Removing more values than Max restricted path consistency for the same cost

Filtering techniques are essential to e ciently look for a so lution in a constraint network CN They remove some local inconsisten cies and so reduce the search space However a given local consistency has to be not too expensive if we want to use it to e ciently prune the search tree during search Hence for a long time it has been considered that the best choice is the limited local consistency achieved by forward checking However more recent works show that main taining arc consistency which is a more pruningful local consistency during search outperforms forward checking on hard and large constraint networks It is very likely that maintaining an even more pruningful local consistency may pay o on very hard problems A comparison of the local consistencies more pruningful than arc consistency that can be used on large CNs has been done in The conclusion of this work is that Max restricted path consistency Max RPC is one of the most promising local consistencies It deletes far more values than arc consistency with reasonable cpu time requirements For all the local consistencies more pruningful than Max RPC there exists CNs on which achieving them is really prohibitive In this paper we show that Max RPC is not the limit we have to do not exceed to guarantee a reasonable cpu time We pro pose a new ltering algorithm called Max RPCEn which prunes more values than an algorithm achieving Max RPC It requires less constraint checks than Max RPC the Max RPC algorithm proposed in and has almost the same cpu time requirements