A Coloring Theorem for Inaccessible Cardinals
A dissertation presented to the faculty of the College of Arts and Sciences of Ohio University
In partial fulfillment of the requirements for the degree Doctor of Philosophy
Douglas J. Hoffman December 2013 c 2013. Douglas J. Hoffman. All Rights Reserved. 2
This dissertation titled A Coloring Theorem for Inaccessible Cardinals
by DOUGLAS J. HOFFMAN
has been approved for the Department of Mathematics and the College of Arts and Sciences
Todd Eisworth Associate Professor of Mathematics
Robert Frank Dean, College of Arts and Sciences 3 Abstract
HOFFMAN, DOUGLAS J., Ph.D., December 2013, Mathematics A Coloring Theorem for Inaccessible Cardinals (73 pp.) Director of Dissertation: Todd Eisworth This dissertation proves a coloring theorem for inaccessible cardinals having a stationary subset S ⊂ {δ < κ : cf(δ) = λ} that does not reflect in any inaccessible cardinal for a fixed regular uncountable cardinal λ. The theorem shows we can give
2 an “approximate” yes to κ 6→ [κ]κ. Recent work of Eisworth has produced partition theorems in the case of successors of singular cardinals, and this dissertation extends his ideas to inaccessible cardinals with a stationary subset that does not reflect in any inaccessible cardinal and consists of singular cardinals of uncountable cofinality. Club- guessing sequences, minimal walks and elementary submodels are used in the proof of the theorem. First, the existence of a nice S-club guessing sequence is established
κ when S ⊆ Sλ . After proving the existence of this club-guessing sequence, we look at constructing a club-guessing sequence when S consists of inaccessible cardinals and
C is an S-club sequence satisfying ⊗C . Then, we study weakly normal ideals and the effects of a least function on an ideal. The behavior of a least function can have a strong affect on the ideal. When a least function fails to be regressive in a rather strong fashion, this failure forces the ideal to be weakly normal. After investigating weakly normal ideals, the ideal idp(C, I) is introduced, and a coloring is constructed to prove the main theorem in this paper. This coloring has the property that the range of any unbounded subset of κ is almost κ. Using this coloring, we show the
2 ideal idp(C, I) is weakly κ-saturated if κ → [κ]κ holds. 4
To my wife, Beth, and to my parents, Michael and Lynnae 5 Acknowledgments
I would like to thank my wife, Beth, for her love and support through the years. She has been with me for this stage of my life. From my first day of graduate school to the writing of this paper, she has been with me every step of the way and supporting me any way she can. It takes a lot of patience to hear someone talk about something only a few people would find interesting. I would like to thank my parents, Michael and Lynnae Hoffman. They let me do my own thing and taught me to always give it your best effort. Without their support, none of this would be possible. Also, I would like to express my gratitude to my advisor, Dr. Todd Eisworth. His help and advice was not limited to the writing of this dissertation. He got me interested in combinatorial set theory and introduced me to partition theorems. With his help, I was able to explore a rich fragment of combinatorial set theory and con- tribute to the theory. Under his guidance, I was able to go from a graduate student in mathematics to a mathematician. I would also like to thank the members of my dissertation committee for investing time in this manuscript. Chris Baily-Brown, Tammy Matson and Kelly Pero have my gratitude. They have helped me numerous times over the years with forms and deadlines I would have forgotten about. Lastly, I would like to thank Karen Ernst at Hawkeye Community College. She showed me that it was okay to be enthusiastic about mathematics, and doing so, she set me on the path to become a mathematician. 6
Table of Contents
Abstract...... 3
Dedication...... 4
Acknowledgments...... 5
1 Introduction...... 7 1.1 What Has Been Done So Far...... 7 1.2 Goals For The Paper...... 15
2 Existence of Nice Club-Guessing Sequences...... 17 2.1 Club-Guessing Sequences...... 17 2.2 Nice Club-Guessing Sequences: Uncountable Cofinality...... 23
2.3 Constructing A Club-Guessing Sequence Using ⊗C ...... 31
3 Weakly Normal Ideals...... 37 3.1 An Introduction to Ideals and Filters...... 37 3.2 Introduction to Weak Normality...... 39 3.3 Weak Normality Arises...... 41
4 Partition Results...... 55 4.1 Motivation For The Main Theorem...... 55 4.2 Minimal Walks...... 55
4.3 The Ideal idp(C, I)...... 59 4.4 The Main Theorem...... 63
Bibliography...... 72 7
1 Introduction
1.1 What Has Been Done So Far
The ultimate goal of this dissertation is prove a coloring theorem for certain infinite cardinals. In particular, our focus will be on the negation of the square- bracket relation in the realm of inaccessible cardinals. There has been a lot of work done with the negation of the square-bracket relation. Todorˇcevi´c’swork in [15] describes what happens in the case of the successor of a regular cardinal. Eisworth’s work [5, 6, 4] has produced results in the case of successors of singular cardinals. Chapters III and IV of [13] are dedicated to square-bracket partition relations on inaccessible cardinals. Our goals for this paper focus on extending the techniques used on successors of singular cardinals to inaccessible cardinals. Our work builds on a long line of research. In [15], Todorˇcevi´ccreated his method of minimal walks and established a relation between stationary reflection and the failure of a square-bracket relation. Shelah created a plethora of tools utilized in the case of the successor of a singular cardinal. Such tools include club-guessing sequences, club-guessing ideals and scales. In a series of papers, Eisworth has obtained many results dealing with the successors of a singular cardinals. Eisworth’s result in [4] is Todorˇcevi´c-like; it relates the failure of a square-bracket relation to stationary reflection. To understand the failure of the square-bracket relation, we start with a brief look at ordinary partition relations. Then, we will see that the failure of an ordinary partition relation leads us to a weak form of homogeneity: the square-bracket relation. From there, we look at the results of Todorˇcevi´cand Eisworth, and end with the goals of this paper. 8
To begin, we look at the following famous theorem in combinatorial set theory: Ramsey’s Theorem.
THEOREM 1.1.1 (Ramsey’s Theorem, [11]) Suppose c :[ω]2 → {0, 1}. Then
2 there is an H ⊆ ω such that f [H] is constant.
The function c is referred to as a coloring, and the subset H is said to be homo- geneous for c. A natural question to ask is to what extent does Ramsey’s theorem generalize to uncountable cardinals? Before we consider generalizations of Ramsey’s theorem, we will remind the reader of arrow notation introduced by Erd˝osand Hajnal in [8].
DEFINITION 1.1.1 Let γ be an ordinal, and let κ, λ, τ be cardinals. The symbol
λ κ → (τ)γ is used to indicate that the following assertion holds: For an arbitrary set X of cardinality κ and for every γ-partition f :[X]λ → γ, there is a set Y ⊆ X and ordinal ν < γ such that |Y | = τ and
f”[Y ]λ = {ν}.
Relations of this form are called ordinary partition relations. Ramsey’s theo- rem can be restated using this notation.
THEOREM 1.1.2 (Ramsey’s Theorem)
2 ω → (ω)2
When an ordinary partition relation holds, we know that no matter how [X]λ is partitioned into γ pieces, there will always be a set Y ⊆ X of size τ and ν < γ such that set [Y ]λ is contained in the νth piece of the partition. When dealing with 9 partition relations, we sometimes think of a partition of [κ]λ as a coloring of subsets of κ of size λ. Back to generalizing Ramsey’s theorem. One can ask if there are other cardinals for which Ramsey’s theorem holds? Equivalently, is there an uncountable cardinal κ for which
2 κ → (κ)2
ℵ0 2 holds? Sierpi´nksiproved it does not hold in general by showing 2 6→ (ℵ1)2. His proof was generalized to show that if Ramsey’s theorem holds for an un- countable κ, then κ has to be a strong limit cardinal. Another easy argument shows
2 that Ramsey’s theorem fails for singular cardinals, so if κ → (κ)2, then κ has to be
2 strongly inaccessible. Much more is known: uncountable cardinals for which κ → (κ)2 are called weakly compact, and they have been studied extensively. Now, we consider the situation when an ordinary partition relation fails. When
λ λ κ → (µ)2 fails, there is a way to partition [κ] into two sets such that for any Y ⊆ κ with |Y | = µ,[Y ]λ intersects both parts of the partition. Said another way, there is a function c :[κ]2 → {0, 1} such that for any Y ⊆ κ with |Y | = µ, we have
λ rng(f [A] ) = {0, 1}.
When an ordinary partition relation fails, it says the partition is complicated in the sense it takes on both colors when restricted to each unbounded subset. If we increase the number of colors, we get a more drastic failure of Ramsey’s theorem. To study the failure of an ordinary partition relation, we turn to another partition relation called the square-bracket relation. Again, the notation is due to Erd˝os and Hajnal [8].
DEFINITION 1.1.2 Given cardinals κ, λ and µ and ordinal γ, the symbol
λ κ → [µ]γ 10
indicates the following assertion holds: For an arbitrary set X of cardinality κ and for any arbitrary partition f :[X]λ → γ, there is a set Y with |Y | = µ for which f”[Y ]λ 6= γ.
Compared to ordinary partition relations, the square-bracket relation is a weak form of homogeneity. When studying square-bracket relations, we are more interested in when it fails. If we have
λ κ 6→ [µ]γ , then there is a set X of cardinality κ and a partition of f :[X]λ → γ such that for any set Y ⊆ X with |Y | = µ, we have
f”[Y ]λ = γ.
These colorings are so complicated that they take on all possible values whenever the coloring is restricted to a set of size µ. The negation of a square-bracket relation at a cardinal κ yields a strong failure of Ramsey’s theorem by giving us a coloring that remains complicated when restricted to any unbounded subset of κ.
ℵ0 2 Sierpi´nksi’sproof of 2 6→ (ℵ0)2 actually shows
ℵ0 2 2 6→ [ℵ1]2.
In his proof, Sierpi´nksiused an arbitrary well-order of R and the regular ordering
of R to define the coloring. The coloring was defined as follows: for x0 < x1 ∈ R,
f({x0, x1}) = 0 if the well-order and regular order on R agreed, and 1 otherwise. If
2 2 H ⊂ R with |H| = ℵ1 and f [H] = {0} or f [H] = {1}, then H with the regular ordering or reverse ordering would be well-ordered. It can be shown that any such set is at most countable, and we have a contradiction.
ℵ0 2 Sierpi´nksi’sproof of 2 6→ [ℵ1]2 can be generalized as follows.
THEOREM 1.1.3 (Sierpi´nksi,[14]) For each infinite κ ≥ ω,
κ + 2 2 6→ [κ ]2. 11
A later result was obtained by Erd˝osin [8].
THEOREM 1.1.4 If κ ≥ ω and 2κ = κ+, then
+ + 2 κ 6→ [κ ]κ+ .
This theorem differs from the previous two by increasing the number of colors in
+ ℵ0 the partition from 2 to κ . Under the Continuum Hypothesis ( 2 = ℵ1), the above theorem states that
2 ℵ1 6→ [ℵ1]ℵ1 .
The question whether ℵ 6→ [ℵ ]2 holds without the use of the Continuum Hypothesis 1 1 ℵ1 was a major open question in set theory until it was settled by Stevo Todorˇcevi´cin [15].
THEOREM 1.1.5 (Todorˇcevi´c,[15]) Suppose ω < λ = cf(λ) has a non-reflecting stationary subset, then
2 λ 6→ [λ]λ.
A stationary subset S ⊆ κ is non-reflecting if S ∩ α is not stationary in α for all limit α < κ. It can be shown that if κ is a regular cardinal, then the set
κ+ + Sκ := {δ < κ : cf(δ) = κ}
is a non-reflecting stationary subset of κ+. Along with this fact, Todorˇcevi´c’stheorem shows for every regular cardinal κ ≥ ω that
+ + 2 κ 6→ [κ ]κ+ .
In particular, by fixing κ = ℵ0, Todorˇcevi´c’stheorem yields the following corollary.
COROLLARY 1.1.1 (Todorˇcevi´c,[15])
2 ℵ1 6→ [ℵ1]ℵ1 . 12
In Todorˇcevi´c’sproof, he used a non-reflecting stationary subset, S ⊆ λ, to create a sequence C = hCδ : δ < λi such that
i. Cδ is a club subset of δ with 0 ∈ Cδ for each δ < λ, and
ii. for each limit δ < λ, Cδ ∩ S = ∅.
Todorˇcevi´ccalls a sequence of this form a C-sequence. Todorˇcevi´cuses C- sequences to produce, for each pair of ordinals α < β < λ, a decreasing (hence finite) sequence of ordinals beginning with β and ending with α. These sequences are produced using his technique of minimal walks, and he uses them to define a complicated partition of λ which then yields the negative square-bracket partition relation.
κ+ Todorˇcevi´c’stheorem applies to many cardinals but not all of them. Since Sκ is a non-reflecting stationary subset for a regular κ, the question whether a square-bracket relation holds for the successor of a regular cardinal is settled. But what about the successor of a singular cardinal? When working with the successor of a singular cardinal, we run into trouble be- cause it is consistent (assuming large cardinals) that every stationary subset of such cardinal reflects ([4]). In the context of successors of singular cardinals, new tech- niques and tools needed to be developed. Collaborating with Shelah, Eisworth pro- duced some results in this situation. In [5], Eisworth showed when µ is a singular cardinal,
+ + 2 µ 6→ [µ ]µ+ if and only if
+ + 2 µ 6→ [µ ]θ for arbitrary large θ < µ.
And, in [4], Eisworth was able to give concrete conditions for the negation of the square-bracket relation. 13
+ + 2 THEOREM 1.1.6 (Eisworth, [4]) If µ is a singular cardinal and µ → [µ ]µ+ , then every collection of fewer than cf(µ) stationary subsets of µ+ reflects simultane- ously.
In order to get these results in the case of the successor of a singular cardinal, Eisworth used a multitude of mathematical tools. From Shelah, Eisworth used scales and club-guessing sequences, and from Todorˇcevi´c,he used Todorˇcevi´c’smethod of minimal walks. Using these tools, Eisworth constructed a coloring with the property that the range of any unbounded subset of µ+ is a large set relative to a specific ideal. Then, he shows the existence of this ideal is impossible if a square-bracket relation holds. An important tool used in Eisworth’s proof of Theorem 1.1.6 was the S-club guessing sequence. An S-club guessing sequence is a sequence of subsets that can “guess” when a club is club “a lot.” In the definition of an S-club guessing sequence, the terms “guesses” and “a lot” are defined in a rigorous manner.
DEFINITION 1.1.3 Suppose κ is an uncountable cardinal and S ⊆ κ is a station- ary subset. We say hCδ : δ ∈ Si is an S-club guessing sequence if
i. For each δ ∈ S, Cδ ⊆ δ is club in δ.
ii. For each club E ⊆ κ, there are stationarily many δ ∈ S such that Cδ “guesses” E ∩ δ.
There are a few ways that an S-club guessing sequence can “guess” a club. When we want to “guess” a club, we typically require one of the following:
1. Cδ ⊆ E,
2. Cδ \ E is bounded in δ,
3. δ = sup(E ∩ nacc(Cδ)), or 14
4. δ = sup(Cδ ∩ E).
An ordinal α is a non-accumulation point for a set A if α ∈ A and sup(α∩A) < α. For a set A, we denote the set of non-accumulation points of A by nacc(A). Depending on the situation, we may require more from our S-club sequence. For
κ instance, if S ⊆ Sλ for some regular λ < κ, we may require that otp(Cδ) = λ. By
requiring this condition, we can ensure that |Cδ| = λ for each δ ∈ S. To see why S-club guessing sequences are of some interest, we consider the well- known principle ♦(S).
DEFINITION 1.1.4 Let κ be a cardinal and S ⊆ κ be a stationary subset of κ.
♦(S) asserts the existence of a sequence hCδ : δ ∈ Si with Cδ ⊂ δ such that for every X ⊆ κ, there are stationarily many δ ∈ S such that
X ∩ δ = Cδ.
Notice that ♦(S) easily implies the existence of an S-club guessing sequence, so one can think of the club-guessing principle as a weak form of diamond. The surprising fact is that very often these club-guessing sequences are provable in ZFC. In [13], Shelah gave various conditions for when certain club-guessing sequences
exist. For example, when S ⊆ Sℵ2 , there is an S-club guessing sequence hC : δ ∈ Si ℵ0 δ
that guesses clubs of ℵ2 in way that for each club E ⊆ ℵ2, there are stationarily many
δ ∈ S such that Cδ ⊆ E. Eisworth and Shelah use certain club-guessing sequences to construct various complicated colorings at successors of singular cardinals. We will prove the existence of analogous club-guessing sequences for weakly inaccessible cardinals, and then use them to construct complicated colorings in the case where the inaccessible cardinal has a stationary subset which does not reflect in an inaccessible cardinal. 15
1.2 Goals For The Paper
Our first goal is to provide a correct proof of the following club-guessing sequence result claimed by Shelah.
THEOREM 1.2.1 (Shelah, [13]) Suppose κ is (weakly) inaccessible, λ < κ is an
κ uncountable regular cardinal, and S is a stationary subset of Sλ . Then there is an
S-club sequence hCδ : δ ∈ Si such that
(i) Cδ is club in δ with order-type λ,
(ii) hcf(α): α ∈ nacc(Cδ)i is strictly increasing with limit δ, and
(iii) for every club E ⊆ κ, there are stationarily many δ ∈ S with Cδ ⊂ E.
After proving Claim 1.2.1, the goal of the paper is to prove our main theorem.
Main Theorem. Suppose
i). κ is an inaccessible cardinal,
ii). ω < λ = cf(λ) < κ, and
κ iii). S ⊆ Sλ is a stationary subset that does not reflect in any inaccessible.
Then there is c :[κ]2 → κ and an ideal I associated with the sequence from Claim
2 ∗ 1.2.1 such that for every unbounded subset A ⊆ κ, we have rng(c [A] ) ∈ I .
The other smaller goals of the paper will serve to prove the main theorem. We will break these goals down by chapter. In Chapter 2, we establish the existence of a nice S-club guessing sequence as in Claim 1.2.1. Also in Chapter 2, we investigate the existence of S-club guessing sequences for Mahlo cardinals κ where S consists of a stationary set of inaccessible cardinals. 16
In Chapter 3, we study weak normality in ideals. Weak normality is a property that generalizes normality, and in some way, is stronger than normality. We determine when weak normality and semi-weak normality coincide in an ideal. We show the behavior of a least function for an ideal affects the weak normality of the ideal. We conclude the chapter by looking at two different methods of creating a weakly normal ideal. In Chapter 4, we begin by formally introducing Todorˇcevi´c’smethod of minimal walks. Then, we discuss an ideal due to Shelah that will be crucial in our proofs: idp(C, I). After characterizing the elements of this ideal, we use a club-guessing sequence from Chapter 2 to define the coloring c that establishes the main theorem.
2 A corollary to the Main Theorem is if κ → [κ]κ, then our ideal is weakly κ-saturated. 17
2 Existence of Nice Club-Guessing Sequences
2.1 Club-Guessing Sequences
The main goal of this chapter is to prove the following theorem:
THEOREM 2.1.1 (Shelah, [13]) Suppose κ is (weakly) inaccessible, λ < κ is an
κ uncountable regular cardinal, and S is a stationary subset of Sλ . Then there is an
S-club sequence hCδ : δ ∈ Si such that
(i) Cδ is club in δ with order-type λ,
(ii) hcf(α): α ∈ nacc(Cδ)i is strictly increasing with limit δ, and
(iii) for every club E ⊆ κ there are stationarily many δ ∈ S with Cδ ⊂ E.
This theorem was stated by Shelah in Cardinal Arithmetic, but the proof given was incorrect. A similar result dealing with successors of singular cardinals was also claimed in Cardinal Arithmetic; the flawed proof in this situation was repaired by Eisworth and Shelah in [7]. We adapt their techniques to give the above result for inaccessible cardinals. Before presenting our proof, we discuss club-guessing sequences in general and outline some of the techniques we will be using. We begin with some vocabulary.
DEFINITION 2.1.1 Suppose C is a club subset of an ordinal δ. Then
(1) acc(C) = {α ∈ C : α = sup(C ∩ α)}, and
(2) nacc(C) = C \ acc(C).
For a set C, acc(C) denotes the set of accumulation points of C and nacc(C) is the set of non-accumulation points. 18
DEFINITION 2.1.2 Let κ be a cardinal and S ⊆ κ be a stationary set of limit
ordinals. We say that a sequence C = hCδ : δ ∈ Si is an S-club system if Cδ is a club subset of δ for each δ ∈ S.
When S = κ, we call the S-club sequence a C-sequence.
DEFINITION 2.1.3 Suppose κ is an uncountable cardinal and S ⊆ κ is a station- ary subset. We say hCδ : δ ∈ Si is an S-club guessing sequence if C is an S-club sequence such that for each club E ⊆ κ, there are stationarily many δ ∈ S such that
Cδ “guesses” E ∩ δ.
There are a few ways an S-club guessing sequence can “guess” a club. When we want to “guess” a club, we typically require one of the following:
1. Cδ ⊆ E,
2. Cδ \ E is bounded in δ,
3. δ = sup(E ∩ nacc(Cδ)), or
4. δ = sup(Cδ ∩ E).
The following concepts are due to Shelah and will be used in our proof. Rather than using their original names, we will use the more descriptive names (due to Kojman in [10]).
DEFINITION 2.1.4 Let C and E be sets of ordinals with E closed. Define
Drop(C,E) = {sup(E ∩ α): α ∈ C \ min(E) + 1}.
If α ∈ nacc(C), then we define Gap(α, C), the gap in C determined by α, by
Gap(α, C) = (sup(C ∩ α), α). 19
If C and E are both subsets of some cardinal κ and heα : α < κi is a C-sequence, then for each α ∈ nacc(C) ∩ acc(E), we define
F ill(α, C, E) = Drop(eα,E) ∩ Gap(α, C).
These operations are used to modify a given S-club sequence in an attempt to transform it into a sequence that guesses clubs. We first lay out a few properties of the Drop operation.
FACT 2.1.1 Suppose κ is a cardinal, E ⊆ κ is a club, δ ∈ acc(E) and C is club in δ. Then
i). Drop(C,E) ⊆ E.
ii). Drop(C,E) is club in δ.
Proof: The proof of these facts is fairly straightforward. The first fact is true because sup(E ∩ α) ∈ E as E is club. For the second fact, we start by showing Drop(C,E) is unbounded in δ. Let γ < δ and choose α ∈ C and β ∈ E such that
γ < β < α.
This can be done because C is club in δ and δ ∈ acc(E). From this, we have
γ < β ≤ sup(E ∩ α).
As sup(E ∩ α) ∈ Drop(C,E), we have Drop(C,E) is unbounded in δ. Next, we show Drop(C,E) is closed. Suppose ξ < cf(δ) and
haα : α < ξi ⊂ Drop(C,E) is a strictly increasing sequence with
a = sup{aα : α < ξ}. 20
For each α < ξ, there is bα ∈ C such that
aα = sup(E ∩ bα).
Since C is closed in δ, we have b = sup{bα : α < ξ} ∈ C. From this, we get a ≤ sup(E ∩ b). We actually have equality because if a < sup(E ∩ b), then there would be α < ξ such that a < aα. So Drop(C,E) is club in δ.
The following theorem by Shelah is one of the first results for club-guessing se- quences. This proof is a prototype of most results relating to the existence of a club-guessing sequence. The general idea behind the proof is if there is no club- guessing sequence with the desired properties, then we correct the sequence using the Drop. Now, if this new sequence does not meet the requirements, then we correct the new sequence using the Drop. Then, we argue that eventually one of these corrected sequences has the required properties.
THEOREM 2.1.2 (Shelah, [13]) Suppose λ < κ are regular cardinals with λ+ <
κ κ, and S ⊆ Sλ is stationary. Then there exists an S-club guessing sequence hCδ : δ ∈ Si such that for any club E ⊆ κ, the set
{δ ∈ S : Cδ ⊂ E} is stationary.
Proof: Assume the contrary and suppose there is no sequence hCδ : δ ∈ Si with the
+ desired property. We will construct a sequence hEξ : ξ < λ i of clubs of κ and a ξ sequence of club-guessing sequences C to arrive at a contradiction. To start with,
0 0 let E0 = κ and C be a sequence of subsets of κ such that Cδ ⊂ δ and otp(Cδ) = λ for each δ ∈ S. ζ When ξ = ζ + 1, C does not have the property we are looking for. So, there is a
0 club Eζ ⊆ κ such that ζ 0 {δ ∈ S : Cδ ⊂ Eζ } 21
is non-stationary, or
0 ζ 0 δ ∈ S ∩ Eζ ⇒ Cδ 6⊂ Eζ .
0 ζ We define Eξ = acc(Eζ ∩ Eζ ). To define C , we will use the Drop operation. If
δ ∈ acc(Eξ) ∩ S, then we let
ξ Cδ = Drop(Cδ,Eξ).
ξ If δ ∈ nacc(Eξ) ∩ S, then let Cδ = Cδ. When ξ is a limit ordinal, we let
\ Eξ = Eζ . ζ<ξ
ξ And, we modify C just like above. Let \ E = acc Eξ , ξ<λ+ + and let δ ∈ S ∩ E. Since δ ∈ S ∩ E, we know δ ∈ S ∩ Eξ for all ξ < λ , which means
ξ Cδ 6⊂ Eξ
+ + ξ for all ξ < λ . As δ ∈ acc(Eξ) for all ξ < λ , it follows that Cδ = Drop(Cδ,Eξ).
+ For each α ∈ Cδ, there is ξα < λ such that for any ξ ≥ ξα,
sup(Eξα ∩ α) = sup(Eξ ∩ α).
+ This follows from the fact that hsup(Eξ ∩ α): ξ < λ i is a non-increasing sequence
+ of ordinals and must be eventually constant. Since otp(Cδ) = λ < λ , we know
∗ + ξ = sup{ξα : α ∈ Cδ} + 1 < λ .
By construction, for any α ∈ Cδ, we have
sup(Eξ∗ ∩ α) = sup(Eξ∗+1 ∩ α) ∈ Eξ∗+1. 22
This in turn shows that
ξ∗ ξ∗ Cδ = Drop(Cδ ,Eξ∗ ) ⊂ Eξ∗+1,
contradicting the choice of Eξ∗+1.
Theorem 2.1.2 establishes the existence of a club-guessing sequence on a lot of cardinals, which is nice, but there is room for improvement. When κ is the successor of a singular cardinal, Theorem 2.1.2 can be improved, which is stated as Claim 2.6 in [13]. The new club-guessing sequence still guesses clubs in the previous fashion with a stronger condition on the non-accumulation points.
THEOREM 2.1.3 (Shelah, claimed in [13], correct proof in [7]) Suppose κ =
+ κ µ , ω < λ = cf(µ) < µ, and S ⊆ Sλ is stationary. Then there is a sequence hCδ : δ ∈ Si such that
(i). otp(Cδ) = λ,
(ii). hcf(α): α ∈ nacc(Cδ)i is a strictly increasing sequence in δ, and
(iii). for every club E ⊆ κ, there are stationarily many δ ∈ S such that Cδ ⊂ E.
Club-guessing sequences from with properties (i)-(iii) are called “nice club-guessing sequences.” Theorem 2.1.3 shows that nice club-guessing sequences exist when κ is the successor of a singular cardinal of uncountable cofinality. As mentioned earlier, we give a correct version of Theorem 2.1.3 in the context of inaccessible cardinals. Shelah also investigates club-guessing sequences at Mahlo cardinals, taking a look at the existence of S-club guessing sequences when S is a stationary set of inaccessible cardinals. The results he obtains are stated without proof, and it is not clear if they are correct. In the last section of this chapter, we will achieve a partial rescue of one of his claims. 23
2.2 Nice Club-Guessing Sequences: Uncountable Cofinality
In this section, we prove Theorem 1.2.1 from the introduction. Our first step is to prove the following lemma:
LEMMA 2.2.1 Suppose κ is an inaccessible cardinal, λ < κ is an uncountable reg-
κ ular cardinal, and S ⊆ Sλ is stationary. Then there exists an S-club sequence such that:
(i). |Cδ| < δ, and
(ii). for every club E ⊂ κ, there are stationarily many δ ∈ S such that for all τ < δ,
{α ∈ nacc(Cδ) ∩ E : cf(α) > τ} is unbounded in δ.
κ Proof: Let S ⊆ Sλ is stationary. By way of contradiction, assume that there is no
S-club sequence with property (ii). Let hCδ : δ ∈ Si be any S-club sequence such
that |Cδ| < δ. We will construct by recursion on n < ω
n n • C := hCδ : δ ∈ Si, an S-club sequence,
n • τδ < δ, a regular cardinal,
n • δ < δ, and
• En ⊆ κ, a club in κ.
0 0 0 n n Let C = hCδ : δ ∈ Si, E0 = κ, δ = 0 and τδ = 0. Suppose the objects C , En, δ
n and τδ have been constructed. Now we describe the n + 1 stage of the construction.
0 By our assumption, there is a club En ⊆ κ such that
n 0 {δ ∈ S : ∀τ < δ({α ∈ nacc(Cδ ) ∩ En : cf(α) > τ} is unbounded)}
1 is non-stationary. So, there is another club En ⊆ κ such that
1 n 0 En ∩ {δ ∈ S : ∀τ < δ({α ∈ nacc(Cδ ) ∩ En : cf(α) > τ} is unbounded)} = ∅ 24
1 or if δ ∈ En ∩ S, then there are τ, < δ such that
n 0 α ∈ nacc(Cδ ) ∩ En \ ( + 1) ⇒ cf(α) ≤ τ.
0 1 n+1 n+1 Define En+1 = acc(En ∩ En ∩ En), and δ , τδ < δ be least and τ corresponding to the above comment.
We say δ is active at stage n + 1 if δ ∈ acc(En+1). If δ ∈ S is inactive at stage
n+1 n n+1 n n+1 n n + 1, we let Cδ = Cδ , δ = δ and τδ = τδ . Suppose δ is active at stage n + 1. We say α < δ needs attention at stage n + 1 if
n n+1 α ∈ nacc(Cδ ) ∩ acc(En+1) \ δ + 1.
n+1 Any ordinal α < δ that needs attention at stage n + 1 has cofinality less than τδ by our earlier comment.
n+1 n n n To construct Cδ , we first we let Dδ = Drop(Cδ ,En+1). The set Dδ is club in
n δ as δ ∈ acc(En+1) and Cδ is club in δ. If α needs attention at stage n + 1, then
n α ∈ acc(En+1) and α = sup(α ∩ En+1). We also have α ∈ nacc(Dδ ), for if this isn’t
n the case, then we get a contradiction to fact that α ∈ nacc(Cδ ). This fact allows us
n to conclude that Fill(α, Dδ ,En+1) is defined for any α that needs attention at stage n + 1. Now, we define
n+1 n [ n Cδ = Dδ ∪ {Fill(α, Dδ ,En+1): α needs attention)}.
n+1 n n+1 n+1 The set Cδ is unbounded in δ because Dδ ⊂ Cδ , and it is closed because Cδ
n n+1 was built by adding closed sets into the gaps of Dδ . We also have |Cδ | < δ because
n+1 n n+1 n |Cδ | ≤ |Cδ | + τδ · |Cδ | < δ.
n n n The previous inequality holds for |Dδ | ≤ |Cδ | < δ, we added in at most |Dδ | many n+1 n+1 sequences of length at most τδ . C has been constructed to fulfill our needs, and the recursion can go on. 25
T Let E = n<ω En. For each δ ∈ acc(E) ∩ S, we define
n δ = sup{δ : n < ω} and
n θδ = sup{|Cδ | : n < ω}.
n Each δ and θδ is strictly less than δ because ω < cf(δ) and |Cδ | < δ. Next, we let
ηδ = max{δ, θδ} + 1,
and we have ηδ < δ. Using Fodor’s theorem, we can pass to a stationary subset
1 ∗ 1 S ⊆ S ∩ acc(E) such that ηδ = η for each δ ∈ S .
1 ∗ + Now, we can choose δ ∈ S such that |η | divides otp(E ∩ δ). As E ⊂ acc(En) for all n, we have δ is active at all stages of the construction. Since δ ∈ acc(E) and |η∗|+ divides otp(E ∩ δ), |E ∩ δ \ η∗| ≥ |η∗|+.
As |η∗|+ > η∗, we can find an ordinal γ such that
• γ ∈ E
• η∗ < γ < δ and
S n • γ 6∈ n<ω Cδ .
Now, we construct the sequence hαn : n < ωi of ordinals by
n αn = min(Cδ \ γ), and prove this sequence is strictly decreasing.
S n By definition, αn is strictly greater than γ because γ 6∈ n<ω Cδ . This means
n n+1 ∗ αn is an element of nacc(Cδ ). From the definition of αn, we know δ < η < αn.
There are two cases to consider with αn: either αn needs attention at stage n+1 or not. 26
Case 1: αn does not need attention at stage n + 1.
Since αn does not need attention at stage n + 1, we have αn 6∈ acc(En+1), and by letting βn = sup(En+1 ∩ αn) < αn, we have, as γ ∈ E ⊂ En+1, γ ≤ βn < αn. Also, βn
n n+1 is an element of Dδ ⊂ Cδ , so αn+1 ≤ βn < αn.
Case 2: αn needs attention at stage n + 1.
n In this situation, αn ∈ nacc(Cδ)∩acc(En+1), and Fill(αn,Dδ ,En+1) is defined and club in αn. Since γ < αn, we have
n γ < αn+1 ≤ min(Fill(αn,Dδ ,En+1)) < αn
and αn+1 < αn. In both cases, we have αn+1 < αn, and hαn : n < ωi is a strictly decreasing sequence of ordinals, a contradiction. Hence, there is an S-club guessing sequence with the desired properties.
Even though the bound on this sequence is not uniform, we can prove there is a sequence with a uniform bound on all the terms and have the same guessing property. In the upcoming proofs, we will be working with property (ii) quite a bit, so we will introduce some ad-hoc terminology. Let E be a club of κ. We say C guesses E at δ if
{α ∈ E ∩ nacc(Cδ): cf(α) > τ} is unbounded in δ for each τ < δ.
COROLLARY 2.2.1 Suppose κ is an inaccessible cardinal, λ < κ is an uncountable
κ regular cardinal, and S ⊆ Sλ is stationary. There is a C-sequence C = hCδ : δ ∈ Si and µ < κ such that
(i). |Cδ| < µ, and 27
(ii). for every club E ⊆ κ, there are stationarily many δ ∈ S such that for all τ < δ,
{α ∈ nacc(Cδ) ∩ E : cf(α) > τ} is unbounded in δ.
Proof: Let C be the S-club guessing sequence from Lemma 2.2.1. Assume the contrary. Suppose for each µ < κ, there is a club Eµ ⊆ κ such that if C guesses Eµ at δ, then
|Cδ| ≥ µ.
Now, let E = {δ < κ :(∀α < δ)(δ ∈ Eα)} which is club in κ. Since there are stationarily many places where C guesses E, we can pick δ ∈ E where C guesses E
at δ. Since C guesses E at δ, it follows that C guesses Eα at δ for each α < δ. This in turn implies that
|Cδ| ≥ α
for all α < δ, or
|Cδ| ≥ δ.
This contradicts (i) of Lemma 2.2.1, and the corollary is proved.
The next lemma shows the S-club sequence from Lemma 2.2.1 implies the exis- tence of a nice S-club guessing sequence.
LEMMA 2.2.2 Suppose κ is an inaccessible cardinal, λ < κ is an uncountable reg-
κ ular cardinal, and S ⊆ Sλ is a stationary subset. Then the following are equivalent:
(1). There is an S-club system hCδ : δ ∈ Si such that
(i). |Cδ| < δ, and
(ii) for every club E ⊆ κ, there are stationarily many δ ∈ S such that for all
τ < δ, {α ∈ nacc(Cδ) ∩ E : cf(α) > τ} is unbounded in δ.
(2). There is an S-club system hCδ : δ ∈ Si and µ < κ such that 28
(i). |Cδ| < µ, and
(ii) for every club E ⊆ κ, there are stationarily many δ ∈ S such that for all
τ < δ, {α ∈ nacc(Cδ) ∩ E : cf(α) > τ} is unbounded in δ.
(3). There is an S-club system hCδ : δ ∈ Si such that
(i). otp(Cδ) = λ,
(ii). hcf(α): α ∈ nacc(Cδ)i is strictly increasing sequence with limit δ, and
(iii). for every club E ⊆ κ, there are stationarily many δ ∈ S such that Cδ ⊂ E.
Proof: For (1) ⇒ (2), this is just Corollary 2.2.1. Now, we show (2) ⇒ (3). Our proof and notation is similar to the proof of Theorem 1 in [2]. To start, we prove if C is as in (2), then there is a club E∗ ⊆ κ such that for every club E ⊆ κ, there are stationarily many δ ∈ S where C guesses
∗ ∗ acc(E ) at δ and Drop(Cδ,E ) ⊂ E Suppose this is not the case. By recursion, we will construct a µ-sequence of clubs
Eξ of κ as follows:
Case 1: ξ = 0: We start with E0 = κ,
T Case 2: ξ is a limit: We let Eξ = ζ<ξ Eζ .
∗ Case 3: ξ = ζ + 1: From our assumption, Eζ cannot be our E . This means we
0 1 0 can find clubs Eζ and Eζ such that for all δ ∈ Eζ , if C guesses acc(Eζ ) at δ, then
1 Drop(Cδ,Eζ ) 6⊆ Eζ ,
or there is α ∈ Cδ \ min(Eζ ) + 1 such that
1 sup(Eζ ∩ α) 6∈ Eζ . 29
0 1 We define the next club Eξ by Eξ = acc(Eζ ∩ Eζ ∩ Eζ ). T Now, let E = ξ<µ Eξ, which is club in κ, so by (ii) of (2), we can find δ ∈ S where C guesses E at δ and δ > µ. Since C guesses E at δ, it follows that δ ∈ E, and from
0 the definition of E, we know δ ∈ Eξ for each ξ < µ. Also, we have C guesses acc(Eξ)
at δ because E ⊆ acc(Eξ). Hence, for each ξ < µ, there is α ∈ Cδ \ min(Eξ) + 1 with
1 sup(Eξ ∩ α) 6∈ Eξ ⊇ Eξ+1.
As hEξ : ξ < µi is a decreasing sequence of clubs, for each α ∈ Cδ \ min(E) + 1, it
follows that hsup(Eξ ∩ α): ξ < µi is non-increasing sequence and must be eventually
constant. For each α ∈ Cδ, there are γα < δ and ξα < µ such that if ξα ≤ ξ, then
sup(Eξ ∩ α) = γα.
Since |Cδ| < µ, it follows that
∗ ξ = sup{ξα : α ∈ Cδ} < µ.
Since C guesses acc(Eξ∗ ) at δ, there is α ∈ Cδ \ min(Eξ∗ ) + 1 such that
sup(Eξ∗ ∩ α) 6∈ Eξ∗+1.
∗ Yet ξ ≥ ξα, and we have
sup(Eξ∗ ∩ α) = γα = sup(Eξ∗+1 ∩ α) ∈ Eξ∗+1,
a contradiction. With the existence of such E∗, we can continue our proof. If C guesses acc(E∗)
at δ, we can construct Dδ by recursion on λ. Before we start, for each δ for which C
∗ δ guesses acc(E ), fix an increasing λ-sequence hτξ : ξ < λi which is cofinal in δ.
δ First, for ξ = 0, let α0 = 0. For ξ = ζ + 1, choose
δ ∗ δ αξ ∈ {α ∈ nacc(Cδ) ∩ acc(E ): cf(α) > τξ } 30
δ δ ∗ such that αζ < αξ. This is possible since C guesses acc(E ) at δ. When ξ is a limit,
we let αξ = sup{αζ : ζ < ξ}, and our construction can continue. In the end, we define
Dδ = {αξ : ξ < λ}.
∗ From our method of construction, when C guesses acc(E ) at δ, Dδ has the fol- lowing properties:
∗ (i). Dδ ⊆ Drop(Cδ,E )
(ii). Dδ is club in δ with otp(Dδ) = cf(δ), and
(iii). hcf(α): α ∈ nacc(Dδ)i is increasing and cofinal in δ.
∗ ∗ Fact (i) holds because Dδ ⊂ Cδ ∩ acc(E ). If α ∈ Dδ, then sup(E ∩ α) = α and
α ∈ Cδ. Fact (ii) and (iii) hold by how we constructed Dδ. For other δ ∈ S, we can pick Dδ to be a subset of δ satisfying (ii) and (iii). It is easy to see hDδ : δ ∈ Si is an S-club system meeting the requirements of (3), and (2) ⇒ (3). It is clear to see that (3) ⇒ (1).
Recall, the goal of this section was to prove the following theorem:
Theorem 2.1.1: Suppose κ is (weakly) inaccessible, λ < κ is an uncountable regular
κ cardinal, and S is a stationary subset of Sλ . Then there is an S-club sequence hCδ : δ ∈ Si such that
(i) Cδ is club in δ with order-type λ,
(ii) hcf(α): α ∈ nacc(Cδ)i is strictly increasing with limit δ, and
(iii) for every club E ⊆ κ, there are stationarily many δ ∈ S with Cδ ⊂ E.
Combining Lemma 2.2.2 and Corollary 2.2.1, we get the existence of a nice S-
κ club sequence on an inaccessible cardinal κ when S ⊆ Sλ for a regular cardinal λ of uncountable cofinality. This establishes Theorem 2.1.1. 31
Finding a nice S-club guessing sequences when S is a stationary subset of ordinals with countable confinality is a bit tricky. Club subsets of an ordinal with countable cofinality only need to be unbounded, and this fact causes trouble. It is an open question whether nice S-club guessing sequences exist when S is a stationary subset consisting of ordinals with countable cofinality. In [7], Eisworth and Shelah constructed a “well-formed” S-club sequence to replace a nice S-club guessing sequence. At a first glance, it seems possible for there to be a well-formed S-club guessing sequence when S is a stationary subset consisting of ordinals of countable cofinality. The S-club guessing results obtained in this section used arguments very similar to those used in the case of successors of a singular cardinals. The next step would be to find conditions for the existence of a well- formed S-club guessing sequence when S consists of ordinals of countable cofinality. This leads to the following questions.
κ QUESTION 2.2.1 Suppose κ is an inaccessible cardinal and S ⊆ Sω is a stationary subset of κ. Does there exist a well-formed S-club guessing sequence?
A positive answer would be a first step towards a partition theorem when S consists of ordinals of countable cofinality.
2.3 Constructing A Club-Guessing Sequence Using ⊗C
In this section, we will focus on constructing a club-guessing sequence when the stationary subset consists of regular cardinals. The proof uses a club-sequence having the following property:
⊗C : For every club E ⊆ κ, there are stationarily many δ ∈ S such that E ∩ δ \ Cδ is unbounded in δ 32
Under what situations is there an S-club sequence satisfying ⊗C ? Suppose λ <
κ κ are regular cardinals, S ⊆ Sλ is stationary, and C is an S-club sequence with
otp(Cδ) = λ for each δ ∈ S. Then ⊗C holds. Why? If E ⊆ κ is a club, then
{δ < κ : otp(E ∩ δ) = δ}
is club in κ. Now, for any δ ∈ S ∩ {δ < κ : otp(E ∩ δ) = δ}, we have otp(E ∩ δ) = δ
and otp(Cδ) = λ < δ. Since otp(Cδ) = λ < δ = otp(E ∩ δ), it follows that E ∩ δ \ Cδ
is unbounded in δ. Hence, ⊗C holds.
What makes ⊗C an interesting property is that it is SOMETIMES possible to
find an S-club sequence satisfying ⊗C when S is a stationary subset of inaccessible cardinals:
CLAIM 2.3.1 (Shelah, [13]) Suppose κ is an inaccessible cardinal and there is a sequence hSi : i < κi of stationary subsets of κ such that for each δ < κ, Si ∩ δ is not stationary in δ for some i < δ. Then for every stationary S ⊆ κ consisting of
inaccessible cardinals, there is an S-club system satisfying ⊗C .
Along with conditions for the existence of an S-club sequence satisfying ⊗C , Shelah
showed the existence of an S-club sequence on κ satisfying ⊗C implies κ cannot be weakly compact. This was done by way of constructing a function c :[κ]2 → ω that has an infinite range on every unbounded subset of κ (see Claim 4.9 in [13]).
The property ⊗C was introduced in Claim 2.12 of Chapter III of [13]. Claim 2.12 states the existence of a nice S-club guessing sequence from ⊗C and conditions for
this modified club-guessing sequence to satisfy ⊗C , but there is a problem with Claim 2.12. We give a partial rescue to Claim 2.12 with Lemma 2.3.1, but before proving the lemma, we need two facts about Drop.
FACT 2.3.1 Suppose E ⊆ κ is club, δ ∈ acc(E) and Cδ is club in δ. If β ∈
acc(Drop(Cδ,E)), then β ∈ Cδ. 33
Proof: Given β ∈ acc(Drop(Cδ,E)), there is an increasing sequence hβi : i <
cf(β)i ⊂ Drop(Cδ,E) converging to β. As βi ∈ Drop(Cδ,E), there is αi ∈ Cδ such that
βi = sup(E ∩ αi)
for each i < cf(β). Since hβi : i < cf(β)i is increasing and converges to β, it follows
that hαi : i < cf(β)i is increasing and converges to β. As Cδ is club in δ and αi → β, we have β ∈ C.
Using Fact 2.3.1, we get a slightly more useful fact.
FACT 2.3.2 Suppose E ⊆ κ is a club, δ ∈ acc(E), Cδ is club in δ. If β ∈
Drop(Cδ,E) and β 6∈ Cδ, then β ∈ nacc(Drop(Cδ,E)).
Proof: Since β ∈ Drop(Cδ,E), we know either β ∈ nacc(Drop(Cδ,E)) or β ∈
acc(Drop(Cδ,E)). From Fact 2.3.1 and knowing β 6∈ Cδ, the only possibility is
β ∈ nacc(Drop(Cδ,E)).
Now, we can prove the lemma.
LEMMA 2.3.1 (Shelah, [13]) Suppose κ is an inaccessible cardinal, S is a station- ary subset of {α < κ : α is an inaccessible}, and there is an S-club system C that sat-
∗ ∗ isfies ⊗C . There is a club E ⊆ κ such that for hDδ : δ ∈ Si with Dδ = Drop(Cδ,E ) has the following property
(1) For every club E ⊂ κ, there are stationarily many δ ∈ S such that
δ = sup(E ∩ nacc(Dδ)).
Proof: Assume the contrary, and suppose for every club E of κ, property (1) fails. This means for each club E ⊆ κ, there is a club F of κ such that
δ 6= sup(F ∩ nacc(Drop(Cδ,E))) 34
for every δ ∈ S. Now, we will construct an ω-sequence, hEn : n < ωi, of clubs of κ
δ and βn for each δ ∈ S.
δ For n = 0, we define E0 = λ and β0 = 0 for each δ ∈ S.
δ When n = k + 1, suppose Ek and βk for each δ ∈ S have been defined. Since
0 property (1) fails to hold for any club of κ, there is a club Ek ⊆ κ such that for any δ ∈ S, we have
0 δ 6= sup(Ek ∩ nacc(Drop(Cδ,Ek))).
δ For each δ ∈ S, there is γk < δ such that
0 δ sup(Ek ∩ nacc(Drop(Cδ,Ek))) = γk.
Next, we let
δ δ δ βk+1 = max{γk, βk} + 1 and