Multi-type branching processes with immigration generated by point processes

Basic Information

• Paper ID: 2411.12474
• Title: Multi-type branching processes with immigration generated by point processes
• Authors: Martin Minchev, Maroussia Slavtchova-Bojkova
• Classification: math.PR (Probability Theory)
• Publication Date: November 2024 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2411.12474

Abstract

Building upon Sevastyanov's pioneering work on homogeneous Poisson immigration branching processes, this paper explores the application of general point process frameworks beyond Poisson processes in modeling immigration in branching processes. The authors demonstrate how to generalize the technical methods from the Poisson immigration case to the immigration case driven by determinantal point processes (DPP), providing new theoretical frameworks and analytical tools for branching process theory.

Research Background and Motivation

Problem Definition

1. Core Problem: Traditional branching process immigration models are primarily based on Poisson processes, which have equidispersion assumptions and time-homogeneity constraints that limit practical applications.
2. Significance: Branching processes have broad applications in biology, epidemiology, finance, and other fields. More flexible immigration mechanisms can better describe real-world phenomena.
3. Existing Limitations:
   • The equidispersion assumption of Poisson processes (variance equals mean) often does not hold in practice
   • Time-homogeneity constraints limit model applicability
   • Lack of a unified theoretical framework for handling general point process immigration

Research Motivation

• Establish a branching process immigration theory framework based on general point processes
• Provide mathematical tools that transcend Poisson process limitations
• Offer more flexible modeling choices for practical applications

Core Contributions

1. Theoretical Framework: Establishes a unified theoretical framework based on point process Laplace functionals for analyzing branching process immigration driven by general point processes.
2. Generating Function Derivation: Provides explicit expressions for probability generating functions in cases of DPP, Cox processes, and fractional Poisson process immigration.
3. Asymptotic Behavior Analysis: Proves limit theorems for multi-type branching processes with DPP immigration in subcritical, critical, and supercritical cases.
4. Moment Calculations: Derives exact expressions for the mean and covariance of processes under DPP immigration.
5. Error Correction: Corrects technical errors in references 17, Theorem 8 and 36, Theorem 3.

Methodology Details

Problem Formulation

Study the asymptotic properties of multi-type continuous-time Markov branching processes $Z(t) = (Z_1(t), \ldots, Z_d(t))$ driven by immigration times generated by a general point process $\Phi$, where:

• $Z_i(t)$ represents the number of particles of type $i$ at time $t$
• Immigration occurs at jump times $T_1 < T_2 < \cdots$ of the point process $\Phi$
• Each immigration brings an independent and identically distributed particle population $I^{(k)} \sim I$

Theoretical Framework

Point Process Foundations

Define a point process $\Phi = \sum_{i \leq \kappa} \delta_{X_i}$ with key characteristics:

• Intensity Measure: $\Lambda(B) = E[\Phi(B)]$
• Laplace Functional: $L_\Phi(f) = E[e^{-\int f d\Phi}] = E[e^{-\sum_{i \leq \kappa} f(X_i)}]$
• Joint Intensity: $\rho_k(x_1, \ldots, x_k)$

Generating Function Representation

The generating function of the branching process with immigration can be expressed as: $G_{Z(t)}(s) = L_\Phi(f_t), \quad f_t(x) = -\ln(G_{Z_\times(t-x)}(s))1_{\{x \leq t\}}$

where $Z_\times$ is the underlying branching process without immigration.

Special Point Process Cases

Determinantal Point Process (DPP)

For a $(\Lambda, K)$-DPP, the joint intensity is: $\rho_n(x_1, \ldots, x_n) = \det(K(x_i, x_j))_{1 \leq i,j \leq n}$

The Laplace functional is: $L_\Phi(f) = 1 + \sum_{n \geq 1} \frac{(-1)^n}{n!} \int_{\mathbb{R}_+^n} D(x) \prod_{i=1}^n (1-e^{-f(x_i)}) \Lambda(dx)$

Cox Process

For a Cox process with directing measure $\eta$: $L_\Phi(f) = E\left[\exp\left(-\int_{\mathbb{R}_+} (1-e^{-f(x)}) \eta(dx)\right)\right]$

Fractional Poisson Process

For a fractional Poisson process with parameters $(\beta, \lambda)$: $L_\Phi(f) = E\left[\exp\left(-\lambda \int_{\mathbb{R}_+} (1-e^{-f(t)}) dY_\beta(t)\right)\right]$

where $Y_\beta$ is the inverse of a $\beta$-stable subordinator.

Main Theoretical Results

Moment Properties (Theorem 5.1)

For $(\Lambda, K)$-DPP immigration:

1. Mean: $E[Z_i(t)] = \int_{(0,t]} K(x,x) E[Z_{\times,i}(t-x)] \Lambda(dx)$
2. Covariance: $\text{Cov}(Z_i(t), Z_j(t)) = \int_{(0,t]} K(x,x) E[Z_{\times,i}(t-x)Z_{\times,j}(t-x)] \Lambda(dx)$$- \int_{(0,t]^2} K^2(x,y) E[Z_{\times,i}(t-x)]E[Z_{\times,j}(t-y)] \Lambda(dx)\Lambda(dy)$

Asymptotic Behavior (Theorem 5.2)

Let $\rho$ be the Perron-Frobenius root of the generating matrix $A$. If $\int_{(0,\infty)} e^{-\rho x} K(x,x) \Lambda(dx) < \infty$, then there exists an $\mathbb{R}^d$-valued random variable $W$ such that: $\frac{Z(t)}{e^{\rho t}} \xrightarrow{d} W \quad (t \to \infty)$

Refined Asymptotic Results (Theorem 5.3)

For stationary DPP with $\lambda(x) \sim \lambda_\infty e^{\delta x}$:

1. Supercritical Case ($\delta > \max\{\rho, 0\}$): $\frac{Z_i(t)}{e^{\delta t}} \xrightarrow{L^2} A_i = K^* \lambda_\infty \int_0^\infty e^{-\delta x} E[Z_{\times,i}(x)]dx$
2. Critical Growth ($\rho > 0, \delta = \rho$): $\frac{Z_i(t)}{te^{\delta t}} \xrightarrow{L^2} A_i' = K^* \lambda_\infty \langle u, E[I] \rangle v_i$
3. Critical Case ($\rho = 0, \delta = 0$): $\frac{Z(t)}{t} \xrightarrow{d} Y v$ where $Y \sim \Gamma(K(0,0)\lambda_\infty \beta, 1/Q)$
4. Subcritical Case ($\rho < 0, \delta = 0$): $Z(t) \xrightarrow{d} X$ where the Laplace transform of $X$ is given by equation (6.17).

Technical Innovations

Unified Framework

• Establishes a unified approach for handling general point process immigration through Laplace functionals
• Avoids the complexity of traditional methods requiring explicit computation of point count distributions within intervals

Technical Generalization

• Cleverly combines dominated convergence theorem and martingale theory to handle the complex correlation structure of DPP
• Utilizes Hadamard inequality to control determinant growth, ensuring integral controllability

Error Correction

• Identifies and corrects technical errors in literature regarding eigenvector normalization
• Clarifies discrepancies in eigenvector definitions across different references

Related Work

This paper builds upon the following important works:

1. Sevastyanov (1957): Foundational work on homogeneous Poisson immigration branching processes
2. Mitov et al. (2018): In-depth study of non-homogeneous Poisson immigration
3. Slavtchova-Bojkova et al. (2022-2023): Extensions to multi-type cases
4. Rabehasaina & Woo (2021): Application of characteristic function methods

Compared to existing work, the main advantages of this paper are:

• Provides a general theoretical framework transcending Poisson processes
• Offers more concise and unified derivation methods
• Corrects technical errors in the literature

Conclusions and Discussion

Main Conclusions

1. Establishes a branching process theory framework based on general point process immigration
2. Proves complete asymptotic theory for DPP immigration cases
3. Provides practical computational tools and methods

Limitations

1. Technical Conditions: Some results require strong moment conditions and integrability assumptions
2. Computational Complexity: The determinantal structure of DPP becomes computationally complex in high dimensions
3. Practical Applications: Lacks concrete application examples and numerical experiments

Future Directions

1. Study more general reducible cases
2. Develop numerical computation methods and algorithms
3. Explore specific applications in biology and epidemiology

In-Depth Evaluation

Strengths

1. Theoretical Depth: Provides rigorous and complete mathematical theory, filling an important gap in the field
2. Methodological Innovation: Application of Laplace functional methods simplifies traditionally complex derivation processes
3. Technical Rigor: Detailed proofs with appropriate technical handling
4. Practical Value: Provides more flexible tools for practical modeling

Weaknesses

1. Application-Oriented: Emphasizes theoretical derivation while lacking demonstration of practical applications
2. Numerical Aspects: No computational methods or numerical experiments provided
3. Readability: High technical threshold for non-specialist readers

Impact

1. Academic Contribution: Provides important theoretical extension to branching process theory
2. Methodological Value: Laplace functional methods may inspire related research
3. Application Potential: Provides new theoretical foundation for practical problem modeling

Applicable Scenarios

• Population dynamics modeling requiring consideration of correlation between immigration events
• Epidemic transmission models with repulsive or clustering characteristics
• Jump-diffusion process modeling in financial markets

References

The paper cites 50 important references, primarily including:

• Athreya & Ney (1972): Classical textbook on branching processes
• Sevastyanov (1957, 1971): Foundational work on immigration branching processes
• Baccelli et al. (2024): Point process theory reference
• Mitov et al. (2018): Recent important related work

Overall Assessment: This is a high-quality theoretical mathematics paper that makes important contributions to branching process theory. While emphasizing theory over applications, the established theoretical framework possesses significant academic value and application potential.