Interaction and observation, categorically

This paper proposes to use dialgebras to specify the semantics of interactive systems in a natural way. Dialgebras are a conservative extension of coalgebras. In this categorical model, from the point of view that we provide, the notions of observation and interaction are separate features. This is useful, for example, in the specification of process equivalences, which are obtained as kernels of the homomorphisms of dialgebras. As an example we present the asynchronous semantics of the CCS.

💡 Research Summary

The paper introduces dialgebras as a categorical framework that cleanly separates interaction (inputs) from observation (outputs) in the semantics of interactive systems. While algebras model operations on data and coalgebras model observable behavior, a dialgebra combines both by providing a single morphism f : F X → B X, where F generates “experiments” (the possible inputs an external observer may perform) and B collects the resulting observations (outputs, internal τ‑steps, etc.). This structure generalizes Mealy machines (I × X → O × X) and can be seen as a conservative extension of both algebras (when B = Id) and coalgebras (when F = Id).

A morphism between dialgebras (h : X → Y) must satisfy g ∘ F h = B h ∘ f, ensuring that the image of an experiment under h produces the same observation as first observing then mapping. The kernel of any such morphism yields an equivalence relation on X, called dialgebraic bisimilarity. This relation captures the intuitive idea that two states are equivalent when they react identically to every possible experiment and the resulting successor states are themselves equivalent.

The authors apply this theory to the asynchronous CCS (Calculus of Communicating Systems). In the standard labelled transition system (LTS) semantics, inputs, outputs, and internal τ‑actions are all treated uniformly as labels, which obscures the fact that an external observer cannot directly see input actions in an asynchronous setting. By choosing F to encode “sending a message on a channel” (the experiment) and B to encode the observable outputs and τ‑steps, the dialgebra f maps each state to the set of possible observations resulting from each experiment.

The main technical result (Theorem 1) shows that the dialgebraic bisimilarity coincides exactly with strong asynchronous bisimilarity, a well‑studied equivalence for CCS where input actions are invisible to the observer. The proof proceeds by constructing a suitable dialgebra homomorphism that collapses states related by asynchronous bisimilarity, and then demonstrating that its kernel yields the same relation. Consequently, the dialgebraic approach provides a clean categorical characterisation of asynchronous equivalence without having to modify the underlying LTS.

Beyond this example, the paper discusses how dialgebras differ from bialgebras (which pair an algebra and a coalgebra on the same carrier). In a bialgebra, the algebraic part builds terms while the coalgebraic part observes them; a dialgebra, by contrast, directly returns observations from experiments, merging the two perspectives into a single operation. This makes dialgebras especially suitable for scenarios where the syntax of experiments (e.g., pressing buttons on a vending machine, sending packets in a network protocol) is unrelated to the internal syntax of the system being observed.

Related work is surveyed: dialgebras have previously appeared in the specification of data types, but their use for programming‑language semantics and process equivalences is novel. The authors also hint at future extensions: replacing the polynomial functor F with more sophisticated functors (e.g., probability distributions, monadic effects) could model probabilistic or quantum systems, and the categorical machinery would still provide a kernel‑based notion of equivalence.

In summary, the paper proposes a mathematically elegant and conceptually clear method for modelling interactive computation: by treating experiments and observations as separate functorial components, dialgebras enable a natural definition of behavioral equivalence that respects the asymmetry between input and output inherent in many concurrent and asynchronous systems. This contributes both to the theory of process calculi and to the broader categorical semantics of programming languages.