betaFeedback

WorksMathematics

Kernel retracts of co-sober spaces

Abstract

Kernel retracts of co-sober spaces need not be co-sober, even when the ambient space is countably infinite and compact. We construct a co-sober T0T_0-space YY and a continuous idempotent map k ⁣:Y→Yk\colon Y\to Y satisfying k(y)≤Yyk(y)\leq_Y y in the specialization order, whose image with the subspace topology is not co-sober. The construction starts from a countable compact T1T_1 KC anti-Hausdorff space of van Douwen type and adjoins an infinite chain. We classify the kk-irreducible compact saturated subsets of the extension and show that each is a point saturation. The kernel operator collapses the added chain to a distinguished point while fixing the original space, thereby recovering the non-co-sober seed as its image. This answers negatively two questions of Xu and Zhao, concerning continuous kernel operators and arbitrary retracts, and contrasts with the known preservation of co-sobriety by continuous closure operators.

Introduction

Sobriety and co-sobriety are distinct properties in non-Hausdorff topology and domain theory. The notion of a co-sober space was introduced by Escardó, Lawson and Simpson in connection with a dual form of the Hofmann–Mislove theorem; see [2]. In a co-sober space, every kk-irreducible compact saturated set is the saturation of a unique point. It is easy to give a co-sober space which is not sober (see, e.g., [6 Example 2.1]). Escardó, Lawson and Simpson asked whether every sober space is co-sober ([2 Problem 9.7]). In [6], Wen and Xu gave a negative answer to this problem by proving that Isbell’s lattice constructed in [3] equipped with the lower topology is sober but not co-sober. Based on Wen and Xu’s result, Shen et al. [5] further obtained a dcpo whose Scott topology is sober but not co-sober.

A basic theme in the theory is the preservation of co-sobriety under natural topological constructions (see [6, 7, 8]). Wen and Xu proved, among other things, that closed subspaces and saturated subspaces of co-sober spaces are co-sober, and that the image of a co-sober space under a continuous closure operator is co-sober [6]. Motivated by this result, Xu and Zhao asked whether the analogous assertion holds for kernel operators [7 Question 9.5]. With the convention

x≤Xy  ⟺  x∈cl⁡X{y},x \le_X y \iff x \in\operatorname{cl}_X\{y\},

a continuous map k:X→Xk:X \to X is a kernel operator if

k2=k,k(x)≤Xx (x∈X).k^2 = k,\quad k(x) \le_X x\ (x \in X).

Email addresses: xiqxu2002163.com (Xiaoquan Xu), [email protected] (Fengshuo Xu), [email protected] (Haitao Ji), [email protected] (Dong Qiu). They also asked whether an arbitrary retract of a co-sober space must be co-sober [7 Question 9.6]. We answer both questions negatively within the class of countable compact spaces. The construction separates two requirements: co-sobriety of the ambient space and failure of co-sobriety in the kernel image. We start with a countable compact T1T_1 KCKC anti-Hausdorff seed ZZ which is not co-sober. We then attach a countable chain to ZZ. The chain is topologized so that all non-principal compact saturated sets whose projection covers the whole seed admit explicit nontrivial decompositions and hence fail to be kk-irreducible. The remaining kk-irreducible compact saturated sets project onto proper kk-irreducible compact subsets of ZZ, and those are forced to be singletons. This makes the extension co-sober. Finally, a kernel operator collapses the added chain to one distinguished point of ZZ while fixing the old copy of ZZ, so that its image recovers the non-co-sober seed.

It is useful to formulate the conclusion in the language of retracts. If r:X→Ar:X\to A and i:A→Xi:A\to X are continuous, ri=id⁡Ari=\operatorname{id}_A, and ir≤id⁡Xir\leq\operatorname{id}_X, we shall call AA a kernel retract of XX. Dually, the condition id⁡X≤ir\operatorname{id}_X\leq ir corresponds to the closure-operator situation; one may naturally call such an AA a closure retract. The known positive theorem for closure operators and our negative theorem for kernel operators therefore exhibit a genuine asymmetry between these two kinds of order-compatible retracts.

Our main result is the following.

Theorem 1.1. There exist a countably infinite compact co-sober T0T_0-space YY and a continuous kernel operator k:Y→Yk:Y\to Y such that the subspace k(Y)k(Y) is not co-sober. Consequently, a kernel retract, and hence a retract, of a co-sober space need not be co-sober.

The proof uses the equality form of kk-irreducibility, not the stronger condition involving inclusions in unions, and all images carry the subspace topology. The sliced almost-disjoint family and the seed topology are due to van Douwen; the infinite-chain extension and the analysis of its compact saturated sets provide the kernel counterexample. Section 3 records the seed construction and classifies its proper kk-irreducible compact subsets. Sections 4–5 define the extension, determine its specialization order, and construct the kernel operator. Sections 6–8 establish co-sobriety of the extension and failure of co-sobriety of its image.

Preliminaries

Compactness is used without any separation assumption. We write N={0,1,2,…}\mathbb{N}=\{0,1,2,\ldots\} and [N]ω[\mathbb{N}]^\omega for the family of all infinite subsets of N\mathbb{N}.

Let XX be a topological space. A subset S⊆XS\subseteq X is saturated if it is the intersection of all open sets containing it. If XX is T0T_0, this is equivalent to SS being an upper set with respect to the specialization order. We write

sat⁡X(x)={y∈X:x≤Xy}\operatorname{sat}_X(x)=\{y\in X:x\leq_X y\}

for the saturation of a point xx.

A nonempty compact saturated set K⊆XK\subseteq X is called kk-irreducible if

K=K0∪K1,K0,K1 compact saturated,K=K_0\cup K_1,\quad K_0,K_1\text{ compact saturated},

implies K=K0K=K_0 or K=K1K=K_1. The family of all such sets is denoted by K-Irr(X)\mathrm{K\text{-}Irr}(X). A T0T_0-space XX is co-sober if for every K∈K-Irr(X)K\in\mathrm{K\text{-}Irr}(X) there exists a unique point x∈Xx\in X such that

K=sat⁡X(x).K=\operatorname{sat}_X(x).

Recall that a space is a KCKC-space if every compact subset is closed, and it is anti-Hausdorff if every two nonempty open subsets intersect. A space is called a USUS-space if every convergent sequence has a unique limit. A Fréchet space means a Fréchet–Urysohn space: for every A⊆XA\subseteq X and x∈cl⁡XAx\in\operatorname{cl}_X A, a sequence in AA converges to xx. We shall use the following observation of Künzi and van der Zypen [4 Lemma 1].

Lemma 2.1 (Künzi and van der Zypen). Every Fréchet USUS-space is a KCKC-space.

Proof. Let KK be a compact subset of a Fréchet US-space XX, and let x∈cl⁡XKx \in\operatorname{cl}_{X} K. By the Fréchet property there is a sequence (xn)(x_n) of points of KK converging to xx. Since KK is compact, the sequence has a cluster point y∈Ky \in K. Using the Fréchet property again, one can choose a subsequence of (xn)(x_n) converging to yy. Since XX is a US-space, limits of convergent sequences are unique, and hence x=y∈Kx=y \in K. Thus KK is closed.

Lemma 2.2. Let XX be a non-singleton compact T1T_1 KC anti-Hausdorff space. Then XX is not co-sober.

Proof. Since XX is T1T_1, every subset is saturated. Suppose

X=K0∪K1X = K_0 \cup K_1

with K0K_0 and K1K_1 nonempty compact subsets. By the KC property, both KiK_i are closed. If both are proper, then X∖K0X \setminus K_0 and X∖K1X \setminus K_1 are disjoint nonempty open sets, contradicting anti-Hausdorffness. Hence XX is kk-irreducible. Since XX is non-singleton while every point saturation in a T1T_1-space is a singleton, XX is not co-sober.

A van Douwen-type seed space

We use the sliced maximal almost-disjoint construction of van Douwen. We first state the required combinatorial result and then record the properties of his seed space needed for the extension.

A sliced MAD theorem of van Douwen

The following is the case of van Douwen’s slicing theorem for maximal almost-disjoint (MAD) families on a countable set; see [1 Sections 5–7]. The underlying set is countable; the family itself is not asserted to be countable.

Theorem 3.1 (van Douwen). There exist an infinite family A⊆[N]ω\mathcal{A} \subseteq[\mathbb{N}]^{\omega} and a map ℓ:A→N\ell: \mathcal{A} \to\mathbb{N} such that:

(M1)(M1) distinct a,b∈Aa,b \in\mathcal{A} have finite intersection;

(M2)(M2) for every infinite S⊆NS \subseteq\mathbb{N} there exists a∈Aa \in\mathcal{A} such that a∩Sa \cap S is infinite;

(M3)(M3) if

A(S)={a∈A:∣a∩S∣=∞}\mathcal{A}(S) = \{a \in\mathcal{A} : |a \cap S| = \infty\}

is infinite, then

ℓ[A(S)]=N.\ell[\mathcal{A}(S)] = \mathbb{N}.

Equivalently, for every n∈Nn \in\mathbb{N} there exists a∈A(S)a \in\mathcal{A}(S) with ℓ(a)=n\ell(a) = n.

Remark 3.2. Conditions (M1) and (M2) say that A\mathcal{A} is an ω\omega-MAD family. Condition (M3) is precisely the slicing condition denoted by (E) in van Douwen’s construction; see [1 p. 151]. Its existence in ZFC follows from van Douwen’s slicing theorem, in particular Theorem 7.1 of [1], after identifying the countably infinite set occurring there with N\mathbb{N}.

For a∈Aa \in\mathcal{A} and B⊆aB \subseteq a, call

B∪{ℓ(a)}B \cup\{\ell(a)\}

a branch piece. Put Z=NZ = \mathbb{N} and let the complements of all branch pieces form a subbase for a topology τZ\tau_Z on ZZ.

Proposition 3.3. The space Z=(N,τZ)Z = (\mathbb{N}, \tau_Z) is countably infinite, compact, T1T_1, KC and anti-Hausdorff. Moreover every proper closed subset of ZZ is contained in a set of the form

⋃a∈Fa∪E,\bigcup_{a \in F} a \cup E,

where F⊆AF \subseteq\mathcal{A} and E⊆NE \subseteq\mathbb{N} are finite.

Proof. We record the parts of the sliced van Douwen construction that will be used later.

First, ℓ\ell is onto: apply (M3) to S=NS=\mathbb{N}. Consequently each singleton {n}\{n\} occurs as a branch piece, and hence ZZ is T1T_1.

If UU is open and ℓ(a)∈U\ell(a)\in U, then a∖Ua\setminus U is finite. It is enough to check this for subbasic open sets. If a subbasic open set containing ℓ(a)\ell(a) is the complement of B∪{ℓ(b)}B\cup\{\ell(b)\}, then b≠ab\ne a; by (M1), its complement meets aa in only finitely many points. The property is preserved under finite intersections and arbitrary unions. It follows that every branch piece B∪{ℓ(a)}B\cup\{\ell(a)\} is compact: an open cover contains one member covering ℓ(a)\ell(a), and that member misses only finitely many points of BB.

Next, every proper closed subset CC is small in the stated sense. Pick x∉Cx\notin C. Since Z∖CZ\setminus C is open, there is a basic neighbourhood of xx, given by a finite intersection of subbasic open sets, contained in Z∖CZ\setminus C. Taking complements shows that CC is contained in a finite union of branch pieces and hence in ⋃a∈Fa∪E\bigcup_{a\in F}a\cup E for suitable finite FF and EE. On the other hand, ZZ itself is not small: if FF and EE are finite, choose a∈A∖Fa\in\mathcal{A}\setminus F and then a point of aa outside E∪⋃FE\cup\bigcup F, using (M1).

Every proper closed set is compact, because it is a closed subset of a finite union of compact branch pieces and finitely many points. To see that ZZ is compact, take an open cover and choose one nonempty member UU. The complement Z∖UZ\setminus U is a proper closed, hence compact, subset and therefore has a finite subcover; adjoining UU gives a finite cover of ZZ.

Van Douwen’s argument for the topology determined by such an ω\omega-MAD family and slicing shows that ZZ is Fréchet and US; see [1 pp. 149–151], in particular properties (8)–(10) and condition (E). Hence Lemma 2.1 implies that ZZ is a KC-space. Finally, if two nonempty open sets U,VU,V were disjoint, then Z∖UZ\setminus U and Z∖VZ\setminus V would be proper closed small sets whose union is ZZ, contradicting the fact that ZZ is not small. Thus ZZ is anti-Hausdorff.

Proposition 3.4. If K⊊ZK\subsetneq Z is a kk-irreducible compact subset, then KK is a singleton.

Proof. By the KC property, KK is closed. By Proposition 3.3, there are a finite set F⊆AF\subseteq\mathcal{A} and a finite E⊆ZE\subseteq Z such that

K⊆⋃a∈Fa∪E.K\subseteq\bigcup_{a\in F}a\cup E.

For each a∈Fa\in F, define

Qa={K∩a,∣K∩a∣<∞,(K∩a)∪{ℓ(a)},∣K∩a∣=∞.Q_a= \begin{cases} K\cap a, & |K\cap a|<\infty,\\ (K\cap a)\cup\{\ell(a)\}, & |K\cap a|=\infty. \end{cases}

If K∩aK\cap a is infinite, an injective enumeration of K∩aK\cap a converges to ℓ(a)\ell(a) by the branch-convergence property in the preceding proof. Since KK is closed, ℓ(a)∈K\ell(a)\in K. Hence every QaQ_a is contained in KK, and it is compact: the finite case is immediate and the infinite case is a branch piece. Together with the finitely many singletons belonging to K∩EK\cap E, these sets give a finite cover of KK by compact subsets of KK. Since ZZ is T1T_1, all of them are saturated. Repeated use of kk-irreducibility shows that KK must coincide with one of these pieces.

A finite piece containing at least two points splits into two nonempty proper finite compact subsets, so a finite kk-irreducible piece is a singleton. If QaQ_a is infinite, choose x∈K∩ax\in K\cap a with x≠ℓ(a)x\ne\ell(a). Then

Qa={x}∪((K∩a)∖{x}∪{ℓ(a)})Q_a=\{x\}\cup\bigl((K\cap a)\setminus\{x\}\cup\{\ell(a)\}\bigr)

is a union of two nonempty proper compact subsets, again contradicting the kk-irreducibility of KK. Thus only the singleton case is possible.

By Lemma 2.2 and Proposition 3.3, ZZ itself is not co-sober.

The infinite-chain extension

Fix z0∈Zz_0\in Z. Let

Y=Z⊔N.Y=Z\sqcup\mathbb{N}.

Write the old points as μ(z)\mu(z), z∈Zz \in Z, and the new chain points as pnp_n, n∈Nn \in\mathbb{N}.

For O⊆YO \subseteq Y, put

U(O)={z∈Z:μ(z)∈O},A(O)={n∈N:pn∈O},U(O)=\{z \in Z:\mu(z) \in O\},\quad A(O)=\{n \in\mathbb{N}:p_n \in O\},

and for U⊆ZU \subseteq Z define

R(U)=Z∖(U∪{z0}).R(U)=Z\setminus(U\cup\{z_0\}).

We declare OO to be open if and only if:

(E1) U(O)U(O) is open in ZZ;

(E2) A(O)A(O) is an upper set in N\mathbb{N} with its usual order;

(E3) if z0∈U(O)z_0 \in U(O), then A(O)=NA(O)=\mathbb{N};

(E4) if A(O)≠∅A(O)\ne\varnothing, then U(O)≠∅U(O)\ne\varnothing and R(U(O))R(U(O)) is compact in ZZ.

Proposition 4.1. Conditions (E1)–(E4) define a topology on YY.

Proof. The empty set and YY satisfy the conditions. Let O,PO,P be open. Conditions (E1)–(E3) are stable under intersections. If A(O∩P)≠∅A(O\cap P)\ne\varnothing, then both U(O)U(O) and U(P)U(P) are nonempty. Since ZZ is anti-Hausdorff, U(O)∩U(P)≠∅U(O)\cap U(P)\ne\varnothing, and

R(U(O)∩U(P))=R(U(O))∪R(U(P))R(U(O)\cap U(P))=R(U(O))\cup R(U(P))

is compact. Thus O∩PO\cap P is open.

Now let O=⋃i∈IOiO=\bigcup_{i\in I}O_i, where each OiO_i is open. Conditions (E1)–(E3) are immediate. If A(O)≠∅A(O)\ne\varnothing, choose i0i_0 with A(Oi0)≠∅A(O_{i_0})\ne\varnothing. Then U(O)≠∅U(O)\ne\varnothing and

R(U(O))=R(U(Oi0))∩(Z∖U(O)).R(U(O))=R(U(O_{i_0}))\cap(Z\setminus U(O)).

Since Z∖U(O)Z\setminus U(O) is closed and R(U(Oi0))R(U(O_{i_0})) is compact, the right-hand side is compact. Hence arbitrary unions are open.

Lemma 4.2. The map μ:Z→Y\mu:Z\to Y is a topological embedding.

Proof. The old trace U(O)U(O) of every open O⊆YO\subseteq Y is open in ZZ. Conversely, let U⊆ZU\subseteq Z be open. If z0∉Uz_0\notin U, then μ(U)\mu(U) is open in YY. If z0∈Uz_0\in U, then

μ(U)∪{pn:n∈N}\mu(U)\cup\{p_n:n\in\mathbb{N}\}

is open: condition (E4) holds because R(U)=Z∖UR(U)=Z\setminus U is a closed subset of the compact space ZZ. In both cases the old trace is exactly UU.

Specialization order and the kernel operator

Proposition 5.1. For z,w∈Zz,w\in Z and m,n∈Nm,n\in\mathbb{N} one has

μ(z)≤Yμ(w)⟺z=w,\mu(z)\le_Y\mu(w)\Longleftrightarrow z=w,
pn≤Ypm⟺n≤m,p_n\le_Y p_m\Longleftrightarrow n\le m,
μ(z0)≤Ypnfor every n,\mu(z_0)\le_Y p_n\quad\text{for every }n,

and there are no other comparabilities between an old point and a new point.

Proof. The first assertion follows from Lemma 4.2 and the T1T_1 property of ZZ. Condition (E2) gives pn≤Ypmp_n\le_Y p_m whenever n≤mn\le m. For the converse, the set

On=μ(Z∖{z0})∪{pj:j≥n}O_n=\mu(Z\setminus\{z_0\})\cup\{p_j:j\ge n\}

is open by (E1)–(E4), since its residual old set is empty. If m<nm<n, then pn∈Onp_n\in O_n and pm∉Onp_m\notin O_n. Thus pn≤Ypmp_n\le_Y p_m if and only if n≤mn\le m.

Condition (E3) says that every open neighbourhood of μ(z0)\mu(z_0) contains every pnp_n, so μ(z0)≤Ypn\mu(z_0)\le_Y p_n. If z≠z0z\ne z_0, the open set μ(Z∖{z0})\mu(Z\setminus\{z_0\}) contains μ(z)\mu(z) and no chain point. Conversely, for every old point μ(z)\mu(z) there is an open set containing a prescribed pnp_n but omitting μ(z)\mu(z): use old trace Z∖{z}Z\setminus\{z\} when z≠z0z\ne z_0, and Z∖{z0}Z\setminus\{z_0\} when z=z0z=z_0, together with a suitable tail. Hence no new point specializes to an old point.

Corollary 5.2. The space YY is T0T_0.

Define

k(μ(z))=μ(z) (z∈Z),k(pn)=μ(z0) (n∈N).k(\mu(z))=\mu(z)\ (z\in Z),\quad k(p_n)=\mu(z_0)\ (n\in\mathbb{N}).

Proposition 5.3. The map k:Y→Yk:Y\to Y is a continuous kernel operator.

Proof. Idempotence is immediate. By Proposition 5.1,

k(pn)=μ(z0)≤Ypn,k(p_n)=\mu(z_0)\le_Y p_n,

while every old point is fixed. Hence k(y)≤Yyk(y)\le_Y y for every y∈Yy\in Y.

For continuity, let OO be open. If z0∉U(O)z_0\notin U(O), then

k−1(O)=μ(U(O)).k^{-1}(O)=\mu(U(O)).

If z0∈U(O)z_0\in U(O), then

k−1(O)=μ(U(O))∪{pn:n∈N}.k^{-1}(O)=\mu(U(O))\cup\{p_n:n\in\mathbb{N}\}.

Both sets satisfy (E1)–(E4), so kk is continuous.

Compact saturated sets in the extension

Put

D=sat⁡Y(μ(z0))={μ(z0)}∪{pm:m∈N},D=\operatorname{sat}_Y(\mu(z_0))=\{\mu(z_0)\}\cup\{p_m:m\in\mathbb{N}\},
Tn=sat⁡Y(pn)={pm:m≥n},T_n=\operatorname{sat}_Y(p_n)=\{p_m:m\ge n\},
Pn={μ(z):z≠z0}∪{pm:m≥n}.P_n=\{\mu(z):z\ne z_0\}\cup\{p_m:m\ge n\}.

Lemma 6.1. The sets DD, TnT_n, PnP_n and YY are nonempty compact saturated subsets of YY.

Proof. The sets DD and TnT_n are point saturations and hence supercompact. The set PnP_n is saturated by Proposition 5.1. Let U\mathcal{U} be an open cover of PnP_n and choose O0∈UO_0\in\mathcal{U} with pn∈O0p_n\in O_0. By (E2), O0O_0 contains every pmp_m with m≥nm\ge n. By (E4), R(U(O0))R(U(O_0)) is compact. The set μ(R(U(O0)))\mu(R(U(O_0))) is compact by Lemma 4.2 and is contained in PnP_n. Hence finitely many members of U\mathcal{U} cover it. Together with O0O_0, they cover PnP_n.

Finally, let V\mathcal{V} cover YY. Compactness of the old copy μ(Z)\mu(Z) yields finitely many members covering all old points. One of them contains μ(z0)\mu(z_0) and therefore, by (E3), contains every chain point. Thus YY is compact.

Define

π:Y⟶Z,π(μ(z))=z,π(pn)=z0.\pi:Y\longrightarrow Z,\quad\pi(\mu(z))=z,\quad\pi(p_n)=z_0.

The same inverse-image computation as in Proposition 5.3 shows that π\pi is continuous.

Lemma 6.2 (Full-projection classification). Let S⊆YS\subseteq Y be saturated and suppose π[S]=Z\pi[S]=Z. Then either S=YS=Y or S=PnS=P_n for some n∈Nn\in\mathbb{N}.

Proof. For each z≠z0z\ne z_0, the fibre π−1(z)\pi^{-1}(z) consists only of μ(z)\mu(z), so all these old points belong to SS. If μ(z0)∈S\mu(z_0)\in S, saturation and Proposition 5.1 force every pnp_n into SS, and hence S=YS=Y.

Assume μ(z0)∉S\mu(z_0)\notin S. Since z0∈π[S]z_0\in\pi[S], the set

NS={m:pm∈S}N_S=\{m:p_m\in S\}

is nonempty. Saturation makes NSN_S an upper set of N\mathbb{N}, so it has a least element nn and equals {m:m≥n}\{m:m\ge n\}. Thus S=PnS=P_n.

There are strict decompositions

Y=D∪P0,Pn=Tn∪Pn+1 (n∈N).Y=D\cup P_0,\quad P_n=T_n\cup P_{n+1}\ (n\in\mathbb{N}).

All terms are nonempty compact saturated sets by Lemma 6.1, and both pieces in each union are proper.

Corollary 6.3. Neither YY nor any PnP_n is kk-irreducible.

Co-sobriety of the extension

Theorem 7.1. The space YY is co-sober.

Proof. Let S∈K-Irr(Y)S \in\mathrm{K\text{-}Irr}(Y). We first show that π[S]\pi[S] is a kk-irreducible compact subset of ZZ. It is nonempty and compact by continuity of π\pi, and it is saturated because ZZ is T1T_1.

Suppose

π[S]=A∪B,\pi[S] = A \cup B,

where AA, BB are nonempty compact subsets of ZZ. Since ZZ is KC, both AA and BB are closed. Put

SA=S∩π−1(A),SB=S∩π−1(B).S_A = S \cap\pi^{-1}(A), \quad S_B = S \cap\pi^{-1}(B).

Both sets are nonempty. They are saturated because AA and BB are saturated in the T1T_1-space ZZ and inverse images of saturated sets under continuous maps are saturated. They are compact because they are closed subsets of the compact set SS. Since

S=SA∪SB,S = S_A \cup S_B,

kk-irreducibility gives S=SAS = S_A or S=SBS = S_B, and therefore π[S]=A\pi[S] = A or π[S]=B\pi[S] = B. Hence π[S]\pi[S] is kk-irreducible.

If π[S]=Z\pi[S] = Z, Lemma 6.2 gives S=YS = Y or S=PnS = P_n for some nn, contradicting Corollary 6.3. Therefore π[S]≠Z\pi[S] \ne Z. By Proposition 3.4,

π[S]={q}\pi[S] = \{q\}

for some q∈Zq \in Z.

If q≠z0q \ne z_0, the fibre π−1(q)\pi^{-1}(q) consists only of μ(q)\mu(q), and hence

S={μ(q)}=sat⁡Y(μ(q)).S = \{\mu(q)\} = \operatorname{sat}_Y(\mu(q)).

Suppose q=z0q = z_0. If μ(z0)∈S\mu(z_0) \in S, saturation gives D⊆SD \subseteq S, whereas S⊆π−1(z0)=DS \subseteq\pi^{-1}(z_0) = D; hence S=D=sat⁡Y(μ(z0))S = D = \operatorname{sat}_Y(\mu(z_0)). If μ(z0)∉S\mu(z_0) \notin S, then SS consists of chain points. Its set of levels is a nonempty upper subset of N\mathbb{N}, hence equals {m:m≥n}\{m : m \ge n\} for its least element nn. Thus

S=Tn=sat⁡Y(pn).S = T_n = \operatorname{sat}_Y(p_n).

In all cases SS is a point saturation. Conversely, every point saturation is supercompact and hence kk-irreducible: in any cover by compact saturated subsets, a member containing the point contains its entire saturation. Thus the proof gives the exact classification

K-Irr(Y)={{μ(q)}:q∈Z∖{z0}}∪{D}∪{Tn:n∈N}.\mathrm{K\text{-}Irr}(Y) = \{\{\mu(q)\} : q \in Z \setminus\{z_0\}\} \cup\{D\} \cup\{T_n : n \in\mathbb{N}\}.

By Corollary 5.2, each representing point is unique, so YY is co-sober.

The kernel image is not co-sober

By definition,

k(Y)=μ(Z).k(Y) = \mu(Z).

By Lemma 4.2, the restriction

μ:Z⟶k(Y)\mu: Z \longrightarrow k(Y)

is a homeomorphism. Since ZZ is non-singleton, compact, T1T_1, KC and anti-Hausdorff, Lemma 2.2 shows that ZZ, and hence k(Y)k(Y), is not co-sober.

Theorem 8.1. There exist a countably infinite compact co-sober T0T_0-space YY and a continuous kernel operator k:Y→Yk : Y \to Y such that k(Y)k(Y), endowed with the subspace topology, is not co-sober. Hence the answer to [7 Question 9.5] is negative.

Since k2=kk^{2}=k, the image equals the fixed-point set,

k(Y)=Fix⁡(k),k(Y)=\operatorname{Fix}(k),

and k:Y→k(Y)k:Y\to k(Y) together with the inclusion k(Y)↪Yk(Y)\hookrightarrow Y is a retraction. Moreover the composite inclusion–retraction is exactly the kernel operator k≤id⁡Yk\leq\operatorname{id}_{Y}.

Corollary 8.2. A kernel retract of a co-sober space need not be co-sober. In particular, a retract of a co-sober space need not be co-sober, so the answer to [7 Question 9.6] is negative.

Concluding remarks

The construction exhibits a sharp contrast between closure and kernel retracts. The image of a co-sober space under a continuous closure operator is co-sober [6], whereas Theorem 8.1 shows that the analogous assertion for a kernel operator fails. In retract language, closure retracts preserve co-sobriety, while kernel retracts do not.

The infinite chain supplies the decompositions used in this example. Every saturated set whose projection covers all of ZZ is, by Lemma 6.2, either YY or one of the PnP_n, and each has a nontrivial compact saturated decomposition:

Y=D∪P0,Pn=Tn∪Pn+1.Y=D\cup P_{0},\qquad P_{n}=T_{n}\cup P_{n+1}.

All remaining kk-irreducible compact saturated sets have proper projection to the seed, and the seed classification then collapses that projection to a singleton. These decompositions allow co-sobriety of the ambient space to coexist with non-co-sobriety of its kernel image.

The result suggests a broader study of preservation properties under order-compatible retracts. In particular, it is natural to compare closure retracts, kernel retracts and arbitrary retracts for other non-Hausdorff properties arising in domain theory.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

No research data were used in this theoretical study.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this manuscript, the authors used ChatGPT (OpenAI) for language editing, organization of the exposition, literature cross-checking, and assistance with checking mathematical arguments. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

References

  1. [1]E. K. van Douwen, An anti-Hausdorff Fréchet space in which convergent sequences have unique limits, Topology Appl. 51 (1993) 147–158. https://doi.org/10.1016/0166-8641(93)90147-6.
  2. [2]M. Escardó, J. D. Lawson, A. Simpson, Comparing Cartesian closed categories of (core) compactly generated spaces, Topology Appl. 143 (2004) 105–145. https://doi.org/10.1016/j.topol.2004.02.011.
  3. [3]J. Isbell, Completion of a construction of Johnstone, Proc. Amer. Math. Soc. 85 (1982) 333–334. https://doi.org/10.1090/S0002-9939-1982-0656096-4.
  4. [4]H.-P. A. Künzi, D. van der Zypen, Maximal (sequentially) compact topologies, in: W. Gähler, G. Preuss (Eds.), Categorical Structures and Their Applications, World Scientific, 2004, pp. 173–187. https://doi.org/10.1142/9789812702418_0013.
  5. [5]C. Shen, G. Wu, X. Xi, D. Zhao, Sober Scott spaces are not always co-sober, Topology Appl. 282 (2020) 107316. https://doi.org/10.1016/j.topol.2020.107316.
  6. [6]X. Wen, X. Xu, Sober is not always co-sober, Topology Appl. 250 (2018) 48–52. https://doi.org/10.1016/j.topol.2018.10.003.
  7. [7]X. Xu, D. Zhao, Some open problems on well-filtered spaces and sober spaces, Topology Appl. 301 (2021) 107540. https://doi.org/10.1016/j.topol.2020.107540.
  8. [8]B. Zhao, Z. He, K. Wang, On co-sober spaces, Houston J. Math. 49(4) (2023) 971–988.DOI

Paper details

vlnc:W_sxt45b7q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT (OpenAI)

View statement
Looking for related papers…
Contents