Abstract. We demonstrate how to solve constraint satisfaction problems (CSPs) with DNA computing. Assuming that DNA operations can be faulty, we estimate error probability of our algorithm. We show that for any k-CSP, there is a polynomial-time DNA algorithm with bounded probability of error. Thus, k-CSPs belong to a DNA analogue of BPP. 1

One of the most serious problems in DNA computing is that basic DNA operations are faulty. Many DNA computation models use operations based on annealing and magnetic-beads separation which sometimes produce undesirable results. Some models use reliable operations only but their computational power is too weak. The purpose of this paper is to nd a good trade-o between the robustness of DNA operations and their computational power. We present a robust DNA computation model that can solve computationally hard problems. We prove that (i) this model can solve PSPACE- complete problems, and (ii) any computational problem that can be solved with this model is in PSPACE.