SUBSYSTEMS OF SET THEORY AND ANALYSIS by Harvey Martin Friedman SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY August, 1967 Signature of Author.,.......i.-.... Department of Mathematics August , 1967 Certified by ... rw. ..-........... Thesis Supervisor L ;1 by....... Accepted -. ivS 0 ..... - - - · .. -· - . - I. ·. ·0 0 0· Chairman, Departmental Committee on Graduate Students 2 SUBSYSTEMS OF SET THEORY AND ANALYSIS by Harvey Martin Friedman "Submitted to the Department of Mathematics on August , 1967 in partial fulfillment of the requirements for the degree of Doctor of Philosophy." Abstract Usual set theory is formulated in terms of closure conditions. A typical example is the power set axiom, which asserts closure under the operation of power set. In Chapter I we consider set theory based on closure conditions applied only to definable sets. We formalize this set theory and call it ZF*. Our principal result of Chapter I is that, provably in first-order arithmetic, ZF is consistent if ZF* is. Our proof of this theorem uses a Skolem hull construction and a syntactic transformation. In Chapter II, we consider three theories of hyperarithmetic analysis, A1 -CA, 21 -AC, and 1 -DC. : These are called theories of hyperarithmetic analysis primarily because the hyperarithmetic sets form a minimum w-model for each of them. 1 1 We first show that £1 -DC is a conservative extension of 1-CA for purely 21 sentences. The proof is by means of an inner model construction. Careful attention has to be paid to limit the axioms we use to prove relevant sentences about hyperarithmetic sets. We then show that there are theorems of which are not theorems of finding a suitable sentence 1 -AC. S 1-DC This is done by first and considering the auxiliary theories :1 -AC + S and :-1DC + S. We then obtain our inde- pendence result via Gdel's Theorem, by showing that Con(Z1 -AC + S) in £1-AC of is provable 1 -DC + S. Last, we show that is a conservative extension, for purely T, 21 sentences, a natural subsystem of predicative analysis. The proof uses an inner model construction on certain auxiliary theories. Thus, a model for each finite subsystem of Z1 -AC is obtained as an inner model of a model of an extension, by the negation of an instance of induction, of a corresponding finite subsystem of T. Thus non-standard models are implicit in the construction. Chapter III is concerned with hierarchies (based on the Jump operator) on recursive linear orderings. Let X be the set of recursive linear orderings which have no hyperarithmetic descending chains. Joseph Harrison showed that there are elements of X, which are not well-orderings, on which there are hierarchies. We first show that under certain weak conditions on a recursive linear ordering, that if there is a hierarchy on it, then it must be in X. Finally, we establish the existence of a recursive linear ordering which is in X, yet on which there are no hierarchies. The proofs of these assertions use certain Lemmaswhich are proved in the following indirect way: one assumes the Lemmais false, and then forms a theory consisting of the negation of the Lemma together with certain true sentences; then one shows that the resultingtheory proves its own consistency. Thesis Supervisor: Gerald E. Sacks Title: Professor of Mathematics TABLE OF CONTENTS Title Page ......... ................... 1 Abstract ................................................. 2 Biographical Note........................................ 6 Acknowledgments.......................................... 7 Bibliography . ............ ................................. 8 Introduction...................................... 9 Chapter I 1. General Situation............... 2. Remarks on Terminology and Notation............ 16 3. Axioms......................... 0 0 4. Outline of Proof of Main Theorem a 0 a 5. Developmentof Ordinals......... 6. Development of 7. ...............15 ,, 0 9 00 00 0 0 & 16 0 .0 00 & * 0 18 0 a 0 * 0* 00 & 0 0 21 L............... * 0 0 0 00 0* * 0 0 25 The System ZF*' ................ 0 0 0 0 00 00 & & 0 32 8. The System ZF' ................ . 0 & * 00 90 0 0 0 37 9. The Skolem Argument............. 0 0 0 0 &0 0* * & 0 38 ZF, Parameterless ZF, and 0 0 & * *6 0. & 0 0 10. ZF*. 40 5 Chapter II 48 1. Definitions of Systems.......... 2. Preliminary Lemmas.............. 3. Conservative Extension Result... 4. Independence Result.............00000000000.000*00 65 5. Relation between Predicative and Hyperarithmetic Analysis..................... ..0..00 .0. 0..000 52 0.....0.000...0 .. 59 68 ,....0...........0 Chapter III*............................. .00...0.00..0.000S 77 6 BIOGRAPHICAL NOTE The author was born in Chicago, Illinois in September, l948. The author attended Highland Park High School in Highland Park, Illinois from September, 1961 to June, 1964, and Massachusetts Institute of Technology from September, 1964 through August, 1967. 7 ACKNOWLEDGMENTS The author wishes to express thanks to Professors G. Sacks, S. Feferman, and G. Kreisel for their valuable conversations; to Professors Kreisel and Feferman for their valuable correspondence; and to Professor Sacks for his remarkable patience and encouragement. 8 BIBLIOGRAPHY 1. S. Feferman, Systems of Predicative Analysis, Journal of Symbolic Logic, vol. 29 (1964), pp. 1-27. 2. Joseph Harrison, Dissertation, Some Applications of Recursive Pseudo-Well Orderings, August 1966, pp. 4-7, 31-33. 3. S. C. Kleene, Introduction to Metamathematics, 4. G. Kreisel, Survey of Proof Theory, to appear. 5. W. W. Tait, Normal Derivability in Classical Logic, to appear. 1950. 9 INTRODUCTION Set theory is usually formulated in terms of closure conditions. A typical example is the power set axiom, which asserts closure under the operation of power set. In Chapter I, we consider set theory based on closure conditions applied only to definable sets. We formalize this set theory and call it ZF*, and we give a consistency proof of ZF relative to ZF*. A direct method presents itself for obtaining this relative consistency result; namely, to use a constructible set construction, and prove within ZF* the relativized to the constructible sets of each instance of ZF. This is, of course, in analogy with the method of proof for the consistency of ZF + AxC relative to ZF. However, an examination of the basic principles needed for such a constructible set construction to go through reveals the need for the least counterexample principle for ordinals to be provable in ZF*. By the least counterexample principle for ordinals, we mean the schema (3 a)Pa -> (34a)Pa, where P is any formula. It does not appear that this schema is derivable in ZF*, even if restricted to have only one free variable, a. P is Of course, in ZF, the schema is derivable by means of a closure condition applied to all sets as follows: an a. Then form [l0fea & P, assume (3 a)Pa, and fix such and use Foundation to obtain ("a)Pa. This illustrates the basic difference between ZF*, in that the closure condition, necessarily provable in ZF* 3 only when lpea & Pal, a ZF and is is given a definition. 10 Such a direct attack seems hopeless. An outline of our proof can be found in Section 4 of Chapter I. Towards the end of Chapter I, we show how to add elements "on top of" a model of ZF, to obtain nonstandard models of ZF*. By this means, we obtain results concerning independence from ZF*. In Chapter II, we consider three theories of hyperarith- metic analysis. These are CA A1 -CA, I1 -AC, and Z1-DC. Here refers to comprehension axiom; AC, to axiom of choice; DC to dependent choice. The hyperarithmetic sets form a minimum w-model for each of these theories. We first prove that £1-Dc 1-cA is a conservative extension of for purely 1 sentences. The proof is by means of an inner model construction. We show that given a model of 1-CA, the submodel of all sets hyperarithmetic submodel satisfies 1 -DC. if we then take in a fixed set, this Careful attention has to be paid to the way in which the notion of relative hyperarithmeticity is formalized. The formulation in terms of hierarchies seems to be the correct one here (not ). The usual proof that the sets hyperarithmetic in a fixed set always form an of 2E-DC is too crude for our purposes. w-model It uses the comparability of all recursive well-orderings, which is a principle too strong to be provable in know, in A 1-CA, A1 -CA. However, if we that given orderings have hierarchies (based on the Jump operator) on them, we can then conclude, in AlCA, their comparability. Thus, the: key point of our proof of the conservative extension result is the Judicious use of 11 hierarchies on orderings. Our inner model construction can, in the standard way, be 1_-DC rela- transformed into a finitary consistency proof of tive to A1-CA. The second result of Chapter II is the independence of £-DC from 1-AC. It would seem to be the case that a Cohen type argument would be not only useful here, but perhaps necesWe found that quite the contrary is true. sary. Our proof does not use a Cohen type argument, and Cohen type arguments do not seem to be helpful here, since it seems difficult to find Cohen type models (starting with the standard model, the hyperarithmetic sets, of following property: Z1 -AC) which do not have the any arithmetical predicate (in x) satis- fied to have a solution in the new model has a solution hyper- arithmetic (in x) in the new model. This property can be seen, from our Theorem 1 of Chapter II, to imply that the new model satisfies 1 -DC. Instead, we use G8del's theorem. and consider the auxiliary theories We show that So if But + S Z-1_DC 1 We choose a sentence Z1 -AC + S and proves the consistency of 1_-DC+ S = Z1 -AC + S, then Z1 -DC + S S is chosen to be a true sentence; so S -1DC + S. Z11 -AC + S. is inconsistent. 1_DCf Notice that the assumption Con(M -DC + S) 1_AC. is needed for the independence. We do not know if there is a finitary independence proof (i.e., a finitary proof of consistency of Z1 -AC + ~F in z 11 -oDC). relative to -AC, for some F that is provable 12 The key property of B, 21 -DC £1 -DC + B, one can construct, in key property of the sentence S is equivalent, in IT2 sentence. S is that for any 1 n2 sentence, an W-model for is that the theory B. Z1-AC + S 21 -DC, to an extension of induction by a intuitively says that for all sets x, recursive well-ordering in x proves -DC + S proves S -> every has a hierarchy starting from More information may be obtained than independence. see that The Con(Il-AC + S); Con(21-AC + S), hence which is a purely x. We £i-DC 21 sentence. Furthermore, an examination of our proof yields that C:-AC together with only a finite number of instances of no parameter £1 -DC 2-DC is needed to prove + S 1 3 proves S -> Con(l1 -AC + S). an w-model for Z1 -AC + S. 1 we can conclude that there is a purely provable in 1 Finally, From all this sentence which is ZI-AC together with a finite number of instances of no parameter 1-DC, but which is not an w-consequence of £1-AC. (An instance Z -DC, except the hypothesis, of no parameter 1 -DC is the same as (f)( 3 g)A(f,g), arithmetical withno free variables must have other than A f and g.) The last result in Chapter II is concerned with the relation between 1 -AC ZI predicative analysis, T. and a certain natural subsystem of T represents the first of predicative analysis. We show that extension of T for purely U1 1 -AC cO levels is a conservative sentences. This result, together with the known characterization of the provable ordinals of T, gives the provable ordinals of Z1 -AC. Fur- thermore, our proof of conservative extension uses an inner 13 model construction, which can be transformed into a finitary 1 i-AC proof of consistency of relative to T. Now T known to have a predicative consistency proof; and so, must is then, 21-AC. The relative consistency result is somewhat surprising, since the minimum -model of the minimum w-model of T. ZI-AC is so much larger than With this in mind, it is not surprising that nonstandard models (i.e., non-w-models) must be essential in our proof. each finite subsystem of In the proof, we obtain a model of £1 -AC as an inner model of a model of an extension, by the negation of an instance of induction, of a corresponding finite subsystem of T. In Chapter III, we consider which elements of hierarchies. We generalize W* have to include recursive linear orderings whose field is not necessarily lize hierarchies, W* w. We also genera- so that at successor we merely have a set in which the Jump of the set at the predecessor is recursive; at limits, we have a set in which the effective union of the previous sets is recursive. Harrison proved that 3 neW*-W on which there are hierarchies, in the less general sense W*, W and hierarchies. He left open whether every hierarchy. neW* has a We answer it in the negative for our general notion of hierarchy and the less general notion of W*. We also show, under weak conditions, that if hierarchy, then n has a neW*. The proofs use certain Lemmas which are proved in the 14 following highly indirect way: we assume the Lemma is false, and form a theory by adding true sentences to the negation of the Lemma; we then show that the resulting theory proves its own consistency. 15 CHAPTER 1. General Situation. axiom I Suppose we are given a comprehension (x1 )...(xn)(3 y)Rxl...xny. We are interested here in forming the derived schema consisting of the axioms [(3 tx1 )(3 x2 )...(3 !xn)(FlX 1 &...& Fnxn) (3x where the ) (3x2 ) .. . Fi (3xn ) (FlX1 & F2 x2 &...& Fn -> n & (y)Rx 1 ... xn), are formulae with only the free variable xi. We are purposely vague about the general situation (what is a comprehension axiom?), since we have only looked at this derived schema when the original axiom is drawn from a natural set of axioms, such as ZF, or analysis, or other naturally occurring systems. More specifically, we will look at the schema of schema formed by taking the union of all the schema defined above corresponding to each of the comprehension axioms of ZF. We will not perturb the other (non-comprehension) axioms of except in minor ways. We call this derived theory inessentially modify the Replacement schema in ZF ZF, ZF*. (We for convenience, so that each instance is appropriately placed in the form (x1 )...(Xn)(3y)Rxy, so that we may pass to the derived schema in the manner above.) The axioms of ZF* ZF and are spelled out in detail, in an elegant form, in the next section. We are interested in the relation between ZF and ZF* as axiomatic theories. Our main result is that IFET ConZF* -> system of 2. ConZF. It is obvious that ZF* is a sub- ZF. Remarks on Terminolo and Notation. The only (standard) symbols that can occur in a formula of " e; 2 2-ary relation symbols "=", the ZF (or ZF*) are the 2 quantifiers (3x) and (x); the propositional connectives; and variables xi, zi, ui, vi, etc. Yi, Everything else is nonstandard; when nonstandard symbols occur in a formula, they are meant to be expanded out in such a way that the mere occurrence of a non-standard symbol implies existence of the corresponding set. For example, x = (3 z) (w)(wez would be y is an abbreviation for (3 u)(weu & uz)) & x - z]. Also, say, (3y)[ (z)(zy) & yex]. 3. Axioms. 0. Axioms for predicate calculus with equality. 1. Extensionality. 2. Infinity. 3. Power set. (xO)(3 l)(X2 )(x2 ex1 - (x3 )(x3ex 2 -> 4. Sum set. 5. Replacement variables Then e x x For ZF, we have (x0 (3 xO)( 3 (x0)( and xl) e x xl ) schema. y = 2 )(x2 ex & (x1 )(x1 e x0 -> (x 2 ) ( x 2 Let ( ex l Axy - 2 e x 1 ). x 1 U {x1 le x0 )). x3 x0 )). (3 x3 )(x3 ex0 &3 x2 ex )). be a formula with the free and possibly more free variables Xl,..,x n . 17 (X)''' (n) (Y)[(3 Y 1 )(Y2 )(Y2 E Y1 - (3 Y3 )(AY3 Y2 & (Y4) (Ay3 Y4 -> (The domain is is an instance. Y4 = Y2 ) & Y 3 E ))J and the axiom asserts the y range of the partial function, A'y3 y 2 = AY 3 y2 & (z) (Ay3 z -> z = Y2 ), on the domain Foundation. (xO)(xO 6. y, exists.) (3 x l ) (x l 0 -> & (x2 )(x 2 ex -> ex x2 x O ))). It is clear that by the usual process of making partial functions into total functions axiom schema 5 is the same as the usual formulation in the present context. Now let formula be a formula of 1 free variable. Then the Ax (3 y)(x)(x e y - Ax) For ZF* Same as ZF. 1. Same as ZF. 2. Infinity. (x1 Fin(y)). 5. (x1 A 6. Fin(y) y xO - (x2 )(x2 e x1 e x2 & Ax2 )), xe O -> CA -- > ( 3 xO)(x -> Ax 2 )), A A CA - with 1 free variable, B ZF. e xO 1) ( 3 x 0 )(x 1 )(x 1l x O - with 1 free variable. x0 - (3 x2 )(Ax2 & Bx2 x1 & (x3 )(Bx2 x 3 -> Same as [X1 with 1 free variable. Replacement schema. The instances are E xU is a finite ordinal. Sum set. The instances are 2 ) ( xI 1 will be defined later. Power set. The instances are 4. (3x E xO & (Xl)(X (3xo)(0 Intuitively it means 3. CA. we have 0. & (y)(y E x 0 is abbreviated as CA -> (3 xo)(xl ) x3 = Xl))), for with exactly 2 free variables. ) 18 Our formulation of Power set in Remarks: ZF* is seen to be equivalent to the derived schema of Power set in ZF (as given in General Situation) by noticing 1) that if (3 y)Fy, if then CA, CA, where A These latter are obtained (3!x)(y)(yEx - Ay). then ZF*, by axiom 1 of and 2) that (3y)(Fy & xy) is The same remark applies Extensionality. to Sum Set. the same idea yields that the union of the Essentially derived schema of the instances of Replacement in ZF is equi- valent, in the present context, to the schema (CA1 &C B(x2,x fxl A 1x, A) >>-(3xO)(xlExO (3x2 )(Ax 2 & [x lA2 3,, x ... [x l An 3) & (x3 )(B(x2 ,x 3,[x fA1x,... {XIAnX}) where B free variables, n + 2 has -- > X3 = A 1) ) ) and the the present context) in Replacement in ZF*. A, {x A2 x3 ,.. ment in Bxy (CA1 &...& CAn) -> as ., x Anx}). ZF* is obvious.) NOTE: ENTConZF* - -> ConZF. ZF* ZF*, setting B(X2 ,xl,fxAXX, 1 "In the present context" ZF*." Outline of Proof of Main Theorem. developing in (in (That the above schema contains Replace- means "using the other axioms of 4. have 1 But the above is easily seen to follow from that instance of 5 of as Ai We want to show this schema is contained free variable. A , The main theorem is The first step in proving this is an adequate definition of ordinals, which 19 turns out to be a much more delicate matter than for ZF, due to the lack of certain key instances of Replacement in By an adequate definition of ordinals in ZF*. ZF*, we mean a definition of ordinals such that provably in ZF*, members of ordinals are ordinals, and (for a natural definition in of w) w ZF* is an ordinal, and ordinals are comparable by e, and the e-relation on the ordinals is transitive, and antisymmetric, and antireflexive, and every ordinal has a (natural) successor, except possibly the greatest ordinal. turns out that in ordinal ZF* (It even we can prove there is no greatest for our definition of ordinal given later.) The definition is made and Lemma 1 establishes the above properties for it in the next section; we even obtain more: provably in that, ZF*, the new definition of ordinal coincides with the usual definition given in ZF, on definable sets. (This is made precise in Lemma 1, f).) Next we develop an adequate definition of Among the properties of the predicate x have, provable in w L, ZF*, L every member of an the new definition of x L L ZF*. needed, we must x E L is L, coincides with the usual definition of constructibility for definable x, definable well-ordering of within and a L. With this machinery, an apparently straightforward "proof" of our main result comes to mind. prove the relativized to L Namely, Just to of each instance of by taking least counterexamples ZF of various things in G6del established the relativization of the axioms of in ZF* ZF*, as ZF to 20 L within ZF. But a moment's reflection will reveal that one can hardly expect that ZF* will prove any general least counterexample principles; i.e., one may well be able to prove in (3 x)(xeL & Px), ZF* that (3ix)(xeL & Px) in ZF*, yet not be able to prove where the definable well-ordering of U is defined in terms of In fact, we do not even L. see how to prove each instance of the relativized of L, within ZF* to ZF*! In order to get our main result, we form an auxiliary system ZF*' C ZF*, whose definition depends on a certain crucial transformation on sentences, T. the property that each instance of This subsystem has ZF*' semi-relativized (semi-relativization is a certain modification of relativization) to L, is provable in auxiliary theory ZF + V = L is to ZF' ZF*. ZF*. We form another which is related to It turns out that ZF*' about as ZF' ZF. It also turns out that in the theory obtained by semi-relativizing each comprehension axiom of ZF*' and retaining the other axioms, one can give a Skolem hull argument for each finite subsystem of ZF' that proves the existence of a (suitably definable) model of this finite subsystem, and hence its consistency. Putting all this together we get a finitary proof that (n) ZF COn(ZF n ) . So FENTConZF* -> ConZF. 21 5. Development of Ordinals. We define Ordx = Trans(x) & e-Conn(x) & (x is semi-closed under succession) & there are no 3-chains in & x, i.e., Ord(x) = (y)(z)((ycx & zEy) -> & zx) (y)(z)((yE-x - - yZ v y = V ZE zcx) z)) & (y)(yex-> (3z) [ (w)(WEz -- (wey v w=y)) & (zex v z=x)I) &: (y)(z)(w)([yez & ZE & (wey V w=y)]). We define y Q x) -> Ord'(x) = Ordx & (y)((Ordy & (y x v y=x)) - Ord"(x) - Ord'(x) & (y)(Ord'y -> (xC_y v yC x)). Ord" will be our notion It is obvious that (x) (y)(Ord"x & Ord"y -> For Ord(x), we define of ordinal in (xc y v yEx v y'=X)) is successor of y ZF*. x if and only if y =x U {x. Lemma 1: The following are Theorems of ZF*, where Ax has 1 free variable: a) If Ordx & yx, b) If Ord'x c) For x then and with yEx Ordy. then Ord(x), Ord'y. and not a successor, ltJx = x. x For Ord(x), x a succe ssor, x = y d) If C)rd"(x) e If yr f) If yr = g) Ord"( (w). (Explained 1below.) h) If = {xAxI, Proof : v y=w. then Ord'( y), xlAx}, and a) then wEx We can't have So z EX WEx, Ax -> wey. aj nd To see and so Ord"(y). Ord"(y). Ord"(x), yx. yew v y=w t.Jx = y. Ord"(y). Claim Trains(y). we have 3-chain. and yex fxiAx}, Ord(y ), then y r= Trans (x), and yi, we have Let zy, By then Ord t" Vly). wz. Th,en by -Conn(x), w Ey V yEW because we would h; ave a -Conn (y), let zEcw V WEZ v w=z zey, wey. by T' hen -Conn(x). 22 zIex or z U {zi = x. z U f{zey or z Lt (zz = y z lJ yez Ut {z}t, then 3-chain yey, yy, So yey. Ord(y). for if z=x, then xC zex v z=x. But Hence since if yz, But By zex. So yey. and hence y, zey v z=y v yez. e-Conn(x), a,b,cex, Towards showing Ord'(y), let By Trans(x), z C x. y, Ord(z). z4x, U {z. x. and we have a 3-chain in z' yz which yields the yey, bec, (cea v c=a), a,b,c,ey. Then Now suppose ab, By a), we have But if The first yields the 3-chain yez v yz. the second yields zey, yez, z=x; b) yez U (z}. or we have e-Conn(x), By zex, and zey. Hence semi-closed, let y Towards showing then yez, Y Ey. Ord(x), Let c) (y)(y U with x every member of a member of exists. if Ux {y) ex. But yU member of x if Lix exists. no 3-chains. So Ux C z, is a member of a member of d) xC By z, then y _ z. c), then y '. x. x. x, So lJx exists and is yC z. If zex, Let If Ord'(z). z . x, then then x = z U {zl, since x zx, But wz, If x. if it exists. since if Ord'(y). x, since also every x =-Ux, is a member of a member of x, So UxC x. By semi-closure of {y}, and so yeyU then again Ux C_ x, z is a member of x yex. Now let By Trans(x), x). yi has But every member of then w is a member of z. If z z x. zex v z=x. If z=x, Then yez v zy .v x z=y. Hence x y v 23 e) Let Ord' (y), y = {xJAx . 1) All elements of Ord . For every members of y have Uy Since are If z uez, yC y_ y xC z v z x. 2) Ord". If y = u uEz v u=z. So y Some element of The Suppose If u then u C z. u=z, then zC x. contains z. z. z C y. proved that any be any Either all y D z. Suppose the second holds, i.e., some Then clearly z Then let is not a successor, by c) we z. we have then Ord". or some member of Then if y, and so Ord'(z), are we have xEy the first holds. y Then either Ord' (unique) xEy is ord". y is not Ord". We have such that every member is e-least element of is definable, and has every member an y Ord" is which is not Ord" Ord". by suitable use of Replacement schema of ust (There is a least ZF*, and Foundation.) Just apply 1) to obtain a contradiction. f) We merely have to show all definable Ords are Ord'. As in e) we go down to a definable Ordinal all of whose members are Ord". Let Ordz. We wish to show 3 wEy with If z either w w U z C y, z=w, y be such an ordinal, w·z. then y = zEy v z=y. Suppose But since Ord"w, z=w v zw wc)y v w U and so (wU fw]). Hence wU out, so zC wl {w], and so w] = y. zey. So xfBxl. Let zfy & yz. we have zcy. If wlU w U {w]Ey. So w z = w U {w) v zEW U (w], Then w. z, w z. w) = y Now then Ord" we have zey. Now then since The first is Ord'(w tl (w}) either one implying y, z v z w f {wJ)C_z v z C w U {w). and since z 24 g) We now explain Axiom 3. and (z)(zex -> (3 w)(z=w U Ordx & (3y) (y=xU I xi). Fin(x) = Ordx & (3y)(x=y Ordy) wl)). We let x w is a successor be the x0 in Ord = Ax. 2 of ZF*. Now clearly show is definable, w Ord(w). 1) Trans(w). Let successor Ord or 0. successor Ord or 0 Ord SC) or 0, and so new. Since Let n Then x x is transitive, and every member of x is a x is a is a successor is a finite Ord (i.e., s-Conn (w) . Suppose 3 new nn & mn & nm. Tt ake There is an Fin(x)), k d 0, any -least such k and mew, lm then m=l But if 1 1 1 meW, k. and Foundation. by Foundation lew . We must have, for some mel, then mck. If k=lU Il me'W, k n & then = mvlU mek. If l) em,i .e., Contradic tion, by the lack of 3-chains in Ords, and the comparability directly and call it is a sulccessor Ord, and m k & knm & (lem v me1 v m=l). 3) n ZF* em v 0=m, for any since and by Trans(w), k:=mv kem. such that for some -least by Replacement in and Trans (m). Hence of 1 and m. e under succession. This is insured Semi-closur from Axiom 2 of 4) n ,m, rEW. have xen. XEW. 2) Now and so we merely have to rer, contradict No 3-chains · By Trans(r), if ren. Ord(r). ZF*. Suppose nem, m er, rn we have nr. If r=n, v r=n, By Trans(r) we also have rr. where again, These we 25 Trans ~Jy). 1) wey; hence any Ord". both So 3) z,wE Uy, If w' w or Then zew', Semi-closure. Let Ord"(w), u&Jy. then zez' Iy,ww'EY,z',w' Without loss of generality, v w=z. zew v wz Hence zewey, Ord"(w). Then ze Uy. or ZJfzew. z U {z}Ew So z'. wEw'. for some zew then uEw, wey. has y). z z' C w'. assume z Jy, If uez s-Conn 2) So Ord~(Uy). We have only to show h) z U FzIE Uy. In the first case, In the second suppose some ucy A) so w C u, but wu. has either w=u or has Hence weu, and z U {z} = we Uy. ucy Every B) Uy = w = z u w. Then clearly (z3. A) and B) are exhaustive. 4) a,b,c e Uy. cec. If 6. Then c=a, Develo a,b,c nt of are then aec, aeb, bec, Suppose No 3-chains. L. Ord", cec. of course, aEc. Contradicts If where then cea, Ord(c). We wish to define a class of sets, which has a definable well-ordering, L, and (cea v c=a), and provably so in L, ZF*. We are interested in will not be an object. the predicate xL. We let n,m,r,p,q be special variables We let a,p,,y,... be special variables Ord"x. We write a+l for a 'U {a}. for elements of for sets We let X variable for limit (non-successor and non-null) x with be a special Ord"'s. w. 26 x= M ( a We say -> (3f)(Domf = a+l ) U f(x) = f(X)) & ()(Bea+l -> & f(0) = & (X)(Xca+l f (+l) = Fodo(f(B)) & XEX x (3 xo)(3 n)(x=(zlzey & <y,ey> Fodo(y) = x=f(a)), where n(z)[xo]l & FinSeq(x,y))], <yey> n(z)[x O satisfies the formula with the structure <y,ey> ] which is the sequence starting with z, means G6del at the sequence of elements of the domain n number where y, (z,xo) , followed by the sequence xO. Remark on formalization: We formalize the satisfaction relation in ZF* the same way we do in Terminoloy Remarks on and Notation. Lemma 2: For each Czyl...Yn' and Ax Bzw with only free variables shown, the If and y ={xjAx. and if Y1,'''yYn E y, n(z)[Yl1 ,...,Y n]} tzlzey & <Yey> is the Gdel n ZF*: y}, y U a well-ordering of and for each with 1 free variable, following is provable in where Also see 2. ZF. z= (zlzy & CY C. number of Thus x, is then 1 ...yn] Fodo(x) set of all sets first-order definable over Bzw means the for definable sets x. Proof: both Suppose Czy 1 (zlzey & Czy1 l.. and is (3 w)(wez & way 1 ). {zlzey & <ys by Replacement on the definable y. <ysy> = is routine. nlzvyl ] , for zy. > We note that = n(z)[yl] We want to show ZF* certain sets definable in terms of the members of ZF*, Czy1 The proof in the case of What complicates it in the case of provably exist in exist y ZF is that may not and also that the theory of Gdel numbering may not be formalizable in ZF*. The latter is not 27 the case, since w is definable, and hence by Replacement and Foundation, induction on we may define + and x, w provably holds in ZF*, and also and prove the relevant properties. What comprehension axioms are needed to establish our equivalence? Apparently, what's involved is ust that the theory of finitely hereditary sequences of elements of numbers provably in Yl,y2 ,y3 Ey, n,mew, <{yl,n],{n,mll],[y2 ,Y3,fnfl>." ZF*, Y1,Y 2,y 3, and for sentences there exists the sequence Such sentences can clearly be proved by suitable instances of Replacement in definable sets and natural ZF* have the intended interpretation. For instance, we must verify provability in like "for every y n,m. ZF* for One assumes in ZF* that such a sentence in false, and goes to definable counterexamples Y1 ,Y2 ,y3 and n,m via the definable well-ordering of y ! (y, Bzw. Lemma 3: a) [a = xAx3 & (=a v pea) & (3x)(x=M(B))] -> 3f fying b) (3x Each instance of the following is provable in the conditions given in the definition of a o) xJAx (3 !x0 )(x0 =M(a)). -> xfAx3 -> (Xo=M(P)). Proof: Assume have 2 functions a = {xjAx & (3 x)(x=M(a)). Suppose we f,g satisfying conclusion, take, using Replacement and Foundation in way, B that 3f f(B) satis- x=M(B). Pea = Also ZF*: to be the and g, -least element of fg and ZF* a + 1 B is definable. use Replacement on $ to get the Suppose fyl (3 y)(ye g. We in the usual with the property satisfying definition of g(P). Then f x=M(a), with Lim(B). Then & for all f 28 satisfying x=M(a), f(y) = y)l. definition of sum set in ZF* U. to get a union, the definition of x=M(a), clearly contrary to hypothesis, f(B) Now suppose when of f = 6+1. In any f f(B) = U. satisfying But this is g(B). By hypothesis, f(6) is fixed varies over the functions satisfying the definition x=M(a). Clearly f(8+l) = Fodo(f(b)) fying the definition of independent of f, Clearly x=M(a), again contradicting B f(6+1) f satis- is also the definition of i. exists. Now suppose for some B, and so for any 0. So no such such We can apply Bea, part a) false. Take least and apply above, since least such B is definable. This concludes part a). b) (3!xo)(xo=M(p)) Now suppose = a. for all We conclude the proof of Lemma 3 by obtaining a contradiction. Suppose ZF* write but not for We may assume this without loss of generality, by taking least counterexamples. in BEa, on a M(B) Lim(a). Using Replacement we can get the set of all M(p)'s, Pea. for that with x xo=M() Thus we may take the union by sum set in (We if it is unique.) ZF* and call this Now it is not hard to see, under our hypothesis that using replacement there is an PEa, and consisting of only (a ,U>, and that this is the required definition of from a). f U=M(a). Suppose So a = T-l. <P,M(P)>'s, f in the (3 xO )(xo=M(a)). Uniqueness comes It is easy to see that definable and by our hypotheses, f(y) is definable. y is It is U. 29 obvious that elements of and that P(f(y)) Fodo(f(y)) P(f(y)). 0 are elements of exists by power set in ZF*. P(f(y)) Furthermore, exists since it can be gotten by replacement on Proceed as above to get an appropriate Fodo(f(y)) = M(a). a = Fodo(f(y)) Uniqueness follows from a). f to give The case is trivial. We want to insure in ordering of L ZF* there being a definable well- (among other things). This insurance is easily obtained by a natural definition of L in ZF, but not in We have no choice but to complicate the definition of L ZF*. by adding on conditions. We define xEL, approximately as (3 a)(3y)(y=M(a) & xy). But this is not good enough for our purposes. extra conditions on this a (z)(z=M(a) --> z=y). Whenever say y: (a & xz)] -> (0)(z)([z=M(3) 1) and We define 5 x and v a and =a)), have such a y, we O(x) = a. 2) and if yea, For all B,y < a, then there is a unique corresponding M(B) 3) The M(B)'s, Bea, 4) For every O(z) < a. 5) Also if q M(y). and M(a) zEM(a), we have z,w y=M(B), M(a), then are transitive sets. (3 p)(O(z) = ) and [ (zeM(O(w)) -> Now, there is a usual definable mapping O(z)<O(w)]. F in full set theory (identity this with the 2-ary relation F(x) = y) mapping the constructible ZF* sets 1-1 may well not be able to prove into ordinals. Of course, (x)(3 y)(F(x) = y). ition 5) will be that (the 2-ary relation) F is a 1-1 Cond- 30 function when restricted to domain (x)(xM(a) -> Ord"(F(x))). M(a), (This is the and F which, in full set theory, assigns (GCdel numbers in the form of ordinals) to each constructible set a sequence of ordinals, the first being the rank of the set in the constructible the rest of the sequence codes in, via F hierarchy, a+l; on the sets in M(a), how the constructible set in question is first-order defined over M(a).) xeL = (3a)(3 y)(vy=M(a)& XEy & a We define and y satisfy conditions 1)-5) above). We define Lemma 4: x < y = xEL & yeL & F(x)eF(y), for The following are provable in yL). (y)(yeM(O(x)) -> Proof: we have Let Also nition of yL, since the definition of (3a)(O(z) = a). yL, (y)(yex -> if yeL), xeL if O(y) let • O(x) . zM(O (y)). Then for suppose not. rest of condition 4) follows for since y x does, and To show zeM(O(x)) . and that part of Lemma 4, Then Then we get a xL. The M(O(y)) C_ M(O(x)). O(y), M(O(y)) M(O(y))C_ M(O(x)). (y)(yex -> have (assuming yx) xeL, because of condition i4) satisfies5) (i.e., y, together with y xcL. Towards verifying 4 ) in contradiction via condition 2) in tl he definition of y, then satisfy 1), 2), and 3) in the defi- O(z) _ O(x), being satisfied for as in 5). By 4) in the definition of yM(O(x)). y, M(O(y)), O(y) ZF*: F yeL), that yeL. notice by yM(O(x)), Trans(M(O(x)) we and so by the first 31 is provable in (x)(xEM(B) --> xcL), exists unambiguously by Lemma 3.) Proof of Lemma 5: y Now is a limit. = y. ZF*. Then xM(y). Let ). xM(6+l With this well-ordering such that there is no assuming there is an x. of a, y Case 1. y, There is a definable and we may use this to definably M(6) the finite sequences of elements of ZF*, in Let + 1. in the natural way, proving in x does not satisfy and hence by the definition of M(5), <, well ordering on well-order, is definable in M(B) wEL. Also aljac & M(a) Form ay, for some Case 2. xEL. (Note that ZF*. Take e-least member, and call it conclusion}. xeM(a), B = {xjAx3 -> 1 free variable, Ax, For each Lemma 5: that it is a well-ordering. ZF* M(6+l), M(+1), we take a least, in M(a) satisfying condition 1, This least {aClaE+l & xEM(a)I and so we consequently can form ZF*, is definable in x and take the e-least member, thereby obtaining a contradiction. So every xcM(6+l) possesses a (unique) O(x). Conditions 2)-4) are treated similarly, taking definable counterexamples and using definable well-orderings. The proof of 5), after taking least counterexamples, is much like our indication of construction of a definable well-ordering M(+l) ) on the basis of one for To show weL, Lemma 6: it suffices to prove EL. wM(w+l). The proof ZF. HZF (x)(x=uW-_(x=w)), where to the predicate above. M(5), is like the proof of this fact in of A is A relativized 32 Proof: 7. Left to the reader. The System ZF*'. We define a transformation mapping formula in prenex form in the standard notation (described in 2. Remarks on Terminology and Notation) into formulae which contain the '<( symbol. If B to be the usual prenex form for is in prenex form define B. to be the T Take identity on formulae with no quantifiers, and take to be is B- T((3 xi)Bxi) (3 xi)(T(Bxi)& (xj)(xj< x i -> T(B-xj))).T((xi)Bxi) (xi)(T(Bxi) v (3 xj)(xj < x i & T(B-xj))). proved by induction that T(B-) for any prenex B. and XT(B) It is easily are equivalent Recall that the interpretation of xj ( xi is xL & xeL & F(xj)eF(x i ). We form ZF*' as follows: Extensionality & Foundation & Infinity & Power Set & Sum Set & Modified Replacement. The latter is the only difference between other axioms are the same. follows: in ZF*' ZF* in ZF*' Any instance of Replacement in provided that the is obvious that of and Replacement (z)(z < y --)>T(Bxz)). Bxy Lemma 7: ZF*' Axy ZF* ZF*. The is as is an instance be of the form T(Bxy) & having only 2 free variables. It ZF*' C ZF*. Extensionality & Foundation & Infinity are theorems when relativized to Proof: L. For (Infinity)' take x0 to be w. For (Extensionality)' and (Foundation)' ust note from Lemma 4 that & yx) -> yEL. (xeL 33 The Power set and Sum set axioms (of Lemma 8: ZF* when relativized to theorems of to x = will be equivalent to L zeL & wL. yfAyl. The relativized yjyEL & (Ay)'1 & xL. (z C w)' observe that the relation L. x = In power set, we have Proof: are ZF*) is equivalent to xL & x = xO the set of all ylyeL & then there is a set xoEL with subsets z such that zeL. Now the hypothesis x of is definable, and so x IP(x). We use replacement on O(y)'s with all yeL yeL with x y M(0) C m(a). with and yEL to get the set of all yElP(x). This is a definable set of .J have MJtl), is definable, and so, U yeM(J), The required set is in tells us has a definable power set, x IP(x) and so it has a union, Ord", zC w & So we have to verify that if (Ay)'), that Now x 1a, because if of all subsets y by Lemma 2; hence xoEL, then of x by Lemma 5. The relativized of Sum Set is checked similarly. Lemma 9: If all (3 y)(y < z & Ayxl...xn ), and Ea = fxjBxl, (That is, if A then 3 O(y), with O(z), O(x1 ),..., O (xn) y < z & Ayxl...x n . any formula with the free variables shown, any formula with 1 free variable, the above is provable in Proof: Z,Xl,...,xn provably a M(a) Lemma 10: O(y) to this lemma, and then go to definable But we obtain a contradiction, since there is for these supposed definable counterexamples. The semi-relativized of each of Replacement in is a theorem of ZF*). ust assumes there are counterexamples One counterexamples. B ZF*. ZF*' 34 Proof: ZF*', We take a particular e.g., that one whose (u)(u < y -->T(( We let D Axy instance of Replacement in 3 z)(w)Czwxu)), where be a definable domain, ZF* that there is a set xcL with xeD, y SL T(( 3 z)(w)Czwxy) & is C DEL. of all is the unique is quantifier-free. We wish to show in yeL y such that for some with (Axy)'. This is easily seen to be equivalent to finding a set SL of all yEL such that for some 1) xEL & yL & (3 Z))L(w)L(Cwxy V (3 WI)L(W, < w & & (')L(Z' < z xD, (3 w)L(~Cztwxy & (wt)L(wI < W -> -> Cztwtxy)) & (u)L{U < y ->(z)LE (3w)L(~Czwxu & (w')L(W'< w -> (3z')L(zt < z & (w)L(Cz'wxu v (3 w t)L(w Convenient Notation: If in 1), X then let X,Y] and ending with Let xeD. Czw'XY)) X and Y t Czwxu)) v < w & 'Cztwtxu))]]. are expressions occurring be the subformula of 1) beginning with Y. U = union of the O(y)'s such that 1) holds for some We proceed to place definable bounds on the quantifiers above in such a way that the new formula is equivalent to 1) for y xD, yMJ). (Note that for each xeD there is at most 1 satisfying 1).) for Let fl(<x,y>) zeL with of the range of Let O(w) for E(W)LCz'w'xy] fl f2 (<x,y,z>) wL be undefined with on if 1 is false: be otherwise. Define O(z) t!1 = union D x MJ). be undefined if "[Czwxy, E(w)L,I~Czwtxy]; be Czw'xy], otherwise. Define 35 L2'2 = Let otherwise bie ,Czw ' xy. (See Lemma 9.) on Define f4 on f5 union of the range of Let f 7 (<x,y>) f6 on f7 O(z) on x M) O(w) x M4). r(w')L,Cz'wxy]; (u)L,Cz 'w'xu]; Cz'w'xu]. be undefined if for zL with with f8 wL U9 = union of the range of fO(<x,y,u,z,w>) uO(wt) with union of the range of if = D x M(U) x MJll) x M(U4) x MJ ). 5 other- Define U 7 = union on f1 0 f9 (Z )L,"CZ 'w'xu]; 3w)L,~Cz'w'xu] . ~P[( Define D x M(U) x M(U7 ). f9 (<x,y,u,z>) be undefined if otherwise be [(3 w)L,Cz'w'xy]; if D x M(U). U 8 = union of the range of otherwise be =4 wt < w & .Cz'w'xy. Define U 6 = ,[u < y, f8 (<x,y,u>) otherwise be be x M(U1). be undefined be undefined if with of the range of D x M) on D x M) with PO(w') pO(u) [(z')L,Cz'wxyl]; otherwise. Define tJ be undefined f 6 (<x,y,z, ',w>) otherwise be Let w' < w & with [CCz'wxy,Cztw'xyj otherwise. Define U union of the range of Let with ) be undefined if f5 (<,y,z,2 '>) be O(w) for wL Let w' U3 = union of the range of 4[z' < z, Cz'w'xy, with Let wise be D x M ) x Me 1 ). O(w' ) for union of the range of Let on D x M(U) x M( l1 ) x M(J2). Let f4(<x,y,z>] 'O(z') f2 f3 (<x,y,z,w>) be undefined if (w')L(W' < w -> Czw'xy); f3 of union of the range .[(3w)L,Czw'xu]; and [Czwxu, Czwtxu]. Define on D x MJ) x MU be undefined if 7 ) x MJ ). [(w')L,Czw'xu]; w' < w & "Czw'xu. Define U on D x M)) x M 8 7) xM 8) = x Ml 9). 36 Let fll(<x,y,u,z>) 0O(z')with otherwise be union of the range of Let be undefined if [zI < z,CzwtIxu]. fll on f 1 2 (<xy,u,z,z'>) otherwise be O(w) for '[ (3zt,)L,Czw'wxu]; D x M(j) x M J7 ) x M be undefined wEL with U12 = union of the range of Define fl2 = Ull 8 ). 4 [(w)L,v Cz'w'xu]; if '[Cz'wxu,,Cz'w'xu . Define on D x Ml) x M(U 7) x MJ 8 ) x MJl)Let f13(<x,y,u,z,z',w>) otherwise be PO(w') with union of the range of w' < w & fl ,[(3 w,)L,._Cz'wIxu]; be undefined if on Cztwtxu. Define U = 3 D x MVJ) x M(U 7 ) x MPJ 8 ) x Mu11) X M12) Note that by suitable instances of Replacement in all of the above are provably well-defined. ULi is definable, that for xD, so that each yMJ(U), Note that each It is easily seen it is the case that to the predicate Bxy 1 < i < 13 M(Ui) C L. ZF*, Axy is equivalent obtained by placing the bounds Mi), on the appropriate quantifiers in 1). Now each instance of the following is provable in If a = x < y xjAx), iff and x < y a limit, then for M(a), A of and of 1 free variable. of the schema is like the proof in well-ordering x ZF. Use the definable Then relativizing the quantifiers occurring in the expansions of the "<"ts for xED. Bxy, The proof in M(a). Now let; V = max~J,Ui,O (D)). occur in yEM(a), holds when the quantifiers in the definition are relativized to ZF* a ZF*: to M(V+w), that we get the same predicate as Hence we have shown our S we wanted to show Bxy, 37 originally EL, is first-order definable over EM(V+w+l), V + w + 1 hence 8. The System M(V+w), and definable. ZF'. Making use of the transformation T defined in the previous section, we form ZF' as follows: first, Extensionality & Foundation & Infinity & Power Set and Sum Set axioms of ZF. x) (x)(xEL) & (: In addition, we have Ord"(x)) & (x)(3 y)(z)(yEx & (zEx -> (Ordx -> in Replacement ZF' z y)). Let will be the following: Bxyyl...; Yn be a formula in prenex form with only the free variables s hown. Then (Yn)(x) (3 X1) ( 2 ) (x2 eX1 (Yl). (3 x3)(x3 T (Bx3x2Y1.. 'Yn) & (x4)(x4 < x2 -> instance. Lemma 11: First, we wish to show in This is trivial for Suppose T(A) - A prenex form with prenex form. any A quantifiers. B B ...Yn)))), )T(Bx i ZF' each instance of with no quantifiers. A having (3 xi)(Axi). Then n + 1 quantifiers in T(A) - A n + 1 ,Axj )) is provable in provable for T(B) quantifiers. is (xj)(xj < x i v T(A-x))). Now FZFT(Axi) T(Axj) we have is equivalent to T(A-xj) We have to check that in ZF t'. be in prenex f orm with is for all A ZF' We then wish to show that Then we will have shown in prenex form in Let Suppose n A is provable in is provable with T(A) - A s an 1 3 x4Y ZF' D ZF. Proof: T(A) - A. & (3xi ) (Ax ZF '. i) (3xi ) (Ax i (3xi)(T(Axi ) Ax i . & Since zFT(A-xj) & (xj)(xj < <x1 Xi V 38 Define Cxy to be equivalent with n T(Ay & y=x), (x4 )(x4 ( x2 -> ZF'. But ZFT(Ax Choose any such 2 & xi. = x3 ) 2 wff of (3 x)(xEx0 & Cxx2 & Ax 2 & X2 = x 3. Ax 2 & x2 exO). Take O(xi), and use the above theorem of M(O(xi)) a transforration on a ~T(Ay & y=x))))), is (equivalent to) an axiom ZFI ,(x)(3 xl)(x2 )(x2 exl of This is, of course, (x0 )(3 x1 )(x 2 )(x2Exl quantifiers. Now of y = x & T(Ay). ZF' Now assume and set So (3 xi)Axi. xO = M(O(xi)), to get the set of all elements having the property A. (M(O(xi)) is defined and has required properties since (x)(xeL) is an axiom of and ZF*, previously.) Hence by xeL is formalized as in one of the axioms of ZF', ZF', there is a (-least member. Hence we have shown by induction the equivalence between and A, in ZF'. eliminating the 9. This has the effect of provably in T's in the axioms of ZF', The Skolem Argument. We wish to show where ZFn is the first enumeration of them. suppose But then n axioms of and so T(A) ZF' ZF' 3 ZF. (n) ZF* Con(ZFIn) ZF' in some natural If we succeed in showing this, then uCon(ZF). Then ~Con(ZF'). Then (3n)ENTCon(ZF). Since ENT (3 n) Con(ZFA). is formalizable in ZF*, (3 n) IF*Con(ZFn). Hence (3 n)(FzF*Con(ZFA) & FZF* Con(ZFn)), and so Con(ZF*). Hence ConZF* -> ConZF. We give, without loss of generality, a Skolem closure argument within ZF* to give, provably in is a model for 1) Extensionality 3) Infinity in ZF, 4) in Power Set in ZF, 2) ZF*, a set which Foundation ZF, 5) in Sum Set in ZF, ZF, 39 6) (x)(xEL), 7) z y)), 9) 1) in 7. (x)(Ordx -> Let D(xyT) The System free predicate E(zwxyT). C, ZF*I Ord"x), 8) (x)(3y)(z)(yEx& be the formula obtained from taking and replacing the 4-place quantifier- with some 5-place quantifier-free predicate (3 x)(xED & D(x,y,T)). (D)(T)(3S)(y)(yES The construction, in ZF*, of the model of these 9 sentences will be much like a Skolem construction in which the . initial model is At each stage and we take the union as x:EL, We simultaneously U in define n an n, we throw in some sets ranges over and Sn . w. We are interested S nEw n Sn = M(an). S = M(/) = 0. Consider, for each xS , n (zEy -> aO = . the <-least yx with (z) z x). Consider, for each XESn, PL() Consider, for each x,yESn, - set of all y x with yeL. element of x not in x y, the <-least y. Consider, for each that with XESn, UL(x) = set of all TSn, the unique yL such (3 z)(zex & yEZ). Consider, for each semi-relativized We continue of 9) to L. SL satisfying the (Call this 9)"). "considering" through 9)", closing Sn in effect, under the "Skolem functions" for 9)", in such a way that, as in 7. The System ZF*, we have that the Skolem func- tions produce values definable in terms of the arguments. 40 We take an+l = (union of the considered above) + w. Take O(z)'s Sn+ for the z's M(an+l). We can then use appropriate instances of Replacement in ZF* in combination with the definable well-ordering <, show that if the n, Sn and an are not well-defined to for each then there are definable counterexamples to our construction in the following sense: for some specifically definable sets, the sets corresponding to them that we considered above do not But this is impossible by Lemmas 7, exist. our construction is well-defined in 8, 9, and 10. So ZF*. i Sn is an M(a), a definable, a a new In particular, it is transitive. It also contains Now our model limit. Due to the absoluteness of the definition of definition of in ZF*, w in M(a)'s, a a limit, and of the it is easily seen putting all this together, that the sentences 1)-9) are true when the quantifiers range over since L w. M(a) Furthermore, is definable, the definition of satisfaction and the induction on to prove, in M(a). a ZF*, are easily developable in that ZF*, in order Con( 1)-9) ). From the remarks at the beginning of this section, we immediately have Theorem 1: 10. F ConZF* -- > ConZF. ENT ZF. Parameterless partially ZF, and ZF*. to further consideration This section is devoted of the system ZF* Chapter 1, and partially to some other subsystems of of ZF. We define a sentence of set theory to be arithmetical if it is the relativized of some sentence of set theory to Corollary 1: is a conservative extension of ZF w. ZF* for arithmetical sentences. Proof: Let A ,A) -> Con(ZF* + be arithmetical, and Con(ZF + IZFA. by modifying the proof of A) .A, ZF' + A, our Lemmas carry over. Con(ZF* + A), Due to Lemma 6, respectively. Now since and so ~Con(ZF + as ZF*I, ZF' Theorem 1 slightly; Just redefine the systems ZF*' + We can show all of .A), we have zF*A. Our next Theorem concerns sentences of the form (x)(3 !y)Axy, arbitrary, with only 2 free variables, that A ZF*. are provable in Now in which define, provably, in ZF, everything and which is 1-1. Another is (x)(3 y)(y Theorem 2: ZF Not so in ((Axy & Axz) -> Then Proof: We let z). C we construct a model for for X Q ZF* + XC, ZF*. R given an arbitrary model is a 2-ary relation on as follows: The domain is to be is the rationals. The 2-ary relation, Sxy, defined as yeX; is not provable in X, Note is a first-order theory with equality.) We define where C x) & (x)(y)(z) (All models are assumed to be equality models. ZF* that = be a sentence of the above form, and ZF, G\ = <X,R>, where pi . ZF*. Thus, (x)(3 1y)(Axy & y C = (x)(3 y)(y = IP(x)). be any formula with only free variables shown, and let y a Skolem function which moves An example is {x}). Let Axy there are many such sentences Rxy true if if xX, x,yeX; x < y yeQ. if X U Q, is x,yEQ; false, if xEQ, U We claim ZF* +C. satisfies First, we show that the elements of indistinguishable in for G in the sense that if xicQ, yjEX, then so does Q Q, same order relations in are Axl...xnYl...ym holds AZl...znyl...y for if the two sequences of rationals, zieQ 3 in xi, m have the i (i.e., there is a 1-1 order To see this, it suffices to show that, given preserving map). such a pair of similar sequences of rationals, there is an X which keeps the elements of automorphism of fixed, and which maps, in an order-preserving way, the sequence xi onto zi. And such an automorphism is easily given by any map which fixes the elements of itself, which maps the X into the xi yj. j (x)(3,y)(Axy). Then by indistinguish- Now supposeB it; is clear that for ability, and maps the rationals 1-1 onto some yX, for otherwise we would have hence for Axz So 8 Q3 x 2 , xl x1 , x0 E Q. a(k satisfies axiom in the domain. of 0 ~ = S satisfies ZF*. (x2 )(x2 x 0 - x 2 ex 1), Then either - x 2 cx 1 ), and so since In the second case, we have 1/2(x0 + x1 ) and x,yEQ, xI and xO EX, or In the first case, we can conclude that (x2 )(x2 EXO xO = x 1. for C. To verify 1 of ZF*, suppose for for Axy by indistinguishability, assuming xQ. does not satisfy Clearly Axy But now I claim that z = y + 1. since yX, A(x+l,y), C we have xEQ, x1 . But then x0 = x 1 . 1=ZF, we have 1/2(x 0 + x 1 ) < xO iff 43 To verify 2 of ZF*, set 1d3 =0 see that x U Also, x) al from EX0 , since to . (3 in 6 remains unchanged for x axiom 2 of easy 0 xeX, f. in when we pass a. of x0 are the . Also, the subsets in at in ZF* . Putting this together, we see that is satisfied in "in the same way" as it E 4 , To see that 3 of ZF* is satisfied by satisfies CA. suppose Then the unique element defined in CA by indistinguishability. We let this element be we are looking for is a power set of claim that y = P(x) in the model have to show that (2 , in subsets of x x. . We does the trick. We But this is in x What are the members in 3 are the . in ZF* is checked similarly. To see that axiom 6 of If y must be in the model ( satisfies y = P(x). and the only subsets of Axiom 4 of xeX U Q. x aL obvious, since the only members of y to . is in of is or any of its members, are identical with the corres- X0 , ponding elements in ( eX, It is same as . Also, the members in same as the members of of x0 = w of xeX, take ZF* xO is satisfied in g, let in foundation as an R-least memberof x in ® . If xeQ, take x0 to be , in aL (or 2). Obviously I2j p e x. Axiom 5 of CA defines in ZF* 63 is the most complicated. is again interested in the range in 6 e X. Call it D. The set that Then we are of the partial function in , x3 =x 1 ), on the domain Bx2 x 1 & (x3 )(Bx2 x3 -> X2 with if S(x2 ,D), S(x2 ,D) X Bx2 x & (x3 )(Bx2 x 3 -> x3 =x1 ), then x1 EX. and , Cx2xl Tl holds in Replacement in the model Replacement , in 6 , such that, for Cx2xl holds in . x2 EX, Then by we would have (this instance of) It remains to show with the free variables shown, Axl...x ,n al which holds in Bx1 ...x n there is a formula , iff Cx2 xl and we would be done. that for each formula holds in Now every x2 EX, and so, by indistinguishability, has Now suppose there is a formula xl D. iff Ax1 .. . x n xi X. when It suffices to prove by induction that for any formula AXl...n, into Q, there is a formula Bxil ... xik with xiEX iff A 1x...n To see this XiExj, Axl...x iDom(f), instances of by n (3v)(v=v) Put quantifier-free, take JEDom(f), by or by if Axl...xn f(i) if to 2) replacing all xi = xi, or xi=xj, (3v)(v=v) B by 1) replacing all instances x =xi, iEDom(f), J 3) replacing all instances of xjxj, i,JcDom(f), or xiExj, n =Bxil...Xik. iff be the formula obtained from A of k iEDom(f), we have if f(i) i Dom(f), xi for i til<i<n) such that for any sequence x...x Dom(f)1, fill< i < n & i 3 from and for any partial function f xi ex , j (3v)(v=v) xi=xJ, (3v)(v=v) f(i) < f(J), f(J), or Dom(f), if not; if not, respectively. in prenex form, and suppose our claim is true for all formulae with less quantifiers. We may assume that Axlx. . . n is (xO)CxOxl ., .xn since the existential case follows from this case by taking 45 negations. into Q. Let f fill be a partial function from Now 2 finite partial functions g,h i < n) w -> from Q are said to be of the same type if 1) they have the same D, 2) g(x) < g(y) domain iff h(x) < h(y). Consider the set of partial functions on which are identical to a finite # f on ill i n. f. = of types represented in this set. Pick a repre- Let Di, 1 i inductive hypothesis for k, X 1 n, = fi (l) and for any for Cx0 xl...xn xO...xn {PlP Then we take B indistinguishability, q) = to be fi; for I for k}, {r o Dom(fi), (13 < i.e., each Di xeX with 1eDom(fi), we have -Xp DiXpl. ffl,f2,,f. be the formula given by the has exactly the free variables x O .• < i < n) There are only sentative from each type, and call this set fl filO for CxOxl...xn r n & r (xo)D1 & lDom(fi), ~ iff Dom(fi ) } . Di. By it is easily seen that B and A satisfies the conclusion of our claim for the function f. This concludes the proof of Theorem 2. We now define some new subsystems of the same as ZF except for the Replacement instead, only allows the to have n + 2 other ones. ZF. Axy ZF n schema. is to be ZF n, in the Replacement schema at most free variables; x It is easy to see that and ZF* y and possibly n is a subsystem of ZF© . We also have Corollary 2: Proof: ZF* ZF 0 . Consider the model of ZF* constructed in the proof of Theorem 2. (ly & (z)(z0 -> -> ZFO provable in . (x)(3y)(0ex Consider the sentence A (zEx <-> zy)))). But our model of We believe strongly that stronger conjecture that ZFO the details of a proof of ZF* does not satisfy A. ZF, and in fact in the ZFn+ l, ZFi ZFO ZF A is obviously for all n. Although have not yet been carried out, we can give the definition of a very promising model of ZFO+ (3 x)(y)(3 z)(--(zy <-We let M(a) z=x)). be the minimal model of ZF. any Cohen generic set of natural numbers over MS (a) be the corresponding Cohen model for We let FS We let M(a). ZF + V S be We let L. be the set of all sets of natural numbers which are finitely different from difference from S Recall that S (i.e., whose symmetric is finite). R(I+1) Consider the sets = U IP(R(1)), R(X) = xeMS(a) R(1). such that for any formula Ayzx1 , we have, in let X be the set of all such x. xeX such that any finite combination of union and power set on x in 1MS(a),[zl(3y)(yex& A(x,y,FS))) f MS (a) gives a set in For each , let lative hierarchy up to Define, for each R'(a) n Y, fg (z) R' () 5 We let We define fX)= be the set of all be the unique rank in the cumu- a function f and by the equations f+l(x) y), MS (a). whose domain is {YfyeY & (3 z)(zex & y U fx xcZ as We X. in the model 5, Y xl. [(3B3)(<a d x Range(fg )) & xEY]. 47 We conjecture that <Z,e> is the desired model. We feel that a detailed analysis of the relations between the theories Z distinguish each n, hopefully by finding natural sentences to ZFn from ZF n+ , would involve non-trivial applications of the notion of forcing. In particular, careful attention seems to be required as to the model-theoretic properties of models of theories obtained by forcing; e.g., the definability or indistinguishability of various elements of the models constructed. CHAPTER II 1. Definitions of Systems. We will have for our language, number variables, n,m,p,q,r; set variables, x,y,z,w,u,v,...; the relation nex between numbers and sets; and the constant number "O" which can hold only between variables of the same type. We introduce function variables in the usual way by defining them in terms of sets of natural numbers in the usual We will have full number theory at our disposal, since way. all systems considered here will have the unrestricted induction axiom schema (called I) where A [AO & (n)(An -> An')] --> (n)An, is any formula with possibly free variables of both kinds, and can have both number and set quantification. Also, all systems considered here will have the axiom of extensionality (that any two sets with the same members are equal), and we will tacitly assume that the system we will call I includes this axiom. Thus I will be the unrestricted axiom schema of induction plus the axiom of extensionality. In addition, all systems considered will have the recursively enumerable comprehension axiom schema (called ReCA), 3 y)(n)(ney- (3k)T-e,n,k,xl, .. . ,) ) 1 (x1)... (xl)(e)( The predicates Ti are understood to be written out in the usual way with bounded quantifiers. (Or, we may have instead introduced them as primitive, and defined them by adding axioms of primitive recursion.) The predicates TA above are what Kleene would call the characteristic T above are what Kleene would call T,f where , function for the set, x i. where fi See Kleene is [3], p. 291. Before we get into the mathematics of the systems we will be considering, we will state, informally, some propositions concerning what can be done in the system Proposition 1: We can, in I + ReCA, I + ReCA. Justify all uses of coding normally found in the development of hierarchy theory. Thus, we may Justify the use of such symbols as (respectively (2n + l * (X)n where pn 3 m+llnex & mey], x,y> and and {mlpm+lex], is the nth prime). We define the relation between functions and sets alluded to above, as l 2 n+ f(n) = m iff n,m> f, where n,m> = 3m+l. Proposition 2: In ReCA + I, we may provably perform "collapse of like quantifiers". In other words, the usual way of collapsing 2 successive universal set quantifiers (or function quantifiers) into one can be completely Justified on the basis of ReCA + I. Def. 1: essentially A predicate A(n,xl, . .,. xp ) of n is said to be 1TI if it is in prenex form followed by a matrix T(e,n,m,xl,...,P,fl,...,f ,nl,...,nk), where there are no set quantifiers in A, and the fi occur as universal function quantifiers, in any order, mixed together with possibly number quantifiers (ni) and function quantifiers in (3 nj) A.) and (3 m). (No existential An essentially 4 formula is Just the prenex form of the negation of an essentially T11 5o formula. Proposition 3: For every arithmetical predicate (with (f)(3 m)T(e,n,m,xl,...,x 2, parameters) there is a predicate nl, . . .,n p ) which is, provably in I + ReCA, equivalent. ) There is also a predicate (3f)(m),T(i,n,m,xl...,x,·nl,...,np which is, provably in Proposition 4: y = (x)( xi+l ) - 3 I + ReCA, equivalent. as the (x0 ,...,x,) a sequence = Jump of xi . ( ) (x) Define such that (x)(1) (Thus, y -th Jump of x x. = x and is a predicate of 3 Then in variables, defined in the usual way.) I + ReCA we = (x)()). may prove (3!y)(u This is proved by induction on 2. Consider L) (x)(p)(e) (n) (f ) (3m)T2(e,n,m,f,x) (3g)(r)~-T2(p,n,r,g,x)I -- > (3y)(n)(ney (f)(3m)T2(e,n,m,f,x))}. 2) The schema, where A (x)((n)[Anx (x)(e)((n)( 3 f)(m)-T 3) 1, is essentially 2 (3 y)(n)(ny- Bnx] -> B Anx)}, is essentially I1 l(e,O,m,n,f,x) -> (3 y)(n)(y)n is a function& (m)~T 2 1 (e,0O,m,n,y, x ))) (x)((n)( 3 f)A(n,f,x) -> 4) A(n,yn,x))), where A 3 g)(m)T (x)(e)((f)( 5) (m)-T 3 (e,,m,y n is essentially Z 1. 3 (e,0,m,f,g,x) -> where A (f)(3 y)(n)(yo = f & yn+ l ,x) ) ). (x)((f)(3g)A(f,g,x)-> 6) (3y)(n)((Yn) is a function & is essentially 2 1 (f)(3y)(n)(yo = f & A(Yn·Yn+lX))) having only the free variables f,g,x. We call I + ReCA + 1), the pure We call I + ReCA + 1), the pure A; II + ReCA + 2), A1-CA; + ReCA + 2), I 51 essentially E1 -AC; I + ReCA + 4), I + ReCA + 3), pure A-CA; essentially 41-AC; I + ReCA + 5), pure :1 -DC; I + ReCA + 6), essentially 2 -DC. NOTE: "CA" is supposed to mean "comprehension axiom"; "AC", axiom of choice; "DC", dependent choices. Proposition 5: In pure predicate, there is a pure 21-AC, for every essentially 1 predicate 11 (f)(3n)T(,m,n,...) which is provably equivalent. To see this, first use Proposition 2 to collapse adjacent like quantifiers. Then use pure £1 -AC to interchange a number and function quantifier, and then use Proposition 2 and then pure Z1-AC, etc. Proposition 6: Pure 21 -AC D pure The idea is that, in pure m Al-CA, there is a solution to one of two n1 uses pure Z1 -AC set in pure A-CA 1 in the 1-CA. one has that for each predicates, and one to form a Skolem function. Then the required is obtained, in pure Z 1 -AC, recursively ump of the Skolem function. Proposition 7: Pure £1 -AC = essentially £1 -AC. By Proposition 5, one has only to consider the case (n)(3 f)(3 g)(m)fT--> 3 the 1 £1 -AC f and g Skolem function. But we may collapse quantifiers in the usual way, and apply pure to get a Skolem function, which will have recursive in it the Skolem function wanted in the implication above. A similar argument shows Proposition 8: Pure 2 1-AC D essentially i-CA. 52 4-CA Conjecture: We do not know whether pure A-CA or whether pure essentially A-CA, Z1 -AC - essentially = - Z-AC, or whether AI-CA, and we conjecture that none of the three statements is correct. Proposition 9: call essentially Pure 21 -DC -> 1_AC, and pure 2Z-AC. (By Prop. 5, we 2:-AC, just 1 -AC.) £1 -DC - pure 1 -DC. Hence also Proposition 10: Essentially Proposition 11: We can prove in ReCA + I the recursion theorem (with parameters). 2. Preliminary Letmmas. We will eventually show that Z1 -DC is a conservative extension for purely 21 sentences of pure hi-CA. We define, in I + ReCA, several notions. Def. 1: P(n,m) is defined as the CGdel number (in the usual 1-1 onto G8del numbering of pairs of natural numbers) of the pair <n,m>. Def. 2: epX is defined in the usual way as the nth partial re- cursive function in x. Note that in I + ReCA we can prove (n)(x)(3 c) Def. 3: C {0,l' & (m) RLOX(n) is defined as "Range (cpX) (meDom(ep)) & Cpx defines a linear ordering," where o defines a linear ordering means that 1) (m)(cpX(P(m,m))= 0), 2) (p)(q)(r) (Cppx(P(p,q)) . 1 & PX(P(q,r)) - 1) -> 3) (p)(q)(qnx(P(p,q)) - 1 -> - 1)). ,x (P(p,r)) cpX(P(q,p)) - 0. 4) (p)(q)(p = 1 v cp(P(q,p))= 1). We define Xcp(P(p,q)) = q v RLO(n) =- RLO0(n). 53 Def. 4: define We define n q = RLOX(n) & (cpn(P(pq)) p < n x = , q = RLOX(n) & n(P(p,q)) < we use the subscript n When 1 v p = q). n . instead of We 1. denotes the empty set.) We define WX(n) Suc (pq) = n RLOX(n) & q < p onx(P) .SUx (p))- - n n HY n n n condition ye (P(r,s) s < xP & re(z)s = ) n Y= ). Thus H x(Z) n z n xq )· & xP n ((Z)p)(l) n (x)p r n (q)(p RLOX(n) & 4 variables asserting that starting from (r)(r < xP -> RLO(nX) & (p) (m)(Suc nx (m,) (Z) (P) -> Def. 8: v (mey & Lim x( ) = RLOX(n) & (3q)(q < (z) m ) & (p) (lim x(P) -> (x)p n (p)(o & n Suc x(p) = (3 q)Suc n(p,q). De . 7: & (y)(3 m)(y= m < xr))). n (r) (rey -> Def. 6: RLOX(n) as = & is the predicate of is a hierarchy on the RLO, n, It will be useful later on to include the (3p) (kE(z)p)), in the definition above. (k)(kez -> is the predicate of 5 variables asserting that HYx( ) "I z is a hierarchy on the starting from Def. 9: x y y. RLO, nx Thus if p is defined as n x, uR to but not including then (z)p= 0. "x is recursive in y" in the usual way. Def. 10: Inx < ImXI proper imbedding of means nx into RLOX (n) & RLOx (m ) & mX, i.e., < (p) (q) { (p nxq is a -- > I 54 f(p) m xf(q)) & ([p<xq = In X I < ImxI = InXl = ImXI Def. 11: (f)] -> p E Range means that the range of the InX y & q m mXI v InX W. is (HX (z)) & & HZ (x) & (x)(y)(z)([WZ(n) n I + ReCA. Also, provable in y = x HZ nz (y)] -> y=x) is - (y)] [W' Z(n) & HZ_(x) & HZz np np is also provable. Proof: element n,x,y,z with I + ReCA. By Given (P I W p74(Y)P q. then, since ()q above TZ) Lemma 1: (Y)q Range (f))1. < ImX . (3n)(3z)(RLOX(n) & HypX (y) f E in WZ(n), HZz(y). n WX(n), Form we have a < -least n Clearly we must not have (s)(s < If Suc x(qs), n n 0 x(q). n lim x(q), If n xq -> (X) s = (Y)s), we have then since (X)S = (Y)S (X)q we have = (Y)q. The following is provable in pure Lemma 2: wZ(n) & wZ(m) & three holds: a) A1-CA: if (3x) (4 x)).* then exactly one of the following there is a unique imbedding of the ordering nz onto a proper initial segment of the ordering mz and 55 there is no imbedding of the ording ment of the ordering mZ onto an initial seg- . nz there is a unique imbedding of the ordering n z b) whole ordering m, onto the and vice versa, and there are no imbeddings onto proper initial segments in either direction. c) the interchanging of n with Furthermore, denote x(1) m and ( ). •fx for some I In a), if there is an imbedding, a proper initial segment of g Then all the depending on n (p), Suc z(p) n and of n onto I + ReCA, (klf(k) g(k)), and Running through the 3 cases, p. in a straightforward way, lim z(p) n f, then it must be unique, since m in is another, form, take an n-least member, O e w. x, m. Proof: if a). for the result of applying iterations of the Jump operator to above maps may be found in n using the linearity of n and m, we get a contradiction. Same with c). of If there is an imbedding, g, segment of with f n, take -1 g 1, m and note that onto an initial -l g 1 must disagree somewhere. Then follow the procedure above. If there is an imbedding of then there is one from m to that we can form inverses by n onto the whole ordering m, n by taking inverse. ReCA.) (Note And there are no proper imbeddings, using least counterexample argument above. So the main thing is the existence of these imbeddings. We define a function f z(p) greatest k such that 56 (3q)(Suck z(p,q)), where Suck (This is well defined, since is k iterations of (p)(q)(r)(Suc n -> r = q)). There is always a greatest Suc. z(p,r) & Suc z(p,q) ] n k, since otherwise we would have an arithmetically defined chain through n z , and finitely many iterations of ReCA (+ I) would realize this claim as an object, and would contradict WZ(n). Fix n,m, with WZ(n), WZ(m), (x) (Zz (x)). Let f (p) be n n abbreviated f(p). We claim that, for each p, either there is an imbedding of the ordering the ordering n, so that m onto an initial segment S keS---> k < p, of or there is an n imbedding of kJk < zp) of n, onto an initial segment of m, and these imbeddings are to be found < T(x)p(l(f(P)+l)). (This is the result of applying 10(f(p)+l) iterations of the Jumpoperator to (x)p.) Weassumethis is false,and, as usual, form plclaim is false}. We take an n-least clearly claim is true for f(r) + 1 = f(q). member, q. If q Suc If (q), then (q,r), then certainly So there exists a comparison mapping, < T()r(lO(f(r)+l)), between the segment of m. 0 n up to r, and It is then easy to see that there is a (unique) comparison map between the segment of n up to q, and m, IT(x(lO(f(r)+l)) The limit case is similarly easy; the uniqueness of these compar- ison maps is used heavily, and that the property of being a comparison map is low arithmetical, and that the function f is low arithmetical. 57 Lemma : az nd WY(n) The following is provable WY (m), InY I < lmy I, and (3z) (z < T(x)p (o (f(f)+l)) & HY (z)), mY in n, is in I + ReCA: an d If (3x) (HY (x)), nY wh ere p of the range of the imbedding of then is the l.u.b. in and f n, defined in the proof of Lemma 2. f y Proof: Similar to the proof of Lemma 2. Hy that if One notes, e.g., has a solution, it must be unique. np If Lemma 4: of the image of the nZ and p is the l.u.b. in then imbedding, and (3y)(HZz(y)), m < InZI, m where f is f z (3x)(H z (x) & y < T() (l(f(P)+l))), n np is provable in Proof: I + ReCA. One can prove first, as in Lemma 2, that (3x) (H (x)). Then, again like Le mna with one can prove that HZZ(x), This 2, starting with this y < T(x)(lO(f()+l)) x (0: np course, it is provable in ReCA + (3x)(HZz(x) -> [ that np (3!x) HZ (x)).) np Reas (nx ) Def. 12: (3q) Suck n. (q,P)-> X (n) & (p) (Onx(P) v Lim x(p) v (3k) RLO x(P,q) & limn (q) ) & (q)(q < xp )] . (3r -)(O (r)) &C(P ) (Reas (n) reads Lemma 5:(z)(n)(x)(y)(m)([Reas(mZ) "n & Hzz n W Z (m) -> (3 p)(x < T( y is provable in )p I + ReCA. (3q) (Suc is reasonable.") (x) &HZz(y)& m & there are infinitely many q >mzp)), 58 Assume Reas(mz), ffZ (x), H z (y), WZ (m). Then n zleast member there is a set, w, which has no mZ-least member.It is Proof: then easy to see, using I + ReCA, that (3g)(k)(g(k+l) ( g(k)). n (p) (k)((x)p We will show that T (Y)g(k~ BY Reas(mZ), this is sufficient. For, we take, in I + ReCA, pl(3k)(~((x)p < T(Y)g(k)))•. Wecan do this, since the predicate we are taking the extension of is arithmetical in call it 0 1) q, by x,y. WZ(n). And we take an clearly We then obtain a contradiction. z T(Y)g(k)) (k)(z T( Y )g (k ) ' If lim If (g(k)), then by Reas(mz), we have r,s with () s(r) Suc lim(s), and so and so80 z T()g(k) (q,r). (x)(l) T(Y)g(k1), z T(Y)s (X)r < T ( Then all k > 0, )g Sucz K T(Y)q (q ) , use The case when If (k)' (q) some Sucrz(g(k),s), all and hence lime then So clearly k for the following general reason: (y)(l) z(g(k)), 0m (g(k)), also easy. Suc all such p, z (q). To get a contradiction for this case, it suffices to show that 2) n-least if k. (Y)g(k) But then (X)q T(y)g(k), p < zq, then is easy. When Reas(mz) as in Case 1) above. m 3) lim > (X)1 a set is (q). Let T(Y)g(k+10) (x)1 , k be arbitrary. We know () ) - ( < zq Furthermore, the question of whether is a question low arithmetical in the uniqueness of Hierarchies on RLO's, , due to and even hierarchies 59 on initial segments of in (Y)g(k+lO)' RLO's. and hence Hence (x)q is low arithmetical (x)q will re recursive in mZ, g(k+l10)' Hence, by reasonableness of we have (,X)q < T(Y)g(k). Lemma 6: "The predicate WX(n) Proof: We have, in I1 for predicates in ReCA + I, x, is the Kleene normal form (3 f)(m)4T(e,n,m,f,x)). Now there is an explicit recursive function 3 WX(g(k,n))]. Hence g, for which we can (k)(n)E(f)(3m)T(k,n,m,f,x)- ReCA + I, that B) such that (n)(f)(3m)T(n,O,m,f,x) - (3 f)(m)~'.T(e,n,f,x)), is provable in (f)(3 m)T(m,O,m,f,x) of predicate x" and we use that to formalize the So assume A) (n)(WX(n) prove in in I + ReCA. provable in Lemma. E1 is not I + ReCA, n is 1 since the in x, by A) and B), and Kleene's normal form Theorem is provable. Now substitute 3. e for to get a contradiction. Conservative Extension Result. Theorem 1: S n is any purely E 1 -DC + S where Proof: pretation J Given any model k S of pure A1 -CA + S, sentence, there exists a model M' is (3 y)(g)(3n)T(g,O,n,y,z), A model of pure A1 -CA + S X, J,+,x,O,'; 2) having same +,x,O,, and a binary relation We define a new model HypY(M) = M', R(J,x) where of for some e. consists of an inter- of the natural numbers, and with a set of objects, XEX. M together R(j,x), JEJ, as 1) having same but restricted to those 60 JEJ, xX that with Hyp(x). M Now since M S, choose y (g)(3 n)T(i,O,n,y,g). Then clearly M' M (e,O,n,y,g). Mt So We claim M' (g)(3 n)T S. = E 1-DC. It is sufficient to show that the relativized of each axiom of a theorem of pure such A 1 -CA. X-1DC to the predicate Hyp is (We define the relativized, T(A), of a formula A, to the predicate HypY(x) by T(A) v T(B), T(A & B) = T(A) & T(B), T(A) -= T(A),T(3xA)= T(A v B) = (3 x)(T(A) & ypY(x)), T((x)A) = (x)(HypY(x) -> Q, Q quantifier free, T(A)), T(Q) = T(3nA) = (3n)T(A), T((n)A) = (n)(T(A))). In other words, the universal closure, obtained by inserting the universal quantifier axiom of 1) (y), of the relativized of each -1_DC to HypY is a Theoremof pure Induction. It is instance of induction to clear HypY ion, and so is provable in pure 2) ReCA. To see this, let ny first 10'k l . l1 -CA. integers it suffices (x)(HypY (x) -> Hypy (x ( l i, So set Define a new ordering x() ) )). < T(1) k = f y(i) + 2. If mx by adding on the on top of n, and make,in the trivial way, the ordering my Imy relativized of each is again an instance of induct- Hy (z), W Y(n), x < T has a greatest element, not, set k = 2. the To prove the relativized of ReCA, to prove, in pure A1-CA, that If that A1-CA. total, so that RLOY (m) and Iny l < Now clearly by Lemma 4, (3w)(Hmy(w)), and so clearly (3w)(HY y(w)), m by I + ReCA. But, also by Lemma 4, z(1) < Tw . Hence 3) HypY(z(l)). Suppose (f)(HpY(f) -> (e,O,m,f,g,x))), where the predicate nO . . . ,n k n l Pn n (3g)(HypY(g)& (m)~T 3 HypY(x). Let HypY(f). We consider is a member of a finite sequence k 2 0, such that I) (i)WY(ni). II) sequence of sets xO ,xl,...,x j , J k, For any such that (i)(HY (xi ) ) , ny i 3 we have that and z0 = f another sequence and (i < J)(m)T 3 such that each (e,O,m,z ,X). i+ 1 J < k, sequence xo,xl,...,x, zi with zirx i , III) For any (i < J)(RHY(xi)), it is (z) & p has infinitely (3 z)(3p)[Hy (nj+1 )p Zi ~~T~~i, ¶1+1• TZmany q > many q > y p & 3Zozl,S, zz j j+ l TXi' Zj+1 TZ' i n.+1 not the case that )(m)T3(eO (s < with ,m,ZsZs+l,1x)]. Also consider Qn n O ,.. . , n such that 3 sequence of sets with I) (i)HYy(xi). II) 3 a sequence z i that each zi < Txi III) -1 Let is a member of a finite sequence O, Reas(ny ), , k xOXl,...,Xk n zo = f and and (i)(m)-T3 (e,O,m,zi,zi+ 1 ,x • j < k. It is not the case that Et • TXJ+l & HY (z) & p such (3 z)(3 p) has infinitely many q > (n3 , )y & 3 i_ TXi' ,Zj+l, ,Zl,...,z j ZJ+1 TZ' with (S < )(m) ~,T 3 (e ,O,m,z s ,zs+1 ,x) ] . We first note that that Qn is a 21 Pn is a predicate in We wish to show, in TI predicate in f,x,y f,x,y. I + ReCA, that (n)(Pn - Qn). and ) · 62 Suppose Pn. Assume that nO,nl,...,nk have the properties mentioned in the definition of (i)(3 xi)HYy (xi). take i For, if for some i, '(3xi)HYy (xi), then to be least such (induction), and then by (i)WY(ni) (q)([WY(q) & (3 z)(Hy (z))] -> we have that qY and, in fact, qY is imbedded into infinitely many points must be Pn. We claim that • T some w p > with itely many points s > yr. condition III) of Pn negative.) ni yS. niy H J = i-l. when Pn, z having with (w), where r (We allow For Just apply condition 2) of II) of i with 1.u.b. Hence the y < InYl), Hy (z) has infin- We claim that this violates (f)(3 g)(m)-uT relativized to condition ny qY HypY , use that Pn, and that J to be and that Hypy (f). In (s)(s < i -> (3 xi)HY(xi)), n½ and use the x i. So we want to show with of HYy(xi). II) in i Pn. III) in Qn Qn, Qn and we may choose XO,xl,...,xk follows from 2) in the definition follows from III) in <T plus the have unique solutions observation that the predicates HY (n Pn p in (X)p. So Pn -- Suppose and taking i > Qn. Qn. We first to show least with WY (ni), (i)WY(ni). we contradict in the same way as the argument above for one uses II) of Qn, Pn -> Using III) Qn. Lemma 5, of Qn For, and the relativization of (f)(3g)(m)~T. 63 (i)WY(ni). So Pn Now II) of ni of hierarchies on Pn III) of follows from II) of because of Qn by uniqueness WY(ni). follows from III) of for the same reason. Qn Next, we show that we can eliminate the parameters x Pn in and Qn, 1 respectively, let el, e2 have f and and have the same meaning and still be, and For, since Hypy E1. WY(el) (x ) Hypy and , TZ WY(e2 ) & (3z)(HYy(z) & x and (f ) e1 with Gdel number k) & (3w)(HYy(w) & f T with G8del 2 number 1). P'n Then take HY (w) & x < T Gdel with (x)(f)(w)(z)(HYy(z) & to be k & f number with G'del Tw ey P(n,f,x)). Take -> number 1 (Q(n,x,f) & Hy (w) & x number with Gdel ). Tz Q'n Gcdel number with el,e2, k , Here, (3x)(3f)(3w)(3z) to be and k & f T are constants, 2 not variables. Now, Q'n Q'n - P'n. clearly So define (m)(Rm -> (3p)(WY(p) & InYlJ of pY 1. is IpYl). up to Rp). Also, Rm - (3n)(Q'n & ImYl WY (x)), and Fix such a We wish to form R p. So InYl). By So l. So by Lemma 6, WY(p) & (n)(Q'n -> A -CA, Call this set W Y(p), let pO segment since this is [<r,s>P's & ~.(thesegment of rl(Bs)(<r,s> e S!. since [<r,s>lQ's & Isl ithe rj<jls)1, form this (these) set(s). T = is r ). We can, using pure also equivalent to WY(n)), (n)('n-> pY S. be the up to Let < y-least py upper bound of this set. Let pY segment of bound for Inyl < T, qY|). ment of q q have WV(q), up to and including po1. take q = p. I1). Hence, clearly q the If there is no upper In any case, clearly Also, clearly ()(3n)(Q'n up to with (n)(Q'n -> InYl 2 -> the seg- (j)(3 z)(Ky(z)). (3 w)(Hy (w)). The only hard case is when We wish to show q qY has no greatest point. But we can form, in pure {<,J>l(3z)(Hq (z) & Jez)), ition, has the 1 1 I 1 defin- since it, besides this definition A1-CA, (<,J>j(z)(H (z) -> jez)), I and we may, arithmetically in this set, get a w with Hy (w). qy Finally, we claim that the sequence (not necessarily unique) needed to verify the relativized z1 -DC Hyp y, can to HYy (W), and hence would qY have what we wanted all along, namely satisfying HypY . To see be defined arithmetically in w with this, it suffices to show that any finite sequence n,...,nk satisfying the conditions in the definition of extended to nl,nl,...,nk+l, satisfying Pn, Pn and also that there are sequences satisfying the definition of with. Take To see the latter, let (k|f a(Z)k, and let k ey ISYI = Ithe segment of f < Tz with be a l.u.b. up to k can be Pn to begin H Yy(z), W(e). Take s with with an appropriate finite number of points added on top so as to make, by the Lemmas, (3 w)(H y(w) & z s = e. Take nO = s. Tw ) . If there is no l.u.b., take Then the sequence <no> satisfies the 65 the definition of Pn. The same trick allows one to see how to extend a sequence nO, n l' , .*,. nO n l ,.. nk l nk + (f)(3 g)(m)~T satisfying Pn satisfying Pn, 2-AC. <nO,nl,...,nk> satis- Qn. Independence Result. independent of to an for one uses the fact that relativizes, and use that fies (the definition of) 4. nk 21-DC We wish to show here that 1 i-AC It suffices to show that is I £-DC. We do this by showing that, for a suitably chosen sentence S with Con(S + -DC), we can prove in Con(S + 2:-AC). For then, if proves -DC), S + Con(S + 1-_DC S + 4i-DC 1 -DC that then and hence by Gdel's S+ DC theorem, would be inconsistent. Lemma 1: If S is any sentence in prenex form starting with universal number and set quantifiers from left, followed by existential number and set quantifiers, followed by a matrix Pr(X1-DC + S containing only number quantifiers, then (coding into natural numbers) Proof: w-model satisfying S"). Collapse all the universal number quantifiers in the left part of quantifiers S, and collapse all the existential number on the right part of S (before the matrix) that S becomes (3 m) into the matrix, to get S' = (n)(x)( 3 m)( 3 y)A(n,x,m,y). can talk of ( (z)(3 w)C(z,w), being a closure of i.e., LC Dom(4B) & C(z,w))). so Then push (n)(x)( 3 y)B(n,x,y). Now in set theory, given two w-models (w a 3 at ML and , we under the sentence and (z)(z Dom(fl) -> (3w) 66 Instead of talkingof real w-models, talk of codings of them into a set of natural numbersin a natural (arithmetical) Then we can prove in way. the n (x)(3 y)B(,x,y), k < n. sentences This depends (n)(x)(3 y)B(n,x,y) is provable in heavily on that £1-DC + S. Now, consider the arithmetical predicate D(z,w) where y is a (coding of) pair (k,x), w a (coding of) pair (x)(3 y)B(,x,y), D, p < k+l." "z is (k+l,y) x under the ow applying k Z1-DC sentences to this predi- we obtain a sequence of dependent choices in which the desired w-model of S can be obtained recursively. S Now suppose we found a true sentence Lemma 1, = is a (coding of) finite w-model which is a closure of the (coding of) finite w-model cate and for n a finite closure of this w-model under -model, 3 any finite for any 2z-DC + S that such I + S that £1 -AC. (remember all w-models provably satisfy consistent, since Then we would be would be provably consistent in I + S done, since proves in the form for S I) and ZI-DC + S i1 -DC + S is is true. By well-known techniques, ReCA is finitely axiomatizable, since we need only consider the case of two parameters. We take S to be the conjunction of this finite axiomatization with the sentence (3y)(X (y))). Then (x)(n)(Wx(n) -> S clearly of the proper form. We define (y) -> (W*)X(n) = RLOX(n)& (y)(HypX nX-least element of We know that y, S + I provided y ReCA + I, 3 an 0). r so we can use Lemmas of the previous chapter about provability in ReCA + I. is 67 Lemma 2: n with The following is provable in WX(n), nx Proof: in (W*)X(m) & ~WX (m ), onto an initial segment of that if WX(p) is an mX. S guarantees that WX(n). (n)(3 f)(m)4T(e,O,m,f,n,x). The predicate of (3z)I x(z)) & ReasX(p) is with But ust on the basis of Now suppose p, f(m) (W*)X(n) & (3z)(Hxx(z)), then the conclusion of the Lemma is true. n we have P(w) We can prove, like Lemma 2 of the previous section, I + ReCA, (3z)(Hxx(z)) for every there is a coding of a function f:w -> such that whenever imbedding of S + I: satisfies it, 3 £1, k and so, since every such that (3z)(HxX(z)) & Reasx(k) & Wx (k.). Then by Lemma5, (z)(Hx',(z) -> Clearly we have (n)( 3 f)( 3 z )[l x (z) p Hyp (z)). & (m)"hT(e,O,m,f,n,x)3. Collapsing the quantifiers in the usual way, and putting the (n)(3 h)(m) result in Kleene Normal Form, we end up with ~T(q,O,h,n,x). Furthermore, any Skolem function for this sentence would have, recursive in it, a Skolem function for (n)(3 f)(m)'T(e,O,m,f,n,x). that So it remains to show, in S + I, (n)(3h)(m)~T(q,O,m,h,n,x) has a Skolem function. Now,there is a standard recursive function F (n)(RLOX(F(n)) & any set for which there is no element, has recursive in it a solution & any solution h is no h such that F(n)X-least to (m)~.T(q,O,m,h,n,x) has recursive in it a set for which there F(n)X-least element). So Now consider the (n)((W*)X(F(n)) & ~wX(F(n))). El-predicate Pk a coding of a function from w into = RLOX(k) & (3f)(n)(f P(w) such that f(n) is 68 an imbedding of this into F(n)X). predicate holds of all 1 for some f kx r . Let X g(n) g n, f(n) WX(k), Then we have a function is an imbedding of Then recursively in X, f, ---> P(w), of a function rx into we can find a such that, for each n, is a non-empty set for which there is no element. and so holds be any non-empty set for which there is no rX-least member. coding with WX(r), by Lemma 6. with such that for each F(n)X k By Lemma 2 of this section, F(n)X-least Hence, by the special property of the recursive function F, we may obtain the desired Skolem function for (n)(3 h)(m)T(q,O,m,h,n,x) recursively in Thus we have verified that I + S g. proves E1 -AC, and we immediately have Theorem 2: 5. 1i'-DC is independent of £1 -AC. Relation Between Predicative and Ivyperarithmetic Analysis. £1-A C (or i-CA 1 or E-DC) are considered reasonable formulations of so-called hyperarithmetic analysis, in view of the fact that they are natural systems whose minimum w-model consists of exactly the hyperarithmetic sets of natural numbers. In this section we compare 1 -AC with a system T which represents the formalization of a small part of predicative analysis (see Feferman, E The system (y)(3x)( is the Gdel T is (x)), where ). I + ReCA + the infinite list of axioms n varies, and k is fixed, and number of a natural well-ordering of type k cO. 69 (Thus the infinite list of sentences is obtained by changing n). We will show that T £1 -AC with respect to all purely is a conservative extension of N2 sentences. (ReCA is formalized as one sentence, in the standard way, i.e., using only 2 parameters xO x1 .) It is easily seen by well-known techniques that the proof of conservative extension can be made finitary. So we obtain a finitary proof of Con(ZE1-AC) relative to Con(T). Finitary generally means here, in PRA (primitive recursive arithmetic). A widely used index of complexity of an axiomatic theory is how large is the least upper bound of its provable ordinals. a In the present context, an ordinal is said to be a provable ordinal of a given fixed theory, if there is an n with RLO(n) and n has order type a and the theory proves W(n). In T1 that we will prove, view of the conservative extension for it is clear that the provable ordinals of the provable ordinals of 1 -AC are exactly T. It follows from work of Feferman [l) and Tait least upper bound on the provable ordinals of T represented by the so-th critical function at 0. l], pp. 14-16). 5] that the is the ordinal (See Feferman Furthermore, it also follows from their work that in PRA + rule of primitive recursive induction on the natural RLO corresponding to the £0 -th critical function at O, a consistency proof may be given for T. So, by our results, such a consistency proof may be given for By an instance of induction on kn 1-AC. we mean a statementof 70 the form [A(q) & (p)(tp < > (p)(p < i n-> n & (m)(m< Ap), where q p -> Am)] -> has Ok(q), and Ap)P is any A formula (in the language of Analysis, described at the beginning of this chapter), with possibly free variables, and A is in prenex form. By the complexity of a formula in prenex form, we will mean here the total number of quantifiers occurring. Let Comp(A) be the complexity of For each integer m, on A. there is a natural predicate T_(n) GCdel numbers of formulae of complexity T_(n) says that the formula with GCdel number n, such that n is true. m The formalization of these truth predicates involve placing formulae in a weak Kleene normal form (i.e., no attention is paid to the form of the quantifiers, but only that the matrix be the T-predicate of the appropriate variables). By the reflection principle, for a theory language) of complexity (n)([n Pr(S,n)] -> m, S, (of this we mean the single sentence Gdel number of a sentence of complexity (i & T_(n)). We call this Rm (s). m We need some facts from proof theory which follow from work of Tait (see [5]), Feferman (see [1], p. 23) and Kreisel (see [43 ). Fact 1: 3 primitive recursive functions F, G, and H such that the following are provable in PRA: 1) For each m,n > O, F(m,n) an instance of induction on complexity < H(m), kG(n) gives the Gdel applied with no parameters. number of to a formula of 71 2) For each the theory m,n we have that Rm(In) In + the formula represented by defined as the subsystem of I J Pn F(m,n). is In consisting of induction < n. applied only to formulae of complexity Let is provable in be the sentence "for every purely Y1 (possibly with parameters), (3 2)(q)((q < _- predicate > Pq) & k < _p & (r)(r > (2 = p v( k -> -Pr))))." We let T' = T + k Jp. the infinite list In the theory T, the axioms (y)(3x)H (x), as n kn varies, have bounded complexity. Choose (y)(3 x)(HY conjunction of any sentence c' such that the (x)) with ReCA and kn any purely Z some fixed constant m = c'. c'. (x) + ReCA + J. kp has complexity < c!, We are interested in Fact 1, when c = H(c'). We let (y)(3x)Hy Jp, sentence and any TP = In + We define new theories for m = c Applying Fact 1 P we get Fact 2: 3 primitive recursive functions F the following 1) instance and G such that are provable in PRA: For each n > 0, of induction on F(n) kG(n) gives the Gdel number of an applied to a formula with no parameters of complexity < c. 2) For each in the theory n > O, we have that Rc,(In) is provable In + the formula represented by F(n). Now, since Comp(A) = Comp(-A), we see that formally implies Con(TP n + B) within TP n + B, R c,(In) where p is any 72 B integer, and is any purely formula represented by if TP + B 4 sentence. So hence the F(n) must not be provable in is consistent, by Gdel's matter what n and p T + B Theorem. This is so, no are. hew n Pew Fact :3 F, G, primitive recursive functions PRA, under the assumption Con(T + B), B sentence, we can derive, for some constant such that, in some fixed 4 c, (n)(p)Con(T + B + .(the sentence "F(n)")) & (n)(F(n) represents an instance of induction on kG(n) formula with no parameters of complexity by finitary Con(Zl-AC + B). means, that It suffices to show (n)Con(z1-AC with only Proof: induction c). we can conclude, From Con(T' + B) Lemma : • applied to a In + B). For each n, we get a model for "B + 1-AC with only I" by applying the inner model technique to the theory TG(nc3) + B + .(thesentence"F(n+c )" ). n+c3 Now on k F(n+c3 ) , G(n+c 3 ) parameters. represents a certain instance of induction and let this induction be applied to Now B says property); we fix such an ivized of each instance of Qs, no (3f)(f satisfies some specific f. 1 Then we wish to show the relat- "B + 1-AC with only induction In 1n 73 Rx = (3s)(3y)(Hf to the predicate, (y) & x < T & (r)(r < s k ks TG(n + c 3 ) + B + (the sentence n+c 3 "F(n+c3)"). This would give a consistency proof of "B + El-AC -> Qr)), with only "F(n+c3)") . is provable in relative to I n Con(TG(n+c) + B + n+c3 ,(the sentence This in turn is immediately generalizable to a consistency proof of Z1 -AC + B relative to Con(T' + B). Actually, it suffices to consider the case when n > 5. This lower bound will be convenient later. since the predicate and R has small complexity compared to is available in I provably relative to In So, certainly all instances of R c3 , TG(n+c 3 ) n+c3 n+c 3 Certainly, B provably relativizes, since the numbers, and hence all the arithmetical relations, remain unchanged by taking this inner model, by fiat. For similar reasons, clearly ReCA model. holds in this inner The inner model is non-empty, since Qq, since we assumed that induction on Q where is false. To verify the last, and most important axiom of with only in In", ReCA + In Hence unique y "B + 1-AC first note that since n 2 5, we have that, we can prove (a)(A < _G(n+c3) -> k TG(n+c). n+c 3 O.(q), k W(i) (3ty)H! (y)) is provable in ki From now on, whenever above by by well-known techniques. Q < _G(n+c3 ), k H f . Then also we denote the (2)(p)(t: ap < G(n+c3 ) & p & k _(n+c k 3 )] -> Hf < TH) k k k TG (n+ c ) is provable in n+c (p)(3 g)A(p,g,x) holds relativized to Now suppose that the predicate R, where we also have Rx. We wish to conclude, TG(n+c3 ) + B + ,"F(n+c3)" that 3 in the theory Skolem n+c function for the above sentence, h, (p)(3 g)(A(p,g,x) & Rg), So we have consider an integer nition of see that Rg. such that s A Rh. arithmetical. But satisfying the properties Since induction fails on s < _ G(n+c 3 ). in the defi- c3 k( Q, for we It is straightforward to see, using U AU W(k) and ReCA, that (p) (3s)(s 1) Hf (A(p,g,x)) & (t)(t <_s -> k A(p,g,x)). (3ts) t kt wit]h kt Hf p H < kt Ps s). G(n+c3 ) & (r)(r < (3 p)(stuff to the right of (3!s) Qr)]). (s)(Ps -> Applying the axiom k we get a l.u.b. in [(s < _ J G(n+c) k, call it ko < _G(n+c3). Then clearly k ko = G(n+c3 ), then 11 Hence consider the It is clear that > s ) T which holds iff and 11 can be written in a natural way in s -> form (since t < holds of < TH f does not have a It is also easy to see that the part of this sentence to the right of predicate 3 ) & (3g< TH ) < _G(n+c k ks ko, on the (r)(r < kt 0 -> with Ps. Qr). Now So if k (r)(r < _G(n+c3 ) -> k s Qr), contradicting 75 that induction fails on k 0 < _G(n+c 3 ). So G(n+c3 ) with Q. (r)(r < _k0 -> Hence, since Qr), we have (r)(r < kk 0 -> ar). A similar argument shows that G(n+c3) cannot be reached by a finite number of iterations of successor from ko, in the ordering k. But by sentence 1), it is clear that a Skolem function for our original sentence can be found recursive in at most a few jumps of H . Hence a Skolem function can be found satisfying the predicate R. In retrospect what we have shown is that for each sufficiently large n, " TG(n+c3 n+c 3 ) with only I + B" is consistent if (F(n+c3 ))" is. Consequently any + B + provable in 1 -AC £1-AC must also be provable in 1T'sentence T'. We now observe Lemma 2: Proof: in T. even T' T. It suffices to show that each Jp is provable But this is clear, since it is well known that T ust I + ReCA) proves induction for any formula on (or k, p for each p. TC- Z11-AC. Lemma3: Proof: (y)(3x)H y We have to show that for each (x) is provable in Z1 -AC. Let p, the sentence A(q) be the predi- kp cate (y)(3 x)H y k ordering kp. is provable in (x), and we apply induction to A on the q (Remember, the full schema of induction on E21-AC). If Suc_(r,q), then clearly Ik kp A(q) -> 76 A(r), by taking J-umps, using ReCA. 1 im_ (q) Suppose & (r) k (r < _q -> k A(r)). Then (r)(r < _q -> k ( 3 x)H Y kr (x)). form a Skolem function for this sentence, within W e can 1 1 -AC. When we do, we can find, recursive in at most a few Jumps of the Skolem function, a z with Hy (z). This is true of all kq and so we have A(q ). (q)(q < _p -> k A(q)). Hence by induction on kp, we have By a similar argument to the above, we obtain (y)(3x)Hy (x). kp Wehave immediately Theorem 3: sentences. £1 -AC is a conservative extension of T for 1 2 77 CHAPTER III In this chapter, we consider the question of ust which recursive linear orderings can have certain structures placed It is convenient to consider on them; namely, hierarchies. more general notions of recursive linear orderings and hierarchies, than in Chapter II. We say, in this chapter, RLO+(n) iff ape defines a recursive linear ordering (as in Chapter II) whose field is p Field(n), we let the name for the subordering of n, whose field is all q < p. RLO+. recursively enumerable. For any Thus np H+(x) as 1) (k)(kex -> We define Field(n)). itself is an 2) (q)(q -> with (3 m)(ke(x)m & m E [(On(q) & 9 < T(x)q) v (x)q)]). we can define the satisfaction Note that in ReCA+ I, We can do this, since we relation for (codings of) w-models. can prove in q be T(X)q) v (limn(q) & [P(r,s) s < nq & (SueC(q,r) & (X)(1) r e (x)S} < Field(n) np ReCA + I that (x)(J)(3y)(y H e (x), we define the corresponding Mx as (a coding of) the sets < T define Reas+(e) RLO(e) by in some x ( j) ) . Given (a coding of an w-model) (x)m, m Field(n). We the same as in Chapter II, except replace RLO+(e). We define lim(e) as (p)(3 q)(p E Field(e) -> P < eq). We define except replace W+(e) RLO(e) and by W*+(e) the same as in Chapter II, RLO+(e). 78 We say that NW(e,X) founded with respect to 2) RLO+(e) & e iff i.e., X, Lemma 1: Let have x for corresponding M x, (3n)(nEX 1) there is no e-least member of is not well- & n E Field(e)) X. He+(x), Reas+(e), lim(e). we have Mx ~ W*+(e). Then In fact this I + ReCA. Lemma is provable in Proof: Weconsider the theory T = I + ReCA+ (3x) (3e) (they satisfy hypotheses of the Lemma but not the conclusion). We will show that T, In the theory ing proves its own consistency. T fix such an Mx . We will show, in M It is clear that ) W*(e), = W(r) & < TX Hr(X) & M Fix such an r. z < (Y)t) (y)s with s and hence rW(r). So Mz J Y with Lim e such that 1) Consider and with M and holds. with NW(e,Y). (s ). (This NW(Y, e). (y)s. (t)(t < es There is a -> For, since M M l= Hr (X) & W(r), must be in the model Y, (n) & (z)t = We cla im that for these values, our claim about satisfaction in X, X and an s E Field(e)). 2) H+ (z), & Reas+ (e) & (ReCA + I), r there is a set Reas(e). n = e . We set g= ReCA. there is an is taken to imply implicitly set M -~w,(n)). Now choose We can do this, using T. (3z)(3n)(%H(z) & Reas+ lim(n) & corresponding M z ~ M f I. It remains to show M M M = correspond- and that this T, Also, since lim(e), we have Since x, e M z, s ). So certainly W*+(e and M ~ W(e ) we see that Mz /= (Mz certainly l W*+ 79 Mz e M, (es )). Note that since lim(e). This completes the proof that Hence T proves Con(T), Theorem, T. since the Soundness Theorem can be formalized and proved in So, by GCdels M T ReCA + I. is inconsistent. So the Lemma has been established. In fact, in view of the inconsistency, the ReCA + I. Lemma is provable in Theorem 1: of If Reas+(e), W* Reas+(e) and W*+(e) and W*+(e) RLO + (e)), then there is no Proof: If lim(e), corresponding M, are identical notions, since x H(x). with then if there was such an by Lemma 1, has W*+(e). Mx since every hyperarithmetic set is in must have If Mx (of course, in view M x, then However, (because W+(e)) we .W*+(e), which is a contradiction. . -lim(e), then by Reas(e), there is an lime(m) & there are only finitely many n > em. m So with W* + (e m) . Then argue as above. Let NTWO(n) be RLO+(n)& no tail i.e., (p)(p Field(n) -> of n the subordering on is well-ordered, qlq > np) is not well-founded). Lemma 2: NTWO(n), If Reas+(e) & H(x) & corresponding M x Mx then I + ReCA + S, where S is the sentence of Chapter II. Proof: (n)(WZ(n) -> NTWO(n). Let zeM. We wish to show Mz i ReCA + I + (3w)(HZ(w))). Clearly.lim(e), since Mz Hence Mz I + ReCA. = 8o Clearly (3s)(s e Field(e) & z n e Field(e) {qls eq model Mx such that the subordering of en) Let e defined on is satisfied not to be well-founded in the satisfies that there is a hierarchy y Mx Then . T(x)s). (in the generalized sense of this chapter) on a non-well-founded RLO+, k, the subordering of T' where Mx w M in view of m r e Field(n) Choose subordering of n NW(k,X). Now consider the n t to If (3q)(q > np & NTWO(nq)). r > np with and the .W+ (n). N1 predicate, Pt from I + ReCA + S: {s p • ns < nrj is not well- determined by We can do this since subordering of (y)p (3w)( (w) & I + ReCA + (3x) Field(n), then NTWO(n) & W*+(n) & p founded. (n)(Wz(n) -> The following is provable in Lemma 2: Proof: defined above, such that Then the obvioas generalization of Lemma Ok (p). 5 in Chapter II gives Ty)), e r t < nr & the If is well-founded. P has no solutions, we are done, for then NTWO(nr). [2])that has shown Harrison (see which has a solution in every Field(n), where Nl predicate W*+(n), has an n-least solution. His proof uses only principles provable in I + ReCA, orderings. excepting the comparability of recursive wellIn Chapter II, we showed this comparability Lemma is provable in I + ReCA + S, by Lemma 2 of Section 2. So there is an call it way r q, n-least solution to the predicate Pt, and we can prove this in was chosen, it is clear that I + ReCA + S. q > np. By the And by the way q is defined, it is clear that Lemma 4: ing | Mx Let NTWO(q). Reas+(e), H(x), lim(e). Then correspond- ~NTWO(e). Proof: T = I + ReCA + (3e)(3x) (they Consider the theory satisfy hypotheses, but not conclusion). As in Lemma 1, we wish to show that T proves its own consistency. So, we argue in T, that if e and T, that M MI + ReCA. f (3z)(3n)( (z) & Reas+(n) & It remains to show M lim(n) & corresponding Mz . NTWO(n)). By Lemma 2 of this chapter, M 1 -AC. we have M < eP (k)(k Now for each p f (k)(k < ep -> - > NWM(ep .(Y)k) M with M M NTWO(ep), (3Y) ) ' So again, (p)(p with Hence (3X)NW(ep,X)). Hence M (q > ep & (3Y)(k)(k < eq -> Z S + I + ReCA. Field(e) By Lemma 3 of this chapter, ReCA + S + W*+(e). and we will T. Clearly, as in Lemma 1, M C are chosen so M = Mx; that they violate this Lemma, then let show, in x M i (p)(p E Field(e) -> NW(eq, (Y)k))), by M Field(e) -> since M 1 (3q) I + 1-AC, we obtain a (Z)p is (Y,q) with q > ep & (k)( < eq -- > NW(eq,(Y)k))). Fix such a 2) Z T(X)r. Z, Let and let s > er (x)s. There is a natural (t)(t < e s -> r with have 1) M y K T()s (Y)t = (x)t). claim about satisfaction in We set M. r Field(e), = NTWO(s). such that Consider H z = y, n = e s, (y), and in our 82 Clearly lim(es) & Reas+ (es). It remains to show corres- My ponding NTWO(es). We have is in TWO(s), M Hence My - M. Z e My. and So every set AZ TWO(eS). This completes the proof of the self-consistency proving of T, and hence the inconsistency of T. Hence Lemma4 must be true. Lemma 5: There is an e with W*+(e) & Field(e) = [nln is even) such that 1) (3x)H(x), 2) (y)(H(y) -> (3X)(Xi, y(10) & NW(e,X))). Proof: We define a total recursive function F on indices of the partial recursive functions. We define G(n) to be the Gadel number of the RLO (field w) associated with the sentence "RLO(n) v (x)(~.H(x))." So for every the Gdel domain number of some {nln is even} RLO. Let defined by F(n) be the P < G(n)q By the recursion theorem, there is an Fix such an field is e. Then (nin e is even). is the Gdel y(10))). Hence W*+(e). (3x)(NW(e,X) & Hyp(X)), e G(n) RLO+ is with iff 2p <F(n)2q. tpe = PF(e) with number of an It is clear that (H(y) & y < T X(10))) & (y)(H+(y) -> n, T1 RLO+ whose (X)(NW(e,X) -> (3y) (3x)(NW(e,X) & X For, if not, then W+(e), and contradicting the 1st conjunct of the above conjunction. We claim ~W(e). For, if not, then (3y) (H+(y)), contradicting the 2nd conJunct of the above conjunction. So by the 1st conjunct, we have the properties (3y)(H+(y)). So stated in this Lemma. e has all 83 Theorem 2: ,) Take e W*(n) as in Lemma 5. e RLO with 1) the ordering ural Field(n) = w copies of (hence Field(n) = n to be the nat- is an initial segment, 2) (i.e., e x e). 'H(x). Then (3 y)(NW(e,y) & y E Mx). found recursively in such a n Take corresponds to n w, 3) the ordering Now suppose of with n such that (x) (H n (x)). Proof: But There are Mx Hence there is a y, Mx I NTWO(n). by Lemma 4. z which can be with the property that no tail is well-founded with respect to this contradicts NTWO(n) z. And z E Mx . But