Forcing Axioms and the Definabilty of the Nonstationary Ideal on \omega_1

Assume for a contradiction that there is a \Sigma_1- formula \varphi and a set A \subset \omega_1 such that \forall T \in P(\omega_1) (T is stationary iff \varphi(T,A)). Let \delta be our Woodin cardinal.

Let

and let g be \mathbb{P}_{<\delta}-generic over V where \mathbb{P}_{<\delta} is the associated full stationary tower. Let us pick a generic filter g which contains S_0, which is possible by the stationarity of S_0. As usual we can form the generic elementary embedding in the universe V[g]:

for a transitive inner model M of V[g]. Membership in the generic filter g for the stationary tower forcing can be characterized using j, namely we have that

thus S_0 \in g yields j" H(\omega_2)^V \in j(S_0). In particular j"H(\omega_2)^V is transitive, and as H(\omega_2)^V is the transitive collapse of j"H(\omega_2)^V, we obtain that j"H(\omega_2)^V=H(\omega_2)^V and that the critical point crit(j) of the elementary embedding j must be \ge \omega_2^V. As H(\omega_2)^V \in j(S_0), |H(\omega_2)^V|=\aleph_1 in M, so crit(j)=\omega_2^V.

We have P(\omega_1) \cap V \in M. Indeed by \mathsf{BMM}, 2^{\aleph_1}=\aleph_2, so let f: \omega_2 \rightarrow P(\omega_1) \cap V be a bijection with f \in V. Then j(f) \upharpoonright \omega_2^V =f and j(f) \in M so P(\omega_1) \cap V = ran(f) \in M.

It is a theorem of A. D. Taylor (see [Ta]) that \mathsf{MA_{\omega_1}} implies that \hbox{NS}_{\omega_1} is not \omega_1-dense. As P(\omega_1) \cap V has size |\omega_2^V|=\aleph_1 in M, there is a stationary set D \subset \omega_1 in M such that T \backslash D is stationary for every stationary T \in P(\omega_1) \cap V. By Theorem 2.5.8 of [La2], M^{<\delta} \cap V[g] \subset M, which gives us that V[g] is a stationary set preserving extension of V. Further it is still true in V[g] that D is stationary and T \backslash D is stationary for all T \notin \hbox{NS}_{\omega_1}^{V}.

In the next step we use the ordinary club shooting forcing \mathbb{P}_{\omega_1 \backslash D} over V[g] to shoot a club through the complement of D. The forcing will not destroy any stationary subsets from V \cap P(\omega_1):

So V[g,h] is a stationary set preserving extension of V. But now by our hypothesis and by elementarity of j:V \rightarrow M we get that

and hence

as \varphi is \Sigma_1, and consequentially

In V[g,h] the set D is nonstationary, thus

This statement is \Sigma_1 with parameter A \subset \omega_1 in the language of set theory, and as \mathsf{BMM} is assumed to hold true in V we conclude that

which is a contradiction.

Let V=L(\mathbb{R})[g], where g is \mathbb{P}_{\text{max}}-generic over L(\mathbb{R}). Suppose for a contradiction that there is a \mathbb{P}_{\text{max}}-name for a subset of \omega_1, \dot{A}, a \Sigma_1-formula \varphi(-,\dot{A}) and a \mathbb{P}_{\text{max}}-condition p= (p,J,b) such that

By [LLS], the conclusion of the theorem is true for every A \in P(\omega_1)^{L(\mathbb{R})}, so we may assume that \dot{A}^g=A \in P(\omega_1)^{V} \backslash L(\mathbb{R}). It is well-known that any such A \subset \omega_1 gives rise to a \mathbb{P}_{\text{max}}-generic filter g_A:=\{ (p,I,a) \in \mathbb{P}_{\text{max}} \, : \, p generically iterates to some p_{\omega_1}=(p_{\omega_1}, \hbox{NS}_{\omega_1}^V, A) \} and L(\mathbb{R})[g_A]=L(\mathbb{R})[g]. Hence we assume without loss of generality that the \dot{A} above is the name for such a distinguished subset of \omega_1 which gives rise to the \mathbb{P}_{\text{max}}-generic filter.

Now let q be as in the last Claim and for g our \mathbb{P}_{\text{max}}-generic over L(\mathbb{R}) we let q \in g which we can assume by density. By the choice of p and q<p, q \in g we have that

Let j: q \rightarrow q^{\ast}=(q^{\ast},K,A) be the unique iteration of q of length \omega_1 given by g which sends c to A. By standard \mathbb{P}_{\text{max}}-arguments, K=NS_{\omega_1}^V \cap q^{\ast}. By elementarity

and as \varphi is \Sigma_1,

In other words j(T) is stationary in V and for all S \in j((J^+)^{p^{\ast}}), S \backslash j(T) is stationary in V. Hence

Now let L denote the ideal generated from \hbox{NS}_{\omega_1}^V \cup \{j(T)\}, i.e. X \in L iff (X \backslash j(T)) \cap C = \emptyset for some club C. It is easy to see that L is a normal ideal and j((J^+)^{p^{\ast}}) \subset L^+, so j(J)=L \cap j(p^{\ast}).

Now we take a countable elementary substructure of (H(\omega_2^V), L, A) and transitively collapse it. This results in a \mathbb{P}_{\text{max}} condition (r, \bar{L},a) (see [Wo], Lemma 3.14) which gives us:

But now if we consider the generic iteration of r given by g,

of length \omega_1, then r^{\ast} \models \varphi(k(T), A) and so V \models \varphi(k(T), A). Yet k(T) is nonstationary in V=L(\mathbb{R})[g]. But r <_{\mathbb{P}_{\text{max}}} p which is a contradiction.

(Sketch, following the proof of [S], Theorem 3.2 ) We will only need to show by induction on j\le \alpha that for any N\in \mathcal T_{\vec{S}}, if j\in N, then:

(*)_N For all i<j, i\in N and for all p\in N\cap P_j and q\in P_i is (N,O_i) generic condition extending p\upharpoonright i, there is an r\in P_j such that r is (N, P_j)-generic condition extending p and r\upharpoonright i=q.

The statement (*)_N is identical to the statement (*) in the proof of [S], Theorem 3.2. It can be checked that the original proof also works here.

((b) \rightarrow (a)) Follow from the definition of the forcing.

((a) \rightarrow (b)) Fix an i and assume without loss of generality that i is even. Write i= \alpha \cdot ({\beta+1}) +2j and suppose for a contradiction that j is not an element of B_{\beta}. By the definition of an \vec S coding forcing, we must have added a club through S_{\alpha (\beta+1)+2j+1} instead. Let \vec T be the sequence \langle T_k \mid k<\kappa \rangle, where T_k=S_{k} if k\neq i and T_i=\eta_i\setminus S_{i}. It follows from Theorem 3.2 again that \vec T is mutually stationary. We will prove that \mathbb{P} is \vec T-preserving to derive a contradiction. Indeed we shall see that \vec T-perservation implies that \eta_i \setminus S_i must remain stationary after forcing with \mathbb{P}, yet \mathbb{P} \Vdash `` S_i contains an \omega-club" which is impossible.

To see that \vec{T} preserving forcings preserve the stationarity of every S_{\eta_i} \in \vec{T}, we only need to note that for any name \dot C of a subset of \eta_i which is unbounded and \omega-closed, and any countable elementary substructure N which contains \dot C and for which \sup (N\cap \eta_i)\in S_{\eta_i}, any (N,\mathbb{P})-generic condition q, q forces \dot C\cap ( S_{\eta_i})\neq \emptyset.

Next we show by induction that each Q_{\beta} is forced to be \vec T-preserving. Work in V[G_\beta]. If \dot Q_{\beta}/G_\beta is proper, then it is also \vec T-preserving. Otherwise, (1) holds. Now \dot{Q}_\beta/G_{\beta} is a countable support product of shooting club forcing. Fix any N\in \mathcal T_T which is a countable substructure of H(\theta)^{V[G_\beta]}. For any p\in N\cap \dot Q_{\beta}, we can construct a countable decreasing sequence of conditions \langle p_i \mid i<\omega \rangle meeting all dense set in N. Define q by setting q(i) to be the closure of \bigcup_{n<\omega} p_n(i) if i\in N and trivial otherwise. Note any nontrivial q(i) is equal to \bigcup_{n<\omega} p_n(i)\cup \{\sup(N\cap \eta_i)\}, where \eta_i=\sup (S_i) is a regular cardinal. As N\in \mathcal T_{T} and no \beta,\alpha,j witness that \beta\cdot(\alpha+1)+2j=i or \beta\cdot(\alpha+1)+2j+1=i, we have \sup(N\cap \eta_{i})\in S_i, whenever q(i) is non-trivial. Hence q is a condition.

Let \epsilon_\alpha=\Sigma_{\beta<\alpha}l(\beta) for any \alpha\le\delta. We define a countable support iteration \mathbb{P}'= \langle P_\alpha',\dot Q_\alpha' \mid \alpha <\epsilon_\delta \rangle. By induction on \alpha< \delta, we give the definition of P'_{\epsilon_\alpha}. We will also verify that P'_{\epsilon_\alpha} is forcing equivalent to P_\alpha and P'_{\epsilon_\alpha} is a \eta_0-\vec S coding.

If \alpha is limit. Then P'_{\epsilon_\alpha} is the countable support limit of P'_{\gamma} for \gamma<\epsilon_\alpha. Hence P'_{\epsilon_\alpha} is the countable support limit of P'_{\epsilon_\gamma} for \gamma<\alpha. As P'_{\epsilon_\gamma} is forcing equivalent to P_{\gamma}. P'_{\epsilon_\alpha} is forcing equivalent to the countable support limit of P_{\gamma} for \gamma<\alpha. Hence P'_{\epsilon_\gamma} is forcing equivalent to P_{\alpha}.

If \alpha=\beta+1. Let \epsilon_\alpha=\epsilon_\beta+l(\alpha). Now \dot{Q}_\beta is a name for a forcing iteration \langle P^\alpha_\gamma,\dot{Q}^\alpha_\gamma \mid \gamma<l(\alpha) \rangle. Now as P_\beta is forcing equivalent to P'_{\epsilon_\beta}. We can view \dot{Q}_\alpha as a P'_{\epsilon_\beta} name. By induction on \gamma< l(\alpha), we can set \dot{Q}'_{\epsilon_\beta+\gamma} be P'_{\epsilon_\beta+\gamma} name of \dot Q^\alpha_\gamma and verify that P'_{\epsilon_\beta+\gamma} is forcing equivalent to P'_{\epsilon_\beta}*P^\alpha_\gamma.

It remains to check that P'_{\epsilon_\alpha} is a \eta_0-\vec S coding. Fix a \beta<\epsilon_\alpha, with \dot{Q}_\beta is not forced to be proper. There is \gamma such that \beta\in [\epsilon_\gamma, \epsilon_{\gamma+1}). \beta' is inaccessible cardinal greater then \eta_ \gamma. Let \beta= \epsilon_\gamma+\beta'. As \epsilon_\gamma< \eta_\gamma and \beta is inaccessible. \beta=\beta' is inaccessible. Now P'_{\beta} is forcing equivalent to P_{\gamma}*P^\gamma_\beta. P_{\gamma} is of size less than \beta and P^\gamma_\beta is forced to be forcing equivalent to a forcing of size less of equal to \beta. Hence P'_{\beta} is also forcing equivalent to a forcing of size less of equal to \beta. The rest of (2) is routine to check.

This follows from the definition. Note that being a club is an absolute property between transitive models.

By Lemma 3.6, for any i, S_i contains a club if there is some stage \alpha such that \dot{Q}_\alpha satisfies (2) in the definition. But \dot{Q}_\alpha forces S codes P(\alpha)^{L[G_\alpha]} in L[G_{\alpha+1}]. Then by Lemma 3.9, S codes P(\alpha)^{L[G_\alpha]} in L[G].

We first choose a sequence \langle S_ \alpha\mid \alpha\in \mathop{\textrm{Lim}}(\delta)\rangle uniformly in \alpha< \delta satisfying

• S_\alpha\subset \alpha

• If \alpha is a regular cardinal, then S_\alpha is stationary co-stationary in \alpha\cap \textrm{Cof}(\omega).

The existence of such a sequence \langle S_ \alpha\mid \alpha\in \mathop{\textrm{Lim}}(\delta)\rangle follows from the fact that \diamondsuit_{\lambda} holds in L for any L-cardinal \lambda and is a routine construction.

Now we define the forcing poset P. The forcing P=\langle P_\alpha, \dot{Q}_\alpha \mid \alpha<\delta \rangle will be a countable support iteration of length \delta. We require the size of each iterand to be smaller than \delta. As a consequence P satisfies \delta-cc. We demand that \dot{Q}_\alpha is trivial unless \alpha is an inaccessible cardinal and P_\alpha\subset L_\alpha. We split into two cases if \alpha is inaccessible.

• (\alpha is Mahlo) We follow an idea from Goldstern-Shelah [GS]. Choose a A\subset \omega_1 and a \Sigma_1 formula \psi(x,y). A and \psi is chosen in a bookkeeping way so that during the whole iteration, each pair (A,\psi) will be dealt with unboundedly many times. Since \delta is reflecting, whenever there is a \alpha-\vec S_{>\alpha} coding forcing poset Q which adds a witness to \psi(x,A), there is such a Q of size less than \delta. In this case, let \dot{Q}_\alpha be the name of Q and use this forcing at stage \alpha. When there is no such Q, we set \dot{Q}_\alpha to be trivial forcing.

• (\alpha is inaccessible but not Mahlo) \dot{Q}_\alpha is trivial unless \vert P_\alpha\vert\le \alpha. Work in L[G_\alpha], where G_\alpha is a L-P_\alpha generic filter. \vert P_\alpha\vert= \alpha is inaccessible. We choose \langle B_\beta\mid \beta <2^\alpha\rangle to be an enumeration of P(\alpha)^{V[G_\alpha]}. Let Q be the forcing

  \prod_{i\in B_\beta}Club(S_{\alpha\cdot(\beta+1)+2i}) \times \prod_{i\notin B_\beta}Club(S_{\alpha\cdot(\beta+1)+2i+1})

  with countable support. Then we force with Q at stage \alpha.

It follows from the definition of P that for any \alpha <\delta, the forcing \dot{Q}_{\alpha} is forced to be a \alpha-\vec S coding forcing. Applying Lemma 3.7, we know that P is a \vec{S}-coding forcing. Moreover, for any \alpha Mahlo, the tail P /P_\alpha is a \alpha-\vec S_{>\alpha} coding.

Now as P is \vec S-preserving, P preserves \omega_1. On the other hand, P is \delta-c.c and P preserves cardinals above \delta. By the definition of the forcing in the inaccessible but not Mahlo case, all cardinals below \delta are collapsed to \omega_1. In summary, \omega_1^{L^P}=\omega_1^L and \omega_2^{L^P}=\delta.

We now present the \Sigma_1 definition of stationary subsets of \omega_1 over (H(\omega_2), \in, \{\omega_1\}). Let \psi(x) describe the following statement: there are objects A and M such that

• (i) A= \langle C_i \mid i \in [\alpha, \beta) \rangle is a sequence with C_i is an \omega-club in \sup (C_i \cap Ord)=\eta_i.

• (ii) All \eta_i are regular cardinals in L. \eta_\alpha=\alpha is inaccessible in L.

• (iii) M is a transitive ZFC^- model with \omega_1=\omega_1^M, ((\alpha^+)^L)^M= (\alpha^+)^L, and x\in M.

• (iv) M\models x \subset \omega_1 is stationary.

• (v) M\models A is a \vec S codes for P(\alpha).

It is routine to check (i), (iii), (iv) and (v) are all \Sigma_1 over \langle H(\omega_2),\in, \{\omega_1\} \rangle. For (ii), as L[G]\models BPFA, we can apply a trick of Todorcevic (cf. Proof of Lemma 4, [To02]) to get \Sigma_1 formulas \psi_0 (x) and \psi_1 (x) such that H(\omega_2)\models \psi_0(\beta) if and only if \beta is a regular cardinal in L and H(\omega_2)\models \psi_1(\beta) if and only if \beta is an inaccessible cardinal in L.

Here \psi_0(x) is the formula describing the existence of a specialization function of the tree T_x, where T_x is derived from the canonical global square sequence in L. It is a consequence of \mathsf{BPFA} that x is uncountable regular in L if and only if such a specialization function exists. Now \psi_1(x) is a formula says \alpha is regular and a limit of regular cardinals in L.

It is now clear that \psi(x) is \Sigma_1 over \langle H(\omega_2),\in, \{\omega_1\} \rangle. What is left is to show that \psi(x) indeed characterizes stationarity.

Assume that we are at stage \alpha, thus the model we have produced so far is W_1[H_{\omega_2}][ I_{\alpha}] and we force with \dot{\mathbb{S}}_{\alpha} which is a ccc forcing of size \aleph_1 in W_1[H_{\nu}][I_{\alpha}] for \nu \le \alpha. Consider some block \vec{h}^{\zeta}, \zeta \ge \alpha in the universe W_1[H_{\omega_2}][\mathbb{R}_{\alpha} \ast \dot{\mathbb{S}}_{\alpha}]. The latter universe is obtained via the iteration

and the right hand side can be rewritten as

As we can switch the order in products, the latter can be written as

and consequentially the W_1[H_{\nu}]-generic filter H_{\nu, \omega_2} for the tail (\mathbb{Q}_{\zeta} \slash \mathbb{Q}_{\nu} \, : \, \zeta> \nu, \zeta < \omega_2) remains generic over the model W_1[(\mathbb{Q}_{\zeta} \, : \, \zeta \le \nu) \ast (\mathbb{R}_{\alpha} \ast \dot{\mathbb{S}}_{\alpha})]. This means in particular that the generically added Suslin trees \vec{h}^{\zeta}, \zeta > \nu for Tennenbaum’s forcing for adding a Suslin tree which are elements in H_{\nu, \omega_2} remain generic even over the ground model W_1[H_{\nu} \ast I_{\alpha+1}]. Now Tennebaum’s forcing is computed in an absolute way in every universe with the same \omega_1, and trivially, every generic filter for it is a Suslin tree. Hence we obtain that every h^{\zeta}_{\eta}, \zeta > \nu, \eta< \omega_1 is a Suslin tree in W_1[H_{\nu} \ast I_{\alpha+1} ], thus every h^{\zeta}_{\eta}, \zeta \ge \alpha \ge \nu is a Suslin tree in W_1[H_{\omega_2} \ast I_{\alpha+1}] as claimed.

If S is stationary then the rules of the iteration guarantee that there is such an \alpha < \omega_2 with the desired properties. On the other hand if there is an \alpha < \omega_2 such that the \alpha-th block \vec{h}^{\alpha} sees a certain 0,1-pattern then by the last Lemma, this pattern must come from the first case in the definition of our iteration. Hence S has to be stationary.

That \hbox{NS}_{\omega_1} is saturated is clear as W_2[I_{\omega_2}] is a ccc extension of W_0 and \hbox{NS}_{\omega_1} is saturated in W_0 and ccc indestructible. The proof that \mathsf{MA_{\omega_1}} holds in W_2[I_{\omega_2}] is also clear as a standard computation yields that the continuum is \aleph_2 in W_2[I_{\omega_2}]. Hence, it is sufficient to show that \mathsf{MA_{\omega_1}} holds for ccc posets of size \aleph_1. Let \mathbb{P} \in W_2[I_{\omega_2}] be such, then there is a stage \nu < \omega_2 such that \mathbb{P} \in W_2[I_{\nu}]. The rules of the iteration yield that we will consider \mathbb{P} unboundedly often after stage \nu. Thus there will be a stage \alpha< \omega_2 such that \mathbb{P} is considered by the bookkeeping F and \mathbb{R}_{\alpha} \ast \mathbb{P} is an element of W_1[H_{\nu}] for \nu \le \alpha, and hence we used \mathbb{P} in the iteration (\mathbb{R}_{\eta} \, : \, \eta < \omega_2), so \mathsf{MA_{\omega_1}} holds.

In order to see that every S \in W_2[I_{\omega_2}] has a \Sigma_1(\vec{C}, \vec{T}^0)-definition we exploit the fact that the trees (h_{\alpha} \, : \, \alpha < \omega_2) are \Sigma_1(\vec{C}, \vec{T}^0)-definable in W_1[I_{\omega_2}]. We claim that the following \Sigma_1(\vec{C}, \vec{T}^0)-formula \varphi(S) defines being stationary in W_2[I_{\omega_2}]:

• \varphi(S) if and only if there exists a triple of (M_1, M_2, M_3) of transitive models of size \aleph_1 such that M_1 \subset M_2 \subset M_3, M_1 is a suitable model and M_2 sees a full pattern on the trees in some \omega_1-block \vec{T}^{\alpha}. This pattern itself yields an \omega_1-block of trees \vec{h}^{\beta} and M_3 sees a full pattern on \vec{h}^{\beta} and this pattern is the characteristic function for S.

Note that the formula \varphi(S) is of the form \exists M_1,M_2,M_3 \,\sigma(M_1,M_2,M_3,S) and \sigma is \Delta_1 as all the statements in \sigma are of the form M_i \models... which is a \Delta_1-formula.

By absoluteness and the way we defined our iteration, it is clear that if S \subset \omega_1 is stationary in W_2[I_{\omega_2}], then \varphi(S) holds.

On the other hand, if \varphi(S) is true and M_1,M_2 and M_3 are witnesses to the truth of \varphi(S), then, as they see full patterns, their local patterns must coincide with the patterns in the real world W_2[I_{\omega_2}]. But the last Lemma ensures that the patterns of W_2[I_{\omega_2}] characterize stationarity, so the proof is finished.