Take any introductory calculus course and you will encounter the following fact: every bounded increasing [sequence](/page/Sequence) converges. This is stated as though it were obvious, but it is not. The statement is *false* for the rational numbers $\mathbb{Q}$. Consider the sequence of rational truncations of $\sqrt{2}$: \begin{align*} a_1 &= 1, \quad a_2 = 1.4, \quad a_3 = 1.41, \quad a_4 = 1.414, \quad \ldots \end{align*} Each term is rational, the sequence is strictly increasing, and it is bounded above by $2$. Yet it has no limit in $\mathbb{Q}$: the number it "wants" to converge to, $\sqrt{2}$, is not rational. The rationals have a *hole* at $\sqrt{2}$, and analysis — the study of [limits](/page/Limit), derivatives, integrals, and infinite [series](/page/Series) — collapses if the number line has holes. Every $\varepsilon$-$\delta$ argument, every appeal to a limit, every construction of the integral, implicitly assumes that these holes do not exist. The real numbers $\mathbb{R}$ are the unique ordered field in which no such holes remain. [example: Irrationality of $\sqrt{2}$] We recall the proof that $\sqrt{2} \notin \mathbb{Q}$ to make the deficiency of $\mathbb{Q}$ precise. Suppose for contradiction that $\sqrt{2} = p/q$ with $p, q \in \mathbb{Z}$, $q \neq 0$, and $\gcd(p, q) = 1$. Then $p^2 = 2q^2$, so $p^2$ is even, which forces $p$ to be even. Write $p = 2k$. Then $4k^2 = 2q^2$, so $q^2 = 2k^2$, and by the same argument $q$ is even. This contradicts $\gcd(p, q) = 1$. Therefore no rational number squares to $2$. Now the sequence $a_n = \lfloor 10^n \sqrt{2} \rfloor / 10^n$ is a sequence in $\mathbb{Q}$ satisfying $|a_n - a_m| \le 10^{-\min(n,m)}$ for all $n, m \in \mathbb{N}$, so it is Cauchy in $\mathbb{Q}$. But $|a_n - \sqrt{2}| \le 10^{-n}$, and since $\sqrt{2} \notin \mathbb{Q}$, there is no $L \in \mathbb{Q}$ with $a_n \to L$ in the usual sense. The Cauchy property and convergence come apart in $\mathbb{Q}$. [/example] This failure has a name: $\mathbb{Q}$ is *not complete* as a [metric space](/page/Metric%20Space). The solution — patching every hole — is the Cauchy completion of $\mathbb{Q}$, and the result is $\mathbb{R}$. ## The Number System Tower Before constructing $\mathbb{R}$, it is worth placing it in context. Mathematics builds number systems by extending previous ones to solve problems the old system cannot handle. We start from $\mathbb{N} = \{1, 2, 3, \ldots\}$ (natural numbers), which supports addition and multiplication but not subtraction: $1 - 2 \notin \mathbb{N}$. Adjoining negatives and zero produces $\mathbb{Z}$, the integers. The integers support subtraction but not division: $1/3 \notin \mathbb{Z}$. The fraction field construction, forming equivalence classes of pairs $(p, q)$ with $q \neq 0$, produces $\mathbb{Q}$, the rationals. The rationals form a field, and they are ordered: for any two rationals $p$ and $q$, exactly one of $p < q$, $p = q$, $p > q$ holds, and the order is compatible with addition and multiplication. The rationals are dense in themselves: between any two rationals there is another. Yet they have holes, as the example above shows. Passing to the Cauchy completion produces $\mathbb{R}$, which closes all remaining gaps. [remark: The Tower Perspective] The extension $\mathbb{N} \to \mathbb{Z} \to \mathbb{Q}$ each time solves a specific algebraic failure. The extension $\mathbb{Q} \to \mathbb{R}$ solves a *topological* (or metric) failure: the absence of limits for Cauchy sequences. There is no algebraic equation that forces $\mathbb{R}$; rather, analysis forces it. [/remark] ## Definition ### Why Cauchy Sequences? To construct $\mathbb{R}$, we need to name the holes in $\mathbb{Q}$. A hole is a place "toward which" rational sequences are heading without arriving. What distinguishes such sequences from divergent ones? The Cauchy condition: the terms get close to *each other*, regardless of whether they converge to anything in $\mathbb{Q}$. [definition: Cauchy Sequence in $\mathbb{Q}$] A sequence $(a_n)_{n \in \mathbb{N}}$ with $a_n \in \mathbb{Q}$ for all $n$ is a **[Cauchy sequence](/page/Cauchy%20Sequence)** if for every rational $\varepsilon > 0$ there exists $N \in \mathbb{N}$ such that \begin{align*} |a_n - a_m| < \varepsilon \quad \text{for all } n, m \ge N. \end{align*} [/definition] Every convergent sequence in $\mathbb{Q}$ is Cauchy: if $a_n \to L$ then $|a_n - a_m| \le |a_n - L| + |L - a_m| < 2\varepsilon$ for large $n, m$. The converse fails in $\mathbb{Q}$, and that failure is precisely what we wish to repair. The idea is simple: we *declare* the limit to exist by treating the Cauchy sequence itself as the limiting object. Two Cauchy sequences that are approaching the same "hole" should be identified — they represent the same missing number. This leads to an [equivalence relation](/page/Equivalence%20Relation). [definition: Equivalence of Cauchy Sequences] Two Cauchy sequences $(a_n)$ and $(b_n)$ in $\mathbb{Q}$ are **equivalent**, written $(a_n) \sim (b_n)$, if \begin{align*} \lim_{n \to \infty} |a_n - b_n| = 0. \end{align*} [/definition] One checks that $\sim$ is indeed an equivalence relation: reflexivity is immediate; symmetry is immediate; transitivity follows from the triangle inequality $|a_n - c_n| \le |a_n - b_n| + |b_n - c_n| \to 0$. [definition: Real Numbers] The set of **real numbers** $\mathbb{R}$ is the set of equivalence classes of Cauchy sequences of rational numbers: \begin{align*} \mathbb{R} := \{ [(a_n)] : (a_n) \text{ is a Cauchy sequence in } \mathbb{Q} \}, \end{align*} where $[(a_n)]$ denotes the equivalence class of $(a_n)$ under $\sim$. [/definition] Every rational number $q \in \mathbb{Q}$ embeds into $\mathbb{R}$ via the constant sequence: $q \mapsto [(q, q, q, \ldots)]$. This embedding preserves addition, multiplication, and order, so $\mathbb{Q}$ sits inside $\mathbb{R}$ as a subfield. The non-rational reals are the classes of Cauchy sequences that do not converge in $\mathbb{Q}$. ### Arithmetic on $\mathbb{R}$ With the real numbers defined as equivalence classes, the arithmetic operations must be defined on representatives and shown to be well-defined. [definition: Addition and Multiplication on $\mathbb{R}$] Let $x = [(a_n)]$ and $y = [(b_n)]$ be real numbers. Define: \begin{align*} x + y &:= [(a_n + b_n)], \\ x \cdot y &:= [(a_n \cdot b_n)]. \end{align*} [/definition] Before using these operations, we must verify that they do not depend on the choice of representative sequences — that replacing $(a_n)$ by an equivalent sequence $(a_n')$ leaves the sum and product unchanged. [quotetheorem:3225] For addition: $|(a_n + b_n) - (a_n' + b_n')| \le |a_n - a_n'| + |b_n - b_n'| \to 0$ as $n \to \infty$, since both differences tend to $0$ by hypothesis. For multiplication: write $a_n b_n - a_n' b_n' = a_n(b_n - b_n') + b_n'(a_n - a_n')$. Cauchy sequences in $\mathbb{Q}$ are bounded — if $(a_n)$ is Cauchy then $|a_n| \le M$ for some $M > 0$, and similarly $|b_n'| \le M'$ — so $|a_n b_n - a_n' b_n'| \le M|b_n - b_n'| + M'|a_n - a_n'| \to 0$. [definition: Order on $\mathbb{R}$] We say $x = [(a_n)] > 0$ if there exists $\varepsilon \in \mathbb{Q}$ with $\varepsilon > 0$ and $N \in \mathbb{N}$ such that $a_n > \varepsilon$ for all $n \ge N$. For general $x, y \in \mathbb{R}$, declare $x < y$ iff $y - x > 0$. [/definition] One must check that this order is well-defined and makes $\mathbb{R}$ into a **totally ordered field**: the order is compatible with addition ($x < y \implies x + z < y + z$ for all $z$) and with multiplication ($x > 0$ and $y > 0$ implies $xy > 0$). For the order axioms, note that if $(a_n)$ and $(a_n')$ are equivalent representatives and $a_n > \varepsilon$ eventually, then $|a_n - a_n'| < \varepsilon/2$ eventually, forcing $a_n' > \varepsilon/2$ eventually, so positivity is independent of the representative. The compatibility with addition follows directly from the same inequality applied to $(a_n + c_n) - (b_n + c_n) = a_n - b_n$; and multiplication compatibility reduces to showing that if $a_n > \varepsilon$ and $b_n > \delta$ eventually, then $a_n b_n > \varepsilon\delta$ eventually. <!-- illustration-needed: the number line ℝ with ℚ dense inside it — show a segment of the line with rational points highlighted densely, and the "holes" (like √2, π) filled in by the completion --> ## Completeness The central property of $\mathbb{R}$ — the reason the entire construction was undertaken — is that Cauchy sequences now converge. [quotetheorem:3226] [explanation: Why This Is Not Circular] It might seem circular to say "Cauchy sequences in $\mathbb{R}$ converge" when $\mathbb{R}$ was built from Cauchy sequences. The subtlety is that the Cauchy sequences used in the construction are sequences in $\mathbb{Q}$, whereas the completeness theorem is about Cauchy sequences in $\mathbb{R}$. A Cauchy sequence in $\mathbb{R}$ is a sequence of equivalence classes of rational Cauchy sequences — each term $x_n$ is itself an equivalence class $[(r_{n,k})_{k \in \mathbb{N}}]$ of rational numbers, where we use a double subscript $(n, k)$ to keep the two indices separate. Showing that the sequence $(x_n)$ converges requires a diagonal argument: for each $n$, choose a rational representative $r_{n,k}$ and pick a specific rational $q_n$ close to $x_n$ (say, within $1/n$). The resulting sequence $(q_n)$ of rationals is Cauchy in $\mathbb{Q}$, and its equivalence class $[(q_n)] \in \mathbb{R}$ is the limit of $(x_n)$. [/explanation] Completeness can be equivalently stated in terms of upper bounds: [quotetheorem:3227] [definition: Supremum and Infimum] Let $S \subset \mathbb{R}$ be non-empty. An element $M \in \mathbb{R}$ is the **supremum** of $S$, written $M = \sup S$, if: 1. $M$ is an upper bound for $S$: $s \le M$ for all $s \in S$. 2. $M$ is the least such bound: if $M'$ is an upper bound for $S$, then $M \le M'$. Equivalently, $M = \sup S$ iff $M$ is an upper bound and for every $\varepsilon > 0$ there exists $s \in S$ with $s > M - \varepsilon$. The **infimum** $\inf S$ is defined dually as the greatest lower bound. [/definition] The equivalence of the Cauchy criterion and the least upper bound property is one of the foundational theorems of real analysis. Neither is more "primary" — each implies the other, given the ordered field axioms. [explanation: Connecting Cauchy Completeness to Least Upper Bounds] Given Cauchy completeness, we can prove the least upper bound property: let $S \subset \mathbb{R}$ be non-empty and bounded above by some $M_0$. Pick any $s_0 \in S$. Define a sequence of intervals $[l_n, u_n]$ by bisection: if $(l_n + u_n)/2$ is an upper bound for $S$, set $u_{n+1} = (l_n + u_n)/2$ and $l_{n+1} = l_n$; otherwise set $l_{n+1} = (l_n + u_n)/2$ and $u_{n+1} = u_n$. The midpoints form a Cauchy sequence (the interval lengths halve each step), so they converge by Cauchy completeness, and the limit is $\sup S$. Conversely, given the least upper bound property, a Cauchy sequence $(x_n)$ is bounded, and $L = \inf_{N} \sup_{n \ge N} x_n$ exists by the LUB property; one verifies that $x_n \to L$. [/explanation] [example: Supremum of an Irrational Set] Let $S = \{x \in \mathbb{Q} : x^2 < 2\}$. This set is non-empty ($1 \in S$) and bounded above in $\mathbb{R}$ (e.g., $2$ is an upper bound). The least upper bound property guarantees $\sup S$ exists in $\mathbb{R}$. We claim $\sup S = \sqrt{2}$. First, $\sqrt{2}$ is an upper bound: if $x \in S$ then $x^2 < 2$, so $x < \sqrt{2}$ (since $x$ is rational and hence not equal to $\sqrt{2}$, and $x^2 < 2 = (\sqrt{2})^2$ with $x, \sqrt{2} > 0$). Second, $\sqrt{2}$ is the least upper bound: for any $\varepsilon > 0$, we must find $x \in S$ with $x > \sqrt{2} - \varepsilon$. By the density of $\mathbb{Q}$ in $\mathbb{R}$ (established below), there exists a rational $q$ with $\sqrt{2} - \varepsilon < q < \sqrt{2}$. Since $q < \sqrt{2}$ we have $q^2 < 2$, so $q \in S$, and $q > \sqrt{2} - \varepsilon$. Thus $\sup S = \sqrt{2} \in \mathbb{R} \setminus \mathbb{Q}$. In $\mathbb{Q}$, the same set $S \cap \mathbb{Q} = S$ has no supremum in $\mathbb{Q}$: any rational upper bound $r$ satisfies $r^2 \ge 2$, and one can find a rational strictly between $\sqrt{2}$ and $r$ that is still an upper bound for $S$, contradicting $r$ being the least. [/example] ## The Archimedean Property and Density of $\mathbb{Q}$ Having built $\mathbb{R}$ and established completeness, we can derive two fundamental properties that describe the fine structure of the real line. The first might be called the "no infinitely large" property: no real number is larger than every natural number. This seems obvious, but it is a *consequence* of completeness — it fails in other complete ordered fields that contain infinitely large elements (non-Archimedean fields). [quotetheorem:737] [explanation: Proof via the Least Upper Bound Property] Suppose for contradiction that $x \ge n$ for all $n \in \mathbb{N}$. Then $\mathbb{N}$ is a non-empty subset of $\mathbb{R}$ bounded above by $x$. By the least upper bound property, $M = \sup \mathbb{N}$ exists in $\mathbb{R}$. Since $M - 1 < M = \sup \mathbb{N}$, the element $M - 1$ is not an upper bound for $\mathbb{N}$, so there exists $n_0 \in \mathbb{N}$ with $n_0 > M - 1$. But then $n_0 + 1 \in \mathbb{N}$ satisfies $n_0 + 1 > M$, contradicting $M$ being an upper bound. This contradiction shows $\mathbb{N}$ is unbounded above, so no such $x$ exists. [/explanation] [remark: Equivalent Formulations] The [Archimedean property](/page/Archimedean%20Property) is equivalent to: (i) for every $\varepsilon > 0$, there exists $n \in \mathbb{N}$ with $1/n < \varepsilon$; (ii) for every $x > 0$ and $y > 0$, there exists $n \in \mathbb{N}$ with $nx > y$. These equivalences follow immediately from the main statement. [/remark] The second structural property says that $\mathbb{Q}$ is "spread evenly" throughout $\mathbb{R}$: no real number has a neighbourhood containing no rationals. [quotetheorem:740] [explanation: Proof via the Archimedean Property] Since $b - a > 0$, by the [Archimedean property](/theorems/737) there exists $n \in \mathbb{N}$ with $n > 1/(b - a)$, i.e., $n(b - a) > 1$. By the Archimedean property applied to $na$, there exists $m \in \mathbb{Z}$ with $m - 1 \le na < m$ (take $m = \lfloor na \rfloor + 1$, choosing the smallest integer strictly greater than $na$). Then $q = m/n$ satisfies: \begin{align*} q = \frac{m}{n} > \frac{na}{n} = a, \end{align*} and since $m \le na + 1 < na + n(b - a) = nb$, \begin{align*} q = \frac{m}{n} < b. \end{align*} So $q \in \mathbb{Q}$ with $a < q < b$. [/explanation] Density of $\mathbb{Q}$ in $\mathbb{R}$ means that every real number can be approximated arbitrarily well by rationals. This is the sense in which the rational approximations $1, 1.4, 1.41, 1.414, \ldots$ "capture" $\sqrt{2}$: every open interval around $\sqrt{2}$ contains a rational. It is also the key tool used in the supremum example above — the existence of a rational between any two reals is what allows us to exhibit elements of $S$ close to $\sqrt{2}$. ## Uniqueness: Any Complete Ordered Field is $\mathbb{R}$ We have *constructed* $\mathbb{R}$. But we should ask: is this construction canonical, or could a different method produce a different complete ordered field? The answer is unambiguous: up to a canonical isomorphism, there is only one complete ordered field. [quotetheorem:3228] The proof sketch: given a complete ordered field $F$, the element $1_F$ is the multiplicative identity, $1_F + 1_F$ maps to $2$, and so on; this defines an embedding $\mathbb{N} \hookrightarrow F$, which extends to $\mathbb{Z}$ and then to $\mathbb{Q}$. The embedding $\mathbb{Q} \hookrightarrow F$ is unique (there is only one ordered field homomorphism from $\mathbb{Q}$ to any ordered field). Since $F$ is complete, every Cauchy sequence in $\mathbb{Q}$ has a limit in $F$, and by density of $\mathbb{Q}$ in $F$ (which follows from the Archimedean property of $F$) every element of $F$ is such a limit. So the map $\mathbb{R} \to F$ defined by sending $[(a_n)]$ to the limit of $(a_n)$ in $F$ is a well-defined bijective ordered field homomorphism. [remark: What Uniqueness Means Practically] The uniqueness theorem says we are not making an arbitrary choice when we "define $\mathbb{R}$." Whether we use Cauchy sequences, Dedekind cuts, or any other construction, we will always get the same mathematical object. The particular construction is scaffolding; $\mathbb{R}$ is determined by the axioms. The identification of any specific complete ordered field with $\mathbb{R}$ is canonical: any two such fields come with a unique order-preserving field isomorphism between them, so speaking of "the" real numbers is justified. [/remark] ## Constructing Key Reals as Cauchy Sequences The abstract definition of $\mathbb{R}$ becomes concrete when we identify specific real numbers as Cauchy sequences. [example: $\sqrt{2}$ as a Cauchy Sequence] The number $\sqrt{2}$ is the equivalence class of any Cauchy sequence of rationals whose squares approach $2$. One explicit sequence: let $a_1 = 1$ and \begin{align*} a_{n+1} = \frac{a_n}{2} + \frac{1}{a_n} \end{align*} (the Newton–Raphson iteration for $\sqrt{2}$). Each $a_n$ is rational (since $a_1 \in \mathbb{Q}$ and the recurrence uses only field operations). We show $(a_n)$ is Cauchy and that its square approaches $2$. Note that $a_n^2 - 2 = \left(\frac{a_{n-1}}{2} + \frac{1}{a_{n-1}}\right)^2 - 2 = \frac{(a_{n-1}^2 - 2)^2}{4a_{n-1}^2}$. To bound this, we need $a_{n-1} \ge 1$ for $n \ge 2$. We verify this inductively: $a_1 = 1$, and if $a_{n-1} \ge 1$ then $a_n = a_{n-1}/2 + 1/a_{n-1} \ge 1$ by AM-GM (since $a_{n-1}/2 + 1/a_{n-1} \ge 2\sqrt{(a_{n-1}/2)(1/a_{n-1})} = \sqrt{2} \ge 1$). So $a_n \ge 1$ for all $n \ge 1$. Since $a_1 = 1$ gives $a_1^2 - 2 = -1$, we get $a_2 = 3/2$ and $a_2^2 - 2 = 1/4$. The error $|a_n^2 - 2|$ satisfies \begin{align*} |a_n^2 - 2| = \frac{(a_{n-1}^2 - 2)^2}{4a_{n-1}^2} \le \frac{1}{4}(a_{n-1}^2 - 2)^2 \quad (\text{since } a_{n-1} \ge 1). \end{align*} Starting from $|a_2^2 - 2| = 1/4$, the error decreases doubly exponentially: $|a_n^2 - 2| \le (1/4)^{2^{n-2}}$ for $n \ge 2$. In particular $|a_n^2 - 2| \to 0$, meaning $a_n^2 \to 2$ in $\mathbb{Q}$. Since all $a_n \ge 1$, we have $|a_n - a_m| = |a_n^2 - a_m^2|/(a_n + a_m) \to 0$, so $(a_n)$ is Cauchy in $\mathbb{Q}$. Its equivalence class $[(a_n)] \in \mathbb{R}$ satisfies $[(a_n)]^2 = [(a_n^2)] = [(2, 2, 2, \ldots)] = 2$, so $[(a_n)] = \sqrt{2}$. [/example] [example: Euler's Number $e$ as a Cauchy Sequence] Define the sequence of partial sums \begin{align*} s_n = \sum_{k=0}^{n} \frac{1}{k!} = 1 + 1 + \frac{1}{2} + \frac{1}{6} + \cdots + \frac{1}{n!}. \end{align*} Each $s_n \in \mathbb{Q}$ (finite sum of rationals). We show $(s_n)$ is Cauchy. For $m > n$, \begin{align*} |s_m - s_n| = \sum_{k=n+1}^{m} \frac{1}{k!} \le \sum_{k=n+1}^{\infty} \frac{1}{k!}. \end{align*} For $k \ge n + 1$, note $k! \ge (n+1)! \cdot (n+2)^{k-(n+1)}$, so \begin{align*} \sum_{k=n+1}^{\infty} \frac{1}{k!} \le \frac{1}{(n+1)!} \sum_{j=0}^{\infty} \frac{1}{(n+2)^j} = \frac{1}{(n+1)!} \cdot \frac{n+2}{n+1} \le \frac{2}{(n+1)!}. \end{align*} Since $(n+1)! \to \infty$, the right side tends to $0$, so for any $\varepsilon > 0$ we can choose $N$ such that $2/(n+1)! < \varepsilon$ for all $n \ge N$. Thus $|s_m - s_n| < \varepsilon$ for all $m > n \ge N$, confirming $(s_n)$ is Cauchy in $\mathbb{Q}$. Euler's number is defined as $e = [(s_n)] \in \mathbb{R}$. It is transcendental (not a root of any polynomial with rational coefficients), though proving this requires more work. [/example] [example: Non-example — $\mathbb{Q}$ is Not Complete] The sequence $a_n = \sum_{k=1}^n 1/k$ (partial sums of the harmonic series) is a sequence in $\mathbb{Q}$ that is *not* Cauchy: for any $N$, choosing $m = 2n$ with $n \ge N$ gives \begin{align*} a_{2n} - a_n = \sum_{k=n+1}^{2n} \frac{1}{k} \ge \sum_{k=n+1}^{2n} \frac{1}{2n} = \frac{n}{2n} = \frac{1}{2}. \end{align*} So $|a_{2n} - a_n| \ge 1/2$ for all $n$, and the sequence diverges to $+\infty$ in $\mathbb{R}$. A subtler non-example: the set of computable real numbers $\mathbb{R}_{\mathrm{comp}}$ consists of those real numbers whose decimal expansions can be generated by a Turing machine. While $\mathbb{Q} \subset \mathbb{R}_{\mathrm{comp}}$ and $\mathbb{R}_{\mathrm{comp}}$ contains $\sqrt{2}$, $e$, $\pi$, and all algebraic numbers, it is a [countable set](/page/Countable%20Set). Since $\mathbb{R}$ is uncountable (by Cantor's diagonal argument), almost every real number is *not* computable. The computable reals are dense in $\mathbb{R}$ but do not form a complete ordered field: there exist Cauchy sequences of computable reals whose limit is non-computable. [/example] ## The Dedekind Cut Construction The Cauchy-sequence construction we have followed is the preferred one for analysis and for generalization: the same procedure — forming equivalence classes of Cauchy sequences — yields the completion of *any* [metric space](/page/Metric%20Space), as studied on the page [Complete Metric Spaces](/page/Complete%20Metric%20Space). The same idea underlies the construction of [Polish spaces](/page/Polish%20Space) and the $p$-adic numbers $\mathbb{Q}_p$. There is an alternative construction, due to Dedekind, which is historically important and conceptually elegant: [definition: Dedekind Cut] A **Dedekind cut** is a pair $(L, R)$ of non-empty disjoint sets with $L \cup R = \mathbb{Q}$ satisfying: 1. If $p \in L$ and $q < p$, then $q \in L$ (L is closed downward). 2. $L$ has no greatest element. 3. $R$ is determined by $L$: $R = \mathbb{Q} \setminus L$. The real number corresponding to the cut $(L, R)$ is thought of as the unique "point" separating $L$ from $R$. [/definition] For example, $\sqrt{2}$ corresponds to the cut $L = \{q \in \mathbb{Q} : q \le 0 \text{ or } q^2 < 2\}$. The two constructions produce isomorphic ordered fields, as the uniqueness theorem guarantees. [remark: Dedekind vs. Cauchy] Dedekind cuts make the ordered structure of $\mathbb{R}$ transparent — the order is literally set inclusion. Cauchy sequences make the metric and topological structure transparent — convergence is built in from the start. For analysis, the Cauchy construction generalizes better. For foundational arithmetic, Dedekind cuts have certain elegance. The choice is one of emphasis, not correctness. [/remark] ## Topology and Uncountability The real line carries a natural metric, and it is worth pausing to say why the absolute value is the right choice. Any metric on $\mathbb{R}$ must interact well with the algebraic structure: distances should be translation-invariant ($d(x + z, y + z) = d(x, y)$ for all $z$) and scale-compatible ($d(cx, cy) = |c| \cdot d(x, y)$). The function $d(x, y) = |x - y|$ is the unique metric with both properties that assigns distance $1$ to the unit interval. It is also exactly the metric inherited from the absolute value on $\mathbb{Q}$ used to define Cauchy sequences, making the construction of $\mathbb{R}$ self-consistent. [definition: Standard Metric on $\mathbb{R}$] The **standard metric** on $\mathbb{R}$ is the function \begin{align*} d: \mathbb{R} \times \mathbb{R} &\to [0, \infty) \\ (x, y) &\mapsto |x - y|. \end{align*} [/definition] With the metric in hand, we want to describe which subsets of $\mathbb{R}$ are "open" — those around which every point has some room to move. The open intervals are the natural building blocks: every point in $(a, b)$ is strictly between $a$ and $b$, so it has room on both sides. Taking arbitrary unions of open intervals generates all open sets, giving $\mathbb{R}$ a [topology](/page/Topology) that encodes [continuity](/page/Continuity), convergence, and compactness in a uniform language. [definition: Open Intervals and Standard Topology on $\mathbb{R}$] An **open interval** in $\mathbb{R}$ is a set of the form $(a, b) = \{x \in \mathbb{R} : a < x < b\}$ for $a < b$ in $\mathbb{R}$ (or the unbounded variants $(-\infty, b)$, $(a, \infty)$, $\mathbb{R}$ itself). The **standard topology** $\tau_{\mathrm{std}}$ on $\mathbb{R}$ is the topology generated by the open intervals: $U \subset \mathbb{R}$ is open iff every point of $U$ is contained in an open interval contained in $U$. [/definition] Under this topology, $\mathbb{R}$ is a [complete metric space](/page/Complete%20Metric%20Space) (by the completeness theorem), [separable](/page/Separable) (the rationals form a countable [dense subset](/page/Dense%20Subset)), and metrizable. It is also connected — cannot be written as a disjoint union of two non-empty open sets — and locally compact. [quotetheorem:758] [explanation: Cantor's Diagonal Argument] Suppose for contradiction that $[0, 1]$ is countable and list its elements as $x_1, x_2, x_3, \ldots$. Write each $x_n$ in its decimal expansion: $x_n = 0.d_{n,1}d_{n,2}d_{n,3}\ldots$. Define a new number $y = 0.e_1 e_2 e_3 \ldots$ where $e_n = 5$ if $d_{n,n} \neq 5$ and $e_n = 6$ if $d_{n,n} = 5$. Then $y \in [0, 1]$, but for each $n$, $y$ differs from $x_n$ in the $n$-th decimal place, so $y \neq x_n$. This contradicts the assumption that $(x_n)$ lists all elements of $[0, 1]$. [/explanation] This uncountability is in sharp contrast with $\mathbb{Q}$, which is countable. The "extra" elements of $\mathbb{R}$ — those not in $\mathbb{Q}$ — are the irrational numbers, and they form an uncountable dense subset of $\mathbb{R}$. ## Connection to Metric Completion and Polish Spaces The construction of $\mathbb{R}$ from $\mathbb{Q}$ is the prototype for a general operation in analysis. [definition: Completion of a Metric Space] Let $(X, d)$ be a metric space. Its **completion** $(\hat{X}, \hat{d})$ is a [complete metric space](/page/Complete%20Metric%20Space) together with an isometric embedding $\iota: X \to \hat{X}$ such that $\iota(X)$ is dense in $\hat{X}$, and such that every isometric embedding of $X$ into a complete metric space factors through $\iota$ uniquely. [/definition] The completion always exists and is unique up to isometric isomorphism. The construction mirrors exactly what we did for $\mathbb{R}$: form equivalence classes of Cauchy sequences in $X$ under the equivalence $(x_n) \sim (y_n)$ iff $d(x_n, y_n) \to 0$, define $\hat{d}([(x_n)], [(y_n)]) = \lim_n d(x_n, y_n)$, and embed $X$ via constant sequences. Taking $X = \mathbb{Q}$ with the metric $d(p, q) = |p - q|$ recovers the construction of $\mathbb{R}$. But the same procedure applies everywhere: the completion of the rationals under the $p$-adic metric $|p - q|_p$ gives $\mathbb{Q}_p$, the $p$-adic numbers; the completion of the space of continuous functions under the $L^2$ norm gives $L^2$; and so on. A [complete metric space](/page/Complete%20Metric%20Space) that is also separable is called a **[Polish space](/page/Polish%20Space)**. The real line $\mathbb{R}$ is a Polish space: complete (by the completeness theorem) and separable (since $\mathbb{Q}$ is countable and dense). Polish spaces are the natural setting for descriptive set theory and for many parts of probability theory. [remark: Generalisation Roadmap] $\mathbb{R}$ is the simplest Polish space. The theory of Polish spaces abstracts the key properties — completeness and separability — that make analysis on $\mathbb{R}$ work. If you understand $\mathbb{R}$ through the Cauchy-completion lens, you already understand the mechanism behind the completion of $L^p$ spaces, Sobolev spaces, and abstract metric spaces. The general construction appears on the [Complete Metric Space](/page/Complete%20Metric%20Space) and [Polish Space](/page/Polish%20Space) pages. [/remark] ## References W. Rudin, *Principles of Mathematical Analysis*, 3rd ed. (1976). McGraw-Hill. Chapter 1 constructs $\mathbb{R}$ via Dedekind cuts; Chapter 3 develops Cauchy sequences and the completeness criterion. T. Tao, *Analysis I*, 3rd ed. (2016). Springer. Chapters 5–6 give the Cauchy-sequence construction with full foundational detail. E. Landau, *Foundations of Analysis* (1930). Chelsea Publishing. The classical axiomatic development from Peano axioms through $\mathbb{R}$. G. Cantor, "Über die Ausdehnung eines Satzes aus der Theorie der trigonometrischen Reihen" (1872). The original Cauchy-completion construction of $\mathbb{R}$. R. Dedekind, *Stetigkeit und irrationale Zahlen* (1872). The Dedekind-cut construction.