You are on page 1of 28

Galois Connections in Category

Theory, Topology and Logic


W. Gahler
Abstract
The notion of a Galois connection is important in different branches of
mathematics. It even is used for defining basic notions in several theories.
In this paper the role of Galois connections is demonstrated in reference to
known results and moreover in presenting new ones.
AMS Mathematics Subject Classification: 06D35,18B35,03E12.
Key words: Galois connection, Non-classical logic, MV-algebra, Partially
ordered monad, Fuzzy filter.

1. Different kinds of Galois connections


Standard kind. G. Birkhoff ([1]) introduced a notion of Galois connection, related to power sets. Later on 0. Ore ([15]) and J. Schmidt
([18]) studied a generalization of this notion, which can be considered
as the standard kind of a Galois connection. Its "covariant form" is
defined as follows: A pair of isotone mappings f : (X, :S) -----t (Y, :S)
and g : (Y, :S) -----t (X, :S) with (X, :S) and (Y, :S) partial ordered sets, is
a Galois connection provided the equivalence f( x ) :S y -:::::::? x :S g(y)
holds for all x E X and y E Y . (!,g) is a Galois connection if and
only if for each y E Y the supremum U{x EX I j(x) :S y} exists and
coincides with g(y). Therefore, g will be called the sup-inverse of f.
Categorical notions of Galois connections. The following kind
of Galois connections, pointed out by H. Herrlich and M. Husek ([10]),
is more general. In this case instead of isotone mappings f : (X, :S
) -----t (Y, :S) and g : (Y, :S) -----t (X, :S) two concrete functors <p : A -----t B
and 'ljJ : B -----t A are given, where A and B are concrete categories over
425
K. Denecke et al. (eds.), Galois Connections and Applications, 425-452.
2004 Kluwer Academic Publishers.

426

W. Gahler

a category X. The pair (r.p, 1/J ) is called a concrete Galois connection


provided the following equivalence holds
r.pA

B E MorB

~ 'lj;B

E MorA

for A E ObA, B E ObB and f E MorX. We obtain the classical


kind if we take (X,~) and (Y, ~) as the concrete categories A and
B, respectively, with X and Y the sets of objects and the partial
orderings of X and Y as the related sets of morphisms and interpreting
X trivially as a category having only one object and only the identity
morphism.
A further kind of Galois connection, also discussed in [10], is that
of an adjoint situation, which consists of a pair (<p, 1/J ) of covariant
functors <p : A ---+ B and 1/J : B ---+ A and two natural transformations
'T] : idA ---+ 1/J o <p and c : <p o 1/J ---+ id 8 (the unit and counit of the adjoint
situation) such that idcp = Ecp o <p'TJ and id'l/1 = 'lj;E: o 'Tl'l/1

2. Galois connections and partially ordered monads


There are interesting applications of the standard kind of Galois connections in general topology. Two examples related to partially ordered monads will be presented in this section.
Partially ordered monads. In the general theory of convergence
structures the notion of a partially ordered monad is basic. It is defined
by means of the notion of an almost complete semilattice. By an
almost complete semilattice we mean a poset (X , ~) in which all nonempty suprema exist. Infima of subsets, even of pairs {x, y } , need not
exist.
Let acSLAT denote the category of almost complete semilattices, where
the morphisms are the mappings between almost complete semilattices
which preserve non-empty suprema. By a partially ordered monad
(over SET) (cf. [3, 6]) we mean a quadruple <P = (r.p, ~' ry , p,) which
fulfills the following conditions:
<P consists at first of a covariant functor (r.p, ~) :SET---+ acSLAT, X H
( r.pX, ~) with r.p : SET ---+ SET the underlying set functor. Moreover
<I> consists of two natural transformations 'T] = ('TJx )x EObSET and p, =
(p,x )x EObSET of mappings 'T]x : X ---+ r.pX and f.Lx : r.pr.pX ---+ r.pX
respectively, such that the triple (r.p, ry, p,) is a monad over SET. Finally
we assume:

Galois Connections in Category Theory, Topology and Logic

427

(MO) rpX is empty in case X is empty.


(Ml) For any set X and each pair of different elements x and y of X,
the infimum of rJx(x) and TJx(Y) does not exist.
(M2) For all mappings f, g: Y ------t rpX from f ~ g it follows J-Lx o rpf ~
J-Lx o rpg, where ~ is defined argumentwise with respect to the
partial ordering of rpX.
(M3) For each set X, J-Lx: (rprpX,
suprema.

~) ------t

(rpX , ~)preserves non-empty

The partial orderings ~ of the sets rpX are considered as finer relations. For each set X, the elements of rpX are called rp- objects on X,
and the minimal elements of rpX also ultra objects.
For each non-empty subset A of rpX the supremum V M exists.
MEA

Hence, whenever for a subset A of rpX a lower bound in rpX exists,


the infimum A M of this subset exists.
MEA

For each non-empty set X we denote the supremum

V TJx(x)

by

xEX

TJx[X]. A rp-object M on X for which M ~ TJx[X] holds, is called


stratified.
A rp-object Mona set X for which M < TJx(x) holds for some x EX,
is called a microobject at x. Because of condition (Ml), x is uniquely
associated to M. Of course, microobjects are special stratified <pobjects.

Proposition 1 ([6]) If there is a microobject at an element x of X,


then there is a microobject at any element y of X.
Microobjects are in some sense properly finer than points. They may
exist or may not. If there are no microobjects, then condition (Ml)
can be formulated as follows:
(Ml') For each set X, TJx : X ------t rpX is an injection and all values
TJx(x) are ultra objects on X.
The classical case of a partially ordered monad is that of the partially
order-ed filter- monad (F, ~' 7], J-L), which is given as follows: F is the

428

W. Gahler

filter functor-, which assigns to each set X the set FX of all (proper)
filters on X. ::; indicates that the sets FX are equipped with the finer
relations of filters, that is , the inversion of the inclusion. ry and p, are
natural transformations consisting of all mappings 'r/x : X --+ FX and
fJ,x: FFX--+ FX respectively, where for each x EX, 'r!x (x) = {M ~
X I X E M} and for each filter .c on FX , Mx(.C) =
M. In

un

AE.CMEA

this classical case microobjects do not exist.


In the general case of a partially ordered monad <]) = (<p, ::; , ry , p,) there
are the following examples of Galois connections.
Example 1. Let f : X --+ <pY and e : Y --+X be mappings such that
foe= ryy. Then the sup-inverse of P,x o<p f : (<pX ,::; )--+ (<pY ,::; )
exists and is denoted by
<p;;: (<pY, ::;) --+ (<pX, ::;).
Example 2. For each surjection f : X --+ Y , the sup-inverse of the
acSLAT-morphism <pf : (<pX , ::;) --+ (<pY , ::;) exists, denoted by <p- f :
(<pY, ::;) --+ (<pX, ::;).

Partially ordered submonads. Let<]) = (<p, ::; , ry , p,) be a partially


ordered monad. By a partially ordered submonad of <]) we mean a
partially ordered monad W = (<p' , ::; , ry' , p,') such that

(1) (<p', ::;) :SET--+ acSLAT is a subfunctor of (<p, ::;)


(in particular, for each set X and each non-empty subset A of
<p'X the suprema of A with respect to (<p'X , ::; ) and to (<pX, ::;)
coincide).

(2) (<p', ry' , p,') is a submonad of (<p , ry, p,).


An example always exists: the partially ordered submonad of <]) of
all stratified <p-objects. For the related subfunctor <p1 of <p we have
<p'X = {ME <pX I M::; 'r/x[X] }.

3. Some topological notions


Let <]) = (<p, ::;, 7], p,) be a partially ordered monad.
A <P-converyence str-ucture on a set X is a subset T of <pX x X such
that, writing M--+ x instead of (M , x) E T , we have

(Cl) 'r!x(x ) --+ x for all

.T

EX,

Galois Connections in Category Theory, Topology and Logic

429

(C2) M ---+ x and N::; M imply N---+ x, and


(C3) M---+ x implies M V TJx(x) ---+ x.
A <P-convergence structure Tis called a <P-limit structure if instead of
(C3) the following condition is fulfilled:
(C3') M ---+ x and N---+ x imply M V N---+ x.
In the filter case the <P-convergence structures are the convergence
structures in sense of D. C. Kent, introduced first in [13], and the
<P-limit structures are the limit structure in the usual sense, proposed
first by H.-J. Kowalsky in [14].
A mapping p : X ---+ rpX is called a <P-pretopology on X, provided that
TJx (x) ::; p( x) holds for all x E X. p can be interpreted as the special
<P-limit structure T = { (M, x) I M ::; p(x) } . For each <P-pretopology
p : X ---+ rpX, nb : rpX ---+ rpX defined by
nb = J1x o rpp,

is called the neighbourhood operator of p.


A <P-pretopology p is called a <P-topology provided that nb o p = p or,
equivalently, nb o nb = nb holds, that is, nb is idempotent.
Let T be a <P-convergence structure on a set X and let t 1 and t 2 be
the first and second projection ofT, respectively.

rpX

rpX

Fig. 1. The projections t1 and t2 ofT


and related cp-morphisms

For the mapping e : X ---+ T, defined by e(x) = (TJx(x), x) for all


x E X, we have h o e = TJx. Because of Example 1 of a Galois
connection the sup-inverse rp~t 1 of J1xorpt 1 exists. Since t 2 is surjective,
because of Example 2 of a Galois connection moreover the sup-inverse

430

W. Gahler

cp- t 2 of cpt 2 exists. We introduce the closure operator cl : cpX ---+ cpX
ofT ([3]) by means of the sup-inverse cp;t 1 defining

cl is a hull operator, that is, M ::::; elM holds for all M E cpX . If
M = elM, then M is called closed. Since cl is a hull operator, we
have that the infimum of any set of of closed cp-objects on X is closed,
as far as this infimum exists.
As neighbourhood operator nb : cpX ---+ cpX of the <I>-convergence
structure T we mean the neighbourhood operator of the associated <I>pretopology p ofT, which is defined by p(x) = V M for all x E X.
M--tx

nb is also a hull operator and can be introduced by means of the


sup-inverse cp- t 2 ([6]) by

This shows that the notions of a closure operator and of a neighbourhood operator are in some sense dual. More exactly, here we can
speak of an associated pair of Galois connections. In general topology both notions of closure operator and of neighbourhood operator
are used for different reasons. This differs from the classical case of
a topological space and even of a pretopological space in which each
of the notions of a closure operator and of a neighbourhood operator determines uniquely the other one. This simple situation in the
classical case depends on the fact that for the interior and the closure of subsets of a space we have X\ intM = cl(X \ M), where
intM = { x E X I M E p( x)}. Because of this reason, classical pretopological spaces are sometimes called closure spaces. That in general
such a one-to-one correspondence does not exist, will be demonstrated
at the end of this paper by examples of different <I>-pretopologies which
have one and the same closure operator.
In the following let T be a <I>-convergence structure on a set X and let
nb be its neighbourhood operator. A cp-object M on X is called open
provided that M = nbM.

Proposition 2 ([6]) The set 0 of all open cp-objects on X fulfills the


following condition

431

Galois Connections in Category Theory, Topology and Logic

(0) 0 is closed with respect to all non-empty suprema and all infima,
as far as these infima exist.
Each if>-topology p can be characterized by the set 0 of all open <pobjects on X by taking

(1)

p(x) =
MEO,ryx(x):SM

for all x EX.

Proposition 3 ([6]) Let 0 be a subset of <pX which fulfills the condition (0) and let p be the mapping defined by (1). Then the following
are equivalent:

(a) p is a if>-topology with 0 the set of all open <p-objects on X.

(b) For each M

E <pX we have

nbM =

1\

N.

NEO,M:SN

4. Basic structures in non-classical logics


There is a close relation between the general topology and the different
types of non-classical first order predicate logics, given by means of
the associated basic structures of these logics. Concerning this logics
we refer to [11].

Frame case. A first kind of basic structures is given by frames . A


frame is a non-degenerate complete lattice (L , ~) which fulfills the
complete distributivity law

i EI

i EJ

Non-degenerate means that the minimal element 0 and the maximal


element 1 differ.
In a frame (L, ~), by means of the implications

a ----+ y = max { x E L

I a 1\ x

y }

W . Gahler

432

for each a E La Galois connection is given, consisting of the mappings


fa : L ---+ L and ga : L---+ L , where fa(x) = a 1\ x and ga(Y) = a---+ y.
Clearly, fa(x) :S y +-+ x :S ga(Y) holds for all x , y E L.
Quantale case. A structure more general than that of a frame , is
that of a commutative quantale (L, :::; , *) (cf. Rosenthal's paper [17]) .
It is defined as a non-degenerate complete lattice which is equipped
with a commutative and associative binary operation * such that the
complete distributivity law

iEI

iEI

holds. We always assume that the maximal element 1 of the underlying


lattice is the unit element with respect to *
Analogously as in case of a frame , in a commutative quantale implications a---+ bare defined and Galois connections are introduced. The
only difference is that 1\ is replaced by *, that is, here we have

a ---+ y = max { x E L

I a * x :S

y }

and for the mappings fa : L ---+ L and ga : L ---+ L, given by fa (x) = a*X
and ga(Y) =a---+ y, we have fa(x) :S y +-+ x :S ga(Y) for all x , y E L.
Clearly, a frame is a commutative quantale with 1\ : L 2 ---+ L the
related binary operation.
Further examples of commutative quantales are given by means of the
continuous t-norms * : 2 ---+ L. They are defined as follows:
(T1) L is the unit interval [0, 1] equipped with the usual ordering.
(T2) * is a continuous commutative and associative binary operation
with unit element 1.
(T3) * is order preserving, that is, a :::; band c :::; dimply a* c :::; b *d.
Because of the continuity of*, for at-norm the complete distributivity
law with respect to * holds.
We note some special examples:
Godel t-norm
a* b =min{ a, b}
Product t-norm
a * b = a b (product of reals)
Lukasiewicz t-norm
a* b = max{O , a+ b- 1}

Galois Connections in Category Theory, Topology and Logic

433

Case of Girard monoids. A commutative quantale which fulfills the


condition of double negation -.(-a) =a, is a complete Girard monoid.
Note that the negation of a is given by a = a --t 0.
For the Godel t-norm and the product t-norm we have
if a> 0
if a= 0.
Hence, for these t-norms the condition of double negation is not fulfilled and they therefore do not define Girard monoids.
Case of MY-algebras. By a complete MV-algebra we mean a complete Girard monoid (L , ::; , *) which is divisible, that is, for all a, b E L
from a ::; b it follows a = b * c for some c E L.
Examples of complete MY-algebras:
(1) The Lukasiewicz t-norm *defines a complete MY-algebra ([0, 1],::;
'* ).
(2) All complete Boolean algebras are complete MY-algebras. They
can be characterized as those complete MY-algebras (L, ::;, *)for which
* is idempotent, that is, a * a = a for all a E L. In this case, then *
equals 1\.
Related logics. In the following there are listed basic structures
together with related non-classical logics:
intuitionistic logic
Frames
Commutative quantales
monoidallogic
Complete Girard-monoids - linear logic
Complete MY-algebras
Lukasiewicz logic
The following is well-known in first order logic:
Proposition 4 ( [11]) A formula is valid with respect to all nondegenerate complete MV-algebras if and only if it is valid with respect
to ([0, 1], ::;, *) with* th e Lukasiewicz t-norm.
Proposition 5 ([16]) A formula is valid with respect to all nondegenerate complete Boolean algebras if and only if it is valid with
respect to the Boolean algebra consisting only of 0 and 1.

434

W. Gahler

5. Fuzzy filters in the frame case


There are different kinds of fuzzy filters. The following deals with one
of these kinds.
If (L, :S::) is a complete lattice, then for each set X the mappings f E
Lx are called fuzzy subsets of X. For each a E L let a denote the
constant fuzzy subset of X with value a.
Let L = (L, :S::) be a non-degenerate frame. A fuzzy filter on X in the
frame case, also called an L-filter, is a mapping M : Lx --* L such
that
(Fl) M(O) = 0 and M(I) = 1,
(F2) M(f

1\

g)= M(f)

1\

M(g) for all f,g

Lx.

In the following, a definition of the related partially ordered fuzzy


filter-monad (FL, :::;;, ry, J-t) , also called the partially ordered L-filter monad, is given.
For each set X, FLX consists of all L-filters on X. The partial ordering
on each set FLX is given by

M :S:: N ~ M(f)

:::=:

N(f) for all f E Lx.

For each mapping f : X --* Y, the mapping FLf : FLX --* FLY
assigns to each M E FLX the fuzzy filter FLf(M) on Y given by
FLf(M)(g) = M(g of) for all g E Lx .
For each set X, each x E X and f E Lx let 'TJx(x )(f) = f(x) , and
for each E FLFLX and f E Lx let J-tx(.C)(f) = .C(ef) , where ef :
FLX--* L is the mapping M r-+ M(f).
We distinguish some types of fuzzy filters. An L-filter M on a set X
is called
bounded
if M(a) :::;; a holds for all a E L and
homogeneous
if M(a) =a holds for all a E L.
As a specialization of the general property of being stratified, here we
have that an L-filter on a set X is
stratified
if and only if M(a) :::=: a holds for all a E L.
Each fuzzy filter coarser than a bounded fuzzy filter is also bounded.
Each fuzzy filter finer than a stratified fuzzy filter is also stratified.

435

Galois Connections in Category Theory, Topology and Logic

Proposition 6 There are three partially ordered submonads of (FL , ::;


, TJ, f.L), defined by all bounded fuzzy filters , all homogeneous fuzzy filters
and all stratified fuzzy filters, respectively.

The following is related to suprema and infima of fuzzy filters .

Proposition 7 Let X be a set. For the supremum of a non-empty


subset A of FLX we have

(v

M)(f) =

M EA

A M(f)

(2)

M EA

for all f E Lx . If the infimum of a subset A of FLX exists, then

(A

M)(f) =

MEA

h 1\ 1\ fnS!

Ml , ... ,MnEA

for all f E LX.

6. Principal fuzzy filters in the frame case


Let X be a non-empty set. For each mapping
with a ::; Uf by

E Lx

and each a E L

[/, a[(g) = { ;

iff g '
if f ::; g =!= I ,
if g =I

for all g E Lx, a bounded fuzzy filter [f, a] is defined, called a principal
fuzzy filter on X (first defined in [4]).

Proposition 8 For each bounded fuzzy filter M in the frame case


the infimum M 1\ TJx[X] exists, called the homogenization of M.
Since for each f E Lx the principal fuzzy filter [f, Uf] is bounded,
its homogenization [ f , Uf) 1\ TJx[X] exists. For any f E Lx and all
g E Lx we have [f](g) =
V (Uf 1\a) vng . Clearly, [OJ= TJx[X].
f /\a5:c g

W. Gahler

436

In the following some representations of fuzzy filters by means of principal fuzzy filters are given.
Proposition 9 ([8]) Let X be a set of at least two elements. Then
faT each fuzzy filter M on X we have:
M is the infimum of all principal fuzzy filters [f, a] coaT"ser than M ,
that is, of all principal fuzzy filters [f, a] for which a ::; M (f) 1\ U.f
holds:

1\

[f, a].

a ~M(f) 1\ Uf

Moreover then faT each stmtified fuzzy filter M on X we have:


M is the infimum of all fuzzy filters [.f] which are coarser than M ,
that is, of all fuzzy filters [.f] for which U.f ::; M (f) holds:

M=

1\

[f].

Uf~M(f)

Proposition 10 ([8]) Let X be a set of only one elem ent, then we


have:
Each bounded fuzzy filter on X can be represented analogously as in
Pmposition 9 by means of pTincipal fuzzy filters.
Non of the stmtified, non-homogeneous fuzzy filters on X can be represented in this way.
For any non-empty set X we have [ 0]

1\ [.f, nf ].

/ ELX

7. Fuzzy filters in the quantale case


In the following let L = ( L , ::;, *) be a non-degenerate commutative
quantale. A fuzzy filt er on a set X in the quantale case, also called an
L-filter, is a mapping M : Lx -+ L such that

(Ql) M(O)

= 0 and

M(l ) = 1.

(Q2) f ::; g implies M(J) ::; M(g) for all f ,g E L x.


(Q3) M(J) * M(g) ::; M(J *g) for all f ,g E L x .

Galois Connections in Category Theory, Topology and Logic

437

If* is the infimum 1\ in (L , ::;), then the notion of a fuzzy filter in both
the frame case and in the quantale case coincide. Hence, the quantale
case is more general than the frame case.
In the quantale case, the related partially ordered fuzzy filter monad
(.ri, ::;, ry, fJ), also called the L-filter monad, is defined in the same way
as in the frame case. In particular, the L-filters Tlx(x) : Lx ---* L are
the same mappings as in the frame case.
Moreover, the notions of bounded, homogeneous and stratified fuzzy
filter are defined in the same way as in the frame case. There exist the
partially ordered fuzzy filter submonads of (FL, ::;, ry, 11) defined by all
bounded, homogeneous and stratified L-filters, respectively.
For each set X the supremum of a non-empty subset A of FLX has
the same representation as in the frame case given in Proposition 7
by equation (2). It follows that for each non-empty set X , T]x [X] =
V Tfx(x) also is a fuzzy filter in the quantale case.

xEX

For the infima of fuzzy filters in the quantale case we have:

Proposition 11 Let X be a set. If the infimum of a subset A of FLX


exists, then

(A

M)(J) =

h fn S:: f

MEA

M1 , ... ,MnEA

for all f E Lx.

The following is an easy consequence of Proposition 11.

Proposition 12 Let X be a non-empty set and A a linearly ordered


subset of (FLX, ::;). Then the infimum 1\ M exists and for each
f E Lx we have

(A
MEA

M)(J)

MEA

M(J) .

MEA

A bounded fuzzy filter on a set X will be called a bounded ultra fuzzy


filter if there is no properly finer bounded fuzzy filter on X. Analogously the notions of homogeneous ultra fuzzy filter and of stratified
ultra fuzzy filter are defined.

438

W. Gahler

It follows by means of Zorn's lemma from Proposition 12 that:

Proposition 13 Let X be a non-empty set. In the quantale case we


have: Each bounded fuzzy filter has a finer bounded ultra fuzzy filter.
Analogously, each homogeneous and each stratified fuzzy filter has a
finer homogeneous and a finer stratified ultra fuzzy filter, respectively.

Remark. Clearly, this proposition also holds in the frame case. In


this case because of Proposition 8 we even have that each bounded
ultra fuzzy filter is a homogeneous fuzzy filter.

8. Principal fuzzy filters in the quantale case


In the quantale case we introduce a notion of principal fu zzy filter in
defining the mappings [f, a] : Lx ~ L exactly as in the frame case.
However only some of these mappings appear as fuzzy filters in the
quantale case. We have the following :

Proposition 14 For each set X , each f E Lx and each a :::; Uf


the mapping [f, a] is a fuzzy filter in the quantale case if and only if
f :::; f * f or a * a = 0 holds.
Clearly in the frame case we especially have f = f 1\ f for all f E Lx.
The following counterexample is obtained as an immediate consequence of Proposition 14.

Proposition 15 Let a and b be elements of L such that b * b < b <


a :::; a* a holds. Moreover, let X be a set of at least two elements and
x 0 an element of X. Then [f, a] with f E Lx defined by

f(x) = { :

if

X= Xo

if

X -::j::. Xo

for all x E X, is not a fuzzy filter in the quantale case.

Galois Connections in Category Theory, Topology and Logic

439

Under some conditions, in the quantale case, homogeneous fuzzy filters [f] exist.

* f or Uf *Uf = 0
holds and there is an x EX with f(x) = Uj, the fuzzy filter infimum
in the quantale case

Proposition 16 For any f E Lx for which f :::; f

[f]

[f,Uf] !\1Jx[X]

exists and for each g E Lx we have [ f ](g)

V (uf *a) v ng.


f*a~g

9. Distinguished and adjoint fuzzy filters, the frame


case
Let L = (L, :::;) be a non-degenerate frame. An L-filter M on a set X
is said to be distinguished ([9]) provided that

M(f

v g)

M(f)

v M(g)

holds for all f, g E Lx.


In the filter case distinguished means that this filter F is prime, that
is for non-empty subsets M and N of X from M UN E Fit follows
M E F or N E F. Hence, in this case distinguished means that F is
an ultra filter.
In the general frame case the property of a fuzzy filter to be distinguished is more complex. As will be shown in the example in Section
12, there are distinguished fuzzy filters which are neither homogeneous ultra fuzzy filters nor stratified ultra fuzzy filters and on the
other hand there are homogeneous ultra fuzzy filters which fail to be
distinguished.
In the following the zero-condition (Z) of L is important. (Z) means
that
a > 0 and b > 0 imply a!\ b > 0

for all a, b E L. This condition is needed for proving some types of


compactifications (cf. [9]). Clearly this condition is fulfilled in case

440

(L,

W. Gahler

~)

is a complete chain.

Proposition 17 ([9]) If L fulfills the condition (Z), then each stratified ultra fuzzy filter is distinguished.
If L is a non-degenerate complete chain, then each distinguished homogeneous fuzzy filter is a homogeneous ultra fuzzy filter.
Remark. ([9]) Propositions 13 and 17 imply that under the condition
(Z) each fuzzy filter has a finer distinguished fuzzy filter (which may
not be homogeneous).
In the following let an antitone involution c : L --t L be fixed. Clearly,
for all a, bEL we have a V b = c(c(a) 1\ c(b)).
By means of c each distinguished fuzzy filter M on X can be assigned
a further fuzzy filter Me defined by
Me(!) = c(M(c o !))

for all f E Lx. Me is called the c-adjoint of M.

Proposition 18 The c-adjoint of each distinguished fuzzy filter M is


also a distinguished fuzzy filter and we have (Me)e = M .
If M and N are distinguished fuzzy filters, then M ~ N implies
Me 2: Ne.
From the example in Section 12 we will obtain, that there are distinguished fuzzy filters which differ from their c-adjoints and also some
which do not.
If M and N are distinguished fuzzy filters and M = Ne , hence
also Me = N, then {M , N} will be called a c-adjoint pair. In case
M = Me, the fuzzy filter M will be called c-selfadjoint.

Proposition 19 Let M be a distinguished fuzzy filter. If one of the


fuzzy filter-s M and Me is bounded the other- one is stratified and vice
ver-sa. M is homogeneous if and only if Me is homogeneous.
Clearly, if we restrict ourselves to the partially ordered submonad of
all bounded fuzzy filters or of the partially ordered submonad of all

Galois Connections in Category Theory, Topology and Logic

441

stratified fuzzy filters, then for each non-homogeneous distinguished


fuzzy filter, its c-adjoint does not appear.

10. Distinguished and adjoint fuzzy filters, the


quantale case
In the following let L = (L , ~' *) be a non-degenerate commutative
quantale. If * differs from /\ , for defining the notion of distinguished
fuzzy filter in the quantale case always an antitone involution c : L -t
L is to be fixed.
If not specified we always assume in the following that such a mapping
c is given. Let E9 denote the binary operation on L defined for all
a, bEL by
a E9 b = c(c(a)

* c(b)).

If L is a Girard monoid, then as c we can take the negation -, : L -t L


If L = [0, 1], cis given by c(a) = 1- a and *is a continuous t-norm,
then E9 is the t-conorm associated to * Both notions oft-norm and tconorm, introduced by Schweitzer and Sklar ([19]) , play an important
role in fuzzy logic.
For a fixed antitone involution c of L, an L-filter on a set X is called
c-distinguished provided that
M(f E9 g) ~ M(f) E9 M(g)

holds for all f, g E Lx. Notice that in this definition only an inequality
appears, which is in accordance with the inequality in condition (Q3)
of the definition of a fuzzy filter in the quantale case.
Analogously as in the frame case, to each c-distinguished fuzzy filter
M can be associated a further fuzzy filter Me defined by M e(!) =
c(M(c of)) for all f E Lx. Me is called the c-adjoint of M. Propositions 18 and 19 hold analogously in the quantale case. The notions
c-adjoint pair and c-selfadjoint fuzzy filter are defined analogously as
in the frame case.
In the example which will be presented in Section 13 all c-distinguished
fuzzy filters M are c-selfadjoint.

442

W . Gahler

11. "Basic example"


In this section a special example of fuzzy filters in the frame case will
be considered. For different reasons, this example has been already
considered in [9] in the general case and at first , restricted to homogeneous fuzzy filters also in [2] . We are here especially interested in
the distinguished fuzzy filters in this example.
Let X = {0, 1} and L = {0, ~ ' 1} . There are 9 fuzzy subsets on X ,
and among them there are 6 fuzzy subsets which are non-constant.
They are presented in Fig. 2.

L J

fs

Fig. 2. All nonconstant mappings of X= {0, 1} into L = {0, ~ ' 1}.


There are 25 fuzzy filters in this example. They are shown in Fig. 3,
in which finer filters are situated more downwards. In particular there
are
9 bounded, non-homogeneous fuzzy filters , indicated by small circles,
11 homogeneous fuzzy filt ers, indicated by boldfaced dots, and
5 stratified, non-homogeneous fuzzy filt ers, indicated by small dots.
Whereas in this example all the principal fuzzy filters are the bounded,
non- homogeneous fuzzy filters , their homogenizations as well as H1s =
[!I] A [fs], H16 = [!I] A [!6], Hs2 = [is] A [!2], H52 = [!6]/\ [!2] are
the homogeneous fuzzy filters .
M1 = [!4]/\ [is], Mo1 = [fs] 1\ [!6]
Moreover, Mo = [h] A [!6],
and 5 0 = [!I] A [fs]A [f6], S1 = [!2] A [fs] A [!6] are the stratified,
non-homogeneous fuzzy filters.
Proposition 20 Th e following hold:
(1) M0 and M 1 are the only micro fuzzy filters and the only stratified
ultra fuzzy filters .
(2) TJx(O), 7Jx(1), H16 and Hs2 are the only homogeneous ultra fuzzy
filters.

Galois Connections in Category Theory, Topology and Logic

443

(3) In this example [h, 1], [!4 , 1], 7Jx(O), 7Jx(1), M0 and M1 are the
distinguished fuzzy filters, indicated in Fig. 3 additionally by overlines.
(4) {M 0 , [h, 1]} and {M 1 , [!4 , 1]} are adjoint pairs.
(5) Both fuzzy filter-s 1Jx(O) and 7Jx(1) are selfadjoint.
(0, 0]

TJx(

=[h]

Fig. 3. Graph of all fuzzy filters on X = {0, 1}


with L = {0, ~' 1} in the frame case.
Notice the following:
7Jx(O) and 7Jx(1) fail to be stratified ultra fuzzy filters.
H16 and H52 are homogeneous ultra fuzzy filters which are not distinguished.
The distinguished fuzzy filters [h, 1] and [f4 , 1] are neither homogeneous ultra fuzzy filters nor stratified ultra fuzzy filters.

12. Analogous lukasiewicz case


As in the preceding section let X = {0, 1} and let L be the set {0, ~, 1}.
We equip L with the usual ordering and with the binary operation *

444

W . Gahler

defined by

0 0 0 0
1
1
2 0 0 2
1 0 21 1

By a fuzzy filter here we mean an L-filter with L the quantale (L, :::; , *).
There are 22 fuzzy filters in this example. They are shown in Figure
4, where the finer fuzzy filters are situated more downwards. In particular, there are
11 bounded, non-homogeneous fuzzy filters, indicated by small circles,
and
11 homogeneous fuzzy filters , indicated by bold faced dots.
Stratified fuzzy filters which are non-homogeneous, in particular microfuzzy filters, do not exist.
From Proposition 14 it follows:
The mappings [!5, 1], [!6, 1], [f5] and [f6] are excluded as fuzzy filters with respect to (L , :::; , *).

Proposition 21

There are four bounded non-homogeneous fuzzy filters which are not
single principal fuzzy filters. They are:

834

[h, ~],

1\

[!4, ~]

B36

854

[f5, ~]

1\

[f4, ~],

B56

=
=

[/s, ~]
[f5, ~]

1\
1\

[!6, ~],
[!6, ~].

Notice that for the principle fuzzy filters [!,a] appearing in this example in some cases only the condition f :::; f * f (e.g. for [h, 1]) and
in some other cases only the condition a* a = 0 (e.g. for [h , ~]) is
fulfilled.

445

Galois Connections in Category Theory, Topology and Logic

[0,0]

'TJX (0)

=[h]

Fig. 4. Graph of all fuzzy filters on X = {0, 1}


with L = {0, ~' 1} in the Lukasiewicz case.

Among the principal fuz zy fil ters there are three which are homogeneous, that is [ 0] =
~] , [!I] = [/I ,~] and [h] = [h , ~ ].
The further homogeneous fuzzy filters are [h] = T,lx(O), [!4 ] = T,lx(1)
and

[J,

H12 = [!1]
H3

1\

[h, ~]

[h] , H14 = [!I]


1\

1\

[0], H4 = [j4 , ~ ]

[j4 , ~] ,

1\

H32 = [h, ~]

[OJ , H34 = [h, ~]

1\

1\

[h] ,

[!4,

The property of a fuzzy filter to be distinguished here we understand


with respect to the associated binary operation EB of * defined by
means of the antitone involution c with c(a) = 1 - a , that is we have
EB

1
2

1
2

1
2

1 1

1 1 1

446

W. Gahler

Proposition 22 In this example the ultra fuzzy filters 1Jx(O), 7Jx(l)


and H12are the only distinguished fuzzy filters (indicated in Fig. 4 additionally by overlines). They are selfadjoint.

13. Balanced fuzzy filters


Let L = (L, :::;) be a non-degenerate frame. An L-filter will be called
balanced ([9]) whenever a finer distinguished homogeneous fuzzy filter
exists.
All balanced fuzzy filters are bounded. In the "basic example" H16
and H52 are homogeneous fuzzy filters which are not balanced.
Proposition 23 ([9]) Let L be a non-degenerate chain. Then we have:

(1) All balanced fuzzy filters define a partially ordered submonad of


the partially ordered fuzzy filter monad.

(2) All homogeneous balanced fuzzy filters define a partially ordered


submonad of the partially ordered fuzzy filter monad.

14. Fuzzy convergence structures


In the following let L = (L, :::;, *) be a non-degenerate commutative
quantale and let <I> = ( cp, :::; , 1], J.L) be the partially ordered L- filter
monad or one of its partially ordered submonads. The special cases
we will consider are listed in the following table. We describe these
cases only by noting the related fuzzy filters.

Galois Connections in Category Theory, Topology and Logic

447

Fuzzy filters of <P

Case
Case
Case
Case
Case

(A)
(S)
(B)
(H)
(B')

all
all
all
all
all

fuzzy filters
stratified fuzzy filters
bounded fuzzy filters
homogeneous fuzzy filters
balanced fuzzy filters with
(L, :S:) a non-degenerate complete chain
and*= 1\
all homogeneous balanced fuzzy filters with
Case (H')
(L, :S:) a non-degenerate complete chain
and*= 1\.
By the interior operator of a fuzzy pretopology p : X ---+ <pX we mean
the mapping int : Lx ---+ Lx defined for all f E Lx and x E X by
(intf)(x) = p(x)(J).

(3)

According to the following proposition, the interior operator of a fuzzy


pretopology can be characterized independently of this fuzzy pretopology.
Proposition 24 The interior operator int of a fuzzy pretopology fulfills
the following conditions:
(1) intf :S: f holds for all f E Lx and inti= I.
(2) f :S: g implies intf :S: intg for all f, g E Lx.
(3) intf * intg :::; int(J *g) holds for all f , g E Lx .
On the other hand, each mapping int : Lx ---+ Lx which fulfills these
conditions, is the interior operator of the fuzzy pretopology defined by
equation (3). The interior operator int characterizes a fuzzy topology
if and only if additionally into int = int.

This proposition shows that fuzzy pretopologies and fuzzy topologies


can be characterized only by fuzzy sets. Clearly in this proposition
the partial ordering and the products of fuz zy subsets are defined
argumentwise.

448

W. Gihler

In the frame case conditions (2) and (3) are equivalent to the condition
(2)'

intf 1\ intg = int(f 1\ g) for all j, g E Lx.

In a <I>-pretopological space a mapping f is called open provided that


intf =f.
Proposition 25 The set

of all open fuzzy sets f E Lx of a <I>pretopology fulfills the condition

(0')

!1

is closed with respect to all suprema and all finite products


* * fn of fuzzy sets.

Each <I>-topology p can be characterized as a subset T of Lx which fulfills condition (0') by taking intf =
V g for all f E Lx.
g"S_ f,gET

Remark. In the frame case, (0') is a well-known condition for characterizing a <I>-topology. Notice that (0') and condition (0) in Proposition 3 look completely different.
Proposition 26 The set r of all open fuzzy sets of a fuzzy topology
equipped with the argumentwise defined partial ordering and the argumentwise defined product *, is a quantale.

Remark. Notice that for each subset M of r the supremum of


M in the lattice (r, :::;) is the argumentwise defined supremum x H
V f (x) (x E X) and the infimum of M in (r , :::;) is the interior of the
! EM

argumentwise defined infimum x

1\ f(x) (x

X).

! EM

16. Closure operator of a fuzzy convergence structure


Let L = (L, :=:;;, *) be a non-degenerate commutative quantale and let
<I> = (rp, :::;, ry, p,) be specified as in Section 15. Moreover, let T be a
<I>-convergence structure on a set X and let as usual t 1 : T ----+ <pX be
the first projection ofT.

Galois Connections in Category Theory, Topology and Logic

449

Proposition 27 The sup-inverse 'P;t 1 : (/)X ---+ (/JT of J.Lx o'Pt 1 : (/JT ---+
'PX is given by

(4)

for all M E 'PX and h E LT, wheTe joT each f E Lx, ef : 'PX ---+ L is
defined by ef (N) = N (f).
Because of this proposition the closure operator cl
given as follows.

'PX ---+ (/)X is

Proposition 28 For each M E (/)X and f E LX we have

(elM)(!)

(5)

N(h)NCJn) ~ f (x)
for all N -tx

In the following we assign to each finite subset { JI, ... , fn} of Lx


as its fuzzy set closuTe a fuzzy set cl (!1, ... , fn) E LX defined for all
X EX by

cl (h, , fn)(x)

V(N(h) * * N(.fn))

(6)

N-tx

Some properties of this fuzzy set closure are the following.

(1) h * * fn = 0 implies cl(JI,, ... , fn) = 0,


(2) h :::; 91, ... , fn:::; 9n implies cl(h, ... , fn) :::; cl(91, ... , 9n) and
(3) h * ... * fn :::; cl (!1, ... 'fn) and cl (!1, ... 'fn, 91, ... '9m) :::;
cl (h, .. . , fn) * cl (91, ... , 9m) holds for all JI, ... , fn, 91, ... , 9m E Lx.

By means ofthe notion offuzzy set closure, equation (4) can be written
as follows:

(ciM)(.f)

cl (JI ,.. ..Jn) :S;

M(h* *fn)
J

(7)

450

W. Gahler

Because of the result on finer ultra <p-objects in Proposition 13, we


have:

Proposition 29 The <p-objects N appearing in equations (5) and (6)


can be assumed to be ultra <p-objects.
The frame case. In the following we assume that * equals 1\. In this
case we have more results. Some are even more simple. For instance,
for all JI, ... , fn E Lx it follows that cl (JI , ... , j~) = cl (fi/\ .. .1\fn).
Hence, in the frame case, we only need the notion of fuzzy set closure
for single fuzzy sets. Thus instead of (6) we only need
(clf)(x)

V N(.f).

(8)

N -+ x

Proposition 30 In the frame case the equations (4) , (5) and (7) can
be reduced, respectively, to
<p~t1(M)(h)

(ciM)(.f)

e9 o

v M(g) ,
v M(g),
t1 ::; h

N(g ) ~

(ciM)(.f)

(9)

f( x)

for all JV -4x

M(g).

clg :S: f

Because of Proposition 29 in the equations (8) and (9) the <p-objects

N can be assumed to be ultra <p-objects.

Proposition 31 In the cases (A) and (S) under the assumption that
(L, ~) fulfills the condition (Z) and in the cases (B') and (H') (in which
(L, ~ ) even is a complete chain) we have cl M V ciN = cl (M V N)
for all M,N E <pX .
Hence, under the assumption on <I> stated in Proposition 31 we have
that the supremum of finitely many closed <p-objects is closed.

Examples. In the "basic example" in the cases (A), (B) and (B') as
well as in the "analogous Lukasiewicz case" we have: p and q defined

Galois Connections in Category Theory, Topology and Logic

451

by p(O) = [!I], q(O) = [h, ~] and p(l) = q(l) = 7Jx(l) are different
<I>-pretopologies which have one and the same closure operator.

References
[1] G. Birkhoff, Lattice theory, Amer. Math. Soc. Colloq. Publ. 25 , 1940.
[2] P. Eklund, W. Gahler, Completions and compactifications by means
of monads, in: Fuzzy Logic, State of Art, Kluwer (1993) 39-56
[3] W. Gahler, Monadic topology- a new concept of generalized topology,
in: Recent Developments of General Topology and its Applications,
Akademie Verlag 1992, 136-149.
[4] W. Gahler, The general fuzzy filter approach to fuzzy topology, I,
Fuzzy Sets and Systems 76 (1995) 205-224.

[5] W. Gahler, The general fuzzy filter approach to fuzzy topology, II,
Fuzzy Sets and Systems 76 (1995) 225-246.

[6] W. Gahler, Monadic convergence structures, Seminarberichte aus dem


Fachbereich Mathematik, Fernuniversitat Hagen 67 (1999) 111-130.
[7] W. Gahler, General topology - the monadic case, examples, applications, Acta Math. Hungar. 88 (4) (2000) , 279-290.
[8] W. Gahler, Extension structures and completions in topology and
algebra, Seminarberichte aus dem Fachbereich Mathematik, Fernuniversitat Hagen (2001).
[9] W. Gahler, P. Eklund, Extension structures and compactifications,
Universitat Bremen, Mathematik-Arbeitspapiere Nr. 54 (2000) 181205.

452

W . Gahler

[10] H. Herrlich, M. Husek, Galois connections categorically, J. Pure Appl.


Algebra 68 (1990) 165-180.
[11] U. Hohle, Monoidal logic, in: Fuzzy Systems in Computer Science,
Vieweg Verlag 1994, 223-234.
[12] U. Hohle, Locales and L-topologies, Universitat Bremen, MathematikArbeitspapiere Nr. 48 (1997) 223-250.
[13] D. C. Kent, On convergence groups and convergence uniformities,
Fundamenta Math. 60 (1967) 213-222.
[14] H.-J. Kowalsky, Limesraume und Komplettierung, Math. Nachr. 12
(1954) 301-340.
[15] 0. Ore, Galois connections, Trans. Amer. Math. Soc. 55 (1944)
493-513.
[16] H. Rasiowa, R. Sikorski, The mathematics of metamathematics,
Paristwowe Wydawnictwo Naukowe 1963.
[17] H. I. Rosenthal, Quantales and their applications, Pitman Research
Notes in mathematics 234, Longman, Burnt Mill, Harlow, 1990.
[18] J. Schmidt, Beitrage zur Filtertheorie II, Math. Nachr. 10 (1953)
197-232.
[19] B. Schweitzer, A. Sklar, Probabilistic metric spaces, North-Holland,
1983.

Author's address:
Werner Gahler,
Scheibenbergstr. 37,
12685 Berlin

You might also like