arXiv++ Combinatorics

Let $X$ be a set of $n$ points of norm at most $1$ in the Euclidean space $R^k$, and suppose $\varepsilon>0$. An $\varepsilon$-distance sketch for $X$ is a data structure that, given any two points of $X$ enables one to recover the square of the (Euclidean) distance between them up to an {\em additive} error of $\varepsilon$. Let $f(n,k,\varepsilon)$ denote the minimum possible number of bits of such a sketch. Here we determine $f(n,k,\varepsilon)$ up to a constant factor for all $n \geq k \geq 1$ and all $\varepsilon \geq \frac{1}{n^{0.49}}$. Our proof is algorithmic, and provides an efficient algorithm for computing a sketch of size $O(f(n,k,\varepsilon)/n)$ for each point, so that the square of the distance between any two points can be computed from their sketches up to an additive error of $\varepsilon$ in time linear in the length of the sketches. We also discuss the case of smaller $\varepsilon>2/\sqrt n$ and obtain some new results about dimension reduction in this range. In particular, we show that for any such $\varepsilon$ and any $k \leq t=\frac{\log (2+\varepsilon^2 n)}{\varepsilon^2}$ there are configurations of $n$ points in $R^k$ that cannot be embedded in $R^{\ell}$ for $\ell < ck$ with $c$ a small absolute positive constant, without distorting some inner products (and distances) by more than $\varepsilon$. On the positive side, we provide a randomized polynomial time algorithm for a bipartite variant of the Johnson-Lindenstrauss lemma in which scalar products are approximated up to an additive error of at most $\varepsilon$. This variant allows a reduction of the dimension down to $O(\frac{\log (2+\varepsilon^2 n)}{\varepsilon^2})$, where $n$ is the number of points.

We study a natural discrete Bochner-type inequality on graphs, and explore its merit as a notion of curvature in discrete spaces. An appealing feature of this discrete version seems to be that it is fairly straightforward to compute this notion of curvature parameter for several specific graphs of interest - particularly, abelian groups, slices of the hypercube, and the symmetric group under various sets of generators. We further develop this notion by deriving Buser-type inequalities (a la Ledoux), relating functional and isoperimetric constants associated with a graph. Our derivations provide a tight bound on the Cheeger constant (i.e., the edge-isoperimetric constant) in terms of the spectral gap, for graphs with nonnegative curvature, particularly, the class of abelian Cayley graphs - a result of independent interest.

Let $G_{\infty}=(C_m^d)_{\infty}$ denote the graph whose set of vertices is $\{1,..., m\}^d$, where two distinct vertices are adjacent iff they are either equal or adjacent in $C_m$ in each coordinate. Let $G_{1}=(C_m^d)_1$ denote the graph on the same set of vertices in which two vertices are adjacent iff they are adjacent in one coordinate in $C_m$ and equal in all others. Both graphs can be viewed as graphs of the $d$-dimensional torus. We prove that one can delete $O(\sqrt d m^{d-1})$ vertices of $G_1$ so that no topologically nontrivial cycles remain. This improves an $O(d^{\log_2 (3/2)}m^{d-1})$ estimate of Bollobás, Kindler, Leader and O'Donnell. We also give a short proof of a result implicit in a recent paper of Raz: one can delete an $O(\sqrt d/m)$ fraction of the edges of $G_{\infty}$ so that no topologically nontrivial cycles remain in this graph. Our technique also yields a short proof of a recent result of Kindler, O'Donnell, Rao and Wigderson; there is a subset of the continuous $d$-dimensional torus of surface area $O(\sqrt d)$ that intersects all nontrivial cycles. All proofs are based on the same general idea: the consideration of random shifts of a body with small boundary and no- nontrivial cycles, whose existence is proved by applying the isoperimetric inequality of Cheeger or its vertex or edge discrete analogues.