Randomized greedy: new variants of some classic approximation algorithms

We consider the performance of two classic approximation algorithms which work by scanning the input and greedily constructing a solution. We investigate whether running these algorithms on a random permutation of the input can increase their performance ratio. We obtain the following results:

1. Johnson's approximation algorithm for MAX-SAT is one of the first approximation algorithms to be rigorously analyzed. It has been shown that the performance ratio of this algorithm is 2/3. We show that when executed on a random permutation of the variables, the performance ratio of this algorithm is improved to 2/3 + c for some c > 0 This resolves an open problem of Chen, Friesen and Zhang [JCSS 1999]. (See also the paper by Poloczek and Schnitger in these proceedings for related results on this algorithm and its variants).

2. Motivated by the above improvement, we consider the performance of the greedy algorithm for MAX-CUT whose performance ratio is 1/2. Our hope was that running the greedy algorithm on a random permutation of the vertices would result in a 1/2 + c approximation algorithm. However, it turns out that in this case the performance of the algorithm remains 1/2. This resolves an open problem of Mathieu and Schudy [SODA 2008].