Chemical Reaction Networks and Stochastic Local Search

Abstract

Stochastic local search can be an effective method for solving a wide variety of optimization and constraint satisfaction problems. Here I show that some stochastic local search algorithms map naturally to stochastic chemical reaction networks. This connection highlights new ways in which stochasticity in chemical reaction networks can be used for search and thus for finding solutions to problems. The central example is a chemical reaction network construction for solving Boolean formula satisfiability problems. Using an efficient general-purpose stochastic chemical reaction network simulator, I show that direct simulation of the networks proposed here can be more efficient, in wall-clock time, than a somewhat outdated but industrial-strength commercial satisfiability solver. While not of use for practical computing, the constructions emphasize that exploiting the stochasticity inherent in chemical reaction network dynamics is not inherently inefficient – and indeed I propose that stochastic local search could be an important aspect of biological computation and should be exploited when engineering future artificial cells.