Similarity and bisimilarity notions appropriate for characterizing indistinguishability in fragments of the calculus of relations

Motivated by applications in databases, this paper considers various fragments of the calculus of binary relations. The fragments are obtained by leaving out, or keeping in, some of the standard operators, along with some derived operators such as set difference, projection, coprojection, and residuation. For each considered fragment, a characterization is obtained for when two given binary relational structures are indistinguishable by expressions in that fragment. The characterizations are based on appropriately adapted notions of simulation and bisimulation.

💡 Research Summary

The paper investigates the expressive power of various fragments of the calculus of binary relations, focusing on the problem of indistinguishability: given two finite binary relational structures, can they be distinguished by any expression built from a selected set of relational operators? The authors systematically define a large family of fragments by starting from a core fragment C that contains the constants empty (0) and identity (1₀) together with union, intersection, and composition, and then optionally adding any subset of the derived operators set‑difference (‑), projection (π₁, π₂), co‑projection (¯π₁, ¯π₂), left and right residuals (/ and ), the universal constant 1, the diversity constant 0₀, and complement (c). Each fragment is denoted C(F) where F is the set of added operators/constants.

A central technical device is the notion of degree of an expression, which measures the maximal nesting depth of composition, projection, co‑projection, and residual operators. Operators ∪, ∩, and set‑difference do not increase degree, while ◦, /, , π, and ¯π each add one. For a fragment F and a natural number k, the set C(F)ᵏ consists of all expressions in that fragment whose degree does not exceed k. This mirrors the quantifier rank in first‑order logic and provides a natural bound for finite‑round games.

The authors then introduce k‑bounded simulations and k‑bounded bisimulations tailored to the presence or absence of set‑difference. When set‑difference is available, the appropriate notion is a symmetric bisimulation: a relation R between the node sets of two structures that satisfies forward and backward conditions for all operators of the fragment, with the degree bound k ensuring that only paths of length ≤ k are examined. When set‑difference is omitted, only a one‑directional simulation is needed, reflecting the fact that without set‑difference the language cannot express negation of a relation.

A key contribution is the construction of characteristic expressions χₖ(G) for any finite structure G and any degree bound k. χₖ(G) is built recursively from the relations of G using the operators of the fragment, and it captures exactly the information that cannot be distinguished by any expression of degree ≤ k. The authors prove that two structures G₁ and G₂ satisfy χₖ(G₁) = χₖ(G₂) iff there exists a k‑bounded (bi)simulation linking them. Consequently, indistinguishability with respect to the whole fragment (i.e., equality of the results of all expressions) is equivalent to the existence of a k‑bounded (bi)simulation for some sufficiently large k. This yields a Hennessy–Milner‑type theorem for each fragment: for finite structures, logical equivalence in the fragment coincides with (bi)simulation equivalence.

The paper also analyses the role of the constant 1. Even when 1 is not explicitly present, it can be expressed as 0/0 or 0\0 using the residuals; however, these definitions have degree 1, whereas the literal constant 1 has degree 0. This distinction leads to a nuanced classification of fragments: those where 1 is present at degree 0, those where it is only expressible at degree 1, and those where it is completely absent. This classification influences the definition of (bi)simulation and the construction of characteristic expressions.

From a computational standpoint, the authors show that for all fragments considered (which always include intersection), indistinguishability can be decided in polynomial time. The algorithm proceeds by computing the maximal k‑bounded (bi)simulation relation, which can be done by a fixed‑point iteration reminiscent of standard bisimulation checking. In contrast, if one were to consider the extreme fragment consisting solely of composition (without intersection, union, or any other operator), the problem collapses to automaton equivalence and becomes PSPACE‑complete. Thus the presence of intersection (and consequently conjunction in the corresponding modal view) is crucial for the tractability results.

Beyond the theoretical contributions, the paper motivates practical applications in database query processing and structural indexing. In many modern data management scenarios—Web graphs, Linked Data, RDF stores, and XML trees—queries are naturally expressed as relational algebra expressions over binary relations. If a query uses only a certain fragment of the calculus, then data blocks that are indistinguishable under that fragment can be grouped together, enabling pre‑computation of indexes, cache sharing, and query‑plan optimizations. The (bi)simulation checks provided by the paper thus become a foundational tool for such indexing schemes.

In summary, the paper delivers a comprehensive framework for characterizing indistinguishability across a wide spectrum of relational‑algebra fragments. By introducing degree‑bounded (bi)simulations, constructing characteristic expressions, and proving Hennessy–Milner‑type theorems, it bridges the gap between logical expressiveness and algorithmic verification. The results are both theoretically elegant—extending modal‑logic bisimulation techniques to a richer algebraic setting—and practically relevant for optimizing relational queries on graph‑structured data. Future work may explore fragments without intersection, infinite structures, or dynamic settings where (bi)simulation must be maintained under updates.