On Modal Logics for Qualitative Possibility in a ... - Semantic Scholar
278
On Modal Logics for Qualitative Possibility in
Petr Hajek Dagmar Harmancova Institute of Computer Science Academy of Sciences 18207 Prague, Czech Republic e-mail: {hajek,dasa }@uivt.cas.cz
Abstract Within the possibilistic approach to uncer tainty modeling, the paper presents a modal logical system to reason about qualitative (comparative) statements of the possibility (and necessity) of fuzzy propositions. We re late this qualitative modal logic to the many valued analogues MVS5 and MVKD45 of the well known modal logics of knowledge and be lief 55 and KD45 respectively. Completeness results are obtained for such logics and there fore, they extend previous existing results for qualitative possibilistic logics in the classical non-fuzzy setting. Keywords: Possibilistic Logic, Fuzzy Logic, Qualitative Possibility, Many-valued Modal
Logic.
1
INTRODUCTION
In the recent past, a lot of effort has been put in re lating numerical and symbolic approaches to uncer tain reasoning. Numerical formalisms attach degrees of belief to propositions. Belief degrees are understood as a measure on the set of possible worlds (possible state descriptions) that assigns to every proposition the measure of the set of worlds in which the proposi tion is true. Therefore, uncertainty measures are not truth-functional, as it is welt known and established, i.e. the measure of a compound formula can not be in general obtained as a function of the measures of its subformulas. Possibilistic logic (cf. e.g. [Dubois & Prade, 88]) is a particular numerical formalism based on the use of the so-called possibility and necessity measures that provide to what extent a crisp piece of knowledge can be considered plausible and certain respectively. Even if formulas bear numerical possibilities, we may be in terested not in the values themselves but only in their comparison, i.e. in formulas such as A D), thus T'lf-(j)B for some j :S i, by the in duction hypothesis. On the other hand, T0 f- (i)B implies TB f- (i)B and by the induction hypothe sis, TBJf-(i)B. Hence i = max{jj B JITJ T E W}, i =II B II- T hi s completes the proof of this lemma • and of the Main Lemma 4.7. Corollary 4.13
MVSS is complete with respect to the
given semantics. 5
THE LOGIC
Q F L2
(b) (3TRiT0)(Tf-(i)C)
The QF L2 comparative modality
(DA --t DB),
On Modal Logics for Qualitative Possibility in a Fuzzy Setting
DA ._... DDA, (j)DA ..... D(j)DA, OA ..... DOA, (j)OA
._...
D(j)OA and (1)0True
are }-tautologies. Proof: We only prove the first formula (axiom K). The rest are easily proved by straightforward compu tations. It suffices to show that
II DA--+ DB 11211 D(A--+ B) II. II DA-+DB II= I(minw{ 17r(w) v IIAIIw},minw•{l - 1r(w') v IIBIIw•l) = maxwminw•I(l-7r(w) VIIA llw, 1-1r(w') VIIBllw,) 2: min wi(l-7r(w) VJIAIIw,l-7r(w) V IIB IIw), and in the other hand II D(A --+ B) II= minw{l-1r(w) V I(ll A llw, IIBllw)}. Thus if we prove We have in the one hand
I(l-1r(w) v IIAIIw,1-1r(w) v IIBIIw) 2: 2: 1- 1r(w) v I(I/AIIw, I!BII w)
•
--+
V £(2: i)O(E A (i)(A A
Proo[- If i = 0 it is obvious. If i = maxv ( II A 11, A1r(v)) > 0 then, it is easy to prove that, there ex ists vo and E such that II A JJ,o A7r( vo ) = i, vol f-E and 7r ( vo ) =II OE II· Therefore II A llvo A7r(vo) = IIAIIvo A IIOE II= i. Thus II E A (i)(A A OE) llvo= 1, IIE A(i)(A AOE) IIvo A7r (vo ) = i, and therefore • II O(E A(i)(AA OE)) 11 2: i. Now
we
are
ready
to
present
our
axioms
of
MVKD45. Definition 5.6 The modal logic MVKD45 has the following axioms: •
axioms of propositional calculus (as above)
•
axioms of KD45:
D(A--+ B)--+ (DA--+ DB), DA ..... DDA, OA ._... DOA
the lemma will be proved. But observe that: •
I 0. Then for some E, To 1- (> O)O(E 1\ (j)(C A OE)). Put D = (E 1\ (j)(C A OE)). Let To 1- (ir.)DBk , k = 1, ... , n;
To 1- D(ik)D Bk and To 1- D/\�=t(ik)DBk (note (ik)DBk is a B-formula!). Since MVKD45 proves OD-+ (DH--+ O(D A H)) for H, D being B-formulas, we get, for H = 1\(ik)DBk, To 1- OD - O( D A H), thus To 1- (> O)OD -+ (> O)(D A H), thus To 1(> O)O(D A H), and therefore D 1\ H is consistent3. then that
3Recall that a formula A is MV K D45-consistent
MV K D45 fj-.A.
if
284
Hajek, Hannancova, Esteva, Garcia, and Godo
Consequently, D is consistent with the set of all To provable formulas of the form (i)DB, completing this • theory we get our T. This last lemma enables us to define our model as fol lows. For each i and C such that To 1(i)OC let Tc be a theory T as in (b) in the above lemma. We define the model (W, lh 1r ) such that the set of worlds is W ={To} U {Tc I C}; forT E W, the forcing relation is defined by TH-( i)p iff T f- ( i)p, and finally, the possibility distribution is given by 1r(T) = i i!JT r E 1\ (i)OE. Definition 5.8
more natural choice would be II DA II= minw ( 1r ( w ) ) II A llw) but then to have duality of D and 0 we should have II OA II= maxw(7r & II A llw) (strict conj u nction). Unfortunately, for this semantics of D one of the basic axioms of modal logic, namely D(A -+ B) -+ (DA -> DB), is not a tautology. --->
(3) To find an elegant (non-pedestrian) axiomatization of QF L2 in its own language still remains. As a matter of fact, it is worth noticing that some of the axioms of Farinas and Herzig's QP L logic, e.g. • • •
Completeness is obtained by proving next main lemma in a similar way as in lemma 4.12. Lemma 5.9
For each i, B, and for eachT E W, Tlf--Ci)B iff T r (i)B.
•
(A<JB) v (B
On Modal Logics for Qualitative Possibility in
Petr Hajek Dagmar Harmancova Institute of Computer Science Academy of Sciences 18207 Prague, Czech Republic e-mail: {hajek,dasa }@uivt.cas.cz
Abstract Within the possibilistic approach to uncer tainty modeling, the paper presents a modal logical system to reason about qualitative (comparative) statements of the possibility (and necessity) of fuzzy propositions. We re late this qualitative modal logic to the many valued analogues MVS5 and MVKD45 of the well known modal logics of knowledge and be lief 55 and KD45 respectively. Completeness results are obtained for such logics and there fore, they extend previous existing results for qualitative possibilistic logics in the classical non-fuzzy setting. Keywords: Possibilistic Logic, Fuzzy Logic, Qualitative Possibility, Many-valued Modal
Logic.
1
INTRODUCTION
In the recent past, a lot of effort has been put in re lating numerical and symbolic approaches to uncer tain reasoning. Numerical formalisms attach degrees of belief to propositions. Belief degrees are understood as a measure on the set of possible worlds (possible state descriptions) that assigns to every proposition the measure of the set of worlds in which the proposi tion is true. Therefore, uncertainty measures are not truth-functional, as it is welt known and established, i.e. the measure of a compound formula can not be in general obtained as a function of the measures of its subformulas. Possibilistic logic (cf. e.g. [Dubois & Prade, 88]) is a particular numerical formalism based on the use of the so-called possibility and necessity measures that provide to what extent a crisp piece of knowledge can be considered plausible and certain respectively. Even if formulas bear numerical possibilities, we may be in terested not in the values themselves but only in their comparison, i.e. in formulas such as A D), thus T'lf-(j)B for some j :S i, by the in duction hypothesis. On the other hand, T0 f- (i)B implies TB f- (i)B and by the induction hypothe sis, TBJf-(i)B. Hence i = max{jj B JITJ T E W}, i =II B II- T hi s completes the proof of this lemma • and of the Main Lemma 4.7. Corollary 4.13
MVSS is complete with respect to the
given semantics. 5
THE LOGIC
Q F L2
(b) (3TRiT0)(Tf-(i)C)
The QF L2 comparative modality
(DA --t DB),
On Modal Logics for Qualitative Possibility in a Fuzzy Setting
DA ._... DDA, (j)DA ..... D(j)DA, OA ..... DOA, (j)OA
._...
D(j)OA and (1)0True
are }-tautologies. Proof: We only prove the first formula (axiom K). The rest are easily proved by straightforward compu tations. It suffices to show that
II DA--+ DB 11211 D(A--+ B) II. II DA-+DB II= I(minw{ 17r(w) v IIAIIw},minw•{l - 1r(w') v IIBIIw•l) = maxwminw•I(l-7r(w) VIIA llw, 1-1r(w') VIIBllw,) 2: min wi(l-7r(w) VJIAIIw,l-7r(w) V IIB IIw), and in the other hand II D(A --+ B) II= minw{l-1r(w) V I(ll A llw, IIBllw)}. Thus if we prove We have in the one hand
I(l-1r(w) v IIAIIw,1-1r(w) v IIBIIw) 2: 2: 1- 1r(w) v I(I/AIIw, I!BII w)
•
--+
V £(2: i)O(E A (i)(A A
Proo[- If i = 0 it is obvious. If i = maxv ( II A 11, A1r(v)) > 0 then, it is easy to prove that, there ex ists vo and E such that II A JJ,o A7r( vo ) = i, vol f-E and 7r ( vo ) =II OE II· Therefore II A llvo A7r(vo) = IIAIIvo A IIOE II= i. Thus II E A (i)(A A OE) llvo= 1, IIE A(i)(A AOE) IIvo A7r (vo ) = i, and therefore • II O(E A(i)(AA OE)) 11 2: i. Now
we
are
ready
to
present
our
axioms
of
MVKD45. Definition 5.6 The modal logic MVKD45 has the following axioms: •
axioms of propositional calculus (as above)
•
axioms of KD45:
D(A--+ B)--+ (DA--+ DB), DA ..... DDA, OA ._... DOA
the lemma will be proved. But observe that: •
I 0. Then for some E, To 1- (> O)O(E 1\ (j)(C A OE)). Put D = (E 1\ (j)(C A OE)). Let To 1- (ir.)DBk , k = 1, ... , n;
To 1- D(ik)D Bk and To 1- D/\�=t(ik)DBk (note (ik)DBk is a B-formula!). Since MVKD45 proves OD-+ (DH--+ O(D A H)) for H, D being B-formulas, we get, for H = 1\(ik)DBk, To 1- OD - O( D A H), thus To 1- (> O)OD -+ (> O)(D A H), thus To 1(> O)O(D A H), and therefore D 1\ H is consistent3. then that
3Recall that a formula A is MV K D45-consistent
MV K D45 fj-.A.
if
284
Hajek, Hannancova, Esteva, Garcia, and Godo
Consequently, D is consistent with the set of all To provable formulas of the form (i)DB, completing this • theory we get our T. This last lemma enables us to define our model as fol lows. For each i and C such that To 1(i)OC let Tc be a theory T as in (b) in the above lemma. We define the model (W, lh 1r ) such that the set of worlds is W ={To} U {Tc I C}; forT E W, the forcing relation is defined by TH-( i)p iff T f- ( i)p, and finally, the possibility distribution is given by 1r(T) = i i!JT r E 1\ (i)OE. Definition 5.8
more natural choice would be II DA II= minw ( 1r ( w ) ) II A llw) but then to have duality of D and 0 we should have II OA II= maxw(7r & II A llw) (strict conj u nction). Unfortunately, for this semantics of D one of the basic axioms of modal logic, namely D(A -+ B) -+ (DA -> DB), is not a tautology. --->
(3) To find an elegant (non-pedestrian) axiomatization of QF L2 in its own language still remains. As a matter of fact, it is worth noticing that some of the axioms of Farinas and Herzig's QP L logic, e.g. • • •
Completeness is obtained by proving next main lemma in a similar way as in lemma 4.12. Lemma 5.9
For each i, B, and for eachT E W, Tlf--Ci)B iff T r (i)B.
•
(A<JB) v (B
