Sei sulla pagina 1di 28

Duality for Bounded Lattices

Gerard Allwein Chrysa s Hartonas


Visual Inference Laboratory Depts. of Philosophy and Mathematics
Lindley Hall Sycamore Hall
Indiana University Indiana University
Bloomington, Indiana 47405 Bloomington, Indiana 47405
gtall@cs.indiana.edu hartona@ucs.indiana.edu
August 25, 1993

Abstract
We present a duality theorem for bounded lattices that improves
and strengthens Urquhart's topological representation for lattices.
Rather than using maximal, disjoint lter-ideal pairs, as Urquhart
does, we use all disjoint lter-ideal pairs. This allows not only for
establishing a bijective correspondance between lattices and a cer-
tain kind of doubly ordered Stone Spaces (Urquhart), but for a full
duality result. We provide, in the sequel, a treatment of congruences
and epimorphisms, as well as of sublattices, proving relevant duality
theorems. The paper is concluded with a sectional representation of
lattices, imbedding a general lattice in a sheaf space of lattices.

1 Introduction
This paper falls in the tradition of Stone Spaces initiated by Marshall Stone's
[6],[7]topological representation of Boolean algebras and distributive lattices.1 As
is well known, Priestley [5] provided an alternative representation for distribu-
tive lattices by means of Ordered Stone Spaces. Urquhart, in his [8], extended
this result to general, bounded lattices, using doubly-ordered Stone Spaces.
Urquhart's work was subsequently used by Allwein and Dunn [2] to construct
Kripke models for the multiplicative/additive fragment of Linear Logic. As was
pointed out in that paper, the duality work presented here is consistent with the
Kripke models and in fact, the work in Allwein [1] extends the duality theory
presented in this paper to include the Kripke models.
1 Many thanks to J. Michael Dunn for the frequency and quality of discussions over these
matters.

1
A lattice duality is expressed as a pair of contravariant adjoint functors be-
tween a category of lattices and a category of topological spaces. The Stone
duality uses Boolean lattices and Stone spaces. The Priestley duality uses dis-
tributive lattices and Ordered Stone spaces where there is a quasiorder on the
points of the space. Urquhart does not present a full duality but is able to map
lattices bijectively to doubly ordered spaces, i.e., an ordered topological space
with two quasiorders. The reason for the failure to achieve a full duality was
the maximalization process of Zorn's lemma used by Stone and Priestley and
subsequently by Urquhart. There is no reason in the lattice case to use this
lemma and in fact dropping it is necessary to achieve the full duality. Urquhart
[8], in the absence of a full duality theorem, showed that the congruence lattice
is isomorphic to the lattice of R-increasing subspaces, where R is a relation he
de nes on the space. We show that, in e ect, every subspace is R-increasing,
using the correlation between Urquhart's L-spaces and our EL-spaces.
Much of the rst part of the paper is taken up with presenting the lattice
duality. The doubly ordered space of Urquhart must now have two complete
meet semilattices imposed on it. This is directly attributable to the need for
every point of the space to determine a closed set in both of the orders in the
sense that the closed set can be had by taking all the points above that particular
point in the order. There is a further density condition (coupling) that also
must be imposed which relates the two orders. Stone had no order on his space
because maximal ideals are not ordered. The order for Priestley was essentially
the subset order on prime ideals. The orders of Urquhart are the subset orders
on lters and ideals. Our orders are the same as Urquhart however they are
also complete meet semilattice orders. As in Priestley, the inverse images of
morphisms for one of the categories (lattices or spaces) become maps of the
other category in the duality. It is adherence to this principle that forces our
dropping of Zorn's lemma for maximalization from Urquhart's work.
The duality for lattices we present is subsequently extended to a duality
theorem for the lattice of congruences. Urquhart [8], in the absence of a full
duality theorem, showed that the congruence lattice is isomorphic to the lattice
of R-increasing subspaces, where R is a relation he de nes on the space. We show
that, in e ect, every subspace is R-increasing, using the correlation between
Urquhart's L-spaces and our EL-spaces.
While Urquhart's L-spaces become Priestley's CTOD-spaces in the distribu-
tive case, this is no longer the case with the duality we present here. We show,
however, that in the distributive case our EL-spaces contain a CTOD-space as
a distinguished subspace and provide conditions under which the dual space of
a lattice is a space arising from a distributive lattice.
Representation of congruences allows for a construction of a sheaf space
of lattices in a canonical way. We present this construction and show that a
given lattice can be canonically imbedded in the lattice of global sections of its
associated sheaf space.

2
2 Preliminary De nitions
Urquhart [8] proved an objects-only duality theorem for bounded lattices, using
the doubly-ordered Stone Space X = (X; 1 ; 2 ) of disjoint, maximal lter-
ideal pairs. To represent joins, Urquhart de ned a pair of maps r and ` on
subsets of the space increasing in the 1-order and 2-order, respectively. In
the distributive case, r and ` are the set complement operations and a disjoint,
maximal lter-ideal pair is a pair consisting of a prime lter and its complement,
a prime ideal.The maps r and ` form a Galois connection from the lattice of
1-increasing subsets to the lattice of 2-increasing subsets of the space. With
representation map de ned by a 7! a = fxja 2 x1g, where a is an element of
the lattice and x1 is the lter part of the lter-ideal pair x, Urquhart topologized
the space via the subbasis
f? aja 2 Lg [ f?r bjb 2 Lg
and characterized the image of the representation function as the collection of
all doubly-closed (both A and rA closed in the topology) stable sets (stability
meaning that `rA = A).
We retain from Urquhart's paper the use of the maps r and ` and the way
of topologizing the dual space of a lattice. Dropping the restriction to maxi-
mal lter-ideal pairs, we identify the appropriate structure on the dual space
and show that it is sucient to characterize abstractly a certain category of
doubly-ordered, compact, totally separated spaces, which we call enhanced lat-
tice spaces (EL-spaces), as the category dual to the category of bounded, general
lattices.
Some preliminary de nitions are given below, followed by a sketch of the
representation result. A good portion of the rest of the paper is taken up with
the proof of the representation and duality theorems. The remaining part is
devoted to extensions of the result, as already mentioned, including a duality
for the congruence lattice and the lattice of sublattices and a representation of
lattices by continuous sections.
2.1 Complete Doubly Ordered Sets
De nition 2.1 A structure X = (X; 1 ; 2) is a doubly ordered set (do-
set) if
1. i is a partial order on X==i , and
2. for all x; y 2 X x =1 y and x =2 y =) x = y,
where x =i y is de ned as x 1 y and y 1 x.
X is a complete doubly ordered set if each of X==i is a complete meet
semi-lattice.

3
We will be only interested in complete double orders in the sequel.
The relations =i are obviously equivalence relations on the set X. We let
[x]i be the respective equivalence classes, [x]i = fyj x =i yg. Extending the
usual notation, we write either [x)i or x "i for the principal cone over x in the
i -order. More generally, for a subset C  X we write
C "i= [C)i = fy 2 X j 9x 2 C x i yg:
Assume X is a do-set and de ne the functions, as in Urquhart [8]2,
` : P(X) ?! P(X); r : P(X) ?! P(X)
on a subset C of X by
`C = fx 2 X j 8y 2 X(x 1 y ) y 62 C)g (1)
rC = fx 2 X j 8y 2 X(x 2 y ) y 62 C)g (2)
Lemma 2.2 ` and r form a Galois connection
` : L("2 )op ?! L("1 ) r : L("1 ) ?! L("2)op
between the lattice L("1 ) of 1 -increasing subsets of X and the lattice L("2 ) of
the 2 -increasing subsets of X , i.e. for all 1-increasing C and 2-increasing D,
C  `D () D  rC
De nition 2.3 1. Assume X is a do-set and let C  X . We say that C is
`-stable i `rC = C and r-stable i r`C = C .
2. An
V
1-increasing subset AV  X is 1 -closed if eitherVA = ;, or else
V
( 1 A) "1V= A, where by 1 A we mean any element in 1 (A==1 ). Simi-
larly for 2 -closed sets.
3. A
V
subset A  X is -closed if A is
V -closed and rA is
V 1-increasing 1
2
2 -closed.

Lemma 2.4 Considering the Galois connection


` : L("2 )op ! L("1 ); r : L("1 ) ! L("2 )op
if C; D are `-stable sets, C 2 L("1 ); D 2 L("2 ), then
1. rC  ?D =) rC  rD
2. C  ?rD =) C  D
3. C  D and rC  rD =) C = D. 2
2` and r are reversed with respect to the distinction between left and right adjoints in
category theory but we will remain consistent with Urquhart's work.

4
2.2 Doubly Ordered Spaces
De nition 2.5 A doubly ordered topological space is a complete doubly
ordered set together with a topology, X = (X; T ; 1; 2). Given such a space, a
subset C  X is called strongly double-closed, or a dc-set i
1. both C and rC are closed in the given topology, and
V
2. C is -closed.
If X is a doubly ordered space, the dcl-sets of X are the strongly double-
closed, `-stable sets. We let X  be the set of dcl-sets of X .
De nition 2.6 Assume X is a doubly ordered space. Call X doubly order-
separated i for all x; y,
x 61 y =) 9C 2 X  (x 2 C ^ y 62 C)
and
x 62 y =) 9C 2 X  (x 2 rC ^ y 62 rC)
De nition 2.7 A doubly ordered space X is coupled i for all x; y, if there is
z such that x 1 z and y 2 z , then [x]1 \ [y]2 6= ;.
With these preliminary de nitions we can now de ne the notion of an en-
hanced L-Space.
De nition 2.8 An enhanced L-space is a doubly ordered space
X = (X; T ; X  ; 1; 2 )
where
1. X= =1 and X= =2 are both complete, meet semilattices,
2. X is doubly order-separated, coupled and compact,
3. X  [ rX  is a subbasis for closed sets of the topology T (where rX  is the
set of rU , with U 2 X  ), and
4. If C; D  X are dcl-sets, then C \ D and `(rC \ rD) are also dcl.
De nition 2.9 Enhanced L-spaces form a category ELSpace with morphisms
f : X ! Y , where X and Y have the Galois connections r; ` and r0 ; `0, respec-
tively, such that
1. f preserves i ,
2. f ?1 takes Y  into X  and rY  into rX  , and

5
3. If B 2 Y  , then f ?1 rB = r0 f ?1 B , and if D 2 rY  then f ?1 `D = `0 f ?1 D.
We list also the following auxiliary de nitions for later use.
De nition 2.10 Let S be a doubly ordered space and F a family of dcl subsets
of S (doubly closed and `-stable sets). Call F a separating family i
8x; y 2 S (x 61 y =) 9C 2 F (x 2 C ^ y 62 C)):
De nition 2.11 A doubly compact set C is one where C and rC are both
compact.

3 Representation Theory
3.1 The Duality BLat ?! ELSpace
Suppose given a lattice A = (A; ^; _). Let A be the set of all lter-ideal pairs
x = (x1; x2), where x1 \ x2 = ;. A is to be the underlying set of the dual space
of A.
Let i , for i = 1; 2, be the partial orderings on A determined by inclusions,
x i y i xi  yi . It is easy to see that (A ; 1 ; 2) is a complete doubly
ordered set. Let ` and r be the Galois connection of section 2.1. De ne a map
: A ! P(A) by
a = fx 2 A j a 2 x1g
De ne a topology T on this do-set by taking as a subbasis the collection B given
by [
? A [ (?r A) = f? aga2A f?r aga2A
where the \?" sign indicates set complement.
The proof of the following lemma will be delayed for a later section.
Lemma 3.12 The structure SA = (A ; T ; B; 1; 2) is an enhanced L-space,
called the dual space of the lattice A. 2
Conversely, given a space X = (S; T ; X  ; 1; 2), where X  is the set of dcl-
subsets (doubly closed and `-stable), X  forms a lattice with operations de ned
by
C ^ D = C \ D; C _ D = `(rC \ rD)
The lattice L(X ) = (X  ; ^; _) is called the dual lattice of X .
Theorem 3.13 There is an equivalence of categories
BLat ?! S ELSpaceop ELSpaceop ?! L BLat

where a bounded lattice A is isomorphic to the dual lattice of its dual space,
LS A, and an enhanced L-space X is isomorphic to the dual space of its lattice
of dcl-sets (its dual lattice), SLX . 2

6
More speci cally, the function is an isomorphism of a lattice A with the
dual lattice of its dual space LS A. Conversely, given an EL-space X , de ne the
maps 1 ; 2 by
1 x = fC 2 X  j x 2 C g 2 x = fC 2 X  j x 2 rC g
It is easily seen that ( 1 x; 2x) is a disjoint lter-ideal pair. De ne then
: X ?! SLX ; x = ( 1 x; 2 x)
The map is an order-homeomorphism of the EL-space X with the dual space
of its dual lattice SLX .
3.1.1 Duality for Objects
Theorem 3.14 Compact Set Theorem. Every doubly compact, `-stable sub-
set of a compact, doubly order-separated space, X , is closed.
The proof of the Compact Set Theorem is exactly as in Urquhart where he
has doubly closed in place of our doubly compact.
Corollary 3.15 If X is a compact, doubly ordered space and F is a separating
family containing X and ; and closed under ^ and _^ , then F is the family of
all doubly closed `-stable sets in X .
Theorem 3.16 Continuous Map Theorem. Let f : X ! Y be an order iso-
morphic continuous function. If X is compact and Y is doubly order separated,
then f is a homeomorphism.
Proof. All we need show is the continuity of f ?1 . We can easily see that f
maps subbasis elements to subbasis elements and commutes with the operations
l; r; [; \. The latter is so because f is an order-isomorphism. So for l-stable
subbasis element C and r-stable subbasis element D, frC = rfC and flD =
lfD since l and r are de ned solely by the 1 and 2 orders. Because of
this and
T f being at least
T an injection, we have for subbasis elements C, and
D, f fCiji 2 I g = ffCi ji 2 I g and f(C [ D) = fC [ fD. Now, if C is
a subbasis element in X , then C is doubly closed and hence doubly compact.
Therefore, fC is doubly compact since the image of a compact set under a
continuous map is compact, and also the fact that f commutes with l and r by
virtue of being an order-isomorphism. Since Y is doubly order-separated, fC is
doubly-closed from the Compact Set Theorem and is hence a subbasis element
of Y from the preceding corollary. Any closed set in X can be generated from
subbasis elements by the use of arbitrary intersections and nite unions, but f
already distributes across those, hence the image under f of all closed sets in X
can be generated by the image of subbasis elements under f, and those are all
closed. Therefore, f ?1 is continuous. 2

7
Lemma 3.17 If A = (A; ^; _) is a lattice, then the EL-space is de ned by
A = (A ; T ; (A ) ; 1; 2 ) where T is the topology generated by taking the set
(A ) (= A) as a subbasis, and 1 and 2 are the subset orders on lters and
ideals respectively.
Proof: The proof that A is compact is nearly the same as Urquhart's. We rst
show compactness via a reductio argument using Alexander's Subbase Lemma:
\Let (X; T ) be a topological space generated by a subbasis. Then X
is compact if every open cover of X by members of the subbasis has a
nite subcover."
Let P be a subfamily of ?(A ) [?r((A ) ) (which is the open set subbasis)
such that P covers A but no nite subfamily of P covers A . We de ne
P1 = faj ? a 2 P g and P2 = fbj ? r b 2 P g. Then for any collection
fa1; . . .; am g  P1 and fb1 ; . . .; bng  P2, there is some x 2 A such that
x 62 ? a1 [ . . . [ ? am [ ?r b1 [ . . . [ ?r bn, hence x 2 a1 \ . . . \ am \
r b1 \ . . . \ r bn. Therefore, a1 ^ . . . ^ am 2 x1 and b1 _ . . . _ bn 2 x2 by
de nition of . Also by de nition, the lter generated by P1 and the ideal
generated by P2 are disjoint. That is, if there existed some element d such that
a1 ^ . . . ^ am  d  b1 _ . . . _ bn for some selection of a1; . . .; am 2 P1 and
b1; . . .; bn 2 P2, then they could never be separated by a disjoint pair x. Call
this lter-ideal pair (r; ). Then for all S a 2 P1 and b 2 P2, we have r 2S a
and  2 r b giving us that (r; ) 62 P. This is so because (r; ) 2 P
implies (r; ) 2 ? a for some a 2 P1, but then a 2 r and hence (r; ) 2 a
which is a contradiction.
We must cover the pathological cases. Assume  = ;, if r is not proper, then
F 2 r and a1 ^ . . . ^ am  F for some a1 ; . . .; am 2 P1. So F = a1 \ . . .\ am .
But F = ;, hence ? a1 [ . . . [ ? am covers A which is a contradiction. On
the other hand, if r is proper, then z = (r; (c]) for some c is a disjoint pair.
Also, z 2 a for all a 2 P1. Therefore z 62 A ? a for all a 2 P1 and P is not
a cover and we have a contradiction.
Assume r = ;, if  is not proper, then T 2  and T  b1 _ . . . _ bn
for some b1; . . .; bn 2 P2. So r T = r b1 \ . . . \ r bn. But r T = ;, hence
?r b1 [ . . . [ ?r bn covers A which is a contradiction. On the other hand,
if  is proper, then z = ([c); ) for some c is a disjoint pair. Also, z 2 b for
all b 2 P2. Therefore z 62 r b for all b 2 P2 and P is not a cover which is a
contradiction.
From Alexander's subbase theorem, A is compact.
We show that A = =1 and A = =2 are complete meet semilattices. We do
1. The elements Tof A = =T1 are equivalence classes fx; yjx =1Tyg. Thus for
C  A , we de ne 1 C = fy1 jy 2 C g which is a lter: T a; b 2 1 C i V for all
y 2 C, a; b 2 y1 i for all y 2 C, a ^ b 2 y1 i a ^ b 2 1 C. We de ne 1 with
^ \
C = fxj 1
C = x1g: (3)
1

8
T V
V y 2 C,V ( 1 C; y2) 2 1 C.
This set is not empty if C is not empty since for any
We must show that the subbasis elements are 1 and 2 closed. Note that
we can rede ne with
a = fxj[a)  x1g: (4)
This gives us
^ \
a = fxj 1
a = x1g
1
^
= fxj[a)  x1 g
1
= fxj[a) =1 xg

since [a)  x1 implies [a) \ x2 = ;. Therefore, V x 2 fxj[a)  x1g implies


([a); x2) 2 fxj[a)  x1g. Note that for a = F, 1 a = ; = a. Because we
have a bounded lattice, for a 6= F there is always at least one pair ([a); (F]).
From this we get
^
[ a)1 = [fxj[a) = x1g)1
1
= fyj[a)  y1 g
= a
V
Showing [ 2 r a)2 = r a is similar.
We must show that A is coupled. This is easy, if x 1 z and y 2 z then
(x1; y2) is a disjoint pair. Hence [x]1 \ [y]2 6= ;.
A is doubly order-separated. Therefore conditions (1), (2) and (4) of an
enhanced l-space are met. The family f aja 2 Ag ful lls the conditions for the
preceding corollary, hence condition (3) is met. 2
Theorem 3.18 If A is a lattice, then a 7! a is an isomorphism between A
and LS A.
If a 6 b then ([a); (b]) is a disjoint pair and hence a 6= b giving us that
is an order-monomorphism. From the preceding corollary, we get that is
onto. 2
T
Theorem 3.19 Let X be an EL-space, U is a lter inTX  i U = [x)1 for
some element x 2 X . Also, V is an ideal in X  i rV = [x)2 for some
element x 2 X .
Proof: Since our T space is compact, and U has the nite intersection property,
we know that U is non-empty. To simplify the exposition, we are going to

9
V
V a bit and allow the expression 1 C V1 x to mean
cheat on notation V that every
element of 1 C isV underneath x in the 1 order. Let V1 C 1 1 P to mean
every element of 1 C is underneath every V element of 1 P in the 1 . These
slips will
V be justi ed by noting that x; y 2 1 C implies x =1 y and similarly
with 1 P. Let ^
P = fxj8C 2 U( C 1 x)g: (5)
V 1
We note
V that 1 P exists because the 1 order
V isC acomplete
V P. Themeet semilattice.
Also,
V 1 P  P since for all C 2 U, we have 1V 1 1 result is that
[ 1 P)1 =VP, the principle
T cone determined by 1 P is the set P itself. Now
we show [ 1 P)1 = U.
^ ^
y 2 [ P)1 i P 1 y
1 1
i 9x 2 P(x 1 y)
^
i 9x 2 P; 8C 2 U( C 1 x and x 1 y)
^ 1
i 8C 2 U(y 2 [ C)1)
1
i 8C 2\U(y 2 C)
i y 2 U:

Hence
V
we let x 2 1 P and say that U is
T V -closed. Similarly, T rV is
V -closed.
1
2
2
We de ne the functions 1 and 2 by
1 x = fC jx 2 C and C 2 X  g 2 x = fC jx 2 rC and C 2 X  g: (6)
We de ne the function by
x = ( 1 x; 2 x): (7)
That is we send every point into a pair of sets. The left hand side is a lter in
the lattice of subbasis elements and the right-hand side is an ideal: C; D 2 1 x
i x 2 C; D i x 2 C \ D i C \ D 2 1 x. Also, C; D 2 2 x i x 2 rC; rD i
x 2 rC \ rD i l(rC \ rD) 2 2 x i C _^ D 2 2 x. Also, 1 x \ 2 x = ; since
x 2 rC implies x 62 C and visa versa.
Theorem 3.20 X= =1 is isomorphic to the collection of all lters of subbasis
elements under the subset order. X= =2 is isomorphic to the collection of all
ideals of subbasis elements under the subset order.

10
Proof: Let us construe the map, 1 , as being a function from X= =1 to the
collection of lters. 1 is injective, i.e., ?1 x = x, from the separation condition
on the space. T T
To show is 1-1, we let x 2 U such that U = [x)1 forV lter U in
the lattice X  . That is, we de ne ?1 for lter U with ?1 U = T U and
VT 1 1
[x]1 = 1 U. We show ?1 x =1 x. Assume x 61 ?1 x, then for some y, we
1

have ?1 x =1 y from the preceding lemma and x 61 y. Hence there is some
subbasis element C such that x 2 C and y 62 C. But then C 2 x by de nition
and y 62 ?1 x which is a contradiction.
T Therefore x 1 ?1 x and we also
have x 2 C for all C 2 , hence x 2 x and ?1 x 1 x.
We have to work a little harder to get that 1 is surjective. We show, that
?1
V1 T Right to left is easy. Let C 2 U and VletTx 2 1?1 U.
1 U = U. V T T U,Since
U  [
T1U  C, we get 1 U) 1 , and from the previous lemma, [ 1 U) 1 = and
x 2 C. Hence C 2 1 x giving us U  1 1?1 U.
To go the other way, let C 2 1 1?1 U. Then 1?1 U  C and hence [ 1?1 U)1 
TC Ufrom the fact that C is an `-stable set. FromTthe previous lemma, we have
 C. Using T our GaloisTconnection, rC  r U. Therefore, for all x 2 rC,
we have x 2 r U and x 62 U. Hence there exists some Dx 2SU such that x 62
Dx which leads us to conclude x 2 ?Dx . We now have rC  f?Dx jx 2 rC g.
From compactness, there is some integer n such that rC  ?Dx1 [ . . . [ ?Dxn .
Using DeMorgan set laws and properties of stable sets, we get rC  r(Dx1 \ . . .\
Dxn ) and hence Dx1 \ . . . \ Dxn  C. Since U is a lter and Dx1 \ . . .\ Dxn 2 U,
we have C 2 U.
Checking the order preservation property of 1 is easy. Let x 1 y, then for
all elements C 2 X  , we have x 2 C implies y 2 C. Therefore, 1 x  2 y.
That X= =2 is isomorphic to the collection of all ideals of subbasis elements
is similar. 2
Proof: If X is an enhanced l-space, then X is order-homeomorphic to SLX . 2
De nition 3.21 f : X ! Y is an order-embedding when
(x 1 y i fx 1 fy) and (x 2 y i fx 2 fy): (8)
Proof: We prove the following items to show that X and SLX are order-
homeomorphic.
1. is order-embedding.
2. is onto.
3. is continuous.
1: is an order-embedding: x 1 y i 8C(x 2 C implies y 2 C) i 1 x  1 y
i x 1 y. Similarly, x 2 y i x 2 y. Since 1 and 2 are both 1-1, is
an order embedding.

11
2: is onto. Take any lter-ideal pair (U; V ) of the dual lattice X  . From the
previous theorem, we get that 1 and 2 are onto. What we need to know is that
?1 (U; V ) = 1?1 U \ 2?1 V 6= ;. If 1 x = fC jx 2 C g and 2 x = fDjx 2 rDg
where C and rD are `-stable and r-stable subbasis elements T respectively,
T then it
must be the case that if (U; V ) are to identify an point, U \ rV 6= ; (apply
r pointwise to V ). Hence, following Urquhart's lead again, let P = fC jC 2
U g [ frDjD 2 V g. P has the nite intersection property since otherwise for
some C 2 U and D 2 V we have C \ T rD = ; (noteT that U and V are are
composed of subbasis elements and so U and rV exist). In that case,
C  ?rD. From our properties on stable we have C  D, we have a
T Psetssince
contradiction.
VT Hence thereVTis some z 2 X is compact. But then
z 2 [ 1 U)1 and z 2 [ 2 rV )2 from lemma bf Put The Damn Lemma
Reference In Here. Hence for some x 2 1?1 U and y 2 2?1 V , we have x 1 z
and y 2 z. Since X is coupled, 1?1 U \ 2?1 V 6= ;.
For 3 it is sucient to show that the subbasis elements C and r C are
closed in the inverse map ?1 for all C 2 X  .
?1 C = fx 2 X j 1 x 2 C g
= fx 2 X jC 2 1 xg
= fx 2 X jx 2 C g
= C

and
?1 r C = fx 2 X j x 2 r C g
= fx 2 X jC 2 2 xg
= fx 2 X jx 2 rC g
= rC

We have all the facts we need to use the Continuous Map Theorem. The
function is order-isomorphic and continuous, X is compact since it is an EL-
space, and SLX is an EL-space by the de nition of S and so is doubly order-
separated. Hence we conclude that is a homeomorphism. 2
3.1.2 Morphisms
Now we must show that morphisms in the category BLat (Bounded Lattices)
can be translated into morphisms of the category ELSpace (Enhanced L-
Spaces). We need two functors
S : BLat ! ELSpaceop L : ELSpaceop ! BLat: (9)

12
The operation of S and L on objects has already been de ned by the preceding
section. S A is the dual space of the lattice A and LX is the dual lattice of the
space X . We will now have to think about how to map the arrows. A morphism
in the category of bounded lattices preserves the lattice operations and the
bounds. A morphism in the category of ELSpace is a little more complicated.
We recall from section 2.2 that an ELSpace morphism f : X ! Y must
satisfy
1. f preserves 1 and 2 .
2. f ?1 maps Y  to X  and rY  to rX  , i.e., f is continuous.
3. f ?1 commutes with l and r on subbasis elements.
Theorem 3.22 Let h : A ! B be a BLat morphism, then Sh : SB ! SA is
an ELSpace morphism where Sh = h?1 .
Proof: Sh is well-de ned since the h?1 applied to a disjoint pair is still a
disjoint pair and is uniquely determined by h. That Sh preserves 1 and 2
is seen by noting that these are just the subset orders on lters and ideals. We
rename Sh?1 as LSh and note that
x 2 LSh a i Shx 2 a i a 2 (Shx)1 i ha 2 x1 i x 2 ha: (10)
and
x 2 LShr a i Shx 2 r a i a 2 (Shx)2 i ha 2 x2 i x 2 r ha: (11)
LShA = B since A = TA and B = TB . Then x 2 LSh TA i x 2 hTA
i x 2 TB .
Last, we must show commutation of LSh with l and r. From above, we have
x 2 LShr a i x 2 r ha i x 2 rLSh a. Also, x 2 LShlr a i x 2 LSh a i
x 2 ha i x 2 l(r ha) i x 2 l(LShr a).
2
Theorem 3.23 Let f : X ! Y be an ELSpace morphism, then Lf : Y  ! X 
is a BLat morphism where Lf = f ?1 .
Proof: Lf maps Y to X since f is everywhere de ned on X, and Lf ; = ; so
Lf preserves the bounds. Also by de nition, Lf maps Y  to X  . Lf preserves
the lattice operations on Y  : Lf(C \ D) = LfC \ LfD, and Lfl(rC \ rD) =
l(rLfC \ rLfD); the rst is by the properties of inverse functions and the second
is from f ?1 commuting with l and r. We will use the notation (C; D) 2 x to

13
denote that x is a lter-ideal pair in a lattice of sets and that C is in the lter
half and D is in the ideal half. Note that
(C; D) 2 SLf x i (LfC; LfD) 2 x
i x 2 LfC and x 2 rLfD
i x 2 LfC and x 2 LfrD
i fx 2 C and fx 2 rD
i (C; D) 2 fx:

2
We need next to show that S and L preserve identity and composition arrows.
This is easy, we let h : A ! B for any homomorphismh work on pairs of elements
(a; b) such that h(a; b) = (ha; hb). Then we get for any pair of lattice elements
(a; b) and any lter-ideal pair x where (a; b) 2 x means a 2 x1 and b 2 x2 :
(a; b) 2 (SidA )x i idA(a; b) 2 x
i (a; b) 2 x
i (a; b) 2 idS A x:

and for any point x and l-stable subbasis element C


x 2 (LidX )C i idX x 2 C
i x 2 C
i x 2 idLX C:

Next, we do composition. Let h : A ! B and k : B ! C be morphisms in BLat.


Then
(a; b) 2 S(k  h)x i (a; b) 2 (k  h)?1 x
i (k  h)(a; b) 2 x
i (kha; khb) 2 x
i (ha; hb) 2 k?1x
i (a; b) 2 h?1k?1 x
i (a; b) 2 (Sh  Sk)x

That is, S is a contravariant functor. Let f : X ! Y and g : Y ! Z , then


x 2 L(g  f)C i (g  f)x 2 C

14
i gfx 2 C
i fx 2 g?1C
i x 2 f ?1 g?1 C
i x 2 (Lf  Lg)C

Hence, L is a contravariant functor also. eorem. There is a bijection between


morphisms in BLat and morphisms in ELSpace and the following diagrams
commute.

A h -B X f -Y
A B X X
? - LS?B ? - SL?Y
LS A SLX
LSh SLf
Proof: The diagram chases were done in the two preceding theorems show-
ing that and when thought of as transformations are indeed natural. and
are the identity transformations as shown in the preceding section. The counit
arrow at lattice A is denoted A and the unit arrow at space X is denoted X .
These are components of the natural transformations. We compress the identity
results of the preceding section with

(a; b) 2 (S A  S A )x i (a; b) 2 S A ( S A x)
i ( A a; Ab) 2 S A x
i x 2 Aa and x 2 r Ab
i (a; b) 2 x:

which shows that


S A  S A = 1S A: (12)
Also,
x 2 LX  L( X )(A; B) i L X (A; B) 2 x
i x 2 A and x 2 rB:

15
which shows that
LX  L X = 1LX : (13)
The functors S and L are bijective on hom sets. We rst treat S. Let
g; k 2 BLat(A; B) and g 6= k. Then for some a, we have ga 61 ka or visa versa.
Without loss of generality, we choose the former. Hence there is some pair x
such that ga 2 x1 and ka 2 x2 yielding that a 2 (Sgx)1 but that a 62 (Skx)1 .
Therefore S is injective on hom sets.
To show that S is onto, let h 2 ELSpace(S B; S A). We will show there exists
some k 2 BLat(A; B) such that Sk = h. Since h 2 ELSpace(S B; S A), we have
Lh 2 BLat(LS A; LS B from Theorem 3.23. De ne ka = b i (Lh) a = b,
hence LSk = Lh. We have the following chain: a 2 ((Sk)x)1 i ka 2 x1
i x 2 ka i x 2 (LSk) a i x 2 (Lh) a i hx 2 a i a 2 (hx)1. And
a 2 ((Sk)x)2 i ka 2 x2 i x 2 r ka i x 2 r(LSk) a i x 2 (LSk)r a i
x 2 (Lh)r a i hx 2 r a i a 2 (hx)2 . Therefore Sk = h.
Now we work on L. First, we show that L is injective: Let f; h 2 ELSpace(X ; Y )
and f 6= h. Then for some x, we have one of the conditions fx 61 hx or
fx 62 hx or hx 61 fx or hx 62 fx. Assume we have the rst, then there exists
some C such that fx 2 C and hx 62 C. Hence x 2 (Lf)C and x 62 (Lh)C and so
Lf 6= Lh. If we have the second condition, then there exists some C such that
fx 2 rC and hx 62 rC. Hence x 2 L(fr)C and x 62 L(hr)C. Since Lf and Lh
commute with r, we have x 2 r(Lf)C and x 62 r(Lh)C. This means there exists
some y such that x 2 y and y 2 (Lh)C. Since r(Lf)C is r-stable, y 2 r(Lf)C
and y 62 (Lf)C. Therefore Lf 6= Lh.
We prove that L is onto by letting k 2 BLat(LY ; LX ) and show that there
exists some h 2 ELSpace(X ; Y ) such that Lh = k. From k 2 BLat(LY ; LX ),
we get Sk 2 ELSpace(SLY ; SLX from Theorem 3.22. De ne hx = y i
Sk x = y, hence SLh = Sk. We have the following chain where Lh(C; rD)
means allow Lh to work on each member of the pair (C; rD) : x 2 Lh(C; rD)
i hx 2 (C; rD) i (C; D) 2 hx i (C; D) 2 SLh x i (C; D) 2 Sk x i
k(C; D) 2 x i x 2 k(C; rD). Hence Lh = k. 2
Theorem 3.24 BLat and ELSpace are dual to each other.

4 Congruences, Epimorphisms and Subspaces


It follows from general properties of dualities (cf appendix) that if L a R is
a duality C ! Dop , then L takes monomorphisms in C to epimorphisms in
D, and epimorphisms in C to monomorphisms in D. Since congruences on
lattices are kernels of lattice epimorphisms, it is clear that there is a one-to-one
correspondence between congruences on a given lattice A and EL-subspaces of
its dual space X = S A. Such an argument is not directly available, however, in
the case of Urquhart's duality for bounded lattices, precisely because Urquhart's
duality theorem does not extend to a duality for morphisms. In his treatment

16
of congruences and epimorphisms Urquhart does obtain a partial duality for
morphisms, by appealing to a relation R (on L-spaces), de ned as the least
transitive relation containing the relations R1; R2, de ned as follows:
xR1y () [x 1 y and 8w(x 1 w and y 2 w =) w 2 y)]
xR2y () [x 2 y and 8w(x 2 w and y 1 w =) w 1 y)]
He concludes with the following theorem:
Theorem 4.25 (Urquhart) Let A1; A2 be lattices and X1 ; X2 their dual L-
spaces. There is a one-one correspondence between epimorphisms  : A1 ! A2
and embeddings : X2 ! X1, where (X2 ) is an R-increasing L-subspace of X1.
2
Urquhart proves that every L-subspace of the form (X2) is R-increasing, but
one is left with the impression that congruences are set to correspondance with
only such L-subspaces and that, presumably, there are L-subspaces that fail
this property. We show that this is not the case, by arguing that every L-
subspace of the dual L-space of a given lattice is obtained as the image of a
morphism as in the theorem above. This allows for the conclusion that, in
both Urquhart's duality and our extension of it to a full duality of morphisms
as well, the congruence lattice is dual to the lattice of subspaces (L-subspaces,
or EL-subspaces respectively), in a way to be made precise in the sequel. We
begin on a more basic level, gradually leading to this conclusion.
Notation 4.26 For a lattice A, we will denote by X = SA its dual EL-space,
and by X^ = S( ^ A) its dual L-space. We will consistently use the hat notation
for subsets consisting of maximal points.
De nition 4.27 Let A be a lattice an X = SA its dual EL-space. Consider
Y  X in the subspace topology and doubly-ordered with the restrictions of i .
De ne the maps `0 and r0 on Y . Call Y a compatible subspace of X i
C 2 X  =) r0(Y \ C) = Y \ rC D 2 rX  =) `0 (Y \ D) = Y \ `D
To simplify arguments we rst note that
Lemma 4.28 The compatibility condition on the Galois connections is equiva-
lent to the requirement that for any C 2 X  and D 2 rX  ,
Y \ C #2  (Y \ C) #Y2 ; Y \ D #1  (Y \ D) #Y1
where the superscript Y on the downarrows indicates restriction of the downclo-
sure in Y  X .
The converse inclusions are always true. The lemma follows by noticing that r
and ` can be alternatively de ned by rC = X n C #2 and `D = X n D #1. The
rest follows from de nitions. 2

17
De nition 4.29 For a congruence  on A, a lter-ideal pair x in the dual space
is called -closed i if a 2 xi and a  b, then b 2 xi, for each i = 1; 2. We let
X   X be the set of -closed members of X .
Assume X is the dual EL-space of A.
Proposition 4.30 Let Y  X be a compatible subspace. Then
1. The set  = fY \ C jC 2 X  g is a lattice, in fact a quotient of X  
=A
2. Y generates a congruence Y on A, de ned by
a Y b i Y \ a = Y \ b
3.  = A=Y
4. Y  X Y , i.e. every y 2 Y is Y -closed.
Write CY := Y \ C and de ne CY ^ DY = (C \ D)Y and, where `0 ; r0 is
the Galois connection on Y , de ne join by CY _ DY = `0 (r0CY \ r0 DY ). Using
compatibility it follows CY _ BY = Y \ (A _ B) = (A _ B)Y . To see why,
notice that `0 (Y \ (rA \ rD)) = Y \ `(rC \ rD). This follows from the fact
that rC \ rD = r`(rC \ rD), where `(rC \ rD) 2 X  , so that rC \ rD 2 rX  ,
and then by the compatibility condition. For part (ii), consider the map  :
X  !  de ned by C = Y \ C.  is a lattice homomorphism and its kernel
is clearly ker() = Y . Thus   = L=Y follows, too. For the last part, notice
rst that by compatibility the congruence Y can be alternatively de ned by
Y \ r a = Y \ r b. Now if y 2 Y; a 2 y1 and a Y b, then the de nition of
Y implies y 2 b, hence b 2 y1 . Use the alternative de nition of Y to get that
y2 is Y -closed as well. 2
The following lemma lays the ground for correlating congruences to sub-
spaces.
Lemma 4.31 Let X be an EL-space, dual to a lattice L = X  , and Y  X in
the subspace topology.
1. Y is an EL-subspace i
V
(a) Y is a complete subset of X , i.e. for C  Y; i C  Y , for i = 1; 2.
(b) the Galois connection `0 a r0 on Y is compatible with the connection
` a r on X (Y is a compatible subspace)
(c) Y is a compact subset of X .
2. Y is an L-subspace i it is a compact and compatible subspace consisting
of maximal points (i.e. points x such that for all y, x 1 y and x 2 y
implies x = y).

18
We rst show that the conditions imply that Y is an L-space (EL-space, respec-
tively). S
Since Y is in the subspace topology, fY \ C gC 2X  fY \ DgD2rX  is a
subbasis for closed sets. The compatibility hypothesis implies that the sets
Y \ C, with C 2 X  , are `0 -stable sets, since
`0 r0(Y \ C) = `0 (Y \ rC) = Y \ `rC = Y \ C
Compactness implies that the family fY \ C jCdoubly-closed and `-stable g is
the family of all `0 -stable sets (this follows from a lemma due to Urquhart a
modi cation of which applies in the case of the extended duality theorem as
well). Furthermore, each Y \ C is closed and r0 (Y \ C) = Y \ rC is closed, since
C is doubly closed. Thus Y \ C is doubly closed. It is straightforward that Y
inherits the order-separation property, and compactness is already assumed for
Y . Letting Y  be the collection of `0 -stable sets, take Y \ C; Y \ D 2 Y  . Then
r0(Y \ (C \ D)) = Y \ r(C \ D) is closed and similarly for `0 (r0 (Y \ C) \ r0(Y \ D)).
It follows from the above that if Y consists of maximal points, then Y is an
L-space. And if Y is a complete subset, then it is an EL-space.
For the converse, notice that, in particular, the set X^  X of all maximal
points is an L-space. The compatibility condition follows from the fact that X^
is the set of all maximal points.
Speci cally, if x^ 2 X^ \ C #2, let x^ 2 y 2 C. Use Zorn's lemma to get a
maximal pair y^ above y in both orderings. Then y 1 y^ implies that y^ 2 C,
since C is 1-increasing. Thus x^ 2 (X^ \ C) #X2^ and X^ is a compatible subset.
That X^  X is an L-space comes, of course, as no surprise, since X^ is precisely
the dual L-space of the given lattice L. Call X^ the characteristic L-subspace
of the dual EL-space X of the lattice L. It is clear that if Y^ is an L-subspace
of X , then Y^ is contained in an EL-subspace Y , of which it is the characteristic
L-subspace. Hence Y^ is a compatible subset of Y , by the previous argument.
Compatibility of Y in X follows from the fact that the inclusion Y ,! X is an
EL-morphism. Since the compatibility relation is clearly transitive, it follows
that Y^ is a compatible subset of X . 2
We can now argue that the congruence lattice is isomorphic to the lattice
of EL-subspaces (respectively, L-subspaces) of its dual EL-space (respectively,
L-space).
Proposition 4.32 Let A be a bounded lattice, X its dual EL-space and X  = A
its dual lattice. If  is a congruence on X  , then the set of -closed lter-ideal
pairs is an EL-subspace X  ,! X . Furthermore, every EL-subspace Y is of the
form Y = X  , for some congruence  on X  .
Completeness of the set X  follows from the fact that the intersection of
-closed lters, or ideals, is again -closed.
For compatibility, it is enough to show that r (X  \ C)  X  \ rC (the
other direction is immediate and the argument for ` and ` will be similar). Let

19
p : L ! L= be the canonical projection and X = S A; X = S A= the dual
spaces, with p = Sp. Let also ` a r be the Galois connection on X.
Let x 2 r (X  \ C)  X  , a -closed pair, and suppose x 62 rC. Let x be
the pair with xi = f[a]ja 2 xi g. If x 62 rC, then x 62 p? 1 rC = rp? 1C, hence
there is y such that x 2 y and y 2 p? 1 C. Then y (where yi := faj[a] 2 yi g)
is a -closed pair and x 2 y 2 C, contradiction.
Compactness follows from continuity of p and compactness of X.
To see that every EL-subspace is of the form X  , if Y ,! X is an EL-
subspace, let  be the congruence Y it induces (by the compatibility condition)
and observe that Y = X  . 2
EL-subspaces have a lattice structure, easily seen once they are considered
in the form X ; X . It is straightforward to verify that setting
^ _
X  X := X _ X  X := X ^
the subspaces of X become a complete lattice.
We summarize the duality for congruences and EL-subspaces in the following
theorem, where A and X are dual.
Theorem 4.33 The collection of EL-subspaces Sub(X ) of an EL-space X form
a complete lattice isomorphic to the lattice Quot(A) of quotient lattices of its
dual lattice A and dual to the lattice Con(A) of congruences on A. 2
Remains to see that the same conclusion applies in the case of L-spaces, with
the congruence lattice of a lattice A being dual to the lattice of L-subspaces of
its dual L-space X^. This will follow from the following lemma.
Lemma 4.34 Let Y^ ,! X be an L-subspace and  the congruence it generates .
Then Y^ is the characteristic L-subspace X^ of the EL-subspace of -closed pairs
in X .
Y^ generates a congruence on the lattice A via
a  b i Y^ \ a = Y^ \ b
since, as argued before, Y^ is a compatible subset of X (where X is the dual
EL-space of A). The conclusion follows directly from the fact that the dual
lattice  of Y^ is exactly the dual lattice A= of X  . 2
In fact, if p : A ! A= is the canonical epimorphism, and, letting S be^
the functor such that S( ^ A) is the dual L-space of A, if q : S(
^ A= ) ,! S(
^ A)
?1 ? ^ ^
is the imbedding qx = (p x1; p x2), then q(S(A= )) = Y . Hence every L-
1

subspace of the dual L-space X^ of a lattice A is obtained from a congruence


on A. Consequently, every L-subspace of X^ is R-increasing, where R is the
relation introduced by Urquhart as the smallest transitive relation containing
the relations R1; R2 de ned as in (1).

20
Since every L-subspace of the dual L-space X^ = S(
^ A) is of the form X^  , and
the X form a lattice in the way previously indicated, it is clear that L-subspaces

of X^ form a lattice. Thus we can conclude:
Theorem 4.35 The congruence lattice Con(A) is dual to the lattice of L-
subspaces Sub(X^) of the dual L-space X^ of A. 2
4.1 Distributive Lattices
It is a familiar fact that in a distributive lattice an ideal maximal among such
disjoint from a xed lter is prime, and similarly for lters. Consequently, with
duality using only maximal lter-ideal pairs, it is not hard to show, as does
Urquhart, that a lattice is distributive just in case every point in its dual L-space
consists of a prime lter and its complement. In this sense, an L-space in the
distributive case is exactly a CTOD-space (compact, totally order disconnected
space). If all lter-ideal pairs are used there is no hope for a characterization
in terms of some property of the points of the space, since a copy of every lter
and every ideal (paired with the trivial ideal or lter respectively) is present in
the space. What can be shown is that
Theorem 4.36 Let A be a lattice, X , X^ its dual EL and L-spaces respectively,
and X   = X^  
= A the dual lattices. Then
1. The following are equivalent
(a) every C 2 X  is a compatible subset of X
(b) every C 2 X  is an EL-subspace
2. The following are equivalent
(a) A is distributive
(b) every C 2 X  generates a congruence via D C E i C \ D = C \ E
Furthermore, if any of the rst two (equivalent) conditions holds, then A is
distributive.
If X  is replaced by X^  in the statements above, then all four conditions are
equivalent to each other.
We show equivalence of two of the conditions above independently of the topo-
logical reprersentation.
Lemma 4.37 For a lattice A the following are equivalent
1. A is distributive

21
2. for any a 2 A, the relation de ned by
b a c i a ^ b = a ^ c
is a congruence on A.
Distributivity of A clearly entails the other condition. Conversely, assuming
every a 2 A generates a congruence in the way indicated, none of the lattices
K1; K2 below can be a sublattice of A

 1 HHHH   1 @



aH b Hc a @@c
HH
HH 0  b XX
XXX 0 ??
K1 K2
Otherwise, by assumption, a 2 K1  A generates a congruence on K1 (the re-
striction of the congruence it generates on A). If A=a is the quotient lattice,
then the canonical epimorphism p : A ! A=a should be a lattice morphism.
However, p(b _ c) = p(1) 6= pb _ pc = p0, since a ^ 1 = a 6= 0 = a ^ 0. Similarly
K2 is not a sublattice of A. Hence A is distributive. 2
For the proof of the theorem now, it is clear that if C 2 X  is a compatible
subset, then C is an EL-subspace, since C is also closed and complete. The
converse is obvious. Assuming compatibility, by previous proposition, every
C 2 X  generates a congruence in the way indicated, hence by the lemma, A is
a distributive lattice.
Finally we show that if we consider C 2 X^  , then distributivity of A^ is
equivalent to compatibility of each C 2 X^  .
First, suppose the lattice is distributive. Then for any D 2 X^  ; D = D #2 .
To see why, notice that r and ` can be alternatively de ned by rD = X^ n D #2
and `E = X^ n E #1. If A is distributive, then for every x = (x1 ; x2) 2 X, ^
x1 is a prime lter and x2 its complement. It then follows that for any y,
x 1 y () y 2 x. Thus for D 2 X^  , it follows D = D #2 . Then
rC (C \ D) = C n D #C2 = C n (C \ D #2) = C n D #2= C n D
It then follows that C \ rD = C \ (X^ n D) = C n D = rC (C \ D). Similarly for
` and `C .
For the converse, enough to show that, assuming every C 2 X^  is a compat-
ible subset, for any C; D; E 2 X^  we have C ^ (D _ E)  (C ^ D) _ (C ^ E).

22
Using the compatibility assumption, this is equivalent to showing that
`C rC (C \ (D [ E))  `r(C \ (D [ E)) (14)
If x is in the left hand side then by de nitions we have
8y 2 C [x 1 y =) 9z [y 2 z and z 2 C \ (D [ E)]] (15)
If x is not in the right hand side, let y be such that x 1 y and y 2 r(C \(D [E)).
Such y is in C (since C is 1-increasing). Then there is z 2 C such that y 2 z
and z 2 C \ (D [ E), contradicting the assumption that y 2 rC (C \ (D [ E)).
Compatibility of the sets C 2 X^  is clearly equivalent to the assertion that
every C 2 X^  is an L-subspace. 2
4.2 Sectional Representation
As an application of the duality for congruences discussed above, we give a
representation of bounded lattices by sheaf spaces.
4.2.1 Sheaf Spaces of K-Algebras
We recall rst the de nition of a sheaf space of sets.
De nition 4.38 1. A concrete sheaf or sheaf space is a triple < F; ; X >
where F; X are topological spaces and  : F ! X is a local homeomor-
phism (i.e. for every a 2 F there is a neighborhood Ua , and a neighborhood
Va such that the restriction jUa : Ua ! Va is a homeomorphism).
2. F is called the total space,  the projection and X the base space.
3. For each x 2 X , the set Fx = ?1 x is called the stack of x
4. For any open subset U  X , a section on U is a continuous map
 : U ! F such that  : U ! U is the identity on U . A global
section is a section on X .
5. The set of all sections on U is denoted by ?(U; F).
6. For each space X , the category C(X ) of concrete sheaves on X is the
category with morphisms  :< F; ; X >?!< G; ; X > the continuous
functions  : F ! G such that  = .
The notion of a sheaf space of sets above generalizes to that of a sheaf space
of algebras (K-algebras, or more generally algebras of a certain similarity type
). Speci cally, we require that
1. for each x 2 X, the stack Fx = ?1 x over x is an algebra of type ,

23
2. for each open subset U  X, the sections ?(U; F) form an algebra of the
same type, under pointwise operations, i.e. if f is an n-ary operation and
1 ; . . .; n : U ! F are sections, then f(1 ; . . .; n) is a section on U,
de ned on x 2 U by
f(1 ; . . .; n)(x) = f(1 x; . . .; nx)
4.2.2 Sheaf Spaces for Bounded Lattices
Let A be a lattice and X = S A its dual space. For each x 2 X, let fi gi2I
be theWcollection of all nontrivial
V congruences such that x 2 X i 6= X , and put
 =
x  , so that X x =
i2I i

i2IX i is the smallest EL-subspace containing
x.
Lemma 4.39 A is a subdirect product of the (Ax)x2X .
Q a 7!< [a]x >x2X is injective. Clearly, composition with
Verify that the map
a projection px : x2X Ax ?! Ax is the natural quotient map A ! Ax .
Furthermore, for each x, X x  X i  X , hence A is not isomorphic to any of
the Ax . 2
We construct Fa sheaf space < F; p; X > as follows.
We let F = A , and p : F ! X the natural projection [a] 7! x.
x2X x x
Topologize X by taking as a subbasis the collection of the subsets
Xa;b = fxj a x bg
For each a 2 A, let [a] : X ! F be the map [a](x) = [a]x.
Now topologize F by taking as a subbasis the collection f[a]U ga2A, for each
open subset U of X and where [a]U is the pointwise image of U under [a]. This
is a standard construction about which we recall from [3] the following
Proposition 4.40 With topology imposed as above,
1. the maps [a] with a 2 A are continuous and the map u : A ! ?(X; F)
de ned by u : a 7! [a] is a well de ned lattice homomorphism.
2. < F; p; X > is a sheaf space of lattices. 2
For each x, the stack p?1x = Fx = Ax is the quotient lattice obtained from
A by factoring by x. If U  X is an open subset, then the sections ?(U; F)
form a lattice, with operations, given ;  : U ! F,
 ^  : U ! F; x 7! x ^ x = [a]x ^ [b]x = [a ^ b]x
 _  : U ! F; x 7! x _ x = [a]x _ [b]x = [a _ b]x
Letting (X; F) be the collection of global sections  such that there is an
a 2 A and for all x; x = [a]x, (X; F) is a sublattice of the lattice of global
sections and we have:

24
Proposition 4.41 u : A ! ?(X; F) is a lattice imbedding, and A = (X; F).
u is a homomorphism and injectivity follows from the fact that if [a]x = [b]x for
all x 2 X, then for all x 2 X we have C \X x = D \X x , where C = a; D = b
and is the representation function. Now if y 2 C, then from C \X y = D \X y
it follows y 2 D, and similarly for the other direction. Hence C = D and by
injectivity of , a = b follows. 2
In the sectional representation above we used as a base space the space
consisting of all disjoint lter ideal pairs. The same argument will go through
if the space is restricted to the collection of all maximal lter-ideal pairs, since
lemma 2.17 will still be true. In the particular case where A is distributive, its
dual L-space is a compact totally order disconnected space (CTOD space), in
the sense of Priestley [5]. The sheaf topology on the base space coincides, in
this case, with the representation topology.
4.3 Sublattices and Quotient EL-Spaces
Let X and A be dual, A  = X  (the lattice of dcl-sets). We show in this section
that the lattice of sublattices of A is isomorphic to the lattice of quotient EL-
spaces of its dual EL-space X .
If K ,! A is a sublattice, K induces an equivalence relation on X de ned by
x  y (modK) () K \ x = K \ y
where K \ x = K \ (x1 ; x2) := (K \ x1; K \ x2 ). The factor set X is doubly
ordered via
x i y () K \ xi  K \ yi
We denote the equivalence classes by x or, for more clarity x = K \ x. Letting
[x]i = [K \ x]i = fx0 j x =i x0 g = fx0j K \ x0i = K \ xig
be the =i equivalence classes in X ==i := Xi , the latter become complete meet
semi-lattices, with binary meets
^
[K \ x]i [K \ y]i = [K \ (x \ y)]i
i
for each i = 1; 2. Letting ` a r be the Galois connection on the doubly ordered
 1; 2 ), and  the canonical epimorphism
set (X;
 : X ?! X;  x 7! K \ x
where, recall, K \ x := x denotes the equivalence class of x 2 X, we have
Lemma 4.42  : X ?! X is an ELSpace epimorphism, where X is considered
in the quotient topology.

25
 is the image, under the duality, of the inclusion K ,! A. From familiar
facts, the quotient topology

on X coincides with the representation topology.2
Conversely, if X ?! Y is an ELSpace epimorphism, then  determines
a sublattice of the dual lattice X  of X . Where C 2 X  , call C -stable
if ?1 C = C, and similarly for rD 2 rX  . The proof of the following is
straightforward.
Lemma 4.43 If X ?!  Y is an ELSpace epimorphism, the -stable dcl-sets
of X form a sublattice K ,! X  of its dual lattice. 2
It easily follows from the above that
Theorem 4.44 The lattice of sublattices Sub(A) of a lattice A is dual to the
lattice Quot(X ) of quotient spaces of its dual space X .
By the above discussion, every quotient EL-space of X is of the form X=K , for a
sublattice K of its dual lattice. The lattice structure on Quot(X ) is the obvious
one. 2

26
5 Appendix 1: Duality
5.1
Assume L a R : C ?! Dop is a duality of categories C and D. We list some
well known properties of the adjunction:
1. L, having a right adjoint, preserves colimits in C, i.e. colimits in C are
taken to limits in Dop .
2. Dually, R preserves limits, hence colimits in Dop are taken to limits in C.
3. Each of L and R, being part of an equivalence of categories, is full and
faithful.
Corollary 5.45 Let L a R : C ?! Dop be a duality of categories C and D.
Then
f Lf
1. a ! b is an epi in C i Lb ?! La is monic in D, and
h Lh La is an epi in D.
2. a ,! b is monic in C i Lb ?! 2
The proof follows from the fact that L is full and faithful.
5.2 Subobjects and Quotient Objects
We assume a duality of categories C and D as above. For a category X, we
write (A #e X) for the category of epis with domain A and (X "m A)for the
category of monics with codomain A.
Corollary 5.46 If L a R is a duality of the categories C and D, then there is
an equivalence of the categories
L a R : (A #e C) 
= (D "m LA)
and
L a R : (C "m A) = (LA #e D)
We write Quot(A) for the set of quotients of an object A in a category, and
Sub(A) for its set of subobjects. Recall that quotient objects and subobjects
are ordered by f  g i f factors through g, i.e. i there is an arrow h in the
category such that f = gh. Notice that f  g in C just in case f op  gop in Cop
where monics turn to epis and vice-versa. It follows from the previous corollary
that
Corollary 5.47 If C is dual to D, then the partial order of quotient objects
Quot(A) in C is isomorphic to the partial order of subobjects Sub(LA) in D.
Similarly, the partial order of subobjects Sub(A) in C is isomorphic to the partial
order of quotients Quot(LA) in D. 2

27
In particular
Corollary 5.48 If Quot(A) is a lattice in C (i.e. (A #e C) has products and
coproducts) then so is Sub(LA) and they are isomorphic as lattices. Similarly,
if Sub(A) is a lattice, then so is Quot(LA) and they are isomorphic as lattices.
2

References
1. Allwein, Gerard. Duality of Algebraic and Kripke Models for Lin-
ear Logic, Dissertation, August 1992, Indiana University.
2. Allwein, Gerard and J. Michael Dunn. Kripke Models for Linear Logic, to
appear in Journal of Symbolic Logic, 1993.
3. Davey, B. A. Sheaf Spaces and Sheaves of Universal Algebras, Math.
Zeitschrift 134, pp. 275-290, 1973.
4. Davey, B.A. and H.A. Priestley. Introduction to Lattices and Order,
Cambridge University Press, 1990.
5. Priestley, H. A. Ordered Topological Spaces and the Representation of Dis-
tributive Lattices, Proceedings of the London Mathematical Society (3) 24
(1972) pp.507-530.
6. Stone, Marshall. The Theory of Representations for Boolean Algebras,
Transactions of the American Mathematical Society, 40, pp. 37-111.
7. Stone, Marshall. Topological Representations of Distributive Lattices and
Brouwerian Logics, C asopsis pro Pestovani Matematiky a Fysiky, 67, pp.
1-15.
8. Urquhart, Alasdair. A topological representation theory for lattices, Alge-
bra Universalis 8 (1978) pp. 45-58.

28

Potrebbero piacerti anche