• No results found

3 The Compactness Theorem (1st proof)

N/A
N/A
Protected

Academic year: 2021

Share "3 The Compactness Theorem (1st proof)"

Copied!
6
0
0

Loading.... (view fulltext now)

Full text

(1)

3

The Compactness Theorem (1st proof )

We review here the Henkin proof of the compactness theorem, as given in most logic classes; except that we abstract away all mention of syntactical proof. The interplay now is between finite satisfiability and existence of a model.

Fix a language L.

Let ⌃ be a set of L-sentences. We write A ✏ ⌃ (A models ⌃, or A is a model of ⌃) if, for any 2⌃, A✏ .

AnL-sentence is said to bea logical consequence of ⌃ifA ✏⌃implies A ✏ for every L-structure A.

Notationally, it will be convenient to write ⌃✏ when is a logical conse-quence of some finite subset of ⌃. We will see later that in fact ⌃✏ i↵

is a logical consequence of ⌃.

written⌃✏ ,ifA ✏⌃impliesA✏ for everyL-structureA.For⌃infinite,

⌃✏ means that there is a finite ⌃0 such that 0 ✏ . 4

is calledlogically valid, written ✏ , if A ✏ for every L-structure A. A set ⌃ of L-sentences is said to be satisfiable if it has a model, i.e. there is an L-structure A such that A ✏ ⌃. ⌃ is said to be finitely satisfiable (f.s.) if any finite subset of ⌃ is satisfiable.

⌃ is said to be complete if, for any L-sentence , 2⌃ or¬ 2⌃.

Exercise 3.1. Let ↵,↵1, . . . ,↵n, , 1, . . . , n, be closed L-terms, P, f

L-symbols forn-ary predicate andn-ary function, correspondingly, and (v0, v1, . . . , vn) an L-formula with free variablesv0, v1, . . . , vn.Prove that

1. ↵l ✏ l↵;

2. ↵l , l ✏↵l ;

3. ✏↵l↵;

4. ↵1 l 1, . . . ,↵nl n, P(↵1, . . . ,↵n)|=P( 1, . . . , n);

5. ↵l ,↵1 l 1, . . . ,↵nl n, f(↵1, . . . ,↵n)l↵|= f( 1, . . . , n)l ;

4This definition will be convenient for the proof, and will be seen later to be equivalent

(2)

6. ( ,↵1, . . . ,↵n)|= 9v0 (v0,↵1, . . . ,↵n).

Definition 3.1. (1)-(5) are the axioms of equality. A binary relation satis-fying these laws is called a congruence.

A set of L-sentences⌃ is said to be deductively closedif

⌃✏ implies 2⌃.

Exercise 3.2. (i) If ⌃0 and 0 then ;

(ii) A complete f.s. ⌃ is deductively closed.

Proposition 1 (Lindenbaum’s Theorem). For any f.s. set ofL-sentences⌃

there is a complete f.s. set of L-sentences⌃# such that #.

Proof Let

S ={⌃0 :⌃⌃0 a f.s. set of L-sentences }.

Clearly S satisfies the hypothesis of Zorn’s Lemma, so it contains a maximal element ⌃# say. This is complete for otherwise, say 2/ # and ¬ 2/ #. By maximality neither { }[⌃# nor {¬ }[# is f.s.. Hence there exist finite S1 ⌃# and S2 # such that neither { }[S1 nor {¬ }[S2 is satisfiable. However, S1[S2 ✓⌃#,finite, so has a model, A say. But either A ✏ , soA{ }[S1, orA ✏¬ , soA ✏{¬ }[S2, a contradiction.

Remark In class we saw another proof of Lindenbaum’s theorem, if the language is countable or more generally well-ordered. Namely, let { n} enu-merate all sentences of L. Define n recursively: if ⌃S{ 1, . . . , n 1, n} is consistent, let n = n; otherwise let n = ¬ n. One verifies easily that

⌃S{ n:n = 1,2, . . .} is complete and f.s.

Indeed the Henkin proof of compactness (or completeness) does not require the axiom of choice, provided the symbols of the language itself are well-ordered. This is an advantage over the ultraproduct proof in the next section.

Exercise 3.3. Let us allow 0-place relation symbols (Ri : i 2 I); they are also called propositional symbols. Assume there are no other symbols; so that a structure consists just of assigning a truth value to each Ri. Thus a structure is just an element of the I-fold product {0,1}I.

1. In this case, show that a complete f.s. set of sentences ⌃ determines a model M(⌃) of ⌃, simply by assigning 1 to Ri if Ri 2 ⌃ and 0 otherwise.

(3)

2. Define a topology on {0,1}I by letting a basic open set have the form B(i0, . . . , ik;⌫0, . . . ,⌫k) ={f :f(i0) = ⌫0, . . . , f(ik) =⌫k}

where i0, . . . , ik 2 I and ⌫0, . . . ,⌫k 2 {0,1}. Tychono↵’s theorem as-serts that this topology is compact. Prove this using this case of the compactness theorem.

A set ⌃ of L-sentences is said to be witnessing if for any sentence in ⌃ of the form 9v'(v) there is a closedL-term such that '( )2⌃.

Exercise 3.4. Any complete f.s. witnessing set of L-sentences has a closed L-term.

Hint: Consider the L-sentence9v v lv.

An L-structure A is called minimal if it has no proper substructure.

Exercise 3.5. Show that A is a minimal L-structure i↵ for every a 2 A there is a closed L-term such that A =a.

Let us say that a model of ⌃ iscanonical if it is minimal as an L-structure. 5

Proposition 2. For any complete, witnessing, f.s. set⌃ofL-sentences there is a canonical model A of ⌃.

Proof Let⇤ be the set of closed terms of L. This is nonempty by 3.4. For ↵, 2⇤ define ↵v i↵ ↵l 2⌃.

This is an equivalence relation by 3.1.1- 3.1.3. and 3.2.

For ↵2⇤, let ˜↵ denote the v-equivalence class containing ↵. Let A={↵˜: ↵2⇤}.

This will be the domain of our modelA.We want to define relations, functions and constants of L onA.

Let P be an n-ary relation symbol of L and ↵1, . . . ,↵n2⇤. Define h↵˜1, . . . ,↵˜ni 2PA i↵ P(↵1, . . . ,↵n)2⌃.

By 3.1.4 the definition does not depend on the choice of representatives in the v-classes.

(4)

For a unary function symbol f of L of arity m and ↵1, . . . ,↵m 2⇤ define fA(˜↵1, . . . ,↵m) = ˜˜ ⌧, where ⌧ =f(↵1, . . . ,↵m).

By 3.1.5 this is well-defined.

Finally, for a constant symbol, cA is just ˜c.

We now prove by induction on complexity of anL-formula '(v1, . . . , vn) that (⇤) A✏'(˜↵1, . . . ,↵n) i˜ ↵ '(↵1, . . . ,↵n)2⌃.

For atomic formulas we have this by definition. If '= ('1^'2) then

A ✏('1(˜↵1, . . . ,↵n)˜ ^'2(˜↵1, . . . ,↵n)) i˜ ↵A✏'1(˜↵1, . . . ,↵n) and˜ A✏'2(˜↵1, . . . ,↵n)˜ i↵ (by induction hypothesis) '1(↵1, . . . ,↵n),'2(↵1, . . . ,↵n) 2 ⌃ i↵ (by 3.2) ('1(↵1, . . . ,↵n)^'2(↵1, . . . ,↵n))2⌃. Which proves (*) in this case.

The case '=¬ is proved similarly. In case ' =9v

A ✏ 9v (v,↵1, . . . ,˜ ↵n) i˜ ↵ there is 2 ⇤ such that A ✏ ( ˜,↵1, . . . ,˜ ↵n) i˜ ↵

there is 2 ⇤ such that ( ,↵1, . . . ,↵n) 2 ⌃. The latter implies, by 3.1.6 and 3.2, that 9v (v,↵1, . . . ,↵n) 2 ⌃, and the converse holds because ⌃ is witnessing. This proves (*) for the formula and finishes the proof of (*) for all formulas.

Finally notice that (*) implies that A✏⌃.

We sometimes need to expand or reduce our language.

Let L be a language with non-logical symbols {Pi}i2I [{fj}j2J [{ck}k2K and L0 L with non-logical symbols {Pi}i

2I0 [{fj}j2J0 [{ck}k2K0 (I0 ✓ I, J0 J, K0 K). Let

A=hA;{PiA}i2I;{fjA}j2J;{cAk}k2Ki and

A0 =hA;{PiA}i2I0;{fjA}j2J0;{cAk}k2K0i.

Under these conditions we callA0 the L0-reduct of Aand, correspondingly,

A is an L-expansion of A0.

RemarkObviously, under the notations above for anL0-formula'(v1, . . . , vn) and a1, . . . , an 2A

(5)

Exercise 3.6. Let, for each i2N, ⌃i denote a set ofL sentences. Suppose

⌃0 ✓⌃1 ✓. . .⌃i. . . and each ⌃i is finitely satisfiable.

Then the union of the chain, Si2N⌃i, is finitely satisfiable.

Theorem 1(Compactness Theorem).Any finitely satisfiable set ofL-sentences

⌃ is satisfiable. Moreover, ⌃ has a model of cardinality less or equal to |L|.

Proof We introduce new languages Li and complete set of Li-sentences ⌃i (i = 0,1, . . .). Let L0 =L. By Lindenbaum’s Theorem there exists ⌃0 ◆ ⌃, a complete set of L0-sentences.

Given f.s. ⌃i in languageLi, introduce the new language Li+1 =Li[{c : a one variable Li-formula} and the new set of Li+1 sentences

⌃⇤i =⌃i[{(9v (v)! (c )) : a one variable Li-formula}. Claim. ⌃⇤

i is f.s. Indeed, for any finite S ✓ ⌃⇤i let S1 = S\⌃i and take a model A of S1 with domain A, which we assume well-ordered. Assign constants to symbols c as follows:

c = ⇢

the first element in (A) if (A)6=; the first element in A if (A) = ; . Denote the expanded structure A⇤. By the definition, for all (v),

A⇤ 9v (v)! (c ).So AS. This proves the claim.

Let ⌃i+1 be a complete f.s. set of Li+1-sentences containing ⌃⇤i.

Take ⌃⇤ = Si2N⌃i. This is finitely satisfiable by 3.6. By construction one sees immediately that ⌃⇤ is also witnessing and complete set of sentences

in the language SLi = L +{new constants}. Proposition 2 gives us the canonical model, A⇤,of .The reduct of Ato language L is a model of .

The cardinality of the model we constructed is less or equal to |L| (see also Exercise 1.1).

As noted in the logic class, the contrapositive of the compactness theorem is the statement that logical consequence is intrinsically finitary:

(6)

Exercise 3.7. Show that a sentence is a logical consequence of some finite subset of a set ⌃ of sentences, if and only if it is a logical consequence of⌃. The following exercises draw some corollaries of compactness.

Exercise 3.8. Let T = T h(N,+,·,0,1). Let L0 ={+,·, c} be the language

obtained by adjoining a new constant symbol, T0 = T S{c 6= 0, c 6= 1, c 6= 1 + 1,· · ·}. Show that T0 has a model A0, and let A be the L-reduct of A0.

Show thatA is a model ofT, and is not minimalL-structure. Conclude that A, B are not isomorphic. This proves Skolem’s theorem, that the natural numbers are not characterized by their first-order theory.

Exercise 3.9. Show that there exists a model A of T h(R,+,·) containing an infinitesimal element, i.e. an element a such that for anyn 2N,A|= 0< na < 1.

Exercise 3.10. Assume the language has at least one constant symbol. Let

⌃ be a set of quantifier-free sentences. Assume ⌃ is satisfiable and that for any quantifier-free sentence , either 2 ⌃ or ¬ 2 ⌃. Show that there exists a unique minimalL-structure, up to isomorphism, which is a model of

⌃.

Exercise 3.11. Assume, for each n 2 N, that T has a model with at least n elements. Let be any set. Show thatT has a model Awhose universe A satisfies |A| | |. (Hint: introduce new constant symbols ci for i2 , and sentences ci 6=cj; use compactness.)

This was proved by Tarski, and called by him theupwardL¨owenheim-Skolem theorem.

References

Related documents

We give a combinatorial proof of the rank-unimodality of the poset of order ideals of a product of chains of lengths 2, n, and m, and find a symmetric chain decomposition in the

Unless stated otherwise, we shall use small letters such as x to denote positive inte- gers or functions or elements of a set, capital letters such as X to denote sets, and

Sauer and Spencer in their work [5] conjectured that if the minimum degree of G is at least 2n/3 then G contains every graph of order n with maximum degree of at most 2..

Consider the following question: given n coins with distinct integer weights between 1 gram and n grams, each labeled by a distinct integer label between 1 and n, what is the

By restricting Fan’s condition to some induced subgraphs of 2- connected graphs, Ning and Zhang [9] extended Theorem 1 as follows.... This can be easily deduced from the

(4) Extension theorems are often used for optimal linear codes problem, especially to prove the nonexistence of linear codes with certain parameters.. Moreover, the extended matrix of

Corr´ adi and Hajnal [2] proved that if G is a graph of order at least 3k with minimum degree at least 2k, then G contains k independent cycles.. This result has inspired much

Specifically, we prove that for any graph H (with some notable exceptions), and for any graph G, G = H, there is an edge-coloring of G with k colors which contains no induced