The P-Ideal Dichotomy and the Souslin Hypothesis

John Krueger is visiting Toronto these days, and in a conversation today, we asked ourselves how do one prove the Abraham-Todorcevic theorem that PID implies SH. Namely, that the next statement implies that there are no Souslin trees:

Definition. The P-ideal Dichotomy asserts that for every uncountable set Z, and every P-Ideal I over [Z]0, exactly one of the following holds:

  • There exists an uncountable AZ such that [A]0I;
  • There exists a sequence Znn<ω such that n<ωZn=Z, and [Zn]0I= for all n<ω.

It turns out that the proof is quite easy, so let us give the full details.

Suppose that (T,<T) is a Souslin tree. Denote x:={yTy<Tx}. Then, define I:={X[T]0xT(Xx) is finite }.

It is clear that I is an ideal. Let us show that moreover it is a P-Ideal, namely, that I is closed under taking pseduo-unions of countable sets.

Lemma. For every {Xnn<ω}I, there exists some XI such that XnX is finite for all n<ω.
Proof. As I is downward closed, we shall avoid trivialities and assume that {Xnn<ω} are mutually disjoint. Fix a bijection ψ:ωn<ωXn . Let δ:=sup{ht(x)xn<ωXn}. Evidently, δ<ω1. Then, let S:={xTht(x)δ+1}, and note that S is countable.  Next, for all xS, define fx:ωω by letting for all n<ω: fx(n):=min{m<ωXnxψm}.

As S is countable, we may pick a function f:ωω such that {n<ωfx(n)>f(n)} is finite, for all xS. Now, let X:={Xnψ[f(n)]n<ω}. Then, for every n<ω, XnXψ[f(n)] is finite. Assume towards a contradiction that XI. Then, in particular, there exists some xT such that Xx is infinite. Since ht(y)<δ+1 for all yX, there exists then some z of height δ+1 which is compatible with x and Xz is infinite. Thus, without loss of generality, we may assume that xS.
As fxf while Xnx is finite for all n<ω, let us pick a large enough n<ω such that fx(n)<f(n) and (Xx)Xn. Pick zXxXn, and let m:=ψ1(z). Since zXnx, we get that fx(n)>m. In particular, f(n)>m, and so by definition of X, we get that ψ(m)X, contradicting the choice of z=ψ(m) is in X. ◻

So I is a P-ideal, and we may invoke the dichotomy.

Alternative I. There exists an uncountable AT such that [A]0I. Let us show that A contains an uncountable antichain.
Proof. Since A is uncountable, while the levels of the tree are countable, we can fix an injection φ:ω1A such that ht(φ(α))α<ω1 is a strictly-increasing sequence.
Define a coloring f:[ω1]2{0,1} by letting f(α,β)=0 iff φ(α) is <T-incomparable with φ(b). Then, by the Dushnik-Miller theorem ω1(ω1,ω+1)2, one of the following holds.

  • There exists an uncountable Bω1 such that f[B]2={0}. Clearly, φ[B] is an uncountable antichain.
  • There exists a subset Bω1 of order-type ω+1 such that f[B]2={1}. Let β:=max(B). Then φ[Bβ][A]0I. In particular, (φ[Bβ]φ(β)) is finite, contradicting the fact that φ[Bβ]φ(β). So, this case is void. ◻

Alternative II. There exists a sequence Znn<ω such that n<ωZn=T, and [Zn]0I= for all n<ω. Let us show that T contains an uncountable chain.
Proof. Fix n<ω such that Zn is uncountable, and define a coloring f:[Zn]2{0,1} by letting f(x,y)=0 iff x and y are <T-comparable. Then, by the Dushnik-Miller theorem ω1(ω1,ω)2, one of the following holds.

  • There exists an uncountable BZn such that f[B]2={0}. Clearly, such a B is an uncountable chain.
  • There exists some A[Zn]0 such that f[A]2={1}. Then, by [Zn]0I=, we know that AI, and there exists some xT such that Ax is infinite. In particular, we can find distinct y,zA such that y<Tx and z<Tx. Recalling that (T,<T) is a tree, we conclude that y,z are <T-comparable, contradicting the fact that f(y,z)=1. So this case is void. ◻

For completeness, we mention that James Hirschorn proved that the strong dichotomy principle ()c entails that all Aronszajn trees are special.

This entry was posted in Blog, Expository and tagged , . Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *