42.03.01 · mathematical-logic / set-theory-forcing

The ZFC Axioms and the Cumulative Hierarchy

shipped3 tiersLean: none

Anchor (Master): Kunen 2011 *Set Theory* (College Publications) Ch. I §14 (the cumulative hierarchy V_α, rank, ∈-recursion) and §15-16 (the Reflection Principle and its metamathematical consequences); Jech 2003 *Set Theory* 3e (Springer) Ch. 6 (the von Neumann hierarchy) and Ch. 12 (reflection); Zermelo 1930 *Fund. Math.* 16 (Grenzzahlen, models V_κ)

Intuition Beginner

Almost everything in mathematics can be built out of one humble ingredient: the collection. A number, a function, a triangle, a probability — each can be coded as a collection of simpler things, which are themselves collections, all the way down. Set theory is the study of these collections, called sets, and the surprising fact is that a short list of ground rules about them is enough to support the entire building. The single relation you need is "belongs to": one set either is or is not a member of another.

You might think the safe rule is "any property you can describe carves out a set." Picture the property "is a set that does not belong to itself." Most everyday collections satisfy it. Now ask whether the collection of all such sets belongs to itself. If it does, then by its own description it should not; if it does not, then it fits the description and so should. Either way you reach a contradiction. This is Russell's paradox, and it shows the carefree rule is broken. The repair is to stop assuming every description gives a set and instead lay down explicit, modest rules for building sets from sets you already have.

Those rules are the ZFC axioms. They are deliberately cautious: you may pair two sets, pool a family into its union, form the collection of all subsets, separate out the members of an existing set that satisfy a property, and guarantee one infinite set to get started. None of these lets you scoop up "everything at once," which is exactly what caused the trouble.

The picture that makes the whole system feel inevitable is a tower built in stages. Start with nothing. At each new stage, allow every collection whose members appeared at earlier stages. You never reach a top, but every set lives at some stage. This staged tower is the cumulative hierarchy, and the rest of this unit is about why it captures precisely the sets the axioms describe.

Visual Beginner

The cumulative hierarchy is a tower of stages. Each stage contains every collection you can form using only the items from the stages below it. The picture shows the bottom stages and the rule that builds the next stage from the one before.

   stage 3   { everything whose members come from stage 2 }   ... huge
              ^
              |  collect all sub-collections of the stage below
   stage 2   { {}, {a}, {b}, {a,b}, ... }        items: 4
              ^
              |
   stage 1   { {} , {a} }                          items: 2
              ^
              |   here a = {} is the only item available
   stage 0   {  }   (nothing yet)                  items: 0

   the tower never stops; every set sits at some finite or
   infinite stage, and a set's "rank" is the first stage that
   contains it

Read the tower from the bottom up. Stage $0$ is empty. At each step you look at everything already present and form all possible sub-collections of it; those become the new stage. Because stage $1$ already has $2$ items, stage $2$ has $2 \times 2 = 4$ sub-collections, stage $3$ has $2^{4} = 16$ , and the counts explode upward.

stage	how many items it holds
stage $0$	$0$
stage $1$	$2$
stage $2$	$4$
stage $3$	$16$

The key idea: a set never appears before all of its members have appeared. The first stage where a set shows up is called its rank, and ranking every set is what ties the loose rules together into one orderly universe.

Worked example Beginner

Let us build the first few stages by hand, starting from absolutely nothing, and find the rank of a small set. The only raw material is the empty collection, written $\emptyset$ — the collection with no members at all.

Step 1. Stage $0$ is empty: it has no items. So at stage $0$ there is nothing to collect.

Step 2. Build stage $1$ . We collect every sub-collection of what stage $0$ offered. The only sub-collection of nothing is the empty collection itself. So stage $1 = {\emptyset}$ , a tower level with exactly one item, namely $\emptyset$ .

Step 3. Build stage $2$ . Now we have one item available, $\emptyset$ , so the sub-collections are: the empty one, and the one containing $\emptyset$ . So stage $2 = {\emptyset, {\emptyset}}$ , with two items.

Step 4. Build stage $3$ . We have two items available, so there are $2 \times 2 = 4$ sub-collections to form, including the new set ${{\emptyset}}$ and the full set ${\emptyset, {\emptyset}}$ .

Step 5. Find a rank. Consider the set $x = {{\emptyset}}$ . Its one member is ${\emptyset}$ , which first appeared at stage $2$ . A set appears one stage after its members, so $x$ first appears at stage $3$ . Its rank is $3$ .

What this tells us: with no ingredients except the empty collection, the staged rule already manufactures a rich supply of distinct sets, and each one earns a definite rank — the first stage that holds it. This single number will organise the whole universe in the formal sections.

Check your understanding Beginner

Formal definition Intermediate+

The background language is first-order logic with equality over a single binary relation symbol $\in$ (cross-ref 42.01.03 pending); the only nonlogical primitive is membership, and every other notion ( $\subseteq$ , $\emptyset$ , ordered pair, function) is defined from it. Variables range over sets; there are no urelements. A class is not an object of the theory but a syntactic abbreviation: for a formula $φ (x)$ with parameters, the expression ${x : φ (x)}$ stands for $φ$ used as a membership predicate, and $y \in {x : φ (x)}$ abbreviates $φ (y)$ . A class that is equal to a set (some $z$ with $\forall x (x \in z \leftrightarrow φ (x))$ ) is a set; otherwise it is a proper class. The universal class $V = {x : x = x}$ is a proper class, as is the Russell class $R = {x : x \in / x}$ .

The ZFC axioms are the following sentences and schemas ^{[Kunen Ch. I]}.

Extensionality. $\forall x \forall y (\forall z (z \in x \leftrightarrow z \in y) \to x = y)$ . Sets are determined by their members.

Foundation (Regularity). $\forall x (x \neq = \emptyset \to \exists y \in x (y \cap x = \emptyset))$ . Every nonempty set has an $\in$ -minimal member.

Pairing. $\forall x \forall y \exists z \forall w (w \in z \leftrightarrow w = x \lor w = y)$ .

Union. $\forall F \exists u \forall x (x \in u \leftrightarrow \exists S \in F (x \in S))$ .

Power Set. $\forall x \exists y \forall z (z \in y \leftrightarrow z \subseteq x)$ .

Separation (Comprehension) schema. For each formula $φ (z, p)$ not containing $y$ free, $\forall x \forall p \exists y \forall z (z \in y \leftrightarrow (z \in x \land φ (z, p)))$ . One may separate a subset of an existing set by any property; one may not form ${z : φ (z)}$ outright. This is the precise repair of naive comprehension that disarms Russell's paradox: ${z \in x : z \in / z}$ exists for every set $x$ , but is not a member of $x$ , so no contradiction arises.

Replacement schema. For each formula $ψ (z, w, p)$ functional in $w$ , $\forall x \forall p ((\forall z \in x \exists! w ψ) \to \exists y \forall w (w \in y \leftrightarrow \exists z \in x ψ))$ . The image of a set under a definable class function is a set.

Infinity. $\exists x (\emptyset \in x \land \forall y \in x (y \cup {y} \in x))$ . There is an inductive set.

Choice (AC). Every set of nonempty pairwise-disjoint sets has a choice set; equivalently every set is well-orderable. Choice is stated here but its theory — the equivalents (Zorn, well-ordering) and independence — is developed separately in 42.03.05.

The theory ZF is the list without Choice; ZFC adjoins it. Ordered pairs are coded à la Kuratowski, $(x, y) := {{x}, {x, y}}$ , which satisfies $(x, y) = (x^{'}, y^{'}) \leftrightarrow (x = x^{'} \land y = y^{'})$ ; the Cartesian product $x \times y = {(a, b) : a \in x, b \in y}$ is a set by Separation applied to $P (P (x \cup y))$ . A relation is a set of ordered pairs, a function a relation that is single-valued, and an ordering a relation with the appropriate reflexivity/transitivity/antisymmetry. All of relation theory thus lives inside the set universe.

Counterexamples to common slips Intermediate+

" $V = {x : x = x}$ is a set." If $V$ were a set, Separation would yield the Russell set $R = {x \in V : x \in / x}$ , which both is and is not a member of itself. So $V$ is a proper class; the same argument shows the class of all singletons, the class of all ordinals, and $R$ itself are proper classes.
"Foundation forbids only $x \in x$ ." Foundation forbids every finite or infinite descending $\in$ -chain $\dots \in x_{2} \in x_{1} \in x_{0}$ , not merely the one-step loop $x \in x$ . The two-step loop $x \in y \in x$ and the infinite descent are also ruled out, because the set ${x_{0}, x_{1}, x_{2}, \dots}$ would have no $\in$ -minimal element.
"Replacement is just Separation in disguise." Separation only carves subsets out of a given set and so cannot increase rank or cardinality; Replacement can. The set ${ω, P (ω), P (P (ω)), \dots}$ and the ordinal $ω + ω$ require Replacement: the relevant collection is the image of $ω$ under a class function but is not a subset of any set produced without it.
" ${\emptyset}$ and $\emptyset$ are the same because both are 'empty'." $\emptyset$ has no members; ${\emptyset}$ has exactly one member, namely $\emptyset$ . By Extensionality they differ, since $\emptyset \in {\emptyset}$ but $\emptyset \in / \emptyset$ . The rank of $\emptyset$ is $0$ and the rank of ${\emptyset}$ is $1$ .

Key theorem with proof Intermediate+

The cumulative hierarchy is defined by transfinite recursion on the ordinals, and the signature theorem is that, granting Foundation, it exhausts the universe: every set lives at some level, so $V = ⋃_{α} V_{α}$ . This is what licenses proofs "by induction on rank" throughout set theory.

Definition (cumulative hierarchy). By recursion on ordinals, $$ V_0 = \varnothing, \qquad V_{\alpha+1} = \mathcal P(V_\alpha), \qquad V_\lambda = \bigcup_{\alpha < \lambda} V_\alpha \ \ (\lambda \text{ limit}). $$ Each $V_{α}$ is a set (Power Set at successors, Replacement and Union at limits). The $V_{α}$ are transitive and $\subseteq$ -increasing. The rank of a set $x$ is $rank (x) =$ the least $α$ with $x \in V_{α + 1}$ , equivalently $rank (x) = sup {rank (y) + 1 : y \in x}$ .

Theorem (the hierarchy exhausts $V$ ). Assume the axioms of ZF, including Foundation. Then every set belongs to some $V_{α}$ ; that is, $V = ⋃_{α \in Ord} V_{α}$ , and every set has an ordinal rank.

Proof. Call a set $x$ grounded if $x \in V_{α}$ for some ordinal $α$ . First note that $x$ is grounded if and only if every member of $x$ is grounded. If $x \in V_{α}$ , then since each $V_{α}$ is transitive every $y \in x$ lies in $V_{α}$ as well, so members of grounded sets are grounded. Conversely, suppose every $y \in x$ is grounded, with $rank (y)$ defined; by Replacement the image ${rank (y) + 1 : y \in x}$ is a set of ordinals, so it has a supremum $β$ . Then every $y \in x$ lies in $V_{β}$ , hence $x \subseteq V_{β}$ , so $x \in P (V_{β}) = V_{β + 1}$ and $x$ is grounded.

Now suppose toward a contradiction that some set is not grounded. By Foundation applied to a suitable transitive set, choose a set $z$ that is not grounded but all of whose members are grounded: form the transitive closure $trcl (x)$ of an ungrounded $x$ (a set, built by Union and Replacement over the $ω$ -iterated unions), let $A \subseteq trcl (x) \cup {x}$ be the nonempty set of its ungrounded members together with $x$ , and apply Foundation to $A$ to get an $\in$ -minimal $z \in A$ . By minimality, no member of $z$ is in $A$ , so every member of $z$ is grounded. But then $z$ is grounded by the equivalence above — contradicting $z \in A$ . Hence no ungrounded set exists, and every set lies in some $V_{α}$ with a well-defined rank. $□$

Bridge. This theorem is the foundational reason set theory can argue by induction on rank: every set sits at a definite ordinal level, so a property holding at all lower levels and propagating upward holds everywhere, which is exactly the $\in$ -induction principle proved in the next section. It generalises ordinary induction on $ω$ to the whole universe, and it builds toward the construction of the ordinals 42.03.02 and cardinals 42.03.03, which are themselves the index set $Ord$ and certain ranks inside this tower. The central insight is that Foundation and the hierarchy are two faces of one fact — the membership relation is well-founded — and this is dual to the way the ordinals well-order the levels from below; putting these together, the universe $V$ is not a completed totality but the union of an unbounded, increasing sequence of sets, a picture that appears again in the Reflection Principle, where finite fragments of the theory of $V$ are already true in some level $V_{α}$ .

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Prove that each $V_{α}$ is transitive (every member of a member is a member), by induction on $α$ .

Hint

Transitivity is preserved by power set and by union. Handle $0$ , successor, and limit separately.

Answer

By transfinite induction. Base: $V_{0} = \emptyset$ is vacuously transitive. Successor: assume $V_{α}$ transitive and let $x \in V_{α + 1} = P (V_{α})$ , so $x \subseteq V_{α}$ . If $y \in x$ then $y \in V_{α}$ , and transitivity of $V_{α}$ together with $V_{α} \subseteq P (V_{α}) = V_{α + 1}$ (every member of $V_{α}$ is a subset of $V_{α}$ , hence in $P (V_{α})$ ) gives $y \in V_{α + 1}$ . Limit: $V_{λ} = ⋃_{α < λ} V_{α}$ ; if $x \in V_{λ}$ then $x \in V_{α}$ for some $α < λ$ , and by induction every $y \in x$ lies in $V_{α} \subseteq V_{λ}$ . Rubric: full credit for the inclusion $V_{α} \subseteq V_{α + 1}$ at the successor step and correct handling of all three cases.

Exercise 4 (medium, symbolic).

Show that the Russell class $R = {x : x \in / x}$ is a proper class. Then show, using Foundation, that $R = V$ — that is, no set is a member of itself.

Hint

For the first part, suppose $R$ is a set and ask whether $R \in R$ . For the second, apply Foundation to ${x}$ .

Answer

If $R$ were a set, then $R \in R$ holds if and only if $R \in / R$ by the defining condition, a contradiction; so $R$ is a proper class. For $R = V$ : let $x$ be any set and apply Foundation to ${x}$ , which is nonempty, to get an $\in$ -minimal member. The only member is $x$ , and minimality says $x \cap {x} = \emptyset$ , i.e. $x \in / x$ . Hence every set satisfies $x \in / x$ , so $x \in R$ for all $x$ and $R = V$ . (Thus under Foundation the Russell condition is satisfied by everything, yet collecting everything still fails — the obstruction is comprehension, not the predicate.) Rubric: full credit for the self-membership contradiction and the Foundation-on- ${x}$ argument.

Exercise 5 (medium, symbolic).

Verify the Kuratowski pairing axiom: prove $(x, y) = (x^{'}, y^{'})$ implies $x = x^{'}$ and $y = y^{'}$ , where $(a, b) := {{a}, {a, b}}$ .

Hint

Use Extensionality. Split into the cases $x = y$ and $x \neq = y$ .

Answer

Suppose ${{x}, {x, y}} = {{x^{'}}, {x^{'}, y^{'}}}$ . Case $x = y$ : then $(x, y) = {{x}}$ , a singleton, so the right side is a singleton too, forcing ${x^{'}} = {x^{'}, y^{'}}$ , hence $x^{'} = y^{'}$ and ${x^{'}} = {x}$ , giving $x^{'} = x = y = y^{'}$ . Case $x \neq = y$ : then ${x} \neq = {x, y}$ , so $(x, y)$ has two distinct members. The singleton member ${x}$ must equal the singleton member of the right side; if ${x} = {x^{'}, y^{'}}$ then $x^{'} = y^{'} = x$ , making the right side a singleton, contradiction, so ${x} = {x^{'}}$ and $x = x^{'}$ . Then ${x, y} = {x^{'}, y^{'}} = {x, y^{'}}$ with $x \neq = y$ forces $y = y^{'}$ . Rubric: full credit for both cases and use of Extensionality to compare the singleton components.

Exercise 6 (medium, symbolic).

Prove that Replacement implies Separation, given the other axioms. (So Separation is redundant in the official list, though it is kept for clarity.)

Hint

Given $x$ and $φ$ , define a class function that is the identity on members satisfying $φ$ and sends the rest somewhere fixed; handle the empty case separately.

Answer

Fix a set $x$ and a formula $φ (z)$ . If no $z \in x$ satisfies $φ$ , the desired set is $\emptyset$ (a set by Separation-free means: $\emptyset = {z \in x : z \neq = z}$ is replaced below, or take Pairing then Replacement). Otherwise fix some $z_{0} \in x$ with $φ (z_{0})$ and define the class function $ψ (z, w) :\equiv (φ (z) \land w = z) \lor (\neg φ (z) \land w = z_{0})$ . This is functional: each $z \in x$ has a unique image $w$ (itself if $φ (z)$ , else $z_{0}$ ). By Replacement the image $y = {w : \exists z \in x ψ (z, w)}$ is a set, and $y = {z \in x : φ (z)}$ since every output is some $z \in x$ with $φ (z)$ (including $z_{0}$ ). Hence the Separation instance for $φ$ holds. Rubric: full credit for the functional class definition with the fixed fallback $z_{0}$ and the empty-case handling.

Exercise 7 (hard, symbolic).

Prove the principle of $\in$ -induction from Foundation: if $φ$ is a formula such that $\forall x ((\forall y \in x φ (y)) \to φ (x))$ , then $\forall x φ (x)$ .

Hint

Suppose $φ$ fails somewhere. Pass to the transitive closure of a witness and apply Foundation to the set of failures inside it.

Answer

Suppose the inductive hypothesis $\forall x ((\forall y \in x φ (y)) \to φ (x))$ holds but $φ (a)$ fails for some $a$ . Let $t = trcl ({a})$ be the transitive closure (a set, formed by iterating Union over ${a}, ⋃ {a}, \dots$ and taking the union of the $ω$ -sequence via Replacement). The set $F = {z \in t : \neg φ (z)}$ is nonempty since $a \in F$ . By Foundation choose an $\in$ -minimal $z \in F$ . For each $y \in z$ : since $t$ is transitive, $y \in t$ , and by minimality $y \in / F$ , so $φ (y)$ . Thus $\forall y \in z φ (y)$ , and the inductive hypothesis gives $φ (z)$ , contradicting $z \in F$ . Hence no failure exists and $\forall x φ (x)$ . Rubric: full credit for passing to the transitive closure, applying Foundation to the failure set, and using transitivity to feed the inductive hypothesis.

Exercise 8 (hard, symbolic).

Let $κ$ be an inaccessible cardinal (regular and a strong limit; this anticipates 42.03.03). Sketch why $V_{κ}$ satisfies Replacement, and hence is a model of all of ZFC. (You may use that $V_{κ}$ already models the other axioms.)

Hint

A definable function on a set of size $< κ$ has image of size $< κ$ by regularity; bound the ranks of the image using strong-limit-ness.

Answer

Work inside $V_{κ}$ and take an instance of Replacement: a set $a \in V_{κ}$ and a class function $F$ (definable with parameters in $V_{κ}$ ) defined on $a$ . Since $a \in V_{κ}$ we have $∣ a ∣ < κ$ (as $∣ V_{α} ∣ < κ$ for $α < κ$ , using that $κ$ is a strong limit so each power-set step stays below $κ$ ). The image $F [a]$ has $∣ F [a] ∣ \leq ∣ a ∣ < κ$ . Each $F (z)$ lies in some $V_{β_{z}}$ with $β_{z} < κ$ ; by regularity of $κ$ , $sup_{z \in a} β_{z} < κ$ , call it $β$ . Then $F [a] \subseteq V_{β}$ and $∣ F [a] ∣ < κ$ , so $F [a] \in V_{β + 1} \subseteq V_{κ}$ . Thus the image is a set in $V_{κ}$ , verifying Replacement. With the other axioms holding in $V_{κ}$ (Power Set from strong-limit-ness, Infinity since $κ > ω$ , Choice inherited from $V$ ), $V_{κ} ⊨ ZFC$ . Rubric: full credit for using regularity to bound the supremum of ranks and strong-limit-ness to keep cardinalities below $κ$ .

Advanced results Master

The hierarchy makes the universe a well-ordered union of sets, and the deeper structural facts concern how the theory of $V$ is approximated from within and why this approximation is exactly as strong as it can consistently be.

Theorem 1 ( $\in$ -recursion). Let $G$ be a class function. There is a unique class function $F$ on $V$ satisfying $F (x) = G (x, F ↾ x)$ for all $x$ , where $F ↾ x = {(y, F (y)) : y \in x}$ ^{[Kunen Ch. I]}. The proof builds approximations $F_{α}$ on $V_{α}$ by recursion on rank, shows any two approximations agree on their common domain by $\in$ -induction, and takes the union. Rank itself is the instance $G (x, h) = sup {h (y) + 1 : y \in x}$ , and the transitive closure, the von Neumann ordinals, and the constructible hierarchy $L_{α}$ of 42.03.06 are all defined this way. $\in$ -recursion is the universe-wide form of transfinite recursion: the well-foundedness of $\in$ guaranteed by Foundation is exactly the input that makes recursive definitions over $V$ legitimate.

Theorem 2 (the Lévy-Montague Reflection Principle). For every formula $φ (v)$ of the language of set theory and every ordinal $β$ , there is a limit ordinal $α > β$ such that $φ$ is absolute between $V_{α}$ and $V$ : for all $a \in V_{α}$ , $φ^{V_{α}} (a) \leftrightarrow φ (a)$ ^{[Kunen Ch. I]}. More strongly, for any finite list $φ_{1}, \dots, φ_{n}$ the class of such $α$ is closed and unbounded. The proof takes the finitely many subformulas, uses Replacement to define for each existential subformula a function $α \mapsto$ (least level adding a witness when one exists in $V$ ), and iterates $ω$ times to a closure ordinal at which every existential subformula true in $V$ already has a witness in $V_{α}$ ; a Tarski-Vaught test then yields absoluteness. Reflection is a theorem schema, one instance per finite formula list, and cannot be internalised to a single sentence " $V$ reflects every formula" without a truth predicate, which Tarski's theorem forbids.

Theorem 3 (non-finite-axiomatisability of ZF). ZF is not finitely axiomatisable: no finite subset of ZF (indeed no single sentence consistent with ZF) proves all of ZF ^{[Jech Ch. 12]}. Given a finite fragment $T \subseteq ZF$ , Reflection produces, provably in ZF, an ordinal $α$ with $V_{α} ⊨ T$ ; so ZF proves " $T$ has a set model," whence ZF proves $Con (T)$ . If some finite $T$ proved all of ZF, then $T$ would prove $Con (T)$ , contradicting Gödel's second incompleteness theorem (any consistent recursively axiomatised theory interpreting arithmetic cannot prove its own consistency). The same mechanism shows ZF proves the consistency of each of its finite fragments while — if consistent — never proving $Con (ZF)$ itself.

Theorem 4 (unprovability of $Con (ZFC)$ ). If ZFC is consistent, then ZFC does not prove $Con (ZFC)$ , and a fortiori does not prove the existence of a set model of ZFC, equivalently does not prove that any $V_{κ} ⊨ ZFC$ ^{[Jech Ch. 12]}. By Theorem 3, asserting "there is an $α$ with $V_{α} ⊨ ZFC$ " exceeds ZFC: it implies $Con (ZFC)$ . The least $κ$ with $V_{κ} ⊨ ZFC$ — when one exists — need not be inaccessible (it may be of cofinality $ω$ by a Löwenheim-Skolem-and-collapse argument), but the existence of an inaccessible $κ$ , giving $V_{κ} ⊨ ZFC$ for full second-order reasons, is a genuine large-cardinal strengthening beyond ZFC. This is the first rung of the large-cardinal hierarchy and the reason Reflection cannot be pushed to a single internal schema.

Theorem 5 (rank arithmetic and closure of $V_{α}$ ). For sets $x, y$ : $rank ({x, y}) = max (rank (x), rank (y)) + 1$ , $rank (P (x)) = rank (x) + 1$ , $rank ((x, y)) = max (rank (x), rank (y)) + 2$ , and $rank (⋃ x) \leq rank (x)$ . Consequently $V_{α}$ for $α$ a limit is closed under Pairing, Union, and Separation, and $V_{α}$ for $α > ω$ a limit additionally models Infinity; $V_{ω}$ is exactly the hereditarily finite sets and models ZFC minus Infinity. These rank computations are what make the levels usable as models and underlie the consistency arguments above.

Synthesis. The cumulative hierarchy is the central insight unifying every theme of foundational set theory: Foundation makes $\in$ well-founded, well-foundedness licenses $\in$ -recursion and $\in$ -induction, and these in turn construct the very ordinals that index the hierarchy, so the tower defines the apparatus that defines the tower. The foundational reason the universe is the increasing union $V = ⋃_{α} V_{α}$ rather than a completed object is that each level is a set while the whole is a proper class, and this is exactly what the Reflection Principle exploits: any finite portion of the theory of the proper-class $V$ is already true in some set-level $V_{α}$ , which generalises the naive idea of "taking a large enough stage" into a precise closed-unbounded reflection schema. Putting these together with Gödel's second incompleteness theorem, the schematic — never single-sentence — character of Reflection becomes inevitable: ZFC proves the consistency of each finite fragment of itself (by reflecting it into some $V_{α}$ ) yet cannot prove its own consistency, so it cannot prove that any single $V_{κ}$ models all of ZFC. The hierarchy is dual to the ordinal sequence that climbs it, and the bridge from this unit to the rest of the chapter is that ordinals 42.03.02, cardinals 42.03.03, the constructible universe $L$ 42.03.06, and forcing extensions are all constructions inside this tower, each one a controlled way of building or shrinking the levels $V_{α}$ while keeping the membership relation well-founded.

Full proof set Master

Proposition 1 (rank is well-defined and ordinal-valued). Under ZF, the assignment $rank (x) = sup {rank (y) + 1 : y \in x}$ defines a function $rank : V \to Ord$ , and $x \in V_{α} \leftrightarrow rank (x) < α$ .

Proof. Existence and uniqueness of $rank$ as a class function come from $\in$ -recursion (Theorem 1) with $G (x, h) = sup (ran (h ↾ x) + 1)$ ; the image is a set of ordinals by Replacement, so its supremum is an ordinal, and the values are ordinals by induction on rank. For the equivalence, argue by induction on $α$ . At $α = 0$ both sides fail. At a successor $α = γ + 1$ : $x \in V_{γ + 1} = P (V_{γ})$ iff $x \subseteq V_{γ}$ iff every $y \in x$ has $rank (y) < γ$ (induction) iff $sup {rank (y) + 1 : y \in x} \leq γ$ iff $rank (x) \leq γ$ iff $rank (x) < γ + 1$ . At a limit $λ$ : $x \in V_{λ}$ iff $x \in V_{γ}$ for some $γ < λ$ iff $rank (x) < γ < λ$ for some such $γ$ iff $rank (x) < λ$ . $□$

Proposition 2 ( $V_{α}$ is transitive and $\subseteq$ -increasing). For all ordinals $α \leq β$ , $V_{α}$ is transitive and $V_{α} \subseteq V_{β}$ .

Proof. Transitivity is Exercise 3. For monotonicity, induct on $β$ . If $β = γ + 1$ and $V_{α} \subseteq V_{γ}$ for $α \leq γ$ , then since $V_{γ}$ is transitive each $x \in V_{γ}$ satisfies $x \subseteq V_{γ}$ , so $x \in P (V_{γ}) = V_{γ + 1}$ ; thus $V_{γ} \subseteq V_{γ + 1}$ and composing gives $V_{α} \subseteq V_{γ + 1}$ . The case $α = β$ is identity; at limits $V_{α} \subseteq ⋃_{γ < λ} V_{γ} = V_{λ}$ for $α < λ$ directly. $□$

Proposition 3 (Reflection, single-formula form). For every formula $φ (v)$ and every ordinal $β_{0}$ there is a limit $α > β_{0}$ with $\forall a \in V_{α} (φ^{V_{α}} (a) \leftrightarrow φ (a))$ .

Proof. List the subformulas $φ = φ_{0}, \dots, φ_{m}$ closed under subformula. For each existential subformula $φ_{i} = \exists w φ_{j} (w, v)$ define $H_{i} (a)$ to be the least $γ$ such that $V_{γ}$ contains a witness $w$ with $φ_{j} (w, a)$ if any exists in $V$ , and $0$ otherwise; $H_{i}$ is a class function by Replacement. Define an increasing $ω$ -sequence $β_{0} < β_{1} < \dots$ by letting $β_{n + 1}$ exceed $β_{n}$ and exceed $sup {H_{i} (a) : a \in V_{β_{n}}, i \leq m}$ (a set, by Replacement). Let $α = sup_{n} β_{n}$ , a limit. Then for every $a \in V_{α}$ and every existential subformula, if a witness exists in $V$ one exists in $V_{α}$ . By induction on subformula complexity (the Tarski-Vaught condition), $φ_{i}^{V_{α}} (a) \leftrightarrow φ_{i} (a)$ for all $i$ and all $a \in V_{α}$ ; the case $φ_{i} = φ$ is the claim. Atomic and Boolean steps are immediate from $V_{α} \subseteq V$ ; the existential step uses witness reflection. $□$

Proposition 4 (reflection yields consistency of finite fragments). For each finite $T \subseteq ZF$ , $ZF ⊢ \exists α (V_{α} ⊨ T)$ , hence $ZF ⊢ Con (T)$ .

Proof. Let $T = {σ_{1}, \dots, σ_{k}}$ and let $φ = σ_{1} \land \dots \land σ_{k}$ . By Proposition 3 there is a limit $α$ with $φ$ absolute between $V_{α}$ and $V$ . Each $σ_{i}$ is a ZF theorem, so $σ_{i}$ holds in $V$ , hence $σ_{i}^{V_{α}}$ holds; thus $V_{α} ⊨ T$ . Since $V_{α}$ is a set, " $T$ has a model" is provable in ZF, and by the (arithmetised) completeness theorem the existence of a model entails $Con (T)$ , so $ZF ⊢ Con (T)$ . If a single finite $T$ proved all of ZF, then $T ⊢ Con (T)$ , contradicting Gödel's second incompleteness theorem; hence ZF is not finitely axiomatisable. $□$

Connections Master

The von Neumann ordinals are the transitive sets well-ordered by $\in$ , constructed by the $\in$ -recursion and rank machinery established here; they are co-produced as 42.03.02 and serve as the index set $Ord$ that climbs the cumulative hierarchy. The foundational reason ordinals and the hierarchy are inseparable is that $rank$ takes ordinal values and the ordinals are exactly the ranks of the well-ordered "spine" of $V$ , so the construction of the levels $V_{α}$ and the construction of the indices $α$ are a single transfinite recursion.
Cardinals and cardinal arithmetic 42.03.03 live inside this tower as distinguished ordinals (initial ordinals), and the rank-arithmetic and closure facts of Theorem 5 are what let $V_{κ}$ serve as a model when $κ$ is inaccessible; the large-cardinal hierarchy begins exactly where Theorem 4 stops, with the assertion " $\exists κ V_{κ} ⊨ ZFC$ " that ZFC cannot prove. This unit owns the ambient hierarchy; that unit owns the cardinal invariants measured inside it.
The Axiom of Choice, stated here as the ninth ZFC axiom, has its equivalents (Zorn's lemma, the well-ordering theorem, Tychonoff) and its independence developed in 42.03.05; the well-ordering theorem is what lets every set be assigned a position in the ordinal spine, making the rank-and-hierarchy picture compatible with the choice-driven constructions, and Gödel's constructible universe $L$ 42.03.06 is the choice-respecting inner-model refinement of the very cumulative tower defined in this unit, replacing $P (V_{α})$ by the definable power set at each successor step.

Historical & philosophical context Master

The axiomatic treatment of sets began with Ernst Zermelo's 1908 axiomatisation, motivated by the paradoxes of Russell (1901) and Burali-Forti and by the need to secure his 1904 well-ordering theorem. Zermelo's Separation (Aussonderung) replaced Frege's inconsistent unrestricted comprehension, blocking Russell's paradox by allowing only the carving of subsets from existing sets. Abraham Fraenkel and Thoralf Skolem independently added the Replacement schema in the early 1920s, completing what is now ZF; Skolem also clarified that the axioms are first-order schemas in the language with $\in$ .

The cumulative-hierarchy picture and the rank function are due to John von Neumann (1923-1928), who introduced the ordinals as transitive sets and defined the stratification of the universe by transfinite iteration of the power set; Dmitry Mirimanoff had anticipated the well-foundedness idea in 1917. Ernst Zermelo's 1930 paper Über Grenzzahlen und Mengenbereiche ^{[Zermelo 1930]} cast the universe as an open-ended sequence of models $V_{κ}$ indexed by inaccessible "boundary numbers," giving the modern picture in which reflection and large cardinals are continuous with the basic hierarchy. The Axiom of Foundation, isolating the well-founded universe, was formulated by von Neumann and Zermelo and shown by them to make $V = ⋃_{α} V_{α}$ .

The Reflection Principle in its schematic form was proved by Azriel Lévy and Richard Montague around 1960, and Lévy used it to establish that ZF is not finitely axiomatisable. Its connection to Gödel's 1931 second incompleteness theorem — that ZF proves the consistency of each finite fragment of itself yet, if consistent, never proves $Con (ZF)$ — is the precise sense in which the cumulative hierarchy is inexhaustible from within. Kunen's textbook treatment ^{[Kunen Ch. I]} and Jech's ^{[Jech Ch. 6]} are the standard modern developments of this material.

Bibliography Master

@book{kunen2011set,
  author    = {Kunen, Kenneth},
  title     = {Set Theory},
  series    = {Studies in Logic},
  volume    = {34},
  publisher = {College Publications},
  year      = {2011}
}

@book{jech2003set,
  author    = {Jech, Thomas},
  title     = {Set Theory},
  edition   = {3},
  series    = {Springer Monographs in Mathematics},
  publisher = {Springer},
  year      = {2003}
}

@article{zermelo1930grenzzahlen,
  author  = {Zermelo, Ernst},
  title   = {{\"U}ber Grenzzahlen und Mengenbereiche: Neue Untersuchungen {\"u}ber die Grundlagen der Mengenlehre},
  journal = {Fundamenta Mathematicae},
  volume  = {16},
  year    = {1930},
  pages   = {29--47}
}

@article{zermelo1908untersuchungen,
  author  = {Zermelo, Ernst},
  title   = {Untersuchungen {\"u}ber die Grundlagen der Mengenlehre I},
  journal = {Mathematische Annalen},
  volume  = {65},
  year    = {1908},
  pages   = {261--281}
}

@article{vonneumann1928axiomatisierung,
  author  = {von Neumann, John},
  title   = {Die Axiomatisierung der Mengenlehre},
  journal = {Mathematische Zeitschrift},
  volume  = {27},
  year    = {1928},
  pages   = {669--752}
}

@article{levy1960axiom,
  author  = {L{\'e}vy, Azriel},
  title   = {Axiom schemata of strong infinity in axiomatic set theory},
  journal = {Pacific Journal of Mathematics},
  volume  = {10},
  year    = {1960},
  pages   = {223--238}
}

@article{montague1961semantic,
  author  = {Montague, Richard},
  title   = {Semantic closure and non-finite axiomatizability I},
  journal = {Infinitistic Methods (Warsaw)},
  year    = {1961},
  pages   = {45--69}
}

@article{godel1931formal,
  author  = {G{\"o}del, Kurt},
  title   = {{\"U}ber formal unentscheidbare S{\"a}tze der Principia Mathematica und verwandter Systeme I},
  journal = {Monatshefte f{\"u}r Mathematik und Physik},
  volume  = {38},
  year    = {1931},
  pages   = {173--198}
}

Prerequisites

none — this is a leaf unit

Tier anchors

beginner: Halmos 1960 *Naive Set Theory* (Van Nostrand) §1-3 for the membership picture and Russell's paradox; Enderton 1977 *Elements of Set Theory* (Academic Press) Ch. 1-2 for the informal axiom tour read by hand
intermediate: Kunen 2011 *Set Theory* (College Publications) Ch. I §1-9 (the ZFC axioms stated one by one, classes vs sets, ordered pairs and relations set-theoretically, Foundation and ∈-induction); Jech 2003 *Set Theory* 3e (Springer) Ch. 1 for the parallel development
master: Kunen 2011 *Set Theory* (College Publications) Ch. I §14 (the cumulative hierarchy V_α, rank, ∈-recursion) and §15-16 (the Reflection Principle and its metamathematical consequences); Jech 2003 *Set Theory* 3e (Springer) Ch. 6 (the von Neumann hierarchy) and Ch. 12 (reflection); Zermelo 1930 *Fund. Math.* 16 (Grenzzahlen, models V_κ)

References

Kunen, K. — Set Theory · College Publications, Studies in Logic 34 (2011, revised edition). Chapter I develops axiomatic set theory: §1-2 motivate the axioms (why naive comprehension fails, Russell's paradox) and fix first-order logic with the single binary relation ∈ as the background language; §3-9 state and explain the ZFC axioms one at a time (Extensionality, Foundation/Regularity, Comprehension/Separation schema, Pairing, Union, Replacement schema, Power Set, Infinity, Choice); §6-7 develop classes versus sets (the proper class {x : φ(x)}, V as a proper class), ordered pairs à la Kuratowski, relations, functions, and orderings inside the theory; §14 constructs the cumulative hierarchy V_α (V_0 = ∅, V_{α+1} = P(V_α), V_λ = ∪_{α<λ} V_α), proves every set has a rank and that V = ∪_α V_α follows from Foundation, and develops ∈-induction and ∈-recursion; §15-16 prove the Lévy-Montague Reflection Principle (every finite list of formulas reflects to some V_α) and draw the connection to the unprovability of Con(ZFC) inside ZFC via Gödel's second incompleteness theorem.
Jech, T. — Set Theory · Springer Monographs in Mathematics, 3rd millennium edition (2003). Chapter 1 lists the axioms of ZFC and develops the elementary theory of sets, relations, functions, and orderings; Chapter 6 constructs the von Neumann cumulative hierarchy and the rank function, proving V = ∪_α V_α from the Axiom of Foundation and characterising the V_α as the levels of the well-founded universe; Chapter 12 proves the Reflection Principle in the Lévy-Montague form and uses it to show ZFC is not finitely axiomatisable and that no V_α (for α a limit of cofinality ω, say) is provably a model of all of ZFC.
Zermelo, E. — Über Grenzzahlen und Mengenbereiche · *Fundamenta Mathematicae* 16 (1930), 29-47. Zermelo introduces the cumulative-hierarchy picture of the set-theoretic universe stratified by the von Neumann ordinals, identifies the 'boundary numbers' (inaccessible cardinals κ) for which V_κ is a model of second-order ZFC, and frames the universe as an open-ended succession of such models — the source of the modern V_α stratification and the model-theoretic reading of reflection.

Estimated time

beginner: 20m
intermediate: 50m
master: 90m