arXiv subjects

Pei-Sen Li

Publications and source records attributed to Pei-Sen Li.

16 recordsLinked to original sources

arXiv2026

A Pathwise Approach to the Strong Feller Property and Irreducibility of Nonlinear Branching Processes

We study the strong Feller property and irreducibility for continuous-state nonlinear branching processes defined as solutions to stochastic differential equations with jumps. Due to boundary degeneracy and discontinuous jump coefficients, classical methods do not apply. We develop a pathwise approach combining state-dependent time change, truncated auxiliary processes, and localized coupling to establish these two properties. As applications, we obtain exponential convergence to a unique quasi-stationary distribution in the absorbing case, and uniform exponential ergodicity in the non-absorbing case. This pathwise approach is flexible and can be adapted to a broader class of jump-diffusions without relying on specific coefficient structures.

math.PR↗

arXiv2026

Yaglom limits of continuous-state branching processes in Brownian random environment

In this paper, we investigate the asymptotic behavior of continuous-state branching processes in a Brownian random environment (CBBRE) conditioned on non-extinction. For the subcritical case, we prove the existence of the Yaglom limit and derive an explicit representation of its Laplace transform using Kummer confluent hypergeometric functions. Notably, we demonstrate that the Yaglom limit is strictly independent of the initial state of the process across all three subcritical regimes: weakly, intermediately, and strongly subcritical.

math.PR↗

arXiv2026

Matrix-Decoupled Concentration for Autoregressive Sequences: Dimension-Free Guarantees for Sparse Long-Context Rewards

Sequence-level evaluations in autoregressive Large Language Models (LLMs) rely on highly dependent token generation. Establishing tight concentration bounds for these processes remains a challenge due to two fundamental bottlenecks in existing frameworks: (i) classical inequalities typically separate dependency structures from target sensitivities, leading to a scalar collapse that inflates the variance proxy to a suboptimal $\mathcal{O}(N)$ for sparse terminal rewards; (ii) conversely, while certain spatial methods achieve tighter bounds, they lack the strictly causal filtration required by sequential generation, rendering them inapplicable to the autoregressive setting. To resolve both bottlenecks, we establish a sharp McDiarmid-type inequality for dependent sequences, governed strictly by the exact matrix-vector multiplication of the causal dependency resolvent and the target sensitivity vector. This Matrix-Decoupled Concentration (MDC) framework natively recovers optimal constants for Markov chains and exploits directed $d$-separation to yield order-optimal bounds for causal trees. Crucially, by exactly preserving the coordinate-wise sparsity of rewards within a strictly causal framework, MDC mathematically prevents scalar collapse, guaranteeing a dimension-free $\mathcal{O}(1)$ variance proxy and providing a rigorous mathematical justification for the stability of long-context reasoning.

cs.LG↗

arXiv2026

A localized coupling approach to interacting continuous-state branching processes

We introduce a class of continuous-state branching processes with immigration, predation and competition, which can be viewed as a combination of the classical Lotka-Volterra model and continuous-state branching processes with competition that were introduced by Berestycki, Fittipaldi, and Fontbona (Probab. Theory Relat. Fields, 2018). This model can be constructed as a unique strong solution to a class of two-dimensional stochastic differential equations with jumps. We establish sharp conditions for the uniform ergodicity in the total variation of this model. Our proof relies on a novel, localized Markovian coupling approach, which is of its own interest in the ergodicity theory of Markov processes with interactions.

math.PR↗

arXiv2025

Stochastic integral representations for the Ray-Knight theorem of the Levy forest

We present a simple stochastic integral representation for the local times of the height process of a spectrally positive Levy process stopped at a hitting time. From the representation we derive a strong stochastic equation for the local time process of the type of Bertoin and Le Gall (Illinois J. Math., 2006) and Dawson and Li (Ann. Probab., 2012). This leads to a representation of the Ray-Knight theorem of Le Gall and Le Jan (Ann. Probab., 1998) and Duquesne and Le Gall (Asterisque, 2002), which codes the genealogical forest of a continuous-state branching process. The results extend those in the recent work of Aidekon et al. (Sci. China Math., 2024) for a Brownian motion with a local time drift.

math.PR↗

arXiv2024

Uniqueness Problem for the Backward Differential Equation of a Continuous-State Branching Process

The distributional properties of a multi-dimensional continuous-state branching process are determined by its cumulant semigroup, which is defined by the backward differential equation. We provide a proof of the assertion of Rhyzhov and Skorokhod (Theory Probab. Appl., 1970) on the uniqueness of the solutions to the equation, which is based on a characterization of the process as the pathwise unique solution to a system of stochastic equations.

math.PR↗

arXiv2023

Exponential ergodicity of branching processes with immigration and competition

We study the ergodic property of a continuous-state branching process with immigration and competition. The exponential ergodicity in a weighted total variation distance is proved under natural assumptions. The main theorem applies to subcritical, critical and supercritical branching mechanisms, including all those of stable types. The proof is based on the construction of a Markov coupling process and the choice of a nonsymmetric control function for the distance. Those are designed to identify and to take the advantage of the dominating factor from the branching, immigration and competition mechanisms in different parts of the state space. The approach provides a way of finding a lower bound of the ergodicity rate.

math.PR↗

arXiv2023

Quasi-stationary distribution for continuous-state branching processes with competition

We study quasi-stationary distribution of the continuous-state branching process with competition introduced in Berestycki, Fittipaldi and Fontbona\ (Probab. Theory Relat. Fields, 2018). This process is constructed as the unique strong solution to a stochastic integral equation with jumps. An important example is the logistic branching process constructed in Lambert (Ann. Appl. Probab., 2005). We establish the strong Feller property,trajectory Feller property, Lyapunov condition, weak Feller property and irreducibility, respectively. These properties together allow us to prove that when the competition term is strong enough near $+\infty$, then there is a unique quasi-stationary distribution, which attracts all initial distributions with exponential rates.

math.PR↗

arXiv2020

Time-changed spectrally positive Lévy processes starting from infinity

Consider a spectrally positive Lévy process $Z$ with log-Laplace exponent $Ψ$ and a positive continuous function $R$ on $(0,\infty)$. We investigate the entrance from $\infty$ of the process $X$ obtained by changing time in $Z$ with the inverse of the additive functional $η(t)=\int_{0}^{t}\frac{{\rm d} s}{R(Z_s)}$. We provide a necessary and sufficient condition for infinity to be an entrance boundary of the process $X$. Under this condition, the process can start from infinity and we study its speed of coming down from infinity. When the Lévy process has a negative drift $δ:=-γ<0$, sufficient conditions over $R$ and $Ψ$ are found for the process to come down from infinity along the deterministic function $(x_t,t\geq 0)$ solution to ${\rm d} x_t=-γR(x_t) {\rm d} t$, with $x_0=\infty$. When $Ψ(λ)\sim cλ^α$, with $λ\rightarrow 0$, $α\in (1,2]$, $c>0$ and $R$ is regularly varying at $\infty$ with index $θ>α$, the process comes down from infinity and we find a renormalisation in law of its running infimum at small times.

math.PR↗

arXiv2020

On the entrance at infinity of Feller processes with no negative jumps

Consider a non-explosive positive Feller process with no negative jumps. It is shown in this note that when infinity is an entrance boundary, in the sense that the entrance times of the process remain bounded when the initial value tends to infinity, the process admits a Feller extension on the compactified state space $[0,\infty]$. Moreover, when started from infinity, the extended Markov process on $[0,\infty]$ leaves infinity instantaneously and stays finite, almost-surely. Arguments are adapted from a proof given by O. Kallenberg for diffusions. We also show that the process started from $x$ converges weakly towards that started from infinity in the Skorokhod space, when $x$ goes to infinity.

math.PR↗

arXiv2020

Integral functionals for spectrally positive Levy processes

We find necessary and sufficient conditions for almost sure finiteness of integral functionals of spectrally positive Lévy processes. Via Lamperti type transforms, these results can be applied to obtain new integral tests on extinction and explosion behaviors for a class of continuous-state nonlinear branching processes.

math.PR↗

arXiv2019

Exponential ergodicity for general continuous-state nonlinear branching processes

By using the coupling technique, we present sufficient conditions for the exponential ergodicity of general continuous-state nonlinear branching processes in both the $L^1$-Wasserstein distance and the total variation norm, where the drift term is dissipative only for large distance, and either diffusion noise or jump noise is allowed to be vanished. Sufficient conditions for the corresponding strong ergodicity are also established.

math.PR↗

arXiv2018

A general continuous-state nonlinear branching process

In this paper we consider the unique nonnegative solution to the following generalized version of the stochastic differential equation for a continuous-state branching process. \beqnn X_t \ar=\ar x+\int_0^tγ_0(X_s)\dd s+\int_0^t\int_0^{γ_1(X_{s-})} W(\dd s,\dd u)\cr \ar\ar\qquad+\int_0^t\int_{0}^\infty\int_0^{γ_2(X_{s-})} z\tilde{N}(\dd s, \dd z, \dd u), \eeqnn where $W(\dd t,\dd u) $ and $\tilde{N}(\dd s, \dd z, \dd u)$ denote a Gaussian white noise and an independent compensated spectrally positive Poisson random measure, respectively, and $γ_0,γ_1$ and $γ_2$ are functions on $\mbb{R}_+$ with both $γ_1$ and $γ_2$ taking nonnegative values. Intuitively, this process can be identified as a continuous-state branching process with population-size-dependent branching rates and with competition. Using martingale techniques we find rather sharp conditions on extinction, explosion and coming down from infinity behaviors of the process. Some Foster-Lyapunov type criteria are also developed for such a process. More explicit results are obtained when $γ_i, i=0, 1, 2$ are power functions.

math.PR↗

arXiv2018

A continuous-state polynomial branching process

A continuous-state polynomial branching process is constructed as the pathwise unique solution of a stochastic integral equation with absorbing boundary condition. The extinction and explosion probabilities and the mean extinction and explosion times are computed explicitly, which are also new in the classical branching case. We present necessary and sufficient conditions for the process to extinguish or explode in finite times. In the critical or subcritical case, we give a construction of the process coming down from infinity. Finally, it is shown that the continuous-state polynomial branching process arises naturally as the rescaled limit of a sequence of discrete-state processes.

math.PR↗

arXiv2017

Nonlinear branching processes with immigration

The nonlinear branching process with immigration is constructed as the pathwise unique solution of a stochastic integral equation driven by Poisson ran- dom measures. Some criteria for the regularity, recurrence, ergodicity and strong ergodicity of the process are then established.

math.PR↗

arXiv2016

Perturbations of continuous-time Markov chains

The equivalence of regularity of a Q-matrix with its bounded perturbations is proved and a integration by parts formula is established for the associated Feller minimal transition functions.

math.PR↗