Computational Geometry Seminar
Wednesday, March 28th, 2007, 16:10-18:00
Room 309
Schreiber Building
Let S be an n-point set in R^d, and let epsilon be a parameter between 0
and 1. A "weak epsilon-net" for S is another set of points N, such that,
for every subset S' of S of size at least epsilon*n, the convex hull of S'
containts at least one point of N. The problem is to construct weak
epsilon-nets that are as small as possible. For convenience, let r =
1/epsilon, so our bounds increase with r.
It is known that every planar set S has a weak 1/r-net of size O(r^2), and
every set S in R^d has a weak 1/r-net of size O(r^d * polylog(r)). For
certain special cases better bounds are known: If S is planar and in
convex position, then S has a weak 1/r-net of size O(r*polylog(r)). And if
S is a set of points along the "moment curve" in R^d, then S also has a
weak 1/r-net of size O(r*polylog(r)).
In this talk we present improved constructions for the above special
cases, achieving almost-linear bounds that involve alpha(r), the inverse
Ackermann function. Namely, we show that every planar set S in convex
position has a weak 1/r-net of size O(r * alpha(r)). And every point set S
along the moment curve in R^d has a weak 1/r-net of size O(r *
2^poly(alpha(r))), where the degree of the polynomial depends on d.
Our constructions result from a reduction to a problem which we call
"stabbing interval chains with j-tuples". We derive barely-superlinear
upper bounds for this problem, involving functions alpha_i(n) of the
inverse Ackermann hierarchy. We also derive almost-matching lower bounds.
Joint work with Noga Alon, Haim Kaplan, Micha Sharir, and Shakhar
Smorodinsky.