Hypercompletions of Riesz Spaces

Author(s): Wolfgang Filter

Source: Proceedings of the American Mathematical Society, Vol. 109, No. 3 (Jul., 1990), pp.
775-780
Published by: American Mathematical Society
Stable URL: http://www.jstor.org/stable/2048219 .
Accessed: 19/01/2015 17:03

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to
Proceedings of the American Mathematical Society.

http://www.jstor.org

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
PROCEEDINGS OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 109, Number 3, July 1990

HYPERCOMPLETIONS OF RIESZ SPACES

WOLFGANG FILTER

(Communicatedby John B. Conway)

ABSTRACT. It is shown that each Riesz space with separatingorder continuous

dual can be embedded in a unique " e-hypercompletion,"where e is a fixed
weak unit of the extended order continuous dual.

1. INTRODUCTION

In [8] I investigated-in terms of an internal characterization-those Riesz

spaces which can be represented as bands of spaces of measures; I call these
spaces hypercomplete. The present paper is devoted to the question of how
far Archimedean Riesz spaces can be embedded in a "hypercompletion." It
follows already from [7, Theorem A] that each Riesz space L with separating
order continuous dual Ln (i.e. (Ln)0 = {0}) embeds order densely in some
hypercompleteRiesz space M, but uniqueness is not yet guaranteed. The main
result given here is that each Riesz space L with (Ln)0 = {O} can be embedded
order densely in a unique " e-hypercompletion"where e is some weak unit of
the extended order continuous dual F(L) of L. But we will see in an example
that, in general, dependence on the weak unit e cannot be avoided.

2. RESULTS
L always denotes an Archimedean Riesz space.
For convenience, the definitions of extended order continuous dual and hy-
percompleteness follow.
Let D be the set of all order dense ideals of L. Then F(L) U=U '(, is
the extended ordercontinuousdual of L (see [10]), where 4 c In and q Jn
are identified if they coincide on I n J. With algebraic and order structure
defined by using representatives, F(L) is a universally complete Riesz space
[10, 1.5], and for each 4 E F(L) there exists a greatest L[f] E (D such that
c- L[K]n[ 10, 1.3].
If e is a weak unit of F(L) and R c L- is a set of components of e such
that for each upward-directedfamily (xl) from L+ for which sup g(x1) < oo
Received by the editors September1, 1989.
1980 Mathemnatics SubjectClassification(1985 Revision). Primary46A40.
?
1990 American Mathematical Society
0002-9939/90 $1.00 + $.25 per page

775

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
776 WOLFGANG FILTER

for every g E R there exists supx1 in L, then call (e, R) an hc-pair of L.

Call L hypercompleteif there exists an hc-pair of L.
In the sequel let e be a weak unit of F(L). I denote by Re the set of all
components of e in Lj.
Before reading the appropriatedefinition, note that it would not be senseful
to call a Riesz space M e-hypercompletion of L if L is order dense in M
and e is the first component of an hc-pair of M, since in this case M would
not be unique: If X(?X) denotes the set of all real-valued a-additive measures
on some a5-ringJ of subsets of a set X, and h(J) denotes the bounded
measures on M , then for L = /h(J) and e = ix , both M1 -=/(() and
MV2= 1f(M) would satisfy this bad definition. One has to impose an extra
condition for this reason.
If M is an ArchimedeanRiesz space and if /: L -* M is an injective Riesz
homomorphism such that XL is order dense in M, I denote by V,M, the
canonical Riesz isomorphism
F(M) -' F(L), F- 4 ?

([10, 2.6]).
Definition 1. A hypercomplete Riesz space M is called e-hypercompletionof
L if there exists an injective Riesz homomorphism q: L , M such that
1. XL is an order dense Riesz subspace of M;
2. we have 1 (g) E Mn for all g E R ,and (Vj (e), VY (Re)) iS
an hc-pair of M.
Roughly, M is an e-hypercompletionof L iff L is order densely embedded
nM, ReCMne and (e, Re) is an hc-pair of M.
It follows from condition 1 and [8, 2.3] that only spaces with (L-)? = {O}
can possess an e-hypercompletion. For this reason, we assume Ln to be sepa-
rating from now on.
If Y is a locally compact Stonian space, let X(Y) denote the set of normal
Radon measures on Y.
Proposition2. If M1, M2 are e-hypercompletionsof L, then M1 and M2 are
canonically Riesz isomorphic, i.e. iffor k E { 1, 2}, qk5:L -* Mk satisfies the
definition above, then there exists a Riesz isomorphism v: MI - M2 such that
V 0?1 = 02
Proof. Set VWk := . By the Ogasawara-Maedarepresentation theorem
there exist a compact hyperstonianspace Z and a Riesz isomorphism U: F(L) -*
COO(Z) with u(e) = I z . Then U0 ol/k: F(Mk) -COO (Z) is a Riesz isomorphism
with u o v/k(Yk (e)) = 1 Set

Yk := U suppuo (YIk(g))
gC R(,

Uk: F(Mk) -
(U ?
Co(Yk), I ) Vk)(4) Y,

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
 HYPERCOMPLETIONS
OF RIESZSPACES 777

By [7, Theorem A], there exists a Riesz isomorphism Vk: Mk -+ 1(Yk) such
that (bk' Uk, vk) is a w7 (e)-mr of Mk to which I1 (Re) is associated;
observe that vk is onto by the hypercompletenessof Mk. But Y, = Y2, and
hence v := v I0a v is a Riesz isomorphism.
Fix x c L. Then for all 4 E F(L) for which 4(x) is defined:

J uXlydv, (l0x) =f u 1 dv, (Xl1x)= (_ 711)(x) = 4(x)

- J
= Y2u~jy
dV2(X2X).

Hence v1(qIx) = v2($2x) and thus v o 1 = 02 E]

Let Ge denote the ideal of L- generated by Re.
Theorem3. (Ge)- is the e-hypercompletionof L.
Proof. By [7, Theorem A] there exists an e-mr (Y, u, v) of L to which Re
is associated. Then (lI, {kgJg E Re}) is an hc-pair of X41(Y), which implies
that X1/{(Y) is an e-hypercompletionof L.
Since obviously Ge = CC(Y) (the set of continuous functions on Y with
compact support), it follows from the Riesz representationtheorem that (Ge)
= (Y)
The uniqueness was proved in Proposition 2. Ei
Of course, L is its own e-hypercompletion if (e, Re) is an hc-pair of L.
As we have seen in the proof of the precedingtheorem, the e-hypercompletion
of L can always be realized as the space X(Y), for some locally compact hy-
perstonian space Y. Another beautiful description of it in terms of extended
L1(8)-spaces-where ,u is a normal measure on a compact Stonian space-was
obtained by Y. A. Abramovich [15].
Copying the proof of Theorem 3, we get
Corollary 4. If R c Re with e = supgER g,then (GR)n is hypercomplete,where
GR standsfor the ideal of L- generated by R. o

But (GR)n = (Ge?) need not always hold: If L = ,b(S), e = ix, R =

{ i 4A c M}, then (Ge)n = (g) while (GR)n =
The following example shows that, in general, to different e's there belong
different e-hypercompletions.
Example. Let L := f?. Then L- = fI and F(L) = RN. For each weak unit
N~~~~~~~~~
e of lN , set

v (e)= {A c NIE e(n) < oo}

Me := {h CE 1N Ih(n)Ie(n) < oc for all A C X(e)

nEA

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
778 WOLFGANG FILFER

Then Me is the e-hypercompletionof L, where the embedding map 0 is just

the identity map on L.
It follows that if v: Me , Me is a Riesz isomorphism preservingthe em-
bedding of L, we must have VIL = id, and hence, by the order continuity
of v, v = id, which implies Me = Me . But if, for example, e, , e2 E L-
then the condition M M is easily seen to be equivalent to the existence of
positive reals C, D with Ce? < e2 < Del, which allows us to construct many
nonisomorphic e-hypercompletionsof L. n

Let us briefly consider the situation where Ln possesses weak units.

If e E L , then e is the greatest element of R , and since R c ((G)n
by Theorem 3, we have
Proposition5. If e E L_ then (e, {e}) is an hc-pair of (G ) E.
Let M be a Dedekind complete Riesz space, and denote, for x E M, by
S(x) the ideal of M generated by x. Then, if f, g are weak units of M,
the Ogasawara-Maedatheorem implies the existence of a Riesz isomorphism
X: S(f) - S(g) with Of = g.
An immediate consequence of this remark is that for weak units e1, e2 of
Ln , the corresponding e-hypercompletionsare Riesz isomorphic (but of course
not canonically, i.e. not preservingthe embedding of L):
Proposition 6. If e1, e2 are weak units of L-, then Ge and Ge are Riesz
isomorphic,and hence the same holdsfor (G )n and (GQ . a
We can even say that an e-hypercompletionbelonging to e c L- is in some
sense minimal. Consider another weak unit f of F(L) . Then e: inf{e, f}
is a weak unit of L-,
n and since S(f) is isomorphic to G (the remark above
e,
applied to F(L)), Gf embeds order densely in Ge, which implies that (Ge
embeds in (G )n, and hence (G)< embeds in (Gf)n, too.

3. AN APPLICATION
I now brieflysketch an application;a more detailed presentationwill be given
in a subsequent paper.
Several authors (cf. [1-5, 9, 11-14]) have studied so-called Hellinger inte-
grals to get integral representationsof the duals of spaces of finitely additive
or a-additive measures; here a real-valued function q, defined on some ring
of sets, X, is called Hellinger integrable with respect to a measure ,u having
domain J if lim 4E q(A)u(A) exists in R, where _v runs throughthe set
of partitions of X , directed under refinement. But the representationtheorems
these authors get are not very satisfying in some sense, because set theoretical
assumptions have to be made (see e.g. [12]), and one must necessarily restrict
to "small"spaces of measures as Keisler has shown [9, Theorem 4]. A method
proposed here to overcome these difficulties is to work with generalized parti-

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
 HYPERCOMPLETIONS OF RIESZ SPACES 779

tions as describedbelow; this method works not only in spaces of measures but
in Riesz spaces L with separatingdual Ln:
3 denotes the set of all countable disjoint systems from Re . Setting g,:
supgc,/ g for sv E 3, one can define an order relation on 3 by
v' -<i7 :
gX < gw, and for all h E 97 either
inf{h, g,} = 0 or there is g E -v with h < g.
Let 9- be the section filterof 3 with respectto -< . For x E L and q: Re R
set Xx (v ) := EgEl' 0(g)g(x) if the series is summable, and Xx(v ) := 00,
otherwise (-V c 39). One can call X x-integrable if Xx(S9) converges in IR;
in this case, set
J dx limOx (Y9).
The following representationtheorem may serve to justify the definition of
generalized Hellinger integraljust given (observe that this theorem applies in
particularto spaces of measures since they are hypercompleteby [8, 2.5]):
Theorem7. For each 4 E F(L) whichis extendableto M[e] ( where M denotes
the e-hypercompletionof L), there exists a boundedmap X: Re -* IR such that
X is x-integrable and 4(x) = f X dx for all x E L[fl.
Proof (idea of). Let (Y, u, v) be the e-mr of M to which Re C Mn is
associated; then vM = JX(Y), by [7, Theorem A] and the hypercompleteness
of M.
For g E Re, g :&0, set q5(g) := u4(yg), where yg is a point of supp ug in
which lkil attains its minimum on supp kg; for g = 0, set q(g) :=0.
Since kX is definedon 4b(Y) = v(M[e]), kX is bounded by [6, 1.6.lb)], and
hence X is likewise. The rather technical verification that X has the required
properties is omitted here. f1
Every 4 E F(L) can be representedin the same way, but q will no longer be
bounded in general. The advantageof the boundedness of q is that in this case
one obtains the same Hellinger integral when working only with finite disjoint
systems from Re (at least for x E L[e]).
ACKNOWLEDGMENT

I am very grateful to the referee for a number of useful comments.

REFERENCES

1. W. D. L. Appling, Concerninga class of linear transformations,J. London Math. Soc. 44

(1969), 385-396.
2. , An isomorphismand isometrytheoremfor a class of linear functionals,Trans. Amer.
Math. Soc. 199 (1974), 13 1-140.
3. W. C. Bell, A decompositionof additiveset functions,Pacific J. Math. 72 (1977), 305-311.
4. , ApproximateHahn decompositions,uniformabsolute continuity and uniform inte-
grability,J. Math. Anal. Appl. 80 (1981), 393-405.

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions
780 WOLFGANG FILTER

5. W. C. Bell and M. Keisler, A characteriation of the representableLebesguedecomnposition

properties, Pacific J. Math. 83 (1979), 185-186.
6. C. Constantinescu, Dualitv in measuretheoryv,Lecture Notes in Math., vol. 796, Springer-
Verlag, Berlin, Heidelberg, and New York, 1980.
7. W. Filter, Atomicaland atomnfieeelemiientsof a Ries space, Arch. Math. 52 (1989), 580-587.
8. , JivpercompleteRies spaces, Atti Sem. Mat. Fis. Univ. Modena (to appear).
9. M. Keisler, Integral representationfor elemiients
of the dual of ha(S, X), Pacific J. Math.
83 (1979), 177-183.
10. W. A. J. L,uxemburg and J. J. Masterson, ,In extension of the conceptof the orderdual of a
Ries7 space, Canad. J. Math. 19 (1967), 488-498.
11. J. S. MacNerney, Finitelv additiveset functions,Houston J. Math. 6 (1980).
12. R. D. Mauldin, 4 representationtheoremiifor the second dual of C[O, 1], Studia Math. 46
(1973), 197-200.
13. , 7he continliuunhypothesis,integrationand duals of miieasur
e spaces, Illinois J. Math.
19 (1975), 33-40.
14. in miieasuiretheory, Adv. in Math. 27
, Somne effects of set-theoreticalassumnptions
(1978), 45-62.
15. Y. A. Abramovich, ,I special class of vectorlattices and its application to hvpercomplete
spaces, preprint.

MATHEMATIK, ETH-ZENTRUM, CH-8092 ZiJRICH, SWITZERLAND

This content downloaded from 128.235.251.160 on Mon, 19 Jan 2015 17:03:24 PM

All use subject to JSTOR Terms and Conditions