On necessary conditions of rational-infinite divisibility for distributions with non-zero discrete parts

A. A. Khartov · 2025

Read On necessary conditions of rational-infinite divisibility for distributions with non-zero discrete parts by A. A. Khartov (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07244.txt.

Introduction

Alexey Khartov, in a solo 2025 paper, studies a new class of probability distributions called rational-infinitely divisible. A distribution joins this class if some infinitely divisible distribution splits into it convolved with another such distribution. Convolution combines two random effects into one total. The class is much wider than the classic infinitely divisible family and already has uses in several fields.

Khartov focuses on members that have a nonzero discrete part, meaning some probability sits on single points. He proves two necessary conditions for such members. In particular, the characteristic functions, which encode a distribution in frequency form, of both the distribution and its discrete part must stay away from zero. Conditions like these help researchers test whether a given law can belong to the class at all. The paper continues a run of results that map the borders of this enlarged family.

Catalog record

CREATORA. A. Khartov
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07244

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On necessary conditions of rational-infinite divisibility for distributions with non-zero discrete parts A. A. Khartov Abstract We consider the new class 𝑸 \boldsymbol{Q} of rational-infinitely (or quasi-infinitely) divisible distribution functions on the real line. By definition, F ∈ 𝑸 F\in\boldsymbol{Q} if there are some infinitely divisible distribution functions F 1 F_{1} and F 2 F_{2} such that F 1 = F ∗ F 2 F_{1}=F*F_{2} , where “ ∗ * ” is the convolution. The characteristic function of such F F admits the Lévy–Khintchine-type representation with a “signed spectral measure”. The class 𝑸 \boldsymbol{Q} is a significant extension of the family of infinitely divisible distribution functions and it have already found some applications in several areas. So there is an active interest in this class. In particular, a lot of results have recently appeared on the problem of belonging to the class 𝑸 \boldsymbol{Q} in terms of characteristic functions. In the paper, we continue this series of results by proposing two necessary conditions for distribution functions from 𝑸 \boldsymbol{Q} with non-zero discrete parts. Namely, the characteristic functions of such a distribution function and its discrete part are always separated from zero. 1 1 footnotetext: Institute for Information Transmission Problems (Kharkevich Institute) of Russian Academy of Sciences, Bolshoy Karetny per. 19, build.1, 127051 Moscow, Russia. 2 2 footnotetext: Smolensk State University, 4 Przhevalsky st., 214000 Smolensk, Russia. 3 3 footnotetext: ITMO University, 49 Kronverksky Pr., 197101 Saint-Petersburg, Russia. Keywords and phrases : infinite divisibility, rational-infinite divisibility, quasi-infinite divisibility, characteristic functions, the Lévy–Khintchine-type representation. 1 Introduction Let F be a distribution function on the real line \mathbb{R} . Let f be the characteristic function of F , i.e. \displaystyle f(t):=\int_{\mathbb{R}}e^{itx}\,dF(x),\quad t\in\mathbb{R}. We call a distribution function F (and the corresponding probability law) rational-infinitely divisible if there are some infinitely divisible distribution functions F_{1} and F_{2} on \mathbb{R} such that F_{1}=F*F_{2} , where “ * ” denotes the convolution operation. This equality may be written in the form f(t)=f_{1}(t)/f_{2}(t) , t\in\mathbb{R} , where f_{1} and f_{2} denote the characteristic functions of F_{1} and F_{2} , respectively. Let us denote by \boldsymbol{Q} the class of all rational-infinitely divisible distribution functions. It is clear that \boldsymbol{Q} is an extension of the class \boldsymbol{I} of all infinitely divisible distribution functions. There are some examples of distribution functions from \boldsymbol{Q}\setminus\boldsymbol{I} in the classical monographs [ 10 ] (p. 81–83) and [ 21 ] (p. 165). The classes \boldsymbol{I} and \boldsymbol{Q} have very similar properties. So it is seen from the definition that the characteristic function f of any F\in\boldsymbol{Q} doesn’t have zeroes on \mathbb{R} and it admits the Lévy–Khinchine-type representation : \displaystyle f(t)=\exp\biggl\{it\gamma+\int_{\mathbb{R}}\Bigl(e^{itx}-1-it\sin(x)\Bigr)\dfrac{1+x^{2}}{x^{2}}\,dG(x)\biggr\},\quad t\in\mathbb{R}, (1) with the shift parameter \gamma\in\mathbb{R} , and the spectral function G:\mathbb{R}\to\mathbb{R} of bounded total variation on \mathbb{R} . The function G is assumed to be right-continuous at every point of the real line with the condition G(-\infty)=0 . Note that G is non-monototic in general. Nevertheless, the spectral pair (\gamma,G) is uniquely determined by f and hence by F as for the class \boldsymbol{I} (see [ 10 ] , p. 80). Next, let f be a characteristic function represented by formula ( 1 ) with some \gamma and G satisfying the above-mentioned conditions. Following Lindner and Sato [ 20 ] , the corresponding distribution function F (and the corresponding probability law) is called quasi-infinitely divisible . Due to the Hahn–Jordan decomposition for G , it is not difficult to show that F is rational-infinitely divisible. The class \boldsymbol{Q} is actively studied now. The main reference is the paper [ 19 ] by Lindner, Pan, and Sato, which contains the detailed analysis of this class. Multivariate analog of \boldsymbol{Q} is considered in [ 2 ] , [ 5 ] , and [ 6 ] . There are interesting applications in theory of stochastic processes (see [ 20 ] and [ 25 ] ), number theory (see [ 23 ] ), physics (see [ 7 ] and [ 9 ] ), and insurance mathematics (see [ 26 ] ). The class \boldsymbol{Q} penetrates statistics (see [ 24 ] ). This short note is connected with the problem of belonging to the class \boldsymbol{Q} . The general statement of the problem is to obtain the necessary and/or sufficient conditions for which a given distribution function F belongs to \boldsymbol{Q} . For instance, there is the following simple fact (see [ 8 ] , Theorem 4.3.7): if a probability law has a mass >1/2 at some point, then its distribution function belongs to \boldsymbol{Q} . By now a number of deeper results have accumulated. We will give a brief survey of them now. Let F be a distribution function on \mathbb{R} . According to the Lebesgue decomposition theorem, F admits the representation: \displaystyle F(x)=c_{d}F_{d}(x)+c_{a}F_{a}(x)+c_{s}F_{s}(x),\quad x\in\mathbb{R}, (2) where F_{d} , F_{a} , and F_{s} are discrete, absolutely continuous and continuous singular distribution functions, respectively. Here the coefficients c_{a} , c_{a} , and c_{s} are non-negative constants such that c_{d}+c_{a}+c_{s}=1 . The characteristic function f of F is represented accordingly: \displaystyle f(t)=c_{d}f_{d}(t)+c_{a}f_{a}(t)+c_{s}f_{s}(t),\quad t\in\mathbb{R}, (3) where f_{d} , f_{a} , and f_{s} are the characteristic functions corresponding to F_{d} , F_{a} , and F_{s} , respectively. The most of existing results are formulated in terms of characteristic functions. The first rather general result was obtained by Lindner, Pan, and Sato in [ 19 ] (Theorem 8.1, p. 30) for the case c_{d}=1 and discrete lattice F=F_{d} . It states that such F belongs to the class \boldsymbol{Q} if and only if f(t)\neq 0 for any t\in\mathbb{R} . The criterion for arbitrary discrete F (exactly the case c_{d}=1 ) was obtained in [ 11 ] (see also [ 1 ] ): F\in\boldsymbol{Q} if and only if f is separeted from zero, i.e. \inf_{t\in\mathbb{R}}|f(t)|>0 . This criterion generalizes the previous one, because the absolute value |f(\,\cdot\,)| is a periodic continuous function on \mathbb{R} for lattice F and therefore, in the lattice case, f is zero-free on the period segment (and hence on \mathbb{R} ) if any only if it is separated from zero. By the way, there is a recent result in paper [ 17 ] , where this criterion is formulated in terms of the support points of the discrete distribution and their probabilities (see also [ 16 ] ). Next, in the paper [ 4 ] , Berger and Kutlu considered the case c_{d}>0 , c_{a}\geqslant 0 , and c_{s}=0 . They proved that, under these assumptions, F\in\boldsymbol{Q} if and only if 1) f(t)\neq 0 for any t\in\mathbb{R} and 2) \inf_{t\in\mathbb{R}}|f_{d}(t)|>0 . Conditions 1) and 2) together are equivalent to one condition 3) \inf_{t\in\mathbb{R}}|f(t)|>0 . We note that this result is a generalization of an earlier criterion by Berger (see [ 3 ] ), where F_{d} is assumed to be lattice. Next, it is asserted in [ 15 ] that the previous criterion is valid under a more general assumption of the dominated continuous singular part. More precisely, let us assume that c_{d}>0 , c_{a}\geqslant 0 , c_{s}\geqslant 0 , and define \mu_{d}:=\inf_{t\in\mathbb{R}}|f_{d}(t)| . We additionally assume that c_{s}<c_{d}\mu_{d} if \mu_{d}>0 and c_{s}=0 if \mu_{d}=0 . Then F\in\boldsymbol{Q} if and only if conditions 1) and 2) hold together or equivalent condition 3) is satified. It is important to note,

Read the complete text