Abstract: We prove that an integrated simple random walk, where random walk and integrated random walk are conditioned to return to zero, has asymptotic probability $n^{-1/2}$ to stay positive. This question is motivated by so-called random polymer models and proves a conjecture by Caravenna and Deuschel.
Introduction
In the study of stochastic processes, "persistence" refers to the probability that a process remains on one side of a boundary—typically zero—over a long duration. For a standard random walk, this probability generally follows a power law $n^{-\theta}$, where $\theta$ is the persistence exponent. This paper investigates the persistence of an integrated simple random walk (IRW) conditioned to return to its origin, a structure known as an integrated random walk bridge.
The research addresses a specific conjecture proposed by Caravenna and Deuschel (2009) regarding the behavior of random polymer chains with Laplace interactions. While it was hypothesized that the persistence exponent for such a bridge should be $1/2$, a rigorous mathematical proof remained elusive until now.
The Integrated Random Walk Model
Let $S_n$ be a simple random walk, defined as the sum of independent symmetric Bernoulli variables. The integrated random walk, denoted $A_n$, represents the cumulative sum of the positions of $S_n$:
$$A_n = \sum_{i=1}^{n} S_i$$
Geometrically, $A_n$ can be viewed as the area under the path of the random walk. The authors focus on the integrated random walk bridge, where the process is "pinned" at both ends. Specifically, the system is conditioned such that at time $4n$, both the walk and its integral return to zero ($S_{4n} = 0$ and $A_{4n} = 0$). The goal is to determine the probability that $A_k \geq 0$ for all $k \in \{1, \dots, 4n\}$ under these boundary conditions.
Methodological Framework
To solve this problem, the authors developed several novel probabilistic techniques to bridge the gap between discrete random walks and continuous stochastic calculus:
1. Discrete Local Limit Theorem (LLT)
A primary challenge in this study was the lack of a joint Local Limit Theorem for the pair $(S_n, A_n)$ in a discrete setting. While continuous versions existed, the authors derived a new LLT for simple random walks with Bernoulli steps. They utilized a "three-domain approach" involving characteristic functions to rigorously define the joint distribution.
2. Three-Period Partitioning
The proof strategy involves segmenting the total time $4n$ into three distinct intervals: the first $n$ steps, a middle $2n$ segment, and the final $n$ steps. By analyzing the "pinned" ends separately, the authors could demonstrate that the paths from both ends meet in the middle with a predictable probability.
3. Time Reversion and Adjoint Processes
To handle the conditioning at the terminal end ($4n$), the authors employed adjoint processes and time reversion. This allowed them to treat the end of the bridge with the same mathematical rigor as the beginning, ensuring symmetry in the analysis of the path's persistence.
4. Convergence to Brownian Motion
The authors proved a Central Limit Theorem (CLT) for the pinned discrete process, showing that as $n$ increases, the scaled discrete process converges to a pair consisting of Brownian motion and integrated Brownian motion. This convergence allows the discrete problem to be solved using the properties of continuous Gaussian processes.
Key Insights and Results
The Impact of Double Conditioning
One of the most significant findings is the sensitivity of the persistence exponent to boundary conditions. For an unconditioned integrated random walk, the persistence exponent is $1/4$. Interestingly, if only the walk is conditioned to return to zero ($S_{4n}=0$), the exponent remains $1/4$. However, when both the walk and its integral are conditioned to return to zero ($S_{4n}=0, A_{4n}=0$), the exponent doubles to $1/2$.
Main Theorem
The paper's central result confirms the Caravenna-Deuschel conjecture. Formally, the authors prove:
$$\mathbb{P}(A_1 \geq 0, \dots, A_{4n} \geq 0 \mid A_{4n} = S_{4n} = 0) \sim n^{-1/2}$$
This result demonstrates that under the condition of staying positive, the process $(S_n, A_n)$ does not merely stay near the origin but scales naturally ($S_n \sim n^{1/2}$ and $A_n \sim n^{3/2}$), which is essential for the path to successfully "bridge" the two zero-points.
Conclusion and Future Directions
By providing a rigorous proof for the $n^{-1/2}$ persistence probability, this work solidifies the theoretical foundation for random polymer models with fixed boundary conditions. While the current proof is tailored to simple random walks with Bernoulli steps, the methodology—particularly the "gluing" of conditioned processes and the new discrete LLT—provides a framework that may be applicable to more general step distributions and complex boundary value problems in stochastic processes.
Summary updated