Chapter 1. Informal introdution to the axioms of ZF.

1.1. Extension. Our conception of sets comes from set of objects that we know well such as N, Q and R, and subsets we can form from these determined by their properties. Here are two very simple examples: { r R | 0 r } = { r R | r has a square root in R } { x | x = x } = { n N | n even, greater than two and a prime }. (This last example gives two denitions of the empty set , ). The notion of set is an abstraction of the notion of a property. So if X , Y are sets we have that x ( x X x Y ) = X = Y . One can immediately see that this implies that the empty set is unique. 1.2. Unions. Once we have some sets to play with we can consider sets of sets, for example: { xi | i I } for I some set of indices. We also have the set that consists of all objects that belong to some xi for i I . The indexing set I is not relevant for us, so { xi | i I } = iI xi can be written X , where X = { xi | i I }. Typically we will use a, b, c, w, x, z , . . . as names of sets, so we can write x= Or simply: { z | z x } = { w | z such that w z and z x }.

x = { w | z x w z }.

(To explain this again we have y=

{y | y x} =

{ z | y x z y }.

So we have already changed our way of thinking about sets and elements as dierent things to thinking about all sets as things of the same sort, where the elements of a set are simply other sets.) 1.3. Power sets If x is a set then P (X ) = { Z | Z X }, the set of all subsets of x is also a set. Proposition. (Cantor) There is no surjection f : X P (X ) (a function f : X P (X ) such that for all Z P (X ) there is x X such that f (x) = Z . Proof. If f was such a surjection we could consider { x X | x / f (x) }, = A, lets say. As A X we have that A P (X ) and so there is some a X such that A = f (a) since f is a surjection. But look, now we have that a A if and only if a / A, uma contradiction! Corollary. The collection of all sets is not a set. Proof. Lets call the collection of all sets V . If V was a set then P (V ) would be a set as well. But P (V ) would be the set of all subsets of V , so for all x P (V ) we would have x V and hence P (V ) V . But in that case there would be a surjection f : V P (V ) given by x x if x P (V ) and x if x / P (V ). Contradiction!

ver 1.1, 12/2/06

Propoganda slogan: There is no universal set! This corollary is sometimes called Cantors paradox . Russells paradox can be obtained in a similar way by taking f to be the identity function in the proof of the proposition above to see that there can be no collection { x | x / x }. (That is, if { x | x / x }, = r, lets say, was a set, then the question of whether r r or r / r would give a similar contradiction.) Clearly these are not true paradoxes, but rather memorably/impressive facts. Russells paradox shows that we cannot hope to use the extensions of all mathematical properties as sets some are inherently contradictory and this gives us one reason to need to develop a satisfactory theory of sets. However I really ought to say that set theory began and continues to be interesting and useful not to avoid paradoxes or to give secure foundations to mathematics, but as a part of mathematics (just like its other parts) with its own theory, questions, techinques, etc., and mathematical links with other areas of mathematics. 1.4. The cummulative hierarchy. We will see what we get if we start with the empty set successively take power sets and unions. We begin with V0 = and repeatedly apply P . V0 V1 V2 V3 = = P (V0 ) = {} = P (V1 ) = {, { } } = P (V2 ) = {, { }, {{ }}, {, {} }} . . . { Vn | n nite }.

After all of the nite steps we collect together what we have so far and get: V = Now we can begin to apply P again: V+1 = P (V ), V+2 = P (V+1 ), . . . One can continue in this way generating sets in ones imagination for ever.

(The Vs are so-named for von Neumann , the rst person to think about this hierarchy.) Denition. A set is a member of some V . The universe of sets (generated from ) is the collection V = { x | ( indexes a step in the iteration & x V ) }. As a shorthand we can talk about V = { V | indexes a step in the iteration }, but as we have already seen, V is not a set and this union is not the union of a set we have that each V is a set but we cannot properly take the union of them and form V as a set . So we have to think of V in some sense as an uncompleted entity, and hence, V really is just an abbreviation for us well only use it (later) in occasions such as x V by which we formally will mean there is an such that x V . Nevertheless it is useful to have a name in hand for this collection of all sets. Proposition. If comes before as an index then V V . Proof. This is clear if = . It is also clear if V was generated by the process of taking a union (because V { V | is before as an index } and so if x V we have that x { V | is before as an index }).

However, if V = P (V ) for some we have to show that V V = V P (V ) in order to complete the proof. 2

So let me interupt the proof to give an important denition. 1.5: Transitive sets. Denition. A set x is transitive if u v x = u x (if and only if u x = u x, if and only if x x). To nish the proof it suces to show that each V is transitive. Because, as V V we have x V = x V , and if V is transitive we have x V = x V . Thus x V = x V , so x P (V ) = V . But as the following proposition shows, all V are transitive. Proposition. (i) is transitive. (ii) If x is transitive then P (x) is transitive. (iii) Unions of transitive sets are also transitive. Proof. (i) and (iii) are both trivial. (Can you see why?) (ii) If u v P (x), we have u v x and thus u x. But since x is transitive we have that u x. Hence u P (x). NOTE! To say that x is transitive is not the same thing as saying that is a transitive relation on x. ( is a transitive relation on x if every element of x is transitive.) We have implicitly used the principle of induction that ( < Pr( )) Pr() Pr(), where Pr(.) is a property of our indices. We will use this principle to show the following: Proposition. (-induction) x (u x (u)) (x) x (x), for any property . Proof. We show by induction on that x V (x). Let us give this property of , x V (x), the name Pr(). By our induction principle it is sucient to show for each that we have < Pr( )) Pr(). So, suppose that < Pr( ). Take x V . We have two cases. Either V = P (V ) for some less that , when x V and for all u x we have (u). Then using the antecedente for x we have (x). Because x was arbitrary we have x V (x), i.e., Pr(). Or V = < V , when there is some < such that x V . As we already know Pr( ) we have (x). Again, because x was arbitrary we have x V (x), i.e., Pr(). As was arbitrary we have ( < Pr( )) Pr() . And now we can deduce that Pr(), and hence x V (x), and thus x (x).

In order to avoid doing too many proofs like this we use a principle, Foundation, equivalent to -induction:

( ) If A is a property of sets and there is some x such that A(x) then there is some x such that A(x) and for all u x we do not have A(u). Symbolically: x A(x) = x A(x) & y x A(y ) Note: we thus have the idea that for anything there is a rst stage at which that thing happens. 3

We have two remaining problems:

(1) What are the properties of this hierarchy (2) For how long does it go on? Answer to (2) (part (a)). 1.6 Innity. We need the natural numbers. So here they are : 0= 1 = {0} 2 = { 0, 1 } 3 = { 0, 1, 2 } . . . Note if we dene s(x) = x { x } we have s(n) = n + 1 for each n. We also want a set of natural numbers: N = = { 0, 1, 2, . . . }. Thus we would like to dene = { x | 0 x & u x s(u) x }.

But are there any innite sets? Well, yes, but we have to stipulate that they exist. Innity. x ( x & u x s(u) x). (Note. V is such an x.) Other notions of innity. (1) Innite = not nite: x n x n wher x u if and only if f : x y and f is a bijection. (In words: there is some x not in a bijection with any natural number.) But to say this we need functions which we havent yet dened. (2) Dedekind Innity: x u x (x = u & x u). (In words: there is a set which has a proper subset of the same size as the set.) If a set is Dedekind innite it is innite. To show, in the other direction that an innite set is Dedekind innite requires a small amount of the Axiom of Choice (to show that if x is not nite then x contains a countable subset). 1.7. The language of set theory and properties. Properties are the properties that are denable in a language, the language of set theory, LST, with sets as parameters. In fact we can choose a very simple language for the language of set theory. (i) variables: u, v , x, z , . . . (ii) symbols for relations: =, (iii) propositional logical connectives: & (and), v (or), (implies), (not), (if and only if) (iv) quantiers: x (for all x), x (there is some x). Note: if we have a set a we have that { x | a x } is a property and not a set. With luck you understand this language informally and can argue using it. (For a more formal introduction to rst order logic take a look at, e.g., van Dalens Logic and Structure.) We are also going to use some natural abbreviations: 4

x=z x /z x z (x) x z (x) !x (x) ,

in in in in in

place place place place place

of of of of of

(x = z ) (x z ) x (x z (x)) x (x z & (x)) x ((x)) & z ((z ) z = x)

and so on and so forth. (Also I will often write or instead of v to avoid orthographic confusion between v and v .) Introduction of symbols for relations and functions. If I have a formula (x) in LST then I can introduce a new symbol R (x) for a relation and an axiom x(R (x) (x)). This doesnt change the expressive power of the language (what one can say in the language). For example: one can write x z if and only if u x u z , where is our new symbol. Similarly, if we have (x0 , . . . , xk1 , z ) and we can show (from some axioms) that, writing x for x0 , . . . , xk1 , x !z ( x , z ), we can introduce a symbol f (z ) for a function and a axiom x , y (( x , y ) f ( x ) = y ). Again this doesn not change what the language can express. Im next going to give some examples of use of this language. As we have already encountered most of the Zermelo-Fraenkel axioms it seems like a good idea to use them as our rst examples. 1.8 The axioms of Zermelo-Fraenkel (ZF) set theory The axioms are (briey): Extension Empty set Pairs (if we have two sets we also have the set to which belong exactly the two sets) Innity Unions Power sets Comprehension (or formation of subsets) Collection (not discussed up to now) -indution or Foundation: If A is non-empty then A has an element which is mininal in (minimal). Zermelo produced all of the axioms except Collection and so the system without Collection is called Zermelo (Z) set theory. [Bibliographic aside re history of development of axioms: see Michael Hallet, Cantorian Set Theory and the Limitation of Size.] Exercise. Show that Z holds in V+ . Now I give them again, using LST. (I also include a couple of comments.) Extension. x y z (z x z y x = y ). Empty set. x y (y / x).

By the axiom of extentionality !x y (y x). Thus we can introduce a new symbol for a constant such that y y / . (A constant is merely a function of 0 variables.) 5

Pairs. x y !z u (u z u = x v u = y ).

We can introduce a symbol for the function { ., . } with u (u { x, y } u = x v u = y ). (The pair of two sets is unique by extension.) Now we can see that {} { , { } } and so 1 2 (introducing the symbols 1 and 2 for constants in the obvious way for these sets). Unions. x z (u z y x (u y )).

Again we can introduce a symbol,

for a function and we have x (u

x y x u y ).

Power Set. x !z (y z u y (u x)).

Once more we introduce a symbol P with the eect that z P (x) z x. Innity. x ( x & y x (y { y } x)).

We introduce a symbol s for a new function and set s(y ) = y { y }, and dene by u x (( x & y x (y { y } x)) u x). With two further steps we can say that = { x | x & y x (y { y } x) }.

Comprehension and classes. By a property of sets I mean some formula ( x, a ) where a = a0 , . . . , an1 are names for some specic sets. The property is true of b0 , . . . , bm1 = b if and only if ( b , a ) is true. The extension of a property is the collection of sets of which the property is true. This collection is a subcollection of V , the collection of all sets. We call the extension of a property a class . Thus a class is something of the form { x | (x, a ) }. Axiom of Comprehension. For all formulas (x, y ) in LST we have x a0 , . . . , an1 y z (z y z x & (z, a )). Note, this axiom is in fact an axiom schema: we have one axiom for each formula (x, y ) in LST, and not merely one single axiom. In eect this schema says that for every x we have that each x { z | (z, a ) } is a set. Thus a class { x | (x, a ) } is a set (in the sense that z x (x z (x, a )) if and only if { x | (x, a ) } V for some . I.e., { x | (x, a ) } is a set if and only if it is bounded in the hierarchy. It is very useful (as we shall see) to be able to use various classes as informal abbreviations. Everyone does it and so shall we. However, you should bear in mind that the classes we use are only abbreviations and their use can always be eliminated in favour of talking about the formulars of LST that dene them (cf. the above remarks about V ). For example, if A = { x | (x, a ) } is a class, we can use b A s an abbreviation for (b, a ) being true, but one cannot think about whether A A is true. It is not false, but doesnt have any meaning . Similarly, P (A) does not have any meaning. If you still have questions or doubts about how to use or think about classes wait a short while and after we have used them a little things should become clearer. 6

 Foundation. For all formulas (x, a ) in LST y (y, a ) y (y, a ) & z y (z, a ). Using classes as abbreviations this says that if A is a class and is non-empty then A has an minimal element. How high is the hierarchy? (b) Suppose that x u y (x, y, a ). For each x there is a rst x such that y Vx (x, y, a ) (by the axiom of foundation). So we have a function sending u to { x | x u } given by x x . Thus { x | x u } is a set of levels, and hence there is a such that x < for all x u. So we have Vx V for all x u. If we write v for V we see that v x u y v (x, y, a ). But wait! We need an axiom (well, actually a schema of axioms, again) to do the collecting that gave us the set of levels we used above. Axiom of Collection. For each formula (x, y, z ) in LST (with free variables z ) we have the following: a u(x u y (x, y, a ) v x u y v (x, y, a )). Surprisingly this schema implies that the hierarchy carries on for a very long time. Note that feeding the u through the one gets that the axiom can be re-written as: for each formula (x, y, z ) in LST and tuple a of sets one has x u y (x, y, a ) u v x u y v (x, y, a ). Principle of Replacement For each formulas (x, y, z ) in LST (with free variables z ) we have the following: a (u !v (u, v, a ) x y z (z y w x (w, z, a ))).

Using classes as abbreviations again this says, If F is a operation denable in V then for every set x we have que F xx (the image of the restriction of F to x) is also a set. Some authors prefer to use Replacement in place of Collection in formulating the axioms of ZF. I think that it is better to use Collection (I think it is simpler to understand what is going on in it, in part because one does not have to use classes to formulate it), but it does not make any dierence formally which you use because Replacement follows from Collection and Comprehension. (Proof. For each a the lhs gives for each x that we have u x (!)v (u, v, a ). Using Collection we have s u x v s (u, v, a ). But the y that we need is a subcollection denable from s and so exists (is a set) by Comprehension.)

1.9. Other mathematially useful notions. Now we can introduce various other mathematically useful notions, giving denitions of them in LST. Denition Ordered pairs. (x, y ) = {{ x }, { x, y }}. Proposition. (a) For all x and y , (x, y ) is a set. (b) For all x, y , u and v , (x, y ) = (u, v ) x = u & y = v . 7

Proof. (a) By the axiom of Pairs (applied three times). (b) By extensionality. Note: if x = y the denition gives that (x, x) = { { x } } and we have to check this case as well. Denition. Cartesian product of two sets. x y = { (a, b) | a x & b y }. Proposition. For all x and y , x y is a set. Proof. For all x and y , x y is a denable subcollection of P (P ( { x, y })) and so is a set by the axioms of comprehension, power set, unions and pair. Denition. Relations and functions. A relation on the product x y is a subset of x y , i.e., an element P (x y ). So we can write, equally, Rel(x, y ) P (x y ) as an aide memoire . Functions are (just) certain relations. The collection of functions from x to y is written
x

y = { f Rel(x, y ) | a x !b y (a, b) f }.

Proposition. For all x and y , Rel(x, y ) and x y are sets. (We could carry on in this way, dening, for example, dom(z ) = { a | b (a, b) z } z and im(z ) = { b | a (a, b) z } z , and afterwards proving that, given that Rel(z ) x z a, b (x = (a, b)), we have Rel(z ) z dom(z ) im(z ).) Let us write f : x y when f x y , dom(f ) = x and im(f ) y . We would like to use the notation f (a) when f x y , a x and (a, b) f . If we are adding a name for a function denable in the language we ought to give a meaning to any f (a). So we set f (a) = b if a dom(f ), and f (a) = if a / dom(f ). Here we are using f in two senses, one as the name of the set (the function from x to y ), and the other as a new symbol in an extension of the language by a function denable in LST. So we need to set f (a) (f in the second sense) equal to if a / dom(f ) (f in the rst sense). In practice we will rapidly come to ignore such distinctions. Now we can dene injective, surjective and bijective for functions f x y , and so we can write the axiom of innity, for example, as: Innity. x n f x n f is bijective. Now, also, we can introduce the Axiom of Choice. Axiom of Choice. x ((y x y = ) f x x y x f (y ) y ).

This says that for all sets of non-empty sets there is a function that chooses a member of each one. Well see more about AC in chapter 3. Other similar useful denitions. If I and { yi | i I } are sets we have: General Cartesian Products. iI yi = { f : I { yi | i I } | i I f (i) Yi }.

Disjoint unions. iI yi = { (i, a) | a yi }. (Sometimes one sees iI yi written for iI yi .) Quotient of a set by an equivalence relation. R is an equivalence relation on x if R is reexive, y x (y, y ) R, R is symmetric, y, z x (y, z ) R (z, y ) R and R is transitive, w, y, z x (w, y ) R & (y, z ) R (w, z ) R. If R is an equivalence relation on x we have x/R = { y P (x) | a x b (b y (a, b) R) }. Proposition. Each iI yi , iI yi and x/R is a set. 8

Theorem. The collection of all sets, V , is closed under (i) denable subsets (ii) unions, (iii) products indexed b sets, (iv) disjoint unions indexed by sets, (v) spaces of functions (vi) power sets (vii) quotients and also contains N i.e., an object N with a function s : N N which satises Peanos axioms (which are give the rules that we expect for the conduct of the successor function on the natural numbers).

Using this theorem we can construct more or less any mathematical object (and it is easy to augment the theorem in order to show how to carry out other mathematical constructions in the theory of sets). These are no great shakes as you can see from the following examples. We can dene + and . on the natural numbers inductively. First of all + n + (m + 1) = s(n + m); 0 } { 1 } N { 2 } and afterwards . n.(m + 1) = n.m + n. Then we can dene Z as N { 0 }{ ( N 3), with (i, y ) < (j, z ) if i < j , or i = j = 0 and z < y , or i = j = 2) and y < z , for example. (But clearly we could also choose any other reasonable scheme.) We can then give the laws of +, . on Z. Next one can dene Q as Z N / where is a relation (a, b) (c, d) if a.d = b.c. After this one can dene R via Dedekind cuts. Or one could equally well, for example. dene { r R | 0 < r < 1 } via binary expansions, or via continued fractions, and take R = Z { r R | 0 < r < 1 }. (You can nd more information about continued fractions in any good book on the elementary theory of numbers, such as, for example, Alan Bakers excellent A Consise Introduction to the Theory of Numbers , Cambridge University Press, pp. 41-42.) As you will have seen by now, these constructions are more or less trivial and one has a great deal of freedom in the details of how one chooses to carry them out.