Motivation Polish- and Banach-representability Representability inc0and`1
Representations of ideals in Banach spaces
Barnabás Farkas1
2joint work with
3Piotr Borodulin-Nadzieja2 and Grzegorz Plebanek2
4First Brazilian Workshop in Geometry of Banach Spaces
1Kurt Gödel Research Center, University of Vienna
2Mathematical Institute, University of Wrocław
Motivation Polish- and Banach-representability Representability inc0and`1
Classical summable ideals
Definition
Leth:ω→[0,∞)be a sequence such thatP
n∈ωh(n) =∞.
Then thesummable ideal associated tohis Ih=
A⊆ω:X
n∈A
h(n)<∞
(anFσ P-ideal).
We have to change the definition:A∈ Ih0 if the sum of(h(n))n∈A
isunconditionally convergent, that is, the net XhA=
sh(F) =X
n∈F
h(n) :F ∈[A]<ω
is convergent.
But we still have the same family of ideals:A∈ Ih0 iffA∈ I|h|. . .
Motivation Polish- and Banach-representability Representability inc0and`1
Classical summable ideals
Definition
Leth:ω→[0,∞)be a sequence such thatP
n∈ωh(n) =∞.
Then thesummable ideal associated tohis Ih=
A⊆ω:X
n∈A
h(n)<∞
(anFσ P-ideal).
Whyh:ω →[0,∞)? What ifh:ω→R?
We have to change the definition:A∈ Ih0 if the sum of(h(n))n∈A
isunconditionally convergent, that is, the net XhA=
sh(F) =X
n∈F
h(n) :F ∈[A]<ω
is convergent.
But we still have the same family of ideals:A∈ Ih0 iffA∈ I|h|. . .
Classical summable ideals
Definition
Leth:ω→[0,∞)be a sequence such thatP
n∈ωh(n) =∞.
Then thesummable ideal associated tohis Ih=
A⊆ω:X
n∈A
h(n)<∞
(anFσ P-ideal).
Whyh:ω →[0,∞)? What ifh:ω→R?
We have to change the definition:A∈ Ih0 if the sum of(h(n))n∈A
isunconditionally convergent, that is, the net XhA=
sh(F) =X
n∈F
h(n) :F ∈[A]<ω
is convergent.
But we still have the same family of ideals:A∈ Ih0 iffA∈ I|h|. . .
Motivation Polish- and Banach-representability Representability inc0and`1
Generalized summable ideals
h : ω → ?
Definition
LetGbe a Polish Abelian group andh:ω→Gsuch thatP his not convergent. Then thegeneralized summable ideal
associated toGandhis IhG =n
A⊆ω:X
hA is convergento .
We say that an idealJ onωisrepresentable inGif there is anh:ω →Gsuch thatJ =IhG. IfCis a class of groups then J is C-representableif it is representable in aG∈C.
Motivation Polish- and Banach-representability Representability inc0and`1
Generalized summable ideals
h : ω → ?
Definition
LetGbe a Polish Abelian group andh:ω→Gsuch thatP his not convergent. Then thegeneralized summable ideal
associated toGandhis IhG =n
A⊆ω:X
hA is convergento .
anh:ω →Gsuch thatJ =IhG. IfCis a class of groups then J is C-representableif it is representable in aG∈C.
Motivation Polish- and Banach-representability Representability inc0and`1
Generalized summable ideals
h : ω → ?
Definition
LetGbe a Polish Abelian group andh:ω→Gsuch thatP his not convergent. Then thegeneralized summable ideal
associated toGandhis IhG =n
A⊆ω:X
hA is convergento .
We say that an idealJ onωisrepresentable inGif there is anh:ω →Gsuch thatJ =IhG. IfCis a class of groups then J is C-representableif it is representable in aG∈C.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Example
J is representable inRn(or inTn) iffJ is a summable ideal.
J is representable inRω (or inTω) iffJ is the intersection of countable many summable ideals. These ideals are not necessarilyFσ.
Example
If(Gk)k∈ω is a sequence of non-trivial discrete Abelian groups, thenJ is representable inQ
k∈ωGk iffJ is representable inZω2 iff there is a family{Xk:k ∈ω} ⊆[ω]ω such that
J = n
A⊆ω:∀k ∈ω|A∩Xk|< ω o
.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Example
J is representable inRn(or inTn) iffJ is a summable ideal.
Example
J is representable inRω (or inTω) iffJ is the intersection of countable many summable ideals. These ideals are not necessarilyFσ.
Example
If(Gk)k∈ω is a sequence of non-trivial discrete Abelian groups, thenJ is representable inQ
k∈ωGk iffJ is representable inZω2 iff there is a family{Xk:k ∈ω} ⊆[ω]ω such that
J = n
A⊆ω:∀k ∈ω|A∩Xk|< ω o
.
Examples
Example
J is representable inRn(or inTn) iffJ is a summable ideal.
Example
J is representable inRω (or inTω) iffJ is the intersection of countable many summable ideals. These ideals are not necessarilyFσ.
Example
If(Gk)k∈ω is a sequence of non-trivial discrete Abelian groups, thenJ is representable inQ
k∈ωGk iffJ is representable inZω2 iff there is a family{Xk:k ∈ω} ⊆[ω]ω such that
J = n
A⊆ω:∀k ∈ω|A∩Xk|< ω o
.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Example
Thedensity zero ideal Z =
A⊆ω: |A∩n|
n →0
=
A⊆ω: |A∩[2n,2n+1)|
2n →0
(anFσδ P-ideal) is representable inc0:
Examples
h(0) =(0,0, 0 , 0 , 0 , . . .) h(1) =(0,1, 0 , 0 , 0 , . . .) h(2) =(0,0,1/2, 0 , 0 , . . .) h(3) =(0,0,1/2, 0 , 0 , . . .) h(4) =(0,0, 0 ,1/4, 0 , . . .) h(5) =(0,0, 0 ,1/4, 0 , . . .) h(6) =(0,0, 0 ,1/4, 0 , . . .) h(7) =(0,0, 0 ,1/4, 0 , . . .)
... ... ... ... ... ... IfA⊆ω thenP
hAis convergent ⇐⇒
X
n∈A
h(n) =
0,|A∩[2,4)|
2 ,|A∩[4,8)|
4 , . . .
∈c0 ⇐⇒ A∈ Z.
Motivation Polish- and Banach-representability Representability inc0and`1
Why are we doing this?
(i) Our approach reveals some “geometric” properties of ideals and therefore it can be helpful in classifying certain classes of ideals.
(ii) These methods can be useful in providing new interesting examples of non-pathological analytic P-ideals (see later).
(iii) Representability of certain ideals in a Banach space can be seen as a combinatorial property of the space itself and this may lead us to develop new methods in the theory of Banach spaces.
Polish- and Banach-representability
Theorem
J is Polish-representable iffJ is an analytic P-ideal.
Theorem
J is Banach-representable iffJ is a non-path. analytic P-ideal.
Motivation Polish- and Banach-representability Representability inc0and`1
Analytic P-ideals
A functionϕ:P(ω)→[0,∞]is
asubmeasure(onω) ifϕ(∅) =0,ϕis monotonic, subadditive, andϕ({n})<∞for everyn∈ω;
lower semicontinuous(lsc, in short) if ϕ(X) =limn→∞ϕ(X ∩n)for eachX ⊆ω.
Ifϕis an lsc submeasure then let Fin(ϕ)=
n
A⊆ω:ϕ(A)<∞o
(anFσ ideal).
Exh(ϕ)= n
A⊆ω: lim
n→∞ϕ(A\n) =0 o
(anFσδ P-ideal).
Analytic P-ideals
Theorem (Mazur, Solecki)
LetJ be an ideal onω. Then the following are equivalent:
(i) J is an analyticP-ideal.
(ii) J =Exh(ϕ)for some (finite) lsc submeasureϕ.
(iii) There is a Polish group topology onJ (with respect to4)
such that the Borel structure of this topology coincides with the Borel structure inherited fromP(ω).
Furthermore,J is anFσ P-ideal iffJ =Fin(ϕ) =Exh(ϕ)for some lsc submeasureϕ.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Classical summable ideals.
(Generalized) Density ideals: Let~µ= (µn)n∈ω be a sequence of (sub)measures onω with pairwise disjoint finite supports with lim supn→∞µn(ω)>0. Then the (generalized) density ideal associated to~µis
Z~µ=
A⊆ω:µn(A)→0 . The trace of the null ideal:
tr(N)=
A⊆2<ω:λ{f ∈2ω:∃∞n f n∈A}=0 . Farah’s ideal:
JF =
A⊆ω:X
n∈ω
min{n,|A∩[2n,2n+1)|} n2 <∞
.
Tsirelson ideals.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Classical summable ideals.
(Generalized) Density ideals: Let~µ= (µn)n∈ω be a sequence of (sub)measures onω with pairwise disjoint finite supports with lim supn→∞µn(ω)>0. Then the (generalized) density ideal associated to~µis
Z~µ=
A⊆ω:µn(A)→0 .
tr(N)=
A⊆2<ω:λ{f ∈2ω:∃∞n f n∈A}=0 . Farah’s ideal:
JF =
A⊆ω:X
n∈ω
min{n,|A∩[2n,2n+1)|} n2 <∞
.
Tsirelson ideals.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Classical summable ideals.
(Generalized) Density ideals: Let~µ= (µn)n∈ω be a sequence of (sub)measures onω with pairwise disjoint finite supports with lim supn→∞µn(ω)>0. Then the (generalized) density ideal associated to~µis
Z~µ=
A⊆ω:µn(A)→0 . The trace of the null ideal:
tr(N)=
A⊆2<ω:λ{f ∈2ω:∃∞n f n∈A}=0 .
Farah’s ideal: JF =
A⊆ω:X
n∈ω
min{n,|A∩[2n,2n+1)|} n2 <∞
.
Tsirelson ideals.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Classical summable ideals.
(Generalized) Density ideals: Let~µ= (µn)n∈ω be a sequence of (sub)measures onω with pairwise disjoint finite supports with lim supn→∞µn(ω)>0. Then the (generalized) density ideal associated to~µis
Z~µ=
A⊆ω:µn(A)→0 . The trace of the null ideal:
tr(N)=
A⊆2<ω:λ{f ∈2ω:∃∞n f n∈A}=0 . Farah’s ideal:
JF =
A⊆ω:X
n∈ω
min{n,|A∩[2n,2n+1)|}
n2 <∞
.
Motivation Polish- and Banach-representability Representability inc0and`1
Examples
Classical summable ideals.
(Generalized) Density ideals: Let~µ= (µn)n∈ω be a sequence of (sub)measures onω with pairwise disjoint finite supports with lim supn→∞µn(ω)>0. Then the (generalized) density ideal associated to~µis
Z~µ=
A⊆ω:µn(A)→0 . The trace of the null ideal:
tr(N)=
A⊆2<ω:λ{f ∈2ω:∃∞n f n∈A}=0 . Farah’s ideal:
JF =
A⊆ω:X
n∈ω
min{n,|A∩[2n,2n+1)|}
n2 <∞
.
Tsirelson ideals.
Motivation Polish- and Banach-representability Representability inc0and`1
Non-pathological ideals
A submeasureϕis non-pathologicalif for everyA⊆ω ϕ(A) =sup
µ(A) :µis a measure on P(ω) and µ≤ϕ . An analytic P-idealJ isnon-pathologicalifJ =Exh(ϕ)for some non-pathological lsc submeasureϕ.
An analytic P-idealJ is non-pathological iffJ X ≤K Z(i.e. there is anf :ω→X such thatf−1[A]∈ Z for everyA∈ J) for everyX ∈ P(ω)\ J.
Motivation Polish- and Banach-representability Representability inc0and`1
Non-pathological ideals
A submeasureϕis non-pathologicalif for everyA⊆ω ϕ(A) =sup
µ(A) :µis a measure on P(ω) and µ≤ϕ . An analytic P-idealJ isnon-pathologicalifJ =Exh(ϕ)for some non-pathological lsc submeasureϕ.
Theorem (Hrušák)
An analytic P-idealJ is non-pathological iffJ X ≤K Z(i.e.
there is anf :ω→X such thatf−1[A]∈ Z for everyA∈ J) for everyX ∈ P(ω)\ J.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in c
0Proposition
An idealJ is representable inc0iff there is a family
{µk :k ∈ω}of measures onωsuch thatJ =Exh(ϕ)where ϕ=sup{µk :k ∈ω}and{k :n∈supp(µk)}is finite for eachn.
Density ideals are representable inc0.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in c
0Proposition
An idealJ is representable inc0iff there is a family
{µk :k ∈ω}of measures onωsuch thatJ =Exh(ϕ)where ϕ=sup{µk :k ∈ω}and{k :n∈supp(µk)}is finite for eachn.
Example
Density ideals are representable inc0.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in c
0We show thattr(N), Farah’s ideal, and Tsirelson ideals are not representable inc0.
IfJ is representable inc0andtotally bounded(that is, if J =Exh(ϕ)thenϕ(ω)<∞), thenJ is a generalized density ideal.
Corollary
tr(N)is not representable inc0. Proof:
1.tr(N)is not a generalized density ideal (because e.g. it is summable-like).
2. Hrusak+Hernandez-Hernandez:tr(N)is totally bounded (because e.g.s(tr(N)) =ω).
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in c
0We show thattr(N), Farah’s ideal, and Tsirelson ideals are not representable inc0.
Proposition
IfJ is representable inc0andtotally bounded(that is, if J =Exh(ϕ)thenϕ(ω)<∞), thenJ is a generalized density ideal.
Corollary
tr(N)is not representable inc0. Proof:
1.tr(N)is not a generalized density ideal (because e.g. it is summable-like).
2. Hrusak+Hernandez-Hernandez:tr(N)is totally bounded (because e.g.s(tr(N)) =ω).
Representability in c
0We show thattr(N), Farah’s ideal, and Tsirelson ideals are not representable inc0.
Proposition
IfJ is representable inc0andtotally bounded(that is, if J =Exh(ϕ)thenϕ(ω)<∞), thenJ is a generalized density ideal.
Corollary
tr(N)is not representable inc0. Proof:
1.tr(N)is not a generalized density ideal (because e.g. it is summable-like).
2. Hrusak+Hernandez-Hernandez:tr(N)is totally bounded (because e.g.s(tr(N)) =ω).
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in c
0Theorem
If a tallFσ ideal is representable inc0iff it is a summable ideal.
Corollary
Farah’s ideal and Tsirelson ideals are not representable inc0.
Representability in c
0Theorem
If a tallFσ ideal is representable inc0iff it is a summable ideal.
Corollary
Farah’s ideal and Tsirelson ideals are not representable inc0.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in `
1Theorem
There is a tallFσ P-ideal (theRademacher-ideal) which is representable in`1but not a summable ideal.
Proof (sketch): Our first, still unfinished (but 99% working,) idea was to build blocks of the form
(0, . . . ,0, ε0/an, . . . , εn−1/an,0, . . .)whereεi ∈ {±1}andanis an appropriate sequence tending to∞. Problem: If we add all 2npossible new elements to our family, then we should understand the following constant:
infnmax{kP
Fk:∅6=F⊆S}
P{kvk:v∈S} :∅ 6=S∈[X\ {0}]<ωo .
Piotr’s refinement: At thenth stage we add “Rademacher-like” vectors only and then we can apply Khintchine’s inequality etc.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in `
1Theorem
There is a tallFσ P-ideal (theRademacher-ideal) which is representable in`1but not a summable ideal.
idea was to build blocks of the form
(0, . . . ,0, ε0/an, . . . , εn−1/an,0, . . .)whereεi ∈ {±1}andanis an appropriate sequence tending to∞. Problem: If we add all 2npossible new elements to our family, then we should understand the following constant:
infnmax{kP
Fk:∅6=F⊆S}
P{kvk:v∈S} :∅ 6=S∈[X\ {0}]<ωo .
Piotr’s refinement: At thenth stage we add “Rademacher-like” vectors only and then we can apply Khintchine’s inequality etc.
Motivation Polish- and Banach-representability Representability inc0and`1
Representability in `
1Theorem
There is a tallFσ P-ideal (theRademacher-ideal) which is representable in`1but not a summable ideal.
Proof (sketch): Our first, still unfinished (but 99% working,) idea was to build blocks of the form
(0, . . . ,0, ε0/an, . . . , εn−1/an,0, . . .)whereεi ∈ {±1}andanis an appropriate sequence tending to∞. Problem: If we add all 2npossible new elements to our family, then we should understand the following constant:
infnmax{kP
Fk:∅6=F⊆S}
P{kvk:v∈S} :∅ 6=S∈[X\ {0}]<ωo .
Piotr’s refinement: At thenth stage we add “Rademacher-like” vectors only and then we can apply Khintchine’s inequality etc.
Representability in `
1Theorem
There is a tallFσ P-ideal (theRademacher-ideal) which is representable in`1but not a summable ideal.
Proof (sketch): Our first, still unfinished (but 99% working,) idea was to build blocks of the form
(0, . . . ,0, ε0/an, . . . , εn−1/an,0, . . .)whereεi ∈ {±1}andanis an appropriate sequence tending to∞. Problem: If we add all 2npossible new elements to our family, then we should understand the following constant:
infnmax{kP
Fk:∅6=F⊆S}
P{kvk:v∈S} :∅ 6=S∈[X\ {0}]<ωo .
Piotr’s refinement: At thenth stage we add “Rademacher-like”
vectors only and then we can apply Khintchine’s inequality etc.
Motivation Polish- and Banach-representability Representability inc0and`1
Questions
Question
General version: Is there any reasonable characterization of Rω-, orc0-, or`1-representable ideals?
Reasonable versions:
Which ideals can be covered by a summable ideal?
Are all`1-representable idealsFσ? Question
Assume that all non-pathological analytic P-ideals are representable in a Banach spaceX. Does it imply thatX is universal for the class of all separable Banach spaces?
Question
What can we say about ideals representable in (completions of) weak topologies?