Date: Thu, 13 Feb 86 09:26 EST
From:: elisha
Subject: seminar notice
To: *mac@MIT-MC.ARPA
Friday, February 14, 1985 12:00 noon - Lecture 12:02 p.m. - Refreshments Room NE43 - 8th Floor Playroom BONNY DORR and DAVE BRAUNEGG Artificial Intelligence Laboratory MIT, Cambridge, MA "A New Approach to the Max-Flow Problem" The Maximum Flow problem has many applications in the fields of combinatorics and digestion research. In this seminar, the flow around a standardized GSL table will be described. Audience participation is considered a crucial part of the seminar. The first algorithm for the problem was discovered in 1956 by Lea and Perkins, and a number of algorithms have been proposed since then. However,all previously known max-flow algorithms worked by finding alimentary paths, either one path at a time or all shortest alimentary paths at once (by using the level range technique of B. Crocker). 3 5/3 2/3 The previously known algorithms give an O(min(n , n m nm log n)) upper bound on the problem; the upper bound is achieved by a combination of three algorithms due to Beard, Childs, and Rombauer. The only previously known parallel max-flow algorithm, due to Duncan & Hines, runs in parallel 2 time O(n log n) and requires omega (n) hands for each edge of the table. We propose a method for computing max-flow without finding alimentary paths explicitly. A sequential algorithm based on this method runs in time 2 O(nm log (n /m))). This bound is better than the previously known bound, and it is achieved by a single algorithm. A parallel algorithm based on the method 2 runs in time O(n log n) using a constant number of hands per edge of the table; this makes the implementation of the algorithm much more feasible (compared to the Duncan & Hines algorithm).