Completeness

A space with no missing points, where every bunching sequence actually lands on something inside the space.

Why Does This Exist?

Iterative methods produce sequences that bunch together: successive iterates get closer and closer. Bunching alone does not guarantee they approach anything real. In the rationals, approximations $1, 1.4, 1.41, 1.414$ to $\sqrt{2}$ are Cauchy yet have no rational limit.

Completeness closes that gap. A space is complete when every Cauchy sequence converges to a point in the space. The reals are complete; Banach spaces power fixed-point solvers because of it.

This page defines Cauchy versus convergent and states the fixed-point payoff. Metric language is in metric spaces and normed instances in norms.

The one idea you need first

A sequence is Cauchy when its terms get arbitrarily close to each other: for every $\epsilon > 0$ there is $N$ with $m, n > N$ implying $d(x_m, x_n) < \epsilon$ . Convergent implies Cauchy; the reverse needs completeness.

Think of It Like This

A net under trapeze artists

Cauchy means the artists swing closer together with each pass. Completeness means a net hangs below: wherever they bunch, something catches them. An incomplete space is a missing net panel; the artists converge over the hole and fall through.

The rationals miss the panel at $\sqrt{2}$ . The reals Mie the full net. Where the analogy stops: nets catch bodies, but completeness also supports theorems like Banach iteration that need the limit to feed back into the map.

How It Actually Works

In a metric space $(X, d)$ , completeness means: if $(x_n)$ is Cauchy then some $x$ in $X$ has $x_n \to x$ . $\mathbb{R}$ with $| \cdot |$ is complete. $(0,1)$ with the same metric is not: $x_n = 1/n$ is Cauchy but its limit $0$ lies outside.

Banach fixed-point theorem: a contraction $T$ with $d(Tx, Ty) \le q\, d(x,y)$ for $q < 1$ on a complete space has exactly one fixed point, reached by iterating $x_{n+1} = T(x_n)$ from any start, with error $\le q^n d(x_1, x_0)/(1-q)$ .

Worked example

Solve $x = \cos x$ by $x_{n+1} = \cos x_n$ from $x_0 = 0.5$ . Values: $0.8776, 0.6390, 0.8027, 0.6948, 0.7682$ . Gaps: $0.3776, 0.2386, 0.1637, 0.1079, 0.0734$ , shrinking by about $0.7$ each step. The limit $0.7391$ lies in $[0,1]$ , which is complete, so iteration is justified. On $(0,1)$ open at $0$ the same logic would still land inside here, but $x_n = 1/n$ shows why openness generally breaks the guarantee.

Watch Out For

Calling Cauchy convergent without checking the space

Symptom: declaring an optimizer converged because steps shrink, while iterates drift toward a boundary outside the feasible set. Fix: verify the feasible set is complete (closed inside a complete space) before reading Cauchy as convergent.

Iterating a non-contraction and expecting Banach rates

Symptom: $q \ge 1$ yet linear convergence is claimed; errors stall at $0.05$ for fifty steps. Fix: estimate a local Lipschitz constant numerically; shrink the region or change the map until $q < 1$ holds.

The Quick Version

Cauchy means terms bunch together; convergent means they land on a point in the space.
Complete spaces keep every Cauchy limit inside, as the reals do for $\sqrt{2}$ approximations.
$(0,1)$ is incomplete because $1/n$ aims at missing $0$ .
Contractions on complete spaces converge from any start with geometric error bounds.
Shrinking steps alone prove nothing until completeness of the set is confirmed.