Chapter 13 The First-Order Theory Of Linear Orderings
CHAPTER 13 THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
$1. THE EHRENFEUCHT-FRAISSE GAMES AND FIRST-ORDER THEORIES In Chapters 6 and 7, where we introd...
CHAPTER 13 THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
$1. THE EHRENFEUCHT-FRAISSE GAMES AND FIRST-ORDER THEORIES In Chapters 6 and 7, where we introduced and analyzed Ehrenfeucht games, we made many references to the implications of that analysis for the first-order theory of linear orderings. We will begin this section, and this chapter, by making the connection between games and languages; that is, we will prove that A B if and only if A = B. The implications of this connection will be discussed in 913.2. Although the statement of the theorem above mentions only statements of L , its proof will proceed by induction on formulas. Thus we will need to define a version of the Ehrenfeucht game which permits parameters. As we observed in Chapter 6, the games can be played with any two structures of the same type; later, in 513.6, we will play games with k-partitioned orderings, which are just linear orderings with k unary relations. The formulation of Ehrenfeucht games we now present is sufficiently general to embrace all of these extensions of the original definitions. To simplify the notation in the coming definitions, it is convenient to separate notationally the distinguished elements of a structure from the structure itself. Thus while we have let A denote the structure ( A ; ( R i l i €I } ; ( a j l , j E J ) ) , w e w i l l n o w l e t A d e n o t e ( A ; R , , . . .,R,)anduse(A,a,, ...,a k ) or (A, a) for what was formerly denoted A. Another comment on notation : Whenever we speak of structures in general, we will use the notation A ; however, when we specialize the discussion to linear orderings, we will forget the boldface type and write A as in earlier chapters.
-
DEFINITION 13.1 : Let A and B be structures of the same type 2. That is, A is ( A , R , , . . . ,R P ) and B is ( B , S , , . . . ,S,), where Ri is a z(i)-ary relation on A and Siis a z(i)-ary relation on B for each i. Let a = (a,, a 2 ,. . . ,ak) and b = ( b , ,b 2 , . . . ,b k ) be sequences of elements of A and B, respectively. Let n be a fixed natural number. A play of the game G,( (A, a), (B, b)) consists 247
13.
248
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
of an ordered sequence of n repetitions of the following: PLAYER I chooses an element of either A or B and PLAYER II chooses an element of the other. The element of A selected at the rth turn is denoted a k + tand the element of B selected at the tth turn is denoted b,,,. We say that PLAYER 11 has won the play of the game if for each i, 1 I i I p , and each sequence r( l), r(2),. . . , r ( s ( i ) )of numbers between 1 and k + n,
*
* . ,ar(r(i))) E R i
if and only if
. ,br(T(i)))~Si;
otherwise we say that PLAYER I has won the play of the game. We say that PLAYER 11 has a winning strategy in the game G,( (A, a), (B,b)) if there are functions f ; ,#,, . . . ,f, such that
(i) the domain of .f, is the set of all ordered t-tuples of elements of A u B; (ii) given c 1 , c 2 , .. . ,c, E A u B (representing the choices of PLAYER I at the first t turns) f,(cl,c2,. . . , C J E A
if C , E B
f,(cl,c2, . . . , c , ) t B
if c , E A ;
and
(iii) i f c , , ~ , ,. . . , ~ , E Abandi if we define
and
for every t I n, then for every i, 1 I i Ip , and for every sequence r ( l ) , r(2),. . . , r ( z ( i ) )of numbers between 1 and k + n
..
< ~ r ( ~ ) - a r ( 2 j* ~, a r ( r ( i j ) ) E
Otherwise we say that
R,
if and only if
PLAYER I
...
9
br(,(i))) E S i .
has a winning strategy.
We observe that Definition 13.1 coincides with Definition 6.2 in the case where A and B are linear orderings and k = 0, with a slight exception. The exception is simply that the game G,(A,B) is now defined for any linear orderings A and B,and PLAYER II always has a winning strategy in G,(A,B). [Note, however, that PLAYER 11 does not have a winning strategy in G,((A,a,,a,),(B,b,,b,)) unlessa, < A a, ifandonlyifb, < B b 2 . ]
1.
THE
EHRENFEUCHT-FRAYS&
GAMES AND FIRST-ORDER THEORIES
249
NOTATION : We write G,((A, a), (B, b)) E 11 if PLAYER II has a winning strategy in that game; otherwise we write G,( (A, a), (B, b)) E I.
The following lemmas are easily verified. LEMMA 13.2: (1) C,( (A, a), (B.b)) E I1 ifandonly ifC,( (B, b), (A, a)) E 11. (2) I f f is an isomorphism ,fronz A onto B and if ,f(aj)= bj f o r all j , 1 5 j 5 k, then G,((A, a), (B,b)) E 11. A LEMMA 13.3:
I f G,((A,a), (B,b))E I1 and 0 5 m < n, then G,((A, a), (B, b)) E 11.
THEOREM 13.4: G , , and
(i) for every a
E
A
(A, a), (B, b)) E I1 if and only
if
A there is a b E B such that G,((A, a, a), (B,b, b ) )E 11,
(ii) for every h E B there is an a E A such that G,( (A, a, a),(B, b, b ) )E 11.
A
Note that the form of the conditions in Theorem 13.4 is different from that in Theorem 6.6. However, it is easily seen from Definition 13.1 above that G,((A,a), (B,b ) )E I1 if and only if both G,(A'", B t b ) E I1 and G,(A'",B'b) E 11, so that the conditions here and in Theorem 6.6 agree for k = 0 when A and B are linear orderings. (When A and B are not linear orderings, the conditions of Theorem 6.6 do not make sense at all.) LEMMA 13.5 : Suppose that G,( (A, a), (B, b)) E I1 and G,( (B, b), (C,c)) E 11. Then G,( (A, a), (C,c)) E 11. A
The proofs of 13.2, 13.3, 13.4, and 13.5 are similar to those of 6.3, 6.4, 6.6, and 6.7, respectively, and are left to the reader. The upshot of the three lemmas is that we can define an equivalence relation, for each k 2 0 and each n 2 0 and each type z, on structures of the form (A, a), where A is a structure of type z and a is an ordered k-tuple of elements of A, by specifying that (A*a ) - k , n (B, b, if C,((A,a), (B, b))EII. In the case where A and B are linear orderings and k = 0, we have that A B if and only if A -,B according to Definition 6.8. (We will often suppress the k in - k , n and write -,.) As in Chapter 6,
250
13.
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
we say that two structures are C-equivalent or --equivalent if they are -,-equivalent for all n. EXERCISE 13.6: Let A and B be linear orderings. Let a , < a 2 < . . . < ak and b, < b, < . . < bk be elements of A and B, respectively. Define A , = A‘“’, Ak = A’ ak and A j = [ u j , u j + for each j , 1 5 j < k, and similarly define B , = BSb1,B , = B’bk and B j = [bj, bj+l]B for each j, 1 I j < k. Show that C,((A,a), ( B ,b ) ) E I I if and only if G,(Aj,Bj)~IIfor every j 5 k. *
-
The proof that A B if and only if A = B will proceed by showing that A - n B if and only if A -,, B. Intuitively, A =,, B if any statement of complexity ~n that is true in A is also true in B. The appropriate measure of complexity is not the number of quantifiers in a formula, but the depth of the quantifiers. For example, the statements (W((3Y)P(X, Y ) A (VZ)P(X, z)),
(3X)(3Y)(P(X, Y ) A ( V 4 ( 1 (P(x, 4 * P(Z, Y )1)) have the same number (and kind) of quantifiers. The first has quantifier depth 2; the second has quantifier depth 3. In the first, neither of ( 3 y ) and (Vz) is within the scope of the other; in the second, (Vz) is in the scope of (3y), which is, in turn, in the scope of (3x). (The “scope” of a quantifier is a term whose precise definition can be found references [31 and [7] of Chapter 12.) DEFINITION 13.7: We define the quantifier depth qd(4) of a formula 4 by induction on formulas d as follows:
DEFINITION 13.8: Let Lk,,be the set of formulas ofL whose free variables are among u l , u 2 , . . . ,uk and whose quantifier depth is at most n. Let A and B be interpretations of L and let u t , a 2 , . . . , a k €A and b t , b 2 ,. . . ,bkEB. We define (A,a) E k , , (B,b) if for every &u,, u 2 , . . . , u k ) E Lk.,,we have
A k 4[a1,a2,.. . ,ak]
if and only if
If k = 0, we write A -,, B insteud qf A q n B.
B k 4[bl, b2, . . . ,bk].
1.
THE EHRENFEUCHT-FRAi‘SSk GAMES AND FIRST-ORDER THEORIES
251
B if and only if A - n B EXERCISE 13.9: Assuming that for every n, A (Corollary 13.12),show that the statement (3x)(3y)(P(x, y) A ( V z ) ( i (P(x,z ) A P(z, y ) ) ) )considered above is not logically equivalent to any statement of quantifier depth 2. LEMMA 13.10: For each k and ii there is a finite subset @k,n of Lk,nsuch that every forniula of Lk,n is logically equivalent to a ,formula of @ k , n . Proof: It is easily verified by induction that every formula of Lk,n is a Boolean combination of formulas of Lk,n-l and of formulas of the form (Qy)+(u,,u,,. . . ,uk,y),where Q can be either 3 or V. By replacement of bound variables, all of the latter formulas are logically equivalent to formulas of the form (Quk+l)+(ul, u 2 , . . . ,tikruk+ 1), where +(ul, u 2 , . . . ,u k + 1) is in L k + 1 , n - 1.
We now proceed by induction on n to show that the desired conclusion holds for all k. For n = 0, the formulas of Lk,nare all Boolean combinations of atomic formulas in the variables u l , u,, . . . ,u k . Since the number of predicate letters is finite, there are finiteIy many such atomic formulas. Hence, by Exercise 12.5, we can define for each k an appropriate finite subset Qk.0 of Lk,o.For n > 0, we may assume that ak,,and @)k+ 1, n- have been defined and are finite; then every formula of L,’nis logically equivalent to a Boolean combination of formulas from ( D k , n - l and of formulas (Quk+l)+(ul,...,’k+~), where ~ ( U I , . . ? U k + l ) is in @ k + l , n - l . Hence, by Exercise 12.5, we can find an appropriate finite subset @)k,n of Lk,n. We are now ready to state and prove the main theorem, which relates the Ehrenfeucht games to first-order languages. Note that the proof of the lemma above depends on the assumption that the language has only finitely many predicate letters; the theorem below, which uses the lemma above, is false without this assumption, as we will show near the end of 513.4. THEOREM 13.11 : Let n 2 0 and k 2 0 be given, let A and B be structures of the same type and let a , , a,, . . . , a k €A and b , , b,, . . . ,b k E B. Then (A, a) = k , n (€3, b)
if and only if
(A, a)
-
k.n
(B, b).
We proceed by induction on n to show that the equivalence holds for all k. If n = 0, then assuming that A = ( A ,R , , . . . ,R,) and B = ( B , S,, . . . ,S,), it is clear that Go((A, a), (B,b))E 11 if and only if for each i and for each sequence r(l),r(2),. . . , r ( z ( i ) )of numbers between 1 and k
Proof:
(ar(~j,ar,,,, . . . ,ar(r(i)))E Ri
if and only if
(br(lj,br(2),. . * ,br(r,ijj> E Si
252
13.
THE FIRST-ORDER THEORY OF LINEAR ORDERINCS
if and only if for every formula 4(t',,u , , . . . ,U , ) E @k,O A b 4[a,,a,, . . . vak]
if and only if
A I= 4 [ U l i Q ~ ., . . -ak]
if and only if
B != +[b,, b,, . . . ,b,]. Assume that the claim has been proven for n and suppose that (A, a) -,,+ (B, b). Let 4 be in @k,,,+ ,. Then either 4 is of the form (Qu,, where or 4 is a Boolean combination of formulas in @k,,, and formulas of this form. It clearly suffices to show that B =!
4[bl, b,,
. . . ,bk]
holds for all formulas 4 in @k,n+ of the form ( Q U k + J$ since the "if and only if" is preserved under Boolean combinations of formulas. And since (Vy)$ is logically equivalent to 1 ( 3 y ) i $ , we may assume that 4 is of the form ( 3 4 , ,)$, where $ has quantifier depth at most n. Since (A,a) -,,+ (B,b), for each ~ E there A is a b E B such that (A, a, u ) -,, (B, b, b). Hence by the induction hypothesis, (A, a, a) -,, (B, b, b ) ; thus for each a E A there is a b E B such that A b $ [ a , , . . . ,a,,u]
if and only if
B 1 $ [ b l , . . , ,b,, b].
Hence if A I=4[a,,a,, . . . ,a,], then €3 k 4 [ b , , b,, . . . ,bk]. Similarly, if B k 4 [ b , ,b,, . . . ,b,], then A k 4[al, a,, . . . ,uk]. Hence, for each 4 E @ k , n + 1 , A 1 d [ a ~ , U , ,. . . >a,]
if and only if
,
B k 4 [ b l ,b,, . . . ,b k ] .
Conversely, suppose that (A,a) Q,,+ (B, b). We must show that there is a formula 4 ~ @ , , , ,such + that it is not the case that
,
A 1 4 [ a l , u 2 , . . . ,ak]
if and only if
B 1 4[b1, . . . ,b,].
Since (A, a) *,,+, (B, b), there is (without loss of generality) an element U E A such that for each b E B, (A, a, a) * ,,(B,b, b). By the induction hypothesis, there is, for each b E B, a formula qbE @)k+ ," such that A != $b[a13a2,. . . ,ak,a]
B y $b[bi,b,,. . . ,b,,b]. Since @ k + , , , , is finite, it follows that {$blbEB) is finite. Let $ be the conjunctionA{$,(bEB}.ThenA k $ [ a l , U 2 ,..., a,,a] butB y $[b1,h2,..., b,,h] for any b E B. Let 4 be ( h k + l ) $ . Then A 1 4 [a ,,a ,, . . . ,a,] and B y 4 [h ,b, , . . . ,bk], although 4 E a,.,, + This proves the converse and thus the theorem. and
,
-
COROLLARY 13.12: (1) For each n, A (2) A = B ifand only if A B. W
E,, B
if and only
if
A
-,, B.
Thus two structures are elementarily equivalent if and only if they are G-equivalent.
2.
253
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
82. THE EHRENFEUCHT GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
As a consequence of the equivalence between logical equivalence and G-equivalence, we can now interpret logically various results of Chapters 6 and 7. For example, from Corollary 6.12 it follows that o is logically equivalent to o 5 . CI for any order type CI. In particular, no statement of the first-order language distinguishes between w and w 5.
+
+
THEOREM 13.13: (Tarski [20]) an elementary class.
The class W of well-orderings is not
Proof: W is not closed under elementary equivalence, although by Exercise 12.7.2, every elementary class is closed under elementary equivalence. I
By Exercise 6.11, if A is a finite linear ordering with at least 2" - 1 elements, then A -"o o*.Hence any statement true in o o* is also true in some finite linear ordering. Moreover, given any finite set of statements true in o ID*, their conjunction, which is also true in w o*, must be true in some finite linear ordering. O n the other hand, o o* is not elementarily equivalent to any finite A since if A has m elements, a statement asserting the existence of at least m 1 distinct elements would be true in o + o*but not in A . Thus although o + w* can be distinguished from each finite A by a first-order statement, no one first-order statement distinguishes o + o*from them all. This is expressed by saying that o o*, or more properly Th(o w*), is not finitely axiomatizable.
+
+
+
+ +
+
+
+
DEFINITION 13.14: Given a theory T we say that T is finitely axiomatizable if there is a statement 4 E T such that for every 9 E T , 4 logically implies 9. The statement 4 is called an axiomatization of T. Given a structure B, we say that B is finitely axiomatizable if Th(B) is finitely axiomatizable, and if 4 is an axiomatization of Th(B), we also say that is an axiomatization of B.
Note that
4 is an axiomatization of B if and only if for every structure 4, then C = B.
C of the same type as B, if C !=
PROPOSITION 13.15: o
+ o* is not Jinitely axiomarizable.
I
13.
254
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
Note that 4 is an axiomatization of a theory T if and only if for every interpretation C of the language of T , C k T if and only if C b 4. Thus, for example, the conjunction 4Lof (1) ( ' J X ) ( V Y ) ( W X < y A Y < 2 --+ x < 4, (2) ( W V Y ) ( X < Y 7 ( Y < x )1, and (3) ( V X ) ( \ J Y ) ( X < Y v x = Y" Y < x ) +
is an axiomatization of the theory of linear orderings. Although w + o*is not finitely axiomatizable, w is finitely axiomatizable. For every model of the conjunction of the statements (1)
4L.t
(2) ( W 3 Y ) ( X < Y A ( V Z ) ( l ( X < z A z < Y ) ) ) , (3) (3X)(VY)(X< Y v x = Y ) , (4) (VX)((3Y)(Y< x ) --+ (3Y)(Y < x A ( W l ( Y < z A
-=x ) ) ) )
has order type w + 6 . c1 for some c1 and so is elementarily equivalent to o by Corollary 6.12 and Corollary 13.12. PROPOSITION 13.16: w is Jinitely axiomatizable.
We note that the quantifier depth of the above axiomatization of w is 3; hence, if B =30then B = w. This fact, and the fact that w is finitely axiomatizable with an axiomatization of quantifier depth 3, can also be explained combinatorially, so we make a brief digression to Ehrenfeucht games.
-
DEFINITION 13.17: A structure A is said to be hitary if there is an n such that B A implies B A for all structures B of the same type. If A is finitary, then we will write f(A) = m if m is the least IZ for which B -" A implies B A for all structures B of the same type.
-
Before commenting on this definition, we note that using Corollary 13.12,Theorem 6.16 says that given any linear ordering C there is a countable linear ordering elementarily equivalent to it. This conclusion, the Lowenheim-Skolem Theorem, is also true for arbitrary structures, as pointed out at the end of 812.2. By the Lowenheim-Skolem Theorem, to show that A is finitary it suffices to show that B -,, A implies B A for all countable structures B of the same type; for if C is uncountable, then C B for some countable B, so that C -,, A implies that B w nA, which implies that B A and therefore C A.
-
- -
-
2.
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
255
EXAMPLES: (1) q is finitary, for if B m 3 q, then B must be dense and have no endpoints; hence if B is countable, it follows, by Cantor’s Theorem 2.5, that B v q. Then Lemma 6.3.2 implies that B mnq for all n, or, equivalently, that B q. O n the other hand, since 5 - 2 q , we conclude that f ( q ) > 2, so that f ( q ) = 3. (2) w is finitary, for if B m 3 w , then B must have a first element, every element of B must have an immediate successor, and every element of B, except the first, must have an immediate predecessor. Thus, if b E B is not among the first w elements of B, then cF(b)‘v 5, so that B ‘v w + 5 . a for some order type a. Since B w by Corollary 6.12, we conclude that w is finitary. On the other hand, ((D + o)-2 w so that f ( w )= 3. (3) o + w* is not finitary, for, given any n, 2” -,, (w + a*)by Exercise 6.11, but 2“ * n + (w + a*)by Exercise 6.10.2. (4) Each finite ordering is finitary, for, by Exercise 6.10.2, if k < 2” - 1 and A -” k, then A = k so that A k ; hence f(k) I n.
-
-
-
EXERCISE 13.18: (1) Show that 5 is finitary and that f ( 5 ) = 3. (2) Show that w + w is finitary and that f(o+ a)= 5. (3) Show that w* ( + w is finitary and that f ( o * + 5 + w ) = 5. [Note: You will have to state and prove analogues of Corollary 6.12 for 5, w + o,and w* + 5 + w.]
+
This completes the digression to Ehrenfeucht games. We showed earlier that if B = 3 w then B = w by using a specific axiomatization of w . Alterw, so B o by Example natively, if B - 3 w, then, by Corollary 13.12, B 2 above; hence, by Corollary 13.12, B = w. Furthermore, since @ 0 , 3 is finite, &4 E @0,310 1 4} is an axiomatization of w. This is an example of the following fact. m3
-
PROPOSITION 13.19 : A structure A is finitely axiomatizable if and only cf it is finitary. Furthermore, if , f ( A )= n, then A has an axiomatization of quantijier depth n but has no axiomatization of quantijier depth smaller than n. Proof : The first statement follows from Corollary 13.12. If f(A) = n, then as in the discussion above, A has an axiomatization of quantifier depth n. If A had an axiomatization of quantifier depth m < n, it would follow that f(A) I m, contrary to assumption. H
Applying Proposition 13.19 to conclude that w is finitely axiomatizable is more efficient, but does not of course give an explicit axiomatization of
256
13.
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
w. It does, however, guarantee that no axiomatization of o has smaller quantifier depth than @<,, .
COROLLARY 13.20:
q and
5 are Jinitely axiomatizable. A
c, +
+
EXERCISE 13.21 : ( I ) Find axiomatizations of q, o o,and o* 5 o that have minimal quantifier depth. (2) Describe an axiomatization for each n that has minimal quantifier depth.
+
Corollaries 7.10 and 7.21 have the following logical interpretation. THEOREM 13.22: (Lauchli and Leonard) (1) I f # is a sfatement true in some scattered linear ordering A , then M k 4 .for some M E A. . (2) If 4 is a statement true in some linear ordering A , then M I. # for some M E A.
Proof: Given # of quantifier depth n, use Corollary 7.10 (or 7.21) to find an M E A. (or M E . X ) such that M m n A . Then, by Corollary 13.12, M = " A , so that M k 4.
Note that the proof of Theorem 13.22 actually shows more, namely, given any (scattered) linear ordering A and any n, there is a single (scattered) in which is true every statement of quantifier linear ordering M in A (or Ao) depth at most n which is true in A . In 513.6, we will prove the following stronger version of Theorem 13.22.
THEOREM 13.23: (Amit, Myers, Schmerl, Shelah) Let A be a linear ordering. Then for each n there is a finitely axiomatizable M E Af such that A =,, M . If' A is scattered, then M can be chosen f r o m Ao. Theorem 13.22 can be used to show that various linear orderings are not finitely axiomatizable. For example, assume that cow is finitely axiomatizable. It follows from Theorem 13.22 that ow= M for some M E do since the axiomatization of coo must be true in some M E Ao.Choose M of smallest rank such that cow = M . Let M' be the subordering of M consisting of all limit points of M , that is, all elements of M that have no immediate predecessor. Now it is easily verified by induction on A. that if M E -H0 and M' is the set of limit points of M then M' E A. and rF(M') < rF(M). Hence coo f M'. Thus PLAYER I has a winning strategy in Gn(mw, M') for
2.
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
257
some n. But thenPLAYER I has the following winning strategy in G , + 2 ( ~MW) :, In his first n moves he plays only on the limit points of ww and M , which have order types ww and M ' , respectively, following his winning strategy in Gn(wW, M ' ) ; since PLAYER 11 must have chosen a non-limit point, PLAYER I can win in two more moves. This contradiction proves the following result. COROLLARY 13.24: ww is not jinitely axiomatizubfe. It is now natural to ask whether every ordinal a < ww is finitely axiomatizable. Note that we cannot automatically expect a positive answer since, as we have seen, dodoes contain non-finitely axiomatizable linear orderings. We will show that every ordinal a < owis finitely axiomatizable, and that, moreover, an axiomatization of M can be obtained effectively from a description of a. When we speak here of a description of a, we mean of course the Cantor Normal Form a = wn . a,
+ wn-' .
tE,-
1
+ . . . + w . u1 + a,,
where a, # 0. We will show how to construct an axiomatization of M from its description. To prove that the axiomatizations @ a that we present for each M really are axiomatizations, we will need lo show that if A b @a, then A = a. To do so, we will define, for each x , a class &(a) of a-like linear orderings and show that for each linear ordering A , A is G-equivalent to a if and only if A E ~ ( c c )Then, . by virtue of the equivalence of G-equivalence and elementary equivalence, to show that CD, is an axiomatization of a it will suffice to show that if A k CDa, then A E d ( x ) . We first define each d ( w " ) ,by induction on n. Then wO-likeorderings all have order type 1. For n = 1, we have already seen that A o if and only if A c o 5 . a for some a,and so we define these to be the w-like orderings. Note that both 1 and o are finitary, and, in fact, f ( w ) = 3. Now assume that we have defined, for each k < n, the class d ( w k )of wk-likeorderings, that we have shown that A E d ( w k )if and only if A cok, and that we have shown that wk is finitary and that f ( w k )5; 2k + 1. Moreover, we assume that for each k < n, A E d ( w k )if and only if A = 1 i E W ), where W is cok- I like and each 4 is w-like, and that A E d ( o k )ifand only if A = Kli E W } , where W is d i k e and each is wk-l-like. (Note that for k = 1 both of these equivalences hold.) Now suppose that A wn.L,et A' be the set of all limit points of A . Then A' must have order type in d ( ~ " for - ~if)G,2 , - l ( A ' , o " - l )E I, then, by sticking to his winning strategy in G2n-l(A',w"-1),PLAYER I can force
-
+
N
I{
-
w.
I{
13.
258
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
to a non-limit point and thus win in G2,,+,(A,w").Between each pair of successive limit points of A, the interval must be o-like. Thus A = x{l4(\i E W } ,where W is o"-'-like and each is o-like. Since W is on-I like, it can be written in the form W = 51j E V } , where V is o-like and each V, is wnV2-like.Thus A = c ( c { w l i E y } l j E V } ; by the induction W , ( iE 5 ) is on-'-like, so that A = c { X j ( jE V ] ,where hypothesis, each V is d i k e and each X j is w"-'-like. More explicitly, PLAYER II
I{
I{
-
where each l4(. and each W,' is w"-'-like. Thus we define &(a")to be all linear orderings of this form. Clearly, if A E &(on),then A o" by the 2n + 1. Thus the following is correct. induction hypothesis. Also, f ( w " ) I PROPOSITION 13.25: Define a sequence { d ( o " ) l n< Q } of sets of linear orderings, by induction on n, as follows:
(0) d(1) = all linear orderings of order rype 1 ; (1) d ( o )= all linear orderings of order type w + 5 . a for ci arbitrary; (n) &(on)= all linear orderinys of form
where euch
4, and each W,' A
is in &(on-'). Then, ,for each n,
E &(d)
Moreouer, f(o")= 2n + 1.
if and only if
A
-
on.
Proof : We have verified everything except that f(o")2 2n + 1. This follows from the fact that o"r r 2 , , o"+ w" (Theorem 6.18) which is not on-like. . u2 + . * . + Now let cc < owbe arbitrary and write a as w"' . a , + on2 . ak in Cantor Normal Form. We say that A is a-like if and only if A can be written in the form (link
where each Wyt is an w"[-likeordering, and we let &(a) be the set of all cc-like orderings. THEOREM 13.26: For every a < ow,A over, a is jnitary.
E &(a)
i f and only i f A
-
a. More-
2.
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
-
-
259
Proof: If A E &(a), it is clear that A a. Assume that A CI and that a = wnl . a, . . . wnk. u k .Then PLAYER I in his first c { a j l 1 I j I k } moves chooses the first elements of each of the aj copies of con, for all j. If between two successive choices of PLAYER 11 in A the interval is not co"J-like for the appropriate n j , then PLAYER 11 is punished in 2nj 1 more moves. Hence A E ~ ( c I )Moreover . f ( u ) I 2n, + 1 + c{ujI 1 Ij I k } .
+
+
+
We will now pattern our axiomatization of each a < owon the constructions above; that is to say, we will axiomatize anby saying that its limit points are w"-'-like. To do Ihis, it will be very helpful to introduce the idea of relativizing formulas of L . Relativization is useful when one wants to speak about a substructure of a given structure as a structure. Thus, for example, given a linear ordering A and an element a E A , one can write a statement about A'" that says that every element of A'" has a successor in A'"; such a statement would be simply (Vx)(3y)(x < y), since this says about any linear ordering that every element has a successor. If one wants to write something about A and a that says that A'" has this property, one would use instead the formula (vX)(X
+
( 3 y ) ( y< U
A X
< y)),
observing that A k (VX)(X < u
+
(3y)(y < u A
x < y))[a] A'" 1 (Vx)(Iy)(x< y).
if and only if
We say that the new formula is the old one relativized to the formula e(w,u ) = w < u, since every quantifier ( Q w ) is interpreted as applied to { w ( w < u } . Now for the general formulation. DEFINITION 13.27: Let O(w,vl, u 2 , . . . ,uJ be a formula of L. We define for every formula 4 of L the relativization by induction on formulas:
+@
(1) If 4 is atomic, then 4' is 4, (2) If 4 is i $then , 4' is i($@). ( 3 ) I f 4 is $1 A $ 2 , $1 " $ 2 , IC/I $ 2 , Or $1 $ 2 , then $'is $ i @ A $2', v $2', + 4b2', or $,' ++ $2', respectively. (4) If 4 is (3y)$, then 4' is (3y)(,O(y,u l , . . . ,u,) A $'). ( 5 ) If 4 is (Vy)$, then 4' is (vy)(O(y,u l , . . . ,0,) + $7. +
Thus, for example, the statement (Vx)[(Vy)(y < X ( W ( Y < < 4) (3X')(('dy)(y< x' -+ (3z)(y < Z < X')) +
+
A X
< X')]
260
13.
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
-
is exactly the statement (Vx)(3s')(x < x') relativized to the formula (Vy)(y < w -, (3z)(y < z < w)) and thus says of a linear ordering that it has no last limit point. Similarly, the statement (3x)(Vx')(.x I x') relativized to ( V y ) ( y < w + ( 3 z ) ( y< z < w ) ) says of a linear ordering that it has a first limit point. More generally, given any statement 4, the relativization of 4 to ( V y ) ( y < w -+ ( 3 z ) ( y < z < w ) ) says of a linear ordering that its limit points have property 4. We now explain this more precisely. Given a structure A and elements a,, a 2 , . . . ,a, E A , let (A,ul, . . . ,a,)" = jblA I= 8 [ b , a , , . . . , a , ] } . Then what 4' says about ( A , a l ,. . . ,a,) is that 4 holds in ( A ,a,, . . . ,a,)'. This is a corollary of the following lemma, which is proved by induction on formulas. LEMMA 13.28 : Let @(w,v,, . . . ,c,) be a formula of L. Then for any structure A , any formula 4(x1, x 2 , . . . ,x,), any a , , . . . ,a, E A, and any b l , . . . ,b, E ( A , a , ,. . . ,a,)',
A
if and only if 4'[bl,. . . , b n r a i . .. . , a t ] (A,a,, . . . ,at)' 1 4 [ b , , . . . ,&I. A
COROLLARY 13.29: Let O(w, u l , . . . ,u,) be a formula of L . Then .for any structure A , any statement 4, and any a , , . . . ,a, E A,
A I. 4'[al, . . . .a,]
if and only
if
(A, a,, . . . ,a,)' I. 4.
A
Thus, for example, if A is a linear ordering and 4 is a true statement about the interval [a,, a 2 ] of A, then, choosing O(w, u,,u,) to be u , I w I u 2 , we see that A k @[al, a2] since 4' makes the same assertion about A that 4 makes about [a1,a 2 ] . COROLLARY 13.30: Let O ( w ) be a formula of L. Then for any structure A and any statement 4,
A b
4'
if and only if
A' k
4. A
Thus, for example, if A is a linear ordering, then 4 is a true statement about the limit points of A if and only if A I= 4', where B(w) is ( V y ) ( y< w -+ (&)(y < z < w ) ) . Note that for any formula 4, qd(4') = qd(4) qd(O), so that if 8 is quantifier-free, the quantifier depth of 4' is the same as that of 4. Also note that if 4 is a statement and O ( w , u l , . . . ,u,) is a formula, then the free variables of 4' are u,, . . . ,u, but not w.
+
2.
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
261
The substructures of A we discussed in our treatment of relativization are all definable, so it seems appropriate to define that notion before returning to our discussion of ordinals.
DEFINITION 13.31 : Let A be a linear ordering and let B G A . We say
that B is definable if there is a formula O(w) such that b E B if and only if A k O[b]. Let c1,c2, . . . ,ct E A ; we say that B is definable over c 1 , c 2 ,. . . ,c, if there is a formula O(w,y l , y,, . . . , y,) such that b E B if and only if A k O[b,ci,cz,. . . ,cJ.
Thus, for example, the set A’ of limit points of A is definable by the formula ( V y ) ( y < w -+ ( 3 z ) ( y< z < w)), the interval [ b , , b21A is definable over b,, h, by the formula y, 5 \v 5 y,, and the interval [a, + ) A is definable over a by the formula w I y , . The results above can then be interpreted as follows. COROLLARY 13.32: (1) Let B be a dejnable subordering of A . Then, given any statement 4, we can ejj%ctively,find a statement q5A such that
Bk
4
f and only fi
A k
4A.
(2) Let B G A be dejnable over. c,, c,, . . . ,ct. Then, given any statement can efectively find a forniula 4 A ( y l ,y,, . . . ,y,) such that
4, we
Bk
4
i f and on/.) if
A k 4*[c,,c2, . . . ,c,].
Proof : Take to be @, where O defines B (over c,, c,, . . . ,c,), and apply Corollaries 13.29 and 13.30.
Given a linear ordering A, which subsets B s A are definable? Suppose, for example, that B G g and B is definable: If b , , b , ~g, then there is an automorphism f of g such that f ( b , )= b,. Hence, by Exercise 12.6, for any formula 8, q k 8[b,] if and only if q k 8[b,]. Thus B must be either @ or all of g; that is, g has no definable proper subsets. The same is true of 4, and, in fact, whenever A is transitive, A has no definable proper subsets. Which subsets of o are definable? For a complete analysis, see the exercise below. EXERCISE 13.33: (1) Show that each element of o is definable (where a € A is definable if ( a ) is definable) and that each finite subset of o is definable; show that the complement of each finite subset of o is definable. (2) Show that the only definable subsets of o are its finite and cofinite subsets. [Hint: Look at o 4.1
+
13. T H E FIRST-ORDER
262
THEORY OF LINEAR ORDERINGS
(3) Suppose that 4 is a statement which is true in infinitely many nonisomorphic finite linear orderings. Show that 4 is true in n for all but finitely many n.
Now let O1 be the axiomatization of o discussed earlier; note that qd(@,,)= 3. Let O(w) be the formula (Vy)(3z)(y < w + y < z < w ) discussed above, which says that w has no immediate predecessor. Then A k Qle if and only if A' I=a,. But A' = A', so that A b Ole if and only if A' is d i k e if is an axiomatization of 0'. and only if A E d ( w 2 ) .Thus THEOREM 13.34: For each n > 1, LteJine @, axiomatization of on.Moreover, qd(@,,)= 2n + 1.
Then @,, is an
=
Proof: By induction on n, as in the argument for n = 2. Moreover, since qd(@,) = 3 and qd(O) = 2, we can show by induction on n, using the fact that qd(de) = qd(4) qd(O), that qd(@,,) = 2n + 1 for all n. W
+
Note that the quantifier depth of the axiomatization a,,of W" is precisely
f ( d )as , predicted by Proposition 13.25, so that these axiomatizations are
optimal. . u2 + . . . + w " ~. a k , Turning now to the general case c( = w"' . a, + on2 we see that we want to axiomatize CY by saying that there are certain points such that the intervals between them are wk-like;for this we need to relativize to the formula ~(w,u,,u2) that is v 1 I w < u 2 . We thus define @(a) to be ( 3 x 1 ')
' ' '
(jXi,)(3X12)
A(X1'
' ' '
(jX:,)
' ' '
5 y)
(3Xik){(Vy)(X,' '
. . A @ e ( w . d 1 . x l 2 )A ae(w,x,2,x22)
nl
. . .A
(]Xik)
< " ' < xi, < X1' < . . . < XZ2 < . . * < X l k < . . < X i k )
A @e(w,x,',Xl2) A . A
' ' '
n1 @ ~ ( W , X , ~ . X ~ ~ )
nk
A"
A,
. .A ae(w%x:2.x,3) nz
"2
.A ( D o ( ~ ~ ~ : k _ A @e'(wdk) nk
nk
1,
where B'(w, vl) is u I I w. The following is then easily verified. THEOREM 13.35 : For each ordinal a < ww,@(a)is an axiomatization of
CY.
A
We conclude this section with several theorems of a technical nature specific to linear orderings that result from the equivalence between logical equivalence and G-equivalence. These theorems will be used in $813.4 and 13.5. Recall that for every k and n, Lk,,,consists of all formulas of L whose free variables are among u l , . . . , uk and whose quantifier depth is at most n. We showed that there is a finite subset @k,,, of Lk,,, such that every formula
2.
GAMES AND THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
263
of Lk,,is logically equivalent to a formula of @ k , , ; thus if 4 E a,,,, then there 4. Now let A be a structure, let is a II/ E @ k , , that is logically equivalent to i a,, . . . ,ak E A , and let @ = @(A,a , , . . . ,a,) be the conjunction of all 4 E @ k , n such that A 1 b[a,, . . . ,a,]. Then A 1 @[a,, . . . ,a,] and for any B and b , , . . . ,b, E B we have (A,a,, . . . ,ak) =,,, (B, b , , . . . ,bk) if and Only if B b @ [ b , , . . . ,b,]. We can thus find a finite set {Qi(ul, . . . ,u,)l1 I i I t } of formulas of Lk,, that completely describe all of the =,,,-equivalence classes. That is, for each structure A and elements a,, . . . , ak E A, there is a unique such that A 1 Qi[a,, . . . ,a,] and, moreover,
B k a i [ b , , . . . , bk]
if and Only if
(B,b , , . . . ,bk) - k , n (A,al, . . . ,a,).
For k = 0 this says that there is for each n a finite set {Oil1 I i I t} of statements of quantifier depth at most n that describe all =,-equivalence classes of linear orderings. That is, for each structure A, there is a unique misuch that A 1 mi and, moreover,
B 1 ai
if and only if
B =,A.
This much we can say about arbitrary structures, not just linear orderings; with linear orderings we can carry this analysis much further. Note first of all that each Q imust specify the ordering of the variables. We will work with the case where @ = mi logically implies that u , < u2 < . . . < u,; any permutation can be dealt with similarly. Let A k @ [ a , , . . . , a k ] , so that, in particular, a , < a2 < . . . < a,. Let A , , , 4 , , . . . ,A,- ,, A , be the intervals (t, a,], [a,, a,], . . . , [a,- ,,a,], [a,, +) of A . By the discussion above, we can find statements Y o , 'PI,. . . ,y k that completely describe the =,-equivalence j I k - 1, let YF be the relativizaclasses of A , , A , , . . . , A , . For each j . 1 I tion of Y jto e(w, u j , ujt = ( u j I w Iu j + ,), so that B k Y F [ b j ,b j +, ] if and only if [ b j ,b j +,] 1 Y j If and only if [ b j ,b j +, ] =, [ a j , a j + , ] . Let Y o Abe the u,) and let Y / be the relativization of relativization of Y oto O(w, uL) = (w I yk to o ( w , u k ) = (0, 5 w). Let y ( u 1 , . , . ,u k ) be the formula (01 < u 2 < ' ' ' < u k ) A YoA(u1) A YIA(ul, u,) A . . . A Y f - .1(uk- u,) A y t ( u k ) . It is easily verified that Y(u,,. . . ,u k ) characterizes the =,,,-equivalence class of @,a,, . . . ,a,); that is, A k Y [al' . . . ,ak] and, for any B and b , , . . . ,b, E B,B 1 Y [ b , , . . . ,b,] if and only if (B,b , , . . . ,b,) E k , , (.4,a,, . . . ,a,). Furthermore, since each Y j E Lk,nand relativization to quantifier-free formulas does not increase quantifier rank, we conclude that Y E Lk,,. We summarize this discussion in the following lemma.
,,
LEMMA 13.36: Let A be a linear ordering and let a , < a, < . ' ' < ak be elements of A . Then there is a ,formula Y(u,, . . . ,u k ) E L,,, such that A k Y [a . . . ,a,] and such that B 1 Y [ b , , . . . , bk] if and OnlJ) if (B,b , , . . . ,bk)-kk,,(A,al,. . . ,Uk).
264
13.
THE FIRST-ORDER THEORY OF LINEAR ORDERINGS
Furthermore, Y can be chosen to be a conjunction of v1 < vz < . . ' < v k with formulas Y ~ ~ ( U Y~ l A)( v, l rvz), . . . ,Yt- ,(uk- 1,ok)r y,"(uk) so that for any linear ordering B and elements b,, . . . , b, E B we have B
YoA[b,]
if andonly if if and only if
(+,b,] =n(+,u,], [bk, +)
B 1 Y',"[bk] and fbr each j , I 5 j 5 k - 1, B =! Y P [ b , , b j + l ]