randomness is linear in space（随机性是线性空间）.pdfVIP

Randomness is Linear in Space Noam Nisan y David Zuckerman z Abstract We show that any randomized algorithm that runs in space S and time T and uses pol y S random bits can b e simulated using only O S random bits in space S and time T pol y S A deterministic simulation in space S follows Of indep endent interest is our main technical to ol a pro cedure which extracts randomness from a defective random source using a small additional numb er of truly random bits Intro duction The relative p ower of deterministic and randomized algorithms is a basic question in complexity theory Despite much eort very little is known In this pap er we consider this question when the complexity measured is space Is randomized space S stronger than deterministic space S While several nontrivial deterministic simulations of randomizedspace are known BNS N N this question is still completely op en No simulation of randomized space S is known which uses less than O S 2 deterministic space a simulation which can b e achieved by Savitchs theorem S Indeed from Savitchs pro of it follows that a language accepted by a randomized space S machine using R random bits is also accepted by a deterministic space S log R S machine There is only one result that improves this b ound for some R Namely Ajtai Komlos and Szemeredi showed that any randomized space S algorithm using only O S 2 log S random bits can b e simulated deterministically in space S AKS In this pap er we improve up on this result and give a deterministic simulation of algorithms using pol y S random bits What we obtain is a pseudorando