The Fixed-Point Theorem and Its Applications

Statement and Proof

Statement of the theorem

Banach fixed-point theorem. Let XX be a closed, nonempty subset of R\mathbb{R}, and f:X→Xf : X \to X a kk-contracting mapping, with 0≤k<10 \leq k < 1. Then ff has a unique fixed point cc in XX, and for every u0u_0 in XX, the sequence defined by un+1=f(un)u_{n+1} = f(u_n) converges to cc.

Three hypotheses, none superfluous: XX closed — a segment [a ; b][a \,;\, b] does the job perfectly —, stability f(X)⊂Xf(X) \subset X, and strict contraction k<1k < 1.

Banach's statement, published in 1922, actually holds in any complete metric space, which makes it one of the most widely reused tools in analysis.

Proof

Existence. Start from an arbitrary point u0u_0 of XX and set un+1=f(un)u_{n+1} = f(u_n); stability f(X)⊂Xf(X) \subset X guarantees that the sequence remains well defined in XX.

Applying the contraction property to two consecutive terms gives

∣un+1−un∣=∣f(un)−f(un−1)∣≤k ∣un−un−1∣\lvert u_{n+1} - u_n \rvert = \lvert f(u_n) - f(u_{n-1}) \rvert \leq k\,\lvert u_n - u_{n-1} \rvert

hence, by induction, ∣un+1−un∣≤kn ∣u1−u0∣\lvert u_{n+1} - u_n \rvert \leq k^n\,\lvert u_1 - u_0 \rvert.

For p>np > n, the triangle inequality and the sum of a geometric series then give

∣up−un∣≤∑j=np−1∣uj+1−uj∣≤∣u1−u0∣∑j=np−1kj≤kn1−k ∣u1−u0∣\lvert u_p - u_n \rvert \leq \sum_{j=n}^{p-1} \lvert u_{j+1} - u_j \rvert \leq \lvert u_1 - u_0 \rvert \sum_{j=n}^{p-1} k^j \leq \frac{k^n}{1 - k}\,\lvert u_1 - u_0 \rvert

The upper bound tends to 00 as nn tends to infinity, independently of pp: the sequence (un)(u_n) is Cauchy. Since R\mathbb{R} is complete, it converges to a real number cc, and since XX is closed, cc belongs to XX.

A contraction mapping being continuous, taking the limit in un+1=f(un)u_{n+1} = f(u_n) gives c=f(c)c = f(c).

Uniqueness. Suppose there are two fixed points cc and dd. Then

∣c−d∣=∣f(c)−f(d)∣≤k ∣c−d∣\lvert c - d \rvert = \lvert f(c) - f(d) \rvert \leq k\,\lvert c - d \rvert

so (1−k) ∣c−d∣≤0(1 - k)\,\lvert c - d \rvert \leq 0. Since 1−k>01 - k > 0, it follows that ∣c−d∣=0\lvert c - d \rvert = 0, that is, c=dc = d.

Every hypothesis is necessary

Hypothesis removed Counterexample What fails
XX closed f(x)=x2f(x) = \dfrac{x}{2} on ]0 ; 1]\left]0 \,;\, 1\right] the limit 00 leaves XX
f(X)⊂Xf(X) \subset X f(x)=x2+3f(x) = \dfrac{x}{2} + 3 on [0 ; 1][0 \,;\, 1] the sequence leaves the interval
strict k<1k < 1 f(x)=x+1xf(x) = x + \dfrac{1}{x} on [1 ; +∞[\left[1 \,;\, +\infty\right[ no fixed point
XX nonempty — nothing to prove

Classic pitfall

Forgetting to check stability f(X)⊂Xf(X) \subset X. It is the step people most readily skip, and yet it is the one that requires the real work in an exercise: you must establish bounds on ff over XX, often through a study of its variations. Without it, the sequence of iterates is not even defined.