Reachability in Restricted Chemical Reaction Networks

TL;DR

This paper investigates the computational complexity of reachability problems in various restricted models of Chemical Reaction Networks, providing complexity classifications and algorithms for different subclasses.

Contribution

It characterizes the complexity of reachability in restricted CRNs, including polynomial-time solvability and NP-completeness results, advancing understanding of computational limits in molecular systems.

Findings

Reachability is PSPACE-complete with bimolecular reactions in certain restrictions.

Polynomial-time algorithms exist for feed-forward CRNs without void/autogenesis rules.

Adding void/autogenesis rules or multiple sources/consumers makes reachability NP-complete.

Abstract

The popularity of molecular computation has given rise to several models of abstraction, one of the more recent ones being Chemical Reaction Networks (CRNs). These are equivalent to other popular computational models, such as Vector Addition Systems and Petri-Nets, and restricted versions are equivalent to Population Protocols. This paper continues the work on core \emph{reachability} questions related to Chemical Reaction Networks; given two configurations, can one reach the other according to the system's rules? With no restrictions, reachability was recently shown to be Ackermann-complete, which resolved a decades-old problem. In this work, we fully characterize monotone reachability problems based on various restrictions such as the allowed rule size, the number of rules that may create a species ($k$-source), the number of rules that may consume a species ($k$-consuming), the volume, and whether the rules have an acyclic production order (\emph{feed-forward}). We show PSPACE-completeness of reachability with only bimolecular reactions in two-source and two-consuming rules. This proves hardness of reachability in a restricted form of Population Protocols. This is accomplished using new techniques within the motion planning framework. We give several important results for feed-forward CRNs, where rules are single-source or single-consuming. We show that reachability is solvable in polynomial time as long as the system does not contain special \emph{void} or \emph{autogenesis} rules. We then fully characterize all systems of this type and show that with void/autogenesis rules, or more than one source and one consuming, the problems become NP-complete. Finally, we show several interesting special cases of CRNs based on these restrictions or slight relaxations and note future significant open questions related to this taxonomy.