Abstract
The purpose of this paper is to describe a sorting network of size $O(n n)$ and depth $O(n)$. A natural way of sorting is through consecutive halvings: determine the upper and lower halves of the set, proceed similarly within the halves, and so on. Unfortunately, while one can halve a set using only $O(n)$ comparisons, this cannot be done in less than log $n$ (parallel) time, and it is known that a halving network needs $1/2 n n$ comparisons. It is possible, however, to construct a network of $O(n)$ comparisons which halves in constant time with high accuracy. This procedure ($\epsilon$-halving) and a derived procedure ($\epsilon$-nearsort) are described below, and our sorting network will be centered around these elementary steps.
Users
Please
log in to take part in the discussion (add own reviews or comments).