Hyperconvexity in Partial Metric Spaces

• Hyperconvexity in partial metric spaces generalizes classical hyperconvexity by allowing nonzero self-distances to capture partial information.
• AP-hyperconvexity and nodal hyperconvexity are distinct notions that modify the intersection conditions of closed balls in the partial metric context.
• Metric-induced variants like p^m, d_m, and D reveal varied fixed-point and convexity properties, exposing structural challenges in generalization.

Hyperconvexity in partial metric spaces is a topic at the intersection of metric geometry, category theory, and the generalization of fixed-point and extension principles. While classical hyperconvexity, as introduced by Aronszajn and Panitchpakdi, enjoys a robust set of structural properties—including being a metric absolute retract, allowing Hahn–Banach–type extensions, guaranteeing fixed points for nonexpansive self-maps, and total convexity—transferring this notion to the setting of partial metrics introduces significant divergence, structural loss, and a proliferation of inequivalent candidate definitions (Bugajewski et al., 5 Jan 2026).

1. Foundations: Partial Metrics and Classical Hyperconvexity

A partial metric on a set $U$ is a map $p:U \times U \to [0,\infty)$ satisfying:

• (P1) $x=y$ iff $p(x,x) = p(y,y) = p(x,y)$,
• (P2) $p(x,x) \le p(x,y)$,
• (P3) $p(x,y) = p(y,x)$,
• (P4) $p(x,y) \le p(x,z) + p(z,y) - p(z,z)$ for all $x,y,z \in U$.

Partial metrics generalize ordinary metrics by admitting nonzero self-distances ($p(x,x) \ge 0$), which capture partial information and measure of definedness. Setting $p(x,x)=0$ recovers classical metrics.

Three associated metrics, each with distinct roles in the theory, can be derived from a partial metric $p$:

• $p^m(x,y) = 2p(x,y) - p(x,x) - p(y,y)$,
• $$d_m(x,y) = \max\{p(x,y) - p(x,x),\, p(x,y) - p(y,y)\}$$,
• $D(x,y) = p(x,y)$ if $x \ne y$, and 0 if $x = y$.

In the classical metric context, a space $(H,d)$ is hyperconvex if every family of closed balls $\{_d(x_i,r_i)\}_{i \in I}$ with $d(x_i,x_j) \le r_i + r_j$ for all $i,j$ has nonempty intersection. This property underpins extension, injectivity, and fixed-point results central to modern nonlinear analysis.

2. Direct Partial Metric Analogues: AP- and Nodal Hyperconvexity

Translating hyperconvexity to partial metric spaces is nontrivial. The naïve extension of the Aronszajn–Panitchpakdi definition, requiring intersections for families of $p$-balls, forces $p$ to be a metric (all self-distances zero), yielding no genuinely new spaces [(Bugajewski et al., 5 Jan 2026), Remark 3.2]. Thus, alternative notions emerge:

• AP-hyperconvexity: For every family $\{{}_p(x_i, r_i)\}$ with $p(x_i, x_j) \le r_i + r_j$, there exists $z$ such that $p(z, x_i) \le p(x_i, x_i) + r_i$ for all $i$.
• Nodal hyperconvexity: For such a family, there exists $z$ such that $p(z, x_i) \le p(z, z) + r_i$ for all $i$.

In the metric case, these coincide (as all self-distances vanish), but in the partial metric setting they are logically distinct, as demonstrated by counterexamples [(Bugajewski et al., 5 Jan 2026), Examples 2.3–2.4]:

• Nodal hyperconvexity does not imply AP-hyperconvexity and vice versa.
• Every finite set can be endowed with both AP- and nodally hyperconvex partial metrics, while finite hyperconvex metric spaces reduce to singletons.

3. Metric-Induced Hyperconvexities and Interrelations

An alternative approach defines hyperconvexity for partial metric spaces via the associated metrics:

• $p^m$-hyperconvexity: $(U, p^m)$ is hyperconvex.
• $d_m$-hyperconvexity: $(U, d_m)$ is hyperconvex.
• $D$-hyperconvexity: $(U, D)$ is hyperconvex.

The following implications hold:

• $p^m$-hyperconvexity ⇒ nodal hyperconvexity.
• $d_m$-hyperconvexity ⇒ both AP-hyperconvexity and nodal hyperconvexity.

However, these implications are not reversible, with examples demonstrating that the induced metric hyperconvexities may not coincide with the direct analogues. For instance:

• On $\mathbb{R}$ with $p(x,y) = 1 + |x-y|$, both $p^m$ and $d_m$ yield hyperconvex metrics, but $D$ does not [(Bugajewski et al., 5 Jan 2026), Example 3.4].
• In certain subsets of $\mathbb{R}^3$ with specific partial metrics, $p^m$-hyperconvexity holds but $d_m$-hyperconvexity does not [(Bugajewski et al., 5 Jan 2026), Example 3.5].

4. Structural Obstacles: Failure of Classical Properties

Hyperconvexity in classical metric spaces is characterized by several hallmark properties:

• Absolute retract (injectivity) and metric Hahn–Banach-type extension,
• Baillon–Sine–Soardi fixed-point property for nonexpansive self-maps on bounded sets,
• Total convexity and completeness.

In the partial-metric setting, direct analogues fail to exhibit these. For instance:

• Insisting on the classical intersection criterion trivializes the structure to genuine metrics [(Bugajewski et al., 5 Jan 2026), Remark 2.1].
• Neither AP- nor nodal hyperconvexity guarantees fixed points for nonexpansive maps (e.g., flip map on a two-point partial metric space has no fixed point although both AP- and nodal hyperconvexity hold) [(Bugajewski et al., 5 Jan 2026), Example 5.6].
• AP- and nodally hyperconvex spaces need not be complete in the partial-metric sense, nor totally convex [(Bugajewski et al., 5 Jan 2026), Example 3.3].

This indicates that none of the partial-metric hyperconvexity notions fully inherit the rich structure of the classical metric case.

5. Concrete Examples and Notions Comparison

The distinctions between the various forms of hyperconvexity can be summarized as follows:

Property	Metric hyperconvex	AP-hyperconvex	Nodal hyperconvex	$p^m$-hyperconvex	$d_m$-hyperconvex	$D$-hyperconvex
Total convexity	Yes	No	No	Yes	Yes	Yes
Metric completeness	Yes	No	No	Yes	Yes	Yes
Fixed-point property	Yes (bounded)	No	No	Yes	Yes	Yes
Injective object in p-Met	Yes	No	No	–	–	–
Non-trivial new examples	–	Yes	Yes	No	No	No

Key examples [(Bugajewski et al., 5 Jan 2026), Section 5.1]:

• All finite sets support AP- and nodally hyperconvex partial metrics, in contrast to metrics.
• AP and nodal hyperconvexity are not equivalent in finite examples.
• The three metric-induced variants ($p^m$, $d_m$, $D$) are in general distinct and exhibit nontrivial failure of implications.

6. Open Problems and Research Directions

Several central questions remain unresolved [(Bugajewski et al., 5 Jan 2026), Section 5.3]:

1. Given that every real Banach space with its canonical partial metric is AP-hyperconvex, does this extend to nodal hyperconvexity (including over $\mathbb{C}$)?
2. For which spaces does $p^m$-hyperconvexity guarantee AP-hyperconvexity?
3. Under what additional restrictions does $d_m$-hyperconvexity imply $p^m$-hyperconvexity?
4. Is it possible to formulate a single definition of partial-metric hyperconvexity that inherits all desirable properties from the metric case, or does the inherent structure preclude such a unification?

7. Synthesis and Outlook

Hyperconvexity in partial metric spaces lacks a unique canonical extension of the metric theory. Attempts to generalize the Aronszajn–Panitchpakdi notion invariably lose at least one of the classical properties: completion, total convexity, fixed-point property, or injectivity. Each of the competing definitions—AP-hyperconvexity, nodal hyperconvexity, and the metric-induced variants—captures different aspects but fails to replicate the full suite of structural features present in the classical case. The existence of a genuinely satisfactory analogue or a proof of its impossibility remains an open and compelling challenge in the field (Bugajewski et al., 5 Jan 2026).