Higher-Dimensional Automata

• Higher-dimensional automata (HDAs) are finite precubical sets with designated start and accept cells that model simultaneous events and true concurrency.
• They support modular design through interface-based gluing, enabling rational operations such as serial, parallel compositions, and Kleene-plus.
• HDAs provide a unified operational and algebraic framework for non-interleaving concurrent systems, linking automata theory with algebraic topology and category theory.

Higher-dimensional automata (HDAs) extend the classical theory of finite automata to explicitly model true concurrency—multiple events occurring simultaneously. HDAs are formalized as finite precubical sets (presheaves over labeled cubes) equipped with distinguished start and accept cells. They recognize languages of finite interval pomsets (partially ordered multisets), closed under subsumption (order-refinement), and support rational operations such as union, serial (gluing) composition, parallel composition, and Kleene-plus. HDAs provide an operational and algebraic foundation for analyzing non-interleaving concurrent systems, with applications to distributed systems, Petri nets, event structures, and beyond. The central result is a Kleene-type theorem: the class of languages recognized by finite HDAs coincides precisely with the rational closure under these operations, and the geometric structure underlying HDAs enables sophisticated constructions, including those defined by algebraic topology and categorical composition (Fahrenberg et al., 2022).

1. Formal Definition: Precubical Sets and HDAs

An HDA is a presheaf over the labeled precube category $\mathsf{Sq}$, whose objects are concurrency lists $U$ (totally ordered, labeled sets), and whose morphisms $d_{A,B}: V \hookrightarrow U$ encode injective order- and label-preserving embeddings, with the complement $U\setminus V = A \cup B$ partitioned into "not yet started" ($A$) and "already terminated" ($B$) events.

A precubical set $X$ is a functor $X: \mathsf{Sq}^\mathrm{op} \to \mathsf{Set}$, assigning to each $U$ the set $X[U]$ of cells corresponding to the active events $U$ in that cell. Face maps $\delta_{A,B}: X[U] \to X[V]$ reflect the activation or termination of events, with functorial compatibility across dimensions.

An HDA is a finite such precubical set, together with distinguished subsets of start and accept cells:

• $X_\perp$, $X^\top \subseteq \bigcup_U X[U]$, typically located in degree $U = \varnothing$.

Cells "carry" their active event conclist, and transitions correspond to starting or terminating events—modeled geometrically via cubes of higher dimension.

2. Interfaces and HDA Composition

Interfaces generalize HDAs to track active event sets across composition, essential for modular design and rational operations. HDAs with interfaces (iHDAs) equip each cell with source ($S$) and target ($T$) interfaces, recorded as subsets of active events. Gluing two HDAs $X$ and $Y$ along a shared interface $U$ (the set of events currently active) is executed by identifying those cells in both HDAs whose active-event conclist matches $U$.

Technically, gluing in the base category $\mathsf{Sq}$ lifts to a colimit in the presheaf category, producing the composite HDA:

• Cells in $X * Y$ are those in $X \sqcup Y$, with specified faces identified along $U$.

This yields rational serial composition. The use of interfaces enables complex constructions such as cylinders (for pushouts and "idle step" insertions) and spiders (for loop closure and star operators), inheriting key mechanisms from algebraic topology and categorical algebra (Fahrenberg et al., 2022).

3. Interval Pomsets, Subsumption, and Recognized Languages

The semantic domain of an HDA is the set of interval pomsets with interfaces (ipomsets). Formally,

• $P = (P, <, \preceq, S, T, \lambda)$, where $P$ is a finite event set; $<$ is a strict interval order (no $2+2$ suborder); $\preceq$ refines incomparability in $<$, giving a secondary event order; $S$ (source interface) is minimal under $<$, $T$ (target interface) maximal; $\lambda: P \to \Sigma$ labels events.

A morphism $f: P \to Q$ is a subsumption if $f$ is bijective, preserves labels and interfaces, reflects precedence, and preserves event order on incomparable events. The subsumption order $P \sqsubseteq Q$ captures increased concurrency (more causal relationships or more independence).

Languages recognized by HDAs are subsets $L \subseteq \{\text{interval ipomsets}\}$ that are subsumption-closed: if $P \sqsubseteq Q$ and $Q \in L$, then $P \in L$. The down-closure $$L^\downarrow = \{P \mid \exists Q \in L:\, P \sqsubseteq Q\}$$ is central in the rational expression semantics.

4. Rational Operations: Union, Gluing, Parallel, Kleene-Plus

Let $L, M$ be subsumption-closed languages of interval ipomsets. The rational operations are:

• Union: $L \cup M$.
• Serial Gluing: $$L * M = \{P * Q \mid P \in L, Q \in M, T_P = S_Q\}^\downarrow$$, where $P * Q$ glues $P$ and $Q$ along matching interfaces.
• Parallel Composition: $$L \parallel M = \{P \parallel Q \mid P \in L, Q \in M\}^\downarrow$$, with $P \parallel Q$ disjoint union, minimal cross-precedences, and ordering $P$ events before $Q$ under $\preceq$.
• Kleene Plus: $L^+ = \bigcup_{n \geq 1} L^n$, iterated serial composition. Kleene-star ($L^*$) is omitted due to the unbounded growth of dimension in non-trivial $\epsilon$-closures.

The rational closure of singleton pomset languages $\{\varepsilon\}, \{a\}, \{\bullet a\}$, etc., under these operators, matches exactly the class of languages recognized by finite HDAs—establishing a full algebraic correspondence (Fahrenberg et al., 2022).

5. Kleene Theorem for HDAs and Proof Outline

Theorem (Kleene):

A language $L$ of finite interval ipomsets is regular (i.e., $L = \operatorname{Lang}(X)$ for some finite HDA $X$) if and only if $L$ is rational (generated from singleton languages under union, serial/parallel composition, and Kleene-plus).

Proof Overview:

• Regular $\implies$ Rational: For HDA $X$ of dimension $n$, construct a classical finite automaton $G(X)$ over discrete ipomsets of size $\leq n$. Use the classical Kleene theorem to represent $L = \operatorname{Lang}(X)$ as a rational expression combining gluing, parallel, and iteration on the underlying automaton.
• Rational $\implies$ Regular: Each rational operation corresponds to a finite HDA construction: union is coproduct, parallel is tensor product, serial and iteration require gluing (via colimits in the presheaf category), cylinders, and spider constructions for loop closure.

Critical technical devices include:

• Cylinders: capture homotopical gluing with lifting properties (cofibration/fibration analogues).
• Spider Construction: splits/identifies accept and start interfaces for iteration.
• The inductive height of rational expressions and systematic application of these constructions yields the finite HDA realizing the language.

HDAs thus occupy the same place for concurrent interval pomsets as finite automata do for regular word languages (Fahrenberg et al., 2022).

6. HDAs as a Universal Model for Concurrency

HDAs strictly generalize classical automata and asynchronous automata: in dimension $1$ they reduce to ordinary finite automata; dimension $2$ captures asynchronous automata (commuting squares encoding independent processes). For higher dimension, HDAs capture autoconcurrency, durative events, and phenomena inaccessible to trace-based or interleaving models.

Every safe Petri net or event structure embeds into an HDA, and interval orders precisely encode system executions. The Kleene correspondence ensures that HDA-recognizable interval pomset languages coincide with the rational closure under the four operators.

Topological and categorical tools used in HDAs suggest new methodologies for concurrency theory:

• Cylinders and path-lifting: control insertion/collapse of silent steps and modular composition.
• (Co)fibration patterns: support compositional reasoning.

HDAs as presheaves fit into a broader category-theoretic context: coalgebraic automata, open maps, model categories, and factorization systems, enabling deep connections with other operational and logical frameworks.

7. Implications, Applications, and Research Directions

The geometric, algebraic, and categorical structure of HDAs allows:

• Direct modeling and minimization of concurrent systems through topological abstraction and cube collapses, preserving homotopy, trace category, and homology invariants (Kahl, 2015).
• Explicit representation of independence, concurrency, and causal ordering in system executions.
• Translation and analysis of Petri nets, event structures, shared-variable systems, and other concurrent formalisms into a unified operational and semantic framework.
• The algebraic-topological operations underlying HDA constructions lay the foundation for future Kleene-type correspondence in more sophisticated non-interleaving and topologically enriched concurrency models.

HDAs provide a comprehensive operational and algebraic theory of concurrency, encapsulating the essential behaviors of distributed systems, and exactly characterizing the class of rational, subsumption-closed interval pomset languages recognized by finite concurrent automata (Fahrenberg et al., 2022).