(

6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


On full Souslin trees
Saharon Shelah

Institute of Mathematics
The Hebrew University of Jerusalem
91904 Jerusalem, Israel
and
Department of Mathematics
Rutgers University
New Brunswick, NJ 08854, USA
and
Mathematics Department
University of Wisconsin Madison
Madison, WI 53706, USA
August 17, 2011
Abstract
In the present note we answer a question of Kunen (15.13 in [Mi91])
showing (in 1.7) that
it is consistent that there are full Souslin trees.

We thank the NSF for partially supporting this research under grant #144-EF67.
Publication No 624.
0
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 1
0 Introduction
In the present paper we answer a combinatorial question of Kunen listed in
Arnie Millers Problem List. We force, e.g. for the rst strongly inaccessible
Mahlo cardinal , a full (see 1.1(2)) Souslin tree and we remark that
the existence of such trees follows from V = L (if is Mahlo strongly
inaccessible). This answers [Mi91, Problem 15.13].
Our notation is rather standard and compatible with those of classical
textbooks on Set Theory. However, in forcing considerations, we keep the
older tradition that
a stronger condition is the larger one.
We will keep the following conventions concerning use of symbols.
Notation 0.1 1. , will denote cardinal numbers and , , , , , will
be used to denote ordinals.
2. Sequences (not necessarily nite) of ordinals are denoted by , ,
(with possible indexes).
3. The length of a sequence is g().
4. For a sequence and an ordinal g(), is the restriction of
the sequence to (so g() = ). If a sequence is a proper
initial segment of a sequence then we write (and has
the obvious meaning).
5. A tilde indicates that we are dealing with a name for an object in
forcing extension (like x

).
1 Full -Souslin trees
A subset T of
>
2 is an tree whenever ( is a limit ordinal and) the
following three conditions are satised:
) T, if T then T,
T implies

0),

1) T, and
for every T and < such that g() there is T such
that and g() = .
A Souslin tree is a tree T
>
2 in which every antichain is of size less
than .
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 2
Denition 1.1 1. For a tree T
>
2 and an ordinal we let
T
[]
def
= T

2 and T
[<]
def
= T
>
2.
If is limit then we dene
lim

T
[<]
def
=

2 : ( < )( T).
2. An tree T is full if for every limit ordinal < the set lim

(T
[<]
)
T
[]
has at most one element.
3. An tree T
>
2 has true height if for every T there is


2 such that
and ( < )( T).
We will show that the existence of full Souslin trees is consistent assuming
the cardinal satises the following hypothesis.
Hypothesis 1.2 (a) is strongly inaccessible (Mahlo) cardinal,
(b) S < : is a strongly inaccessible cardinal is a stationary set,
(c) S
0
is a set of limit ordinals,
(d) for every cardinal S,
S
0

holds true.
Further in this section we will assume that , S
0
and S are as above and
we may forget to repeat these assumptions.
Let as recall that the diamond principle
S
0

postulates the existence of a

sequence =

: S
0
) (called a
S
0

sequence) such that

2
(for S
0
) and
(

2)[ the set S
0
: =

is stationary in ].
Now we introduce a forcing notion Q and its relative Q

which will be
used in our proof.
Denition 1.3 1. A condition in Q is a tree T
>
2 of a true
hight = (T) < (see 1.1(3); so is a limit ordinal) such that
| lim

(T
[<]
) T
[]
| 1 for every limit ordinal < ,
the order on Q is dened by T
1
T
2
if and only if
T
1
= T
2

(T
1
)>
2 (so it is the endextension order).
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 3
2. For a condition T Q and a limit ordinal < (T), let

(T) be the
unique member of lim

(T
[<]
) T
[]
if there is one, otherwise

(T) is
not dened.
3. Let T Q. A function f : T lim
(T)
(T) is called a witness for T
if ( T)( f()).
4. A condition in Q

is a pair (T, f) such that T Q and f : T

lim
(T)
(T) is a witness for T,
the order on Q

is dened by (T
1
, f
1
) (T
2
, f
2
) if and only if
T
1

Q
T
2
and ( T
1
)(f
1
() f
2
()).
Proposition 1.4 1. If (T
1
, f
1
) Q

, T
1

Q
T
2
and
() either
(T
1
)
(T
2
) is not dened or it does not belong to rang(f
1
)
then there is f
2
: T
2
lim
(T
2
)
(T
2
) such that (T
1
, f
1
) (T
2
, f
2
) Q

.
2. For every T Q there is a witness f for T.
Proof Should be clear.
Proposition 1.5 1. The forcing notion Q

is (< )complete, in fact

any increasing chain of length < has the least upper bound in Q

.
2. The forcing notion Q is strategically -complete for each < .
3. Forcing with Q adds no new sequences of length < . Since |Q| = ,
forcing with Q preserves cardinal numbers, conalities and cardinal
arithmetic.
Proof 1) It is straightforward: suppose that (T

, f

) : < ) is an
increasing sequence of elements of Q

. Clearly we may assume that <

is a limit ordinal and
1
<
2
< (T

1
) < (T

2
). Let T

=

<
T

and = sup
<
(T

). Easily, the union is increasing and the T

is a full
tree. For T

let
0
() be the rst < such that T

and let
f

() =

f

() :
0
() < . By the denition of the order on Q

we get that the sequence f

() :
0
() < ) is increasing and hence
f

() lim

(T

). Plainly, the function f

witnesses that T

has a true height

, and thus (T

, f

) Q

. It should be clear that (T

, f

) is the least upper

bound of the sequence (T

, f

) : < ).
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 4
2) For our purpose it is enough to show that for each ordinal < and a
condition T Q the second player has a winning strategy in the following
game (

(T, Q). (Also we can let Player I choose T

for odd.)
The game lasts moves and during a play the players, called
I and II, choose successively open dense subsets T

of Q and
conditions T

Q. At stage < of the game:

Player I chooses an open dense subset T

of Q and
Player II answers playing a condition T

Q such that
T
Q
T

, ( < )(T

Q
T

), and T

.
The second player wins if he has always legal moves during the
play.
Let us describe the winning strategy for Player II. At each stage <
of the game he plays a condition T

and writes down on a side a function

f

such that (T

, f

) Q

. Moreover, he keeps an extra obligation that

(T

, f

)
Q
(T

, f

) for each < < .

So arriving to a non-limit stage of the game he takes the condition (T

, f

)
he constructed before (or just (T, f), where f is a witness for T, if this is
the rst move; by 1.4(2) we can always nd a witness). Then he chooses
T

Q
T

such that (T

) = (T

) + and (T

)
[(T

)]
= lim
(T

)
(T

).
Thus
(T

)
(T

) is not dened. Now Player II takes T

+1

Q
T

from
the open dense set T
+1
played by his opponent at this stage. Clearly

(T

)
(T
+1
) is not dened, so Player II may use 1.4(1) to choose f
+1
such
that (T

, f

)
Q
(T
+1
, f
+1
) Q

.
At a limit stage of the game, the second player may take the least upper
bound (T

, f

) Q

of the sequence (T

, f

) : < ) (exists by 1)) and then

apply the procedure described above.
3) Follows from 2) above.
Denition 1.6 Let T

be the canonical Qname for a generic tree added by

forcing with Q:

Q
T

T : T G

Q
.
It should be clear that T

is (forced to be) a full tree. The main point is

to show that it is Souslin and this is done in the following theorem.
Theorem 1.7
Q
T

is a Souslin tree.
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 5
Proof Suppose that A

is a Qname such that

Q
A

is an antichain ,
and let T
0
be a condition in Q. We will show that there are < and a
condition T

Q stronger than T
0
such that T

Q
A

[<]
(and thus
it forces that the size of A

is less than ).
Let A

be a Qname such that

Q
A

= T

: ( A

)( ) or ( A

)( ) .
Clearly,
Q
A

is dense open.
Let be a suciently large regular cardinal (
7
(
+
)
+
is enough).
Claim 1.7.1 There are S and B (H(), , <

) such that:
(a) A

, A

, S, S
0
, Q, Q

, T
0
B,
(b) |B| = and
>
B B,
(c) B = .
Proof of the claim: First construct inductively an increasing continuous
sequence B

: < ) of elementary submodels of (H(), , <

) such that
A

, A

, S, S
0
, Q, Q

, T
0
B
0
and for every <
|B

| =

< , B

, and

B
+1
.
Note that for a club E of , for every S E we have
|B

| = ,
>
B

, and B = .
Let S and B (H(), , <

) be given by 1.7.1. We know that

S
0

holds, so x a
S
0

sequence =

: S
0
).
Let
1

def
= T Q : T is incompatible (in Q) with T
0
or:
T T
0
and T decides the value of A

(T)>
2 and
( T)( T)( & T
Q
A

).
Claim 1.7.2 1

is a dense subset of Q.
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 6
Proof of the claim: Should be clear (remember 1.5(2)).
Now we choose by induction on < a continuous increasing sequence
(T

, f

) : < ) Q

B.
Step: i = 0
T
0
is already chosen and it belongs to Q B. We take any f
0
such that
(T
0
, f
0
) Q

B (exists by 1.4(2)).
Step: limit
Since
>
B B, the sequence (T

, f

) : < ) is in B. By 1.5(1) it has the

least upper bound (T

, f

) (which belongs to B).

Step: = + 1
First we take (the unique) tree T

of true height (T

) = (T

) + such
that T

(T

)>
2 = T

and:
if (T

) S
0
and
(T

)
/ rang(f

) then (T

)
[(T

)]
= lim
(T

)
(T

)
(T

)
,
otherwise (T

)
[(T

)]
= lim
(T

)
(T

).
Let T

Q1

be strictly above T

(exists by 1.7.2). Clearly we may choose

such T

in B. Now we have to dene f

. We do it by 1.4, but additionally

we require that
if T

then ( T

)( f

() & T
Q
A

).
Plainly the additional requirement causes no problems (remember the de-
nition of 1

and the choice of T

) and the choice can be done in B.

There are no diculties in carrying out the induction. Finally we let
T

def
=

<
T

and f

<
f

.
By the choice of B and we are sure that T

is a tree. It follows from

1.5(1) that (T

, f

) Q

, so in particular the tree T

has enough branches

(and belongs to Q).
Claim 1.7.3 For every lim

(T

) there is < such that

( < (T
+1
))(T
+1

Q
A

).
Proof of the claim: Fix lim

(T

) and let
S

def
= S
0
: (T

) = and

= .
Plainly, the set S

is stationary in (remember the choice of ). By the

denition of the T

s (and by lim

(T

)) we conclude that for every

S

(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 7
if

(T
+1
) is dened then ,=

(T

) =

(T
+1
).
But =

(as S

). So look at the inductive denition: necessarily for

some

we have

= f

), i.e. = f

). Now,

=

<
T

and hence for some () < , we have

T
()
. By Fodor lemma we nd

< such that the set

S

def
= S

: () =

is stationary in . Consequently we nd

such that the set

S
+

def
= S

is stationary (in ). But the sequence f

) :

< ) is increasing,
and hence the sequence is its limit. Now we easily conclude the claim
using the inductive denition of the (T

, f

)s.
It follows from the denition of A

and 1.7.3 that

T

Q
A

(remember that A

is a name for an antichain of T

), and hence
T

Q
|A

| < ,
nishing the proof of the theorem.
Denition 1.8 A tree T is S
0
full, where S
0
, if for every limit <
if S
0
then T
[]
= lim

(T),
if S
0
then |T
[]
lim

(T)| 1.
Corollary 1.9 Assuming Hypothesis 1.2:
1. The forcing notion Q preserves cardinal numbers, conalities and car-
dinal arithmetic.
2.
Q
T

>
2 is a Souslin tree which is full and even S
0
full .
[So, in V
Q
, in particular we have:
for every < < , for all T

2 there is T

2 such that
, and for a limit ordinal < , lim

(T
[<]
) T
[]
is either empty
or has a unique element (and then S
0
).]
(
6
2
4
)


r
e
v
i
s
i
o
n
:
1
9
9
6
-
0
8
-
1
3







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
8
-
1
3


[Sh:624] August 17, 2011 8
Proof By 1.5 and 1.7.
Of course, we do not need to force.
Denition 1.10 Let S
0
, S . A sequence (C

) : < limit) is
called a squared diamond sequence for (S, S
0
) if for each limit ordinal <
(i) C

a club of disjoint to S,
(ii)

2,
(iii) if acc(C

) then C

= C

and

,
(iv) if S then

: C

S
0
) is a diamond sequence.
Proposition 1.11 Assume (in addition to 1.2)
(e) there exist a squared diamond sequence for (S, S
0
).
Then there is a Souslin tree T
>
2 which is S
0
full.
Proof Look carefully at the proof of 1.7.
Corollary 1.12 Assume that V = L and is Mahlo strongly inaccessible.
Then there is a full Souslin tree.
Proof Let S < : is strongly inaccessible be a stationary non-
reecting set. By Beller and Litman [BeLi80], there is a square C

: <
limit) such that C

S = for each limit < . As in Abraham Shelah

Solovay [AShS 221, 1] we can have also the squared diamond sequence.
References
[AShS 221] Uri Abraham, Saharon Shelah, and R. M. Solovay. Squares with
diamonds and Souslin trees with special squares. Fundamenta
Mathematicae, 127:133162, 1987.
[BeLi80] Aaron Beller and Ami Litman. A strengthening of Jensens
square principles. The Journal of Symbolic Logic, 45:251264,
1980.
[Mi91] Arnold W. Miller. Arnie Millers problem list. In Haim Judah,
editor, Set Theory of the Reals, volume 6 of Israel Mathemati-
cal Conference Proceedings, pages 645654. Proceedings of the
Winter Institute held at BarIlan University, Ramat Gan, Jan-
uary 1991.