True Self-Avoiding Walks (TSAW)

• TSAW is a non-Markovian stochastic process on integer lattices that assigns exponentially decreasing transition probabilities to repeatedly visited sites or edges.
• It exhibits super-diffusive scaling in one dimension—converging to the Tóth–Werner process—and standard Brownian motion in higher dimensions.
• Analytical tools such as the Ray–Knight theorem and local time formalism are central to understanding TSAW dynamics and its connection to event-chain Monte Carlo algorithms.

The true self-avoiding walk (TSAW) is a non-Markovian stochastic process on the integer lattice, characterized by its propensity to avoid regions it has visited repeatedly by assigning exponentially decreasing transition probabilities to frequently traversed sites or edges. Distinct from simple self-avoiding walks, which strictly prohibit revisits, the TSAW incorporates a “soft” memory, leading to super-diffusive scaling in one dimension and diffusive scaling in higher dimensions. The rigorous analysis of TSAW, particularly its scaling limits and universal exponents, forms a central subject in modern probability and statistical mechanics.

1. Mathematical Definition and Local-Time Formalism

TSAW on $\mathbb{Z}$ is defined by assigning the $n$th position $X_n$ transition probabilities that penalize revisiting previously accessed sites or bonds. The construction uses:

• Edge local times: For $k\in\mathbb{Z}$,

$$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$

• Site local time:

$$L(n, k) = \sum_{i=0}^n 1_{X_i = k} = E_n^+(k) + E_n^-(k) + 1_{X_n=k}$$

• Transition probability at time $n$:

$$P(X_{n+1} = X_n + 1 \mid \mathcal{F}_n) = \frac{\exp\left[ -\beta (E_n^+(X_n) + E_n^-(X_n+1)) \right]} { \exp\left[ -\beta (E_n^+(X_n) + E_n^-(X_n+1)) \right] + \exp\left[ -\beta (E_n^+(X_n-1) + E_n^-(X_n)) \right] }$$

where $\lambda = e^{-\beta} \in (0,1)$ penalizes repeated edge crossings (Kosygina et al., 16 Feb 2025, Maggs, 2024). In $d$ dimensions ($n$0), the process is equivalently described by a rate function $n$1 governing movement along edges, possibly in continuous time, via

$n$2

where $n$3 is the time-integrated local occupation of site $n$4 up to time $n$5 (Horvath et al., 2010, Horvath et al., 2010).

2. Scaling Laws, Limiting Processes, and Ray–Knight Structure

In $n$6, the TSAW displays super-diffusive scaling and converges, under suitable rescaling, to the Tóth–Werner true self-repelling motion (TSRM), a process defined via its local-time fields and intricate interaction between reflected/absorbed Brownian motions:

• The spatial coordinate rescales as $n$7,
• The local time at the current position rescales as $n$8,
• The scaling exponents are $n$9 for displacement and $X_n$0 for local-time "height" (Kosygina et al., 16 Feb 2025, Maggs, 2024).

The limiting process $X_n$1 is continuous but non-Markovian, being encoded as the projection of a space-filling curve $X_n$2 constructed from a two-sided Brownian motion in the $X_n$3 variable, reflected and absorbed on intervals parameterized by local-times and merge-absorption points (joint generalized Ray–Knight theorem).

In $X_n$4, by contrast, TSAW is diffusive with variance proportional to $X_n$5, and its scaling limit is Brownian motion with nondegenerate covariance, confirming the original predictions of Amit–Parisi–Peliti (Horvath et al., 2010, Horvath et al., 2010).

3. Analytical and Probabilistic Tools

A critical component in establishing functional limit theorems for TSAW is the "joint generalized Ray–Knight theorem" (GRKT). For one-dimensional models:

• The discrete Ray–Knight curve at $X_n$6 is

$X_n$7

where $X_n$8 is the $X_n$9st hitting time of $k\in\mathbb{Z}$0, and $k\in\mathbb{Z}$1.

• Multi-curve convergence in law is established, and merge/absorption points track where local-time fields coalesce.

Tightness and identification of finite-dimensional distributions follow by Laplace functional analysis of hitting-times and inversion of occupation-time formulas (Kosygina et al., 16 Feb 2025). Martingale decompositions of local-time difference processes underpin submartingale properties required in delicate gambler's-ruin type estimates.

In higher dimensions, the proof relies on non-reversible Kipnis–Varadhan theory and the enhanced graded-sector condition, employing martingale approximations and resolvent bounds for additive functionals of the environment process as viewed from the walker (Horvath et al., 2010, Horvath et al., 2010).

4. Universality, Extensions, and Higher-Dimensional Behavior

The super-diffusive exponents $k\in\mathbb{Z}$2 (displacement) and $k\in\mathbb{Z}$3 (local height) are argued to be universal for a broad class of one-dimensional self-repelling walks (Kosygina et al., 16 Feb 2025, Maggs, 2024). The functional limit approach "inverts" the Ray–Knight framework: the only possible scaling limits consistent with limiting Ray–Knight curves are versions of the Tóth–Werner process.

In $k\in\mathbb{Z}$4, both the TSAW and its continuous-space analogue, the self-repelling Brownian polymer, exhibit diffusive scaling; the limiting process is Brownian motion, and the non-trivial local-time interactions are encoded in the stationary ergodic Gibbs measure for the environment seen from the walker. The scaling in $k\in\mathbb{Z}$5 is conjectured to have logarithmic corrections, but no rigorous results establish a polymer-type scaling limit (Horvath et al., 2010, Horvath et al., 2010). Extensions include continuous-time analogues and non-nearest-neighbor walks.

The table below summarizes scaling exponents and limiting behaviors:

Dimension $k\in\mathbb{Z}$6	Displacement Scaling	Limiting Process
$k\in\mathbb{Z}$7	$k\in\mathbb{Z}$8	Tóth–Werner TSRM
$k\in\mathbb{Z}$9	$$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$0	Brownian motion
$$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$1 (conjectural)	Gaussian (logarithmic corrections)	Not rigorously identified

5. Connections to Event-Chain Monte Carlo and Algorithmic Implications

Recent work demonstrates a direct connection between TSAW and the event-chain Monte Carlo (ECMC) algorithm in one dimension (Maggs, 2024). In tension-free harmonic chains, ECMC's statistics of active-particle displacement and update counts exactly replicate those of TSAW end-to-end separation and local-time distributions, including the hallmark $$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$2 super-diffusive exponent.

Perturbations such as linear factor-fields ("pressure" tuning) restore symmetry and optimal algorithmic convergence (dynamical exponent $$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$3), while non-harmonic potentials and inhomogeneities show robust persistence of TSAW universality. The mapping between TSAW and ECMC suggests that the unlikely crossing probabilities and coalescence times in TSAW analysis correspond to large-scale mixing and relaxation phenomena in irreversible Markov chain samplers.

Numerical studies confirm analytic predictions for dynamic exponents, mixing times, and the robustness of core singularities in the scaling limit, highlighting the emergence of a novel universality class in one-dimensional chains distinguished by TSAW-like dynamics (Maggs, 2024).

6. Open Problems and Ongoing Research Directions

Key open questions include:

• The universality of TSAW scaling limits for more general self-interacting walks and environments where Ray–Knight theorems are available, including site-repulsion and multi-cookie walks (Kosygina et al., 16 Feb 2025).
• The rigorous identification of polymer-type scaling limits and possible non-trivial universality classes in dimensions $$E_n^+(k) = \#\{i<n : X_i=k,\, X_{i+1}=k+1\},\quad E_n^-(k) = \#\{i<n : X_i=k,\, X_{i+1}=k-1\}$$4 and higher, as predictions from physics remain unproven (Horvath et al., 2010, Horvath et al., 2010).
• The detailed study of mixing times, autocorrelation structures, and relaxation dynamics for TSAW and ECMC in non-equilibrium and rugged energy landscapes (Maggs, 2024).
• The mathematical understanding of persistent anomalous features (such as the cusp singularity in displacement distribution under broken balance) within the ECMC–TSAW correspondence.

These problems connect the theory of self-interacting random walks, generalized Ray–Knight frameworks, and the algorithmic design of irreversible Markov chains, underscoring the interdisciplinary nature of the TSAW paradigm.