Bulletin of the Section of Logic Volume 18/2 (1989), pp. 63–71 reedition 2005 [original edition, pp. 63–71] Nobu-Yuki Suzuki INTERMEDIATE LOGICS CHARACTERIZED BY A CLASS OF ALGEBRAIC FRAMES WITH INFINITE INDIVIDUAL DOMAIN From semantical point of view, an essential nature of predicate logics depends on the fact that individual domains are infinite – not potentially but actually infinite. Ono [2] proved that a class C of algebraic frames characterizes an intermediate predicate logic (an IPL) if and only if there exists at least one algebraic frame (P, U ) in C whose individual domain U is infinite. Hence, a class of algebraic frames each of which has an infinite individual domain always characterizes an IPL. Our starting point is the following question: What kind of an IPL does such a class characterize? We say such an IPL is ω + -complete. In Section 1, we will define a syntactical property which we call the pseudo-relevance property (PRP, for short), which can be regarded as a weak version of Craig’s interpolation property. We will show that if L is an algebraically complete IPL with the above PRP, then L is ω + -complete. Next, we will show that the ω + -completeness of an IPL L implies PRP of L, provided that K(≡ ¬¬∀x(p(x) ∨ ¬p(x))) is provable in L. By making use of this result, we will show that completeness does not always imply ω + -completeness in Section 2. In the following, we assume the readers’ familiarity with Ono [2]. 64 Nobu-Yuki Suzuki 1. We will repeat some terminology and notation. We refer readers to Ono [2] for details. As usual, we fix a first-order language L, which contains neither constants nor function symbols (see Ono [2]). For each non-empty set U , we denote by L[U ] the language obtained from L by adding the name u of each u ∈ U as new constants. We sometimes identity L[U ] with the set of all sentences of L[U ]. A pair (P, U ) of a non-degenerate pseudo-Boolean algebra P and a non-empty set U is said to be an algebraic frame if P is λ-complete, where λ is the cardinality of U . An assignment f of an algebraic frame (P, U ) is a mapping of the set of all atomic sentences in L[U ] to P = <P, ∪, ∩, →, 0, 1 >. We extend it to a mapping of L[U ] to P as follows: (A1) f (A ∨ B) (A2) f (A ∧ B) (A3) f (A ⊃ B) (A4) f (¬A) (A5) f (∀xA(x)) (A6) f (∃xA(x)) = = = = = = f (A) ∪ f (B), f (A) ∩ f (B), f (A) → f (B), fT(A) → 0, Su∈U f (A(u)), u∈U f (A(u)). A formula A (of L) is said to be valid in an algebraic frame (P, U ) if f (A) = 1 holds for every assignment f of (P, U ), where A is the universal closure of A. The set of all formulas of L that are valid in an algebraic frame (P, U ) is denoted by L+ (P, U ). As is well-known, L+ (P, U ) contains all formulas provable in the intuitionistic predicate logic H, and is closed under modus ponens, the rule of generalization and the rule of substitution. An intermediate predicate logic (an IPL) L is said to be (algebraically) complete if there exists a class C of algebraic frames such that T L = (P,U )∈C L+ (P, U ). Definition 1.1. An algebraic frame (P, U ) is said to be an ω + -frame if + U is infinite. Let L be an IPL. L is said T to be ω +-complete if there exists + a class C of ω -frames such that L = (P,U )∈C L (P, U ). It is obvious that ω + -completeness implies completeness. In Section 2, we will prove that ω + -completeness does not follow from completeness. Next, we will introduce a syntactical property which plays an important rôle in this paper. Intermediate Logics Characterized by a Class of Algebraic Frames with... 65 Definition 1.2. Let L be an IPL. L is said to have the pseudo-relevance property (PRP) if L satisfies the following condition (*): (*) For every formula A and B (in L) which contain no predicate variable in common, if A ⊃ B is provable in L then either ¬A or B is provable in L. We remark here that PRP can be regarded as a weak version of Craig’s interpolation property (see e.g., Ono [3]). We can define PRP analogously for intermediate propositional logics. However, it is well-known that every intermediate propositional logic has PRP (see Komori [1]). On the other hand, PRP is not trivial for IPL’s. Theorem 1.3. For every intermediate propositional logic J other than the classical one, there exist at least countably infinite predicate extensions of J each of which does not have PRP. Proof. Let an intermediate propositional logic J other than the classical one be fixed. Define formulas Yn (n ≥ 1), Zn (n ≥ 1), and N Rn (n ≥ 1) by Ym ≡ p1 (x); m V ≡ ( ¬pi (x)) ∧ pm (x) (m ≥ 2), Zn ≡ Y1 i=1 n V ∃xYi , and i=1 N Rn ≡ Zn ⊃ (q ∨ ¬q), where each pi (i = 1, 2, . . .) is a monadic predicate variable and q is a propositional variable. Since each N Rn (n ≥ 2) is provable in the classical predicate logic C, we see that J∗ +N Rn (n ≥ 2) is an IPL, where J∗ +N Rn is the logic obtained from the minimum predicate extension J∗ of J by adding N Rn as a new axiom. Since the associated propositional formula α(Zn ) of each Zn (n ≥ 2) is logically equivalent to “absurdum” in H, each α(N Rn ) is provable in J. Hence, for each n (n ≥ 2), J∗ + N Rn is a predicate extension of J. Since J∗ + N Rn is strictly weaker than C, neither ¬Zn nor q ∨ ¬q is provable in J∗ + N Rn , and hence J∗ + N Rn fails to have PRP. It remains to prove that J∗ + N Rm 6= J∗ + N Rn if m 6= n. Let S2 be the three-valued linear pseudo-Boolean algebra. Then, if n > m ≥ 2, an algebraic frame (S2 , {1, 2, . . . , m}) validates N Rn but not N Rm . Hence, J∗ + N Rn ⊂ L+ (S2 , {1, 2, . . . , m}) 6∈ N Rm . Therefore, J∗ + N Rm 6= J∗ + N Rn . q.e.d. 66 Nobu-Yuki Suzuki Since there are uncountably many intermediate propositional logics, Theorem 1.3 provides us with uncountably many ILP’s without PRP. For complete IPL’s, PRP is a sufficient condition to be ω + -complete. That is, Theorem 1.4. complete. Let L be a complete IPL. If L has PRP, then L is ω + - Proof. Assume that L is complete with respect to a class C of algebraic frames. Let D be the T class of ω + -frames in C. We will show that if L has PRP, then L = (P,U )∈D L+ (P, U ). Recall that if L = T L+ (P, U ), then L is ω + -complete. Suppose that L is a proper (P,U )∈D T subset of (P,U )∈D L+ (P, U ). Then there exists a sentence A such that (1) A is not provable in L. (2) A is valid in every ω + -frame in C. Define sentences F1 , F2 and F in by F1 F2 F in ≡ ∀xr(x, x) ∧ ∀x∀y(r(x, y) ∨ r(y, x)), ≡ ∀x∀y∀z(r(x, y) ∧ r(y, z) ⊃ r(x, z)), and ≡ F1 ∧ F2 ⊃ ¬∀x∃y¬r(y, x), where r is binary predicate variable. Ono [2, Theorem 2.6] proved that F in is valid in every algebraic frames whose individual domain is finite. Hence, A ∨ F in is valid in every algebraic frame in C. It follows from this that A ∨ F in is provable in L. Since A ∨ F in ⊃ (¬F in ⊃ A) is provable in H, ¬F in ⊃ A is provable in L. Without loss of generality, we can assume that ¬F in and A contain no predicate variable in common. Recall that ¬¬F in is not provable in C. So, ¬¬F in ≡ ¬(¬F in) is not provable in L. From this and (1), it follows that L does not have PRP. q.e.d. The converse of Theorem 1.4 does not always hold (see Section 2). However, the converse holds under some additional conditions. Let K be the formula ¬¬∀x(p(x) ∨ ¬p(x)), where p is a monadic predicate variable. Note that K is provable in C. Theorem 1.5. Let L be an IPL in which K is provable. If L is ω + complete, then L has PRP. In the rest of this Section, we will prove Theorem 1.5. Our proof proceeds similarly to the case of intermediate propositional logics in Komori [1, Theorem 2.1]. He made use of Glivenko’s theorem. The condition that, “K Intermediate Logics Characterized by a Class of Algebraic Frames with... 67 is provable” has a closed relationship to Glivenko’s theorem in IPL’s as follows. Fact 1.6. (Umezawa [5]) Let L be an IPL. The following two conditions on L are equivalent: (i) K is provable in L, (ii) For every formula A, ¬A is provable in C if and only if ¬A is provable in L. Lemma 1.7. Let a formula A ≡ A(x1 , . . . , xn ) contains no free variables other than x1 , . . . , xn . If ¬A is not provable in C, then, for every ω + -frame (P, U ), there exist an assignment f of (P, U ) and elements u1 , . . . , un ∈ U such that f (A(u1 , . . . , un )) = 1P where 1P is the greatest element of P . Proof. Since U is infinite, C is complete with respect to an ω + -frame (2, U ), where 2 is the two-valued Boolean algebra, i.e., 2 = {0, 1}. Hence, there exist an assignment g of (2, U ) and elements u1 , . . . , un ∈ U such that (3) g(A(u1 , . . . , un )) = 1 in 2. Define a mapping f of the set of all atomic sentences in L[U ] to P by 1P if g(p(v1 , . . . , vm )) = 1, f (p(v1 , . . . , vm )) = 0P otherwise, for each m−ary predicate variable p and each vi ∈ U (i = 1, . . . , m), where 0P is the least element of P . Extending f as an assignment of (P, U ), we have that for every sentence C of L[U ], 1P if g(C) = 1, f (C) = 0P otherwise. It follows from (3) that f (A(u1 , . . . , un )) = 1P . Proof of Theorem 1.5. Suppose that ¬A and B contain no predicate variable in common, and none of them is provable in L. Without loss of generality, we can assume that A and B contain no free variables other than x1 , . . . , xn . We write A(x1 , . . . , xn ) and B(x1 , . . . , xn ) for A and B respectively. Then, there exist an ω + -frame (P, U ), an assignment f1 of (P, U ) and elements u1 , . . . , un ∈ U such that (4) L ⊂ L+ (P, U ), and 68 Nobu-Yuki Suzuki (5) f1 (B(u1 , . . . , un )) 6= 1P in P . By Fact 1.6, ¬A(x1 , . . . , xn ) is not provable in C. Hence, by Lemma 1.7, there exist an assignment f2 of (P, U ) and elements v1 , . . . , vn ∈ U such that (6) f2 (A(v1 , . . . , vn )) = 1P in P . Define mappings σ1 and σ2 of the set of atomic sentences in L[U × U ] to L[U ] by σ1 (p((a1 , b1 ), . . . , (am , bm ))) = p(a1 , . . . , am ), and σ2 (p((a1 , b1 ), . . . , (am , bm ))) = p(b1 , . . . , bm ), for every m-ary predicate variable p and every (ai , bi ) ∈ U × U (i = 1, . . . , m). By induction, σ1 and σ2 can be extended to mappings of L[U ×U ] to L[U ] such that for every C ∈ L[U × U ], (7) σ1 (C((a1 , b1 ), . . . , (am , bm ))) = C(a1 , . . . , am ), and (8) σ2 (C((a1 , b1 ), . . . , (am , bm ))) = C(b1 , . . . , bm ). Next, f is defined to be a mapping which sends each atomic sentence p((a1 , b1 ), . . . , (am , bm )) of L[U × U ] for m-ary predicate variable p and for (ai , bi ) ∈ U × U (i = 1, . . . , m) to the following element of P : f1 (p(a1 , . . . , am )) if p occurs in B, f (p((a1 , b1 ), . . . , am , bm ))) = f2 (p(b1 , . . . , bm )) otherwise. We extend f to an assignment of (P, U ) by induction. Then, we can prove that for every C ∈ L[U × U ], (9) if every predicate variable in C appears in B, f (C) = f1 (σ1 (C)), (10) if every predicate variable in C appears in A, f (C) = f2 (σ2 (C)). Thus, by (7), (9) and (5), we have that f (B((u1 , v1 ), . . . , (un , vn ))) = f1 (B(u1 , . . . , un )) 6= 1P . And, by (8), (10) and (6), f (A((u1 , v1 ), . . . , (un , vn ))) = f2 (A(v1 , . . . , vn )) = 1P . Therefore, f ((A ⊃ B)((u1 , v1 ), . . . , (un , vn ))) 6= 1P . It follows from this that A ⊃ B is not valid in (P, U × U ). Since U is infinite, U × U can be identified with U as an individual domain. Thus, we have A ⊃ B 6∈ L+ (P, U ). Hence, A ⊃ B is not provable in L by (4). q.e.d. Intermediate Logics Characterized by a Class of Algebraic Frames with... 69 Theorem 1.5 provides us with a powerful device for proving non-ω + completeness of a given IPL. It suffices to show that K is provable in it and it does not have PRP. Is it possible to eliminate the condition that K is provable in L? The answer is negative (see Section 2). On the other hand, there exists an ω + -complete IPL with PRP in which K is not provable, namely the intuitionistic predicate logic H. 2. In this Section, we will compare completeness, ω + -completeness and PRP. Theorem 2.8. There exists a complete IPL which is not ω + -complete. That is, ω + -completeness does not follow from completeness. Proof. Let L1 = L+ (2, ω) ∩ L+ (S2 , {0}), where ω is the first infinite ordinal, which we will identify with the set {i; i < ω}. Clearly, L1 is a complete IPL. It is a routine to check that K is valid both in (2, ω) and in (S2 , {0}). Hence, K is provable in L1 . If L1 is ω + -complete, L1 must have PRP by Theorem 1.5. So, it suffices to show that L1 does not have PRP. Let p and q be a monadic predicate variable and a propositional variable respectively. In (2, ω), q ∨ ¬q is valid, hence so is (∃xp(x) ∧ ∃x¬p(x)) ⊃ (q ∨ ¬q). In (S2 , {0}), ∃xp(x) ∧ ∃x¬p(x) is always assigned to the least element of S2 . Hence (∃xp(x) ∧ ∃x¬p(x)) ⊃ (q ∨ ¬q) is valid in (S2 , {0}). It follows from this that (∃xp(x) ∧ ∃x¬p(x)) ⊃ (q ∨ ¬q) is provable in L1 . On the other hand, neither ¬(∃xp(x) ∧ ∃x¬p(x)) nor (q ∨ ¬q) is provable in L1 . q.e.d. Theorem 2.9. There exists an ω + -complete IPL without PRP. That is, PRP does not follow from ω + -completeness. Proof. Define a set Q and an order on Q by Q = {(x, y); x = 1, 2 and y = 1, 2, . . . , ω} ∪ {(0, 0)} and (x1 , y1 ) (x2 , y2 ) if and only if x2 ≤ x1 and y2 ≤ y1 , where ≤ is the natural order on ω∪{ω}. Then, Q together with coincides with the complete pseudo-Boolean algebra illustrated in Figure 1. We write Q for this algebra. 70 Nobu-Yuki Suzuki ◦ (0,0) ◦ (1,1) @ @◦ (2,1) (1,2) ◦ @ @◦ (3,1) @ @◦ (2,2) @ @ @ @ (3,2) ◦ · @ @ · · (ω, 1) · ◦ · · (ω, 2) ◦ Figure 1 Clearly, L2 = L+ (Q, ω) is an ω + -complete IPL. We will show that L2 does not have PRP. Let Lin be the formula (q ⊃ r) ∨ (r ⊃ q), where q and r are propositional variables. it can be easily verified that Lin is not provable in L2 . Next, define an assignment f of (Q, ω) by f (p(i)) = (i + 1, 1) for each i ∈ ω. Then we have that f (¬¬K) = (ω, 1). hence, ¬¬K is not provable in L2 . It remains to show that ¬K ⊃ Lin is valid in (Q, ω). By an easy calculation, we have that (1, 1) f (Lin) for every assignment f of (Q, ω). Verify that (ω, 1) f (p(i) ∨ ¬p(i)) holds for every assignment f of (Q, ω). Hence (ω, 1) f (∀x(p(x) ∨ ¬p(x))). Thus f (¬K) = f (¬∀x(p(x) ∨ ¬p(x))) (ω, 1) → (ω, 2) = (1, 2), where → is the relative pseudo-complementation of Q. Therefore f (¬K ⊃ Lin) = (0, 0) for every assignment f of (Q, ω). Thus we have that ¬K ⊃ Lin is valid in (Q, ω). q.e.d. Recall that K 6∈ L+ (Q, ω). It follows that in Theorem 1.5, we cannot eliminate the condition “K is provable in L”. We illustrate the situation simply in Figure 2. Note that each implication (→) was proved to be proper. Intermediate Logics Characterized by a Class of Algebraic Frames with... 71 ω + − complete + “K is provable” ↓ PRP + complete ↓ + ω − complete ↓ complete Figure 2 Acknowledgement. The author would like to express his thanks to Professor Hiroakira Ono and Professor Mitio Takano for their comments on the earlier versions of this article. He also wishes to thank Professor Masazumi Hanazawa and Mr. Minoru Narui for discussions. References [1] Y. Komori, Logics without Craig’s interpolation property, Proceedings of the Japan Academy 54 (1978), pp. 46–48. [2] H. Ono, A study of intermediate predicate logics, Publications of Research Institute for Mathematical Sciences, Kyoto University 8 (1972), pp. 619–649. [3] H. Ono, Some problems in intermediate predicate logics, Reports on Mathematical Logic 21 (1987), pp. 55–67. [4] H. Rasiowa, An algebraic approach to non-classical logics, Studies in Logic and the Foundation of Mathematics 78, North-Holland Publishing Company, Amsterdam-London, 1974. [5] T. Umezawa, On some properties of intermediate logics, Proceedings of the Japan Academy 35 (1959), pp. 575–577. Mathematical Institute Tôhoku University Sendai 980, Japan