TheoremBase

We first derive from Gödel's class existence axioms classes that reverse ordered pairs and that test membership conditions on chosen entries of tuples, then rewrite the formula into a normal form whose only atomic parts are set-in-set and set-in-class memberships, and finally build the class by induction on the number of connectives and quantifiers. Uniqueness for one variable follows from extensionality.

Proof

The argument follows Mendelson's proof of the class existence theorem (Introduction to Mathematical Logic, Chapter 4, Proposition 4.4 and the propositions before it), reorganised. It is a proof scheme: for each fixed formula, and for each fixed metatheoretic number used below, the inductions below (which are inductions in the metatheory) produce a proof in the theory.

Throughout, letters a,b,c,p,q,ya,b,c,p,q,y in Step 0 denote sets. We use two facts about tuples that hold by their definition: every tuple and every ordered pair of sets is a set, and ⟨x1,…,xn,xn+1⟩=(⟨x1,…,xn⟩,xn+1)\langle x_{1},\dots,x_{n},x_{n+1}\rangle=(\langle x_{1},\dots,x_{n}\rangle,x_{n+1}), in particular ⟨a,b,c⟩=((a,b),c)\langle a,b,c\rangle=((a,b),c) and ⟨x1⟩=x1\langle x_{1}\rangle=x_{1}. All equivalences below are obtained by instantiating the universally quantified conditions of the axioms at these sets; no injectivity of pairs is needed. We also use that some set exists, by the axiom of the empty set; call this fact (S). Each clause of the class existence axioms applies to arbitrary classes XX and YY.

Step 0. Derived class constructions.

Lemma A (reversing pairs). For every class XX there is a class ZZ such that (a,b)∈Z(a,b)\in Z if and only if (b,a)∈X(b,a)\in X, for all sets a,ba,b.

By (B5) applied to XX there is X1X_{1} with (p,c)∈X1(p,c)\in X_{1} if and only if p∈Xp\in X, for all sets p,cp,c; taking p=(a,b)p=(a,b) gives ⟨a,b,c⟩∈X1\langle a,b,c\rangle\in X_{1} if and only if (a,b)∈X(a,b)\in X. By (B7) applied to X1X_{1} there is X2X_{2} with ⟨a,b,c⟩∈X2\langle a,b,c\rangle\in X_{2} if and only if ⟨a,c,b⟩∈X1\langle a,c,b\rangle\in X_{1}. By (B6) applied to X2X_{2} there is X3X_{3} with ⟨a,b,c⟩∈X3\langle a,b,c\rangle\in X_{3} if and only if ⟨b,c,a⟩∈X2\langle b,c,a\rangle\in X_{2}; by the previous condition (instantiated at b,c,ab,c,a) this holds if and only if ⟨b,a,c⟩∈X1\langle b,a,c\rangle\in X_{1}, that is, if and only if (b,a)∈X(b,a)\in X. So the swap of the first two entries of a triple is the composite of one transposition and one rotation. Finally, by (B4) applied to X3X_{3} there is ZZ with p∈Zp\in Z if and only if there is a set cc with (p,c)∈X3(p,c)\in X_{3}. For p=(a,b)p=(a,b) we have (p,c)=⟨a,b,c⟩(p,c)=\langle a,b,c\rangle, so (a,b)∈Z(a,b)\in Z if and only if there is a set cc with (b,a)∈X(b,a)\in X. The condition (b,a)∈X(b,a)\in X does not involve cc, and by (S) some set cc exists; hence (a,b)∈Z(a,b)\in Z if and only if (b,a)∈X(b,a)\in X.

Lemma R (testing one entry against a passenger). For every metatheoretic n≥1n\ge1, every ii with 1≤i≤n1\le i\le n and every class WW there is a class ZZ such that, for all sets x1,…,xn,yx_{1},\dots,x_{n},y, we have (⟨x1,…,xn⟩,y)∈Z(\langle x_{1},\dots,x_{n}\rangle,y)\in Z if and only if (xi,y)∈W(x_{i},y)\in W.

By induction on nn. For n=1n=1 we have i=1i=1 and ⟨x1⟩=x1\langle x_{1}\rangle=x_{1}, so Z=WZ=W works. Suppose the lemma holds for nn, and let 1≤i≤n+11\le i\le n+1. Write t=⟨x1,…,xn⟩t=\langle x_{1},\dots,x_{n}\rangle, a set; then (⟨x1,…,xn+1⟩,y)=((t,xn+1),y)=⟨t,xn+1,y⟩(\langle x_{1},\dots,x_{n+1}\rangle,y)=((t,x_{n+1}),y)=\langle t,x_{n+1},y\rangle.

Case i≤ni\le n. The induction hypothesis gives Z0Z_{0} with (t,y)∈Z0(t,y)\in Z_{0} if and only if (xi,y)∈W(x_{i},y)\in W. By (B5) applied to Z0Z_{0} there is Z1Z_{1} with ⟨a,b,c⟩=((a,b),c)∈Z1\langle a,b,c\rangle=((a,b),c)\in Z_{1} if and only if (a,b)∈Z0(a,b)\in Z_{0}. By (B7) applied to Z1Z_{1} there is ZZ with ⟨a,b,c⟩∈Z\langle a,b,c\rangle\in Z if and only if ⟨a,c,b⟩∈Z1\langle a,c,b\rangle\in Z_{1}. Then ⟨t,xn+1,y⟩∈Z\langle t,x_{n+1},y\rangle\in Z if and only if ⟨t,y,xn+1⟩∈Z1\langle t,y,x_{n+1}\rangle\in Z_{1}, if and only if (t,y)∈Z0(t,y)\in Z_{0}, if and only if (xi,y)∈W(x_{i},y)\in W.

Case i=n+1i=n+1. By (B5) applied to WW there is Z1Z_{1} with ⟨a,b,c⟩∈Z1\langle a,b,c\rangle\in Z_{1} if and only if (a,b)∈W(a,b)\in W. By (B6) applied to Z1Z_{1} there is ZZ with ⟨a,b,c⟩∈Z\langle a,b,c\rangle\in Z if and only if ⟨b,c,a⟩∈Z1\langle b,c,a\rangle\in Z_{1}. Then ⟨t,xn+1,y⟩∈Z\langle t,x_{n+1},y\rangle\in Z if and only if ⟨xn+1,y,t⟩∈Z1\langle x_{n+1},y,t\rangle\in Z_{1}, if and only if (xn+1,y)∈W(x_{n+1},y)\in W.

Lemma Q (one entry in a class). For every metatheoretic n≥1n\ge1, every ii with 1≤i≤n1\le i\le n and every class WW there is a class ZZ such that ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if xi∈Wx_{i}\in W, for all sets x1,…,xnx_{1},\dots,x_{n}.

By (B5) applied to WW there is W1W_{1} with (a,b)∈W1(a,b)\in W_{1} if and only if a∈Wa\in W. Lemma R applied to nn, ii and W1W_{1} gives Z0Z_{0} with (⟨x1,…,xn⟩,y)∈Z0(\langle x_{1},\dots,x_{n}\rangle,y)\in Z_{0} if and only if (xi,y)∈W1(x_{i},y)\in W_{1}, that is, if and only if xi∈Wx_{i}\in W. By (B4) applied to Z0Z_{0} there is ZZ with p∈Zp\in Z if and only if there is a set yy with (p,y)∈Z0(p,y)\in Z_{0}. Taking p=⟨x1,…,xn⟩p=\langle x_{1},\dots,x_{n}\rangle: p∈Zp\in Z if and only if there is a set yy with xi∈Wx_{i}\in W, which by (S) holds if and only if xi∈Wx_{i}\in W.

Lemma P (two distinct entries as a pair in a class). For every metatheoretic n≥1n\ge1, all distinct i,ji,j with 1≤i,j≤n1\le i,j\le n and every class WW there is a class ZZ such that ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if (xi,xj)∈W(x_{i},x_{j})\in W, for all sets x1,…,xnx_{1},\dots,x_{n}.

By induction on nn. For n=1n=1 there are no distinct i,ji,j and there is nothing to prove. Suppose the lemma holds for nn, let i≠ji\ne j be in 1,…,n+11,\dots,n+1, and write ⟨x1,…,xn+1⟩=(⟨x1,…,xn⟩,xn+1)\langle x_{1},\dots,x_{n+1}\rangle=(\langle x_{1},\dots,x_{n}\rangle,x_{n+1}).

Case i,j≤ni,j\le n. The induction hypothesis gives Z0Z_{0} with ⟨x1,…,xn⟩∈Z0\langle x_{1},\dots,x_{n}\rangle\in Z_{0} if and only if (xi,xj)∈W(x_{i},x_{j})\in W. By (B5) applied to Z0Z_{0} there is ZZ with (p,q)∈Z(p,q)\in Z if and only if p∈Z0p\in Z_{0}; with p=⟨x1,…,xn⟩p=\langle x_{1},\dots,x_{n}\rangle and q=xn+1q=x_{n+1} this is the claim.

Case j=n+1j=n+1, so i≤ni\le n. Lemma R applied to nn, ii and WW gives ZZ with (⟨x1,…,xn⟩,y)∈Z(\langle x_{1},\dots,x_{n}\rangle,y)\in Z if and only if (xi,y)∈W(x_{i},y)\in W; take y=xn+1y=x_{n+1}.

Case i=n+1i=n+1, so j≤nj\le n. Lemma A applied to WW gives W′W' with (a,b)∈W′(a,b)\in W' if and only if (b,a)∈W(b,a)\in W. Lemma R applied to nn, jj and W′W' gives ZZ with (⟨x1,…,xn⟩,y)∈Z(\langle x_{1},\dots,x_{n}\rangle,y)\in Z if and only if (xj,y)∈W′(x_{j},y)\in W', that is, if and only if (y,xj)∈W(y,x_{j})\in W; take y=xn+1y=x_{n+1}.

Reduction to class parameters. Let φ\varphi be as in the clause comprehension of the statement, and suppose some parameter is a set variable uu, assigned a set aa. By renaming bound variables (as in Stage (1a) below, with fresh set variables; equivalent under every assignment by elementary logic) we may assume that no quantifier of φ\varphi binds uu, so every occurrence of uu in φ\varphi is free. Choose a class variable UU occurring nowhere in φ\varphi and different from all parameters (the list of class variables is infinite by Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §symbols), and let φ′\varphi' result from φ\varphi by replacing every occurrence of uu by UU. An atomic part containing uu, such as u∈tu\in t, t∈ut\in u, u=tu=t or t=ut=u, becomes the same part with UU in place of uu, which is again an atomic formula by Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §formula, and no quantifier is added or changed. Hence φ′\varphi' is predicative, its free variables are those of φ\varphi with uu replaced by UU, and in particular its free set variables other than set-variable parameters are among x1,…,xnx_{1},\dots,x_{n}. The set aa is a class, by Classes, Membership, Sets and Proper Classes §set; assign it to UU. Read as in Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §reading, each atomic part of φ′\varphi' then says about aa exactly what the corresponding part of φ\varphi says, and the quantifiers of both formulas range over sets through the same set variables, none of which is uu or UU; so φ′\varphi' and φ\varphi express the same statement about x1,…,xnx_{1},\dots,x_{n}, aa and the remaining parameters. Doing this for each set-variable parameter in turn yields a predicative formula whose parameters are all class variables, distinct from x1,…,xnx_{1},\dots,x_{n}, and which, with each new class variable assigned the set assigned to the parameter it replaced, expresses the same statement as φ\varphi; a class ZZ obtained for it at these classes therefore satisfies the conclusion of the clause comprehension of the statement for φ\varphi. So for the proof of the clause comprehension of the statement we may, and through the end of Step 2 do, assume that Y1,…,YmY_{1},\dots,Y_{m} are all class variables, and we keep writing φ\varphi for the formula.

Step 1. Normal form.

Call a formula normal if it is predicative, is built by the rules of Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §formula using only ¬\neg, ∧\wedge and ∃\exists over set variables, every atomic part has the form s∈ts\in t with ss a set variable and tt either a set variable different from ss or a class variable, and no two occurrences of ∃\exists in it bind the same variable. Its degree is the number of occurrences of ¬\neg, ∧\wedge and ∃\exists in it. Abbreviations (∀\forall, ⇒\Rightarrow, ⇔\Leftrightarrow) are understood unfolded as in Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §formula.

Let φ\varphi be as in the clause comprehension of the statement, with the parameters Y1,…,YmY_{1},\dots,Y_{m} all class variables as arranged above. Since φ\varphi is predicative, no class variable is quantified in it, so every class variable occurring in φ\varphi is free and hence among Y1,…,YmY_{1},\dots,Y_{m}. We transform φ\varphi in three stages; in each, a fresh variable means a set variable occurring nowhere in the current formula, different from x1,…,xnx_{1},\dots,x_{n} and from every variable introduced before (the list of set variables is infinite by Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §symbols).

Stage (1a). Replace each part ∃v ψ\exists v\,\psi, one at a time, by ∃v′ ψ′\exists v'\,\psi', where v′v' is fresh and ψ′\psi' results from ψ\psi by replacing the free occurrences of vv by v′v'. By elementary logic (renaming of bound variables, with no capture since v′v' is fresh) the result is equivalent to the original under every assignment. Afterwards no two quantifiers bind the same variable and none binds any of x1,…,xnx_{1},\dots,x_{n}.

Stage (1b). Replace each atomic part s∈ts\in t in which ss is a class variable, or ss and tt are the same set variable, by ∃z (z=s∧z∈t)\exists z\,(z=s\wedge z\in t) with zz fresh. This is equivalent for every assignment of classes SS to ss and TT to tt (with S=TS=T if s,ts,t are the same variable), read as in Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §reading: if S∈TS\in T, then SS is a set by Elements Are Sets, and the Empty Set and the Pair of Two Sets Exist and Are Unique §elements, so the set z=Sz=S witnesses z=Sz=S and z∈Tz\in T; conversely, if some set zz satisfies z=Sz=S and z∈Tz\in T, then S∈TS\in T by substitutivity of equality (Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §symbols). The new part z∈tz\in t has a set variable zz on the left, different from tt.

Stage (1c). Replace each atomic part s=ts=t (with s,ts,t variables of any kind, including those created in (1b)) by ∀z (z∈s⇔z∈t)\forall z\,(z\in s\Leftrightarrow z\in t) with zz fresh. For classes S,TS,T assigned to s,ts,t: if S=TS=T then for every set zz, z∈Sz\in S if and only if z∈Tz\in T, by substitutivity of equality (Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §symbols); conversely, if this holds for every set zz then S=TS=T by the axiom of extensionality. The new atomic parts z∈sz\in s and z∈tz\in t have the set variable zz on the left, different from ss and tt. This stage creates no equality and no membership with a class variable on the left.

Each replacement substitutes for an atomic part a formula with the same free variables and with new quantifiers over fresh set variables only, so by the replacement property of elementary logic the whole formula stays equivalent to φ\varphi under every assignment, stays predicative, and keeps its free variables. The result φ∗\varphi^{*} is therefore a normal formula, equivalent to φ\varphi for all sets x1,…,xnx_{1},\dots,x_{n} and classes Y1,…,YmY_{1},\dots,Y_{m}, whose free set variables are among x1,…,xnx_{1},\dots,x_{n}, whose free class variables are among Y1,…,YmY_{1},\dots,Y_{m}, and none of whose quantifiers binds any of x1,…,xnx_{1},\dots,x_{n}.

Step 2. Induction on the degree.

Fix classes Y1,…,YmY_{1},\dots,Y_{m}. We prove, by strong induction on the metatheoretic number kk, the claim C(k)C(k): for every normal formula ψ\psi of degree kk whose class variables are among Y1,…,YmY_{1},\dots,Y_{m}, every n≥1n\ge1 and every list x1,…,xnx_{1},\dots,x_{n} of distinct set variables that contains every free set variable of ψ\psi and no variable bound by a quantifier of ψ\psi, there is a class ZZ with ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if ψ\psi, for all sets x1,…,xnx_{1},\dots,x_{n}. The induction hypothesis is used for all such nn and lists, not only the given one.

Degree 00, ψ\psi is s∈ts\in t with s,ts,t distinct set variables. Both are free, so s=xis=x_{i} and t=xjt=x_{j} with i≠ji\ne j. By (B1) there is EE with (a,b)∈E(a,b)\in E if and only if a∈ba\in b. Lemma P applied to nn, ii, jj and EE gives ZZ with ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if (xi,xj)∈E(x_{i},x_{j})\in E, that is, if and only if xi∈xjx_{i}\in x_{j}.

Degree 00, ψ\psi is s∈Yls\in Y_{l} with ss a set variable. Then s=xis=x_{i} for some ii, and Lemma Q applied to nn, ii and YlY_{l} gives the class.

Degree k≥1k\ge1, ψ\psi is ¬χ\neg\chi. The formula χ\chi is normal of degree k−1k-1 with the same free and bound variables, so the induction hypothesis gives Z0Z_{0} for χ\chi and the same list. By (B3) applied to Z0Z_{0} there is ZZ with p∈Zp\in Z if and only if p∉Z0p\notin Z_{0} for every set pp; taking the set p=⟨x1,…,xn⟩p=\langle x_{1},\dots,x_{n}\rangle, this holds if and only if χ\chi fails, that is, if and only if ψ\psi.

Degree k≥1k\ge1, ψ\psi is (χ∧θ)(\chi\wedge\theta). Both χ\chi and θ\theta are normal of degree less than kk, their free set variables are free in ψ\psi and their bound variables are bound in ψ\psi, so the same list serves for both; the induction hypothesis gives Z1Z_{1} for χ\chi and Z2Z_{2} for θ\theta. By (B2) applied to Z1Z_{1} and Z2Z_{2} there is ZZ with p∈Zp\in Z if and only if p∈Z1p\in Z_{1} and p∈Z2p\in Z_{2}; at p=⟨x1,…,xn⟩p=\langle x_{1},\dots,x_{n}\rangle this holds if and only if χ\chi and θ\theta.

Degree k≥1k\ge1, ψ\psi is ∃y χ\exists y\,\chi with yy a set variable. Since yy is bound in ψ\psi, it is not in the list, so x1,…,xn,yx_{1},\dots,x_{n},y are n+1n+1 distinct set variables. Every free set variable of χ\chi is free in ψ\psi or equals yy, so it lies in the extended list. Since ψ\psi is normal, yy is not bound again inside χ\chi, and the other bound variables of χ\chi are bound in ψ\psi, so the extended list contains no bound variable of χ\chi; and χ\chi is normal of degree k−1k-1. (If yy does not occur free in χ\chi nothing changes.) The induction hypothesis, applied to χ\chi with the list x1,…,xn,yx_{1},\dots,x_{n},y, gives WW with ⟨x1,…,xn,y⟩∈W\langle x_{1},\dots,x_{n},y\rangle\in W if and only if χ\chi, for all sets x1,…,xn,yx_{1},\dots,x_{n},y. By (B4) applied to WW there is ZZ with p∈Zp\in Z if and only if there is a set yy with (p,y)∈W(p,y)\in W. Since (⟨x1,…,xn⟩,y)=⟨x1,…,xn,y⟩(\langle x_{1},\dots,x_{n}\rangle,y)=\langle x_{1},\dots,x_{n},y\rangle, we get ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if there is a set yy with χ\chi, which by Formulas of the Language of Class Theory, Free Variables, Predicative Formulas and Abbreviations §reading is the reading of ψ\psi.

Conclusion of the clause comprehension of the statement. Given φ\varphi, the list x1,…,xnx_{1},\dots,x_{n} and classes Y1,…,YmY_{1},\dots,Y_{m}, Step 1 yields the normal formula φ∗\varphi^{*} for which the list satisfies the hypotheses of C(k)C(k), with kk the degree of φ∗\varphi^{*}. Step 2 gives ZZ with ⟨x1,…,xn⟩∈Z\langle x_{1},\dots,x_{n}\rangle\in Z if and only if φ∗\varphi^{*}, hence, by the equivalence of Step 1, if and only if φ(x1,…,xn,Y1,…,Ym)\varphi(x_{1},\dots,x_{n},Y_{1},\dots,Y_{m}), for all sets x1,…,xnx_{1},\dots,x_{n}.

Step 3. Uniqueness, the clause unique of the statement. Let n=1n=1 and let ZZ and Z′Z' both satisfy the condition. For every set x1x_{1}, x1∈Zx_{1}\in Z holds if and only if φ(x1,Y1,…,Ym)\varphi(x_{1},Y_{1},\dots,Y_{m}), if and only if x1∈Z′x_{1}\in Z'. By the axiom of extensionality, which requires agreement on sets only, Z=Z′Z=Z'.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…