Metric Map (Nonexpansive Map)¶

A function between metric spaces that never increases pairwise distance, equivalently a Lipschitz map with constant at most one.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2269
Origin domain: metric geometry
Subdomain: nonlinear analysis
Aliases: Nonexpansive map, Nonexpanding map, Short map, 1-Lipschitz map

Core Idea¶

A metric map, usually called a nonexpansive map, is a function \(f:(X,d_X)\to(Y,d_Y)\) satisfying

\[ d_Y(f(x),f(x'))\le d_X(x,x') \qquad\text{for every }x,x'\in X. \]

It is exactly a Lipschitz map whose Lipschitz constant is at most one. The invariant is global pairwise distance nonincrease: the map may preserve a distance or shrink it, but it may not stretch any pair. This definition is standard in metric fixed-point theory and metric geometry.^[1]^[2]

The name “metric map” is less stable internationally than “nonexpansive map” or “short map,” so the inequality, not the surface label, locks the identity. The structure matters because identity maps and composites remain nonexpansive, making metric spaces and these maps a category, while iteration inherits an a priori bound on the separation of trajectories.

Structural Signature¶

Source metric space: a set \(X\) with a declared metric \(d_X\).
Target metric space: a set \(Y\) with a declared metric \(d_Y\).
Underlying function: a single-valued assignment \(f:X\to Y\).
Global pair test: every pair \(x,x'\), not merely nearby pairs, is checked.
Unit Lipschitz bound: \(d_Y(fx,fx')\le d_X(x,x')\).
Composition closure: if \(f\) and \(g\) are nonexpansive, \(g\circ f\) is nonexpansive.
Iteration bound: a nonexpansive self-map does not increase the distance between synchronized iterates.
Equality boundary: preserving every distance makes an isometric embedding; having a nonexpansive inverse forces equality.

Recognition test. Identify both metrics and calculate the supremal distance ratio over distinct inputs. The map qualifies exactly when every ratio is at most one. A local derivative norm bounded by one can certify the condition in suitable normed smooth settings, but continuity alone cannot.

What It Is Not¶

Nonexpansive does not mean strictly contractive. A contraction has some common constant \(q<1\), and on a complete space Banach's theorem then supplies a unique attracting fixed point. A nonexpansive map may have Lipschitz constant exactly one, multiple fixed points, no fixed point, or nonconvergent iterates. Rotation of a circle is the simplest warning.

It is not an isometry: an isometry preserves all distances, while a nonexpansive map may collapse distinct points or whole regions. It is not merely continuous, uniformly continuous, or topology-preserving. The unit distance bound implies uniform continuity, but many uniformly continuous maps expand some distances.

It is also not the opposite convention sometimes called a metric map in niche texts, where a real-valued function satisfies a lower distance inequality. This entry follows the frozen target and the dominant nonexpansive/1-Lipschitz identity.

Scope of Application¶

In fixed-point theory, nonexpansive self-maps sit at the limiting boundary of contraction methods. Results require additional geometry—convexity, compactness, normal structure, uniform convexity, or hyperconvexity—because the unit bound alone supplies no fixed point.^[2] Iterative schemes such as Krasnosel'skii–Mann averaging are designed precisely for this weaker regime.

In metric geometry, short maps are natural morphisms because they respect the observable meaning of distance without demanding invertibility. Distance-to-a-set functions, nearest-point projections in Hilbert space, retractions onto convex sets under appropriate hypotheses, and quotient-like collapses supply recurring cases.

In optimization and monotone operator theory, resolvents and proximal mappings are often firmly nonexpansive, a stronger inequality that implies nonexpansiveness. In probability, couplings and Markov operators may contract selected metrics, but a stochastic operator belongs here only after a specific metric and the global unit bound are established.

Clarity¶

The abstraction separates three questions that ordinary “stability” language blurs. Does the map avoid magnifying input error? Does repeated application converge? Does a fixed point exist? Nonexpansiveness answers only the first. Convergence and existence need extra structure.

The decisive diagnostic is pairwise. For \(f(x)=|x|\) on the real line, the reverse triangle inequality gives \(\lvert |x|-|y|\rvert\le |x-y|\), so the map is nonexpansive even though it is not linear or injective. For \(f(x)=2x\), the ratio is two, so continuity and smoothness do not rescue it.

Manages Complexity¶

One inequality replaces an unbounded collection of case-specific perturbation estimates. From it, every pair of trajectories satisfies

\[ d(f^n(x),f^n(y))\le d(x,y), \]

by induction. Approximation error is therefore never amplified merely by applying the map repeatedly. Composition also lets a long pipeline be verified componentwise: if every stage is nonexpansive, the whole pipeline is.

The compression deliberately discards how much individual pairs shrink, whether directions are preserved, and whether images are convex or closed. Those variables must be restored for rates, inverse reconstruction, or fixed-point guarantees.

Abstract Reasoning¶

If \(f:X\to Y\) and \(g:Y\to Z\) are nonexpansive, then

\[ d_Z(gf(x),gf(x'))\le d_Y(f(x),f(x'))\le d_X(x,x'), \]

so composition is closed. The identity map has equality, establishing categorical identities.

If \(f\) is bijective and both \(f\) and \(f^{-1}\) are nonexpansive, applying the two inequalities yields

\[ d_Y(fx,fy)\le d_X(x,y)\le d_Y(fx,fy), \]

so \(f\) is an isometry. This is a useful boundary inference: invertible “no expansion in either direction” is exactly distance preservation.

If a sequence \(x_n\to x\), then \(d(fx_n,fx)\le d(x_n,x)\to0\), proving continuity directly. The converse fails, locating nonexpansiveness strictly below general continuity and above stronger contraction classes.

Knowledge Transfer¶

The same inequality transfers literally among metric geometry, nonlinear functional analysis, optimization, and metric fixed-point theory. What changes is the chosen metric and the source of the map; the pairwise test and composition proof do not change.

The parent Function (Mapping) carries source, target, evaluation, and composition across domains. Metric supplies the pairwise quantitative structure. The named candidate remains domain-specific because the unit Lipschitz comparison literally presupposes two metric spaces. “This policy is nonexpansive” outside such a space is metaphor until a metric is defined.

Examples¶

Orthogonal projection. The nearest-point projection \(P_C\) onto a nonempty closed convex subset of a Hilbert space is nonexpansive; in fact, it is firmly nonexpansive. The domain geometry is essential.^[3]
Absolute value. \(f:\mathbb R\to\mathbb R\), \(f(x)=|x|\), is nonexpansive by the reverse triangle inequality but is not injective.
Constant map. \(f(x)=y_0\) has output distance zero for every pair. It is nonexpansive and maximally collapsing, proving that the class is broader than isometries.
Rotation. A rotation of Euclidean space is an isometry and hence nonexpansive, but its iterates need not converge. Nonexpansiveness is not an attractor claim.
Failure case. Scaling by two on a normed space expands every nonzero separation by two and fails the recognition test.

Structural Tensions¶

Stability vs. convergence: errors do not grow, yet iterates may cycle. Diagnostic: analyze asymptotic regularity or averaging separately.
Weakness vs. fixed-point power: the unit bound is broad, so fixed-point conclusions require geometric hypotheses. Diagnostic: list completeness, convexity, compactness, or normal structure before citing a theorem.
Distance loss vs. information loss: shrinking is permitted, including collapse. Diagnostic: test injectivity and lower Lipschitz bounds independently.
Coordinate derivative vs. global metric: a local Jacobian bound may establish the property only under suitable path geometry. Diagnostic: return to the global pair inequality.
Autonomy vs. Function-plus-Metric closure: the parents supply ingredients but not the named unit-Lipschitz morphism class and its iteration calculus. Diagnostic: require the exact bound and closure consequences.

Structural–Framed Character¶

The entry is strongly structural: its identity is one invariant inequality, and its categorical and iterative consequences are independent of coordinates. Its framing is nevertheless mathematical. “Distance,” Lipschitz constant, and fixed-point hypotheses must be literal, not evaluative or institutional.

Structural Core vs. Domain Accent¶

The portable core is controlled transformation: outputs do not separate more than inputs. The domain accent is exact metric comparison for every pair. Removing the metrics leaves the general ideas of Function (Mapping), constraint, or robustness, not a nonexpansive map.

Because the rule occurs within many mathematical and computational subfields but always as the same metric-space construct, breadth does not convert the named object into a prime. The domain-specific residual consists of the unit Lipschitz class, categorical closure, and metric fixed-point boundary.

Function (Mapping) is the proposed minimal parent: a nonexpansive map is a strict function specialization constrained by two metrics. Metric supplies the comparison structure, while Continuity is a guaranteed consequence rather than the best parent. Constraint and robustness are interpretive neighbors but do not define the class.

Relationships to Other Abstractions¶

Current abstraction Metric Map (Nonexpansive Map) Domain-specific

Parents (1) — more general patterns this builds on

Metric Map (Nonexpansive Map) is a kind of Function (Mapping) Prime

Function (Mapping) is the proposed minimal parent: a nonexpansive map is a strict function specialization constrained by two metrics.

Hierarchy path (1) — routes to 1 parentless root

Metric Map (Nonexpansive Map) → Function (Mapping)

Neighborhood in Abstraction Space¶

Metric Map (Nonexpansive Map) sits in a sparse region of the domain-specific corpus (82^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Metric Geometry & Approximation (13 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Contraction mapping: requires a uniform constant strictly below one.
Firmly nonexpansive map: satisfies a stronger inner-product or averagedness inequality.
Isometry: preserves every distance exactly.
Lipschitz map: may have any finite Lipschitz constant, including one greater than one.
Uniformly continuous map: controls small input changes but can expand large or moderate distances.
Metric projection: a particular construction that is nonexpansive only under appropriate geometry.

References¶

[1] Mohamed A. Khamsi and William A. Kirk, An Introduction to Metric Spaces and Fixed Point Theory, Wiley, 2001, especially chapter 8, https://doi.org/10.1002/9781118033074.ch8. registry ↩

[2] William A. Kirk and Brailey Sims, eds., Handbook of Metric Fixed Point Theory, Springer, 2001, https://doi.org/10.1007/978-94-017-1748-9. registry ↩a ↩b

[3] Heinz H. Bauschke and Patrick L. Combettes, Convex Analysis and Monotone Operator Theory in Hilbert Spaces, 2^nd ed., Springer, 2017, https://doi.org/10.1007/978-3-319-48311-5. registry ↩