Effective use of memory in linear space best first search

Classical best-first search algorithms like A* require space exponential in length of the solution path. To overcome this limitation algorithms like IDA* and RBFS regenerate parts of problem space trading space for time. These algorithms are probably too modest using only a tiny amount of memory today’s computers can offer. We present a novel, RBFS based algorithm which uses all available memory. This is achieved by keeping the generated nodes in memory and pruning some of them away when the memory runs out. We describe three pruning methods and experiments confirming the advantage of our approach.