Derivatives on Graphs for the Positive Calculus of Relations with Transitive Closure

We prove that the equational theory of the positive calculus of relations with transitive closure (PCoR*) is EXPSPACE-complete. Here, PCoR* terms consist of the following standard operators on binary relations: identity, empty, universality, union, intersection, composition, converse, and reflexive transitive closure (so, PCoR* terms subsume Kleene algebra and allegory terms as fragments). Additionally, we show that the equational theory of PCoR* extended with tests and nominals (in hybrid logic) is still EXPSPACE-complete; moreover, it is PSPACE-complete for its intersection-free fragment. To this end, we design derivatives on graphs by extending derivatives on words for regular expressions. The derivatives give a finite automata construction on path decompositions, like those on words. Because the equational theory has a linearly bounded pathwidth model property, we can decide the equational theory of PCoR* using these automata.

💡 Research Summary

The paper establishes that the equational theory of the Positive Calculus of Relations with Transitive Closure (PCoR*) is EXPSPACE‑complete, and it extends this result to the version enriched with tests (as in Kleene Algebra with Tests) and nominals (from hybrid logic). PCoR* is a relational algebraic system that includes the constants identity (1), empty (0), universal (⊤), and the binary operators composition (;), union (+), intersection (∩), together with converse (⌣) and reflexive transitive closure (*). Because it subsumes both Kleene algebra and allegories, its equational theory is strictly more expressive than the regular‑expression equivalence problem (PSPACE‑complete) and the graph‑homomorphism problem (decidable but with non‑elementary complexity in naïve approaches).

The authors’ main technical contribution is a novel “derivative on graphs” construction that generalizes Brzozowski’s and Antimirov’s derivatives for regular expressions. They first introduce labeled Kleene lattice (KL) terms, which are KL expressions equipped with multiple input and output ports (interfaces). For a given set of interface labels L, the graph derivative computes a left‑quotient of the language of runs, analogous to the word‑level left‑quotient. Crucially, the derivative operation is compositional: it can be applied locally to each component of a graph’s path decomposition and then recombined using a gluing operator (⊙). This compositionality is formalized in Theorem 5.5, which shows that the derivative of a large graph can be obtained from the derivatives of its smaller “pre‑glued” bags.

The paper proves two structural properties of PCoR* terms: every term has treewidth at most 2 (Proposition 2.8) and, more importantly for the algorithm, a linear‑bounded pathwidth model property (Proposition 2.9). These properties guarantee that any graph representing a PCoR* term admits a path decomposition whose bag size grows linearly with the term size. Using the derivative construction on such a decomposition yields a finite automaton whose number of states is exponential in the product of the pathwidth and the term size, i.e., 2^{O(k·|t|)} where k is the pathwidth bound.

To decide an equation t = s, the authors translate each side into a two‑way alternating finite automaton (2AFA) via the derivative construction, then intersect the two automata, and finally convert the result into a nondeterministic finite automaton (NFA). The emptiness test for this NFA can be performed in space exponential in the term size, giving an EXPSPACE upper bound. For the lower bound, they give a many‑one reduction from the universality problem for regular expressions with intersection, which is known to be EXPSPACE‑complete. The reduction encodes the intersection structure and the Kleene star into PCoR* terms, showing that deciding whether a PCoR* equation holds is at least as hard as the EXPSPACE‑complete problem.

The framework is robust enough to incorporate additional operators. Tests (boolean predicates) are encoded as special labeled subgraphs, while nominals (named vertices) are represented by unique interface labels. Both extensions fit seamlessly into the derivative‑based automata construction, preserving the EXPSPACE upper bound. Moreover, when the intersection operator is absent (or its “intersection width” is fixed), the construction yields a PSPACE‑complete decision procedure, matching the known complexity for Kleene algebra without intersection.

In comparison with prior work, Brunet and Pous used Petri automata to show EXPSPACE‑completeness for Kleene lattices with transitive closure, but they left the full PCoR* case open. This paper fills that gap by providing a derivative‑based method that works directly on graphs, leading to a conceptually simpler and more implementable algorithm. The authors also correct an earlier mistake in the decomposition lemmas of their conference version by switching from one‑way to two‑way alternating automata.

The paper is organized as follows: Section 2 introduces the syntax of PCoR* terms, graphs with bi‑interfaces, and the notions of pathwidth and treewidth. Section 3 extends KL terms with interface labels. Section 4 defines graph derivatives for labeled KL terms. Section 5 builds the automata from derivatives and proves EXPSPACE‑completeness for KL terms. Section 6 shows how to encode tests, nominals, and the universal constant ⊤, extending the result to full PCoR*. Section 7 contains the technical proof of the decomposition theorem for derivatives. Section 8 discusses related work and future directions.

Overall, the paper delivers a decisive complexity classification for the equational theory of PCoR*, introduces a powerful derivative‑on‑graphs technique, and demonstrates its applicability to richer logics with tests and nominals, thereby advancing the understanding of relational algebras, automata theory, and logical model theory.