Multiple monotonicity of discrete distributions: the case of the Poisson model (Q2002570): Difference between revisions

For an integer $k\geq1$, the function $p$ is said to be $k$-monotone if $(-1)^k\Delta^{(k)}p(i)\geq0$ for all $i$, where $\Delta^{(k)}$ represents the forward difference operator applied $k$ times. This generalizes monotonicity (corresponding to $k=1$) and convexity ($k=2$), and has been well studied in the literature. The main result of this paper is that the mass function of a Poisson random variable with mean $\lambda$ is $k$-monotone if and only if $\lambda\leq\lambda_k$, where $\lambda_k$ is the smallest root of the $k$th Laguerre polynomial given by \[ L_k(x)=\sum_{j=0}^k(-1)^j\binom{k}{j}\frac{x^j}{j!}, \] for $x\geq0$. This also implies conditions under which a mixed Poisson distribution has a $k$-monotone mass function, and shows that no Poisson distribution with $\lambda>0$ has a mass function which is $k$-monotone for all $k$. The proof relies on properties of the roots of generalized Laguerre polynomials. In part, this paper corrects an error in the earlier work [the third author, ibid. 11, No. 1, 1--49 (2017; Zbl 1357.62164)].