The Critical Patch Size Problem in Random Graphs

Abstract: The problem of {\it critical patch size} -- a threshold condition for population persistence -- is investigated in the context of discrete habitats, modeled as graphs with a distinguished subset of vertices acting as sinks. These sinks impose boundary-like constraints analogous to Dirichlet conditions in continuous domains. The population proliferates locally at the vertices and diffuse across the network through the graph Laplacian. In the sinks the population cannot survive. The Dirichlet eigenvalue of the habitat is defined as the smallest eigenvalue of the principal submatrix of the Laplacian obtained by removing the rows and columns associated with sink vertices. This spectral parameter governs the habitat's viability: survival occurs when the Dirichlet eigenvalue of the habitat lies below a critical reaction-to-diffusion ratio. We study survival conditions for a sequence of random habitats built on binomial random graphs. We establish a law of large numbers for the corresponding sequence of Dirichlet eigenvalues and prove the emergence of a sharp threshold phenomenon: with high probability, a large random habitat is either viable or non-viable, depending on whether the reaction-to-diffusion ratio lies below or above this threshold. Our results provide the first general spectral theory for critical patch size on graphs, with implications for ecology, synthetic biology, and the modeling of processes on brain connectomes.

• The paper proves that the principal Dirichlet eigenvalue converges to sₙ · pₙ, defining a spectral threshold that determines habitat viability.
• It rigorously establishes sharp thresholds for persistence and extinction by analyzing different scaling regimes in Erdős-Rényi random graphs.
• The work bridges discrete spatial models with classical ecological theory, offering practical probabilistic bounds for species survival in network-structured habitats.

The Critical Patch Size Problem in Random Graphs

Introduction and Motivation

The paper addresses the longstanding ecological question of critical patch size—the minimum habitat size required to ensure population persistence—by reframing it in the context of discrete, network-structured environments. Rather than relying on classical PDE-based reaction-diffusion models, the authors investigate this threshold phenomenon on random graphs, where dispersal occurs through the graph Laplacian and extinction is modeled by imposing Dirichlet (absorbing) conditions at a subset of sink nodes. This approach connects applied ecology, spectral graph theory, and probabilistic combinatorics, yielding a treatment of persistence thresholds informed by the spectral properties of random habitats.

Model Formulation

The system is defined on a sequence of random graphs $G_n$ with $n$ vertices chosen from the Erdős-Rényi ensemble $\mathcal{G}(n, p_n)$. A set $S_n$ of $s_n$ nodes acts as absorbing boundaries (sinks) where population survives nowhere ($u_i = 0$ for $i \in S_n$). The population dynamics are governed by a discrete analog of the reaction-diffusion equation: $$\partial_t \mathbf{u} + L(G_n)\mathbf{u} = \rho\,\mathbf{u}, \quad \text{with}\ u_{i}|_{i\in S_n} = 0$$ where $L(G_n)$ is the graph Laplacian, and $\rho$ encodes the local reaction-to-diffusion ratio. Persistence (non-extinction) occurs if and only if the smallest eigenvalue $n$0 of the Dirichlet (sink-constrained) Laplacian—i.e., the principal submatrix with rows and columns indexed by non-sink nodes—satisfies $n$1.

The critical habitat size is thereby defined as the smallest $n$2 ensuring population survival with high probability, and the authors seek sharp probabilistic bounds on this critical threshold as a function of the number of sinks and the random graph parameters.

Main Results

Law of Large Numbers for the Dirichlet Eigenvalue

The authors prove a strong concentration phenomenon for the principal Dirichlet eigenvalue in large random graphs with many sinks. Specifically, under the scaling $n$3 (where $n$4 is the edge probability), the Dirichlet eigenvalue $n$5 satisfies

$n$6

Thus, the spectral threshold for persistence is governed asymptotically by the expected "boundary degree"—that is, the average number of connections from non-sink nodes to the sink set.

This result implies a sharp viability threshold: for a large random habitat, with high probability all instances are either viable or inviable, depending on whether the reaction-to-diffusion parameter $n$7 lies below or above $n$8.

Sharp Thresholds and Finite-Size Effects

The paper rigorously establishes threshold behavior for different regimes:

• When $n$9 and $\mathcal{G}(n, p_n)$0, the smallest eigenvalue tends to zero and persistence is impossible.
• In the quasi-dense regime ($\mathcal{G}(n, p_n)$1 and $\mathcal{G}(n, p_n)$2), the Dirichlet eigenvalue is tightly concentrated near $\mathcal{G}(n, p_n)$3.
• The threshold $\mathcal{G}(n, p_n)$4 splits the parameter space: for $\mathcal{G}(n, p_n)$5, habitats are almost surely viable; for $\mathcal{G}(n, p_n)$6, they are almost surely non-viable.

The authors also derive tail bounds (via Chernoff inequalities) for the smallest and average boundary degree, and thus provide explicit bounds on the probability that a random habitat fails to provide persistence.

Probabilistic Bounds on Critical Size

The analysis extends to provide quantitative finite-size criteria analogous to the KiSS (Kierstead–Slobodkin–Skellam) result in classical ecology. For fixed $\mathcal{G}(n, p_n)$7 sinks and extinction parameter $\mathcal{G}(n, p_n)$8, they show that with high probability at least a fraction $\mathcal{G}(n, p_n)$9 of random habitats with $S_n$0 vertices and $S_n$1 edges are viable if

$S_n$2

with $S_n$3.

These bounds connect the critical patch size directly to the graph structure (via average boundary degree), generalizing continuous-space spectral results to complex, stochastically assembled discrete habitats.

Analytical Techniques

The work's analysis combines:

• Spectral bounds: Employing classical Rayleigh quotient arguments to bracket the Dirichlet eigenvalue between the minimal and mean boundary degree.
• Concentration of measure: Leveraging Chernoff bounds and Borel–Cantelli to demonstrate almost sure convergence.
• Graph scaling regimes: Distinguishing connected and quasi-dense phases of Erdős-Rényi graphs, mapping the existence and positivity of the spectrum to basic results in random graph theory.
• Monotonicity of graph properties: Using monotone property coupling to transfer results between random graph models (e.g., $S_n$4 and $S_n$5).

The authors clarify that their method avoids recourse to the full machinery of matrix concentration inequalities; elementary probabilistic tools suffice to capture the essential behavior.

Implications

Theoretical:

The results provide a foundational spectral characterization of persistence in strongly stochastic, network-structured habitats. The law of large numbers for the Dirichlet eigenvalue bridges discrete spatial ecology and random matrix theory, facilitating the analysis of extinction and survival thresholds in complex systems.

Practical:

The framework is directly relevant for:

• Ecology: Modeling species persistence in fragmented or patchy landscapes where dispersal is mediated by an irregular, variable network.
• Synthetic biology: Informing the design of robust, compartmentalized bioengineered systems where functional units must survive despite imposed boundaries.
• Neuroscience: Analyzing network diffusion phenomena in brain connectomes, especially regarding failure propagation or survivability of neural subpopulations.

Future Directions:

Potential research avenues include extension to more structured or correlated graphs, incorporation of alternative local dynamics (e.g., Allee effects), and analysis of finite-ensemble corrections or rare-event statistics relevant for small systems.

Conclusion

This paper establishes a rigorous, probabilistic spectral theory for the critical patch size in random networks, demonstrating that, in large random graphs with extensive boundaries, the persistence threshold is governed by the expected boundary degree between viable and sink nodes. The results unify and generalize classical continuous-space critical patch size paradigms, providing new tools for the analysis and design of complex networked ecological, biological, and physical systems.

Reference: "The Critical Patch Size Problem in Random Graphs" (2604.00624)