@

Appl. Gen. Topol. 26, no. 1 (2025), 71-85

doi:10.4995/agt.2025.20840

edUPV

Infinity valued partial metrics

Steve Matthews a and Homeira Pajoohesh b

This paper is dedicated to the memory of Ralph Kopperman (1942-2021)

a Coventry, U.K. (redpolo1@googlemail.com)

b Department of Mathematics, Medgar Evers College, CUNY, Brooklyn, NY 11225,
USA (hpajoohesh@mec.cuny.edu)

Communicated by Ó. Valero

Abstract
A partial metric is a generalised metric incorporating non-zero self-distance.
It was introduced by Matthews in Computer Science, and connected well with
Kopperman’s long standing interest in generalised metrics and non-Hausdorff
topology. In this article we further generalise partial metrics by incorporating
∞ as a distance. We prove that a so-called infinite partial metric induces a
partial metrizable topology, but not in general vice verse.

2020 MSC: 54H30; 68U03; 54E99.

Keywords: partial metrics; infinite distance functions.

1. An Introduction and Motivation for Partial Metric Spaces

1.1. Self-distance in metric spaces. To introduce and motivate partial metrics we have to
approach them from both the perspectives of mathematics and computer science until, so to
speak, they meet in the middle. To his great credit Kopperman was not only a very accomplished
mathematician, but also able to step outside his own discipline to recognise a question had
arisen for the essence of mathematics. In 1975 Matthews’ first degree was an unsurprising one

Received 6 December 2023 – Accepted 18 September 2024

 S. Matthews and H. Pajoohesh

in UK mathematics as it contained an introductory course on metric and topological spaces.

Seemingly implied to students was that any ’interesting’ topology was most probably Hausdorff
separable. That is x 6= y if and only if there exist disjoint open sets containing each of x
and y. When expressed using a metric d : U × U → [0, ∞), ∀x, y : x = y ⇔ d(x, y) = 0.
Had Hausdorff separability become the norm in applications of general topology? Kopperman,
whom the authors of this paper worked with closely on partial metrics, thought that non-
Hausdorff separability had uses yet to be discovered. The pioneering use of non-Hausdorff
topology in Scott’s theory of domains [14] for modelling the denotational semantics of computer
programming languages [15] had long since justified Kopperman’s position. What interest was
there among mathematicians for non-Hausdorff topology?
Matthews had no reason to challenge this established state of affairs until studying for his
doctorate in Computer Science at Warwick University UK in 1980. This was an application in
programming language design which involved a metric space of infinite sequences where distance
is defined as follows. For any two sequences having longest common initial segment of length
n their distance is 2−n . In accordance with metric space reflexivity, ∀x . d(x, x) = 2−∞ = 0.
Unfortunately this reflexive property does not extend to sequences x of length n < ∞ where
if we define d(x, x) = 2−n self-distance is not zero. Why we need to model distance for finite
sequences in order to study infinite sequences is the crux of an impending problem. Are we
somehow to define a larger metric space to include finite sequences? To his utter frustration
this was the ultimately futile approach tried by Matthews for a long time. With the benefit of
hindsight there is a simple reason why this happened.

1.2. Ontology in mathematics and computer science. Consider the ontology of the data
studied in each of mathematics and computer science. In mathematics statements are made
about well defined entities. In computer science however each data item has to be constructed
bit by bit, and so, in general, is not well defined during a computation. A computer program
has to be written and executed to construct each data item. Computer science thus has to
model both the well defined final result of constructing a data item as well as the history of the
developing construction. To mathematically model each infinite sequence on a computer in our
example the means (inspired by Scott) and assumed by Matthews was defined to be a series of
so-called approximations. Each approximation is a ’finite’ initial segment of the infinite sequence
being constructed.
Now we can begin to see what a metric space can model in the context of computer science.
In accordance with the metric reflexive axiom self-distance can be defined to be zero when there

Appl. Gen. Topol. 26, no. 1 (2025) 72

 Infinity valued partial metrics

is no more work to do, that is the data item is, in the sense of mathematics, well defined. In
our working example here a ’finite’ sequence is not an end product of computation, leading us
to ask for an additional ontology to mathematically model partially defined data and partially
defined computations. An analogy is a time-lapse video recording each partially built state of the
construction of a new building. Thus, as new buildings cannot be instantly created, a necessary
ontological question to ask is whether each photo taken is showing an entity existing fully (i.e.
a building site) or a partially existing building? As the term partial metric suggests Matthews
eventually settled upon defining, so to speak, partially defined metric spaces. This being achieved
by dropping the metric reflexivity axiom. At this point in our discussion it is interesting to
reflect briefly upon Dana Scott’s ground breaking mathematical theory of computation [15].
His immense achievement was to give the first mathematical model for the untyped λ-calculus
which formed the underlying logic for early models of computation. Scott (as we understand it)
mathematically modelled data items and their computation treating each partially computed
data item as a well defined mathematical entity. In contrast our (much more limited) work
approaches computation from the alternative perspective of partializing, so to speak, well defined
mathematics, for which we gratefully acknowledge here Scott’s personal interest. We must be
clear that Matthews’ work was restricted to the study of metric spaces in computer science while
Scott has a much more general and comprehensive theory of computation.

1.3. The history of a computation. We have now argued our case for the difficulties in
using metric spaces for modelling computation. More generally put, as Scott shows in his work,
the reason is that a topological approach to modelling partiality in computation requires non-
Hausdorff separability. In our working example of computing an infinite sequence using a series
of finite sequences we have yet to explain how this relates to non-Hausdorff separability. A
history here begins with the empty sequence hi adding data values one by one. It is a simple
linear model of computation where each value added can never be changed or removed. Let V
be a set of data values. For each n ∈ ω let xn : {0, ..., n − 1} → V be a finite sequence of length
n, also denoted hxn0 , . . . , xnn−1 i.
Here we should pause for an ontological reflection. In using finite sequences we are, as is
customary in mathematics, presuming them to be a totally defined concept. What we really
want is to consider each as being a partially defined infinite sequence termed a partial sequence.
The important distinction between a finite sequence and a partial sequence is that the former
is completely defined while the latter (as far as we can presently say) is only partially defined.

Appl. Gen. Topol. 26, no. 1 (2025) 73

 S. Matthews and H. Pajoohesh

Hence our interest in weak notions of separability. We now introduce the following to model
partial sequences.

Definition 1.1. Let V⊥ := V ∪ {⊥} where ⊥ 6∈ V is termed the undefined value. For each
n ∈ ω, xn : ω → V⊥ is defined to be a partial sequence over V if

∀i < n . xni ∈ V ∧ ∀i ≥ n . xni = ⊥

Each xn : ω → V is defined to be a well defined sequence. The length of a partial sequence

x (denoted |x|) is defined to be the largest n ∈ ω such that xn−1 ∈ V . The length of a well
defined sequence is ∞. The set of all partial and well defined sequences over V is denoted V⊥∞ .

Definition 1.2. For each V the information (partial) ordering vV ⊆ V⊥∞ × V⊥∞ is such that

x v y ⇔ |x| ≤ |y| ∧ ∀i < |x| . xi = yi

The chain x0 v x1 v . . . v x is the history of the computation of x.

Clearly there is an isomorphism between our two approaches to modelling finite and partial
sequences. The first approach sees an infinite sequence as the ’limit’ of a series of finite sequences
while the second sees a partial sequence as an approximation to a well defined sequence. With
the partial modelling approach we can say that if x v y then y is (in general) ’separable’ from
x but, most importantly, x is not separable from y. This is simply because each data value of x
is in the same position in y, but not (in general) vice versa. And thus the inherent symmetry of
the usual -ball topology of metric spaces is most unlikey to work for our partial metric spaces.
All in all, akin to Kopperman, we ask whether traditional totally defined mathematics can be
generalised to incorporate approaches from the partially defined perspective.

1.4. Defining partial metric spaces. We have argued so far that an asymmetric notion of
separability has been raised by the theory of computation. If x v y then x cannot (and can
never) be separated from y while y is separable from x having (in general) more data values.
Thus, as we shall soon see, T0 -separability became our way forward. If x v y what is the
difference between distances for x and those for y? As we observed earlier self-distance for finite
sequences decreases as computation progresses toward completion and zero self-distance. And
so, at last, we reach the following definition of a metric generalised to incorporate non-zero
self-distance.

Definition 1.3. A partial metric [9, 3] is a function p : U × U → [0, ∞) such that

(1) ∀x, y ∈ U . p(x, x) ≤ p(x, y)

Appl. Gen. Topol. 26, no. 1 (2025) 74

 Infinity valued partial metrics

(2) ∀x, y ∈ U . p(x, y) = p(y, x)

(3) ∀x, y ∈ U . p(x, x) = p(x, y) = p(y, y) ⇒ x = y
(4) ∀x, y, z ∈ U . p(x, z) ≤ p(x, y) + p(y, z) − p(y, y)

Although each partial metric p is symmetric we can define a partial ordering as follows.
x v y ⇔ p(x, x) = p(x, y). A metric can be defined as a partial metric for which all self-
distances are zero. In the context of computer science we can understand each p(x, x) as being
a quantification of the extent to which x is complete. A simple example of a partial metric
is for three-valued logic. Let p : {⊥, F, T }2 → {0, 1} be the function such that p(x, y) = 0 iff
x = y = F or x = y = T . For our working example of sequences let pV : V⊥∞ × V⊥∞ → [0, ∞)
be such that pV (x, y) = 2−min{|x|,|y|} .
The usual notion of -ball from metric spaces can be generalised based on the quasi metric qp
defined by qp (x, y) = p(x, y) − p(x, x) which induces a T0 topology on X as follows:
τp = {W ⊆ X : if x ∈ W ⇒ ∃ > 0, N (x) ⊆ W }, where N (x) = {y ∈ X : p(x, y) − p(x, x) ≤
}; see [9] and [7] .

Example 1.4. For the partial metric p(x, y) = x ∨ y on the real numbers, one can verify that
τp = {(−∞, a) : a ∈ R} is the lower topology on R.

The last axiom of partial metric implies that τp is a T0 topological space. However, not every
T0 topology is induced by a partial metric. As is mentioned in [1], every point of a set equipped
with a partial metrizable topology must have a countable local base. It is straightforward to
prove that the cofinite topology on the set of real numbers is a T0 topology such that none of
its points has a countable local base and therefore, is not a partial metrizable topology.

1.5. A Fuss about Nothing. Partial metrics illustrate our long awaited meeting point of
mathematics and computer science. From Kurt Gödel’s Incompleteness Theorem we can see
the unavoidable development of partially defined data in the rise of the computer age. The
undefined nothing value ⊥ finally takes its rightful place in mathematics. Our particular choice
of approach to Section 1 of our article is a consequence of the pivotal role played by the notion
of weak separability in both topology and computer science. Our work on partial metric spaces
is an example of how the separability of data in computation is not a binary consideration. We
can no longer (in general) say that for any x, y either x = y or x 6= y and there is no other
possibility. Yet Hausdorff separability is binary as (by definition) either x = y or there exist
disjoint open sets containing each of x and y to separate them. The experience of computer
science is that in general disjoint open sets are not tenable as, may we put it, (in general) data

Appl. Gen. Topol. 26, no. 1 (2025) 75

 S. Matthews and H. Pajoohesh

items are only partially separable. T0 -separability and partial metric spaces were sufficient in
their time, but the vast complexity of computational models today call for more sophisticated
notions of weak (i.e. non-Hausdorff) separability. Kopperman [8] appreciated the importance
of non-binary separability in topology, and so encouraged Matthews’ work on partial metric
spaces. Computer science has, for its own reasons, been heralding the call for more research into
non-Hausdorff topology. Kopperman was a mathematicians having a broad insight to call for
more research into non-Hausdorff general topology. This article now considers a generalisation
of partial metrics. The motivation for this work is to explore a type of partial metric for Beer’s
work in [2].

2. Partial metrics valued into [0, ∞]

In this section we define a new generalization of partial metrics and prove that the topology
induced by these generalized partial metrics is partial metrizable.
Partial metrics have been generalized in several ways for instance in [7] partial metrics were
generalized by allowing p : X × X → V , where V is a value quantale or a value lattice. There, it
was shown that each continuous poset has a partial metric p on it valued into a value quantale
such that its Scott and lower topologies can be obtained from the partial metric p.
In [6], partial metrics on `-groups (lattice ordered groups) were discussed. There it was shown
if Pn is the variety of `-groups such that the function P : G×G → G defined by p(x, y) = n(x∨y),
n ∈ N is a partial metric, then Pn ⊆ En , where En is the variety of `-groups that their nth power
commutes; if G ∈ En then na + nb = nb + na for all a, b ∈ G. There it was shown that if n
is prime then the two varieties are equal. Later in [5], it was shown that the two varieties are
equal if and only if n is prime. This paper led to more research on varieties of `-groups related
to partial metrics, for example see [4] and [12]. These papers built another bridge between
computer science and order theory.
In [11] O’Neill let partial metrics take values in the real numbers rather than [0, ∞). In
[10] this generalization of partial metrics was referred to as dualistic partial metrics and in [16]
generalization of Banach fixed point theorems for these partial metrics were proved.
Here we consider a new generalization of partial metrics and then we prove that the induced
topology by these generalized partial metrics is partial metrizable.

Definition 2.1. A function e : X × X → [0, ∞] is called an infinite partial metric if it obeys all
of the axioms of partial metrics except that it is valued into [0, ∞] rather than [0, ∞).

Appl. Gen. Topol. 26, no. 1 (2025) 76

 Infinity valued partial metrics

Most of these generalizations of all types of metrics are generalizing real numbers into the
structures more general than the real numbers. But in [2], Beer considers metrics into infinite
real numbers ,[0, ∞], rather than the reals with the property of ∞+∞ = a+∞ = ∞
non-negative
0, if a = 0;
and a · ∞ = .
∞, if a ∈ (0, ∞]
Here we consider partial metrics valued into infinite real numbers so that [0, ∞] has the same
a+∞
algebraic properties plus b+∞ = 1 for every a, b ∈ [0, ∞).

Lemma 2.2. If p : X × X → [0, ∞] is an infinite partial metric, then, there is at most one
element with self distance of infinity.

Proof. Suppose p(a, a) = p(b, b) = ∞ for a, b ∈ X. Then, from p(a, a) ≤ p(a, b) we conclude
p(a, b) = ∞. Thus, p(a, a) = p(b, b) = p(a, b) = ∞. Then the last property of partial metrics
implies a = b. 

Let p : X × X → [0, ∞] be an infinite partial metric. Define

τp,∞ = {W ⊆ X : x ∈ W ⇒ ∃ ∈ (0, ∞)(N∞ (x) ⊆ W )}, where N∞ (x) = {y ∈ X : p(x, y) ≤
p(x, x) + }.
It is straightforward to prove the next lemma and we leave it to the reader.

Lemma 2.3. If p : X × X → [0, ∞] is an infinite partial metric, then τp,∞ is a topology on X.

We need the following lemma for the proof of our main theorem, Theorem 2.5.

Lemma 2.4. Let a, b, c, d ∈ [0, ∞). If d ≤ b, c and a + d ≤ b + c then

(1) ad ≤ bc
1 1 1
(2) a + d ≥ b + 1c .
a d b c
(3) 1+a + 1+d ≤ 1+b + 1+c .

Proof. Either b ≤ c or c ≤ b. Without lose of generality assume b ≤ c.

Part (a): Let δ = b − d. Then,

ad = a(b − δ) ≤ (b + c − d)(b − δ) = (c + δ)(b − δ) = bc + bδ − cδ − δ 2 ≤ bc + cδ − cδ − δ 2 ≤ bc.

1 1 1 1 1 1 b+c b+c b+c 1 1

Part (b): a + d ≥ b+c−d + d = c+δ + b−δ = (b−δ)(c+δ) ≥ bc+cδ−δ 2 −cδ
≥ bc = b + c

Part (c): An algebraic calculation shows that the inequality of this part is equivalent to prove
that a + d + 2ad + abd + acd ≤ b + c + 2bc + abc + bcd.

Appl. Gen. Topol. 26, no. 1 (2025) 77

 S. Matthews and H. Pajoohesh

By the assumption of the Lemma, a + d ≤ b + c and by Part (a) we have 2ad ≤ 2bc.
1 1 1
So it is enough to prove that abd + acd ≤ abc + bcd. By Part (b), c + b ≤ d + a1 . Thus,
abcd( 1c + 1b ) ≤ abcd( a1 + d1 ) or abd + acd ≤ abc + bcd.


Theorem 2.5. If e : X × X → [0, ∞] is an infinite partial metric, then, τe,∞ is a partial

metrizable topology on X.

Proof. Suppose e : X × X → [0, ∞] is an infinite partial metric. We prove that there is a

partial metric p : X × X → R such that τp = τe,∞ . Define p : X × X → [0, ∞) by p(x, y) =

 e(x,y) if e(x, y) < ∞;
1+e(x,y),
.
2, if e(x, y) = ∞
x
Since f (x) = 1+x is an increasing function, e satisfies the first two and last properties of partial
metrics. For the triangle inequality, note that if ∞ ∈ {e(x, y), e(y, z)}, then p(x, z) ≤ p(x, y) ∨
p(y, z) and p(y, y) ≤ p(x, y)∧p(y, z). Thus, p(x, z)+p(y, y) ≤ p(x, y)∨p(y, z)+p(x, y)∧p(y, z) =
p(x, y)+p(y, z). On the other hand, since e(x, z)+e(y, y) ≤ e(x, y)+e(y, z), ∞ 6∈ {e(x, y), e(y, z)}
implies ∞ 6∈ {e(x, y), e(y, z), e(x, z), e(y, y)}. Thus, {e(x, y), e(y, z), e(x, z), e(y, y)} ⊆ R. There-
fore, by Lemma 2.4, p(x, z) + p(y, y) ≤ p(x, y) + p(y, z).
Now we show that τp = τe,∞ . First we prove τp ⊆ τe,∞ . Note that if a ∈ W ∈ τp then
there exists an  > 0 such that a ∈ N (a) = {y ∈ X : p(a, y) − p(a, a) < } ⊆ W . It is
enough to show that N ∞ ∞
 (a) ⊆ W . Let y ∈ N  (a). If e(a, a) = ∞ then e(a, y) = e(a, a) = ∞.
2 2
Thus p(a, y) = p(a, a) = 2 and so, y ∈ N (a) ⊆ W . For the case that e(a, a) < ∞ if y ∈
e(a,y) e(a,a)
N∞ (a), then e(a, y) < ∞ as e(a, y) ≤ e(a, a) + . Then p(a, y) − p(a, a) = 1+e(a,y) − 1+e(a,a) =
e(a,y)−e(a,a) 
(1+e(a,y))(1+e(a,a)) ≤ e(a, y) − e(y, y) ≤ 2 < . Consequently, y ∈ N (a) ⊆ W and the proof of
τp ⊆ τe,∞ is complete.
Next suppose W ∈ τe,∞ . In order to prove that W ∈ τp , we show that for every a ∈ W
there is an r > 0 such that Nr (a) ⊆ W . Since a ∈ W ∈ τe,∞ there is an  > 0 such that
a ∈ N∞ (a) ⊆ W . If e(a, a) = ∞ then obviously e(a, y) ≤ e(a, a) +  for every y ∈ X. In
this case N (a) = X = W ∈ τp . For the case that e(a, a) < ∞, then e(a, y) < ∞ and so,
e(a,a) e(a,y)
p(a, a) = 1+e(a,a) < 1 and p(a, y) = 1+e(a,y) < 1. Thus, 1 − p(a, a) > 0. Now we define the
radius r as following:
x
f : [0, 1−p(a, a)) → [0, ∞) defined by f (x) = 1−p(a,a)−x is a one-to-one function and its inverse
(1−p(a,a))x
is f −1 (x) = 1+x) . Let s = f −1 ((1 − p(a, a))) and r = s ∧ (1 − p(a, a)). We prove that
Nr (a) ⊆ W . It is enough to show that if y ∈ Nr (a) then y ∈ N∞ (a) or e(a, y)−e(a, a) ≤ . Since

Appl. Gen. Topol. 26, no. 1 (2025) 78

 Infinity valued partial metrics

e p p(a,y) p(a,a) p(a,y)−p(a,a)

p= 1+e , we have e = 1−p . Thus, e(a, y) − e(a, a) = 1−p(a,y) − 1−p(a,a) = (1−p(a,y))(1−p(a,a)) ≤
r r f (r) f (s) (1−p(a,a))
(1−p(a,y))(1−p(a,a)) ≤ (1−p(a,a)−r)(1−p(a,a)) = (1−p(a,a)) ≤ (1−p(a,a)) = (1−p(a,a)) = . Thus, y ∈
Nr (y) implies y ∈ N∞ (a) ⊆ W . Therefore, Nr (a) ⊆ W and so, W ∈ τp . Hence τe,∞ ⊆ τp .
Consequently, τe,∞ = τp . 

3. pure infinite partial metric

Recall that if τ is a topology on a set X and a, b ∈ X, by a ≤τ b we mean a ∈ cl(b). If p is

an (infinite) partial metric on a set X, x ≤ y implies p(x, y) = p(x, x) and τp being T0 implies
(X, ≤τp ) is a poset. We recall that two elements of a poset are incomparable if neither of them
is less than or equal to the other one.

Definition 3.1. An infinite partial metric p on a set X is called pure infinite partial metric if
there are two incomparable elements x, y of (X, ≤τp ) such that p(x, y) = ∞.

In this section we characterize pure infinite partial metrics.

In Theorem 2.5, was shown that every topology induced by a pure extended partial metric,
is partial metrizable. So, the following question is raised:
Question: Is every partial metrizable topology the topology induced by a pure infinite partial
metric?
In the example below it can be seen that the answer is no.

Example 3.2. Let X = {x, y, a, ⊥}. Define p : X × X → R such that

p(x, y) = 1, p(x, x) = p(y, y) = 0, p(x, a) = p(y, a) = p(a, a) = 1, p(x, ⊥) = p(y, ⊥) = p(a, ⊥)
= p(⊥, ⊥) = 2. Now we show that τp is not pure infinite partial metrizable.
The triangular property is the only axiom we want to verify as it is straightforward to see the
other axioms hold. It is easy to see that p(x, y) + p(m, m) ≤ p(x, m) + p(m, y) where m ∈ {a, T }.
Also due to the fact that switching x and y yields the same function it is enough to verify the
following inequalities which we leave it to the reader:
• p(x, ⊥) + p(a, a) ≤ p(x, a) + p(a, ⊥)
• p(x, ⊥) + p(y, y) ≤ p(x, y) + p(y, ⊥)
• p(x, a) + p(⊥, ⊥) ≤ p(x, ⊥) + p(a, ⊥)
• p(x, a) + p(y, y) ≤ p(x, y) + p(y, a)
• p(⊥, a) + p(x, x) ≤ p(⊥, x) + p(x, a)
Note that by definition of p we have ⊥≤τp a ≤τp x, y.

Appl. Gen. Topol. 26, no. 1 (2025) 79

 S. Matthews and H. Pajoohesh

Now we prove that τp is not pure infinite partial metrizable. By the way of contradiction
suppose there is a pure infinite partial metric e : X × X → [0, ∞]. Thus, there are two elements
that are not comparable whose distance is infinity. Since x and y are the only two elements that
are not comparable, we must have e(x, y) = ∞. From e(x, y) + e(a, a) ≤ e(x, a) + e(y, a). Since
e(x, y) = ∞, either e(x, a) = ∞ or e(a, y) = ∞. without lost of generality assume e(x, a) = ∞.
Since a ≤τp x, we have e(a, a) = ∞. Now from e(x, a) + e(⊥, ⊥) ≤ e(x, ⊥) + e(⊥, a) = 2e(⊥, ⊥),
we conclude e(⊥, ⊥) = ∞. Thus, e(a, ⊥) = e(⊥, ⊥) = ∞ = e(a, a) which is a contradiction as
a 6=⊥.

For a partial
 metric space (X, p) define
X, if p(x, x) < ∞ for every x ∈ X;
X∗ = .
X \ {T }, if p(T, T ) = ∞

Lemma 3.3. If A is an open set of X ∗ with respect to subspace topology, then either A is an
open set of X or A = X ∗ .

Proof. Suppose
 A is an open set of X ∗ , then A = Ā ∩ X ∗ , where Ā is an open set of X. Thus,
Ā \ {⊥}, if X ∗ 6= X and ⊥∈ Ā;
A= .
Ā, otherwise
Note that if ⊥∈ Ā, then there is an  > 0 such that N∞ (⊥) ⊆ Ā. Since N∞ (⊥) = X, we
have Ā = X and so, A = X ∗ . 

Define an equivalence relation on X ∗ as follows:

x∆y if and only if p(x, y) < ∞
Since p(x, x) < ∞ for every x ∈ X ∗ , the relation ∆ is reflexive and obviously it is symmetric.
Next we show that ∆ is transitive. Suppose x∆y and y∆z. Then, p(x, y) < ∞ and p(y, z) < ∞.
Thus, p(x, z) ≤ p(x, z) + p(y, y) ≤ p(x, y) + p(y, z) < ∞, implies x∆z.
Now the equivalence classes of ∆ form a partition for X ∗ . We denote the equivalence class of
x ∈ X ∗ by [x].

Lemma 3.4. For every x ∈ X ∗ , the set [x] is a clopen subset of X ∗ .

Proof. If there is only one equivalence class then [x] = X ∗ and obviously [x] is a clopen subset
of X ∗ . So assume that there are more than one equivalence class. First we prove [x] is closed.
Suppose a 6∈ [x]. Then, p(a, x) = ∞. We show that if y ∈ N1∞ (a), then p(y, x) = ∞. Note

Appl. Gen. Topol. 26, no. 1 (2025) 80

 Infinity valued partial metrics

that ∞ = p(a, x) + p(y, y) ≤ p(a, y) + p(y, x). Thus, p(a, y) + p(y, x) = ∞. Since p(a, y) ≤
p(a, a) + 1 < ∞. we have, p(x, y) = ∞ and so, N1∞ (a) ⊆ X \ [x]. Consequently, [x] is closed.
Now we prove that [x] is open. Assume a ∈ [x]. If y ∈ N1∞ (a), then p(a, y) ≤ p(a, a) + 1 < ∞.
We show that y ∈ [x] or p(y, x) < ∞. By triangle inequality, p(x, y)+p(a, a) ≤ p(x, a)+p(a, y) ≤
p(x, a) + p(a, a) + 1 < ∞. Thus, p(x, y) < ∞ and so, y ∈ [x]. 

Corollary 3.5. If p is an infinite partial metric on X, then either p(X ∗ , X ∗ ) ⊆ [0, ∞) or X ∗

is disconnected with respect to subspace topology.

Proof. Note that either X ∗ has one equivalence class or more than one. If it has one equivalence
class, then p(X ∗ , X ∗ ) ⊆ [0, ∞). If there are more than one equivalence class, then there are
more than one clopen sets. 

Definition
 3.6. For a partial metrizable topology τ on a set X, define
X, if (X, ≤τ ) does not have smallest element;
X̂ = .
X \ {⊥}, if ⊥ is the smallest element of (X, ≤ )
τ

Theorem 3.7. Let τ be a partial metrizable topology on a set X. Then τ is pure infinite partial
metrizable if and only if X̂ is a disconnected topology with respect to the subspace topology.

Proof. Suppose τ = τe,∞ for the pure infinite partial metric e. Note that ≤τ =≤τe,∞ , as τ = τe,∞ .
So, X̂ = X ∗ . Thus, there are incomparable elements x, y ∈ X such that e(x, y) = ∞. Since x
and y are not comparable, x, y ∈ X ∗ . Thus, e(X ∗ , X ∗ ) ⊆ [0, ∞]. On the other hand by Corollary
3.5 either e(X ∗ , X ∗ ) ⊆ [0, ∞) or X ∗ is disconnected. Consequently, X ∗ = X̂ is disconnected.
For the converse suppose X̂ is disconnected and τ = τp for the partial metric p. Let A be a
non-empty clopen set of X̂ with A 6= X̂. By Lemma 3.3 A is an open set of X with respect to
τp . Define 
p(x, y), if x, y ∈ A or x, y ∈ X̂ \ A;
e : X × X → [0, ∞] by e(x, y) = .
∞, otherwise
One can easily see that e(x, x) = ∞ if and only if x ∈ X \ X̂. We will leave it to the reader
to verify that e is an infinite partial metric. We show that e is pure infinite partial metric. Note
that for every m ∈ A and every n ∈ X̂ \ A we have e(m, n) = ∞ and m and n are incomparable
as they are in two different open sets. Thus, e is pure infinite partial metric.
Next we show that τp = τe,∞ . Let H ∈ τp and x ∈ H. Thus, there is an r > 0 such that
Nr (x) ⊆ H. Note that there are three cases: x ∈ X \ X̂, x ∈ A, or x ∈ X̂ \ A. In the case
that x ∈ X \ X̂, we have p(x, x) = p(x, y) for every y ∈ X. Thus, Nr (x) = X ⊆ H and

Appl. Gen. Topol. 26, no. 1 (2025) 81

 S. Matthews and H. Pajoohesh

so, H = X and H ∈ τe,∞ . If x ∈ A, then x ∈ H ∩ A ∈ τp . Thus, there is an  ∈ [0, ∞)

such that x ∈ {y : p(x, y) − p(x, x) ≤ } ⊆ H ∩ A. It is enough to show that x ∈ {y :
e(x, y) ≤ e(x, x) + } ⊆ H ∩ A ⊆ H. Note that e(x, x) 6= ∞ implies that e(x, y) 6= ∞ for every
y ∈ {y : e(x, y) ≤ e(x, x) + }. Thus, in this case e(x, x) = p(x, x) and e(x, y) = p(x, y). Thus,
y ∈ {y : e(x, y) ≤ e(x, x) + } implies y ∈ {y : p(x, y) − p(x, x) ≤ } ⊆ H ∩ A. If x ∈ X̂ \ A, a
similar argument works. Thus, τp ⊆ τe,∞ .
For the other inclusion, suppose H ∈ τe,∞ and x ∈ H. Thus, there is an r ∈ [0, ∞) such that
x ∈ {y : e(x, y) ≤ e(x, x) + r} ⊆ H. If x ∈ X \ X̂, then Nr∞ (x) = X ⊆ H and so, H ∈ τp . So,
assume x ∈ X̂. Then, either x ∈ A or x ∈ X̂ \A. Without lost of generality we can assume x ∈ A.
Since A is open in τp , there is an s ∈ [0, ∞) such that x ∈ {y : p(x, y) − p(x, x) ≤ s} ⊆ A. Let
 = r∧s. We show that {y : p(x, y)−p(x, x) ≤ } ⊆ H. Note that if a ∈ {y : p(x, y)−p(x, x) ≤ }
then since x and a both will be in A, e(x, a) = p(x, y) and e(x, x) = p(x, x). Thus, from
p(x, a)−p(x, x) <  we have e(x, a) ≤ e(x, x)+ ≤ e(x, x)+r. Thus, a ∈ {y : e(x, y) ≤ e(y, y)+r}
and so a ∈ H. Consequently, x ∈ {y : p(x, y) ≤ p(y, y) + } ⊆ H. Therefore, H ∈ τp . So,
τp = τe,∞ . 

Recall that a preordered set (X, ≤) is called connected if for any a, b ∈ X, there exists a finite
sequence (a = x1 , · · · , xn = b) such that xi and xi+1 are comparable for every i = 1, · · · , n − 1.
If (X, ≤) is a preordered set, a ∈ X, and A ⊆ X, then ↑ a = {x ∈ X : a ≤ x}, ↓ a = {x ∈ X :
S S
x ≤ a}, ↑ A = a∈A ↑ a, and ↓ A = a∈A ↓ a.
In [13] it is shown that a preordered set (X, ≤) is connected if and only if X =↑↓ {a}∪ ↑↓↑↓
{a} ∪ · · · for every a ∈ X.

Theorem 3.8. Let τ be a partial metrizable topology on a set X. If τ is pure infinite partial
metric, then X̂ is a disconnected poset with respect to ≤τ .

Proof. Suppose τ = τe,∞ for the pure infinite partial metric e. Thus, there are incomparable
elements x, y ∈ X such that e(x, y) = ∞. We prove that x 6∈↑↓ {y}∪ ↑↓↑↓ {y} ∪ · · · . For
contrary assume that x ∈↑↓ {y}∪ ↑↓↑↓ {y} ∪ · · · . Thus there are y1 , · · · , yn such that y =
y0 ≥ y1 ≤ y2 ≥ y3 ≤ · · · x = yn . Thus, for for every i ∈ {0, 1, · · · , n}, either p(yi , yi+1 ) = yi
or p(yi , yi+1 ) = yi+1 . Since p(x, x) < ∞ for every x ∈ X̂, we conclude p(yi , yi+1 ) < ∞ for
for every i ∈ {0, 1, · · · , n}. If we apply the triangularity of partial metrics n − 1 time we have
p(y, x) ≤ ni=0 p(yi , yi+1 )− ni=1 p(yi , yi ) < ∞ which is a contradiction. Therefore, ↑↓ {y}∪ ↑↓↑↓
P P

{y} ∪ · · · 6= X and so, X̂ is a disconnected poset with respect to ≤τ . 

By Theorem 3.7 the corollary below is immediate.

Appl. Gen. Topol. 26, no. 1 (2025) 82

 Infinity valued partial metrics

Corollary 3.9. Let τ be a partial metrizable topology on a set X. If X̂ is a disconnected topology

with respect to the subspace topology, then X̂ is a disconnected poset with respect to ≤τ .

In the next example one can see that the converse of Theorem 3.8 is not true.
1
 Let X = { n : n ∈ N \ {1}} ∪ {0}. Define p : X × X → [0, 1] as follows:
Example 3.10.
 1 , if {x, y} = {0, n1 },
n−1
p(x, y) = .
x ∨ y, otherwise
1 1
One can verify that x ≤τp y if and only if x = n and y = m with n ≤ m. Thus, the subset
{ n1 : n ∈ N} of X is a chain and 0 is incomparable with any element of it. Therefore, X̂ is a
disconnected poset with respect to ≤τp which does not have smallest element. So, X̂ = X.
We prove that X̂ = X with respect to τp is a connected topology. Assume H is a clopen set
1 1
containing 2. Since H is closed it contains all elements greater than or equal 2. Since H is
open, it contains all of the non-zero real numbers in X that are less than or equal to 12 . Thus,
H ⊇ X̂ \ {0}. It is straightforward to verify that 0 ∈ cl(H) = H. Therefore, H = X̂ = X and
consequently X̂ = X is connected.

Definition 3.11. Let p : X × X → [0, ∞] be a infinite partial metric. We call the sequence {xn }
a Cauchy sequence if limm,n→∞ p(xm , xn ) exists. We say a Cauchy sequence {xn } converges to
an element a if limm,n→∞ p(xm , xn ) = p(a, a). We call X complete, if every Cauchy sequence
converges.

Here we show that the fixed point theorem of Matthews of partial metric spaces, work for
infinite partial metrics.

Theorem 3.12. Let p : X × X → [0, ∞] be a complete infinite partial metric and f : X → X be

such that p(f (x), f (y)) ≤ cp(x, y) where 0 < c < 1. Then there is an a ∈ X such that f (a) = a.
Moreover, if X has more than one element, p(a, a) = 0.

Proof. If X = {a} then every function has a fixed point. So, suppose X has more than one
element. By Lemma 2.2, there is an element b ∈ X such that p(b, b) < ∞. Then if one follows
Matthews’s proof in [9], {f n (b)} is a Cauchy sequence approaching a unique number u with
f (u) = u with p(u, u) = 0. 

4. Conclusion

This article has explained how partial metrics arose from the need to model partial data in
computation. This clicked with Kopperman’s interest in non-Hausdorff topology. Generalising

Appl. Gen. Topol. 26, no. 1 (2025) 83

 S. Matthews and H. Pajoohesh

partial metric spaces by introducing non zero self-distance is a counterpart to other generalisa-
tions such as pseudo metric and quasi metric spaces. All such generalisations raise fascinating
questions in non-Hausdorff topology, and further still into the ontologies of mathematics waiting
to be discovered. As a talented topologist Kopperman pursued these very questions.

Acknowledgements. The authors would like to thank Michael Bukatin for meeting with us
to discuss the paper.

References
[1] S. Assaf and K.Pal, Partial metric spaces with negative distances and fixed point theorems, Topology Proc.
49 (2017), 19–40.
[2] G. Beer, The structure of infinite real-valued metric spaces, Set-Valued Var. Anal. 21, no.4 (2013), 591–602.
[3] M. Bukatin, R. Kopperman, S. Matthews and H.Pajoohesh, Partial metric spaces, American Mathematical
Monthly 116 (2009), 708–718.
[4] M. Darnel and W. C. Holland, Minimal non-metabelian varieties of `-groups that contain no nonabelian
o-groups, Communications in Algebra 42 (2014), 5100–5133.
[5] M. Darnel, W. C.Holland, and H. Pajoohesh, The relationship of partial metric varieties and commuting
powers varieties, Order 30, no.2 (2013), 403–414.
[6] C. Holland, R. Kopperman, and H. Pajoohesh, Intrinsic generalized metrics, Algebra Universalis 67, no. 1
(2012), 1–18.
[7] R. Kopperman, S. Matthews, and H. Pajoohesh, Universal partial metrizability, Applied General Topology
5 (2004), 115–127.
[8] R. Kopperman, All topologies come from generalised metrics, American Mathematical Monthly 95, no. 2
(1988).
[9] S. Matthews, Partial metric topology, Proc. 8th summer conference on topology and its applications, ed S.
Andima et al., New York Academy of Sciences Annals, 728 (1994), 183–197.
[10] S. Oltra and O. Valero, Banach’s fixed point theorem for partial metric spaces, Rend. Istit. Mat. Univ. Trieste
36, no. 1-2 (2004), 17–26.
[11] S. J. O’Neill, Partial metrics, valuations and domain theory, Proc. 11th Summer Conference on General
Topology and Applications. Ann. New York Acad. Sci. 806 (1996), 304–315.
[12] H.Pajoohesh, The relationship of partial metric varieties and commuting powers varieties II, Algebra Uni-
versalis, 73, no. 3-4 (2015), 291–295.
[13] H.Pajoohesh, T0 functional Alexandroff topologies are partial metrizable, Applied General Topology 25, no.
2 (2024), 305–319.
[14] D. Scott, Outline of a mathematical theory of computation, Oxford University Computing Laboratory,
PRG02, 1970.
[15] J. E. Stoy, The Scott-Strachey Approach to Programming Language Theory, MIT Press 1977.

Appl. Gen. Topol. 26, no. 1 (2025) 84

 Infinity valued partial metrics

[16] O. Valero, On Banach fixed point theorems for partial metric spaces, Applied General Topology 6, no. 2
(2005), 229–240.

Appl. Gen. Topol. 26, no. 1 (2025) 85