Title :
Definability on a random 3-CNF formula
Author :
Atserias, Albert
Author_Institution :
Univ. Politecnica de Catalunya, Barcelona, Spain
Abstract :
We consider the question of certifying unsatisfiability of random 3-CNF formulas. At which densities can we hope for a simple sufficient condition for unsatisfiability that holds almost surely? We study this question from the point of view of definability theory. The main result is that first-order logic cannot express any sufficient condition that holds almost surely on random 3-CNF formulas with n2-α clauses, for any irrational positive number α. In contrast, it can when the number of clauses is n2+α, for any positive α. As an intermediate step, our proof exploits the planted distribution for 3-CNF formulas in a new technical way. Moreover, the proof requires us to extend the methods of Shelah and Spencer for proving the zero-one law for sparse random graphs to arbitrary relational languages.
Keywords :
computational complexity; formal logic; graph theory; computational complexity; first-order logic; random 3-CNF formula; relational language; sparse random graph; Layout; Linear programming; Logic; Optimization methods; Physics; Polynomials; Sufficient conditions; Testing;
Conference_Titel :
Logic in Computer Science, 2005. LICS 2005. Proceedings. 20th Annual IEEE Symposium on
Print_ISBN :
0-7695-2266-1
DOI :
10.1109/LICS.2005.14