Fritz

Propositional quantifiers are quantifiers binding proposition

letters, understood as variables. This Element introduces
propositional quantifiers and explains why they are especially
interesting in the context of propositional modal logics. It
surveys the main results on propositionally quantified modal
logics which have been obtained in the literature, presents Philosophy and Logic
a number of open questions, and provides examples of
applications of such logics to philosophical problems.

Propositional Quantifiers
Propositional
Quantifiers
About the Series Series Editors
This Cambridge Elements series provides Bradley
an extensive overview of the many and Armour-Garb
varied connections between philosophy SUNY Albany
and logic. Distinguished authors provide
an up-to-date summary of the results of
current research in their fields and give
Frederick Kroon
The University of
Auckland
Peter Fritz

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

their own take on what they believe are
the most significant debates influencing
research, drawing original conclusions.

Cover image: JESPER KLAUSEN /

SCIENCE PHOTO LIBRARY/
Getty Images ISSN 2516-418X (online)
ISSN 2516-4171 (print)
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Elements in Philosophy and Logic
edited by
Bradley Armour-Garb
SUNY Albany
Frederick Kroon
The University of Auckland

PROPOSITIONAL QUANTIFIERS

Peter Fritz
University of Oslo
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467

Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781009532747
DOI: 10.1017/9781009177740
© Peter Fritz 2024
This publication is in copyright. Subject to statutory exception and to the provisions
of relevant collective licensing agreements, no reproduction of any part may take
place without the written permission of Cambridge University Press & Assessment.
When citing this work, please include a reference to the DOI 10.1017/9781009177740
First published 2024
A catalogue record for this publication is available from the British Library.
ISBN 978-1-009-53274-7 Hardback
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

ISBN 978-1-009-17773-3 Paperback

ISSN 2516-418X (online)
ISSN 2516-4171 (print)
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this
publication and does not guarantee that any content on such websites is, or will
remain, accurate or appropriate.
 Propositional Quantifiers

Elements in Philosophy and Logic

DOI: 10.1017/9781009177740
First published online: May 2024

Peter Fritz
University of Oslo

Author for correspondence: Peter Fritz, peter.fritz@ifikk.uio.no

Abstract: Propositional quantifiers are quantifiers binding proposition

letters, understood as variables. This Element introduces propositional
quantifiers and explains why they are especially interesting in the context
of propositional modal logics. It surveys the main results on
propositionally quantified modal logics which have been obtained in the
literature, presents a number of open questions, and provides examples of
applications of such logics to philosophical problems.

Keywords: modal logic, propositional quantifier, propositionally quantified

modal logic, second-order propositional modal logic, sentential quantifier

© Peter Fritz 2024

ISBNs: 9781009532747 (HB), 9781009177733 (PB), 9781009177740 (OC)
ISSNs: 2516-418X (online), 2516-4171 (print)
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Contents

1 Propositional Quantifiers 1

2 Relational Frames 22

3 Beyond Relational Frames 51

4 Conclusion 88

List of Abbreviations 93

References 94
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Propositional Quantifiers 1

1 Propositional Quantifiers
1.1 Introduction
The simplest languages of formal logic are propositional. These languages
provide sentential letters and connectives with which we can represent, for
example, the conditional structure of a sentence like the following:

(1) If you invented the smiley, then I am the pope.

Letting s stand for you inventing the smiley, p for me being the pope, and using
→ to represent the conditional in (1), this can be formalized as follows:

(2) s → p

Other familiar sentential connectives, such as negation, conjunction, disjunc-

tion, and the biconditional can be treated similarly, for which I will use the
symbols ¬, ∧, ∨, and ↔ respectively.
It is often supposed that the connectives just mentioned are truth-functional.
For example, the truth-value of ¬p is plausibly determined by the truth-value of
p: ¬p is true just in case p is false, and ¬p is false just in case p is true. But many
applications of propositional logics across different disciplines require sen-
tential connectives which are not truth-functional. Standard examples involve
modal and epistemic notions, such as the following two:

(3) Necessarily, Jupiter is a planet.

(4) Kushim believes correctly that barley is a grain.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Necessity and belief are not truth-functional: some but not all truths are nec-
essary, and some but not all truths (and falsehoods) may be believed by a
given agent. Nevertheless, it is straightforward to extend the language of prop-
ositional logic to capture these statements as well. For the first, it suffices to
introduce a sentential operator □ for necessity, and for the second, a sentential
operator Bk for being believed by Kushim:

(5) □r
(6) Bk g ∧ g

In (5) and (6), r stands for Jupiter being a planet, and g for barley being a grain.
The study of logical systems with sentential operators which are not truth-
functional is known as modal logic; see, for example, Hughes and Cresswell
(1996) and Blackburn et al. (2001).
All three of the examples just mentioned are particular, as opposed to general.
For example, (4) attributes to Kushim a particular (correct) belief. But in many
 2 Philosophy and Logic

cases, it is important to be able to use quantifiers to express generality. For

example, we might want to say not just that this belief of Kushim’s is correct,
but that all of Kushim’s beliefs are correct. Or, to put the same point slightly
differently, we might want to say:

(7) Everything Kushim believes is true.

Similarly, instead of attributing necessity specifically to Jupiter being a planet,

we might want to say generally that necessity is not trivial, in the sense that
there are examples of necessities. That is:

(8) Something is necessary.

As a final illustration, note that when I say in (1) (“If you invented the smiley,
then I am the pope”) that I am the pope, I am simply saying something which is
patently false. I could equally have said that I am the king or queen of England,
or made any other absurd claim. More generally, I might say that any absurdity
is the case, or simply – and absurdly – that everything is the case:

(9) If you invented the smiley, then everything is the case.

Standard propositional languages do not provide any quantifiers, so there is

no useful way of regimenting these three quantified claims in these languages.
But it would take little to allow for such quantification: We would only have to
allow ourselves to use sentential letters like p, q, and r as variables, and to bind
them by a universal quantifier ∀ and an existential quantifier ∃, analogous to the
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

familiar case of the quantifiers of first-order logic which bind individual vari-
ables. With such quantifiers, (7), (8), and (9) are straightforwardly formalized
as follows:

(10) ∀p(Bk p → p)
(11) ∃p□p
(12) s → ∀q q

Logic is not just about formalizing statements, but also about capturing log-
ical properties of, and relationships among, such statements. For example,
(3) (“Necessarily, Jupiter is a planet”) is an instance of the existential claim
(8) (“Something is necessary”), and so the latter follows intuitively from the
former. Thus, we would expect ∃p□p to be counted as a logical consequence of
□r. Using instances of classical propositional reasoning, this follows straight-
forwardly by a schematic principle of existential introduction, in particular the
following instance:
 Propositional Quantifiers 3

(13) □r → ∃p□p

Analogously, s → p can be obtained from s → ∀q q using the following

instance of universal instantiation:

(14) ∀q q → p

By classical propositional reasoning, s → ∀q q and ∀q q → p give us s → p,

as required.
Such quantifiers, binding variables which occupy the position of formu-
las, are often called propositional quantifiers, for example by Kripke (1959)
and Fine (1970). This Element is about these quantifiers in the context of
propositional languages, their logic, and their applications in philosophy. This
introductory Section 1 explains first, in Section 1.2, why the job of proposi-
tional quantifiers cannot obviously be done by the more familiar quantifiers of
first-order logic. Section 1.3 gives a partial explanation of why propositional
quantifiers are rarely encountered: in the context of many logical systems, they
are redundant in at least one of two different senses of redundancy. Section 1.4
explains why there is nevertheless a range of interesting settings in which prop-
ositional quantifiers are not redundant, and in which they are usefully studied.
The main examples for this are propositional modal languages along the lines
mentioned earlier. Consequently, the following Sections 2 and 3 of this Ele-
ment are concerned with the resulting propositionally quantified modal logics.
Section 1.5 provides an outlook on these sections and explains why they are
structured according to different styles of set-theoretic model theories for such
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

logics. A final Section 1.6 of the present section gives a brief overview of the
historical development of propositional quantifiers in formal logic.
This Element presents a continuous narrative, but it is also possible to read
the various sections selectively and out of order. Figure 1 presents a depend-
ency diagram indicating which sections depend on which other sections. The
label “PML” indicates that starting with Section 2, familiarity with the basics
of propositional modal logic will be assumed. The central logical theory of
propositionally quantified modal logics is developed in the section numbers
highlighted in bold.

1.2 Why Propositional Quantifiers?

Before we get any further into the theory of propositional quantifiers, it is worth
considering why one might want to use these quantifiers in the first place. In
many contexts, quantificational claims can be captured straightforwardly using
first-order logic, which is an extremely well-understood and very well-behaved
 4 Philosophy and Logic

Figure 1 Dependency among sections

formal system. Can’t we just use first-order logic to formalize the preceding
examples?
In first-order logic, an atomic formula consists of an application of a pred-
icate F to a finite number of arguments t1, . . . , tn , forming the statement
Ft1 . . . tn . In the simplest case, the arguments are individual variables x1, . . . , xn ,
which can be bound by first-order quantifiers ∀ and ∃. Thus, the claim that every
echidna is happy can be formalized as follows, with the obvious interpretations
of the predicate letters:

(15) ∀x(Ex → Hx)

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

How might we use such first-order quantifiers to formalize, for example, (8)
(“Something is necessary”)? We cannot just replace a propositional variable by
a first-order variable: □ is a sentential operator, and so □ can only be applied
to formulas, and not to individual variables; thus the string ∃x□x is ill-formed.
There are two natural options to overcome this difficulty. First, we might use a
necessity predicate N instead of a necessity operator □. With N, we might use
the following formula of first-order logic:

(16) ∃xNx

Second, we might introduce a truth predicate T. We can then attribute necessity

to x by stating that x is necessarily true. With T, we might then use the following
formula of first-order modal logic:

(17) ∃x□Tx
 Propositional Quantifiers 5

Consider first the option of using modal predicates. Using a predicate N

instead of a sentential operator □ has a knock-on effect for the formalization
of particular necessity claims. As noted, we want our formalization to capture
that (8) (“Something is necessary”) follows from (3) (“Necessarily, Jupiter is a
planet”). But it is not clear how ∃xNx could be seen to be a logical consequence
of □r. The obvious response to this difficulty is to reconsider the formalization
of (3), and use N instead of □. However, we can again not simply exchange N
and □, since they take different types of expressions as arguments. In order to
be able to apply N, Jupiter being a planet must be expressed by an individual
term instead of a formula.
The problem would be solved if we could turn any formula, such as Fj,
into a corresponding individual term. A natural idea is therefore to introduce
a device which effects this transformation. So, let us consider an extension of
the language of first-order logic in which for every formula φ, there is an indi-
vidual term [φ]. For example, (3) (“Necessarily, Jupiter is a planet”) can then
be formalized in more detail as follows:

(18) N [Fj]

Now, ∃xNx follows straightforwardly from N [Fj] using existential introduction

in first-order logic.
On the modal predicate approach, there are also cases which call out for a
truth predicate. Consider (7) (“Everything Kushim believes is true”). As in the
previous case, we might use a first-order quantifier instead of a propositional
quantifier, and a predicate K instead of the sentential operator Bk to capture
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

the notion of being believed by Kushim. With this, we can state that Kushim
believes something using the formula ∃xKx. But it is not possible to formalize
(7) along the lines of ∀p(Bk p → p); in particular, the string ∀x(Kx → x) is
ill-formed, since x cannot take the position of a formula. The simplest way to
address this deficiency is to introduce a truth predicate T. We can then propose
to use the following formula:

(19) ∀x(Kx → Tx)

The truth predicate T serves as something of an inverse of the propositional

abstraction device [. . .] with which we obtain an individual term [φ] from a
formula φ: The first turns an individual term into a corresponding formula, and
the second turns a formula into a corresponding individual term.
Thus, the modal predicate approach leads us to introducing a truth predicate.
But, as we have seen, with a truth predicate we could also avoid introducing
modal predicates and write, for example, □Tx instead of Nx. It is also instructive
 6 Philosophy and Logic

to consider the inferential relationships between the various claims on this sec-
ond approach. To capture that (8) (“Something is necessary”) follows from (3)
(“Necessarily, Jupiter is a planet”), we now would want ∃x□Tx to follow from
□r. Again, this means going from a sentential expression r to an individual var-
iable x, which is naturally effected using the propositional abstraction device
[. . .]. For example, one might appeal to the following schematic principle
governing truth, where φ may be any formula:

(20) φ ↔ T [φ]

(This principle is closely related to the schematic biconditionals discussed by

Tarski [1983 [1933]], but note that the latter deal with sentences rather than
propositions.) Let φ be r, and consider the left-to-right direction: r → T [r].
By standard axiomatic modal reasoning, using the rule of necessitation and the
distributivity axiom for □, if r → T [r] is derivable, then so is □r → □T [r].
From □r, we therefore obtain □T [r], and so ∃x□Tx by existential introduction.
Thus, although we can use first-order quantifiers instead of propositional
quantifiers, both of the ways of doing so sketched here require further logi-
cal resources, in particular a truth predicate T and the propositional abstraction
device [. . .]. One of them allows us to continue to use modal operators, as
standard in modal logic; the other replaces them with modal predicates. It is
worth noting that one might also endorse a hybrid approach which provides
both modal operators and modal predicates; this could be justified by arguing
that there is a philosophically important distinction between “necessarily” and
“necessary” which gets conflated in standard uses of modal logic.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

We have seen that there are ways of formalizing the quantificational exam-
ples of Section 1.1 in first-order logic. However, I hope to have illustrated
that they come with certain complexities. In contrast, formalizations using
propositional quantifiers are extremely simple. The complexities of first-order
approaches may earn their keep by allowing for the formulation of theories
which do better on other dimensions of theoretical virtue. This is not the place
to try to settle these issues. For defenses of different versions of the first-order
approach, see Halbach and Welch (2009) and Bealer (1998); for exchanges on
the relative merits of the first-order and propositional approaches, see Anderson
(1987) and Bealer (1994), as well as Menzel (2024) and Williamson (2024).
One straightforward reason for investigating propositional quantifiers is
therefore their simplicity. Another has to do with the ontological commitments
of different forms of quantification. When we regiment claims like (8) (“Some-
thing is necessary”) using first-order quantifiers, it is natural to take these
quantifiers to range over propositions. Existential claims like ∃xNx and ∃x□Tx
therefore commit us to an ontology of propositions, namely to the existence
 Propositional Quantifiers 7

of certain things – propositions – which are necessary. Nominalists, according

to whom there are no propositions, will disagree with this claim. They might
instead appeal to propositional quantifiers and argue that there is a way of
interpreting the existential propositional quantification ∃p□p which makes it
an existential generalization of □r, without entailing the existence of propo-
sitions. On this view, propositional quantifiers should not be understood as
ranging over propositions. How, then, should they be understood? Accord-
ing to one version of this view, propositional quantifiers should be thought
of as a new, sui generis, form of quantification. In informal discussion, we
might still paraphrase them using the available English constructions, but these
paraphrases should – in the words of Frege (1892) – be taken with a grain of
salt. Alternatively, we could introduce into English new constructions which
better correspond to propositional quantifiers, as suggested by Prior (1971, sec-
tion 3.4) and Grover (1972). In any case, on this use of logics with propositional
quantifiers, there is a sense in which propositional quantifiers do not quantify
over propositions. For more on these kinds of views, see Fritz and Jones (2024).
In order to avoid suggesting that propositional quantifiers quantify over
propositions, such quantifiers are sometimes also called “sentential quan-
tifiers”; see, for example, Künne (2003). Although this label brings out clearly
that the relevant quantifiers bind variables which take sentential position, it also
invites the suggestion that they are quantifiers ranging over sentences. In many
cases, this would be a misunderstanding. There are substitutional readings of
propositional quantifiers, and model-theoretic constructions in which the truth-
conditions of propositional quantifiers are specified in terms of metatheoretic
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

quantification over sentences; such approaches will be discussed in Section 3.7.

But unless it is explicitly indicated that such a reading is intended, prop-
ositional (sentential) quantifiers should also not be taken as ranging over
sentences.

1.3 Redundancy
I have argued that propositional quantifiers are natural and useful logical con-
cepts, which cannot straightforwardly be replaced by more familiar forms
of quantification in formal logic. Yet, propositional quantifiers are relatively
underexplored. In this section, we observe one important reason for this: in
many contexts, propositional quantifiers are redundant in at least one of two
senses of redundancy.
To illustrate the first sense of redundancy, consider classical propositional
logic. Assume that the formulation of this logic under consideration includes
two logical sentential constants, ⊤ and ⊥, with ⊤ and ¬⊥ being provable.
 8 Philosophy and Logic

Since all sentential operators of classical propositional logic are truth-

functional, any two materially equivalent formulas can be replaced, salva ver-
itate, in any context. In this sense, classical propositional logic is extensional.
Because of this, we can simulate propositional quantification straightforwardly:
A universal quantification ∀pφ is true just in case φ is true under any interpre-
tation of p. By extensionality, since only the truth-value of φ matters, this is the
case if and only if φ is true when p is replaced by ⊤ or ⊥, for which we will write
φ[⊤/p] and φ[⊥/p], respectively. Instead of ∀pφ, we can therefore simply write
φ[⊤/p] ∧ φ[⊥/p]. Similarly, instead of ∃pφ, we can write φ[⊤/p] ∨ φ[⊥/p].
The argument just sketched is essentially semantic, as it appeals to interpre-
tations of the proposition letters. It is worth noting that a more careful version
of the argument can be carried out deductively, using the principles of clas-
sical propositional logic and standard principles of elementary quantification
for propositional quantifiers. The argument also extends to many other exten-
sional logics, including classical first-order logic, second-order logic, and any
extensions of these systems by generalized quantifiers: in all of them, adding
propositional quantifiers is redundant along the lines sketched earlier. It will
therefore be worth developing this argument a little more carefully and gen-
erally. We focus only on the universal propositional quantifier; the argument
for the existential propositional quantifier is analogous and can also be derived
from the universal case on the assumption that one quantifier is the dual of the
other.
Let L∗ be a logical language, the formulas of which are defined by a standard
recursion, and let ⊢∗ be a proof system for L∗ . We write ⊢∗ φ for a formula φ
of L∗ being provable in ⊢∗ . We assume the following:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

• We have defined the notion of an occurrence of a propositional variable

p being free in a formula φ, and the notion of a formula ψ being free for
a propositional variable p in a formula φ. (These are standard notions of
logical systems; see Section 2.1 for definitions in the case of propositionally
quantified modal logic.) If the latter condition is satisfied, we write φ[ψ/p]
for the result of replacing every free occurrence of p in φ by ψ.
• L∗ is closed under the Boolean connectives, including ⊤ and ⊥, as well as
the universal propositional quantifier ∀p, so that ∀pφ is a formula whenever
φ is a formula.
• The formulas provable in ⊢∗ include all classical tautologies and are closed
under modus ponens (MP), the rule that if ⊢∗ φ and ⊢∗ φ → ψ, then ⊢∗ ψ.
• ⊢∗ includes the principles of universal instantiation (UI) and universal gen-
eralization (UG) for propositional quantifiers. So, if ψ is free for p in φ,
then
 Propositional Quantifiers 9

⊢∗ ∀pφ → φ[ψ/p],

and if p is not free in φ, then

⊢∗ φ → ψ only if ⊢∗ φ → ∀pψ.

• L∗ is extensional according to ⊢∗ , in the sense that whenever ψ and χ are

free for p in φ, then:
⊢∗ (ψ ↔ χ) → (φ[ψ/p] ↔ φ[ χ/p])

It follows from these assumptions that the formulas provable in ⊢∗ are closed
under uniform substitution, in the sense that φ[ψ/p] is provable whenever φ
is provable and ψ is free for p in φ; the argument is analogous to the proof of
Proposition 2.2.2.
The elimination of propositional quantifiers can be stated more rigorously by
defining a mapping ·∗ from L∗ to the sublanguage of L∗ in which no universal
propositional quantifiers occur. This mapping is defined recursively, with only
one nontrivial clause:
(∀pφ)∗ := φ∗ [⊤/p] ∧ φ∗ [⊥/p]

To say that the mapping is recursive and all other cases are trivial is just to say
that ⊤∗ := ⊤, (¬φ)∗ := ¬(φ∗ ), and so on.
We can now prove that this elimination succeeds by showing that it maps any
formula to a provably equivalent formula which does not contain any universal
propositional quantifiers:

Claim For every formula φ ∈ L∗ , ⊢∗ φ ↔ φ∗ .

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Argument The argument is by induction on the complexity of φ. By the con-

struction of the mapping ·∗ , it suffices to consider just the case of universal
propositional quantifiers, and show that ⊢∗ ∀pφ ↔ (∀pφ)∗ on the assumption
that ⊢∗ φ ↔ φ∗ . We consider the two directions of the biconditional separately.
For the left-to-right direction, note first that by UI, tautologies, and MP,

⊢∗ ∀pφ → φ[⊤/p] ∧ φ[⊥/p].

Second, by induction hypothesis (IH), ⊢∗ φ → φ∗ . So, for + being ⊤ or ⊥, it

follows by uniform substitution that

⊢∗ φ[+/p] → φ∗ [+/p].

Using tautologies and MP, we obtain, from these two observations,

⊢∗ ∀pφ → φ∗ [⊤/p] ∧ φ∗ [⊥/p].

This is ⊢∗ ∀pφ → (∀pφ)∗ , as required.

 10 Philosophy and Logic

For the right-to-left direction, note first the following two instances of
extensionality:

⊢∗ ( p ↔ ⊤) → (φ∗ ↔ φ∗ [⊤/p])
⊢∗ ( p ↔ ⊥) → (φ∗ ↔ φ∗ [⊥/p])

By tautologies and MP, we conclude:

⊢∗ φ∗ [⊤/p] ∧ φ∗ [⊥/p] → φ∗

By IH, ⊢∗ φ∗ → φ, and so:

⊢∗ φ∗ [⊤/p] ∧ φ∗ [⊥/p] → φ

Since p is not free in the antecedent of this conditional, we can apply UG:

⊢∗ φ∗ [⊤/p] ∧ φ∗ [⊥/p] → ∀pφ,

which is ⊢∗ (∀pφ)∗ → ∀pφ, as required. This concludes the argument.

It is routine to show that many standard logical systems, including classi-

cal propositional logic, first-order logic, and second-order logic, all satisfy the
assumptions of this result when propositional quantifiers are added to the syn-
tax and the deductive principles UI and UG are added to a standard axiomatic
proof system. The crucial condition of extensionality can be shown by induc-
tion on the complexity of formulas. In all these logics, propositional quantifiers
are therefore redundant in the sense that they can be eliminated along the lines
developed here.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

The eliminability of propositional quantifiers shows that for many purposes,

it is pointless to add propositional quantifiers. However, not every property
one may be interested in is preserved under the elimination. Most importantly,
the elimination of propositional quantifiers leads to an exponential increase in
the size of formulas. Consequently, the elimination of propositional quantifiers
may not preserve matters of computational complexity of the logic in question.
This can be illustrated by the basic case of classical propositional logic. The
problem of determining whether a formula of classical propositional logic is
satisfiable (i.e., is true under some assignment of truth values to the proposi-
tion letters) is NP-complete, whereas the corresponding satisfiability problem
for classical propositional logic with propositional quantifiers is PSPACE-
complete. (For further details, including definitions of these complexity classes,
see Blackburn et al. [2001, pp. 514–516]. That PSPACE-completeness comes
apart from NP-completeness is conjectured, but has not been proven.) Clas-
sical propositional logic with propositional quantifiers is also known as QBF
(“quantified Boolean formulas”); for applications in artificial intelligence and
 Propositional Quantifiers 11

related fields, see Shukla et al. (2019). There are also applications of proposi-
tional quantifiers to the axiomatics of classical propositional logic; we discuss
them in Section 1.6.
In second-order logic, propositional quantifiers can also be seen to be redun-
dant in a second sense: Second-order logic adds to first-order logic variables
taking the position of predicates, and quantifiers binding these variables. For
example, in second-order logic, we can state ∀X(Xc). For a detailed intro-
duction to second-order logic, see Shapiro (1991). Just as predicate constants
can have any arity, so can second-order variables. Depending on the formula-
tion of second-order logic, this may not just include the cases of polyadic and
monadic variables, but also the limiting case of zero-adic, or nullary, variables.
A nullary predicate takes no arguments to form an atomic predication, so we
can understand it as a propositional constant. Analogously, a nullary second-
order variable can be understood as a propositional variable. Propositional
quantifiers can therefore be understood as nullary second-order quantifiers.
In the context of a suitable formulation of second-order logic, propositional
quantifiers are therefore also redundant in a second sense: they are already
included.
The second sense of redundancy extends to higher-order extensions of
second-order logic. This is important since third- and higher-order logics do not
(without further assumptions) satisfy the extensionality requirement of the first
redundancy argument. This is easy to see: In third-order logic, a standard pred-
icate F can not only be applied to individual terms to form an atomic formula,
but also serve itself as the argument of a (higher-order) predicate Y. Here, Y may
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

be a constant or a variable; if it is a variable, it may also be bound by a quan-

tifier. To illustrate, in this case we may construct a formula ∀Y(YF → ∀xFx).
Since propositional variables p and q can be understood as nullary second-order
variables, there is a third-order variable X which takes p or q as an argument.
Consider now the following claim:

( p ↔ q) → (Xp ↔ Xq)

Although p and q might have the same truth-value, the two variables need
not stand for the same proposition. Thus one of them might have a property
X which the other lacks. This is an informal semantic argument, but it can
easily be turned into a rigorous model-theoretic proof to show that in third-
and higher-order logic (without any extensionality assumptions), this instance
of the extensionality schema cannot be derived. In this sense, the step from
second- to third- and higher-order logic takes us outside the realm of extension-
ality. In third- and higher-order logic, propositional quantifiers are therefore
not redundant in the first sense, but they are redundant in the second sense:
 12 Philosophy and Logic

while they are not eliminable, they are already included as nullary second-order
quantifiers.

1.4 Fragments of Higher-Order Logic

The two arguments for redundancy discussed in the last section explain why it is
often uninteresting to add propositional quantifiers to a given logical language:
If the language is sufficiently restrictive to be extensional, like propositional,
first-order, and second-order logic, propositional quantifiers are eliminable. If
the language is sufficiently inclusive, like second-, third-, and higher-order
logic, then propositional quantifiers are already included as nullary second-
order quantifiers. In fact, it may seem that these two arguments cover all
interesting logical systems. But this is not so.
There are logics which are non-extensional without incorporating full third-
or higher-order logic. The most well-known cases are propositional modal log-
ics, which add to the language of classical propositional logic one or more
(non-truth-functional) sentential operators. Such operators can be understood
as third-order constants, taking nullary second-order variables (i.e., proposi-
tional variables) and complex terms of the same syntactic type (i.e., formulas)
as arguments. If we ignore the difference between variables and constants –
which is immaterial in the absence of quantifiers – the languages of proposi-
tional modal logics can therefore be seen as certain quantifier-free fragments of
the languages of third-order logic. (One might object that third-order constants
are non-logical, whereas modal operators are logical. But in practice, it is not
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

easy to see what this difference should amount to, especially as modal logics
are used in a variety of different applications, on different interpretations of the
modal operators.)
If the languages of propositional modal logic are fragments of the language
of third-order logic, then so are their extensions by propositional quantifiers:
they can be understood as obtained by adding to the relevant quantifier-free
fragment the special case of nullary second-order quantifiers. In such a con-
text, propositional quantifiers need not be redundant in either of the two
senses discussed earlier: the first redundancy argument does not apply, since
propositional modal logics are typically not extensional, and the second redun-
dancy argument does not apply since they do not already include propositional
quantifiers (or any other form of quantification).
One might wonder: Why bother with all these fragments? Why not just work
in third-, or better, higher-order logic? In some cases, this may well be the best
option, but it will not be the best option in all cases. Third- and higher-order log-
ics are complex systems, which may well introduce significant complications.
 Propositional Quantifiers 13

If the full flexibility of these systems is not required, it may well be a signifi-
cant advantage to make do with a more restrictive fragment. For example, many
second- and higher-order logics are essentially incomplete: the valid formulas,
on a suitable notion of validity, cannot be exactly captured by any proof system.
In technical terms, the validities are not recursively enumerable, essentially due
to the incompleteness results of Gödel (1931). By restricting the language to the
formulas of a propositional modal logic, this issue is very often avoided, as very
many propositional modal logics are computationally much more tractable.
In systematically investigating the principles of a logic of, for example,
belief, there are therefore good reasons not to always work in full higher-order
logic, but sometimes also in more restrictive fragments, like propositional logic
with an extra unary sentential operator. In some cases, however, the restrictions
of this particular fragment may prove to be too limiting. Our initial examples,
such as (7) (“Everything Kushim believes is true”), illustrate this, showing that
in many applications, it is natural to quantify over the arguments of modalities.
Such quantification does not require the full power of third-order logic: we
can expand propositional modal logic more carefully using just the required
propositional (i.e., nullary second-order) quantifiers. As we will see later in
this Element, in some but not all cases, such quantifiers preserve the tracta-
bility of propositional modal logics. Just like propositional modal logics, their
extensions by propositional quantifiers provide a trade-off between strength
and tractability. Which systems are most suitable will depend on the particular
application.
Propositional modal logics are the main setting in which propositional quan-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

tifiers have been studied, and they will also be the focus of this Element. But
there are other settings in which they play an important role. In particular,
there are other proper fragments of third- and higher-order logics which include
propositional quantifiers, and in which they are not eliminable. For example,
propositional quantifiers can be added to first-order modal logic. Such a lan-
guage has proven useful in investigating the interpolation property in first-order
modal logics; see the discussion of Fine (1979) by Kripke (1983), as well as
Fitting (2002). Another example is the extension of modal logics by proposi-
tional quantifiers and third-order quantifiers binding variables in the position of
sentential operators. A philosophical application of such a system can be found
in Fritz (2023c).
In the following, I will call a propositional modal logic with propositional
quantifiers a “propositionally quantified modal logic”. In the literature, this
label is used interchangeably with “quantified propositional modal logic”.
The reason for adopting the former rather than the latter is that the former
makes clear that only propositional quantifiers added. Logics with further
 14 Philosophy and Logic

quantifiers are therefore explicitly excluded, such as the one just mentioned
which incorporates certain third-order quantifiers.

1.5 Outlook
For the reasons discussed in the previous subsection, the two subsequent sec-
tions of this Element will focus on propositional quantifiers in the context of
propositional modal logic. It makes little sense to consider such extensions
without having already studied propositional modal logics on their own. Con-
sequently, from Section 2 on, I will assume familiarity with basic definitions
and results in propositional modal logic.
The material to be discussed in the remaining two sections is organized
according to different model-theoretic approaches. The main reason for this
is a matter of expository convenience: different model-theoretic approaches
require rather different mathematical tools. The sections aim to introduce the
relevant model-theoretic concepts, to consider the logics to which the relevant
models give rise from a deductive perspective, and to illustrate the usefulness
of these ideas in applications. For reasons of space, it will only be possible to
state many of the most important results; in these cases, proofs can be found
in the references provided. This Element therefore aims to be an introduction
and a survey rather than a comprehensive textbook on propositional quan-
tifiers. Although many interesting results have been obtained on propositionally
quantified modal logics, many natural questions have not been considered. A
number of basic results on such questions will be established here; in these
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

cases, proofs are either routine or are provided in detail. Many further ques-
tions remain open, and many specific results await a more systematic treatment.
Each section therefore also states some important open questions concerning
the relevant model-theoretic approach.
Section 2 discusses the most well-known models for propositional modal
logic, based on sets of possible worlds and relations of accessibility between
them; such structures are known as relational frames or Kripke frames. On
such frames, propositional quantifiers can be interpreted straightforwardly as
ranging over sets of worlds. Section 2.1 begins by laying out the syntax of
propositionally quantified modal logics. Section 2.2 sets up proof systems
in a general manner, and formulates the axioms and rules of classical logic
for propositional quantifiers. Section 2.3 introduces relational frames and the
resulting notion of normality for modal logics. To illustrate the usefulness
of propositional quantifiers in applications, this section also briefly discusses
the knowability paradox, and shows how its conclusion can be derived for-
mally using the proof systems defined here. Section 2.4 considers two further
 Propositional Quantifiers 15

principles which are valid on relational frames, namely a version of the Barcan
formula and a principle of atomicity which is unique to the case of propositional
quantifiers. Section 2.5 discusses the well-known modal logic S5, in which
propositional quantifiers behave especially well, and lead to an axiomatizable
logic. As a second illustration of an application of propositional quantifiers,
this section also discusses arguments to the effect that possible worlds can be
understood as special propositions, and shows how these arguments might be
formalized using propositional quantifiers. Section 2.6 turns to the more techni-
cal question of (recursive) axiomatizability of propositionally quantified modal
logics on classes of relational frames, and notes that propositional quantifiers
often lead to unaxiomatizability. Only in relatively few cases is it possible to
give complete axiomatizations; these cases and the corresponding complete-
ness questions are discussed in Section 2.7. The final Section 2.8 discusses
some model-theoretic beginnings, focusing in particular on transformations of
relational frames and their interactions with propositional quantifiers.
Section 3 turns to interpretations of propositionally quantified modal lan-
guages beyond relational frames. Section 3.1 begins by generalizing relational
frames to neighborhood frames, which are still based on possible worlds. Sec-
tion 3.2 goes further and generalizes neighborhood frames to models based
on complete Boolean algebras. Section 3.3 discusses a generalization which
can be applied to any frames based on possible worlds or models based on
complete Boolean algebras, which is to distinguish either one of the worlds,
or a filter of the algebra. Section 3.4 discusses a second way of generalizing
models based on complete Boolean algebras, namely by weakening the require-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

ment for the algebra to be complete. Section 3.5 considers the corresponding
generalization of relational and neighborhood frames, which involves limiting
which sets of possible worlds are counted as propositions. All of the generaliza-
tions of these five sections are well-known in the case of propositional modal
logics, but many of them have been little explored in the presence of prop-
ositional quantifiers. Section 3.6 considers possible worlds models in which
the domain of propositional quantification varies from world to world, adapt-
ing well-known model-theoretic ideas for modal predicate logic. Section 3.7
notes that propositional quantifiers need not be interpreted as ranging over any
domain of propositions at all, as there are formally coherent ways of inter-
preting propositional quantifiers as substitutional, and so as ranging – in some
sense – over sentences. The final Section 4 notes that even if propositional quan-
tifiers are interpreted as ranging over a domain of propositions supplied by a
model, and the model is to validate the principles of classical logic, the domain
of quantification does not have to constitute a Boolean algebra. It concludes
by using propositional quantifiers to discuss certain paradoxes of propositional
 16 Philosophy and Logic

attitudes and propositional individuation. A list of abbreviations is included at

the end.
The brief discussion of non-Boolean models in Section 4 also points to the
biggest omission of this Element: we will not consider propositional quantifiers
in the context of nonclassical propositional logics. The argument for the elim-
inability of propositional quantifiers in Section 1.3 depends on the principles
of classical logic. In certain nonclassical systems, including intuitionistic and
relevant logics, propositional quantifiers cannot be eliminated from proposi-
tional logic even without any additional modal operators. Apart from the study
of propositional quantifiers in the context of propositional modal logics, their
most detailed investigations in the literature can be found in the context of such
nonclassical logics. Although for reasons of space, these nonclassical investiga-
tions will not be discussed in this Element, references to the relevant literature
can be found at the end of the next section.

1.6 Historical Overview

While later sections will make references to the relevant literature, the aim
there will be to present the state of the art, rather than to proceed historically.
It will therefore be useful to provide some historical context, both on earlier
developments and on important motivating applications.

Early Work (1879–1955) The theory of quantification in modern sym-

bolic logic goes back to the Begriffsschrift of Frege (1879). The syntax of
Frege’s system is not defined rigorously, but the introduction of quantifiers by
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Frege (1879, p. 19) is plausibly general enough to encompass propositional

quantification. Indeed, Frege employs propositional quantifiers himself in cor-
respondence with Russell from 1904; see Frege (1980, pp. 158–166). The use
Russell and Frege make of propositional quantifiers is interesting: They discuss
a proposal by Russell to define negation by letting ¬φ stand for φ → ∀pp. It is
now not uncommon to define ¬φ as standing for φ → ⊥, and we can think of
Russell’s proposal as combining this with a definition of ⊥ as ∀pp. Effectively,
this defines the contradictory ⊥ as standing for the claim that everything is the
case. Although it may initially seem odd to define the negation of a statement
φ as the claim that φ materially implies a contradiction, we have already seen
in the first example (1) (“If you invented the smiley, then I am the pope”) that
something very similar is felicitous in English. In his letter, Frege hesitates to
endorse Russell’s definition of negation, but his reservations concern only the
relative priority of the relevant connectives, and not the intelligibility of prop-
ositional quantification, which Frege notes is easily expressed in the notation
of his Begriffsschrift.
 Propositional Quantifiers 17

Shortly after this correspondence, propositional quantifiers occur in pub-

lished work by Russell (1906, p. 192). However, Russell quickly moves from a
higher-order logic along the lines sketched in Section 1.4 to a much more com-
plicated system, the ramified type theory of Russell (1908) and Whitehead and
Russell (1910–13). Although ramified systems still contain quantifiers which
can be understood as propositional quantifiers, they are all in a certain sense
restricted; in this sense, these systems do not offer any unrestricted proposi-
tional quantifiers. Russell’s evolving views on logic – including propositional
quantification – can be traced through his published and unpublished writings
in this period, which are now available as Russell (2014).
An early philosophical application of propositional quantifiers occurs in the
work of Frank Ramsey. Ramsey (1923, 1927) proposes a version of the defla-
tionary theory of truth, according to which asserting that it is true that Caesar
was murdered amounts to asserting that Caesar was murdered. Ramsey notes
that such an elimination of the truth predicate is harder to effect in quantified
contexts, such as the earlier example of (7) (“Everything Kushim believes is
true”). However, Ramsey argues that this need for a truth predicate seems to be
a peculiarity of English, and not an essential matter concerning truth: using a
propositional variable p, we can state that that for every p, if Kushim believes
p, then p. In unpublished work, Ramsey (1991, p. 9, note 7) makes a similar
point explicitly using a propositional quantifier. Formalizing (7) using propo-
sitional quantifiers as (10), as proposed here, is therefore clearly in the spirit
of Ramsey’s proposal. Ramsey’s deflationism is anticipated in posthumously
published work, of 1904 by Brentano (1966, pp. 45–48) and of 1915 by Frege
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

(1979, pp. 251–252), but neither of them uses any symbolism in these notes,
and therefore also no propositional quantifiers.
A few years after Ramsey’s discussion, propositional quantifiers play a sig-
nificant role in the first edition of Lewis and Langford (1959 [1932], see ch. VI,
sect. 6). Lewis and Langford develop proof systems for propositional modal
logic, and note that they intend not to include the principle which may be
rendered in modern notation as follows:

(21) □( p → q) ∨ □( p → ¬q)

This is a version of the well-known principle of conditional excluded middle –

see, for example, Lewis (1973, pp. 79–83) – for strict implication. Lewis and
Langford note that they not only intend this principle not to be derivable, but in
fact to assert that it has false instances. This motivates them to introduce prop-
ositional quantifiers, with which the relevant claim is straightforwardly made
using a prefix of existential quantifiers binding p and q. Introducing standard
principles of quantification for propositional quantifiers, they go on to derive
 18 Philosophy and Logic

various formulas of propositionally quantified modal logic. Among the deriva-

ble formulas is a version of our earlier example (11) in which □ is formulated as
the dual of ^, namely ∃p¬^¬p; see Lewis and Langford (1959 [1932], p. 185,
20.21). In other discussions of modal logic until the late 1950s, propositional
quantifiers are rarely isolated. But in this period, modal logic is often devel-
oped in what would strike us today as very ambitious logical settings, in some
cases including first- and second-order quantifiers, which, as noted in Sec-
tion 1.3, subsume propositional quantifiers; for examples, see Barcan (1947),
Carnap (1947, esp. pp. 45, 179, & 191–193), and Bayart (1958, 1959), the last
of which are now available in translation and with commentary by Cresswell
(2015).
The definability of ⊥ in terms of propositional quantifiers suggested by
Russell’s definition of negation may seem to be nothing more than a neat
trick. But such definitions play an important role in the formulation of pro-
tothetic, a version of higher-order logic developed by Lesniewski (1929).
Using also quantifiers binding variables in the position of unary sentential
operators, Tajtelbaum-Tarski (1923) shows that conjunction, and so all other
truth-functional connectives, are definable using just the biconditional connec-
tive and universal quantifiers. In general, propositional quantifiers are found
in a number of early formulations of propositional logic, including one by
Łukasiewicz and Tarski (1930). In subsequent years, considerable ingenuity is
expended on finding the shortest axiomatization of classical propositional logic
in the presence of suitable quantifiers, culminating in the discovery by Mere-
dith (1951) of the sufficiency of the single axiom oo⊥ → op, with o an operator
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

variable and ⊥ defined as ∀pp, building on work by Łukasiewicz (1951).

Systems of protothetic which include a suitable axiom, such as the law of sub-
stitution discussed by Tajtelbaum-Tarski (1923), are extensional in the sense of
Section 1.3. This means that as long as a sufficient basis of truth-functional con-
nectives is included, propositional quantifiers can be eliminated, as discussed
by Church (1956, pp. 151–154) along the lines presented in Section 1.3. By
the 1950s, protothetic and investigations of axiomatizations of classical propo-
sitional logic using propositional quantifiers are already out of step with the
trajectory of symbolic logic. This sentiment is expressed clearly by Myhill
(1957, p. 118), who, in a review of a textbook on logic by Prior (1955), criticizes
Prior for devoting attention to such “bizarreries” as protothetic. Nevertheless,
propositional quantifiers receive an explicit defense by Church (1962), and an
extension of protothetic to a fuller propositional type hierarchy is discussed
by Henkin (1963) and Andrews (1963), albeit only as a stepping stone toward
developing new methods for a more standard higher-order logic.
 Propositional Quantifiers 19

Active Years (1955–1975) Starting in the late 1950s and continuing through
the 1960s, great advances are made in the study of modal logic. As we have seen
earlier, in such modal contexts, propositional quantifiers play a more substantial
role, and it is therefore not surprising to see a renewed interest in proposi-
tional quantifiers in this period. Essentially, these investigations return to the
applications of propositional quantifiers pioneered by Ramsey, and Lewis and
Langford, some three decades earlier. This is not to say that no studies of prop-
ositional quantifiers in intensional systems are considered in this intervening
period – see, for example, Łoś (1948), an English review of which was pub-
lished by Suszko (1949) – but only that they did not seem to have elicited much
excitement.
Arguably the most influential advocate of propositional quantifiers in the
period of their resurgence is Arthur Prior. Prior (1955, pp. 190–192) already
briefly discusses questions arising from the combination of propositional quan-
tifiers and modal operators, addressing an argument of Łukasiewicz (1951).
Two years later, Prior (1957) considers a range of interpretations of modal
operators, including temporal, alethic, epistemic, and deontic ones. Prior (1957,
pp. 130–131) briefly mentions an example involving propositional quantifica-
tion into an epistemic context, which prompts Cohen (1957, p. 231, fn. 6) to
challenge Prior to explain how his system avoids a form of the liar paradox.
Drawing on observations in a review of Koyré (1946) by Church (1946), Prior
(1958b) develops a more explicit deductive account of a propositionally quan-
tified modal logic, and uses it to discuss the paradox. Prior (1961) continues
this discussion; we return to it in the final Section 4.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Propositional quantifiers continue to play an important role in Prior’s work

after 1957, including the posthumously published Prior (1971). Especially
noteworthy is the proposal to reduce possible worlds or instants of time to
propositions in Prior (1967, p. 79). Prior (1968, pp. 205–207) develops this
suggestion further, and formulates a number of related open problems in propo-
sitionally quantified modal logic. Very soon after this, Bull (1968, 1969),
Kaplan (1970b), and Fine (1969, 1970) undertake systematic investigations of
propositionally quantified modal logic. Model-theoretically, these discussions
employ the famous possible worlds models of Kripke (1959, 1963a). In such
models, propositional quantifiers can straightforwardly be evaluated by letting
them range over sets of worlds in a given model; in fact, such an interpretation
of propositional quantifiers is already discussed by Kripke (1959, pp. 12–13).
Propositional quantifiers can also be found in a number of philosophical
applications in this period. An important example arises from a resurgence of
interest in Ramsey’s deflationism about truth; see, for example, Sellars (1960)
and Heidelberger (1968). (While Sellars does not refer to Ramsey, he mentions
 20 Philosophy and Logic

Carnap (1942), who in turn cites Ramsey (1923). It is also worth noting that Sel-
lars is likely to have been familiar with propositional quantification in modal
logic, since, as noted by deVries (2020), he studied modal logic with Lang-
ford around the time of publication of the first edition of Lewis and Langford
(1959 [1932]).) Combined with a defense of propositional quantifiers by Grover
(1970, 1972), this leads to a version of the deflationary theory of truth called
the prosentential theory of truth by Grover et al. (1975). For further examples
of uses of propositional quantifiers in this period, see Fitch (1963).
In a few short years, from 1967 to 1970, an immense amount of progress is
made on propositionally quantified modal logic. Many of the most fundamental
results in the area are established in this period, which will be covered in detail
later, with references to the original publications. The outstanding achievement
in this period is an article of Fine (1970), which focuses on propositionally
quantified modal logic interpreted over possible worlds models. It is hard to
convey a sense of just how many important contributions are contained in the
eleven short pages of this article, including axiomatizations, model construc-
tions, and results on completeness, decidability, and the failure of recursive
axiomatizability. The multitude of references to it in this Element will give
something of an impression of its importance. However, as a consequence of
the wealth of results in this short article, proofs are often only sketched in the
barest outlines.

Later Developments (1975–present) After an explosion of work on propo-

sitional quantifiers around the year 1970, interest in the topic seems to drop as
rapidly as it arose. Two reasons may contribute to the waning of propositional
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

quantifiers at this point: First, Arthur Prior, the chief proponent of proposi-
tional quantifiers in the 1960s, dies in 1969. Second, Quine (1986 [1970])
argues influentially against both modal and higher-order logic (where proposi-
tional quantifiers count as an instance of the latter). From a Quinean standpoint,
modal logic with propositional quantifiers manages to combine two conceptual
confusions in one very simple package.
Whatever the reasons for the lack of interest in propositional quantifiers in
modal logic, little work is done in the field for roughly the next twenty years.
This changes from the mid 1990s, as illustrated by publications of Ghilardi
and Zawadowski (1995), Kaminski and Tiomkin (1996), Kremer (1997c),
Antonelli and Thomason (2002) and ten Cate (2006). Especially in the last
decade, one finds a marked resurgence of higher-order modal logic in philos-
ophy, in no small part due to the influence of Williamson (2013). A number
of interesting philosophical arguments require propositional quantifiers but no
other higher-order quantifiers, and can therefore naturally be formulated in
 Propositional Quantifiers 21

propositionally quantified modal logic. For examples in this Element, see the
discussion of the knowability paradox in Section 2.3, the reduction of pos-
sible worlds to propositions in Section 2.5, the contingency of propositional
existence in Section 3.6, and the paradox of Epimenides in Section 4.
Over the last 60 years, many important results on propositionally quantified
modal logics are obtained, and many interesting applications are discussed.
Some examples of applications will be discussed in what follows, involving
epistemic and alethic modalities. Propositional quantifiers are also used in the
context of applications involving other modalities, such as deontic (see Prior
(1958a), Lokhorst (2006) and Rönnedal (2020)) and temporal (see Kesten and
Pnueli (2002) and French and Reynolds (2003)).
In these respects, propositionally quantified modal logic resembles standard
propositional modal logic. However, due to periods of neglect, the proposi-
tionally quantified case is comparatively understudied, and especially lacking
in general and systematic results. This is illustrated by the fact that a complete-
ness theorem for the most straightforward algebraic model theory of the most
straightforward propositionally quantified modal logic is only established very
recently, by Holliday (2019). The field is rife with similar longstanding open
questions, which are open not because of intrinsic mathematical difficulty, but
because of a lack of interest. In order to encourage research on these questions,
I will highlight some of them in the following sections.

Nonclassical Logics The preceding historical overview only covers proposi-

tional quantifiers in modal logic, as well as motivating philosophical discus-
sions. As noted in Section 1.5, it therefore omits the other important context
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

in which propositional quantifiers have been discussed, which is the study of

nonclassical logics. Since nonclassical logics are beyond the scope of this Ele-
ment, I won’t attempt to trace the relevant developments in any detail. But for
the benefit of the reader who wants to delve into propositional quantifiers in
nonclassical logics, I list some relevant publications here.
For propositional quantifiers in intuitionistic logic, see Myhill (1953),
Prawitz (1965, pp. 67–68), Gabbay (1974), Löb (1976), Sobolev (1977),
Kreisel (1981), Troelstra (1981), Pitts (1992), Połacik (1993, 1998a,b),
Skvortsov (1997), Kremer (1997b, 2018), Zach (2004), Ferreira (2006), Baaz
and Preining (2008), Zdanowski (2009), and Sørensen and Urzyczyn (2010).
For propositional quantifiers in relevant logic, see Anderson and Belnap,
Jr (1961), Anderson (1972), Routley and Meyer (1973, sect. 13–16), Ander-
son et al. (1992, ch. VI), Kremer (1993), Kremer (1997a), Goldblatt and Kane
(2010), and Badia (2019).
 22 Philosophy and Logic

There appear to be few investigations of propositional quantifiers in nonclas-

sical settings apart from intuitionistic and relevant logics, although see Baaz
and Veith (2000) and Baaz et al. (2000) for exceptions.

2 Relational Frames
2.1 Propositionally Quantified Modal Languages
We start with a formal definition of a language of propositionally quantified
modal logic. There are a number of choice of points in setting up this definition.
First, it is clear that the language will have to be based on a choice of atomic
proposition letters. Such letters could be divided into nonlogical constants and
variables, but we won’t need such a division: in all the applications discussed
in this section, the role of nonlogical constants can be played by free variables.
So we start from a countably infinite set of variables Φ. For elements of Φ, we
use lowercase Roman letters, in particular p, q, and r, and the results of adding
various indices to these letters.
Second, different choices of Boolean connectives may be included. Propo-
sitional modal logics are often defined using some small functionally complete
set of connectives, that is, a set of connectives such that every polyadic func-
tion on truth-values is expressed by some complex formula. This is because the
modal logics of interest typically do not distinguish between formulas which
are equivalent by the principles of classical propositional logic. For example,
if a modal logic does not distinguish between formulas of the form ¬φ and
φ → ⊥ in any context, then there is no reason to include ¬ as a connective of
the language in addition to → and ⊥; instead, we can simply introduce “¬φ”
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

as a way of abbreviating “φ → ⊥”. The main reason for limiting the number
of Boolean connectives is that it simplifies many definitions and proofs, and
we follow this common practice here, using just → and ⊥ as Boolean connec-
tives. For concreteness, we use other Boolean connectives as abbreviations, as
follows:

¬φ := φ → ⊥
φ ∨ ψ := ¬φ → ψ
φ ∧ ψ := ¬(φ → ¬ψ)
φ ↔ ψ := (φ → ψ) ∧ (ψ → φ)
⊤ := ¬⊥

The choice of → and ⊥ may seem a strange one, since ⊥ is not an especially
natural connective; in particular, there is no obvious way of rendering it in
English. From the perspective of translating the formal language into English,
it would be more natural to start with, for example, ¬ and ∧, which are very
 Propositional Quantifiers 23

naturally read as “not” and “and”. But → and ⊥ have two formal advantages:
→ is the most important connective for stating standard axiomatic principles,
so including it allows us to avoid using defined connectives in the statement
of these principles. And various arguments are simplified by ensuring that ⊤
and ⊥ contain no free variables (or quantifiers), which motivates including ⊥ as
primitive. It is worth noting that in the general context of classical modal logics
as introduced in the next section, the choice of primitive Boolean connectives
can make a substantial difference; see Makinson (1973) and Segerberg (1982)
for discussion.
Quantifiers present a choice-point similar to that of Boolean connectives. In
many logics, the existential and universal quantifiers are treated as each other’s
duals, so that, for example, formulas of the form ∃pφ and ¬∀p¬φ are not dis-
tinguished in any context. We take the universal quantifier ∀ as primitive, and
introduce the existential quantifier as the following abbreviation:

∃pφ := ¬∀p¬φ

It is worth noting that the definition of ⊥ as ∀pp discussed in Section 1.6 would
be available to us for much of the following. However, we still include ⊥ as a
primitive connective, for two reasons: First, as noted, it can be helpful for ⊤ and
⊥ to contain neither free variables nor quantifiers. Second, in discussing stand-
ard (i.e., quantifier-free) propositional modal logics, we still need to ensure that
the set of Boolean connectives is functionally complete.
Finally, our languages will contain modal operators. Most discussions of
modal logics only consider including a single unary operator □. But from the
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

perspective of applications, unary unimodal logics are very limiting, as urged

already by Scott (1970). First, many natural applications require multiple modal
operators. For example, already the simple claim that every truth can be known
requires two modal operators. Second, many natural modal connectives are pol-
yadic, such as the (binary) counterfactual conditional €; see Lewis (1973). We
therefore set up our formal language relative to a choice of operators, which we
call a modal signature. This will be specified by a set O of modal operators, and
a function ρ mapping every one of them to its arity. For example, a language
containing a unary (necessity) operator □ and a binary (counterfactual) con-
nective € will be specified using the modal signature ⟨{□, €}, ρ⟩, where
ρ(□) = 1 and ρ(€) = 2.
All of these choices lead us to the following definition:

Definition 2.1.1 A modal signature is a pair σ = ⟨O, ρ⟩, where O is a set, and
ρ : O → N.
 24 Philosophy and Logic

Let Φ be a countably infinite set of propositional variables. Given a modal

signature σ = ⟨O, ρ⟩, Lσ is the smallest set such that:

(i) If p ∈ Φ, then p ∈ Lσ .
(ii) ⊥ ∈ Lσ .
(iii) If φ, ψ ∈ Lσ , then (φ → ψ) ∈ Lσ .
(iv) If o ∈ O and φ1, . . . , φρ(o) ∈ Lσ , then oφ1 . . . φρ(o) ∈ Lσ .
(v) If p ∈ Φ and φ ∈ Lσ , then ∀pφ ∈ Lσ .
σ is the set of quantifier-free formulas of Lσ , that is, the smallest set
Lqf
σ instead of Lσ .
satisfying conditions (i)–(iv) for Lqf

A few clarifications and remarks on this definition are in order:

First, the letter “p” is not itself assumed to be a propositional variable (i.e., a
member of Φ); rather, it is used as a variable in the metalanguage (mathematical
English) to stand for a member of Φ. Nevertheless, we will use such letters later
to specify particular formulas, such as □( p → q) → (□p → □q). The intention
in such uses is that p and q are distinct members of Φ; since it won’t matter
which elements are chosen, they are not further specified.
Second, lowercase Greek letters like “φ” and “ψ” are used as metalinguistic
variables ranging over formulas. In clauses (i)–(v), concatenation is indicated
using juxtaposition. To illustrate this, consider clause (iii): Using “φ” and “ψ”
as metalanguage variables over formulas (i.e., elements) of Lσ , “(φ → ψ)” is
used to denote the result of concatenating (, φ, →, ψ, and ). Strictly speaking,
we have not said what (, →, and ) are. But since it does not matter, as long as
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

they are pairwise distinct – just as it does not matter exactly what the elements
of Φ are – we will leave this unspecified.
Third, according to clause (iv), a formula obtained using an application of
a binary modal operator o has the form oφψ. However, some modal opera-
tors, such as €, are more commonly used infix rather than prefix: it is more
common to write (φ € ψ) rather than €φψ. We therefore allow the former
notation as an abbreviation for the latter formula. When not required to resolve
any structural ambiguities, we drop the parentheses, as we do for →. Among
binary operators, we assume that ∧ and ∨ bind stronger than any other. So, to
illustrate, we write p → q ∧ r to abbreviate ( p → (q ∧ r)).
Finally, the notion of a free occurrence of a variable is defined as usual:
an occurrence of p is bound if it is in the scope of a quantifier ∀p, and free
otherwise. A variable p is free in φ if there is a free occurrence of p in φ. A
formula ψ is free for p in φ if no free occurrence of a variable in ψ becomes
bound when every free occurrence of p in φ is replaced by ψ. If ψ is free for
p in φ, we write φ[ψ/p] for the result of this replacement. We generalize this
 Propositional Quantifiers 25

last notation to finite sequences, writing φ[ψ1 /p1, . . . , ψn /pn ] for the result of
simultaneously replacing every free occurrence of pi in φ by ψi , assuming ψi is
free for pi in φ (for 1 ≤ i ≤ n, with p1, . . . , pn being pairwise distinct).

2.2 Modal Logics Based on Classical Logic

Having defined the languages of propositionally quantified modal logics, we
turn to the logics themselves. As common in modal logic, we will think of
a logic primarily as a set of formulas. Such a set will be required to satisfy
certain closure conditions, which correspond to axioms or rules in a Hil-
bert calculus. Other proof-theoretic approaches are possible, but we won’t
consider them here. For examples involving propositional quantifiers, see
the tableaux systems of Kripke (1959), Bull (1969), Rönnedal (2019, 2020)
and Blackburn et al. (2020), and the (labeled) natural deduction system of
Pascucci (2019).
The conditions on logics which we will assume for the most part ensure the
derivability of any principle of classical propositional logic and any principle of
elementary quantification theory for propositional quantifiers. There are many
different ways of imposing this requirement. We proceed as follows: First, we
demand that every tautology is included. Tautologies are the formulas involv-
ing only propositional variables and Boolean connectives which are true under
every truth-value assignment; by the soundness and completeness of classical
propositional logic, they are also just the theorems of classical propositional
logic. Second, we demand that the relevant set is closed under the rule of modus
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

ponens. With this rule, we could modify the first condition, and require only a
finite set of axioms to be included from which all tautologies are derivable using
modus ponens. This choice points won’t matter in the following, so we simply
include all tautologies; note that the definition of being a tautology using truth-
tables constitutes a decision procedure. Turning to quantifiers, we require the
inclusion of every instance of universal instantiation. This condition encap-
sulates the idea that if a given condition holds for every proposition, then it
holds as well for the proposition expressed by any given formula ψ. Finally,
we require closure under universal generalization: If a conditional φ → ψ
is contained in the set and p is not free in φ, then φ → ∀pψ must be con-
tained as well. This condition encapsulates the idea that propositional variables
stand for arbitrary propositions, and the thought that if a condition ψ is guar-
anteed to hold for an arbitrary p (on the assumption of φ), then it must hold
for every p (assuming φ). These four conditions suffice to ensure that the basic
principles of classical logic are included. We therefore call the relevant logics
classical:
 26 Philosophy and Logic

Definition 2.2.1 Given any modal signature σ, define the following conditions
on a set Λ ⊆ Lσ :

(Taut) φ ∈ Λ whenever φ is a tautology.

(MP) If φ ∈ Λ and φ → ψ ∈ Λ, then ψ ∈ Λ.
(UI) ∀pφ → φ[ψ/p] ∈ Λ whenever ψ is free for p in φ.
(UG) If φ → ψ ∈ Λ and p is not free in φ, then φ → ∀pψ ∈ Λ.

A classical propositionally quantified modal logic is a set Λ which satisfies all

four of these conditions.

Strictly speaking, the notion of a classical propositionally quantified modal

logic is relative to the signature σ, but since the signature will usually be clear
from context, we don’t usually mention this explicitly. In the literature, the
notion of a classical propositionally quantified modal logic is rarely isolated.
Instead, stronger conditions are usually imposed, along lines which we will
discuss below. For an exception in the unary unimodal case, see Ding (2018,
p. 220), who uses the label “Π-logic”.
From the four constraints of classical propositionally quantified modal log-
ics, we can show that any such logic contains all of the familiar principles
of elementary quantification theory for propositional quantifiers. We can also
show that any such logic is closed under uniform substitution: any free proposi-
tional variable may be replaced by any formula free for it in the relevant context.
The following result gives some examples of such quantificational principles,
including those used in the axiomatic system of Fine (1970).
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Proposition 2.2.2 If Λ is a classical propositionally quantified modal logic,

then it satisfies all of the following conditions:

(1) If φ is a tautology, then φ[ψ/p] ∈ Λ.

(2) If φ ∈ Λ, then ∀pφ ∈ Λ.
(3) If φ ∈ Λ and ψ is free for p in φ, then φ[ψ/p] ∈ Λ.
(4) φ → ∀pφ ∈ Λ, whenever p is not free in φ.
(5) ∀p(φ → ψ) → (∀pφ → ∀pψ) ∈ Λ.
(6) φ[ψ/p] → ∃pφ ∈ Λ, whenever ψ is free for p in φ.
(7) ∀pφ ↔ ∀q(φ[q/p]) ∈ Λ, whenever q is free for p in φ.

Proof We consider the first three items; the rest can be deduced from these by
standard arguments.
(1): Assume φ is a tautology. Then so is ⊤ → φ, which is therefore a member
of Λ, by Taut. So with UG, ⊤ → ∀pφ ∈ Λ. As ⊤ ∈ Λ by Taut, ∀pφ ∈ Λ
 Propositional Quantifiers 27

follows with MP. By UI, ∀pφ → φ[ψ/p] ∈ Λ. (ψ must be free for p in φ, as φ

is quantifier-free.) So by MP, φ[ψ/p] ∈ Λ.
(2): Assume φ ∈ Λ. Since p → (⊤ → p) is a tautology, it follows with (1)
that φ → (⊤ → φ) ∈ Λ. So ⊤ → φ ∈ Λ by MP, whence ⊤ → ∀pφ ∈ Λ by
UG. Since ⊤ is a tautology, it follows with Taut and MP that ∀pφ ∈ Λ.
(3): Assume that φ ∈ Λ and ψ is free for p in φ. Then ∀pφ ∈ Λ by (2). By
UI, ∀pφ → φ[ψ/p] ∈ Λ, and so φ[ψ/p] ∈ Λ with MP. □

As a special case of (3), we obtain the observation that relabeling free vari-
ables does not affect membership in a classical propositionally quantified modal
logic Λ: if Λ contains φ, then Λ also contains φ[q/p], as long as q is free for p
in φ. Further, from (7) we can see the equivalence of a universally quantified
statements with the result of relabeling the bound variable of the outermost
quantifier. By a straightforward induction, this can be extended to relabeling
the bound variables of any quantifiers which are not in the scope of any modal
operators. But we cannot remove the restriction on modal operators, since we
are not assuming that formulas which are equivalent by the lights of Λ are
interchangeable within modal operators. This is intentional: At this point, we
don’t want to impose any limitations on the distinctions which can be drawn
by modal operators. For example, we will consider a binary operator of iden-
tity = in Section 4. We don’t want to assume at this point that, for example,
(∀pp) = (∀qq), even if it is plausible. Consequently, we cannot assume that
bound variables can always be relabeled. It is therefore important that such
a relabeling is not built into the definition of φ[ψ/p], as it sometimes is, for
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

example, by Bull (1969, pp. 257–258). In many kinds of propositionally quan-

tified modal logics (in particular normal and congruential ones, defined later),
bound variables can be relabeled, but this must be established on the basis of
the particular constraints on modal operators.
Having defined the notion of a classical propositionally quantified modal
logic, it will be useful to define a corresponding notion for the quantifier-free
case. For our purposes, it will be useful not just to consider sets of quantifier-
free formulas satisfying Taut and MP, but sets which are also closed under
uniform substitution: Since a given proposition letter p stands for an arbitrary
proposition, a formula φ should only be contained in a logic if φ[ψ/p] is con-
tained as well. There was no need to impose this requirement in the presence
of quantifiers, as shown by Proposition 2.2.2. Without quantifiers, this natural
condition must be imposed separately:

Definition 2.2.3 Given any modal signature σ, a classical modal logic is a set
σ which satisfies Taut, MP, and the following condition:
Λ ⊆ Lqf
 28 Philosophy and Logic

σ , then φ[ψ/p] ∈ Λ.
(US) If φ ∈ Λ, p ∈ Φ, and ψ ∈ Lqf

The terminology used here comes apart somewhat from established termi-
nology, which uses the label “modal logic” for what we are calling a “classical
modal logic”; see, for example, Segerberg (1971, p. 8). Segerberg also uses
the label “classical modal logic” for a different notion, which we call “con-
gruential modal logic” in Section 3.1, following Makinson (1973, p. 196). The
reason for our use of “classical” for the notion just defined is that it corresponds
to the previous notion of classicality in the propositionally quantified case. As
noted, there are modal logics based on non-classical propositional logics, and
in Section 3.5, we will see logics which weaken the principles associated with
the quantifiers. Because of these reasons, we require the qualifier “classical”.
So far, we have taken an abstract approach to modal logics as sets of formu-
las satisfying certain closure conditions. One might wonder why we haven’t
considered any proof systems. In fact, each closure condition considered here
corresponds to a schematic axiom or rule, and a constraint like classicality can
be turned into a proof-theoretic concept. I will illustrate the point using the
notion of a classical propositionally quantified modal logic, but it is easy to see
that it applies as well to classical modal logics and similar concepts defined
below.
First, we note that for every set of formulas Γ ⊆ Lσ , there is a unique small-
est classical propositionally quantified modal logic including Γ, namely the
intersection of all classical propositionally quantified modal logics including Γ.
This intersection is guaranteed to be well-defined, since Lσ trivially includes Γ
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

and satisfies the four closure conditions. The smallest classical propositionally
quantified modal logic including Γ can alternatively be characterized as the set
of theorems of the proof system which contains as axioms the tautologies and
instances of the schema of universal instantiation (as in Taut and UI), as well
as the elements of Γ, and as rules modus ponens and universal generalization
(as in MP and UG). We therefore define:

Definition 2.2.4 For any set Γ ⊆ Lqf σ , the classical modal logic axiomatized
σ
by Γ, written C Γ, is the smallest classical modal logic including Γ. For any
set Γ ⊆ Lσ , the classical propositionally quantified modal logic axiomatized
by Γ, written CσΠ Γ, is the smallest classical propositionally quantified modal
logic including Γ.

σ γ . . . γ instead of Cσ Γ.
If Γ is a finite set {γ1, . . . , γn }, we also write C(Π) 1 n (Π)
σ
In particular, C(Π) is the smallest classical (propositionally quantified) modal
logic. For brevity, we usually omit mention of the modal signature σ in Cσ ,
 Propositional Quantifiers 29

unless it is not clear from context. All of these conventions will be applied to
analogs of this definition for normal and congruential modal logics as well.
A set of formulas is a classical propositionally quantified modal logic if and
only if it is the classical propositionally quantified modal logic axiomatized by
Γ, for some set Γ. This observation gives us a deductive formulation of classi-
cal propositionally quantified modal logics. It also follows that every classical
propositionally quantified modal logic Λ is axiomatized by some set, in par-
ticular by Λ itself, so in this sense, every such logic is axiomatizable. In this
sense, being axiomatizable is an uninteresting notion. In later discussions of
axiomatizability, we will therefore consider a more demanding notion. If we
think of a logic (set of formulas) Λ as given axiomatically, we say that φ is a
theorem of, or derivable in, Λ if φ ∈ Λ.

2.3 Relational Frames and Normality

The remainder of this section focuses on the most well-known model theory
for modal languages: the relational frames of Kripke (1963a). Such relational
frames can be defined for a language of any modal signature, as it is done by
Blackburn et al. (2001, p. 20), but they are especially natural in the case of
unary modal operators. In Section 3.1, we consider a generalization of rela-
tional frames which extends straightforwardly to polyadic modal operators. For
the remainder of this section, we only consider unary modal signatures, where
a signature ⟨O, ρ⟩ is unary just in case ρ maps every element of O to 1. In this
case, we omit mention of ρ and simply write LO . If O is a finite set {□1, . . . , □n },
we abbreviate this further to L□1 ...□n (and correspondingly in analogous con-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

texts that follow). Analogous to the definition of ∃ as the dual of ∀, we introduce

the following abbreviation for the dual of □:

^φ := ¬□¬φ

This convention is extended in the obvious way to □1 and ^1 , and so on.

Depending on the interpretation of □, there is often a corresponding natural
interpretation of ^. For example, in contexts in which □ is interpreted as a
notion of necessity, ^ can usually be interpreted as the corresponding notion
of possibility.
Kripke’s definition of relational frames combines two ideas: First, a rela-
tional frame is based on a set of (possible) worlds or points, each of which
settles the truth-value of every formula. (This notion of truth is relative to an
interpretation of the atomic proposition letters – we consider this in a moment.)
Second, for every (unary) modal operator □, a relational frame contains a binary
relation of accessibility R□ among worlds which determines the interpretation
 30 Philosophy and Logic

of □, where □φ is true in a world just in case φ is true in every accessible

world. It is worth noting already at this point that these two ideas are sepa-
rable: Section 3.1 will discuss a class of models which are based on worlds
but not accessibility relations, and Section 3.2 will mention a class of models
in which the interpretation of modal operators involves accessibility relations,
but no worlds.
The interpretation of formulas in relational frames is relative to a function
which maps every proposition letter to the set of worlds in which it is true. Since
proposition letters are often considered as constants, such a function is often
called a valuation function; when a frame is enriched with such a function it is
often called a model. Here we conceive of proposition letters as variables, so we
call such a function an assignment function instead. With this, the interpretation
of propositional quantifiers at a world is straightforward: ∀pφ is true relative
to an assignment function a just in case φ is true relative to every assignment
function which agrees with a on every propositional variable with the possi-
ble exception of p. This interpretation can already be found in Kripke (1959).
An alternative approach restricts the admissible assignment functions, allow-
ing only certain sets of worlds to serve as the interpretation of a propositional
variable; we return to this in Section 3.5.
To state the truth-conditions of universal and modal formulas, we introduce
the following conventions: a[x/p] is the function mapping p to x and every
propositional variable q , p to a(q). (Note that this convention is distinct from
the convention of writing φ[ψ/p] for the result of replacing every free occur-
rence of p in φ by ψ.) R(w) is the set of elements v such that Rwv; for any set x,
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

R[x] is the set of elements v such that Rwv for some w ∈ x. We can now define
relational frames formally as follows:

Definition 2.3.1 A relational frame (for a unary modal signature O) is a struc-

ture F = ⟨W, R□ ⟩□∈O such that W is a set and R□ ⊆ W × W for every □ ∈ O. An
assignment function (for F) is a function a : Φ → P(W ). A formula φ ∈ LO
being true relative to F, w ∈ W, and a, written F, w, a ⊩ φ, is defined by the
following recursive clauses:

F, w, a ⊩ p if and only if w ∈ a( p)
F, w, a ⊮ ⊥
F, w, a ⊩ φ → ψ if and only if F, w, a ⊩ φ only if F, w, a ⊩ ψ
F, w, a ⊩ ∀pφ if and only if F, w, a[x/p] ⊩ φ for all x ⊆ W
F, w, a ⊩ □φ if and only if F, v, a ⊩ φ for all v ∈ R□ (w)

From this, two notions of validity are derived, letting C be a class of relational
frames:
 Propositional Quantifiers 31

F ⊩ φ if F, w, a ⊩ φ for all w ∈ W and a : Φ → P(W )

C ⊩ φ if F ⊩ φ for all F in C

Relational frames provide a very flexible model theory for modal languages:
Every class of relational frames C determines a logic, namely the set of formu-
las valid on every frame in C. For example, in order to ensure the validity of
the principle □p → p, we may focus on relational frames in which R□ is reflex-
ive, and to ensure the validity of the principle ∃p(□p ∧ ¬p), we may focus on
relational frames in which R□ is irreflexive. This is a general feature of model
theories for modal languages, so we define generally:

Definition 2.3.2 If Γ is a set of formulas and C is a class of structures such that

a relation of validity ⊩ has been defined between members of the two classes,
Γ(C) is the set of formulas γ ∈ Γ such that S ⊩ γ for all structures S in C.

We call Lσ (C) the propositionally quantified modal logic of C, and Lqf σ (C)

the modal logic of C. We can show that the (propositionally quantified) modal
logic of any class of relational frames is a classical (propositionally quantified)
modal logic. In fact, the logics of classes of relational frames are guaranteed
to satisfy two additional conditions, which lead to the following notion of
normality:

Definition 2.3.3 For any set Λ ⊆ LO , we define the following conditions:

(K) □( p → q) → (□p → □q) ∈ Λ, for all □ ∈ O.

(Nec) If φ ∈ Λ and □ ∈ O, then □φ ∈ Λ.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Λ is a normal (propositionally quantified) modal logic if it is a classical

(propositionally quantified) modal logic satisfying K and Nec.

Since normality plays a central role in the remainder of this section,

we abbreviate “normal (propositionally quantified) modal logic” as “NML”
(“NPQML”). It is a standard result in modal logic that the modal logic of any
class of relational frames is normal. As the next proposition notes, this extends
to the propositionally quantified case:

Proposition 2.3.4 For any class C of relational frames, LO (C) is an NPQML.

Proof It is routine to show that LO (C) satisfies each of the six conditions. □

It is worth noting that in contrast to the notion of classicality, normality is not

motivated independently of the model theory of relational frames. This is illus-
trated by the fact that in the early literature on modal logic, at least two other
notions of normality were employed; see McKinsey and Tarski (1948, p. 7),
 32 Philosophy and Logic

Kripke (1963a, p. 67), and Lemmon (1977 [1966], p. 30). It is only in sys-
tematic explorations of relational frame semantics, such as Makinson (1966,
p. 379) and Segerberg (1971, p. 12), that the now-common notion of normal-
ity emerged (initially called “semi-normality” by Makinson). In the literature
on propositionally quantified modal logics, a definition of normality for the
unimodal case can be found in Ding (2018, 2021a) and Holliday (2019).
Analogous to the case of classicality, normality gives rise to a notion of
normal logics axiomatized by a given set of axioms:

Definition 2.3.5 For any set Γ ⊆ LO qf

, the NML axiomatized by Γ, written KO Γ,
is the smallest NML including Γ. For any set Γ ⊆ LO , the NPQML axiomatized
by Γ, written KOΠ Γ, is the smallest NPQML including Γ.

Just as a class of relational frames determines a set of formulas (the set of

formulas valid on the class), a set of formulas determines a class of relational
frames: the class of relational frames on which every member of the set is valid.
Again, this idea extends to any language and class of structures among which
a notion of validity has been defined:

Definition 2.3.6 If Γ is a set of formulas and C is a class of structures such that

a relation of validity ⊩ has been defined between members of the two classes,
then C(Γ) is the class of structures S in C such that S ⊩ γ for all γ ∈ Γ.

We write R for the class of all relational frames (for a given modal signature),
and say that R(Γ) is the class of relational frames defined by Γ. Since every NML
defines a class of relational frames, and every class of relational frames gives
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

rise to an NPQML, we obtain a model-theoretic way of extending NMLs by prop-

ositional quantifiers: Given an NML Λ, we may consider the NPQML of validities
on the class of frames validating Λ, namely, LO (R(Λ)). Adapting notation of
Fine (1970), we introduce the following abbreviation (with O determined by
context):

Definition 2.3.7 If Λ is an NML, then Λπ+ is LO (R(Λ)).

At this point, it is worth mentioning two other notations for proposition-

ally quantified modal logics used in the literature: First, Fine (1970) introduces
the notation Λπ alongside Λπ+, defined in terms of a wider class of mod-
els. We return to this notation in Section 3.5, when we consider the relevant
models. Second, Bull (1969) defines, for a propositional modal proof system
S, a proof system SΠ obtained by adding rules corresponding to UI and UG.
Recent authors including Ding (2018) and Holliday (2019) have adapted this
notation to propositional modal logics Λ; they write ΛΠ for the smallest NPQML
including Λ. On this notation, KΓΠ is what we here write KΠ Γ. The reason why
 Propositional Quantifiers 33

I don’t follow this notation here is the following: Given an NML Λ, there are sev-
eral ways of adding propositional quantifiers. For example, we might consider
the smallest classical propositionally quantified modal logic including Λ, or the
smallest NPQML including Λ. Adding Π as a subscript to K makes clear that the
normal extension is intended. The downside of this notation is that it makes
it awkward to specify normal propositionally quantified extensions when the
relevant NML is not given in the form KΓ. The most prominent cases are the
Lewis systems S4 and S5. This awkwardness can be mitigated by introducing
the convention of writing SΠ 4 and SΠ 5 instead of KΠ S4 and KΠ S5, respec-
tively, and analogously for their extensions. Similarly, some NMLs have names
of the form KX, without X being an axiom or list of axioms; an example is K3,
which we discuss in Section 2.6. In these cases, we also write KΠ X for KΠ KX.
Classes of relational frames, normal modal logics, and the relation of valid-
ity give rise to many questions which have been central to the study of modal
logic. Such questions are often phrased in terms of soundness and complete-
ness: A normal modal logic Λ is said to be sound with respect to a class of
relational frames C if every member of Λ is valid on C, and complete if every
formula valid on C is a member of Λ. Establishing these properties is especially
interesting if Λ is specified as KΓ, for a finite set of axioms Γ.
For every class of relational frames C, there is a unique normal modal logic
which is sound and complete with respect to it, namely LO qf
(C); this is the logic
determined by C. But conversely, it is not obvious that every normal modal logic
KΓ is sound and complete with respect to some class of frames. If it is, then it is
sound and complete in particular with respect to R(Γ), and so KΓ = LO qf
(R(Γ)).
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

KΓ is always sound with respect to R(Γ), but it is not obvious that KΓ should
always be complete with respect to R(Γ). That is, it may be that some formula
φ is valid on every relational frame in the class defined by Γ, even though
φ cannot be derived from Γ, in the sense that φ < KΓ. In fact, Thomason
(1974) and Fine (1974) showed that there are normal modal logics which are
not determined by any class of relational frames. Such logics are often called
Kripke-incomplete.
Although not every normal modal logic is Kripke-complete, very many nat-
ural normal modal logics are so. This includes every normal unimodal logic
axiomatized by a subset of the following common modal axioms:

D □p → ^p
T □p → p
B p → □^p
4 □p → □□p
5 ^p → □^p
 34 Philosophy and Logic

For example, KT is sound and complete with respect to the class of relational
frames on which T is valid, which contains just those relational frames with
a reflexive accessibility relation, and K is sound and complete with respect to
the class of all relational frames.
Much of the remainder of this section will be concerned with these kinds of
questions in the propositionally quantified modal case. As it turns out, in the
presence of propositional quantifiers, the situation changes drastically: KΠ Γ is
incomplete with respect to R(Γ) for very many natural choices of axioms Γ.
This incompleteness can have a number of different sources, which we explore
in the following.
Before turning to these model-theoretic questions in subsequent sections, it
is worth noting a purely deductive question which appears not to have been
considered in the literature. Say that a propositionally quantified modal logic
Π is a conservative extension of a modal logic Λ if the quantifier-free theorems
of Π are just the theorems of Λ; that is, if Π ∩ LO qf
= Λ. One would not expect
that adding propositional quantifiers governed by UI and UG to a given propo-
sitional modal logic would allow the derivation of any additional purely modal
(i.e., quantifier-free) principles. And indeed, it is not very difficult to show the
following, as I do in Fritz (unpublished):

Proposition 2.3.8 For every quantifier-free set Γ, CΠ Γ is a conservative

extension of CΓ.

However, there is a closely related question which is more difficult to settle.

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

This is the analogous question for normal logics, namely the question whether
KΠ Γ is always a conservative extension of KΓ. Note that this is not settled
by the previous result: if we let ∆ be KΓ, then C∆ is KΓ, and so CΠ KΓ is a
conservative extension of KΓ. But it doesn’t obviously follow that KΠ Γ is a
conservative extension of KΓ: KΠ Γ does, but CΠ KΓ does not obviously allow
quantified axioms to be necessitated. It is easy to see that if KΓ is the logic of
a class of relational frames, this class will also validate KΠ Γ, whence KΠ Γ is
a conservative extension of KΓ. But, as I show in Fritz (unpublished), there
are cases of Kripke-incomplete NMLs KΓ in which the incompleteness can be
demonstrated deductively with the addition of quantifiers:

Proposition 2.3.9 For some quantifier-free set Γ, KΠ Γ is a nonconservative

extension of KΓ.

To conclude this section, we illustrate the usefulness of proof systems

for propositionally quantified modal logics in applications with an argument
 Propositional Quantifiers 35

known the knowability paradox. Informally, this is an argument for the sur-
prising conclusion that if every truth is knowable, then every truth is known. It
was first published by Fitch (1963), but it is due to Church (2009 [1945]) who
formulated it in a referee report for an earlier version of Fitch’s article. Many
detailed discussions of the argument can be found in an edited volume on the
argument, Salerno (2009). Informally, the argument goes as follows:

Reasoning contrapositively, assume that some truth p is not known. We argue

that in this case, it is an unknowable truth that p is an unknown truth. The
argument assumes that the following principles of knowledge hold necessar-
ily: first, a conjunction is known only if the conjuncts are known, and second,
only truths are known. We can then argue as follows: If it were known that p
is an unknown truth, then, first, p would be known, and second, it would be
known that p is unknown. From the second of these consequences, it follows
that p is unknown, which contradicts the first consequence. So it is not pos-
sibly known that p is an unknown truth, even though p is an unknown truth.
Therefore, there is an unknowable truth, namely that p is an unknown truth.
Since p is arbitrary, we can conclude generally that if there is any unknown
truth, then there is an unknowable truth. Contrapositively, if every truth is
knowable, then every truth is known.

This argument is clearly sufficiently complicated for formalization to shed

light on the inferential assumptions required to arrive at the conclusion. The
argument is also very naturally formalized using propositional quantifiers. In
fact, Church and Fitch employed propositional quantifiers in their original for-
mulations of the argument. The most straightforward formalization uses two
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

modal operators: a unary operator K for “it is known that” and a unary operator
□ for “it is necessary that”. Then the conclusion of the knowability paradox can
straightforwardly be stated as follows:

(KNOWABILITY PARADOX) ∀p( p → ^Kp) → ∀p( p → Kp)

Even just regimenting the conclusion of this argument using propositional

quantifiers is illuminating: although the informal statement and argument
appealed to truth, no property of truth needs to invoked in the formal state-
ment using propositional quantifiers. (For more on this point, see Section 1.6.)
But the full potential of formalization is only realized once we regiment the
informal argument as a deductive proof. From such a deduction, we can glean
exactly which assumptions are needed to carry out the argument. It turns out
that all that is required, in addition to the assumptions encoded in NPQMLs, is
the axiom Kp → p which states that knowledge is factive:
 36 Philosophy and Logic

Proposition 2.3.10 KNOWABILITY PARADOX is provable in K□K

Π Kp → p.

Proof By the following sketch of a derivation (eliding some details of straight-

forward inferences in which quantifiers are not involved):

(1) K( p ∧ ¬Kp) → (Kp ∧ K¬Kp) KK

(2) K¬Kp → ¬Kp Kp → p
(3) ¬K( p ∧ ¬Kp) 1, 2
(4) □¬K( p ∧ ¬Kp) 3, Nec
(5) ∀p( p → ^Kp) → (( p ∧ ¬Kp) → ^K( p ∧ ¬Kp)) UI
(6) ∀p( p → ^Kp) → ( p → Kp) 4, 5
(7) ∀p( p → ^Kp) → ∀p( p → Kp) 6, UG □

2.4 The Barcan Formula and Atomicity

Returning to the sources of incompleteness of NPQMLs with respect to classes of
relational frames, we begin with two relatively straightforward cases. They are
two important schematic principles governing the interactions between modal
operators and propositional quantifiers which are valid on all relational frames,
but not guaranteed to be contained in NPQMLs.
The first principle is a propositional version of the Barcan formula of first-
order modal logic, named after Barcan (1946). This is the following schema,
where φ ∈ LO and □ ∈ O:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

(Bc) ∀p□φ → □∀pφ

The status of this schematic principle in the propositionally quantified setting

is analogous to its status in the setting of first-order modal logic: First, every
instance of Bc is valid on every relational frame. This follows by the fact that
propositional quantifiers range over the same domain at every world, namely
the powerset P(W ). Second, every instance of the converse of Bc, that is, the
schema □∀pφ → ∀p□φ, is derivable in every NPQML. Third, for any modality
□, the instances of Bc for □ are derivable in any NPQML which contains the
modal principle B for □. The latter two of these claims, concerned with deriv-
ability, can be established by straightforward analogs of standard derivations
in the first-order case, as found, for example, in Hughes and Cresswell (1996,
pp. 245–247). Finally, some instances of Bc are not derivable in certain NPQMLs
(not containing B for the relevant modality). This can be shown using models
presented below, in Sections 3.1 and 3.6.
 Propositional Quantifiers 37

The second kind of principle arises from the fact that for every world w of a
relational frame, the domain of propositions P(W ) contains the singleton {w}.
From an algebraic perspective, such propositions can be seen as atomic ele-
ments; we return to this perspective in Section 3.2. We therefore call singletons
of worlds atomic propositions. Such propositions have special properties which
can be expressed in a propositionally quantified language. This leads to a clus-
ter of related principles which are valid on relational frames but which are not
in general derivable in NPQMLs.
For the first such principle, note that for every world w, the atomic propo-
sition {w} is true in w and no other (accessible) world. Thus, at w, every true
proposition q is strictly implied by {w}, in the sense that necessarily, if {w}
then q. Relational frames therefore validate the following schematic principle,
which says that there is a proposition p with this feature (for any □ ∈ O):

(At) ∃p( p ∧ ∀q(q → □( p → q)))

For another example, assume that v is accessible from w, and consider the
atomic proposition {v}. From the perspective of w, {v} is possible, and for
every proposition p, {v} either strictly implies p or strictly implies the negation
(complement) of p. {v} therefore satisfies a condition which we can formulate
in the object language as follows, for any proposition expressed by a formula φ:

Q(φ) := ^φ ∧ ∀r(□(φ → r) ∨ □(φ → ¬r))

Here, r is some propositional variable not free in φ. If a proposition q is possible

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

from the perspective of a world w, then it is true in some accessible world v,

which means that {v} is a proposition satisfying Q which strictly implies q.
So, relational frames validate the following principle, which states that there is
always such a proposition p:

(At′) ^q → ∃p(Q( p) ∧ □( p → q))

At and At′ intuitively capture aspects of the availability of atomic proposi-

tions in domains of propositional quantifiers interpreted over relational frames.
We therefore call them principles of atomicity (an informal term which is not
meant to have any precise definition). In the next section, we will see that
in an NPQML with sufficiently strong modal principles, At′ follows from At
in the sense that the logic contains the latter only if it contains the former.
But the question how the two principles relate is much less straightforward
in weaker NPQMLs. Although principles like At and At′ were already consid-
ered by Prior (1967, pp. 79–82), there does not appear to exist any systematic
 38 Philosophy and Logic

investigation into their relations in arbitrary NPQMLs. I therefore consider this

an open problem in propositionally quantified modal logic:

Open Question 1 In the context of which NPQMLs does At entail At′ (and vice
versa), in the sense that the logic contains the former only if it contains the
latter?

Similar questions arise for further natural principles of atomicity. These prin-
ciples include a number of variants of At and At′ in which □ is strengthened
to a more demanding condition. By way of illustration, consider At. First, in
a bimodal setting with modalities □1 and □2 , we may not only consider an
instance of At for each modality, but also the following combined principle:

∃p( p ∧ ∀q(q → □1 ( p → q) ∧ □2 ( p → q)))

It is easy to see that this is derivable in any NPQML containing the relevant two
instances of At: if p1 and p2 are witnesses of At for □1 and □2 , respectively,
then p1 ∧ p2 is a witness of the combined principle. While this variant is easily
seen to be derivable, the matter is different with a second variant, which iterates
the modality as follows:

∃p( p ∧ ∀q(q → □□( p → q)))

It is far from clear that an NPQML which contains At must also contain this
iterated principle (unless, of course, it contains an iteration principle like 4).
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

2.5 S5 and the Reduction of Possible Worlds

In this section, we consider the special case arising from the well-known normal
unimodal logic S5. S5 is strong and simple, and is often considered a contender
for the correct modal logic of necessity in metaphysics. Apart from the principle
T, according to which what is necessary is true, it includes the principles 4
and 5, with which it follows that being necessary and being possible are both
noncontingent matters. The name “S5” is taken from Lewis and Langford (1959
[1932]), and indicates that it is the fifth in a series of systems they define, using
axioms first considered by Becker (1930).
S5 can be axiomatized as KT5, or alternatively as KT4B. It defines the class
E of relational frames in which the accessibility relation is an equivalence rela-
tion: a reflexive, symmetric, and transitive relation on the set of worlds. S5 is
Kripke-complete, and so complete with respect to E. It is also complete with
respect to the more restrictive class U of relational frames in which the accessi-
bility relation is universal. This fits a metaphysical conception of the intended
 Propositional Quantifiers 39

notion of necessity as being maximally broad; see Kripke (1980 [1972], p. 99)
and Williamson (2013).
Unusually, the results on S5 can be extended to propositional quantifiers.
First, the propositionally quantified modal logics of E and U are also the same,
as we will show in Section 2.8 using the standard technique of generated sub-
frames. Second, the resulting propositionally quantified modal logic also has
very natural axiomatizations. Since S5 contains B, any NPQML which includes
S5 already contains Bc. In Section 3.2, we will show that such an NPQML
need not include the atomicity principles discussed in the last section. But the
axiomatization of S5π+, that is, L□ (R(S5)), can be completed by adding At:

Theorem 2.5.1 SΠ 5At is sound and complete with respect to U, the class of
relational frames with a universal accessibility relation.

This result was shown by Kaplan (1970b) and Fine (1970). Soundness is
straightforward to establish using Proposition 2.3.4. Fine notes two ways of
establishing completeness: one is by a canonical model construction, and the
other by a quantifier-elimination argument using certain additional logical con-
stants. Kaplan employs the second method. Kaplan and Fine don’t provide
many details of these proofs, although related discussion can be found in
Kaplan (1970a) and Fine (1969, 1972).
As indicated in Section 1.6, an important motivation for the investigation
of S5π+ can be found in the works of Prior (1967, 1968). One of Prior’s con-
cerns was the task of giving an account of possible worlds in philosophical
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

theorizing. Naturally, structures such as relational frames containing elements

which are informally called “possible worlds” may be used for mathematical
purposes without any illuminating philosophical account of possible worlds:
in these contexts, possible worlds may be pure sets, or any other elements.
But in many philosophical contexts, a more substantial use is made of possible
worlds.
Prior considered the idea that a possible world may be considered as a propo-
sition, namely the proposition which describes the relevant world in its entirety.
According to Copeland (2006), this idea had been discussed by Prior and Mer-
edith as early as 1953, with reference to the Tractatus of Wittgenstein (1921),
and was stated in print in Prior (1962). The idea continued to occupy Prior’s
attention, and is discussed at length in work which was unfinished at the time
of his death, and which was published posthumously in Prior and Fine (1977).
Using propositional quantifiers and the technical developments discussed
in this section and the previous one, it is easy to see how the reduction of
worlds to propositions may be carried out: Assume that SΠ 5At is correct for
 40 Philosophy and Logic

the intended notion of necessity in the sense that all of its theorems are true
under any interpretation of the free propositional variables. (As discussed by
Williamson (2013, section 3.3), this is the notion of logical truth of Tarski (2002
[1936]).) Prior’s idea can then be understood as proposing that talk of possi-
ble worlds can be understood as talk of propositions corresponding to possible
worlds. To correspond to a possible world, a proposition has to describe a pos-
sible and complete way for things to be, that is, to be possible and to settle the
truth of every proposition. In this modal context, a proposition p can be under-
stood to settle a proposition q as being true if it strictly implies it, namely, if
□( p → q), and to settle q as being false of it strictly implies its negation, that
is, if □( p → ¬q). That is, a proposition p corresponds to a possible world just
in case it satisfies condition Q defined in the previous section. If p is a such
world-proposition, we understand a proposition q to be true in p just in case p
settles q as being true, that is, just in case □( p → q).
We can now use SΠ 5At to show that this way of regimenting talk of possible
worlds vindicates central assumptions about the relationship between possible
worlds and propositions. Most importantly, we can prove what Menzel and
Zalta (2014) call the fundamental principle of world theory: the claim that a
proposition is possible just in case it is true in some possible world. On Prior’s
proposal, this can be regimented as the following sentence of L□ :

∀q(^q ↔ ∃p(Q( p) ∧ □( p → q)))

This sentence is provable in SΠ 5At: it is easy to see that it is valid on U, so the

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

claim follows from Theorem 2.5.1.

This sketch of Prior’s proposal is somewhat of a simplification: In actual fact,
Prior held that the existence of propositions is in general a contingent matter.
This conflicts with SΠ 5At, and in particular with certain instances of the Barcan
formula and its converse. Weakening the logic to admit the contingent existence
of propositions introduces substantial complications, as discussed by Fine in
the postscript of Prior and Fine (1977). We return to some of these issues in
Section 3.6.
Setting aside difficulties arising from contingent existence, Prior’s proposal
is similar to many other theories of possible worlds, which conceive of possi-
ble worlds as special propositions, states of affairs, properties, sets, classes, or
various similar entities; see, for example, Plantinga (1974), Stalnaker (1976),
and Adams (1981). Prior’s proposal has the advantage of receiving an elegant
formalization using propositional quantifiers. But as Menzel and Zalta (2014,
p. 336) note, these proposals typically depend on a substantial assumption,
namely that there are the required propositions (states of affairs/properties/…)
 Propositional Quantifiers 41

to play the role of worlds. This applies to Prior’s proposal as well: the claim
that a proposition is possible just in case it is true in some world(-proposition)
depends essentially on the inclusion of the atomicity principle At. Using the
models of Section 3.2, we can show that this equivalence cannot be derived in
SΠ 5. This poses the question: Why think that At is (necessarily) true?
It turns out that the matter can easily be settled if a way of quantifying
plurally over propositions is added to L□ . The relevant plural propositional
quantifiers are analogous to the plural (first-order) quantifiers of Boolos (1984).
By a variant of arguments of Gallin (1975), Fine (1980), and Menzel and Zalta
(2014), it can be shown that At becomes derivable in the presence of such plu-
ral propositional quantifiers. I discuss this derivation in more detail in Fritz
(2023c). This result illustrates that just as in the case of standard proposi-
tional modal languages, the restrictiveness of propositionally quantified modal
languages is both a boon and a burden: it makes results like Theorem 2.5.1
possible, but also bars certain interesting arguments from being formulated.
Another instance of this important lesson will be mentioned in Section 4.
Finally, it is worth noting that a reduction of possible worlds to propositions
can also be carried out in the context of modal logics weaker than S5. However,
further complexities arise in these settings, in particular since we may no longer
be able to rely on a completeness result corresponding to Theorem 2.5.1, as
we will see in the next section. For recent discussion of such more general
reductions of possible worlds to propositions in higher-order languages, see
Dorr et al. (2021, section 1.6) and Bacon (2024, section 7.5).
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

2.6 Axiomatizability
Earlier, we noted that every NPQML is axiomatizable in the sense of being the
NPQML axiomatized by itself. This sense of axiomatizability is therefore uninter-
esting. A more interesting notion of axiomatizability is that of being recursively
enumerable, which requires being relatively well-behaved from a computa-
tional perspective: a set of formulas Λ is recursively enumerable if there is a
computer program which produces only theorems of Λ, and, for every theorem
Λ, produces it after some finite number of computational steps. For example, if
Γ is a finite set of axioms, then KΠ Γ is recursively enumerable. So, since S5π+
can be axiomatized as SΠ 5At (i.e., KΠ T5At), it is recursively enumerable. In
fact, S5π+ is computationally even better behaved, as it is decidable, which
means that there is a computer program which takes any formula φ ∈ L□ as
input, and outputs after a finite number of computational steps the answer to
the question whether φ ∈ S5π+:
 42 Philosophy and Logic

Theorem 2.6.1 S5π+ is decidable.

This result was also established by Kaplan (1970b) and Fine (1970). Fine
notes that it follows from the quantifier-elimination argument used to prove
completeness. As shown by Ding (2021a), this argument can be extended to
show that KD45π+ is decidable as well. Fine observes that there is a sec-
ond way of proving Theorem 2.6.1, which is also used by Kaplan. It proceeds
by extending the standard translation of propositional modal logic to propo-
sitional quantifiers. The standard translation shows that propositional modal
languages can be thought of as fragments of first-order languages: worlds
become individuals, propositional variables become predicates, and modal
operators become first-order quantifiers restricted to accessible worlds. This
extends straightforwardly to propositional quantifiers, which become monadic
second-order quantifiers. The translation can therefore be defined as follows,
given a first-order variable x which serves as the world of evaluation of the rele-
vant formula, and assuming that Xp is a distinct monadic second-order variable,
for every propositional variable p:

px := Xp x
⊥x := ⊥
(φ → ψ)x := φx → ψ x
(□φ)x := ∀y(R□ xy → φy )
(∀pφ)x := ∀Xp (φx )
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

The target language of monadic second-order logic is straightforwardly inter-

preted over relational frames understood as standard models, with the set of
worlds serving as the individual domain. By a standard abuse of notation,
we are here using R□ also as a predicate of the monadic second-order object
language, taking it to be interpreted using the relation R□ of the relevant frame.
An induction shows that a propositionally quantified formula φ is valid on
a relational frame just in case its translate φx is valid on the frame as well,
in the sense of being true under any assignment function. This means that
for any class of relational frames C, LO (C) can be seen as a fragment of
monadic second-order logic interpreted over the frames in C. (Because of this,
propositionally quantified modal logic, interpreted on relational frames, is also
sometimes called “second-order propositional modal logic”, for example, by
Kaminski and Tiomkin (1996), ten Cate (2006), and Belardinelli et al. (2018).)
This means that the former is no more complex than the latter. The decidability
of S5π+ therefore follows from the decidability of the monadic second-order
validities over U. When interpreted over U, any formula R□ xy becomes trivially
 Propositional Quantifiers 43

true, and so can be replaced by ⊤. The monadic second-order validities over U

can therefore be thought of as monadic second-order logic without non-logical
constants, which is well-known to be decidable, by observations going back to
Löwenheim (1915).
The resulting argument for the decidability of S5π+ effectively proceeds by
showing that a formula φ ∈ L□ is a member of S5π+ if and only if it is valid
on every frame in U up to a certain finite size determined by the syntactic com-
plexity of φ. It thereby also naturally generalizes the proof of the decidability
of S5 by Parry (1933). This mode of argument can be extended to several other
strong NPQMLS, as shown by Fritz (forthcoming). To state the relevant result,
let worlds w and v of a relational frame F be duplicates if the function mapping
w and v to each other, and every other world to itself, is an automorphism of
the whole frame (an isomorphism from the frame to itself). Being a duplicate
is an equivalence relation. Let the diversity of a class of frames C be the supre-
mum – if it exists – of the numbers of equivalence classes of point-generated
subframes of frames in C. It can then be shown that:

Proposition 2.6.2 If Γ ⊆ LO is finite and R(Γ) has finite diversity, then

LO (R(Γ)) is decidable.

Various decidability results can be derived from this. For example, with the
fact that every normal extension of K5 is finitely axiomatizable, as shown
by Nagle and Thomason (1985), it follows that Λπ+ is decidable for every
unimodal NML Λ containing the axiom 5. The general formulation of this obser-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

vation might give a misleading impression: despite its simplicity, axiom 5

is very strong, and in general only highly restrictive logics define classes of
relational frames with finite diversity. In fact, the vast majority of NMLs of inter-
est define classes of relational frames whose propositionally quantified modal
logic is not recursively enumerable. Fine (1970) already noted this for several
examples, including KT4, also known as S4. Fine’s argument proceeds by a
reduction of second-order arithmetic into the relevant NPQMLs. The details of
these arguments were eventually published by Garson (1984).
Kremer (1993) notes that Fine’s unaxiomatizability results can be strength-
ened, to show that second-order logic – i.e., the validities of a full second-order
language over standard models – can be reduced to these NPQMLs. This means
that from a computational perspective, these NPQMLs are equivalent to second-
order logic; in technical parlance, they are recursively isomorphic to second-
order logic (see Kremer (1993) for a definition of this concept). According to
Kremer (1997b, p. 530), this stronger result was already proven by Fine and
Kripke “shortly after the publication of” Fine (1970). Kaminski and Tiomkin
 44 Philosophy and Logic

(1996) published a proof of a generalization of this result, which is further gen-

eralized in Fritz (forthcoming). The full result is complicated to state, but we
note the following important consequence:

Proposition 2.6.3 Any unimodal NPQML included in Bπ+, K2.1π+, or

K4.2Wπ+ is recursively isomorphic to second-order logic.

Here, B is KTB, K2.1 is KT4MG1J1, and K4.2W is KG1W, where the

axioms not yet introduced are as follows:

M □^p → ^□p
G1 ^□p → □^p
J1 □(□( p → □p) → p) → p
W □(□p → p) → □p

It is worth noting that K4.2W proves 4 and J1. J1 is also known as Grz; W
is also known as GL, after Gödel and Löb, as it plays an important role in
provability logic; see Boolos (1985).
There is one important group of NMLs which the results mentioned so far do
not cover, namely those (apart from S5) containing the following axiom:

Lem0 □(( p ∧ □p) → q) ∨ □((q ∧ □q) → p)

This axiom enforces a weak form of linearity on relational frames. Fine (1970)
noted that the propositionally quantified unimodal logic of the singleton class
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

containing the relational frame consisting of the natural numbers under their
natural order (weak or strict) is decidable. This follows by the standard trans-
lation using the corresponding result about monadic second-order logic, which
had been shown by Büchi (1962). In Fritz (forthcoming), this observation is
extended to show the following result:

Proposition 2.6.4 If Λ is S4.3.1 or one of its normal unimodal extensions, or

K4.3Z, then Λπ+ is decidable.

Here, S4.3.1 is KT4Lem0 N1 and K4.3Z is K4Lem0 Z, where the axioms

not yet introduced are as follows:

N1 □(□( p → □p) → p) → (^□p → p)

Z □(□p → p) → (^□p → □p)

N1 is also known as Dum.

 Propositional Quantifiers 45

Büchi’s result was strengthened by Rabin (1969), and Fine (1970) asserts
that this can be used to establish the decidability of S4.3π+ (with S4.3 being
KT4Lem0 ). But according to Kaminski and Tiomkin (1996), this is incorrect,
as it follows from results of Shelah (1975) and Gurevich and Shelah (1983)
that S4.3π+ is in fact not recursively enumerable. In Fritz (forthcoming), this
observation is extended to show the following result, where K3 is KT4Lem0 M:

Proposition 2.6.5 For any normal unimodal logic Λ included in K3, Λπ+ is
not recursively enumerable.

As the results presented here illustrate, failure of recursive enumerability is

usually preserved under enlargements of the relevant class of frames, while
recursive enumerability is often preserved under restrictions of the relevant
class of frames. Among many NMLs Λ, the question whether Λπ+ is recursively
enumerable therefore divides the space of NMLs into two halves, separated by
what Ding (2021a) calls the axiomatizability boundary, where Λ is sufficiently
strong to be above the boundary just in case Λπ+ is recursively enumerable.
As noted in Fritz (forthcoming), such a boundary cannot be drawn among arbi-
trary NMLs, since there are cases in which Λ′ extends Λ, even though Λπ+ is
recursively enumerable while Λ′ π+ is not. The cases of logics used to show
this involve infinitely many axioms, which suggests the following apparently
open question, further discussed in Fritz (forthcoming):

Open Question 2 Are there finite sets Γ ⊆ ∆ ⊆ L□qf such that KΓπ+ is
recursively axiomatizable while K∆π+ is not?
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

So far, we have only discussed unimodal logics. In the multimodal case,

computational tractability becomes an even rarer phenomenon. For example,
even though the unimodal logic S5 defines a class of relational frames whose
propositionally quantified modal logic is decidable, once we admit two modal-
ities □ and ⊠ each governed by S5, the resulting propositionally quantified
modal logic is again recursively isomorphic to second-order logic. This nor-
mal bimodal logic is known as the fusion of S5 with S5, and written S5 ⊗ S5.
To define it, we write T⊠ for the variant of the axiom T in which □ is replaced
by ⊠, that is, ⊠p → p, and similarly for other axioms and modalities. Then
S5 ⊗ S5 can be defined as the normal bimodal logic KT□ 5□ T⊠ 5⊠ .

Proposition 2.6.6 (S5⊗S5)π+ is recursively isomorphic to second-order logic.

This was shown by Antonelli and Thomason (2002); proofs can also be found
in Kuhn (2004) and Belardinelli et al. (2018). The result is strengthened in Fritz
 46 Philosophy and Logic

(2020) to the logic of so-called products of two frames with a universal acces-
sibility relation. In contrast, it is shown in Fritz (forthcoming) that if S5 ⊗ S5 is
strengthened by the addition of the principle □p → ⊠p, the result of which is
there called a linear fusion, the resulting NML defines a class of frames whose
propositionally quantified modal logic is again decidable. In the case of episte-
mic logic, axiomatizability results have also been obtained for propositionally
quantified multimodal logics involving operators of public announcement or
common knowledge; see Belardinelli et al. (2016) and Belardinelli et al. (2018),
respectively.
Many problems concerning the axiomatizability of propositionally quan-
tified modal logics of classes of relational frames remain open. A number of
them are noted in Fritz (forthcoming, section 7), including the following (which
is only very partially addressed by the results just mentioned):

Open Question 3 Among NMLS Λ such that Λπ+ is decidable, which fusions
define classes of frames whose propositionally quantified modal logic is decid-
able?

A number of variations on this question can be formulated as well, where

decidability is replaced by recursive enumerability, or fusions are replaced by
products.
Reflecting the literature it summarizes, this section has focused mainly on
relatively coarse classifications of propositionally quantified modal logics in
terms of decidability, recursive axiomatizability, and recursive isomorphism
to second-order logic. It would be interesting to obtain a more fine-grained
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

classification:

Open Question 4 What are the computational complexity classes of the

various decidable NPQMLs?

2.7 Completeness Results

In the previous section, we saw that the propositionally quantified modal logics
of many natural classes of relational frames are not recursively enumerable. So,
for many propositional modal logics, the addition of propositional quantifiers
introduces essential incompleteness with respect to the model theory of rela-
tional frames. Completeness results with respect to classes of relational frames
are therefore much rarer in the presence of propositional quantifiers than in
their absence.
We have already discussed the most important such completeness result,
which is the completeness of SΠ 5At with respect to relational frames with
 Propositional Quantifiers 47

a universal accessibility relation. This shows that SΠ 5At is S5π+; see The-
orem 2.5.1. Recall that one proof of this result proceeds by a quantifier-
elimination argument. This technique can be developed further to apply to
extensions of S5π+, to prove the following result. It follows from a more
general result of Ding (2018), using algebraic models which we consider in
Section 3.2.

Proposition 2.7.1 Every NPQML which includes S5π+ is complete with respect
to the class of relational frames it defines.

It is worth noting that some of these normal extensions of S5π+ are not of
the form Λπ+, for any normal modal logic Λ. For example, we can consistently
extend S5π+ using the claim that there is a contingent truth: ∃p( p ∧ ¬□p).
This claim is not provable in Λπ+, for any consistent NML Λ: by Makinson
(1971), every consistent NML is valid on a one-element relational frame, in
which every truth is necessary. The characterization of the extensions of S5 by
Scroggs (1951) and Gärdenfors (1973) therefore requires further elaboration to
be adapted to the propositionally quantified case. For such an elaboration, see
Ding (2018).
Using algebraic methods, Ding (2021a) has also extended the completeness
result via quantifier elimination in the other direction, of weaker NMLs, and
shown the following result:

Proposition 2.7.2 KΠ D45BcAt is complete with respect to the class of rela-

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

tional frames it defines, namely, the class of relational frames with a serial,
transitive and Euclidean accessibility relation.

It is important to be clear that this result does not follow straightforwardly

from anything we have observed so far: It follows from Proposition 2.6.2 that
KD45π+ is decidable and so that there is some recursive way of axiomatizing
it. But this does not obviously mean that KΠ D45BcAt is such an axiomatiza-
tion. In general, there is no reason to expect that for every NML KΓ, if KΓπ+
is decidable then it is completely axiomatized by KΠ ΓBcAt. Indeed, a coun-
terexample can be found in which Γ is finite, as we now show using the case of
KT4Lem0 J1, which is also known as K3.1. The proof is based on an incom-
pleteness result for first-order modal logic based on S4M, which is the result of
adding M to S4, namely, KT4M. This result was already announced in Kripke
(1967); a proof can be found in Hughes and Cresswell (1996, pp. 265–270).
 48 Philosophy and Logic

Proposition 2.7.3 K3.1π+ is decidable, but not completely axiomatized by

KΠ 3.1BcAt (i.e., KΠ T4Lem0 J1BcAt).

Proof We note that ^∀p( p → □p) is valid on R(S4M): Call a world of a rela-
tional frame with a reflexive accessibility relation final if it can only access
itself. Consider any relational frame on which S4M is valid. By standard argu-
ments, it can be shown that the accessibility relation of such a frame is reflexive,
transitive, and relates every world to some final world; for details, see Cha-
grov and Zakharyaschev (1997, p. 82). From the last condition, the validity of
^∀p( p → □p) is immediate.
K3.1 is an extension of S4M, so it follows that K3.1π+ contains ^∀p( p →
□p) as well. In Section 3.5, we will be able to show that we cannot derive
this formula in KΠ 3.1BcAt; see Proposition 3.5.8. So KΠ 3.1BcAt is incom-
plete with respect to R(K3.1) (the class of relational frames it defines). And
since K3.1 is a normal extension of S4.3.1, it follows by Proposition 2.6.4 that
K3.1π+ is decidable. □

In light of this result, it is interesting to ask how one might provide perspic-
uous axiomatizations of recursively enumerable logics of the form Λπ+ for
NMLs Λ. As noted by Ding (2021a, p. 1196) the syntactic reductions involved
in quantifier-elimination arguments require very strong modal logics, and it
is not to be expected that this technique is applicable to all decidable logics.
Fortunately, another approach is available, which is the construction of canon-
ical models, used by Fine (1970) to establish the completeness of SΠ 5At with
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

respect to U. In unpublished work, Yipu Li and Yifeng Ding (pc) use this tech-
nique to provide a complete axiomatization of the propositionally quantified
modal logic of any class of relational frames which has finite diversity and
which is defined by a set of Sahlqvist formulas Γ, in terms of Γ, Bc, At, and
one further axiom.
There is an important observation to be drawn from the discussion in this
section and the previous one: for many NMLs Λ, it is highly ambiguous to
speak of “propositionally quantified Λ”. Taking the case of S4, namely, KT4,
as an example, we must distinguish between the weaker SΠ 4 and the stronger
S4π+. Since the former is recursively enumerable and the latter is not, there
are infinitely many NPQMLs linearly ordered in strength between these two log-
ics. All of these logics will be conservative extensions of S4, and could in this
sense be counted as “propositionally quantified S4”. In fact, there is even a
decidable NPQML extending S4π+ which is a conservative extension of S4. This
follows via the standard translation from the completeness of S4 with respect
to a single countable reflexive and transitive tree – see Blackburn et al. (2001,
 Propositional Quantifiers 49

pp. 353–355) – and a decidability result for monadic second-order logic on

such a tree, often called SωS, by Rabin (1969). As discussed by Zach (2004),
analogous observations can be made for a number of further classes of trees.
Zach notes that it would be interesting to provide perspicuous axiomatizations
of such logics. In similar cases, of multimodal temporal logics interpreted over
the natural numbers, axiomatizations of the resulting decidable propositionally
quantified modal logics are provided by Kesten and Pnueli (2002) and French
and Reynolds (2003).

2.8 Model Theory

An important class of techniques in the study of modal logic concern trans-
formations of relational frames which preserve the validity of formulas. Some
of these techniques continue to be applicable in the presence of propositional
quantifiers, while others become inapplicable. In this section, we consider some
of the most important examples.
We start with the technique of constructing generated subframes. As Black-
burn et al. (2001, p. ix) stress, one of the distinguishing features of modal logic
interpreted on relational frames is locality: the truth of a formula in a world w
only depends on the features of w and the worlds which can be reached from
w by finite paths along the accessibility relations. This means that a formula is
valid on a world in a frame just in case it is valid on that world in the subframe
which only contains these reachable worlds. This preservation result is unaf-
fected by the inclusion of propositional quantifiers, as noted by van Benthem
(1983, p. 187). The next definition and result make this observation precise.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Definition 2.8.1 For any relational frame F = ⟨W, R□ ⟩□∈O and world w ∈ W,
the point-generated subframe Fw is ⟨W′, R□′ ⟩□∈O , where

W′ is the set of v ∈ W such that there is a finite sequence w = w0, w1, . . . , wn =

v, where for all i < n, wi R□ wi+1 for some □ ∈ O, and
R□ is R□ ∩ W′ × W′, for all □ ∈ O.
′

Proposition 2.8.2 For every relational frame F, world w, assignment function

a, and φ ∈ LO ,

F, w, a ⊩ φ iff Fw, w, a′ ⊩ φ,

where a′ maps any p ∈ Φ to a( p) ∩ W′.

Proof By induction on the complexity of φ. □

 50 Philosophy and Logic

This result can be refined, by considering formulas of a certain modal depth n

(the maximal number of nested modal operators), and subframes which include
only the worlds reachable in n steps from the generating worlds. We won’t
need such a refined version here; details can be found in ten Cate (2006) and
Fritz (forthcoming). With generated subframes, it is easy to see that E, the class
of unimodal frames in which the accessibility relation is an equivalence rela-
tion, and U, the class of unimodal frames in which the accessibility relation is
universal, have the same propositionally quantified modal logic, that is, S5π+:

Proposition 2.8.3 L□ (E) = L□ (U).

Proof ⊆ is immediate, so assume that φ < L□ (E). Then there is a frame F in

E such that F ⊮ φ. Let w be a world witnessing F ⊮ φ; by Proposition 2.8.2,
Fw ⊮ φ. The worlds of Fw are the members of the equivalence class of w, so
Fw is in U. Thus φ < L□ (U). □

As a contrast, we now consider bounded morphisms, also known as pseudo-

epimorphisms or p-morphism. A bounded morphism is a certain kind of
function from one frame to another; if it is surjective (onto), it can be shown
that any propositional modal formula valid on the former frame is valid on the
latter frame. It turns out that this result breaks down when propositional quan-
tifiers are included. Thus, bounded morphisms do not play an important role in
propositionally quantified modal logic. We therefore skip a general definition
of this concept, and only consider a pair of frames which are easily shown to
be related by a bounded morphism: Let F and F ′ be unimodal relational frames
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

whose accessibility relations are universal, with F containing two worlds and
F ′ containing one world. There is just one function mapping the worlds of F to
the world of F ′, and it is a surjective bounded morphism. Consequently, every
propositional modal formula valid on F is valid on F ′. It is easy to see that this
does not hold for formulas with propositional quantifiers: ∃p( p ∧ ¬□p) is valid
on F but not valid on F ′.
The case of bounded morphisms illustrates that propositional modal lan-
guages become more expressive when propositional quantifiers are added. This
idea can be made more precise by considering the notion of a modally definable
class of frames: a class of relational frames C is definable in a given language L
if there is a set of formulas Γ of L which defines C. A well-known case of a class
of relational frames which is not definable in L□qf is the class of frames in which
the accessibility relation is irreflexive. This is easy to define with propositional
quantifiers, using the formula ∃p(□p ∧ ¬p). Yet, even with propositional quan-
tifiers, not every class of frames is definable. This follows already by cardinality
 Propositional Quantifiers 51

considerations, but we can also note a concrete example: by Proposition 2.8.3,

U is not definable in L□ , but it is easily defined in first-order logic (consid-
ering relational frames as models of first-order logic as in the discussion of
the standard translation in Section 2.6). A more detailed investigation of the
classes of relational frames definable in L□ , along with a prenex normal form
theorem, can be found in ten Cate (2006). For a closer look at propositionally
quantified formulas expressing first-order conditions on relational frames, see
Zhao (2023).
Bounded morphisms are closely related to bisimulations. A bisimulation is a
certain kind of relation between the worlds of two relational models (i.e., frames
with assignment functions) which guarantees that the same formulas of Lqf are
true in any two connected worlds. As in the case of bounded morphisms, this
guarantee does not extend to formulas of L involving quantifiers. However,
the definition of a bisimulation can be adapted in a natural way to take into
account the additional resources of propositional quantifiers. Such a variant
definition can be found in Fritz (forthcoming, section 3.3); a related notion is
presented in Belardinelli et al. (2018, section 5.1). It is also possible to provide
a different way of interpreting ∀p and ∃p on relational frames which makes
them invariant under bisimulation; see French (2006) and Steinsvold (2020).
The resulting notion of bisimulation quantifiers is connected to the study of the
modal µ-calculus, which also extends propositional modal logics by quantifier-
like connectives. These connectives are written as µp and νp, and interpreted
in terms of fixed points; for more, see Bradfield and Stirling (2007).
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

3 Beyond Relational Frames

3.1 Neighborhood Frames
We begin our exploration of further models for propositionally quantified
modal logics with a number of generalizations of relational frames. These gen-
eralizations are well known in the quantifier-free case; see Segerberg (1971)
and Hansson and Gärdenfors (1973) for illuminating general discussion. In
many cases, these models have not been considered in the presence of proposi-
tional quantifiers, and many fundamental questions remain open. We start with
neighborhood frames. Neighborhood frames are also known as Scott-Montague
frames, after Scott (1970) and Montague (1970). Chellas (1980) calls them
minimal models. A detailed discussion of them can be found in Pacuit (2017).
Neighborhood frames arise as a natural generalization of relational frames.
First, note that relational frames specify, for each world w, which proposi-
tions fall under any given modality □, namely those propositions (i.e., sets of
worlds) which include R□ (w), the set of worlds accessible via the relation for □.
 52 Philosophy and Logic

This puts various constraints on the set of propositions which fall under the
modality □ at any given world. For example, for every world w, the set of
propositions which include R□ (w) is closed under conjunction (i.e., intersec-
tion). The resulting constraints ensure that the logic of any class of relational
frames is normal.
However, in various applications, normality is undesired. For example, you
might endorse an epistemic theory on which knowledge is not closed under
conjunction, so that an agent might know p and know q, without knowing p∧q.
This is an instance of a much-debated principle of closure in epistemology; see
Hawthorne (2004) for in-depth discussion. Similarly, a deontic theory might
deny that an obligation to bring p about entails an obligation to bring p ∨ q
about, which is another entailment licensed by normal modal logics. A well-
known example for this is due to Ross (1941), who notes that it does not seem
to follow from an agent being obliged to post a letter that the agent is obliged
to post it or burn it.
To obtain models which don’t underwrite these inferences of normal modal
logics, we should allow for the set of propositions which are determined to fall
under □ not to obey the relevant closure conditions. The simplest way to intro-
duce this flexibility is to specify directly as part of the model which propositions
fall under □ at a given world. So, instead of an accessibility relation R□ , we may
simply use a function N□ which maps every world w to the set of propositions
which fall under □ at w. For historical reasons to do with connections to topol-
ogy, to which we return shortly, such a function N□ is known as a neighborhood
function, from which the terminology of neighborhood frames is derived.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Relational frames can straightforwardly be understood as a special case of

neighborhood frames, since every accessibility relation R□ corresponds to a
unique neighborhood function, namely the function which maps every world w
to the set of propositions which include R□ (w). When convenient, we will there-
fore treat relational frames as (special) neighborhood frames. It is also easy to
see that not every neighborhood function N□ can be obtained from an acces-
sibility relation R□ in the way just indicated: for example, let N□ (w) be empty
for every world w of a non-empty set W. Further, the resulting neighborhood
frame will fail to validate □⊤, and so its logic will not be normal.
Neighborhood frames generalize straightforwardly to polyadic modal oper-
ators: If ◦ is an n-ary modal operator, then the neighborhood function for ◦
maps each world w to a set of sequences of propositions of length n, contain-
ing those sequences of propositions which are to be related by ◦ at w. Similar
to relational frames in the unary case, many possible worlds model theories
of polyadic modal operators can be understood as special cases of neighbor-
hood frames. Examples include the many model theories of counterfactuals
 Propositional Quantifiers 53

based on possible worlds, such as those of Stalnaker (1968) and Lewis (1973).
For discussions of propositional quantifiers in such a context, see Lewis (1973,
pp. 45–47) and Besnard et al. (1997).
Having motivated neighborhood frames conceptually, we define them as
follows:

Definition 3.1.1 A neighborhood frame (for a modal signature σ = ⟨O, ρ⟩) is

a structure F = ⟨W, N◦ ⟩◦∈O such that W is a set and N◦ : W → P(P(W )ρ(◦) )
for every ◦ ∈ O. Assignment functions, truth, and validity are all defined as in
Definition 2.3.1, with the exception of the recursive clause for truth of modal
formulas, which is now as follows:

F, w, a ⊩ ◦φ1 . . . φn if and only if ⟨Jφ1 KF,a, . . . , Jφn KF,a ⟩ ∈ N◦ (w),

where n = ρ(◦) and Jφi KF,a = {v ∈ W : F, v, a ⊩ φi }.

We write N for the class of all neighborhood frames (for a given modal
signature).
Although neighborhood frames do not enforce the principles of normality,
they do impose constraints which go beyond those captured by classicality
as defined in Section 2.2. It turns out that the principal constraint imposed
on the logic of any class of neighborhood frames is that if it is valid on the
class that two formulas are materially equivalent, then these formulas can be
replaced in any context without changing the validity of the surrounding for-
mula. Adapting terminology from algebra, we call this feature congruentiality.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

(Recall from Section 2.2 that “classical” is sometimes used instead of “congru-
ential”, a usage which we don’t follow since we require the label “classical” for
a weaker condition.) Formally, this constraint can be formulated as follows:

Definition 3.1.2 For any set Λ ⊆ Lσ , we define the following condition:

(Cong) If ◦ ∈ O, n = ρ(◦), and φi ↔ ψi ∈ Λ for all i < n, then

◦φ0 . . . φn−1 ↔ ◦ψ0 . . . ψn−1 ∈ Λ.

Λ is a congruential (propositionally quantified) modal logic if it is a classical

(propositionally quantified) modal logic satisfying Cong.

Analogous to the normal case, we abbreviate “congruential (propositionally

quantified) modal logic” as “CML” (“CPQML”). Further, the standard result that
the modal logic of any class of neighborhood frames is congruential can be
extended to the propositionally quantified case, as the following proposition
notes:
 54 Philosophy and Logic

Proposition 3.1.3 For any class C of neighborhood frames, Lσ (C) is a CPQML.

Proof Analogous to the proof of Proposition 2.3.4, considering Cong instead

of K and Nec. □
Finally, we introduce – as in the normal case – labels for congruential logics
axiomatized by any given set of formulas:

Definition 3.1.4 For any set Γ ⊆ Lqf σ , the CML axiomatized by Γ, written Eσ Γ,

is the smallest CML including Γ. For any set Γ ⊆ Lσ , the CPQML axiomatized
by Γ, written EσΠ Γ, is the smallest CPQML including Γ.

Neighborhood frames relate to CMLs in very much the same way in which
relational frames relate to NMLs. We have already seen Proposition 3.1.3, which
corresponds to Proposition 2.3.4. E is also complete with respect to N, just as
K is complete with respect to R. Furthermore, Gerson (1975b) showed that
there are CMLs which are not sound and complete with respect to any class of
neighborhood frames, just as there are NMLs which are not sound and complete
with respect to any class of relational frames.
Turning to propositionally quantified modal logics, the corresponding com-
pleteness question with respect to arbitrary neighborhood frames appears not to
have been considered. Some basic observations are easily made. We start with
the two schematic principles discussed in Section 2.4. First, even though the
domain of propositional quantification is still independent of the world of eval-
uation, the Barcan formula Bc can be falsified on neighborhood frames. In fact,
the same applies to instances of the converse Barcan formula, which are deriv-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

able in NPQMLs. These observations mirror corresponding results concerning

neighborhood semantics for modal predicate logic; see Arló Costa (2002) and
Pacuit (2017, section 3.2). Second, the atomicity principle At (for any modality
□) is also not valid on all neighborhood frames.

Proposition 3.1.5 None of the following formulas is valid on N:

(i) □∀p□p → ∀p□□p (an instance of the converse of Bc)

(ii) ∀p□□p → □∀p□p (an instance of Bc)
(iii) ∃p( p ∧ ∀q(q → □( p → q))) (i.e., any instance of At)

Proof Let W = {0, 1}. For each formula, we construct a neighborhood frame
on W on which it can be falsified. For (i), interpret □ as negation, that is, let
N□ (w) = {x ⊆ W : w < x} for all w ∈ W; then (i) is false in every world. For
(ii), let N□ (0) = {∅, W} and N□ (1) = {{0}, {1}}; then (ii) is false in 1. For (iii),
let N□ (w) = ∅ for all w ∈ W; then (iii) is false in every world. □
 Propositional Quantifiers 55

With none of the formulas of Section 2.4 being valid on all neighborhood
frames, it is natural to wonder whether EΠ is complete with respect to N. But
this can easily be shown not to be the case. First, □⊤ → At is valid on N for
every unary modality □: if □⊤ is true in a world w, then {w} again witnesses the
truth of At. Second, we will see in the next sections that there are extensions
of EΠ which are normal, and therefore contain □⊤, but which do not contain
At. Any formula of the form □⊤ → At is therefore an example of the incom-
pleteness of EΠ with respect to N. In fact, L□ (N) is not recursively enumerable.
We return to this observation at the end of this section, when we will be able
to derive it from a corresponding result, due to Kremer (1997c), for a more
restricted class of frames which we will encounter shortly.
So far, we have considered two ways in which neighborhood frames offer
greater flexibility than relational frames: First, they weaken the logical con-
straints, allowing us to extend our scope from normal to congruential modal
logics. Second, they very naturally extend to polyadic modal operators. There
is a third way in which neighborhood frames offer greater flexibility than
relational frames: there are normal unimodal logics which are not sound and
complete with respect to any class of relational frames, but sound and complete
with respect to classes of neighborhood frames. This was shown by Gerson
(1975a, 1976). Nevertheless, there are also normal unimodal logics which are
not sound and complete with respect to any classes of neighborhood frames,
which was also shown by Gerson (1975b).
Neighborhood frames whose logic is normal are easily characterized using
the standard order-theoretic notion of a filter on a powerset. A set F ⊆ P(W )
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

is a filter just in case the following three conditions are satisfied:

(i) W ∈ F
(ii) If x ∈ F and y ∈ F, then x ∩ y ∈ F.
(iii) If x ∈ F and y ⊆ W, then x ∪ y ∈ F.

For any unary modal signature O, let a neighborhood frame F = ⟨W, N□ ⟩□∈O
be a filter neighborhood frame just in case for every □ ∈ O and w ∈ W, N□ (w)
is a filter. Let F be the class of filter neighborhood frames.
Proposition 3.1.6 A neighborhood frame F = ⟨W, N□ ⟩□∈O for a unary modal
signature O is a filter neighborhood frame if and only if the LO
qf
-formulas valid
on F form an NML (in which case the LO -formulas valid on F form an NPQML).

Proof The left-to-right direction follows from the fact that if F is filter neigh-
borhood frame, then W ∈ N□ (w), and (W \x) ∪ y, x ∈ N□ (w) only if y ∈ N□ (w),
for all w ∈ W and x, y ⊆ W. The right-to-left direction follows from the fact
that every NML contains □⊤, (□p ∧ □q) → □( p ∧ q), and □p → □( p ∨ q). □
 56 Philosophy and Logic

What kinds of NPQMLs are determined by classes of filter neighborhood

frames? Since the logic of any filter neighborhood frame is normal, it fol-
lows with the derivability of the converse Barcan formula in NPQMLs that any
instance of this schema is also valid on any class of filter neighborhood frames.
Furthermore, it is easy to see that every instance of At is also valid on any
such class: as in the case of relational frames, the relevant existential claim
is always witnessed by the singleton of the world of evaluation. This tells us
that despite Proposition 3.1.5 (iii), neighborhood frames cannot be used to con-
struct a model theory for any NPQML which does not include At. The matter is
different for the Barcan formula, as the following proposition shows:

Proposition 3.1.7 The following instance of Bc is not valid on F:

∀p□(□p → p) → □∀p(□p → p)

Proof Let F be the neighborhood frame ⟨N, N□ ⟩, where for each n ∈ N, N□ (n)
is the set of cofinite subsets of N, that is, the set of x ⊆ N such that N\x is finite.
Since the cofinite sets form a filter, F is a filter neighborhood frame.
J□p → pKF,a is cofinite, whether a( p) is cofinite or not. So F ⊩ ∀p□(□p →
p). However, if a( p) = N\{n}, then F, n, a ⊮ □p → p. So, F ⊩ ¬∀p(□p → p),
whence F ⊩ ¬□∀p(□p → p). Thus F ⊮ ∀p□(□p → p) → □∀p(□p → p). □

As we noted earlier, relational frames can be thought of as a special class

of (filter) neighborhood frames. Many other kinds of possible worlds model
theories of modal languages can be thought of as special classes of neighbor-
hood frames as well. Another example is the case of topologies, used as models
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

of propositional languages with one unary modal operator. Such models in

fact predate relational frames, going back to Tang (1938), McKinsey (1941),
and McKinsey and Tarski (1944). This topological interpretation is especially
interesting for us, since Kremer (1997c, 2018) has investigated them as models
of L□ .
For our purposes, it will suffice to introduce just the basic ideas. More
on topology in general can be found in textbooks like Dugundji (1966); for
an overview of topological modal logic see van Benthem and Bezhanishvili
(2007); and for a discussion of topological model in the context of neighbor-
hood frames see Pacuit (2017, section 1.4.1). Similar to the notion of a filter, a
topological space on a set W is a set T ⊆ P(W ) satisfying the following three
conditions:

(i) W ∈ T and ∅ ∈ T.
∩
(ii) If X is a finite subset of T, then X ∈ T.
∪
(iii) If X is a subset of T, then X ∈ T.
 Propositional Quantifiers 57

Such a topological space can be understood as determining a neighborhood

frame, namely the frame ⟨W, N□ ⟩, where for all x ⊆ W and w ∈ W:

x ∈ N□ (w) if and only if w ∈ s ⊆ x for some s ∈ T

(In topological terms, and adopting the algebraic perspective which we will
consider in the next section, this means that □ is interpreted as the interior
operation of the topological space.) Let T be the class of neighborhood frames
determined by topological spaces. All such frames are filter neighborhood
frames; moreover, the modal logic of T turns out to be exactly S4.
Kremer (1997c) introduces the label S4πt for L□ (T). His results entail the
following proper inclusions:

Proposition 3.1.8 SΠ 4At ⊊ S4πt ⊊ S4π+.

The improper inclusions are easy corollaries of results which we have

already noted. That S4πt is a proper subset of S4π+ follows from Kremer’s
observation that certain instances of Bc are not valid on T. That SΠ 4At is a
proper subset of S4πt follows from the following result, which Kremer estab-
lishes by reducing second-order arithmetic to S4πt, together with the obvious
fact that SΠ 4At is recursively enumerable:

Proposition 3.1.9 S4πt is not recursively enumerable.

This is the result we can use to establish that L(N) is not recursively axioma-
tizable, since L(T) is reducible to L(N), in the sense that there is a computable
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

function f : L → L such that φ ∈ L(T) if and only if f (φ) ∈ L(N). In fact, L(T)
is reducible to the logic of any class of neighborhood frames which includes
all (neighborhood frames determined by) topological spaces. The class of filter
neighborhood frames F is another example of such a class we have encountered.
The reduction itself is easy to define. First, we introduce a formula which
encodes that the axiomatic principles of S4 hold on any interpretation of the
proposition letters:

(ρS4 ) ∀p∀q(□⊤ ∧ (□( p → q) → (□p → □q)) ∧ (□p → p) ∧ (□p → □□p))

In the context of congruential modal logics, □⊤ corresponds to the rule of

necessitation; the four conjuncts therefore correspond to the rule of necessi-
tation and the axioms K, T and 4. Let ·∗ be the recursive mapping from L□ to
L□ whose only nontrivial condition is:

(□φ)∗ := □(ρS4 → φ∗ )
 58 Philosophy and Logic

Finally, let φ† := ρS4 → φ∗ . We will show that ·† reduces L(T) to L(N).

To do so, we require a way of restricting any neighborhood frame to worlds
which validate ρS4 . So, for any neighborhood frame F = ⟨W, N⟩, let FS4 be the
neighborhood frame ⟨W′, N′⟩, such that:

W′ = JρS4 KF
N′(w) = {x ∩ W′ : x ∈ N(w)}

The central claims needed to establish the reduction are collected in the
following lemma:

Lemma 3.1.10 Let F = ⟨W, N⟩ be a neighborhood frame.

(i) F ⊩ ρS4 if and only if F is in T.

(ii) For all w ∈ W′ and assignment functions a, F, w, a ⊩ φ∗ if and only if
FS4 , w, a′ ⊩ φ, where a′( p) = a( p) ∩ W′, for all p ∈ Φ.
(iii) FS4 ⊩ ρS4 , whence FS4 is in T (by i).

Proof (i) follows from Pacuit (2017, p. 24, proposition 1.1).

(ii) By induction on the complexity of φ.
(iii) Considering the four conjuncts individually, using ii. □
Although this won’t be needed in the following, it is worth noting that ρS4
is valid on a neighborhood frame just in case S4 is valid on it. With the three
results of Lemma 3.1.10, the reduction follows:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Proposition 3.1.11 For every class C of neighborhood frames which includes

T, L(T) is reducible to L(C).

Proof We show, for any φ ∈ L□ , that φ ∈ L(T) if and only if φ† ∈ L(C).

If φ < L(T), then there is a neighborhood frame F in T (i.e., a topological
space), on which φ is not valid. By Lemma 3.1.10(i), ρS4 is valid on F. Con-
sequently, FS4 is F, and so by Lemma 3.1.10(ii), φ∗ fails to be valid on F as
well. Since C includes T, φ† < L(C).
If φ† < L(C), then there is a neighborhood frame F, world w, and assign-
ment function a such that F, w, a ⊮ φ† . First, it follows that F, w, a ⊩ ρS4 ,
which means that w is contained in FS4 , and with Lemma 3.1.10(iii), that FS4
is in T. Second, F, w, a ⊮ φ∗ ; since w is contained in FS4 , it follows with
Lemma 3.1.10(ii) that FS4, w, a′ ⊮ φ. Thus φ < L(T). □

Using Kremer’s result stated earlier as Proposition 3.1.9, we obtain:

Corollary 3.1.12 Neither L(N) nor L(F) is recursively axiomatizable.

 Propositional Quantifiers 59

As Kremer notes, it is an open question whether his result can be strength-

ened to recursive isomorphism to second-order logic:

Open Question 5 Is S4πt recursively isomorphic to second-order logic?

Given our observations here, a positive answer will settle the matter as well
for L(N) and L(F). If the answer is negative, separate analogous questions arise
for these logics.

3.2 Complete Boolean Algebras

Our next generalization of relational frames exchanges the representation of
propositions by sets of possible worlds with a more abstract, algebraic, struc-
ture. To motivate this idea, it is helpful to consider a variant definition of
neighborhood frames, which is mathematically equivalent to the one presented
earlier. Recall that neighborhood frames simply stipulate, for each world and
modality of arity n, which sequences of propositions (sets of worlds) of length
n are related by the modality at that world. Equivalently, we can think of a
neighborhood frame as stipulating, for each modality of arity n and sequence
of propositions of length n, at which worlds these propositions are related by
that modality. The set of these worlds can be thought of as the proposition that
these proposition are related by that modality. So, formally, a neighborhood
function N◦ for a modality ◦ of arity n can also be represented as a function
∗◦ mapping any sequence of n propositions to a proposition. It is easy to see
that every neighborhood function N◦ determines a unique function ∗◦ , and vice
versa. To illustrate this informally, if □ is interpreted as being known (by a cer-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

tain agent), then the relevant function ∗□ can be understood as mapping every
proposition x to the proposition that x is known.
A neighborhood frame can therefore equivalently be presented as a structure
⟨W, ∗◦ ⟩◦∈O , where W is a set and ∗◦ is a function from P(W )ρ(◦) to P(W ), for
all ◦ ∈ O. On this presentation, the evaluation clause for modal operators turns
into the following condition:

F, w, a ⊩ ◦φ1 . . . φn if and only if w ∈ ∗◦ (⟨Jφ1 KF,a, . . . , Jφn KF,a ⟩).

In fact, on this presentation of neighborhood frames, it makes sense not to take

the relation of truth ⊩ as basic and define the interpretation function J·K◦ in
terms of it, but to reverse the order: That is, it makes sense to recursively define
a function J·K◦ which assigns to each formula the proposition it expresses, and
then derive the truth relation ⊩ by saying that F, w, a ⊩ φ if and only if w ∈
JφKF,a . The clause for a modality ◦ then becomes a simple application of the
function ∗◦ used to interpret it:
 60 Philosophy and Logic

J◦φ1 . . . φn KF,a = ∗◦ (⟨Jφ1 KF,a, . . . , Jφn KF,a ⟩)

This makes clear that neighborhood semantics is in one sense a maximally

permissive generalization of relational frames: modal operators are interpreted
using completely arbitrary functions on propositions.
The other truth-conditional clauses can be given a corresponding pres-
entation in terms of the interpretation function, with Boolean connectives
interpreted using corresponding set-theoretic operations. If we used ¬ and ∧
as the primitive Boolean connectives, these would be the operations of com-
plement relative to the set of worlds, and intersection, respectively. The choice
of ⊥ and → as primitive makes the corresponding operations unfortunately
somewhat less intuitive:

JpKF,a = a( p)
J⊥KF,a = ∅
Jφ → ψKF,a = (W \JφKF,a ) ∪ JψKF,a

We focus just on the quantifier-free setting for the moment, and get back to
quantifiers shortly.
We have arrived at an algebraic perspective on neighborhood frames. (For
more on algebraic approaches to logic, see Halmos and Givant (1998), Dunn
and Hardegree (2001), and Rasiowa and Sikorski (1963).) Let a powerset alge-
bra be a structure A = ⟨A, 0, ⊐⟩, where A = P(W ) for some set W, 0 = ∅, and
x ⊐ y = (W \x) ∪ y for all x, y ⊆ W. (Here, “0” need not stand for the number
zero; context will always disambiguate between the algebraic and numerical
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

reading.) From an algebraic perspective, we can think of a neighborhood frame

as extending a powerset algebra by arbitrary n-ary functions on the underlying
powerset. Once the set-theoretic operations are explicitly represented in the
algebra, the set-theoretic nature of the elements of the algebra becomes imma-
terial. Instead of neighborhood frames, we can therefore equivalently consider
any algebra isomorphic to a powerset algebra, extended by arbitrary polyadic
functions on the underlying set. This raises the question: which structural con-
straints are imposed by requiring an algebra ⟨A, 0, ⊐⟩ to be isomorphic to a
powerset algebra? The well-known answer to this question is the following:

Proposition 3.2.1 An algebra ⟨A, 0, ⊐⟩ is isomorphic to a powerset algebra if

and only if it is a complete atomic Boolean algebra.

For a proof, and rigorous definition of the relevant algebraic concepts, see
a textbook treating Boolean algebras, such as Davey and Priestley (2002) and
Givant and Halmos (2009). Here, we will have to limit ourselves to a brief dis-
cussion of these concepts. First, we can think of a Boolean algebra as an algebra
 Propositional Quantifiers 61

A = ⟨A, 0, ⊐⟩, where A is a set, 0 ∈ A, and ⊐ a binary function on A, such that

on the present interpretation of Boolean formulas on such an algebra, any two
formulas which are equivalent by classical propositional logic are interpreted
as the same element of the algebra. As a consequence, we can define, in any
Boolean algebra, operations −, ⊔, and ⊓ which interpret ¬, ∨, and ∧, as well as
a distinguished element 1 interpreting ⊤. For example, − is defined as mapping
x to x ⊐ 0, and 1 is defined as −0.
It remains to define completeness and atomicity. For completeness, we first
define an order ≤ among elements of A, stipulating that x ≤ y if and only
if x = x ∧ y. This order is sometimes called entailment, but note that it is a
relation among elements of the algebra, which represent propositions, and not
among formulas. Intuitively, it orders propositions according to strength, so
that x ≤ y just in case x is at least as strong as y. It can be shown that ≤ is a
partial order, which means that it is reflexive, transitive, and antisymmetric.
(See Davey and Priestley (2002) for definitions of such basic order-theoretic
notions.) Furthermore, it can be shown that ≤ forms a lattice, which means
that any two elements x and y have a least upper bound and a greatest lower
bound. A is complete just in case every set X ⊆ A has a least upper bound and
a greatest lower bound. For atomicity, an element a of A is defined as an atom
just in case a is distinct from 0, and there is no other element x ∈ A such that
x ≤ a. A is atomic just in case for every x , 0, there is an atom a ≤ x.
The requirement for A to be Boolean ensures that every theorem of classical
propositional logic is always interpreted as the top element 1, under any inter-
pretation of the proposition letters, which is the algebraic correspondent of the
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

notion of being valid on a frame. Requiring A to be Boolean therefore plays

a crucial role in ensuring the validity of tautologies. (Although see Section 4
for less restrictive ways of achieving this end.) It turns out that the constraints
of completeness and atomicity are not required to guarantee the principles
of CMLs. So, using the algebraic perspective, we can generalize the class of
neighborhood frames simply by dropping the requirement for the underlying
Boolean algebra to be complete and atomic. More formally, we therefore make
the following definition:

Definition 3.2.2 A Boolean algebra expansion (for a modal signature σ =

⟨O, ρ⟩) is a structure A = ⟨A, 0, ⊐, ∗◦ ⟩◦∈O such that ⟨A, 0, ⊐⟩ is a Boolean alge-
bra and ∗◦ : Aρ(◦) → A for every ◦ ∈ O. An assignment function is a function
a : Φ → A. The interpretation of quantifier-free formulas is defined recursively
using the following clauses:
 62 Philosophy and Logic

JpKA,a = a( p)
J⊥KA,a = 0
Jφ → ψKA,a = JφKA,a ⊐ JψKA,a
J◦φ1 . . . φn KA,a = ∗◦ (⟨Jφ1 KA,a, . . . , Jφn KA,a ⟩)

Validity is defined by letting A ⊩ φ if JφKA,a = 1 for all a : Φ → A.

The label of “Boolean algebra expansion” is taken from Ding and Holliday
(2020). We abbreviate it using “BAE” in the following. As in the case of neigh-
borhood frames, the logic of every class of BAEs is a CML. However, unlike the
case of neighborhood frames, the converse holds for BAEs: every CML is the
logic of some class of BAEs. This is straightforward to show, using the standard
construction of a Lindenbaum–Tarski algebra. In the terminology of Chagrov
and Zakharyaschev (1997), BAEs therefore provide an “adequate” model theory
of CMLs.
We are finally ready to return to languages involving propositional quan-
tifiers. First, note that the evaluation clause of propositional quantifiers in
neighborhood frames can equivalently be formulated by stating that the propo-
sition expressed by a universal propositional quantification is the conjunction
(intersection) of its instances:
∩
J∀pφKF,a = x∈P (W ) JφKF,a[x/p]

In a powerset P(W ), the intersection of any set X ⊆ P(W ) is its greatest lower
bound (under the entailment order ≤). This fits well with the common idea that
universal quantifications are conjunctions of their instances, and the fact that
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

the binary conjunction operation ⊓ in Boolean algebras maps any two elements
to their greatest lower bound. This suggests that algebraically, propositional
quantifiers can be interpreted using greatest lower bounds as well. In the fol-
∧
lowing, we notate greatest lower bounds using . We must, of course, ensure
that the relevant greatest lower bounds always exist. The simplest way to do
so is to assume that the algebra is complete, in the sense defined earlier. (We
consider a weaker, but more complicated, requirement in Section 3.4.) With
this, Definition 3.2.2 is straightforwardly extended to propositional quantifiers,
at least for complete BAEs, which we call CBAEs. (We extend properties of Bool-
ean algebras to BAEs in the obvious way; for example, a BAE is complete just in
case its underlying Boolean algebra is complete.)

Definition 3.2.3 For any CBAE A = ⟨A, 0, ⊐, ∗◦ ⟩◦∈O , the interpretation of

formulas of Definition 3.2.2 is extended to Lσ by the following recursive clause:
∧
J∀pφKA,a = x∈A JφKA,a[x/p]
 Propositional Quantifiers 63

Again, this is a straightforward generalization of neighborhood frames, with

neighborhood frames falling out as the special case of atomic CBAEs. Fur-
thermore, CBAEs suffice to guarantee congruentiality, even in the presence of
propositional quantifiers:

Proposition 3.2.4 For any class C of CBAEs, Lσ (C) is a CPQML.

Proof The argument is routine except for the case of UG. For this case, we
∧
appeal to the following law of complete Boolean algebras: y∈Y x ⊐ y = x ⊐
∧
Y. This follows from the infinite distributivity laws of complete Boolean
algebras; see (the dual of) Givant and Halmos (2009, p. 47, lemma 3). □

Requiring algebras to be complete makes interpreting propositional quan-

tifiers easy. But it breaks the simple and general completeness proofs for CMLs
with respect to BAEs: typically, Lindenbaum–Tarski algebras are incomplete,
and it is not obvious how to complete them without affecting the interpretation
of propositional quantifiers, as discussed by Holliday (2019). In fact, there are
not just CPQMLs, but also NPQMLs which are incomplete with respect to the class
of CBAEs they define, and therefore not determined by any class of CBAEs. It is
worth noting that this does not straightforwardly follow from the fact, shown
by Litak (2004), that there are NMLs which are not determined by any class of
CBAEs: recall from Proposition 2.3.9 that the extension of such an NML need not
be conservative, and so need not itself fail to be determined by any class of of
CBAEs.
The existence of NPQMLs which are incomplete with respect to any class of
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

CBAEs can be shown using two results of Ding (2021a,b). First, Ding (2021b,
p. 53, theorem 3.2.16) shows that there are instances of the following schema
which are not derivable in KΠ D45:

(4∀ ) ∀p□φ → □∀p□φ

(We return to the kinds of models used in this proof in Section 3.6.) Second,
Ding (2021a, p. 1162, theorem 3.1) shows that all instances of 4∀ are valid on
the class of CBAEs defined by KΠ D45. (Moreover, Ding (2021a) shows that
KΠ D4∀ 5 is sound and complete with respect to this class.) Therefore:

Proposition 3.2.5 KΠ D45 is not the logic of any class of CBAEs.

Most of the existing work on CBAEs is concerned with normal logics, despite
the fact that CBAEs constitute a natural model theory for the wider class
of congruential logics. As in the case of neighborhood frames, BAEs whose
 64 Philosophy and Logic

logic is normal have to satisfy certain constraints. In particular, since □⊤ and

□( p ∧ q) ↔ (□p ∧ □q) are contained in every NML for every modality □, such
BAEs must satisfy the following conditions, for every modality □:

(Normality) ∗□ 1 = 1
(Additivity) For all elements x and y, ∗□ (x ⊓ y) = (∗□ x ⊓ ∗□ y).

Conversely, these constraints suffice to guarantee normality, as the next propo-

sition notes. (BAEs satisfying these constraints are often called “Boolean alge-
bras with operators”, for example, by Blackburn et al. (2001, p. 277). In the
present context, this is confusing, since BAEs are literally Boolean algebras
with arbitrary extra operators (functions), and so arguably most deserving of
this label.)

Proposition 3.2.6 A BAE A for a unary modal signature O satisfies normality

and additivity for every □ ∈ O if and only if the LO
qf
-formulas valid on A form
an NML (in which case the L -formulas valid on A form an NPQML).
O

Proof The left-to-right direction follows from the fact that if A satisfies nor-
mality and additivity, then its validities contain □⊤ (and so the LO
qf
-formulas
valid on A are closed under necessitation), and the axiom K□ . The right-to-left
direction follows from the fact that every NML contains □⊤ and □( p ∧ q) ↔
(□p ∧ □q). □
In the context of CBAEs, it is natural to consider a generalization of additivity
to arbitrary greatest lower bounds:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

∧ ∧
(Complete Additivity) For every set of elements X, ∗□ X= {∗□ x : x ∈ X}.

In certain cases, this stronger condition leads to additional validities. This can
be illustrated using results by Ding (2021a), who shows that KΠ D4∀ 5 – which
is complete with respect the class of CBAEs it defines – is incomplete if this
class is restricted to CBAEs whose modal operation is completely additive. This
is witnessed by certain instances of the Barcan formula. With Bc, 4 ∀ becomes
derivable; in fact, Ding shows:

Proposition 3.2.7 KΠ D45Bc is sound and complete with respect to a class of

CBAEs whose modal function satisfies complete additivity.

Ding also considers At, and shows that At becomes valid if, in addition to
complete additivity, we restrict ourselves to atomic CBAEs. Such CBAEs are sim-
ply relational frames, algebraically presented. Consequently, Proposition 2.7.2
shows the soundness and completeness of KΠ D45BcAt with respect to the
 Propositional Quantifiers 65

class of such CBAEs it defines. Conversely, At can be falsified on nonatomic

CBAEs, which shows that At is not derivable from KΠ D45Bc.
Corresponding results on At have been obtained in the stronger setting of S5,
in which they take on an especially simple form. Recall from Section 2.5 that
SΠ 5At is sound and complete with respect to the class of relational frames with
a universal relation. Algebraically, a universal accessibility relation determines
an operation ∗□ on the power set P(W ) which maps 1 to itself, and every other
proposition to 0. This treatment of □ is straightforwardly extended to arbitrary
BAEs, and it is easy to see that any such function ∗□ satisfies normality and com-
plete additivity. Such algebras were already considered by Lewis and Langford
(1959 [1932], p. 501), where they are attributed to Paul Henle. We therefore
define:

Definition 3.2.8 A Henle algebra is a BAE ⟨A, 0, ⊐, ∗□ ⟩ such that for all x ∈ A:

1
 if x = 1
∗□ (x) =
0 otherwise


Complete and atomic Henle algebras are simply relational frames with a uni-
versal accessibility relation, algebraically presented. We therefore know from
Theorem 2.5.1 that At is valid on such Henle algebras. In fact, we know that the
logic of such Henle algebras is exactly SΠ 5At. But what happens in nonatomic
complete Henle algebras? Bull (1969, p. 260) observed that At is not valid on
any complete Henle algebra which is not atomic. However, it is easy to see
that the axioms T and 5 are valid on all complete Henle algebras, from which
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

it follows that SΠ 5 is sound with respect to complete Henle algebras. Along

these lines, Bull concludes:

Proposition 3.2.9 At is not a theorem of SΠ 5.

This result, together with the completeness of SΠ 5At with respect to com-
plete atomic Henle algebras, suggests two very natural conjectures: first, that
completeness is preserved when At is removed and nonatomic algebras are
admitted, and second, that completeness is preserved when At is replaced by
its negation ¬At, and only atomless algebras are considered. These very natural
conjectures were only recently confirmed by Holliday (2019), who established
the following companions to Theorem 2.5.1:

Proposition 3.2.10 SΠ 5 is sound and complete with respect to complete Henle

algebras, and SΠ 5¬At is sound and complete with respect to complete atomless
Henle algebras.
 66 Philosophy and Logic

The results of Ding and Holliday concern two very strong NMLs, namely
KD45 and S5. This is no accident, since the proof methods used by both authors
heavily depend on the quantifier elimination technique of Kaplan (1970b) and
Fine (1970). Fine had already extended these methods to SΠ 5, which he used to
establish the decidability of this logic. These methods also allow Ding (2021a)
to establish decidability for all of the particular extensions of KΠ D45 discussed
by him. However, it is far from clear that these quantifier elimination results
can be extended to weaker NMLs. Consequently, many open questions remain,
many of which will likely require very different techniques. A basic example
is the following:

Open Question 6 Is the propositionally quantified modal logic of the class of

all CBAEs recursively enumerable?

Depending on the answer to the this question, many variants and further
questions arise. For example, if the answer to this question is positive, is EΠ
complete with respect to the class of all CBAEs? Is the answer the same if we
consider unary modal signatures, and CBAEs whose modal functions satisfy nor-
mality and (complete) additivity? Can the logic of such restricted classes of
CBAEs be completely axiomatized using KΠ ? Are the answers to any of these
questions dependent on the modal signature? For further discussion of open
questions, see Holliday (2019) and Ding (2021a).
Before moving on to another topic, it is worth noting an alternative approach
to CBAEs. Recall how neighborhood frames can be considered as the special case
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

of atomic CBAEs. (Similarly, relational frames can be considered as the special

case of atomic CBAEs with modal functions satisfying normality and complete
additivity.) This follows from Proposition 3.2.1, which in algebraic termi-
nology is a representation theorem, showing that complete atomic Boolean
algebras can be represented using powerset algebras. There is a more general
representation theorem, which shows that all complete Boolean algebras can be
represented as the completions of arbitrary partial orders, or, equivalently, as
the regular open subalgebras of arbitrary topological spaces. The former rep-
resentation theorem is often attributed to Tarski (1935), and the latter to Tarski
(1937) and MacNeille (1937).
In the representation of complete Boolean algebras using partial orders, the
elements of the order can be understood as (possibly incomplete) possibilities,
and so as a generalization of the concept of possible worlds. Furthermore, the
evaluation of formulas can be formulated in terms of being made true by a pos-
sibility, and even the accessibility relations of relational frames can be applied
to possibilities to interpret unary modal operators when considering algebras
 Propositional Quantifiers 67

whose modal functions satisfy normality and additivity. (Along with neigh-
borhood frames, these models therefore exemplify the earlier claim that the
model-theoretic ideas of possible worlds and accessibility relations are entirely
separable.) Such a possibility semantics for propositional modal languages
was proposed by Humberstone (1981); the general theory and connection to
algebraic models is developed in detail by Holliday (forthcoming). These mod-
els can be extended to provide an elegant model theory for propositionally
quantified modal logic; see Holliday (2021, section 5.1) for an illustration.

3.3 Pointed Frames, and Matrices

There is one more way in which we can generalize the model theories and
modal logics we have considered so far. Consider first again the simplest case
of unimodal relational frames; we return to BAEs in a moment. In such frames,
worlds play two roles: First, and most obviously, the worlds accessible from a
given world play the role of representing the ways things could be (according to
the modality under consideration, from the perspective of the given world). For
example, if □ is interpreted as being known by a particular agent, the worlds
accessible from the world of evaluation represent the ways things could be,
given the agent’s knowledge at the given world. But worlds also play a second
role, which is to represent the way things could logically be. In the example just
mentioned, a frame might contain a world which can only access itself, as well
as a world which can access itself as well as another world. As a consequence,
the logic of the frame allows both for the agent to be omniscient, as well as for
them not to be omniscient, in the sense that neither ∀p( p → □p) nor its negation
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

is included in the logic of the frame.

These two roles played by worlds are already partly separated in relational
frames by the accessibility relation, since not every world is possible (from
the perspective of a given world) according to the standard of the relevant
object-language modality. But in relational frames, every world must serve to
constrain the logic, since validity is defined as truth in all worlds of the frame
(under every assignment function). In this sense, all worlds of relational frames
are logically possible worlds. It is therefore impossible to use worlds to con-
strain the behaviour of modal operators without also constraining the logic. For
example, if we want to represent an agent not knowing that knowledge is fac-
tive, that is, if we want a world to verify the principle ¬□∀p(□p → p), we have
to admit a world which is not accessible from itself. But this means that the
factivity of knowledge, ∀p(□p → p), will not be included in the logic of the
frame. More abstractly, the requirement for all worlds of a frame to count as
logically possible worlds forces logics of relational frames to be closed under
necessitation.
 68 Philosophy and Logic

However, these limitations are easily overcome: In a given frame, we can

distinguish certain worlds as logically possible, simply by including in the defi-
nition of a frame a subset of worlds which are to count as logically possible.
Then we can, for example, consider a class of frames with distinguished worlds
in which every distinguished world can access itself, without imposing the same
requirement on all worlds. On the resulting class of frames, ∀p(□p → p) is
valid, without □∀p(□p → p) thereby becoming valid as well. This separation
of the two roles of possible worlds was in fact present in the original definition
of relational frames of Kripke (1963a). Kripke requires there to be just a single
distinguished world, but this choice is easily seen to be immaterial when we
consider logics of classes of frames: The logic of any class C of frames with
a set of distinguished worlds is the same as the logic of the class of frames
obtained from frames in C by restricting the set of distinguished worlds to a
singleton. Here we follow Kripke and distinguish a single world. We call the
resulting frames pointed.
The addition of a distinguished world extends straightforwardly to the case
of neighborhood frames. For brevity, we introduce pointed relational and
neighborhood frames together:

Definition 3.3.1 A pointed relational/neighborhood frame is a structure F =

⟨W, w0, X◦ ⟩◦∈O such that ⟨W, X◦ ⟩◦∈O is a relational/neighborhood frame, and
w0 ∈ W. Assignment functions and truth are defined as in Definition 2.3.1/3.1.1.
Validity on a frame is defined as follows:

F ⊩ φ if F, w0, a ⊩ φ for all a : Φ → P(W ).

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

What kind of logics do we get from classes of pointed frames? We have

already seen that in the relational case, they need not be normal, as they
need not be closed under necessitation. Similarly, in the neighborhood case,
they need not be congruential. However, the logic of any class of pointed
relational/neighborhood frame must include some normal/congruential modal
logic, namely the logic of the class of relational/neighborhood frames which
we get from omitting the distinguished worlds. Consequently, the logic of any
class of pointed frames must contain the smallest normal/congruential logic.
The logic of any class of pointed frames must also be classical. These two
constraints lead to the following definition:

Definition 3.3.2 A set Λ ⊆ Lσ is a quasi-normal/quasi-congruential (proposi-

tionally quantified) modal logic if it is a classical (propositionally quantified)
modal logic including some normal/congruential (propositionally quantified)
modal logic.
 Propositional Quantifiers 69

We abbreviate “quasi-normal/quasi-congruential (propositionally quan-

tified) modal logic” as “QNML/QCML(QNPQML/QCPQML)”. As discussed by
Segerberg (1971), classes of pointed relational/neighborhood frames give us
QNMLs/QCMLs, and we now note that the analogous claim holds in the presence
of propositional quantifiers:

Proposition 3.3.3 For any class of pointed relational/neighborhood frames C,

Lσ (C) is a QNPQML/ QCPQML.

Proof That Lσ (C) includes an NPQML/CPQML follows from Propositions 2.3.4

and 3.1.3. Classicality is routine, considering each of the four conditions. □

Pointed frames are often not considered since many of the formal results
on relational and neighborhood frames transfer straightforwardly to pointed
frames. But there are many applications of modal logics in which the addi-
tional flexibility of pointed frames is important. An example from philosophy
is the logic of indexical operators, such as “now” and “actually”, investigated
by Kaplan (1989 [1977]) and Crossley and Humberstone (1977). An example
from mathematics is the QNML GLS, which Solovay (1976) showed to capture
the notion of provability from the axioms of Peano arithmetic according to true
arithmetic, on the provability interpretation of modal logic; this is discussed in
more detail in Boolos (1985). In the context of neighborhood frames, pointed
frames also allow the construction of model theories of some lesser known
Lewis systems, including S2 and S3, since the models of Kripke (1965) for
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

these systems can be formulated as pointed neighborhood frames.

In many cases, the problem of the axiomatizability of a propositionally
quantified modal logic of a class of pointed frames can be reduced to the
known problem of the axiomatizability of the logic of a corresponding class of
unpointed frames. Exemplarily, we consider the case of (any class of) pointed
relational frames in which for some fixed modality □, R□ is universal. First, let
O be a unary modal signature, and define O′ to be the signature obtained from O
by adding one further unary operator △. For any pointed relational frame F for
σ, let F ′ be the relational frame for σ ′ obtained by removing the distinguished
world w0 from F and adding an accessibility relation R △ such that, in general,
wR0 v if and only if v = w0 . We show that there are simple recursive mappings
which reduce the logics of F ′ and F to each other. One direction is straight-
forward, as any formula φ ∈ Lσ is valid on F just in case △φ is valid on F ′.
′
For the converse direction, define φ ′ for every φ ∈ Lσ recursively using the
single nontrivial clause (△φ)′ := □( p0 → φ ′), where we assume that p0 does
not already occur in any formula φ under consideration. We can now show that
 70 Philosophy and Logic
′
φ ∈ Lσ is valid on F ′ just in case ∀p0 (Q( p0 ) ∧ p0 → □φ ′) is valid on F, where
Q is used as defined in Section 2.4.
Consider now the case of BAEs. Recall that BAEs can be understood as gen-
eralizing neighborhood frames by dropping the requirement of the (implicit)
underlying Boolean algebra to be complete and atomic. Such algebraic struc-
tures therefore need not contain atoms, which would correspond to possible
worlds. We can therefore not easily adapt the idea of distinguishing a sin-
gle world. But as noted earlier, instead of pointed frames we can also work
with frames with sets of distinguished elements. Sets of worlds correspond to
elements of a Boolean algebra, so we can extend frames with distinguished ele-
ments to BAEs by distinguishing one element x0 of the algebra. Validity is then
defined by letting a formula φ be valid just in case its interpretation is entailed
(according to the Boolean order ≤) by x0 , under every assignment function.
The distinguished element x0 determines the set of propositions ↑x0 it entails.
A formula is then valid just in case, for every assignment function, it is inter-
preted as a member of this set. By construction, ↑x0 is guaranteed to be a filter,
where the notion of a filter is generalized to arbitrary Boolean algebras as fol-
lows: F ⊆ A is a filter of ⟨A, 0, ⊐⟩ if 1 ∈ F, x ⊓ y ∈ F for any x, y ∈ F, and
x ⊔ y ∈ F for any x ∈ F, y ∈ A. This ensures that the quantifier-free validities
form a classical modal logic. In fact, ↑x0 is by definition a principal filter, which
means that it is not just closed under binary greatest lower bounds, but arbitrary
∧
greatest lower bounds: X ∈ ↑x0 for all X ⊆ ↑x0 . In the quantifier-free case,
this additional requirement is not needed to ensure that the resulting logic is
classical. So, in the quantifier-free case, the notion of a pointed neighborhood
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

frame can be generalized to the notion of a BAE extended by a filter F, where a

formula counts as valid just in case its interpretation is a member of F, under
every assignment function. This leads us to the notion of a matrix, which goes
back to the very beginnings of work in symbolic logic, as noted by Łukasiewicz
and Tarski (1930). In the context of modal logics, matrices were used as early
as McKinsey (1941).

Definition 3.3.4 A (C)BAE-matrix is a structure M = ⟨A, 0, ⊐, F, ∗◦ ⟩◦∈O such that

⟨A, 0, ⊐, ∗◦ ⟩◦∈O is a (C)BAE, and F is a filter of ⟨A, 0, ⊐⟩. Assignment functions
and evaluation are defined as in Definition 3.2.2. Validity on a matrix is defined
as follows:

A ⊩ φ if and only if JφKA,a ∈ F.

From this definition, it is easy to see that the notion of a BAE can be
understood as the special case of a BAE-matrix with the singleton filter {1}.
 Propositional Quantifiers 71

The logic of any class of BAE-matrices is a QCML. In fact, as in the case of

BAEs and CMLS, the logics of classes of BAE-matrices are all and only the QCMLs;
see Hansson and Gärdenfors (1973). Turning to propositionally quantified log-
ics, we restrict ourselves to CBAEs as in the previous section, to guarantee that
greatest lower bounds are always defined. This does not suffice, however, to
ensure classicality, as we now demonstrate. The proof employs the notion of
an ultrafilter, which is a filter which contains, for every element x, either x or
−x, but not both.

Proposition 3.3.5 There is a CBAE-matrix M for the unary unimodal signature

such that L□ (M) is not classical.

Proof Let ⟨A, 0, ⊐⟩ be an infinite complete Boolean algebra, and U a nonprin-

cipal ultrafilter. (Using the axiom of choice, it can be shown that there must be
such a filter.) Define ∗□ : A → A such that for all x ∈ A:

x
 if x ∈ U
∗□ (x) =
 −x otherwise

Let M = ⟨A, 0, ⊐, U, ∗□ ⟩. Then L□ (M) does not satisfy UG, since M ⊩ □p, but
M ⊮ ∀p□p. □

The problem can be overcome by requiring the filters of CBAE-matrices to be

principal:

Proposition 3.3.6 For any class C of CBAE-matrices with principal filters,

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Lσ (C) is a QCPQML.

Proof The crucial case of UG is as in the proof of Proposition 3.2.4, using the
fact that principal filters are closed under arbitrary greatest lower bounds. □

Along the lines of Proposition 3.2.6, it is easy to conclude from this that if
the modal functions of a class of CBAE-matrices with principal filters satisfy
normality and additivity, the logic of the class is a QNPQML.

3.4 Incomplete Boolean Algebras

So far, we have only interpreted propositional quantifiers on complete BAEs.
Completeness immediately guarantees that universal propositional quantifiers
can be interpreted as greatest lower bounds, since every set has a greatest lower
bound. But we really only need a weaker condition, which is that greatest lower
bounds exist for sets of propositions determined by open formulas. That is, we
need that for every formula φ, variable p, and assignment function a, the set
 72 Philosophy and Logic

of elements JφKa[x/p] , for arbitrary elements x, has a greatest lower bound. This
motivates the following definition:

Definition 3.4.1 A BAE A = ⟨A, 0, ⊐, ∗◦ ⟩◦∈O is quantifiable if there is a function

J·K · which satisfies the evaluation conditions of Definitions 3.2.2 and 3.2.3. If
so, validity is defined as in Definition 3.2.2.

We call quantifiable BAEs QBAEs. A simple way of demonstrating that QBAEs

need not be complete considers the nonmodal case, that is, the limiting case
of the empty modal signature. In this case, it can be shown that every Bool-
ean algebra A = ⟨A, 0, ⊐⟩ is quantifiable. This follows along the lines of the
elimination of propositional quantifiers in Section 1.3: it can be shown that
JφKA,a[1/p] ⊓ JφKA,a[0/p] is always the greatest lower bound of {JφKA,a[x/p] : x ∈
A}, whence the latter always exists. We will see more interesting examples in
a moment, in connection with Proposition 3.4.3.
Naturally, not every BAE is quantifiable. For a simple example, add to the
algebra of finite and cofinite subsets of N a bijection ∗□ from the algebra to the
set of singletons of even numbers. Then the interpretation of ∀p□p requires the
set of singletons of even numbers to have a greatest lower bound, which does
not exist. However, on such non-quantifiable BAEs, one can at least interpret
the combined modal-quantificational propositional quantifier [∀p] investigated
by Holliday and Litak (2018), whose interpretation is given by the following
condition:



 1 if JφKA,a[x/p] = 1 for all x ∈ A
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

J[∀p]φKA,a =
 0 otherwise


Returning to quantifiable BAEs, we note that their restrictions suffice to ensure

congruentiality:

Proposition 3.4.2 For any class C of QBAEs, Lσ (C) is a CPQML.

Proof A straightforward generalization of Proposition 3.2.4, using the fact that

the relevant infinite distributive law holds in all Boolean algebra, whenever the
relevant greatest lower bounds are defined. □

Furthermore, the constraints on QBAEs are weak enough to allow us to prove

also that every CPQML is the logic of a class of QBAEs. QBAEs therefore stand to
CPQMLs as BAEs stand to CMLs: the logics of QBAEs are all and only the CPQMLs.
In particular, every CPQML is (sound and) complete with respect to the class of
QBAEs it defines.
 Propositional Quantifiers 73

Proposition 3.4.3 Any CPQML is Lσ (C) for some class C of QBAEs.

Proof Via a standard construction of a Lindenbaum–Tarski algebra, using the

fact that in any CPQML, provable equivalence is a congruence with respect to the
Boolean and modal connectives, and formulas which are alphabetical variants
in bound variables are provably equivalent. □

With Proposition 3.2.6, we get as an immediate corollary the same claim for
NPQMLs and QBAEs satisfying normality and additivity:

Corollary 3.4.4 Any NPQML is Lσ (C) for some class C of QBAEs satisfying
normality and additivity.

As further immediate corollaries, we get that EΠ is the propositionally quan-

tified modal logic of the class of all QBAEs, and KΠ is the propositionally
quantified modal logic of the class of all QBAEs which satisfy normality and
additivity. It is also worth noting that the QBAEs used in the proof of Propo-
sition 3.4.3 are all based on a countable atomless Boolean algebra, which is
unique up to isomorphism, and incomplete. This provides us with many more
examples of BAEs which are incomplete but nevertheless quantifiable.

3.5 General Frames

We saw earlier that neighborhood frames can be thought of as concrete repre-
sentations of complete and atomic BAEs. In the context of propositional modal
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

logics without quantifiers, there is a straightforward way of generalizing the

definition of these frames so that they serve as concrete representations of all
BAEs. The resulting structures are known as general frames. As the name sug-
gests, they bear a certain correspondence to the general models for higher-order
logic of Henkin (1950). In the context of propositional modal logics without
quantifiers, they were first used by Thomason (1972). As Thomason notes, it is
interesting that in the context of propositionally quantified modal logics, they
were already used slightly earlier, by Bull (1969) and Fine (1970). In this con-
text, the notion of a general frame arises in two subtly different forms. We start
with the quantifier-free case.
By a famous representation theorem of Stone (1936), every Boolean alge-
bra is isomorphic to a field of sets, that is, a subset of a power set which
is closed under finite set-theoretic operations (e.g., intersection and relative
complement). With this representation theorem, the concrete representation of
complete and atomic BAEs using neighborhood frames can be extended to arbi-
trary BAEs by adjoining to the frame a field of sets A. Effectively, we can think
 74 Philosophy and Logic

of A as delimiting which sets of possible worlds are to count as propositions.

We only need to ensure that every formula expresses a proposition in A, under
every assignment function whose image is included in A (which also guarantees
that A is a field of sets). The definition extends straightforwardly to relational
frames, so we define the two concepts together:

Definition 3.5.1 A general relational/neighborhood frame (for a modal signa-

ture σ = ⟨O, ρ⟩) is a structure F = ⟨W, A, X◦ ⟩◦∈O such that ⟨W, X◦ ⟩◦∈O is a
relational/neighborhood frame and A ⊆ P(W ), such that JφKF,a ∈ A for every
formula φ ∈ Lqf σ and a : Φ → A, on the standard truth-conditional interpre-

tation of J·K · given by Definitions 2.3.1 and 3.1.1. Validity is defined by letting
F ⊩ φ if F, w, a ⊩ φ for all w ∈ W and a : Φ → A.

In the case of general relational frames, the connection to BAEs whose

modal functions satisfy normality and additivity can also be spelled out alge-
braically, using a generalization of Stone’s representation theorem due to
Jónnson and Tarski (1951, 1952). The correspondences between general rela-
tional/neighborhood frames and the relevant classes of BAEs have been inves-
tigated in great generality using the tools of category theory; see Thomason
(1975) and Došen (1989).
By Stone’s representation theorem, general neighborhood frames and BAEs
behave exactly alike as models of CMLs. First, every general neighborhood
frame F = ⟨W, A, N◦ ⟩◦∈O corresponds to some BAE based on the field of sets
A, with ∗ derived from N as discussed in Section 3.2. Conversely, since any
Boolean algebra is isomorphic to some field of sets A ⊆ P(W ), any BAE B
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

based on this Boolean algebra corresponds to a general neighborhood frame F

based on W and A. Furthermore, the evaluation of quantifier-free formula is pre-
served by the correspondence, in the sense that for any isomorphism f from A
to B, assignment function a : Φ → A, and φ ∈ Lqf , f (JφKF,a ) = JφKB,f◦a . Gen-
eral frames therefore add only mathematical convenience in the presentation of
BAEs.
The situation changes when we add propositional quantifiers. For example,
consider a CBAE A, and a corresponding general neighborhood frame F. When
interpreting propositional quantifiers on F = ⟨W, A, N◦ ⟩◦∈O , we might naturally
employ the standard truth-conditions of quantifiers, restricted to the designated
algebra of propositions:

F, w, a ⊩ ∀pφ if and only if F, w, a[x/p] ⊩ φ for all x ∈ A.

Along the lines discussed in Section 3.2, this means that the evaluation clause
for propositional quantifiers is the following:
 Propositional Quantifiers 75
∩
J∀pφKF,a = x∈A JφKF,a[x/p]

But note that the evaluation clause on the algebraic side interprets proposi-
tional quantifiers using greatest lower bounds. Transferring this condition to
the setting of general frames, we obtain the following condition:
∧
J∀pφKF,a = x∈A JφKF,a[x/p]

When considering a (standard) neighborhood frame as a BAE, the underly-

ing Boolean algebra is a power set algebra, in which greatest lower bounds
are always intersections. But in arbitrary fields of sets, greatest lower bounds
may come apart from intersections. This means that for general neighborhood
frames, the two evaluation conditions for propositional quantifiers may come
apart. In the presence of propositional quantifiers, there are therefore two nat-
ural conceptions of general frames. (We will shortly see concrete examples
where they come apart.)
The notion of general frames on which propositional quantifiers are eval-
uated using greatest lower bounds is simply a concrete representation of the
notion of a BAE, where we will have to restrict ourselves to frames corre-
sponding to QBAEs in order for propositional quantifiers to be interpretable. The
resulting model theory can be seen as a propositionally quantified modal ver-
sion of the (constant domain version of the) admissibility semantics developed
for propositionally quantified relevant logics by Goldblatt and Kane (2010) and
for modal predicate logics by Goldblatt (2011). In contrast, the conception of
general frames on which propositional quantifiers are given the standard truth-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

conditional interpretation leads to a class of models which come apart from

QBAEs, so we consider these for the remainder of this section. These were also
the models investigated by Bull (1969) and Fine (1970).

Definition 3.5.2 A general relational/neighborhood frame F = ⟨W, A, X◦ ⟩◦∈O

is quantifiable if JφKF,a ∈ A for every formula φ ∈ Lσ and a : Φ → A, where
Definition 3.5.1 is extended to quantified formulas using the following clause:

F, w, a ⊩ ∀pφ if and only if F, w, a[x/p] ⊩ φ for all x ∈ A.

We abbreviate “quantifiable general relational/neighborhood frame” as

“QGRF”/“QGNF”.
In the quantifier-free case, general frames are easy to construct: We can start
with any set of sets X ⊆ P(W ) and accessibility relations or neighborhood
functions on W. Since we can think of the various connectives of Lqf as inter-
preted using operators on P(W ), we can simply close X under these operators,
that is, we can take the smallest subset A ⊆ P(W ) closed under these operators.
 76 Philosophy and Logic

The same is not possible for quantifiers: The interpretation of a quantified for-
mula depends on the domain of quantification, namely, the very field of sets
we are trying to construct. Consequently, we cannot construct a quantifiable
general frame simply by starting from any set and closing it to include all
the “required” propositions: which propositions are required depends on the
algebra of propositions, which is just being determined.
How do QGNFs relate to QBAEs? It is easy to see that if a general neighborhood
frame is quantifiable, then the BAE to which it corresponds is quantifiable as
∩
well. This follows from the fact that for any field of sets A and X ⊆ A, if X ∈
∩
A, then X is the greatest lower bound of X in A. This is a useful observation
since it allows us to construct further examples of BAEs which are quantifiable
without being complete: Recall that the examples of such BAEs arising from
Proposition 3.4.3 are all atomless. In contrast, many useful examples of general
relational and neighborhood frames are based on incomplete atomic fields of
sets, and some of them are easily shown to be quantifiable. The QGRF S defined
in the following discussion is an example.
Every QGNF corresponds to a QBAE, but the converse is not the case. Indeed,
there is a CBAE which does not correspond to any QGNF. Let A be a BAE for
the unary unimodal signature based on a complete atomless field of sets A ⊆
P(W ), with a function ∗□ defined as in the proof of Proposition 3.3.5, where
U = {x ∈ A : a ∈ x} for some a ∈ W. (Since A is atomless, U is a nonprincipal
ultrafilter.) Let F be the general neighborhood frame corresponding to A, based
∩
on W and A. Then {J□pKF,a[x/p] : x ∈ A} is {a}. Since A is atomless, {a} < A.
Thus, F is not quantifiable: the intersection needed to interpret ∀p□p is not an
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

element of the underlying field of sets.

QGNFs can therefore be seen as a properly special case of QBAEs, and anal-
ogously for QGRFs and QBAEs satisfying normality additivity. It follows that
quantifiable general frames guarantee the relevant deductive principles:

Proposition 3.5.3 For any class C of QGRF/ QGNF, Lσ (C) is an NPQML/ CPQML.

Proof Routine except for the case of UI, which follows using the fact that
JφKF,a is guaranteed to be a member of A. □
General frames have mostly been considered in the form of general relational
frames. It is easy to see that all instances of the Barcan formula Bc are valid on
QGRFs. In contrast, At is not valid on all QGRFs. This follows from the following
completeness results of Bull (1969) (the last two) and Fine (1970), together with
Proposition 3.2.9 on the underivability of At:

Theorem 3.5.4 KΠ Bc, KΠ TBc, SΠ 4Bc and SΠ 5 are complete with respect to
the classes of QGRFs they define.
 Propositional Quantifiers 77

The enumerative nature of this result immediately suggests the question

whether it can be extended to all NPQMLs containing (every instance of) Bc. That
is, is every NPQML containing Bc complete with respect to the class of QGRFs
it defines, and so sound and complete with respect to some class of QGRFs? In
unpublished work, Yipu Li and Yifeng Ding (pc) show that this is indeed the
case.
For the four logics mentioned in Theorem 3.5.4, Fine also asserts a stronger
claim: For any unimodal NML Λ, Fine defines Λπ as the set of L□ -formulas
valid on the class of QGRF based on frames which validate Λ. He then asserts
that KΠ Bc is Kπ, and similarly for the other three logics. He states that this
can be proven using the canonical model technique. Furthermore, he asserts
that these four logics are undecidable.
Fine’s announcement of these results are noteworthy for two reasons. First,
the completeness claims are genuinely stronger than Theorem 3.5.4, as there are
cases in which the QGRFs based on frames validating a certain set of quantifier-
free axioms Γ validate a quantified principle which is not validated on the bigger
class of QGRFs which validate Γ. We show this in Proposition 3.5.6.
Second, Fine announces an even stronger completeness result, which is the
corresponding completeness result for every NML which is canonical, where
being canonical means being valid on its canonical frame. This claim can be
understood in two ways, depending on which canonical frame is intended. On
the one hand, since we are interested in the claim that Λπ is KΠ ΛBc, we might
understand an NML Λ to be canonical if it is valid on the canonical frame for
KΠ ΛBc. On this understanding of canonicity, Fine’s claim is confirmed by the
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

work of Li and Ding. On the other hand, an NML Λ is normally understood to be

canonical just in case it is valid on its own canonical frame, that is, the canonical
frame for Λ. Fine’s claim is false on this second way of understanding it. We
show this in Proposition 3.5.7.
Both of these observations on Fine’s results can be established using the same
model construction, to which we now turn. We will also use it to complete the
proof of Proposition 2.7.3; as noted there, the construction adapts the proof
strategy of an incompleteness result in modal predicate logic.
Let S be the general relational frame ⟨N, S, ≤⟩, where S is the set of finite
and cofinite subsets of N. Equivalently, x ∈ S if and only if there is some n ∈ N
such that for all m > n, m ∈ x if and only if n ∈ x. We can think of the members
of S as the propositions whose truth-value settles at some world n, in the sense
of staying the same in every world accessible from n.

Lemma 3.5.5 S is a QGRF validating S4M.

 78 Philosophy and Logic

Proof To show that S is a QGRF, we need to show that, in general, JφKF,a ∈ S.

This follows from the fact that whenever a( p) is settled at n, for every variable
p free in φ, JφKF,a is settled at n, which can be shown by an induction on the
complexity of φ. Only the case of quantifiers is interesting, which follows from
the uniform structure of the frame. (This can be shown more rigorously using
a version of the notion of generated subframes – stated in Definition 2.8.1 – for
general frames. For further details, see Fritz (unpublished).)
Recall that S4M strengthens S4 by the McKinsey axiom M, which is □^p →
^□p. The validity of S4 is immediate since ≤ is reflexive and transitive. For M,
note that S, n, a ⊩ □^p requires a( p) to settle to being true, which guarantees
the truth of ^□p. □

We can now establish our first claim, that requiring the underlying frame to
validate certain principles can be more demanding than requiring the general
frame itself to validate these principles, and that this can be reflected in the
validities on the relevant class of structures:

Proposition 3.5.6 ^∀p( p → □p) is valid on every QGRF which is based on a

frame validating S4M, but not on every QGRF which validates S4M.

Proof A routine argument shows that in every relational frame validating S4M,
every world w can access a world v such that R□ (v) = {v}; this ensures the
validity of ^∀p( p → □p). However, S, which we have just shown to validate
S4M, does not validate ^∀p( p → □p), since it contains no world at which all
propositions are settled. □
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Next, we can show that the second reading of Fine’s claim concerning
canonical NMLs is incorrect:

Proposition 3.5.7 S4M is a canonical NML, but S4Mπ is not completely

axiomatized by SΠ 4MBc (i.e., KΠ T4MBc).

Proof S4M is known to be canonical; see Hughes and Cresswell (1996,

pp. 131–134). (This does not contradict the fact that the M axiom alone is
not canonical, which was shown by Goldblatt (1991).) As shown in the proof
of Proposition 3.5.6, ^∀p( p → □p) is a member of S4Mπ, but not valid
on S. Since S validates S4M, it validates SΠ 4MBc, whence ^∀p( p → □p)
is not a theorem of SΠ 4MBc. So, SΠ 4MBc does not completely axiomatize
S4Mπ. □

We can also use S to prove an assertion used in the proof of Proposition 2.7.3:
 Propositional Quantifiers 79

Proposition 3.5.8 KΠ 3.1BcAt (i.e., KΠ T4Lem0 J1BcAt) does not prove

^∀p( p → □p).

Proof Given the preceding results, it suffices to show that KΠ 3.1BcAt is valid
on S. The only interesting case is J1, that is, the formula □(□( p → □p) →
p) → p. Arguing contrapositively, consider any world n in which p is false, on
a given variable assignment a. There are two options: either p is settled as false
at some m ≥ n, or p settles as true at m + 1 for some m ≥ n at which p is false.
(For brevity, we conflate p and a( p) in laying out the argument.) In both cases,
it is easy to see that □( p → □p) → p is false in m, whence □(□( p → □p) → p)
is false in n. Therefore J1 is valid on S. □

The requirement of being quantifiable ensures that general frames validate

UI. But the interpretation of Lσ on general frames is well-defined without this
requirement. In fact, we can even relax the requirement of general frames that
JφKF,a should be a member of the propositional domain, for every quantifier-
free formula φ. Depending on how much this requirement is relaxed, the
corresponding instances of UI might fail, so the relevant models will not in
general lead to classical propositionally quantified modal logics. We therefore
don’t consider them in any detail here, but it should be noted that such wider
classes of frames and correspondingly weaker logics were already considered
by Fine (1970), who establishes certain completeness results for them as well.
More recently, these results were extended by Belardinelli et al. (2018), who
also consider the bimodal case. For a philosophical application of such weaker
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

logics, see Bacon et al. (2016).

3.6 Variable Propositional Domains

So far, the frames we have considered effectively operate with a single propo-
sitional domain, over which propositional quantifiers range independently of
the world of evaluation. However, there are applications where this feature
is too restrictive. Consider the view that according to the notion of neces-
sity operative in metaphysics, it is a contingent matter what things there are.
Williamson (2013) calls this view contingentism, and the opposing view neces-
sitism. Contingentism is attractive since it matches our pretheoretic judgements
about simple cases such as the following: Ludwig Wittgenstein could have
had a daughter. But since he actually had no children, there is plausibly noth-
ing which could have been his daughter. So, there could have been something
which does not actually exist, in the sense that actually, there is nothing identi-
cal to it. Conversely, Ludwig Wittgenstein could himself have failed to be born,
 80 Philosophy and Logic

in which case he would plausibly not have existed, in the sense that there would
not have been anything identical to him. In this sense, then, there is something
which might have failed to exist.
Some philosophers have not only endorsed contingentism, but also argued
that propositions are existentially dependent on the individuals they are about;
see, for example, Prior (1967), Fine (1977), Adams (1981), and Stalnaker
(2012). For example, they would argue that the proposition that Ludwig Witt-
genstein is a philosopher existentially depends on Ludwig Wittgenstein: had
there not been Ludwig Wittgenstein, there would also not have been the
proposition that Ludwig Wittgenstein is a philosopher. Similarly, had Lud-
wig Wittgenstein been the father of a daughter, the proposition that she is
his daughter would have existed, but since she does not actually exist, nei-
ther does the proposition that she is his daughter. Adapting Williamson’s
terminology, Fritz (2016) calls the view that propositional existence is con-
tingent propositional contingentism, and the opposing view propositional
necessitism. Contingentism has therefore been argued to lead to proposi-
tional contingentism. This argument was already made by Prior, and Prior
accommodated this propositional contingentism in his propositionally quan-
tified modal logic; this led to his System Q in Prior (1957, chapter 5).
Prior assumed that nonexistent propositions are neither true nor false, which
led to a complicated and, from a modern perspective, unusual system. In
the following, we will consider some less intrusive modifications to the
models considered so far which nevertheless make room for propositional
contingentism.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Prima facie, constructing models for propositional contingentism may seem

relatively easy: one may simply enhance frames based on possible worlds by a
form of the variable domains of Kripke (1963b). That is, we modify the general
frames of the previous section by replacing the single domain of quantification
A by a function D which maps every world to a domain of propositions over
which the propositional quantifiers range at that world. One version of such a
model theory was already proposed by Fine (1970, pp. 344–345). However, it
is natural to require the propositional domains at each world w to be closed in
very much the same sense as it was imposed on the domains of quantifiable
general frames: any proposition expressed by a formula using parameters from
the domain of w must itself be in the domain of w. (Here, a parameter is a
proposition used as the interpretation of a free variable.) This requirement cor-
responds to a natural comprehension principle for propositional contingentism;
see the principle discussed by Fritz (2017, p. 211), which is adapted from a cor-
responding principle of higher-order modal logic in Williamson (2013, p. 284).
The definition of validity also requires some thought. In general frames, the
 Propositional Quantifiers 81

definition of validity considers only assignment functions which map proposi-

tional variables to elements in the propositional domain A. In variable domain
frames, the propositional domain is dependent on the world, so this require-
ment needs to be refined. One natural way of doing so defines φ to be valid on
a frame just in case φ is true in every world w under any assignment function
mapping every propositional variable to an element of the propositional domain
of w:

Definition 3.6.1 A variable domain relational/neighborhood frame (for a modal

signature σ = ⟨O, ρ⟩) is a structure F = ⟨W, D, X◦ ⟩◦∈O such that ⟨W, X◦ ⟩◦∈O is
a relational/neighborhood frame and D : W → P(P(W )), such that JφKF,a ∈
D(w) for every formula φ ∈ Lσ , w ∈ W, and a : Φ → D(w), on the stand-
ard truth-conditional interpretation of J·K · given by Definitions 2.3.1 and 3.1.1,
except for the following clause which is used for quantified formulas:

F, w, a ⊩ ∀pφ if and only if F, w, a[x/p] ⊩ φ for all x ∈ D(w).

Validity is defined as follows:

F ⊩ φ if F, w, a ⊩ φ for all w ∈ W and a : Φ → D(w).

In the following, we abbreviate “variable domain relational/neighborhood

frame” as “VDRF”/“VDNF”. We can confirm that these model-theoretic construc-
tions suffice to ensure classicality. Since VDRFs are effectively special cases of
VDNFs, it suffices to consider the following claim:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Proposition 3.6.2 For any class C of VDNFs, Lσ (C) is a classical proposition-

ally quantified modal logic.

Proof Routine. For the validity of UI, we use the fact that JφKF,a ∈ D(w) for
every formula φ ∈ Lσ , w ∈ W, and a : Φ → D(w). □

VDNFs generalize QGNFs in a natural way, and many of the remarks on

QGNFs made in the previous section apply here as well. For example, the truth-
conditional interpretation of quantifiers could be replaced by an algebraic one,
and the resulting models would constitute a natural version of the admissibil-
ity semantics of Goldblatt (2011) for propositionally quantified modal logics.
Likewise, as in the case of quantifiable general frames, one cannot construct
variable domain frames just by starting with a stock of propositions assigned
to each world and then closing under definability in Lσ ; again, a stable interpre-
tation of quantified formulas requires a fixed domain (function). It is therefore
 82 Philosophy and Logic

useful to consider a more specific construction, with which a range of nontrivial

variable domain frames can be defined.
The construction we will consider is based on work of Fine (1977) and Stal-
naker (2012), and developed in Fritz (2017). It builds on the very simple case
of relational frames with a universal accessibility relation. To determine prop-
ositional domains, one starts with an equivalence system, which is a function ≈
which associates each world w with an equivalence relation ≈w among worlds.
Informally, v ≈w u is interpreted as v and u being indistinguishable from the
perspective of w. A domain function D is then derived by letting D(w) con-
tain just those propositions which do not distinguish between indistinguishable
worlds. This determines a VDRF as long as ≈ satisfies a certain condition of
“quasicoherence”, stated in the following definition. The condition makes use
of the notion of an automorphism of an equivalence system ≈; this is a function
f on the set of worlds such that for all worlds w, v, and u: v ≈w u if and only if
f (v) ≈f(w) f(u).

Definition 3.6.3 An equivalence system on a set W is a function ≈ mapping

every w ∈ W to an equivalence relation ≈w .
≈ is quasicoherent if for all w, v, u ∈ W such that v ≈w u, there is an
automorphism f of ≈ such that f (v) = u and f ⊆ ≈w .
F≈ is the structure ⟨W, D, R⟩ such that R = W × W, and D is the function
mapping every w ∈ W to:

D(w) = {x ⊆ W : if v ≈w u, then v ∈ x if and only if u ∈ x}

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

As the term “quasicoherent” indicates, there is a somewhat stronger condi-

tion of being coherent. Coherence adds to the definition of quasicoherence the
clause that f (w) = w. It follows from results in Fritz (2017, pp. 212–213) that:

Proposition 3.6.4 If ≈ is a quasicoherent equivalence system, then F≈ is a

VDRF.

With this concrete way of constructing variable domain frames, we can easily
show that both the Barcan formula Bc and its converse have instances which
are not valid on VDRFs. This is expected, as these formulas are hallmarks of
(propositional) necessitism; see Williamson (2013, chapter 2) for discussion.
To state these observations, note that on frames with a universal accessibility
relation, the following definition allows us to state in the object language that
the proposition expressed by any given formula φ exists, in the sense of being
in the domain of the world of evaluation:
 Propositional Quantifiers 83

Eφ := ∃q□(q ↔ φ),

where q is chosen to be some variable which is not free in φ.

Proposition 3.6.5 The following instances of Bc and its converse are not valid
on VDRFs:

(1) ∀p□□Ep → □∀p□Ep

(2) □∀pEp → ∀p□Ep

Proof Let ≈ be the equivalence system based on {1, 2, 3} such that v ≈w u if

and only if v = u, or w = 1 and {v, u} ⊆ {2, 3}. ≈ is coherent, so F≈ is a VDRF.
(1) is false in 1, and (2) is false in 2. □
Since all instances of the converse of Bc are derivable in NPQMLs, it follows
that the logic of VDRFs is not normal. In fact, even the logic of the class of
frames F≈ for coherent equivalence systems ≈ – which includes the strong NML
S5 – is not normal. The culprit is the rule of necessitation: although Ep is valid
on VDRFs, its necessitation is not valid, as demonstrated by the frame used in
the proof of Proposition 3.6.5.
The failure of necessitation creates difficulties for constructing axiomatic
systems. One solution is to axiomatize the validities according to a more
demanding notion of validity which removes the constraint on assignment
functions, defined as follows:

F ⊩′ φ if F, w, a ⊩ φ for all w ∈ W and a : Φ → P(W ).

https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

It is easy to see that questions about ⊩-validity can straightforwardly be reduced

to questions about ⊩′-validity, as the following equivalence holds for any VDNF
F and formula φ with free variables p1, . . . , pn :
F ⊩ φ if and only if F ⊩′ ∀p1 . . . ∀pn φ.

By axiomatizing the ⊩′-validities of a class of frames, one thereby also obtains

a syntactic characterization of the ⊩-validities.
The ⊩′-validities of any class of VDRFs are easily seen to be closed under
necessitation, and therefore more amenable to a standard axiomatic treatment.
However, they need not include all instances of the axiom schema UI; for
example, ∀pEp → Ep is not ⊩′-valid on the frame used in the proof of Propo-
sition 3.6.5. To axiomatize the ⊩′-validities of a class of VDRFs, one might
therefore follow the model of axiomatizations of modal predicate logics based
on variable domain frames, which are often based on axiom systems of free
logic. Along the lines of the approach taken in Hughes and Cresswell (1996,
chapter 16), one might restrict UI by an existence predicate. As we have seen,
 84 Philosophy and Logic

such a predicate E can be defined in the context of VDRFs with a single univer-
sal accessibility relation. The restricted version of UI can then be formulated
as follows:

(RUI) ∀pφ → (Eψ → φ[ψ/p]) whenever ψ is free for p in φ.

The matter is more difficult in the context of VDRFs generally, or even VDNFs,
where the definition of Eφ is not guaranteed to express existence of the relevant
proposition. In these contexts, one might instead adopt the approach of Kripke
(1963b), and use an axiom schema consisting of certain closed instances of UI:

(CUI) ∀q(∀pφ → φ[q/p]) whenever q is free for p in φ.

However, subtleties involving the permutation of quantifiers emerge on this

approach; see Fine (1983) and Hughes and Cresswell (1996, pp. 304–309).
Whichever approach is taken, propositionally quantified modal logics also
pose special difficulties in terms of axiomatizability. Recall from Section 2
that many classes of relational frames have a propositionally quantified modal
logic which is not recursively enumerable. The case of relational frames with a
universal accessibility relation emerged as an exception, as the propositionally
quantified modal logic of this class of frames is decidable. Since quasicoher-
ent equivalence systems are based on such frames, the question arises whether
the resulting class of VDRFs has a recursively enumerable logic. Fritz (2017,
pp. 211–212) shows that this is not the case, and the result extends to the class
of coherent equivalence systems as well:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Proposition 3.6.6 If C is the class of VDRFs F≈ , for ≈ a quasicoherent/coherent

equivalence system, then L□ (C) is recursively isomorphic to second-order
logic.

Equivalence systems can be expanded to accommodate nontrivial accessi-

bility relations or neighborhood functions, a generalization which has not yet
been investigated systematically. For an application of such structures to the
logic of awareness, see Fritz and Lederman (2016). The resulting VDRFs can
also be used to establish results concerning NPQMLs; for example, Ding (2021b)
uses them to establish that 4∀ is not derivable in KΠ D45, a result mentioned in
Section 3.2.
There are also many open questions concerning variable domain frames
beyond those determined by equivalence systems. One example appears to be
the following:
 Propositional Quantifiers 85

Open Question 7 Is the propositionally quantified modal logic of the class of

all VDRFs recursively enumerable?

As usual, there are a number of natural variations on this question, which

also appear to be open. For example, we might ask about the logic of the class
of VDNFs, and whether the answers to these questions depend on the modal
signature.

3.7 Substitutional Quantifiers

In this section, we consider a substitutional interpretation of propositional
quantifiers, on which a universal quantification is true just in case all of its
substitution instances are true. Philosophical discussion of a substitutional
interpretation of propositional quantifiers can be found especially in the debate
on deflationism about truth, mentioned in Section 1.6; see, for example, Grover
et al. (1975), van Inwagen (2002), and Christensen (2011). For an example of
an application of propositional quantifiers – to the logic of necessity and aprior-
ity – which assumes a substitutional interpretation, see Tharp (1989). Here, we
focus on how a substitutional interpretation of propositional quantifiers could
be formalized model-theoretically.
In the context of relational or neighborhood frames, the intended truth con-
ditions of a propositional quantifier on the substitutional interpretation are easy
enough to state: ∀pφ is intended to be true just in case φ[ψ/p] is true for every
ψ ∈ Lσ free for p in φ. However, it is well known that these truth conditions
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

cannot serve the usual purpose of a recursive determination of which formulas

are true relative to which worlds and assignment functions. To illustrate this,
consider the case of the formula ∀pp. According to the truth conditions just
stated, this is true if and only if every one of its substitution instances is true.
But one of these substitution instances is ∀pp itself, so one of the conditions for
the truth of ∀pp is the truth of ∀pp itself.
There are different ways of solving this difficulty. One option is to limit
the substitution instances, which is suggested by Grover (1972). For exam-
ple, one might restrict the substitution instances to quantifier-free formulas,
to Boolean combinations of proposition letters, or to just proposition letters. A
substitutional interpretation of propositional quantifiers restricted to quantifier-
free instances in the context of a logic of (un)awareness can be found in Halpern
and Rêgo (2009), and a similar treatment of a quantified “public announce-
ment” operator is investigated in Ågotnes et al. (2016). (See also van Ditmarsch
(2023) for a general survey of work in dynamic epistemic logic involving
operators which incorporate some aspect of propositional quantification, often
 86 Philosophy and Logic

involving restricted substitutional clauses.) These relatively straightforward

treatments of substitutional quantifiers typically lead to logics which are non-
classical, as not all instances of UI are guaranteed to be valid, similar to the
logics mentioned briefly at the end of Section 3.5. There are various ways of
trying to improve on these limitations. For example, Grover (1973) extends
the capabilities of propositional quantifiers on a substitutional interpretation by
stratifying propositional variables into an infinite hierarchy of levels, reminis-
cent of the ramified type theory of Whitehead and Russell (1910–13). Another
example is Halpern and Rêgo (2013), whose refined models make the relevant
substitution instances relative to the world of evaluation, in some ways similar
to the variable domains discussed in Section 3.6.
There is, however, an alternative approach, which retains the truth conditions
of substitutional propositional quantifiers as stated earlier. On this alternative
approach, these truth conditions do not play a part in a recursive determination
of truth, but serve as holistic constraints on admissible interpretation func-
tions used to define the relevant class of models. Such a model construction is
briefly mentioned by Bull (1969, p. 258), and considered in more detail by Gab-
bay (1971), who uses it to develop models for NPQMLs based on neighborhood
frames. For present purposes, it suffices to illustrate the idea using models based
on relational frames. Such models might take the form ⟨W, R□, ⊩⟩□∈O , where
⟨W, R□ ⟩□∈O is a relational frame. ⊩ is then required to be a relation between
members of W and formulas of Lσ satisfying the intended truth conditions. In
the case of proposition letters and Boolean operators, these are the standard
truth conditions stated in Definition 2.3.1. For propositional quantifiers, they
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

are the substitutional truth conditions, which can be stated as follows:

w ⊩ ∀pφ if and only if w ⊩ φ[ψ/p] for all ψ ∈ Lσ free for p in φ.

On this approach, it is a substantial claim that there are any models at all. For
a given relational frame, we cannot simply assume that there is any relation ⊩
of truth satisfying the desired truth conditions. In certain cases, it can be shown
that there is such a relation; for example, Kripke (1976, p. 332) notes that a
suitable relation ⊩ can always be found if all operators are truth-functional.
This can be shown using the eliminability of propositional quantifiers in such a
setting, discussed in Section 1.3. In the context of his models based on neigh-
borhood frames, Gabbay (1971) establishes completeness results, which entail,
for every formula φ which is not derivable in the relevant proof system, the
existence of a model in which φ is falsified. It is important to note, however, that
without an independent proof of the consistency of the relevant proof system,
such a completeness result does not entail the existence of any model.
 Propositional Quantifiers 87

A substitutional interpretation of propositional quantifiers in modal logic is

especially congenial on a substitutional interpretation of the modal operators.
Such an interpretation of the unary operator □ was proposed by McKinsey
(1945). On one elegant form of this proposal, □φ is true just in case any uni-
form substitution of the free variables in φ produces a true formula. Combined
with the substitutional interpretation of quantifiers, this interpretation can be
defined as follows:

Definition 3.7.1 Let a valuation be a function v from L□ to {0, 1}. v is sub-

stitutional if it satisfies the following constraints, where p1, . . . , pn are the free
variables in φ:

v(⊥) = 0
v(φ → ψ) = 1 if and only if v(φ) = 1 only if v(ψ) = 1
v(□φ) = 1 if and only if v(φ[ψ1 /p1, . . . , ψn /pn ]) = 1 for all ψ1, . . . , ψn ∈ L□
free for p1, . . . , pn , respectively, in φ
v(∀pφ) = 1 if and only if v(φ[ψ/p]) = 1 for all ψ ∈ L□ free for p in φ

At first glance, it might appear as if in the context of substitutional valuations,

□φ becomes interchangeable with ∀p1 . . . ∀p1 φ, where p1, . . . , pn are the free
variables in φ. For example, consider a formula φ with a single free variable
p. Then the conditions for a valuation v to map □φ to 1 are the same as the
conditions for v to map ∀pφ to 1; namely, v must map φ[ψ/p] to 1 for every
ψ free for p in φ. However, this does not mean that we can eliminate □ using
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

propositional quantifiers in the context of substitutional valuations. The reason

is that there is a crucial syntactic difference between □φ and ∀pφ: the variable
p remains free in □φ but not in ∀pφ. This means that the two formulas embed
differently. For example, the constraints on substitutional valuations require
∀p¬□p to be mapped to 0 but ∀p¬∀pp to be mapped to 1.
Are there substitutional valuations of L□ ? Even in the quantifier-free case,
where we use L□qf instead of L□ and omit the last clause of Definition 3.7.1, it
is far from clear that there are any substitutional valuations. In fact, the 42nd
question on the list of questions in mathematical logic by Friedman (1975)
asks whether there is a substitutional valuation in the quantifier-free case, with
Friedman conjecturing that there is such a valuation. The existence of such a
valuation was independently established by Prucnal (1979), and by Kit Fine in
unpublished work. Fine’s proof is to be published as part of Bacon and Fine
(forthcoming). Friedman also asks whether there is a unique substitutional val-
uation which maps every p ∈ Φ to 1, and conjectures that the answer is negative.
This further question on uniqueness remains open in the quantifier-free case.
 88 Philosophy and Logic

More generally, one might ask whether every mapping from p ∈ Φ to {0, 1} has
a unique extension to a substitutional valuation of L□qf . In the propositionally
quantified case, all of these questions remain open:

Open Question 8 Is there a substitutional valuation of L□ ? If so, does every

mapping from p ∈ Φ to {0, 1} have a unique extension to a substitutional
valuation of L□ ?

Bacon and Fine (forthcoming) consider a version of this question, and con-
jecture that there is a substitutional valuation of L□ . It is worth noting that
Bacon and Fine’s definition of a substitutional valuation is slightly different,
since their language contains both propositional variables and nonlogical con-
stants, and their conditions for the modal operator and quantifier refer to closed
substitution instances only. Moreover, while their constraints on quantified
statements match the definition given here insofar as it considers substitutions
of the bound propositional variable, their constraints on modal statements of
the form □φ vary not the interpretation of free variables in φ but of the propo-
sitional constants in φ. This leads to substantial differences, even within the
shared fragment of the language not involving any propositional constants.
For example, in Bacon and Fine’s setting, a substitutional valuation must map
( p → □p) ∧ (¬p → □¬p) to 1 (taking p to be a variable, as before), whereas
on the definition presented here, it must map this formula to 0. It is also worth
noting that it is not immediately obvious whether substitutional valuations as
defined earlier verify all classical principles, that is, all theorems of CΠ . In
contrast, in the quantifier-free case, it follows from results of McKinsey (1945)
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

that any substitutional valuation verifies every theorem of S4M. If the answer
to the first part of Question 8 is positive, there are therefore natural follow-up
questions concerning which logical principles are guaranteed to be verified by
substitutional valuations, and whether there are substitutional valuations which
verify, for example, every theorem of SΠ 4M.

4 Conclusion
We have seen that propositional quantifiers constitute a natural extension of
a number of logical systems, in particular propositional modal logics. The
resulting propositionally quantified modal logics can be interpreted over many
well-known classes of models for propositional modal logics. Many of the
standard questions concerning propositional modal logics arise as well for
propositionally quantified modal logics. Some results extend from the stand-
ard quantifier-free setting to the setting with propositional quantifiers, but often
they do not do so, or do not do so without nontrivial adjustments. Over many
 Propositional Quantifiers 89

classes of models, propositional quantifiers constitute a substantial addition,

with further classes of models becoming definable and the logics of vari-
ous classes of models increasing in computational complexity, in a significant
number of cases becoming not recursively axiomatizable.
Almost all of the models discussed in this Element effectively represent
propositions as elements of a Boolean algebra: They either explicitly start from
a Boolean algebra, or are based on a set of worlds W and effectively represent
propositions as elements of a field of sets A ⊆ P(W ). Even the purely substi-
tutional interpretation considered in Section 3.7, in which sentences are only
assigned 0 or 1, can be thought of as based on a conception of the propositions
as truth-values, and so as the elements of the two-element Boolean algebra. The
fact that all of our models operate with an underlying algebra which is Boolean
plays a role in ensuring that the logics to which they give rise are classical. For
example, if the underlying algebra is Boolean, then every tautology is always
interpreted as the top element 1 of the algebra, and so valid on the relevant
model. However, it is important to be clear that making the underlying algebra
Boolean is not dictated by the constraints of classicality.
To illustrate this very abstractly, consider again the notion of a BAE-matrix
M = ⟨A, 0, ⊐, F, ∗◦ ⟩◦∈O , stated in Definition 3.3.4. The definition of a BAE-
matrix requires ⟨A, 0, ⊐⟩ to be a Boolean algebra, and F to be a filter. This
ensures that every tautology is interpreted as 1, and so as a member of F,
which in turn means that tautologies are valid. But we might also impose this
requirement more directly: We might simply let ⟨A, 0, ⊐⟩ be an arbitrary alge-
braic structure, with 0 a nullary and ⊐ a binary function, while requiring that
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

0 < F and x ⊐ y ∈ F just in case x ∈ F only if y ∈ F. This will also guar-

antee that all tautologies are members of F, and thereby guarantee the validity
of tautologies. This abstract algebraic approach can be extended to quantifiers;
for an example, see Lewitzka (2015). There are many variations on this general
idea; for another example in the more general context of higher-order logic, see
Muskens (2007).
Considering these more general algebraic models is especially natural in
exploring theories of propositional identity. We have not specifically consid-
ered an identity connective here, but such a connective does fall under the
general notion of a modal operator we have been working with: we can think of
a propositional identity connective = simply as a binary modal operator, in the
wide sense of modality employed here. Such an identity operator may be unfa-
miliar, but it has many interesting applications. For example, as mentioned in
Section 1.6, a version of the Epimenides paradox is naturally formalized using
propositional quantifiers.
 90 Philosophy and Logic

To illustrate this, let E be an operator formalizing “is said by Epimenides”.

Assume that Epimenides says that everything Epimenides says is false. That is,
assume Eδ, where δ is the following sentence:

(δ) ∀p(Ep → ¬p)

Is everything Epimenides says false? That is, δ? Instantiating δ with δ, we

obtain:

(γ) Eδ → ¬δ

We assumed Eδ, so ¬δ follows, but this contradicts the assumption. So, ¬δ. But
then there is some p such that Ep and p. What is this p? It cannot be δ, since
¬δ. But we have only assumed that Epimenides says δ. We seem to be forced,
paradoxically, to the conclusion that there is something else which Epimenides
says, even if he only utters “∀p(Ep → ¬p)”:

(β) ¬∀q(Eq ↔ q = δ)

As the present sketch suggests, this argument can be carried out in a relatively
weak propositionally quantified modal logic, using plausible deductive princi-
ples for = and no particular assumptions about E. This was well known to the
founders of symbolic logic, and a derivation along these lines can be found, for
example, in Hilbert and Ackermann (1938, pp. 114–121). The paradox, in its
formalization using propositional quantifiers, was also discussed at length by
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

the pioneer of propositional quantifiers, Prior (1958b, 1961).

In the presence of a modality of necessity □, applying necessitation and
existential generalization to β yields the following corollary:

(α) ∃p¬^∀q(Eq ↔ (q = p))

Thus, there is something – δ – which Epimenides cannot uniquely say. Var-

ious alternative interpretations of E yield other paradoxical conclusions. For
example, we might read E as “is entertained”, and conclude that there is some
proposition, namely the proposition that everything entertained is false, which
cannot be uniquely entertained. This modal version of the paradox was also
noted by Kaplan (1995), who observed the validity of α on a very general class
of possible worlds models. Kaplan described this observation as a problem
for possible world semantics. However, as pointed out by Anderson (2009)
and Lindström (2009), the deducibility of this principle from plausible axio-
matic principles shows that the problem is not obviously tied to any particular
 Propositional Quantifiers 91

model theory. The issue has recently received renewed attention; see Bacon
et al. (2016), Holliday and Litak (2018), Bacon and Uzquiano (2018), Ding
and Holliday (2020), and Uzquiano (2021).
Returning to non-Boolean models, such models are especially interesting for
exploring views on which propositions are very finely individuated. However,
we need to be cautious in doing so, for two reasons: First, once we reject the
identities of Boolean algebras, we must reconsider the interdefinability of the
operators ⊤, ⊥, ¬, ∧, ∨, → and ↔. For example, on a fine-grained conception
of propositions, the formulas ( p → ⊥) → q and (q → ⊥) → p may not always
express the same proposition (on a given interpretation of p and q). In such a
situation, it is not clear that we can use (φ → ⊥) → ψ to abbreviate φ ∨ ψ: the
former may simply fail to express disjunction. In fact, in such a setting, it is not
clear that we can make any such abbreviations. We may then have to take all
of the operators as primitive.
Second, the restrictiveness of languages like Lσ involving only propositional
quantifiers may make fine-grained theories of propositions more plausible than
they are. This is because prima facie plausible principles of fine individuation
of propositions can be shown to be inconsistent using higher-order quantifiers,
by an argument due to Russell (1903, appendix B) and Myhill (1958). In a
higher-order language, using X and Y as variables which take the position of
unary sentential operators, a natural version of the idea that propositions are
structured leads to the following principle:

∀X ∀Y ∀p(Xp = Yp → X = Y )
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

This, however, can be shown to be inconsistent in a natural proof system of

classical third-order logic, using the argument of Russell and Myhill. Since the
argument cannot be formulated in a purely propositionally quantified modal
language, limiting ourselves to such languages may therefore make various
theories of propositional identification seem more attractive than they actu-
ally are. As discussed in Section 1.4, the restrictiveness of languages like Lσ
has both advantages and disadvantages. It may well be that for the purpose
of exploring theories of propositional identity, it is better to work in richer lan-
guages. Such languages need not necessarily provide a full infinite hierarchy of
types. For example, in Fritz (2023c) I develop a theory of propositional identity
using a fragment of third-order logic which includes propositional quantifiers
alongside quantifiers binding variables in the position of sentential opeators.
Is a structured theory of propositions consistent in a merely proposition-
ally quantified setting? The consistency question is most naturally posed
as concerning a logic of propositional identity which consists of just the
elementary principles of identity:
 92 Philosophy and Logic

(RI) p = p
(LL) p = q → (φ → φ[q/p]), whenever q is free for p in φ

For any modal signature σ = ⟨O, ρ⟩ which includes the binary identity con-
nective =, let Iσ be CΠ
σ RILL, the classical propositionally quantified modal

logic axiomatized by RI and LL. A natural theory of structured propositions

can be formulated using the following schemas, where µ and ν may be any
propositional connective (i.e., Boolean operator or member of O):

(S1 ) µp1 . . . pn , νq1 . . . qm , whenever µ and ν are distinct

(S2 ) µp1 . . . pn = µq1 . . . qn → pi = qi

To illustrate these two principles: By S1 , no conditional proposition is an iden-

tity proposition. By S2 , conditional propositions can only be identical if their
antecedents are the same and their consequents are the same. It follows immedi-
ately from observations in Fritz (2023b) that S1 and S2 are jointly consistent in
Iσ . What is not obvious is whether we can introduce corresponding principles
for propositional quantifiers:

(S3 ) µp1 . . . pn , ∀qφ

(S4 ) ∀pφ = ∀pψ → ∀p(φ = ψ)

In the presence of higher-order quantifiers, S1 − S4 are jointly inconsistent;

indeed, Church (1984) essentially notes that S4 is inconsistent by itself. (See
also Fritz (2023a,b) for discussion of a number of variant principles of propo-
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

sitional structure and their consistency.) In a more restricted language like Lσ ,

consistency appears to be an open question.
In the course of the journey through the landscape of logics with propo-
sitional quantifiers which this Element has taken, we have seen many areas
calling for further development. I have highlighted some examples of par-
ticularly intriguing open questions in the hope of inspiring others to explore
propositional quantifiers, especially in the context of modal logics. It therefore
seems fitting to conclude with a final open question:

Open Question 9 Are S1 − S4 jointly consistent in Iσ ?

 Abbreviations
The following abbreviations are used with the following meanings, with page
numbers in parentheses referring to the introduction of the relevant concept:

BAE: Boolean algebra expansion (p. 62)

CBAE: complete Boolean algebra expansion (p. 62)
CML: congruential modal logic (p. 53)
CPQML: congruential propositionally quantified modal logic (p. 53)
NML: normal modal logic (p. 31)
NPQML: normal propositionally quantified modal logic (p. 31)
QBAE: quantifiable Boolean algebra expansion (p. 72)
QGNF: quantifiable general neighborhood frame (p. 75)
QGRF: quantifiable general relational frame (p. 75)
QCML: quasi-congruential modal logic (p. 69)
QCPQML: quasi-congruential propositionally quantified modal logic (p. 69)
QNML: quasi-normal modal logic (p. 69)
QNPQML: quasi-normal propositionally quantified modal logic (p. 69)
VDNF: variable domain neighborhood frame (p. 81)
VDRF: variable domain relational frame (p. 81)
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 References
Ågotnes, T., H. van Ditmarsch, and T. French. The undecidability of quantified
announcements. Studia Logica, 104:597–640, 2016.
Adams, Robert Merrihew. Actualism and thisness. Synthese, 49:3–41, 1981.
Anderson, Alan Ross. An intensional interpretation of truth-values. Mind,
81:348–371, 1972.
Anderson, Alan Ross, and Nuel D. Belnap, Jr. Enthymemes. The Journal of
Philosophy, 58:713–723, 1961.
Anderson, Alan Ross, Nuel D. Belnap, Jr, and J. Michael Dunn. Entailment:
The Logic of Relevance and Necessity, volume II. Princeton: Princeton
University Press, 1992.
Anderson, C. Anthony. Bealer’s Quality and Concept. Journal of Philosophical
Logic, 16:115–164, 1987.
Anderson, C. Anthony. The lesson of Kaplan’s paradox about possible world
semantics. In Joseph Almog and Paolo Leonardi, editors, The Philosophy of
David Kaplan, pages 85–92. Oxford: Oxford University Press, 2009.
Andrews, Peter B. A reduction of the axioms for the theory of propositional
types. Fundamenta Mathematicae, 52:345–350, 1963.
Antonelli, G. Aldo. and Richmond H. Thomason. Representability in second-
order propositional poly-modal logic. The Journal of Symbolic Logic,
67:1039–1054, 2002.
Arló Costa, Horacio. First order extensions of classical systems of modal logic:
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

The role of the Barcan schemas. Studia Logica, 71:87–118, 2002.

Baaz, Matthias, and Norbert Preining. Quantifier elimination for quantified
propositional logics on Kripke frames of type ω. Journal of Logic and
Computation, 18:649–668, 2008.
Baaz, Matthias, and Helmut Veith. An axiomatization of quantified proposi-
tional Gödel logic using the Takeuti–Titani rule. In Samuel R. Buss, Petr
Hájek, and Pavel Pudlák, editors, Logic Colloquium ’98, volume 13 of Lec-
ture Notes in Logic, pages 91–104. Cambridge: Cambridge University Press,
2000.
Baaz, Matthias, Agata Ciabattoni, and Richard Zach. Quantified propositional
Gödel logics. In Michel Parigot and Andrei Voronkov, editors, Logic for
Programming and Automated Reasoning. 7th International Conference,
LPAR 2000, volume 1955 of Lecture Notes in Artificial Intelligence, pages
240–256. Berlin: Springer, 2000.
 References 95

Bacon, Andrew. A Philosophical Introduction to Higher-Order Logics. New

York: Routledge, 2024.
Bacon, Andrew, and Kit Fine. The logic of logical necessity. In Romina Padró
and Yale Weiss, editors, Saul Kripke on Modal Logic. Springer, forthcoming.
Bacon, Andrew, and Gabriel Uzquiano. Some results on the limits of thought.
Journal of Philosophical Logic, 47:991–999, 2018.
Bacon, Andrew, John Hawthorne, and Gabriel Uzquiano. Higher-order free
logic and the Prior–Kaplan paradox. Canadian Journal of Philosophy,
46:493–541, 2016.
Badia, Guillermo. Incompactness of the ∀1 fragment of basic second-order
propositional relevant logic. Australasian Journal of Logic, 16:1–8, 2019.
Barcan, Ruth C. A functional calculus of first order based on strict implication.
The Journal of Symbolic Logic, 11:1–16, 1946.
Barcan, Ruth C. The identity of individuals in a strict functional calculus of
second order. The Journal of Symbolic Logic, 12:12–15, 1947.
Bayart, Arnould. La correction de la logique modale du premier et second ordre
S5. Logique et Analyse, 1:28–44, 1958.
Bayart, Arnould. Quasi-adéquation de la logique modale de second ordre S5 et
adéquation de la logique modale de premier ordre S5. Logique et Analyse,
2:99–121, 1959.
Bealer, George. Property theory: The type-free approach v. the Church
approach. Journal of Philosophical Logic, 23:139–171, 1994.
Bealer, George. Propositions. Mind, 107:1–32, 1998.
Becker, Oskar. Zur Logik der Modalitäten. Jahrbuch für Philosophie und
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

phänomenologische Forschung, 11:497–548, 1930.

Belardinelli, Francesco. Hans van Ditmarsch, and Wiebe van der Hoek.
Second-order propositional announcement logic. In John Thangarajah,
Karl Tuyls, Catholijn Jonker, and Stacy Marsella, editors, Proceedings of the
15th International Conference on Autonomous Agents and Multiagent Sys-
tems (AAMAS 2016), pages 635–643. New York: Association for Computing
Machinery, 2016.
Belardinelli, Francesco. Wiebe van der Hoek, and Louwe B. Kuijer. Second-
order propositional modal logic: Expressiveness and completeness results.
Artificial Intelligence, 263:3–45, 2018.
Besnard, Philippe. Jean-Marc Guinnebault, and Emmanuel Mayer. Proposi-
tional quantification for conditional logic. In Dov M. Gabbay, Rudolf Kruse,
Andreas Nonnengart, and Hans Jürgen Ohlbach, editors, Qualitative and
Quantitative Practical Reasoning, Lecture Notes in Computer Science 1244,
pages 183–197. Berlin: Springer, 1997.
 96 References

Blackburn, Patrick. Maarten de Rijke, and Yde Venema. Modal Logic, vol-
ume 53 of Cambridge Tracts in Theoretical Computer Science. Cambridge:
Cambridge University Press, 2001.
Blackburn, Patrick. Torben Braüner, and Julie Lundbak Kofod. Remarks on
hybrid modal logic with propositional quantifiers. In Peter Hasle, David
Jakobsen, and Peter Øhrstrøm, editors, The Metaphysics of Time: Themes
from Prior, pages 401–426. Aalborg: Aalborg Universitetsforlag, 2020.
Boolos, George. To be is to be a value of a variable (or to be some values of
some variables). The Journal of Philosophy, 81:430–449, 1984.
Boolos, George. The Logic of Provability. Cambridge: Cambridge University
Press, 1985.
Bradfield, Julian. and Colin Stirling. Modal µ-calculi. In Patrick Blackburn,
Johan van Benthem, and Frank Wolter, editors, Handbook of Modal Logic,
pages 721–756. Amsterdam: Elsevier, 2007.
Brentano, Franz. The True and the Evident. London: Routledge and Kegan
Paul, 1966.
Büchi, J. Richard. On a decision method in restricted second order arithmetic.
In Ernest Nagel, Patrick Suppes, and Alfred Tarski, editors, Logic, Meth-
odology and Philosophy of Science: Proceedings of the 1960 International
Congress, pages 1–11. Redwood City, CA: Stanford University Press, 1962.
Bull, R. A. On possible worlds in propositional calculi. Theoria, 3:171–182,
1968.
Bull, R. A. On modal logics with propositional quantifiers. The Journal of
Symbolic Logic, 34:257–263, 1969.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Carnap, Rudolf. Introduction to Semantics. Cambridge, MA: Harvard Univer-

sity Press, 1942.
Carnap, Rudolf. Meaning and Necessity: A Study in Semantics and Modal
Logic. Chicago, IL: The University of Chicago Press, 1947.
Chagrov, Alexander. and Michael Zakharyaschev. Modal Logic, volume 35 of
Oxford Logic Guides. Oxford: Clarendon Press, 1997.
Chellas, Brian F. Modal Logic: An Introduction. Cambridge: Cambridge Uni-
versity Press, 1980.
Christensen, Ryan. Propositional quantification. Russell: The Journal of
Bertrand Russell Studies, 31:109–120, 2011.
Church, Alonzo. Review of The Liar by Alexandre Koyré. The Journal of
Symbolic Logic, 11:131, 1946.
Church, Alonzo. Introduction to Mathematical Logic. Princeton, NJ: Princeton
University Press, 1956.
Church, Alonzo. Mathematics and logic. In Ernest Nagel, Patrick Suppes,
and Alfred Tarski, editors, Logic, Methodology and Philosophy of Science.
 References 97

Proceedings of the 1960 International Congress, pages 181–186. Redwood

City, CA: Stanford University Press, 1962.
Church, Alonzo. Russell’s theory of identity of propositions. Philosophia
Naturalis, 21:513–522, 1984.
Church, Alonzo. Referee reports on Fitch’s “A Definition of Value”. In Joe
Salerno, editor, New Essays on the Knowability Paradox, pages 13–20.
Oxford: Oxford University Press, 2009 [1945].
Cohen, L. Jonathan. Can the logic of indirect discourse be formalised? The
Journal of Symbolic Logic, 22:225–232, 1957.
Copeland, B. Jack. Meredith, Prior, and the history of possible world semantics.
Synthese, 150:373–397, 2006.
Cresswell, M. J. Arnould Bayart’s modal completeness theorems. Logique et
Analyse, 229:89–142, 2015.
Crossley, John N., and Lloyd Humberstone. The logic of “actually”. Reports
on Mathematical Logic, 8:11–29, 1977.
Davey, B. A. and H. A. Priestley. Introduction to Lattices and Order. Cam-
bridge: Cambridge University Press, second edition, 2002.
deVries, Willem. Wilfrid Sellars. In Edward N. Zalta, editor, The Stanford
Encyclopedia of Philosophy. Metaphysics Research Lab, Stanford Univer-
sity, 2020.
Ding, Yifeng. On the logics with propositional quantifiers extending S5Π. In
Guram Bezhanishvili, Giovanna D’Agostino, George Metcalfe, and Thomas
Studer, editors, Advances in Modal Logic, volume 12, pages 219–235.
College Publications, 2018.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Ding, Yifeng. On the logic of belief and propositional quantification. Journal

of Philosophical Logic, 50:1143–1198, 2021a.
Ding, Yifeng. Propositional Quantification and Comparison in Modal Logic.
PhD thesis, University of California, Berkeley, 2021b.
Ding, Yifeng, and Wesley H. Holliday. Another problem in possible world
semantics. In Nicola Olivetti, Rineke Verbrugge, and Sara Negri, editors,
Advances in Modal Logic, volume 13. College Publications, 2020.
Dorr, Cian, John Hawthorne, and Juhani Yli-Vakkuri. The Bounds of Possibil-
ity: Puzzles of Modal Variation. Oxford: Oxford University Press, 2021.
Došen, Kosta. Duality between modal algebras and neighborhood frames.
Studia Logica, 48:219–234, 1989.
Dugundji, James. Topology. Boston: Allyn and Bacon, 1966.
Dunn, J. Michael, and Gary M. Hardegree. Algebraic Methods in Philosophical
Logic. Oxford: Clarendon Press, 2001.
Ferreira, Fernando. Comments on predicative logic. Journal of Philosophical
Logic, 35:1–8, 2006.
 98 References

Fine, Kit. For some proposition and so many possible worlds. PhD thesis,
University of Warwick, 1969.
Fine, Kit. Propositional quantifiers in modal logic. Theoria, 36:336–346,
1970.
Fine, Kit. In so many possible worlds. Notre Dame Journal of Formal Logic,
13:516–520, 1972.
Fine, Kit. An incomplete logic containing S4. Theoria, 40:23–29, 1974.
Fine, Kit. Properties, propositions and sets. Journal of Philosophical Logic,
6:135–191, 1977.
Fine, Kit. Failures of the interpolation lemma in quantified modal logic. The
Journal of Symbolic Logic, 44:201–206, 1979.
Fine, Kit. First-order modal theories II – Propositions. Studia Logica, 39:159–
202, 1980.
Fine, Kit. The permutation principle in quantificational logic. Journal of
Philosophical Logic, 12:33–37, 1983.
Fitch, Frederic B. A logical analysis of some value concepts. The Journal of
Symbolic Logic, 28:135–142, 1963. Reprinted in Salerno (2009).
Fitting, Melvin. Interpolation for first-order S5. The Journal of Symbolic Logic,
67:621–634, 2002.
Frege, Gottlob. Begriffsschrift, eine der arithmetischen nachgebildete Formel-
sprache des reinen Denkens. Halle a. S.: Louis Nebert, 1879.
Frege, Gottlob. Über Begriff und Gegenstand. Vierteljahresschrift für wis-
senschaftliche Philosophie, 16:192–205, 1892.
Frege, Gottlob. Posthumous Writings. Oxford: Basil Blackwell, 1979. Edited
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

by Hans Hermes, Friedrich Kambartel and Friedrich Kaulbart.

Frege, Gottlob. Philosophical and Mathematical Correspondence. Oxford:
Basil Blackwell, 1980. Edited by Gottfried Gabriel, Hans Hermes, Friedrich
Kambartel, Christian Thiel and Albert Veraart.
French, Tim. Bisimulation quantified modal logics: Decidability. In Guido
Governatori, Ian Hodkinson, and Yde Venema, editors, Advances in Modal
Logic, volume 6, pages 147–166. College Publications, 2006.
French, Tim, and Mark Reynolds. A sound and complete proof system for
QPTL. In Philippe Balbiani, Nobu-Yuki Suzuki, Frank Wolter, and Michael
Zakharyaschev, editors, Advances in Modal Logic, volume 4, pages 127–
147. College Publications, 2003.
Friedman, Harvey. One hundred and two problems in mathematical logic. The
Journal of Symbolic Logic, 40:113–129, 1975.
Fritz, Peter. Propositional contingentism. The Review of Symbolic Logic,
9:123–142, 2016.
 References 99

Fritz, Peter. Logics for propositional contingentism. The Review of Symbolic

Logic, 10:203–236, 2017.
Fritz, Peter. Propositional quantification in bimodal S5. Erkenntnis, 85:455–
465, 2020.
Fritz, Peter. Propositional potentialism. In Federico L. G. Faroldi and Fred-
erik Van De Putte, editors, Kit Fine on Truthmakers, Relevance and Non-
Classical Logic, volume 26 of Outstanding Contributions to Logic, pages
469–502. Cham: Springer, 2023a.
Fritz, Peter. Operands and instances. The Review of Symbolic Logic, 16:188–
209, 2023b.
Fritz, Peter. The Foundations of Modality: From Propositions to Possible
Worlds. Oxford: Oxford University Press, 2023c.
Fritz, Peter. Axiomatizability of propositionally quantified modal logics on
relational frames. The Journal of Symbolic Logic, forthcoming.
Fritz, Peter. Nonconservative extensions by propositional quantifiers and modal
incompleteness. Unpublished.
Fritz, Peter, and Nicholas K. Jones, editors. Higher-Order Metaphysics.
Oxford: Oxford University Press, 2024.
Fritz, Peter, and Harvey Lederman. Standard state space models of unaware-
ness. Electronic Proceedings in Theoretical Computer Science, 215:141–
158, 2016. Proceedings of the 15th Conference on Theoretical Aspects of
Rationality and Knowledge (TARK 2015).
Gabbay, Dov M. Montague type semantics for modal logics with proposi-
tional quantifiers. Zeitschrift für mathematische Logik und Grundlagen der
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Mathematik, 17:245–249, 1971.

Gabbay, Dov M. On 2nd order intuitionistic propositional calculus with full
comprehension. Archiv für mathematische Logik und Grundlagenforschung,
16:177–186, 1974.
Gallin, Daniel. Intensional and Higher-Order Modal Logic. Amsterdam:
North-Holland, 1975.
Garson, James W. Quantification in modal logic. In Dov M. Gabbay and Franz
Guenthner, editors, Handbook of Philosophical Logic, volume II, pages 249–
307. Dordrecht: D. Reidel, 1st edition, 1984.
Gerson, Martin. An extension of S4 complete for the neighbourhood seman-
tics but incomplete for the relational semantics. Studia Logica, 34:333–342,
1975a.
Gerson, Martin. The inadequacy of the neighbourhood semantics for modal
logic. The Journal of Symbolic Logic, 40:141–148, 1975b.
 100 References

Gerson, Martin. A neighbourhood frame for T with no equivalent relational

frame. Zeitschrift für mathematische Logik und Grundlagen der Mathematik,
22:29–34, 1976.
Ghilardi, Silvio, and Marek Zawadowski. Undefinability of propositional quan-
tifiers in the modal system S4. Studia Logica, 55:259–271, 1995.
Givant, Steven, and Paul Halmos. Introduction to Boolean Algebras. New
York: Springer, 2009.
Goldblatt, Robert. The McKinsey axiom is not canonical. The Journal of
Symbolic Logic, 56:554–562, 1991.
Goldblatt, Robert. Quantifiers, Propositions and Identity. Admissible Seman-
tics for Quantified Modal and Substructural Logics. Cambridge: Cambridge
University Press, 2011.
Goldblatt, Robert, and Michael Kane. An admissible semantics for proposition-
ally quantified relevant logic. Journal of Philosophical Logic, 39:73–100,
2010.
Grover, Dorothy L. Topics in Propositional Quantification. PhD thesis, Uni-
versity of Pittsburgh, 1970.
Grover, Dorothy L. Propositional quantifiers. Journal of Philosophical Logic,
1:111–136, 1972. Reprinted in Grover (1992).
Grover, Dorothy L. Propositional quantification and quotation contexts. In
Hugues Leblanc, editor, Truth, Syntax and Modality, pages 101–110.
Amsterdam: North-Holland, 1973.
Grover, Dorothy L. A Prosentential Theory of Truth. Princeton: Princeton
University Press, 1992.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Grover, Dorothy L., Joseph L. Camp, Jr., and Nuel D. Belnap, Jr. A prosen-
tential theory of truth. Philosophical Studies, 27:73–125, 1975. Reprinted in
Grover (1992).
Gurevich, Yuri, and Saharon Shelah. Monadic theory of order and topology in
ZFC. Annals of Mathematical Logic, 23:179–198, 1983.
Gärdenfors, Peter. On the extensions of S5. Notre Dame Journal of Formal
Logic, 414:277–280, 1973.
Gödel, Kurt. Über formal unentscheidbare Sätze der Principia Mathematica
und verwandter Systeme I. Monatshefte für Mathematik und Physik, 38:173–
198, 1931.
Halbach, Volker, and Philip Welch. Necessities and necessary truths: A pro-
legomenon to the use of modal logic in the analysis of intensional notions.
Mind, 118:71–100, 2009.
Halmos, Paul, and Steven Givant. Logic as Algebra. The Mathematical Asso-
ciation of America, 1998.
 References 101

Halpern, Joseph Y., and Leandro C. Rêgo. Reasoning about knowledge of

unawareness. Games and Economic Behaviour, 67:503–525, 2009.
Halpern, Joseph Y., and Leandro C. Rêgo. Reasoning about knowledge of
unawareness revisited. Mathematical Social Sciences, 65:73–84, 2013.
Hansson, Bengt, and Peter Gärdenfors. A guide to intensional semantics. In
Modality, Morality and Other Problems of Sense and Nonsense: Essays ded-
icated to Sören Halldén, pages 151–167. Lund: CWK Gleerup Bokförlag,
1973.
Hawthorne, John. Knowledge and Lotteries. Oxford: Clarendon Press, 2004.
Heidelberger, Herbert. The indispensability of truth. American Philosophical
Quarterly, 5:212–217, 1968.
Henkin, Leon. Completeness in the theory of types. The Journal of Symbolic
Logic, 15:81–91, 1950.
Henkin, Leon. A theory of propositional types. Fundamenta Mathematicae,
52:323–344, 1963.
Hilbert, D., and W. Ackermann. Grundzüge der theoretischen Logik. Berlin:
Springer, second edition, 1938.
Holliday, Wesley H. A note on algebraic semantics for S5 with propositional
quantifiers. Notre Dame Journal of Formal Logic, 60:311–332, 2019.
Holliday, Wesley H. Possibility semantics. In Melvin Fitting, editor, Selected
Topics from Contemporary Logics, pages 363–476. London: College Publi-
cations, 2021.
Holliday, Wesley H. Possibility frames and forcing for modal logic. The
Australasian Journal of Logic, forthcoming.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Holliday, Wesley H., and Tadeusz Litak. One modal logic to rule them all? In
Guram Bezhanishvili, Giovanna D’Agostino, George Metcalfe, and Thomas
Studer, editors, Advances in Modal Logic, volume 12, pages 367–386.
College Publications, 2018.
Hughes, G. E., and M. J. Cresswell. A New Introduction to Modal Logic.
London: Routledge, 1996.
Humberstone, Lloyd. From worlds to possibilities. Journal of Philosophical
Logic, 10:313–339, 1981.
Jónnson, Bjarni, and Alfred Tarski. Boolean algebras with operators, part I.
American Journal of Mathematics, 73:891–939, 1951.
Jónnson, Bjarni, and Alfred Tarski. Boolean algebras with operators, part II.
American Journal of Mathematics, 74:127–162, 1952.
Kaminski, Michael, and Michael Tiomkin. The expressive power of second-
order propositional modal logic. Notre Dame Journal of Formal Logic,
37:35–43, 1996.
 102 References

Kaplan, David. S5 with multiple possibility. The Journal of Symbolic Logic,

35:355–356, 1970a.
Kaplan, David. S5 with quantifiable propositional variables. The Journal of
Symbolic Logic, 35:355, 1970b.
Kaplan, David. Demonstratives. In Joseph Almog, John Perry, and Howard
Wettstein, editors, Themes from Kaplan, pages 481–563. Oxford: Oxford
University Press, 1989 [1977]. Completed and circulated in mimeograph in
the published form in 1977.
Kaplan, David. A problem in possible-world semantics. In Walter Sinnott-
Armstrong, Diana Raffman, and Nicholas Asher, editors, Modality, Moral-
ity, and Belief, pages 41–52. Cambridge: Cambridge University Press,
1995.
Kesten, Yonit, and Amir Pnueli. Complete proof system for QPTL. Journal of
Logic and Computation, 12:701–745, 2002.
Koyré, Alexandre. The liar. Philosophy and Phenomenological Research,
6:344–362, 1946.
Kreisel, Georg. Monadic operators defined by means of propositional quan-
tification in intuitionistic logic. Reports on Mathematical Logic, 12:9–15,
1981.
Kremer, Philip. Quantifying over propositions in relevance logic: Nonaxioma-
tisability of primary interpretations of ∀p and ∃p. The Journal of Symbolic
Logic, 58:334–349, 1993.
Kremer, Philip. Defining relevant implication in a propositionally quantified
S4. The Journal of Symbolic Logic, 62:1057–1069, 1997a.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Kremer, Philip. On the complexity of propositional quantification in intuition-

istic logic. The Journal of Symbolic Logic, 62:529–544, 1997b.
Kremer, Philip. Propositional quantification in the topological semantics for
S4. Notre Dame Journal of Formal Logic, 38:295–313, 1997c.
Kremer, Philip. Completeness of second-order propositional S4 and H in
topological semantics. The Review of Symbolic Logic, 11:507–518, 2018.
Kripke, Saul A. A completeness theorem in modal logic. The Journal of
Symbolic Logic, 24:1–14, 1959.
Kripke, Saul A. Semantical analysis of modal logic I: Normal modal prop-
ositional calculi. Zeitschrift für mathematische Logik und Grundlagen der
Mathematik, 9:67–96, 1963a.
Kripke, Saul A. Semantical considerations on modal logic. Acta Philosophica
Fennica, 16:83–94, 1963b.
Kripke, Saul A. Semantical analysis of modal logic II: Non-normal modal prop-
ositional calculi. In J . W. Addison, Leon Henkin, and Alfred Tarski, editors,
 References 103

The Theory of Models. Proceedings of the 1963 International Symposium at

Berkeley, pages 206–220. Amsterdam: North-Holland, 1965.
Kripke, Saul A. Review of Algebraic semantics for modal logics II by E. J.
Lemmon. Mathematical Reviews, 0205835 (34 #5661), 1967.
Kripke, Saul A. Is there a problem about substitutional quantification? In
Gareth Evans and John McDowell, editors, Truth and Meaning: Essays in
Semantics, pages 325–419. Oxford: Clarendon Press, 1976.
Kripke, Saul A. Naming and Necessity. Cambridge, MA: Harvard Univer-
sity Press, 1980 [1972]. First published in Semantics of Natural Language,
edited by Donald Davidson and Gilbert Harman, pages 253–355, 763–769,
Dordrecht: D. Reidel, 1972.
Kripke, Saul A. Review of Failures of the interpolation lemma in quantified
modal logic by Kit Fine. The Journal of Symbolic Logic, 48:486–488, 1983.
Kuhn, Steven. A simple embedding of T into double S5. Notre Dame Journal
of Formal Logic, 45:13–18, 2004.
Künne, Wolfgang. Conceptions of Truth. Oxford: Clarendon Press, 2003.
Lemmon, E. J. The “Lemmon Notes”. An Introduction to Modal Logic. Oxford:
Basil Blackwell, 1977 [1966]. In collaboration with Dana Scott, edited by
Krister Segerberg, completed and circulated in 1966.
Lewis, Clarence Irving, and Cooper Harold Langford. Symbolic Logic. New
York, NY: Dover, second edition, 1959 [1932]. Republication of the first
edition published by The Century Company in 1932.
Lewis, David. Counterfactuals. Oxford: Basil Blackwell, 1973.
Lewitzka, Steffen. Denotational semantics for modal systems S3–S5 extended
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

by axioms for propositional quantifiers and identity. Studia Logica, 103:507–

544, 2015.
Lesniewski, Stanisław. Grundzüge eines neuen Systems der Grundlagen der
Mathematik. Fundamenta Mathematicae, 14:1–81, 1929.
Lindström, Sten. Possible worlds semantics and the liar. In Joseph Almog and
Paolo Leonardi, editors, The Philosophy of David Kaplan, pages 93–108.
Oxford: Oxford University Press, 2009.
Litak, Tadeusz. Modal incompleteness revisited. Studia Logica, 76:329–342,
2004.
Lokhorst, Gert-Jan C. Propositional quantifiers in deontic logic. In Lou Goble
and John-Jules Meyer, editors, Deontic Logic and Artificial Normative Sys-
tems, DEON 2006, volume 4048 of Lecture Notes in Artificial Intelligence,
pages 201–209. Berlin: Springer, 2006.
Łoś, Jerzy. Logiki wielowartościowe a formalizacja funkcji intensjonalnych.
Kwartalnik Filozoficzny, 17:59–78, 1948.
 104 References

Łukasiewicz, Jan. On variable functors of propositional arguments. Proceed-

ings of the Royal Irish Academy, 54:25–35, 1951.
Łukasiewicz, Jan, and Alfred Tarski. Untersuchungen über den Aussagen-
kalkül. Comptes Rendus des séances de la Societé des Sciences et des
Lettres de Varsovie, 23:30–50, 1930. Reprinted in English translation by
J. H. Woodger as “Investigations into the Sentential Calculus” in Alfred
Tarski, Logic, Semantics, Metamathematics, pages 38–59, Oxford: Claren-
don Press, 1956.
Löb, M. H. Embedding first order predicate logic in fragments of intuitionistic
logic. The Journal of Symbolic Logic, 41:705–718, 1976.
Löwenheim, Leopold. Über Möglichkeiten im Relativkalkül. Mathematische
Annalen, 76:447–470, 1915.
MacNeille, H. M. Partially ordered sets. Transactions of the American Mathe-
matical Society, 42:416–460, 1937.
Makinson, David. On some completeness theorems in modal logic. Zeitschrift
für mathematische Logik und Grundlagen der Mathematik, 12:379–384,
1966.
Makinson, David. Some embedding theorems for modal logic. Notre Dame
Journal of Formal Logic, 12:252–254, 1971.
Makinson, David. A warning about the choice of primitive operators in modal
logic. Journal of Philosophical Logic, 2:193–196, 1973.
McKinsey, J. C. C. A solution of the decision problem for the Lewis systems
S2 and S4, with an application to topology. The Journal of Symbolic Logic,
6:117–134, 1941.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

McKinsey, J. C. C. On the syntactical construction of systems of modal logic.

The Journal of Symbolic Logic, 10:83–94, 1945.
McKinsey, J. C. C., and Alfred Tarski. The algebra of topology. Annals of
Mathematics, 45:141–191, 1944.
McKinsey, J. C. C., and Alfred Tarski. Some theorems about the sentential
calculi of Lewis and Heyting. The Journal of Symbolic Logic, 13:1–15, 1948.
Menzel, Christopher. Pure logic and “higher-order” metaphysics. In Peter Fritz
and Nicholas K. Jones, editors, Higher-Order Metaphysics. Oxford: Oxford
University Press, 2024.
Menzel, Christopher, and Edward N. Zalta. The fundamental theorem of world
theory. Journal of Philosophical Logic, 43:333–363, 2014.
Meredith, C. A. On an extended system of the propositional calculus. Proceed-
ings of the Royal Irish Academy, 54:37–47, 1951.
Montague, Richard. Universal grammar. Theoria, 36:373–398, 1970.
Muskens, Reinhard. Intensional models for the theory of types. The Journal of
Symbolic Logic, 72:98–118, 2007.
 References 105

Myhill, John. On the interpretation of the sign ‘⊃’. The Journal of Symbolic
Logic, 18:60–62, 1953.
Myhill, John. Review of Formal Logic by A. N. Prior. The Philosophical
Review, 66:117–120, 1957.
Myhill, John. Problems arising in the formalization of intensional logic.
Logique et Analyse, 1:78–83, 1958.
Nagle, Michael C., and S. K. Thomason. The extensions of the modal logic K5.
The Journal of Symbolic Logic, 50:102–109, 1985.
Pacuit, Eric. Neighborhood Semantics for Modal Logic. Short Textbooks in
Logic. Cham: Springer, 2017.
Parry, W. T. Zum Lewisschen Aussagenkalkül. Ergebnisse eines Mathematis-
chen Kolloquiums, 4:15–16, 1933.
Pascucci, Matteo. Propositional quantifiers in labelled natural deduction for
normal modal logic. Logic Journal of the IGPL, 27:865–894, 2019.
Pitts, Andrew M. On an interpretation of second order quantification in first
order intuitionistic propositional logic. The Journal of Symbolic Logic,
57:33–52, 1992.
Plantinga, Alvin. The Nature of Necessity. Oxford: Clarendon Press, 1974.
Połacik, Tomasz. Operators defined by propositional quantification and their
interpretation over Cantor space. Reports on Mathematical Logic, 27:67–79,
1993.
Połacik, Tomasz. Pitts’ quantifiers are not topological quantification. Notre
Dame Journal of Formal Logic, 39:531–545, 1998a.
Połacik, Tomasz. Propositional quantification in the monadic fragment of
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

intuitionistic logic. The Journal of Symbolic Logic, 63:269–300, 1998b.

Prawitz, Dag. Natural Deduction. A Proof-Theoretical Study, volume 3
of Stockhom Studies in Philosophy. Stockholm: Almqvist & Wiksell,
1965.
Prior, Arthur N. Formal Logic. Oxford: Clarendon Press, 1955.
Prior, Arthur N. Time and Modality. Oxford: Clarendon Press, 1957.
Prior, Arthur N. Escapism: the logical basis of ethics. In A. I. Melden, editor,
Essays in Moral Philosophy. Seattle: University of Washington Press, 1958a.
Prior, Arthur N. Epimenides the Cretan. The Journal of Symbolic Logic,
23:261–266, 1958b.
Prior, Arthur N. On a family of paradoxes. Notre Dame Journal of Formal
Logic, 2:16–32, 1961.
Prior, Arthur N. Wspólczesna logika w Anglii. Ruch Filozoficzny, 21:251–256,
1962.
Prior, Arthur N. Past, Present and Future. Oxford: Clarendon Press, 1967.
Prior, Arthur N. Egocentric logic. Noûs, 2:191–207, 1968.
 106 References

Prior, Arthur N. Objects of Thought. Oxford: Clarendon Press, 1971. Edited by

P. T. Geach and A. J. P. Kenny.
Prior, Arthur N., and Kit Fine. Worlds, Times and Selves. London: Duckworth,
1977.
Prucnal, Tadeusz. On two problems of Harvey Friedman. Studia Logica,
38:247–262, 1979.
Quine, W. V. Philosophy of Logic. Cambridge, MA: Harvard University Press,
second edition, 1986 [1970]. First edition published in 1970.
Rabin, Michael O. Decidability of second-order theories and automata on
infinite trees. Transactions of the American Mathematical Society, 141:1–35,
1969.
Ramsey, Frank P. Critical notice of Tractatus Logico-Philosophicus by Ludwig
Wittgenstein. Mind, 32:465–478, 1923. Reprinted in Ramsey (1931).
Ramsey, Frank P. Facts and propositions. Proceedings of the Aristotelian
Society, Supplementary Volumes, 7:153–170, 1927. Reprinted in Ramsey
(1931).
Ramsey, Frank P. The Foundations of Mathematics, and other Logical Essays.
London: Routledge and Kegan Paul, 1931. Edited by R. B. Braithwaite and
G. E. Moore.
Ramsey, Frank P. On Truth. Original Manuscript Materials (1927–1929) from
the Ramsey Collection at the University of Pittsburgh. Dordrecht: Kluwer,
1991. Edited by Nicholas Rescher and Ulrich Majer.
Rasiowa, Helena, and Roman Sikorski. The Mathematics of Metamathematics.
Warsaw: Państwowe Wydawnictwo Naukowe, 1963.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Rönnedal, Daniel. Semantic tableau versions of some normal modal systems

with propositional quantifiers. Organon F, 26:505–536, 2019.
Rönnedal, Daniel. The moral law and the good in temporal modal deontic logic
with propositional quantifiers. Australasian Journal of Logic, 17:22–69,
2020.
Ross, Alf. Imperatives and logic. Theoria, 7:53–71, 1941.
Routley, Richard, and Robert K. Meyer. The semantics of entailment. In
Hugues Leblanc, editor, Truth, Syntax and Modality, pages 199–243.
Amsterdam: North-Holland, 1973.
Russell, Bertrand. The Principles of Mathematics. Cambridge: University
Press, 1903.
Russell, Bertrand. The theory of implication. American Journal of Mathemat-
ics, 28:159–202, 1906.
Russell, Bertrand. Mathematical logic as based on the theory of types. Ameri-
can Journal of Mathematics, 30:222–262, 1908.
 References 107

Russell, Bertrand. Toward “Principia Mathematica”. 1905–08. The Collected

Papers of Bertrand Russell. London: Routledge, 2014. Edited by Gregory H.
Moore.
Salerno, Joe, editor. New Essays on the Knowability Paradox. Oxford: Oxford
University Press, 2009.
Scott, Dana. Advice on modal logic. In Karel Lambert, editor, Philosophical
Problems in Logic. Some Recent Developments, pages 143–173. Dordrecht:
D. Reidel, 1970.
Scroggs, Schiller Joe. Extensions of the Lewis system S5. The Journal of
Symbolic Logic, 16:112–120, 1951.
Segerberg, Krister. An Essay in Classical Modal Logic, volume 13 of
Filosofiska Studier. Uppsala: Uppsala Universitet, 1971.
Segerberg, Krister. Classical Propositional Operators. An Exercise in the
Foundations of Logic. Oxford: Clarendon Press, 1982.
Sellars, Wilfrid. Grammar and existence: A preface to ontology. Mind, 69:499–
533, 1960.
Shapiro, Stewart. Foundations without Foundationalism: A Case for Second-
order Logic. Oxford: Clarendon Press, 1991.
Shelah, Saharon. The monadic theory of order. Annals of Mathematics,
102:379–419, 1975.
Shukla, Ankit, Armin Biere, Martina Seidl, and Luca Pulina. A survey on
applications of quantified boolean formulas. In IEEE 31st International Con-
ference on Tools with Artificial Intelligence (ICTAI 2019), pages 78–84.
IEEE Computer Society, Conference Publishing Services, 2019.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Skvortsov, D. Non-axiomatizable second order intuitionistic propositional

logic. Annals of Pure and Applied Logic, 86:33–46, 1997.
Sobolev, S. K. The intuitionistic propositional calculus with quantifiers (in
Russian). Matematicheskie Zametki, 22:69–76, 1977.
Solovay, Robert M. Provability interpretations of modal logic. Israel Journal
of Mathematics, 25:287–304, 1976.
Stalnaker, Robert. A theory of conditionals. In Nicholas Rescher, editor, Studies
in Logical Theory, pages 98–112. Oxford: Blackwell, 1968.
Stalnaker, Robert. Possible worlds. Noûs, 10:65–75, 1976.
Stalnaker, Robert. Mere Possibilities. Princeton: Princeton University Press,
2012.
Steinsvold, Christopher. Some formal semantics for epistemic modesty. Logic
and Logical Philosophy, 29:381–413, 2020.
Stone, M. H. The theory of representation for Boolean algebras. Transactions
of the American Mathematical Society, 40:37–111, 1936.
 108 References

Suszko, R. Review of Many-Valued Logics and the Formalization of Inten-

sional Functions by Jerzy Łoś. The Journal of Symbolic Logic, 14:64–65,
1949.
Sørensen, Morten H. and Pawel Urzyczyn. A syntactic embedding of predicate
logic into second-order propositional logic. Notre Dame Journal of Formal
Logic, 51:457–473, 2010.
Tajtelbaum-Tarski, Alfred. O wyrazie pierwotnym logistyki. Przegląd Filo-
zoficzny, 26:68–89, 1923. Reprinted in English translation by J. H. Woodger
as “On the Primitive Term of Logistic” in Alfred Tarski, Logic, Semantics,
Metamathematics, pages 1–23, Oxford: Clarendon Press, 1956.
Tang, Tsao-Chen. Algebraic postulates and a geometric interpretation for the
Lewis calculus of strict implication. Bulletin of the American Mathematical
Society, 44:737–744, 1938.
Tarski, Alfred. Zur Grundlegung der Bool’schen Algebra. Fundamenta Math-
ematicae, 24:177–198, 1935.
Tarski, Alfred. Über additive und multiplikative Mengenkörper und Mengen-
funktionen. Sprawozdania z Posiedzeń Towarzystwa Naukowego Warsza-
wskiego, Wydział III Nauk Matematyczno-fizycznych (=Comptes Rendus des
Séances de la Société des Sciences et des Lettres de Varsovie, Classe III),
30:151–181, 1937.
Tarski, Alfred. The concept of truth in formalized languages. In John Corcoran,
editor, Logic, Semantics, Metamathematics, pages 152–278. Indianapolis:
Hackett, 1983 [1933]. Translation of an article published in Polish in 1933
and German in 1935.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

Tarski, Alfred. On the concept of following logically. History and Philosophy of

Logic, 23:155–196, 2002 [1936]. Originally published in Polish and German
in 1936.
ten Cate, Balder. Expressivity of second order propositional modal logic.
Journal of Philosophical Logic, 35:209–223, 2006.
Tharp, Leslie. Three theorems of metaphysics. Synthese, 81:207–214, 1989.
Thomason, S. K. Semantic analysis of tense logics. The Journal of Symbolic
Logic, 37:150–158, 1972.
Thomason, S. K. An incompleteness theorem in modal logic. Theoria, 40:30–
34, 1974.
Thomason, S. K. Categories of frames for modal logic. The Journal of Symbolic
Logic, 40:439–442, 1975.
Troelstra, A. S. On a second order propositional operator in intuitionistic logic.
Studia Logica, 40:113–139, 1981.
Uzquiano, Gabriel. Elusive propositions. Journal of Philosophical Logic,
50:705–725, 2021.
 References 109

van Benthem, Johan. Modal Logic and Classical Logic. Naples: Bibliopolis,
1983.
van Benthem, Johan, and Guram Bezhanishvili. Modal logics of space. In
Marco Aiello, Ian Pratt-Hartmann, and Johan van Benthem, editors, Hand-
book of Spatial Logics, pages 217–298. Dordrecht: Springer, 2007.
van Ditmarsch, Hans. To be announced. Information and Computation, 292:1–
42, 2023.
van Inwagen, Peter. Generalizations of homophonic truth-sentences. In Richard
Schantz, editor, What Is Truth?, volume 1 of Current Issues in Theoretical
Philosophy, pages 205–222. Berlin: De Gruyter, 2002.
Whitehead, Alfred North, and Bertrand Russell. Principia Mathematica, Vol-
umes 1–3. Cambridge: Cambridge University Press, 1910–1913.
Williamson, Timothy. Modal Logic as Metaphysics. Oxford: Oxford University
Press, 2013.
Williamson, Timothy. Menzel on pure logic and higher-order metaphysics.
In Peter Fritz and Nicholas K. Jones, editors, Higher-Order Metaphysics.
Oxford: Oxford University Press, 2024.
Wittgenstein, Ludwig. Logisch-philosophische Abhandlung. In Wilhelm
Oswald, editor, Annalen der Naturphilosophie, volume 14, pages 185–262.
Leipzig: Unesma, 1921.
Zach, Richard. Decidability of quantified propositional intuitionistic logic and
S4 on trees of height and arity ≤ ω. Journal of Philosophical Logic, 33:155–
164, 2004.
Zdanowski, Konrad. On second order intuitionistic propositional logic without
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

a universal quantifier. The Journal of Symbolic Logic, 74:157–167, 2009.

Zhao, Zhiguang. Sahlqvist correspondence theory for second-order proposi-
tional modal logic. Journal of Logic and Computation, 33:577–598, 2023.
 Acknowledgments
For questions, comments, and discussion, thanks to Andrew Bacon, Adam
Bjorndahl, Sam Carter, Nevin Climenhaga, Yifeng Ding, Kit Fine, Salvatore
Florio, Dmitri Gallow, Simon Goldstein, Jeremy Goodman, Wesley Holliday,
Yipu Li, Gillian Russell, Juhani Yli-Vakkuri, and the participants at courses
on propositional quantifiers at the Nordic Logic Summer School 2022 held at
the University of Bergen, at NASSLLI 2022 held at the University of Southern
California, at ESSLLI 2023 held at the University of Ljubljana, and at Beijing
Normal University in 2022 and 2023. Special thanks to Valentin Goranko,
Lloyd Humberstone, and two anonymous reviewers who read drafts and sent
me detailed comments. Finally, I am grateful to the editors of the Logic and
Philosophy series of Elements, Bradley Armour-Garb and Frederick Kroon,
for the opportunity to publish in this series, and for their patience.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Philosophy and Logic

Bradley Armour-Garb
SUNY Albany
Bradley Armour-Garb is chair and Professor of Philosophy at SUNY Albany. His books
include The Law of Non-Contradiction (co-edited with Graham Priest and J. C. Beall,
2004), Deflationary Truth and Deflationism and Paradox (both co-edited with J. C. Beall,
2005), Pretense and Pathology (with James Woodbridge, Cambridge University Press,
2015), Reflections on the Liar (2017), and Fictionalism in Philosophy (co-edited with Fred
Kroon, 2020).

Frederick Kroon
The University of Auckland
Frederick Kroon is Emeritus Professor of Philosophy at the University of Auckland. He has
authored numerous papers in formal and philosophical logic, ethics, philosophy of
language, and metaphysics, and is the author of A Critical Introduction to Fictionalism
(with Stuart Brock and Jonathan McKeown-Green, 2018).

About the Series

This Cambridge Elements series provides an extensive overview of the many and varied
connections between philosophy and logic. Distinguished authors provide an up-to-date
summary of the results of current research in their fields and give their own take on what
they believe are the most significant debates influencing research, drawing original
conclusions.
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press
 Philosophy and Logic

Elements in the Series

Set Theory
John P. Burgess
Higher-Order Logic and Type Theory
John L. Bell
Proofs and Models in Philosophical Logic
Greg Restall
Gödel’s Incompleteness Theorems
Juliette Kennedy
Classical First-Order Logic
Stewart Shapiro and Teresa Kouri Kissel
Logical Consequence
Gila Sher
Temporal Logics
Valentin Goranko
The Many Faces of Impossibility
Koji Tanaka and Alexander Sandgren
Relevance Logic
Shay Allen Logan
Propositional Quantifiers
Peter Fritz
https://doi.org/10.1017/9781009177740 Published online by Cambridge University Press

A full series listing is available at: www.cambridge.org/EPL