Explicitly Non-Standard Uniserial Modules

P. C. Eklof University of California, Irvine S. Shelah Hebrew University and Rutgers University

October 6, 2003

Abstract A new construction is given of non-standard uniserial modules over certain valuation domains; the construction resembles that of a special Aronszajn tree in set theory. A consequence is the proof of a sucient condition for the existence of non-standard uniserial modules; this is a theorem of ZFC which complements an earlier independence result.
modified:1995-09-04

Introduction
This paper is a sequel to [ESh]. Both papers deal with the existence of non-standard uniserial modules over valuation domains; we refer to [ESh] for history and motivation. While the main result of the previous paper was an independence result, the main results of this one are theorems of ZFC, which complement and extend the results of [ESh]. We are interested in necessary and sucient conditions for a valuation domain R to have the property that there is a non-standard uniserial R-module of a given type J/R. (Precise denitions are given below.) The question is interesting only when R is uncountable, and since additional complications arise for higher cardinals, we conne ourselves to rings of cardinality 1 . Associated to any type J/R is an invariant, denoted (J/R), which is a member of a Boolean algebra D(1 ) (equal to P(1 ) modulo the lter of closed unbounded sets). For example, if R is an almost maximal valuation domain, then (J/R) = 0 for all types J/R; but there are natural and easily dened examples where (J/R) = 1.
Thanks to Rutgers University for its support of this research through its funding of the rst authors visits to Rutgers. Partially supported by Basic Research Fund, Israeli Academy of Sciences. Pub. No. 461

revision:1995-09-04

461

2 It is a fact that if (J/R) = 0, then there is no non-standard uniserial R-module of type J/R (cf. [ESh, Lemma 5]). In [ESh] we showed that the converse is independent of ZFC + GCH; the consistency proof that the converse fails involved the construction of a valuation domain R associated with a stationary and co-stationary subset of 1 that is, 0 < (J/R) < 1. The existence of such sets requires a use of the Axiom of Choice; no such set can be explicitly given. Thus without attempting to give a mathematical denition of natural we could say that for natural valuation domains, R, it is the case that for every type J/R, (J/R) is either 0 or 1. For natural valuation domains, it turns out that the converse is true: if there is no non-standard uniserial R-module of type J/R, then (J/R) = 0. This is a consequence of the following result which is proved below (for all valuation domains of cardinality 1 ): if (J/R) = 1, then there is a non-standard uniserial R-module of type J/R. (Theorem 12.) This vindicates a conjecture made by Barbara Osofsky in [O1, (9), p. 164]. (See also the Remark following Theorem 12.) The proof of Theorem 12 divides into several cases; the key new result which is used is a construction of a non-standard uniserial module in the essentially countable case; this construction is done in ZFC and is motivated by the construction of a special Aronszajn tree. (See Theorem 7.) Moreover, the uniserial constructed is explicitly non-standard in that there is an associated special function which demonstrates that it is non-standard. This special function continues to serve the same purpose in any extension of the universe, V, of set theory, so the module is absolutely non-standard. In contrast, this may not be the case with non-standard uniserials constructed using a prediction (diamond) principle. (See the last section.) The rst author would like to thank L. Salce and S. Bazzoni for their critical reading of a draft of this paper, and A. Kechris for a helpful conversation on absoluteness.

revision:1995-09-04

modified:1995-09-04

Preliminaries
For any ring R, we will use R to denote the group of units of R. If r R we will write x y (mod r) to mean x y rR. A module is called uniserial if its submodules are linearly ordered by inclusion. An integral domain R is called a valuation domain if it is a uniserial R-module. If R is a valuation domain, let Q denote its quotient eld; we assume Q = R. The residue eld

461

3 of R is R/P , where P is the maximal ideal of R. [FS] is a general reference for modules over valuation domains. If J and A are R-submodules of Q with A J, then J/A is a uniserial R-module, which is said to be standard. A uniserial R-module U is said to be non-standard if it is not isomorphic to a standard uniserial. Given a uniserial module U , and a non-zero element, a, of U , let Ann(a) = {r R: ra = 0} and let D(a) = {r 1 R: r divides a in U }. We say U is of type J/A if J/A D(a)/Ann(a). This is well-dened in that if b is another non-zero element of U , = then D(a)/Ann(a) D(b)/Ann(b). For example, U has type Q/R if and only if U is = divisible torsion and the annihilator ideal of every non-zero element of U is principal. (But notice that there is no a U with Ann(a) = R.) It is not hard to see that if U has type J/A, then U is standard if and only if it is isomorphic to J/A. We will only consider types of the form J/R; it is a consequence of results of [BFS] that the question of the existence of a non-standard uniserial R-module of type J1 /A can always be reduced to the question of the existence of a non-standard uniserial of type J/R for an appropriate J. From now on we will assume that R has cardinality 1 . We always have ()
modified:1995-09-04

1 J = <1 r R

for some sequence of elements {r : < 1 } such that for all < , r |r . If J is countably generated, then U is standard, so generally we will be assuming that J is not countably generated; then it has a set of generators as in (), where, furthermore, r does not divide r if < . If lim(1 ), let ()
1 J = < r R. def

revision:1995-09-04

By results in [BS] every uniserial module U , of type J/R, is described up to isomorphism by a family of units, {e : < < 1 } such that () e e e (mod r )

for all < < < 1 . Indeed, U is a direct limit of submodules a R where Ann(a ) = 1 r R; then a R r R/R and U is isomorphic to a direct limit of the r R/R, where = 1 1 1 1 1 1 the morphism from r R/R to r R/R takes r to e r if a = e r r a . If U is given by (), then U is standard if and only if there exists a family {c : < 1 } of units of R such that () c e c (mod r )

461

4 for all < < 1 . Indeed, if the family {c : < 1 } satisfying () exists, then mul1 tiplication by the c gives rise to isomorphisms from r R/R to a R, which induce an isomorphism of J/R with U .

Essentially Countable Types

1 Denition. Suppose J = <1 r R as in (*). Call the type J/R essentially uncountable if for every < 1 there exists > such that r R/r R is uncountable. Otherwise, J/R is essentially countable; this is equivalent to saying that there is a < 1 such that for all < < 1 , r R/r R is countable. Say that J/R is strongly countable if for all < 1 , R/r R is countable; clearly, a strongly countable type is essentially countable. It is easily seen that the notions of being essentially or strongly countable are welldened, that is, independent of the choice of the representation (*). If the residue eld of R is uncountable, then, except in trivial cases, the types J/R have to be essentially uncountable; but if the residue eld is countable, the question is more delicate.

Proposition 1 If the residue eld of R is uncountable, then every type J/R such that J is not countably generated is essentially uncountable.
modified:1995-09-04

1 Proof. Let J = <1 r R as in (*). It suces to prove that if < , then r R/r R 1 is uncountable. But r R/r R R/tR where t = r r P . So we have tR P R, = and hence (R/tR)/(P/tR) R/P , the residue eld of R. Since R/P is uncountable, so = is R/tR. 2

Theorem 2 For any countable eld K there are valuation domains R 1 and R2 , both of cardinality 1 with the same residue eld K and the same value group, whose quotient elds, Q1 and Q2 , respectively, are generated by 1 but not countably many elements, and such that Q1 /R1 is essentially uncountable and Q2 /R2 is strongly countable. Proof. Let G be the ordered abelian group which is the direct sum <1 Z ordered anti-lexicographically; that is, n > 0 if and only if n > 0, where is maximal such that n = 0. In particular, the basis elements have their natural order and if < , then k < in G for any k Z. Let G+ = {g G: g 0}. Let R = K[[G]], that is, R = { g kg X g : kg K, a well-ordered subset of G+ }, with the obvious addition and multiplication (cf. [O1, p. 156]). Given an element y = g g kg X of R, let supp(y) = {g : kg = 0}; let p-supp(y) = { 1 : g supp(y) whose projection on Z is non-zero}. Dene v(y) = the least element of supp(y). If X G, then y|X is dened to be gX kg X g . Let y| = y|{g G+ : g < }.

461

revision:1995-09-04

5 Let R1 = {y R: p-supp(y) is nite}. Then R1 is a valuation domain since p1 supp(xy ) p-supp(x) p-supp(y). Let R2 be the valuation subring of R1 generated by {X g : g G}. We have Qj = <1 X Rj for j = 1, 2. Now Q1 /R1 is essentially uncountable since for all > , X R1 /X R1 contains the 20 elements of the form (n)X (n+1)
n

modified:1995-09-04

(with p-supp = {}) where is any function: 2. R1 has cardinality 20 ; if 20 > 1 , to get an example of cardinality 1 , choose a valuation subring of R1 which contains all the monomials X g (g G) and 1 of the elements n (n)X (n+1) for each . We claim that Q2 /R2 is essentially countable. Let K[G] be the subring of R generated by {X g : g G+ }; thus K[G] consists of the elements of R with nite support; we shall refer to them as polynomials. R2 consists of all elements of the form xy 1 where x and y are polynomials and v(x) v(y). We claim that R2 /X R2 is countable for any < 1 . There are uncountably many polynomials, but we have to show that there are only countably many truncations xy 1 |. Given polynomials x and y with v(x) v(y), there is a nitely generated subgroup 1id Zi of G (with 1 < 2 < . . . < d ) such that x and y are linear combinations of monomials X g with g 1id Zi . More precisely, there exist k, r and a (k+r)-tuple (a1 , . . . , ak+r ) of elements of K and k + r linear terms tj of the form
d

tj =
i=1

nij vi
d i=1

(nij Z, vi variables) such that if we let tj () denote () x=

k j=1

nij i , then

aj X tj () and y =

r j=k+1

aj X tj () .

revision:1995-09-04

Finally, there is q d such that q is maximal with q < . Now, consideration of the algorithm for computing xy 1 shows that, for xed (a1 , . . . , ak+r ) and tj , there are linear terms
d

s =
i=1

mi v i

(mi Z, ) and elements c K such that for any strictly increasing sequence = 1 , . . . , d , if x and y are as in ( ), then xy 1 =

c X s () .

461

6 For any q d, only certain of the s involve only variables vi with i q (i.e. mi = 0 if i > q); say these are the s with T (T ). If is such that i < i i q, then s () xy 1 | = . T c X There are only countably many choices for q, d, k and r in , (a1 , . . . , ak+r ) K k+r , and for 1 < ... < q < . Therefore, there are only countably many possibilities for the truncations xy 1 |. 2 By the rst part of the following, the type Q2 /R2 of the previous theorem must be strongly countable; on the other hand, there are types which are essentially countable but not strongly countable. Proposition 3 (i) If Q/R is essentially countable, then it is strongly countable. (ii) For any countable eld K, there is a valuation domain R with residue eld K which has a type J/R which is essentially countable but not strongly countable.
1 Proof. (i) Since Q/R is essentially countable, we can write Q = <1 r R such that for all < , r R/r R is countable. We claim that R/r0 R is countable, which clearly is equivalent to R/r R countable for all < 1 . Suppose not. There is a < 1 such 2 2 that r = r0 t for some t R (since r0 Q). But then r0 R/r R R/tr0 R, which is = uncountable since R/r0 R is uncountable, and this contradicts the choice of the r . (ii) Let G = 1 Z, ordered anti-lexicographically. Let R = K[[G]] (cf. proof of Theorem 2), and let R be the smallest valuation subring of R containing all the g 1 +1 . Then the proof that monomials X (g G). Let J = <1 r R where r = X R/X R is countable for all < is the same as in Theorem 2. But r R/r R = R/r0 R = R/X 1 R is clearly uncountable. 2

modified:1995-09-04

Remark. More generally, referring to a dichotomy in [O1, Prop. 7, p. 155], if the type J/R is essentially countable and falls into Case (A), then J/R is strongly countable; if it falls into case (B), then it is not strongly countable.

revision:1995-09-04

Gamma Invariants
A subset C of 1 is called a cub short for closed unbounded set if sup C = 1 and for all Y C, sup Y 1 implies sup Y C. Call two subsets, S1 and S2 , of 1 equivalent i there is a cub C such that S1 C = S2 C. Let S denote the equivalence class of S. The inclusion relation induces a partial order on the set, D(1 ), of equivalence classes, i.e., S1 S2 if and only if there is a cub C such that S1 C S2 C. In fact, this induces a Boolean algebra structure on D(1 ), with least element, 0, the equivalence class of sets disjoint from a cub; and greatest element, 1, the equivalence class of sets

461

7 containing a cub. We say S is stationary if S = 0, i.e., for every cub C, C S = . We say S is co-stationary if 1 \ S is stationary. Given R and a type J/R, where J is as in (*), dene (J/R) to be S, where S = { lim(1 ): R/ < r R is not complete} where the topology on R/ < r R is the metrizable linear topology with a basis of neighborhoods of 0 given by the submodules r R ( < ). This denition is independent of the choice of the representation of J as in (*) see [ESh]. For any limit ordinal < 1 , let
TJ/R = { u : < : < < (u R , and u u r R)}; that is, TJ/R consists of sequences of units which are Cauchy in the metrizable topology on R/ < r R. Let L consist of those members of TJ/R which have limits in R, i.e. J/R L = { u : < TJ/R : u R s.t. < (u u r R)}. J/R

Note that (J/R) = S where

S = { lim(1 ): TJ/R = L }. J/R
modified:1995-09-04

If J is not countably generated, then CH implies that (J/R) = 1, since the completion of R/ < r R has cardinality 20 > 1 . An 1 -ltration of R by subrings is an increasing chain {N : 1 } of countable subrings of R such that R = 1 N , and for limit , N = < N . Dene (J/R) = E where
E = { lim(1 ): u : < TJ/R s.t. f R < s.t. u f N (mod r )}. /

revision:1995-09-04

Again, it can be shown that the denition does not depend on the choice of {r : < 1 } or of {N : < 1 }. Notice that (J/R) (J/R) since if TJ/R = L , then we J/R can let f be a limit of u1 : < . In [ESh, Theorem 7] it is proved that if (J/R) = 0, then there is a non-standard uniserial R-module. Theorem 4 Suppose J/R is essentially countable. Then (i) (J/R) = 0; (ii) (J/R) = 1.

461

8
1 Proof. Without loss of generality we can assume that J = <1 r R where r0 R/r R is countable for all < 1 . (i) We can also assume that the 1 -ltration of R by subrings, R = <1 N , has the property that for all , N contains a complete set of representatives of r0 R/r R for each < . For any lim(1 ), and any u : < in TJ/R , let f = u1 . To show that 0 E , it suces to show that u f N (mod r ) for all < . Now u f = u u1 1 / 0 (mod r0 ), since u u0 (mod r0 ), by denition of TJ/R . Say u u1 1 = y r0 R. By the 0 assumption on N , there exists a N such that y a (mod r ). Then u f = u u1 = 0 1 + y 1 + a (mod r ), and 1 + a N since N is a subring of R. (ii) To show that (J/R) = 1, it suces to show that for all limit ordinals < 1 , R/< r R is not complete. Assuming that it is complete, we shall obtain a contradiction by showing that r0 R/r R is uncountable. Fix a ladder on , i.e., a strictly increasing sequence n : n whose sup is . For each function : 2, dene u = u : < TJ/R as follows: if m < m+1 , then

u =
im

(i)ri .

Clearly u r0 R, and if > , where k < k+1 , then m k and

modified:1995-09-04

u u =
i=m+1

(i)ri rm+1 R r R.

Since R/ < r R is assumed to be complete, for each there is an element u R which represents the limit of u : < in R/ < r R. To obtain a contradiction, we need only show that if = , then u u r R. Without loss of generality there exists m / such that |m = |m and (m) = 0, (m) = 1, then u um+1 rm+1 R; and

revision:1995-09-04

u u m+1 rm+1 R; / but um+1 u m+1 = rm rm+1 R, so u u rm+1 R r R. 2 /

Special Aronszajn trees

This section contains standard material from set theory. (See, for example, [J, 22] or [Dr, Ch. 7, 3].) It is included simply to provide motivation for the notation and proof in the next section.

461

9 A tree is a partially ordered set (T , <) such that the predecessors of any element are well ordered. An element x of T is said to have height , denoted ht(x) = , if the order-type of {y T : y < x} is . The height of T is dened to be sup{ht(x) + 1: x T }. If T is a tree, a branch of T is a maximal linearly ordered initial subset of T ; the length of a branch is its order type. If T is a tree, let T = {y T : ht(y) = }. We say that a tree T is a -Aronszajn tree if T is of height , |T | < for every < , and T has no branch of length . A tree T of height 1 is a special Aronszajn tree if T is countable for all < 1 and for each < 1 there is a function f : T Q such that ($) whenever x T and y T and x < y, then f (x) < f (y).

Notice that a special Aronszajn tree is an 1 -Aronszajn tree, since an uncountable branch would give rise to an uncountable increasing sequence of rationals. Knigs Lemma implies that there is no -Aronszajn tree. However, there is an o 1 -Aronszajn tree: Theorem 5 There is a special Aronszajn tree. Proof. Let < denote the set of all functions from { 1 : < } to . We shall construct T and f by induction on < 1 such that T is a countable subset of < and the partial ordering is inclusion, i.e., if x T and y T then x < y if and only if < and y| = x. Finally, T will be dened to be <1 T . Let T0 = {}, f0 () = 0, T1 = {0} , and f1 : T1 Q be onto (0, ). Suppose now that T and f have been dened for all < such that for all < < : ( ) for any > 0, and x T there is y T such that x < y and f (y) < f (x) + . There are two cases. In the rst case, if is a successor ordinal, = + 1, let T = {x {( , n)} : n , x T }.

revision:1995-09-04

modified:1995-09-04

Dene f so that for every x T , {f (x {( , n)}) : n } = {r Q : r > f (x)}. Clearly ( ) continues to hold. In the second case, is a limit ordinal. Choose a ladder n : n on . For each < , x T and k > 0, by inductive hypothesis ( ) there exists a sequence yn : n s.t. n > such that yn Tn , x < yn < ym for n < m and fn (yn ) < f (x)+(1/k1/n). Let y[, x, k] = n yn < . Let T consist of one such y[, x, k] for each , x, k. Dene f (y[, x, k]) = f (x) + 1/k. Then it is clear that ( ) still holds. 2

461

10

Special Uniserial Modules

1 Denition. Suppose U is a uniserial module of type J/R where J = <1 r R as in (*). For each > 1 , x an element a of U such that Ann(a ) = r R (so that the submodule a R of U is isomorphic to R/r R r R/R). Let I be the set of all = 1 1 R-module isomorphisms : a R r R/R. We say that {f : 1 } is a special family of functions for U if for each < 1 , f : I Q such that whenever < and I extends I , then f () < f ().

Lemma 6 If U has a special family of functions, then U is non-standard. Proof. Suppose there is an isomorphism : U J/R. Then for every < 1 , restricts 1 to an isomorphism of a R onto r R/R. But then f ( ): < 1 is an uncountable strictly increasing sequence of rationals, a contradiction. 2 With this lemma as justication, we will say that U is explicitly non-standard if U has a special family of functions. If the uniserial module uniserial module U , of type J/R, is described up to isomorphism by a family of units, {e : < < 1 } as in (), then it is clear that U is explicitly non-standard if and only if for every < 1 , there is a function
modified:1995-09-04

f : (R/r R) Q such that whenever < and c , c R satisfy c c e (mod r ), then f (c ) < f (c ).

($$)

revision:1995-09-04

(Here, and hereafter, we abuse notation and regard f and f as functions on R .) Note that we have a tree, T, such that T = (R/r R) and the partial ordering is given by: c + r R < c + r R < and c c e (mod r ). Assume < . Each c has at least one successor of height , namely c e , and if r R/r R is countable, then c has only countably many successors of height . For each c T , its unique predecessor in T is c (e )1 . (Here again we abuse notation and write, for example, c for an element of T instead of c + r R.) Without loss of generality we can assume that the r are such that for all < 1 , r R/r+1 R is innite. (Just choose a subsequence of the original r s if necessary.) Thus for all < 1 , there is an innite subset W of R such that for all u = v W , u 1 (mod r ) and u v (mod r+1 ).

461

11 Theorem 7 If J/R is an essentially countable type, then there is an explicitly nonstandard uniserial R-module of type J/R. Proof. We will rst give the construction in the case when J/R is strongly countable, and afterward indicate the modications needed for the general case. Thus T is assumed countable for all < 1 . We will dene, by induction on , e for < < as in () and, at the same time, the maps f : (R/r R) Q for < . We will do this so that ($$) holds and the following condition is satised for all < < 1 : for any > 0, m , cj T , and cj T (j = 1, . . . , m) such that cj < cj , there exists u R such that u 1 (mod r ) and f (ucj ) < f (cj ) + for all j = 1, . . . , m.

, )

modified:1995-09-04

Note that the cj determine the cj cj cj (e )1 (mod r ) and ucj is another successor of cj of height . For any given > 0, < < , m , and cj R (j = 1, . . . , m), there exist innitely many u as in ( , ), since we can decrease as much as we like. Suppose we have dened e and f for all < < satisfying the inductive hypothe ses. Let un : n enumerate representatives of all the elements of (R/r R) . Also, let q : q enumerate all instances of ( , ), for all < , with each instance repeated innitely often. More precisely, we enumerate (with innite repetition) all tuples of the form 1 = , , cj + r R: j = 1, . . . , m n with n \ {0}, < , and cj R . We will dene f as the union of a chain of functions f,k into Q, each with a nite domain. When k is even we will concentrate on insuring that the domain of f will be T ; and when k is odd, we will work at satisfying the conditions ( , ). Suppose rst that = + 1 and dene e = 1 and e = e for < . Suppose that f,i has been dened for i < k, and assume rst that k is even. Let n be minimal such that un dom(f,k1 ). Let dom(f,k ) = dom(f,k1 ) {un } and let f,k (un ) be any / rational greater than f (un (e )1 ) (= f (un )). Now suppose k is odd; say k = 2q + 1. Its easy to see that its enough to construct f to satisfy ( , ). So if q is an instance of ( , ) for < , let f,k = f,k1 . Otherwise, suppose q is the instance of ( , ) given by 1 j , c + r R: j = 1, . . . , m. n

461

revision:1995-09-04

12 Since W is innite (see above), there is a unit u such that u 1 (mod r ) and ucj / dom(f,k1 ) for j = 1, . . . , m. Then dene f,k to be the extension of f,k1 with domain = dom(f,k1 ) {ucj : j = 1, ..., m} such that f,k (ucj ) = f (cj (e )1 ) + 1 . 2n

Now we consider the case where is a limit ordinal. Fix a ladder n : n on . We are going to dene units e n by induction such that e n e m em (mod rn ) n whenever n < m < . This will easily determine the sequence e : < such that for all < < , e e e (mod r ); then () will be satised for e : < . For simplicity of notation, let en denote e n . Suppose weve already dened f,k1 and ek such that for all x dom(f,k1 ), fk (xe1 ) < f,k1 (x). k (Recall that if x T , then xe1 is the unique predecessor of x in Tn .) If k is even, we n proceed as in the even case above (when is a successor). If k = 2q + 1 and q is 1 , , cj + r R: j = 1, . . . , m n we can assume since each instance is repeated innitely often that < k . Thus e = e k ek is dened. Note that cj (e )1 cj e1 (ek )1 (mod r ), so we can apply k ( ,k ) [with cj k = cj e1 ] and obtain a unit w 1 (mod r ) such that for all j = k 1, . . . , m 1 fk (wcj e1 ) < f (cj (e )1 ) + . k 2n Moreover, since there are innitely many such w, we can choose one so that the elements wcj (j = 1, . . . , m) do not belong to dom(f,k1 ). Let these be the new elements of the domain of f,k and dene f,k (wcj ) = f (cj (e )1 ) + 1 . 2n
modified:1995-09-04

revision:1995-09-04

Now we will dene ek+1 (for k odd or even). For each x dom(f,k ) we have committed ourselves to f (x) (= f,k (x)) and to the predecessor of x in Tk (= xe1 ); we need to k choose ek+1 so that x and its predecessor, xe1 , in Tk+1 satisfy ($$). k+1 k+1 Let e = ek (ek )1 . The desired element ek+1 will have the form ue for some unit u 1 (mod rk ). Choose < f,k (x) fk (xe1 ) for each x dom(f,k ). Apply k ( k ,k+1 ) to this and xe1 Tk , xe 1 Tk+1 (x dom(f,k )). (Note that xe1 < k k xe 1 by choice of e .) This gives us v 1 (mod rk ) such that for all x fk+1 (vxe 1 ) < fk (xe1 ) + k < f,k (x).

461

13 Then we let ek+1 = v 1 e , and we have completed the inductive step. This completes the proof in the strongly countable case. We turn now to the general (essentially countable) case. In this case, R/r0 R may be uncountable; let Z be a complete set of representatives of (R/r0 R) . Fix z0 Z. We rst dene, by induction on , f (c ) or, more precisely, f (c + r R) for all c R such that c z0 e (mod r0 ). 0 We do the construction exactly as in the previous strongly countable case; this will work since there are only countably many cosets c + r R such that c z0 e (mod r0 ) since 0 r0 R/r R is countable. Having done this, the e are determined. We claim that there is no family {c : < 1 } satisfying (). Indeed, suppose we had such a family. Let z Z be such that c0 z (mod r0 ). Then for all < 1 , c ze (mod r0 ). Hence the family {z0 z 1 c : < 0 1 } satises () and also satises z0 z 1 c z0 e (mod r0 ); but this is impossible by 0 construction. 2

Consequences
Now we consider some of the general consequences, for the question of the existence of non-standard uniserials, of the results of the previous sections. First of all, we can construct non-standard uniserial modules associated to any residue eld of cardinality 1 . Proposition 8 (i) For any countable eld K, there exists a valuation domain R of cardinality 1 with residue eld K such that there is an explicitly non-standard uniserial module of type Q/R. (ii) For any eld K of cardinality 1 , there exists a valuation domain R of cardinality 1 with residue eld K such that there is a non-standard uniserial module of type Q/R. Proof. Part (i) is an immediate consequence of Theorem 2 and Theorem 7. Part (ii) follows from (i) in the case of a countable K and from the Osofsky construction in the case of an uncountable K (cf. [O1]; see also [ESh, Theorem 11]). 2 The following improves [ESh, Corollary 15], in that it is a theorem of ZFC rather than a consistency result. It shows that the condition (J/R) > 0 is not necessary for the existence of a non-standard uniserial of type J/R. Proposition 9 There is a valuation domain R of cardinality 1 such that (Q/R) = 0 and there is a non-standard uniserial R-module of type Q/R.

461

revision:1995-09-04

modified:1995-09-04

14 Proof. Let R be such that Q/R is essentially countable (cf. Theorem 2). By Theorem 4(i), (Q/R) = 0, but there is a non-standard uniserial R-module of type Q/R by Theorem 7. 2 The following sums up some old results which we want to combine with results proved here. Theorem 10 Suppose that R is a valuation domain of cardinality 1 . (i) If CH does not hold and J/R is an essentially uncountable type, then there is a non-standard uniserial R-module of type J/R. (ii) If CH holds and (J/R) = 1, then there is a non-standard uniserial R-module of type J/R. Proof. Part (i) is Theorem 8 of [ESh]. Part (ii) is because the weak diamond principle, 1 (1 ), is a consequence of CH (see [DSh]) and this implies that there exists a nonstandard uniserial of type J/R when (J/R) = 1 (see [ESh, Proposition 3] or [FrG]). 2 Now we can completely handle the cases when either CH fails, or = 1. Theorem 11 If CH does not hold, then for every valuation domain R of cardinality 1 , and every type J/R such that (J/R) = 0, there is a non-standard uniserial R-module of type J/R. Proof. Use Theorem 7 for the essentially countable case, and Theorem 10(i) otherwise. 2 Remark. Thm. 14]. This result shows that CH is needed for the independence result in [ESh,

modified:1995-09-04

Theorem 12 For every valuation domain R of cardinality 1 and every type J/R, if (J/R) = 1, then there is a non-standard uniserial R-module of type J/R.

revision:1995-09-04

Proof. If CH fails, use the previous theorem. If CH holds, use Theorem 10(ii). 2 Remark. Osofskys original conjecture ([O1, (9), p. 164], restricted to valuation domains of cardinality 1 , said in our notation that there is a non-standard uniserial R-module of type J/R if and only if (J/R) = 1. This is now seen to be true assuming CH. On the other hand, it cannot be true in this form assuming CH, since CH implies the weak diamond principle for some co-stationary subsets of 1 (cf. [EM, VI.1.10]) Indeed, as in the proof of [ESh, Prop. 3] it is possible to construct R with a type J/R where (J/R) = S and 1 (S) holds; so there is a non-standard uniserial R-module of type J/R (cf. [ESh, proof of Prop. 3]). On the other hand, to construct such an R

461

15 one has to begin with the stationary and co-stationary set S, so such rings will not be natural, i.e. will not be ones ordinarily met in algebraic contexts. Recall, from [ESh], that it is in the case when the hypotheses of the previous theorems fail i.e., when CH holds and (J/R) < 1 (and non-zero) that the independence phenomena occur.

Absoluteness
Finally, let us briey discuss absoluteness. Consider Theorem 11; if CH fails and (J/R) = 0, we always have a non-standard uniserial module of type J/R, but there are two separate constructions involved. In one case, when J/R is essentially countable, we construct an explicitly non-standard uniserial. If the universe of set theory is extended to a larger universe (with the same 1 ) this module remains non-standard because the special family of functions remains a special family for U in the extension of the universe. In the essentially uncountable case we use the fact that (J/R) = 1 ( [ESh, Theorem 8]) and construct our non-standard uniserial U as in [ESh, Theorem 7]. In this case too U remains non-standard in an extension of the universe (preserving 1 ). The reason here is more subtle; relative to a xed 1 -ltration of R by subrings, N , the e we construct satisfy the following property for every lim(1 ) and every c R :
modified:1995-09-04

(#c, )

c : < N [ < (t R(c c e = r t))]

461

revision:1995-09-04

It is a theorem of ZFC that if (#c, ) holds for U (dened by the e ) for all c, , then U is non-standard. Now (#c, ) is, by a coding argument, a 1 statement (with parameters 1 in the ground model) about . Hence, by a theorem of Mostowski (cf. [Dr, Thm. 7.13, p. 160]), it remains true in an extension of the universe, so U remains non-standard. On the other hand, in the proof of Theorem 12, there is one additional case: when J/R is essentially uncountable and (J/R) = 0 (so CH holds). In this case the existence of a non-standard uniserial is proved using the weak diamond principle, which is a consequence of CH. Here the U we construct may not remain non-standard in an extension of the universe. Consider for example that R is constructed as in [ESh, Theorem 14], but with (J/R) = 1. If P is the forcing dened in the proof there, then P is proper, so it preserves 1 and, moreover, in the P-generic extension U is standard. (Of course, in the generic extension we can construct another non-standard module.)

16

References
[BFS] S. Bazzoni, L. Fuchs, and L. Salce, The hierarchy of uniserial modules over a valuation domain, preprint. [BS] S. Bazzoni and L. Salce, On non-standard uniserial modules over valuation domains and their quotients, J. Algebra 128 (1990), 292-305.

[DSh] K. Devlin and S. Shelah, A weak version of which follows from 20 < 21 , Israel J. Math. 29, 239247 (1978). [Dr] [EM] F. R. Drake, Set Theory: An Introduction to Large Cardinals, NorthHolland (1974). P. C. Eklof and A. H. Mekler, Almost Free Modules, North-Holland (1990).

[ESh] P. C. Eklof and S. Shelah, On a conjecture regarding non-standard uniserial modules, to appear in Trans. Amer. Math. Soc. [FrG] B. Franzen and R. Gbel, Nonstandard uniserial modules over valuation domains, o Results in Math. 12 (1987), 8694.
modified:1995-09-04

[FS]

L. Fuchs and L. Salce, Modules Over Valuation Domains, Marcel Dekker (1985).

[FSh] L. Fuchs and S. Shelah, Kaplanskys problem on valuation rings, Proc. Amer. Math. Soc. 105 (1989), 2530. [J] [O1] T. Jech, Set Theory, Academic Press (1978). B. L. Osofsky, Constructing Nonstandard Uniserial Modules over Valuation Domains, Azumaya Algebras, Actions, and Modules, Contemporary Mathematics 124 (1992), 151164. B. L. Osofsky, A construction of nonstandard uniserial modules over valuation domains, Bull. Amer. Math. Society, 25 (1991), 8997. S. Shelah, Nonstandard uniserial module over a uniserial domain exists, Lecture Notes in Math vol. 1182, Springer-Verlag (1986), pp. 135150.

revision:1995-09-04

[O2] [Sh]

461