SEMISELECTIVE COIDEALS

ILIJAS FARAH

In this note we give an answer to the following problem of Todorcevic:

Find out the combinatorial essence behind the fact that the family JP of the
ground-model infinite sets of integers in a Perfect-set forcing extension has the
property that for any Borel/: [N]a ->{0, 1} there exists an AeJt such t h a t /
is constant on [A]m (see [7], [13]). In other words, one needs to capture the
combinatorial properties of the family Jf of ground-model subsets of M which
assure that it diagonalizes all Borel partitions. It turns out that the notion
which results from our analysis of this problem is a bit more optimal than the
older notion of a "happy family" (or selective coideal) introduced by A.R.D.
Mathias [16] long ago in order to extend the well-known theorems of Galvin-
Prikry [6] and Silver [25] (see Theorems 3.1 and 4.1 below). We should remark
that these Mathias-style extensions can indeed be as useful in the applications
as the original partition theorems. For example, one such application (where
the original partition theorem of Galvin Prikry and Silver does not seem to
fit) was recently found by Todorcevic ([28]) in order to supply a new proof of
the famous Bourgain Fremlin-Talagrand theorem ([2]). Other applications
can be found in the so-called parametrized partition calculus (see e.g.,
[17], [19], [29], [38]). One can also use these Mathias-style extensions of the
Galvin Prikry and Silver theorems to give a new proof of the well-known
perfect-tree theorem of J. Stern ([39], see also [38, §C]).
We have organized our paper so that it can be read by two kinds of mathe-
maticians, those unfamiliar with the forcing technique and the others. In §1
(on which the latter sections do not depend and which can be skipped by
readers not familiar with forcing) we prove that Mathias' notion of selective
coideal is too strong to be preserved by the Perfect-set forcing. In §2 we
introduce the notion of a semiselective coideal and compare it with the stronger
notion of a selective coideal. Sections 3 and 4 are independent from each other,
but they both rely on §2. In §3 we use topological methods to prove that a
coideal is semiselective exactly when the abstract Baire property and the Ram-
sey property, naturally associated with it, coincide (see Theorem 3.1 below).
In §4 we prove that a coideal Jt is semiselective if, and only if, the Mathias
poset Jfjr- associated with J f has the Prikry property and if, and only if, Ji'#>
has the Mathias property (i.e. subsets of a generic set are generic). We also
show that semiselective coideals are preserved by the Perfect-set forcing and
describe a situation in which every semiselective coideal diagonalizes all defin-
able sets.
We denote finite sets of positive integers by s,t,u,. . . and infinite sets of
positive integers by A, B, C, Symbols [N]<a> (resp. [N]01) denote families
of all finite (resp. infinite) sets of positive integers. The following notation will

[MATHEMATIKA, 45 (1998), 79-103]

80 I. FARAH

be useful:

A/s = A\{l,. . . , maxs},

A/n = A\{\,...,n).

The symbol s E t means that s is an initial segment of t. We consider two

topologies on [N]"\ the separable metric given by basic open sets

and the exponential whose basic open sets are

[s,A] = {BcM:s^B and B&
A family 34? of sets of integers is a coideal if its complement ^(N)\Jf is an
ideal; in other words, coideal is a family which satisfies the following two
axioms:
(10) F, 3 F 2 and F2eM? implies F^eM';
(11) if F, u F 2 is in J f then either F, is in JV or F 2 is in Jt.
For an AeJt we define

A set #•£ [N]"8 is Jf-Ramsey if for every [5, ^ ] such that ^ e^f there is Be^f f A
such that [J, 5] is either included in or disjoint from SC. It is M'-Ramsey null
if it is Jf-Ramsey but the set [s, B] as above is always disjoint from it. The
algebra of Jt-Ramsey sets and its ideal of ^f-Ramsey null sets are denoted by
&{#?) and &0(Jf) respectively. A s e t f c R x [^]° is perfectly 3f-Ramsey if
for every perfect P^U and every [s, A] such that AeJ4? there is a perfect Q^P
and a Bejf \ A such that Q x [s, B] is either included in or disjoint from 3C. It
is perfectly M -Ramsey null if it is #f -Ramsey but the set Q x [s, B] as above is
always disjoint from it. The algebra of perfectly Jf-Ramsey sets and its ideal
of perfectly Jf-Ramsey null sets are denoted by &>&(#?) and ^^0(J^) respec-
tively. Using these notions one can state Todorcevic's question avoiding the
terminology of forcing (see [19], [17]):
Which combinatorial properties of the family [N]m of all infinite sets of inte-
gers are responsible for the fact that all Borel sets are perfectly [N\a-Ramsey1
We will always assume that ideals are proper and they include the Frechet
ideal [f^] <ffl , i.e., that coideals are closed under finite changes and differ from
&>(N). With this convention, every coideal can be naturally considered as a
poset ordered by almost inclusion, s * . So we will adopt the terminology
of forcing and talk about dense open subsets, incompatible elements, maximal
antichains, etc., of JtiP, always referring to the poset < J?, £ *>. The same applies
to the poset 0> of all perfect subsets of reals ordered by the inclusion—this is
the so-called Perfect-set (or Sacks) forcing—and to the poset & x Jf with the
product ordering.
Let Fi ^ F 2 2 F 3 ^ . . . be an infinite decreasing sequence of sets of integers.
Then a set F^ is a diagonalization of this sequence if Fx/n^Fn for all neF^.
By A <01 we denote the set of all finite sequences of elements of A. A family
of nonempty perfect sets {Ps: se{0, 1}<<O} is a fusion sequence if for every 5
 SEMISELECTIVE COIDEALS 81

sets P,AO and PSA) are disjoint perfect subsets of Ps, and the diameter of Ps
converges to 0 as the length of s approaches infinity. The perfect set
u n p/r.= n u p.
J£{0,1}°* neco nea) se2"
is the fusion of this family. The ^-operation is denned for a family
[Fs: se N <m} of sets of reals by
<a
>)= U « 0 F/[n.
/N N

The author gratefully offers an acknowledgement to Stevo Todorcevic for help,

attention, and support while working on these problems, and to Alan Dow for
his interest in this work and for a conversation which led us to construct the
example given in Remark 4.1. He also thanks Ontario for a Graduate Schol-
arship and the Science Foundation of Serbia for partial support.

§1. Preserving selectivity. Recall the following well-known notion intro-

duced by Mathias in [16].

DEFINITION 1.1. A coideal J f s f N ] ' 0 is selective if every decreasing

sequence in M' has a diagonalization in Jf.

We note that in [16] selective coideals were called happy families. It is easy
to see that [N]m is an example of a selective coideal but there are less obvious
examples (see [16], [28]). One might be led to expect that selectivity is the
notion we seek by the following well-known results.

THEOREM 1.1 (Mathias). Analytic sets are J4?-Ramsey for every selective
coideal Jf.

THEOREM 1.2. If °U is a selective ultrafilter then in a Perfect-set forcing

extension it still generates a selective ultrafilter.

For the first, see [16; Theorem 4.3]. In an unpublished manuscript of

Mathias [42] one finds an interesting history of the second result which we
reproduce here following the suggestion of the referee. In the course of a
lecture at Berkeley in 1967 Harvey Friedman announced that in an extension
by one Sacks real, every infinite subset of N contains or is disjoint from one
in the ground model, and sketched a proof. In discussion after the lecture,
Prikry showed that Friedman's proof was defective and Mathias found a correct
one. Around 1971 Kunen observed in a letter to Mathias that the latter's proof
of Friedman's statement would establish that a selective ultrafilter in the ground
model generated one in an extension by one Sacks real. In reply to Kunen's
letter, Mathias remarked that Kunen's observation would imply that where a
is Sacks-generic over V and b is ^Vgeneric (see §4) over V for a selective
ultrafilter °U in V, then a and b are automatically mutually generic. Later on,
Miller ([17]) found this curious fact's elegant consequence: All Borel sets are
82 I. FARAH

perfectly Ramsey. In his letter Kunen also claimed, but later retracted, the
corresponding preservation result for arbitrary Sacks extensions (see [40;
p. 93]). Later work (see [1], [7], [26]) established the truth of that claim and
its connection to a deep combinatorial fact known under the name of Halpern-
Lauchli theorem (see Remark 4.1 below, also [13], [26; §6]).
It is not difficult to see that the standard proof of Theorem 1.2 gives that
in the Perfect-set forcing extension a selective coideal still generates a coideal
(see Corollary 4.2 below). However, the following lemma shows that selectivity
is destroyed whenever the coideal Jf has infinite cellularity.

LEMMA 1.1. If M" is a selective coideal which has an infinite antichain then
in the Perfect-set forcing extension it does not generate a selective coideal.

Proof. Assume that Jf has an infinite antichain {Xn}. We can further

assume that Xi's are pairwise disjoint (by replacing Xn with Xn =
Xn\\Si=\Xi), and that they are bijectively enumerated as B, (?e{0, 1 }<<0). So
we can recursively embed the binary tree <{0, 1}<O), c ) into <Jf, c > : let
As=\j Bt, for se{0, \}<l°.

Then we have the following:

(1) .4,2 4, for all ( E S ;
(2) A,r\As= 0 for all t^s of the same length.
Let r be a new real in a forcing extension of V, and in the extension let {Fn}
be the following sequence (recall that r \n is the sequence of first n digits
of r):
Fn = Ar[n.
Then this is a decreasing sequence of elements of Jtf. We claim that no ground-
model infinite subset of a> diagonalizes it, moreover

CLAIM. IfX^co is infinite and almost included in all Fn's, then re V[X].

Proof Let
T={te{0, \y°:XnA, is infinite},
This set is obviously downwards closed. It is infinite, because all initial seg-
ments of r belong to it. We claim that r is the unique infinite branch of T: if
xe {0, 1 } a is distinct from r and all its initial segments are in T, look at maximal
te T such that both x and r extend it. Then both X n A,* 0 and Xr\At*\ are
infinite, but this contradicts to the choice of X. So r is definable from X, as
required.

So in V[r] we have a sequence {Fn} in Jf with no diagonalization in jiff,

and therefore Jf is destructible. This proves Lemma 1.1.

We say that a family of sets of integers generates a coideal if its upwards

closure is a coideal.
 SEMISELECTIVE COIDEALS 83

THEOREM 1.3. A selective coideal is preserved by the Perfect-set forcing if

and only if it is generated by finitely many selective ultrafilters.

Proof. Assume Jf is a selective coideal which in a Perfect-set forcing

extension generates a selective coideal. Let j ^ / c ^ f be a maximal family such
that for all A # B in si:
(1) A n B is finite; and
(2) 3tf \ A has the f.i.p. (finite intersection property).
By Lemma 1.1, the family s/ is finite, say s#={A\,. . ., An). Then for every
CeJt there is Aes/ such that AnCeJV, and <sUi=J^\A, is a selective
ultrafilter on At for /= 1,. . . , n. Therefore $e is generated by \J;=\ *%i, as
required.
Let Jt= U"=i ^ , , where ^,'s are distinct selective ultrafilters. By Theorem
1.2 in a Perfect-set forcing extension all ^,'s remain selective. To see that Jf
itself is preserved, it is enough to see that the condition from Definition 1.1 is
preserved "locally", i.e., that it remains true on a dense subset of Jf. But this
obviously reduces to the fact that each °Ui is selective in an extension.

§2. Semiselectivity If 3>n is a sequence of families of sets of integers, then

a set Fx is a diagonalization of it if F^/n is in 3>n for all neF^. (In other
words, if there are sets Fne&n such that Fx is their diagonalization.)

DEFINITION 2.1. A coideal #P £ [N \° is semiselective if for every sequence

{3>n} of dense open subsets of ,W the family of all diagonalizations of this sequence
is dense in #C.

Thus, semiselectivity is a weakening of selectivity where we require cr-distri-

butivity instead of cr-closedness. Note that in [11] Kunen uses the term semi-
selective for a weakening of selectivity different from ours—he retains a-
closedness but relaxes the Q+ property (for definitions see the paragraph before
Theorem 2.3 below).

THEOREM 2.1. Let #? be a semiselective coideal. (a) The family of all sets
homogeneous for some partition o/[^J] 2 into two pieces is Jtf'-Ramsey, (b) Ana-
lytic sets are ^-Ramsey and perfectly M'-Ramsey. (c) The algebras of ^C-
Ramsey and perfectly $f-Ramsey sets are a-algebras closed under the si-
operation.

Proof (a) Let [N]2 = A: o u/:i be a partition. Define

{«} *A^K0 or {«} x ^ c ^ , } .
Then each Qsn is a dense subset of 3ff, so we can pick a set As^f such that
A/ne3>n for all n; by definitions, this means that A is min-homogeneous, i.e.,
that there is a sequence ;„ (neA) such that {«, m)eKin for all n<m in A. Let
Ao= {neA: /„ = ()} and Ax = {neA: im= 1}; these two sets are homogeneous and
one of them is in Jf.
84 I. FARAH

Parts (b) and (c) will be proved in Sections 3 and 4, by using different
methods, but we will prove special cases of (b) and (c) now to illustrate typical
applications of semiselective coideals. Applications of the following theorem,
(i.e., its version when jf = [^J]<1)), are found in several areas of mathematics,
like better quasi-ordering theory ([18]), partition calculus ([6]), convergence
theory in function spaces ([23]) and Borel graph theory ([10]). Recall that a
family J* of finite sets is a barrier on A if every infinite subset of A has an
initial part in J* and J*" is an antichain in the c ordering. For a set B<^A let
J^ \B denote the family J* n [B]<co. The following theorem was proved by
Nash-Williams and Galvin in the case when JF = [N]O>, and by Mathias in the
case when J f is selective.

THEOREM 2.2 (Semiselective Nash-Williams; Galvin Lemma). Let Jf be

a semiselective coideal. If' 3F is a family of finite subsets ofN, then there is a
such that ^ \B is either empty or it includes a barrier.

Proof. We say AsJF accepts s if every element of [s, A] has an initial

segment in J5', A rejects s if no element of [s, A]n J^f accepts s, and A decides
for s if A either rejects or accepts s. The following statements are obvious.
(1) If A accepts (rejects) s then every B in Jf \A accepts (rejects) s.
(2) For all s, A there is a B in Jf \ A which either accepts or rejects s.
(3) If A accepts s then A accepts J U { « ) for all n in A/s.
An analogue of (3) for "rejects" is easily seen to be false—A can reject s while
accepting s u {«} for some neA/s (e.g., if J* = {t: net}).

CLAIM 1. If A rejects s, then the set ofneA such that A accepts s u {«} is
not in Jt.

Proof. Suppose the contrary, that the set B of all such n is in Jf. Then
B accepts s, contrary to our assumption and (1).

CLAIM 2. There is a B in Jf which decides for each of its finite subsets.

Proof. Let
:C decides for s}.
Then sets Q)n = (Lax *•<« ^ are dense open in Jf, so if B is their diagonalization
then by (2) above it decides for all of its finite subsets.

Let B be as in Claim 2 above. If B accepts 0 then & f B includes a barrier,

so we can assume B rejects 0. For s^B such that B rejects it let
Ss={CeH\B: Crejects s u {n} for all neC/s},
and let <fs = J f \B otherwise. By Claim 1 sets Ss are dense below B, so let
C e ^ f \B be a diagonalization of {Ss}. Then it follows (by induction on the
length of s) that C rejects all of its finite subsets; in particular every .s £ C is
not in J% so 3F \ C is empty. This finishes the proof.
 SEMISELECTIVE COIDEALS 85

We will now prove a parametrized version of Theorem 2.2. For an

xe {0, 1 }w and ke hi by JC ["k we denote the sequence of first k digits of x. For
a perfect set P c {0, 1}*" by 7V we denote the set of a\\x\k for jceP and keN,
and by 7> \k we denote the set of all ueTP of length k; it ueTP\k then we
write \u\=k. The space (0, I}1" is considered with the metric d(x,y) =
l/(A(x,y)+ 1), where A(x, y) is the minimal k such that x \k^=y \k.

T H E O R E M 2.3. Let JC be a semiselective coideal. For every 3P'<=

{0, 1 } < < B X [ N ] < < B there is a perfect P^{0, I} 00 and AeJt so that one of the
following two alternatives applies:
(a) for all xeP and all infinite B<^A there are integers k, I such that the pair
(x\k,Bn{l,..., /}> is in &; or
(b) the set 7>x [A]<0> is disjoint from &.

LEMMA 2.1. If 3>n (ne N) are dense open subsets of 3P* $f then the set
®X = {(P, i ) : ( P n [ « ] , A/n>e@nforallneA andueTP \n}
is dense open in SP x Jf.

Proof. For a set £ £ & x j f let 7T2(<f) denote the set of aWAeJf such that
<P, ^> is in S for some Pe^ 9 . Note that n2{2n) is a dense open subset of J f
for all «. Find a maximal antichain J < / | C J 2 ( S | ) and for each Aestf\ fix a
perfect set P = P< }(A) such that <JP, A/I) is in ^ , . Then let S>'2 be a family
of all <g, B)e&2 such that fi^^ for some Aes/i and g c f ( i ) . Note that
this A=A(B) is unique, if it exists. Find a maximal antichain jrf2 included
in K2(3''2), and for each Be^2 find perfect sets Pm(B), P<i>(5) of diameter
^ 1 / 2 included in P< y{A(B)). Proceeding in this way, we define maximal anti-
chains s/n in Jt and for each Aes/n disjoint perfect sets PS(A) (se{0, l}"^ 1 )
such that:
(1) <Ps(A),A/n) is in £>„;
(2) the diameter of PS(A) is at most 1/n; and
(3) if A<=Bfor AestfnandBe,9i?m (n>m) then PS(A)^P,(B) for all S3t.
Let
$„ = {Ce j f : C is almost included in some member of s/n}.
Sets $„ are dense open in jf, so they have a diagonalization ,4^ in ^f. Then
for each n there is Anejrfn such that ^ K diagonalizes the sequence {An}, and
sets /*" = Ps (An) form a fusion sequence such that TP»s \ n has only one element.
Let Prc be the fusion of {/*"}; then for each «6v4 and ueTP \n we have ? such
that

so the pair </" x , /ioo> is in ^oo.

Proof of Theorem 2.3. For a pair (u,s}e{0, 1} <<U , perfect Q ^ { 0 , I}* 0 ,

and v4 6 ^ we say (Q, A) accepts <M, i> if ueTQ and for every xeQn [u] and
5 e [ i , ^ ] there are integers k, I such that {x \k, Bn {I, . . ., l}}e^. We say
(Q, A) rejects <w, s> if </\ S> does not accept <M, i> for all perfect PsQ and
86 I. FARAH

\A, and that (Q, A} decides for <w, s} if it either accepts or rejects it.
The following statements are obvious.
(1) If <g, A} accepts (rejects) <«, s}, then every </>, B} such that P is a
perfect subset of Q n [M] and BeJt \ A also accepts (rejects) <«, .?>.
(2) For all <w, s>, ^ and g there is a perfect P^Qn[s) and a j5e«# M
such that </>, fi> decides for <w, J>.
(3) If <2>^> accepts {u,s} then it also accepts <H, ,?U{H}> for all

(4) If <2>^> rejects <w, J> then the set of all neA/s such that <(?, A}
accepts <«, J u {«}> is not in #C
(5) If (Q, A} accepts (rejects) <«, s} then it also accepts (rejects) (v, s) for
all vsu in !Te.

CLAIM. There is a pair (£), A} which decides for every <«, s}eTQx [A]<<!>
such that max (s)<\u\.

Proof. Let
S>»= {<Q, A): <Q,A) decides for all <w, s ) e r e x [/4]<ffl
such that max (s) ^ | u\ = «}.
Then (2) implies 3n is a dense open subset of ^ x Jf, so let <g, A) be as
guaranteed by Lemma 2.1. Then (Q, A} is as required, because </*, ^4> decides
for <w, s) if, and only if, ( ? n [M], y4> does.

Let (Q, A} be as in the Claim. If it accepts < < >, 0 > , then it satisfies (a),
so we can assume it rejects < <>, 0 > . Define sets SUJS for <M, s}eTQx [A]<0>
as follows:
§UJS= {(P, B)e0>xjf \A: P^Qn [u] and (P, B/n) rejects <M, J U {«}>},
if <g, ^> rejects <w, J> and gu,s = ^ x Jf otherwise. By (4) the sets

are dense open in Jt \A, so let (P, B} be as guaranteed by Lemma 2.1. We

claim that <P, 5> satisfies (b). Assume the contrary, and let
<M, j>eJ* n (7>x [5] <a) ) be such that max(j) is minimal (taking max(0) =
0). Then ( f n [u], B} accepts < < >, s}, so in particular J # 0 . Let n = max(s)
and t = s\{n}. Then if o c u is of length max(/), (P, B} rejects <u, t}, therefore
<Pn [M], B} is not in <?„,,—a contradiction.

Now we will investigate the relationship between selective and semiselective

coideals. We say that a coideal Jf is Ramsey if it diagonalizes all two-
dimensional partitions of its elements, and that it has the Q+-property if for
every As3f and every partition of A into finite sets there is a Be Jf \A which
intersects every piece of the partition in at most one point. It follows from the
definition that JC is selective if, and only if, it is cr-closed and has the Q+-
property, and that jff is semiselective if, and only if, it is <r-distributive and
 SEMISELECTIVE COIDEALS 87

has the (2+-property. Also, Ramsey coideals always have the g + -property.
[For every partition of N into disjoint pieces there is a partition of [N]2 such
that every homogeneous set is either included in one piece or it intersects each
piece in at most one point.] So we have the following implications:
selective => semiselective => Ramsey.
Examples 2.1 and 2.2 below demonstrate that none of these arrows is reversible
(and it is well-known that in general Q + -property does not imply Ramseyness).
On the other hand, by [16; Theorem 0.10], an ultrafilter % is selective if, and
only if, it is Ramsey if, and only if, every partition of [f^l]2 into two pieces has
a homogeneous set in <%, so we have:

THEOREM 2.4. An ultrafilter is Ramsey if, and only if, it is selective and if,
and only if, it is semiselective.

Example 2.1. A semiselective coideal which is not selective. Our working

copy of M will be N x |^J. Let J be the ideal of all subsets of N x P*J having
finite intersections with all vertical sections {«} x fo|. We claim the coideal Jf =
^(N)\J is semiselective, yet not selective. To see that Jf is not selective,
consider the sets
An = [n, oo)xN.
All An's are in Jf, but every diagonalization of the sequence {An} has to be in
J. On the other hand, J is a nowhere dense ideal, i.e., every Ae#f has an
infinite subset B such that J \ B equals the Frechet ideal; therefore J is
semiselective.

Example 2.2. A Ramsey coideal which is not semiselective. A generating

family for this coideal is {Au: u is a finite sequence of reals} such that ^4<> =
N and
Au*r(l;eM) is an almost disjoint family of subsets of Au and for
every analytic J>/E[^4 M ]' U there is £ so that [^4UA,;]'0 is either
disjoint from or included in s/.
Such family is constructed by the recursion as follows: Let s/% (<§eR) be an
enumeration of all analytic subsets of [N]a. If Au is chosen, pick an almost
disjoint family B^ ( ^ e R ) of subsets of Au. Using Silver's theorem for every
%eM find ^ / § 9 ^ which is either disjoint from or included in j / % . By the
construction, the family {Au} generates a Ramsey coideal. Dense open sets Q)n
generated by {Au :\u\=n) verify that this coideal is not CT-distributive.
Example 2.2 can be modified to obtain semiselective coideals with special
properties; see e.g., Example 4.1.

§3. The abstract Baire property. In [19] Pawlikowski used the work of
Marczewski ([35]) and Morgan ([36]) on the abstract Baire property to analyse
the algebra of perfectly [N]ra-Ramsey sets. In this section we will apply his
methods in the context of semiselective coideals. We should note that [19]
88 I. FARAH

originated in a challenge to find elementary proofs for the results in [17] about
perfectly Ramsey sets.
Throughout this section we assume #P is a semiselective coideal, and 3? will
denote the family of all perfect (nonempty) subsets of a Cantor cube {0, \}a.
For a family of sets of integers s/ we define a family Exp(,s/) by
ExpCO = {[.?, A]:Aes/}.
We always assume N is in s4 so that Exp(stf) extends the standard basis for
the separable metric topology on [N]01. The family Exp(^f) for a coideal j f
is rarely a basis for some topology, as the following easy fact shows.

F A C T 3.1. Exp (stf) is a basis for the topology if, and only if, for all A,
Best? which are not almost disjoint there is a Cestf included in their intersection.

DEFINITION 3.1. Let @) be a family of subsets of some set Y. A set 2£<~ Y:

is ^-open if it is equal to the union of some subfamily of 0&;
is ^-nowhere dense if every U in & has a subset in & disjoint from 3L';
has the abstract ^-Baire property if every Ue& has a subset in i% which is
either included in or disjoint from 9C; and
has the ^-Baire property // there is a 3$-open set & such that 3Ct±(9 is 3H-
nowhere dense.
It is easy to see that the ^-Ramsey property implies the abstract Exp( Jf )-
Baire property, which in turn implies the Exp( JC )-Baire property. We will see
that in the case when ^C is semiselective we can say more. By & x Exp(Jf)
denote the family of all subsets of IR x [N]® of the form P*[s,A] for a perfect
Pand [s, A]eExp(Jff).
In the case when Jt = [N]0>, (b)-(d) of the following theorem were proved
by Ellen tuck in [3], (e)-(g) were proved by Pawlikowski in [19], and in the
case when Jt is a selective ultrafilter (b)-(d) were proved by Louveau in [14].

THEOREM 3.1. Let M" be a coideal. Then the following are equivalent.
(a) J f is semiselective.
(b) The Jf-Ramsey sets are exactly the sets with the abstract Exp( Jf )-Baire
property.
(c) All Exp(Jf )-nowhere dense sets are M'-Ramsey.
(d) The following three families of subsets of[H]m coincide:
3tf -Ramsey null sets;
Exp(Jf )-nowhere dense sets; and
Exp(J^)-meager sets.
(e) The perfectly Jf-Ramsey sets are exactly the sets with the abstract
0> x Exp(^f )-Baire property.
(f) All 0> x E\p(Jf)-nowhere dense sets are perfectly M'-Ramsey.
(g) The following three families of subsets oflN]01 coincide:
perfectly M'-Ramsey null sets;
2? x ExpiJtf7 )-nowhere dense sets; and
3P x Exp(jf )-meager sets.
 SEMISELECTIVE COIDEALS 89

LEMMA 3.1. If ^C is semiselective, then the ideal 0lo(Jtf') is a a-ideal and

the algebra fftiJC) is a a-algebra.

Proof. Let s/n be a sequence of j>f-Ramsey null sets and fix [t, B] for
BeJtf; we can assume t = 0. For ,se[N ] <ffl define dense open subsets of Jf
@s={Aeje: [s, A] n s/n = 0 for ail n^\s\}.

In C is a diagonalization of the sequence &s {i.e., a diagonalization of the

sequence S>n=f]mMS<n 3>n) then [0, C] is disjoint from {J<s/n.
Let s/n be a sequence of Jf -Ramsey sets and fix [t, B] for BeJC Then
either there is CEJ#* \ B such that [t, C] is included in some $in or a proof
similar to the above gives CeJf \B such that [?, C] is disjoint from [Jstfn.

LEMMA 3.2. If Jf is semiselective, then the ideal 0>{%o(3tf>) is a a-ideal and

the algebra ^ 2 ( J f ) is a a-algebra.

Proof. Let j?/n be a sequence of perfectly ^f-Ramsey null sets and fix
{P, [t, B]} in 0>x Exp(Jf). We can assume t = 0. For 1se[N]<0) and ueTP
so that max(.s) ^ | u \ = m define a dense open subset of ^x Jit
3>s= {(Q, A): AeJf and Q*[s, A] n jtfn = 0 for all
then an application of Lemma 2.1 gives us <g, AyeSd^ so that the set Q x [?,
avoids U-^n-
Let j?/B be a sequence of perfectly jf-Ramsey sets andfix[t, 2?]eExp
and Pe0>. Then either there are CeJt? \B and a perfect g ^ p such that
Q x [?, C] is included in some sin or a proof similar to the above gives CeJt \B
and a perfect QsP such that g x [t, C] is disjoint from {Js/n.
LEMMA 3.3. If JV is semiselective, then the ffl-Ramsey property is equiva-
lent to the abstract Exp(,W )-Baire property.

Proof. Let 6 be a set with the abstract Exp(J>f )-Baire property. Fix [t, B]
such that BeJt. Assuming t = 0 makes this proof only notationally simpler,
therefore we will do so. For ^e[M]<fl) let
[s, B]^(9 or [s, B]n(9 = 0
or [s, C] is neither included in nor disjoint from (9 for all Ce Jf \ B).
Then each ids is dense open so let CeJf \B be such that C/se3ss for all J S C .
Let J S be the set all all s such that [s, C]s(9 and let J^, be the set of all s
such that [s,C]n& = 0. If J% \C(J5", f C) includes a barrier on C, then
[ 0 , C ] £ 0 ( [ 0 , C ] n C = 0 , resp.), so by Theorem 2.2 we can assume [C]<0>
is disjoint from J^o u f | . But by the abstract Baire property of (9 there is an
se[C] <<o andDeJf fCso that [s,i)]c© Or [s,D]n(9 = 0, s o i e f i u f , -
a contradiction. Therefore (9 is Jf-Ramsey. The other direction is trivial.

LEMMA 3.4. If Jf is semiselective, then the perfectly JF-Ramsey sets are

exactly the sets with the abstract 3P x Exp(jf )-Baire property.
90 I. FARAH

Proof. Assume & has the 0> x Exp(Jf )-Baire property. Fix [t,A]e
Exp(Jf) and Pe0>; without a loss of generality t = 0. Forse[N]<o> and ueTP
let
) : BeJf, £>£P is perfect, Qx[s, B]^& or Qx [s, B]nO =
or R x [s, C] is neither included in nor disjoint from G
for all CeJf\B and perfect /? £ Q],
= D ^-
w e 7"
max

By Lemma 2.1 let C e j f t # and a perfect R^P be such that </?n[w]

C/s)e@n for all w e ^ f« and J £ { 1 , . . . , « } . Let

If </>, C> and J^o O^i) satisfy (a) of Theorem 2.3 (with A replaced by C)
then P x [ 0 , C] £ 0 (P x [ 0 , C] n ^ = 0 , resp.), so we can assume (b) of this
theorem applies, and J% u J^i is disjoint from TP x [C]<ffl. But by the abstract
^ x Exp (^f )-Baire property of (9 this is impossible, therefore (9 is perfectly
Jf-Ramsey. The other direction follows by the definitions.

LEMMA 3.5. If' #F is not semiselective, then there is an Exp(Jf)-nowhere

dense set (9 which is not Jf-Ramsey.

Proof. Let s/n be a sequence of maximal antichains in Jf with no diagonal-

ization in Jf. Let & be the family of all diagonalizations of this sequence; it
is not ^-Ramsey because it is disjoint from Jf, yet every element of Jf has
an infinite subset in (9. On the other hand, (9 is Exp(^f)-nowhere dense: For
any [t, A]eExp(Jf) we can find neA such that A/n is not in s/n. Pick Bes4n
so that BnAeJf and m>n in A/B; extend t to J SO that n,mss. Then
[s, A n B] is disjoint from 0.

Proof of Theorem 3.1. We have proved (a) => (b), (a) => (e) and (c) => (a),
and the implications (d) => (c), (g) => (d), (f) => (c), (e) => (b), are obvious.
(b) => (c). If a set is both Exp(Jf )-nowhere dense and ^-Ramsey, then it
has to be Jf -Ramsey null.
(e) => (f). Similar to (b) => (c) above.
(a) => (d). If Jf is semiselective, then by Lemma 1 Jf-Ramsey null sets
form a cr-ideal, and by (c) all Exp(Jf)-nowhere dense sets are Jf-Ramsey;
therefore Exp(Jf )-nowhere dense sets form a cr-ideal, and they coincide with
Exp(jf )-meager sets.
(a) => (g). This proof is analogous to the proof of (a) => (d) above.

To prove Theorem 2.1, we will make use of the following classical result
whose proof can be found e.g., in [30], [31] or [28]. If 3b is a cr-algebra of
subsets of [N]w and 380 is its a-ideal, we say that a pair (JSi, &0} is a Marczewski
 SEMISELECTIVE COIDEALS 91

pair if for every SC£ [N f there is <P(^) 3SC in J 1 such that <W £<D(#*)Vf is in
^ if, and only if, it is in ^ o -

THEOREM 3.2 (Marczewski). ^4« algebra & which together with one of its
a-ideals forms a Marczewski pair is closed under the srf-operation.

Proof of Theorem 2.1(b). By Theorem 2.2 open sets are Jf-Ramsey, so it

suffices to prove that 3$(JP) is closed under the ^/-operation. The following
is a version of the proof given by Todorcevic in [28, Section 9] for the case
when #e is selective. The first step is to prove that the Continuum Hypothesis,
CH, implies (<%(jtf), <M0(Jf)} is a Marczewski pair. The second step makes
use of the following well-known result of Platek ([34]):

THEOREM 3.3. The use of the Continuum Hypothesis can be eliminated

from the proof of any statement involving only quantification over the reals and
possibly some fixed set of reals as a predicate.

CLAIM. If CH holds and J f is a semiselective coideal, then (@(Jf),

3#o(3f)} is a Marczewski pair.

Proof. First observe that Lemma 3.5 implies the following:

(M) If [sn, An](neN) and [t, B] are in Exp(Jf) and the set
[t, 2?]\U»eN [$„, An] is Exp(Jf)-nowhere dense, then for
some N the set [t, B] n [sn, An] is not Exp(^f)-nowhere dense.

Since CH implies that size of Exp(^f) equals Xi, (M) says that, in the terminol-
ogy of [19], Exp (Jf) is M-like. Therefore, by Lemma 3.1 and [19, §2] the pair
^ 0 ( ^ ) > is a Marczewski pair.

Now fix Jf-Ramsey sets Fs (se N <<o ). Statement "st?{Fs) is not Jf-Ramsey"
is false under CH by the Claim and Marczewski's theorem, and it is of the
form to which Platek's theorem applies. Therefore the set s4{Fs) is Jf-Ramsey,
as required.

Remark 3.1. A natural question is whether one can prove

iM0(^)) is a Marczewski pair without the use of an additional set-theoretic
axiom such as CH. The answer to this question is negative: There can be a
selective coideal without this property. More precisely, Alan Dow observed
([33]) that under some additional Set-Theoretic axiom (see [37, §8]) there is a
closed subset X of the remainder of the Cech-Stone compactification of the
integers, N*, and two relatively open disjoint subsets U, V of X such that
whenever U' and V are open subsets of N* such that U' nX= [/and V nX =
V, then U' and V are not disjoint. This X can moreover be chosen so that
the corresponding coideal #f=\)x'\% selective, and it is not difficult to use U
and V as above to construct a subset of 0>(H) witnessing that
is not a Marczewski pair.
92 I. FARAH

Now we turn to the perfectly Jf -Ramsey part of Theorem 2.1.

Proof of Theorem 2.1(c). In Theorem 2.3 we proved that open sets are
perfectly Jf-Ramsey, so it remains to prove SP^iJf) is closed under the .co-
operation. The scenario is analogous to one used in the proof of the part (b)
above.

CLAIM. If CH holds and Jf is a semiselective coideal, then

'o(Jf )> is a Marczewski pair.
Proof Using Pawlikowski's terminology introduced before, & is M-like
(by [19, Lemma 28]) so since Exp(Jf) is M-like and the sizes of & and Exp(Jf)
are equal, 3P x Exp(Jf) is M-like as well ([19, Lemma 2.7]) therefore by Lemma
3.2 and [19, §2] the conclusion follows.
By Marczewski's theorem and the Claim, CH implies the desired conclusion.
By Platak's theorem, the use of CH can be eliminated from this proof.
Note that the Exp (Jf )-Baire property implies the abstract Exp (Jf )-Baire
property only when Jf is an ultrafilter or when its dual ideal is nowhere dense—
i.e., for every Aejf there is a BsJf \A such that [B]a is included in Jf. This
is because if there are disjoint A and B in Jf such that for every CeJf \B the
set [C]'° is not included in Jf then the set
0 = U {M u CT:C<=fiand C£jf}
is Exp (Jf )-open, yet not Jf-Ramsey, since for all CeJf\B the set [0, C] is
neither disjoint from nor included in &.
We will now characterize those Jf for which there is a family Jf'SJf such
that the Exp (Jf')-Baire property is equivalent to the Jf-Ramsey property. Let
Jf be a semiselective coideal. Consider the following statements.
(El) Jf is Ellentuck: There is a dense Jf'^jf such that Exp (Jf') forms
a basis for a topology.
(E2) There is a dense Jf'^Jf such that the Jf-Ramsey property is equiva-
lent to the Exp (Jf')-Baire property.
(E3) There is a dense Jf'^Jf such that all A, BeJf' are either almost
disjoint or their intersection is in Jt.
THEOREM 3.4. The statements (E1)-(E3) are all equivalent.
Proof. (E3) o (El). By the Fact 1 above, the family J f ' u {N} verifies
(El) if, and only if, family M" verifies (E3).
(E3) => (E2). Let ^f" be as in (E3); we claim that it verifies (E2) as well.
Obviously all Jf-Ramsey sets have the Exp (Jf')-Baire property, so we can
concentrate on proving the other implication.
CLAIM 1. Exp (3tf")-open sets are Jf-Ramsey.
Proof Let & be an Exp (Jf')-open set andfix[/, B] for Bejf. For s3t
define a dense open subset of Jf by
for all CeJf \A}.
 SEMISELECTIVE COIDEALS 93

By the semiselectiveness, there is a CeJf" \B such that Ce2s for all sst
included in C. We claim that (9 is relatively open on [t, C] even in the standard
metric topology, more precisely if we let & = {s: [s, C] ^0} then (9 n [t, C] =
\Jse.r [s] ^ [U C]. Only " c " requires a proof, so pick De(9 n [s, C]. Then
there is a [M, /)'] ^0 such that D'eJtf" and Z)e[w, £>']; but by our assumption
on jtf" the set Cn D' is in Jt and therefore weJ*, SO De{Jse^[s], as required.
Since by Theorem 2.1 open sets in the metric topology are Jf-Ramsey, this
ends the proof.

Let ,s/ be a set with the Exp( Jt ')-Baire property. Then there is an Exp(Jt ')-
open set (9 such that ,a/A(9 is Exp(Jf')-meager; but then ,4A0 is Exp(^f')-
nowhere dense and the Jf -Ramsey property of s/ reduces to the ^-Ramsey
property of (9, i.e., to Claim 1.

(E2) => (E3). Assume tf" satisfies (E2) and enumerate 3/t" as {Ai)^<x-
Then recursively pick a sequence B^(^<X) such that for all §< 77<X we have
(1) BseX",
(2) B^As,
(3) B^^*B/: or fi,, and 5 5 are almost disjoint.
Set £„ is constructed by applying (E2) to sets &={Ji<n[B^]co and
[0,An]. Then M"'={B^<X satisfies (E3).

COROLLARY 3.1. Every selective ultrafilter is Ellentuck.

We can say slightly more, namely

LEMMA 3.6. If in M' there are no uncountable antichains then tf is

Ellentuck.

Proof. In the case when 3tf \A does not have the f.i.p. for all AeJtf the
proof proceeds by transfinite induction and the construction of 3tf" as in (El)
is very similar to that of Example 2.2. To prove Lemma in the general case,
let
3>o= {Ae^f: Jf\Ahas the f.i.p.},
3>i = {Ae Jt?: no set in J>F f ^ i s i n ^ o } -
Then !3ou3>i is dense open in Jf, so we can pick an almost disjoint family
A,, (n < a) of sets in Jf so that each s/n is in &0 u 2\. Then by the above and
Corollary 3.1 there is a dense Jf'„in 2? \An for each n which satisfies (c); let
M" be the union of all jtf"n's. It is easy to check that Jf' is as required.

LEMMA 3.7. The Continuum Hypothesis implies that every semiselective

coideal Jf whose complement is a P-ideal is Ellentuck.

Proof. This follows by [8; Theorem 2].

94 I. FARAH

§4. Semiselectivity and forcing. The reader is referred to [12] for the gen-
eral theory of forcing, and to [32] for Perfect-set (or Sacks) forcing. Recall
that if Jf is a selective coideal, then forcing with (Jf, £*> adjoins no new
reals to the universe and its generic filter is a selective ultrafilter. We will prove
that semiselective coideals still retain this desirable property. From now on,
yC will always denote a semiselective coideal.

LEMMA 4.1. The following are equivalent for a family Jf of sets of integers.
(a) Jf generates a semiselective coideal.
(b) Forcing with <Jf, <= *> adjoins no new reals and its generic filter is a
selective ultrafilter.

Proof. (=>). By our definition, the poset <Jf, £*> is cr-distributive, so it

adds no new reals. Let ^U be a name for a generic subset of JP; by genericity it
is an ultrafilter, and the semiselective version of Nash-Williams, Galvin lemma
implies that every partition of [N]2 into two pieces has a homogeneous set in
<W. By Theorem 2.4, <% is selective.
(<=). If #f does not generate a semiselective coideal, then there is a sequence
{s/n} of maximal antichains in (Jif, £*> with no diagonalization in Jif. In
the forcing extension by <^f, £*> let An be the unique element of s/n n 'W;
then the sequence {An} has no diagonalization in <W, and therefore "11 is not
selective.

The Jf-Ramsey part of Theorem 2.1 (b) and (c) easily follows from Lemma
4.1 by an absoluteness argument (see [16; §4]), but we choose to give a more
informative proof here. We need some definitions first.

DEFINITION 4.1. If J? is a coideal then let Jt.# be the Mathias poset

associated to Jf. Conditions are pairs <s, A > such that max s < min A and A e J^.
The ordering is defined by
(s,A}^(t, B) if, and only if, S3t, A^B and s\t^B.

If Jf is an ultrafilter poset M^ is denoted by ^ V and called the Prikry poset.

Poset Jl' x> has the Prikry property if for every sentence of the forcing lan-
guage q> and every condition <5, A} there is a BeJtf \A such that (s, B} decides
<p. It has the Mathias property if every subset of an Jl'^-generic set is M#-
generic.

Let % be a canonical name for the <^f, £*>-generic ultrafilter. Then it is

a routine to check that the poset Ji^ is equivalent to the two-step iteration of
posets <Jf, £*> and 0>4, hence it is of the form cr-distributive * ccc.
In the case when Jf is a normal ultrafilter on a measurable cardinal, (J?2)
of the following theorem was introduced and proved by Prikry in [20] and
{JtV) was introduced and proved by Mathias in [15], while (JtT) and ( ^ 3 )
for a selective coideal on M were proved by Mathias in [16]. As far as we
know, even the equivalence of {J12) and (^#3) below is new.
 SEMISELECTIVE COIDEALS 95

THEOREM 4.1. For a coideal #? the following are equivalent.

{MY) 2f is semiselective.
{Ml) M # has the Prikry property.
{M3) Mjr has the, Mathias property.

Proof of {MY) => {Ml). Assume j f is semiselective, and fix a sentence of

the forcing language <p and a condition <j, A > in M#•. For 13 s let 2>, be the
set of all Be J4? \ A such that either <7, B) decides <p or <?, C> does not decide
q> for a\\Ceje\B. Sets 3>t are dense open in <Jf, s*>, so let BeJf\A be
such that B/negi, for all wei? and a l W s { 1 , . . . , « } . Let
&rl = {/a^: (t, B} forces <p},
^ 2 = {t3s: (t, B) forces—up}.
Then let CeJtf \ B be as guaranteed by the semiselective Nash-Williams, Galvin
lemma as applied to to J^i and to 3Fi. We claim that <J, C> decides (p. To
prove this, it suffices to check that if <7i, Ci> and </2, C2> are different exten-
sions of (s, C> deciding <p, then they both decide it in the same way. So
assume the contrary that <?i, C{) forces cp and <?2, C2) forces its negation.
Then t\e^F\ and ^ e J 2 ^ , and both are finite subsets of C. This together with
the choice of C means that every infinite subset of C has an initial segment
both in J^i and in J ^ ; so we can pick two compatible conditions one of which
forces 0 and the other forces its negation—a contradiction. So (s, C> does
decide (p.

Proof of {MY) => {MV). Assume 3fC is semiselective. In Lemma 4.2 below
we prove a statement which immediately implies {MY). For X^N let

LEMMA 4.2. The filter G{X) is ^-generic if, and only if, the filter GM{X)
is M.»-generic {over a model of a large enough part o / Z F C , of course).

Proof. {=>). Follows from the decomposition of Mathias forcing M^e as

(<=). For a dense open subset S of My? let 3>{8) be the family of all A e J f
such that the family

includes a barrier on A.

CLAIM. If $ is dense open in Mx- then 3i{$) is dense open in <Jf, £*>.

Proof. Pick (s, A)eJf, and let 2), (for tss) be the set of all Be JV f A
such that either <f, B} is in <f or <?, C> is not in S for all C6.?f t 5 - Sets 9,
are dense open in (jf \A, ^ * > ,
96 I. FARAH

so let Bed? \A be their diagonalization. By the semiselective Nash-Williams,

Galvin Lemma applied to the family

obtain Cejf \ B such that either C is in Q){8) or no finite subset of C is in 'S.

If the second possibility applies, then we find </, D}eS which extends <0, C>.
Let « = min C/t and DeQ), be such that B/n^D; then we must have (t, B)e£
and therefore te'S—a contradiction. So the first possibility applies here and
C is in

If G(X) is Jf-generic, then by Claim for every dense open subset § of M*

there is a Be3){£) n G(Z) such that X ^ A The family J*" denned as above
remains a barrier in any forcing extension (because well-foundedness is abso-
lute) so there is a tcX such that (t, A) is in S. But </, /!> is in GM(X), so
is generic.

Proof of ~\ {J?X) => —i (^#2), ~i (M3). Assume Jf is not semiselective, and

let s/n be a sequence of maximal antichains in <jf, c*> such that no element
of Jf is a diagonalization of this sequence. Let X be a canonical name for an
^#jf<-generic subset of N, and let An be a name for the (unique) element of sfn
such thatX<=*An.
To see that (J/3) fails, we claim that X is forced not to be a diagonalization
of this sequence. Suppose the contrary, that <7, B) forces
X/n^An for all neX.
Pick neB/t such that fi/n is not included in any element of s4n, so there
is Anes/n such that BnAneJ^ and there is meA/n\A. Then the condition
<? n {«, m}, B n ^n> forces that ^4W = ^4n and that meX/n\An—a contradiction.
Therefore X is forced not to be a diagonalization of {^4«}. But X has an infinite
subset Y which is a diagonalization of {An}, and the latter set cannot be Ji^-
generic.
To see that {J(2) fails, assume further that s/n+] refines srfn for all n;
namely, that every element of s/n+i is included in a unique element of j / n .
Let ^) be the statement.
((p) If h is the least ne^T such that X/n is not included in An, then the size
k of the set X n {1,...,«— 1} is an even number.
Since X does not diagonalize the sequence s/n, n is well-defined. Assume
Jl,w has the Prikry property, so let < 0 , A) be a condition which decides <p.
Find w, <mi<n2<m2 as follows: «] is the minimal neA such that A/n is not
included in any member of sin. Let A\ estfn be such that An A\ is in Jf, and
find m, >«, in .4 V4i. The set A' = A riA\\{\,. .., Wi} does not diagonalize
s/n, so let n2 be the minimal neA' such that A'/n is not included in any member
of sfn. Pick A2esin and let m2>n2 in A'\A2. Let A" = A'nA2\{l,. .., m2).
Then {{nl,ml}, A'} decides n = nt and k = 0 (so <p is forced), while
<{«,, n2, m2}, A"} decides M = M2 and k'=l (so —\<p is forced).

Remark. The importance of the above theorem is twofold—on the one

hand, it gives us a new forcing notion with nice properties which is not of the
 SEMISELECTIVE COIDEALS 97

usual form a -closed * ccc, and which even need not be proper—see Example
4.1 below. On the other hand, it gives us an alternative definition of semiselec-
tivity and shows that this is a very natural notion. The following fact is also
worth mentioning.

COROLLARY 4.1. If Jl.# has the Prikry property and/or the Mathias prop-
erty then in the Perfect-set forcing extension it retains this property.

Proof. This follows by Theorem 4.1 and by the fact that Perfect-set forcing
preserves semiselectivity (Theorem 4.2 below).

Proof of Theorem 2.1, M-Ramsey part. (b). Let s4 be an analytic set and
let [t, C] be such that Cejf; we can, and will, assume t = 0 and C= M. There
is a tree r c [ ^ ] < £ O x ^ < a ) such that B is in sf if, and only if, the tree
TB={seN<m:(Bn\s\,s}eT}
has an infinite branch. Find AeJt such that the condition {0,A} decides
whether Tx (X is a name for an ./^-generic set) has an infinite branch or not.
So in the first case let / be a name for the leftmost branch of Tx and in the
second case let H be a name for the rank mapping witnessing that Tx is well-
founded, i.e., an order-reversing mapping from Tx into ordinals. Since the tree
T is countable, there is a countable sequence of dense open subsets {$„} of
Jijr such that if G = G.^(X) intersects all of them then val G (/) (valc (H),
resp.) is an infinite branch through Tx (is a rank mapping for Tx, resp.). It
remains to find XeJf such that G^{X) is sufficiently generic in the above
sense. But by Lemma 4.2 it suffices that G(X) intersects dense open subsets
3i{$n) of Jf, and therefore by the semiselectivity there is Z e J f such that
Gj,(X) intersects all <gn's. Fix such an X.
If < 0 , A} forces Tx has an infinite branch, then the G^(X ^interpretation
/ o f / i n G^(X) is an infinite branch of Tx, so Xes/. If < 0 , A} forces Tx is
well-founded, then the G^(X^interpretation H of H witnesses Tx is well-
founded, so X^stf. Moreover, by the Mathias property the same applies to all
infinite subsets of X, and therefore [X\° is either included in or disjoint from
si.
(c). Assume (Fs: seN *<0> are Jf -Ramsey sets and [t, A] is such that AeJt?.
We need to provide Be Jtf \ A such that [t, B] is either included in or disjoint
from s/(Fs), and it is easy to see that we can assume t = 0 and A = N. For
N andseN<0> let

n,s= {Be*?: V«c {1,. . . , n}[u, B] cf s or [u, B] n F,= 0},

\s\ tin

Since each 3FS is ^f-Ramsey, sets Sn are dense open in Jf, so they have a
diagonalization B in jf. then Fs is clopen when restricted to [B]m for all s, so
st?(Fs) is analytic when restricted to [B]w and by part (b) there is a CeJ? \B
such that [C]a is either included in or disjoint from
98 I. FARAH

Now we turn to the forcing proof of perfectly Jf -Ramsey part of Theorem

2.1. This reduces to proving that semiselective coideals generate semiselective
coideals in the Perfect-set forcing extension. For definitions see [13] or [26].

THEOREM 4.2. If' .ff is a semiselective coideal then in the Perfect-set forcing
extension it generates a semiselective coideal.

Proof. By Lemma 4.1, it suffices to prove that the iteraction

Sacks * <Jf, <=*> forces % is selective. But this iteration can be considered as
the product of <Jf, £*> and Sacks because the definition of (jf, £*> is not
changed by the Sacks forcing, and therefore the conclusion follows from the
well-known fact that Sacks forcing preserves selective ultrafilters (see Theorem
1.1).

COROLLARY 4.2. In the Perfect-set forcing extension the family of ground-

model sets of integers diagonalizes all Borel partitions.

Proof. Clearly, the coideal [N]a is selective, so by Theorem 4.2 it remains

semiselective in an extension, and therefore by Theorem 2.1(b) the conclusion
follows.

Remark 4.1. All of the above results are true in the case of adding any
number of side-by-side Sacks reals with countable supports; this can be proved
by using the above methods, or by the direct fusion argument using the infinit-
ary version of the Halpern-Laiichli theorem proved by Laver ([13]), or more
precisely, its semiselective version that we will now formulate. If {Tn} is a
sequence of perfect trees and A^N then by (x)"=i Tn\A we denote the set
[JieA n ^ = 1 Tn(i) (here T(i) denotes the /-th level of the tree T). For a coideal
J? consider the following statement.
HL a ) (j | f) For every AeJtif, sequence of perfect trees T", and
/ : (x)"=i Tn->{0, 1} there is a fleJf \A and perfect subtrees
P"^T" such that /assumes a constant value on (x),T=i P" \B.
Then HLffl (Jf) for some coideal J f is equivalent to the following.
#e generates a coideal in any forcing extension by a countable support
product of Perfect-set forcing.
[An explanation of this equivalence in the case when jf=[N](O is given in
[13], where (1) and (2) correspond respectively to HL r a (Jf) and the above
preservation statement.] We note that HL (B (Jf) is true whenever ^f is semi-
selective by Theorem 6.8 of [26], where HL ffl (jf) for a selective ultrafilter J f
was proved.
Let us recall some definitions before we turn to the proof of the remaining
part of Theorem 2.1. To every analytic subset #J of {0, X)01 x [f^]<B we associate
the tree 7 ^ on {0, 1} < '° x [N]<<0 x [N]<co such that ^ equals the set of all
<x, B~) for which the tree
Tx,B={seN<a:(x\\s\,Bn{l,...,\s\},s}eT}
 SEMISELECTIVE COIDEALS 99

has an infinite branch. This gives a name for the set s/ in a forcing extension.
Note that a pair <j, A} is not in s/ if, and only if, tree TyB is well founded,
if, and only if, there are a countable ordinal a and a strictly decreasing mapping
H of TyiA into a. The minimal such a is called the rank of 7 ^ .

Proof of Theorem 2.1, perfectly JF-Ramsey part, (b). Fix an analytic set
,s/c:{o, l}<Bx[N]<B, a perfect f c {o, I}01, and [f, A] (Aejf). We have to find
a perfect Q^Pand Be Jf f^ such that g x [s, B] is either included in or disjoint
fromj/. Wecan, and will, assume P={0, \}a,t = 0 and A = H. Let j(X) be
the canonical name for a Sacks (Mathias, resp.)-generic real. By Theorem 4.2
and the already proved Jf-Ramsey part of Theorem 2.1, there is a perfect Q
and BeJt such that Q forces set {j>} x [i?]" is either included in or disjoint
from s4. Like in the proof of Jf-Ramsey part of this theorem above, let T be
the tree on {0, 1}<<B x [N]<0> x M<0) such that
s/ = {<j, X}: 7 ^ has an infinite branch}.
There are countably many dense open subsets {&>„} of the product Sacks x JI*
relevant to the decision of the statement l\y,X}eji/", namely these are the
sets deciding the infinite branch of the tree Tyj or a strictly decreasing mapping
of this tree into some countable ordinal. By the preservation of semiselective
coideals by Sacks forcing (Theorem 4.2) and Lemma 4.2, we can translate Z&n's
to dense open subsets &„ of the poset Sacks x ye, so that (j, X} is
(9n, Sacks x t#jr>-generic if, and only if, <y, G(X)> is <<fn, Sacks x j f >-gen-
eric. By Lemma 2.1, there are a perfect i?£ Q and CeJf \B such that for all
neC, all ueTR \n we have
n

<Rn[u],C/n>ef) gn,

and therefore (j, D} is Sacks x ^^-generic for all yeR and all D^C. In
particular, either fix[Cfc^ Or R x [C]01 n s/ = 0, depending on what
<Q, <0, B}> forces.
(c). Fix perfectly Jf -Ramsey sets F,!={0, l f x [ N f (ie^ < < o ), a perfect
P^{0, 1}", and [t, A] such that AeJC. We have to find a perfect g g p and
a Bejf \A such that 2 X [?, ^] is either included in or disjoint from s/(Fs).
We can, and will, assume P= {0, 1}", s = 0 and y4 = N. For a positive integer
nlet
®n={(P, B}: P perfect, fieJf, and for all wS {1,. . . «}
either P x [u, A] cFs or P x [«, ^] n F,= 0 } .
Then by Lemma 2.1 there are a perfect i? and CeJ^ such that for all neC,
teTR \n, w={l,...,«} and seN" the set

(Rn[t])x[u,C]
is either disjoint from or included in Fs, namely Fs is relatively clopen when
restricted to R x [C]w. Therefore the set s4{Fs) is analytic when restricted to
the same set, and the desired conclusion follows from part (b) of this theorem.
100 I. FARAH

Now we describe a situation in which all definable sets are #? -Ramsey for
every semiselective Jf. In this section we will use the word "definable" as a
synonym for "an element of L(U)". For the other definitions see [9].

THEOREM 4.3. If there is a supercompact cardinal and Jdf is a semiselective

coideal, then all definable sets are J f - Ramsey and perfectly M-Ramsey.

Shelah and Woodin ([24]) proved that if there is a supercompact cardinal

then all definable sets are Ramsey. Using a version of their result, Todorcevic
proved the following.

THEOREM 4.4. If there is a supercompact cardinal, then every selective

ultrafilter is [Nf-generic over L(U),

In other words, all definable sets are ^-Ramsey for every selective ultrafilter
<W. This unpublished result appears in [27] and since we shall need this result
in order to supply a proof of Theorem 4.3, we include its proof with
Todorcevic's kind permission. To state the lemma of Shelah and Woodin on
which the proof of Theorem 4.4 is based we need the following:

DEFINITION 4.2. Let M be a countable elementary submodel of some struc-

ture of the form He which contains a poset 0* and a 2P-name f for a real. Then
we say that M is (L(U), ^-correct if for every (M, 3P)-generic filter G<=,gP r\M
and every formula cp(x, p) with parameter p in M the formula <p(val<j (r) p) is
true in L(U), if, and only if, there is a condition in G which forces this. We shall
say that the truth in L(R) is unchangeable by forcing if the following condition
is satisfied.
(*) For every poset 3? there is a large enough 0 so that there are stationarily
many countable elementary submodels M of He which are (L(R), &)
correct.

LEMMA 4.3 ([4], [24], [5]). If there is a supercompact cardinal, then truth
in L(K) is unchangeable by forcing.

We will also need the following lemma.

LEMMA 4.4. Assume that truth in L(U) is unchangeable by forcing. If X

is a ccc Baire space and f: X -*U is continuous then the f-inverse image of any
set of reals from L(U) has the property of Baire.

Proof. Fix a set of reals A from L(U) and 6 which verifies (*) for 3P =
ro (X). Every set of reals from L(U) is determined by a formula q>(- , •) and
a finite sequence of parameters p which are either reals or ordinals in the sene
that x belongs to A if, and only if, q>{x,p) is true in L(U) (see [9]). So let
(p(- , •) andp be such that
A = {xeU: q>{x,p) is true in L(R)}.
If G is a name for a ^"-generic filter, then let f be the ^-name for a real such
that
f e [s] if, and only if, / ~ ' [s] e G,
 SEMISELECTIVE COIDEALS 101

for all basic intervals s. Let M be an (L(R), ^)-correct countable elementary

submodel of He and such that everything relevant is in M. Let B be the Boolean
value of the formula of the ^-forcing language which says that <p(f,p) is true
in L(U). Then B is a regular open subset of X.

CLAIM. f~\A)AB is a meager subset of X.

Proof. Suppose that it is not; say,/~'04)\B is a nonmeager. So we can

pick
xe(f~\A)\B)nf] {U: U is a dense open subset of X and UeM}.
Since X is a ccc space it is easily seen that G={Ue!P: xeU} is a (0>, M)-
generic filter. Note also that valc (r) =f(x) is an element of A, so the formula
<jo(valG (f),p) is true in L(U). This contradicts the assumption that M is
(L(K), ^)-correct, since clearly B is not in G. The assumption that
B\f~l(A) is nonmeager would lead to a similar contradiction.

Since B is a regular open subset of X, this proves Lemma 4.5.

Proof of Theorem 4.4. Let °U be a selective ultrafilter and let X be the

topological space [N]01 whose basis is the family Exp (•$?) as defined in §3. Then
by Corollary 3.1 this is a topology. By Theorem 3.3 X is a Baire space and
subsets of X with the Baire property are exactly the ^-Ramsey sets. Moreover
X is a ccc space because SP® is a ccc poset. Let/:X-» [N]ra be the identity
mapping; then/is continuous if [N]a is considered with its separable metric
topology, and by Lemma 4.4 every set of reals SC in L(U) is ^-Ramsey. So if
such an 3C is, moreover, dense in <[^J]°\ £*>, then there is Ae^l such that
[AY^SC. Therefore W is (L(U), [N]ffl)-generic.

Proof of Theorem 4.3. Let / : [N]1"->•{(), 1} be definable. Force with

<Jf, c:*>, and let t be a generic selective ultrafilter. Forcing with a small
poset preserves supercompactness, so by Theorem 4.4 the ultrafilter % is generic
over L(U). By the same result, the family of all A s N such that/is constant
on A is a dense open subset of [M]01, so % intersects it; therefore Jf intersects
it as well, and there is an AeJt such that/is constant on A. Since there are
no new reals in an extension, A is in the ground model and this proves the Jf-
Ramsey part of our theorem. The perfectly Jf-Ramsey part follows from the
fact that semiselectivity is preserved in the Perfect-set forcing extension, and
the unchangeability of theory of L(U) by Sacks forcing.

We finish this section with an example that has been promised after the
proof of Theorem 4.1 above.

Example 4.1. A semiselective coideal Jf such that the poset Jlye is not
proper. For a stationary, costationary subset S of W] let ./(S) be Jensen's
poset of all countable closed subsets of S ordered by the end-extension (see
[12, p. 250]). It is cr-distributive yet it destroys the stationarity of co^\S and
therefore it is not proper. Observe that J{S) is a cr-distributive tree of height
102 I. FARAH
(O\ with no uncountable branches such that every one of its nodes has contin-
uum many incomparable successors. By an extension (literally) of the construc-
tion of Example 2.2, for every such tree T there is a Ramsey coideal J^(T) such
that the regular open algebras of Jf(T) and Tare isomorphic; in particular, ^f
isCT-distributive.But every cr-distributive Ramsey coideal is semiselective, so
this finishes the construction.

§5. Concluding remarks We would like to point out some analogies

between §3 and §4. Note that the fact that Exp(Jf )-Baire property implies the
Jf-Ramsey property roughly corresponds to the poset Jl#> having the Prikry
property, while the fact that Exp(^f )-nowhere dense sets are Jf-Ramsey null
roughly corresponds to Jl^- having the Mathias property. "Roughly" because
e.g. for a condition [t, B] to decide the statement "Xestf" (where X is a name
for an ^#jr-generic and srf is an analytic set) it is not necessary that [t, B]Qstf
or [t, B]nstf = 0. An alternative approach was taken by Matet [41], who has
studied Jf-Ramsey property for a selective coideal 3ff using games.
Many partition theorems, like canonical ones of Pudlak-Rodl [22], Mathias
[16, §6] and Promel Voigt [21] have their semiselective versions. Functions to
which these results apply are exactly the functions which have the abstract
Exp (Jf)-Baire property, i.e., such that the inverse images of open sets have the
abstract Exp (jf )-Baire property (see §3). Their proofs rely on the following
strengthening of Theorem 3.3(b) which is of the independent interest as an
analogue of a classical theorem of Baire (see [30], also [16; 6.2]).

LEMMA 5.1. If Jf is a semiselective coideal and f: [N]a-*M has the

abstract Exp (Jfy Baire property, then there is AeJf such that f\[A]m is
continuous.

The proof of the semiselective version of a partition theorem can be

obtained either by mimicking the original proof, from the theorem itself (like
e.g., in [26, Theorem 6.8]) or by using forcing arguments. An explanation of
the phenomenon of the existence of (semi)selective versions of partition theor-
ems lies in Theorems 4.3 and 4.4 and the fact that partition theorems can
usually be formulated as the statements of L(U)

References
1. J. Baumgartner and R. Laver. Iterated Perfect-Set Forcing, Ann. Math. Logic, 17 (1979),
271-288.
2. J. Bourgain, D. H. Fremlin and M. Talagrand. Pointwise compact sets of Baire-measurable
functions. Amer. J. Math., 100 (1978), 846-886.
3. E. Ellentuck. A new proof that analytic sets are Ramsey. J. Symbolic Logic, 39 (1974), 161-
165.
4. M. Foreman, M. Magidor and S. Shelah. Martin's maximum. Ann. of Math., 127 (1988), 1-
47.
5. Q. Feng, M. Magidor and W. H. Woodin. Universally Baire sets of reals. MSRI Publications,
26 (1992), 203-242.
6. F. Galvin and K. Prikry. Borel sets and Ramsey's theorem. J. Symbolic Logic, 38 (1973), 193-
198.
 SEMISELECTIVE COIDEALS 103
7. J. D. Halpern and D. Pincus. Partitions of products. Trans. Amer. Math. Soc, 267 (1981),
549-568.
8. W. Just and A. Krawczyk. On certain Boolean algebras 8?((o)/I. Trans. Amer. Math. Soc,
285 (1984), 411 429.
9. A. Kanamori. The higher infinite (Springer-Verlag, 1995).
10. A. Kechris, S. Solecki and S. Todorcevic. Borel chromatic numbers (1995). To appear in
Advances in Math.
11. K. Kunen. Some points in 0N. Math. Proc. Cambridge Phil. Soc, 80 (1976), 385-398.
12. K. Kunen. An Introduction to Independence Proofs (North-Holland, 1980).
13. R. Laver. Products of Infinitely Many Perfect Trees. J. London Math. Soc. (2), 29 (1984),
385-396.
14. A. Louveau. Demonstration topologique de theoremes de Silver et Mathias. Bull. Sci. Math.,
98 (1974), 97-102.
15. A. R. D. Mathias. On sequences generic in the sense of Prikry. J. Austral. Math. Soc, 15
(1973), 403-414.
16. A. R. D. Mathias. Happy Families. Ann. Math. Logic, 12 (1977), 59-111.
17. A. W. Miller. Infinite combinatorics and definability. Ann. Pure Appl. Logic, 41 (1989), 179-
203.
18. C. St. J. A. Nash-Williams. On well-quasi-ordering of transfinite sequences. Proc. Camb. Phil.
Soc, 61 (1965), 33-39.
19. J. Pawlikowski. Parametrized Ellentuck theorem. Topology and its Applications, 37 (1990), 65-
73.
20. K. Prikry. Changing measurable into accessible cardinals. Dissertationes Math. (Rozprawy
Matematycne), 68 (1970), 5-52.
21. H. J. Promel and B. Voigt. Canonical forms of Borel-measurable mappings A: [a)]w -> R. Jour.
Comb. Theory, ser. A, 40 (1985), 409-417.
22. P. Pudlak and V. Rodl. Partition theorems for systems of finite subsets of integers. Discrete
Mathematics, 38 (1982), 67-73.
23. H. P. Rosenthal. A characterization of Banach spaces containing /'. Proc. Natl. Acad. Sci.
USA, 71 (1974), 2411-2413.
24. S. Shelah and W. H. Woodin. Large cardinals imply that every reasonably definable set of
reals is Lebesgue measurable. IsraelJ. Math., 70 (1990), 381-394.
25. J. Silver. Analytic sets are Ramsey. J. Symbolic Logic, 35 (1970), 60-64.
26. S. Todorcevic and I. Farah. Some Applications of the Method of Forcing (Mathematical Insti-
tute, Belgrade and Yenisei, Moscow, 1995).
27. S. Todorcevic. Lecture notes from a course given in Toronto, summer 1993.
28. S. Todorcevic. Topics in topology. (Springer Lecture Notes in Mathematics No. 1642, 1997).
29. J. Llopis and S. Todorcevic. Parametrized polarized partition relations. Preprint, 1958.
30. K. Kuratowski. Topology, Vol. I (Academic Press, 1966).
31. A. S. Kechris. Classical Descriptive Set Theory, (Springer-Verlag, 1995).
32. G. Sacks. Forcing with perfect closed sets. Axiomatic set theory, Proceedings of Symposia in
Pure Mathematics 13/1 (American Mathematics Society, Providence, 1971), pp. 331-355.
33. A. Dow. Personal communication (October, 1996).
34. R. Platek. Eliminating the Continuum Hypothesis. / . Symb. Logic, 34 (1969), 219-225.
35. E. Marczewski (Szilprajn). Sur une classe de fonctions de W. Sierpihski et la classe corre-
spondante d'ensembles. Fund. Math., 24 (1935), 17-34.
36. J. C. Morgan. On general theory of point sets II. Real Anal. Exchange, 12 (1986/87).
37. S. Todorcevic. Partition problems in topology (American Math. Soc, Providence, 1989).
38. S. Todorcevic. Compact sets of Baire class-l functions, (1997). Preprint.
39. J. Stern. A Ramsey theorem for trees, with an application to Banach spaces. Israel J. Math.,
29 (1978), 179-188.
40. J. Ketonen. On the existence of /"-points in Cech-Stone compactification of integers. Fund.
Math., (1976), 91 94.
41. P. Matet. Happy families and completely Ramsey sets. Arch. Math. Logic, 32 (1993), 151 —
171.
42. A. R. D. Mathias. A notion offorcing (1985). Preprint.

I. Farah, 03E05. MATHEMATICAL LOGIC AND

Department of Mathematics, FOUNDATIONS; Set theory; Combi-
York University, natorial set theory.
4700 Keele Street,
North York, Ontario,
Canada, M3J 1P3. Received on the \lth of September, 1997.