SAT Encoding for Set of At-Most-K Cardinality Constraints with Ladder Shape

| Source: arXiv AI

Tags: SAT Encoding, At-Most-K Constraints, Sequential Counter, Nurse Rostering, Anti-bandwidth Problem

A new SAT encoding for ladder-shaped At-Most-K constraints uses fewer variables and clauses and outperformed CP and MIP solvers on anti-bandwidth and nurse rostering problems.

Details

Ladder AMK constraints are sequences of at-most-K cardinality constraints that share central variables and differ at the ends. Encoding each one separately bloats SAT instances. The proposed encoding splits the sequence into blocks, encodes each block as a whole with a Sequential Counter, and adds clauses linking adjacent blocks to keep equivalence. Compared with a prior BDD-based encoding for Ladder AMO, it needs fewer variables and clauses, and the authors say it is the first such encoding for k>1. On single Ladder AMO and AMK constraints it reduces encoding size and solve time, and on the Anti-bandwidth and Nurse Rostering problems it beat alternative encodings as well as Constraint Programming and Mixed Integer Programming solvers.