Random Walk Pinning Model Overview

• The Random Walk Pinning Model (RWPM) is a framework that reweights random walk or renewal trajectories using exponential pinning rewards, defining a Gibbs measure with a partition function and associated free energy.
• It explores various formulations—including discrete, continuous, and renewal approaches—to analyze localization–delocalization transitions through control parameters such as the summed pinning potential and critical thresholds.
• Key methodologies like block-splitting, local-CLT estimates, and renewal representations connect RWPM to broader topics such as sparse random environments, scaling limits, and disorder relevance.

Random Walk Pinning Model (RWPM) denotes a family of pinning and wetting models in which a random-walk or renewal trajectory is reweighted by an exponential reward for contacts: visits to selected levels, contacts with a wall, or coincidence with another random walk. In the cited literature, these models share a common thermodynamic structure: a Gibbs weight built from a contact functional, a partition function, a free energy defined through asymptotic logarithmic growth, and a localization–delocalization transition characterized by whether that free energy is positive (Caputo et al., 2014, Berger et al., 2010, Berger et al., 10 Sep 2025). The term is therefore used for several closely related, but not identical, constructions.

1. Canonical formulations

A discrete multi-level formulation considers an integer-valued, symmetric, irreducible random walk $X=\{X_0,\dots,X_L\}$ with mean zero and variance

$\sigma^2=\sum_k k^2 p(k),$

conditioned to start and end at $0$ and to stay nonnegative: $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$ Given a nonnegative pinning sequence $\epsilon=(\epsilon_0,\epsilon_1,\dots)$, the pinning potential is

$$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$

and the partition function is

$$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$

Its free energy

$$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$

exists by subadditivity; localization means $F(\sigma,\epsilon)>0$, while delocalization means $F(\sigma,\epsilon)=0$ (Caputo et al., 2014).

A continuous-time formulation replaces the static substrate by a moving catalyst. Let $\sigma^2=\sum_k k^2 p(k),$0 and $\sigma^2=\sum_k k^2 p(k),$1 be independent continuous-time random walks, with $\sigma^2=\sum_k k^2 p(k),$2 interpreted as quenched disorder. The Hamiltonian is the overlap time

$\sigma^2=\sum_k k^2 p(k),$3

and the quenched partition function is

$\sigma^2=\sum_k k^2 p(k),$4

The quenched free energy

$\sigma^2=\sum_k k^2 p(k),$5

exists almost surely in $\sigma^2=\sum_k k^2 p(k),$6, is nonnegative, and defines a critical point

$\sigma^2=\sum_k k^2 p(k),$7

The localized phase is $\sigma^2=\sum_k k^2 p(k),$8, and the delocalized phase is $\sigma^2=\sum_k k^2 p(k),$9 (Berger et al., 10 Sep 2025).

A renewal-based disordered pinning formulation uses a renewal process $0$0 with inter-arrival law $0$1, disorder variables $0$2, Hamiltonian

$0$3

and grand canonical partition function

$0$4

This model enters RWPM theory through exact correspondences with random walks in sparse random environments and through critical-window limits (Poisat, 2024, Wei et al., 2024).

2. Multi-level pinning and wetting thresholds

For the conditioned nonnegative walk with potential $0$5, the central control parameter is

$0$6

The principal result is a sharp localization–delocalization criterion: there exists a universal constant $0$7 such that delocalization holds when $0$8, and localization holds when $0$9. In the latter regime one in fact proves $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$0. Equivalently, the critical value $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$1 lies in $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$2 independently of the detailed shape of $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$3 (Caputo et al., 2014).

The proof of delocalization is organized around a recursive bound on the partition function. The path is decomposed according to the last return to $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$4 before $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$5 and the first return after $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$6, which produces a convolution structure. The multi-level potential is then decoupled by Jensen’s inequality by writing $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$7 as a convex combination of single-level pinning terms with strengths $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$8. The remaining task is a uniform estimate on the single-level partition functions $$\Omega_{0,L}^{+,0}=\{X:X_0=X_L=0,\;X_n\ge 0\ \forall n\}.$$9, obtained by induction at the critical diffusive scale

$\epsilon=(\epsilon_0,\epsilon_1,\dots)$0

with a base case controlled by fine local-CLT estimates and an induction step based on Fourier-type or combinatorial splitting at time $\epsilon=(\epsilon_0,\epsilon_1,\dots)$1 (Caputo et al., 2014).

This threshold is explicitly compared with the Bargmann–Jost–Pais criterion for the absence of negative bound states for the radial Schrödinger equation. In the pinning model, the quantity $\epsilon=(\epsilon_0,\epsilon_1,\dots)$2 plays the role of $\epsilon=(\epsilon_0,\epsilon_1,\dots)$3, and the small-$\epsilon=(\epsilon_0,\epsilon_1,\dots)$4 delocalization condition is the discrete analogue of the Jost–Pais bound (Caputo et al., 2014).

The same strategy extends to self-avoiding paths $\epsilon=(\epsilon_0,\epsilon_1,\dots)$5 in $\epsilon=(\epsilon_0,\epsilon_1,\dots)$6, weighted by $\epsilon=(\epsilon_0,\epsilon_1,\dots)$7 with $\epsilon=(\epsilon_0,\epsilon_1,\dots)$8 large. Contacts are counted by

$\epsilon=(\epsilon_0,\epsilon_1,\dots)$9

Up to replacing $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$0 by $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$1, the same threshold holds. The argument requires additional Peierls-type estimates to control overhangs and to show that, with overwhelming probability, the path touches each vertical line at most once, thereby recovering an approximate random-walk structure (Caputo et al., 2014).

3. Moving catalysts, quenched disorder, and Harris-type phase diagrams

In the continuous-time RWPM, the reward is proportional to the time spent by $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$2 on the quenched trajectory $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$3. For transient walks, this model exhibits a critical inverse temperature $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$4, with $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$5 below criticality and $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$6 above it. Comparison with the homogeneous case gives $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$7 and $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$8 (Berger et al., 10 Sep 2025).

A detailed disorder-relevance theory is available for transient $$\Phi(X)=\sum_{j\ge 0}\epsilon_j N_j(X),\qquad N_j(X)=|\{0<n<L:X_n=j\}|,$$9-stable walks on $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$0, with $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$1, $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$2. For $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$3, disorder is relevant: there is $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$4 such that

$$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$5

with a matching upper bound from the companion paper. In the marginal case $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$6, disorder is always relevant, independently of the slowly varying prefactor $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$7, and explicit lower bounds on $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$8 are given according to whether $$Z_{0,L}^{\Phi,+,0}=\sum_{X\in\Omega_{0,L}^{+,0}} w(X)e^{\Phi(X)}, \quad w(X)=\prod_{n=1}^L p(X_n-X_{n-1}).$$9 with $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$0, $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$1, or $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$2. For $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$3, disorder is irrelevant for small $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$4, in particular $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$5 for $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$6 small enough, although the disorder still has a non-trivial effect on the free-energy curve for every $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$7 (Berger et al., 10 Sep 2025, Berger et al., 10 Sep 2025).

The $$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$8 regime is not completely trivial. One has

$$F(\sigma,\epsilon)=\lim_{L\to\infty}\frac1L\log Z_{0,L}^{\Phi,+,0}$$9

for every $F(\sigma,\epsilon)>0$0, there exists $F(\sigma,\epsilon)>0$1 such that $F(\sigma,\epsilon)>0$2 for $F(\sigma,\epsilon)>0$3, and at annealed criticality the constrained partition function satisfies

$F(\sigma,\epsilon)>0$4

in $F(\sigma,\epsilon)>0$5-probability (Berger et al., 10 Sep 2025).

Earlier results for continuous-time RWPM on $F(\sigma,\epsilon)>0$6 established two structural features. First, in $F(\sigma,\epsilon)>0$7 disorder smooths the phase transition: for all $F(\sigma,\epsilon)>0$8,

$F(\sigma,\epsilon)>0$9

so the quenched transition is at least of second order, even when the annealed transition is first order. Second, at low temperature,

$F(\sigma,\epsilon)=0$0

which shows that disorder changes the large-$F(\sigma,\epsilon)=0$1 correction to the free energy (Berger et al., 2010).

4. Renewal representations, sparse environments, and annealed–quenched comparisons

A central structural fact is that several RWPM variants admit renewal representations. In the continuous-time moving-catalyst model, the constrained partition function can be written as a weighted renewal with kernel

$F(\sigma,\epsilon)=0$2

and random weights

$F(\sigma,\epsilon)=0$3

This representation underlies coarse-graining, second-moment estimates, and disorder-relevance results (Berger et al., 10 Sep 2025).

A different exact dictionary appears in the connection between the random pinning model and one-dimensional random walks in sparse random environments. For a fixed renewal environment $F(\sigma,\epsilon)=0$4, a classical martingale or ruin-estimate argument gives

$F(\sigma,\epsilon)=0$5

and averaging over $F(\sigma,\epsilon)=0$6 yields

$F(\sigma,\epsilon)=0$7

Thus the grand canonical partition function of the pinning model coincides with the mean number of returns to the origin of a random walk in a random sparse environment averaged on the randomness location. This identity translates pinning criticality into integrability thresholds for return times in annealed and partially annealed setups (Poisat, 2024).

The question whether annealed and quenched transition points coincide is model dependent rather than universal. In a $F(\sigma,\epsilon)=0$8-dimensional directed walk near a corrugated wall with i.i.d. site disorder $F(\sigma,\epsilon)=0$9, the first moment gives an annealed critical point

$\sigma^2=\sum_k k^2 p(k),$00

while the second moment yields a corner-localization threshold $\sigma^2=\sum_k k^2 p(k),$01. Writing $\sigma^2=\sum_k k^2 p(k),$02, annealed and quenched transitions coincide exactly when

$\sigma^2=\sum_k k^2 p(k),$03

Poissonian and Gaussian potentials satisfy $\sigma^2=\sum_k k^2 p(k),$04, whereas asymmetric bimodal disorder may lie on either side of the criterion (Xu et al., 21 Jul 2025).

5. Scaling limits and pathwise properties

RWPM analysis is not confined to free energies. For pinning and wetting models based on random walks with finite variance, an optimal integrability result is available for the diffusively rescaled maximum

$\sigma^2=\sum_k k^2 p(k),$05

Under the free law, meander law, bridge law, excursion law, and analogous laws conditioned to avoid zero, one has

$\sigma^2=\sum_k k^2 p(k),$06

Equivalently, for some $\sigma^2=\sum_k k^2 p(k),$07,

$\sigma^2=\sum_k k^2 p(k),$08

As an application, for every $\sigma^2=\sum_k k^2 p(k),$09 and every $\sigma^2=\sum_k k^2 p(k),$10, the family of diffusive rescalings $\sigma^2=\sum_k k^2 p(k),$11 is tight on $\sigma^2=\sum_k k^2 p(k),$12 (Caravenna, 2018).

The proof uses a regeneration–excursion decomposition of the zero-level set, together with a general tightness criterion based on tightness of bulk and final excursion laws and on uniform tail-smallness of the excursion maximum. Local-limit estimates, conditioned-walk asymptotics, martingale maximal inequalities, and fluctuation identities are the key inputs (Caravenna, 2018).

At marginal relevance, the theory reaches a genuine critical-window limit. For a disordered pinning model induced by a random walk with mean zero, unit variance, vanishing third cumulant, and finite fourth moment, one has

$\sigma^2=\sum_k k^2 p(k),$13

The critical inverse-temperature scale is

$\sigma^2=\sum_k k^2 p(k),$14

If $\sigma^2=\sum_k k^2 p(k),$15 lies in the refined critical window

$\sigma^2=\sum_k k^2 p(k),$16

then the point-to-point partition random measure converges to a unique limiting random measure $\sigma^2=\sum_k k^2 p(k),$17, the Critical Disordered Pinning Measure (Wei et al., 2024).

The same limiting object also appears in a continuous counterpart involving a mollified stochastic heat equation and a critical stochastic Volterra equation. This identifies a common critical scaling structure for discrete pinning and a continuum noise-driven model (Wei et al., 2024).

6. Structural themes and interpretive scope

Across the cited formulations, several techniques recur. The discrete multi-level problem relies on local-CLT estimates, spectral or Jensen decoupling, and recursive block-splitting at the scale $\sigma^2=\sum_k k^2 p(k),$18 (Caputo et al., 2014). The moving-catalyst model uses renewal representations, coarse-graining, Paley–Zygmund estimates, fractional-moment bounds, size-biased laws, Poisson constructions, and stochastic domination (Berger et al., 10 Sep 2025, Berger et al., 10 Sep 2025). Low-temperature and smoothing results exploit tilting of the law of the disorder walk and block decompositions around jump times (Berger et al., 2010).

A common misconception is that disorder relevance, or coincidence of quenched and annealed critical points, should be universal within pinning models. The results assembled here point in the opposite direction. In the moving-catalyst RWPM, the answer depends sharply on the decay exponent: $\sigma^2=\sum_k k^2 p(k),$19 gives relevance, while $\sigma^2=\sum_k k^2 p(k),$20 gives irrelevance for small $\sigma^2=\sum_k k^2 p(k),$21 (Berger et al., 10 Sep 2025). In the directed corrugated-wall model, coincidence may or may not occur depending on the disorder distribution $\sigma^2=\sum_k k^2 p(k),$22 through a second-moment criterion involving $\sigma^2=\sum_k k^2 p(k),$23 (Xu et al., 21 Jul 2025).

The broad significance of RWPM is therefore not a single universal phase diagram, but a set of related contact-interaction models in which localization can be studied with unusual precision. Depending on the formulation, the model connects to wetting of conditioned walks, self-avoiding contour models, sparse random environments, stochastic heat-flow limits, and the discrete analogue of the Bargmann–Jost–Pais criterion (Caputo et al., 2014, Poisat, 2024, Wei et al., 2024).