Wheeler Deterministic Finite Automata

• Wheeler DFAs are a subclass of deterministic finite automata that impose a unique total order on states based on co-lexicographic rules, enabling structured and efficient processing.
• They support succinct indexing via BWT-like representations and use interval semantics to facilitate fast substring and path queries.
• Their design allows for near-linear determinization and linear-time minimization, offering practical benefits in compressed data structures and automata analysis.

to=arxiv_search 大发快三的 亚洲男人天堂 天天中彩票派奖 天天中彩票粤ness 大发快三有json-simplified ിക്കേണ്ട { "^^^^^^^^3query3 "^^^^^^^^3\3 DFA^^^^^^^^3\3 OR ^^^^^^^^3\3 automata^^^^^^^^3\3 "max_results": ^^^^3\3query3^^^^, "sort_by": "submittedDate" } Wheeler deterministic finite automata are deterministic finite automata whose states admit a total order compatible with both the co-lexicographic order of source-to-state path labels and the transition structure. In this setting, state order is not an auxiliary annotation but a structural invariant: it supports BWT-like representations, interval semantics for path queries, and efficient pattern matching on the substring closure of the recognized language. The regular languages admitting such an automaton are the Wheeler languages, a proper subclass of the regular languages and, in fact, a subclass of the star-free languages (&&&^^^^3query3^^^^&&&, &&&^^^^3\3^^^^&&&, &&&^^^^3 OR \3^^^^&&&).

^^^^3\3^^^^. Definition and order-theoretic foundations

Let PRESERVED_PLACEHOLDER_^^^^3query3^^^^ be a DFA, with PRESERVED_PLACEHOLDER_^^^^3\3^^^^ equipped with a total order PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^. The co-lexicographic order on $\Sigma^*$ compares words from right to left: $\epsilon \prec \alpha$ for every non-empty $\alpha$, and if $\alpha=\alpha' a$ and $\beta=\beta' b$, then

$$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$

For each state $u$, one writes

PRESERVED_PLACEHOLDER_^^^^3\3query3^^^^

the set of strings reaching PRESERVED_PLACEHOLDER_^^^^3\3\3^^^^ from the source.

A Wheeler DFA is a DFA for which there exists a total order PRESERVED_PLACEHOLDER_^^^^3\3 OR \3^^^^ on PRESERVED_PLACEHOLDER_^^^^3\33^^^^ such that PRESERVED_PLACEHOLDER_^^^^3\3^^^^4 for every PRESERVED_PLACEHOLDER_^^^^3\3^^^^5, and whenever PRESERVED_PLACEHOLDER_^^^^3\3^^^^6 and PRESERVED_PLACEHOLDER_^^^^3\3^^^^7, the following hold: if PRESERVED_PLACEHOLDER_^^^^3\3^^^^8, then PRESERVED_PLACEHOLDER_^^^^3\3^^^^9; if PRESERVED_PLACEHOLDER_^^^^3 OR \3query3^^^^, PRESERVED_PLACEHOLDER_^^^^3 OR \3\3^^^^, and PRESERVED_PLACEHOLDER_^^^^3 OR \3 OR \3^^^^, then PRESERVED_PLACEHOLDER_^^^^3 OR \33^^^^. An equivalent global formulation orders states by the sets PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^4: define

PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^5

A DFA is Wheeler if and only if PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^6 is a total order; in that case, it is the unique Wheeler order (&&&^^^^3query3^^^^&&&).

A direct consequence is input consistency: all incoming edges of a state carry the same label. This permits the definition of a well-defined incoming-label map PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^7 on states other than the source, and it is one of the reasons Wheeler automata align so closely with BWT-style indexing. The order is therefore simultaneously an order on states, an order on incoming contexts, and a constraint on admissible transition geometries (D'Agostino et al., 2021).

^^^^3 OR \3^^^^. Wheeler languages, convexity, and canonical structure

A Wheeler language is a regular language recognized by some Wheeler automaton. Wheeler NFAs and Wheeler DFAs have the same expressive power, so the language class can be defined using either model. The class is strictly smaller than the class of regular languages, and it sits inside the star-free languages (&&&^^^^3query3^^^^&&&, &&&^^^^3 OR \3^^^^&&&).

One of the central structural properties of Wheeler DFAs is path coherence. If the states are arranged in Wheeler order, then for every interval of states PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^8 and every word PRESERVED_PLACEHOLDER_^^^^3 OR \3^^^^9, the set of states reachable from some state in $\Sigma^*$^^^^3query3^^^^ by reading $\Sigma^*$^^^^3\3^^^^ is again an interval. This interval behavior is the automata-theoretic analogue of the suffix-array interval property behind the BWT and FM-index (D'Agostino et al., 2021).

The language-theoretic counterpart is a refinement of Myhill–Nerode equivalence. For a language $\Sigma^*$^^^^3 OR \3^^^^, the usual right-context equivalence is

$\Sigma^*$3

For Wheeler languages, one refines this to an input-consistent, co-lex convex equivalence $\Sigma^*$4: $\Sigma^*$5 when $\Sigma^*$6, the two words end with the same character, and every prefix lying between them in co-lex order remains in the same $\Sigma^*$7-class. The resulting theorem is a Myhill–Nerode theorem specialized to Wheeler languages: $\Sigma^*$8 is Wheeler if and only if $\Sigma^*$9 has finite index, and the minimum WDFA recognizing $\epsilon \prec \alpha$^^^^3query3^^^^ has one state per $\epsilon \prec \alpha$^^^^3\3^^^^-class (D'Agostino et al., 2021).

A complementary characterization states that a regular language is Wheeler if and only if all monotone sequences in $\epsilon \prec \alpha$^^^^3 OR \3^^^^ become eventually constant modulo $\epsilon \prec \alpha$3. This places the Wheeler condition squarely at the interface between order theory and right-congruence theory (D'Agostino et al., 2021).

3. Determinization and minimization

Determinization within the Wheeler class is substantially more controlled than general NFA determinization. Given a Wheeler NFA $\epsilon \prec \alpha$4, the canonical Wheeler determinization $\epsilon \prec \alpha$5 has state set

$\epsilon \prec \alpha$6

where $\epsilon \prec \alpha$7 is the interval of vertices reachable by $\epsilon \prec \alpha$8 from the source. This interval automaton satisfies

$\epsilon \prec \alpha$9

Accordingly, Wheeler determinization has polynomial, in fact near-linear, blow-up, in contrast with the exponential blow-up of general NFA determinization (&&&^^^^3\3query3^^^^&&&).

Recent work sharpened the algorithmic side of this picture. Given a sorted Wheeler NFA $\alpha$^^^^3query3^^^^, its Wheeler determinization can be constructed in

$\alpha$^^^^3\3^^^^

time, improving the earlier $\alpha$^^^^3 OR \3^^^^ bound. Since $\alpha$3, $\alpha$4, and $\alpha$5, this is $\alpha$6 on sorted inputs, and for $\alpha$7 it is the first linear-time Wheeler determinization algorithm. The bound is tight: there are sorted inputs for which the minimum WDFA has $\alpha$8 edges (&&&^^^^3\3query3^^^^&&&).

Minimization is equally distinctive. For general DFAs, Hopcroft’s algorithm gives $\alpha$9 time. For WDFAs, the minimum equivalent WDFA can be computed in linear $\alpha=\alpha' a$^^^^3query3^^^^ time. The algorithm exploits the fact that WDFA equivalence classes form runs in Wheeler order and identifies their borders through a sparse border graph, rather than through general partition refinement. On de Bruijn WDFAs built from real DNA datasets, an implementation “reduces the number of nodes from ^^^^3\3^^^^4% to 5^^^^3\3^^^^% at a speed of more than ^^^^3\3^^^^ million nodes per second” (&&&^^^^3\3^^^^&&&).

4. Succinct indexing and pattern matching

The primary algorithmic importance of WDFAs lies in succinct indexing. A Wheeler automaton with alphabet size $\alpha=\alpha' a$^^^^3\3^^^^ can be stored in just $\alpha=\alpha' a$^^^^3 OR \3^^^^ bits per edge and indexed for substring queries with auxiliary succinct data structures. In this representation, WDFAs support optimal-time pattern matching queries on the substring closure of the language they recognize (&&&^^^^3\3^^^^&&&).

The interval semantics given by path coherence is the crucial enabler. Pattern matching generalizes FM-index backward search: a pattern corresponds to a contiguous interval of states in Wheeler order, and extension by a letter corresponds to an interval update. This viewpoint has supported several WDFA analogues of classical string-indexing primitives (&&&^^^^3\3^^^^4&&&).

One notable example is the generalization of matching statistics from strings to WDFAs. That work introduces a notion of LCP array for Wheeler automata, built from extremal backward contexts associated with states, and proves that the pattern matching statistics of a pattern $\alpha=\alpha' a$3 can be computed in time $\alpha=\alpha' a$4 using an $\alpha=\alpha' a$5-word data structure (&&&^^^^3\3^^^^5&&&).

A more recent development targets the cache behavior of Wheeler indexes. The Graph Suffix Array for WDFAs uses the ordering of infimum and supremum strings, together with a suffix-array-like search strategy, to reduce the number of I/O operations. The data structure uses $\alpha=\alpha' a$6 RAM words and locates occurrences of a pattern $\alpha=\alpha' a$7 of length $\alpha=\alpha' a$8 with

$\alpha=\alpha' a$9

I/Os, where $\beta=\beta' b$^^^^3query3^^^^ is the number of unary paths traversed in the final graph-navigation phase. Empirically, this structure uses up to ^^^^3\3^^^^5 times the space of forward search, but it can be 5^^^^3query3query3^^^^ times faster and process a single character in less than 3 ns on deterministic Wheeler pangenome graphs (&&&^^^^3\3^^^^4&&&).

5. Recognition, hardness, and structural boundaries

Two recognition problems must be distinguished. The first asks whether a given DFA is itself Wheeler. The second asks whether the language of a given DFA is Wheeler, possibly via a different equivalent WDFA. The second is the more subtle language-level problem.

For DFAs with $\beta=\beta' b$^^^^3\3^^^^ states and $\beta=\beta' b$^^^^3 OR \3^^^^ transitions, Wheeler-language recognition can be solved in

$\beta=\beta' b$3

where $\beta=\beta' b$4 is the co-lex width of the minimal DFA. This improves the earlier $\beta=\beta' b$5 bound. The same work proves a conditional matching lower bound: unless SETH fails, the problem cannot be solved in strongly subquadratic time. In the NFA case, the same problem is PSPACE-complete (&&&^^^^3query3^^^^&&&).

The broader complexity landscape is sharply stratified. On DFAs and reduced NFAs, several Wheeler-related decision problems are polynomial. But when one moves to arbitrary NFAs or allows the alphabet order itself to vary, hardness appears quickly. In particular, deciding whether a given NFA is Wheeler is NP-complete; deciding whether the language of a given NFA is Wheeler is PSPACE-complete; and deciding whether a DFA or a language is generalized Wheeler, meaning Wheeler for some alphabet order, is NP-complete (D'Agostino et al., 2021).

A recent refinement removes dependence on a specific alphabet order by introducing universally Wheeler languages, those that are Wheeler with respect to all orders of the alphabet. The class $\beta=\beta' b$6 of Strictly Locally Testable languages is strictly included in $\beta=\beta' b$7. At the same time, $\beta=\beta' b$8 is not closed under complement, and the languages for which both $\beta=\beta' b$9 and $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$^^^^3query3^^^^ are in $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$^^^^3\3^^^^ are exactly the Definite or Reverse Definite languages. Deciding whether a regular language given by a DFA is in $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$^^^^3 OR \3^^^^ can be done in quadratic time, and this is optimal unless SETH fails (&&&^^^^3\3^^^^9&&&).

6. Extensions and broader developments

WDFAs occupy the width-$$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$3 point of a broader theory of co-lex ordering. For arbitrary DFAs, the canonical maximum co-lex order is a partial order whose width measures how far the automaton is from being Wheeler. Faster prefix-sorting algorithms compute this structure in $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$4 and $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$5 time for arbitrary DFAs, and in $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$6 time for acyclic DFAs. In this framework, a WDFA is exactly a DFA of co-lex width $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$7 (&&&^^^^3 OR \3query3^^^^&&&).

The Wheeler viewpoint also extends beyond recognizers. Sequential Wheeler transducers combine a Wheeler DFA on the input side with a monotonicity requirement on outputs. This class is closed under composition, Wheeler languages are closed under inverse image of Wheeler transductions, and a Myhill–Nerode-style theorem characterizes exactly the functions realizable by a sequential Wheeler transducer via a refined congruence $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$8 (&&&^^^^3 OR \3^^^^&&&).

At the automata-theoretic level, Wheeler bisimulations refine classical bisimulation by enforcing convexity with respect to Wheeler order. Standard bisimulation is not sufficient: its minimal quotient can fail to be Wheeler. Wheeler bisimulations, by contrast, induce a unique minimal Wheeler NFA, and in the deterministic case they recover the minimum Wheeler deterministic automaton of a language. The corresponding quotient can be built in linear time (&&&^^^^3 OR \3 OR \3^^^^&&&).

Finally, random-generation and succinct-representation questions have also been developed specifically for WDFAs. A uniform generator for WDFAs with $$\alpha \prec \beta \iff (a \prec b)\ \vee\ (a=b \wedge \alpha' \prec \beta').$$9 states, $u$^^^^3query3^^^^ transitions, and alphabet size $u$^^^^3\3^^^^ runs in $u$^^^^3 OR \3^^^^ expected time and $u$3 worst-case time with high probability, using constant working space, for all alphabets of size $u$4. The same work proves that

$u$5

bits are necessary and sufficient to encode a WDFA with $u$6 states and alphabet of size $u$7, up to an additive $u$8 term (&&&^^^^3 OR \33^^^^&&&).

Wheeler DFAs are therefore best understood not merely as an automata subclass but as a canonical interface between regular-language structure, co-lex order, and compressed indexing. Their distinctive features are the uniqueness of the Wheeler order, interval semantics for path queries, tractable determinization and minimization inside the Wheeler class, and a rapidly expanding theory linking automata, language classes, transductions, bisimulations, and succinct data structures.