Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Propositional calculus wikipedia , lookup
Mathematical logic wikipedia , lookup
Modal logic wikipedia , lookup
Intuitionistic logic wikipedia , lookup
Jesús Mosterín wikipedia , lookup
Model theory wikipedia , lookup
Mathematical proof wikipedia , lookup
Law of thought wikipedia , lookup
Laws of Form wikipedia , lookup
Mathematical Social Sciences 38 (1999) 215–240 Group knowledge is not always distributed (neither is it always implicit)q Wiebe van der Hoek*, Bernd van Linder 1 , John-Jules Meyer Utrecht University, Department of Computer Science, P.O. Box 80.089, 3508 TB Utrecht, The Netherlands Abstract In this paper we study the notion of group knowledge in a modal epistemic context. Starting with the standard definition of this kind of knowledge on Kripke models, we show that this definition gives rise to some quite counter-intuitive behaviour. Firstly, using a strong notion of derivability, we show that group knowledge in a state can always, but trivially, be derived from each of the agents’ individual knowledge. In that sense, group knowledge is not really implicit, but rather explicit knowledge of the group. Thus, a weaker notion of derivability seems to be more adequate. However, adopting this more ‘local view’, we argue that group knowledge need not be distributed over (the members of) the group: we give an example in which (the traditional concept of) group knowledge is stronger than what can be derived from the individual agents’ knowledge. We then propose two additional properties on Kripke models: we show that together they are sufficient to guarantee ‘distributivity’, while, when leaving one out, one may construct models that do not fulfil this principle. 1999 Elsevier Science B.V. All rights reserved. Keywords: Group knowledge; Kripke models 1. Introduction In the field of AI and computer science, the modal system S5 is a well-accepted and by now familiar logic to model the logic of knowledge (cf. Halpern and Moses, 1985; Meyer and van der Hoek, 1995b). Since the discovery of S5’s suitability for epistemic logic, many extensions and adaptations of this logic have been proposed. In this paper we look into one of these extensions, namely a system for knowledge of m agents *Corresponding author. Tel.: 131-30-2531454; fax: 131-30-2513791. E-mail address: [email protected] (W. van der Hoek) q This is an improved and extended version of van der Hoek et al. (1995). 1 Currently at ABN-AMRO Bank, Amsterdam. 0165-4896 / 99 / $ – see front matter 1999 Elsevier Science B.V. All rights reserved. PII: S0165-4896( 99 )00013-X 216 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 containing a modality for ‘ group knowledge’. Intuitively, this ‘group knowledge’ (let us write G for it) is not the knowledge of each of the agents in the group, but the knowledge that would result if the agents could somehow ‘combine’ their knowledge. The intuition behind this notion is best illustrated by an example. Let the formula w denote the proposition that P ± NP. Assume that three computer scientists are working on a proof of this proposition. Suppose that w follows from three lemmas: c1 , c2 and c3 . Assume that scientist 1 has proved c1 and therefore knows c1 . Analogously, for agents 2 and 3 regarding c2 and c3 5 (c1 ∧ c2 ) → w, respectively. If these computer scientists would be able to contact each other at a conference, thereby combining their knowledge, they would be able to conclude w. This example also illustrates the relevance of communication with respect to this kind of group knowledge: the scientists should somehow transfer their knowledge through communication in order to make the underlying implicit knowledge explicit. In this paper, we try to make the underlying notions that together constitute G explicit: What does it mean for a group of, say, m agents to combine their knowledge? We start by giving and explaining a clear semantical definition of group knowledge, as given by Halpern and Moses (1985). In order to make some of our points, we distinguish between a global and a local notion of (deductive and semantic) consequence. Then we argue that the defined notion of group knowledge may show some counter-intuitive behaviour. For instance, we show in Section 3 that, at the global level, group knowledge does not add deductive power to the system: group knowledge is not always a true refinement of individual knowledge. Then, in Section 4, we try to formalize what it means to combine the knowledge of the members of a group. We give one principle (called the principle of distributivity: it says that G that is derived from a set of premises can always be derived from a conjunction of m formulae, each known by one of the agents) that is trivially fulfilled in our setup. We also study a special case of this principle (called the principle of full communication), and show that this property is not fulfilled when using standard definitions in standard Kripke models. In Section 5 we impose two additional properties on Kripke models: we show that they are sufficient to guarantee full communication; they are also necessary in the sense that one can construct models that do not obey one of the additional properties and that at the same time do not verify the principle under consideration. In Section 6 we round off. Proofs of theorems are provided in Appendix A. 2. Knowledge and group knowledge Halpern and Moses (1985) introduced an operator to model group knowledge. Initially, this knowledge in a group was referred to as ‘implicit knowledge’ and indicated with a modal operator I. Since in systems for knowledge and belief, the phrase ‘implicit’ had already obtained its own connotation (cf. Levesque, 1984), later on the term ‘distributed knowledge’ (with the operator D) became the preferred name for the group knowledge we want to consider here (cf. Halpern and Moses, 1992). Since we do not want to commit ourselves to any fixed terminology we use the operator G to model this kind of ‘group knowledge’. From the point of view of communicating agents, G- W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 217 knowledge may be seen as the knowledge being obtained if the agents were fully able to communicate with each other. Actually, instead of being able to communicate with each other, one may also adopt the idea that the G-knowledge is just the knowledge of one distinct agent, to whom all the agents communicate their knowledge [this agent was called the ‘wise man’ by Halpern and Moses (1985); a system to model such communication was proposed by van Linder et al. (1994)]. We will refer to this reading of G (i.e., in a ‘send and receiving context’) as ‘a receiving agent’s knowledge’. We start by defining the language that we use. Definition 2.1. Let P be a non-empty set of propositional variables, and m [ N be given. The language + is the smallest superset of P such that: if w,c [ + then ¬w,(w ∧ c ),Ki w,Gw [ + (i # m). The familiar connectives ∨ , → , ↔ are introduced by definitional abbreviation in the def def usual way; ¡ 5p ∨ ¬p for some p [ P and ' 5¬¡. The intended meaning of Ki w is ‘agent i knows w ’ and Gw means ‘w is group knowledge of the m agents.’ We assume the following ‘standard’ [the system below was originally introduced by Halpern and Moses (1985)] inference system S5 m (G) for the multi-agent S5 logic extended with the operator G. Definition 2.2. The logic S5 m (G) has the following axioms: A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 any axiomatization for propositional logic (Ki w ∧ Ki (w → c )) → Ki c Ki w → w Ki w → Ki Ki w ¬Ki w → Ki ¬Ki w Ki w → Gw (Gw ∧ G(w → c )) → Gc Gw → w Gw → GGw ¬Gw → G¬Gw On top of that, we have the following derivation rules: R1 R2 £w,£w → c ⇒ £c £w ⇒ £Ki w, for all i # m. In words, we assume a logical system (A1,R1) for rational agents. Individual knowledge, i.e. the knowledge of one agent, is moreover supposed to be veridical (A3). The agents are assumed to be fully introspective: they are supposed to have positive (A4) as well as negative (A5) introspection. In the receiving agent’s reading, the axiom A6 may be understood as the ‘communication axiom’: what is known to some member of W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 218 the group is also known to the receiving agent. Note how the axiom A6 and the rule R2 also induce the principle of necessitation on the operator G. The other axioms express that the receiver has the same reasoning and introspection properties as the other agents of the group. In the group knowledge reading, A6 declares what the members of the group are; the other axioms enforce this group knowledge to obey the same properties that are ascribed to the individual agents. The derivability relation £ S5 m (G ) , or £ for short, is defined in the usual way. That is, a formula w is said to be provable, denoted £w, if w is an instance of one of the axioms or if w follows from provable formulae by one of the inference rules R1 and R2. To relate this notion of group knowledge to the motivating example of Section 1, in which the group is able to derive a conclusion w from the lemmas c1 , c2 and c3 , each known by one of the agents, let us consider the following derivation rule which is to be understood to hold for all w,ci , i # m: R3£(c1 ∧ ? ? ? ∧ cm ) → w ⇒ £(K1 c1 ∧ ? ? ? ∧ Km cm ) → Gw. Theorem 2.3. Let the logic S5 m (G)[R3 /A6] be as S5 m (G), with axiom A6 replaced by the rule R3. Then, the two systems are equally strong: for all w, S5 m (G)£w ⇔S5 m (G)[R3 /A6]£w. From the proof of this theorem, as given by Meyer and van der Hoek (1995b) (p. 263), one can even derive that this equivalence is obtained for any multi-modal logic that is normal with respect to G, i.e. for any logic X containing the axioms A7 and the rules R1 and R29:£w ⇒ £Gw we have an equivalence between X 1 A6 and X 1 R3. We now define two variants of provability from premises: one in which necessitation on premises is allowed and one in which it is not. Definition 2.4. Let c be some formula, and let F be a set of formulae. Using the relation £ of provability within the system S5 m (G) we define the following two relations £ 1 and £ 2 7 2 + 3 + : F £ 1 c (F £ 2 c ) ⇔ 'w1 , . . . ,wn with wn 5 c, and such that for all 1 # i # n: • • • • either wi [ F or there are j,k , i with wj 5 wk → wi or wi is an S5 m (G) axiom or wi 5 Kk s where s is such that H s 5 wj with j , i in the case of £ 1 , 2 s in the case of £ 2 . From a modal logic point of view, one establishes: 5£ 1 w ⇔ 5£ 2 w, such that £w is unambiguously defined as 5£ 1 w or 5£ 2 w, and £(w1 ∧ . . . ∧ wn ) → c ⇔ w1 , . . . ,wn £ 2 c ⇒ w1 , . . . ,wn £ 1 c. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 219 In other words, the deduction theorem holds for £ 2 , not for £ 1 . Note that the relation £ is more liberal than £ 2 in the sense that £ 1 allows for necessitation on premises, where £ 2 only applies necessitation to S5 m (G)-theorems. Our definition of £ 1 deviates from standard terminology of premises: in classical logic, the formulas w1 , . . . ,wn are called premises in a derivation of c if (w1 ∧ . . . ∧ wn ) → c is derivable. Thus, premises in our sense in the context of £ 1 play the role of logical axioms [cf. Kaneko and Nagashima (1997), where a related discussion on ‘logical’ versus ‘nonlogical axioms’ leads to a number of (in-)definability results in epistemic logic]. Also, one can prove the following connection between the two notions of derivability from premises. 1 Theorem 2.5. Let hK1 , . . . ,Km j * be the set of sequences of epistemic operators. Then hXw u X [ hK1 , . . . ,Km j * ,w [ F j£ 2 c ⇔F £ 1 c. From Meyer and van der Hoek (1995a) (Lemma A1), for the one agent case without group knowledge (i.e., S5 1 ), one even obtains hKw u w [ F j£ 2 c ⇔ F £ 1 c. From an epistemic point of view, when using £ 2 , we have to view a set of premises F as a set of additional given (i.e., true) formulae [non-logical axioms, in the terminology of Kaneko and Nagashima (1997)], whereas in the case of £ 1 , F is a set of known formulae (‘logical axioms’). For instance, when describing a game and the knowledge that is involved (see also Bacharach et al., 1997), it would be convenient to express the mutual knowledge of the players as premises under the £ 1 notion of derivability. Definition 2.6. A Kripke model } is a tuple } 5 kS,p,R 1 , . . . ,R m l where 1. S is a non-empty set of states, 2. p : S 3 P → h0,1j is a valuation to propositional variables per state, 3. for all 1 # i # m, R i 7 S 3 S is an (accessibility) relation. For any s [ S and i # m, with R i (s) we mean ht [ S u R i stj. We refer to the class of Kripke models as _m or, when m is understood, as _. The class of Kripke models in which the accessibility relations R i are equivalence relations, is denoted by 6 5 m (or by 6 5). Definition 2.7. The binary relation * between a formula w and a pair },s consisting of a model } and a state s in } is inductively defined by: },s* p⇔p (s, p) 5 1, },s*w ∧ c ⇔ },s*w and },s*c, },s*¬w ⇔ },s± u w, },s*Ki w ⇔ },t*w for all t with (s,t) [ Ri , },s*Gw ⇔ },t*w for all t with (s,t) [ R 1 > . . . > Rm . For a given w [ + and } [ 6 5, nw m 5 ht u },t*w j. For sets of formulae F, },s*F is 220 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 defined by: },s*F ⇔ },s*w for all w [ F. In this paper, a model } is always an 65-model. The intuition behind the truth definition of G is as follows: if t is a world which is not an epistemic alternative for agent i, then, if the agents would be able to communicate, all the agents would eliminate the state t. This is justified by the idea that the actual, or real state, is always an epistemic alternative for each agent (in 65-models, R i is reflexive; or, speaking in terms of the corresponding axiom, knowledge is veridical). Using the wise-man metaphor: this man does not consider any state which has already been abandoned by one of the agents. A formula w is defined to be valid in a model } iff },s*w for all s [ S; w is valid with respect to 65 iff } *w for all } [ 6 5. The formula w is satisfiable in } iff },s*w for some s [ S; w is satisfiable with respect to 65 iff it is satisfiable in some } [ 6 5. A set of formulae F is valid with respect to 65 iff each formula w [ F is valid with respect to 65. For a class of models # we define two relations between a set 2 of formulae and a formula: F * 1 # w iff ; } [ # ( } *F ⇒ } *w ) and F * # w iff ; } [ # ;s( },s*F ⇒ },s*w ). If the class symbol # is left out, we always mean 6 5m . Theorem 2.8. (Soundness and Completeness.) Let w be some formula, and let F be a set of formulae. The following soundness and completeness results hold. 1. £w ⇔ *w, 2. F £ 2 w ⇔ F * 2 w, 3. F £ 1 w ⇔ F * 1 w. 3. Group knowledge is not always implicit Within the system S5 m (G), all agents are considered ‘equal’: if we make no additional assumptions, they all know the same. Lemma 3.1. For all i, j # m, we have: £Ki w ⇔ £Kj w. Remark 3.2. One should be sensitive for the comment that Lemma 3.1 is a meta statement about the system S5 m (G), and as such it should be distinguished from the claim that one should be able to claim within the system S5 m (G) that all agents know the same: i.e., we do not have (nor wish to have) that £Ki w ↔Kj w, if i ± j. Considering £w as ‘w is derivable from an empty set of premises’ (cf. our remarks following Definition 2.4), Lemma 3.1 expresses ‘ When no contingent fact is known beforehand, all agents know the same’. In fact, one easily can prove that ‘ When no contingent facts are given, each agent knows exactly the S5 m (G)- theorems’, since we have, for each i # m: £w ⇔ £Ki w W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 221 Interestingly, we are able to prove a property like that in Lemma 3.1 even when the G operator is involved. Theorem 3.3. Let X and Y range over hK1 ,K2 , . . . ,Km ,Gj. Then: £Xw ⇔ £Yw. Theorem 3.3 has, for both the reading as group knowledge as well as that of a receiving agent for G, some remarkable consequences. It implies that the knowledge in the group is nothing else than the knowledge of any particular agent. We can also phrase this in terms of implicit knowledge. First, let us agree upon what it means to say that group knowledge is implicit: Definition 3.4. Given a notion of derivability £ a (where we will typically take a [ h 1 , 2 j), we say that group knowledge is strongly implicit (under £ a) if there is a formula w for which we have £ a Gw, but for all i # m Z a Ki w. It is said to be weakly implicit under £ a if there is a set of premises F for which F £ a Gw, but F Z a Ki w, for all i # m. Thus, Theorem 3.3 tells us that, using £ of S5 m (G), group knowledge is not strongly implicit! Although counter-intuitive at first, Theorem 3.3 also invites one to reconsider the meaning of £w as opposed to a statement F £ 1 w. When interpreting the case where F 5 5 as ‘initially’ or, more loosely, ‘nothing has happened yet’ (in particular, when no communication of contingent facts has yet taken place), it is perhaps not too strange that all agents know the same and thus that all group knowledge is explicitly present in the knowledge of all agents. With regard to derivability from premises, we have to distinguish the two kinds of derivation introduced in Definition 2.4. Lemma 3.5. Let w be a formula, F a set of formulae and i some agent. • F £ 1 Gw ⇔F £ 1 Ki w, • F £ 2 Ki w ⇒ F £ 2 Gw, • F £ 2 Gw £F £ 2 Ki w. The first clause of Lemma 3.5 states that for derivability from premises in which necessitation on premises is allowed, group knowledge is also not implicitly but explicitly present in the individual knowledge of each agent in the group (thus F may be considered an initial set of facts that are known to everybody of the group). The second and third clause indicate that the notion of group knowledge is relevant only for derivation from premises in which necessitation on premises is not allowed. In that case, group knowledge has an additional value over individual knowledge. A further investigation into the nature of group knowledge when necessitation on premises is not allowed is the subject of the next section. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 222 4. Group knowledge is not always distributed In the previous section we established that group knowledge is in some cases not implicit, but explicit. In particular, for provability per se and for derivation from premises with necessitation on premises, group knowledge is rather uninteresting. The only case where group knowledge could be an interesting notion on its own is that of derivability from premises without necessitation on premises. For this case, we want to formalize the notion of distributed knowledge. Although intuitively clear, a formalization of distributed knowledge brings a number of hidden parameters to the surface. Informally, we say that the notion of group knowledge is distributed if the group knowledge (apprehended as a set of formulae) equals the set of formulae that can be derived from the union of the knowledge of the agents that together constitute the group. The following quote is taken from a recent dissertation (Borghuis, 1994): Implicit knowledge is of interest in connection with information dialogues: if we think of the dialogue participants as agents with information states represented by epistemic formulae, then implicit knowledge precisely defines the propositions the participants could conclude to during an information dialogue . . . Borghuis (1994) means with ‘implicit knowledge’ what we call ‘group knowledge’. We will see in this section that using standard epistemic logic one cannot guarantee that group knowledge is precisely that what can be concluded during an information dialogue. To do so, we will first formalize the notion of distributivity using the notions of derivability used so far. 4.1. A deductive approach To formalise what we mean with distributed group knowledge, let us have a second look at derivation rule R3 first. It says that, if w follows from the conjunction of w1 , . . . ,wm , then, if we also have K1 w1 ∧ ? ? ? ∧ Km wm , we are allowed to conclude that w is known by the group, i.e. then also Gw. Now, we are inclined to call group knowledge distributed if the converse also holds, that is, if everything that is known by the group does indeed follow from the knowledge of the group’s individuals. Let us, for convenience, say that group knowledge is perfect (over the individuals) if both directions hold. What we now have to do is be precise about the phrase ‘follows from’ and about the role that can be played by premises here. Let us summarize our first approximation to the crucial notions in a semi-formal definition: Definition 4.1 (first approximation). 1. We say that G represents complete group knowledge if we have that for every w and w1 , . . . ,wm such that (K1 w ∧ ? ? ? ∧ Km wm ) can be derived, and also the implication (w1 ∧ ? ? ? ∧ wm ) → w, then we also have that Gw can be derived. 2. The group knowledge represented by G is called distributed if for every w such that W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 223 Gw is derivable we can find w1 , . . . ,wm such that: (1) together they imply w and (2) the conjunction (K1 w ∧ ? ? ? ∧ Km w ) is derivable. 3. If both items above hold we call the group knowledge represented by G perfect group knowledge. Observation 4.2. In S5 m (G), using £, G-knowledge is both complete group knowledge and distributed. The proof of Observation 4.2 shows that it holds for a trivial reason. For distributivity, if £Gw, by Theorem 3.3, we also have £K1 w, . . . ,£Km w, and then distributivity follows from £(w ∧ ? ? ? ∧ w ) → w. To show that G models complete group knowledge, one reasons as follows: from £K1 w1 ∧ ? ? ? ∧ Km wm and £(w1 ∧ ? ? ? ∧ wm ) → w one obtains £Gw1 ∧ ? ? ? ∧ Gwm and £G((w1 ∧ ? ? ? ∧ wm ) → w ) from which £Gw immediately follows. So, to obtain distributivity in a non-trivial way, we should allow for premises. However, we know from the previous section that we then have at least two notions of derivability to consider. Unfortunately, as the following observation summarizes, even when allowing premises, we trivially obtain distributivity. Observation 4.3. If the notion of derivability used in Definition 4.1 is to be F £ 1 c or F £ 2 c, we again obtain distributivity of G-knowledge in a trivial way. Let us try to understand the source of Observation 4.3. For both w 5 1 , 2 , if F £ w Gw, then we also have F £ w w, and hence F £ w (¡ ∧ ? ? ? ∧ ¡) → w. Obviously, we also have F £ w (K1 ¡ ∧ ? ? ? ∧ Km ¡), justifying the claim that group knowledge is distributed. From the two Observations 4.2 and 4.3 above, we conclude that, in order to have an interesting notion of distributed group knowledge, we have to allow for a subtle use of premises and of notions of derivability. Let us therefore introduce the following notation. Let k [ h0,1j be a characteristic function in the sense that, for every set F, k ? F 5 5 if k 5 0, and F else. Now, we want to distinguish three types of derivability: one to obtain the (K1 w1 ∧ ? ? ? ∧ Km wm ) formulas (the ‘knowledge conclusions’), one to obtain (w1 ∧ ? ? ? ∧ wm ) → w (the ‘logical conclusions’) and one to obtain formulas of type Gw (the ‘group knowledge conclusions’). For each type of derivability, we may choose to allow for premises or not, and we also choose between £ 1 and £ 2 . Definition 4.1. Let kk ,kl ,kg [ h0,1j and let K, L and G be variables over h 1 , 2 j. For each tuple L 5 kkk ,K,kl ,L,kg ,Gl we now define the following notions: we say that L-derivable group knowledge is complete if, for all sets of premises F and every formula w : kg ? F £ G Gw ⇐'w1 , . . . ,wm : [kk ? F £ K (K1 w1 ∧ ? ? ? ∧ Km wm ) & kl ? F £ L (w1 ∧ ? ? ? ∧ wm ) → w ]; (1) W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 224 we say that L-derivable group knowledge is distributed, if for every set of premises F, and every w : kg ? F £ G Gw ⇒ 'w1 , . . . ,wm : [kk ? F £ K (K1 w1 ∧ ? ? ? ∧ Km wm ) & kl ? F £ L (w1 ∧ ? ? ? ∧ wm ) → w ]. (2) If both (1) and (2) hold for a tuple L and every F and w, we say that the knowledge represented by G is perfect with respect to L. Thus, we have specialized our first attempt to define perfect group knowledge, in notions depending on the choices in L. There are, in principle, 64 such notions, but the following observations give a systematic account of them: the reader may also consult the summarising Table A.1 in Appendix A. Observation 4.4. For 49 possibilities for L 5 kkk ,K,kl ,L,kg ,Gl one can show that group knowledge is indeed distributed—but the proofs show that this is only established in a ‘trivial’ manner. In the observation above, ‘trivial’ refers to the fact distributivity is proven in a way similar to that in Observations 4.2 and 4.3 – for an exhaustive proof, the reader is referred to Appendix A. Those cases are summarized in Table A.1, with an a, b, c or d in the ‘yes’ column. Observation 4.5. For the 13 of the remaining 15 possibilities for L 5 kkk ,K,kl ,L,kg ,Gl one can show that group knowledge is not distributed; they are summarized in Table A.1 with the items e, f and g in the ‘no’ column. The remaining interesting cases are L 5 k1, 2 ,0,L,k1 ,1l which we take as a starting point for the next section. Note that, in fact, this is only one case, since if kl 5 0, the choice of L 5 2 , 1 is arbitrary. Also note that, apart from the technical results of the observations above, this remaining L seems to make sense: we do not distinguish between kk and kg ( 5 1), we take premises seriously but we only consider the case where K 5 G 5 2 , so that group knowledge can be called implicit, according to Section 3. Then, according to Observation 4.3 the only sensible choice for kl is 0. Let us formulate the notion of distributivity for the remaining L’s: F £ 2 Gw ⇒ 'w1 , . . . ,wm : F £ 2 (K1 w1 ∧ ? ? ? ∧ Km wm ) & £(w1 ∧ ? ? ? ∧ wm ) → w. (3) Thus, the conclusion w should be an S5 m (G)-consequence of the lemma’s w1 , . . . ,wm , for Gw and (K1 w1 ∧ ? ? ? ∧ Km wm ) that are derivable using the same rules: using premises but without applying necessitation to them. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 225 4.2. A semantic approach If we try to reformulate (3) semantically, we obtain If Gw is true in every situation that verifies F, then there should be w1 , . . . ,wm such that (w1 ∧ ? ? ? ∧ wm ) → w is valid and (K1 w1 ∧ ? ? ? ∧ Km wm ) is true in every situation verifying F. However, we think that the order of quantification in the phrase above deserves some more attention. Consider F 5 hK1 p ∨ K2 pj. Indeed, we have F * 2 Gp, but does p follow from formulas w1 ,w2 such that (K1 w1 ∧ K2 w2 ) is true in every situation },s for F ? No, obviously not (see Observation 4.6 below or its proof in Appendix A); this is too strong a requirement: instead of finding such w1 and w2 such that every situation },s for F satisfies (K1 w1 ∧ K2 w2 ), however we are able to find such formula for every situation that verifies F, since such a situation },s either satisfies K1 p or K2 p, so that we can take (w1 ,w2 ) to be either ( p,¡) or (¡, p). Observation 4.6. Although we have hK1 p ∨ K2 pj£ 2 Gp, there exist no w1 ,w2 with £(w1 ∧ w2 ) → p and hK1 p ∨ K2 pj£ 2 (K1 w1 ∧ K2 w2 ). Corollary 4.7. Group knowledge is not distributed in the sense of (3). So, according to the latter Corollary, also the remaining choices for L do not yield a notion of distributed knowledge. Let us thus push our argument about the order of quantification one step further, by introducing a principle of ‘full communication’: We say that group knowledge satisfies the principle of full communication if in every situation in which Gw is true one can find w1 , . . . ,wm such that (K1 w1 ∧ ? ? ? ∧ Km wm ) also holds in that situation, and, moreover, (w1 ∧ ? ? ? ∧ wm ) → w is a valid implication. Phrased negatively, group knowledge does not satisfy the principle of full communication if there is a situation in which the group knows something, i.e. Gw, where w does not follow from the knowledge of the individual group members. We will now show that in the logic S5 m (G), group knowledge does not satisfy the principle of full communication, either. First, note that the semantic counterpart of full communication reads as follows: ; },s,w 'w1 , . . . wm : [ },s*Gw ⇒ },s*(K1 w1 ∧ ? ? ? ∧ Km wm ) & (w1 ∧ ? ? ? ∧ wm )* 2 w ]. (4) Definition 4.8. Let } be a Kripke model with state s, and i # m. The knowledge set of i in },s is defined by: K(i, },s) 5 hw u },s*Ki w j. Then, one easily verifies that (4) is equivalent to 226 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 },s*Gw ⇒ <K(i, },s)* 2 w. i (5) Property (5) expresses that G-knowledge can only be true at some world s if it is derivable from the set that results when putting all the knowledge of all the agents in s together. For multi-agent architectures in which agents have the possibility to communicate (like, for instance, van Linder et al., 1994) principle (5) is rather relevant. This so-called principle of full communication captures the intuition of fact discovery (cf. Halpern and Moses, 1990) through communication. The principle of full communication formalizes the intuitive idea that it is possible for one agent to become a wise man by communicating with other agents: these other agents may pass on formulae from their knowledge that the receiving agent combines to end up with the knowledge that previously was implicit. As such, the principle of full communication seems highly desirable a property for group knowledge. It is questionable whether group knowledge is of any use if it cannot somehow be upgraded to explicit knowledge by a suitable combination of the agents’ individual knowledge sets, probably brought together through communication. Coming back to the example of the three computer scientists and the question whether P ± NP, if there is no way for them to combine their knowledge such that the proof results, it is not clear whether this statement should be said to be distributed over the group at all. Unfortunately, in the context of S5 m (G), the principle of full communication does not hold. The following counterexample describes a situation in which group knowledge cannot be upgraded to explicit knowledge (thereby answering a question raised by van Linder et al., 1994). Counterexample 4.9. Let the set P of propositional variables contain the atom q. Consider a Kripke model } such that • S 5 hx,x9,z,z9j, • p (q,x) 5 p (q,x9) 5 1, p (q,z) 5 p (q,z9) 5 0 and for all p [ P, p ( p,x) 5 p ( p,x9) and p ( p,z) 5 p ( p,z9) (the precise value, 0 or 1, of p does not matter), • R 1 is given by the solid lines in Fig. 1, and R 2 is given by the dashed ones (reflexive arrows are omitted). Observation 4.10. It holds that (see Appendix A for a justification): Fig. 1. A finite model for two agents. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 227 • K(1, },x) 5 K(2, },x), • },x*Gq, • q[ ⁄ K(1, },x) < K(2, },x) 5 K(1, },x). From this last observation it follows that the principle of full communication does not hold in this model: although the formula q is group knowledge, it is not possible to derive this formula from the combined knowledge of the agents 1 and 2. Thus it is not possible for one of these agents to become a wise man through receiving knowledge from the other agent. In terms of Borghuis: the model of Counterexample 4.9 shows that one can have group knowledge of an atomic fact q, although this group knowledge will never be derived during a dialogue between the agents that are involved. The ‘reason’ for this in our example is that the agents have no means in their language to distinguish between x and x9, or between z and z9. In Section 5 we will see that such a negative result can already be obtained in a model in which there are no such ‘copies of the same world’ (see also Fig. 3). Corollary 4.11. In S5 m (G), group knowledge does not satisfy the principle of full communication. 5. Models in which group knowledge is distributed In this section we will characterize a class of Kripke models, in which G-knowledge is always distributed, in the sense that they satisfy the principle of full communication. To be more precise, the models that we come up with will satisfy },s*Gw ⇔ <K(i, },s)* 2 w. i (6) Note that the principle (6) indeed is that of (5), combined with the validity of axiom A6. Definition 5.1. A Kripke model } 5 kS,p,R 1 , . . . ,R m l is called finite if S is finite; moreover, it is called a distinguishing model if for all s,t [ S with s ± t, there is a ws 1 ,t 2 such that },s*ws 1 ,t 2 and },t± u ws 1 ,t 2 . When considering an S5 1 -model as an epistemic state of an agent, it is quite natural to use only distinguishing models: such a model comprises all the different possibilities the agent has. One can show that for S5, questions about satisfiability of finite sets of formulae, and hence that of logical consequence of a finite set of premises, can be decided by considering only finite distinguishing models. In fact, one only needs to require such models to be distinguishing at the propositional level: any two states in such a simple S5-model differ in assigning a truth value to at least one atom (see, e.g., Meyer and van der Hoek, 1995b, Section 1.7). In the multiple-agent case, this distinguishing requirement needs to be lifted from the propositional level to the whole language +. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 228 The nice feature of finite distinguishing models is that sets of states can be named: Lemma 5.2. Let } 5 kS, . . . l be a finite distinguishing model. Then: ;X , S'aX [ + ;x [ S ( },x*aX ⇔x [ X). For a given set X, we call aX the characteristic formula for X. Lemma 5.3. Let } 5 kS,p,R 1 , . . . ,R m l be a given finite distinguishing model, with s [ S. Suppose that Z 7 S is such that for all z [ Z we have },z* z. Moreover, suppose X1 , . . . ,Xn 7 S are such that R i (s) 7 (X1 < ? ? ? < Xn < Z). Then },s*Ki (¬aX 1 → (¬aX 2 → ( . . . (¬aXn → z ) . . . ))). If one tries to link up the two lemmas above with the principle of full communication, one may take the following point of view. First of all, note that, given a state s, all agents at least share one epistemic alternative, which is s itself. Now, consider the set of epistemic alternatives for agent i in s, R i (s). Relative to agent j we can partition R i (s) into two subsets: a set Yj 7 R i (s) of worlds that are considered as epistemic alternatives for agent j, and a set Xj 5 Ri (s)\Yj of worlds that are possible alternatives for i, but not for j. Let aXj and aYj be the characteristic formulas for Xj and Yj , respectively. Then, agent i knows that the real world must be in either one of Xj and Yj , i.e. agent i knows that ¬aXj → aYj , and agent j knows ¬aXj . Thus, agent j may inform agent i so to speak which of i’s alternatives can be given up. Now, if all the agents h would communicate this information Xh to agent i, then finally agent i would end up with those epistemic alternatives that are considered possible by all of the agents. This argument is made more precise in the (proof of) the following theorem. Theorem 5.4. Let } be a finite distinguishing model. Then, in }, G-knowledge is distributed, in the sense that it satisfies (6). We finally observe that the two properties of Definition 5.1 are in some sense also necessary for distributivity. Let us say that in } group knowledge satisfies the principle of full communication if, for all states s and formulae w, Eq. (6) is satisfied. Then, we know already from Fig. 1 and Observation 4.10 that there is a finite 6 5 2 model in which this principle of full communication is not valid. One might wonder if there is a distinguishing model that falsifies the same principle (this model should then, by virtue of Theorem 5.4, be infinite). It is tempting to apply the following reasoning to the model } of Fig. 1. Take the canonical model } c for S5 m (G). There is a set G 5 hc u },x*c j which is a world in this canonical model } c , where },x is as in Fig. 1. G is a maximal consistent set, and hence, a member of the canonical model } c . Also, one has } c ,G *(K1 a ↔K2 a ), for all a, } c ,G *¬K1 q and also } c ,G *Gq. Thus, } c ,G demonstrates that the principle of full communication is violated in the canonical W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 229 model. This model } c moreover is infinite, and distinguishing, by definition ( its set of states is the set of all maximal consistent sets). However, we have to mention here that } c is in general not a model of the kind as defined in Definition 2.6; the true definition of },s*Gw is ;t(s,t) [ R G ⇒ },t*w, where R G can only be guaranteed to be some subset of R 1 > ? ? ? > R m . For more on this, the reader is referred to van der Hoek and Meyer (1992). But still, we claim the existence of an 6 5 2 model 1 that is distinguishing, in which G-knowledge is defined as in our Definition 2.7, and in which the principle of full communication does not hold. This model 1 is provided in Definition A.4 in Appendix A. Let us conclude this section by summarizing the discussion of the last paragraphs: Observation 5.5. • There is a finite, not distinguishing model, in which group knowledge does not fulfil the principle of full communication. • There is a distinguishing model, not finite, in which group knowledge does not fulfil the principle of full communication. 6. Conclusion We have given formal definitions of what it means for group knowledge to be implicit or to be distributed relative to a group. Basically, group knowledge is implicit, if there is some formula known by the group, but not by any of its members. Here, one has to be precise about the role to be played by premises, and about the notion of derivability that is used to conclude that something is known. It appears that, under the standard notion of derivability in S5 m (G), and when no premises are allowed, group knowledge is not implicit. The same holds if one allows for applications of the Necessitation rule to premises. Thus, the only case in which group knowledge may be implicit in a group is the case in which premises describe a situation rather than the knowledge present beforehand. Next, we studied whether group knowledge is always distributed. We made the notion of distributivity precise by saying that a formula is distributed (given some set of premises) if it can somehow, using some notion of derivability and some related set of premises, be derived from the union of the agents’ knowledge. We argued that there was one way of choosing the parameters such that distributivity is not trivialised or directly absent. We gave a semantic account of a variant of distributivity (called the principle of full communication): we then were able to show that this principle was still not guaranteed, in general. Then, we gave two properties of Kripke models that guarantee that group knowledge does allow full communication; we also showed that both properties are, in some sense, needed: giving up one of the properties enables one to falsify the principle mentioned. We think it quite remarkable that such an appealing and natural notion like group knowledge as discussed in this paper can give rise to rather complicated behaviour or 230 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 properties. For instance, it is also well known that the canonical models for a logic like S5 m (G) only have the property that the accessibility relation for G, say R G , is a subset of R 1 > ? ? ? > R m , and cannot be guaranteed to be equal to this intersection. In other words, this equality is not modally definable, although one can find a complete axiomatization for those models that satisfy it. For more on this, the reader is referred to van der Hoek and Meyer (1992). In that paper, the authors also show that definability of R 1 > ? ? ? > R m 5 R G is achieved when using graded modal logic. But in this logic, the assumption of distinguishability of a model seems to be unnatural; graded modal logic is typically designed to be able to reason about numbers (and hence copies) of worlds. Also, in van der Hoek and Meyer (1997) the same authors demonstrate that proving an epistemic logic involving both group knowledge and common knowledge complete with respect to S5 m (G)-like models can be quite complicated. Again, distinguishability of the models has to be given up. To be more precise, given a state },s for c, the technique of filtration in modal semantics often yields a finite distinguishing model } 9 with state s9, such that } 9,s9 still verifies c, which might suggest af first sight that finite distinguishing models are always obtained. It is not guaranteed, however, that this model } is in the appropriate class, i.e. if } is an 6 5m (G)-model, } 9 need not be. Finally, in a recent manuscript (Lomuscio and Ryan, 1997) a rather surprising property of group knowledge was mentioned in the context of so-called Interpreted Systems, systems that are widely accepted as a general semantics for distributed systems (this is also a major approach in Fagin et al., 1995). In such a system, one obtains Gw ↔w, for all w. Since such a system is by definition distinguishing (the states usually differ already on the objective level), in the finite versions the principle of full communication would be automatically obtained, but, at the same time, it might be a too strong, and perhaps optimistic principle: all facts about the world would be eventually obtained by the agents, would they be able to fully communicate. Acknowledgements This work was partially supported by ESPRIT WG No. 23531 (FIREworks) and the Human Capital and Mobility network EXPRESS. The contents of this paper evolved by fruitful discussions with and comments from several people, and can thus be conceived as a good example of a body of distributed knowledge, obtained by repeated and fruitful communication. More particularly, Mark Ryan and Elias Thijsse gave useful comments on previous versions of this paper. Frank de Boer gave helpful suggestions to simplify the presentation of the model of Fig. 3. We also thank the participants of LOFT2 (the Second Conference on Logic and the Foundations of the Theory of Games and Decisions) for the inspiring discussion on this paper. Finally, the comments of two anonymous referees of Mathematical Social Sciences helped to further improve our presentation. Appendix A. Proofs Proof of Theorem 2.5. Our proof is inspired by that of Lemma A1 of Meyer and van W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 231 der Hoek (1995a), where a similar property is proven for an ‘S5 1 -like’ logic. Let us abbreviate hX u X [ hK1 , . . . ,Km j 1 j to W (the set of non-empty words over the alphabet hK1 , . . . Km j) and, for a set of premises F, let us write WF for hXw u X [ W,w [ F j. We first prove the following lemma: Lemma A.1. For all F,w,i # m: WF £ 2 c ⇔WF £ 2 Ki c. Proof. The ‘⇐’ direction follows immediately from an application of R1 to Ki c and axiom A3. The ‘ ⇒ ’ direction is proven by induction on the length l(p ) of a shortest £ 2 -proof p 5 kc1 , . . . cl(p ) 5 c l of c from WF, i.e. we show that, for all n and all c, If WF £ 2 c has a shortest proof of length n, then WF £ 2 Ki c. (A.1) • If l(p ) 5 1 then 1.1.either c [ WF, i.e. c 5 Xw for a sequence X 5 Kj 1 Kj 2 . . . Kj t [ W and w [ F. Obviously, Ki Xw [ WF, and hence we also have WF £ 2 Ki c. 1.2.Or c is an axiom of S5 m (G). By necessitation, we also have £Ki c and thus WF £ 2 Ki c. • Suppose that (A.1) holds for all WF £ 2 a with l(p ) # k(k $ 1) and suppose now that WF £ 2 c has a shortest proof p 5 kc1 , . . . ck 11 ( 5 c )l. 2.1.We do not have to consider the case that c [ WF, since then p would not be a shortest proof. The same holds for the case c being an S5 m (G)-axiom. 2.2.Suppose c 5 Kj b with £b. By necessitation, we then also have £Ki b and we prove Ki c 5 Ki Ki b by applying A4 and R1, yielding £Ki c and thus WF £ 2 Ki c. 2.3.The only other possibility is that ck is obtained by an application of modus ponens to cj and cj → ck 11 , j # k. Using the induction hypothesis on the proofs for cj and cj → ck11 , we obtain WF £ 2 Ki cj , WF £ 2 Ki (c → ck 11 ). Applying R1 and A2 to the latter two conclusions yields WF £ 2 Ki ck 11 , i.e. WF £ 2 Ki c. j We now reformulate and prove Theorem 2.5: Theorem 2.5. Let hK1 , . . . Km j 1 be the set of non-empty sequences of epistemic operators. Then hXw u X [ hK1 , . . . Km j 1 ,w [ F j£ 2 c ⇔F £ 1 c. Proof. (‘ ⇒ ’): Suppose that WF £ 2 c. Since £ 1 extends £ 2 we then also have WF £ 1 c. Moreover, we also have that F £ 1WF (meaning that F £ 1 Xw for any X [ W,w [ F ). Combining F £ 1Wf and WF £ 1 c then yields F £ 1 c. (‘⇐’): Suppose that F £ 1 c, with p 1 5 kc1 . . . ck ( 5 c )l a £ 1 -proof of c from F. Let `(p ) be the number of applications of the Necessitation rule in a proof p to non-S5 m (G)-theorems. We prove, for all n and all c : If F £ 1 c has a proof p 1 with `(p 1 ) 5 n, then WF £ 2 c. (A.2) • If `(p 1 ) 5 0, we replace every step in p 1 of the form wj , which is justified by 232 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 wj [ F, by the three steps Ki wj ,Ki wj → wj ,wj to obtain a proof k (p 1 )2 for WF £ 2 w. (Note that the resulting proof k (p 1 )2 indeed only uses premises from WF.) • Suppose that (A.2) holds for all p 1 with `(p 1 ) # k, and we have a F £ 2 c with a proof p 1 with `(p 1 ) 5 k 1 1, say p 1 5 kc1 . . . cr , cr 11 , . . . ck 11 ( 5 c )l, where cr 11 is obtained from ch (h # r) as a result of the last Necessitation step to previous formulas. So, cr 11 5 Kj ch , for some j # m. Thus, if p 9 1 is the proof for F £ 1 ch , then `(p 9 1 ) # k. The induction hypothesis yields WF £ 2 ch . By Lemma A.1, we obtain WF £ 2 Kj c. Note that the remaining part s 1 of the £ 1 -proof p 1 with s 1 5 k(ck 11 ( 5 c ) . . . c r 1 1( 5 Kj ch )l from F does not involve Necessitation to formulas in this sequence, so `(s 1 ) 5 0. Thus, as above, we can transform it in a £ 2 -proof k (s 1 )2) for Kj c from WF (replacing each assumption in s 1 of the form cv ( [ F ) by the triple K1 cv ( [ WF ),K1 cv → cv ,cv ). j Proof of Theorem 2.8. • This is a standard result in modal logic; see, for instance, Hughes and Cresswell (1984) or Fagin et al. (1995) and Meyer and van der Hoek (1995b) for this specific multi-agent epistemic logic S5 m (G). This property is also referred to as ‘weak soundness and completeness’. • This is the standard notion of adding premises in modal logic. Proofs of this ‘strong soundness and completeness’ result can again be found in Hughes and Cresswell (1984) or Fagin et al. (1995) and Meyer and van der Hoek (1995b). • For the ⇒ -direction, one only has to check that applying Necessitation to premises in F is * 1 -valid. The argument for ⇐ runs as follows. Suppose that we have F and w for which F Z 1 w. By Theorem 2.5, we then have WF Z 2 w, where W is defined as in the proof for Theorem 2.5 above. Thus, we find some Kripke model } and world s for which },s*WF and },s± u w. Using a standard argument in modal logic (see, for instance, Theorem 2.12 of Meyer and van der Hoek (1995a)), one may assume that the model } is generated from s: for all w [ S there is a path → v 5 ks 5 w 1 ,w 2 , . . . w n 5 wl such that for each w i and w i 11 (1 # i , n) there is some vi # m with R vi w i w i 11 . We now claim that } *F and } ± u w, which is equivalent to F± u 1 w. To show that } *F, let c [ F and let w be an arbitrary element in the set → of states, S. We know that },s*WF. Let the path v be as above, and let, as above again, for vi # m, R vi w i w i 11 . Now, Kv 1 Kv 2 . . . Kvn 21 c is an element of WF, so },s*Kv 1 Kv 2 . . . Kvn 21 c. But then, also },w*c, which was to be proven. The fact that } ± u w follows immediately from the fact that },s± u w. j On Observations 4.4 and 4.5. Table A.1 should be read as follows: the first column gives the name of the corresponding horizontal line; the second column give possible tuple L and then, the ‘yes’ column give a reason why this tuple gives rise to the claim that the corresponding notion of group knowledge indeed may be called distributed (see the proof of Observation 4.4 below). The ‘no’ column, to the contrary, argues why the notion cannot be called distributed (see the proof of Observation 4.5). Then, the same exercise is done for the tuple L9 that is obtained from L by changing kk from 0 to 1. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 233 Table A.1 Table summarizing Observations 4.4 and 4.5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 kk K kl L kg G Yes 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 a,b,c a,c No e e,f a,b,c a,b,c e e c c b f c c b b a,b,c,d a,c,d e e,f a,b,c,d a,b,c,d e e c,d c,d b f c,d c,d b b kk K kl L kg G Yes 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 b,c c No h g b,c b,c h g c c b g c c b b b,c c d d b,c b,c d d c c b,d d c c b,d b,d Proof of Observation 4.4. We claim that the following tuples L 5 kkk ,K,kl ,L,kg ,Gl make the notion of group knowledge distributed – but in a trivial way. (a) Tuples of type k0,K,0,L,0,Gl, i.e. no premises allowed. This was already observed in Observation 4.2, and the argument is given also in Section 4.1. (b) Suppose that kg 5 kl , and that G 5 L or G 5 2 ,L 5 1 . Then, if kg ? F £ G Gw, then also kg ? F £ G w and, under the specified conditions, also kl ? F £ L (¡ ∧ ? ? ? ∧ ¡) → w. Of course, one also has kk ? F £ K (K1 ¡ ∧ ? ? ? ∧ Km ¡), independent of kk and K. Note how this item generalizes (the proof of) Observation 4.3. (c) Tuples L for which kg 5 0 (note how this item generalizes item (a)). Under this L, suppose that £Gw. Then, by Theorem 3.3, we have £(K1 w1 ∧ ? ? ? ∧ Km wm ) and hence kk ? F £ K (K1 w1 ∧ ? ? ? ∧ Km wm ). Also, one has kl £ L (w ∧ ? ? ? ∧ w ) → w. Thus, by this 234 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 item, we have judged half of all the possibilities to obtain distributed group knowledge as trivial. (d) Here, K 5 1 and kg 5 kk . The crucial observation now is that if kg ? F £ G Gw, then kk ? F £ K Ki w. Obviously, we also have kl ? £ L (w ∧ ? ? ? ∧ w ) → w. Proof of Observation 4.5. The following tuples L 5 kkk ,K,kl ,L,kg ,Gl make the notion of group knowledge to be not distributed. (e) kk 5 kl 5 0, kg 5 1. Let F 5 hGpj, where p is an atom. Suppose that we would have w1 , . . . ,wm for which £(K1 w1 ∧ ? ? ? ∧ Km wm ) and £(w1 ∧ ? ? ? ∧ wm ) → p. Then (using A3 and R1) we would also have £p, which we do not. (f) kg 5 1, G 5 1 and, as a first subcase, suppose kk 5 0 and L 5 2 . Let F 5 h p,Gp → qj. We have F £ 1 Gq. Suppose for this subcase that there are w1 , . . . wm with £ K (K1 w1 ∧ ? ? ? ∧ Km wm ) and h p,Gp → qj£ 2 (w1 ∧ ? ? ? ∧ wm ) → q. Then we would have h p,Gp → qj£ 2 q. But the fact that h p,Gp → qj± u 2 q is easily verified, and then 2 Theorem 2.8 yields h p,Gp → qj£ q. (g) As a second subcase to the previous item, suppose, on top of kg 5 1, G 5 1 , that kk 5 1, K 5 2 and that either kl 5 0 or L 5 2 . As in the previous item, we then have that if there would be w1 , . . . wm for which h p,Gp → qj£ K (K1 w1 ∧ ? ? ? ∧ Km wm ), and kl ? h p,Gp → qj£ K (w1 ∧ ? ? ? ∧ wm ) → q, then we would also have h p,Gp → qj£ 2 q, which we do not, as argued above. (h) The remaining tuples are L 5 kkk ,K,kl ,L,kg ,Gl 5 k1, 2 ,0,L,1, 2 l, with L 5 2 2 , 1 . That they do not guarantee distributivity of group knowledge follows from Observation 4.6. Proof of Observation 4.6. In Observation 4.6 we claimed the following: Although we have hK1 p ∨ K2 pj£ 2 Gp, there exist no w1 ,w2 with £(w1 ∧ w2 ) → p and hK1 p ∨ K2 pj£ 2 (K1 w1 ∧ K2 w2 ). Consider the model of Fig. 2 (reflexive arrows are omitted; R 1 is denoted with the solid line, R 2 with a dashed one). Suppose that there would be formulas w1 and w2 of the type described above. Since we both have },s*K1 p ∨ K2 p and },t*K1 p ∨ K2 p, we have },s*(K1 w1 ∧ K2 w2 ) and },t*(K1 w1 ∧ K2 w2 ). Thus, we have Fig. 2. The model proving Observation 4.6. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 235 },u*w1 ∧ w2 . However, this is not compatible with £(w1 ∧ w2 ) → p, thus such w1 and w2 do not exist. j Proof of Corollary 4.7. For F 5 hK1 p ∨ K2 pj we have F £ 2 Gp, but we also saw that there are no w1 and w2 such that £(w1 ∧ w2 ) → p and for which F £ 2 (K1 w1 ∧ K2 w2 ), which is sufficient to disprove distributivity for the tuples L 5 k1, 2 ,0,L,1, 2 l, L 5 1 , 2 . j Proof of Observation 4.10. Observation 4.10 claimed about 1 that: 1. K(1, },x) 5 K(2, },x), 2. },x*Gq, 3. q [ ⁄ K(1, },x) < K(2, },x) 5 K(1, },x). Proof. We first state and prove the following lemma. Lemma A.2. For all u [ hx,zj, and all w : },u*w ⇔ },u9*w. Proof of Lemma A.2. Using induction on w. For atoms in P this follows by the definition of the valuation p. The proof for the connectives ∧ , ∨ and ¬ is straightforward. So let us assume the statement has been proven for formula c and consider w 5 K2 c. If },x*K2 c, we have that },u*c for u 5 x,z. The induction hypothesis then yields that },u9*c for u 5 x,z, and thus },x9*K2 c. Similar arguments hold for the other direction, as well as for the world z in the model, and also for w 5 K1 c,Gc. End of Proof of Lemma A.2. j Now, the first item of Observation 4.10 is proven as follows. Suppose that c [ K(2, },x), that is, },x*K2 c. Thus, by definition of K2 , we have },u*c for u 5 x,z. Then, by the previous lemma, we also have that z9 verifies c. But then, obviously, we have },x*K1 c. To see the second item, observe that (x,v) [ R 1 > R 2 ⇔v 5 x. And, since q is true in x, we have },x*Gq. Finally, since q is false in z9, we have that K1 q is false at x, so q [ ⁄ K(1, },x). The equality K(1, },x) < K(2, },x) 5 K(1, },x) follows immediately from the first item. j Proof of Lemma 5.2. This is easily first obtained for singletons X 5 hxj. Since } 5 kS, . . . l is finite, we can enumerate the worlds that differ from x as s 1 ,s 2 , . . . s n . Let ax be (wx 1,s 12 ∧ wx 1,s 22 ∧ . . . ∧ wx 1,s 2n ). Then, obviously },s*ax ⇔s 5 x. Subsequently, one puts aX 5 k x[X ax . j Proof of Lemma 5.3. Let t [ R i (s). We have to show that },t*¬aX 1 → (¬aX 2 → ( . . . (¬aXn → z ) . . . )) (*). Note that, for arbitrary x [ S and formulas b, we have ( },x*¬aXj → b )⇔(x [ ⁄ Xj ⇒ },x* b ). So, (*) is equal to t[ ⁄ X1 ⇒ (t [ ⁄ X2 ⇒ ( . . . (t [ ⁄ Xn ⇒ },t* z ) . . . )) (**). Now, (**) is true because t [ (X1 < ? ? ? < Xn < Z) and, for all x [ S, x [ Z ⇒ },x* z. j 236 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 Proof of Theorem 5.4. For the ‘⇐’ direction, suppose < i K(i, },s)* 2 w, and let t be such that (s,t) [ R 1 > . . . R m . Then, for every b [ < i K(i, },s)* 2 w we have },t* b and, thus, },t*w. By definition of t, we may conclude that },s*Gw. For the ‘ ⇒ ’-direction, suppose },s*Gw. Let us consider R 1 (s): we can cover it with possibly overlapping regions: let Z 7 R 1 (s) be R 1 (s) > ? ? ? > R m (s), and for every agent i, let Xi 5 hx [ R 1 (s) u x [ ⁄ R i (s)j. Thus, Xi is the set of worlds considered possible by agent 1, but not by agent i. If the agents were able to communicate, agent i could tell agent 1 that each x [ Xi is not a good epistemic alternative, so that the agent 1 can eliminate all worlds x that appear in some set Xi , thus ending up in exactly those alternatives in R 1 (s) > ? ? ? > R m (s). Formally, we put Lemma 5.3 to work. Note that R 1 (s) 7 (X1 < ? ? ? < Xm < Z). By Lemma 5.2, we find a characteristic formula aXi for every Xi , and, by definition of Z, we have },z*w, for all z [ Z. Then, Lemma 5.3 guarantees that },s*K1 (¬aX 1 → (¬aX 2 → ( . . . (¬aXn → w ) . . . ))) and, by definition of Xi and aXi , we also have },s*Ki ¬aXi (for, if },s± u Ki ¬aXi there would be a t [ R i (s) with },t*aXi , i.e. t [ Xi , which contradicts the definition of Xi ). Thus, h¬aX 1 → (¬aX 2 → ( . . . (¬aXn → w ) . . . ),¬aX 1 ,¬aX 2 , . . . ¬aXm j 7 < i K(i, },s), and thus < i K(i, },s)* 2 w. j Proof of Observation 5.5. • The proof of this claim is provided by the model of Fig. 1. • This follows from Theorem A.5, presented below. Before formally introducing our next Kripke model, we need some preliminary notions and notation. Definition A.3. We now assume that the set of atomic propositions P consists of an atom q together with infinitely atoms pi (i $ 1). In the following, we will use infinite binary sequences for the representation of worlds in our model. Here, a binary sequence is either an infinite sequence s 5 ks 1 ,s 2 , . . . l, or a finite sequence s 5 ks 1 ,s 2 , . . . ,s n l with each s i [ h0,1j. In the finite case, we say that the length l(s) of s is n. Let us denote the set of all the infinite sequences by IBS, and the finite sequences by FBS. It is convenient to define, for a finite sequence u 5 ku 1 , . . . ,u n l and an arbitrary binary sequence v the concatenation u;v of u and v, as u;v 5 ku 1 , . . . u n ,v1 ,v2 , . . . l. Also, for a finite sequence u, we define u 1 5 u, and u n 11 5 u;u n (n . 1). For any finite sequence u of length n, u 5 ku 1 , . . . u n l, let the infinite sequence u v [ IBS be ku 1v ,u 2v , . . . l, where u iv 5 u i mod n . Another way of defining u v would be as the solution of the equivalence u v 5 u;u v . Thus, k0,1,1l v 5 k0,1,1,0,1,1,0,1,1, . . . l, k1l v 5 k1,1,1, . . . l and k0l v 5 k0,0,0, . . . l. We call a sequence t a suffix of a given sequence u iff for some finite sequence v we have u 5 v;t (v may be the empty sequence e ). Finally, the equivalence relation ; s on IBS holds between sequences with an equal suffix, that is, u ; s t iff there is some v [ IBS and some u 1 ,t 1 [ FBS such that u 5 u 1 ;v and t 5 t 1 ;v. Given this relation ; s , for any u [ IBS, [u] 5 ht [ IBS u u ; s tj. Definition A.4. Let 1 5 kS,p,R 1 ,R 2 l be the Kripke model of Fig. 3, that is, S 5 S1 < S2 < S3 < S4 7 IBS, where S1 5 [k1l v ], S2 5 [k0l v ], S3 5 [k0,0,1l v ], and S4 5 [k1,1,0l v ]. W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 237 Fig. 3. An infinite distinguishing model for two agents. We have chosen the worlds x, z, x9 and z9 as named members of the sets S1 , S2 , S3 and S4 , respectively. We can choose for these worlds the representatives of the equivalence class that defines the Si that contains that world: then, for instance, we get x 5 k1l v , x9 5 k0,0,1l v , and so on. Then, we put p (u)(q) 5 1 iff u [ S1 < S3 , and p (u)( pi ) 5 1 iff u i 5 1. Thus, for instance, in each state u from S1 , q is true, and, although some of the p’s may be false in such a world in S1 , there is some index k u [ N, for which all pi ’s are true, if i $ k u , since u ; s k1l v . The relations R 1 and R 2 are indicated by the straight and dotted lines, respectively, and formally defined as follows. R 1 is the equivalence relation with the two equivalence classes S1 < S4 and S2 < S3 , whereas the equivalences classes of R 2 are S1 < S2 and S3 < S4 . Note that the model 1 is a distinguishing infinite model. The relation between the model 1 just defined and the model } of Fig. 1 is already partly suggested by the names of the distinguished worlds. For instance, in both cases, x and x9 are worlds in which q is false, and which are both R 1 and R 2 related to a world of z-type, but not R 1 > R 2 -related. However, in 1, it is not the case that x and x9 verify the same formulas, which is imposed by the mere fact that 1 is distinguishing (in this case, note that 1,x* p1 whereas 1,x9*¬p1 ). However, we will be able to prove that, although x and 238 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 x9 are not equal, the cluster S1 is ‘similar’ to S3 , whereas S2 is in the same sense related to S4 . We are now ready to state the main result about the model 1. Theorem A.5. Let 1 be the model defined as in Definition A.4, let x [ S1 , z [ S2 , x9 [ S3 and z9 [ S4 . Then: 1. K(1, 1,x) 5 K(2, 1,x), 2. },x*Gq, 3. q [ ⁄ K(1, },x) < K(2, },x) 5 K(1, },x). The proof of Theorem A.5 proceeds by combining a number of lemmas. First we need an additional definition. Definition A.6. Let n [ N, n $ 1. The language +n consists of the formulas of which the only propositional atoms are q, p1 , . . . , pn . Also, two sequences u,v [ IBS are nequivalent, written u ; n v if for all i # n, u i 5 vi . Now we can state the lemma that guarantees that the clusters S1 and S3 behave similarly (as do S2 and S4 ). Lemma A.7. Let n $ 1 be given. Then, for every world in S1 there is an n-equivalent world in S3 , and vice versa. The same relation holds between S2 and S4 . Put more formally, we have: ;n(;x [ S1 'x9 [ S3 (x ; n x9) & ;x9 [ S3 'x [ S1 (x9 ; n x)) & ;n(;z [ S2 'z9 [ S4 (z ; n z9) & ;z9 [ S4 'z [ S2 (z9 ; n z)). Proof. Note that a cluster Si contains all sequences with a common suffix. So, as an example, take x [ S1 . We know that x 5 kx 1 ,x 2 , . . . ,x n l;v;k1l v for some finite sequence v. But, then, obviously, by putting x9 5 kx 1 ,x 2 , . . . ,x n l;v;k0,0,1l v we have that x9 [ S3 and x9 ; n x. j Lemma A.8. Let n $ 1. Then: ;x [ S1 ;x9 [ S3 (x ; n x9⇔;w [ +n ( 1,x*w ⇔ 1,x9*w )), & ;x [ S2 ;z9 [ S4 (z ; n z9⇔;w [ +n ( 1,z*w ⇔ 1,z9*w )). (9) (10) Proof. For the main ‘⇐’ of (9), suppose that 1,x*w ⇔ 1,x9*w, for all w. We can now in particular take w 5 F (x) 5 n 1#i #n ki (x i , pi ), where ki (x i , pi ) 5 pi if x i 5 1, and ki (xi , pi ) 5 ¬pi else. This choice F (x) for w exactly encodes the first n bits of x, and, since by assumption 1,x9*F (x), we must have x9 ; n x. A similar argument proves ‘⇐’ in (10). The other direction is proven simultaneously for (9) and (10), by induction on the complexity of w. Let us concentrate on the first conjunct and assume that x ; n x9 and, for the base case, that w is a propositional atom. Then either it is q (which is W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 239 falsified by x as well as by x9) or it is some pi with i # n and, since x ; n x9 these two worlds also agree on pi . The proof for conjunctions and negations is straightforward, so let us now assume that w 5 K1 c, for some c [ +n with 1,x*K1 c. To reach a contradiction, assume that 1,x9*¬K1 c. It means that there is some z [ S2 with 1,z*¬c. Using the induction hypothesis on (10), we find some z9 [ S4 for which 1,z9*¬c. However, we have R 1 xz9, which indeed contradicts 1,x*K1 c. We conclude that 1,x9*K1 c. j Proof of Theorem A.5. 1. Suppose c [ ⁄ K(1, 1,x). Then, either there is an s [ S1 with 1,s*¬c and, in that case, we have c [ ⁄ K(1, 1,x) or there is a z9 [ S4 with 1,z9*¬c. Let n be such that c [ +n . By definition of the model 1, there is a z [ S2 , with z ; n z9. By Lemma A.8, we find that 1,z*¬c, and thus, by definition of R 2 , c [ ⁄ K(2, 1,x). 2. We have 1,x*Gq since q is true in the cluster S1 , which are exactly the R 1 > R 2 accessible worlds from x. 3. The atom q is not an element of K(1, 1,x), since q is not true in z9 [ S4 , with R 1 xz9. Proof of Observation 5.5. • This is proven by the model of Fig. 1 and Observation 4.10. • This follows from the model of Fig. 3 and Theorem A.5. References Bacharach, M.O.L., Gerard-Varet, L.A., Mongin, P., Shin, H.S. (Eds.), 1997. Epistemic Logic and the Theory of Games and Decisions, Kluwer, Dordrecht. Borghuis, T., 1994. On the interpretation of modalities in typed l-calculus. PhD thesis, Technical University Eindhoven. Fagin, R., Halpern, J.Y., Moses, Y., Vardi, M.Y., 1995. Reasoning About Knowlege, MIT Press, Cambridge, MA. Halpern, J.Y., Moses, Y., 1985. A guide to the modal logics of knowledge and belief. In: Proceedings of IJCAI-85, Los Angeles, CA, pp. 480–490. Halpern, J.Y., Moses, Y.O., 1990. Knowledge and common knowledge in a distributed environment. J. ACM 37 (3), 549–587. Halpern, J.Y., Moses, Y., 1992. A guide to the completeness and complexity for modal logics of knowledge and belief. Artificial Intelligence 54, 319–379. Hughes, G.E., Cresswell, M.J., 1984. A Companion to Modal Logic, Methuen, London. Kaneko, M., Nagashima, T., 1997. Axiomatic indefinability of common knowledge in finitary logic. In: Bacharach, M.O.L., Gerard-Varet, L.A., Mongin, P., Shin, H.S. (Eds.), Epistemic Logic and the Theory of Games and Decisions, Kluwer, Dordrecht, pp. 69–93. Levesque, H., 1984. A logic of implicit and explicit belief. In: Proceedings of the National Conference on Artificial Intelligence, pp. 198–202. Lomuscio, A., Ryan, M., 1997. A note on the relation between interpreted systems and Kripke models. Unpublished manuscript. Meyer, J.-J., van der Hoek, W., 1995a. A default logic based on epistemic states. Fundamenta Informaticae 23 (1), 33–65. 240 W. van der Hoek et al. / Mathematical Social Sciences 38 (1999) 215 – 240 Meyer, J.-J., van der Hoek, W., 1995b. Epistemic Logic for AI and Computer Science, No. 41 in Cambridge Tracks in Theoretical Computer Science, Cambridge University Press. Van der Hoek, W., Meyer, J.-J., 1992. Making some issues of implicit knowledge explicit. Int. J. Found. Comput. Sci. 3 (2), 193–224. Van der Hoek, W., Meyer, J.-J., 1997. A complete epistemic logic for multiple agents – combining distributed and common knowledge. In: Bacharach, M.O.L., Gerard-Varet, L.A., Mongin, P., Shin, H.S. (Eds.), Epistemic Logic and the Theory of Games and Decisions, Kluwer, Dordrecht, pp. 65–68. 1995. Group knowledge isn’t always distributed. In: Van der Hoek, W., van Linder, B., Meyer, J.-J.Ch. (Eds.), Proceedings of the Bi-annual Israelian Symposium on the Foundation of AI ’95, pp. 191–200. Van Linder, B., van der Hoek, W., Meyer, J.-J.Ch., 1994. Communicating rational agents. In: Nebel, B., Dreschler-Fischer, L. (Eds.), KI-94: Advances in Artificial Intelligence, Lecture Notes in Computer Science (subseries LNAI, Vol. 861, Springer, New York, pp. 202–213.