Algebraic classifications of regular tree languages 413
D
Σ,k
(X) is generated freely over Def
Σ,k
by {x/δ
Σ,k
(X) | x ∈ X}. It follows from the results
of [33] that Def
Σ,k
is the Σ-VFA corresponding to Def
Σ,k
.
For any given k ≥ 0andX,thequestion“T (A) ∈ Def
Σ,k
(X)?” is again decidable for any
ΣX-recognizer A =(A,α,F); assuming that A is minimal, one has just to check whether
the algebra A is k-definite. On the other hand, it is not immediately clear that the question
“T (A) ∈ Def
Σ
(X)?” is decidable. However, Heuter [33, 34] has shown that if A has n states
and T (A) is definite, then T (A)is(n −1)-definite. Furthermore,
´
Esik [21] has shown that A
is k-definite if and only if the unary algebra (A, Tr(A)) is k-definite. Hence, the definiteness
question of regular tree languages can be reduced to the well-known question of whether a
given unary algebra is definite. Let us also note that the paper [21] contains many further
results about definite tree automata. In particular, it is shown that the class of definite tree
automata is closed under cascade compositions, and various equational characterizations are
considered.
A language is reverse definite [7] if there is a k ≥ 0 such that for any word w ∈ X
∗
of
length ≥ k, w ∈ L if and only if the prefix of w of length k is in L. In the definition of
a generalized definite language [31] a prefix condition and a suffix condition are combined.
These notions can be extended to trees as follows.
12.4 Example The counterparts of prefixes of words are subtrees. However, two differences
should be noted. Firstly, a tree may have several subtrees of a given height, and secondly,
to take into account the whole frontier part of a tree up to a certain depth, we have to also
consider the subtrees of lesser height.
For any k ≥ 0, let sub
k
(t)={s ∈ sub(t) | hg(t) <k}. For example, if f (g(c),f(x, g(c)))
then sub
0
(t)=∅,sub
1
(t)={c, x},sub
2
(t)={c, x, g(c)},sub
3
(t)={c, x, g(c),f(x, g(c))} and
sub
k
(t)=sub(t) for all k ≥ 4.
For any k ≥ 0andanyX,letρ
Σ,k
(X) be the equivalence on T
Σ
(X) defined by
sρ
Σ,k
(X) t if and only if sub
k
(s)=sub
k
(t)(s, t ∈ T
Σ
(X)).
It is obvious that ρ
Σ,k
(X) ∈ FCon
Σ
(X) with a congruence class for every possible collection
sub
k
(t) of subtrees of height <k. For any homomorphism ϕ : T
Σ
(X) →T
Σ
(Y ), any k ≥ 0
and any s, t ∈ T
Σ
(X), if sub
k
(s)=sub
k
(t) then clearly also sub
k
(sϕ)=sub
k
(tϕ). This
implies by Proposition 10.5 that Γ RDef
Σ,k
= {[ρ
Σ,k
(X))}
X
is a principal Σ-VFC for each
k =0, 1,..., and hence Γ RDef
Σ
=
k≥0
ΓRDef
Σ,k
is a Σ-VFC.
The ΣX-tree languages saturated by ρ
Σ,k
(X)aresaidtobereverse k-definite (k ≥ 0).
Let RDef
Σ,k
(X)denotethesetofallreversek-definite ΣX-tree languages. Then RDef
Σ,k
=
{RDef
Σ,k
(X)}
X
is the Σ-VTL corresponding to the Σ-VFC Γ RDef
Σ,k
.AΣX-tree language
is reverse definite if it is reverse k-definite for some k ≥ 0. Let RDef
Σ
=
k≥0
RDef
Σ,k
be
the Σ-VTL of all reverse definite Σ-tree languages. Of course, RDef
Σ
=ΓRDef
t
Σ
.
For any h, k ≥ 0andX,letγ
Σ,h,k
(X)=ρ
Σ,h
(X) ∩ δ
Σ,k
(X). It is again easy to see that
ΓGDef
Σ,h,k
= {[γ
Σ,h,k
(X))}
X
is a principal Σ-VFC for any pair h, k ≥ 0, and that the family
of these principal Σ-VFCs is directed. Therefore their union Γ GDef
Σ
=
h≥0,k≥0
ΓGDef
Σ,h,k
isalsoaΣ-VFC.
For any h, k ≥ 0andX,aΣX-tree language is called generalized (h, k)-definite if it is
saturated by γ
Σ,h,k
(X), and it is generalized definite if it is generalized (h, k)-definite for some
h, k ≥ 0. Hence, T ⊆ T
Σ
(X) is generalized (h, k)-definite, if s ∈ T if and only if t ∈ T ,for