arXiv subjects
Publications and source records attributed to Roberto Imbuzeiro Oliveira.
We give a new, elementary proof of a key inequality used by Rudelson in the derivation of his well-known bound for random sums of rank-one operators. Our approach is based on Ahlswede and Winter's technique for proving operator Chernoff bounds. We also prove a concentration inequality for sums of random matrices of rank one with explicit constants.
Consider any random graph model where potential edges appear independently, with possibly different probabilities, and assume that the minimum expected degree is omega(ln n). We prove that the adjacency matrix and the Laplacian of that random graph are concentrated around the corresponding matrices of the weighted graph whose edge weights are the probabilities in the random model. While this may seem surprising, we will see that this matrix concentration phenomenon is a generalization of known results about the Erös-Rényi model. In particular, we will argue that matrix concentration is implicit the theory of quasi-random graph properties. We present two main applications of the main result. In bond percolation over a graph G, we show that the Laplacian of the random subgraph is typically very close to the Laplacian of G. As a corollary, we improve upon a bound for the spectral gap due to Chung and Horn that was derived via much more complicated methods. In inhomogeneous random graphs, there are points X_1,...,X_n uniformly distributed on the interval [0,1] and each pair is connected with probability p kappa(X_i,X_j). We show that if \ln n/n<< p<< 1 and kappa is bounded, then the adjacency matrix of the random graph is close to an integral operator defined in terms of kappa. Our main proof tool is a new concentration inequality for matrix martingales that generalizes Freedman's inequality for the standard scalar setting.
We study random k-lifts of large, but otherwise arbitrary graphs G. We prove that, with high probability, all eigenvalues of the adjacency matrix of the lift that are not eigenvalues of G are of the order (D ln (kn))^{1/2}, where D is the maximum degree of G. Similarly, and also with high probability, the "new" eigenvalues of the Laplacian of the lift are all in an interval of length (ln (nk)/d)^{1/2} around 1, where d is the minimum degree of G. We also prove that, from the point of view of Spectral Graph Theory, there is very little difference between a random k_1k_2 ... k_r-lift of a graph and a random k_1-lift of a random k_2-lift of ... of a random k_r-lift of the same graph. The main proof tool is a concentration inequality for sums of random matrices that was recently introduced by the author.
We consider Kac's random walk on $n$-dimensional rotation matrices, where each step is a random rotation in the plane generated by two randomly picked coordinates. We show that this process converges to the Haar measure on $\mathit{SO}(n)$ in the $L^2$ transportation cost (Wasserstein) metric in $O(n^2\ln n)$ steps. We also prove that our bound is at most a $O(\ln n)$ factor away from optimal. Previous bounds, due to Diaconis/Saloff-Coste and Pak/Sidenko, had extra powers of $n$ and held only for $L^1$ transportation cost. Our proof method includes a general result of independent interest, akin to the path coupling method of Bubley and Dyer. Suppose that $P$ is a Markov chain on a Polish length space $(M,d)$ and that for all $x,y\in M$ with $d(x,y)\ll1$ there is a coupling $(X,Y)$ of one step of $P$ from $x$ and $y$ (resp.) that contracts distances by a $(ξ+o(1))$ factor on average. Then the map $μ\mapstoμP$ is $ξ$-contracting in the transportation cost metric.
Consider a balls-in-bins process in which each new ball goes into a given bin with probability proportional to f(n), where n is the number of balls currently in the bin and f is a fixed positive function. It is known that these so-called balls-in-bins processes with feedback have a monopolistic regime: if f(x)=x^p for p>1, then there is a finite time after which one of the bins will receive all incoming balls. Our goal in this paper is to quantify the onset of monopoly. We show that the initial number of balls is large and bin 1 starts with a fraction α>1/2 of the balls, then with very high probability its share of the total number of balls never decreases significantly below α. Thus a bin that obtains more than half of the balls at a "large time" will most likely preserve its position of leadership. However, the probability that the winning bin has a non-negligible advantage after n balls are in the system is ~ C n^{1-p}, and the number of balls in the losing bin has a power-law tail. Similar results also hold for more general functions f.