 |
Rice University
Department of Computer Science
presents
Xiaoxu Wang
Master of Science Thesis Defense
Abstract
Analysis on the Assignment Landscape of 3-SAT Problems
The complexity of random 3-SAT problems is one of the fundamental
problems in computer science theory. 3-SAT problems shift from
satisfiable to unsatisfiable at the crossover point, and show high
complexity around it. All assignments in a random SAT instance
constitute a landscape. Using the CUDD package and Image Computation
method, we conduct some experiments on the assignment landscape and
the solution set. The crossover point can be obtained from the
analytical computation of the spin model in statistical physics. This
thesis confirms the clustering behavior of the solution set, another
important conclusion drawn from the statistical model. Our
experimental results indicate that the solution set does break into
several clusters when the density falls into a specific range. As the
density grows, the number of clusters increases and peak at density
3.2. As the density keeps increasing, the number of clusters in the
solution set drops and non-solution basins in the assignment landscape
emerge. These non-solution basins can cause high complexity in
search-based SAT solvers at the densities right below the crossover
point.
Friday, May 14 at 1:00 p.m. in DH 2014
--- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- |
|
| |