Partial Metric Spaces Overview

Updated 2 May 2026

Partial metric spaces are defined as sets with a distance function that allows nonzero (or negative) self-distances and a modified triangle inequality.
They are applied in domain theory, computer science, and similarity scoring, enabling models that capture incomplete or asymmetric information.
Extending standard metrics, these spaces support unique convergence, compactness, and fixed point properties that enhance both theoretical and practical analyses.

A partial metric space is a set equipped with a distance function that allows nonzero self-distances but preserves a weakened form of the triangle inequality and other essential features of metric geometry. This framework, originally introduced by Matthews (1994), is motivated by applications in computer science, domain theory, and areas where incomplete or partial information is naturally modeled by nonzero self-distance. The structure has developed to encompass a spectrum of generalizations, links to other generalized metric spaces, fixed point principles, topological properties, and category theory.

1. Formal Definitions and Foundational Structures

Let $X$ be a nonempty set. A partial metric is a function $p:X\times X\to\mathbb{R}$ (often $[0,\infty)$ , but negative values are also permitted in some frameworks) satisfying for all $x,y,z\in X$ the axioms:

$p(x,x) \le p(x,y)$ .
$p(x,y) = p(y,x)$ .
$x=y \iff p(x,x) = p(x,y) = p(y,y)$ .
$p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ .

The pair $(X,p)$ is a partial metric space (Bayram et al., 2023, Abodayeh et al., 2015, Imamura, 2019, Bugajewski et al., 2020, Assaf et al., 2014). This framework generalizes ordinary metric spaces, recovered by enforcing $p(x,x)=0$ for all $p:X\times X\to\mathbb{R}$ 0 and the standard triangle inequality.

The induced metric $p:X\times X\to\mathbb{R}$ 1 is a genuine metric. Convergence in $p:X\times X\to\mathbb{R}$ 2 is defined so that $p:X\times X\to\mathbb{R}$ 3 if $p:X\times X\to\mathbb{R}$ 4 and, for proper convergence, $p:X\times X\to\mathbb{R}$ 5 (Bugajewski et al., 2020). The natural topology is $p:X\times X\to\mathbb{R}$ 6 but not Hausdorff in general.

Partial metrics admit negative self-distances (O'Neill partial metrics) (Assaf et al., 2015, Assaf et al., 2014), enabling applications to similarity scores and constructions with negative weights.

Further generalizations include:

Partial metric type spaces (K-partial metric): $p:X\times X\to\mathbb{R}$ 7 for fixed $p:X\times X\to\mathbb{R}$ 8 (Gaba, 2018).
Partial $p:X\times X\to\mathbb{R}$ 9-metric spaces: generalized to functions $[0,\infty)$ 0 with analogous axioms (Assaf et al., 2015).

2. Topology, Compactness, and Completion

Open balls in $[0,\infty)$ 1 are defined by $[0,\infty)$ 2 (Bugajewski et al., 2020). Sequential compactness and compactness coincide: every compact cover has a finite subcover if and only if every sequence has a convergent subsequence (Bugajewski et al., 2020). A Hausdorff compact partial metric space is metrizable via a compatible metric.

Completeness is determined via Cauchy sequences:

A sequence $[0,\infty)$ 3 is Cauchy if $[0,\infty)$ 4 exists and is finite.
Completeness: every Cauchy sequence properly converges to some $[0,\infty)$ 5, equivalently $[0,\infty)$ 6 is complete (Abodayeh et al., 2015, Bugajewski et al., 2020).

Cauchy completions require care: symmetric denseness (where both $[0,\infty)$ 7 is close to $[0,\infty)$ 8 and $[0,\infty)$ 9 to $x,y,z\in X$ 0) gives a unique (up to isometry) completion with a universal extension property (Imamura, 2019). Asymmetric completions exist as well, but may fail this universal property and lead to non-isometric completions.

3. Fixed Point Theorems

Partial metric spaces support a robust theory of fixed points, unifying and extending principles from metric fixed point theory.

Banach Contraction Principle: For a complete partial metric space $x,y,z\in X$ 1, a self-map $x,y,z\in X$ 2 with $x,y,z\in X$ 3 ( $x,y,z\in X$ 4) admits a unique fixed point (often with $x,y,z\in X$ 5) (Ahmadullah et al., 2016, Bugajewski et al., 2020, Abodayeh et al., 2015, Assaf et al., 2015, Gaba, 2018).
Generalized Contractive Mappings: "Functional contractive" and admissibility-based conditions: $x,y,z\in X$ 6, with $x,y,z\in X$ 7 normal and $x,y,z\in X$ 8 an auxiliary function (e.g., mean self-distance). These unify earlier linear contraction schemes and enable non-linear hypotheses (Turinici, 2012, Gaba, 2018).
Ordered Partial Metric Spaces: Contractive maps that preserve a partial order and act on comparable elements only admit fixed points under mild monotonicity and comparability conditions (Aydi, 2011).
Multi-Map and Relation-Theoretic Schemes: Results extend to families of operators (e.g., common fixed points for commuting contractions) and to settings where a binary relation replaces the metric's symmetry or order (Ahmadullah et al., 2016).
Caristi–Kirk Type Theorems: Fixed points arise for self-maps that decrease a lower semicontinuous potential function, extending to $x,y,z\in X$ 9-metric spaces and resolving previous gaps in proofs (Abodayeh et al., 2015).
Negative or Nonzero Self-Distances: Fixed points with $p(x,x) \le p(x,y)$ 0 are admitted, e.g., for non-expansive or orbitally contractive mappings in strong partial metric spaces (Assaf et al., 2014, Assaf et al., 2015).

For all such results, the restriction to the bottom set $p(x,x) \le p(x,y)$ 1 may be necessary; on $p(x,x) \le p(x,y)$ 2, $p(x,x) \le p(x,y)$ 3 is a genuine metric. Many theorems reduce to classical fixed point theory on $p(x,x) \le p(x,y)$ 4 (Bugajewski et al., 2020).

4. Relationships to Other Metric-Like Spaces

Partial metric spaces interface with a range of generalizations:

Partial $p(x,x) \le p(x,y)$ 5-metrics, partial $p(x,x) \le p(x,y)$ 6-metrics: Relaxed triangle inequalities incorporating scaling or additional points (Karahan et al., 2018, Gaba, 2018).
Rectangular and $p(x,x) \le p(x,y)$ 7-generalized metrics: Higher-arity triangle inequalities via auxiliary points, supporting further generalizations (Gaba, 2018, Karahan et al., 2018).
$p(x,x) \le p(x,y)$ 8-metric spaces: Allow comparison via minimal self-distances (entries in the triangle axiom) (Abodayeh et al., 2015); every partial metric is an $p(x,x) \le p(x,y)$ 9-metric, but not conversely.
Probabilistic partial metric spaces: Spaces enriched in quantaloids of diagonals between distance distributions generalize probabilistic metrics via categorical frameworks (He et al., 2018).

A summary table of some relations is given below.

Structure	Key Axioms / Relationship	Canonical Example / Metric Reduction
Partial metric	$p(x,y) = p(y,x)$ 0, relaxed triangle	$p(x,y) = p(y,x)$ 1 ( $p(x,y) = p(y,x)$ 2 metric, $p(x,y) = p(y,x)$ 3)
$p(x,y) = p(y,x)$ 4-metric	min-self-distance in triangle inequality	$p(x,y) = p(y,x)$ 5 when $p(x,y) = p(y,x)$ 6 is a partial metric
Partial $p(x,y) = p(y,x)$ 7-metric	$p(x,y) = p(y,x)$ 8, $p(x,y) = p(y,x)$ 9 auxiliary points in triangle	Recovers partial metric for $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 0, $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 1
Rectangular metrics	$x=y \iff p(x,x) = p(x,y) = p(y,y)$ 2, suitable $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 3 parameterization	Classical rectangular metrics for $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 4
Probabilistic partial	Diagonals in quantale of distance distrib.	Enriched category frameworks, Lawvere quantale

5. Convergence and Summability in Partial Metric Spaces

Statistical convergence and Cesàro summability have analogues in partial metric spaces, incorporating the possibility $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 5 (Bayram et al., 2023).

Statistical Convergence of Order $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 6: For $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 7, $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 8-convergent to $x=y \iff p(x,x) = p(x,y) = p(y,y)$ 9 if

$p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 0

Strong $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 1-Cesàro Summability of Order $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 2:

$p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 3

implies statistical convergence of higher order.

$p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 4-statistical convergence: For sequences of variable window sizes adapting the density notions.

These extend classical analysis methods to realms where nonzero self-distance encodes partial or incomplete information.

6. Hyperconvexity and Geometric Properties

Multiple hyperconvexity notions arise in partial metric spaces (Bugajewski et al., 5 Jan 2026).

AP-hyperconvexity: For any family of closed balls with $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 5, there exists $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 6 such that $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 7.
Nodal hyperconvexity: $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 8 for all $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ 9.
Associated metric hyperconvexity: Hyperconvexity with respect to induced metrics, e.g., $(X,p)$ 0, $(X,p)$ 1, or $(X,p)$ 2.

Unlike the classical case, these properties may not coincide, and standard links between hyperconvexity, completeness, and fixed-point properties may fail. Counterexamples show that AP- and nodal hyperconvexity generally diverge, and even bounded AP-hyperconvex spaces need not have the fixed-point property.

7. Significance and Applications

Partial metric spaces and their extensions offer a uniform framework for a variety of fixed point phenomena, topologies, and analytic techniques relevant in logic programming, computational semantics, domain theory, bioinformatics (similarity scoring), and stochastic metric structures. Their intrinsic flexibility, via nonzero or negative self-distances, and the suite of available fixed-point and convergence results enables applications and unifications not achievable in standard metric frameworks. Ongoing research seeks to refine geometric properties (especially hyperconvexity), further bridge connections with category theory, and extend fixed point results to multi-valued, nonself, or probabilistic settings (He et al., 2018, Bugajewski et al., 5 Jan 2026).