Sei sulla pagina 1di 15

Variants of First-Order Modal Logi

Marta Cialdea Mayer1 and Serenella Cerrito2


1 Universita Roma Tre, Dipartimento di Informati a e Automazione
2 Universite de Paris-Sud, L.R.I.
In this paper we study proof pro edures for some variants
of rst order modal logi s, where domains may be either umulative
or freely varying and terms may be either rigid or non-rigid, lo al or
non-lo al. We de ne both ground and free variable tableau methods,
parametri with respe t to the variants of the onsidered logi s. The
treatment of ea h variant is equally simple and is based on the annotation
of fun tional symbols by natural numbers, onveying some semanti al
information on the worlds where they are meant to be interpreted.

Abstra t.

1 Introdu tion
First order modal logi s onstitute a deli ate and di ult subje t (see for example [12 for an overview). In prin iple, their semanti s ould simply be obtained
by extension of the modal propositional semanti s: a rst order modal stru ture an be built on the base of a set of rst order lassi al interpretations (the
\possible worlds"), onne ted by a binary relation (the a essibility relation).
However, there are a number of di erent possibilities that an be onsidered,
on erning for example:

The obje t domains: is there any relation between the universes of the different worlds? They an be required to be the same for all worlds ( onstant
domains), they an bear no relation one to the other (varying domains), or
the obje t domains an vary, but monotoni ally, i.e. if w0 is a essible from
w, then the obje t domain of w is in luded in the domain of w0 ( umulative
domains).
The designation of terms: is the denotation of terms the same in all worlds
or an it vary? When the answer is positive, then designation is taken to be
rigid, otherwise it is non-rigid (often, mixed approa hes are of interest, too,
where the interpretations of some symbols are rigid, others are not).
The existen e of obje ts: does the extension of any ground term belong to
every obje t domain or not? If the answer is positive, then we assume terms
to be lo al, otherwise terms are non-lo al.

So, several variants of quanti ed modal logi s (QMLs) are possible, just by hoosing di erent ombinations of the ases onsidered above. We all them DDE variants (Domains/Designation/Existen e variants) of QML. Obviously, the logi
also depends on its propositional kernel, so that we have a four-dimensional

spa e of possible QMLs. As often happens with modal logi s, di erent hoi es
are appropriate for di erent appli ations. A dis ussion an be found in [12 as
well as [11.
Some DDE variants an easily be given a proof theoreti hara terization,
some are harder to treat. The easiest approa h is obviously obtained just by
adding the prin iples of lassi al logi to a propositional modal system. What
we obtain then is a logi with umulative domains, rigid designation and lo al
terms. In fa t, su h a logi validates the onverse of Bar an Formula (CBF),
that hara terizes umulative domains: 28xA ! 8x2A. Rigid designation
and lo ality of terms are onsequen es of the instantiation rule of lassi al logi
8xA ! A[t=x. For example, 2p( ) follows from the instantiation rule and p( ) ^
8x(p(x) ! 2p(x)), but it is not a logi al onsequen e of the latter formula alone,
if the denotation of an vary from world to world (the example is from [16).
Besides the axiomati hara terization, some rst order modal logi s with
umulative domains, rigid designation and lo al terms have been given sequent
and tableau al uli [9, 10, 15, natural dedu tion [5, matrix proof pro edures
[19, resolution style al uli [1, 7, and translation based pro edures [3, 18.
Now, while translation methods are general enough to treat also other DDE
variants of QML (for instan e, [3 deals with onstant domains logi s, where designation may be rigid as well as non rigid, and [18 handles all the DDE variants
of QML), the dire t methods are less exible. For instan e, it an be proved
that onstant domain logi s annot be treated by standard tableau systems ( ut
free sequent al uli). The addition of pre xes labelling tableau nodes solves this
problem [9, as matrix methods do [19: in both kinds of al uli in fa t it is possible to analyse more than one possible world at a time, and this allows the proof
to \go ba k and forth" (the same me hanism solves the problem of symmetri
logi s). In [9 all the variants of QML on erning the obje t domains of possible
worlds are treated by pre x tableau methods and some suggestions on how to
treat the varying domains ase in al uli without pre xes an be found. A dire t
approa h dealing with both varying domains and non-rigid symbols has a representative in [16, whi h de nes a resolution method for epistemi logi s, where
terms an be annotated by a \bullet" onstru tor distinguishing rigid terms from
non-rigid ones. A di erent dire tion is followed by [11, where the language of
modal logi is enri hed by means of a predi ate abstra tion operator, in order to
apture di eren es on the denotation of terms, and a tableau proof pro edure is
presented for su h a logi , with no restri tion on the domains of possible worlds.
In general, dire t proof methods an treat the less easy variants of QML
either by modi ations of lassi al and/or modal rules in an often umbersome
manner (if the axiomati approa h is taken), or else by atta hing some kind of
semanti al information to the proof stru tures.
The aim of this work is to provide a treatment of some variants of QML,
where the expli it intrusion of semanti s in the proof stru tures is kept as simple
and minimal as possible. We propose al uli in the tableau style, that allow us
to treat all the DDE variants of QML (that are formally de ned at the end
of this se tion), with the ex lusion of onstant domain logi s, that we believe

annot be aptured in the same style. Namely, we deal with the ombinations
umulative/varying domains, rigid/non rigid designation, lo al/non lo al terms,
for the analyti logi s K, D, T, K4, S4. The treatment of ea h DDE variant is
equally simple and is based on the annotation of fun tional symbols by natural
numbers, resulting in a minimal departure from standard tableau methods (i.e.
ut-free sequent systems). A similar me hanism is used in [7, 8, to obtain skolemlike normal forms and de ne modal resolution proof pro edures.
We de ne both ground tableau al uli { alled T abG (Se tion 2), dealing with
losed formulae { and free variable tableau al uli { alled T abFV (Se tion 3).
In both ases, the al uli are parametri with respe t to the DDE variants of the
onsidered logi . The treatments di er only in the annotation me hanism and
a suitable lassi ation of terms a ording to the annotation of their symbols,
that determines, in T abG , whi h appli ations of the -rule are orre t and, in
T abFV , whi h substitutions are allowed. Both lasses of al uli are sound and
omplete. Due to spa e restri tions, proofs are only sket hed (Se tion 4).
In the rest of this se tion, we brie y re all the syntax and semanti s of QML.
A rst order modal language L is onstituted by logi al symbols (propositional
onne tives, quanti ers, modal operators and a ountable set X of individual
variables), a non empty set LP of predi ate symbols and a set LF of fun tional
symbols with an asso iated arity. Constants are onsidered as fun tional symbols
with null arity. Terms are built using symbols from LF and X . The indu tive
de nition of terms and formulae is as usual.
A rst order modal interpretation M of a language L is a tuple hW; w0 ; R; D; ;
; i su h that:

{
{
{
{
{

W is a non empty set (the set of \possible worlds");


w0 is a distinguished element of W ;
R is a binary relation on W (the a essibility relation);
D is a non empty set (the obje t domain);
is a fun tion assigning to ea h w 2 W a non empty subset of D, the domain
of w: (w)  D;
{  represents the interpretation of onstants and fun tional symbols in the
language: for every world w 2 W and n-ary fun tional symbol in the language
(with n  0), (w; f ) 2 Dn ! D.
{  is the interpretation of predi ate symbols: if p is a n-ary predi ate symbol
and w 2 W , then  (w; p)  Dn is a set of n-tuples of elements in D.
The a essibility relation R of a modal stru ture an be required to satisfy
additional properties, hara terizing di erent logi s: we onsider seriality (D),
re exivity (T), transitivity (K4), both re exivity and transitivity (S4). The
logi K makes no additional assumption on R.
The interpretation fun tion  is extended to terms in the usual way, and,
by an abuse of notation, (w; t) denotes the interpretation of t in w. If M =
hW; w0 ; R; D; ; ; i is an interpretation of the language L, the language of
the model M, LD , is obtained from L by addition of a \name" for ea h d 2 D,
i.e. a new onstant d. The interpretation M is assumed to interpret names in

su h a way that that for every d 2 D and w 2 W , (w; d) = d. Note that the
interpretation of names d is always rigid.
Below we take :, ^ and _ as the only primitive propositional onne tives.
The relation j= between an interpretation M = hW; w0 ; R; D; ; ;  i, a world
w 2 M and a losed formula in LD is de ned indu tively as follows:

M; w j= p(t1 ; :::; tn ) i h(w; t1 ); :::; (w; tn )i 2 (w; p).


M; w j= :A i M; w 6j= A.
M; w j= A ^ B i M; w j= A and M; w j= B .
M; w j= A _ B i M; w j= A or M; w j= B .
M; w j= 8xA i for all d 2 (w), M; w j= A[d=x
M; w j= 9xA i there exists d 2 (w) su h that M; w j= A[d=x
M; w j= 2A i for any w0 2 W su h that hw; w0 i 2 R, M; w0 j= A
M; w j= 3A i there is a w0 2 W su h that hw; w0 i 2 R and M; w0 j= A
A losed formula A is true in M i M; w0 j= A, and it is valid i it is true in
any interpretation M.
Let M = hW; w0 ; R; D; ; ;  i be a rst order modal interpretation. We shall
1.
2.
3.
4.
5.
6.
7.
8.

deal with the following DDE variants of QML:

Cumulative domains: for all w; w0 2 W , if wRw0 then (w)  (w0 ).


Varying domains: no restri tions on (w).
Rigid designation: for all w; w0 2 M and for every fun tional symbol f ,
(w; f ) = (w0 ; f ).
Non-rigid designation: no restri tions on the relation between (w; f ) and
(w0 ; f ), for any w 6= w0 and f 2 LF .
Lo al terms: for all w 2 M and for every fun tional symbol f , if d1 ; :::; dn 2
(w), then (w; f )(d1 ; :::; dn ) 2 (w).
Non-lo al terms: no restri tion on the losure of domains (w) with respe t
to the appli ation of the designation of fun tional symbols.

For the sake of simpli ity, we treat ea h variant as a di erent logi . However,
mixed approa hes are also possible. For instan e, we an establish that onstant
symbols denote rigidly and fun tional symbols (of arity greater than 0) do not.
In the al uli des ribed in the rest of this work, ea h symbol would then be
treated a ording to the rules of the orresponding logi .
Note that, if terms are lo al, the denotation of every ground term t is in every
domain (w). However, if domains are allowed to vary without restri tions, if
M; w j= 9x3p(f (x)), and d is the element of (w) su h that M; w j= 3p(f (d)),
then the denotation of f (d) is not ne essarily in domain of the world w0 a essible
from w and su h that M; w0 j= p(f (d)), unless d 2 (w0 ).

2 Ground Tableaux
In this se tion we illustrate the me hanism that allows us to deal with some DDE
variants of QML. We show how to treat the ases of umulative and varying

domains, where designation of terms an be either rigid or non-rigid, and terms


either lo al or not. In the following, we assume that a given logi has been xed,
with a propositional basis that is one of the analyti logi s K, D, K4, T, or
S4, and a rst order variant.
The tableau systems presented in the sequel work with formulae in negation normal form: negation over non-atomi formulae is onsidered as a de ned
symbol. In other terms, modal formulae are built out of literals (atoms and
negated atoms) by use of ^, _, 2, 3 and the quanti ers 8 and 9. Negation over
non-atomi formulae is de ned as usual.
Below, we present the ground al ulus T abG. Formulae o urring in a ground
tableau for a set S of modal formulae are built over an extension L+ of the language L of S , where onstants and fun tional symbols an be annotated by a
numeri al supers ript, that is alled the level of the symbol. Symbols with supers ripts are alled annotated symbols. L+ extends L with two di erent in nite
sets of onstants: a set A ontaining a onstant for ea h level k 2 IN and a set
C , ontaining an in nite number of onstants for ea h level k > 0. Moreover, L+
ontains the annotated versions of the onstants and fun tional symbols in L.
Pre isely:

De nition 1. Let L be a modal language, A = fa00; a11 ; :::::g and C disjoint sets
of annotated onstants di erent from any symbol in L, su h that A ontains
exa tly one onstant akk for ea h k 2 IN, and C ontains a denumerable set of
onstants kk1 ; kk2 ; ::: annotated with level k, for ea h k  1. Then the labelled
+
extension of L is the language L+ su h that L+
P = LP and LF is:
LF [ C [ A [ ff k j f 2 LF and k 2 INg
The nodes of a ground tableau for a set of losed formulae in the modal
language L are labelled sets of the form n : S where n is a natural number,
alled the depth of the node, and S is a set of losed modal formulae in the
language L+ . Similarly to the sequent al ulus in [17, in this approa h there is
no labelling of single formulae, but of whole sets of formulae.
Symbol annotations and node depths are related: let t be the o urren e of
a term in a node n : S , that we an assume des ribes a given world w; then the
levels of the symbols in t tell us whether t has a denotation in the domain of w,
and whether its denotation an be assumed to be the same as the denotation of
another o urren e of the same term t (with possibly di erent annotations).
Next, we de ne the initial tableau for S and an operation that is used in
some propositional expansion rules.

De nition 2.
1. If S is a set of modal formulae, then the initial tableau for S is the single
node 1 : S  , where:
lo al terms and rigid designation: S  is obtained from S by annotating ea h fun tional symbol with level 0.
lo al terms and non-rigid designation: S  is obtained from S
by annotating ea h o urren e of a fun tional symbol outside the
s ope of any modal operator with level 1.

non lo al terms: S  = S
2. The base language of the tableau rooted at 1 : S  is the language L of S .
The language of the tableau is the labelled extension L+ of L.
3. If S is a set of modal formulae and n 2 IN, then:
lo al terms: S n is obtained from S by annotating ea h non-annotated
o urren e of a fun tional symbol outside the s ope of any modal
operator with level n;
non lo al terms: S n = S .
The main intuition behind the above notions is that symbols annotated with
level 0 are always symbols of the original language (and have not been introdu ed
by means of the -rule, de ned below), their interpretation is the same in every
possible world and the domain of every world is losed with respe t to the
appli ation of the fun tions they denote. Thus, symbols of level 0 o ur only
in the al ulus for the DDE variant with rigid designation and lo al terms.
In the ase of non-rigid designation and lo al terms, on the ontrary, a symbol
of the base language L an o ur with di erent levels in a tableau bran h: when
o urring with level n, it a tually refers to the denotation of the symbol in the
world orresponding to the node with depth n in that bran h. Symbols o urring
outside the s ope of a modal operator are initialized with level 1, the depth of
the initial node. Symbols of LF o urring inside a modal operator re eive their
levels only when, by appli ation of a propositional rule, they ome to the surfa e
(by means of operation 3 in De nition 2).
In the ase of non lo al terms, on the ontrary, no assumption is made on
the denotation of symbols of the base language: they have no level in the initial
node of the tableau and never get one.
We onsider a simple set of propositional tableau rules (others may be hosen
as well), shown in Table 1, where S; S 0 are sets of modal formulae (in the language
of the tableau), 2S stands for f2G j G 2 S g, S 0 is a set of non-boxed modal
formulae, omma is set union. The propositional part of the tableau systems we
onsider onsists of the lassi al rules, and: the rule K in systems K, D and T,
the rule 4 in K4 and S4, the rule T in T and S4, the rule D in D.
Adopting a terminology from [13, the rules that do not modify the depth of
the node are alled stati , the others are alled dynami . So, a non-annotated
symbol an obtain a supers ript only with the appli ation of dynami rules.
As an be noted, the depth n of a node n : S in a tableau \ ounts" the
number of dynami rules applied in the bran h before n : S . And, in fa t, the
labelling of nodes is not essential and the depth of a node ould be de ned as
a meta-level on ept. For this reason the present approa h is not in the style of
pre xed tableaux.
The next de nition establishes the relation between the depth of a node and
levels of symbols, by means of the on ept of domain of a node, that is needed
to formulate the quanti er tableau rules.

De nition 3. Let L be a modal language and L+ its labelled extension.


1. A term t in L+ is in depth n i every symbol in t has a level and t ontains
only symbols with:

Classi al rules

( )

n : A ^ B; S
n : A; B; S

( )

n : A _ B; S
n : A; S
n : B; S

 -rules
(D )

n : 2S; S
n + 1 : S n+1
0

(T )

n : 2A; S
n : An ; 2A; S

 -rules
(K )

n : 3A; 2S; S
n + 1 : An+1 ; S n+1
0

Fig. 1.

(4 )

n : 3A; 2S; S
n + 1 : An+1 ; S n+1 ; 2S
0

Propositional expansion rules

varying domains and non-rigid designation: level n;


varying domains and rigid designation: level n and level 0;
umulative domains: level k  n.
2. If n : S is a tableau node, then its domain is:
dom(n : S ) = fapp g [ ft j t o urs in S and t is in depth ng,
where
varying domains, rigid designation and lo al terms: if there
is at least a onstant in L then p = 0, otherwise p = n;
varying domains and either non-rigid designation or non-lo al
terms: p = n;
umulative domains: p = 1
Roughly speaking, the obje t denoted by a term in dom(n : S ) (and o urring
in a tableau) an be assumed to exist in the domain of the world \ orresponding"
to the node n : S . Su h terms are formed by use of symbols of level n itself,
symbols of level 0 in the ase of rigid designation (i.e. symbols of the base
language L), and symbols having lower level in the ase of umulative domains,
e.g. onstants denoting obje ts existing in the domain of some previous world,
therefore existing also in the onsidered world (in the ground version of the
al ulus we are presenting in this se tion, the only fun tional symbols with non
null arity are those of the base language L).
Note that an annotated term may be in more than one depth. Moreover, if
t 2 dom(n : S ), then t is a ground term, it does not ontain any non-annotated
symbol and every subterm t0 of t is in dom(n : S ).
The quanti ers expansion rules in T abG are the following:

( )

n : 8xA; S
n : A[t=x; 8xA; S

( )

n : 9xA; S
n : A[ n =x; S

In the -rule, t is any ground term in dom(n : 8xA; S ). In the -rule, is a


onstant in C that does not o ur in fAg [ S . Note that needs not be new in
the whole tableau: the lo al restri tion on the -rule orresponds exa tly to the
eigenvariable ondition in sequent al uli.
Sin e dom(n : 8xA; S ) is never empty, even if n : 8xA; S does not ontain
any ground term, the -rule an always be applied at least on e to su h a node.
The onstants of the set A (see De nition 1) in fa t are used to take into a ount
the fa t that the obje t domains are never empty, in mu h the same way as an
additional onstant is used to form Herbrand domains in lassi al logi when
there are no ground terms. The di eren e, here, is that we may need to have an
in nite number of su h onstants, one for ea h depth. If domains are umulative,
then one is enough: it denotes any obje t in the initial world w0 , that exists in any
other world a essible from w0 . Also if domains an vary but the base language L
ontains at least a onstant and terms are assumed to be lo al, one additional
onstant is enough: in fa t, if terms of L are lo al, denotes an obje t existing
in every world, and the onstant a00 will be taken to denote su h an obje t. If
none of the previous ases applies, then it may be the ase that any two obje t
domains have empty interse tion, so we need a di erent onstant to name an
arbitrary obje t of ea h domain.
In T abG, on e that an o urren e of a symbol gets a level, it never hanges.
In parti ular, new onstants introdu ed by the -rule never hange their level
and the only symbols that may o ur with di erent annotations in a tableau
node are symbols from the base language, when we are in the ase of non-rigid
designation and lo al terms. In that ase, f n is the denotation of f in the world
orresponding to the node of depth n, that an be di erent from the denotation
of the same symbol in another world. In onsideration of this fa t, the notion of
omplementary literals does not ignore the level of symbols.

De nition 4. Two literals P and :Q in the language L+ of a tableau are omplementary i P and Q are identi al, in luding their annotations. A node is
losed if it ontains two omplementary literals. A tableau is losed i all its
leaves are losed.
For example, p( 1 ) and :p( 1 ) are omplementary, but p( 1 ) and :p( 2 ) are not.
In the ase of umulative domains, rigid designation and lo al terms, the
al ulus is not di erent from the tableau system des ribed in [9, Chapter 7
(where, however, the language does not ontain any fun tional symbol with non
null arity). In fa t, every fun tional symbol in a tableau is annotated and n : S
never ontains symbols of level greater than n. So every term o urring in n : S
is in dom(n : S ).

Here follow some examples. The tableau on the left below proves the validity of CBF in the umulative domain logi s, using the K -rule. Sin e there
are no fun tional symbols in LF , there is no distin tion to be made about the
designation of terms. The last step in the varying domain systems would not be
allowed: 8xp(x) an be instantiated only with terms in depth 2. Moreover, sin e
there are no hoi e points above and the last expansion an only be repla ed by
an appli ation of the -rule on the onstant a22 (there are no onstants in LF ),
CBF annot be proved valid in the varying domains systems.
1 : :(28xp(x) ! 8x2p(x))
( )
1 : 28xp(x); :8x2p(x)
( )
1 : 28xp(x); :2p( 1 )
(K )
2 : 8xp(x); :p( 1 )
( )
2 : p( 1 ); 8xp(x); :p( 1 )

1 : :(8x2p(x) ! 28xp(x))
( )
1 : 8x2p(x); :28xp(x)
( )
1 : 2p(a1 ); 8x2p(x); :28xp(x)
(k )
2 : p(a1 ); :8xp(x)
(

)
2 : p(a1 ); :p( 2 )

A similar reasoning shows that no instan e of the Bar an formula (BF)

8x2A ! 28xA, that hara terizes onstant domains, is provable. The tableau

above on the right is an attempt, using again the K -rule and the instan e where
A = p(x). In this ase there is no di eren e between the umulative and varying
domains variants of the al ulus. Sin e with the appli ation of the K -rule the
universal formula is ne essarily lost, the tableau annot be losed.
Below, on the left, there is a proof of 2(8xp(x) ! p( )) (in any propositional
basis) with rigid designation (either varying or umulative domains: in both ases
the term 0 is in any depth) and lo al terms, showing that, in this variant of
the systems, the rule of instantiation is ne essarily valid. The same formula has
a di erent proof (on the right, below) in the ase of non rigid designation (and
lo al terms).
1 : :2(8xp(x) ! p( 0 ))
( )
2 : :(8xp(x) ! p( 0 ))
( )
2 : 8xp(x); :p( 0 )
( )
2 : p( 0 ); 8xp(x); :p( 0 )

1 : :2(8xp(x) ! p( ))
( )
2 : :(8xp(x) ! p( 2 ))
( )
2 : 8xp(x); :p( 2 )
( )
2 : p( 2 ); 8xp(x); :p( 2 )

And it annot be proved when terms are not ne essarily lo al. Here follows a
failed attempt:
1 : :2(8xp(x) ! p( ))
( )
2 : :(8xp(x) ! p( ))
( )
2 : 8xp(x); :p( )
( )
2 : p(a22 ); 8xp(x); :p( )
The next example is a (failed) attempt to show that p( ); 8x(p(x) ! 2p(x)) `

2p( ) in the ase of non-rigid designation and lo al terms:

1 : 8x(p(x) ! 2p(x)); p( 1 ); :2p( )


( )
1 : 8x(p(x) ! 2p(x)); p( 1 ) ! 2p( 1 ); p( 1 ); :2p( )
( )
1 : 8x(p(x) ! 2p(x));
1 : 8x(p(x) ! 2p(x));
2p( 1 ); p( 1 ); :2p( )
:p( 1 ); p( 1 ); :2p( )
(k )
2 : p( 1 ); :p( 2 )
The leftmost leaf is losed, but the rightmost is not, be ause p( 1 ) and :p( 2 )
are not omplementary. And in fa t 2p( ) is not a logi al onsequen e of p( )
and 8x(p(x) ! 2p(x)) in the ase of non-rigid designation and lo al terms. If
terms are not lo al, even the rst inferen e is not orre t. In fa t, the tableau
is initialized without any annotation on symbols and the -rule an never be
applied to a term with no levels.

3 Free variable tableaux


In this se tion we introdu e T abFV , i.e. free-variable tableaux for the onsidered
logi s. The de nition of quanti er rules is similar to the lassi al ase, but for
the annotation on symbols that is taken into a ount in the de nition of modal
substitutions. First of all, we de ne a further extension of the labelled extension
of a language (De nition 1), by addition of parameters and skolem fun tions:

De nition 5. Let L be a modal language, P = fz1n1 ; z2n2 ; :::g an in nite denumerable set of new annotated symbols, alled parameters, and F a denumerable
set of new annotated fun tion symbols su h that, for ea h i; j 2 IN, P ontains
in nitely many parameters of level i and F ontains in nitely many fun tion
symbols of level i and arity j . Then the labelled free variable extension of L is
the language Lfv su h that Lfv
P = LP and
+
Lfv
F = LF [ P [ F
The levels of parameters will be shown only when relevant. The de nition
of the depth of a term (De nition 3) arries along to terms in Lfv with no
modi ation. The next de nition introdu es the notion of modal substitution,
where a parameter z an be mapped to a term t in Lfv only if the annotation of
symbols in t \agrees" with the level of z . Sin e the agreement depends on whi h
depths the term is, the di erent DDE variants of QML orrespond to di erent
restri tions on modal substitutions.

De nition 6. A modal substitution  is a fun tion from P to terms in Lfv ,


that is the identity almost everywhere, and su h that if (z n ) = t (and z has
level n) then t is in depth n. Substitutions are denoted as usual by expressions
of the form ft1 =z1; :::; tm =zmg.

Uni ers, i.e. solutions of uni ation problems, and most general uni ers
(m.g.u.) are de ned like in the lassi al ase. The following de nition is also
borrowed from lassi al tableau al uli.

De nition 7. Let n1 : S1 ; :::; nk : Sk be all the leaves of a free variable tableau

T , and, for ea h i = 1; :::; k, let Pi and :Qi be a pair of literals in S i . If the modal

substitution  is a solution of the uni ation problem P1 = Q1 ; :::; Pk = Qk , then


 is an atomi losure substitution for T . If it is a most general solution for
su h a uni ation problem, then it is a most general atomi losure substitution
for T .
Besides the propositional expansion rules, that are the same as in T abG (but
for the fa t that parameters and skolem fun tions an o ur in tableau nodes),
the expansion rules of T abFV are:

Quanti er expansion rules :


( )

n : 8xA; S
n : A[z n =x; 8xA; S

( )

n : 9xA; S
n
n : A[f (z1n; :::; zk n )=x; S

In the -rule, z is a new parameter. In the -rule, f n is a new skolem fun tion,
i.e. a symbol of level n in F that does not o ur in f9xAg[ S , and z1 n ; :::; zk n
are all the parameters with level n in 9xA; S .

Most General Atomi Closure Substitution Rule :


If T is a tableau for a set S of senten es in L and the modal substitution
 is a most general atomi losure substitution for T , then the tableau T ,
obtained by applying  to T , is a tableau for S .
Note that the skolem term f n (z1n ; :::; zk n ) introdu ed by an appli ation of
the -rule ontains only parameters with level n, and the skolem fun tion f is

required to be \lo ally" new (i.e. new in the node). This is enough to ensure that
the rule is sound. In fa t, the role of the restri tion on f and the parameters
in the term is to prevent a modal substitution  to make f n (z1n ; :::; zk n ) equal
to some other parameter z o urring in the premise of the -rule, n : 9xA; S .
If z has level n then  (f n (z1n ; :::; zk n )) =  (z ) is impossible, be ause z is one
of z1n ; :::; zk n . The level m of z annot be greater than n, sin e a node never
ontains any symbol with level greater than its depth. Therefore, if m 6= n, then
m is ne essarily stri tly less than n, and, by de nition, a modal substitution
annot map a variable with level m < n to a term ontaining a symbol of level
n, that is not in depth m in any DDE variant of QML.
The de nition of losed tableau (De nition 4) stays un hanged for T abFV
(modulo the possible presen e of parameters and skolem fun tions in terms).
Note that in the al uli T abFV de ned here, losed tableaux are built up by rst
applying expansion rules, then omputing an m.g.u. allowing to simultaneously
lose all the leaves, and applying the substitution rule to the tableau so far

onstru ted just on e. Assuming a single, nal appli ation of the substitution
rule makes the soundness proof simpler. However, like in the lassi al ase, an
alternative version of the rule an be onsidered, where any m.g.u. of a pair of
literals in a single leaf an be applied. Sin e any most general atomi losure
substitution an be found by repeated appli ations of su h an alternative rule,
this se ond version of the al uli is equivalent to the rst one.
As an example, below, we give again a proof of the validity of CBF in the
umulative domain logi s, using the K -rule, but this time in the al ulus T abFV .
1 : :(28xp(x) ! 8x2p(x))
( )
1 : 28xp(x); :8x2p(x)
( )
1 : 28xp(x); :2p( 1 )
(K )
2 : 8xp(x); :p( 1 )
( )
2 : p(z12 ); 8xp(x); :p( 1 )

Here, 1 is a (0-ary) skolem fun tion. Sin e 1 is in depth 2, it su es to


apply the substitution f 1 =z12 g to lose the tableau. In the ase of varying domain
systems, su h a substitution is illegal, be ause 1 is not in depth 2.
As a further example, the tableau below is a T abFV proof of the validity of the
formula 28x9yp(x; y ) ! 9v 29wp(v; w) in the logi K with umulative domains
(sin e LF is empty, there is no di eren e to be made about the designation of
terms).
1 : :(28x9yp(x; y ) ! 9v 29wp(v; w))
( )
1 : 28x9y p(x; y ); :9v 29wp(v; w)
(

)
1 : 28x9y p(x; y ); :29wp(z11 ; w)
(K )
2 : 8x9y p(x; y ); :9wp(z11 ; w)
( )
2 : 9y p(z22 ; y ); 8x9y p(x; y ); :9wp(z11 ; w)
( )
2 : p(z22 ; g 2 (z22 )); 8x9y p(x; y ); :9wp(z11 ; w)
( )
2 : p(z22 ; g 2 (z22 )); 8x9y p(x; y ); :9wp(z11 ; w); :p(z11 ; z32 )

Sin e domains are umulative, the term z11 is in any depth k  1, thus, in
parti ular, it is in depth 2. Hen e fz11 =z22 ; g 2 (z11 )=z32 g is a modal substitution,
and onstitutes a most general atomi losure substitution. Its appli ation yields
a losed tableau. Note that in the varying domains option, no substitution an
lose the tableau, sin e z11 is not in depth 2, and neither fz11 =z22 g nor fz22 =z11 g
are legal substitutions. Indeed, the onsidered formula is not valid in this ase.

4 Soundness and Completeness of T abG and T abF V


In this se tion we brie y sket h the main ideas underlying the soundness and
ompleteness proofs for both systems T abG and T abFV .
In order to show that T abG is sound, we indu tively show that every tableau
for a satis able set of formulae S has always an open leaf, i.e. a satis able leaf,

where a node n : S is satis ed by a world w in M i M; w j= S (ignoring


annotations) and for every ground term t 2 dom(n : S ), M(w; t) 2 (w). The
latter requirement is exploited to show that if the premise of a -rule is satis able,
then so is its expansion. And it must be guaranteed to hold both in the initial
tableau and when a dynami rule is applied.
In the ompleteness proof for T abG, we rst onsider a al ulus where the
-rule is repla ed by its \wasteful" variant  :
(  )

n : 8xA; S
n : A[t=x; 8xA; S

where t is any ground term in depth


n of the language of the tableau L+

A tableau in this modi ation of the al ulus will be alled a  -tableau. Completeness is proved by onstru tion of a \synta ti al model" of any set of senten es that has no losed  -tableau, and the onstru tion obviously exploits the
wasteful  -rule. Then we show that any  -expansion in a losed  -tableau an
be repla ed by a -expansion (using a suitable term from the set of symbols A introdu ed in De nition 1), thus obtaining a losed tableau in the original ground
al ulus T abG .
The synta ti al model of the unrefutable set of formulae S is built in two
steps. First of all an interpretation M of the extension L+ of the language L
of S is onstru ted, where however annotations are not ignored: a symbol f i is
onsidered di erent from f j when i 6= j . The frame of M is a (possibly in nite
and in nitely bran hing) tree, whose elements are saturated sets of formulae
(the de nition of saturation is similar to the standard one, obviously referring to
the al uli we are onsidering). The initial world w0 is a saturation of S  , where
1 : S  is the initial tableau for S . Ea h element w of the frame is asso iated with
a natural number depth(w), that measures its distan e from the root.
The elements and  of M make ea h possible world w a kind of Herbrand
interpretation: the domain (w) is the set of ground terms in dom(depth(w); w),
where terms that di er (even only) in the annotation of their symbols are different elements of the domain. Annotated fun tional symbols are always given a
rigid interpretation, in su h a way that, if a term t ontains only annotated symbols, then (w; t) = t for all w 2 W . The treatment of non-annotated symbols
di ers in the ases of lo al and non-lo al terms. Finally, an atom P is true at w
i P 2 w. It an be shown that, for all w 2 W , M; w j= A for all A 2 w, where,
like before, symbols o urring with di erent levels are onsidered di erent. For
this reason, M is not yet an interpretation of the language L, and we annot
dire tly on lude that M j= S . From the interpretation of the (possibly in nite)
set of symbols f 1 ; f 2 ; f 3 ; ::: where f 2 LF , the interpretation of f is extra ted
for every w 2 W . Su h a new interpretation satis es all the requirements of the
di erent DDE variants of QML we are onsidering, and is indeed a model of S .
The soundness and ompleteness of T abFV is established by showing the
equivalen e between the al uli T abFV and T abG. Therefore, sin e T abG is sound
and omplete, S is unsatis able if and only if there is a losed T abFV tableau
for S . The proof is based on a synta ti al transformation of T abG -tableau proofs
into tableau proofs in T abFV and vi eversa.

5 Con luding Remarks


In this paper we provide al uli treating di erent variants of rst order modal
logi (varying and umulative domains, rigid and non rigid designation, lo al
and non lo al terms), by addition of a minimal semanti al annotation to tableau
methods obtained by extending propositional modal systems with the lassi al
rules for quanti ers. We give both a ground formulation of the al uli (T abG )
and free variable tableau systems (T abFV ). In both ases, the di erent DDE
variants are obtained in a simple and modular way, just by hoosing among
alternative de nitions of a few basi operations. In parti ular, in the ase of
T abFV al uli, a rst order variant is essentially hara terized by the notion
of modal substitution. Di erently from other approa hes su h as for instan e
[15, 2, however, term uni ation does not depend on the propositional kernel
of the logi (i.e. on the a essibility relation) and does not involve any omplex
or strongly semanti ally dependent operation: what is additionally needed, with
respe t to lassi al uni ation, is some numeri al \greater than" test.
Although we have treated ea h DDE variant as a di erent logi , some mixed
approa hes are possible with respe t to the denotation of symbols: for example,
it an be established that onstant symbols (proper names) are rigid while fun tional symbols are not, so that de nite des riptions do not ne essarily denote
rigidly. Ea h symbol is then to be treated a ording to the orresponding variant
in the onstru tion of the initial tableau and in the appli ation of dynami rules.
We believe that equality an easily be treated in the same style.
In the ase of the (proof-theoreti ally) easiest DDE variant assuming umulative domains, rigid designation and lo al terms, the al uli T abG are a tually
the same as what is obtained by a simple addition of lassi al quanti er rules
to modal propositional ones (su h as, for example, in [9, Chapter 7). However,
their T abFV ounterpart is not exa tly the same as proposed in [10, where a
tableau al ulus with free variables, dynami skolemization and uni ation is
de ned, that is orre t and omplete for this variant of QML. There, the new
skolem term f (z1 ; :::; zn ) introdu ed by an appli ation of the -rule is su h that
z1 ; :::; zn are all the parameters o urring in the bran h. Exploiting the annotation on symbols, the -rule in T abFV is somewhat more liberal, requiring only
parameters having the same level as the depth of the node to be the arguments
of the skolem fun tions. Probably, the -rule an be further relaxed, in the style
of [14, 6, 4. The details, however, still have to be worked out.
The (ground) pre xed tableaux presented in Chapter 8 of [9, in the three
di erent variants of onstant, umulative and varying domains, asso iate pre xes
with onstants introdu ed by appli ations of the -rule in a way that bears
some resemblan e with T abG and its labelling me hanism. The present approa h,
however, is not in the style of pre xed tableaux, where pre xes are asso iated
to ea h formula (not the whole set of \a tive" formulae), are more omplex
stru tures than integers and play a fundamental role in modal expansion rules.
A stri ter relation links the present work with [7, 8, where the same kind of
annotation is used to restri t uni ation in the ontext of modal resolution.
However, being the skolem-like transformation stati in that ase, it works well

with logi s su h as K and D, where formulae annot move from their modal
ontext, but requires a preliminary propositional transformation of formulae
when dealing with re exive and/or transitive logi s, where the modal degree of
a formula is subje t to hange. The dynami introdu tion of parameters and
skolem terms in tableau systems makes things easier: a quanti er is treated only
when it omes to the surfa e, i.e. when, after possible hanges due to appli ations
of the T or 4 -rule, its modal ontext is settled.

Referen es
1. M. Abadi and Z. Manna. Modal theorem proving. CADE-86, 172{189, 1986.
2. A. Artosi, P. Benassi, G. Governatori, and A. Rotolo. Shakespearian modal logi :
A labelled treatment of modal identity. Advan es in Modal Logi , 1{21, 1998.
3. Y. Au ray and P. Enjalbert. Modal theorem proving: an equational viewpoint. J.
of Logi and Computation, 2:247{297, 1992.
4. M. Baaz and C.G. Fermuller. Non-elementary sppedups between di erent versions
of tableaux. Tableaux '95, 217{230, 1995.
5. D. Basin, M. Matthews, and L. Vigano. Labelled modal logi s: Quanti ers. J. of
Logi , Language and Information, 7(3):237{263, 1998.
6. B. Be kert, R. Hahnle, and P. H. S hmitt. The even more liberalized -rule in free
variable semanti tableaux. In Pro . of the 3rd Kurt Godel Colloquium KGC'93,
108{119, 1993.
7. M. Cialdea. Resolution for some rst order modal systems. Theoreti al Computer
S ien e, 85:213{229, 1991.
8. M. Cialdea and L. Fari~nas del Cerro. A modal Herbrand's property. Z. Math.
Logik Grundlag. Math., 32:523{539, 1986.
9. M. Fitting. Proof Methods for Modal and Intuitionisti Logi s. Reidel Publ. Co.,
1983.
10. M. Fitting. First-order modal tableaux. J. of Automated Reasoning, 4:191{213,
1988.
11. M. Fitting. On quanti ed modal logi . Fundamenta Informati ae, 39, 1999.
12. J. W. Garson. Quanti ation in modal logi . In D. Gabbay and F. Guenthner,
editors, Handbook of Philosophi al Logi , vol. II, 249{307. Reidel Publ. Co., 1984.
13. R. Gore. Tableau methods for modal and temporal logi s. In M. D'Agostino,
G. Gabbay, R. Hahnle, and J. Posegga, editors, Handbook of tableau methods.
Kluwer, 1999.
14. R. Hahnle and P. H. S hmitt. The liberalized -rule in free variable semanti
tableaux. J. of Automated Reasoning, 13:211{222, 1994.
15. P. Ja kson and H. Rei hgelt. A general proof method for rst-order modal logi .
IJCAI '87, 942{944, 1987.
16. K. Konolige. Resolution and quanti ed epistemi logi s. CADE 86, 199{208, 1986.
17. G. Mints. Indexed systems of sequents and ut-elimination. J. of Philosophi al
Logi , 26:671{696, 1997.
18. H. J. Ohlba h. Semanti s based translation methods for modal logi s. J. of Logi
and Computation, 1(5):691{746, 1991.
19. L. A. Wallen. Automated Dedu tion in Non lassi al Logi s: E ient Matrix Proof
Methods for Modal and Intuitionisti Logi s. MIT Press, 1990.

Potrebbero piacerti anche