How to solve larger CP instances with constraint-based local search?

Combinatorial optimization problems are ubiquitous in numerous practical applications. Yet most of them are challenging, both from computational complexity and programming standpoints. Local search is one of the main approaches to address these problems. However, it often requires sophisticated incremental algorithms and data structures, and considerable experimentation. This paper proposes a constraint-based, object-oriented, architecture to reduce the development time of local search algorithms significantly. The architecture consists of declarative and search components. The declarative component includes invariants, which maintain complex expressions incrementally, and differentiable objects, which maintain properties that can be queried to evaluate the effect of local moves. Differentiable objects are high-level modeling concepts, such as constraints and functions, that capture combinatorial substructures arising in many applications. The search component supports various abstractions to specify heuristics and meta-heuristics. We illustrate the architecture with the language Comet and several applications, such as car sequencing and the progressive party problem. The applications indicate that the architecture allows for very high-level modeling of local search algorithms, while preserving excellent performance.