Does $\sqrt{2}$ exist? The set $S = \{x \in \mathbb{Q} : x^2 < 2\}$ is nonempty ($1 \in S$) and bounded above ($2$ is an upper bound). If the least upper bound of $S$ existed *in $\mathbb{Q}$*, it would have to be $\sqrt{2}$ — but $\sqrt{2}$ is irrational. In $\mathbb{Q}$, the set $S$ has upper bounds but no *least* upper bound: there is no smallest rational number that exceeds every element of $S$. The rationals have a gap at $\sqrt{2}$ — and at $\pi$, and at $e$, and at every irrational number. The **completeness axiom** of $\mathbb{R}$ asserts that these gaps do not exist in the real numbers: every nonempty set of reals that is bounded above has a least upper bound (supremum) *in $\mathbb{R}$*. This single axiom is the foundation of real analysis. Without it, bounded monotone sequences might not converge, the [Intermediate Value Theorem](/theorems/180) might fail, and [Cauchy sequences](/page/Cauchy%20Sequence) might have no limit. Every theorem in IA Analysis that goes beyond pure algebra — every theorem that requires "taking [limits](/page/Limit)" — ultimately depends on the completeness of $\mathbb{R}$. ## Definitions ### Bounds [definition:Upper Bound] Let $S \subseteq \mathbb{R}$ be nonempty. A real number $M$ is an **upper bound** for $S$ if $x \leq M$ for all $x \in S$. The set $S$ is **bounded above** if it has at least one upper bound. [/definition] [definition:Lower Bound] A real number $m$ is a **lower bound** for $S$ if $x \geq m$ for all $x \in S$. The set $S$ is **bounded below** if it has at least one lower bound. The set $S$ is **bounded** if it is bounded above and below. [/definition] If $M$ is an upper bound for $S$ and $M' > M$, then $M'$ is also an upper bound. So the set of upper bounds of $S$ is an interval $[M_0, \infty)$ for some $M_0$ — or is it? The question is whether this interval has a left endpoint: is there a *smallest* upper bound? In $\mathbb{Q}$, the answer can be no (as the $\sqrt{2}$ example shows). In $\mathbb{R}$, the answer is always yes. ### Supremum and Infimum [definition:Supremum] Let $S \subseteq \mathbb{R}$ be nonempty and bounded above. The **supremum** (or **least upper bound**) of $S$, written $\sup S$, is the real number $K$ satisfying: \begin{align*} &\text{(i) } x \leq K \text{ for all } x \in S \quad \text{($K$ is an upper bound)}, \\ &\text{(ii) if } M < K \text{ then } M \text{ is not an upper bound of } S \quad \text{($K$ is the least such)}. \end{align*} Equivalently, condition (ii) says: for every $\varepsilon > 0$, there exists $x \in S$ with $x > K - \varepsilon$. [/definition] [definition:Infimum] Let $S \subseteq \mathbb{R}$ be nonempty and bounded below. The **infimum** (or **greatest lower bound**) of $S$, written $\inf S$, is the real number $k$ satisfying: \begin{align*} &\text{(i) } x \geq k \text{ for all } x \in S, \\ &\text{(ii) for every } \varepsilon > 0, \text{ there exists } x \in S \text{ with } x < k + \varepsilon. \end{align*} [/definition] The equivalence in condition (ii) — between "$K$ is the *least* upper bound" and "elements of $S$ get *arbitrarily close* to $K$ from below" — is the key to using suprema in practice. It transforms the abstract "least upper bound" into a concrete approximation tool. [example:Suprema Of Intervals And Finite Sets] $\sup(0, 1) = 1$ (no element of $(0, 1)$ equals $1$, but elements get arbitrarily close). $\sup[0, 1] = 1$ (here $1 \in [0, 1]$, so the supremum is attained). $\sup\{1/n : n \in \mathbb{N}\} = 1$ (attained at $n = 1$). $\inf\{1/n : n \in \mathbb{N}\} = 0$ (not attained — no $n$ gives $1/n = 0$, but $1/n$ gets arbitrarily close to $0$). The supremum need not belong to the set. Whether $\sup S \in S$ depends on whether $S$ is closed "at the top." The supremum always exists as a real number (by the completeness axiom), but it may or may not be a *member* of $S$. [/example] ## The Completeness Axiom [definition:Completeness Axiom] **(Least Upper Bound Property.)** Every nonempty subset of $\mathbb{R}$ that is bounded above has a supremum in $\mathbb{R}$. [/definition] This axiom is what *defines* $\mathbb{R}$ — or more precisely, what distinguishes $\mathbb{R}$ from $\mathbb{Q}$. Both $\mathbb{Q}$ and $\mathbb{R}$ are ordered fields (they support addition, multiplication, and comparison), but only $\mathbb{R}$ is *complete*. The completeness axiom fills every gap in $\mathbb{Q}$: wherever the rationals have a "missing point" (like $\sqrt{2}$), the reals have a supremum that plugs the hole. The infimum version follows automatically: if $S$ is nonempty and bounded below, consider $-S = \{-x : x \in S\}$, which is bounded above. Then $\inf S = -\sup(-S)$. ### The Approximation Property The most useful consequence of the definition — and the tool used in virtually every proof involving suprema — is the **approximation property**: for every $\varepsilon > 0$, there exists $x \in S$ with $\sup S - \varepsilon < x \leq \sup S$. This is simply condition (ii) restated. It says that the supremum can be *approximated from within $S$* to any desired accuracy. [example:Using The Approximation Property] Let $A = \{x \in \mathbb{R} : x^2 < 2\}$. We claim $\sup A = \sqrt{2}$. Setting $K = \sup A$: **$K^2 \leq 2$:** Suppose $K^2 > 2$. Then $K^2 - 2 > 0$. Choose $\delta > 0$ small enough that $(K - \delta)^2 > 2$ (possible by [continuity](/page/Continuity) of $x^2$). Then $K - \delta$ is also an upper bound for $A$: if $x \in A$, then $x^2 < 2 < (K - \delta)^2$, so $x < K - \delta$ (since both are positive). This contradicts $K = \sup A$ being the *least* upper bound. So $K^2 \leq 2$. **$K^2 \geq 2$:** Suppose $K^2 < 2$. Then $2 - K^2 > 0$. Choose $\delta > 0$ small enough that $(K + \delta)^2 < 2$. Then $K + \delta \in A$, contradicting $K$ being an upper bound. So $K^2 \geq 2$. Therefore $K^2 = 2$, i.e., $K = \sqrt{2}$. This argument — "assume $K$ is too big, find a smaller upper bound; assume $K$ is too small, find an element of $S$ above it" — is the standard template for computing suprema. The approximation property provides the elements of $S$ needed to reach a contradiction. [/example] ## Consequences of Completeness The completeness axiom, despite its simple statement, implies every major structural property of $\mathbb{R}$ that distinguishes it from $\mathbb{Q}$. We derive the most important consequences. ### The Archimedean Property For any real $x > 0$ and $y > 0$, there exists $n \in \mathbb{N}$ with $nx > y$. Equivalently: $\mathbb{N}$ is unbounded in $\mathbb{R}$, and for any $\varepsilon > 0$, there exists $n$ with $1/n < \varepsilon$. The idea: if $\mathbb{N}$ were bounded, completeness would give $K = \sup \mathbb{N}$. The approximation property provides $n_0 > K - 1$, so $n_0 + 1 > K$ — but $n_0 + 1 \in \mathbb{N}$, contradicting the supremum. Therefore $\mathbb{N}$ is unbounded. The Archimedean property says there are no "infinitely large" or "infinitely small" real numbers — the natural numbers eventually exceed any real, and the reciprocals $1/n$ eventually undercut any positive real. This fails in non-Archimedean ordered fields (like the hyperreals), where infinitesimals exist. ### Density of $\mathbb{Q}$ in $\mathbb{R}$ Between any two distinct real numbers lies a rational number. Equivalently, every open interval $(a, b)$ with $a < b$ contains a rational. The argument uses the Archimedean property twice: first to choose $n$ with $1/n < b - a$ (so the rationals $k/n$ are spaced closely enough to land inside $(a, b)$), then to find the right integer $k$ with $a < k/n < b$. The irrationals are also dense (since $a + \sqrt{2}/n$ is irrational for any rational $a$), so $\mathbb{Q}$ and $\mathbb{R} \setminus \mathbb{Q}$ are both dense in $\mathbb{R}$ — intricately interwoven at every scale. ### The Monotone Convergence Theorem [quotetheorem:626] This is the most important single application of completeness to sequences. The idea: for an increasing bounded sequence $(a_n)$, the set $S = \{a_n : n \in \mathbb{N}\}$ is bounded above, so $K = \sup S$ exists by completeness. The approximation property gives $a_N > K - \varepsilon$ for some $N$; monotonicity then forces $a_n \geq a_N > K - \varepsilon$ for all $n \geq N$, and $a_n \leq K$ gives $|a_n - K| < \varepsilon$. Without completeness, the theorem fails: the increasing bounded sequence $1, 1.4, 1.41, 1.414, \ldots$ (decimal approximations to $\sqrt{2}$) has no limit in $\mathbb{Q}$. ### The Nested Intervals Theorem If $[a_1, b_1] \supseteq [a_2, b_2] \supseteq [a_3, b_3] \supseteq \cdots$ is a nested sequence of closed bounded intervals with $b_n - a_n \to 0$, then $\bigcap_{n=1}^{\infty} [a_n, b_n] = \{c\}$ for a unique real number $c$. The idea: $(a_n)$ is increasing and bounded above (by $b_1$), so it converges to $\alpha = \sup\{a_n\}$ by the [Monotone Convergence Theorem](/theorems/626). Similarly $(b_n)$ decreases to $\beta = \inf\{b_n\}$. Since $b_n - a_n \to 0$, $\alpha = \beta = c$, and this common value is the unique point in the intersection. The nested intervals theorem is the engine behind the bisection proof of the [Bolzano-Weierstrass Theorem](/theorems/628): repeatedly halve a bounded interval, choosing the half containing infinitely many sequence terms, to produce a convergent subsequence. ## Equivalent Formulations of Completeness The completeness axiom can be stated in several equivalent ways. Each formulation captures a different facet of "no gaps in $\mathbb{R}$": **Least upper bound property** (our axiom): every nonempty bounded-above subset has a supremum. **Monotone convergence**: every bounded monotone sequence converges. **Nested intervals**: nested closed intervals with shrinking lengths have nonempty intersection. **[Bolzano-Weierstrass](/theorems/628)**: every bounded sequence has a convergent subsequence. **[Cauchy completeness](/page/Cauchy%20Sequence)**: every Cauchy sequence converges. **Dedekind completeness**: every Dedekind cut determines a real number. These are all equivalent over the Archimedean ordered fields — assuming the field is Archimedean and ordered, any one of these properties implies all the others. The proofs form a cycle: least upper bound $\Rightarrow$ monotone convergence (proved above) $\Rightarrow$ nested intervals (via monotone limits of endpoints) $\Rightarrow$ Bolzano-Weierstrass (via bisection) $\Rightarrow$ Cauchy completeness (bounded Cauchy has convergent subsequence, Cauchy property forces full convergence) $\Rightarrow$ least upper bound (construct $\sup S$ as the limit of a Cauchy sequence of increasingly tight upper bounds). The equivalence means that completeness is a single, robust property of $\mathbb{R}$ — it doesn't matter which formulation you take as the axiom, since the others all follow. The least upper bound property is the standard choice because it is the most elementary (it involves only [sets](/page/Set) and inequalities, not sequences or limits), but the [Cauchy completeness](/page/Cauchy%20Sequence) formulation is the one that generalises to [metric spaces](/page/Metric%20Space) (where suprema are meaningless but Cauchy sequences make sense). ## Supremum and Infimum in Practice ### Computing Suprema of Sequences and [Functions](/page/Function) For a [sequence](/page/Sequence) $(a_n)$, $\sup\{a_n : n \in \mathbb{N}\}$ is the smallest number $K$ with $a_n \leq K$ for all $n$. For a function $f: E \to \mathbb{R}$, $\sup_{x \in E} f(x) = \sup\{f(x) : x \in E\}$. In both cases, the approximation property is the essential tool: to show $\sup = K$, prove (i) $f(x) \leq K$ for all $x$, and (ii) for every $\varepsilon > 0$, find $x_0$ with $f(x_0) > K - \varepsilon$. [example:Supremum Of A Sequence] Let $a_n = 1 - 1/n$ for $n \geq 1$. Then $a_n < 1$ for all $n$ (so $1$ is an upper bound) and $a_n = 1 - 1/n > 1 - \varepsilon$ for $n > 1/\varepsilon$ (the approximation property). So $\sup\{a_n\} = 1$. Note that $\sup\{a_n\} = 1$ but $\max\{a_n\}$ does not exist — no term equals $1$. The supremum always exists (by completeness); the maximum exists only if the supremum is attained. [/example] ### Suprema and the $\varepsilon$-$\delta$ Framework The supremum and infimum pervade the $\varepsilon$-$\delta$ definitions of analysis. The *supremum norm* $\|f\|_\infty = \sup_{x \in E} |f(x)|$ measures the "worst-case" size of a function and is the metric that makes [uniform convergence](/page/Uniform%20Convergence) precise: $f_n \to f$ uniformly iff $\|f_n - f\|_\infty \to 0$. The *oscillation* of $f$ on an interval $[a, b]$ is $\sup_{x,y \in [a,b]} |f(x) - f(y)|$, and a function is [uniformly continuous](/page/Continuity%20(Real%20Analysis)) iff its oscillation on intervals of width $\delta$ tends to $0$ as $\delta \to 0$. The [Riemann integral](/page/Riemann%20Integral) is defined via upper and lower sums, which are suprema and infima of $f$ on partition subintervals. ### The $\limsup$ and $\liminf$ For any bounded [sequence](/page/Sequence) $(a_n)$, the **limit superior** and **limit inferior** are defined by \begin{align*} \limsup_{n \to \infty} a_n = \lim_{n \to \infty} \sup_{k \geq n} a_k, \qquad \liminf_{n \to \infty} a_n = \lim_{n \to \infty} \inf_{k \geq n} a_k. \end{align*} The sequence $(\sup_{k \geq n} a_k)$ is decreasing (removing terms can only decrease the supremum) and bounded below, so it converges by the [Monotone Convergence Theorem](/theorems/626). Similarly $(\inf_{k \geq n} a_k)$ is increasing and bounded above. The $\limsup$ is the largest cluster point of the sequence, and $\liminf$ is the smallest. The sequence converges if and only if $\limsup = \liminf$, in which case the common value is the limit. The $\limsup$ appears in the [Cauchy-Hadamard formula](/page/Power%20Series) for the [radius of convergence](/theorems/273) of a power series: $1/R = \limsup |a_n|^{1/n}$. It is the natural tool whenever a quantity depends on the "worst-case" or "asymptotically largest" behaviour of a sequence — situations where the ordinary limit may not exist but the $\limsup$ always does.