Linear datalog and bounded path duality of relational structures
We study which constraint satisfaction problems (CSPs) are solvable in NL. In particular, we identify a general condition called bounded path duality, that explains all the families of CSPs previously known to be in NL. Bounded path duality captures the class of constraint satisfaction problems that can be solved by linear Datalog programs, i.e., Datalog programs with at most one IDB in the body of each rule. We obtain several alternative characterizations of bounded path duality.