Qualified Tree Automata

• Qualified tree automata are tree-based models that incorporate structured successor sets or additional memory, counters, and logical constraints beyond plain tuples.
• They encompass various frameworks such as alternating parity, VTAM, GPTA, and holonomic automata, each tailoring qualification to different computational and expressive needs.
• These models facilitate optimized decision procedures and efficient enumeration techniques, underscoring their significance in verifying logical and quantitative tree properties.

Searching arXiv for recent and foundational work on tree automata variants relevant to qualified tree automata. Qualified tree automata is a broad label for tree-automata formalisms in which acceptance is not determined solely by an unstructured tuple of child states, but is constrained by additional structure on successor sets, auxiliary memory, counters, variable annotations, weights, or syntactic residual information. A plausible formalisation, stated explicitly for enriched trees, is an alternating parity automaton $$\mathcal{A}=\langle Q,\Sigma,\delta,q_0,\Omega\rangle$$ over $\bbS$-enriched trees in which each transition formula $\delta(q,a)$ is interpreted over a structured successor fibre rather than over a plain successor tuple. Other strands of the literature realise the same general idea through visibly tree automata with memory and constraints, Parikh and one-counter extensions, holonomic or differential tree automata, EU-automata for QCTL and MSO, tree variable automata for query answering, and expression-derived constructions such as $k$-position, follow, equation, and $k$-C-continuation automata (Blumensath, 21 Feb 2025, 0804.3065, Herrmann et al., 2024, Herrmann et al., 2024, Manssour et al., 2024, Laroussinie et al., 2024, Amarilli et al., 2018, Mignot et al., 2014).

1. Conceptual scope

The literature suggests that “qualified tree automata” is best understood as an umbrella notion rather than as a single standardized automaton class. The common invariant is that a local move is qualified by information richer than the plain ranked-symbol interface of a classical finite tree automaton. In the enriched-tree setting, the qualification is a logical constraint on a structured successor set. In memory-based models, the qualification is a visible storage discipline together with equality or disequality tests on memories or brother subtrees. In counter models, it is the way counters are accumulated, copied, reset, or threaded globally through the tree. In query-oriented models, it is the annotation of nodes by valuations of logical variables. In expression-based constructions, it is the fact that states are identified with positions, follow sets, continuations, or partial derivatives.

This broad reading also clarifies what qualified tree automata are not. They are not restricted to one acceptance mechanism, one arity discipline, or one descriptive formalism. The represented classes range from bottom-up recognizers with tree-shaped memory to alternating parity automata on arbitrary finite branching, from weighted devices with rational size-dependent transition weights to automata that arise from regular tree expressions. What unifies them is the replacement of purely local, unqualified successor inspection by a disciplined additional structure.

2. Enriched successors and logic-parameterized transitions

The most explicit general framework is coalgebraic. A transition system is given by a map

$\suc : S \to \bbS S$

for a polynomial functor

$\bbS X=\sum_{i\in I}X^{D_i}.$

An element of $\bbS X$ is a pair $(i,s)$ with $i\in I$ and $\bbS$0, so each node carries not merely a set of successors but a $\bbS$1-indexed family of successors. A $\bbS$2-labelled $\bbS$3-enriched transition system is

$\bbS$4

where $\bbS$5 labels states and $\bbS$6 is the structured successor set of $\bbS$7. This covers finitely branching trees, bounded branching of cardinality $\bbS$8, sibling structures obtained from countable relational structures, and successors equipped with probability measures. The associated unravelling $\bbS$9 turns such systems into $\delta(q,a)$0-enriched trees, and an enriched tree is precisely a transition system isomorphic to its unravelling.

Over these trees, the transition relation of an automaton is itself logical. Fix a polynomial functor $\delta(q,a)$1 and a family of logics with polarities $\delta(q,a)$2 over $\delta(q,a)$3. An alternating $\delta(q,a)$4-automaton over $\delta(q,a)$5-enriched trees is

$\delta(q,a)$6

with finite state set $\delta(q,a)$7, parity map $\delta(q,a)$8, and transition function $\delta(q,a)$9. In a run $k$0, the qualification condition is

$k$1

so the whole structured successor object, labelled by sets of successor states, must satisfy the formula attached to $k$2. Classical alternating parity tree automata are recovered when $k$3 and $k$4 is propositional over the fixed child positions. More expressive choices of $k$5 let transitions talk about sibling relations, chains, counting congruences, measures, or other fibre-level structure.

This framework is matched by a parameterised fixed-point logic $k$6. The main equivalence theorem states that a language $k$7 is $k$8-definable iff it is recognized by an $k$9-automaton; the pure fragment $k$0 corresponds to pure $k$1-automata; and the alternation-free fragment $k$2 corresponds to weak $k$3-automata. The translations are effective. The same machinery yields characterisations of $k$4, $k$5, $k$6, $k$7, $k$8, and $k$9 on enriched trees. It is also used to give a simplified proof of Muchnik’s theorem and variants for $\suc : S \to \bbS S$0, $\suc : S \to \bbS S$1, $\suc : S \to \bbS S$2, and $\suc : S \to \bbS S$3. In this line of work, qualification is literally a predicate on the structured successor fibre (Blumensath, 21 Feb 2025).

3. Memory, visibility, and constrained transitions

A second major reading of qualified tree automata is memory-based. Tree automata with one memory are bottom-up devices

$\suc : S \to \bbS S$4

whose transitions have the form

$\suc : S \to \bbS S$5

with $\suc : S \to \bbS S$6. The memory is tree-shaped rather than stack-shaped, and transitions may manipulate it by push-like, pop-like, duplication, or equality-testing operations. This generalizes both pushdown automata and Bogaert–Tison tree automata with equality or disequality constraints between brother subtrees.

Visibly Tree Automata with Memory (VTAM) impose a visibility discipline analogous to visibly pushdown automata. The input signature is partitioned into

$\suc : S \to \bbS S$7

and every rule belongs to one visible category: PUSH, one of the POP$\suc : S \to \bbS S$8 forms, or one of the INT forms. The effect on memory is therefore syntactically determined by the input symbol. This restriction yields determinization, closure under Boolean operations, PTIME-complete emptiness, PTIME membership, and EXPTIME-complete inclusion and universality.

The constrained variants refine the qualification mechanism further. In VTAM$\suc : S \to \bbS S$9, transitions may require $\bbS X=\sum_{i\in I}X^{D_i}.$0 or $\bbS X=\sum_{i\in I}X^{D_i}.$1 for child memories. A general theorem gives decidable emptiness for VTAM$\bbS X=\sum_{i\in I}X^{D_i}.$2 when, for every VTAM$\bbS X=\sum_{i\in I}X^{D_i}.$3 automaton $\bbS X=\sum_{i\in I}X^{D_i}.$4 and state $\bbS X=\sum_{i\in I}X^{D_i}.$5, the memory language $\bbS X=\sum_{i\in I}X^{D_i}.$6 is effectively regular and the equivalence classes of $\bbS X=\sum_{i\in I}X^{D_i}.$7 are finite and enumerable. If $\bbS X=\sum_{i\in I}X^{D_i}.$8 is an arbitrary regular binary relation on memories, membership becomes NP-complete and emptiness is undecidable even for VTAM$\bbS X=\sum_{i\in I}X^{D_i}.$9. Two specific relations are especially important. For syntactic equality, VTAM$\bbS X$0 has decidable emptiness, NP-complete membership, closure under union, and failure of effective closure under complement. For structural equality $\bbS X$1, which ignores labels and checks only shape, VTAM$\bbS X$2 has decidable emptiness, PTIME membership, determinization, and full Boolean closure. Adding Bogaert–Tison brother constraints on the input tree yields BTVTAM$\bbS X$3, which still admits determinization, Boolean closure, and decidable emptiness. The paper also exhibits languages of perfectly balanced binary trees, powerlists, and red-black trees as natural examples of what these qualifications can express (0804.3065).

4. Counters, global storage, and differential weights

Counter-based variants show that qualification can be quantitative rather than purely logical or memory-based. Global Parikh tree automata (GPTA) work over decorated alphabets $\bbS X$4, accumulate a single global counter vector

$\bbS X$5

and test a semilinear constraint once for the whole tree. Non-global PTA and PTAR instead propagate counter configurations along root-to-leaf paths, copying them at branching and checking the semilinear constraint at leaves. The language

$\bbS X$6

is recognized by a $\bbS X$7-PTA but not by GPTA, whereas

$\bbS X$8

is recognized by a $\bbS X$9-GPTA but not by PTA. Hence GPTA and non-global PTA are incomparable. For non-global PTA, non-emptiness is undecidable for $(i,s)$0, membership remains decidable, and linear PTAR regains decidable non-emptiness through a reduction based on spinal computation trees and Parikh automata on words (Herrmann et al., 2024).

Global One-Counter Tree Automata (GOCTA) impose a different storage discipline: a single global counter is passed through the tree in lexicographical order and is never duplicated at branchings. The input tree is generated by configurations in $(i,s)$1, and every step acts at the lexicographically first state occurrence. This global discipline is incomparable with ordinary one-counter tree automata (OCTA), which copy the counter to all children. GOCTA recognize languages such as $(i,s)$2, where a global equality of symbol counts is enforced across the whole tree, but they do not recognize $(i,s)$3, which OCTA do recognize. Emptiness for GOCTA is undecidable, yet membership is in P. A key normal-form lemma states that for a normalized, $(i,s)$4-accepting GOCTA and any accepted tree $(i,s)$5, there exists a successful computation $(i,s)$6 with

$(i,s)$7

This polynomial counter bound underlies the bounded-behaviour automaton used for the polynomial-time membership test (Herrmann et al., 2024).

A third quantitative direction is given by holonomic or differential tree automata. Here a ranked symbol $(i,s)$8 carries a transition matrix $(i,s)$9 whose entries are rational functions in the size of the parent tree and the sizes of its $i\in I$0 immediate subtrees, taken from

$i\in I$1

The weight of a tree is computed bottom-up by

$i\in I$2

and the associated census generating function

$i\in I$3

is exactly a differentially algebraic power series. Conversely, every differentially algebraic series is realized by some holonomic tree automaton. The model strictly generalizes weighted tree automata, yields effective constructions from rational dynamical systems, and has a decision procedure for equality of automata obtained via differential-algebraic zeroness testing (Manssour et al., 2024).

5. Logical characterisations and query-oriented automata

EU-automata provide a quantified view of qualification on infinite trees of arbitrary finite arity. Their basic transition atoms are EU-constraints

$i\in I$4

where $i\in I$5 is an existential multiset requirement and $i\in I$6 is a universal residual set. A multiset $i\in I$7 over states satisfies $i\in I$8 iff $i\in I$9 and the support of $\bbS$00 is contained in $\bbS$01. Alternating EU-automata allow positive Boolean combinations of such constraints and parity acceptance; non-alternating EU-automata restrict transitions to disjunctions. The model is tailored to arbitrary finite branching because transitions quantify over successor-state multisets rather than over fixed child indices. The paper develops algorithms for union, intersection, complement, projection, alternation removal, membership, and emptiness, and uses them to obtain optimal decision procedures for QCTL satisfiability and model checking. It also proves that any QCTL formula with $\bbS$02 quantifier alternations can be translated to a formula with at most one quantifier alternation, with a $\bbS$03-exponential blow-up, and that any MSO formula can be translated into one with at most four quantifier alternations and only two second-order-quantifier alternations, again with a $\bbS$04-exponential blow-up (Laroussinie et al., 2024).

Tree Variable Automata (TVA) supply another, more operational, interpretation. A $\bbS$05-TVA is a nondeterministic bottom-up tree automaton

$\bbS$06

whose leaves or nodes are annotated by subsets of a finite variable set $\bbS$07. In the binary case, a valuation $\bbS$08 annotates each leaf by $\bbS$09, and a run is accepting when the root ends in a final state under that annotation. The model captures MSO queries with free variables. The paper introduces homogenization into $\bbS$10-states and $\bbS$11-states, compiles TVAs into complete structured DNNFs of width $\bbS$12, and obtains enumeration with dynamic updates. For an unranked $\bbS$13-TVA $\bbS$14 with state space $\bbS$15 and an unranked $\bbS$16-tree $\bbS$17, satisfying assignments can be enumerated with

$\bbS$18

where $\bbS$19 is the current assignment being output. For fixed MSO queries with free first-order variables, this yields linear preprocessing, constant delay, and $\bbS$20 updates. The same work proves a lower bound showing that there exists an MSO query $\bbS$21 such that any enumeration algorithm for $\bbS$22 under relabelings must satisfy

$\bbS$23

This suggests a distinct operational sense of qualification: the automaton is structurally restricted so as to support efficient enumeration and maintenance rather than merely recognition (Amarilli et al., 2018).

6. Expression-derived constructions, reduction, and comparative significance

Regular tree expressions induce several canonical automata constructions. For a linear tree expression $\bbS$24, the $\bbS$25-position tree automaton $\bbS$26 has one state $\bbS$27 for each child position $\bbS$28 of each occurrence of a non-constant symbol $\bbS$29, plus the initial state $\bbS$30. Its transitions are determined by the position functions $\bbS$31 and $\bbS$32. The follow automaton $\bbS$33 quotients these states by equality of follow sets. The equation tree automaton $\bbS$34 uses partial derivatives $\bbS$35 as states, and the $\bbS$36-C-continuation automaton $\bbS$37 uses continuations $\bbS$38. These are the tree analogues of the Glushkov position automaton, the Ilie–Yu follow automaton, Antimirov’s equation automaton, and Champarnaud–Ziadi’s c-continuation automaton for words. The morphic relations are preserved in the tree setting: $\bbS$39 By contrast, $\bbS$40 and $\bbS$41 are incomparable in general. This is a syntactic form of qualification: the same language can be recognized by states interpreted as positions, follow sets, continuations, or residual equations (Mignot et al., 2014).

Reduction theory provides a final comparative perspective. For ranked nondeterministic tree automata, transition pruning and quotienting can be organized around downward and upward simulations, trace inclusions, and their lookahead approximations. A pruning relation $\bbS$42 is good for pruning (GFP) when it preserves the language, and an equivalence is good for quotienting (GFQ) when quotienting preserves the language. The Heavy framework alternates useless-state removal, quotienting, and GFP pruning using polynomial-time approximations of trace inclusion. The corresponding paper explicitly notes that, although it does not mention qualified tree automata, its theory is relevant for any extension of tree automata that still has a regular tree language semantics. This suggests that any qualified variant with finite-state local structure should admit an analogous reduction theory once appropriate qualified simulations, qualified trace inclusions, and GFP/GFQ proofs are in place (Almeida et al., 2015).

Taken together, these lines of work indicate that qualified tree automata are best viewed as a research program rather than a single automaton definition. In one branch, qualification means logical predicates over enriched successor fibres; in another, it means visible memory operations and structural constraints; in another, pathwise or global counter flow; in another, rational size-dependent weights; in another, quantifier-sensitive successor constraints for temporal and monadic second-order logics; and in another, syntactic information extracted from regular tree expressions. A plausible implication is that the term is most useful when it names this common shift from plain successor tuples to semantically structured local transitions, while leaving the specific qualification mechanism to the underlying model (Blumensath, 21 Feb 2025).