No Better Ways to Generate Hard NP Instances than Picking Uniformly at Random

by Leonid A. Levint

Datum	Value	Source
TITLE	No Better Ways to Generate Hard NP Instances than Picking Uniformly at Random	SVM HeaderParse 0.2
AUTHOR NAME	Leonid A. Levint	SVM HeaderParse 0.2
AUTHOR AFFIL	Computer Science department; Boston University	SVM HeaderParse 0.2
AUTHOR ADDR	111 Cummington St. Boston	SVM HeaderParse 0.2
ABSTRACT	Distributed NP (DNP) problems are ones supplied with probability distributions of in-stances. We can consider their hardness for typical instances rather than just for the worst case (which may be extremely rare). Reductions between such problems must ap-proximately preserve the distributions. A number of papers show completeness of sev-eral natural DNP problems in the class of aJl DNP problems with P-time computable distributions. This approach has been criti-cized as too restrictive: hard instances may be generated with samplable but not com-putable in P-time distributions. There were doubts whether naturul problems (which all have simple distributions) may be complete for the class of all NP problems with sam-plable distributions. We show that every DNP problem com-plete for P-time computable distributions is also complete for all samplable distributions. This rather surprising observation makes the concept of average case NP completeness ro-	SVM HeaderParse 0.2
CITATIONS	12 found	ParsCit 1.0