You are on page 1of 6

A natural interpretation of fuzzy set theory

Mamoru SHIMODA Shimonoseki City University Shimonoseki, Japan mamoru-s@shimonoseki-cu .ax:.jp


Abstract
We present a new and natural interpretation of fuzzy set theory in a cumulative Heyting valued model for intuitionistic set thoery. By this interpretation we can consistently get various notions and properties of fuzzy sets, fuzzy relations and fuzzy mappings. We can consider notions such as operations of fuzzy subsets of different universes, fuzzy relations and mappings between fuzzy subsets. and intuitionistic negation are the most natural basic set operations, and max-min composition is the most natural operation for composition of relations. We distinguish a generalized fuzzy set, a fuzzy subset of a crisp set (called a universe for the fuzzy subset), and a membership function of a fuzzy set on a crisp set (universe). There is a natural correspondence between fuzzy subsets of a crisp set and mappings from the crisp set to H, which preserves inclusion and basic set operations. We can naturally define basic set operations of fuzzy subsets of different universes and define inclusion relation between them. Thus we obtain a general theory of fuzzy sets where fuzzy subsets of different universes are treated in a natural and uniform way. We can also define relations and mappings between fuzzy subsets of crisp sets, and give characterization of defining properties of equivalence relation, partial order, and linear order. There have been proposed several definitions of fuzzy mappings or fuzzy functions, one of which identifies fuzzy mapping with fuzzy relation ([l, 2 , 4, 7,8, 10, 111). Our interpretation of fuzzy mappings seems to be unique and quite natural. The extension principle of Zadeh is expressed as a theorem on images of mappings for natural extensions of mappings between crisp sets. More details can be seen in [13, 141, where the extention of fuzzy set operations discussed in Subsection 3.2 is not treated.

As far as fuzzy sets and fuzzy relations are considered as extensions of crisp sets and relations, this interpretation seems to be most natural.
1. Introduction
In ordinary fuzzy set theory, fuzzy sets axe identified with mappings into the unit interval [0, 1 1 of

real numbers, and the propeties or operations of fuzzy seta and relations are defined by equations or inequalities ([17,18]).Various definitionshave been proposed on the basic operations such as intersection, union, complement of fuzzy sets, and composition and inverse of fuzzy relations etc. (cf. e.g., 13, 4, 5, 6, 7, 8,11)). Sometimes only the defining equations are presented without explanation. We interpret fuzzy sets and fuzzy relations in the model V H introduced by the author [12],a Heyting valued model for intuitionistic set thoery, where N is a complete Heyting algebra and is considered the set of truth values in the model. V H is a kind of so-called sheaf model, cumulatively constructed by transfinite iteration of power sheaf construction over H.In the interpretation the canonical embedding from the class of all crisp sets into the model plays an important role. Though the model is similar to the sheaf models in [9,15,16]and so on, our interpretation is original and unique.

2. The Heyting valued model


2.1. Basic set constructions

In the model we can easily define basic notions and


operations of sets and relations. By the canonin d embedding we can obtain most of the standard defining equations of basic notions and operations of fuzzy sets and relations. It shows that min, max

Let H be a complete Heyting algebra with operations and constants A, V, A, V, +,7 ,0 , 1, and I. We construct an H-valued model for the extended intuitionistic set theory, which is a first order intuitionistic logic with predicates E,= and E (called an existence predicate) together with axioms of intuitionistic set theory. Let V be the class of d l crisp sets and On be the class of all ordinals.

0-7803-7@78-3/0l/$10.00 (C)u)ol IEEE.

Page: 493

Definition 2.1. The H-valued model V H is constructed as follows. For every ordinal a,Vf is defined by induction:

bH = 9,
V .=

U VF
P<a

(if a is a limit ordinal),


: !Du+H,

The assignment from each crisp set to its check set is called the canonical embedding. For a bounded ,a,) and crisp sets 2 1 ,.. . ,2n , formula cp(a1, cp(zl,...,zn) holds iff II~J(Z~,.-. , 2 , ) 1 1 = 1, and l(~( lzn) ~ holds l, iff * Ilp(Zi,... * * ,Zn)[l = 0.

v,H,,= { U = (IuI,Eu); 1.1


'Ducvf,
ThenVH=

Pairs, couples (ordered pairs), (cartesian) products, unions, and power sets in V Hare defined as follows. Definition 2.4. Let u,v E Vu.

EUEH, lul(z) 5 Eu A E z (Vz E Du)}.

U V,".
&On

(1) l ) { ~ , v = } ~ { u , ~ )E{u,v}" , = Eu A Ev,

We identify1 . 1 with U. For a sentence cp of V H , the Heyting value llpll E H is defined as follows.

{u,v}" :2 t--) 1. (2) ( U , V ) H = ttulu}Hl t % V } H ) H . (3) D(u E v ) = { ( 2 ,y)" ;2 E Du, y E Dv},


E(u
U ) = Eu A E v ,

The basic set operations are defined as follows.

Definition 2.5.

Let u,v E VH.

(1) D(u nxv)= 'Du U 'Dv, E(u nHv) = Eu A Ev, U nHv :z I + llz E U I A ~1 1 2E ~ 1 1 . (2) !D(uUHv) = 'Du U 'Dv, E(u U " V ) = Eu V Ev, u U H v:2 t ) 1. E U11 v 1 1 2E VI[. (3) D(u\"v) = 'Du U Dv, E(u\"v) = Eu V Ev,
U \ ~ V:z c-) 1 1 3 :E

uII A

llz E U[/.

a sentence cp of V*, cp is valid in V H if IIpII = 1. W e say u is a subset ofv in V H and write u E v if


IIU

An element of the model is called a set in VH.For

The superscripts and commas may be omitted. A l l axioms of (extended) intuitionistic set theory are valid in VH.

2.2. Relations in the model

c~

I= I

A
ZCDU

(+I + 112E VI)

= 1.

If U E w and (U w). If U

E U , we say u and v are similar v and Eu = Ev, U and v are said to be equivalent (U w U).
v

A relation in V H is a subset of a cartesian product in VH. For R,u,v E Vn,R is a relation from u to v in V H if R is a subset of U x v in V H ,that is, R g u x v. We often write zRy instead of (zy)E R. For each set U in VHl the identity relation 1, on U
is defined bv:

Definition 2.3. For every crisp set 2 in V, the check set j! E V His defined recursively by:
D(5) = { f i ; ~ E z},

D(Iu)= ((2z);zE Du), E&) = Eu, 1, :(m) c-)U ( . ) .

Ei:= 1, i: : fi C ) 1.

0-7803-7078-3/ol/$lO.Ml (C)u#)l IEEE.

Page: 494

Definition 2.0.

Let R, S E V H . (1) The composition S o R is defined by:

D(S 0 R ) = { ( z z ) ; X , z E V,"}, E(S 0 R ) = ER A ES, S o R : (zz)c)IlEly(sRy A ySz)II,


where Q is an ordinal satisfying 'DRU IDS E V,". ( 2 ) The inverse relation R-' is dehed by:

relation R ' o R is symmetric, and in addition if 2 is total R-' o R is reflexive, and if it is univalent R-' o R is transitive.

2 . 3 . Mappings in the model A mapping in V H is a total and univalent relation in V H .For U , U,f E V H ,f is called a mapping from U to v in V H if f is a total and univalent relation from U to v in v H . We write f : U +v an V H if f is a mapping from U to v in V H .Hence for every f in V H ,f :U 3 v in V Hiff

D(R-') = { ( z y ) ; z , yE V,"}, E(R-') = E R , R-' : (4 I ) llY&Il, where Q is a suitable ordinal satisfying DR C V z .


The definitions are not affected by the choice of the ordinal a. If R and S are relations from U to II and from v to w respectively, then S o R is a relation from U to 20 and R-' is a relation from v to U . Deflnition 2.7. Let R be a relation from U to v in V*. (1) For A E V H ,the image R(A) is defined by:

f C U x U, I, E f '

of,

and f o f '

I,.

We often write f(z)= y instead of ( z y ) E f or zfy. An injection (resp. a sujection) is an injective (resp. surjective) mapping, and a bijection is an injective and surjective mapping. A composition of injections (resp. surjections or bijections) is also an injection (resp. a surjection or a bijection). For a relation f from U to v , f is an injection iff I , fi: f ' o f and f o f ' E I,, is a surjection iff I , E f-'of and fof-' R. I,, and f is a bijection iff I , fi: f ' 0 f and f o f ' M I,. Hence for a mapping f : U 3 v in v H , f i s injective iff I , R. f - l o f and is surjective iff f o f ' R. I,. For a mapping f in VH, f ' o f is an equivalence relation. We can define images and inverse images of m a p pings similarly to those o f relations. Most of the basic properties of relations and mappings of usual set theory also hold in the model.

D(R(A))= Dv, E(R(A))= ER A EA, R(A) :y t3 \1(3z E A)zRyll.


(2) For B E V H , the inverse image R-'(B) is defined by:

D(R-'(B)) = 'Du, E(R-'(B)) = E R A EB, R-'(B) : z c-)11(3y E B)zRyll.


A relation R from U t o v in V w is called total if (Vz E u)(3y E v)(zRy) is valid, and is surjective if (Vg E v)(3z E u)(zRy)is valid. R is injective if (Vz,y E u)(Vz E v)(zRzA yRz -P = y ) is valid, and if (Vz E u)(Vy,z E v)(zRg A zRz -P y = z ) is valid, R is called univalent. Then R is total 8 I , C R-' 0 R, R is surjective iff I, C R o R-', R is injective iff R-' o R E I,, and R is univalent iff R o R-' C I,.
A relation R on U (that is, a relation from U to U ) is called wflemke if (Vz E u)(z&) is valid. Similarly R is symmetric if VxVy(xRy 3 y & ) is valid, is tnrrnsitiue if VxVyVz(zRyA y R z + zRz) is valid, is a n t i s y " e t r i c i f V z V g ( z R y ~ y R z 3 2 = y) is valid, and is connected if ( V z E u)(Vy E ~ ) ( z RVyU&) is valid. Then R is reflexive 8 I, E R, is symmetric 8R-' C R i f fR-' kr R, is transitive iff R o R C R, is antisymmetric 8 RnR-' E I,, and is connected iff U x U R U R"l.

3. The natural interpretation


3.1. Natural interpretation of fuzzy sets

In the following, let X, Y, 2 be crisp sets.


. 1 . Every set in V H is an H-fuzzy Deflnition 3 set. For an H-fuzzy set A, the mapping

PA =

:x+H;

2c )1 1 5

E All

is called the membership function of A on X . An H-fuzzy subset of X is a subset of X in V H .

Obviously A is an H-fuzzy subset of X iff A for every A E V H . Lemma 3 . 1 .

Cx

Every mapping from X to H is the membership function of some H-fuzzy subset of X .

An equivalence relation is a reflexive, symmetric, and transitive relation, an order relation is a reflexive, antisymmetric, and transitive relation, and a h e a r oder is a connected order relation. Hence a relation R on U is an equivalence relation iff I , 5 R, R-' E R, and R o R E R. If R is a

Hence there is a natural correspondence between H-fuzzy subsets of X and mappings from X to H. The natural correspondence preserves inclusion (order) and the basic set operations.

Page: 4%

Definition 3.2. Let p, U : X +H . (1) p U iff p ( z ) 5 Y ( Z ) for all z E X. (2) The mappings P A Y , p V u , p\u, y p : X+H a m defined by:
p A U :z
p

Definition 3.4. For p : X H , Y : Y +U , themappings p A u , p V u , p \ u o n X U Y = Z a r e defined as follows.

+p ( z ) A ~(z),

v U : z c)p ( 2 ) v U@),
c )-9p(z).

p\~:~t)p(~)A-u(z),
- p :5

= (P12)A c1 v v = (P tZ) v (:U tZ), c1\ = ( P tZ) \ . (


c1 A

Yz),
-+ H,v : Y -+HI and
(Z

/a
x nY ) eY ) . ( 2 E x \ Y) ( Z E x n Y)
E EX

x e Y = (x\ Y) u (Y\ x).


0

Lemma 3.4.

Let p : X

Theorem 1 . Let A, B be H-fuzzy sets and all the membership functions below be on X .

p(z) (1) p * Y : z c r { A U(Z)


(2) p v v : z *

(Z

(1) If A and B are H-fuzzy subsets of X I then


A E B iff P A 5 pB, and A B iff P A = p ~ .
N

(2)

PAnB

= P A A CIS,

PAUS

PA\B

= P A \ PB,

PW\A

= P A v pS1 =?PA.

P(4 p(z) v

44

Y(Z)

( z E Y \X).

P(2) (2 E p(z) A ~ ( z ) ( Z E 0 (z E

x\Y) x nY )

Y \ X).
and

3.2. Extension of fuzzy set operations

Proposition 3.5. u:Y +H.

Assume that p : X

+H

Now we consider the relation between H-fuzzy sets and mappings from different crisp sets to H.

For p : X

+H,define p r y : Y 3 H by:
pA
U (resp. p V U) works as the inhimum (resp. supremum) of p and U in the set of equivalence classes with respect to N.

p [Yis called the restriction of p on Y. Obviously IL = P a d ( P t y ) t z = P W n 2).

tx

Definition 3 . 3 . For p : X +H, U : Y -+ H, the relations p 5 U and p 11 U are defined as follows.

P3V
P H U

Theorem 2. Let A, B be sets an Vw,p~ be the membership function of A on X , and p~ be the membership function of B on Y.

P S VtX, pduanduip.

(1) A n x E B n y i f l p ~ 5 p E , and A nX -B np iflp~ N ~ B .


(2) In addition, assume that A and B are H-fuzzy subsets o f X and Y respectively. Then

The relation 5 is a preorder compatible with the equivalence relation N. If X = Y, then p 5 U becomes p 5 u and p N u becomes p = Y . Lemma 3 . 2 . Let p : X

+H and u : Y +H.

C -

B i f fP A 5 PS, and B i f l p A N- p ~ .

Theorem 3 . Assume A, B are H f u z z ysubsets of X,Y respectively, and P A , C(B are the membership functions on X and on Y respectively. Let p ~ m ~A , LJB PA\B , be the membership functions of An B , AU B, A\ B on X U Y reappeetively.
(1) pAnB = P A A P E (2) PAUB = PA V PS.

(3)

PA\B

=PA \ PE-

Now we extend Definition 3.2(2) and define operations of mappings from diflerent sets to H.

Hence there is a natural correpondence between Hfuzzy subsets of merent crisp sets and mappings from those crisp sets to H, which preserves order and operations.

0-7803-7078-3/0l/$lO.MI (C)zoOlIEEE.

Page: 496

3.3. Relations between fuzzy subsets

Every relation in V H is called an H-fvzzy relation. Definition 3.5. For R,u,v E V H ,R is an H fuzzy relation from U to v if it is a relation from U to v in V H . For every R E V H ,the membership function of R from X to Y is the membership function of R on X x Y, hence it is the mapping
PR :

Proposition 3 . 8 . Let A, B be H-fuzzy subsets of X , Y respectively, R be an H-fuzzy relation from A to B , and P A , p~g, p~ be their membership functions on X , Y,X x Y respectively. (1) R is total ifl V y E Y p ~ ( ~ =y ~ ) A ( x for ) all
2

x X y +H ; ( x y )

c-) II(zY)" E

RII.

An H-fuzzy relation from X to Y is an H-fuzzy subset of X x Y .

ally E Y . (3) R is injective i f fp ~ ( x zA ) ~ R ( Y Z> ) 0 implies x = y for a l l x , y E X and z E Y . (4) R is univalent i f f~ R ( X Y A ) ~ R ( x z> ) 0 implies y = z forallxEX andy,tEY.

E (2) R is sujective iff v S E x p ~ ( x y= ) ~ B ( Y f)o r

x.

If A, B are H-fuzzy subsets of X, Y respectively, then obviously an H-fuzzy relation from A to B is an H-fuzzy relation from X to Y . For every set R in V H ,R is an H-fuzzy relation from A to B iff
p R b y ) 5 P A (2)A p B (y) for all 2

x,y E y,

where P A , p ~p~ , be the membership functions on X , Y , X x Y respectively. Proposition 3.6. Let A, B , C be H-fuzzy subsets of X , Y , Z respectively, R be an H-fuzzy relation from A to B , S be an H-fuzzy relation from B to C , and P R , ps, P S ~ R~, R - Ibe the membership funcX x Z, Y x X respectively. tions on X x Y,Y x 2, (1) The composition S o R is an H-fuzzy relation from A to C , and for all x E X , z E Z pSoR(XZ) =
YY

Theorem 4. Let A be an H-fuzzy subset of X , R be an H-fuzzy relation on A, and pa, p~ be their membership functions on X , X x X respectively. ( 1 ) R is reflexive i f f p ~ ( ~= 5~ ) A ( x f)o r all x E X . (2) R is symmetric i f l p ~ ( z y = ) ~ R ( Y X f)o r a11 2, yE ( 3 ) R is tmnsitive ifl p ~ ( t yA )~R(YZ) 5 p~(Zz) f o r all x,y , z E X . (4) R is antisymmetric ifffor all x,y E X, p ~ ( 2 yA) p ~ ( y > ~ 0) implies x = y . ( 5 ) R is connected i f ff o r all 5 , y E X , pR(XY) v PR(YX) = P A ( X ) A p A ( 9 ) -

x.

If A = X and H = [0,1], the conditions are almost same to the usual definitions in fuzzy literature, but the condition ( 5 ) and the definition of linear order is stronger than ordinary definitions.
3.4. Fuzzy mappings and extension principle

( P R ( 2 Y ) A PS(Yz)).

(2) The inverse relation R ' is an H - f u z q relation @m B to A, and for all x E X , y E Y


p R - l ( ! / z ) = pR(XY).

These are extensions of the defining equations of max-min composition and those of inverse for ordinary fuzzy relations.
Proposition 3.7. Let A, B be H-fuzzy subsets of X , Y respectively, R be an H-fuzzy relation from A to B, and p~ be its membership function on X x Y . ( 1 ) For every C E V H ,the image R(C) i s a subset of B a n V H ,and for all y E Y
pR(C)(Y) =
2EX

Every mapping in V His called an H-fuzzy mapping. Definition 3 . 6 . An H-fuzzy mapping from U to v is a mapping from U to v in VH.An H-fizzy mapping from x to Y is a mapping from X to P in VH. Obviously an H-fuzzy mapping f r o m U to v is a total and univalent H-fuzzy relation f r o m U to U. Theorem 5. Let A, B be H-fuzzy subsets of X , Y respectively, and P A , ps be their membership function on X , Y respectively. Suppose f E V H and 1c, = pf i s its membership fvnctaon o n X x Y . Then f is an H-fuzzy mapping fi-om A to B i f $ satisfies the following three conditions:

( P C ( z )A P R ( X Y ) ) ,

where pc and ~ R ( c ) are the membership functions on X and on Y respectively. (2) For every D E V H ,the inverse image R-'(D) is a subset of A in V H ,and for all x E X
PR-I(D)(X)=

(R)$(xY) 5 P A ( z ) A P B ( Y ) f o r all 2 E x, E yt (E) '$'(zY) = p A ( X ) f o r all 2 E x,

YY

V ( P D ( Y )A I L R ( X Y ) )
VEY

(U) $(xy) A $(zz) > 0 implies y = z for all x E X , and y, z E Y.


Conversely, i f a crisp mapping $ : X x Y 3 H satisfies the three conditions, there i s an H - f u z q mapping f from A t o B whose membership function on X x Y is identical with $.

where p~ and ~ R - I ( D ) are the membership functions on Y and o n X respectively.

0-7803-7@78-3/Ol/$lO~~ ( C ) U ) o l IEEE.

Page: 497

Proposition 3 . 9 . Let A, B be H-fuzzy subsets of X , Y respectively, f be an H-fuzzy mapping from A to E , and pf be its membership function on X x Y . (1) For every C E V H ,the image f ( C ) is a subset of B in V H ,and for ally E Y
Pf(C)(Y)= =EX

[5] J. Fodor and R. R. Yager, "Fuzzy set-theoretic operators and quantifiers," Fundamentals of f i z z y sets, Kluwer Academic Publishers, Dordrecht, 2000, pp 125-193.

( P C ( 4 A Pf

(4),

[6]S. Gottwald, "F'uzzy uniqueness of fuzzy mappings," Fuzzy Sets and Systems, Vol. 3, 1980, pp 49-74.
[7]S. Gottwald, Fuzzy Sets and fizzy Logic: Foundations of Applimtions - from a Mathematical Point of View, Vieweg, Wiesbaden, 1993.

where p c and pf(c) are the membership functions on X and on Y respectively. (2) For every D E V H ,the inverse image f-'(D) i s a subset of A in V H ,and for all x E X
Pf-1(&)

DEY

(PdY) A Pf (SY)),

[SI S. Gottwald, "Fundamentals of fuzzy relation calculus," Fuzzy Modelling: Paradigms and Pmctice, Kluwer Academic Publishers, Boston, 1996,pp 25-47.
[9] R. J. Grayson, "Heyting-valued models for intuitionistic set theory," Applications of sheaves (Lecture Notes in Math. 753),Springer, Berlin, 1979,pp 402414.

where p~ and Pf-l(D) are the membership f i n c tions on Y and on X respectively.

Theorem 6. Let cp :X
Then the check set

--+

Y be a crisp mapping.

i s a n H - f i z z y mapping from

X to Y , and the followings hold. (1) For every A E V H ,the image $(A) is an H fuzzy subset of Y , and for all y E Y
&(A)(y)

9(=)=v

2EX

[lo] W. C. Nemitz, "Fuzzy relations and fuzzy functions," &zy Sets and Systems, Vol. 19, 1986,pp 177-191. [ll] S. V. Ovchinnikov, "Structure of fuzzy binary relations," f i z z y Sets and Systems, Vol. 6, 1981,pp 169-195.

where PA and are the membership functions on X and on Y respectively. (2) For every B E V H ,the inverse image @-'(B) i s an H-fuzzy subset of X , and for all x E X

[12]M.S b o d a , "Categorical aspects of Heytingvalued models for intuitionistic set theory," Comment. Math. Univ. Sancti Pauli, Vol. 30,

where p~ and p p i ( ~ an? ) the membership functions on Y and on X respectively.

1981, DD ._ _ 17-35. [13]M. Shmoda, "A natural interpretation of


fuzzy sets and fuzzy relations," submitted,

(1) shows that the extension principle of Zadeh holds.

1998. [14]M. Shimoda, "A natural interpretation of fuzzy mappings," submitted, 2000.
[15] G. Takeuti and S. Titani, "Heyting valued universes of intuitionistic set theory," Logic Symposium Hakone 1979, 1980 (Lecture Notes in Math. 891), Springer, Berlin, 1981, pp 192-

References [l] S . S. L. Chang and L. A. Zadeh, "On fuzzy


mapping and control," IEEE h n s . on Systems, Man, and Cybernetics, Vol. 2, 1972,pp

30-34. [2]M. Demirci, Fuzzy functions and their fundamental properties, h z y Sets and Systems Vol. 106,1999, pp 239-246.

306. [16]G. Takeuti and S. Titad, "Fuzzy logic and a t h .Logic, Vol. 32, fuzzy set theory," Arch. M 1992,pp 1-32. [17]L. A. Zadeh, "Fuzzy sets," Infomation and Control, Vol. 8, 1965,pp 338-353. [18]L. A. Zadeh, "Similarity relations and fuzzy orderings," Information Sciences, Vol. 3, 1971, pp 177-200.

[3]D. Dubois, W. Ostasiewicz, and H. Prade, "F'uzzy sets: history and basic notions," findamentals of f i z z y sets, Kluwer Academic Publishers, Dordrecht, 2000,pp 21-124.

[4]D. Dubois and H. Prade, M

y Sets and Systems: Theory and Applicatio~zs,Academic Press,New York, 1980.

0-7803-7078-3/0V$l0.00 (C)zoOl IEEJL

Page: 498

You might also like