arXiv:submit/5729843 [math.LO] 13 Jul 2024
Preconditionals
W ESLEY H. H OLLIDAY1
Abstract: In recent work, we introduced a new semantics for conditionals,
covering a large class of what we call preconditionals. In this paper, we
undertake an axiomatic study of preconditionals and subclasses of preconditionals. We then prove that any bounded lattice equipped with a preconditional can be represented by a relational structure, suitably topologized,
yielding a single relational semantics for conditional logics normally treated
by different semantics, as well as generalizing beyond those semantics.
Keywords: conditionals, Heyting algebras, ortholattices, orthomodular lattices, Sasaki hook, indicatives, counterfactuals, flattening, relational frames
1 Introduction
Conditionals in their different flavors—material, strict, indicative, counterfactual, probabilistic, constructive, quantum, etc.—have long been of central interest in philosophical logic (see Egré and Rott 2021 and references
therein). In this paper, based on a talk at Logica 2023, we further investigate a new approach to conditionals introduced in our recent work on the
representation of lattices with conditional operations (Holliday 2023, § 6).
We define a preconditional → on a bounded lattice to be a binary operation satisfying five natural axioms, which we show to be independent
(Section 2.1). We also consider the properties of the associated negation
defined by ¬a = a → 0 (Section 2.2). Familiar examples of bounded lattices equipped with a preconditional include Heyting algebras (Section 2.3),
ortholattices with the Sasaki hook (Section 2.4), and Lewis-Stalnaker-style
conditional algebras satisfying the so-called flattening axiom (Section 2.5).
We characterize these classes axiomatically in terms of additional independent axioms they satisfy beyond those of preconditionals.
We then show (Section 3) that every bounded lattice equipped with a preconditional can be represented using a relational structure (X, ⊳), suitably
topologized. This yields a single relational semantics for conditional logics
1 Forthcoming in The Logica Yearbook 2023, ed. Igor Sedlár, College Publications. Thanks
to Yifeng Ding, Matt Mandelkern, Guillaume Massas, and Snow Zhang for helpful comments.
1
Wesley H. Holliday
normally treated by different semantics, as well as a generalization beyond
those semantics. We conclude (Section 4) with some suggested directions
for further development of this approach to conditionals.
2 Preconditionals
2.1 The axioms and their independence
The definition of a preconditional from Holliday 2023 was discovered through
an attempt to axiomatize the class of lattices with an implication operation
amenable to a relational representation described in Section 3. However,
here we will begin with axiomatics and turn to representation only at the end.
Definition 1 Given a bounded lattice L, a preconditional on L is a binary
operation → on L satisfying the following for all a, b, c ∈ L:
1. 1 → a ≤ a;
2. a ∧ b ≤ a → b;
3. a → b ≤ a → (a ∧ b);
4. a → (b ∧ c) ≤ a → b;
5. a → ((a ∧ b) → c) ≤ (a ∧ b) → c.
Arguably all of the axioms are intuitively valid for both indicative conditionals and counterfactual conditionals in natural language, but we will not
make that case here. Instead, let us begin with the following easy check.
Fact 1
The axioms of preconditionals are mutually independent.
Proof. For each axiom, we provide a lattice with a binary operation → in
which the axiom does not hold but it is easy to check that the other axioms do.
For axiom 1, consider the two-element lattice on {0, 1} with → defined
by a → b = 1. Since 1 → 0 = 1 6≤ 0, axiom 1 does not hold.
For axiom 2, consider the lattice on {0, 1} with → defined by a → b = 0.
Since 1 ∧ 1 = 1 6≤ 0 = 1 → 1, axiom 2 does not hold.
For axiom 3, consider the lattice on {0, 1} with → defined by a → b = b.
Since 0 → 1 = 1 6≤ 0 = 0 → 0 = 0 → (0 ∧ 1), axiom 3 does not hold.
For axiom 4, consider the lattice with → on the left of Figure 1. Since
0 → (1 ∧ 0) = 0 → 0 = 1 6≤ 1/2 = 0 → 1, axiom 4 does not hold.
Finally, for axiom 5, consider the lattice with → on the right of Figure 1.
Since 1/2 → ((1/2 ∧ 0) → 0) = 1/2 → (0 → 0) = 1/2 → 0 = 1 6≤ 0 =
0 → 0 = (1/2 ∧ 0) → 0, axiom 5 does not hold.
2
Preconditionals
1
→
0
1/2
1
1/2
0
1
0
0
1/2
1/2
1/2
1/2
1
1/2
1/2
1
→
0
1/2
1
0
0
1
0
1/2
0
1
1/2
1
0
1
1
0
Figure 1: Left: lattice L and → demonstrating independence of axiom 4.
Right: → demonstrating independence of axiom 5.
2.2 Precomplementation
The preconditional axioms settle some basic properties of the negation operation defined from → by ¬x := x → 0.
Proposition 1 Let L be a bounded lattice with a preconditional →. Then
defining ¬x := x → 0, we have:
1. a ≤ b implies ¬b ≤ ¬a;
2. ¬1 = 0.
Proof. For part 1, if a ≤ b, then we have
b→0
≤ b → ((b ∧ a) → 0) by axiom 4 of preconditionals
≤ (b ∧ a) → 0 by axiom 5 of preconditionals
≤ a→0
since b ∧ a = a from a ≤ b.
Part 2 is immediate from axiom 1 of preconditionals.
Following Holliday 2023, we call a unary operation ¬ satisfying parts 1
and 2 of Proposition 1 a precomplementation. Given a precomplementation,
we can induce a preconditional as follows—an idea to which we will return
in the context of ortholattices in Section 2.4.
Proposition 2 Let L be a bounded lattice equipped with a precomplementation ¬. Then the binary operation → defined by
a → b := ¬a ∨ (a ∧ b)
is a preconditional.
3
Wesley H. Holliday
Proof. Using the definition of →, the axioms of preconditionals become:
1. ¬1 ∨ (1 ∧ a) ≤ a;
2. a ∧ b ≤ ¬a ∨ (a ∧ b);
3. ¬a ∨ (a ∧ b) ≤ ¬a ∨ (a ∧ (a ∧ b));
4. ¬a ∨ (a ∧ (b ∧ c)) ≤ ¬a ∨ (a ∧ b);
5. ¬a ∨ (a ∧ (¬(a ∧ b) ∨ ((a ∧ b) ∧ c))) ≤ ¬(a ∧ b) ∨ ((a ∧ b) ∧ c).
Axiom 1 holds given the assumption that ¬1 = 0. Axioms 2-4 follow from
the axioms for lattices. Axiom 5 holds given the assumption that a ∧ b ≤ a
implies ¬a ≤ ¬(a ∧ b).
2.3 Heyting implication
As suggested in Section 1, several familiar conditional operations are examples of preconditionals. Our first example is the Heyting implication in
Heyting algebras. Consider the following axioms:
• modus ponens (MP): a ∧ (a → b) ≤ b;
• weak monotonicity: b ≤ a → b.
Fact 2
1. Modus ponens is independent of the axioms of preconditionals plus
weak monotonicity.
2. Weak monotonicity is independent of the axioms of preconditionals
plus modus ponens.
Proof. For modus ponens, consider the lattice with → in Figure 2. We have
b ∧ (b → a) = b ∧ 1 = b 6≤ a, so modus ponens does not hold, but one can
check that the other axioms do.
For weak monotonicity, consider the two-element lattice on {0, 1} with
→ defined by a → b = a ∧ b. Since 1 6≤ 0 = 0 → 1, weak monotonicity
does not hold, but one can check that the other axioms do.
We can characterize Heyting implications as preconditionals satisfying
the two axioms above.
4
Preconditionals
1
b
a
0
→
0
a
b
1
0
1
0
0
0
a
1
1
1
a
b
1
1
1
b
1
1
1
1
1
Figure 2: (L, →) demonstrating independence of modus ponens in Fact 2.1.
Proposition 3 For any bounded lattice L and binary operation → on L,
the following are equivalent:
1. → is a Heyting implication, i.e., for all a, b, c ∈ L,
a ∧ b ≤ c iff a ≤ b → c;
2. → is a preconditional satisfying modus ponens and weak monotonicity;
3. → satisfies axioms 3 and 4 of preconditionals, modus ponens, and
weak monotonicity.
Proof. The implication from 1 to 2 is straightforward and the implication
from 2 to 3 is immediate.
From 3 to 1, supposing a ≤ b → c, we have a ∧ b ≤ (b → c) ∧ b ≤ c by
modus ponens. Conversely, supposing a ∧ b ≤ c, we have
a
≤
≤
b → a by weak monotonicity
b → (a ∧ b) by axiom 3 of preconditionals
≤
≤
b → (a ∧ b ∧ c) by our assumption that a ∧ b ≤ c
b → c by axiom 4 of preconditionals.
Fact 3 Axioms 3 and 4 of preconditionals, modus ponens, and weak monotonicity are mutually independent.
Proof. For axiom 3, we can again use the two-element lattice on {0, 1} with
→ defined by a → b = b, as in the proof of Fact 1.
For axiom 4, consider the three-element lattice on {0, 1/2, 1} with →
defined as follows: x → y = 1 if x = y and otherwise x → y = y. Then
0 → 0 = 1 6≤ 1/2 = 0 → 1/2, so axiom 4 does not hold. However, one
can check that axiom 3, modus ponens, and weak monotonicity hold.
5
Wesley H. Holliday
For modus ponens, consider again the two-element lattice on {0, 1} with
→ defined by a → b = 1. Since 1 ∧ (1 → 0) = 1 ∧ 1 = 1 6≤ 0, modus
ponens does not hold, but the other axioms clearly do.
Finally, for weak monotonicity, consider again the two-element lattice
on {0, 1} with → defined by a → b = 0. Since 1 6≤ 0 = 1 → 1, weak
monotonicity does not hold, but the other axioms clearly do.
A natural weakening of modus ponens to consider is that the derived ¬
operation is a semicomplementation: a ∧ (a → 0) = 0. Let us say that a
proto-Heyting implication is a preconditional satisfying weak monotonicity
and semicomplementation. The implication used in the proof of Fact 2.1 is
a proto-Heyting implication that is not a Heyting implication.
2.4 Sasaki hook
For our second example, an ortholattice is a bounded lattice L equipped
with a unary operation ¬, called an orthocomplementation, satisfying
• antitonicity: a ≤ b implies ¬b ≤ ¬a,
• semicomplementation: a ∧ ¬a = 0, and
• involution: ¬¬a = a.
From these properties, one can derive excluded middle (a ∨ ¬a = 1)2 and
De Morgan’s laws (¬(a ∨ b) = ¬a ∧ ¬b and ¬(a ∧ b) = ¬a ∨ ¬b).
In an ortholattice, the Sasaki hook is the binary operation defined by
s
a → b := ¬a ∨ (a ∧ b) = ¬(a ∧ ¬(a ∧ b)).
The following is immediate from Proposition 2.
Corollary 1 In any ortholattice, the Sasaki hook is a preconditional.
Next we add axioms on a preconditional → to characterize the Sasaki
hook. First, note that one half of the equation a → b = ¬a ∨ (a ∧ b), where
¬ is now defined from →, already follows from the preconditional axioms.
Lemma 1
For any preconditional → on a bounded lattice, we have
(a → 0) ∨ (a ∧ b) ≤ a → b.
2 Since
a ≤ a ∨ ¬a and ¬a ≤ a ∨ ¬a, we have ¬(a ∨ ¬a) ≤ ¬a ∧ ¬¬a = 0, so
¬0 ≤ ¬¬(a ∨ ¬a) = a ∨ ¬a. Finally, 1 ≤ ¬¬1 and ¬1 = 1 ∧ ¬1 = 0, so 1 ≤ ¬0, which
with the previous sentence implies 1 ≤ a ∨ ¬a.
6
Preconditionals
Proof. We have a → 0 ≤ a → b by axiom 4 of preconditionals and a ∧ b ≤
a → b by axiom 2 of preconditionals, so (a → 0) ∨ (a ∧ b) ≤ a → b.
To prove the reverse inequality, we assume that the negation defined by
→ is an involutive semicomplementation.
Proposition 4 For any bounded lattice L and binary operation → on L,
the following are equivalent:
1. → is a preconditional with a ∧ (a → 0) = 0 and (a → 0) → 0 = a;
2. L equipped with ¬ defined by ¬a := a → 0 is an ortholattice and →
coincides with the Sasaki hook: a → b = ¬a ∨ (a ∧ b).
Proof. From 2 to 1, that → is a preconditional follows from Corollary 1.
That a ∧ (a → 0) = 0 and (a → 0) → 0 = a follows from the definition of
¬ and the assumption that ¬ is an orthocomplementation.
From 1 to 2, first we show that ¬ is an orthocomplementation. Both
a ∧ ¬a = 0 and ¬¬a = a follow from our assumptions on → and the
definition of ¬. That a ≤ b implies ¬b ≤ ¬a is given by Proposition 1.1.
Finally, we show a → b = ¬a ∨ (a ∧ b). The right-to-left inclusion is
given by Lemma 1, so it only remains to show a → b ≤ ¬a ∨ (a ∧ b):
x ∧ ¬y ≤ ¬y
⇒ ¬¬y ≤ ¬(x ∧ ¬y) by antitonicity for ¬
⇒ y ≤ ¬(x ∧ ¬y) by involution for ¬
⇒ x → y ≤ x → ¬(x ∧ ¬y) by axiom 4 of preconditionals
⇒ x → y ≤ x → ((x ∧ ¬y) → 0) by definition of ¬
⇒ x → y ≤ (x ∧ ¬y) → 0 by axiom 5 of preconditionals
⇒ x → y ≤ ¬(x ∧ ¬y) by definition of ¬
⇒ x → y ≤ ¬x ∨ y by De Morgan’s law and involution for ¬
⇒ a → (a ∧ b) ≤ ¬a ∨ (a ∧ b) substituting a for x, a ∧ b for y
⇒ a → b ≤ ¬a ∨ (a ∧ b) by axiom 3 of preconditionals.
An ortholattice is orthomodular if a ≤ b implies b = a ∨ (¬a ∧ b). In
fact, as observed by Mittelstaedt (1972), orthomodularity is equivalent to
the Sasaki hook satisfying modus ponens.
Lemma 2 (Mittelstaedt) An ortholattice L is orthomodular if and only if
a ∧ (¬a ∨ (a ∧ b)) ≤ b for all a, b ∈ L.
7
Wesley H. Holliday
Combining Lemma 2 with Proposition 4 yields the following.
Proposition 5 For any bounded lattice L and binary operation → on L,
the following are equivalent:
1. → is a preconditional satisfying modus ponens and (a → 0) → 0 =
a;
2. L equipped with ¬ defined by ¬a := a → 0 is an orthomodular lattice
and → coincides with the Sasaki hook: a → b = ¬a ∨ (a ∧ b).
Figure 3 summarizes the relations between the classes of preconditionals
covered so far (OL and OML stand for ortho- and orthomodular lattices).
We also add the classical material implication of Boolean algebras, which is
equivalent to Heyting implication with involution of ¬ and to Sasaki hook in
orthomodular lattices with weak monotonicity (by Proposition 3, Lemma 2).
classical material implication
(a → 0) → 0 = a
b≤a→b
Heyting implication
Sasaki hook in OML
(a → 0) → 0 = a
b≤a→b
a ∧ (a → b) ≤ b
a ∧ (a → b) ≤ b
proto-Heyting implication
preconditionals with MP
Sasaki hook in OL
a ∧ (a → b) ≤ b
(a → 0) → 0 = a
b≤a→b
preconditionals with semicomp
a ∧ (a → 0) = 0
preconditionals
Figure 3: classes of preconditionals.
8
Preconditionals
2.5 Lewis-Stalnaker-style conditionals
The third example of lattices with preconditionals that we will consider are
Boolean algebras with Lewis-Stalnaker-style conditionals (Stalnaker 1968,
Lewis 1973) satisfying the axiom of flattening (Mandelkern in press, §§ 6.4.16.4.2, citing Cian Dorr, p.c.):
a → ((a ∧ b) → c) = (a ∧ b) → c.
Axiom 5 of preconditionals is simply the left-to-right inclusion.
Lewis-Stalnaker-style (set-)selection function semantics in effect treats
a conditional a → b as the result of applying an a-indexed normal modal
operator a to b. That is, there is a binary relation Ra between worlds, and
a → b is true at w iff all Ra -accessible worlds from w make b true. Further
constraints are imposed so that the relations Ra can be derived from wellfounded preorderings of the set of worlds: wRa v iff v is one of the closest
a-worlds to w according to the well-founded preorder 6w associated with
w. But for our purposes here, the key aspects of the Lewis-Stalnaker (set)selection function semantics are captured by the following definition.
Definition 2 A selection frame is a pair (W, {RA }A⊆W ) where W is a
nonempty set and each RA is a binary relation on W satisfying the following
for all w, v ∈ W and A ⊆ W :
1. success: if wRA v, then v ∈ A;
2. centering: if w ∈ A, then wRA v iff v = w.
Such a frame is functional if it satisfies the following:
3. if wRA v and wRA u, then v = u.
Such a frame is strongly dense if it satisfies the following:
4. if wRA∩B v, then ∃u: wRA u and uRA∩B v.
Strong density says that instead of conditioning on a stronger proposition, one can first condition on a weaker one, then condition on the stronger
one, and end up in the same state as one would reach by conditioning on
the stronger proposition straightaway. Though not all Lewis-Stalnaker-style
frames that satisfy success, centering, and functionality are strongly dense,
the following are, as observed by Boylan and Mandelkern (2022).
9
Wesley H. Holliday
Example 1 Given a well-ordered set (W, 6), for any w ∈ W and A ⊆ W ,
let wRA v iff v is the first world in A according to 6 such that w 6 v. Then
(W, {RA }A⊆W ) is a strongly dense, functional selection frame.
Proposition 6 For any strongly dense selection frame (W, {RA }A⊆W ),
the operation →R defined by
A →R B := A B = {w ∈ W | for all v ∈ W, wRA v ⇒ v ∈ B}
is a preconditional on the Boolean algebra ℘(W ).
Proof. We must check the following for → = →R :
1. W → A ⊆ A; 2. A ∩ B ⊆ A → B; 3. A → B ⊆ A → (A ∩ B);
4. A → (B ∩ C) ⊆ A → B; 5. A → ((A ∩ B) → C) ⊆ (A ∩ B) → C.
Condition 1 follows from centering (in particular, the right-to-left direction of the biconditional in centering), as does condition 2 (but now the
left-to-right direction of the biconditional in centering); condition 3 follows from success; and condition 4 is immediate from the definition of
→. For condition 5, suppose w ∈ A → ((A ∩ B) → C) and wRA∩B v.
Then by strong density, there is a u such that wRA u and uRA∩B v. Since
w ∈ A → ((A ∩ B) → C) and wRA u, we have u ∈ (A ∩ B) → C, which
with uRA∩B v yields v ∈ C. This shows that w ∈ (A ∩ B) → C.
The key principles validated by selection frames beyond the axioms of
preconditionals are modus ponens,
• identity: a → a = 1, and
• normality: (a → b) ∧ (a → c) ≤ a → (b ∧ c).
Functional frames also validate
• negation import: ¬(a → b) ≤ a → ¬b.
Proposition 7 Let B be a finite Boolean algebra equipped with a preconditional → satisfying modus ponens, identity, normality, and negation import.
c the isomorphism from B to ℘(W ).
Let W be the set of atoms of B and (·)
For a ∈ B, define
wRba v iff for all b ∈ B, w ≤ a → b implies v ≤ b.
Then (W, {RA }A⊆W ) is a strongly dense, functional selection frame, and
(B, →) is isomorphic to (℘(W ), →R ).
10
Preconditionals
Proof. First, we check the following:
1. success: if wRba v, then v ≤ a;
2. centering: if w ≤ a, then wRba v iff v = w;
3. functionality: if wRba v and wRba u, then v = u;
4. strong density: if wR[
v, then ∃u: wRba u and uR[
v.
a∧b
a∧b
For success, given w ≤ a → a from identity, wRba v implies v ≤ a.
For centering, assume w ≤ a. Modus ponens for → yields wRba w. Then
the rest of centering follows given functionality, which we prove next.
For functionality, if wRba v, then we claim w ≤ a → v. For if w 6≤ a → v,
then since w is an atom, we have w ≤ ¬(a → v) and hence w ≤ a → ¬v
by negation import, contradicting wRba v. Then since w ≤ a → v, if wRba u,
then u ≤ v, which implies u = v given that v is an atom.
For strong density, assume wR[
v. Let
a∧b
^
x = {y ∈ B | w ≤ a → y}.
We claim that x 6= 0. Otherwise there are y1 , . . . , yn such that
w ≤ (a → y1 ) ∧ · · · ∧ (a → yn ) and y1 ∧ · · · ∧ yn = 0.
But then by normality and axioms 4-5 of preconditionals,
w ≤ a → (y1 ∧ · · · ∧ yn ) = a → 0 ≤ (a ∧ b) → 0,
contradicting wR[
v. Hence x 6= 0, so there is an atom u ≤ x, and by
a∧b
construction of x, wRba u. To show uR[
v, suppose u ≤ (a ∧ b) → c.
a∧b
Then w ≤ a → ((a ∧ b) → c), for otherwise w ≤ a → ¬((a ∧ b) → c)
using negation import, in which case u ≤ ¬((a ∧ b) → c) by construction
of x, which contradicts u ≤ (a ∧ b) → c given modus ponens. Then since
w ≤ a → ((a ∧ b) → c), we have w ≤ (a ∧ b) → c by axiom 5 of
preconditionals, which with wR[
v implies v ≤ c. Thus, uR[
v.
a∧b
a∧b
Finally, the proof that a\
→b=b
a →R bb is just like the usual proof for a
normal modal box.
This kind of result can be generalized
V beyond finite algebras (e.g.,
V to complete
and atomic algebras, assuming {a → bi | i ∈ I} ≤ a → {bi | i ∈ I})
and beyond Boolean algebras, but we will not do so here.
Finally, let us return to the Heyting and Sasaki examples of Sections 2.3
and 2.4, respectively, with identity, normality, and negation import in mind.
11
Wesley H. Holliday
Proposition 8
1. Heyting implications satisfy identity, normality, and negation import.
2. Proto-Heyting implications satisfy identity and negation import, but
not necessarily normality.3
3. In orthomodular lattices, Sasaki hook satisfies normality, but not necessarily negation import.
4. In ortholattices, Sasaki hook satisfies identity but not necessarily normality.4
Proof. Part 1 is standard. For part 2 and identity, by weak monotonicity and
axiom 3 of preconditionals, 1 ≤ a → 1 ≤ a → (a ∧ 1) = a → a. For
negation import, by weak monotonicity, b ≤ a → b, so ¬(a → b) ≤ ¬b
by Lemma 1.1, and ¬b ≤ a → ¬b by weak monotonicity again, so indeed
¬(a → b) ≤ a → ¬b. For a proto-Heyting implication that does not satisfy
normality, consider the lattice on the left of Figure 4, and define → such that
for any elements x and y: 1 → x = x; x → 0 = 0 if x 6= 0; 0 → 0 = 1;
x → y = 1 if x 6= 1, y 6= 0, and (x, y) 6= (a, d); and a → d = d. Then
(a → b) ∧ (a → c) = 1 ∧ 1 = 1 but a → (b ∧ c) = a → d = d, so normality
does not hold, but one can check that the other axioms hold.
For part 3, in an orthomodular lattice, ¬x ≤ y implies y ≤ ¬x ∨ (x ∧ y).
Then since ¬a ≤ (a → b) ∧ (a → c), we have
(a → b) ∧ (a → c) ≤
≤
=
¬a ∨ (a ∧ (a → b) ∧ (a → c))
¬a ∨ (a ∧ b ∧ c) by Lemma 2
a → (b ∧ c)
by definition of Sasaki hook.
To see that negation import is not necessarily satisfied, consider the modular
lattice M4 with elements {0, a, b, c, d, 1} such that a, b, c, d are incomparable in the lattice order, ¬a = b, and ¬c = d. Then for the Sasaki hook we
have ¬(a → c) = ¬(¬a ∨ (a ∧ c)) = ¬(¬a ∨ 0) = ¬¬a = a, whereas
a → ¬c = ¬a ∨ (a ∧ ¬c) = ¬a ∨ (a ∧ d) = ¬a ∨ 0 = ¬a.
For part 4, identity for Sasaki hook is just excluded middle. For a failure
of normality, consider the ortholattice in Figure 4. Then for the Sasaki hook,
3 Moreover, there are normal proto-Heyting implications that are not Heyting implications,
such as the implication used in the proof of Fact 2.1.
4 Moreover, there are non-orthomodular lattices in which the Sasaki hook satisfies normality,
such as the lattice O6 (the “benzene ring”).
12
Preconditionals
1
1
a
a
c
b
d
¬a
c
b
0
¬b
¬c
0
Figure 4: Left: lattice for the proof of Proposition 8.2. Right: an ortholattice
in which the Sasaki hook violates normality for Proposition 8.4.
(a → b)∧(a → c) = (¬a∨(a∧b))∧(¬a∨(a∧c)) = (¬a∨b)∧(¬a∨c) =
1 ∧ 1 = 1, whereas a → (b ∧ c) = ¬a ∨ (a ∧ b ∧ c) = ¬a ∨ 0 = ¬a.
Though the conditionals in Proposition 8 are normal if they satisfy modus
ponens, this is not the case for preconditionals in general.5 An instructive
example comes from the following probabilistic interpretation. Given W =
{0, . . . , 10}, define for each w ∈ W a measure µw : ℘(WP
) → [0, 1] by
µw ({w}) = .9, µw ({v}) = .01 for v 6= w, and µw (A) = v∈A µw ({v})
for non-singleton A ⊆ W . Then for A, B ⊆ W with A 6= ∅, let
A → B = {w ∈ W : µw (B | A) ≥ .9},
where as usual µw (B | A) = µw (A ∩ B)/µw (A), and ∅ → B = W .
Proposition 9 The operation → just defined is a preconditional on ℘(W )
satisfying modus ponens but not normality.
Proof. First observe that A ∩ (A → B) ⊆ B, because if w ∈ A, then
since µw ({w}) = .9, we can have µw (B | A) ≥ .9 only if w ∈ B.
This also shows that axiom 1 of preconditionals holds. For axiom 2, if
w ∈ A ∩ B, then again since µw ({w}) = .9, we have µw (B | A) ≥ .9, so
5 Note that preconditionals satisfying normality but not modus ponens can easily be obtained
from strongly dense selection frames that do not satisfy centering, as well as from the examples
in Footnotes 3 and 4.
13
Wesley H. Holliday
w ∈ A → B. Axioms 3 and 4 also clearly hold. For axiom 5, suppose
w ∈ A → ((A ∩ B) → C), so µw ((A ∩ B) → C | A) ≥ .9. If w ∈ A,
then by modus ponens, we have w ∈ (A ∩ B) → C, as desired. So suppose
w 6∈ A. Further suppose for contradiction that w 6∈ (A ∩ B) → C, i.e.,
µw (A ∩ B ∩ C)/µw (A ∩ B) < .9, so |A ∩ B ∩ C| < |A ∩ B|. We claim
that (A ∩ B) → C ⊆ A ∩ B. Consider any x ∈ W \ (A ∩ B) with x 6= w.
Since x 6∈ A ∩ B, w 6∈ A ∩ B, x 6= w, and |W | = 11, we have |A ∩ B| ≤ 9,
which with |A ∩ B ∩ C| < |A ∩ B| implies µx (C ∩ A ∩ B)/µx (A ∩ B) ≤
.08/.09 < .9, so x 6∈ (A ∩ B) → C. Thus, (A ∩ B) → C ⊆ A ∩ B,
which implies (A ∩ B) → C ⊆ A ∩ B ∩ C by modus ponens, which with
µw ((A ∩ B) → C | A) ≥ .9 implies µw (A ∩ B ∩ C | A) ≥ .9, which in
turn implies µw (C | A ∩ B) ≥ .9 and hence w ∈ (A ∩ B) → C.
Finally, for the failure of normality, 0 ∈ {1, . . . , 10} → {1, . . . , 9} and
0 ∈ {1, . . . , 10} → {2, . . . , 10}, but 0 6∈ {1, . . . , 10} → {2, . . . , 9}.
3 Relational representation of lattices with preconditionals
Having hopefully shown the interest of the class of preconditionals from an
axiomatic perspective, let us return to the semantic origin of preconditionals.
Given a set X, a binary relation ⊳ on X, and A, B ⊆ X, define6
A →⊳ B = {x ∈ X | ∀y ⊳ x (y ∈ A ⇒ ∃z ⊲ y : z ∈ A ∩ B)}.
(1)
As shown in Holliday 2023, the operation c⊳ defined by c⊳ (A) = X →⊳ A
is a closure operator, so its fixpoints ordered by inclusion
W form a complete
lattice
L(X,
⊳)
with
meet
as
intersection
and
join
as
⊳ {Ai | i ∈ I} =
S
c⊳ ( {Ai | i ∈ I}). If A, B ∈ L(X, ⊳), then A →⊳ B ∈ L(X, ⊳),7 so we
may regard →⊳ as an operation on L(X, ⊳). The following is easy to check.
Fact 4
The operation →⊳ on L(X, ⊳) is a preconditional.
Example 2 In the relational frame shown at the top of Figure 5, the arrow
from y to x indicates x ⊳ y, etc. Reflexive loops are assumed but not shown.
6 In Holliday 2023, we denoted the operation defined in (1) by ‘։ ’ in order to distinguish
⊳
it from a different operation denoted by ‘→⊳ ’. Since we do not need that distinction here, we
will use the cleaner ‘→⊳ ’ for the operation defined in (1).
7 Obviously A → B ⊆ X → (A → B). For the reverse, suppose x 6∈ A → B.
⊳
⊳
⊳
⊳
Hence there is some y ⊳ x such that y ∈ A and for all z ⊲ y, we have z 6∈ A ∩ B. It follows
that there is no w ⊲ y with w ∈ A →⊳ B. Thus, if x 6∈ A →⊳ B, then there is a y ⊳ x
such that for all w ⊲ y, w 6∈ A →⊳ B, which shows that x 6∈ X →⊳ (A →⊳ B).
14
Preconditionals
Transitive arrows are not assumed. Note that y ∈ G but y 6∈ B →⊳ G,
contra weak monotonicity. Also note that y ∈ (P →⊳ 0) →⊳ 0 = 0 →⊳
0 = 1, but y 6∈ P , contra involution. So →⊳ is neither Heyting nor Sasaki.
y
w
x
z
G
B
P
R
O
R = {x}
O = {z}
G = {x, y}
B = {w, z}
P = {x, z}
→⊳
0
R
O
G
P
B
1
0
1
B
G
O
0
R
0
R
1
1
G
P
G
R
R
O
1
B
1
O
B
P
O
G
1
1
G
1
G
R
G
P
1
1
1
P
1
P
P
B
1
B
1
O
B
1
B
1
1
1
1
1
1
1
1
Figure 5: A relational frame (top) giving rise to a lattice with preconditional
(bottom).
In fact, every complete lattice with a preconditional can be represented
by such a frame (X, ⊳). More generally, we have the following.
Theorem 1 (Holliday 2023, Theorem 6.3) Let L be a bounded lattice and
→ a preconditional on L. Then where
P = {(x, x → y) | x, y ∈ L} and (a, b) ⊳ (c, d) if c 6≤ b,
there is a complete embedding of (L, →) into (L(P, ⊳), →⊳ ), which is an
isomorphism if L is complete.
Let us take this a step further with a topological representation. Important precedents for the non-conditional aspects of this representation can be
found in Urquhart 1978, Allwein and Hartonas 1993, Ploščica 1995, and
Craig, Haviar, and Priestley 2013. Given a bounded lattice L and a preconditional →, define FI(L, →) = (X, ⊳) as follows: X is the set of all pairs
(F, I) such that F is a filter in L, I is an ideal in L, and for all a, b ∈ L:
if a ∈ F and a ∧ b ∈ I, then a → b ∈ I.
15
Wesley H. Holliday
Call such an (F, I) a consonant filter-ideal pair. We define (F, I) ⊳ (F ′ , I ′ )
if I ∩ F ′ = ∅. Finally, given a ∈ L, let b
a = {(F, I) ∈ X | a ∈ F }, and let
S(L, →) be FI(L, →) endowed with the topology generated by {b
a | a ∈ L}
(cf. Bezhanishvili and Holliday 2020).
Theorem 2 For any bounded lattice L and preconditional → on L, the
map a 7→ b
a is
1. an embedding of (L, →) into (L(FI(L, →)), →⊳ ) and
2. an isomorphism from L to the subalgebra of (L(FI(L, →)), →⊳ ) consisting of elements of L(FI(L, →)) that are compact open in the space
S(L, →).
Proof. For a ∈ L, let ↑a (resp. ↓a) be the principal filter (resp. ideal) generated by a. First we claim that for any a, b ∈ L, (↑a, ↓a → b) ∈ X. For
suppose c ∈ ↑a and c ∧ d ∈ ↓a → b, so a ≤ c and c ∧ d ≤ a → b. Then
c→d ≤
≤
c → (c ∧ d) by axiom 3 of preconditionals
c → (a → b) by axiom 4, since c ∧ d ≤ a → b
=
≤
c → ((a ∧ c) → b) since a ≤ c
(a ∧ c) → b by axiom 5
=
a → b since a ≤ c,
so c → d ∈ ↓a → b. Since b = 1 → b by axioms 1 and 2 of preconditionals,
it follows that (↑1, ↓b) = (↑1, ↓1 → b) ∈ X as well.
Using the above facts, the proof that b
a belongs to L(FI(L, →)) and that
a 7→ b
a is injective and preserves ∧ and ∨ is the same as in the proof of
Theorem 4.30.1 in Holliday 2023. Also note that b
1 = X and b
0 = c⊳ (∅).
b
\
→ b,
Next we show that a → b = b
a →⊳ b. First suppose (F, I) ∈ a\
′ ′
′ ′
(F , I ) ⊳ (F, I), and (F , I ) ∈ b
a, so a ∈ F ′ . Since (F, I) ∈ a\
→ b, we
have a → b ∈ F , which with (F ′ , I ′ ) ⊳ (F, I) implies a → b 6∈ I ′ , which
with a ∈ F ′ and the definition of X implies a∧b 6∈ I ′ . Now let F ′′ = ↑a∧b
and I ′′ = ↓(a ∧ b) → 0. Then (F ′′ , I ′′ ) ∈ X, (F ′ , I ′ ) ⊳ (F ′′ , I ′′ ), and
→ b,
(F ′′ , I ′′ ) ∈ a[
∧ b. Thus, (F, I) ∈ b
a →⊳ bb. Conversely, if (F, I) 6∈ a\
′ ′
so a → b 6∈ F , then setting (F , I ) = (↑a, ↓a → b), we have (F ′ , I ′ ) ∈ X
and (F ′ , I ′ ) ⊳ (F, I). Now consider any (F ′′ , I ′′ ) such that (F ′ , I ′ ) ⊳
(F ′′ , I ′′ ), so a → b 6∈ F ′′ . Then by axiom 2 of preconditionals, a ∧ b 6∈ F ′′ ,
so (F ′′ , I ′′ ) 6∈ a[
∧ b. Thus, (F, I) 6∈ b
a →⊳ bb.
16
Preconditionals
The proof of part 2 is the same as the proof of Theorem 4.30.2 in Holliday
2023.
Finally, we can characterize the spaces equipped with a relation ⊳ that
are isomorphic to S(L, →) for some (L, →). Let X be a topological space
and ⊳ a binary relation on X. Let COFix(X, ⊳) be the set of all compact
open sets of X that are also fixpoints of c⊳ . If COFix(X, ⊳) is a lattice with
meet as ∩ and join as ∨⊳ , F is a filter in this lattice, and I is an ideal, then
we can speak of (F, I) being a consonant filter-ideal pair as defined above
Theorem 2, using →⊳ in the definition. Given x ∈ X, let
F(x) = {U ∈ COFix(X, ⊳) | x ∈ U }
I(x) = {U ∈ COFix(X, ⊳) | ∀y ⊲ x y 6∈ U }
and note that (F(x), I(x)) is a consonant filter-ideal pair from COFix(X, ⊳).
For if U ∈ F(x) and U ∩ V ∈ I(x), then x ∈ U but for all y ⊲ x, y 6∈ U ∩ V ,
which implies that for all y ⊲ x, y 6∈ U →⊳ V , so U →⊳ V ∈ I(x).
Proposition 10 For any space X and binary relation ⊳ on X, there is
a bounded lattice L with preconditional → such that (X, ⊳) and S(L, →)
are homeomorphic as spaces and isomorphic as relational frames iff the
following conditions hold for all x, y ∈ X:
1. x = y iff (F(x), I(x)) = (F(y), I(y));
2. COFix(X, ⊳) contains X and c⊳ (∅), is closed under ∩, ∨⊳ , and
→⊳ , and forms a basis for X;
3. each consonant filter-ideal pair from COFix(X, ⊳) is (F(x), I(x)) for
some x ∈ X;
4. x ⊳ y iff I(x) ∩ F(y) = ∅.
The proof is very similar to that of Proposition 3.21 in Holliday 2022,
but we provide the adapted proof here for convenience.
Proof. Suppose there is such an L. It suffices to show S(L, →) satisfies conditions 1–4 in place of (X, ⊳). That condition 2 holds for COFix(S(L, →))
and S(L, →) follows from the proof of Theorem 2. Let ϕ be the isomorphism a 7→ b
a from L to COFix(S(L, →)) in Theorem 2, which induces a
17
Wesley H. Holliday
bijection (F, I) 7→ (ϕ[F ], ϕ[I]) between consonant filter-ideal pairs of L
and of COFix(S(L, →)). Conditions 1, 3, and 4 follow from the fact that
for any x = (F, I) ∈ S(L, →), (ϕ[F ], ϕ[I]) = (F(x), I(x)).
(2)
First, b
a ∈ ϕ[F ] iff a ∈ F iff x ∈ b
a iff b
a ∈ F(x). Second, b
a ∈ ϕ[I] iff
a ∈ I, and we claim that a ∈ I iff b
a ∈ I(x), i.e., for all (F ′ , I ′ ) ⊲ (F, I),
(F ′ , I ′ ) 6∈ b
a, i.e., a 6∈ F ′ . If a ∈ I and (F, I) ⊳ (F ′ , I ′ ), then a 6∈ F ′
by definition of ⊳. Conversely, if a 6∈ I, let F ′ = ↑a and I ′ = ↓a → 0,
so (F ′ , I ′ ) is consonant by the proof of Theorem 2, (F, I) ⊳ (F ′ , I ′ ) since
a 6∈ I, and a ∈ F ′ . Thus, b
a 6∈ I(x). This completes the proof of (2).
Now for condition 1, given x, y ∈ S(L, →) with x = (F, I) and y =
(F ′ , I ′ ), we have (F, I) = (F ′ , I ′ ) iff (ϕ[F ], ϕ[I]) = (ϕ[F ′ ], ϕ[I ′ ]) iff
(F(x), I(x)) = (F(y), I(y)); similarly, for condition 4, (F, I) ⊳ (F ′ , I ′ ) iff
I ∩F ′ = ∅ iff ϕ[I]∩ϕ[F ′ ] = ∅ iff I(x)∩F(y) = ∅. Finally, for condition 3,
if (F , I) if a consonant filter-ideal pair from COFix(S(L, →)), then setting
x = (ϕ−1 [F ], ϕ−1 [I]), we have x ∈ S(L, →) and (F , I) = (F(x), I(x)).
Assuming X satisfies the conditions, COFix(X, ⊳) is a bounded lattice
with a preconditional by Fact 4, and we define a map ǫ from (X, ⊳) to
S(COFix(X, ⊳), →⊳ ) by ǫ(x) = (F(x), I(x)). The proof that ǫ is a homeomorphism using conditions 1–3 is analogous to the proof of Theorem 5.4(2)
in Bezhanishvili and Holliday 2020. That ǫ preserves and reflects ⊳ follows
from condition 4.
4 Conclusion
We have seen that preconditionals encompass several familiar classes of
conditionals, including Heyting implication, the Sasaki hook, and LewisStalnaker style conditionals satisfying flattening. Lattices with these implications are therefore covered by the general representation in Theorem 2. A
natural next step is to try to obtain nice characterizations of the relationaltopological duals of these kinds of algebras, as well as of more novel kinds—
such as lattices with normal preconditionals—not to mention going beyond
representation to categorical duality. We hope that the delineation of preconditionals and their relational-topological representation may help to provide
a unified view of a vast landscape of conditionals arising in logic.
18
Preconditionals
References
Allwein, G., & Hartonas, C. (1993). Duality for bounded lattices. (Indiana
University Logic Group, Preprint Series, IULG-93-25 (1993))
Bezhanishvili, N., & Holliday, W. H. (2020). Choice-free Stone duality.
The Journal of Symbolic Logic, 85(1), 109-148.
Boylan, D., & Mandelkern, M.
(2022).
Logic and information sensitivity.
(ESSLLI 2022 lecture notes,
https://mandelkern.hosting.nyu.edu/ESSLLI2022Day3.pdf)
Craig, A. P. K., Haviar, M., & Priestley, H. A. (2013). A fresh perspective
on canonical extensions for bounded lattices. Applied Categorical
Structures, 21, 725-749.
Egré, P., & Rott, H.
(2021).
The Logic of Conditionals.
https://plato.stanford.edu/archives/win2021/entries/logicconditionals/.
(In The Stanford Encyclopedia of Philosophy
(Winter 2021 Edition))
Holliday, W. H. (2022). Compatibility and accessibility: lattice representations for semantics of non-classical and modal logics. In D. F. Duque
& A. Palmigiano (Eds.), Advances in Modal Logic, Vol. 14. London:
College Publications.
Holliday, W. H. (2023). A fundamental non-classical logic. Logics, 1,
36-79.
Lewis, D. (1973). Counterfactuals. Oxford: Basil Blackwell.
Mandelkern, M. (in press). Bounded Meaning: The Dynamics of Interpretation. Oxford: Oxford University Press.
Mittelstaedt, P. (1972). On the interpretation of the lattice of subspaces
of Hilbert space as a propositional calculus. Zeitschrift für Naturforschung, 27a, 1358–1362.
Ploščica, M. (1995). A natural representation of bounded lattices. Tatra
Mountains Mathematical Publication, 5, 75-88.
Stalnaker, R. C. (1968). A theory of conditionals. In N. Rescher (Ed.),
Studies in Logical Theory (pp. 98–112). Oxford: Blackwell.
Urquhart, A. (1978). A topological representation theory for lattices. Algebra Universalis, 8, 45-58.
Wesley H. Holliday
University of California, Berkeley
USA
E-mail: wesholliday@berkeley.edu
19