Connecting the q-multiplicative convolution and the finite difference convolution
The Walsh additive and Grace-Szegö multiplicative polynomial convolution on \(f,g\in \mathbb{C}_{n}[x]\), denoted by \(\boxplus^{n}\) and \(\boxtimes^{n}\) respectively, are defined by \[ f\boxplus^{n} g = \frac{1}{n!} \sum^{n}_{k=0}\partial^{k}_{x}f\cdot (\partial_{x}^{n-k}g)(0) \quad \text{and} \quad f\boxtimes^{n} g = \sum^{n}_{k=0}\binom{n}{k}^{-1} (-1)^{k} f_{k}g_{k}x^{k}, \] where \(f_{k}\) and \(g_{k}\) denote the coefficient of \(x^{k}\) of \(f\) and \(g\), respectively. It is known that the additive convolution can only increase root mesh, which is defined as the minimum absolute difference between any pair of roots of a given polynomial. As well, for polynomials with only non-negative roots, the multiplicative convolution can only increase logarithmic root mesh. Properties of logarithmic mesh preservation are also known for the \(q\)-multiplicative convolution. This convolution is defined as follows, where \(p_{k}\) and \(r_{k}\) are the coefficients of \(x^{k}\) of \(p\) and \(r\), \[ p\boxtimes_{q}^{n} r=\sum^{n}_{k=0}\binom{n}{k}^{-1}_{q}q^{-\binom{k}{2}}(-1)^{k}p_{k}r_{k}x^{k}, \] with \((x)_{q}\) meaning \(\frac{1-q^{x}}{1-q}\). Writing \(\operatorname{lmesh}(p)\) for the minimum ratio between any pair of non-zero roots of \(p\), the result is that, for \(p\), \(r\) polynomials with real roots, one has \(\operatorname{lmesh}(p\boxtimes_{q}^{n}r)\ge q\) whenever \(\operatorname{lmesh}(p)\ge q\) and \(\operatorname{lmesh}(r)\ge q\) for some \(q\in (1,\infty)\). In this paper, the authors show that the \(b\)-additive convolution preserves the space of polynomials with root mesh at least \(b\). This gives a characterization of finite difference operators which preserve root mesh at least \(b\). The \(b\)-additive convolution of \(p\) and \(r\) is \[ p\boxplus_{b}^{n} r=\frac{1}{n!}\sum^{n}_{k=0}\Delta_{b}^{k} p\cdot (\Delta_{b}^{n-k}r)(0), \] where \(\Delta_{b}\) is a finite \(b\)-difference operator defined by \[ \Delta_{b}: p\mapsto \frac{p(x)-p(x-b)}{b}. \] A real-rooted polynomial \(p\) is \(b\)-mesh if the minimum non-negative difference of any pair of roots of \(p\) is at least \(b\). In this situation, one writes \(\operatorname{mesh}(p)\ge b\). The following result answers an open question: \begin{itemize} \item Let \(p\) and \(r\) be polynomials of degree at most \(n\) such that \(\operatorname{mesh}(p) \ge b\) and \(\operatorname{mesh}(r) \ge b\), for some \(b\in (0,\infty)\). Then, \(\operatorname{mesh}(p\boxplus_{b}^{n} r) \ge b\). \end{itemize} The proof is based in a way to pass root properties of the \(q\)-multiplicative convolution to the \(b\)-additive convolution. To formulate it one needs some notation. For \(b\ge 0\) and \(q\ge 0\) consider the following bases of \(\mathbb{C}[x]\): \begin{align*} v_{q,b}^{k}&= \frac{(1-x)(1-q^{b}x)\dotsm (1-q^{(k-1)b}x)}{(1-q)^{k}},\\ \nu_{b}^{k}&=x(x+b)(x+2b)\dotsm (x+(k-1)b), \end{align*} and define the following generalized ``exponential map: \[ E_{q,b}: \nu_{b}^{k}\mapsto v_{q,b}^{k}. \] The main result, which gives an analytic link between the \(b\)-additive and \(q\)-multiplicative convolutions is the following: \begin{itemize} \item Fix \(b\ge 0\) and let \(p\), \(r\) be polynomials of degree \(n\). We have the following, where convergence is uniform on compact sets: \[ \lim_{q\to 1}(1-q)^{n} \Bigl[E_{q,b}(p)\boxtimes_{q^{b}}^{n} E_{q,b}(r)\Bigr] (q^{x})=p\boxplus_{b}^{n}r. \] \end{itemize} The corresponding result in the classical case, which gives an analytic connection between the additive and multiplicative convolutions is \begin{itemize} \item Let \(p,r\in\mathbb{C}[x]\) be of degree at most \(n\). We have the following: \[ \lim_{q\to 1}(1-q)^{n} [E_{q,0}(p)\boxtimes^{n} E_{q,0}(r)] (q^{x})=p\boxplus^{n}r. \] \end{itemize} Finally, the authors prove that the \(b\)-additive convolution preserves a root-interlacing property for \(b\)-mesh polynomials. Let \(f,g\in \mathbb{R}[x]\) two real-rooted polynomials with positive leading coefficients and roots \(\alpha_{1}\ge\dotsb\ge\alpha_{n}\) and \(\beta_{1}\ge\dotsb\ge \beta_{m}\), respectively. If \(n-m\in \{0,1\}\), then one says that the roots of \(g\) interlaces the roots of \(f\) whenever \[ \alpha_{1}\ge \beta_{1}\ge\alpha_{2}\ge \beta_{2}\ge\alpha_{3}\ge\dotsb \] Then one has \begin{itemize} \item Let \(f,g\in \mathbb{R}_{n}[x]\) be \(b\)-mesh polynomials of degree \(n\). Let \(T_{g}: \mathbb{R}_{n}[x]\to \mathbb{R}_{n}[x]\) be the real linear operator defined by \(T_{g}: r\mapsto r\boxplus_{b}^{n}g\). Then, \(T_{g}\) preserves the set of polynomials whose roots interlace the roots of \(f\). \end{itemize}
- Elements of Pólya-Schur theory in the finite difference setting
- scientific article; zbMATH DE number 3507071 (Why is no real title available?)
- Multivariate stable polynomials: theory and applications
- Suffridge's convolution theorem for polynomials and entire functions having only real zeros
- The Story of the Binomial Theorem
This page was built for publication: Connecting the \(q\)-multiplicative convolution and the finite difference convolution
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2006060)
