Persistence in perturbed contact models in continuum

This paper demonstrates that perturbed contact models on locally compact metric spaces, even when the critical balance between birth and death rates is violated, typically avoid extinction and admit a family of invariant measures describable via the Feynman-Kac formula.

The Gist

Imagine a vast, bustling city where people (particles) are constantly being born and dying. In this city, the rules of life are simple:

• Birth: If you have neighbors, you are more likely to have a child nearby.
• Death: People die at a certain rate, which can vary from neighborhood to neighborhood.

For a long time, scientists studying this city believed that for the population to survive forever without exploding or dying out, the "birth rate" and "death rate" had to be in a perfect, delicate balance. They called this the "Critical Regime." It's like a tightrope walker; if the wind (death rate) gets even a little bit stronger in one spot, the walker falls, and the whole city collapses into extinction.

The Big Question
The authors of this paper asked: What if the balance isn't perfect? What if there are local "disasters"—areas where the death rate is suddenly much higher than usual? Does the whole city die out, or can it survive?

The Discovery: Resilience, Not Fragility
The paper says: The city survives.

Even if there are "local disasters" (areas with high mortality), the population doesn't vanish. Instead, the population just adjusts. It's like a river flowing around a large rock. The water (the population) gets a little turbulent and changes shape around the rock, but the river keeps flowing. The "disaster" doesn't stop the flow; it just perturbs it.

How They Proved It (The Metaphors)

1. The "Shadow" of the Disaster (The Feynman-Kac Formula):
   To understand how the population behaves, the authors used a mathematical tool called the Feynman-Kac formula. Think of this as a "time-lapse camera" that tracks every possible path a person could take through the city over time.

   • In a normal city, a person's path is just a random walk.
   • In this "disaster" city, the camera adds a "shadow" to the path. If a person walks through a high-death zone, their "shadow" gets dimmer (representing the risk of dying).
   • The authors showed that even with these shadows, you can still calculate a stable, long-term average of where people will be. The "shadow" doesn't make the person disappear; it just changes the probability of them being in certain spots.

2. The Chain Reaction (Hierarchical Equations):
   The city is complex. To understand the whole population, you can't just look at one person; you have to look at pairs, groups of three, groups of four, and so on.

   • The authors built a "chain" of equations. They solved the problem for one person first (using the time-lapse camera).
   • Then, they used that solution to solve for pairs, then groups of three, and so on, step-by-step (induction).
   • They proved that this chain doesn't break, even with the high-death zones. The math holds together, meaning a stable population distribution exists.

3. The "Heavy Tail" vs. "Light Tail" (Why it works):
   The paper mentions that in some small cities (low dimensions), the population only survives if the "dispersal kernel" (how far people move to have children) has "heavy tails."

   • Light Tail: People only have children very close to home. If a disaster hits a neighborhood, everyone there dies, and no one from far away can replace them.
   • Heavy Tail: People can have children far away. If a disaster hits one spot, people from distant, safe spots can move in and repopulate the area.
   • The authors show that even with local high-death rates, as long as the "heavy tail" rule is met (or the dimension is high enough), the population finds a new, stable equilibrium.

The Bottom Line
The paper proves that local catastrophes do not necessarily lead to total extinction.

In the world of these mathematical models, a population is much tougher than previously thought. You don't need a perfect, global balance between birth and death to have a stable society. You can have "rough patches" where death is high, and the system will simply reorganize itself into a new, stable state. The "invariant measure" (the stable state) still exists; it's just a slightly different version of the original, adapted to the local dangers.

In short: The system is robust. A local disaster is a bump in the road, not a cliff edge.

Technical Summary

1. Problem Statement

The paper addresses a fundamental question in the theory of interacting particle systems: Can local disasters (excess mortality) lead to the extinction of a population modeled by a contact process in a continuous space?

• Context: Previous research (e.g., [18, 19]) established the existence of invariant measures (stationary states) for contact processes on locally compact metric spaces, but only under a critical regime condition. This condition requires a precise stochastic balance between birth and death rates.
• The Gap: In many biological or physical scenarios, this balance is violated locally due to perturbations (e.g., a localized area with higher death rates). The authors investigate whether the system remains persistent (i.e., retains invariant measures) when the critical balance is broken by a non-negative local perturbation in the death rate.
• Specific Challenge: The evolution of correlation functions in this perturbed regime is governed by a nonlocal convolution operator perturbed by a negative potential. The authors must prove that this perturbation does not destroy the existence of a family of invariant measures.

2. Mathematical Framework and Model

The authors consider a continuous-time contact process on a locally compact separable metric space $X$ (e.g., $\mathbb{R}^d$).

• State Space: Configurations $\gamma \in \Gamma(X)$, representing finite or countably infinite sets of particles.

• Generator: The process is defined by a generator $L$ acting on functions $F: \Gamma \to \mathbb{R}$:
  $$(LF)(\gamma) = \sum_{x \in \gamma} U(x) (F(\gamma \setminus x) - F(\gamma)) + \int_X \sum_{x \in \gamma} a(y, x) (F(\gamma \cup y) - F(\gamma)) m(dy)$$
  where:

  • $U(x) = V(x) + W(x)$ is the death rate.
  • $V(x)$ is the base death rate satisfying the critical regime condition.
  • $W(x) \geq 0$ is the local perturbation (excess mortality).
  • $a(y, x)$ is the dispersal kernel (birth rate density).
  • $m$ is a locally finite Borel measure.

• Key Assumptions:

  1. Critical Regime for $V$: There exists a strictly positive function $\Psi(x)$ such that the birth and death rates balance for the unperturbed system: $\int a(x, y)\Psi(y)m(dy) = V(x)\Psi(x)$.
  2. Transience Condition: The associated Markov jump process (with generator $L_f$ derived from the critical rates) is transient. Specifically, two independent trajectories starting at different points drift apart, ensuring the integral of their interaction decays sufficiently fast.
  3. Integrability of Perturbation: The perturbation $W(x)$ satisfies an integrability condition relative to the transient process: $\sup_x \mathbb{E}_x [\int_0^\infty W(X(t)) dt] < \infty$.

3. Methodology

The authors employ a hierarchical approach based on correlation functions and semigroup theory, adapting techniques from Schrödinger operator theory.

• Correlation Functions: Instead of solving for the measure directly, they analyze the system of $n$-point correlation functions $k^{(n)}_t$. The time evolution is governed by a recursive chain of equations:
  $$\frac{\partial k^{(n)}_t}{\partial t} = S_n k^{(n)}_t + f^{(n)}_t$$
  where $S_n = \hat{L}^*_n - W_n$. Here, $\hat{L}^*_n$ is the adjoint of the unperturbed generator, and $W_n$ is a multiplication operator by the sum of perturbations.
• Operator Analogy: The operator $S_n$ is analogous to an $n$-particle Schrödinger operator with a potential $-W_n$. The problem reduces to finding the stationary solution ($S_n k^{(n)} + f^{(n)} = 0$).
• Feynman-Kac Formula: To solve the equation for the first correlation function ($n=1$), the authors utilize the Feynman-Kac formula. The stationary solution is expressed as the limit of the evolution of a constant initial state under the perturbed semigroup:
  $$k^{(1)}_\rho(x) = \rho \mathbb{E}_x \left[ \exp\left( -\int_0^\infty W(X(s)) ds \right) \right]$$
  This formula explicitly links the stationary density to the "survival probability" of the perturbation along the trajectory of the unperturbed process.
• Inductive Construction: Once the first correlation function is established, the authors use an inductive procedure to construct solutions for higher-order correlation functions ($n \geq 2$). They prove that the series of solutions converges and satisfies specific growth bounds.

4. Key Contributions and Results

Theorem 3.1 (Main Result):
Under the stated assumptions (measurability, regularity, critical regime for $V$, transience, and $W$-integrability), the following hold:

1. Existence of Invariant Measures: There exists a one-parameter family of invariant probability measures $\{\mu_\rho\}_{\rho > 0}$ for the perturbed contact process.
2. Persistence: The local violation of criticality (due to $W(x) > 0$) does not lead to extinction. The system persists, and the stationary state is merely perturbed, not destroyed.
3. Explicit Bounds: The correlation functions $k^{(n)}_\rho$ satisfy the bound:
   $$k^{(n)}_\rho(x_1, \dots, x_n) \leq D H^n (n!)^2$$
   where $H$ is the constant from the transience condition and $D$ depends on the parameter $\rho$.
4. Convergence: If the system starts from a Poisson measure with intensity $\rho$, the time-evolved correlation functions converge to the stationary correlation functions as $t \to \infty$.

Technical Novelty:

• The paper extends the existence of invariant measures beyond the critical regime.
• It demonstrates that the "criticality" condition is robust against local positive potentials (excess mortality), drawing a parallel to spectral theory where local positive potentials do not change the essential spectrum of the Laplacian.
• It provides a constructive proof using the Feynman-Kac formula for the first correlation function and an inductive scheme for the hierarchy.

5. Significance

• Biological Modeling: The results are crucial for modeling population dynamics and epidemic spread in heterogeneous environments. It proves that a population can survive and reach a stationary state even if specific regions have higher mortality rates, provided the dispersal kernel allows for sufficient mixing (transience) and the perturbation is not too "heavy" (integrability condition).
• Mathematical Physics: The work bridges the gap between contact processes and Schrödinger operator theory. It shows how tools from quantum mechanics (Feynman-Kac, spectral properties) can be adapted to stochastic interacting particle systems in continuous space.
• Robustness of Stationarity: It challenges the notion that a strict balance between birth and death is necessary for stationarity, showing instead that a "perturbed balance" is sufficient for the existence of invariant measures in high dimensions ($d \geq 3$) or with heavy-tailed dispersal kernels.

In summary, the paper rigorously proves that local disasters do not necessarily cause global extinction in continuum contact models, provided the underlying unperturbed system is transient and the perturbation is integrable with respect to the process's trajectory.