Artificial Conflict Sampling for Real Satisfiability Problems
James H. Davenport, Ali Kemal Uncu · 2022
We outline some preliminary ideas on a guided theory assignment of variables in a real (QF_NRA) satisfiability problem. One objective of this approach is to mix the topdown approach of cylindric algebraic decomposition and the bottom-up approach of partial theory assignments of modern SAT/SMT solvers. We use equational constraints and a single strict inequality at a time to artificially create conflicting variable assignment traces, which can later be used in conflict resolution to enrich the constraints of the original satisfiability problem.