Searcharxiv⌕ Search

arXiv subjects

Amir Dembo

Publications and source records attributed to Amir Dembo.

75 records · Page 5Linked to original sources

Thick points for intersections of planar sample paths

Let $L_n^{X}(x)$ denote the number of visits to $x \in {\bf Z}^2$ of the simple planar random walk $X$, up till step $n$. Let $X'$ be another simple planar random walk independent of $X$. We show that for any $0<b<1/(2 π)$, there are $n^{1-2πb+o(1)}$ points $x \in {\bf Z}^2$ for which $L_n^{X}(x)L_n^{X'}(x)\geq b^2 (\log n)^4$. This is the discrete counterpart of our main result, that for any $a<1$, the Hausdorff dimension of the set of {\it thick intersection points} $x$ for which $\limsup_{r \to 0} {\mathcal I}(x,r)/(r^2|\log r|^4)=a^2$, is almost surely $2-2a$. Here ${\mathcal I}(x,r)$ is the projected intersection local time measure of the disc of radius $r$ centered at $x$ for two independent planar Brownian motions run till time 1. The proofs rely on a `multi-scale refinement' of the second moment method. In addition, we also consider analogous problems where we replace one of the Brownian motions by a transient stable process, or replace the disc of radius $r$ centered at $x$ by $x+rK$ for general sets $K$.

math.PR↗

Critical Behavior in Lossy Source Coding

The following critical phenomenon was recently discovered. When a memoryless source is compressed using a variable-length fixed-distortion code, the fastest convergence rate of the (pointwise) compression ratio to the optimal $R(D)$ bits/symbol is either $O(\sqrt{n})$ or $O(\log n)$. We show it is always $O(\sqrt{n})$, except for discrete, uniformly distributed sources.

math.PR↗

Random polynomials having few or no real zeros

Consider a polynomial of large degree n whose coefficients are independent, identically distributed, nondegenerate random variables having zero mean and finite moments of all orders. We show that such a polynomial has exactly k real zeros with probability n^{-b+o(1)}$ as n --> infinity through integers of the same parity as the fixed integer k >= 0. In particular, the probability that a random polynomial of large even degree n has no real zeros is n^{-b+o(1)}. The finite, positive constant b is characterized via the centered, stationary Gaussian process of correlation function sech(t/2). The value of b depends neither on k nor upon the specific law of the coefficients. Under an extra smoothness assumption about the law of the coefficients, with probability n^{-b+o(1)} one may specify also the approximate locations of the k zeros on the real line. The constant b is replaced by b/2 in case the i.i.d. coefficients have a nonzero mean.

math.PR↗