Bulletin of the Section of Logic Volume 32/4 (2003), pp.

165177

George Tourlakis Francisco Kibedi

A MODAL EXTENSION OF FIRST ORDER CLASSICAL LOGICPart I

Abstract We formalize a fragment of the metatheory of classical rst order logic by adding a new connective, the modal 2, whose intended interpretation is the (classical) metalogical predicate is a theorem ( ). We then illustrate how to employ this modal extension of classical logic to write equational proofs of classical theorems, and prove (in Part II) its completeness and soundness with respect to Kripke models. We also prove that the extension meets our objective: For any classical formulae A and B , we can prove 2A 2B modally i we can prove A B classically.

Keywords: First order logic, modal logic, equational logic, calculational logic, consistency, completeness, Leibniz rule, derivability conditions, provability predicate, Kripke models.

1. Introduction
First order equational, or calculational, logic was proposed in [4], [6], [7] and was shown to be sound and complete in [13], [14], [15]. As a practitioners tool, it relies heavily on Leibnizs principle of replacing equals by equals allowing the user to prove assertions in the same manner that one veries the equality or inequality of two expressions in high school algebra
1 This

research was partially supported by NSERC grant No. 8820

166

George Tourlakis and Francisco Kibedi

or trigonometry, namely, by constructing a conjunctional2 chain of equalities and inequalitiesin the case of logic, equivalences () and implications () respectively. A classical formal proof in the equational style ought to look like annotation annotation annotation A1 A2 ... An (1) normally written vertically, where each occurrence of stands for one of the formal connectives or . Such a conguration establishes the derivability of An if A1 is known to be a theorem. More generally, it establishes the derivability of A1 An (A1 An if all the are ). In practice, one often needs to break this chain of formal equivalences and implications and invoke a rule of inference that will spawn a new chain, disjoint from the original. For example, A1
annotation

...

annotation

B (x)B

annotation

...

annotation

An

(2)

The above cannot be either of or , but is the classical metalogical (here also the two-sided works). Thus the conguration (2) proves An from A1 , however it does so using two connected components of type (1). It is disconnected at the point where one moves into the metatheory and applies . We believe that from a methodological point of view it is desirable to have a new proof calculus that is suciently powerful to simulate classical equational proofs such as (2) above without having to disconnect the proof by the insertion of the metalogical connectives or , as the latter are often confused with the formal and by inexperienced users. We expect that such a calculus will eliminate this potential confusion by simplifying and further mechanizing the process of writing equational proofs, as the user will now exclusively deal with the formal and . Similar comments were contained in [8] were it was, essentially, suggested that one formalize as the modal 2, extending in some natural way the classical logic (such an extension was restricted to the propositional case in loc. cit.). Our goal in this paper is to present a simple modal framework within which we can write equational proofs of classical theorems3 in a manner
a = b < c = d < e, meaning a = b and b < c and c = d and d < e. arbitrary modal equational proof may also be disconnected. However we are only interested in modal proofs that simulate classical equational proofs. Such simulations will be always connected.
3 The 2 E.g.,

A Modal Extension of First Order Classical LogicPart I

167

that the only connectives used are the formal and . In this framework the two formal chains in (2) that were disconnected at will be rendered by the connected modal chain

2An (3) To this end we devise a particular modal extension of classical rst order logic that simulates generalization in a natural manner and prove that it meets our main conservation requirement: For any classical formulae A and B and classical theory T , we have that T {A} proves B classically i T proves 2A 2B modally. Fulllment of this requirement allows one to replace any, possibly disconnected, classical equational proof by a connected modal equational proof: First, replace all classical and by the classical metalogical and respectively (a valid step by modus ponens). Then replace every A B (respectively, A B ) by 2A 2B (respectively, 2A 2B ). It is immediate by the conservation requirement that, conversely, a connected modal chain of type (3) can be replaced by a classical, possibly disconnected, chain of type (2). Modal extensions of propositional ([2], [8], [9], [10], [12], [17]) and predicate ([1], [3], [5], [9], [10], [11]) classical logic are not new, however, previous ones dier in at least two signicant ways from our approach. One, they do not address, but we do, the needs of the user, investigating instead general or specic philosophical and metamathematical issues. Two, their 2 either denotes (intuitively) a form of necessity, or it denotes provability in the narrow sense of G odels provability predicate for Peano arithmetic (or for ZF). As far as we know, our approach is the rst that uses 2 to simulate classical rst order provability in general. This entails some departures from the norm: Firstly, the semantics of Section 5 (in Part II) leads to models that are similar to but not identical to varying domain models. Secondly, we require 2A to be a sentence for all formulae A (contrast, e.g., with [1], [3], [5]), a choice that we justify later on. There are many variants of modal logic to choose from as a starting point. We build upon a rst order version of logic K4, adding just one axiom schema ((M3) of Section 2) that simulates classical generalization A (x)A. The conservation requirement determines the form of (M3): ... ...

2A1

annotation

annotation

2B 2(x)B

annotation

annotation

168

George Tourlakis and Francisco Kibedi

2A 2(x)A.4 This axiom plays a role in making G odel completeness work (cf. Lemma 6.3 in Part II). Motivation for keeping the two modal axioms of K4 is straightforward: Axiom (M1) in Section 3 simulates classical modus ponens (cf. 4.10). Axiom (M2), of less obvious intuitive value, is technically expedient, for example, towards the proof of weak necessitation and inner Leibniz rule. These two axioms are the counterparts of L obs derivability conditions DC2 and DC3 respectively that one encounters in proofs of G odels second incompleteness theorem ([2], [12], [16]), this observation providing immediate peace of mind with respect to their relative consistency with the classical axioms. By the way, our axiom (M3) too is true when 2 is interpreted as G odels provability predicate.5 This observation, once more, puts to rest any consistency worry. We do not need the reection principlethat is, axiom schema 2A A of logic S4and we do not include it. Indeed, it is a trivial exercise to show that it is underivable in our logic, for one can easily build a countermodel (using an irreexive frame relation) by the techniques of Section 5 of Part II. We also prefer to simulate the inference if A, then 2A (corresponding to L obs DC1) by hiding it inside the axioms, adopting only classical primitive rules of inference. The paper is split into two parts. Part I (Sections 24) deals with syntactic issues. Part II (Sections 56) introduces semantics, proves the soundness and strong completeness of the proposed logic with respect to Kripke structures, and proves that the main conservation requirement is met.

2. The Language of Modal Logic

Terms are built from the object variables x, y, z, x , y , z51 , . . ., and whatever nonlogical symbols may be available in any particular theory of interest, such as constants a, b, c, a , c13 , . . ., and functions f, g, h, f , h101 . . ., exactly as in classical rst order logic.
4 Since 2A is closed for all A, one may view (M3) as a special case of the Barcan formula. 5 True abuses language here, and it means: If P (x) is G odels provability predicate for Peano arithmetic, and if A denotes the (formal) G odel number of the formula A, then one can prove P ( A ) P ( (x)A ) in some appropriate conservative extension of Peano arithmetic ([16]).

A Modal Extension of First Order Classical LogicPart I

169

Formulae (well-formed modal formulaewfmf) are built, via the standard inductive denitions, from atomic formulaewhich may involve predicate symbols P, Q, R, P54 , . . .and the primary logical symbols. The latter are the Boolean variables p, q, p , p , q13 , . . ., and the connectives: , , , , 2, (, ), =, , and the comma. We note two slight deviations from the standard denitions: One is that we add an induction clause if A is formula, then so is (2A). The other is the inclusion of Boolean variables (traditionally used in equational logics to facilitate applications of the Leibniz rule) and the Boolean constants (formal true) and (formal false). Every Boolean symbol is an atomic formula. A is the scope of the leftmost 2 in (2A). Every object variable in the scope of a 2 is bound; thus, (2A) is closed (a sentence). The motivation regarding object variables is our intended intuitive interpretation of 2 as the classical , and therefore as the classical |= as well. When we say |= A classically, we mean that for all structures where we interpret A, and for all value-assignments to the free object variables of A, the formula is true. Thus the variables in a statement such as |= A are implicitly universally quantied and are unavailable for substitutions. However, Boolean variables in a wfmf are always free. If a formula does not contain 2, then we say it is a classical formula or a well-formed formula, or w. Additional symbols , , and are introduced metalinguistically in the usual manner. The usual metalinguistic rules of how to omit brackets apply: highest priority symbols are the unary , , , 2; all associativities are right. We denote substitutions into variables of a wfmf A by the meta expressions A[x := t] (term substituted into all occurrences of the free object variable x) and A[p := B ] (formula substituted into all occurrences of the Boolean variable p). In the the rst type we disallow substitution if proceeding would have some variable of the term t captured by a quantier. In the second type we proceed regardless. In our notation the symbols [x := t] and [p := B ] have the least scope, thus, e.g., (x)A[y := t]if allowedmeans (x)(A[y := t]). If is a set of wfmf, then 2 denotes {2A : A }. We say that 2A is the boxed version of A and 2 is the boxed version of .

170

George Tourlakis and Francisco Kibedi

3. Axioms
We shall call our rst order modal logic M3 , the 3 indicating the presence of three modal axioms. Definition 3.1. The set of axioms of M3 is 2, where consists of all instances of the following schemata. (1) All tautologies (2) (x)A A[x := t], provided the substitution is allowed (3) A (x)A, provided x is not free in A (4) (x)(A B ) (x)A (x)B (5) x = x (6) s = t (A[x := s] A[x := t]), provided the substitutions are allowed (M1) 2(A B ) 2A 2B (M2) 2A 22A (M3) 2A 2(x)A There are two primary rules of inference. Modus ponens (MP) if A and A B , then infer B , and generalization (Gen) if A, then infer (x)A, for any object variable x. We shall always work within a mathematical theory, generically denoting its set of nonlogical axioms by T . Examples of T are ZFC, Peano arithmetic, or something totally wild (including wfmfs), or . In the latter case we have a pure theory, i.e., we are doing just logic. Definition 3.2. [-proofs and -theorems] We say that a formula A is a -theorem of T based on a (possibly empty) set of additional assumptions, and write T Ai there is a -proof of An from T . By such a proof we understand a sequence of formulae A1 , . . . , An such that A is An and each Ai satises one of the following conditions: (1) Ai 2 T 2T (2) There are numbers j, k < i such that Ak is Aj Ai (3) There is a number j < i such that Ai is (x)Aj . The corresponding direct recursive denition of -theorems is to say that A is a -theorem i it satises (1) (using A for Ai ) or there is a -theorem B , such that B A is also a -theorem, or A is (x)B and B is a -theorem. We omit writing (or T ) if it is empty.

A Modal Extension of First Order Classical LogicPart I

171

Having added the boxed versions of all the axioms in T and , it is unnecessary to include if A, then 2A as a primary rule, since we can obtain a form of this inference as a derived rule (4.2 below). We also note that there is a subtle but important dierence between writing A and A. A, namely, A is the same as 2

4. Some metatheorems
Metatheorem 4.1. [Tautology Theorem] If A1 , . . . , An |=taut B ,6 then A1 , . . . , An T B for any T . Metatheorem 4.2. [Derived Rule: Weak Necessitation (WN)] If T A, then T 2A, provided = 2 or = 2 for some . Proof. Induction on -theorems. (1) If A T , then 2A 2( T ), and we are done. If A is 2B for some B T , then we have T 2B , but also T 2B 22B , by (M2). Thus T 22B by MP. (2) If A , then we proceed as in (1). (3) Let T A, and also T B and T B A. We have T 2B and T 2(B A) by induction hypothesis (I.H.). Then we have T 2B 2A by (M1) and MP. Using MP again, we get T 2A. (4) Let T C , and A be (x)C . By I.H., T 2C , hence T 2(x)C by (M3) and MP. Corollary 4.3. If
T

A, then

2 A.

We call an inference rule weak if in order to obtain its conclusion we must place restrictions on how the premises were derived. Otherwise the rule is strong. For example, Gen is strong for we place no conditions on the hypothesis A. Metatheorem 4.4. [Outer Deduction Theorem] For any formulae A, B and any set of formulae , if + A T B with a condition, then T A B . The condition is that a + A-proof of B exists that contains no generalization steps C (x)C , for any x that is free in A.
6A , . . . , A | n =taut B indicates that A1 , . . . , An tautologically imply B . That is the 1 same as saying that |=taut A1 . . . An B , i.e., that A1 . . . An B is a tautology.

172

George Tourlakis and Francisco Kibedi

N.B. + A is an abbreviation of {A}. Proof. By induction on + A-theorems B obtained via + A-proofs that satisfy the condition. (1) If B is in one of , 2, T , or 2T , then T B . Now, B |=taut A B . So we get T A B by 4.1, from which T A B . (2) Suppose B is in . Then T B . Since B |=taut A B , as above, we have T A B . (3) Suppose B is A. Then A B is the tautology A A. Hence T A B , and so T A B . (4) Suppose + A T C and + A T C B . By I.H., T A C and T A (C B ). Since A C, A (C B ) |=taut A B , we have T A B . (5) Finally, let + A D and (x)D be B . By I.H., A D, hence (x)(A D) by Gen. Axiom (4) now yields (x)A (x)D (i)

via MP. The fact that D (x)D was employed in the proof of B means that x is not free in A. Thus, by axiom (3) and 4.1, (i) yields A (x)D. Metatheorem 4.5. [Inner Generalization] 2A 2(x)A.

Proof. The direction is by 2((x)A A) (boxed axiom (2)), by (M1) and MP. The direction is (M3). Remark 4.6. The qualiers outer and inner are used with respect to the classical logic that M3 extends. Thus, an inner rule is not a rule of M3 at all. Rather it is a theorem schema of M3 that simulates a rule of classical logic according to the (soon to be proved) main conservation requirement. Here inner generalization simulates classical generalization on classical w A: A and (x)A are mutually derivable. Note however that 4.5 applies to all wfmf A not only to classical w A. The qualier outer applied to a derived rule indicates that, unlike inner, this is a bona de derived rule of M3 . Metatheorem 4.7.
T

(x)(A B ) (x)A (x)B .

A Modal Extension of First Order Classical LogicPart I

173
T

Metatheorem 4.8. [2-monotonicity] If 2B . Metatheorem 4.9. [2 over ] Proof.

2(A B ), then

2A

2(A B ) (2A 2B ).

2(A B ) 4.8 plus |=taut (A B ) A B 2(A B ) (M1) 2A 2B 2(A B ) (2B 2A) and are done by 4.1.

We similarly prove

Remark 4.10. 2(A B ) (2A 2B ) is the counterpart of the equanimity rule of [6], [13], [14], [15], namely A B, A B . This follows from Boolean manipulation, the modal theorem 2(A B ) 2A 2B (exercise) and the main conservation requirement. Similarly, (M1) is inner MP, capturing the classical A B, A B . Metatheorem 4.11. T (x)A (x)B . [Outer -monotonicity] If
T

A B , then

Metatheorem 4.12. [Inner -monotonicity] 2(A B ) 2 (x)A (x)B

Proof.

2(A B ) (M3) 2 (x)(A B ) boxed axiom (4), and 2-monotonicity (4.8 2 (x)A (x)B

Metatheorem 4.13.[Inner Leibniz Rule]

T

2(A B ) 2 C [p := A] C [p := B ]

174

George Tourlakis and Francisco Kibedi

Proof. We prove the claim by induction on the formula C . Basis: If C is one of q (other than p), p, , , then the result follows trivially. If C is P (t1 , . . . , tn ) for some n-ary predicate symbol P (possibly the logical =) and some terms t1 , . . . , tn , then again the result follows trivially. For example, in the latter case we are asked to verify 2(A B ) 2 P (t1 , . . . , tn ) P (t1 , . . . , tn ) which follows from 4.1 and axiom 2 P (t1 , . . . , tn ) P (t1 , . . . , tn ) . Induction steps: (1) If C is D or D G, then the result follows by tautological implication via the obvious I.H. For example, 2(D[p := A] D[p := B ]) 2(D[p := A] D[p := B ]) by 2-monotonicity. Hence 2(A B ) 2(D[p := A] D[p := B ]) by I.H. and tautological implication. (2) If C is (x)D, then we calculate as follows: 2(A B ) I.H. 2 D[p := A] D[p := B ] (M3) 2 (x) D[p := A] D[p := B ] 4.7 + 2-monotonicity (4.8) 2 (x)D[p := A] (x)D[p := B ] We are done since (x) D[p := A] is the same as (x)D [p := A]. (3) If C is 2D, then we calculate as follows: 2(A B ) I.H. 2 D[p := A] D[p := B ] (M2) 22 D[p := A] D[p := B ] 4.9 + 2-monotonicity (4.8) 2 2D[p := A] 2D[p := B ] We are done since 2 D[p := A] is the same as (2D)[p := A]. Corollary 4.14. 2(A B ) 2C [p := A] 2C [p := B ] .

Metatheorem 4.15. [Inner -Introduction] If A has no free x, then 2(A B ) 2(A (x)B ).

A Modal Extension of First Order Classical LogicPart I

175

Proof.

2(A B ) inner -monotonicity (4.12) 2 (x)A (x)B Leibniz (4.14): axioms (2, 3) yield 2((x)A A) 2 A (x)B 2(A B ) 2((x)A B )

Corollary 4.16. [Inner -Introduction] If B has no free x, then

Remark 4.17. Each of the implications in 4.15 and 4.16 is promoted to an equivalence by tautological implication and the fact that the other direction holds. For example, 2(A B ) 2(A (x)B ) by 2-monotonicity (4.8) and the tautological consequence (A (x)B ) (A B ) of an instance of axiom (2). Due to lack of space we present here only two examples of use of M3 as a tool for writing equational proofs of classical theorems. Example 4.18. [-swap] To prove the classical (x)(y )A and (y )(x)A are mutually derivable we prove instead 2(x)(y )A 2(y )(x)A. 2(x)(y )A gen (4.5) 2(y )A gen 2A gen 2(x)A gen 2(y )(x)A The classical equational proof of the above is totally disconnected since cannot be replaced by in the -introduction direction. Example 4.19. What if we want the classical (x)(y )A (y )(x)A instead? We can do this by proving and directions separately, followed by tautological implication. E.g., for the direction we verify

176

George Tourlakis and Francisco Kibedi

2 (x)(y )A (y )(x)A : 2 (y )A A inner -mon. 2 (x)(y )A (x)A inner -intro. 2 (x)(y )A (y )(x)A

References
[1] Sergei Artemov and Franco Montagna, On First-Order Theories with Provability Operator, J. of Symb. Logic 59/4 (1994), pp. 11391153. [2] George Boolos, The unprovability of Consistency; An essay in modal logic, Cambridge University Press, Cambridge, 1979. [3] George Boolos and Vann McGee, The degree of the set of sentences of predicate provability logic that are true under every interpretation, J. of Symb. Logic 52/1 (1987), pp. 165171. [4] Edsger W. Dijkstra and Carel S. Scholten, Predicate Calculus and Program Semantics, Springer-Verlag, New York, 1990. [5] Melvin Fitting and Richard L. Mendelsohn, First-Order Modal Logic, Kluwer Academic Publishers, Dordrecht, 1998. [6] David Gries and Fred B. Schneider, A Logical Approach to Discrete Math, Springer-Verlag, New York, 1994. [7] David Gries and Fred B. Schneider, Equational propositional logic, Information Processing Letters 53 (1995), pp. 145152. [8] David Gries and Fred B. Schneider, Adding the Everywhere Operator to Propositional Logic, J. Logic Computat. 8/1 (1998), pp. 119129. [9] G. E. Hughes and M. J. Cresswell, An Introduction to Modal Logic, Methuen and Co. Ltd., 1968, London. [10] G. E. Hughes and M. J. Cresswell, A New Introduction to Modal Logic, Routledge, 1996, London. [11] Franco Montagna, The predicate modal logic of provability, Notre Dame J. of Formal Logic 25 (1984), pp. 179189. [12] C. Smory nski, Self-Reference and Modal Logic, SpringerVerlag, New York, 1985. [13] G. Tourlakis, A Basic Formal Equational Predicate Logic-Part I, Bulletin of the Section of Logic 29:1/2 (2000), pp. 43-56.

A Modal Extension of First Order Classical LogicPart I

177

[14] G. Tourlakis, A Basic Formal Equational Predicate Logic-Part II, Bulletin of the Section of Logic 29:3 (2000), pp 7588. [15] G. Tourlakis, On the soundness and Completeness of Equational Predicate Logics, J. Logic Computat. 11/4 (2001), pp. 623653. [16] G. Tourlakis, Lectures in Logic and Set Theory, Volume 1, Mathematical Logic, (2003), Cambridge University Press, Cambridge. [17] Johan van Benthem, Modal Logic and Classical Logic, Bibliopolis, Napoli, 1983.

Department of Computer Science York University Toronto, Ontario M3J 1P3 Canada gt@cs.yorku.ca Department of Mathematics and Statistics York University Toronto, Ontario M3J 1P3 Canada francisco kibedi@hotmail.com