Between reduced powers and ultrapowers, II

We need to prove that PFA implies \mathcal P({\mathbb N})/\operatorname{Fin} is not isomorphic to an ultraproduct of Boolean algebras associated with a nonprincipal ultrafilter on {\mathbb N}. By [van1989there] (see [Sh:1057, Corollary 1.9]), PFA implies that there are no tie points in \mathcal P({\mathbb N})/\operatorname{Fin}, while there are tie points in an ultraproduct of countable atomless Boolean algebras by Lemma 6.4.

If {\mathcal I} does not include the Frèchet filter then \mathcal P({\mathbb N})/{\mathcal I} has atoms, and therefore cannot be isomorphic to \mathcal P({\mathbb N})/\operatorname{Fin}. We may therefore assume that {\mathcal I} incudes the Frèchet filter.

In the model obtained by adding at least \mathfrak c^+ Cohen reals to a model of ZFC, suppose that \mathfrak B is a projectively definable Boolean algebra. By Lemma 6.5, there are no true tie points in \mathfrak B. By Lemma 6.4, in every model of \operatorname{ZFC} there is a true tie point in any ultraproduct of countable atomless Boolean algebras.

Fix a theory T with the robust order property. We will prove that the Levy collapse of the continuum to \aleph_1 followed by \mathbb H_E for E provided by Galvin’s theorem forces both conclusions of Theorem C.

These proofs have a common initial segment that we now present.

In the extension by the Levy collapse of the continuum to \aleph_1, let \kappa>\mathfrak c be a regular cardinal (\kappa=\aleph_2 will do). Let E be the poset as guaranteed by Galvin’s theorem stated at the beginning of this section. We will prove that the Levy collapse followed by \mathbb H_E is the forcing notion as promised in the statement of Theorem C.

Fix an ultrafilter \mathcal U on {\mathbb N}, a sequence (A_n) of countable structures in the language of T, and a sequence (B_n) of countable models of T. Proposition 4.15 implies that \mathbb H_E adds a strictly increasing map from E into \prod_{\mathcal U} (B_n,\triangleleft_\varphi) whose range is linearly ordered by \triangleleft_\varphi. By the choice of E, there exists a \kappa-\triangleleft_\varphi-chain or a \kappa^*-\triangleleft_\varphi-chain in \prod_{\mathcal U} (B_n,\triangleleft_\varphi).

On the other hand, by Theorem 4.1 there are neither \kappa-chains nor \kappa^*-chains in \prod_{\operatorname{Fin}} (A_n, \triangleleft_\varphi).

From this point on the proofs of (?) and (?) differ.

(?) We need to prove that \prod_{\mathcal U} B_n is not isomorphic to an elementary submodel of \prod_{\operatorname{Fin}} A_n. Since elementary embeddings preserve \triangleleft_\varphi, this is immediate from the fact that the former contains a \kappa- or \kappa^*-\triangleleft_\varphi-chain and the latter does not.

(?) Suppose in addition that \varphi is quantifier-free. Then all embeddings preserve \triangleleft_\varphi, and \prod_{\mathcal U} B_n cannot be isomorphic to a submodel of \prod_{\operatorname{Fin}} A_n.

Suppose that A is a separable C^*-algebra and \mathcal U is an ultrafilter on {\mathbb N}. If \mathcal U is principal, then (A\otimes C(K))^{\mathcal U} is isomorphic to A\otimes C(K) while A^\infty is nonseparable. We may therefore assume that A is infinite-dimensional and that \mathcal U is nonprincipal.

The theory of infinite-dimensional C^*-algebras has the order property witnessed by an atomic formula ([FaHaSh:Model1, Lemma 5.3]). Therefore the theory of A\otimes C(K) has the robust order property, and Theorem C (?) implies that (A\otimes C(K))^{\mathcal U} does not embed into B^\infty for any C^*-algebra B.

Fix models A_n\models T for n\in {\mathbb N} and suppose \varphi is an atomic formula. If \bar a_n and \bar b_n are tuples of the appropriate sort in A_n such that A_n\models \varphi(\bar a_n, \bar b_n), then (writing \bar a for the element of the product that has the representing sequence (\bar a_n)), we have \prod_n A_n\models \varphi(\bar a, \bar b) and moreover for any filter \mathcal F on {\mathbb N} we have \prod_{\mathcal F} A_n\models\varphi(\bar a, \bar b). This also applies to the negation of \varphi, and the assertion follows immediately.

Let C:={\mathbb N} and define the relation \triangleleft as follows. If m<n are in {\mathbb N} and n=\sum_j d_j(n) 3^j is the ternary expansion of n (so that d_j(n)\in \{0,1,2\} for all j) then let m\triangleleft n if d_m(n)=1, n\triangleleft m if d_m(n)=2, and let m and n unrelated if d_m(n)=0. The structure (C,\triangleleft) has the following property resembling the random graph:

1. (*) If F and G, are disjoint finite subsets of C, then there exists n\in C such that m\in F implies m\triangleleft n, m\in G implies n\triangleleft m, and m\notin F\cup G implies that m and n are unrelated.

To see this, let n:=\sum_{m\in F} 3^m+\sum_{m\in G} 2\cdot 3^m.

Given the property (*) of (C,\triangleleft), every countable (A,\triangleleft') can be isomorphically embedded into (C,\triangleleft) by recursion. Since (*) is a first-order property, it is absolute between transitive models of a sufficiently large fragment of \operatorname{ZFC} (see e.g., [Ku:Set, Lemma II.4.3]).

Fix A, I enumerated in the increasing order as \xi(i), for i<m, F(\xi), for i<m, and an ordering \leq on E:=A\cup \bigcup_{i<m} F(\xi(i)).

We will write \ll for \ll_s. It is clear from the definition that x\ll y implies x\leq y and that \ll and \leq agree on A\cup F(i) for every i. Therefore \ll is antisymmetric and reflexive, and it will suffice to prove that it is transitive.

Towards this end, fix x,y, and z such that x\ll y and y\ll z. Then x\leq y and y\leq z, and therefore x\leq z. If x and z belong to A\cup F(\xi(i)) for some i, then x\ll z by (?). Therefore if at least one of x\in A or z\in A holds then x\ll z, and we may assume

1. (xz) x\in F(\xi(i)) and z\in F(\xi(j)) for some distinct i and j.

If y\in A then x\ll z by (?). Similarly, if there exists a\in A such that x\leq a and a\leq y, then x\ll z. Also, if there exists a\in A such that y\leq a and a\leq z, then x\ll z. We can therefore assume (in addition to (xz) above) that

1. (y) y\in F(\xi(n)) for some n

and that both x\ll y and y\ll z are witnessed by instances of (?).

The proof now reduces to the analysis of the ordering of the set \{i,j,n\}. The following claim will help when discussing the possible cases.

Back to our proof. With i,j, and n as in (xz) and (y), if i\leq n\leq j, then part (?) of Claim implies that x\ll z. If i<j<n, then the witnessing sequence for x\ll y contains t\in F(\xi(j)), such that x\ll t and t\ll y. But then (since \ll implies \leq) t\leq z, and t\ll z since both t and z belong to F(\xi(j)). A proof in the case when n<i<j is similar and uses part (?) of the Claim. This proves our claim in the case when i<j.

The proof in the case when i>j is analogous.

Fix x\ll_t y in A\cup \bigcup_{\xi\in s} F(\xi). A glance at Definition 3.1 shows that we may assume that x\in F(\xi) and y\in F(\eta) for some distinct \xi and \eta in s, since in any other situation x\ll_s y follows immediately. If there is z\in A such that x\leq z\leq y, then clearly x\ll_w y whenever \{\xi,\eta\}\subseteq w\Subset I. We may therefore assume that there is a t-walk with endpoints x and y. We denote the ‘steps’ of s by x(\zeta), for \zeta\in t and \xi\leq \zeta\leq \eta. Lemma 3.3 implies that x(\zeta), for \zeta\in s and \xi\leq \zeta\leq \eta is an s-walk with endpoints x and y, hence x\ll_s y.

Let \mathbb P be the poset of all X\subseteq \kappa such that 0\in X which satisfy the condition *_{(\xi,\eta)} for all distinct \xi and \eta in X, ordered by the inclusion. This poset is clearly closed under unions of chains, and therefore Zorn’s lemma implies that it has a maximal element, X. (Readers not fond of the Axiom of Choice will notice that since \kappa is well-ordered, the existence of X can be proved by recursion in ZF.) We claim that X satisfies the requirements. Towards obtaining a contradiction, suppose that X is not cofinal in \kappa.

If X has no maximal element, then let \alpha:=\sup(X) and Y:=X\cup\{\alpha\}. In order to verify that Y belongs to \mathbb P, it suffices to verify *_{(\xi,\alpha)} for all \xi\in X. Fix \xi\in X. We need to prove that there exists t such that there is no t-walk with endpoints in F(\xi) and F(\alpha). Let \eta:=\min(X\setminus (\xi+1)) and (using *_{(\xi,\eta)}) let s\Subset \kappa be such that \{\xi,\eta\}=\{\min( s),\max(s)\} and there is no s-walk with endpoints F(\xi) and F(\eta). Let t:=s\cup \{\alpha\}. Then \{\xi,\alpha\}=\{\min(t), \max(t)\} and Lemma 3.3 implies there is no t-walk with endpoints in F(\xi) and F(\alpha).

Now suppose that X has a maximal element, \xi. Assume for a moment that there are \eta>\xi and s\Subset \kappa which witnesses that *_{(\xi,\eta)} holds. We claim that X\cup \{\eta\} belongs to \mathbb P. Fix \zeta\in X such that \zeta<\xi. Since X\in \mathbb P, there is t\Subset \kappa such that \{\zeta,\xi\}=\{\min(t),\max(t)\} and there is no t-walk with endpoints in F(\zeta) and F(\xi). With w:=s\cup t we have \{\zeta,\eta\}=\{\min(w),\max(w)\} and Lemma 3.3 implies that there is no w-walk with endpoints in F(\zeta) and F(\eta). Therefore *_{(\zeta,\eta)} holds. Since \zeta was an arbitrary element of X below \xi (and since *_{(\xi,\eta)} holds), this implies X\cup \{\eta\} is an element of \mathbb P strictly larger than X; contradiction.

It will therefore suffice to find \eta>\xi such that *_{(\xi,\eta)} holds. Assume that such \eta does not exist. Then for every \eta>\xi and every s\Subset \kappa which satisfies \{\xi,\eta\}=\{\min(s),\max(s)\} there is an s-walk with endpoints in F(\xi) and F(\eta). For each s fix a walk, x(s,\zeta), for \zeta\in s, with endpoints in F(\xi)=\min(s) and F(\eta)=\max(s).

Let \mathcal U be an ultrafilter on \operatorname{Fin}_\kappa which for every s\Subset \kappa includes the set \{t\Subset \kappa\vert s\subset t\} (such \mathcal U exists, since the family of sets of this form has the finite intersection property). Fix \xi\leq \zeta<\kappa. Since F(\zeta) is finite, there exists a unique y(\zeta) \in F(\zeta) such that \{s\vert x(s,\zeta)=y(\zeta)\}\in \mathcal U. Clearly y(\zeta), for \xi\leq \zeta<\kappa, is a \kappa-chain or a \kappa^*-chain (as decided by \mathcal U); contradiction.

Therefore there exist \eta>\xi and s\Subset \kappa such that \{\xi,\eta\}=\{\min(s),\max(s)\} such that there is no s-walk with endpoints in F(\xi) and F(\eta). As already pointed out, this implies X\cup \{\eta\}\in \mathbb P; contradiction.

Therefore any maximal element of \mathbb P is cofinal in \kappa, as required.

The proof of this theorem will occupy most of the present section. For \mathbb H_E see Definition 4.2, (?) is Lemma 4.4, (?) is Lemma 4.6, and (?) is Theorem 4.13.

Let D_q:=\bigcup_{i<m} D_i and n_q:=\max_{i<m} n_i. If i<m is such that n_i= n_q, then for a\in D_i let f_q(a)=f_i(a). Then f_q(a) is well-defined for a\in R by (?). For i<m such that n_i<n_q and for a\in D_i\setminus R, let (with \max\emptyset=0)

for n_i\leq_E j<n_q. This defines q\in \mathbb H_E. We will prove that q\leq p_i for all i<m.

Clearly, q and p_i satisfy (?) of Definition 4.2 for all i<m. Fix i<m.

If n_i=n_q then (?) of Definition 4.2 is vacuous, hence q\leq_E p_i.

Suppose n_i<n_q. To check that q\leq p_i, we need to verify (?) of Definition 4.2. Fix a and b in D_i such that a\leq_E b. If there is no c\in R such that c\leq_E b, then for all j\in [n_i,n_q) we have f_q(a)(j)=f_q(b)(j)=0. If there is c\in R such that c\leq_E b, then \{c\mid c\leq_E a\}\subseteq \{c\mid c\leq_E b\} and by the definition of f_q we have f_q(a)(j)\leq f_q(b)(j).

Thus (?) of Definition 4.2 holds, and q\leq p_i.

Fix a family p_\xi, for \xi<\kappa, in \mathbb H_E. By the \Delta-system lemma and passing to a subfamily of the same cardinality, we may assume that there exists R\Subset E such that D_\xi\cap D_\eta=R for all distinct \xi and \eta below \kappa. By the pigeonhole principle (using the assumption that \kappa has uncountable cofinality), we may also assume that there exists n such that n_\xi=n for all \xi. Also, since there are only finitely many possibilities for f_\xi(a), for a\in R, we may assume that the functions f_\xi agree on R and therefore we are in the situation of Lemma 4.3. Therefore, after this refining argument, Lemma 4.3 implies that every finite subset of \{p_\xi\mid \xi<\kappa\} has a common lower bound.

By genericity, G intersects all dense sets defined in Lemma 4.5 and therefore \Upsilon is a strictly increasing map from E into (\prod_k k ,\leq^*).

The identity map from \mathbb H_{E} into \mathbb H_{E'} is clearly an order-embedding. It suffices to prove that there exists a reduction (or projection) \pi\colon \mathbb H_{E'}\to \mathbb H_{E}: A map such that for every p\in \mathbb H_{E'} we have p\leq \pi(p) and every q\in \mathbb H_{E} such that q\leq \pi(p) is compatible with p ([Ku:Set, Lemma III.3.72]). Let

Clearly, p\leq \pi_E (p). If q\leq \pi(p), then D_q\cap D_p=D_p\cap E and f_p(a)(j)=f_q(a)(j) for all a\in D_p\cap D_q and all j<n_p. By Lemma 4.3, p and q are compatible.

Since C as in Lemma 2.2 is universal, \mathbb H_E forces that \prod_{\operatorname{Fin}} (A_n, \triangleleft_n) has a \kappa-chain for some sequence (A_n,\triangleleft_n) (not necessarily in the ground model) if and only if \mathbb H_E forces that (C^{\mathbb N},\triangleleft^*):=\prod_{\operatorname{Fin}} (C,\triangleleft) has a \kappa chain. It will therefore suffice to prove the theorem with the additional assumption that (A_n,\triangleleft_n)=(C,\triangleleft) for all n.

Assume that \dot f_\xi, for \xi<\kappa, is a name for a \kappa-chain in (C^{\mathbb N},\triangleleft^*). (We emphasize that this means that for all \xi<\eta, \mathbb H_E forces both \dot f_\xi\triangleleft^* \dot f_\eta and \dot f_\eta\ntriangleleft^* \dot f_\xi.) The ccc-ness of \mathbb H_E implies that for every limit ordinal \xi there exists a countable E(\xi)\subseteq E such that \dot f_\xi, \dot f_{\xi+1}, and \dot f_{\xi+2} are \mathbb H_{E(\xi)}-names. Since \kappa is regular and \lambda^{\aleph_0}<\kappa for all \lambda<\kappa, the \Delta-system lemma for countable sets implies that any family of \kappa countable sets includes a \Delta-system of cardinality \kappa. By passing to a subfamily, we may assume that the sets E(\xi) form a \Delta-system with a countable root A.

For every limit ordinal \xi fix q_\xi\in \mathbb H_E and n\in {\mathbb N} such that

By the pigeonhole principle and passing to a subfamily if necessary, we may assume that n as in (?) is the same for all \xi. Lemma 4.7 implies that \mathbb H_{E(\xi)} is a regular subordering of \mathbb H_E. We may therefore assume that q_\xi\in \mathbb H_{E(\xi)} and we have

Writing q_\xi=(D_\xi,n_\xi, f_\xi), let F(\xi):=D_{\xi}\setminus A. Note that each F(\xi) is finite.

Fix \xi<\kappa for a moment and fix a generic filter G\subseteq \mathbb H_{A\cup F(\xi)} such that q_\xi\in G. Recall that the domain of C is {\mathbb N}, and is therefore equipped with a well-ordering (as a matter of fact, the first well-ordering known to man). For j\in{\mathbb N} we can let h_\xi(j) be the least element of C (in this well-ordering) such that

for some r in the quotient \mathbb H_E(A\cup F(\xi),\Upsilon_G)/G (see Lemma 4.10).

This defines h_\xi\in C^{\mathbb N} in V[G]. Use the Maximality Principle ([Ku:Set, Theorem IV.7.1]) to choose an \mathbb H_{A\cup F(\xi)}-name \dot h_\xi for this function.

The pairs (q_\xi,\dot h_\xi) are indexed by limit ordinals below \kappa. We re-enumerate them preserving the order and obtain conditions q_\xi and names \dot h_\xi for \xi<\kappa. Since \mathbb H_E has the ccc, some condition q\in \mathbb H_E forces that \kappa of the q_\xi’s belong to the generic filter. Therefore q forces that the family of all h_\xi such that q_\xi belongs to the generic filter is a \kappa-chain in (C^{\mathbb N}, \triangleleft^*).

By Proposition 3.9, there exists a cofinal X\subseteq \kappa such that for any two distinct elements \xi<\eta of X there is s\Subset \kappa such that \{\xi,\eta\}=\{\min(s),\max(s)\} and there is no s-walk with endpoints in F(\xi) and F(\eta).

Fix \xi\in X, let \xi':=\min(X\setminus (\xi+1)), and fix s\Subset \kappa such that \{\xi,\xi'\}=\{\min(s),\max(s)\} and there is no s-walk with endpoints in F(\xi) and F(\xi'). We will analyze the relation between the names \dot h_{\xi} and \dot h_{\xi'}.

Let \xi(j), for j<n,(This n is unrelated to the n appearing in ?. No danger of confusion here.) be an increasing enumeration of s, so that in particular \xi(0)=\xi and \xi(n-1)=\xi'. Consider the depletion \ll_s of \leq_E given by A and F(\eta), for \eta\in s. By Lemma 3.4, \ll_s is a partial ordering on

For i<j<n let

ordered by \ll_s. Note that if |i-j|>1 then A(i,j) is ordered by a restriction of the depletion associated with s which can differ from the natural depletion \ll_{(\xi(i), \xi(j))} (Remark 3.7). However, for i<n-1, Lemma 3.8 implies that the restriction of \ll_s to A(i,i+1) agrees with the depletion \ll_{\{\xi(i),\xi(i+1)\}}, which by Lemma 3.2 agrees with the ordering induced from E. By Lemma 4.7, we have \mathbb H_{A(i,i+1)}\lessdot \mathbb H_E.

Since A' is ordered by \ll_s, which possibly disagrees with the ordering of A' inherited from E, the posets \mathbb H_{A'} and \mathbb H_E are possibly unrelated. On the other hand, each A(i,j) is a subordering of A' and therefore Lemma 4.7 implies \mathbb H_{A(i,j)}\lessdot \mathbb H_{A'}.

Since \dot h_{\xi(i)} and \dot h_{\xi(j)} are \mathbb H_{A(i,j)}-names and \mathbb H_E forces \dot h_{\xi(i)}\triangleleft^* \dot h_{\xi(i+1)} for all i<n-1, we have that \mathbb H_{A'} forces \dot h_{\xi(i)}\triangleleft^* \dot h_{\xi(i+1)} for all i<n-1. By transitivity, \mathbb H_{A'} forces \dot h_{\xi(0)}\triangleleft^* \dot h_{\xi(n-1)}. Since \dot h_{\xi(0)} and \dot h_{\xi(n-1)} are \mathbb H_{A(0,n-1)}-names and \mathbb H_{A(0,n-1)}\lessdot \mathbb H_{A'}, we have that \mathbb H_{A(0,n-1)} forces \dot h_{\xi(0)}\triangleleft^* \dot h_{\xi(n-1)}.

Since there is no s-walk whose endpoints are some x\in F(\xi(0)) and some y\in F(\xi(n-1)), we have that x\ll_s y implies there is a\in A such that x\leq a\leq y, and that y\ll_s x imples there is a\in A such that y\leq a\leq x. This means that the assumptions of Lemma 4.12 with E, A, B, and D replaced with A\cup F(\xi(0))\cup F(\xi(n-1)), A\cup F(\xi(0)), A\cup F(\xi(n-1)), and A, respectively, (sorry!) are satisfied. Therefore if G\subseteq \mathbb H_A is a generic filter then the quotient \mathbb H_{A\cup A(0,n-1)}/G is forcing-equivalent to the product

Most importantly, the names \dot h_{\xi(0)} and \dot h_{\xi(n-1)} are added by the two factors of this product. By Lemma 4.11, there exist a condition p_\xi\in \mathbb H_A and an \mathbb H_A-name \dot g_\xi (recall that \xi=\xi(0)) such that

Since \mathbb H_A is countable, there is q\in \mathbb H_A such that Y=\{\xi\in X\vert p_\xi=q\} is a cofinal subset of X (and of \kappa). Therefore q forces that \mathbb H_A adds a strictly increasing \kappa-chain \dot g_\xi, for \xi<\kappa, to ({\mathbb N}^{{\mathbb N}},\leq^*). Since A is countable, \mathbb H_A cannot add more than \mathfrak c reals; contradiction.

The first part is almost obvious, but proving it in some detail will also provide a proof of the second part.

By Lemma 2.1, there are \eta\in {\mathbb N}^{{\mathbb N}} and \Phi\colon \prod_{\operatorname{Fin}} (n,\leq^*)\to \prod_{\operatorname{Fin}} (\eta(n),<^*) such that f\leq^* g and g\nleq^* f implies (\forall^\infty n) \Phi(f)(n)<\Phi(g)(n). Since A_n is a model of T, there exists a \triangleleft_\varphi-chain C_n of length \eta(n) in A_n. By identifying this chain with (\eta(n),\leq), we obtain \Phi\colon \prod_{\operatorname{Fin}} (n, \leq^*)\to\prod_{\operatorname{Fin}} (A_n,\triangleleft_\varphi) such that f\leq^* g and g\nleq^* f implies \Phi(f)(n)\triangleleft_\varphi \Phi(g)(n) and \Phi(g)(n)\ntriangleleft_\varphi \Phi(f)(n) for all but finitely many n. By composing the embedding of E into \prod_{\operatorname{Fin}} (n,\leq^*) provided by Theorem 4.1 with \Phi, we obtain an \mathbb H_E-name for an embedding of \Xi\colon E\to \prod_{\operatorname{Fin}} (A_n,\triangleleft_\varphi) (this proves the first part; read on for the proof of the second part) that in addition has the property that a<_E b implies

Let \mathcal U be a nonprincipal ultrafiter on {\mathbb N} and let \pi_{\mathcal U} denote the quotient map from \prod_{\operatorname{Fin}} A_n to \prod_{\mathcal U} A_n. Then the displayed formula implies that the restriction of \pi_{\mathcal U} to \Xi[E] is strictly increasing. The range of this map is the ultraproduct of the \triangleleft_\varphi-chains C_n, and therefore linearly ordered by Ł oś’s Theorem. This proves the second part.

Every ultrafilter in a countable atomless Boolean algebra is a true tie point, since the generating sets of order type \omega can be chosen by recursion. Suppose \prod_{\mathcal U} C_n is an ultraproduct of countable atomless Boolean algebras. If \mathcal U is principal, then \prod_n C_n is isomorphic to one of the C_n’s and the assertion follows from the first sentence of this proof.

Now assume \mathcal U is nonprincipal. For every C_n fix a true tie point p_n and linearly ordered generating sets {\mathcal A}_n and \mathcal B_n for the ideal C_n\setminus p_n. Then (C_n, {\mathcal A}_n,\mathcal B_n) is an expansion of C_n to the language with two additional unary predicates. Each one of these structures satisfies the following: Both {\mathcal A}_n and \mathcal B_n are linearly ordered, A\wedge B=\emptyset for all A\in {\mathcal A}_n and B\in \mathcal B_n, and for every X\in C_n either X or its complement belongs to {\mathcal A}_n\cup \mathcal B_n. These are all first-order statements, and they imply that the complement of {\mathcal A}_n\cup \mathcal B_n is an ultrafilter.

The ultraproduct \prod_{\mathcal U} (C_n,{\mathcal A}_n,\mathcal B_n) is an expansion of \prod_n C_n and by Ł oś’s Theorem the sets {\mathcal A}:=\prod_{\mathcal U} {\mathcal A}_n and \mathcal B:=\prod_{\mathcal U} \mathcal B_n generate ideals of \prod_{\mathcal U} C_n whose complement is a true tie point.

Suppose that \kappa\geq \mathfrak c^+ and let \mathbb C_\kappa denote the poset for adding \kappa Cohen reals. Fix an n\in {\mathbb N} and \mathbf\Sigma^1_n-formulas \varphi, \varphi_\wedge, \varphi_\vee, and \varphi_{\setminus}, which define \mathfrak B. We will only need the \mathbf \Delta^1_{n+1} formula \varphi_{<}, that defines the relation a<b in \mathfrak B. By passing to an intermediate forcing extension, without a loss of generality we may assume that the reals coding these formulas belong to the ground model.

By genericity, no nonprincipal ultrafilter on {\mathbb N} in the forcing extension is generated by fewer than \kappa subsets of {\mathbb N}. (This is well-known, but here is a sketch of the proof: After adding \kappa Cohen reals, for every {\mathcal X}\subseteq \mathcal U of cardinality less than \kappa there is a Cohen real Y generic over V[{\mathcal X}]. For every infinite X\subseteq {\mathbb N}, the set of all Y\subseteq {\mathbb N} such that X\cap U and X\setminus Y are both infinite is comeager. Therefore {\mathcal X} does not ‘decide’ whether Y\in \mathcal U or {\mathbb N}\setminus Y\in \mathcal U.) Assume p is a true tie point in \mathfrak B and let {\mathcal A} and \mathcal B be the linearly ordered (modulo {\mathcal I}) sets whose complements generate \mathfrak B\setminus p. By genericity, at least one of {\mathcal A} and \mathcal B has cofinality greater than \mathfrak c. By interchanging {\mathcal A} and \mathcal B, we may assume that the cofinality of {\mathcal A} is \kappa>\mathfrak c.

The proof is completed by Kunen’s isomorphism of names argument ([kunen1969inaccessibility]; see §? for disambiguation) that we now sketch.

Suppose that \dot f_\xi, for \xi<\kappa, is a \mathbb C_\kappa-name for a strictly increasing chain cofinal in {\mathcal A}. In particular, f_\xi, for \xi<\mathfrak c^+, is a name for a strictly increasing chain in {\mathcal A}/{\mathcal I}. This will suffice to obtain a contradiction. Since each \dot f_\xi is a name for a real, it is coded by a sequence of antichains and the union, denoted D_\xi, of the supports of all conditions in these antichains is a countable subset of \kappa. Since \mathfrak c^+ is regular and \lambda<\mathfrak c^+ implies \lambda^{\aleph_0}<\mathfrak c^+, by passing to a cofinal subset we may assume that the sets D_\xi form a \Delta-system with root R. By a counting argument and passing to a cofinal subset again we may assume that the restrictions of \dot f_\xi to R agree, and that \dot f_\xi and \dot f_\eta are isomorphic for all \xi and \eta. This means that for \xi<\eta there is an automorphism \Phi_{\xi\eta} of \mathbb C_\kappa that sends \dot f_\xi to \dot f_\eta, for any two \xi<\eta<\mathfrak c^+. However, since \mathbb C_\kappa forces that \varphi_<(\dot f_\xi, \dot f_\xi) is true and the real coding the asymmetric formula \varphi_< is in the ground model, this is a contradiction that completes the proof.

It is well-known that every model of T of cardinality \mathfrak c is isomorphic to an elementary submodel of an ultrapower of a countable model of T. This follows from the results of [Sh:c, Chapter VI.5] or [Sh:954]. Therefore any M\in \operatorname{Ult}_T which is \preceq-universal is also \preceq-universal for \operatorname{Mod_{\mathfrak c}}(T), and similarly for \hookrightarrow-universality.

In [FaHaSh:Model2, Theorem 5.6] it was proved that if T is stable then all ultrapowers A^\mathcal U for a countable (or separable) A and \mathcal U\in \beta{\mathbb N}\setminus {\mathbb N} are saturated, and therefore isomorphic.

The existence of \kappa as stated implies there is no \hookrightarrow-universal linear order of cardinality \mathfrak c by [Sh:409, Theorem 3.10]. By [Sh:409, Theorem 5.5], this implies that \operatorname{Mod_{\mathfrak c}}(T) has no \preceq-universal element. As explained in [Sh:409], if the SOP is witnessed by a quantifier-free formula, then \operatorname{Mod_{\mathfrak c}}(T) has no \hookrightarrow-universal element.

This is [Sh:500, Theorem 2.12].

This is [Sh:1029, Theorem 2.9 (i) and (ii)]. For the quantifier-free case, use part (ii) and the \lambda-(\eta,k,m)-olive property from [Sh:1029, Definition 2.3 (ii)].

We will prove the \preceq-case, starting with the following.

Suppose towards contradiction that \operatorname{Mod_{\mathfrak c}} has a \preceq-basis \mathbb B of cardinality less than 2^\kappa. Since \operatorname{Ult}_T has 2^\mathfrak c elements ([Sh:954]), we can fix \mathbb A\subseteq \operatorname{Ult}_T of cardinality 2^\kappa. With M_A, for A\in \mathbb A, as provided by Claim Proof, for every A there is N_A\in \mathbb B such that M_A\preceq N_A. Since |\mathbb B|<2^\kappa, the conclusion of Claim Proof fails; contradiction.

A proof of the \hookrightarrow-case of Proposition is analogous, using the modification of the Claim in which it is allowed that N_\xi\hookrightarrow N_\xi'.