Decompositions and eigenvectors of Riordan matrices (Q2133678): Difference between revisions

For a field \(\mathbb{F}\) of real or complex numbers, let \(\mathbb{F}[[x]]\) be the ring of formal power series over \(\mathbb{F}\). A Riordan matrix \(A = [a_{ij}]_{i,j \in \mathbb{N}}\) over \(\mathbb{F}\) is an infinite lower triangular matrix whose entries are determined by \(a_{ij} = [x^i]g(x)F(x)^j\) or \(g(x)F(x)^j =\sum_{i \geq j} a_{ij} x^i\) for some \(g, F \in \mathbb{F}[[x]]\) such that \(g(0) \not = 0, F(0) = 0, F'(0) \not= 0\). The Riordan matrix is denoted by \(A = (g(x), F(x)) \) or simply \(A =(g, F)\). As an example, the Pascal matrix, where \(a_{ij}=\binom{i}{j}\) and whose \(j\)-th column has the formal power series as its generating function \[\frac{1}{1-x}\left(\frac{x}{1-x}\right)^j=\sum_{i \geq j}\binom{i}{i} x^i, ~j \geq 0,\] is a Riordan matrix given by \(A=(\frac{1}{1-x},\frac{x}{1-x})\). It is known that the set of all Riordan matrices with entries in \(\mathbb{F}\) forms a group under the usual matrix multiplication in terms of generating functions, \((g,F)(h,L) =(g\cdot h(F), L(F)),\) called the Riordan group over \(\mathbb{F}\) and denoted by \(\mathcal{R}(\mathbb{F})\). The paper studies the algebra of the Riordan group. In particular, by studying decomposition of Riordan matrices, the authors give a new factorization of a Riordan matrix in terms of almost-Riordan arrays. It is also shown that if a Riordan matrix \(A\) is an \(n \times n\) pseudo-involution, then the singular values of \(A\) must come in reciprocal pairs. Further, it is described a complete analysis of existence and nonexistence of the eigenvectors of Riordan matrices. An eigenvector of a Riordan matrix \((g, F)\) of the form \(h_k = (0, \ldots , 0, h_k, h_{k+1}, \ldots)^T\) with \(h_k \not = 0 \) is an eigenvector of level \(k\), and the corresponding eigenvalue is \(\lambda_k = g(0)([x]F)^k\). A full set of eigenvectors is defined to be a set of eigenvectors \(\{h_0, h_1, \ldots , h_k, \ldots\}\), in which every possible eigenvector level is achieved. As a main result of the paper, the authors prove that the Riordan group \(\mathcal{R}(\mathbb{F})\) has a partition \[\mathcal{R}(\mathbb{F}) = \mathcal{R}_{\mathrm{full}} \cup \mathcal{R}_{\mathrm{none}} \cup \bigcup_{k=0}^{\infty} \mathcal{R}_k,\] where \(\mathcal{R}_{\mathrm{full}}\) are Riordan matrices having full sets of eigenvectors, \(\mathcal{R}_{\mathrm{none}}\) have no eigenvectors and \(\mathcal{R}_k\) have a one-dimensional set of eigenvectors, all of which are of level \(k\). Finally, for a given nonzero vector \(v\), the authors analyze the Riordan matrices in the stabilizer Riordan subgroup for which \(Av = v\), that is to say, they establish conditions to obtain Riordan matrices \(A=(g,F)\), for which \(Av = v\). The method the authors use is the study of properties of Riordan matrices by means of formal power series from the viewpoint of linear algebra. Along the way, the authors also obtain known results, whose proofs are new and simpler.