Soroush Ebadian


A Simple Randomized Algorithm for All Nearest Neighbors


Given a set P of n points in the plane, the all nearest neighbors problem asks for finding the closest point in P for each point in the set. The following folklore algorithm is used for the problem in practice: pick a line in a random direction, project all points onto the line, and then search for the nearest neighbor of each point in a small vicinity of that point on the line. It is widely believed that the expected number of points needed to be checked by the algorithm in the vicinity of each point is O( √ n) on average. We confirm this conjecture in affirmative by providing a careful analysis on the expected number of comparisons made by the algorithm. We also present a matching lower bound, showing that our analysis is essentially tight.