From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. Automated Deduction Multi-Modal Time and for Logic a relations have particular properties for a given of modal logic system. Knowledge Recently, H.J Ohlbach in [18.1, 19], A. Herzig in [10], Y. Auffray in [0], Y. Auffray and P. Enjalbert in [1] , L. Farifias del Cerro Olivier Gasquet I.R.I.T - Universit# Paul Sabatier, 118, route de Narbonne, F-31062 Toulouse cedex, France e-mail: gasquet@irit, fr and A. Herzig in [4] have devised new proof methods for modal logics based on a translation into first order logic with specific equational theories, according to the system Introduction and logical background under concern. Multimodal logics are an appropriate The wayto get a proof methodfor a particular frameworkto represent agents, actions, time, modallogic is, first to exhibit the associated knowledge... (see [8J, [9], [14], [18], [25] equational for corresponding unification algorithm. In these some of its uses). For example Knowledge may be represented by means of theory and then to define the papers ([4, 19, 1]) the method has been the modal operator [K] and Time by [T]; and defined for monomodallogics, [K]p maybe read as "p is known"and [T]p as one "p will be true tomorrow". A typical modal Our aim is to extend this method towards formula is [K][T]p meaning that it is known complex multimodal logics containing one or that p will be true tomorrow(whichis a priori more interaction axioms, i. e. ones involving different from [T][KIp). several modal operators 1. An example is the The semantics of these logics is based in inclusion axiom [K1]p-->[K2]p, which can be Kripke models [15], where to each modal read as "if agent 1 knows p then agent 2 modal operator where only is used. operator is associated a binary relation over a set of so-called possible worlds. These 1 It seems that for multimodal logicswithout interaction axioms, the deduction problem is, essentiallya superposition of thoseof the subsystems. 38 From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. knowsp". A solution for this axiom has been equational theory; in section 4 we define a given in [17]. unification algorithm; finally, in section 5, we We have been specially interested in prove the termination, automated deduction methods for modal logics with a persistence [KI[T]p~[TI[K]p, soudness and completenessof this algorithm. axiom of the form : 1 The multimodal logic which has been used in Wewill use a simplified logic system [5], [ 18], for example.Its intuitive meaningin in order to clarify the presentation. The a "knowledge and time" interpretation is: "if it language of our logic is defined as usual is knownthat tomorrow p will be true then ([12]); it is that of classical first-order logic tomorrow it will be knownthat p is true". plus the modal connectives [K] and [T]. That is, knowledgeabout the future persists. Its axiomatics is based on that of first-order It can be shownthat someof these multimodal versions logics can be represented in an equational of the system Q2 for both the connectives [K] and IT] (see again [12]), plus order-sorted framework,with essentially two the axiom IKIITIp~[TI[K]p sorts that we call symbols, possessing an equational persistence axiom as in [5]. Its semantics is axiomcapturing the persistence. based in Kripke models M=<W,RK, RT, D, Followingthis, the two central points of this m>where: Wis a set of possible worlds, D is paper are: to present a wayto deal with such a the domainof objects, m is the interpretation logical axiom by translating it into an function mapping variables equational theory and to define a unification to D and n-ary n, and, RKand predicates to subsets of WxD algorithmfor it. In the sequel, wedefine first RTare accessibility relations (over W)which a multimodal logic with persistence axiom; verify: then we give the target logic and a translation i) RTand RKare serial 3, and from the multimodallogic into the target one, involving a particular equational theory. In 2 Q (or KD) is based on the system K and the axiom section 3 we give a formal definition of the []P*<>P. 3 A binary relation R is serial iff for any wE W, there exists w’e Wsuch that (w,w’)~ 39 From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. 5, tr([KJA) = VX:K([X] tr(A)) ii) RToRK_c RKoRT4 (called relational persistence henceforth) tr([T]A) = VY:T([Y] tr(A)), Proofs which are not given in this paper can tr(<T>A) = 3Y:T([Y] tr(A)), be found in 161. tr(<K>A)= 3Y:K([Y] tr(A)); where, as usual, <K> (resp. <T>) 2 The target logic and the translation abbreviates ~[K]~ (resp. ~[T]~); The language of the target logic (that wewill and is homomorphic w.r.t, the other call TLhenceforth) is as follows: connectives(--,, v .... ). It is based on classical connectives, a set of Now,to every (relational) Kripke model <W, predicate symbolsand three sets of variables FK, FT, D, m>, we associate a (functional) VCL, VT and VK. That x is a variable of Vs, model of TL: for t amongCL, T, K, is denoted by x:s. where Wand D are as above; and FK (resp. FT) is a set of total functions from Wto W Formulasare defined by: whose graphs cover that of RK(resp. that of p(xl:sl ..... xn:sn) is a formula, where si is RT), moreover: CL, T,or K and p is a predicate symbol. -’7’t~ FT,Vk~FK, ~k’~ FK, 3t’~ FT : 11 A and B are formulas, x:s and y:s’ are tok = k’ot’ 6 (this functional persistence variables where s is CL, T,or K, and s’ is captures the relational persistence RKand RT). T,or K; then ~A, AvB, A^B, A---~B, A~--)B, - m maps VKto FK, VT to FT, VCLto Vx:s A, 3x:s A, and ly:s’]A are formulas. D and n-ary predicates to subsets of WxDn; Wepresent here, very briefly, the translation and the sem~mticsof TL. Thetranslation tr is definedby: tr(A) =A,if Ais an atom. 4 Where o is relation 5 X:K denotes that X is of sort K. 6 tok is defined by: tok(w) = k(t(w)), o denotes function composition but in the reverse order w.r.t, the usual one. Wechoose this convention in order to keep close to relation composition and to stress the "path" between two points of W. composition; (a,b)~ RToRKiff ]c~ W; (a,c)¢ RT and (c,b)e 40 From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. and if wedefine an operator ¯ such that Now,given a formula of the original m(o)=o, then [X:T][Y:K]A is equivalent logic, we want to check whether or not it is a [X:T ¯ Y:K]A(see [4]). theorem; summarizing our results, we have showedin [6] that: Note that the seriality of both RKand RT enables the totality of functions of FKand FT. Satisfiability -this formula is satisfiable relational in functional modelsis semantics if and only if its translation is satisfiable defined as in relational ones, exceptfor: in the in the functional semantics of TL$; -there exists a clausal normalform in M,wsat ly:slA iff M,m(y)(w)sat TL¢(via a skolemizationstep); m(y:s) is in Fs, wheres is T or -the classical resolution principle ([23]) can be extended to our case, by applying The next step of our work consists in proving a strengthening unification in the algebra of terms over the theorem which signature that we shall present below. expresses that wecan restrict our attention to the functional Example: models which enable the The formula [T]<K><K>A becomes, after construction of two operators ¢~ and fir,, translation, having the following property: ?V’X:T 3Y:KqZ:K [X ¯ Y ¯ Z]p, which is V t~ FT,V k~ FK:tok = ~ (t,k)o¢~ where the type of ~ (resp. FKXFT~FT (resp. (t,k), skolemized in of ~f4¢ ) [X:T ° fK(X) ° gK(X)]p, where fK and FKXF’r~FK). are skolem functions mappingsort T into sort K. As one can see, these operators are a sort of Skolem functions. This literal can be unified with another, say, Adding them to the [Y. Z]-~p language of TL to get TL’I’, we can showthat if andonly if: the formula [X:TI[Y:K]A is equivalent to a:T ¯ fK(a) ° gK(a) and Y:K¯ Z:T I ~-C(X,Y). ~(X,Y)I unifiable under the equational theory defined below. 41 From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. (Z-)terms, (Z-)algebras and (Z-)substi-tutions 3 The order-sorted equational are defined as usual (see [11] or [26]). framework Consequently, the signature Overthis signature, weconsider the following E we use is equational theory denoted E and defined as formally: follows: is a set of sort symbols, S={T,K,TOP,CL} (x:TOP with the followingsort structure: T _< TOP and K _< TOP z’TOP U MFunT U MFunK), is a set of operators of our problem persistence axiom : (x:K,y:T) U (CFun ¯ and the main feature previous section). ~,~v} y:TOP) x ¯ (y ¯ z), i.e. ¯ is associative; (Sort CL denotes the domain of objects; cf F={., ¯ ¯ ~ (x,y) =E x.y, axiom allows some kind of permutation of with the types. following declarations : ¯ : TOP,TOP-+TOP (its interpretation will be We borrow definitions of unificand, function composition) E-equivalence of unificands (denoted by ~), ¢’~ : K,T~ T ; system and disjunction of systems from [13]. ~¢Cv : K,T-OK; The empty substitution will be denoted e and for each f of CFun, f: Kk,TI,CLm --> T ~7 will denote a unificand f : Kk,TI,CLm-+CL, where k+l+m=arityof f, solution. without any and for each f of MFunT(resp. of MFunK), f : Kk,TI,CLm--~T, (resp. f : Kk,TI,CLm-’-~K)7 4 The Algorithm As usual we add to the signature three sets of Presentation of the algorithm is inspired by VTand VK, of sorts CL, T and both [2] and [13], its input is the equation to variables VCL, be solved, denoted by r=?s. K respectively. The following rules are used to transform a unificand, whichis a conjunction of equations to be solved, 7 These are three sets of skolem functions, one for each sort of terms introduced during the skolemization step we talked about in the previous section; neither translation, nor skolemization introduces ternts of sort TOP, this is why we do not use variables of sort TOP. into a "simpler" E-equivalent one in the following way: 42 and From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. considering all the rules of the form e/fi that Occur-check (OCC) x=?t A U can be applied on the equation e=(r=?s), or s--?r, of the unificand eAU, then e AU ;Ifx~ Vi=l...n (fi A U). Let Sd be the set of decomposableoperators of the signature, i.e such that Vf~ Sd: Clash of sortsl (CST1) (t:S=?U’V)A ;for Clash of sorts2 (CST2) :T°. ¯ ¯ °um:T)^ (tl :Ket2:K".¯ ¯ *tn:K=?u I :Tou2 ~V For us, Sd=CFunUMFunTU MFunKU{ ff%,~ }, only 8. ¯ is not decomposable V Decomposition (D) f(p 1 ..... pn)=?f(q1 ..... qn) (pl=?qlA...Apn=?qn) s=T or K 7 {f(pl ..... pn)=?f(ql ..... qn)} {pl=?qlA...Apn=?qn} and f~g, {f(pl ..... pn)=?g(ql ..... qm)} t and tax. 7 Eliminate (ELIM) x:s=?t:s ^ U ^ U[x~t] Clash of symbols (CSB) f(p1 ..... pn)=?g(q1 ..... qm) ^ ; Vf, g~Sd, f~g ; If x~ t and t~x (for S=T or K). Classical unification (UC) (t:s ¯ u=?t’:s ¯ v)A ; (for S=T or K) 7 (t=?t’AU=?V)^ Persistence (PER) Deletion (DEL) t=?t A U (tl:K°...*tn-l:K ° tn:K ° z:T ¯ u=?v) A U (tl°...°tn.1 °z’°x’*u=?vA tn=?ff(c(z’:T,x’:K) A Z=?~(Z’,X’))A 8 Decomposability of elements of CFunUMFunTUMFunK is given by lemma2.1.1 of [31 about free function symbols, that of ~ and ~ is easy to establish thanks to the rewrite systemR. 43 wherez’ and x’ are newvariables. From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. Actually, the aim of the PERrule is to bring of new operators. Weclaim that this method up a subterm of sort T to the leftmost will be helpful for manyother logic systems. occurrence More precisely, of the left-hand side term. it seems that many of the Sometimes, it will be possible to apply rule useful multimodallogic systems can be treated UCon the transformed equation as well as it this way, i.e. was possible on the initial one; this leads to problems to deduction redundancyin the generated set of solutions. theories, through, eventually, introduction of Therefore, the algorithm can be optimized by new operators. forcing problems to apply rule PER n times by reducing their deduction under equational This approach will face of theories combination; of consecutively. unification The algorithm we have presented above is structures...; sound, complete and terminates.. Wewill not structures; with non disjoint universes. Weare give the proofs, by lack of space, but they can currently be found in 161. theories in groups or other algebraic with more complex sort investigating such equational associated to complex multimodal logics. Conclusion and perspectives References In this paper we have exposed howto reduce a logical [0l Yves Auffray, "Rrsolution modale et logique des problem to a unification chemins", Th~se de l’Universit~ [1] Yves Auffray & Patrice Enjalbert, problem, and we have presented an algorithm which solves the latter. "Modal theorem proving : An equational view point", IJCAI, This algorithm computes a complete set of unifiers de Caen, 1989 1989 [2l H.-J.Biirckert, and "Unification theory", in K.H.Bl~ius & H.-J.Btlrckert, terminates, thus we get a terminating, sound Artificial in Artificial and complete theorem prover for the logic of 13] H.-J.Biirckert, knowledgeand time we are interested in (see "Deduction Systems in Intelligence", Intelligence, 1989 "On Equational theories, Unification, and (Un)Decidability", in Kirchner editor, "UNIFICATION",p69-115, 141 and[61). Ellis HorwoodSeries Academic Press, 1990 This extension of previous works has [4] Luis Farifias del Cerro & Andreas Herzig, "Deterministic Modal Logic for Automated Deduction", 9th European Conference on been madepossible by means of introduction Artificial 44 Intelligence, Stockholm, 1990 From: AAAI Technical Report FS-93-01. Compilation copyright © 1993, AAAI (www.aaai.org). All rights reserved. [5] Paul Gaborit, Alix Potet, Claudette Sayettat. [17] Madeleine Lescot and Fran~oise Debart, "Semantics and Validation Procedures of a "Extension de la logique des chemins pour des Multi-modal Logic for the Formalization of Multi-agents Universes", 9th European L.I.U.C, Conference on Artificial Stockholm, logiques multimodales", Les cahiers du Intelligence, 1990 expert system for aircrafts", [6] Olivier Gasquet, "Planification multi-agents: Technical Report I.R.I.T Proceedings of Orlando (USA). Pergamon Press, n°90-23 R, 1991. [19] G. S. Makanin, "The problem of solvability Toulouse, France, 1990 103. English translation of the 3rd ACMSymposium on Sbornik, the Principles of Distributed Systems, p50- in Math USSR 32, 1977 [20] Hans-JOrgen Ohlbach, "A Resolution Calculus for 81, 1984 Modal Logics", Thdse de l’Universitd 191 David Harel, "’Dynamic Logic", In Dov Gabbay and Frans Gilnthner editors, Kaiserslautern, Ilanbook of Modal Logics", Proceedings of 9th CADE, [101 Andreas Herzig, "Raisonnement automatique en LNCS310, Springer Verlag, 1988 logique modale et algorithme d’unification", 122] Jean-Pierre P6cuchet, ’*Equations avec constantes Thdse de l’Universit~ Paul Sabatier, 1989 et algorithme de Makanin", Thdse de [ 111 G6rard P.Huet & D.C.Oppen, "Equations and l’Universit~ rewrite rules : a survey", Formal Language the resolution 1980 principle", JACM12, 1965 [24] Michael Rusinowitch, ’*D6monstration "A Companion to automatique - Techniques de r66criture", Modal Logic", Methuen & Co. Lid, 1986 Inter6ditions, [ 13] Claude Kirchner, "Order-sorted equational unification", 5th Int. Conf. on Logic Programming, Seattle, de Rouen, 1981 [23] J.A.Robinson, "A machine oriented logic based on theory: Perspectives and ()pen Problem.~’, 1121 G.Hugues et M.J.Cresswell, de 1988 [21] Hans-Jtirgen Ohlbach, ’*A Resolution Calculus for Logic, D. Rudel, 1984 Academic Press, of equations in a free semigroup", Mat. Sbornik, [8] J. Halpern, "Knowledge and commonknowledge", Philosophical avoidance the World Congress on Expert Systems, r6solution et algorithmes d’unification", Proceedings 1990 [18] MamedeLima Marques, "A collision Paris, 1989 [25] Christel Seguin, "Une &udelogique de la coop6ration", Revue Fratu;aise d’lntelligence 1988 Artificielle, [14] Sarit Kraus and Daniel Lelunann, "Knowledge, 2, 1990 [26] JOrg H.Siekmann, ’*Universal unification", 7th belief and time", In Laurent Kott editor, Int. Conf. on Automated Deduction., Springer Automata, Languages and Programing, Verlag, 1984 Lecture Notes in Computer Science, 226, p186-195, Springer Verlag, 1986. [15] Saul Kripke, "Semantical Analysis of Modal Logic 1, Normal Propositional Calculi", Zeitschrift fiir mathematische Logik und Grunlagen der Mathematik, 9, 1963 I161 P. Lescanne, "Computer experiments with the REVETerm Rewriting Systems Generator", lOt.h Annual ACMSymposium on Principles of Progranlming Languages, 1983 45