5

3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


6, 2003
Uniformization, choice functions and well orders in the class of trees.
BY
SHMUEL LIFSCHES and SAHARON SHELAH*
Institute of Mathematics, The Hebrew University of Jerusalem, Jerusalem, Israel
ABSTRACT
The monadic second-order theory of trees allows quantication over elements and over arbitrary
subsets. We classify the class of trees with respect to the question: does a tree T have a denable
choice function (by a monadic formula with parameters)? A natural dichotomy arises where the
trees that fall in the rst class dont have a denable choice function and the trees in the second
class have even a denable well ordering of their elements. This has a close connection to the
uniformization problem.
0. Introduction
The uniformization problem for a theory T in a language L can be formulated as follows:
Suppose T (

Y )(

X)(

X,

Y ) where is an L-formula and

X,

Y are tuples of variables. Is there
another L-formula

such that
T (

Y )(

X)[

(

X,

Y ) (

X,

Y )] and T (

Y )(!

X)

(

X,

Y )?
Here ! means there is a unique.
The monadic second-order logic is the fragment of the full second-order logic that allows quanti-
cation over elements and over monadic (unary) predicates only. The monadic version of a rst-order
language L can be described as the augmentation of L by a list of quantiable set variables and
by new atomic formulas t X where t is a rst order term and X is a set variable. The monadic
theory of a structure / is the theory of / in the extended language where the set variables range
over all subsets of [/[ and is the membership relation.
Given a tree T we may ask the following question: is there a sequence

P of subsets of T and a
formula (x, X,

Z) in the monadic language of trees such that
T [= (a, A,

P) [A ,= & a A] T [= (X)(y)[X ,= (y, X,

P)] and
T [= (a, A,

P) (b, A,

P) a = b ?
* The second author would like to thank the U.S.Israel Binational Science Foundation for par-
tially supporting this research. Publ. 539
1
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


If the answer is positive we will say that T has a (monadically) denable choice function (with
parameters) and that denes a choice function from non-empty subsets of T. Note that if we
let (x, Y ) be the formula that says if Y is not empty then x Y then a negative answer to
the choice function problem for T implies a negative answer to the uniformization problem for the
monadic theory of T (with being a counter-example).
dealing with the choice function problem we split the class of trees into two natural parts, wild
trees and tame trees and prove the following:
Theorem. Let T be a tree. If T is wild or T embeds
>
2 then there is no denable choice
function on T (by a monadic formula with parameters). If T is tame and does not embed
>
2 then
there is even a denable well ordering of the elements of T by a monadic formula (with parameters)
(x, y,

P).
Looking at the denitions and proofs we observe that a tree is tame [wild] if and only if its completion
is tame [wild] and that the counter-examples for the choice function problem are either anti-chains
or linearily ordered subsets of T. Hence we can prove:
Conclusion. Let T be a tree and T

be its completion. Then the following are equivalent:

a) For some n, l < , for every anti-chain/branch A of T there is a monadic formula
A
(x, X,

P
A
)
with quantier depth n and l parameters from T, that denes a choice function from non empty
subsets of A.
b) There is a monadic formula, with parameters, (x, y,

P) that denes a well ordering of the
elements of T.
c) There is a monadic formula, with parameters,

(x, y,

P

) that denes a well ordering of the

elements of T

.
The paper continues the work by Gurevich-Shelah ([GuSh]) who answered negatively a question by
Rabin ([Ra]), by showing that the answer for the choice function problem is negative in
>
2.
The positive results on the existence of a denable well ordering (3,5) are elementary and do not
require knowledge of monadic logic. The negative results (2,3,4) are based on understanding of
some composition theorems that hold for the monadic theory of trees. These facts are collected in
1.
More details and Historical background can be found in [Gu] and [GuSh].
1. Composition Theorems
In this section we will dene partial theories and establish the technical tools that will be applied
later. We will formalize composition theorems that will enable to compute the partial theory of a
tree from partial theories of its parts. Using such theorems enables to prove that if for example a
dense chain does not have denable choice function then a tree with a dense branch does not have
a denable choice function.
Denition 1.1. (T, ) is a tree if is a partial order on T and for every T, : is
linearily ordered by .
Note, a chain (C, <) and even a set without structure I is a tree.
Denition 1.2. Let T be a tree
2
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


1. X T is a convex subset if , X and T implies X. If T is a chain we use the
term a convex segment or just a segment.
2. (S, ) is a subtree of (T, ) if S T and S is a convex subset of T.
3. B T is a sub-branch of T if B is convex and linearily ordered.
4. B T is a branch of T if B is a maximal sub-branch of T.
5. A T is an initial segment of T if A is a sub-branch that is downward closed. is above an
initial segment A if A .
6. For T, T

is the sub-tree ( T : , ). T
>
is the sub-tree (T

, ). For
A T an initial segment, T
A
and T
>A
are dened naturally.
7. For T we deote by suc() or suc
T
() the set of immediate successors of (which may be
empty).
8. For , T we denote the intersection of and in T by . This may be a member of T or
an initial segment of T, in any case the meaning of is natural and [ ] is
used only when is an element [an initial segment].
9. If there is an T that satises ( T)[ ] we say that T has a root and denote by
root(T).
10. , T are incomparable in T if neither nor . X T is an anti-chain of T if X
consists of pairwise incomparable elements of T.
11. A gap in T is a pair (A, B) where A B = , A B is a sub-branch, A is an initial segment,
(so A, B ), A without a -maximal element, B without a -minimal element, and for
some T for every A and B we have & , are incomparable.
12. Filling a gap (A, B) in T is adding a node to T such that A , B and
for every as in (11) we have .
Denition 1.3. The full binary tree is the tree (
>
2 , ) where for sequences ,
>
2,
means is an initial segment of .
Denition 1.4. The monadic language of trees L is the monadic version of the language of
partial orders . Usually means smaller than or equal but when we restrict ourselves to chains
(linearily ordered sets) we use < and . For simplicity, we add to L the predicate sing(X) saying
X is a singleton so that we can quantify only over subsets. Note that everything that is dened
in 1.2 is denable in L.
Next we dene, following [Sh], the partial theories of a tree T. These are nite approximations
of the monadic theory of T. Th
n
(T;

P) is essentially the monadic theory of (T,

P, ) restricted to
sentences of quantier depth n.
Denition 1.5. For any tree T,

A T(T)
lg(

A)
, and a natural number n, dene by induction
t = Th
n
(T;

A).
for n = 0:
t =
_
(

X) : (

X) L, (

X) quantier free, T [= (

A)
_
.
for n = m+ 1:
t =
_
Th
m
(T;

A

B) : B T(T).
T
n,l
is the set of all formally possible Th
n
(T;

P) where T is a tree and l(

P) = l.
3
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Fact 1.6. (A) For every formula (

X) L there is an n such that from Th
n
(T;

A) we can
eectively decide whether T [= (

X).
(B) If m n then Th
n
(T;

A) can be eectively computed from Th
m
(T;

A).
(C) Each Th
n
(T;

A) is hereditarily nite, and we can eectively compute the set T
n,l
of formally
possible Th
n
(T,

A).
Next we recall the composition theorem for linear orders which states that the partial theory
of a chain can be computed from the partial theories of its convex parts. This allows us to sum
partial theories formally.
Denition 1.7. If C, D are chains then C + D is any chain that can be split into an initial
segment isomorphic to C and a nal segment isomorphic to D.
If C
i
: i < ) is a sequence of chains then

i<
C
i
is any chain D that is the concatenation of
segments D
i
, such that each D
i
is isomorphic to C
i
.
Theorem 1.8 (composition theorem for linear orders).
(1) If l(

A) = l(

B) = l(

A

) = l(

B

) = l, and
Th
m
(C,

A) = Th
m
(C

,

A

) and Th
m
(D,

B) = Th
m
(D

,

B

)
then
Th
m
(C +D, A
0
B
0
, . . . , A
l1
B
l1
) = Th
m
(C

+D

, A

0
B

0
, . . . , A

l1
B

l1
).
(2) If Th
m
(C
i
,

A
i
) = Th
m
(D
i
,

B
i
), l(

A
i
) = l(

B
i
) = l for each i < , then
Th
m
_

i<
C
i
,
i
A
1,i
, . . . ,
i
A
l1,i
_
= Th
m
_

i<
D
i
,
i
B
1,i
, . . . ,
i
B
l1,i
_
.
Proof. By [Sh] Theorem 2.4 (where a more general theorem is proved), or directly by induction
on m.

Notation 1.9.
(1) t
1
+t
2
= t
3
means: for some m, l < , t
1
, t
2
, t
3
T
m,l
(remember denition 1.5) and
if
t
1
= Th
m
(C, A
0
, . . . , A
l1
) and t
2
= Th
m
(D, B
0
, . . . , B
l1
)
then
t
3
= Th
m
(C +D, A
0
B
0
, . . . , A
l1
B
l1
).
By the previous theorem, the choice of C and D is immaterial.
(2)

i<
Th
m
(C
i
,

A
i
) is Th
m
(

i<
C
i
,
i<
A
1,i
, . . . ,
i<
A
l1,i
).
(3) If D is a subchain of C and X
1
, . . . , X
l1
are subsets of C then Th
m
(D, X
0
, . . . , X
l1
) abbre-
viates Th
m
(D, X
0
D, . . . , X
l1
D).
(4) We use abbreviations as

P

Q,
i

P
i
and

P C. The meanings should be clear.
(5) For C a chain, a < b C and

P C we denote by Th
n
(C;

P)
[a,b)
the theory Th
n
([a, b);

P
[a, b)).
The class of trees has some weaker (but sucient for our purpose) composition theorems. First
we dene the composition of subtrees of the full binary tree following [GuSh] and quote the the
respective composition theorem.
4
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Denition 1.10. Let M
>
2 be a tree. A grafting function on M is a function g satisfying the
following conditions:
(a) Dom(g) M 0, 1,
(b) if (x, 0) Dom(g) then x

0) , M and if (x, 1) Dom(g) then x

1) , M,
(c) every value g(x, d) of g (d 0, 1) is a tree
>
2.
A composition of a tree M and a grafting function g is the tree
M
_
x

d)

y : (x, d) Dom(g), y g(x, d)

_
.
Theorem 1.11 (composition theorem for binary trees). Let M
>
2 be a tree, N
>
2
be the composition of M and a grafting function g,

X N and n < . Then, there is m = m(n) <
(eectively computable from n) such that from Th
m
(M;

X,

L
g
(n,

X),

R
g
(n,

X)) we can eectively
compute Th
m
(N;

X) where
L
g
t
(n,

X) :=
_
x M : (x, 0) Dom(g), Th
n
(g(x, 0),

X) = t
_

L
g
(n,

X) :=
_
L
g
t
(n,

X) : t a formally possible n theory
_
and

R
g
(n,

X) is dened similary by replacing L, 0 with R, 1.
Proof. This is theorem 2 in 2.3. of [GuSh]. The language that is used there is dierent from our
L but all the mentioned symbols are monadically inter-denable (with some additional parameters)
with our , (For example the relation X is an immediate left successor of Y is easily denable
from and the parameter A :=
>
2 : (
>
2)[ =

0)]). Thus the translation of
[GuSh]s proof is clear.

The next three theorems allow us to compute a partial theory Th

n
(T;

X) from partial theories
of sub-structures of T. The proofs are by induction on n noting that Th
0
(T;

P) can express only
statements as P
i
P
j
, P
i
P
j
and P
i
= P
j
and that Th
n+1
is a collection of n-theories. Everything
is basically the same as in the previous case and we will not elaborate beyond that.
Theorem 1.12 (composition theorem for general successors). Let T be a tree,

X T and
A T an initial segment (i.e. linearily ordered by and downward closed).
For every x above A (x , A and y A y x) denote by T
A,x
the sub-tree
y T : (z)[z x & z y & z above A].
We say that x and y are equivalent above A if x and y are above A and T
A,x
= T
A,y
(compare with
denition 4.1), nally let T
i
: i I
A
list the equivalence classes above A (its a disjoint union of
sub-trees).
Then for every n < , there is m = m(n) < (eectively computable from n) such that from
Th
m
(T
A
;

X) and Th
m
(I
A
;

P
A
(n,

X)) we can eectively compute Th
n
(T;

X) where
T
A
:= y T : y not above A
P
A
t
(n,

X) := i I
A
: Th
n
(T
i
;

X) = t,

P
A
(

X) := P
A
t
(n,

X) : t a formally possible n theory,
and Th
m
(I
A
;

P
A
(n,

X)) is the m-theory of a set without sructure i.e. in the monadic language of
equality.
(A natural case is when for some y T we have A = z : z y, x
i
: i I = suc(y) and
T
i
= T
xi
).

5
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Theorem 1.13 (composition theorem for branches). Let T be a tree, B T a branch,

X T and n < . (B

, ) is the chain that is obtained by adding nodes to ll the gaps in B

remember 1.2(12), (so B

is contained in the completion of B). And let T

be the tree obtained by

replacing the branch B by B

Then there is m = m(n) < (eectively computable from n) such that from Th
m
(B

;

P
B

(n,

X))
we can eectively compute Th
n
(T;

X) where
for B

, T
B

:= T

P
B

t
(n,

X) :=
_
B : Th
m
(T
B

;

X) = t
_
P
B

t
(n,

X) :=
_
P
B

t
(n,

X) : t a formally possible n theory
_
.
Moreover, if

Y B then from Th
m
(B

;

P
B

(n,

X),

Y ) we can eectively compute Th
n
(T;

X,

Y ).

Notations 1.14. For stating the next composition theorem we need a considerable amount of
notations.
Let T be a tree, by F:
>
2 T is an embedding we mean F is 1-1 and for ,
>
2,
F() F(), we also assume that T has a root and F(root(
>
2)) = root(T).
let S T be F

(
>
2), it is a tree (but not necesarily a subtree of T) that can be identied with
>
2.
For x = F() S dene x
0
[x
1
] S to be F(

0)) [F(

1))].
For Y S an anti-chain (hence an anti-chain of T) let Bush(Y ) := x T : (y Y )[x y] (its a
subtree of T) and let Bush
S
(Y ) := Bush(Y ) S (its a subtree of S).
For every y S denote y
0
y
1
by y
i
. It may be an element of T or an initial segment but remember
the convention in 1.2(8).
For every y S we dene some subtrees of T
y
(some of them may be trivial if for example y = y
i
):
0) T
0
(y):=T
y
.
1) T
1
(y):=
_
x T : (y
i
x) & (z ,= y)[(z x)&(y z y
i
)]
_
, [These are the elements that split
from the segment (y, y
i
) ].
2) T
2
(y):=
_
x T : (y x) & (z)[(z y
i
)&(z x) (z y)]
_
. (If y
i
is an initial segment replace
z y
i
with z y
i
), [These are the elements that split from y but not from the segment (y, y
i
) ].
3) T
3
(y):=
_
x T : (y
0
x) & (z ,= y
i
)[(z x)&(y
i
z y
0
)]
_
. (If y
i
is an initial segment replace
(z ,= y
i
) with (z)(y
i
z) ), [These are the elements that split from the segment (y
i
, y
0
) ].
4) T
4
(y):=
_
x T : (y
1
x) & (z ,= y
i
)[(z x)&(y
i
z y
1
)]
_
. (If y
i
is an initial segment replace
(z ,= y
i
) with (z)(y
i
z) ), [These are the elements that split from the segment (y
i
, y
1
) ].
5) T
5
(y):=
_
x T : (y
i
x) & (z)[(z x)&(z y
0
z y
1
) (z y
i
)]
_
. (If y
i
is an initial segment
replace z y
i
with z y
i
), [These are the elements that split from y
i
but not from the segments
(y
i
, y
0
) and (y
i
, y
1
)].
6) T
6
(y):= T
y
0.
7) T
7
(y):= T
y
1.
For y S,

P T,

t = t
0
, t
1
. . . , t
7
), t
i
a possible n-theory, we have y Q
t
Th
n
(T
0
(y);

P) =
t
0
&. . . & Th
n
(T
7
(y);

P) = t
7
. For y , S we have y Q

.
Finaly let

Q(n,

P) be Q
t
:

t a possible sequence of n-theories)

)
Note that every anti-chain Y is denable from Bush
S
(Y ) and S is denable from

Q.
Theorem 1.15 (composition theorem for embeddings). Following the above notations, let
T be a tree and F:
>
2 T an embedding.
6
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Then for every Y S an anti-chain, y Y ,

P T and n < , there is m = m(n) < (eec-
tively computable from n) such that from Th
m
(Bush
S
(Y ); y,

Q(n,

P)) we can eectively compute
Th
n
(T; y, Y,

P).

2. Dense linear orders

Every nite set A has a denable well ordering by a formula with [A[ parameters. This is not
the case for innite models.
Claim 2.1. Let A be an innite set without structure. Then there is no denable choice function
on A. Moreover, if [A[ > 2
l
then no formula with l parameters denes a choice function on A.
Proof. Let

P A and suppose (x, X,

P) denes a choice function on an innite A. Let B A
be an indiscernible set with respect to (belonging to)

P of size 2. Then, for every b
1
, b
2
B,
A [= (b
1
, B,

P) i A [= (b
2
, B,

P), a contradiction. The second part is clear.

A chain C that embeds a dense linear order (hence the rational order Q) does not have a
denable choice function. The proof is by applying a Ramsey-like theorem for additive colourings
from [Sh].
Denition 2.2. (a) A colouring of a chain C is a function f from the set of unordered pairs of
distinct elements of C, into a nite set I of colours.
(b) The colouring f is additive if for x
i
< y
i
< z
i
C (i = 1, 2),
[f(x
1
, y
1
) = f(x
2
, y
2
), f(y
1
, z
1
) = f(y
2
, z
2
)] f(x
1
, z
1
) = f(x
2
, z
2
).
In this case a partial operation + is dened on I, such that for x < y < z C, f(x, z) = f(x, y) =
f(y, z). (Compare with 1.9(1)).
(c) A subchain D C is homogeneous (for f) if there is an i
0
I such that for every x < y D,
f(x, y) = i
0
.
Theorem 2.3. If f is an additive colouring of a dense chain C, by a nite set I of colours, then
there is an interval of C which has a dense homogeneous subset.
Proof. This is theorem 1.3. in [Sh].

Claim 2.4. Let (C, <) be a linear order that embeds a dense linear order. Then there is no
denable choice function on C.
Proof. Let

P C and suppose (x, X,

P) denes a choice function on C. Let n be so that from
Th
n
(C; x, X,

P) we know if (x, X,

P) holds and naly let D C be dense (in itself). By 2.3 there is
an A D, dense inside an interval of D, hence in itself, homogeneous with respect to the colouring
f(a, b) = Th
n
(C;

P)
[a,b)
, (Remember the notation 1.9(5)).
Let t

be the constant theory Th

n
(C;

P)
[a,b)
for every a < b in A. Let Z be the set of integers and
X A, X := x
n
: n Z be of order type Z. Suppose our choice function picks x
m
from X, i.e.
C [= (x
m
, X,

P).
7
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


We assume for simplicity of notations that inf(X) and sup(X) belong to C and denote inf(X) by
0 and sup(X) by 1. So Th
n
(C;

P)
X
= Th
n
(C;

P)
(0,1)
.
Letting t
0
be Th
n
(C;

P)
{x:x0}
, and t
1
be Th
n
(C;

P)
{x:x1}
we get:
Th
n
(C;

P) = t
0
+

kZ
Th
n
(C;

P)
[x
k
,x
k+1
)
+t
1
= t
0
+

kZ
t

+t
1
Now denote:
t

0
:= Th
n
(C; x
m
, X,

P)
{x:x0}
( = Th
n
(C; , ,

P)
{x:x0}
),
t

1
:= Th
n
(C; x
m
, X,

P)
{x:x1}
( = Th
n
(C; , ,

P)
{x:x1}
),
t

:= Th
n
(C; x
l
, X,

P)
[x
k
,x
k+1
)
for k ,= l, ( = Th
n
(C; , x
k
,

P)
[x
k
,x
k+1
)
) and
t
(l)
:= Th
n
(C; x
l
, X,

P)
[x
l
,x
l+1
)
( = Th
n
(C; x
l
, x
l
,

P)
[x
l
,x
l+1
)
).
Clearly t
0
determines t

0
, t
1
determines t

1
, t

0
and t

1
do not depend on m and t

determines t

and
t
(l)
. We also have, for every l Z:
Th
n
(C; x
l
, X,

P) = t

0
+

jZ,<
t

+t
(l)
+

jZ,>
t

+t
1
But, by homogeneity, we get for every k, l Z:
1) t
(k)
= t
(l)
,
2) Th
n
(C; x
l
, X,

P)
(0,x
l
)
=

jZ,<
t

=

jZ,<
t

= Th
n
(C; x
k
, X,

P)
(0,x
k
)
,
3) Th
n
(C; x
l
, X,

P)
(x
l
,1)
=

jZ,>
t

=

jZ,>
t

= Th
n
(C; x
k
, X,

P)
(x
k
,1)
.
It follows that Th
n
(C; x
m
, X,

P) = Th
n
(C; x
l
, X,

P) for every l Z, but chooses x
m
from X,
(and can be computed from Th
n
) a contradiction.

3. Scattered orders
A scattered order is a linear order that does not embed a dense order. We will dene Hdeg, the
Hausdor degree of scattered chains, and show that a scattered chain (C, <
C
) has a denable well
ordering if Hdeg(C) < and that Hdeg(C) there is no denable choice function on C.
Denition 3.1. We dene by recursion the Hausdor degree of a scattered chain (C, <
C
):
Hdeg(C) = 0 i C is nite
Hdeg(C) = i
<
Hdeg(C) ,= and C =

iI
C
i
where I is well ordered or inversely well
ordered and for every i I,
<
Hdeg(C
i
) = .
Hdeg(C) i ( < )(Hdeg(C) > ) ( limit).
Claim 3.2. (1) Let C be a scattered chain with Hdeg(C) = , C

the completion of C and

D C

. Then C

and D are scattered and Hdeg(D) Hdeg(C

) = .
(2) Let C be a scattered chain. Hdeg(C) is well dened (i.e. it is an ordinal ).
Proof. (1) By induction on .
(2) By [Ha].

Claim 3.3. Let C be a scattered chain with Hdeg(C) = n. Then there are

P C, lg(

P) = n1,
and a formula (depending on n only)
n
(x, y,

P) that denes a well ordering of C.
8
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Proof. By induction on n = Hdeg(C):
n 1: Hdeg(C) 1 implies (C, <
C
) is well ordered or inversely well ordered. A well ordering of
C is easily denable from <
C
.
Hdeg(C) = n + 1: Suppose C =

iI
C
i
and each C
i
is of Hausdor degree n. By the in-
duction hypothesis there are a formula
n
(x, y,

Z) and a sequence

P
i
: i I) with

P
i
C
i
,

P
i
= P
i
1
, . . . , P
i
n1
) such that
n
(x, y,

P
i
) denes a well ordering of C
i
.
Let for 0 < k < n, P
k
:=
iI
P
i
k
(we may assume that the union is disjoint) and P
n
:= C
i
: i even.
We will dene an equivalence relation by x y i
_
i
(x C
i
y C
i
).
and [x], (the equivalence class of an element x), are easily denable from P
n
and <
C
. We can
also decide from P
n
if I is well or inversely well ordered (by looking at subsets of C consisted of
nonequivalent elements) and dene <

to be < if I is well ordered and the inverse of < if not.

n+1
(x, y, P
1
, . . . , P
n
) will be dened by:

n+1
(x, y,

P)
_
x , y & x <

_
x y &
n
(x, y, P
1
[x], . . . , P
n1
[x])

n+1
(x, y,

P) well orders C.

Next we prove that a scattered orders of innite Hdeg dont have a denable choice function
(hence a well ordering).
Denition 3.4. We dene for every n < a model /
n
in the language consisted of a binary
relation <
n
:
a) The universe of /
n
, which will be denoted by M
n
, is the tree
n
.
b) Let, for every
n
, <

be a linear ordering of suc() :=

k) : k < such that if lev()

is even then k < l

k) <

l), and if lev() is odd then k < l

l) <

k).
(So <

orders suc() with order type if is in an even level and with order type

if is in an
odd level).
c) <
n
is the lexicographic order induced by the orders <

of immediate successors.
(M
n
, <
n
) is hence a chain. Note, the usual partial order on
n
(being an initial segment), is
not denable in /
n
.
Denition 3.5. We dene by induction the scattered chains C
n
and C

n
:
C
1
:= , C

1
:=

,
C
2
:=

i

, C

2
:=

i
,
and in general:
C
n
:=

i
C

n
, C

n
:=

i
C
n
.
Denition 3.6. f: /
n
C is an embedding of /
n
in a scatterd chain (C, <
C
) if f is 11 and
<
n
f() <
C
f()
Fact 3.7. Let C be a scattered chain with Hdeg(C) n + 1. Then there is an embedding
f: /
n
C.
Proof. Clearly the following hold:
() For a scattered chain C: Hdeg(C) = n [C
n
C or C

n
C].
() /
n
/
n+1
() There is an embedding g: /
n
C
n
.
9
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Now assume Hdeg(C) = n + 1 and use (). In the case C
n+1
C we have by () an embedding
g: /
n+1
C and by () an embedding f: /
n
C. In the case C

n+1
C we have, by the
denition of C

n+1
, C
n
C

n+1
and by () an embedding f: /
n
C.

Conclusion 3.8. Let C be a scattered chain with Hdeg(C) . Then, for every n < there is
an embedding of /
n
into C.

Lemma 3.9. If C is scattered and Hdeg(C) then no monadic formula (x, X,

P) denes a
choice function on C.
Proof. Assume towards a contradiction that there is

P C, lg(

P) = l and (x, X,

P) denes a
choice function on C. Let m be so that from Th
m
(C; x, X,

P) we can decide if C [= (x, X,

P). As
in the proof of 2.4 it is enough to nd an B C, of order type Z, homogeneous with respect to the
colouring f(a, b) = Th
m
(C;

P)
[a,b)
. Let
n > [Th
m
(D;

Q) : D a chain ,

Q D, l(

Q) = l[ = [T
m,l
[
and f: /
n
C be an embedding. Let T C be the image of f and we will identify T with
n

and the submodel (T, <

C
) (C, <
C
) with the model (
n
, <
n
)
Notation: We will write < instead of <
C
and its restriction <
n
. Given <
n
= T we will
write Th
m
[, ) instead of Th
m
(C;

P)
[,)
. T

and T
>
are the usual subsets of
n
= T
We will begin to thin out the tree T =
n
, in order to obtain a quite homogeneous subtree A T
going down with the levels. Arriving to a node , we will have dened A

for every suc()

and will dene A

by thinning out suc() to a set B

and taking A

: B

. A

will
satisfy the following:
() [ < A

, lev() = lev()] Th
m
[, ) depends only on lev( )
Assume w.l.o.g that n is odd.
Step 1: for every
n
with lev() = n 1 pick out an innite set B

such that
k < l B

Th
m
(C;

P)
[

k,

l)
= t

(note that k < l <

k) <

l)), let k

be the second element of B

. Let A

be

k) : k

k B

, this is a subtree of T and () clearly holds.

Step 2: Given
n
with lev() = n 2 we have dened B

, k

and A

for every suc().

Pick out an innite B
0

so that () will hold for lev() = lev() = n1, lev( ) = n 2 i.e.

k > l B
0

Th
m
[

k),

l)) = t

(suc() are ordered as

). Thin out B
0

to an innite B
1

so that () will hold for lev() =

lev() = n, lev( ) = n 2 i.e.
k > l B
1

, =

k), =

l) Th
m
[

),

k

)) is constant
Why does it suce to look only at e.g.

) ? because by the choice of t

and A

we have
t

+t

= t

hence for every < T we can break the paths [,

)) and [,

l)), for l A

,
10
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


into three parts: rst from to then from to its rst successor in A

, and then to

) or

l) (this is why we chose k

to be the second element of A

), but adding the last theory does not

change the sum hence Th
m
[,

)) = Th
m
[,

l)) for every l A

. By a similar argument we
can show that for every l A

we have Th
m
[

), ) = Th
m
[

l), ).
Next, thin out B
1

to get B

so that () will hold for lev() = lev() = n, lev( ) = n 1 i.e.

k B

, =

k) t

is constant
let k

to be the second element of B

. Dene the subtree A

to be A

k
: k

k B

.
Clearly A

satises ().
Step n 1: we have reached e), the root of
n
. B
0
e
, B
1
e
, . . . , B
(n1)(n2)
e
= B
e
are dened as
before, taking care of () for all the possibilities of the form lev() = lev() = k, lev( ) = l
(some thinning outs are not necessary as they have been taken care of in previous steps), k
e
, t
e
and
A
e
= A are dened as well.
Final Step: By our construction, for every < in A, with lev() = lev(), Th
m
(C;

P)
[,)
depends only on lev( ) and we dene t
k
by:
t
k
:= Th
m
[, ) where < , lev() = lev() = n, lev( ) = n k
By our choice of n we have some k < l n with t
k
= t
l
. Lets show how to get a suitable
homogeneous subset B of T (C) from this.
Example 1. t
1
= t
2
Pick A with lev() = n2. The successors of in A have order type

and for every successor

of in A, its successors have order type . Dene:
B
1
:=
_

l)

l
) : l A

, l > k

_
and
B
2
:=
_

k) : k A

k
_
and let B = B
1
B
2
.
Clearly B
1
has order type

, B
2
has order type and B has order type Z. Moreover, for every
< B
1
we have Th
m
[, ) = t
1
(since lev( ) = n 1) and for every < B with B
2
we have Th
m
[, ) = t
2
(since lev( ) = n 2). By t
1
= t
2
we conclude:
( < B)
_
Th
m
(C;

P)
[,)
= t
1

.
Finding a homogeneous subset of C of order type Z, we can proceed as in claim 2.2 to get a
contradiction to (x, X,

P) denes a choice function on C.
Example 2. t
2
= t
3
Pick A with lev() = n 3. The successors of in A have order type and for every successor
of in A, its successors have order type

. Let :=

) (lev() = n 2), for l > k

l
:=

l) (lev(
l
) = n 2) and
l
:=
l

k

l
) (lev(
l
) = n 1). Dene:
B
1
:=
_

l
) : l A

, l > k

_
(B
1
has order type

). To dene B
2
we let, for l A

,
l
:=

l) (
l
are extensions of and
with lev(
l
) = n 1) and then extend each
l
to a
l
dened by
l
:=
l

k

l
). So
B
2
:=
_

l
: l A

_
11
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


and it has order type . B := B
1
B
2
has order type Z and we can easily check that for every

1
<
2
B we have Th
m
[
1
,
2
) = t
3
(as
1

2
= so lev(
1

2
) = n 3) and for every

1
<
2
B
2
we have Th
m
[
1
,
2
) = t
2
(as
1

2
= so lev(
1

2
) = n 2). By t
2
= t
3
we
conclude:
( < B)
_
Th
m
(C;

P)
[,)
= t
2

.
and we proceed as before.
What we did in both examples can be described as follows: we xed a node A and a successor
of , we extended the other successors of and the successors of in a canonical way, ( is
extended to

k

)) to nodes of level n. The result is a homogeneous subset of C of order type Z.

General case. l + 1 < r, t
l
= t
r
Let , A be such that lev() = lev() = n and lev( ) = n r, so Th
m
[, ) = t
r
. Then nd
A with < < , lev() = n, lev( ) = n (l + 1) and lev( ) = n r. What we get is
the following equation:
t
r
= Th
m
[, ) = Th
m
[, ) +Th
m
[, ) = t
l+1
+ t
r
but t
r
= t
l
hence
() t
l
= t
l+1
+t
l
Imitate this computation: let , A be such that lev() = lev() = n and lev() = n(l +1),
so Th
m
[, ) = t
r
and nd A with < < , lev() = n, lev( ) = n (l + 1) and
lev( ) = n l. What we get is the following equation:
t
l+1
= Th
m
[, ) = Th
m
[, ) +Th
m
[, ) = t
l+1
+t
l
hence
() t
l+1
= t
l+1
+ t
l
Combining () and () we get t
l+1
= t
l
. Now proceed as in example 1 (if l is odd) or as in example
2 (if l is even) by taking canonical extensions of successors to get the required homogeneous subset
B of order type Z.

Conclusion 3.10. For every m, l < there is an n < such that if C is a scattered chain and
Hdeg(C) n + 1 then C does not have a denable choice function by a formula with quantier
depth m and with l parameters.
Proof. Let n be larger than [T
m,l
[. Now if Hdeg(C) n+1 then we can embed
n
into C and
immitate the previous proof.

4. Wild trees
Intuitively, wild trees are trees that have a large amount of splitting (4.2(1)(i)) or have compli-
cated branches (4.2(1)(ii)(iii)), the next two denitions state this formally. Wild trees dont have
a denable choice function (4.6).
12
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Denition 4.1. Let (T, ) be a tree
(1) If A is an initial segment of T then top(A) is x T : (t A)[t x]. (Its a tree).
(2) Let A be an initial segment of T then the binary relation
0
A
on T A is dened by
x
0
A
y (t A)[t x t y]
(Its an equivalence relation that says x and y break A in the same place).
(3) Let A be an initial segment of T then the binary relation
1
A
on T A is dened by
x
1
A
y [x
0
A
y] & (z)[z x & z y & z
0
A
x]
(Its an equivalence relation that divides for every initial segment B A top(B)/
0
B
into
disjoint subtrees).
Denition 4.2. (1) A tree T is called wild if either
(i) sup
_
[top(A)/
1
A
[ : A T an initial segment
_

0
or
(ii) There is a branch B T and an embedding f: QB or
(iii) All the branches of T are scattered linear orders but sup
_
Hdeg(B) : B a branch of T
_
.
(2) A tree T is tame for (n

, k

) if the value in (i) is n

, (ii) does not hold and the value in (iii)

is k

(3) A tree T is tame if T is tame for (n

, k

) for some n

, k

.
Claim 4.3. If T is a wild tree and (1)(i) of 4.2 holds then no monadic formula (x, X,

P) denes
a choice function on T.
Proof. We will use the composition theorem for general successors 1.12.
Suppose (x, X,

P) denes a choice function on T and Th
n
(T; x, X,

P) computes . For an initial
segment A T let top(A)/sim
1
A
= T
i
: i I
A
, by our assumption, for every l < there is an
initial segment A T such that [I
A
[ > l. Choose a large enough l (see below) and a corresponding
A and for every i I
A
pick x
i
T
i
.
If l is larger than the number of possible theories (= [T
n,l(

P)
[) then there are i ,= j I
A
such
that Th
n
(T
i
; x
i
,

P) = Th
n
(T
j
; x
j
,

P) and lets assume that we have chosen such an l. Now let

R
1
= x
i
x
i
, x
j


P and

R
2
= x
j
x
i
, x
j


P. Apply 1.12: clearly
Th
m
(T
A
;

R
1
) = Th
m
(T
A
;

R
2
) = Th
m
(T
A
; , ,

P)
and easily
Th
m
(I
A
;

Q
A
(n,

R
1
) = Th
m
(I
A
;

Q
A
(n,

R
2
)
but by 1.12 these theories determine Th
n
(T; x
i
, x
i
, x
j
,

P) and Th
n
(T; x
j
, x
i
, x
j
,

P) hence
T [= (x
i
, x
i
, x
j
,

P) T [= (x
j
, x
i
, x
j
,

P)
a contradiction.

Claim 4.4. If T is a wild tree and (1)(ii) of 4.2 holds then no monadic formula (x, X,

P) denes
a choice function on T.
13
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Proof. Let B T be a branch that embeds Q. We will apply 1.13 and translate the choice
function on T to a choice function on B but by 2.4 there is no denable choice function on B.
So assume that (x, X,

P) denes a choice function on T and is determined by Th
n
(T; x, X,

P). By
1.13 there is an m < , a chain B

with (B, ) (B

, ) and a sequence of parameters

Q B

such
that from Th
m
(B

;

Q) we can compute Th
n
(T;

P). Dene, for B, f(, ) = Th
m
(B

;

Q)
[,)
.
f is an additive colouring hence by 2.3 there is X =
i

iZ
, of order type Z, homogeneous with
respect to f. As in the proof of 2.4 we have:
i, j Z T

(B

, X,

Q) = T

(B

, X,

Q)
and (by the moreover clause in 1.13) this implies
i, j Z T

(T;

, X,

P) = T

(T;

, X,

P).
Hence
i, j Z [T [= (

, X,

P) T [= (

, X,

P)]
and this contradicts chooses an element from X.

Claim 4.5. If T is a wild tree and (1)(iii) of 4.1 holds then no monadic formula (x, X,

P)
denes a choice function on T.
Proof. Similar to the previous proof.
By (1)(iii) for every m < there is a branch B T with Hdeg(B) > m. Use 1.13, 3.10 and the
proof of 3.9 to nd, for a suitable branch B, a homogeneous subset that contradicts the assumption
that (x, X,

P) denes a choice function on T.
The details are left to the reader.

We conclude
Theorem 4.6. T is a wild tree T does not have a monadically denable choice function.
Moreover, every candidate fails to choose from either linearily ordered subsets (4.4, 4.5) or anti-
chains (4.3).

5. Tame trees
By [GuSh]
>
2 does not have a denable choice function. To know if a tame tree T has a
denable choice function we just have to ask if there is an embedding of f:
>
2 T. If such an
embedding exists we use [GuSh] to show that T does not have one, if not, T has even a denable
well ordering.
Claim 5.1. Let T be a tree and F:
>
2 T be a tree embedding. Then no monadic formula
(x, X,

P) denes a choice function on T.
14
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Proof. We will use [GuSh] 1.15 and the notations of 1.14. First, we may assume w.l.o.g that T
has a root (adding a root will not eect the existence of a choice function) and that F(root(
>
2)) =
root(T). Now apply the proof in 5 of [GuSh]. From the proof there we learn that for every

Q
>
2
and m < there is an innite anti-chain Y
>
2 such that for every y Y there is y

,= y Y
with Th
m
(Bush >
2
(Y ); y,

Q) = Th
m
(Bush >
2
(Y ); y

, Y,

Q). In our context (F

(
>
2) = S T)
the result has the form:
() for every

Q S and m < there is an innite anti-chain Y S such that for every y Y
there is y

,= y Y with Th
m
(Bush
S
(Y ); y,

Q) = Th
m
(Bush
S
(Y ); y

,

Q).
Let (x, X,

P) be a candidate for a denition of a choice function on T and suppose Th
n
(T; x, X,

P)
decides . Let m < and

Q =

Q(n,

P) be as in 1.15 and Y S be the anti-chain from ().
Suppose T [= (y, Y,

P), by () we have y

Y as in there. Now Th
m
(Bush
S
(Y ); y,

Q) =
Th
m
(Bush
S
(Y ); y

,

Q) and by 1.15
Th
n
(T; y, Y,

P) = Th
n
(T; y

, Y,

P)
hence
T [= (y, Y,

P) T [= (y

, Y,

P)
hence fails to dene a choice function on T.

Denition 5.2. Let T be a tree. For T we dene by recursion a rank function rk() by:
rk() + 1 there are
1
,
2
T with
1
and
2
such that
1
,
2
are incomparable in
T and rk(
1
), rk(
2
)
If rk() is not dened we stipulate rk() = .
Fact 5.3. (1) T rk() rk() where has the obvious meaning.
(2)
>
2 is not embeddable in a tree T for every T, rk() ,=
Lemma 5.4. Let T be a tame tree. If
>
2 is not embeddable in T then there are

Q T and a
monadic formula (x, y,

Q) that denes a well ordering of T.
Proof. Assume T is (n

, k

) tame, recall denitions 4.1 and 4.2 and remember that for every
x T, rk(x) is well dened (i.e. < ). We will partition T into a disjoint union of sub-branches,
indexed by the nodes of a well founded tree and reduce the problem of a well ordering of T to a
problem of a well ordering of .
Step 1. Dene by induction on a set

Ord (this is a our set of indices), for every

dene a tree T

T and a branch A

.
= 0 :
0
is ), T

is T and A

is a branch (i.e. a maximal linearily ordered subset) of T.

= 1 : Look at (T A

)/
1
A

, its a disjoint union of trees and name it T

i
: i < i

), let

1
:= i) : i < i

and for every i)

1
let A
i
be a branch of T
i
.
= +1 : For

denote (T

)/
1
A
by T

i
: i < i

, let

i) :

, i < i

and choose A

i
to be a branch of A

i
.
limit: Let

=

Ord :
<

,
<
T

,= , let for

=
<
T

and
A

a branch of T

. (T

may be empty).
Now, at some stage [T[
+
we have

= and let =
<

. Clearly A

: is a
partition of T into disjoint sub-branches.
Notation: having two trees T and , to avoid confusion, we use x, y, s, t for nodes of T and , ,
for nodes of .
15
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Step 2. We want to show that

= hence is a well founded tree. Note that we made no

restrictions on the choice of the A

s and we add one now in order to make the above statement

true. Let

i) dene A
,i
to be the sub-branch t A

: (s A

i
)[rk(t) rk(s)] and

,i
to be rk(t) for some t A
,i
. By 5.5(1) and the inexistence of a stricly decreasing sequence of
ordinals, A
,i
,= and
,i
is well dened. Note also that s A

i
rk(s)
,i
.
Proviso: For every and i < i

the sub-branch A

i
contains every s T

i
with rk(s) =

,i
.
Following this we claim: does not contain an innite, stricly increasing sequence. Otherwise let

i<
be one, and choose s
n
A
n,n+1(n)
(so s
n
A
n
). Clearly rk(s
n
) rk(s
n+1
) and by the
proviso we get
rk(s
n
) = rk(s
n+1
) rk(s
n+1
) > rk(s
n+2
)
therefore rk(s
n
)
n<
contains an innite, stricly decreasing sequence of ordinals which is absurd.
Step 3. Next we want to make x and y belong to the same A

denable.
For each choose s

, and let Q T be the set of representatives. Let h: T

d
0
, . . . , d
n

1
be a colouring that satises: h
A

= d
0
and for every

i) , h
A

i
is
constant and, when j < i and s

j

0
A
s

j
we have h
A

i
,= h
A

j
. This can be done
as T is (n

, d

) tame.
Using the parameters D
0
, . . . , D
n

1
(x D
i
i h(x) = d
i
), we can dene

x, y A

by x, y are
comparable and the sub-branch [x, y] (or [y, x]) has a constant colour.
Step 4. As every A

has Hausdor degree at most k

, we can dene a well ordering of it using

parameters P

1
, . . . , P

k
and by taking

P to be the (disjoint) union of the

P

s we can dene a partial

ordering on T which well orders every A

.
By our construction if and only if there is an element in A

that breaks A

i.e. is above a
proper initial segment of A

. (Caution, if T does not have a root this may not be the case for )
and a < n

number of i)s and we may need parameters for expressing that). Therefore, as by step
3 being in the same A

is denable, we can dene a partial order on the sub-branches A

(or the
representatives s

) by A

.
Next, note that is an immediate successor of in is denable as a relation between s

and
s

hence the set A

+

:= A

i
is denable from s

. Now the order on A

induces an order
on s

i
/
0
A
which is can be embedded in the complition of A

hence has Hdeg k

. Using
additional parameters Q

1
, . . . , Q

, we have a denable well ordering on s

i
/
0
A
. As for the
ordering on each
1
A
equivalence class (nite with n

elements), dene it by their colours (i.e.

the element with the smaller colour is the smaller according to the order).
Using

D,

P, Q and

Q =

we can dene a partial ordering which well orders each A

+

in such a
way that every x A

is smaller then every s

i
.
Summing up we can dene (using the above parameters) a partial order on subsets of T that well
orders each A

, orders sub-branches A

, A

when the indices are comparable in and well orders

all the immediate successors sub-branches of a sub-branch A

.
Step 5. The well ordering of T will be dened by x < y
a) x and y belong to the same A

and x < y by the well order on A

; or
b) x A

, y A

and ; or
c) x A

, y A

, = in (dened as a relation between sub-branches),

i) ,

j)
and s

i
< s

j
in the order of A
+

.
16
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


Note, that < is a linear order on T and every A

is a convex and well ordered sub-chain. Moreover

< is a linear order on and the order on the s

s is isomorphic to a lexicographic order on .

Why is the above (which is clearly denable with our parameters) a well order? Because of the
above note and because a lexicographic ordering of a well founded tree is a well order, provided
that immediate successors are well ordered. In detail, assume X = x
i

i<
is a stricly decreasing
sequence of elements of T. Let
i
be the unique node in such that x
i
A
i
and by the above note
w.l.o.g i ,= j
i
,=
j
. By the well foundedness of and clause (b) we may also assume w.l.o.g
that the
i
s form an anti-chain in . Look at
i
:=
1

i
which is constant for innitely many is
and w.l.o.g equals to for every i. Ask:
() is there is an innite B such that i, j B x
i

0
A
x
j
?
If this occurs we have
1
,= with
1
such that for some innite B

B we have i
B

1

i
. (use the fact that
1
A
is nite). W.l.o.g B

= and we may ask if () holds for

1
. Eventually, since does not have an innite branch, we will have a negative answer to (). We
can conclude that w.l.o.g there is such that i ,= j x
i
,
0
A
x
j
i.e. the x
i
s break A

in
dierent places.
Dene now
i
to be the unique immediate successor of such that
i

i
. The set S = s
i

i<
A
+

is well ordered by the well ordering on A

+

and by clause (c) in the denition of <, x

i
> x
j

i
>
j
so S is an innite stricly decreasing subset of A
+

a contradiction.
This nishes the proof that there is a denable well order of T.

Finally we can conclude:

Theorem 5.5. Let T be a tree. If T is wild or T embeds
>
2 then there is no denable choice
function on T (by a monadic formula with parameters). If T is tame and does not embed
>
2 then
there even a denable well ordering of the elements of T by a monadic formula (with parameters)
(x, y,

P).

As mentioned in the introduction, a tree is tame [wild] [embeds

>
2] if and only if its completion is
tame [wild] [embeds
>
2]. Moreover looking at the proofs of 4.3, 4.4, 4.5 and 5.1 we note that the
counter-examples for the choice function problem are either anti-chains or linearily ordered subsets
of T. We conclude:
Conclusion 5.6. Let T be a tree and T

be its completion. Then the following are equivalent:

a) For some n, l < , for every anti-chain/branch A of T there is a monadic formula
A
(x, X,

P
A
)
with quantier depth n and l parameters from T, that denes a choice function from non empty
subsets of A.
b) There is a monadic formula, with parameters, (x, y,

P) that denes a well ordering of the
elements of T.
c) There is a monadic formula, with parameters,

(x, y,

P

) that denes a well ordering of the

elements of T

17
5
3
9


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
4
-
2
0







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
4
-
2
0


REFERENCES
[BL] J.R. B

UCHI and L.H. LANDWEBER, Solving sequential conditions by nite-state strategies,

Transactions of the American Mathematical Society, vol. 138 (1969), pp. 295311.
[Ha] F. HAUSDORFF, Grundz uge einer Theorie der geordnetn Mengen, Math. Ann., vol. 65
(1908), pp. 435505.
[Gu] Y. GUREVICH, Monadic Secondorder Theories, Model Theoretic Logics, (J. Barwise
and S. Feferman, editors), SpringerVerlag, Berlin 1985, pp. 479506.
[GuSh] Y. GUREVICH and S. SHELAH, Rabins Uniformization Problem, The Journal of
Symbolic Logic, vol. 48 (1983), pp. 11051119.
[Ra] M.O. RABIN, Decidability of second-order theories and automata on innite trees, Trans-
actions of the American Mathematical Society, vol. 141 (1969), pp. 135.
[Sh] S. SHELAH, The monadic Theory of Order, Annals of Mathematics, ser. 2, vol. 102
(1975), pp. 379419.
18