Decidable and undecidable fragments of first-order branching ...
Decidable and undecidable fragments of first-order branching ...
Decidable and undecidable fragments of first-order branching ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Decidable</strong> <strong>and</strong> <strong>undecidable</strong> <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> <strong>branching</strong><br />
temporal logics<br />
Ian Hodkinson, Frank Wolter, <strong>and</strong> Michael Zakharyaschev<br />
(e-mails:ÑÓºººÙ¸ÛÓÐØÖÒÓÖÑغÙÒ¹ÐÔÞº¸<br />
Department <strong>of</strong> Computing, Imperial College<br />
180 Queen’s<br />
ÑÞ׺кºÙµ<br />
Gate, London SW7 2BZ, U.K.;<br />
Institut für Informatik, Universität Leipzig<br />
Augustus-Platz 10-11, 04109 Leipzig, Germany;<br />
Department <strong>of</strong> Computer Science, King’s College<br />
ÓÒØØÙØÓÖÅÐÖÝ×Ú<br />
Str<strong>and</strong>, London WC2R 2LS, U.K.<br />
Ü´·µ¾¼¾½ ÑÞ׺кºÙ ÈÓÒ´·µ¾¼¾¾ ÔØÓÓÑÔÙØÖËÒ¸ÃÒ³×ÓÐиËØÖÒ¸ÄÓÒÓÒϾʾÄ˸ÍÃ<br />
Abstract<br />
In this paper we analyze the decision problem for <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> extensions <strong>of</strong><br />
<strong>branching</strong> time temporal logics such as computational tree logicsÌÄ<strong>and</strong>ÌÄ£or Prior’s Ockhamist<br />
logic <strong>of</strong> historical necessity. On the one h<strong>and</strong>, we show that the one-variable <strong>fragments</strong><br />
<strong>of</strong> logics like <strong>first</strong>-<strong>order</strong> C T L£—such as the product <strong>of</strong> propositional C T L£with simple propositional<br />
modal logicË, or even the one-variable bundled <strong>first</strong>-<strong>order</strong> temporal logic with sole<br />
temporal operator ‘some time in the future’—are <strong>undecidable</strong>. On the other h<strong>and</strong>, it is proved<br />
that by restricting applications <strong>of</strong> <strong>first</strong>-<strong>order</strong> quantifiers to state (i.e., path-independent) formulas,<br />
<strong>and</strong> applications <strong>of</strong> temporal operators <strong>and</strong> path quantifiers to formulas with at most one free variable,<br />
we can obtain decidable <strong>fragments</strong>. The same arguments show decidability <strong>of</strong> ‘non-local’<br />
propositionalÌÄ£, in which truth values <strong>of</strong> propositional atoms depend on the history as well<br />
as the current time. The positive decidability results can serve as a unifying framework for devising<br />
expressive <strong>and</strong> effective time-dependent knowledge representation formalisms, e.g., temporal<br />
description or spatio-temporal logics.<br />
0
1 Introduction<br />
Temporal logics <strong>of</strong> <strong>branching</strong> time originate in philosophy <strong>and</strong> computer science. In philosophy,<br />
they formalize reasoning about indeterminate future; see e.g., [18, 19, 5, 20]. In computer science,<br />
they are used to reason about state transition systems (computations <strong>of</strong> reactive systems, agents in an<br />
unpredictable environment, etc.); see e.g., [6, 8, 16]. Families <strong>of</strong> different logics have been constructed<br />
in both disciplines, <strong>and</strong> in both cases most <strong>of</strong> the logics are propositional (the only exceptions we know<br />
<strong>of</strong> are [4, 22, 23]).<br />
The main aim <strong>of</strong> this paper is to investigate the computational behavior—at least on the level <strong>of</strong><br />
decidability—<strong>of</strong> <strong>first</strong>-<strong>order</strong> <strong>branching</strong> temporal logics (FOBTLs).<br />
The starting point <strong>of</strong> our investigation lies in recent work [14] on <strong>first</strong>-<strong>order</strong> linear temporal logic<br />
(FOLTL), which showed that although very weak <strong>fragments</strong> (say, the monadic two-variable fragment)<br />
<strong>of</strong> st<strong>and</strong>ard FOLTLs can be highly <strong>undecidable</strong>, by restricting applications <strong>of</strong> temporal operators<br />
to formulas with at most one free variable, we obtain a fragment with much better computational<br />
behavior. This ‘monodic’ fragment can be axiomatized in a natural way [25] <strong>and</strong> becomes decidable<br />
if its ‘purely <strong>first</strong>-<strong>order</strong> part’ is restricted to a decidable fragment <strong>of</strong> <strong>first</strong>-<strong>order</strong> logic.<br />
Of course, FOBTLs inherit bad computational properties <strong>of</strong> FOLTLs. For example, the guarded<br />
L£<br />
two-variable fragment <strong>of</strong> most FOBTLs is not even recursively enumerable; cf. [14]. Unfortunately<br />
(<strong>and</strong> to our surprise) it turned out that the situation is much worse. As will be shown in Section 3,<br />
the one-variable fragment <strong>of</strong> <strong>first</strong>-<strong>order</strong> CT L£is <strong>undecidable</strong>. This is a monodic fragment with very<br />
simple decidable <strong>first</strong>-<strong>order</strong> part; it can be reformulated as the product <strong>of</strong> propositional CT L£with<br />
propositionalË. Even the one-variable fragment <strong>of</strong> the bundled FOBTL [5] with sole temporal<br />
operator ‘now or some time in the future’ is not decidable.<br />
These ‘negative’ results become less surprising when we realize that <strong>first</strong>-<strong>order</strong> (bundled) CT<br />
allows a form <strong>of</strong> quantification in three dimensions, something that is known to be associated with<br />
undecidability [17, 13]. A natural way to limit the interaction between the three dimensions is to restrict<br />
applications <strong>of</strong> <strong>first</strong>-<strong>order</strong> quantifiers to state (i.e., path-independent) formulas, <strong>and</strong> applications<br />
<strong>of</strong> temporal operators <strong>and</strong> path quantifiers to formulas with at most one free variable. The resulting<br />
fragment contains full propositional CT L£<strong>and</strong> full <strong>first</strong>-<strong>order</strong> logic. In this paper, we show in a<br />
similar way to FOLTL that restricting its <strong>first</strong>-<strong>order</strong> part to a decidable <strong>first</strong>-<strong>order</strong> fragment yields a<br />
decidable monodic fragment <strong>of</strong> <strong>first</strong>-<strong>order</strong> CT L£. A number <strong>of</strong> ‘positive’ results <strong>of</strong> this sort will be<br />
proved in Section 4.<br />
These results are interesting per se, <strong>and</strong> also because they identify a limit beyond which monodic<br />
<strong>fragments</strong> <strong>of</strong> temporal logics are no longer decidable. More practically, the positive results serve as<br />
a unifying framework for devising expressive <strong>and</strong> effective time-dependent knowledge representation<br />
formalisms, say, temporal description or spatio-temporal logics; for details see [15].<br />
2 Syntax <strong>and</strong> semantics<br />
There are a number <strong>of</strong> approaches to constructing temporal logics based on the <strong>branching</strong> time<br />
paradigm; see, e.g., [5, 7, 26, 10]. Many <strong>of</strong> the resulting languages are <strong>fragments</strong> <strong>of</strong> the language<br />
we call here Q P CT L£, quantified CT L£with past operators. It is obtained in the st<strong>and</strong>ard way by<br />
extending the language P L <strong>of</strong> classical predicate logic (without equality or function symbols) with<br />
the binary temporal operatorsË(Since) <strong>and</strong>Í(Until) <strong>and</strong> the path universal quantifier.<br />
The intended models for Q P CT L£are based on ω-trees. Recall that a tree is a strict partial <strong>order</strong><br />
Wcontaining a unique-minimal point (the root <strong>of</strong>) <strong>and</strong> such that for all w¾W, the set<br />
1
0<br />
I´wµ<br />
0c<br />
v¾W : vwis linearly <strong>order</strong>ed by. A full branch <strong>of</strong>is a maximal linearly-<strong>order</strong>ed subset <strong>of</strong><br />
W. An ω-tree is a tree whose full branches are all <strong>order</strong>-isomorphic toÆ.<br />
Q P CT L£(as well as its sublanguages to be introduced below) is interpreted in structures <strong>of</strong> the<br />
formÅÅHDI«, whereWis an ω-tree, H is a set (bundle) <strong>of</strong> full branches <strong>of</strong>with<br />
ËHW, D is a non-empty I´wµ<br />
I´wµ I´wµ<br />
set called the domain <strong>of</strong>Å, <strong>and</strong> I is a function associating I´wµ<br />
with I´vµ<br />
every<br />
moment <strong>of</strong> time w¾W a <strong>first</strong>-<strong>order</strong> P L-structure<br />
I´wµDP the state <strong>of</strong>Åat moment w. (Here, the Pi<br />
are predicates on D interpreting the predicate symbols<br />
P i <strong>of</strong> P L <strong>and</strong> the ci<br />
are elements <strong>of</strong> D interpreting its constants.) We require that c c i i for<br />
any wv¾W—i.e., that constants are ‘rigid’. The branches in the bundle H are called histories. If H<br />
contains all full branches <strong>of</strong>, we say thatÅis a full tree model, or simply a tree model.<br />
An assignment in D is a functionfrom the set <strong>of</strong> individual variables to D. (So assignments too<br />
are ‘rigid’. We extendto constants via´cµcI´wµfor any w¾W . The truth relation´Åhwµϕ,<br />
for w¾h¾H (or simply´hwµϕifÅis understood), is defined as follows:<br />
¯´hwµP i´y 1yµiff I´wµP i´´y 1µ´yµµ, where the y i are variables or constants<br />
(this is the so-called ‘local’ approach in that there is no dependence on h),<br />
¯´hwµxψ iff´hwµψfor every assignmentin D that may differ fromonly on x,<br />
¯´hwµχËψ iff there is vw such that´hvµψ <strong>and</strong>´huµχ for every u¾´vwµ,<br />
where´vwµu¾W : vuw,<br />
¯´hwµχÍψ iff there is v¾h such that vw,´hvµψ<strong>and</strong>´huµχfor<br />
ϕϕ¿FϕϕÍϕϕϕ<br />
every u¾´wvµ, ¯´hwµψ iff´h¼wµψfor all h¼¾H such that w¾h¼,<br />
plus the usual clauses for the Booleans. For a formula ϕ´¯xµ<strong>and</strong> a tuple ā <strong>of</strong> elements <strong>of</strong> D, we write<br />
´Åhwµϕ´āµif´Åhwµϕwhere´¯xµā. We will use the following st<strong>and</strong>ard abbreviations:<br />
¿FϕÍϕ¾Fϕ¿Fϕ¾·F ϕϕ¾Fϕ¿·F<br />
Thus,¿F can be read as ‘some time in the future’,¾·F as ‘from now on’,as ‘at the next moment’<br />
or ‘tomorrow’, <strong>and</strong>as ‘there exists a history in H ’.<br />
We will also be considering the following sublanguages <strong>of</strong> Q P CT L£:<br />
¯Q CT L£—that is, Q P CT L£without the past temporal operatorË;<br />
¯Q CT L£F—that is, Q CT L£in which the binary operatorÍis replaced by¾·F ;<br />
¯Q P CT L—the fragment <strong>of</strong> Q P CT L£in which temporal operators occur only in the form<br />
´ψ 1Íψ 2µ,´ψ 1Ëψ 2µ,´ψ 1Íψ 2µ, or´ψ 1Ëψ 2µ.<br />
To introduce one more fragment, Q P CT L s , we need the definition <strong>of</strong> state <strong>and</strong> path formulas:<br />
¯all formulas without path quantifiers <strong>and</strong> temporal operators are state formulas;<br />
¯the set <strong>of</strong> state formulas is closed under the Booleans, path <strong>and</strong> <strong>first</strong>-<strong>order</strong> quantifications;<br />
¯every state formula is a path formula;<br />
2
¯the set <strong>of</strong> path formulas is closed under the Booleans <strong>and</strong> the temporal operators;<br />
¯if ψ is a path formula, thenψ is a state formula.<br />
So for a state formula ψ, whether´hwµψdoes not depend on h. Now, Q P CT L s consists <strong>of</strong><br />
all state formulas in Q P CT L£. The main difference between full Q P CT L£<strong>and</strong> Q P CT L s is that<br />
<strong>first</strong>-<strong>order</strong> quantifiers in Q P CT L s can be applied only to formulas which are history-independent.<br />
(That Q P CT L s only contains state formulas is not important for decidability, since a path formula<br />
ϕ is satisfiable<br />
¾Fx´ÓÖÖØÑ´xµÐÚÖØÑ´xµµ ¾Fx´ÓÖÖØÑ´xµÐÚÖØÑ´xµµ x¾F´ÓÖÖØÑ´xµÐÚÖØÑ´xµµ iff the state formulaϕ is satisfiable.) It should<br />
be clear that<br />
¢ Q P CT L<br />
¢<br />
s contains<br />
propositional P CT L£<strong>and</strong> Q P CT L. As examples, we give three formulas in the various <strong>fragments</strong>,<br />
trying to express that every <strong>order</strong>ed item is delivered in one day.<br />
Q P CT L£Q P CT L s Q P CT L<br />
For any <strong>of</strong> the languages L introduced above, denote byÄ(respectively,Ä) the set <strong>of</strong> all L-<br />
formulas that are true at all points in all histories under every assignment in every bundled (respectively,<br />
full) tree model. Such formulas are said to be valid, <strong>and</strong> a formula is satisfiable if its negation<br />
is not valid. Thus,ÉÈÌÄ£is the set <strong>of</strong> Q P CT L£-formulas valid in bundled tree models, while<br />
ÉÈÌÄ£is the set <strong>of</strong> Q P CT L£-formulas valid in all tree models. The logicÉÈÌÄ£is the <strong>first</strong><strong>order</strong><br />
version <strong>of</strong> the bundled Ockhamist logic <strong>of</strong> historical necessity; cf. e.g. [10].<br />
Q P CT L£-satisfiability in bundled tree models can be reduced to satisfiability in full tree models.<br />
Indeed, given a Q P CT L£-formula ϕ, we take a propositional variable q not occurring in ϕ <strong>and</strong> denote<br />
by ϕthe result <strong>of</strong> replacing each subformula <strong>of</strong> ϕ <strong>of</strong> the formψ by´¿F¾Fqψµ. Note that if ϕ<br />
is in Q P CT L s then so is ϕ(however, ϕ¾QP CT L).<br />
LEMMA 1. ϕ is satisfiable in a bundled tree model iff¿F¾Fqϕis satisfiable in a full tree model.<br />
Pro<strong>of</strong>. The implication´´µis easily seen. We prove´µµ. Using a Löwenheim–Skolem argument<br />
(cf. [5]), we may assume ϕ to be satisfied in a modelÅwith a countable bundle H . We assume that<br />
H is infinite, leaving the (easy) other case to the reader. Let h 0h 1be an enumeration <strong>of</strong> H . We<br />
convertÅinto a full tree modelņ <strong>and</strong> define a truth-relation for q in it inductively as follows: Put<br />
´h 0wµq for all w¾h 0 . Suppose we have already defined truth <strong>of</strong> q in´h iwµ, for all in. Consider<br />
h n·1. There must be a w¾h n·1 such that the distance from w to each h i , in·1, is2 (the distance<br />
is the length <strong>of</strong> the shortest path from w to a point in h i ). Then we put´h n·1w¼µq if w¼w <strong>and</strong><br />
w¼¾h n·1.<br />
Say that a full branch h <strong>of</strong>Åis marked if there is w¾h such that´hw¼µq for all w¼w,<br />
w¼¾h. One can easily see that h is marked iff h¾H. In particular, if h¾H <strong>and</strong> for each n, t n is the<br />
least element <strong>of</strong> hÒËmn h m , then´ht nµq <strong>and</strong>t 0t 1is infinite, so h is not marked. Now one<br />
can prove by induction that for every subformula ψ <strong>of</strong> ϕ <strong>and</strong> every´hwµ, we have´Åhwµψ iff<br />
´Å†hwµψ. It follows that¿F¾Fqϕis satisfied inņ .<br />
3 Undecidable <strong>fragments</strong><br />
The following theorems indicate some limits beyond which one cannot hope to find decidable <strong>fragments</strong><br />
<strong>of</strong> <strong>first</strong>-<strong>order</strong> temporal logics.<br />
Given a <strong>first</strong>-<strong>order</strong> temporal language L <strong>and</strong>ω, we denote by Lthe-variable fragment <strong>of</strong> L<br />
(i.e., every formula in Lcontains at mostdistinct individual variables). And by L mo we denote the<br />
3
monadic fragment <strong>of</strong> L (i.e., the set <strong>of</strong> formulas which contain only unary predicates <strong>and</strong> propositional<br />
variables). Both the two-variable <strong>and</strong> the monadic <strong>fragments</strong> <strong>of</strong> classical (non-temporal) <strong>first</strong>-<strong>order</strong><br />
logic are known to be decidable <strong>and</strong> have the finite model property; see [3] <strong>and</strong> references therein.<br />
The computational behavior <strong>of</strong> the corresponding <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> temporal logics turns out to<br />
be quite different. From linear time results (Theorem 2 <strong>of</strong> [14]) we easily obtain:<br />
THEOREM 2. For any <strong>of</strong> the FOBTLsÄintroduced above,ÄL 2L mo is not recursively enumerable.<br />
Another well-behaved fragment <strong>of</strong> classical predicate logic is the guarded fragment <strong>of</strong> [1]. The<br />
corresponding fragment T GF <strong>of</strong> FOBTL is obtained by restricting the Q P CT L£quantification<br />
formation rule to:<br />
¯if xy are tuples <strong>of</strong> variables, G´xyµis an atomic formula, ϕ´xyµ¾TGF , <strong>and</strong> every free<br />
variable occurring in ϕ´xyµoccurs in G´xyµas well, theny´G´xyµϕ´xyµµis in T GF .<br />
The set T GF is called the guarded fragment <strong>of</strong> FOBTL. Again in contrast to the case <strong>of</strong> classical<br />
predicate logic, we have the following consequence <strong>of</strong> Theorem 73 <strong>of</strong> [14]:<br />
THEOREM 3. For any <strong>of</strong> the FOBTLsÄabove,ÄL 2T GF is not recursively enumerable.<br />
The main result <strong>of</strong> this section is the following:<br />
THEOREM 4. The one-variable <strong>fragments</strong> <strong>of</strong> the logicsÉÌÄ£F <strong>and</strong>ÉÌÄ£F are <strong>undecidable</strong>. Hence,<br />
so are the one-variable <strong>fragments</strong> <strong>of</strong>ÉÈÌÄ£,ÉÌÄ£,ÉÈÌÄ£, <strong>and</strong>ÉÌÄ£.<br />
Pro<strong>of</strong>. We only considerÉÌÄ£F ; the other case follows from Lemma 1. The pro<strong>of</strong> involves a<br />
rather indirect reduction <strong>of</strong> the following tiling problem. An instance is a finite set T <strong>of</strong> square tiles,<br />
the edges <strong>of</strong> each τ¾T being colored Left´τµRight´τµUp´τµDown´τµ(see Fig. 2 in Appendix). 1<br />
T is a yes-instance <strong>of</strong> the problem iff for each τ¾T there is a T -tiling <strong>of</strong>¢that uses τ—a map<br />
f :¢T with Right´f´ijµµLeft´f´i·1jµµ, Up´f´ijµµDown´f´ij·1µµfor all ij¾,<br />
<strong>and</strong> (without loss <strong>of</strong> generality) f´00µτ. It is well known that this tiling problem is <strong>undecidable</strong>—<br />
see, e.g., [2].<br />
To code this problem intoÉÌÄ£F, we go via a construction originating in algebraic logic, which<br />
has already been used to prove undecidability <strong>of</strong> 3-dimensional modal logics [13]. It was shown in<br />
[12] that there is no algorithm to decide whether a finite Tarskian relation algebra representable. We<br />
do not wish to assume any knowledge <strong>of</strong> algebraic logic here, so we will state some consequences <strong>of</strong><br />
this result without using terms from that field. Given a tiling instance T , we can effectively construct<br />
a certain finite signature L´Tµ<strong>of</strong> binary relation symbols, <strong>and</strong> a set satisfying´pqrµ¾<br />
CL´Tµ3<br />
Cµ´qrpµ¾C (<strong>and</strong> with further properties to be explained later). We call C the set <strong>of</strong> consistent<br />
triples. Results in [12] imply that if T is a yes-instance then there exists a countable L´Tµ-structure<br />
N satisfying:<br />
Ïp¾L´Tµp´xyµÏq¾L´TµÒpq´xyµ¡Îp¾L´Tµxy ¯xy p´xyµ,<br />
Ï´pqrµ¾C ¯xy p´xyµ¡°z q´yzµr´zxµ¡¡, for all qr¾L´Tµ.<br />
Thus, each pair <strong>of</strong> points in N is labelled by a unique p¾L´Tµ; only consistently-labelled triangles<br />
occur in N, but these must occur wherever possible. We will write a one-variable Q CT L£F-sentence<br />
ϕ T , essentially the Yankov–Fine formula <strong>of</strong> a frame derived from C, that has a bundled model iff T<br />
is a yes-instance. This will establish undecidability <strong>of</strong>ÉÌÄ£F.<br />
1 Strictly, we also require that there is a ‘white’ tile in T whose sides are all colored white; no other tile has any white<br />
edges. This is for technical reasons in [12] not relevant here.<br />
4
In Q CT L£F, truth values <strong>of</strong> atomic formulas are history-independent (cf. Section 2). This is<br />
inconvenient here, so we temporarily drop this ‘locality’ restriction. We introduce three unary relation<br />
symbols p 0p 1p 2 for each p¾L´Tµ. Let ϕ T be the conjunction <strong>of</strong><br />
Ïp ¾·Fx p 0´xµÏqp q 0´xµ¡Îp¾L´Tµ¿·Fx p 0´xµ<strong>and</strong><br />
pq¾L´Tµ qr¾L´Tµ pr¾L´Tµ<br />
similarly for p 1p 2<br />
¾·Fx ´Ï´pqrµ¾C p 0´xµµ°x´q 1´xµr 2´xµµ¡for all<br />
¾·Fx ´Ï´pqrµ¾C q 1´xµµ°¿·F´r 2´xµp 0´xµµ¡for all<br />
¾·Fx ´Ï´pqrµ¾C r 2´xµµ°´p 0´xµq 1´xµµ¡for all<br />
If T is a yes-instance, take N as above. We identify (notationally) N with its domain. Choose an<br />
ω-treeWwith a countably infinite set H <strong>of</strong> full branches such thatËHW. Let π 0 be the<br />
identity map on N. Choose µ : ωN such that µ 1´aµis infinite for all a¾N, <strong>and</strong> define π 1 : WN<br />
by π 1´wµµ´ht´wµµ, where ht´wµv¾W : vw. 1<br />
Choose π 2 : HN such thatË´π<br />
for all a¾N. Now we can define values for the p i j in a modelÆwith domain N, underlying tree,<br />
<strong>and</strong> bundle H , by pulling back from N. For a¾N, h¾H,w¾h, <strong>and</strong><br />
1´wµµ 0´aµµ 2´hµµ<br />
pqr¾L´Tµ, we let<br />
´Æhwµp 0´aµiff Np´π 1´wµπ ´Æhwµq 1´aµiff Nq´π 2´hµπ ´Æhwµr 2´aµiff Nr´π 0´aµπ It can be checked that ϕ T is true inÆat the root.<br />
We now show how to codeÆinto an <strong>of</strong>ficial Q CT L£F-model with ‘local’ history-independent<br />
atoms. We exploit the fact that the values <strong>of</strong> the atoms inÆdepend on only two coordinates. r 2<br />
depends on aw, which is acceptable already. p 0 depends on wh (actually, on π 2´hµ) but not a; we<br />
can express this by coding h as π 2´hµ, so reducing the dependence to w <strong>and</strong> a domain point. And q 1<br />
depends on ha but not w; by regarding q 1 as true at ha iff it is in fact true at aw for c<strong>of</strong>initely many<br />
w¾h, we again reduce the dependence to wa.<br />
In detail, we translate ϕ T into a Q CT L£F -sentenceϕ T <strong>of</strong> the signature <strong>of</strong> ϕ T plus a unary relation<br />
symbol Q. For pqr¾L´Tµ, we define formulas<br />
2´xµ 1´xµ Q´xµµ p 0´xµx´p 0´xµ¿·F¾·F<br />
q 1´xµ¿·F¾·F q<br />
r 2´xµr <strong>and</strong> letϕ T be the result <strong>of</strong> replacing each subformula p 0´xµ<strong>of</strong> ϕ T byp 0´xµ, <strong>and</strong> similarly for q 1r 2 .<br />
We form a Q CT L£F-structureÅwith the same tree, bundle, <strong>and</strong> domain asÆ. Enumerate H as<br />
h 0h 1, <strong>and</strong> for nω let h nh nÒËmn h m , a c<strong>of</strong>inite subset <strong>of</strong> h n . For h¾H, w¾h, a¾N,<br />
<strong>and</strong> pqr¾L´Tµ, define<br />
iff aπ 2´hµ,<br />
2´aµµW<br />
Q´aµ ¯´ÅhwµQ´aµiff there is g¾H with w¾g <strong>and</strong> aπ 2´gµ—so´Åhwµ¿·F¾·F ¯´Åhwµp 0´aµiff there is g¾H with w¾g,´Ægwµp ¯´Åhwµq 1´aµiff there is g¾H with w¾g <strong>and</strong>´Ægwµq ¯´Åhwµr 2´aµiff´Æhwµr 2´aµ.<br />
0´aµ, <strong>and</strong> π 2´gµa,<br />
1´aµ,<br />
These values do not depend on h, soÅis a bona fide Q CT L£F -structure. It can now be checked that<br />
´Åhwµp 0´aµiff´Æhwµp 0´aµ, <strong>and</strong> similarly forq 1r . Hence,ϕ T is true inÅ.<br />
5<br />
2
For the converse,<br />
1<br />
we<br />
2<br />
will need more details about L´Tµ<strong>and</strong> C from [12]. There is a ‘converse’<br />
map p˘p defined on L´Tµ, with ˘ppfor all p. L´Tµcontains a relation symbol τ for each tile<br />
τ¾T (we identify the two, so that TL´Tµ), the converse ˘τ <strong>of</strong> each tile τ, <strong>and</strong> also relation symbols<br />
012·1 1 1·1 1 201100220 (<strong>and</strong><br />
iµ˘<br />
others that do not concern us here), all distinct.<br />
Each p¾L´Tµhas a start index st´pµ<strong>and</strong> an end index end´pµ. The start <strong>and</strong> end indices <strong>of</strong>i j are<br />
ij, respectively; the start <strong>and</strong> end indices <strong>of</strong>i·1 i , <strong>and</strong> 1 i are i, <strong>and</strong> the tiles have start index 1 <strong>and</strong><br />
end<br />
¯´1001<br />
index 2. We have g˘<br />
i jg ji , ˘ii, <strong>and</strong>´·1 1 i . C satisfies:<br />
¯If´pqrµ¾C then end´pµst´qµ, end´qµst´rµ, end´rµst´pµ, <strong>and</strong>´qrpµ´˘r˘q˘pµ¾C.<br />
¯For all pq¾L´Tµ, p˘q iff´pqiµ¾C for some i3.<br />
¯For all p¾L´Tµ,´2001pµ¾C iff p¾T .<br />
1 1µ´2002·1 2µ¾C.<br />
¯for τυ¾T ,´ῠ·1 1τµ¾C iff Right´υµLeft´τµ, <strong>and</strong>´τ 1 2ῠµ¾C iff Down´τµUp´υµ.<br />
Now assume thatϕ T is true (at the root, say) in some Q CT L£F-modelÅwith domain D, tree<br />
W, <strong>and</strong> bundle H . We will sketch a pro<strong>of</strong> that T is a yes-instance. Let τ 00¾T be given; we<br />
will read <strong>of</strong>f fromÅatiling with τ 00 at the origin. As notation, for pqr¾L´Tµ, a¾D, h¾H,<strong>and</strong><br />
w¾h, we write´awhµpqr iff´Åhwµp 0´aµq 1´aµr 2´aµIt follows from the form <strong>of</strong>p 0<br />
<strong>and</strong> <strong>of</strong> ϕ T that<br />
¯for all awh, there are unique pqr¾L´Tµwith´awhµpqr; moreover,´pqrµ¾C,<br />
¯for all´pqrµ¾C there are ahw with´awhµpqr,<br />
¯whenever´awhµpqr <strong>and</strong>´a¼whµp¼q¼r¼, we have pp¼,<br />
0hµ<br />
¯whenever´awhµpqr <strong>and</strong>´pq¼r¼µ¾C, there is a¼¾D with´a¼whµpq¼r¼.<br />
Analogous facts hold for q 1r 2 . We use these facts below without explicit mention.<br />
Asϕ T is true inÅ, <strong>and</strong>´2001τ00µ¾C, there are x 0¾D, h¾H, <strong>and</strong> y 0¾h with´x 0y 2001τ 00 . From this, <strong>and</strong> since´20022µ¾C, there must be u 0¾D with´u 0y 0hµ20022.<br />
Similarly,´10011µ¾C, so there is v 0¾h with y 0v 0 <strong>and</strong>´x 0v 0hµ10011. In this way, by<br />
induction, for each i¾we can find x iu i¾D <strong>and</strong> y 0y iv i¾h with<br />
<strong>and</strong>´x i·1v ihµ1001 1 1 ,<br />
¯´u iy ihµ20022 <strong>and</strong>´u iy i·1hµ2002·1 2 .<br />
See Fig. 3 in Appendix.<br />
Consider now´x iy jhµ(ij¾). We have´x iy jhµpqr for some unique´pqrµ¾C. Since<br />
´u jy jhµ20022, we have p20. Since´x iv ihµ10011, we have q01. By the properties<br />
<strong>of</strong> C, the only possibility for r is a tile, say τ i j .<br />
It remains to show that´ijµτi j is a tiling <strong>of</strong>¢(with τ 00 at the origin <strong>of</strong> course). Take<br />
ij¾; we check that Right´τi jµLeft´τi·1jµ. We will use the points x ix i·1v iy j , <strong>and</strong> one more<br />
point t¾h. Let m be whichever <strong>of</strong> v iy j has greatest height in h. Since´x imhµp01r for some<br />
pr, there must exist tm with t¾h <strong>and</strong>´x ithµ10011. Since´111µ¾C, there is g¾H<br />
with t¾g <strong>and</strong>´x itgµ111. Note that v iy j¾g.<br />
We consider some triples formed from the above with h <strong>and</strong> g (see Fig. 4 in Appendix). First,<br />
we show that´x iv igµ111. Let´x iv igµpqr, say, for unique´pqrµ¾C. We know that<br />
6<br />
¯´x iv ihµ10011
jgµ<br />
Also,´x iv ihµ10011, so r1 as well. So we must have p1<br />
as required. In the same way we can successively show that´x i·1v igµ1·1 1 1 1 ,´x iy i j·1 1 τ i·1j . Therefore,´˘τ i j·1 1τi·1jµ¾C, <strong>and</strong> the cited properties <strong>of</strong><br />
C yield Left´τi·1jµRight´τi jµ, as required. That Up´τi jµDown´τij·1µis shown similarly.<br />
We have shown that T is a yes-instance <strong>of</strong> the cited tiling problem iffϕ T has a bundled Q CT L£F -<br />
model. The undecidability <strong>of</strong> the tiling problem now yields undecidability <strong>of</strong> satisfiability <strong>of</strong>ÉÌÄ£F -<br />
sentences.<br />
´x itgµ111, so q1.<br />
˘τ i j1τi j , <strong>and</strong>´x i·1y jgµ˘τ<br />
REMARK 5. A more general version <strong>of</strong> this result can be formulated in terms <strong>of</strong> products <strong>of</strong> propositional<br />
modal logics; see [9] <strong>and</strong> references therein. Namely, there is no decidable set L <strong>of</strong> formulas<br />
such thatâÌÄ£FLË¢ÌÄ£F . The same holds for the propositional bundled case as well.<br />
4 <strong>Decidable</strong> <strong>fragments</strong><br />
The ‘negative’ results <strong>of</strong> Section 3 can be ‘explained’ by the fact that all the <strong>undecidable</strong> <strong>fragments</strong><br />
there are in a sense ‘three-dimensional’, which is <strong>of</strong>ten a cause <strong>of</strong> bad computational properties. The<br />
three-variable fragment <strong>of</strong> classical <strong>first</strong>-<strong>order</strong> logic is <strong>undecidable</strong> even without equality [17], <strong>and</strong><br />
products <strong>of</strong> three propositional modal logics are usually <strong>undecidable</strong> [13]. In Theorem 4 we also have<br />
quantification in three dimensions: temporal operators, path quantifiers <strong>and</strong> the domain quantification.<br />
A natural way to reduce the interaction between the dimensions is to restrict <strong>first</strong>-<strong>order</strong> quantification<br />
to state formulas—that is to work with the language Q P CT L s —<strong>and</strong> to limit the scope <strong>of</strong> the<br />
path quantifiers (hence also <strong>of</strong> the temporal operators) to formulas with1 free variable; cf. [14, 24].<br />
DEFINITION 6 (MONODIC FORMULAS). Let Q P CT L<br />
s<br />
1<br />
be the set <strong>of</strong> all Q P CT L s -formulas ϕ such<br />
that any subformula <strong>of</strong> ϕ <strong>of</strong> the formψ, ψ 1Íψ 2 , or ψ 1Ëψ 2 has at most one free variable. Such<br />
formulas ϕ will be called monodic <strong>and</strong> Q P CT L<br />
s<br />
1<br />
the monodic fragment <strong>of</strong> Q P CT L s .<br />
It should be clear from the definition that Q P CT L<br />
s<br />
1<br />
contains full propositional CT L£<strong>and</strong> the<br />
full <strong>first</strong>-<strong>order</strong> (non-temporal) language. The latter means, in particular, that the monodic <strong>fragments</strong><br />
<strong>of</strong> the logics under consideration are still <strong>undecidable</strong>.<br />
The main aim <strong>of</strong> this section is to prove a satisfiability criterion for the monodic formulas (Theorem<br />
8) <strong>and</strong> then apply it in <strong>order</strong> to obtain various decidable <strong>fragments</strong> <strong>of</strong> FOBTLs. As in [14], the<br />
idea is to encode models in structures called quasimodels <strong>and</strong> then express the statement ‘there exists<br />
a quasimodel satisfying a given monodic sentence’ as a monadic second-<strong>order</strong> sentence.<br />
In what follows we assume that ϕ¾Q P CT L s 1<br />
. Denote by sub n´ϕµthe closure under negation<br />
<strong>of</strong> the set <strong>of</strong> state subformulas <strong>of</strong> formulas in ϕ containingn free variables; con´ϕµis the set <strong>of</strong><br />
constants in ϕ. Without loss <strong>of</strong> generality, we may identify ψ <strong>and</strong>ψ, so sub n´ϕµis finite. For every<br />
formula ψ´xµχ´xµin sub 1´ϕµwith a free variable x, we reserve a unary predicate P ψ´xµ, <strong>and</strong> for<br />
every sentence ψχ in sub 0´ϕµwe fix a propositional variable p ψ . P ψ´xµ<strong>and</strong> p ψ are called the<br />
surrogates <strong>of</strong> ψ´xµ<strong>and</strong> ψ, respectively. Given a state formula ψ, denote by ψ the result <strong>of</strong> replacing<br />
all its maximal subformulas <strong>of</strong> the formχ with their surrogates. Thus, ψ contains neither temporal<br />
operators nor path quantifiers at all—it is a P L-formula.<br />
Let x be a variable not occurring in ϕ <strong>and</strong> let sub x´ϕµψxy: ψ´yµ¾sub 1´ϕµ. Define a type<br />
for ϕ as a subset t <strong>of</strong> sub x´ϕ) such that ψχ¾t iff ψ¾t <strong>and</strong> χ¾t, for every ψχ¾sub x ϕ, <strong>and</strong><br />
ψ¾t iff ψ¾t, for every ψ¾sub x ϕ. Given a type t for ϕ <strong>and</strong> a constant c¾con´ϕµ, the pairtcis<br />
called an indexed type for ϕ. We will not be distinguishing between a type t <strong>and</strong> the conjunctionÎt<br />
<strong>of</strong> formulas in it.<br />
7
DP0c0<br />
Suppose that T is a set <strong>of</strong> types for ϕ, <strong>and</strong> T contc:c¾con´ϕµis a set <strong>of</strong> indexed types<br />
such thatt :tc¾T conT . Then the pairTT conis called a state c<strong>and</strong>idate for ϕ. Consider<br />
a <strong>first</strong>-<strong>order</strong> P L-structure<br />
(1)<br />
<strong>and</strong> suppose that a¾D. The set t´aµψ¾sub x´ϕµ:ψa℄is clearly a type for ϕ. Say that<br />
realizes a state c<strong>and</strong>idateTT conif Tt´aµ:a¾D<strong>and</strong> T conÅt´cµc«:c¾con´ϕµ.<br />
A state c<strong>and</strong>idate is realizable if somerealizes it.<br />
LetWbe an ω-tree. A state function for ϕ overis a map f associating with each w¾W<br />
a realizable state c<strong>and</strong>idate f´wµT wT wfor con ϕ.<br />
For every subformula ψ´yµ<strong>of</strong> ϕ, every full branch h in, every w¾h, <strong>and</strong> every function r<br />
mapping each w¾W to a type for ϕ, define inductively a ‘formula’ cond´ψhrwµ:<br />
¯cond´ψhrwµis ‘ψxy¾r´wµ’ if ψ´yµis a state formula;<br />
¯if ψ 1ψ 2 is not a state formula, cond´ψ 1ψ 2hrwµis ‘cond´ψ 1hrwµcond´ψ 2hrwµ’;<br />
¯ifψ is not a state formula, cond´ψhrwµis ‘cond´ψhrwµ’;<br />
¯cond´ψ 1Íψ 2hrwµis ‘vw´v¾hcond´ψ 2hrvµu¾´wvµcond´ψ 1hruµµ’;<br />
¯cond´ψ 1Ëψ 2hrwµis ‘vw´cond´ψ 2hrvµu¾´vwµcond´ψ 1hruµµ’.<br />
Since every Q P CT L s -formula is built from state formulas using only the Booleans <strong>and</strong> the temporal<br />
operators, this is well-defined. Let f be a state function for ϕ overW, with f´wµT wT con<br />
for w¾W . By a run in f we mean a function r from W into the setËw¾W T w such that<br />
¯r´wµ¾T w , for all w¾W ,<br />
¯for allψ¾sub x ϕ <strong>and</strong> w¾W, we haveψ¾r´wµiffh¿wcond´ψhrwµ.<br />
We call f a quasimodel for ϕ overif the following conditions hold:<br />
¯for each c¾con´ϕµ, the function r c defined by r c´wµt, fortc¾T con<br />
w , w¾W, is a run in f ,<br />
¯for all w¾W <strong>and</strong> t¾T w , there exists a run r in f such that r´wµt.<br />
Say that ϕ is satisfied in f if there are w¾W <strong>and</strong> t¾T w such that ϕ¾t.<br />
THEOREM 7. A Q P CT L s 1<br />
-sentence ϕ is satisfiable in a full tree model based onWiff it is<br />
satisfied in a quasimodel for ϕ over.<br />
Pro<strong>of</strong>. Similar to the pro<strong>of</strong> <strong>of</strong> Theorem 14 in [14]. For´µµ, we construct a quasimodel from a model<br />
in the obvious way. For´´µ, given a quasimodel f for ϕ overW, we take a cardinal κ <strong>and</strong><br />
form a set R <strong>of</strong> κ copies <strong>of</strong> each run in f . Since the language P L is countable <strong>and</strong> does not contain<br />
equality, it follows from classical model theory that if κ is large enough then each f´wµis realized by<br />
a modelw with domain R , <strong>and</strong> with tw´rµr´wµfor all r¾R. LetÅR´wµw¾W. We<br />
may now check by induction that for all state subformulas ψ´¯xµ<strong>of</strong> ϕ, all ¯r in R , <strong>and</strong> all hw, we have<br />
wψā℄iff´Åhwµψā℄for any h¿w. It follows that ϕ is satisfied inÅ.<br />
In the pro<strong>of</strong> <strong>of</strong> the next theorem we show that the existence <strong>of</strong> a quasimodel satisfying ϕ can be<br />
expressed as a monadic second-<strong>order</strong> formula, which yields the following satisfiability criterion:<br />
w<br />
8
THEOREM 8. Let LQ P CT L<br />
s 1<br />
<strong>and</strong> suppose that there is an algorithm that decides for any L-<br />
sentence ϕ whether an arbitrarily-given state c<strong>and</strong>idate is realizable. Then the satisfiability problem<br />
for L-sentences in both bundled <strong>and</strong> full tree models<br />
ψ´xµ<br />
is decidable.<br />
Pro<strong>of</strong>. We consider only full tree models, leaving the bundled case<br />
ψ´xµ<br />
to Lemma 1. For each ψ¾sub x´ϕµ,<br />
let R ψ be a unary predicate variable, <strong>and</strong> for each type t for ϕ, let<br />
R R<br />
χ t´xµψ¾t<br />
ψ¾sub x´ϕµÒt<br />
The formula χ t´xµsays that the R ψ´xµdefine the type t at x. For any subformula ψ´yµ<strong>of</strong> ϕ, an<br />
individual variable x, <strong>and</strong> a set variable h, define inductively a formula γ´ψhxµ<strong>of</strong> monadic second<strong>order</strong><br />
logic by taking γ´ψhxµR ψ´xµif ψ is a state formula, <strong>and</strong> for non-state formulas taking:<br />
¯γ´ψ 1ψ 2hxµγ´ψ 1hxµγ´ψ 2hxµ;<br />
¯γ´ψhxµγ´ψhxµ;<br />
ρx<br />
¯γ´ψ 1Íψ 2hxµy¾h´xyγ´ψ ¯γ´ψ 1Ëψ 2hxµy´yxγ´ψ<br />
con´xµ 2hyµz´xzyγ´ψ 1hzµµµ;<br />
2hyµz´yzxγ´ψ <br />
1hzµµµ.<br />
β´hxµγ´ψhxµ¡<br />
PTT χ t´xµ¡x x´ϕµRψ´xµ°h TT con¾Σ<br />
s´xµ<br />
ψ¾sub<br />
s¼´xµ ψρx<br />
t´xµ¡<br />
P<br />
ss¼P ϕ¾ËTx x<br />
PTT con´xµ<br />
s¼¾Σ<br />
R PTT con´xµχ t´xµµ tc¾T con<br />
PTT R ψ´ρχ<br />
Denote by β´hxµamonadic second-<strong>order</strong> formula saying that h is a full branch containing x. Let Σ<br />
be the set <strong>of</strong> all realizable state c<strong>and</strong>idates for ϕ (Σ can be constructed effectively), <strong>and</strong> let P s (s¾Σ)<br />
be a unary predicate variable. Then<br />
says that the R ψ´xµdefine a run through realizable state c<strong>and</strong>idates in Σ defined with the help <strong>of</strong> the<br />
P s . Finally, we define a monadic second-<strong>order</strong> sentence σ ϕ by:<br />
P<br />
s¾Σ<br />
sxs¾Σ<br />
t¾T<br />
TT c¾con´ϕµ<br />
con´xµ<br />
x´ϕµ<br />
con¾Σ<br />
ψ¾sub TT con¾Σ<br />
TT con¾Σ<br />
t¾T<br />
ψ¾sub x´ϕµ<br />
It is not hard to check thatσ ϕ iff ϕ is satisfied in a quasimodel over. It remains to recall<br />
that the monadic second-<strong>order</strong> theory <strong>of</strong> countably <strong>branching</strong> trees is decidable (this can easily be<br />
shown by reduction to the monadic second-<strong>order</strong> theory <strong>of</strong> two successor functions, which is decidable<br />
[21]). A Löwenheim–Skolem–Tarski argument (see Theorem 14 <strong>of</strong> the Appendix) will show that any<br />
satisfiable Q P CT L 1 s -formula has a model based on a countably <strong>branching</strong> tree.<br />
As a consequence we obtain<br />
9
THEOREM 9. The following <strong>fragments</strong> are decidable:<br />
¯the two-variable fragment <strong>of</strong>ÉÈÌÄ£Q P CT L s 1<br />
;<br />
¯the two-variable fragment <strong>of</strong>ÉÈÌÄ£Q P CT L s 1<br />
;<br />
¯the monadic fragment <strong>of</strong>ÉÈÌÄ£Q P CT L s 1<br />
;<br />
¯the monadic fragment <strong>of</strong>ÉÈÌÄ£Q P CT L s 1<br />
;<br />
¯the guarded fragment <strong>of</strong>ÉÈÌÄ£Q P CT L s 1<br />
;<br />
¯the guarded fragment <strong>of</strong>ÉÈÌÄ£Q P CT L p 1 .<br />
A similar construction yields the following theorem answering a question that has puzzled many<br />
temporal logicians since [11]:<br />
THEOREM 10. It is decidable whether an arbitrarily-given propositional P CT L£-formula is satisfiable<br />
in a ‘non-local’ full tree model—where truth values <strong>of</strong> atoms may depend on the branch <strong>of</strong><br />
evaluation. The same holds for non-local bundled tree models.<br />
5 Conclusion<br />
This paper may be regarded as a beginning <strong>of</strong> systematic research into the computational behavior<br />
<strong>of</strong> <strong>first</strong>-<strong>order</strong> <strong>branching</strong> time logics. We have obtained decidability results <strong>of</strong> two kinds. The most<br />
striking bad news is that the one-variable <strong>fragments</strong> <strong>of</strong> logics containingÉÌÄ£F <strong>and</strong>ÉÌÄ£F are<br />
<strong>undecidable</strong>, which contrasts with the situation in many other modal <strong>and</strong> temporal <strong>first</strong>-<strong>order</strong> logics<br />
(cf. [14, 24]). The good news is that there are still ways <strong>of</strong> obtaining decidable <strong>fragments</strong> with nontrivial<br />
interaction between <strong>first</strong>-<strong>order</strong> quantifiers, path quantifiers <strong>and</strong> temporal operators. In this<br />
paper, we considered the case when the <strong>first</strong>-<strong>order</strong> quantifiers are applied only to state formulas, while<br />
the path-quantifiers <strong>and</strong> temporal operators are applicable to formulas with1 free variable.<br />
Another way to obtain decidable <strong>fragments</strong> is to restrict further the formulas that can have free<br />
variables. For example, using quasimodels <strong>and</strong> a mosaic technique one can prove a result similar to<br />
Theorem 8 for bundled models <strong>and</strong> the fragment <strong>of</strong> Q CT L£in whichmay be applied to formulas<br />
with at most one free variable,Í,<strong>and</strong>are applicable only to sentences, <strong>and</strong> there are no restrictions<br />
on <strong>first</strong>-<strong>order</strong> quantification (for a precise formulation see [15]).<br />
Two other promising ways are to allow arbitrary <strong>first</strong>-<strong>order</strong> quantification, but to restrict applications<br />
<strong>of</strong> path quantifiers to closed formulas <strong>and</strong> those <strong>of</strong> linear temporal operators to formulas with<br />
1 free variable, <strong>and</strong> vice versa. However, they remain open for investigation.<br />
Acknowledgements<br />
The work <strong>of</strong> the <strong>first</strong> <strong>and</strong> third authors was partially supported by UK EPSRC grant GR/R45369/01<br />
“Analysis <strong>and</strong> mechanisation <strong>of</strong> decidable <strong>first</strong>-<strong>order</strong> temporal logics” (the third author was also partially<br />
supported by grant 99-01-00968 from the Russian Foundation for Basic Research). The work<br />
<strong>of</strong> the second author was supported by Deutsche Forschungsgemeinschaft (DFG) grant Wo583/3-1.<br />
We are grateful to S. Bauer, A. Kurucz, A. Rabinovich, M. Reynolds, C. Stirling, <strong>and</strong> A. Zanardo for<br />
stimulating discussions <strong>and</strong> comments.<br />
10
References<br />
[1] H. Andréka, J. van Benthem, <strong>and</strong> I. Németi. Modal languages <strong>and</strong> bounded <strong>fragments</strong> <strong>of</strong> predicate<br />
logic. J. Philosophical Logic, 27:217–274, 1998.<br />
[2] R. Berger. The undecidability <strong>of</strong> the domino problem, volume 66 <strong>of</strong> Memoirs. Amer. Math. Soc.,<br />
Providence, RI, 1966.<br />
[3] E. Börger, E. Grädel, <strong>and</strong> Yu. Gurevich. The Classical Decision Problem. Perspectives in<br />
Mathematical Logic. Springer, 1997.<br />
[4] J. Burgess. The unreal future. Theoria, 44:157–179, 1978.<br />
[5] J. Burgess. Logic <strong>and</strong> time. J. Symbolic Logic, 44:566–582, 1979.<br />
[6] E. Clarke, E. Emerson. Synthesis <strong>of</strong> synchronization skeletons for <strong>branching</strong> time temporal<br />
logic. In Proc. IBM Workshop on Logic <strong>of</strong> Programs, pages 52–71, Springer-Verlag, 1981.<br />
[7] E. Emerson. Temporal <strong>and</strong> modal logic. In J. van Leeuwen, editor, H<strong>and</strong>book <strong>of</strong> Theoretical<br />
Computer Science, pages 996–1076, 1990.<br />
[8] E. Emerson <strong>and</strong> J. Halpern. ‘Sometimes’ <strong>and</strong> ‘not never’ revisited: on <strong>branching</strong> versus linear<br />
time. J. ACM, 33:151–178, 1979.<br />
[9] D. Gabbay, A. Kurucz, F. Wolter, <strong>and</strong> M. Zakharyaschev. Many-Dimensional Modal Logics:<br />
Theory <strong>and</strong> Applications. Elsevier, 2002. SeeØØÔ»»ÛÛۺ׺кºÙ»×Ø»ÑÞ.<br />
[10] D. Gabbay, M. Reynolds, <strong>and</strong> M. Finger. Temporal Logic: Mathematical Foundations <strong>and</strong><br />
Computational Aspects, Volume 2. Oxford University Press, 2000.<br />
[11] Y. Gurevich <strong>and</strong> S. Shelah. The decision problem for <strong>branching</strong> time logic. J. Symbolic Logic,<br />
50:668–681, 1985.<br />
[12] R. Hirsch <strong>and</strong> I. Hodkinson. Representability is not decidable for finite relation algebras. Trans.<br />
Amer. Math. Soc., 353:1403–1425, 2001.<br />
[13] R. Hirsch, I. Hodkinson, <strong>and</strong> A. Kurucz. On modal logics betweenââÃ<strong>and</strong>ˢˢË.<br />
J. Symbolic Logic, 2001. (In press.)<br />
[14] I. Hodkinson, F. Wolter, <strong>and</strong> M. Zakharyaschev. <strong>Decidable</strong> <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> temporal<br />
logics. Annals <strong>of</strong> Pure <strong>and</strong> Applied Logic, 106:85–134, 2000.<br />
[15] I. Hodkinson, F. Wolter, <strong>and</strong> M. Zakharyaschev. Monodic <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> temporal<br />
logics: 2000–2001 A.D. In R. Nieuwenhuis <strong>and</strong> A. Voronkov, editors, Logic for Programming,<br />
Artificial Intelligence <strong>and</strong> Reasoning, number 2250 <strong>of</strong> LNAI, Springer, 2001, pages 1–23.<br />
[16] R. Ladner <strong>and</strong> J. Reif. The logic <strong>of</strong> distributed protocols. In J. Halpern, editor, Theoretical<br />
Aspects <strong>of</strong> Reasoning about Knowledge, pp.207–222, Morgan Kaufmann, 1986.<br />
[17] R. Maddux. The equational theory <strong>of</strong> CA 3 is <strong>undecidable</strong>. J. Symbolic Logic, 45:311–316, 1980.<br />
[18] A. Prior. Past, Present <strong>and</strong> Future. Oxford University Press, 1967.<br />
11
[19] A. Prior. Now. Nous, 2:101–119, 1968.<br />
[20] O. Øhrstrøm <strong>and</strong> P.F.V. Hasle. Temporal Logic: From Ancient Ideas to Artificial Intelligence.<br />
volume 57 <strong>of</strong> Studies in Logic <strong>and</strong> Philosophy. Kluwer Academic Publishers, 1996.<br />
[21] M. O. Rabin. Decidability <strong>of</strong> second <strong>order</strong> theories <strong>and</strong> automata on infinite trees. American<br />
Mathematical Society Transactions, 141:1–35, 1969.<br />
[22] S. Woelf. Combinations <strong>of</strong> tense <strong>and</strong> modality for predicate logic. J. Philosophical Logic,<br />
28:371–398, 1999.<br />
[23] M. Wooldridge <strong>and</strong> M. Fisher. A <strong>first</strong>-<strong>order</strong> <strong>branching</strong> time logic <strong>of</strong> mutli-agent systems. In<br />
B. Neumann, editor, Proc. 10th European Conference on Artificial<br />
SeeØØÔ»»ÛÛۺ׺кºÙ»<br />
Intelligence (ECAI92), pages<br />
234 – 238, Chichester, 1992. John Wiley <strong>and</strong> Sons.<br />
[24] F. Wolter <strong>and</strong> M. Zakharyaschev. <strong>Decidable</strong> <strong>fragments</strong> <strong>of</strong> <strong>first</strong>-<strong>order</strong> modal logics. J. Symbolic<br />
Logic, 66:1415–1438, 2001.<br />
[25] F. Wolter <strong>and</strong> M. Zakharyaschev. Axiomatizing the monodic fragment <strong>of</strong> <strong>first</strong>-<strong>order</strong> temporal<br />
logic. Annals <strong>of</strong> Pure <strong>and</strong> Applied Logic (In print.)<br />
×Ø»ÑÞ.<br />
[26] A. Zanardo. Branching-time logic with quantification over branches: the point <strong>of</strong> view <strong>of</strong> modal<br />
logic. J. Symbolic Logic, 61:1–39, 1996.<br />
12
Appendix<br />
1<br />
2¾·FxP´xµ<br />
A monodic Q C T L£F -sentence with only uncountable models<br />
4xyz´´xxµ´xyyzxzµµ<br />
P´xµ¿·F¾·Fy´P´yµyxµ¿·F¾·Fy´P´yµyxµ¡<br />
P´yµ¿·F¾·Fz´P´zµzyµxy℄¿·F¾·Fz´P´zµzyµxy℄¡<br />
1xy 3¾·Fx<br />
We take a signature with relation symbols P (unary) <strong>and</strong>(binary). Let ϕ be the conjunction <strong>of</strong> the<br />
following Q CT L£F-sentences:<br />
ϕ is monodic <strong>and</strong> only uses the reflexive temporal operators¾·F¿·F . The idea is that ϕ forces an<br />
embedding <strong>of</strong> 2 ω full branches into the domain.<br />
LEMMA 11. ϕ has a full tree model.<br />
Pro<strong>of</strong>. Letω 2be<br />
<br />
the tree withthe <strong>order</strong>ing <strong>of</strong> proper initial segment. We identify full<br />
branches <strong>of</strong>with elements<br />
Ð <br />
<strong>of</strong> ω 2. Let Dω 2. Define a linear<br />
ËË Ð¸<br />
<strong>order</strong>on D in the st<strong>and</strong>ard way.<br />
First, the nodes <strong>of</strong>are <strong>order</strong>ed ‘left–right’ by ‘projecting’ down onto a horizontal axis, as shown in<br />
Figure 1.<br />
<br />
000 001 010 011 100 101 110 111<br />
00<br />
01<br />
10 11<br />
0<br />
1<br />
¹<br />
ε<br />
Figure 1: Ordering the nodes <strong>of</strong>ω 2<br />
The full branches h <strong>of</strong>are then inserted in this <strong>order</strong>ing in their natural places: h goes to the left <strong>of</strong><br />
all t¾ω 2 such that tˆ0¾h, <strong>and</strong> to the right <strong>of</strong> all t¾ω 2 such that tˆ1¾h. This defineson D.<br />
Now for each t¾ω 2 let M t be the structure with domain D, withinterpreted as above, <strong>and</strong><br />
with M tP´dµiff P´yµ¿·F¾·Fz´P´zµzyµxy℄¿·F¾·Fz´P´zµzyµxy℄¡<br />
dt, for d¾D. We obtain a modelÅ´D´M t : t¾ω 2µµ.<br />
Write ε for the root <strong>of</strong>. We check that´Åhεµϕ for any h. For sentence 1 above, let h be<br />
any branch <strong>of</strong>. We require<br />
´Åhεµxy ¿·F<br />
We will take ‘x’ to be h¾D. Then for any d¾D with´Åhεµ¿·F P´dµ, we require´Åhεµ<br />
¿·F¾·Fz´P´zµzdµhd. Such a d must be inω 2. Suppose that there is u¾h with<br />
M vz´P´zµzdµfor all v¾h with vu. Choose v¾h higher than d. Since M vP´vµ, we<br />
¸¸¸<br />
13
´Åhtµ¿·F¾·Fy´P´yµytµ (2)<br />
have vd. By definition <strong>of</strong>we have vd iff hd. So hd, as required. The other side (xy)<br />
is similar.<br />
Sentence 2,¾·FxP´xµ, is clearly true, since M tP´tµfor any t¾ω 2.<br />
so´Ågtµ<br />
For sentence 3,¾·Fx P´xµ¿·F¾·Fy´P´yµyxµ¿·F¾·Fy´P´yµyxµ¡, take<br />
h, t¾h, d¾D. If´ÅhtµP´dµthen dt, so we can assume td. We require<br />
<strong>and</strong> similarly on the other side with yt. So take g to be any branch containing tˆ0. By definition <strong>of</strong><br />
, every utˆ0 insatisfies ut in D. Thus,´Ågtˆ0µ¾·Fy´P´yµytµ, ¿·F¾·Fy´P´yµytµ, yielding (2). The second half <strong>of</strong> sentence 3 is h<strong>and</strong>led similarly, using tˆ1.<br />
Sentence 4 is trivially true.<br />
LEMMA 12. Any full tree model <strong>of</strong> ϕ has uncountable domain.<br />
Pro<strong>of</strong>. Assume that for some ω-treeW,Å´D´M w : w¾Wµµis a full tree model <strong>of</strong> ϕ.<br />
We can assume that ϕ is true at the root <strong>of</strong>. By sentence 2, for all w¾W we may pick d w¾D with<br />
M wP´d wµ.<br />
We define nodes w t¾W (for t¾ω 2) by induction ont. We let w ε be the root <strong>of</strong>, where ε<br />
is the empty sequence inω 2. Let t¾ω 2 <strong>and</strong> assume that w t is defined. By sentence 3, there is a<br />
full branch h <strong>of</strong>with w t¾h <strong>and</strong>´Åhw tµ¿·F¾·Fy´P´yµyd wtµ. So there is w tˆ0¾h<br />
with w tˆ0w t <strong>and</strong>´Åhw tˆ0µ¾·Fy´P´yµyd wtµ. Similarly, there are g¿w tˆ1w t with<br />
´Ågw tˆ1µ¾·Fy´P´yµyd wtµ.<br />
For each full branch β <strong>of</strong>ω 2, pick a full branch ˆβ <strong>of</strong>containingw t : t¾β. Sentence<br />
βy℄¡<br />
1 tells<br />
us that there is d β¾D with<br />
´Åˆβw εµy ¿·F P´yµ¿·F¾·Fz´P´zµzyµd βy℄¿·F¾·Fz´P´zµzyµd Let βγ be distinct branches <strong>of</strong>ω 2; we claim that d βd γ . Let t be the highest node in βγ. We<br />
may assume without loss <strong>of</strong> generality that tˆ0¾β <strong>and</strong> tˆ1¾γ. Now,´Åˆβw wt<br />
P´yµ<br />
εµy ¿·F ¿·F¾·Fz´P´zµzyµd βy℄¡. We have´Åˆβw εµ¿·F P´d wtµsince M wtP´d wtµ. Hence,<br />
´Åˆβw εµ¿·F¾·Fz´P´zµzd wtµd wtµ<br />
βd By definition <strong>of</strong> w tˆ0 , we have´Åˆβw tˆ0µ¾·Fy´P´yµyd wtµ. So certainly,<br />
´Åˆβw εµ¿·F¾·Fz´P´zµzd Hence,´Åˆβw εµd βd wt . That is, M wεd βd wt Similarly, M wεd γd wt . By sentence 4,<br />
is transitive in M wε , so M wεd βd γ . Sinceis irreflexive, d βd , as claimed.<br />
Thus,Dd β : β a branch <strong>of</strong>ω 22 , proving the lemma.<br />
ω<br />
γ<br />
REMARK 13. We can make any model <strong>of</strong> ϕ have uncountable <strong>branching</strong> factor by adding a conjunct<br />
¾·Fx´Q´xµy´yxyxQ´yµµµplus the statement thatis rigid.<br />
14
2 Q P C T L£-formulas with countable models<br />
On the other h<strong>and</strong>, if we restrict quantifiers to apply only to state formulas, we get a downward<br />
Löwenheim–Skolem–Tarski theorem:<br />
THEOREM 14. Let ϕ be a Q P CT L s -sentence, <strong>and</strong> suppose that ϕ is satisfiable in a full tree model.<br />
Then ϕ is satisfiable in a full tree model with countable tree <strong>and</strong> domain.<br />
Pro<strong>of</strong>. LetÅbe any full tree model. We may viewÅ´DIµas a three-sorted <strong>first</strong>-<strong>order</strong> structure,<br />
with sorts for the domain, the tree, <strong>and</strong> the set <strong>of</strong> all full branches. Taking a countable elementary<br />
substructure <strong>of</strong> this yields a bundled tree modelÆ´0D 0HI 0µwhose domain D 0 , tree0, <strong>and</strong><br />
bundle H are countable. It is easy to translate Q P CT L£-formulas to three-sorted <strong>first</strong>-<strong>order</strong> formulas<br />
with the same meaning. It follows that for any ā in D 0 , h¾H , w¾h, <strong>and</strong> any Q P CT L£-formula<br />
ψ´¯xµ, we have´Åhwµψ´āµiff´Æhwµψ´āµ.<br />
Now letÆ´0D 0I 0µbe<br />
´Ågwµϕ´āµ´µ´Ægwµϕ´āµ<br />
the full tree model based onÆ. We claim that for all formulas ϕ´¯xµ<br />
as in the formulation <strong>of</strong> the theorem, all full branches g <strong>of</strong>0 (g may not be in H ), all w¾g, <strong>and</strong> all<br />
ā in D 0 , we have<br />
The pro<strong>of</strong> is by induction on ϕ. The atomic, boolean, <strong>and</strong> temporal cases are easy <strong>and</strong> we omit<br />
them. Consider the caseϕ´¯xµ<strong>and</strong> inductively assume the result for ϕ. If´Ægwµϕ´āµ, it is<br />
easily seen that´Ågwµϕ´āµ. Conversely, if´Ågwµϕ´āµ, pick h¾H containing w.<br />
Clearly,´Åhwµϕ´āµ, so´Æhwµϕ´āµ. So there is h¼¾H with´Æh¼wµϕ´āµ. Thus,<br />
´Åh¼wµϕ´āµ. Inductively,´Æh¼wµϕ´āµ. So´Ægwµϕ´āµ, as required.<br />
Finally consider the casexϕ´xȳµ, for a state formula ϕ for which we assume the result inductively.<br />
If´Ægwµxϕ´xāµ, then´Ægwµϕ´bāµfor some b¾D 0 . Inductively,´Ågwµ<br />
ϕ´bāµ, so´Ågwµxϕ´xāµ. Conversely, suppose that´Ågwµxϕ´xāµ. Pick h¾H containing<br />
w. Then asxϕ´xȳµis a state formula,´Åhwµxϕ´xāµ. So´Æhwµxϕ´xāµ,<br />
whence´Æhwµϕ´bāµfor some b¾D 0 . Then´Åhwµϕ´bāµ. Since ϕ is a state formula,<br />
so´Ægwµxϕ´xāµas<br />
´Ågwµϕ´bāµ. Inductively,´Ægwµϕ´bāµ,<br />
3 Diagrams for Theorem 4<br />
required.<br />
Up´τµ<br />
Down´τµ<br />
Left´τµ Right´τµ<br />
τ<br />
Figure 2: a tile τ<br />
15
Ú´yµ<br />
Ô<br />
Ú×<br />
Ô<br />
Ú ×<br />
ÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔ Ô ÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔÔ Ú Ô× Ú×<br />
× × ×<br />
× × × ×<br />
Ú × Ú<br />
Ú×<br />
× × ×<br />
x 2 u 2 x 1<br />
× × ×<br />
Ú×<br />
<br />
2001τ 02<br />
2001τ 01<br />
2002·1 2<br />
2002·1 2<br />
10011 1001 1 1<br />
v 0<br />
2002·1 2<br />
0<br />
2001τ<br />
10 u 12001τ 00 y20022<br />
2001τ 10<br />
0<br />
10011 1001 1<br />
v 1<br />
1<br />
2002·1 2 2001τ 0 1<br />
20022 y 1<br />
10011 1001 1 1<br />
v 2<br />
y 2001τ<br />
2<br />
2<br />
20022<br />
1<br />
y 2<br />
v 1<br />
y 1<br />
Figure 3: The defined points<br />
10011 1001 1 1<br />
20022<br />
x 0 u 0 x 1 u 1<br />
x 2<br />
¹ ´xµ<br />
0<br />
16
±±±±±±±±± ×<br />
±±±±±±±±±<br />
× × ±±±±±±±±±<br />
±±±±±±±±± ±±±±±±±± ± × ×<br />
±±±±<br />
×<br />
×<br />
× ×<br />
histories H<br />
h<br />
tree T<br />
××<br />
t<br />
v i<br />
10011<br />
y jm<br />
g<br />
10011<br />
2001τ i j<br />
111<br />
˘τ i j1τ i j<br />
111<br />
2001τ i·1j<br />
1001 1 1<br />
domain D<br />
x i x i·1<br />
˘τ i j·1 1 τ i·1j<br />
1·1 1 1 1<br />
Figure 4: Checking Right´τ i jµLeft´τ i·1jµ.<br />
17