Founded World Views with Autoepistemic Equilibrium Logic

Defined by Gelfond in 1991 (G91), epistemic specifications (or programs) are an extension of logic programming under stable models semantics that introducessubjective literals. A subjective literal al-lows checking whether some regular literal is true in all (or in some of) the stable models of the program, being those models collected in a setcalledworld view. One epistemic program may yield several world views but, under the original G91 semantics, some of them resulted from self-supported derivations. During the last eight years, several alternative approaches have been proposed to get rid of these self-supported worldviews. Unfortunately, their success could only be measured by studying their behaviour on a set of common examples in the literature, since no formal property of "self-supportedness" had been defined. To fill this gap, we extend in this paper the idea of unfounded set from standard logic programming to the epistemic case. We define when a world view is founded with respect to some program and propose the foundedness property for any semantics whose world views are always founded. Using counterexamples, we explain that the previous approaches violate foundedness, and proceed to propose a new semantics based on a combination of Moore's Autoepistemic Logic and Pearce's Equilibrium Logic. The main result proves that this new semantics precisely captures the set of founded G91 world views.

• 28 publications
• 28 publications
• 1 publication
07/23/2019

About epistemic negation and world views in Epistemic Logic Programs

In this paper we consider Epistemic Logic Programs, which extend Answer ...
07/22/2019

Founded (Auto)Epistemic Equilibrium Logic Satisfies Epistemic Splitting

In a recent line of research, two familiar concepts from logic programmi...
08/17/2021

Thirty years of Epistemic Specifications

The language of epistemic specifications and epistemic logic programs ex...
12/20/2018

Splitting Epistemic Logic Programs

Epistemic logic programs constitute an extension of the stable models se...
08/05/2020

eclingo: A solver for Epistemic Logic Programs

We describe eclingo, a solver for epistemic logic programs under Gelfond...
10/01/2020

Constraint Monotonicity, Epistemic Splitting and Foundedness Could in General Be Too Strong in Answer Set Programming

Recently, the notions of subjective constraint monotonicity, epistemic s...
08/06/2021

1 Introduction

The language of epistemic specifications, proposed by Gelfond in 1991 [4], extends disjunctive logic programs (under the stable model [6] semantics) with modal constructs called subjective literals. Using these constructs, it is possible to check whether a regular literal is true in every stable model (written ) or in some stable model (written ) of the program. For instance, the rule:

 a←¬Kb (1)

means that must hold if we cannot prove that all the stable models contain . Subjective literals have been incorporated as an extension of the Answer Set Programming (ASP) paradigm [14, 18] in different solvers and implementations – see [12] for a recent survey. The definition of a “satisfactory” semantics for epistemic specifications has proved to be a non-trivial enterprise, as shown by the list of different attempts proposed so far [2, 4, 5, 9, 23, 24, 25]. The main difficulty arises because subjective literals query the set of stable models but, at the same time, occur in rules that determine those stable models. As an example, the program consisting of:

 b←¬Ka (2)

and (1) has now two rules defining atoms and in terms of the presence of those same atoms in all the stable models. To solve this kind of cyclic interdependence, the original semantics by Gelfond [4] (G91) considered different alternative world views or sets of stable models. In the case of program (1)-(2), G91 yields two alternative world views111For the sake of readability, sets of propositional interpretations are embraced with rather than ., and , each one containing a single stable model, and this is also the behaviour obtained in the remaining approaches developed later on. The feature that made G91 unconvincing, though, was the generation of self-supported world views. A prototypical example for this effect is the epistemic program consisting of the single rule:

 a←Ka (3)

whose world views under G91 are and . The latter is considered as counter-intuitive by all authors222This includes Gelfond himself, who proposed a new variant in [5] motivated by this same example and further modified this variant later on in [9]. because it relies on a self-supported derivation: is derived from by rule (3), but the only way to obtain is rule (3) itself. Although the rejection of world views of this kind seems natural, the truth is that all approaches in the literature have concentrated on studying the effects on individual examples, rather than capturing the absence of self-supportedness as a formal property. To achieve such a goal, we would need to establish some kind of derivability condition in a very similar fashion as done with unfounded sets [3] for standard logic programs. To understand the similarity, think about the (tautological) rule . The classical models of this rule are and , but the latter cannot be a stable model because is not derivable applying the rule. Intuitively, an unfounded set is a collection of atoms that is not derivable from a given program and a fixed set of assumptions, as happens to in the last example. As proved by [13], the stable models of any disjuntive logic program are precisely its classical models that are founded, that is, that do not admit any unfounded set. As we can see, the situation in (3) is pretty similar to but, this time, involves derivability through subjective literals. An immediate option is, therefore, extending the definition of unfounded sets for the case of epistemic programs – this constitutes, indeed, the first contribution of this paper.

Once the property of founded world views is explicitly stated, the paper proposes a new semantics for epistemic specifications, called Founded Autoepistemic Equilibrium Logic (FAEEL), that fulfills that requirement. In the spirit of [2, 25], our proposal actually constitutes a full modal non-monotonic logic where becomes the usual necessity operator applicable to arbitrary formulas. Formally, FAEEL is a combination of Pearce’s Equilibrium Logic [20], a well-known logical characterisation of stable models, with Moore’s Autoepistemic Logic (AEL) [17], one of the most representative approaches among modal non-monotonic logics. The reason for choosing Equilibrium Logic is quite obvious, as it has proved its utility for characterising other extensions of ASP, including the already mentioned epistemic approaches in [2, 25]. As for the choice of AEL, it shares with epistemic specifications the common idea of agent’s introspection where means that is one of the agent’s beliefs. The only difference is that those beliefs are just classical models in the case of AEL whereas epistemic specifications deal with stable models instead. Interestingly, the problem of self-supported models has also been extensively studied in AEL [11, 15, 19, 21], where the formula , analogous to (3), also yields an unfounded world view333Technically, AEL is defined in terms of theory expansions but each one can be characterised by a canonical S5-model with the same form of a world view [16, 22]. . Our solution consists in combining the monotonic bases of AEL and Equilibrium Logic (the modal logic KD45 and the intermediate logic of Here-and-There (HT) [8], respectively), but defining a two-step models selection criterion that simultaneously keeps the agent’s beliefs as stable models and avoids unfounded world views from the use of the modal operator . As expected, we prove that FAEEL guarantees the property of founded world views, among other features. Our main result, however, goes further and asserts that the FAEEL world views of an epistemic program are precisely the set of founded G91 world views. We reach, in this way, an analogous situation to the case of standard logic programming, where stable models are the set of founded classical models of the program.

The rest of the paper is organised as follows. Section 2 and 3 respectively revisit the background knowledge about equilibrium logic and epistemic specifications necessary for the rest of the paper. Section 4 introduces the foundedness property for epistemic logic programs. In Section 5, we introduce FAEEL and show that its world views precisely coincide with the set of founded G91 world views. Finally, Section 6 concludes the paper.

2 Background

We begin recalling the basic definitions of equilibrium logic and its relation to stable models. We start from the syntax of propositional logic, with formulas built from combinations of atoms in a set with operators and in the usual way. We define the derived operators ,  ,   and  .

A propositional interpretation is a set of atoms . We write to represent that classically satisfies formula . An is a pair (respectively called “here” and “there”) of propositional interpretations such that ; it is said to be total when . We write to represent that satisfies a formula  under the recursive conditions:

• [ topsep=2pt]

• iff

• iff and

• iff or

• iff both (i) and (ii) or

As usual, we say that is a model of a theory , in symbols , iff for all . It is easy to see that iff classically. For this reason, we will identify simply as and will use ‘’ indistinctly. By we denote the set of all classical models of . Interpretation is a stable (or equilibrium) model of a theory iff and there is no such that . We write to stand for the set of all stable models of . Note that by definition.

3 G91 semantics for epistemic theories

In this section we provide a straightforward generalisation of G91 allowing its application to arbitrary modal theories. Formulas are extended with the necessity operator according to the following grammar:

 \fF::=⊥∣a∣\fF1∧\fF2∣\fF1∨\fF2∣\fF1→\fF2∣Kφ for any atom a∈\and.

An (epistemic) theory is a set of formulas as defined above. In our context, the epistemic reading of is that “ is one of the agent’s beliefs.” Thus, a formula is said to be subjective if all its atom occurrences (having at least one) are in the scope of . Analogously, is said to be objective if does not occur in . For instance, is subjective, is objective and none of the two.

To represent the agent’s beliefs we will use a set of propositional interpretations. We call belief set to each element and belief view to the whole set . The difference between belief and knowledge is that the former may not hold in the real world. Thus, satisfaction of formulas will be defined with respect to an interpretation , possibly , that accounts for the real world: the pair is called belief interpretation (or interpretation in modal logic KD45). Modal satisfaction is also written (ambiguity is removed by the interpretation on the left) and follows the conditions:

• [ topsep=2pt]

• ,

• iff , for any atom ,

• iff and ,

• iff or ,

• iff or , and

• iff for all .

Notice that implication here is classical, that is, is equivalent to in this context. A belief interpretation is a belief model of iff for all and all – additionally, when , we further say that is an epistemic model of and abbreviate this as . Belief models defined in this way correspond to modal logic KD45 whereas epistemic models correspond to S5.

Take the theory corresponding to rule (1). An epistemic model must satisfy: or , for all . We get three epistemic models from , , , and and the rest of cases must force true, so we also get and . In other words, has the same epistemic models than . Note that rule (1) alone did not seem to provide any reason for believing , but we got three epistemic models above satisfying . Thus, we will be interested only in some epistemic models (we will call world views) that minimize the agent’s beliefs in some sense. To define such a minimisation we rely on the following syntactic transformation provided by [24].

[Subjective reduct] The subjective reduct of a theory  with respect to a belief view , also written , is obtained by replacing each maximal subformula of the form by: , if ; by , otherwise. Notice that is a classical, non-modal theory. Finally, we impose a fixpoint condition where, depending on whether each belief set is required to be a stable model of the reduct or just a classical model, we get G91 or AEL semantics, respectively.

[AEL and G91 world views] A belief view  is called an AEL-world view of a theory iff , and is called a G91-world view of iff .

[Example 3 revisited] Take any such that . Then, with and . None of the two satisfy so cannot be fixpoint for G91 or AEL. If instead, we get , whose classical models are and , but only the former is stable. As a result, is the unique AEL world view and the unique G91 world view.

Take now the theory corresponding to rule (3). If we get and so is an AEL and G91 world view. If , the reduct becomes , a classical tautology with unique stable model . As a result, is the other AEL world view, while is the a second G91 world view.

As we can see, the difference between AEL and G91 is that we use classical instead of stable models, respectively. It is well known that adding the excluded middle axiom for all atoms makes equilibrium logic collapse into classical logic. This leads us to the following result. is an AEL world view of some theory  iff is a G91-world view of .

4 Founded world views of epistemic specifications

As we explained in the introduction, world view of is considered to be “self-supported” in the literature but, unfortunately, there is no formal definition for such a concept, to the best of our knowledge. To cover this lack, we proceed to extend here the idea of unfounded sets from disjunctive logic programs to the epistemic case. For this purpose, we focus next on the original language of epistemic specifications [4] (a fragment of epistemic theories closer to logic programs) on which most approaches have been actually defined.

Let us start by introducing some terminology. An objective literal is either an atom , its negation  or its double negation . A subjective literal is any of the formulas444We focus here on the study of the operator , but epistemic specifications also allow a second operator whose relation to is also under debate and, for this reason, we leave it future work. , or where an objective literal. A literal is either an objective or a subjective literal, and is called negative if it contains negation and positive otherwise. A rule is a formula of the form

 a1∨…∨an←B1∧…∧Bm (4)

with , and , where each is an atom and each is a literal. For any rule like (4), we define its body as and its head , which we sometimes use as the set of atoms . When , and the rule is a constraint, whereas if then and the rule is a fact. The set collects all atoms occurring in positive objective literals in the body while collects all atoms occurring in positive subjective literals. An epistemic specification or program is a set of rules. As with formulas, a program without occurrences of  is said to be objective (it corresponds to a standard disjunctive logic program with double negation).

[Unfounded set] Let be a program and a belief view. An unfounded set with respect to and is a non-empty set of pairs where, for each , we have that and are sets of atoms and there is no rule with satisfying:

1. [itemsep=3pt,topsep=2pt]

2. with .

The definition works in a similar way to standard unfounded sets [13, Definition 3.1]. In fact, the latter corresponds to the first three conditions above, except that we use to check , as it may contain now subjective literals. Intuitively, each represents some potential belief set (or stable model) and is some set of atoms without a “justifying” rule, that is, there is no allowing a positive derivation of atoms in . A rule like that should have a true (condition 1) but not because of positive literals in (condition 2) and is not used to derive other head atoms outside (condition 3). The novelty in our definition is the addition of condition 4: to consider a justifying rule, we additionally require not using any positive literal in the body such that atom also belongs to any of the unfounded components in .

[Founded world view] Let be a program and be a belief view. We say that is unfounded if there is some unfounded-set  s.t., for every , we have and . is called founded otherwise.

When is an objective program, each pair corresponds to a standard unfounded set of some potential stable model  in the traditional sense of [13].

Given the single disjunctive rule suppose we check the (expected) world view . For and , rule satisfies the four conditions and justifies . The same happens for . So, is founded. However, suppose we try with instead. For we can form and and in both cases, the only rule in the program, , violates condition 3. As a result, is unfounded due to the set .

To illustrate how condition 4 works, let us continue with Example 3. ex:self-supporting.rule Theory is also a program. Given belief set we can observe that makes unfounded because the unique rule in does not fulfill condition 4: we cannot derive from a rule that contains . On the other hand, the other G91 world view, , is trivially founded.

Since Definition 4 only depends on some epistemic program and its selected world views, we can raise it to a general property for any epistemic semantics.

[Foundedness] A semantics satisfies foundedness when all the world views it assigns to any program are founded.

Approaches proposed after G91 do remove unfounded world views in the examples studied in the literature, but unfortunately, this does not mean that they generally satisfy foundedness. Let us consider a common counterexample. Take the epistemic logic program:

 a∨ba←Kbb←Ka (\program???)

whose G91-world views are and . These are, indeed, the two cases we analysed in Example 4. is again founded because keeps justifying both possible pairs, that is, and . However, for we still have the unfounded set which violates condition 3 for the first rule as before, but also condition 4 for the other two rules. Note how allows us to spot the root of the derivability problem: to justify in we cannot use because is part of the unfounded structure in the other pair , and vice versa. Since the variants by Gelfond in [5] (G11) and Kahl et al. [9] (K15) also assign the unfounded world view to (in fact, they coincide with G91 for this program), we can conclude that G11 and K15 do not satisfy foundedness either.

A more elaborated strategy is adopted by the recent approaches by Fariñas et al.  [2] (F15) and Shen and Eiter [23] (S17), that treat the previous world views as candidate solutions555In [2], these candidate world views are called epistemic equilibrium models while selected world views receive the name of autoepistemic equilibrium models., but select the ones with minimal knowledge in a second step. This allows removing the unfounded world view in Example 4, because the other solution provides less knowledge. Unfortunately, this strategy does not suffice to guarantee foundedness, since other formulas (such as constraints) may remove the founded world view, as explained below.

ex:unfounded Take the program . The constraint rules out because the latter satisfies . In G91, G11, F15 and S17, only world view is left, so knowledge minimisation has no effect. However, is still unfounded in since constraints do not affect that feature (their empty head never justifies any atom).

As a conclusion, semantics F15 and S17 do not satisfy foundedness either.

5 Founded Autoepistemic Equilibrium Logic

We present now the semantics proposed in this paper, introducing Founded Autoepistemic Equilibrium Logic (FAEEL). The basic idea is an elaboration of the belief (or KD45) interpretation already seen but replacing belief sets by HT pairs. Thus, we extend now the idea of belief view to a non-empty set of HT-interpretations and say that is total when for all of them, coinciding with the form of belief views we had so far. Similarly, a belief interpretation is now redefined as , or simply , where is a belief view and stands for the real world, possibly not in . Next, we redefine the satisfaction relation from a combination of modal logic KD45 and HT. A belief interpretation satisfies a formula , written , iff:

• [topsep=2pt]

• ,

• iff , for any atom ,

• iff and ,

• iff or ,

• iff both: (i) or ; and (ii) or , where .

• iff for all .

For total belief interpretations, this new satisfaction relation collapses to the one in Section 3 (that is, KD45). Interpretation is a belief model of iff for all and all – additionally, when , we further say that is an epistemic model of , abbreviated as .

[Persistence] implies .

A belief model just captures collections of HT models which need not be in equilibrium. To make the agent’s beliefs correspond to stable models we impose a particular minimisation criterion on belief models.

We define the partial order for belief interpretations and when the following three conditions hold:

1. and , and

2. for every , there is some , with .

3. for every , there is some , with .

As usual, means and . The intuition for is that contains less information than for each fixed component. As a result, implies for any formula without implications other than .

A total belief interpretation is said to be an equilibrium belief model of some theory  iff is a belief model of and there is no other belief model of such that . By we denote the set of equilibrium belief models of . As a final step, we impose a fixpoint condition to minimise the agent’s knowledge as follows. A belief view  is called an equilibrium world view of  iff: c+x* W  =    T (W, T) ∈EQB[ Γ] &

ex:self-supporting.rule2 Back to , remember its unique founded G91-world view was . It is easy to see that because and no smaller belief model can be obtained. Moreover, is an equilibrium world view of  since no other satisfies . The only possibility is but it fails because there is a smaller belief model satisfying . As for the other potential world view , it is not in equilibrium: we already have because the smaller interpretation also satisfies . In particular, note that and, thus, clearly satisfies .

The logic induced by equilibrium world views is called Founded Autoepistemic Equilibrium Logic (FAEEL). A first important property is: FAEEL satisfies foundedness.

A second interesting feature is that equilibrium world views are also G91-world views though the converse may not be the case (as we just saw in Example 5). This holds, not only for programs, but in general for any theory:

For any theory , its equilibrium world views are also G91-world views of .

In other words, FAEEL is strictly stronger than G91, something that, as we see next, is not the case in other approaches in the literature.

The following program:

 a∨bc←Ka⊥←¬c (5)

has no G91-world views, but according to G11, K15, F15 and S17 has world view . This example was also used in [1] to show that these semantics do not satisfy another property, called there epistemic splitting.

ex:unfounded Take again program  whose G91-world views were and . Since is unfounded, it cannot be an equilibrium world view (Theorem 5), leaving as the only candidate (Theorem 5). Let us check that this is in fact an equilibrium world view. First, note that because there is no model of such that . In fact, it is easy to see that is not a model of the rule if for any . Symmetrically, we have that too. Finally, we have to check that no other can form an equilibrium belief model. For the case , it is easy to check that does not satisfy . For , we have that because, for instance, the smaller is a model of .

Theorems 5 and 5 assert that any equilibrium world view is a founded G91-world view. The natural question is whether the opposite also holds. In Examples 5, 5 and 5 we did not find any counterexample, and this is in fact a general property, as stated below.

Given any program , its equilibrium world views coincide with its founded G91-world views.

An interesting observation is that in all the original examples of epistemic specifications [4, 7] used by Gelfond to introduce G91, modal operators occurred in the scope of negation. Negated beliefs never incur unfoundedness, so this feature could not be spotted using this family of examples. In fact, under this syntactic restriction, FAEEL and G91 coincide.

For any theory where all occurrences of are in the scope of negation, we have that the equilibrium world views and the G91-world views coincide.

Proposition 5 also holds for semantics [24, 25] that are conservative extensions of G91, as well as for G11. Apart from foundedness, [1] recently proposed other four properties for semantics of epistemic specifications. We analyse here three of them, omitting the so-called epistemic splitting due to lack of space.

1. supra-ASP holds when, for any objective theory , either has a unique world view or and has no world view.

2. supra-S5 holds when every world view of a theory is also an S5-model of  (that is, ).

3. subjective constraint monotonicity holds when, for any theory and any subjective constraint , we have that is a world view of iff both is a world view of and is not an S5-model of .

FAEEL satisfies supra-ASP, supra-S5 and subjective constraint monotonicity. All semantics discussed in this paper satisfy the above first two properties but most of them fail for subjective constraint monotonicity, as first discussed in [10]. In fact, a variation of Example 5 can be used to show that K15, F15 and S17 do not satisfy this property.

ex:nog91 Suppose we remove the constraint (last rule) from 5 getting the program . All semantics, including G91 and FAEEL, agree that 5 has a unique world view . Suppose we add now a subjective constraint . This addition leaves G91 and FAEEL without world views (due to subjective constraint monotonicity) the same happens for G11, but not for K15, F15 and S17, which provide a new world view  not obtained before adding the subjective constraint.

Tables 1 and 2 show a list of examples taken from Table 4 in [2] and their world views according to different semantics.

6 Conclusions

In order to characterise self-supported world-views, already present in Gelfond’s 1991 semantics [4] (G91), we have extended the definition of unfounded sets from standard logic programs to epistemic specifications. As a result, we proposed the foundedness property for epistemic semantics, which is not satisfied by other approaches in the literature. Our main contribution has been the definition of a new semantics, based on the so-called Founded Autoepistemic Equilibrium Logic (FAEEL), that satisfies foundedness. This semantics actually covers the syntax of any arbitrary modal theory and is a combination of Equilibrium Logic and Autoepistemic Logic. As a main result, we were able to prove that, for the syntax of epistemic specifications, FAEEL world views coincide with the set of G91 world views that are founded. We showed how this semantics behaves on a set of common examples in the literature and proved that it satisfies other three basic properties: all world views are S5 models (supra-S5); standard programs have (at most) a unique world view containing all the stable models (supra-ASP); and subjective constraints just remove world views (monotonicity). FAEEL also satisfies the property of epistemic splitting as proposed in [1], but we leave the proof and discussion for future work, together with a formal comparison with other approaches.

References

• [1] Cabalar, P., Fandinno, J., Fariñas del Cerro, L.: Splitting Epistemic Logic Programs. arXiv e-prints arXiv:1812.08763 (Dec 2018)
• [2]

Fariñas del Cerro, L., Herzig, A., Su, E.I.: Epistemic equilibrium logic. In: Proc. of the Intl. Joint Conference on Artificial Intelligence (IJCAI’15). pp. 2964–2970. AAAI Press (2015)

• [3] Gelder, A.V., Ross, K.A., Schlipf, J.S.: The well-founded semantics for general logic programs. J. ACM 38(3), 620–650 (1991)
• [4] Gelfond, M.: Strong introspection. In: Dean, T.L., McKeown, K. (eds.) Proceedings of the AAAI Conference. vol. 1, pp. 386–391. AAAI Press/The MIT Press (1991)
• [5] Gelfond, M.: New semantics for epistemic specifications. In: LPNMR. Lecture Notes in Computer Science, vol. 6645, pp. 260–265. Springer (2011)
• [6] Gelfond, M., Lifschitz, V.: The stable model semantics for logic programming. In: Proc. of the 5th Intl. Conference on Logic Programming (ICLP’88). pp. 1070–1080 (1988)
• [7] Gelfond, M., Przymusinska, H.: Reasoning on open domains. In: Proc. of the Intl. Conf. on Logic Programming and Non-Monotonic Reasoning (LPNMR’93). pp. 397–413. MIT Press (1993)
• [8] Heyting, A.: Die formalen Regeln der intuitionistischen Logik. Sitzungsberichte der Preussischen Akademie der Wissenschaften, Physikalisch-mathematische Klasse pp. 42–56 (1930)
• [9] Kahl, P., Watson, R., Balai, E., Gelfond, M., Zhang, Y.: The language of epistemic specifications (refined) including a prototype solver. Journal of Logic and Computation (2015)
• [10] Kahl, P.T., Leclerc, A.P.: Epistemic logic programs with world view constraints. In: ICLP (Technical Communications). OASICS, vol. 64, pp. 1:1–1:17. Schloss Dagstuhl - Leibniz-Zentrum fuer Informatik (2018)
• [11] Konolige, K.: On the relation between default and autoepistemic logic. Artif. Intell. 35(3), 343–382 (1988)
• [12] Lecrerc, A.P., Kahl, P.T.: A survey of advances in epistemic logic program solvers. In: Proc. of the 11th Intl. Workshop on Answer Set Programming and other Computer Paradigms (ASPOCP’18) (2018)
• [13] Leone, N., Rullo, P., Scarcello, F.: Disjunctive stable models: Unfounded sets, fixpoint semantics, and computation. Inf. Comput. 135(2), 69–112 (1997)
• [14] Marek, V., Truszczyński, M.: Stable models and an alternative logic programming paradigm, pp. 169–181. Springer-Verlag (1999)
• [15] Marek, V.W., Truszczynski, M.: Relating autoepistemic and default logics. In: KR. pp. 276–288. Morgan Kaufmann (1989)
• [16] Moore, R.C.: Possible-world semantics for autoepistemic logic. In: Proceedings of the Non-Monotonic Reasoning Workshop, Mohonk Mountain House, New Paltz, NY 12561, USA, October 17-19, 1984. pp. 344–354. American Association for Artificial Intelligence (AAAI) (1984)
• [17] Moore, R.C.: Semantical considerations on nonmonotonic logic. Artif. Intell. 25(1), 75–94 (1985), https://doi.org/10.1016/0004-3702(85)90042-6
• [18] Niemelä, I.: Logic programs with stable model semantics as a constraint programming paradigm. AMAI 25, 241–273 (1999)
• [19] Niemelä, I.: Constructive tightly grounded autoepistemic reasoning. In: IJCAI. pp. 399–405. Morgan Kaufmann (1991)
• [20] Pearce, D.: A new logical characterisation of stable models and answer sets. In: NMELP. Lecture Notes in Computer Science, vol. 1216, pp. 57–70. Springer (1996)
• [21] Schwarz, G.: Autoepistemic logic of knowledge. In: Nerode, A., Marek, V.W., Subrahmanian, V.S. (eds.) Logic Programming and Non-monotonic Reasoning, Proceedings of the First International Workshop, Washington, D.C., USA, July 1991. pp. 260–274. The MIT Press (1991)
• [22] Schwarz, G.: Minimal model semantics for nonmonotonic modal logics. In: Logic in Computer Science, 1992. LICS’92., Proceedings of the Seventh Annual IEEE Symposium on. pp. 34–43. IEEE (1992)
• [23] Shen, Y., Eiter, T.: Evaluating epistemic negation in answer set programming (extended abstract). In: Proc. of the Intl. Joint Conference on Artificial Intelligence (IJCAI’17). pp. 5060–5064 (2017)
• [24] Truszczyński, M.: Revisiting epistemic specifications. In: Logic Programming, Knowledge Representation, and Nonmonotonic Reasoning. Lecture Notes in Computer Science, vol. 6565, pp. 315–333. Springer (2011)
• [25] Wang, K., Zhang, Y.: Nested epistemic logic programs. In: LPNMR. Lecture Notes in Computer Science, vol. 3662, pp. 279–290. Springer (2005)

Appendix A Proof of Proposition 5

5 Just note that, for atomic , we have that iff which implies that . The rest of the proof follows by induction in the structure of .

Appendix B Proof of Theorem 5

The proof of Theorem 5 rely on the definition of S5-equilibrium models. We first show that S5-equilibrium models are unfounded-free and then that autoepistemic world views are S5-equilibrium models. Then, we obtain that autoepistemic world views are unfounded-free as a corollary. We start by defining S5-interpretations  as sets of . An S5-interpretations is said to be total iff it satisfies that every   is total, that is, . We say that an S5-interpretations  is a -model of a formula  iff for every .

Given S5-interpretations and , we write iff the following two condition hold:

1. for every , there is some , with .

2. for every , there is some , with .

As usual, we write iff and .

A total S5-interpretation is said to be a S5-equilibrium model of some theory  iff is a -model of and there is no other -model  of such that .

S5-equilibrium model are similar to epistemic equilibrium model in the sense of [2], being the major difference that, in our approach, different evaluations in the here world can be used to minimise the same evaluation in the there world. As a result, we conjeture that every S5-equilibrium model is also an epistemic equilibrium model in the sense of [2]. On the other hand, the converse does not hold in general. For instance, if we take the program 4 from Example 4, we have that is an epistemic equilibrium model in the sense of [2], but not an S5-equilibrium model in our approach. This shows that epistemic equilibrium model in the sense of [2] are not unfounded-free (in the same way that world views or autoepistemic equilibrium model in this approach are not unfounded-free). The following result shows that S5-equilibrium model are, in fact, unfounded-free.

Any S5-equilibrium model of any program  is unfounded-free.

Let be some S5-equilibrium model of some program  and suppose, for the sake of contradiction, that it is not unfounded-free. Then, there is a unfounded-set  for  with respect to  such that, for every , we have and . Let l ?C? C l C l W’ &=&& I,I && I,I ∈W and no X,I ∉S
&&∪& I∖X,I && I,I ∈W and X,I ∈S Since is non-empty and for all , we have that and, since is an S5-equilibrium model of , it must be that is not an -model of . Hence, there is some rule such that is an -model of while is not. Besides, the latter implies that there is some such that and, thus, that one of following conditions must hold:

1. and , or

2. and .

Note that the latter is a contradiction with the fact that is an -model of and, thus, the former must hold. Furthermore, implies and, since is an -model of , that . Hence, and there is an atom such that . By construction, this implies that, since with and, thus, one of the following conditions must hold:

1. ,

2. , or

3. , or

4. .

The first condition cannot hold because we have that . Furthermore, , we have that
holds
iff
iff
iff
iff or
iff
which does not hold. Hence, the forth condition cannot hold either.

Assume now that . Then, and, thus, we have that which is a contradiction with 1. Therefore, it must be that holds. Pick some atom . But then, there is some such that and with . This implies that and, thus, that which is a contradiction with the fact that . Consequently, must be unfounded-free.

Let be a theory and  be a model of . Then, is an -model of .

Let be a theory and be an S5-interpretation such that is a equilibrium model of for every . Then, is an S5-equilibrium model of .

First note that, since is a equilibrium model of for every , it follows that is a model of for every and, thus, that for every formula . In its turn, this implies that is an S5-model of . Suppose now, for the sake of contradiction, that is not an S5-equilibrium model of  and, thus, that there is some -model  of such that . Hence, there is such that . Furthermore, since is a equilibrium model of it follows that is not a model of . Hence, there is a formula  and such that . This implies that is not an -model of which is a contradiction with the fact that is an -model of . Consequently, must be an S5-equilibrium model of .

Note that, in general, the converse of Proposition B does not hold. For instance, is the unique S5-equilibrium model of the the theory while is not a equilibrium model of it.

Let be an autoepistemic world view of some theory . Then, it satisfies the following two conditions:

1. is an S5-equilibrium model, and

2. there is no propositional interpretation such that model is an equilibrium model of and .

Note that, in general, the converse of Corollary B does not hold. For instance, theory has no autoepistemic world view while we have that is an S5-equilibrium model and is not a equilibrium model of it. To see that is not autoepistemic world view, note that is not a equilibrium model of .

5 From Corollary B, we have that is an S5-equilibrium model and, from Theorem B, this implies that is unfounded-free.

Appendix C Proof of Theorem 5

The proof of Theorem 5 rely on the definition of weak autoepistemic world views. We first show that every autoepistemic world view is also a weak autoepistemic world view and then that weak autoepistemic world views coincide with G91-world views. Then, we obtain that autoepistemic world views are G91-world views as a corollary. Let us start by defining semi-total interpretations. We say that an interpretation is semi-total iff every   is total, that is, . It is easy to see that, every total interpretation is semi-total but not vice-versa.

A total interpretation is said to be a weak equilibrium model of some theory  iff is a model of and there is no other semi-total model of such that .

Note that every equilibrium model is also a weak equilibrium model, but not vice-versa. For instance, is a weak equilibrium model of but not a equilibrium model.

A S5-interpretation  is called a weak autoepistemic world view of  iff it satisfies the following two conditions:

1. is a weak equilibrium model of for every , and

2. there not exists any propositional interpretation such that is a weak equilibrium model of and .

Every autoepistemic world view is also a weak autoepistemic world view.

Since every equilibrium model is also a weak equilibrium model, it only remains to be shown that if is an autoepistemic world view then

1. there not exists any propositional interpretation such that is a weak equilibrium model of and .

Suppose, for the sake of contradiction, that there is some propositional interpretation such that is a weak equilibrium model of and . Since is an autoepistemic world view, this implies that is not a equilibrium model of and, thus, that there is some non-semi-total model of such that . Hence, there is some such that . Let . Then, we have that and, since is an autoepistemic world view of and , we have that is a equilibrium model of . These two facts together imply that is not a model of . Hence, there is a formula  such that is not a model of and, thus, there is such that . On the other hand, since is a model of , it follows that for every . Hence, it follows that and and that . However, since , this implies that , which is a contradiction with the fact that is a model of . Consequently, is a weak autoepistemic world view.

Let be a formula and be a semi-total interpretation. Then, is a model of iff is a model of .

Assume that . Then, we have that if and only if for every iff is a S5-model of iff iff . Then, by induction in the structure of , we get that iff . Finally, we have that is a model of iff and for every iff and for every iff is a model of .

Let be a propositional theory and be some interpretation. Then, is a model of iff is a of for every .

By definition, we have that is a model of iff is a model of for all . Furthermore, is a model of iff and for every