M4P49 Model Theory Mathematics Imperial College London These notes are based on a course of lectures given by Dr Pál during Autumn Term 2008 at Imperial College London. In general the notes follow the lectures very closely and only few changes were made mostly to make the typesetting easier. These notes have not been checked by Dr Pál and should not be regarded as ocial notes for the course. In particular, all the errors are made by me. However, I don't take any responsibility for their consequences; use at your own risk (and attend lectures, they are fun). Please email me at [email protected] with any comments or corrections. Anton Stefanek & James Haydon December 2008 Contents 1 Naive set theory 3 1.1 First steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Orderings and Ordinals 5 1.3 Cardinals of sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4 Hierarchy of Cardinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Languages and structures 10 2.1 Formulas and satisfaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.2 Examples of theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.3 Gödel's Compactness theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 3 Quantier elimination 26 2 1 Naive set theory 1.1 First steps All objects in naive set theory are sets; use capital letters for sets and small letters for sets contained in them. Use the shorthand x∈X for x is an element of X . The axioms/denitions of naive set theory: • We do not distinguish between sets containing the same elements, this is the axiom of extensiability: X = Y ↔ ∀x(x ∈ X ↔ x ∈ Y ). • Empty set: ∃X(∀y(y ∈ / X)). It can be shown that • X as above is unique, call it ∅. Denition of the subset relation: X ⊆ Y ↔ ∀x(x ∈ X → x ∈ Y ). • Denition of the union operation: ∀X∀Y ∃Z(x ∈ Z ↔ (x ∈ X ∨ x ∈ Y )). Can show that • Z as above is unique, it is the union of X and Y and is denoted X ∪Y. Unordered pairs: ∀x∀y∃Z(u ∈ Z ↔ (u = x ∨ u = y)). Denote the unordered pair by elements of a set X {x, y}. We will also x1 , x2 , . . . , then can be listed as write {x, x} as {x}. we denote this set as In general if the {x1 , x2 , . . . }. Construction of the natural numbers 1 = {∅}, 2 = 1 ∪ {1} = {∅, {∅}}, . . . n = n − 1 ∪ {n − 1}. More set theory • Ordered pair of x and y: (x, y) = {x, {x, y}}. • Union of a set: ∀X∃Y (z ∈ Y ↔ ∃y(y ∈ X ∧ z ∈ y)). Y is the union of the elements of the elements of X . Denote the union of X by ∪X . Note that ∪{x, y} = x ∪ y and so the rst denition of union can be made redundant. This notation is sometime used with an index set: {Xi | i ∈ I} or {Xi }i∈I . We say that the elements are indexed by I . • Intersection: ∩X = {x | ∀y(y ∈ X → x ∈ y)}. • Cartesian products: X × Y = {z | ∃x∃y(x ∈ X ∧ y ∈ Y ∧ z = (x, y))}. This is just the set of ordered pairs. • R • A is a on X × Y iff R ⊆ X × Y . R ⊆ X × Y is a function iff relation relation Write xRy for (x, y) ∈ R. relation function ∀x(x ∈ X → ∃!y(y ∈ Y ∧ xRy)). • A function f : X → ∪X is a choice function iff ∀x(x ∈ X → f (x) ∈ x). 3 choice function Axiom of choice. ∀x(x ∈ X → x 6= ∅) → ∃ The set of choice functions is denoted X × Y → Π{X, Y }. Two sets A and B are Πi∈I Xi . choice function f : X → ∪X . This is because there is a natural correspondance of the same size if they are in bijection. iX : X → P (X), y 7→ {y}. Note that there is a natural injection Theorem 1.1. There doesn't exist a surjective map s : X → P (X). Proof. s : X → P (X) be ∃a ∈ X, s(a) = Y . There are two By contradiction: let is surjective, • • if if a∈Y a∈ /Y then: then: a∈ / s(a) a ∈ s(a) surjective. Let Y = {x ∈ X | x ∈ / s(x)}. Because a∈ / Y. a∈Y. and so and so so both cases lead to a contradiction. Russel's paradox. Proof. The collection Again by contradiction: Let s cases: C of all sets is not a set. Y = {x ∈ C | x ∈ x} ⊆ C . So Y should be a well dened set, but • Y ∈Y • Y ∈Y Y ∈Y. Y ∈ / Y. implies that implies Both a contradiction. Theorem 1.2 (Bernstein). If there exists injections f : X → Y and g : Y → X then there exists a bijection h : X → Y . Proof. We need two lemmas. Lemma 1.3. Let X be any set and let h : P (X) → P (X) be a function such that ∀A ⊆ B ⊆ X, h(A) ⊆ h(B). Then there exists T ⊆ X such that h(T ) = T . Proof. Let T = ∪{A ⊆ X | A ⊆ h(A)}. • T ⊆ h(T ): If x ∈ T , then there is some A ⊆ X such that x ∈ A and A ⊆ h(A). A ⊆ T and so h(A) ⊆ h(T ), thus, x ∈ A ⊆ h(A) ⊆ h(T ). • h(T ) ⊆ T : From T ⊆ h(T ) we deduce h(T ) ⊆ h(h(T )). So by denition of T , h(T ) ⊆ T . Lemma 1.4. Let f : X → Y and g : Y → X be two maps. Let ∗ : P (X) → P (X) be given by the rule: ∗(X) = A∗ , A∗ = X − g(Y − f (A)). Then A ⊆ B ⊆ X implies that A∗ ⊆ B ∗ . Note that this formula is the complement of the complement. Drawing a picture helps. Proof. Suppose A ⊆ B ⊆ X. Then f (A) ⊆ f (B) ⇒ Y − f (A) ⊃ Y − f (B) ⇒ g(Y − f (A)) ⊃ g(Y − f (B)) ⇒ X − g(Y − f (A)) ⊆ X − g(Y − f (B)). 4 Now we can prove the theorem. Let Dene T h: X → Y : be a subset of ( h(x) = Then X f (x) g −1 |g(Y ) (x) such that if if T ∗ = T , i.e. T = X −g(Y −f (T )). x ∈ T, x ∈ X − T. X − T = g(Y − f (T )) ⊆ g(Y ). • f |T is injective (???), • g|Y −f (T ) is injective (???), So h is injective. It is also surjective (???) and hence bijective. inj inj Trichotomy principle ∀X∀Y (∃X → Y ∨ ∃Y → X). This has no simple proof as one needs the axiom of choice and lots of machinery about the ordinals. 1.2 Orderings and Ordinals A binary relation (i) (ii) (iii) ≤ X on partial ordering is a partial ordering ∀x, y ∈ X , if x ≤ x, (x ≤ y ∧ y ≤ x) → x = y , (x ≤ y ∧ y ≤ z) → x ≤ z . (P (X), ⊆). For example For all partial ordering ≤ we introduce another binary relation < dened as follows: x < y ↔ (x ≤ y ∧ x 6= y). This satises (i)' (ii)' ¬(x < x), (x < y) → ¬(y < x), and property (iii) as above. The relations ≤ and < uniquely determine each other, so we use both of them when we have a partial ordering. A partial ordering is a (iv) complete ordering Such a y Then (X, ≤) is well ordered if for every ∅ 6= Y ⊆ X, ∃y ∈ Y such that ∀x ∈ Y, y ≤ x. min(Y ) if it exists. Let (X, ≤) be a woset (= well ordered set). Let E⊆X such that E = X. Proof. For a contradiction, assume E= 6 X . Then X − E 6= ∅ and so let x = min(X − E). y < x, y ∈ E (otherwise x isn't minimum). by (i). Therefore, for all For two ordered sets (X, ≤), (X 0 , ≤0 ) a function f : X → X 0 is an order isomorpsim if f Then is bijective ∀x, y ∈ X , x ≤ y ↔ f (x) ≤ f (y). In this case we write Claim. Proof. Let (X, ≤) f: X ∼ =X be a woset, 0 . Y ⊆ X, f : X ∼ = (Y, ≤ ∩ Y × Y ). Then ∀x ∈ X, x ≤ f (x). E = {x ∈ X | f (x) < x} 6= ∅. Then if x0 = min(E), f (x0 ) < x0 since f (f (x0 )) < f (x0 ) as f is an order isomorphism. This shows that f (x0 ) ∈ E , but contradicts that x0 = min(E). Assume that x0 ∈ E . this well ordered min(X) ∈ E , ∀x ∈ X(∀y(y < x → y ∈ E) → x ∈ E). x 6= min(X) and complete ordering is unique so we call it the minimum and denote it Transnite induction (i) ordering ) if ∀x, y ∈ X, x ≤ y ∨ y ≤ x. The ordered set (ii) (or just an Then 5 order isomorpsim Claim. (X, ≤), (X 0 , ≤0 ) f: X ∼ = X 0. Let isomorphism Proof. be wosets. (X, ≤) ∼ = (X 0 , ≤0 ) then there exists a unique order f −1 ◦ g : X ∼ = X . So we have reduced the problem to X = X and we need to show that h = idX (i.e. that f = g ). By previous claim: x ≤ h(x), and −1 using the same argument with h gives us h(x) ≤ x. Hence by trichotomy h(x) = x. segment Let f: X ∼ = X 0, g : X ∼ = X 0. If Then 0 For any woset (X, ≤) and any a ∈ X , Xa is the segment of X dened as: Xa = {y ∈ X | y < a}. Claim. Proof. There does not exist an order isomorphism Suppose for a contradiction that one exists, f (a) ∈ / Xa , Claim. X∼ = Xa for any f: X ∼ = Xa . a ∈ X. By the above, a ≤ f (a), but then a contradiction. Let (X, ≤) be a woset and let A = {Xa | a ∈ X}. Then (X, ≤) ∼ = (A, ⊆). Proof. f : X ∼ = A is given by f (a) = Xa . ordinal A woset (X, ≤) the ordering on Claim. (i) (ii) Let α is an X ordinal if ∀a ∈ X, Xa = a. Let n is α be Let α For (ii), (α + 1)α = α. an ordinal. an ordinal, then for all Proof. Let b ∈ αa . b} = αb = b. Claim. Then a ∈ α, the segment αa is an ordinal. (αa )b = {x ∈ αa | x < b} = {x ∈ α | x < a and x < b} = {x ∈ α | x < be an ordinal. If Y $α and Y is an ordinal then Y = αa for some Proof. For the reverse inclusion let If but this is a contradiction. Hence Let a = min(α − Y ), then αa ⊆ Y . a < b then a ∈ αb so a ∈ Yb and a ∈ Y , b < a and b ∈ αa . Claim. If α, β Then also an ordinal. Proof. ∅ = αmin(α) = min(α∈ α, this proves (i). Claim. a < b ↔ Xa $ Xb ↔ a b. be an ordinal, then, α 6= ∅ → ∅ ∈ α. α + 1 = α ∪ {α} is For example, In this case is canonical and so we drop it from the notation. are ordinals then Proof. If a ∈ α ∩ β x < a} = (α ∩ β)a . then α∩β α a = a = βa , a∈Y. b ∈ Y . Then Yb = b = αb . b ≤ a. But b 6= a so is also an ordinal. i.e. {x ∈ α | x < a} = a = {x ∈ β | x < a} = {x ∈ α ∩ β | Theorem 1.5. Let α, β be ordinals. If α 6= β then one is a segment of the other. Proof. ¬(α ⊆ β ∨ β ⊆ α) ↔ (α ∩ β 6⊃ α ∧ α ∩ β 6⊃ β). So assume this is false, then α and of β (also a = αa = βb = b. α ∩ β, which is an ordinal (by above), must be an initial segment of by above). So there exists a ∈ α and b ∈ β such that α ∩ β = αa = βb . Then then a ∈ α ∩ β = αa ⇒ a < a, a contradiction. Hence Theorem 1.6. If α and β are ordinals, α ∼ = β , then α = β . 6 a = b ∈ α ∩ β. But Proof. Let f: α∼ = β. We show that f = idα . Let E = {x ∈ α | f (x) 6= x}. Assume that f (a), E 6= ∅ and let a = min(E). Then for all x < a, f (x) = x. Hence αa = βf (a) = a = a contradiction. Claim. For ordinals α and β, α $ β ↔ α ∈ β. Proof. • α $ β ⇒ α = βa | a ∈ β ⇒ α = βa = a ∈ β . • α ∈ β ⇒ α = βα $ β . Let On denote the collection of all ordinals. On Proposition 1.7. (i) On is well ordered w.r.t. ∈. (ii) On is not a set. NO PROOF PROVIDED Proof. ∈ is an ordering on On . (i)' α ∈ / α: α ∈ α ⇒ α = a ∈ α ⇒ α = αa ⇒ α ∼ = one of α's segments. (ii)' α ∈ β → β ∈ / α: Otherwise β ∈ α ∈ β ⇒ β $ β . (iii)' α ∈ β ∧ β ∈ γ → α ∈ γ : As β ∈ γ , β ⊆ γ and so α ∈ β ⊆ γ . Now we show it is well ordered: Let C be a non-empty collection of ordinals. dene D = {α ∈ β | α is in C} ⊆ β . didn't get what followed. (i) First we show that Let β∈C and Theorem 1.8. Let (X, ≤) be a woset such that for all a ∈ X the segment Xa is isomorphic to an ordinal. Then X is also isomorphic to an ordinal. Proof. ∀a ∈ X, ∃ga : Xa ∼ = Z(a) ordinals are equal and therefore where ga Z(a) is an ordinal. Z(a) Then is unique as isomorphic is also unique. Dene W = {Z(a) | a ∈ X}. Then W is a set, and Z : X → W, a 7→ Z(a) Claim. x, y ∈ X, x < y → Z(x) $ Z(y). Proof. Xx = (Xy )x ordinal Z(y) and so and so is also a function. gy |Xx : Xx ∼ = Z(y)gy (x) is an isomorphism. Z(y)gy (x) is a segment of an ordinal. So Z(y)gy (x) = Z(x) $ Z(y) by uniqueness of f . Z is an order isomorphism and so (W, ⊆) is (W, ⊆) is an ordinal. If x, y ∈ X and x < y , then Z(y)gy (x) = gy (x) since Z(y) is an ordinal. Z(x) = Z(y)gy (x) so Z(x) = gy (x). Thus From the claim we deduce that a woset. We now only need to show that WZ(y) = {Z(y) | Z(x) $ Z(y)} = {Z(x) | x < y}, [as Z We already saw that is an order iso.] = {gy (x) | x < y} = gy (Xy ) = Z(y). Theorem 1.9. Every woset is order isomorphic to a unique ordinal. Proof. Uniqueness is clear. Existence is proved using transnite induction. Let E = {a ∈ X | Xa is ∼ = to an ordinal}. (X, <) is some woset. By previous theorem this strategy works, i.e. it is enough to show E = X . So suppose E 6= X , then let x = min(X − E). Then for all a < x, Xa ∼ = an ordinal. Xa = (Xx )a hence Xx is an ordinal by previous and so x ∈ E , a contradiction. where that 7 cardinal limit ordinal ω Denition. An ordinal α is a cardinal if ∀β < α there is no bijection f : β → α. An ordinal α is a limit ordinal if α 6= β + 1 for any ordinal β . We denote the set of all nite ordinal by ω . Claim. (i) The elements of (ii) ω (iii) If Proof. to α. ω are cardinals. is a cardinal. κ is a cardinal and ω⊆κ then κ is a limit ordinal. Only prove (iii) since this is a naive theory. Let We dene f: α→α+1 f (0) = α, α ≥ ω. Need to show that α + 1 is bijective as follows: f (n) = n − 1 if 0 < n ∈ ω, and otherwise f (ξ) = ξ (w ≤ ξ < α). This is like in Hilbert's Hotel. α, β ∈ On . Let Dene an ordering < on {0} × α ∪ {1} × β by x, y ∈ {0} × α and x < y, x < y ⇔ x, y ∈ {1} × β and x < y, x ∈ {0} × α and y ∈ {1} × β. Claim. < is a well ordering. Proof. Let Otherwise ∅ 6= S ⊆ {0} × α ∪ {1} × β . If S ∩ {0} × α 6= ∅ then min(S ∩ {0} × α) S ⊆ {1} × β . Hence min(S) exists as β is well ordered. Now dene αuβ as the unique ordinal isomorphic to ({0} × α ∪ {1} × β, <). is min(S). Note that: • u on nitie ordinals coincides with usual addition. • (α u β) u γ = α u (β u γ), but not always α u β 6= β u α. partially ordered set chain upper bound maximal element partially ordered set is a pair (X, ≤) where X is a set and ≤ is a partial ordering on X . A subset is a chain if (Y, ≤) is an ordered set. An element x ∈ X is an upper bound for Y ⊆ X if ∀y ∈ Y, y ≤ x. A maximal element in X is an A Y ⊆X element x∈X such that ∀y ∈ X, x ≤ y → x = y . Theorem 1.10. The following are equivalent: AC ∀X((y ∈ X → y 6= ∅) → ∃ choice function f : X → ∪X). (Axiom of choice) ZL For all partially ordered set (A, ≤) for which ∅ 6= A and ∀Y ⊆ A, Y is a chain →∃ upper bound a∈A for Y, then A has a maximal element. (Zorn's lemma) WO Every set can be well ordered. (Well ordering principle) Proof will come later. 1.3 Cardinals of sets From now on we assume the axiom of choice. Proposition 1.11. for any set X there exists a unique cardinal λ such that there exists a bijection i : X → λ. 8 Proof. X i: X → λ λ (by previous theorem). Choose λ is a cardinal: 0 0 Assume it isn't, then there exists an ordinal λ < λ such that ∃j : λ → λ a bijection. Then −1 0 j ◦ i : X → λ is a bijection, a contradiction since λ is the smallest such ordinal. Uniqueness is a consequence of the fact that is κ, λ are cardinals that are in bijection then κ = λ. Indeed if κ < λ then λ cannot be a cardinal. can be well ordered, so smallest such λ λ The cardinal for some ordinal (ordinals are well ordered), then as above is called the cardinality of X. It is denoted cardinality |X|. |X| Proposition 1.12. Let X, Y be non-empty sets. Then The following are equivalent: (i) |X| ≤ |Y |, (ii) ∃ injection i : X → Y , (iii) ∃ surjection j : Y → X . (Write ' A→B to say that A and B ' A→B are in bijection.) Proof. ⇒ (i) (ii) ⇒ (ii) (iii) ' ∃i : X → |X|, as |X| ≤ |Y |, |X| ⊆ |Y | and hence there is an injection j : |X| → |Y |. ' is a bijection h : |Y | → Y . Now h ◦ j ◦ i : X → Y is injective. Let x0 ∈ W be arbitrary and let f : X → Y be a surjective map. Dene ( f −1 (y) if y ∈ f (X) g(y) = x0 otherwise. There f is injective, g is well dened and is surjective. j : Y → X be surjective. Then there exists a surjection f : |Y | → |X|. Let x ∈ |X| −1 and take g(x) = min(f (x)). So we have a well dened map g : |X| → |Y | that is injective −1 −1 (f (x) ∩ f (y) 6= ∅ if x 6= y ). Let U ⊆ |Y | be the image of this map g . Then U is a woset and so is order isomorphic to a unique ordinal α. As |X| is a cardinal, |X| ≤ α. Lemma 1.13. Let λ ∈ On , U ⊆ λ, α ∼ = U , α ∈ On . Then α ≤ λ. because (iii) ⇒ (i) Let Proof. Suppose λ<α i(x) ∈ U 0 , λ is isomorphic i: λ → U0 $ U. then some order isomorphism to and initial segment of Let x = min(λ − U 0 ). α and so there exists Then i(x) < x a contradiction. since Now we can prove the trichotomy principle: ∀X∀Y (∃X ,→ Y ∨ ∃Y ,→ X). Proof. Either 1.4 Hierarchy of Cardinals |X| ≤ |Y |, and then ∃X ,→ Y , or |Y | ≤ |X| and then ∃Y ,→ X . Lemma 1.14. For any cardinal κ there exists a cardinal λ such that λ % κ. Proof. Let 2κ = |P(κ)|. There is no surjection κ → P(κ), hence we don't have κ 2 > κ. As On and so For any cardinal κ let κ+ denote the smallest cardinal such that sup(U ) ≥ α for all κ+ > κ. Let U be a set of ordinals. α ∈ On − U . Let sup(U ) denote the smallest ordinal with the α ∈ U . We dene inductively, for every ordinal α ∈ On a cardinal ωα is not a set, there is some property κ ≥ |P(κ)| as follows: ω0 = ω, 9 κ+ sup(U ) and for β < α, ( ωβ+ ωα = supβ<α (ωβ ) Claim. Proof. There exists a unique way to attach to each α = β + 1, if otherwise. α an ωα such that the above holds. Both uniqueness and existence by usual transnite induction. 2 Languages and structures Denition. (i) (ii) (iii) rst order language L is the following data: + a set F , called the function symbols, and their arities nf ∈ N + a set R called the relation symbols, nR ∈ N for all R ∈ R, a set C , called the constant symbols. A We also assume that for all f ∈ F, F ∩ C = F ∩ R = R ∩ C = ∅. Examples are: (i) The language of rings Lring : F = {+, −, ·}, n+ = n− = n· = 2, R = ∅, C = {0, 1}. (ii) The language of orders, Lor : F = ∅, R = {<}, n< = 2, C = ∅. (iii) The language of sets Lsets : R = {∈}, n∈ = 2, F = {P(·)}, nP(·) = 1. Denition. Let L be a language. An L-structure M is the following: M , called the universe /domain /underlying set a function f : M nf → M for all f ∈ F , M a relation R ⊆ M nR for all R ∈ R, M an element c ∈ M for all c ∈ C . (i) a non-empty set (ii) (iii) (iv) We denote this by M, M = (M, f M , RM , cM ). For example, for all rings there is an Denition. Let respectively. An (i) of M Lring -structure. L be a language and let M and N be L-structures with universes M L-embedding η : M → N is an injection η : M → N such that η(f M (a1 , . . . , anf )) = f N (η(a1 ), . . . , η(anf )) 10 for all f ∈F and a1 , a2 , . . . , anf ∈ M, and N (ii) (iii) (a1 , a2 , . . . , anR ) ∈ RM iff (η(a1 ), . . . , η(anR )) ∈ RN η(cM ) = cN for all c ∈ C . L-embedding is called L-embedding, then we say that M A bijective Denition. The symbols an L-isomorphism. is a substructure (syntax) of the language of L for all If N R ∈ R, M ⊆ N and the inclusion N is an extension of M. map is an or are the following: F ∪ R ∪ C, v1 , v2 , . . . , vn , . . . , equation symbol =, boolean operators ∨, ∧, ¬, quantiers ∀ and ∃, parentheses ( and ), (i) the elements of (ii) variable symbols (iii) (iv) (v) (vi) (vii) comma ,. Denition. The set Denition. The set of (i) (ii) (iii) L∗ called the words L-terms Dene T0 L consits of all nite strings of is the smallest subset c ∈ T for all c ∈ C , vi ∈ T for all i ∈ N+ , if t1 , t2 , . . . , tnf ∈ T and f ∈ F Denition. of then T of L∗ L. Clearly L ⊆ L∗ . such that f (t1 , t2 , . . . , tnf ) ∈ T . as T0 := {c | c ∈ C} ∪ {vi | i ∈ N+ }. By induction on i, we dene for all i≥0 Ti+1 = Ti ∪ {f (t1 , . . . , tnf ) | f ∈ F Claim. T = Proof. S i∈N t1 , . . . , tnf ∈ Ti }. Ti . By induction on i we get Ti ⊆ T for all i satises the conditions (i)(iii) in the denition of For example, and ·(+(a, b), c) and therefore T and hence S i∈N S Ti ⊆ T. i∈N Ti . T ⊆ The set is a term in the language of rings. We will simplify it as S i∈N Ti (a + b) · c and similarly we will simplify all terms. That is, we will not write all the parentheses and use the inx notation for binary functions in Denition Lring . (Interpretation of terms) . Let M L-structure and let ū = (u1 , . . . , un ) be a ā = (a1 , a2 , . . . , an ) ∈ M n , an L-term s M an element of ū, we dene s (ā) as follows: be an nite list of variable symbols without repetition. Given an such that every variable (symbol) appearing in s s s (ii) is a constant symbol This is a recursive denition. i∈N such that Claim. Nn is M c, then s (ā) = c , M if is the variable ui (i = 1, . . . , n) then s (ā) = ai , if is the term f (t1 , . . . , tnf ) where f is a function symbol M M M then s (ā) = f (t1 (ā), . . . , tM nf (ā)). (i) if (iii) s M then Proof. of L and t1 , t2 . . . , tnf are terms, It can be seen that it is well-dened by considering the smallest s ∈ Ti . Let s, ū be as above and let η : M → N be an L-embedding. sN (η(ā)) is dened and sN (η(ā)) = η(sM (ā)). By induction on depth, that is minimal i such that 11 s ∈ Ti . If η(ā) = (η(a1 ), η(a2 ), . . . , η(an )) ∈ 2.1 Formulas and satisfaction Denition. (i) (ii) Let L be a language. We say that t1 = t2 , where t1 , t2 ∈ T , R(t1 , . . . , tnR ) where R ∈ R Denition. and L-formulas The set of φ ∈ L∗ t1 , t2 , . . . , tnf is an atomic L-formula if φ is either: are terms. is the smallest subset W of L∗ containing all atomic formulas such that ϕ ∈ W then ¬(ϕ) ∈ W , ϕ, ψ ∈ W then (ϕ) ∨ (ψ), (ϕ) ∧ (ψ) ∈ W , ϕ ∈ W then ∃x(ϕ) ∈ W and ∀x(ϕ) ∈ W . (i) if (ii) if (iii) if Claim. [Formula induction] Let if it is true for formulas variables in Proof. Let x. W0 Then Φ ϕ, ψ Φ be a claim which is true for atomic formulas and such that, then it holds for (ϕ) ∨ (ϕ), (ϕ) ∧ (ψ), ¬(ϕ), ∃x(ϕ) and ∀x(ϕ) for all is true for all formulas. be the set of all atomic formulas. For all i > 0, we dene by recursion Wi+1 := Wi ∪ {(ϕ) ∨ (ψ), (ϕ) ∧ (ψ), ¬(ϕ), ∃x(ϕ), ∀x(ϕ) | x a variable}. S W = i∈N Wi . Formula induction now follows from induction on i. Note that We now introduce abbreviations: abbreviation stands for ϕ→ψ ϕ↔ψ ϕ∧ψ ∀x(ϕ) ¬ϕ ∨ ψ (ϕ → ψ) ∧ (ψ → ϕ) ¬(¬(ϕ) ∨ ¬(ψ)) ¬(∃x(¬ϕ)) These make sense because the formulas are logically equivalent. So we may drop the cases and ∀x(ϕ) Denition. For all ϕ∈W we dene the set ϕ is an atomic formula, let V (ϕ) V (ϕ ∨ ψ) = V (ϕ) ∪ V (ψ) , V (¬ϕ) = V (ϕ), V (∃xϕ) = V (ϕ) − {x}. (i) if (ii) (iii) (iv) Claim. Proof. ϕ∧ψ when we prove something using formula induction. The set V (ϕ) V (ϕ) free variables of of ϕ recursively as follows: be the set of variable symbols in ϕ, is well-dened. By formula induction. What we need is that in the induction step, we can tell in which case (ii)(iv) we are in. Denition. A variable v appearing in the formula ϕ is bound if it appears below a quantier in the formation tree. A variable can be both free and bound, for example: Denition variables . (Truth of formulas) ū = (u1 , u2 , . . . , un ). Let ∃x(¬(x = 0)) ∧ (x · 0 = 0). Let M be an L-structure and let ϕ be a formula with free ā = (a1 , a2 , . . . , an ) ∈ M n . We recursively dene M |= ϕ(ā) as follows: (i) if (ii) if (iii) if (iv) if ϕ ϕ ϕ ϕ is is is is M t1 = t2 then M |= ϕ(ā) iff tM 1 (ā) = t2 (ā), M M R(t1 , . . . , tnR ) then M |= ϕ(ā) iff (tM 1 (ā), . . . , tnR (ā)) ∈ R , ¬ψ then M |= ϕ(ā) iff M 6|= ψ(ā), i.e. M |= ψ(ā) does not hold, ψ ∨ θ then M |= ϕ(ā) iff either M |= ψ(ā) or M |= θ(ā), 12 ϕ is ∃x(ψ) x∈ / V (ϕ)). (v) if M |= ϕ(ā) then M |= ϕ(ā) iff there exists b ∈ M is well-dened, that is, it is a unique relation for all Denition. ϕ∈W Formula is a sentence if V (ϕ) = ∅. M |= ψ(ā, b) such that (note that L-structures M. In this case we write M |= ϕ instead of M |= ϕ(ā). In words: M satises ϕ(ā) M. or ϕ(ā) is true in Denition. symbols An L-theory is a set of L-sentences. M |= T ) if M |= ϕ for all ϕ ∈ T . Denition. The full theory of an is denoted by Denition. L-structure M We say that is a model of the theory is the set of all sentences ϕ T (in M |= ϕ. It M ≡ N) if such that Th(M). Two L-structures M and N elementary equivalent are (in symbols Th(M) = Th(N ). Theorem 2.1. Suppose there exists an L-isomorphism j : M → N . Then M ≡ N . Proof by formula induction. Let ϕ V (ϕ) = (u1 , . . . , un ), ā = (a1 , . . . , an ) ∈ M n . be a formula, M |= ϕ(ā) ⇔ N |= ϕ(j(ā)) and 2.2 j(ā) = (j(a1 ), . . . , j(an )). This is true for atomic formulas. Examples of theories Complete Orderings L< : ∀x¬(x < x), R = {<}, n< = 2. Axioms: ∀x∀y∀z((x < y ∧ y < z) → x < z), Dense orderings without endpoints (DLO) L same as above with added axioms: ∀x∀y(x < y → ∃z(x < z ∧ z < y)), Abelian groups F = {+}, R = ∅, C = {0}, n+ = 2. ∀x(∃y(x < y) ∧ ∃z(z < x)). Axioms: ∀x(0 + x = x), ∀x∀y∀z(x + (y + z) = (x + y) + z), ∀x∀y(x + y = y + x), ∀x∃y(x + y = 0). Torsion-free divisible abelian groups For all ∀x∀y(x < z ∨ x = y ∨ y < x). 0<n∈N (DAG) L as in Abelian groups with: two axioms: ∀x((x + · · · + x = 0) → x = 0), | {z } n times ∀x∃y(y + · · · + y = x) | {z } n times and ∃x(¬(x = 0)). For example, Q is a model of this theory. Rings and Fields Lring = {+, −, · · · , 0, 1}. guage {0, +} Axioms for rings are as for abelian groups in lan- with: ∀x∀z(x − y = z ↔ x = y + z) ∀x(x · 0 = 0) ∀x∀y∀z(x(yz) = (xy)z) ∀x(x · 1 = 1 · x = x) ∀x∀y∀z(x(y + z) = xy + yz) ∀x∀y∀z((x + y)z = xz + yz) 13 elementary equivalent And for elds add: ∀x∀y(xy = yx) ∀x(x 6= 0 → ∃y(x · y = x)) Example Theories: Z is a ring, Algebraically closed elds Q, R, C are elds. (ACF) Field axioms with, for all n∈N such that n > 0: ∀a0 . . . ∀an−1 ∃x(xn + an−1 xn−1 + · · · + a0 = 0) Example: C. Ordered elds Lor = Lring ∪ L< . orderings L< Axioms are those of elds in Lring plus those of complete plus two additional axioms: ∀x∀y∀z((x < z) → (x + z < y + z)) ∀x∀y∀z((x < y ∧ z > 0) → (xz < yz)) Models: R and Denition. such that 2.3 Let T M |= T . Q for example. be an L-theory. We say that T is satisable if there exists an L-structure M Gödel's Compactness theorem Theorem 2.2 . T is satisable if and only if for all nite subsets S of T , S is satisable. (Gödel) Proof later, remark that (⇒) is trivial. Denition. Claim. X Let ≡U x Let I [x]R . ai ≡U i∈I When reference to the ultralter Claim. ≡U Q {Xi | i ∈ I} a collection i∈I Xi dened by be an index set and be the binary relation on Y Proof. is denoted is a disjoint union of equivalence classes. Denition. I. Equivalence class of Y bi ⇔ of sets. Let U be an ultralter on {i ∈ I | ai = bi } ∈ U. i∈I U is clear we just write ≡. is an equivalence relation. (All products and set denitions are over i ∈ I .) • I ∈ U so ≡ is symetric. • ≡ is obviously reexive. Q Q Q Q • Let ai , bi , ci ∈ Xi . Then {ai = ci } ⊇ {ai = bi } ∩ {bi = ci }. Q Q If ai ≡ bi then {ai = bi } ∈ U . Q Q If bi ≡ ci then {bi = ci } ∈ U . Hence {ai = bi } ∩ {bi = ci } ∈ U and so {ai = ci } ∈ U (both since U Q Q ai ≡ ci . Note that the proof didn't use that ultraproduct U was an is a lter). Thus ultra lter. Denition. Let {Mi | i ∈ I} be a collection of L-structures. We dene the L-structure M = Q ( i∈I Mi )/U , called the ultraproduct of Mi (w.r.t. U ) as follows: Q Q (i) the underlying set of ( i∈I Mi )/U is M = ( i∈I Mi )/U , Q M Mi (ii) for all constant symbols c ∈ L we dene c as the equivalence class of , i∈I c 14 L and equivalence classes ā1 , . . . , ānR ∈ M represented by Q Q Rn∈ nR M 1 1 2 R , we have (ā1 , . . . , ānR ) ∈ R if {i ∈ I | (ai , . . . , ai ) ∈ a , a , . . . , a i∈I i i∈I i i∈I i Mi R } ∈ U, for all function symbols f¯ ∈ L and equivalence classes ā1 , . . . , ānf ∈ M represented by Q Q Q n a1 , . . . , a f ∈ i∈I Mi we dene f M (ā1 , . . . , ānf ) to be the equivalence class of Qi∈I iMi 1 i∈I ni f (ai , . . . , ai ). i∈I f (iii) for every relation symbol Q (iv) Claim. M is well-dened. Proof. (a) (b) The claims (ā1 , . . . , ānR ) ∈ RM , f M (ā1 , . . . , ānf ) ∈ M , are independent of the choice of representatives. Take another set of representatives and the set Q i∈I b1i , . . . , Q i∈I nR or nf V = \ {i ∈ I | aki = bki } ∈ U. k=1 V ∩ {i ∈ I | (a11 , . . . , ani R ) ∈ RMi } = V ∩ A = V ∩ {i ∈ I | (b1i , . . . , bni R ) ∈ RMi } = V ∩ B . have A ∈ U iff A ∩ V ∈ U iff B ∩ V ∈ U iff B ∈ U and therefore (a). Now for all i ∈ V , we Then We have n n f Mi (a1i , . . . , ai f ) = f Mi (b1i , . . . , bi f ) and so Y f Mi (a1i , . . . ) ≡U i∈I Y f Mi (b1i , . . . ) i∈I and therefore (b). ϕQ∈ W with free ū = (u1 , . . . , un ). Let ā = (a1 , . . . , an ) ∈ M n . Choose Q variables 1 n 1 n n tives i∈I ai , . . . , i∈I ai for a1 , a2 , . . . , an respectively. Let a¯i = (ai , . . . , ai ) ∈ Mi . Q Theorem 2.3 (o± Lemma). We have ( i∈I Mi )/U |= ϕ(ā) if and only if Let representa- {i ∈ I | Mi |= ϕ(a¯i )} ∈ U. Proof. • By formula induction. If ϕ is an atomic formula t1 = t2 we have M M |= ϕ(ā) ⇔ tM 1 (ā) = t2 (ā) Mi i ⇔ {i ∈ I | tM 1 (a¯i ) = t2 (a¯i )} ∈ U ⇔ {i ∈ I | Mi |= ϕ(a¯i )} ∈ U. If ϕ is the atomic formula R(t1 , . . . , tnR ) then Mi Mi i M |= ϕ(ā) ⇔ {i ∈ I | (tM }∈U 1 (a¯i ), . . . , tnR (a¯i )) ∈ R ⇔ {i ∈ I | Mi 6|= ϕ(a¯i )} ∈ U. • If ϕ is the formula ¬ψ then M |= ¬ψ(ā) ⇔ M 6|= ψ(ā) ⇔ {i ∈ I | Mi |= ψ(a¯i )} ∈ /U ⇔ {i ∈ I | Mi 6|= ψ(a¯i )} ∈ U ⇔ {i ∈ I | Mi |= ¬ψ(a¯i )} ∈ U ⇔ {i ∈ I | M |= ϕ(a¯i )} ∈ U. 15 [by induction] [as U is an ultralter] bki • If ϕ is the formula ψ∨θ then M |= (ψ ∨ θ)(ā) ⇔ M |= ψ(ā) or M |= θ(ā) ⇔ {i ∈ I | Mi |= ψ(a¯i )} ∈ U or {i ∈ I | Mi |= θ(a¯i )} ∈ U ⇔ {i ∈ I | Mi |= ψ(a¯i )} ∪ {i ∈ I | Mi |= θ(a¯i )} ∈ U ⇔ {i ∈ I | Mi |= (ψ ∨ θ)(a¯i )} ∈ U. • If ϕ is the formula ∃xψ then M |= ∃xψ(ā) ⇔ exists b ∈ M such that M |= ψ(ā, b) Y Y ⇔ exists bi ∈ Mi such that {i ∈ I | Mi |= ψ(a¯i , bi )} ∈ U i∈I [by induction] i∈I ⇔ {i ∈ I | Mi |= ∃xψ(a¯i )} ∈ U. Corollary 2.4. If ϕ is a sentence, then M |= ϕ if and only if {i ∈ I | Mi |= ϕ} ∈ U . Denition. For every set I let I <ω = {S ⊆ I | |S| < ω}. Theorem 2.5. Let T be a theory. For every set S ∈ T <ω has a model MS . Recall Gödel's compactness theorem: A theory T is satisable if and only if for all nite subsets Proof of Gödel's compactness theorem. S of T, S is satisable. S ⊆ T let S have a model MS |= S . Let ϕ ∈ T let ϕ̂ = {S ∈ I | ϕ ∈ S} ⊆ I (all nite sub-theories of T in which ϕ appears). Take F = {ϕ̂ ⊆ I | ϕ ∈ T } ⊆ P(I). Then F has the nite intersection property: If ϕ ˆ1 , . . . , ϕˆn ∈ F then {ϕ1 , . . . , ϕn } is a nite sub-theory of T in which each of ϕ1 , . . . , ϕn appears, hence {ϕ1 , . . . , ϕn } ∈ ϕ̂i for all 1 ≤ i ≤ n, and so Tn Q {ϕ1 , . . . , ϕn } ∈ i=1 ϕ̂i . Hence there exists an ultralter U ⊇ F. Dene M = ( i∈I Mi )/U . For all ϕ ∈ T we have ϕ̂ ∈ F ⊆ U . Then for all i ∈ ϕ̂, Mi |= ϕ since Mi |= i and ϕ ∈ i. Hence ϕ̂ ⊆ {i ∈ I | Mi |= ϕ} ∈ U and then by the corollary to o± lemma, M |= ϕ. I = T <ω (all nite sub-theories of For any nite T) and for any Now an application. Denition. A graph G is a pair relation. The elements of that there is an edge V (V, E) are called between v and w where vertices if V is a set and E ⊆ V × V is a symmetric binary E are called edges. We say and the elements of vEw. Denition. Let 0 < k ∈ N, we say that G = (V, E) can be coloured with k colours if there exists f : V → k such that f (v) 6= f (w) for every v, w ∈ V such that vEw. We call f a colouring (with k colours). a function Denition. 0 E∩V ×V Clearly if with k 0 A subgraph of a graph (V, E) is another graph (V 0 , E 0 ) such that V0 ⊆ V and E0 ⊆ . G0 is a subgraph of G and G can be coloured with k colours, then G0 can be coloured colours. Denition. Graph G = (V, E) is nite if |V | < ω (and hence |E| < ω ). Theorem 2.6 (Erdos and de Bruin). A graph G can be coloured with k colours iff every nite subgraph of G can be coloured with k colours. Proof. Lgraph = {E}, nE = 2, E ∈ R. xRy → yRx. Let L be the language such that R = {E, R0 , . . . , Rk−1 }, i = 0, . . . , k − 1, F = ∅, C = {cg | g ∈ V } (where G = (V, E)). Let T be The only if part is trivial. For the oposite direction, dene The theory of graphs is nE = 2 and nRi = 1 for the L-theory consisting of the following formulas: 16 (o) (i) (ii) cg0 6= cg1 for g0 6= g1 . cg0 Ecg1 whenever there is an edge between g0 and g1 , V ∀x(R0 (x) ∨ R1 (x) ∨ · · · ∨ Rn−1 (x)) ∧ i6=j ¬(Ri (x) ∧ Rj (x)), each edge has a colour and only one. (iii) If ∀x∀y(xEy → M |= T then V i∈k M (M, E ¬(Ri (x) ∧ Ri (y)), ) is a graph. If colouring and under the map T is S⊆T We claim that g 7→ cM g , neighbours don't have the same colour. f : M → k such that f (v) = i iff M |= RiM (v) then f G can be considered as a subgraph of (M, E M ). is a satisable. By Gödels compactness theorem, we only need to show that every S contains (ii) and (iii) (this only makes it V 0 ⊆ V be the set of vertices g such that cg appears in any of the formulas S . Then V 0 is nite. Let G0 be the graph (V 0 , E ∩ V 0 × V 0 ). By denition, G0 is a subgraph G and hence by assumption there exists a colouring f : V 0 → k . Let MS be the following nite subset is satisable. We may assume that harder to satisfy). Let in of L-structure: • underlying set of MS is V , • cM g = g for all g ∈ V , • g0 Eg1 iff g0 , g1 ∈ V 0 and there is an edge between g0 and g1 0 of E ), • if g ∈ V 0 then g ∈ RiM iff f (g) = i, • if g ∈ / V 0 then g ∈ R0M and g ∈ / RiM for i = 1, . . . , k − 1. Obviously (either G0 or G by the denition Ms |= S . Ambrus Pál: I think it's the language courses that mess this room up. Not to mention that they move at the snails pace. Still needed stu about nite intersection property (FIP) for proof of Gödel's compactness theorem. Denition. Let I be an index set. Let G ⊆ P(I) be a system of sets. Then G has the nite intersection property if for every nite collection of elements of G, their intersection is not empty. Filters have the nite intersection property by denition. Lemma 2.7. Let G ⊆ P(I) be a system of sets having the FIP. Then there exists an ultralter U ⊆ P(I) such that G ⊆ U . Proof. G0 ⊆ P(I) having the FIP and G0 ⊇ G. Then A is partially ordered with respect to inclusion. If C ⊆ A is a chain, then ∪C ∈ A and therefore every chain has an upper bound and hence by Zorn's lemma A has a maximal element U ∈ A. We claim that U is an 0 0 ultralter. We have ∩∅ = ∅ and therefore ∅ ∈ U . If x, y ∈ U , take U = {x ∩ y} ∪ U : if U does not have the FIP then there exist x1 , . . . , xn ∈ U such that x1 ∩ · · · ∩ xn ∩ x ∩ y = ∅ and therefore U does not have the FIP as x1 , . . . , xn , x, y ∈ U . Therefore U 0 has the FIP and so U 0 ⊆ U as U is 0 0 maximal. Therefore x ∩ y ∈ U . If x ∈ U , take y ⊇ x and U = {y} ∪ U . If U does not have the FIP, then there exist x1 , . . . , xn such that x1 ∩ · · · xn ∩ y = ∅ and so also x1 ∩ · · · ∩ xn ∩ x = ∅, 0 0 a contradiction as U has the FIP. Therefore y ∈ U . So U is a lter. If U ⊇ U and U is a lter, 0 0 0 then U ∈ A and so U ⊆ U and so U = U and U is an ultralter. Let A Denition. be the set of all We will call |M | the cardinality of the L-structure M. Theorem 2.8. Let T be an L-theory. Assume that T has an innite model. Then for every cardinal κ there exists a model M of T such that |M | ≥ κ. Consider the formula ∃x1 ∃x2 . . . ∃xn ( ^ ¬(xi = xj ) ∧ ∀y( i6=j n _ l=1 17 (y = xi ))), saying I have exactly Proof. L0 be the L0 theory Let Dene an n elements. language L ∪ {cα | α ∈ κ} T 0 = T ∪ {cα 6= cβ | for all where cα α, β ∈ κ is a new constant symbol for all such that α ∈ κ. α 6= β}. T 0 has a model. We prove the latter by using compactness by showing 0 that every nite subset S ⊆ T has a model. Let I ⊆ κ be the set of all α ∈ κ such that cα appears in any formula S . Pick an innite model M of T . Then there is a map f : κ → M such that f |I 0 M is injective. We make M an L -structure by setting cα = f (α) for all α ∈ κ. Then M |= S . It is enough to show that Denition. Let η : M → N be an L-embedding. We say that η is an elementary embedding M |= ϕ(ā) iff N |= ϕ(η(â)) for every ϕ ∈ W with free variables (u1 , u2 , . . . , un ) and ā = (a1 , a2 , . . . , an ) ∈ M n . If M is a substructure of N and the inclusion map is an elementary embedding, then we write M ≺ N and we say that M is an elementary substructure of N and N is an elementary extension of M. if If there exists an elementary embedding substructure, we do not get η:M→N then M ≡ N. But if M≡N and M is a M ≺ N. Denition. Let M be an L-structure and let LM = L ∪ {cm | m ∈ M } be a language we get from L by adding a new constant symbol cm for all m ∈ M . Then M is automatically an LM -structure M where cm = m for all m ∈ M . The atomic diagram Diag0 (M) of M is the set of all variable free sentences ϕ of LM such that M |= ϕ. The elementary diagram Diag(M) of M is the set of all sentences ϕ of LM such that M |= ϕ. Obviously Diag0 (M) ⊆ Diag(M). Denition. with k We say that a graph G is k -chromatic for some positive integer k k − 1 colours. if it can be coloured colours and cannot be coloured with k , let Gk be the graph whose vertices is the set 2k + 1 and for any i, j ∈ 2k + 1, there is an edge between them iff |i − j| = 1 or |i − j| = 2k . Show that Gk is 3-chromatic. Let U be an ultralter on the set of positive integers which does not contain nite sets (nonQ principal ultralter). Prove that 0<k<ω Gk /U is 2-chromatic. Note that all vertices of Gk have degree 2. We can write this in the language of graphs: (i) For every positive integer (ii) ϕ := ∀x∃y0 ∃y1 (xEy0 ∧ xEya ∧ x 6= y0 ∧ x 6= y1 ∧ y0 6= y1 ∧ ∀x(xEz → z = y1 ∨ z = y1 )). Q We know that Gk |= ϕ and so by o± lemma, 0<k<ω Gk /U |= ϕ. The following formula say that I am not a circle of length at most n: ^ ^ xi Exi+1 ) → ¬(x0 Exn−1 ) θn := ∀x0 ∀x1 ∀x2 · · · ∀xn−1 ( xi 6= xj ) ∧ ( i∈n−1 i6=j Then GQ k |= θn if k > n. Gk /U |= θn . If U 3 {k ∈ ω | Gk |= θn } ⊇ {k ∈ ω | k > n} ∈ U and so by Lo± lemma Lemma 2.9. Let N be an L-structure such that N |= Diag0 (M). Then there is the L-embedding η : N → M. Proof. Dene η: M → N as η(m) = cN m m ∈ M . For all n, m ∈ M such that m 6= n, we have M |= ¬(cm = cn ). Therefore N N |= ¬(cm = cn ) and so cN m 6= cn and so η(m) 6= η(n). n M Let f ∈ F of L and m = (m1 , . . . , mnf ) ∈ M f , f (m1 , m2 , . . . , mnf ) = m0 and thereN fore M |= cm0 = f (cm1 , . . . , cmn ) and therefore N |= cm0 = f (cm1 , . . . , cmn ) and so cm = 0 f f N N N f (cm1 , . . . , cmn ). for all f 18 R ∈ R of L and m = (m1 , . . . , mnR ) ∈ M nR then (m1 , . . . , mnR ) ∈ RM iff M |= R(cm1 , . . . , cmnR ) N iff N |= R(cm1 , . . . , cmn ) iff (η(m1 ), . . . , η(mnR )) ∈ R . R If If c∈C in L, m = cM , M |= cm = c then for N |= cm = c and so and so η(m) = cN . Proposition 2.10. Let N be an LM -structure such that N |= Diag(M). Then there is an elementary embedding η : M → N . Proof. N |= Diag0 (M) which is a subset of Diag(M), for which we constructed an η : M → N . It is enough to show that this η is an elementary embedding. n 0 Let ϕ be a formula of L with V (ϕ) = (u1 , . . . , un ) and let ā = (a1 , . . . , an ) ∈ M . Let ϕ be the formula of LM we get from ϕ by writing cai for ui for all i. V (ϕ) is meaningful when we consider ϕ as a formula in L or LM . If I consider ϕ as an LM formula, still V (ϕ) = (u1 , . . . , un ). Clearly ϕ0 has no free variables and so is a sentence in LM . We know that embedding Lemma 2.11. We have N |= ϕ(ā) as an L-structure iff N |= ϕ0 (as an LM -structure). Proof. By routine formula induction. Now the proposition follows by applying the lemma both to M |= ϕ0 iff N |= ϕ0 iff M and N: We have M |= ϕ(ā) N |= ϕ(ā). iff Theorem 2.12 (Going up Lowenheim-Skolem Theorem). Let M be an innite L-structure. Then for every cardinal κ ≥ ω there exists an L-structure N such that there exists an elementary embedding η : M → N and |N | ≥ κ. Proof. an LM It is enough to construct an LM structure theory which has an innite model M N and so such that Diag(M) N |= Diag(M). But Diag(M) is |N | ≥ κ. has a model Lemma 2.13. Let M be an L-structure, ∅ 6= X ⊆ M and κ = max(|X|, |L|, ω). Then there exists an L-structure N < M such that X ⊆ N and |N | ≤ κ. Proof. Let N0 := X ∪ {cM | c ∈ C}. By recursion we dene Ni+1 := Ni ∪ {f M (x1 , . . . , xnf | xi ∈ Ni N= S and f ∈ F)}. Ni . Claim. |N | ≤ κ. Set Proof. i∈ω By induction we prove that |Ni | ≤ κ and so N → Q Ni injection and |N0 | ≤ |X| + |L| ≤ κ. QQ nf nf nf The induction step: Ni+1 → Ni =κ f ∈F Ni , |Ni | ≤ κ κ + κ · κ ≤ κ. | Q Ni | ≤ ω · κ ≤ κ: We have Construction of Let then cN ∈ N ), X⊆N This satises all the desired properties: (u1 , . . . , un ) |Ni+1 | ≤ |N0 | + |F| · κ ≤ N: • the underlying set is N , • cN := cM for all c ∈ C (if cM ∈ N0 • f N := f M |N nf for all f ∈ F , • RM := RM ∩ N nR . Claim. and and N ⊆ M. M ⊆ N be L-structures. Then for every atomic formula ϕ ∈ W ā = (a1 , . . . , an ) ∈ M n we have M |= ϕ(ā) iff N |= ϕ(ā). with V (ϕ) = and Lemma 2.14 (Tarski-Vaugh lent: • M ≺ N; . Let M ⊆ N be L-structures. The following are equiva- criterion) 19 • For all ϕ ∈ W with V (ϕ) = (u1 , . . . , un ) and ā = (a1 , . . . , an ) ∈ M n , there exists c ∈ N such that N |= ϕ(ā, c) Proof. ⇐⇒ there exists b ∈ M such that N |= ϕ(ā, b). M ≺ N . We need to show that if there exists c ∈ N such that N |= ϕ(ā, c) b ∈ M such that N |= ϕ(ā, b). Assume there is c ∈ N such that N |= ϕ(ā, c) implies N |= ∃vϕ(ā) and so M |= ∃vϕ(ā) as a1 , . . . , an ∈ M and M ≺ N and so there exists b ∈ M such that M |= ϕ(ā, b). Now we assume that the criterion holds and we will show by formula induction that for all ϕ ∈ W n with V (ϕ) = (u1 , . . . , un ) and ā = (a1 , . . . , an ) ∈ M we have M |= ϕ(ā) iff N |= ϕ(ā). Assume rst that implies there exists • For atomic formulas this is true because M ⊆ N and by the above claim. • ϕ is ¬ψ : M |= ϕ(ā) iff M 6|= ψ(ā) iff (by induction) N 6|= ψ(ā) iff N |= ϕ(ā). • ϕ is ψ ∨ θ: M |= ϕ(ā) iff M |= ψ(ā) or M |= θ(ā) iff (by induction) N |= ψ(ā) or N |= θ(ā) iff N |= ϕ(ā). • ϕ is ∃vψ : M |= ϕ(ā) iff there exists b ∈ M such that M |= ψ(ā, b) iff (by induction) N |= ψ(ā, b) iff (by the criterion) N |= ∃vψ(ā) iff N |= ϕ(ā). Theorem 2.15 (Going down Lowenheim-Skolem theorem). Let M be an L-structure. Let κ ≥ |L| + ω and let X ⊆ M such that |X| = κ. Then there exists N < M such that X ⊆ N and |N | = κ. Proof. For every ϕ ∈ W with V (ϕ) = (u1 , . . . , un , v) we say that f : M n → M is a Skolem function n for ∃vϕ if for all ā = (a1 , . . . , an ) ∈ M we have M |= ϕ(ā, f (ā)) if M |= ∃vϕ(ā). Claim. For all ϕ and v as above there is a Skolem function f for ∃vϕ. Proof. We have a well ordering ( f (ā) := < on M (axiom of choice) and min{b ∈ M | M |= ϕ(ā, b)} min(M ) if this set is not empty, otherwise. L0 = L ∪ {f∃vϕ } be the language we get from L by adding a new function symbol for all ϕ 0 and v as above, nf∃vϕ = n = #V (ϕ) − 1. Then |L | ≤ |L| + ω(|L| + ω) ≤ |L| + ω . We make M 0 an L -structure by interpreting the new function symbol f∃vϕ by a Skolen function for ∃vϕ, that M 0 is f∃vϕ = Skolem function for ∃vϕ. By the lemma above there is an L substructure N ⊆ M such that X ⊆ N and |N | ≤ max(|X|, |L|, ω) = κ and also |N | ≥ |X| = κ and so |N | = κ. Claim. As an L-structure, N ≺ M. Let Proof. If ϕ ∈ W with V (ϕ) = (u1 , . . . , un ) and ā = b ∈ M such that M |= ϕ(ā, b), then by the deniM M 0 tion of the Skolem function, M |= ϕ(ā, f∃vϕ (ā)). We have f∃vϕ (ā) ∈ N as N is an L -substructure and so there exists c ∈ N such that M |= ϕ(ā, c). We verify the Tarski-Vaugh criterion. (a1 , . . . , an ) ∈ N n such that there exists This proves the theorem. Corollary 2.16. Let T be an L-theory, assume that T has an innite model. Then for all cardinals κ ≥ |L| + ω there is an L-structure M such that M |= T and |M| = κ. Proof. subset exists We know that there exists an 0 X⊆M M ≺ M0 such that such that such that M0 |= T |X| = κ and apply the going down theorem |M| = κ. As M ≡ M0 we have M |= T . Denition. A (satisable) For example, Th(M) Denition. A sentence M |= ϕ. L-structure M0 L-theory T is complete if all models of T to and X |M| ≥ κ. Take a M to get: there and are elementarily equivalent. is complete. ϕ is a consequence T |= ϕ. of a theory We write this as 20 T if for every model M of T we have We remark that an Denition. κ An L-theory T L-theory T is is complete iff for every sentence κ-categorical for a cardinal κ ϕ, either T |= ϕ if any two models of T |= ¬ϕ. or T of cardinality are isomorphic. Proposition 2.17 (Vaugh's test). Let T be a satisable L-theory what is κ-categorical for some innite cardinal κ ≥ |L| and has only innite models. Then T is complete. Proof. M, N be models of T . As |M|, |N | ≥ ω , by the going down theorem there exist M0 , N 0 0 0 0 0 0 such that M ≡ M, N = N , |M | = |N | = κ. We have M |= T and so M |= T and also N |= T 0 0 0 0 0 and so N |= T and so by assumption M ' N (they are isomorphic) and so M ≡ N and M ≡ N. Let Denition. Dene ACFp = theory of algebraically closed elds of characteristic p, where p is a prime, with axioms • ACF and =theory Dene ACF0 • 1 + 1 + · · · + 1 = 0 (p ACF and times). of algebraically closed elds of characteristic ¬(1 + 1 + · · · + 1 = 0) any p 0 with axioms number of times. Theorem 2.18. The following theories are complete (i) dense orderings (DLO), (ii) divisible torsion-free abelian groups (DAG), (iii) algebraically closed elds of characteristic p, p = 0 allowed (ACFp ). Proof. In each case we use Vaught's test: ω -categorical: Let M and N be two models of cardinality ω0 . Let M = {m1 , . . . , mk , . . . } N = {n1 , n2 , . . . , nk , }. The following is called the back and forth argument. By induction, we dene subsets Xn ⊆ M and Yn ⊆ N and bijection fn : Xn → Yn such that (i) m1 , m2 , . . . , mn ∈ Xn and Xn is nite, (ii) n1 , n2 , . . . , nn ∈ Yn and Yn is nite, (iii) fn is an isomorphism, i.e. it is an order-preserving map, (iv) Xn ⊇ Xm , Yn ⊇ Ym if n ≥ m, (v) fn |Xm = fm for all n ≥ m. S S Once we have constructed these objects, we are done as M = n∈ω Xn and N = n∈ω Yn , S f = n∈ω fk is an isomorphism f : M → N . So take n = 1 and have X1 = {m1 }, Y1 = {n1 } and f (m1 ) = n1 . For the induction step, assume that Xn , Yn , fn are already constructed. (i) DLO is and Let Xn0 := Xn ∪ {mn+1 }. Claim. Proof. There exists an embedding Let a = mn+1 . If a ∈ Xn , fn0 : Xn → N such that fn0 |Xn = fn . then there is nothing to show. Otherwise there are three cases (i) (ii) a<b b<a b ∈ Xn ), b ∈ Xn ), b0 , b1 ∈ Xn (for all (for all such that b0 < a < b1 and for any b ∈ Xn , b ≤ b0 N |= DLO, there exists c ∈ N such that c < b for all b ∈ Yn , b < c for all b ∈ Yn , fn (b0 ) < c and c < fn (b1 ) 0 0 respectively. Let fn (b) = c, and so fn has the required properties. (iii) there exists Because Then we get Yn+1 := fn0 (Xn0 ) ∪ {nn+1 }. 21 or b1 ≤ b. Using the same argument we can construct an embedding (fn0 )−1 . gn+1 such that gn+1 |fn0 (Xn0 ) = Dene Xn+1 := gn+1 (Yn+1 ) ⊇ Xn0 ⊇ Xn and take fn+1 = (gn+1 )−1 . These satisfy the required properties. κ-categorical for all κ > ω : If G |= DAG then G is a Q-vector space as for all g ∈ G, ka/kb = a/bg = h and so k(a/(kb)g) = 1/bg and so |G| ≥ ω . Remark: G1 , G2 |= DAG, G1 ' G2 as groups iff G1 ' G2 as vector spaces. a a Dene for b ∈ Q, a, b ∈ Z, b g as the unique h ∈ G such that bh = ag . Hence the Q-vector space structure is an iso. invariant of G as a group. We need to show that if G1 , G2 are two Q-vector spaces, |G1 | = |G2 | = κ > ω then G1 ' G2 . Recall that X ⊆ G is a basis of G if (∗) every nite subset X is linearly independent, () every element g ∈ G can be written as a linear combination of some nitely many elements of X . (ii) DAG is Lemma 2.19. Every vector space has a basis. Proof. Let X be a linearly independent set (condition (∗)), which is maximal with respect to inclusion. There is such a set by Zorn's lemma. If g ∈ G such that g is not a linear combination Y = X ∪ {g} is linearly independent, a contradiction. there exists GP has a basis X , then |X| ≤ G ≤ |X| · |Q| g = xi ∈X λi xi , λi ∈ Q for all i and there is an If G→ a since every X is not a basis, then (in the sense of ()). Then g ∈ G can be written as injection |Q|S| | S⊆X If X Conclusion: (iii) S is P(X) hence |G| ≤ ω . If X ` andnso it is ` nite and ` n n n∈ω X and so | n∈ω X | ≤ | n∈ω |X || = n∈w |X| = ω|X|. a Q-vector space then |G| = κ > ω iff G has a basis X with is nite, then the index set is innite, then if S→ G is ` |X| = κ: ⇐ κ ≤ |G| ≤ ω · κ = ω + κ = max(ω, κ) = κ, ⇒ |X| = λ 6= κ and λ ≤ |G| ≤ max(ω, λ), a contradiction. Therefore if G1 , G2 have bases X1 , X2 with |X1 | = |X2 |, then there is a unique isomor0 0 phism ϕ : G1 → G2 for all bijections ϕ : X1 → X2 such that ϕ |X1 = ϕ. ACFp is κ-categorical for κ > ω (nite elds are not algebraically closed and therefore ACFp has no nite models). Let K, L be models of ACFp of cardinality κ. Let K = {aα | α ∈ κ}, L = {bα | α ∈ κ}. It will be sucient to construct Kα ≤ K and Lα ≤ L and fα : Kα → Lα (for all α ∈ κ) by transnite induction such that (i) {ap | β < α} ⊆ Kα , |Kα | ≤ |α| + ω , (ii) {cp | β < α} ⊆ Lα , |Lα | ≤ |α| + ω , (iii) Kβ ⊆ Kα , Lβ ⊆ Lα for all β < α, (iv) fα |Kβ = fβ for all β < α. Construction: For α = 0, let K0 and L0 be the prime elds of K and L respectively: Fp if p > 0 and Q otherwise. Therefore there exists an isomorphism f0 : K0 → L0 , |K0 | = |L0 | ≤ ω that satises all the required properties. For the induction step, assume that α Kβ , Lβ , fβ are all constructed for all β < α. First case: is a limit ordinal. Then Kα = [ Kβ , Lα = β<α [ Lβ β<α S |Kβ |, |Lp | ≤ |α| + ω for all β < α and therefore |Kα |, |Lα | ≤ |α| + ω . Take fα = β<α fβ and so fα is an isomorphism from Kα to Lα and (Kα , Lα , fα ) satises (i)-(iv): aβ ∈ Kα for and 22 β < α as there aβ ∈ Kβ 0 ≤ Kα . all The second case: fβ0 : Kβ0 → L Proof. (a) exists such that β < β0, since α is a limit ordinal and therefore α = β + 1. Let Kβ0 = Kβ (aβ ). Claim. fβ0 |Kβ = fβ . There exists embedding such that Two cases: g ∈ Kβ [x] be the monic polynomial which is the minimal fβ (g) ∈ Lβ [x] is an irreducible polynomial and so there exists b ∈ L which is a root of fβ (q) and there exists unique fβ0 : Kβ0 → Lβ (b) isomorphism 0 0 such that fβ (aβ ) = b and hence by construction fβ |Kβ = fβ . aβ is algebraic over polynomial for (b) β0 < α aβ . Kβ . Let Then aβ is transcendental over Kβ . Let Lβ = algebraic closure of Lβ in L. Then |Lβ | ≤ |Lp |+ω ≤ |β|+ω < κ and therefore there exists b ∈ L−Lβ such that b is transcendental 0 0 0 over Lβ and there exists a unique isomorphism fβ : Kβ → Lβ (b) such that fβ (aβ ) = b 0 and fβ |Kβ = fβ . |Lα | ≤ |Kβ0 | + ω ≤ |α| + ω . Using the same argument 0 −1 . Set as above we can construct an embedding gα : Lα → K such that gα |f 0 (Kβ ) = (fβ ) β −1 Kα = gα (Lα ), fα = (gα ) . Hence (iii)-(iv). As aβ ∈ Kα , bβ ∈ Lα we get (i) and (ii). Let Lα = fβ0 (Kβ0 )(bβ ). By construction Lecturer quote: This was quite a lot of work to get to this, compactness, going up, going down, but maybe you don't apreciate how powerful it is. theorem proving machine. This theorem is a machine: a Lemma 2.20. Let T be an L-theory and ϕ be a sentence in L such that T |= ϕ. Then there exists a nite S ⊆ T such that S |= ϕ. Proof. that S ⊆ T there exists L-structure M such S ∪ {¬ϕ} is satisable. By compactness, T 0 = T ∪ {¬p} is satisable and M of T such that M |= ¬p, a contradiction. Assume for contradiction that for all nite sets M |= S ∪ {¬ϕ}, i.e. so there exists a model Theorem 2.21 (Theorem of Ax). Let ϕ be a sentence in Lring . Then the following are equivalent: (i) ACF0 |= ϕ, (ii) C |= ϕ, (iii) for innitely many prime numbers p, Fp |= ϕ, (iv) for all but nitely many primes p ACFp |= ϕ. Proof. C |= ACF0 . (ii) implies (i) as ACF0 is complete. |= ϕ then there exists S ⊆ ACF0 such that |S| < ω N ∈ N such that for every prime p > N , ACFp |= S . This implies (i) implies (ii) as For (i) implies (iv), if ACF0 and S |= ϕ and so there exists that ACFp |= ϕ for all ϕ > N. For (iv) implies (iii): as Fp |= ACFp For (iii) implies (i): Assume ACF0 ACFp |= ¬ϕ and hence 6|= ϕ. Fp |= ϕ p/ |= ¬ϕ as ACF0 is complete. Therefore Fp |= ¬ϕ for all but nitely for all but nitely many Then ACF0 for all but nitely many primes. contradiction. a Let F be a eld. A polynomial map Φ over F is an n-tuple Φ = (Φ1 , Φ2 , . . . , Φn ) Φi ∈ F ∈ [x1 , x2 , . . . , xn ] for i = 1, 2, . . . , n. Then Φ furnishes a map F n → F n , also by Φ, given by the rule such that Φ(z̄) = (Φ1 (z̄), Φ2 (z̄), . . . , Φn (z̄)) for all p, Denition. denoted Therefore many z̄ = (z1 , z2 , . . . , zn ) ∈ F n , 23 Theorem 2.22. Let Φ be a polynomial map over C. If Φ : Cn → Cn is injective, then it is surjective. Proof. Claim. Φ Let be a polynomial map over Fg . If Φ : Fng → Fng is injective, then it is surjective. Proof. In fact, any map Claim. Φ Let Fng → Fng injective is surjective as be a polynomial map over Fp . If |Fng | < ω . n Φ : Fp → Fp n is injective, then it is surjective. Proof. We have Φ = (Φ1 , Φ2 , . . . , Φn ) and so there is a power g of p such that Φi ∈ Fg [x1 , x2 , . . . , xn ] i = 1, 2, . . . , n. By the above, Φ : Fngm → Fngm is surjective for all m ∈ N. Therefore n n Φ : Fp → Fp is also surjective as [ n Fp = Fngm . for m∈N Let Φ Φi = be as above and P j aij x̄j , where J : {1, 2, . . . , n} → N J(1) x̄j := x1 Let |J| > k . Claim. We say that We also say that Φ = (Φ1 , . . . , Φn ) Φ has n variables. There exists a sentence for any polynomial map it is injective. Denition. Φ with n ϕn,k is in C C of in has degree Lring ≤ k if aij = 0 for all i = 1, . . . , n and F we have F |= ϕn,k iff Φ : F n → F n is surjective, if such that for every eld variables and degree By the theorem of Ax, A class L-structure M nitely axiomatisable · · · xJ(n) . n |J| = maxnl=1 J(i). Denition. axomatisable J(2) · x2 is a multiindex and ≤k the map C |= ϕn,k . L-structures is axomatisable M |= T . if there is an L-theory T such that an if and only if T axiomatises the class C . A class C of L-structures is nitely axiomatisable L-theory T which axiomatises the class C . In this case we say that if there is a nite Lemma 2.23. Assume that the class of models of the L-theory T is nitely axiomatisable. Then there is a nite subset S ⊆ T which axiomatises the class of the models of T . Proof. Let {ϕ1 . . . . , ϕn } axiomatise this class. Then {ϕ} also axiomatises the ϕ1 ∧ · · · ∧ ϕn . By denition, T |= ϕ. So by compactness there exists nite S ⊆ T We claim that S axiomatises the class: • If M |= T then M |= S : M |= T ⇒ M |= S is easy. For M |= ϕ ⇒ M |= T (as {ϕ} axiomatises models of T ). • If M |= S then M |= ϕ. ϕ = S |= ϕ. class where such that the reverse implication M |= S ⇒ Lecture quote: I have not had enough coee so was doing the proof slowly and this made it harder to prove. Theorem 2.24. The following classes are axiomatisable but not nitely axiomatisable. (i) Torsion free abelian groups. (ii) Fields of characteristic 0. (iii) Algebraically closed elds. Proof. 24 (i) Recall that the axioms {ϕn | n ∈ ω} where ϕn T is for torsion free abelian groups are those of abelian groups and ∀x((x + · · · + x = 0) → x = 0) (n summands). T axiomatises the class. S⊆T I = {n ∈ ω | ϕn ∈ S}, this set is nite. So there is a prime p > n for all n ∈ I . Then Z/pZ |= S . Let T be the axioms of elds with {ϕp | p a prime}. This axiomatises the class of elds of characteristic 0, where ϕn is ¬(1 + · · · + 1 = 0) (p summands). Suppose it is nitely axiomatisable, say nite S ⊆ T , (using lemma) S axiomatises the class, then take I = {p prime | ϕp ∈ S}, then I is nite. and there is a prime l > p for all p ∈ I . Then Fl |= S . Assume that the class is nitely axiomatisable. Then by the lemma there is a nite which axiomatises the class. Let (ii) (iii) ACF axiomatises this class. Using same argument as above, it is enough to construct a eld Kp p such that: Kp has no non-trivial nite extensions of degree < p. (b) Kp has an extension of degree p. Let Kp = {α ∈ F̄l | p does not divide [Fl (α) : Fl ]} ⊆ F̄l . Claim. Kp ⊆ F̄l . for every prime (a) Warning: High doses of Galois theory ahead! Lecturer Quote: If we actually want to prove something big about Algebra with model theory then the proof often uses some deep Algebra. Proof of claim. [Comments in square brackets are my own additions, might be wrong] [Let α, β ∈ Kp , we want to show that the eld operations between α, β Fl (α, β).] We have the following diagram of extensions: keep them in Kp , so study Fl (α, β) II u II u II uu u II u u I u u Fl (α) Fl (β) JJ t JJ t t JJ tt JJ tt JJ t tt Fl Where Fl (α)|Fl and Fl (β)|Fl are nite Galois. So [by Fl (α, β)|Fl is nite Galois and there exists an [injective a theorem of Galois theory course] group morphism]: Gal(Fl (α, β)|Fl ) → Gal(Fl (α)|Fl ) × Gal(Fl (β)|Fl ) (1) γ 7→ γ|Fl (α) × γ|Fl (β) (2) γ(Fl (α)) ⊆ Fl (α) and γ(Fl (β)) ⊆ Fl (β).) For some γ ∈ Gal(Fl (α, β)|Fl ), γ|Fl (α) = id then γ(α) = α. If γ|Fl (β) = id then γ(β) = β . Then γ = id as α and β generate the extension Fl (α, β)|Fl . [Hence injectivity.] So [Fl (α, β) : Fl ] divides [Fl (α) : Fl ][Fl (β) : Fl ] by the fundamental theorem of Galois (Note that If theory [Galois theory tells us that these dimensions are the sizes of their Galois groups and group theory says that the injective image is a subgroup that then divides the cardinality of the α, β are assumed to be in Kp ], p does not divide [Fl (α)Fl ][Fl (β) : [Fl (α, β) : Fl ] either. Hence all elements of Fl (α, β) are in Kp . [Indeed if η ∈ Fl (α, β) then [Fl (η) : Fl ] divides [Fl (α, β)|Fl ] (tower law). So p does not divide [Fl (η) : Fl ] and so η ∈ Kp . In particular this holds for α + β, αβ, α−1 , −α, so Kp is a eld, a subeld of F̄l .] group by Lagrange]. [Since Fl ] and so does not divide Claim. If K 0 ≤ Kp is a nite extension of 25 Fl then p does not divide [K 0 : Fl ]. Proof of claim. tion By the primitive element theorem, 0 [K : Fl ] = [Fl (α) : Fl ] Claim. If L ≤ F̄l Proof of claim. is not divisible by is a nite extension of Kp K 0 = Fl (α) for some α ∈ Kp . By deni- p. and p does not divide [L : Kp ] then L = Kp . By the primitive element theorem, or by induction on the number of gen- L as an extension of Kp , we may assume that L = Kp (α) for some α ∈ F̄l . n Let f (x) = a0 + a1 x + · · · + an x ∈ Kp [x] be a minimal polynomial of α [over Kp ]. Let 0 K = Fl (a0 , . . . , an ) ≤ Kp . Since [K 0 : Fl ] is nite [nitely generated by algebraic elements] it 0 is not divisible by p by the previous claim. Considering f (x) as an element of the ring K [x], 0 0 it is irreducible as it is in Kp [x]. Also p does not divide [Kp (α) : Kp ] = deg(f ) = [K (α) : K ] 0 0 0 [by assumption]. Now, by the tower law, [Fl (α) : Fl ] divides [K (α) : K ][K : Fl ] [Indeed [Fl (α) : Fl ] divides [K 0 (α) : Fl (α)][Fl (α) : Fl ] = [K 0 (α) : Fl ] = [K 0 (α) : K 0 ][K 0 : Fl ]]. Hence p does not divide [Fl (α) : Fl ] and so α ∈ Kp and L = Kp (α) = Kp . erators of Claim. Kp Proof. has an extension of degree p. α ∈ F̄l such that [Fl (α) : Fl ] = p. Then [Kp (α) : Kp ] ≤ p: [Kp (α) : Kp ] < p. Then Kp (α) = Kp by above and α ∈ Kp . There exists Assume Now we can nally nish the proof: such that [L : Kp ] < p K̄p = F̄l and so if there exists a nite extension then there exists such an extension in F̄l L|Kp aswel. 3 Quantier elimination Denition. such that Let L be a language, ϕ ∈ W , V (ϕ) = ū = (u1 , . . . , un ). M |= ϕ iff for all ā = (a1 , . . . , an ) ∈ M , M |= ϕ(ā). Claim. Let Proof. Let ϕ ϕ be a formula such that M |= ∀xϕ By denition, be a formula, V (ϕ) = ū, V (ϕ) = (ū, x), M |= ϕ iff for all b ∈ M, for all iff Let M ϕ, L-structure M |= ∀xϕ. ā ∈ M n , M |= ϕ(ā, b) then we dene the closure of be an denoted ϕ̄ is iff M |= ϕ. ∀u1 ∀u2 · · · ∀un ϕ. Lemma 3.1. ϕ̄ is a sentence and M |= ϕ̄ iff M |= ϕ. Proof. By the above claim with induction on Remark. It is important to note that clearly not |V (ϕ)|. ¬ϕ 6= ¬(ϕ̄). For example for ϕ x = 0, we have ∀¬(x = 0) is ¬∀x(x = 0). Notation. Lemma 3.2. If ϕ ∈ W is a sentence, Γ ⊆ W and Γ ∪ {ϕ} is not satisable, then Γ |= ¬ϕ. Proof. but Let M |= Γ. M |= ¬ϕ as ϕ is If M |= ¬, then M |= Γ ∪ {ϕ}, a sentence. then Γ ∪ {ϕ} is satisable.Therefore M 6|=, Lemma 3.3. If ∆ is a theory, Γ ∪ W and Γ ∪ ∆ is not satisable, then there exist nitely many ϕ1 , . . . , ϕn ∈ ∆ such that Γ |= ¬(ϕ1 ∧ · · · ∧ ϕn ). Proof. Γ ∪ ∆ is not satisable, so by compactness, ϕ1 , . . . , ϕn ∈ ∆. Therefore Γ ∪ {ϕ1 ∧ · · · ∧ ϕn } Γ ∪ {ϕ1 , . . . , ϕn } is not satisable for some Γ |= ¬(ϕ1 ∧ · · · ∧ ϕn } by the is not satiable and so previous lemma. Notation. Let L be a language, ϕ ∈ W with V (ϕ) = x̄ = (x1 , . . . , xn ) and let c̄ = (c1 , . . . , cn ) be n-tuple of constant symbols not appearing in L. Let ϕ(c̄) denotethe formula we get from ϕ by 0 writing ci instead of xi for all i = 1, 2, . . . , n. Let L = L ∪ {c1 , . . . , cn }. an 26 Lemma 3.4. Let Γ be a set of L-formulas and assume that Γ |= ϕ(c̄). Then Γ |= ϕ. Proof. L-structure such that M |= Γ and let ā = (a1 , . . . , an ) ∈ M n . Let M0 be the L −structure we get from M by setting cM = ai for all i = 1, . . . , n. Then M0 |= Γ and so by i 0 induction M |= ϕ(c) and therefore M |= ϕ(ā) as our choice of ā was arbitrary. Hence M |= ϕ. Let M be an 0 Denition. A formula ϕ ∈ W is open if it has no quantiers. Let T be an L-theory. We say that T has quantier elimination if for every formula in ϕ ∈ W , there is an open ψ ∈ W such that V (ϕ) = V (ψ) and T |= ϕ ↔ ψ . For example, take a (n × n) matrix (aij ). The following says it is invertible: ∃b11 ∃b12 . . . ∃bnn ∀k n X aij bjk = δik i,j=1 where δij is the Kronecker delta. This is equivalent to Kramer's rule: det(aij ) 6= 0 iff X (−1)sign(π) a1,π(1) 6= 0. π∈Sn And this has no universal quantiers, so it's easy to check. The quantier elimination states that there always exists a universal Kramer's rule. Lemma 3.5. Let T be an L-theory and assume that for every open formula ϕ ∈ W with V (ϕ) = (u1 , u2 , . . . , un , x) there is an open formula ψ with V (ψ) = (u1 , . . . , un ) such that T |= ∃xϕ ↔ ψ . Then T has quantier elimination. Proof. We verify the condition for quantier elimination for all • ϕ • ϕ ϕ∈W by formula induction. ϕ is open and T |= ϕ ↔ ϕ. ¬ψ : then let θ ∈ W be open such that V (ψ) = V (θ) and T |= ψ ↔ θ. Then ¬θ is open and V (¬θ) = V (θ) = V (ψ) = V (¬ψ) and T |= ¬ψ ↔ ¬θ . • ϕ is ψ ∨ θ: • ϕ is ∃xψ : let ψ 0 ∈ W be open with V (ψ 0 ) = V (ψ) and T |= ψ ↔ ψ 0 . Then V (∃xψ 0 ) = V (∃xψ) = V (ψ 0 ) − {x} = V (ψ) − {x} and T |= ϕ ↔ ∃xψ . 0 The formula ∃xψ is not open, but by assumption there is an open θ ∈ W such that V (θ) = 0 V (ψ ) − {x} and T |= ∃xψ 0 ↔ θ. Then T |= ϕ ↔ θ and V (θ) = V (ϕ). is atomic: then is Denition. ϕ∈W Let T be an L-theory. T∀ denote the L-theory consisting of all sentences ϕ̄ where T |= ϕ. Then the theory T∀ is called the (set of ) universal Let is an open formula such that consequences of T . Lemma 3.6. If M ⊆ N are L-structures and N |= T , then M |= T∀ . Proof. If ϕ is open such that Denition. ϕ̄ ∈ T∀ then N |= ϕ. Hence M |= ϕ and so M |= ϕ̄. L-theory T has algebraically prime models if for every model M of T∀ , there is L-embedding i : M → N such that for every model N 0 of T and for every L-embedding g : M → N 0 there is an L-embedding k : N → N 0 such that j = k ◦ i. In this case we say that M is an algebraically prime extension of M. a model N An and an Denition. Let M ⊆ N be L-structure. We say that M is simply closed in N , in symbol M ≺S N if for every open formula ϕ ∈ W , V (ϕ) = (u1 , u2 , . . . , un , x) and ā = (a1 , a2 , . . . , an ) ∈ M n if N |= ∃xϕ(ā) then M |= ∃xϕ(ā). Theorem 3.7. Let L be a language such that C 6= ∅, let T be an L-theory such that (i) T has algebraically prime models, (ii) if M ⊆ N and M |= T , N |= T then M ≺S N . 27 open Then T has quantier elimination. Proof. ϕ ∈ W with V (ϕ) = (u1 , u2 , . . . , un , x), there ψ ∈ W with V (ψ) = V (ϕ) − X and T |= ∃xϕ ↔ ψ . 0 Let c̄ = (c1 , . . . , cn ) be an n-tuple of constant symbols not in L and L = L ∪ {c1 , . . . , cn }. Let ∆ 0 be a set of variable free formulas in L . It will be enough to show that for every open is an open Lemma 3.8. Let M, N be L0 -structures such that M |= T and N |= T . Assume that (i) M |= ∃xϕ(c̄), (ii) if ψ∆ and M |= ψ then N |= ψ . Then N |= ∃ϕ(c̄). Proof. The intersection of the underlying sets of all L0 -substructures for a given L0 structure is not empty, because it contains the interpretation of constant symbols. Hence it is the underlying set of the smallest respectively. Proof. L0 substructure: Claim. M 0 and N 0 Let M0 , N 0 denote the smallest L0 -substructure in M, N are isomoprhic. M 0 = {tM ∈ M | t is a term without variable symbols} and N 0 dened similarly. The 0 0 0 0 M N obvious map from M to N is i : M → N dened by i(t ) = t . The map i is well-dened: if t1 M = t2 M then M |= t1 = t2 . By assumption, N |= t1 = t2 and so N M N N M tN 1 N = t2 , so i(t1 ) = t1 = t2 = i(t2 ). M M M It is a homomorphism: if t0 = f (t1 , . . . , tM nf ) then M |= t0 = f (t1 , . . . , tnf ) and so N |= N N N M N N N N t0 = f (t1 , . . . , tnf ) and so t0 = f (t1 N, . . . , tN nf ) and so i(t0 ) = t0 = f (t1 , . . . , tnf ) = N M M f (i(t1 ), . . . , (tnf )). Let We also have N M M M M (tM iff M |= R(t1 . . . , tnR ) iff N |= R(t1 , . . . , tnR ) iff (i(t1 ), . . . , i(tn )) ∈ 1 , . . . , t nR ) ∈ R R R . i is obviously surjective (by the denitions of M 0 and N 0 ) and is also injective N N M and so t1 = t2 and so N |= t1 = t2 and so M |= t1 = t2 and so t1 = t2 M. Let M00 as M i(tM 1 ) = i(t2 ) M0 . We may assume that M ⊆ M00 ⊆ M (by the L0 -embedding k : M00 → N such that k|M 0 = i. 00 00 whose underlying set is k(M ). Then N ∼ = M00 and hence be algebraically prime extension of condition). By denition, there exists an N 00 ⊆ N be the L0 -substructure N |= T as M00 |= T . 00 By assumption M ≺S M and N 00 ≺S N , M |= ∃xϕ(c̄) implies M00 |= ∃xϕ(c̄). 00 ∼ 00 0 M = N as L -structures, N 00 |= ∃xϕ(c̄) and so by ≺S , N |= ∃xϕ(c̄). Let 00 Take ∆ = {δ ∈ ∆ | T ∪ {∃xϕ(c̄)} |= δ}. Clearly Because T ∪ {∃xϕ(c̄)} |= ∆0 . Lemma 3.9. T ∪ ∆0 |= ∃xϕ(c̄). Proof. 0 Let M |= T ∪ ∆ . We need to show that M |= ∃ϕ(c̄). T ∪ {∃xϕ(c̄)} ∪ ∆00 is satisable. Proof. Suppose not. δ = ¬δ1 ∨ · · · ∨ ¬δn . Let ∆00 = {δ ∈ ∆ | M |= δ}. Claim. T ∪ {∃xϕ(c̄)} |= ¬(δ1 ∧ · · · ∧ δn ) for δ1 , . . . , δn ∈ ∆00 . δ ∈ ∆ and T ∪ {∃xϕ(c̄)} |= δ and so δ ∈ ∆0 . Then Then N |= T ∪ {∃xϕ(c̄)} ∪ ∆00 . If δ ∈ ∆ and N |= δ (condition of the previous 00 Otherwise. So M |= ¬δ and so ¬δ ∈ ∆ and so N |= ¬δ . Therefore, by the previous lemma, M |= ∃xϕ(c̄). Let Therefore lemma) then M |= δ . δ1 , δ2 , . . . , δn ∈ ∆0 such that T ∪ {δ1 , . . . , δn } |= ∃ϕ(c̄). 0 Therefore if δ = δ1 ∧ · · · ∧ δn then T ∪ {δ} |= ∃xϕ(c̄). As δ ∈ ∆ , T ∪ {∃xϕ(c̄)} |= δ and so 0 T |= ∃xϕ(c̄) → δ and so T |= δ ↔ ∃xϕ(c̄). Let δ be the open L-formula we get from δ by writing ui instead of ci for i = 1, 2, . . . , n and hence T |= ∃xϕ ↔ δ 0 . By compactness, there are nitely many 28 Denition. Let L be a language and let ϕ ∈ W . We say that ϕ is a logical truth, in symbols |= ϕ, L-structure M, M |= ϕ. We say that two formulas ϕ, ψ ∈ W are logically equivalent |= ϕ ↔ ψ . Let W 0 be the set of open formulas. if for every if Denition. A formula The conjunction /disjunction of the formulas ϕ1 , . . . ϕn is ϕ1 ∧· · ·∧ϕn and ϕ1 ∨· · ·∨ϕn . is in conjunctive /disjunctive normal form if it is a conjunction/disjunction of ϕ ∈ W0 disjunction/conjunction of atomic formulas and negation of atomic formulas. For example (u ∈ x¬(u ∈ y)) ∨ (¬(u ∈ x) ∧ u ∈ y) ∨ x = y is a disjunctive normal form. Also (u ∈ x ∨ u ∈ y ∨ x = y) ∧ (¬(u ∈ x) ∨ ¬(u ∈ y) ∨ x = y) is a conjunctive normal form. Theorem 3.10. For every ϕ ∈ W 0 there is a ψ ∈ W 0 in conjunctive/disjunctive normal form such that V (ϕ) ⊇ V (ψ) and |= ϕ ↔ ψ . Proof. The proof is not trivial and we don't have enough time to go through it. Corollary 3.11. Let M ⊆ N be L-structures. Assume that for all ϕ ∈ W 0 with V (ϕ) = (u1 , . . . , un , x) which is the conjunction of atomic formulas and negation of atomic fomulas and M |= ∃xϕ if N |= ∃xϕ. Then M ≺S N . Proof. Let ψ ∈ W0 be an arbitrary. Then |= ψ ↔ ϕ where ϕ = ϕ1 ∨ · · · ∨ ϕn is the conjunction of atomic formulas and negation of atomic formulas. If N |= ∃xψ , then N |= ∃xϕ as |= ∃xψ ↔ ∃xϕ M |= ∃xϕ and so M |= ∃xψ . and so N |= ∃xϕi for some i. Hence M |= ∃xϕi and so Theorem 3.12. DAG has quantier elimination. Proof. Let L = {0, +} be the language of abelian groups. We say that commutative monoid (with the cancellation property) if the following: an L-structure M is a • 0 + a = a, • a + b = b + a, • (a + b) + c = a + (b + c), • (the cancellation property) a + c = b + c → a = b. These formulas are universal consequences of the theory of abelian groups. That is, if M is a commutative monoid with the cancellation property. 0 Let M be a commutative monoid. Let M := {x − y | x, y ∈ M }, 0 := 0 − 0. 0 0 0 0 (x − y) + (x − y ) := (x + x ) − (y + y ) makes M 0 an L-structure. Claim. M 0 is a commutative monoid. M |= DAG∀ , then Proof. Let ≡ be the following relation on M 0 : x − y ≡ x0 − y 0 iff there exists The operation x∈M such that x + y 0 + z = x0 + y + z x − y + z = x0 − y 0 + z x − y = x0 − y 0 . Denition. Let M be an L-structure. A binary relation equivalence relation if: (i) ≡ is an equivalnce relation, 29 ≡ on M is an admissible (L-admissible) f ∈ F and ā = (a1 , . . . , anf ) ∈ M nf and b̄ = (b1 , . . . , bnf ) ∈ M nf if ai ≡ bi for i = 1, . . . , nf then f M (ā) = f M (b̄), n n for all R ∈ R, ā = (a1 , . . . , anR ) ∈ M R , for all b̄ = (b1 , . . . , bnR ) ∈ M R if ai ≡ bi for all M M i = 1, . . . , nR , then ā ∈ R iff b̄ ∈ R . (ii) for every (iii) Denition. (i) (ii) (iii) Let M and ≡ be as above. Let M/ ≡ denote the L-structure dened as follows: M/ ≡ (the underlying set of M/ ≡) is the set of equivalence classes of ≡, cM/≡ is the equivalence class of cM for all c ∈ C , n 0 if f ∈ F and ā = (a1 , . . . , anf ) ∈ (M/ ≡) f and ai is represented as ai ∈ M , M 0 0 is the equivalence class of f (a1 , . . . , anf ), R ∈ R and b̄ = (b1 , . . . , bnR ) ∈ b01 , . . . , b0nR representatives of bi . (iv) if Claim. Proof. The L-structure M/ ≡ (M/ ≡)nR then b̄ ∈ RM/≡ iff then f M/≡ (ā) (b01 , . . . , b0nR ) ∈ RM for some is well-dened. Trivial. L-structure M/ ≡ is called the quotient L-structure (with respect to ≡). Claim. ≡ is an L-admissible equivalence relation and the quotient structure M 0 / ≡ is an abelian The group. Proof. Exercise. Denition. dened as Grothendieck group K(M ) of M is M 0 / ≡. There is a natural map i : M → K(M ) x 7→ [x − 0]≡ . Denition. ϕ(a) + ϕ(b) Claim. The ϕ : M 1 → M2 a, b ∈ M1 . A map for all between commutative monoids is a homomorphism if ϕ(a+b) = i : M → K(M ) is a homomorphism of monoids. Moreover, for every j : M → G where G is a commutative group, there exists a unique j 0 : K(M ) → G j 0 ◦ i = j . Finally, if M has the cancellation property, then i : M → K(M ) is injective. The above map homomorphism such that Proof. Exercise. 30
© Copyright 2026 Paperzz