Sharp and Exact Tail Estimates for Multitype Poissonian Galton Watson Processes and Inhomogeneous Cluster Processes

Théo Leblanc · 2025

Read Sharp and Exact Tail Estimates for Multitype Poissonian Galton Watson Processes and Inhomogeneous Cluster Processes by Théo Leblanc (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-2507-08462.txt.

Introduction

Theo Leblanc, writing alone in 2025, proves sharp bounds on the size of branching family trees. The trees model processes where each parent has a random number of offspring, across several types. Such trees also sit inside Hawkes processes, which model events that trigger more events. The paper bounds how fast the chance of a huge total tree size falls off.

The bounds are stated in terms of the average offspring matrix and the weights of a chosen summary. Type-by-type bounds follow for Hawkes processes, with better results in high dimensions. Bounds for clustered point patterns reflect how the pull between events fades. The results give analysts tools for rare-event risks in branching systems.

Catalog record

CREATORThéo Leblanc
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE91 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-08462

Browse the full library · About the Tower of Babel

Full text

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

Sharp and Exact Tail Estimates for Multitype Poissonian Galton Watson Processes and Inhomogeneous Cluster Processes Théo Leblanc Abstract . We prove exponential moments for linear combinations of the number of individuals of each type of a whole multitype Poissonian Galton Watson process. We give sharp estimates for such quantities, which depend on the expectation of the Poissonian offspring distribution and the linear combination. We apply this result to derive type specific exponential moments for multitype Hawkes processes, improving existing results when the dimension is large. We also prove exponential estimates for tails of inhomogeneous Poissonian clusters. These estimates depend on the decay of the interaction functions defining the clusters and are starting points to derive probabilistic properties on branching processes such as Hawkes processes. 2020 Mathematics Subject Classification 60J80, 60E15, 60G55, 60J85, 47J26 1. Introduction Galton Watson processes are one of the simplest examples of branching processes. They are starting points for many more complex processes. Galton Watson processes describe a whole family coming from a single ancestor. The key property of Galton Watson processes is that each individual reproduces independently of the others and has a random number of children according to a common offspring distribution, see [ 1 , 7 ] . In the present article we focus on the case where the offspring distribution is a Poisson distribution. This particular case remains quite general and fits the case of Hawkes processes, which have applications in biology, epidemiology, finance, neuroscience, earthquake modeling to name but a few, see for example [ 8 , 12 , 10 , 2 ] . The link between Poissonian Galton Watson processes and Hawkes processes lies in Poissonian clusters. Hawkes processes admit a cluster decomposition, where clusters arrive according to a Poisson process and each cluster consists in a Galton Watson tree embedded in the real line, see [ 9 ] . In this paper we derive estimates for the Laplace transform and for the tails of multitype Galton Watson processes and clusters. The branching properties of such objects lead to fix point equations for the Laplace transforms. In [ 16 ] the author studied these fixed point equations for general Galton Watson processes. Restricting ourselves to Poissonian processes allows us to derive sharper and more specific bounds. For Hawkes processes, the usual way to derive exponential moments for the number of points in a compact interval is to use cluster decomposition and sum all the contributions of each cluster. In previous work, see for example [ 13 ] and [ 6 ] , decay assumptions on the interaction functions were used to ensure that the sum of the contributions of each cluster converges. In [ 11 ] it was shown that in fact, the sum of the contributions only depends on the matrix \bm{H} gathering the L^{1} norms of the interaction functions, regardless of the decay of these functions. More precisely, the only relevant assumption is \spr(\bm{H})<1 , where \spr stands for the spectral radius, the weakest possible assumption since it is required for the process to exist in a stationary version, [ 4 ] . An important result established by [ 11 ] is the derivation of exponential moments for a whole multitype Galton Watson with offspring distribution {P}\!{o}\!{i}\!{s}(\bm{H}) . Even though the bound in [ 11 ] is quantitative, it quickly becomes unadapted in an high dimensional framework. Indeed, when the number of types (for example the number of neurons when a Hawkes process models brain activity) is large, one is typically interested in the number of points coming from one or from a small number of types. The bound in [ 11 ] becomes less significant since it is a bound for all the types at once and depends on the total number of types which could go to infinity. In this paper we derive exponential estimates for Poissonian Galton Watson processes that are type specific, quantitative and stated under the optimal assumption, i.e. \spr(\bm{H})<1 . This result applies to Hawkes processes, leading to quantitative and type specific exponential estimates which drastically improve the global bound proved in [ 11 ] when the dimension is large. In this work we also derive tail estimates for Galton Watson processes in terms of the generations of the individuals. More precisely, we provide sharp and non asymptotic estimates for the exponential moment of individuals whose generation is at least n with n\to\infty . This result is also stated in a type specific framework where the information about types is kept. We then study structural, topological and exact properties of the Laplace transform of Poissonian Galton Watson processes, leading for example to a precise description of the set of vectors for which the exponential moments are finite. Lastly we provide tail estimates for inhomogeneous Poissonian clusters in terms of the temporal embedding of the clusters, also in a type specific framework. Temporal tail estimates, for clusters, are derived in the following three cases. First when the interaction functions are exponentially decaying, second when the decay follows a power law, improving previous results where compactly supported interaction functions where considered in the single type case, [ 13 ] . And lastly when the interactions functions have bounded support where we obtains the better bound by taking advantage of this strong assumption. Estimates for the tail of clusters are the starting point for many probabilistic results for Hawkes processes such as mixing properties, see [ 3 ] , or regenerative properties, see [ 5 ] . The main originality of this work is that it goes beyond the homogeneous case: the results for Poissonian clusters are stated for inhomogeneous clusters, meaning that the interactions functions may depend on both s\in\mathds{R} and t\in\mathds{R} (non convolutive framework) and not just on the delay t-s (convolutive framework) between the current time t and the time s of the previous individual. Similar cases were considered by [ 14 ] where the authors also studied properties of the Laplace transform. The non convolutive framework allows more flexibility and improves the possibilities for applications in neurosciences, finance etc. The present paper is structured as follows. In Section 2 we introduce the objects and notations used in this paper. In Section 3 we present sharp bounds on the exponential moments of the number of points of given types in a Poissonian Galton Watson process. Then we show how exponential moments for Poissonian Galton Watson processes applies to Hawkes processes. Finally we also present results about tail estimates in terms of generations of the individuals for Poissonian Galton Watson processes. In Section 4 we state tail estimates for Poissonian cluster processes. In Section 5 we perform the exact study of the topological properties of the Laplace transform of Galton Watson processes. Technical results on bivariate convolution are needed in Section 4 , they are presented and proved in Appendix A . Finally, the proofs can be found in Section 6 . 2. Settings Let \mathds{N} the set of non negative integers and \mathds{N}^{*}=\mathds{N}\setminus\{0\} . Let \mathds{R} the set of real numbers and \mathds{R}_{+}=[0,\infty) the set of non negative real numbers. Consider M\in\mathds{N}^{*} and denote by \bm{M} the set [\![1,M]\!] . When, in the sequel, Poissonian Galton Watson processes are considered, in Sections 3

Read the complete text