• Nie Znaleziono Wyników

(1)Independen e-pre alibers of measure algebras D.H.Fremlin &amp

N/A
N/A
Protected

Academic year: 2021

Share "(1)Independen e-pre alibers of measure algebras D.H.Fremlin &amp"

Copied!
6
0
0

Pełen tekst

(1)

Independen e-pre alibers of measure algebras

D.H.Fremlin & G.Plebanek

University of Essex, Col hester, England;Wro law University

Notation Wefollow Fremlin 06?. In parti ular, 



is the usualmeasure onf0;1g



,N



its null ideal and (B



;



) its measure algebra.

1De nition(a)LetAbeaBooleanalgebraand, ardinals. Wesaythat(;)isan

independen e-pre aliberpairof Aif whenever ha

 i

<

is afamilyof distin telements

of A thenthere isa 2[℄



su h thatfa



: 2 g isBoolean-independent. If (;) isan

independen e-pre aliberpairof(A;) wewillsaythatisanindependen e-pre aliber

of (A;).

(b)Let(A;) beameasurealgebraand, ardinals. Wesaythat(;)isameasure-

independen e-pre aliber pairof (A;) if whenever ha

 i

<

is a familyin A su h that

inf

<<



(a

 4a



) > 0 then there is a 2 [℄



su h that fa



:  2 g is Boolean-

independent. If (;) isa measure-independen e-pre aliber pair of (A;) we willsaythat

 is a measure-independen e-pre aliberof (A;).

Remark Of ourseanyindependen e-pre aliber (pair)ofa measurealgebraisameasure-

independen e-pre aliber (pair). If (;) is a measure-independen e-pre aliber pair of a

totally nite measure algebra (A;), and 

!

<  for every  < , then (;) is an

independen e-pre aliber pair of A. PPP If ha

 i

<

is any family of distin t elements of A,

then fa



: <g must have density  for the measure metri , and therefore must have a

metri allyisolated subset of size , whi h will have a Boolean-independent subset of size

. QQQ (See Proposition 11 below.)

2 Proposition! is a measure-independen e-pre aliber of every probability algebra.

proofLet(A;) beaprobabilityalgebraandha

n i

n2N

asequen einAsu hthat(a

m 4a

n )

Æ > 0 whenever m 6= n. Let u 2 L 2





be a luster point of ha

n i

n2N for T

s (L

2





;L 2



 ), and

set =[[0<u<1℄℄. Then 6=0. PPP??? Otherwise, u=a for some a. Now there must be

in nitely manyn su h that



(a4a

n )

R

(u a

n

)(a (1na))<

Æ

2 ,

whi h is impossible. XXXQQQ

We an therefore hoose indu tively a stri tly in reasing sequen e hn

k i

k2N

su h that

0 < (a

n

k

\b) < b whenever 0 6= b and b belongs to the algebra generated by

f g[fa

n

i

: i<kg; in whi h ase fa

n

k

:k2Ng will be Boolean-independent.

3 Lemma Let  be an un ountable ardinal and (A;) a probability algebra. Let D

be a losed subalgebra of A su h thatA is relatively atomless over D, anda2AnD; set

Æ =(a;D). Then there are a sequen e he

n i

n2N

in Aand a 2A su h that

TypesetbyA

M S-T

E X

(2)



e

n

= 1

2

for every n

thee

n

aresto hasti ally independent of ea h other and D

a, belong to the losed subalgebra Bgenerated by D[fe

n

:n2Ng

taking : B ! B to be the measure-preserving automorphism su h that

d =d for every d2D, e

0

=1n e

0 , e

n

=e

n

for n>0, then =

\e

0

1na, ne

0

a



 Æ.

proofDe neu2L 1

(D)bysayingthat R

d

u=(a \d)foreveryd 2D. Setd

0

=[[u>

1

2

℄℄.

Lete 0

0

abesu hthat(e 0

0

\d) = 1

2



(d\d

0

)foreveryd2D(331B);lete 0 0

0

1nabesu h

that(e 0

0

\d) = 1

2



(dnd

0

)for everyd 2D;sete

0

=(d

0 ne

0

0 )[e

00

0

,sothat(e

0

\d) = 1

2



d

for every d 2D. Now hoose e

n

, for n 1, su h that (e

n

\d)= 1

2



d for every d in the

algebrageneratedbyD[fe

i

:i <ng, andabelongstothe losed subalgebraB generated

byD[fe

n

: n2Ng. [To seethat thisis possible, startfromanysequen e hb

n i

n2N

in Aof

elementsof measure 1

2

sto hasti allyindependentof ea hotherandofD,andletBbethe

losed subalgebraof A generatedbyD[fa;e

0 g[fb

n

:n2Ng; useFremlin 02, 333Cto

seethat(B; B) anbeidenti edwiththeprobabilityalgebrafreeprodu tof(D

1

; D

1 )

and(B

!

;

!

),where D

1

isthe subalgebraof AgeneratedbyD[fe

0

g.℄ Be ause (B; B)

isequallyisomorphi totheprobabilityalgebrafree produ tof(D; D)and(B

!

;

! ),we

have a unique measure-preserving automorphism : B !B su h that d =d for every

d2D, e

0

=1n e

0

and e

n

=e

n

for n>0.

Set 0

=d

0 na,

0 0

=and

0

. Thenat leastone of 0

, 00

hasmeasureat least 1

2

Æ; allthis



; set =



[



, so that 2B and = .

If



= 0

then



 d

0

\e

0 ,



 d

0 ne

0

a, \e

0

= 0

is disjoint from a, ne

0

=

0

 aand   =2 0

Æ. If



= 00

then



\e

0

=0,

\e

0

=



e

0 nd

0

=e 00

0

1na,

ne

0

= 00

a and   =2  00

Æ.. So we have what we need.

4 Corollary If  is an in nite ardinal, (A;) a Maharam homogeneous probability

algebra, and ha

 i

<

a family in A su h that inf

<<



(a

 4a



) =Æ >0, then there are

a family he

n i

<;n2N

, a fun tion :! and a family h

 i

<

in A su h that



e

n

= 1

2

for all , n,

he

n i

<;n2N

is sto hasti allyindependent,

a

() ,



belong to the losed subalgebra B of A generated by fe

n

: <,

n2Ng for every ,

if 



: B ! B is the measure-preserving automorphism de ned by setting



 (e

0

)=1ne

0

and 

 (e

n )=e

n

for all (;n)6=(;0), then 





=

 ,







 1

2

Æ for every ,



\e

0

1na

() , ne

0

a

()

for every .

proof If D A is a losed subalgebra with Maharam type less than , then there must

be a  <  su h that (a



;D)  1

2

Æ. We an therefore hoose D

 , e

n ,

 ,



and ()

indu tively,asfollows. D

0

=f0;1g. GiventhatD



isthe losedsubalgebraofAgenerated

by fe

n

:  < , n 2 Ng, let () <  be su h that (a

()

;D

 ) 

1

Æ; using Lemma 3,

(3)

hoosee

n

,of measure 1

2

,sto hasti allyindependent of ea hotherandof D



,and

 ,su h

that

a

() and



belong to the losed subalgebra D

+1

generatedbyD

 [fe

n :

n2Ng,







 1

2 Æ,



\e

0

1na

() ,

 ne

0

a

() ,

taking

 :D

+1

!D

+1

to be the measure-preserving automorphism su h

that d = d for every d 2 D

 ,

 e

0

= 1n

 e

0 ,

 e

n

= e

n

for n> 0, then

 =

 .

LetB=D



be the losedsubalgebra of Ageneratedby thesto hasti allyindependent

family he

n i

<;n2N

. This works.

5 PropositionIf ,  are in nite ardinals, !

2

and (;) is a measure-pre aliber

pairof atotally nitemeasurealgebra(A;) thenitisameasure-independen e-pre aliber

pair of (A;).

proof (a) To begin with, suppose that (A;) is a Maharam homogeneous probability

algebra, that (;) is a measure-pre aliber pair of (A;) and that ha

 i

<

is a family in

A su h that (a

 4a



)  Æ > 0 whenever  < . Take he

n i

<;n2N

, B, :  !  and

h

 i

<

asinCorollary4. Forea h <letJ





betheminimalsubsetofN su hthat



belongs to the losed subalgebra generated by fe

j

:j 2Jg, and set L



=f:(;n)2Jg.

AsinFremlin06?, 534Je,(;0)2= J





. ByHajnal'sFreeSet Theoremthereis a

0 2[℄



su h that  2= L



for all distin t ,  2

0

. If I, J  C are disjoint nite sets and

=inf

2I[J



, then is sto hasti ally independent of the e

0

, for  2I[J, and



(inf

2J a

() n sup

2I a

()

)(inf

2I (e

0 na

() )\ inf

2J a

() ne

0 ))

(inf

2I (



\e

0

)\ inf

2J

 ne

0 ))

=(  \ inf

2I e

0 n sup

2J e

0 )=2

#(I[J)



 :

Now,be ause(;) isameasure-pre aliberpairof(A;), thereisa 2[

0



su h that

f



:2 g is entered, in whi h ase ha

() i

2

is Boolean-independent.

(b) For the general ase, we have only to note that, for a totally nite measure al-

gebra (A;), (;) is an (independen e-)measure-pre aliber pair of (A;) i it is an

(independen e-)measure-pre aliber pair of every Maharam homogeneous prin ipal ideal

of (A;) ( f. Fremlin 06?, 524Ha).

6 Proposition If !     and (;) is a measure-independen e-pre aliber pair of

every probabilityalgebra thenitis ameasure-pre aliber pair of every probabilityalgebra.

proofLet(A;) beaprobabilityalgebraandha

 i

<

afamilyinAsu hthatinf

<



a



=

Æ >0. Let(C;) betheprobabilityalgebrafreeprodu tof(A;) and(B



;



),andhe

 i

<

a sto hasti ally independent family in B



of elements of measure 1

2

. Set



=a

 e

 for

ea h ; then ( 

 4



) = (a

 [a



)  Æ whenever  < . There is therefore a 2 [℄



su h thath i is Boolean-independent, in whi h ase fa : 2 g must be entered.

(4)

7 Lemma Let n 1 be an integer,  a regular un ountable ardinal, X  (f0;1g n

)



a losed set, and f



: f0;1g n

! f0;1g a fun tion. Set f(x) = hf



(x())i

<!

1

for x 2 X.

If f[X℄ = f0;1g



, there are i < n and C 2 [℄



su h that 

iC

[X℄ = f0;1g C

, where



iC

(x)()=x()(i) for x2X,  2C, i<n.

proof Indu e on n. If n = 1 then every f



has to be surje tive, therefore bije tive, and

f is a bije tion, so we an take C = , i = 0. For the indu tive step to n+1, identify

(f0;1g n+1

)



with (f0;1g n

)



 f0;1g



, so that we have X  (f0;1g n

)



 f0;1g



and

f



: f0;1g n

f0;1g ! f0;1g su h that f[X℄ = f0;1g



, where f(x;y)()= f



(x();y())

for  < , x 2 (f0;1g n

)



, y 2 f0;1g



. Let X 0

 X be a minimal losed set su h that

f[X 0

℄=f0;1g



.

SetY =fy:(x;y)2X 0

g, so that Y is a losed subset of f0;1g



. Set

A=f : <; 8y 2Y 8I 2[℄

<!

9y 0

; y 00

2Y;

y 0

I =y 00

I =yI; y 0

()=0; y 00

()=1g:

If I 2 [A℄

<!

and u 2 f0;1g I

there is a y 2 Y su h that yI = u (indu e on maxI), so

fyA:y 2Yg=f0;1g A

and if #(A)= we antake C =A,i =n and stop. Otherwise,

for  2 nA take y



2 Y, I

 2 [℄

<!

and v() 2 f0;1g su h that y() = v() whenever

y 2 Y and yI



= y



I



. Let J 2 [℄

<!

, u 2 f0;1g J

be su h that B = f :  2 nA,

I



=J, y



I



=ug has ardinal . For 2B, s2f0;1g n

setg



(s)=f



(s;v()).

Now onsider f(x;y): (x;y)2 X 0

, yJ = ug. This is a non-empty open subset of X 0

and fX 0

is irredu ible, so there is an open ylinder set V  f0;1g



su h that yJ = u

whenever (x;y) 2 X 0

and f(x;y) 2 V. Let K 2 [℄

<!

be su h that V is determined

by oordinates in K. Set D = BnK, Z = fxD : (x;y) 2 X 0

g, g(z) = hg



(z())i

2D

for z 2 Z. If w 2 f0;1g D

, there is a w 0

2 V su h that w 0

D = w; there is an element

(x;y)2X 0

su h that f(x;y)=w 0

; now yJ =u so yB=vB and

g



(x())=f(x();y())=w 0

()=w()

for every  2D, that is, g(xD)=w. Thus g[Z℄=f0;1g D

, while Z (f0;1g n

)



.

By the indu tive hypothesis, there are C 2 [D℄



and an i < n su h that 

iC [Z℄ =

f0;1g C

; now, re-interpreting the formula 

iC

appropriately, we have 

iC

[X℄ = f0;1g C

and the indu tion ontinues.

8 PropositionLetM beaset, aregularun ountable ardinalandX a losedsubset

of f0;1g M

. If there is a ontinuous surje tion h :X ! f0;1g



, then there is a C 2 [M℄



su h thatx 7!xC :X !f0;1g C

is surje tive.

proof Shrinking X if ne essary, we may suppose that h is irredu ible. For ea h  < ,

fx : x 2 X, h(x)() = 1g is an open-and- losed set in X, so there is a nite set I



 

su h that h(x)() = h(x 0

)() whenever x, x 0

2 X and xI



= x 0

I



. Let A 2 [℄



be

su h that hI

 i

2A

is a onstant-size -system with root I say. Let u be any member of

fxI :x2Xg;thenthereisa ylinder setV f0;1g



su h thatxI =uwheneverx2X

and h(x)2 V; shrinking A slightlyif ne essary, we an suppose that V is determinedby

oordinatesin nA. For  2A,set J



=I



nI and de nef



:f0;1g J



!f0;1gby saying

that f (v) =h(x)() whenever x 2 X, xI = u and xJ =v. Now observe that for any

(5)

w 2 f0;1g A

we have a w 0

2 V extending w, so that there is an x 2 X with h(x) = w 0

,

xI =u and f

 (xJ



)=w() for every 2A.

We an therefore apply Lemma 7 to fx

S

2A J



: x 2 Xg, identifying S

2A J

 with

nAwherenisthe ommonvalueof #(J



)for 2A,toseethatthereisanun ountable

C  S

2A J



su h thatfxC :x2Xg=f0;1g



.

9CorollaryIf!

1

hasHaydon'spropertythen!

1

isameasure-independen e-pre aliber

of every probabilityalgebra.

proofLet (A;) be aprobabilityalgebraand ha

 i

<!

1

a familyin A su h thatinf

<<!

1



(a



;a



)>0. Thenwehave a Radonprobabilitymeasure  on f0;1g

!

1

de ned by saying

that

fx:xI =ug=(inf

2I;u()=1 a

 n sup

2I;u()=0 a

 )

whenever I 2[!

1

<!

and u 2 f0;1g I

. Let Z be the support of . Sin e  has Maharam

type !

1

, there is a ontinuous surje tion from Z onto [0;1℄

!

1

; so we an nd a losed

subset Z 0

of Z and a ontinuous surje tion h : Z 0

! f0;1g

!

1

. By Proposition 8, there

is a C 2 [!

1

!

1

su h that fzC : z 2 Z 0

g = f0;1g C

. But this means that ha

 i

2C is

Boolean-independent.

10 Proposition Suppose that there is a family hW

 i

<!1 in N

!1

su h that every

losedsubset off0;1g

!

1

n S

<!

1 W



is s attered. Then!

1

isnotameasure-independen e-

pre aliber of every probability algebra.

proof As in Fremlin 06?, 534N (following Plebanek 97), there is a zero-dimensional

ompa tHausdor spa eX su hthat!

1

2Mah

R

(X)butthereisno ontinuoussurje tion

from X onto f0;1g

!

1

. Let  be a Maharam homogeneous Radon probabilitymeasure on

X withMaharam type !

1

and (A;) itsmeasure algebra. Be ause X is zero-dimensional,

thereisafamilyhK

 i

<!1

ofopen-and- losedsetsinX su hthat(K

 4K

 )

1

3

whenever

 < ; now hK

 i

<!

1

has no Boolean-independent subfamily of size !

1

, so hK



 i

<!

1 has

no Boolean-independent subfamily of size !

1 .

11 Proposition Suppose that  and  are ardinals su h that max(2

;

!

) < f 

2



. Then  is an independen e-pre aliber of every totally nite measure algebra.

proof D

zamonja & Plebanek p04, Theorem 6.3.

Remark Note that if we have only



!

< f2



then  is a measure-pre aliber of every measure algebra (D

zamonja & Plebanek p04,

4.7; Fremlin p06?, 524V), therefore a measure-independen e-pre aliber of every proba-

bility algebra (D

zamonja & Plebanek p04, 6.6). If 

!

<  for every  <, then any

family of size  in any metri spa e has a dis rete subfamily of size , so  will be an

independen e-pre aliber of every probability algebra. But the result here also overs the

ase =2 2

=! .

(6)

12ProblemFromFremlin 06?,534L,andPropositions2,5and6above,weseethat

for  !

2

the followingare equiveridi al:

 is a measure-independen e-pre aliber of every probability algebra;

 is a measure-pre aliber of every probability algebra;

 has Haydon's property.

By Proposition 10 and the remarks in the notes to x534 of Fremlin 06?, it is possible

that

!

1

is a measure-pre aliber of every probabilityalgebra,

!

1

is not a measure-independen e-pre aliber of every probabilityalgebra.

ByCorollary9,weseethatif!

1

hasHaydon'spropertythenitisameasure-independen e-

pre aliber of every probabilityalgebra. Butwedo notknow whetherthe onverse istrue.

Referen es

Dzamonja M. & Plebanek G. [p04℄ `Pre alibre pairs of measure algebras', 2.4.04

Fremlin D.H. [02℄ Measure Theory, Vol. 3: Measure Algebras. Torres Fremlin, 2002.

FremlinD.H.[03℄MeasureTheory,Vol. 4: Topologi alMeasureTheory. TorresFremlin,

2003.

Fremlin D.H. [06?℄ Measure Theory, Vol. 5: Set-theoreti measure theory, in prepara-

tion. Partialdraftsavailableviahttp://www.essex.a .u k/m ath s/st aff /fr eml in/m t.h tm;

paragraph numbers indi ated here are not to be trusted.

Plebanek G. [97℄ `Non-separable Radon measures and small ompa t spa es', Funda-

menta Math. 153(1997) 25-40.

Cytaty

Powiązane dokumenty

A bstract. The property of weak compactness for sequences of finite Borel measures on the real line is extended to a sequence of families of Borel measures on R and discussed in

The paper aims at characterizing a class of these distributions p of random variables X t for which at a suitable choice of norming operators An and vectors

[r]

W tej ostatniej robią się otwory, roślinkę wsadza się do przedziurawionego korka i umocowuje się j ą zapomocą waty, poczem korek wraz z rośliną wsadza

CCQE large axial mass problem multinucleon ejection hypothesis sources of multinucleon nal states microscopic models  an overview a simplied Marteau model generation of nal

• The objective reality in WSS operating are different types of undesirable events which cause the deterioration of water quality (final product) and lower the level

If CH holds, then there is a compact 0-dimensional, hereditarily separable and hereditarily Lindel¨ of space X and a Radon proba- bility measure µ on X such that the measure algebra

Provided that f is polynomial-like we prove that the harmonic measure on J is mutually absolutely contin- uous with the measure of maximal entropy if and only if f is