Amir Dembo

Stanford University, Palo Alto, California, United States

Are you Amir Dembo?

Claim your profile

Publications (168)143.27 Total impact

  • Source
    Amir Dembo, Ofer Zeitouni
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider the quadratic optimization problem $$F_n^{W,h}:= \sup_{x \in S^{n-1}} ( x^T W x/2 + h^T x )\,, $$ with $W$ a (random) matrix and $h$ a random external field. We study the probabilities of large deviation of $F_n^{W,h}$ for $h$ a centered Gaussian vector with i.i.d. entries, both conditioned on $W$ (a general Wigner matrix), and unconditioned when $W$ is a GOE matrix. Our results validate (in a certain region) and correct (in another region), the prediction obtained by the mathematically non-rigorous replica method in Y. V. Fyodorov, P. Le Doussal, J. Stat. phys. 154 (2014).
    09/2014;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider recurrence versus transience for models of random walks on domains of $\mathbb{Z}^d$, in which monotone interaction enforces domain growth as a result of visits by the walk (or probes it sent), to the neighborhood of domain boundary.
    06/2014;
  • Source
    Sourav Chatterjee, Amir Dembo
    [Show abstract] [Hide abstract]
    ABSTRACT: We present a general technique for computing large deviations of nonlinear functions of independent Bernoulli random variables. The method is applied to compute the large deviation rate functions for subgraph counts in sparse random graphs. Previous technology, based on Szemer\'edi's regularity lemma, works only for dense graphs. Applications are also made to exponential random graphs and three-term arithmetic progressions in random sets of integers.
    01/2014;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Suppose that ${\mathcal G}$ is a finite, connected graph and $X$ is a lazy random walk on ${\mathcal G}$. The lamplighter chain $X^\diamond$ associated with $X$ is the lazy random walk on the wreath product ${\mathcal G}^\diamond = {\mathbb Z}_2 \wr {\mathcal G}$, the graph whose vertices consist of pairs $(\underline{f},x)$ where $\underline{f}$ is a $\{0,1\}$-labeling of the vertices of ${\mathcal G}$ and $x$ is a vertex in ${\mathcal G}$. In each step, $X^\diamond$ moves from a configuration $(\underline{f},x)$ by updating $x$ to $y$ using the transition rule of $X$, and if $x \ne y$, replacing $f_x$ and $f_y$ by two independent uniform random bits. The mixing time of the lamplighter chain on the discrete torus ${\mathbb Z}_n^d$ is known to have a cutoff at a time asymptotic to the cover time of ${\mathbb Z}_n^d$ if $d=2$, and to half the cover time if $d \geq 3$. We show that the mixing time of the lamplighter chain on ${\mathcal G}_n(a)={\mathbb Z}_n^2 \times {\mathbb Z}_{a \log n}$ has a cutoff at $\psi(a)$ times the cover time of ${\mathcal G}_n(a)$ as $n \to \infty$, where $\psi$ is a weakly decreasing map from $(0,\infty)$ onto $[1/2,1)$. In particular, as $a > 0$ varies, the threshold continuously interpolates between the known thresholds for ${\mathbb Z}_n^2$ and ${\mathbb Z}_n^3$. Perhaps surprisingly, we find a phase transition (non-smoothness of $\psi$) at the point $a_*=\pi r_3 (1+\sqrt{2})$, where high dimensional behavior ($\psi(a)=1/2$ for all $a>a_*$) commences. Here $r_3$ is the effective resistance from $0$ to $\infty$ in ${\mathbb Z}^3$.
    12/2013;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: For normally reflected Brownian motion and for simple random walk on growing in time d-dimensional domains, d>=3, we establish a sharp criterion for recurrence versus transience in terms of the growth rate.
    12/2013;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this short note, we present a theorem concerning certain "additive structure" for the level sets of non-degenerate Gaussian fields, which yields the multiple valley phenomenon for extremal fields with exponentially many valleys.
    10/2013;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We prove a Large Deviations Principle (LDP) for systems of diffusions (particles) interacting through their ranks, when the number of particles tends to infinity. We show that the limiting particle density is given by the unique solution of the approriate McKean-Vlasov equation and that the corresponding cumulative distribution function evolves according to the porous medium equation with convection. The large deviations rate function is provided in explicit form. This is the first instance of a LDP for interacting diffusions, where the interaction occurs both through the drift and the diffusion coefficients and where the rate function can be given explicitly. In the course of the proof, we obtain new regularity results for a certain tilted version of the porous medium equation.
    11/2012;
  • Source
    Anirban Basak, Amir Dembo
    [Show abstract] [Hide abstract]
    ABSTRACT: We show that the empirical eigenvalue measure for sum of $d$ independent Haar distributed $n$-dimensional unitary matrices, converge for $n \to \infty$ to the Brown measure of the free sum of $d$ Haar unitary operators. The same applies for independent Haar distributed $n$-dimensional orthogonal matrices. As a byproduct of our approach, we relax the requirement of uniformly bounded imaginary part of Stieltjes transform of $T_n$ that is made in [Guionnet, Krishnapur, Zeitouni; Theorem 1].
    Electronic communications in probability 08/2012; · 0.49 Impact Factor
  • Source
    Amir Dembo, Sumit Mukherjee
    [Show abstract] [Hide abstract]
    ABSTRACT: Consider random polynomial $\sum_{i=0}^n a_i x^i$ of independent mean-zero normal coefficients $a_i$, whose variance is a regularly varying function (in $i$) of order $\alpha$. We derive general criteria for continuity of persistence exponents for centered Gaussian processes, and use these to show that such polynomial has no roots in $[0,1]$ with probability $n^{-b_\alpha+o(1)}$, and no roots in $(1,\infty)$ with probability $n^{-b_0+o(1)}$, hence for $n$ even, it has no real roots with probability $n^{-2b_\alpha - 2b_0+o(1)}$. Here $b_\alpha =0$ when $\alpha \le -1$ and otherwise $b_\alpha \in (0,\infty)$ is independent of the detailed regularly varying variance function and corresponds to persistence probabilities for an explicit stationary Gaussian process of smooth sample path. Further, making precise the solution $\phi_d(x,t)$ to the $d$-dimensional heat equation initiated by a Gaussian white noise $\phi_d(x,0)$, we confirm that the probability of $\phi_d(x,t)\neq 0$ for all $t\in [1,T]$, is $T^{-b_{\alpha} + o(1)}$, for $\alpha=d/2-1$.
    08/2012;
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We provide an explicit formula for the limiting free energy density (log-partition function divided by the number of vertices) for ferromagnetic Potts models on uniformly sparse graph sequences converging locally to the d-regular tree for d even, covering all temperature regimes. This formula coincides with the Bethe free energy functional evaluated at a suitable fixed point of the belief propagation recursion on the d-regular tree, the so-called replica symmetric solution. For uniformly random d-regular graphs we further show that the replica symmetric Bethe formula is an upper bound for the asymptotic free energy for any model with permissive interactions.
    Communications in Mathematical Physics 07/2012; · 1.97 Impact Factor
  • Source
    Anirban Basak, Amir Dembo
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider the ferromagnetic Ising model on a sequence of graphs $G_n$ converging locally weakly to a rooted random tree. Generalizing [Montanari, Mossel, Sly '11], under an appropriate "continuity" property, we show that the Ising measures on these graphs converge locally weakly to a measure, which is obtained by first picking a random tree, and then the symmetric mixture of Ising measures with + and - boundary conditions on that tree. Under the extra assumptions that $G_n$, of uniformly bounded degrees, are edge-expanders, and ergodicity of the simple random walk on the limiting tree, we show that the local weak limit of the Ising measures conditioned on positive magnetization, is the Ising measure with + boundary condition on the limiting tree. We confirm the "continuity" and ergodicity properties in case of limiting (multi-type) Galton Watson trees, and the edge-expander property for the corresponding configuration model graphs.
    05/2012;
  • Source
    Amir Dembo, Jian Ding, Fuchang Gao
    [Show abstract] [Hide abstract]
    ABSTRACT: Let $S_n^{(2)}$ denote the iterated partial sums. That is, $S_n^{(2)}=S_1+S_2+ ... +S_n$, where $S_i=X_1+X_2+ ... s+X_i$. Assuming $X_1, X_2,....,X_n$ are integrable, zero-mean, i.i.d. random variables, we show that the persistence probabilities $$p_n^{(2)}:=\PP(\max_{1\le i \le n}S_i^{(2)}< 0) \le c\sqrt{\frac{\EE|S_{n+1}|}{(n+1)\EE|X_1|}},$$ with $c \le 6 \sqrt{30}$ (and $c=2$ whenever $X_1$ is symmetric). The converse inequality holds whenever the non-zero $\min(-X_1,0)$ is bounded or when it has only finite third moment and in addition $X_1$ is squared integrable. Furthermore, $p_n^{(2)}\asymp n^{-1/4}$ for any non-degenerate squared integrable, i.i.d., zero-mean $X_i$. In contrast, we show that for any $0 < \gamma < 1/4$ there exist integrable, zero-mean random variables for which the rate of decay of $p_n^{(2)}$ is $n^{-\gamma}$.
    Annales de l Institut Henri Poincaré Probabilités et Statistiques 05/2012; 49(3). · 0.93 Impact Factor
  • Source
    Amir Dembo, Andrea Montanari, Nike Sun
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider homogeneous factor models on uniformly sparse graph sequences converging locally to a (unimodular) random tree T, and study the existence of the free energy density phi, the limit of the log-partition function divided by the number of vertices n as n tends to infinity. We provide a new interpolation scheme and use it to prove existence of, and to explicitly compute, the quantity phi subject to uniqueness of a relevant Gibbs measure for the factor model on T. By way of example we compute phi for the independent set (or hard-core) model at low fugacity, for the ferromagnetic Ising model at all parameter values, and for the ferromagnetic Potts model with both weak enough and strong enough interactions. Even beyond uniqueness regimes our interpolation provides useful explicit bounds on phi. In the regimes in which we establish existence of the limit, we show that it coincides with the Bethe free energy functional evaluated at a suitable fixed point of the belief propagation (Bethe) recursions on T. In the special case that T has a Galton-Watson law, this formula coincides with the non-rigorous "Bethe prediction" obtained by statistical physicists using the "replica" or "cavity" methods. Thus our work is a rigorous generalization of these heuristic calculations to the broader class of sparse graph sequences converging locally to trees. We also provide a variational characterization for the Bethe prediction in this general setting, which is of independent interest.
    The Annals of Probability 10/2011; · 1.38 Impact Factor
  • Source
    Amir Dembo, Fuchang Gao
    [Show abstract] [Hide abstract]
    ABSTRACT: Let p_n denote the persistence probability that the first n iterated partial sums of integrable, zero-mean, i.i.d. random variables X_k, are negative. We show that p_n is bounded above up to universal constant by the square root of the expected absolute value of the empirical average of {X_k}. A converse bound holds whenever P(-X_1>t) is up to constant exp(-b t) for some b>0 or when P(-X_1>t) decays super-exponentially in t. Consequently, for such random variables we have that p_n decays as n^{-1/4} if X_1 has finite second moment. In contrast, we show that for any 0 < c < 1/4 there exist integrable, zero-mean random variables for which the rate of decay of p_n is n^{-c}.
    01/2011;
  • Source
    Amir Dembo, Nike Sun
    [Show abstract] [Hide abstract]
    ABSTRACT: Let T be a rooted supercritical multi-type Galton-Watson (MGW) tree with types coming from a finite alphabet, conditioned to non-extinction. The lambda-biased random walk (X_t, t>=0) on T is the nearest-neighbor random walk which, when at a vertex v with d(v) offspring, moves closer to the root with probability lambda/[lambda+d(v)], and to each of the offspring with probability 1/[lambda+d(v)]. This walk is recurrent for lambda>=rho and transient for 0<lambda<rho, with rho the Perron-Frobenius eigenvalue for the (assumed) irreducible matrix of expected offspring numbers. Subject to finite moments of order p>4 for the offspring distributions, we prove the following quenched CLT for lambda-biased random walk at the critical value lambda=rho: for almost every T, the process |X_{floor(nt)}|/sqrt{n} converges in law as n tends to infinity to a reflected Brownian motion rescaled by an explicit constant. This result was proved under some stronger assumptions by Peres-Zeitouni (2008) for single-type Galton-Watson trees. Following their approach, our proof is based on a new explicit description of a reversing measure for the walk from the point of view of the particle (generalizing the measure constructed in the single-type setting by Peres-Zeitouni), and the construction of appropriate harmonic coordinates. In carrying out this program we prove moment and conductance estimates for MGW trees, which may be of independent interest. In addition, we extend our construction of the reversing measure to a biased random walk with random environment (RWRE) on MGW trees, again at a critical value of the bias. We compare this result against a transience-recurrence criterion for the RWRE generalizing a result of Faraud (2011) for Galton-Watson trees.
    ELECTRONIC JOURNAL OF PROBABILITY 11/2010; · 0.79 Impact Factor
  • Source
    Amir Dembo, Andrea Montanari
    [Show abstract] [Hide abstract]
    ABSTRACT: Many problems of interest in computer science and information theory can be phrased in terms of a probability distribution over discrete variables associated to the vertices of a large (but finite) sparse graph. In recent years, considerable progress has been achieved by viewing these distributions as Gibbs measures and applying to their study heuristic tools from statistical physics. We review this approach and provide some results towards a rigorous treatment of these problems. Comment: 111 pages, 3 eps figures, Lecture notes for the 2008 Brazilian School of Probability (to appear in BJPS without Sections 1.3 and 6)
    Brazilian Journal of Probability and Statistics 10/2009; · 0.44 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We study the asymptotic behavior of the appropriately scaled and possibly perturbed spectral measure $\mu$ of large random real symmetric matrices with heavy tailed entries. Specifically, consider the N by N symmetric matrix $Y_N^\sigma$ whose (i,j) entry is $\sigma(i/N,j/N)X_{ij}$ where $(X_{ij}, 0<i<j+1<\infty)$ is an infinite array of i.i.d real variables with common distribution in the domain of attraction of an $\alpha$-stable law, $0<\alpha<2$, and $\sigma$ is a deterministic function. For a random diagonal $D_N$ independent of $Y_N^\sigma$ and with appropriate rescaling $a_N$, we prove that the distribution $\mu$ of $a_N^{-1}Y_N^\sigma + D_N$ converges in mean towards a limiting probability measure which we characterize. As a special case, we derive and analyze the almost sure limiting spectral density for empirical covariance matrices with heavy tailed entries.
    Communications in Mathematical Physics 12/2008; · 1.97 Impact Factor
  • Source
    Amir Dembo, Andrea Montanari
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider ferromagnetic Ising models on graphs that converge locally to trees. Examples include random regular graphs with bounded degree and uniformly random graphs with bounded average degree. We prove that the "cavity" prediction for the limiting free energy per spin is correct for any positive temperature and external field. Further, local marginals can be approximated by iterating a set of mean field (cavity) equations. Both results are achieved by proving the local convergence of the Boltzmann distribution on the original graph to the Boltzmann distribution on the appropriate infinite random tree. Comment: Published in at http://dx.doi.org/10.1214/09-AAP627 the Annals of Applied Probability (http://www.imstat.org/aap/) by the Institute of Mathematical Statistics (http://www.imstat.org)
    The Annals of Applied Probability 04/2008; · 1.37 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We consider the Fluctuation Dissipation Theorem (FDT) of statistical physics from a mathematical perspective. We formalize the concept of "linear response function" in the general framework of Markov processes. We show that for processes out of equilibrium it depends not only on the given Markov process X(s) but also on the chosen perturbation of it. We characterize the set of all possible response functions for a given Markov process and show that at equilibrium they all satisfy the FDT. That is, if the initial measure is invariant for the given Markov semi-group, then for any pair of times s<t and nice functions f,g, the dissipation, that is, the derivative in s of the covariance of g(X(t)) and f(X(s)) equals the infinitesimal response at time t and direction g to any Markovian perturbation that alters the invariant measure of X(.) in the direction of f at time s. The same applies in the so called FDT regime near equilibrium, i.e. in the limit s going to infinity with t-s fixed, provided X(s) converges in law to an invariant measure for its dynamics. We provide the response function of two generic Markovian perturbations which we then compare and contrast for pure jump processes on a discrete space, for finite dimensional diffusion processes, and for stochastic spin systems.
    Annales de l Institut Henri Poincaré Probabilités et Statistiques 10/2007; · 0.93 Impact Factor
  • Source
    Amir Dembo, Andrea Montanari
    [Show abstract] [Hide abstract]
    ABSTRACT: The (two) core of an hyper-graph is the maximal collection of hyper-edges within which no vertex appears only once. It is of importance in tasks such as efficiently solving a large linear system over GF[2], or iterative decoding of low-density parity-check codes used over the binary erasure channel. Similar structures emerge in a variety of NP-hard combinatorial optimization and decision problems, from vertex cover to satisfiability. For a uniformly chosen random hyper-graph of m=n\rho vertices and n hyper-edges, each consisting of the same fixed number l >= 3 of vertices, the size of the core exhibits for large n a first order phase transition, changing from o(n) for rho> rho_c to a positive fraction of n for rho<rho_c, with a transition window size Theta(n^{-1/2}) around rho_c>0. Analyzing the corresponding `leaf removal' algorithm, we determine the associated finite size scaling behavior. In particular, if rho is inside the scaling window (more precisely, rho = rho_c+r n^{-1/2}, the probability of having a core of size Theta(n) has a limit strictly between 0 and 1, and a leading correction of order Theta(n^{-1/6}). The correction admits a sharp characterization in terms of the distribution of a Brownian motion with quadratic shift, from which it inherits the scaling with n. This behavior is expected to be universal for wide collection of combinatorial problems.
    The Annals of Applied Probability 03/2007; · 1.37 Impact Factor

Publication Stats

4k Citations
143.27 Total Impact Points

Institutions

  • 1988–2012
    • Stanford University
      • • Department of Mathematics
      • • Department of Statistics
      • • Department of Electrical Engineering
      • • Information Systems Laboratory
      Palo Alto, California, United States
  • 2001
    • University of California, Berkeley
      • Department of Statistics
      Berkeley, CA, United States
  • 1984–1995
    • Technion - Israel Institute of Technology
      • Electrical Engineering Group
      Haifa, Haifa District, Israel
  • 1991
    • University of Haifa
      H̱efa, Haifa District, Israel
  • 1990
    • Massachusetts Institute of Technology
      • Laboratory for Information and Decision Systems
      Cambridge, MA, United States
  • 1987–1988
    • Brown University
      • Department of Applied Mathematics
      Providence, RI, United States