Decompositions of CSBPs via Poissonian Intertwining

Clément Foucart, Olivier Hénard · 2025

Read Decompositions of CSBPs via Poissonian Intertwining by Clément Foucart, Olivier Hénard (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-2503-12103.txt.

Introduction

Clément Foucart and Olivier Hénard reframe how branching populations split into a core and a skeleton, in a paper from 2025. Continuous-state branching processes model populations that grow and die in bulk. Their skeletal splits expose the family lines that survive to a later time.

The tool is intertwining: a linking kernel that lets one process shadow another. Each population gets a partner pair, one continuous part with the same law and one counting part whose count, given the first, follows a Poisson rule. A threshold on the linking dial splits two regimes. Past it, the classic skeleton results return. Below it, in growing populations, a birth term appears as steady immigration.

Catalog record

CREATORClément Foucart, Olivier Hénard
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE101 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-12103

Browse the full library · About the Tower of Babel

Full text

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

Decompositions of CSBPs via Poissonian Intertwining Clément Foucart, Olivier Hénard Abstract We revisit certain decompositions of continuous-state branching processes (CSBPs), commonly referred to as skeletal decompositions, through the lens of intertwining of semi-groups. Precisely, we associate to a CSBP X X with branching mechanism ψ \psi a family of ℝ + × ℤ + \mathbb{R}_{+}\times\mathbb{Z}_{+} -valued branching processes ( X λ , L λ ) (X^{\lambda},L^{\lambda}) , indexed by a parameter λ ∈ ( 0 , ∞ ) \lambda\in(0,\infty) , that satisfies an intertwining relationship with X X through the Poisson kernel with parameter λ ​ x \lambda x . The continuous component X λ X^{\lambda} has the same law as X X , while the discrete component L λ L^{\lambda} , conditionally on X t λ X^{\lambda}_{t} , has a Poisson distribution with parameter λ ​ X t λ \lambda X^{\lambda}_{t} . The law of ( X λ , L λ ) (X^{\lambda},L^{\lambda}) depends on the position of λ \lambda within [ 0 , ∞ ) = [ 0 , ρ ) ∪ [ ρ , ∞ ) [0,\infty)=[0,\rho)\cup[\rho,\infty) , where ρ \rho is the largest positive root of ψ \psi . When λ ≥ ρ \lambda\geq\rho , various well-known results concerning skeleton decompositions are recovered. In the supercritical case ( ρ > 0 \rho>0 ), when λ < ρ \lambda<\rho , a novel phenomenon arises: a birth term appears in the skeleton, corresponding to a one-unit proportional immigration from the continuous to the discrete component. Along the way, the class of continuous-time branching processes taking values in ℝ + × ℤ + \mathbb{R}_{+}\times\mathbb{Z}_{+} is constructed. Keywords. Continuous-state branching process, Intertwining, Esscher transform, Two-type branching process, Explosion 1 Introduction This paper investigates how Continuous-State Branching Processes (CSBPs) can be decomposed into \mathbb{R}_{+}\times\mathbb{Z}_{+} -valued branching processes using intertwining relationships between Markovian semigroups. An intertwining relationship is a property of commutation of semigroups, or generators, through some kernel. Formally, two semigroups P_{t} and Q_{t} are said to be intertwined with respect to a kernel \Lambda if they satisfy the following relationship \Lambda P_{t}=Q_{t}\Lambda,\hskip 9.24994pt\forall t\geq 0. (1) Intertwining is intimately connected to the problem of determining whether a function of a Markov process remains Markovian, as explored by Pitman and Rogers [ PR81 ] . It also arises in Markovian filtering, see e.g. Kurtz and Ocone [ KO88 ] , Kurtz [ Kur98 ] , and Kurtz and Nappo [ KN11 ] , where Markov mapping theorems are established in the setting of martingale problems. Numerous examples of intertwined Markov semigroups have been discovered, including those presented by Carmona et al. [ CPY98 ] and Pal and Shkolnikov [ PS13 ] . The intertwining theory has been further developed in various directions, becoming a fundamental tool in stochastic processes. They play for instance an important role in the study of strong stationary times, see for instance Diaconis and Fill [ DF90 ] , Miclo [ Mic20 ] and Arnaudon et al. [ ACPM24 ] . Intertwining likewise appears in analyzing certain interacting particle systems, see e.g. Floreani et al. [ FJRW24 ] . It also plays a key role in the so-called lookdown construction of the genealogy of branching populations, see for instance Kurtz [ Kur18 ] and Etheridge and Kurtz [ EK19 ] for recent works in this direction. The concept of decomposition of branching processes into “skeletons” dates back to Harris [ Har48 ] , who showed that any supercritical Bienaymé-Galton-Watson process with positive probability of extinction can be embedded in a two-type branching process, see also Athreya-Ney’s book [ AN04 , Chapter I-Section 12] . In this framework, one type represents individuals with infinite line of descent (called prolific or immortal individuals), while the other corresponds to mortal individuals. This decomposition has become a key tool in branching process theory. In the context of Continuous State Branching Processes (CSBPs), the prolific individuals form a discrete branching process, while the non-prolific ones form a continuum, as explored by Bertoin et al. [ BFM08 ] , see also Lambert and Uribe Bravo, [ LUB24 ] , for a recent work in this direction in the context of splitting trees. The framework with spatial motion has also garnered considerable attention, we refer for instance to Eckhoff et al. [ EKW15 ] , Fekete et al. [ FFK20 ] and the references therein. In a seminal paper, Duquesne and Winkel [ DW07 ] introduced a nested family of discrete trees with edge lengths that is consistent under Poissonian sampling of the leaves; they proved that any such family embeds a Lévy real tree encoding the genealogy of a CSBP. In a similar spirit, Abraham and Delmas [ AD12 ] have constructed a tree-valued Markov process, where evolving skeletons serve as the hprimary objects. See also Abraham and Delmas [ AD09 ] and Abraham et al. [ ADV10 ] for closely related studies. More recently, these decompositions have been revisited using Poissonization techniques applied to the stochastic differential equation with jumps solved by a CSBP, as discussed by Fekete et al. [ FFK19 ] . In all these works, Esscher transforms of the branching mechanism play a central role. To summarize, given an eventually positive branching mechanism \psi , and denoting by \rho its largest positive root, a skeleton decomposition consists in “grafting” subcritical continuous masses — evolving with a branching mechanism given by an Esscher transform of \psi at the right of \rho — onto a discrete branching process, so that the total continuous mass is a CSBP governed by \psi . Although not always highlighted in the works previously cited – see however [ BFM08 , Page 722] and [ LUB24 , Page 1283] – a common feature in such skeletal decompositions lies on the fact that the joint process encoding both the discrete component and the continuous mass satisfies the branching property. Two-type branching processes taking values in \mathbb{R}_{+}\times\mathbb{Z}_{+} come therefore naturally into play. To our knowledge, no general treatment of this class of processes has been presented before. We summarize their fundamental properties in Section 2 , see Theorem 2.1 and Proposition 2.3 , and recall well-known facts about one-dimensional branching processes. The \mathbb{R}_{+}\times\mathbb{Z}_{+} -valued branching processes are constructed in Section 4 by using classical results from the theory of Markov processes. Our main results are presented in Section 3 . We start by establishing an intertwining relationship, through the Poisson kernel with parameter \lambda x>0 , between the generator of a CSBP (which may be immortal, namely its branching mechanism \psi can be negative) and that of a specific \mathbb{R}_{+}\times\mathbb{Z}_{+} -valued branching process (X^{\lambda},L^{\lambda}) , see Theorem 3.1 . We show then that a relationship of the form ( 1 ) holds between P_{t} the semigroup of (X^{\lambda},L^{\lambda}) and Q_{t} the semigroup of the CSBP (\psi) , see Theorem 3.4 . The process (L_{t}^{\lambda},t\geq 0) is the so-called skeleton . The decomposition of the CSBP follows from an application of Pitman-Rogers theorem, see Theorem A , and states the following: if X^{\lambda} starts from x>0 and the initial distribution of the skeleton L^{\lambda} has a Poisson law with parameter \lambda x , the first coordinate projection X^{\lambda} is a CSBP (\psi) started from x and for any t\geq

Read the complete text