arXiv subjects

Amir Dembo

Publications and source records attributed to Amir Dembo.

At least 19 recordsLinked to original sources

arXiv2026

The law of (1+1)D SOS with an area tilt in a wedge

Motivated by the study of the level lines of the $(2+1)$D Solid-On-Solid (SOS) model above a floor, near the corners of the box, we derive the limit law of an ensemble of $K$ curves from a $(1+1)$D SOS model, with an area tilt, and above a wedge-shaped floor in $\{-N,\ldots,N\}$. We show that there exist explicit critical points $\alpha_0=1>\alpha_1>\ldots>\alpha_K>0$ such that, for each $r \geq 1$, along the intervals $\pm(\alpha_{r} N,\alpha_{r-1} N)$, the bottom $K+1-r$ curves, rescaled by $(N^{2/3},N^{1/3})$, tend to the law of a Geometrically-Area-Tilted Ensemble of non-crossing Brownian paths (Brownian GATE), independently across those $2K$ intervals. All other curves, centered and rescaled by $(N,\sqrt{N})$, tend to a product of $K$ suitable Brownian bridges.

math.PR↗

arXiv2026

The Gaussian Wave for Graphs of Finite Cone Type

We show that for any infinite tree of finite cone type satisfying a mild expansion condition, the only typical process on its vertices with covariance induced by the Green's function is the Gaussian wave. This generalizes a result of Backhausz and Szegedy, who proved this for the infinite regular tree of degree $d\geq 3$. We do this by giving a reduction to a statement concerning the distribution of the inner product of our process with columns of the Green's function, which in turn are straightforward to calculate. As a consequence, for random bipartite biregular graphs, the distribution of local neighborhoods of eigenvectors must approximate the Gaussian wave. Moreover, for generic configuration models including random lifts, the local distribution of a uniformly chosen eigenvector from any arbitrarily small spectral window likewise converges to the Gaussian wave.

math.PR↗

arXiv2025

Dynamics for spherical spin glasses: Gibbs distributed initial conditions

We derive the coupled non-linear integro-differential equations for the thermodynamic limit of the empirical correlation and response functions in the Langevin dynamics at temperature $T$, for spherical mixed $p$-spin disordered mean-field models, initialized according to a Gibbs measure for temperature $T_0$, in the replica-symmetric (RS) or $1$-replica-symmetry-breaking (RSB) phase. For any $T_0=T$ above the dynamical phase transition point $T_c^{\rm dyn}$ the resulting stationary relaxation dynamics coincide with the FDT solution for these equations, while for lower $T_0=T$ in the $1$-RSB phase, the relaxation dynamics coincides with the FDT solution, now concentrated on the single spherical band within the Gibbs measure's support on which the initial point lies.

math.PR↗

arXiv2025

The LZ78 Source

We study a family of processes generated according to sequential probability assignments induced by the LZ78 universal compressor. We characterize entropic and distributional properties such as their entropy and relative entropy rates, finite-state compressibility and log loss of their realizations, and the empirical distributions that they induce. Though not quite stationary, these sources are "almost stationary and ergodic;" similar to stationary and ergodic processes, they satisfy a Shannon-McMillan-Breiman-type property: the normalized log probability of their realizations converges almost surely to their entropy rate. Further, they are locally "almost i.i.d." in the sense that the finite-dimensional empirical distributions of their realizations converge almost surely to a deterministic i.i.d. law. However, unlike stationary ergodic sources, the finite-state compressibility of their realizations is almost surely strictly larger than their entropy rate by a "Jensen gap". We present simulations demonstrating the theoretical results. These sources allow to gauge the performance of sequential probability models, both classical and deep learning-based, on non-Markovian non-stationary data. As such, we apply realizations of the LZ78 source to the study of in-context learning in transformer models.

cs.IT↗

arXiv2024

Disordered Gibbs measures and Gaussian conditioning

We study the law of a random field $f_N(\boldsymbol{\sigma})$ evaluated at a random sample from the Gibbs measure associated to a Gaussian field $H_N(\boldsymbol{\sigma})$. In the high-temperature regime, we show that bounds on the probability that $f_N(\boldsymbol{\sigma})\in A$ for $\boldsymbol{\sigma}$ randomly sampled from the Gibbs measure can be deduced from similar bounds for deterministic $\boldsymbol{\sigma}$ under the conditional Gaussian law given that $H_N(\boldsymbol{\sigma})/N=E$ for $E$ close to the derivative $F'(\beta)$ of the free energy (which is the typical value of $H_N(\boldsymbol{\sigma})/N$ under the Gibbs measure). In the more challenging low-temperature regime we restrict to $k$-RSB spherical spin glasses, proving a similar result, now with a more elaborate conditioning. Namely, with $q_i$ denoting the locations of the non-zero atoms of the Parisi measure, in addition to specifying that $H_N(\boldsymbol{\sigma})/N=E$, here one needs to also condition on the energy and its gradient at points $\mathbf{x}_1,\ldots,\mathbf{x}_k$ such that $\langle \mathbf{x}_i,\mathbf{x}_j\rangle/N=q_{i\wedge j}$ and $\langle \mathbf{x}_i,\boldsymbol{\sigma}\rangle/N\approx q_{i}$. Like in the high-temperature phase, the energy and gradient values on which one conditions are also specified by the model's Parisi measure. We apply our general results to two important problems from statistical physics. That is, computing the Franz-Parisi potential at any temperature and, reducing certain asymptotics of Langevin dynamics with initial conditions distributed according to the Gibbs measure, to the more manageable problem of studying dynamics with non-random initial conditions and conditional disorder.

math.PR↗

arXiv2023

Potts and random cluster measures on locally regular-tree-like graphs

Fixing $\beta \ge 0$ and an integer $q \ge 2$, consider the ferromagnetic $q$-Potts measures $\mu_n^{\beta,B}$ on finite graphs ${\sf G}_n$ on $n$ vertices, with external field strength $B \ge 0$ and the corresponding random cluster measures $\varphi^{q,\beta,B}_{n}$. Suppose that as $n \to \infty$ the uniformly sparse graphs ${\sf G}_n$ converge locally to an infinite $d$-regular tree ${\sf T}_{d}$, $d \ge 3$. We show that the convergence of the Potts free energy density to its Bethe replica symmetric prediction (which has been proved in case $d$ is even, or when $B=0$), yields the local weak convergence of $\varphi^{q,\beta,B}_n$ and $\mu_n^{\beta,B}$ to the corresponding free or wired random cluster measure, Potts measure, respectively, on ${\sf T}_{d}$. The choice of free versus wired limit is according to which has the larger Potts Bethe functional value, with mixtures of these two appearing {as limit points on} the critical line $\beta_c(q,B)$ where these two values of the Bethe functional coincide. For $B=0$ and $\beta>\beta_c$, we further establish a pure-state decomposition by showing that conditionally on the same dominant color $1 \le k \le q$, the $q$-Potts measures on such edge-expander graphs ${\sf G}_n$ converge locally to the $q$-Potts measure on ${\sf T}_{d}$ with a boundary wired at color $k$.

math.PR↗

arXiv2023

KPZ-type equation from growth driven by a non-Markovian diffusion

We study a stochastic PDE model for an evolving set $\mathbb{M}(t)\subseteq\mathbb{R}^{\mathrm{d}+1}$ that resembles a continuum version of origin-excited or reinforced random walk. We show that long-time fluctuations of an associated height function are given by a regularized Kardar-Parisi-Zhang (KPZ)-type PDE on a hypersurface in $\mathbb{R}^{\mathrm{d}+1}$, modulated by a Dirichlet-to-Neumann operator. We also show that for $\mathrm{d}+1=2$, the regularization in this KPZ-type equation can be removed after renormalization. To our knowledge, this gives the first instance of KPZ-type behavior in Laplacian growth, which was asked about (for somewhat different models) in Parisi-Zhang '84 and Ramirez-Sidoravicius '04.

math.PR↗

arXiv2023

A flow-type scaling limit for random growth with memory

We study a stochastic Laplacian growth model, where a set $\mathbf{U}\subseteq\mathbb{R}^{\mathrm{d}}$ grows according to a reflecting Brownian motion in $\mathbf{U}$ stopped at level sets of its boundary local time. We derive a scaling limit for the leading-order behavior of the growing boundary (i.e. "interface"). It is given by a geometric flow-type PDE. It is obtained by an averaging principle for the reflecting Brownian motion. We also show that this geometric flow-type PDE is locally well-posed, and its blow-up times correspond to changes in the diffeomorphism class of the growth model. Our results extend those of Dembo-Groisman-Huang-Sidoravicius '21, which restricts to star-shaped growth domains and radially outwards growth, so that in polar coordinates, the geometric flow transforms into a simple ODE with infinite lifetime. Also, we remove the "separation of scales" assumption that was taken in Dembo-Groisman-Huang-Sidoravicius '21; this forces us to understand the local geometry of the growing interface.

math.PR↗

arXiv2023

Spectral measure for uniform $d$-regular digraphs

Consider the matrix $A_{\mathcal{G}}$ chosen uniformly at random from the finite set of all $N$-dimensional matrices of zero main-diagonal and binary entries, having each row and column of $A_{\mathcal{G}}$ sum to $d$. That is, the adjacency matrix for the uniformly random $d$-regular simple digraph $\mathcal{G}$. Fixing $d \ge 3$, it has long been conjectured that as $N \to \infty$ the corresponding empirical eigenvalue distributions converge weakly, in probability, to an explicit non-random limit, %measure $\mu_d$ on $\mathbb{C}$, which is given by the Brown measure of the free sum of $d$ Haar unitary operators. We reduce this conjecture to bounding the decay in $N$ of the probability that the minimal singular value of the shifted matrix $A(w) = A_{\mathcal{G}} - w I$ is very small. While the latter remains a challenging task, the required bound is comparable to the recently established control on the singularity of $A_{\mathcal{G}}$. The reduction is achieved here by sharp estimates on the behavior at large $N$, near the real line, of the Green's function (aka resolvent) of the Hermitization of $A(w)$, which is of independent interest.

math.PR↗

arXiv2023

Regularity method and large deviation principles for the Erdős--Rényi hypergraph

We develop a quantitative large deviations theory for random hypergraphs, which rests on tensor decomposition and counting lemmas under a novel family of cut-type norms. As our main application, we obtain sharp asymptotics for joint upper and lower tails of homomorphism counts in the $r$-uniform Erdős--Rényi hypergraph for any fixed $r\ge 2$, generalizing and improving on previous results for the Erdős--Rényi graph ($r=2$). The theory is sufficiently quantitative to allow the density of the hypergraph to vanish at a polynomial rate, and additionally yields tail asymptotics for other nonlinear functionals, such as induced homomorphism counts.

math.PR↗

arXiv2022

Typical structure of sparse exponential random graph models

We consider general Exponential Random Graph Models (ERGMs) where the sufficient statistics are functions of homomorphism counts for a fixed collection of simple graphs $F_k$. Whereas previous work has shown a degeneracy phenomenon in dense ERGMs, we show this can be cured by raising the sufficient statistics to a fractional power. We rigorously establish the na\"ive mean-field approximation for the partition function of the corresponding Gibbs measures, and in case of "ferromagnetic" models with vanishing edge density show that typical samples resemble a typical Erd\H{o}s--R\'enyi graph with a planted clique and/or a planted complete bipartite graph of appropriate sizes. We establish such behavior also for the conditional structure of the Erd\H{o}s--R\'enyi graph in the large deviations regime for excess $F_k$-homomorphism counts. These structural results are obtained by combining quantitative large deviation principles, established in previous works, with a novel stability form of a result of [5] on the asymptotic solution for the associated entropic variational problem. A technical ingredient of independent interest is a stability form of Finner's generalized H\"older inequality.

math.PR↗

arXiv2022

Capacity of the range of random walk: The law of the iterated logarithm

We establish both the $\limsup$ and the $\liminf$ law of the iterated logarithm (LIL), for the capacity of the range of a simple random walk in any dimension $d\ge 3$. While for $d \ge 4$, the order of growth in $n$ of such LIL at dimension $d$ matches that for the volume of the random walk range in dimension $d-2$, somewhat surprisingly this correspondence breaks down for the capacity of the range at $d=3$. We further establish such LIL for the Brownian capacity of a $3$-dimensional Brownian sample path and novel, sharp moderate deviations bounds for the capacity of the range of a $4$-dimensional simple random walk.

math.PR↗

arXiv2022

On the limiting law of line ensembles of Brownian polymers with geometric area tilts

We study the line ensembles of non-crossing Brownian bridges above a hard wall, each tilted by the area of the region below it with geometrically growing pre-factors. This model, which mimics the level lines of the $(2+1)$D SOS model above a hard wall, was studied in two works from 2019 by Caputo, Ioffe and Wachtel. In those works, the tightness of the law of the top $k$ paths, for any fixed $k$, was established under either zero or free boundary conditions, which in the former setting implied the existence of a limit via a monotonicity argument. Here we address the open problem of a limit under free boundary conditions: we prove that as the interval length, followed by the number of paths, go to $\infty$, the top $k$ paths converge to the same limit as in the free boundary case, as conjectured by Caputo, Ioffe and Wachtel.

math.PR↗

arXiv2021

Diffusions interacting through a random matrix: universality via stochastic Taylor expansion

Consider $(X_{i}(t))$ solving a system of $N$ stochastic differential equations interacting through a random matrix $\mathbf J = (J_{ij})$ with independent (not necessarily identically distributed) random coefficients. We show that the trajectories of averaged observables of $(X_i(t))$, initialized from some $μ$ independent of $\mathbf J$, are universal, i.e., only depend on the choice of the distribution $\mathbf{J}$ through its first and second moments (assuming e.g., sub-exponential tails). We take a general combinatorial approach to proving universality for dynamical systems with random coefficients, combining a stochastic Taylor expansion with a moment matching-type argument. Concrete settings for which our results imply universality include aging in the spherical SK spin glass, and Langevin dynamics and gradient flows for symmetric and asymmetric Hopfield networks.

math.PR↗

arXiv2021

Upper Tail For Homomorphism Counts In Constrained Sparse Random Graphs

Consider the upper tail probability that the homomorphism count of a fixed graph $H$ within a large sparse random graph $G_n$ exceeds its expected value by a fixed factor $1+δ$. Going beyond the Erdős-Rényi model, we establish here explicit, sharp upper tail decay rates for sparse random $d_n$-regular graphs (provided $H$ has a regular $2$-core), and for sparse uniform random graphs. We further deal with joint upper tail probabilities for homomorphism counts of multiple graphs $H_1,\ldots, H_k$ (extending the known results for $k=1$), and for inhomogeneous graph ensembles (such as the stochastic block model), we bound the upper tail probability by a variational problem analogous to the one that determines its decay rate in the case of sparse Erdős-Rényi graphs.

math.PR↗

arXiv2020

Everything is a Race and Nakamoto Always Wins

Nakamoto invented the longest chain protocol, and claimed its security by analyzing the private double-spend attack, a race between the adversary and the honest nodes to grow a longer chain. But is it the worst attack? We answer the question in the affirmative for three classes of longest chain protocols, designed for different consensus models: 1) Nakamoto's original Proof-of-Work protocol; 2) Ouroboros and SnowWhite Proof-of-Stake protocols; 3) Chia Proof-of-Space protocol. As a consequence, exact characterization of the maximum tolerable adversary power is obtained for each protocol as a function of the average block time normalized by the network delay. The security analysis of these protocols is performed in a unified manner by a novel method of reducing all attacks to a race between the adversary and the honest nodes.

cs.CR↗

arXiv2020

Averaging Principle and Shape Theorem for a Growth Model with Memory

We present a general approach to study a class of random growth models in $n$-dimensional Euclidean space. These models are designed to capture basic growth features which are expected to manifest at the mesoscopic level for several classical self-interacting processes originally defined at the microscopic scale. It includes once-reinforced random walk with strong reinforcement, origin-excited random walk, and few others, for which the set of visited vertices is expected to form a "limiting shape". We prove an averaging principle that leads to such shape theorem. The limiting shape can be computed in terms of the invariant measure of an associated Markov chain.

math.PR↗

arXiv2020

Dynamics for spherical spin glasses: disorder dependent initial conditions

We derive the thermodynamic limit of the empirical correlation and response functions in the Langevin dynamics for spherical mixed $p$-spin disordered mean-field models, starting uniformly within one of the spherical bands on which the Gibbs measure concentrates at low temperature for the pure $p$-spin models and mixed perturbations of them. We further relate the large time asymptotics of the resulting coupled non-linear integro-differential equations, to the geometric structure of the Gibbs measures (at low temperature), and derive their FDT solution (at high temperature).

math.PR↗