Work in ZF, without choice. Read safe set construction, ordinals and well-founded recursion. Arithmetic notation is introduced through the constructions below; a negative integer or fraction is not assumed to be a primitive set-language term. Completeness of the real numbers is not proved in this chapter.
Motivation
Infinity asserts the existence of an inductive set, but does not name a unique natural-number object. Subtraction and division introduce further problems: natural pairs may represent the same integer difference, and integer pairs may represent the same fraction. Quotients solve these problems only after the equivalence relations and operations have been proved compatible.
We will construct the objects in that order. First isolate the least inductive set and prove its induction principle. Then derive the arithmetic needed to justify the two quotients. The goal is not a particular coding of minus three; it is a number system whose laws and embeddings follow from explicit set constructions.
Definitions
Definition 1.5.1
(Inductive sets and )
Write and . A set is inductive if it contains zero and is closed under . Infinity supplies such an . Separation inside and its power set defines
Definition 1.5.2
(Addition and multiplication on )
Define natural operations by recursion on the second argument:
All values remain in . Comparisons use its ordinal order.
Definition 1.5.3
(Integers and rationals as quotients)
On put when . The integer set is , with class intended as the formal difference .
After its ordered-ring laws are established, put
The notation means this class; positive denominators are part of the definition. Both quotients exist by Separation and Replacement applied to subsets of a fixed set, without choosing a representative from each class.
Theorem/proposition
Theorem 1.5.4
(The least inductive set and recursive arithmetic)
is independent of the initial inductive set , is the least inductive set, and satisfies induction for every subset. Its elements are ordinals, every nonzero element is a successor, and successor is injective. Recursive addition and multiplication make it a commutative semiring, with additive cancellation and multiplicative cancellation by nonzero factors.
Theorem 1.5.5
(Integer differences form an ordered integral domain)
The integer relation is an equivalence relation. Operations
are well defined and make an ordered integral domain. The order is exactly when . Natural numbers embed by .
Concept lensStructural
Integers as the universal way to subtract
The pair construction is characterized by a property rather than by its coding. If is an abelian group and satisfies and , then is the unique additive map sending to . It is well defined because gives , and unique because . Any other construction with this property is isomorphic to by a unique additive bijection fixing the embedded naturals.
Theorem 1.5.6
(Rational fractions form the smallest ordered field)
The rational relation is an equivalence relation, and
make an ordered field. Every ordered field receives a unique order-preserving field embedding of sending one to one. No assertion that is order complete follows.
Proof sketch or proof idea
Proof
(Leastness, induction and the semiring laws)
Zero belongs to every inductive subset of . If belongs to all of them, so does , and still belongs to . Thus is inductive. For any other inductive , the intersection is an inductive subset of , so . This proves leastness and independence. If a subset of contains zero and is successor closed, leastness makes it all of , proving induction. The same argument applies to a definable property by Separation.
Induction shows every is an ordinal and : zero has both properties, and the ordinal successor preserves them. Thus is a transitive set of ordinals. The subset consisting of zero and all successors of elements is inductive and contained in , hence equals . Zero is not a successor, and ordinal successor is injective since . Well-founded recursion on now supplies the displayed operations as actual set functions, uniquely, rather than presupposing an infinite list of outputs.
Here are the arithmetic checks needed later. Induction on gives . Induction on gives and ; induction on using these identities gives commutativity. Addition cancels by induction on the common second summand, using successor injectivity at each step. Ordinal order satisfies exactly when for some : the forward direction follows by induction on using the successor description, and the reverse by successive addition. Hence addition preserves strict order.
For multiplication, induction on gives , rearranging finite sums with the proved addition laws. Induction on also gives ; using this and the zero case, induction on proves multiplication commutative. Consequently distributivity holds on either side. Induction on in now proves associativity by distributivity. The identities with zero and one follow directly from recursion. A product of positive naturals is positive: repeatedly adding a positive never gives zero. If , write with positive; then for positive , proving multiplicative cancellation. These checks establish the semiring and order facts used in the quotients.
Proof
(Integer operations respect the balance relation)
The integer relation is reflexive and symmetric. For transitivity, and give ; cancel to get . Adding equivalent pairs preserves equivalence by adding their balance equations. Negation exchanges the two sides. For products it suffices to change one representative. If , compare the two product pairs with a fixed . Their cross-sums agree because
Symmetry handles a change of the other factor.
Ring identities follow on representatives from these natural identities. Addition is componentwise, negation gives , and one is . Distributivity expands into the same four summands on both sides. In associating three product pairs, the positive component on both sides is and the negative component is , proving associativity explicitly. The integer embedding is injective by its balance equation. Every class has form or , since the natural coordinates are comparable. The proposed order is unchanged by equivalent representatives by adding their balance equations and canceling equal summands; it is total and invariant under addition. Its positive classes are exactly for positive, so products of positives remain positive. Taking signs reduces a product of two nonzero integers to that case, proving absence of zero divisors.
Proof
(Fractions form an ordered field with a universal embedding)
The fraction relation is reflexive and symmetric. If and , multiplying gives ; cancellation of nonzero in the integer domain gives . Thus it is transitive. If and , then
These prove addition and multiplication respect classes. Negation plainly does too. Associativity, commutativity and distributivity follow by putting both sides over the same positive product denominator and using the integer laws. Zero and one are and . If is nonzero, its inverse is when is positive, and when is negative; denominators remain positive and multiplication gives one. The integer embedding is injective. Cross-multiplication by positive denominators preserves inequalities, proving order independence, trichotomy and compatibility with addition and positive multiplication.
In any ordered field , repeated addition of its positive one never gives zero. Send a natural to copies of one, an integer difference to the corresponding difference, and a fraction to the corresponding quotient. The balance relations make these maps well defined. Positive integers and positive fractions map to positive elements, proving injectivity and order preservation. Field operations force all these values in any one-preserving embedding, proving uniqueness. This gives a precise universal meaning to rather than privileging its chosen pair coding.
Proof X-ray
A quotient needs laws before notation
Writing or does not yet permit arithmetic. Each construction first identifies equivalent representatives, proves the operations survive that identification, and only then transfers algebra and order. Integer cancellation is essential for rational transitivity; using fraction cancellation before that point would be circular.
The restriction to nonzero denominators is exactly what makes that cancellation available.
Counterexample mode
Allowing a zero denominator breaks transitivity
Suppose the fraction relation were defined by the same rule on all of . Then for every pair, because . In particular and , but , since . The relation is not transitive, so no quotient field arises. The transitivity proof fails at the step that cancels the middle denominator, here . Restricting denominators to nonzero integers, and in this construction to positive ones, repairs the definition.
Worked examples
Worked example 1.5.7
(A negative integer is a class, not a selected pair)
The pairs and represent the same integer because . Its additive inverse is . Multiplying by returns , while multiplying by gives . The intended minus sign has been proved compatible with the coding. No arbitrary selection of one pair per class was needed to define the integer set or its operations.
Worked example 1.5.8
(Different fraction calculations agree before reduction)
The sum is by the common-denominator formula, and this equals because . One may instead represent as ; the same formula then gives , which also equals because . The well-definedness calculation proves this agreement for every representative choice. A lowest-terms convention is useful for printing but is not part of the field construction.
Worked example 1.5.9
(Rational density leaves a square-root gap)
There is no rational with . Choose a representation with positive and least possible. If were odd its square would be odd, so makes even; writing then makes even as well. Dividing both by two contradicts minimality of . Parity follows directly by induction from the alternatives or .
The set is nonempty and bounded above by two, but has no rational supremum. An upper bound must be positive. If , choose a small positive rational so , making fail to be an upper bound. If , choose small enough that remains positive and its square still exceeds two; then is a smaller upper bound. Such choices may be made as for sufficiently large natural by the polynomial inequalities just stated. The equality case was excluded. Thus the ordered-field laws and rational density do not supply completeness.
Common mistakes
Common mistake
(Infinity alone does not give the least inductive set)
Infinity alone gives an inductive set, not the least one without further separation. Induction for was derived from leastness rather than assumed for an arbitrary inductive set.
Common mistake
(Equivalent pairs are different representatives)
Equivalent pairs are usually different pairs. A quotient operation must yield the same class, not necessarily the same displayed representative. Positive denominators remove an order-sign ambiguity but do not make fraction representatives unique.
Common mistake
(Choice-free is not the same as canonical or complete)
The construction is in ZF and uses no choice. It does not say all quotient sets have a canonical representative, nor does it make rational numbers complete merely because all permitted field operations exist.
Summary
Separation extracts and its induction principle from Infinity. Recursion and induction establish arithmetic, whose cancellation laws justify integer differences and rational fractions. The resulting ordered field embeds uniquely in every ordered field, while the square-root gap shows why a separate real-number construction is still needed.
Exercises
Quick check
(Independence of the Infinity witness)
Explain why intersecting with any other inductive proves independent of the chosen Infinity witness. Where is Separation used?
Solution · Answer
contains zero and is successor closed, so is an inductive subset of and must contain the defined . Hence lies in every inductive and is the unique least one. Power Set supplies the set of candidate subsets of ; Separation defines the satisfying the displayed universal condition inside , and also forms intersections and definable induction subsets. No intersection of an alleged set of all sets is used.
Quick check
(Associativity of integer products)
Verify the common positive and negative components in associating a product of three integer pairs.
Solution · Answer
First multiplying and , then , gives positive component and negative component . Expanding yields exactly the two sums in the proof. Grouping the last two pairs first gives those same sums by distributivity and commutativity. Thus even the actual expanded pairs agree after rearrangement, which is stronger than merely showing their classes equal.
Quick check
(A rational inverse)
Give the inverse of with positive denominator and verify its product is one using the equivalence relation.
Solution · Answer
The inverse is . Their product is , equal to because . Both denominators are positive. Writing would represent the same intended value only after extending a notation whose present definition requires positive denominators.
Quick check
(The characteristic of an ordered field)
Prove an ordered field has characteristic zero. Why can a finite field not carry an order compatible with field addition and multiplication?
Solution · Answer
One is positive in an ordered field, so induction makes every positive finite sum of ones positive and therefore nonzero. The characteristic is thus zero. In a finite field the additive multiples of one repeat, so a positive multiple is zero; such a compatible order would contradict the preceding positivity argument.
Quick check
(Least denominators)
Prove each rational class has a unique representative with positive and least among denominators of that class. Does this imply every quotient family has a choice-free selector?
Solution · Answer
Identify each positive integer denominator with the unique positive natural index of its embedded form . The resulting denominator-index set is a nonempty subset of , so it has a least element, giving the least denominator . For that fixed , equality of and gives , and cancellation gives . Thus the pair is unique and definable for this particular quotient. The proof uses its numerical well-order and algebraic cancellation; an arbitrary quotient need not provide either feature, so no general choice principle follows.