---
title: 'Riordan Group: Theory and Extensions'
url: https://www.emergentmind.com/topics/riordan-group
type: topic
---

# Riordan Group: Theory and Extensions

Searching arXiv for recent and foundational papers on the Riordan group, including structure, involutions, higher-dimensional variants, and generalizations.
The Riordan group is the group of infinite lower-triangular matrices encoded by pairs of formal power series, or equivalently a group of weighted composition operators on generating functions. In the classical setting over a field \(K\) of characteristic \(0\), its elements are pairs \((g(x),\phi(x))\) with \(g\) multiplicatively invertible and \(\phi\) compositionally invertible, and the group law combines formal multiplication with formal composition. This framework is central in enumerative combinatorics, but it also supports Hopf-algebraic, Lie-theoretic, profinite, and multivariate extensions [1706.01323].

## 1. Definition and operator model

Let \(F=K[[x]]\). Two basic groups are
\[
F_0=\{g(x)\in F:g_0\neq 0\},
\qquad
F_1=\{\phi(x)\in F:\phi(x)=\phi_1x+\phi_2x^2+\cdots,\ \phi_1\neq 0\}.
\]
A Riordan array is a pair \((g(x),\phi(x))\in F_0\times F_1\), equivalently an infinite lower-triangular matrix \(R=(d_{n,k})_{n,k\ge 0}\) with entries
\[
d_{n,k}=[x^n]\,g(x)\phi(x)^k.
\]
Because \(\phi(x)^k\) starts with \(x^k\), one has \(d_{n,k}=0\) for \(n<k\), and the diagonal entries are nonzero. The group law is
\[
(g,\phi)\cdot(u,\psi)=\bigl(g\cdot(u\circ \phi),\ \psi\circ \phi\bigr),
\]
with identity \((1,x)\) and inverse
\[
(g,\phi)^{-1}=\left(\frac{1}{g\circ \bar\phi},\,\bar\phi\right),
\]
where \(\bar\phi\) is the compositional inverse of \(\phi\). In the literature, the same objects also appear in \((d,h)\) notation and as \(T(f\mid g)\) [1706.01323][1803.06872].

The operator interpretation is fundamental. The pair \((g,\phi)\) acts on a formal power series \(A(x)\) by
\[
A(x)\longmapsto g(x)\,A(\phi(x)).
\]
Matrix multiplication coincides exactly with composition of these weighted composition operators. At the level of sets, the Riordan group is a semidirect product
\[
R \cong F_0 \ltimes F_1,
\]
with exact sequence
\[
1 \longrightarrow F_0 \longrightarrow R \longrightarrow F_1 \longrightarrow 1.
\]
This already exhibits the two structural ingredients of Riordan theory: multiplicative deformation through \(g\), and substitution through \(\phi\) [1706.01323].

## 2. Canonical subgroups and internal decompositions

Several natural subgroups organize the internal structure of the Riordan group. The Appell, or Toeplitz, subgroup is
\[
\mathcal A=\{(g(x),x)\},
\]
the Associated, or Lagrange, subgroup is
\[
\mathcal L=\{(1,\phi(x))\},
\]
and the Bell subgroup is
\[
\mathcal B=\{(g(x),xg(x))\}.
\]
The Appell subgroup is normal, \(\mathcal A\cong F_0\), \(\mathcal L\cong F_1\), and every Riordan array factors as
\[
(g,\phi)=(g,x)\cdot(1,\phi),
\]
so \(R\cong \mathcal A\rtimes \mathcal L\). A distinguished subgroup is the monic Riordan group \(R^{(1)}\), where
\[
g(x)=1+g_1x+g_2x^2+\cdots,
\qquad
\phi(x)=x+\phi_2x^2+\cdots;
\]
its matrices have diagonal entries all equal to \(1\), and over a field of characteristic \(0\) it is a closed subgroup of the prounipotent group of lower triangular matrices with all diagonal entries \(1\), hence a Lie subgroup in the pro-Lie sense [1706.01323].

This decomposition also clarifies why certain classical families recur throughout the subject. The Appell subgroup captures pure multiplicative transforms, the Associated subgroup captures pure substitutions, and the Bell subgroup interpolates between the two. In later work, these same patterns reappear in generalized, double, triple, and multivariate Riordan-type groups, where one modifies either the substitution side, the coefficient ring, or the matrix indexing set.

## 3. Commutator structure, involutions, and reversibility

An element \((g,f)\) is an involution if
\[
(g,f)^2=(1,x),
\]
equivalently
\[
f(f(x))=x,
\qquad
g(x)\,g(f(x))=1.
\]
In the inverse-limit approach to finite Riordan groups \(R_n\), involutions admit an inductive construction level by level, and every nontrivial involution can be parametrized by an arbitrary power series. A concrete constraint is that if \(T(f\mid g)\) is an involution and \(g(x)=\sum_{n\ge 0}g_nx^n\), then \(g_2=0\); in \((d,h)\) notation this becomes \(h_2=h_1h_3\). The subset
\[
\mathcal L_0=\{T(f\mid g)\in R:g_2=0\}
\]
is a subgroup. The commutator subgroup is
\[
[R,R]=\{(d,h)\in R:d_0=1,\ h_1=1\},
\]
and every element of \([R,R]\) is itself a commutator. Consequently,
\[
R/[R,R]\cong K^\ast\times K^\ast.
\]
For the subgroup generated by involutions, if \(J\) denotes the set of Riordan involutions, then every element of \(\langle J\rangle\) is a product of at most four involutions, this bound is sharp for \(R_n\) with \(n\ge 2\) and for \(R\), and
\[
\langle J\rangle \cong [R,R]_0\rtimes K,
\qquad
[R,R]_0=[R,R]\cap \mathcal L_0,
\]
where \(K\cong \mathbb Z_2\times \mathbb Z_2\) is the Klein four-group of diagonal involutions [1803.06872].

The derived series is also explicit. If \(R^{(n)}\) denotes the \(n\)-th commutator subgroup, then
\[
R^{(n)}=\left\{(d,h)\in R:
d\in 1+x^{2^n-n}K[[x]],
\ h\in x+x^{2^n}K[[x]]
\right\},
\qquad n\ge 1,
\]
and every element of \(R^{(n)}\) is a commutator of elements in \(R^{(n-1)}\). For the Associated subgroup \(A=\{(1,h)\}\), one has
\[
A^{(n)}=G_{2^n},
\qquad
G_k=\{(1,h):h(x)\in x+x^kK[[x]]\},
\]
so the first nontrivial term of the substitution component doubles in degree along the derived series [2108.00486].

A later refinement studies reversibility. An element is reversible if it is conjugate to its inverse, and strongly reversible if it is a product of two involutions. In the Riordan group, not every reversible element is strongly reversible, and every product of two Riordan involutions is of the form \(\pm C\), where \(C\) is a commutator. This separates reversibility from involution length in a way that does not collapse to the classical linear-group pattern [2601.11187].

## 4. Classical arrays, transformations, and pseudo-involutions

The most familiar Riordan array is the Pascal matrix
\[
\left(\frac{1}{1-x},\,\frac{x}{1-x}\right),
\]
whose entries are \(\binom{n}{k}\). More generally,
\[
\left(\frac{1}{1-ax},\,\frac{x}{1-ax}\right)
\]
forms a one-parameter subgroup, with multiplication corresponding to addition of the parameter \(a\). The Riordan action on a generating function \(A(x)\) gives
\[
B(x)=g(x)A(\phi(x)),
\]
so the Pascal array implements the binomial transform. Another standard example is
\[
(1-x,\ x(1-x)),
\]
whose inverse is
\[
(c(x),\ xc(x)),
\qquad
c(x)=\frac{1-\sqrt{1-4x}}{2x},
\]
placing the Catalan generating function inside the Bell subgroup. These examples illustrate the basic Riordan philosophy: lower-triangular matrices encode structured transforms of generating functions [1706.01323].

Pseudo-involutions form a related but distinct class. With
\[
M=(1,-z),
\]
a Riordan matrix \(L\) is a pseudo-involution if \(LM\) has order \(2\). Equivalently, \((g(z),-f(z))\) is a pseudo-involution if
\[
g(z)\,g(-f(z))=1,
\qquad
-f(-f(z))=z.
\]
Such elements satisfy the simple inverse formula
\[
A^{-1}=MAM.
\]
Two complementary constructions are available. First, if \(g(z)=1+g_1z+g_2z^2+\cdots\) with \(g_1\neq 0\), then there exists a unique \(f(z)\) such that \((g(z),f(z))\) is a pseudo-involution. Second, if \(-f(z)\) has compositional order \(2\), then the set of all \(g(z)\) such that \((g(z),f(z))\) is a pseudo-involution is an infinite subgroup of \(\mathbb C[[z]]\) under multiplication. Explicit families in the paper have first columns given by the modified Lucas numbers, the Fibonacci numbers, and the convolved Fibonacci numbers [2105.12671].

Barry also showed that pseudo-involutions generate involutions systematically. For every pseudo-involution \(P\) and every Riordan element \((g(x),f(x))\),
\[
(g(x),f(x))^{-1}\cdot P\cdot (g(-x),f(-x))
\]
is an involution. Applied to Riordan arrays arising from orthogonal polynomials, this yields a three-parameter family of involutions indexed by \(r,s,t\), and the resulting row sums admit continued-fraction descriptions tied to the corresponding three-term recurrences [2207.02719].

## 5. Hopf-algebraic, Lie, and profinite viewpoints

Barry’s Hopf-algebraic treatment recasts the Riordan group through coordinate rings. The coordinate ring of the multiplicative group \(F_0\) gives a Hopf algebra \(H_0\), the coordinate ring of the composition group \(F_1\) gives a Hopf algebra \(H_1\), and the Riordan group appears as a group of characters on the corresponding coordinate algebras. At the level of coordinate rings,
\[
C(R)\cong H_0\otimes H_1.
\]
This viewpoint is one reason the Riordan group can serve as a toy model for renormalization in scalar field theory. The same paper also emphasizes inverse limits of finite Riordan groups \(R_n\), the Lie-theoretic nature of the monic subgroup, and the relation with formal diffeomorphisms and Faà di Bruno type Hopf structures [1706.01323].

Over finite fields, the Riordan group becomes a profinite object. For a finite field \(\mathbb F_q\), one has
\[
R(\mathbb F_q)=H(\mathbb F_q)\rtimes N(\mathbb F_q),
\]
where \(N(\mathbb F_q)\) is the Nottingham group. The groups \(R(\mathbb F_q)\) are profinite, and for characteristic \(p\) they are pro-\(p\). They are topologically finitely generated, and for \(p>2\) their lower central series satisfies
\[
\gamma_n(R(\mathbb F_q))=
H_{t_n}(\mathbb F_q)\rtimes N_{t_n+1}(\mathbb F_q),
\qquad
t_n=n+\left\lfloor\frac{n-2}{p-1}\right\rfloor.
\]
They also have finite width. For index-subgroups \(R(I,J)\), the Hausdorff dimension with respect to the filtration \(G_n=H_n(\mathbb F_p)\times N_n(\mathbb F_p)\) is
\[
\dim_H(R(I,J))=\tfrac12\bigl(\mathrm{ldense}(I)+\mathrm{ldense}(J)\bigr),
\]
and the index-subgroup spectrum contains the interval \([0,\tfrac12]\) together with explicitly described rational sets [2012.10895].

Characteristic \(2\) exhibits a distinct profile. The Riordan group over \(\mathbb F_2\) is a split extension of the Appell subgroup by the Nottingham group, its lower central series satisfies
\[
\gamma_n(\mathcal R(\mathbb F_2))
\cong
\mathcal A_{2n-3}\ltimes \gamma_n(\mathcal N(\mathbb F_2)),
\qquad n\ge 2,
\]
and, over any commutative ring with identity,
\[
\mathcal R^{ab}(\mathbb D)\cong \mathbb D^\*\times \mathbb D\times \mathcal N^{ab}(\mathbb D).
\]
This isolates the Appell contribution and the substitution-group contribution inside the abelianization [2511.21639].

## 6. Multivariate, higher-order, and generalized Riordan families

In higher dimensions, O’Farrell replaces one-variable series by
\[
F_d=K[[x_1,\dots,x_d]]
\]
and invertible formal maps \(g=(g_1,\dots,g_d)\). The \(d\)-dimensional Riordan group is
\[
R_d(K)\cong F_d^\times\rtimes G_d^{\mathrm{op}},
\]
with multiplication
\[
(f,g)(f',g')=\bigl(f\cdot(f'\circ g),\ g'\circ g\bigr).
\]
Its matrices are indexed by monomials \(m,n\) in \(x_1,\dots,x_d\), with
\[
M(f,g)_{m,n}=[m]\bigl(f\cdot(n\circ g)\bigr).
\]
The multivariate Fundamental Theorem states that this matrix representation is a semigroup homomorphism, so the operator \(u\mapsto f\cdot(u\circ g)\) remains the governing mechanism in several variables [2009.14130].

A recent multivariate example is the multivariate Pascal matrix. If \(L\) denotes the matrix of multidimensional binomial coefficients, then for every integer \(p\),
\[
L^p=\mathcal R\bigl(G(\mathbf z),X_1(\mathbf z),\dots,X_n(\mathbf z)\bigr),
\qquad
G(\mathbf z)=\prod_{j=1}^n\frac{1}{1-pz_j},
\qquad
X_i(\mathbf z)=\frac{z_i}{1-pz_i}.
\]
In particular, the infinite multivariate Pascal matrix itself is an element of the multivariate Riordan group [2605.08892].

Another direction allows the first component to be a formal semi-Laurent series. The generalized Riordan group
\[
RL_1(\mathbb K)=\{(g,f): g\in \mathcal L_s(\mathbb K)\setminus\{0\},\ \operatorname{ord}f=1\}
\]
retains the classical product
\[
(g,f)*(h,k)=\bigl(g\cdot(h\circ f),\ k\circ f\bigr),
\]
and the classical Riordan group is precisely the order-zero layer \(RL_1^0\). These groups are genuinely new: \(RL_1(\mathbb C)\) is not isomorphic to \(RL_1^0(\mathbb C)\). The paper also gives \(RL_1(\mathbb C)\) an infinite-dimensional Lie group structure modeled on \(s_L\times s\), with Lie algebra
\[
rl_1(\mathbb C)=\mathcal L_s(\mathbb C)\times z\mathbb X(\mathbb C)
\]
and bracket
\[
[(\gamma_1,\kappa_1),(\gamma_2,\kappa_2)]
=
\bigl(\gamma_2'\kappa_1-\gamma_1'\kappa_2,\ \kappa_2'\kappa_1-\kappa_1'\kappa_2\bigr)
\]
[2509.04160].

Higher-order Riordan-type groups extend the pair \((g,f)\) to tuples. The Double Riordan group contains a subgroup isomorphic to the ordinary Riordan group via
\[
(g,f)\longmapsto \left(g(z^2),\,z,\,\frac{f(z^2)}{z}\right),
\]
the type-1 almost Appell subgroup [2410.10709]. The Triple Riordan group uses 4-tuples \((g,f_1,f_2,f_3)\) with \(g\in \mathbf R[[x^3]]\) and \(f_i\in x\mathbf R[[x^3]]\), and group law
\[
(g,f_1,f_2,f_3)\cdot(G,F_1,F_2,F_3)
=
\bigl(g\,G(h),\ f_1F_1(h),\ f_2F_2(h),\ f_3F_3(h)\bigr),
\qquad
h=(f_1f_2f_3)^{1/3}
\]
[2412.05461]. The Sprugnoli group is another recent extension: its elements are triples \((g,f_1,f_2)\) with \(g\in F_0\), \(f_1\in F_1\), and \(f_2\in F_1\cap xR[[x^2]]\), its columns are built from sequence bisections and vertically stretched Riordan arrays, and it admits a production-matrix characterization with two alternating \(A\)-type sequences [2605.16633].

Taken together, these constructions show that the Riordan group is both a concrete combinatorial tool and a flexible algebraic template. Its classical form remains the reference point, but modern work treats it as one member of a broader family of power-series matrix groups governed by multiplication, substitution, and increasingly elaborate symmetry conditions.

Source: https://www.emergentmind.com/topics/riordan-group