betaFeedback

WorksMathematics

Compact metric spaces as exponentiable metric compact Hausdorff spaces

Abstract

The class of compact metric spaces (metric spaces which induce a compact topology), despite being an important class of metric spaces, is not closed under many natural constructions. For instance, infinite metric products of compact metric spaces will often fail to induce a compact topology, while on the other hand, the topological product of any family of compact metric spaces is always a compact topological space, evidencing a discrepancy between topological and metric products. The notion of metric compact Hausdorff space – compact Hausdorff spaces equipped with a lower semi-continuous metric – provides a class of spaces which keep track of metric data and topological data separately, thereby forming a complete and cocomplete category, while including all compact metric spaces. Our main contribution is that this inclusion can be characterised via universal property: the compact metric spaces are the exponentiable metric compact Hausdorff spaces with respect to a natural monoidal structure on the respective category.

Introduction

A fruitful theory of (topological) spaces should allow for constructing new spaces out of a family of old ones; it is also often desirable for two given constructions to commute with one other (see also [5]). Prominent examples of this phenomena are the so-called “convenient categories of spaces” [23], that is, categories of spaces which are Cartesian closed (in the sense that they do admit a canonical function space [X,Y][X,Y] for all spaces XX and YY), for this property implies that “products commute with all quotients”.

As it is well-known, many important categories of topology are not Cartesian closed, motivating the study of those spaces XX – called exponentiable spaces – which do admit such a function space [X,Y][X,Y] for all YY; put more formally, those spaces XX such that the endofunctor −×X- \times X has a right adjoint [X,−][X,-]. For instance, the exponentiable topological spaces are precisely the core-compact spaces [9], and the exponentiable metric spaces are the totally quasi-convex spaces [7]. In this context, it might seem rather disappointing that the exponentiable compact Hausdorff spaces are trivial: as shown in [6], a compact Hausdorff space is exponentiable in the category CH of compact Hausdorff spaces and continuous maps if and only if it is finite. The situation does not improve if we move to the category OrdCH of Nachbin’s ordered compact Hausdorff spaces with monotone and continuous maps [19]: it is not difficult to prove that in this category the exponentiable objects are essentially the finite ones as well (see Corollary 2.11).

2020 Mathematics Subject Classification. Primary 54E45 54C35 54D30 54D15 18F60.

Key words and phrases. metric space, metric compact Hausdorff space, exponentiable, totally quasi-convex, Hausdorff metric, ultrafilter metric, Mulholland inequality, Minkowski inequality.

Surprisingly, we have more luck considering metric compact Hausdorff spaces and appropriate morphisms. These spaces constitute a simultaneous generalisation of ordered compact Hausdorff spaces and of compact metric spaces; in a nutshell, a metric compact Hausdorff space is given by a set equipped with a metric and a compact Hausdorff topology (not necessarily induced by the metric) which are compatible in an appropriate sense. We stress that every compact metric space is also a metric compact Hausdorff space, and several arguments usually applied to compact metric space can be used in this context as well (see, for instance, Proposition 1.12). This fact makes us confident that metric compact Hausdorff spaces provide a flexible generalisation of compact metric spaces. Metric compact Hausdorff spaces and the corresponding category were introduced in [12] based on [24]. However, with different motivation, similar combinations of topology and metric were also investigated in [16, 3, 4, 20].

Our main contribution is that the inclusion of compact metric spaces into metric compact Hausdorff spaces can be characterised via universal property: the compact metric spaces are the exponentiable metric compact Hausdorff space with respect to a natural monoidal structure on the category of metric compact Hausdorff spaces and morphisms. We start by characterising those metric spaces which are exponentiable with respect to a suitable monoidal structure on the category of metric spaces. We obtain an easily checkable characterization (Theorem 2.5) which allows us to prove that exponentiability for certain Minkowski tensors is equivalent to total quasi-convexity (Corollary 2.6). Building on this, we obtain our main result (Theorem 2.19): the exponentiable (with respect to a suitable monoidal structure) separated metric compact Hausdorff spaces are precisely the compact metric spaces. Hence, upholding the importance of metric compact Hausdorff spaces, the main result of our paper shows that compact metric spaces arise naturally as an important subclass of all separated metric compact Hausdorff spaces.

Funding

Funding. The authors acknowledge the partial support by CIDMA under the Portuguese Foundation for Science and Technology (FCT, https://ror.org/00snfqn58) Multi-Annual Financing Program for R&D Units, grants UID/4106/2025 and UID/PRR/4106/2025. The second author acknowledges the partial support by the Center for Mathematics of the University of Coimbra (CMUC, https://doi.org/10.54499/UID/00324/2025) under the Portuguese Foundation for Science and Technology (FCT), Grants UID/00324/2025 and UID/PRR/00324/2025.

Preliminaries

In this paper, we use the designation metric in the generalised sense of [14]; namely as a distance function

d:X×X⟶[0,∞]d: X \times X \longrightarrow[0,\infty]

such that d(x,x)=0d(x,x)=0 and d(x,y)+d(y,z)≥d(x,z)d(x,y)+d(y,z) \ge d(x,z), for all x,y,z∈Xx,y,z \in X. A metric space is a pair (X,d)(X,d) consisting of a set XX and a metric dd on XX, but we often write simply XX instead of (X,d)(X,d). A metric map f:X→Yf:X \to Y between metric spaces consists of a function f:X→Yf:X \to Y between the sets XX and YY satisfying d(x1,x2)≥d(f(x1),f(x2))d(x_1,x_2) \ge d(f(x_1),f(x_2)), for all x1,x2∈Xx_1,x_2 \in X. The category of metric spaces and metric maps is denoted by Met.

Example 1.1. Of particular importance is the metric space [0,∞][0,\infty] with the metric μ\mu defined by μ(u,v)=max⁡{v−u,0}\mu(u,v)=\max\{v-u,0\}, for all u,v∈[0,∞]u,v \in[0,\infty].

Since our notion of metric is asymmetric; that is, we generally have d(x,y)≠d(y,x)d(x,y) \ne d(y,x) for some pairs x,yx,y in a metric space XX, we have two distinct notions of open balls and closed balls, denoted by

B◃(x,r)={y∣d(y,x)<r}B▹(x,r)={y∣d(x,y)<r}B^{\triangleleft}(x,r)=\{y\mid d(y,x)<r\} \qquad B^{\triangleright}(x,r)=\{y\mid d(x,y)<r\}
B‾◃(x,r)={y∣d(y,x)≤r}B‾▹(x,r)={y∣d(x,y)≤r}\overline{B}^{\triangleleft}(x,r)=\{y\mid d(y,x)\le r\} \qquad \overline{B}^{\triangleright}(x,r)=\{y\mid d(x,y)\le r\}

which are respectively called the lower and upper open/closed balls with center x∈Xx\in X and radius r∈[0,∞]r\in[0,\infty]. By definition, the lower and upper open balls form a sub-basis for the topology on XX induced by dd, hence, the collection of all symmetric open balls

B(x,r)=B◃(x,r)∩B▹(x,r)={y∣d(x,y)<r & d(y,x)<r}B(x,r)=B^{\triangleleft}(x,r)\cap B^{\triangleright}(x,r)=\{y\mid d(x,y)<r\ \&\ d(y,x)<r\}

with center x∈Xx\in X and radius r∈[0,∞]r\in[0,\infty] form a basis for this topology. As usual, a metric space is called compact whenever this topology is compact. We also note that this topology is Hausdorff if and only if XX is separated, that is,

(0=d(x,y) & 0=d(y,x))⟹x=y(0=d(x,y)\ \&\ 0=d(y,x))\Longrightarrow x=y

for all $x,y\in X.

The category Met has a terminal object, namely the one-element space 11, and for metric spaces X,YX,Y, the product X×YX\times Y of XX and YY is given by the Cartesian product X×YX\times Y of the underlying sets equipped with the max-metric defined by

d((x1,y1),(x2,y2))=max⁡{d(x1,x2),d(y1,y2)},d((x_1,y_1),(x_2,y_2))=\max\{d(x_1,x_2),d(y_1,y_2)\},

for all x1,x2∈Xx_1,x_2\in X and $y_1,y_2\in Y.

A metric space XX is exponentiable in Met whenever the functor −×X:Met→Met-\times X:\mathrm{Met}\to\mathrm{Met} has a right adjoint, which we denote by (−)X(-)^X. The category Met is not Cartesian closed, and below Theorem 1.4 we recall the characterisation of exponentiable metric spaces obtained in [7] (see also [8]) via a quasi-convexity condition:

Definition 1.2. Let XX be a metric space.

  1. We say XX is totally quasi-convex whenever, for all x1,x2∈Xx_1,x_2\in X and all u1,u2∈[0,∞]u_1,u_2\in[0,\infty] such that d(x1,x2)=u1+u2d(x_1,x_2)=u_1+u_2, for all ε>0\varepsilon>0, there exists z∈Xz\in X such that d(x1,z)≤u1+εd(x_1,z)\le u_1+\varepsilon and d(z,x2)≤u2+εd(z,x_2)\le u_2+\varepsilon.

  2. We say XX is totally convex if for all x,y∈Xx,y\in X and all u1,u2∈[0,∞]u_1,u_2\in[0,\infty] with d(x,y)=u1+u2d(x,y)=u_1+u_2, there exists z∈Xz\in X such that d(x,z)≤u1d(x,z)\le u_1 and d(z,y)≤u2d(z,y)\le u_2.

Remark 1.3. Note that, thanks to the triangle inequality, one necessarily has d(x,z)=u1d(x,z)=u_1 and d(z,y)=u2d(z,y)=u_2 when u1+u2<∞u_1+u_2<\infty in the definition of a totally convex space. Clearly, every totally convex metric space is totally quasi-convex, and one easily verifies the converse for compact metric spaces.1

Theorem 1.4 ([7]). The following assertions are equivalent, for a metric space XX.

(i) XX is exponentiable in Met.

(ii) For all x1,x2∈Xx_1,x_2\in X and all u1,u2∈[0,∞]u_1,u_2\in[0,\infty],

inf⁡z∈X(max⁡{d(x1,z),u1}+max⁡{d(z,x2),u2})≤max⁡{d(x1,x2),u1+u2}.\inf_{z\in X}\left(\max\{d(x_1,z),u_1\}+\max\{d(z,x_2),u_2\}\right)\le\max\{d(x_1,x_2),u_1+u_2\}.

(iii) XX is totally quasi-convex.

For metric spaces XX and YY, another important metric on the Cartesian product X×YX \times Y of the sets XX and YY is the plus-metric defined by

d((x1,y1),(x2,y2)):=d(x1,x2)+d(y1,y2),d((x_1,y_1),(x_2,y_2)) := d(x_1,x_2) + d(y_1,y_2),

for all x1,x2∈Xx_1,x_2 \in X and y1,y2∈Yy_1,y_2 \in Y. We denote the resulting metric space by X⊕YX \oplus Y. In contrast to the categorical product, this construction makes Met\mathbf{Met} a closed symmetric monoidal category (see [14]); here, the right adjoint [X,−]+[X,-]_+ of −⊕X-\oplus X is given by

[X,Y]+={f:X→Y∣f is metric},d(f,g)=sup⁡x∈Xd(f(x),g(x))[X,Y]_+ = \{f:X \to Y \mid f\text{ is metric}\}, \qquad d(f,g) = \sup_{x\in X} d(f(x),g(x))

for all metric spaces XX and YY and metric maps f,g:X→Yf,g:X \to Y.

Below we investigate the interaction of exponentiability with two well-known constructions of metric spaces. Here we will use here the following notation: for a metric space XX, A,B⊆XA,B \subseteq X and ε>0\varepsilon> 0,

CA,B,ε={z∈X∣∃x∈A,y∈B. (d(x,z)≤u1+ε&d(z,y)≤u2+ε)}.C_{A,B,\varepsilon} = \{z \in X \mid\exists x \in A, y \in B.\,(d(x,z) \leq u_1 + \varepsilon\mathbin{\&} d(z,y) \leq u_2 + \varepsilon)\}.

First, we consider the metric space H(X)\mathcal{H}(X) of all subsets of XX, with the asymmetric Hausdorff metric defined by

d(A,B)=sup⁡y∈Binf⁡x∈Ad(x,y),d(A,B) = \sup_{y\in B}\inf_{x\in A} d(x,y),

for all $A,B \subseteq X ([21, 11], but see also [15]). We note that, since d(x,y)=d({x},{y})d(x,y) = d(\{x\},\{y\}) for all x,y∈Xx,y \in X, the canonical map X→H(X)X \to\mathcal{H}(X), x↦{x}x \mapsto\{x\} is an isometric embedding.

Proposition 1.5. Let XX be a metric space. Then XX is totally quasi-convex if and only if H(X)\mathcal{H}(X) is totally quasi-convex.

Proof. Assume first that XX is totally quasi-convex. Let A,B⊆XA,B \subseteq X and u1,u2∈[0,∞]u_1,u_2 \in[0,\infty] with d(A,B)=u1+u2d(A,B) = u_1 + u_2. Let ε>0\varepsilon> 0. One easily verifies that

d(A,CA,B,ε)≤u1+εandd(CA,B,ε,B)≤u2+ε.d(A,C_{A,B,\varepsilon}) \leq u_1 + \varepsilon\quad\text{and}\quad d(C_{A,B,\varepsilon},B) \leq u_2 + \varepsilon.

Assume now that H(X)\mathcal{H}(X) is totally quasi-convex. Let x1,x2∈Xx_1,x_2 \in X and u1,u2∈[0,∞]u_1,u_2 \in[0,\infty] with d(x1,x2)=u1+u2d(x_1,x_2) = u_1 + u_2. Let ε>0\varepsilon> 0. By hypothesis, there exists a subset C⊆XC \subseteq X so that

u1+ε>u1+ε2≥d({x1},C)=sup⁡z∈Cd(x1,z)u_1 + \varepsilon> u_1 + \frac{\varepsilon}{2} \geq d(\{x_1\},C) = \sup_{z\in C} d(x_1,z)

and

u2+ε>u2+ε2≥d(C,{x2})=inf⁡z∈Cd(z,x2).u_2 + \varepsilon> u_2 + \frac{\varepsilon}{2} \geq d(C,\{x_2\}) = \inf_{z\in C} d(z,x_2).

By the second inequality, there is some z0∈Cz_0 \in C with d(z0,x2)≤u2+εd(z_0,x_2) \leq u_2 + \varepsilon, by the first inequality, we also have d(x1,z0)≤u1+εd(x_1,z_0) \leq u_1 + \varepsilon. □

For a metric space XX, the set U(X)\mathcal{U}(X) of all ultrafilters on XX becomes a metric space with distance

d(x,y)=sup⁡A∈x,B∈yinf⁡x∈A,y∈Bd(x,y),d(\mathfrak{x},\mathfrak{y}) = \sup_{A\in\mathfrak{x},B\in\mathfrak{y}}\inf_{x\in A,y\in B} d(x,y),

for all x,y∈U(X)\mathfrak{x},\mathfrak{y} \in\mathcal{U}(X) (see [13 Section III.5]). In particular, for all x,y∈Xx,y \in X, we have d(x˙,y˙)=d(x,y)d(\dot{x},\dot{y}) = d(x,y), making the map X→U(X)X \to\mathcal{U}(X), x↦x˙x \mapsto\dot{x} an isometric embedding. We also note that the metric space U(X)\mathcal{U}(X) is in general not compact: for instance, if XX is discrete, then U(X)\mathcal{U}(X) is discrete.

Proposition 1.6. Let XX be a metric space. Then the following assertions are equivalent.

(i) XX is exponentiable, i.e. totally quasi-convex.

(ii) U(X)\mathcal{U}(X) is totally convex.

(iii) U(X)\mathcal{U}(X) is totally quasi-convex.

Proof. Assume first that XX is exponentiable. Let x,y∈U(X)\mathfrak{x},\mathfrak{y} \in\mathcal{U}(X) with d(x,y)=u1+u2d(\mathfrak{x},\mathfrak{y}) = u_1+u_2. The collection

{CA,B,ε∣A∈x,B∈y,ε>0}\{C_{A,B,\varepsilon} \mid A \in\mathfrak{x}, B \in\mathfrak{y}, \varepsilon> 0\}

is a filter base on XX, which can be extended to an ultrafilter z∈U(X)\mathfrak{z} \in\mathcal{U}(X). For each A∈xA \in\mathfrak{x}, C∈zC \in\mathfrak{z} and ε>0\varepsilon> 0, we have C∩CA,X,ε≠∅C \cap C_{A,X,\varepsilon} \ne\varnothing, hence, there are x∈Ax \in A and z∈Cz \in C with d(x,z)≤u1+εd(x,z) \le u_1+\varepsilon. This proves d(x,z)≤u1+εd(\mathfrak{x},\mathfrak{z}) \le u_1+\varepsilon, for each ε>0\varepsilon> 0, and therefore d(x,z)≤u1d(\mathfrak{x},\mathfrak{z}) \le u_1. A similar argument shows that d(z,y)≤u2d(\mathfrak{z},\mathfrak{y}) \le u_2. Therefore U(X)\mathcal{U}(X) is totally convex.

Clearly, if U(X)\mathcal{U}(X) is totally convex, then U(X)\mathcal{U}(X) is totally quasi-convex.

Assume now that U(X)\mathcal{U}(X) is totally quasi-convex and let x,y∈Xx,y \in X with d(x,y)=u1+u2d(x,y) = u_1+u_2 and let ε>0\varepsilon> 0. By hypothesis, there exists z∈U(X)\mathfrak{z} \in\mathcal{U}(X) such that d(x,z)≤u1+ε2d(x,\mathfrak{z}) \le u_1+\frac{\varepsilon}{2} and d(z,y)≤u2+ε2d(\mathfrak{z},y) \le u_2+\frac{\varepsilon}{2}. Therefore, for every C∈zC \in\mathfrak{z},

inf⁡z∈Cd(x,z)<u1+εandinf⁡z∈Cd(z,y)<u2+ε.\inf_{z\in C}d(x,z) < u_1+\varepsilon\quad\text{and}\quad\inf_{z\in C}d(z,y) < u_2+\varepsilon.

Put D={z∈X∣d(x,z)≤u1+ε}D=\{z\in X\mid d(x,z)\le u_1+\varepsilon\}. The first inequality above implies that D∈zD\in\mathfrak{z}, and then the second one entails that there exits some z∈Dz\in D with d(z,y)≤u2+εd(z,y)\le u_2+\varepsilon. We conclude that XX is totally quasi-convex. □\square

In the remainder of this section we will consider the main objects of interest for this paper: metric spaces equipped with an additional compact Hausdorff topology.

Definition 1.7. A metric compact Hausdorff space XX consists of a compact Hausdorff space XX together with a lower semicontinuous metric

d:X×X⟶[0,∞],d:X\times X\longrightarrow[0,\infty],

that is, the metric dd is continuous with respect to the topology on [0,∞][0,\infty] generated by the basic closed sets [0,u][0,u] (u∈[0,∞]u\in[0,\infty]).

Example 1.8. The metric space [0,∞][0,\infty] of Example 1.1 becomes a metric compact Hausdorff space with the Euclidean topology on [0,∞][0,\infty], that is, the topology generated by the basic closed sets [0,u][0,u] and [u,∞][u,\infty] (u∈[0,∞]u\in[0,\infty]). We note that this topology differs from the one induced by the metric μ\mu, which is not compact.

Example 1.9. For each metric space XX, the metric space U(X)\mathcal{U}(X) becomes a metric compact Hausdorff space with the Zariski topology (also called hull-kernel topology, see [10]), that is, the topology with basic open sets

{x∈U(X)∣A∈x}(A⊆X).\{x\in\mathcal{U}(X)\mid A\in x\}\quad(A\subseteq X).

We find it worthwhile to point out that this topology only depends on the underlying set of XX and not on its metric.

Example 1.10. Recall from [19] that an ordered compact Hausdorff space is a compact Hausdorff space XX equipped with an order relation ≤\le (i.e. a reflexive and transitive relation on XX) such that

{(x,x′)∣x≤x′}⊆X×X\{(x,x')\mid x\le x'\}\subseteq X\times X

is closed in the product space $X\times X.

Every ordered compact Hausdorff space can be viewed as a metric compact Hausdorff space via the metric dd defined by

d(x,y)={0if x≤y,∞if x≰y,d(x,y)= \begin{cases} 0 & \text{if } x\leq y,\\ \infty& \text{if } x\not\leq y, \end{cases}

for all x,y∈Xx,y\in X.

Remark 1.11. If XX be a metric compact Hausdorff space, then the closed balls are closed but the open balls need not be open (see Proposition 1.14). Furthermore, if XX is separated, then the metric topology refines the topology of XX (see [3 Lemma 1.4]).

The following result is a slight generalisation of Remark 1.3.

Proposition 1.12. Totally quasi-convex metric compact Hausdorff spaces are totally convex.

Proof. Let XX be a totally quasi-convex metric compact Hausdorff space, and let x1,x2∈Xx_1,x_2\in X and u1,u2∈[0,∞]u_1,u_2\in[0,\infty] such that d(x1,x2)=u1+u2d(x_1,x_2)=u_1+u_2. Hence, for all ε>0\varepsilon>0, there exists z∈Xz\in X such that d(x1,z)≤u1+εd(x_1,z)\leq u_1+\varepsilon and d(z,x2)≤u2+εd(z,x_2)\leq u_2+\varepsilon. For every ε>0\varepsilon>0, we consider the set

Aε={y∈X∣d(x1,y)≤u1+ε and d(y,x2)≤u2+ε}A_\varepsilon=\{y\in X\mid d(x_1,y)\leq u_1+\varepsilon\text{ and }d(y,x_2)\leq u_2+\varepsilon\}
=d(x1,−)−1([0,u1+ε])∩d(−,x2)−1([0,u2+ε]),=d(x_1,-)^{-1}([0,u_1+\varepsilon])\cap d(-,x_2)^{-1}([0,u_2+\varepsilon]),

since the metric d:X×X→[0,∞]d:X\times X\to[0,\infty] is lower semicontinuous, AεA_\varepsilon is closed in XX. By hypothesis, Aε≠∅A_\varepsilon\neq\varnothing for each ε>0\varepsilon>0, and since Aε⊆AδA_\varepsilon\subseteq A_\delta whenever ε≤δ\varepsilon\leq\delta, the collection of these sets is a filter basis. Thus, we may choose an ultrafilter z\mathfrak{z} which contains these sets.

Since XX is compact Hausdorff, z\mathfrak{z} converges to (a unique) z∈Xz\in X. Hence, for every ε>0\varepsilon>0, we have that z∈Aεz\in A_\varepsilon, that is, d(x1,z)≤u1+εd(x_1,z)\leq u_1+\varepsilon and d(z,x2)≤u2+εd(z,x_2)\leq u_2+\varepsilon. Thus, we obtain d(x1,z)≤u1d(x_1,z)\leq u_1 and d(z,x2)≤u2d(z,x_2)\leq u_2.

Remark 1.13. As it is shown in [12], a metric compact Hausdorff space can be equivalently described as a metric space XX, equipped with a compact Hausdorff topology, so that the ultrafilter convergence α:U(X)→X\alpha:\mathcal{U}(X)\to X is a metric map. This description allows for a slightly shorter argument in the proof of Proposition 1.12, based on Proposition 1.6. If d(x1,x2)=u1+u2d(x_1,x_2)=u_1+u_2, then there is some ultrafilter z\mathfrak{z} on XX with d(x1,z)≤u1d(x_1,\mathfrak{z})\leq u_1 and d(z,x2)≤u2d(\mathfrak{z},x_2)\leq u_2. Therefore d(x1,α(z))≤u1d(x_1,\alpha(\mathfrak{z}))\leq u_1 and d(α(z),x2)≤u2d(\alpha(\mathfrak{z}),x_2)\leq u_2.

The following result follows from standard arguments in metric spaces.

Proposition 1.14. Let XX be a compact Hausdorff space equipped with a metric d:X×X→[0,∞]d:X\times X\to[0,\infty]. Then the following assertions are equivalent.

(i) The upper/lower balls of the metric space XX are open in the topology of XX.

(ii) The symmetric open balls of XX are open in the topology of XX.

(iii) The map d:X×X→[0,∞]d:X\times X\to[0,\infty] is continuous with respect to the Euclidean topology of [0,∞][0,\infty].

We denote by MetCH the category of metric compact Hausdorff spaces and metric and continuous maps. Both the max-metric and the plus-metric endow MetCH with a symmetric monoidal structure; however, as we show in the next section, MetCH not closed for either of these structures. For a metric space XX, the function d:X×X→[0,∞]d: X \times X \to[0,\infty] is continuous with respect to the topology on XX induced by the metric and the Euclidean topology on [0,∞][0,\infty], therefore every compact separated metric space is a metric compact Hausdorff space when equipped with the topology induced by the metric. We denote the full subcategory of Met formed by all compact separated metric spaces by CompMet, and therefore we have a full embedding

CompMet⟶MetCH.\mathrm{CompMet} \longrightarrow\mathrm{MetCH}.

From Proposition 1.14 and Remark 1.11 we obtain the following observation.

Lemma 1.15. A separated metric compact Hausdorff space is compact metric if and only if any (and hence all) of the three assertions of Proposition 1.14 hold.

We conclude this section with a metric version of Urysohn’s Lemma essentially obtained in [16 Theorem 2.4] (see also [4 Theorem 1.6]).

Theorem 1.16. Let XX be a metric compact Hausdorff space and A,B⊆XA, B \subseteq X closed subsets. Put r:=inf⁡{d(x,y)∣x∈A,y∈B}r := \inf\{d(x,y) \mid x \in A, y \in B\}. Then there exists a continuous metric map f:X→[0,r]f: X \to[0,r] with A⊆f−1(0)A \subseteq f^{-1}(0) and B⊆f−1(r)B \subseteq f^{-1}(r).

Corollary 1.17. The set {[0,r]∣0<r<∞}\{[0,r] \mid0 < r < \infty\} and the space [0,∞][0,\infty] are initially dense in MetCH.

Main results

To treat various monoidal structures on Met and MetCH simultaneously, we consider here an arbitrary ordered commutative monoid structure ⊗\otimes on [0,∞][0,\infty] with neutral element 00. Furthermore, we assume that ⊗\otimes satisfies Mulholland’s inequality2 [17, 18] (a generalization of Minkowski’s inequality (Example 2.2)),

(u1+v1)⊗(u2+v2)≤(u1⊗u2)+(v1⊗v2),(u_1+v_1)\otimes(u_2+v_2) \leq(u_1\otimes u_2)+(v_1\otimes v_2),

for all u1,u2,v1,v2∈[0,∞]u_1,u_2,v_1,v_2 \in[0,\infty].

Indeed, this inequality ensures that the function d:(X×Y)×(X×Y)→[0,∞]d: (X \times Y) \times(X \times Y) \to[0,\infty] defined by

d((x,y),(x′,y′)):=X(x,x′)⊗Y(y,y′),d((x,y),(x',y')) := X(x,x') \otimes Y(y,y'),

for all x,x′∈Xx,x' \in X and y,y′∈Yy,y' \in Y, is a metric on the product space X×YX \times Y. Naturally, the reflexivity condition holds, and so does the triangle inequality: we have

d((x,y),(x′,y′))=X(x,x′)⊗Y(y,y′)≤(X(x,x′′)+X(x′′,x′))⊗(Y(y,y′′)+Y(y′′,y′))≤(X(x,x′′)⊗Y(y,y′′))+(X(x′′,x′)⊗Y(y′′,y′))=d((x,y),(x′′,y′′))+d((x′′,y′′),(x′,y′)).\begin{aligned} d((x,y),(x',y')) &= X(x,x') \otimes Y(y,y') \\ &\leq(X(x,x'')+X(x'',x')) \otimes(Y(y,y'')+Y(y'',y')) \\ &\leq(X(x,x'')\otimes Y(y,y''))+(X(x'',x')\otimes Y(y'',y')) \\ &= d((x,y),(x'',y''))+d((x'',y''),(x',y')). \end{aligned}

for all x,x′,x′′∈Xx,x',x'' \in X and y,y′,y′′∈Yy,y',y'' \in Y. We denote this metric space by X⊗YX \otimes Y, and we note that this defines a symmetric monoidal structure on the category Met.

Lemma 2.1. If ⊗\otimes is an ordered commutative monoid structure on [0,∞][0,\infty] satisfying Mulholland’s inequality, then:

  • we have a+b≥a⊗b≥max⁡(a,b)a+b \geq a\otimes b \geq\max(a,b) for all a,b∈[0,∞]a,b \in[0,\infty],

  • ⊗\otimes defines a metric map [0,∞]⊗[0,∞]→[0,∞][0,\infty] \otimes[0,\infty] \to[0,\infty].

Proof. For the inequalities, simply note that, for all a,b∈[0,∞]a,b \in[0,\infty],

a+b=(a⊗0)+(0⊗b)≥(a+0)⊗(0+b)=a⊗b,a+b=(a\otimes0)+(0\otimes b)\geq(a+0)\otimes(0+b)=a\otimes b,

and that monotonicity of ⊗\otimes guarantees a⊗b≥a⊗0=aa\otimes b\geq a\otimes0=a, and a⊗b≥0⊗b=ba\otimes b\geq0\otimes b=b. The second reduces to proving that μ(a⊗c,b⊗d)≤μ(a,b)⊗μ(c,d)\mu(a\otimes c,b\otimes d)\leq\mu(a,b)\otimes\mu(c,d) for all a,b,c,d∈[0,∞]a,b,c,d\in[0,\infty]. This follows by monotonicity and Mulholland’s inequality. □\square

This yields numerous consequences. For one, we have that the identity functions A⊕B→A⊗B→A×BA\oplus B\to A\otimes B\to A\times B are metric maps, and therefore the topology on A⊗BA\otimes B induced by the metric is the product topology. It also follows that ⊗:[0,∞]×[0,∞]→[0,∞]\otimes:[0,\infty]\times[0,\infty]\to[0,\infty] is a continuous map, and therefore this tensor product defines a symmetric monoidal structure on MetCH\mathrm{MetCH} as well.

Additionally, we also find that the map −⊗u:[0,∞]→[0,∞]-\otimes u:[0,\infty]\to[0,\infty] preserves infima for all u∈[0,∞]u\in[0,\infty]. Thus, −⊗u-\otimes u has a right adjoint with respect to the “greater or equal relation” on [0,∞][0,\infty], which we denote by [u,−]⊗:[0,∞]→[0,∞][u,-]_{\otimes}:[0,\infty]\to[0,\infty]. In other words, Mulholland’s inequality ensures ⊗\otimes is a commutative quantale on [0,∞][0,\infty] with order relation ≥\geq and unit 00 (see [22]).

We call a metric (compact Hausdorff) space ⊗\otimes-exponentiable whenever −⊗X-\otimes X has a right adjoint, denoted by [X,−]⊗[X,-]_{\otimes}.

Example 2.2. (a) We may consider ⊗=+\otimes=+ and ⊗=max⁡\otimes=\max.

(b) For 1<p<∞1<p<\infty, we may consider u⊗pv=up+vppu\otimes_p v=\sqrt[p]{u^p+v^p}, thanks to Minkowski’s inequality:

(u1+v1)p+(u2+v2)pp≤u1p+u2pp+v1p+v2pp.\sqrt[p]{(u_1+v_1)^p+(u_2+v_2)^p}\leq\sqrt[p]{u_1^p+u_2^p}+\sqrt[p]{v_1^p+v_2^p}.

which additionally states that the equality holds if and only if u1v2=u2v1u_1v_2=u_2v_1. We point out that ⊗1=+\otimes_1=+ and lim⁡p→∞⊗p=max⁡\lim_{p\to\infty}\otimes_p=\max.

(c) From [18 Theorem 1], we deduce that if g:[0,∞]→[g(0),∞]g:[0,\infty]\to[g(0),\infty] is a strictly increasing, continuous, convex and geoconvex3 function, then the binary operation defined by x⊗y=g−1(g(x)+g(y)−g(0))x\otimes y=g^{-1}(g(x)+g(y)-g(0)) is an ordered commutative monoid structure on [0,∞][0,\infty] satisfying Mulholland’s inequality. We have g(x)=exg(x)=e^x as an example that is not a power function.

(d) [2 Table 2.6] lists several more examples4, as one may check.

Remark 2.3. By general adjunction rules, there is a bijection [X,Y]⊗≅Met(X,Y)[X,Y]_{\otimes}\cong\mathrm{Met}(X,Y), and the diagrams

[X,Y]⊗⊗X→counitY≅↓↑≅Met(X,Y)⊗X→evY\begin{CD} [X,Y]_{\otimes}\otimes X @>{\mathrm{counit}}>> Y \\ @V{\cong}VV @AA{\cong}A \\ \mathrm{Met}(X,Y)\otimes X @>{\mathrm{ev}}>> Y \end{CD}

and

Y→ηMet(X,Y⊗X)≅↑↑≅[X,Y⊗X]⊗→unitMet(X,Y⊗X)\begin{CD} Y @>{\eta}>> \mathrm{Met}(X,Y\otimes X) \\ @A{\cong}AA @AA{\cong}A \\ [X,Y\otimes X]_{\otimes} @>{\mathrm{unit}}>> \mathrm{Met}(X,Y\otimes X) \end{CD}

commute. Here, the canonical map η:Y→Met(X,Y⊗X)\eta:Y\to\mathrm{Met}(X,Y\otimes X) sends y∈Yy\in Y to the metric map defined by x↦(y,x)x\mapsto(y,x). For an ⊗\otimes-exponentiable metric compact Hausdorff space, the same assertions hold, with MetCH(X,Y)\mathrm{MetCH}(X,Y) in place of Met(X,Y)\mathrm{Met}(X,Y). Henceforth, we take the underlying set of [X,Y]⊗[X,Y]_{\otimes} to be equal to Met(X,Y)\mathrm{Met}(X,Y) respectively MetCH(X,Y)\mathrm{MetCH}(X,Y), and the unit and the counit to be given by η\eta and ev\mathrm{ev}, respectively.

Below we give a slight generalisation of Theorem 1.4.

Theorem 2.4. If XX be a metric space, then the following are equivalent:

(i) XX is ⊗\otimes-exponentiable in Met.

(ii) For all x1,x2∈Xx_1,x_2 \in X and all u1,u2∈[0,∞]u_1,u_2 \in[0,\infty],

inf⁡z∈X((d(x1,z)⊗u1)+(d(z,x2)⊗u2))≤d(x1,x2)⊗(u1+u2).\inf_{z\in X} \left((d(x_1,z)\otimes u_1)+(d(z,x_2)\otimes u_2)\right) \le d(x_1,x_2)\otimes(u_1+u_2).

If XX is ⊗\otimes-exponentiable in Met, then the metric of [X,Y]⊗[X,Y]_\otimes is given by

d(f,g)=sup⁡x,x′∈X[d(x,x′),d(f(x),g(x′))]⊗.(1)d(f,g)=\sup_{x,x'\in X}[d(x,x'),d(f(x),g(x'))]_\otimes. \tag*{(1)}

Proof. For a ⊗\otimes-exponentiable metric space XX, its counit ev⁡:[X,Y]⊗⊗X→Y\operatorname{ev}: [X,Y]_\otimes\otimes X\to Y is a metric map, hence

d(f,g)≥sup⁡x,x′∈X[d(x,x′),d(f(x),g(x′))]⊗⏟=u.d(f,g)\ge\underbrace{\sup_{x,x'\in X}[d(x,x'),d(f(x),g(x'))]_\otimes}_{=u}.

Consider now the metric space Z={0,1}Z=\{0,1\} with d(0,1)=ud(0,1)=u and d(1,0)=∞d(1,0)=\infty. Then the map

X⊗Z⟶Y,(x,0)↦f(x), (x,1)↦g(x)X\otimes Z\longrightarrow Y,\qquad(x,0)\mapsto f(x),\ (x,1)\mapsto g(x)

is metric, hence also the map

Z⟶[X,Y]⊗,0↦f, 1↦gZ\longrightarrow[X,Y]_\otimes,\qquad0\mapsto f,\ 1\mapsto g

is metric, and therefore u≥d(f,g)u\ge d(f,g). We conclude that a metric space XX is ⊗\otimes-exponentiable in Met if and only if the distance defined by (2.1) satisfies the triangle inequality, for all metric spaces YY. With the same proof as in [7] one verifies that this is indeed equivalent to the assertion (ii) above. In fact, an easy calculation shows that (ii) guarantees the dd satisfies the triangle inequality, for all metric spaces YY.

Assume now that XX is ⊗\otimes-exponentiable. We consider the maps

φ1:X⟶[0,∞],x↦d(x1,x),\varphi_1:X\longrightarrow[0,\infty],\quad x\mapsto d(x_1,x),
φ2:X⟶[0,∞],x↦d(x1,x)⊗u1,\varphi_2:X\longrightarrow[0,\infty],\quad x\mapsto d(x_1,x)\otimes u_1,
φ3:X⟶[0,∞],x↦inf⁡z∈X((d(x1,z)⊗u1)+(d(z,x)⊗u2)).\varphi_3:X\longrightarrow[0,\infty],\quad x\mapsto\inf_{z\in X}\left((d(x_1,z)\otimes u_1)+(d(z,x)\otimes u_2)\right).

for each x1,x2∈Xx_1,x_2\in X and u1,u2∈[0,∞]u_1,u_2\in[0,\infty]. Then φ1,φ2,φ3\varphi_1,\varphi_2,\varphi_3 are in fact metric maps since, for all x∈Xx\in X, all u∈[0,∞]u\in[0,\infty] and all sets II, the maps

d(x,−):X⟶[0,∞],−⊗u:[0,∞]⟶[0,∞],inf⁡:[0,∞]I⟶[0,∞]d(x,-):X\longrightarrow[0,\infty],\qquad-\otimes u:[0,\infty]\longrightarrow[0,\infty],\qquad\inf:[0,\infty]^I\longrightarrow[0,\infty]

are metric. For all metric maps φ,φ′:X→[0,∞]\varphi,\varphi':X\to[0,\infty] and u∈[0,∞]u\in[0,\infty], we have u≥d(φ,φ′)u\ge d(\varphi,\varphi') if and only if, for all x,x′∈Xx,x'\in X, u⊗d(x,x′)≥μ(φ(x),φ′(x′))u\otimes d(x,x')\ge\mu(\varphi(x),\varphi'(x')), which in turn is equivalent to (u⊗d(x,x′))+φ(x)≥φ′(x′)(u\otimes d(x,x'))+\varphi(x)\ge\varphi'(x'), for all x,x′∈Xx,x'\in X. From that we conclude that u1≥d(φ1,φ2)u_1\ge d(\varphi_1,\varphi_2) and u2≥d(φ2,φ3)u_2\ge d(\varphi_2,\varphi_3). Since dd satisfies the triangle inequality, u1+u2≥d(φ1,φ3)u_1+u_2\ge d(\varphi_1,\varphi_3). Therefore

(u1+u2)⊗d(x1,x2)=((u1+u2)⊗d(x1,x2))+φ1(x1)≥φ3(x2)=inf⁡z∈X((d(x1,z)⊗u1)+(d(z,x2)⊗u2)).\begin{aligned} (u_1+u_2)\otimes d(x_1,x_2)=((u_1+u_2)\otimes d(x_1,x_2))+\varphi_1(x_1) \\ &\ge\varphi_3(x_2)=\inf_{z\in X}\left((d(x_1,z)\otimes u_1)+(d(z,x_2)\otimes u_2)\right). \end{aligned}

We remark that it is even possible to characterise ⊗\otimes-exponentiability for a space XX in terms of a more useful metric condition on XX. Intuitively, XX must simultaneously “almost” satisfy the converses of the triangle inequality and Mulholland’s inequality.

To make this intuition precise, we fix the following notation for a subset S⊆XS \subseteq X:

Bε(S)={x∈X∣∃s∈S:d(x,s)<ε and d(s,x)<ε},B_{\varepsilon}(S)=\{x\in X\mid\exists s\in S : d(x,s)<\varepsilon\text{ and }d(s,x)<\varepsilon\},
B‾ε(S)={x∈X∣∃s∈S:d(x,s)≤ε and d(s,x)≤ε}.\overline{B}_{\varepsilon}(S)=\{x\in X\mid\exists s\in S : d(x,s)\leq\varepsilon\text{ and }d(s,x)\leq\varepsilon\}.

Additionally, we consider the continuous function

Ωd,u1,u2:[0,d]⟶[0,∞],r⟼r⊗u1+(d−r)⊗u2\Omega_{d,u_{1},u_{2}}:[0,d]\longrightarrow[0,\infty],\quad r\longmapsto r\otimes u_{1}+(d-r)\otimes u_{2}

and note that we have Ω(r)≥d⊗(u1+u2)\Omega(r)\geq d\otimes(u_{1}+u_{2}) for all rr, by Mulholland’s inequality. With this in mind, we define

Md,u1,u2=Ω−1(d⊗(u1+u2))={r∈[0,d]∣r⊗u1+(d−r)⊗u2=d⊗(u1+u2)}.M_{d,u_{1},u_{2}}=\Omega^{-1}\bigl(d\otimes(u_{1}+u_{2})\bigr)=\{r\in[0,d]\mid r\otimes u_{1}+(d-r)\otimes u_{2}=d\otimes(u_{1}+u_{2})\}.

Note that M0,u1,u2={0}M_{0,u_{1},u_{2}}=\{0\}, M∞,u1,u2=[0,∞]M_{\infty,u_{1},u_{2}}=[0,\infty], Md,0,0=[0,d]M_{d,0,0}=[0,d], and Md,∞,u2=Md,u1,∞=[0,d]M_{d,\infty,u_{2}}=M_{d,u_{1},\infty}=[0,d]. These are the cases d=0d=0, d=∞d=\infty, u1+u2=0u_{1}+u_{2}=0 and u1+u2=∞u_{1}+u_{2}=\infty, respectively. We call these the boundary conditions.

Theorem 2.5. Assume that a⊗x<b⊗xa\otimes x<b\otimes x whenever a<ba<b and x<∞x<\infty.5 The following are equivalent for a metric space XX:

  1. XX is ⊗\otimes-exponentiable.

  2. for all x,y∈Xx,y\in X, all u1,u2∈[0,∞]u_{1},u_{2}\in[0,\infty] and all ε>0\varepsilon>0, there exists z∈Xz\in X with d(x,z)+d(z,y)≤d(x,y)+εd(x,z)+d(z,y)\leq d(x,y)+\varepsilon and d(x,z)∈B‾ε(Md(x,y),u1,u2)d(x,z)\in\overline{B}_{\varepsilon}(M_{d(x,y),u_{1},u_{2}}).

Proof. We use Theorem 2.4. To prove (i) implies (ii), let x,y∈Xx,y\in X, let u1,u2∈[0,∞]u_{1},u_{2}\in[0,\infty], and let ε>0\varepsilon>0. Under the boundary conditions, there is nothing to show; hence, we assume 0<d(x,y)<∞0<d(x,y)<\infty, and 0<u1+u2<∞0<u_{1}+u_{2}<\infty.

We write M=Md(x,y),u1,u2M=M_{d(x,y),u_{1},u_{2}}; if B‾ε(M)=[0,d(x,y)]\overline{B}_{\varepsilon}(M)=[0,d(x,y)], there is nothing to check; else, we take

η1=min⁡Ω∣[0,d(x,y)]∖B‾ε(M),η2=(d(x,y)+ε)⊗(u1+u2),\eta_{1}=\min\Omega\big|_{[0,d(x,y)]\setminus\overline{B}_{\varepsilon}(M)},\qquad\eta_{2}=(d(x,y)+\varepsilon)\otimes(u_{1}+u_{2}),

highlighting that if Ω(a)≤η1\Omega(a)\leq\eta_{1}, then a∈B‾ε(M)a\in\overline{B}_{\varepsilon}(M).

Taking ω=min⁡(η1,η2)−d(x,y)⊗(u1+u2)>0\omega=\min(\eta_{1},\eta_{2})-d(x,y)\otimes(u_{1}+u_{2})>0, by ⊗\otimes-exponentiability we find z∈Xz\in X with

d(x,z)⊗u1+d(z,y)⊗u2≤d(x,y)⊗(u1+u2)+ω=min⁡(η1,η2).d(x,z)\otimes u_{1}+d(z,y)\otimes u_{2}\leq d(x,y)\otimes(u_{1}+u_{2})+\omega=\min(\eta_{1},\eta_{2}).

We have d(x,y)−d(x,z)≤d(z,y)d(x,y)-d(x,z)\leq d(z,y), from which we deduce that Ω(d(x,z))≤η1\Omega(d(x,z))\leq\eta_{1}, and hence d(x,z)∈B‾ε(M)d(x,z)\in\overline{B}_{\varepsilon}(M). Moreover, we also have

(d(x,z)+d(z,y))⊗(u1+u2)≤d(x,z)⊗u1+d(z,y)⊗u2≤η2,(d(x,z)+d(z,y))\otimes(u_{1}+u_{2})\leq d(x,z)\otimes u_{1}+d(z,y)\otimes u_{2}\leq\eta_{2},

from which we deduce that d(x,z)+d(z,y)≤d(x,y)+εd(x,z)+d(z,y)\leq d(x,y)+\varepsilon.

To show (ii) implies (i), let x,y∈Xx,y\in X, u1,u2∈[0,∞]u_{1},u_{2}\in[0,\infty], and let η>0\eta>0. With nothing to prove under the boundary conditions, we assume 0<d(x,y)<∞0<d(x,y)<\infty, and 0<u1+u2<∞0<u_{1}+u_{2}<\infty.

Let ε1=η2>0\varepsilon_{1}=\frac{\eta}{2}>0. Since Ω\Omega is a continuous function defined on a compact interval, there exists ε2>0\varepsilon_{2}>0 such that if r∈B‾ε2(M)r\in\overline{B}_{\varepsilon_{2}}(M), then Ω(r)<d(x,y)⊗(u1+u2)+ε1\Omega(r)<d(x,y)\otimes(u_{1}+u_{2})+\varepsilon_{1}. Indeed, for all x∈Mx\in M, there exists δ>0\delta>0 such that for all yy with ∣x−y∣<δ\lvert x-y\rvert<\delta, we have Ω(y)−Ω(x)<ε1\Omega(y)-\Omega(x)<\varepsilon_{1}. Now, by compactness, we find finitely many δ\delta; take ε2\varepsilon_{2} to be the smallest.

So, with ε=min⁡(ε1,ε2)>0\varepsilon=\min(\varepsilon_{1},\varepsilon_{2})>0, via (ii) we find z∈Xz\in X such that d(x,z)∈B‾ε2(M)d(x,z)\in\overline{B}_{\varepsilon_{2}}(M) and d(x,z)+d(z,y)≤d(x,y)+ε1d(x,z)+d(z,y)\leq d(x,y)+\varepsilon_{1}. Thus, we have that

d(x,z)⊗u1+d(z,y)⊗u2≤d(x,z)⊗u1+(d(x,y)−d(x,z)+η2)⊗u2≤Ω(d(x,z))+η2≤d(x,y)⊗(u1+u2)+η.\begin{aligned} d(x,z) \otimes u_1 + d(z,y) \otimes u_2 \le d(x,z) \otimes u_1 + \left(d(x,y) - d(x,z) + \frac{\eta}{2}\right) \otimes u_2 \\ &\le\Omega(d(x,z)) + \frac{\eta}{2} \le d(x,y) \otimes(u_1 + u_2) + \eta. \end{aligned}

as desired. □

As a corollary, we characterise ⊗p\otimes_p-exponentiability:

Corollary 2.6. If p>1p > 1, a metric space XX is ⊗p\otimes_p-exponentiable if and only if XX is totally quasi-convex.

Proof. As noted in Example 2.2(b), we have Md,u1,u2={d⋅u1u1+u2}M_{d,u_1,u_2} = \left\{d \cdot\frac{u_1}{u_1+u_2}\right\} for 0<u1+u2<∞0 < u_1+u_2 < \infty and 0<d<∞0 < d < \infty, by Minkowski’s inequality. One then immediately checks that Theorem 2.5(ii) is equivalent to total quasi-convexity. □

Corollary 2.7. Under the hypothesis of Theorem 2.5, if Md,u1,u2=∅M_{d,u_1,u_2} = \varnothing outside the boundary conditions, then a metric space is ⊗\otimes-exponentiable if and only if it is discrete.

Remark 2.8. We conjecture that Md,u1,u2=∅M_{d,u_1,u_2} = \varnothing outside the boundary conditions for all tensors not of the form ⊗p\otimes_p.

In the remainder of this section we analyse ⊗\otimes-exponentiability for metric compact Hausdorff spaces; we start by providing a necessary condition.

Proposition 2.9. Let XX be a metric compact Hausdorff space. Then (1) ⟹\Longrightarrow (2) ⟹\Longrightarrow (3) below.

(1) XX is ⊗\otimes-exponentiable on MetCH.

(2) For every metric compact Hausdorff space YY, MetCH⁡(X,Y)\operatorname{MetCH}(X,Y) is a closed subspace of the product space ∏x∈XY\prod_{x\in X}Y.

(3) The lower/upper open balls are open in the topology of XX.

Proof. Assume first (1). For every x∈Xx \in X, ev⁡x:[X,Y]⊗→Y\operatorname{ev}_x : [X,Y]_{\otimes} \to Y is an arrow in MetCH, in particular it is continuous, hence we obtain a canonical injective and continuous map MetCH⁡(X,Y)→∏x∈XY\operatorname{MetCH}(X,Y) \to\prod_{x\in X}Y. By compactness, this map is a closed embedding.

Assume now (2). Hence, for each x∈Xx \in X, the map πx:MetCH⁡(X,[0,∞])→[0,∞]\pi_x : \operatorname{MetCH}(X,[0,\infty]) \to[0,\infty], f↦f(x)f \mapsto f(x) is continuous. For every ε<ε0\varepsilon< \varepsilon_0 and U⊆XU \subseteq X open with B▹(x0,ε0)⊆UB^{\triangleright}(x_0,\varepsilon_0) \subseteq U we put

CU,ε:={h:X→[0,∞] in MetCH∣h(x0)=0 and h(z)≥ε for all z∈Uc}.C_{U,\varepsilon}:=\{h:X\to[0,\infty]\text{ in MetCH}\mid h(x_0)=0\text{ and }h(z)\ge\varepsilon\text{ for all }z\in U^c\}.

By Theorem 1.16, CU,ε≠∅C_{U,\varepsilon} \ne\varnothing; moreover,

CU,ε∩CV,δ⊇CU∩V,max⁡{ε,δ}.C_{U,\varepsilon} \cap C_{V,\delta} \supseteq C_{U\cap V,\max\{\varepsilon,\delta\}}.

Therefore there exists an ultrafilter x\mathfrak{x} on MetCH⁡(X,[0,∞])\operatorname{MetCH}(X,[0,\infty]) such that CU,ε∈xC_{U,\varepsilon} \in\mathfrak{x} for all ε<ε0\varepsilon< \varepsilon_0 and U⊆XU \subseteq X open with B▹(x0,ε0)⊆UB^{\triangleright}(x_0,\varepsilon_0) \subseteq U. By compactness, x\mathfrak{x} converges to some f∈MetCH⁡(X,[0,∞])f \in\operatorname{MetCH}(X,[0,\infty]). We have the following facts.

(1) f(x0)=0f(x_0)=0. In fact, for every A∈xA \in\mathfrak{x}, A∩CX,ε02≠∅A \cap C_{X,\frac{\varepsilon_0}{2}} \ne\varnothing, that is, there is some k∈Ak \in A with k(x0)=0k(x_0)=0. Therefore πx0(x)\pi_{x_0}(\mathfrak{x}) is the principal ultrafilter generated by 0. Since πx0(x)→f(x0)\pi_{x_0}(\mathfrak{x}) \to f(x_0), we conclude that f(x0)=0f(x_0)=0.

(2) f(x)<ε0f(x)<\varepsilon_0 for every x∈B▹(x0,ε0)x \in B^{\triangleright}(x_0,\varepsilon_0). In fact, since ff is a metric map,

f(x)=f(x)−f(x0)≤X(x0,x)<ε0.f(x)=f(x)-f(x_0)\le X(x_0,x)<\varepsilon_0.
f(y)≥ε0for every y∉B▹(x0,ε0).(2)f(y) \ge\varepsilon_0 \quad\text{for every } y \notin B^{\triangleright}(x_0,\varepsilon_0). \tag*{(2)}

To see this, note first that, by Theorem 1.16, for every ε<ε0\varepsilon< \varepsilon_0 there exists an arrow hε:X→[0,∞]h_\varepsilon:X \to[0,\infty] in MetCH with hε(x0)=0h_\varepsilon(x_0)=0 and hε(y)≥εh_\varepsilon(y)\ge\varepsilon. Put

U=⋃ε<ε0hε−1([0,ε[).U=\bigcup_{\varepsilon<\varepsilon_0}h_\varepsilon^{-1}([0,\varepsilon[).

Then U⊆XU\subseteq X is open, B▹(x0,ε0)⊆UB^{\triangleright}(x_0,\varepsilon_0)\subseteq U and y∉Uy\notin U.

Let now ε<ε0\varepsilon<\varepsilon_0. For every A∈xA\in\mathfrak{x}, A∩CU,ε≠∅A\cap C_{U,\varepsilon}\ne\varnothing, that is, there is some k∈Ak\in A with k(y)≥εk(y)\ge\varepsilon. Therefore [ε,∞]∈πy(x)[\varepsilon,\infty]\in\pi_y(\mathfrak{x}), since πy(x)→f(y)\pi_y(\mathfrak{x})\to f(y), we conclude that f(y)≥εf(y)\ge\varepsilon. Consequently, f(y)≥ε0f(y)\ge\varepsilon_0.

Putting everything together, we have shown that B▹(x0,ε0)=f−1([0,ε0[)B^{\triangleright}(x_0,\varepsilon_0)=f^{-1}([0,\varepsilon_0[) is open in XX. A similar argument shows that B◃(x0,ε0)B^{\triangleleft}(x_0,\varepsilon_0) is open in XX. □

Using Lemma 1.15 and Proposition 2.9, we deduce immediately the following result.

Theorem 2.10. Let XX be a ⊗\otimes-exponentiable separated metric compact Hausdorff space. Then XX is a compact metric space.

Using Example 1.10, we conclude the following fact for ordered compact Hausdorff spaces.

Corollary 2.11. Let XX be an exponentiable ordered compact Hausdorff space. Then, for every x∈Xx\in X, the sets ↓x={y∈X∣y≤x}\downarrow x=\{y\in X\mid y\le x\} and ↑x={y∈X∣x≤y}\uparrow x=\{y\in X\mid x\le y\} are open in XX. In particular, if XX is antisymmetric, then XX is finite.

Now that we have a better understanding of the topological structure of ⊗\otimes-exponentiable (separated) metric compact Hausdorff spaces, we expose their metric structure.

Proposition 2.12. A ⊗\otimes-exponentiable metric compact Hausdorff space is also ⊗\otimes-exponentiable in Met.

Proof. The underlying set of [X,Y]⊗[X,Y]_\otimes is necessarily given by MetCH⁡(X,Y)\operatorname{MetCH}(X,Y), and the same argument as in the proof of Theorem 2.4 shows that the distance of [X,Y]⊗[X,Y]_\otimes is given by (1), using that the space ZZ is finite. In fact, as in the proof of Theorem 2.4, we show that an exponentiable metric compact space satisfies Condition (ii) of Theorem 2.4. First observe that, since XX is ⊗\otimes-exponentiable, the metric d:X×X→[0,∞]d:X\times X\to[0,\infty] is even continuous (see Propositions 2.9 and 1.14). Since ⊗:[0,∞]×[0,∞]→[0,∞]\otimes:[0,\infty]\times[0,\infty]\to[0,\infty] is continuous, we obtain immediately that the metric functions φ1,φ2:X→[0,∞]\varphi_1,\varphi_2:X\to[0,\infty] defined in the proof of Theorem 2.4 are continuous. Regarding φ3:X→[0,∞]\varphi_3:X\to[0,\infty], observe first that the map

ψ:X×X⟶[0,∞],(z,x)⟼(d(x1,z)⊗u1)+(d(z,x)⊗u2)\psi:X\times X\longrightarrow[0,\infty],\qquad(z,x)\longmapsto(d(x_1,z)\otimes u_1)+(d(z,x)\otimes u_2)

is continuous. With [0,∞]X[0,\infty]^X denoting the exponential in the category of topological spaces and continuous maps (which can be taken since XX is compact Hausdorff), its exponential mate

ψ‾:X⟶[0,∞]X\overline{\psi}:X\longrightarrow[0,\infty]^X

is also continuous. Furthermore, inf⁡:[0,∞]X→[0,∞]\inf:[0,\infty]^X\to[0,\infty] is continuous since XX is compact (see the proof of [1 Proposition 4.1]), therefore φ3=inf⁡⋅ψ‾\varphi_3=\inf\cdot\overline{\psi} is continuous. As in the proof of Theorem 2.4, we conclude that

(u1+u2)⊗d(x1,x2)≥inf⁡z∈X((d(x1,z)⊗u1)+(d(z,x2)⊗u2))(u_1+u_2)\otimes d(x_1,x_2)\ge\inf_{z\in X}\bigl((d(x_1,z)\otimes u_1)+(d(z,x_2)\otimes u_2)\bigr)

for all x1,x2∈Xx_1,x_2 \in X and all u1,u2∈[0,∞]u_1,u_2 \in[0,\infty].

We will now also see that the assertion (ii) of Theorem 2.4 implies ⊗\otimes-exponentiability in MetCH, for compact separated metric spaces.

Lemma 2.13. Let XX be a metric space and YY be a metric compact Hausdorff space. Then the subset Met⁡(X,Y)⊆∏x∈XY\operatorname{Met}(X,Y) \subseteq\prod_{x\in X}Y is closed in the product space ∏x∈XY\prod_{x\in X}Y.

Proof. Let f∈Y∣X∣f \in Y^{|X|} with f∉Met⁡(X,Y)f \notin\operatorname{Met}(X,Y). Then there exist x1,x2∈Xx_1,x_2 \in X with d(x1,x2)<d(f(x1),f(x2))d(x_1,x_2) < d(f(x_1),f(x_2)). Put ε=d(x1,x2)\varepsilon=d(x_1,x_2), then d(f(x1),f(x2))=ε+ηd(f(x_1),f(x_2))=\varepsilon+\eta with η>0\eta>0. By Theorem 1.16, there exists h:X→[0,∞]h:X\to[0,\infty] in MetCH with h(f(x1))=0h(f(x_1))=0 and h(f(x2))=ε+ηh(f(x_2))=\varepsilon+\eta. Put U=h−1([0,η2))U=h^{-1}([0,\frac{\eta}{2})) and V=h−1(ε+η2,∞]V=h^{-1}(\frac{\varepsilon+\eta}{2},\infty], and note that UU and VV are open subsets of XX with f(x1)∈Uf(x_1)\in U and f(x2)∈Vf(x_2)\in V. Moreover, for all y1∈Uy_1\in U and y2∈Vy_2\in V, d(y1,y2)≥h(y2)−h(y1)>εd(y_1,y_2)\ge h(y_2)-h(y_1)>\varepsilon. Therefore W=πx1−1(U)∩πx2−1(V)W=\pi_{x_1}^{-1}(U)\cap\pi_{x_2}^{-1}(V) is an open neighbourhood of ff with W∩Met⁡(X,Y)=∅W\cap\operatorname{Met}(X,Y)=\varnothing. We conclude that Met⁡(X,Y)\operatorname{Met}(X,Y) is closed in Y∣X∣Y^{|X|}.

Corollary 2.14. For every metric space XX and every metric compact Hausdorff space YY, we have that Met⁡(X,Y)\operatorname{Met}(X,Y) is a metric compact Hausdorff space with the structure inherited from ∏x∈XY\prod_{x\in X}Y.

Lemma 2.15. If XX is a compact separated metric space and YY is a metric compact Hausdorff space, then the canonical map η:Y→Met⁡(X,Y⊗X)\eta:Y\to\operatorname{Met}(X,Y\otimes X) is continuous.

Lemma 2.16. Let XX and YY be compact separated metric spaces. Then the map ev⁡:Met⁡(X,Y)⊗X→Y\operatorname{ev}:\operatorname{Met}(X,Y)\otimes X\to Y is continuous.

Proof. Let f:X→Yf:X\to Y be a metric map, x∈Xx\in X and V⊆YV\subseteq Y an open neighbourhood of f(x)f(x). Since YY is a compact metric space, there is some ε>0\varepsilon>0 so that B(f(x),2ε)⊆VB(f(x),2\varepsilon)\subseteq V. Then W:=πx−1(B(f(x),ε))×B(x,ε)W:=\pi_x^{-1}(B(f(x),\varepsilon))\times B(x,\varepsilon) is an open neighbourhood of (f,x)(f,x) and, for all (h,z)∈W(h,z)\in W,

d(h(z),f(x))≤d(h(z),h(x))+d(h(x),f(x))≤2εd(h(z),f(x))\le d(h(z),h(x))+d(h(x),f(x))\le2\varepsilon

and d(f(x),h(z))≤2εd(f(x),h(z))\le2\varepsilon. Therefore ev⁡(W)⊆V\operatorname{ev}(W)\subseteq V.

With the help of Corollary 1.17, we extend Lemma 2.16 to all metric compact Hausdorff spaces YY.

Lemma 2.17. Let XX be a compact separated metric space and YY be a metric compact Hausdorff space. Then the map ev⁡:Met⁡(X,Y)⊗X→Y\operatorname{ev}:\operatorname{Met}(X,Y)\otimes X\to Y is continuous.

Proof. Let φ:Y→[0,r]\varphi:Y\to[0,r] be a continuous and metric map. First observe that map Met⁡(X,φ)⊗1X:Met⁡(X,Y)⊗X⟶Met⁡(X,[0,r])⊗X, (f,x)↦(φf,x)\operatorname{Met}(X,\varphi)\otimes1_X:\operatorname{Met}(X,Y)\otimes X\longrightarrow\operatorname{Met}(X,[0,r])\otimes X,\ (f,x)\mapsto(\varphi f,x) is continuous (and metric). Secondly, the diagram

Met⁡(X,Y)⊗X→ev⁡YMet⁡(X,φ)⊗1X↓↓φMet⁡(X,[0,r])⊗X→ev⁡[0,r]\begin{CD} \operatorname{Met}(X,Y)\otimes X @>{\operatorname{ev}}>> Y \\ @V{\operatorname{Met}(X,\varphi)\otimes1_X}VV @VV{\varphi}V \\ \operatorname{Met}(X,[0,r])\otimes X @>{\operatorname{ev}}>> [0,r] \end{CD}

commutes. Since [0,r][0,r] is a compact separated metric space, ev⁡:Met⁡(X,[0,r])⊗X→[0,r]\operatorname{ev}:\operatorname{Met}(X,[0,r])\otimes X\to[0,r] is continuous (Lemma 2.16) and therefore also ev⁡⋅(Met⁡(X,φ)⊗1X)\operatorname{ev}\cdot(\operatorname{Met}(X,\varphi)\otimes1_X). Since the cone of all metric and continuous maps Y→[0,r]Y\to[0,r] separates the points of YY (Corollary 1.17), ev⁡:Met⁡(X,Y)⊗X→Y\operatorname{ev}:\operatorname{Met}(X,Y)\otimes X\to Y is also continuous.

Lemma 2.18. If XX is a compact metric space, ⊗\otimes-exponentiable in Met, and YY is a metric compact Hausdorff space, then [X,Y]⊗[X,Y]_{\otimes} is a metric compact Hausdorff space.

Proof. First note that, for all x,x′∈Xx,x' \in X, the composite

[X,Y]⊗×[X,Y]⊗→ev⁡x×ev⁡x′Y×Y→d[0,∞]→[X(x,x′),−]⊗[0,∞][X,Y]_{\otimes} \times[X,Y]_{\otimes} \xrightarrow{\operatorname{ev}_{x} \times\operatorname{ev}_{x'}} Y \times Y \xrightarrow{d} [0,\infty] \xrightarrow{[X(x,x'),-]_{\otimes}} [0,\infty]

is lower semicontinuous, and hence

[X,Y]⊗×[X,Y]⊗→d∏x,x′∈X[0,∞]→sup⁡[0,∞][X,Y]_{\otimes} \times[X,Y]_{\otimes} \xrightarrow{d} \prod_{x,x'\in X}[0,\infty] \xrightarrow{\sup} [0,\infty]

is lower semicontinuous as well. □

Putting everything together:

Theorem 2.19. We consider a continuous commutative quantale structure ⊗\otimes on [0,∞][0,\infty] with neutral element 00 satisfying Mulholland’s inequality, that is,

(u1+u2)⊗(v1+v2)≤(u1⊗v1)+(u2⊗v2),for all u1,u2,v1,v2∈[0,∞].(u_1+u_2)\otimes(v_1+v_2) \le(u_1\otimes v_1)+(u_2\otimes v_2), \qquad\text{for all }u_1,u_2,v_1,v_2\in[0,\infty].

Then, a separated metric compact Hausdorff space is ⊗\otimes-exponentiable in MetCH if and only if it is compact metric and ⊗\otimes-exponentiable in Met.

Corollary 2.20. In particular, a separated metric compact Hausdorff space is

  • ++-exponentiable in MetCH if and only if it is compact metric.

  • ⊗p\otimes_p-exponentiable in MetCH if and only if it is compact metric and totally convex, for 1<p≤∞1 < p \le\infty.

References

CIDMA, Department of Mathematics, University of Aveiro, 3810-193 Aveiro, Portugal

Email address: [email protected]

CMUC, Department of Mathematics, University of Coimbra, 3000-143 Coimbra, Portugal

Email address: [email protected]

References

  1. [1]Marco Abbadini and Dirk Hofmann, Barr-coexactness for metric compact Hausdorff spaces, Theory and Applications of Categories 44 (2025), no. 6, 196–226.arxiv.org/abs/2408.07039
  2. [2]Claudia Alsina, Maurice J. Frank, and Berthold Schweizer, Associative functions. triangular norms and copulas, Hackensack, NJ: World Scientific, 2006 (English).
  3. [3]Itai Ben Yaacov, Topometric spaces and perturbations of metric structures, Logic and Analysis 1 (2008), no. 3–4, 235–272.
  4. [4]———, Lipschitz functions on topometric spaces, Journal of Logic and Analysis (2013), 1–21.
  5. [5]Herschel L. Bentley, Horst Herrlich, and Robert Lowen, Improving constructions in topology, Category theory at work (Horst Herrlich and Hans-Eberhard Porst, eds.), Research and Exposition in Mathematics, vol. 18, Heldermann Verlag, Berlin, 1991, With Cartoons by Marcel Erné, pp. 3–20.
  6. [6]Francesca Cagliari and Sandra Mantovani, Local homeomorphisms as the exponentiable morphisms in compact Hausdorff spaces, Topology and its Applications 41 (1991), no. 3, 263–272.DOI
  7. [7]Maria Manuel Clementino and Dirk Hofmann, Exponentiation in V-categories, Topology and its Applications 153 (2006), no. 16, 3113–3128.DOI
  8. [8]Maria Manuel Clementino, Dirk Hofmann, and Isar Stubbe, Exponentiable functors between quantaloid-enriched categories, Applied Categorical Structures 17 (2009), no. 1, 91–101.
  9. [9]Brian J. Day and G. Max Kelly, On topologically quotient maps preserved by pullbacks or products, Proceedings of the Cambridge Philosophical Society 67 (1970), no. 03, 553–578.DOI
  10. [10]Leonard Gillman and Meyer Jerison, Rings of continuous functions, The University Series in Higher Mathematics. Princeton-Toronto-London-New York: D. Van Nostrand Company, Inc. IX, 1960 (English).
  11. [11]Felix Hausdorff, Grundzüge der Mengenlehre, Veit & Comp, Leipzig, 1914.
  12. [12]Dirk Hofmann and Carla D. Reis, Convergence and quantale-enriched categories, Categories and General Algebraic Structures with Applications 9 (2018), no. 1, 77–138.arxiv.org/abs/1705.08671
  13. [13]Dirk Hoffmann, Gavin J. Seal, and Walter Tholen (eds.), Monoidal Topology. A Categorical Approach to Order, Metric, and Topology, Encyclopedia of Mathematics and its Applications, vol. 153, Cambridge University Press, Cambridge, jul 2014, Authors: Maria Manuel Clementino, Eva Colebunders, Dirk Hoffmann, Robert Lowen, Rory Lucyshyn-Wright, Gavin J. Seal and Walter Tholen.
  14. [14]F. William Lawvere, Metric spaces, generalized logic, and closed categories, Rendiconti del Seminario Matematico e Fisico di Milano 43 (1973), no. 1, 135–166, Republished in: Reprints in Theory and Applications of Categories, No. 1 (2002), 1–37.DOI
  15. [15]Sandro Levi, Roberto Lucchetti, and Jan Pelant, On the infimum of the Hausdorff and Vietoris topologies, Proceedings of the American Mathematical Society 118 (1993), no. 3, 971–978 (English).DOI
  16. [16]Eva Matoušková, Extensions of continuous and Lipschitz functions, Canadian Mathematical Bulletin 43 (2000), no. 2, 208–217.DOI
  17. [17]Theodore Mostkzin, Sur le produit d’espaces métriques, Comptes rendus du Congrès International des Mathématiciens, Oslo, vol. 2, 1936, pp. 137–138.
  18. [18]Hugh P. Mulholland, On generalizations of Minkowski’s inequality in the form of a triangle inequality, Proceedings of the London Mathematical Society s2-51 (1949), no. 1, 294–307.DOI
  19. [19]Leopoldo Nachbin, Topology and Order, Van Nostrand Mathematical Studies, vol. 4, D. Van Nostrand, Princeton, N.J.-Toronto, Ont.-London, 1965, Translated from the Portuguese by Lulu Bechtolsheim.
  20. [20]Enrico Pasqualetto, Timo Schultz, and Janne Taipalus, A categorical perspective on extended metric-topological spaces, Tech. report, 2026.arxiv.org/abs/2601.07505
  21. [21]Dimitrie Pompeiu, Sur la continuité des fonctions de variables complexes, Annales de la Faculté des Sciences de l’Université de Toulouse pour les Sciences Mathématiques et les Sciences Physiques. 2ième Série 7 (1905), no. 3, 265–315.
  22. [22]Kimmo I. Rosenthal, Quantales and their applications, Pitman research notes in mathematics series, no. 234, Longman Scientific & Technical, Harlow, Essex, 1990.
  23. [23]Norman E. Steenrod, A convenient category of topological spaces, Michigan Mathematical Journal 14 (1967), no. 2, 133–152.DOI
  24. [24]Walter Tholen, Ordered topological structures, Topology and its Applications 156 (2009), no. 12, 2148–2157.DOI

Paper details

vlnc:W_7czdw7yk
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

Not recorded for this paper

View statement
Looking for related papers…
Contents