LARGE CARDINALS AND GAP-1 MORASSES ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN 8 Abstract. We present a new partial order for directly forcingmorasses to exist 0 thatenjoysasignificanthomogeneityproperty. Wethenusethisforcinginareverse 0 Eastoniterationtoobtainanextension universewithmorassesateveryregularun- 2 countablecardinal,whilepreservingalln-superstrong(1≤n≤ω),hyperstrongand 1-extendible cardinals. In the latter case, a preliminary forcing to make the GCH n a holdisrequired. Ourforcingyieldsmorassesthatsatisfyanextrapropertyrelatedto J thehomogeneityofthepartialorder;werefertothemasmangroves andprovethat 2 theirexistenceisequivalenttotheexistenceofmorasses. Finally,weexhibitapartial 1 order that forces universal morasses to exist at every regular uncountable cardinal, andusethis toshow that universalmorassesareconsistent withn-superstrong,hy- ] perstrong, and 1-extendible cardinals. This all contributes to the second author’s O outer model programme, theaimofwhichistoshowthat L-likeprinciplescanhold L inouter modelswhichnevertheless containlargecardinals. . h t a m §1. Introduction. Morasses are combinatorial structures formu- [ lated by Jensen to abstract out properties of L useful for proving 1 cardinal transfer theorems. Originally thought to be such complex v 2 objects that one should not attempt to understand them outside 1 the context of their construction in L (according to [11]), they were 9 opened up to broader study and deeper understanding by Velleman 1 . [12], who amongst other things presented a particularly elegant par- 1 0 tial order with which one can force morasses to exist. This lends 8 itself to use in the second author’s outer model programme in which, 0 as a counterpoint to the well-known inner model programme, the : v goal is to force to obtain L-like outer models for various large car- i X dinals. Indeed, in [7] it is shown that one may force morasses to r exist at every regular uncountable cardinal, while preserving a given a n-superstrong or hyperstrong cardinal. Inthisarticlewe extend theseresults. Inadditionton-superstrong and hyperstrong cardinals we consider 1-extendible cardinals, for which a careful consideration of forcing the GCH as a preliminary step is appropriate. Further, we exhibit a new forcing partial or- der to obtain the morasses, which allows us to preserve all cardinals of the considered kinds. This is achieved by modifying Velleman’s forcing to give it a degree of homogeneity. Finally, we show how one may force universal morasses to exist. This new forcing again Andrew Brooke-Taylor was supported for this research by Austrian Science Fund (FWF) project P16790-N04. Sy-David Friedman was supported by FWF projects P16790-N04 and P19375-N18. 1 2 ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN fits in with the techniques of [7], and so we obtain that universal morasses can be forced while preserving an n-superstrong, hyper- strong or 1-extendible cardinal. This answers a question of Pereira to the authors. The reader may be aware that the existence of morasses has been shown by Velleman [13] to be equivalent to the existence of much simpler objects called simplified morasses. However, while these lat- ter structures are much more manageable, the only known way to obtain them in L is via morasses themselves. Thus, to stay more di- rectly in keeping with the goalof forcing L-like models, we stick with morasses. On the other hand, we do use Velleman’s [12] axiomati- sation of morasses, which already strips away some of the detail of morasses as defined by Jensen. This will have ramifications when we force to obtain universal morasses — see section 9 below. §2. Preliminaries: the GCH and 1-Extendible Cardinals. Our forcing constructions will rely on having the GCH hold in the ground model, so we discuss here how to attain it by a preliminary forcing. The iterated forcing we use to obtain the GCH is a fairly standard, natural one, at each stage collapsing the old 2ℵα to ℵ . α+1 The main issue is to show that forcing with this particular partial order does preserve the large cardinals we are interested in. In many cases, this was achieved in [7]; we show here that similar arguments go through for 1-extendible cardinals. Following [10], we denote by Fn(I,J,λ) the set of partial functions from I to J with cardinality less than λ. Definition 1. The GCH Partial Order P is the reverse Easton iteration of hQ˙ |α ∈ Ordi, in which direct limits are taken only at α strongly inaccessible cardinals and inverse limits are taken at other limitstages, where Q˙ is the canonicalP -nameforFn(ℵ ,2,ℵ ), α α α+1 α+1 with ℵ to be evaluated in V[G ]. α+1 α For the rest of this section P = P will denote the GCH Partial Ord Order, with P denoting the iteration after α stages, G = G α Ord denoting a P-generic over V, G denoting G↾P , and Q denoting α α α Fn(ℵ ,2,ℵ ) in V[G ]. α+1 α+1 α Before proceeding with the analysis of this iteration, let us re- call the basic properties of the forcing poset Fn(ℵ ,2,ℵ ). It is α+1 α+1 ℵ -closed and (2ℵα)+-cc, so the only cardinals collapsed by forcing α+1 with it are those κ such that ℵ < κ ≤ 2ℵα. Of course, since α+1 Fn(ℵ ,2,ℵ ) is equivalent to Fn(ℵ ,2ℵα,ℵ ), a surjection α+1 α+1 α+1 α+1 ℵ → 2ℵα is added, and all such κ are indeed collapsed. By a α+1 nice names argument, the continuum function is unchanged by this forcing at and above the ground model 2ℵα, and 2ℵα+1 in the exten- sion will be the ground model 22ℵα. By ℵ -closure, the contin- α+1 uum function is unchanged below ℵ , and after the forcing we have α 2ℵα = ℵ . α+1 LARGE CARDINALS AND MORASSES 3 Now, to the GCH partial order itself. To demonstrate that ZFC is preserved by this class forcing, we show that it is tame (see [6]). Lemma 2. The GCH Partial Order is tame. Proof. By the Factor Lemma (see for example [8], Lemma 21.8), P may be written as P ∗P˙α, where P˙α names the iteration starting α with Q . Each iterand in P˙α is forced to be ℵ -closed, and inverse α α+1 limits are taken everywhere except at inaccessibles of V, which will remain inaccessible after forcing with P (indeed, it can be shown α by induction that P has a dense suborder of size at most iV ). α α+1 We therefore have that for all α 1 (cid:13) P˙α is ℵ -closed, Pα α+1 and this implies that P is tame — see [6], Lemmas 2.22 and 2.31. ⊣ Lemma 3. Let α be an ordinal or Ord. Then for every ordinal γ ≤ α, iV[Gα] = ℵV[Gα]. γ γ Proof. As in the proof of Lemma 2 above, P may be factorised α as P ∗P˙[γ,α), where P˙[γ,α) names the iteration starting with Q , and γ γ we have 1 (cid:13) P˙[γ,α) is ℵ -closed. Pγ γˇ+1 It hence suffices to show that iV[Gγ] = ℵV[Gγ] for every γ. This may γ γ be proven by induction on γ, with the successor step following from the discussion of Fn(ℵ ,2,ℵ ) above. ⊣ α+1 α+1 In particular, these two lemmas combine to show that forcing with P yields a model of ZFC+GCH. In [7], it is shown that when forcing with the GCH forcing P, any given hyperstrong or n-superstrong cardinal may be preserved. In fact, the same proof essentially gives the following stronger state- ment. Theorem 4. Forcing with the GCH partial order P preserves all hyperstrong cardinals and n-superstrong cardinals, for all n ∈ ω+1. Proof. Observe that each iterand in the GCH forcing is very ho- mogeneous: given a condition p ∈ Q = Fn(ℵ ,2,ℵ ) and a α α+1 α+1 Q -generic g, there is an automorphism π of Q such that p ∈ π“g. α α Such strong homogeneity is known to hold for an entire iteration if it is forced to hold for each iterand; see [5] for the details. It follows that all the arguments for Theorem 4.2 of [7], which in par- ticular choose P-generics containing a certain master condition for each large cardinal κ, can be carried out within V[G]. Therefore, all large cardinals of the listed kinds are preserved in V[G]. ⊣ We shall show that the same result is true for 1-extendible cardi- nals. Recall their definition: Definition 5. A cardinal κ is 1-extendible if there is a λ > κ and an elementary embedding j : H → H with critical point κ. κ+ λ+ 4 ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN Equivalently, κ is1-extendible if andonly if there isa λ > κand an elementary embedding V → V with critical point κ. The above κ+1 λ+1 definition will be more convenient for our purposes, however, as H κ+ forms a model of ZFC minus the Power Set axiom, which we shall refer to as ZFC−. 1-extendible cardinals lie in consistency strength between superstrong and hyperstrong cardinals (see [9], Proposition 26.11) and play an interesting role in the outer model programme, as Cummings and Schimmerling [3] have shown them to be essentially the strongest large cardinals compatible with the principle (cid:3). Theorem 6. For any model V of ZFC, there is a class-generic extension V[G] of V such that V[G] (cid:15) GCH, and in which every 1-extendible cardinal of V remains 1-extendible. Proof. We shall of course force with the GCH forcing P. Let κ be 1-extendible in V, let j : H → H in V witness this fact, and κ+ λ+ let G be P-generic over V. We shall show that j may be lifted to an embedding j∗ : HV[G] → HV[G] witnessing the 1-extendibility of κ in κ+ λ+ V[G]. The GCH forcing P may be factorised as P ∗Pκ, with Pκ a κ+- κ closed forcing. Hence, HV[G] = HV[Gκ]. Similarly, HV[G] = HV[Gλ]. κ+ κ+ λ+ λ+ Now because κ is inaccessible, P has cardinality κ and lies in H . κ κ+ Recall that forcing over a model of ZFC− yields a model of ZFC−, and note thatG is generic over H for the set forcing P . We claim κ κ+ κ that in fact HV[Gκ] = HV [G ], the generic extension of the model κ+ κ+ κ HV of ZFC− by G . For this we need to show that if σ is a P -name κ+ κ κ in V such that σ ∈ HV[Gκ], then there is a name τ ∈ HPκ such Gκ κ+ κ+ that σ = τ . But now every element of H can be obtained as Gκ Gκ κ+ the Mostowski collapse of a relation on κ given by a subset of κ×κ (and this process does not appeal to the Power Set Axiom). Every subset of κ×κ in the extension V[G ] has a nice name of the form κ ˇ {(α,β)}×A (α,β) (α,β)∈κ×κ [ where each A is an antichain in P . Since |P | = κ, such nice (α,β) κ κ names lie in HV . Therefore, every subset of κ×κ in V[G ] is also in κ+ κ H [G ], and consequently we indeed have that HV[Gκ] = HV [G ]. κ+ κ κ+ κ+ κ Ofcourse byelementarity, λisalsoinaccessible, andsoalso HV[Gλ] = λ+ HV [G ]. λ+ λ We are thus reduced to showing that j : H → H may be lifted κ+ λ+ to an elementary embedding j∗ : H [G ] → H [G ]. We wish to κ+ κ λ+ λ applySilver’s technique ofliftingembeddings(see[2], Section9),and must show that j“G ⊆ G . But now j↾P is the identity function, κ λ κ so j“G = G ⊂ G . Therefore j indeed lifts to j∗ : HV[G] → HV[G], κ κ λ κ+ λ+ and so κ is 1-extendible in V[G]. ⊣ In Section 8, we will show that 1-extendible cardinals κ may be preserved while forcing morasses to exist. To show that the forcing LARGE CARDINALS AND MORASSES 5 is pretame from the perspective of H , we will first force GCH as κ+ above, andusethefactthatafterourGCHforcing,V[G] = L[G],and V[G] further H = L [A], where Ais taken to bea class predicate over κ+ κ+ H for the Cohen set added at stage κ. This gives a stratification κ+ of our model H of ZFC− into sets, as is required for the proof that κ+ the forcing relation is definable for a pretame class forcing — see [6], Theorem 2.18. Only then will we be able to conclude that the forcing relation for our forcing is definable. However, p (cid:13) ϕ(σ ,... ,σ ) will only be definable relative to A, 0 n−1 and so the usual lifting lemma will not suffice to lift the embedding j witnessing 1-extendibility, as j need not respect arbitrary class predicates. We show here that in fact j does respect the predicate A given by the generic, giving a mild strengthening of Theorem 6. The crux of the proof will be the definability of the forcing relation for P , which from the point of view of H is a class forcing. The κ+1 κ+ usual argument for pretame class forcing (Theorem 2.18 of [6]) is not applicable, since we don’t yet have a stratification of H into κ+ sets. Nevertheless, we will be able to demonstrate the definability of forcing for P because it is such a well-behaved forcing partial κ+1 order. Definition 7. The language L is the language obtained from STG the language of set theory by adding a single unary predicate G. Lemma 8. Let κ ∈ M be an inaccessible cardinal, and let Q de- κ note Fn(κ+,2,κ+), the forcing at stage κ in the GCH forcing above. Then the forcing relation (cid:13) ϕ for forcing over H with Q is defin- κ+ κ able over H for L formulae ϕ, where the predicate G is to be κ+ STG interpreted as the Q -generic, and the Truth Lemma holds: a formula κ is true in the extension if and only if it is forced by some condition in the generic. Proof. We follow the expository style of [10], defining a relation (cid:13)∗ ϕ for each ϕ, which will then be seen to be equivalent to (cid:13) ϕ. We will make use of the closure properties of Q to make this definition κ for atomic formulae, from which point it may be extended to all formulae as usual. As a first step, we give a definition for forcing the predicate G to hold: define p (cid:13)∗ G(σ) to mean (1) ∀q ≤ p∃r ≤ q∃s ≥ r(r (cid:13)∗ σ = sˇ), that is, it is dense below p to force σ to be something specific greater than yourself. If p ∈ G, then by genericity there is some r ∈ G and some s greater than or equal to r and consequently also in G such that r (cid:13)∗ σ = sˇ, whence σ ∈ G. Conversely, suppose σ ∈ G for G G every G ∋ p; let q ≤ p and fix some G ∋ q. Once we have shown that the forcing relation is definable for atomic formulae, it will follow from the truth lemma there is some r′ ∈ G such that r′ (cid:13) σ = (σˇ ). G 6 ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN Taking r ∈ G such that r ≤ q, r ≤ r′, and r ≤ σ , we have that r G makes (1) hold with σ as s. Hence, (1) is indeed a formal definition G encapsulating the statement that for every generic containing p, σ G is in G, and it is clear that the argument extends to prove the Truth Lemma valid for formulae involving the predicate G. With forcing the predicate G so defined, we can move on to show that the forcing relation is indeed definable for atomic formulae. Let σ and τ be Q -names in H , and let κ κ+ R = trcl({σ,τ})∩P, σ,τ the set of conditions hereditarily appearing in either σ or τ. Note that since the names σ and τ are in H , |R | ≤ κ. Let d = κ+ σ,τ σ,τ dom(r); then d is a subset of κ+ of size at most κ. r∈Rσ,τ σ,τ Now, for conditions q such that d ⊆ dom(q), we may recursively σ,σ S define the evaluation of σ at q by val(σ,q) = {val(ρ,q)|∃X ⊆ dom(q)(hρ,q↾Xi ∈ σ)}. Observe that if G is any Q -generic over H containing such a q, κ κ+ then σ = val(σ,q). Likewise, for any G which is Q -generic over G κ H and any element q of G with domain containing d , we have κ+ σ,σ val(σ,q) = σ . Thus, the evaluation of any name in the generic G extension may be entirely determined by a single condition. With this in hand, we may define p (cid:13)∗ σ ∈ τ ←→ ∀q ≤ p dom(q) ⊇ d → val(σ,q) ∈ val(τ,q) σ,τ and (cid:0) (cid:1) p (cid:13)∗ σ = τ ←→ ∀q ≤ p dom(q) ⊇ d → val(σ,q) = val(τ,q) . σ,τ Then clearly p (cid:13)∗ σ ∈ τ if and only if σ ∈ τ for every Q -generic (cid:0) G G κ (cid:1) G ∋ p, and p (cid:13)∗ σ = τ if and only if σ = τ for every Q -generic G G κ G ∋ p. Moreover, if σ ∈ τ , then any p ∈ G with dom(p) ⊇ d G G σ,τ will force σ = τ, and the Truth Lemma follows. ⊣ We are now ready for our strengthening of Theorem 6. Let use denote by G(κ) the generic from stage κ of G, that is, the Q -generic κ associated to G. Theorem 9. Let V be a model of ZFC and let κ be a 1-extendible cardinal in V with 1-extendibility witnessed by an elementary em- bedding j : H → H . Let V[G] be a P-generic extension of V. κ+ λ+ Then there is a G′ ⊂ V[G] which is P-generic over V such that V[G] = V[G′] and the lift j∗ of j to HV[G′] (as in the proof of The- κ+ orem 6) is elementary between the L -structures hHV[G′],G′(κ)i STG κ+ and hHV[G′],G′(λ)i. λ+ Proof. In Theorem 6 the lift j∗ was constructed in HV[Gκ], after κ+ observing that Pκ is κ+-closed and hence does not affect H . We κ+ now claim that the Pκ generic Gκ may be chosen so that this same j∗ is also elementary for formulae in the language L . By the STG LARGE CARDINALS AND MORASSES 7 Truth Lemma, every L sentence ϕ true in hHV[Gκ+1],G(κ)i = STG κ+ hHV[G],G(κ)i is forced (over HV[Gκ]) to be true by some p ∈ Q . κ+ κ+ κ Now by the definability of the forcing relation, we have p (cid:13) ϕ(σ ,... ,σ ) ↔ j∗(p) (cid:13) ϕ(j∗(σ ),... ,j∗(σ )). Qκ 0 n Qλ 0 n Therefore, ifj∗“G(κ) ⊆ G(λ),wewillhavethatj∗ iselementaryfrom hHV[G],G(κ)i to hHV[G],G(λ)i. But now |G(κ)|V[Gλ] = (κ+)V[Gλ] < κ+ λ+ λ,soj∗“G(κ)isapairwise-compatiblesetofconditionsinFn(λ,2,λ)V[Gλ] (indeed, in Fn(λ,2,λ)V[Gκ]) of size less than λ. Moreover, this set lies in V[G ], since j, G and G(κ) all do. Hence, m = j∗“G(κ) λ κ is a single condition in Fn(λ,2,λ)V[Gλ] = Q , and so if G is chosen λ S such that this master condition lies in G(λ), then we will indeed have j∗“G(κ) ⊂ G(λ). This is easy to arrange by modifying G(λ) to obtain G′(λ). Let q be the element of G(λ) with dom(q) = dom(m), and define ψ : Q → Q changing partial functions by switching λ λ their values between 0 and 1 on points where q and m disagree, that is, 0 if q(α) 6= m(α) and f(α) = 1 ψ(f)(α) = 1 if q(α) 6= m(α) and f(α) = 0  f(α) otherwise for all f ∈ Q and α ∈ dom(f). Clearly ψ is an (involutive) auto- λ   morphism of Q in V[G], whence G′(λ) = ψ“G(λ) is Q -generic over λ λ V[G ], and V[G ] = V[G ∗G′(λ)]. The “tail” generic Gλ+1 will of λ λ+1 λ course still be Pλ+1-generic over V[G ∗ G′(λ)]. Hence, considering λ P as P ∗ Q ∗ Pλ+1, we may take G′ = G ∗ G′(λ) ∗ Gλ+1, which λ λ λ is P-generic over V and satisfies V[G] = V[G′]. Moreover, G′(λ) contains the master condition m determined by G(κ) = G′(κ), so j∗“G′(κ) ⊂ G′(λ), and so j∗ is elementary from hHV[G′],G′(κ)i to κ+ hHV[G′],G′(λ)i, as required. ⊣ λ+ §3. Basic definitions. The definition for morasses that we use will be that of Velleman [12]. In particular, we retain the notation M.1–M.7 for axioms of a morass, used both there and in Devlin [4]. We also follow Velleman in separating out the following subsidiary definition, although we change the terminology slightly to emphasise the order preservation property. Definition 10. A function π : α → β between two ordinals is a successor, limit, zero and order preserving (SLOOP) function if • for all γ < α, π(γ +1) = π(γ)+1, and • for all limit γ < α, π(γ) is also a limit, and • π(0) = 0, and • for all γ < δ < α, π(γ) < π(δ). Note in particular that the composition of SLOOP functions will yield a SLOOP function. 8 ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN Morasses will be defined based on a set S of ordered pairs of or- dinals. For such an ordered pair x = hα,βi, let us denote α by l(x) (the level of x), and β by o(x) (the order of x). Also, note that we here use the word “tree” in the liberal sense where others might use “forest”: our tree will have many root nodes. Definition 11. For any uncountable regular cardinal κ, a (κ,1)- morass (or simply morass when κ is clear from the context) consists of: i. a subset S of (κ×κ)∪({κ}×κ+), and ii. A tree order p≺ on S, and iii. For every pair hx,yi of elements of S with x p≺ y, a function π : o(x)+1 → o(y)+1, xy such that the following conditions hold. Left-alignment: For each α ≤ κ, let θ = {β|hα,βi ∈ S}. Then α θ is in fact an ordinal. Moreover, θ = κ+, and for α < κ, α κ 0 < θ < κ. α Monotonicity: For x and y in S, x p≺ y implies l(x) < l(y). Commutativity: If x p≺ y p≺ z then π ◦π = π . yz xy xz M.1: For each pair x,y ∈ S with x p≺ y, π is an SLOOP func- xy tion, and π (o(x)) = o(y). xy M.2: Suppose x p≺ y ∈ S and ν < o(x). Let w = hl(x),νi and z = hl(y),π (ν)i. Then w p≺ z and π = π ↾ (ν +1). xy wz xy M.3: For all y ∈ S, {l(x)|x p≺ y} is closed in l(y). M.4: For all y ∈ S, if o(y) + 1 6= θ (as defined for Left- l(y) alignment above), then {l(x)|x p≺ y} is unbounded in l(y). In particular, if α is a successor ordinal, then θ = 1. α M.5: For all y ∈ S, if {l(x)|x p≺ y} is unbounded in l(y), then o(y) = {π “o(x)|x p≺ y}. xy M.6: Suppose x p≺ y ∈ S and o(x) is a limit ordinal. Let ν = S sup(π “o(x)) and let z = hl(y),νi. Then x p≺ z and π ↾ xy xz o(x) = π ↾ o(x). xy M.7: Suppose x p≺ y ∈ S, l(x) < α < l(y), o(x) is a limit ordinal, and o(y) = sup(π “o(x)). If xy ∀ν < o(x)∃γ(hα,γi p≺ hl(y),π (ν)i), xy then there is a γ such that hα,γi p≺ y. Note in particular axiom M.5, in the case where l(y) = κ: for any τ < κ+, we have τ expressed as the (increasing, by commutativity) union of the images of the maps π ↾ o(x) for x p≺ hκ,τi. Further xhκ,τi note that this isn’t just a variant of κ many things “adding up” to κ, and then τ being bijective with κ, but something more direct: the maps π are order preserving, and so the ordinals mapping xhκ,τi into τ must to some extent reflect the structure of τ. For example, if τ is a successor ordinal and so has a largest element, then for x p≺ hκ,τi with l(x) sufficiently large, o(x) must also have a largest element. Jensen’s orignal definition in fact called for even stronger LARGE CARDINALS AND MORASSES 9 preservation properties; see [4] for details. Also observe that for any x p≺ y in S, o(x) ≤ o(y) since by M.1, π is a strictly order- xy preserving function from o(x)+1 to o(y)+1. When we force morasses to exist in the presence of multiple large cardinals, it will be convenient for the sake of preservation of the large cardinal property to use a partial order which lends itself to homogeneity arguments (see Section 6 below). The upshot will be that the generic morass satisfies a useful property not possessed by the morasses obtained by forcing with Velleman’s partial order de- fined in [12]. We give here a name for the kind of morass that we obtain. Definition 12. Suppose M = hS,p≺,hπxyixp≺yi is a (κ,1)-morass. An ordinal α < κ is a mangal of M if for all x,y ∈ S with x p≺ y and l(x) < α < l(y), there is a z ∈ S with l(z) = α such that x p≺ z p≺ y. If the set of mangals of M is cofinal in κ, we say that M is a κ-mangrove, or simply mangrove when κ is clear from the context. Animmediatequestioniswhether theexistence ofamorassimplies theexistence ofamangrove. Itturnsoutthattheanswerisyes, aswe shall show in Section 5, using Velleman’s theorem that the existence of a morass is equivalent to a certain forcing axiom. Another natural question is what the set of mangals of a morass can look like. A first result in this direction is the following. Proposition 13. For any morass M, the set of mangals of M is closed in κ. Proof. Suppose that γ < κ is a limit point of the set of mangals of a morass M = hS,p≺,πi, and let x,y ∈ S satisfy x p≺ y and l(x) < γ < l(y). For each mangal α of M such that l(x) < α < γ, there is a z with l(z ) = α and x p≺ z p≺ y. But then by M.3, there α α α must be some z with l(z ) = γ and x p≺ z p≺ y; thus, γ is also a γ γ γ mangal of M. ⊣ In particular, if M is a mangrove, the set of mangals of M is a closed unbounded subset of κ. §4. Forcing morasses to exist. As is the case for many com- binatorial structures, one can force with a partial order of partial morasses to get a morass in the generic extension, as we shall show below. We use a partial order similar to those described in [7] and [12]. However, ourswill differ inthat we strengthen the requirements for an extension of a condition, ensuring that each condition of the generic goes up to a mangal of the ultimate morass. Naturally, this will make our generic morass a mangrove, whereas it is not hard to check that a generic morass for Velleman’s partial order in [12] will be far from a mangrove; indeed, it will be κ-branching at the node h0,0i, and so cannot have any mangals at all! However, many of the 10 ANDREW D. BROOKE-TAYLOR AND SY-DAVID FRIEDMAN details of our proof will remain essentially the same as in that paper. In particular, we repeatedly use the basic construction given there for extending morass conditions, tweaking it to fit the particular requirements in each case. 4.1. Definitions. Let x p≺ y denote that y is an immediate p≺- i successor of x. Definition 14. For any uncountable regular cardinal κ, a (κ,1)- morass condition consists of: i. a subset S of ((λ+1)×κ)∪({κ}×κ+) for some λ < κ, and ii. A tree order p≺ on S, and iii. For every pair hx,yi of elements of S with x p≺ y, a function π : o(x)+1 → o(y)+1, xy such that 1. Left-alignmentholdsforall α ≤ λ, andthe setS ={β|hκ,βi ∈ S} contains 0 and is closed under ordinal successors and predeces- sors. 2. Let f be the order-preserving bijection from ot(S) to S. Then ot(S) ≤ θ , and for each ν < ot(S), hλ,νi p≺ hκ,f(ν)i. λ 3. Monotonicity, Commutativity, M.1, M.2, M.3 and M.5 hold. Axiom M.4 holds for those y ∈ S such that l(y) ≤ λ. Axioms M.6 and M.7 hold for those x and y such that x p≺ y and i l(y) ≤ λ. Forour analysis, we will needto extend thenewnotionofa mangal to also be applicable to morass conditions. Definition 15. Let p = hS,p≺,πi be a morass condition. An or- dinal α ≤ λ is a mangal of p if for all x,y ∈ S with x p≺ y and l(x) < α < l(y), there is a z ∈ S with l(z) = α such that x p≺ z p≺ y. Ourdefinitionofamorassconditionisthesameasthatusedin[12]; our partial order will differ in the definition of ≤. For this reason, we continue to refer to these conditions as morass conditions, even though we shall call our partial order the mangrove forcing. Also note that, since it arises frequently and the meaning is fairly clear, we shall consistently abuse notation, writing x ∈ p to mean that x is an element of the set S for the morass condition p. A few comments about the definition of morass conditions are in order at this point. Note that requiring Axiom M.2 when l(y) = κ implicitly entails imposing the condition on S and the various π xy that for any x p≺ y with l(y) = κ, π “(o(x)+1) ⊂ S. Requiring M.7 xy for x p≺ y is the same as positing the non-existence of α fitting the i antecedent of that axiom, since x p≺ hα,γi p≺ y contradicts x p≺ y. i Observethatbyrequirement1above,ot(S)mustbealimitordinal; hence by requirement 2, θ > 1, and so by M.4, λ is a limit ordinal. λ Also, from M.2, requirement 2, and the fact that p≺ is a tree order, we get that for x p≺ y with l(x) = λ and l(y) = κ, π = f ↾ o(x)+1. xy