The Completeness Axiom

The Completeness Axiom is invisible machinery behind almost every "obvious" fact in calculus. The Intermediate Value Theorem — that a continuous function taking a negative and a positive value must vanish somewhere — is proved by taking the supremum of the set of points where the function is still negative; without completeness that supremum might not exist and the theorem would be false, exactly as it fails over $\mathbb{Q}$. The Bolzano–Weierstrass Theorem (every bounded sequence has a convergent subsequence) and the Monotone Convergence Theorem both reduce to completeness via nested intervals, the same device used in Example 2. Cauchy sequences converging — the precise sense in which $\mathbb{R}$ has "no missing limit points" — is logically equivalent to completeness together with the Archimedean property. Beyond pure analysis, completeness underlies numerical analysis, where bisection and Newton's-method convergence proofs implicitly assume the reals have no gaps for the iterates to fall into. In probability theory, the existence of suprema and infima of bounded families of random variables (needed to define essential suprema, and to build measure-theoretic expectations) rests on the same axiom. In functional analysis and Banach space theory, completeness of $\mathbb{R}$ is the base case that gets generalized to completeness of metric spaces, and ultimately to Banach and Hilbert spaces central to quantum mechanics and PDE theory. In computer science and cryptography, floating-point and interval-arithmetic systems are explicitly incomplete approximations to $\mathbb{R}$, and understanding precisely how and why they diverge from true completeness (rounding, representable gaps) is essential to certifying numerical algorithms and error bounds.

Ask the tutor: Can you show me why the Completeness Axiom, together with the Archimedean Property, is exactly equivalent to every Cauchy sequence of reals converging?

Think of $\mathbb{Q}$ as a number line riddled with pinholes: it looks continuous, but certain sets of rationals — like $\{x \in \mathbb{Q} : x^2 < 2\}$ — pile up toward a point that is not there. The set has plenty of upper bounds in $\mathbb{Q}$ (any rational bigger than $\sqrt{2}$ works), but no smallest one, because between any candidate upper bound and $\sqrt{2}$ you can always squeeze in a strictly smaller rational upper bound. The supremum is trying to exist at exactly the missing point. The Completeness Axiom says: postulate a number system where this can never happen. Whenever a set has upper bounds at all, one of them is guaranteed to be the least, and that least bound is a genuine element of the system. Geometrically, this is the statement that the real line has no punctures — it is a true continuum, not a fine dust of points with gaps between clusters.

The Completeness Axiom is phrased in terms of suprema, but it forces an identical statement about infima. This equivalence is proved once and then used freely.

Example 1: The square root of a non-square positive integer exists in $\mathbb{R}$, using completeness directly. We show $\sqrt{3}$ exists as a real number — deliberately choosing $3$, not $2$, to avoid the reader pattern-matching to a memorized special case. Let

$S$ is nonempty: $0 \in S$ since $0^2 = 0 < 3$. $S$ is bounded above: $2 \in S$'s upper-bound set since if $x > 2$ then $x^2 > 4 > 3$, so no element of $S$ exceeds $2$; thus $2$ is an upper bound. By the Completeness Axiom, $\alpha := \sup S$ exists in $\mathbb{R}$, and $\alpha > 0$ since $1 \in S$ (as $1^2 = 1< 3$), forcing $\alpha \ge 1$. We claim $\alpha^2 = 3$, ruling out $\alpha^2 < 3$ and $\alpha^2 > 3$ separately. Case $\alpha^2 < 3$: We produce $h > 0$ with $(\alpha + h)^2 < 3$, contradicting that $\alpha$ is an upper bound of $S$. For $0 < h < 1$,

Choose

(the right side is positive since $3 - \alpha^2 > 0$ and $2\alpha+1>0$). Then $h(2\alpha+1) \le 3 - \alpha^2$, so $(\alpha+h)^2 < \alpha^2 + (3-\alpha^2) = 3$. Hence $\alpha + h \in S$, but $\alpha + h > \alpha = \sup S$ — contradicting that $\alpha$ is an upper bound. So $\alpha^2 < 3$ is impossible. Case $\alpha^2 > 3$: We produce $k>0$ with $\alpha - k$ still an upper bound of $S$, contradicting that $\alpha$ is the least upper bound. Take

Then

So $(\alpha-k)^2 > 3$. Now for any $x \in S$ with $x \ge 0$ and $x^2 < 3 < (\alpha-k)^2$, since both $x, \alpha - k > 0$ ($\alpha - k > 0$ because $k<\alpha$) and squaring is strictly increasing on nonnegatives, $x < \alpha - k$. So $\alpha - k$ is an upper bound of $S$ strictly less than $\alpha$ — contradicting minimality of $\alpha$. So $\alpha^2 > 3$ is impossible. By trichotomy of order, $\alpha^2 = 3$. Since $\alpha \ge 1 > 0$, $\alpha$ is the nonnegative real square root of $3$: $\alpha = \sqrt{3}$. This is Existence of Square Roots proved directly from completeness; no such $\alpha$ exists in $\mathbb{Q}$ because $\sqrt3$ is irrational (same style of proof as for $\sqrt2$).

Example 2: Nested closed intervals with shrinking length have a common point — completeness via infima. Let $(I_n)_{n=1}^\infty$ be a sequence of closed bounded intervals $I_n = [a_n, b_n]$ with $I_1 \supseteq I_2 \supseteq I_3 \supseteq \cdots$ (nested), where $a_n \le b_n$ for all $n$. We show $\bigcap_{n=1}^\infty I_n \neq \emptyset$, using an asymmetric explicit example alongside the general argument: take

Check nesting directly: $a_n = 1 - 1/n$ is increasing in $n$ (as $n$ grows, $1/n$ shrinks, so $a_n$ grows), and $b_n = 2 + 1/n^2$ is decreasing in $n$. So $a_1 \le a_2 \le a_3 \le \cdots$ and $b_1 \ge b_2 \ge b_3 \ge \cdots$, and since $a_n < 2 \le b_n$ for every $n$ (as $a_n<1<2\le b_n$), each $I_n$ is a genuine nonempty closed interval and $I_{n+1}\subseteq I_n$.

General argument. Let $A = \{a_n : n \ge 1\}$. This set is nonempty and bounded above (by $b_1$, since $a_n \le b_n \le b_1$ for all $n$ — the last inequality because $(b_n)$ is decreasing). By the Completeness Axiom, $\alpha := \sup A$ exists in $\mathbb{R}$. Claim: $a_n \le \alpha \le b_n$ for every $n$, so $\alpha \in \bigcap_n I_n$. The inequality $a_n \le \alpha$ holds because $\alpha$ is an upper bound of $A \ni a_n$. For $\alpha \le b_n$: fix $n$. For every $m \ge n$, nesting gives $a_m \le b_m \le b_n$ (using $I_m \subseteq I_n$, hence $a_m\ge a_n$ and $b_m\le b_n$, and $a_m\le b_m$). For $m < n$, nesting the other way gives $a_m \le a_n \le b_n$ (since $I_n \subseteq I_m$ forces $a_m \le a_n$). Either way, $a_m \le b_n$ for every $m$, so $b_n$ is an upper bound of $A$. Since $\alpha = \sup A$ is the least upper bound, $\alpha \le b_n$. So $a_n \le \alpha \le b_n$ for every $n$, i.e. $\alpha \in I_n$ for every $n$. Hence $\alpha \in \bigcap_{n=1}^\infty I_n \neq \emptyset$. In the concrete instance, $A = \{1 - 1/n : n\ge1\}$ is bounded above by $b_1 = 3$, and one checks $\sup A = 1$ (since $1-1/n \to 1$ from below and $1$ is an upper bound: $1 - 1/n < 1$ always). Also $\inf$ of the $b_n$'s over the tail is $2$ (as $1/n^2 \to 0$), and indeed $1 \le 2$, consistent with $\alpha = 1 \in [a_n,b_n]$ for all $n$: e.g. $I_1 = [0,3]$, $I_5 = [0.8, 2.04]$, and $1$ lies in every one of them. This nested-interval fact is the seed of The Archimedean Property and of compactness arguments built later on closed bounded intervals.