LENGTH OF BOOLEAN ALGEBRAS AND ULTRAPRODUCTS

MENACHEM MAGIDOR AND SAHARON SHELAH Abstract. We prove the consistency with ZFC of the length of an ultraproduct of Boolean algebras is smaller than the ultraproduct of the lengths. Similarly for some other cardinal invariants of Boolean algebras.

0. Introduction On the length of Boolean algebras (the cardinality of linearly ordered subsets) see Monk [M1], [M2] (and Denition 1.1 below). In Shelah [Sh 345, 1] it is said that Koppelberg and Shelah noted that by the Lo theorem for s an ultralter D on and Boolean algebras Bi (i < ) we have () | Length(Bi )/D| Length( Bi /D), and
i< i<

i < Length(Bi )

|
i<

i /D| < Length(

Bi /D).

modified:1998-05-27

D. Peterson noted that the indicated proof fails, but holds for regular ultralters (see [Pe97]). Now the intention in [Sh 345] was for Length+ , i.e. ()+ | Length+ (Bi )/D| Length+ ( Bi /D), where Length (B) is the rst cardinal not represented as the cardinality of a linearly ordered subset of the Boolean Algebra (the only dierence being the case the supremum is not attained). Here we prove that the statement () may fail (see Theorem 1.3 and Proposition 1.6). The situation is similar for many cardinal invariants. Of course, if () fails then (using ultraproducts of (H(), Bi ) : i < or see e.g. Roslanowski, Shelah [RoSh 534, 1]) we have {i < : Length+ (B) is a limit cardinal } D, and Length+ (Bi )/D is like for some successor
i< i< + i<

revision:1998-05-01

cardinal . Hence {i < : Length+ (Bi ) is a regular cardinal} D

433

MENACHEM MAGIDOR AND SAHARON SHELAH

hence {i : Length+ (Bi ) is an inaccessible cardinal} D, so the example we produce is in some respect the only one possible. (Note that our convention is that inaccessible means regular limit (> 0 ), not necessarily strong limit.) More results on cardinal invariants of ultraproducts of Boolean algebras can be found in [Sh 462], [Sh 479], [RoSh 534] and [Sh 620], [RoSh 651]. This paper is continued for other cardinal invariants (in particular spread) in [ShSi 677]. We thank Otmar Spinas and Todd Eisworth for corrections and comments. 1. The main result Denition 1.1. (1) For a Boolean algebra B, let its length, Length(B), be sup{|X| : X B, and X is linearly ordered (in B) }. (2) For a Boolean algebra B, let its strict length, Length+ (B), be sup{|X|+ : X B, and X is linearly ordered (in B)}. Remark 1.2. (1) In Denition 1.1, Length+ (B) is (equivalently) the rst such that for every linearly ordered X B we have |X| < . (2) If Length+ (B) is a limit cardinal then Length+ (B) = Length(B); and if Length+ (B) is a successor cardinal then Length+ (B) = (Length(B))+ . Theorem 1.3. Suppose V satises GCH above (for simplicity), is measurable, < , is + -hypermeasurable (somewhat less will suce), F is the function such that F () = the rst inaccessible > , and = F () is well dened, and < , > 22 . Then for some forcing notion P not collapsing cardinals, except those in the interval (+ , ) [so in VP we have ++ = = F V ()], and not adding subsets to , in VP , we have: () in VP the cardinal is a strong limit of conality , () for some strictly increasing continuous sequence i : i < of (strong limit) singular cardinals > with limit , each i =: F V (i ) is still inaccessible and for any normal ultralter D V on we have: i /D has order type ++ = F V ().
i<

revision:1998-05-01

modified:1998-05-27

Denition 1.4. A forcing notion Q is directed -complete if: for a directed quasi-order I (so (s0 , s1 I)(t I)(s0 I t & s1 t)) of cardinality < , and p = pt : t I such that pt Q and s I t ps Q pt , there is p Q such that t I pt Q p.

433

LENGTH OF BOOLEAN ALGEBRAS AND ULTRAPRODUCTS

Proof Without loss of generality for every directed complete forcing notion Q of cardinality at most satisfying the -c.c., in VQ the cardinal is still -hypermeasurable. [Why? If supercompact, use Laver [L], if is just -hypermeasurable see more in Gitik Shelah [GiSh 344].] Let Q be the forcing notion adding Cohen subsets to , i.e., {f : f a partial function from to {0, 1}, |Dom(f )| < }. In V, let R = Levy(+ , < ) = {f : f a partial two place function such that [f (, i) dened 0 < < & i < + & f (, i) < ] and |Dom(f )| < + } (so R collapses all cardinals in (+ , ) and no others, so in VR the ordinal becomes ++ ). Clearly R is + complete and hence adds no sequence of length of members of V. In VQ , there is a sequence D = Di : i < of normal ultralters on as in [Mg4] and, g = gi,j : i < j , gi,j H() witness this (that is Di (VQ ) /Dj , in fact Di is equal to gi,j /Dj in the Mostowski collapse of (VQ ) /Dj ). Let D = D i : i , g = g i,j : i < j be Qnames of that a Qname A ofa subset of is an object of size such sequences. Note , i.e., it consists of a sequence of sequences of members of Q, say pi,j : j < : i < , and function f : {0, 1} such that each {pi,j : j < } is a maximal antichain of Q and pi,j Q i A f (i, j) = 1. So the set of members of Q and the set of Qnames of subsets of are the same in V and in VR . So in VRQ the sequence D still gives a sequence of normal ultralters as required in [Mg4] as witnessed by g = gi,j : i < j . g ) (from there) for changing the conality of Also the Magidor forcing P(D, to (not collapsing cardinals not adding subsets to , the last is just by xing the rst element in the sequence) is the same in VQ and VRQ and has the same set of names of subsets of . We now use the fact that P(D, g ) + -c.c. (see [Mg4]). Let P = (Q R) P(D, g ), so again every satises the g g Q P(D, g )-name involves only decisions so also VQP(D,) , V(QR)P(D,) same subsets of . So our only problem is to check conclusion () have the of Theorem 1.3. Let D V be any normal ultralter on (so this holds also in VR , VRQ , (VRQ )P(D) ).
g Claim 1.4.1. In VQP(D,) , the linear order

modified:1998-05-27

F (i )/D has true conality

revision:1998-05-01

F () = . Proof of the Claim: Clearly Q for i < we have /D i is well ordered (as Di is 1 complete) and
Q

for i < we have /D i has cardinality 2 and even i< 2|i| /D i has cardinality 2

433

MENACHEM MAGIDOR AND SAHARON SHELAH

[Why? As = < and Di is a uniform ultralter on . In details, let h : > 2 be one-to-one, and for each 2 dene g by g (i) = h( i). Then = 2 {i < : g (i) = g (i)} is a bounded subset of

and hence its complement belongs to Di but |{g (i) : 2}| = 2|i| ]. Consequently, for some F we have
Q

F ()/D i is isomorphic to . <

If we look at the proof in [L] (or [GiSh 344]) which we use above, we see that w.l.o.g. F is the F above (and so does not depend on i). So let f i, be Q-names such that
Q

for i < and < , f i, < F () and F ()/D i . fi, /D i is the -th function in < Di and let Bi Di \
i< j<i

In VQ let D =

Dj be as in [Mg4] (you can also

produce them straightforwardly), so Bi : i < is a sequence of pairwise disjoint subsets of . Dene f for < as follows: f Bi = fi, Bi and f ( \
i<

Bi ) is constantly zero. F (). Let D , B i , for < < the set

So f : < is <D increasing and conal in

f be Qnames forced to be as above. Then Q A, = { < : f () < f ()} belongs to D . Now in VQ , one of the properties of Magidor forcing P(D, g ) is that
g P(D,)

for every A D =
i<

Di

revision:1998-05-01

for every i < large enough we have i A (where i : i < is the increasing continuous sequence conal in which P( g ) adds). D, Since for every p P(D, g ), for some q p we have (recall from [Mg4] q (i) is the set which q says belongs to (when q does not forces a that F i value to i )) [F q (i) A, for every i < large enough],
g hence in VQP(D,) , for < < the set {i < : (f (i ) < f (i ))} is F (i ). bounded, i.e., f (i ) : i < : < is <J increasing in bd i<

433

LENGTH OF BOOLEAN ALGEBRAS AND ULTRAPRODUCTS

f F (i ). W.l.o.g. p i< forces that F (i ) < i+1 , and so by [Mg4] (possibly increasing p), we have, for some function h, Dom(h) = , h(i) [i]<0 , that above p, we know: f (i) depends on the value of j for j {i} h(i). So we can dene a function : f
g P(D,)

On the other hand, in VQ , assume p

for some i < , possibly i is and is a possible value (above p) of f (i)}. So f () < F (), hence (in VQ ) for some , f <D f and consequently g p f <J f . So f : < is, in VQP(D,) , <J increasing and bd bd conal in F (i ) which is more than enough for 1.4.1.
i< g Note that 1.4.1 holds in V(RQ)P(D ,) too (remember that any sequence ,) g i < F (i ) : i < in V(QR)P(D is bounded by a function f g V F () (see [Mg4]) and also is itself in VQP(D,) ). Reg g But why, if i < F (i ) and i : i < V(RQ)P(D ,) , do we have that + (this means < )? i /D has cardinality i< g It suces to prove this inequality in the universe V1 = VQP(D,) . Now V1 /D is well founded, hence there is an isomorphism from V1 /D onto a transitive class which we now call M and let j be the isomorphism g (= the Mostowski collapse). As in VQP(D,) is strong limit > , clearly < j( : i < /D) < ; and as D is normal, and i : i < is increasing continuous, we have j( i : i < /D) = . As i < F (i ) we have (by the Lo theorem): s

f () = sup{ :

modified:1998-05-27

M |= j( i : i < /D) is an ordinal smaller than the rst inaccessible >. But the property not weakly inaccessible is preserved by extending the universe (from M to V1 ). So we nish. 1.3 Remark 1.5. (1) The proof has little to do with our particular F . Assume F : and we add () F () = for = j(F ) (), j an appropriate elementary embedding.

revision:1998-05-01

Then all the proof of 1.3 works except possibly the last sentence, which use some absoluteness of the denition of F . (2) We can also vary R. g (3) Let i = F (i ). In V = V(QR)P(D,) we have = < , moreover ( < )[| |< < ] and is strong limit. Hence if in V , P is a forcing notion satisfying the -c.c. of cardinality < , and D

433

MENACHEM MAGIDOR AND SAHARON SHELAH

is an ultralter on extending D then (in (V )P ) we have: ultraproduct i /D is like.

the

Proposition 1.6. Suppose i : i < is a sequence of (weakly) inaccessible cardinals > , D an ultralter on , and the linear order (i , <)/D is
i<

like, regular. (1) There are Boolean algebras Bi (for i < ) such that: (a) Length(Bi ) = Length+ (Bi ) = i , (b) Length+ ( Bi /D) = , (c) if = + then Length(
i< i<

Bi /D) = . Bi, , Bi, increasing

(2) Assume i is regular > , |Bi | = i , Bi =

continuous in (the Bi , Bi, are Boolean algebras of course) and we have ( < i )(|Bi, | < i ) and ()0 if i < , < i , cf() = + then: Bi, < Bi (< is complete subalgebra sign) i.e. ()1 i ,Bi, if x Bi \ Bi, , x = 0 then for some y we have: B (a) y Bi, and y = 0 (b) for any y Bi, such that Bi |= y y & y = 0:
modified:1998-05-27

Bi |= y x = 0 & x y = 0 Then Length+ (

Bi /D) .

xi, xi, for < < .

revision:1998-05-01

Let Bi, be the subalgebra of Bi generated by {xi, : < , < }. Now clause (a) holds immediately, and the inequality in clause (b) holds by the Lo theorem, and the other inequality follows by part (2) of s the proposition. Lastly clause (c) follows. (2) W.l.o.g. the set of members of Bi is i , and the set of elements of Bi, is an initial segment. Let Si = { : < i , the set of members of Bi, is and cf() = + }

433

LENGTH OF BOOLEAN ALGEBRAS AND ULTRAPRODUCTS

Let (B, < , S) =

(Bi , <i , Si )/D, with <i the order on the ordinals < i

( is reserved for the order in the Boolean algebra). So (|B|, < ) is like (where |B| is the set of elements of B). Let yi : i < be an < increasing sequence of members of B. Let S = { < : cf() = + and {yi : i < } has in (|B|, < ) a least upper bound which we call y and it belongs to S}. Now clearly (read [Sh 420, 1] if you fail to see; or assume 2 < and [Sh 111, 2.3 p. 269]): S is stationary. Note: is enough, as if X B is linearly ordered by <B , let yi X for i < be pairwise distinct; as < is like w.l.o.g. yi : i < is < increasing, and let S be as above. For each S apply ()1 B,B{y:y< y } from the assumption to y , y (holds by Lo theorem) and get yi . Then s 2 of S and an element y apply Fodor lemma, and get a stationary subset S such that for every i S 2 we have yi = y . Now the set of yi for i S 2 is independent (check or see [Sh:92, 4.1]). 1.6 So putting Theorem 1.3 and Proposition 1.6 together Conclusion 1.7. Assume CON(ZFC+ some is + hypermeasurable, for 1 some strongly inaccessible 1 > ). Then it is consistent that for some , and sequence Bi : i < of Boolean algebras and ultralter D on we have, for some : (a) Length(Bi )/D = +
i< def

modified:1998-05-27

(b) Length(
i<

Bi /D) = .

433

revision:1998-05-01

Remark 1.8. (1) We can say more on when ultraproducts of free products of Boolean algebras has not too large length. (2) We can use the disjoint sum of Bi, : < i instead. (3) In the proof of 1.6 we actually have Depth+ (Bi ) = Length+ (Bi ) and Depth+ (B) = Length+ (B), and so similarly without +; where Depth+ (B) = sup{|X|+ : X B is well ordered} Depth(B) = sup{|X| : X B is well ordered}. So the parallel of 1.6 holds for Depth instead Length. (4) Recal c(B) is the cellularity of a Boolean algebra b, i.e., sup{|X| : X is a set of pairwise disjoint non zero elements}. If in the proof of 1.6 i dening B we replace xi, xi, by 4 xi, xi, = 0

MENACHEM MAGIDOR AND SAHARON SHELAH

then c(Bi ) = i = c+ (Bi ), c(B) = + , = c+ (B). (Same proof.) (5) We can also get the parallel to 1.6 for the independence number. Let B i be the Boolean algebra generated by {xi, : < and < i } freely except (5 ) xi, xi, = 0 if < < , < , < . Let Ii be the ideal of B i which {xi, : < , < i } generates. Clearly it is a maximal ideal. Let Bi, be the ideal of B i generated by {xi, : < and < }. Again w.l.o.g. the universe of Bi is i and let Ci = { < i : for x Bi we have: x < i x Bi, x Bi, }. It is a club of i . The Bi, are not Boolean subalgebras of Bi , just Boolean subrings; now ()0 in proposition 1.6 is changed somewhat. We will have P Bi = Ii and (B, < , P ) = (Bi , < , P Bi )/D. i
i<

We know: () P is a maximal ideal of B ( by Lo Theorem) s () if i < , < i is a limit ordinal and Ci then for any x P Bi there are x0 < , x0 P Bi and x1 P Bi disjoint to all members of P Bi which are <i and x = x0 x1 . Similarly for B add if you like Qi = Ci i .
modified:1998-05-27

References
[GiSh 344] Moti Gitik and Saharon Shelah. On certain indestructibility of strong cardinals and a question of Hajnal. Archive for Mathematical Logic, 28:3542, 1989. [L] Richard Laver. Making the supercompactness of indestructible under directed closed forcing. Israel J. of Math., 29:385388, 1978. [Mg4] Menachem Magidor. Changing conality of cardinals. Fund. Math., XCIX:6171, 1978. [M1] J. Donald Monk. Cardinal Invariants of Boolean Algebras. Lectures in Mathematics. ETH Zurich, Birkhauser Verlag, Basel Boston Berlin, 1990. [M2] J. Donald Monk. Cardinal Invariants of Boolean Algebras, volume 142 of Progress in Mathematics. Birkhuser Verlag, BaselBostonBerlin, 1996. a [Pe97] Douglas Peterson. Cardinal functions on ultraproducts of Boolean algebras. Journal of Symbolic Logic, 62:4359, 1997. [RoSh 534] Andrzej Roslanowski and Saharon Shelah. Cardinal invariants of ultrapoducts of Boolean algebras. Fundamenta Mathematicae, 155:101151, 1998. math.LO/9703218. [RoSh 651] Andrzej Roslanowski and Saharon Shelah. Forcing for hL and hd. Colloquium Mathematicum, 88:273310, 2001. math.LO/9808104. [Sh:92] Saharon Shelah. Remarks on Boolean algebras. Algebra Universalis, 11:7789, 1980.

433

revision:1998-05-01

LENGTH OF BOOLEAN ALGEBRAS AND ULTRAPRODUCTS

Saharon Shelah. On power of singular cardinals. Notre Dame Journal of Formal Logic, 27:263299, 1986. [Sh 345] Saharon Shelah. Products of regular cardinals and cardinal invariants of products of Boolean algebras. Israel Journal of Mathematics, 70:129187, 1990. [Sh 420] Saharon Shelah. Advances in Cardinal Arithmetic. In Finite and Innite Combinatorics in Sets and Logic, pages 355383. Kluwer Academic Publishers, 1993. N.W. Sauer et al (eds.). 0708.1979. [Sh 479] Saharon Shelah. On Monks questions. Fundamenta Mathematicae, 151:119, 1996. math.LO/9601218. [Sh 462] Saharon Shelah. -entangled linear orders and narrowness of products of Boolean algebras. Fundamenta Mathematicae, 153:199275, 1997. math.LO/9609216. [Sh 620] Saharon Shelah. Special Subsets of cf() , Boolean Algebras and Maharam measure Algebras. Topology and its Applications, 99:135235, 1999. 8th Prague Topological Symposium on General Topology and its Relations to Modern Analysis and Algebra, Part II (1996). math.LO/9804156. [ShSi 677] Saharon Shelah and Otmar Spinas. On incomparability and related cardinal functions on ultraproducts of Boolean algebras. Mathematica Japonica, 52:345358, 2000. math.LO/9903116. Institute of Mathematics, The Hebrew University, Jerusalem 91904, Israel E-mail address: [email protected] Institute of Mathematics, The Hebrew University, Jerusalem 91904, Israel, and Rutgers University, Mathematics Department, New Brunswick, NJ 08854, USA E-mail address: [email protected]

[Sh 111]

433

revision:1998-05-01

modified:1998-05-27