© 2015 Thomas Ernst, licensee De Gruyter Open.
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License.
Research Article Open Access
Thomas Ernst
Factorizations for q-Pascal matrices of two variables
DOI 10.1515/spma-2015-0020
Received July 15, 2015; accepted September 14, 2015
Abstract:In this second article on q-Pascal matrices, we show how the previous factorizations by the sum- mation matrices and the so-called q-unit matrices extend in a natural way to produce q-analogues of Pascal matrices of two variables by Z. Zhang and M. Liu as follows
Φn,q(x, y) =
⎛
⎝ (︃
i j )︃
q
xi−jyi+j
⎞
⎠
n−1
i,j=0
.
We also find two different matrix products for
Ψn,q(x, y; i, j) =
⎛
⎝ (︃i+ j
j )︃
q
xi−jyi+j
⎞
⎠
n−1
i,j=0
.
Keywords: q-Pascal matrix; q-unit matrix; q-matrix multiplication
1 Introduction
Once upon a time, Brawer and Pirovino [2, p.15 (1), p. 16(3)] found factorizations of the Pascal matrix and its inverse by the summation and difference matrices. In another article [7] we treated q-Pascal matrices and the corresponding factorizations. It turns out that an analoguous reasoning can be used to find q-analogues of the two variable factorizations by Zhang and Liu [13]. The purpose of this paper is thus to continue the q-analysis- matrix theme from our earlier papers [3]-[4] and [6]. To this aim, we define two new kinds of q-Pascal matrices, the lower triangular Φn,qmatrix and the Ψn,q, both of two variables. To be able to write down addition and subtraction formulas for the most important q-special functions, i.e. the q-exponential function and the q- trigonometric functions, we need the q-additions. These addition formulas were first published in different notation by Jackson [9] and Exton [8]. In one formula of the present paper we use this q-addition. This paper is organized as follows: In section 2 we give the definitions for q-calculus and definitions and a simple theorem for the matrices.
In section 3 we give the factorization and inverse of the extended q-Pascal matrix Φn,q(x, y). Finally, in section 4, we give the factorizations for the generalized symmetric q-Pascal matrix Ψn,q(x, y).
2 Definitions
For a full description of all definitions, see the recent book [5].
Thomas Ernst:Department of Mathematics, Uppsala University, P.O. Box 480, SE-751 06 Uppsala, Sweden, E-mail:
thomas@math.uu.se
Definition 1. The power function is defined by q ≡e . We always assume that 0 < q < 1. Let δ > 0 be an arbitrary small number. We will use the following branch of the logarithm: −π + δ < Im (log q) ≤ π + δ.
This defines a simply connected space in the complex plane.
The variables a, b, c, . . .∈C denote certain parameters. The variables i, j, k, l, m, n will denote natural numbers except for certain cases where it will be clear from the context that i will denote the imaginary unit.
The q-analogues of a complex number a and the factorial function are defined by:
{a}q≡ 1 − qa
1 − q , q∈C∖{1}, (1)
{n}q!≡
n
∏︁
k=1
{k}q, {0}q!≡1, q∈C. (2) Gauss’ q-binomial coefficients are given by
(︃n k )︃
q
≡ {n}q!
{k}q!{n− k}q!. (3)
Definition 2. Let α and β be elements of a ring. The NWA q-addition is given by [1], [5], [6], [10], [11] :
(α⊕qβ)n≡
n
∑︁
k=0
(︃n k )︃
q
αkβn−k, n = 0, 1, 2, . . . (4)
Definition 3. If f (x) ∈ C[x], the polynomials with complex coefficients, the function ϵ : C[x] ↦→ C[x] is defined by
ϵf(x)≡f(qx). (5)
We now leave q-calculus and turn our attention to the matrix definitions. In order to be able to write down certain q-matrix multiplication formulas, the following definition will be convenient.
Definition 4. Let A and B be two n × n matrices, with matrix index aijand bij, respectively. Then we define
ABf,q(i, j)≡
n−1
∑︁
m=0
aimbmjqf(m,i,j). (6)
Whenever we use a q-matrix multiplication, we specify the corresponding function f (m, i, j).
Remark1. This q-matrix multiplication will be used in formulas (38) and (39).
The following matrices, which are used for intermediary calculations, have a relatively simple structure. In section 3 we will encounter similar q-dependent matrices, which enable a multitude of similar formulas.
Definition 5. The matrices In, Sn(x), Wn(x, y) , Dn(x) and Un(x, y) [13, p. 171] are defined by
In≡diag(1, 1, . . . , 1), (7)
Sn(x)(i, j)≡
{︃xi−j, if j ≤ i,
0, if j > i, (8)
Wn(x, y; i, j)≡
{︃xi−jyi+j, if j ≤ i,
0, if j > i, (9)
Dn(x; i, i)≡1, i = 0, . . . , n − 1, Dn(x; i + 1, i)≡−x, for i = 0, . . . , n − 2,
Dn(x; i, j)≡0, if j > i or j < i − 1. (10)
Un(x, y; i, i)≡y−2i, i = 0, . . . , n − 1; Un(x, y; i + 1, i)≡−xy−2i−1,
for i = 0, . . . , n − 2; Un(x, y; i, j)≡0, if j > i or j < i − 1. (11) The matrices Snand Dnare used in definition (19).
Definition 6. The extended q-Pascal matrix Φn,q(x, y) is given by
Φn,q(x, y; i, j)≡ (︃i
j )︃
q
xi−jyi+j. (12)
Theorem 2.1. A q-analogue of [13, p. 170].
Φn,q(x1, y1)Φn,q(x2, y2) = Φn,q(x1
y2⊕qx2y1, y1y2), y2≠ 0. (13) Proof.
Φn,q(x1, y1)Φn,q(x2, y2)(i, j) =
n−1
∑︁
k=0
x1i−kyi1+k (︃
i k
)︃
q
x2k−jyk+j2 (︃
k j )︃
q
= (︃
i j )︃
q
(y1y2)i+j
n−1
∑︁
k=0
(︃
i− j k− j
)︃
q
(x1
y2)i−k(x2y1)k−j= (︃
i j )︃
q
(y1y2)i+j(x1
y2⊕qx2y1)i−j.
(14)
Definition 7. The matrices Pn,q(x), Pn,k,q(x), Pk,q* (x) and Pn,k,q* (x) are defined by
Pn,q(x; i, j)≡ (︃i
j )︃
q
xi−j, i, j = 0, . . . , n − 1, (15)
Pn,k,q(x)≡
[︃ In−k 0T 0 Pk,q(x)
]︃
, (16)
Pk,q* (x; i, j)≡ (︃
i j )︃
q
(qx)i−j, i, j = 0, . . . , k − 1, (17)
Pn,k,q* (x)≡
[︃ In−k 0T 0 Pk,q* (x)
]︃
. (18)
The summation matrix Gn,k(x) and the difference matrix Fn,k(x) are defined by
Gn,k(x)≡
[︃ In−k 0T 0 Sk(x)
]︃
, k = 3, . . . , n,
Fn,k(x)≡
[︃ In−k 0T 0 Dk(x)
]︃
, k = 3, . . . , n, Fn,n(x)≡Dn(x), n > 2.
(19)
Let the two matrices Ik,q(x) and Ek,q(x) be given by
Ik,q(x; i, i)≡1, i = 0, . . . , k − 1, Ik,q(x; i + 1, i)≡x(qi+1− 1), i ≤ k − 2, Ik,q(x; i, j)≡0 for other i, j.
Ek,q(x; i, j)≡ ⟨j+ 1; q⟩i−jxi−j, i ≥ j, Ek,q(x; i, j)≡0 for other i, j.
(20)
In,k,q(x)≡
[︃ In−k 0T 0 Ik,q(x)
]︃
, In,n,q(x)≡In. (21)
En,k,q(x)≡
[︃ In−k 0T 0 Ek,q(x)
]︃
, En,n,q(x)≡In. (22)
We call In,k,q(x) the q-unit matrix function. We will use a slightly q-deformed version of the D- and F-matrices:
Dk,q* (x; i, i)≡1, i = 0, . . . , k − 1, Dk,q* (x; i + 1, i)≡−xqi+1, i ≤ k − 2,
Dk,q* (x; i, j)≡0, if j > i or j < i − 1. (23)
Fn,k,q* (x)≡
[︃ In−k 0T 0 Dk,q*(x)
]︃
. (24)
Gk,q* (x; i, j)≡
{︃q(i−j+12 )+j(i−j)xi−j, if j ≤ i,
0, if j > i,, Gn,k,q* (x)≡
[︃ In−k 0T 0 Gk,q* (x)
]︃
.
(25)
The inverse of Pk,q* (x) is given by
(Pk,q* (x))−1(i, j) = (︃
i j )︃
q
(−x)i−jq(i−j+12 ), i, j = 0, . . . , k − 1. (26)
3 Factorization of the extended q-Pascal matrix Φn,q(x, y)
We start this section with a couple of lemmata to be able to make quick proofs of the factorization theorems.
Lemma 3.1. Four inverse relations.
Wn(x, y) = Un(x, y)−1; Fn,k(x) = Gn,k(x)−1, k = 3, . . . , n. (27)
In,k,q(x)−1= En,k,q(x); Fn,k,q* (x)−1= Gn,k,q* (x). (28) The following three lemmata enable a step by step proof of (33).
Lemma 3.2. [7] A q-analogue of [12, p.53 (1)]. If n ≥ 3, the q-Pascal matrix Pn,q(x) can be factorized by the summation matrices and by the q-unit matrices as
Pn,q(x) =
3
∏︁
k=n
(︀In,k,q(x)Gn,k(x))︀ Gn,2,q* (x), (29)
where the product is taken in decreasing order of k.
Proof. Use the same technique as in Brawer & Pirovino [2], but use the q-unit matrices and the q-Vandermonde theorem.
Lemma 3.3. [7]
In,n−1,q(x)Pn,n−1,q(x) = Pn,n−1,q* (x), n ≥ 1. (30)
Proof. Use the q-Pascal triangle in the end.
Lemma 3.4. A q-analogue of [13, p.173].
Wn(x, y)Pn,n−1,q* (x
y) = Φn,q(x, y), n ≥ 1. (31)
Proof. For n = 1, Pn,n−1,q* (x) = Inand Wn(x, y) = Φn,q(x, y). Let n > 1. The matrix element (Wn(x, y)Pn,n−1,q* (x
y)(i, j) =
i
∑︁
k=j
xi−jyi+j (︃
k− 1 j− 1 )︃
q
qk−j= xi−jyi+j (︃
i j )︃
q
= Φn,q(x, y)(i, j), j ≥ 1. (32)
For j = 0, Wn(x, y)Pn,n−1,q* (xy; i, 0) = (xy)i= Φn,q(x; i, 0).
Theorem 3.5. A q-analogue of [13, p.174]. If n≥ 3, the extended q-Pascal matrix Φn,q(x, y) can be factorized by the W matrix, the summation matrices and by the q-unit matrices as
Φn,q(x, y) = Wn(x, y)
∏︁3 k=n−1
(︂
In,k,q(x y)Gn,k(x
y) )︂
Gn,2,q* (x
y), (33)
where the product is taken in decreasing order of k.
Proof. Use (31). At each step the q-unit matrix maps Pn,k,q* (xy) to Pn,k,q(xy) by (30) and to Gn,k(xy) by (29).
Example1. The case n = 4:
⎛
⎜
⎜
⎜
⎝
1 0 0 0
xy y2 0 0
x2y2 {2}qxy3 y4 0 x3y3 {3}qx2y4 {3}qxy5 y6
⎞
⎟
⎟
⎟
⎠
=
⎛
⎜
⎜
⎜
⎝
1 0 0 0
xy y2 0 0
x2y2 xy3 y4 0 x3y3 x2y4 xy5 y6
⎞
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎝
1 0 0 0
0 1 0 0
0 xy(q − 1) 1 0
0 0 xy(q2− 1) 1
⎞
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎝
1 0 0 0
0 1 0 0
0 xy 1 0
0 (xy)2 xy 1
⎞
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎝
1 0 0 0
0 1 0 0
0 0 1 0
0 0 xyq 1
⎞
⎟
⎟
⎟
⎠ .
(34)
Theorem 3.6. A q-analogue of [13, p.174]. The inverse of the extended q-Pascal matrix is given by Φn,q(x, y)−1= Fn,2,q*(︂ x
y )︂ n
∏︁
k=3
(︂
Fn,k
(︂ x y )︂
En,k,q
(︂ x y
)︂)︂
Un(x, y). (35)
Proof. Use formulas (27), (28) and (29).
4 Higher q-Pascal matrices
In this section we treat matrices of a slightly different form.
Definition 8. A q-analogue of the matrix Qnin [12, p.55]. The matrix Rn,q(x) is defined by Rn,q(x; i, j)≡
(︃i j )︃
q
xi+j, i, j = 0, . . . , n − 1. (36)
A q-analogue of the matrix Ψn[x, y] in [13, p.175]. The generalized symmetric q-Pascal matrix Ψn,q(x, y) is defined by
Ψn,q(x, y; i, j)≡ (︃
i+ j j
)︃
q
xi−jyi+j, i, j = 0, . . . , n − 1. (37)
Theorem 4.1. A q-analogue of [13, p.175]. In the following two formulas we use the q-matrix multiplication with f(m, i, j) = m2.
Ψn,q(x, y) = Rn,q(xy)Φn,q(y,1
x)T, (38)
Ψn,q(x, y) = Φn,q(x, y)Pn,q(y
x)T. (39)
Proof. By the first q-Vandermonde theorem,
Rn,q(xy)Φn,q(y,1
x)T(i, j) =
n−1
∑︁
k=0
(︃
i k
)︃
q
(xy)i+kyj−kx−j−k (︃
j k
)︃
q
qk2 = (︃
i+ j j
)︃
q
xi−jyi+j. (40)
Φn,q(x, y)Pn,q(y
x)T(i, j) =
n−1
∑︁
k=0
(︃
i k
)︃
q
xi−jyi+j (︃
j k )︃
q
qk2 = (︃
i+ j j
)︃
q
xi−jyi+j. (41)
We have found two different matrix products for the same function.
Example2. The case n = 4:
⎛
⎜
⎜
⎜
⎜
⎝
1 yx
y2 x2
y3 x3
xy {2}qy2 {3}qy3
x {4}qy4 x2
x2y2 {3}qxy3 (︀4
2
)︀
qy4 (︀5
3
)︀
q y5
x
x3y3 {4}qx2y4 (︀5
2
)︀
qxy5 (︀6
3
)︀
qy6
⎞
⎟
⎟
⎟
⎟
⎠
=
⎛
⎜
⎜
⎜
⎝
1 0 0 0
xy x2y2 0 0
x2y2 {2}qx3y3 x4y4 0 x3y3 {3}qx4y4 {3}qx5y5 x6y6
⎞
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎝ 1 yx
y2 x2
y3 x3
0 x12 {2}q y
x3 {3}qy2 x4
0 0 x14 {3}q y x5
0 0 0 x16
⎞
⎟
⎟
⎟
⎠
=
⎛
⎜
⎜
⎜
⎝
1 0 0 0
xy y2 0 0
x2y2 {2}qxy3 y4 0 x3y3 {3}qx2y4 {3}qxy5 y6
⎞
⎟
⎟
⎟
⎠
⎛
⎜
⎜
⎜
⎝
1 yx yx22
y3 x3
0 1 {2}qy
x {3}qy2 x2
0 0 1 {3}qyx
0 0 1
⎞
⎟
⎟
⎟
⎠
. (42)
We note that we needed a q-matrix multiplication for the two last formulas and thus no inverse was available.
References
[1] W. A. Al-Salam, q-Bernoulli numbers and polynomials, Math. Nachr 17 (1959), 239–260.
[2] R. Brawer and M. Pirovino, The linear algebra of the Pascal matrix, Linear Algebra Appl. 174 (1992), 13–23.
[3] T. Ernst, q-Leibniz functional matrices with applications to q-Pascal and q-Stirling matrices, Adv. Stud. Contemp. Math., Kyungshang 22 (2012), 537-555.
[4] T. Ernst, q-Pascal and q-Wronskian matrices with implications to q-Appell polynomials, J. Discrete Math., (2013), Article ID 450481, 10 p.
[5] T. Ernst, A comprehensive treatment of q-calculus, Birkhäuser 2012.
[6] T. Ernst, An umbral approach to find q-analogues of matrix formulas, Linear Algebra Appl. 439 (2013), 1167–1182.
[7] T. Ernst, Faktorisierungen von q-Pascalmatrizen (Factorizations of q-Pascal matrices), Algebras Groups Geom. 31 (2014), no. 4, 387-405
[8] H. Exton, q-Hypergeometric functions and applications, Ellis Horwood 1983.
[9] F.H. Jackson, A basic-sine and cosine with symbolical solution of certain differential equations, Proc. Edinburgh Math. Soc.
22 (1904), 28–39.
[10] P. Nalli, On a calculation procedure similar to integration, (Sopra un procedimento di calcolo analogo all integrazione) (Italian), Palermo Rend 47 (1923), 337–374.
[11] M. Ward, A calculus of sequences, Amer. J. Math. 58 (1936), 255–266.
[12] Z. Zhang, The linear algebra of the generalized Pascal matrix, Linear Algebra Appl. 250 (1997), 51–60.
[13] Z. Zhang and M. Liu, An extension of the generalized Pascal matrix and its algebraic properties, Linear Algebra Appl. 271 (1998), 169–177.