Correlated Site Percolation Problem

• Correlated site percolation is a process where lattice site occupancy is statistically linked, leading to unique scaling laws and critical behavior.
• It exhibits non-classical critical exponents (e.g., β≈1, γ≈3.9 in 3D) that distinguish it from independent site percolation models.
• Finite-size scaling and simulation methods are crucial for extracting thresholds and exponents, highlighting the impact of spatial correlations on connectivity.

Correlated site percolation describes percolative phenomena where the occupation states of lattice sites are statistically correlated, in contrast to classical Bernoulli percolation in which site occupations are independent. The paradigmatic instance of the Correlated Site Percolation Problem (CSPP) is the percolation of sites not visited by a random walk (RW) on a hypercubic lattice, a model that realizes precise long-range spatial correlations with explicit scaling exponents distinct from the uncorrelated universality class (Levi et al., 2024, Kantor et al., 2019, Levi et al., 18 Jan 2026). CSPP also encompasses a broader class of models, including short-range positively associated lattices, Ising and habitat aggregation constructions, correlated random networks, and models with explicit algebraic, geometric, or conditional dependencies.

1. Model Definitions and Key Mechanisms

The prototypical CSPP, as formulated in (Levi et al., 2024, Kantor et al., 2019), is defined on a $d$-dimensional hypercubic lattice of linear size $L$ with periodic boundary conditions. A single RW of ${\cal N}=uL^d$ steps is performed; every site visited at least once is considered "removed," while unvisited sites are declared "occupied." The field of surviving (unvisited) sites, $\{\sigma_i\}$, exhibits strong spatial correlations: for $r=|i-j|\ll L$, the connected part of the two-point correlation scales as $\langle \sigma_i \sigma_j \rangle_c \sim 1/r^{d-2}$ for $d>2$ (Levi et al., 2024, Kantor et al., 2019).

Other correlated site percolation frameworks include:

• Positive-association models, where short-range (e.g., nearest-neighbor) correlations are imposed by explicit functions of (possibly overlapping) random variables, but real-space correlations decay rapidly (Dianati et al., 2013, Huth et al., 2014).
• Field-theoretic and graph-based generalizations, as in the degree-correlated configuration model or Ising/Potts droplets, where an underlying Hamiltonian or graph structure induces nontrivial correlations (Coniglio et al., 2016, Allard et al., 2015, Cao et al., 2012).
• Models with algebraic-long-range disorder, where the assignment of site states is a deterministic or stochastic function of auxiliary fields with specified spatial power spectra (Schrenk et al., 2013).

2. Statistical Correlations and Universality

In CSPP generated by RW removal, the marginal probability that a site remains unvisited decays exponentially with the normalized walk length, $p(u) \sim \exp(-A_d u)$, with $A_d$ the asymptotic fraction of new sites visited per step (Kantor et al., 2019). The field of occupancies is non-Gaussian with dominant two-point correlations decaying as $1/r^{d-2}$, which is a defining feature for universality (Levi et al., 2024).

The spatial correlation structure controls the universality class of the percolative phase transition. As established by the Weinrib-Halperin criterion, correlations decaying as $1/r^a$ with $a<d$ generically alter the scaling exponents if $a<2/\nu_{\mathrm{uncorr}}$; in the RW-vacant-site model, this mechanism is realized with $a=d-2$ and thus strongly modifies universality for $3\leq d<6$ (Levi et al., 2024, Kantor et al., 2019).

Other correlated percolation models display a range of behaviors:

• In positively associated, finitely correlated models, the percolation threshold shifts (typically lowers for positive correlations), but critical exponents remain unchanged from the uncorrelated case, as correlation length is finite (Dianati et al., 2013, Huth et al., 2014).
• In models with algebraically decaying correlations or conditional constraints (e.g., Pauli exclusion, aggregation rules, Ising droplets), both thresholds and exponents may be modified, or even discontinuous transitions may arise (Coniglio et al., 2016, Cao et al., 2012, Maksymenko et al., 2012, Schrenk et al., 2013).

3. Critical Behavior, Exponents, and Scaling Laws

In CSPP defined by unvisited sites of random walks, a sharp phase transition occurs at a critical walk length $u_c$, separating regimes with ($u<u_c$) and without ($u>u_c$) a system-spanning cluster of vacant sites. Key observables and their scaling are:

• Order parameter: $$P_\infty(u) = \lim_{L\to\infty} \langle M_1 \rangle / L^d \sim (u_c - u)^\beta$$ for $u \to u_c^-$.
• Mean finite cluster size: $S(u) \sim (u_c-u)^{-\gamma}$.
• Correlation length: $\xi \sim |u-u_c|^{-\nu}$, with critical exponent $\nu=2/(d-2)$ for $3\leq d\leq 6$ (Weinrib’s long-range value).
• Fractal dimension: $d_f = d-\beta/\nu$. At $u_c$ in $d=3$, $d_f=5/2$ (Levi et al., 18 Jan 2026, Levi et al., 2024).
• Cluster size distribution: For large clusters ($s\gg 1$), $n_s \sim s^{-\tau}$, with $\tau=1+d/d_f$.

Representative exponent values for $d=3$:

• $u_c\approx3.15$, $\nu=2.00$, $\beta=0.99\pm0.04$, $\gamma=3.90\pm0.05$, $d_f=2.5$, $\tau=1.2$ (Levi et al., 2024, Levi et al., 18 Jan 2026).

These exponents differ markedly from uncorrelated site percolation; e.g., in $d=3$, uncorrelated exponents are $\beta\approx0.41$, $\gamma\approx1.80$, $\nu\approx0.88$ (Levi et al., 2024, Coniglio et al., 2016). As $d\to6$, CSPP exponents converge toward mean-field values.

For large clusters ($r\gg 1$), the $r$th largest cluster mass at criticality scales as $M_r \sim L^{5/2}/r^{5/6}$ in $d=3$ (Levi et al., 18 Jan 2026). This scaling is a direct consequence of the anomalous fractal dimension and the broad power-law distribution of cluster sizes characteristic of CSPP universality.

4. Methodologies for Critical Threshold and Exponent Extraction

Critical parameters and exponents in CSPP are extracted through robust finite-size scaling (FSS) analyses (Levi et al., 2024, Kantor et al., 2019, Levi et al., 18 Jan 2026):

• Threshold location: The ratio $R(L,u) = \langle M_1 \rangle / \langle M_2 \rangle$ becomes independent of $L$ at $u_c$; intersection points of $R$ vs.\ $u$ curves for different $L$ determine $u_c$.
• Exponent measurement: At $u=u_c$, $M_1 \sim L^{d_f}$; fits to $\log M_1$ vs.\ $\log L$ yield $d_f$. Given $\nu$, $\beta$ follows from $d_f=d-\beta/\nu$. The mean finite cluster size $S(L,u)$ is collapsed across sizes using $S(L,u)=L^{\gamma/\nu}\tilde S[(u_c-u)L^{1/\nu}]$ to obtain $\gamma$.

Such approaches are essential due to the non-Bernoulli nature of site correlations, which preclude the use of standard independence-based binomial convolutions.

5. Relation to Other Correlated Percolation Models

CSPP generated by RW removal is part of a broader taxonomy of correlated percolation:

Model Type	Correlation Decay	Universality Class	Typical Threshold/Scaling
RW-removal (CSPP)	Power-law, $r^{-(d-2)}$	New (Weinrib)	$u_c\approx 3$, $\nu=2/(d-2)$
Finite-range, positive	Nearest-neighbor	Uncorrelated	Lowered $p_c$, but $\nu$ unchanged
Ising/Potts droplets	Exp./Power-law	Thermal (Ising/Potts)	Threshold tied to $T_c$
Field-based algebraic	$r^{2H}$ (2D landscapes)	$H$-dependent (Schrenk et al., 2013)	Continuously varying exponents

Long-range correlations (as in the RW model, power-law fields, or voter processes with slow decay) can drive the emergence of new universality classes with genuinely non-classical exponents (Levi et al., 2024, Kantor et al., 2019, Ráth et al., 2017, Schrenk et al., 2013, Coniglio et al., 2016). In contrast, finite-range or exponentially decaying correlations typically only shift thresholds (Dianati et al., 2013, Huth et al., 2014). Iterative and algorithmically defined models (e.g., cluster recoloring, habitat aggregation) can interpolate between these regimes, with scaling behavior sensitive to the induced correlation structure (Li et al., 2023, Huth et al., 2014).

6. Implications and Outstanding Problems

Correlated percolation exposes the profound sensitivity of global connectivity to the details of local correlation structure. The CSPP realized by RW removal provides a clean and tractable paradigm for studying long-range critical correlations, applicable to physical processes ranging from enzyme degradation and epidemic trace percolation to random environment and porous medium connectivity (Levi et al., 2024, Levi et al., 18 Jan 2026, Kantor et al., 2019). Open research directions include detailed characterization of cluster statistics beyond the leading exponents, transport and conductivity phenomena on correlated vacant sets, behavior in $d=2$ (where there is no sharp threshold), and analytical extensions to cases with multiple or interacting walkers (Kantor et al., 2019, Levi et al., 18 Jan 2026).

Rank-size laws—such as the scaling $M_r \sim L^{d_f}/r^{d_f/d}$—reveal that CSPP at threshold supports a hierarchy of large, fractal clusters rather than a unique dominant entity, contrasting classical percolation (Levi et al., 18 Jan 2026). This feature may offer practical advantages for incipient cluster analysis and network vulnerability diagnostics.

7. Summary of Key Scaling Relations and Results

CSPP exponents and scaling functions for $3\le d<6$ are as follows (Levi et al., 2024):

• Correlation length: $\xi\sim|u-u_c|^{-\nu}$, $\nu=2/(d-2)$.
• Order parameter: $P_\infty\sim(u_c-u)^{\beta}$, with $\beta\approx 1$.
• Mean finite cluster size: $S\sim(u_c-u)^{-\gamma}$, $\gamma\approx 4/(d-2)$.
• FSS forms for observables:

$$P_\infty(L,u) = L^{-\beta/\nu} \widetilde{P}[(u_c-u)L^{1/\nu}], \quad S(L,u) = L^{\gamma/\nu} \widetilde{S}[(u_c-u)L^{1/\nu}]$$

• For $d=3$: $u_c\approx 3.15$, $\nu=2$, $\beta=0.99\pm0.04$, $\gamma=3.90\pm0.05$, $d_f=2.5$.
• At threshold, the mean mass of the $r$-th largest cluster: $M_r \sim L^{5/2}/r^{5/6}$ ($d=3$).

These results quantitatively distinguish the CSPP universality class from both short-range-correlated and Bernoulli percolation. The general theoretical and computational framework developed for CSPP is extendable to many correlated percolation processes of current physical and mathematical interest (Levi et al., 2024, Levi et al., 18 Jan 2026, Coniglio et al., 2016, Schrenk et al., 2013).