Free Askey–Wilson functionals and geometric last passage percolation on a strip

Włodek Bryc, Kamil Szpojankowski, Jacek Wesołowski · 2025

Read Free Askey–Wilson functionals and geometric last passage percolation on a strip by Włodek Bryc, Kamil Szpojankowski, Jacek Wesołowski (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01879.txt.

Introduction

Wlodek Bryc and co-authors extend formulas for a growth model on a strip using new free-probability tools in this 2025 paper. Last passage percolation tracks the best path through random weights. Confining the grid to a strip gives the model a steady state.

Earlier work found that steady state and integral formulas for its generating functions. The pair of objects introduced here, free Askey-Wilson functionals, widen those formulas. The wider range covers boundary settings that were out of reach before. With them, the full phase diagram of long-run behavior falls out. Probability models of this kind echo traffic, queues, and crystal growth. Exact phase maps help all those fields at once.

Catalog record

CREATORWłodek Bryc, Kamil Szpojankowski, Jacek Wesołowski
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE135 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-01879

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Free Askey–Wilson functionals and geometric last passage percolation on a strip Włodek Bryc, Kamil Szpojankowski, Jacek Wesołowski Abstract . Barraquand, Corwin, and Yang [ Barraquand et al., 2024 ] established that geometric last passage percolation (LPP) on a strip of ℤ 2 \mathbb{Z}^{2} has a unique stationary measure. Building on this, Barraquand [ Barraquand, 2024 ] derived explicit contour integral formulas for the model’s multipoint probability generating function. In this paper, we introduce free Askey–Wilson functionals and use them to extend these generating function formulas. Our framework yields explicit expressions valid over a broader range of boundary parameters than previously accessible. This generalization allows us to determine the full phase diagram that characterizes how the large-scale asymptotics of the stationary measure depend on the boundary conditions. In addition, we prove a Poisson approximation for the stationary measure when the parameters vary with the strip width. 1. Introduction Stationary measures of out-of-equilibrium systems in finite domains typically exhibit a nontrivial dependence on boundary conditions. A classical example is the asymmetric simple exclusion process (ASEP) on N sites, coupled to boundary reservoirs (open ASEP). In this work, we demonstrate that analogous phenomena arise in the stationary measure of the directed geometric last passage percolation (LPP) model on a strip, recently introduced in [ Barraquand et al., 2024 ] . A technical reason underlying this similarity in phase diagrams is that, somewhat unexpectedly, the two models are linked through integral representations of their generating functions, first discovered in the fan region of ASEP in [ Bryc and Wesołowski, 2017 ] , and more recently for geometric LPP in [ Barraquand, 2024 ] . The objective of the present work is to deepen the theoretical understanding of such representations and to extend their applicability to the full range of boundary parameters. 1.1. Geometric last passage percolation on a strip For N=1,2,\dots , we consider a strip \mathbb{S}_{N}:=\{(j,n):n\in\mathbb{Z}_{\geq 0},j\in\{n,n+1,\dots,n+N\}\}\subset\mathbb{Z}_{\geq 0}^{2} of width N+1 of the \mathbb{Z}^{2} lattice with vertices (j,n) . Vertices (n,n) for n\geq 0 are called the left boundary vertices; vertices (n+N,n) for n\geq 0 are called the right boundary vertices and the remaining vertices (j,n) for n\geq 0 and n<j<n+N are called the bulk vertices (at the n -th level). We write X\sim Geo(p) if a random variable X has the geometric distribution P(X=n)=p^{n}(1-p) , n=0,1,\dots with parameter 0\leq p<1 . Definition 1.1 (geometric LPP) . Let \mathsf{a}\in(0,1) be the bulk parameter and let \mathsf{c}_{1},\mathsf{c}_{2}\geq 0 be boundary parameters such that \mathsf{a}\mathsf{c}_{1},\mathsf{a}\mathsf{c}_{2}<1 . Let \left(\omega_{j,n}\right)_{1\leq n\leq j\leq n+N} be a sequence of independent geometric random variables indexed by the vertices of the strip above level 0 . We assume that at the boundaries of the strip, \omega_{n,n}\sim Geo(\mathsf{a}\mathsf{c}_{1}) , \omega_{n,n+N}\sim Geo(\mathsf{a}\mathsf{c}_{2}) and that in the bulk n<j<n+N we have \omega_{j,n}\sim Geo(\mathsf{a}^{2}) , n=1,2,\dots . For a given initial condition G(i,0)\geq 0 , 0\leq i\leq N with G(0,0)=0 , the geometric LPP is the sequence of random variables G(j,n) indexed by the vertices (j,n) of the strip, which solve the following recursion: for j\geq 0 , n\geq 1 , (1.1) G(j,n)=\omega_{j,n}+\max\{G(j-1,n),G(j,n-1)\},\quad n\leq j\leq n+N, with the convention that G(n-1,n)\equiv 0 and G(n+N,n-1)\equiv 0 for the vertices in \mathbb{Z}^{2} that are outside of the strip \mathbb{S}_{N} . In particular, G(j,1)=\max_{1\leq k\leq j}\{G(k,0)+\sum_{i=k}^{j}\omega_{k,1}\} , j=1,\dots,N+1 . We are interested in the laws of the vectors \mathbf{G}_{n}:=(G(j+n,n)-G(n,n))_{j=0,\dots,N} along the horizontal paths at level n . Definition 1.2 (stationary measure) . We say that the law \nu on \mathbb{Z}_{\geq 0}^{N} is a stationary measure for the LPP if under the initial condition \mathbf{G}_{0} that has law \nu and is independent of \left(\omega_{j,n}\right)_{1\leq n\leq j\leq n+N} , the law of vector \mathbf{G}_{n} is given by the same law \nu for any n\geq 1 . Barraquand, Corwin and Yang [ Barraquand et al., 2024 , Theorem 1.3] established a remarkable result that if \mathsf{c}_{1},\mathsf{c}_{2}\in(0,1/\mathsf{a}) , then the geometric LPP has a unique stationary measure, which is given as a marginal of the following two-layer ensemble. For \mathbf{m}=(m_{1},\dots m_{N}) , \mathbf{n}=(n_{1},\dots,n_{N})\in\mathbb{Z}_{\geq 0}^{N} consider two sequences of partial sums \mathbf{L}_{1},\mathbf{L}_{2} given by \mathbf{L}_{1}(j)=\sum_{k=1}^{j}m_{k} , \mathbf{L}_{2}(j)=\sum_{k=1}^{j}n_{k} for j=1,\dots,N , with \mathbf{L}_{1}(0)=\mathbf{L}_{2}(0)=0 . With \mathsf{a}\in(0,1) , \mathsf{c}_{1},\mathsf{c}_{2}\in(0,1/\mathsf{a}) . we assign the probability measure to the pair of sequences (\mathbf{L}_{1},\mathbf{L}_{2}) using the weight function (1.2) \mathbb{Q}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}_{N}(\mathbf{L}_{1},\mathbf{L}_{2})=\mathsf{a}^{\mathbf{L}_{1}(N)}\mathsf{a}^{\mathbf{L}_{2}(N)}(\mathsf{c}_{1}\mathsf{c}_{2})^{\max_{1\leq j\leq N}(\mathbf{L}_{2}(j)-\mathbf{L}_{1}(j-1))}\mathsf{c}_{2}^{\mathbf{L}_{1}(N)-\mathbf{L}_{2}(N)}. which according to [ Barraquand et al., 2024 , Proposition 2.18] (see also Lemma 3.1 ) can be normalized by the finite constant (1.3) \mathcal{Z}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}(N)=\sum_{\mathbf{L}_{1},\mathbf{L}_{2}}\mathbb{Q}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}_{N}(\mathbf{L}_{1},\mathbf{L}_{2}). Theorem 1.1 ( [ Barraquand et al., 2024 , Theorem 1.6] ) . If \mathsf{a}\in(0,1) , \mathsf{a}\mathsf{c}_{1},\mathsf{a}\mathsf{c}_{2}\in(0,1) then the marginal law \nu of \mathbf{L}_{1} under the normalized measure (1.4) \mathbb{P}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}_{\rm Geo}(\mathbf{L}_{1},\mathbf{L}_{2})=\frac{1}{\mathcal{Z}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}(N)}\mathbb{Q}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}_{N}(\mathbf{L}_{1},\mathbf{L}_{2}) is the unique stationary measure of the geometric LPP. More recently, Guillaume Barraquand [ Barraquand, 2024 ] derived explicit integral formulas for the d -point probability generating function. In the simplest case, d=1 , his formula is equivalent to the following: (1.5) G_{N}(t):=\sum_{\mathbf{L}_{1},\mathbf{L}_{2}}t^{2\mathbf{L}_{1}(N)}\mathbb{Q}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}_{N}(\mathbf{L}_{1},\mathbf{L}_{2})=\frac{1-\mathsf{c}_{1}\mathsf{c}_{2}}{2\pi}\int_{-2}^{2}\,\frac{\sqrt{4-y^{2}}}{\mathfrak{h}_{\mathsf{a}t}^{N}(y)\,\mathfrak{h}_{\mathsf{c}_{1}/t}(y)\,\mathfrak{h}_{\mathsf{c}_{2}t}(y)}\,{\textnormal{d}}y, where (1.6) \mathfrak{h}_{\alpha}(y)=1+\alpha^{2}-\alpha y. provided that \mathsf{c}_{1},\mathsf{c}_{2}<1 , \mathsf{a}t<1 , \mathsf{c}_{1}<t<1/\mathsf{c}_{2} . The series on the left-hand side of ( 1.5 ) converges when \mathsf{a}\mathsf{c}_{1},\mathsf{a}t,\mathsf{a}\mathsf{c}_{2}t^{2}<1 , so according to [ Barraquand, 2024 ] formula ( 1.5 ) holds in this generality after the right-hand side is replaced by its analytic continuation in parameters \mathsf{c}_{1},\mathsf{c}_{2} . Our goal is to make this analytic continuation more explicit, so that we can use it to prove limit theorems by analyzing the Laplace transform \mathbb{E}[e^{2s\mathbf{L}_{1}(N)/N}]=G_{N}(e^{s/N})/\mathcal{Z}^{\left(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}\right)}=G_{N}(e^{s/N})/G_{N}(1). It turns out that formula ( 1.5 ) yields a concise expression that remains valid over a larger range of parameters than allowed in ( 1.5 ). Proposition 1.2 . If 0<\mathsf{a},\mathsf{a}t,\mathsf{a}\mathsf{c}_{1},\mathsf{a}\mathsf{c}_{2}t^{2}<1 , then (1.7) \sum_{N=1}^{\infty}z^{N-1}G_{N}(t)=\frac{B}{\mathsf{a}(t-\mathsf{c}_{1}B)(1-\mathsf{c}_{2}tB)}, where B=B(z,\mathsf{a}t)<1 is the smaller root of the quadratic equation B+\frac{1}{B}=\frac{1-z}{\mathsf{a}t}+\mathsf{a}t with 0\leq z<(1-\mathsf{a}t)^{2} which is small enough to ensure that B\mathsf{c}_{1}<t,B\mathsf{c}_{2}t<1 . In this paper, however, we adopt a different approach, working directly with ( 1.2 ). For parameters satisfying 0<\mathsf{a}\mathsf{c}_{1},\mathsf{a}t,\mathsf{a}\mathsf{c}_{2}t^{2}<1 , Theorem 1.4 allows us to replace ( 1.5 ) with (1.8) G_{N}(t)=\mathbb{L}^{\left(\mathsf{c}_{1}/t,\mathsf{c}_{2}t\right)}[\mathfrak{h}_{\mathsf{a}t}^{-N}], where (1.9) \mathbb{L}^{\left(a,b\right)}[f]=\frac{1}{2\pi{\textnormal{i}}}\oint_{|w|=\rho}f(w+1/w)\frac{1-w^{2}}{(1-aw)(1-bw)}\frac{{\textnormal{d}}w}{w},\quad\rho,|a|\rho,|b|\rho<1, is the moment functional for the Askey-Wilson polynomials [ Askey and Wilson, 1985 ] with parameters (a,b,0,0) and q=0 , extended to an appropriate class of analytic functions. By analyzing the representation ( 1.8 ) for the Laplace transform \mathbb{E}\left[e^{2s\mathbf{L}_{1}(N)/N}\right]=G_{N}(e^{s/N})/G_{N}(1) , we derive the phase diagram for the geometric LPP, which closely parallels the phase diagram known for the ASEP [ Derrida et al., 1993 ] , as predicted by universality; see Fig. 1 . (A similar phase diagram also appears in [ Yang, 2024 ] for a different model.) The large-scale limit of \mathbf{L}_{1}(N)/N depends on the boundary parameters \mathsf{c}_{1} and \mathsf{c}_{2} as detailed below. 1 1 {\mathsf{c}_{1}} {\mathsf{c}_{2}} (0,0) H H \rho=\tfrac{\mathsf{a}\mathsf{c}_{2}}{1-\mathsf{a}\mathsf{c}_{2}} L L \rho=\tfrac{\mathsf{a}}{\mathsf{c}_{1}-\mathsf{a}} \rho=\tfrac{\mathsf{a}}{1-\mathsf{a}} I \frac{1}{\mathsf{a}} \frac{1}{\mathsf{a}} Figure 1. The limit \frac{1}{N}\mathbf{L}_{1}(N)\to\rho=\rho(\mathsf{a},\mathsf{c}_{1},\mathsf{c}_{2}) and the boundary parameters \mathsf{c}_{1},\mathsf{c}_{2}\in(0,1/\mathsf{a}) . Below the hyperbola \mathsf{c}_{1}\mathsf{c}_{2}=1 , the functional \pi^{t} in ( 1.12 ) is given as an integral with respect to a positive Askey–Wilson measure. For (\mathsf{c}_{1},\mathsf{c}_{2})\in I:=(0,1]^{2} the limit does not depend on the boundary parameters. The line between the points (1,1) and (1/\mathsf{a},1/\mathsf{a}) separates the regions L and H with the low and high average and the limit of \frac{1}{N}\mathbf{L}_{1}(N) on that line is random, ( 1.10 ). Theorem 1.3 . Let \mathsf{a}\in(0,1) and \mathsf{c}_{1},\mathsf{c}_{2}\in(0,1/\mathsf{a}) . (i) If \mathsf{c}_{1},\mathsf{c}_{2}\leq 1 , then \mathbf{L}_{1}(N)/N\to\mathsf{a}/(1-\mathsf{a}) in probability. (ii) If \mathsf{c}_{1}>1 and

Read the complete text