pplications

numerative
A
ombinatorics

[Link]
Enumerative Combinatorics and Applications ECA 4:3 (2024) Article #S2S1
[Link]

An Invitation to the Riordan Group

Dennis E. Davenport† , Shakuan K. Frankson† , Louis W. Shapiro† , and Leon C. Woodson‡

† Department of Mathematics , Howard University

Email: [Link]@[Link], [Link]@[Link], lshapiro@[Link]

‡ Departmentof Mathematics, Morgan State University

Email: [Link]@[Link]

Received: September 23, 2023, Accepted: March 5, 2024, Published: March 22, 2024
The authors: Released under the CC BY-ND license (International 4.0)

Abstract: The Riordan group is a group of infinite lower triangular matrices that are defined by two generating
functions, g and f . The k th column of each matrix has a generating function related to gf k . There are many
applications of Riordan arrays, among other things they can be used to count combinatorial objects and to
prove combinatorial identities.

Keywords: Riordan arrays; Generating Functions

2020 Mathematics Subject Classification: 05A05, 05A10, 05A15

1. Introduction
The original idea for the Riordan group evolved from a set of combinatorial examples and from Rota’s umbral
calculus. We could prove various identities very neatly using what is now called the fundamental theorem of
Riordan arrays (FTRA) [42]. This also gives us the group structure. Since many of our identities came from
John Riordan’s famous book, ”Combinatorial Identities” [39] and he had recently passed away, we named the
group after him. The group structure allowed us a systematic way to invert identities which had been a major
theme in his book. Several years later in 1994 an important paper by Renzo Sprugnoli [46] appeared which
widely extended the range of examples using the Riordan group framework.
Following this, there were many talented other contributors from Italy. The next group to contribute was
led by Gi-Sang Cheon and a strong group of mathematicians from South Korea. In the ensuing years, hundreds
of papers have been published with contributors from all over the world.
A valuable bibliography through 2016 is due to Sprugnoli [47]. Typing in ”Riordan array” on your browser
will lead to several hundred articles. An introductory text is Paul Barry’s ”Riordan Array Primer” [6] and
a research-level book is ”The Riordan Group and Applications” [41]. There is also an annual international
conference and some YouTube videos [53, 54].
This survey article on Riordan arrays is divided into six sections, not counting the introduction and the
conclusion. Section 2 is on Preliminary Results and includes the Fundamental Theorem of Riordan Arrays,
the Riordan group, and using A− and Z−sequences to construct Riordan arrays. In Section 3, there are some
interesting applications of Riordan arrays, such as counting paths and trees. Section 4 discusses subgroups
and the relatively new result (2021) by A. Luzón, M. Morón and L. Felipe Prieto-Martı́nez describing the
commutator subgroup of the Riordan group, [29]. There are many types of Riordan arrays. In Sections 2-4, we
consider ordinary Riordan arrays. In Section 5, we write about the well-known exponential Riordan arrays and,
in Section 6, we write about generalized Riordan arrays and double Riordan arrays. Section 7 is on involutions
and pseudo-involutions. Most of Subsection 7.1 is new.

2. Preliminary Results
There is a folklore saying in the mathematics community to the effect that,
“Combinatorial identities are pretty boring except for the one that you are working on.”
 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Here is an identity that we will use to introduce many of the ideas of the Riordan group. One item of interest
here is that the right-hand side is always 2, for n ≥ 2.
X n n − k  k
(−1) 22n−2k = 2
n−k k
k≥0

As a first step, we try out a few small values of n. For instance if n = 6 we get
       
6 6 6 5 6 4 6 3
64 − 16 + 4− = 2.
6 0 5 1 4 2 3 3
For the moment ignore the negative signs and the powers of 2 and we get the coefficients 1, 6, 9, 2.
Similarly n = 5 gives us 1, 5, 5 and n = 4 yields 1, 4, 2. Collecting the lower terms give us the table
 
1
1 
 
1 2 
 
1 3 .
 
1 4 2 
 
1 5 5 
1 6 9 2

For the second step write the identity as a matrix times a column vector. We want a triangular matrix and
would like 1”s on the main diagonal so we get
    
1 0 0 0 0 0 0 1 1
0 1 0 0 0 0 0  2  2
    
−2 0 1 0 0 0 0  4  2
   

 0 −3 0 1 0 0 0  8  = 2 .
   

2
 0 −4 0 1 0 0 16
 
 2
 
0 5 0 −5 0 1 0 32
   2
−2 0 9 0 −6 0 1 64 2

We can only display a few rows but we have here an infinite lower triangular matrix and infinite column
vectors.
For the third step, we find the generating functions for the columns of the matrix and the two-column
vectors.
2
The left most column (the zeroth column) has the generating function 1−z
1+z 2 := g. The next column has the
2 2
1−z 2 1−z z
generating function z (1+z 2 )2 and the next is z (1+z 2 )3 . If we let f = 1+z 2 then the columns have the generating
2 1
functions g, gf, gf , ... The column vector for the powers of 2 is and the column of almost all 2’s has the
1−2z
1+z
generating function 1−z. .
For the fourth step, we look at the matrix times column vector multiplication in terms of generating functions
and we have
2 3
1 − z2 1 − z2 1 − z2 1 − z2
 
z z z
1· + · ·2+ · ·4+ · · 8 + ···
1 + z2 1 + z2 1 + z2 1 + z2 1 + z2 1 + z2 1 + z2
( 2 3 )
1 − z2
 
z z z
= 1 + · 2 + · 4 + · 8 + ···
1 + z2 1 + z2 1 + z2 1 + z2
1 − z2 1 1 − z2 1 + z2 1+z
= 2
·  = 2
· 2
= as desired.
1+z 1−2 z
2
1 + z 1 − 2z + z 1−z
1+z

T 1
We can generalize what we just did by replacing the column vector [1, 2, 4, 8, · · · ] ←→ 1−2z with [a0 , a1 , a2 ,
T 1+z
· · · ] ←→ A (z) and [1, 2, 2, 2, · · · ] ←→ 1−z with [b0 , b1 , b2 , · · · ] ←→ B (z).
Now our computation becomes
   
a0 b0
a1  b1 
g, gf, gf 2 , · · · a2  = b2 
     
.. ..
   
. .
a0 g + a1 gf + a2 gf 2 + · · · = B (z)

ECA 4:3 (2024) Article #S2S1 2

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

g a0 + a1 f + a2 f 2 + a3 f 3 + · · ·


=
gA (f ) = B (z)
or g (z) A (f (z)) = B (z) .
This modest observation is very important and is called the Fundamental Theorem of Riordan Arrays
which is abbreviated as the FTRA. More formally we have the following, which was first stated by Shapiro,
Getu, Woan, and Woodson in [42]. But first, we need some notation.
P∞ P∞
Definition 2.1. Let g(z) = k=0 gk z k and f (z) = k=1 fk z k , where g0 6= 0 and f1 6= 0. Let dn,k be the

k
coefficient of z n
in g(z)
f (z) . Then D = (dn,k )n,k≥0 is a Riordan array. We denote the Riordan array D by
D = g(z), f (z) = g, f .
P∞
Theorem 2.1. (The Fundamental Theorem of Riordan Arrays): Let A(z) = k=0 ak z k and B(z) =
P∞ k T T
k=0 bk z and let A and B be the column vectors A = (a0 , a1 , a2 , · · · ) and B = (b0 , b1 , b2 , · · · ) . Then
(g, f )A = B, if and only if B(z) = g(z)A(f (z)).
As a simple application, we can use FTRA to find row sums of Riordan arrays. For instance, going back to
our example.
First, the row sums of the matrix without the minus signs are the Lucas numbers. To find row sums let
1 T 1+z 2 z
A (z) = 1−z ←→ [1, 1, 1, · · · ] . Without the minus signs we will have g = 1−z 2 and f = 1−z 2 . For the Lucas
2
1+z 2 3 4 5 6
numbers, we have the generating function B (z) = 1−z−z 2 = 1 + z + 3z + 4z + 7z + 11z + 18z + · · · .

By the FTRA we have

1 + z2 1 + z2
 
z 1 1
2
, 2
= ·  
1−z 1−z 1−z 1 − z2 1 − z
2 1−z
2
1+z 1
= · 1−z−z 2
1 − z2 1−z 2
1 + z2
=
1 − z − z2
= B (z) .
There is another identity we can treat. Returning the minus signs and taking row sums we have,
1 − z2 1 − z2
 
z 1 1
2
, 2
= 2 z
1+z 1+z 1−z 1 + z 1 − 1+z 2

1 − z2 1
= 1−z+z 2
1 + z2 1+z 2
2
1−z
=
1 − z + z2
 1 − z − 2z 2 − z 3 + z 4 + z 5 
= 1+z .
1 − z6
Thus the sums after the first are a repeating six-cycle 1, −1, −2, −1, 1, 1. This identity was a problem on the
1932 Tripos exam and appeared in Hardy’s Pure Mathematics on page 445, [21].
What happens if we multiply one Riordan array by a second (G, F )? The k th column of (G, F ) has the
generating function GF k and we can apply the FTRA to express (g, f ) GF k = gG (f ) F k (f ) .
Since this applies for each column of (G, F ) we have a multiplication rule for Riordan arrays.
(g, f ) ∗ (G, F ) = (gG (f ) , F (f )) .
Now suppose that f (z) = f1 z + f2 z 2 + f3 z 3 + · · · and that f1 6= 0. Then f is a unit element in the domain of
formal power series and has a compositional inverse which we denote as f (z) or as f .
With the requirements that g0 6= 0 and f1 6= 0 we now have a group structure and we call this group the
Riordan group, which will be denoted by R. The identity is the usual matrix identity and is (1, z) . For the
inverse we have ! !
−1 1 1
(g (z) , f (z)) =
, f (z) =
, f .
g f (z) g f
Theorem 2.2. Let (g, f ) and (G, F ) be two Riordan arrays. Then the operation *, given by (g, f ) ∗ (G, F ) =
(g(z)G(f (z)), F (f (z))) is the usual matrix multiplication
  which is an associative binary operation, (1, z) is the
1
identity element and the inverse of (g, f ) is g (f )
,f .

ECA 4:3 (2024) Article #S2S1 3

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Using FTRA we can easily prove many combinatorial identities and the group structure gives a systematic
way to invert identities.
The Riordan Group has many interesting and important subgroups. The set of all elements (g, f ), such
that g is an even function (with nonzero constant term) and f is an odd function, is called the Checkerboard
Subgroup. The terminology comes from the fact that (g, f ) has the appearance of a checkerboard. We also
say that a generating function or an array is aerated if it has alternating zeros. Other subgroups are given in
Section 3.
Given any Riordan array, every element of the array that is not in the leftmost column can be written as a
linear combination of elements in the row that is directly above starting from the preceding column, see [40].
In addition, every element in the zeroth column other than the zeroth element can be expressed as a linear
combination of the elements from the row that is directly above, see [40] and [33]. Hence, we can construct
Riordan arrays by using the rows of the matrix. The following theorem tells us how to construct a Riordan
array using its rows.
Theorem 2.3. Let D = (dn,k ) be an infinite lower triangular matrix. Then D is a Riordan array if and only
if there exist two sequences A = {ai }∞ ∞
i=0 and Z = {zi }i=0 with a0 6= 0 and z0 6= 0 such that

∞
X
dn+1,k+1 = aj dn,k+j ; k, n = 0, 1, 2, ... (1)
j=0
X∞
dn+1,0 = zj dn,j ; n = 0, 1, 2, ... (2)
j=0

The entry d0,0 is given and must be nonzero, but we often set it equal to 1. The sequences {ai }∞
i=0 and
{zi }∞
i=0 , respectively, are called the A−sequence and Z−sequence of the Riordan array D.
The following theorem shows us how to find the A− and Z−sequences of a Riordan array if we know the
generating functions that determine the array.


Theorem 2.4. Let D = g(z), f (z) be a Riordan array. Let A be the generating function of the A−sequence
and Z the generating function of the Z−sequence. Then
!
z 1 1
A(z) = and Z(z) = · 1−
,
f (z) f (z) g f (z)

where f is the compositional inverse of f .

See [6] and [33] for more information about A− and Z−sequences of Riordan arrays.
Another method of finding the A− and Z−sequences of Riordan arrays is using the production (or Stieltjes)
matrix.
Definition 2.2. Let (g, f ) be a Riordan array. The production matrix or Stieltjes matrix P is given by:

P = (g, f )−1 · (g, f ),

where (g, f ) is the truncated Riordan array with the first row omitted.
Theorem 2.5. Let R
= (g, f ) be a Riordan array. Then the production matrix P for R is of the form
Z A tA t2 A ... , where Z is the generating function for the Z−sequence and A is the generating function
for the A−sequence of R.
The background needed to use the Riordan group is very modest, a little abstract algebra, an acquaintance
with generating functions, and a knowledge of some basic combinatorial objects such as the Catalan and
Fibonacci numbers and Pascal’s Triangle. The next example shows how to derive new results using some
elementary tools.

Example 2.1. Recall that given any positive integer n a Schröder path is a lattice path from the origin (0, 0)
to (2n, 0) with steps (1, 1), (1, −1), and (2, 0) that do not go below the x-axis. We start with the following figure
of Schröder paths with no level steps at odd heights.

ECA 4:3 (2024) Article #S2S1 4

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

1
1
 
1 9  0 1 

 2 0 1 

1 8 46 
 0 3 0 1 

1 6 29 132

 5 0 5 0 1 ··· 

 0 10 0 6 0 1 
 
1 5 21 86 355  15 0 21 0 8 0 1 
0 36 0 29 0 9 0 1
 
1 3 10 36 137 543 ...
1 2 5 15 51 188 731

This yields a double Riordan array rather than a Riordan array. This is just suggestive at this point, details
are in Section 5.2. However, taking just rows at even heights or at odd heights yields two Riordan arrays which
we call the distillates. Let E be the matrix whose columns are the rows at even height and O the matrix whose
columns are the rows at odd height. Then the first few rows of each are
   
1 1
2 1   3 1 
   
5 5 1 · · ·
  10 6 1 · · ·
E= 
 , O =  36 29 9
 .
15 21 8 1 1 
  
51 86 46 11 1  137 132 57 121 1 
... ...

An obvious question is what are the g and f functions such that E = (g, f )? This problem can be solved using
relations between the functions, see [17]. There are also two tools that can be used to assist us in answering the
question, OEIS [43] and the Riordan calculator [38]. We leave it to the reader to answer this question.

3. Riordan Arrays, Trees, and Path Counts

In this section, we look at some more elementary applications of the Riordan group, look at some subgroups,
and discuss the A and Z sequences. The Catalan numbers count ordered trees and we sketch these trees for
n = 0, 1, 2, 3, and 4 edges.

1 1 2 5 14

Figure 1: Counting Ordered Trees

Let’s count vertices by height and we get the following matrix where n is the number of edges and k is the
height. As usual, we show just the first few rows of an infinite matrix. We include more rows than usual because
the k th row has interesting divisibility properties if 2k − 1 is a prime.
 
1 0 0 0 0 0 0 0
 1 1 0 0 0 0 0 0 
 
 2 3 1 0 0 0 0 0 
 
 5 9 5 1 0 0 0 0   = C, zC 2


V := 

 14 28 20 7 1 0 0 0 

 42 90 75 35 9 1 0 0 
 
 132 297 275 154 54 11 1 0 
429 1001 1001 637 273 77 13 1

ECA 4:3 (2024) Article #S2S1 5

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Similarly, we can classify leaves by height and we get a similar matrix.

 
1 0 0 0 0 0 0 0
 0 1 0 0 0 0 0 0 
 
 0 2 1 0 0 0 0 0 
 
 0 5 4 1 0 0 0 0  = 1, zC 2


L := 
 0 14
 14 6 1 0 0 0 

 0 42 48 27 8 1 0 0 
 
 0 132 165 110 44 10 1 0 
0 429 572 429 208 65 12 1

These two are connected by the equation TL = V where

 
1 0 0 0 0
 1 1 0 0 0 
 
 2 1 1 0 0 
 
 5 2 1 1 0 
T := 
 14
 = (C, z) .
 5 2 1 1 

 42 14 5 2 1 1 
 
 132 42 14 5 2 1 1 
429 132 42 14 5 2 1 1

With this example in hand, there are two directions to explore. One is subgroups of the Riordan group,
which we will explore in Section 3.
Here are the five basic equations for counting UUR trees. These are the trees where the possible up-degrees
are the same for each vertex.

TL = V (3)
0
V = (zT ) (4)
L1 = zA0 (zT ) (5)
k
Lk = (L1 ) (6)
1
L = (7)
1 − L1
The generating functions involved are as follows. The generating function for a class of trees is T = n≥0 tn z n
P
where tn is the number of such trees with n edges. We essentially are assigning a z to each edge. The generating
function V counts these same trees but with a marked vertex. Similarly, L is the generating function for these
trees but with a marked leaf. Then L1 is the generating function for these trees with a marked leaf at height
1, where A(z) is the generating function of the A-sequence. Also, Lk counts these trees with a marked leaf at
height k.
The following is a sketch of the proof for Equation (3): Consider a tree with a marked vertex v. Then cut
the tree at v to produce two smaller trees. The lower subtree is a tree with a marked leaf where v was. The
upper tree was sitting atop v. This translates to V = T L.

!
cut at a vertex v
,
lower tree with marked leaf upper tree

Here is a sketch of the proof for Equation (4): Since a tree with n edges has n + 1 vertices we want
P 0
P n n+1 0
n≥0 (n + 1) t n z = t
n≥0 n z = (zT ) .
We explain Equation (6) as follows. Obviously L = 1+L1 +L2 +L3 +· · · . But L2 = L21 and indeed Lk = Lk1 .

ECA 4:3 (2024) Article #S2S1 6

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

The following is a sketch of the proof for Equation (7): The A-sequence gives the number of possibilities for
the updegree of a vertex. The number or weight for updegree k is ak . Often it is just ak = 0 meaning that no
vertex has updegree k. Dually ak = 1 means that a vertex of updegree k is permitted. For instance a Motzkin
tree is one where each vertex can have updegree 0,1, or 2. Thus it has the A-sequence 1, 1, 1, 0, 0, 0, · · · and A
function A (z) = A = 1 + z + z 2 . Complete binary trees have the A sequence 1, 0, 1, 0, 0, 0, · · · and A = 1 + z 2 .
However incomplete binary trees can have a right or a left single edge so the A function for these is 1 + 2z + z 2 .
For ordered trees, we have T = C the Catalan number generating function. Then
 
X
n
X 1 2n n
V = (n + 1) Cn z = (n + 1) z
n+1 n
X 2n 
1
= zn = B = √ .
n 1 − 4z

By Equation (3), we have L = B/C = B+1 2 .

We have now proved two interesting results. The first is that for nontrivial ordered trees half of the vertices
are leafs. The companion equation which follows from B = 1 + 2zCB is B−1 2 = zBC = L − 1.
Next, we want to compute the total height of all the leafs of all the trees with n edges. A quick count of
the trees with 2 or 3 edges yields a total height 4 for the two trees with 2 edges and 16 for the five trees with
3 edges. The key observation is that if we pick a leaf at height k then there are k − 1 vertices between it and
the root. Pick one of these k − 1 vertices and cut the path at that point. The appropriate generating function
is L (L − 1) . But for ordered trees, we then have
B z X
L (L − 1) = · zBC = zB 2 = = 4n−1 z n .
C 1 − 4z
n≥1

The total leaf height is 4n−1 for trees with n edges. This translates directly to all Dyck paths of 2n steps. The
total height of all peaks is also 4n−1 .
We can obtain similar results for the many other types of UUR trees. We will work through one example
here and comment on a few others.
The second example could be called Traffic Light trees. We consider ordered trees where the edges above a
vertex can be either red or green. If both are present we require that all the green edges are to the left of all
the red edges. Thus a vertex of updegree k has k + P1 possibilities for the edges 2above it ranging from all green
to all red. Thus the generating function is A(z) = n≥0 (n + 1) z n = 1/ (1 − z) . Differentiating gives
−3
A0 (z) = 2 (1 − z)
2

From this it follows that f = z (1 − z) . Hence, z = f 1 − f 2 . Looking at the first few values of L1 which are
0, 2, 6, 24, 110, · · · suggest that L1 = 2g − 2 where
 
3 1 X 1 3n + 1 n
g = 1 + zg = = z ,
1 − zg 2 n≥0 3n + 1 n
the generating function for ternary numbers.P
2 3n+2

0 3n+1


It turns out that the T function is g 2 = n≥0 3n+2 z n . Thus V = (zT ) = zn.
P
n n≥0 n
1
We can express L as either V /T or as 1−L 1
where

2 2z
L1 = zA0 (zT ) = z · 3 = 3 = 2zg 3 = 2 (g − 1) = 2z + 6z 2 + 24z 3 + 110z 4 + 546z 5 + · · ·
(1 − zT ) (1 − zg 2 )
For the Traffic Light trees we have
 
1 0 0 0 0

 0 2 0 0 0 

L :=  0 6 4 0 0  = (1, 2(g − 1))
 
 0 24 24 8 0 
0 110 132 72 16

and  
1 0 0 0 0
 2 2 0 0 0 
 = g 2 , 2 (g − 1) .
 

V := 
 7 10 4 0 0 
 30 50 32 8 0 
143 260 208 88 16

ECA 4:3 (2024) Article #S2S1 7

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

We now go back to the ordered trees and ask what the average number of leafs at height 1. Since L1 = zC 2
we have
1 2n


[z n ] zC 2 n+1 n
= 1 2n
= 1 for n ≥ 1 since C = 1 + zC 2 .
[z n ] C n+1 n

Thus, for all trees with n edges, there is on average one leaf of height 1 per tree. In terms of Dyck paths, the
number of hills, which are subsequences U D starting and ending on the x-axis is on average 1.
By way of contrast, we have Motzkin trees where every vertex can have updegree 0, 1, or 2.
The generating function counting Motzkin trees is denoted m (z) or m. The defining equation for this
generating function is√ m = 1 + zm + z 2 m2 . The three terms correspond to the root having degree 0, 1, or
1−2z−3z 2
2. Thus m = 1−z− 2z = 1 + z + 2z 2 + 4z 3 + 9z 4 + 21z 5 + 51z 6 + · · · = n
P
p 2 n≥0 mn z . Note that
2
2z m = 1 − z − (1 + z) (1 − 3z) so the radius of convergence is 1/3 where 1 − 3z = 0 occurs. The ratio test
then says that lim mmn+1n
= 3. The A-sequence is 1, 1, 1, 0, 0, 0, ... and thus A (z) = 1 + z + z 2 , which implies
n→∞
A0 (z) = 1 + 2z. With the L1 equation we get

L1 = zA0 (zm) = z (1 + 2zm)

Then [z n ] (z (1 + 2zm)) = 2mn−2 for n ≥ 2
 2
2mn−2 mn−2 mn−1 1 2
Hence = 2· · →2· = .
mn mn−1 mn 3 9

So, the average number of leaves at height 1 approaches just 2/9.

To check this we see that
2m10 2 · 2188
= ' 0.282 12
m12 15511
2m20 50 852 019
= 2· ' 0.253 78
m22 400 763 223
showing the convergence, albeit slowly to 2/9 ' 0.22222.
Note that for the Traffic Light trees
1 3n+1


[z n ] L1 [z n ] 2 (g − 1) 2 · 3n+1 n 2n + 2 2
n
= n 2
= 2 3n+2

= → .
[z ] T [z ] g 3n+2 2
3n + 1 3

4. Subgroups
Many subgroups of R have been studied, both for their combinatorial and algebraic properties. For example,
the Bell subgroup, given by {(g, f ) ∈ R : f = zg} = {(g, zg)}, the Associated subgroup, given by {(g, f ) ∈ R :
g = 1} = {(1, f )}, and the Appell subgroup, given by {(g, f ) ∈ R : f = z} = {(g, z)}. Note that for all functions
f and g, (g, f ) = (g, z) ∗ (1, f ). Thus R is the semidirect product of the Associated and Appell subgroups.
Here is a small list of subgroups.

Typical Element Name

(g, zg) Bell subgroup
(1, f ) Associated subgroup
(g, z) Appell subgroup
(f 0 , f ) 
 Derivative subgroup
zf 0
f ,f

Hitting time subgroup
even
g , f
odd Checkerboard subgroup
g , zg b
a
(a, b)-Bell subgroup

The Bell subgroup is an (a, b)-Bell subgroup with a = 1 and b = 1. Another common well-known Riordan
group subgroup that is used often in this article is the 2−Bell subgroup which is an (a, b)-Bell subgroup with
a = 1 and b = 2. The Pascal matrix is obviously in the Bell subgroup and less obviously in the Hitting time
subgroup. The
  matrix L is in the Associated subgroup. The matrix V is in the 2-Bell subgroup as is the matrix
1 z
,
1−z (1−z)2 from section one.
Of these seven subgroups, the only one that is a normal subgroup is the Appell subgroup. The verifications
of the subgroups mentioned here are subgroups is usually straightforward. Here is an example.
Theorem 4.1. The derivative subgroup is indeed a subgroup.

ECA 4:3 (2024) Article #S2S1 8

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

d
Proof. Since dz (z) = 1 The identity (1, z) is in the subgroup.
To show closure we multiply (f 0 , f ) and (F 0 , F ) and their product is (f 0 , f ) (F 0 , F ) = (f 0 F 0 (f ) , F (f )) which
is in the subgroup since
d
(F (f (z))) = F 0 (f (z)) · f 0 (z) .
dz
Similarly, !
d −1 1
f (z) = 1/f 0 f (z) so (f 0 , f ) =
, f (z) = f (z)0 , f (z) .




dz 0
f f (z)

The other subgroup verifications are similar or even less involved and are left to the reader.
A natural question to ask is, what is the commutator subgroup of R? In 2021, Ana Luzón, Manuel Morón,
and L. Felipe Prieto-Martı́nez proved the following, see [29].
Theorem 4.2. The commutator subgroup of R, denoted by [R, R], is given by [R, R] = {(g, f ) ∈ R : g0 =
1 and f1 = 1}. Moreover, every element in [R, R] is a commutator.

Hence, the elements in [R, R] are those Riordan arrays with 1s down the main diagonal.

5. Exponential Riordan Arrays

The other major class of examples and applications of Riordan arrays involve exponential generating functions.
n
We now associate a sequence (a0 , a1 , a2 , ...) with the generating function A(z) = n≥0 an zn! .
P
n n n
If A(z) = n≥0 an zn! and B(z) = n≥0 bn zn! are multiplied, then C(z) = A(z)B(z) = n≥0 cn zn! , where
P P P
Pn
cn = k=0 nk ak bn−k .


P∞ n P∞ n
Let g(z) = g = n=0 gn zn! , with g0 6= 0 and f (z) = f = n=1 fn zn! , with f1 6= 0. Then the exponential
g(z)(f (z))k
Riordan array they generate is denoted by [g, f ], where the k th column has the generating function k! .
Here are five well-known exponential Riordan arrays.
 
1
1 1 
z
 
1. [e , z] = 1 2 1
 

1 3 3 1
...
The Pascal matrix is a rare example of an array that is both an ordinary Riordan array and an exponential
Riordan array.
 
1
0 1 
 
  
0 1 1 

2. [1, ez − 1] = S(n, k) = 0 1 3 1 
n,k≥0
 
0 1 7 6 1 
 
0 1 15 25 10 1
...
The Stirling numbers of the second kind S(n, k) count the number of ways to partition a set with n
elements into k nonempty blocks. The row sums of the array are the Bell numbers and S(n, k)k! counts
the number of onto functions from an n − set to a k − set.
 
1
0 1 
 
h  i   
0 1 1 

1
3. 1, ln 1−z = s(n, k) = 0 2 3 1
 
n,k≥0

0 6 11 6 1 
 
0 24 50 35 10 1
...
These are the signless Stirling numbers of the first kind s(n, k). These count the number of permutations
in Sn whose disjoint decomposition has k cycles.

ECA 4:3 (2024) Article #S2S1 9

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

 
1
0 1 
 
1 0 1 
 
2
4. [D(z), z] =  3 0 1 

9 8 6 0 1 
 
44 45 20 10 0 1
...
n
e−z
P∞
D(z) = n=0 dn zn! = 1−z , is the derangement series. Recall that a derangement is a permutation with
no fixed points and (dn )n≥0 = (1, 0, 1, 2, 9, 44, ...). The k th column, for k = 1, 2, 3, ... counts permutations
with k fixed points.
5. The telephone exchange.
Let Tn be the number of ways that n people can be off the phone or on with one other person. If k counts
the number of people not engaged in a phone conversation, then we get the following exponential Riordan
array
 
1
0 1 
 
1 0 1 
 
z 2  0 3 0 1 
[e 2 , z] = 
 
3 0 6 0 1 

 0 15 0 10 0 1 
 
15 0 45 0 15 0 1
...
The row sums (1, 1, 2, 4, 10, 26, 76, ...) = (Tn )n≥0 . Tn is also the number of solutions of π 2 = ι in the
symmetric group on n letters.
There is an FTRA for exponential Riordan arrays where we get a group structure. The identity element is [1, z]
and " # " #
−1 1 1
[g (z) , f (z)] =
, f (z) =
, f .
g f (z) g f
The subgroup definitions are unchanged as well.
Given an exponential generating function [g, f ], the generating function for the row sums is [g, f ]ez =
g(z)ef (z) .

Example 5.1. The row sums of the Stirling numbers of the first kind start 1, 1, 2, 6, 24, and the assumption is
that we are looking at (n!)n≥0 . To prove this we have
 
1
 X zn
ez = 1 · eln( 1−z ) =
1
1, ln n! .
1−z n!
n≥0

6. Other Types of Riordan Arrays

We have introduced both ordinary Riordan arrays and exponential Riordan arrays. In this section, we will now
generalize these ideas. Given a sequence r = (r0 , r1 , ...), we will define generating functions g(z; r) and f (z; r).
The Riordan array defined by these generating functions will be called an r-Riordan array and the sequence r
will be called a denominator sequence. Much of Section 6.1 can be found in T. Wang and W. Wang [51] and L.
Woodson [52].

6.1 r - Riordan Arrays

Let r = (r0 , r1 , r2 , ...), where r0 = 1, rn 6= 0 for n ≥ 1. Then we define
∞
z z2 z3 X zn
g(z; r) = a0 + a1 + a2 + a3 + · · · = an
r1 r2 r3 n=0
rn

and
∞
z z2 z3 X zn
f (z; r) = b1 + b2 + b3 + · · · = bn .
r1 r2 r3 n=1
rn

ECA 4:3 (2024) Article #S2S1 10

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Define the r - Riordan matrix M by

| | | |
 
2 3
M = (g(z; r), f (z; r)) = (g, f ; r) = g
 g rf1 g fr2 g fr3 ...
| | | |

For a fixed sequence r, we have the following property: the set of (g, f ; r) forms a group when a0 6= 0 and b1 6= 0.
In the case of ordinary Riordan arrays we have r = (1, 1, 1, ...) and when r = (n!)∞ n=0 we get the exponential
Riordan arrays. Many of the subgroups given in the ordinary Riordan group are defined similarly for r-Riordan
arrays. For example, the Appell subgroup is the set of elements of the form (g, z; r) and the Associated subgroup
is the set of elements of the form (1, f ; r).
(2n)!
Example 6.1. Let rn = 2n .
 
1
 1 1 
 
 4
  10 1 
 √ √
sec2 ( z), tan2 ( z); (2n)!

n = 34 154 35 1 
2  
 496 3520 1344 84 1 
 
11056 112816 63580 6468 165 1
···
The decomposition of this matrix into the product of an Associated matrix with an Appell matrix is
√ √
sec2 ( z), z; (2n)!/2n ∗ 1, tan2 ( z); (2n)!/2n , i.e.

  
1 1
 1 1  0 1 
  
 4 6 1  0 4 1 
  
 34 60 15 1  0 34 20 1
  
 
 496 952 280 28 1  0 496 504 56 1 
  
11056 22320 7140 840 45 1 0 11056 16960 3108 120 1
··· ···

6.1.1 Generalized Binomial Coefficients

In this subsection, we will give the definition of the generalized binomial coefficient. These coefficients will be
used later when we discuss
Pnsummation functions.
PnThroughout this subsection,

r will be a fixed sequence.
Definition. [x + y]n = k=0 rk rrn−k xk y n−k = k=0 nk r xk y n−k , where nk r = rk rrn−k


n n
.

Theorem 6.1. If
X zn X zn
A(z) = an and B(z) = bn ,
rn rn
n≥0 n≥0

then
∞  
X zn X n
A(z)B(z) = dn , where dn = ak bn−k .
rn k rn
n≥0 k=0

n n



Note that if r = {n!}, then k r = k and [x + y]nr = (x + y)n . When r = {1}, then
n
X
[x + y]nr = xk y n−k = xn + xn−1 y + ... + xy n−1 + y n .
k=0

6.1.2 Summation Functions

zn
P∞
Define the summation function E(z; r) = E(z) = n=0 rn . Also, define the linear functional σr by
X zn X zn
σr A(z) = an+1 , where A(z) = an .
rn rn
n≥0 n≥0

Then, σr · E(z; r) = E(z, r).

For example, when r = {n!n!}, σr = DzD, where D is the differential operator.

ECA 4:3 (2024) Article #S2S1 11

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

∞ ∞ ∞ ∞
X zn X zn X z n−1 X zn
E(z, n!n!) = =⇒ DzD = Dz =D
n=0
n!n! n=0
n!n! n=1
n!(n − 1)! n=1
n!(n − 1)!
∞ ∞
X z n−1 X zn
= = = E(z, n!n!).
n=1
(n − 1)!(n − 1)! n=0 n!n!
A classical well-known example using summation functions is
    
1 1 1
1 1  1   2 
    
1 2 1  1   4 
    
1 3 3 1  1  =  8 
    
1 4 6 4 1    1   16 
   

1 5 10 10 5 1  1   32 
··· ··· ···
 
1 z 1 1
where the entries are represented either by 1−z , 1−z ; 1 ∗ 1−z = 1−2z or by (ez , z; n!) ∗ ez = [ez , z] ∗ ez = e2z .
The summation function E(z, r) is also the row sum function for Riordan arrays. That is, the Riordan array
(g, f ; r) has row sum g · E(f, r). Where (E, z; r) is the generalized Pascal triangle.
 0

0 r
 
1 1



 0 r 1 r

 
 
2 2 2




 
 0 r 1 r 2 r 
 
 3

3

3

3


 
 0 r 1 r 2 r 3 r 
 
 
4 4 4 4 4






0 r 1 r 2 r 3 r 4 r
···

Note that, for a summation function E, (E, z; r) ∗ E = E 2 .

Here are a few examples.
1 1 1 2
1. ( 1−z , z; 1) ∗ 1−z = ( 1−z )

2. (ez , z, n!) ∗ ez = (ez )2 = e2z

√ √ √
3. (I0 (2 z), z; (n!)2 ) ∗ I0 (2 z) = I02 (2 z) where I0 (z) is the modified Bessel function of order 0.
√
To find the entries of (I0 (2 z), z; (n!)2 ), we look at the generating functions for the first few columns:
√ z 2
z 3
z4
I0 (2 z) = 1 + z + 2!2! + 3!3! + 4!4! + ...
√ 4z 2 9z 3 4
25z 5
zI0 (2 z) = z + 2!2! + 3!3! + 16z4!4! + 5!5! + ...
z 2 √ z 2
9z 3
36z 4
100z 5
2!2! I0 (2 z) = 2!2! + 3!3! + 4!4! + 5!5! + ...
z3 √ z3 16z 4 100z 5 400z 6
3!3! I0 (2 z) = 3!3! + 4!4! + 5!5! + 6!6! + ...

 
1
1 1 
 
√ 1 4 1 
(I0 (2 z), z; (n!)2 ) =  
1 9
 9 1 · · ·

1 16 36 16 1 
1 25 100 100 25 1
√ 6z 2 20z 3 70z 4
Note that I02 (2 z) = 1 + 2z + 2!2! + 3!3! + 4!4! + ..., which we see below:
    
1 1 1
1 1  1   2 
    
1 4 1  1   6 
    
1 9 9 1   1  =  20 
    
1 16 36 16 1   1   70 
    
1 25 100 100 25 1  1  252
··· ··· ···

ECA 4:3 (2024) Article #S2S1 12

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

√ √ √
Thus, we have confirmed that (I0 (2 z), z; (n!)2 ) ∗ I0 (2 z) = I02 (2 z). Without the generating functions we
have the famous identity
n  2  
X n 2n
= .
k n
k=0

6.1.3 Appell Riordan Arrays

Riordan arrays of the form (g, z; r) are Appell Riordan arrays. We have the following results:
1. (g(z), z; r) + (h(z), z; r) = ((g + h)(z), z; r)
2. a(g(z), z; r) = (ag(z), z; r) for any constant a

3. (z, z; r)n = (z n , z; r), n ∈ Z+ . We define (z, z; r)0 = (1, z; r)

Note that some of the matrices from equations 1, 2, and 3 above are not necessarily Riordan.
n
Theorem 6.2. Let g(z; r) = n≥0 gn zrn and A = (z, z; r). Then, g(A) = (g(z), z; r).
P

Proof.
X An X gn X  zn   X zn 
g(A) = gn = (z, z; r)n = gn , z; r = gn , z; r = (g(z), z; r)
rn rn rn rn
n≥0 n≥0 n≥0 n≥0

An immediate corollary of this theorem is that an Appell matrix (g(z), z; r) is diagonally reducible to
(E(z, r), z; r). That is to say, the matrix is the diagonally-enhanced generalized Pascal triangle.
 0

0 r a0

 
 1
1


 0 a1 1 r a0

 r 
 
 2
2

2


 0 r a2 1 r a1 2 r a0
 

 
 3
3

3

3


 0 r a3 1 r a2 2 r a1 3 r a0
 

 


 4 a 4


a 4


a 4


a 4


a 
0 r 4 1 r 3 2 r 2 3 r 1 4 r 0
···
The following example looks at an element in the Bell subgroup:
 
1
1 1 
 
1 4 1 
 
1 12 9 1
(ez , zez ; n!) = 
 

1 32 54 16 1 
 
1 80 270 160 25 1 
 
1 192 1215 1280 375 36 1
···
Note that for r ≡ 1 = (1, 1, 1, ...), the matrix is not Riordan. The column-generating functions of this matrix
are
| | | | |
 
1 z z2 z3

1−z (1−2z)2 (1−3z)3 (1−4z)4 ··· .
| | | | |
n n
3n−2 , n3 4n−3 , ... A combinatorial interpretation will




So, the coefficients of the columns are 1, 2n−1 , 1 2
be the following: let the zeroth column be 1. For the first column, you must choose an executive committee
of 1 person and form all possible committees. The committees can be empty. For the second column, you
must choose an executive committee of 2 people and forming all possible committees and subcommittees of the
committees. The third column will represent choosing an executive committee of 3 people and form all possible
committees, subcommittees, sub-subcommittees, and so on for the following columns.

ECA 4:3 (2024) Article #S2S1 13

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

6.2 Double Riordan Array

In a Riordan array, we use one multiplier function. Consider the case where we use two multiplier functions.
So, if g gives column zero and f1 and f2 are the multiplier functions, then the first column is gf1 , the second
is gf1 f2 , the third is gf1 f2 f1 , and so on. The set of double Riordan arrays is not closed under multiplication.
However, if we require that g be an even function and f1 and f2 odd functions, then we can develop an analog
of FTRA and obtain a group structure.
P∞ P∞ P∞
Definition 6.1. Let g(z) = k=0 g2k z 2k , f1 (z) = k=0 f1,2k+1 z 2k+1 , and f2 (z) = k=0 f2,2k+1 z 2k+1 , with
g0 6= 0, f1,1 6= 0, and f2,1 6= 0. Then the double Riordan matrix (or array) of g, f1 and f2 , denoted by (g; f1 , f2 ),
is given by
(g; f1 , f2 ) = (g, gf1 , gf1 f2 , gf12 f2 , gf12 f22 , · · · ).
The set of all double Riordan matrices is denoted as DR. Note that all of these matrices are aerated.
√
1− 1−4z
Example 6.2. Let C(z) = 2z . This is the generating function for the Catalan numbers. The following
is a double Riordan array.
 
Let (g, f1 , f2 ) = C(z 2 ), zC 2 (z 2 ), z

g(f1 f2 )2 gf1 (f1 f2 )2 g(f1 f2 )3 gf1 (f1 f2 )3 ···

g gf1 gf1 f2 gf1 (f1 f2 )

1 
0 1 
1 0 1 
 
0 3 0 1 
.. 
 
.
2 0 3 0 1

0 9 0 5 0 1 
 
5 0 9 0 5 0 1 
0 28 0 20 0 7 0 1 
 
14 0 28 0 20 0 7 0 1
...
P∞ 2k
Theorem P∞6.3. (The 2k+1
Fundamental TheoremP∞ of Double Riordan Arrays): Let g(z) = k=0 g2k z ,
2k+1
f1 (z) = k=0 f1,2k+1 z , and f2 (z) = k=0 f2,2k+1 z .
P∞ ∞ T T
Case 1: If A(z) = k=0 a2k z 2k and B(z) = k=0 b2k z 2k , and A = (a0p
P
, 0, a2 , 0, · · · ) and B = (b0 , 0, b2 , 0, · · · )
are column vectors. Then (g, f1 , f2 )A = B if and only if B(z) = g(z)A( f1 (z)f2 (z)).
P∞ 2k+1
P∞ 2k+1
Case
p 2: If p A(z) = k=0 a2k+1 z and B(z) = k=0 b2k+1 z with (g, f1 , f2 )A = B, then B(z) =
g(z) f1 /f2 A( f1 (z)f2 (z)).
Definition 6.2. Let (g, f1 , f2 ) √and (G,p F1 , F2 ) be elements
√ pof DR. Then√
(g; f1 , f2 )(G; F1 , F2 ) = (gG( f1 f2 ); f1 /f2 F1 ( f1 f2 ), f2 /f1 F2 ( f1 f2 )).
The following theorem is analogous to Theorem 2.
Theorem 6.4. (DR, ∗) is a group.
Proof. The matrix (1; z, z) is the identity. √ Matrix multiplication is associative.
Let (g; f1 , f2 ) be in DR and let h = f1 f2 and also denote by h̄ the compositional inverse of h. Then
((1/g(h̄); z h̄/f1 (h̄), z h̄/f2 (h̄)) is the inverse of (g; f1 , f2 ).
Theorem 6.5. Let A={(g; f1 , f2 ) ∈ DR : g = 1} and
B1 ={(g; f1 , f2 ) ∈ DR : f1 = zg} and B2 ={(g; f1 , f2 ) ∈ DR : f2 = zg}. Then B1 , B2 and A are subgroups of
DR.
Theorem 6.6. {(g; z, z) ∈ R} is a normal subgroup of DR and DR is the semidirect product of {(g; z, z) ∈ R :
f = z} and A.
Theorem 6.7. The following are subgroups.
1. {(f 0 , f, cf ) ∈ DR : c ∈ IR, c > 0} and {(f 0 , cf, f ) ∈ DR : c ∈ IR, c > 0}
2. {(g, czg, f ) ∈ DR : c ∈ IR, c > 0} and {(g, f, czg) ∈ DR : c ∈ IR, c > 0}
The subgroups are called the derivative subgroups and the C-Bell Subgroups, see [10]. We see that DR has
some of the same subgroup properties as R. For more about the Double Riordan Array, see [17].
A Riordan array has one Z−sequence and one A−sequence. It is interesting to note that there are three
known ways to row construct elements in DR. One uses two Z−sequences and one A−sequence and the other
two use one Z−sequence and two A−sequences. One found by He is given in the following theorem, see [24].

ECA 4:3 (2024) Article #S2S1 14

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Theorem 6.8. Let D = (g; f1 , f2 ) = (dn,k ) be an infinite lower triangular matrix. Then D is a Double Riordan
array if and only if there exist three sequences A1 = {a1,i }∞ ∞ ∞
i=0 , A2 = {a2,i }i=0 and Z = {zi }i=0 with a1,0 6= 0,
a2,0 6= 0, and z0 6= 0 such that for each n
∞
X
dn,2k−1 = a1,j dn−1,2k+2(j−1) ; k = 0, 1, 2, ...,
j=0
X∞
dn,2k = a2,j dn−2,2k+2(j−1) ; k = 0, 1, 2, ..., and
j=0
X∞
dn,0 = zj dn−2,2j .
j=0

The following theorem is similar to Theorem 4 in Section 1. It allows us to find the A− and Z−sequences
using the generating functions that define the Double Riordan Array.
P∞ P∞
Corollary
P∞6.1. 2kLet D = (g; f1 , f2 ) be an element of DR. Let A1 (t) = k=0 a1,k t2k , A2 (t) = k=0 a2,k t2k , and
Z(t) = √k=0 zk t respectively be the generating functions for the A1 −sequence, A2 −sequence, and Z−sequence.
Let h = f1 f2 . Then

f1 (h)
A1 (t) =
h
t2
A2 (t) = 2
h !
1 g(0)
Z(t) = 2 · 1−
.
h g h

The next two theorems give other row constructions of Double Riordan Arrays, see [10] and [18].
Theorem 6.9. Let D = (g; f1 , f2 ) = (dn,k ) be an infinite lower triangular matrix. Then D is a Double Riordan
array if and only if there exist three sequences A = {ai }∞ ∞ ∞
i=0 , Z0 = {z0,i }i=0 and Z1 = {z1,i }i=0 with a0 6= 0,
z0,0 6= 0, and z1,0 6= 0 such that for each n
∞
X
dn,k = aj dn−2,k+2(j−1) ; k = 2, ...,
j=0
X∞
dn,1 = z1,j dn−2,2j+1 , and
j=0
X∞
dn,0 = z0,j dn−2,2j .
j=0
P∞ 2k
P∞ 2k
Corollary 6.2.P∞ Let D = (g; f1 , f2 ) be an element of DR. Let A(t) = k=0 ak t , Z0 (t) = k=0 z0,k t ,
2k
and Z1 (t) = √t respectively be the generating functions for the A−sequence, Z0 −sequence, and
k=0 z1,k
Z1 −sequence. Let h = f1 f2 . Then

t2
A(t) = 2
h !
1 g(0)
Z0 (t) = 2 · 1 −
.
h g h
 
1 f1,1 h
Z1 (t) = 1 − .
(h)2 k(h)

Theorem 6.10. Let D = (g; f1 , f2 ) = (dn,k ) be an infinite lower triangular matrix. Then D is a Double
Riordan array if and only if there exists three sequences A1 = {a1,i }∞ ∞ ∞
i=0 , A2 = {a2,i }i=0 and Z = {zi }i=0 with
a1,0 6= 0, a2,0 6= 0, and z0 6= 0 such that for each n
∞
X
dn,2k−1 = a1,j dn−1,2k+2(j−1) ; k = 0, 1, 2, ...,
j=0

ECA 4:3 (2024) Article #S2S1 15

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

∞
X
dn,2k = a2,j dn−1,2k+2(j−1) ; k = 0, 1, 2, ..., and
j=0
X∞
dn,0 = zj dn−1,2j .
j=0

P∞ P∞
Corollary
P∞6.3. 2kLet D = (g; f1 , f2 ) be an element of DR. Let A1 (t) = k=0 a1,k t2k , A2 (t) = k=0 a2,k t2k , and
Z(t) = √k=0 zk t respectively be the generating functions for the A1 −sequence, A2 −sequence, and Z−sequence.
Let h = f1 f2 . Then
t
A1 (t) =
f1 h
f1 (h)
A2 (t) =
h !
1 1 g(0)
Z(t) = · −
.
h f1 g h f1

In Section 1, we discussed the production matrix for Riordan Arrays. For Double Riordan Arrays two
production matrices were found, see [18].

Definition 6.3. Let (g; f1 , f2 ) be a double Riordan array. The production matrix of the first kind P1 is given
by:
P1 = (g; f1 , f2 )−1 · (g; f1 , f2 ),
where (g; f1 , f2 ) is the truncated double Riordan array with the first row omitted.
Definition 6.4. Let (g; f1 , f2 ) be a double Riordan array. The production matrix of the second kind P2 is
given by:
P2 = (g; f1 , f2 )−1 · (g; f1 , f2 ),

where (g; f1 , f2 ) is the truncated double Riordan array with the first two rows omitted.
Theorem 6.11. Let D = (g; f1 , f2 ) be a double Riordan

array. Then the production matrix of the second kind
P2 for D is of the form Z0 zZ1 A zA z 2 A ... , where Z0 is the first Z−sequence, Z1 is the second
Z−sequence, and A is the A−sequence for D from Corollary 2.
Theorem 6.12. Let D = (g; f1 , f2 ) be a double Riordan array.

Then the production matrix of the first kind
P1 for D is of the form Z A1 zA2 z 2 A1 z 3 A2 z 4 A1 ... , where Z is the Z−sequence, A1 is the first
A−sequence, and A2 is the second A−sequence for D from Corollary 3.
Example 6.3. For a combinatorial example, consider Schröder paths with no level steps at odd heights. See
the below grid.

1 9

1 8 46

1 6 29 132

1 5 21 86 355

1 3 10 36 137 543

1 2 5 15 51 188 731

ECA 4:3 (2024) Article #S2S1 16

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Arranging these numbers as a lower triangular array we get the following DR matrix.
1 1
   
 0 1  1  0 
 2 0 1  0 1  −2 
   

 0 3 0 1 
 −3 0 1 0



..  −1 ..
D= =⇒ D =  5
  
 5 0 5 0 1 . 
  0 −5 0 1 . 

 0 10 0 6 0 1   0 8 0 −6 0 1 
−13 −8
   
 15 0 21 0 8 0 1  0 19 0 0 1 
0 36 0 29 0 9 0 1 −21 −9 0 1
   0 0 25 0

... ...

 
0 1
2 0 1 
 
0 3 0 1 
 
5 0 5 0 1 
 
 ..
D=0 10 0 6 0 1 .


15 0 21 0 8 0 1 
 
0 36 0 29 0 9 0 1 
 
51 0 86 0 46 0 11 0 1 
 
 0 137 0 132 0 57 0 12 0 
...

 
2 0 1

 0 3 0 1 


 5 0 5 0 1 


 0 10 0 6 0 1 

 ..
D =  15 0 21 0 8 0 1 .


 0 36 0 29 0 9 0 1 
 
 51 0 86 0 46 0 11 0 1 
 
 0 137 0 132 0 57 0 12 0 
 
188 0 355 0 235 0 80 0 14 
...

Now we compute P1 and P2 .

P1 = D−1 ∗ D,
P2 = D−1 ∗ D.
As a result we get,
 
0 1
2 0 1 
 
0 1 0 1 
 
−1 0 2 0 1 
 
 ..
P1 =  0 0 0 1 0 1 .


 1 0 −1 0 2 0 1 
 
0 0 0 0 0 1 0 1 
 
−1 0 1 0 −1 0 2 0 1 
 
0 0 0 0 0 0 0 1 0 
...

Z − seq : (0, 2, 0, −1, 0, 1, 0, −1, ...)

A1 − seq : (1, 0, 1) A2 − seq : (1, 0, 2, 0, −1, 0, 1, ...)

ECA 4:3 (2024) Article #S2S1 17

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

 
2 0 1
0 3 0 1 
 
1 0 3 0 1 
 
0 1 0 3 0 1 
 
 ..
P2 = 0 0 1 0 3 0 1 .


0 0 0 1 0 3 0 1 
 
0 0 0 0 1 0 3 0 1 
 
0 0 0 0 0 1 0 3 0 
 
0 0 0 0 0 0 1 0 3 
...

Z1 − seq : (2, 0, 1) Z2 − seq : (3, 0, 1)

A − seq : (1, 0, 3, 0, 1)
The question if we can define a Riordan array and Riordan group of matrices with more than two multiplier
functions was affirmed in [17] without giving an explicit definition. In [24], He gave the following definition for
matrix multiplication of matrices P∞ with k multiplier functions, such matricesPwe call k-Riordan arrays. In the
∞
k-Riordan group, we let g(t) = n=0 gkn tkn and for each 1 ≤ i ≤ k, fi (t) = n=0 fi,kn+1 tkn+1 . For k-Riordan
k
Q
arrays (g, f1 , f2 , ..., fk ) and (G, F1 , F2 , ..., Fk ), multiplication is defined as follows. Let h(t) = fi (t). Then
i=1

(g, f1 , f2 , ..., fk ) · (G, F1 , F2 , ..., Fk ) =

s s
k (t) k
 
f k fk (t)
p  p  p
k k 1 k k
g(t) · G h(t) , · F1 h(t) , · · · , · Fk h(t) .
h(t) h(t)

7. Involutions and Pseudo-involutions

In this section, after some definitions, we look at a few famous examples, then a few less famous examples.
Then we give five methods of finding pseudo-involutions.

Definition 7.1. Let G be a group with identity e. Then an element g ∈ G is an involution ⇐⇒ g 2 = e.

Definition 7.2. Let M = (1, −z). Then an element A in the Riordan group is a pseudo-involution (P I) ⇐⇒
AM or M A is an involution.
Recallthat the Bell subgroup of the Riordan group are the elements of the form (g, zg) or alternatively of
the form fz , f . More generally, there is the k−Bell subgroup with elements of the form g, zg k where k is a

positive integer.
Among the examples of pseudo-involutions given are two in the Bell subgroup, one in the 3-Bell subgroup,
three more in the 2-Bell subgroup, and one which is not in any k-Bell subgroup. Many of the g functions are
well known in the arena of enumerative combinatorics. Displayed are the first five or so rows of each example
along with some commentary.
The first example is Pascal’s matrix
 
1 0 0 0 0 0 0
 1 1 0 0 0 0 0 
 
   1 2 1 0 0 0 0 
1 z  
P = , = 1 3 3 1 0 0 0 · · · 
1−z 1−z 
 1 4 6 4 1 0 0


 
 1 5 10 10 5 1 0 
1 6 15 20 15 6 1

There is no need to discuss the importance of this matrix. This indeed is almost the prototype for Riordan
group theory. It is in the Bell subgroup.

ECA 4:3 (2024) Article #S2S1 18

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

The RNA matrix  

1 0 0 0 0 0 0

 1 1 0 0 0 0 0 


 1 2 1 0 0 0 0 

(g, zg) = 
 2 3 3 1 0 0 0 ··· 


 4 6 6 4 1 0 0 

 8 13 13 10 5 1 0 
17 28 30 24 15 6 1
Here g = 1 + zg + z 2 g (g − 1) . One combinatorial interpretation is as Motzkin paths with no peaks. The g
sequence counts the number of possible RNA secondary structures for a chain of length n. Once again this
matrix is in the Bell subgroup.
The Catalan companion
 
1 0 0 0 0 0 0
 1 1 0 0 0 0 0 
 
 2 4 1 0 0 0 0 
C, zC 3 = 

 
 5 14 7 1 0 0 0 · · · 

 14 48 35 10 1 0 0 
 
 42 165 154 65 13 1 0 
132 572 637 350 104 16 1

This is an element in the 3-Bell subgroup.

Here are three pseudo-involutions in the 2-Bell subgroup:

   
1 0 0 0 0 1 0 0 0 0

 2 1 0 0 0 


 2 1 0 0 0 

2

 6 6 1 0 0 ···  2

 6 6 1 0 0 ··· 
r, zr = , B, zB = ,

 22 30 10 1 0 


 20 30 10 1 0 

 90 146 70 14 1   70 140 70 14 1 
··· ···

 
1 0 0 0 0

 2 1 0 0 0 

6 6 1 0 0 ···
t3 , zt23 = 

 


 24 30 10 1 0 

 110 152 70 14 1 
···

The three generating functions are the big Schröder numbers,

r, the central binomial coefficients, B , and the
ternary numbers doubled t3 . Since r = 1 + z r + r2 it has as it’s palindrome γ = x + x2 with darga(γ) =
1 + 2 = 3. The darga and γ function will be discussed in Corollary E.
2
B 4x2
For B we have B = 1 + 4z · 1+B so γ = 1+x and the darga is (2 + 2) − (0 + 1) = 3. For the third example we
t3 +1 t3 +3t2 +3t +1 3 2
start with T3 = 1 + zT33 substitute T3 = that t3 = 1 + z 3 34 3
2 and find giving us γ = x +3x 4+3x+1
with darga(γ) = 3 + 0 = 3.
We comment that the row sums of the first example (r, zr2 ) are the Central Delannoy numbers (see more
in [1], [49]).
The Fibonacci numbers provide a non-Bell example of a pseudo-involution:
 
1 0 0 0 0 0 0
  1 1 0 0 0 0 0
q  
 
1−5z−5z 2
1−  2 4 1 0 0 0 0
 
1 1−z−z 2 
F = 
2
,  =  3 14
 7 1 0 0 0 ··· 
1−z−z 2  5 50 35 10

 1 0 0 

 8 190 160 65 13 1 0 
13 778 720 360 104 16 1

For any of these examples, a set of questions arise. Given a generating function g, presumably of combinatorial
interest, how is f found? What is a combinatorial interpretation for f ? What is the combinatorial meaning of

ECA 4:3 (2024) Article #S2S1 19

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

the other entries in the matrix other than the leftmost column? How are the A−sequences, Z−sequences, and
B− sequences related? In this paper, we will answer some of these questions for a large number of examples
involving k-Bell arrays and their relatives.
Here are a few ways to construct pseudo-involutions. The following construction is due to Candice Marshall,
see [31] and [32]. Given a function g = 1+g1 z +g2 z 2 +g3 z 3 +..., we want to find a function f such that (g, f ) is a
pseudo-involution. We will say that f is a companion
  to g and we will abbreviate pseudo-involution to PI. Here
1−g
is one method; Let G = g − 1. Then f = −G g . As an example let g be the GF for the ternary numbers
1 3n+1 n

3
so that g = 1 + zg 3 = 1 + z + 3z 2 + 12z 4 + · · · = n≥0 3n+1 z . Then G = g − 1 = zg 3 = z (G + 1) and
P
n
3 3 3
substituting G for z we find that z = G (1 + z) so that G (z) = G = z/ (1 + z) . Since 1−g zg
g = − g = −zg
2

we have that
−zg 2
f = −G −zg 2 = − 2 3 5


3 = zg · g = zg
(1 − zg 2 )


Here is first part of the PI g, zg 5
 
1 0 0 0 0 0
 1 1 0 0 0 0 
 
3 6 1 0 0 0 · · ·
g, zg 5 = 



 12 33 11 1 0 0 
 
 55 182 88 16 1 0 
273 1020 627 168 21 1


The same reasoning gives the more general result that if g = 1 +
zg k then g, zg 2k+1 is a PI. When k = 1 we
have the Pascal matrix and when k = 2 the Catalan PI C, zC 3 is the result.
Here is how this formula for f is derived. If f is the PI companion then −f is an involution and thus

(g, −f ) (g, −f ) = (1, z) = (g · g (−f ) , −f (−f ))

So g · (g (−f )) = 1
1
and g (−f ) = g ◦ (−f ) =
g
1 1−g
G (−f ) = (g − 1) ◦ (−f ) = − 1 =
g g
 
1−g
Hence − f = G
g

7.1 The Master Theorem and Five Corollaries

The master theorem uses the simple fact that a conjugate of an involution is an involution. Recall also that if
M = (1, −z) , the identity matrix with alternating signs, then Z is a PI if and only if ZM is an involution, as
is M Z. The corollaries have been proven previously but now we present a unifying approach.
Theorem 7.1. Let Z be a PI. Then if X = (g, f ) is any element in the Riordan group then XZM X −1 M is
also a PI.
Proof.

ZM is an involution.
X (ZM ) X −1 is also an involution.
Thus XZM X −1 M is a PI.

It is convenient to denote M X −1 M as Xb and with this notation XZ X

b is a PI. For generating functions it
is also convenient to define fb = −f (−z) = (−z) ◦ f ◦ (−z) .

Corollary 7.1. If X is any Riordan group element, then X X

b and XX
b are PIs.

Proof. Let Z = I.
Example 7.1. A classic example has the Stirling numbers of the first kind as
  
1 1
X= , ln
1−z 1−z

ECA 4:3 (2024) Article #S2S1 20

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

whose inverse yields the Stirling numbers of the second kind. Then M X −1 M = X b are the unsigned Stirling
numbers of the second kind. The product X X b has various names: the Lah numbers, preferential arrangements,
ordered partitions, or more informally, a horse race with possible ties. Here, in matrix form are the first few
rows
     
1 0 0 0 0 1 0 0 0 0 1 0 0 0 0
1 1 0 0 0 1 1 0 0 0
   2 1 0 0 0
  
2
 3 1 0 0 · 1 3 1 0 0 =  6
    6 1 0 0
 6 11 6 1 0 1 7 6 1 0  24 36 12 1 0
24 50 35 10 1 1 15 25 10 1 120 240 120 20 1

Let us examine the row 6,6,1. There are 3 horses in a race. Without ties, there are 3! = 6 possibilities. There
are 3 ways to have two of the horses tie for first place and another 3 ways to have a tie for second place. There
is just one way for all three horses in a tie for first.
The PI XX
b is also interesting. Here
     
1 0 0 0 0 1 0 0 0 0 1 0 0 0 0
1 1 0 0 0  1 1 0 0 0
  2 1 0 0 0
   
1 3
 1 0 0 ·  2
  3 1 0 0 =  6  6 1 0 0 
1 7 6 1 0  6 11 6 1 0  26 36 12 1 0
1 15 25 10 1 24 50 35 10 1 150 250 120 20 1

Here the left-hand column now counts horse races where not only can you have ties but one or more horses can
drop out. For instance with 3 horses there are 13 ways with no dropouts, 3 · 3 = 9 with one dropout, 3 with 2
horses dropping out, and 1 where, tragically, all three horses had to drop out. The total is 13 + 9 + 3 + 1 = 26,
the (3, 0) entry.
Here we have    " #
1 1 z z 1 z
XX b= , ln [e , e − 1] = 2,
1−z 1−z (1 − z) 1 − z
for the Lah case while

ez
     
z z 1 1 1
[e , e − 1] , ln = , ln
1−z 1−z 2 − ee 2 − ez
Note that all these matrices are in the derivative subgroup.
P∞   
1 h
Corollary 7.2. Let h(z) = k=1 hk z k with h1 6= 0. Then 1−h , h 1−h is a PI.
 
1 z
Proof. Let X = (1, h) and take Z = P = 1−z , 1−z .

This version is quite convenient and is discussed with many examples in [25] and [12].
There is the folklore heuristic principle that
1
= ”T otal case GF ”
1 − Connected part GF
As a prime example, we want to find the PI involving the Fibonacci numbers. The generating function is
1
g=F =
1 − z − z2
so we take
h (z) = z + z 2
There is a vague duality between the Fibonacci and Catalan numbers. One way to express it is to note that
c = z + z2
zC

so that
1
F = ◦ zC.
c
1−z
 
1 z
Using X = 1−z , 1−z we have
 
2

1 z
ZX Zb = 1, z + z , (1, zC)
1−z 1−z

ECA 4:3 (2024) Article #S2S1 21

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

z + z2
 
1
= , (1, zC)
1 − z − z2 1 − z − z2
z + z2 z + z2
  
1
= , · C
1 − z − z2 1 − z − z2 1 − z − z2
= (F, (F − 1) C (F − 1))
The companion function
r !
1 1 − 5z − 5z 2
(zC) ◦ (F − 1) = 1−
2 1 − z − z2
z + 3z 2 + 9z 3 + 32z 4 + 126z 5 + 538z 6 + 2429z 7 + 11 412z 8 + O z 9


=

Looking at (zC) ◦ (F − 1) we see that that we obtain only positive integer coefficients.
The first few rows of this pseudo-involution are
 
1 0 0 0 0 0
1 1 0 0 0 0
 
2 4 1 0 0 0
 .
3 14 7 1 0 0
 
5 50 35 10 1 0
8 190 160 65 13 1
If we let h = 2z + z 2 then we obtain the Pell numbers 1, 2, 5, 12, 29, 70, ... with
1
g= .
1 − 2z − z 2
The PI companion is
√
4 − 4H
2− √
F = =1− 1−H
r2 r
h 1 − 2h
= 1− 1− =1−
1−h 1−h
r
1 − 4z − 2z 2
= 1− .
1 − 2z − z 2
The surprise comes starting with the coefficient of z 8 .
r
1 − 4z − 2z 2
F = 1−
1 − 2z − z 2
7233 8 26 183 9
= z + 3z 2 + 9z 3 + 28z 4 + 90z 5 + 299z 6 + 1025z 7 + z + z + ...
2 2
An open question is when does the companion have integer coefficients?
Corollary 7.3. Let A and B both be PI. Then ABA is also a PI.
Note that usually, AB is not a PI.
If X is a PI, then X −1 = M XM or equivalently X = M X −1 M = X. b
Thus if A and B are both PIs then so is AB A = ABA. Then it is easy to establish that any palindromic
b
expression in PI is also a PI. This includes AIA = AA, ABCBA, ABCDDCBA, ...
There are many examples in [25] and here is one more.
     
1+z z 1+z z
Example 7.2. Let A = 1−z , z and B = 1, 1−z . The product ABA = 1−3z+2z 2 , 1−z . Here are the first
few rows of this PI.  
1 0 0 0 0
 4 1 0 0 0 
 
 10 5 1 0 0 
 
 22 15 6 1 0 
46 37 21 7 1
The leftmost column counts rows of n squares where the leftmost squares can be red or green, followed pos-
sibly by some purple squares finishing with zero or one brown square. The 10 possibilities when n = 2 are
RR,RG,GR,GG,RP,GP,PP,RB,GB, and PB.
For successive columns add one new color per column.
The sequence 1, 4, 10, 22, 46, 94, ... has the OEIS reference A033484 where more information can be found.

ECA 4:3 (2024) Article #S2S1 22

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

Here are two examples using the main theorem itself.

   
1 b = (1 + Az, z) . For our seed PI we use Z = 1, z
Example 7.3. Let X = 1−Az , z so that X 1−z and then
  
1 z
XZ X
b = ,z 1, (1 + Az, z)
1 − Az 1−z
 
1 z
= , (1 + Az, z)
1 − Az 1 − z
 
1 1 + (A − 1) z z
= · , .
1 − Az 1−z 1−z

For instance if A = 3 we get the following PI.

 
1 0 0 0 0
   6 1 0 0 0 
1 + 2z 1 z  
· , = 21 7 1 0 0 
1 − 3z 1 − z 1 − z  
 66 28 8 1 0 
201 94 36 9 1

A visit to the OEIS leads to the fact that the sequence in the leftmost column counts the number of maximal
cliques in the series of Hanoi graphs. This is easily verified and it is a rather pretty sequence of graphs.
 
z
Example 7.4. We again use Z = 1, 1−z and this time let X = (C (−z) , z) . The resulting PI is
 
C (−z) z
  , .
C 1−zz 1−z

Reasonable enough? Here are the first eight rows of the PI.
 
1 0 0 0 0 0 0 0
 −2 1 0 0 0 0 0 0
 
 1
 −1 1 0 0 0 0 0
 −10 0 0 1 0 0 0 0
 
 5
 −10 0 1 1 0 0 0
 −92 −5 −10 1 2 1 0 0
 
 4 −97 −15 −9 3 3 1 0
−1117 −93 −112 −24 −6 6 4 1

Reasonable assumptions do not save us from a combinatorial nightmare.

Corollary 7.4. We recall
 that if (g, f ) is a PI, g = 1 + g1 z + g2 z 2 + · · · , g1 6= 0, and G = g − 1. Then the

companion is f = −G 1−g
g .

We can rephrase this as

(g, f ) = (g, z) (1, f )

 
z

= (g, z) (1, G) 1 (1, −z) 1, G (1, −z)
1+z
 
z \
= (g, z) (1, G) 1, (1, G).
1+z

Thus
z
f = (−z) ◦ G ◦ (−z) ◦ ◦G
1+z
b◦ z
= G ◦ G.
1+z
Corollary 7.5. Suppose g = 1 + zγ (g) z + 1. Then the f companion is

γ (g)
f =z .
gγ (1/g)

ECA 4:3 (2024) Article #S2S1 23

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

z
Proof. Let G = g − 1 = zγ (g) = zγ (G + 1) so that z = Gγ (z + 1). Then G
b=
γ(z−1) and

b◦ z
f = G ◦G
1+z
z z
= ◦ ◦ zγ (g)
γ (z − 1) 1 + z
z G
= ◦
γ (z − 1) 1 + G
G 1
=  
1+Gγ G −1
1+G

g−1 1 γ (g)
= =z .
g γ (1/g) gγ (1/g)

Corollary 7.6. Suppose in addition that γ (x) is a palindrome of darga d then f = zg d−1 i.e. a member of
the (d − 1)-Bell subgroup. The darga is the sum of the degrees of the initial and final non-zero terms of the
palindrome.

First, we look at an example, then the proof.

2 1 2
Example
7.5. Let
r = 1 + z(r + r ). Then γ (x) = x + x is a palindrome of darga 1 + 2 = 3. Thus
r, zr3−1 = r, zr2 is a pseudo-involution. This is a well-known example as r is the generating function for
the large Schröder numbers.

Proof. Since we have a palindrome of darga d it follows that xd γ (1/x) = γ (x) . Hence

γ (g) g d−1 γ (g) g d−1 γ (g)

z =z d =z = zg d−1 .
gγ (1/g) g γ (1/g) γ (g)

See [12] for more examples.

8. Conclusion
The first applications of Riordan arrays were in providing quick proofs for somewhat involved combinatorial
identities and the ability to invert combinatorial identities using the group structure. Since then, many new
connections to other parts of mathematics have been found. These include
• Riordan Lie theory (Cheon, Luzón, Morón, Prieto-Martinez, Song [16])
• connections with the Banach fixed point theorem (Luzón, Morón [28])

• A-matrices (Barry, Merlini, Rogers, Sprugnoli, Verri [4, 33])

• summation methods (He [23])
• orthogonal polynomial theory (Barry, Mwafise, Viennot [3, 5, 7, 50])
• the RNA secondary structure (Evans, Nkwanta [20])

• directed animals (Barcucci, Del Lungo, Pergola, Pinzani [2])

• Riordan graphs and posets (Cheon, Curtis, Kwon, Mwafise [14])
• the Riemann hypothesis (Cheon, Kim [15])
• pattern avoidance (Burstein, Lankham, Merlini, Sprugnoli [11, 34])

• Somos sequences and elliptic functions (Barry, Mwafise [4, 8, 35])

• total positivity (Chen, Liang, Mao, Mu, Slowik, Wang [13, 30, 44]),

ECA 4:3 (2024) Article #S2S1 24

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

to name a few.
We hope this invitation will interest and equip the reader to go further. Despite, or because of, the simplicity
of the concepts, much has developed since the first paper in 1991 [42]. Their ability to represent and manipulate
sequences, generate combinatorial identities, and model intricate phenomena demonstrates their significance
in advancing research and applications across diverse disciplines. As mathematicians and researchers delve
deeper into Riordan arrays’ rich and unexplored territory, we can anticipate even more profound insights and
practical implications. For those who want to learn more about Riordan arrays, there is an annual international
conference on Riordan arrays, an undergraduate textbook written by Paul Barry, [6], a Springer monograph by
Shapiro, Sprugnoli, Barry, et al [41], and two Youtube presentations by Melkamu Zeleke [53, 54].

References
[1] C. Banderier and S. Schwer, Why Delannoy numbers?, J. Statist. Plann. Inference 135 (2005), 40–54.
[2] E. Barcucci, A. Del Lungo, E., Pergola, and R. Pinzani, Directed animals, forests and permutations, Discrete
Math. 204 (1999), 41—71.
[3] P. Barry, On a transformation of Riordan moment sequences, J. Integer Seq., Article 18.7.1.
[4] P. Barry, Riordan arrays, the A-matrix, and Somos-4 sequences, [Link]
2019.
[5] P. Barry, Riordan arrays, orthogonal polynomials as moments, and Hankel transforms, J. Integer Seq. 14
(2011), Article 11.2.2.
[6] P. Barry, Riordan Arrays: A Primer, Logic Press, Raleigh, 2016.
[7] P. Barry and A. Mwafise, Classical and semi-classical orthogonal polynomials defined by Riordan arrays,
and their moment sequences, J. Integer Seq. 21(2018) Article 18.1.5.
[8] P. Barry and A. Mwafise, Exponential Riordan arrays and Jacobi elliptic functions, Linear and Multilinear
Algebra 70 (2021), 5770–5789.
[9] M. Bóna, Handbook of Enumerative Combinatorics, Taylor and Francis Group, Boca Raton, 2015.
[10] D. Branch, D. Davenport, S. Frankson, J. Jones, and G. Thorpe, A and Z Sequences for Double Riordan
Arrays, Springer Proceedings in Mathematics and Statistics 388 (2022), 33–46.
[11] A. Burstein and I. Lankham, Restricted patience sorting and barred pattern avoidance, in Permutation
Patterns, LMS Lecture Note Series, No. 376, Cambridge University Press, 2010, pp. 233-–257.
[12] A. Burstein and L. Shapiro, Pseudo-involutions in the Riordan group, J. Integer Seq. 25 (2022), Article
22.3.6.
[13] X. Chen, H., Liang, and Y. Wang, Total positivity of Riordan arrays, European J. Combin. 46 (2015),
68–74.
[14] G. Cheon, B. Curtis, G. Kwon, and A. Mwafise, Riordan posets and associated incidence matrices, Linear
Algebra Appl. 632. 10.1016/[Link].2021.10.002.
[15] G.-S. Cheon and H. Kim, Mertens equimodular matrices of Redheffer type, Linear Algebra Appl. 572 (2019),
252–272.
[16] G. Cheon, A. Luzón, M. Morón, L. Prieto-Martinez, and M. Song, Finite and infinite dimensional Lie
group structures on Riordan groups, Adv. in Math. 319 (2017), 522–566.
[17] D. Davenport, L. Shapiro, and, L. C. Woodson, The double Riordan array, Electron. J. Combin. 18 (2011),
1–16.
[18] D. Davenport, F. Fall, J. Francis, and T. Lee, Production matrices for double Riordan arrays, Springer
Proceedings in Mathematics and Statistics, Accepted.
[19] R. Donaghey and L. W. Shapiro, A survey of the Motzkin numbers, J. Combin. Theory Ser. A (1977),
291–301.
[20] J. Evans and A. Nkwanta, Linear trees, lattice walks, and RNA arrays, Appl. Math. (2023), 200–220.
[21] G. Hardy, A Course of Pure Mathematics, 7th edition, University Press, Cambridge, England, 1938
[22] F. Harary and R. C. Read, The enumeration of tree-like polyhexes, Proc. Edinburgh Math. Soc. 17 (1970),
1–13.
[23] T.-X. He, Methods for the summation of series, CRC Press, 2022.
[24] T.-X. He, Sequence characterizations of double Riordan arrays and their compressions, Linear Algebra
Appl. 549 (2018), 176–202.
[25] T.-X. He and L. W. Shapiro, Palindromes and pseudo-involution multiplication, Linear Algebra Appl. 593
(2020), 1–17.

ECA 4:3 (2024) Article #S2S1 25

 Dennis E. Davenport, Shakuan K. Frankson, Louis W. Shapiro, and Leon C. Woodson

[26] T.-X. He and R. Sprugnoli, Sequence characterizations of Riordan arrays, Discrete Math. 309 (2009),
3962–3974.
[27] A. Kuznetkov, I. Pak, and A. Postnikov, Trees associated with the Motzkin numbers, J. Combin. Theory
Ser. A, 76 (1996), 145–147.
[28] A. Luzón and M. Morón, Ultrametrics, Banach’s fixed point theorem and the Riordan group, Discrete
Appl. Math. 156(14) (2008), 2620–2635.
[29] A. Luzón, M. Morón, and L. Prieto-Martı́nez, Commutators and commutator subgroups of the Riordan
group, Adv. in Math. 428 (2023), Article 109164.
[30] J. Mao, L. Mu, and Y. Wang, Yet another criterion for the total positivity of Riordan arrays, Linear Algebra
Appl. 634 (2022), 106–111.
[31] C. A. Marshall, Construction of Pseudo-Involutions in the Riordan Group, Congr. Numer. 229 (2017),
343–351.
[32] C. A. Marshall, Construction of Pseudo-Involutions in the Riordan Group, Ph.D. Dissertation, Morgan
State University, 2017.
[33] D. Merlini, D. G. Rogers, R., Sprugnoli, and M. C. Verri, On some alternative characterizations of Riordan
arrays, Can. J. Math 49 (1997), 301–320.
[34] D. Merlini and R. Sprugnoli, Algebraic aspects of some Riordan arrays related to binary words avoiding a
pattern, Theoret. Comput. Sci. 412 (2011), 2988-–3001
[35] A. Mwafise, Riordan Arrays, Elliptic Functions, and their Applications, Ph.D. Dissertation, Waterford
Institute of Technology, 2017.
[36] P. Peart and W. Woan, Generating Functions via Hankel and Stieltjes Matrices, J. Integer Seq. 3 (2000),
Article 00.2.1.
[37] P. Peart and L. Woodson, Triple factorization of some Riordan matrices, Fibonacci Quart. 31 (1993),
121–128.
[38] Riordan Calculator [Link]
[39] R. Riordan, Combinatorial Identities, Wiley Series in Probability and Mathematical Statistics, New York,
1968.
[40] D. G. Rogers, Pascal triangles, Catalan Numbers and renewal arrays, Discrete Math. 22 (1978), 301–310.
[41] L. W. Shapiro, R. Sprugnoli, P. Barry, G. Cheon, T. He, D. Merlini, W. Wang, The Riordan Group and
Applications, Springer Monographs in Mathematics, Cham, Switzerland, 2022.
[42] L. W. Shapiro, S. Getu, W. Woan, and L. C. Woodson, The Riordan Group, Discrete Appl. Math. 34
(1991), 229–239.
[43] Sloane’s Online Encyclopedia of Integer Sequences, [Link]
[44] R. Slowik, Some (counter) examples on totally positive Riordan arrays, Linear Algebra Appl. 594 (2020),
117-123.
[45] R. Slowik, When a word in Riordan involutions is a Riordan involution, Aequat. Math. 96 (2022), 843–847.
[46] R. Sprugnoli, Riordan arrays and combinatorial sums, Discrete Math. 132 (1994), 267–290.
[47] R. Sprugnoli, A bibliography on Riordan arrays, [Link]
Renzo-Sprugnoli/publication/228562578_A_bibliography_on_Riordan_arrays/links/
5813801208aedc7d8961e320/[Link].
[48] R. Stanley, Enumerative Combinatorics , Volume 2, Cambridge University Press, 1999.
[49] R. A. Sulanke, Objects counted by the central Delannoy numbers, J. Integer Seq. 6 (2003), Article 3.1.5.
[50] X. Viennot, Combinatorial theory of orthogonal polynomials and continued fractions, Lecture notes, https:
//[Link]/wa_res/files/cours_imsc19_ch0.pdf, 2019.
[51] W. Wang and T. Wang, Generalized Riordan arrays, Discrete Math. 308 (2008), 6466–6500.
[52] L. Woodson, Infinite Matrices, Cn −Functions and Umbral Calculus, Ph.D. Dissertation (1988), Howard
University.
[53] M. Zeleke, Riordan Arrays and Their Applications in Combinatorics Part 1, April 19, 2012, [Link].
com.
[54] M. Zeleke, Riordan Arrays and Their Applications in Combinatorics Part 2, April 19, 2012, [Link].
com.

ECA 4:3 (2024) Article #S2S1 26