New Perspectives on Semiring Applications to Dynamic Programming

TL;DR

This paper develops a general semiring-based framework for dynamic programming that extends classical problems to include cost and counting variants, introducing a new algebraic operation and demonstrating fixed parameter tractability.

Contribution

It introduces a semiring extension approach for dynamic programming problems, allowing cost and solution counting, and proposes a novel $ riangle$-product operation for solution enumeration.

Findings

Framework applies to NP-hard problems like Connected-Dominating-Set and CSPs.

Proves fixed parameter tractability with respect to clique-width and tree-width.

Enables counting solutions of minimal costs, an overlooked problem.

Abstract

Semiring algebras have been shown to provide a suitable language to formalize many noteworthy combinatorial problems. For instance, the Shortest-Path problem can be seen as a special case of the Algebraic-Path problem when applied to the tropical semiring. The application of semirings typically makes it possible to solve extended problems without increasing the computational complexity. In this article we further exploit the idea of using semiring algebras to address and tackle several extensions of classical computational problems by dynamic programming. We consider a general approach which allows us to define a semiring extension of any problem with a reasonable notion of a certificate (e.g., an NP problem). This allows us to consider cost variants of these combinatorial problems, as well as their counting extensions where the goal is to determine how many solutions a given problem admits. The approach makes no particular assumptions (such as idempotence) on the semiring structure. We also propose a new associative algebraic operation on semirings, called $\Delta$-product, which enables our dynamic programming algorithms to count the number of solutions of minimal costs. We illustrate the advantages of our framework on two well-known but computationally very different NP-hard problems, namely, Connected-Dominating-Set problems and finite-domain Constraint Satisfaction Problems (CSPs). In particular, we prove fixed parameter tractability (FPT) with respect to clique-width and tree-width of the input. This also allows us to count solutions of minimal cost, which is an overlooked problem in the literature.