Set Theory

Ordinals

Reading preferences

Optional display controls need JavaScript. All reading content and navigation work without it.

Source file content/set-theory/ordinals/ordinals.tex

Source file content/set-theory/ordinals/introduction.tex

Introduction

In chapter “Steps towards Z”, we postulated that there is an infinite-th stage of the hierarchy, in the form of stagesinf (see also our axiom of Infinity). However, given stagessucc, we can't stop at the infinite-th stage; we have to keep going. So: at the next stage after the first infinite stage, we form all possible collections of sets that were available at the first infinite stage; and repeat; and repeat; and repeat; dots

Implicitly what has happened here is that we have started to invoke an “intuitive” notion of number, according to which there can be numbers after all the natural numbers. In particular, the notion involved is that of a transfinite ordinal. The aim of this chapter is to make this idea more rigorous. We will explore the general notion of an ordinal, and then explicitly define certain sets to be our ordinals.

Source file content/set-theory/ordinals/idea.tex

The General Idea of an Ordinal

Consider the natural numbers, in their usual order:

Natural numbers in their usual order

The diagram expands a source loop. In that loop, the current loop value successively takes the values zero through five. Each displayed value is less than the next. After five the diagram continues, and so on. Thus, from left to right: zero is less than one, one is less than two, two is less than three, three is less than four, four is less than five, and the sequence continues. End diagram.

Nodes

    Edges

      Elements

      • loop values. 0,1,2,3,4,5\xsourcerole: ordered values.
      • less than separators. <<sourcerole: relation separators; relation speech: is less than.
      • continuation. \ldotssourcerole: unbounded continuation.
      source 10

      We call this, in the jargon, an ω\omegasource-sequence. And indeed, this general ordering is mirrored in our initial construction of the stages of the set hierarchy. But, now suppose we move 00source to the end of this sequence, so that it comes after all the other numbers:

      Positive natural numbers followed by zero

      In the source loop, the current loop value successively takes the values one through five. Each loop value is ordered before the next. After five the positive-number sequence continues, and so on. The entire continuing sequence is ordered before the final zero. The resulting order has all positive natural numbers first and zero last. End diagram.

      Nodes

        Edges

          Elements

          • positive loop values. 1,2,3,4,5\xsourcerole: ordered values.
          • loop less than separators. <<sourcerole: relation separators; relation speech: is less than.
          • positive continuation. \ldotssourcerole: unbounded continuation.
          • final less than. <<sourcerole: relation separator; relation speech: is less than.
          • final zero. 00sourcerole: last value.
          source 24

          We have the same entities here, but ordered in a fundamentally different way: our first ordering had no last element; our new ordering does. Indeed, our new ordering consists of an ω\omegasource-sequence of entities (1,2,3,4,5,1, 2, 3, 4, 5, \ldotssource), followed by another entity. It will be an ω+1\omega+1source-sequence.

          We can generate even more types of ordering, using just these entities. For example, consider all the even numbers (in their natural order) followed by all the odd numbers (in their natural order):

          Even natural numbers followed by odd natural numbers

          The diagram reads left to right. The first displayed value is zero; it is ordered before two, which is ordered before four. Four is ordered before the later even natural numbers, and so on. The entire continuing even sequence is ordered before one, which is ordered before three. Three is ordered before the later odd natural numbers, and so on. It places every even natural number in increasing order before every odd natural number in increasing order. End diagram.

          Nodes

            Edges

              Elements

              • even sequence. role: first ordered sequence; displayed values: 0; 2; 4; continuation: yes.
              • odd sequence. role: second ordered sequence; displayed values: 1; 3; continuation: yes.
              • relation separators. role: less than between successive values; relation speech: is less than.
              source 45

              This is an ω\omegasource-sequence followed by another ω\omegasource-sequence; an ω+ω\omega+\omegasource-sequence.

              Well, we can keep going. But what we would like is a general way to understand this talk about orderings.

              Source file content/set-theory/ordinals/wo.tex

              Well-Orderings

              The fundamental notion is as follows:

              Definition one in this chapter

              The relation <<source well-orders AAsource iff it meets these two conditions:

              1. <<source is connected, i.e., for all a,bAa, b \in Asource, either a<ba < bsource or a=ba = bsource or b<ab < asource;

              2. every non-empty subset of AAsource has a <<source-minimal element, i.e., if XA\emptyset \neq X \subseteq Asource then (mX)(zX)zm(\exists m \in X)(\forall z \in X)z \nless msource

              It is easy to see that three examples we just considered were indeed well-ordering relations.

              Exercise one in this chapter

              Section “The General Idea of an Ordinal” in chapter “Ordinals” presented three example orderings on the natural numbers. Check that each is a well-ordering.

              Here are some elementary but extremely important observations concerning well-ordering.

              Proposition one in this chapter

              If <<source well-orders AAsource, then every non-empty subset of AAsource has a unique <<source-least member, and <<source is irreflexive, asymmetric and transitive.

              Proof

              If XXsource is a non-empty subset of AAsource, it has a <<source-minimal element mmsource, i.e., (zX)zm(\forall z \in X)z \nless msource. Since <<source is connected, (zX)mz(\forall z \in X)m \leq zsource. So mmsource is the <<source-least element of XXsource.

              For irreflexivity, fix aAa \in Asource; the <<source-least element of {a}\{a\}source is aasource, so aaa \nless asource. For transitivity, if a<b<ca < b < csource, then since {a,b,c}\{a, b, c\}source has a <<source-least element, a<ca < csource. Asymmetry follows from irreflexivity and transitivity

              Proposition two in this chapter

              If <<source well-orders AAsource, then for any formula ϕ(x)\phi(x)source:

              if (aA)((b<a)ϕ(b)ϕ(a)), then (aA)ϕ(a).\text{if }(\forall a \in A)((\forall b < a)\phi(b) \lif \phi(a))\text{, then }(\forall a \in A)\phi(a).source

              Proof

              We will prove the contrapositive. Suppose ¬(aA)ϕ(a)\lnot(\forall a \in A)\phi(a)source, i.e., that X={xA:¬ϕ(x)}X = \Setabs{x \in A}{\lnot\phi(x)} \neq \emptysetsource. Then XXsource has an <<source-minimal element, aasource. So (b<a)ϕ(b)(\forall b < a)\phi(b)source but ¬ϕ(a)\lnot \phi(a)source.

              noindent This last property should remind you of the principle of strong induction on the naturals, i.e.: if (nω)((m<n)ϕ(m)ϕ(n))(\forall n \in \omega)((\forall m < n)\phi(m) \lif \phi(n))source, then (nω)ϕ(n)(\forall n \in \omega)\phi(n)source. And this property makes well-ordering into a very robust notion.Footnote: A reminder: all formulas can have parameters (unless explicitly stated otherwise).

              Source file content/set-theory/ordinals/iso.tex

              Order-Isomorphisms

              To explain how robust well-ordering is, we will start by introducing a method for comparing well-orderings.

              Definition two in this chapter

              A well-ordering is a pair A,<\tuple{A, <}source, such that <<source well-orders AAsource. The well-orderings A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source are order-isomorphic iff there is a bijection f:ABf \colon A \to Bsource such that: x<yx < ysource iff f(x)f(y)f(x) \lessdot f(y)source. In this case, we write A,<B,\ordeq{\tuple{A, <}}{\tuple{B, \lessdot}}source, and say that ffsource is an order-isomorphism.

              noindent In what follows, for brevity, we will speak of “isomorphisms” rather than “order-isomorphisms”. Intuitively, isomorphisms are structure-preserving bijections. Here are some simple facts about isomorphisms.

              Lemma one in this chapter

              Compositions of isomorphisms are isomorphisms, i.e.: if f:ABf \colon A \to Bsource and g:BCg \colon B \to Csource are isomorphisms, then (gf):AC(g \circ f) \colon A \to Csource is an isomorphism.

              Exercise two in this chapter

              Prove lemma one in chapter “Ordinals”.

              Proof

              Left as an exercise.

              Corollary one in this chapter

              XY\ordeq{X}{Y}source is an equivalence relation.

              Proposition three in this chapter

              If A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source are isomorphic well-orderings, then the isomorphism between them is unique.

              Proof

              Let ffsource and ggsource be isomorphisms ABA \to Bsource. We will prove the result by induction, i.e.\ using proposition two in chapter “Ordinals”. Fix aAa\in Asource, and suppose (for induction) that (b<a)f(b)=g(b)(\forall b < a)f(b) = g(b)source. Fix xBx \in Bsource.

              If xf(a)x \lessdot f(a)source, then f1(x)<af^{-1}(x) < asource, so g(f1(x))g(a)g(f^{-1}(x)) \lessdot g(a)source, invoking the fact that ffsource and ggsource are isomorphisms. But since f1(x)<af^{-1}(x) < asource, by our supposition x=f(f1(x))=g(f1(x))x =f(f^{-1}(x)) = g(f^{-1}(x))source. So xg(a)x \lessdot g(a)source. Similarly, if xg(a)x \lessdot g(a)source then xf(a)x \lessdot f(a)source.

              Generalising, (xB)(xf(a)xg(a))(\forall x \in B)(x \lessdot f(a) \liff x \lessdot g(a))source. It follows that f(a)=g(a)f(a) = g(a)source by the proposition on extensionality strictlinearorders. So (aA)f(a)=g(a)(\forall a \in A)f(a) = g(a)source by proposition two in chapter “Ordinals”.

              noindent This gives some sense that well-orderings are robust. But to continue explaining this, it will help to introduce some more notation.

              Definition three in this chapter

              When A,<\tuple{A, <}source is a well-ordering with aAa \in Asource, let Aa={xA:x<a}A_a = \Setabs{x \in A}{x < a}source. We say that AaA_asource is a proper initial segment of AAsource (and allow that AAsource itself is an improper initial segment of AAsource). Let <a<_asource be the restriction of <<source to the initial segment, i.e., <Aa2\funrestrictionto{\mathord{<}}{A_a^2}source.

              noindent Using this notation, we can state and prove that no well-ordering is isomorphic to any of its proper initial segments.

              Lemma two in this chapter

              If A,<\tuple{A, <}source is a well-ordering with aAa \in Asource, then A,<Aa,<a\ordneq{\tuple{A, <}}{\tuple{A_a, <_a}}source

              Proof

              For reductio, suppose f:AAaf \colon A \to A_asource is an isomorphism. Since ffsource is a bijection and AaAA_a \subsetneq Asource, using proposition one in chapter “Ordinals” let bAb \in Asource be the <<source-least element of AAsource such that bf(b)b \neq f(b)source. We'll show that (xA)(x<bx<f(b))(\forall x \in A)(x<b \liff x < f(b))source, from which it will follow by the proposition on extensionality strictlinearorders that b=f(b)b = f(b)source, completing the reductio.

              Suppose x<bx < bsource. So x=f(x)x = f(x)source, by the choice of bbsource. And f(x)<f(b)f(x) < f(b)source, as ffsource is an isomorphism. So x<f(b)x < f(b)source.

              Suppose x<f(b)x < f(b)source. So f1(x)<bf^{-1}(x) < bsource, since ffsource is an isomorphism, and so f1(x)=xf^{-1}(x) = xsource by the choice of bbsource. So x<bx < bsource.

              Our next result shows, roughly put, that an “initial segment” of an isomorphism is an isomorphism:

              Lemma three in this chapter

              Let A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source be well-orderings. If f:ABf \colon A \to Bsource is an isomorphism and aAa \in Asource, then fAa:AaBf(a)\funrestrictionto{f}{A_{a}} : A_a \to B_{f(a)}source is an isomorphism.

              Proof

              Since ffsource is an isomorphism:

              f[Aa]=f[{xA:x<a}]=f[{f1(y)A:f1(y)<a}]={yB:yf(a)}=Bf(a)\funimage{f}{A_a} &= \funimage{f}{\Setabs{x \in A}{x < a}}\\ &= \funimage{f}{\Setabs{f^{-1}(y) \in A}{f^{-1}(y) < a}} \\ &= \Setabs{y \in B}{y \lessdot f(a)} \\ &=B_{f(a)}source

              And fAa\funrestrictionto{f}{A_a}source preserves order because ffsource does.

              Our next two results establish that well-orderings are always comparable:

              Lemma four in this chapter

              Let A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source be well-orderings. If Aa1,<a1Bb1,b1\ordeq{\tuple{A_{a_1}, <_{a_1}}}{\tuple{B_{b_1}, \lessdot_{b_1}}}source and Aa2,<a2Bb2,b2\ordeq{\tuple{A_{{a_2}}, <_{a_2}}}{\tuple{B_{{b_2}}, \lessdot_{b_2}}}source, then a1<a2 iff b1b2{a_1} < {a_2} \text{ iff }{b_1} \lessdot {b_2}source

              Proof

              We will prove left to right; the other direction is similar. Suppose both Aa1,<a1Bb1,b1\ordeq{\tuple{A_{a_1}, <_{a_1}}}{\tuple{B_{b_1}, \lessdot_{b_1}}}source and Aa2,<a2Bb2,b2\ordeq{\tuple{A_{{a_2}}, <_{a_2}}}{\tuple{B_{{b_2}}, \lessdot_{b_2}}}source, with f:Aa2Bb2f \colon A_{{a_2}} \to B_{{b_2}}source our isomorphism. Let a1<a2{a_1} < {a_2}source; then Aa1,<a1Bf(a1),f(a1)\ordeq{\tuple{A_{a_1}, <_{a_1}}}{\tuple{B_{f({a_1})}, \lessdot_{f({a_1})}}}source by lemma three in chapter “Ordinals”. So Bb1,b1Bf(a1),f(a1)\ordeq{\tuple{B_{b_1}, \lessdot_{b_1}}}{\tuple{B_{f({a_1})}, \lessdot_{f({a_1})}}}source, and so b1=f(a1){b_1} = f({a_1})source by lemma two in chapter “Ordinals”. Now b1b2{b_1} \lessdot {b_2}source as ffsource's domain is Bb2B_{{b_2}}source.

              Theorem one in this chapter

              Given any two well-orderings, one is isomorphic to an initial segment (not necessarily proper) of the other.

              Proof

              Let A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source be well-orderings. Using Separation, let

              f={a,bA×B:Aa,<aBb,b}.f = \Setabs{\tuple{a, b} \in A \times B}{ \ordeq{\tuple{A_a, <_a}}{\tuple{B_b, \lessdot_b}}}.source

              By lemma four in chapter “Ordinals”, a1<a2a_1 < a_2source iff b1b2b_1 \lessdot b_2source for all a1,b1,a2,b2f\tuple{a_1, b_1}, \tuple{a_2, b_2} \in fsource. So f:dom(f)ran(f)f \colon \dom{f} \to \ran{f}source is an isomorphism.

              If a2dom(f)a_2 \in \dom{f}source and a1<a2a_1 < a_2source, then a1dom(f)a_1 \in \dom{f}source by lemma three in chapter “Ordinals”; so dom(f)\dom{f}source is an initial segment of AAsource. Similarly, ran(f)\ran{f}source is an initial segment of BBsource. For reductio, suppose both are proper initial segments. Then let aasource be the <<source-least element of Adom(f)A \setminus \dom{f}source, so that dom(f)=Aa\dom{f} = A_asource, and let bbsource be the \lessdotsource-least element of Bran(f)B \setminus \ran{f}source, so that ran(f)=Bb\ran{f} = B_bsource. So f:AaBbf \colon A_a \to B_bsource is an isomorphism, and hence a,bf\tuple{a, b} \in fsource, a contradiction.

              Source file content/set-theory/ordinals/vn.tex

              Von Neumann's Construction of the Ordinals

              the theorem on woalwayscomparable gives rise to a thought. We could introduce certain objects, called order types, to go proxy for the well-orderings. Writing ord(A,<)\ordtype{A, <}source for the order type of the well-ordering A,<\tuple{A, <}source, we would hope to secure the following two principles:

              ord(A,<)=ord(B,) iff A,<B,ord(A,<)<ord(B,) iff A,<Bb,b for some bB\ordtype{A, <} = \ordtype{B, \lessdot} & \text{ iff } \ordeq{\tuple{A, <}}{\tuple{B, \lessdot}}\\ \ordtype{A, <} < \ordtype{B, \lessdot}& \text{ iff }\ordeq{\tuple{A, <}}{\tuple{B_b, \lessdot_b}}\text{ for some }b \in Bsource

              Moreover, we might hope to introduce order-types as certain sets, just as we can introduce the natural numbers as certain sets.

              The most common way to do this---and the approach we will follow---is to define these order-types via certain canonical well-ordered sets. These canonical sets were first introduced by von Neumann:

              Definition four in this chapter

              The set AAsource is transitive iff (xA)xA(\forall x \in A)x \subseteq Asource. Then AAsource is an ordinal iff AAsource is transitive and well-ordered by \insource.

              noindent In what follows, we will use Greek letters for ordinals. It follows immediately from the definition that, if α\alphasource is an ordinal, then α,α\tuple{\alpha, \in_\alpha}source is a well-ordering, where α={x,yα2:xy}\in_\alpha = \Setabs{\tuple{x, y} \in \alpha^2}{x \in y}source. So, abusing notation a little, we can just say that α\alphasource itself is a well-ordering.

              Here are our first few ordinals:

              ,{},{,{}},{,{},{,{}}},\emptyset, \{\emptyset\}, \{\emptyset, \{\emptyset\}\}, \{\emptyset, \{\emptyset\}, \{\emptyset, \{\emptyset\}\}\}, \ldotssource

              You will note that these are the first few ordinals that we encountered in our Axiom of Infinity, i.e., in von Neumann's definition of ω\omegasource (see section “Infinity” in chapter “Steps towards Z”). This is no coincidence. Von Neumann's definition of the ordinals treats natural numbers as ordinals, but allows for transfinite ordinals too.

              As always, we can now ask: are these the ordinals? Or has von Neumann simply given us some sets that we can treat as the ordinals? The kinds of discussions one might have about this question are similar to the discussions we had in section “Philosophical Reflections” in chapter “Relations”, section “Some Philosophical Reflections” in chapter “Arithmetization”, section “Dedekind's “Proof” of the Existence of an Infinite Set” in chapter “Infinite Sets”, and section “Selecting our Natural Numbers” in chapter “Steps towards Z”, so we will not belabour the point. Instead, in what follows, we will simply use “the ordinals” to speak of “the von Neumann ordinals”.

              Source file content/set-theory/ordinals/basic.tex

              Basic Properties of the Ordinals

              We observed that the first few ordinals are the natural numbers. The main reason for developing a theory of ordinals is to extend the principle of induction which holds on the natural numbers. We will build up to this via a sequence of elementary results.

              Lemma five in this chapter

              Every element of an ordinal is an ordinal.

              Proof

              Let α\alphasource be an ordinal with bαb \in \alphasource. Since α\alphasource is transitive, bαb \subseteq \alphasource. So \insource well-orders bbsource as \insource well-orders α\alphasource.

              To see that bbsource is transitive, suppose xcbx \in c \in bsource. So cαc \in \alphasource as bαb \subseteq \alphasource. Again, as α\alphasource is transitive, cαc \subseteq \alphasource, so that xαx \in \alphasource. So x,c,bαx, c, b \in \alphasource. But \insource well-orders α\alphasource, so that \insource is a transitive relation on α\alphasource by proposition one in chapter “Ordinals”. So since xcbx \in c \in bsource, we have xbx \in bsource. Generalising, cbc \subseteq bsource

              Corollary two in this chapter

              α={βα:β is an ordinal}\alpha = \Setabs{\beta \in \alpha}{\beta \text{ is an ordinal}}source, for any ordinal α\alphasource

              Proof

              Immediate from lemma five in chapter “Ordinals”.

              The rough gist of the next two main results, theorem “Transfinite Induction” in chapter “Ordinals” and theorem “Trichotomy” in chapter “Ordinals”, is that the ordinals themselves are well-ordered by membership:

              Theorem: Transfinite Induction

              [Transfinite Induction] For any formula ϕ(x)\phi(x)source:

              if αϕ(α), then α(ϕ(α)(βα)¬ϕ(β))\text{if }\exists \alpha \phi(\alpha)\text{, then }\exists \alpha(\phi(\alpha) \land (\forall \beta \in \alpha) \lnot \phi(\beta))source

              where the displayed quantifiers are implicitly restricted to ordinals.

              Proof

              Suppose ϕ(α)\phi(\alpha)source, for some ordinal α\alphasource. If (βα)¬ϕ(β)(\forall \beta \in \alpha) \lnot \phi(\beta)source, then we are done. Otherwise, as α\alphasource is an ordinal, it has some \insource-least element which is ϕ\phisource, and this is an ordinal by lemma five in chapter “Ordinals”.

              noindent Note that we can equally express theorem “Transfinite Induction” in chapter “Ordinals” as the scheme:

              if α((βα)ϕ(β)ϕ(α)), then αϕ(α)\text{if }\forall \alpha((\forall \beta \in \alpha)\phi(\beta) \lif \phi(\alpha))\text{, then }\forall \alpha\phi(\alpha)source

              just by taking ¬ϕ(α)\lnot\phi(\alpha)source in theorem “Transfinite Induction” in chapter “Ordinals”, and then performing elementary logical manipulations.

              Theorem: Trichotomy

              [Trichotomy] αβα=ββα\alpha \in \beta \lor \alpha = \beta \lor \beta \in \alphasource, for any ordinals α\alphasource and β\betasource.

              Proof

              The proof is by double induction, i.e., using theorem “Transfinite Induction” in chapter “Ordinals” twice. Say that xxsource is comparable with yysource iff xyx=yyxx \in y \lor x = y \lor y \in xsource.

              For induction, suppose that every ordinal in α\alphasource is comparable with every ordinal. For further induction, suppose that α\alphasource is comparable with every ordinal in β\betasource. We will show that α\alphasource is comparable with β\betasource. By induction on β\betasource, it will follow that α\alphasource is comparable with every ordinal; and so by induction on α\alphasource, every ordinal is comparable with every ordinal, as required. It suffices to assume that αβ\alpha \notin \betasource and βα\beta \notin \alphasource, and show that α=β\alpha = \betasource.

              To show that αβ\alpha \subseteq \betasource, fix γα\gamma \in \alphasource; this is an ordinal by lemma five in chapter “Ordinals”. So by the first induction hypothesis, γ\gammasource is comparable with β\betasource. But if either γ=β\gamma = \betasource or βγ\beta \in \gammasource then βα\beta \in \alphasource (invoking the fact that α\alphasource is transitive if necessary), contrary to our assumption; so γβ\gamma \in \betasource. Generalising, αβ\alpha \subseteq \betasource.

              Exactly similar reasoning, using the second induction hypothesis, shows that βα\beta \subseteq \alphasource. So α=β\alpha = \betasource.

              noindent As such, we will sometimes write α<β\alpha <\betasource rather than αβ\alpha \in \betasource, since \insource is behaving as an ordering relation. There are no deep reasons for this, beyond familiarity, and because it is easier to write αβ\alpha \leq \betasource than αβα=β\alpha \in \beta \lor \alpha = \betasource.Footnote: We could write α¯β\alpha \mathrel{\underline{\in}} \betasource; but that would be wholly non-standard.

              Here are two quick consequences of our last results, the first of which puts our new notation into action:

              Corollary three in this chapter

              If αϕ(α)\exists \alpha\phi(\alpha)source, then α(ϕ(α)β(ϕ(β)αβ))\exists \alpha(\phi(\alpha) \land \forall \beta(\phi(\beta) \lif \alpha \leq \beta))source. Moreover, for any ordinals α,β,γ\alpha, \beta, \gammasource, both αα\alpha \notin \alphasource and αβγαγ\alpha \in \beta \in \gamma \lif \alpha \in \gammasource.

              Proof

              Just like proposition one in chapter “Ordinals”.

              Exercise three in this chapter

              Complete the “exactly similar reasoning” in the proof of theorem “Trichotomy” in chapter “Ordinals”.

              Corollary four in this chapter

              AAsource is an ordinal iff AAsource is a transitive set of ordinals.

              Proof

              Left-to-right. By lemma five in chapter “Ordinals”. Right-to-left. If AAsource is a transitive set of ordinals, then \insource well-orders AAsource by theorem “Transfinite Induction” in chapter “Ordinals” and theorem “Trichotomy” in chapter “Ordinals”.

              Now, we glossed theorem “Transfinite Induction” in chapter “Ordinals” and theorem “Trichotomy” in chapter “Ordinals” as telling us that \insource well-orders the ordinals. However, we have to be very cautious about this sort of claim, thanks to the following result:

              Theorem: Burali-Forti Paradox

              [Burali-Forti Paradox] There is no set of all the ordinals

              Proof

              For reductio, suppose OOsource is the set of all ordinals. If αβO\alpha \in \beta \in Osource, then α\alphasource is an ordinal, by lemma five in chapter “Ordinals”, so αO\alpha \in Osource. So OOsource is transitive, and hence OOsource is an ordinal by corollary four in chapter “Ordinals”. Hence OOO \in Osource, contradicting corollary three in chapter “Ordinals”.

              noindent This result is named after Cesare Burali-Forti. But, it was Cantor in 1899---in a letter to Dedekind---who first saw clearly the contradiction in supposing that there is a set of all the ordinals. As van Heijenoort explains:

              Burali-Forti himself considered the contradiction as establishing, by reductio ad absurdum, the result that the natural ordering of the ordinals is just a partial ordering. (Jean van Heijenoort, 1967, p. 105)

              Setting Burali-Forti's mistake to one side, we can summarize the foregoing as follows. Ordinals are sets which are individually well-ordered by membership, and collectively well-ordered by membership (without collectively constituting a set).

              Rounding this off, here are some more basic properties about the ordinals which follow from theorem “Transfinite Induction” in chapter “Ordinals” and theorem “Trichotomy” in chapter “Ordinals”.

              Proposition four in this chapter

              Any strictly descending sequence of ordinals is finite.

              Proof

              Any infinite strictly descending sequence of ordinals α0>α1>α2>\alpha_0 > \alpha_1 > \alpha_2 > \ldotssource has no <<source-minimal member, contradicting theorem “Transfinite Induction” in chapter “Ordinals”.

              Proposition five in this chapter

              αββα\alpha \subseteq \beta \lor \beta \subseteq \alphasource, for any ordinals α,β\alpha, \betasource.

              Proof

              If αβ\alpha \in \betasource, then αβ\alpha \subseteq \betasource as β\betasource is transitive. Similarly, if βα\beta \in \alphasource, then βα\beta \subseteq \alphasource. And if α=β\alpha = \betasource, then αβ\alpha \subseteq \betasource and βα\beta \subseteq \alphasource. So by theorem “Trichotomy” in chapter “Ordinals” we are done.

              Proposition six in this chapter

              α=β\alpha = \betasource iff αβ\ordeq{\alpha}{\beta}source, for any ordinals α,β\alpha, \betasource.

              Proof

              The ordinals are well-orders; so this is immediate from Trichotomy (theorem “Trichotomy” in chapter “Ordinals”) and lemma two in chapter “Ordinals”.

              Exercise four in this chapter

              Prove that, if every member of XXsource is an ordinal, then X\bigcup Xsource is an ordinal.

              Source file content/set-theory/ordinals/replacement.tex

              Replacement

              In section “Von Neumann's Construction of the Ordinals” in chapter “Ordinals”, we motivated the introduction of ordinals by suggesting that we could treat them as order-types, i.e., canonical proxies for well-orderings. In order for that to work, we would need to prove that every well-ordering is isomorphic to some ordinal. This would allow us to define ord(A,<)\ordtype{A, <}source as the ordinal α\alphasource such that A,<α\tuple{A, <} \isomorphic \alphasource.

              Unfortunately, we cannot prove the desired result only the Axioms we provided introduced so far. (We will see why in section “The Strength of Replacement” in chapter “Replacement”, but for now the point is: we can't.) We need a new thought, and here it is:

              Axiom: Scheme of Replacement

              [Scheme of Replacement] For any formula ϕ(x,y)\phi(x, y)source, the following is an axiom:

              for any AAsource, if (xA)∃!yϕ(x,y)(\forall x \in A)\lexists![y][\phi(x,y)]source, then {y:(xA)ϕ(x,y)}\Setabs{y}{(\exists x \in A)\phi(x,y)}source exists.

              noindent As with Separation, this is a scheme: it yields infinitely many axioms, for each of the infinitely many different ϕ\phisource's. And it can equally well be (and normally is) written down thus:

              Definition five in this chapter

              For any formula ϕ(x,y)\phi(x,y)source which does not contain “BBsource”, the following is an axiom:

              A[(xA)∃!yϕ(x,y)By(yB(xA)ϕ(x,y))]\forall A[(\forall x \in A)\lexists![y][\phi(x,y)] \lif \exists B\forall y (y \in B \liff (\exists x \in A)\phi(x,y))]source

              On first encounter, however, this is quite a tangled formula. The following quick consequence of Replacement probably gives a clearer expression to the intuitive idea we are working with:Footnote: A term is an expression which picks out exactly one object (on any completion of its free variables). For example, “\emptysetsource” is a term which picks out the empty set; “{x}\{x\}source” is a term which picks out xxsource's singleton (whatever xxsource might be); “xyx \cup ysource” is a term which picks out the union of xxsource and yysource (whatever they might be).

              Corollary five in this chapter

              For any term τ(x)\tau(x)source, and any set AAsource, this set exists:

              {τ(x):xA}={y:(xA)y=τ(x)}.\Setabs{\tau(x)}{x \in A} = \Setabs{y}{(\exists x \in A)y = \tau(x)}.source

              Proof

              Since τ\tausource is a term, x∃!yτ(x)=y\forall x \lexists![y][\tau(x) = y]source. A fortiori, (xA)∃!yτ(x)=y(\forall x \in A)\lexists![y][\tau(x) = y]source. So {y:(xA)τ(x)=y}\Setabs{y}{(\exists x \in A)\tau(x) = y}source exists by Replacement.

              noindent This suggests that “Replacement” is a good name for the Axiom: given a set AAsource, you can form a new set, {τ(x):xA}\Setabs{\tau(x)}{x \in A}source, by replacing every member of AAsource with its image under τ\tausource. Indeed, following the notation for the image of a set under a function, we might write τ[A]\funimage{\tau}{A}source for {τ(x):xA}\Setabs{\tau(x)}{x \in A}source.

              Crucially, however, τ\tausource is a term. It need not be (a name for) a function, in the sense of section “Functions as Relations” in chapter “Functions”, i.e., a certain set of ordered pairs. After all, if ffsource is a function (in that sense), then the set f[A]={f(x):xA}\funimage{f}{A} = \Setabs{f(x)}{x \in A}source is just a particular subset of ran(f)\ran{f}source, and that is already guaranteed to exist, just using the axioms of Z\Zminussource.Footnote: Just consider {yf:(xA)y=f(x)}\Setabs{y \in \bigcup \bigcup f}{(\exists x \in A)y = f(x)}source. Replacement, by contrast, is a powerful addition to our axioms, as we will see in chapter “Replacement”.

              Source file content/set-theory/ordinals/milestone.tex

              ZF\ZFminussource: a milestone

              The question of how to justify Replacement (if at all) is not straightforward. As such, we will reserve that for chapter “Replacement”. However, with the addition of Replacement, we have reached another important milestone. We now have all the axioms required for the theory ZF\ZFminussource. In detail:

              Definition six in this chapter

              The theory ZF\ZFminussource has these axioms: Extensionality, Union, Pairs, Powersets, Infinity, and all instances of the Separation and Replacement schemes. Otherwise put, ZF\ZFminussource adds Replacement to Z\Zminussource.

              noindent This stands for Zermelo--Fraenkel set theory (minus something which we will come to later). Fraenkel gets the honour, since he is credited with the formulation of Replacement in 1922, although the first precise formulation was due to Thoralf Skolem (1922).

              Source file content/set-theory/ordinals/ordtype.tex

              Ordinals as Order-Types

              Armed with Replacement, and so now working in ZF\ZFminussource, we can finally prove the result we have been aiming for:

              Theorem five in this chapter

              Every well-ordering is isomorphic to a unique ordinal.

              Proof

              Let A,<\tuple{A, <}source be a well-order. By proposition six in chapter “Ordinals”, it is isomorphic to at most one ordinal. So, for reductio, suppose A,<\tuple{A, <}source is not isomorphic to any ordinal. We will first “make A,<\tuple{A, <}source as small as possible”. In detail: if some proper initial segment Aa,<a\tuple{A_a, <_a}source is not isomorphic to any ordinal, there is a least aAa \in Asource with that property; then let B=AaB = A_asource and =<a\mathord{\lessdot} = \mathord{<_a}source. Otherwise, let B=AB = Asource and =<\mathord{\lessdot} = \mathord{<}source.

              By definition, every proper initial segment of BBsource is isomorphic to some ordinal, which is unique as above. So by Replacement, the following set exists, and is a function:

              f={β,b:bB and βBb,b}f = \Setabs{\tuple{\beta, b}}{b \in B\text{ and }\ordeq{\beta}{\tuple{B_b, \lessdot_b}}}source

              To complete the reductio, we'll show that ffsource is an isomorphism αB\alpha \to Bsource, for some ordinal α\alphasource.

              It is obvious that ran(f)=B\ran{f} = Bsource. And by lemma four in chapter “Ordinals”, ffsource preserves ordering, i.e., γβ\gamma \in \betasource iff f(γ)f(β)f(\gamma) \lessdot f(\beta)source. To show that dom(f)\dom{f}source is an ordinal, by corollary four in chapter “Ordinals” it suffices to show that dom(f)\dom{f}source is transitive. So fix βdom(f)\beta \in \dom{f}source, i.e., βBb,b\ordeq{\beta}{\tuple{B_b, \lessdot_b}}source for some bbsource. If γβ\gamma \in \betasource, then γdom(f)\gamma \in \dom{f}source by lemma three in chapter “Ordinals”; generalising, βdom(f)\beta \subseteq \dom{f}source.

              This result licenses the following definition, which we have wanted to offer since section “Von Neumann's Construction of the Ordinals” in chapter “Ordinals”:

              Definition seven in this chapter

              If A,<\tuple{A, <}source is a well-ordering, then its order type, ord(A,<)\ordtype{A, <}source, is the unique ordinal α\alphasource such that A,<α\ordeq{\tuple{A, < }}{\alpha}source.

              Moreover, this definition licenses two nice principles:

              Corollary six in this chapter

              Where A,<\tuple{A, <}source and B,\tuple{B, \lessdot}source are well-orderings:

              ord(A,<)=ord(B,) iff A,<B,ord(A,<)ord(B,) iff A,<Bb,b for some bB\ordtype{A, <} = \ordtype{B, \lessdot}&\text{ iff }\ordeq{\tuple{A, <}}{\tuple{B, \lessdot}}\\ \ordtype{A, <} \in \ordtype{B, \lessdot}&\text{ iff }\ordeq{\tuple{A, <}}{\tuple{B_b, \lessdot_b}}\text{ for some }b \in Bsource

              Proof

              The identity holds by proposition six in chapter “Ordinals”. To prove the second claim, let ord(A,<)=α\ordtype{A, <} = \alphasource and ord(B,)=β\ordtype{B, \lessdot} = \betasource, and let f:βB,f \colon \beta \to \tuple {B, \lessdot}source be our isomorphism. Then:

              αβ iff fα:αBf(α) is an isomorphism iff A,<Bf(α),f(α) iff A,<Bb,b for some bB\alpha \in \beta&\text{ iff }\funrestrictionto{f}{\alpha} \colon \alpha \to B_{f(\alpha)}\text{ is an isomorphism}\\ &\text{ iff }\ordeq{\tuple{A, <}}{\tuple{B_{f(\alpha)}, \lessdot_{f(\alpha)}}}\\ &\text{ iff }\ordeq{\tuple{A, <}}{\tuple{B_b, \lessdot_b}}\text{ for some $b \in B$}source

              by proposition three in chapter “Ordinals”, lemma three in chapter “Ordinals”, and corollary two in chapter “Ordinals”.

              Source file content/set-theory/ordinals/opps.tex

              Successor and Limit Ordinals

              In the next few chapters, we will use ordinals a great deal. So it will help if we introduce some simple notions.

              Definition eight in this chapter

              For any ordinal α\alphasource, its successor is α+=α{α}\ordsucc{\alpha} =\alpha \cup \{\alpha\}source. We say that α\alphasource is a successor ordinal if β+=α\ordsucc{\beta} = \alphasource for some ordinal β\betasource. We say that α\alphasource is a limit ordinal iff α\alphasource is neither empty nor a successor ordinal.

              noindent The following result shows that this is the right notion of successor:

              Proposition seven in this chapter

              For any ordinal α\alphasource:

              1. αα+\alpha \in \ordsucc{\alpha}source;

              2. α+\ordsucc{\alpha}source is an ordinal;

              3. there is no ordinal β\betasource such that αβα+\alpha \in \beta \in \ordsucc{\alpha}source.

              Proof

              Trivially, αα{α}=α+\alpha \in \alpha \cup \{\alpha\} = \ordsucc{\alpha}source. Equally, α+\ordsucc{\alpha}source is a transitive set of ordinals, and hence an ordinal by corollary four in chapter “Ordinals”. And it is impossible that αβα+\alpha \in \beta \in \ordsucc{\alpha}source, since then either βα\beta \in \alphasource or β=α\beta = \alphasource, contradicting corollary three in chapter “Ordinals”.

              noindent This also licenses a variant of proof by transfinite induction:

              Theorem: Simple Transfinite Induction

              [Simple Transfinite Induction] Let ϕ(x)\phi(x)source be a formula such that:

              1. ϕ()\phi(\emptyset)source; and

              2. for any ordinal α\alphasource, if ϕ(α)\phi(\alpha)source then ϕ(α+)\phi(\ordsucc{\alpha})source; and

              3. if α\alphasource is a limit ordinal and (βα)ϕ(β)(\forall \beta \in \alpha)\phi(\beta)source, then ϕ(α)\phi(\alpha)source.

              Then αϕ(α)\forall \alpha \phi(\alpha)source.

              Proof

              We prove the contrapositive. So, suppose there is some ordinal which is ¬ϕ\lnot\phisource; let γ\gammasource be the least such ordinal. Then either γ=\gamma = \emptysetsource, or γ=α+\gamma = \ordsucc{\alpha}source for some α\alphasource such that ϕ(α)\phi(\alpha)source; or γ\gammasource is a limit ordinal and (βγ)ϕ(β)(\forall \beta \in \gamma)\phi(\beta)source.

              noindent A final bit of notation will prove helpful later on:

              Definition nine in this chapter

              If XXsource is a set of ordinals, then lsub(X)=αXα+\supstrict(X) = \bigcup_{\alpha \in X} \ordsucc{\alpha}source.

              noindent Here, “lsub” stands for “least strict upper bound”.Footnote: Some books use “sup(X)\text{sup}(X)source” for this. But other books use “sup(X)\text{sup}(X)source” for the least non-strict upper bound, i.e., simply X\bigcup Xsource. If XXsource has a greatest element, α\alphasource, these notions come apart: the least strict upper bound is α+\ordsucc{\alpha}source, whereas the least non-strict upper bound is just α\alphasource. The following result explains this:

              Proposition eight in this chapter

              If XXsource is a set of ordinals, lsub(X)\supstrict(X)source is the least ordinal greater than every ordinal in XXsource.

              Proof

              Let Y={α+:αX}Y = \Setabs{\ordsucc{\alpha}}{\alpha \in X}source, so that lsub(X)=Y\supstrict(X) = \bigcup Ysource. Since ordinals are transitive and every member of an ordinal is an ordinal, lsub(X)\supstrict(X)source is a transitive set of ordinals, and so is an ordinal by corollary four in chapter “Ordinals”.

              If αX\alpha \in Xsource, then α+Y\ordsucc{\alpha} \in Ysource, so α+Y=lsub(X)\ordsucc{\alpha} \subseteq \bigcup Y = \supstrict(X)source, and hence αlsub(X)\alpha \in \supstrict(X)source. So lsub(X)\supstrict(X)source is strictly greater than every ordinal in XXsource.

              Conversely, if αlsub(X)\alpha \in \supstrict(X)source, then αβ+Y\alpha \in \ordsucc{\beta} \in Ysource for some βX\beta \in Xsource, so that αβX\alpha \leq \beta \in Xsource. So lsub(X)\supstrict(X)source is the least strict upper bound on XXsource.