5.1. Fuzzy Set Theory in FTT
The fuzzy set theory can be easily formulated using the means of the fuzzy type theory. A fuzzy set in a universe , is obtained by the interpretation of some formula . Thus, if a model and an assignment p to variables are given, then the interpretation of in is the function , which is a fuzzy set in the universe.
To simplify the explanation, we will not distinguish between a fuzzy set represented by a formula and its interpretation as a fuzzy set. Hence, by abuse of language we will simply say “a fuzzy set ” and not “a formula whose interpretation is a fuzzy set in the universe ”.
Several basic formal definitions of operations on fuzzy sets are the following:
where
.
Note that the empty and universal fuzzy sets and have a type. This means that for each type there are different empty and universal fuzzy sets. This is correct because fuzzy sets are, in fact, functions, which holds also for these special fuzzy sets being identified with constant functions having all values equal to 0 or 1.
The formula
expresses that the fuzzy set
is nonempty. The complement
of a fuzzy set
is represented by a formula
We also need to specify crisp sets: a fuzzy set is
crisp if it can be characterized by a formula
The formula
assigns a truth value to any fuzzy set
. Hence, saying that
is crisp means that
, which means that the following is provable:
Clearly, is true in any model in which the membership degree for all .
Lemma 2. For all fuzzy sets:
- (a)
and.
- (b)
.
- (c)
.
- (d)
Proof. Let .
(a) This immediately follows from and .
(b) From the definition, we obtain
using the properties of FTT. But since the formula
is crisp, we have
. From this, using Rule (R) and generalization, we obtain
from which
follows.
(c) From Lemma 1g it follows that
from which (b) follows using generalization.
(d) The proof can proceed semantically: let be an arbitrary model and p an assignment. Then is a crisp set in which the membership degree of any element is either 1 or 0.
Let us denote . Then iff, either or , iff . □
A
singleton (of type
), i.e., a one-element set containing an element of type
, is specified in FTT as follows:
In words: the interpretation of (
14) is a (fuzzy) set of all elements
equal in the degree 1 to a given element
. By the separation of
(interpretation of ≡), all such elements are classically equal to
u, and hence every interpretation of
is a one-element set
.
Lemma 3. Let be a formula defining a singleton set and be a formula of type α.
- (a)
.
- (b)
.
Proof. - (a)
is obvious.
- (b)
This follows from the following sequence of provable formulas:
, and
.
□
5.2. Transfer of Selected Concepts of AST into Fuzzy Set Theory
In this and the following subsections, we will translate several concepts developed in AST and rough set theory into the language of the fuzzy type theory. The outcome is a unified formulation of similar concepts from different theories. Then, when choosing a proper model, we immediately obtain the theory of rough fuzzy sets. Moreover, we will also see that there is a close parallel between topological concepts developed in AST based on the indiscernibility relation and the basic concepts of rough (fuzzy) set theory. This is especially interesting if we realize that AST has been developed earlier than rough set theory and arises from foundations very different from those of the former.
Let us extend the language of FTT by a new fuzzy equality , , that fulfills the following axioms for all :
- (EV1)
,
- (EV2)
,
- (EV3)
,
- (EV4)
.
Axioms (EV1)–(EV3) are the standard axioms of any fuzzy equality. Axiom (EV4) says that the fuzzy equality is separated. In the sequel, we will omit the type at the symbol ≈ and assume that it is always clear from the context.
To simplify the explanation, we will not introduce a special theory but write and understand that is provable in some theory, in which, at least, axioms (EV1)–(EV4) are valid.
Definition 3. Letbe fuzzy sets (formulas) andan element.
- (i)
A figure of a fuzzy set
is defined by - (ii)
A monad
of an element u is defined by - (iii)
A property characterizing a fuzzy set to be a figure is represented by the formula
By
-conversion, a
figure of is a fuzzy set
At the same time, by
-conversion, a
monad of is a fuzzy set
We say that a fuzzy set
is a
figure (Note that in the fuzzy set theory, such a fuzzy set is called
extensional (w.r.t. ≈).) if
Lemma 4. Letand.
- (a)
and .
- (b)
.
- (c)
, i.e.,is a figure.
- (d)
.
- (e)
Ifthen.
- (f)
.
Proof. (a) After rewriting, we obtain
Due to the definition of ∅, the formula on the right-hand side is equivalent to ⊤, from which (a) follows. For the proof is analogous.
(b) This follows from
(c) Using
-conversion, we obtain from (
17) that
must be provable. The right-hand side of (
18), however, is equivalent to
which is provable using (EV3), the properties of &, and quantifiers.
(d) We start with axiom (EV3):
Using generalization, the properties of Δ and quantifiers we obtain
Conversely, by [
12] (Lemma 3(a)), we can prove the opposite implication. Then
and by generalization and [
3] (Theorem 11), we obtain
After substitution of the definition of singleton we obtain
which is (d) after substituting the definition of a figure.
(e) follows immediately from the definition of and the properties of strong conjunction &.
(f) follows from the provable formula
by the properties of FTT (
-abstraction). □
Lemma 5. Letbe a fuzzy set. The the following is equivalent.
- (a)
.
- (b)
.
- (c)
.
Proof. (a)⇒ (b): Applying (
17), we obtain
(b)⇒ (a): Since
, using the properties of FTT we can prove that
. Joining this and (
19), we obtain the first equivalence after application of the properties of FTT.
The equivalence between (b) and (c) follows by the properties of FTT if we realize that . □
Lemma 6. Letbe a fuzzy set.
- (a)
iff.
- (b)
.
Proof. (a) Let
. Then
from which
i.e.,
. The converse inclusion follows from Lemma 4(c).
The converse implication follows from Lemma 4(c) and Rule (R).
(b) This follows immediately from Lemma 4(c) and (b). □
Lemma 7. , i.e., a monad of an elementis a figure.
Proof. By Lemma 6(a) we must prove that
After substitution of the definitions of monad and figure, we have
which using [
12] (Lemma 3(a)) gives (
20). □
Definition 4. Letbe fuzzy sets (formulas) andan element. Then the following concepts can be introduced:
- (i)
Separability of two fuzzy sets is characterized by the formula - (ii)
are separable if.
- (iii)
A special case is separability of an element u from X: - (iv)
Closure of a fuzzy set
X is a fuzzy set
Remark 3. - (a)
Formula (21) means that if fuzzy setsare separated, then if u belongs toin a non-zero degree and v belongs toin a non-zero degree then they cannot be equal in a non-zero degree. Interpretation of this formula, however, can be many-valued, i.e., we can have two fuzzy sets separable only by some degree. Full separability is obtained if provability ofis assured—cf. item (ii). - (b)
Formula (22) is, in fact, different from Formula (21). The special case of the latter is. For obvious reasons, however, we will use the same symbol both for separability of two fuzzy sets and separability of an element from a fuzzy set, if no misunderstanding can occur.
If follows from (22) that - (c)
Closure of X is a fuzzy set of elements, to which there are elementsfrom the figure, and an elementfrom the monad ofthat are fuzzy equal in a non-zero degree.
The following is immediate.
Lemma 9. Letand.
- (a)
, i.e.,is a figure.
- (b)
.
- (c)
.
- (d)
.
- (e)
If X is a figure then.
We start with the following provable formula (based on the transitivity of ≈, the properties of FTT and quantifiers):
where the right-hand side of the implication is equivalent to
. The left-hand side is obtained from the provable implication
in which the provable formula
was used in the left-hand side of (
27). Joining the latter with (
26) we obtain the implication
which, after applying generalization, is equivalent to (
25).
(b) Using Lemma 4(b) and the properties of FTT, we obtain the provable formula
Applying two times substitution to the right-hand side of this implication and using transitivity of ⇒, we obtain the formula
Finally, using generalization we obtain a formula that is equivalent to (b).
(c) is a consequence of (a) and Lemma 6.
(d) holds if , where is a variable.
The inclusion right to left follows from (b). The opposite inclusion is equivalent to
Using (c) and Rule (R), (
28) becomes
Using prenex operations, we can rewrite (
29) into
Let us denote
. To prove (
30), we start with the provable formula
Now, we can consider formulas
to be equivalent to ⊤ since otherwise they can be equivalent to ⊥ and, consequently, (
30) is trivially provable. Using the property
and the properties of quantifiers, we obtain from (
31) the formula
After realizing equivalent substitutions, we have
where
. We conclude from (
32) that
Renaming bound variables in (33) and adding the formula
which proves (d).
(e) follows immediately from Lemma 6 using Rule (R). □
Definition 5. Letbe fuzzy sets (formulas) andan element.
- (i)
- (ii)
A fuzzy set Y is dense
in X if
Theorem 2. Letand. Then
- (a)
- (b)
.
Proof. (a) follows from Definition 5(i) using Lemma 5.
(b) is obtained by rewriting (a). □
By this theorem, the interior of a fuzzy set X is obtained as a union of all monads contained in it.
Lemma 10. Letand. A fuzzy set Y is dense in X ifand Proof. This can be obtained by detailed rewriting of Definition 5(ii). □
5.3. Rough Fuzzy Sets in FTT
We can define rough fuzzy sets using the formalism of FTT. In the resulting theory we obtain generalization of the original rough set theory.
Definition 6. The following formulas define special properties of fuzzy sets.
- (i)
Upper approximation of a fuzzy set: - (ii)
Lower approximation of a fuzzy set:
When realizing that a monad of
in (
16) is just an equivalence class of
with regard to ≈, Formulas (
36) and (
37) are just Formulas (
7) and (
8) rewritten in the language of FTT. This becomes obvious also from the following lemma.
Lemma 11. Letand.
- (a)
.
- (b)
.
We conclude that Definition 6 indeed defines the concepts of upper and lower approximation of Definition 2 in the formalism of FTT.
Lemma 12. Let.
- (a)
, i.e.,is a figure.
- (b)
, i.e.,is a figure.
Proof. (a) We want to prove that
First note that, after rewriting and by the properties of quantifiers, we obtain
Furthermore, using the transitivity of ≈ and the properties of implication, we can prove that
Using generalization and quantifier properties, we obtain
From this, using generalization twice and (
39), we obtain (
38).
(b) We start with the following provable formula:
Furthermore,
. By transitivity of implication and formal adjunction, we obtain from these two formulas that
Finally, by generalization and properties of quantifiers we obtain
which is (b). □
Theorem 3. Letand.
Then Proof. - L.1
- L.2
- L.3
□
By this theorem, we see that the concepts of upper approximation from the rough set theory and that of a figure of a (fuzzy) set X from AST are equivalent. It is important to emphasize, however, that the motivation of both concepts is different. While in rough set theory, the goal was to introduce an approximation of a set using an equivalence relation, in AST, the goal was to characterize shapes of objects using the indiscernibility relation. Unlike rough set theory, the latter is an infinitary concept in AST. Namely, it is a mathematical model of the situation in which originally different points begin to merge if any imaginable crisp criteria to discern the objects fail.
The following lemma shows that the basic known properties of rough sets are universally valid.
Lemma 13. Let.
- (a)
and.
- (b)
and.
- (c)
.
- (d)
.
- (e)
.
- (f)
.
- (g)
and.
Proof. (a) Both inclusions follow from the substitution axioms and the reflexivity of ≈, namely
:
(b) follows from (a). The other equality follows from , where the right-hand side is equivalent to ⊥.
(c) follows from Theorem 3 and Lemma 4(f).
(d) will follow if we prove that
This formula is implied by the following two provable implications using the quantifier properties:
(g) The first formula follows from the provable formula
which is obtained after applying
(cf. [
12]).
The second formula is proved analogously. □
Theorem 4. Letbe a fuzzy set and, i.e., it is a figure..
- (a)
,
- (b)
Proof. (a) by Lemma 13(a).
Let
be a figure and
. Then
From this we obtain
using the properties of quantifiers and other properties of FTT. The latter implies the lemma.
(b) by (a), . By Theorem 3, . If X is a figure then by Lemma 6. Using Rule (R), we have and thus . □
In the same way as in rough set theory we can introduce the fuzzy boundary region and the interior of a fuzzy set.
Definition 7. Let. A boundary region
of X is the fuzzy set It is easy to prove the following.
Lemma 14. Let.
- (a)
If X is a figure, i.e.,, then.
- (b)
Proof. This is obtained by rewriting the corresponding formulas and using the properties of quantifiers. Namely, for all
, we obtain the following sequence of provable formulas:
The lemma is then obtained by -abstraction. □
Theorem 5. Let. Then Proof. By Lemma 13(a),
. Let
. Then
, from which
. On the basis of this and using definition of lower approximation, we derive the following sequence of equivalences:
The theorem follows from the last equivalence using Lemmas 6 and 12. □
Thus, by this theorem the concepts of interior and lower approximation are equivalent.
Theorem 6. Let. Then the following is equivalent.
- (a)
.
- (b)
.
- (c)
.
Proof. (a)⇒ (b): Then . By the properties of FTT, this implies that , which gives (b) by Lemma 13(a).
(b)⇒ (a) follows from the definition difference of fuzzy sets and the fact that .
(b)⇒ (c): Let . By Lemma 13(a), this implies that . Furthermore, by Theorem 3, which implies that using Rule R and we conclude that from which follows by Lemma 6(a).
(c)⇒ (b) follows from Lemma 14(a). □
By this theorem, figures have an empty boundary. Following the rough set theory, a fuzzy set is rough if its boundary is non-empty (
), i.e., if it is not a figure.
5.4. Model
Let us demonstrate the above notions on a model of FTT. We will construct a model
(cf. (
12)) where the algebra of truth values
is the standard ukasiewicz MV
Δ-algebra and, furthermore,
where
,
, is the operation of biresiduation in
, and
is the fuzzy equality (
11). The sets
are supposed to contain all functions
so that all formulas
are assured to have a value in
(Note that such a definition of a model of FTT generalizes the concept of a safe model of fuzzy predicate logic introduced by Hájek in [
9].).
For simplicity, we will put where p is an assignment of values from to variables. Hence, axioms (EV1)–(EV4) are trivially true in and, therefore, it is a model of our theory. Fuzzy sets on arbitrary universe are functions from the set that are obtained as interpretation of certain formulas .
Let us define the function
by
where
are parameters. The function is a simple trapezoidal function depicted in
Figure 1.
Proposition 1. Letbe a formula whose interpretation in the modelis a fuzzy set with the membership function, provided thatas well as. Thenis a figure in the model, i.e.,.
Proof. We must check that for all
,
To prove this inequality is a tedious but straightforward task. □
Let us consider a fuzzy set .
- (i)
Let the interpretation
(a set). Then interpretation
of a figure of
X is a fuzzy set depicted in
Figure 2.
- (ii)
Let the interpretation
be
where
. This fuzzy set is depicted in
Figure 3. Then interpretation
of a figure of
X is a fuzzy set depicted in
Figure 4.