See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233402669

Base Tree Property

Article  in  Order · November 2012

DOI: 10.1007/s11083-013-9316-2 · Source: arXiv

CITATIONS READS
4 56

3 authors, including:

Michal Doucha Michael Hrušák

The Czech Academy of Sciences Universidad Nacional Autónoma de México
47 PUBLICATIONS   228 CITATIONS    126 PUBLICATIONS   907 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Ultrafilter, almost disjoint families and similar structures View project

Strong measure zero View project

All content following this page was uploaded by Michael Hrušák on 18 February 2014.

The user has requested enhancement of the downloaded file.

 BASE TREE PROPERTY

BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

arXiv:1211.3350v3 [math.LO] 3 Mar 2013

Abstract. Building on previous work from [1] we investigate σ-

closed partial orders of size continuum. We provide both an inter-
nal and external characterization of such partial orders by showing
that (1) every σ-closed partial order of size continuum has a base
tree and that (2) σ-closed forcing notions of density c correspond
exactly to regular suborders of the collapsing algebra Coll(ω1 , 2ω ).
We further study some naturally ocurring examples of such par-
tial orders.

Introduction
A partially ordered set (P, ≤) is σ-closed if every countable decreas-
ing sequence of elements of P has a lower bound. In this note we study
σ-closed partial orders of size continuum. Orders of this type natu-
rally arise in combinatorial and descriptive set-theory, topology and
analysis.
An essential example is the collapsing algebra Coll(ω1 , 2ω ), i.e. the
completion, in the sense of Boolean algebra, of the complete binary
tree of height ω1 . This forcing notion has several presentations:
• (Fn(ω1 , {0, 1}, ω1), ⊇) - ordering for adding a new subset of ω1 ,
• (Fn(ω1 , R, ω1 ), ⊇) - ordering for the consistency of the contin-
uum hypothesis,
• (Fn(2ω , {0, 1}, ω1), ⊇) - ordering for adding c-many subsets of
ω1 ,
• the natural ordering for adding a ⋄-sequence,
• Jech’s forcing for adding a Suslin tree by countable conditions.

2000 Mathematics Subject Classification. 03E15, 03E17, 03E05, 03E35.

Key words and phrases. Forcing, Boolean algebras, Base tree.
The research of the second author was partially supported by NSF grant DMS
0801114 and grant IAA100190902 of Grant Agency of the Academy of Sciences of
the Czech Republic.
The research of the third author was partially supported by PAPIIT grant
IN102311 and CONACYT grant 177758.
1
2 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

All these orderings are forcing equivalent, in fact, they have isomor-
phic base trees (see Theorem 2.1 for the term base tree). We refer to
[13] for definitions of these orderings.
Consider now the set [ω]ω of all infinite sets of natural numbers
ordered by inclusion. This order is not σ-closed, but it is also not sep-
arative.1 The separative quotient of ([ω]ω , ⊆) are the positive elements
in the Boolean algebra P(ω)/f in. In [1] the surprising fact that also
P(ω)/f in has a base tree was established. It was then studied in [8],
[10], [19].
Since then many other naturally occuring examples were studied
([2],[3]) and in each case the methods of [1] were used to prove the
corresponding Base Tree Theorem.
In this note we prove this general fact for all partial orders with a
dense σ-closed subset of size continuum. We also identify the σ-closed
forcings of size continuum as the regular subalgebras of the collapsing
algebra Coll(ω1 , 2ω ).
We then present some of the standard examples and review the rel-
evant published results.
We note that similar but somewhat more general notions were stud-
ied in [12].

1. Main results
The height of a partial order (P, ≤), h(P ) shortly, is the minimal
cardinality of a system of open dense subsets of P such that the inter-
section of the system is not dense. An equivalent definition involves
maximal antichains: h(P ) is equal to the minimal cardinality of a sys-
tem of maximal antichains from P that do not have a common refine-
ment. For a Boolean algebra B we define h(B) as the height of the
ordering (B \ {0}, ≤), where ≤ is the canonical ordering on B. If B
is complete, it coincides with its distributivity number. Recall that the
distributivity number of B is the least cardinal κ such that there ex-
matrix hu(α, β) W
ists a W
V : β ∈ Iα , αV< κi of elements from B such that
Q
α<κ β∈Iα u(α, β) 6= f ∈ α<κ Iα α<κ u(α, f (α)). We will deal mostly
with non-atomic orderings but for completeness we allow atomic order-
ings in the definition too. Thus, if (P, ≤) is atomic, i.e. there is a set of
minimal elements such that every other element is above one of them,
then we set h(P ) = ∞. Note that for non-atomic orderings height is
always a regular cardinal.

1Recallthat a partial order P is separative if whenever p, q are elements of P

such that p 6≤ q, there is an r ∈ P such that r ≤ p and r ⊥ q.
 BASE TREE PROPERTY 3

The height is a forcing invariant, that means every dense subset of

an ordering has the same height. In particular, h(P ) = h(RO(P )).
Fact 1.1. For an ordering P , h(P ) is the minimal cardinal κ such that
forcing with P adds a new function from κ to ordinals. In particular,
forcing with P preserves all cardinals less or equal to κ.
An ordering P is homogeneous in h (homogeneous in height) if for
every p ∈ P h(↓ p) = h(P ). The following proposition shows that every
partial order can be decomposed into factors homogeneous in height.
For complete Boolean algebras there is a canonical such decomposition.
Proposition
Q 1.2. Let B be a complete Boolean algebra. Then B ∼ =
b∈I B ↾ b, where I is a partition of unity and B ↾ b is homogeneous
in height for every b ∈ I.
Moreover, h(B ↾ a) 6= h(B ↾ b) if a 6= b for a, b ∈ I.
W
Proof. Let A be the set of all atoms, then B ↾ A is the first factor
homogeneous in the height ∞. W
Next, we work W with an atomless complete algebra B0 = B ↾ (− A)
(B ∼ = B0 × B ↾ A). T Let (Dα )α<h(B0 ) be the system of open dense
subsets of B0 such that α<h(B0 ) Dα is not dense. Let A1 be the subset
T
of elements of B0 witnessing the non-density, i.e. ↓ a∩ α<h(B0 ) Dα = ∅
for every a ∈ A1 . We claim that for every a ∈ A1 B ↾ a is homogeneous
in the height (with height h(B0 )). Assume not, then there is some
a ∈ A1 and b < a such that h(B ↾ b) < h(B ↾ a). Thus, there is a
system (Sα )α<h(B↾b) of open dense subsets of B ↾ b with a non-dense
intersection below b. However, if we set Dα = Sα ∪ B0 \ ↓ b then we get
a system of open dense subsets in B0 without a dense intersection less
than h(B0 ), that is aWcontradiction.
W We take the join A1 of all elements from A1 and the factor B ↾
A1 is homogeneous
W in height. We continue with the remainder B1 =
B0 ↾ (− A1 ) and in the same way get a set A2 of elements witnessing
the non-density of the intersection of a system of open dense subsets of
size h(B1 ). It is possible that h(B1 ) = h(B0 ). In this case, we joinWthe
elements of A2 with the elements of A1 . In the opposite case, B1 ↾ A2
is a new factor homogeneous in height.
We continue similarly until we treat all elements of B. We end up
with the desired decomposition. 
Definition 2 (Base tree property). An ordering (P, ≤) has the Base
Tree Property (we shall shortly say it has the BT-property) if it contains
a dense subset D ⊆ P with the following three properties:
- it is atomless; i.e. for every d ∈ D there are elements d1 , d2 ∈ D
below d such that d1 ⊥ d2
4 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

- it is σ-closed
- |D| ≤ c
It can be easily seen that assuming the Continuum Hypothesis,
all partial orders with the BT-property are forcing equivalent with
Coll(ω1 , 2ω ) and, consequently have a base tree. In fact, the follow-
ing is true in ZFC.
Theorem 2.1 (The base tree theorem). Let (P, ≤) be an ordering
homogeneous in height with the BT-property. Then there are h(P )
maximal antichains (Tα )α<h(P ) ⊆ P such that:
S
(i) (T = α<h(P ) Tα , ≥) is a tree of height h(P ), where Tα is the α-th
level of the tree,
(ii) each t ∈ T has c immediate successors,
(iii) T is dense in P .
T is called a base tree of P .

Proof. We need to work with some dense subset guaranteed by the

definition of the BT-property rather than with P itself. To avoid in-
troducing new symbols and sets, we assume P itself has the properties.
We use the definition of height. So we can find a system (Aα )α<h(P )
of dense open subsets with a non-dense intersection. However, we need
to ensure the intersection to be T empty. For this, we will work in the
completion RO(P ). Suppose α<h(P ) Aα is not empty. Consider the
WT
element a = ( α<h(P ) Aα ) ∈ RO(P ). Since P is homogeneous in
height, also RO(P ) is homogeneous in height, and there is a system
(Āα )α<h(P ) of dense open sets of P below a of the same size such that
their intersection is not dense below a. We replace ↓ a ∩ Aα by Āα (i.e.
(Aα \ ↓ a)∪ Āα ). We get a new system (A′α )α<h(P ) of dense open subsets
of P with a non-dense intersection. If this intersection
WT is again non-
empty, we again consider the element a ≥ b = ( α<h(P ) A′α ) ∈ RO(P )
and continue similarly. We can repeat this procedure until we get
the desired system (Bα )α<h(P ) of dense open subsets with an empty
intersection.
Next, we extract from each dense open set Bα a maximal antichain
Cα . We claim that for every p ∈ P there is at least one maximal
antichain Cα and elements a, b ∈ Cα such that p is compatible with both
of them: suppose that for some p ∈ P and for every α < h(P ) there
is only one element cα from Cα that is compatible with p. However, p
is then, in fact, below cα (since if p  cα then there is a p0 ≤ p that
is disjoint with cα but necessarily compatible with another element of
 BASE TREE PROPERTY 5
T T
Cα ). This means that p ∈ α<h(P ) ↓ Cα ⊆ α<h(P ) Bα and that is a
contradiction with the fact that the intersection is empty.
Before constructing the levels of T we modify the antichains into a
system (Dα )α<h(P ) where Dβ refines Dα if α < β. This can be easily
done if we set Dα to be a common refinement of (Cγ )γ≤α and (Dγ )γ<α .
The levels of the tree T will be maximal antichains. What we need
to take care of is to ensure that T is dense and that every element of T
has c immediate successors. We begin by showing that for each element
p ∈ P there is an antichain Dα with c-many elements compatible with
p. There is some Dα0 and elements d0 , d1 ∈ Dα0 compatible with p,
i.e. there are elements p0 ≤ d0 , p1 ≤ d1 below p. Then again there is
some Dα1 and elements d00 , d01 , d10 , d11 ∈ Dα1 , the first two compatible
with p0 , the last two with p1 (note that this is the place where we need
the antichains to be refining; since in general there would be some
Dβ1 with compatible elements with p0 and some Dβ2 with compatible
elements with p1 but in our case we can take α1 to be sup{β1 , β2 }).
We again get pζ ≤ p for each ζ ∈ 2 {0, 1}. We continue until we get
an appropriate pζ ≤ p for each ζ ∈ <ω {0, 1}. For every ξ ∈ ω {0, 1} we
have a descending chain p ≥ pξ↾{0} ≥ . . . pξ↾n ≥ . . . with a lower bound
pξ (due to σ-closedness). Moreover, pξ1 ⊥ pξ2 for ξ1 6= ξ2 . Thus we
see that there is a maximal antichain of size c below p; we denote it
A(p). Each such pξ is compatible with some element dξ of Dα where
α = sup{αn : n ∈ ω}. And again ξ1 6= ξ2 implies dξ1 6= dξ2 .
Let Pα = {p ∈ P S : p is compatible with c-many elements of Dα }.
We see that P = α<h(P ) Pα . Since |Pα | ≤ c for each α there is an
injective mapping fα : Pα → ω 2 such that pfα (p) ≤ p for every p ∈ Pα ,
where pfα (p) is from the construction above. For each α for which Pα is
non-empty we complete the antichain {pfα (p) : p ∈ Pα } into a maximal
antichain Rα .
Now we are ready to start the construction of the base tree. We set
T0 = D0 and for each α + 1 we set Tα+1 to be the common refinement
of Dα+1 , A(p) for all p ∈ Tα and Rα if it exists. For α limit, Tα is a
common refinement of (Tγ )γ<α .
Note that by refining A(p) for all p ∈ Tα we ensure that each element
of the tree has c-many immediate successors and by refining Rα ’s that
T is dense. This finishes the proof. 
Corollary 2.2. For an ordering (P, ≤), the following statements are
equivalent:
(i) P has the BT-property,
(ii) P has a dense subset with the BT-property,
(iii) Every dense subset of P has the BT-property,
6 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

(iv) RO(P ) has the BT-property.

Proof. Note that (i)⇒(ii) and (iii)⇒(iv) follow from the definition. It
suffices to prove (ii)⇒(iii), (iv)⇒(i) is then a consequence.
We need to find a dense subset of a given dense subset that is, more-
over, σ-closed and of size c, atomlessness is clear.
Assuming (ii), we have a base tree T , we are given a dense subset D
and we show that there is a σ-closed dense subset S ⊆ D.
We make S from maximal antichains. For every S t ∈ T0 we find a
maximal antichain At ⊆ D below the element t. t∈T0 At is the first
maximal antichain S0 .
Then for every s ∈ S0 we find a maximal antichain S Ms ⊆ T below s.
Let P1 ⊆ T be a maximal antichain from T refining s∈S0 As and T1 .
Again, for Severy p ∈ P1 we find a maximal antichain Ap ⊆ D from D,
the union p∈P1 Ap is S1 .
Isolated steps are treated similarly. We need not omit Pα to be
refining the tree level Tα . Then we refine it to Sα ⊆ D.
For a limit α we take a refinement Pα of all Pβ ’s for β < α (which is
also a refinement of Sβ ’s) together with Tα . Then we again refine it to
Sα ⊆ D. S
The resulting set S = α<h(P ) Sα is dense and σ-closed. We ensured
density by refining all levels of T . For σ-closedness observe that for
every countable descending chain s0 ≥ s1 ≥ . . . from S, where sn ∈
Sαn , there is an interwined descending chain p0 ≥ p1 ≥ . . . such that
p0 ≥ s0 ≥ p1 ≥ s1 ≥ . . ., where pn ∈ Pαn . This interwined chain has
a lower bound p in Pα , where α = sup{αn : n ∈ ω}, and p has some
successor s ∈ S. 
In other words, having a σ-closed dense set is preserved by forcing
equivalence among separative partial orders of size continuum. On
the other hand, Zapletal in [20] has constructed a model in which the
Continuum Hypothesis holds and there are two forcing equivalent sep-
arative partial orders of size ℵ2 one σ-closed and the other without a
σ-closed dense set. One has to wonder whether such a pair exists in
ZFC.
Question 2.3. Are there, in ZFC, two separative partial orders which
are forcing equivalent such that one is σ-closed and the other does not
have a σ-closed dense set? Can such partial orders be found as dense
subsets of the completion of Fn(ω1 , c+ )?
Finally, using this internal characterization of the partial orders with
the BT-property one can easily deduce the following external charac-
terization.
 BASE TREE PROPERTY 7

Theorem 2.4.
(1) Let (P, ≤) be an ordering with the BT-property. Then RO(P )
is a regular subalgebra of Coll(ω1 , c).
(2) Let B be a complete atomless regular subalgebra of Coll(ω1 , c).
Then B has the BT-property.
Proof. We prove (1): Let D ⊆ P be its dense subset witnessing the
BT-property. Then D × Fn(ω1 , {0, 1}, ω1) with induced product or-
dering clearly has the BT-property, the height is ω1 , thus it is forcing
equivalent with the complete Boolean algebra Coll(ω1 , c). Note that
there is a regular embedding e : D → D × Fn(ω1 , {0, 1}, ω1) defined as
e(d) = (d, 1) where 1 is the biggest element in Fn(ω1 , {0, 1}, ω1), i.e.
the empty set. e is extended onto ē : RO(P ) → Coll(ω1 , c) mapping
RO(P ) on a regular subalgebra of Coll(ω1 , c).

Now we prove (2): We use the result of [17] that every atomless game-
closed complete Boolean algebra of density at most c has a σ-closed
dense subset (of size at most c in fact). We note that the converse is
trivial; see also [17]. Since Fn(ω1 , {0, 1}, ω1) contains a σ-closed dense
subset it is game-closed. Now let B be any complete atomless regular
subalgebra, it is still game-closed and of density at most c. Thus it has
a σ-closed dense subset which is atomless and of size at most c thus it
witnesses that B has the BT-property. 

3. Classical examples
The Boolean algebra P(ω)/f in is a prototype of an ordering with
the BT-property. Recall the definitions of the cardinal invariants p, t
([6]). It was proved recently by M. Malliaris and S. Shelah ([15]) that
p = t. We shall discuss these cardinal invariants on other orderings
too.
The second fundamental example is (Dense(Q), ⊆), where Dense(Q)
is a set of all dense subsets in rationals. The situation here is sim-
ilar with the previous example, it is not separative and the ordering
(Dense(Q), ⊆) itself does not satisfy the BT-property. The separative
modification is (Dense(Q), ⊆nwd ), where A ⊆nwd B if A \ B is nowhere
dense in Q, has the BT-property. This ordering is studied in [2].
Let pQ , tQ , hQ be the cardinal invariants of (Dense(Q), ⊆nwd ) defined
in the same way as their counterparts in ([ω]ω , ⊆∗ ). It was proved in
[2] that pQ = p and tQ = t (thus pQ = tQ = t) whereas hQ and h are
incomparable in ZFC, hQ < h and hQ > h are both consistent (see [2]
and [7]); and hQ = h too, of course.
8 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

For the third example, let A be the Cantor algebra, i.e. the algebra of
all clopen subset of 2ω , and consider the countable product Aω modulo
the ideal Fin ⊆ Aω , where Fin = {f ∈ Aω : |{n : f (n) 6= 0}| < ω}. It
satisfies the BT-property, moreover, Aω /Fin is homogeneous.
t(Aω /Fin) = t and h(Aω /Fin) ≤ min{h, add(M)} ([3]) and it is
consistent that h(Aω /Fin) < h ([3],[9]).
For any Boolean algebra B let us consider an infinite product B ω . Let
J be an ideal on ω. By IJ ⊆ B ω we denote the ideal {f ∈ B ω : {n ∈ ω :
f (n) 6= 0} ∈ J}. The quotient algebra B ω /IJ consists of equivalence
classes where f, g ∈ B ω are equivalent if {n : f (n) 6= g(n)} ∈ J
(f △ g ∈ IJ equivalently). We state and prove a simple criterion for
when such a product has the BT-property.
Theorem 3.1. Let B be a Boolean algebra and J an ideal on ω. Then
the reduced product B ω /IJ has the BT-property if and only if B con-
tains a dense subset of size c and (either P(ω)/J is σ-closed or J is a
maximal ideal), and (either P(ω)/J or B is atomless).
Proof. Since B ω /IJ contains a dense subset of size c if and only if B
contains a dense subset of size less or equal to c the requirement on the
cardinality is satisfied.
Suppose that P(ω)/J is not σ-closed. Let (Xn )n∈ω be a descending
chain of infinite subsets of ω such that the chain ([Xn ])n∈ω does not
have a lower bound in P(ω)/J, where [Xn ] is the equivalence class
containing Xn . We define a descending chain ([fn ])n∈ω ⊆ B ω /IJ as
follows: fn (i) = 1 if i ∈ Xn and fn (i) = 0 otherwise (it is the image of
the chain ([Xn ])n∈ω via the regular embedding of P(ω)/J into B ω /IJ ).
Suppose that it has a lower bound [f ]. Then the support of f , i.e. the
set {i : f (i) 6= 0}, would determine a lower bound for ([Xn ])n∈ω .
Next we use the fact mentioned in [3] that B ω /IJ can be written as
an iteration P(ω)/J ⋆ B ω /U̇, where U̇ is a name for an ultrafilter added
by P(ω)/J. For [f ] ∈ B ω /IJ we define Φ([f ]) = ({i : f (i) 6= 0}, [f˙ ]),
where [f˙ ] is a name for an equivalence class containing f in B ω /U̇. Φ
is easily verified to be a dense embedding which proves the fact.
Now observe that an ultrapower of any Boolean algebra is σ-closed.
For a countable descending chain we can choose representatives of
T support f0 = ω, support f1 ⊇ sup-
equivalence classes (fn )n∈ω so that
port f2 ⊇ support f3 ⊇ . . . and n∈ω support fn = ∅ since the ultrafilter
is non-principal. Then we set f (i) = fn (i) if n is the smallest number
such that i ∈ support fn \ support fn+1 . f clearly determines the lower
bound for the chain. Hence, we conclude that B ω /IJ is σ-closed since
an iteration of two σ-closed forcings is.
 BASE TREE PROPERTY 9

To check atomlessness, if P(ω)/J is atomless then for any f ∈ B ω ,

where the support of f is not in J, we can always split the support of f
into two disjoint infinite sets both outside of J, restrict f to these sets
and make two disjoint elements of B ω /IJ below [f ]. If B is atomless
then we can always find two disjoint successors coordinatewise. Finally,
suppose that P(ω)/J has an atom [A] and B has an atom b. Then
f ∈ B ω defined so that f (n) = b for n ∈ A and f (n) = 0 for n ∈ / A
determines an atom in B ω /IJ . 

4. On classes of ideals ordered by reverse inclusion

We shall deal with orderings that consist of ideals on ω of some type
ordered by reverse inclusion. We assume that all such ideals extend the
ideal of finite subsets of ω. Since every ideal on ω can be considered as
a subset of the Cantor space we can speak about the topological, resp.
measure-theoretical characterizations of such ideals.

4.1. Non-tall ideals. An ideal I on ω is tall if for every X ∈ [ω]ω

there is infinite Y ⊆ X that belongs to I. Consider the set T of all
non-tall ideals on ω ordered by reverse inclusion.
First of all, this ordering is not separative. However, for every
A ∈ [ω]ω consider the ideal IA of all subsets of ω that have a finite
intersection with A. IA is a non-tall ideal and B ⊆∗ A implies IB ⊇ IA .
Moreover, for every non-tall ideal I and some infinite set A almost
disjoint with every element of I, IA ⊇ I. Thus we see that ([ω]ω , ⊆∗ )
is isomorphic with a dense subset of (T, ⊇) and of its seprative mod-
ification showing that the separative modification of (T, ⊇) has the
BT-property, however it is forcing equivalent to ([ω]ω , ⊆∗ ).

4.2. Fσ ideals. Consider the ordering of all Fσ ideals on ω denoted as

F ordered by reverse inclusion. The study of this ordering was initiated
by C. Laflamme in [14] and also studied in [11].
It is immediate from the definition that F is σ-closed. To show that
it is atomless, consider any Fσ ideal I. Since it is not maximal there is
a subset A ⊆ ω such that neither A nor ω \ A belong to I. We extend
I by adding A to obtain an ideal IA ; similarly, we obtain an ideal Iω\A .
TheyS are disjoint, we must prove they are Fσ . We do it for IA . Write I
as n Fn where each Fn is closed, thus compact. S The mapping π that
sends X to X ∪ A is continuous, thus π[I] = n π[Fn ] is still Fσ and
the downward closure of π[I] is still Fσ and equal to IA . Since there
are only c-many Fσ ideals we just proved that F has the BT-property.
However, we do not know what the height of this ordering is.
10 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

4.3. Summable ideals. Consider the ordering (c+ 1 ∗

0 \ ℓ , ≤ ) where c0
+
1
is the set of all sequences of positive reals that tend to zero and ℓ the
set of all sequences of reals whose sum converges. The order relation
≤∗ is almost domination, i.e. f¯ ≤∗ ḡ if {n : gn > fn } is finite. The
investigation of this ordering was initiated by P. Vojtáš in [18]. (c+ 0 \
ℓ1 , ≤∗ ) is not separative but we will show that the separative quotient
is isomorphic to the set IΣ of all summable ideals ordered by inverse
inclusion.
¯ + 1
PAn ideal I is summable if there exists f ∈ c0 \ ℓ such that I = {A :
n∈A f (n) < ∞}. Note that any summable ideal is an Fσ P -ideal,
thus IΣ is a subordering of F.
We check that IΣ has the BT-property. Let us verify atomlessness.
Let I be a summableP ideal determined by a sequence (an )∞ n=0 , and let
A ∈ I. Then a
i∈ω\A i diverges; we divide ω \ A into two infinite
subsets B1 and B2 such that the appropriate sums both diverge. We
make new sequences (bn )∞ ∞
n=0 and (cn )n=0 so that bi = ai for i ∈ A ∪
B1 and bi = zi for i ∈ B2 , where (zn )∞ n=0 is an arbitrary converging
sequence of positive reals. (cn )∞ n=0 is defined similarly, just B1 and B2
∞ ∞
change their roles. Both (bn )n=0 and (cn )n=0 diverge. We denote the
appropriate summable ideals Ib and Ic . It is clear that Ib , Ic ⊇ I and
that they are disjoint.
Let (Ij )j∈ω be an increasing (in inclusion) sequence of summable
ideals. Let (ajn )∞n=0 be the sequence of positive reals that determines the
ideal PIj . We may assume that (a0n )∞ 1 ∞
n=0 ≥ (an )n=0 ≥ . . .. Let n0 be such
that j≤n0 a0j > 1. We set an = a0n for n ≤ n0 . Then we find a n1 > n0
P 1
such that nj=n a1 > 1 and set an = a1n for n0 < n ≤ n1 . And so on
0 +1 j
to obtain the whole sequence (an )n∈ω so that (an )n∈ω ≤∗ (ajn )n∈ω for all
j ∈ ω.
To verify separativness, consider ideals Ia and Ib corresponding to
sequences (an )∞ ∞
n=0 and (bn )n=0 , such that Ia + Ib , P i.e. there is a set
P ∈ Ib which does not belong to Ia . That means k∈B bk < ∞ but
B
k∈B ak = ∞. If ω \B belongs to Ia then Ia and Ib are already disjoint,
∞
if this is not that case thenP we make a new sequence (cn )n=0 such that
cn = an for n ∈ B and k∈ω\B ck < ∞. The corresponding ideal Ic is
below Ia and disjoint with Ib .
It is easy to check that if (an )∞ ∞ ∞
n=0 ≈sep (bn )n=0 , i.e. ∀(cn )n=0 ∈
(c+ 1 ∗ ∞ ∞ ∞ ∞
0 \ ℓ , ≤ )((cn )n=0 ⊥ (an )n=0 ⇔ (cn )n=0 ⊥ (bn )n=0 ), then (an )n=0
∞
∞
and (bn )n=0 determine the same summable ideal and the mapping P Φ:
+ 1 ∗ ∞
(c0 \ ℓ , ≤ ) → (IΣ , ⊇), defined as Φ((cn )n=0 ) = {A ⊆ ω : n∈A cn <
∞}, is an onto homomorphism of orderings preserving the disjointness
 BASE TREE PROPERTY 11

relation. And the preimage of each summable ideal is precisely an

equivalence class of sequences in ≈sep .
Proposition 4.1. t((c+ 1 ∗
0 \ ℓ , ≤ )) = t.
Proof. To simplify the notation we will write ā instead of (an )∞ n=0 . Let
(āα )α<κ be a descending chain of sequences from (c+ 0 \ ℓ 1
, ≤∗
) of length
κ < t. We use the methods from [5] to show it has a lower bound.
For each α < κ let hα : ω → ω be a function such that ∀n ∈
ω( hα1(n) ≤ āα,n ). Since κ < t ≤ b, there is a function h ∈ ω ω that
almost dominates all hα ’s, i.e. h ≥∗ hα for all α < κ.
P for each α < κ let fα : ω → ω be a function such that
Similarly,
∀n ∈ ω( fα (n)≤i<fα (n+1) āα (i) > 1). Since κ < t ≤ b, there is a function
f ∈ ω ω that almost dominates all fα ’s, i.e. f ≥∗ fα for all α < κ. Define
g ∈ ω ω recursively so that g(0) = f (0) and g(n+1) P= f (g(n)+1). Note
that for every α < κ and all but finitely many n’s g(n)≤i<g(n+1) āα (i) >
1 since g(n) < fα (g(n)) < fα (g(n) + 1) ≤ g(n + 1). We denote In the
interval [g(n), g(n + 1)).
For every n, we denote Fn the following P set of functions {F : dom(F ) ⊆
In ∧ rng(F ) ⊆ { 2|I1n | , 2|I2n | , . . . , 1} ∧ i∈dom(F ) F (i) > 12 }. Let F =
S
n∈ω Fn . In the following we shall treat F as ω.
For every āα , let
Xα = {F : ∃n(F ∈ Fn ∧ ∀i ∈ dom(F )(F (i) ≤ āα (i))}
An easy pigeon-hole type argument shows that it is infinite for every
α < κ. It is also clear that Xβ \ Xα is finite for α < β. Since κ < t,
there is a lower bound X ⊆ F. By shrinking it if neccessary, we can
assume that |X ∩ Fn | ≤ 1 for every n. Finally, we define a sequence ā
as follows:
For every m ∈ ω, if there exists F ∈ X such that m ∈ dom(F ) then
1
we set ā(m) = F (m). Otherwise, we set ā(m) = h(m) . It is now easy
to check that ā is the desired lower bound.
To prove the converse, let us at first prove that t((c+ 1 ∗
0 \ ℓ , ≤ )) ≤ b.
Suppose the contrary. Let (bα )α<b be a system of almost increasing
functions from ω ω without an upper bound, π : N × N → N a bijection
1
and (ln )∞ 1
n=0 a strictly decreasing sequence from ℓ such that ln < n for
every n. We define a descending chain of sequences from (c+ 1 ∗
0 \ ℓ ,≤ )
(āα )α<b as follows: āα (π(1, k)) = lk for k ≤ bα (0), for l > bα (0) we
set āα (π(1, l)) = 1l ; generally, āα (π(n, k)) = lk for k ≤ bα (n − 1), for
l > bα (n − 1) we set āα (π(n, l)) = 1l .
Let ā be a lower bound for this chain. Define a function f by f (n) =
min{k : ā(π(n, k)) > lk }. It is easy to check that f almost dominates
(bα )α<b , a contradiction.
12 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

Now assume that t < t((c+ 1 ∗ ω

0 \ ℓ , ≤ )). Let (Xα )α<t ⊆ [ω] be a
ω
descending chain without a lower bound. We define fα ∈ ω for every
α < t so that fα (n) = k such that |Xα ∩ [fα (n − 1), fα (n))| ≥ n + 1.
Since t < t((c+ 1 ∗ ω
0 \ ℓ , ≤ )) ≤ b, we can again find g ∈ ω such that for
every α < t and for almost all n’s |Xα ∩ [g(n − 1), g(n))| ≥ n + 1.
Define a chain (āα )α<t of sequences as follows: āα,n = k1 if n ∈
Xα ∩ [g(k − 1), g(k)); if no such k exists then let āα,n = ln .
Finally, let ā be a lower bound for this descending chain and define
a lower bound X = {n : ān > ln } for the chain (Xα )α<t . 
4.4. Meager and null ideals. Next we consider the class of all mea-
ger ideals M and the set of all ideals N of measure zero; i.e. those ideals
that are meager sets and null sets respectively in the Cantor space
topology. Simultaneously, we study the set of all hereditary meager
MH and hereditary null NH ideals, where an ideal I is hereditary mea-
ger (null) if for every X ∈ I + the restriction I ↾ X = {A ∈ I : A ⊆ X}
is meager (null) in the Cantor space 2X .
It is obvious they are both σ-closed. We show they are atomless,
what the separative quotient for meager ideals is and that there is no
dense subset in both of these orderings that has cardinality c. In fact,
there are 2c mutually disjoint elements in both orderings. Let us note
that by ≈sep we denote the “separative equivalence” (in the ordering
of meager ideals); i.e. I ≈sep J if and only if for any meager ideal K,
K is disjoint with I iff K is disjoint with J.
We will use the following characterizations of meager and null ideals.
Proposition 4.2 (Talagrand; see for example Theorem 4.1.2 [4]). An
ideal I on ω is meager ifSand only if there is a partition (Pi )i∈ω of ω
into finite sets such that i∈A Pi ∈ I iff A is finite.
Proposition 4.3 (Bartoszyński; Theorem 4.1.3 [4]). An ideal I on ω
is null if and only if there exists an infinite system {An : n ∈ ω} such
that
(1) each An is a S finite setSconsisting of finite subsets of ω
(2) ∀m
P 6
= n ∈ ω( An ∩ Am = ∅)
(3) n∈ω µ{X ⊆ ω : ∃a ∈ An (a ⊆ X)} < ∞
(4) for every X ∈ I ∃∞ n∃a ∈ An (X ∩ a = ∅)
where µ is the Lebesgue measure on the Cantor space.
Proposition 4.4. There is a mapping Φ : (M, ⊇) → (MH , ⊇) such
that ∀I ∈ M(Φ(I) ⊇ I ∧ Φ(I) ≈sep I).
Note that it follows that for any meager ideal I, Φ(I) is the least
element in the equivalence class of ≈sep containing I.
 BASE TREE PROPERTY 13

Proof. For a meager ideal I consider the set

I˜ = {A ⊆ ω : I ↾ A is not meager}
Let (Pn )n∈ω be the partition of ω witnessing I is meager (from 4.2).
I˜ is a hereditary meager ideal containing I. To see that it is meager
check that (Pn )n∈ω still works. Let A ∈ I˜+ be arbitrary. Since A ∈ / I˜
we have I ↾ A is meager, so there is a partition (Qn )n∈ω of A into finite
S
sets such that i∈C Qi ∈ I iff C is finite. If I˜ ↾ A were not meager then
S
there would be an infinite set C ⊆ ω such that B = i∈C Qi ∈ I˜ ↾ A.
I ↾ B would have to be S nonmeager but then there would be an infinite
set D ⊆ C such that i∈D Qi ∈ I ↾ A, a contradiction.
It remains to prove that Φ(X) ≈sep X for any X ∈ M. But obviously
if a meager ideal I is compatible with a meager ideal J, then Φ(I) is
compatible with Φ(J) and so also with J. Each equivalence class of
meager ideals has its minimal element, the corresponding hereditary
meager ideal. 
Question 4.5. We do not know whether the previous proposition also
holds true for the class of null ideals. Let I be a null ideal, is I˜ = {A ⊆
ω : I ↾ A is not null} a (hereditary) null ideal?
It is easy to prove that I˜ = {A ⊆ ω : I ↾ A is not null} is an ideal
though. It is immediate that I˜ extends I. Let us check that it is
downward closed. Let A ∈ I˜ and B ⊆ A. Suppose that B ∈ ˜ Then
/ I.
I ↾ B is null. We use Proposition 4.3 to obtain an infinite system
{An : n ∈ ω} witnessing it. It follows from that proposition that
the same system would witness that I ↾ A is null as well, which is a
contradiction.
Let A, B ∈ I,˜ we may assume they are disjoint. Then realize that
X → (X ∩ A, X ∩ B) is a measure preserving homeomorphism from
A∪B to A×B and it follows from the Fubini theorem that I ↾ A×I ↾ B,
so also I ↾ A∪B, is not null. Thus I˜ is closed under taking finite unions.
Corollary 4.6. (M, ⊇) and (N, ⊇) are atomless, not separative, the
separative quotient of (M, ⊇) is isomorphic to the ordering (MH , ⊇) of
all hereditary meager ideals via the mapping Φ.
Proof. To prove they are atomless, let I be an arbitrary meager ideal,
let A and B be two infinite subsets of ω such that A∪B = ω and neither
A nor B is in Φ(I) (Φ(I) is meager, thus not maximal). Extend I by
A and by B to obtain two disjoint ideals XA and XB that are easily
verified to be meager. The proof for null ideals is similar.
We claim they are not separative. Consider some maximal ideal
M on odd natural numbers and the ideal F of finite sets on even
14 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

numbers. Then let I = {A ∪ B : A ∈ M ∧ B ∈ F } and J =

{A ∪ B : A is an arbitrary subset of odd natural numbers ∧ B ∈ F }
be two ideals, both easily verified to be meager and null. However,
I and J are equivalent in the separative modification for both classes
(meager and null) of ideals.
On the other hand, if I and J are two hereditary meager ideals such
that I + J, then there is an infinite set A ∈ J that is not in I. Let K
be an ideal generated by I ∪ {ω \ A}, it is clearly meager and Φ(K) ⊇ I
is disjoint with J.
It remains to show that Φ defines an isomorphism between the sep-
arative quotient of (M, ⊇) and an ordering (MH , ⊇). But Φ obviously
preserves the inclusion relation and the disjointness between ideals and
from the previous proposition, Φ(I) ≈sep I for any I ∈ M, so we are
done. 
Finally, we show there is no dense subset of these orderings with size
c, thus preventing them to have the BT-property.
Theorem 4.7. There are 2c ideals that are both meager and null and
that are mutually disjoint (in both (M, ⊇) and (N, ⊇)).
In particular, neither (M, ⊇) nor (N, ⊇) has the BT-property.
Proof. Let (In )n∈ω be a partition of ω into intervals such that |In | =
n + 1. For any X ⊆ ω let X I be the set
[
{k ∈ ω : ∃n(k is the m-th element of In }
m∈X

It is clear that if J is an ideal on ω then {X I : X ∈ J} is a base of an

ideal; we shall denote this ideal as IJ .
Now let M be the set of all maximal ideals on ω, its size is 2c . For
any J ∈ M, we make an ideal IJ and obtain a system I of 2c ideals.
The disjointness of two ideals J1 , J2 ∈ M is easily seen to be preserved
for IJ1 and IJ2 .

Claim 1 I is a system of meager ideals.

The interval partition (In )n∈ω works for all ideals from I. Assume
that some set set X ∈ IJ , where IJ ∈ I, contains a union of infinitely
many intervals. It is easy to check that once it contains the whole
interval In then it contains all previous intervals. Thus, we conclude
that X = ω which is a contradiction.

Claim 2 I is a system of null ideals.

 BASE TREE PROPERTY 15

We use characterization of null ideals from 4.3. For m ≤ n, let im n

be the m-th element of In . Let An = {an = {inm : n ≤ m ≤ 2n − 1}}.
These sets satisfy the first three conditions from 4.3. Let X ∈ IJ ,
where IJ ∈ I, be a given set. X ⊆ Y I for some Y ∈ J. Then it is easy
to check that An ∩ X = ∅ for n ∈ ω \ Y and |ω \ Y | = ω; thus we also
verified the last fourth condition and proved that every IJ is null. 

5. Products
If P and Q are two orderings with the BT-property then their prod-
uct P × Q has again the BT-property and the height is less or equal
to the minimum of heights of the original orderings. To see this, just
realize that if B is a regular subalgebra of C then h(B) ≥ h(C), and
that RO(P ) is a regular subalgebra of RO(P × Q). The same holds for
countable products and iterations. Let us mention the case of prod-
ucts of ([ω]ω , ⊆∗ ). By hα , for 2 ≤ α ≤ ω we denote h(([ω]ω , ⊆∗ )α ). It
immediately follows that t(([ω]ω , ⊆∗ )α ) = t for any 2 ≤ α ≤ ω. Shelah
and Spinas in [16] proved that consistently for any n ∈ ω hn+1 < hn .
However, the following question remains open.
Question 5.1. Does it hold in ZFC that hω = min{hn : n ∈ ω}?

References
[1] B. Balcar, J. Pelant, P. Simon, The space of ultrafilters on N covered by nowhere
dense sets, Fund. Math. 110 (1980), 11-24
[2] B. Balcar, F. Hernandéz-Hernandéz, M. Hrušák, Combinatorics of dense sub-
sets of rationals, Fund. Math. 183(2004), 59-79
[3] B. Balcar, M. Hrušák, Distributivity of the algebra of regular open subsets of
βR \ R∗ , Topol. Appl. 149 (2005), 1-7
[4] T. Bartoszyński, H. Judah, Set Theory: On the Structure of the Real Line, A.
K. Peters, Wellesley, 1995
[5] T. Bartoszyński, M. Scheepers, Remarks on small sets related to trigonometric
series, Topol. Appl. 64 (1995), 133-140
[6] A. Blass, Combinatorial Cardinal Characteristics of the Continuum, In: Hand-
book of Set Theory, Springer, Dordrecht, 2010
[7] J. Brendle, Van Douwen’s diagram for dense sets of rationals, Ann. Pure Appl.
Logic 143, Issues 13(2006), 54-69
[8] P. Dordal, A model in which the base matrix tree has no cofinal branches, J.
Symb. Logic 52 (1987), Issue 3, 651-664
[9] A. Dow, The regular open algebra of βR \ R is not equal to the completion of
P(ω)/fin, Fund. Math. 157 (1998), 33-41
[10] A. Dow, Tree π-Bases for βN − N in Various Models, Topol. Appl. 33 (1989),
3-19
[11] W. Just, A. Krawczyk, On certain Boolean algebras P(ω)/I, Trans. Amer.
Math. Soc. 285 (1984), Number 1, 411429
 16 BOHUSLAV BALCAR, MICHAL DOUCHA, AND MICHAEL HRUŠÁK

[12] B. König, Dense subtrees in complete Boolean algebras, Math. Logic Quart. 52
(2006), Issue 3, 283-287
[13] K. Kunen, Set Theory: An Introduction to Independence Proofs, North-
Holland, 1980
[14] C. Laflamme, Forcing with filters and complete combinatorics, Ann. Pure Appl.
Logic 42 (1989), Issue 2, 125-163
[15] M. Malliaris, S. Shelah, Cofinality spectrum theorems in model theory, set the-
ory and general topology, preprint, http://arxiv.org/abs/1208.5424
[16] S. Shelah, O. Spinas, The distributivity numbers of finite products of
P (omega)/f in, Fund. Math. 158 (1998), 81-93
[17] B. Veličković, Playful Boolean algebras, Trans. Amer. Math. Soc. 296 (1986),
Number 2, 727-740
[18] P. Vojtáš, Boolean isomorphism between partial orderings of convergent and
divergent series and infinite subsets of N, Proc. Amer. Math. Soc. 117 (1993),
235-242
[19] S.W. Williams, Trees, Gleason spaces, and Coabsolutes of βN−N, Trans. Amer.
Math. Soc. 271 (1982), Number 1, 83-100
[20] J. Zapletal, On the existence of a σ-closed dense subset, Commentat. Math.
Univ. Carol. 51 (2010), issue 3, 513-517

The Center for Theoretical Study, Charles University in Prague,

Jilská 1, Prague, Czech republic
E-mail address: [email protected]

Institute of mathematics, Academy of Sciences of the Czech repub-

lic, Žitná 25, Prague, Czech republic
E-mail address: [email protected]

Instituto de Matemáticas, UNAM, Apartado Postal 61-3, Xangari,

58089, Morelia, Michoacán, México.
E-mail address: [email protected]

View publication stats