Metainduction in Operational Set Theory

Transcription

1 Metainduction in Operational Set Theory Luis E. Sanchis Department of Electrical Engineering and Computer Science Syracuse University Syracuse, NY sanchis November 6, Introduction. Metainduction is a type of primitive construction, in the sense of Sanchis [3], that can be applied in any standard system of set theory, but it is crucial in those systems we call operational. Essentially, it describes a construction that takes place outside the system, but involves operations or sets available in the theory. On the basis of such a construction new operations and predicates are defined, always outside the system. The critical point in the process is the formal introduction of those operations and predicates (as symbols) with axioms which are derived from the construction. Clearly, this procedure provides a machinery where we can extend a given theory with new operations and predicates that are controlled by the new axioms. These extensions can be applied at any time and as frequently as the system operator may wish. Note that, in general, the extensions derived by metainduction are not conservative, for the new axioms are inductive, and there is no machinery available to eliminate the new symbols We deal with systems of set theory formalized in a classical first order logic involving set predicates (in particular and =), set operations, propositional connectives, local quantifiers, and global quantifiers. A local formula is a formula that does not contain global quantifiers. A formula φ is operational if it is of the form ( Y 1 )...( Y n )φ, where 0 n and φ is a local formula. 1

2 2 We say that a system of set theory is operational if all the axioms are operational. For example, the critical replacement axiom in standard set theory is not operational. On the other hand the separation axiom is operational in some cases, hence it may be a part of an operational system. In an operational system metainduction is a crucial procedure to introduce new operations and predicates. In fact, in such systems the usual procedures via existencial axioms, in particular the so-called replacement axiom, are not available. The idea of allowing primitive constructions, in particular metainduction constructions, was introduced in [3] as a resource in the partially formalized system G The restriction to operational axioms is only a part of a more general restriction concerning the universe of set theory. As explained in [3] we reject the idea of such universe as a complete totality. We call this restriction the predicativity condition. Notwithstanding the predicativity condition, we allow formulas involving global quantifiers. The point is that we need global universal quantifiers to express properly the axioms, and under the specific restriction to operational axioms those quantifiers play a very limited role. Furthermore, we allow classical logic in dealing with both global and local expressions. This is a change from the position taken in [3], where we required intuitionistic logic for global formulas In this paper we assume a fixed given operational system OST where the following assumptions are valid In OST it is possible to generate new operations by substitution with given operations, variables and constants. Hence, if f is a k-ary set operation and V 1,..., V n are given terms, where the variables occur in a list X = X 1,..., X k, we can introduce a new set k-ary set operation f with the axiom: f (X) = f(v 1,..., V k ). This axiom is clearly operational. In general, if W is a term where the variables occur in the list X, we can introduce an operation f with the axiom f(x) = W Given a local formula φ where the set variables occur in the list X = X 1,..., X k it is possible to introduce a new predicate p with the axiom: p(x) φ. This axiom is operational, in the sense defined above, and the new predicate is

3 3 primitive in the same sense that and = are primitive. In other words, any rule or specification in the system that refers to primitive predicates apply to the new predicate p. In particular p(v 1,..., V k ) is an atomic formula in the usual sense The replacement rule in [3], Chapter 2, is also available in OST. This means that all the operations introduced there (in particular those in page 33-34) are available The assumptions described above are very week indeed, and they do not include sets which are usually considered to be essential en set theory, in particular the set ω. In fact, the purpose of this paper is to describe a machinery where the system OST can be extended in order to provide a more satisfactory theory of sets. On the other hand we do not intend to bring in all the essential constructions in classical set theory (see 1.1.1), in particular the power set operation P(X) for any arbitrary set (although we propose a well known construction for the set P(ω)). Still, the extensions via metainduction provide a theory where many important constructions are in fact available. Some of these applications come from [3], but here we are more systematic, stressing their generality and uniformity. We shall show that the applications are essentially inductive and induce general rules of proof by induction. 2. Metainduction. Metainduction is a construction that depends on a given set operation ρ and a given set Z, both available in OST. The construction itself is independent and takes place outside OST A ρ-branch for Z is a (potentially infinite) sequence of sets Z 0, Z 1,..., Z k,... (k a numeral), where Z 0 Z and Z k+1 ρ(z k ). While ρ is a fixed given operation, Z is not necessarily a fixed set, and can be assumed to be an arbitrary set. On the other hand ρ is not necessarily a unary operation, and may contain extra arguments that behave as parameters in the construction. To make explicit this situation we should write ρ(z,x) where X is a list of extra variables X 1,..., X k. In general we do not take advantage of this notation, but the reader should keep in mind this more general formulation. In one application below one extra parameter W is in fact brought explicitly into the notation (see 5.1.1) Note that, essentially, the definition says that Z k+1 is determined by choice among the elements of ρ(z k ). So here we are assuming a weak informal version of the axiom of choice, in fact close to the usual axiom of dependent choices. This does not contradict our basic assumptions, because the choices take place one

4 4 by one and in each case over a given set ρ(z k ). This process is predicative because it does not assume the universe of sets as a complete totality If ρ(z) = then Z 0 is undefined and the ρ-tree is empty. In general, if for some k, Z k is defined and ρ(z k ) =, then Z k+1 is undefined and we say that the ρ-branch halts at k. Otherwise, Z k+1 is defined. We say that the sets Z 0, Z 1,... are ρ-residuals of Z. The set Z is not necessarily a residual The structure determined by the totality of all ρ-branches we call the ρ-tree for Z induced by ρ. More precisely, we mean that the ρ-tree consists of all the ρ-branches when we assume they are generated simultaneously. The ρ-residuals in the branches are also ρ-residuals in the ρ-tree. We say that ρ is the handle for the ρ-branches and also for the ρ-tree. From now on we omit the prefix ρ unless it is necessary or convenient At this stage we postulate a completion of the ρ-tree, that contains all the residuals and all the branches, even if some of them are infinite. More precisely, the completion looks at the ρ-tree as a complete totality ( exactly in the way we claim it is not legitimate to look at the universe of sets). The completion of a ρ-tree is admissible as long as it does not involve the universe as a complete totality From the completion of a ρ-tree we can derive sets, operations and predicates, as we show in the applications below. On the other hand, it is a fact that the constructions we propose in this paper can be properly defined inside any standard set theory, say the Zermelo-Fraenkel theory. But such theories are not operational and contain inpredicative components, via the replacement axiom, the foundation axiom and the general power set construction, that we do not assume available in OST We describe below a number of applications of metainduction, where from given operation ρ and set Z we introduce the ρ-tree construction, take the completion and define new operations (eventually sets) and predicates. These operations and predicates are defined in the contex determined by the ρ-tree, so they are not meaningful definitions inside OST (although they may be available in some standard impredicative systems of set theory, as explained in 2.2.1). Rather, we derive axioms from the construction, and these axioms are formulas in OST, and in fact become axioms there. This is a two steps process where first we define operations or predicates in the metainductive construction, and then write axioms for the operations or predicates in OST. In every application the axioms are operational in the formal sense

5 5 explained above (see 1.1). 3. Transitive closure. In the first application of metainduction we assume an operation ρ (in OST) and define a new metainductive operation tc ρ. tc ρ (Z) = the set of all ρ-residuals of Z, where Z is an arbitrary set in the universe of operational set theory. We call tc ρ the transitive ρ-closure operation. The set tc ρ (Z) comes essentially from the completion of the metainduction construction induced by the operation ρ and the set Z. It is in fact a definition in the construction and it is meaningless outside the construction (see 2.2.2) The next step introduces the operation tc ρ (in fact a symbol) into OST with the proper axioms, where X, V, Z are arbitrary sets in OST: TC 1: ρ(z) tc ρ (Z). TC 2: X tc ρ (Z)..ρ(X) tc ρ (Z). TC 3: Z V ( X V )ρ(x) V..tc ρ (Z) V. Here, as in every application of metainduction, we must show that the axioms follow indeed from the definition o tc ρ. This argument takes place outside OST, although it may involve known properties of the operation ρ, that come from OST. For example, in the example above concerning the operation tc ρ, we note that if Z 0 ρ(z), then certainly Z 0 is a residual in some ρ-branch, hence ρ(z) tc ρ (Z) and TC 1 holds. Furthermore, if X is a residual then Y is also a residual whenever Y ρ(x) hence TC 2 holds. Finally, if V is a set that satisfies the closure properties in axiom TC 3, then in the process that generates a ρ-branch of Z each generated residual is an element of V. We conclude that tc ρ (Z) V, so TC 3 holds. Technically, the axioms are the universal closure of such assertions. Still, we refer to them as axioms. The whole application consists in the introduction of a new primitive operation tc ρ and three new axioms in OST. We call these axioms the Peano axioms for the operation tc ρ. As usual, from the Peano Axioms we can derive a rule of proof by residual induction Residual Induction(in OST) :Le p be a set predicate that satisfies the following condition for a given set Z:

6 6 (i) p(z) ( X tc ρ (Z))(p(X)..( Y ρ(x))p(y )) It follows that: (ii) ( X tc ρ (Z))p(X) This follows immediately from axion TC 3, by setting the set V by separation in the form: X V X tc ρ (Z) p(x), and it follows from the assumptions that tc ρ (Z) V, in fact, that tc ρ (Z) = V The preceding argument depends only on the axioms of OST, so we have proved in OST that the implication (i) (ii) is valid for every set Z, any operation ρ in OST, and any predicate p defined in OST. This result supports an inductive procedure, where in order to prove that p(x) holds whenever X tc ρ (Z), we need only to prove (i) holds. In the proof of (i) the condition p(z) is usually given as an assumption. To prove the second part of (i) we assume typically that p(x) holds for some X tc ρ (Z) (the induction hypothesis), and prove that p(y ) holds for every Y ρ(x) An important example of transitive closure is the set ω, with the usual meaning in standard set theory. To represent the natural numbers we introduce the operations X + = {X} (the successor) and ρ(x) = X ++ (the handle). We represent 0 with + and n + 1 with X + if X is the representation of n. Now we set we set ω = tc ρ ({ }. The ρ-tree induced by ρ consists of only one infinite branchs of the form: +, ++, +++,..., and the completion of this tree is the set ω of natural numbers. The Peano axioms for the set for ω take the following traditional form: NT 1: + ω NT 2: X ω X + ω NT 3: ( V ( X V )X + V..ω V. In these axioms the variables X and V range over the universe of sets. It follows from that in order to prove p(x) for every X ω we need only to prove p( + ) and p(x) p(x + ) for every X ω. As usual, we refer to this procedure as mathematical induction.

7 7 4. Bar Induction. Bar induction is, in some sense, the inverse of mathematical induction (see 3.2). In fact, it is quite different from the residual construction, because it introduces a predicate, rather than an operation. Still, it is another application of metainduction Again, we start with a set operation ρ from OST, and Z is an arbitrary set. We define a predicate WF ρ as follows: WF ρ (Z) every ρ-branch of Z halts. We read WF ρ (Z) as Z is ρ-well-founded. Again, this is a metainductive definition that has no meaning in OST, although it can be formalized inside an impredicative system (see 2.2.1) As usual, we have to write axioms in order to introduce this predicate in OST. Closure Axiom : WF ρ (Z) ( X ρ(z))wf ρ (X) Foundation Axiom : WF ρ (Z) Z V..( Y V )V ρ(y ) = These are axioms in OST, intended to be valid for arbitrary sets V, X, Z. Still, we must show that they indeed follow from the metainductive construction. The closure axiom is trivial noting that if for some X ρ(z), X is not well-founded, then there is a non-halting branch for X. Clearly, this means there a non-halting branch for Z, contradicting WF ρ (Z). The foundation axiom is more involved. Assume given sets V, Z, where WF ρ (Z) holds and Z V. We want to show there is Y V such that V ρ(y ) =. If V ρ(z) =, we take Y = Z and we have finished. Otherwise, we generate a ρ-branch of Z, say Z 0, Z 1,..., Z k,... but we impose the following condition for any k: if V ρ(z k ), then Z k+1 V ρ(z k ) (and otherwise, Z k+1 ρ(z k ), as usual). Now we note that there is a k such that ρ(z k ) = (the branch halts at k). It follows that there is a first k with the property that Z k V and V ρ(z k ) =. Hence, we take Y = Z k The argument in that validates the foundation axiom takes place in the metainduction construction and depends on the definition of the predicate WF ρ. Still, the axiom itself is a formula in OST, which now becomes an axiom in OST. With the two new axioms in OST, we can use them to prove formally (in OST) formal properties of bar induction, noting that residual induction (3.1.1) is also

8 8 available. For example, we can prove the following theorem: WF ρ (Z) ( X tc ρ (Z))WF ρ (X). This follows by residual induction, taking as p the predicate WF ρ and using the closure axiom. On the other hand, there is not yet a rule of proof by bar induction We proceed now to prove a general rule of bar induction, that involves an arbitrary predicate p (from OST, compare with 3.1.1). First, given a predicate p in OST we define (also in OST) the following predicate p : p (X) ( Y ρ(x))p(y )..p(x). In particular, if ρ(x) =, then p (X) p(x) Bar Induction Rule: Let p be a predicate in OST, and Z a set such that the following condition is satisfied: (i) WF(Z) ( X tc ρ (Z))(p (X) p(x)) It follows that: (ii) ( X tc ρ (Z))p(X). In order to prove (in OST) that (i) (ii) we assume (i) and introduce by separation the following set V : X V X tc ρ (Z) p(x). We shall show that V =, so (ii) holds. To get contradiction, let X V. Noting that WF ρ (Z) holds by (i), it follows from that WF ρ (X) also holds, hence by the foundation axiom there is X V, such that WF ρ (X ) also holds, and ρ(x ) V =. Furthermore, ρ(x ) tc ρ (Z) holds, hence ( Y ρ(x ))p(y ) also holds. From (i) it follows that p(x ) holds, and this is a contradiction because X V The preceding result supports an induction procedure to prove assertions of the form: ( X tc ρ (Z)p(X) for a given predicate p and a given set Z. In fact, if we know WF ρ (Z) holds, we need only to assume p (X) and prove p(x) for any arabitrary X tc ρ (Z). The assumption p (X) is here the induction hypothesis (compare with 3.1.2). 5. Numerical Functions. We have discussed two applications of metainduction: transitive closure and bar induction. In the first we introduce an operation

9 9 tc ρ (Z) and prove that an inductive procedure is available to prove statements of the form ( X tc ρ (Z))p(X). In the second we introduce a predicate WF ρ (Z) and show that a different inductive procedure is available to prove a similar statement under the assumption that WF ρ (Z) holds Now we extend the scope of metainduction with a different application where we are concerned with numerical functions, which are functions in the usual sense of set theory: sets of single valued ordered pairs. The formal definition is available in OST via the predicate NF(F, W) that means F is a numerical function over the set W: NF(F, W) FU(F) do(f) = ω ra(f) W For the notation in this definition see [3] Chapter 3. In this definition F is an ordinary set symbol (like X, Y, V,...), which we intend to use to denote functions We want to introduce in OST, via metainduction, a set operation F such that the following axiom is satisfied for any sets W, F: F F(W) NF(F, W) The basic idea is quite simple. We identify the numerical functions in F(W) with the ρ-branches generated by a handle ρ, defined in such a way that every branch determines exactly one numerical function over W, and every numerical function is determined exactly by one branch. Via the completion of the ρ-tree we get the set of all ρ-branches, hence the set of all numerical functions over W. The set W is a parameter in the construction, that appears as an argument in the operation ρ, so we have now a binary operation ρ(z, W) (see 2.1). In this section we assume that W. In the following the symbols n is a set variable that denotes an element of the set ω when use in positions that require a set argument. Otherwise, it is a syntactical symbol that denotes numerals We take as initial set Z = and define ρ(, W) = { } W. Hence, hence in every ρ-branch the first residual Z 0 is <, w>, where w W. To complete the definition we define ρ(z, W) for the case Z = <n, w >, where n ω and w W. We set ρ(<n, w >) = {n + } W. For example, ρ(<4, w >) = {5} W. Clearly, ρ is an operation in OST. It follows that a ρ-branch under these definitions takes the form: <0, w 0 >, <1, w 1 >, <2, w 2 >,..., <n, w n >,...

10 10 where w 0, w 1, w 2,..., w n,... are arbitrary elements of W (eventually all the same). Note that when n occurs as a subscript it is a numeral, and not an element of ω. The completion of the ρ-tree induces the completion of each ρ-branch, which is a set of ordered pairs, in fact it is a numerical function over the set W. So the completion of the tree induces the set of all such functions. It should be clear that this construction does not involve the universe of sets We denote by F(W) the set of all numerical functions on W, as determined by the preceding metainductive construction. This construction is a ρ-tree, where the behavior of the operation ρ is determined by the rules above. The completion of the tree is an objective structure where each numerical function occurs as a ρ-branch (more precisely, occurs as the set of all residuals in some particular branch) To introduce F in OST we must determine the (operational) axioms that control the operation F. Note that the argument to support the axioms takes place in the metainduction construction, although the axioms are written in the language of OST. This is, of course, the same situation we have found in the applications above (see 3.1 and 4.1.2). The first axiom is a trivial consistency property, and simply says that the elements of the set F(W) are numerical functions on W. This follows because the elements of F are (the completions of) ρ-branches, and each one of these is a numerical function by construction. Consistency Axiom: F F(W) NF(F, W) This axioms is necessary, for example, whenever we introduce a set F with the assumption F F(W), as it provides a minimal basic information about the meaning of such characterization In the other direction, a second axioms appears to be necessary where if a set F is found to have the necessary properties, we can assert that it is an element of F(W). We start with a weak version of this condition. Weak Completeness Axiom: NF(F, W) F F(W)

11 11 Clearly, we need this axiom, but it is weak in the sense that it determines that F F(W), but it fails to take advantage of the metainductive construction that supports the set F(W). For example, the fact that the elements of F are actually generated by free choices (see 2.1.1) plays no role in the axiom. So, we propose next a stronger second version of the axiom, where under given circunstances we assert the existence of some F F(W), that satisfies an extra condition that depends on a given binary predicate p (from OST) The strong completeness axiom requires a technical construction, available in OST (see [3], definition 2.4.1). If F is a numerical function, and n is a natural number, then F n is also a function where do(f n) = n and for i < n, F n(i) = f(i). This construction is usually referred to as the restriction of F to n. The strong completeness axiom takes the following form, where p is a given predicate: Strong Completeness Axiom: ( F F)( n ω)[( X W)p(F n, X)..p(F n, F(n))] This axiom is operational, and it is intended to be valid for any any set W. As explained above, the predicate p is given, so here we have an axion schema that becomes an axion once the predicate p has been determined The argument that supports this axiom involves a simple application of the definition of metainduction. We assume the predicate p and a non-empty set W, and want to prove that there is F F(W) that satisfies the axiom. To prove there is such a function F we generate a branch <0, w 0 >,..., <n, w n >,... as follows. Suppose that at n there is in fact some X W such that p(f n, X) holds. If this is the case we proceed to choose w n = X, where X is arbitrarily chosen among those sets in W that satisfy the condition. If it is not the case, we take w n arbitrarily. By applying systematically this strategy we get a numerical function F that satisfies the axiom. Note that the operation F (that comes from the metainduction construction) is essential for the argument. For this operation provides the local existential quantifier that it is required for the axiom to be operational Furthermore, the argument in is essentially recursive, for the value of F(n) depends on the values F(0),..., F(n 1), although it is not completely determined by such values. In fact, the determination involves a choice, so we have a combination of recursion with choice.

12 The weak completeness axiom follows (in OST) from the strong completeness axiom. To prove this assume F is a numerical function over W, so the predicate NF(F, W) holds. We want to show, using the strong completeness axiom, that F F(W). To apply the axiom we take a predicate p defined: p(y, X) F (do(y )) = X. From the strong completeness axiom it follows that there is F F(W) such that whenever for some n there is X W such that p(f n, X) holds, then p(f( n, F(n)) holds, hence (noting that do(f n) = n), F (n) = F(n). Now, using induction on n, it follows that F (n) = F(n) for every n, hence F = F and F F(W). This proof can be formalized in OST using mathematical induction on n from The set P(ω) = the power set of ω, can be derived from the operation F. For example, we can introduce this set with the following axioms, where h is an auxiliary operation: X h(f) X do(f) F(X) = 0. Z P(ω) ( F F({0, 1})Z = h(f) Both operations follow from the replacement rule in OST using 1.2.3, and noting that h is derived by separation, and furthermore Z = h(f) can be expressed as Z {h(f)} The set Z P(ω) in the axiom above is actually generated via the strong completeness axiom, and this means it can be forced to satisfy a previous inductive condition defined by a predicate p. As explained in , this process is essentially recursive, although it involves an element of choice. 6. Recursion. Whenever there is a rule of proof by induction we have an associate recursion rule, where an operation f can be defined (in the metainduction construction) and the values f(x) of the operation are derived in a manner similar to the proof of p(x) in the corresponding induction rule. Such a rule may not be well defined in somes cases. 6.1 The first rule we must consider is residual induction in Here, the purpose of the recursion rule is to introduce an operation f with domain tc ρ (Z) where the value of f(y ) for Y tc ρ (Z) is determined by the value of f(x) whenever Y ρ(x). In principle, such a rule is not well defined because for a given Y there are

13 13 many residuals X such that Y ρ(x ) and the value of f(y ) will depend on which X is used. In other words, a recursion rule in this situation is not single-valued Obviously, we can overcome the difficulty in the preceding paragraph by considering a case where ρ is strongly single-valued. This happens when tc ρ (Z) is in fact the set ω in 3.2. Here the recursion rule takes the following form: the value of f(x ++ ) is determined from the value of f(x + ), and the value of f( + ) is given. This is, of course, the well-known rule of primitive recursion (see also 5.3.1) In the case of bar induction, the order of dependency is reversed, and there is no ambiguity in the recursion. Here the value of f(x) is determined from the values of f(y ) when Y ρ(x). This form of recursion is known in the literature as bar recursion The application of metainduction to numerical functions induces a form of induction combined with recursion, via the strong completeness axiom (see ). This axioms depends on a given predicate p, so it is essentially inductive. At the same time asserts the existence of a function F F({0, 1}), where the value of F(n) is determined, up to some point, from the values in F (n), and this is essentially recursion Operational set theory by itself, as explained in the introduction, is an extremely week theory where not even the existence of an infinite set can be proved. By allowing metainductive constructions we move to a different system where a substantial portion of standard set theory can be formalized. From metainduction we get not only the set ω, but also the set P(ω), hence the set P(ω ω), which is the set of all countable relations, in particular the countable well-orders. Finally, we get the set of all countable ordinals, hence the ordinal ω 1, the first non-countable ordinal. Still, the general power set operation is missing. 7 Why Operational? We have explained in the introduction our reservations concerning the notion of a universe of sets as a complete totality. We discuss now a pair of examples where sets are introduced in violation of this predicativity condition, and in one case the result is the actual inconsistency of the theory. We note first, that in most systems of set theory the introduction of a set V with the properties of the universe (say via the axiom Z V) is inconsistent The first example is the set R in the Russell paradox, which is introduced with the axiom: Z R Z Z.

14 14 This appears to be a correct operational axiom, but it implies the inconsistent assertion R R R R. As a consequence, systems of set theory are always designed with sufficient restrictions to avoid the introduction of the set R. We want to argue that the axiom as given, independently of its consequences, involves the universe of sets and it is not admissible in operational set theory. In fact, the reference to the universe is implicit in the axiom, in the sense that the axiom does not determine the extension of the set R. The axiom only describes when a given set Z is an element of R. To determine the extension we must check each and every element of the universe, and only in this way we really know which is the extension of the set R. In other words, the axiomatic characterization of the membership relation is not sufficient to determine the extension of the set R. This deficiency of the axiom is crucial when we consider the situation where we have local quantifiers of the form ( Y R)φ or ( Y R)φ where φ is some formula. The meaning of these assertions depends on the totality of elements in the set R, and the axiom only say that such a totality is determined by the elements in the universe that satisfy the axiom. In other words, the quantifiers are in fact global quantifiers, although they are written with the notation for local quantifiers. In principle, global quantification is impredicative, as it implies a reference to the universe as a complete totality The preceding argument can be read in two different ways. In classical set theory the problem with the local quantifiers is irrelevant, for all quantifiers are in principle global and the axiom is rejected because it is inconsistent, not because it involves a quantification over the universe. On the other hand, in operational set theory we may say that it is not at all surprising that an axiom so obviously impredicative is inconsistent Several devices have been created in order to avoid the introduction of the paradoxical set R. This result is obtained by imposing some restrictions, in particular via the separation axiom, which is fairly strong, and also via the replacement axiom that it is weaker. These are reasonable safeguards, but can be avoided simply by not enforcing them. This is what happens in the second example The second example is more important, because it involves the power set axiom, which is crucial in classical set theory. In fact, most of the constructions derived above via metainduction can be handled, without much problem, in classical set theory by means of the power set operation. We write P(X) = the power set of X, which can be described by the following

15 15 axiom: Z P(X) ( Y Z)Y X. The problem with this axiom is essentially the same: the axiom fails to determine the extension of the set P(X), which is essentially determined by the universe. On the other hand, the axiom is universally assumed to be consistent, and there is no expectation, in classical or non-classical set theory, that the axiom may eventually turn out to be inconsistent. This happens, simply, because the power set axiom ignores both the separation axiom and the replacement axiom. On the other hand, applications of separation or replacement actually involve applications of the power set, and would be impossible without such applications. In this way, while substantial restrictions are imposed via separation and replacement, they can be easily violated via the power set, which is not subjected to such restrictions These arrangements appear to be essentially oportunistic. The power set is a crucial part of standard set theory, that in some sense generates the whole universe of sets. Unless it is shown to be inconsistent it will continue to play a fundamental role. When we started this project of operational set theory we expected to find a construction to support the power set, and in fact a construction was proposed in [3], Chapter 8. Unfortunately, now we think the construction involves the universe, and we have decided to reject it (see also Section 8.5 in [3]). On the other hand, we have expanded the constructions available in [3] via the general notion of metainduction proposed in this note. As expanded here metainduction provides support for P(ω), and also for P(ω ω). This set is essentially the set of all countable relations, in particular of all countable well-orders, so we have also the set of all countable ordinals, hence the ordinal ω 1, the first uncountable ordinal. 7.3 We can always consider operational set theory as a subtheory of Zermelo- Fraenkel, with special properties. In particular, the fact that the theory is operational imposes considerable proof theoretical restrictions, that it will be useful to explore (for example, in the direction of Theorem 9.5 in [3]). Holmes [1] makes some interesting suggestions about operational set theory. On the other hand Maybery [2] rejects our argument about the power set operation (see 7.2). REFERENCES

16 16 [1 ] Holmes, M.R. (1998). Review of [3]. Mathematical Reviews 98: [2 ] Mayberry, J. (1998). Review of [3]. The Journal of Symbolic Logic 63: [3 ] Sanchis, L.E.(1996). Set Theory - An Operational Approach. Gordon and Breach Publishers.