Random walk and parallelism in local search

The use of randomness in local search is a widespread technique applied to improve algorithm performance. A property of noise strategies, i.e., strategies that make use of randomness, is that there is an optimal amount of randomness that enables the algorithm to achieve on optimal performance. The same property holds also for another technique introduced in local search, that is the parallel application of more than one local move. In this work, we investigate the outcome of the combination of both the techniques. We show that in the case of random 3-SAT instances a synergistic effect comes into play and the resulting technique outperforms the single ones.