Can a contraction be pseudocontractive?

Yes. Every contraction is non-expansive and therefore pseudocontractive.

Does pseudocontractivity guarantee a unique solution?

No. The identity mapping satisfies it and fixes every point.

What does strict pseudocontractivity add?

A common definition multiplies the residual-squared term by a constant smaller than one. The precise convention and theorem should be stated.

AI Glossary / Mathematics and optimization / Advanced

Pseudocontractive operator

Definition

A pseudocontractive operator is a mapping T on a Hilbert space, or a suitable subset, for which ||Tx−Ty||² ≤ ||x−y||² + ||(I−T)x−(I−T)y||² for every pair of inputs. Equivalently, the residual operator I−T is monotone.

What it means

The name sounds like a promise that repeated application brings points closer together. It is a weaker condition. The residual term on the right allows a pseudocontractive mapping to increase the distance between two inputs.

Here I means the identity mapping, which leaves an input unchanged. The residual at x is x−Tx: the difference between the current point and the point proposed by the mapping. Monotonicity concerns the inner product between differences in residuals and differences in inputs. It does not simply mean that every output coordinate increases.

The concept appears in fixed-point methods and mathematical optimization. It helps describe families of mappings for which carefully designed algorithms can be studied. It does not, by itself, establish that an unmodified iteration converges or that a fixed point exists.

How it works

Expand the squared residual difference in the definition and cancel matching terms. The condition becomes ⟨Tx−Ty,x−y⟩ ≤ ||x−y||². Rearranging gives ⟨(I−T)x−(I−T)y,x−y⟩ ≥ 0, which is the monotonicity condition for I−T.

Every non-expansive mapping is pseudocontractive: its output distance is already at most the input distance, so adding a nonnegative residual term preserves the inequality. The reverse implication fails.

For a real scalar affine mapping T(x)=ax+b, pseudocontractivity requires a≤1. Non-expansiveness requires |a|≤1. A large negative slope can therefore satisfy the first condition while violating the second.

Researchers often add continuity, Lipschitz bounds, a nonempty fixed-point set or restrictions on iteration parameters. Those assumptions belong to particular convergence results. A theorem about strictly pseudocontractive mappings also has a stronger condition than ordinary pseudocontractivity and should not be silently transferred.

Three practical examples

An optimization researcher checks whether a proposed update has a monotone residual before selecting a fixed-point algorithm.
A numerical methods developer uses a simple expansive pseudocontractive mapping to test whether documentation overstates a solver's convergence guarantee.
A technical reviewer separates a proven mapping property from the additional assumptions needed for a distributed iteration to converge.

Think of it this way

Think of an instruction that can overshoot its destination but whose correction has a consistent mathematical orientation. The orientation may be useful when designing a moderated response, even though following the full instruction repeatedly can be unstable.

This analogy does not define the property. In several dimensions, orientation is measured through an inner product, and the required inequality must hold for every pair of permitted inputs.

Worked example

A mapping that expands distances

Let T(x)=−2x on the real line and write d=x−y. The output difference is −2d, and the residual difference is 3d. The defining inequality becomes 4d² ≤ d²+9d², which always holds. Thus T is pseudocontractive even though it doubles distances.

Its only fixed point solves x=−2x, so x=0. Starting at 1, the direct iteration produces 1, −2, 4, −8 and diverges in magnitude.

For this particular mapping, a relaxed update with weight θ is (1−θ)x+θT(x)=(1−3θ)x. Choosing θ=1/2 gives the sequence 1, −1/2, 1/4, −1/8, which converges to zero. More generally, this scalar example converges when 0<θ<2/3 because |1−3θ|<1.

That calculation establishes a result for this mapping. It does not prove that the same relaxation weight works for every pseudocontractive operator.

Illustrative case study

A fictional software laboratory builds a teaching library of fixed-point methods. A draft tutorial labels every pseudocontractive update as distance reducing. A reviewer inserts T(x)=−2x and immediately exposes the mistake.

The team revises the tutorial to display the operator inequality, the fixed-point equation and the iteration rule separately. It then shows how relaxation changes this example's behavior while leaving its fixed point unchanged.

The laboratory also adds T(x)=x+1, which is pseudocontractive but has no fixed point on the real line. This prevents a second misleading implication: a mapping classification alone does not promise an available solution.

Common misconceptions and mistakes

Pseudocontractive means contractive. It includes mappings that expand distances.
A fixed point must exist. Translation by a nonzero constant supplies a counterexample.
Monotone residual means coordinatewise increasing in every dimension. The relevant definition is an inner-product inequality.
Pseudomonotone and pseudocontractive are interchangeable. They describe different mathematical properties.
Any relaxation establishes convergence. Valid parameters depend on the mapping and the theorem's assumptions.

Business use

This is mainly a technical assurance term for teams assessing optimization software or algorithm research. Its business relevance is the quality of claims about stability and convergence.

Ask suppliers which mapping satisfies the condition, on what domain, and which complete theorem supports their iteration. A numerical example can reveal a false general claim, but successful examples cannot prove the inequality for every input pair. Keep algorithm assumptions visible when translating research into software requirements.

Check your understanding

A mapping doubles input distances but satisfies the pseudocontractive inequality. Is that contradictory?

Answer: No. The residual term permits expansion. For T(x)=−2x, the inequality reads 4d²≤10d², even though direct fixed-point iteration diverges from any nonzero start.

Related terms

Fixed point
monotone operator
non-expansive mapping
contraction mapping
firmly non-expansive mapping
relaxed iteration

Learn next

Read non-expansive mapping and fixed-point iteration. Compare the scalar conditions for T(x)=ax+b, then calculate which relaxation weights make the example T(x)=−2x contractive.

Fixed-point iteration

Sources

Alakoya and colleagues, On a system of monotone variational inclusion problems with fixed-point constraint, Definition 2.5: https://doi.org/10.1186/s13660-022-02782-4. Primary text provides the inequality and monotone-residual equivalence.

Aalto Dictionary of Machine Learning: https://aaltodictionaryofml.github.io/ADictML_Main.pdf. Its displayed example T(x)=x−max(0,x) is actually non-expansive; this entry replaces that misclassification with the independently calculated example above.

Previous: Pseudocolumn Next: Pseudocounts