Riordan Arrays: Theory and Applications

• The Riordan-array framework is an infinite lower-triangular matrix system defined by two power series, bridging combinatorics and algebra.
• It features a rich group structure with identifiable subgroups such as the Appell, Lagrange, and Bell groups, facilitating structured factorizations and recurrence relations.
• The theory underpins combinatorial identities, umbral calculus methods, and extensions to multivariable arrays and pseudo-involutions.

A Riordan array is an infinite lower-triangular matrix determined by two power series, capturing a unifying algebraic-combinatorial formalism at the intersection of combinatorics, algebra, and power series. The Riordan-array framework encompasses a group structure, functional calculus, umbral and Hopf-algebraic interpretations, generalizations to higher dimensions and multiple parameters, and deep connections to combinatorial identities and recursive structures. The following exposition presents a rigorous account of the framework, methods, and key phenomena underpinning the modern theory of Riordan arrays.

1. Definition and Algebraic Structure

Let $K$ be a commutative ring. A proper Riordan array is encoded by a pair of formal power series with $g(x) \in K[[x]]$, $g_0 \neq 0$ and $f(x) \in K[[x]]$, $f_1 \neq 0$, typically $f(0)=0$. The $(n,k)$ entry is

$$r_{n,k} = [x^n]\, g(x)\,(f(x))^k\,, \quad n,k \geq 0\,.$$

The set of all such arrays forms a group under the multiplication

$$(g,f) \cdot (h,\ell) = (g(x)\,h(f(x))\,,\, \ell(f(x)))\,,$$

with identity $(1,x)$ and inverse

$$(g,f)^{-1} = \left(\frac{1}{g(\bar f(x))},\, \bar f(x)\right)$$

where $\bar f$ is the compositional inverse of $f$. This is known as the Riordan group (Barry, 2017).

Principal subgroups include the Appell subgroup $(g,x)$, the Lagrange (associated) subgroup $(1,f)$, and the Bell subgroup $(g,\,x\,g(x))$. The group admits a standard semidirect product decomposition and associates a unique action on suitably indexed column vectors via generating functions.

2. Recurrence Systems and Generating Functions

A Riordan array encodes two canonical recurrence systems, parameterized by generating functions $A(x), Z(x)$: $$\begin{aligned} f(x) &= x A(f(x)) \qquad &(A\text{-sequence}) \ g(x) &= \frac{1}{1-x Z(f(x))} \qquad &(Z\text{-sequence}) \end{aligned}$$ Correspondingly, the entries satisfy

$$d_{n+1,k+1} = \sum_{j=0}^{n-k} a_j\, d_{n,k+j}, \quad d_{n+1,0} = \sum_{j=0}^{n} z_j\, d_{n,j}\,.$$

The fundamental theorem for Riordan arrays (FTRA) asserts that for any formal power series $h(x)$,

$(g,f) \cdot h(x) = g(x)\,h(f(x))\,,$

meaning the array acts on power series via multiplication and substitution (Barry, 2017, Burstein et al., 2021).

The array’s columns are determined by the OGF $g(x)\,f(x)^k$. The bivariate generating function is

$$B(x,y) = \sum_{n,k \geq 0} r_{n,k} x^n y^k = \frac{g(x)}{1-y f(x)}\,.$$

3. Subgroup Structure, Factors, and Half-Array Operations

The Riordan group contains important subgroups and supports canonical factorizations:

• Appell subgroup: $(g,x)$—lower-triangular Toeplitz matrices.
• Lagrange subgroup: $(1,f)$.
• Bell (hitting-time) subgroup: $(x h'(x), h(x))$ for generic $h(x)$.
• k-Bell subgroups and pseudo-involutions: Elements $(g, x g^k)$ for integer $k$ are central to the classification of involutions and pseudo-involutions (when $f(-z)=-f(z)$ and $g(-f(z))=1/g(z)$) (Burstein et al., 2021).

Half-array constructions extract and analyze "halves" of Riordan arrays:

• Vertical half: $V_{n,k} = T_{2n-k, n}$.
• Horizontal half: $H_{n,k} = T_{2n, n+k}$. Both result in new Riordan arrays with explicit formulae for their parameters in terms of compositional inverses and derivatives, leading to canonical factorizations: $V = (x \phi'(x), \phi(x)) \cdot (g(\phi(x)), x)\,,$

$$H = (x \phi'(x), \phi(x)) \cdot (g(\phi(x)) f(\phi(x)), x)$$

with $\phi(x) = \text{Rev}[x^2/f(x)]$ (Barry, 2019).

These structures generalize to r-shifted central triangles, with entries $a_{2n + r, n+k+r}$ for a base array $A = (g(t), f(t))$. Each $M^{(r)}$ is itself a Riordan array, with explicit factorization: $$M^{(r)} = (\phi'(x) f(\phi(x))^{r-1}, \phi(x)) \cdot (g(x), x f(x))$$ where $\phi(x)=\mathrm{Rev}(x f(x))$ (Barry, 2019). Consecutive shifts are related by simple multiplier matrices: $M^{(r+1)} = (f(\phi(x)), x) \cdot M^{(r)}$

4. Extensions and Generalizations

Multivariable and Multiple Riordan Arrays

The one-dimensional Riordan framework generalizes to $d$-dimensional arrays parameterized by $(G(x), F_1(x), ..., F_d(x))$, with group law

$(G, F) \cdot (H, L) = (G (H \circ F), L \circ F)$

where $F = (F_1, ..., F_d)$ (O'Farrell, 2020). Entries are indexed and constructed via multi-indices and monomial products.

A further generalization defines multiple Riordan arrays with $g(x)$ and a cyclic sequence $f_1, ..., f_\ell$ such that the $k$-th column is generated by $g(x) H(x)^q f_1(x)...f_r(x)$ with $k = q \ell + r$ (He, 5 Apr 2025). The multiple Riordan group and its semigroup structure are fully developed, supporting one $A$-sequence and $\ell$ $Z$-sequences in the characterization of array entries.

Quasi-Riordan and Almost-Riordan Arrays

Quasi-Riordan arrays are defined via matrix representations $[g, f]$ with $k$th column generated by $t^k f(t)$, extending the group law to encompass vertical recurrence structures and new transforms (He, 2022). This extends to almost-Riordan arrays, a supergroup with triples $(a(x), g(x), f(x))$, allowing more general transforms connected to Hankel and Toeplitz phenomena (Barry, 2016).

Central and One-pth Arrays

The central Riordan description considers matrices whose $(n,k)$ entry is the coefficient of $x^{n-k}$ in $g(x) f(x)^n$, shown to be equivalent to vertical halves of classical arrays with specific group multiplication and inversion formulas. This construction captures central coefficients in classical combinatorial triangles (Barry, 2020).

One-pth Riordan arrays generalize half-array notions to arbitrary $p$ via nested compositions and Lagrange inversion formulas (He, 2020), with explicit generating-function machinery for constructing new convolution identities and algebraic families (e.g., ternary and higher-ary analogues).

5. Combinatorial and Algebraic Applications

Riordan arrays encode combinatorial structures such as binomial coefficients, Catalan numbers, ballot/path models, Schröder and Motzkin numbers, and orthogonal (including Laurent biorthogonal) polynomials. The fundamental action $(g,f) \cdot h(x) = g(x) h(f(x))$ yields, for each array, a machinery for generating identities and bijections among combinatorial classes (Barry, 2017, Barry, 2013, Barry, 13 Apr 2025).

Key tools include:

• Symmetric minors and integrable models: Square symmetrizations of Riordan arrays yield principal minors equating to Robbins numbers and numbers associated to the $20$-vertex model (Barry, 2024).
• Analytic recursions: The vertical and horizontal recurrences form a dual calculus for extracting identities, with vertical relations expressing $d_{n,k}$ in terms of $g_m$ and horizontal in terms of the $A$- and $Z$-sequences.
• Hopf-Algebraic structure: The Riordan group is the character group of a bicrossed Hopf algebra $H=H_0 \otimes H_1$, bringing algebraic tools from quantum field theory (renormalization, Dyson-Schwinger equations) into the Riordan setting (Barry, 2017).
• Umbral calculus: Both classical and symbolic/abelian umbral calculus yield fundamental theorems and recurrences as direct consequences of Abel's identity, enabling complex and fractional array powers, non-recursive entry formulas, and universal vertical-diagonal recurrences (Agapito et al., 2011, Agapito et al., 2015).

6. Pseudo-involutions, Duality, and Structural Richness

Pseudo-involutions $(g, f)$ (where $f(-z) = -f(z)$ and $g(-f(z)) = 1/g(z)$) are classified via palindromic generating functions and k-Bell subgroups, with combinatorial implications and B-functions (column recursions) expressible through Chebyshev polynomials (Burstein et al., 2021, Marshall et al., 2021). Duality phenomena manifest as interplay between k-Bell pseudo-involutions and their Bell-group "aerated" cousins, with explicit duality relations between their B-functions.

Recursive matrices and dual Riordan arrays, constructed from doubly infinite extensions and anti-transposed submatrices, encapsulate further algebraic symmetries. The reversion of power series is structurally central, yielding convolutional recurrences for families such as the Patalan and super-Patalan numbers (Richardson, 2016).

7. Outlook, Generality, and Theoretical Unification

The Riordan-array framework unifies a broad universe of combinatorial arrays, algebraic operations, and functional transformations, providing a mature theory that bridges enumerative combinatorics, formal group laws, umbral and symbolic methods, representation theory, orthogonal polynomials, and mathematical physics. Generalizations to higher-dimensions, multiple parameters, and nonstandard ring settings are fully developed and actively extended (O'Farrell, 2020, He, 5 Apr 2025).

Layered constructions (half-arrays, central arrays, one-pth arrays) and Hopf-algebraic, umbral, and group-theoretic methodologies position Riordan arrays as a fundamental organizing structure. The toolkit encompasses summation and convolution identities, group dualities, matrix factorizations, combinatorial bijections, and connections to special functions, revealing the Riordan-array calculus as a canonical "language of structure" for an exceptionally broad class of algebraic and combinatorial phenomena.