[comp.theory] Open problem??? Voronoi diag of sorted pts

eppstein@glacier.ics.uci.edu (David Eppstein) (11/27/90)

In article <17966@yunexus.YorkU.CA> s442215@nexus.YorkU.CA (Binhai Zhu) writes:
> If n points on the plane are already sorted according to their
> x-coordinates(not according to their y-coordinates).  Can you compute
> the Voronoi Diagram of these points in o(nlogn) time?

No.  If you could, you could find the nearest pair of points.  But this
is known to be impossible, by a reduction from the integer smallest gap
problem.  More specifically, suppose you have a collection of n
integers.  For each integer a_i, place a point at (i/n,a_i).  The
points end up sorted by x-coordinate, and the closest pair of points
gives the closest pair of integers.  But the closest integer problem has an
Omega(n log n) lower bound [A.C. Yao, FOCS 1989, 308-313].

The question is more interesting when the points are given in two sorted
orders, one per coordinate.  Then the problem for the usual L_2 metric
remains open, but for the L_1 (equivalently L_inf) metric the Voronoi
diagram can be computed in O(n log log n) time [Chew & Fortune, 13th
Symp. Math. Prog., Japan, 1988].
--
David Eppstein    Info & Computer Science, UC Irvine    eppstein@ics.uci.edu