arXiv:2102.06067v1 [math.LO] 11 Feb 2021
CO-QUANTALE VALUED LOGICS
DAVID REYES AND PEDRO H. ZAMBRANO
A BSTRACT. In this paper, we propose a generalization of Continuous Logic ([BBHU08]) where the distances take values in suitable co-quantales (in the way as it was proposed in [Fla97]). By
assuming suitable conditions (e.g., being co-divisible, co-Girard
and a V-domain), we provide, as test questions, a proof of a version of the Tarski-Vaught test (Proposition 3.35) and Łoś Theorem
(Theorem 3.62) in our setting.
Keywords. metric structures, lattice valued logics, co-quantales,
co-Girard, co-divisibility, domains, Tarski-Vaught test, Łoś theorem.
AMS classification 2010. 03C95, 03B60, 03C90, 03C20, 06F07, 18B35
1. I NTRODUCTION
S. Shelah and J. Stern proved in [SS78] that a first order attempt
of study of classes of Banach Spaces has a “bad” behavior (this has
a very high Hanf number, having a model-theoretical behavior similar to a second order logic of binary relations). This led to develop
a suitable logic beyond first order logic, in order to do a suitable
model-theoretic analysis of Banach Spaces.
In [CK66], C. C. Chang and H. J. Keisler proposed a new logic with
truth values within a compact Hausdorff topological space, which
was the first time where the term Continuous Logic appeared. They
developed basics on Mathematical Logic in this book, but by then
Model Theory had not been very developed (Morley’s first order
categoricity theorem had just been proved, and there was no stability theory by then). For some reason, people did not continue
working on this kind of logic until it was rediscovered in the 90’s
by W. Henson and J. Iovino (see [HI02, Iov99]) and later by I. BenYaacov et al (see [BBHU08]), but in the particular case by taking the
Date: February 12, 2021.
The first author wants to thank the second author for the time devoted to advise
his undergraduate thesis, where this paper is one of its fruits. The second author
wants to thank Universidad Nacional de Colombia for the grant “Convocatoria
para el apoyo a proyectos de investigación y creación artśtica de la sede Bogotá de
la Universidad Nacional de Colombia - 2019”.
1
2
D. REYES AND P. ZAMBRANO
truth values in the unitary interval [0, 1], focusing on the study of
structures based on complete metric spaces (e.g. Hilbert spaces together with bounded operators -see [AB09]-, Banach spaces, Probability spaces -see [BH04]-). This logic is known as Continuous Logic
Because of some technical reasons, people working on Continuous
Logic have to consider strong assumptions on the involved operators (e.g. boundness) in order to axiomatize classes of metric structures in this logic. This took us to the notion of Metric Abstract Elementary Class (see [HH09, Zam11]) for being able to deal with nonaxiomatizable -in Continuous Logic- classes of complete metric structures. However, this approach does not consider topological spaces
in general.
Independently, Lawvere provided in [Law73] a framework in Category Theory for being able to consider a logic with generalized
truth values in order to study metric spaces from this point of view.
However, there is no a deep model-theoretic study in this paper.
There is an attempt of a study of first order Model Theory for
Topological Spaces (see [FZ80]), but it was just suitable to study
particular algebraic examples like Modules and Topological Groups,
due to the algebraic nature of first order logic. Moreover, this approach was left as a model-theoretic study of general Topological
Spaces but it was the beginning of the model-theoretic study of Modules (see [Pre88]).
Quantales are a suitable kind of lattices introduced for being able
to deal with locales (a kind of lattices which generalizes the ideal of
open sets of a topological space) and multiplicative lattices of ideals
from Ring Theory and Functional Analysis (e.g., C∗ -algebras and
von Neumann algebras). Considering the contravariant notions in
a quantale (which we will called co-quantales), Flagg gave in [Fla97]
a way to deal with topological spaces as pseudo metric spaces where
the distance takes values in a suitable quantale.
In this paper, we will propose a generalization of Continuous Logic
by defining distances with values in value co-quantales together with
suitable assumptions (e.g., being co-divisible, co-Girard and a Vdomain).
This paper is organized as follows: In the second section we will
provide basic facts in co-quantales. In the third section, we introduced our approach to co-quantales valued logics, analogously as it
is done in Continuous Logic but by considering distances with values in a suitable co-quantale. As test questions, we provide a proof
CO-QUANTALE VALUED LOGICS
3
of a version of the Tarski-Vaught test (Proposition 3.35) and a version of Łoś Theorem (Theorem 3.62). A difference between our approach and Continuous Logic lies on the fact that we can provided
a version of Łoś Theorem for D-products (before doing the quotient
to force a D-product being an actual metric space -which is called
a D-ultraproduct in Continuous Logic-). We notice that the same
proof for D-products works for D-ultraproducts in our setting. As
consequences of Łoś Theorem, in a similar way as in first order and
Continuous logics, we provide a proof of a version of Compactness
Theorem (Corollary 3.64) and of the existence of ω1 -saturated models (Proposition 3.69).
2. VALUE
CO - QUANTALES AND
C ONTINUITY S PACES
In this section, we provide some basics on value co-quantales.
2.1. Valued lattices.
Definition 2.1. Given L a complete lattice and x, y ∈ L, we say that x
is
V co-well below y (denoted by x ≺ y), if and only, if for all A ⊆ X, if
A ≤ x then there exists a ∈ A such that a ≤ y.
Remark 2.2. In [0, ∞], x ≺ y agrees with x < y.
Lemma 2.3. ([Fla97]; Lemma 1.2) Let L be a complete lattice, then for all
x, y, z ∈ L we have that
(1) y ≺ x implies y ≤ x
(2) z ≤ y and y ≺ x imply z ≺ x
(3) y ≺ x and x ≤ z imply y ≺ z.
Lemma 2.4. ([Fla97]; Lemma V
1.3) If L is a complete lattice, then for all
A ⊆ L and x ∈ L we have that A ≺ x if and only if there exists a ∈ A
such that a ≺ x.
DefinitionV
2.5. A complete lattice L is said to be completely distributive if a = {b : a ≺ b} provided that a ∈ L.
Lemma 2.6. ([Fla97]; Lema 1.6]) Given L a completely distributive lattice
and x, y ∈ L provided that x ≺ y, there exists z ∈ L such that x ≺ z y
z ≺ y.
Definition 2.7. A value lattice is a completely distributive lattice L provided that
(1) 0 ≺ 1
(2) if δ, δ′ ∈ L satisfy 0 ≺ δ and 0 ≺ δ′ , then 0 ≺ δ ∧ δ′ .
4
D. REYES AND P. ZAMBRANO
Examples 2.8.
(1) The 2-valued Boolean algebra 2 := {0, 1}, where
0 < 1.
(2) The ordinal number ω + 1 := {0, 1, ..., ω} together with the usual
ordering.
(3) The unit interval I = [0, 1] with the usual real ordering.
(4) ([0, 1], ≥) = ([0, 1], ≤)op (which we will denote by E).
(5) ([0, ∞], ≤usual).
(6) ([FK97]; pg 115-117) Given a set R, let us denote Pfin (R) = {X ∈
P(X) : X is finite} and for all X ∈ Pfin (R) we denote ↓ (X) = {Y ∈
P(X) : Y ⊆ X}.
Given Ω(R) = {p ∈ P(Pfin (R)) : X ∈ p implies ↓ (X) ⊆ p}, then
(Ω(R), ⊇) is a valued lattice.
2.2. Value co-quantales. In this paper, we do not work with the
usual notion of quantale. We consider the contravariant notion (which
we call co-quantale) because this approach allows us to work with a
notion of distance (pseudo-metric) with values in the co-quantale
(see [Fla97]), in an analogous way as the metric structures given in
Continuous Logic. We know that this is not the standard way to
study quantales, but we chose this setting in order to do a similar
study as it is done in Continuous Logic.
Definition 2.9. A co-quantale V is a complete lattice provided with a
commutative monoid structure (V, +) such that
(1) The minimum element 0 of V is the identity of (V, +); i.e., a+0 = a
for all a ∈ V and
V
V
(2) for all a ∈ V and (bi )i∈I ∈ V I , a + i∈I bi = i∈I (a + bi )
Proposition 2.10. ([Fla97]; Pg 6]) If V is a co-quantale, then for all a, b, c ∈
V we have that
(1) a + 1 = 1
(2) a ≤ b implies c + a ≤ c + b
Proposition 2.11. V
([Fla97]; Thrm 2.2]) Given V a co-quantale and a, b ∈
· b := {r ∈ V : r + b ≥ a}. Therefore, for all c ∈ V we have
V, define a −
that
· b ≤ c if and only if a ≤ b + c
(1) a −
· b) + b
(2) a ≤ (a −
· b≤a
(3) (a + b) −
·
(4) a − b = 0 if and only if a ≤ b
· (b + c) = (a −
· b) −
· c = (a −
· c) −
· b
(5) a −
·
·
·
(6) a − c ≤ (a − b) + (b − c)
CO-QUANTALE VALUED LOGICS
5
· is the left adjoint of + and it preserves categorical limits,
Since −
we have the following fact.
Fact 2.12. Let V be a co-quantale,
then for any
W
W sequence (bi )i∈I and any
· a = i∈I (bi −
· a)
element a ∈ V we have that ( i∈I bi ) −
Lemma 2.13. Given a co-quantale V, for any a, b ∈ V, a ≤ b implies
· a≥ c−
· b and a −
· c≤b−
· c.
that for all c we have that c −
Proof. Let c ∈ V, then by Proposition 2.11 (2) we may say that c ≤
· a) + a. So, c ≤ (c −
· a) + b whenever a ≤ b (by Proposition 2.10
(c −
· b ≤ c−
· a.
(2)), and so by Proposition 2.11 (1) this is equivalent to c −
On the other hand, since a ≤ b then
· c) + c (Proposition 2.11 (2))
b ≤ (b −
· c) + c ( since a ≤ b)
a ≤ (b −
· c ≤ b−
· c (Proposition 2.11 (1))
a−
Proposition 2.14. Given
a ∈ V and a sequence (bi )i∈I
V V a co-quantale,
W
·
·
in V, we have that a − i∈I bi = i∈I (a − bi ).
V
W
· i∈I bi ≤ i∈I (a −
· bi ) implies that
Proof.WBy Proposition
a−
V 2.11 (1), V
W
·
·
a ≤ i∈I (a − bi ) + i∈I bi = i∈I (( i∈I (a − bi )) + bi ), which holds
by Proposition 2.11 (2)
W and Proposition 2.10 (2) allow us to say that
·
· bi )) + bj for any j ∈ J.
a ≤ (a − bj ) + bj ≤ ( i∈I (a −
V
W
· bi )) +
Since j ∈ I was taken arbitrarily, then a ≤ i∈I (( i∈I (a −
bi ).
Definition 2.15. A (V, +, 0) co-quantale is said to be a value co-quantale
if V is a value lattice.
Definition 2.16. Given V a value co-quantale, let set V+ = {ǫ ∈ V : 0 ≺
ǫ}, and we call it the positives filter of V.
Lemma 2.17. ([Fla97]; Thrm 2.9]) If (V, +, 0) is a value co-quantale,
given ǫ ∈ V+ there exists δ ∈ V+ such that δ + δ ≺ ǫ
By an obvious inductive argument, we can prove the following
fact.
Corollary 2.18 ([LRZ18]; Remark 2.26). Given any ǫ ∈ V+ and n ∈
n times
z }| {
+
N \ {0}, there exists θ ∈ V such that nθ := θ + · · · + θ ≺ ǫ.
Fact 2.19. ([Fla97];
V Thrm 2.10]) GivenVa value co-quantale V and p ∈ V,
p = p + 0 = p + {ǫ ∈ V : 0 ≺ ǫ} = {p + ǫ : 0 ≺ ǫ}.
6
D. REYES AND P. ZAMBRANO
Lemma 2.20. ([Fla97]; Thrm 2.11]) If V is a value co-quantale and p, q ∈
V such that q ≺ p, then there exist r, ǫ ∈ V such that 0 ≺ ǫ, q ≺ r and
r+ǫ ≺p
Example 2.21. The 2-valued Boolen algebra 2 = {0, 1} is a value coquantale, by taking + := ∨
Example 2.22. ([0, ∞], ≤, +) is a value co-quantale. Notice that the oper· a (a ∈ [0, ∞]) is given by b −
· a := max{0, b − a}. We denote
ation − −
this example by D.
Example 2.23. There are several ways to define a value co-quantale with
underlying value lattice E. One of them is by taking + as the usual real
product, another one consists by taking + := ∨. Also, the Łukasiewicz’s
addition +L provides a value co-quantale structure with underlying lattice
E -where a +L b = ∧{0, a + b − 1} by taking ∧ in ([0, 1], ≥)-. Denote the
previous examples by E∗ , E∨ and EL respectively.
Fact 2.24. ([FK97]; pg 115-117) Given a non empty set R, (Ω(R), ⊇, ∩) is
a value co-quantale. In particular, if (X, τ) is a topological space, (Ω(τ), ⊇
, ∩) is a value co-quantale. This last example is called the free local associate to (X, τ).
Definition 2.25. A co-quantale (V, ≤, +) is said to be co-divisible if for
all a, b ∈ V, a ≤ b implies that there exists c ∈ V such that b = a + c.
Lemma 2.26. A co-quantale (V, ≤, +) is co-divisible,if and only, if for all
· a).
a, b ∈ V, a ≤ b implies b = a + (b −
· a ⊣ a + −.
Proof. This follow from − −
The following property allows us to approximate 0 by means of a
N-indexed sequence.
Definition 2.27. ([LRZ18]) Given a co-quantale V, we say that it has the
Sequential Approximation From Above property (shortly, SAFA), if
and only, if there is a sequence (un )n∈N such that
V
(1) n∈N un = 0.
(2) for all n ∈ N, 0 ≺ un
(3) for all n ∈ N, un+1 ≤ un
2.2.1. Co-Girard value co-quantales. In this section, we give the basic
notions and results relative to co-Girard value co-quantales, which
allows to consider a kind of pseudo complement relative to some
· This notion will allow
fixed element b (a dualizing element) and −.
to prove our test questions (Proposition 3.35 and Theorem 3.62).
CO-QUANTALE VALUED LOGICS
7
Definition 2.28. Given a co-quantale (V, +), an element d ∈ V is said
to be a dualizing element, if and only if, for all a ∈ V we have that
· (d −
· a).
a=d−
Definition 2.29. A co-quantale V is said to be co-Girard, if and only, if
it has a dualizing element.
2.3. Continuity spaces. In this section, we will provide some basics on continuity spaces, a framework given in [Fla97, FK97] in order
to generalize (pseudo) metric spaces but considering distances with
values in general co-quantales.
Definition 2.30. Given X 6= ∅ a set, V a value co-quantale and a mapping
d : X × X → V, the pair (X, d) is said to be a V-continuity space, if and
only if,
(1) (reflexivity) for all x ∈ X, d(x, x) = 0, and
(2) (transitivity) for any x, y, z ∈ X, d(x, y) ≤ d(x, z) + d(z, y).
If there is no confusion about which co-quantale V we are considering,
we may say that (X, d) is just a continuity space.
Example 2.31. Let V := (2, , ≤, ∨) and X 6= ∅ be a set. A 2-continuity
space (X, d) codifies a binary relation R := {(x, y) ∈ X × X : d(x, y) = 0}
on X which is reflexive and transitive.
Example 2.32. Given D := ([0, ∞], ≤, +), a D-continuity space is just
a pseudo metric space (where the distance betwwen two elements might be
infinite).
Example 2.33. Given (X, τ) a topological space and a, b ∈ X, define
d(a, b) := {A ⊆finite τ : for all U ∈ A, a ∈ U implies b ∈ U}. (X, d) is a
Ω(τ)-continuity space.
Example 2.34. Given a value co-quantale V, define d : V × V → V by
· a. (V, d) is a V-continuity space.
d(a, b) := b −
Proposition 2.35. ([FK97]; Pg 119]) Given a value co-quantale V and
(X, dX), (Y, dY ) V-continuity spaces, define
dX×Y : (X × Y) × (X × Y) → V
by dX×Y ((x1 , y1), (x2 , y2 )) := dX (x1 , x2 ) ∨ dY (y1 , y2 ), then (X × Y, dX×Y )
is a V-continuity space.
Remark 2.36. There is another way to provide to a value co-quantale a V· b) ∨ (b −
· a) for
continuity space structure by defining dsV (a, b) := (a −
s
all a, b ∈ V. dV is called the symmetric distance for V. Notice that for
8
D. REYES AND P. ZAMBRANO
all a ∈ V we have that dsV (a, 0) = dsV (0, a) = a:
· 0) ∨ (0 −
· a)
dsV (a, 0) = (a −
· 0) ∨ 0 (by Proposition 2.11 (4) and 0 = min V)
= (a −
· 0
= a−
^
·
=
{r : r + 0 ≥ a} (by definition of −)
^
=
{r : r ≥ a}
= a
2.3.1. The topology of a V-continuity space. In this subsection, we will
give some basics on the underlying topology of a V-continuity space.
Note 2.37. Throughout this subsection, V will denote a value co-quantale.
Definition 2.38. Given a V-continuity space (X, d), ǫ ∈ V+ and x ∈ X
define Bǫ (x) := {y ∈ X : d(x, y) ≺ ǫ} (which we call the disc with radius
ǫ centered in x.
Definition 2.39. A subset U of a V-continuity space (X, d) is said to be
open, if and only if, given x ∈ U there exists some ǫ ∈ V+ such that
Bǫ (x) ⊆ U.
Fact 2.40. ([Fla97]; Thrm 4.2) The family of open subsets in a V-continuity
space (X, d) is closed under finite interesections and arbitrary unions. Also,
∅ and X are open sets.
Definition 2.41. Given (X, d) a V-continuity space, the family of open
sets of (X, d) determines a topology on X, which we will call the topology
induced by d and we will denote it by τd .
Lemma 2.42. Given (X, d) a V-continuity space, for all x ∈ X and ǫ ∈ V+
we have that the disc Bǫ (x) is an open set of X.
V
Proof. Let y ∈ Bǫ (x), so d(x, y) ≺ ǫ. By Fact 2.19 d(x, y) = {d(x, y)+
δ : 0 ≺ δ} ≺ ǫ, then by Lemma 2.4 there exists δ ∈ V+ such that
d(x, y) + δ ≺ ǫ. We may assure that Bδ (y) ⊆ Bǫ (x): If z ∈ X satisfies
d(y, z) ≺ δ, then d(x, z) ≤ d(x, y) + d(y, z) ≤ d(x, y) + δ ≺ ǫ.
Fact 2.43. Given a V-continuity space (X, d), the family of open discs
forms a base for τd .
Definition 2.44. V
Given a V-continuity space (X, d), A ⊆ X and x ∈ X,
define d(x, A) := {d(x, a) : a ∈ A}.
Proposition 2.45. Given a V-continuity space (X, d), then a subset A ⊆
X is τd -closed, if and only if, for all x ∈ X we have that d(x, A) = 0 implies
x ∈ A.
CO-QUANTALE VALUED LOGICS
9
Proof. Suppose that A ⊆ X is τd -closed and let x ∈ X be such that
d(x, A) = 0. In case that x ∈
/ A, since A is τd -closed, there exists
+
c
ǫ ∈ V such that Bǫ (x) ⊆ A = {y ∈ X : y ∈
/ A}. Since d(x, A) :=
V
{d(x, a) : a ∈ A} = 0 ≺ ǫ, by Fact 2.4 there exists a ∈ A such that
d(x, a) ≺ ε, so a ∈ Bǫ (x) ∩ A (contradiction).
On the other hand, let A ⊆ X be such that for all x ∈ X, d(x, A) = 0
implies x ∈ A. Suppose that y ∈ X belongs to the adherence of A, so
by Fact 2.43 given any ǫ ∈ V+ , we have that
V A∩Bǫ (y) 6= ∅. Therefore,
+
for any ε ∈VV we have that d(y, A) := {d(y, a) : a ∈ A} ≤ ε, so
d(y, A) ≤ {ε : 0 ≺ ε} = 0, hence by hypothesis we may say that
y ∈ A. Therefore, A is closed.
Corollary 2.46. Given a V-continuity space (X, d), the topological closure
of A ⊆ X is given by cl(A) := A = {y ∈ X : d(y, A) = 0}.
Definition 2.47. Given a V-continuity space (X, dX), ǫ ∈ V+ and x ∈ X,
define the closed disc of radius ε centered in x by Cǫ (x) := {y ∈ X :
dX (x, y) ≤ ǫ}.
Fact 2.48. ([FK97]; Lemma 3.2 (2)) Let (X, dX ) be a V-continuity space
and x ∈ X. The family {Cǫ (x) : ǫ ∈ V+ } determines a fundamental system
of neighborhoods around x.
Definition 2.49. Given a V-continuity space (X, d), define the dual distance d⋆ : X × X → V relative to d by d⋆ (x, y) := d(y, x). In general,
if we add ⋆ as a superscript to any topological notion, it means that it is
related to the distance d⋆ ; e.g., the topology induced by d⋆ is denoted by τ⋆d .
Proposition 2.50. Given a V-continuity space (X, d), x ∈ X and ǫ ∈ V+,
C⋆ǫ (x) is τd -closed.
Proof. Let x ∈ X and ǫ ∈ V+ , so by Proposition
2.45 it is enough to
V
⋆
check that for any y ∈ X, d(y, Cǫ (x)) := {d(y, a) : a ∈ C⋆ǫ (x)} = 0
implies that y ∈ C⋆ǫ (x). Let y ∈ X be such that d(y, C⋆ǫ (x)) = 0 and
δ ∈ V+ . Since d(y, C⋆ǫ (x)) = 0 ≺ δ, by Fact 2.4 there exists z ∈ C⋆ǫ (x)
such that d(y, z) ≺ δ, so d⋆ (x, y) := d(y, x) ≤ Vd(y, z) + d(z, x) =
d(y,V
z) + d⋆ (x, z) ≤ δ + ǫ, therefore d⋆ (x, y) ≤ {ǫ + δ : 0 ≺ δ} =
ǫ + {δ : 0 ≺ δ} = ǫ + 0 = ǫ, therefore y ∈ C⋆ǫ (x).
Definition 2.51. Given a V-continuity space (X, d), define the symmetric space relative to (X, d) by (X, ds ), where ds (x, y) := d(x, y)∨d⋆ (x, y).
In general, we will denote the topological notions related to ds by adding
the superscript s.
Proposition 2.52. ([FK97]; Lemma 3) Given a V-continuity space (X, d),
for the topology τs induced by ds we have that U ⊆ X belongs to τs , if and
only if, there exist V, W ⊆ X such that V ∈ τd , W ∈ τ⋆ and U = V ∩ W.
10
D. REYES AND P. ZAMBRANO
Lemma 2.53. ([FK97]; Lemma 3,4) If (X, d) is a V-continuity space, (X, τs )
satisfies the following separation properties:
(1) (pseudo-Hausdorff) For all x, y ∈ X, if x ∈
/ {y} according to (X, τd )
then there exist U, V ⊆ X such that x ∈ U, y ∈ V, U ∈ τd , V ∈ τ⋆
and U ∩ V = ∅.
(2) (regularity) For all x ∈ X and A ⊆ X, if A ∈ τd and x ∈ A then
there exist U, C ⊆ X such that U is τd -open, C es τ⋆ -closed and
x ∈ U ⊆ C ⊆ A.
2.3.2. V-domains. In order to give a version of Łoś Theorem in our
setting, following [CK66, BBHU08], we need to consider compact
and Hausdorff topological spaces. The setting which involves these
assumptions in continuity spaces corresponds to V-domains.
Definition 2.54. A V-continuity space (X, d) is said to be T0 , if and only
if, for any x, y ∈ X, d(x, y) = 0 and d(y, x) = 0 implies x = y.
Remark 2.55 ([FK97], pg 120). A continuity space (X, d) is T0 , if and
only if, (X, τd ) is T0 as a topological space and (X, τsd) is Hausdorff.
Definition 2.56. A V-continuity space (X, d) is said to be a V − domain,
if and only if, it is T0 and (X, τsd ) is compact.
Remark 2.57. Let (X, d) be a V-domain, therefore by definition (X, τsd ) is
compact. Since (X, d) is T0 , by Remark 2.55 (X, τsd ) is Hausdorff.
The importance of the previous properties lies on the fact that
these allow us to provide a proof of a version of Łos’s Theorem in
the logic that we will introduce in this paper (Theorem 3.62). We
will provide some examples which satisfy these properties.
Proposition 2.58. ([FK97]; Thrm 4.14) The following examples are domains:
(1) 2 = ({0, 1}, 0 ≤ 1, ∨).
(2) ([0, 1], ≤, +).
(3) The quantale of errors ([0, 1], ≥, ⊗), where a ⊗ b := max{a + b −
1, 0}.
(4) The quantale of fuzzy subsets associated to a set X1 .
(5) The free local associated to a set X: (Ω(X), ⊇, ∩)2.
3. VALUE
CO - QUANTALE LOGICS
In this section, we will introduce a logic with truth values within
value co-quantales, generalizing Continuous Logic (see [BBHU08],
1
2
It is denoted by Λ(X)in [FK97]
Tt is denoted by Γ (X) in [FK97]
CO-QUANTALE VALUED LOGICS
11
where the truth values are taken in the unitary interval [0, 1], which
is a particular case of our setting).
Throughout the rest of this paper, we assume some technical conditions (Definitions 2.25, 2.29 and 2.56) that we need for providing
a proof of a version of Tarski-Vaught test -Proposition 3.35- and a
version of Łoś Theorem -Theorem 3.62- for the logics introduced in
this paper. At some point, we require to work with the symmetric
distance dsV of V.
Assumption 3.1. Throughout this section, we assume that V is a value
co-quantale which is co-divisible, co-Girard and a V-domain.
3.1. Modulus of uniform continuity. Modulus of uniform continuity are introduced in [BBHU08] as a technical way of controlling from
the language the uniform continuity of the mappings considered
in Continuous Logic. In this subsection, we develop an analogous
study of modulus of uniform continuity but in the setting of mappings valued in value co-quantales.
Remark 3.2. Given (M, dM ), (N, dN) V-continuity spaces and
(x1 , y1 ), (x2, y2 ) ∈ M×N, we define dM×N ((x1 , y1 ), (x2 , y2)) := dM (x1 , y1 )∨
dN (x2 , y2 ). By Proposition 2.35, (M × N, dM×N ) is a V-continuity space.
Definition 3.3. (c.f. [BBHU08]; pg 8) Given a mapping f : M → N
between two V-continuity spaces (M, dM ) and (N, dN), we say that ∆ :
V+ → V+ is a modulus of uniform continuity for f, if and only if, for
any x, y ∈ M and any ǫ ∈ V+ , dM (x, y) ≤ ∆(ǫ) implies dN (f(x), f(y)) ≤
ǫ.
As a basic consequence we have the following fact.
Proposition 3.4. Given (M, dM ), (N, dN), (K, dK) V-continuity spaces
and f : M → N, g : N → K uniformly continuous mappings, ∆ and Θ
modulus of uniform continuity for f and g respectively, then ∆ ◦ Θ is a
modulus of uniform continuity for g ◦ f.
Definition 3.5. Given a sequence of mappings (fn )n∈N with domain (M, dM)
and codomain (N, dN) (both of them V-continuity spaces), we say that
(fn)n∈N uniformly converges to a mapping f : M → N, if and only
if, for all ǫ ∈ V+ there exists n ∈ N such that for any m ≥ n and for any
x ∈ M we may say that dN (fm (x), f(x)) ≤ ǫ.
It is straightforward to see that uniform convergence behaves well
with respect to composition of mappings.
12
D. REYES AND P. ZAMBRANO
Proposition 3.6. Let (M, dM ), (N, dN), (K, dK ) be V-continuity spaces,
f : M → N, (fn)n∈N be a sequence of mappings from M to N, g : N → K,
and (gn )n∈N be a sequence of mappings from N to K such that (fn)n∈N
uniformly converges to f and (gn )n∈N uniformly converges to g. If g is
uniformly continuous, then (gn ◦ fn )n∈N uniformly converges to g ◦ f.
V
W
3.1.1. Uniform continuity of and . The following fact is very important because, as in Continuous Logic, it allows us to control
V (by
usingWdirectly the language) the uniform continuity of both (inf)
and (sup), understood as quantifiers (in an analogous way as in
Continuous Logic).
Proposition 3.7. Let (M, dM ), (N, dN) be V-continuity spaces, f : M ×
N → V be a uniformly continuous mapping provided with a modulus of
+
uniform continuity ∆ : V+ →
W V , then ∆ is also
V a modulus of uniform
continuity
for
the
mappings
:
M
→
V
and
f : M → V defined by
W
V f
x 7→ y∈N f(x, y) and x 7→ y∈N f(x, y) respectively.
Proof. Let ǫ ∈ V be such that 0 ≺ ǫ, y ∈ N and a, b ∈ M be such that
dM (b, a) ≤ ∆(ǫ). Then,
dM×N ((b, y), (a, y)) := dM (b, a) ∨ dN (y, y) = dM (b, a)
≤ ∆(ǫ)
Since ∆ is a modulus of uniform continuity for f, then
· f(b, y) ≤ dV (f(b, y), f(a, y))
f(a, y) −
≤ ǫ
By Proposition 2.11 (1) we may say
f(a, y) ≤ f(b, y) + ǫ ≤
_
f(b, z) + ǫ
z∈N
Since y ∈ N was taken arbitrarily, then
_
_
f(a, z) ≤
f(b, z) + ǫ
z∈N
z∈N
and by Proposition 2.11 (1)
_
_
_
_
·
·
f(a) −
f(b) =
f(a, z) −
f(b, z) ≤ ǫ.
f
f
z∈N
z∈N
In a similar way we prove the related statement for
V
f.
As an immediate consequence, we have the following useful facts.
CO-QUANTALE VALUED LOGICS
13
Corollary 3.8. Given an arbitrarily set I 6= ∅ and I-sequences (ai )i∈I , (bi)i∈I
in a V-continuity
(M, dM ), if ǫ ∈ V+ satisfies
d
W spaceW
V
VV (ai , bi ) ≤ ǫ for all
i ∈ I, then dV ( i∈I ai , i∈I bi ) ≤ ǫ and dV ( i∈I ai , i∈I bi ) ≤ ǫ.
Corollary 3.9. Given a V-continuity space (M, dM ), I 6= ∅, and a Isequence of mappings (fi : M → V)i∈I , if ∆ : V+ → V+ is a modulus
of uniform continuity forWfi ( i ∈ I), then ∆Vis also a modulus of uniform W
continuity for both V i fi : M → V and i fi : M → V defined by
x 7→ i∈I fi (x) and x 7→ i∈I fi (x), respectively.
Proposition 3.10. Let (M, dM ), (N, dN) be V−continuity spaces, f :
M × N → V, (fn )n∈N a sequence of mappings
W from M ×
N to V such
that (fn )n∈N uniformly converges to f, then
uniformly
y∈N fn (x, y)
n∈N
V
W
converges to y∈N f(x, y) and
uniformly converges to
y∈N fn (x, y)
n∈N
V
y∈N f(x, y).
Proof. Since by hypothesis (fn)n∈N uniformly converges to f, given
ǫ ∈ V+ there exists n ∈ N such that if m ≥ n, then dV (fm (x, y), f(x, y)) ≤
ǫ for any (x, y) ∈ M × N. For a fixed x ∈ M and m ≥ n, define the
sequences (fm (x, y))y∈N and
W (f(x, y))y∈N,Wwhich satisfy the hypothesis of Corollary 3.8, so dV ( y∈N fm (x, y), y∈N f(x, y)) ≤ ǫ whenever
m ≥ n. Since this holds for all ǫ ∈ V+ , we got the uniform convergence desired.
V
In an analogous way, we prove the respective statement for .
3.2. Some basic notions. In first order logic, an n-ary relation in a
set A is defined as a subset of An . In this way, a tuple (a1 , · · · , an )
might belong to A or not. We may codify this by using characteristic
functions, dually, by the discrete distance from a tuple in An to R.
In Continuous Logic, an n-ary relation in A is understood according
to this second approach by taking a uniformly continuous mapping
R : An → [0, 1]. In this setting, we generalize this approach replacing
[0, 1] by a suitable value co-quantale V.
All topological notions about V are relative to the symmetric topology of V.
Definition 3.11.
W Given a V-continuity space (M, dM ) and A ⊆ M, define
diam(A) := {dM (a, b) : a, b ∈ A}. (which we will call the diameter of
A.
3.2.1. Continuous structures. Given a V-continuity space (M, dM ) with
diameter p ∈ V, we define a continuous structure with underline Vcontinuity space (M, dM ) as a tuple M = ((M, dM), (Ri )i∈I , (fj)j∈J , (ck)k∈K ),
where:
14
D. REYES AND P. ZAMBRANO
(1) For each i ∈ I, Ri : Mni → V is a uniformly continuous mapping (which we call a predicate), with modulus of uniform
continuity ∆Ri : V+ → V+ . In this case, ni < ω is said to be
the arity of Ri .
(2) For each j ∈ J, fj : Mmj → M is a uniformly continuous mapping with modulus of uniform continuity ∆Fj : V+ → V+ . In
this case, mj < ω is said to be the arity of Fj .
(3) For each k ∈ K, Ck is an element M.
3.2.2. Languages for continuous structures. For a fixed continuous structure M := ((M, dM ), (Ri)i∈I , (fj)j∈J , (ck)k∈K ), we will define the language associated to M in the natural way, as follows.
Predicate symbols: Ri 7→ (Pi , ni , ∆Ri ) (i ∈ I)
Function symbols: fj 7→ (Fj, nj , ∆fj ) (j ∈ J)
Constant symbols: ck 7→ ek (k ∈ K).
This set of non logical symbols is denoted
W V by NLM .
Let us denote by LG := {d} ∪ X ∪ C ∪ { , } (which we call logical
symbols), where:
• X = {xi : i ∈ N} is a countable set of variables.
• C is the set of all uniformly continuous mappings with domain Vn and codomain V (1 ≤ n < ω). As in Continuous Logic, we understand a uniformly continuous mapping
u : Vn → V as a connective.
• d is a symbol, which we will interpret as the V-valued distance given in (M, dM ). This symbol will play the role of the
equality in first order logic, in a similar way as we do in Continuous Logic.
Definition 3.12. Given a continuous structure M
:=
((M, dM), (Ri)i∈I , (fj)j∈J , (ck )k∈K ), we define the language based on M as
LM := NLM ∪ LG. We will drop M if it is clear from the context.
We define the notion of terms as follows.
Definition 3.13. Given a language based on a continuous structure L, we
define the notion of L-term recursively, as follows:
• Any variable and any constant symbol is an L-term.
• Given L-terms t1 , ..., tn and a function symbol f ∈ L of arity n,
ft1, ..., tn is an L-term.
Definition 3.14. An L-term is said to be closed, if and only if, it is built
without use of variables.
Now, we provide the notion of L-formulae in this new setting. We
mimic the analogous notion given in Continuous Logic.
CO-QUANTALE VALUED LOGICS
15
Definition 3.15. Given L a language based on a continuous structure, we
define the notion of L-formula recursively, as follows:
• Given L-terms t1 , t2, dt1 t2 is an L-formula.
• Given L-terms t1 , ...tn and a predicate symbol P ∈ L of arity n,
Pt1 , ...tn is an L-formula.
• Given L-formulas ψ1 , ..., ψm and a connective (i.e., a uniformly
continuous mapping) a : Vm → V, then aψ1 , ..., ψm is an Lformula.
V
W
• Given an L-formula ψ and a variable x, both xψ and xψ are
L-formulas.
Remark 3.16. Let V be a co-Girard value co-quantale and b ∈ V be a
· x. Denote the usual, dual and symdualizing element. Denote x ′ := b −
∗
s
metric distances in V by d, d and d respectively. Notice that the mapping
· : V → V defined by (b −
· )(x) := b −
· x is uniformly continub−
ous (relative to the symmetric topology) provided with modulus of uniform
continuity idV+ . In fact, given x, y ∈ V we have that
· y
d(y, x) = x −
· (b −
· x)) −
· y) (b is a dualizing element)
= (b −
· ((b −
· x) + y) (by Prop. 2.11 (5))
= b−
· (y + (b −
· x)) (+ is commutative)
= b−
· y) −
· (b −
· x) (by Prop. 2.11 (5))
= (b −
′ ·
′
= y −x
= d(x ′ , y ′ )
Therefore, d(y, x) = d(x ′ , y ′) = d∗ (y ′ , x ′ ).
Exchanging the role of y and x above, we may say that d(x, y) = d∗ (y, x) =
d∗ (x ′ , y ′ ) = d(y ′ , x ′). Since ds (x, y) = d(y, x) ∨ d∗ (y, x) = d∗ (y ′ , x ′ ) ∨
· ) respects the distance ds and therefore
d(y ′ , x ′ ) = ds (y ′ , x ′ ), then (b −
it is uniformly continuous relative to the symmetric topology provided with
idV+ as a modulus of uniform continuity.
W
V
Notation W
3.17. The
subsequences
x
and
x of an L-fórmula can be
V
written as x and x , respectively.
d(t1, t2) denotes the sequence dt1 t2 .
Definition 3.18. An L-formula
W φ is said
V to be quantifier-free, if and only
if, there are no appearances of x and x inside φ.
Definition 3.19. An appearance of a variable x inside
W anVL-fórmula φ is
said to be free whenever it is not under the scope of x o x inside φ.
Definition 3.20. An L-formula φ is said to be an L-sentence, if and only
if, all appearances of variables are not free.
16
D. REYES AND P. ZAMBRANO
Notation 3.21. φ(x1 , ..., xn) means that the variables that appear free in
φ are among x1 , ..., xn.
3.2.3. L-structures. Let L be a language based on a V-continuous structure M := ((M, dM), (Ri )i∈I , (fj)j∈J , (ck )k∈K ). Given a V-continuity
space (N, dN ) with diameter (Definition 3.11) at most diam(M), we
will interpret the symbols in NLM in (N, dN) as follows:
• For any predicate symbol P of arity n and modulus of uniform continuity ∆P , associate a uniformly continuous mapping PN : Nn → V with modulus of uniform continuity ∆P .
• For any function symbol F of arity m and with modulus of
uniform continuity ∆F , associate a uniformly continuous mapping
FN : Nm → N with modulus of uniform continuity ∆F .
• For any constant symbol e, associate an element eN ∈ N.
Also, the logical symbol d is interpreted in (N, dN) as the distance
d := dN .
N
Definition 3.22. Given a language L based on a continuous structure
M := ((M, dM), (Ri )i∈I , (fj)j∈J , (ck )k∈K ) and a V-continuity space (N, dN),
the continuous structure obtained by interpreting the symbols of L on N =
((N, dN) as above, N := ((N, dN), (PiN )i∈I , (FN
j )j∈J , (ek )k∈K ), is said to be
an L-structure.
3.2.4. Semantics. Given an L-structure N and A ⊆ N, we extend the
language L by adding new constant symbols ca (a ∈ A), interpreting
cN
a := a. Abusing of notation, we will write a instead of ca , but
understood as a constant symbol. Let us denote this language by
L(A).
Definition 3.23. Given an L−structure N, for any L − term t we define
recursively its interpretation in N, denoted by tN , as follows:
(1) If t is a constant symbol c, define tN := cN .
(2) If t is a variable x, define tN : N → N as the identity function of N.
(3) If t is of the form ft1 , ...tn provided that f is an n-ary function symN
bol and t1 (x), ..., tn(x) are L-terms, define tN := fN (tN
1 , ..., tn ) as
N N
N N
N
m
the mapping f (t1 , ..., tn ) : N → N where (a) 7→ f (t1 (a), ..., tN
n (a))
m
for all a ∈ N .
Definition 3.24. Let N be an L-structure. We define recursively the interpretation of L(N)-sentences in N, as follows.
N
(1) (d(t1, t2))N := dN (tN
1 , t2 ), where t1 , t2 are L(N)-terms
CO-QUANTALE VALUED LOGICS
17
N
(2) (P(t1, ..., tn))N := PN (tN
1 , ..., tn ), where P is an n-ary predicate
symbol and t1 , · · · , tn are L(N)-terms.
N
(3) (u(φ1 , .., φn))N := u(φN
1 , ..., φn ) for any uniformly continuous
n
mapping
(connective)
u : V → V and all L(N)-sentences φ1 , ..., φn.
W
W
(4) (Vx φ)N := Va∈N φN (a), whenever φ(x) is an L(N)-formula.
(5) ( x φ)N := a∈N φN (a), whenever φ(x) is an L(N)-formula.
Analogously as in Continuous Logic, all terms and all formulae
have a modulus of uniform continuity, which do not depend of the
structures.
Proposition 3.25. Given L a language based on a continuous structure,
φ(x1 , ..., xn) an L − formula and t(x1 , ..., xm) an L-term, then there exist
∆φ : V+ → V+ and ∆t : V+ → V+ such that for any L-structure N, ∆φ is
a modulus of uniform continuity for φN and ∆t is a modolus of continuity
for tN .
Proof. The basic cases are given by definition, since constant symbols’s interpretations can be viewed as constant functions, variables
are interpreted as the identity function and predicate symbols are
interpreted as a uniformly continuous mapping with the respective
modulus of uniform continuity. The connective case corresponds
to compose uniformly continuous mappings (and we get the desired result by Proposition 3.4), and the quantifier cases follow from
Proposition 3.7.
Definition 3.26. Given L a language based on a continuous structure and
M, N L − structures, we say that M is an L-substructure of N, if and
only if, M ⊆ N and the interpretations of all non logical symbols and of d
in M correspond to the respective restrictions of the interpretations in N of
those symbols.
3.2.5. L-conditions. Fix L a language based on a continuous structure.
Definition 3.27. (c.f. [BBHU08]; Def 3.9) Given φ1 (x1 , ..., xn), φ2 (x1 , ..., xn)
L − formulas, we say that φ1 is logically equivalent to φ2 , if and
only if, for any L-structure M and any a1 , ..., an ∈ M we have that
M
φM
1 (a1 , ..., an) = φ2 (a1 , ..., an).
Definition 3.28. Let φ1 (x1 , ..., xn), φ2 (x1 , ..., xn) be L-formulas and M be
an L-structure, we define the logical distance between φ1 and φ2 relative
to M as follows:
W
M
d(φ1 , φ2 )M := {dV (φM
1 (a1 , ...an), φ2 (a1 , ...an)) : a1 , ...an ∈ M}
The logical distance between φ1 , φ2 is defined as follows:
18
D. REYES AND P. ZAMBRANO
d(φ1 , φ2 ) :=
W
{d(φ1 , φ2 )M : M is an L − structure}
We define the notion of satisfiability in an L-structure in an analogous ways as in Continuous Logic, by using the notion of L-conditions.
Definition 3.29. (c.f. [BBHU08]; pg 19) An L-condition E is a formal
expression of the form φ = 0, where φ(x1 , ..., xn) is an L-fórmula. An
L-condition E is said to be closed if it is of the form φ = 0, where φ
is an L-sentence. Given an L-formula φ(x1 , ..., xn), the related condition
E : φ(x1 , ..., xn) = 0 is denoted by E(x1 , ..., xn).
Definition 3.30. Given φ(x1 , ..., xn) an L-fórmula, M an L-structure and
a1 , .., an ∈ M, the L-condition E(x1 , ..., xn) : φ(x1 , ..., xn) = 0 is said to
be satisfied in M for a1 , ..., an, if and only if, φM (a1 , ..., an) = 0. We
denote this by M |= E(a1 , ..., an).
Notation 3.31. Given φ, ψ L-formulae, we denote by φ = ψ the L· ψ) ∨ (ψ −
· φ) = 0 and we denote by φ ≤ ψ the L-condition
condition (φ −
·
φ − ψ = 0.
Definition 3.32. An L-theory is a set of closed L-conditions.
Definition 3.33. Given an L-theory T and an L-structure M, we say that
M is a model of T , if and only if, for any L-condition E ∈ T we have that
M |= E.
3.3. Tarski-Vaught test.
Definition 3.34. (c.f. [BBHU08]; Def 4.3)Let M, N be L-structures.
(1) We say that M is elementary equivalent to N (denoted by M ≡
N), if and only if, any L-sentence ϕ satisfies ϕM = ϕN .
(2) Let M be an L-substructure of N. We say that M is an L-elementary
substructure of N (denoted by M 4 N), if and only if, any Lformula ϕ(x1 , ..., xn) satisfies ϕM (a1 , ..., an) = ϕN (a1 , ..., an) for
all a1 , ..., an ∈ M. In this case, we also say that N is an Lelementary extension of M.
We will provide a version of the well-known Tarski-Vaught test,
as a equivalence of being an L-elementary substructure, as it holds
in both first order and Continuous logics. We need to assume that V
is co-Girard (Definition 2.29).
Proposition 3.35. (Tarski-Vaught test, c.f. [BBHU08] Prop 4.5) Assume
that V is a co-Girard value co-quantale and let b a dualizing element of V.
Let M, N be L-structures such that M ⊆ N. The following are equivalent:
(1) M 4 N.
CO-QUANTALE VALUED LOGICS
19
(2) For any L-formula ϕ(x, x1 , ..., xn) and a1 , ..., an ∈ M, we have
that
V
{ϕM (c, a1, ..., an) : c ∈ M} =
V
{ϕN (c, a1, ..., an) : c ∈ N}
Proof. Suppose that M 4 N. Let ϕ(x, x1 , ..., xn) be an L-formula and
a1 , ...an ∈ M, so
^
^
{ϕM (c, a1, ..., an) : c ∈ M} = ( ϕ(x, a1, ..., an))M
x
(by Definition 3.24(5))
^
= ( ϕ(x, a1, ..., an))N
x
(since M 4 N)
^
=
{ϕN (c, a1, ..., an) : c ∈ N}
(by Definition 3.24(5)).
On the other hand, suppose that for any L-formula ϕ(x, x1 , ..., xn)
and any a1 , ..., an ∈ M we have that
^
{ϕM (c, a1, ..., an) : c ∈ M} =
^
{ϕN (c, a1, ..., an) : c ∈ N}
By Definition 3.34 (2), we need to do an inductive argument on
L-formulas in order to prove M 4 N. It is straightforward to see that
M ⊆ N guarantees the basic cases, and from the hypothesis ifVfollows
the inductive step by using connectives and the
Wquantifier , so we
have just to check the inductive step by using . Let ϕ(x, x1 , ..., xn)
be an L-formula and a1 , ..., an ∈ M. Let b ∈ V be a dualizing ele· is uniformly continuous in
ment. Notice that by Remark 3.16 b −
the symmetric topology and so it is a connetive. Theferore,
20
(
D. REYES AND P. ZAMBRANO
_
x
ϕ(x, x1 , ..., xn))M (a1 , ..., an) =
_
{ϕ(c, a1, ..., an) : c ∈ M}
(by Definition 3.24 (4))
_
· (b −
· ϕM (c, a1, ..., an)) : c ∈ M}
=
{b −
(b is a dualizing element)
^
·
· ϕM (c, a1, ..., an) : c ∈ M}
= b−
{b −
(by Proposition 2.14)
^
·
· ϕN (c, a1, ..., an) : c ∈ N}
= b−
{b −
(hypothesis induction on ϕ
and by applying this statement to
· ϕ(x, x1 , ..., xn) and
b−
^
· ϕ(x, x1 , ..., xn)
(b −
x
· is a connective by Remark 3.16-)
-b −
_
· (b −
· ϕN (c, a1, ..., an)) : c ∈ N}
=
{b −
(by Proposition 2.14)
_
=
{ϕN (c, a1, ..., an) : c ∈ N}
(b is a dualizing element)
_
= ( ϕ(x, x1 , ..., xn))N (a1 , ..., an)
x
(by Definition 3.24)
3.4. D-products and Łoś Theorem in co-quantale valued logics. Chang
and Keisler ([CK66] ) defined some logics with truth values on Hausdorff compact topological spaces. In that context, they provided a
version of Łos’ Theorem, which implies a Compactness Theorem in
their logic and the existence of saturated models (as it holds in first
order logic). This approach is rediscovered in [BBHU08], but by taking the particular case of truth values in the unit interval [0, 1]. We
propose to generalize the version of Łoś Theorem in our context of
value co-quantale valued logics, as a test question of the logics proposed in this paper.
CO-QUANTALE VALUED LOGICS
21
3.4.1. D-limits. Let us fix V a V-domain value co-quantale provided
with its symmetric topology. By Remark 2.57, (V, τs) is compact and
Hausdorff, therefore we may apply Lemma 3.37 to the symmetric
topology of V. Let I be a non empty set and D an ultrafilter over I.
Remark 3.36. We need to assume that V is provided with its symmetric
· q) ∨ (q −
· p), because we need to guarantee that
distance dsV (p, q) := (p −
s
·
·
p−q ≤ dV (p, q), which might fail for the original distance d(p, q) := q−p
· p ≤ d(p, q) always holds for both the original
in V. The inequality q −
and the symmetric distances of V.
From now, for the sake of simplicity, let us denote the symmetric distance
of V by dV .
The following is a very known fact about convergence of sequences
in Hausdorff Compact topological spaces.
Lemma 3.37. ([CK66]; Thrm 1.5.1.) If (X, τ) is a Hausdorff compact topological space, given a sequence (xi )i∈I in X there exists a unique x ∈ X such
that for any neighborhood V of x, then {i ∈ I|xi ∈ V} ∈ D.
Fact 2.48 and Lemma 3.37 allow us to give the following notion of
convergence in our setting.
Definition 3.38. Given (ai )i∈I a sequence in V, the unique a ∈ V which
satisfies that for any ǫ ∈ V+ we have that {i ∈ I|dV (a, ai) ≤ ǫ} ∈ D is
said to be the D-ultralimit of the sequence (ai )i∈I , which we denote it by
limi,D ai .
Definition 3.39. Given ǫ ∈ V+ , define
A(ǫ) := {j ∈ I|dV (limi,D ai , aj) ≤ ǫ}.
Proposition 3.40. Let (ai )i∈I be a sequence in V and b ∈ V.
(1) If there exists A ∈ D such that for all j ∈ A we have that b ≤ aj,
then b ≤ limi,D ai .
(2) If dV is the symmetric distance of V and there exists A ∈ D such
that for all j ∈ A we have that b ≥ aj, then b ≥ limi,D ai .
Proof.
(1) It is enough to prove that for any ǫ ∈ V such that 0 ≺
· limi,D ai ≤ ǫ, because V is completely
ǫ we have that b −
·
distributiveV(Definition 2.5) and then we would have that b −
limi,D ai ≤ {ǫ ∈ V : 0 ≺ ǫ} = 0 and by Proposition 2.11 (4)
b ≤ limi,D ai holds, as desired.
Let ǫ ∈ V+ , so by definition of limi,D ai we know that {j ∈
I : dV (limi,D ai , aj) ≤ ǫ} =: A(ǫ) ∈ D. By hypothesis A ∈ D,
22
D. REYES AND P. ZAMBRANO
therefore A(ǫ) ∩ A ∈ D and so there exists j ∈ A(ǫ) such
· limi,D ai ≤
that b ≤ aj (because j ∈ A). Notice that aj −
dV (limi,D ai , aj) ≤ ǫ (by Remark 3.36 and since j ∈ A(ǫ)). By
· limi,D ai ≤
Lemma 2.13 and since b ≤ aj , we may say that b −
·
aj − limi,D ai ≤ ǫ
(2) It is enough to prove that whenever 0 ≺ ǫ we have that
· b ≤ ǫ. As above, A(ǫ) ∈ D. Since A ∈ D, there
limi,D ai −
exists j ∈ A(ǫ) such that aj ≤ b, since dV is the symmetric dis· aj ≤
tance of V and by Remark 3.36 we have that limi,D ai −
· aj ) ∨ (aj −
· limi,D ai ) ≤ ǫ; by
dV (limi,D ai , aj) := (limi,D ai −
· b ≤
Lemma 2.13 and since aj ≤ b we have that limi,D ai −
· aj ≤ dV (limi,D ai , aj) ≤ ǫ, as desired.
limi,D ai −
The following fact is a kind of converse of the previous result, by
assuming co-divisibility (Definition 2.25).
Proposition 3.41. Suppose that V is co-divisible (Definition 2.25) and
that dV is the symmetric distance of V. Let (ai )i∈I be a sequence in V and
· limi,D ai . Therefore, there
b ∈ V such that limi,D ai ≤ b and 0 ≺ b −
exists A ∈ D such that i ∈ A, ai ≤ b.
· limi,D ai , therefore by Lemma 2.6 there
Proof. By hypothesis 0 ≺ b −
· limi,D ai . By Lemma 2.3
exists some ǫ ∈ V such that 0 ≺ ǫ ≺ b −
·
(1) we may say 0 ≺ ǫ ≤ b − limi,D ai . By taking A(ǫ) := {j ∈
I : dV (limi,D ai , aj) ≤ ǫ}, we have that A(ǫ) ∈ D (by definition of
limi,D ai ). By Remark 3.36 and definition of A(ǫ), for all j ∈ A(ǫ)
· limi,D ai ≤ dV (limi,D ai , aj ) ≤ ǫ ≤ b −
· limi,D ai .
we have that aj −
By Proposition 2.11 (1) and Lemma 2.26 (by hypothesis, V is co· limi,D ai ) + limi,D ai = b, so
divisible), we may say that aj ≤ (b −
A := A(ǫ) is the required set.
Lemma
in X. Therefore,
V
W 3.42. Let K 6= ∅ a set and (ak )k∈K bea K-sequence
+
·
(
(a
−
a
))
=
0,
if
and
only
if,
for
all
ǫ
∈
V
there
exists k ∈ K
k
k∈K
l∈K
W l
V
W
·
·
such that l∈K al − ǫ ≤ ak . Also, k∈K ( l∈K (ak − al ))
V = 0, if and only
· ǫ ≤ l∈K al .
if, for all ǫ ∈ V+ there exists k ∈ K such that ak −
V
W
· ak )) = 0 and let ǫ ∈ V+ (i.e.,
Proof. Suppose that k∈K ( l∈K (al −
V
W
· ak )) we may
0 ≺ ǫ), by definition of ≺ and since 0W≥ k∈K ( l∈K (al −
·
say that there exists
W
W k ∈ K such thatW l∈K (al − ak ) ≤ ǫ. By Fact 2.12,
·
·
·
− ak , so ( l∈K al ) − ak ≤ ǫ, and by Propol∈K (al − ak ) = ( l∈K al )W
· ǫ ≤ ak .
sition 2.11 we have that ( l∈K ak ) −
Conversely,
suppose
that
for
all
ǫ
∈ V+ there exists k ∈ K such that
W
·
l∈K al − ǫ ≤ ak . By Proposition 2.11 (1) and Fact 2.12 we may say
CO-QUANTALE VALUED LOGICS
23
W
W
· ak ) = ( l∈K al ) −
·
that l∈KV
(al −
W
V ak ≤ ǫ. Therefore (by Proposi· ak )) ≤ {ǫ ∈ V : 0 ≺ ǫ} = 0
tion 2.4), k∈K ( l∈K (al −
idea works for proving the second statement. Suppose
V A similar
W
· l )) = 0 and let ǫ ∈ V+ (i.e., 0 ≺ ǫ). By definition of ≺
(
(a
k −a
k∈K
l∈K
W
·
there exists k ∈ K such
Vthat l∈K (ak −al ) ≤ ǫ, and by Proposition 2.14
·
we may say that
V ak − l∈K al ≤ ǫ. By Proposition 2.11 (1), this implies
·
that ak − ǫ ≤ l∈K al .
· ≤
Conversely,
suppose that for all ǫ ∈ V+ exist k ∈ K such that ak −ǫ
V
Wl∈K al , so by Proposition
V 2.11 (1) and Proposition 2.14 we have that
·
·
(a
−
a
)
=
a
−
l
k
Vl∈K Wk
V l∈K al ≤ ǫ, therefore (by Proposition 2.4)
·
{ǫ ∈ V : 0 ≺ ǫ} = 0.
k∈K ( l∈K (ak − al )) ≤
Remark 3.43. The following fact is very important to deal with the quantifier cases in the proof of Łoś Theorem in this setting. The idea of these
proofs is quite similar to the one presented in [BBHU08], but adapted to
our setting.
Proposition 3.44. (c.f.[BBHU08]; Lemma 5.2) Let dV be the symmetric
distance of V. Let S 6= ∅ and (Fi )i∈I be a sequence of mappings with domain
V S and codomain V, then: V
W{limi,D Fi (x) : x ∈ S} ≥ limi,D (W{Fi (x) : x ∈ S}) and
{limi,D Fi (x) : x ∈ S} ≤ limi,D ( {Fi (x) : x ∈ S}).
Also, given ǫ ∈ V+ there exist
W sequences (bi )i∈I and (ci )i∈I in S such that
limi,D Fi (bi ) + ǫ ≥ limi,D (V {Fi (x) : x ∈ S}) and
· ǫ ≤ limi,D ( {Fi (x) : x ∈ S}), whenever there exist B, C ∈
limi,D Fi (ci )V−
W
V
W
· i (y))) = 0 for all i ∈ B and y∈S ( x∈S (Fi (y)−
·
D such that y∈S ( x∈S (Fi (x)−F
Fi (x))) = 0 for all i ∈ C.
V
Proof. Let ri := {Fi (x) : x ∈ S}, r := limi,D ri and ǫ ∈ V+ . Define
A(ǫ) = {j ∈ I : dV (r, rj) ≤ ǫ}, so by Definition of limi,D ri we may say
that A(ǫ) ∈ D. Notice that if j ∈ A(ǫ) and by Remark 3.36 we have
· rj ≤ dV (r, rj) ≤ ǫ. So, by Proposition 2.11 (1) it follows that
that r −
· ǫ ≤ rj .
r ≤ rj + ǫ and then r −
V
·
Let x ∈ S, then r − ǫ ≤ rj := {Fj (y) : y ∈ S} ≤ Fj (x). Since
·
A(ǫ) ∈ D, by Proposition 3.40 (1) we have that
V r − ǫ ≤ limi,D Fi (x),
·
and since it holds for any x ∈ S then r − ǫ ≤ {limi,D Fi (x) : x ∈ S}.
By Proposition
2.11 (1) and by commutativity of +, it follows that
V
r ≤ {limi,D Fi (x) : x ∈ S} + ǫ.VSince 0 ≺ ǫ was taken arbitrarily, by
Fact 2.19 we may say that r ≤ {limi,D Fi (x)|x ∈ S}.
W
Let si := {Fi (x) : x ∈ S}, s := limi,D si and ǫ ∈ V+ . By taking
A′ (ǫ) = {j ∈ I : dV (s, sj ) ≤ ǫ}, then A′ (ǫ) ∈ D (definition of limi,D si ).
· s ≤ dV (s, sj ) ≤ ǫ
Given j ∈ A′ (ǫ), by Remark 3.36 we have that sj −
24
D. REYES AND P. ZAMBRANO
and by Proposition 2.11 (1) we may say that sj ≤ s + ǫ. Given x ∈ S,
it follows that Fj (x) ≤ sj ≤ s + ǫ. Since dV is symetric, by Proposition 3.40 (2) and since A′ (ǫ) ∈ D, we have
W that limi,D Fi (x) ≤ r + ǫ;
since x ∈ S was taken arbitrarily, then {limi,D Fi (x)|x
W ∈ S} ≤ r + ǫ.
Since 0 ≺ ǫ is arbitrary, by Fact 2.19 we have that {limi,D Fi (x)|x ∈
S} ≤ r.
Let us continue with the proof of the two last facts. Let ǫ ∈ V+ , so by
Lemma 2.17 there exists some θ ∈ V+ such that 0 ≺Vθ y θW+ θ ≤ ǫ. By
·
hypothesis, suppose there exists B ∈ D such that y∈S ( x∈S (Fi (x) −
Fi (y))) = 0 for all i ∈ B.
W Let i ∈ B, since 0 ≺ θ and by Lemma 3.42
·
take bi ∈ S such that x∈S
W Fi (x) − θ ≤ Fi (bi ), and so by Proposi· Fi (bi ) ≤ θ. If i ∈
tion 2.11 we may say that x∈S Fi (x) −
/ B, choose bi
as any element in S. Let B(θ) := {i ∈ I : dV (limi,D Fi (bi ), Fi (bi )) ≤ θ},
so by definition ofWlimi,D Fi (bi ) we know that B(θ) ∈ D. Therefore, if
j ∈ B ∩ B(θ) then x∈S Fj (x) ≤ limi,D Fi (bi ) + ǫ. In fact, if j ∈ B ∩ B(θ)
then
_
_
· limi∈I Fi (bi ) ≤ dV ( Fj (x), limi∈I Fi (bi ))
Fj (x) −
x∈S
x∈S
(by Remark 3.36)
_
≤ dV ( Fj (x), Fj(bj )) + dV (Fj (bj ), limi∈IFi (bi ))
x∈S
=
"
_
x∈S
· Fj (bj)
Fj(x) −
!
∨
·
Fj (bj ) −
_
Fj (x)
x∈S
+dV (Fj (bj), limi∈I Fi (bi ))
_
· Fj (bj) + dV (Fj(bj ), limi∈I Fi (bi ))
=
Fj(x) −
x∈S
≤ θ+θ
≤ ǫ.
W
·
Therefore, x∈S Fj (x) −
W limi∈I Fi (bi ) ≤ ǫ, and by Proposition 2.11
(1) we may say that x∈S Fj (x) ≤ limi∈I Fi (bi ) + ǫ. Hence,
W since
B ∩ B(θ) ∈ D by Proposition 3.40 (2) we have that limi,D ( {Fi (x) :
x ∈ S}) ≤ limi,D Fi (bi ) + ǫ.
Let us construct the sequence (ci )i∈I as follows: Let ǫ ∈ V+ . By
hypothesis and Lemma 3.42,
V there for all j ∈ C there exists cj ∈
· θ ≤ x∈S Fj (x), where θ ∈ V+ satisfies θ +
S such that Fj (cj ) −
V
· x∈S Fj (x) ≤
θ ≤ ǫ. By Proposition 2.11 (1), it follows that Fj(cj ) −
!#
CO-QUANTALE VALUED LOGICS
25
V
V
θ; since x∈S Fj (x) ≤ Fj (cj), it implies that dV (Fj (cj), x∈S Fj (x)) ≤
θ. If j ∈
/ C, take cj as any element of S. Let C(θ) := {i ∈ I :
dV (limi∈I Fi (ci ), Fi (ci )) ≤ θ}. Therefore, if j ∈ C ∩ C(θ) we have that
^
^
·
limi,D Fi (ci ) −
Fj (x) ≤ dV (limi,D Fi (ci ),
Fj (x))
x∈S
x∈S
≤ dV (limi,D Fi (ci ), Fj(cj)) + dV (Fj(cj ),
^
Fj(x))
x∈S
≤ θ+θ
≤ ǫ
V
· ǫ ≤ x∈S Fj (x).
By Proposition 2.11 (1), we have that limi,D Fi (ci ) −
· ǫ ≤
Since CV∩ C(θ) ∈ D and by Proposition 3.40 (1), limi,D Fi (ci ) −
limi,D ( {Fi (x) : x ∈ S}).
3.4.2. D-product and D-ultraproduct of spaces and mappings.
Proposition 3.45. Let dV be the symmetric distance of V. If (Mi )i∈I is a
sequence of Vcontinuity spaces such that for
Qall i ∈ I all distances dMi are
symmetric, then in the cartesian product i∈I Mi , the relation ∼ defined
by (xi )i∈I ∼ (yi )i∈I , if and only if, limi,D dMi (xi , yi ) = 0, is an equivalence
relation.
Proof. Reflexivity follows trivially and symmetry follows from symmetry
of all dMi . Let us focus on the transitivity. Let (xi )i∈I , (yi )i∈I , (zi )i∈I ∈
Q
i∈I Mi be such that limi,D dMi (xi , yi ) = 0 and limi,D dMi (yi , zi ) = 0.
Since V is a value co-quantale, it is enough to check that for any
ǫ ∈ V+ we have that limi,D dMi (xi , zi ) ≤ ǫ. Let ǫ ∈ V+, so by
Lemma 2.17 there exists θ ∈ V+ such that θ + θ ≤ ǫ. Since 0 ≺ θ
and by definition of limi,D dMi (xi , yi ) and limi,D dMi (yi , zi ), A(θ) :=
{i ∈ I : dV (limi,D dMi (xi , yi ), dMi (xi , yi ) ≤ θ} and B(θ) := {i ∈ I :
dV (limi,D dMi (yi , zi ), dMi (yi , zi ) ≤ θ} belong to D. By hypothesis,
limi,D dMi (xi , yi ) = 0 y limi,D dMi (yi , zi ) = 0. Notice that by Proposition 2.11 (4) and since 0 = min V, it follows that
dV (limi,D dMi (xi , yi ), dMi (xi , yi ))
· dMi (xi , yi ),
= ∨{limi,D dMi (xi , yi ) −
· limi,D dMi (xi , yi )}
dMi (xi , yi ) −
(dMi is assumed to be symmetric)
· dMi (xi , yi ), dMi (xi , yi ) −
· 0}
= ∨{0 −
= ∨{0, dMi (xi , yi )}
= dMi (xi , yi )
26
D. REYES AND P. ZAMBRANO
Therefore, A(θ) = {i ∈ I : dMi (xi , yi ) ≤ θ} and B(θ) = {i ∈ I :
dMi (yi , zi ) ≤ θ}. So, if i ∈ A(θ) ∩ B(θ) then dMi (xi , zi ) ≤ dMi (xi , yi ) +
dMi (yi , zi ) ≤ θ + θ ≤ ǫ. Since A(θ) ∩ B(θ) ∈ D, by Lemma 3.40 (2)
we may say that limi,D dMi (xi , zi ) ≤ ǫ.
Remark 3.46. In general, we do not require that all continuity spaces in the
sequence (Mi )i∈I are symmetric, where in that case ∼ might not be an equivalence relation. We just need this requirement if (Mi , dMi ) := (V, dM ) for
all i ∈ I. As we will see in the following proposition,
Q we can provided a
continuity space structure to the cartesian product i∈I Mi , without assuming the symmetry on dMi .
Proposition 3.47. Suppose that dV is the Q
symmetric distance. If (Mi )i∈I
is a sequence of V Q
− continuityspaces,
(
i∈I Mi , dD ) is a V-continuity
Q
space, where dD : i∈I Mi × i∈I Mi → V is defined by ((xi )i∈I , (yi )i∈I ) 7→
limi,D dMi (xi , yi ).
Q
Proof. Given (xi )i∈I ∈ i∈I Mi , then dQi∈ ((xi )i∈I , (xi )i∈I ) = limi,D dMi (xi , xi ) =
limi,D 0 = 0.
Q
Let (xi )i∈I , (yi )i∈I , (zi )i∈I ∈ i∈I Mi , ai := dMi (xi , yi ), bi := dMi (xi , zi ),
ci := dMi (yi , zi ), a := limi,D dMi (xi , yi ), b := limi,D dMi (xi , zi ) and
c := limi,D dMi (yi , zi ). We want to see that a ≤ b + c, which by
· b ≤ c. Let ǫ ∈
Proposition 2.11 (1) it is enough to prove that a −
+
V , so by Corollary 2.18 there exists some θ ∈ V+ such that θ +
θ + θ ≺ ǫ. By Lemma 2.3 (1), θ + θ + θ ≤ ǫ. By definition of
limi,D dMi (xi , yi ) =: a, limi,D dMi (xi , zi ) =: b and limi,D dMi (yi , zi ) =:
c, A := {i ∈ I : dV (a, ai ) ≤ θ}, B := {i ∈ I : dV (b, bi) ≤ θ} and
C := {i ∈ I : dV (c, ci ) ≤ θ} belong to D. So, A ∩ B ∩ C ∈ D. Let
i ∈ A ∩ B ∩ C, so by Proposition 2.11 (1) and (2) we may say that
· b) ≤ (a −
· ai ) + (ai −
· bi ) + (bi −
· b). Since dMi satisfies tran(a −
· bi ≤ ci . Since
sitivity, by Proposition 2.11 (1) we have that ai −
·
·
·
i ∈ C, we may say that ci − c ≤ ∨{ci − c, c − ci } = dV (c, ci) ≤ θ.
By Proposition 2.11 (1), it follows that ci ≤ c + θ. By monotonic· b ≤ (a −
· ai ) + (bi −
· b) + (c + θ). Since i ∈ A(θ) ∩ B(θ),
ity, a −
· ai ≤ ∨{a −
· ai , ai −
· a} = dV (a, ai ) ≤ θ and
we also have that a −
·
·
·
· b ≤ (a −
· ai ) +
bi − b ≤ ∨{bi − b, b − bi } = dV (b, bi) ≤ θ. Hence, a −
· b) + (c + θ) ≤ θ + θ + (c + θ) = θ + θ + θ + c ≤ ǫ + c. Since
(bi −
· b ≤ c.
ǫ ∈ V+ was chosen arbitrarily, it follows that a −
Definition 3.48. We call D-product of the sequence of V-continuity
spaces
Q
(Mi , dMi )i∈I to the V-continuity space (MD , dMD ) := ( i∈I Mi , dD ) defined in Proposition 3.47.
Lemma 3.49. Let dV be the symmetric distance of V. If (Mi , dMi ) is a
sequence of V-continuity spaces such that dMi is symmetric for all i ∈ I,
CO-QUANTALE VALUED LOGICS
27
Q
then
i∈I Mi / ∼, dD is a V-continuity space, provided that dD is defined by ([(xi )i∈I ], [(yi )i∈I ]) 7→ limi,D dMi (xi , yi ), where ∼ is defined as in
Proposition 3.45.
Proof. First, let us check that ((xi )i∈I , (yi )i∈I ) 7→Qlimi,D dMi (xi , yi ) is
well-defined. Let (ai )i∈I , (bi )i∈I , (ci)i∈I , (di )i∈I ∈ i∈I Mi be such that
(ai )i∈I ∼ (bi )i∈I and (ci )i∈I ∼ (di )i∈I . By Fact 2.48 and the definition of
D-ultralimits, in order to prove that limi,D dMi (ai , ci ) = limi,D dMi (bi , di )
it is enough to prove that for every ǫ ∈ V+ we have that
{i ∈ I : dV (limi,D dMi (ai , ci ), dMi (bi , di )) ≤ ǫ} ∈ D.
Let ǫ ∈ V+ , so by Proposition 2.17 there exists θ ∈ V+ such that
θ + θ ≺ ǫ. By Lemma 2.3 (1), we may say that θ + θ ≤ ǫ. In
similar way we can prove that there exists some δ ∈ V+ such that
δ + δ ≤ θ. Since (ai )i∈I ∼ (bi )i∈I , by definition of ∼ we have that
limi,D dMi (ai , bi ) = 0, hence for all γ ∈ V+ we have that {i ∈ I :
dV (limi,D dMi (ai , bi ), dMi (ai , bi )) ≤ γ} ∈ D. By Remark 2.36, we may
say that {i ∈ I : dMi (ai , bi ) ≤ γ} ∈ D. Since δ ∈ V+ , in particular
A := {i ∈ I : dMi (ai , bi ) ≤ δ} ∈ D. Analogously, since (ci )i∈I (di )i∈I
we may say that B := {i ∈ I : dMi (ci , di ) ≤ δ} ∈ D. So, if i ∈ A ∩ B
then
dMi (ai , ci ) ≤ dMi (ai , bi ) + dMi (bi , di ) + dMi (di , ci )
(by transitivity of dMi )
≤ δ + dMi (bi , di ) + δ
(since i ∈ A ∩ B)
≤ dMi (bi , di ) + θ
(by Proposition 2.10 and since δ + δ ≤ θ)
In a similar way we may say that dMi (bi , di ) ≤ dMi (ai , ci ) + θ, when· dMi (ai , ci ) ≤ θ
ever i ∈ A ∩ B. By Proposition 2.11 (1), dMi (bi , di ) −
· Mi (bi , di ) ≤ θ, therefore dV (dMi (ai , ci ), dMi (bi , di )) ≤
and dMi (ai , ci )−d
θ. Notice that by definition of limi,D dMi (ai , ci ), A(θ) := {i ∈ I :
dV (limi,D dMi (ai , ci ), dMi (ai , ci )) ≤ θ} ∈ D. So, if i ∈ A ∩ B ∩ A(θ) we
28
D. REYES AND P. ZAMBRANO
have that
dV (limi,D dMi (ai , ci ), dMi (bi , di )) ≤ dV (limi,D dMi (ai , ci), dMi (ai , ci ))
+dV (dMi (ai , ci ), dMi (bi , di ))
(by transitivity of dV )
≤ θ+θ
(since i ∈ A ∩ B ∩ A(θ))
≤ ǫ
(by Proposition 2.10 and since
θ + θ ≤ ǫ)
Definition 3.50. Let (M
Qi , dMi )i∈I be a sequence of V-continuity spaces.
The V-continuity space i∈I Mi / ∼ provided with the symmetric distance
dMD ([(xi )i∈I ], [(yi)i∈I ]) := limi,D dMi (xi , yi ) is called the D-ultraproduct
of the sequence (Mi , dMi )i∈I .
Definition 3.51. (c.f. [BBHU08]; pag 25) Let (Mi )i∈I , (Ni )i∈I be sequences of V-continuity spaces and K ∈ V be such that K ∈ V greater
than the diameter of all the considered spaces. Given a fixed n ∈ N \ {0}
and (fi : Mni → Ni )i∈I a sequence of uniformly continuous mappings provided with the same modulus of uniform continuity. The mapping fD :
n
N
(MD , dM
D ) → (ND , dD ) defined by
((x1i )i∈I , ..., (xni)i∈I ) 7→ (fi (x1i , ..., xni))i∈I
is said to be the D-product of the sequence (fi : Mni → Ni )i∈I
Proposition 3.52. Let V be a co-divisible value co-quantale such that dV
is the symmetric distance of V. If (fi : Mni → Ni )i∈I is a sequence of uniformly continuous mappings with the same modulus of uniform continuity
∆ : V+ → V+ , then ∆ is also a modulus of uniform continuity for the
n
N
D-product fD : (MD , dM
D ) → (ND , dD ).
Proof. For Q
the sake of simplicity, let us take n = 1. Let x = (xi )i∈I , y =
M
+
(yi )i∈I ∈
i∈I Mi such that dD (x, y) ≤ ∆(ǫ), where ǫ ∈ V (i.e.,
limi,D dMi (xi , yi ) ≤ ∆(ǫ)). By Lemma 3.41, there exists A ∈ D such
that for all i ∈ A we have that dMi (xi , yi ) ≤ ∆(ǫ). Since ∆ is a modulus of uniform continuity for fi for all i ∈ I, in particular we may say
that dNi (fi (xi ), fi (yi ) ≤ ǫ, whenever i ∈ A. Since dV is the symmetric distance of V and by Lemma 3.40 (2), dN
D ((fi )i∈I (x), (fi )i∈I (y)) ≤
ǫ.
CO-QUANTALE VALUED LOGICS
29
3.4.3. Łoś Theorem in value co-quantale logics. In order to prove a version of Łoś Theorem in this setting, we do not require that the involved distances of the metric structures are necessarily symmetric. Because of that, it is enough to consider the D-product of a sequence of V-continuity spaces (Mi , dMi )i∈I instead of its respective
D-ultraproduct, like we need to do in Continuous Logic for assuring that the obtained distance is actually symmetric. However, we
need to consider the respective quotient of D-powers of (V, dV ) as
a V-symmetric continuity space. From now, we assume that (V, dV )
is a compact, Hausdorff, co-divisible value co-quantale, where dV is
the symmetric distance of V.
Definition 3.53. Given (M, dM ) a V-continuity space where dM is symmetric, define the ultrapower of (M, dM ) as the D-ultraproduct
Q
i∈I M/ ∼, dD of the constant sequence ((M, dM ))i∈I
Q
Definition 3.54.
Given
a
D-ultraproduct
M
/
∼:=
(
D
i∈I Mi )/ ∼, the
Q
mapping θ : i∈I Mi → MD / ∼ defined by (xi )i∈I 7→ [(xi )i∈I ] is called the
canonical mapping.
Notation 3.55. We denote by (VD / ∼, dV/∼ ) the D-ultrapower of the Vcontinuity space (V, dV ) (where dV is the symmetric distance).
Definition 3.56. We say that two V-continuity spaces (X, dX ) and (Y, dY )
are V-equivalent, if and only if, there exists a bijection f : (X, dx ) →
(Y, dY ) such that dX (x, y) = dY (f(x), f(y)) for any x, y ∈ X. In this case,
we say that f is an V-equivalence.
Proposition 3.57. (c.f. [BBHU08]; pg 26) Let (V, dV ) a value co-quantale
provided that dV is the symmetric distance, then the D-ultrapower (VD / ∼
, dV/∼ ) is V−equivalent to (V, dV ).
Proof. Defining T : V → V/ ∼ by x 7→ [(x)i∈I ], we have that T is
a V-equivalence. In fact, T is injective: Let x, y ∈ V be such that
T (x) = T (y), so by definition of ∼ (Proposition 3.45) we have that
dV/∼ (T (x), T (y)) = dV/∼ ([(x)i∈I], [(y)i∈I]) = limi,D dV (x, y) = dV (x, y) =
· y) ∨ (y −
· x) = 0, therefore x −
· y = 0 and y −
· x = 0. By Proposi(x −
tion 2.11 (4) we may say that x ≤ y and y ≤ x, and so x = y.
In order to prove that T is surjective, let [(xi )i∈I ] ∈ VD . Let us see
that T (limi,D xi ) := [(limi,D xi )i∈I ] = [(xi )i∈I ] (i.e., we will have that
limi,D dV (limi.D xi , xi ) = 0)): let ǫ ∈ V+ , therefore dV (0, dV (limi.D xi , xi )) =
· dV (limi.D xi , xi ), dV (limi.D xi , xi ) −
· 0} = ∨{0, dV (limi.D xi , xi )} =
∨{0 −
dV (limi.D xi , xi ), and by definition of limi.D xi we may say that {i ∈ I :
dV (limi.D xi , xi ) ≤ ǫ} ∈ D. Hence, {i ∈ I : dV (0, dV (limi.D xi , xi )) ≤
ǫ} ∈ D and then limi,D dV (limi.D xi , xi ) = 0.
30
D. REYES AND P. ZAMBRANO
T preserves distances: In fact, dVD (T (x), T (y)) = dVD ([(x)i∈I], [(y)i∈I]) :=
limi,D (dV (x, y))i∈I = dV (x, y).
Therefore, T is an equivalence.
Remark 3.58. Notice that the mapping T ′ : VD / ∼ → V defined by
[(xi )i∈I ] 7→ limi,D xi is the inverse of T .
Fact 3.59. Given a sequence of V-continuity spaces ((Mi , dMi ))i∈I provided
that all distances dMi are symmetric, the canonical mapping θ :
Q
i∈I Mi → MD / ∼ is uniformly continuous with modulus of uniform
continuity idV+ .
Definition 3.60. Suppose that V is co-divisible and let L be a language
based in a continuous structure. Given a sequence (Mi )i∈I of L-structures,
define the D-product of (Mi )i∈I as the L-structure MD with underlying
V-continuity space (MD , dD ), defined as follows:
Q
(1) For a predicate symbol R ∈ L, define R i∈I Mi := T ′ ◦ θ ◦ RD ,
where RD is the D-product of the mappings (RMi )i∈I , θ the canonical mapping given in Definition 3.54 and T ′ the mapping defined
in Remark 3.58.
Q
(2) For a function symbol F ∈ L, defined F i∈I Mi as the D-product of
the mappings (FMi )i∈I .
Q
(3) For a constant symbol c ∈ L, define c i∈I Mi := (cMi )i∈I .
Remark 3.61. Notice that Theorem 3.52 guarantees that the interpretations of the symbols of L given above have the same modulus of uniform
continuity given by the language.
Theorem 3.62. (Łoś Theorem; c.f. [BBHU08] Thrm 5.4). Let (V, dV ) be
a co-divisible V-domain. If (Mi )i∈I is a sequence of L-structures, then for
any L-formula φ(x1 , ..., xn) (if φ has quantifiers, we require that its interpretations in any L-structure Mi satisfyQthe hypothesis of Proposition 3.42)
and any tuple ((a1i)i∈I , ..., (ani)i∈I ) ∈ ( i∈I Mi )n , we have that
φMD ((a1i)i∈I , ..., (ani)i∈I ) = limi,D φMi (a1i , ..., ani)
Proof. We proceed by induction on L-formulae.
(1) φ : d(x1 , x2 )
(d(x1 , x2 ))M ((a1i)i∈I , (a2i)i∈I ) := dM ((a1i )i∈I , (a2i)i∈I )
= limi,D dMi (a1i , a2i)
(definition of a distance in a product)
= limi,D (d(x1 , x2 ))Mi (a1i, a2i ))
CO-QUANTALE VALUED LOGICS
31
(2) φ : R(x1 , ..., xn), where R is a predicate symbol in L.
(R(x1 , ..., xn)((a1i)i∈I , ..., (ani)i∈I ))M := (R((a1i)i∈I , ..., (ani)i∈I ))M
= T ′ ◦ θ((RMi (a1i , ..., ani))i∈I )
(by Definition 3.60 (1))
= limi,D RMi ((a1i, ..., ani))
(by definition of θ and T ′ )
(3) φ : u(σ1 , ..., σm)(x1 , ..., xn), where u : Vm → V is a uniformly
continuous mapping and σ1 , · · · σm are L-formulae such that
Mi
1
n
1
n
σM
k ((ai )i∈I , ..., (ai )i∈I ) = limi,D σk (ai , ..., ai ) for all k ∈ {1, ..., m}
(induction hypothesis). For the sake of simplicity, denote
((a1i)i∈I , ..., (ani)i∈I ) =: a and (a1i , ..., ani) = ai .
M
(u(σ1 , ...σm))M (a) := u(σM
1 (a), ..., σm (a))
Mi
i
= u(limi,D σM
1 (ai ), ..., limi,D σm (ai ))
(induction hypothesis)
i
Define bi,k := σM
k (ai ) for any k ∈ {1, · · · , m} and i ∈ I. Notice
that
{i ∈ I : dV (u(limi,D bi,1 , ..., limi,D bi,m ), u(bi,1, ..., bi,m)) ≤ ǫ}
contains the set
{i ∈ I : dV n ((limi,D bi,1 ..., limi,D bi,m ), (bi,1, ..., bi,m)) ≤ ∆(ǫ)}
n
_
= i ∈ I : {dV (limi,D bi,k , bi,k )} ≤ ∆(ǫ) ,
k=1
whenever ∆ is a modulus of uniform continuity for u.
Notice
that this previous set belongs to D, because it contains
Tn
{i
∈ I|dV (limi,D bi,k , bi,k ) ≤ ∆(ǫ)}, which belongs to D by
k=1
definition of a D-limit and since D is an ultrafilter.
Therefore,
Mi
Mi
Mi
i
u(limi,D σM
1 (ai ), ..., limi,D σm (ai )) = limi,D u(σ1 (ai ), · · · σm (ai ))
limi,D (u(σ1 , · · · , σm ))Mi (ai ).
W
(4) φ : x ϕ(x, x1 , ..., xn)
Let ϕ(x, x1 , ..., xn) be an L-formula such that
ϕM ((bi )i∈I , (a1i)i∈I , ..., (ani)i∈I ) = limi∈I ϕMi (bi , a1i , ..., ani)
32
D. REYES AND P. ZAMBRANO
Q
for all (bi )i∈I ∈ i∈I Mi (induction hypothesis). For the sake
of simplicity, denote a := ((a1i)i∈I , ..., (ani)i∈I ), ai := (a1i , ..., ani)
and b := (bi )i∈I . So,
_
x
ϕ(x, x1 , ..., xn)
!M
(a) =
_
x
ϕ(x, a)
!M
_
=
{ϕM (b, a) : b ∈ M}
(by Definition 3.24 (4))
_
=
{limi,D ϕMi (bi , ai ) : b ∈ M}
(induction hypothesis)
_
≤ limi,D {ϕMi (bi , ai ) : b ∈ M}
(by Proposition 3.44)
By Proposition 3.44, given ǫ ∈ V+ there exists a sequence
Q
(cj)j∈I of tuples cj := (cji )i∈I ∈ i∈I Mi such that
limi,D (
_
{ϕMi (bi , ai ) : b ∈ M}) ≤ limi,D ϕMi (cii , ai ) + ǫ
_
≤
{limi,D ϕMi (bi , a(i)) : b ∈ M} + ǫ
where the last inequality
follows from monotonicity of ≤
Q
i
and since (ci )i∈I ∈ i∈I Mi . Since ǫ ∈ V+ is arbitrary, we have
W
W
that limi,D {ϕMi (bi , ai ) : b ∈ M} ≤ limi,D {ϕMi (bi , ai ) :
b ∈ M}.
by antisymmetry
W Therefore,
W of ≤ weMmay say that
Mi
limi,D {ϕ (b(i), aiW
) : b ∈ M} = {limi,D ϕW i (b(i), a(i)) :
Mi
b ∈ M}. Notice
W that {ϕ (biM, ai ) : b ∈ M} =W {ϕ(c, ai ) : c ∈
Mi }, then ( xWϕ(x, x1 , ..., xn)) (a) = limi,D {ϕ(c, ai ) : c ∈
Mi } = limi,D ( x ϕ(x, x1 , ..., xn))Mi (ai ), as desired.
V
(5) φ : x ϕ(x, x1 , ..., xn)
Let ϕ(x, x1 , ..., xn) be an L-formula such that
ϕM ((bi )i∈I , (a1i)i∈I , ..., (ani)i∈I ) = limi∈I ϕMi (bi , a1i , ..., ani)
CO-QUANTALE VALUED LOGICS
33
Q
for all (bi )i∈I ∈
i∈I Mi . For the sake of simplicity, denote
1
n
a := ((ai )i∈I , ..., (ai )i∈I ), ai := (a1i, ..., ani) and b := (bi )i∈I . So,
!M
!M
^
^
ϕ(x, x1 , ..., xn)
(a) =
ϕ(x, a)
x
x
^
=
{ϕM (b, a) : b ∈ M}
(by Definition 3.24 (5))
^
=
{limi,D ϕMi (bi , ai ) : b ∈ M}
(induction hypothesis)
^
≥ limi,D {ϕMi (bi , ai ) : b ∈ M}
(by Proposition 3.44)
By Proposition 3.44 and by hypothesis, given ǫ ∈ V+ there
Q
exists a sequence (bj)j∈I of tuples bj := (bji )i∈I ∈ i∈I Mi such
that
limi,D (
^
{ϕMi (bi , ai ) : b ∈ M}) + ǫ ≥ limi,D ϕMi (bii , ai )
^
≥
{limi,D ϕMi (bi , ai ) : b ∈ M}
follows from the fact that (bii )i∈I ∈
Qwhere the last inequality
+
ǫ ∈ V was takenVarbitrarily, we have that
i∈I M
Vi . Since
Mi
limi,D {ϕ V
(bi , ai ) : b ∈ M} ≥ {limi,D ϕMi (bi , ai ) : b ∈
M} ≥ limi,D {ϕMi (b(i), ai) : b ∈ M}. By antisymmetry of ≤,
we may say that
^
^
limi,D {ϕMi (bi , ai ) : b ∈ M} = {limi,D ϕMi (bi , ai ) : b ∈ M}.
V M
V
i
Since, V {ϕMi (bi , ai ) : b ∈ M} = {ϕ
V (c, ai ) : c ∈ Mi },
M
then ( Vx ϕ(x, x1 , ..., xn)) (a) = limi,D {ϕ(c, ai ) : c ∈ Mi } =
limi,D ( x ϕ(x, x1 , ..., xn))Mi (ai ), as desired.
Note 3.63. In case that we want to work in symmetric spaces, the same
argument as above
to prove a version of Łoś Theorem for the DQ works
ultraproduct
i∈I Mi ∼ .
3.4.4. Some consequences of Łoś Theorem. In this setting, Łoś Theorem implies a version of Compactness Theorem and the existence
of some kind of ω1 -saturated models, as it holds in both first order
and Continuous logics.
34
D. REYES AND P. ZAMBRANO
First, we provide a proof of Compactness Theorem, up to Łoś Theorem.
Corollary 3.64. (Compactness Theorem, c.f. [BBHU08] Thrm 5.8) Let
L be a language based on a continuous structure. Let T be an L− theory
which conditions satisfy the hypothesis of Theorem 3.62 and C be a class of
L-structures. Therefore, if T is finitary satisfiable in C, then there exists a
D-product of structures in C that is a model of T.
Proof. Let Λ be the collection of all finite subsets of T. By hypothesis,
given λ ∈ Λ with λ := {E1 , ..., En}, there exists some Mλ ∈ C such that
Mλ |= Ek for all k ∈ {1, ..., n}.
Fixed an L-condition E ∈ T, define S(E) := {λ ∈ Λ : E ∈ λ}.
Notice that S(E1) ∩ ... ∩ S(En ) 6= ∅ ( λ := {E1 , · · · , En} ∈ S(E1 ) ∩
... ∩ S(En )), therefore {S(E) : E ∈ T} satisfies the Finite Intersection
Property. Q
Let D be an ultrafilter over Λ extending {S(E) : E ∈ T}.
Let M := λ∈Λ Mλ be the respective D-product of the sequence of Lstructures (Mλ )λ∈Λ . Given E ∈ T, where E : ψ = 0 (ψ an L-sentence).
Notice that for all λ ∈ S(E) we have that Mλ |= E; i.e., ψMλ = 0. Since
by construction S(E) ∈ D, by Łoś Theorem (Theorem 3.62) it follows
that limλ,D ψMλ = 0. Notice that ψM = limλ,D ψMλ = 0, so M |= E.
Therefore, M |= T.
Keisler showed in [Kei64] the existence of saturated structures by
using ultraproducts. In [BBHU08], there is a proof of an analogous
result to Keisler’s construction by using metric ultraproducts. In the
following lines, we provide a proof of this result in the logic propposed in this paper, supposing that V is provided with the symmetric distance (abusing of the notation, we will denote by dV ) and
co-divisible value co-quantale satistying the SAFA Property (Definition 2.27).
Definition 3.65. Let L be a language based on a continuous structure,
Γ (x1 , ..., xn) be a set of L-conditions and M be an L-structure. We say that
Γ (x1 , ..., xn) is satisfiable in M, if and only if, there exist a1 , ..., an ∈ M
such that M |= E(a1 , ..., an) for all E(x1 , · · · , xn ) ∈ Γ (x1 , ..., xn).
Definition 3.66. Let L be a language based on a continuous structure ,
M be an L-structure and κ be an infinite cardinal. We say that M is κsaturated, if and only if, given A ⊆ M such that |A| < κ and Γ (x1 , ..., xn)
a set of L(A)-conditions with parameters in A, it holds that if Γ (x1 , ..., xn)
is finitely satisfiable in M then Γ (x1 , ..., xn) is satisfiable in M.
Definition 3.67. An ultrafilter D over K 6= ∅ is said to be countablyincomplete,
if and only if, there exists some {An : n ∈ N} ⊆ D such that
T
n∈N An = ∅.
CO-QUANTALE VALUED LOGICS
35
Proposition 3.68. Let D be an coutably-incomplete ultrafilter over K 6= ∅,
then there exists a countableTsubcollection {Jn : n ∈ N} of D such that
Jn+1 ⊆ Jn for all n ∈ N and n∈N Jn = ∅.
Proof. By hypothesis,
there exists a subcollection {An : n ∈ N} ⊆ D
T
of D tal que n∈N An = ∅. Define J0 := A0 and Jn+1 := Jn ∩ An+1 for
any n ∈ N \ {0}.
Proposition 3.69. (c.f. [BBHU08]; Prop. 7.6) Let V be a compact, Hausdorff, co-divisible value co-quantale satisfying SAFA, provided with the
symmetric distance dV . Let L be a countable language based on a continuous structure and D be a countably-incomplete ultrafilter over a non empty
set Λ. Given any Λ-sequence of L-structures (Mλ )λ∈Λ , its D-product MD
is ω1 -saturated, assuming that all L-formulae satisfy the hypothesis in Łoś
Theorem (Theorem 3.62).
Proof. For the sake of
Qsimplicity, let us analyze L-conditions with one
variable x. Let A ⊆ λ∈Λ Mλ be countable and Γ (x) be a set of L(A)conditions with parameters in A which is finitely satisfiable in MD .
We will prove that Γ (x) is satisfiable in MD .
Since L is countable, let Γ (x) := {ψn (x) : n < ω} be an enumeration
of Γ (x). Since D is contably-incomplete, by Proposition 3.68 there
exists a sequence
(Jn )n∈N of elements in D such that Jn+1 ⊆ Jn for all
T
n ∈ N and n∈N Jn = ∅.
By
Q hypothesis, for all k ∈ N the set {ψ1 (x), ..., ψk(x)} is satisfiable
in λ∈Λ MλQ
. By Łoś Theorem (Theorem 3.62), there exists some a :=
(aλ )λ∈Λ ∈
λ∈Λ Mλ such that for all n ∈ {1, · · · , k} we have that
MD
+
λ
ψn Mλ (a) = limλ,D ψM
n (aλ ) = 0. Therefore, given ǫ ∈ V we may
λ
say that {λ ∈ Λ : dV (0, ψM
n (aλ )) ≤ ǫ} ∈ D. Since dV is the symmetric
λ
distance, by Remark 2.36 we have that {λ ∈ Λ : ψM
n (aλ ) ≤ ǫ} ∈
D. Since V satisfies SAFA Property (Definition 2.27), there exists a
sequence (uk )k∈N in V such that
V
(1) n∈N un = 0.
(2) for all n ∈ N, 0 ≺ un
(3) for all n ∈ N, un+1 ≤ un
λ
Therefore, {λ ∈ Λ : ψM
u } ∈ D for all l ∈ N. This implies that
n V(aλ ) ≤
Wk l
Ak := {λ ∈ Λ : Mλ |= x∈Mλ n=1 ψn (x) ≤ uk+1 } ∈ D whenever k ∈
N. Define the sequence (Xn )n∈N of elements in D as follows: X0 := Λ
V
W
and Xk := Jk ∩ Ak if k ∈ N \ {0}. Notice that x∈Mλ kn=1 ψn (x) ≤
W
Vk+1
⊆ An for all n ∈ N and
x∈Mλ
n=1 ψn (x) ≤ uk+2 ≤ uk+1 , then An+1 T
then Xn+1 ⊆ Xn for all n ∈ N. Notice that n∈N Xn = ∅, therefore
if λ ∈ Λ there exists kλ ∈ N such that kλ := max{n ∈ N : λ ∈ Xn }.
36
D. REYES AND P. ZAMBRANO
Q
Define a := (aλ )λ∈Λ ∈
λ∈Λ Mλ as follows: In case that kλ = 0 ,
take
a
as
any
element
in
Mλ , otherwise take aλ ∈ Mλ such that
W Mλ
λ
{ψn (aλ ) : n ≤ kλ } ≤ ukλ . So, if k ∈ N then for any n ∈ N such
λ
that k ≤ n and λ ∈ Xn we have that n ≤ kλ , hence ψM
k (aλ ) ≤ ukλ ≤
unQ. Since Xn ∈ D, by Łoś Theorem (Theorem 3.62) we have that
M
MD
ψk λ∈Λ λ (a) = limλ,D
Qψk (aλ ) = 0. Since ψk (x) ∈ Γ (x) was taken
arbitrarily, then a ∈ λ∈Λ Mλ realizes Γ (x), as desired.
R EFERENCES
[AB09]
C. Argoty and A. Berenstein. Hilbert spaces with unitary operators.
Math. Log. Q., 55:37–50, 2009.
[BBHU08] I. BenYaacov, A. Berenstein, C.W. Henson, and A. Usvyatsov. Model
theory for metric structures. In Model Theory with Applications to Algebra
and Analysis (London Mathematical Society Lecture Notes Series), volume
349, pages 315–427. Cambridge University Press, 2008.
[BH04]
A. Berenstein and C.W. Henson. Model theory of probability spaces
with an automorphism, 2004. arXiv:math/0405360v1.
[CK66]
C. C. Chang and H. J. Keisler. Continuous model theory. Princenton University Press, 1966.
[FK97]
R. Flagg and R. Kopperman. Continuity spaces: Reconciling domains
and metric spaces. Theoret. Comput. Sci., 177:111–138, 1997.
[Fla97]
R. Flagg. Quantales and continuity spaces. Algebra universalis, 37:257–
276, 1997.
[FZ80]
J. Flum and M. Ziegler. Topological model theory. In Lecture Notes in
Mathematics, volume 769. Springer-Verlag Berlin, 1980.
[HH09]
A. Hirvonen and T. Hyttinen. Categoricity in homogeneous complete
metric spaces. Arch. Math. Logic, 48:269–322, 2009.
[HI02]
C.W. Henson and J. Iovino. Ultraproducts in analysis. In Analysis and
Logic, vol 262 of London Mathematical Society Lecture Notes Series, pages
1–115. Cambridge University Press, 2002.
[Iov99]
J. Iovino. Stable Banach spaces and Banach space structures, I: Fundamentals. In X. Caicedo and C. Montenegro, editors, Models, Algebras,
and Proofs. Marcel Dekker, New York, 1999.
[Kei64]
H.J. Keisler. Ultraproducts and saturated models. Indag. Math., 26:178–
186, 1964.
[Law73] F. W. Lawvere. Metric spaces, generalized logic, and closed categories.
Rendiconti del seminario matematico e fisico de Milano, 43:135–166, 1973.
[LRZ18] M. Lieberman, J. Rosicky, and P. Zambrano. Tameness in generalized
metric structures, 2018. preprint, https://arxiv.org/abs/1810.02317,
submitted.
[Pre88]
M. Prest. Model theory and modules. In London Mathematical Society
Lecture Notes Series, volume 130. Cambridge University Press, 1988.
[SS78]
S. Shelah and J. Stern. The Hanf number of the first order theory of
Banach Spaces. Trans. Amer. Math. Soc., 244:147–171, 1978.
[Zam11] P. Zambrano. Around Superstability in Metric Abstract Elementary Classes.
PhD thesis, Universidad Nacional de Colombia, 2011.
CO-QUANTALE VALUED LOGICS
37
Email address: davreyesgao@unal.edu.co
D EPARTAMENTO DE M ATEMÁTICAS , U NIVERSIDAD N ACIONAL
AK 30 # 45-03 CÓDIGO POSTAL 111321, B OGOTA , C OLOMBIA .
DE
C OLOM -
BIA ,
Email address: phzambranor@unal.edu.co, phzambranor@gmail.com
D EPARTAMENTO DE M ATEMÁTICAS , U NIVERSIDAD N ACIONAL
AK 30 # 45-03 CÓDIGO POSTAL 111321, B OGOTA , C OLOMBIA .
BIA ,
DE
C OLOM -