Preprint of June 2016. Forthcoming in Advances in Modal Logic, Vol. 11.
Locales, Nuclei, and Dragalin Frames
Guram Bezhanishvili 1
Department of Mathematical Sciences
New Mexico State University
Wesley H. Holliday 2
Department of Philosophy &
Group in Logic and the Methodology of Science
University of California, Berkeley
Abstract
It is a classic result in lattice theory that a poset is a complete lattice iff it can be
realized as fixpoints of a closure operator on a powerset. Dragalin [9,10] observed
that a poset is a locale (complete Heyting algebra) iff it can be realized as fixpoints
of a nucleus on the locale of upsets of a poset. He also showed how to generate a
nucleus on upsets by adding a structure of “paths” to a poset, forming what we call
a Dragalin frame. This allowed Dragalin to introduce a semantics for intuitionistic
logic that generalizes Beth and Kripke semantics. He proved that every spatial locale
(locale of open sets of a topological space) can be realized as fixpoints of the nucleus
generated by a Dragalin frame. In this paper, we strengthen Dragalin’s result and
prove that every locale—not only spatial locales—can be realized as fixpoints of the
nucleus generated by a Dragalin frame. In fact, we prove the stronger result that for
every nucleus on the upsets of a poset, there is a Dragalin frame based on that poset
that generates the given nucleus. We then compare Dragalin’s approach to generating
nuclei with the relational approach of Fairtlough and Mendler [11], based on what we
call FM-frames. Surprisingly, every Dragalin frame can be turned into an equivalent
FM-frame, albeit on a different poset. Thus, every locale can be realized as fixpoints of
the nucleus generated by an FM-frame. Finally, we consider the relational approach of
Goldblatt [13] and characterize the locales that can be realized using Goldblatt frames.
Keywords: nucleus, locale, Heyting algebra, intuitionistic logic, lax logic
1
Introduction
A well-known result of Shehtman [30] (cf. [23]) shows that there are intermediate logics that cannot be characterized by Kripke frames [21]. This incompleteness result renewed interest in the earlier topological semantics for intuitionistic
1
2
guram@math.nmsu.edu.
wesholliday@berkeley.edu.
2
Locales, Nuclei, and Dragalin Frames
logic due to Tarski [33]. It remains a famous open problem of Kuznetsov [22]
whether every intermediate logic is topologically complete. Dragalin [9,10]
made the important but somewhat neglected observation that by generalizing
Kripke frames in a way inspired by Beth semantics [4], one obtains a semantics
for intuitionistic logic that is at least as general as topological semantics.
Dragalin frames are triples (S, ≤, D) where (S, ≤) is a poset and
D : S → ℘(℘(S)) satisfies natural conditions stated below. 3 Dragalin called
each X ∈ D(x) a path starting from x. In the literature on Beth semantics,
‘path’ suggests a linearly ordered subset of (S, ≤), so we will instead call X a
development starting from x and elements of X stages of the development. For
the poset (S, ≤), we use the following standard notation for Y ⊆ S and y ∈ S:
•
↑Y = {z ∈ S | ∃y ∈ Y : y ≤ z} and ↑y = ↑{y};
•
↓Y = {z ∈ S | ∃y ∈ Y : z ≤ y} and ↓y = ↓{y}.
A subset U of S is an upset (upward closed set) if x ∈ U implies ↑x ⊆ U .
A downset (downward closed set) is defined dually. If X ⊆ ↓Y , so ∀x ∈ X
∃y ∈ Y : x ≤ y (every stage of development in X is extended by a stage of
development in Y ), then we say that X is bounded by Y .
Definition 1.1 A Dragalin frame is a triple F = (S, ≤, D) where (S, ≤) is a
poset and D : S → ℘(℘(S)) (a Dragalin function) is such that for all x, y ∈ S:
(1◦ ) ∅ 6∈ D(x);
(2◦ ) if y ∈ X ∈ D(x), then ∃z ∈ X: x ≤ z and y ≤ z;
(3◦ ) if x ≤ y, then ∀Y ∈ D(y) ∃X ∈ D(x): X ⊆ ↓Y ;
(4◦ ) if y ∈ X ∈ D(x), then ∃Y ∈ D(y): Y ⊆ ↓X.
A Dragalin frame is normal if D(x) 6= ∅ for all x ∈ S.
Conditions (1◦ )–(4◦ ) admit intuitive intepretations. (1◦ ) says that the
empty set is not a development of anything. (2◦ ) says that every stage y
in a development starting from x is at least compatible with x, in that x and y
have a common extension z. Dragalin also mentions the stronger condition:
(2◦◦ ) if X ∈ D(x), then X ⊆ ↑x,
so the stages in a development starting from x are extensions of x. Next, (3◦ )
says that if at some “future” stage y a development Y will become available,
then it is already possible to follow a development that is bounded by Y .
Dragalin also mentions the stronger condition:
(3◦◦ ) if x ≤ y, then D(y) ⊆ D(x),
3 Dragalin used the term ‘Beth-Kripke frame’. To give due credit to Dragalin, we introduce
the term ‘Dragalin frame’ instead. Note that Dragalin started with a preordered set (S, ≤),
but there is no loss of generality in starting with a poset (see Remark 2.6 and Theorem 3.5).
Also note that Dragalin worked with downsets in (S, ≤). Others, including Goldblatt [13,14]
and Fairtlough and Mendler [11], work instead with upsets. It proves to be more convenient
to flip Dragalin’s approach to use upsets than to flip the other approaches to use downsets.
Thus, we will work with upsets, at the expense of another flip of perspective in Theorem 2.8.
Bezhanishvili and Holliday
3
so developments available at future stages are already available. Finally, (4◦ )
says that we “can always stay inside” a development, in the sense that for every
stage y in X, we can follow a development Y from y that is bounded by X. A
stronger notion of “staying inside” comes from replacing Y ⊆ ↓X with Y ⊆ X:
(4◦◦ ) if y ∈ X ∈ D(x), then ∃Y ∈ D(y): Y ⊆ X.
In §3 we will see that (2◦◦ ), (3◦◦ ), and (4◦◦ ) can be assumed without loss of
generality. This motivates the following definition.
Definition 1.2 A Dragalin frame is standard if it satisfies (2◦◦ )–(4◦◦ ).
To use Dragalin frames to give semantics for the language of propositional
logic, we use models M = (S, ≤, D, V ) where (S, ≤, D) is a Dragalin frame and
V assigns to each propositional variable an upset V (p) of (S, ≤) with the property that x ∈ V (p) iff ∀X ∈ D(x) X ∩V (p) 6= ∅, i.e., each development starting
from x hits the interpretation of p. The forcing clauses for the connectives are
the same as in intuitionistic Kripke semantics except for ⊥ and ∨:
•
M, x
⊥ iff D(x) = ∅;
•
M, x
ϕ ∨ ψ iff ∀X ∈ D(x) ∃y ∈ X: M, y
ϕ or M, y
ψ.
What is going on here is that we are evaluating formulas not in the full Heyting algebra Up(S, ≤) of upsets of (S, ≤), as in Kripke semantics, but rather in
the Heyting algebra of just those upsets U such that x ∈ U iff ∀X ∈ D(x)
X ∩ U 6= ∅. As Dragalin explained, and as we review in §2, the function D
gives rise to a nucleus on the Heyting algebra Up(S, ≤), and we are evaluating
formulas in the Heyting algebra of fixpoints of this nucleus. 4 This idea of
evaluating formulas as fixpoints of a closure operator, of which a nucleus is a
special case, and interpreting disjunction by taking the closure of the union appears in semantics for substructural logic [27, §12.2] and in recent philosophical
discussions of the relation between intuitionistic and classical logic [28,29].
Nuclei play an important role in pointfree topology [18]. The pointfree
generalization of a topological space is a complete Heyting algebra, also known
as a locale, 5 and nuclei on a locale describe sublocales of the locale. The locale
Up(S, ≤) is a very special locale (see Remark 2.7). Shehtman’s result shows
that not all intermediate logics are complete with respect to such locales. On
the other hand, Dragalin proved that every locale can be represented as the
algebra of fixpoints of a nucleus on some Up(S, ≤). This representation theorem
is related to the classic result in lattice theory that every complete lattice can
be represented as the lattice of fixpoints of a closure operator on a powerset.
Dragalin’s representation theorem motivates the notion of a nuclear frame
that we introduce in §2. The question is then how a nucleus on Up(S, ≤) can
be realized more concretely. Dragalin frames do so with the function D, and
4 Dragalin [9,10] used the term ‘completion operator’ instead of ‘nucleus’. In this paper, we
follow the now standard terminology and notation from pointfree topology.
5 Locales are also known as frames in the pointfree topology literature, but in this paper we
use the term ‘frame’ as it is used in the modal logic literature.
4
Locales, Nuclei, and Dragalin Frames
Dragalin proved that every spatial locale (locale of open sets of a topological
space) can be represented as fixpoints of the nucleus generated by D for some
Dragalin frame. In §3, we prove that every locale—not only spatial locales—can
be represented as fixpoints of the nucleus generated by a Dragalin frame. In
fact, we prove the stronger result that for every nucleus on the upsets of a poset,
there is a Dragalin frame based on that poset that realizes the given nucleus.
In §4, we compare Dragalin’s approach to generating nuclei with the relational
approach of Fairtlough and Mendler [11], based on what we call FM-frames.
Surprisingly, every Dragalin frame can be turned into an equivalent FM-frame,
albeit on a different poset. Thus, every locale can be represented as fixpoints
of the nucleus generated by an FM-frame. Finally, in §5, we consider the
relational approach of Goldblatt [13] and characterize the locales representable
using Goldblatt frames. We conclude in §6 with directions for further research.
2
Nuclear Frames
Our basic objects of study will be nuclei on Heyting algebras and locales
[24,25,12,18].
Definition 2.1 A nucleus on a Heyting algebra H is a function j : H → H
such that for all a, b ∈ H: a ≤ ja (inflationary); jja = ja (idempotent);
j(a ∧ b) = ja ∧ jb (multiplicative). A nucleus is dense if j0 = 0.
A nuclear algebra is a pair H = (H, j) where H is a Heyting algebra and j
is a nucleus on H. It is a localic nuclear algebra if H is a locale.
The following result is well known (see, e.g., [10, p. 71]).
Theorem 2.2 If H = (H, ∧, ∨, →, 0, j) is a nuclear algebra, then the algebra of fixpoints Hj = (Hj , ∧j , ∨j , →j , 0j ) is a Heyting algebra where Hj =
{a ∈ H | a = ja} is the set of fixpoints of j in H and for a, b ∈ Hj :
(i) a ∧j b = a ∧ b;
(ii) a ∨j b = j(a ∨ b);
(iii) a →j b = a → b;
(iv) 0j = j0.
If H is a localic nuclear algebra, then Hj is a locale, where for X ⊆ Hj :
W
V
W
V
(vi) j X = j X.
(v) j X = X;
We will often abuse notation and conflate Hj and Hj .
Remark 2.3 An important example of a nucleus on a Heyting algebra is the
nucleus of double negation ¬¬, where ¬x = x → 0. It is well known that
the algebra of fixpoints of double negation as in Theorem 2.2 forms a Boolean
algebra, which is complete if the original Heyting algebra is complete. For
example, since the Heyting algebra Up(S, ≤) of upsets in a poset is a locale,
the algebra of fixpoints in (Up(S, ≤), ¬¬) is a complete Boolean algebra. Going
from a poset to a complete Boolean algebra in this way is a standard technique
in set theory for relating forcing posets to Boolean-valued models [32].
The nuclei on a Heyting algebra H are naturally ordered by j ≤ k iff ja ≤ ka
for all a ∈ H. It is well known that in the case of a locale L, the collection
Bezhanishvili and Holliday
5
N (L) of all nuclei on L with the natural ordering is itself a locale (see, e.g., [12,
Th. 2.20], [18, Prop. II.2.5]). Meets in N (L) are computed pointwise, whereas
joins are more difficult to describe.
There are several families of nuclei that can generate all nuclei in N (L).
Here we focus on the following: given a ∈ L, define the function wa on L by
wa b = (b → a) → a.
(1)
It can be verified that wa is a nucleus. Note that when a = 0, w0 is the nucleus
of double negation. 6 One of the special roles of these nuclei is shown by the
following observation of Simmons [31, p. 243], which we will utilize in Theorem
3.5. To keep the paper self-contained, we include a proof.
Lemma 2.4 (Simmons) Given a locale L and a nucleus j on L,
^
j = {wja | a ∈ L}.
Proof. First, observe that for any nucleus k on L,
k ≤ wa iff ka ≤ a.
(2)
From left to right, if k ≤ wa , then ka ≤ wa a = (a → a) → a = a. From
right to left, the multiplicativity of k implies k(b → a) ≤ kb → ka, and the
inflationarity of k yields (b → a) ≤ k(b → a). It follows that kb ∧ (b → a) ≤
kb ∧ k(b → a) ≤ ka and hence kb ≤ (b → a) → ka. Thus, if ka ≤ a, then we
have kb ≤ (b → a) → ka ≤ (b → a) → a = wa b for every b ∈ L, so k ≤ wa .
It follows from (2) that j is a lower bound of {wja | a ∈ L}. To see that it
is the greatest, suppose k is also a lower bound of {wja | a ∈ L}, so for every
a ∈ L, we have k ≤ wja . Then (2) implies ka ≤ ja for every a ∈ L, so k ≤ j.
Therefore, j is the greatest lower bound.
✷
In this paper, we are interested in nuclear algebras in which the underlying
Heyting algebra is the locale Up(S, ≤) of upsets of some poset (S, ≤), in which
implication is defined by U → V = {x ∈ S | ↑x ∩ U ⊆ V }. The locale
Down(S, ≤) of downsets of (S, ≤) is defined dually.
Definition 2.5 A nuclear frame is a triple F = (S, ≤, j) where (S, ≤) is a
poset and j is a nucleus on Up(F) := Up(S, ≤). We say that F is dense if j is
dense. The nuclear algebra of F is the nuclear algebra (Up(F), j).
Remark 2.6 One could also allow the relation ≤ in Definition 2.5 to be a
preorder. However, as is well known, for any preordered set (S, ≤), taking its
quotient with respect to the equivalence relation defined by x ∼ y iff x ≤ y
and y ≤ x produces a poset (S ′ , ≤′ ), called the skeleton of (S, ≤), such that
Up(S ′ , ≤′ ) is isomorphic to Up(S, ≤). Thus, any preordered nuclear frame
(S, ≤, j) can be turned into a partially ordered nuclear frame (S ′ , ≤′ , j ′ ) such
that their nuclear algebras are isomorphic.
6 As in the case of double negation, for any nuclear algebra of the form (H, w ), its algebra
a
of fixpoints is a Boolean algebra (see [12, p. 330], [18, p. 51]).
6
Locales, Nuclei, and Dragalin Frames
Remark 2.7 Nuclear frames generate only a special class of localic nuclear
algebras. Recall
W that for a locale L, an element a ∈ L is completely join-prime
if from a ≤ X it follows that a ≤ x for some x ∈ X. Let J ∞ (L) be the set of
completely join-prime elements of L. We call a locale L Alexandroff if J ∞ (L) is
join-dense in L, i.e., each element of L is the join of completely join-prime elements below it. Then L is Alexandroff iff L is isomorphic to the locale of upsets
of a poset (see, e.g., [8,5]) and hence to the locale of open sets in an Alexandroff
space. Thus, nuclear frames generate exactly the localic nuclear algebras based
on Alexandroff locales, which we call Alexandroff nuclear algebras.
Although not every localic nuclear algebra can be represented as the nuclear
algebra of a nuclear frame, nonetheless every locale can be represented as the
algebra of fixpoints in the nuclear algebra of a nuclear frame. To keep the paper
self-contained, we include a proof of this important result from [10, p. 75].
Theorem 2.8 (Dragalin) A poset P is a locale iff there is a dense nuclear
frame F such that P is isomorphic to the algebra of fixpoints in the nuclear
algebra of F.
Proof. From right to left, since Up(F) is a locale, the algebra of fixpoints in
the nuclear algebra of F is a locale by Theorem 2.2.
From left to right, suppose P is a locale. Let S = P \ {0} and ≤ be the
restricted order. We will build a nuclear frame F whose poset is (S, ≥). Since
Up(S, ≥) = Down(S, ≤), we can work with the locale of downsets in the poset
(S, ≤). Define a unary function j on this locale by
_
jX = ↓ X,
(3)
W
where X is the join of X in P , which exists since P is complete, and ↓ indicates the downset
in (S, ≤).W It is easyWto see that j is inflationary, idempotent,
W
and
that
↓
(X
∩
W
WY ) ⊆ (↓ WX) ∩ (↓ Y ) for X, Y ∈ Down(S, ≤). To see that
↓ (X ∩WY ) ⊇ (↓ X) W
∩ (↓ Y ), suppose that
W a∈S
W is in the right hand side,
so a ≤ X and a ≤ Y , whence a ≤ ( X) ∧ ( Y ). By the join-infinite
distributive law for locales,
_
_
_
( X) ∧ ( Y ) = {x ∧ y | x ∈ X, y ∈ Y },
W
so a ≤ {x ∧ y | x ∈ X, y ∈ Y }. Since
W X and Y are downsets,
W we have {x ∧ y |
x
∈
X,
y
∈
Y
}
⊆
(X
∩Y
)∪{0},
so
{x∧y
|
x
∈
X,
y
∈
Y
}
≤
((X ∩Y )∪{0}) =
W
W
W
(X ∩ Y ). Thus, a ≤ (X ∩ Y ) and hence a ∈ ↓ W
(X ∩ Y ). Therefore, j is a
nucleus. To see that j is dense, observe that j∅ = ↓ ∅ = ↓0 = ∅ since 0 6∈ S.
Finally, we must check that our original locale P is isomorphic to the algebra
of fixpoints in the nuclear algebra of F. Observe that the fixpoints of j in the
nuclear algebra of F are exactly the principal downsets in (S, ≤) plus ∅. Thus,
the map sending each x to ↓x is the desired isomorphism.
✷
Remark 2.9 The proof technique of Theorem 2.8 is related to a standard
technique in set theory, whereby one goes from a complete Boolean algebra to
Bezhanishvili and Holliday
7
a poset by deleting the bottom element [32]. The poset thereby obtained is a
separative poset (if y 6≤ x, then ∃y ′ ≤ y ∀y ′′ ≤ y ′ : y ′′ 6≤ x). Moreover, the
nucleus j defined in (3) above is the nucleus of double negation [6].
Theorem 2.8 shows that nuclear frames suffice to represent arbitrary locales.
However, since nuclear frames are a mix of the concrete (S, ≤) and the algebraic
j, it is natural to ask if we can replace the nucleus j with more concrete data
from which j can be recovered. We will answer this question in the next section.
3
Dragalin Frames
The Dragalin frames of Definition 1.1 replace the nucleus j in a nuclear frame
(S, ≤, j) with the function D : S → ℘(℘(S)). As shown by Dragalin [10, pp. 7273], this D indeed gives rise to a nucleus, as in Proposition 3.1. Given its
importance in our story, we include a proof of this result.
Proposition 3.1 (Dragalin) Given a Dragalin frame F = (S, ≤, D), define
a function [Di on Up(F) by
[DiU = {x ∈ S | ∀X ∈ D(x) : X ∩ U 6= ∅}.
(4)
(i) [Di is a nucleus on Up(F);
(ii) [Di is a dense nucleus iff F is normal.
We call (Up(F), [Di) the nuclear algebra of F.
Proof. For part (i), to see that U ∈ Up(F) implies [DiU ∈ Up(F), suppose
x ∈ [DiU and x ≤ y. For each Y ∈ D(y), by (3◦ ) there is an X ∈ D(x) with
X ⊆ ↓Y . Since x ∈ [DiU , X ∩ U 6= ∅, which with U ∈ Up(F) and X ⊆ ↓Y
implies Y ∩ U 6= ∅. Since this holds for each Y ∈ D(y), we have y ∈ [DiU .
For inflationarity, for any X ∈ D(x), there is a z ∈ X with x ≤ z by
(1◦ )–(2◦ ). So if x ∈ U ∈ Up(F), then z ∈ U , so X ∩ U 6= ∅. Hence x ∈ [DiU .
For idempotence, suppose x 6∈ [DiU , so there is an X ∈ D(x) such that
X ∩ U = ∅. We claim that X ∩ [DiU = ∅. For any y ∈ X, by (4◦ ) we have a
Y ∈ D(y) such that Y ⊆ ↓X. For reductio, suppose there is a z ∈ Y ∩ U . Then
since Y ⊆ ↓X, there is a z ′ ∈ X such that z ≤ z ′ , which with z ∈ U ∈ Up(F)
implies z ′ ∈ U . But then z ′ ∈ X ∩ U , contradicting X ∩ U = ∅ from above.
Hence Y ∩ U = ∅, so y 6∈ [DiU . Since this holds for all y ∈ X, X ∩ [DiU = ∅,
which with X ∈ D(x) implies x 6∈ [Di[DiU .
Finally, [Di is monotonic (A ⊆ B implies [DiA ⊆ [DiB) by its definition,
so [Di(U ∩ U ′ ) ⊆ [DiU ∩ [DiU ′ . Conversely, if we can show that A ∩ [DiA′ ⊆
[Di(A ∩ A′ ), then by two applications of this fact, plus monotonicity and idempotence, [DiU ∩ [DiU ′ ⊆ [Di([DiU ∩ U ′ ) ⊆ [Di[Di(U ∩ U ′ ) ⊆ [Di(U ∩ U ′ ). So
suppose x ∈ A ∩ [DiA′ . Then for X ∈ D(x), there is a y ∈ X ∩ A′ . Therefore,
by (2◦ ), there is a z ∈ X with x ≤ z and y ≤ z. Since x ∈ A, y ∈ A′ , and
A, A′ ∈ Up(F), we have z ∈ A ∩ A′ , so X ∩ A ∩ A′ 6= ∅. Thus, x ∈ [Di(A ∩ A′ ).
For part (ii), by (4) we have that [Di∅ = {x ∈ S | D(x) = ∅}, so [Di∅ = ∅
([Di is dense) iff D(x) 6= ∅ for all x ∈ S (F is normal).
✷
8
Locales, Nuclei, and Dragalin Frames
Example 3.2 For any poset (S, ≤) and x ∈ S, define D(x) = {↑y | x ≤ y}.
One can easily check that (S, ≤, D) is a standard Dragalin frame as in Definition
1.2. Observe that x ∈ [DiU iff ∀y ≥ x ∃z ≥ y: z ∈ U , so [Di is the nucleus of
double negation (recall Remark 2.3).
The obvious next question is: which nuclear frames can be generated by
Dragalin frames as in Proposition 3.1? In addition, in light of Theorem 2.8,
another obvious question is: which locales can be generated as the algebra of
fixpoints in the nuclear algebra of a Dragalin frame?
Dragalin gave a partial answer to the second question. We provide a sketch
of his proof [10, pp. 75-76] of Theorem 3.3 to convey the main idea, but we omit
the details since we will prove a more general result below. Recall that a spatial
locale is a locale isomorphic to the locale of open sets of a topological space,
and a normal Dragalin frame (S, ≤, D) is one in which D(x) 6= ∅ for all x ∈ S.
Theorem 3.3 (Dragalin) If L is a spatial locale, then there is a normal Dragalin frame F such that L is isomorphic to the algebra of fixpoints in the nuclear
algebra of F.
Proof. [Sketch] Given a T
topological space (X, Ω) and x ∈ X, a B ⊆ Ω is a
local basis of x if (i) x ∈ B and (ii) whenever x ∈ U ∈ Ω, there is a V ∈ B
with V ⊆ U . From (X, Ω), we define (S, ≤, D) where S = Ω \ {∅},
S U ≤ V
iff U ⊇ V , and D(U ) = {B | ∃x ∈ U : B is a local basis of x and
B ⊆ U }.
Dragalin showed that (S, ≤, D) is a normal Dragalin frame. In addition, for
the locale Ω(X) of opens of (X, Ω), he showed that f : Ω(X) → Up(S, ≤)[Di
defined by f (U ) = {V ∈ S | U ≤ V } is an isomorphism.
✷
In the other direction, it is not the case that for every Dragalin frame F,
the algebra of fixpoints in the nuclear algebra of F is spatial.
Example 3.4 Consider the Dragalin frame (S, ≤, D) where (S, ≤) is the poset
associated with the complete infinite binary tree and D gives the nucleus of
double negation as in Example 3.2. Then the algebra of fixpoints in the nuclear
algebra (Up(S, ≤), [Di) is a complete atomless Boolean algebra [16, Ex. 2.40].
But a complete Boolean algebra is spatial iff it is atomic.
Indeed, every locale can be realized as the algebra of fixpoints in the nuclear
algebra of a Dragalin frame. To prove this, it would suffice to show that the
nuclear frame used in the proof of Theorem 2.8 can be generated by a Dragalin
frame. We will prove the following stronger result.
Theorem 3.5 Given any nuclear frame (S, ≤, j), there is a standard Dragalin
frame (S, ≤, D) such that j = [Di.
To prove Theorem 3.5, we will utilize Simmons’s result (Lemma 2.4) that
every nucleus on a locale L is the meet, in N (L), of nuclei of the wa type. We
will apply this to the locale L = Up(S, ≤). First, we show how nuclei of the wa
type on Up(S, ≤) can be realized by a Dragalin frame (S, ≤, D).
Lemma 3.6 Given a poset (S, ≤) and A ∈ Up(S, ≤), let wA be the nucleus on
Up(S, ≤) defined as in (1) and define DA (x) = {↑x′ \ A | x′ ∈ ↑x \ A}. Then:
Bezhanishvili and Holliday
9
(i) (S, ≤, DA ) is a standard Dragalin frame;
(ii) [DA i = wA .
Proof. For (i), DA satisfies (1◦ ) because if x′ ∈ ↑x \ A, then x′ ∈ ↑x′ \ A,
which shows that ∅ 6∈ D(x). Clearly DA also satisfies (2◦◦ ), as well as (3◦◦ ),
for if x ≤ y and y ′ ∈ ↑y \ A, then y ′ ∈ ↑x \ A. Finally, for (4◦◦ ), suppose
y ∈ X ∈ D(x), so X = ↑x′ \ A for some x′ ∈ ↑x \ A. Let Y = ↑y \ A. Since
y ∈ X and hence y 6∈ A, we have y ∈ Y , which implies Y ∈ D(y). Moreover,
since y ∈ X and hence y ∈ ↑x′ , we have ↑y \ A ⊆ ↑x′ \ A, so Y ⊆ X.
For (ii), observe that x ∈ wA U = (U → A) → A iff for all x′ ≥ x, if x′ 6∈ A,
then there is an x′′ ≥ x′ such that x′′ ∈ U \ A. This is equivalent to the
condition that for all x′ ∈ ↑x \ A, we have (↑x′ \ A) ∩ U 6= ∅. That is in turn
equivalent to the condition that for all X ∈ DA (x), X ∩ U 6= ∅, which is finally
equivalent to x ∈ [DA iU .
✷
Next we show that we can build up meets of nuclei from Dragalin frames.
Lemma 3.7 Given a family {jα }α∈I of nuclei on Up(S, ≤) and a family
{Dα }α∈I of Dragalin functions on (S, ≤) such that
x ∈ jα U iff x ∈ [Dα iU,
(5)
the function D defined by
D(x) =
[
Dα (x)
(6)
α∈I
is a Dragalin function such that
^
jα U iff x ∈ [DiU.
x∈
(7)
α∈I
Moreover, if each Dα is standard, then so is D.
Proof. We first prove (7). Since meets of nuclei are computed pointwise and
meets in Up(S, ≤) are intersections, we have
\
^
^
jα U.
jα U =
jα U =
α∈I
Now suppose x 6∈
T
α∈I
α∈I
jα U , so there is some α ∈ I such that x 6∈ jα U . Then by
α∈I
(5), we have x 6∈ [Dα iU , so there is some X ∈ Dα (x) with X ∩ U = ∅. By
(6), X ∈ D(x), which with X ∩ U = ∅ implies x 6∈ [DiU . Conversely, suppose
x 6∈ [DiU , so there is some X ∈ D(x) with X ∩ U = ∅. Then by (6), there is
some α ∈ I such that X ∈ Dα (x), which
T with X ∩ U = ∅ implies x 6∈ [Dα iU ,
which implies x 6∈ jα U by (5), so x 6∈
jα U .
α∈I
Next, we show that D satisfies (1◦ )–(4◦ ), assuming that each Dα does.
Clearly if (1◦ ) holds for each Dα , then it holds for D. The same is true of (2◦ ).
10
Locales, Nuclei, and Dragalin Frames
For (3◦ ), suppose x ≤ y and Y ∈ D(y). Then Y ∈ Dα (x) for some α ∈ I.
Applying (3◦ ) for Dα , there is an X ⊆ ↓Y such that X ∈ Dα (x) and hence
X ∈ D(x), so (3◦ ) also holds for D. Similarly, for (4◦ ), if y ∈ X ∈ D(x), then
X ∈ Dα (x) for some α ∈ I, in which case (4◦ ) for Dα gives us a Y ⊆ ↓X such
that Y ∈ Dα (y) and hence Y ∈ D(y), so (4◦ ) also holds for D. It is also easy
to see that if each Dα satisfies (2◦◦ )–(4◦◦ ), then so does D.
✷
We can now put the pieces together to prove Theorem 3.5.
Proof. By Lemmas 2.4, 3.6, and 3.7, we have:
^
^
j = {wjA | A ∈ Up(S, ≤)} = {[DjA i | A ∈ Up(S, ≤)} = [Di,
where D is defined from the DjA ’s as in Lemma 3.7. By Lemma 3.6, each DjA
is standard, so D is as well by Lemma 3.7.
✷
Putting together Theorems 2.8 and 3.5 and Proposition 3.1(ii), we obtain
the following.
Corollary 3.8 A poset P is a locale iff there is a standard normal Dragalin
frame F such that P is isomorphic to the algebra of fixpoints in the nuclear
algebra of F.
In §4, we shall see that an analogue of Corollary 3.8 holds for frames that
replace Dragalin’s function D with a partial order . For the purposes of
comparing these frames, we will use the fact that every Dragalin frame can be
turned into one satisfying a property stronger than (2◦◦ ).
Definition 3.9 A Dragalin frame F = (S, ≤, D) is convex if for all x ∈ S and
X ∈ D(x), we have X = ↑x ∩ ↓X.
Remark 3.10
(i) If a Dragalin frame F = (S, ≤, D) is convex, then X ∈ D(x) implies x ∈ X.
For if X ∈ D(x), then by (1◦ ), there is a y ∈ X = ↑x ∩ ↓X, so x ≤ y and
y ∈ ↓X, which implies x ∈ ↓X and hence x ∈ ↑x ∩ ↓X = X.
(ii) A convex F typically does not satisfy (3◦◦ ). Consider x, y ∈ S such that
D(y) 6= ∅. By convexity each Y ∈ D(y) is such that Y ⊆ ↑y, and by (i)
each X ∈ D(x) is such that x ∈ X, so if x < y, then D(x) ∩ D(y) = ∅.
(iii) By contrast, every convex F satisfies (4◦◦ ). Suppose y ∈ X ∈ D(x), so by
(4◦ ) there is a Y ∈ D(y) such that Y ⊆ ↓X. By convexity, Y ⊆ ↑y and
X = ↑x ∩ ↓X. Since y ∈ X ∈ D(x) implies x ≤ y by convexity, we have
↑y ⊆ ↑x. Thus, Y ⊆ ↑x, which with Y ⊆ ↓X implies Y ⊆ ↑x ∩ ↓X = X.
Proposition 3.11 For each Dragalin frame F, there is a convex Dragalin
frame G such that the nuclear algebras of F and G are isomorphic. Moreover,
if F is normal, then so is G.
Proof. Given a Dragalin frame F = (S, ≤, D), define G = (S, ≤, D′ ) by:
D′ (x) = {↑x ∩ ↓X | X ∈ D(x)}.
Bezhanishvili and Holliday
11
We claim that for all U ∈ Up(S, ≤), [DiU = [D′ iU . Suppose x ∈ [DiU and
consider some ↑x ∩ ↓X ∈ D′ (x), so X ∈ D(x). Since x ∈ [DiU and X ∈ D(x),
there is a y ∈ X ∩ U . Then by (2◦ ), there is a z ∈ X with x ≤ z and y ≤ z.
From z ∈ X and x ≤ z, we have z ∈ ↑x ∩ X ⊆ ↑x ∩ ↓X. From y ∈ U and
y ≤ z, we have z ∈ U . Therefore, (↑x ∩ ↓X) ∩ U 6= ∅ and hence x ∈ [D′ iU .
Conversely, if x 6∈ [DiU , so there is an X ∈ D(x) such that X ∩ U = ∅, then
since U is an upset, (↑x ∩ ↓X) ∩ U = ∅ and ↑x ∩ ↓X ∈ D′ (x), so x 6∈ [D′ iU . It
follows that the nuclear algebras of F and G are isomorphic.
Next we show that G satisfies (1◦ )–(4◦ ), so it is a Dragalin frame. Since G
is convex, (2◦ ) is immediate. For (1◦ ) for D′ , if X ∈ D(x), then by (1◦ )–(2◦ )
for D, ↑x ∩ X 6= ∅, so ↑x ∩ ↓X 6= ∅ and hence ∅ 6∈ D′ (x).
For (3◦ ) for D′ , suppose x ≤ y and Y ′ ∈ D′ (y), so Y ′ = ↑y ∩ ↓Y for some
Y ∈ D(y). Then by (3◦ ) for D, there is an X ∈ D(x) such that X ⊆ ↓Y .
Setting X ′ = ↑x ∩ ↓X, we have X ′ ∈ D′ (x), and we claim that X ′ ⊆ ↓Y ′ ,
which will establish (3◦ ) for D′ . If a ∈ X ′ , then a ∈ ↓X, so there is a b with
a ≤ b ∈ X. Then since X ⊆ ↓Y , there is a c with b ≤ c ∈ Y . Given Y ∈ D(y),
it follows by (2◦ ) for D that there is a z ∈ Y such that y ≤ z and c ≤ z. Thus,
a ≤ b ≤ c ≤ z ∈ ↑y ∩ Y ⊆ ↑y ∩ ↓Y , so a ∈ ↓(↑y ∩ ↓Y ) = ↓Y ′ .
Next we prove (4◦◦ ). Suppose y ∈ X ′ ∈ D′ (x), so X ′ = ↑x ∩ ↓X for some
X ∈ D(x). We need a Y ′ ∈ D′ (y) with Y ′ ⊆ X ′ , so we need a Y ∈ D(y) with
↑y∩↓Y ⊆ ↑x∩↓X. Since y ∈ X ′ , y ∈ ↓X, so there is a z with y ≤ z ∈ X. Given
X ∈ D(x), it follows by (4◦ ) for D that there is a Z ∈ D(z) such that Z ⊆ ↓X.
Then given y ≤ z, it follows by (3◦ ) for D that there is a Y ∈ D(y) such that
Y ⊆ ↓Z, which with Z ⊆ ↓X implies Y ⊆ ↓X, which in turn implies ↓Y ⊆ ↓X.
Since y ∈ X ′ , y ∈ ↑x, so we also have ↑y ⊆ ↑x. Therefore, ↑y ∩ ↓Y ⊆ ↑x ∩ ↓X.
Finally, it is obvious that if F is normal, then G is normal too.
✷
4
Fairtlough-Mendler Frames
In this section, we consider another way of replacing the nucleus j in a nuclear
frame with more concrete data. Fairtlough and Mendler [11] (also see [1])
give a semantics for an intuitionistic modal logic called propositional lax logic
with a modality
obeying the axioms of a nucleus. The frames used in their
semantics therefore provide another method for representing nuclear algebras.
Definition 4.1 An FM-frame (Fairtlough-Mendler frame) is a tuple F =
(S, ≤, , F ), where ≤ and are preorders on the set S, is a subrelation
of ≤, and F ∈ Up(F) := Up(S, ≤). We say that F is normal if F = ∅, and F is
partially ordered if ≤ (and hence ) is a partial order.
Each FM-frame gives rise to a nuclear algebra. To see this, for each FMframe F = (S, ≤, , F ), let Up(F)F = {U ∈ Up(F) | F ⊆ U } be the relativization of Up(F) to F . Then Up(F)F is a locale, where the operations
∧, ∨, → on Up(F)F are the restrictions of the corresponding operations on
Up(F), and F is the bottom element of Up(F)F . When F = ∅, we obviously have Up(F)F = Up(F). In order to define a nucleus on Up(F)F , recall
that for a binary relation R on S, x ∈ S, and U ⊆ S, it is customary to define
12
Locales, Nuclei, and Dragalin Frames
R(x) = {y ∈ S | xRy} and let
✷R (U ) = {x ∈ S | R(x) ⊆ U } and ✸R (U ) = {x ∈ S | R(x) ∩ U 6= ∅}.
Now consider the following operator on Up(F)F :
✷≤ ✸ U = {x ∈ S | ∀y(x ≤ y ⇒ ∃z(y z & z ∈ U ))}.
Fairtlough and Mendler use this operator to interpret the modality
of lax
logic in FM-frames. The following proposition is essentially their soundness
result for lax logic [11, p. 9], which we prove for the reader’s convenience.
Proposition 4.2 (Fairtlough-Mendler) If F = (S, ≤, , F ) is an FMframe, then (Up(F)F , ✷≤ ✸ ) is a nuclear algebra, which we call the nuclear
algebra of F. Moreover, if F = ∅, then ✷≤ ✸ is a dense nucleus.
Proof. For any U ∈ Up(F)F , since is a reflexive subrelation of ≤, we
have U ⊆ ✷≤ ✸ U . Since ≤ is reflexive and is transitive, we also have
✷≤ ✸ ✷≤ ✸ U ⊆ ✷≤ ✸ U . Clearly ✷≤ ✸ (U ∩ V ) ⊆ ✷≤ ✸ U ∩ ✷≤ ✸ V .
Conversely, for any U, V ∈ Up(F)F , if x ∈ ✷≤ ✸ U ∩ ✷≤ ✸ V and x ≤ y,
then there is a z with y z ∈ U and hence y ≤ z. So x ≤ z, which with
x ∈ ✷≤ ✸ V implies that there is a w with z w ∈ V and hence z ≤ w. Then
since z ∈ U ∈ Up(F)F , we have w ∈ U ∩ V . Thus, x ∈ ✷≤ ✸ (U ∩ V ). Finally,
since ≤ is reflexive, ✷≤ ✸ ∅ = ∅, so ✷≤ ✸ is dense if F = ∅.
✷
Example 4.3 If ≤ = in an FM-frame (S, ≤, , F ), then ✷≤ ✸ is the nucleus
of double negation (cf. [11, p. 23] and [3]).
Remark 4.4 Let us now consider extracting a nuclear frame rather than a
nuclear algebra from an FM-frame. If the FM-frame F = (S, ≤, , F ) is normal,
then (S, ≤, ✷≤ ✸ ) is a preordered nuclear frame as in Remark 2.6, which we
can turn into a partially ordered nuclear frame by taking the skeleton, and
the nuclear algebra of this nuclear frame is isomorphic to that of F. Now
suppose F is not normal. Define the preordered nuclear frame (S − , ≤− , j)
where S − = S \ F , ≤− is the restriction of ≤ to S − , and j is defined for
U ∈ Up(S − , ≤− ) by jU = ✷≤ ✸ (U ∪F )\F . The nuclear algebra of (S − , ≤− , j)
is then isomorphic to that of F, and once again we can turn (S − , ≤− , j) into a
partially ordered nuclear frame as in Remark 2.6.
Turning FM-frames into partially ordered FM-frames is more difficult. It is
not clear how to define the skeleton F′ = (S ′ , ≤′ , ′ , F ′ ) of an FM-frame F =
(S, ≤, , F ) such that (Up(F)F , ✷≤ ✸ ) and (Up(F′ )F ′ , ✷≤′ ✸′ ) are isomorphic
nuclear algebras. The difficulty is in defining ′ , as standard ways of defining
a new binary relation on a quotient do not work in this case. Below we take
a different approach by unwinding instead of collapsing clusters to produce a
partially ordered FM-frame F† . While the nuclear algebra of F† will be “larger”
than that of F, their algebras of fixpoints will be isomorphic.
Proposition 4.5 For any FM-frame F, there is a partially ordered FM-frame
F† such that the algebra of fixpoints in the nuclear algebra of F is isomorphic
to that of F† . Moreover, F is normal iff F† is normal.
Bezhanishvili and Holliday
13
Proof. Given an FM-frame F = (S, ⊑, , F ), define F† = (S † , ⊑† , † , F † ) as
follows (in this proof, ≤ is the usual ordering on N):
•
S † = {hx, ti | x ∈ S, t ∈ N};
•
hx, ti ⊑† hx′ , t′ i iff either [x = x′ and t ≤ t′ ] or [x ⊑ x′ and t < t′ ];
•
hx, ti † hx′ , t′ i iff either [x = x′ and t ≤ t′ ] or [x x′ and t < t′ ];
•
F † = {hx, ti ∈ S † | x ∈ F }.
Observe that ⊑† and † are partial orders. Moreover, F = ∅ iff F † = ∅, so F
is normal iff F† is normal.
Let j be the nucleus associated with F and j † the nucleus associated with
†
F . Let g : S † → S be defined by g(x, t) = x. We claim that the function G
that maps each fixpoint U of j † to G(U ) = g[U ] is an isomorphism between
the algebras of fixpoints in the nuclear algebras of F† and F.
First, toward showing that G sends fixpoints of j † to fixpoints of j, we show
that for every hx, ti ∈ S † and U ∈ Up(F† )F † :
hx, ti ∈ j † U iff x ∈ jg[U ].
(8)
From left to right, if x′ ⊒ x, then hx′ , t + 1i ⊒† hx, ti. Since hx, ti ∈ j † U ,
there is hx′′ , t′′ i † hx′ , t + 1i such that hx′′ , t′′ i ∈ U . This implies x′′ x′
and x′′ ∈ g[U ]. Hence we have shown that x ∈ jg[U ]. From right to left,
if hx′ , t′ i ⊒† hx, ti, then x′ ⊒ x. Since x ∈ jg[U ], there is an x′′ x′ with
x′′ ∈ g[U ]. It follows that there is some s ∈ N such that hx′′ , si ∈ U . Since U ∈
Up(F† )F † , it follows that hx′′ , t′ + s + 1i ∈ U . Given x′′ x′ and t′ + s + 1 > t′ ,
we have hx′′ , t′ + s + 1i † hx′ , t′ i. Therefore, hx, ti ∈ j † U .
Now we can see that if U is a fixpoint of j † , then g[U ] is a fixpoint of j. To
see that jg[U ] ⊆ g[U ], observe that if x ∈ jg[U ], then by (8) and the assumption
that U is a fixpoint of j † , we have hx, ti ∈ j † U ⊆ U , whence x ∈ g[U ].
Second, we claim that G is surjective. Suppose V is a fixpoint of j. Then
we claim that g −1 [V ] is a fixpoint of j † , which with g[g −1 [V ]] = V , given by
the surjectivity of g, will show that G is surjective. We begin by showing that
g −1 [V ] ∈ Up(F† )F † . To see that g −1 [V ] ∈ Up(F† ), observe that if hx, ti ∈
g −1 [V ] and hx′ , t′ i ⊒† hx, ti, then x ∈ V and x′ ⊒ x, which with V ∈ Up(F)
implies x′ ∈ V and hence hx′ , t′ i ∈ g −1 [V ]. To see that g −1 [V ] ∈ Up(F† )F † ,
observe that since g[F † ] = F ⊆ V , we have F † ⊆ g −1 [g[F † ]] ⊆ g −1 [V ]. Next,
we must show that j † g −1 [V ] ⊆ g −1 [V ]. If hx, ti ∈ j † g −1 [V ], then by (8),
x ∈ jg[g −1 [V ]] = jV = V , using that V is a fixpoint of j. Hence hx, ti ∈ g −1 [V ].
Third, we show that G preserves and reflects order: for any fixpoints U and
V of j † , we have U ⊆ V iff g[U ] ⊆ g[V ]. If U ⊆ V and x ∈ g[U ], then for some
t ∈ N, hx, ti ∈ U ⊆ V , so x ∈ g[V ]. Conversely, suppose U 6⊆ V , so there is an
hx, ti ∈ U \ V . Then x ∈ g[U ], and since V = j † V , hx, ti 6∈ j † V , which with (8)
implies x 6∈ jg[V ] and hence x 6∈ g[V ]. So g[U ] 6⊆ g[V ].
Thus, G is an isomorphism between the algebras of fixpoints of F† and F.✷
Turning preordered FM-frames into partially ordered Dragalin frames does
not require the unwinding in the previous proof. In this case we may collapse
14
Locales, Nuclei, and Dragalin Frames
clusters without difficulty, by first turning a preordered FM-frame into a partially ordered nuclear frame as in Remark 4.4 and then turning that nuclear
frame into a Dragalin frame as in Theorem 3.5.
Remark 4.6 In the case of a normal FM-frame F = (S, ≤, , F ), there is an
even more direct approach: we can define a standard Dragalin frame D =
(S ′ , ≤′ , D) where (S ′ , ≤′ ) is the skeleton of (S, ≤) and D([x]) = {⇑y | x ≤ y},
where [x] is the equivalence class of x and ⇑y = {[z] ∈ S ′ | y z}. For lack of
space, we omit the proof that the nuclear algebras of F and D are isomorphic.
It is rather surprising that we can also go in the other direction, from
Dragalin to FM-frames. This we do by “enlarging” the underlying poset.
Theorem 4.7 For any Dragalin frame D, there is an FM-frame F such that
the nuclear algebras of D and F are isomorphic. Moreover, D is normal iff F
is normal.
Proof. By Proposition 3.11, we may assume that the Dragalin frame D is
convex and therefore satisfies (4◦◦ ) by Remark 3.10(iii). First we give the
proof assuming that D is normal and then show how to modify the proof to
lift this assumption. The construction is similar to the construction of intuitionistic relational frames from intuitionistic neighborhood frames in [19] (cf.
the construction of birelational frames from monotonic neighborhood frames in
[20]). Given a normal Dragalin frame D = (S, ≤, D), we define an FM-frame
F = (S ′ , ≤′ , ′ , F ′ ) with F ′ = ∅ as follows:
•
S ′ = {(x, X) | x ∈ S, X ∈ D(x)};
•
(x, X) ≤′ (y, Y ) iff x ≤ y;
•
(x, X) ′ (y, Y ) iff y ∈ X and Y ⊆ X.
Clearly ≤′ is a preorder. To see that ′ is a preorder, X ∈ D(x) implies x ∈ X
by convexity and Remark 3.10(i), so ′ is reflexive. For transitivity, if (x, X) ′
(y, Y ) ′ (z, Z), then Z ⊆ Y ⊆ X, so Z ⊆ X, and z ∈ Y ⊆ X, so z ∈ X.
Hence (x, X) ′ (z, Z). Finally, ′ is a subrelation of ≤′ : if (x, X) ′ (y, Y ),
then by (2◦◦ ), y ∈ X ∈ D(x) implies x ≤ y, so (x, X) ≤′ (y, Y ).
Define f : Up(D) → Up(F) by
f (U ) = {(x, X) | x ∈ U, X ∈ D(x)}.
It is routine to check that f is an isomorphism between Up(D) and Up(F). To
show it is an isomorphism between the nuclear algebras of D and F, we show:
f ([DiU ) = ✷≤′ ✸′ f (U ).
(9)
Suppose (x, X) ∈ f ([DiU ), so x ∈ [DiU and X ∈ D(x). Consider any (y, Y )
such that (x, X) ≤′ (y, Y ), so x ≤ y. Then from x ∈ [DiU we have y ∈ [DiU ,
since [DiU is an upset whenever U is. Given y ∈ [DiU and Y ∈ D(y), there is a
z ∈ Y such that z ∈ U . Given z ∈ Y ∈ D(y), by (4◦◦ ) there is a Z ∈ D(z) such
that Z ⊆ Y . Thus, (y, Y ) ′ (z, Z). Then since z ∈ U , we have (z, Z) ∈ f (U ).
Hence we have shown that (x, X) ∈ ✷≤′ ✸′ f (U ).
Bezhanishvili and Holliday
15
Conversely, suppose (x, X) 6∈ f ([DiU ), so x 6∈ [DiU . Then there is a
Y ∈ D(x) with Y ∩ U = ∅ (but Y 6= ∅ by (1◦ )). Therefore, (x, X) ≤′ (x, Y ),
and for any (z, Z) such that (x, Y ) ′ (z, Z), we have z ∈ Y and hence z 6∈ U ,
so (z, Z) 6∈ f (U ). Thus, x 6∈ ✷≤′ ✸′ f (U ). This completes the proof of (9).
If D = (S, ≤, D) is not normal, define the FM-frame F = (S ′ , ≤′ , ′ , F ′ ) by:
•
S ′ = {(x, X) | x ∈ S, X ∈ D(x)} ∪ {(x, ∅) | D(x) = ∅} ∪ {m}; F ′ = {m};
•
(x, X) ≤′ (y, Y ) iff x ≤ y; and m is the maximum of ≤′ ;
•
(x, X) ′ (y, Y ) iff y ∈ X and ∅ 6= Y ⊆ X;
•
for all (x, ∅) ∈ S ′ , (x, ∅) ′ (x, ∅), (x, ∅) ′ m, and m ′ m.
By (1◦ ), for all x ∈ S, ∅ 6∈ D(x), so adding (x, ∅) to S ′ when D(x) = ∅ does
not cause any ambiguity. Note that ≤′ and ′ are still preorders, and ′ is a
subrelation of ≤′ . Since F is an FM-frame, its nuclear algebra is based on the
locale Up(F)F ′ . Define g : Up(D) → Up(F)F ′ by
g(U ) = {(x, X) | x ∈ U, X ∈ D(x)} ∪ {(x, ∅) | x ∈ U, D(x) = ∅} ∪ {m}.
As in the case of f above, it is routine to check that g is an isomorphism
between Up(D) and Up(F)F ′ . To see that it is an isomorphism between the
nuclear algebras of D and F, the proof of (9) above works for g in place of f
with only small additions. In particular, we must show that (x, ∅) ∈ g([DiU )
iff (x, ∅) ∈ ✷≤′ ✸′ g(U ). In fact, for any (x, ∅) ∈ S ′ , both (x, ∅) ∈ g([DiU )
and (x, ∅) ∈ ✷≤′ ✸′ g(U ). To see the first, since D(x) = ∅, for any U ∈ Up(D)
we have x ∈ [DiU and hence (x, ∅) ∈ g([DiU ). To see the second, consider
any (y, Y ) such that (x, ∅) ≤′ (y, Y ). Then x ≤ y, which with D(x) = ∅ and
(3◦ ) implies D(y) = ∅, so (y, Y ) = (y, ∅). By construction, (y, ∅) ′ m and
✷
m ∈ g(U ). Thus, (x, ∅) ∈ ✷≤′ ✸′ g(U ).
From Theorems 3.5 and 4.7, we obtain the following.
Corollary 4.8 For any nuclear frame F, there is an FM-frame G such that
the nuclear algebras of F and G are isomorphic.
This is weaker than what we had for Dragalin frames in Theorem 3.5, which
showed that we can always go from a nuclear frame to a Dragalin frame based on
the same poset. The following example shows that when going from a nuclear
frame to an FM-frame, changing the underlying poset may be unavoidable.
Example 4.9 If ≤ is identity, so Up(S, ≤) = ℘(S), then = ≤, so ✷≤ ✸ is the
identity nucleus on ℘(S) and any relativization thereof. Yet for any non-trivial
Boolean algebra, there is a nucleus distinct from the identity nucleus.
Putting together Corollary 3.8, Theorem 4.7, and Proposition 4.5, we obtain
the following analogue of Corollary 3.8.
Corollary 4.10 A poset P is a locale iff there is a partially ordered normal
FM-frame F such that P is isomorphic to the algebra of fixpoints in the nuclear
algebra of F.
16
5
Locales, Nuclei, and Dragalin Frames
Goldblatt Frames
A nucleus on a Heyting algebra is a special case of a dual operator on a Heyting
algebra, a unary function that preserves all finite meets (including 1). Following
the tradition in modal logic, we denote such a function by ✷.
Definition 5.1 A modal Heyting algebra is a pair (H, ✷) where H is a Heyting
algebra and ✷ is a dual operator. It is a modal locale if H is a locale.
Typical examples of modal locales come from intuitionistic modal frames [34].
Definition 5.2 An IM-frame (intuitionistic modal frame) is a triple F =
(S, ≤, R) where (S, ≤) is a poset and R ⊆ S 2 is such that ≤ ◦ R ◦ ≤ = R.
The condition that ≤ ◦ R ◦ ≤ = R guarantees that for U ∈ Up(F), we also
have ✷R U ∈ Up(F). It is straightforward to check that F+ := (Up(F), ✷R ) is a
modal locale. A natural question, then, is whether there are conditions on an
IM-frame F that are equivalent to F+ being a nuclear algebra. These conditions
were identified by Goldblatt [13, pp. 500-01]. Recall that a relation R ⊆ S 2 is
dense if whenever xRy, there is a z ∈ S such that xRzRy.
Lemma 5.3 (Goldblatt) Let F = (S, ≤, R) be an IM-frame.
(i) R is a subrelation of ≤ iff U ⊆ ✷R U for each U ∈ Up(F).
(ii) R is dense iff ✷R ✷R U ⊆ ✷R U for each U ∈ Up(F).
Remark 5.4 Goldblatt [13] did not assume the full IM-frame condition that
≤ ◦ R ◦ ≤ = R, but only the weaker condition that R ◦ ≤ = R, which is still
sufficient for ✷R to be a function on Up(S, ≤). 7 Relative to frames satisfying
the weaker condition, the property that ✷R ✷R U ⊆ ✷R U for each U ∈ Up(F)
corresponds to a “pseudo-density” condition on R [13, p. 501], rather than
density. But any frame satisfying the weaker condition can be turned into an
IM-frame such that their associated modal locales are isomorphic (by simply
defining a new relation R′ = ≤ ◦ R ◦ ≤).
Following Goldblatt’s notation, we will use ≺ for a dense subrelation of ≤.
Definition 5.5 A Goldblatt frame is an IM-frame F = (S, ≤, ≺) such that ≺
is a dense subrelation of ≤.
For a Goldblatt frame F = (S, ≤, ≺), the modal operator ✷≺ is a nucleus
by Lemma 5.3, which gives us the following.
Proposition 5.6 (Goldblatt) If F = (S, ≤, ≺) is a Goldblatt frame, then
(Up(F), ✷≺ ) is a nuclear algebra, which we call the nuclear algebra of F.
Remark 5.7 A special case of the frames considered by Goldblatt [13]—with
≺ not only dense, but also serial—appears in the semantics for intuitionistic
epistemic logic in [2,26], which treats ✷≺ as an intuitionistic knowledge modality. Note that in a Goldblatt frame, ✷≺ is a dense nucleus iff ≺ is serial. The
logic IEL+ of [2,26] is exactly the logic of a dense nucleus. This is the extension
7
A still weaker sufficient condition is that R ◦ ≤ ⊆ ≤ ◦ R [7].
Bezhanishvili and Holliday
17
of propositional lax logic with the axiom ¬ ⊥, which Fairtlough and Mendler
[11, Thm. 4.5] prove is the propositional lax logic of normal FM-frames.
We have seen that for every nuclear frame F, there is a Dragalin frame and
an FM-frame whose nuclear algebras are isomorphic to that of F (Theorem 3.5
and Corollary 4.8). Let us now consider for which nuclear frames there is a
Goldblatt frame with an isomorphic nuclear algebra.
First, we define a necessary and sufficient condition for a modal locale (L, ✷)
to be isomorphic to F+ for some IM-frame F (recall Remark 2.7).
Definition 5.8 A modal locale (L, ✷) is perfect
is Alexandroff and ✷ is
V if L V
completely multiplicative: for every X ⊆ L, ✷ X = {✷x | x ∈ X}.
As is well known, if a function on a complete lattice is completely multiplicative, then it admits an adjoint.
Lemma 5.9 Given a modal locale (L, ✷)
V with ✷ completely multiplicative, the
function ✸∗ : L → L defined by ✸∗ a = {x ∈ L | a ≤ ✷x} is a left adjoint of
✷: for all a, b ∈ L, ✸∗ a ≤ b iff a ≤ ✷b.
Lemma 5.9 is used in the proof of the following characterization.
Theorem 5.10 Let (L, ✷) be a modal locale.
(i) (L, ✷) is isomorphic to F+ for an IM-frame F iff (L, ✷) is perfect.
(ii) (L, ✷) is isomorphic to F+ for a Goldblatt frame F iff (L, ✷) is a perfect
nuclear algebra.
Proof. For part (i), for any IM-frame F, it is easy to see that F+ is perfect.
Conversely, suppose (L, ✷) is perfect, so L is Alexandroff. Let S = J ∞ (L) (see
Remark 2.7) and ⊑ be the dual of the restriction of the order ≤ on L to S. It
is well known that the function f : L → Up(S, ⊑) defined by f (a) = {x ∈ S |
a ⊑ x} is an isomorphism between L and Up(S, ⊑) (see, e.g., [8,5]).
If ✷ is a completely multiplicative operator on L, then we define R on S
by xRy iff y ≤ ✸∗ x. To see that F := (S, ⊑, R) is an IM-frame, suppose
x ⊑ yRz ⊑ u. Then u ≤ z, z ≤ ✸∗ y, and y ≤ x. Therefore, u ≤ ✸∗ y
and ✸∗ y ≤ ✸∗ x, yielding u ≤ ✸∗ x, so xRu. To see that f is an isomorphism
between (L, ✷) and F+ , it only remains to show that f (✷a) = ✷R f (a). Suppose
x ∈ f (✷a) and xRy. Then ✷a ⊑ x, so x ≤ ✷a, and y ≤ ✸∗ x, so by Lemma 5.9,
y ≤ ✸∗ x ≤ a and hence a ⊑ y. Thus, y ∈ f (a), which shows x ∈ ✷R f (a).
Conversely, if x ∈
/ f (✷a), then ✷a 6⊑ x, so x 6≤ ✷a and hence ✸∗ x 6≤ a by
Lemma 5.9. Since L is Alexandroff, there is a y ∈ S such that y ≤ ✸∗ x and
y 6≤ a. Hence xRy and a 6⊑ y, so y 6∈ f (a), whence x ∈
/ ✷R f (a).
For part (ii), apply part (i) and Lemma 5.3.
✷
Note that when L is the locale of upsets in a poset P , the poset (J ∞ (L), ⊑)
constructed in the proof of Theorem 5.10 is isomorphic to P . Thus, the following is an immediate corollary of Theorem 5.10.
Corollary 5.11 Given any nuclear frame F = (S, ≤, j) with j completely multiplicative, there is a Goldblatt frame G = (S, ≤, ≺) such that j = ✷≺ .
18
Locales, Nuclei, and Dragalin Frames
Using the above results, we can also characterize the locales that can be
realized as the algebra of fixpoints in the nuclear algebra of a Goldblatt frame.
Theorem 5.12 A locale L is isomorphic to the algebra of fixpoints in the nuclear algebra of a Goldblatt frame iff L is completely distributive.
Proof. From right to left, it suffices to observe that in the proof of Theorem
2.8, if the poset P is a completely distributive locale, then the nucleus j defined
in (3) is completely multiplicative, whence Corollary 5.11 gives us the desired
Goldblatt frame. The proof that j is completely multiplicative is just like the
original proof that j is multiplicative, which used the join-infinite distributive
law for locales, but now we use the completely distributive law.
From left to right, by Theorem 5.10, the nucleus in the nuclear algebra of any
Goldblatt frame is completely multiplicative, and the underlying locale of the
nuclear algebra of any Goldblatt frame is Alexandroff and hence completely
distributive. Thus, it suffices to show that for any localic nuclear algebra
L = (L, j) with L completely distributive and j completely multiplicative, its
algebra of fixpoints Lj is completely distributive. Let {xϕ,ψ | ϕ ∈ Φ, ψ ∈ Ψϕ }
be a doubly indexed family of elements from
, F the set of functions f
V Lj W
assigning
to
each
ϕ
∈
Φ
some
f
(ϕ)
∈
Ψ
,
and
the operations in L, and
ϕ
d
F
and
the operations in Lj . Then we have:
^ _
l G
xϕ,ψ by definition of Lj
j
xϕ,ψ =
ϕ∈Φ ψ∈Ψϕ
ϕ∈Φ ψ∈Ψϕ
=j
^ _
xϕ,ψ by complete multiplicativity of j
ϕ∈Φ ψ∈Ψϕ
=j
_ ^
xϕ,f (ϕ) by complete distributivity of L
f ∈F ϕ∈Φ
=
G l
xϕ,f (ϕ) by definition of Lj ,
f ∈F ϕ∈Φ
so Lj is completely distributive.
✷
Remark 5.13 While every Alexandroff locale is completely distributive, there
are completely distributive non-Alexandroff locales, such as the interval [0, 1].
6
Conclusion
We now have a complete picture of the ability of Dragalin frames, FM-frames,
and Goldblatt frames to represent nuclear algebras and, via their algebras of
fixpoints, to represent locales. This is summarized in the following table: 8
8 We put the dash next to FM because we must typically change the underlying poset
of a nuclear frame in order to find an FM-frame with an isomorphic nuclear algebra (see
Theorem 4.7 and Example 4.9). Recall that for a nuclear frame (S, ≤, j), its associated locale
is Up(S, ≤), whereas for an FM frame (S, ≤, , F ), its associated locale is Up(S, ≤)F .
Bezhanishvili and Holliday
Frames
Dragalin
FM
Goldblatt
Nuclear Frames
all
–
completely
multiplicative j
Nuclear Algebras
Alexandroff
Alexandroff
perfect
19
Locales
all
all
completely
distributive
The frames studied in this paper are not the only frames in the literature for
representing nuclear algebras. Goldblatt [14] introduces localic cover systems
and proves that every locale can be realized as the algebra of fixpoints in the
nuclear algebra of a localic cover system, by showing that the nucleus j as in the
proof of Theorem 2.8 can be generated by one of his cover systems. In future
work [6], we will present a detailed comparison of this “cover” perspective and
the “development” perspective of Dragalin, thereby relating Scott-Montaguestyle neighborhood semantics with Beth-style path semantics.
We will also explain in future work how Dragalin frames (S, ≤, D) for intuitionistic propositional logic extend to modal Dragalin frames (S, ≤, D, R) for
intuitionistic modal logic, where R is a binary relation on S that interacts with
≤ and D in a natural way. This provides an intuitionistic generalization of the
recently studied “possibility semantics” for classical modal logic [17,15,16,3].
Returning to the logical angle with which we began, it is an open problem
whether every intermediate logic is the logic of some class of locales. Given the
results of this paper, we can equivalently rephrase the problem as follows: is
every intermediate logic the logic of some class of Dragalin frames?
Acknowledgement
We wish to thank the three anonymous AiML referees for their valuable comments, Dana Scott for helpful references, and the Department of Mathematical
Sciences at NMSU and the Group in Logic at UC Berkeley for travel support.
References
[1] Alechina, N., M. Mendler, V. de Paiva and E. Ritter, Categorical and Kripke semantics
for constructive S4 modal logic, Lecture Notes in Computer Science 2142 (2001),
pp. 292–307.
[2] Artemov, S. and T. Protopopescu, Intuitionistic epistemic logic (2014),
arXiv:1406.1582v2 [math.LO].
[3] van Benthem, J., N. Bezhanishvili and W. H. Holliday, A bimodal perspective on
possibility semantics, Journal of Logic and Computation (Forthcoming).
[4] Beth, E., Semantic construction of intuitionistic logic, Mededelingen der Koninklijke
Nederlandse Akademie van Wetenschappen 19 (1956), pp. 357–388.
[5] Bezhanishvili, G., Varieties of monadic Heyting algebras. Part II: Duality theory, Studia
Logica 62 (1999), pp. 21–48.
[6] Bezhanishvili, G. and W. H. Holliday, Development frames (2016), in preparation.
[7] Božic, M. and K. Došen, Models for normal intuitionistic modal logics, Studia Logica
43 (1984), pp. 217–245.
[8] Davey, B. A., On the lattice of subvarieties, Houston Journal of Mathematics 5 (1979),
pp. 183–192.
20
Locales, Nuclei, and Dragalin Frames
[9] Dragalin, A. G., “Matematicheskii Intuitsionizm: Vvedenie v Teoriyu Dokazatelstv,”
Matematicheskaya Logika i Osnovaniya Matematiki, “Nauka”, Moscow, 1979.
[10] Dragalin, A. G., “Mathematical Intuitionism: Introduction to Proof Theory,”
Translations of Mathematical Monographs 67, American Mathematical Society,
Providence, RI, 1988.
[11] Fairtlough, M. and M. Mendler, Propositional lax logic, Information and Computation
137 (1997), pp. 1–33.
[12] Fourman, M. P. and D. S. Scott, Sheaves and logic, in: M. P. Fourman, C. J. Mulvey
and D. S. Scott, editors, Applications of Sheaves, Springer, Berlin, 1979 pp. 302–401.
[13] Goldblatt, R., Grothendieck topology as geometric modality , Zeitschrift für
Mathematische Logik und Grundlagen der Mathematik 27 (1981), pp. 495–529.
[14] Goldblatt, R., Cover semantics for quantified lax logic, Journal of Logic and
Computation 21 (2011), pp. 1035–1063.
[15] Holliday, W. H., Partiality and adjointness in modal logic, in: R. Goré, B. Kooi and
A. Kurucz, editors, Advances in Modal Logic, Vol. 10, College Publications, London,
2014 pp. 313–332.
[16] Holliday, W. H., Possibility frames and forcing for modal logic (2015), UC Berkeley
Working Paper in Logic and the Methodology of Science.
URL http://escholarship.org/uc/item/5462j5b6
[17] Humberstone, L., From worlds to possibilities, Journal of Philosophical Logic 10 (1981),
pp. 313–339.
[18] Johnstone, P. T., “Stone Spaces,” Cambridge Studies in Advanced Mathematics 3,
Cambridge University Press, Cambridge, 1982.
[19] Kojima, K., Relational and neighborhood semantics for intuitionistic modal logic,
Reports on Mathematical Logic 47 (2012), pp. 87–113.
[20] Kracht, M. and F. Wolter, Normal monomodal logics can simulate all others, Journal of
Symbolic Logic 64 (1999), pp. 99–138.
[21] Kripke, S. A., Semantical analysis of intuitionistic logic I, in: J. N. Crossley and M. A. E.
Dummett, editors, Formal Systems and Recursive Functions, North-Holland Publishing
Company, Amsterdam, 1965 pp. 92–130.
[22] Kuznetsov, A. V., On superintuitionistic logics, in: Proceedings of the International
Congress of Mathematicians (Vancouver, B. C., 1974), Vol. 1 (1975), pp. 243–249.
[23] Litak, T., A continuum of incomplete intermediate logics, Reports on Mathematical
Logic 36 (2002), pp. 131–141.
[24] Macnab, D. S., “An Algebraic Study of Modal Operators on Heyting Algebras with
Applications to Topology and Sheafification,” Ph.D. thesis, University of Aberdeen
(1976).
[25] Macnab, D. S., Modal operators on Heyting algebras, Algebra Universalis 12 (1981),
pp. 5–29.
[26] Protopopescu, T., Intuitionistic epistemology and modal logics of verification, in:
W. van der Hoek, W. H. Holliday and W. Wang, editors, Logic, Rationality, and
Interaction, Springer, Berlin, 2015 pp. 295–307.
[27] Restall, G., “An Introduction to Substructural Logics,” Routledge, New York, 2000.
[28] Rumfitt, I., On a neglected path to intuitionism, Topoi 31 (2012), pp. 101–109.
[29] Rumfitt, I., “The Boundary Stones of Thought: An Essay in the Philosophy of Logic,”
Oxford University Press, Oxford, 2015.
[30] Shehtman, V. B., Incomplete propositional logics, Doklady Akademii Nauk SSSR 235
(1977), pp. 542–545 (Russian).
[31] Simmons, H., A framework for topology, in: A. Macintyre, L. Pacholski and J. Paris,
editors, Logic Colloquium ’77, North-Holland, Amsterdam, 1978 pp. 239–251.
[32] Takeuti, G. and W. M. Zaring, “Axiomatic Set Theory,” Springer-Verlag, New York,
1973.
[33] Tarski, A., Der Aussagenkalkuül und die Topologie, Fundamenta Mathematicae 31
(1938), pp. 103–134.
[34] Wolter, F. and M. Zakharyaschev, The relation between intuitionistic and classical modal
logics, Algebra and Logic 36 (1997), pp. 73–92.