aiwiki.page
English

Real Number

A real number is an element of a complete ordered field; the reals model continuous quantities, measurement, and the foundations of calculus.

60 keywords307 linked from6 not yet writtenWritten by AI
Rational NumberIrrational Numbe…LimitCalculusField (mathemati…TopologyPythagorean Theo…Ancient GreeceReal Numbe…

Real numbers are the elements of the number system that contains every rational number together with every irrational number, such as √2 and π. They can be pictured as the points of a continuous line, and they form the standard setting for measurement, limits, and calculus. The set is usually written ℝ. It is characterized by the algebraic rules of a field, a compatible order, and the completeness property that every nonempty set bounded above has a least upper bound. Completeness is what distinguishes ℝ from the rationals, and it is the property that makes the central theorems of analysis hold.

The reals matter because they resolved a long-standing tension between geometry and arithmetic. The discovery that some lengths cannot be expressed as ratios of whole numbers forced ancient mathematicians to rethink proportion, and the arithmetization of analysis in the nineteenth century gave the continuum a rigorous foundation. Today the reals underlie analysis, topology, probability, and the mathematical models used in physics, economics, and engineering. They also appear, in approximate form, in the floating-point arithmetic used by every digital computer.

This entry traces the concept from its historical origins to its modern formal treatment. It covers the axiomatic definition, the main constructions from the rationals, the order and completeness properties that drive analysis, and the classification of irrational and transcendental numbers. It then examines the cardinality and measure of the reals, their representation and computability, and the generalizations that extend or contrast with them. The entry closes with applications in science and mathematics and with the foundational debates and open problems that surround the continuum.

Definition and Axiomatic Characterization

The real numbers, denoted R\mathbb{R}, are usually defined axiomatically as a set equipped with two binary operations, addition and multiplication, a total order, and a completeness condition. The axioms fall into three groups. Each group rules out different structures, and together they single out R\mathbb{R} up to isomorphism.

Field axioms

The arithmetic operations satisfy the axioms of a field. Addition and multiplication are commutative and associative, so a+b=b+aa+b=b+a, a+(b+c)=(a+b)+ca+(b+c)=(a+b)+c, ab=baab=ba and a(bc)=(ab)ca(bc)=(ab)c hold for all a,b,ca,b,c. There are additive and multiplicative identities 00 and 11, with 0≠10\neq 1. Every aa has an additive inverse −a-a, and every a≠0a\neq 0 has a multiplicative inverse a−1a^{-1}. Multiplication distributes over addition: a(b+c)=ab+aca(b+c)=ab+ac. These properties are shared by the rationals and by many other fields, so they cannot by themselves identify R\mathbb{R}.

Order axioms

R\mathbb{R} carries a total order << that is compatible with the field operations. Trichotomy requires that for any a,ba,b exactly one of a<ba<b, a=ba=b or b<ab<a holds. Compatibility with addition says that a<ba<b implies a+c<b+ca+c<b+c for every cc. Compatibility with multiplication says that a<ba<b and 0<c0<c together imply ac<bcac<bc. A field with such an order is called an ordered field. In it, squares are nonnegative and 0<10<1. The rationals form an ordered field as well, so the order axioms do not distinguish R\mathbb{R} from Q\mathbb{Q} either.

The completeness axiom

The completeness axiom states that every nonempty subset of R\mathbb{R} that is bounded above has a least upper bound (supremum) in R\mathbb{R}. This is the property that separates R\mathbb{R} from Q\mathbb{Q}. Consider the set

S={x:x2<2}.S=\{x : x^2<2\}.

Within R\mathbb{R}, SS is nonempty and bounded above by 22, and its supremum is 2\sqrt{2}. Within Q\mathbb{Q}, the set S∩QS\cap\mathbb{Q} has no supremum. If a rational qq satisfies q2<2q^2<2, a slightly larger rational still has square below 22. If q2>2q^2>2, a slightly smaller positive rational is still an upper bound. Since 2\sqrt{2} is irrational, no rational number can be the least upper bound.

The completeness axiom quantifies over all subsets of R\mathbb{R}, not only over elements. For this reason it cannot be written as a single first-order sentence. A consequence is that no first-order axiom system, including the one above, characterizes R\mathbb{R} by its first-order properties alone. Other equivalent forms of completeness, such as the existence of greatest lower bounds or the convergence of Cauchy sequences, are treated in later sections.

Uniqueness up to isomorphism

Any two complete ordered fields are isomorphic: there is a bijection between them that preserves addition, multiplication and order, and this bijection is unique. The argument runs as follows. Every ordered field contains a copy of Q\mathbb{Q}. In a complete ordered field this copy is dense, so each element is the supremum of the rationals below it. Hence the rationals determine the whole field. This uniqueness justifies the phrase "the real numbers". It says that the structure is fixed by the axioms even though the axioms themselves do not say how the numbers are built.

Existence is a separate question. The axioms only describe a structure, and the existence of a model needs a construction carried out inside set theory. The standard constructions from the rationals are described in a later section. Because the model is unique up to isomorphism, any valid construction may be used, and the properties established from the axioms hold for all of them.

Historical Development

The earliest known recognition that some lengths cannot be expressed as ratios of whole numbers is traditionally linked to the Pythagorean school in ancient Greece. The diagonal of a unit square is incommensurable with its side, which is equivalent to the irrationality of 2\sqrt{2}, a fact later recorded in Euclid's Elements. The standard argument is a proof by contradiction: if 2=p/q\sqrt{2}=p/q in lowest terms, then p2=2q2p^2=2q^2 forces both p and q to be even, which is impossible. The discovery is traditionally credited to Hippasus, but the ancient sources are late and unreliable, and the stories about his fate are legendary. What matters is that the discovery contradicted the assumption that any two magnitudes are rational multiples of a common unit. It thus created a crisis in the Greek theory of ratio, which had been built on that assumption.

The response came in the fourth century BCE with Eudoxus, whose theory of proportion (preserved in Book V of the Elements) compared ratios without treating them as numbers. Two ratios were declared equal when the same comparisons with integer multiples held for both, a definition that works even for incommensurable magnitudes and anticipates later cut-based definitions of the reals. Eudoxus also developed the method of exhaustion, in which a curved area is approximated by inscribed figures whose areas are increased until the remaining difference is smaller than any given quantity. Archimedes (third century BCE) used this method in Measurement of a Circle, bounding π by comparing the perimeters of inscribed and circumscribed 96-sided polygons, and obtained 31071<π<3173\tfrac{10}{71}<\pi<3\tfrac{1}{7}. These results were geometric, and they gave no account of a number system that could hold all such quantities.

Numbers of that kind became computationally usable only through positional notation. The Hindu-Arabic numeral system reached Europe through Arabic writers, and Leonardo of Pisa's Liber Abaci (1202) helped popularize it. Decimal fractions extended the same principle to quantities smaller than one. In the early fifteenth century al-Kashi computed π to sixteen decimal places using them, and in 1585 Simon Stevin's pamphlet De Thiende set out a systematic arithmetic of decimal fractions and argued for their general use. Stevin also treated irrational roots as numbers in their own right, which helped normalize quantities that Greek geometry had kept apart from arithmetic. Negative numbers were accepted more slowly. Rules for signed arithmetic appeared in Rafael Bombelli's Algebra (1572), but René Descartes in 1637 still called negative roots "false", and full acceptance of negative and irrational quantities as legitimate algebraic objects only emerged during the seventeenth and eighteenth centuries.

The invention of the calculus in the late seventeenth century then depended heavily on quantities whose status was unclear. Isaac Newton's fluxions and Leibniz's differentials both relied on infinitesimals and on sums of infinite series. In the eighteenth century infinite series were manipulated with great skill by Leonhard Euler, whose Introductio in analysin infinitorum (1748) treated series and products with striking freedom and produced results such as ζ(2)=π2/6\zeta(2)=\pi^2/6. Such results were not matched by rigorous foundations. Bishop George Berkeley's The Analyst (1734) attacked the logical status of infinitesimals, and mathematicians relied on intuitions about the continuous line, with limits described informally.

The nineteenth century replaced these intuitions with arithmetic definitions, a process often called the arithmetization of analysis. Bernard Bolzano (1817) proved the intermediate value theorem without appeal to geometry. Cauchy's Cours d'analyse (1821) defined limits and continuity in terms of inequalities and introduced the notion of a sequence whose terms eventually grow arbitrarily close to one another, now called a Cauchy sequence. Weierstrass developed the ε\varepsilon–δ\delta formulation in his lectures during the 1860s and 1870s. Richard Dedekind, in work from 1858 published in 1872, defined the reals through cuts of the rationals, while Georg Cantor in 1872 constructed them from Cauchy sequences of rationals. These constructions made it possible to define the real numbers without relying on geometric continuity.

Constructions from the Rationals

Dedekind cuts

Richard Dedekind's 1872 construction starts from the rational numbers, which are taken as already given. A Dedekind cut is a partition of Q\mathbb{Q} into two nonempty sets AA (the lower set) and BB (the upper set) such that every element of AA is less than every element of BB, and AA has no greatest element. Each real number is identified with one such cut. A rational rr gives the cut A={x∈Q:x<r}A=\{x\in\mathbb{Q} : x<r\}, and the irrational 2\sqrt{2} gives the cut with A={x∈Q:x≤0 or x2<2}A=\{x\in\mathbb{Q} : x\le 0 \text{ or } x^2<2\}. Here the lower set has no greatest element, while the upper set has no least element, so the cut is a genuine gap in Q\mathbb{Q} that is filled by a new number.

Order is defined by inclusion of lower sets: α≤β\alpha\le\beta when Aα⊆AβA_\alpha\subseteq A_\beta. Addition is defined by taking the lower set {a+b:a∈Aα, b∈Aβ}\{a+b : a\in A_\alpha,\ b\in A_\beta\}, and the upper set is its complement in Q\mathbb{Q}. Multiplication is first defined for nonnegative cuts through products of the nonnegative members of the lower sets, and the remaining sign cases are reduced to this case by sign rules. Each step requires checking that the result is again a cut, and the field axioms are then verified one by one. Completeness comes from the fact that the union of a nonempty family of lower sets that is bounded above is again a lower set without a greatest element, which gives the supremum.

Cauchy sequences

Georg Cantor's construction, also published in 1872, works with sequences instead of partitions. A sequence (qn)(q_n) of rationals is a Cauchy sequence if for every rational ε>0\varepsilon>0 there is an NN such that ∣qm−qn∣<ε|q_m-q_n|<\varepsilon whenever m,n>Nm,n>N. Two Cauchy sequences are declared equivalent when their termwise difference tends to 0, and a real number is defined as an equivalence class of such sequences. Addition and multiplication are defined termwise, and these operations are well defined on classes because Cauchy sequences are bounded. A rational rr is embedded as the class of the constant sequence (r,r,r,…)(r,r,r,\ldots).

The inverse of a nonzero class is the most delicate step. A class is nonzero when its sequences do not tend to 0, which forces the terms to stay eventually bounded away from 0, so the termwise reciprocal is again a Cauchy sequence. The order is defined by saying x<yx<y when y−xy-x is represented by a sequence that is eventually bounded below some positive rational. The construction is designed so that Cauchy sequences of rationals acquire limits within the new system, which is the property that the construction is meant to deliver.

Decimal expansions and the case of 0.999…

A decimal viewpoint represents each real in [0,1][0,1] by a digit sequence a1,a2,…a_1,a_2,\ldots with value ∑k≥1ak10−k\sum_{k\ge1} a_k 10^{-k}, following the positional notation that spread from the Hindu-Arabic numeral system. This representation is not a definition by itself, because the same number can have two expansions. The number 0.999…0.999\ldots is the limit of the partial sums 0.9, 0.99, 0.999,…0.9,\ 0.99,\ 0.999,\ldots, and the difference from 1 is 10−n10^{-n}, which tends to 0. Hence 0.999…0.999\ldots and 1.000…1.000\ldots denote the same real number, and the same phenomenon occurs for every terminating decimal m/10nm/10^n. All other reals have a unique decimal expansion.

Arithmetic on expansions is also not carried out digit by digit from the left, since a carry can propagate through infinitely many places. For this reason the decimal picture is made rigorous by the limit definition, and the truncations of an expansion form a Cauchy sequence of rationals. This links the decimal viewpoint directly to Cantor's construction.

Constructions and axioms

The Dedekind reals and the Cantor reals are different sets, since a cut is a set of rationals and a Cantor real is a class of sequences. Both satisfy the complete ordered field axioms, and any two complete ordered fields are isomorphic through a unique isomorphism. This uniqueness theorem means that either construction, or any other valid model, can serve as R\mathbb{R}, and the choice does not change any property that depends only on the axioms. Under the isomorphism, the cut for 2\sqrt{2} corresponds to the class of a Cauchy sequence of rationals whose squares converge to 2.

The constructions themselves are carried out inside set theory. They require sets of rationals, sequences, and equivalence classes, and the rationals are in turn built from the integers and the natural numbers. The existence of the reals is therefore a theorem relative to a background theory such as ZFC, and the axioms of the real numbers are satisfied by the constructed objects within that framework.

Order, Density, and the Archimedean Property

Density of the rationals and irrationals

Between any two distinct real numbers lie both a rational number and an irrational number. In other words, both subsets are dense in R\mathbb{R}.

To find a rational between x<yx<y, choose a natural number nn with 1/n<y−x1/n<y-x, which the Archimedean property guarantees. Let m=⌊nx⌋+1m=\lfloor nx\rfloor+1. Then m>nxm>nx, so m/n>xm/n>x, and m≤nx+1m\le nx+1, so m/n≤x+1/n<ym/n\le x+1/n<y. Hence m/n∈(x,y)m/n\in(x,y).

To find an irrational, note that (x/2, y/2)(x/\sqrt{2},\,y/\sqrt{2}) has positive length and therefore contains a nonzero rational qq. Then x<q2<yx<q\sqrt{2}<y, and q2q\sqrt{2} is irrational, since otherwise 2=(q2)/q\sqrt{2}=(q\sqrt{2})/q would be rational.

Density does not imply that a set is large. The rationals are countable, while R\mathbb{R} is uncountable, yet the rationals are dense in R\mathbb{R}.

The Archimedean property

The Archimedean property states that for any real numbers x>0x>0 and yy, there is a natural number nn with nx>ynx>y. Equivalently, the set of natural numbers is unbounded above in R\mathbb{R}. The principle appears as a definition in Book V of Euclid's Elements and was used by Eudoxus and Archimedes, who gave it its name.

In R\mathbb{R}, this property is not an independent axiom but a consequence of the least upper bound property. If N\mathbb{N} were bounded above, then s=sup⁡Ns=\sup\mathbb{N} would exist. Since s−1s-1 is not an upper bound, some natural nn satisfies n>s−1n>s-1, so n+1>sn+1>s. But n+1∈Nn+1\in\mathbb{N}, which contradicts the choice of ss.

Two consequences follow. First, R\mathbb{R} has no nonzero infinitesimals: for every ε>0\varepsilon>0 there is nn with 1/n<ε1/n<\varepsilon, so no positive real is smaller than all of the numbers 1/n1/n. Second, every real number is bounded by a natural number: for each xx there is n∈Nn\in\mathbb{N} with n>xn>x.

The property alone does not single out R\mathbb{R}, because Q\mathbb{Q} is also Archimedean. Non-Archimedean ordered fields, such as the hyperreals, do contain infinitesimals, so the property must be combined with completeness to characterize the reals.

Absolute value and the metric

The absolute value is defined by ∣x∣=x|x|=x when x≥0x\ge 0 and ∣x∣=−x|x|=-x otherwise. It satisfies ∣xy∣=∣x∣ ∣y∣|xy|=|x|\,|y| and the triangle inequality ∣x+y∣≤∣x∣+∣y∣|x+y|\le |x|+|y|.

The distance d(x,y)=∣x−y∣d(x,y)=|x-y| is nonnegative, equals zero only when x=yx=y, is symmetric, and satisfies d(x,z)≤d(x,y)+d(y,z)d(x,z)\le d(x,y)+d(y,z). These properties make R\mathbb{R} a metric space, which is the setting for defining limits and continuity through ε\varepsilon-δ\delta conditions. The same distance is the basis for the Euclidean distance on Rn\mathbb{R}^n.

Intervals, supremum, and infimum

For real numbers a≤ba\le b, the intervals are: the closed interval [a,b][a,b], the open interval (a,b)(a,b), and the half-open intervals [a,b)[a,b) and (a,b](a,b]. Unbounded intervals include (a,∞)(a,\infty), [a,∞)[a,\infty), (−∞,b](-\infty,b], and (−∞,b)(-\infty,b), and R\mathbb{R} itself is (−∞,∞)(-\infty,\infty). The degenerate interval [a,a][a,a] is a single point, and (a,a)(a,a) is empty.

A set S⊆RS\subseteq\mathbb{R} is bounded above if some MM satisfies x≤Mx\le M for all x∈Sx\in S, and such an MM is an upper bound. The supremum sup⁡S\sup S is the least upper bound, meaning an upper bound that is less than or equal to every other upper bound. Dually, the infimum inf⁡S\inf S is the greatest lower bound. The completeness axiom guarantees that every nonempty set bounded above has a supremum in R\mathbb{R}, and such a supremum is unique when it exists.

Equivalently, s=sup⁡Ss=\sup S if and only if ss is an upper bound and, for every ε>0\varepsilon>0, some x∈Sx\in S satisfies x>s−εx>s-\varepsilon. The supremum need not belong to SS. For example, sup⁡{1−1/n:n∈N}=1\sup\{1-1/n : n\in\mathbb{N}\}=1, but 1∉S1\notin S, so SS has no maximum. A set has a maximum exactly when its supremum lies in the set. The empty set has no supremum in R\mathbb{R}, since every real number is an upper bound for it, and no unbounded set has a supremum in R\mathbb{R}.

Completeness and the Core Theorems of Analysis

Within the ordered-field framework, the least upper bound property of R\mathbb{R} is equivalent to several other completeness statements, and analysis uses these forms interchangeably. The most common is Cauchy completeness. A Cauchy sequence is one whose terms eventually cluster arbitrarily tightly, and R\mathbb{R} is complete in the sense that every such sequence has a limit inside R\mathbb{R}. With the usual distance ∣x−y∣|x-y|, this makes R\mathbb{R} a complete metric space. Cauchy's 1821 Cours d'analyse made this criterion central, and the equivalence with the supremum property shows that the two formulations describe the same gap-free line.

The sequence and compactness theorems follow from this completeness. The Bolzano–Weierstrass theorem states that every bounded sequence of real numbers has a convergent subsequence. The Heine–Borel theorem, in its covering form due to Borel (1895), states that every open cover of a closed bounded interval [a,b][a,b] has a finite subcover, so that such intervals are compact. In R\mathbb{R}, the compact sets are exactly the closed and bounded ones. The nested interval property says that a decreasing sequence of nonempty closed bounded intervals has nonempty intersection, and if the lengths tend to zero, that intersection is a single point. The monotone convergence theorem says that a bounded increasing sequence converges to its supremum, and a bounded decreasing one converges to its infimum.

The intermediate value theorem and the extreme value theorem concern continuous functions on closed intervals. If ff is continuous on [a,b][a,b], then ff takes every value between f(a)f(a) and f(b)f(b). Bolzano proved this in 1817 by a bisection argument, which is equivalent to the existence of a supremum. The extreme value theorem states that such an ff is bounded and attains both a maximum and a minimum on [a,b][a,b]. Both results depend on completeness, because the proofs locate a point whose existence requires a limit or supremum to be present in the number system.

Restricting these results to Q\mathbb{Q} makes the gaps visible. Consider the decimal truncations of 2\sqrt{2}: 1, 1.4, 1.41, 1.414, and so on. This sequence is Cauchy and bounded in Q\mathbb{Q}, and it is increasing, yet it has no rational limit, so the monotone convergence theorem and Cauchy completeness both fail. The nested intervals [1,2]⊃[1.4,1.5]⊃[1.41,1.42]⊃⋯[1,2] \supset [1.4,1.5] \supset [1.41,1.42] \supset \cdots have empty intersection among rationals, while in R\mathbb{R} their intersection is {2}\{\sqrt{2}\}. The rational points of [0,2][0,2] also violate the compactness theorems. Set Un={q∈Q:∣q−2∣>1/n}U_n = \{q \in \mathbb{Q} : |q-\sqrt{2}| > 1/n\} for n≥1n \ge 1. These sets cover Q∩[0,2]\mathbb{Q} \cap [0,2], since no rational equals 2\sqrt{2}, but any finite subfamily misses the rationals within 1/N1/N of 2\sqrt{2}, so no finite subcover exists. Bolzano–Weierstrass fails here too, because the rational approximations to 2\sqrt{2} form a bounded sequence with no rational subsequential limit.

The same obstructions appear in the main theorems. For f(x)=x2−2f(x)=x^2-2 on Q∩[1,2]\mathbb{Q} \cap [1,2], the values f(1)=−1f(1)=-1 and f(2)=2f(2)=2 have opposite signs, yet ff has no rational zero, so the intermediate value theorem fails. For f(x)=1/(x2−2)f(x)=1/(x^2-2) on Q∩[0,2]\mathbb{Q} \cap [0,2], the function is continuous at every rational point, since the denominator never vanishes there, but it is unbounded near 2\sqrt{2}. It therefore has no maximum, and the extreme value theorem fails. Each failure occurs because the missing value 2\sqrt{2} lies inside the interval but outside Q\mathbb{Q}.

Irrational and Transcendental Numbers

The irrationality of 2\sqrt{2} has the classical proof by contradiction. If 2=p/q\sqrt{2} = p/q in lowest terms, then p2=2q2p^2 = 2q^2, so pp is even, and then qq must also be even, which contradicts lowest terms. Euler established in 1737 that ee and e2e^2 are irrational. Johann Heinrich Lambert proved the irrationality of π\pi in 1761, and his memoir was published in 1768. His method expanded tan⁡x\tan x as a continued fraction and showed that this expression cannot be rational when xx is a nonzero rational. Since tan⁡(π/4)=1\tan(\pi/4) = 1, the number π/4\pi/4, and therefore π\pi, must be irrational. In the same work Lambert conjectured that both ee and π\pi are transcendental, and he sketched an argument for π\pi that was not complete. Later proofs of irrationality, including Hermite's proof for π\pi, used different techniques, but the logic of a contradiction from a rational assumption remained the basic pattern.

A real number is algebraic if it is a root of a nonzero polynomial with integer coefficients, and transcendental if it is not algebraic. Every rational number is algebraic, since p/qp/q is a root of qx−pqx - p. Numbers such as 2\sqrt{2} are algebraic irrationals, but algebraicity is not the same as rationality. Lambert was the first to treat transcendental numbers as numbers that are not roots of polynomials with rational coefficients. Even so, for most of the eighteenth and early nineteenth centuries it was not known whether any transcendental numbers existed at all.

Joseph Liouville settled that question in 1844. He showed that a nonzero algebraic number α\alpha of degree nn cannot be approximated too closely by rationals: there is a constant c>0c > 0 with ∣α−p/q∣≥c/qn|\alpha - p/q| \ge c/q^n for all rationals p/qp/q with q>0q > 0. Real numbers that admit rational approximations violating this bound for every nn must be transcendental. Liouville produced explicit examples of this kind, first through continued fractions and later in 1851 through decimal expansions. The best-known decimal example, Liouville's constant, has a 1 in the decimal place n!n! and zeros elsewhere. Its transcendence shows that some numbers can be classified in a fully explicit way, but this class is very thin and does not account for most transcendental numbers.

The first natural constants to be proved transcendental were ee and π\pi. In 1873 Charles Hermite proved that ee is transcendental, using auxiliary functions built from the exponential function, which is closely related to the Padé approximants he studied. In 1882 Ferdinand von Lindemann extended Hermite's method to π\pi. He used Euler's identity eiπ=−1e^{i\pi} = -1: if π\pi were algebraic, then iπi\pi would be algebraic and nonzero, which contradicts the transcendence of eiπ=−1e^{i\pi} = -1 under the Hermite–Lindemann theorem. Lindemann's result therefore implied that the circle cannot be squared with compass and straightedge. Karl Weierstrass generalized the theorem in 1885 to the statement that eαe^{\alpha} is transcendental for every nonzero algebraic α\alpha. In 1929 Gelfond proved that eπe^{\pi} is transcendental.

Transcendental numbers are not rare in the measure-theoretic sense, and a counting argument shows this. Each nonzero integer polynomial has finitely many roots, and there are only countably many such polynomials, so the set of algebraic numbers is countable. Because R\mathbb{R} is uncountable, as Cantor proved, the complement of the algebraic numbers in R\mathbb{R} is uncountable, and almost all real numbers are transcendental. This existence result is non-constructive: it does not identify any particular transcendental number. Cantor's argument came after Liouville's explicit examples, and the two approaches together show that transcendental numbers are both abundant and difficult to identify one by one.

Many specific constants remain unresolved. It is not known whether the Euler–Mascheroni constant is irrational, and no proof exists either way. It is also unknown whether π\pi and ee are algebraically independent. Mathematicians have not been able to prove that e+πe + \pi or eπe\pi is transcendental, although both are expected to be. The gap between what is known about individual numbers and what is true of almost all reals is one of the central features of this area of number theory.

Cardinality and Measure

Cantor's proof that ℝ is uncountable uses a diagonal construction on the interval [0,1][0,1]. Suppose the reals in [0,1][0,1] could be listed as x1,x2,x3,…x_1, x_2, x_3, \dots, each written as a decimal expansion. Build a new number y=0.y1y2y3…y = 0.y_1y_2y_3\dots by choosing each digit yny_n to differ from the nn-th digit of xnx_n. Restricting the digits to avoid 0 and 9 (for example, using 5 unless the diagonal digit is 5, in which case using 6) prevents the same number from having two decimal expansions, which is the technical point that makes the argument work for reals. The resulting yy lies in [0,1][0,1] but differs from every xnx_n, so no such list exists. Cantor published this argument in 1891, and it is treated in detail in Cantor's diagonal argument. Since [0,1][0,1] is uncountable, so is ℝ. The contrast with the natural numbers, which can be listed, is the basis of the distinction between countable and uncountable sets.

The notion of cardinality measures the size of sets by the existence of bijections. The cardinality of ℕ is ℵ0\aleph_0, and the cardinality of ℝ is denoted c\mathfrak{c} (the "cardinality of the continuum") or 2ℵ02^{\aleph_0}. Cantor's theorem, proved by a diagonal argument in general form, states that the power set of any set has strictly greater cardinality than the set itself. Applied to ℕ, this shows that P(N)\mathcal{P}(\mathbb{N}) is larger than ℕ, and binary expansions give a bijection between P(N)\mathcal{P}(\mathbb{N}) and [0,1][0,1], so c=2ℵ0\mathfrak{c} = 2^{\aleph_0}. Several other sets have the same cardinality as ℝ: the open interval (0,1)(0,1) is in bijection with ℝ through a simple trigonometric or rational function, and ℝⁿ can be put in bijection with ℝ for every finite nn, for example by interleaving digits. These bijections are only set-theoretic. They do not preserve continuity, distance, or dimension, so counting alone does not capture the geometry of ℝⁿ.

Cardinality does not distinguish all uncountable sets from all countable ones in a way useful for analysis, and measure theory supplies a finer notion of size. Lebesgue measure generalizes length: it assigns b−ab - a to an interval (a,b)(a,b) and extends this to a large class of subsets of ℝ, the Lebesgue measurable sets. Not every subset of ℝ is measurable under the usual axioms, a fact tied to the axiom-of-choice. Countable sets always have measure zero, because each point has length zero and countable unions of null sets are null. The rationals therefore have Lebesgue measure zero even though they are dense in ℝ. Measure zero does not require countability, however: the Cantor set is uncountable yet has measure zero. This leads to the phrase "almost everywhere": a property holds almost everywhere if the set where it fails has measure zero. For example, the indicator function of the rationals equals zero almost everywhere, so its Lebesgue integral over [0,1][0,1] is zero, even though it differs from the zero function at every rational point. Such statements are central to measure-theory and to the Lebesgue integral.

The continuum-hypothesis is Cantor's conjecture that no set has cardinality strictly between ℵ0\aleph_0 and c\mathfrak{c}, that is, that c=ℵ1\mathfrak{c} = \aleph_1. Cantor formulated it in 1878, and David Hilbert listed it first among his problems in 1900. The question was settled in the sense that it cannot be decided from the standard axioms. Kurt Gödel showed in 1940 that the continuum hypothesis cannot be refuted from Zermelo–Fraenkel set theory with the axiom of choice (ZFC), by constructing an inner model LL in which it holds. Paul Cohen showed in 1963, using his method of forcing, that the negation of the continuum hypothesis is also consistent with ZFC. Together these results establish that the hypothesis is independent of ZFC, so neither it nor its negation can be proved there, provided ZFC itself is consistent. Cohen's method has since been applied extensively in set-theory, and the outcome has led to a continuing debate over whether the reals have a determinate cardinality or whether additional axioms are needed to fix it.

Representation and Computation

Every real number has a positional expansion in any integer base b≥2b \ge 2, as described in positional notation. Long division shows that a rational number has an eventually periodic decimal expansion, because dividing by a fixed denominator leaves only finitely many possible remainders, so the digits must eventually repeat. The converse also holds: any eventually periodic expansion represents a rational. For example, 1/7=0.142857‾1/7 = 0.\overline{142857}, and a terminating expansion is the special case in which the repeating block is zero. Irrational numbers such as 2=1.41421356…\sqrt{2} = 1.41421356\ldots never settle into a repeating pattern. The same criterion applies in base 2, which explains why many decimal fractions with finite expansions, such as 1/101/10, have infinite binary expansions: 1/10=0.00011‾21/10 = 0.0\overline{0011}_2. Only reals with terminating expansions have two expansions in a given base, one ending in zeros and one ending in nines, as in 0.999…=10.999\ldots = 1.

A different representation is the continued fraction, in which every real xx can be written as x=a0+1a1+1a2+⋯x = a_0 + \cfrac{1}{a_1 + \cfrac{1}{a_2 + \cdots}}, with an integer a0a_0 and positive integers aia_i. Rationals have finite continued fractions and irrational numbers have infinite ones. The truncations of the fraction, called convergents, are the best rational approximations to xx in a precise sense. Lagrange's theorem states that the continued fraction of a real number is eventually periodic exactly when the number is a quadratic irrational, a root of a quadratic equation with integer coefficients. Thus 2=[1;2‾]\sqrt{2} = [1; \overline{2}], while e=[2;1,2,1,1,4,1,1,6,…]e = [2; 1, 2, 1, 1, 4, 1, 1, 6, \ldots] follows a visible pattern and π\pi has no known regular pattern in its partial quotients.

Turing's 1936 paper "On Computable Numbers" introduced the notion of a computable real. A real is computable if some Turing machine can produce its digits, or approximations of it to any requested accuracy. Familiar constants such as 2\sqrt{2}, ee, and π\pi are computable, and so are all algebraic numbers. Computable reals are closed under the field operations and under many standard functions. Since every Turing machine can be described by a finite string of symbols, there are only countably many computable reals, and so almost all real numbers are not computable. Explicit non-computable reals exist, for example Chaitin's halting probability. Deciding whether a computable real given by a program equals zero is not possible in general, which limits what can be automated. The broader context is computability theory.

In practice, computers do not manipulate reals directly. Floating-point arithmetic, standardized as IEEE 754, stores a finite set of binary numbers, each with a fixed-width significand and an exponent. In double precision the significand has 53 bits, which corresponds to roughly 15 to 17 significant decimal digits, and the unit roundoff is 2−53≈1.1×10−162^{-53} \approx 1.1 \times 10^{-16}. Each arithmetic operation returns the nearest representable value, so the computed result equals the exact result multiplied by 1+δ1 + \delta with ∣δ∣≤2−53|\delta| \le 2^{-53}. The consequences are concrete: 0.1 has no finite binary expansion, so 0.1+0.20.1 + 0.2 yields 0.30000000000000004 in double precision; addition is no longer associative; and cancellation between nearly equal quantities destroys significant digits, as in evaluating (1−cos⁡x)/x2(1 - \cos x)/x^2 for small xx. The study of how such errors grow and how algorithms can be designed to control them belongs to numerical analysis and numerical stability. Because the set of floating-point values is finite, it does not form a field.

Two methods aim to restore reliability. Interval arithmetic represents a quantity by an interval that is guaranteed to contain its true value. Each operation produces an enclosing interval, with lower bounds rounded down and upper bounds rounded up, which is the approach popularized by Ramon Moore in 1966. Its main weakness is the dependency problem: for x∈[1,2]x\in[1,2], the expression x−xx - x evaluates to [−1,1][-1, 1] rather than 0, so widths grow unless the computation is restructured or subdivided. Interval methods are used in verified numerics and in computer-assisted proofs. Exact real arithmetic takes a different approach by representing each real as a procedure that delivers approximations to any precision on demand, refining them only when needed. This makes results correct to arbitrary precision, but comparisons such as x=0x = 0 or x<yx < y may fail to terminate when the true value lies exactly on the boundary. Implementations therefore use three-valued logic or precision cut-offs to handle such cases.

Generalizations and Related Number Systems

The real numbers are not the end of the number-system hierarchy. Several constructions extend, complete, or reorganize them. The most familiar is the passage to the complex numbers, C=R(i)\mathbb{C} = \mathbb{R}(i), obtained by adjoining a root of x2+1=0x^2 + 1 = 0 to ℝ. Since C\mathbb{C} is a two-dimensional real vector space and every complex number is algebraic over ℝ, it is a quadratic field extension of ℝ and is in fact the algebraic closure of ℝ. The fundamental theorem of algebra states that every nonconstant complex polynomial has a complex root. As a consequence, up to isomorphism, ℝ and ℂ are the only finite extensions of ℝ. Gauss gave a proof in his 1799 doctoral dissertation, and later proofs draw on analysis, topology, or Galois theory.

Real closed fields give the algebraic characterization that explains why ℝ behaves as it does. An ordered field is real closed when every positive element has a square root and every polynomial of odd degree has a root. ℝ is real closed, and so is the field of real algebraic numbers, which is a proper subfield of ℝ. Any real closed field that is Archimedean is isomorphic to ℝ, so the completeness of ℝ is not needed in this algebraic formulation for the Archimedean case. Alfred Tarski showed that the first-order theory of real closed fields is complete and decidable, which allows many geometric and polynomial questions over ℝ to be settled by mechanical procedures.

The extended real line adjoins two points, −∞-\infty and +∞+\infty, to ℝ and extends the order so that each of them is larger or smaller than every real number. This set is not a field, because expressions such as ∞−∞\infty - \infty remain undefined. It is nevertheless convenient in analysis and in measure theory, where it allows limits, suprema, and integrals of unbounded or divergent quantities to be stated uniformly. Its order topology makes it compact, which is a practical advantage when limits must always exist.

Non-standard extensions add infinitesimals to ℝ. Abraham Robinson's hyperreal numbers, developed in the 1960s, are built by an ultrapower construction on sequences of reals. Łoś's theorem then lets statements about ℝ transfer to the larger ordered field, so that hyperreal numbers can be used to reason about derivatives and integrals with infinitesimal quantities. The surreal numbers, introduced by John Conway in the 1970s and popularized by Donald Knuth's 1974 expository work, form a much larger ordered field. They contain ℝ, the ordinal numbers, and infinitesimals, and they form a proper class rather than a set. Both extensions are real closed fields and are not complete in the sense of the least upper bound property. Neither is a complete ordered field of the kind characterized by the uniqueness theorem.

The p-adic numbers provide a different completion of ℚ. For each prime p, the p-adic absolute value of a nonzero rational is p−kp^{-k}, where k is the exponent of p in its factorization. This absolute value satisfies ∣x+y∣p≤max⁡(∣x∣p,∣y∣p)|x+y|_p \le \max(|x|_p, |y|_p), a stronger form of the triangle inequality that makes the space non-Archimedean. Kurt Hensel introduced these numbers in 1897, and the completion of ℚ with respect to ∣⋅∣p|\cdot|_p is the field Qp\mathbb{Q}_p. Alexander Ostrowski proved in 1916 that every nontrivial absolute value on ℚ is equivalent either to the usual absolute value or to some p-adic absolute value. Thus, up to equivalence, the completions of ℚ are exactly ℝ and the fields Qp\mathbb{Q}_p.

Finally, ℝ serves as the scalar field for the theory of normed spaces. The space Rn\mathbb{R}^n with the Euclidean norm ∥x∥=x12+⋯+xn2\|x\| = \sqrt{x_1^2 + \cdots + x_n^2} is a complete metric space, and the Heine–Borel theorem holds in it. Because all norms on a finite-dimensional real vector space are equivalent, Rn\mathbb{R}^n is complete under any norm. A Banach space is a complete normed vector space, and ℝ is its simplest one-dimensional example. Infinite-dimensional examples include sequence spaces such as ℓ2\ell^2 and the continuous functions on [0, 1] with the supremum norm. Those with norms arising from an inner product form the subclass of Hilbert spaces.

Applications in Science and Mathematics

ℝ as the domain of analysis

Within mathematics, ℝ is the standard domain and codomain for the functions studied in calculus. Derivatives and integrals are defined by limits taken over real variables, and the completeness of ℝ ensures that the limiting processes used in these definitions have values inside the number system. Continuous functions on closed intervals inherit the properties that make optimization and root-finding possible, which is why ℝ is the usual setting for mathematical analysis.

Differential equations use ℝ in two ways. Time is a real variable, and the unknown is a function of real space and time coordinates. The heat equation ∂u∂t=α∂2u∂x2\frac{\partial u}{\partial t} = \alpha \frac{\partial^2 u}{\partial x^2}, for example, describes a temperature field u(x,t)u(x,t) with xx and tt ranging over real intervals. Solutions are typically real-valued functions, and existence and uniqueness theorems depend on the metric structure of ℝ and ℝⁿ.

Fourier analysis extends this role. A periodic function on an interval is expanded as a sum of sines and cosines, which is the Fourier series approach. On the whole real line, the Fourier transform assigns to a function of a real variable another function of a real frequency variable. Both constructions take ℝ as their basic index set and value set.

Probability and statistics

A real-valued random variable is a measurable function from a probability space into ℝ. Its behaviour is described by the cumulative distribution function F(x)=P(X≤x)F(x) = P(X \le x), which is nondecreasing and tends to 0 and 1 as xx goes to −∞-\infty and +∞+\infty. Continuous distributions are given by densities, and probabilities are computed as integrals over subsets of ℝ, typically with respect to Lebesgue measure.

The normal distribution is the most widely used example. For a mean μ∈R\mu \in \mathbb{R} and a standard deviation σ>0\sigma > 0, its density is

f(x)=1σ2π e−(x−μ)2/(2σ2),x∈R,f(x) = \frac{1}{\sigma\sqrt{2\pi}}\, e^{-(x-\mu)^2/(2\sigma^2)}, \qquad x \in \mathbb{R},

and it integrates to 1 over the whole line. Measurement errors, sample statistics and many physical noise models are analysed through distributions of this kind, so the real line is both the sample space of the quantity and the place where its probability is assigned.

Physical quantities and the SI

Length, mass, time and temperature are modelled as real multiples of units. The International System of Units changed its foundations on 20 May 2019, when the kilogram, ampere, kelvin and mole were defined by fixing exact numerical values of the Planck constant, the elementary charge, the Boltzmann constant and the Avogadro constant. The metre is tied to the speed of light, fixed at 299 792 458 m/s, and the second is tied to a fixed caesium frequency of 9 192 631 770 Hz.

These fixed values are exact decimal or integer numbers, and in that sense they are rational. The quantities they define are real-valued in the model, but every laboratory measurement yields a finite decimal approximation with a stated uncertainty. In continuum mechanics, the same idealization is used at the macroscopic scale: density, velocity, pressure and stress are treated as real-valued fields over spatial and temporal coordinates in ℝ³ and ℝ, which allows the use of calculus on material bodies.

Parameters, losses, and precision in machine learning

Model parameters in neural networks are vectors of real numbers, and the training objective, or loss, is a real-valued function of those parameters. Gradient-based optimization updates the parameters using derivatives of the loss. In computers, however, these real values are stored in floating-point formats with finite bit widths. Single precision (FP32) has 8 exponent bits and 23 fraction bits, half precision (FP16) has 5 exponent bits and 10 fraction bits, and bfloat16 (BF16) keeps the 8-bit exponent of FP32 but has only 7 fraction bits.

Precision limits affect training in concrete ways. The smallest normal FP16 value is about 6.1×10−56.1 \times 10^{-5}, so small gradients can underflow to zero. To counter this, mixed-precision training usually scales the loss by a large factor, commonly a power of two, before backpropagation, and then removes the scaling in the weight update. Most forward and backward operations run in 16-bit arithmetic, while accumulators, loss values and master copies of the weights stay in FP32. The reason is visible in the spacing of FP16 numbers: near 1.0 the gap between representable values is 2−10≈9.77×10−42^{-10} \approx 9.77 \times 10^{-4}, so a tiny update of order 10−1010^{-10} applied directly in FP16 would be rounded away. BF16 avoids overflow more readily because its exponent range matches FP32, at the cost of fewer significant digits.

Foundational Debates and Open Problems

Platonism, formalism, and finitism

Most working analysts treat R\mathbb{R} as a completed totality whose existence does not depend on any particular notation or construction, a position usually called mathematical Platonism. Rival positions dispute this. Formalists, following Hilbert's program, treat the reals as rules for manipulating symbols and seek to justify those rules by proving the consistency of the formal system. Finitists accept only objects and arguments that can be carried out in finitely many steps, so they hesitate to treat an infinite decimal expansion as a finished object. These disagreements rarely change the calculus results that practitioners use every day. They concern what those results are about and which proofs count as legitimate.

Intuitionism and Brouwer's continuum

Intuitionism, founded by L. E. J. Brouwer, rejects the law of excluded middle and rebuilds the continuum from different primitives. Brouwer introduced choice sequences, which are infinite sequences whose terms are fixed one at a time, either freely or under a law. Brouwer introduced the continuity principle, which roughly asserts that any function from choice sequences to the natural numbers is continuous, so the value at a given sequence depends only on a finite initial segment. Within this framework, every real-valued function on the continuum comes out continuous. Brouwer also needed a further result, the uniform continuity of continuous functions on closed intervals, which classical analysis proves outright. These principles alone do not suffice to recover that part of traditional analysis that Brouwer considered intuitionistically sound, so he added the bar theorem. Even within intuitionism, the freely proceeding choice sequences remain controversial.

Constructive analysis

Errett Bishop's 1967 Foundations of Constructive Analysis showed that a constructive approach could reach most of real analysis. He showed that, contrary to previous perceptions, large swaths of mathematics could be developed constructively with only minor changes from the classical theory. Bishop objected to the free use of the law of the excluded middle and other nonconstructive principles, which he called "idealistic". In his approach, an existence claim must come with a method for producing the object, and operations must be computable in principle. The Bishop–Bridges revision of 1985 kept this program, linking constructive analysis to computability theory and to certified numerical methods.

Kronecker and the debate over existence proofs

Leopold Kronecker, active in the 1880s, regarded the integers as the secure foundation of mathematics and was skeptical of Cantor's set theory. He doubted that a proof showing an object must exist, without exhibiting it, delivered a genuine mathematical object. This objection is the historical root of the modern question of how far nonconstructive arguments may be trusted when they concern infinite collections. Classical analysis accepts such arguments, while constructive schools limit them, and the two camps still disagree about which theorems are meaningful.

Physical continuity

Whether physical space and time are continuous is a question about nature that mathematics cannot settle. Measurements always have finite precision, so experiments can never fully distinguish a continuum from a very fine discrete structure. Some approaches to quantum gravity, such as loop quantum gravity, predict discrete spectra for geometric quantities near the Planck length, roughly 1.6×10−351.6 \times 10^{-35} m. Other approaches retain continuous models, and the use of R\mathbb{R} in physics thus remains a modeling choice.

Open problems in the arithmetic of the reals

A central open question is whether π\pi is a normal number. A real number is normal in base bb if every finite block of kk digits appears with limiting frequency b−kb^{-k}. Although Borel proved that almost all real numbers are normal in the measure-theoretic sense, establishing the normality of specific mathematical constants remains one of the outstanding problems in number theory. At the end of 2025, a new computational record reached 314 trillion decimal digits, and the digit statistics agree with normality, yet no proof exists. Computations have also shown that π\pi appears normal in bases 10 and 16, but whether it is absolutely normal, that is, normal in every base, remains open. The same gap applies to ee and 2\sqrt{2}, since so far none of these has been proved normal in any base. Conjectures extend further: every irrational algebraic number is conjectured to be normal, but only weak lower bounds on digit counts are currently known.

References

  1. Transcendental numberen.wikipedia.org
  2. Johann Heinrich Lambert (1728 - 1777) - Biography - MacTutor History of Mathematicsmathshistory.st-andrews.ac.uk
  3. Properly Understanding Lambert's Mémoire on piwww2.cms.math.ca
  4. Recounting the History of Math’s Transcendental Numbersquantamagazine.org
  5. Ferdinand von Lindemann (1852 - 1939) - Biography - MacTutor History of Mathematicsmathshistory.st-andrews.ac.uk
  6. Lindemann%E2%80%93Weierstrass theoremen.wikipedia.org
  7. Liouville's Constant -- from Wolfram MathWorldmathworld.wolfram.com
  8. Auxiliary functions in transcendence proofsarxiv.org
  9. Algebraic and transcendental solutions of some exponential equationsarxiv.org
  10. Non-archimedean analysis on the extended hyperreal line $^*R_d$ and the solution of some very old transcendence conjectures over the field $Q$arxiv.org
  11. Cantor's Diagonal Argumentmathcenter.oxford.emory.edu
  12. Real Numbers are Uncountably Infinite/Cantor's Diagonal Argument - ProofWikiproofwiki.org
  13. Continuum hypothesisen.wikipedia.org
  14. The Continuum Hypothesis (Stanford Encyclopedia of Philosophy)plato.stanford.edu
  15. The Complete Idiot’s Guide to the Independence of the Continuum Hypothesis: Part 1 of <=Aleph_0scottaaronson.blog
  16. A Formal Proof of the Independence of the Continuum Hypothesisarxiv.org
  17. Lebesgue measureen.wikipedia.org
  18. Measure Zero Henry Y. Chan July 1, 2013 1 Measure Zeromath.uchicago.edu
  19. There are Almost No Rational Numbersaustinrochford.com
  20. Almost everywhere — Grokipediagrokipedia.com
  21. 2019 revision of the SIen.wikipedia.org
  22. SI Units Explained: 7 Base Units & 2019 Redefinitionunitconv.org
  23. Mixed Precision Training: FP16, BF16, and Loss Scaling - Interactivembrenndoerfer.com
  24. Mixed Precision Training — How It Worksadhdecode.com
  25. Computability and analysis: the legacy of Alan Turingarxiv.org
  26. Intuitionism in the Philosophy of Mathematics (Stanford Encyclopedia of Philosophy)plato.stanford.edu
  27. Errett Bishopen.wikipedia.org
  28. Bishop's constructive mathematics in nLabncatlab.org
  29. Constructive Analysis in the Agda Proof Assistantarxiv.org
  30. Can $\pi$ generate itself? A Monte Carlo analysis of 314 trillion digitsarxiv.org
  31. The discrepancy of the Champernowne constantarxiv.org
  32. Transcendence Meets Normality: Construction of Transcendentally Normal Numbersarxiv.org
  33. Intuitionism in Mathematicsiep.utm.edu
  34. Sagan numbersarxiv.org
  35. Abelian-normal decimal expansionsarxiv.org