Formalizing set theory in diagonal-free cylindric algebras, searching for the weakest logic with Gdels incompleteness property.

o Draft version!
Hajnal Andrka and Istvn Nmeti e a e 2011. July

Abstract We show that rst-order logic can be translated into a very simple and weak logic, and thus set theory can be formalized in this weak logic. This weak logical system is equivalent to the equational theory of diagonalfree 3-dimensional cylindric algebras, i.e., Boolean algebras with three commuting complemented closure operators. Equivalently, set theory can be formulated in propositional logic with 3 commuting S5 modalities (i.e., in the multimodal logic [S5,S5,S5]). There are many consequences, e.g., free nitely generated Df 3 s are not atomic and [S5,S5,S5] has Gdels o incompleteness property.

Introduction

Tarski in 1953 [13, 14] formalized set theory in the theory of relation algebras. Why did he do this? Because the equational theory of relation algebras (RA) corresponds to a variable-free logic, in other words, to a propositional logic. Tarski got the surprising result that a propositional logic is strong enough to express mathematics, to be the playground for mathematics. The classical view before this result was that propositional logics in general were weak in expressive power, decidable, uninteresting in a sense. By using the fact that set theory can be built in it, Tarski proved that the equational theory of RA is undecidable. This was the rst propositional logic shown to be undecidable. This is why the title of the book [15] is Formalizing set theory without variables.
This contains full proofs for the two main theorems. We make this preprint available in this draft version because so many people expressed strong interest in the proofs. Research supported by the Hungarian National Foundation for scientic research grants T81188.

From the above it is clear that changing RA in Tarskis result to a weaker class of algebras is an improvement of the result and it is worth doing. For more on this see Tarski-Givant [15, pp.892 904 and footnote 17 on p.90]. Relation algebras are halfway between CA3 and CA4 , the classes of 3-dimensional and 4-dimensional cylindric algebras, respectively. We sometimes jokingly say that RA is CA3.8 . Why is RA stronger than CA3 ? Because, the relation algebra reduct of a CA3 is not necessarily an RA, e.g., associativity of relation composition can fail in the reduct. See [5, sec 5.3], and for more in this line see Nmeti-Simon [9]. Why is CA4 stronger than RA? Because RaCA4 RA. e However, RA = SRaCA4 (Maddux, see [5, sec 5.3]), so the equational theories of RA and RaCA4 coincide. Thus Tarski formulated Set Theory, roughly, in CA4 . Draft paragraph. Finite variable fragment hierarchy of FOL. What it says about FOL. Monks result: For all nite n there is a 3-variable formula which is valid but which can be proved (in FOL) with more than n variables only. Intuitively this means that during the proof there are steps when we have to use n independent data (stored in the n variables as in n machine registers). Recent solution of Problem 2.12, related. For example, the associativity of relation product (of binary relations) as formalized in the Ra-reduct of a CAn can be expressed with 3 variables but can be proved only with 4 variables. L3 is essentially incomplete, it does not have a nite Hilbert-style inference system. The inference system |3 belonging to L3 expresses CA3 . Tarskis main idea in [15] is to use pairing functions to form pairs, and so to store two pieces of data in one register. He used this technique to translate usual innite-variable rst-order logic FOL in the three variable fragment of it. From then on, he used that any three-variable FOL-formula about binary relations can be expressed by an RA-equation, [5, sec 5.3]. He needed two registers for storing the data belonging a binary relation and he had one more register for making computations belonging to a proof. Nmeti [6], [7] formalized set theory in CA3 (hence also in the class SA e of semi-associative relation algebras), thus improving Tarskis result. The main idea for this improvement was using the paring functions to store all data always, during every step of a proof, in one register only and so one got two registers to work with in the proofs. In this approach one represents binary relations as unary ones (on pairs). For the execution of this idea see sections 3-5 of the present paper. One of the uses of equality in FOL is that it can be used to express (simulate) substitutions of variables. The reduct SCA3 of CA3 forgets equality dij but retains substitution in the form of the term-denable operations si . The logic j belonging to SCA3 is weaker than 3-variable fragment of FOL. Zaln Gyenis [4] a improved parts of Nmetis result by using SCA3 in place of CA3 . e Df 3 is much weaker than CA3 . Without equality or substitutions, if one has only binary relations, one cannot really use the third variable for anything. Therefore we need at least one ternary relation symbol in order to use the third variable (which we have to use in a formalization of set theory, because the twovariable fragment is decidable). Therefore, while in formalizing set theory in the three-variable fragment of FOL (in CA3 ) we could do with one binary relation 2

symbol, we did not have to change vocabulary during the formalization, in the present equality- and substitution-free case we have to change vocabulary, and we have to pay attention to this new feature of the translation function. What does it mean to formalize set theory? Why set theory? Have we learnt anything about set theory?

A simple logical system: three variable logic without equality or substitutions

The language of this simple system contains three variable symbols, x, y, z, one ternary relational symbol P , and there is one atomic formula, namely P (x, y, z). The logical connectives are , , x, y, z. We denote the set of formulas by Fmd 3 . We will use the derived connectives , , , , too. The proof system d | is a Hilbert style one with the following logical axioms and rules. The logical axioms are the following. Let , Fmd 3 and v, w {x, y, z}. ((1)) , if is a propositional tautology. ((2)) v( ) (v v). ((3)) v. ((4)) vv v. ((5)) v( ) (v v). ((6)) vv v. ((7)) vw wv. The inference rules are Modus Ponens ((MP), or detachment), and Generalization ((G)). This proof system is a direct translation of the equational axiom system of Df 3 . Axiom ((2)) is needed for ensuring that the equivalence relation dened d on the formula algebra by | be a congruence with respect to (w.r.t.) the operation v. It is congruence w.r.t. the Boolean connectives , by axiom ((1)). Axiom ((1)) expresses that the formula algebra factorized with is a Boolean algebra, axiom ((5)) expresses that the quantiers v are operators on this Boolean algebra (i.e., they are additive), axioms ((3)),((4)) express that these quantiers are closure operations, axiom ((6)) expresses that they are complemented closure operators (i.e., the negation of a closed element is closed again). Together with ((5)) they imply that the closed elements form a Boolean algebra, and hence the quantiers are normal operators (i.e., the Boolean zero is a closed element). Finally, axiom ((7)) expresses that the quantiers commute with each other. d We dene Ld3 as the logic with formulas Fmd 3 and with proof system | . The logic Ld3 inherits a natural semantics from rst-order logic (FOL). The d proof system | is sound with respect to this semantics, but it is not complete. Moreover, there is no nite Hilbert-style inference system which would be complete and sound at the same time w.r.t. this semantics (because the quasi-equational theory of RDf 3 is not nitely axiomatizable, see [5] and [2]). We note that in the above system, axiom ((6)) can be replaced with the

following ((8)): ((8)) v( v) (v v). Restricted 3-variable FOL is introduced in [5, Part II, p.157], with proof sys tem |r . If we restrict this system |r to formulas not containing the equality = then we get a system equivalent to our Ld3 . Lets call this system restricted 3-variable FOL without equality. That is, the formulas are those of restricted 3-variable FOL which contain no equality, and we leave out from the axioms of |r the axioms which contain equality. This way we get a proof system with Modus Ponens and Generalization as deduction rules and the following axioms: ((V1)) ((V2)) ((V3)) ((V4)) , if is a propositional tautology. v( ) (v v). v . v, if v does not occur free in .

Lets call this1 equality-free |r . Rule ((V4)) in this system essentially uses variables in its using the notion of free variables of a formula. On the other d hand, no axiom in | needs to use the structure of a formula occurring in d a rule, it is essentially variable-free. So, an advantage of | to equality-free |r is that it is more algebraic, more like propositional logic. On the other hand, equality-free |r contains fewer axioms (it contains only ((V1))-((V4)) as axioms). The logic Ld3 has a neat modal logic form: three commuting S5 modalities. This is denoted as [S5,S5,S5], see [3, p.379, lines 15-20]. One can present this logic in yet one dierent form: Equational logic as the background logic, and the dening axioms of Df 3 as logical axioms.

The following theorem says that this simple logic Ld3 is strong enough for doing all of mathematics in it. It says that we can do set theory in Ld3 as follows: in place of formulas of set theory we use their translated versions Tr() in d Ld3 , and then we use the proof system | of Ld3 between the translated formulas in place of the proof system of FOL between the original formulas of set theory. Moreover, for sentences in the language of set theory, and Tr() mean the same thing (are equivalent) modulo a bridge between the two languages. We need this bridge because the language L of set theory contains only one binary relation symbol and equality, and the language Ld3 contains only one ternary relation symbol P . When f : A B is a function and X A d then f (X) = {f (a) : a X} denotes the image of X under this function f . Theorem 2.1. (Formalizability of set theory in Ld3 ) There is a recursive translation function Tr from the language L of set theory into Ld3 for which the following are true for all sentences in the language of set theory:
1 We note that we also omitted ((V9)) of [5] because v abbreviates v in our approach, so ((V9)) is not needed.

(i) ZF |=
d

d Tr(ZF ) | Tr().

(ii) ZF + |= Tr(),

where

= xyz[(P (x, y, z) (x = y = z (x, y))]. Theorem 2.1 is proved in section 5. The next theorem is a partial completeness theorem for Ld3 . Theorem 2.2. (Partial completeness theorem for Ld3 ) There is a recursive subset K Fmd 3 and there is a recursive function tr mapping all FOL-formulas into K such that the following are true: (i) |= (ii) |= i i
d |

for all K. for all FOL sentences .

|= tr()

d According to the above theorem, the proof system | is complete within K. But is K big enough? Yes, we can prove any valid FOL-formula by translating d d it into K and then proving the translated formula by | . We know that | is not strong enough to prove all valid Fmd 3 formulas. However, we can formulate each sentence in a slightly dierent form, namely as tr() so that this version d of can be now proved by | i it is valid.

Theorem 2.3. (Gdel style incompleteness theorem for Ld3 ) There is a formula o Fmd 3 such that no consistent recursive extension T of is complete. Discussion 2.1. In Theorems 2.1-2.3, at least one at least ternary relation symbol P is needed in the target-language Ld3 , the axiom of commutativity d d ((7)) is needed in the proof system | (because omitting ((7)) from | results decidability of the so obtained proof system, see [7]). We do not know whether complementedness of the closure operators ((6)) is needed or not. Also, two variables do not suce because the satisability problem of the two-variable fragment of FOL is decidable.

Finding QRA-reducts in Df 3

In this section we begin the proof of Theorem 2.1. We show that every Df 3 contains lots of QRAs in them. We do this by dening relation algebra type operations in the term language of Df 3 and proving that these operations form QRAs in appropriate relativizations. Since QRAs are representable, this will amount to a partial representation theorem for Df 3 s, and to partial completeness theorem for Ld3 . We will work in Ld3 in place of Df 3 . There will be parameters in the denitions to come. These will be formulas in Fmd 3 , namely xy , xz with free variables {x, y} and {x, z} respectively, together with two other formulas p0 , p1 with free variables {x, y}. Thus, if you choose xy , xz , p0 , p1 with the above specied free variables then you will arrive at a QRA-reduct of any Df 3 corresponding to these. We get the QRA-reduct

by assuming some properties of the meaning of these formulas, this will be expressed by a formula Ax. In section 5 then we will choose these parameters so that they t set theory, which means that the formula Ax built up from them is provable in set theory. Intuitively, the formulas xy , xz stand for equality x = y, x = z and p0 , p1 will be arbitrary pairing functions. So, choose formulas xy , xz , p0 , p1 with the above specied free variables arbitrarily, they will be parameters of the denitions to come. To simplify notation, we will not indicate these parameters. We now set ourselves to dening the above relation algebra type operations on Fmd 3 . To help readability, we often write just comma in place of conjunction in formulas, especially when they begin with a quantier. E.g., we write x(, ) in place of x( ). Further, True denotes a provably true formula, say True = xy xy . First we introduce notation to support the intuitive meaning of the parameters xy , xz as equality. Denition 3.1. (Simulating equality between variables) . d x = y = xy , . d x = z = xz , . d . . y = z = x(x = y x = z), . d . y = x = x = y, . d . z = x = x = z, . d . z = y = y = z, . d x = x = True, . d y = y = True, . d z = z = True. Denition 3.2. (Simulating substitution with (simulated) equality) . ( x, z) = y(y = z ), ( ) d . ( y, z) = x(x = y ( x, z) ), ( ) ( ) d . ( y, x) = z(x = z ( y, z) ), ( ) ( ) d . ( z, x) = y(y = z ( y, x) ), ( ) ( ) d . ( z, y) = x(x = z ), ( ) d . ( x, x) = y(x = y ), ( ) d . ( y, y) = x(x = y ), ( ) d . ( z, z) = x(x = z ( x, x) ). ( ) ( )
d d

( x, y) = , ( )

Remark 3.3. In FOL, ( y, z) is semantically equivalent with the formula we ( ) get from the formula by replacing x, y with y, z everywhere simultaneously, when xy , xz mean true equality. This is Tarskis trick to simulate substitu6

tions. Etc. Merry-go-round semantically valid but not provable, even with true equality (i.e., in CA3 ). Next we introduce notation supporting intuition about the pairing functions p0 , p1 . First we dene some auxiliary formulas. We will use the notation 2 = {0, 1}, to make the text shorter. Let 2 denote the set of all nite sequences of 0, 1 including the empty sequence <> as well. If i, j 2 then ij denotes their concatenation usually denoted by ij, and |i| denotes the length of i. Further, if k 2, then we write k instead of < k > for the sequence < k > of length 1. Accordingly, 00 denotes the sequence < 0, 0 >. . We are going to dene Fmd 3 -formulas ui = vj for u, v {x, y, z} and i, j 2 . . The intuitive meaning of ui0 ...in = vj0 ...jk is that if p0 , p1 are partial functions then pin . . . pi0 u = pjk . . . pj0 v. As usual in the partial algebra literature, the equality holds if both sides are dened and are equal. E.g., the intuitive meaning . of x0 = y01 is that all of p0 x, p0 y, p1 p0 y exist and p0 x = p1 p0 y. Denition 3.4. (Simulating projections) Let {u, v, w} = {x, y, z}, i, j 2 and k 2. . d . (u<> = v<> ) = u = v, . d (uk = v<> ) = pk ( u, v) , ) . d . (uik = v<> ) = w(ui = w<> , pk ( w, v) ) if i =<>, ) . d . . (ui = vj ) = w(ui = w<> , vj = w<> ) if j =<>, . d . . (xi = xj ) = y(x = y, xi = yj ), . d . . (yi = yj ) = x(x = y, xi = yj ), . d . . (zi = zj ) = x(zi = x<> , zj = x<> ). . . We will omit the index <> in formulas ui = vj , i.e., we write ui = v and . . . u = vi for ui = v<> and u<> = vi resp. if i 2 . So far we did nothing but introduced notation supporting the intuitive meaning of the parameters xy , xz , p0 , p1 as equality and partial pairing functions. Almost any of the concrete formulas supporting this would do, we only had to d x one of them since our proof system | is very weak, it would not prove equivalence of most of the semantically equivalent forms. Now we write up a statement Ax about the parameters using the just introduced notation. Let d H = {i 2 , |i| 3}. Denition 3.5 (pairing axiom Ax). We dene Ax Fmd 3 to be the conjunction of the union of the following nite sets (A1),...,(A4) of formulas: . . . (A1) {ui = vj , vj = wk ui = wk : u, v, w {x, y, z}, i, j, k H} . . . (A2) {ui = vj , uik = uik uik = vjk : u, v {x, y, z}, ik, jk H, k 2} . . . . (A3) {ui = ui , vj = vj w(w0 = ui , w1 = vj ) : u, v, w {x, y, z}, w / {u, v}, i, j H}

. (A4) {w u = w : u, w {x, y, z}}. We are ready to dene our relation-algebra type operations on Fmd 3 . They will have the intended meaning on formulas with one free variable x. Let Fmd 1 3 denote the set of formulas in Fmd 3 with at most one free variable x. Denition 3.6 (relation algebra reduct Dra of Fmd 3 ). Let , Fmd 3 . pair = yp0 yp1 . d . ui = x(x = ui , )
d

if u {y, z} and i 2 . see Figure 3,

d . . . = y(y0 , y1 , x0 = y00 , y01 = y10 , y11 = x1 ), . . d = y(y, y0 = x1 , y1 = x0 ), d . 1 = x0 = x1 , d = pair , d + = . d

d Dra = { Fmd 3 : Ax | 1 for some Fmd 1 }, 3

Dra = Dra, +, , , , 1 .

Figure 3. Illustration of

In the denition of Dra, intuitively 1 means that we close the meaning of with the equivalence relation of equivalent pairs. I.e., let us dene x d . . y = x0 = y0 , x1 = y1 , then 1 means closed with this equivalence relation. If we assume uniqueness of pairs (like in SAx later) then we can write pair in place of 1 in the denition of Dra. The next theorem is the heart of formalizing set theory in Fmd 3 . Let us dene the equivalence relation T on Fmd 3 by
d T = T | ,

where T Fmd 3 . 8

 Theorem 3.7. (i) Dra is an algebra, i.e., the operations +, , , , 1 do not lead out of Dra.

(ii) Dra { : , Fmd 1 }. 3 (iii) Dra/ Ax is a relation algebra. . . (iv) The images of the formulas x1 = x00 and x1 = x01 form a quasi-projection pair in Dra/ Ax . Proof. The proof of the analogous theorem in [7, Thm.9, p.43] goes through with some modications. We indicate here these modications. The proof explanations (CA),. . . , (KV) introduced on [7, p.44] can be used in our proof, too, except that we always have to check whether the explanation uses only the Df-part of (CA). If it uses axiom C7 of CA, then we have to give an alternate proof. Let (Df) denote the part of (CA) which does not use C7. We can use (UV) because it can be derived from Ax and (Df). Of course, throughout . we have to change = to =. We now go through the proof given in [7, pp.46-64] and indicate the changes needed for our proof. All the statements on pp.48-54 beginning with (A1) and ending with (S13) follow from Ax and (Df). In fact, we could just add these statements to Ax since they do not contain formula-schemes denoting arbitrary formulas (such as, e.g., (0) on p.54 does), and so they amount to nitely many formulas only. Because of this, we do not indicate the changes needed in the proofs of these items. We have to avoid statement (0) at the end of p.54 by all means, because it uses axiom C7 of (CA) essentially. Fortunately, we do not really use (0) in the proof, changing to x in some steps will suce for eliminating (0). The proof of (2/a) on p.55 has to be modied, and it can be done as follows. d . Recall from [7] that has at most one free variable x, and u = x(x = u, ) d . when u is dierent from x while x = y(y = x, y). . d Assume x {u, v}. The proof for u, u = v | v is as on [7, p.55]. Then / . d x, x = z | by denition of x . . d y(y = x, y), x = z | . . d y(y = x, y, x = z) | by Ax . d y(y = z, y) | by the case when x {u, v} / d y(z) | z. . d x, x = y | by Ax . . d z(z = x, x, x = y) | by Ax . . d z(z = y, x, x = z) | by the previous case . d z(z = y, z) | by the case when x {u, v} / y.

. d y, y = x | . d y(y = x, y) | x.

by denition of x

. d z, z = x | by Ax . . d y(y = z, z = x, z) | . . d y(y = x, z = y, z) | . d y(y = x, y) | x.

by Ax by the case when x {u, v} /

On p.58, in the last line of the proof of (9) we have to write x in place of . Similarly, on p.59, in lines 7 and 8 we have to change to x and then we can cross out reference to (0). Thats all! QED

Finding CA-reducts in Df 3

Simon [12] denes a CAn -reduct in every QRA, for all n , and also proves that these reducts are representable. We will use, in this paper, the CA3 -reduct of our QRA dened in the previous section, Dra/ Ax . We will use the following stronger form of Ax, just for convenience: Denition 4.1. [strong paring axioms SAx] We dene SAx Fmd 3 to be the conjunction of the union of the nite sets (A1),...,(A4) of Def.3.5 together with the following: . . . . . (A5) { x0 = y0 , x1 = y1 x = y, x0 = x0 x1 = x1 }. The formulas in (A5) express that pairs are unique, and that p0 is dened exactly when p1 is dened. We use (A5) for convenience only, this way formulas will be shorter. We could omit (A5) on the expense that formulas will be longer and more complicated. If we assume SAx then pair Dra for all Fmd 1 . 3 Every QRA has a CA3 -reduct, which is representable, see Simon [12]. The following denition is recalling this CA3 -reduct from [12] in our special case of Dra/ SAx . The denition below is simpler than in [12] because we will assume uniqueness of pairs in SAx, which is not assumed in [12]. We will use the abbreviation d . xi = y(y = xi , y), when Fmd 1 and i 2 . 3
d Beware: | x usually is not true for Fmd 1 . (For the denition of 3 yj see Def.3.6.)

Denition 4.2. [cylindric reduct Dca of Fmd 3 ] d . Triplet = x11 = x11 , see Figure, d d d (0) = 0, (1) = 10, (2) = 11, and for all i, j < 3 and , Fmd 1 3 d . Ti = Tripletx0 Tripletx1 {x0(j) = x1(j) : i = j < 3}, 10

ci = Ti , d . dij = Tripletx1 x1(i) = x1(j) , = Tripletx1 , d + = , d d Dca = { Fmd 3 : SAx | x1 Tripletx1 Dca =
d d

for some Fmd 1 }, 3

Dca, +, , ci , dij : i, j < 3 .

Theorem 4.3. The set Dca is closed under the operations dened in Def.4.2 and the algebra Dca/ SAx is a representable CA3 . Proof. We show that the algebra Dca/ SAx is the CA3 -reduct of the quasid . projective relation algebra Dra/ SAx , as dened in [12]. Let p = x1 = x00 , d . q = x1 = x01 . In the following we will omit referring to SAx , so we will look at p as an element of Dra/ SAx , while only p/ SAx is such. Let
d . e = Tripletx0 x0 = x1 , d . (i) = Tripletx0 x1 = x0(i) ,

d . (i) = Tripletx0 Tripletx1 x0(i) = x1(i) .

Now, one can show that e is the same as (3) in [12, Def.3.1], i.e.,
d SAx | e (qqqq 1).

Similarly, one can show that (i) , (i) are the same as the ones in [12], and Triplet is the same as 1(3) in [12, Def.3.1], e.g.,
d SAx | (1) (eqp), d SAx | (1) ((1) (1) ), d SAx | Triplet (paire).

Thus, our reduct is the same as the 3-reduct dened in [12], and then we can use [12, Thm.3.2,Thm.5.2]. QED

Formalizing set theory in Ld3

The 3-variable restricted fragment L3 of L is dened as follows. Language: three variable symbols, x, y, z, one binary relational symbol , so there is one atomic formula, namely (x, y). Logical connectives: , , x, y, z and u = v for u, v {x, y, z}. We denote the set of formulas by Fm 3 . Derived connectives are , , , . The proof system |3 of L3 is a Hilbert style one dened in [5, Part II, p.157]. The word-algebra of L3 is denoted as Fm3 , it is the absolutely free CA3 -type algebra generated by the formula (x, y). In this section we prove Theorem 2.1 stated in section 2. As a rst step, we will dene a translation function h from Fm 3 to Fmd 3 . This will be analogous 11

to the one dened in [7], but here a novelty is that the vocabulary of Fm 3 is dierent (disjoint) from that of Fmd 3 , and we will have to pay attention to this dierence. Namely, Fm 3 contains one binary relation symbol and equality u = v for u, v {x, y, z} while Fmd 3 contains one ternary relation symbol P and does not contain equality. The formula introduced in section 2 bridges the dierence between the two languages. Namely, is stated in a language which contains both Fm 3 and Fmd 3 and is a denition of P in Fm 3 , so it leads from Fm 3 to Fmd 3 . But is a two-way bridge, because xy x = y |= [(x, y) zP (x, y, z)][x = y = z (P (x, y, z)zP (x, y, z))] , so the equivalent form of provides a denition of equality = and in Fmd 3 , and thus it is a bridge leading from Fmd 3 to Fm 3 . As usual, we begin with denitions. To start, we work in Fm 3 . Recall the concrete denitions of p0 , p1 , and from [7, p.71, p.35]. Dene + = xy[z(p0 ((x, z)), p0 ((y, z))), z(p1 ((x, z)), p1 ((y, z))) x = y] x(yp0 (x, y) yp1 (x, y)). Then p0 , p1 , , + are formulas of Fm 3 . We note that the notation ((u, v)) in . [7] is the same as our ( u, v) introduced in Def.3.2, except that instead of x = y ( ) etc. the notation ((u, v)) uses real equality x = y etc. available in Fmd 3 . Next, we work in Fmd 3 and we get the versions of these formulas that the denition of P provides. We will write out the details. First we x the parameters xy , xz , p0 , p1 occurring in the formula SAx of Fmd 3 . Denition 5.1. [xing the parameters xy , xz of Fmd 3 ] E = z P (x, y, z), D = P (x, y, z) E, xy = zD, xz = yD. The next denition xes the parameters p0 , p1 of Fmd 3 . To distinguish them from their Fm 3 -versions, we will denote them by p0 , p1 . The denition below is a repetition of the denition of the corresponding formulas on p.71 of [7] such that we write E and the above dened concrete xy , xz in place of and x = y, x = z. We will use the notation introduced in section 2 and we will use notation to support set theoretic intuition. Thus, if we introduce a formula denoted as, . . say, x = {y}, then u = {v} denotes the formula ( u, v) (see Denitions 3.2,3.1). ( ) Below, op abbreviates ordered pair (to distinguish it from the formula pair dened earlier.). Denition 5.2. [xing the parameters p0 , p1 of Fmd 3 ] uv = E( u, v) , ( )
d d d d d d

for u, v {x, y, z},

12

. d . x = {y} = z(zx z = y), d . {x}y = z(z = {x}, zy), . d . . x = {{y}} = z(z = {y}, x = {z}),
d . op(x) = yz({z}x y = z) d

x y = z(xz, zy),

. yz[(y x, {y}x, z x, {z}x) y = z], yz(yx zy),

p0 = op(x) {y}x, d . p1 = op(x) [x = {{y}} (y x, {y}x)]. It is not hard to check that (1) , xy x = y |= Ax, + SAx.

We are ready to dene our translation mapping h from Fm 3 to Fmd 3 . For a formula Fmd 2 and i, j 2 we dene 3
d . . ( xi , xj ) = yz(y = xi , z = xj , ( y, z) ). ( ( )

Denition 5.3. [translation mapping h] (i) h : Fm 3 Fmd 1 is dened by the following: 3 h is a homomorphism from Fm3 into Dca such that d h ((x, y)) = E( x1(0) , x1(1) ) . ( (ii) SAx = SAx x(Tripletx1 h ( + )). (iii) We dene the mapping h : Fm 3 Fmd 0 as 3 h() = x(SAx Tripletx1 h ()).
d We say that a translation function f is Boolean preserving w.r.t. | i for d all sentences , Fmd 3 we have that | f ( ) (f () f ()) and d | f ( ) (f () f ()).

Theorem 5.4. Let be a sentence and T be a set of sentences of Fm 3 . Then the following are true.
d (i) T { + } |= h(T ) | h.

(ii) + |= h().
d (iii) h is Boolean preserving and SAx | h() h().

Proof. Proof of (ii): Let M be a model of + . Let V Val =

d V

= {x, y, z} and

M , the set of evaluations of the variables in M. Now, by M |= +

13

we have that pi and pi have the same meaning in M, and they form a pair of pairing functions. Thus one can show by induction that (2) M |= (ui = vj )[k] i . M |= (ui = vj )[k] i k(u)i = k(v)j in M.

We say that a M is a triplet i a11 is dened. If a is a triplet, then we assign an evaluation val(a) Val to a such that val(a) assigns to x, y, z the elements a(0) , a(1) , a(2) respectively. We will prove by induction the following statement: For all Fm 3 and k Val we have (3) M |= h ()[k] i k(x)1 is a triplet and M |= [val(k(x)1 )] .

By (3) and M |= + then we have M |= SAx . Thus, to show M |= h it is enough to show M |= x(Tripletx1 h ). When is a sentence, this follows from (3). This nishes the proof of (ii). Proof of (iii): The proofs of (4) and (6) in [7, p.73], which prove that is Boolean-preserving w.r.t. |3 , work for showing that h is Boolean preserving d , because h has the same structure as . Similarly, the proof of (5) w.r.t. | in [7, p.73] is good for proving the second statement of the present (iii). Proof of (i): First we prove (i) for the special case when T is the empty set. Let d be a sentence of Fm 3 , we want to prove | h() implies + |= . So, assume d d d that | h, i.e., | x(SAx Tripletx1 h ). Then SAx | x(Tripletx1 h ( + ) h ()), i.e., (4)
d
d SAx | h ( + ).

Recall that h is a homomorphism from Fm3 to Dca/ SAx , and the latter is a representable CA3 . Let = + . By (4) we have that the image of is not 1 under h , therefore there is a homomorphism g from Dca/ SAx to a cylindric set algebra C such that the image of h under g is not 1. Let f = gh , then (5) f : Fm3 C
d d

and

f () = 1.

Let U be the base set of C, let R = { s0 , s1 : s f ((x, y))} and dene the model M as U, R . Then for all Fm 3 and s U 3 we have that M |= [s] i s f (). Thus M|= by (5), and we are done with showing + |= . d In the other direction, we have to show that | h implies + |= . By d soundness of the proof system | we have |= h, then by (ii) we have + |= . Since P does not occur in and in + , this means that + |= and we are done. Next, assume that T is a set of sentences of Fm 3 . We want to show that d T { + } |= i h(T ) | h(). Now, T { + } |= i (by compactness of Fm 3 ) T0 { + } |= for some nite T0 T , i 14

 + |= T0 for some nite T0 T , i(by rst part of (i)) d | h( T0 ) for some nite T0 T . Then, by Boolean preserving of h d | h(T0 ) h(), and so by Modus Ponens we get d d h(T0 ) | h(). Conversely, from this we get by the soundness of | that (6) h(T0 ) |= h().

From here on we have to deal with the dierence between the two languages. Let M be an arbitrary model of L3 , so M contains one binary relation, say d M = M, M . Dene P M according to the denition , i.e., P M = { a, a, a : a M } { a, b, c M M M : a, b M }. Let M+ = be the expansion of M with this new relation, and let M reduct of this expansion to the language Ld3 . Assume now M |= T0 { + }. Then + M |= T0 { + }. Thus by (ii) we have that M+ |= h(T0 ), then M |= h() by (6). Thus M+ |= h + , so by (ii) M+ |= , and so M |= and thus, since M was an arbitrary model of L3 T0 { + } |= and we are done. QED
d d

M, M , P M

= M, P M be the

We are almost done with proving Thm.2.1, all we have to do is to use a connection between L and L3 established in [7], [6]. Proof of Thm.2.1: By [7, Lemma 3.1, p.35], or by [6, Lemma 2.2, Remark 2.4, p.25, p.30] there is a recursive function f : Fm 2 Fm 2 for which 3 (7) |= f () , f () = f (), f ( ) = f () f (),

for all sentences , of L . Take such an f and extend it to Fm by letting d f () = f ( ) where is the universal closure of (if n is the smallest number such that the free variables of are among v0 , ..., vn , then is v0 . . . vn ). Then f has the same properties as f and it is dened on the whole of Fm , not only on Fm 2 . Dene Tr : Fm Fmd 3 by Tr() = h(f ()) for all Fm . We show that this translation function satises the requirements of Thm.2.1. First, Tr is recursive because both f and h are such. We dened the parameters p0 , p1 so that (8) ZF |= +
d

holds. Thus ZF |= ZF + + |= f () by the chosen properties of f , and then ZF + |= ZF + + |= hf by Thm.5.4(ii), so ZF + |= Tr 15

for all sentences of L . This is Thm.2.1(ii). To prove Thm.2.1(i), rst we show (9) ZF |= f (ZF ) + + |= f ().

Indeed, assume ZF |= and let M |= f (ZF ) + + . Then M |= ZF by (7), so M |= + + by our assumption ZF |= , so M |= f () by (7). This shows f (ZF ) + + |= f (). Conversely, assume now the latter, and we want to prove ZF |= . Let M |= ZF , then M |= ZF + + by (8), thus M |= f (ZF ) + + by (7), then M |= f () + + by our assumption, and then M |= by (7) again. We have shown that (9) holds. d By combining this (9) with Thm.5.4(i) we get ZF |= i Tr(ZF ) | Tr which is Thm.2.1(i). Later, in section 6, we will also need the following: (10) Tr is Boolean preserving and
d SAx | Tr() Tr.

Indeed, this follows from Thm.5.4(iii) and from (7): Let , Fm 0 . Then d | Tr( ), i by the denition of Tr d | hf ( ), i by (7) d h(f f ), i by Thm.5.4(iii) | d | hf hf , i by the denition of Tr d Tr Tr. | Similarly, d | Tr( ), implies by the denition of Tr d | hf ( ), implies by (7) d | h(f f ), implies by Thm.5.4(iii) d | hf hf , implies by the denition of Tr d | Tr Tr. d d Finally, SAx | h(f ) h(f ), by Thm.5.4(iii), so SAx | h(f ()) d h(f ) by (7), and then SAx | Tr() Tr(), as was to be shown. QED(Thm.2.1)

Free algebras

In this section we prove that the one-generated free Df 3 is not atomic. Theorem 6.1. The one-generated free Df 3 is not atomic. Proof. It is enough to show that the zero-dimensional part of the free Df 3 is not atomic by [5, 1.10.3(i)]: (11) ZdFr1 Df 3 is not atomic implies that Fr1 Df 3 is not atomic.

Let us dene as , i.e. i 16

d | .

It is easy to show that (12) (13) Fr1 Df 3 is isomorphic to Fmd3 / and ZdFr1 Df 3 is isomorphic to Fmd0 / . 3

There is a non-separable formula Fm 0 which is consistent with + (with our concrete pairing formulas), by [7, pp.69-71] (or equivalently by [6, Lemma 2.7, p.34]). Dene = SAx Tr. Then Fmd 0 . We will show that there is no atom below / and the latter 3 is nonzero in Fmd 0 / . Assume the contrary, i.e., that 3 (14) / is an atom below /
d

and we will derive a contradiction. Let

d T = { Fm 0 : | Tr}.

Then T Fm 0 . We will show that T is recursive and it separates the conse quences of from the sentences refutable from which contradicts the choice of , namely that is inseparable. From now on let Fm 0 be arbitrary. Now, / being an atom implies (15)
d | Tr

d | Tr.

From (15) we get that both T and the complement of T are recursively enumerable, so T is recursive. Next we show (16) |= implies T.

Indeed, assume that |= . Then |= . Then, in particular, + |= , d d and so | Tr( ) by Thm.2.1(i). Then | Tr() Tr by (10). By Modus d d Ponens then SAx + Tr() | Tr, i.e., | Tr. By (14) we have (17)
d |

d so we have | Tr, i.e., T as was desired. Next we show

(18)

|=

implies

T. /

Indeed, assume |= . Then () T by the previous case (16), and this means d | Tr(). Now, by (17) and the denition of we have d d | SAx , and thus by | Tr() and (10) d | Tr(). Thus by (15) we get d | Tr, by the denition of T then T. / 17

By the above we have shown that there is no atom below / . It remains to show that the latter is nonzero. This follows from the fact that + has a model. Let M be such that M |= + . Expand this model with P M so that M+ |= also. Then M+ |= Tr() by Thm.2.1(ii), and also M+ |= SAx by M+ |= + . Thus M+ |= , and so M |= where M is the reduct of M+ to the language of . QED

References
[1] Andrka, H., Givant, S. R., Mikuls, Sz., Nmeti, I., and Simon, A., Notions e a e of density that imply representability in algebraic logic. Annals of Pure and Applied Logic 91 (1998), 93-190. [2] Andrka, H., Nmeti, I., and Sain, I., Algebraic Logic. In: Handbook e e of Philosophical Logic Vol. 2, Second Edition, Kluwer, 2001. pp.133-296. http://www.math-inst.hu/pub/algebraic-logic/handbook.pdf [3] Gabbay, D. M., Kurucz, A., Wolter, F., and Zakharyaschev, M., Manydimensional modal logics: theory and applications., Elsevier, 2003. [4] Gyenis, Z., On atomicity of free algebras in certain cylindric-like varieties. Logic Journal of the IGPL 19,1 (2011), 44-52. [5] Henkin, L., Monk, J. D., and Tarski, A., Cylindric Algebras Parts I and II, North-Holland, 1985. [6] Nmeti, I., Logic with three variables has Gdels incompleteness prope o erty - thus free cylindric algebras are not atomic., Mathematical Institute of the Hungarian Academy of Sciences, Preprint No 49/1985, 1985. http://www.math-inst.hu/ nemeti/NDis/NPrep85.pdf [7] Nmeti, I., Free algebras and decidability in algebraic logic., Dissertation e with the Hungarian Academy of Sciences, Budapest, 1986. In Hungarian. English summary is [8]. http://www.math-inst.hu/ nemeti/NDis/NDis86.pdf [8] Nmeti, I., Free algebras and decidability in algebraic logic. Summary in e English. 12pp. http://www.math-inst.hu/ nemeti/NDis/NSum.pdf [9] Nmeti, I., and Simon, A., Relation algebras from cylindric and polyacid e algebras. Logic Journal of the IGPL 5,4 (1997), 575-588. [10] Sayed-Ahmed, T., Tarskian Algebraic Logic., Journal on Relation Methods in Computer Science 1 (2004), pp.3-26. [11] Sayed-Ahmed, T., Algebraic logic, where does it stand today?, The Bulletin of Symbolic Logic 11,4 (2005), pp.465-516. [12] Simon, A., Connections between quasi-projective relation algebras and cylindric algebras., Algebra Universalis 56,3-4 (2007), 263-301. 18

[13] Tarski, A., A formalization of set theory without variables., J. Symbolic Logic 18 (1953), p.189. [14] Tarski, A., An undecidable system of sentential calculus., J. Symbolic Logic 18 (1953), p.189. [15] Tarski, A., and Givant, S. R., Formalizing set theory without variables., AMS Colloquium Publications Vol. 41, AMS, Providence, Rhode Island, 1987.

Hajnal Andrka and Istvn Nmeti e a e Rnyi Mathematical Research Institute e Budapest, Reltanoda st. 13-15 a H-1053 Hungary andreka@renyi.hu, nemeti@renyi.hu

19