Dsat: A Native SAT Solver for Discrete Logic

Researchers have introduced Dsat, a novel native SAT solver designed specifically for discrete logic, addressing limitations in current symbolic reasoning techniques. Traditionally, applications involving discrete variables, such as probabilistic reasoning, planning, and explainable AI, rely on binarizing these variables into Boolean formats to utilize existing SAT solvers. However, this approach often encounters significant computational and semantical challenges. The new Dsat solver extends Boolean logic directly, allowing variables to take arbitrary values while maintaining a design similar to standard Boolean SAT solvers, incorporating features like unit resolution and clause learning that operate natively on discrete variables. Empirical comparisons demonstrate the merits of Dsat against Constraint Satisfaction Problem (CSP) solvers applied to discrete CNFs, Boolean SAT solvers using binarized CNFs, and various hybrid solvers. This development aims to enhance efficiency and accuracy in handling discrete variables within artificial intelligence and logical reasoning frameworks, offering a more direct and potentially superior alternative to traditional binarization methods.