Nonexpansive Operator Theory Fundamentals

• Nonexpansive operator theory is the study of mappings that do not increase distances between points, providing a framework for fixed-point and convergence analysis.
• Firmly nonexpansive and α-averaged operators are characterized by specific inequalities that ensure stability and convergence in optimization algorithms across Hilbert and Banach spaces.
• The interplay between nonexpansive mappings, monotone operator theory, and operator splitting facilitates algorithm design in variational analysis and data-driven methodologies.

Nonexpansive Operator Theory encompasses the study of operators on metric, normed, or more general geometric spaces that do not increase distances between points. Central to this theory are nonexpansive, strictly nonexpansive, and firmly nonexpansive mappings and their structural, dual, and algorithmic interplay with monotone operator theory, fixed-point theory, and convex optimization. The field has substantial cross-pollination with monotone inclusions, operator splitting, variational analysis, and data-driven methodologies for constructing and leveraging nonexpansive maps in both linear and nonlinear regimes.

1. Foundational Classes: Nonexpansive, Strictly Nonexpansive, and Firmly Nonexpansive Maps

Let $X$ be a real Hilbert space with inner product $\langle\cdot,\cdot\rangle$ and norm $\|\cdot\|$. A mapping $T\colon X\to X$ is nonexpansive if

$$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$

$T$ is strictly nonexpansive if $\|T x - T y\| < \|x - y\|$ for all $x\ne y$. This concept admits a weaker form in general metric spaces: $d(Tx, Ty) < \max\{d(x, y), d(x, Tx), d(y, Ty)\}$ for $x \ne y$ (Shi, 28 Oct 2025).

A mapping $\langle\cdot,\cdot\rangle$0 is firmly nonexpansive if any of the following equivalent conditions hold:

• $\langle\cdot,\cdot\rangle$1 for all $\langle\cdot,\cdot\rangle$2,
• $\langle\cdot,\cdot\rangle$3,
• $\langle\cdot,\cdot\rangle$4 is nonexpansive,
• $\langle\cdot,\cdot\rangle$5 is firmly nonexpansive (Bauschke et al., 2011, Bauschke et al., 2011).

These definitions generalize to $\langle\cdot,\cdot\rangle$6-firmly nonexpansive classes: $\langle\cdot,\cdot\rangle$7 for some $\langle\cdot,\cdot\rangle$8 and nonexpansive $\langle\cdot,\cdot\rangle$9 (Bërdëllima et al., 2021, Berdellima, 2020). In non-Hilbertian geometry, the equivalence with $\|\cdot\|$0-averaged mappings may require additional structural assumptions.

2. Firm Nonexpansiveness and Maximal Monotonicity: The Minty Correspondence and Dualities

A set-valued operator $\|\cdot\|$1 is maximally monotone if its graph is maximal with respect to the monotonicity property

$\|\cdot\|$2

Minty established a bijective correspondence: $\|\cdot\|$3

$\|\cdot\|$4

This is fundamental in fixed-point theory and operator splitting: the unique solvability and convergence of various algorithms is often established by passing between firmly nonexpansive $\|\cdot\|$5 and their monotone operator counterparts $\|\cdot\|$6 (Bauschke et al., 2011, Bauschke et al., 2011).

Dualities and self-dualities underpin this structure. For firmly nonexpansive $\|\cdot\|$7, $\|\cdot\|$8 is also firmly nonexpansive; for $\|\cdot\|$9, the dual is $T\colon X\to X$0. Key self-dual properties include strict firm nonexpansiveness, cyclic firm nonexpansiveness (linked to cyclical monotonicity), and paramonotonicity (Bauschke et al., 2011).

3. Nonexpansive Maps in Banach, Metric, and Geodesic Structures

In $T\colon X\to X$1-uniformly convex Banach spaces (with modulus $T\colon X\to X$2), an operator $T\colon X\to X$3 is $T\colon X\to X$4-firmly nonexpansive if

$T\colon X\to X$5

This setting expands the reach of firm nonexpansiveness and connects to $T\colon X\to X$6-averaged and quasi-$T\colon X\to X$7-firmly nonexpansive mappings (where the contractivity defect is only required in directions toward fixed points) (Bërdëllima et al., 2021).

For general metric and geodesic spaces, the notions adapt using “convex combinations” along geodesics (e.g., in CAT(0), Busemann, or W-hyperbolic spaces), and the relevant inequalities encode curvature effects. For example, in CAT(0) spaces, $T\colon X\to X$8-firmly nonexpansive $T\colon X\to X$9 obey

$$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$0

with $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$1 a geometric bilinear form, reducing to the Hilbert-space case for $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$2 (Berdellima, 2020, Ariza-Ruiz et al., 2012).

4. Iterative Convergence, Regularity, and Algorithmic Implications

For $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$3 nonexpansive with fixed points in a Hilbert or uniformly convex Banach space, the Browder–Göhde–Kirk theorem ensures that iterates $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$4 (Picard iteration) converge weakly to a solution, provided the domain is convex, closed, and bounded (Shi, 28 Oct 2025). Krasnoselskii–Mann or Halpern averages can be essential in this setting.

For strictly nonexpansive $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$5 in an arbitrary complete metric space, and under a bounded orbit assumption,

$$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$6

global strong convergence to a unique fixed point is achievable without compactness or convexity (Shi, 28 Oct 2025).

Firmly nonexpansive and quasi-firmly nonexpansive operators (including resolvents) are asymptotically regular: $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$7 and, under uniform convexity or Opial's property, weak convergence (possibly strong for certain regularized projections) to a fixed point is ensured (Bauschke et al., 2011, Bërdëllima et al., 2021, Ariza-Ruiz et al., 2012).

Operator splitting methods (forward-backward, Douglas-Rachford, ADMM, Peaceman-Rachford, etc.) critically rely on the firm nonexpansiveness or strong nonexpansiveness of component operators to guarantee convergence in monotone inclusion, variational inequality, and convex optimization frameworks (Bauschke et al., 2011, Liu et al., 2022, Bauschke et al., 2011).

5. Structural Calculus: Compositions, Convex Combinations, and Extensions

Compositions and convex combinations of (asymptotically regular) firmly nonexpansive operators preserve firm nonexpansiveness and asymptotic regularity under explicit parameter control. In Hilbert or $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$8-uniformly convex spaces,

• The composition $$\|T x - T y\| \le \|x - y\|,\qquad \forall x, y\in X.$$9 of $T$0-firmly nonexpansive maps is $T$1-firmly nonexpansive with $T$2 computable from the constituent $T$3.
• The convex combination $T$4 retains firm nonexpansiveness with parameter $T$5.

For quasi-firmly nonexpansive operators in CAT(0) or $T$6-uniformly convex spaces, similar closure holds, allowing modular algorithm construction (Berdellima, 2020, Bërdëllima et al., 2021, Bauschke et al., 2011).

In the degenerate-metric framework (systems with singular or semi-definite weights), $T$7-firmly nonexpansive and $T$8-averaged operators admit a parallel calculus, ensuring fixed-point convergence for Krasnoselskii–Mann or generalized PPA-type iterations despite the lack of full-rank geometry (Xue, 2021).

6. Generalized and Learned Firmly Nonexpansive Operators

Data-driven frameworks for learning firmly nonexpansive operators enable construction of operators for use in Plug-and-Play (PnP) processing, image denoising, and plug-in splitting algorithms. Under empirical and expected risk minimization with nonexpansivity constraints, convergence (in Γ-sense) to the population minimizer is established, with practical discretization via piecewise-affine schemes and operator-norm constraints (Bredies et al., 2024).

In barycentric or hybrid geometric settings, Bregman-firmly nonexpansive operators generalize classical proximal maps to product geometries (e.g., Euclidean–KL), preserving firm nonexpansiveness in the relevant Bregman metric and thus extending monotone-inclusion analysis to saddle-point and minimax problems (Achab, 2024).

7. Geometric, Structural, and Universality Principles

The scaled relative graph (SRG) encodes operator classes (nonexpansive, contractive, averaged, firmly nonexpansive) as regions in the complex plane, providing a geometric calculus for checking operator properties and handling convergence analyses via inclusions of SRGs in corresponding disks (Ryu et al., 2019).

The Gurarii space construction realizes a universal nonexpansive linear operator that embeds all nonexpansive linear operators between separable Banach spaces up to isometry, establishing a model for operator-theoretic universality and the analysis of invariant substructures in the nonexpansive setting (Garbulińska-Wȩgrzyn et al., 2013).

---

Table: Structural Relationships and Operator Types

Operator Type	Characterization	Key Closure Properties
Nonexpansive	$T$9	Closed under composition
Strictly nonexpansive	$\|T x - T y\| < \|x - y\|$0, $\|T x - T y\| < \|x - y\|$1	Unique fixed point under bounded orbits (Shi, 28 Oct 2025)
Firmly nonexpansive	$\|T x - T y\| < \|x - y\|$2	Minty correspondence with maximally monotone $\|T x - T y\| < \|x - y\|$3; closed under convex comb/composition (Bauschke et al., 2011, Bërdëllima et al., 2021, Bauschke et al., 2011)
$\|T x - T y\| < \|x - y\|$4-averaged ($\|T x - T y\| < \|x - y\|$5)	$\|T x - T y\| < \|x - y\|$6, $\|T x - T y\| < \|x - y\|$7 nonexpansive	Firmly nonexpansive for $\|T x - T y\| < \|x - y\|$8
GAN (generalized averaged nonexp.)	$\|T x - T y\| < \|x - y\|$9	Local/global convergence rates (Lin et al., 2021)

---

• “Firmly nonexpansive mappings and maximally monotone operators: correspondence and duality” (Bauschke et al., 2011)
• “Innovative Method for Proving Iterative Convergence of Strictly Nonexpansive Operators in Bounded Domains” (Shi, 28 Oct 2025)
• “Compositions and convex combinations of asymptotically regular firmly nonexpansive mappings are also asymptotically regular” (Bauschke et al., 2011)
• “Learning Firmly Nonexpansive Operators” (Bredies et al., 2024)
• “On a notion of averaged operators in CAT(0) spaces” (Berdellima, 2020)
• “A Bregman firmly nonexpansive proximal operator for baryconvex optimization” (Achab, 2024)
• “On $x\ne y$0-Firmly Nonexpansive Operators in $x\ne y$1-Uniformly Convex Spaces” (Bërdëllima et al., 2021)
• “Scaled Relative Graph: Nonexpansive operators via 2D Euclidean Geometry” (Ryu et al., 2019)
• “A universal operator on the Gurarii space” (Garbulińska-Wȩgrzyn et al., 2013)
• “Strongly nonexpansive mappings revisited: uniform monotonicity and operator splitting” (Liu et al., 2022)
• “On the nonexpansive operators based on arbitrary metric: A degenerate analysis” (Xue, 2021)
• “Firmly nonexpansive mappings in classes of geodesic spaces” (Ariza-Ruiz et al., 2012)