content/set-theory/spine/spine.tex
1% Part: set-theory2% Chapter: spine34\documentclass[../../../include/open-logic-chapter]{subfiles}56\begin{document}78\olchapter{sth}{spine}{Stages and Ranks}910\olimport{idea}11\olimport{recursion}12\olimport{stagesbasics}13\olimport{foundation}14\olimport{zf}15\olimport{rank}1617\OLEndChapterHook1819\end{document}
content/set-theory/spine/idea.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{valpha}67\olsection{Defining the Stages as the $V_\alpha$s}89In \olref[sth][ordinals][]{chap}, we defined well-orderings and the10(von Neumann) ordinals. In this chapter, we will use these to11characterise the hierarchy of sets \emph{itself}. To do this, recall12that in \olref[sth][ordinals][opps]{sec}, we defined the idea of13successor and limit ordinals. We use these ideas in following14definition:1516\begin{defn}\ollabel{defValphas}17 \begin{align*}18 V_\emptyset &\defis \emptyset\\19 V_{\ordsucc{\alpha}} &\defis \Pow{V_\alpha} & & 20 \text{for any ordinal }\alpha\\21 V_{\alpha} &\defis \bigcup_{\gamma < \alpha} V_\gamma & & 22 \text{when }\alpha\text{ is a limit ordinal}23\end{align*}24\end{defn}25\noindent26This will be a definition by \emph{transfinite recursion} on the27ordinals. In this regard, we should compare this with recursive28definitions of functions on the natural numbers.\footnote{Cf.\ the29definitions of addition, multiplication, and exponentiation in30\olref[sfr][infinite][dedekind]{sec}.} As when dealing with natural31numbers, one defines a base case and successor cases; but when dealing32with ordinals, we also need to describe the behaviour of \emph{limit}33cases. 3435This definition of the $V_\alpha$s will be an important milestone. We36have informally motivated our hierarchy of sets as forming sets by37\emph{stages}. The $V_\alpha$s are, in effect, just those stages.38Importantly, though, this is an \emph{internal} characterisation of39the stages. Rather than suggesting a possible \emph{model} of the40theory, we will have defined the stages \emph{within} our set theory.4142\end{document}
content/set-theory/spine/recursion.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{recursion}6\olsection{The Transfinite Recursion Theorem(s)}78The first thing we must do, though, is confirm that9\olref[valpha]{defValphas} is a successful definition. More generally,10we need to prove that any attempt to offer a transfinite by11(transfinite) recursion will succeed. That is the aim of this section.1213\emph{Warning: this is tricky material}. The overarching moral,14though, is quite simple: Transfinite Induction plus Replacement15guarantee the legitimacy of (several versions of) transfinite16recursion.\footnote{A reminder: all formulas and terms can have parameters (unless explicitly stated otherwise).}1718\begin{defn}19 Let $\tau(x)$ be a term; let $f$ be a function; let $\alpha$ be an ordinal. We say that $f$ is an \emph{$\alpha$-approximation} for $\tau$ iff both $\dom{f} = \alpha$ and $(\forall \beta \in \alpha)f(\beta) = \tau(\funrestrictionto{f}{\beta})$.20\end{defn}2122\begin{lem}[Bounded Recursion]\ollabel{transrecursionfun}23For any term $\tau(x)$ and any ordinal $\alpha$, there is a unique $\alpha$-approximation for $\tau$.24\end{lem}25\begin{proof}26We will show that, for any $\gamma \leq \alpha$, there is a unique27$\gamma$-approximation. 2829We first establish uniqueness. Let $g$ and $h$ (respectively) be $\gamma$- and $\delta$-approximations. A transfinite induction on their arguments shows that30$g(\beta) = h(\beta)$ for any $\beta \in \dom{g} \cap \dom{h} =31\gamma \cap \delta = \min(\gamma, \delta)$. So our approximations are unique (if they exist), and agree on all values.3233To establish existence, we now use a simple transfinite induction34(\olref[ordinals][opps]{simpletransrecursion}) on ordinals35$\delta \leq \alpha$. 3637The empty function is trivially an $\emptyset$-approximation. 3839If $g$ is a $\gamma$-approximation, then $g \cup40\{\tuple{\gamma, \tau(g)}\}$ is a $\ordsucc{\gamma}$-approximation.4142If $\gamma$ is a limit ordinal and $g_\delta$ is a $\delta$-approximation for all $\delta < \gamma$, let $g = \bigcup_{\delta \in \gamma} g_\delta$. This43is a function, since our various $g_\delta$s agree on all values. And44if $\delta \in \gamma$ then $g(\delta) = g_{\ordsucc{\delta}}(\delta) =45\tau(\funrestrictionto{g_{\ordsucc{\delta}}}{\delta}) =46\tau(\funrestrictionto{g}{\delta})$.4748This completes the proof by transfinite induction.49\end{proof}5051If we allow ourselves to define a \emph{term} rather than a function,52then we can remove the bound $\alpha$ from the previous result. In53the statement and proof of the following result, when $\sigma$ is a term, we54let $\funrestrictionto{\sigma}{\alpha} = \Setabs{\tuple{\beta,55\sigma(\beta)}}{\beta \in \alpha}$.5657\begin{thm}[General Recursion]\ollabel{transrecursionschema}58For any term $\tau(x)$, we can explicitly define a term59$\sigma(x)$, such that60$\sigma(\alpha) = \tau(\funrestrictionto{\sigma}{\alpha})$ for any61ordinal~$\alpha$.62\end{thm}6364\begin{proof}65For each $\alpha$, by \olref{transrecursionfun} there is a unique $\alpha$-approximation, $f_\alpha$, for~$\tau$. Define $\sigma(\alpha)$ as $f_{\ordsucc{\alpha}}(\alpha)$. Now:66 \begin{align*}67 \sigma(\alpha) &= 68 f_{\ordsucc{\alpha}}(\alpha) \\&= 69 \tau(\funrestrictionto{f_{\ordsucc{\alpha}}}{\alpha}) \\&= 70 \tau(\Setabs{\tuple{\beta, f_{\ordsucc{\alpha}}(\beta)}}{\beta \in \alpha}) \\&= 71 \tau(\Setabs{\tuple{\beta, f_{\ordsucc{\beta}}(\beta)}}{\beta \in \alpha})\\&=72 \tau(\funrestrictionto{\sigma}{\alpha})73 \end{align*}74noting that $f_{\ordsucc{\beta}}(\beta) = f_{\ordsucc{\alpha}}(\beta)$ for all $\beta < \alpha$, as in \olref{transrecursionfun}. 75\end{proof}76\noindent 77Note that \olref{transrecursionschema} is a \emph{schema}. Crucially, we cannot78expect $\sigma$ to define a function, i.e., a certain kind of79\emph{set}, since then $\dom{\sigma}$ would be the set of all80ordinals, contradicting the Burali-Forti Paradox81(\olref[ordinals][basic]{buraliforti}).8283It still remains to show, though, that \olref{transrecursionschema}84vindicates our definition of the $V_\alpha$s. This may not be85immediately obvious; but it will become apparent with a last, simple,86version of transfinite recursion.8788\begin{thm}[Simple Recursion]\ollabel{simplerecursionschema} 89For any terms $\tau(x)$ and $\theta(x)$ and any set $A$, we can90explicitly define a term $\sigma(x)$ such that:91\begin{align*}92 \sigma(\emptyset) &= A\\93 \sigma(\ordsucc{\alpha}) &= \tau(\sigma(\alpha)) &&94 \text{for any ordinal }\alpha\\95 \sigma(\alpha) &= \theta(\ran{\funrestrictionto{\sigma}{\alpha}})&&96 \text{when }\alpha\text{ is a limit ordinal}97\end{align*}98\end{thm}99100\begin{proof}101We start by defining a term, $\xi(x)$, as follows: 102%t $\xi(x) = A$ if $x$ is not a function whose domain is an ordinal;103%otherwise let:104\[105 \xi(x) = 106 \begin{cases}107 A &\text{if $x$ is not a function whose}\\108 &\hspace{1em}\text{domain is an ordinal; otherwise:}\\109 \tau(x(\alpha)) & \text{if $\dom{x} = \ordsucc{\alpha}$}\\110 \theta(\ran{x}) & \text{if $\dom{x}$ is a limit ordinal}111 \end{cases}112\]113By \olref{transrecursionschema}, there is a term $\sigma(x)$ such that114$\sigma(\alpha) = \xi(\funrestrictionto{\sigma}{\alpha})$ for every115ordinal $\alpha$; moreover, $\funrestrictionto{\sigma}{\alpha}$ is a116function with domain $\alpha$. We show that $\sigma$ has the required117properties, by simple transfinite induction118(\olref[ordinals][opps]{simpletransrecursion}). 119120First, $\sigma(\emptyset) = \xi(\emptyset) = A$. 121122Next, $\sigma(\ordsucc{\alpha}) = \xi(\funrestrictionto{\sigma}{\ordsucc{\alpha}}) = \tau(\funrestrictionto{\sigma}{\ordsucc{\alpha}}(\alpha)) = \tau(\sigma(\alpha))$.123124Last, $\sigma(\alpha) = \xi(\funrestrictionto{\sigma}{\alpha}) = \theta(\ran{\funrestrictionto{\sigma}{\alpha}})$, when $\alpha$ is a limit.125\end{proof}126\noindent127Now, to vindicate \olref[valpha]{defValphas}, just take $A128= \emptyset$ and $\tau(x) = \Pow{x}$ and $\theta(x) = \bigcup x$. At long last, this vindicates the definition of the $V_\alpha$s!{}129130131\end{document}
content/set-theory/spine/stagesbasics.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{Valphabasic}6\olsection{Basic Properties of Stages}78To bring out the foundational importance of the definition of the9$V_\alpha$s, we will present a few basic results about them. We start with a definition:\footnote{There's no standard terminology for ``potent''; this is the name used by \citet{ButtonLT1}.}10\begin{defn}11 The set $A$ is \emph{potent} iff $\forall x((\exists y \in A)x \subseteq y \lif x \in A)$. 12\end{defn}13\begin{lem}\ollabel{Valphabasicprops}14For each ordinal $\alpha$:15\begin{enumerate}16 \item\ollabel{Valphatrans} Each $V_\alpha$ is transitive.17 \item\ollabel{Valphapotent} Each $V_\alpha$ is potent.18 \item\ollabel{Valphacum} If $\gamma \in \alpha$, then $V_\gamma19 \in V_\alpha$ (and hence also $V_\gamma \subseteq V_\alpha$ by20 \olref{Valphatrans})21\end{enumerate}22\end{lem}2324\begin{proof}25We prove this by a (simultaneous) transfinite induction. For26induction, suppose that \olref{Valphatrans}--\olref{Valphacum} holds27for each ordinal $\beta < \alpha$. 2829The case of $\alpha = \emptyset$ is trivial. 3031Suppose $\alpha = \ordsucc{\beta}$. To show \olref{Valphacum}, if32$\gamma \in \alpha$ then $V_\gamma \subseteq V_\beta$ by hypothesis,33so $V_\gamma \in \Pow{V_\beta} = V_\alpha$. To show34\olref{Valphapotent}, suppose $A \subseteq B \in V_\alpha$ i.e., $A35\subseteq B \subseteq V_\beta$; then $A \subseteq V_\beta$ so $A \in36V_\alpha$. To show \olref{Valphatrans}, note that if $x \in A \in37V_\alpha$ we have $A \subseteq V_\beta$, so $x \in V_\beta$, so $x38\subseteq V_\beta$ as $V_\beta$ is transitive by hypothesis, and so $x39\in V_\alpha$. 4041Suppose $\alpha$ is a limit ordinal. To show \olref{Valphacum}, if42$\gamma \in \alpha$ then $\gamma \in \ordsucc{\gamma} \in \alpha$, so43that $V_\gamma \in V_{\ordsucc{\gamma}}$ by assumption, hence44$V_\gamma \in \bigcup_{\beta \in \alpha} V_\beta = V_\alpha$. To show45\olref{Valphatrans} and \olref{Valphapotent}, just observe that a46union of transitive (respectively, potent) sets is transitive47(respectively, potent). 48\end{proof}4950\begin{lem}\ollabel{Valphanotref}51For each ordinal $\alpha$, $V_\alpha \notin V_\alpha$.52\end{lem}5354\begin{proof}55By transfinite induction. Evidently $V_\emptyset \notin V_\emptyset$. 5657If $V_{\ordsucc{\alpha}} \in V_{\ordsucc{\alpha}} = \Pow{V_\alpha}$,58then $V_{\ordsucc{\alpha}} \subseteq V_\alpha$; and since $V_\alpha59\in V_{\ordsucc{\alpha}}$ by \olref{Valphabasicprops}, we have60$V_\alpha \in V_\alpha$. Conversely: if $V_\alpha \notin V_\alpha$61then $V_{\ordsucc{\alpha}} \notin V_{\ordsucc{\alpha}}$6263If $\alpha$ is a limit and $V_\alpha \in V_\alpha = \bigcup_{\beta \in64\alpha}V_\beta$, then $V_\alpha \in V_\beta$ for some $\beta \in65\alpha$; but then also $V_\beta \in V_\alpha$ so that $V_\beta \in66V_\beta$ by \olref{Valphabasicprops} (twice). Conversely, if $V_\beta67\notin V_\beta$ for all $\beta \in \alpha$, then $V_\alpha \notin68V_\alpha$.69\end{proof}7071\begin{cor}72For any ordinals $\alpha, \beta$: $\alpha \in \beta$ iff $V_\alpha \in V_\beta$73\end{cor}7475\begin{proof}76\Olref{Valphabasicprops} gives one direction. Conversely, suppose $V_\alpha \in V_\beta$. Then $\alpha \neq \beta$ by \olref{Valphanotref}; and $\beta \notin \alpha$, for otherwise we would have $V_\beta \in V_\alpha$ and hence $V_\beta \in V_\beta$ by \olref{Valphabasicprops} (twice), contradicting \olref{Valphanotref}. So $\alpha \in \beta$ by Trichotomy.77\end{proof}78\noindent79All of this allows us to think of each $V_\alpha$ as the $\alpha$th80stage of the hierarchy. Here is why.8182Certainly our $V_\alpha$s can be thought of as being formed in an83\emph{iterative} process, for our use of ordinals tracks the notion of84iteration. Moreover, if one stage is formed before the other, i.e.,85$V_\beta \in V_\alpha$, i.e., $\beta \in \alpha$, then our process of86formation is \emph{cumulative}, since $V_\beta \subseteq V_\alpha$.87Finally, we are indeed forming \emph{all} possible collections of sets88that were available at any earlier stage, since any successor stage89$V_{\ordsucc{\alpha}}$ is the power-set of its predecessor $V_\alpha$.9091In short: with $\ZFminus$, we are \emph{almost} done, in articulating92our vision of the cumulative-iterative hierarchy of sets. (Though, of93course, we still need to justify Replacement.)9495\end{document}
content/set-theory/spine/foundation.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{foundation}67\olsection{Foundation}89We are only \emph{almost} done---and not \emph{quite} finished---because nothing in10$\ZFminus$ guarantees that \emph{every} set is in some $V_\alpha$,11i.e., that every set is formed at some stage. 1213Now, there is a fairly straightforward (mathematical) sense in which14we don't \emph{care} whether there are sets outside the hierarchy. (If15there are any there, we can simply ignore them.) But we have motivated16our \emph{concept} of set with the thought that every set is formed at17some stage (see \stageshier{} in \olref[z][story]{sec}). So18we will want to preclude the possibility of sets which fall outside of19the hierarchy. Accordingly, we must add a new axiom, which ensures20that every set occurs somewhere in the hierarchy. 2122Since the $V_\alpha$s are our stages, we might simply consider adding23the following as an axiom:2425\begin{defish}26\emph{Regularity.} $\forall A \exists \alpha\, A \subseteq V_\alpha$27\end{defish}2829This would be a perfectly reasonable approach. However, for reasons30that will be explained in the next section, we will instead adopt an31alternative axiom:3233\begin{axiom}[Foundation]34$(\forall A \neq \emptyset)(\exists B \in A)A \cap B = \emptyset$.35\end{axiom}3637With some effort, we can show (in $\ZFminus$) that Foundation entails Regularity:38\begin{defn}39For each set $A$, let:40\begin{align*}41 \text{cl}_0(A) &= A,\\42 \text{cl}_{n+1}(A) &= \bigcup \text{cl}_n(A),\\43 \text{trcl}(A) &= \bigcup_{n < \omega} \text{cl}_{n}(A).44\end{align*}45We call $\text{trcl}(A)$ the \emph{transitive closure} of $A$.46\end{defn}47\noindent48The name ``transitive closure'' is apt:49\begin{prop}\ollabel{subsetoftrcl}50$A \subseteq \trcl{A}$ and $\trcl{A}$ is a transitive set. 51\end{prop}5253\begin{proof}54Evidently $A = \text{cl}_0(A) \subseteq \text{trcl}(A)$. And if $x55\in b \in \trcl{A}$, then $b \in \text{cl}_n(A)$ for some $n$, so $x56\in \text{cl}_{n+1}(A) \subseteq \text{trcl}(A)$. 57\end{proof}5859\begin{lem}\ollabel{lem:TransitiveWellFounded}60If $A$ is a transitive set, then there is some $\alpha$ such that $A61\subseteq V_\alpha$.62\end{lem}6364\begin{proof}65Recalling the definition of ``$\supstrict(X)$'' from66\olref[ordinals][opps]{defsupstrict}, define two sets:67\begin{align*}68 D &= \Setabs{x \in A}{\forall \delta\ x \nsubseteq V_\delta}\\69 \alpha &= \supstrict\Setabs{\delta}{(\exists x \in A)70 (x \subseteq V_\delta \land (\forall \gamma \in \delta)x \nsubseteq V_\gamma)}71\end{align*}72Suppose $D = \emptyset$. So if $x \in A$, then there is some $\delta$ such that $x \subseteq V_\delta$ and, by the well-ordering of the ordinals, $(\forall \gamma \in \delta)x \nsubseteq V_\gamma$; hence $\delta \in \alpha$ and so $x \in V_\alpha$ by73\olref[spine][Valphabasic]{Valphabasicprops}. Hence $A \subseteq74V_{\alpha}$, as required. 7576So it suffices to show that $D = \emptyset$. For reductio, suppose77otherwise. By Foundation, there is some $B \in D \subseteq A$ such that $D \cap B78= \emptyset$. If $x \in B$ then $x \in A$, since $A$ is transitive,79and since $x \notin D$, it follows that $\exists \delta\ x \subseteq80V_\delta$. So now let81\[82 \beta = \supstrict\Setabs{\delta}{(\exists x \in b)(x \subseteq V_\delta \land (\forall \gamma < \delta)x \nsubseteq V_\gamma)}.83\]84As before, $B \subseteq V_\beta$, contradicting the claim that $B \in85D$. 86\end{proof}8788\begin{thm}\ollabel{zfentailsregularity}89Regularity holds.90\end{thm}9192\begin{proof}93Fix $A$; now $A \subseteq \trcl{A}$ by \olref{subsetoftrcl}, which is transitive. So there is some $\alpha$ such that $A \subseteq \trcl{A} \subseteq V_\alpha$ by \olref{lem:TransitiveWellFounded}94\end{proof}9596These results show that $\ZFminus$ proves the conditional97$\emph{Foundation}\Rightarrow\emph{Regularity}$. In98\olref[spine][rank]{zfminusregularityfoundation}, we will show that99$\ZFminus$ proves $\emph{Regularity}\Rightarrow\emph{Foundation}$. As100such, Foundation and Regularity are \emph{equivalent} (modulo101$\ZFminus$). But this means that, given $\ZFminus$, we can justify102Foundation by noting that it is equivalent to Regularity. And we can103justify Regularity immediately on the basis of \stageshier{}. 104105\end{document}
content/set-theory/spine/zf.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{zf}6\olsection{$\Z$ and $\ZF$: A Milestone}78With Foundation, we reach another important milestone. We have9considered theories $\Zminus$ and $\ZFminus$, which we said were10certain theories ``minus'' a certain something. That certain something11is Foundation. So:1213\begin{defn}14The theory $\Z$ adds Foundation to $\Zminus$. So its axioms are15Extensionality, Union, Pairs, Powersets, Infinity, Foundation, and all16instances of the Separation scheme.1718The theory $\ZF$ adds Foundation to $\ZFminus$. Otherwise put, $\ZF$19adds all instances of Replacement to~$\Z$.20\end{defn}2122Still, one question might have occurred to you. If Regularity is23equivalent over $\ZFminus$ to Foundation, and Regularity's24justification is clear, why bother to go around the houses, and take25Foundation as our basic axiom, rather than Regularity? 2627Setting aside historical reasons (to do with who formulated what and28when), the basic reason is that Foundation can be presented without29employing the definition of the~$V_\alpha$s. That definition relied30upon all of the work of \olref[recursion]{sec}: we31needed to prove Transfinite Recursion, to show that it was justified.32But our proof of Transfinite Recursion employed \emph{Replacement}.33So, whilst Foundation and Regularity are equivalent modulo $\ZFminus$,34they are not equivalent modulo $\Zminus$. 3536Indeed, the matter is more drastic than this simple remark suggests.37Though it goes well beyond this book's remit, it turns out that both38$\Zminus$ and $\Z$ are too weak to define the $V_\alpha$s. So, if you39are working only in $\Z$, then Regularity (as we have formulated it)40does not even make \emph{sense}. This is why our official axiom is41Foundation, rather than Regularity. 4243From now on, we will work in $\ZF$ (unless otherwise stated), without44any further comment. 4546\end{document}
content/set-theory/spine/rank.tex
1\documentclass[../../../include/open-logic-section]{subfiles}23\begin{document}45\olfileid{sth}{spine}{rank}6\olsection{Rank}78Now that we have defined the stages as the $V_\alpha$'s, and we know9that every set is a subset of some stage, we can define the10\emph{rank} of a set. Intuitively, the rank of $A$ is the first moment11at which $A$ is formed. More precisely:1213\begin{defn}\ollabel{defnsetrank}14For each set $A$, $\setrank{A}$ is the least ordinal $\alpha$ such that $A15\subseteq V_\alpha$.16\end{defn}17\begin{prop}\ollabel{ranksexist}18 $\setrank{A}$ exists, for any $A$.19\end{prop}20\begin{proof}21 Left as an exercise.22\end{proof}23\begin{prob}24 Prove \olref[sth][spine][rank]{ranksexist}. 25\end{prob}26\noindent 27The well-ordering of ranks allows us to prove some important results:28\begin{prop}\ollabel{valphalowerrank}29For any ordinal $\alpha$, $V_\alpha = \Setabs{x}{\setrank{x} \in \alpha}$.30\end{prop}3132\begin{proof}33If $\setrank{x} \in \alpha$ then $x \subseteq V_{\setrank{x}} \in34V_\alpha$, so $x \in V_\alpha$ as $V_\alpha$ is potent (invoking35\olref[Valphabasic]{Valphabasicprops} multiple times). Conversely, if $x \in V_\alpha$ then $x \subseteq V_\alpha$, so $\setrank{x} \leq \alpha$; now a simple transfinite induction shows that $x \notin V_\alpha$. 36\end{proof}37\begin{prob}38 Complete the simple transfinite induction mentioned in \olref[sth][spine][rank]{valphalowerrank}.39\end{prob}4041\begin{prop}\ollabel{rankmemberslower}42If $B \in A$, then $\setrank{B} \in \setrank{A}$.43\end{prop}4445\begin{proof}46$A \subseteq V_{\setrank{A}} = \Setabs{x}{\setrank{x} \in47\setrank{A}}$ by \olref{valphalowerrank}.48\end{proof}49\noindent50Using this fact, we can establish a result which allows us to prove51things about \emph{all sets} by a form of induction:5253\begin{thm}[$\in$-Induction Scheme] 54For any formula $\phi$:55\[56 \forall A((\forall x \in A)\phi(x) \lif \phi(A)) \lif \forall A \phi(A).57\]58\end{thm}5960\begin{proof}61We will prove the contrapositive. So, suppose $\lnot \forall A62\phi(A)$. By Transfinite Induction63(\olref[ordinals][basic]{ordinductionschema}), there64is some non-$\phi$ of least possible rank; i.e.\ some $A$65such that $\lnot \phi(A)$ and $\forall x(\setrank{x} \in \setrank{A}66\lif \phi(x))$. Now if $x \in A$ then $\setrank{x} \in67\setrank{A}$, by \olref{rankmemberslower}, so that $\phi(x)$; i.e.\ $(\forall x \in68A)\phi(x) \land \lnot \phi(A)$.69\end{proof}\noindent Here is an informal way to gloss this powerful70result. Say that $\phi$ is \emph{hereditary} iff whenever every71!!{element} of a set is $\phi$, the set itself is $\phi$. Then72$\in$-Induction tells you the following: if $\phi$ is hereditary,73every set is $\phi$.7475To wrap up the discussion of ranks (for now), we'll prove a few76claims which we have foreshadowed a few times. 7778\begin{prop}\ollabel{ranksupstrict}79$\setrank{A} = \supstrict_{x \in A}\setrank{x}$.80\end{prop}8182\begin{proof}83Let $\alpha = \supstrict_{x \in A}\setrank{x}$. By84\olref{rankmemberslower}, $\alpha \leq \setrank{A}$. But if $x \in A$85then $\setrank{x} \in \alpha$, so that $x \in V_\alpha$ by \olref{valphalowerrank}, and hence $A86\subseteq V_\alpha$, i.e., $\setrank{A} \leq \alpha$. Hence87$\setrank{A} = \alpha$.88\end{proof}8990\begin{cor}\ollabel{ordsetrankalpha}91For any ordinal $\alpha$, $\setrank{\alpha} = \alpha$.92\end{cor}9394\begin{proof}95Suppose for transfinite induction that $\setrank{\beta} = \beta$ for96all $\beta \in \alpha$. Now $\setrank{\alpha} = \supstrict_{\beta \in97\alpha}\setrank{\beta} = \supstrict_{\beta \in \alpha}\beta = \alpha$98by \olref{ranksupstrict}.99% First note that $\setrank{\alpha} \neq \beta$ for any $\beta \in100% \alpha$. For otherwise we would have $\beta = \setrank{\beta} \in101% \setrank{\alpha} = \beta$ by \olref{rankmemberslower}, i.e.,102% $\beta \in \beta$, a contradiction. 103%104% Now note that $\alpha \subseteq V_\alpha$. For $\alpha =105% \Setabs{\beta}{\beta \in \alpha}$ by106% \olref[sfr][ordinals][basic]{ordissetofsmallerord}. And if107% $\beta \in \alpha$ then $\beta \subseteq V_{\setrank{\beta}} =108% V_\beta \in V_\alpha$, hence $\beta \in V_\alpha$, by109% \olref[sfr][spine][Valphabasic]{Valphabasicprops} twice. 110\end{proof}111112Finally, here is a quick proof of the result promised at the end of113\olref[foundation]{sec}, that $\ZFminus$ proves the conditional114$\emph{Regularity} \Rightarrow \emph{Foundation}$. (Note that the115notion of ``rank'' and \olref{rankmemberslower} are available for use116in this proof since---as mentioned at the start of this section---they117can be presented using $\ZFminus + \text{Regularity}$.)118119\begin{prop}[working in $\ZFminus + \text{Regularity}$]\ollabel{zfminusregularityfoundation} Foundation holds.120\end{prop}121122\begin{proof}123Fix $A \neq \emptyset$, and some $B \in A$ of least possible rank. If124$c \in B$ then $\setrank{c} \in \setrank{B}$ by125\olref{rankmemberslower}, so that $c \notin A$ by choice of $B$.126\end{proof}127128\end{document}