Contractions and Iterated Sequences

Fixed Points and Contraction Mappings

Fixed point

A fixed point of a function ff is a point that ff does not move:

f(c)=cf(c) = c

Graphically, it is an intersection point of the graph of ff with the line y=xy = x. Solving an equation g(x)=0g(x) = 0 can always be turned into a fixed-point search, for instance by setting f(x)=x+g(x)f(x) = x + g(x) or f(x)=x−g(x)λf(x) = x - \dfrac{g(x)}{\lambda}: this change of viewpoint is what makes iteration possible.

Contraction mappings

A function ff defined on a subset XX of R\mathbb{R} is a contraction — also called kk-contracting — if there exists a constant kk with 0≤k<10 \leq k < 1 such that

∀(x,y)∈X2,∣f(x)−f(y)∣≤k ∣x−y∣\forall (x, y) \in X^2,\quad \lvert f(x) - f(y) \rvert \leq k\,\lvert x - y \rvert

This is exactly the definition of a kk-Lipschitz function, with the additional and decisive requirement k<1k < 1: the function brings points closer together, by a factor of at least kk each time it is applied.

The mean value inequality gives the practical criterion: if ff is differentiable on an interval II with ∣f′(x)∣≤k<1\lvert f'(x) \rvert \leq k < 1, then ff is kk-contracting on II.

Examples and counterexamples

Function Domain Contraction?
f(x)=cos⁡xf(x) = \cos x [0 ; 1][0 \,;\, 1] yes, k=sin⁡1≈0.841k = \sin 1 \approx 0.841
f(x)=x+53f(x) = \dfrac{x + 5}{3} R\mathbb{R} yes, k=13k = \dfrac{1}{3}
f(x)=12(x+2x)f(x) = \dfrac{1}{2}\left(x + \dfrac{2}{x}\right) [1 ; 2][1 \,;\, 2] yes, k=12k = \dfrac{1}{2}
f(x)=x+1f(x) = x + 1 R\mathbb{R} no, k=1k = 1
f(x)=x2f(x) = x^2 [0 ; 2][0 \,;\, 2] no, ∣f′∣\lvert f' \rvert reaches 44

Let us detail the first row: on [0 ; 1][0 \,;\, 1], ∣cos⁡′(x)∣=∣sin⁡x∣≤sin⁡1≈0.841<1\lvert \cos'(x) \rvert = \lvert \sin x \rvert \leq \sin 1 \approx 0.841 < 1, since sine is increasing on this interval. The cosine function is therefore a contraction there.

Classic pitfall

Confusing "contraction" with "strictly brings points closer". The condition

∀x≠y,∣f(x)−f(y)∣<∣x−y∣\forall x \neq y,\quad \lvert f(x) - f(y) \rvert < \lvert x - y \rvert

is weaker than being a contraction, and it is not enough. On [1 ; +∞[\left[1 \,;\, +\infty\right[, the function f(x)=x+1xf(x) = x + \dfrac{1}{x} satisfies f′(x)=1−1x2∈[0 ; 1[f'(x) = 1 - \dfrac{1}{x^2} \in \left[0 \,;\, 1\right[, so it strictly brings every pair of points closer — and yet it has no fixed point, since f(x)=xf(x) = x would require 1x=0\dfrac{1}{x} = 0. What is missing is a uniform constant k<1k < 1.