Abstract
This manuscript contains a small portion of the algebraic theory of orthogonal polynomials developed by Maroni and their applicability to the study and characterization of the classical families, namely Hermite, Laguerre, Jacobi, and Bessel polynomials. It is presented a cyclical proof of some of the most relevant characterizations, particularly those due to Al-Salam and Chihara, Bochner, Hahn, Maroni, and McCarthy. Two apparently new characterizations are also added. Moreover, it is proved through an equivalence relation that, up to constant factors and affine changes of variables, the four families of polynomials named above are the only families of classical orthogonal polynomials.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Orthogonal Polynomials: The Algebraic Theory
Denote by \(\textrm{M}_m(\textrm{A})\) the ring of all square matrices of order m over the commutative ring with unit \(\textrm{A}\). Consider the free module \({{\mathcal {M}}}=\textrm{M}_m(\textrm{A})[X]\) of all “polynomials” in one indeterminate X with coefficients in \(\textrm{M}_m(\textrm{A})\). (Assume that the indeterminate is contained in the centre of \({{\mathcal {M}}}\).) Any free system of polynomials over \(\textrm{M}_m(\textrm{A})\) is a basis of \({{\mathcal {M}}}\). Moreover, the dual of a finite generated free module is a finitely generated free module. However, the dual system associated with a free sequence \((p_n)_{n\ge 0}\) in \({{\mathcal {M}}}\), \(({\textbf{a}}_n)_{n \ge 0}\), is not a system of generators of its algebraic dual, \({{\mathcal {M}}}^*\), in general (cf. [3, VII, Sect. 3, Exercise 10]). But if we consider, for instance, \({{\mathcal {P}}}={{\mathbb {C}}}[X]\), instead of \({{\mathcal {M}}}\), endowed with an “appropriate” topology, then \({{\mathcal {P}}}^*={{\mathcal {P}}}'\) (cf. [23, Exercise 13.1, p. 134]), \({{\mathcal {P}}}'\) being the topological dual \({{\mathcal {P}}}\). Hence
for all \({\textbf{u}} \in {{\mathcal {P}}}'\). (Recall that for every pair of elements \(x \in \textrm{E}\) and \({\textbf{x}} \in \textrm{E}^*\), the element \({\textbf{x}}(x)\) of \(\textrm{E}\) is denoted by \(\left\langle {x, {\textbf{x}}}\right\rangle \), where \(\textrm{E}\) is a left \(\textrm{A}\)-module whose domain of operators is \(\textrm{A}\).) This observation is the cornerstone of Rota’s umbral calculus (cf. [22]) and the algebraic theory of orthogonal polynomials (OP) founded by Maroni (cf. [12,13,14]). For further reading on how to re-establish the “symmetry” between an infinite dimensional vector space and its dual see [4, II, Sect. 6].
This manuscript contains a short exposition of Maroni’s approach on OP and their applicability to the study and characterization of the classical families of OP. A must read on this topic are Maroni [11, 15,16,17,18]. A more detailed historical overview about classical orthogonal polynomials can be find in the works by Maroni or the survey paper [1], among other references. Even a proof of Theorem 3.1 can be found, in one way or another, in the works of Maroni. However, among other questions of pedagogical nature, in this survey we present a cyclical proof of our main results, Theorem 3.1. Moreover, the characterizations (C4) and (C4’) are apparently new ones. While is true that any classical functional is equivalent to one of the canonical forms given in Table 1 below, Theorem 3.2 rigorously reflects this property and, as far as we know, it is not available in the literature. Many results are stated without proof, either because they are simple to prove or because they can be easily found in the literature. Other results that fit in these two categories are, however, proved when the proof methods are different or more attractive, from our point of view, to the existing ones. The positive definite case, widely discussed in the literature, is intentionally omitted. Finally, as the reader will notice, some results are “purely” algebraic. In any case, we will not make distinction between \({{\mathcal {P}}}^*\) and \({{\mathcal {P}}}'\) after Proposition 1.1 below.
1.1 The Spaces \({{\mathcal {P}}}\) and \({{\mathcal {P}}}'\)
It is useful to consider OP as test functions living in an appropriate locally convex space (LCS), which we denote by \({{\mathcal {P}}}\). This LCS is the set of all polynomials (with real or complex coefficients) endowed with a strict inductive limit topology, so that
where \({{\mathcal {P}}}_n\) is the space of all polynomials of degree at most n. (For the sake of simplicity, we do not distinguish between polynomial and polynomial function.) Since \({{\mathcal {P}}}_n\) is a finite dimensional vector space, all its norms are equivalent, so there is no need to specify any particular one. For the development of the theory to be presented here it is not important to know much about the topology (the definition and basic properties of LCS, including inductive limit topologies, can be found, e.g., in Chapter V of Reed and Simon’s book [21], but the reader should keep in mind that the reason why such topology is introduced is because it implies the following fundamental property:
Proposition 1.1
Let \({{\mathcal {P}}}=\text {ind lim}_n\,{{\mathcal {P}}}_n\), as in (1.2), and let \({{\mathcal {P}}}^*\) and \({{\mathcal {P}}}'\) be the algebraic and the topological duals of \({{\mathcal {P}}}\), respectively. Then
Proof
Obviously, \({{\mathcal {P}}}'\subseteq {{\mathcal {P}}}^*\). To prove that \({{\mathcal {P}}}^*\subseteq {{\mathcal {P}}}'\), take \(\textbf{u}\in {{\mathcal {P}}}^*\). From the basic properties of the inductive limit topologies, to prove that \(\textbf{u}\in {{\mathcal {P}}}'\) it suffices to show that the restriction \(\textbf{u}|{{\mathcal {P}}}_n\) is continuous for every n. But this is a trivial assertion, since \(\textbf{u}|{{\mathcal {P}}}_n\) is a linear functional defined on a finite dimensional normed space. \(\square \)
Equality (1.3) means that every linear functional defined in \({{\mathcal {P}}}\) is continuous (for the strict inductive limit topology in \({{\mathcal {P}}}\)). This is a “curious” property, because, for instance, (1.3) is not true for a normed vector space N. Indeed, \(N'=N^*\) if \(\textrm{dim}\,N<\infty \), whilst \(N'\ne N^*\) whenever \(\textrm{dim}\,N=\infty \). Note that being a strict inductive limit of the spaces \({{\mathcal {P}}}_n\), and taking into account that each \({{\mathcal {P}}}_n\) is a proper closed subspace of \({{\mathcal {P}}}_{n+1}\) (so that \({{\mathcal {P}}}\) is indeed an hyper strict inductive limit of the spaces \({{\mathcal {P}}}_n\)), the general theory of LCS ensures that \({{\mathcal {P}}}\) cannot be a metrizable space, and so a fortiori it is not a normed space—or, to be more precise, it is not possible to provide \({{\mathcal {P}}}\) with a norm that generates in it the above inductive limit topology.
In \({{\mathcal {P}}}'\) we consider the weak dual topology, which, by definition, is generated by the family of semi-norms \(s_p:{{\mathcal {P}}}'\rightarrow [0,+\infty [\), \(p\in {{\mathcal {P}}}\), defined by
It turns out that this family of semi-norms \(s_p\) is equivalent to the family of semi-norms \(|\cdot |_n:{{\mathcal {P}}}'\rightarrow [0,+\infty [\), \(n\in {{\mathbb {N}}}_0\), defined by
Indeed, the following proposition holds.
Proposition 1.2
\({{\mathcal {S}}}=\{s_p:p\in {{\mathcal {P}}}\}\) and \({{\mathcal {S}}}_\sharp =\{|\cdot |_n:n\in {{\mathbb {N}}}_0\}\), with \(s_p\) and \(|\cdot |_n\) given by (1.4)–(1.5), are equivalent families of seminorms in \({{\mathcal {P}}}'\), provided \({{\mathcal {P}}}=\text {ind lim}_n\,{{\mathcal {P}}}_n\).
Proof
Given \(p\in {{\mathcal {P}}}\), putting \(p(x)=\sum _{j=0}^na_jx^j\) and \(C(p)=\sum _{j=0}^n|a_j|\), we have
On the other hand, given \(n\in {{\mathbb {N}}}_0\), setting \(p_j(x)=x^j\) (\(j=0,1,\ldots ,n\)), we have
Thus, \({{\mathcal {S}}}\) and \({{\mathcal {S}}}_\sharp \) are equivalent families of semi-norms (see [21, p.126]). \(\square \)
Remark 1.1
Since \({{\mathcal {S}}}_\sharp \) is a countable family of seminorms, then \({{\mathcal {P}}}'\) is a metrizable space, a metric being given by
Moreover, \({{\mathcal {P}}}'\) is a Fréchet space.
1.2 Dual Basis in \({{\mathcal {P}}}^*\)
In \({{\mathcal {P}}}^*\), addition and multiplications by scalars can be defined by
for all \(n\in {{\mathbb {N}}}_0\). \({{\mathcal {P}}}^*\), endowed with these operations, is a vector space over \({{\mathbb {C}}}\). In the vector space \({{\mathcal {P}}}^*\), the identity for the additivity is denoted by \({\textbf{0}}\) and called the zero (or the null element). The zero is therefore defined by the relation \(\left\langle {{\textbf{0}}, x^n}\right\rangle =0\) for all \(n \in {{\mathbb {N}}}_0\). Of course, the elements of \({{\mathcal {P}}}^*\) can not only be added, but also multiplied (the Cauchy product) in order to make the vector space \({{\mathcal {P}}}^*\) into an algebra. Since we work mainly on \({{\mathcal {P}}}^*\) instead of \({{\mathcal {P}}}\), it would be explicitly build bases in \({{\mathcal {P}}}^*\). This makes sense, since (1.3) allows us writing expansions (finite or infinite sums) of the elements of \({{\mathcal {P}}}^*\) in terms of the elements of a given basis, in the sense of the weak dual topology. Such a basis in \({{\mathcal {P}}}^*\) may be achieved in a natural way, by analogy with the case of finitely generated free modules. A simple set in \({{\mathcal {P}}}\) is a sequence of polynomials, \(\{R_n\}_{n\ge 0}\), such that \(\textrm{deg}\,R_n=n\) for every \(n\in {{\mathbb {N}}}_0\) (where \(R_0\equiv \text {const.}\ne 0\)). To any simple set in \({{\mathcal {P}}}\), \(\{R_n\}_{n\ge 0}\), we may associate a dual basis, which, by definition, is a sequence of linear functionals \(\{\textbf{a}_n\}_{n\ge 0}\), being \(\textbf{a}_n:{{\mathcal {P}}}\rightarrow {{\mathbb {C}}}\), such that
where \(\delta _{n,k}\) represents the Kronecker symbol (\(\delta _{n,k}=1\) if \(n=k\); \(\delta _{n,k}=0\) if \(n\ne k\)).
Proposition 1.3
Let \(\{R_n\}_{n\ge 0}\) be a simple set in \({{\mathcal {P}}}\) and \(\{\textbf{a}_n\}_{n\ge 0}\) the associated dual basis. Let \(\textbf{u}\in {{\mathcal {P}}}^*\). Then
in the sense of the weak dual topology in \({{\mathcal {P}}}'\).
Proof
Notice first that the assertion makes sense, according with (1.3). To prove it, fix \(N\in {{\mathbb {N}}}\) and let
be the partial sum of order N of the series appearing in (1.6). We need to show that
Clearly, it suffices to prove that this equality holds for \(p\in \{R_0,R_1,R_2,\dots \}\). Indeed, fix \(k\in {{\mathbb {N}}}_0\). Then, for \(N>k\),
hence \(\;\lim _{N\rightarrow \infty }\langle \textbf{s}_N-\textbf{u},R_k\rangle =0\). \(\square \)
1.3 Basic Operations in \({{\mathcal {P}}}\) and \({{\mathcal {P}}}'\)
Given a functional \(\textbf{u}\in {{\mathcal {P}}}'\), we will denote by
the moment of order n of \(\textbf{u}\). Clearly, if \(\textbf{u}\) and \(\textbf{v}\) are two functionals in \({{\mathcal {P}}}'\) such that the corresponding sequences of moments satisfy \(u_n=v_n\) for all \(n\in {{\mathbb {N}}}_0\), then \(\textbf{u}=\textbf{v}\). Therefore, each functional \(\textbf{u}\in {{\mathcal {P}}}'\) is uniquely determined by its sequence of moments. Define operators \(M_\phi \) and T, from \({{\mathcal {P}}}\) into \({{\mathcal {P}}}\), by
where \(\phi \in {{\mathcal {P}}}\) (fixed) and \('\) denotes derivative with respect to x. Let \(M_\phi '\) and \(T'\) be the corresponding dual operators. For each \(\textbf{u}\in {{\mathcal {P}}}'\), the images \(M_\phi '{} \textbf{u}\) and \(T'{} \textbf{u}\) are elements (functionals) in \({{\mathcal {P}}}'\), hereafter denoted by \(\phi \,\textbf{u}\) and \(D\textbf{u}\).
Definition 1.1
Let \(\textbf{u}\in {{\mathcal {P}}}'\), \(\phi \in {{\mathcal {P}}}\), and \(c\in {{\mathbb {C}}}\).
-
(i)
the left multiplication of \(\textbf{u}\) by \(\phi \), denoted by \(\phi \textbf{u}\), is the functional in \({{\mathcal {P}}}'\) defined by
$$\begin{aligned} \langle \phi \textbf{u}, p\rangle =\langle \textbf{u}, \phi p\rangle ,\quad p\in {{\mathcal {P}}}; \end{aligned}$$ -
(ii)
the derivative of \(\textbf{u}\), denoted by \(D\textbf{u}\), is the functional in \({{\mathcal {P}}}'\) defined by
$$\begin{aligned} \langle D\textbf{u}, p\rangle =-\langle \textbf{u}, p'\rangle ,\quad p\in {{\mathcal {P}}}; \end{aligned}$$
Note that these definitions, introduced by duality with respect to the operators defined in (1.7), are in accordance with those usually given in the Theory of Distributions (this explains the minus sign appearing in the second definition). Note also that
Definition 1.2
(Translation operators). Let \(b\in {{\mathbb {C}}}\).
-
(i)
The translator operator on \({{\mathcal {P}}}\) is \(\tau _b:{{\mathcal {P}}}\rightarrow {{\mathcal {P}}}\) \((p\mapsto \tau _bp)\) defined by
$$\begin{aligned} \tau _bp(x)=p(x-b),\quad p\in {{\mathcal {P}}}; \end{aligned}$$(1.8) -
(ii)
The translator operator on \({{\mathcal {P}}}'\) is \(\varvec{\tau }_b=\tau _{-b}^{\,\prime }\), i.e., \(\varvec{\tau }_b:{{\mathcal {P}}}'\rightarrow {{\mathcal {P}}}'\) is the dual operator of \(\tau _{-b}\), so that
$$\begin{aligned} \langle \varvec{\tau }_b\textbf{u},p\rangle =\langle \textbf{u},\tau _{-b}p\rangle =\langle \textbf{u},p(x+b)\rangle , \quad \textbf{u}\in {{\mathcal {P}}}',\quad p\in {{\mathcal {P}}}. \end{aligned}$$(1.9)
Definition 1.3
(Homothetic operators). Let \(a\in {{\mathbb {C}}}\setminus \{0\}\).
-
(i)
The homothetic operator on \({{\mathcal {P}}}\) is \(h_a:{{\mathcal {P}}}\rightarrow {{\mathcal {P}}}\) \((p\mapsto h_ap)\) defined by
$$\begin{aligned} h_ap(x)=p(ax),\quad p\in {{\mathcal {P}}}. \end{aligned}$$(1.10) -
(ii)
The homothetic operator on \({{\mathcal {P}}}'\) is \(\varvec{h}_a=h_a^{\,\prime }\), i.e., \(\varvec{h}_a:{{\mathcal {P}}}'\rightarrow {{\mathcal {P}}}'\) is the dual operator of \(h_a\), so that
$$\begin{aligned} \langle \varvec{h}_a\textbf{u},p\rangle =\langle \textbf{u},h_ap\rangle =\langle \textbf{u},p(ax)\rangle , \quad \textbf{u}\in {{\mathcal {P}}}',\quad p\in {{\mathcal {P}}}. \end{aligned}$$(1.11)
Proposition 1.4
Let \(\{P_n\}_{n\ge 0}\) be a simple set in \({{\mathcal {P}}}\) and \(\{\textbf{a}_n\}_{n\ge 0}\) its associated dual basis. Let \(a\in {{\mathbb {C}}}\setminus \{0\}\) and \(b\in {{\mathbb {C}}}\). Define
Then \(\{Q_n\}_{n\ge 0}\) is a simple set in \({{\mathcal {P}}}\), and its dual basis, \(\{\textbf{b}_n\}_{n\ge 0}\), is given by
Remark 1.2
The polynomial \(Q_n\) in (1.12) is, indeed,
so that \(Q_n\) is obtained from \(P_n\) by an affine change of the variable, being \(Q_n\) normalized so that it becomes a monic polynomial whenever \(P_n\) is monic.
Proposition 1.5
Let \(\textbf{u}\in {{\mathcal {P}}}'\) and \(p,q\in {{\mathcal {P}}}\setminus \{0\}\), and denote by \(Z_p\) and \(Z_q\) the zeros of p and q, respectively. Then the following property holds:
Proof
Assume that \(p {\textbf{u}}=q {\textbf{u}}={\textbf{0}}\). Since p and q are coprime, there exist polynomials a, b for which \(a\, p +b \, q=1\). Since, for all n, \(\left\langle {p {\textbf{u}}, a\, x^n}\right\rangle =\left\langle {q {\textbf{u}}, b\, x^n}\right\rangle =0\), we see that
The converse is obvious. \(\square \)
1.4 Orthogonal Polynomial Sequences
Definition 1.4
Let \(\textbf{u}\in {{\mathcal {P}}}'\) and let \(\{P_n\}_{n\ge 0}\) be a sequence in \({{\mathcal {P}}}\).
-
(i)
\(\{P_n\}_{n\ge 0}\) is called an orthogonal polynomial sequence (OPFootnote 1) with respect to \(\textbf{u}\) if \(\{P_n\}_{n\ge 0}\) is a simple set and there exists a sequence \(\{h_n\}_{n\ge 0}\), with \(h_n\in {{\mathbb {C}}}\setminus \{0\}\), such that
$$\begin{aligned} \langle \textbf{u},P_mP_n \rangle =h_n\delta _{m,n},\quad m,n=0,1,2,\ldots ; \end{aligned}$$ -
(ii)
\(\textbf{u}\) is called regular (or quasi-definite) if there exists an OP with respect to \(\textbf{u}\).
As usual, denoting by \(u_j=\langle \textbf{u},x^j\rangle \), \(j\in {{\mathbb {N}}}_0\), the moments of \(\textbf{u}\), we define the associated Hankel determinant \(H_n\equiv H_n(\textbf{u})\) as
It is well known that, given \(\textbf{u}\in {{\mathcal {P}}}'\), then \(\textbf{u}\) is regular if and only if
One of the most important characterizations of OP relies upon the fact that any three consecutive polynomials are connected by a very simple relation, expressed as a three-term recurrence relation (TTRR).
Theorem 1.1
Let \(\{\beta _n\}_{n\ge 0}\) and \(\{\gamma _n\}_{n\ge 0}\) be two arbitrary sequences of complex numbers, and let \(\{P_n\}_{n\ge 0}\) be a sequence of (monic) polynomials defined by the three-term recurrence relation
with initial conditions \(P_{-1}(x)=0\) and \(P_{0}(x)=1\). Then there exists a unique functional \(\textbf{u}\in {{\mathcal {P}}}'\) such that
Moreover, \(\textbf{u}\) is regular and \(\{P_n\}_{n\ge 0}\) is the corresponding monic OP if and only if \(\gamma _n\ne 0\) for each \(n\in {{\mathbb {N}}}_0\).
Remark 1.3
Note the relations
1.5 Orthogonal Polynomials and Dual Basis
Since every OP is a simple set of polynomials, it has an associated dual basis in \({{\mathcal {P}}}^\prime \).
Theorem 1.2
Let \(\textbf{u}\in {{\mathcal {P}}}'\) be regular, \(\{P_n\}_{n\ge 0}\) the corresponding monic OP, and \(\{\textbf{a}_n\}_{n\ge 0}\) the associated dual basis. Then:
-
(i)
For each \(n\in {{\mathbb {N}}}_0\), \(\textbf{a}_n\) is explicitly given by
$$\begin{aligned} \textbf{a}_n=\frac{P_n}{\langle \textbf{u},P_n^2\rangle }\, \textbf{u}. \end{aligned}$$As a consequence, \(\{P_n\}_{n\ge 0}\) is a monic OP with respect to \(\textbf{a}_0\), being
$$\begin{aligned} \textbf{u}=u_0\,\textbf{a}_0. \end{aligned}$$ -
(ii)
Let \(\textbf{v}\in {{\mathcal {P}}}'\) and \(N\in {{\mathbb {N}}}_0\) such that
$$\begin{aligned} \langle \textbf{v},P_n\rangle =0 \;\;\text {if}\;\; n\ge N+1. \end{aligned}$$Then,
$$\begin{aligned} \textbf{v}=\sum _{j=0}^{N}\langle \textbf{v},P_j\rangle \, \textbf{a}_j=\phi \,\textbf{u}\;,\quad \phi (x)=\sum _{j=0}^{N}\frac{\langle \textbf{v},P_j\rangle }{\langle \textbf{u},P_j^2\rangle }\, P_j(x). \end{aligned}$$Further, \(\deg \phi \le N\), and \(\deg \phi =N\) if and only if \(\langle \textbf{v},P_N\rangle \ne 0\).
-
(iii)
Let the TTRR fulfilled by \(\{P_n\}_{n\ge 0}\) be (1.18). Then \(\{\textbf{a}_n\}_{n\ge 0}\) fulfills
$$\begin{aligned} x\,\textbf{a}_{n}=\textbf{a}_{n-1}+\beta _n\,\textbf{a}_n+\gamma _{n+1}\,\textbf{a}_{n+1} \;, \quad n\in {{\mathbb {N}}}_0, \end{aligned}$$with initial conditions \(\textbf{a}_{-1}=\textbf{0}\) and \(\textbf{a}_{0}=u_0^{-1}\,\textbf{u}\).
Corollary 1.1
Let \(\{P_n\}_{n\ge 0}\) be a monic OP (with respect to some functional in \({{\mathcal {P}}}'\)) and let \(\textbf{v}\in {{\mathcal {P}}}'\). Then \(\{P_n\}_{n\ge 0}\) is a monic OP with respect to \(\textbf{v}\) if and only if
Theorem 1.3
Under the hypothesis of Proposition 1.4, assume further that \(\{P_n\}_{n\ge 0}\) is a monic OP with respect to the functional \(\textbf{u}\in {{\mathcal {P}}}'\), and let the TTRR fulfilled by \(\{P_n\}_{n\ge 0}\) be (1.18). Then, \(\{Q_n\}_{n\ge 0}\) is a monic OP with respect to
and the TTRR fulfilled by \(\{Q_n\}_{n\ge 0}\) is
with initial conditions \(Q_{-1}(x)=0\) and \(Q_0(x)=1\), where
2 Distributional Differential Equation
The distributional differential equation has the form
where \(\phi \in {{\mathcal {P}}}_2\) and \(\psi \in {{\mathcal {P}}}_1\), and \(\textbf{u}\in {{\mathcal {P}}}'\) is the unknown. Notice that we do not require a priori \(\textbf{u}\) to be a regular functional. We may write
being \(a,b,c,p,q\in {{\mathbb {C}}}\). We also define, for each integer or rational number n,
Notice that \(\psi _n(x)=d_{2n}x+e_n\in {{\mathcal {P}}}_1\). Finally, for each \(\textbf{u}\in {{\mathcal {P}}}'\) and each \(n\in {{\mathbb {N}}}_0\), we set
We begin with the following elementary result.
Lemma 2.1
Let \(\textbf{u}\in {{\mathcal {P}}}'\). Then \(\textbf{u}\) satisfies the distributional differential equation (2.1) if and only if the corresponding sequence of moments, \(u_n=\langle \textbf{u},x^n\rangle \), satisfies the second order linear difference equation
Moreover, if \(\textbf{u}\) satisfies (2.1), then \(\textbf{u}^{[n]}\) satisfies
Notice that if both \(\phi \) and \(\psi \) vanish identically then (2.1) reduces to a trivial equation, so we will exclude this situation from our study.
Lemma 2.2
Let \(\textbf{u}\in {{\mathcal {P}}}'\). Suppose that \(\textbf{u}\) is regular and satisfies (2.1), being \(\phi \in {{\mathcal {P}}}_2\) and \(\psi \in {{\mathcal {P}}}_1\), and assume that at least one of the polynomials \(\phi \) and \(\psi \) is nonzero. Then neither \(\phi \) nor \(\psi \) is the zero polynomial, and
Given a monic polynomial \(P_n\) of degree n (which needs not to belong to an OP), we denote by \(P_n^{[k]}\) the monic polynomial of degree n defined by
where, for a given \(\alpha \in {{\mathbb {C}}}\), \((\alpha )_n\) is the Pochhammer symbol, defined as
Clearly, if \(\{P_n\}_{n\ge 0}\) is a simple set in \({{\mathcal {P}}}\), then so is \(\{P_n^{[k]}\}_{n\ge 0}\). Under such conditions, there is a beautiful relation between the associated dual basis:
where \(\{\textbf{a}_n\}_{n\ge 0}\) and \(\big \{\textbf{a}_n^{[k]}\,\big \}_{n\ge 0}\) are the dual basis in \({{\mathcal {P}}}'\) associated with \(\{P_n\}_{n\ge 0}\) and \(\{P_n^{[k]}\}_{n\ge 0}\), respectively.
Lemma 2.3
Let \(\textbf{u}\in {{\mathcal {P}}}'\), and suppose that \(\textbf{u}\) satisfies the distributional differential equation (2.1), with \(\phi \) and \(\psi \) given by (2.2), being at least one of these polynomials nonzero. Suppose further that \(\textbf{u}\) is regular. Then
Moreover, if \(\{P_n\}_{n\ge 0}\) denotes the monic OP with respect to \(\textbf{u}\), and \(P_n^{[k]}\) is defined by (2.8), then \(\textbf{u}^{[k]}=\phi ^k\textbf{u}\) is regular and \(\{P_n^{[k]}\}_{n\ge 0}\) is its monic OP, for each \(k\in {{\mathbb {N}}}_0\).
We may now establish necessary and sufficient conditions ensuring the regularity of a given functional \(\textbf{u}\in {{\mathcal {P}}}'\) satisfying (2.2). The next results was proved in [10, Theorem 2] (see also [5] in a more general context).
Theorem 2.1
Let \(\textbf{u}\in {{\mathcal {P}}}^\prime \setminus \{\textbf{0}\}\), and suppose that \(\textbf{u}\) satisfies
where \(\phi \) and \(\psi \) are nonzero polynomials such that \(\phi \in {{\mathcal {P}}}_2\) and \(\psi \in {{\mathcal {P}}}_1\). Set
Then, \(\textbf{u}\) is regular if and only if
Moreover, under these conditions, the monic OP \(\{P_n\}_{n\ge 0}\) with respect to \(\textbf{u}\) is given by the three-term recurrence relation
with initial conditions \(P_{-1}(x)=0\) and \(P_0(x)=1\), being
In addition, for each \(n\in {{\mathbb {N}}}_0\), \(P_n\) satisfies the distributional Rodrigues formula
3 Classical Orthogonal Polynomials
The classical functionals are the regular solutions (in \({{\mathcal {P}}}'\)) of the distributional equation (2.1). The corresponding OP are called classical orthogonal polynomials. In this section we present the most significant results concerning this important class of OP.
3.1 Definition and Characterizations
Definition 3.1
Let \(\textbf{u}\in {{\mathcal {P}}}'\). \(\textbf{u}\) is called a classical functional if the following two conditions hold:
-
(i)
\(\textbf{u}\) is regular;
-
(ii)
\(\textbf{u}\) satisfies the distributional differential equation
$$\begin{aligned} D(\phi \textbf{u})=\psi \textbf{u}, \end{aligned}$$(3.1)where \(\phi \) and \(\psi \) are polynomials fulfilling
$$\begin{aligned} \deg \phi \le 2,\quad \deg \psi =1. \end{aligned}$$(3.2)
An OP \(\{P_n\}_{n\ge 0}\) with respect to a classical functional is called a classical OP.
Remark 3.1
According with Lemma 2.2, in the above definition conditions (3.2) may be replaced by the weaker conditions
Theorem 2.1 gives necessary and sufficient conditions for the existence of solutions of the distributional equation (2.1), characterizing also such functionals (and, in particular, solving the question of the existence of classical functionals). Thus, we may state: a functional \(\textbf{u}\in {{\mathcal {P}}}'{\setminus }\{\textbf{0}\}\) is classical if and only if there exist \(\phi \in {{\mathcal {P}}}_2\) and \(\psi \in {{\mathcal {P}}}_1\) such that the following conditions hold:
where we have set \(\phi (x)=ax^2+bx+c\) and \(\psi (x)=px+q\) .
In the next proposition we state several characterizations of the classical OP. For convenience, we introduce the concept of admissible pair of polynomials.
Definition 3.2
\((\phi ,\psi )\) is called an admissible pair if
Introducing this concept makes sense, since according with conditions (ii) in (3.4), only admissible pairs may appear in the framework of the theory of classical OP.
Theorem 3.1
(Characterizations of the classical OP). Let \(\textbf{u}\in {{\mathcal {P}}}^\prime \), regular, and let \(\{P_n\}_{n\ge 0}\) be its monic OP. Then the following properties are equivalent:
-
\(\mathrm{C1.}\) \(\textbf{u}\) is classical, i.e., there are nonzero polynomials \(\phi \in {{\mathcal {P}}}_2\) and \(\psi \in {{\mathcal {P}}}_1\) such that \(\textbf{u}\) satisfies
$$\begin{aligned} D(\phi \textbf{u})=\psi \textbf{u}\;; \end{aligned}$$ -
\(\textrm{C1}'.\) there is an admissible pair \((\phi ,\psi )\) such that \(\textbf{u}\) satisfies
$$\begin{aligned} D(\phi \textbf{u})=\psi \textbf{u}\;; \end{aligned}$$ -
\(\mathrm{C2.}\) (Al-Salam and Chihara) there exist a polynomial \(\phi \in {{\mathcal {P}}}_2\) and, for each \(n\in {{\mathbb {N}}}_0\), complex parameters \(a_n\), \(b_n\) and \(c_n\), with \(c_n\ne 0\) if \(n\ge 1\), such that
$$\begin{aligned} \phi (x)P_n^{\prime }(x)=a_nP_{n+1}(x)+b_nP_n(x)+c_nP_{n-1}(x),\quad n\ge 0 \;; \end{aligned}$$ -
\(\mathrm{C3.}\) (Hahn) \(\Big \{ P_n^{[k]}=\frac{\textrm{d}^k}{\textrm{d}x^k}\frac{P_{n+k}}{(n+1)_k}\Big \}_{n\ge 0}\) is a monic OP for some \(k\in {{\mathbb {N}}}\,\);
-
\(\textrm{C3}'.\) \(\big \{ P_n^{[k]}\big \}_{n\ge 0}\) is a monic OP for each \(k\in {{\mathbb {N}}}\,\);
-
\(\mathrm{C4.}\) there exist \(k\in {{\mathbb {N}}}\) and complex parameters \(r_{n}^{[k]}\) and \(s_{n}^{[k]}\) such that
$$\begin{aligned} P_n^{[k-1]}(x)=P_n^{[k]}(x)+r_{n}^{[k]}P_{n-1}^{[k]}(x)+s_{n}^{[k]}P_{n-2}^{[k]}(x) \; , \quad n\ge 2\;; \end{aligned}$$(3.5) -
\(\textrm{C4}'.\)Footnote 2 for each \(k\in {{\mathbb {N}}}\), there exist parameters \(r_{n}^{[k]}\) and \(s_{n}^{[k]}\) such that (3.5) holds;
-
\(\mathrm{C5.}\) (Bochner) there exist polynomials \(\phi \) and \(\psi \) and, for each \(n\ge 0\), a complex parameter \(\lambda _n\), with \(\lambda _n\ne 0\) if \(n\ge 1\), such that \(y=P_n(x)\) is a solution of the second order ordinary differential equation
$$\begin{aligned} \phi (x) y^{\prime \prime } + \psi (x) y^{\prime } +\lambda _n y =0 ,\quad n\ge 0; \end{aligned}$$ -
\(\mathrm{C6.}\) (Maroni) there is an admissible pair \((\phi ,\psi )\) so that the formal Stieltjes series associated with \(\textbf{u}\), \(S_\textbf{u}(z)=-\sum _{n=0}^{\infty }u_{n}/z^{n+1}\), satisfies (formally)
$$\begin{aligned} \phi (z)S_\textbf{u}^{\prime }(z)=[\psi (z)-\phi ^{\prime }(z)]S_\textbf{u}(z)+\left( \psi ^{\prime } -\ \frac{1}{2}\,\phi ^{\prime \prime }\right) u_0 \;; \end{aligned}$$ -
\(\mathrm{C7.}\) (McCarthy) there exists an admissible pair \((\phi ,\psi )\) and, for each \(n\ge 1\), complex parameters \(h_n\) and \(t_n\) such that
$$\begin{aligned} \phi (P_nP_{n-1})^{\prime }(x)=h_nP_n^2(x)-(\psi -\phi ^{\prime })P_nP_{n-1}(x)+t_nP_{n-1}^2(x) \;; \end{aligned}$$ -
\(\mathrm{C8.}\) (distributional Rodrigues formula) there exist a polynomial \(\phi \in {{\mathcal {P}}}_2\) and nonzero complex parameters \(k_n\) such that
$$\begin{aligned} P_n(x) \textbf{u} = k_n D^n\big (\phi ^n(x)\textbf{u}\big ) \;, \quad n\ge 0 . \end{aligned}$$
Moreover, the polynomials \(\phi \) and \(\psi \) may be taken the same in all properties above where they appear. In addition, let the TTRR fulfilled by the monic OP \(\{P_n\}_{n\ge 0}\) be
(\(P_{-1}(x)=0\); \(P_0(x)=1\)). Write \(\phi (x)=ax^2+bx+c\), \(\psi (x)=px+q\), \(d_n=na+p\), and \(e_n=nb+q\). Then
and the parameters appearing in the above characterizations may be computed explicitly:
Proof
By Lemma 2.3 and Theorem 2.1, C1\(\,\Leftrightarrow \,\)C1\('\), C1\(\,\Rightarrow \,\)C3\('\), and C1\('\,\Leftrightarrow \,\)C8. Clearly, C3\('\,\Rightarrow \,\)C3 and C4\('\,\Rightarrow \,\)C4. We show that C3\('\,\Rightarrow \,\)C4\('\) using the same arguments of the proof of C3\(\,\Rightarrow \,\)C4 given in bellow. The proof of C1\('\,\Leftrightarrow \,\)C6 is left to the reader. Thus, we only need to show that:
(C1 \('\,\Rightarrow \,\) C2). Assume that C1\('\) holds. Fix \(n\in {{\mathbb {N}}}_0\). Since \(\deg (\phi P_n')\le n+1\), then
For each integer number j, with \(0\le j\le n+1\), we deduce
If \(0\le j\le n-2\) we obtain \(\langle \textbf{u},\phi P_n'P_j\rangle =0\), and so \(a_{n,j}=0\). Thus, (3.6) reduces to
where, writing \(\phi (x)=ax^2+bx+c\) and \(\psi (x)=px+q\), \(a_n=na\) (by comparison of coefficients), \(b_n=a_{n,n}\), and \(c_n=a_{n,n-1}\). Setting \(j=n-1\) in (3.7), we deduce
hence
Since, by hypothesis, \((\phi ,\psi )\) is an admissible pair, then we may conclude that \(c_n\ne 0\) for each \(n\ge 1\). Thus C1\('\,\Rightarrow \,\)C2. Notice that taking \(j=n\) in (3.7) yields
hence we deduce the expression for \(b_n\) given in the statement of the theorem:
(C2 \(\,\Rightarrow \,\) C3). Suppose that C2 holds. We will show that \(\{ P_n^{[1]}=P_{n+1}^{\prime }/(n+1)\}_{n\ge 0}\) is a monic OP with respect to \(\textbf{v}=\phi \textbf{u}\). Indeed, for each \(n\in {{\mathbb {N}}}_0\) and \(0\le m\le n\),
Therefore, since \(c_{n+1}\ne 0\) for each \(n\ge 0\), we conclude that \(\{P_n^{[1]}\}_{n\ge 0}\) is a monic OP (with respect to \(\textbf{v}=\phi \textbf{u}\)).
(C3 \(\,\Rightarrow \,\) C4). By hypothesis, \(\{P_n^{[k]}=\frac{\textrm{d}^k}{\textrm{d}x^k}\big (\frac{P_{n+k}}{(n+1)_k}\big )\}_{n\ge 0}\) is a monic OP for some (fixed) \(k\in {{\mathbb {N}}}\). Then there exists \(\beta _n^{[k]}\in {{\mathbb {C}}}\) and \(\gamma _n^{[k]}\in {{\mathbb {C}}}\setminus \{0\}\) such that
Similarly, there exists \(\beta _n\in {{\mathbb {C}}}\) and \(\gamma _n\in {{\mathbb {C}}}\setminus \{0\}\) such that
Changing n into \(n+k\) in (3.9), then taking the derivative of order k in both sides of the resulting equation and using Leibnitz rule on the left-hand side, we find
In this equation, replacing \(xP_n^{[k]}\) by the right-hand side of (3.8), and then changing n into \(n-1\), we obtain (3.5), with
(C4 \(\,\Rightarrow \,\) C1). By hypothesis (3.5) holds. Let \(\{\textbf{a}_n\}_{n\ge 0}\) and \(\{\textbf{a}_n^{[k]}\}_{n\ge 0}\) be the dual bases for \(\{P_n\}_{n\ge 0}\) and \(\{P_n^{[k]}\}_{n\ge 0}\), respectively. By Proposition 1.3, \(\,\textbf{a}_n^{[k]}=\sum _{j\ge 0}\langle \textbf{a}_n^{[k]},P_j^{[k-1]}\rangle \textbf{a}_j^{[k-1]}\) for each \(n\in {{\mathbb {N}}}_0\). Using (3.5), we compute
Hence
Taking the (distributional) derivative of order k in both sides of this equation, and using the relations \(D^j\big (\textbf{a}_n^{[j]}\big )=(-1)^j(n+1)_j\,\textbf{a}_{n+j}\), we obtain
Therefore, since, by Theorem 1.2, \(\textbf{a}_j=\frac{P_j}{\langle \textbf{u},P_j^2\rangle }\,\textbf{u}\) for \(j\in {{\mathbb {N}}}_0\) and, by (1.20), \(\gamma _j=\frac{{\langle \textbf{u},P_j^2\rangle }}{{\langle \textbf{u},P_{j-1}^2\rangle }}\) for \(j\in {{\mathbb {N}}}\), being \(\gamma _j\) the parameter appearing in (3.9), we deduce
where \(\Phi _{n+k+1}\) is a polynomial of degree at most \(n+k+1\), given by
Since \(\Phi _{n+k+1}\) is a (finite) linear combination of polynomials of the simple set \(\{P_j\}_{j\ge 0}\) and \(\gamma _{n+k}\ne 0\), then \(\Phi _{n+k+1}\) does not vanish identically, so \(\Phi _{n+k+1}\in {{\mathcal {P}}}_{n+k+1}\setminus \{0\}\). Setting \(n=0\) and \(n=1\) in (3.10) we obtain the two equations
If \(k=1\) it follows immediately from the first of these equations that C1 holds. Henceforth, assume that \(k\ge 2\). Setting \(n=0\) and \(n=1\) in the definition of \(\Phi _{n+k+1}\) and using the TTRR (3.9), we easily deduce
where \(E_0(\cdot ;k),H_0(\cdot ;k)\in {{\mathcal {P}}}_0\) and \(F_1(\cdot ;k),G_1(\cdot ;k)\in {{\mathcal {P}}}_1\), explicitly given by
Let \(\Delta _2(x)\equiv \Delta _2(x;k)=E_0(x;k)H_0(x;k)- F_1(x;k)G_1(x;k)\), the determinant of the system (3.12). Using (3.11)–(3.13), and taking into account that \(\textbf{u}\) is regular, we may prove that \(\Delta _2\in {{\mathcal {P}}}_2\setminus \{0\}\). Solving (3.12) for \(P_k\) and \(P_{k+1}\) we obtain
Since \(P_k\) and \(P_{k+1}\) are coprime, it follows from (3.14)–(3.15) that any common zero of \(\Phi _{k+1}\) and \(\Phi _{k+2}\) (if there is some) must be a zero of \(\Delta _2\). Let \(\Phi \) be the greatest common divisor of \(\Phi _{k+1}\) and \(\Phi _{k+2}\), i.e.,
Any zero of \(\Phi \) is also a zero of both \(\Phi _{k+1}\) and \(\Phi _{k+2}\), and so it is a zero of \(\Delta _2\). Therefore, \(\Phi \in {{\mathcal {P}}}_2\setminus \{0\}\). (Notice that indeed \(\Phi \not \equiv 0\), since \(\Phi _{k+1}\not \equiv 0\) and \(\Phi _{k+2}\not \equiv 0\).) Moreover, there exist polynomials \(\Phi _{1,k}\) and \(\Phi _{2,k}\), with no common zeros, such that
From (3.11) and (3.16) we deduce
Combining these two equations yields \(\big (\Phi _{1,k}(P_{k+1}+\Phi _{2,k}'\Phi )-\Phi _{2,k}(P_k+\Phi _{1,k}'\Phi )\big )\textbf{u}=\textbf{0}\), and so, since \(\textbf{u}\) is regular, \(\Phi _{1,k}(P_{k+1}+\Phi _{2,k}'\Phi )=\Phi _{2,k}(P_k+\Phi _{1,k}'\Phi )\). Therefore, taking into account that \(\Phi _{1,k}\) and \(\Phi _{2,k}\) are coprime and (3.17) holds, we may ensure that there exists a polynomial \(\Psi \in {{\mathcal {P}}}_1\) such that
Combining equations (3.18) and (3.19) we deduce
From these equations, and using once again the fact that \(\Phi _{1,k}\) and \(\Phi _{2,k}\) are coprime, we conclude, by Proposition 1.5, that \(D(\Phi \textbf{u})=\Psi \textbf{u}\). Thus C4\(\,\Rightarrow \,\)C1. The formulas for \(r_n^{[1]}\) and \(s_n^{[1]}\) given in the statement of the theorem may be derived as follows. We have already proved that C4\(\,\Rightarrow \,\)C1\(\,\Rightarrow \,\)C1\('\,\Rightarrow \,\)C2\(\,\Rightarrow \,\)C3\(\,\Rightarrow \,\)C4, and we see that the polynomials \(\phi \) and \(\psi \) appearing in all these characterizations may be taken the same. As we have seen, the formulas for \(b_n\) and \(c_n\) given in the statement of the theorem hold. We now use these formulas to obtain the expressions for \(r_n^{[1]}\) and \(s_n^{[1]}\). Set \(Q_n=P_n^{[1]}=P_{n+1}'/(n+1)\). By C4, \(P_n=Q_n+r_n^{[1]}Q_{n-1}+s_n^{[1]}Q_{n-2}\) if \(n\ge 2\). Hence, since \(\{Q_n\}_{n\ge 0}\) is a monic OP with respect to \(\textbf{v}=\phi \textbf{u}\), we deduce, for each \(n\ge 2\),
where the third equality holds taking into account C2. Similarly, for each \(n\ge 2\),
(C1\(\,\Rightarrow \,\)C5). By hypothesis, \(D(\phi \textbf{u})=\psi \textbf{u}\), where \(\phi \in {{\mathcal {P}}}_2\), \(\psi \in {{\mathcal {P}}}_1\), and \(\deg \psi =1\) (cf. Lemma 2.2). Fix \(n\in {{\mathbb {N}}}\), and write
Then, for each j such that \(0\le j\le n\),
Since by hypothesis C1 holds, and we have already proved that C1\(\,\Rightarrow \,\)C1\('\,\Rightarrow \,\)C2\(\,\Rightarrow \,\)C3, and in the proof of C2\(\,\Rightarrow \,\)C3 we have shown that \(\{Q_n=P'_{n+1}/(n+1)\}_{n\ge 0}\) is a monic OP with respect to \(\textbf{v}=\phi \textbf{u}\), then \(\langle \phi \textbf{u},P_n'P_j'\rangle =0\) if \(j\ne n\), hence (3.20) reduces to
where \(\lambda _n=-\lambda _{n,n}\). Comparing leading coefficients in (3.21), and setting \(\phi (x)=ax^2+bx+c\) and \(\psi (x)=px+q\), we obtain \(\lambda _n=-n\big ((n-1)a+p\big )=-nd_{n-1}\), hence \(\lambda _n\ne 0\) if \(n\ge 1\) (since C1\(\,\Rightarrow \,\)C1\('\), so \((\phi ,\psi )\) is an admissible pair). Thus C1\(\,\Rightarrow \,\)C5.
(C5\(\,\Rightarrow \,\)C1). By hypothesis, there exists \(\phi ,\psi \in {{\mathcal {P}}}\), and \(\lambda _n\in {{\mathbb {C}}}\), with \(\lambda _n\ne 0\) if \(n\ge 1\), such that \(-\phi P_{n+1}^{\prime \prime }=\psi P_{n+1}'+\lambda _{n+1} P_{n+1}\). Taking in this equation \(n=0\) and \(n=1\) we deduce \(\psi =-\lambda _1P_1\in {{\mathcal {P}}}_1\setminus {{\mathcal {P}}}_0\) and \(\phi =-(\psi P_2'+\lambda _2P_2)/2\in {{\mathcal {P}}}_2\). We will prove that \(D(\phi \textbf{u})=\psi \textbf{u}\) by showing that the actions of the functionals \(D(\phi \textbf{u})\) and \(\psi \textbf{u}\) coincide on the simple set \(\{Q_n\}_{n\ge 0}\). Indeed,
Since at least one of the polynomials \(\phi \) and \(\psi \) is nonzero (because \(\lambda _n\ne 0\)), C1 holds.
(C2\(\,\Rightarrow \,\)C7). Since by hypothesis (C2) holds, we may write
Multiplying (3.22) by \(P_{n-1}\) and (3.23) by \(P_n\) and adding the resulting equalities, we find that \(\phi (P_n P_{n-1})^{\prime }\) is a linear combination of the polynomials \(P_n^2\), \(P_nP_{n-1}\), \(P_{n-1}^2\), \(P_{n+1}P_{n-1}\) and \(P_nP_{n-2}\). Substituting \(P_{n+1}\) and \(P_{n-2}\) by the corresponding expressions given by the TTRR, we deduce
where
Write \(\phi (x)=ax^2+bx+c\) and \(\psi (x)=px+q\). We have already seen that C2\(\,\Leftrightarrow \,\)C1\('\), and while proving C1\('\,\Rightarrow \,\)C2 we have shown that the coefficients \(a_n\), \(b_n\), and \(c_n\) appearing in (3.22) are given by \(a_n=na\), \(b_n=-\frac{1}{2}\psi (\beta _n)\), and \(c_n=-d_{n-1}\gamma _n\). It follows that
where the last equality is easily derived using the expressions for the \(\beta -\)parameters given in the statement of the theorem. Therefore, \(B_n x+C_n=\phi ^{\prime }-\psi \) (independent of n). Finally, substituting (3.25) into (3.24) yields the equation appearing in C7, being \(h_n=A_n=d_{2n-3}\) and \(t_n=D_n=-d_{2n-1}\gamma _n\) for each \(n\ge 1\). Thus C2\(\,\Rightarrow \,\)C7.
(C7\(\,\Rightarrow \,\)C2). Fix an integer \(n\ge 1\). For this n, rewrite the equation in (C7) as
Therefore, since \(P_n\) and \(P_{n-1}\) are coprime, there is \(\pi _{1,n}\in {{\mathcal {P}}}_1\) such that
By comparing the leading coefficients on both sides of equation (3.26) we deduce \(\pi _{1,n}(x)=d_nx+z_n\) for some \(z_n\in {{\mathbb {C}}}\) (and \(d_n=na+p\)). By hypothesis, \((\phi ,\psi )\) is an admissible pair, hence \(d_n\ne 0\) and so \(\deg \pi _{1,n}=1\). Moreover, by the TTRR for \(\{P_n\}_{n\ge 0}\), \(xP_n=P_{n+1}+\beta _nP_n+\gamma _{n}P_{n-1}\). Therefore, (3.26) may be rewritten as
where \(a_n=na\), \(b_n=na\beta _n+z_n-q\), and \(c_n=na\gamma _n+t_n\). To conclude the proof we need to show that \(c_n\ne 0\) for all \(n\ge 1\). Indeed, changing n into \(n+1\) in (3.27) and adding the resulting equation with (3.26), we obtain
Since \(\psi +\phi '=(2a+p)x+q+b\) and taking into account once again the TTRR for \(\{P_n\}_{n\ge 0}\), the last equation may be rewritten as a trivial linear combination of the three polynomials \(P_{n+1}\), \(P_{n}\), and \(P_{n-1}\). Thus, we deduce
Therefore, \(c_n=na\gamma _n+t_n=-d_{n-1}\gamma _n\ne 0\) (since \(n\ge 1\)). This completes the proof. \(\square \)
Remark 3.2
The parameters \(\beta _n\) and \(\gamma _n\) appearing in Theorem 3.1 may be written explicitly in terms of the coefficients of \(\phi \) and \(\psi \) as follows:
Remark 3.3
Using the previous results and Table 1, it is easy to see that (see also [2, 6, 7, 9, 19] and references therein)
3.2 Classification and Canonical Representatives
We all always hear say: up to constant factors and affine changes of variables, there are only four (parametric) families of classical OP, namely, Hermite, Laguerre, Jacobi, and Bessel polynomials. But, what is the rigorous meaning of this statement? The corresponding regular functionals will be denoted by \(\textbf{u}_H\), \(\textbf{u}_L^{(\alpha )}\), \(\textbf{u}_J^{(\alpha ,\beta )}\), and \(\textbf{u}_B^{(\alpha )}\) (resp.) and these will be called the canonical representatives (or canonical forms) of the classical functionals. Their description is given in Table 1. Each one of these functionals fulfils (3.1), being the corresponding pair \((\phi ,\psi )\equiv (\Phi ,\Psi )\) given in the table. The regularity conditions in the table are determined by conditions (ii) appearing in (3.4).
Ultimately, denoting by \([\textbf{u}]\) the equivalent class determined by a functional \(\textbf{u}\in {{\mathcal {P}}}'\) (see [17, Sect. 3.1.2.4, pp. 18–19]), and setting
we will show that
where the parameters \(\alpha \) and \(\beta \) vary on \({{\mathbb {C}}}\) subject to the regularity conditions in Table 1, and \(\sim \) is an equivalence relation in \({{\mathcal {P}}}'\) defined by
We start by proving a proposition that allows to ensure that this equivalence relation preserves the classical character of a given classical functional.
Lemma 3.1
Let \(\textbf{u},\textbf{v}\in {{\mathcal {P}}}'\) and suppose that \(\textbf{u}\sim \textbf{v}\), i.e., (3.28) holds. Suppose that there exist two polynomials \(\phi \) and \(\psi \) such that
Let \(\Phi (x)=K\phi (Ax+B)\) and \(\Psi (x)=KA\psi (Ax+B)\), being \(K\in {{\mathbb {C}}}\setminus \{0\}\). Then
Moreover, if \(\textbf{u}\) is a classical functional, then so is \(\textbf{v}\).
Proof
Since \(\textbf{u}\) and \(\textbf{v}\) fulfill (3.28), then
Therefore, for each \(n\in {{\mathbb {N}}}_0\), we have
Finally, the last sentence stated in the lemma follows by using Theorem 1.3. \(\square \)
Theorem 3.2
(Canonical representatives of the classical functionals). Let \(\textbf{u}\in {{\mathcal {P}}}^\prime \) be a classical functional, so that \(\textbf{u}\) fulfils
where \(\phi (x)=ax^2+bx+c\) and \(\psi (x)=px+q\), subject to the regularity conditions
Then, there exists a regular functional \(\textbf{v}\in {{\mathcal {P}}}'\) such that
where, for each classical functional determined by the pair \((\phi ,\psi )\), the corresponding pair \((\Phi ,\Psi )\) is given by Table 1. More precisely, setting
the following holds:
-
1.
(Hermite) if \(a=b=0\), then:
$$\begin{aligned} \textbf{v}=\big (\varvec{h}_{\sqrt{-p/(2c)}}\circ \varvec{\tau }_{q/p}\big )\textbf{u}=\textbf{u}_{{}_H}; \end{aligned}$$ -
2.
(Laguerre) if \(a=0\) and \(b\ne 0\), then:
$$\begin{aligned} \textbf{v}=\big (\varvec{h}_{-p/b}\circ \varvec{\tau }_{c/b}\big )\textbf{u}=\textbf{u}_{{}_L}^{(\alpha )},\quad \alpha =-1+(qb-pc)/b^2; \end{aligned}$$ -
3.
(Bessel) if \(a\ne 0\) and \(\Delta =0\), then:
$$\begin{aligned} \textbf{v}=\big (\varvec{h}_{2a/d}\circ \varvec{\tau }_{b/(2a)}\big )\textbf{u}=\textbf{u}_{{}_B}^{(\alpha )},\quad \alpha =-2+p/a; \end{aligned}$$ -
4.
(Jacobi) if \(a\ne 0\) and \(\Delta \ne 0\), then:
$$\begin{aligned}{} & {} \textbf{v}=\big (\varvec{h}_{-2a/\sqrt{\Delta }}\circ \varvec{\tau }_{b/(2a)}\big )\textbf{u}=\textbf{u}_{{}_J}^{(\alpha ,\beta )},\\{} & {} \alpha =-1+p/(2a)-d/\sqrt{\Delta },\quad \beta =-1+p/(2a)+d/\sqrt{\Delta }. \end{aligned}$$
Proof
Taking into account Lemma 3.1, the theorem will be proved if we are able to show that, for each given pair \((\phi ,\psi )\), and for each corresponding pair \((\Phi ,\Psi )\) given by Table 1—where the “corresponding pair” \((\Phi ,\Psi )\) is the one in the table such that \(\phi \) and \(\Phi \) have the same degree and their zeros the same multiplicity—, there exist \(A,K\in {{\mathbb {C}}}\setminus \{0\}\) and \(B\in {{\mathbb {C}}}\) such that the relations
hold, for appropriate choices of the parameters \(\alpha \) and \(\beta \) appearing in Table 1 for the Laguerre, Bessel, and Jacobi cases. Indeed, considering the four possible cases determined by the polynomial \(\phi \), we have:
-
1.
Assume \(a=b=0\), i.e., \(\phi (x)=c\). The regularity conditions (3.30) ensure that \(p\ne 0\) and \(c\ne 0\). Therefore, since in this case we require \((\Phi ,\Psi )=(1,-2x)\), from (3.32) we obtain the equations
$$\begin{aligned} 1=Kc,\quad -2=KA^2p,\quad 0=Bp+q. \end{aligned}$$A solution of this system of equations is
$$\begin{aligned} K=1/c,\quad A=\sqrt{-2c/p},\quad B=-q/p, \end{aligned}$$which gives the desired result for the Hermite case, by Lemma 3.1.
-
2.
Assume \(a=0\) and \(b\ne 0\), so that \(\phi (x)=bx+c\). Since in this case we require \((\Phi ,\Psi )=(x,-x+\alpha +1)\), from (3.32) we obtain
$$\begin{aligned} 1=KAb,\quad 0=bB+c,\quad -1=KA^2p,\quad \alpha +1=KA(Bp+q). \end{aligned}$$Solving this system we find
$$\begin{aligned} K=-p/b^2,\quad B=-c/b,\quad A=-b/p,\quad \alpha =-1+(qb-pc)/b^2. \end{aligned}$$Notice that, in this case,
$$\begin{aligned} d_n=p,\quad \phi \left( -\frac{nb+q}{2na+p}\right) =-\frac{b^2}{p}\big (n+\alpha +1\big ), \end{aligned}$$hence the regularity conditions (3.30) ensure that \(p\ne 0\) (and so K and A are well defined, being both nonzero complex numbers) and \(-\alpha \not \in {{\mathbb {N}}}\).
-
3.
Assume \(a\ne 0\) and \(\Delta =0\). Then \(\phi (x)=a\Big (x+\frac{b}{2a}\Big )^2\). In this case we require \((\Phi ,\Psi )=\big (x^2,(\alpha +2)x+2\big )\), hence from (3.32) we obtain
$$\begin{aligned} 1=KA^2a,\quad 0=B+b/(2a),\quad \alpha +2=KA^2p,\quad 2=KA(Bp+q). \end{aligned}$$Therefore, taking into account that \(d=\psi \left( -\ \frac{b}{2a}\right) =(2aq-pb)/(2a)\), we deduce
$$\begin{aligned} K=4a/d^2,\quad B=-b/(2a),\quad A=d/(2a),\quad \alpha =-2+p/a. \end{aligned}$$In this case we have
$$\begin{aligned} d_n=a(n+\alpha +2),\quad \phi \left( -\frac{nb+q}{2na+p}\right) =\frac{d^2}{a(2n+\alpha +2)^2}, \end{aligned}$$hence conditions (3.30) ensure that \(-(\alpha +1)\not \in {{\mathbb {N}}}\) and \(d\ne 0\), and so, in particular, K is well defined, being both K and A nonzero complex numbers.
-
4.
Finally, assume \(a\ne 0\) and \(\Delta \ne 0\). Writing \(\phi (x)=a\Big [\Big (x+\frac{b}{2a}\Big )^2-\frac{\Delta }{4a^2}\Big ]\), since in this case we require \((\Phi ,\Psi )=\big (1-x^2,-(\alpha +\beta +2)x+\beta -\alpha \big )\), from (3.32) we obtain
$$\begin{aligned} \begin{array}{l} -1=KA^2a,\quad 0=B+b/(2a),\quad 1=Ka\Big [\Big (B+\frac{b}{2a}\Big )^2-\frac{\Delta }{4a^2}\Big ], \\ \qquad -(\alpha +\beta +2)=KA^2p,\quad \beta -\alpha =KA(Bp+q). \end{array} \end{aligned}$$A solution of this system of five equations is
$$\begin{aligned} \begin{array}{l} K=-4a/\Delta ,\quad B=-b/(2a),\quad A=-\sqrt{\Delta }/(2a), \\ \alpha =-1+p/(2a)-d/\sqrt{\Delta },\quad \beta =-1+p/(2a)+d/\sqrt{\Delta }. \end{array} \end{aligned}$$(We choose A with the minus sign since whenever \((\phi ,\psi )=(\Phi ,\Psi )\) that choice implies \(A=1\) and \(B=0\), hence \(\textbf{u}=\textbf{v}=\textbf{u}_J^{(\alpha ,\beta )}\), and so it is a more natural choice.) Adding and subtracting the last two equations for \(\alpha \) and \(\beta \), we find \(\alpha +\beta +2=p/a\) and \(\alpha -\beta =-2d/\sqrt{\Delta }\), hence we deduce
$$\begin{aligned} d_n=a(n+\alpha +\beta +2),\quad \phi \left( -\frac{nb+q}{2na+p}\right) =-\frac{\Delta }{a}\frac{(n+\alpha +1)(n+\beta +1)}{(2n+\alpha +\beta +2)^2}, \end{aligned}$$Therefore, conditions (3.30) ensure that \(-(\alpha +\beta +1)\not \in {{\mathbb {N}}}\), \(-\alpha \not \in {{\mathbb {N}}}\), and \(-\beta \not \in {{\mathbb {N}}}\). This completes the proof. \(\square \)
Remark 3.4
It follows from the proof of Theorem 3.2 that the parameters \(\alpha \) and \(\beta \) defined in the statement of this theorem (in cases 2, 3, and 4) fulfil the regularity conditions appearing in Table 1.
The preceding theorem allows us to classify each classical functional according with the degree of the polynomial \(\phi \) appearing in equation (3.1).
Corollary 3.1
Let \(\textbf{u}\) be a classical functional, fulfilling (3.1)–(3.2).
-
(i)
if \(\deg \phi =0\) (hence \(\phi \) is a nonzero constant), then \(\textbf{u}\sim \textbf{u}_H\,\);
-
(ii)
if \(\deg \phi =1\), then \(\textbf{u}\sim \textbf{u}_L^{(\alpha )}\,\) for some \(\alpha \);
-
(iii)
if \(\deg \phi =2\) and \(\phi \) has simple zeros, then \(\textbf{u}\sim \textbf{u}_J^{(\alpha ,\beta )}\,\) for some pair \((\alpha ,\beta )\);
-
(iv)
if \(\deg \phi =2\) and \(\phi \) has a double zero, then \(\textbf{u}\sim \textbf{u}_B^{(\alpha )}\,\) for some \(\alpha \).
The monic OP with respect to the canonical representatives \(\textbf{u}_H\), \(\textbf{u}_L^{(\alpha )}\), \(\textbf{u}_J^{(\alpha ,\beta )}\), and \(\textbf{u}_B^{(\alpha )}\) will be denoted by \(\{{\widehat{H}}_n\}_{n\ge 0}\), \(\{{\widehat{L}}_n^{(\alpha )}\}_{n\ge 0}\), \(\{{\widehat{P}}_n^{(\alpha ,\beta )}\}_{n\ge 0}\), and \(\{{\widehat{B}}_n^{(\alpha )}\}_{n\ge 0}\) (resp.), and they will be called the (monic) Hermite, Laguerre, Jacobi, and Bessel polynomials. (The integral representations of the canonical representatives can be find in [11].) Table 2 summarizes the corresponding parameters appearing in all characterizations presented in Theorem 3.1. In view of Theorem 3.2 and Theorem 1.3, we may now justify a sentence made at the beginning of the section regarding Hermite, Laguerre, Jacobi, and Bessel polynomials. As Maroni said in an interview when asked about Bessel polynomials: “comme dans le roman d’Alexandre Dumas, les trois mousquetaires étaient quatre en réalité”.
Notes
For abbreviation, we continue to write OP for orthogonal polynomial sequence.
It was proved for \(k=1\) by Geronimus (see [8, (42)]).
References
Al-Salam, W.A.: Characterization theorems for orthogonal polynomials. In: Orthogonal Polynomials (Columbus, OH, 1989), 1–24, NATO Adv. Sci. Inst. Ser. C: Math. Phys. Sci., vol. 294. Kluwer Academic Publishers, Dordrecht (1990)
Andrews, G., Askey, R., Roy, R.: Special Functions. Cambridge University Press (1999)
Bourbaki, N.: Algebra I: Chapters 1–3. Translated from the French. Elements of Mathematics (Berlin). Springer, Berlin. Reprint of the 1989 English translation edition (1998)
Bourbaki, N.: Topological vector spaces. Chapters 1–5. Translated from the French by H. G. Eggleston and S. Madan. Elements of Mathematics (Berlin). Springer (1987)
Castillo, K., Mbouna, D., Petronilho, J.: On the functional equation for classical orthogonal polynomials on lattices. J. Math. Anal. Appl. 515, 126390 (2022)
Chihara, T.S.: An Introduction to Orthogonal Polynomials. Gordon and Breach (1978)
Erdélyi, A.: Higher Transcendental Functions, vols. 1–3. McGraw-Hill (1955)
Geronimus, Ya.. L.: On polynomials orthogonal with respect to numerical sequences and on Hahn’s theorem. Izv. Akad. Nauk. 4, 215–228 (1940). ((In Russian.))
Ismail, M.E.H.: Classical and Quantum Orthogonal Polynomials in One Variable. Cambridge University Press (2005)
Marcellán, F., Petronilho, J.: On the solution of some distributional differential equations: existence and characterizations of the classical moment functionals. Integral Transforms Spec. Funct. 2, 185–218 (1994)
Maroni, P.: Fonctions eulériennes. Polynômes orthogonaux classiques. Téch. l’Ingén., Traité Gén. (Sci. Fondam.), A 154, pp. 1–28 (1994)
Maroni, P.: Sur quelques espaces de distributions qui sont des formes linéaires sur l’espace vectoriel des polynômes. In: Brezinski, C., et al. (eds.) Simposium Laguerre, Bar-le-Duc, Lecture Notes in Math., vol. 1171, pp. 184–194. Springer (1985)
Maroni, P.: Le calcul des formes linéaires et les polynômes orthogonaux semiclassiques. In: Alfaro, M., et al. (eds.) Orthogonal Polynomials and Their Applications, Lecture Notes in Math., vol. 1329, pp. 279–290. Springer (1988)
Maroni, P.: Une théorie algébrique des polynômes orthogonaux. Applications aux polynômes orthogonaux semiclassiques, In: Brezinski, C., et al. (eds.) Orthogonal Polynomials and their Applications, Proc. Erice 1990, IMACS, Ann. Comput. Appl. Math., vol. 9, pp. 95–130 (1991)
Maroni, P.: Variations autour des polynômes orthogonaux classiques. C. R. Acad. Sci. Paris, Sér. I 313, 209–212 (1991)
Maroni, P.: Variations around classical orthogonal polynomials. Connected problems. J. Comput. Appl. Math. 48, 133–155 (1993)
Maroni, P.: Fonctions eulériennes. Polynômes orthogonaux classiques. Téch. l’Ingén., Traité Gén. (Sci. Fondam.), A 154, 1–30 (1994)
Maroni, P., da Rocha, Z.: A new characterization of classical forms. Commun. Appl. Anal. 5, 351–362 (2001)
Nikiforov, A.F., Uvarov, V.B.: Special Functions of Mathematical Physics. Birkhauser Verlag, Basel (1988)
Petronilho, J.: Orthogonal polynomials and special functions. arXiv:2111.06478 [math.CA] (2021)
Reed, M., Simon, B.: Methods of Modern Mathematical Physics I: Functional Analysis. Academic Press, New York, London (1972)
Roman, S.: The umbral calculus, Pure and Applied Mathematics, vol. 111. Academic Press, Inc. [Harcourt Brace Jovanovich, Publishers], New York (1984)
Trèves, F.: Topological Vector Spaces, Distributions and Kernels. Academic Press, New York, London (1967)
Acknowledgements
The first author thanks Professor Petronilho’s youngest daughter, Inês, who found these old files and sent them to him. This author also thanks the editor and reviewers for providing constructive feedback to improve the manuscript. This work is supported by the Centre for Mathematics of the University of Coimbra-UIDB/00324/2020, funded by the Portuguese Government through FCT/ MCTES.
Funding
Open access funding provided by FCT|FCCN (b-on).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have not disclosed any competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
My friend J. Petronilho passed away on August 27, 2021. In the academic years 2016/17 and 2020/21 we taught a PhD course on Orthogonal Polynomials. This manuscript goes back to the beginning of 2015 and it was later included in a monograph [20] on the subject that Petronilho finished in 2016. Probably, for this reason or the fact that we had started working on [5], which allows to rewrite many results in a more general framework, we shelved this project until now.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Castillo, K., Petronilho, J. Classical Orthogonal Polynomials Revisited. Results Math 78, 155 (2023). https://doi.org/10.1007/s00025-023-01934-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00025-023-01934-2