1 Introduction
Arguably, the most influential logical analysis of conditional statements is rooted in the work of the philosopher David Lewis [Reference Lewis18, Reference Lewis19]. Inspired by Stalnaker [Reference Stalnaker and Rescher25], Lewis introduced a hierarchy of logics for counterfactual conditionals based on possible worlds semantics. A counterfactual is a statement of the form
to be read as “if it were the case that
$\varphi $
, it would be the case that
$\psi $
.” In a nutshell, Lewis’s idea is that
is true at a world w if, in the closest worlds to w in which
$\varphi $
holds,
$\psi $
also holds. The notion of closeness among worlds is formalized by Lewis in different, but equivalent, ways: by assigning to each world a (pre)order that ranks other worlds by closeness, or a selection function picking the closest worlds in which a certain formula is true, or, most famously, a system of nested spheres, where smaller spheres contain worlds closer to w.
While the literature on Lewis’s logics is vast, their mathematical analysis has only gained traction in recent years, with a proof-theoretical analysis carried on, e.g., in [Reference Girlando, Lellmann, Olivetti and Pozzato15, Reference Girlando, Negri and Sbardolini16, Reference Negri and Sbardolini22], and the recent [Reference Rosella and Ugolini24] where these authors study Lewis’s logics from the logico-algebraic perspective. Grounding on the results in [Reference Rosella and Ugolini24], we will deepen the mathematical foundations of Lewis’s logics by providing categorical dualities and equivalences between all the main models of Lewis’s logics and their algebraic semantics.
In more detail, in [Reference Rosella and Ugolini24] Lewis’s logics are considered as consequence relations, instead of just sets of theorems, and two different kinds of derivation are considered, depending on whether the deductive rules are applied only to theorems (giving a relatively weaker calculus) or to all derivations (yielding a stronger calculus). As it is the case for modal logic [Reference Blackburn, de Rijke and Venema2, Reference Wen27], these two choices correspond to considering two different consequence relations on the original possible-worlds models: a global and a local one (which is the one commonly considered in the literature). The stronger calculi, strongly complete with respect to the global consequence relation, are (strongly) algebraizable in the sense of Blok–Pigozzi [Reference Blok and Pigozzi3], by varieties of Boolean algebras with an extra binary operator
, that yield the equivalent algebraic semantics. The main variety is called
$\mathsf {VA}$
in [Reference Rosella and Ugolini24] (short for variably strict conditional algebras). Intuitively, this means that the global consequence relation is fully and faithfully interpreted in the semantic consequence of
$\mathsf {VA}$
. The local consequence relation is instead shown to be the logic preserving the degrees of truth of the same algebraic models, which essentially means that the local consequence is fully and faithfully interpreted by the lattice-order relation which underlies the algebras in
$\mathsf {VA}$
. Thus, this variety can be meaningfully used to study both versions of Lewis’s logics.
In this work, we circle back from the algebras to the original intended models of Lewis. It will become clear that to properly consider infinite models one should not simply consider sets of worlds, but topological spaces. In this setting, the subsets of the universe representing formulas, specifically, the clopen sets of the topology, play a central role. In more detail, we show three different dual categorical equivalences of
$\mathsf {VA}$
with respect to topological spaces based, respectively, on Lewis’s spheres, Stalnaker’s inspired selection functions, and Lewis’s preorders among worlds. The dualities we show are enrichments of Stone duality between Boolean algebras with homomorphisms and Stone spaces with continuous maps. The constructions we use to move between the different relational structures are based on those introduced by Lewis in [Reference Lewis18]; we here then provide also detailed proofs of some important facts that have been used without proof in the literature, regarding how to translate between these different models. Moreover, by the duality results we also derive alternative proofs of strong completeness for Lewis’s variably strict conditional logics with respect to their intended models.
The reader shall notice that the binary operator
does not straightforwardly inherit the plethora of results on modal operators; for the reader more expert on the algebraic perspective, the models are not Boolean algebras with an operator in the Jónsson–Tarski sense, and
cannot generally be recovered from a unary modal operator, thus these systems also do not inherit the advancements for logics of strict conditionals (conditionals defined by applying a unary modality to the material implication). While there is work on the duality theory of Boolean algebras (or more general algebras) with arbitrary operators [Reference Bezhanishvili, Bezhanishvili and de Groot1, Reference Došen8, Reference Gehrke and Harding11–Reference Gehrke and Priestley14], the existing approaches do not provide for free a duality-theoretic representation of the algebraic models we consider here. Moreover, we wish to point out that our focus is not just to provide a duality theoretic treatment of the algebraic models, but to build the foundations of a precise technical analysis of the existing models of Lewis’s logics, clarifying their mutual relationships.
Lastly, our results shed new light on the role of the limit assumption, a condition on sphere models extensively discussed in the literature. Essentially, it postulates that for each world w and each formula A that holds in some worlds sufficiently close to w (i.e., in some worlds within the system of spheres associated with w), there exists a set of closest worlds to w in which A holds. As a consequence of our theoretical work, we shall demonstrate that the models without the limit assumption are not really “seen” by Lewis’s logics. In particular, for every arbitrary Lewisian sphere model there is a sphere model satisfying the limit assumption which validates exactly the same formulas and deductions, since it carries the same algebraic structure to evaluate the formulas.
After a section of preliminaries, we will develop the duality between
$\mathsf {VA}$
and the topological version of models with the selection function in §3, with topological spheres in §4, and topological orders in §5. In §6, we discuss strong completeness results and the limit assumption.
2 Preliminaries
2.1 Lewis’s logics and their semantics
Lewis bases his interpretation of the counterfactual connective
on the intuitive idea that
is true at some world w if, in the closest worlds to w in which
$\varphi $
is true,
$\psi $
is also true. This leads to the definition of what Lewis calls a “variably strict conditional.” The term “variably” emphasizes that to evaluate counterfactuals with different antecedents at a world, one might need to evaluate the corresponding material implication across different sets of worlds. This also implies that, in general, a counterfactual
cannot be represented as
$\square (\varphi \to \psi )$
for some fixed modality
$\square $
. Lewis formalizes this intuition by assigning to each world w a relational structure intended to represent the relation of closeness relevant to the truth conditions of variably strict conditionals. This relational structure can be represented through different, yet equivalent under certain assumptions, formal frameworks: ordering relations, selection functions, or systems of spheres (see [Reference Lewis18, Reference Lewis19]).
Let us be more precise. The language
is obtained from a denumerable set of propositional variables by extending the language of classical propositional logic,
$\{\wedge , \vee , \neg , 0, 1\}$
, with a binary connective
. A formula
is read as “if it were the case that
$\varphi $
, then it would be the case that
$\psi $
.” As usual, one can define further classical connectives by:
$x \to y : = \neg x \lor y, \;\; x \leftrightarrow y : = (x \to y)\wedge (y \to x)$
. We denote with
the algebra of
-formulas over a fixed denumerable set of variables
$Var$
.
We now recall the three kinds of models used by Lewis for interpreting
-formulas. First, sphere models, which represent a closeness relation among worlds by assigning to each world w a chain of sets of worlds; the smaller a set is in this order, the closer its worlds are to w.
Definition 2.1. A sphere model
$\mathcal {M}$
is a tuple
$\mathcal {M}=\langle W, \mathcal {S}, v \rangle $
, where:
-
1. W is a non-empty set, whose elements are called (possible) worlds;
-
2.
$\mathcal {S}: W\to \mathcal{P}(\mathcal{P}(W))$
is a function assigning to each
$w \in W$
a non-empty chain
$\mathcal {S}(w)$
of subsets of W, i.e., for all
$X, Y \in \mathcal {S}(w)$
, either
$ X\subseteq Y$
or
$Y\subseteq X$
. -
3.
$v: Var \to \mathcal{P}(W)$
is an assignment of the variables to subsets of W, extending in the obvious way on Boolean formulas:
$$ \begin{align*} v(0) = \emptyset;\;\; v(1) = W;\;\; v(\varphi \wedge \psi) = v(\varphi)\cap v(\psi);\\v(\varphi \vee \psi) = v(\varphi)\cup v(\psi);\;\; v(\neg\varphi) = W\setminus v(\varphi) \ \ \end{align*} $$
and for the counterfactual connective:

We now consider the order models, which represent the closeness relation between worlds by assigning to each world w a total preorder that ranks worlds according to their closeness to w.
Definition 2.2. An order model
$\mathcal {M}$
is a tuple
$\mathcal {M}=\langle W, \leq , v \rangle $
, where:
-
1. W is a non-empty set, whose elements are called (possible) worlds;
-
2.
$\leq $
is a function assigning to each element
$w \in W$
a binary relation
$\leq _w \subseteq S_w \times S_w$
which is a total preorder (i.e., reflexive, transitive, and total) of some subset
$S_w \subseteq W$
. -
3.
$v: Var \to \mathcal{P}(W)$
is an assignment of the variables to subsets of W, extending in the obvious way on Boolean formulas, and for the counterfactual:
Lastly, we call
$\alpha $
-models (as in [Reference Lewis19]) the models based on selection functions. These represent the closeness relation among worlds in a more abstract way; specifically, they provide the machinery to select, given a formula
$\varphi $
and a world w, the closest worlds to w where
$\varphi $
is true, denoted as
$f(\varphi ,w)$
.
Definition 2.3. An
$\alpha $
-model
$\mathcal {M}$
is a tuple
$\mathcal {M}=\langle W, f, v \rangle $
, where:
-
1. W is a non-empty set, whose elements are called (possible) worlds;
-
2.
is a function mapping each pair made of a formula and world to a subset of W, satisfying the following constraints for all
and all
$w \in W$
:-
(a)
$f(\varphi , w) \subseteq v(\varphi );$
-
(b) if
$f(\varphi , w) \subseteq v(\psi )$
and
$f(\psi , w) \subseteq v(\varphi )$
, then
$f(\varphi , w)=f(\psi , w);$
-
(c) either
$f(\varphi \vee \psi , w)\subseteq v(\varphi )$
or
$f(\varphi \vee \psi , w)\subseteq v(\psi )$
or
$f(\varphi \vee \psi , w)=f(\varphi , w) \cup f(\psi , w)$
.
-
-
3.
$v: Var \to \mathcal{P}(W)$
is an assignment of the variables to subsets of W, extending in the usual way on Boolean connectives, and moreover
Let
$\mathcal {M}$
be a sphere model
$\mathcal {M}=\langle W, \mathcal {S}, v \rangle $
, an order model
$\mathcal {M}=\langle W, \leq , v \rangle $
, or an
$\alpha $
-model
$\mathcal {M}=\langle W, f, v \rangle $
, and consider a set of
-formulas
$\Gamma $
. We set
If
$\Gamma = \{\gamma \}$
, we drop the parentheses and write
$w \Vdash \gamma $
(or
$\mathcal {M} \Vdash \gamma $
) instead of
$w \Vdash \{\gamma \}$
(or
$\mathcal {M} \Vdash \{\gamma \}$
). It is sometimes convenient to stress to which universe a world belongs; given a model
$\mathcal {M}$
, we then write
Lewis ([Reference Lewis18, Reference Lewis19]) demonstrates the equivalence of these models with respect to the validity of
-formulas (i.e., formulas that are true in every world of the model). This shows that they are different formal representations of the same underlying intuitive semantic ideas. Later in this paper, we will show how some topological counterparts of these models are equivalent in a categorical sense to each other. This will deepen Lewis’s results and reveal the fundamental mathematical form of these equivalences. In order to do that, we recall the needed notions from the logico-algebraic interpretation of Lewis’s work introduced in [Reference Rosella and Ugolini24].
Since sphere models play a central role in Lewis’s book and are treated by him as the main intended models for counterfactuals, for the sake of readability we restrict our attention to sphere models for the next results. Nevertheless, the reader shall keep in mind that all that follows can be rephrased in terms of order models and
$\alpha $
-models as well.
We start by recalling the two different consequence relations over sphere models from [Reference Rosella and Ugolini24]: a local and a global one.
Definition 2.4. Let
$\mathfrak {S}$
be a class of sphere models.
-
1. We define the global
$\mathfrak {S}$
-consequence relation as:
$\Gamma \vDash _{\mathfrak {S},g} \varphi $
if and only if for all models
$\mathcal {M} \in \mathfrak {S}$
,
$\mathcal {M}\Vdash \Gamma $
implies
$\mathcal {M}\Vdash \varphi $
. -
2. We define the local
$\mathfrak {S}$
-consequence relation as:
$\Gamma \vDash _{\mathfrak {S},\ell } \varphi $
if and only if for all models
$\mathcal {M} \in \mathfrak {S}$
and all worlds
$w$
in
$\mathcal {M}$
,
$w\Vdash \Gamma $
implies
$w\Vdash \varphi $
.
When
$\mathfrak {S}$
is the class of all sphere models we write
$\vDash _{g}$
for
$\vDash _{\mathfrak {S},g}$
and
$\vDash _{\ell }$
for
$\vDash _{\mathfrak {S},\ell }$
.
The theorems of the associated logics,
$\texttt {GV}$
and
$\texttt {LV}$
, are the same (see [Reference Rosella and Ugolini24, theorem 2.4]) and coincide with the set of formulas true in all sphere models; i.e., the set of
-formulas
$\varphi $
such that
$\mathcal {M} \Vdash \varphi $
for all sphere models
$\mathcal {M}$
. Lewis indeed introduces axiomatic systems that are complete with respect to sphere models, and in [Reference Rosella and Ugolini24] these are adapted to define two different calculi,
$\texttt {LV}$
strongly complete with respect to the local consequence ([Reference Rosella and Ugolini24, theorem 3.8]),
$\texttt {GV}$
strongly complete with respect to the global consequence ([Reference Rosella and Ugolini24, theorem 3.23]). These systems share the same axioms, modus ponens, and essentially the same extra rule for counterfactuals, except that the latter in
$\texttt {LV}$
is only applied to theorems, while in
$\texttt {GV}$
it applies to all derivations. We denote their consequence relations as
$\vdash _{\texttt {LV}}$
and
$\vdash _{\texttt {GV}}$
, respectively. The interested reader is referred to [Reference Rosella and Ugolini24] for details on the axiomatic systems.
The set of theorems of
$\texttt {LV}$
and
$\texttt {GV}$
is claimed to be by Lewis the “weakest system that has any claim to be called the logic of conditionals” [Reference Lewis18, p. 80]. Lewis then identifies different classes of sphere models, which correspond to different notions of counterfactual conditionals. Among these, the following two are worth mentioning.
Definition 2.5. Let
$\mathcal {M}=\langle W, \mathcal {S}, v \rangle $
be a sphere model.
-
1.
$\mathcal {M}$
is centered if for each
$w \in W$
,
$\{w\} \in \mathcal {S}(w)$
. -
2.
$\mathcal {M}$
is Stalnakerian if for any
$w \in W$
, and any
-formula
$\varphi $
such that
$v(\varphi ) \cap \bigcup S(w) \neq \emptyset $
, there is some
$S \in S(w)$
and
$y \in W$
such that
$v(\varphi ) \cap S = \{y\}$
.
Centered sphere models represent the intended semantics of the system
$\texttt {VC}$
, which, according to Lewis [Reference Lewis19], is the “correct logic of counterfactuals”; Stalnakerian centered sphere models represent a semantics for Stalnaker’s logic of conditionals
$\texttt {VCS}$
[Reference Lewis19]. We call the associated logics
$\texttt {GVC}$
and
$\texttt {LVC}$
, and
$\texttt {GVCS}$
and
$\texttt {LVCS,}$
respectively. For those interested in the other classes of spheres and their corresponding axiomatic systems, we refer to [Reference Lewis19, Reference Rosella and Ugolini24].
2.2 Algebraic semantics: Lewis’s variably strict conditional algebras
We now recall the algebraic semantics of Lewis’s logics introduced in [Reference Rosella and Ugolini24]. Let us first set some notation. Roman bold letters will be used to represent algebras, while the corresponding Roman standard letters will denote their underlying domains. For instance, if
$\mathbf {A}$
is an algebra, then the symbol A will refer to its domain. Given an algebraic language
$\mathcal {L},$
recall that we write
${\mathbf{Fm}}_{\mathcal {L}}$
for its algebra of formulas written over a denumerable set of variables. An equation of the language
$\mathcal {L}$
(or an
$\mathcal {L}$
-equation for short) is a pair
$(p,q)$
of
$\mathcal {L}$
-formulas (i.e., elements of
$Fm_{\mathcal {L}}$
) that we write suggestively as
$p \approx q$
. We write
$Eq_{\mathcal {L}}$
for the set of all
$\mathcal {L}$
-equations. An assignment is a homomorphism (i.e., a function which commutes with all the operations) from the algebra of formulas
${\mathbf{Fm}}_{\mathcal {L}}$
to some
$\mathcal {L}$
-algebra
${\mathbf{A}}$
. An
$\mathcal {L}$
-algebra
${\mathbf{A}}$
satisfies an
$\mathcal {L}$
-equation
$p \approx q$
with an assignment h (and we write
${\mathbf{A}}, h \vDash p \approx q$
) if
$h(p) = h(q)$
in
${\mathbf{A}}$
. An
$\mathcal {L}$
-equation
$p \approx q$
is valid in
${\mathbf{A}}$
(and we write
${\mathbf{A}} \vDash p \approx q$
) if for all assignments h to
${\mathbf{A}}$
,
${\mathbf{A}}, h \vDash p \approx q$
; if
$\Sigma $
is a set of
$\mathcal {L}$
-equations then
${\mathbf{A}} \vDash \Sigma $
if
${\mathbf{A}} \vDash \sigma $
for all
$\sigma \in \Sigma $
. An
$\mathcal {L}$
-equation
$p \approx q$
is valid in a class of
$\mathcal {L}$
-algebras
$\mathsf {K}$
, and we write
$\mathsf {K} \vDash p \approx q$
or
$\vDash _{\mathsf {K}} p \approx q$
, if
${\mathbf{A}} \vDash p \approx q$
for all
${\mathbf{A}} \in \mathsf {K}$
. Moreover, given any set of
$\mathcal {L}$
-equations
$\Sigma \cup \{p \approx q\}$
, and any class of
$\mathcal {L}$
-algebras
$\mathsf {K}$
, we write
$\Sigma \vDash _{\mathsf {K}} p \approx q$
if for every algebra
${\mathbf{A}} \in \mathsf {K}$
, and any assignment h to
${\mathbf{A}}$
, if
$h(p') = h(q')$
for all
$p' \approx q' \in \Sigma $
, then
$h(p) = h(q)$
.
$\Sigma \vDash _{\mathsf {K}} \Delta $
, for
$\Sigma , \Delta $
sets of
$\mathcal {L}$
-equations, is interpreted as
$\Sigma \vDash _{\mathsf {K}} \delta $
for all
$\delta \in \Delta $
.
$\vDash _{\mathsf {K}}$
is called the equational consequence relative to
$\mathsf {K}$
.
Definition 2.6. A Lewis variably strict conditional algebra, or
$\mathsf {V}$
-algebra for short, is an algebra
where
$(C, \wedge , \vee , \neg , 0, 1)$
is a Boolean algebra and
is a binary operation such that, for all
$x, y, z \in C$
:
-
1.

-
2.

-
3.

-
4.

We denote the variety of
$\mathsf {V}$
-algebras with
$\mathsf {VA}$
.
Theorem 2.7 [Reference Rosella and Ugolini24, theorem 4.9].
$\mathsf {VA}$
is the equivalent algebraic semantics of
$\texttt {GV}$
. In particular, for every
,
$\Gamma \vdash _{\texttt {GV}} \varphi $
iff
$\{\gamma \approx 1: \gamma \in \Gamma \} \models _{\mathsf {VA}} \varphi \approx 1$
.
An important consequence of algebraizability is that axiomatic extensions of algebraizable logics are also algebraizable, and if the initial logic has as equivalent algebraic semantics a variety
$\mathsf {V}$
, each axiomatic extension corresponds to a subvariety of
$\mathsf {V}$
. Let us then recall the equivalent algebraic semantics of
$\texttt {GVC}$
and
$\texttt {GVCS}$
. Let
$\mathsf {VCA}$
be the subvariety of
$\mathsf {VA}$
further satisfying:
and
$\mathsf {VCSA}$
the subvariety of
$\mathsf {VCA}$
of Lewis conditional algebras, satisfying:
Corollary 2.8 [Reference Rosella and Ugolini24].
$\mathsf {VCA}$
and
$\mathsf {VCSA}$
are the equivalent algebraic semantics of, respectively,
$\texttt {GVC}$
and
$\texttt {GVCS}$
.
The weaker logics
$\texttt {LV}$
,
$\texttt {LVC}$
, and
$\texttt {LVCS}$
are not algebraizable, however they still can be studied by means of
$\mathsf {V}$
-algebras, as they coincide with the logics preserving degrees of truth of these algebras. Let us be more precise.
Let
$\mathsf {K}$
be a variety of algebras with a lattice reduct; the logic preserving degrees of truth of
$\mathsf {K}$
is defined as follows: for every set
$\Gamma \cup \{\varphi \}$
of formulas in the language of
$\mathsf {K}$
,
for some
$\{\gamma _1, \ldots , \gamma _n\} \subseteq \Gamma $
and
$n \in \mathbb {N}$
. For instance, intuitionistic logic is the logic preserving degrees of truth of Heyting algebras, and the local consequence of the modal logic
$\mathbf {K}$
is the logic preserving degrees of truth of modal algebras. Logics preserving the degrees of truth have been studied in generality in [Reference Font10, Reference Moraschini21, Reference Nowak23]. In [Reference Rosella and Ugolini24] it is shown that all axiomatic extensions of
$\texttt {LV}$
are the logics preserving degrees of truth of the equivalent algebraic semantics of their stronger (or global) counterpart. Specifically, with respect to the systems considered in this work we have the following.
Theorem 2.9 [Reference Rosella and Ugolini24, corollary 4.19].
$\texttt {LV}, \texttt {LVC}$
, and
$\texttt {LVCS}$
are the logics preserving degrees of truth of, respectively,
$\mathsf {VA}$
,
$\mathsf {VCA}$
, and
$\mathsf {VCSA}$
.
2.3 Topological dualities
Stone representation theorem for Boolean algebras is one of the most important results in the field of algebraic logic, connecting algebraic structures to topological spaces. From the logical point of view, this deep result shows a fundamental connection between algebraic and relational models of a logical system. Stone duality has been extended and enriched in many ways over the years, starting from the classical results about Boolean algebras with operators by Jónnson and Tarski [Reference Jónsson and Tarski17]; in this work we proceed along these lines to connect algebraic and relational models of Lewis’s logics.
We here recall the basics of Stone duality for the sake of the reader. The underlying idea of Stone’s result, which is relevant to our investigation as well, is that while finite Boolean algebras are algebras of sets, infinite Boolean algebras are correctly represented by (infinite) sets endowed with a topology, i.e., a collection of special subsets (called open sets) that essentially allow to recover the Boolean algebra. We will phrase some of the results in the language of category theory, the interested reader can check the textbook references [Reference Davey and Priestley7, Reference Mac Lane20] for the unexplained notions.
To each Boolean algebra
${\mathbf{B}}$
, Stone duality associates a topological space
$\mathscr{S}({\mathbf{B}})$
, which is a Stone space (compact and totally disconnected, see [Reference Davey and Priestley7, definition 11.5]). Precisely:
where
$Ul({\mathbf{B}})$
is the set of ultrafilters of
${\mathbf{B}}$
, and
$\tau $
is the Stone topology generated by the basis of clopen sets (i.e., open sets whose complement is also open)
where
In general, the Boolean algebra
${\mathbf{B}}$
can be recovered from
$\mathscr{S}({\mathbf{B}})$
, since
$\mathfrak {s}$
establishes an isomorphism between
${\mathbf{B}}$
and the Boolean algebra of clopen sets
$\{Cl(Ul({\mathbf{B}}), \tau ), \cap , \cup , ^C, \emptyset , Ul({\mathbf{B}})\}$
. Note that if
${\mathbf{B}}$
is finite, its ultrafilters are in one–one correspondence with its atoms, and so
$\mathscr{S}({\mathbf{B}})$
can be seen as the set of atoms of
${\mathbf{B}}$
with the discrete topology (where every subset is a clopen set), and
${\mathbf{B}}$
is isomorphic to the powerset Boolean algebra constructed from its Stone space.
With respect to the maps, let
${\mathbf{B}}, {\mathbf{C}}$
be a pair of Boolean algebras, then given any homomorphism
$h: {\mathbf{B}} \to {\mathbf{C}}$
, Stone duality associates to h a continuous map (i.e., a map such that the pre-image of an open set is open) from
$\mathscr{S}({\mathbf{C}})$
to
$\mathscr{S}({\mathbf{B}})$
defined as
$\mathscr{S}$
is a functor from the category of Boolean algebras with homomorphisms to the category of Stone spaces with continuous maps. Conversely, the so-called adjoint functor to
$\mathscr{S}$
, which we denote by
$\mathscr{S}^{\ -1}$
, associates to each a Stone space
$(S, \tau )$
the Boolean algebra of its clopen subsets
we will write
$A \Rightarrow B$
for
$A^C \cup B$
, and
$A \Leftrightarrow B$
for
$(A \Rightarrow B) \cap (B \Rightarrow A)$
. In order to recover
$(S, \tau )$
, the following map
yields a homeomorphism (i.e., a bijective continuous map with continuous inverse) mapping S to its homeomorphic copy given by the ultrafilters space of the Boolean algebra of its clopen sets. Moreover, given a continuous map
$\varphi $
from a Stone space
$(S, \tau )$
to another Stone space
$(S', \tau ')$
, its dual via Stone duality
$\mathscr{S}^{\ -1}(\varphi )$
is defined as
and it is a homomorphism from
$\mathscr{S}^{\ -1}(S', \tau ')$
to
$\mathscr{S}^{\ -1}(S, \tau )$
.
Composing
$\mathscr{S}$
with
$\mathscr{S}^{\ -1}$
results in isomorphic Boolean algebras in one direction, and homeomorphic Stone spaces in the other. The dual equivalence established by these two functors testifies a deep connection between the algebraic and topological structures.
Notation 2.10. Henceforth we make a point of using the letters
$a, b, c, \ldots $
for the elements of the Boolean algebras, and
$X, Y, Z, \ldots $
for their ultrafilters. To allude to the correspondence established by Stone duality, we will use
$x, y, z, \ldots $
for the points of a Stone space, and
$A, B, C, \ldots $
for its clopen subsets.
The underlying idea of the next section is to appropriately extend Stone duality to
$\mathsf {VA}$
. We observe that
as a binary operator does not distribute over joins (nor meets) on both sides, therefore we cannot just apply the well-known Jónsson–Tarski duality for Boolean algebras with operators [Reference Jónsson and Tarski17], which indeed works for operators distributing over joins on both sides and respecting
$0$
(or, dually, respecting meets and
$1$
); however, since
on the right does distribute over meets and maps
$1$
to
$1$
, we can use the Jónnson–Tarski approach as an inspiration. Dualities for Boolean algebras (or more general structures) with arbitrary operators have been studied in the literature [Reference Bezhanishvili, Bezhanishvili and de Groot1, Reference Došen8, Reference Gehrke and Harding11–Reference Gehrke and Priestley14], however, we cannot straightforwardly derive the duality of interest directly from these results. For instance, in [Reference Bezhanishvili, Bezhanishvili and de Groot1] the authors provide significant advances but consider unary modalities. The recent [Reference Celani, Gruszczyński and Menchón5] provides a duality theoretic treatment of Boolean algebras with a conditional operator
$\rightarrowtail $
, that despite being quite general, does not cover our case since it is assumed to satisfy
$((p \lor q) \rightarrowtail r) \to ((p \rightarrowtail r) \land (q \rightarrowtail r)) = 1$
, which does not hold in our structures (compare with axiom (3) in Definition 2.6).
Moreover, our main results only partly concern the fact that
$\mathsf {V}$
-algebras have a duality-theoretic counterpart; our focus is to develop the mathematical theory of Lewis’s models, and explicitly showing the relationship among the different relational and algebraic models is a key part of our analysis.
Our treatment is also inspired by dualities for bounded distributive lattices with a ternary relation (see, e.g., [Reference Celani, Gruszczyński and Menchón5, Reference Celani6]); we here prefer to use a binary function instead of a ternary relation, in order to highlight the connection with Lewis’s
$\alpha $
-model that indeed use a selection function.
3 Topological
$\alpha $
-models
We first show the duality of
$\mathsf {VA}$
with respect to a topological version of Lewis’s
$\alpha $
-models. Given a Stone space
$(S, \tau )$
, let
$Cl(S)$
denote the set of clopens; we will make use of a binary function
$f : Cl(S) \times S \to \mathcal {P}(S)$
to interpret the dualization of
.
Definition 3.1. Let us call topological
$\alpha $
-model a triple
$(S, \tau , f),$
where
$(S, \tau )$
is a Stone space, and
$f : Cl(S) \times S \to \mathcal {P}(S)$
is a binary function such that, for all clopen sets
$A,B$
of S and
$x,y \in S$
, the following hold:
-
(α1)
$f(A, x) \subseteq A$
; -
(α2)
$f(A, x) \subseteq B$
and
$f(B, x) \subseteq A$
imply
$f(A, x) = f(B, x)$
; -
(α3) either
$f(A \cup B, x) \subseteq A$
, or
$f(A \cup B, x) \subseteq B$
, or
$f(A \cup B, x) = f(A, x) \cup f(B, x)$
; -
(α4)
is clopen; -
(α5)
$f(A, x)$
is closed.
Remark 3.2. Topological
$\alpha $
-models are clearly inspired by Lewis’s
$\alpha $
-models (see Definition 2.3); here we consider a version based on topological spaces, and add the last two conditions. (
$\alpha $
4) is meant to make sure that the counterfactual
is evaluated in an element of the Boolean algebra associated with the Stone space, while (
$\alpha $
5) is a more technical condition, needed in the duality-theoretic machinery. The reader may observe that (
$\alpha $
5) corresponds to the frame relation R being point-closed in the duality between modal algebras and descriptive set frames (see [Reference Blackburn, de Rijke and Venema2, sec. 5.5, pp. 314–315] for a discussion). In fact, (
$\alpha $
5) yields what in descriptive frames is the property of tightness, which here reads as:
Indeed, if
$f(A,x)$
is closed, it can be written as the intersection of all the clopen sets that contain it:
$f(A,x) = \bigcap \{C \in Cl(S): f(A,x) \subseteq C\}$
which easily entails tightness. This fact will be used in Proposition 3.7 and is instrumental in the duality result (Theorem 3.11).
Let us now start from a topological
$\alpha $
-model and define a
$\mathsf {V}$
-algebra, using the postulates used by Lewis [Reference Lewis18] to define the interpretation of the operations; of course here the underlying Boolean algebra is the Boolean algebra of clopen sets given by Stone duality. Given a triple
$\mathfrak {S}= (S, \tau , f)$
, we write
$Cl(\mathfrak {S})$
for the set of clopens. Moreover, given any two clopen
$A,B$
:
Proposition 3.3. Let
$\mathfrak {S} = (S, \tau , f)$
be a topological
$\alpha $
-model;
is in
$\mathsf {VA}$
.
Proof. First observe that
$(Cl (\mathfrak {S}), \cap , \cup , ^C, \emptyset , S)$
is a Boolean algebra via Stone duality. We now prove that the defining equations of
$\mathsf {VA}$
involving
hold, i.e., (1)–(4) in Definition 2.6. For (1), let
$A \in C(\mathfrak {S})$
, then
by (
$\alpha $
1).
We now prove (2), i.e.,
for all
$A,B,C \in Cl (\mathfrak {S})$
. Since
if
then
$f(A,x) \subseteq B$
and
$f(B,x) \subseteq A$
, thus by (
$\alpha $
2) we get that
$f(A,x) = f(B,x)$
. Now,
coincides with
$$ \begin{align*}&(\{x \in S: f(A,x) \subseteq C\} \cap \{x \in S: f(B,x) \subseteq C\})\\ &\quad \cup (\{x \in S: f(A,x) \subseteq C\}^C \cap \{x \in S: f(B,x) \subseteq C\}^C).\end{align*} $$
Since
$f(A,x) = f(B,x)$
, the claim follows from the fact that either
$x \in \{y \in S: f(A,y) \subseteq C\}$
or
$x \in \{y \in S: f(A,y) \subseteq C\}^C$
.
For (3), we verify that for all
$A,B,C \in C (\mathfrak {S})$

It suffices to verify the right-to-left inclusion. Let
$x \in S$
; by (
$\alpha $
3) either
$f(A \cup B,x) \subseteq A$
,
$f(A \cup B,x) \subseteq B$
, or
$f(A \cup B,x) = f(A,x) \cup f(B,x)$
. In the first case we get that
, while in the second one it follows that
. Finally, suppose
$f(A \cup B,x) = f(A,x) \cup f(B,x)$
. We show that then
, or equivalently, that
Indeed, by definition, the latter is equivalent to:
which clearly holds given the hypothesis
$f(A \cup B, x) = f(A,x) \cup f(B,x)$
.
Lastly, it is straightforward from the definition of
that
for all
$A,B,C \in C (\mathfrak {S})$
, which proves (4). We have shown that
$\mathscr{C} (\mathfrak {S}) \in \mathsf {VA}$
.
Vice versa, starting from an algebra
${\mathbf{C}} \in \mathsf {VA}$
, we obtain a topological
$\alpha $
-model on the Stone dual of the Boolean reduct of
${\mathbf{C}}$
,
$\mathscr{S}({\mathbf{C}})$
, with the function defined by the following stipulation:
where we recall that
$\mathfrak {s}(a) = \{Z \in Ul({\mathbf{C}}): a \in Z\}$
, for all
$a \in C, X,Y \in Ul({\mathbf{C}})$
. The reader shall also remember that via Stone duality all clopens of
$\mathscr{S}({\mathbf{C}})$
are of the kind
$\mathfrak {s}(a)$
for some
$a \in {\mathbf{C}}$
.
The following lemma is inspired by [Reference Celani6, lemma 2.5], and it will be often used in the proofs.
Lemma 3.4. Let
${\mathbf{C}} \in \mathsf {VA}$
,
$a, b \in C$
, then the following are equivalent:
-
1.
; -
2.
$Y \in f_C(\mathfrak {s}(a), X)$
implies
$b \in Y$
; -
3.
$f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)$
.
Proof. (1) implies (2) follows directly from the definition of
$f_C$
. We prove (2) implies (1) by contraposition. Suppose
, and let
; then F is a Boolean filter of
${\mathbf{C}}$
. Indeed, it is an upset because of the fact that
is order-preserving on the right and X is upwards closed, it is closed under meet since
distributes over meets on the right and X is closed under meets, and
$1 \in F$
given that
. Moreover, since
$b \notin F$
given that
, by the Boolean Ultrafilter Theorem we obtain that there is an ultrafilter Y of the Boolean reduct of
${\mathbf{C}}$
such that
,
$b \notin Y$
; therefore
$Y \in f_C(\mathfrak {s}(a), X)$
but
$b \notin Y$
.
Finally, the fact that (2) and (3) are equivalent follows from the fact that
$\mathfrak {s}(b) = \{X \in Ul({\mathbf{C}}): b \in X\}$
. This concludes the proof.
Proposition 3.5. Let
${\mathbf{C}} \in \mathsf {VA}$
; then
$\mathscr{A}({\mathbf{C}}) = (\mathscr{S}({\mathbf{C}}), f_C)$
is a topological
$\alpha $
-model.
Proof. Since
$\mathscr{S}({\mathbf{C}})$
is a Stone space by Stone duality, we only have to prove that
$f_C$
satisfies the properties (
$\alpha $
1)–(
$\alpha $
5) in Definition 3.1. Note once again that by Stone duality, every clopen in
$\mathscr{S}({\mathbf{C}})$
is of the kind
$\mathfrak {s}(a)$
for some
$a \in C$
.
For (
$\alpha $
1), let
$a \in C, X \in Ul({\mathbf{C}})$
, we have to verify that
$f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(a)$
. Suppose
$Y \in f_C(\mathfrak {s}(a), X)$
, which by definition is equivalent to
implies
$c \in Y$
, thus since by the first axiom of
$\mathsf {VA}$
:
, we get that
$a \in Y$
and then
$Y \in \mathfrak {s}(a) = \{Z \in Ul({\mathbf{C}}): a \in Z\}$
.
To prove (
$\alpha $
2), we show that for
$a, b \in C, X \in Ul({\mathbf{C}})$
,
$f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)$
and
$f_C(\mathfrak {s}(b), X) \subseteq \mathfrak {s}(a)$
imply
$f_C(\mathfrak {s}(a), X) = f_C(\mathfrak {s}(b), X)$
. Suppose then
$f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)$
and
$f_C(\mathfrak {s}(b), X) \subseteq \mathfrak {s}(A)$
, which by Lemma 3.4 is equivalent to assuming that
and
. Then also
, and by the property (2) of
$\mathsf {V}$
-algebras, we get that:

We prove
$f_C(\mathfrak {s}(a), X) \subseteq f_C(\mathfrak {s}(b), X)$
, the reader shall observe that the other inclusion is proved symmetrically. Assume
$Y \in f_C(\mathfrak {s}(a), X)$
, i.e.,
implies
$c \in Y$
, we show that then
$Y \in f_C(\mathfrak {s}(b), X)$
. Indeed, suppose
, then since also
and X is a Boolean filter, we get that
; therefore
$c \in Y$
, which shows
$Y \in f_C(\mathfrak {s}(b), X)$
.
For (
$\alpha $
3), we verify that, given
$a,b \in C, X \in Ul({\mathbf{C}})$
, either
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) \subseteq \mathfrak {s}(a)$
, or
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) \subseteq \mathfrak {s}(b)$
, or
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) = f_C(\mathfrak {s}(a), X) \cup f_C(\mathfrak {s}(b), X)$
. By property (3) of
$\mathsf {V}$
-algebras, for all
$a,b,c \in C$
:
Since X is an ultrafilter, either
, or
. Using again Lemma 3.4, the first two cases correspond to, respectively,
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) \subseteq \mathfrak {s}(a)$
and
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) \subseteq \mathfrak {s}(b)$
. Consider then the remaining case where
. Then by the definition of
$\leftrightarrow $
and the fact that
$\to $
distributes over
$\land $
we get that also the following elements are in X:

We show that
$f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X) = f_C(\mathfrak {s}(a), X) \cup f_C(\mathfrak {s}(b), X)$
. First, consider
$Y \in f_C(\mathfrak {s}(a), X) \cup f_C(\mathfrak {s}(b), X)$
; without loss of generality, we assume
$Y \in f_C(\mathfrak {s}(a), X)$
, i.e.,
implies
$c \in Y$
. We prove
$Y \in f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X)$
, i.e.,
implies
$c \in Y$
. Indeed, if
, given that also
, we get
and then
$c \in Y$
.
For the other inclusion, suppose that
$Y \notin f_C(\mathfrak {s}(a), X) \cup f_C(\mathfrak {s}(b), X)$
, we prove that
$Y \notin f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X)$
. By assumption, there is
$c \in C$
such that
but
$c \notin Y$
; in more detail, there must be
$c', c" \in C$
such that
but
$c',c" \notin Y$
, then one can take
$c = c' \lor c"$
. Now,
, and also
; thus
is in X too. But
$c \notin Y$
, thus
$Y \notin f_C(\mathfrak {s}(a) \cup \mathfrak {s}(b), X)$
.
(
$\alpha $
4) requires that for all
$a, b \in C$
,
is clopen. Using the definition of
and Lemma 3.4 we get that
$\mathfrak {s}$
is also a homomorphism with respect to
:

which implies that
is clopen.
Lastly, to verify (
$\alpha $
5) we show that each
$f_C(\mathfrak {s}(a), X)$
is closed. We start by proving that
While the left-to-right inclusion is clear, we verify the converse. Let
$Y \in \bigcap \{\mathfrak {s}(b): f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)\}$
, then Y is such that
$f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)$
implies
$Y \in \mathfrak {s}(b)$
for any
$b \in C$
; once again using Lemma 3.4, the latter translates to saying that
implies
$b \in Y$
for any
$b \in C$
, which is exactly the definition of
$Y \in f_C(\mathfrak {s}(a),X)$
. Thus,
$f_C(\mathfrak {s}(a), X) = \bigcap \{\mathfrak {s}(b): f_C(\mathfrak {s}(a), X) \subseteq \mathfrak {s}(b)\}$
, which means that
$f_C(\mathfrak {s}(a), X)$
is an intersection of clopen sets, and therefore it is closed.here
Proceeding towards establishing a duality, we show that the two maps we have introduced,
$\mathscr{C}$
and
$\mathscr{A}$
, are inverses of each other, in the following sense.
Proposition 3.6. Let
${\mathbf{C}} \in \mathsf {VA}$
, then
${\mathbf{C}} \cong (\mathscr{C} \circ \mathscr{A})({\mathbf{C}})$
.
Proof. The isomorphism is given by the Stone map
$\mathfrak {s}$
which also preserves the operation
, indeed as shown in the proof of Proposition 3.5
.
For the converse composition, recall from the preliminaries that
$\beta $
is the Stone homeomorphism mapping a Stone space to its homeomorphic copy given by the ultrafilters space of the Boolean algebra of its clopen sets; we get that:
Proposition 3.7. Let
$\mathfrak {S} = (S, \tau , f)$
be a topological
$\alpha $
-model, and consider
$(\mathscr{A}\circ \mathscr{C})(\mathfrak {S}) = ((\mathscr{S} \circ \mathscr{S}^{\ -1})(S, \tau ), f_{\mathscr{C}(\mathfrak {S})})$
. Then for all
$A \in Cl(S), x,y \in S$
,
$y \in f(A,x) \;\mbox { iff }\; \beta (y) \in f_{\mathscr{C}(\mathfrak {S})}(\beta [A], \beta (x)).$
Proof. Following the definitions, and given that
$\beta [A] = \mathfrak {s}(A)$
, we get:

where the last step is the tightness property discussed in Remark 3.2, which follows from the fact that
$f(A,x) = \bigcap \{C \in Cl(S): f(A,x) \subseteq C\}$
. Indeed, by axiom
$(\alpha $
5),
$f(A,x)$
is closed and hence can be written as the intersection of all the clopen sets that contain it.
We now move to studying the topological counterpart of the algebraic homomorphisms. We remind the reader that in Stone duality such counterpart is given by continuous maps between Stone spaces; in our setting, the continuous maps will have to satisfy some properties involving the function f.
Definition 3.8. Let
$\mathfrak {S} = (S, \tau , f), \mathfrak {S}' = (S', \tau ', f')$
be topological
$\alpha $
-models. An
$\alpha $
-morphism is a map
$\varphi : S \to S'$
such that:
-
1.
$\varphi $
is continuous; -
2.
$y \in f(\varphi ^{-1}[A'], x)$
implies
$\varphi (y) \in f'(A', \varphi (x))$
; -
3.
$y' \in f'(A', \varphi (x))$
implies that there exists
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in f(\varphi ^{-1}[A'], x)$
.
Given a homomorphism of
$\mathsf {V}$
-algebras
$h: {\mathbf{C}} \to {\mathbf{C}}'$
, we will show that its Stone dual
$\mathscr{S}(h)$
, defined as
$\mathscr{S}(h)(U) = h^{-1}[U]$
, is an
$\alpha $
-morphism on the corresponding topological
$\alpha $
-models. Moreover, we will show that given an
$\alpha $
-morphism
$\varphi $
from
$(S, \tau , f)$
to
$(S', \tau ', f')$
, its dual via Stone duality
$\mathscr{S}^{\ -1}(\varphi )$
, defined as
$\mathscr{S}^{\ -1}(\varphi )(A) = \varphi ^{-1}[A]$
, is a homomorphism on the corresponding
$\mathsf {V}$
-algebras.
Proposition 3.9. Given a homomorphism of
$\mathsf {V}$
-algebras
$h: {\mathbf{C}}' \to {\mathbf{C}}$
,
$\mathscr{S}(h)$
is an
$\alpha $
-morphism from
$\mathscr{A}({\mathbf{C}})$
to
$\mathscr{A}({\mathbf{C}}')$
.
Proof.
$\mathscr{S}(h)$
is a continuous map from
$\mathscr{A}({\mathbf{C}})$
to
$\mathscr{A}({\mathbf{C}}')$
by Stone duality. Let us now show that for all
$a' \in C', X,Y \in Ul({\mathbf{C}})$
,
Suppose then that
$Y \in f_C(\mathscr{S}(h)^{-1}[\mathfrak {s}(a')], X)$
. Direct calculations show that
$\mathscr{S}(h)^{-1}[\mathfrak {s}(a')] = \mathfrak {s}(h(a))$
; thus
$Y \in f_C(\mathscr{S}(h)^{-1}[\mathfrak {s}(a')], X)$
if and only if
$Y \in f_C(\mathfrak {s}(h(a')), X)$
if and only if, by definition of
$f_C$
,
implies
$c \in Y$
for any
$c \in C$
. Therefore, consider any
$c' \in C'$
,
$h(c') \in C$
, and we get that
implies
$h(c') \in Y$
; equivalently, for any
$c' \in C'$
,
implies
$c' \in h^{-1}(Y) = \mathscr{S}(h)(Y)$
. We have showed
$\mathscr{S}(h)(Y) \in f_{C'}(\mathfrak {s}(a'), \mathscr{S}(h)(X))$
.
It is left to show that for all
$a' \in C', X \in Ul({\mathbf{C}}), Y' \in Ul({\mathbf{C}}')$
,
$Y'\in f_{C'}(\mathfrak {s}(a'), \mathscr{S}(h)(X))$
implies that there exists
$Y \in Ul(C)$
such that
$\mathscr{S}(h)(Y) = Y'$
and
$Y \in f_C(\mathscr{S}(h)^{-1}[\mathfrak {s}(a')], X)$
. Assume
$Y' \in f_{C'}(\mathfrak {s}(a'), \mathscr{S}(h)(X))$
, i.e., for all
$c' \in C'$
,
implies
$c' \in Y'$
. Equivalently, for all
$c' \in C'$
,
We want to find
$Y \in Ul(C)$
such that
$\mathscr{S}(h)(Y) = Y'$
and
$Y \in f_C(\mathscr{S}(h)^{-1}[\mathfrak {s}(a')], X)$
, equivalently
$Y \in f_C(\mathfrak {s}(h(a')), X)$
, which means that for all
$c \in C$
,
implies
$c \in Y$
. Consider the sets
F is easily seen to be a filter of the Boolean reduct of
${\mathbf{C}}$
, I is a Boolean ideal, and we can show that
$I \cap F = \emptyset $
; indeed, if
$d \in I \cap F$
, we get
$d \leq h(c')$
for some
$c' \notin Y'$
, and
. But then also
, which by (8) above implies
$c' \in Y'$
, a contradiction. Thus we can apply the Boolean Ultrafilter Theorem and obtain an ultrafilter
$Y \in Ul(C)$
such that
$F \subseteq Y$
and
$Y \cap I = \emptyset $
. By construction, for all
$c \in C$
,
implies
$c \in Y$
, since
. We now prove that
$\mathscr{S}(h)(Y) = Y'$
, which will conclude the proof. Consider
$c' \notin Y'$
, then
$h(c') \in I$
, and
$h(c') \notin Y$
, and thus
${c' \notin h^{-1}[Y] = \mathscr{S}(h)(Y)}$
. This implies that
$\mathscr{S}(h)(Y) \subseteq Y'$
. Since
$\mathscr{S}(h)(Y)$
is an ultrafilter and therefore maximal, it follows that
$\mathscr{S}(h)(Y) = Y'$
, completing the proof.
Proposition 3.10. Given an
$\alpha $
-morphism
$\varphi $
from the topological
$\alpha $
-models
$\mathfrak {S}' = (S', \tau ', f')$
to
$\mathfrak {S} = (S, \tau , f)$
,
$\mathscr{S}^{\ -1}(\varphi )$
is a homomorphism from
$\mathscr{C}(\mathfrak {S})$
to
$\mathscr{C}(\mathfrak {S}')$
.
Proof.
$\mathscr{S}^{\ -1}(\varphi )$
is a homomorphism on the Boolean reducts by Stone duality, thus we only need to show that for all
$A,B \in Cl(S)$
,
.
Now,
and
Let
, i.e.,
$f(A, \varphi (x')) \subseteq B$
; we show that
, i.e.,
$f'(\varphi ^{-1}[A], x') \subseteq \varphi ^{-1}[B]$
. Indeed, if
$y' \in f'(\varphi ^{-1}[A], x')$
, by Definition 3.8(2) we get
$\varphi (y') \in f(A, \varphi (x'))$
and then
$\varphi (y') \in B$
, that is,
$y' \in \varphi ^{-1}[B]$
.
Finally, let
, i.e.,
$f'(\varphi ^{-1}[A], x') \subseteq \varphi ^{-1}[B]$
, we verify that
, i.e.,
$f(A, \varphi (x')) \subseteq B$
. If
$y \in f(A, \varphi (x'))$
, by Definition 3.8(3) we get that there exists
$y' \in S'$
such that
$\varphi (y') = y$
and
$y' \in f'(\varphi ^{-1}[A], x')$
; then
$y' \in \varphi ^{-1}[B]$
, that is,
$\varphi (y') = y \in B$
. This shows the second inclusion and concludes the proof.
Using the previous results we proceed to show that Stone duality extends to a duality between the algebraic category
$\mathsf {VA}$
(with objects the algebras in
$\mathsf {VA}$
and morphism the homomorphisms) and a category whose objects are topological
$\alpha $
-models and whose morphisms are
$\alpha $
-morphisms. Let us denote the latter category by
$\mathsf {AM}$
.
We define a map
$\mathscr{A}: \mathsf {VA} \to \mathsf {AM}$
in the following way, for any
${\mathbf{C}}, {\mathbf{C}}' \in \mathsf {VA}, h: {\mathbf{C}} \to {\mathbf{C}}'$
:
$$ \begin{align*} \mathscr{A}({\mathbf{C}}) &= (\mathscr{S}(\mathrm{Bool}({\mathbf{C}})), f_C); \\ \mathscr{A}(h) &= \mathscr{S}(h); \end{align*} $$
where
$\mathrm {Bool}({\mathbf{C}})$
is the Boolean reduct of
${\mathbf{C}}$
. Moreover, we define a map
$\mathscr{C}: \mathsf {AM} \to \mathsf {VA}$
such that, for any
$\mathfrak {S} = (S, \tau , f), \mathfrak {S}' = (S', \tau ', f') \in \mathsf {AM}, \varphi : \mathfrak {S} \to \mathfrak {S}'$
:

Both
$\mathscr{A}$
and
$\mathscr{C}$
can be easily shown to be well-defined functors since they properly extend the Stone duality functors
$\mathscr{S}$
and
$\mathscr{S}^{\ -1}$
: on objects via Propositions 3.3 and 3.5, on morphisms via Propositions 3.9 and 3.10. We now see that they establish a duality between
$\mathsf {VA}$
and
$\mathsf {AM}$
.
Theorem 3.11.
$\mathscr{A}$
and
$\mathscr{C}$
are adjoint functors establishing a dual equivalence between the algebraic category of
$\mathsf {V}$
-algebras and the category of topological
$\alpha $
-models
$\mathsf {AM}$
.
Proof. The dual equivalence between
$\mathsf {V}$
-algebras and
$\mathsf {AM}$
, or equivalently the categorical equivalence between
$\mathsf {V}$
-algebras and the opposite category
$\mathsf {AM}^{\mathrm {{op}}}$
(obtained by reversing the morphisms), is established as an extension of Stone duality. Note that via Propositions 3.9 and 3.10, we get that both
$\mathscr{C}$
and
$\mathscr{A}$
are full and faithful (with respect to the opposite category). It is then left to prove that
$\mathscr{A}$
and
$\mathscr{C}$
are adjoint functors. By Proposition 3.6, the composition
$\mathscr{C} \circ \mathscr{A}$
is naturally isomorphic to the identity functor on
$\mathsf {VA}$
, via the Stone map
$\mathfrak {s}$
. Similarly, by Proposition 3.7 the composition
$\mathscr{A} \circ \mathscr{C}$
is naturally isomorphic to the identity on
$\mathsf {AM}$
, via the Stone homeomorphism
$\beta $
. This completes the proof.
With the following lemmas we can extend the duality to the two most relevant subvarieties of
$\mathsf {VA}$
. We follow Lewis’s ideas in [Reference Lewis18].
Definition 3.12. Let us call a topological
$\alpha $
-model
$\mathfrak {S} = (S, \tau , f)$
as:
-
1. topological
$\alpha _1$
-model if it is such that for each
$A \in Cl(S)$
and
$x \in A$
,
$f(A,x) = \{x\}$
; -
2. topological
$\alpha _2$
-model if it is a topological
$\alpha _1$
-model such that for each
$A \in Cl(S)$
and
$x \in S$
,
$f(A,x)$
contains at most one element.
Lemma 3.13. Let
$\mathfrak {S} = (S, \tau , f)$
be a topological
$\alpha $
-model. Then:
-
1. if
$\mathfrak {S}$
is a topological
$\alpha _1$
-model,
$\mathscr{C}(\mathfrak {S}) \in \mathsf {VCA}$
; -
2. if
$\mathfrak {S}$
is a topological
$\alpha _2$
-model,
$\mathscr{C}(\mathfrak {S}) \in \mathsf {VCSA}$
.
Proof. For the first point, let us start by recalling that
$\mathscr{C}(\mathfrak {S}) \in \mathsf {VCA}$
if and only if
Moreover, by definition,
. Let now
$x \in A \cap B$
; then in particular
$x \in A$
and then if
$\mathfrak {S}$
is a topological
$\alpha _1$
-model,
$f(A, x) =\{x\} \subseteq B$
, which shows the first inclusion. For the second one, let
; if
$x \notin A^C$
, i.e.,
$x \in A$
, once again
$f(A, x) =\{x\}$
, and then
$x \in B$
given that
.
For the second point, we recall that
$\mathscr{C}(\mathfrak {S}) \in \mathsf {VCSA}$
if and only if
, where
Consider any
$x \in S$
; if
$\mathfrak {S}$
is a topological
$\alpha _2$
-model,
$f(A,x)$
contains at most one element. Thus either
$f(A,x) = \emptyset $
, and then x is in both
and
, or
$f(A,x)= \{z\}$
; thus since either
$z \in B$
or
$z \in B^C$
, either
or
.
Conversely:
Lemma 3.14. Let
${\mathbf{C}} \in \mathsf {VA}$
. Then:
-
1. if
${\mathbf{C}} \in \mathsf {VCA}$
then
$\mathscr{A}({\mathbf{C}})$
is a topological
$\alpha _1$
-model; -
2. if
${\mathbf{C}} \in \mathsf {VCSA}$
then
$\mathscr{A}({\mathbf{C}})$
is a topological
$\alpha _2$
-model.
Proof. Suppose
${\mathbf{C}} \in \mathsf {VCA}$
, we show that for each
$a \in C$
and
$X \in Ul({\mathbf{C}})$
such that
$a \in X$
,
$f(\mathfrak {s}(a),X) = \{X\}$
. First, we show
$X \in f(\mathfrak {s}(a),X)$
; let
, then
. Now,
$a, a \to b \in X$
implies
$b \in X$
, which proves
$X \in f(\mathfrak {s}(a),X)$
. We now verify that if
$Y \in f(\mathfrak {s}(a),X)$
holds, then
$Y=X$
. Indeed, let
$b \in X$
; since also
$a \in X$
,
$a \land b \in X$
, and
, thus from
$Y \in f(\mathfrak {s}(a),X)$
we get
$b \in Y$
, which shows
$X \subseteq Y$
. Since X is an ultrafilter and hence maximal, it follows that
$X=Y$
, completing the proof of (1).
For the second point, suppose
${\mathbf{C}} \in \mathsf {VCSA}$
. We prove that
$f_C(\mathfrak {s}(a), X)$
has at most one element for all
$a \in C, X \in Ul({\mathbf{C}})$
. Suppose that both
$Y \in f_C(\mathfrak {s}(a),X)$
and
$Z \in f_C(\mathfrak {s}(a),X)$
hold; this means that
implies
$b \in Y \cap Z$
. Let
$c \in Y$
; since
${\mathbf{C}} \in \mathsf {VCSA}$
,
and hence
or
, since X is a Boolean ultrafilter. If
, then
$c \in Y \cap Z \subseteq Z$
;
instead yields a contradiction, since it entails that
$\neg c \in Y\cap Z \subseteq Y$
. Thus
$c \in Z$
and
$Y \subseteq Z$
. By the maximality of the ultrafilter Y, we have
$Y=Z$
, which concludes the proof.
Hence, the functors
$\mathscr{A}$
and
$\mathscr{C}$
restrict to the full subcategories of topological
$\alpha _1$
-models (and
$\alpha _2$
-models) and
$\mathsf {VCA}$
(
$\mathsf {VCSA}$
), yielding a duality of the subcategories.
Theorem 3.15. The algebraic category of
$\mathsf {VCA}$
-algebras (resp.,
$\mathsf {VCSA}$
-algebras) is dually equivalent to the category of topological
$\alpha _1$
-models (resp.,
$\alpha _2$
-models) with
$\alpha $
-morphisms.
4 Topological spheres
In this section we prove that topological
$\alpha $
-models are categorically equivalent to a category of topological spheres. More precisely, we describe the category whose objects are equivalent in the sense of Lewis [Reference Lewis18] to
$\alpha $
-models. That is, they are defined on the same set, and they define the same algebra: the Boolean algebras have the same domain, and the operations
obtained, respectively, by the spheres and the function f coincide.
Lewis himself notices that in order to obtain equivalent models, one needs to assume the limit assumption: given any point x and its set of spheres, for each proposition
$\varphi $
that intersects the spheres at x, there is a smallest sphere that does so. We will indeed use this assumption to define the category of topological spheres.
Definition 4.1. A topological sphere is a triple
$(S, \tau , \sigma ),$
where
$(S, \tau )$
is a Stone space,
$\sigma : S \to \mathcal {P}(\mathcal {P}(S))$
, and letting
$$ \begin{align*}\Sigma(A,x) := \begin{cases} \bigcap\{U \in \sigma(x) : U \cap A \neq \emptyset\} & \text{if } A \cap \bigcup\sigma(x) \neq \emptyset,\\ \emptyset & \text{otherwise}, \end{cases} \end{align*} $$
the following are satisfied:
-
(S1) for all
$x \in S$
,
$\sigma (x)$
is nested, i.e., for all
$U,V \in \sigma (x)$
,
$U \subseteq V$
or
$V \subseteq U$
; -
(S2) for all
$A,B \in Cl(S)$
,
is clopen; -
(S3) for all
$x \in S$
and
$A \in Cl(S)$
such that
$A \cap \bigcup \sigma (x) \neq \emptyset $
,
$\Sigma (A,x) \in \sigma (x)$
and
$\Sigma (A,x) \cap A \neq \emptyset $
; -
(S4) for all
$x \in S$
,
$A \in Cl(S)$
,
$A \cap \Sigma (A,x)$
is closed.
Remark 4.2. It is clear that the definition of topological spheres is inspired by Lewis’s sphere models. In the next section we will see how to obtain sphere models as in Definition 2.1 from a topological sphere. Moreover, importantly, we will discuss condition (S3), which is Lewis’s limit assumption on sphere models. Indeed, (S3) amounts to asking for the existence of a smallest sphere which intersects a certain proposition A, whenever A intersects the spheres associated with a world x; if such a sphere exists, it coincides with the intersection of all the spheres that intersect A.
We will now define the maps, inspired by Lewis’s work (see both [Reference Lewis18, Reference Lewis19]), that obtain a topological sphere from a topological
$\alpha $
-model and vice versa. Consider a topological sphere
$(S, \tau , \sigma )$
; we associate the triple
$\mathcal {F}((S, \tau , \sigma )) = (S, \tau , f_{\sigma })$
where for all
$A \in Cl(S)$
,
$x \in S$
:
The following is a straightforward consequence of the definitions.
Lemma 4.3. For all
$A \in Cl(S)$
,
$x \in S$
, either
$\Sigma (A,x) = \emptyset $
or
$\Sigma (A,x)$
is the smallest
$U \in \sigma (x), U \cap A \neq \emptyset $
. Moreover:
-
1.
$f_\sigma (A, x) = \emptyset $
iff there is no
$U \in \sigma (x)$
such that
$U \cap A \neq \emptyset $
; -
2.
$f_\sigma (A, x) \neq \emptyset $
iff
$A \cap \bigcup \sigma (x) \neq \emptyset $
.
Proposition 4.4. For any topological sphere
$(S, \tau , \sigma )$
,
$\mathcal {F}((S, \tau , \sigma )) = (S, \tau , f_\sigma )$
is a topological
$\alpha $
-model.
Proof. We have to show that
$f_\sigma $
satisfies the conditions (
$\alpha $
1)–(
$\alpha $
5) of Definition 3.1. (
$\alpha $
1) is easily seen, since
$f_\sigma (A, x) \subseteq A$
for all
$A \in Cl(S)$
by definition.
For (
$\alpha $
2), we proceed to prove that
$f_\sigma (A, x) \subseteq B$
and
$f_\sigma (B, x) \subseteq A$
implies
$f_\sigma (A, x) = f_\sigma (B, x)$
. Let us then assume that
$f_\sigma (A, x) \subseteq B$
and
$f_\sigma (B, x) \subseteq A$
. It follows by the definition and Lemma 4.3 that either both sets are empty (and there is nothing else to prove), or they are both nonempty. Suppose we are in the latter case; thus
$f_\sigma (A, x) = A \cap \Sigma (A, x) \subseteq B$
and
$f_\sigma (B, x) = B \cap \Sigma (B, x) \subseteq A$
. Since
we get that
$\Sigma (B, x) \cap A \neq \emptyset $
; but
$\Sigma (A, x)$
is the least element in
$\sigma (x)$
that intersects A, thus
$\Sigma (A, x) \subseteq \Sigma (B, x)$
. Analogously,
$\Sigma (B, x) \subseteq \Sigma (A, x)$
and therefore
$\Sigma (A, x) = \Sigma (B, x)$
. Thus
$$ \begin{align*}f_\sigma(A, x) &= A \cap \Sigma(A, x) = A \cap \Sigma(A, x) \cap B = A \cap \Sigma(B, x) \cap B\\ & = \Sigma(B, x) \cap B = f_\sigma(B,x).\end{align*} $$
Let us move to (
$\alpha $
3). Given
$A, B \in Cl(S), x \in S$
we show that either
$f_\sigma (A \cup B, x) \subseteq A$
, or
$f_\sigma (A \cup B, x) \subseteq B$
, or
$f_\sigma (A \cup B, x) = f_\sigma (A, x) \cup f_\sigma (B, x)$
. In other words:
$$ \begin{align*}(A \cup B) \cap \Sigma(A \cup B,x) &\subseteq A \mbox{ or } (A \cup B) \cap \Sigma(A \cup B,x)\\& \subseteq B \mbox{ or } (A \cup B) \cap \Sigma(A \cup B,x) = (A \cap \Sigma(A,x)) \cup (B \cap \Sigma(B,x)).\end{align*} $$
Since
$\sigma (x)$
is nested, we can assume without loss of generality that
$\Sigma (A,x) \subseteq \Sigma (B,x)$
. We consider the following cases:
-
1.
$\emptyset \neq \Sigma (A,x) = \Sigma (B,x)$
; then necessarily
$\Sigma (A \cup B,x) = \mathrm {min} \{U \in \sigma (x): U \cap (A \cup B) \neq 0\} = \Sigma (A,x) = \Sigma (B,x)$
. Thus:
$(A \cup B) \cap \Sigma (A \cup B,x) = (A \cap \Sigma (A \cup B,x)) \cup (B \cap \Sigma (A \cup B,x)) = (A \cap \Sigma (A,x)) \cup (B \cap \Sigma (B,x)).$
-
2.
$\emptyset \neq \Sigma (A,x) \subsetneq \Sigma (B,x)$
; it follows that
$\Sigma (A,x) \cap B = \emptyset $
and then
$\emptyset \neq \Sigma (A \cup B,x) = \Sigma (A,x)$
. Hence
$(A \cup B) \cap \Sigma (A \cup B,x) = (A \cap \Sigma (A \cup B,x)) \cup (B \cap \Sigma (A \cup B,x)) = A \cap \Sigma (A, x) \subseteq A.$
-
3.
$\emptyset = \Sigma (A,x) = \Sigma (B,x)$
; in this case also
$\Sigma (A \cup B,x) = \emptyset $
, and then trivially the last condition holds, i.e.,
$(A \cup B) \cap \Sigma (A \cup B,x) = (A \cap \Sigma (A,x)) \cup (B \cap \Sigma (B,x))$
. -
4.
$\emptyset = \Sigma (A,x) \subsetneq \Sigma (B,x)$
; then
$\Sigma (A \cup B,x) = \Sigma (B,x)$
and
$A \cap \Sigma (A \cup B,x) = \emptyset $
, hence
$$ \begin{align*}(A \cup B) \cap \Sigma(A \cup B,x) &= (A \cap \Sigma(A \cup B,x)) \cup (B \cap \Sigma(A \cup B,x))\\& = B \cap \Sigma(B, x) \subseteq B.\end{align*} $$
This proves (
$\alpha $
3). For (
$\alpha $
4), the fact that
is clopen follows from the fact that
is clopen and
and
coincide, indeed:
Lastly,
$f_\sigma (A, x) = A \cap \Sigma (A, x)$
is closed by (S5) in the definition, and (
$\alpha $
5) holds.
For the converse translation, consider a topological
$\alpha $
-model
$(S, \tau , f)$
; we associate the triple
$\mathcal {L}((S, \tau , f)) = (S, \tau , \sigma _f)$
such that for all
$x \in S$
:
$$ \begin{align} \sigma_f(x) = \left\{\bigcup_{A <_x B} f(A,x) : B \in Cl(S)\right\}, \end{align} $$
where
Our definition extends to topological spheres the stipulations in [Reference Lewis18], where Lewis constructs an
$\alpha $
-model from a sphere satisfying the limit assumption. The following properties (1) and (2) are stated without proof in [Reference Lewis18] (and elsewhere in the literature); we include a proof here for the sake of the interested reader.
Lemma 4.5. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
; for all
$A, B \in Cl(S)$
,
$x \in S$
:
-
1. if
$A \subseteq B$
and
$A \cap f(B,x) \neq \emptyset $
, then
$f(A,x) = A \cap f(B,x)$
; -
2. if
$A \subseteq B$
and
$f(B,x) = \emptyset $
, then
$f(A,x) =\emptyset $
; -
3. if
$A \subseteq B$
and
$\emptyset \neq f(B,x) \subseteq A$
, then
$f(A,x) = f(B,x)$
.
Proof. For the first property, suppose
$A \subseteq B$
and
$A \cap f(B,x) \neq \emptyset $
. Notice that
In particular, since
$A \subseteq B$
,
$B = A \cup (B \cap A^C)$
and then
$f(B,x) = f(A \cup (B \cap A^C), x)$
. Thus, by (
$\alpha 3$
) we get the following cases:
-
(i)
$f(B,x) \subseteq A$
, and therefore, since by (
$\alpha 1$
)
$f(A,x) \subseteq A \subseteq B$
, by (
$\alpha 2$
)
$f(B,x) = f(A,x)$
, which implies that
$f(A,x) = A \cap f(B,x)$
; -
(ii)
$f(B,x) \subseteq B \cap A^C \subseteq A^C$
, which yields a contradiction since
$A \cap f(B,x) \neq \emptyset $
, and thus it is never the case; -
(iii)
$f(B,x) = f(A,x) \cup f(B \cap A^C, x)$
, which (together with (
$\alpha 1$
)) directly implies that
$f(A,x) \subseteq A \cap f(B,x)$
. For the other inclusion, let
$y \in A \cap f(B,x)$
; then
$y \in f(A,x) \cup f(B \cap A^C, x)$
. Since
$f(B \cap A^C, x) \subseteq B \cap A^C \subseteq A^C$
, and
$y \in A$
, it follows that
$y \notin f(B \cap A^C, x)$
. Therefore,
$y \in f(A,x)$
, and we have shown that
$f(A,x) = A \cap f(B,x)$
.
Hence, in all the possible cases
$f(A,x) = A \cap f(B,x)$
.
The second property is easier to show. Indeed, suppose
$A \subseteq B$
and
$f(B,x) = \emptyset $
. Then
$f(B,x) = \emptyset \subseteq A$
and, using (
$\alpha 1$
),
$f(A,x) \subseteq A \subseteq B$
; by (
$\alpha 2$
) it follows that
$f(A,x) = f(B,x) =\emptyset $
.
For (3), note that if
$A \subseteq B$
and
$\emptyset \neq f(B,x) \subseteq A$
, then in particular
$A \cap f(B,x) \neq \emptyset $
, and then by (1)
$f(A,x) = A \cap f(B,x)$
. This implies that
$f(A,x) \subseteq f(B,x) \subseteq B$
(by (
$\alpha $
1)). Hence by (
$\alpha $
2) we get
$f(A,x) = f(B,x)$
.
Using the last lemma, we can show the following facts about the relation
$<_x$
. The following also appears in [Reference Lewis18] with some hints to a proof. We offer here a complete and alternative proof which seems to be missing from the literature.
Lemma 4.6. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
; for all
$A, B \in Cl(S)$
,
$x \in S$
:
-
1.
$<_x$
is a total preorder on
$Cl(S)$
; -
2.
$A <_x B$
implies
$B \cap f(A,x) \subseteq f(B,x)$
; -
3.
$A \cap f(B,x) \neq \emptyset $
implies
$A <_x B$
.
Proof. (1) We start by verifying that
$<_x$
is a total preorder on
$Cl(S)$
for each
$x \in S$
. Reflexivity is straightforward from the definition, so let us prove transitivity. We assume that
$A <_x B$
and
$B <_x C$
, i.e.,
and we show that
$A <_x C$
, that is,
$f(C,x) = \emptyset \;\mbox { or }\;\; \emptyset \neq f(A,x) \subseteq f(A \cup C, x)$
. Suppose then that
$f(C,x) \neq \emptyset $
; thus since
$B <_x C$
,
$\emptyset \neq f(B,x) \subseteq f(B \cup C,x)$
, and since
$A <_x B$
also
$\emptyset \neq f(A,x) \subseteq f(A \cup B,x)$
. We need to show that
$f(A,x) \subseteq f(A \cup C, x)$
. First, we observe that
$(A \cup B) \cap f(A \cup (B \cup C), x) \neq \emptyset $
. Indeed, from
$(\alpha 3)$
we get that either
$f(A \cup (B \cup C), x) \subseteq A$
,
$f(A \cup (B \cup C), x) \subseteq B \cup C$
, or
$f(A \cup (B \cup C), x) =f(A,x) \cup f(B \cup C, x)$
. In the first case it is clear that
$f(A \cup (B \cup C), x)$
intersects with
$A \cup B$
. If
$f(A \cup (B \cup C), x) \subseteq B \cup C$
, by Lemma 4.5(3) we get that
$f(A \cup (B \cup C), x) = f(B \cup C, x) \supseteq f(B,x)$
(the last inclusion by hypothesis), and then again it intersects
$A \cup B$
. In the last case,
$f(A \cup (B \cup C), x) \supseteq f(B \cup C, x)$
which again contains
$f(B,x)$
, and so in any case
$f(A \cup (B \cup C), x)$
intersects
$A \cup B$
. Therefore (using Lemma 4.5(1) for the last inclusion):
which implies that
$A \cap f(A \cup B \cup C, x) \neq \emptyset $
. Then also
$(A \cup C) \cap f(A \cup B \cup C, x) \neq \emptyset $
, and again by Lemma 4.5,
$$ \begin{align*}f(A \cup C, x) &= (A \cup C) \cap f(A \cup B \cup C, x)\\& = (A \cap f(A \cup B \cup C, x)) \cup (C \cap f(A \cup B \cup C, x)).\end{align*} $$
Thus, since
$A \cap f(A \cup B \cup C, x) \neq \emptyset $
, we get that
$A \cap f(A \cup C, x)\neq \emptyset $
, and by a last application of Lemma 4.5 we obtain that
$f(A,x) \subseteq f(A \cup C, x)$
which shows that
$A <_x C$
; hence
$<_x$
is transitive for all
$x \in S$
.
For the first point, it is left to show that
$<_x$
is total. Consider
$f(A \cup B,x)$
; if
$f(A \cup B,x) = \emptyset $
, by Lemma 4.5 also
$f(A,x) = f(B,x) = \emptyset $
, and thus
$A <_x B$
and
$B <_x A$
. If
$f(A \cup B,x) \neq \emptyset $
, then by
$(\alpha 3)$
either
$f(A \cup B,x) \subseteq A$
, which by Lemma 4.5(3) yields
$f(A,x) = f(A \cup B,x)$
and then
$A <_x B$
, or
$f(A \cup B,x) \subseteq B$
, which analogously yields
$B <_x A$
, or
$f(A \cup B,x) = f(A,x) \cup f(B,x)$
, and therefore
$A <_x B$
and
$B <_x A$
. In any case, A and B are comparable with respect to
$<_x$
, which is then a total preorder.
(2) Suppose that
$A <_x B$
, i.e.,
Assume first that
$f(B,x) = \emptyset $
, and let us consider two cases: whether
$f(A \cup B,x)$
is empty or not. If
$f(A \cup B,x) = \emptyset $
, by Lemma 4.5 also
$f(A,x) = \emptyset $
and then
$B \cap f(A,x) =\emptyset = f(B,x)$
. If
$f(A \cup B,x) \neq \emptyset $
, from the fact that
$f(B,x) = \emptyset $
, we get that
$f(A \cup B,x) = f(A,x)$
using
$(\alpha 3)$
and Lemma 4.5(3), and, necessarily,
$B \cap f(A \cup B,x) = \emptyset $
by Lemma 4.5(1). Thus
$B \cap f(A,x) = \emptyset = f(B,x)$
.
Assume now that
$\emptyset \neq f(A,x) \subseteq f(A \cup B,x)$
. Then either
$B \cap f(A,x) = \emptyset \subseteq f(B,x)$
or
and then by Lemma 4.5(1)
$f(B,x) = B \cap f(A \cup B,x)$
thus
This completes the proof of (2).
(3) Finally, assume
$A \cap f(B,x) \neq \emptyset $
, which implies that in particular
$f(B,x) \neq \emptyset $
. Consider
$f(A \cup B,x)$
, then by
$(\alpha 3)$
we have three cases:
-
(i)
$f(A \cup B,x) \subseteq A$
; then by Lemma 4.5(3),
$f(A \cup B,x) = f(A,x)$
which is necessarily nonempty, since otherwise by Lemma 4.5(2) also
$f(B,x)$
would be empty, which is a contradiction. Thus we have that
$A <_x B$
. -
(ii)
$f(A \cup B,x) \subseteq B$
; then again by Lemma 4.5(3),
$f(A \cup B,x) = f(B,x)$
and then
$A \cap f(A \cup B,x) = A \cap f(B,x) \neq \emptyset $
. By Lemma 4.5(1) this yields that
$\emptyset \neq f(A, x) \subseteq f(A \cup B, x)$
and then
$A <_x B$
. -
(iii)
$f(A \cup B,x) = f(A,x) \cup f(B,x)$
; thus
$f(A, x) \subseteq f(A \cup B, x)$
and
$f(A, x)$
is nonempty (otherwise we would be in the case above), and once again
$A <_x B$
.
We have shown that in every case
$A <_x B$
, which concludes this proof.
We are ready to show that we can construct a topological sphere from a topological
$\alpha $
-model.
Proposition 4.7. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
;
$\mathcal {L}((S, \tau , f)) = (S, \tau , \sigma _f)$
is a topological sphere.
Proof. We show that
$\sigma _f$
satisfies the properties (S1)–(S4) of Definition 4.1. First, notice that
$\sigma _f$
, defined as
$$ \begin{align*}\sigma_f(x) = \left\{\bigcup_{A <_x B} f(A,x) : B \in Cl(S)\right\} \end{align*} $$
is indeed a map from S to
$\mathcal {P}(\mathcal {P}(S))$
. For (S1), note that for all
$x \in S$
,
$\sigma _f(x)$
is nested, since
$<_x$
is total by Lemma 4.6.
Let now
$\Sigma _f(A,x) = \emptyset $
if
$A \cap \bigcup \sigma _f(x) = \emptyset $
, and
$\Sigma _f(A,x) := \bigcap \{U \in \sigma _f(x): U \cap A \neq \emptyset \}$
otherwise. We prove (S3), i.e., that for all
$x \in S$
and
$A \in Cl(S)$
such that
$A \cap \bigcup \sigma _f(x) \neq \emptyset $
,
$\Sigma _f(A,x) \in \sigma _f(x)$
and
$\Sigma _f(A,x) \cap A \neq \emptyset $
. Suppose then that
$A \cap \bigcup \sigma _f(x) \neq \emptyset $
, hence we will show that
Call
$U_A = \bigcup _{C <_x A} f(C,x)$
. Then
$U_A \in \sigma _f(x)$
by definition. We verify that
$U_A \cap A \neq \emptyset $
. Observe that:
Indeed, the right-to-left direction is obvious since
$f(A, x) \in U_A$
and
$f(A,x) \subseteq A$
. We show the left-to-right one by contraposition. Note that if there is
$U \in \sigma _f(x)$
such that
$A \cap U \neq \emptyset $
, then there is a B such that
$A \cap f(B,x) \neq \emptyset $
and so
$A <_x B$
by Lemma 4.6; thus
$f(A, x) \neq \emptyset $
by the definition of
$<_x$
. Hence, since by hypothesis
$A \cap \bigcup \sigma _f(x) \neq \emptyset $
,
$\emptyset \neq f(A, x) \subseteq A$
(the last inclusion by
$(\alpha 1)$
), and since
$A <_x A$
(
$<_x$
is reflexive by Lemma 4.6), then
$A \cap U_A \neq \emptyset $
.
We now observe that
$U_A$
is the smallest
$U \in \sigma _f(x), U \cap A \neq \emptyset $
; indeed, suppose
$A \cap U_B = A \cap \bigcup _{C <_x B} f(C,x) \neq \emptyset $
. Then
$A \cap f(C,x) \neq \emptyset $
for some
$C <_x B$
, thus by Lemma 4.6
$A <_x C$
; by the transitivity of
$<_x$
, it holds that
$A <_x B$
, thus
$U_A \subseteq U_B$
. It follows that
$U_A = \bigcap \{U \in \sigma _f(x): U \cap A \neq \emptyset \} = \Sigma _f(A, x) \in \sigma _f(x)$
, and also
$\Sigma _f(A, x) \cap A \neq \emptyset $
.
We now proceed to prove (S4), that is, for all
$x \in S$
,
$A \in Cl(S)$
,
$A \cap \Sigma _f(A,x)$
is closed. By what we have shown in the previous point, either
$\Sigma _f(A,x) = \emptyset $
or
$\Sigma _f(A,x) = U_A = \bigcup _{C <_x A} f(C,x)$
. In the first case, the empty set is a closed set in any topological space. In the second case, note that by Lemma 4.6 for all
$C <_x A$
,
$A \cap f(C,x) \subseteq f(A,x) = A \cap f(A,x)$
. Thus, since
$A <_x A$
:
and the latter set is closed, since
$f(A,x)$
is closed by definition of a topological
$\alpha $
-model, A is closed, and closed sets are closed under finite intersection.
Finally, for (S2), the fact that
is clopen for all
$A,B \in Cl(S)$
follows from the fact that
, given that the latter is clopen by definition of a topological
$\alpha $
-model. Let us verify the equality. It is helpful to first recall the definitions:
By (12) above,
$f(A, x) = \emptyset $
if and only if there is no
$U \in \sigma _f(x), U \cap A \neq \emptyset $
. If x is such, then
$\Sigma _f(A,x) = \emptyset $
and therefore
iff
for any
$B \in Cl(S)$
.
Consider now
$x \in S$
such that
$f(A, x) \neq \emptyset $
, and then
iff
$A \cap U_A \subseteq B$
. Hence, also
since
$f(A,x) \subseteq A \cap U_A$
. Conversely, suppose that
, i.e.,
$f(A,x) \subseteq B$
. Then,
since if
$C <_x A$
then
$A \cap f(C,x) \subseteq f(A, x)$
by Lemma 4.6, and so
. This completes the proof.
We now show that if we start from a topological
$\alpha $
-model
$(S, \tau , f)$
, and we apply
$\mathcal {L}$
and then
$\mathcal {F}$
, we obtain exactly the same
$\alpha $
-model.
Proposition 4.8. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
; then
$\mathcal {F}(\mathcal {L}((S, \tau , f))) = (S, \tau , f)$
, i.e.,
$f_{\sigma _{f}} = f$
.
Proof. We verify that given any
$A \in Cl(S), x,y \in S$
,
$y \in f(A,x)$
if and only if
$y \in f_{\sigma _{f}}(A,x)$
. Recall that
$$ \begin{align*}f_{\sigma_{f}}(A,x)& = \emptyset \mbox{ if } A \cap \bigcup\sigma_f(x) = \emptyset,\\f_{\sigma_{f}}(A,x)& = A \cap \displaystyle\bigcap\{U: U \in \sigma_f(X), U \cap A \neq \emptyset\} \mbox{ otherwise}.\end{align*} $$
First, note that
$f(A,x) = \emptyset $
iff there is no
$U \in \sigma _f(x)$
such that
$A \cap U \neq \emptyset $
, as shown in the proof for (S3) of Proposition 4.7. Thus
$f(A,x) = \emptyset $
iff
$f_{\sigma _f} (A,x) = \emptyset $
.
Suppose then that
$f(A, x) \neq \emptyset $
, and let
$y \in f(A,x)$
. Since, again by the proof of (S3) in Proposition 4.7, we have that the smallest sphere intersecting A is
$U_A = \bigcup _{C <_x A}f(C,x)$
, by definition
$f_{\sigma _{f}}(A,x) = A \cap U_A$
. Given that
$y \in f(A, x) \subseteq A$
, and since
$A <_x A$
, we obtain that
$y \in f_{\sigma _{f}}(A,x)$
.
Vice versa, suppose
$y \in f_{\sigma _{f}}(A,x)$
. Since
$f_{\sigma _{f}}(A,x) = A \cap \bigcup _{C <_x A}f(C,x)$
, and if
$C <_x A$
then
$A \cap f(C,x) \subseteq f(A, x)$
by Lemma 4.6,
$y \in f_{\sigma _{f}}(A,x)$
implies
$y \in f(A, x)$
, which concludes the proof.
We now move to describe the morphisms among topological spheres.
Definition 4.9. Consider two topological spheres
$(S, \tau , \sigma ), (S', \tau ', \sigma ')$
; then a map
$\varphi : (S, \tau , \sigma ) \to (S', \tau ', \sigma ')$
is a sphere morphism if:
-
1.
$\varphi $
is continuous; -
2. for all
$y \in S, A' \in Cl(S')$
such that
$\varphi ^{-1}[A'] \cap \bigcup \sigma (x) \neq \emptyset $
,
$y \in \varphi ^{-1}(A') \cap \Sigma (\varphi ^{-1}[A'], x)$
implies
$\varphi (y) \in \Sigma (A', \varphi (x))$
; -
3. for all
$y' \in S', A' \in Cl(S')$
such that
$A' \cap \bigcup \sigma '(\varphi (x)) \neq \emptyset $
,
$y' \in A' \cap \Sigma (A', \varphi (x))$
implies that there is
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in \Sigma (\varphi ^{-1}[A'], x)$
.
Proposition 4.10. Consider two topological spheres
$(S, \tau , \sigma ), (S', \tau ', \sigma ')$
, and let
$\varphi : (S, \tau , \sigma ) \to (S', \tau ', \sigma ')$
be a sphere morphism; then
$\varphi $
is an
$\alpha $
-morphism from
$\mathcal {F}((S, \tau , \sigma ))$
to
$\mathcal {F}((S', \tau ', \sigma '))$
.
Proof. We only need to show the two following properties:
-
1.
$y \in f_\sigma (\varphi ^{-1}[A'], x)$
implies
$\varphi (y) \in f_{\sigma '}(A', \varphi (x))$
; -
2.
$y' \in f_{\sigma '}(A', \varphi (x))$
implies that there exists
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in f_\sigma (\varphi ^{-1}[A'], x)$
.
For (1), assume that
$y \in f_\sigma (\varphi ^{-1}[A'], x)$
. Since
we get that
$\varphi ^{-1}[A'] \cap \bigcup \sigma (x) \neq \emptyset $
, and
$y \in \varphi ^{-1}(A') \cap \Sigma (\varphi ^{-1}[A'], x)$
; by definition of a sphere morphism, this implies
$\varphi (y) \in \Sigma (A', \varphi (x))$
, and of course
$\varphi (y) \in A'$
. Thus, since
$f_{\sigma '}(A', \varphi (x)) = A' \cap \Sigma (A', \varphi (x))$
, we get
$\varphi (y) \in f_{\sigma '}(A', \varphi (x))$
.
For (2) we proceed similarly; suppose
$y' \in f_{\sigma '}(A', \varphi (x))$
. Thus
$A' \cap \bigcup \sigma '(\varphi (x)) \neq \emptyset $
, and by definition of
$f_{\sigma '}(A', \varphi (x))$
,
$y' \in A' \cap \Sigma (A', \varphi (x))$
. Hence, by definition of a sphere morphism, there is a
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in \Sigma (\varphi ^{-1}[A'], x)$
; this yields exactly
$y \in f_\sigma (\varphi ^{-1}[A'], x)$
.
Conversely:
Proposition 4.11. Consider two topological
$\alpha $
-models
$(S, \tau , f), (S', \tau ', f')$
, and let
$\varphi : (S, \tau , f) \to (S', \tau ', f')$
be an
$\alpha $
-morphism; then
$\varphi $
is a sphere morphism from
$\mathcal {L}((S, \tau , f))$
to
$\mathcal {L}((S', \tau ', f'))$
.
Proof. We need to show the following:
-
1. for all
$y \in S, A' \in Cl(S')$
such that
$\varphi ^{-1}[A'] \cap \bigcup \sigma _f(x) \neq \emptyset $
,
$y \in \varphi ^{-1}(A') \cap \Sigma _f(\varphi ^{-1}[A'], x)$
implies
$\varphi (y) \in \Sigma _{f'}(A', \varphi (x))$
; -
2. for all
$y' \in S', A' \in Cl(S')$
such that
$A' \cap \bigcup \sigma _{f'}(\varphi (x)) \neq \emptyset $
,
$y' \in A' \cap \Sigma _f(A', \varphi (x))$
implies that there is
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in \Sigma _f(\varphi ^{-1}[A'], x)$
.
Let us show (1). Assume that
$\varphi ^{-1}[A'] \cap \bigcup \sigma _f(x) \neq \emptyset $
,
$y \in \varphi ^{-1}(A') \cap \Sigma _f(\varphi ^{-1}[A'], x)$
. Then we have that:
$$ \begin{align*}\Sigma_f(\varphi^{-1}[A'], x) = \bigcup_{C<_x \varphi^{-1}[A']}f(C,x).\end{align*} $$
Then
$y \in \varphi ^{-1}(A') \cap f(C,x)$
for some
$C<_x \varphi ^{-1}[A']$
; thus
$y \in f(\varphi ^{-1}(A'), x)$
by Lemma 4.6. Moreover by definition of an
$\alpha $
-morphism, we get that
$\varphi (y) \in f'(A', \varphi (x))$
. Thus
(2) can be verified by easy calculations in a similar fashion; we leave its proof to the reader.
Let us consider the category of topological spheres with sphere morphisms, which we denote by
$\mathsf {SPH}$
. Consider the map
$\mathcal {F}: \mathsf {SPH} \to \mathsf {AM}$
defined as:
$$ \begin{align*} \mathcal{F}((S, \tau, \sigma)) &= (S, \tau, f_\sigma); \\ \mathcal{F}(\varphi) &= \varphi. \end{align*} $$
$\mathcal {F}$
is easily seen to be a functor between the two categories given Proposition 4.10.
Theorem 4.12. The functor
$\mathcal {F}$
establishes a categorical equivalence between the categories of topological
$\alpha $
-models with
$\alpha $
-morphisms and topological spheres with sphere morphisms.
Proof. In order to show that
$\mathcal {F}$
establishes a categorical equivalence, it suffices to say that it is full, faithful and essentially surjective (see, for instance, [Reference Mac Lane20]). Fullness and faithfulness (i.e., injectivity and surjectivity on morphisms among corresponding pairs of objects) are obvious since
$\mathcal {F}$
is the identity on morphisms. The fact that
$\mathcal {F}$
is essentially surjective follows from Proposition 4.8.
Finally, we show that this categorical equivalence specializes to the full subcategories of topological
$\alpha _1$
-models and
$\alpha _2$
-models with respect to the following.
Definition 4.13.
-
1. Let us call centered a topological sphere
$(S, \tau , \sigma )$
such that
$\{x\} \in \sigma (x)$
for all
$x \in S$
. -
2. Let us call Stalnakerian a centered sphere
$(S, \tau , \sigma )$
such that if
$A \cap \bigcup \sigma (x) \neq \emptyset $
, then there exists
$U \in \sigma (x)$
and
$y \in S$
such that
$A \cap U = \{y\}$
.
Lemma 4.14. Let
$(S, \tau , \sigma )$
be a topological sphere. Then
-
1. If
$(S, \tau , \sigma )$
is a centered topological sphere,
$\mathcal {F}((S, \tau , \sigma ))$
is a topological
$\alpha _1$
-model. -
2. If
$(S, \tau , \sigma )$
is a Stalnakerian topological sphere,
$\mathcal {F}((S, \tau , \sigma ))$
is a topological
$\alpha _2$
-model.
Proof. (1) Suppose
$(S, \tau , \sigma )$
is centered, i.e.,
$\{x\} \in \sigma (x)$
for all
$x \in S$
. We need to show that if
$x \in A$
,
$f_\sigma (A,x) = \{x\}$
. Now,
$f_\sigma (A,x) = A \cap \Sigma (A, x)$
. Recall that
$\Sigma (A, x)$
is either empty or the intersection of all
$U \in \sigma (x)$
that intersects with A, and
$\Sigma (A, x)$
belongs to
$\sigma (x)$
if nonempty. Thus if
$x \in A$
, necessarily
$\Sigma (A, x) = \{x\}$
, which implies the claim.
(2) Suppose now
$(S, \tau , \sigma )$
is Stalnakerian. We want to show that for each
$A \in Cl(S)$
and
$x \in S$
,
$f_\sigma (A,x)$
contains at most one element. By hypothesis, if
$A \cap \bigcup \sigma (x) \neq \emptyset $
, then there exists
$U \in \sigma (x)$
and
$y \in S$
such that
$A \cap U = \{y\}$
. Since
$f_\sigma (A,x) = A \cap \Sigma (A, x)$
, either
$f_\sigma (A,x) = \emptyset $
, or otherwise
$A \cap \Sigma (A, x) = \{y\}$
.
Conversely:
Lemma 4.15. Let
$(S, \tau , f)$
be a topological
$\alpha $
-model. Then
-
1. If
$(S, \tau , f)$
is an
$\alpha _1$
-model,
$\mathcal {L}((S, \tau , f))$
is a centered topological sphere. -
2. If
$(S, \tau , f)$
is an
$\alpha _2$
-model,
$\mathcal {L}((S, \tau , f))$
is a Stalnakerian topological sphere.
Proof. (1) First, assume that for each
$A \in Cl(S)$
and
$x \in A$
,
$f(A,x) = \{x\}$
; we show that
$\{x\} \in \sigma _f(x)$
for all
$x \in S$
. Recall that
$$ \begin{align*}\sigma_f(x) = \left\{\bigcup_{B <_x A} f(B,x) : A \in Cl(S)\right\}.\end{align*} $$
Consider a clopen A such that
$x \in A$
(since clopen sets are a base of the Stone space, and singletons are closed, there is surely at least one). Then
$f(A,x) = \{x\}$
; moreover, if
$B <_x A$
, by definition of the preorder
$\emptyset \neq f(B,x) \subseteq f(A \cup B,x)$
, and since
$x \in A\subseteq A \cup B$
, by hypothesis
$f(A \cup B,x) = \{x\}$
and thus
$f(B,x) = \{x\}$
. Hence,
(2) Now, assume that for each
$A \in Cl(S)$
and
$x \in S$
,
$f(A,x)$
contains at most one element. We prove that if
$A \cap \bigcup \sigma _f(x) \neq \emptyset $
, then there exists
$U \in \sigma _f(x)$
and
$y \in S$
such that
$A \cap U = \{y\}$
. Assume
$A \cap \bigcup \sigma _f(x) \neq \emptyset $
; then
$A \cap \Sigma _f(A, x)$
is nonempty, and as shown in the proof of Proposition 4.7,
$\Sigma _f(A, x) = \bigcup _{B <_x A}f(B,x)$
, hence,
In particular, using Lemma 4.6(2),
$f(A, x)$
is nonempty, and thus it contains exactly one element, say
$f(A, x) = \{y\}$
. Moreover, if
$B <_x A$
,
$A \cap f(B,x) \subseteq f(A, x) = \{y\} \subseteq A$
. Therefore,
$A \cap \Sigma _f(A, x) = \{y\}$
and
$\Sigma _f(A, x)$
is the desired element of
$\sigma _f(x)$
.
As a consequence of the duality theorem (Theorem 3.11), its restrictions (Theorem 3.15), the categorical equivalence (Theorem 4.12), and the above Lemmas 4.14 and 4.15, we obtain the following connection between
$\mathsf {V}$
-algebras and topological spheres.
Theorem 4.16. The category of topological spheres is dually equivalent to the algebraic category of
$\kern1pt\mathsf {V}$
-algebras, as testified by the composition of the functors
$\mathcal {F}$
and
$\mathscr{C}$
. Moreover, the restrictions of the composed functors to the full subcategories of, respectively, centered topological spheres and Stalnakerian topological spheres, yield a dual equivalence with respect to
$\mathsf {VCA}$
-algebras and
$\mathsf {VCSA}$
-algebras.
5 Topological orders
We will now explore the connection with the ordered models. Specifically, we will prove that topological
$\alpha $
-models are categorically equivalent to a category of topological orders. As a consequence, we obtain also an equivalence with topological spheres and a duality with
$\mathsf {VA}$
.
These topological orders represent a topological counterpart to Lewis’s order models for variably strict conditionals. As the reader might expect, in order to establish this equivalence we require a limit assumption on the orders. This imposes that for any point x and its associated order
$\prec _x$
, for every formula that intersects the domain or range of
$\prec _x$
, there exist
$\prec _x$
-minimal elements within that intersection.
Definition 5.1. A topological order is a triple
$(S, \tau , \prec )$
, where
$(S, \tau )$
is a Stone space and
$\prec $
is a function assigning to each
$x \in S$
a binary relation
$\prec _x \subseteq S_x \times S_x$
for some subset
$S_x \subseteq S$
, such that:
-
(O1)
$\prec _x$
is a total preorder; -
(O2) for all
$x \in S$
and
$A \in Cl(S)$
such that
$A \cap S_x \neq \emptyset $
, there is a nonempty set of
$\prec _x$
-minimal elements in A, i.e.,
$min_{\prec _x}(A)=\{y \in A : \forall z \in A(\text {if } z \prec _x y\text { then }y \prec _x z)\}\neq \emptyset $
; -
(O3) for all
$A,B \in Cl(S)$
,
is clopen; -
(O4) for all
$x \in S$
,
$A \in Cl(S)$
,
$min_{\prec _x}(A)$
is closed.
Once again inspired by Lewis’s work (see both [Reference Lewis18, Reference Lewis19]), we consider the transformations from a topological order to an
$\alpha $
-model and vice versa. Consider a topological order
$(S, \tau , \prec )$
; we associate the triple
$\mathcal {A}((S, \tau , \prec )) = (S, \tau , f_{\prec }),$
where for all
$A \in Cl(S)$
,
$x \in S$
:
The following lemma is a straightforward consequence of the definition.
Lemma 5.2. For all
$A \in Cl(S)$
,
$x \in S$
:
-
1.
$min_{\prec _x}=\emptyset $
iff
$f_\prec (A, x) = \emptyset $
iff
$A \cap S_x=\emptyset $
; -
2.
$min_{\prec _x}\neq \emptyset $
iff
$f_\prec (A, x) \neq \emptyset $
iff
$A \cap S_x \neq \emptyset $
.
Proposition 5.3. Consider a topological order
$(S, \tau , \prec )$
;
$\mathcal {A}((S, \tau , \prec )) = (S, \tau , f_\prec )$
is a topological
$\alpha $
-model.
Proof. We have to show that
$f_\prec $
satisfies the conditions (
$\alpha $
1)–(
$\alpha $
5) of Definition 3.1. For (
$\alpha $
1),
$f_\prec (A, x) \subseteq A$
for all
$A \in Cl(S)$
by definition.
We proceed to prove (
$\alpha $
2), i.e., for arbitrary
$x \in S$
and
$A, B \in Cl(S)$
,
$f_\prec (A, x) \subseteq B$
and
$f_\prec (B, x) \subseteq A$
implies
$f_\prec (A, x) = f_\prec (B, x)$
. Let us then assume that
$f_\prec (A, x) \subseteq B$
and
$f_\prec (B, x) \subseteq A$
. It follows by the definition and Lemma 5.2 that either both sets are empty (and there is nothing else to prove), or they are both nonempty. Suppose we are in the latter case; thus
$f_\prec (A, x) = min_{\prec _x}(A) \subseteq B$
and
$f_\prec (B, x) = min_{\prec _x}(B) \subseteq A$
. Let
$y \in f_\prec (A, x)$
and
$z \in f_\prec (B, x)$
. Since
$f_\prec (A, x) \subseteq B$
we get
$y \in B$
, and then from
$z \in f_\prec (B, x) = min_{\prec _x}(B)$
it follows that
$z \prec _x y$
; similarly we can show that
$y \prec _x z$
. Since y and z were arbitrary elements of
$f_\prec (A, x)$
and
$f_\prec (B, x)$
, we have that
$f_\prec (A, x)=f_\prec (B, x)$
.
For (
$\alpha $
3) we have to prove that given
$A, B \in Cl(S), x \in S$
, either
$f_\prec (A \cup B, x) \subseteq A$
, or
$f_\prec (A \cup B, x) \subseteq B$
, or
$f_\prec (A \cup B, x) = f_\prec (A, x) \cup f_\prec (B, x)$
. In other words:
$$ \begin{align*}&min_{\prec_x}(A \cup B) \subseteq A \;\;\mbox{ or }\;\; min_{\prec_x}(A \cup B) \subseteq B \;\;\mbox{ or }\\ &min_{\prec_x}(A \cup B) = min_{\prec_x}(A) \cup min_{\prec_x}(B).\end{align*} $$
Assume that
$min_{\prec _x}(A \cup B) \nsubseteq A$
and
$min_{\prec _x}(A \cup B) \nsubseteq B$
, and notice that necessarily
$min_{\prec _x}(A \cup B) \subseteq min_{\prec _x}(A) \cup min_{\prec _x}(B)$
. To prove the converse inclusion, let
$w \in min_{\prec _x}(A) \cup min_{\prec _x}(B)$
, and suppose without loss of generality that
$w \in min_{\prec _x}(A)$
. Since
$min_{\prec _x}(A \cup B) \nsubseteq B$
, there is
$z \in min_{\prec _x}(A \cup B)$
such that
$z \in A$
. Since
$w \in min_{\prec _x}(A)$
,
$w \prec _x z \in min_{\prec _x}(A \cup B)$
, and given that
$w \in A \subseteq A \cup B$
, it follows that
$w \in min_{\prec _x}(A \cup B)$
. We can reason similarly if
$w \in min_{\prec _x}(B)$
. Therefore,
$min_{\prec _x}(A \cup B) = min_{\prec _x}(A) \cup min_{\prec _x}(B)$
.
Finally, (
$\alpha $
4) and (
$\alpha $
5) are direct consequences of (O3) and (O4) in the definition of topological order, together with the fact that
$f_\prec (A, x) = min_{\prec _x}(A)$
.
Conversely, let us show the translation from a topological order to an
$\alpha $
-model, using ideas in [Reference Lewis18]. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
, and recall from (11) in Section 4 that one can define, for each
$x \in S$
, the following total order on clopens:
$A <_x B$
iff
$f(B,x) = \emptyset $
or
$\emptyset \neq f(A,x) \subseteq f(A \cup B,x).$
We associate to
$(S, \tau , f)$
the triple
$\mathcal {O}((S, \tau , f)) = (S, \tau , \prec ^f)$
, where
$ \prec ^f$
associates to each
$x \in S$
the following binary relation
$\prec _x^f$
on
$S_x = \bigcup \{f(A, x) \mid A \in Cl(S)\text { and }f(A, x)\neq \emptyset \}$
:
Proposition 5.4. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
;
$\mathcal {O}((S, \tau , f)) = (S, \tau , \prec ^f)$
is a topological order, and
$f(A, x)=min_{\prec _x^f}(A)$
.
Proof. We verify that for arbitrary
$x \in S$
,
$\prec ^f_x$
satisfies the properties (O1)–(O4) of Definition 5.1.
(O1) Notice first that
$\prec ^f_x$
is indeed a binary relation over a subset
$S_x$
of S. Precisely, by definition,
Let us prove that
$\prec _x^f$
is a total preorder, that is, it is reflexive, transitive, and total. If
$S_x=\emptyset $
there is nothing to show, thus assume
$S_x \neq \emptyset $
. First,
$\prec _x^f$
is reflexive. Indeed, if
$y \in S_x$
then
$y \in f(A, x)$
for some
$A \in Cl(S)$
, and since
$A <_x A$
by Lemma 4.6,
$y \prec _x^f y$
. Moreover,
$\prec _x^f$
is transitive; in fact, assume
$y \prec _x^f z$
and
$z \prec _x^f w$
. This means that there are
$A, B, C, D \in Cl(S)$
such that
$y \in f(A, x)$
,
$z \in f(B, x)$
,
$z \in f(C, x)$
,
$w \in f(D, x)$
, such that
$A <_x B$
and
$C <_x D$
. Now,
$f(B, x) \subseteq B$
by (
$\alpha 1$
), thus
$z \in B \cap f(C, x)$
which is therefore nonempty; by Lemma 4.6(3) we get that
$B <_x C$
. By transitivity of
$<_x$
, it follows that
$A <_x D$
and then
$y \prec _x^f w$
.
Finally, the fact that
$\prec _x^f$
is total follows from
$<_x$
being total (Lemma 4.6). In fact, consider any
$y, z \in S_x$
such that
$y \in f(B, x)$
and
$z \in f(C, x)$
. Either
$B <_x C$
, and then
$y \prec _x^f z$
, or
$C <_x B$
, and so
$z \prec _x^f y$
.
(O2) We now show that for all
$x \in S$
,
$A \in Cl(S)$
, if
$A \cap S_x\neq \emptyset $
, there exists a nonempty set of
$\prec _x^f$
-minimal elements in A. In particular, we show
$\emptyset \neq f(A, x)=min_{\prec _x^f}(A)$
. First, notice that
For the nontrivial direction, if
$A \cap S_x\neq \emptyset $
, then, there is a
$y \in S_x$
and a
$B \in Cl(S)$
such that
$y \in f(B, x) \cap A$
. By Lemma 4.6 we get
$A <_x B$
, and since
$f(B, x) \neq \emptyset $
,
$f(A, x) \neq \emptyset $
by the definition of the order.
Now assume
$A \cap S_x\neq \emptyset $
, and hence
$f(A, x)\neq \emptyset $
. We show that
$f(A, x)$
is indeed the set of
$\prec _x^f$
-minimal elements in A. First,
$f(A, x) \subseteq A$
by (
$\alpha $
1). Now, consider any
$z \in A$
. We need to show that if
$z \prec _x^f y$
for
$y \in f(A, x)$
, then
$ y \prec _x^f z$
. Assume then
$z \prec _x^f y$
for some
$y \in f(A, x)$
. This means that there are
$B, C \in Cl(S)$
such that
$z \in f(B, x)$
,
$y \in f(C, x)$
and
$ B <_x C$
. Observe that
$z \in f(B, x)$
and
$z \in A$
, hence
$A \cap f(B,x)\neq \emptyset $
, and then, by Lemma 4.6, we have that
$A <_x B$
. Since
$y \in f(A, x)$
and
$z \in f(B, x)$
,
$y \prec _x^f z$
as desired. This means that
$min_{\prec _x^f}(A) = f(A, x) \neq \emptyset $
.
(O3) The fact that
is a clopen follows from the fact that
and
coincide, and the former is a clopen by (
$\alpha $
4). By what we proved in the previous point, we have that
$A \cap S_x \neq \emptyset $
iff
$f(A, x)\neq \emptyset $
, and
$f(A, x)=min_{\prec _x^f}(A)$
. Therefore,
(O4) Lastly,
$min_{\prec _x^f}(A)$
is closed for all
$A \in Cl(S)$
since
$min^f_{\prec _x}(A) = f(A, x)$
and
$f(A, x)$
is closed by (
$\alpha $
5).
Note that since we proved that
$f(A, x)=min_{\prec _x^f}(A)$
in (O2), the proof is complete.
We now observe that applying
$\mathcal {O}$
first, and then
$\mathcal {A}$
, coincides with the identity map on topological
$\alpha $
-models.
Proposition 5.5. Consider a topological
$\alpha $
-model
$(S, \tau , f)$
; then
$\mathcal {A}(\mathcal {O}(S, \tau , f)) = (S, \tau , f)$
, i.e.,
$f_{\prec ^f} = f$
.
Proof. By Proposition 5.4, we have that for all
$A \in Cl(S)$
and all
$x \in S$
,
$f(A, x)=min_{\prec _x^f}(A)$
; moreover,
$f_{\prec ^f}(A, x)=min_{\prec _x^f}(A)$
by definition. Therefore
$f(A,x) = f_{\prec ^f}(A, x)$
.
Let us now consider morphisms between topological orders.
Definition 5.6. Consider two topological orders
$(S, \tau , \prec ), (S', \tau ', \prec ')$
; then a map
$\varphi : (S, \tau , \prec ) \to (S', \tau ', \prec ')$
is an order morphism if:
-
1.
$\varphi $
is continuous; -
2. for all
$y, x \in S, A' \in Cl(S')$
,
$y \in min_{\prec _x}(\varphi ^{-1}[A'])$
implies
$\varphi (y) \in min_{\prec ^{\prime}_{\varphi (x)}}(A')$
; -
3. for all
$y' \in S', A' \in Cl(S')$
,
$y' \in min_{\prec ^{\prime}_{\varphi (x)}}(A')$
implies that there is
$y \in S$
such that
$\varphi (y) = y'$
and
$y \in min_{\prec _x}(\varphi ^{-1}[A'])$
.
The above definition closely resembles the one for
$\alpha $
-morphisms; we can indeed see that morphisms between objects in the same category are preserved by the operations
$\mathcal {A}$
and
$\mathcal {O}$
.
Proposition 5.7. Consider two topological orders
$(S, \tau , \prec ), (S', \tau ', \prec ')$
, and
$\varphi : (S, \tau , \prec ) \to (S', \tau ', \prec ')$
a order morphism; then
$\varphi $
is an
$\alpha $
-morphism from
$\mathcal {A}((S, \tau , \prec ))$
to
$\mathcal {A}((S', \tau ', \prec '))$
. Conversely, consider two topological
$\alpha $
-models
$(S, \tau , f), (S', \tau ', f')$
, and let
$\psi : (S, \tau , f) \to (S', \tau ', f')$
be an
$\alpha $
-morphism; then
$\psi $
is an order morphism from
$\mathcal {O}((S, \tau , f))$
to
$\mathcal {O}((S', \tau ', f'))$
.
Proof. The proof follows from the fact that the definitions of
$\alpha $
-morphisms and order morphism coincide, modulo the facts that
$f(A, x)=min_{\prec _x^f}(A)$
and
$f_{\prec }(A, x)=min_{\prec _x}(A)$
.
Let us consider the category of topological orders with order morphisms, which we denote by
$\mathsf {ORD}$
. Consider the map
$\mathcal {A}: \mathsf {ORD} \to \mathsf {AM}$
defined as:
$$ \begin{align*} \mathcal{A}((S, \tau, \prec)) &= (S, \tau, f_\prec); \\ \mathcal{A}(\varphi) &= \varphi. \end{align*} $$
$\mathcal {A}$
is easily seen to be a functor between the two categories given Propositions 5.3 and 5.7.
Theorem 5.8. The functor
$\mathcal {A}$
establishes a categorical equivalence between the categories of topological
$\alpha $
-models with
$\alpha $
-morphisms and topological orders with order morphisms.
Proof. The result follows from the fact that
$\mathcal {A}$
is the identity on morphisms, hence full and faithful, and the fact that
$\mathcal {A}$
is essentially surjective by Proposition 5.5.
As a consequence of the above categorical equivalence (Theorem 5.8) and the duality theorem (Theorem 3.11), we can compose the functors and obtain the following connection between
$\mathsf {V}$
-algebras and topological orders.
Theorem 5.9. The category of topological orders is dually equivalent to the algebraic category of
$\mathsf {V}$
-algebras, testified by the composition of the functors
$\mathcal {A}$
and
$\mathscr{C}$
.
Finally, we show that the categorical equivalence (and hence the duality result) extends to the full subcategories of topological
$\alpha _1$
-models and
$\alpha _2$
-models with respect to the following notions.
Definition 5.10.
-
1. Let us call centered a topological order
$(S, \tau , \prec )$
such that each
$x \in S$
is the minimum with respect to
$\prec _x$
, i.e.,
$x \prec _x y $
for all
$y \in S_x$
and if
$y \prec _x x$
, then
$y = x$
. -
2. Let us call Stalnakerian a centered topological order
$(S, \tau , \prec )$
such that if
$A \cap S_x \neq \emptyset $
, then there is only one
$\prec _x$
-minimal element in A, i.e.,
$|min_{\prec _x}(A)|=1$
.
Lemma 5.11. Let
$(S, \tau , \sigma )$
be a topological order. Then:
-
1. If
$(S, \tau , \prec )$
is a centered topological order,
$\mathcal {A}((S, \tau , \sigma ))$
is a topological
$\alpha _1$
-model. -
2. If
$(S, \tau , \prec )$
is a Stalnakerian topological order,
$\mathcal {A}((S, \tau , \sigma ))$
is a topological
$\alpha _2$
-model.
Proof. (1) Suppose
$(S, \tau , \prec )$
is centered, we need to show that if
$x \in A$
,
${f_\prec (A,x) = \{x\}}$
. By definition of centered order, x is the
$\prec _x$
-minimum. Hence if
$x \in A$
,
$min_{\prec _x}(A)=\{x\}$
, and since
$f_\prec (A,x) = min_{\prec _x}(A)$
the claim follows.
(2) Suppose now
$(S, \tau , \prec )$
is Stalnakerian. We verify that for each
$A \in Cl(S)$
and
$x \in S$
,
$f_\prec (A,x)$
contains at most one element. Recall that if
$A \cap S_x=\emptyset $
, then
$f_\prec (A, x)=min_{\prec _x}(A)=\emptyset $
; while if
$A \cap S_x\neq \emptyset $
then
$|min_{\prec _x}(A)|=1$
, by hypothesis. Therefore, since
$f_\prec (A,x) = min_{\prec _x}(A)$
, we have that
$|f_\prec (A,x)| = 1$
.
Conversely:
Lemma 5.12. Let
$(S, \tau , f)$
be a topological
$\alpha $
-model. Then:
-
1. If
$(S, \tau , f)$
is an
$\alpha _1$
-model,
$\mathcal {O}((S, \tau , f))$
is a centered topological order. -
2. If
$(S, \tau , f)$
is an
$\alpha _2$
-model,
$\mathcal {O}((S, \tau , f))$
is a Stalnakerian topological order.
Proof. (1) First, assume that for each
$A \in Cl(S)$
and
$x \in A$
,
$f(A,x) = \{x\}$
; we show that x is the
$\prec _x$
-minimum. Since S is clopen, then
$f(S, x)=\{x\}$
. Moreover, notice that
$S <_x A$
for all
$A \in Cl(S)$
; indeed,
$S \cup A = S$
, and then
$\emptyset \neq f(S,x) = f(S \cup A, x) = \{x\}$
implies
$S <_x A$
by Lemma 4.6. This, in turn, implies that
$x \prec _x y$
for all
$y \in S_x$
. Moreover, assume
$y \prec_x x$
for
$y \in S_x$
. This means that there are
$A, B \in Cl(S)$
such that
$y \in f(A, x)$
,
$x \in f(B, x)$
and
$A <_x B$
. Since
$x \in f(B, x) \subseteq B$
, then
$x \in A \cup B$
, hence
$f(A \cup B, x)=\{x\}$
, by hypothesis. Then, by
$A <_x B$
, we have that
$f(A, x) \subseteq f(A \cup B, x) = \{x\}$
; and since
$y \in f(A, x)$
, then
$y =x$
. Thus x is the
$\prec _x$
-minimum.
(2) Assume that for each
$A \in Cl(S)$
and
$x \in S$
,
$f(A,x)$
contains at most one element. By Proposition 5.4, we know that for all
$A \in Cl(S)$
,
$f(A, x)=min_{\prec ^f_x}(A)$
. Thus the claim follows, since if
$A \cap S_x \neq \emptyset $
, then
$f(A, x)\neq \emptyset $
, and so
$|min_{\prec _x^f}(A)|=|f(A, x)| = 1$
.
From the previous lemmas, the categorical equivalence between topological sphere models and topological orders and the two considered subclasses easily follow.
Theorem 5.13. The category of centered (Stalnakerian) topological sphere models is categorically equivalent to the class of centered (Stalnakerian) topological orders. Moreover, the full subcategories of, respectively, centered topological orders and Stalnakerian topological orders, are dually equivalent with respect to
$\mathsf {VCA}$
-algebras and
$\mathsf {VCSA}$
-algebras.
6 The limit assumption
In this last section, we circle back to Lewis’s original sphere models. First, as a consequence of the duality theorems, we obtain strong completeness results for Lewis’s logics with respect to topological spheres, both for the global and local consequence. This also provides an alternative proof to strong completeness with respect to all sphere models. Then we will use this intuition to discuss the limit assumption on sphere models.
Consider a topological sphere model
$\mathcal {M} = (S, \tau , \sigma )$
, and let
be any assignment of the variables in
$Var$
to the Boolean algebra of clopen subsets of S. Then of course
$\mathcal {M}_v=(S, \sigma , v)$
is a sphere model satisfying the limit assumption in the sense of Lewis (Definition 2.1).
Definition 6.1. We define the class
$\mathsf {LTS}$
of Lewis topological spheres to be the class of all the sphere models
$\mathcal {M}_v$
obtained from some topological sphere model
$\mathcal {M}$
and some assignment
.
Now,
$\texttt {GV}$
is strongly complete with respect to
$\mathsf {VA}$
(Theorem 2.7). As a consequence of the duality theorem (Theorem 4.16), every algebra in
$\mathsf {VA}$
arises from a topological sphere model and vice versa, which allows us to obtain the following result.
Theorem 6.2 (Topological global strong completeness).
For any
, the following hold:
Proof. The first equivalence is the algebraizability result (Theorem 2.7). For the second one, soundness is easily checked with standard techniques.
We prove completeness by contraposition; assume that
$\Gamma \not \vdash _{\texttt {GV}} \varphi $
, or equivalently,
$\{\gamma \approx 1 : \gamma \in \Gamma \} \not \models _{\mathsf {VA}} \varphi \approx 1$
. That is to say, there is an algebra
${\mathbf{A}} \in \mathsf {VA}$
, and an assignment h from
to
${\mathbf{A}}$
such that
$h(\gamma ) = 1$
for all
$\gamma \in \Gamma $
, but
$h(\varphi ) \neq 1$
. Consider now
$\mathcal {M} = \mathcal {L}(\mathscr{A}({\mathbf{A}}))$
, and take
$\mathcal {M}_v$
where the evaluation v is induced by h, i.e.,
$v= \mathfrak {s} \circ h$
. Then we get that
$\mathcal {M}_v \Vdash \Gamma $
, since each
$\gamma \in \Gamma $
is mapped to the whole universe, but
$\varphi $
is not so
$\mathcal {M}_v \not \Vdash \varphi $
; this yields
$\Gamma \not \models _{\mathsf {LTS},g} \varphi $
and concludes the proof.
Let us now show the analogous result for the weaker calculus.
Theorem 6.3 (Topological local strong completeness).
For any
, the following hold:
Proof. The first equivalence is Theorem 2.9, stating that
$\vdash _{\texttt {LV}}$
is the logic preserving the degrees of truth of
$\mathsf {VA}$
. For the second, soundness can be easily checked directly. For completeness, suppose that
$\Gamma \not \vdash _{\texttt {LV}} \varphi $
, we prove that
$\Gamma \not \models _{\mathsf {LTS},\ell} \varphi $
. We can equivalently assume that there exists
${\mathbf{A}} \in \mathsf {VA}$
, an assignment h of the variables
$Var$
to
${\mathbf{A}}$
and an element
$a \in A$
such that
Without loss of generality, we can consider
${\mathbf{A}}$
to be an algebra obtained from a topological sphere model. Indeed, every algebra in
$\mathsf {VA}$
arises as an algebra defined on a topological
$\alpha $
-model by Proposition 3.6, and every topological
$\alpha $
-model arises by a topological sphere model on the same Stone space and defining the same algebra by Proposition 4.8. One can then see
${\mathbf{A}}$
as the algebra defined on a Lewis topological sphere
$\mathcal {M} = (S, \sigma , h)$
, where all elements are (clopen) subsets of S. Thus (15) above can be rephrased as
Thus, there exists
$w'$
such that
$w'\Vdash \Gamma $
, but
$w'\not \Vdash \varphi $
; hence,
$\Gamma \not \models _{\mathsf {LTS},\ell } \varphi $
, which completes the proof.
The above results imply that Lewis’s logics are strongly sound and complete with respect to the larger class of all sphere models.
Corollary 6.4 (Global and local strong completeness).
For any
, the following hold:
Proof. Soundness can be checked as above. Completeness follows directly from Theorems 6.2 and 6.3, since if a deduction fails in
$\vdash _{\texttt {GV}}$
(or
$\vdash _{\texttt {LV}}$
), its failure is already testified in sphere models in
$\mathsf {LTS}$
, and therefore also in the larger class of all sphere models.
The above results can be extended to the other logics in Lewis’s hierarchy, in particular to the two main extensions considered here. Let
$\mathsf {K}_{\texttt {C}}$
be the class of centered sphere models, and
$\mathsf {K}_{\texttt {CS}}$
be the class of Stalnakerian sphere models.
Theorem 6.5. Let
$\Sigma \in \{\texttt {C},\texttt {CS}\}$
. For all subsets
,
We conclude with a final discussion about the models. Note that all Lewis topological spheres satisfy the limit assumption: we included it in the axiomatization of topological sphere models. While Lewis argues that one “has no right to assume that there always are smallest antecedent-permitting sphere and, within it, a set of closest-antecedent worlds” [Reference Lewis19, p. 20], his logics actually are perfectly captured by the logical behavior of the sphere models that do satisfy the limit assumption. This is witnessed by the above strong completeness theorems (Theorems 6.2 and 6.3), and even more so, by the duality results. In fact, given an arbitrary sphere model in the sense of Lewis, one evaluates formulas on a
$\mathsf {V}$
-algebra whose elements are subsets of the set of its worlds. This
$\mathsf {V}$
-algebra, by the duality theorem (Theorem 4.16), can be seen as an algebra defined over a topological sphere, which does satisfy the limit assumption. In other words, given any sphere model
$\mathcal {M}$
, there is a sphere model
$\mathcal {M}'$
that satisfies the limit assumption, and has isomorphic underlying
$\mathsf {V}$
-algebra; hence, in particular,
$\mathcal {M}'$
satisfies exactly the same validities and deductions.
This suggests that Lewis sphere models possess an excess of expressiveness relatively to Lewis’s logic. This richness, as evidenced by the limit assumption, arises from the models’ ability to encode information that exceeds the expressive boundaries of the considered propositional logics. In particular, in order to isolate the subclass of sphere models that satisfy the limit assumption, one would likely require a more expressive language, such as an infinitary one. Let us elaborate more on this.
As Fine [Reference Fine9], Starr [Reference Starr, Kocurek, Zalta and Nodelman26], and Lewis himself [Reference Lewis18] have observed, the absence of the limit assumption implies the failure of a very intuitive infinitary rule. More precisely, let us consider now an infinitary language over a denumerable set of propositional variables
$\{p_0, p_1, p_2, \dots \}$
which admits infinitary conjunction and disjunction; for instance
$\bigwedge \limits _{i \in \mathbb {N}} p_i = p_0 \wedge p_1 \wedge \dots $
(where the
$p_i$
’s are propositional variables) would be a formula in this language. Furthermore, let us assume that this language can be interpreted within sphere models in the expected way; namely, given a sphere model
$(W, \mathcal {S}, v),$
where infinitary intersections and union are defined:
and the remaining connectives are interpreted as usual. Then, with
$\Gamma \cup \{\psi \}$
being a set of formulas in this infinitary language, the following meta-rule turns out to not be valid in the class of sphere models without the limit assumption (while it does hold in those that do satisfy it):
The failure is testified by Lewis’s counterexample in [Reference Lewis18], which we replicate here for the sake of the reader.
Example 6.6. Consider the sphere model
$\mathcal {M}^* = (W, \mathcal {S}, v),$
where
-
• W is the set of real numbers;
-
• for all
$x \in W$
,
$\mathcal {S}(x)=\{[x, i] : x \leq i \},$
where
$\leq $
is the natural order over the real numbers and
$[x, i]$
is a usual closed interval in the Euclidean topology; -
• v maps the propositional variables to the following open intervals:
$v(p_0)=(0, \infty )$
, and for each
$p_i$
with
$1 \leq i$
,
$v(p_i)=(-\infty , 2^{-i})$
, so for instance
$v(p_1)=(-\infty , \frac {1}{2})$
.
Now, it is easy to realize that given any set of formulas
$\Gamma $
,
$\Gamma \models \bigwedge \Gamma $
is a valid local logical consequence with respect to all the sphere models. Indeed, for every sphere model
$\mathcal {M}$
, for every world w in
$\mathcal {M}$
, if
$w \Vdash \gamma $
for all
$\gamma \in \Gamma $
, then clearly
$w \Vdash \bigwedge \Gamma $
. However, consider the sphere model
$\mathcal {M}^*$
. It is clearly the case that
$\{p_1, p_2, \dots \} \models \bigwedge \limits _{i \in \mathbb {N}} p_i$
. Moreover, the counterfactual formulas
can be checked to hold at
$w = 0$
. However,
is not true at
$0$
since there are certainly points in the sphere associated with 0 where
$p_0$
is true, but there is no point in the model where both
$p_0$
and
$\bigwedge \limits _{i \in \mathbb {N}-\{0\}} p_i$
are true. The failure of the meta-rule depends on the fact that there is no minimal sphere in
$\mathcal {S}(0)$
where
$p_0$
holds, and this entails the failure of the limit assumption.
Let us conclude the section by stressing once again that if one considers the finitary propositional language, the logic corresponding to all sphere models satisfying the limit assumption is exactly the same as the logic defined over the broader class of all sphere models. In this sense, the models without the limit assumption are redundant, as testified also by the duality that does not “see” them.
In fact, the sphere models satisfying the limit assumption occupy a privileged position. They provide a complete and faithful representation of variably strict conditional logic, as their topological counterparts are dually equivalent to the equivalent algebraic semantics of the logic. As far as our study of the logics and their properties is concerned, there exists no independent justification for considering models without the limit assumption. On the other hand, there exists a compelling mathematical rationale for investigating variably strict conditional logics through the lens of models satisfying the limit assumption: these models offer a comprehensive and faithful representation of the logic, as exemplified by the duality results.
7 Conclusions
Our exploration yields a topological study of Lewis’s logics for counterfactual conditionals, revisiting Lewis’s standard possible-worlds semantics through a novel perspective. In particular, we consider topological versions of models based on spheres, orders, and selection functions, and establish their dual equivalence with the algebraic category of
$\mathsf {V}$
-algebras, which is the equivalent algebraic semantics of Lewis’s weakest logic for counterfactuals. Moreover, we extend the results to two main axiomatic extensions, i.e., Lewis’s logic for counterfactuals
$\texttt {VC}$
and Stalnaker’s logic for conditionals
$\texttt {VCS}$
. These duality results unveiled a deeper mathematical understanding of Lewis variably strict conditional logics through the lens of topological spaces. Moreover, they shed new light on the status of the limit assumption and its relationship with the logic.
It is worth stressing that this paper lays the groundwork for several directions of future research. A natural next step is to extend our duality result to the other logics in Lewis’s family, by characterizing the suitable topological models corresponding to their equivalent algebraic semantics, analyzed in [Reference Rosella and Ugolini24]. We also note that this work is only the start of a mathematical analysis of Lewis’s logics. We plan to continue this line of research, using as guide the rich mathematical theory developed for modal logics; we will consider key notions and results (e.g., bisimulations and Sahlqvist theorems) and seek for analogues in the conditional setting.
A further research direction is to consider more general versions of the topological dualities presented here, to advance our understanding of different conditional logics. On one side, one could consider weaker logics than Lewis’s, e.g., the extensions of Burgess’s preferential conditional logic [Reference Burgess4], and study their corresponding algebraic semantics and dual topological spaces in a unifying framework. From another perspective, one could start from the models analyzed here, and consider different topological spaces, other than Stone spaces. This kind of insight may be useful to formalize conditionals based on different nonclassical logics that have an equivalent algebraic semantics with a “good” duality theory (e.g., intermediate logics).
Finally, with respect to the analysis of the limit assumption, a promising direction for future work is to consider conditional logics with an infinitary language, with a theory expressive enough to capture the failure of this condition.
Acknowledgement
We thank the anonymous reviewers for their constructive comments, which helped improve this manuscript.
Funding
G.R. acknowledges financial support from the Italian Ministry of University and Research (MUR) through the PRIN 2022 grant no. 2022ARRY9N (Reasoning with hypotheses: Integrating logical, probabilistic, and experimental perspectives), funded by the European Union (Next Generation EU), and from PLEXUS (Grant Agreement no. 101086295), a Marie Sklodowska-Curie action funded by the EU under the Horizon Europe Research and Innovation Programme.
G.R. also acknowledges partial funding support from the OP JAK project “Knowledge in the Age of Distrust (TRUST)” (No. CZ.02.01.01/00/23_025/0008711).
S.U. acknowledges support from the Ramón y Cajal programme (RyC2021-032670-I), the Spanish project SHORE (PID2022-141529NB-C22), the MOSAIC project (H2020-MSCA-RISE-2020 Project 101007627), and the project PIE CSIC 20235AT019.


