The paper investigates a question about polynomial sequences: when can the coefficients of a family of polynomials, arranged into a triangular grid of numbers, be described by a structured mathematical object called a Riordan array? Riordan arrays are essentially infinite triangular matrices with a very specific internal structure, defined compactly by just two generating functions. The authors focus on polynomial sequences defined by recurrence relations, where each polynomial is built from the previous ones according to a fixed rule. They give a complete answer for sequences defined by two-step recurrences and partial answers for longer recurrences, showing exactly which structural conditions on the recurrence coefficients allow a Riordan array description.
The paper then broadens the scope to orthogonal polynomial sequences, a classical and important family arising throughout mathematics and physics. Even when a Riordan array description is not available, the authors show that a related object called a production matrix always captures the essential structure of the sequence. Specifically, the production matrix always turns out to be a tridiagonal matrix encoding the defining three-term recurrence, and the inverse of the coefficient matrix always encodes the so-called moment sequence associated with the polynomials. The authors illustrate this with concrete examples: Legendre polynomials, which cannot be described by a Riordan array, and Chebyshev polynomials, which can. They also work out explicit formulas for a broad generalization called the Gegenbauer-Humbert polynomials.
Finally, the authors extend everything to a more general setting called d-orthogonal polynomial sequences, where each polynomial satisfies a recurrence involving d plus 2 previous terms rather than just three. They prove that the associated production matrix is always a banded matrix of a specific width determined by d, and that the inverse of the coefficient matrix recovers multiple moment sequences simultaneously. This unifies d-orthogonality with a generalized notion of Riordan arrays, giving a clean structural picture that neatly reduces to the classical case when d equals one.