13
Applied Categorical Structures 9: 139–151, 2001. © 2001 Kluwer Academic Publishers. Printed in the Netherlands. 139 What is a Quotient Map with Respect to a Closure Operator? Dedicated to our friend Horst Herrlich on the occasion of his sixtieth birthday MARIA MANUEL CLEMENTINO Departamento de Matemática, Apartado 3008, 3000 Coimbra, Portugal. e-mail: [email protected] ERALDO GIULI Dip. di Matematica Pura ed Applicata, Università degli Studi di L’Aquila, 67100 L’Aquila, Italy. e-mail: [email protected] WALTER THOLEN Department of Mathematics and Statistics, York University, Toronto, Canada M3J 1P3. e-mail: [email protected] (Received: 26 February 1998; accepted: 8 September 1998) Abstract. It is shown that there is no good answer to the question of the title, even if we restrict our attention to Set -based topological categories. Although very closely related, neither the natural notion of c-finality (designed in total analogy to c-initiality) nor the notion of c-quotient (modelled after the behaviour of topological quotient maps) provide universally satisfactory concepts. More dramatically, in the category T op with its natural Kuratowski closure operator k, the class of k-final maps cannot be described as the class of c-quotient maps for any closure operator c, and the class of k-quotients cannot be described as the class of c-final maps for any c. We also discuss the behaviour of c-final maps under crossing with an identity map, as in Whitehead’s Theorem. In T op, this gives a new stability theorem for hereditary quotient maps. Mathematics Subject Classifications (2000): 18B30, 54B30, 54C10, 18A20. Key words: closure operator, c-initial morphism, c-final morphism, c-quotient, Whitehead’s Theo- rem. 1. Introduction In topology, as in other parts of mathematics, a fundamental device for obtain- ing new structures from old is the formation of quotients. In the category T op of topological spaces, quotient maps are described as regular epimorphisms, that is: as coequalizers of their kernel-pairs. When viewed as a topological category over S et (cf. [12, 14]), quotient maps are simply surjections which are final (= co-cartesian) with respect to the forgetful functor T op S et . Finally, if we think of T op as a category equipped with a closure operator, as given by the usual Kuratowski closure, then we may want to characterize a quotient map p: X Y

What is a Quotient Map with Respect to a Closure Operator?

Embed Size (px)

Citation preview

Applied Categorical Structures9: 139–151, 2001.© 2001Kluwer Academic Publishers. Printed in the Netherlands.

139

What is a Quotient Map with Respect to a ClosureOperator?

Dedicated to our friend Horst Herrlich on the occasion of his sixtieth birthday

MARIA MANUEL CLEMENTINODepartamento de Matemática, Apartado 3008, 3000 Coimbra, Portugal. e-mail: [email protected]

ERALDO GIULIDip. di Matematica Pura ed Applicata, Università degli Studi di L’Aquila, 67100 L’Aquila, Italy.e-mail: [email protected]

WALTER THOLENDepartment of Mathematics and Statistics, York University, Toronto, Canada M3J 1P3. e-mail:[email protected]

(Received: 26 February 1998; accepted: 8 September 1998)

Abstract. It is shown that there is no good answer to the question of the title, even if we restrictour attention toSet-based topological categories. Although very closely related, neither the naturalnotion ofc-finality (designed in total analogy toc-initiality) nor the notion ofc-quotient (modelledafter the behaviour of topological quotient maps) provide universally satisfactory concepts. Moredramatically, in the categoryTop with its natural Kuratowski closure operatork, the class ofk-finalmaps cannot be described as the class ofc-quotient maps for any closure operatorc, and the class ofk-quotients cannot be described as the class ofc-final maps for anyc. We also discuss the behaviourof c-final maps under crossing with an identity map, as in Whitehead’s Theorem. InTop, this givesa new stability theorem for hereditary quotient maps.

Mathematics Subject Classifications (2000):18B30, 54B30, 54C10, 18A20.

Key words: closure operator,c-initial morphism,c-final morphism,c-quotient, Whitehead’s Theo-rem.

1. Introduction

In topology, as in other parts of mathematics, a fundamental device for obtain-ing new structures from old is the formation of quotients. In the categoryT opof topological spaces, quotient maps are described as regular epimorphisms, thatis: as coequalizers of their kernel-pairs. When viewed as a topological categoryover Set (cf. [12, 14]), quotient maps are simply surjections which are final (=co-cartesian) with respect to the forgetful functorT op → Set . Finally, if wethink of T op as a category equipped with a closure operator, as given by the usualKuratowski closure, then we may want to characterize a quotient mapp: X → Y

140 MARIA MANUEL CLEMENTINO ET AL.

as an epimorphism which reflects closedness:B ⊆ Y is closed wheneverp−1(B)

is closed. Is this the ‘right’ general concept?Certainly not. In the topological categorySGph of spatial (= loop-free) graphs

with its natural ‘up-closure’ (which coincides with the Kuratowski closure whenSGph is coreflectively embedded into the category ofCech-pretopological spaces– see [7]), the corresponding notion of quotient no longer characterizes the naturalquotient maps, as given by the regular epimorphisms or the final structures withrespect toSGph→ Set .

On the other hand, there is a natural way of describing quotient maps inSGphvia its natural closure operator↑, namely as↑-final maps. In general,c-final mor-phisms are defined algebraically in total analogy toc-initial morphisms (as in-troduced in [4]). They interact perfectly withc-closed andc-open morphisms viapullback (cf. [11]). However, whilec-initial, c-closed andc-open morphisms de-scribe the ‘right’ maps inT op with c = k the Kuratowski closure,c-finality doesnot characterize quotient maps but hereditary quotient maps. The natural questionthen, whether hereditary quotient maps inT op are describable as thec-final mapsfor any closure operatorc, is answered to the negative (Section 3). Conversely,when definingc-quotients as the ‘surjective’ morphisms reflectingc-closedness forsubobjects, as inT op, then in this way one is not able to characterize the hereditaryquotient maps ofT op (Section 5).

Despite these strikingly negative results, we have a few positive assertions infavour of the notion ofc-finality, in addition to the smooth pullback results givenin [11]. Among these is a general closure-theoretic theorem which, when applied toT op andk, gives the following Whitehead-type result: a mapf : X→ Y for which1Z×f : Z×X→ Z×Y is a hereditary quotient map for every compact HausdorffspaceZ, has this property for every topological spaceZ. We also show that thereis a general concept of localc-compactness which is stable under the formationof c-closed subobjects and ofc-perfectc-final morphisms (Section 7). Finally, ourconsideration of algebraic categories, such as the category ofR-modules with itsnatural closure operators given by preradicals, yields further evidence thatc-finalityis an important and natural categorical notion.

2. Initial and Final Morphisms

2.1. Throughout the paper, we work in a finitely-complete categoryX with aproper and pullback-stable(E,M)-factorization system for morphisms (see, forexample, [1, 3, 7, 10]). For every objectX ∈ X, morphisms inM with codomainX are referred to assubobjectsof X; they are the objects of the full subcategorysubX of X/X, which in fact is a preordered class with finite infima. We use theusual lattice-theoretic notation in subX. Every morphismf : X → Y gives theimage-preimage adjunction

f (−) a f −1(−): subY → subX.

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 141

Hence, for allm ∈ subX andn ∈ subY , one has the inequalities

(1) m ≤ f −1(f (m)) and (2) f (f −1(n)) ≤ nwith ‘∼=’ holding true in (1) in casef is monic (this is easy to see forf ∈ M; fora proof in casef is any monomorphism, see [7]), and with ‘∼=’ holding true in (2)if and only if f ∈ E .

2.2. (Cf. [5, 7]) A closure operator c of X with respect to(E,M) is givenby a family of functionscX: subX → subX (X ∈ X) such thatm ≤ cX(m),cX(m) ≤ cX(n) if m ≤ n, andf (cX(m)) ≤ cY (f (m)) holds for allf : X → Y inX andm,n ∈ subX; one refers to the last inequality as thec-continuity conditionfor f . A subobjectm ∈ subX is c-closed if m ∼= cX(m), and it is c-denseifcX(m) ∼= 1X. A morphismf : X→ Y is c-denseif f (1X) ∈ subY is c-dense. Theoperatorc is calledidempotentif cX(m) is c-closed for allm ∈ subX, X ∈ X,and it isweakly hereditaryif k with cX(m) · k = m is c-dense for allm ∈ subX,X ∈ X. Finally, c is hereditaryif for all m: M → Y , y: Y → X in M one hascY (m) ∼= y−1(cX(y · m)). At times we omit the subscripts if there is no danger ofconfusion.

2.3. Given a closure operatorc of X w.r.t. (E,M), a morphismf : X → Y iscalledc-initial (cf. [4, 7]) if

cX(m) ∼= f −1(cY (f (m)))

for all m ∈ subX (with ‘≤’ holding true for free, by thec-continuity condition andimage-preimage adjunction); andf is calledc-final if

cY (n) ∼= f (cX(f −1(n)))

for all n ∈ subY (with ‘≥’ being satisfied automatically). We note the followingimportant facts:

(1) if e · m = 1 in X, thenm is a c-initial morphism inM and e is a c-finalmorphism inE (see [7], Ex. 2.L, and [11]);

(2) every morphism inM is c-initial if and only if c is hereditary;(3) in view of (2), we callc cohereditaryif and only if every morphism inE is

c-final;(4) c-final morphisms necessarily belong toE (just considern = 1Y in the defin-

ing property), butc-initial morphisms need not be inM;(5) a c-final monomorphism is automaticallyc-initial and lies inE , but may fail

to be an isomorphism – just consider the discrete closure operator in the cate-gory T op of topological spaces with its (surjective, embedding)-factorizationstructure.

142 MARIA MANUEL CLEMENTINO ET AL.

2.4. The following composition-cancellation rules hold true; proofs for the state-ments onc-initiality may be found in [7] (9.2), and the statements onc-finality areproved similarly:

(1) If each of the composable morphismsf andg arec-initial (c-final), theng · fis c-initial (c-final).

(2) If g · f is c-initial, thenf is c-initial, and alsog is c-initial provided thatfbelongs toE .

(3) If g · f is c-final, theng is c-final, and alsof is c-final provided thatg ismonic.

2.5. Consider a pullback diagram

U

p′

f ′V

p

Xf

Y

(1)

in X. The following proposition and its corollary were proved in [11]:

PROPOSITION. (1)If f is c-final andp′ c-initial, thenf ′ is c-final.(2) If f ′ is c-initial and p c-final, thenf is c-initial.

COROLLARY. (1) Let c be hereditary. Then, for everyp: V → Y in M, therestrictionf −1(V )→ V of thec-final morphismf is c-final.

(2) Let c be cohereditary. Then, for everyp: V → Y in E , if the pullbackf ′: U → V of f is c-initial, also f is c-initial.

2.6. As a first example (of a topological category overSet), we consider thecategorySGph of spatial graphs (cf. [7]): objects are sets with a binary reflexiverelation→, and morphisms preserve the relation. With respect to its (surjective,embedding)-factorization structure, consider the up-closure

↑X (M) = {x ∈ X | ∃a ∈ M : a→ x}.↑-initial morphismsf : X → Y are characterized by the condition(x1 → x2 ⇔f (x1) → f (x2)) for all x1, x2 ∈ X (hence these are precisely the initial mapsw.r.t. the forgetful functorSGph→ Set), and the↑-final morphisms are preciselythe natural quotient maps, that is: the surjectionsf : X → Y with (y1 → y2 ⇔∃x1, x2 ∈ X: x1→ x2 andf (xi) = yi) (which are also described as the final mapsw.r.t. SGph→ Set).

We remark that↑-final morphisms do not coincide withc-final morphisms forc the idempotent hull of↑. Indeed,↑-final andc-final morphisms cannot even becompared: consider the set{0,1,2} as a graphY , with the graph structure given

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 143

by the natural order, and consider the graphX arising fromY by removing theedge 0→ 2, and the identity maph: X → Y ; h is c-final but it is not↑-final. Onthe other hand, ifZ is the graph{0,1,1′,2} with the nontrivial edges 0→ 1 and1′ → 2, the mapk: Z→ X, with k(0) = 0, k(1) = k(1′) = 1, k(2) = 2, is↑-finalbut it is notc-final.

Finally we note that the operator↑ is hereditary, but neither idempotent norcohereditary. Next we will show that failure of cohereditariness was to be expecteda priori.

2.7. LetX be any topological category overSet with its (surjective, embedding)-factorization structure (see [1, 12, 14]). We consider the discrete operatord (withdX(M) = M), the indiscrete operatorj (with jX(∅) = ∅ and jX(M) = X forM 6= ∅), and the trivial operatort (with tX(M) = X); then:

PROPOSITION.An arbitrary closure operatorc of a topological categoryX overSet is cohereditary if and only ifc ∈ {d, j, t}.

Proof. The operatorsd, j , t are obviously cohereditary. Conversely, considera cohereditary closure operatorc, so that all surjective morphisms arec-final. Weassumec 6= j , c 6= t and showc = d. With D the two-point discrete object inX(obtained by applying the left adjoint ofX→ Set to {0,1}), we first observe:

(1) cD = dD; indeed, assuming the opposite and considering the ‘switch’ mapD→ D, we would either havecD(∅) 6= ∅ and obtaincD = tD, or cD(∅) = ∅,cD({0}) = D and obtaincD = jD, hencecD ≥ jD in any case; this would im-ply cX(M) ⊇ jX(M) for all M ⊆ X ∈ X, as one easily sees when exploitingc-continuity of the mapsfx : D → X with fx(0) ∈ M andfx(1) = x 6∈ M,and thereforec ∈ {j, t}, in contradiction to our assumption.

Next one considers the two-point indiscrete objectJ in X (applying the rightadjoint ofX→ Set to {0,1}) and shows:

(2) cJ = dJ ; this follows immediately from the fact that the identity mapD→ J

must bec-final, by hypothesis.

Now considering mapsX→ J one easily showscX = dX for all X ∈ X. 22.8. In the categoryT op of topological spaces with its (surjective, embedding)-factorization structure andcX(M) = kX(M) = M the usual (Kuratowski-) closureof M ⊆ X ∈ T op, thek-initial mapsf : X → Y are the ones for whichX carriesthe initial topology with respect tof (i.e. A ⊆ X is closed iffA = f −1(B) forsomeB ⊆ Y closed). Thek-final maps are precisely thehereditary quotient maps(cf. [2, 9, 13]), i.e. those surjective mapsf : X → Y for which every subspaceV ⊆ Y carries the final topology w.r.t. the restrictionf −1(V ) → V of f (so thatB ⊆ V is closed ifff −1(B) ⊆ f −1(V ) is closed).

144 MARIA MANUEL CLEMENTINO ET AL.

We note that the operatork is idempotent and hereditary but (of course) notcohereditary; more significantly, in general,quotient maps fail to bek-final. Infact, it is well known that for a hereditary quotient mapg: X → Y in T op and aspaceZ, the map

f ′ = 1Z × g: Z ×X −→ Z × Y

need not be a quotient map. For example, consider forg the projectionR → Y ,whereY arises fromR by identifying all the integers, and letZ := R\{12, 1

3, . . .}(cf. [9], 2.4.20). Butf ′ is the restriction off = 1R × g along the embeddingZ × Y → R × Y , and, by Whitehead’s Theorem,f is a quotient map. This alsoshows thatc-initiality of p′ is an essential condition in Proposition 2.5(1). Thatc-finality of p is an essential condition in Proposition 2.5(2) is much easier to see (interms ofX = T op, c = k: restrict a bijective map, which is not a homeomorphism,to a single point).

2.9. In the categoryR-Mod of (left-)R-modules with its (epi,mono)-factorizationstructure, every closure operatorc induces apreradical r (that is: a subfunctor ofthe identity functor), namelyrX = cX(0) ≤ X ∈ R-Mod. Conversely, given anyr ,there is a least closure operator minr and a largest closure operator maxr inducingr , namely

minrX(M) = rX +M and maxrX(M) = π−1(r(X/M)),

with π the projectionX→ X/M for M ≤ X ∈ R-Mod.The minr -initial morphisms are thosef : X→ Y with rX = f −1(rY ), and the

minr -final morphismsf : X→ Y are the epimorphisms withf (rX) = rY . Hence,to say thatr is hereditary(which meansrM = M ∩ rX for M ≤ X ∈ R-Mod) isthe same as to say that every monomorphism is minr -initial; it is also equivalent tothe hereditariness of minr , or of maxr (see [7], 3.4). Similarly,cohereditarinessof r(that is:r(X/M) ∼= (M+ rX)/M forM ≤ X ∈ R-Mod) is equivalently describedby the property that every epimorphism is minr -final, i.e. by the cohereditariness ofthe closure operator minr (cf. 2.3(3)); it is also equivalent to the statement minr =maxr (see [7], 5.12).

The maxr -initial morphismsf : X → Y are characterized by the property thatfor all M ≤ X, the induced morphismX/M → Y/f (M) is minr -initial, and everyepimorphism is maxr -final. We observe that maxr -initial morphisms are minr -initial, but not conversely, in general. Indeed, forR = Z andr the torsion radical,consider the additionf : Z×Z→ Z and letM be the diagonal inZ×Z; thenf isminr -initial althoughZ× Z/M → Z/f (M) is not, since its domain is isomorphicto Z and its codomain is isomorphic toZ2. By contrast, minr -final morphisms are(as epimorphisms) trivially maxr -final, while the converse proposition generallyfails (just consider any projection ofZ onto a finite group forr as above).

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 145

We recall that minr is idempotent, and that maxr is idempotent iffr is a rad-ical (i.e. r(X/rX) = 0); by contrast, idempotency ofr (i.e. r(rX) = rX) ischaracterized by weak hereditariness of minr or, equivalently, of maxr (cf. [7],Prop. 3.4).

3. c-Finality does not Describe Topological Quotients

3.1. Since quotient maps in the topological categorySGph are described as the↑-final maps while inT op quotient maps fail to bek-final (cf. 2.6, 2.8), it is natural toask whether there is any closure operatorc of T op such that thec-final morphismsare precisely the quotient maps. Despite the rich supply of closure operators in thecategoryT op, the answer is negative:

THEOREM. The categoryT op has no closure operatorc w.r.t. its (surjective,embedding)-factorization structure, such that thec-final maps are exactly the quo-tient maps.

For the proof we first recall a general fact on closure operators in topologicalcategories overSet (see [7], Theorem 9.1):

3.2. PROPOSITION.If the idempotent hullc of a closure operatorc of a topolog-ical categoryX overSet is hereditary, then alreadyc must be idempotent.

3.3. Proof of the theorem.We assume that there is a closure operatorc suchthat the class ofc-final maps is the class of quotient maps. Thenc 6∈ {d, j, t} (cf.2.7) since otherwise every surjection would bec-final. Now consider the SierpinskispaceS = {0,1} whose only nontrivial closed subset is{1}. Then:

(1) cS 6= jS , since otherwise the identity mapS → J (with J the two-pointindiscrete space) would bec-final;

(2) cS 6= dS , since otherwise the identity mapD → S (with D the two-pointdiscrete space) would bec-final. From (1) and (2) one concludes:

(3) cS = kS (with k denoting the usual closure), sincecS({0}) = S. In fact, if{0} werec-closed, then we would havec ≤ k∗, with k∗ denoting theinverseKuratowski closure, given byk∗X(M) =

⋂ {U |U ⊆ X open,M ⊆ U } = {x ∈X | {x} ∩ M 6= ∅} (cf. [7]); then c would have to coincide with the discreteclosured when restricted to the categoryT op1 of T 1-spaces, so that everysurjection inT op1 would have to be a quotient map. We can now conclude:

(4) cX ≤ kX for everyX ∈ T op. For that it suffices to show that every (k-)closedsetF ⊆ X is c-closed (since thencX(M) ⊆ cX(M) = M = kX(M) forall M ⊆ X). In fact, for a closed setF ⊆ X, we have a continuous mapf : X → S with F = f −1(1), so thatF is c-closed as a pullback of thec-closed set{1}. Next we show:

146 MARIA MANUEL CLEMENTINO ET AL.

(5) c = k. In view of (4), for that it suffices to show that everyc-closed subsetG ⊆X ∈ T op is (k-)closed. ForG ⊆ X c-closed andJ the two-point indiscretespace, consider the mapg: X → J with G = g−1(1), and its (quotient,mono)-factorizationg = m · q, wherem: Y → J is an inclusion.c-finality ofq impliescY (1) = 1, and thenY = D or Y = S andcY = kY . In both cases,the setG, as an inverse image of ak-closed subobject, must bek-closed. From(4) we conclude with 3.2:

(6) c = k. But this contradicts our initial assumption that quotient maps arec-final, sincek-final maps are hereditary quotient maps: see 2.8. This concludesthe proof of Theorem 3.1. 2

4. c-Quotient Morphisms

4.1. The characterization ofk-initial morphisms inT op (see 2.8) can be given inour abstract categorical context 2.1, provided that the closure operatorc of X w.r.t.(E,M) is idempotent:

PROPOSITION. Let c be idempotent. Thenf : X → Y is c-initial if and onlyif everyc-closed subobjecta of X is of the forma ∼= f −1(b) for somec-closedsubobjectb of Y .

Proof.The condition is trivially necessary. Conversely, for everym ∈ subX, weknow thatc(m) is of the formf −1(b) with b ∼= c(b) ∈ subY . Hencef (c(m)) ≤b, evenc(f (m)) ≤ c(f (c(m))) ≤ c(b) ∼= b. Consequently,f −1(c(f (m))) ≤f −1(b) ∼= c(m), as desired. 24.2. Of course, idempotency is an essential assumption in 4.1. For instance,the graph morphismh considered in 2.6 is not↑-initial although it satisfies thecondition of Proposition 4.1.

4.3. The attempt to establish an analogous characterization ofc-final morphisms(as is suggested byT op andk) is only partially successful:

PROPOSITION.Letf : X→ Y bec-final. Then a subobjectb of Y is c-closed ifand only iff −1(b) is c-closed inX.

Proof. In fact, if f −1(b) is c-closed, then

cY (b) = f (cX(f −1(b))) ∼= f (f −1(b)) ≤ b.The ‘only if’ part is well-known. 24.4. We call a morphismf : X→ Y in X a c-quotient if it lies in E and reflectsc-closed subobjects, so thatb is c-closed inY wheneverf −1(b) is c-closed inX.Then one may observe immediately:

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 147

(1) A c-final map is ac-quotient, with the converse statement failing already forX = T op andc = k (cf. 4.3 and 2.8).

(2) Since the notion ofc-quotient depends only onc-closed subobjects (not on theoperatorc itself), it is no loss of generality to requirec to be idempotent in thiscontext, in contrast with the situation ofc-finality (see 2.6).

(3) If each of the composable morphismsf andg arec-quotients, theng · f isalso ac-quotient.

(4) If g·f is ac-quotient, theng is ac-quotient, and alsof is ac-quotient providedthatg is monic.

(5) The property 2.5(1) on the stability under pullback fails if we replacec-final byc-quotient: considerk in T op. Also 2.5(2) does not hold if one assumesp to bea c-quotient instead of beingc-final: consider again, inSGph, the morphismh: X→ Y defined in 2.6. The pullback ofh along itself is obviously↑-initial,althoughh is ↑-quotient but not↑-initial.

4.5. For a preradicalr of R-Mod as in 2.9 and the closure operators minr andmaxr , one shows quite easily that a minr -quotientf : X → Y is already minr -final; indeed,N ≤ Y is minr -closed iff rY ≤ N , so that it suffices to considerN = f (rX) to see thatf (rX) = rY holds. Since every epimorphism is maxr -final, it is in particular a maxr -quotient. Hence, for the two closure operators inquestion, the notionsc-final andc-quotient are equivalent.

5. c-Quotients do not Describe Hereditary Quotient Maps inT op

5.1. In 3.1 we showed that the class ofk-quotient maps inT op cannot be repre-sented as the class ofc-final maps, for anyc. Now we show that the class ofk-finalmaps cannot be represented as the class ofc-quotient maps, for anyc:

THEOREM. The categoryT op has no closure operatorc such that thec-quotientmaps are precisely the hereditary quotient maps.

Proof. Replacingc-quotient byc-final in the proof of Theorem 3.1, we mayuse exactly the same arguments, since we only deal with quotient maps whosecodomain has two points – and therefore they are also hereditary quotient maps–, to conclude thatc = k. But k-quotient maps do not coincide with hereditaryquotient maps, and the result follows. 2

6. Pullbacks ofc-Final Morphisms

6.1. Proposition 2.5(1) gives a sufficient condition for the pullback of ac-finalmap to bec-final. Unfortunately, the given condition (namely, that in the pullbackdiagram (1) the morphismp′ be c-final) is quite restrictive; in this section we

148 MARIA MANUEL CLEMENTINO ET AL.

give sufficient conditions on the objects involved. First we recall that the pullbackdiagram (1) decomposes into two pullback diagrams

U

〈f ′,p′〉

f ′V

〈1V ,p〉

V ×X 1V×fV × Y

Xf

Y

(2)

Since〈1V , p〉 ∈ M, f ′ is c-final if 1V × f is, and ifc is hereditary. Therefore, inwhat follows,we assumec to be hereditaryand consider the problem for whichobjectsZ is 1Z × f c-final wheneverf is c-final.

6.2. Recall (cf. [5, 11]) that a morphismf : X → Y is c-closedif f (cX(m)) ∼=cY (f (m)) for all m ∈ subX. The morphismf is stablyc-closed(stablyc-final)if every pullback off is c-closed (c-final). Note that iff represents ac-closedsubobject,f is actually a stablyc-closed morphism inM; and if f is (stably)c-closed inE , thenf is necessarily (stably)c-final (cf. [11]).

An objectX in X is calledc-Hausdorff (c-compact, resp.) if the morphismδX: X → X × X (!X: X → 1, resp.) is stablyc-closed; in other words ifδX ∈sub(X × X) is c-closed (the projectionX × Y → Y is c-closed for allY ∈ X,resp.). Hence, a stablyc-closed morphismf : X → Y may be described as ac-compact object of the comma-categoryX/Y , andf is c-perfectiff it is c-compactc-Hausdorff inX/Y ; cf. [3].

6.3. For the remainder of this section, we assume that for everyc-HausdorffobjectX there is ac-dense reflexion

βX: X −→ βX

into the full subcategory ofc-Hausdorffc-compact objects ofX. As in [3], we callX c-Tychonoffif X is c-Hausdorff withβX ∈ M. We recall that theHenriksen–Isbell Theoremholds true in this context, which we shall use later on:

PROPOSITION. A morphismf : X → Y with c-Tychonoff objectsX and Y isc-perfect(or stablyc-closed) if and only if

X

f

βXβX

βf

YβY

βY

(3)

is a pullback diagram.

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 149

6.4. The following proposition shows in particular thatc-final morphisms withgood pullback behaviour w.r.t.c-compactc-Hausdorff objects behave well w.r.t.c-Tychonoff objects.

PROPOSITION. Let f : X → Y have the property that for everyc-compactc-Hausdorff objectZ, the morphism1Z × f is c-final. Then this property is ac-tually satisfied for every objectZ for which there exists a stablyc-final morphismq: W → Z withW a c-Tychonoff object.

Proof. For every morphismh: U → V we have two consecutive pullbackdiagrams, with the unnamed arrows being projections:

U ×X1U×f

h×1XV ×X

1V×fX

f

U × Y h×1YV × Y Y

(4)

If Z is c-Tychonoff, thenh := βZ: Z → V := βZ and alsoh× 1X lie in M andarec-initial. Since, by hypothesis, 1V × f is c-final, an application of 2.5(1) to theleft pullback of (4) gives that 1Z × f is c-final.

More generally, now letZ be as in the theorem and chooseh := q: W → Z.Thenh×1Y is c-final by hypothesis onq, and 1W×f is c-final as shown previously.An application of the composition-cancellation rules 2.4(1), (3) to the left part of(4) gives that 1Z × f is c-final, as claimed. 26.5. ForX = T op andc = k, we can strengthen the assertion of 6.4 even further:

THEOREM. A map which is a hereditary quotient map when crossed with theidentity map of a compact Hausdorff space is actually a hereditary quotient mapwhen crossed with the identity map of any topological space.

Proof.Following [8] (Lemma 2.4), we can express any spaceX as a hereditaryquotient of a Tychonoff space: for eachx ∈ X, letXx be the space with underlyingsetX and with all points butx isolated; forx one keeps the same neighbourhoodsas inX; the topological sum

∑y∈X Xy is easily seen to be a Tychonoff space, and

the map

p:∑

y∈X Xy −→ X

(x, y) 7−→ x

is obviously a hereditary quotient map. Now the result follows from Proposi-tion 6.4. 26.6. REMARK. We point out that the hypothesis of the theorem is not satisfied foreveryhereditary quotient map: consider again the hereditary quotient mapf : R→Y of 2.8; then, for the unit intervalI , the map 1I×f fails to be a hereditary quotientmap since, withZ := I\{1/n | n ∈ N}, 1Z × f fails to be a quotient map.

150 MARIA MANUEL CLEMENTINO ET AL.

7. A Note on Local Compactness

7.1. Finally we indicate that the setting of Section 6 allows, to a certain degree, fora categorical treatment of local compactness. Recall (cf. [7, 11]) that a morphismf : X→ Y is c-openif f −1(cY (n)) ∼= cX(f −1(n)) for all n ∈ subY .

DEFINITION. An objectX is calledlocally c-compactif X is c-Tychonoff withβX: X → βX c-open, andX is a stablec-spaceif X is c-Hausdorff and thereexists a locallyc-compact objectY and a stablyc-final morphismq: Y → X.

We note that forX = T op and c = k the Kuratowski closure operator thenotion of localc-compactness takes on the classical meaning of locally compactHausdorff space, while stablek-spaces form a proper subclass ofk-spaces.

7.2. Localc-compactness has the following closure properties w.r.t. morphisms:

PROPOSITION. Let f : X → Y be a stablyc-closed morphism ofc-TychonoffobjectsX andY .

(1) If Y is locally c-compact andf ∈M, thenX is locally c-compact.(2) If X is locally c-compact andf ∈ E , thenY is locally c-compact.

Proof. Consider the pullback diagram (3). According to Theorem 4.3 of [11],βX is c-open whenβY is c-open andf is c-initial, andβY is c-open whenβX isc-open andβf is c-final. Now, if f ∈M we havef c-initial sincec is hereditary;and if f ∈ E , then composition/cancellation rules show thatβf is c-dense, andsinceβf (as a morphism from ac-compact object to ac-separated object, see [3])is alsoc-closed, it must belong toE and is in factc-final (cf. 6.2). 27.3. From Proposition 6.4 we conclude immediately:

COROLLARY. Let f : X → Y have the property that, for everyc-compactc-Hausdorff objectZ, the morphism1Z × f is c-final. Then this property is satisfiedby any stablec-spaceZ.

We point out that, contrary to first impression, the corollary does not lead toWhitehead’s Theorem in Topology, sincec-final maps cannot describe quotientmaps; see also 6.5(1).

Acknowledgements

This work was supported by a NATO Collaborative Grant (no. 940847) and byVolkswagen-Stiftung (RiP Program Oberwolfach, Germany). The first author wouldlike to thank CMUC (Portugal) for partial financial assistance. The second au-thor acknowledges partial financial assistance by MURST (Italy). The third authorwould like to thank NSERC (Canada) for partial financial assistance. In addition

WHAT IS A QUOTIENT MAP WITH RESPECT TO A CLOSURE OPERATOR? 151

we thank Alan Dow for drawing our attention to his paper [8] which led us tothe current formulation of Theorem 6.5. We also thank the anonymous referee forher/his careful reading of the paper.

References

1. Adámek, J., Herrlich, H. and Strecker, G. E.:Abstract and Concrete Categories, Wiley, NewYork, 1990.

2. Arhangel’skiı, A.: Bicompact sets and the topology of spaces,Dokl. Akad. Nauk SSSR150(1963), 9–12; English translation:Soviet Math. Dokl.4 (1963), 561–564.

3. Clementino, M. M., Giuli, E. and Tholen, W.: Topology in a category: Compactness,Portugal.Math.53 (1996), 397–433.

4. Dikranjan, D.: Semiregular closure operators and epimorphisms in topological categories,Suppl. Rend. Circ. Mat. Palermo, Serie II29 (1992), 105–160.

5. Dikranjan, D. and Giuli, E.: Closure operators I,Topology Appl.27 (1987), 129–143.6. Dikranjan, D. and Giuli, E.: Compactness, minimality and closedness with respect to a closure

operator, in:Proc. Internat. Conf. on Categorical Topology, Prague 1988, World Scientific,Singapore, 1989, pp. 284–296.

7. Dikranjan, D. and Tholen, W.:Categorical Structure of Closure Operators, with Applicationsto Topology, Algebra and Discrete Mathematics, Kluwer Acad. Publ., Dordrecht, 1995.

8. Dow, A. and Watson, S.: A subcategory ofTop, Trans. Amer. Math. Soc.337(1993), 825–837.9. Engelking, R.:General Topology, revised and completed edn, Heldermann Verlag, Berlin,

1989.10. Freyd, P. J. and Kelly, G. M.: Categories of continuous functors, I,J. Pure Appl. Algebra2

(1972), 169–191.11. Giuli, E. and Tholen, W.: Openness with respect to a closure operator, Preprint, 1997.12. Herrlich, H.: Topological functors,Gen. Topology Appl.4 (1974), 125–142.13. McDougle, P.: A theorem on quasi-compact mappings,Proc. Amer. Math. Soc.9 (1958), 474–

477.14. Wyler, O.: Top categories and categorical topology,Gen. Topology Appl.1 (1971), 17–28.