L (PhD Thesis Defense) -...

of 39 /39
Quantifier Rank Spectrum of L (PhD Thesis Defense) by Nathanael Leedom Ackerman April 19, 2006

Embed Size (px)

Transcript of L (PhD Thesis Defense) -...

  • Quantifier Rank Spectrum of

    L∞,ω (PhD Thesis Defense)by Nathanael Leedom Ackerman

    April 19, 2006

  • Definition 1. If L is a relational language then

    Lω1,ω(L) is the smallest collection of formulassuch that if φ(x) ∈ Lω1,ω(L) then

    • L ⊆ Lω1,ω(L)

    • ¬φ(x) ∈ Lω1,ω(L)

    • (∀y)φ(x) ∈ Lω1,ω(L)

    • (∃y)φ(x) ∈ Lω1,ω(L)

    and if {ψi(x) : i ∈ ω} ⊆ Lω1,ω(L) where⋃

    i∈ωFreeVariables(φi) is finite then

    • ∧i∈ω ψi(x) ∈ Lω1,ω(L)

    • ∨i∈ω ψi(x) ∈ Lω1,ω(L)

    1

  • Definition 2. If L is a relational language and

    φ(x) ∈ Lω1,ω(L) we define the quantifier rankof φ(x) (qr(φ(x))) by induction:

    • qr(R(x)) = 0 if R is a relation in L.

    • qr(¬φ(x)) = qr(φ(x)).

    • qr(∧i∈ω ψi(x)) = sup{qr(ψi) : i ∈ ω}.

    • qr((∀y)ψ(x)) = qr(ψ(x)) + 1.

    2

  • Definition 3. If M , N are models of the lan-

    guage L then we say M is equivalent to N up

    to α (M ≡α N) if and only if for all φ ∈ Lω1,ω(L)with qr(φ) ≤ α

    M |= φ ⇔ N |= φ

    Definition 4. We say that the quantifier rank

    of M (qr(M)) is α if α is the least ordinal such

    that for all models N of L

    M ≡α N ⇒ (∀β < ω1)M ≡β N

    3

  • We know, by the following theorem of

    Dana Scott, that in the case of countable mod-

    els this is notion is well defined and further the

    quantifier rank of any countable model is itself

    countable.

    Theorem 5 (Scott). If M is a countable model

    of the language L then there is a formula φMof Lω1,ω such that

    • M |= φM

    • For all models N of LN |= φM → N ∼= M

    Further, as we will only be interested in

    countable models for this talk we will assume

    all models are countable and all models have

    countable quantifier rank.

    4

  • Definition 6. Let φ ∈ Lω1,ω(L). We define thequantifier rank spectrum of φ (qr(φ)) to be

    {qr(M) : M |= φ ∧ |M | = ω}

    In this talk we will primarily be interested

    in particularly well behaved formulas.

    Definition 7. Let φ ∈ Lω1,ω. We say that φ isScattered if

    (∀α ∈ qr(φ))|{M : M |= φ ∧ |M | = ω∧ qr(M) = α}| = ω

    Theorem 8 (Morley). Let φ ∈ Lω1,ω. Then φis scattered if and only if |{M : M |= φ and M iscountable}| = ω or ω1 in all forcing extensionsof the universe.

    5

  • The main result of Part I of my thesis and

    the main result of this talk is

    Theorem 9. Let ω ∗α be a limit ordinal. Thenthere is a scattered sentence φω∗α such that

    • Quantifier rank of φω∗α ≤ ω

    • Quantifier rank spectrum of φω∗α is un-bounded in ω ∗ α

    • φω∗α is scattered.

    6

  • Definition 10. Let LP = {Pn : Pn is an n-arypredicate}.

    Definition 11. Let TP be universal closure of

    the following LP sentences:

    (∀i1, · · · in ∈ n)Pn(x1, · · · , xn)→ Pn(xi1, · · · , xin)

    Pn+1(x0, · · · , xn) → Pn(x1, · · · , xn)

    This theory puts a tree structure under

    subseteq (⊆) on the finite subsets of our model.

    7

  • Definition 12. Define the color of a ∈ M (‖a‖)as follows:

    • ¬P (a) ↔ ‖a‖ = −∞

    • P (a) ↔ ‖a‖ ≥ 0

    • If the tree extending a is wellfounded then‖a‖ = sup{‖ab‖ : b ∈ M}

    • ‖a‖ = ∞ otherwise.

    Definition 13. Let M |= TP . Then the Spec-trum of M (Spec(M)) = {‖a‖ : a ⊆ M}

    8

  • Definition 14. Let f : m → ORD such thatf(m +1)+1 = f(m). Then we say that f is a

    slow slant line.

    Definition 15. Let f be a slant line. We say

    that two tuples 〈ai : i ∈ n〉 and 〈bi : i ∈ n〉 arethe same up to f if for all S ⊆ n

    • ‖〈ai : i ∈ S〉‖ = ‖〈bi : i ∈ S‖〉

    • ‖〈ai : i ∈ S〉‖ > f(|S|) and ‖〈bi : i ∈ S‖〉 >f(|S|)

    9

  • Tuples

    Color

    ω ∗ α + 6

    ω ∗ α + 5

    ω ∗ α + 4

    ω ∗ α + 3

    ω ∗ α + 2

    ω ∗ α + 1

    ω ∗ α

    a1 a2 a3 a1a2 a2a3 a1a3 a1a2a3

    fg

    10

  • Definition 16. Let LR = LP ∪ {Ri,j≤ : Ri,j≤ is an

    i + j-ary predicate}.

    We will abuse notation and consider Ri,j≤ as apredicate of two arguments, one of arity i and

    one of arity j.

    Definition 17. Let TR be universal closure of

    the following LR sentences:

    TP

    R≤(x,y) ↔ [[¬P (x)] ∨ [P (x) → P (y)∧(∀a)(∃b)R≤(xa,yb)]]

    This expanded theory TR will be useful

    because we have the following theorem

    Theorem 18. If M |= TR and has no tuples ofcolor ∞ then M |= (∀a, b)R≤(a, b) ↔ ‖a‖ ≤ ‖b‖

    11

  • The “nice” scattered sentences will have what

    we call a Collection of Archetypes. The collec-

    tion of archetypes for a sentence T will consist

    of four pieces of information

    • A set AT(T ) of archetypes

    • A partial order 〈2−AT(T ),≤〉 of on certainpairs of archetypes (called consistent pairs

    of archetypes)

    • A collection BP(T ) of base predicates (alongwith consistent pairs of base predicates

    〈2−BP (T ),≤〉)

    • An “Extra Information” function EIT : AT(T )∪{M : M |= T} → X ×ORD

    12

  • Definition 19. Let T be our theory with nice

    properties in a language L. Further let M |=TP . Then define

    L(M) = L1 ∪ L2 ∪ {Q, R2≤} ∪ {ci : i ∈M}

    Definition 20. Let T (M) be universal closureof the following L(M) sentences:Q:

    • Q(x) ↔ ∨a∈M x = ca

    • Q |= φ(ca1, · · · can) in L2 iffM |= φ(a1, · · · an)

    • Q(x) ∧ ¬Q(y) → ¬U(x,y) where U is anypredicate other than R2≤ and |x| > 0

    13

  • L2 :

    • (∀x)(∃c)Q(c)R2≤(x, c)

    • (∀c)(∃x)¬Q(x) ∧R2≤(x, c)

    Other Axioms:

    • ¬Q |= T2

    • ¬Q |= T1

    • ¬Q |= P1(x) → P2(x)

    14

  • • Homogeneity:For all (A, A∗), (B, B∗) consistent pairs ofbase predicates such that (A, A∗) ≤ (B, B∗),and all m ∈ ω

    ¬Q |=[(∀x)[A1(x) ∧A2∗(x)] →(∃y1, · · ·ym)(B1(x,y) ∧B2∗ (x,y))]

    • Completeness:

    (∀x)(∃y)∨

    (A,A′)∈2−BP (T )(A, A′)(xy)

    15

  • We are going to want our theory T to have

    properties which allow us to prove the following

    Theorem 21. If

    • M, N |= T (M)

    • M |L1 ≡ω∗α N |L1

    • M |L2 ∼= N |L2

    then M ≡ω∗α NTheorem 22. If Spec(M) ⊆ Spec(M) thenthere is a model M ′ |= T (M) such that M ′|L1 ∼=M .

    Theorem 23. If ω ∗ α < Spec(M) which isa limit ordinal then there are M, N such that

    M ≡ω∗α N and Spec(M)∪Spec(N) ⊆ Spec(M).

    16

  • We are now ready to give our definition of a

    collection of archetypes

    (Truth on Atomic Formulas for Archetypes)

    If M |= φ(x) and N |= φ(y) where φ is anarchetype then for every atomic formula ψ,

    M |= ψ(x) iff N |= ψ(y).

    (Truth on Color)

    If φ ∈ AT(T ) and φ(x1, · · · , xn), φ(y1, · · · , yn)then ‖{xi : i ∈ S‖ = ‖{yi : i ∈ S‖ for all S ⊆ n.

    (Truth on Atomic Formulas for Base Predicates)

    If M |= A(x) and N |= B(y) where B is abase predicatethen for every atomic formula

    ψ, M |= ψ(x) iff N |= ψ(y).

    17

  • (Restriction of Arity for Archetypes)

    If φ is an archetype on a tuple x and y is a

    subset of x then we can restrict φ to y and get

    an archetype.

    (Completeness for Archetypes)

    If φ is an archetype which describes a tuple x

    and x∧y extends x then there is some archetypewhich describes x∧y.

    18

  • (Amalgamation for Archetypes)

    If φ and ψ are archetypes which agree on the

    what they force to be true on their common

    domain then there is a consistent extension of

    A and B which forces all “new” colors to be

    −∞.

    (Amalgamation for Base Predicates)

    If A and B are base predicates which agree on

    the what they force to be true on their com-

    mon domain then there is a consistent exten-

    sion of A and B which forces all “new” colors

    to be −∞.

    19

  • (Homogeneity for Base Predicates)

    If B is a base predicate which forces another

    base predicate A to hold and M |= A(a) thenthere are infinitely many extensions {bi : i ∈ ω}of a such that M |= A(a∧b)

    (Uniqueness of Base Predicate)

    This says that each tuple realizes at most one

    base predicate

    (Completeness of Extra Information)

    This says that the extra information predicate

    for a model is just the union of the extra infor-

    mation from each archetype which is realized.

    20

  • Now we come to two of the most important

    properties of a collection of archetypes.

    (Prediction)

    If σ, τ are archetypes such that τ(x,y) forces

    σ(x) then there is an archetype ητ(a) and a

    base predicate Aτ such that

    • M |= (∃x,y)τ(x,y) if and only if M |= (∃a)ητ(a)

    • (∀M |= T ) M |= [ητ(a)∧σ(x)∧Aτ(x,y, z, a)] →τ(x,y).

    If η(a)

    σ(x)

    A(x,y, z, a)

    then η(a)

    σ(x)

    A(x,y, z, a)

    τ(x,y)

    21

  • (Prediction up to a Slant Line)

    If σ, σ′, τ are archetypes such that

    • τ(x,y) forces σ(x)

    • σ and σ′ force the colors on their domainsto be the same up to a slant line sl

    • sl(1) = ω ∗ λ + |xy|+ n

    then there is an archetype ητ |sl(a) and a basepredicate Aτ |sl such that

    • If M |= (∃a)σ′(a) then M |= (∃b)(ητ |sl(b)

    • For all M |= T if M |= [ητ |sl(a) ∧ σ′(x) ∧Aτ |sl(x,y,z, a)]∧τ ′(x,y) then τ and τ ′ forcethe colors of the tuples they describe to be

    the same up to slant line sl

    22

  • (Consistency of Color)

    If (φ, φ′) is a consistent pair archetypes thenany color φ′ forces must be at least as large asthe color φ forces on the same tuple.

    (Consistency of ≤)≤ on consistent archetype pairs is transitiveand if (φ0, φ1) ≤ (ψ0, ψ1) then ψi is the restric-tion of φi to its domain.

    (Restriction of Arity for 2-Seq. of Archetypes)

    If (φ0, φ1)(x,y) ≤ (ψ0, ψ1)(x) and (ζ0, ζ1) is arestriction of (φ0, φ1)(x,y) to x,z with z ⊆ ythen (ζ0, ζ1) ≤ (ψ0, ψ1)

    23

  • (Amalgamation for 2-Sequences of Archetypes)

    If (φ0, φ1) and (ψ0, ψ1) are consistent pairs of

    archetypes which each force the same informa-

    tion on their common domain then the amal-

    gamations which give all “new” tuples color

    −∞ is also a consistent archetype pair and ≤(φ0, φ1) and (ψ0, ψ1).

    24

  • (Homogeneity of 2-Sequences of Archetypes)

    Suppose

    • (σ, σ′), (τ, τ ′), (η, η′) are consistent pairs ofarchetypes

    • (η, η′)(x,y) ≤ (σ, σ′)(x)

    • (η, η′)(x,y) forces (B, B′)(x,y)

    • (τ, τ ′)(x), (σ, σ′)(x) both force (A, A′)(x)

    (where A, A′, B, B′ are base predicates). Thenthere is a consistent pair of archetypes (ζ, ζ′)such that

    • (ζ, ζ′)(x,y) ≤ (τ, τ ′)(x)

    • (ζ, ζ′)(x,y) forces (B, B′)(x,y)

    25

  • (Completeness of 2-Sequences of Base Predicate)

    If (τ, τ ′) is a consistent sequence of archetypessuch that (τ, τ ′) forces (A, A′) and σ, σ′ arearchetypes such that

    • σ(x) forces A(x)

    • σ′(x) forces A′(x)

    • Every color which σ′ forces is at least asgreat as the color σ forces on the same

    tuple

    Then (σ, σ′) is a consistent pair of archetypes.

    26

  • (Extension of 0-Colors)

    Suppose (σ, σ′) is a consistent pair of archetypes.Further assume that τ ′(x,y) forces σ′(x). Then,if τ(x,y) forces σ(x) and forces all “new” tu-

    ples to have color −∞, (τ, τ ′) is a consistentpair of archetypes and (τ, τ ′) ≤ (σ, σ′)

    (Extension of 1-Colors)

    Suppose (σ, σ′) is a consistent pair of archetypes,τ ′(x,y) forces σ′(x) and there is some modelwhich realizes both τ and σ′. Then there isan archetype τ ′ such that (τ, τ ′) is a consistentpair of archetypes and (τ, τ ′) ≤ (σ, σ′)

    Tuples

    Color

    τ σ

    τ ′σ′

    Tuples

    Color

    τ σ

    τ ′σ′

    η

    η′

    27

  • Theorem 24. Let N |= T (M). If

    • (σ0, σ1), (τ0, τ1) ∈ 2−AT(T )

    • (τ0, τ1)(x,y) ≤ (σ0, σ1)(x)

    • τi is realized in N |Li

    then N |= (∀x)(σ0, σ1)(x) → (∃y)(τ0, τ1)(x,y).

    28

  • We start with

    σ(x)

    σ(x)′

    then we have

    σ(x)

    σ(x)′

    B(xy)

    B′(xy)′

    29

  • We know there exists

    σ(x)

    σ(x)′

    B(xy)

    B′(xy)′

    ητ(a)

    ητ ′(b)

    and we have by Prediction

    σ(x)

    σ(x)′

    B(xy)

    B′(xy)′

    ητ(a)

    ητ ′(b)

    A(axyz)

    A′(bxyz′)

    30

  • So we have

    σ(x)

    σ(x)′

    B(xy)

    B′(xy)′

    ητ(a)

    ητ ′(b)

    A(axyz)

    A′(bxyz′)

    E(abxyzz′)

    E′(abxyzz′)

    and this implies

    σ(x)

    σ(x)′

    τ(xy)

    τ ′(xy)′

    ητ(a)

    ητ ′(b)

    A(axyz)

    A′(bxyz′)

    E(abxyzz′)

    E′(abxyzz′)

    31

  • Theorem 25. If

    • (σ0, σ1),(σ′0, σ′1), (τ0, τ1) ∈ 2−AT(T )

    • (τ0, τ1)(x,y) ≤ (σ0, σ1)(x)

    • σ0|sl = σ′0|sl

    then there is a τ ′0 such that (τ ′0, τ1) ≤ (σ′0, σ′1)and τ ′0|sl = τ0|sl

    32

  • Lemma 26. If

    • M, N |= T

    • M ≡ω∗α N

    • Iη∗ω+n = {f : M → N s.t. f is a bijection,|dom(f)| < ω, there exists a slant line sl <(η + 1) ∗ ω such that if M |= σf(dom(f))and N |= τf(range(f)) then σf |sl = τf |sland where sl(|dom(f)|+ n) ≥ η ∗ ω}

    Then 〈Iη : η < ω ∗ α〉 is a sequence of partialisomorphisms which witness that M ≡ω∗α N .

    33

  • Theorem 27. If

    • M, N |= T (M)

    • M |L1 ≡ω∗α N |L1

    • M |L2 ∼= N |L2

    then M ≡ω∗α N

    Proof. Let Iω∗η+n = {f :

    • |dom(f)| < ω,

    • There exists a slant line sl < (η+1)∗ω suchthat if M |= (σ0, σ1)(dom(f)) and N |=(τ0, τ1)(range(f)) then σ0|sl = τ0|sl, τ1 =σ1 and where sl(|dom(f)|+ n) ≥ η ∗ ω}

    34

  • We know that Iη is non-empty for all η <

    ω ∗ α by the previous lemma, and by the pre-vious theorems we know (with out to much

    work) that in fact 〈Iη : η ∈ ω ∗ α〉 has the backand forth property and hence witnesses that

    M ≡ω∗α N

  • 35

  • Theorem 28. If Θ is as in “The Vaught’s Con-

    jecture: A Counter Example” then Θ has a

    collection of archetypes and Θ is scattered.

    Theorem 29. If M, N |= Θ, have no tuples ofcolor∞ and Spec(M)∩ORD,Spec(N)∩ORD ≥ω ∗ α then M ≡ω∗α N

    Theorem 30. If N,Mmodels Θ and if Spec(N) ⊆Spec(M) then there is a model N ′ |= Θ(M)such that N ′|L1 ∼= N and N ′|L2 ∼= M

    36

  • Theorem 31. IfM |= Θ and Spec(M) = {−∞}∪ω ∗ α then

    • Θ(M) has quantifier rank ω

    • Quantifier Rank Spectrum(Θ(M)) is un-bounded in ω ∗ α

    • Θ(M) is Scattered

    37