2 Transitive Models

Observation: If M is transitive and Me is empty”, then e=. This is because if we, then we and eM gives us that wM by transitivity, so Mwe, so Me is not empty”.

Lemma. Assuming that:

  • M is transitive

Then
MExtensionality+Foundation.

Proof. Extensionality: xy(w(wxwy)x=y). Take xy, x,yM. Without loss of generality take zxy. Then zx and xM so zM. So Mzxzy. So M¬w(wxwy).

Foundation: x(xm(mxw¬(wmwx))). Take xM. Mx so x is not empty. So find mx which is -minimal. Then since xM as well, we have mM. Therefore x has an -minimal element in M.

2.1 Absoluteness for transitive models

Definition (Bounded quantifier). We call a quantifier of the form xy,φ or xy,φ a bounded quantifier.

(Defined by xy,φ:=x(xyφ) and xyφ:=x(xyφ)).

Definition (Closed under bounded quantification). A class of formulas Γ is closed under bounded quantification if whenever φ is in Γ, then so are xy,φ and xy,φ.

Definition (Delta0). Δ0 is the smallest class of formulas containing the atomic formulas that is closed under propositional connectives and bounded quantifiers.

Let T be any theory. Then Δ0T is the class of formulas equivalent to a Δ0 formula in T.

Theorem. Δ0 formulas are absolute for transitive models.

Proof. By induction:

Corollary. Assuming that:

  • T is any theory

  • MT is transitive

Then Δ0T-formulas are absolute for M.

Definition (Sigma1, Pi1). A formula is called Σ1 if it is of the form x1,xn,φ where φ is Δ0.

It is called Π1 if it is of the form x1,,xn,φ where φ is Δ0.

(same for Σ1T, Π1T).

Proof. Just definition of the semantics of ,.

Example. What is Δ0?

Definition (Absolute function). Let M a transitive set, and F:MnM (note that this means that M is closed under F). We say F is absolute for M if there is a formula Φ which is absolute for M such that

F(x1,,xn)=zΦ(x1,,xn,z).

Observation: So, if M is closed under pairing (x,yM,{x,y}M), then the pairing operation x,y{x,y} is absolute, and therefore MPairing.

Similarly for union.

Lemma. Assuming that:

Then ψ(x1,,xm):=φ(G1(x1,,xm),,Gn(x1,,xm))H(x1,,xm):=F(G1(x1,,xm),,Gn(x1,,xm))

are absolute for M.

Proof. Check the definitions!

Example. More examples:

Ordinals

x is an ordinal” means x is transitive and (x,) is a well-order.

We know: being well-founded is not expressible in first-order logic (see Example Sheet 1).

Because all transitive models satisfy Foundation, we have that if M is transitive, then

Mx is transitive(x,) is linearly ordered.

characterises ordinals. But this is clearly in Δ0.

So: being an ordinal is absolute for transitive models.

Thus MOrd={xM:Mx is an ordinal}. This is transitive, thus there is αOrd such that α=MOrd.

Also absolute:

Cardinals

x is a cardinal” if and only if

x is an ordinalf,yx,f:yxf is not a surjection

Note that f is not bounded (while yx is bounded).

Observe: this is Π1 and therefore downwards absolute.

Remark.

2.2 Non-absoluteness

Assume that MZFC is transitive and countable. Then

MOrd=α<ω1.

However, MZFC implies Mthere are uncountable cardinals.

Let β<α be such that Mβ is the least uncountable cardinal.

But β is a countable ordinal, so not a cardinal.

Consequence: All cardinals in M except 0 are going to be fake.

So “x is a cardinal” can’t be absolute.

Note. This also shows that “x=P(y)” cannot be absolute:

Take y such that My=P(ω). Then yP(ω), but is countable since yM.

Thus yP(ω). Therefore “x=P(y)” is not absolute.

Recall the general proof strategy mentioned before:

If M is a countable transitive set such that MZFC, then there is a a countable transitive set NM such that NZFC+¬CH.

Question: Is this really solving the original problem? i.e. Con(ZFC)Con(ZFC+¬CH).

It’s not obvious that Con(ZFC) implies that there is a countable transitive model (ctm) of ZFC.

Answer: That’s not only not obvious, but fake.

Let’s prove that Con(ZFC) there is a countable transitive model of ZFC.

Why? Note that Con(ZFC), or Con(T) for any T is Δ0. So, it’s absolute for transitive models.

So if M is a countable transitive model of ZFC, then Con(ZFC) is true, so by absoluteness, MCon(ZFC). So MZFC+Con(ZFC). This contradicts Gödel’s Incompleteness Theorem.

We can get a proof of

Con(ZFC)Con(ZFC+¬CH)

via a trick ( Example Sheet 1).

Lemma (Cohen Lemma). Assuming that:

  • TZFC

Then there is finite TZFC such that if M is a countable transitive model of T, then there is NM such that N is a countable transitive model of T+¬CH.

This reduces the problem to:

Find countable transitive models of T for sufficiently large finite TZFC.

Definition (Hierarchy). We call an assignment αZα a hierarchy if

  • (i) Zα is a transitive set
  • (ii) OrdZα=α
  • (iii) α<βZαZβ
  • (iv) λ limitZλ=α<λZα

If {Zα:αOrd} is a hierarchy, we can define Z:=αOrdZα. This is a proper class as OrdZ. We also define ρZ(x):=min{α:xZα}, a notion of Z-rank.

Paradigmatic example: von Neumann hierarchy Vα, and V is the entire universe.

Theorem 2.1 (Levy Reflection Theorem). Assuming that:

  • Z is a hierarchy

  • φ is a formula

Then there are unboundedly many 𝜃 such that φ is absolute between Z𝜃 and Z.

Proposition 2.2 (Tarski-Vaught Test). Assuming that:

  • M is a substructure of N

Then M is an elementary substructure if and only if for any formula ϕ(v,w¯) and a¯M, if there is bN such that Nϕ(b,a¯), then there is cM such that Nϕ(c,a¯).

TVTΦ:

Let MN and Φ be a collection of formulas closed under subformulas. Then the following are equivalent:

Warm-up: let (M,)ZFC. Find countable NM such that (N,)(M,).

Suppose p¯=(p0,,pn)M and My,ψ(y,p¯). Let w(ψ,p¯) be a witness for this:

Mψ(w(ψ,p¯),p¯)

(if necessary, use Axiom of Choice).

(if M¬y,ψ(y,p¯), then let w(ψ,p¯)=).

Set:

N0:=Ni+1:={w(ψ,p¯):ψ formula and p¯N1<ω}N:=iωNi

Note:

Remark. In general, even if M is transitive, N is not.

For example, if ω1M, then

x,(x is the least countable ordinal)

is true in M.

w(ψ,)=ω1.

So ω1N. But ω1N, since N is countable.

Relevant later!

Also see Example Sheet 1.

Proof of Levy Reflection Theorem. Fix φ and let Φ be its collection of subformulas. This is a finite set!

Need to show: α,𝜃>α such that Z𝜃φZφ.

For each ψΦ and p¯=(p0,,pn), write

o(ψ,p¯):={least α such that yZα with Zψ(y,p¯)if it exists0otherwiseo(p¯):=maxψΦo(ψ,p¯)𝜃0:=α+1𝜃i+1:=sup{o(p¯):p¯Z𝜃i<ω}𝜃:=supiω𝜃i

Then Tarski-Vaught Test implies that Z𝜃 and Z agree on φ.

Corollary. If TZFC is finite, then there is M transitive such that MT.

Proof. Let φ:=ψTψ. Since ZFCφ, we have that φ is true. By Levy Reflection Theorem, we can find 𝜃 such that V𝜃φ. Note V𝜃 is transitive.

Remark about the proof of Levy Reflection Theorem:

Can you do the same if Φ is infinite?

Of course not: otherwise we colud prove that there exists 𝜃 such that V𝜃ZFC, and hence get Con(ZFC).

The problem is the case distinction in the definition of o(ψ,p¯): it requires to check whether y,ψ is true.

Next goal: Obtain some MV𝜃 countable such that Mφ and M is transitive.

TODO

Theorem (Mostowski’s Collapsing Theorem). Let r be a relation on a set a that is well-founded and extensional. Then there exists a transitive set b, adn a bijection f:ab such that (x,ya)(x r yf(x)f(y)). Moreover, b and f are unique.

Proof. See Logic and Set Theory.

Corollary. For every TZFC finite, there is a countable transitive model of T.

Proof. Without loss of generality that T contains the axiom of extensionality. Form MT transitive by Levy Reflection Theorem.

Use warm-up to obtain NM countable. This is extensional and well-founded, so by Mostowski find W transitive such that

(W,)(N,).

Then WT adn |W|=||, so W is countable.

The next few lectures will be spent proving Con(ZFC+CH) using Gödel’s constructible universe.

Absoluteness is preserved under transfinite recursion.

Let F,G,H be three operations.

R(0,x¯):=F(x¯)R(α+1,x¯):=G(α,R(α,x¯),x¯)R(λ,x¯):=H(λ,{R(α,x¯):α<λ},x¯)()

Proof. Attempts: set functions satisfying the ().

R(α,x¯):=y if and only if there exists attempt r with (α,x¯)dom(r) and r(α,x¯)=y.

Note that for F,G,H fixed, there is a finite fragment TF,G,HZFC that proves the recursion theorem instance for F,G,H.

Theorem. If TTF,G,H and F,G,H are absolute for transitive models of T, then so is R defined by ().

Want TF,G,HL1(F,G,H), TF,G,HL2(F,G,H) and TF,G,H proves existence of R.

Convention: We say “T is sufficiently strong” if TZFC is finite and T proves hte existence of all relevant operations such that they are absolute for transitive models of T.

Proof. Observe that by assumption, being an “attempt” is absolute for transitive models of T.

Let MT be transitive.

Note. This uses the fact that “Δ1” concepts are absolute.

Definition (Delta1T property). A property is called Δ1T if it’s both Σ1T and Π1T.

Observe: Δ0T concepts are absolute (upwards from Σ1 and downwards from Π1).

Typical Applications

Bounding a quantifier by operation.

Let F be an operation and T strong enough to prove F is an operation and absolute.

Tx,z,F(x)=zTx,z,z,F(x)=zF(x)=zz=z

Then the quantifiers yF(x) and yF(x) preserve absoluteness.

yF(x)ψz(z=F(x)yzψ)absoluteupwards absolutez(z=F(x)yzψ)absolutedownwards absolute
Applications

2.3 The constructible hierarchy

Fix a set X. Define for each φFml and each pX<ω (parameter)

D(φ,p,X):={wX:Xφ(p,w)}

the subset of X defined by φ with parameter p.

For a sufficiently strong TZFC finite, we have that T proves that D is an absolute operation (see Example Sheet 1).

D(X):={D(φ,p,X):φFml,pX<ω}.

This is absolute for a sufficiently strong theory (use Replacement to get D(X)).

This D(X) is sometimes (misleadingly) called the “definable power set of X” (it is misleading because it is more like a “definable (by X) power set of X”).

(αD(X)φFml,pX<ω,a=D(φ,p,X))

Obvious: D(X)P(X). Also: If X is transitive, then so is D(X).

L0:=Lα+1:=D(Lα)Lλ=α<λLα The constructible hierarchy.

We usually write L:=αOrdLα.

Claim: L is a hierarchy (in the sense of Lecture). See Example Sheet 1.

By closure of absoluteness under transfinite recursion, the L-hierarchy is absolute for transitive models of TZFC where T is strong enough to prove that it exists.

i.e. if MT transitive and αOrdM and MX=Lα, then X=Lα. So

αOrdMLαM.

The main theorem of next lecture will be:

If LZF and MZF transitive, then

αOrdMLαZF.

(Minimal ZF-model).

Some first idea of what the L-hierarchy is like

Clearly, by induction, LαVα, and clearly for nω, Ln=Vn. So Lω=Vω.

Lα+1:=φFmlpLα<ω{D(φ,p,Lα)}.

If αω, then

|Lα+1|0|Lα<ω|=0|Lα|.

Thus |Lα|=|Lα+1|.

Therefore α<ω1, |Lα|=0 and |Lω1|=1.

This means: Vω+1Lω+1 (since the first has size 20, while the second has size 0).

Note: This does not mean VL. (V=L means x,α,xLα).

V=L is called the “axiom of constructibility”.

There is a finite fragment T𝕃 of ZF[!] that proves that all of the operations occuring in the definition of 𝕃, i.e. Fml,X<ω,,D,D are well-defined and absolute.

Thus, if M is a transitive model of T𝕃, then

αOrdM,𝕃αM

and thus 𝕃αM.

So αOrdM𝕃αM.

Axioms of ZF

Structural axioms:

Functional axioms:

Now we check that these hold in 𝕃.

In Lecture 2, we proved Extensionality and Foundation in all transitive structures, so also in 𝕃.

Note that ω satisfies the condition of the axiom of infinity, so any M transitive with ωM will satisfy the axiom of infinity. TODO

Now do pairing and union.

Since the definitions of pairs and unions

z={x,y}z=x

are absolute for transitive models, it’s enough to show that

x,y𝕃,{x,y}𝕃x𝕃x𝕃

If x,y𝕃α, φ(w,x,y):=w=xw=y,

D(φ,(x,y),Lα)={wLα:Lαφ(w,x,y)}={wLα:Lαw=xw}TODO

Powerset axiom.

x,p,w,(wpwx)

The problem here is that is not obviously absolute. In particular, z=P(x) is not absolute.

Consider 𝕃ω+1: we have ω𝕃ω+1 and P𝕃ω+1 is countable.

In 𝕃ω+2, we find

{a𝕃ω+1:aω}

which is the best possible answer to the question “what is the power set of ω?” that 𝕃ω+1 can give, but unlikely to be the correct answer.

Consider instead P(ω)𝕃=:P and define

Ω:={ρ𝕃(a):aP}.

(reminder: ρ𝕃(a) is the least α such that a𝕃α+1)

By Replacement, Ω is a set of ordinals, so find α>Ω. Then P𝕃α.

Therefore P={a𝕃α:aω}𝕃α+1, so P𝕃.

Separation:

p¯,x,s,w,(wswxφ(w,p¯)φ(w,x,p¯))

If x𝕃α, then

D(φ,x,Lα):={wLα:Lαφ(w,x,p¯)}={wLα:Lαwxφ(w,p¯)}=?{w𝕃not a problem:𝕃this is a problemwxφ(w,p¯)}

If φ is not absolute between Lα and L, this won’t work.

Levy Reflection Theorem to the rescue: φ,α,𝜃>α such that φ is absolute between L𝜃 and L.

Thus: form

D(φ,x,L𝜃)={wL𝜃:L𝜃wxφ(w,p¯)}=absolute{wL𝜃:Lwxφ(w,p¯)}

Replacement:

This will be on Example Sheet 2. The proof is a combination of the ideas from power set and separation.

Corollary 2.3 (Minimality). Assuming that:

  • T is a transitive model of ZF

Then for all αTOrd, 𝕃αT. Axiom of TODO.

Remark. Remark on the Axiom of Choice.

Gödel (1938): Con(ZF)Con(ZFC).

Note first that everything we did so far only needed ZF in the universe. We will sketch that 𝕃AC. In fact a strong version of AC known as GLOBAL CHOICE: there is an absolutely definable bijective operation between 𝕃 and Ord.

Sketch: Recursive construction of bijections πα:𝕃αηα for some ordinal ηα, and such that for β<α we have πα|𝕃β=πβ.

If λ is a limit and πα is defined for α<λ, then let

πλ(x):=πα(x)

if x𝕃α.

Suppose α=β+1 and πβ is given by πβ:𝕃βηβ.

Consider Fml×Lβ<ω. Well-order it in order type ηβ via the induced πβ well-order. Then if xLα, say

πα(x):={πβ(x)if xLβηβξif ξ is the ordinal corresponding to the least (φ,p¯) such that x=D(φ,p¯,TODO)

TODO

Cohen:

TZFC finite, TZFC finite such that if M is a countable transitive model of T, then there is NM countable transitive model of T+¬CH. ()

We have seen ( Example Sheet 1) that () implies Con(ZFC)Con(ZFC+¬CH).

Simplified: If M is a countable transitive model of ZFC, then there is NM countable transitive model of ZFCh=¬CH.

Idea: If M is a countable transitive model of ZFC: α:=ω1M; β:=ω2M; f:βP(ω) injection. Force N such that fN and MN.

Observe that there is a countable transitive model N such that fN and MN. Mtcl({f}) is transitive and countable. Thus LSM gives N transitive countable with Mtcl({f})N. Know Ng:βP(ω) is an injection, but no clue what 1 and 2 in N are.

How do we control what we add?