Tree Projections and Structural Decomposition Methods: The Power of Local Consistency and Larger Islands of Tractability

TL;DR

This paper investigates the power of local consistency enforcement in solving NP-hard problems like conjunctive query evaluation and constraint satisfaction, revealing new tractable classes via greedy tree projections.

Contribution

It provides a complete characterization of when local consistency suffices and introduces efficiently recognizable subclasses based on greedy tree projections that extend known tractable classes.

Findings

Enforcing local consistency solves certain classes of queries correctly.

Greedy tree projection subclasses are efficiently recognizable.

These subclasses are larger than previously known islands of tractability.

Abstract

Evaluating conjunctive queries and solving constraint satisfaction problems are fundamental problems in database theory and artificial intelligence, respectively. These problems are NP-hard, so that several research efforts have been made in the literature for identifying tractable classes, known as islands of tractability, as well as for devising clever heuristics for solving efficiently real-world instances. Many heuristic approaches are based on enforcing on the given instance a property called local consistency, where (in database terms) each tuple in every query atom matches at least one tuple in every other query atom. Interestingly, it turns out that, for many well-known classes of queries, such as for the acyclic queries, enforcing local consistency is even sufficient to solve the given instance correctly. However, the precise power of such a procedure was unclear, but for some very restricted cases. The paper provides full answers to the long-standing questions about the precise power of algorithms based on enforcing local consistency. The classes of instances where enforcing local consistency turns out to be a correct query-answering procedure are however not efficiently recognizable. In fact, the paper finally focuses on certain subclasses defined in terms of the novel notion of greedy tree projections. These latter classes are shown to be efficiently recognizable and strictly larger than most islands of tractability known so far, both in the general case of tree projections and for specific structural decomposition methods.