r z + r and therefore we find the following compatibility equation Jωˆ (r) = ω(r)J. Using the notation
un = Hn−1, vn = βn−1 so that γ = u¯n and we have the following expressions n un v 1 0 1 uˆ1 1 0 u¯1 u1 u v2 1 0 1 (r) J = , ω = 0 uˆ2 1 , 0 u2
the compatibility reads vˆ1 1 0 v 1 0 uˆ1 1 0 uˆ1 1 0 1 u¯ˆ1 u u u¯1 vˆ2 1 0 1 1 v2 1 0 uˆ1 u1 0 uˆ2 1 = 0 uˆ2 1 0 u2 u2 0
and we find vˆ uˆ1 = v uˆ1 + u¯1 and 1 u1 1 u1 u1
u¯ˆn u¯ˆn u¯n u¯ˆn + v¯ˆn = +v ¯n , uˆn u¯n u¯n u¯n u¯ˆn uˆn vˆn + =v ¯n + u¯n un for n ∈ N. Thus, we find 1 1 Proposition 9. The squared norms un = Hn−1 and the recursion coe cients vn = βn−1 = pn−1 − pn, satisfy the following system of nonlinear dierence equations u¯ u¯ v¯ˆ − v¯ = − , u¯ˆ uˆ uˆ u¯ˆ vˆ − v¯ = − . u u¯
Finally, recalling ϑηP = ΦP and ΩPˆ = P , we get
ϑηΩ = ΦΩ − ΩΦˆ, so that
Proposition 10. The squared norms un = Hn−1 ful ll u¯ u¯ u¯ˆ ϑ = − . η uˆ u uˆ 18 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
∞ ∞ 2.7. The Toda ows. Given a semi-infinite vector η := {ηl}l=1 ∈ C , for each z ∈ C we define E (z; η) := Q∞ zl l=1 ηl and assume we assume a weight of the form w(z) = v(z)E (z; η), with v η-independent, so that the corresponding moment matrix is ∞ X G = χ(k)χ(k)>v(k)E (k; η). k=0 P∞ t zl Observe that only this first flow preserves the Pearson reduction. Notice also that E = e l=1 l , with tl := log ηl the standard time flows in an integrable hierarchy. We will write η1 = η, normally t1 is denoted by x, t1 = x. We will use ∂ ∂ ϑl := ηl = . ∂ηl ∂tl In particular ϑ1 = ϑη. To ensure the converge of the corresponding series for the moments we require |ηk| ≤ 1 for k ∈ {2, 3,... }. If we take ηk = 1, k ∈ N, then the corresponding Gram matrix is ∞ X > G0 = χ(k)χ(k) v(k). k=0 Moreover, we can write for the deformed Gram matrix ∞ ∞ > Y Λl X l G = E (Λ; η)G0 = G0E (Λ ; η), E (Λ; η) := ηl = exp tlΛ . l=1 l=1 ∞ P Proposition 11. For a functional of the form ρ = δ(z − k)v(z)E (z; η), given l ∈ N, the Gram matrix k=0 satis es l > l (55) ϑlG = (Λ) G = G(Λ ) . l Proof. It follows form ϑlE (z; η) = z E (z; η), indeed ∞ X > l l > l ϑlG = χ(k)χ(k) k v(k)E (k; η) = Λ G = G(Λ ) . k=0 For k ∈ N, let us define the strictly lower triangular matrix (strictly because it has zeros on the main diagonal) −1 Φk := (ϑkS)S . −1 In particular, we denote Φ := Φ1 = (ϑηS)S . Proposition 12. The semi-in nite vector P ful lls
(56) ϑkP = ΦkP. −1 Proof. As P = Sχ then ϑkP = (ϑkS)χ = (ϑkS)S Sχ = ΦP . Proposition 13 (Sato–Wilson equations). The following equations holds > k (57) −ΦkH + ϑkH − HΦk = J H, k ∈ N. Consequently, for k ∈ N and n ∈ N0 we have k k Φk = −(J )−, ϑk log Hn = (J )n,n. DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 19
Proof. Using the Cholesky factorization (2) for the moment matrix we get −1 −> −1 −1 −> −1 −> −1 −> > −> ϑkG = ϑk S HS = −S (ϑkS)S HS + S (ϑkH)S − S HS (ϑkS) S Hence, from the symmetry relation (55) we deduce −1 −1 −> −1 −> −1 −> > −> k −1 −> −S (ϑηS)S HS + S (ϑηH)S − S HS (ϑηS) S = Λ S HS , from where −1 −1> k −(ϑkS)S H + ϑkH − H (ϑkS)S = J H, −1 for J = SΛS , and the result follows.
For k = 1; i.e. for the first flow η1 = η, we have formulas (35), (36) and (37) in terms of the τ-function. Moreover, Proposition 14 (Toda). The following equations hold −1 > (58a) Φ = (ϑηS)S = −Λ γ, −1 (58b) (ϑηH)H = β. In particular, for n, k − 1 ∈ N, we have 1 k k−1 (59a) ϑηpn = −γn, ϑηpn+k = −γn+kpn+k−1, (59b) ϑη log Hn = βn.
The functions qn := log Hn, n ∈ N, satisfy the Toda equations 2 qn+1−qn qn−qn−1 (60) ϑηqn = e − e . For n ∈ N, we also have
ϑηPn(z) = −γnPn−1(z).
> Proof. Equation (58a) is a direct consequence of (57) for k = 1. To obtain (59a) we we write ϑηS = −Λ γS in terms of its subdiagonals, i.e. > [1] > 2 [2] > > [1] > 2 [2] ϑη I + Λ S + (Λ ) S + ··· = −Λ γ I + Λ S + (Λ ) S + ··· that, for k ∈ N, gives [k] k−1 [k−1] ϑηS = −(T− γ)S with S[0] = I. Now, recalling (10) we get (59a). Equation (59b) follows component wise from (58b). As β = p1 − p1 and γ = Hn we deduce that n n n+1 n Hn−1
2 1 1 Hn Hn+1 ϑη log Hn = ϑηpn − ϑηpn+1 = − + , Hn−1 Hn
qn > that is the Toda equation (60) for Hn = e . The last equation follows form ϑηP = ΦP = −Λ γP . Given the τ-function (35) we find Proposition 15 (τ-function expressions). In terms of the τ-function a1 ··· aM τn := Wn M FN ; η , b1 ··· bN we nd τn+1 1 2 τn+1 (61) Hn = , pn = −ϑη log τn, γn = ϑη log τn, βn = ϑη log . τn τn 20 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
Proof. Collect together (37) and (59a). Proposition 16. The following Lax equation holds
ϑηJ = [J+,J]. The recursion coe cients satisfy the following Toda system
(62a) ϑηβn = γn+1 − γn,
(62b) ϑη log γn = βn − βn−1, for n ∈ N0 and β−1 = 0. Consequently, we get 2 (63) ϑη log γn + 2γn = γn+1 + γn−1. Moreover, 2 τn+1τn−1 ϑη log τn = 2 . τn Proof. As J = SΛS−1 is clear that,
ϑηJ = [Φ,J] = [−J−,J] = [J+ − J, J] = [J+,J] From this Lax equation one could derive, component-wise, the Toda system (62). Alternatively, the Toda system (62) could be derived as follows. From (6) and (59a) we obtain 1 1 ϑηβn = ϑηpn − ϑηpn+1 = −γn + γn+1, ϑη log γn = ϑη log Hn − ϑη log Hn−1 = βn − βn−1. Finally,
2 Hn τn+1τn−1 ϑη log τn = γn = = . Hn−1 τn For higher Toda flows, that generically are not consistent with the hypergeometric reduction, we have Proposition 17 (Lax equations). The following Lax equation holds k ϑkJ = [(J )+,J]. Proof. As J = SΛS−1 we have k k k k ϑkJ = [Φk,J] = [−(J)−,J] = [(J )+ − J ,J] = [(J )+,J] Proposition 18 (Wave functions and Zakharov–Shabat equations). The wave function satis es the linear system k ϑkΨ = (J )+Ψ. Moreover, the following Zakharov–Shabat equations hold l k l k ϑk(J )+ − ϑl(J )+ + [(J )+, (J )+] = 0. Proof. The orthogonal polynomials we have k k k k ϑkP = ΦkP = ((J )+ − J )P = ((J )+ − z )P Thus, the wave function fulfills the following linear system k ϑkΨ = ϑk(P )E + P ϑk(E ) = (J )+Ψ. DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 21
Finally, the Zakharov–Shabat equations follow as the compatibility conditions for the previous linear system. ∞ P Proposition 19. Let us assume that G0, the Gram matrix of the functional ρ = δ(z − k)v(z), admits a k=0 Cholesky factorization, and that two semi-in nite matrices Z1 and Z2 are given, such that −1 i) Z1E (Λ, η) is strictly lower triangular, ii) Z2 is upper triangular and iii) Z1(η)G0 = Z2(η). Then, we can ensure that Z1 = Z2 = 0. Proof. We have −1 −1 −1 −> Z1G0 = Z1E (Λ; η) G = Z1E (Λ; η) S HS and we get −1 −1 −> Z1E (Λ; η) S HS = Z2. Consequently, −1 −1 −1 Z1E (Λ; η) S = Z2SH . −1 −1 −1 But Z1E (Λ; η) S is strictly lower triangular while Z2SH is upper triangular. As both terms are equal the only possibility is that both are the zero matrix −1 −1 −1 Z1E (Λ; η) S = Z2SH = 0, and we get the result. Using this results one proves that
Proposition 20 (KP equation). The wave function Ψn satis es the linear equations
ϑmΨn = Pm(ϑη)Ψn with m 1 m−2 m−3 Pm(ϑη) = ϑη − mpnϑη + Un,m−3ϑη + ··· + Un,0 m−3 l 1 1 1 where {Un,j}j=0 are polynomials in ϑηpn, ϑ2pn, . . . , ϑm−1pn. In particular, the following linear equations hold 2 1 ϑ2Ψn = ϑηΨn − 2ϑη(pn)Ψn, 3 ϑ Ψ = ϑ3Ψ − 3ϑ (p1 )ϑ Ψ − ϑ2(p1) + ϑ (p1) Ψ , 3 n η n η n η n 2 η k 2 k n and, consequently, its compatibility condition, which is the KP equation 1 1 2 3 1 2 1 ϑη 4ϑ3(pn) + 6 ϑη(pn) − ϑη(pn) = ϑ2(pn), is ful lled. Remark 8. The weight w for which this KP equation appears is
(a1)z ··· (aM )z z z2 z3 w(z) = η η2 η3 , |η2|, |η3| < 1. Γ(z + 1)(b1)z ··· (bN )z and the corresponding 0-th moment is the series ∞ ∞ k k2 k3 X X (a1)k ··· (aM )k η η2 η3 ρ0 = w(k) = . (b1 + 1)k ··· (bN−1 + 1)k k! k=0 k=0
We see that only for η2 = η3 = 1 we recover a hypergeometric function. 22 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
2.8. Toda–Pearson compatibility. For the compatibility of (9) and (56), that is for the compatibility of the systems P (z + 1) = ΠP (z), ϑη(P (z)) = ΦP (z). we require
ϑη(P (z + 1)) = ϑη(ΠP (z)) = ϑη(Π)P (z) + Πϑη(P (z)) = (ϑη(Π) + ΠJΦ)P (z),
(ϑηP )(z + 1) = ΦP (z + 1) = ΦΠP (z) to be equal and, consequently, we obtain
ϑη(Π) = [Φ, Π]. In the general case the dressed Pascal matrix Π is a lower unitriangular semi-infinite matrix, that possibly has an infinite number of subdiagonals. However, for the case when the weight w(z) = v(z)ηz satisfies the Pearson equation (29), with v independent of η, that is θ(k + 1)v(k + 1)ηk+1 = σ(k)v(k)ηk for k ∈ N0; i.e., θ(k + 1)v(k + 1)η = σ(k)v(k), the situation improves as we have the banded semi-infinite matrix Ψ that models the shift in the z variable as in (41). From the previous discrete Pearson equation we see that σ(z) = ηκ(z) with κ, θ η-independent polynomials in z (64) θ(k + 1)v(k + 1) = ηκ(k)v(k). Proposition 21. Let us assume a weight w satisfying the Pearson equation (29) and, consequently, of the form (33). Then, the Laguerre–Freud structure matrix Ψ given in (38) satis es
−1 > −1 −1 > −1 (65a) ϑη(η Ψ H ) = [Φ, η Ψ H ], −1 −1 (65b) ϑη(ΨH ) = [Φ, ΨH ]. Alternatively, the above equations can be written as follows > −1 > −1 (66a) ϑη(Ψ H ) = [J+, Ψ H ], −1 −1 −1 −1 (66b) ϑη(η ΨH ) = [J+, η ΨH ]. Relations (65a) and (65b) are gauge equivalent. Proof. We look at the compatibility of (41) and (56); i. e. for (65a) we consider σ(z)P (z + 1) = Ψ>H−1P (z), (67) ϑη(P (z)) = ΦP (z), while for (65b) we consider θ(z)P (z − 1) = ΨH−1P (z), (68) ϑη(P (z)) = ΦP (z).
From (67), observing that ϑη(σ) = σ, we deduce > −1 > −1 > −1 ϑη σ(z)P (z + 1) = ϑη(Ψ H P (z)) = ϑη(Ψ H )P (z) + Ψ H ϑη(P (z)) > −1 > −1 = ϑη(Ψ H ) + Ψ H Φ P (z), > −1 ϑη σ(z)P (z + 1) = σ(z)P (z + 1) + σ(z)ΦP (z + 1) = (Φ + I)Ψ H P (z). DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 23 so that > −1 > −1 > −1 ϑη(Ψ H ) − Ψ H = Φ, Ψ H , −1 −1 and recalling ϑη(η ) = −η we obtain (65a). From (68) we find (ϑηθ = 0) −1 −1 −1 ϑη θ(z)P (z − 1) = ϑη(ΨH P (z)) = ϑη(ΨH )P (z) + ΨH ϑη(P (z)) −1 −1 = ϑη(ΨH ) + ΨH Φ P (z), −1 ϑη θ(z)P (z − 1) = θ(z)ΦP (z − 1) = ΦΨH P (z). and (65b) follows immediately. Notice that from (44) this equation (65b) follows immediately: −1 −1 −1 −1 −1 ϑη(ΨH ) = [(ϑηS)S ,SB θ(Λ)S ] = [Φ, ΨH ]. This comment applies similarly to (65a). Using Φ = −J− = J+ − J in (65a) and (43a) we easily get (66a), and similarly from(65b) and (43b) we deduce (66b). −1 −1 Finally, we study the relation between (65a) and (65b). Assume that (65b), ϑη(ΨH ) = [Φ, ΨH ], −1 > > −1 > holds and take its transpose to obtain ϑη(H Ψ ) = −[Φ ,H Ψ ], which is not (65a). Observe that Ψ>H−1 = H(H−1Ψ>)H−1 so that, the following gauge type transformation equation arises > −1 −1 > −1 −1 > −1 −1 > −1 −1 ϑη(Ψ H ) = ϑη(H)(H Ψ )H + Hϑη(H Ψ )H − H(H Ψ )H ϑη(H)H −1 −1 > −1 > −1 > −1 = ϑη(H)H ,H(H Ψ )H − H[Φ ,H Ψ ]H −1 > −1 > −1 = ϑη(H)H − HΦ H , Ψ H . −1 > −1 Now, attending to (57), we have ϑη(H)H − HΦ H = J + Φ so that > −1 > −1 > −1 > −1 ϑη(Ψ H ) = J + Φ, Ψ H = Φ, Ψ H + Ψ H , and (65a) follows immediately. 3. Conclusions and outlook The use of the Gauss–Borel factorization of the moment matrix has been throughly used by Adler and van Moerbeke in their studies of integrable systems and orthogonal polynomials, see [2, 3, 4]. We have ex- tended these ideas and applied them in di erent contexts, CMV orthogonal polynomials, matrix orthogonal polynomials, multiple orthogonal polynomials and multivariate orthogonal [6, 5, 7, 8, 9, 10, 11, 12, 13]. For an general overview see [38]. In this paper, we extended those ideas to the discrete world. In particular we applied this approach to the study of the consequences of the Pearson equation on the moment matrix and Jacobi matrices. For that description a new banded matrix is required, the Laguerre– Freud structure matrix that encodes the Laguerre–Freud relations for the recurrence coe cients. We have also found that the contiguous relations fulfilled generalized hypergeometric functions determining the moments of the weight described for the squared norms of the orthogonal polynomials a discrete Toda hierarchy known as Nijho –Capel equation, see [41]. Further work in this direction are the study of the role of Christo el and Geronimus transformations for the description of the mentioned contiguous relations, and the use of the Geronimus–Christo el transformations to characterize the shifts in the spectral independent variable of the orthogonal polynomials [39]. In [25] these ideas are applied to generalized Charlier, Meixner, and type I Hahn discrete orthogonal polynomials extending the results of [21, 43, 26, 27, 28]. 24 M MAÑAS, I FERNÁNDEZ-IRISARRI, AND OF GONZÁLEZ-HERNÁNDEZ
For the future, we will study the type II generalized Hahn polynomials, and extend these techniques to multiple discrete orthogonal polynomials [14] and its relations with the transformations presented in [18] and quadrilateral lattices [20, 40],
Acknowledgments This work has its seed in several inspiring conversations with Diego Dominici during a research stay at Johannes Kepler University at Linz.
References [1] Vsevolod E. Adler, Alexander I. Bobenko, Yuri B. Suris, Classi cation of Integrable Discrete Equations of Octahedron Type, Inter- national Mathematics Research Notices 2012 no8 (2012) 1822–1889, DOI:10.1093/imrn/rnr083. [2] Mark Adler and Pierre van Moerbeke, Vertex operator solutions to the discrete KP hierarchy, Communications in Mathematical Physics 203 (1999) 185-210 [3] ———-, Generalized orthogonal polynomials, discrete KP and Riemann–Hilbert problems, Communications in Mathematical Physics 207 (1999) 589-620. [4] ———-, Darboux transforms on band matrices, weights and associated polynomials, International Mathematics Research Notices 18 (2001) 935-984. [5] Carlos Álvarez-Fernández, Ulises Fidalgo Prieto, and Manuel Mañas, Multiple orthogonal polynomials of mixed type: Gauss-Borel factorization and the multi-component 2D Toda hierarchy, Advances in Mathematics 227 (2011) 1451–1525. [6] Carlos Álvarez-Fernández, Manuel Mañas, Orthogonal Laurent polynomials on the unit circle, extended CMV ordering and 2D Toda type integrable hierarchies, Advances in Mathematics 240 (2013) 132-193 [7] Carlos Álvarez-Fernández, Gerardo Ariznabarreta, Juan C. García-Ardila, Manuel Mañas, and Francisco Marcellán, Christof- fel transformations for matrix orthogonal polynomials in the real line and the non-Abelian 2D Toda lattice hierarchy, International Mathematics Research Notices 2017 no5 (2017) 1285-1341, DOI:10.1093/imrn/rnw027 . [8] Gerardo Ariznabarreta, Juan C. García-Ardila, Manuel Mañas, and Francisco Marcellán, Matrix biorthogonal poly- nomials on the real line: Geronimus transformations, Bulletin of Mathematical Sciences 9 (2019) 195007 (68 pages) DOI:10.1142/S1664360719500073 (World Scientic) or DOI:10.1007/s13373-018-0128-y3 (Springer-Verlag). [9] Gerardo Ariznabarreta, Juan C. García-Ardila, Manuel Mañas, and Francisco Marcellán, Non-Abelian integrable hierarchies: matrix biorthogonal polynomials and perturbations Journal of Physics A: Mathematical & Theoretical 51 (2018) 205204. [10] Gerardo Ariznabarreta and Manuel Mañas, Matrix orthogonal Laurent polynomials on the unit circle and Toda type integrable systems, Advances in Mathematics 264 (2014) 396-463. [11] ———-, Multivariate orthogonal polynomials and integrable systems, Advances in Mathematics 302 (2016) 628–739. [12] ———-, Christoel transformations for multivariate orthogonal polynomials, Journal of Approximation Theory 225 (2018) 242–283. [13] Gerardo Ariznabarreta, Manuel Mañas, and Alfredo Toledano, CMV Biorthogonal Laurent Polynomials: Perturbations and Christof- fel Formulas, Studies in Applied Mathematics 140 (2018) 333–400. [14] Jorge Arvesú, Jonathan Coussement, and Walter Van Assche, Some discrete multiple orthogonal polynomials, Journal of Compu- tational and Applied Mathematics 153 (2003). [15] R. A. Askey and Adri B. Olde Daalhuis, Generalized hypergeometric function (2010), in Olver, Frank W. J.; Lozier, Daniel M.; Boisvert, Ronald F.; Clark, Charles W. (eds.), NIST Handbook of Mathematical Functions, Cambridge University Press. [16] Jinho Baik, Thomas Kriecherbauer, Kenneth T.-R. McLaughlin, and Peter D. Miller, Discrete Orthogonal Polynomials, Annals of Mathematics Studies 164, Princeton University Press, 2007. [17] Richard Beals and Roderick Wong, Special functions and orthogonal polynomials, Cambridge Studies in Advanced Mathematics 153, Cambridge University Press, 2016. [18] Amílcar Branquinho, Ana Foulquié-Moreno, and Manuel Mañas, Multiple orthogonal polynomials on the step-line, arXiv:2106.12707 [CA] (2021). [19] Peter A. Clarkson, Recurrence coe cients for discrete orthonormal polynomials and the Painlevé equations, Journal of Physics A: Mathematical & Theoretical 46 (2013) 185205. [20] Adam Doliwa, Paolo Maria Santini, and Manuel Mañas, Transformations of quadrilateral lattices, Journal of Mathematical Physics 41 (2000) 944–990. [21] Diego Dominici, Laguerre–Freud equations for generalized Hahn polynomials of type I, Journal of Di erence Equations and Appli- cations 24 (2018) 916–940. [22] ———-, Matrix factorizations and orthogonal polynomials, Random Matrices Theory Applications 9 (2020) 2040003, 33 pp. [23] Diego Dominici and Francisco Marcellán, Discrete semiclassical orthogonal polynomials of class one, Pacific Journal of Mathematics 268 no2 (2012) 389-411. DISCRETE ORTHOGONAL POLYNOMIALS AND CHOLESKY FACTORIZATION 25
[24] ———-, Discrete semiclassical orthogonal polynomials of class 2 in Orthogonal Polynomials: Current Trends and Applications, edited by E. Huertas and F. Marcellán, SEMA SIMAI Springer Series, 22 (2021) 103-169, Springer. [25] Itsaso Fernández-Irrisarri and Manuel Mañas, Pearson Equations for Discrete Orthogonal Polynomials: II. Generalized Hypergeometric Functions and Toda Equations, arXiv:2107.02177 [CA] (2021). [26] Galina Filipuk and Walter Van Assche, Recurrence coe cients of generalized Charlier polynomials and the fth Painlevé equation, Proceedings of American Mathematical Society 141 (2013) 551–62. [27] ———-, Recurrence Coe cients of a New Generalization of the Meixner Polynomials, Symmetry, Integrability and Geometry: Meth- ods and Applications (SIGMA) 7 (2011), 068, 11 pages. [28] ———-, Discrete Orthogonal Polynomials with Hypergeometric Weights and Painlevé VI, Symmetry, Integrability and Geometry: Methods and Applications (SIGMA) 14 (2018), 088, 19 pages. [29] Géza Freud. On the coe cients in the recursion formulae of orthogonal polynomials, Proceedings of the Royal Irish Academy Section A 76 no1 (1976) 1–6. [30] Jarmo Hietarinta, Nalini Joshi and Frank W. Nijho , Discrete Systems and Integrabilty, Cambridge Texts in Applied Mathematics, Cambridge University Press, 2016. [31] Mourad E. H.Ismail, Classical and Quantum Orthogonal Polynomails in One Variable, Encyclopedia of Mathematics and its Ap- plications 98, Cambridge University Press, 2009. [32] Mourad E. H. Ismail and Walter Van Assche, Encyclopedia of Special Functions: The Askey–Bateman Project. Volume I: Univariate Orthogonal Polynomials, Edited by Mourad Ismail, Cambridge University Press, 2020. [33] Edmond Laguerre, Sur la réduction en fractions continues d’une fraction qui satisfait à une équation diérentialle linéaire du premier ordre dont les coe cients sont rationnels. Journal de Mathématiques Pures et Appliquées 4e série, tome 1 (1885) 135–165 . [34] Alphonse P. Magnus, A proof of Freud’s conjecture about the orthogonal polynomials related to |x|ρ exp(−x2m), for integer m, in “Orthogonal polynomials and applications (Bar-le-Duc, 1984)”, Lecture Notes in Mathematics 1171 362–372, Springer, 1985. [35] ———-, On Freud’s equations for exponential weights, Journal of Approximation Theory 46(1) (1986) 65–99. [36] ———-, Painlevé-type dierential equations for the recurrence coe cients of semi-classical orthogonal polynomials, Journal of Compu- tational and Applied Mathematics 57 (1995) 215–237. [37] ———-, Freud’s equations for orthogonal polynomials as discrete Painlevé equations, in “Symmetries and integrability of di erence equations (Canterbury, 1996)”, London Mathematical Society Lecture Note Series 255 228–243, Cambridge University Press, 1999. [38] Manuel Mañas, Revisiting Biorthogonal Polynomials. An LU factorization discussion in Orthogonal Polynomials: Current Trends and Applications, edited by E. Huertas and F. Marcellán, SEMA SIMAI Springer Series, 22 (2021) 273-308, Springer. [39] ———-, Pearson Equations for Discrete Orthogonal Polynomials: III. Christoel and Geronimus transformations, arXiv:2107.02918 [CA] (2021). [40] Manuel Mañas, Adam Doliwa, and Paolo Maria Santini, Darboux transformations for multidimensional quadrilateral lattices. I, Physics Letters A 232 (1997) 99–105. [41] Frank W. Nijho and Hans W. Capel, The direct linearisation approach to hierarchies of integrable PDEs in 2 + 1 dimensions: I. Lattice equations and the dierential-dierence hierarchies. Inverse Problems 6 (1990) 567-590. [42] Arthur F. Nikiforov, Sergei K. Suslov, and Vasilii B. Uvarov, Classical Orhogonal Polynomials of a Discrete Variable, Springer Series in Computational Physics, Springer, 1991. [43] Christophe Smet and Walter Van Assche, Orthogonal polynomials on a bi-lattice, Constructive Approximation 36 (2012) 215–242. [44] Walter Van Assche, Orthogonal Polynomials and Painlevé Equations, Australian Mathematical Society Lecture Series 27, Cam- bridge University Press, 2018.