166 (2000)
On a problem of Steve Kalikow
by
Saharon S h e l a h (Jerusalem and New Brunswick, NJ)
Abstract. The Kalikow problem for a pair (λ, κ) of cardinal numbers, λ > κ (in particular κ = 2) is whether we can map the family of ω-sequences from λ to the family of ω-sequences from κ in a very continuous manner. Namely, we demand that for η, ν ∈
ωλ we have: η, ν are almost equal if and only if their images are.
We show consistency of the negative answer, e.g., for ℵ
ωbut we prove it for smaller cardinals. We indicate a close connection with the free subset property and its variants.
0. Introduction. In the present paper we are interested in the following property of pairs of cardinal numbers:
Definition 0.1. Let λ, κ be cardinals. We say that the pair (λ, κ) has the Kalikow property (and then we write KL(λ, κ)) if there is a sequence hF
n: n < ωi of functions such that
F
n:
nλ → κ (for n < ω) and if F :
ωλ →
ωκ is given by
(∀η ∈
ωλ)(∀n ∈ ω)(F (η)(n) = F
n(η n)) then for every η, ν ∈
ωλ,
(∀
∞n)(η(n) = ν(n)) iff (∀
∞n)(F (η)(n) = F (ν)(n)).
In particular we answer the following question of Kalikow:
Kalikow Problem 0.2. Is KL(2
ℵ0, 2) provable in ZFC?
The Kalikow property of pairs of cardinals was studied in [Ka90]. Several results are known already. Let us mention some of them. First, one can easily notice that
KL(λ, κ) & λ
0≤ λ & κ
0≥ κ ⇒ KL(λ
0, κ
0).
2000 Mathematics Subject Classification: 03E35, 03E05.
Key words and phrases: set theory, forcing, continuity, Kalikow, free subset.
The research was partially supported by the Israel Science Foundation. Publication 590.
[137]
Also (“transitivity”)
KL(λ
2, λ
1) & KL(λ
1, λ
0) ⇒ KL(λ
2, λ
0) and
KL(λ, κ) ⇒ λ ≤ κ
ℵ0.
Kalikow proved that CH implies KL(2
ℵ0, 2) (in fact that KL(ℵ
1, 2) holds true) and he conjectured that CH is equivalent to KL(2
ℵ0, 2).
The question 0.2 is formulated in [Mi91, Problem 15.15, p. 653].
We shall prove that KL(λ, 2) is closely tied with some variants of the free subset property (both positively and negatively). First we present an answer to problem 0.2 proving the consistency of ¬KL(2
ℵ0, 2) in 1.1 (see 2.8 too). Later we discuss variants of the proof (concerning the cardinal and the forcing). Then we deal with a positive answer, in particular KL(ℵ
n, 2), and we show that the negation of a relative of the free subset property for λ implies KL(λ, 2).
We thank the participants of the Jerusalem Logic Seminar 1994/95 and particularly Andrzej Ros lanowski for writing it up so nicely.
Notation. We will use the Greek letters κ, λ, χ to denote (infinite) cardinals and the letters α, β, γ, ζ, ξ to denote ordinals. Sequences of ordinals will be called ¯ α, ¯ β, ¯ ζ with the usual convention that ¯ α = hα
n: n < lg( ¯ α)i etc. Sets of ordinals will be denoted by u, v, w (with possible indexes).
The quantifiers (∀
∞n) and (∃
∞n) are abbreviations for “for all but finitely many n ∈ ω” and “for infinitely many n ∈ ω”, respectively.
1. The negative result. For a cardinal χ, the forcing notion C
χfor adding χ many Cohen reals consists of finite functions p such that for some w ∈ [χ]
<ω, n < ω,
dom(p) = {(ζ, k) : ζ ∈ w & k < n} and rang(p) ⊆ 2 ordered by inclusion.
Theorem 1.1. Assume λ → (ω
1· ω)
<ω2κ, 2
κ< λ ≤ χ. Then
Cχ¬KL(λ, κ) and hence
Cχ¬KL(2
ℵ0, 2).
P r o o f. Suppose that C
χ-names F
˜
n(for n ∈ ω) and a condition p ∈ C
χare such that
p
Cχ“hF
˜
n: n < ωi exemplifies KL(λ, κ)”.
For ¯ α ∈
nλ choose a maximal antichain hp
nα,l¯: l < ωi of C
χdeciding the values of F
˜
n( ¯ α). Thus we have a sequence hγ
α,ln¯: l < ωi ⊆ κ such that p
nα,l¯Cχ
F
˜
n( ¯ α) = γ
α,ln¯.
Let χ
∗be a sufficiently large regular cardinal. Take an elementary submodel M of (H(χ
∗), ∈, <
∗χ∗) such that
• kM k = χ, χ + 1 ⊆ M ,
• hp
nα,l¯: l < ω, n ∈ ω, ¯ α ∈
nλi, hγ
α,ln¯: l < ω, n ∈ ω, ¯ α ∈
nλi ∈ M .
By λ → (ω
1· ω)
<ω2κ(see [Sh 481, Claim 1.3]), we find a set B ⊆ λ of indiscernibles in M over
κ ∪ {hp
nα,l¯: l < ω : n ∈ ω, ¯ α ∈
nλi, hγ
nα,l¯: l < ω : n ∈ ω, ¯ α ∈
nλi, χ, p}
and a system hN
u: u ∈ [B]
<ωi of elementary submodels of M such that (a) B is of order type ω
1· ω and for u, v ∈ [B]
<ω:
(b) κ + 1 ⊆ N
u,
(c) χ, p, hp
nα,l¯: l < ω, n < ω, ¯ α ∈
nλi, hγ
α,ln¯: l < ω, n < ω, ¯ α ∈
nλi ∈ N
u, (d) |N
u| = κ, N
u∩ B = u,
(e) N
u∩ N
v= N
u∩v,
(f) |u| = |v| ⇒ N
u∼ = N
v, and let π
u,v: N
v→ N
ube this (unique) isomorphism,
(g) π
v,v= id
Nv, π
u,v(v) = u, π
u0,u1◦ π
u1,u2= π
u0,u2, (h) if v
0⊆ v, |v| = |u| and u
0= π
u,v(v
0) then π
u0,v0⊆ π
u,v. Note that if u ⊆ B is of order type ω then we may define
N
u= [
{N
v: v is a finite initial segment of u}.
Then the models N
u(for u ⊆ B of order type ≤ ω) have the properties (b)–(h) too.
Let hβ
ζ: ζ < ω
1· ωi be the increasing enumeration of B. For a set u ⊆ B of order type ≤ ω let ¯ β
ube the increasing enumeration of u (so lg( ¯ β
u) = |u|).
Let u
∗= {β
ω1·n: n < ω}. For k ≤ ω and a sequence ¯ ξ = hξ
m: m < ki ⊆ ω
1we define
u[ ¯ ξ] = {β
ω1·m+ξm: m < k} ∪ {β
ω1·n: n ∈ ω \ k}.
Now, working in V
Cχ, we say that a sequence ¯ ξ
˜ is k-strange if (1) ¯ ξ
˜
is a sequence of countable ordinals greater than 0, lg( ¯ ξ
˜
) = k ≤ ω, (2) (∀m < ω)(F
˜
m( ¯ β
u[ ¯˜ξ]m) = F
˜
m( ¯ β
u∗m)).
Claim 1.1.1. In V
Cχ, if ¯ ξ
˜
k
are k-strange sequences (for k < ω) such that (∀k < ω)( ¯ ξ
˜
k
C ¯ ξ
˜
k+1
) then the sequence ¯ ξ
˜ := S
k<ω
ξ ¯
˜
k
is ω-strange.
P r o o f. Should be clear (note that in this situation we have ¯ β
u[ ¯˜ξ]m = β ¯
u[ ¯˜ξm]m).
Claim 1.1.2. p
Cχ“there are no ω-strange sequences”.
P r o o f. Assume not. Then we find a name ¯ ξ
˜
= hξ
˜
m: m < ωi for an ω-sequence and a condition q ≥ p such that
q
Cχ“(∀m < ω)(0 < ξ
˜
m< ω
1& F
˜
m( ¯ β
u[ ¯˜ξ]m) = F
˜
m( ¯ β
u∗m))”.
By the choice of p and F
˜
mwe conclude that
q
Cχ“(∀
∞m)( ¯ β
u[ ¯˜ξ](m) = ¯ β
u∗(m))”,
which contradicts the definition of ¯ β
u[ ¯˜ξ], ¯ β
u∗, Definition 0.1 and the fact that
q
Cχ“(∀m < ω)(0 < ξ
˜
m< ω
1)”.
By 1.1.1, 1.1.2, any inductive attempt to construct (in V
Cχ) an ω-strange sequence ¯ ξ
˜
has to fail. Consequently, we find a condition p
∗≥ p, an integer k < ω and a sequence ¯ ξ = hξ
l: l < ki such that
p
∗Cχ
“ ¯ ξ is k-strange but ¬(∃ξ < ω
1)( ¯ ξ
_hξi is (k + 1)-strange)”.
Then in particular
( ) p
∗Cχ
“(∀m < ω)(F
˜
m( ¯ β
u[ ¯ξ]m) = F
˜
m( ¯ β
u∗m))”.
[It may happen that k = 0, i.e., ¯ ξ = hi.]
For ξ < ω
1let u
ξ= u[ ¯ ξ
_hξi] and w
ξ= u
ξ∪ (u
∗\ {ω
1· k}). Thus w
0= u[ ¯ ξ] ∪ u
∗and all w
ξhave order type ω and π
wξ1,wξ2is the identity on N
wξ\{ω1·k+ξ2}. Let q := p
∗N
w0and q
ξ= π
wξ,w0(q) ∈ N
wξ(so q
0= q). As the isomorphism π
wξ,w0is the identity on N
w0∩ N
wξ= N
w0∩wξ(and by the definition of Cohen forcing), we see that the conditions q, q
ξare compatible.
Moreover, as p
∗≥ p and p ∈ N
∅, we find that both q and q
ξare stronger than p.
Now fix ξ
0∈ (0, ω
1) (e.g. ξ
0= 1) and look at the sequences ¯ β
uξ0and β ¯
u∗. They are eventually equal and hence
p
Cχ“(∀
∞m)(F
˜
m( ¯ β
uξ0m) = F
˜
m( ¯ β
u∗m))”.
So we find m
∗< ω and a condition q
ξ00≥ q
ξ0, q such that (⊗
ξq00,m∗ξ0
) q
ξ00Cχ
“(∀m ≥ m
∗)(F
˜
m( ¯ β
uξ0m) = F
˜
m( ¯ β
u∗m))”
and (as we can increase q
0ξ0) (⊕
ξq00,m∗ξ0
) q
ξ00decides the values of F
˜
m( ¯ β
uξ0m) and F
˜
m( ¯ β
u∗m) for all m ≤ m
∗.
Note that the condition (⊗
ξq00,m∗ξ0
) means that there are NO m ≥ m
∗, l
0, l
1< ω with γ
m¯βuξ0m,l0
6= γ
m¯βu∗m,l1
and the three conditions q
0ξ0, p
m¯βuξ0m,l0
and p
mβ¯u∗m,l1have a common upper bound in C
χ(remember the choice of the p
nα,l¯’s and γ
α,ln¯’s). Similarly, the condition (⊕
ξq00,m∗ξ0
) means there are
NO m ≤ m
∗, l
0, l
1< ω with either γ
βm¯uξ0m,l06= γ
m¯βuξ0m,l1
and both q
ξ00and p
mβ¯uξ0m,l0, and q
ξ00and p
mβ¯uξ0m,l1are compatible in C
χ, or γ
βm¯u∗m,l06=
γ
βm¯u∗m,l1and both q
0ξ0and p
mβ¯u∗m,l0, and q
0ξ0and p
mβ¯u∗m,l1are compatible in C
χ.
Consequently, the condition q
ξ∗0:= q
0ξ0N
w0∪wξ0has both properties (⊗
ξq0∗,m∗ξ0
) and (⊕
ξq0∗,m∗ξ0
) (and it is stronger than both q and q
ξ0).
Now, for 0 < ξ < ω
1let
q
ξ∗:= π
w0∪wξ,w0∪wξ0(q
ξ∗0) ∈ N
w0∪wξ. Then (for ξ ∈ (0, ω
1)) the condition q
ξ∗is stronger than
both q = π
w0∪wξ,w0∪wξ0(q) and q
ξ= π
w0∪wξ,w0∪wξ0(q
ξ0) and it has the properties (⊗
ξ,mq∗ ∗ξ
) and (⊕
ξ,mq∗ ∗ξ
). Moreover for all ξ
1, ξ
2the conditions q
ξ∗1, q
∗ξ2are compatible. [Why? By the definition of Cohen forcing, and π
w0∪wξ2,w0∪wξ1(q
ξ∗1) = q
ξ∗2(chasing arrows) and π
w0∪wξ2,w0∪wξ1is the identity on N
w0∪wξ2∩ N
w0∪wξ1
= N
(w0∪wξ2)∩(w0∪wξ1)(see clauses (e), (f), (h) above).]
Claim 1.1.3. For each ξ
1, ξ
2∈ (0, ω
1) the condition q
∗ξ1∪ q
∗ξ2
forces in C
χthat
(∀m < ω)(F
˜
m( ¯ β
uξ1m) = F
˜
m( ¯ β
uξ2m)).
P r o o f. If m ≥ m
∗then, by (⊗
ξq1∗,m∗ξ1
) and (⊗
ξq2∗,m∗ξ2
) (passing through F ˜ ( ¯ β
u∗m)), we get
q
∗ξ1∪ q
ξ∗2Cχ
“F
˜
m( ¯ β
uξ1m) = F
˜
m( ¯ β
uξ2m)”.
If m < m
∗then we use (⊕
ξq1∗,m∗ξ1
) and (⊕
ξq1∗,m∗ξ2
) and the isomorphism: the values assigned by q
ξ∗1, q
∗ξ2to F
˜
m( ¯ β
uξ1m) and F
˜
m( ¯ β
uξ2m) have to be equal (remember κ ⊆ N
∅, so the isomorphism is the identity on κ).
Look at the conditions
q
ξ1,ξ2:= q
ξ∗1N
wξ1∪ q
ξ∗2N
wξ2∈ N
wξ1∪wξ2. It should be clear that for each ξ
1, ξ
2∈ (0, ω
1),
q
ξ1,ξ2Cχ
“(∀m < ω)(F
˜
m( ¯ β
uξ1m) = F
˜
m( ¯ β
uξ2m))”.
Now choose ξ ∈ (0, ω
1) so large that
dom(p
∗) ∩ (N
wξ\ N
w0) = ∅
(possible as dom(p
∗) is finite, use (e)). Take any 0 < ξ
1< ξ
2< ω
1and put
q
∗:= π
w0∪wξ,wξ1∪wξ2(q
ξ1,ξ2).
(Note: π
w0,wξ1⊆ π
w0∪wξ,wξ1∪wξ2
and π
wξ,wξ2⊆ π
w0∪wξ,wξ1∪wξ2
.) By the isomorphism we get
q
∗Cχ
“(∀m < ω)(F
˜
m( ¯ β
uξm) = F
˜
m( ¯ β
u[ ¯ξ]m))”.
Now look back:
q
∗ξ1≥ q
ξ1= π
w0∪wξ1,w0∪wξ0(q
ξ0) = π
wξ1,wξ0(q
ξ0)
= π
wξ1,wξ0(π
wξ0,w0(q)) = π
wξ1,w0(q) and hence
q
ξ∗1N
wξ1≥ π
wξ1,w0(q) and thus
q
∗N
w0≥ π
w0,wξ1
(q
ξ∗1N
wξ1) ≥ q = p
∗N
w0.
Consequently, by the choice of ξ, the conditions q
∗and p
∗are compatible (remember the definition of q
ξ1,ξ2and q
∗). Now use ( ) to conclude that
q
∗∪ p
∗Cχ
“(∀m < ω)(F
˜
m( ¯ β
u∗m) = F
˜
m( ¯ β
u[ ¯ξ]m) = F
˜
m( ¯ β
uξm))”, which implies that q
∗∪ p
∗Cχ
“ ¯ ξ
_hξi is (k + 1)-strange”, a contradiction.
Remark 1.2. About the proof of 1.1:
(1) No harm is done by forgetting 0 and replacing it by ξ
1, ξ
2.
(2) A small modification of the proof shows that in V
Cχ: If F
n:
nλ → κ (n ∈ ω) are such that
(∀η, ν ∈
ωλ)[(∀
∞n)(η(n) = ν(n)) ⇒ (∀
∞n)(F
n(η n) = F
n(ν n))]
then there are infinite sets X
n⊆ λ (for n < ω) such that (∀n < ω)
∀ν, η ∈ Y
l<n
X
l(F
n(ν) = F
n(η)).
Say we shall have X
n= {γ
n,i: i < ω}. Starting we have γ
0∗, . . . , γ
n∗, . . . In the proof at stage n we have determined γ
l,i(l, i < n) and p ∈ G, p ∈ N
{γl,i:l,i<ω}∪{γn∗,γn+1∗ ,...}. For n = 0, 1, 2 as before. For n + 1 > 2 first γ
0,n, . . . , γ
n−1,nare easy by transitivity of equalities. Then find γ
n,0, γ
n,1as before, and then again duplicate.
(3) In the proof it is enough to use {β
ω·n+l: n < ω, l < ω}. Hence, by 1.2 of [Sh 481] it is enough to assume λ → (ω
3)
<ω2κ. This condition is compatible with V = L.
(4) We can use only λ → (ω
2)
<ω2κ.
Definition 1.3. (1) For a sequence ¯ λ = hλ
n: n < ωi of cardinals we define the property ( ~)
λ¯:
( ~)
¯λfor every model M of a countable language with universe sup
n∈ωλ
nand Skolem functions (for simplicity) there is a sequence hX
n:
n < ωi such that
(a) X
n∈ [λ
n]
λn(actually X
n∈ [λ
n]
ω1suffices)
(b) for every n < ω and ¯ α = hα
l: l ∈ [n + 1, ω)i ∈ Q
l≥n+1
X
l, letting (for ξ ∈ X
n)
M
αξ¯= Sk [
l<n
X
l∪ {ξ} ∪ {α
l: l ∈ [n + 1, ω)} we have:
(L) the sequence hM
αξ¯: ξ ∈ X
ni forms a ∆-system with heart N
α¯and its elements are pairwise isomorphic over the heart N
α¯.
(2) For a cardinal λ the condition ( ~)
λis:
( ~)
λthere exists a sequence ¯ λ = hλ
n: n < ωi such that P
n<ω
λ
n= λ and the condition ( ~)
λ¯holds true.
In [Sh 76] a condition (∗)
λ, weaker than ( ~)
λ, was considered. Now, [Sh 124] continues [Sh 76] to get stronger indiscernibility. But by the same proof (using ω-measurable) one can show the consistency of ( ~)
ℵω+ GCH.
Note that to carry out the proof of 1.1 we need even less than ( ~)
λ: the S
l<n
X
l(in (b) of 1.3) is much more than needed; it suffices to have ¯ β
0∪ ¯ β
1where ¯ β
0, ¯ β
1∈ Q
l<n
X
l.
Conclusion 1.4. It is consistent that 2
ℵ0= ℵ
ω+1and ^
n<ω
¬KL(ℵ
ω, ℵ
n) so ¬KL(2
ℵ0, 2).
Remark 1.5. Koepke [Ko84] continues [Sh 76] to get equiconsistency.
His refinement of [Sh 76] (for the upper bound) works below too.
2. The positive result. For an algebra M on λ and a set X ⊆ λ the closure of X under functions of M is denoted by cl
M(X). Before proving our result (2.6) we remind the reader of some definitions and propositions.
Proposition 2.1. For an algebra M on λ the following conditions are equivalent :
( F)
0Mfor each sequence hα
n: n ∈ ωi ⊆ λ we have
(∀
∞n)(α
n∈ cl
M({α
k: n < k < ω})), ( F)
1Mthere is no sequence hA
n: n ∈ ωi ⊆ [λ]
ℵ0such that
(∀n ∈ ω)(cl
M(A
n+1) cl
M(A
n)),
( F)
2M(∀A ∈ [λ]
ℵ0)(∃B ∈ [A]
ℵ0)(∀C ∈ [B]
ℵ0)(cl
M(B) = cl
M(C)).
Definition 2.2. We say that a cardinal λ has the (F)-property for κ
(and then we write Pr
F(λ, κ)) if there is an algebra M on λ with vocabulary
of cardinality ≤ κ satisfying one (equivalently: all) of the conditions ( F)
iM(i < 3) of 2.1. If κ = ℵ
0we may omit it.
Remember
Proposition 2.3. If V
0⊆ V
1are universes of set theory and V
1|=
¬Pr
F(λ) then V
0|= ¬Pr
F(λ).
P r o o f. By absoluteness of the existence of an ω-branch of a tree.
Remark 2.4. The property ¬Pr
F(λ) is a kind of large cardinal property.
It was clarified in L (remember that it is inherited from V to L) by Silver [Si70] to be equiconsistent with “there is a beautiful cardinal” (terminol- ogy of 2.3 of [Sh 110]), another partition property inherited by L. More in [Sh 513].
Proposition 2.5. For each n ∈ ω, Pr
F(ℵ
n).
P r o o f. This was done in [Sh:b, Chapter XIII], see [Sh:g, Chapter VII]
too, and probably earlier by Silver. However, for the sake of completeness we will give the proof.
First note that clearly Pr
F(ℵ
0) and thus we have to deal with the case when n > 0. Let f, g : ℵ
n→ ℵ
nbe two functions such that if m < n, α ∈ [ℵ
m, ℵ
m+1) then f (α, ·) α : α
1-1→ ℵ
m, g(α, ·) ℵ
m: ℵ
m1-1
→ α are functions inverse to each other.
Let M be the following algebra on ℵ
n: M = (ℵ
n, f, g, m)
m∈ω.
We want to check the condition ( F)
1M: assume that a sequence hA
k: k < ωi
⊆ [ℵ
n]
ℵ0is such that for each k < ω,
cl
M(A
k+1) cl
M(A
k).
For each m < n, the sequence hsup(cl
M(A
k) ∩ ℵ
m+1) : k < ωi is non- increasing and therefore it is eventually constant. Consequently, we find k
∗such that
(∀m < n)(sup(cl
M(A
k∗+1) ∩ ℵ
m+1) = sup(cl
M(A
k∗) ∩ ℵ
m+1)).
By the choice of hA
k: k < ωi we have cl
M(A
k∗+1) cl
M(A
k∗). Let α
0:= min(cl
M(A
k∗) \ cl
M(A
k∗+1)).
As the model M contains individual constants m (for m ∈ ω) we know that ℵ
0⊆ cl
M(∅) and hence ℵ
0≤ α
0. Let m < n be such that ℵ
m≤ α
0< ℵ
m+1. By the choice of k
∗we find β ∈ cl
M(A
k∗+1) ∩ ℵ
m+1such that α
0≤ β. Then necessarily α
0< β. Look at f (β, α
0): we know that α
0, β ∈ cl
M(A
k∗) and therefore f (β, α
0) ∈ cl
M(A
k∗) ∩ ℵ
mand f (β, α
0) < α
0. The minimality of α
0implies that f (β, α
0) ∈ cl
M(A
k∗+1) and hence
α
0= g(β, f (β, α
0)) ∈ cl
M(A
k∗+1),
a contradiction.
Explanation. Better think of the proof below from the end. Let ¯ α = hα
n: n < ωi ∈
ωλ. So for some n(∗), n(∗) ≤ n < ω ⇒ α
n∈ cl
M(α
l: l > n).
So for some m
n> n, {α
n(∗), . . . , α
n−1} ⊆ cl
M(α
n, . . . , α
mm−1) and (∀l < n(∗))(α
l∈ cl
M(α
k: k > n(∗)) ⇒ α
l∈ cl
M(α
k: k ∈ [n, m
n))).
Let w = {l < n(∗) : α
l∈ cl
M(α
n: n ≥ n(∗)). It is natural to aim at:
(∗) for n large enough (say n > m
n(∗)), F
n(hα
l: l < ni) depends just on {α
l: l ∈ [n(∗), n) or l ∈ w} and hF
m( ¯ α m) : m ≥ ni codes
¯
α (w ∪ [n(∗), ω)).
Of course, we are given an n and we do not know how to compute the real n(∗), but we can approximate. Then we look at a late enough end segment where we compute down.
Theorem 2.6. Assume that λ ≤ 2
ℵ0is such that Pr
F(λ) holds. Then KL(λ, ω) (and hence KL(λ, 2)).
P r o o f. We have to construct functions F
n:
nλ → ω witnessing KL(λ, ω).
For this we will introduce functions k and l such that for ¯ α ∈
nλ the value of k( ¯ α) will say which initial segment of ¯ α will be irrelevant for F
n( ¯ α) and l( ¯ α) will be such that (under certain circumstances) elements α
i(for k( ¯ α) ≤ i < l( ¯ α)) will be encoded by hα
j: j ∈ [l( ¯ α), n)i.
Fix a sequence hη
α: α < λi ⊆
ω2 with no repetitions.
Let M be an algebra on λ such that ( F)
0Mholds true. We may assume that there are no individual constants in M (so cl
M(∅) = ∅).
Let hτ
ln(x
0, . . . , x
n−1) : l < ωi list all n-place terms of the language of the algebra M (and τ
01(x) is x) when 0 < n < ω. For ¯ α ∈
ω≥λ (with α
jthe jth element in ¯ α) let
u( ¯ α) = {l < lg( ¯ α) : α
l6∈ cl
M( ¯ α (l, lg(¯ α)))} ∪ {0}
and for l 6∈ u( ¯ α), l < lg( ¯ α) let
f
l( ¯ α) = min{j : α
l∈ cl
M( ¯ α (l, j))},
g
l( ¯ α) = min{i : α
l= τ
ifl( ¯α)−l−1( ¯ α (l, f
l( ¯ α)))}.
For ¯ α ∈
nλ (1 < n < ω) put
k
1( ¯ α) = min((u( ¯ α (n − 1)) \ u(¯ α)) ∪ {n − 1}), k
0( ¯ α) = max(u( ¯ α) ∩ k
1( ¯ α)).
Note that if (n > 1 and) ¯ α ∈
nλ then n − 1 ∈ u( ¯ α) (as cl
M(∅) = ∅) and k
1( ¯ α) > 0 (as always 0 ∈ u( ¯ β)) and k
0( ¯ α) is well defined (as 0 ∈ u( ¯ α)∩k
1( ¯ α)) and k
0( ¯ α) < k
1( ¯ α) < n. Moreover, for all l ∈ (k
0( ¯ α), k
1( ¯ α)) we have α
l6∈ u( ¯ α) by the choice of k
0( ¯ α), hence α
l6∈ u( ¯ α (n − 1)) by the choice of k
1( ¯ α) and thus α
l∈ cl
M( ¯ α (l, n − 1)). Now, for ¯ α ∈
ω>λ, lg( ¯ α) > 1 we define
l( ¯ α) = max{j ≤ k
1( ¯ α) : j > k
0( ¯ α) ⇒ (∀i ∈ (k
0( ¯ α), j))(g
i( ¯ α) ≤ lg( ¯ α))},
m( ¯ α) = max{j ≤ l( ¯ α) : j > max{1, k
0( ¯ α)} ⇒ k
0( ¯ α j) = k
0( ¯ α)}, k( ¯ α) = l( ¯ α m(¯ α)) (if m( ¯ α) ≤ 1 then put k( ¯ α) = −1).
Clearly k( ¯ α) < m( ¯ α) ≤ l( ¯ α) ≤ k
1( ¯ α) < lg( ¯ α).
Claim 2.6.1. For each ¯ α ∈
ωλ, the set u( ¯ α) is finite and : (1) The sequence hk
1( ¯ α n) : n < ωi diverges to ∞.
(2) The sequence hk
0( ¯ α n) : n < ω & k
0( ¯ α n) 6= max u(¯ α)i, if infinite, diverges to ∞. There are infinitely many n < ω with k
0( ¯ α n) = max u(¯ α).
(3) The sequence hl( ¯ α n) : n < ωi diverges to ∞.
(4) The sequences hm( ¯ α n) : n < ωi and hk(¯ α n) : n < ωi diverge to ∞.
P r o o f. Let ¯ α = hα
n: n < ωi ∈
ωλ. By the property ( F)
0Mwe find n
∗< ω such that u( ¯ α) ⊆ n
∗. Fix n
0> n
∗and define
n
1= max{f
n( ¯ α) + g
n( ¯ α) + 2 : n ∈ (n
0+ 1) \ u( ¯ α)}
(so as cl
M(∅) = ∅ we have n
1≥ f
n0( ¯ α)+2 > n
0+3 and for l ∈ (n
0+1)\u( ¯ α), α
l∈ cl
M(α
l+1, . . . , α
n1−1) is witnessed by τ
gfl( ¯α)−l−1l( ¯α)
(α
l+1, . . . , α
fl( ¯α)−1) with f
l( ¯ α), g
l( ¯ α) < n
1− 1).
(1) Note that u( ¯ α n) ∩ (n
0+ 1) = u( ¯ α) for all n ≥ n
1− 1 and hence for n ≥ n
1,
u( ¯ α n) ∩ (n
0+ 1) = u( ¯ α (n − 1)) ∩ (n
0+ 1).
Consequently, for all n ≥ n
1we have k
1( ¯ α n) > n
0. As we could have chosen n
0arbitrarily large we may conclude that lim
n→∞k
1( ¯ α n) = ∞.
(2) Note that for all n ≥ n
1,
either k
0( ¯ α n) = max(u(¯ α)) or k
0( ¯ α n) > n
0. Hence, by the arbitrariness of n
0, we get the first part of (2).
Let l
∗= min(u( ¯ α n
1)\u( ¯ α)) (note that n
1−1 ∈ u( ¯ α n
1)\u( ¯ α)). Clearly l
∗> n
0and α
l∗6∈ u( ¯ α). Consider n = f
l∗( ¯ α) (so l
∗≤ n − 2, n
1≤ n − 1).
Then l
∗∈ u( ¯ α (n − 1)) \ u(¯ α n). As
l
∗∩ u( ¯ α n
1) = l
∗∩ u( ¯ α n − 1) = u(¯ α) (remember the choice of l
∗) we conclude that
l
∗= k
1( ¯ α n) and k
0( ¯ α n) = max u(¯ α).
Now, since n
0was arbitrarily large, we find that for infinitely many n, k
0( ¯ α n) = max u(¯ α).
(3) Suppose that n ≥ n
1. Then we know that k
1( ¯ α n) > n
0and either k
0( ¯ α n) = max u(¯ α) or k
0( ¯ α n) > n
0(see above). If the first possibility takes place then, as n ≥ n
1, we may use j = n
0+ 1 to witness that l( ¯ α n) >
n
0(remember the choice of n
1). If k
0( ¯ α n) > n
0then clearly l( ¯ α n) > n
0.
As n
0could be arbitrarily large we are done.
(4) Suppose we are given m
0< ω. Take m
1> m
0such that for all n ≥ m
1,
either k
0( ¯ α n) = max u(¯ α) or k
0( ¯ α n) > m
0(possible by (2)) and then choose m
2> m
1such that k
0( ¯ α m
2) = max u( ¯ α) (by (2)). Due to (3) we find m
3> m
2such that for all n ≥ m
3, l( ¯ α n) > m
2. Now suppose that n ≥ m
3. If k
0( ¯ α n) = max u(¯ α) then, as l( ¯ α n) > m
2, we get m( ¯ α n) ≥ m
2> m
0. Otherwise k
0( ¯ α n) > m
0(as n > m
1) and hence m( ¯ α n) > m
0. This shows that lim
n→∞m( ¯ α n) = ∞. Now, immediately by the definition of k and (3) above we conclude that lim
n→∞k( ¯ α n) = ∞.
Claim 2.6.2. If ¯ α
1, ¯ α
2∈
ωλ are such that (∀
∞n)(α
1n= α
2n) then (∀
∞n)(l( ¯ α
1n) = l(¯ α
2n) & m(¯ α
1n) = m(¯ α
2n) & k(¯ α
1n) = k(¯ α
2n)).
P r o o f. Let n
0be greater than max(u( ¯ α
1) ∪ u( ¯ α
2)) and such that
¯
α
1[n
0, ω) = ¯ α
2[n
0, ω).
For k = 1, 2, 3 define n
kby
n
k+1= max{f
n( ¯ α
i) + g
n( ¯ α
i) + 2 : n ∈ (n
k+ 1) \ u( ¯ α
i), i < 2}.
As in the proof of 2.6.1, for i = 1, 2 and j < 3 we have:
(⊗
1) (∀n ≥ n
j+1)(k
0( ¯ α
in) = max u(¯ α
i) or k
0( ¯ α
in) > n
j),
(⊗
2) (∀n ≥ n
j+1)(k
1( ¯ α
in) > n
j& l( ¯ α
in) > n
j& m( ¯ α
in) > n
j&
h( ¯ α
in) > n
j),
(⊗
3) (∃n
0∈ (n
1, n
2))(k
0( ¯ α
1n
0) = max u( ¯ α
1) & k
0( ¯ α
2n
0) = max u( ¯ α
2)) (for (⊗
3) repeat arguments from 2.6.1(2) and use the fact that ¯ α
1[n
0, ω) =
¯
α
2[n
0, ω)). Clearly
(⊗
4) (∀n > n
0)(u( ¯ α
1n) \ n
0= u( ¯ α
2n) \ n
0).
Hence, applying (⊗
4) + (⊗
2) + the definition of k
1(−), we conclude that (⊗
5) (∀n ≥ n
1)(k
1( ¯ α
1n) = k
1( ¯ α
2n)),
and then applying (⊗
4) + (⊗
2) + (⊗
5) + the definition of k
0(−), we get (⊗
6) for all n ≥ n
1: either k
0( ¯ α
1n) = max u(¯ α
1) and k
0( ¯ α
2n) =
max u( ¯ α
2), or k
0( ¯ α
1n) = k
0( ¯ α
2n).
Since
(∀n ≥ n
0)(f
n( ¯ α
1) = f
n( ¯ α
2) & g
n( ¯ α
1) = g
n( ¯ α
2))
and by (⊗
2)+(⊗
5)+the choice of n
0+the definition of l(−), we get (compare the proof of 2.6.1)
(⊗
7) (∀n ≥ n
1)(l( ¯ α
1n) = l(¯ α
2n)) and by (⊗
2) + (⊗
7) + (⊗
6) + the definition of m(−),
(∀n ≥ n
3)(m( ¯ α
1n) = m(¯ α
2n) ≥ n
2).
Moreover, now we easily get
(∀n ≥ n
3)(k( ¯ α
1n) = k(¯ α
2n)).
For integers n
0≤ n
1≤ n
2we define functions F
n00,n1,n2:
n2λ → H(ℵ
0) by letting F
n00,n1,n2(α
0, . . . , α
n2−1) (for hα
0, . . . , α
n2−1i ∈
n2λ) be the se- quence consisting of:
(a) hn
0, n
1, n
2i,
(b) the set T
n1,n2of all terms τ
lnsuch that n ≤ n
2− n
1and either l ≤ n
2(we will call it the simple case) or τ
lnis a composition of depth at most n
2of such terms,
(c) hη
αn
2, n, l, hi
0, . . . , i
n−1ii for n ≤ n
2− n
1, i
0, . . . , i
n−1∈ [n
1, n
2) and l such that τ
ln∈ T
n1,n2and α = τ
ln(α
i0, . . . , α
in−1),
(d) hn, l, hi
0, . . . , i
n−1i, ii for n ≤ n
2− n
1, i
0, . . . , i
n−1∈ [n
1, n
2), i ∈ [n
0, n
1) and l such that τ
ln∈ T
n1,n2and α
i= τ
ln(α
i0, . . . , α
in−1),
(e) equalities among appropriate terms, i.e. all tuples hn
0, l
0, n
00, l
00, hi
00, . . . , i
0n0−1i, hi
000, . . . , i
00n00−1ii
such that n
1≤ i
00< . . . < i
0n0−1< n
2, n
1≤ i
000< . . . < i
00n00−1< n
2, n
0, n
00≤ n
2− n
1, l
0, l
00are such that τ
ln00, τ
ln0000∈ T
n1,n2and
τ
ln00(α
i00, . . . , α
i0n0 −1
) = τ
ln0000(α
i000, . . . , α
i00n00 −1
).
(Note that the value of F
n00,n1,n2( ¯ α) does not depend on ¯ α n
0.) Finally we define functions F
n:
nλ → H(ℵ
0) (for 1 < n < ω) by:
if ¯ α ∈
nλ then F
n( ¯ α) = F
k( ¯0α),l( ¯α),n( ¯ α).
As H(ℵ
0) is countable we may think that these functions are into ω. We are going to show that they witness KL(λ, ω).
Claim 2.6.3. If ¯ α
1, ¯ α
2∈
ωλ are such that (∀
∞n)(α
1n= α
2n) then (∀
∞n)(F
n( ¯ α
1n) = F
n( ¯ α
2n)).
P r o o f. Take m
0< ω such that for all n ∈ [m
0, ω) we have α
n1= α
2n, l( ¯ α
1n) = l(¯ α
2n), k( ¯ α
1n) = k(¯ α
2n) (possible by 2.6.2). Let m
1> m
0be such that for all n ≥ m
1,
k( ¯ α
1n) = k(¯ α
2n) > m
0(use 2.6.1). Then, for n ≥ m
1, i = 1, 2 we have
F
n( ¯ α
in) = F
k( ¯0αin),l( ¯αin),n( ¯ α
in) = F
k( ¯0α1n),l( ¯α1n),n( ¯ α
in).
Since the value of F
n00,n1,n2( ¯ β) does not depend on ¯ β n
0and the sequences
¯
α
1n, ¯ α
2n agree on [m
0, ω), we get
F
k( ¯0α1n),l( ¯α1n),n( ¯ α
1n) = F
k( ¯0α1n),l( ¯α1n),n( ¯ α
2n) = F
k( ¯0α2n),l( ¯α2n),n( ¯ α
2n),
and hence
(∀n ≥ m
1)(F
n( ¯ α
1n) = F
n( ¯ α
2n)).
Claim 2.6.4. If ¯ α
1, ¯ α
2∈
ωλ and (∀
∞n)(F
n( ¯ α
1n) = F
n( ¯ α
2n)) then (∀
∞n)(α
1n= α
2n).
P r o o f. Take n
0< ω such that
u( ¯ α
1) ∪ u( ¯ α
2) ⊆ n
0and (∀n ≥ n
0)(F
n( ¯ α
1n) = F
n( ¯ α
2n)).
Then for all n ≥ n
0we have (by clause (a) of the definition of F
n00,n1,n2) l( ¯ α
1n) = l(¯ α
2n) and k(¯ α
1n) = k(¯ α
2n).
Further, let n
1> n
0be such that for all n ≥ n
1, k( ¯ α
1n) > n
0and k
0( ¯ α
1n
1) = max u( ¯ α
1) (exists by 2.6.1) and choose n
2> n
1such that n ≥ n
2implies m( ¯ α
1n) > n
2.
We are going to show that α
1n= α
2nfor all n > n
1. Assume not. Then we have n > n
1with α
1n6= α
2nand thus η
α1n6= η
α2n. Take n
0> n such that η
α1nn
06= η
α2n