Work in ZF without choice, after the natural/integer/rational construction. All bounds in the construction initially use positive rationals. In particular we do not define a distance by taking a supposedly already existing real limit: real numbers are the objects being constructed.
Motivation
The rational square-root gap suggests adding limits, but identifying every sequence with a new number would add too many copies. Two rational Cauchy sequences whose difference tends to zero should determine the same object. Products require boundedness, inverses require a uniform gap from zero, and order requires eventual positive separation rather than positivity of every term.
After these algebraic steps, completeness still needs proof. We will find a supremum by rational bisection. Each step asks whether one rational midpoint bounds the given set; it does not select a new element of that set. This makes the construction explicit in ZF and avoids a hidden countable-choice step.
Definitions
Definition 1.6.1
(Cauchy sequences, null sequences and )
A rational sequence is Cauchy if for every positive rational there is with for at least . It is null if eventually for each such . Let be the set of rational Cauchy sequences, and define
This is a set quotient of a subset of . Addition, multiplication and negation are defined termwise before passing to classes; denotes the class of the constant rational sequence .
Definition 1.6.2
(Positive classes and order)
Call positive when there is a positive rational such that eventually . Define when is positive. The word “eventually” permits any finite initial segment to be ignored; it does not permit a gap shrinking to zero.
Definition 1.6.3
(Dedekind completeness and the Archimedean property)
An ordered field is Dedekind complete if every nonempty subset bounded above has a least upper bound. It is Archimedean if its embedded natural numbers are unbounded above. Sequence convergence and the Cauchy property in an Archimedean ordered field may be tested using positive rational , equivalently all positive field elements.
Theorem/proposition
Theorem 1.6.4
(The Cauchy quotient is an ordered field)
The relation is an equivalence relation, termwise operations are well defined, and every nonzero class has an inverse. The eventual-gap definition gives a compatible total order, extending . The rational subfield is dense and every real is bounded in absolute value by a positive integer.
Theorem 1.6.5
(Completeness without assuming real limits)
Every nonempty bounded-above subset of the constructed has a supremum. Every real Cauchy sequence converges in . The supremum proof uses rational bisection and no selection of a sequence of elements from the bounded set.
Theorem 1.6.6
(Uniqueness of the complete ordered field)
Every Dedekind-complete ordered field is Archimedean and is uniquely order-isomorphic to the constructed by a field isomorphism fixing . Different set codings therefore give the same ordered-field structure up to this unique isomorphism.
Proof sketch or proof idea
Proof
(Field and order on Cauchy classes)
The triangle inequality makes reflexive, symmetric and transitive. Every rational Cauchy sequence is bounded: choose a tail within one of a fixed term and bound its finitely many earlier absolute values as well. If have rational absolute bounds , then
Thus products are Cauchy; sums and negatives are immediate. The same estimate applied to two representatives shows products of equivalent sequences equivalent. Pointwise rational identities give all commutative ring laws on classes, and constant one is not zero.
If is not null, some has arbitrarily late terms of magnitude at least . Choose a Cauchy tail with all differences less than , and one such large term in that tail. Every term of the tail then has magnitude greater than . Define there and, say, zero on the finite prefix. Its tail differences are bounded by , so is Cauchy. The product is eventually one, hence represents one. Inverses are independent of the representative because inverses in a commutative ring are unique: if , then . This proves the field property without assuming an inverse for zero.
A null perturbation changes an eventual lower bound into at worst , so positivity is representative independent. A nonzero Cauchy sequence is eventually bounded away from zero as above, and eventually has one sign: choose a further tail whose differences are less than twice that gap; opposite signs would violate it. Thus exactly one of , positive, or positive holds. Sums and products of positives retain lower bounds and . This gives a total ordered field. Constant sequences preserve rational algebra and order.
Given a representative of and rational , a sufficiently late term satisfies : the representative differences are eventually less than , leaving a positive gap in each inequality. For , write their representative difference eventually at least . Choose a common so each representative differs from its th term by less than on the tail. The rational midpoint then lies strictly between , with gaps at least . This proves density. Boundedness of a representative also gives , and a larger natural integer exists by rational arithmetic, proving the Archimedean property.
Proof
(Supremum by rational bisection, then Cauchy completeness)
Let be nonempty with upper bound . Choose just once and an integer with and . Start with rational bounds , . At each step take their midpoint . If bounds above, replace by ; otherwise replace by . Recursion on defines these sequences from the predicate “is an upper bound of ”, which is a set-theoretic formula. At every step is not an upper bound, is one, and
The width tends to zero using rational inequalities, for . Thus is rational Cauchy. Put . For each fixed , all later lie in . An eventually nonnegative representative has nonnegative class by order trichotomy, so
If some exceeded , choose a positive rational smaller than by density and take width less than . Then , contradicting its upper-bound property. Hence bounds . If , take width smaller than a positive rational below . Then . Since is not an upper bound, some exceeds it, showing is not an upper bound. Therefore is the supremum. No sequence of such was chosen; each use is an individual existential argument.
Dedekind completeness also proves Cauchy completeness. A real Cauchy sequence is bounded by the same finite-prefix/tail argument. Let be the infimum and the supremum of its th tail. These exist by least-upper-bound completeness and negation. The increase and are bounded, so let be their supremum. Nestedness gives for every . A Cauchy tail with pairwise differences less than lies within of its first term, hence has width at most . Thus , and proves convergence. Rational tests suffice by density and the Archimedean property.
Proof X-ray
Why bisection closes without choosing from the set
The supremum argument runs on one invariant: is not an upper bound of , and is. The midpoint test preserves it by construction, and halving drives the width to zero, so both endpoint sequences determine one class squeezed between them. Each half of the supremum claim then uses the invariant at a single late : an element above would exceed the upper bound , while a bound would lie below the non-bound . Nonemptiness and the bound are used only to start the invariant.
Proof
(Every complete ordered field is a copy of the construction)
For uniqueness, first show any Dedekind-complete ordered field is Archimedean. If its naturals had supremum , would not bound them, giving and hence , a contradiction. Its rational subfield is dense: for choose with and the least integer strictly greater than . Such an exists and has a least value because integers bounded below translate into a subset of the well-ordered naturals. Then .
The Cauchy-completeness proof just given uses only ordered-field and supremum properties, so applies to . Map to the -limit of its rational terms. Null differences give equal limits. Sums commute with limits by the triangle inequality; for products use
and boundedness. This makes a field homomorphism fixing ; a positive class has an eventual positive rational lower bound, so its limit is positive, proving order preservation and injectivity. To see surjectivity, for choose the least integer with and put . Then , so this rational Cauchy sequence maps to . No choices are needed because each integer is uniquely specified. Finally an order-preserving map fixing must respect arbitrarily small rational bounds around each , so two such maps cannot differ in the Archimedean . This proves uniqueness.
Proof X-ray
Do not take a real limit before real numbers exist
The quotient is defined by rational statements. Its field and order are proved before suprema, and the supremum is itself a class of rational bisection bounds. Only after that construction do we use real sequence limits. Defining the initial quotient distance as an assumed real limit would conceal the very completeness being established.
Worked examples
Worked example 1.6.7
(The square-root gap is filled by a supremum)
Let now contain nonnegative reals with square below two, and let be its supremum. It lies between one and two. If is below two, a sufficiently small positive rational keeps below two, contradicting the upper bound. If exceeds two, choose so remains positive and its square exceeds two; then bounds , contradicting leastness. Such exist by expanding the quadratic and the Archimedean property. Hence . The earlier irrationality proof shows is a genuinely new class, not an embedded rational.
Worked example 1.6.8
(The recurring decimal is a class identity)
The finite decimal with nines is . Its difference from the constant-one sequence is null, so its class is exactly one. This gives a precise meaning to the recurring decimal: an equivalence class of approximations, not a separate positive number whose gap from one is smaller than every positive rational. The Archimedean property excludes such a positive gap.
Worked example 1.6.9
(Ordered fields need not be Archimedean)
Take the rational-function field , constructed as fractions of polynomials. A nonzero polynomial product has nonzero product leading coefficient, so the polynomial ring is an integral domain and its fraction field exists. Order a nonzero rational function by the sign of its leading-coefficient ratio at large positive . Multiplication preserves positive sign; after making denominators have positive leading coefficient, addition of two positives also has positive leading coefficient, so this is a compatible order. Here exceeds every natural constant . The naturals are bounded but have no supremum, by the argument. Thus ordered-field structure alone does not imply the completeness or uniqueness theorem's hypothesis.
Worked example 1.6.10
(The same number as a Dedekind cut)
A rational cut is a nonempty proper downward-closed subset of with no greatest element. Every real gives the cut : integer bounds give nonemptiness/properness and density excludes a greatest element. Conversely a cut has an upper bound in : any outside exceeds every member, since downward closure would otherwise put in . Let be its real supremum. If belongs to , choose a larger member of , proving . If , is not an upper bound of , so a larger member exists and downward closure puts in . Thus .
This gives a bijection between real Cauchy classes and rational cuts. It preserves strict order: a rational strictly between and belongs to but not when . The identification does not justify multiplying arbitrary cut elements to define cut multiplication. For instance contains every negative integer, whose pairwise products are unbounded above; the set of all such products is not a proper cut. One may transport the already proved field operations across the bijection, or separately prove correct cut formulas with the required sign cases. Matching the objects does not excuse an invalid operation formula.
Worked example 1.6.11
(A canonical approximation needs no choice)
For each and let be the least integer with , and set . The Archimedean integer-bound argument gives this unique integer even when is negative. Then , so the rational sequence is Cauchy and converges to . Every original representative also converges to its own class by the rational-tail estimate. The triangle inequality therefore makes null, so .
Consequently every real class has a uniquely specified rational Cauchy representative of this dyadic form. We have obtained a definable selector for this particular quotient without countable choice. The reason is not that all equivalence relations permit selectors: rational grids, integer well-order after a lower bound, and the proved order single out each approximation. The construction says nothing about computably deciding all the comparisons used to specify it.
Common mistakes
Counterexample mode
Termwise positivity is not positivity
Every term being positive does not make its class positive: a positive null sequence represents zero. A uniform eventual rational gap is essential. Claim: it would suffice to call positive when every term is positive. The sequence is rational, Cauchy and termwise positive, yet it is null, so . Under the proposed definition the class would be positive, contradicting trichotomy, which requires exactly one of , positive, positive. The proposal is not even representative independent: the constant sequence represents the same class and has no positive term. The repaired definition asks for one fixed positive rational with eventually, and no null sequence satisfies it.
Common mistake
(Ignoring infinitely many terms)
Ignoring a finite prefix is legitimate because it gives a null difference. Ignoring infinitely many large terms is not. The inverse construction changes only a finite prefix after proving a lower bound on the whole remaining tail.
Common mistake
(Bisection is not an algorithm)
Bisection asks a classical mathematical predicate about , not a necessarily computable test. No algorithm for comparing every represented real is claimed. Absence of countable choice here relies on rational canonical constructions, not on completeness of arbitrary metric spaces in ZF.
Summary
Rational Cauchy classes form a field through bounded products and tail inverses, and eventual gaps give order. Rational bisection proves supremum completeness without prior real limits or countable selections. Supremum completeness yields Cauchy convergence and the unique complete ordered-field structure, distinguishing from merely ordered fields.
Exercises
Quick check
(Finite prefixes of an inverse)
Explain why the finite-prefix zeros in an inverse sequence do not prevent its product with the original sequence from representing one.
Solution · Answer
After some the representative is nonzero and the inverse is its reciprocal, so the product equals one on that entire tail. Its difference from the constant-one sequence is supported in the finite prefix and is therefore null. Quotient equality ignores exactly that kind of difference; no assertion about an actual reciprocal of a zero prefix term was made.
Quick check
(A positive null sequence)
Give a sequence of strictly positive rationals representing zero, and state the exact stronger condition for a positive class.
Solution · Answer
The sequence is positive termwise and null, so represents zero. Positivity of instead requires one fixed positive rational with every sufficiently late at least . Null perturbations preserve this property after halving the bound, establishing its representative independence.
Quick check
(Where choice might enter)
Locate every choice in the supremum construction. Why does “ is not an upper bound” not require choosing a sequence from ?
Solution
We initially use a single existing and a supplied upper bound. An integer bound can be specified as the least positive integer meeting the two inequalities, and each midpoint decision uniquely determines the next rational pair. At the leastness step, for a fixed one gives a single witness . We do not assemble witnesses for all into a sequence. Hence the proof introduces no countable choice principle.
Quick check
(Uniqueness of limits)
Prove the limit of a Cauchy sequence is unique using rational bounds. Explain where the Archimedean/density property is used.
Solution
If distinct limits existed, density supplies positive rational with . Eventually a sequence term is within of each, so the triangle inequality gives , a contradiction. The ability to find a rational smaller than a positive field gap follows from the Archimedean property and is what makes rational convergence tests sufficient.
Quick check
(Bounded naturals)
Derive the contradiction when the embedded naturals of a Dedekind-complete ordered field are bounded. Why does not contradict the theorem?
Solution · Answer
A supposed supremum has below it, so some natural exceeds . Its successor then exceeds , contradicting that bounds all naturals. In , bounds the naturals but the field is not Dedekind complete, so there is no such . It fails a hypothesis rather than refuting the conclusion.