Modern constraint solving by propagation

Constraint Programming (CP) provides a generic method of solving problems from a wide range of fields, from industrial design to debugging. The major strength of Constraint Programming has been its ability to make use of many different algorithms, efficiently communicating. There are many different techniques in current use for solving large combinatorial problems, including SAT, SMT, ILP and Constraint Programming, each with their own own strengths and weaknesses. In this talk I will compare and contrast the state-of-the-art in constraint programming with each of these different areas. Modern constraint solvers have two major strengths: expressive input languages and the central reasoning algorithms they use, known as propagators. Propagators are a major feature of constraint solvers, and they allow a large range of algorithms, from regular expressions to flow networks, to be bought together in a single efficient framework. In recent years propagators have improved in a number of ways. Many new propagators implement large classes of constraints, such as those expressible by finite automata. Also new techniques, including the use of watched-literal like algorithms from SAT, have helped to greatly improve the performance of constraint solvers. The heavy dependance on propagation algorithms as a central technique has also caused problems and limitations for constraint programming. For example, it is only recently that learning has finally begun to be successful in constraint programming. This talk will explain the state of the art in constraint programming, and show how SAT, SMT and constraint programming are moving toward one unified framework.