Modal logic with names - PhilArchive

G E O R G E G A R G O V AND V A L E N T I N G O R A N K O

MODAL

LOGIC

WITH

NAMES

ABSTRACT. We investigate an enrichment of the propositional modal language with a "universal" modality n having semantics x ~ iq~ iff Vy(y ~ ~0),and a countable set of "names" - a special kind of propositional variables ranging over singleton sets of worlds. The obtained language ~cqc proves to have a great expressive power. It is equivalent with respect to modal definability to another enrichment s176 of L*~ where [ ] is an additional modality with the semantics x ~ []~o iff V y ( y r x ~ y ~ q~). Modeltheoretic characterizations of modal definability in these languages are obtained. Further we consider deductive systems in ~c. Strong completeness of the normal s logics is proved with respect to models in which all worlds are named. Every ~~ axiomatized by formulae containing only names (but not propositional variables) is proved to be strongly frame-complete. Problems concerning transfer of properties ([in]completeness, filtration, finite model property etc.) from ~ to ~ are discussed. Finally, further perspectives for names in multimodal environment are briefly sketched.

1. I N T R O D U C T I O N In a p r o p o s i t i o n a l setting ( m o d a l , t e m p o r a l , d y n a m i c , etc.) n a m e s are p r o p e r t i e s t h a t identify the i n t e n s i o n a l objects completely, i.e. each o f t h e m h o l d s o n l y for a single object (be it a possible world, a t e m p o r a l state o r a d a t a f r a g m e n t , d e p e n d i n g on the semantic i n t e r p r e t a t i o n o f the l a n g u a g e ) a n d can be thus used to n a m e t h a t p a r t i c u l a r object. H i s t o r i c a l l y the idea o f such n a m i n g variables c a n be t r a c e d b a c k to the p i o n e e r i n g w o r k s o f P r i o r (1956) a n d especially o f Bull (1970) w h o i n t r o d u c e d t e m p o r a l reference to p a r t i c u l a r m o m e n t s (or to the c o r r e s p o n d i n g state o f affairs) by m e a n s o f special variables (the c l o c k - v a r i a b l e s o f Prior) which are true at a single time instant. I n Bull (1970) one can find an a x i o m a t i z a t i o n o f a t e m p o r a l system in the l a n g u a g e e x t e n d e d also by a universal m o d a l i t y (i.e. a m o d a l i t y related s e m a n t i c a l l y to the universal relation). In the a r e a o f a p p l i c a t i o n o f p r o p o s i t i o n a l logic to c o m p u t e r science the idea o f n a m e s was first e x p l o r e d ( i n d e p e n d e n t l y o f the d e v e l o p m e n t s in t e m p o r a l logic) by Passy a n d T i n c h e v (1985). In P a s s y a n d T i n c h e v (1991) there are m a n y e x a m p l e s which s h o w t h a t Journal o f Philosophical Logic 22:607 - 636, 1993. 9 1993 Kluwer Academic Publishers. Printed in the Netherlands.

608

GEORGE GARGOV AND VALENTIN GORANKO

this enrichment is quite appropriate for the treatment of problems arising from, or justified as arising from, possible applications (cf. Gargov and Passy (1988) as well). A first attempt to apply named languages to traditional modal logic was made in the main origin of the present paper (Gargov et al., 1987) where the minimal normal modal logic with names was axiomatized and several further properties were stated. Let us note the use of necessity and possibility forms (instead of the universal modality) in the latter paper, borrowed from the treatise on programming logics in Goldblatt (1982). We now find this area ripe for a systematic treatment going from a collection of interesting but loosely connected cases to a kind of a general theory. In our work we try to give a coherent exposition of the theory of names in purely modal environment in order to emphasize just on the effects which names yield. We have chosen to reinstall the universal modality for many reasons, both technical and aesthetical. After the preliminaries we start with expressiveness of the new language ~~Cin Section 3. In Section 4 we investigate modal definability in 5~. The language with names turns out to be equivalent in this respect to another modal language enriched with an additional (recently actively investigated) modality over the inequality relation - the so called difference operator. This equivalence allows for a uniform characterization of modal definability in the spirit of Goldblatt and Thomason (1974). A good demonstration of the expressive power of 2~o is the result that every finite frame is definable in ~ by means of a single pure (not containing propositional variables) formula only. In Section 5 we turn to deductive systems and problems of completeness in ~c. The minimal normal ~, denoted 91 ~ cp[x], if x ~ V(cp); ~0 is valid in 91, denoted 91 ~ q~, if V(~o) = W. D E F I N I T I O N . A n a m e d model ( W, R, V, )~> is surjective if Z is surjective (which, of course, implies NWIt ~ N0). # We say that q~ ~ 5~ is v a l i d i n a m o d e l g J l = if for every Z: C ~ W, (932, X> ~ ~0. N o w validity in a f r a m e is defined as a validity in all models over the frame. Also the notions o f general frame and validity in it are accordingly a d a p t e d in ~ . T h r o u g h o u t the p a p e r we will freely use metavariables: F, G for frames, 5, 15 for general frames, 9]2 for models, 91 for n a m e d models.

3. EXPRESSIVENESS AND F I R S T - O R D E R D E F I N A B I L I T Y IN ~,. There are (at least) two natural extensions to L, ce~ of the translation S T (see van Benthem (1986)) of the m o d a l formulae into the first-order language L1 containing a binary predicate R and a countable set of unary predicates {P0, P1 . . . . }: I. S T " is the standard S T for 5f and obeys the following additional clauses (cf. G a r g o v et al., 1988):

(i) S T ' ( c i ) = ( x = y~) where { Y0, Yl, 99- } is a countable set of individual variables, especially assigned for presenting the names in Lt and x is a fixed variable, different f r o m Y0, Y~. . . . (ii) ST'(Ilcp) = g y S T ' ( c p ) [ y / x ] where y doesn't occur in ST'((p) and x is the variable, fixed to be the only free variable (if any) besides the y's in ST'(~o). N o w one can express the validity in a n a m e d model. Let 9~ = , where 99~ = ~ VxST'(~o). Now, for validity in a model we have: 99l ~ q) iff 9J/' ~ YxVy~a . . . V y ~ S T ' ( q ) ) where % . . . . , % are the names occurring in (p. Let us denote the

612

GEORGE

GARGOV AND VALENTIN

GORANKO

formula Yyil . . . Vy, ST'(q)) by ST(q)). This gives a complete analogy with S T in the standard language ~ and validity in a frame is defined in the same way as for 5r - with the universal closure of the latter formula over all occurring P's. II. Define ST" as follows: let x be a fixed variable;

(i)

P2i+lX; (ii) ST"(ei) = P2ix; (iii) the remaining clauses are the same as above. N o w let 9l = (W, R, V, X) be a named model. Define a corresponding Ll-model 91' = ( W ; R, P0, PI, - 99) such that Pzi = {7~(ci)} and Pzi+l = V(pg) for each i e N. Then 91 ~ q) iff 91' ~ VxST"(q)). For models ST" gives: ifg)~ = (W, R, V ) and ~ ' is obtained as 91' above, evaluating all P2i arbitrarily, then ST"(p,)

'-/

=

9J/~ q) iff 9J/' ~ V x V P 2 q . . .

VP2ik(3!zPzilZ A

...

A

3 ! z P z j ~ ST"(q))) where Q , . . . , cik are the names occurring in ~0. N o w it is natural to denote by ST(o) the formula

V P z i l . . . VPzik(3!zP2i~z /x . . .

A 3!zPzikZ --* ST"(~o)).

It is provably equivalent to the above S T and thus we have again an analogy with S T in 2Z. Finally, validity in a frame is expressed by the universal closure of ST(q)) over all P2~+1 such that Pi occurs in q). The former translation is preferable from the point of view of simplicity, however it is a bit ad hoc while the latter one reflects adequately the idea of names over definable sets. P R O B L E M 1. Find model-theoretic characterizations (in the style of van Benthem (1986)) of the fragments of Lj corresponding to each of the translations above. Is the problem whether an Ll-formula is equivalent to such a translation of a modal formula decidable? Now, a few words about first-order definability. A formula q) ~ 5~ is said to be first-order definable if the class FR(q)) = {F/F ~ q)} is definable by a formula of the first-order language L0 for structures (W, R ) .

MODAL LOGIC WITH NAMES

613

First-order definability is not decidable even for 50 (see Chagrova (1989)) that is why no exact syntactic characterization of this property exists. However, there are several strong and useful sufficient syntactic conditions, e.g., the so called "Sahlqvist forms" and some generalizations (see Sahlqvist (1975), van Benthem (1986)). It is a routine procedure to check that, as a rule, they apply without further ado to 50 ~. Provided V(r = V(Y(~b)) we shall prove that V ( ~ r = V(r'(@ r v(r'( V ( r ( ~ q))) = V ( ~ cp) # 0 hence V(a(q0) # W. Vice versa, let 9J/y o-(q)), i.e. V(~(~ (p)) # 0. Hence, according to Lemma 4.3, V(z(-n q))) = V ( ~ q)), and therefore V(~o) # W. # C O R O L L A R Y 4.5. For every frame F and q) ~ L([-~--]),F ~ ~o l f f F ~r(~o).

#

So, definability in ~ ( [ ~ ) is not stronger than that in L~. II. The opposite is true as well. Indeed, one can easily enforce a propositional variable p to serve as a name in 5e(TCq), putting as an antecedent O ( p /~ ~ - l p ) . More formally, we shall define a translation 7c: 5r ~ ~([-#--1). First the names are coded with variables, by 4: P w C ~ P as follows: ~(Pi) =Dr P2i and ~(ci) =DF P2~+1 for each i E ~. r is accordingly extended to r for each formula ~0 ~ ~c. For each q ~ P u C, put v(q) =Dr O(q /X [-~-l~q). Now let q~ ~ ~ and d~ . . . . . dk be the names occurring in q). Put ~(q0 = ~ v(~(d,)) A . . . A v ( ~ ( d ~ ) ) ~ ~(~o). This syntactic translation yields a semantic one. Let 9J/ = (W, R, V> be a model. We define a model 991~ = (W, R, V~ > where V~ (p) = V(~ l(p)) for each p e P. The following two propositions are easy exercises. P R O P O S I T I O N 4.6. Let 9~ be a model. (1) For every p ~ P, ~ ~ v(p) if V(p) is a singleton and gJl ~ -nv(p) otherwise. (2) For every c ~ C, 93l ~ v(c). #

617


P R O P O S I T I O N 4.7. I f ( W, R, V, ~) is a named model, go ~ ~ V~ is the valuation defined as above then V((p) = V~ (~(q~)).

and

#

For convenience we will use a notion of 2#~-valuation as a valuation both for the variables and names. T H E O R E M 4.8. Let F = ( W, R ) and qo ~ ~c. Then: ( 1 ) f o r every ~c-valuation V in F, V(~o) -= Ve (~((p)); (2) f o r every valuation V in F there exists an ~c-valuation V~ in F such that V~(~o) ~ V(~(cp)). Proof. (1) Let V be an fc-valuation. Then V, (~(d~)) is a singleton, hence Ve (v(~(d~))) = W for i = 1. . . . . k. Therefore Ve (v(~(d l))) /~ ...

A

V~(v(~(d~)))

=

W, so V~(~(~o))

=

V~(~(q~))

=

V(~o).

(2) Let x r V(~(~o)), i.e. x ~ V(v(~(d~))) A . . . /~ V(v(~(dk))) and x r V(~(q0)) hence x ~ V(v(~(di))) for i = 1. . . . . k and, in virtue of 4.6, V(~(d~)) is a singleton, hence V(v(~(d~))) = W. Therefore V(v(~(d~))) A . . . /X V(v(~(dk))) = W whence V(~(cp))= V(~(~0)). Then we can define an 5~ Vo as follows: { V(~(q)) V~(q) =~F

{W}

if ~(q) occurs in rc(cp), otherwise,

where w is an arbitrary fixed element of W and q e P w C. V~ is an # We-valuation and V,(cp) = V(~(cp)) = V(~(q))). So x r V,(cp). C O R O L L A R Y 4.9. Let F = ( W, R ) , x ~ W and q) ~ Sc. Then." (1) F g g0[x] l f f F ~ rc((p)[x]; (2) F ~ q~ / f f F ~ ~(~o).

#

The last result shows that the definability in 5~ is not stronger than that of S([Z]). So we have T H E O R E M 4.10. The languages ~ respect to modal definability.

and S([-~) are equivalent with

4.2. M o d a l definability in 5e([~) We shall characterize modal definability in 2'([5[]) in the modeltheoretic style of Goldblatt and Thomason (1974). All statements

#

618


without proofs below are obtained by simple calculations based on corresponding results from Goranko (1990) where languages containing modalities over a relation (in our case equality) and its complement are investigated. First, we shall define the analog in Lf(Vr of the notion of SA-construction, introduced by Goldblatt and Thomason. D E F I N I T I O N (cf. Goranko (1990), 3.7, 3.10). F' = ( W ' , R ' ) is an C-collapse of the general frame ~ = (W, R, W ) iff F ' is a substructure of F = (W, R ) (i.e. W _ W' and R = R' ~ W2), and there exists a general subframe (5 of ~ such that F '+ ~ (5+ and for each x ~ W', R(x) c_ [R'(x)]~+ where [X]~+ is the least element of (5+, containing X. # In particular, when ~ = (W, R, .r C-collapse of the frame (W, R).

we obtain a definition of

DEFINITION. General ultraproduct of frames is an ultraproduct of the corresponding full general frames (see, e.g. van Benthem (1986) or Goldblatt (1976). # DEFINITION. Let ~ be a class of frames. The modally definable closure of ~ in ~([Z]), [~]e, is the smallest MD in s176 class con# taining g. T H E O R E M 4.11. For every class of frames ~, [~] ~ consists of all isomorphic copies of C-collapses of general ultraproducts of frames from ~. # C O R O L L A R Y 4.12. g is MD in ~ ( [ ~ ) iff it is closed under isomorphisms and ~-collapses of general ultraproducts of frames. # C O R O L L A R Y 4.13. I f E is a A-elementary (defined by a set of firstorder conditions) class then ~ is MD in S ( [ ~ ) iff it is closed under # -collapses. In particular, a first-order property is definable in Y([Z]) iff it is preserved under ~-collapses. (E.g. all universal first-order conditions are.)


619

C O R O L L A R Y 4.14. Every fnite frame is distinguishable (up to iso-

morphism) by a set of Y([~)-formulae. Proof. From 4.13, since no finite frame has proper C-collapses.

#

The essential difference between the above characterization and the classical result of Goldblatt and Thomason is due to the fact that in 5e([~) the notions of generated subframe and disjoint union of frames are trivialized. Again following Goldblatt and Thomason (1974) and Goranko (1990) we can obtain another characterization of the A-elementary classes M D in 5e([]~), which is somewhat more convenient to use. D E F I N I T I O N . A bi-relational frame (W, R, S ) is a nonstandard ~([~)-frame if S u {{x, x}/x e W} = W 2.

#

D E F I N I T I O N . An ~(V~)-morphism is any bi-relational p-morphism of a frame of the type (W, R, r ). # Note that the image of any ~ ( [ ~ ) - m o r p h i s m f is a nonstandard 5r It is standard i f f f is an isomorphism. D E F I N I T I O N . An ultrafiher extension of an 5~([Z])-frame F = (W, R, r is the frame F* = {W*, R*, S*}, denoted ue(F), where W* is the set of ultrafilters in W and R*, S* are canonically defined on R and r respectively. # It is easy to see that ( W * , R*, S * ) is a nonstandard Y([Z])-frame. D E F I N I T I O N . F is an ultrafilter contraction of G iff G ~ ue(F). T H E O R E M 4.15 (cf. Goranko (1990)). A A-elementary class ~ is

MD in 5~(V~) iff ~ is closed under uhrafiher contractions of ~([~)morphic images. In virtue of Th. 4.10 all these results directly apply to ~c- Other results concerning definability in ~ ( [ ~ ) can be found in de Rijke (1989).

#

620


We shall finish this section with a strengthening o f 4.14 which gives an additional evidence on the expressive power of ~ . Denote

P,=

V

#(ci /, c~);

ir l