Hostname: page-component-8448b6f56d-42gr6 Total loading time: 0 Render date: 2024-04-18T04:31:19.557Z Has data issue: false hasContentIssue false

AN ESCAPE FROM VARDANYAN’S THEOREM

Published online by Cambridge University Press:  13 May 2022

ANA DE ALMEIDA BORGES*
Affiliation:
PHILOSOPHY DEPARTMENT UNIVERSITY OF BARCELONA 08001 BARCELONA, SPAIN URL: aborges.eu E-mail: jjoosten@ub.edu URL: joostjjoosten.nl
JOOST J. JOOSTEN
Affiliation:
PHILOSOPHY DEPARTMENT UNIVERSITY OF BARCELONA 08001 BARCELONA, SPAIN URL: aborges.eu E-mail: jjoosten@ub.edu URL: joostjjoosten.nl
Rights & Permissions [Opens in a new window]

Abstract

Vardanyan’s Theorems [36, 37] state that $\mathsf {QPL}(\mathsf {PA})$—the quantified provability logic of Peano Arithmetic—is $\Pi ^0_2$ complete, and in particular that this already holds when the language is restricted to a single unary predicate. Moreover, Visser and de Jonge [38] generalized this result to conclude that it is impossible to computably axiomatize the quantified provability logic of a wide class of theories. However, the proof of this fact cannot be performed in a strictly positive signature. The system $\mathsf {QRC_1}$ was previously introduced by the authors [1] as a candidate first-order provability logic. Here we generalize the previously available Kripke soundness and completeness proofs, obtaining constant domain completeness. Then we show that $\mathsf {QRC_1}$ is indeed complete with respect to arithmetical semantics. This is achieved via a Solovay-type construction applied to constant domain Kripke models. As corollaries, we see that $\mathsf {QRC_1}$ is the strictly positive fragment of $\mathsf {QGL}$ and a fragment of $\mathsf {QPL}(\mathsf {PA})$.

Type
Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited.
Copyright
© The Author(s), 2022. Published by Cambridge University Press on behalf of The Association for Symbolic Logic

1 Introduction

Provability is a fundamental concept in mathematics, logic, and philosophy alike. Gödel proved his famous incompleteness results in [Reference Gödel22] by formalizing provability. Thus, for formal theories like Peano Arithmetic ( $\mathsf {PA}$ ), we have a natural arithmetical predicate $\Box _{\mathsf {PA}}(\cdot )$ that is true exactly for the Gödel numbers of $\mathsf {PA}$ -provable sentences:

(1) $$ \begin{align} \mathsf{PA} \vdash A \iff \mathbb N \vDash \Box_{\mathsf{PA}} (\ulcorner A \urcorner). \end{align} $$

For readability we shall not distinguish formulas from their Gödel numbers or from syntactic terms (numerals) denoting these Gödel numbers in the future.

Gödel observed various provable structural properties of the provability predicate. For example, for any formula A, if $\mathsf {PA}\vdash A$ then $\mathsf {PA}\vdash \Box _{\mathsf {PA}} A$ . Moreover, this can be formalized itself: for any formula A, $\mathsf {PA} \vdash \Box _{\mathsf {PA}} A \to \Box _{\mathsf {PA}} \Box _{\mathsf {PA}} A$ . Using such properties, and after showing how self-reference can be obtained in $\mathsf {PA}$ , Gödel derived his first incompleteness theorem (here presented in a slightly weakened form for $\mathsf {PA}$ ) by observing that the sentence B such that $\mathsf {PA} \vdash B \leftrightarrow \neg \Box _{\mathsf {PA}} B$ can neither be proved nor refuted in $\mathsf {PA}$ , provided that $\mathsf {PA}$ only proves true theorems.

In light of this, it makes sense to design a system that collects all provable structural properties of the provability predicate. The language $\mathcal L_\Box $ of propositional modal logic is optimally suited for this purpose. The formulas of this language are as the ones of propositional logic together with a unary modality $\Box $ that is syntactically treated as negation. Thus we can write $\mathsf {Form}_\Box ::= \bot \mid \mathsf {Prop} \mid \mathsf {Form}_\Box \to \mathsf {Form}_\Box \mid \Box \mathsf { Form}_\Box $ , where $\mathsf {Prop}$ represents propositional symbols.

The modal logical formulas are linked to arithmetic using so-called realizations. A particular realization $\cdot ^{*}$ maps propositional variables to sentences in the language of $\mathsf {PA}$ and this map is extended to any formula by stipulating that it commutes with implication, $\bot $ is mapped to $(0=1)$ , and the modal operator is mapped to formalized provability, i.e., $(\Box A)^{*_{\mathsf {PA}}} = \Box _{\mathsf {PA}} A^{*_{\mathsf {PA}}}$ .

We can now express what it means to be a provable structural property. For example, for any realization $\cdot ^{*}$ we have $\mathsf {PA} \vdash (\Box (p \to q) \to (\Box p \to \Box q))^{*_{\mathsf {PA}}}$ , since $\mathsf {PA} \vdash \Box _{\mathsf {PA}} (p^{*} \to q^{*}) \to (\Box _{\mathsf {PA}} p^{*} \to \Box _{\mathsf {PA}} q^{*} )$ for any formulas $p^{*}$ and $q^{*}$ . The structural properties are thus captured by the modal formulas that are provable under any realization. We call this $\mathsf {PL}(\mathsf {PA})$ , the provability logic of $\mathsf {PA}$ , and write

$$ \begin{align*} \mathsf{PL}(\mathsf{PA}) := \{ A \in \mathcal L_\Box \mid \text{for any }\cdot^{*}, \text{ we have } \mathsf{PA} \vdash A^{*_{\mathsf{PA}}} \}. \end{align*} $$

Via (1) we know that ( $\mathsf {PA} \vdash A^{*_{\mathsf {PA}}}$ for any $\cdot ^{*}$ ) if and only if ( $\mathbb {N} \vDash \Box _{\mathsf {PA}} A^{*_{\mathsf {PA}}}$ for any $\cdot ^{*}$ ). Clearly $A^{*_{\mathsf {PA}}}$ only depends on the value of $\cdot ^{*}$ for the finitely many propositional variables that occur in A, and so the universal quantifier “for any $\cdot ^{*}$ ” can be coded and made internal, making it possible to characterize $\mathsf {PL}(\mathsf {PA})$ by $\{ A \mid \mathbb N \vDash \forall \cdot ^{*} \, \Box _{\mathsf {PA}} A^{*_{\mathsf {PA}}}\}$ . We now observe that $\Box _{\mathsf {PA}} B$ is a $\Sigma ^0_1$ formula. That is, $\Box _{\mathsf {PA}} B$ is of the form $\exists p \, {\mathrm {Prf}}_{\mathsf {PA}}(p, B)$ , where ${\mathrm {Prf}}_{\mathsf {PA}}(\cdot , \cdot )$ is a decidable predicate. Thus, $\mathsf {PL}(\mathsf {PA})= \{ A \mid \mathbb N \vDash \forall \cdot ^{*} \, \Box _{\mathsf {PA}} A^{*_{\mathsf {PA}}}\}$ has a $\Pi ^0_2$ definition. If we moreover realize that $\Box _{\mathsf {PA}} (\cdot )$ is $\Sigma ^0_1$ complete in that any computably enumerableFootnote 1 set U can be defined using a formula $A_U$ as $\{ n \mid \mathbb N \vDash \Box _{\mathsf {PA}} (A_U(n))\}$ , then it seems that there is little hope that the $\Pi ^0_2$ set $\mathsf {PL}(\mathsf {PA})$ allows for a simple characterization.

However, a little miracle happens and by Solovay’s Completeness Theorem [Reference Solovay34] we know that $\mathsf {PL}(\mathsf {PA})$ is decidable, corresponding to what we nowadays call the GödelLöb provability logic $\mathsf {GL}$ . Likewise, Solovay proved that the set of true structural provability principles

$$ \begin{align*} \mathsf{TPL}(\mathsf{PA}) := \{ \varphi \in \mathcal L_\Box \mid \text{for any }\cdot^{*}, \text{ we have } \mathbb N \vDash \varphi^{*_{\mathsf{PA}}} \} \end{align*} $$

is also decidable and described by the well-behaved modal logic $\mathsf {GLS}$ . We observe that, a priori, $\mathsf {TPL}(\mathsf {PA})$ falls outside the arithmetical hierarchy due to Tarski’s [Reference Tarski35] result on the undefinability of arithmetical truth.

After these positive results, there was hope that the nice characterizations could be extended to the realm of quantified modal logic. Thus, the focus switched to the language $\mathcal {L}_{\Box , \forall }$ of quantified modal logic without identity, which contains $\bot $ , relation symbols, Boolean connectives, $\forall x$ , and $\Box $ , as well as the usual abbreviations such as $\Diamond $ and $\exists x$ . We define arithmetical realizations $\cdot ^{*}$ for $\mathcal {L}_{\Box , \forall }$ formulas as before with the only difference that we now map n-ary relation symbols to arithmetical formulas with n free variables and set $(\forall x \, \varphi )^{*_{\mathsf {PA}}}:= \forall y \, \varphi ^{*_{\mathsf {PA}}}$ , where y is the arithmetical variable corresponding to x. The quantified provability logic of $\mathsf {PA}$ can now be defined by analogy (cf. [Reference Boolos11]):Footnote 2

$$ \begin{align*} \mathsf{QPL}(\mathsf{PA}) := \{ A \in \mathcal{L}_{\Box, \forall} \mid \text{for any }\cdot^{*}, \text{ we have } \mathsf{PA} \vdash A^{*_{\mathsf{PA}}} \} \end{align*} $$

as well as

$$ \begin{align*} \mathsf{TQPL}(\mathsf{PA}) := \{ A \in \mathcal{L}_{\Box, \forall} \mid \text{for any }\cdot^{*}, \text{ we have } \mathbb N \vDash A^{*_{\mathsf{PA}}} \}. \end{align*} $$

It is not hard to see that $\Box \forall x \, A \to \forall x \, \Box A$ is in $\mathsf {QPL}(\mathsf {PA})$ , and for some time it was believed that one could obtain $\mathsf {QPL}(\mathsf {PA})$ by adding such principles together with predicate logical reasoning to $\mathsf {GL}$ . A first wrinkle in this hope appeared when Montagna [Reference Montagna32] published a new always provable principle falling outside this easily generated set. Soon after, Artemov [Reference Artemov2] proved that $\mathsf {TQPL}(\mathsf {PA})$ is not even arithmetically definable. Last hopes were scattered when Vardanyan [Reference Vardanyan36] and McGee [Reference McGee31] independently proved that $\mathsf {QPL}(\mathsf {PA})$ is as complex as it can possibly be: $\Pi ^0_2$ complete. Moreover, Boolos and McGee [Reference Boolos and McGee12] showed that $\mathsf {TQPL}(\mathsf {PA})$ is also as complex as possible: $\Pi ^0_1$ complete in the set of true arithmetic.

These negative results are quite robust in various ways. For example, Vardanyan [Reference Vardanyan and Adyan37] showed that restricting the language to the fragment with a single unary predicate and without any modality nesting doesn’t break the $\Pi ^0_2$ completeness.

Restricting the complexity of the realizations to, for example, $\Sigma ^0_1$ formulas doesn’t help either: the corresponding set is still $\Pi ^0_2$ complete, as shown by Berarducci [Reference Berarducci10].

One can consider $\mathsf {QPL}(T)$ for an arbitrary c.e. theory T. It is easy to see that the degenerate case of $\mathsf {QPL}(\mathsf {PA} + \Box _{\mathsf {PA}} \bot )$ is axiomatized by predicate logic together with $\Box \bot $ . However, Visser and de Jonge [Reference Visser and de Jonge38] proved that for most theories T, $\mathsf {QPL}(T)$ is indeed $\Pi ^0_2$ complete. The title of their paper is No Escape from Vardanyan’s Theorem.

Be that as it may, there have already been some “escapes” from Vardanyan’s Theorem. For example, the one-variable fragment of $\mathsf {QPL}(\mathsf {PA})$ is decidable, as shown by Artemov and Japaridze [Reference Artemov and Japaridze4]. Furthermore, an arithmetically complete quantified modal logic was proposed by Yavorsky [Reference Yavorsky, Wolter, Wansing, de Rijke and Zakharyaschev40] (see also [Reference Hao and Tourlakis24]). This logic, called $\mathsf {QGL}^b$ , assumes that in $\Box A$ every free variable of A is bound under the box. In other words, $\mathsf {QGL}^b$ is roughly $\mathsf {QGL}$ extended with the axiom schema $\Box A \to \Box \forall x \, A$ .

Our proposed fragment includes countably many variables and allows for open variables under the box. In order to “escape,” we restrict the language to its strictly positive fragment instead. This work already began in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], where we described the system $\mathsf {QRC_1}$ and proved its decidability. Here we improve on the modal results presented there and prove the arithmetical completeness theorem for $\mathsf {QRC_1}$ , previously left as a conjecture.

There is an ongoing formalizationFootnote 3 of this paper in the Coq Proof Assistant [14], covering Sections 3 and 4 (and part of Section 5) as of May 2022.

1.1 Overview of the paper

We start with a brief overview of strictly positive logics in Section 2, followed by the definition of $\mathsf {QRC_1}$ , our main object of study, in Section 3.

After that the paper is divided into two parts, the first dealing with purely modal results (Sections 46), and the second delving into arithmetic (Sections 710). The two parts are not completely independent, but a reader who wishes to skip to Section 7 need only be familiar with the definition of Kripke model with constant domain (Definition 4.1) and the constant domain completeness theorem for $\mathsf {QRC_1}$ (Theorem 5.11).

In the first part, Section 4 starts by presenting an extended definition of Kripke model that does not depend on inclusive models, with a respective soundness proof for $\mathsf {QRC_1}$ . Then in Section 5 we show that $\mathsf {QRC_1}$ is complete with respect to constant domain Kripke models. Section 6 remarks that $\mathsf {QRC_1}$ is the strictly positive fragment of every logic between $\mathsf {QK}4$ and $\mathsf {QGL}$ , and also of other quantified modal logics such as $\mathsf {QK}4 + \mathsf {BF}$ , where $\mathsf {BF} = \forall x \, \Box A \to \Box \forall x \, A$ is the well-known and arithmetically unsound Barcan Formula.

In the second part, Section 7 explains our arithmetic reading of strictly positive formulas, followed by Section 8, where we prove the arithmetical completeness of $\mathsf {QRC_1}$ . Section 9 then makes use of the arithmetical completeness theorem to show that $\mathsf {QRC_1}$ is a fragment of $\mathsf {QPL}(\mathsf {PA})$ . Section 10, which can be read right after Section 7, shows the arithmetical soundness of $\mathsf {QRC_1}$ with respect to Heyting Arithmetic.

We end with some proposed avenues for future work in Section 11.

2 Strictly positive logics

A key feature of our escape is given by restricting the language. Given variables $x, x_i, \ldots $ and a signature $\Sigma $ fixing the constants $c, c_i, \ldots $ and relation symbols $S, S_i, \ldots $ , the formulas of $\mathsf {QRC_1}$ are built up from $\top $ , n-ary relation symbols applied to n terms (which are either variables or constants), the binary $\land $ , the unary $\Diamond $ , and the quantifier $\forall x$ . The provable judgments in $\mathsf {QRC_1}$ are all of the form $\varphi \vdash \psi $ with $\varphi $ and $\psi $ in the above language.

Our formulas are related to arithmetic through realizations $\cdot ^{\circ }$ that map n-ary predicate symbols to c.e. axiomatizations of theories indexed by n parameters (details can be found in Section 7). Given a $\Sigma ^0_1$ axiomatization $\tau $ of some theory T, we extend $\cdot ^{\circ }$ to non-predicate formulas such that conjunctions are interpreted as the unions of the corresponding axiom sets ( $(\varphi \land \psi )^{\circ _{\tau }} := \varphi ^{\circ _{\tau }} \lor \psi ^{\circ _{\tau }}$ ), and universal quantification corresponds to an infinite union ( $(\forall x \, \varphi )^{\circ _{\tau }} := \exists y \, \varphi ^{\circ _{\tau }}$ , where y is the arithmetical variable corresponding to the modal variable x).

This interpretation is common in the study of reflection calculi [Reference Beklemishev7] and is more general than the usual, finitary one, mentioned in the previous section as $\cdot ^{*}$ . In the latter case, conjunctions of modal formulas are simply interpreted as conjunctions of their arithmetical counterparts. We further define and make use of this finitary notion in Section 8. While it is possible to represent finite extensions of a base theory through the finitary $\cdot ^{*}$ -style realizations, the $\cdot ^{\circ }$ approach allows for the possibility of infinitary extensions.

Using this restricted fragment and infinitary arithmetical interpretation, we define the Strictly Positive Quantified Provability Logic of a theory T as follows.

$$ \begin{align*} \mathsf{QPL}^{\mathsf{SP}}(T) := \{ \langle \varphi , \psi \rangle \mid \forall \cdot^{\circ} \, T \vdash \forall \theta \, (\Box_{\psi^{\circ_{\tau}}} \theta \to \Box_{\varphi^{\circ_{\tau}}}\theta)\}. \end{align*} $$

The main result of this paper is that, for a large class of theories T, the logic $\mathsf {QPL}^{\mathsf {SP}}(T)$ is decidable and given by the system $\mathsf {QRC_1}$ as introduced in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]. As such our paper was inspired by and contributes to three recent developments in the literature: strictly positive logics, reflection calculi, and polymodal logics.

The quintessential polymodal provability logic is $\mathsf {GLP}$ , introduced by Japaridze in 1986 [Reference Japaridze26]. The Reflection Calculus, $\mathsf {RC}$ , was first introduced by Dashkov [Reference Dashkov16] as the set of $\mathsf {GLP}$ -equivalences between strictly positive formulas. It was then axiomatized by Beklemishev [Reference Beklemishev, Bolander, Braüner, Ghilardi and Moss6], and it is the latter formulation that appears in most of the ensuing literature.

Even though $\mathsf {RC}$ has a strictly positive language, it is remarkably expressive, giving rise to an ordinal notation system [Reference Fernández-Duque19] and being an appropriate tool for $\Pi ^0_1$ ordinal analysis [Reference Beklemishev and Pakhomov9]. This is remarkable because, while $\mathsf {GLP}$ is $\mathsf {PSPACE}$ complete [Reference Shapirovsky, Areces and Goldblatt33], $\mathsf {RC}$ has a polynomial-time decision procedure. Thus, at least in the case of $\mathsf {GLP}$ , restricting the language to the strictly positive fragment is very worthwhile.

$\mathsf {QRC_1}$ was inspired by $\mathsf {RC}_1$ , which is the unimodal fragment of $\mathsf {RC}$ . We hoped to emulate $\mathsf {RC}$ ’s success in these two dimensions, obtaining a useful calculus with a simpler complexity than the original ( $\mathsf {QPL}(\mathsf {PA})$ in this case). We already see in this paper that the latter goal was achieved, since $\mathsf {QRC_1}$ is decidable while $\mathsf {QPL}(\mathsf {PA})$ is $\Pi ^0_2$ complete. However, the main expressibility tool available in $\mathsf {RC}$ , the iterated consistency statements (also known as worms [Reference Beklemishev, Chatzidakis, Koepke and Pohlers5]), are not very interesting when there is only one modality. Thus we plan to extend $\mathsf {QRC_1}$ to $\mathsf {QRC}_\Lambda $ in the future.

When considering a strictly positive logic P, one may ask whether there is some modal logic L whose strictly positive fragment $L^{\mathsf {SP}}$ coincides with P (cf. [Reference Beklemishev and Odintsov8]). In that case we would have

(2) $$ \begin{align} \varphi \vdash_P \psi \iff \varphi \vdash_{L^{\mathsf{SP}}} \psi \iff L \vdash \varphi \to \psi , \end{align} $$

where $\varphi $ and $\psi $ are strictly positive formulas.

We know that $\mathsf {RC}$ is the strictly positive fragment of $\mathsf {GLP}$ (in the sense of (2)), and in Section 6 we show that $\mathsf {QRC_1}$ is the strictly positive fragment of $\mathsf {QGL}$ . However, there are strictly positive logics with no such counterpart, such as $\mathsf {RC}$ together with the persistence axiom $\langle \omega \rangle \varphi \vdash \varphi $ . This system was described in [Reference Beklemishev7] as $\mathsf {RC}\omega $ .

The study of strictly positive logics has also been developed in other communities, most notably in the field of description logics, as in [Reference Kikot, Kurucz, Tanaka, Wolter and Zakharyaschev27, Reference Kurucz, Wolter, Zakharyaschev, Beklemishev, Goranko and Shehtman30]. Furthermore, there is a significant body of work done about (non-strictly) positive logics (see [Reference Celani and Jansana13, Reference Dunn17]).

3 Quantified Reflection Calculus with one modality

The Quantified Reflection Calculus with one modality, or $\mathsf {QRC_1}$ , is a sequent logic in a strictly positive predicate modal language introduced in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1].

The free variables of a formula $\varphi $ are defined as usual, and denoted by ${\mathrm {fv}}(\varphi )$ . The expression $\varphi [x {\leftarrow } t]$ denotes the formula $\varphi $ with all free occurrences of the variable x simultaneously replaced by the term t. We say that t is free for x in $\varphi $ if no occurrence of a free variable in t becomes bound in $\varphi [x {\leftarrow } t]$ .

The axioms and rules of $\mathsf {QRC_1}$ are listed in the following definition from [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]. Here we removed the axiom $\Diamond \forall x \, \varphi \vdash \forall x \, \Diamond \varphi $ because it is an easy consequence of the calculus without it.

Definition 3.1 ( $\mathsf {QRC_1}$ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Let $\Sigma $ be a signature and $\varphi $ , $\psi $ , and $\chi $ be any formulas in that language. The axioms and rules of $\mathsf {QRC_1}$ are the following:

  1. (i) $\varphi \vdash \top $ and $\varphi \vdash \varphi $ ;

  2. (ii) $\varphi \land \psi \vdash \varphi $ and $\varphi \land \psi \vdash \psi $ ;

  3. (iii) if $\varphi \vdash \psi $ and $\varphi \vdash \chi $ , then $\varphi \vdash \psi \land \chi $ ;

  4. (iv) if $\varphi \vdash \psi $ and $\psi \vdash \chi $ , then $\varphi \vdash \chi $ ;

  5. (v) if $\varphi \vdash \psi $ , then $\Diamond \varphi \vdash \Diamond \psi $ ;

  6. (vi) $\Diamond \Diamond \varphi \vdash \Diamond \varphi $ ;

  7. (vii) if $\varphi \vdash \psi $ , then $\varphi \vdash \forall x \, \psi $ ( $x \notin {\mathrm {fv}}(\varphi )$ );

  8. (viii) if $\varphi [x {\leftarrow } t] \vdash \psi $ , then $\forall x \, \varphi \vdash \psi $ (t free for x in $\varphi $ );

  9. (ix) if $\varphi \vdash \psi $ , then $\varphi [x {\leftarrow } t] \vdash \psi [x {\leftarrow } t]$ (t free for x in $\varphi $ and $\psi $ );

  10. (x) if $\varphi [x {\leftarrow } c] \vdash \psi [x {\leftarrow } c]$ , then $\varphi \vdash \psi $ (c not in $\varphi $ nor $\psi $ ).

If $\varphi \vdash \psi $ , we say that $\psi $ follows from $\varphi $ in $\mathsf {QRC_1}$ . When the signature is not clear from the context, we write $\varphi \vdash _\Sigma \psi $ instead.

We observe that our axioms do not include universal quantifier elimination. However, this and various other rules are readily available via the following easy lemma.

Lemma 3.2. The following are theorems (or derivable rules) of $\mathsf {QRC_1}$ :

  1. 1. $\forall x \, \forall y \, \varphi \vdash \forall y \, \forall x \, \varphi $ ;

  2. 2. $\forall x \, \varphi \vdash \varphi [x {\leftarrow } t]$ (t free for x in $\varphi $ );

  3. 3. $\Diamond \forall x \, \varphi \vdash \forall x \, \Diamond \varphi $ ;Footnote 4

  4. 4. $\forall x \, \varphi \vdash \forall y \, \varphi [x {\leftarrow } y]$ (y free for x in $\varphi $ and $y \notin {\mathrm {fv}}(\varphi )$ );

  5. 5. if $\varphi \vdash \psi $ , then $\varphi \vdash \psi [x {\leftarrow } t]$ (x not free in $\varphi $ and t free for x in $\psi $ );

  6. 6. if $\varphi \vdash \psi [x {\leftarrow } c]$ , then $\varphi \vdash \forall x \, \psi $ (x not free in $\varphi $ and c not in $\varphi $ nor $\psi $ ).

The following are two useful complexity measures on the formulas of $\mathsf {QRC_1}$ .

Definition 3.3 ( ${\mathrm {d}}_\Diamond $ , ${\mathrm {d}}_\forall $ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Given a formula $\varphi $ , its modal depth ${\mathrm {d}}_\Diamond (\varphi )$ is defined inductively as follows:

  • ${\mathrm {d}}_\Diamond (\top ) := {\mathrm {d}}_\Diamond (S(x_0, \ldots , x_{n-1})) := 0$ ;

  • ${\mathrm {d}}_\Diamond (\psi \land \chi ) := \max \{{\mathrm {d}}_\Diamond (\psi ), {\mathrm {d}}_\Diamond (\chi )\}$ ;

  • ${\mathrm {d}}_\Diamond (\forall x \, \psi ) := {\mathrm {d}}_\Diamond (\psi )$ ;

  • ${\mathrm {d}}_\Diamond (\Diamond \psi ) := {\mathrm {d}}_\Diamond (\psi ) + 1$ .

Given a finite set of formulas $\Gamma $ , its modal depth is ${\mathrm {d}}_\Diamond (\Gamma ) := \max _{\varphi \in \Gamma }\{{\mathrm {d}}_\Diamond (\varphi )\}$ .

The definition of quantifier depth ${\mathrm {d}}_\forall $ is analogous except for:

  • ${\mathrm {d}}_\forall (\forall x \, \psi ) := {\mathrm {d}}_\forall (\psi ) + 1$ and

  • ${\mathrm {d}}_\forall (\Diamond \psi ) := {\mathrm {d}}_\forall (\psi )$ .

The modal depth provides a necessary condition for derivability, which in particular implies irreflexivity.

Lemma 3.4 [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1].

Let $\varphi $ and $\psi $ be formulas in the language of $\mathsf {QRC_1}$ .

  • If $\varphi \vdash \psi $ , then ${\mathrm {d}}_\Diamond (\varphi ) \geq {\mathrm {d}}_\Diamond (\psi )$ .

  • $\varphi \not \vdash \Diamond \varphi $ .

Finally, the signature of $\mathsf {QRC_1}$ can be extended without strengthening the calculus.

Lemma 3.5 [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1].

Let $\Sigma $ be a signature and let C be a collection of constants not yet occurring in $\Sigma $ . By $\Sigma _C$ we denote the signature obtained by including these new constants C in $\Sigma $ . Let $\varphi , \psi $ be formulas in the language of $\Sigma $ . Then, if $\varphi \vdash _{\Sigma _C} \psi $ , so does $\varphi \vdash _\Sigma \psi $ .

4 Relational semantics

$\mathsf {QRC_1}$ was proven sound and complete with respect to relational semantics in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]. Here we extend both of those results in the following ways: we relax the requirement for the adequateness of a frame, and we prove constant domain completeness: that if $\varphi \not \vdash \psi $ then there exists a counter model that, in addition to being finite and irreflexive, also has a constant domain.

We begin by slightly changing the definition of frame and relational model presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]. There, models for $\mathsf {QRC_1}$ were described as a number of first-order models (the worlds) connected through a transitive relation R. We additionally required inclusiveness: that whenever w and u are worlds connected through R, the domain of w be included in the domain of u. We then used the inclusion (identity) function $\iota _{w, u}$ to refer to the element of the domain of u corresponding to an element in the domain of w.

Here, we no longer have the inclusiveness restriction. In fact, as we will see bellow, any configuration of domains is sound as long as the functions $\eta _{w, u}$ relating the domain of w with the domain of u respect the transitivity of R. This is clearly the case if the frame is inclusive and $\eta _{w, u} = \iota _{w, u}$ , so the definitions presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] are a particular case of the ones presented here.

Definition 4.1. A relational model $\mathcal {M}$ in a signature $\Sigma $ is a tuple $\langle W, R, \{M_w\}_{w \in W}, \{\eta _{w, v}\}_{wRv}, \{I_w\}_{w \in W},\{J_w\}_{w \in W} \rangle $ where:

  • W is a non-empty set (the set of worlds, where individual worlds are referred to as $w, u, v$ , etc.);

  • R is a binary relation on W (the accessibility relation);

  • each $M_w$ is a finite set (the domain of the world w, whose elements are referred to as $d, d_0, d_1$ , etc.);

  • if $wRv$ , then $\eta _{w, v}$ is a function from $M_w$ to $M_v$ ;

  • for each $w \in W$ , the interpretation $I_w$ assigns an element of the domain $M_w$ to each constant $c \in \Sigma $ , written $c^{I_w}$ ; and

  • for each $w \in W$ , the interpretation $J_w$ assigns a set of tuples $S^{J_w} \subseteq \wp ((M_w)^n)$ to each n-ary relation symbol $S \in \Sigma $ .

The $\langle W, R, \{M_w\}_{w \in W}, \{\eta _{w, v}\}_{wRv} \rangle $ part of the model is called its frame. We say that the frame (or model) is finite if W is finite, and that it is constant domain if all the $M_w$ coincide and all the $\eta _{w, u}$ are the identity function.

The relevant frames and models will need to satisfy a number of requisites.

Definition 4.2. A frame $\mathcal {F}$ is adequate if:

  • R is transitive: if $wRu$ and $uRv$ , then $wRv$ ; and

  • $\eta $ functions respect transitivity: if $wRu$ and $uRv$ , then $\eta _{w, v} = \eta _{u, v} \circ \eta _{w, u}$ .Footnote 5

A model is adequate if it is based on an adequate frame and it is:

  • concordant: if $wRu$ , then $c^{I_u} = \eta _{w, u}(c^{I_w})$ for every constant c.

Note that in an adequate and rooted model the interpretation of the constants is fully determined by their interpretation at the root.

As in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], we use assignments to define truth at a world in a first-order model. A $w$ -assignment g is a function assigning a member of the domain $M_w$ to each variable in the language. Any $w$ -assignment can be seen as a $v$ -assignment as long as $wRv$ , by composing it with $\eta _{w, v}$ on the left. We write $g^v$ to shorten $\eta _{w, v} \circ g$ when $w$ is clear from the context.

Two $w$ -assignments g and h are x-alternative, written $g \sim _{x} h$ , if they coincide on all variables other than x. A $w$ -assignment g is extended to terms by defining $g(c) := c^{I_w}$ for any constant c. Note that this meshes nicely with the concordant restriction of an adequate model: for any term t, if $wRv$ then $g^v(t) = \eta _{w, v}(g(t))$ .

We are finally ready to define satisfaction at a world. This definition is a straightforward adaptation of the one presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] to our current definition of model. The only difference is in the case of $\Diamond \varphi $ , where we use $g^v$ instead of $g^\iota $ .

Definition 4.3. Let $\mathcal {M} = \langle W, R, \{M_w\}_{w \in W}, \{\eta _{w, u}\}_{wRu}, \{I_w\}_{w \in W}, \{J_w\}_{w \in W} \rangle $ be an adequate model in some signature $\Sigma $ , and let $w \in W$ be a world, g be a $w$ -assignment, S be an n-ary relation symbol, and $\varphi , \psi $ be formulas in the language of  $\Sigma $ .

We define $\mathcal {M}, w \Vdash ^g \varphi $ ( $\varphi $ is true at w under g) by induction on $\varphi $ as follows.

  • $\mathcal {M}, w \Vdash ^g \top $ ;

  • $\mathcal {M}, w \Vdash ^g S(t_0, \ldots , t_{n-1})$ iff $\langle g(t_0), \ldots , g(t_{n-1}) \rangle \in S^{J_w}$ ;

  • $\mathcal {M}, w \Vdash ^g \varphi \land \psi $ iff both $\mathcal {M}, w \Vdash ^g \varphi $ and $\mathcal {M}, w \Vdash ^g \psi $ ;

  • $\mathcal {M}, w \Vdash ^g \Diamond \varphi $ iff there is a $v \in W$ such that $wRv$ and $\mathcal {M}, v \Vdash ^{g^v} \varphi $ ;

  • $\mathcal {M}, w \Vdash ^g \forall x \, \varphi $ iff for all $w$ -assignments h such that $h \sim _{x} g$ , we have ${\mathcal {M}, w \Vdash ^{h} \varphi }$ .

Theorem 4.4 (Relational soundness).

If $\varphi \vdash \psi $ , then for any adequate model $\mathcal {M}$ , for any world $w \in W$ , and for any $w$ -assignment g,

$$ \begin{align*} \mathcal{M}, w \Vdash^g \varphi \Longrightarrow \mathcal{M}, w \Vdash^g \psi. \end{align*} $$

Proof. By induction on the proof of $\varphi \vdash \psi $ , making the same arguments as in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]. Here we highlight only the transitivity axiom, where the transitivity of the $\eta $ functions comes into play, and also remark on the generalization on constants rule.

The transitivity axiom is $\Diamond \Diamond \varphi \vdash \Diamond \varphi $ , so assume that $\mathcal {M}, w \Vdash ^g \Diamond \Diamond \varphi $ . Then there is a world $v$ such that $wRv$ and $\mathcal {M}, v \Vdash ^{\eta _{w, v} \circ g} \Diamond \varphi $ , and also a subsequent world u such that $vRu$ and $\mathcal {M}, u \Vdash ^{\eta _{v, u} \circ (\eta _{w, v} \circ g)} \varphi $ . Since R is transitive, we know that $wRu$ and thus that $\eta _{v, u} \circ (\eta _{w, v} \circ g)$ coincides with $\eta _{w, u} \circ g$ . Then $\mathcal {M}, u \Vdash ^{\eta _{w, u} \circ g} \psi $ , and consequently $\mathcal {M}, w \Vdash ^{g} \Diamond \varphi $ , as desired.

The soundness of the generalization on constants rule, Rule 3.1(x), is the most involved part of the proof presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] and depends on the construction of a model where the interpretation of a constant is changed. Building that model in this context is a simple matter of taking care to propagate that change to all future worlds using the $\eta $ functions.⊣

We end this section by noting that, even though the language of $\mathsf {QRC_1}$ is quite restricted, even its $\forall $ fragment requires counter-models with arbitrarily large domains. For example, the sequent

$$ \begin{align*} \forall x, y \, S(x, x, y) \land \forall x, y \, S(x, y, x) \land \forall x, y \, S(y, x, x) \vdash \forall x, y, z \, S(x, y, z) \end{align*} $$

is unprovable in $\mathsf {QRC_1}$ , but satisfied by every world with at most two domain elements. This reasoning can be extended to any n: if S is an n-ary predicate symbol, let $\varphi _n$ be the conjunction of the $n(n-1)/2$ formulas of the form $\forall x_0, \ldots , x_{n-2} \, S(\ldots , x_0, \ldots , x_0, \ldots )$ , with $x_0$ appearing in every possible pair of positions and every other position filled by a unique variable. Then $\varphi _n$ does not entail $ \psi _n := \forall x_0, \ldots , x_{n-1} \, S(x_0, \ldots , x_{n-1})$ but any world with at most $n-1$ domain elements that satisfies $\varphi _n$ must also satisfy $\psi _n$ .

5 Constant domain completeness

In [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] we proved relational completeness by building a term model that satisfies $\varphi $ and doesn’t satisfy $\psi $ when $\varphi \not \vdash \psi $ . That construction provides a finite, irreflexive, and rooted model with increasing domains. Here we show that it is possible to build a constant domain model instead, that is, a model where the domain of every world is exactly the same. This is extremely useful to prove the arithmetical completeness theorem in Section 8.

Before starting the formal proof, we briefly describe the main idea. The term models we build are such that each world is a pair of sets of closed formulas $p = \langle p^+, p^- \rangle $ . The first set, $p^+$ , is the set of formulas that will be satisfied at that world, or the positive part. The second set, $p^-$ , is the set of formulas that will not be satisfied at that world, or the negative part. All worlds must be well-formed with respect to some finite set of closed formulas $\Phi $ , which means that:

  • p is closed: every formula in $p^+$ and every formula in $p^-$ is closed;

  • p is $\Phi $ -maximal: every formula of $\Phi $ is in either $p^+$ or $p^-$ (and there are no formulas in p but not in $\Phi $ );

  • p is consistent: if $\delta \in p^-$ then $\bigwedge p^+ \not \vdash \delta $ ; and

  • p is fully-witnessed: if $\forall x \, \varphi \in p^-$ then there is a constant c such that ${\varphi [x {\leftarrow } c] \in p^-}$ .

In that case we say that p is $\Phi $ -maximal consistent and fully witnessed, or $\Phi $ -MCW for short (the closeness condition is included in the concept, although in practice almost every formula in this section will be closed). If p and q are pairs, we write $p \subseteq q$ when $p^+ \subseteq q^+$ and $p^- \subseteq q^-$ . Furthermore, if $\Gamma $ is a set of formulas we write $p \subseteq \Gamma $ instead of $p^+ \cup p^- \subseteq \Gamma $ .

We want to have $\Phi $ -MCW pairs where $\Phi $ is closed under subformulas. However, the naive subformulas of $\forall x \, \varphi $ might be open. In order to avoid that, we use the notion of closure with respect to a set of constants C defined in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], where $\varphi [x {\leftarrow } c]$ is a valid subformula of $\forall x \, \varphi $ as long as $c \in C$ .

Definition 5.1 ( $\mathcal {C}\ell _C$ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Given a set of constants C, the closure of a formula $\varphi $ under C, written $\mathcal {C}\ell _C(\varphi )$ , is defined by induction on the formula as such: $\mathcal {C}\ell _C(\top ) := \{\top \}$ ; $\mathcal {C}\ell _C(S(t_0, \ldots , t_{n-1})) := \{S(t_0, \ldots , t_{n-1}), \top \}$ ; $\mathcal {C}\ell _C(\varphi \land \psi ) := \{\varphi \land \psi \} \cup \mathcal {C}\ell _C(\varphi ) \cup \mathcal {C}\ell _C(\psi )$ ; $\mathcal {C}\ell _C(\Diamond \varphi ) := \{\Diamond \varphi \} \cup \mathcal {C}\ell _C(\varphi )$ ; and

$$ \begin{align*} \mathcal{C}\ell_C(\forall x \, \varphi) := \{\forall x \, \varphi\} \cup \bigcup_{c \in C} \mathcal{C}\ell_C(\varphi[x {\leftarrow} c]). \end{align*} $$

The closure under C of a set of formulas $\Gamma $ is the union of the closures under C of each of the formulas in $\Gamma $ :

$$ \begin{align*} \mathcal{C}\ell_C(\Gamma) := \bigcup_{\gamma \in \Gamma} \mathcal{C}\ell_C(\gamma). \end{align*} $$

The closure of a pair p is defined as the closure of $p^+ \cup p^-$ .

Going back to the overview of the completeness proof, suppose that $\varphi \not \vdash \psi $ , (assuming for now that $\varphi $ and $\psi $ are closed). Defining $p := \langle \{\varphi \}, \{\psi \} \rangle $ , the counter-model will be rooted on a $\mathcal {C}\ell _C(p)$ -MCW extension of p, where C is a set of constants to determine. The set of constants C will be used as the domain of the root. Note that p is already closed and consistent, so taking the step to maximality is as simple as deciding whether to add each $\chi \in \mathcal {C}\ell _C(p)$ to the positive or negative part of the root without ruining its consistency. The hard part is doing so in a way that guarantees that the resulting pair is fully witnessed.

The completeness proof shown in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] uses the observation that if $\bigwedge p^+ \not \vdash \forall x \, \chi $ , then also $\bigwedge p^+ \not \vdash \chi [x {\leftarrow } c]$ , as long as c does not appear in either $p^+$ or $\chi $ (Lemma 3.2.6). This suggests a way of sorting the formulas of $\mathcal {C}\ell _C(p)$ into positive and negative: mark a formula as positive if and only if it is a consequence of $p^+$ . This guarantees that there are witnesses for the negative universal formulas as long as there are enough constants to go around. The way to make sure that there are enough constants is precisely the difference between the proof presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] and the proof presented here. To that end, we introduce the following definition.

Definition 5.2 ( ${\mathrm {d}}_{{\mathrm {const}}}$ ).

The number of different constants in a formula $\varphi $ is represented by ${\mathrm {d}}_{{\mathrm {const}}}(\varphi )$ . The maximum number of different constants per formula in a set of formulas $\Gamma $ is defined as ${\mathrm {d}}_{{\mathrm {const}}}(\Gamma ) := \max _{\varphi \in \Gamma }\{{\mathrm {d}}_{{\mathrm {const}}}(\varphi )\}$ .

Note that ${\mathrm {d}}_{{\mathrm {const}}}(\Gamma )$ is not the number of different constants appearing in $\Gamma $ , but the maximum number of different constants in any single formula of $\Gamma $ . For example, ${\mathrm {d}}_{{\mathrm {const}}}(\{S_0(c_0, c_1), S_1(c_2)\}) = 2$ .

We observe that the maximum number of distinct constants per formula in the closure under C of a set of formulas can be bounded by a number that does not depend on C.

Fact 5.3. For any formula $\varphi $ , set of formulas $\Phi $ , and set of constants C:

  • ${\mathrm {d}}_{{\mathrm {const}}}(\varphi ) \leq {\mathrm {d}}_{{\mathrm {const}}}(\varphi [x {\leftarrow } c]) \leq {\mathrm {d}}_{{\mathrm {const}}}(\varphi ) + 1$ ;

  • ${\mathrm {d}}_{{\mathrm {const}}}(\varphi ) \leq {\mathrm {d}}_{{\mathrm {const}}}(\mathcal {C}\ell _C(\varphi )) \leq {\mathrm {d}}_{{\mathrm {const}}}(\varphi ) + {\mathrm {d}}_\forall (\varphi )$ ;

  • ${\mathrm {d}}_{{\mathrm {const}}}(\Phi ) \leq {\mathrm {d}}_{{\mathrm {const}}}(\mathcal {C}\ell _C(\Phi )) \leq {\mathrm {d}}_{{\mathrm {const}}}(\Phi ) + {\mathrm {d}}_\forall (\Phi )$ .

We are now ready to prove a Lindenbaum-like lemma.

Lemma 5.4. Given a finite signature $\Sigma $ with constants C and a finite set of closed formulas $\Phi $ in the language of $\Sigma $ such that $|C|> 2{\mathrm {d}}_{{\mathrm {const}}}(\Phi ) + 2{\mathrm {d}}_\forall (\Phi )$ , if $p \subseteq \mathcal {C}\ell _C(\Phi )$ is a closed consistent pair and $p^+$ is a singleton, then there is a pair $q \supseteq p$ in the language of $\Sigma $ such that q is $\mathcal {C}\ell _C(\Phi )$ -MCW, and ${\mathrm {d}}_\Diamond (q^+) = {\mathrm {d}}_\Diamond (p^+)$ .

Proof. Like in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], we start by defining a pair $q \supseteq p$ such that for each $\chi \in \mathcal {C}\ell _C(\Phi )$ , $\chi \in q^+$ if and only if $p^+ \vdash \chi $ (otherwise $\chi \in q^-$ ). It is easy to see that this pair q is $\mathcal {C}\ell _C(\Phi )$ -maximal consistent, and we have ${\mathrm {d}}_\Diamond (q^+) = {\mathrm {d}}_\Diamond (p^+)$ by Lemma 3.4.

It remains to show that q is fully witnessed. Let $\forall x \, \psi $ be a formula in $q^-$ . We claim that there is $d \in C$ such that d does not appear either in $p^+$ or in $\forall x \, \psi $ . For this it is enough to see that $|C|> {\mathrm {d}}_{{\mathrm {const}}}(\bigwedge p^+) + {\mathrm {d}}_{{\mathrm {const}}}(\forall x \, \psi )$ , which is the same as $|C|> {\mathrm {d}}_{{\mathrm {const}}}(p^+) + {\mathrm {d}}_{{\mathrm {const}}}(\forall x \, \psi )$ because $p^+$ is a singleton by assumption. Since $p^+ \subseteq \mathcal {C}\ell _C(\Phi )$ , we know that ${\mathrm {d}}_{{\mathrm {const}}}(p^+) \leq {\mathrm {d}}_{{\mathrm {const}}}(\mathcal {C}\ell _C(\Phi ))$ , and similarly for $\forall x \, \psi $ . Then by Fact 5.3 we may conclude that ${\mathrm {d}}_{{\mathrm {const}}}(p^+) + {\mathrm {d}}_{{\mathrm {const}}}(\forall x \, \psi ) \leq 2{\mathrm {d}}_{{\mathrm {const}}}(\Phi ) + 2{\mathrm {d}}_\forall (\Phi )$ , which suffices by our assumption on the size of C.

Since $d \in C$ does not appear in either $p^+$ or $\forall x \, \psi $ , we conclude that $p^+ \not \vdash \psi [x {\leftarrow } d]$ by Lemma 3.2.6, and consequently $\psi [x {\leftarrow } d] \in q^-$ as desired.⊣

We now recall the definition of $\hat {R}$ from [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], which is the relation we use to connect the worlds of the term model.

Definition 5.5 ( $p\hat {R}q$ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

The relation $\hat {R}$ between pairs is such that $p\hat {R}q$ if and only if both of following hold:

  • for any formula $\Diamond \varphi \in p^-$ we have $\varphi , \Diamond \varphi \in q^-$ ; and

  • there is some formula $\Diamond \psi \in p^+ \cap q^-$ .

Lemma 5.6 [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1].

The relation $\hat {R}$ restricted to consistent pairs is transitive and irreflexive.

We now see that if $w$ is a $\mathcal {C}\ell _C(\Phi )$ -MCW pair with $\Diamond \varphi \in w^+$ , we can find a $\mathcal {C}\ell _C(\Phi )$ -MCW pair $v$ with $\varphi \in v^+$ and $w\hat {R}v$ . The proof is the same as in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], except that we now use Lemma 5.4 to obtain constant domains throughout the model.

Lemma 5.7 (Pair existence).

Let $\Sigma $ be a signature with a finite set of constants C, and $\Phi $ be a finite set of closed formulas in the language of $\Sigma $ such that $|C|> 2{\mathrm {d}}_{{\mathrm {const}}}(\Phi ) + 2{\mathrm {d}}_\forall (\Phi )$ . If p is a $\mathcal {C}\ell _C(\Phi )$ -MCW pair and $\Diamond \varphi \in p^+$ , then there is a $\mathcal {C}\ell _{C}(\Phi )$ -MCW pair q such that $p\hat {R}q$ , $\varphi \in q^+$ , and ${\mathrm {d}}_\Diamond (q^+) < {\mathrm {d}}_\Diamond (p^+)$ .

Proof. Consider the pair r defined as $r^+ := \{\varphi \}$ and $r^- := \{\delta , \Diamond \delta \mid \Diamond \delta \in p^-\} \cup \{\Diamond \varphi \}$ . It is easy to check that r is consistent and that $p\hat {R}r$ (details can be found in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]), and clearly $r \subseteq \mathcal {C}\ell _C(\Phi )$ . We then use Lemma 5.4 to obtain a $\mathcal {C}\ell _{C}(\Phi )$ -MCW pair $q \supseteq r$ such that ${\mathrm {d}}_\Diamond (q^+) = {\mathrm {d}}_\Diamond (r^+) = {\mathrm {d}}_\Diamond (\varphi ) < {\mathrm {d}}_\Diamond (p^+)$ . We obtain $p\hat {R}q$ as a straightforward consequence of $p\hat {R}r$ .⊣

We can now define an adequate and constant domain model $\mathcal {M}[p]$ from any given finite and consistent pair p such that $\mathcal {M}[p]$ satisfies the formulas in $p^+$ and doesn’t satisfy the formulas in $p^-$ . The idea is exactly the same as in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]: build a term model where each world w is a $\mathcal {C}\ell _{M}(p)$ -MCW pair, and the worlds are related by (a sub-relation of) $\hat {R}$ .

Definition 5.8. Let $\Sigma $ be a signature. Given a finite consistent pair p of closed formulas in $\Sigma $ such that $p^+$ is a singleton,Footnote 6 we define an adequate model $\mathcal {M}[p]$ . Here we will use $\Phi := p^+ \cup p^-$ .

Let C be a set of at least $2{\mathrm {d}}_{{\mathrm {const}}}(\Phi ) + 2{\mathrm {d}}_\forall (\Phi ) + 1$ different constants, including all of the ones appearing in $\Sigma $ and adding more if necessary. The pairs of formulas we work with are in the signature $\Sigma $ extended by C.

We start by defining the underlying frame in an iterative manner. The root is given by Lemma 5.4 applied to C and p, obtaining the $\mathcal {C}\ell _C(\Phi )$ -MCW pair q. Frame $\mathcal {F}^0$ is then defined such that its set of worlds is $W^0 := \{q\}$ , its relation $R^0$ is empty, and the domain of q is $M^0_{q} := C$ .

Assume now that we already have a frame $\mathcal {F}^i$ , and we set out to define $\mathcal {F}^{i+1}$ as an extension of $\mathcal {F}^i$ . For each leaf $w$ of $\mathcal {F}^i$ , i.e., each world such that there is no world $v \in \mathcal {F}^i$ with $w R^i v$ , and for each formula $\Diamond \varphi \in w^+$ , use Lemma 5.7 to obtain a $\mathcal {C}\ell _C(\Phi )$ -MCW pair $v$ such that $w\hat {R}v$ , $\varphi \in v^+$ , and ${\mathrm {d}}_\Diamond (v^+) < {\mathrm {d}}_\Diamond (w^+)$ . Now add $v$ to $W^{i+1}$ , add $\langle w, v \rangle $ to $R^{i+1}$ , define $M^{i+1}_v$ as C, and define $\eta _{w, v}$ as the identity function.

The process described above terminates because each pair is finite and the modal depth of $\mathcal {C}\ell _C(\Phi )$ (and consequently of $w^+$ , for any $w \subseteq \mathcal {C}\ell _C(\Phi )$ ) is also finite. Thus there is a final frame $\mathcal {F}^m$ , for some natural number m. This frame is constant domain by construction, but not transitive. We obtain $\mathcal {F}[p]$ as the transitive closure of $\mathcal {F}^m$ , which is clearly still constant domain. The $\eta $ functions are all the identity in C, thus satisfying the transitivity condition. We conclude that the frame $\mathcal {F}[p]$ is adequate.

In order to obtain the model $\mathcal {M}[p]$ based on the frame $\mathcal {F}[p]$ , let $I_q$ take constants in $\Sigma $ to their corresponding version as domain elements. If $w$ is any other world, let $I_w$ coincide with $I_q$ . This is necessary to make sure that the model is concordant, because q sees every other world, and is sufficient to see that $\mathcal {M}[p]$ is adequate. Finally, given an n-ary predicate letter S and a world $w$ , define $S^{J_w}$ as the set of n-tuples $\langle d_0, \ldots , d_{n-1}\rangle \subseteq (M_w)^n$ such that $S(d_0, \ldots , d_{n-1}) \in w^+$ .

Since everything up until now was meant for closed formulas, and furthermore we are potentially adding new constants to the signature of the formulas we care about, we provide a way of replacing the free variables of a formula with constants.

Definition 5.9 ( $\varphi ^g$ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Given a formula $\varphi $ in a signature $\Sigma $ and a function g from the set of variables to a set of constants in some signature $\Sigma ' \supseteq \Sigma $ , we define the formula $\varphi ^g$ in the signature $\Sigma '$ as $\varphi $ with each free variable x simultaneously replaced by $g(x)$ .

The constant domain model defined above coincides with the non-constant domain definition provided in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] in everything other than that it refers to the stronger Lemmas 5.4 and 5.7 that keep the domain constant. Thus the Truth Lemma holds with exactly the same proof.

Lemma 5.10 (Truth Lemma [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Let $\Sigma $ be a signature. For any finite non-empty consistent pair p of closed formulas in the language of $\Sigma $ , world $w \in \mathcal {M}[p]$ , $w$ -assignment g, and formula $\varphi $ in the language of $\Sigma $ such that $\varphi ^g \in \mathcal {C}\ell _{M_w}(p)$ , we have that

$$ \begin{align*} \mathcal{M}[p], w \Vdash^g \varphi \iff \varphi^g \in w^+. \end{align*} $$

We are now ready to prove the constant domain completeness theorem.

Theorem 5.11 (Constant domain completeness).

Let $\Sigma $ be a signature and $\varphi , \psi $ be formulas in $\Sigma $ . If $\varphi \not \vdash \psi $ , then there is an adequate, finite, irreflexive, and constant domain model $\mathcal {M}$ , a world $w \in W$ , and a $w$ -assignment g such that

$$ \begin{align*} \mathcal{M}, w \Vdash^g \varphi \quad \text{and} \quad \mathcal{M}, w \not\Vdash^g \psi. \end{align*} $$

Proof. As in the original proof of relational completeness presented in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1], define a new constant $c_x$ for each free variable x of $\varphi $ and $\psi $ and let $\Sigma '$ be the signature $\Sigma $ augmented with these new constants and a dummy constant $c_0$ . Let g be the assignment taking each free variable x of $\varphi $ and $\psi $ to $c_x$ and every other variable to $c_0$ . Note that $\varphi ^g \not \vdash _{\Sigma '} \psi ^g$ by Rule 3.1(x) and Lemma 3.5. Build $\mathcal {M} := \mathcal {M}[\langle \{\varphi ^g\}, \{\psi ^g\} \rangle ]$ as described in Definition 5.8, with root $w$ . Then by Lemma 5.10 we have both $\mathcal {M}, w \Vdash ^g \varphi $ and $\mathcal {M}, w \not \Vdash ^g \psi $ , as desired.⊣

6 The strictly positive fragment of $\mathsf {QK}4$ and $\mathsf {QGL}$

Consider the language of full quantified modal logic with no equality nor constant nor function symbols, $\mathcal {L}_{\Box , \forall }$ . We use upper case letters to refer to formulas in this language. A propositional modal logic $\mathsf {S}$ can be extend to a quantified modal logic $\mathsf {QS}$ as described in [Reference Hughes and Cresswell25] by extending the language and adding the following axiom schema and rule:

  • $[\forall \mathsf {E}]$ $\forall x \, A \to A[x {\leftarrow } y]$ (y free for x in A);

  • $[\forall \mathsf {I}]$ if $\mathsf {QS} \vdash A \to B$ , then $\mathsf {QS} \vdash A \to \forall x \, B$ ( $x \notin {\mathrm {fv}}(A)$ ).

The logic $\mathsf {K}4$ is obtained from the smallest normal modal logic $\mathsf {K}$ by adding the 4 Axiom $\Box A \to \Box \Box A$ . The logic $\mathsf {GL}$ is $\mathsf {K}4$ extended with Löb’s Axiom $\Box (\Box A \to A) \to \Box A$ . Dashkov [Reference Dashkov16] showed that $\mathsf {RC}_1$ is the strictly positive fragment of both $\mathsf {K}4$ and $\mathsf {GL}$ (and of any logic between them), in the sense that, if $\varphi $ and $\psi $ are built up from $\top $ , propositional symbols, conjunctions, and diamonds, then $\varphi \vdash _{\mathsf {RC}_1} \psi $ if and only if $\mathsf {K}4 \vdash \varphi \to \psi $ , and similarly for $\mathsf {GL}$ . In this section we show that this equivalence is maintained in the predicate case.

Since $\mathsf {QRC_1}$ includes constants in its language and predicate modal logics are often presented without them, we define a map from the language of $\mathsf {QRC_1}$ to its constant-free subset, replacing constants with fresh variables. Thus, given a finite set of $\mathsf {QRC_1}$ formulas $\Phi $ where the constants appearing in $\Phi $ are $\boldsymbol {c} = c_0, \ldots , c_{n-1}$ , let $\boldsymbol {x} = x_0, \ldots , x_{n-1}$ be fresh variables with respect to $\Phi $ . Then we define $\varphi ^\Phi := \varphi [c_0 {\leftarrow } x_0]\cdots [c_{n-1} {\leftarrow } x_{n-1}]$ for each $\varphi \in \Phi $ , and abbreviate this long substitution with the notation $\varphi [\boldsymbol {c} {\leftarrow } \boldsymbol {x}]$ . Thus we also have $\varphi = \varphi ^\Phi [\boldsymbol {x} {\leftarrow } \boldsymbol {c}]$ .

It is straightforward to iterate Rules 3.1(x) and 3.1(ix) in order to confirm that the above translation is harmless when it comes to provability.

Fact 6.1. Let $\Phi $ be a finite set of $\mathsf {QRC_1}$ formulas such that $\varphi , \psi \in \Phi $ . Then

$$ \begin{align*} \varphi \vdash \psi \iff \varphi^\Phi \vdash \psi^\Phi. \end{align*} $$

The semantics presented here for $\mathsf {QRC_1}$ is a generalization of more traditional semantics for quantified modal logics, since it uses maps between domains. However, for this section, we will only need constant domain models. These can be easily interpreted in the usual framework as described in [Reference Hughes and Cresswell25]. The following is a well-known (and easily provable) result.

Theorem 6.2 (Soundness for $\mathsf {QS}$ [Reference Hughes and Cresswell25]).

If $\mathcal {F} = \langle W, R \rangle $ satisfies every theorem of  $\mathsf {S}$ , then any constant domain model based on $\mathcal {F}$ satisfies every theorem of $\mathsf {QS}$ .

Thus any transitive constant domain model is a $\mathsf {QK}4$ model, and any transitive and conversely well-founded constant domain model is a $\mathsf {QGL}$ model.Footnote 7

We are now ready to show the main theorem of this section.

Theorem 6.3. Let $\varphi $ and $\psi $ be $\mathsf {QRC_1}$ formulas and let $\mathsf {QS}$ be any logic between $\mathsf {QK}4$ and $\mathsf {QGL}$ . Then $\varphi \vdash \psi $ if and only if there is a finite set $\Phi $ such that $\varphi , \psi \in \Phi $ and $\mathsf {QS} \vdash \varphi ^\Phi \to \psi ^\Phi $ .

Proof. For the left-to-right implication, let $\Phi $ be the set of formulas appearing in the proof of $\varphi \vdash \psi $ . It is easy to see by induction on the length of the $\mathsf {QRC_1}$ proof that $\mathsf {QK}4 \vdash \varphi ^\Phi \to \psi ^\Phi $ , which then implies that $\mathsf {QS} \vdash \varphi ^\Phi \to \psi ^\Phi $ .

For the right-to-left implication, we prove the contrapositive. If $\varphi \not \vdash \psi $ , then by Fact 6.1 we know that $\varphi ^\Phi \not \vdash \psi ^\Phi $ for any $\Phi $ including both $\varphi $ and $\psi $ . Let $\mathcal {M}$ be a $\mathsf {QRC_1}$ model satisfying $\varphi ^\Phi $ and not satisfying $\psi ^\Phi $ , as given by Theorem 5.11. This model is transitive, irreflexive, and has finitely-many worlds, which means it is a $\mathsf {QGL}$ model by Theorem 6.2. We conclude that $\mathsf {QGL} \not \vdash \varphi ^\Phi \to \psi ^\Phi $ , and consequently that $\mathsf {QS} \not \vdash \varphi ^\Phi \to \psi ^\Phi $ .⊣

We end with the following observation. The proof of Theorem 6.3 holds for any quantified modal logic that extends $\mathsf {QK}4$ and is sound for any combination of constant-domain, finite, transitive, and irreflexive models. The Barcan Formula ( $\mathsf {BF}$ ), $\forall x \, \Box A \to \Box \forall x \, A$ , is popular in the world of quantified modal logic and is always sound in constant-domain models. Thus the proof of Theorem 6.3 also serves to see that, for example, $\varphi \vdash \psi $ if and only if $\mathsf {QK}4 + \mathsf {BF} \vdash \varphi ^\Phi \to \psi ^\Phi $ . This is curious because $\mathsf {BF}$ does not hold in $\mathsf {QGL}$ and is in fact unsound with the usual provability interpretation of $\Box $ . From this we conclude that it would be worthwhile to study $\mathsf {QRC_1}$ from points of view unrelated to provability.

7 Arithmetical semantics

Recall that the language of arithmetic is that of first-order logic together with the symbols $\{ 0, 1, +, \times , \leq , = \}$ with their usual arities (see [Reference Hájek and Pudlák23] for details). We will rely on $\Sigma ^0_1$ collection throughout the sections on arithmetic, so we take ${\mathrm {I}\Sigma _{1}}$ as our base theory.

Let T be an elementary presented extension of ${\mathrm {I}\Sigma _{1}}$ , meaning that there is a bounded formula ${\mathrm {Ax}}_T(u)$ that is true in the standard model if and only if u is (the Gödel number of) an axiom of T. In this setting it is traditional to define Gödel’s provability predicate $\Box _T \varphi $ as $\exists p \, {\mathrm {Prf}}_T(p, \ulcorner \varphi \urcorner )$ , where

$$ \begin{align*} {\mathrm{Prf}}_T(p, n) &:= \text{sequence}(p) \land p_{|p|-1} = n \land {}\\ & \forall k < |p| \, ( {\mathrm{Ax}}_T(p_k) \lor \exists i, j < k \, \text{MP}(p_i, p_j, p_k) \lor \exists i < k \, \text{Gen}(p_i, p_k) ). \end{align*} $$

Roughly, $\Box _T \varphi $ formalizes that there is a Hilbert-style proof of $\varphi $ , that is, a finite sequence $p_0, \ldots , p_{m}$ such that $p_{m}$ is $\varphi $ and that each $p_k$ is either (the Gödel number of) an axiom of T or follows from previous elements of the sequence through either modus ponens or generalization.

Note that all the functions used in the definition of ${\mathrm {Prf}}_T$ can be naturally defined as bounded formulas in ${\mathrm {I}\Sigma _{1}}$ , and thus ${\mathrm {Prf}}_T$ is itself a bounded formula. This means that $\Box _T \varphi $ is $\Sigma ^0_1$ . We write $\Diamond _T \varphi $ as shorthand for $\neg \Box _T \neg \varphi $ .

In what follows we will be interested in c.e. theories, which are theories whose axioms can be defined by a $\Sigma ^0_1$ formula. It is known by Craig’s Trick [Reference Craig15, Reference Feferman18] that any such theory has an equivalent elementary presentation, allowing us to use the regular definition of $\Box _T$ . However, we will work with specific axiomatizations and thus it is sometimes more convenient to allow ${\mathrm {Prf}}$ to be a $\Sigma ^0_1$ formula. For a given $\Sigma ^0_1$ axiomatization $\tau $ of T, we define $\Box _\tau \varphi := \exists p \, {\mathrm {Prf}}_\tau (p, \ulcorner \varphi \urcorner )$ , where

$$ \begin{align*} {\mathrm{Prf}}_\tau(p, n) := &\text{sequence}(p) \land p_{|p|-1} = n \land \\ & \forall k < |p| \, ( \tau(p_k) \lor \exists i, j < k \, \text{MP}(p_i, p_j, p_k) \lor \exists i < k \, \text{Gen}(p_i, p_k) ) , \end{align*} $$

the only difference from ${\mathrm {Prf}}_T$ being that we use the $\Sigma ^0_1$ formula $\tau $ instead of the bounded ${\mathrm {Ax}}_T$ . Thus ${\mathrm {Prf}}_\tau $ is equivalent to a $\Sigma ^0_1$ formula (provably in ${\mathrm {I}\Sigma _{1}}$ ), and so $\Box _\tau \varphi $ is still $\Sigma ^0_1$ . We define $\Diamond _\tau \varphi $ as $\neg \Box _\tau \neg \varphi $ , as usual.

We wish to interpret the strictly positive formulas in the language of $\mathsf {QRC_1}$ as parameterized axiomatizations of arithmetical theories extending ${\mathrm {I}\Sigma _{1}}$ . Let $\tau $ be a bounded axiomatization of a sound c.e. base theory T extending ${\mathrm {I}\Sigma _{1}}$ .

A realization $\cdot ^{\circ }$ interprets each n-ary relation symbol $S(\boldsymbol {x}, \boldsymbol {c})$ as an $(n+1)$ -ary $\Sigma ^0_1$ formula $\sigma (u, \boldsymbol {y}, \boldsymbol {z})$ in the language of arithmetic, where $\boldsymbol {y}$ matches with $\boldsymbol {x}$ and $\boldsymbol {z}$ matches with $\boldsymbol {c}$ .Footnote 8 We then extend this notion to any formula as follows (we add $\tau (u)$ to the axioms of the interpretation of any relation symbol to guarantee that every theory is an extension of T).

Definition 7.1 ( $\cdot ^{\circ _{\tau }}$ [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

Let $\cdot ^{\circ }$ be a realization such that, for a given n-ary predicate S and terms $\boldsymbol {t}$ , $S(\boldsymbol {t})^{\circ }$ is an $(n+1)$ -ary $\Sigma ^0_1$ arithmetical formula where modal variables $x_k$ are interpreted as $y_k$ and modal constants $c_k$ are interpreted as $z_k$ . We extend this realization to $\mathsf {QRC_1}$ formulas as follows:

  • $\top ^{\circ _{\tau }} := \tau (u)$ ;

  • $S(\boldsymbol {x}, \boldsymbol {c})^{\circ _{\tau }} : = S(\boldsymbol {x},\boldsymbol {c})^{\circ } \lor \tau (u)$ ;

  • $(\psi \land \delta )^{\circ _{\tau }} := \psi ^{\circ _{\tau }} \lor \delta ^{\circ _{\tau }}$ ;

  • $(\lozenge \psi )^{\circ _{\tau }} := \tau (u) \lor (u = \ulcorner \Diamond _{\psi ^{\circ _{\tau }}} \top \urcorner )$ ;

  • $(\forall x_i \, \psi )^{\circ _{\tau }} := \exists y_i \, \psi ^{\circ _{\tau }}$ .

We remark that, if the free variables of $\varphi $ are $x_0, \ldots , x_{n-1}$ and the constants appearing in $\varphi $ are $c_0, \ldots , c_{m-1}$ , then the free variables of $\varphi ^{\circ _{\tau }}$ are u, $y_0, \ldots , y_{n-1}$ , $z_0, \ldots , z_{m-1}$ . For example, if $S(x_0, c_1)^{\circ _{\tau }} = \sigma (u, y_0, z_1) \lor \tau (u)$ , then

$$ \begin{align*} (\Diamond S(x_0, c_1))^{\circ_{\tau}} = \tau(u) \lor (u = \ulcorner \Diamond_{\sigma(u, \dot{y_0}, \dot{z_1}) \lor \tau(u)} \top \urcorner). \end{align*} $$

The dotted variables $\dot {y_0}$ and $\dot {z_1}$ in the expression $u = \ulcorner \Diamond _{\sigma (u, \dot {y_0}, \dot {z_1}) \lor \tau (u)} \top \urcorner $ indicate that $y_0$ and $z_1$ are free variables of this expression that, upon being instantiated by natural numbers n and m, shall be replaced by the numerals $\bar {n}$ and $\bar {m}$ instead.

Since we have $\Sigma ^0_1$ collection, $\varphi ^{\circ _{\tau }}$ is always provably equivalent to a $\Sigma ^0_1$ formula, and it represents the axiomatization of an extension of T.

We briefly inspect our intuitions regarding this arithmetic interpretation and the interaction between $\forall $ and $\Diamond $ . As mentioned in Lemma 3.2, the sequent $\Diamond \forall x \, \varphi \vdash \forall x \, \Diamond \varphi $ is provable in $\mathsf {QRC_1}$ . In the arithmetical reading it says that if the union of many theories (the $\forall x \, \varphi $ part) is consistent, then so is each individual theory. The converse, on the other hand, must not be a consequence of $\mathsf {QRC_1}$ , for it is not in general the case that the union of many consistent theories is itself consistent.

We can now define $\mathcal {QRC}_1(T)$ :

$$ \begin{align*} \mathcal{QRC}_1(T) := \{\varphi(\boldsymbol{x}, \boldsymbol{c}) \vdash \psi(\boldsymbol{x}, \boldsymbol{c}) \mid \forall \cdot^{\circ} \, T \vdash \forall \theta \, \forall \boldsymbol{y}, \boldsymbol{z} \, (\Box_{\psi^{\circ_{\tau}}} \theta \to \Box_{\varphi^{\circ_{\tau}}} \theta) \} , \end{align*} $$

where $\theta $ is a closed formula and $\varphi ^{\circ _{\tau }}, \psi ^{\circ _{\tau }}$ in general depend on $\boldsymbol {y}$ and $\boldsymbol {z}$ .

We will show that

$$ \begin{align*} \mathsf{QRC_1} = \mathcal{QRC}_1(T) \end{align*} $$

for any sound c.e theory T extending ${\mathrm {I}\Sigma _{1}}$ . The left-to-right inclusion was already proved in [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] for ${\mathrm {I}\Sigma _{1}}$ , and the same proof goes through for any such T. The most notable part of the proof is the case of Rule 3.1(vii) ( $\forall $ introduction on the right), which uses $\Sigma ^0_1$ collection. We go into detail on the very similar proof of Theorem 10.3.

Theorem 7.2 (Arithmetical soundness [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1]).

$\mathsf {QRC_1} \subseteq \mathcal {QRC}_1(T)$ .

We dedicate the next section to the proof of the other inclusion.

8 Arithmetical completeness

The proof of arithmetical completeness closely follows the proof of Solovay’s Theorem found in [Reference Boolos11]. The arithmetical semantics used there is different from the one presented in the previous section, allowing only for finite extensions of the base theory instead of arbitrary c.e. ones. This is not a problem because finite axiomatizations are enough to show completeness. We use $\cdot ^{\circ }$ in the general setting and $\cdot ^{*}$ in the finite one. The base theory T considered here is any sound c.e. extension of ${\mathrm {I}\Sigma _{1}}$ . Note that we use the bounded ${\mathrm {Prf}}_T$ formula in the definition of $\cdot ^{*_{T}}$ (depending on an elementary axiomatization $\tau $ of T), as otherwise Solovay’s proof wouldn’t go through.

Definition 8.1 ( $\cdot ^{*_{T}}$ ).

Let $\cdot ^{*}$ be a realization such that, for a given predicate S and terms $\boldsymbol {t}$ , $S(\boldsymbol {t})^{*}$ is an arithmetical formula with the same arity as S where modal variables $x_k$ are interpreted as $y_k$ and modal constants $c_k$ are interpreted as $z_k$ . We extend this realization to $\mathsf {QRC_1}$ formulas as follows:

  • $\top ^{*_{T}} := \top $ ;

  • $S(\boldsymbol {t})^{*_{T}} := S(\boldsymbol {t})^{*}$ ;

  • $(\varphi \land \psi )^{*_{T}} := \varphi ^{*_{T}} \land \psi ^{*_{T}}$ ;

  • $(\Diamond \varphi )^{*_{T}} := \Diamond _T \varphi ^{*_{T}}$ ;

  • $(\forall x_k \, \varphi )^{*_{T}} := \forall y_k \, \varphi ^{*_{T}}$ .

The idea of Solovay’s proof is to embed a Kripke model not satisfying the desired unprovable formula in the language of arithmetic. If the embedding is done correctly, it is possible to prove that a formula is satisfied at a world of the Kripke model exactly when it is a consequence of the representation of that world in the desired theory T.

Given two strictly positive formulas $\varphi $ and $\psi $ such that $\varphi \not \vdash \psi $ in $\mathsf {QRC_1}$ , let $\mathcal {M}_{\varphi , \psi }$ be a finite, irreflexive, and constant domain adequate model satisfying $\varphi $ and not satisfying $\psi $ at the root $1$ under a $1$ -assignment $g_{\varphi , \psi }$ . This model and assignment exist by Theorem 5.11. Since $\mathcal {M}_{\varphi , \psi }$ has constant domain, we refer to $g_{\varphi , \psi }$ and any other i-assignments as just assignments, omitting the relevant world.

We assume that the worlds of $\mathcal {M}_{\varphi , \psi }$ are $W = \{ 1, 2, \ldots , N \}$ , where $1$ is the root. We define a new (adequate) model $\mathcal {M}$ that is a copy of $\mathcal {M}_{\varphi , \psi }$ , except that it has an extra world $0$ as the new root. This world $0$ is connected to all the other worlds through R and has the same domain, constant interpretation, and relational symbol interpretation as $1$ . The functions $\eta _{0, i}$ with $0 < i \leq N$ are all defined as the identity function.

Let $\lambda _i$ be the Solovay sentences for T as defined in [Reference Boolos11], satisfying the following Embedding Lemma.

Lemma 8.2 (Embedding [Reference Boolos11]).

  1. 1. $\mathbb {N} \vDash \lambda _0$ ;

  2. 2. $T \vdash \bigvee _{i \leq N} \lambda _i$ ;

  3. 3. $T \vdash \lambda _i \to \bigwedge _{j \leq N, j \neq i} \neg \lambda _j$ , for $i \leq N$ ;

  4. 4. $T \vdash \lambda _i \to \bigwedge _{j \leq N, iRj} \Diamond _T \lambda _j$ , for $i \leq N$ ;

  5. 5. $T \vdash \lambda _i \to \Box _T \bigvee _{j \leq N, iRj} \lambda _j$ , for $0 < i \leq N$ .

The domain of every world is M. Let m be the size of M, and be a bijection between M and the set of numerals $\{\overline {0}, \ldots , \overline {m - 1}\}$ .

We now define for a given n-ary predicate symbol S and terms $\boldsymbol {t} = t_0, \ldots , t_{n-1}$ the  $\cdot ^{\circledast }$ interpretation as follows:

Here $x_k$ is any $\mathsf {QRC_1}$ variable, $y_k$ is the corresponding T variable, $c_k$ is any $\mathsf {QRC_1}$ constant, $z_k$ is the corresponding T variable, and $\cdot ^{J_i}$ is the denotation of a predicate symbol at world i. Note that $x_k \in {\mathrm {fv}}(S(\boldsymbol {t}))$ if and only if $y_k \in {\mathrm {fv}}(S(\boldsymbol {t})^{\circledast })$ and $c_k$ appears in $S(\boldsymbol {t})$ if and only if $z_k \in {\mathrm {fv}}(S(\boldsymbol {t})^{\circledast })$ . The $\cdot ^{\circledast }$ interpretation is extended to non-predicate formulas as described in Definition 8.1.

Fact 8.3. For any $\mathsf {QRC_1}$ formula $\varphi $ , variable $x_k$ , and constant $c_k$ , we have that $x_k \in {\mathrm {fv}}(\varphi )$ if and only if $y_k \in {\mathrm {fv}}(\varphi ^{\circledast _{T}})$ and that $c_k$ appears in $\varphi $ if and only if $z_k \in {\mathrm {fv}}(\varphi ^{\circledast _{T}})$ .

We further note that $\varphi ^{\circledast _{T}}$ is invariant under replacing variables by themselves modulo m, provably in T.

Lemma 8.4. For any $\mathsf {QRC_1}$ formula $\varphi $ and any arithmetical variable y:

  1. 1. $ T \vdash \varphi ^{\circledast _{T}} \leftrightarrow \varphi ^{\circledast _{T}}[y {\leftarrow } y \ {\mathrm {mod}} \ m] $ ;

  2. 2. $ T \vdash \forall y \, \varphi ^{\circledast _{T}} \leftrightarrow \forall y < m \, \varphi ^{\circledast _{T}} $ .

Proof. The second item is a straightforward consequence of the first, which is proved by an external induction on the complexity of $\varphi $ , noting that T proves $(y \ {\mathrm {mod}} \ m) \ {\mathrm {mod}} \ m = y \ {\mathrm {mod}} \ m$ for any y.⊣

We now prove two versions of a Truth Lemma, one for when $\varphi $ is forced at a world i of $\mathcal {M}$ , and one for when it isn’t. We wish to show that T proves $\varphi ^{\circledast _{T}}$ (respectively $\neg \varphi ^{\circledast _{T}}$ ) as a consequence of $\lambda _i$ , as long as the free variables of $\varphi $ are interpreted in the same way in both settings.

In order to use concise notation, we shall say $\boldsymbol {y}$ instead of $y_0, \ldots , y_{n-1}$ , and similarly for other terms. We also abbreviate iterated substitutions, writing instead of and writing instead of .

Lemma 8.5 (Truth Lemma: positive).

Let $\varphi $ be a $\mathsf {QRC_1}$ formula with free variables $\boldsymbol {x} = x_0, \ldots , x_{n-1}$ and constants $\boldsymbol {c} = c_0, \ldots , c_{n'-1}$ . Then for any world $i \leq N$ and any assignment g,

Proof. By external induction on the complexity of $\varphi $ . The cases of $\top $ and $\land $ are straightforward.

In the case of relational symbols, we assume without loss of generality that the relevant formula is $S(x_0, c_0)$ . If $\mathcal {M}, i \Vdash ^g S(x_0, c_0)$ , then $\langle g(x_0), (c_0)^{I_i} \rangle \in S^{J_i}$ . Reason in T and assume $\lambda _i$ . It suffices to prove

which implies

under the assumption of $\lambda _i$ . We need to find $\langle a_0, a_1 \rangle \in S^{J_i}$ such that

and

. Noting that

is provably equal to

for any b, we pick $a_0 := g(x_0)$ and $a_1 := (c_0)^{I_0}$ . This concludes this step of the proof because $(c_0)^{I_0}$ is equal to $(c_0)^{I_i}$ , for any world i.

For $\forall x_0 \, \varphi $ , assume without loss of generality that the free variables of $\varphi $ are $x_0$ and $x_1$ . If $\mathcal {M}, i \Vdash ^g \forall x_0 \, \varphi $ then for every assignment $h \sim _{x_0} g$ we have $\mathcal {M}, i \Vdash ^h \varphi $ . We wish to show

Reason in T and assume $\lambda _i$ . By Lemma 8.4, it is enough to show

Since $y_0$ , $y_1$ , and $\boldsymbol {z}$ are all different, we can push the substitutions inside and prove

instead. Let $y_0 < m$ be arbitrary. Since

is a bijection, there is $a \in M$ such that

. We define an assignment h such that $h \sim _{x_0} g$ and $h(x_0) := a$ . By assumption, $\mathcal {M}, i \Vdash ^h \varphi $ , so by the induction hypothesis we obtain

This concludes the argument because

and $h(x_1) = g(x_1)$ .

Finally, consider the case of $\Diamond \varphi $ . If $\mathcal {M}, i \Vdash ^g \Diamond \varphi $ , then there is $j \leq N$ such that $iRj$ and $\mathcal {M}, j \Vdash ^g \varphi $ . Reason in T and assume $\lambda _i$ . By Lemma 8.2.4, we obtain $\Diamond _T \lambda _j$ . Then the induction hypothesis under the box gives us the desired .⊣

Lemma 8.6 (Truth Lemma: negative).

Let $\varphi $ be a $\mathsf {QRC_1}$ formula with free variables $\boldsymbol {x} = x_0, \ldots , x_{n-1}$ and constants $\boldsymbol {c} = c_0, \ldots , c_{n'-1}$ . Then for any world $0 < i \leq N$ and any assignment g,

Proof. By external induction on the complexity of $\varphi $ . The cases of $\top $ and $\land $ are straightforward.

For the relational symbols, consider $S(x_0, c_0)$ without loss of generality. If $\mathcal {M}, i \not \Vdash ^g S(x_0, c_0)$ , then $\langle g(x_0), (c_0)^{I_i} \rangle \notin S^{J_i}$ . Reason in T and assume $\lambda _i$ . We obtain $\neg \lambda _j$ for every $j \neq i$ by Lemma 8.2.3, and hence need only show . In other words, we need to check that if $\langle a_0, a_1 \rangle \in S^{J_i}$ , either , or . This follows from our observation that $\langle g(x_0), (c_0)^{I_i} \rangle \notin S^{J_i}$ , taking into account that is equal to for any b, that is injective, and that $(c_0)^{I_0} = (c_0)^{I_i}$ .

Consider now the case of $\forall x_0 \, \varphi $ . We assume without loss of generality that ${\mathrm {fv}}(\varphi ) = \{x_0, x_1\}$ . If $\mathcal {M}, i \not \Vdash ^g \forall x_0 \, \varphi $ , then there is an assignment $h \sim _{x_0} g$ such that $\mathcal {M}, i \not \Vdash ^h \varphi $ . Reason in T and assume $\lambda _i$ . From the induction hypothesis we obtain , which then implies . This is what we wanted, taking into account that $h(x_1) = g(x_1)$ .

Finally, in the case of $\Diamond \varphi $ , assume that $\mathcal {M}, i \not \Vdash ^g \Diamond \varphi $ . Then for every j such that $iRj$ , we have $\mathcal {M}, j \not \Vdash ^g \varphi $ and thus for each such j the induction hypothesis gives us , which put together implies . Reason in T and assume $\lambda _i$ . By Lemma 8.2.5 and our assumption, we obtain $\Box _T \bigvee _{iRj} \lambda _j$ . Taking the previous observation under the box, we conclude , which is precisely .⊣

We are ready to prove something analogous to Solovay’s Theorem, which is the precursor to our desired completeness theorem.

Theorem 8.7. If $\varphi , \psi $ are $\mathsf {QRC_1}$ formulas with free variables $\boldsymbol {x}$ and constants $\boldsymbol {c}$ such that $\varphi \not \vdash \psi $ , we have .

Proof. Recall that $\mathcal {M}$ satisfies $\varphi $ and not $\psi $ at world $1$ under the assignment $g := g_{\varphi , \psi }$ .

Since $\mathcal {M}, 1 \Vdash ^g \varphi $ , we obtain from Lemma 8.5. Since $\mathcal {M}, 1 \not \Vdash ^g \psi $ , we obtain from Lemma 8.6. Thus .

By Lemma 8.2.4 and the fact that $0R1$ , we obtain $T \vdash \lambda _0 \to \Diamond _T \lambda _1$ , so putting this together with the previous observation under the box, .

By Lemma 8.2.1, we know that $\mathbb {N} \vDash \lambda _0$ , and thus by the soundness of T, we know that . Then it must be the case that .⊣

We now define an arithmetical realization $\cdot ^{\circledcirc }$ in the style of Section 7 that behaves like $\cdot ^{\circledast }$ . Recall that if $\tau $ is a $\Sigma ^0_1$ formula axiomatizing T, the realization $\cdot ^{\circ _{\tau }}$ sends $\mathsf {QRC_1}$ formulas to $\Sigma ^0_1$ axiomatizations of theories extending T. So in particular we always have $(\varphi \land \psi )^{\circ _{\tau }} = \varphi ^{\circ _{\tau }} \lor \psi ^{\circ _{\tau }}$ , because $\chi $ is an axiom of the union of the theories axiomatized by $\varphi ^{\circ _{\tau }}$ and $\psi ^{\circ _{\tau }}$ precisely when $\chi $ is an axiom of one of them. This is different from $\cdot ^{*_{T}}$ realizations such as $\cdot ^{\circledast _{T}}$ , which send $\mathsf {QRC_1}$ formulas to generic arithmetical formulas. We then interpret these formulas as finite extensions of T, so in particular $(\varphi \land \psi )^{*_{T}} = \varphi ^{*_{T}} \land \psi ^{*_{T}}$ , because the union of $T + \varphi ^{*_{T}}$ and $T + \psi ^{*_{T}}$ is the same as $T + \varphi ^{*_{T}} \land \psi ^{*_{T}}$ .

We define $\cdot ^{\circledcirc }$ such that $S(\boldsymbol {t})^{\circledcirc } := \tau (u) \lor (u = \ulcorner S(\boldsymbol {t})^{\circledast } \urcorner )$ , and extend it to non-atomic formulas as described in Definition 7.1. Note that $x_k \in {\mathrm {fv}}(\varphi )$ if and only if $y_k \in {\mathrm {fv}}(\varphi ^{\circledcirc _{\tau }})$ and $c_k$ appears in $\varphi $ if and only if $z_k \in {\mathrm {fv}}(\varphi ^{\circledcirc _{\tau }})$ .

Lemma 8.8. For any $\mathsf {QRC_1}$ formula $\varphi $ with free variables $\boldsymbol {x}$ and constants $\boldsymbol {c}$ ,

$$ \begin{align*} T \vdash \forall \theta \, \forall \boldsymbol{y}, \boldsymbol{z} \, ( \Box_{\varphi^{\circledcirc_{\tau}}} \theta \leftrightarrow \Box_T (\varphi^{\circledast_{T}} \to \theta) ) , \end{align*} $$

where $\theta $ is (the Gödel number of) a closed formula.

Proof. By external induction on $\varphi $ . There is nothing to show for $\top $ , because since $\tau $ is a bounded formula, $\Box _{\tau }$ is the same as $\Box _T$ .

The case of relational symbols is a straightforward consequence of the formalized Deduction Theorem [Reference Feferman18].

In the case of $\land $ , we take $\varphi = \psi \land \delta $ , and we omit the variables $\boldsymbol {y}$ and $\boldsymbol {z}$ , as they introduce visual clutter but don’t make the proof any more complex.

( $\to $ ) Reason in T and fix an arbitrary $\theta $ , assuming $\Box _{\psi ^{\circledcirc _{\tau }} \lor \delta ^{\circledcirc _{\tau }}} \theta $ . Then there is a finite sequence $\boldsymbol {\pi } = \pi _0, \ldots ,\pi _n$ with $\pi _n = \theta $ that is a proof of $\theta $ in the theory axiomatized by $\psi ^{\circledcirc _{\tau }} \lor \delta ^{\circledcirc _{\tau }}$ . Each formula $\chi _i$ occurring in $\boldsymbol {\pi }$ that is not a consequence of previous formulas in the sequence through a rule satisfies either $\psi ^{\circledcirc _{\tau }}$ or $\delta ^{\circledcirc _{\tau }}$ . Then we have in particular that either $\Box _{\psi ^{\circledcirc _{\tau }}} \chi _i$ or $\Box _{\delta ^{\circledcirc _{\tau }}} \chi _i$ for each such $\chi _i$ , whence by the induction hypothesis either $\Box _T(\psi ^{\circledast _{T}} \to \chi _i)$ or $\Box _T(\delta ^{\circledast _{T}} \to \chi _i)$ holds. In both cases we have $\Box _T(\psi ^{\circledast _{T}} \wedge \delta ^{\circledast _{T}} \to \chi _i)$ , and thus the proof of $\theta $ can be repeated in T under the assumption of $(\psi \land \delta )^{\circledast _{T}}$ , as desired.

$(\leftarrow )$ Fix $\theta $ , $\boldsymbol {y}$ , and $\boldsymbol {z}$ . By the induction hypothesis (taking $\theta $ to be $\psi ^{\circledast _{T}}$ ) we see that $\Box _{\psi ^{\circledcirc _{\tau }}}\psi ^{\circledast _{T}}$ and likewise $\Box _{\delta ^{\circledcirc _{\tau }}} \delta ^{\circledast _{T}}$ . Thus $\Box _{\psi ^{\circledcirc _{\tau }} \lor \delta ^{\circledcirc _{\tau }}} (\psi ^{\circledast _{T}} \land \delta ^{\circledast _{T}})$ . By assumption we have $\Box _T(\psi ^{\circledast _{T}} \land \delta ^{\circledast _{T}} \to \theta )$ , and since $\varphi ^{\circledcirc _{\tau }}$ extends T, we may conclude $\Box _{\varphi ^{\circledcirc _{\tau }}} \theta $ as desired.

The $\forall $ case follows the same idea as the $\land $ case. Consider $\varphi = \forall x_0 \, \psi $ , with ${\mathrm {fv}}(\psi ) = \{x_0, x_1\}$ without loss of generality. Note that $x_0$ is represented by $y_0$ in T, and this is always a different variable from any z used to represent $\mathsf {QRC_1}$ constants. As there is no further complication with constants, we omit them. Let $\theta $ and l be arbitrary and reason in T.

$(\to )$ If $\Box _{\exists y_0 \, \psi ^{\circledcirc _{\tau }}[y_1 {\leftarrow } l]} \theta $ , then there is a proof $\boldsymbol {\pi } = \pi _0, \ldots , \pi _n$ where $\pi _n = \theta $ and each axiom $\chi _i$ in $\boldsymbol {\pi }$ satisfies $\psi ^{\circledcirc _{\tau }}[y_1 {\leftarrow } l][y_0 {\leftarrow } k_i]$ for some number $k_i$ , and consequently $\Box _T (\psi ^{\circledast _{T}}[y_1 {\leftarrow } l][y_0 {\leftarrow } k_i] \to \chi _i)$ by the induction hypothesis for each i. Then by weakening we conclude $\Box _T (\forall y_0 \, \psi ^{\circledast _{T}}[y_1 {\leftarrow } l] \to \chi _i)$ for each i, and we are done.

$(\leftarrow )$ Assume $\Box _T (\forall y_0 \, \psi ^{\circledast _{T}}[y_1 {\leftarrow } l] \to \theta )$ . By Lemma 8.4 under the box, we obtain $\Box _T (\forall y_0 < m \, \psi ^{\circledast _{T}}[y_1 {\leftarrow } l] \to \theta )$ , where m is the size of the domain of $\mathcal {M}$ . Using the induction hypothesis for each $k < m$ with $\theta := \psi ^{\circledast _{T}}[y_1 {\leftarrow } l][y_0 {\leftarrow } k]$ , $y_0 := k$ , and $y_1 := l$ , we get $\forall k < m \, \Box _{\psi ^{\circledcirc _{\tau }}[y_1 {\leftarrow } l][y_0 {\leftarrow } k]} \psi ^{\circledast _{T}}[y_1 {\leftarrow } l][y_0 {\leftarrow } k]$ , and in particular $\forall k < m \, \Box _{\exists y_0 \, \psi ^{\circledcirc _{\tau }}[y_1 {\leftarrow } l]} \psi ^{\circledast _{T}}[y_1 {\leftarrow } l][y_0 {\leftarrow } k]$ . Then by $\Sigma ^0_1$ collection we can change the order of the quantifier and the box, concluding $\Box _{\exists y_0 \, \psi ^{\circledcirc _{\tau }}[y_1 {\leftarrow } l]} \forall y_0 < m \, \psi ^{\circledast _{T}}[y_1 {\leftarrow } l]$ , which is enough to conclude this part of the proof.

Finally, for the case of $\varphi = \Diamond \psi $ , we start by observing that applying the induction hypothesis to $\bot $ yields $ T \vdash \forall \boldsymbol {y}, \boldsymbol {z} \, ( \Diamond _{\varphi ^{\circledcirc _{\tau }}} \top \leftrightarrow \Diamond _T \varphi ^{\circledast _{T}} ) $ . Note that $(\Diamond \psi )^{\circledcirc _{\tau }} = \tau (u) \lor (u = \ulcorner \Diamond _{\psi ^{\circledcirc _{\tau }}} \top \urcorner )$ , and thus $\Box _{(\Diamond \psi )^{\circledcirc _{\tau }}} \theta $ is equivalent to $\Box _T (\Diamond _{\psi ^{\circledcirc _{\tau }}} \top \to \theta )$ by the formalized Deduction Theorem. The previous observation under the box then suffices to finish the proof.⊣

We are finally ready to prove arithmetical completeness for any sound c.e. theory extending ${\mathrm {I}\Sigma _{1}}$ .

Theorem 8.9 (Arithmetical completeness).

$\mathsf {QRC_1} \supseteq \mathcal {QRC}_1(T)$ .

Proof. Recall the definition of $\mathcal {QRC}_1(T)$ :

$$ \begin{align*} \mathcal{QRC}_1(T) = \{\varphi(\boldsymbol{x}, \boldsymbol{c}) \vdash \psi(\boldsymbol{x}, \boldsymbol{c}) \mid \forall \cdot^{\circ} \, T \vdash \forall \theta \, \forall \boldsymbol{y}, \boldsymbol{z} \, (\Box_{\psi^{\circ_{\tau}}} \theta \to \Box_{\varphi^{\circ_{\tau}}} \theta) \} , \end{align*} $$

where $\theta $ is closed.

We show that if $\varphi \not \vdash \psi $ , then the realization $\cdot ^{\circledcirc }$ defined above is such that

$$ \begin{align*} T \not\vdash \forall \theta \, \forall \boldsymbol{y}, \boldsymbol{z} \, (\Box_{\psi^{\circledcirc_{\tau}}} \theta \to \Box_{\varphi^{\circledcirc_{\tau}}} \theta). \end{align*} $$

Suppose towards a contradiction that T does prove this formula. Then by Lemma 8.8,

$$ \begin{align*} T \vdash \forall \theta \, \forall \boldsymbol{y}, \boldsymbol{z} \, (\Box_T (\psi^{\circledast_{T}} \to \theta) \to \Box_T (\varphi^{\circledast_{T}} \to \theta)). \end{align*} $$

Taking

,

, and

, we conclude

This together with the soundness of T contradicts Theorem 8.7.⊣

We have seen that $\mathsf {QRC_1} = \mathcal {QRC}_1(T)$ for any sound c.e. theory T extending ${\mathrm {I}\Sigma _{1}}$ . Thus $\mathcal {QRC}_1(T)$ is constant over a large class of theories, and for these theories it does not depend on the specific axiomatization $\tau $ chosen. This is similar to the propositional case, but simpler than the full predicate case, where $\mathsf {QPL}(T)$ is known to depend on both T (as shown by Montagna [Reference Montagna32]) and $\tau $ (as shown by Artemov [Reference Artemov3]; see also [Reference Kurahashi28, Reference Kurahashi29]).

9 A decidable fragment of $\mathsf {QPL}(\mathsf {PA})$

As we mentioned before, Vardanyan’s results are very robust and $\Pi ^0_2$ completeness can already be obtained for the language with just one unary predicate symbol and no nested occurrences of $\Box $ , as shown in [Reference Vardanyan and Adyan37]. However, Artemov and Japaridze [Reference Artemov and Japaridze4] managed to carve out a non-trivial decidable fragment of $\mathsf {QPL}(\mathsf {PA})$ : all formulas that are decidable on finite Kripke frames correspond directly to a fragment of $\mathsf {QPL}(\mathsf {PA})$ . They further observed that as a corollary one can conclude the decidability of the one variable fragment of $\mathsf {QPL}(\mathsf {PA})$ .

The above may seem contradictory with Vardanyan’s result on $\Pi ^0_2$ completeness of the fragment of $\mathsf {QPL}(\mathsf {PA})$ with just one unary predicate symbol. However, although it is easy to see that in predicate logic any sentence with just one unary predicate is equivalent to one in the one-variable fragment, this does not hold when modalities are involved, as exhibited by the formula $\forall x \, \forall y \, \Box (P(x) \lor \neg P(y))$ .

In this section we shall see that $\mathsf {QRC_1}$ gives rise to a new decidable fragment of $\mathsf {QPL}(\mathsf {PA})$ . Let $\mathcal {L}_{\Box , \forall }$ be the language of full quantified modal logic, based on the same signatures as the language of $\mathsf {QRC_1}$ and extending the latter with $\to $ , $\bot $ , and the usual abbreviations. We extend the notion of finite arithmetical realization described in Definition 8.1 to $\mathcal {L}_{\Box , \forall }$ as follows:

  • $\bot ^{*_{\mathsf {PA}}} := \bot $ ;

  • $(A \to B)^{*_{\mathsf {PA}}} := A^{*_{\mathsf {PA}}} \to B^{*_{\mathsf {PA}}}$ .

We now define the quantified provability logic of $\mathsf {PA}$ as the set of always provable $\mathcal {L}_{\Box , \forall }$ formulas:

$$ \begin{align*} \mathsf{QPL}(\mathsf{PA}) := \{ A(\boldsymbol{x}, \boldsymbol{c}) \in \mathcal{L}_{\Box, \forall} \mid \text{for any }\cdot^{*}, \text{ we have } \mathsf{PA} \vdash \forall \boldsymbol{y}, \boldsymbol{z} \, A^{*_{\mathsf{PA}}} \} , \end{align*} $$

where $\boldsymbol {y}$ are the arithmetical counterpart of $\boldsymbol {x}$ and $\boldsymbol {z}$ the arithmetical counterpart of $\boldsymbol {c}$ , as before.

Theorem 9.1. Let $\varphi $ and $\psi $ be $\mathsf {QRC_1}$ formulas. Then

$$ \begin{align*} \varphi \vdash \psi \iff \varphi \to \psi \in \mathsf{QPL}(\mathsf{PA}). \end{align*} $$

Proof. The left-to-right implication is proved by induction on $\varphi \vdash \psi $ . We note that the formalized converse Barcan formula is used in the case of the necessitation rule and provable $\Sigma ^0_1$ completeness is used in the case of the transitivity axiom (see [Reference Boolos11] for details on these results).

For the right-to-left implication, we take the contrapositive. Thus, let $\varphi $ and $\psi $ be strictly positive formulas with free variables $\boldsymbol {x}$ and constants $\boldsymbol {c}$ such that $\varphi \not \vdash \psi $ . By Theorem 8.7, we have

and thus $\varphi \to \psi \notin \mathsf {QPL}(\mathsf {PA})$ .⊣

10 Heyting Arithmetic

We end this paper with a foray into intuitionistic arithmetic. We show that $\mathsf {QRC_1}$ is sound with respect to Heyting Arithmetic ( $\mathsf {HA}$ ) and conjecture that it is complete as well.

Let $\eta (u)$ be a $\Sigma ^0_1$ formula naturally axiomatizing $\mathsf {HA}$ .Footnote 9 The $\mathsf {HA}$ -provability of $\varphi $ can thus be expressed by $\Box _\eta \varphi $ , as defined in Section 7.

We observe that a number of standard results in the realm of classical provability logic still hold in the intuitionistic case.

Lemma 10.1 (Derivability conditions [Reference Visser and Zoethout39]).

Let $\tau $ be a $\Sigma ^0_1$ axiomatization of an arithmetical theory T extending $\mathsf {HA}$ , and $\varphi , \psi $ be formulas in the language of arithmetic. Then:

  1. 1. if $T \vdash \varphi $ then $\mathsf {HA} \vdash \Box _\tau \varphi $ ;

  2. 2. $\mathsf {HA} \vdash \Box _\tau (\varphi \to \psi ) \to (\Box _\tau \varphi \to \Box _\tau \psi )$ ;

  3. 3. $\mathsf {HA} \vdash \Box _\tau \varphi \to \Box _\eta \Box _\tau \varphi $ .

Lemma 10.2 (Collection [Reference Fujiwara and Kurahashi21, Proposition 5.13]).

Let $\varphi $ be an arithmetical formula without x as a free variable. Then

$$ \begin{align*} \mathsf{HA} \vdash \forall y < x \, \exists z \, \varphi \to \exists w \, \forall y < x \, \exists z < w \, \varphi. \end{align*} $$

Note that, since $\mathsf {HA}$ proves full collection, ${\mathrm {Prf}}_\eta $ is equivalent to a $\Sigma ^0_1$ formula, provably in $\mathsf {HA}$ .

As in the classical case, we define $\Diamond _\tau \varphi $ as $\neg \Box _\tau \neg \varphi $ when $\tau $ is an axiomatization of an extension of $\mathsf {HA}$ .

We extend a generic realization $\cdot ^{\circ }$ to non-predicate formulas as in the classical case (cf. Definition 7.1). Note that $\varphi ^{\circ _{\eta }}$ is equivalent to a $\Sigma ^0_1$ formula and extends $\mathsf {HA}$ , both of these provably in $\mathsf {HA}$ .

Theorem 10.3 (Arithmetical soundness w.r.t. $\mathsf {HA}$ ).

$$ \begin{align*} \mathsf{QRC_1} \subseteq \{\varphi \vdash \psi \mid \forall \cdot^{\circ} \, \mathsf{HA} \vdash \forall \theta, \boldsymbol{y}, \boldsymbol{z} \, (\Box_{\psi^{\circ_{\eta}}} \theta \to \Box_{\varphi^{\circ_{\eta}}} \theta) \}. \end{align*} $$

Proof. Let $\varphi $ and $\psi $ be formulas such that $\varphi \vdash \psi $ . The proof is similar to the classical case, and proceeds by external induction on $\varphi \vdash \psi $ .

The soundness of $\varphi \vdash \top $ states that $\varphi ^{\circ _{\eta }}$ extends $\mathsf {HA}$ , which can readily be checked by induction on $\varphi $ . The soundness of $\varphi \vdash \varphi $ and of the cut rule are straightforward, and the soundness of $\varphi \land \psi \vdash \varphi $ is a simple weakening.

We proceed with the soundness of the conjunction introduction rule, that if $\varphi \vdash \psi $ and $\varphi \vdash \chi $ then $\varphi \vdash \psi \land \chi $ . Reason in $\mathsf {HA}$ and let $\theta $ be a closed formula, and $\boldsymbol {y}$ and $\boldsymbol {z}$ be arbitrary. Assume $\Box _{\psi ^{\circ _{\eta }} \lor \chi ^{\circ _{\eta }}} \theta $ , taking $\boldsymbol {\pi } = \pi _0, \ldots , \pi _{n-1}$ as a proof of this fact. Some of the $\pi _i$ are axioms of $\psi ^{\circ _{\eta }}$ , some are axioms of $\chi ^{\circ _{\eta }}$ , and some follow from previous steps in the proof through a rule. Let $\{\delta _i\}_{i \in I}$ be the finite set of $\psi ^{\circ _{\eta }}$ axioms appearing in $\boldsymbol {\pi }$ . Thus $\Box _{\psi ^{\circ _{\eta }}} \bigwedge _{i \in I} \delta _i$ and by the induction hypothesis for $\varphi \vdash \psi $ we obtain $\Box _{\varphi ^{\circ _{\eta }}} \bigwedge _{i \in I} \delta _i$ . On the other hand, we know $\Box _{\chi ^{\circ _{\eta }}} (\bigwedge _{i \in I} \delta _i \to \theta )$ by the Deduction Theorem. Through the induction hypothesis for $\varphi \vdash \chi $ we conclude $\Box _{\varphi ^{\circ _{\eta }}} (\bigwedge _{i \in I} \delta _i \to \theta )$ . Putting these two observations together, we obtain the desired $\Box _{\varphi ^{\circ _{\eta }}} \theta $ .

We turn to the necessitation rule, that if $\varphi \vdash \psi $ then $\Diamond \varphi \vdash \Diamond \psi $ . Reason in $\mathsf {HA}$ and let $\theta,\ \boldsymbol{y}$ , and $\boldsymbol {z}$ be arbitrary. Consider the induction hypothesis with $\theta := \neg \top $ (and $\boldsymbol {y}$ , $\boldsymbol {z}$ as given by our assumption): $\Box _{\psi ^{\circ _{\eta }}} \neg \top \to \Box _{\varphi ^{\circ _{\eta }}} \neg \top $ . Taking the contrapositive, we conclude $\Diamond _{\varphi ^{\circ _{\eta }}} \top \to \Diamond _{\psi ^{\circ _{\eta }}} \top $ , and consequently $\Box _\eta (\Diamond _{\varphi ^{\circ _{\eta }}} \top \to \Diamond _{\psi ^{\circ _{\eta }}} \top )$ by Lemma 10.1. Assume now that $\Box _{(\Diamond \psi )^{\circ _{\eta }}} \theta $ . By the Deduction Theorem, we obtain $\Box _\eta (\Diamond _{\psi ^{\circ _{\eta }}} \top \to \theta )$ . Thus our previous observation gives us $\Box _\eta (\Diamond _{\varphi ^{\circ _{\eta }}} \top \to \theta )$ , which is $\Box _{(\Diamond \varphi )^{\circ _{\eta }}} \theta $ by the Deduction Theorem.

Consider the transitivity axiom: $\Diamond \Diamond \varphi \to \Diamond \varphi $ . We start by observing that $(\Diamond \Diamond \varphi )^{\circ _{\eta }}$ is equivalent to $\eta (u) \lor u = \ulcorner \Diamond _\eta \Diamond _{\varphi ^{\circ _{\eta }}} \top \urcorner $ . Note that we can derive $\Box _\eta (\Diamond _\eta \Diamond _{\varphi ^{\circ _{\eta }}} \top \to \Diamond _{\varphi ^{\circ _{\eta }}} \top )$ from Lemma 10.1. Assume now $\Box _{(\Diamond \varphi )^{\circ _{\eta }}} \theta $ . By the Deduction Theorem we have $\Box _\eta (\Diamond _{\varphi ^{\circ _{\eta }}} \top \to \theta )$ . Then we obtain $\Box _\eta (\Diamond _\eta \Diamond _{\varphi ^{\circ _{\eta }}} \top \to \theta )$ by our previous observation, and we finish with one more application of the Deduction Theorem.

The $\forall $ introduction rule on the right states that if $x_0 \notin {\mathrm {fv}}(\varphi )$ and $\varphi \vdash \psi $ , then $\varphi \vdash \forall x_0 \, \psi $ . Reason in $\mathsf {HA}$ and let $\theta , \boldsymbol {y}$ , and $\boldsymbol {z}$ be arbitrary, where $y_0$ does not appear in $\boldsymbol {y}$ . Assume $\Box _{\exists y_0 \, \psi ^{\circ _{\eta }}} \theta $ and let $\boldsymbol {\pi } = \pi _0, \ldots , \pi _{n-1}$ be a proof of this fact. Let $\{\delta \}_{i \in I}$ be the finite set of axioms of $\exists y_0 \, \psi ^{\circ _{\eta }}$ appearing in $\boldsymbol {\pi }$ . Note that $\Box _\eta (\forall i \in I \, \delta _i \to \theta )$ holds by the Deduction Theorem. Then for each $i \in I$ there is $k_i$ such that $\delta _i$ is an axiom of $\psi ^{\circ _{\eta }}[y_0 {\leftarrow } k_i]$ , so in particular $\Box _{\psi ^{\circ _{\eta }}[y_0 {\leftarrow } k_i]} \delta _i$ . We can use the induction hypothesis for each $i \in I$ to conclude $\Box _{\varphi ^{\circ _{\eta }}[y_0 {\leftarrow } k_i]} \delta _i$ , and since $x_0 \not \in {\mathrm {fv}}(\varphi )$ , we also know that $y_0 \not \in {\mathrm {fv}}(\varphi ^{\circ _{\eta }})$ , and thus we obtain $\forall i \in I \, \Box _{\varphi ^{\circ _{\eta }}} \delta _i$ . We now use collection to obtain $\Box _{\varphi ^{\circ _{\eta }}} \forall i \in I \, \delta _i$ , and the result follows from our observation that $\Box _\eta (\forall i \in I \, \delta _i \to \theta )$ , noting that $\varphi ^{\circ _{\eta }}$ extends $\mathsf {HA}$ .

The $\forall $ introduction rule on the left states that if $\varphi [x_0 {\leftarrow } t] \vdash \psi $ , then $\forall x_0 \, \varphi \vdash \psi $ . Let $w$ be the arithmetical counterpart of t (so if t is $x_k$ then $w := y_k$ and if t is $c_k$ then $w := z_k$ ). We have $\varphi [x_0 {\leftarrow } t]^{\circ _{\eta }} = \varphi ^{\circ _{\eta }}[y_0 {\leftarrow } w]$ . Let $\theta , \boldsymbol {y}$ , and $\boldsymbol {z}$ be arbitrary, where $y_0$ appears in $\boldsymbol {y}$ if and only if $x_0$ is a free variable of $\psi $ . We assume $\Box _{\psi ^{\circ _{\eta }}} \theta $ . If t appears in $\varphi $ or in $\psi $ , then the value of w was already fixed when we picked arbitrary $\boldsymbol {y}$ and $\boldsymbol {z}$ . Otherwise, fix $w := 0$ and use the induction hypothesis to obtain $\Box _{\varphi ^{\circ _{\eta }}[y_0 {\leftarrow } w]} \theta $ . It is then clear that $\Box _{\exists y_0 \, \varphi ^{\circ _{\eta }}} \theta $ holds as well.

Finally, the term instantiation and constant elimination rules boil down to variable renaming.⊣

The arithmetical completeness of $\mathsf {QRC_1}$ with respect to $\mathsf {HA}$ remains an open question. We conjecture that the fact that the substitutions in the completeness proofs are of restricted complexity, and the fact that $\mathsf {PA}$ is $\Pi ^0_2$ conservative over $\mathsf {HA}$ (see [Reference Friedman, Scott and Muller20]) can be essential ingredients in a possible completeness proof.

11 Future work

There are many unexplored paths surrounding $\mathsf {QRC_1}$ . It would be worthwhile to extend it to a polymodal language (in analogy with $\mathsf {RC}$ , as in [Reference Beklemishev, Bolander, Braüner, Ghilardi and Moss6]), and to the positive setting (as in [Reference Celani and Jansana13, Reference Dunn17]). Whether this is possible without loosing decidability remains to be seen. A hypothetical $\mathsf {QRC}_\Lambda $ would presumably lead to some interesting applications to $\Pi ^0_1$ ordinal analysis and ordinal notation systems.

There are several proof theoretical properties of interest, such as interpolation, cut-free proofs, fixpoints, etc. One could also strive for uniform completeness.

Moreover, the set of always true $\mathsf {QRC_1}$ sequents should be a productive avenue of study.

The completeness of $\mathsf {QRC_1}$ with respect to $\mathsf {HA}$ remains an open question.

Finally, the current approximation for the computational complexity of $\mathsf {QRC_1}$ is super-exponential space, since the canonical model grows quite large. A more dedicated study might lead to a significant reduction in complexity.

Footnotes

1 Also known as c.e., recursively enumerable, or computably axiomatizable.

2 In general, $\mathsf {QPL}(T)$ may change depending on the chosen axiomatization $\tau $ for T (see [Reference Artemov3, Reference Kurahashi28, Reference Kurahashi29]), but here we omit the axiomatization for simplicity.

4 In [Reference de Almeida Borges, Joosten, Olivetti, Verbrugge, Negri and Sandu1] this was presented as an axiom of $\mathsf {QRC_1}$ , but it is easily provable through the two $\forall $ introduction rules and necessitation.

5 We only require extensional equality.

6 This is without loss of generality, as otherwise we could take the conjunction of every formula in $p^+$ as the new $p^+$ .

7 We assume constant domain models for simplicity, but these results also hold in the inclusive case.

8 In the sections on arithmetic, we always use x for variables of $\mathsf {QRC_1}$ and $y, z, u$ for variables of arithmetic. Furthermore, u is reserved for the Gödel numbers of axioms of theories.

9 $\mathsf {HA}$ has less complex axiomatizations, but $\Sigma ^0_1$ suffices for this section.

References

de Almeida Borges, A. and Joosten, J. J., Quantified reflection calculus with one modality , Advances in Modal Logic, vol. 13 (Olivetti, N., Verbrugge, R., Negri, S., and Sandu, G., editors), College Publications, Cambridge, 2020, pp. 1332.Google Scholar
Artemov, S. N., Nonarithmeticity of truth predicate logics of provability . Doklady Akademii Nauk SSSR , vol. 284 (1985), no. 2, pp. 270271 (in Russian). English translation in Soviet Mathematics Doklady , vol. 33 (1985), pp. 403–405.Google Scholar
Artemov, S. N., Numerically correct logics of provability . Doklady Akademii Nauk SSSR , vol. 290 (1986), no. 6, pp. 12891292 (in Russian).Google Scholar
Artemov, S. N. and Japaridze, G. K., Finite Kripke models and predicate logics of provability, Journal of Symbolic Logic, vol. 55 (1990), no. 3, pp. 1090–1098.Google Scholar
Beklemishev, L. D., The worm principle , Logic Colloquium ’02 (Chatzidakis, Z., Koepke, P., and Pohlers, W., editors), Lecture Notes in Logic, vol. 27, ASL Publications, Cambridge, 2006, pp. 7595.Google Scholar
Beklemishev, L. D., Calibrating provability logic: From modal logic to reflection calculus , Advances in Modal Logic, vol. 9 (Bolander, T., Braüner, T., Ghilardi, T. S., and Moss, L., editors), College Publications, London, 2012, pp. 8994.Google Scholar
Beklemishev, L. D., Positive provability logic for uniform reflection principles . Annals of Pure and Applied Logic , vol. 165 (2014), no. 1, pp. 82105.Google Scholar
Beklemishev, L. D., A note on strictly positive logics and word rewriting systems , Larisa Maximova on Implication, Interpolation, and Definability, (Odintsov, S., editor), Outstanding Contributions to Logic, vol. 15, Springer, Berlin–Heidelberg, 2018, pp. 6170.Google Scholar
Beklemishev, Lev D. and Pakhomov, Fedor N., Reflection algebras and conservation results for theories of iterated truth. Annals of Pure and Applied Logic , vol. 173 (2022), no. 5, pp. 103093.Google Scholar
Berarducci, A., ${\varSigma}_n^0$ -interpretations of modal logic . Bollettino dell’Unione Matematica Italiana , vol. 7 (1989), no. 3-A, pp. 177184.Google Scholar
Boolos, G. S., The Logic of Provability , Cambridge University Press, Cambridge, 1993.Google Scholar
Boolos, G. S. and McGee, V. R., The degree of the set of sentences of predicate provability logic that are true under every interpretation, Journal of Symbolic Logic, vol. 52 (1987), pp. 165–171.Google Scholar
Celani, S. and Jansana, R., A note on the model theory for positive modal logic . Fundamenta Informaticae , vol. 114 (2012), no. 1, pp. 3154.Google Scholar
Coq Development Team, The Coq Proof Assistant (1989–2022), 2022.Google Scholar
Craig, W., On axiomatizability within a system, Journal of Symbolic Logic, vol. 18 (1953), pp. 30–32.Google Scholar
Dashkov, E. V., On the positive fragment of the polymodal provability logic GLP . Mathematical Notes , vol. 91 (2012), nos. 3–4, pp. 318333.Google Scholar
Dunn, J. M., Positive modal logic . Studia Logica , vol. 55 (1995), pp. 301317.Google Scholar
Feferman, S., Arithmetization of metamathematics in a general setting . Fundamenta Mathematicae , vol. 49 (1960), pp. 3592.Google Scholar
Fernández-Duque, D., Worms and spiders: Reflection calculi and ordinal notation systems . IfCoLoG Journal of Logics and their Applications , vol. 4 (2017), no. 10, pp. 32773356.Google Scholar
Friedman, H., Classically and intuitionistically provably recursive functions , Higher Set Theory (Scott, D. S. and Muller, G. H., editors), Lecture Notes in Mathematics, vol. 699, Springer, Berlin, 1978, pp. 2128.Google Scholar
Fujiwara, M. and Kurahashi, T., Refining the arithmetical hierarchy of classical principles, preprint, 2022, arXiv:2010.11527 [math.LO].Google Scholar
Gödel, K., Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I . Monatshefte für Mathematik und Physik , vol. 38 (1931), pp. 173198.CrossRefGoogle Scholar
Hájek, P. and Pudlák, P., Metamathematics of First Order Arithmetic , Springer, Berlin–Heidelberg–New York, 1993.Google Scholar
Hao, Y. and Tourlakis, G., An arithmetically complete predicate modal logic . Bulletin of the Section of Logic , vol. 50 (2021), no. 4, pp. 513541.Google Scholar
Hughes, G. E. and Cresswell, M. J., A New Introduction to Modal Logic , Routledge, London, 1996.Google Scholar
Japaridze, G. K., The modal logical means of investigation of provability , Ph.D. thesis, Moscow State University, Moscow, 1986 (in Russian).Google Scholar
Kikot, S., Kurucz, A., Tanaka, Y., Wolter, F., and Zakharyaschev, M., Kripke completeness of strictly positive modal logics over meet-semilattices with operators, Journal of Symbolic Logic, vol. 84 (2019), no. 2, pp. 533–588.Google Scholar
Kurahashi, T., On predicate provability logics and binumerations of fragments of Peano Arithmetic . Archive for Mathematical Logic , vol. 52 (2013), pp. 871880.Google Scholar
Kurahashi, T., On inclusions between quantified provability logics , Studia Logica , vol. 110 (2022), pp, 165188.Google Scholar
Kurucz, A., Wolter, F., and Zakharyaschev, M., Islands of tractability for relational constraints: towards dichotomy results for the description logic EL , Advances in Modal Logic, vol. 8 (Beklemishev, L. D., Goranko, V., and Shehtman, V., editors), College Publications, Moscow, 2010, pp. 271291.Google Scholar
McGee, V. R., Truth and necessity in partially interpreted languages , Ph.D. thesis, University of California, Berkeley, 1985.Google Scholar
Montagna, F., The predicate modal logic of provability . Notre Dame Journal of Formal Logic , vol. 25 (1984), no. 2, pp. 179189.Google Scholar
Shapirovsky, I., PSPACE-decidability of Japaridze’s polymodal logic , Advances in Modal Logic, vol. 7 (Areces, C. and Goldblatt, R., editors), College Publications, Nancy, 2008, pp. 289304.Google Scholar
Solovay, R. M., Provability interpretations of modal logic . Israel Journal of Mathematics , vol. 28 (1976), pp. 3371.Google Scholar
Tarski, A., The concept of truth in formalized languages , Logic, Semantics and Metamathematics , Hackett, 1983, pp. 152278, English translation of Tarski’s 1936 Der Wahrheitsbegriff in den Formalisierten Sprachen.Google Scholar
Vardanyan, V. A., Arithmetic complexity of predicate logics of provability and their fragments . Doklady Akad. Nauk SSSR , vol. 288 (1986), no. 1, pp. 1114 (in Russian). English translation in Soviet Mathematics Doklady , vol. 33 (1986), pp. 569–572.Google Scholar
Vardanyan, V. A., Bounds on the arithmetical complexity of predicate logics of provability , Questions of Cybernetics: Complexity of Computation and Applied Mathematical Logic (Adyan, S. N., editor), vol. 134, Scientific Council of the USSR Academy of Sciences on the complex problem “Cybernetics”, Moscow, 1988, pp. 4672 (in Russian).Google Scholar
Visser, A. and de Jonge, M., No escape from Vardanyan’s theorem . Archive for Mathematical Logic , vol. 45 (2006), no. 5, pp. 539554.Google Scholar
Visser, A. and Zoethout, J., Provability logic and the completeness principle . Annals of Pure and Applied Logic , vol. 170 (2019), no. 6, pp. 718753.Google Scholar
Yavorsky, R. E., On arithmetical completeness of first-order logics of provability , Advances in Modal Logic, vol. 3 (Wolter, F., Wansing, H., de Rijke, M., and Zakharyaschev, M., editors), World Scientific, Leipzig, 2002, pp. 116.Google Scholar