Power-fractional distributions and branching processes
Gerold Alsmeyer 1, Viet Hung Hoang 2 · 2025
Read Power-fractional distributions and branching processes by Gerold Alsmeyer 1, Viet Hung Hoang 2 (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-2503-18563.txt.
Introduction
Alsmeyer and Hoang study a new family of chance models for branching processes in twenty twenty-five. A branching process tracks how a group grows as each member has a random number of offspring. Simple models give neat formulas, but they cover only a narrow set of cases. The new family aims to keep the neat math while covering more cases.
The authors show that the new models have a heavy tail, so rare large outcomes matter. They prove results for family trees that grow in steady settings and in settings that change at random. They also treat trees that grow in smooth time. The work gives modelers a broader tool with clear formulas.
Catalog record
| CREATOR | Gerold Alsmeyer 1, Viet Hung Hoang 2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-18563
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Power-fractional distributions and branching processes Gerold Alsmeyer 1, Viet Hung Hoang 2 Abstract In branching process theory, linear-fractional distributions are commonly used to model individual reproduction, especially when the goal is to obtain more explicit formulas than those derived under general model assumptions. In this article, we explore a generalization of these distributions, first introduced by Sagitov and Lindo, which offers similar advantages. We refer to these as power-fractional distributions, primarily because, as we demonstrate, they exhibit power-law behavior. Along with a discussion of their additional properties, we present several results related to the Galton-Watson branching process in both constant and randomly varying environments, illustrating these advantages. The use of power-fractional distributions in continuous time, particularly within the framework of Markov branching processes, is also briefly addressed. † † tocauthor: Ivar Ekeland, Roger Temam, Jeffrey Dean, David Grove, Craig Chambers, Kim B. Bruce, and Elisa Bertino † † institute: Inst. Math. Stochastics, Department of Mathematics and Computer Science University of Münster, Orléans-Ring 10, D-48149 Münster, Germany and Faculty of Fundamental Science, Industrial University of Ho Chi Minh City, No. 12 Nguyen Van Bao, Ward 4, Go Vap District, Ho Chi Minh City, Vietnam 1 Introduction It is a well-known fact in the theory of branching processes that the linear-fractional distribution \mathsf{LF}(a,b) with probability generating function (pgf) f(s)\ =\ 1-\left[\frac{a}{1-s}+b\right]^{-1} for parameters a,b>0 with a+b\geqslant 1 , is stable under iterations of f , namely \frac{1}{1-f^{n}(s)}\ =\ \frac{a^{n}}{1-s}+b_{n},\quad b_{n}\,:=\,b\sum_{k=0}^{n-1}a^{k} (1.1) for each n\in\mathbb{N} , and so f^{n} , the n -fold iteration of f , equals the pgf of \mathsf{LF}(a^{n},b_{n}) . But f^{n} is also the pgf of Z_{n} for each n if (Z_{n})_{n\geqslant 0} is a Galton-Watson branching process with one ancestor and offspring law \mathsf{LF}(a,b) . Therefore, all Z_{n} have a linear-fractional distribution, namely Z_{n}\stackrel{{\scriptstyle d}}{{=}}\mathsf{LF}(a^{n},b_{n}) , where \stackrel{{\scriptstyle d}}{{=}} means equality in law. Sagitov and Lindo in ( 14 , Section 14.2) introduced an extended class of possibly defective distributions on \mathbb{N}_{0} that are stable under iterations of their pgf’s. Any such distribution, here called power-fractional law and abbreviated as \mathsf{PF}(\theta,\gamma,a,b) , involves two further parameters \theta\in[-1,1]\backslash\{0\} and \gamma\geqslant 1 . Its pgf has the form \displaystyle f(s)\ =\ \gamma-\left[\frac{a}{(\gamma-s)^{\theta}}+b\right]^{-1/\theta},\quad s\in[0,\gamma). (1.2) where \theta,\gamma,a,b must satisfy the conditions of one of the cases (A1–3) shown in the Table 1 . Case \theta \gamma a,\,b\qquad (A1) \theta\in(0,1] \gamma=1 a,b>0 a+b\geqslant 1 (A2) \theta\in(0,1] \gamma>1 a\in(0,1) \gamma^{-\theta}\leqslant\frac{b}{1-a}\leqslant(\gamma-1)^{-\theta} (A3) \theta\in[-1,0) \gamma\geqslant 1 a\in(0,1) (\gamma-1)^{|\theta|}\leqslant\frac{b}{1-a}\leqslant\gamma^{|\theta|} Table 1: The constraints on the parameters \theta,\gamma,a and b The linear-fractional laws \mathsf{LF}(a,b) are obtained for \theta=\gamma=1 , but the subcritical ones also for \theta=1 and \gamma>1 (see Comment (d) below). With q denoting the minimal fixed point of f in [0,1] , the last constraint in (A2) and (A3) may be restated as \displaystyle b\,=\,\frac{1-a}{(\gamma-q)^{\theta}}\quad\text{and}\quad q\in\,\begin{cases}[0,1]&\text{if }\gamma>1,\\ [0,1)&\text{if }\gamma=1,\end{cases} see ( 14 , Def. 14.1) . One can also readily check that Eq. ( 1.2 ) is equivalent to \displaystyle\frac{1}{(\gamma-f(s))^{\theta}}\ =\ \frac{a}{(\gamma-s)^{\theta}}+b,\quad s\in[0,\gamma). (1.3) The parameter \theta=0 appears as a limiting case, namely, \displaystyle f(s)=\gamma-(\gamma-q)^{1-a}(\gamma-s)^{a} (1.4) or, equivalently, \displaystyle\gamma-f(s)\ =\ (\gamma-q)^{1-a}(\gamma-s)^{a},\quad s\in[0,\gamma). (1.5) This can be deduced from ( 1.2 ) by a continuity argument when observing that \displaystyle\lim_{\theta\to 0}\left[\frac{a}{(\gamma-s)^{\theta}}+\frac{1-a}{(\gamma-q)^{\theta}}\right]^{-1/\theta}\ =\ (\gamma-q)^{1-a}(\gamma-s)^{a}. Here the parameters are \gamma,q and a , and they must satisfy a\in(0,1) and either \gamma=1,\,q\in[0,1) , or \gamma>1,\,q\in[0,1] . We summarize the previous definitions (without repeating the constraints on \gamma,a,b and q ) as \displaystyle f(s)\ =\ \begin{cases}\hfill\displaystyle\gamma\,-\,\frac{\gamma-s}{\big[a+b(\gamma-s)^{\theta}\big]^{1/\theta}}&\text{if }\theta\in(0,1],\\[11.38109pt] \hfill\gamma\,-\,(\gamma-q)^{1-a}(\gamma-s)^{a}&\text{if }\theta=0,\\[2.84526pt] \gamma\,-\,\big[a(\gamma-s)^{|\theta|}+b\big]^{1/|\theta|}&\text{if }\theta\in[-1,0)\end{cases} (1.6) and also note that the iterations of f are easily computed with the help of Eqs. ( 1.3 ) and ( 1.5 ) as \displaystyle f^{n}(s)\ =\ \begin{cases}\hfill\displaystyle\gamma\,-\,\frac{\gamma-s}{\big[a^{n}+b_{n}(\gamma-s)^{\theta}\big]^{1/\theta}}&\text{if }\theta\in(0,1],\\[11.38109pt] \hfill\gamma\,-\,(\gamma-q)^{1-a^{n}}(\gamma-s)^{a^{n}}&\text{if }\theta=0,\\[2.84526pt] \gamma\,-\,\big[a^{n}(\gamma-s)^{|\theta|}+b_{n}\big]^{1/|\theta|}&\text{if }\theta\in[-1,0).\end{cases} (1.7) for any n\geqslant 1 , where b_{n}=b\sum_{k=0}^{n-1}a^{k} as in ( 1.1 ). Due to the leading parameter \theta , Sagitov and Lindo [ 14 ] coined the term theta-branching process for a Galton-Watson process ( \mathsf{GWP} ) with power-fractional offspring law. We continue with some notable facts and supplements in connection with the above definitions: (a) As pointed out in ( 13 , Sect. 1) , Eq.’s ( 1.3 ) and ( 1.5 ) are both special cases of a general functional equation, namely H(f(s))\ =\ aH(s)\,+\,H(f(0)) (1.8) where H(s)=H_{\theta,\gamma}(s)=(\gamma-s)^{-\theta}-\gamma^{-\theta} in ( 1.3 ), ( 1.4 ) and H(s)=H_{0,\gamma}(s)=\log\gamma-\log(\gamma-s) in ( 1.5 ). Regarding the minimal fixed point q of f , it follows that H(q)\ =\ \begin{cases}\displaystyle\frac{H(f(0))}{1-a}&\text{if }a\in(0,1),\\[5.69054pt] \hfil\infty&\text{if }a\geqslant 1\text{ and thus }\theta\in(0,1],\,\gamma=1,\,q=1.\end{cases} (b) Since q equals the smallest fixed point of f in [0,1] , it is also the extinction probability of the associated branching process and therefore the notational choice in accordance with a common convention in the branching process literature. (c) We always have that f(1)\leqslant 1 , and f(1)=1 holds iff one of the cases shown in Table 2 occurs. Notably, b is a unique function of a (and \theta,\gamma ) in the cases (A.2.1) and (A.3.1), giving \displaystyle f(s)\,=\,\gamma\,-\,\bigg[\frac{a}{(\gamma-s)^{\theta}}+\frac{1-a}{(\gamma-1)^{\theta}}\bigg]^{1/\theta} (1.9) and \displaystyle f(s)\,=\,\gamma\,-\,\big[a(\gamma-s)^{|\theta|}+(1-a)(\gamma-1)^{|\theta|}\big]^{1/|\theta|}, (1.10) respectively. Case \theta \gamma a b (A1) \theta\in(0,1] \gamma=1 a>0 b>0,\,b\geqslant 1-a (A2.1) \theta\in(0,1] \gamma>1 a\in(0,1) b=(1-a)(\gamma-1)^{-\theta} (A3.1) \theta\in[-1,0) \gamma\geqslant 1 a\in(0,1) b=(1-a)(\gamma-1)^{|\theta|} Table 2: The constraints on the parameters \theta,\gamma,a and b that are required for f(1)=1 . Thus, a power-fractional law can be defective or, equivalently, assign mass to \infty , in which case the \mathsf{GWP} with this offspring law exhibits explosion in finite time. (d) We have already pointed out that the class of linear-fractional distributions is obtained when \theta=\gamma=1 , i.e. \mathsf{LF}(a,b)\,=\,\mathsf{PF}(1,1,a,b)\quad\text{for all }a,b>0\text{ such that }a+b\geqslant 1. On the other hand, the choice \theta=1 and \gamma>1 obviously also leads to linear-fractional laws. In fact, if \mathsf{LF}(a,b) is subcritical, meaning a>1 , then the associated pgf f has the two fixed points 1 and \widehat{q}=b^{-1}(a+b-1)>1 (in the supercritical case <1 and the extinction probability), and one can readily verify that f satisfies ( 1.3 ) not only for \gamma=1 , but also for \gamma=\widehat{q} . Namely, \displaystyle\frac{1}{\widehat{q}-f(s)}\ =\ \frac{a^{-1}}{\widehat{q}-s}\,+\,\frac{b}{a},\quad s\in[0,\widehat{q}\,), and therefore \displaystyle\mathsf{PF}(1,1,a,b)\ =\ \mathsf{PF}\left(1,\frac{a+b-1}{b},\frac{1}{a},\frac{b}{a}\right)\quad\text{for all }a>1,\,b>0. The reader can check that this aligns with (A2.1) by substituting the pair (a^{-1},b^{-1}(a+b-1)) for (a,\gamma) . Therefore, we conclude that the parametrization of a subcritical linear-fractional law within the class of power-fractional laws is not unique. (e) Let us also mention that, if \theta\neq 0 and \psi(s):=\left[a+b(\gamma-s)^{\theta}\right]^{-1/\theta} , then f^{\prime}(s)\ =\ a\psi(s)^{\theta+1}\quad\text{and}\quad f^{\prime\prime}(s)\ =\ ab(\theta+1)(\gamma-s)^{\theta-1}\psi(s)^{2\theta+1}. The resulting values for \mathsf{m}=f^{\prime}(1) (the mean number of offspring und thus criticality parameter) and f^{\prime\prime}(1) in the three proper cases (A1), (A2.1) and (A3.1) are listed in Table 3 below, keeping in mind that in the last two cases b is uniquely determined by the other parameters. We see that the critical case \mathsf{m}=1 occurs only if \theta>0 and a=1 , and the supercritical case only if \theta>0 and a\in(0,1) . This can be viewed as some evidence for Case (A1) providing the most interesting subclass of power-fractional distributions in the context of branching processes. Case \theta \gamma \mathsf{m}=f^{\prime}(1) f^{\prime}\hskip-1.0pt{}^{\prime}(1) (A1) \theta\in(0,1) \gamma=1 a^{-1/\theta} \infty (A1) \theta=1 \gamma=1 a^{-1} 2a^{-2}b (A2.1) \theta\in(0,1] \gamma>1 a a(1-a)(1+\theta)/(\gamma-1) (A3.1) \theta\in[-1,0) \gamma>1 a a(1-a)(1-|\theta|)/(\gamma-1) (A3.1) \theta\in[-1,0) \gamma=1 a^{1/|\theta|} 0 Table 3: First and second derivative of f at 1 for the proper cases when f(1)=1 . (f) The Sibuya distribution \mathsf{Sib}\hskip 0.6pt(a) for a\in(0,1) , named after Sibuya [