7

2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


On embedding models of arithmetic of
cardinality
1
into reduced powers
Juliette Kennedy

Department of Mathematics
University of Helsinki
Helsinki, Finland
Saharon Shelah

Institute of Mathematics
Hebrew University
Jerusalem, Israel
October 6, 2003
Abstract
In the early 1970s S.Tennenbaum proved that all countable models
of PA

+
1
Th(N) are embeddable into the reduced product N

/F,
where F is the conite lter. In this paper we show that if M is a
model of PA

+
1
Th(N), and |M| =
1
, then M is embeddable
into N

/D, where D is any regular lter on .

1 Preliminaries
Let LA, the language of arithmetic, be the rst order language with non-
logical symbols +, , 0, 1, . N denotes the standard LA structure. We shall
be concerned with the following theories: The theory
1
-Th(N), dened as
the set of all universal formulas true in the standard LA structure N. Hence-
forth we refer to theories satisfying
1
-Th(N) as Diophantine correct. We will
also refer to the theory PA

, which consists of the following axioms (x < y

abbreviates x y x ,= y):
1.) x, y, z((x +y) +z = x + (y +z))

Research partially supported by grant 40734 of the Academy of Finland.

Research partially supported by the United States-Israel Binational Science Founda-

tion. Publication number [728]
1
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


2.) x, y(x +y = y +x)
3.) x, y, z((x y) z = x (y z))
4.) x, y(x y = y x)
5.) x, y, z(x (y +z) = x y +x z)
6.) x((x + 0 = x) (x 0 = 0))
7.) x(x 1 = x)
8.) x, y, z((x < y y < z) x < z)
9.) x x x
10.) x, y(x < y x = y y < x)
11.) x, y, z(x < y x +z < y +z)
12.) x, y, z(0 < z x < y x z < y z)
13.) x, y(x < y z(x +z = y)
14.) 0 < 1 x(x > 0 x 1)
15.) x(x 0)
Thus the theory PA

is the theory of nonnegative parts of discretely

ordered rings. For interesting examples of these models see [2].
2 The Countable Case
We begin by presenting the two embedding theorems of Stanley Tennenbaum,
which represent countable models of PA

by means of sequences of real

numbers. We note that Theorem 1 follows from the
1
-saturation of the
structure N

/T ([1]), however we present Tennenbaums construction as it

constructs the embeddings directly:
We consider rst the reduced power (of LA structures) N

/T, where T
is the conite lter in the boolean algebra of subsets of N. Let A be the
standard LA structure with domain all nonnegative real algebraic numbers.
We also consider the reduced power A

/T.
If f is a function from N to N, let [f] denote the equivalence class of f in
N

/T. We use a similar notation for A

/T. When no confusion is possible,

we will use f and [f] interchangeably.
Theorem 1 (Tennenbaum) Let M be a countable Diophantine correct model
of PA

. Then M can be embedded in N

/T.
Proof. Let m
1
, m
2
, . . . be the distinct elements of M. Let P
1
, P
2
, . . . be
all polynomial equations over N in the variables x
1
, x
2
, . . . such that M [=
2
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


P
i
(x
1
/m
1
, x
2
/m
2
, . . .). Each system of equations P
1
P
n
has a solu-
tion in M. Thus, by Diophantine correctness, there is a sequence of natural
numbers v
1
(n), v
2
(n), . . . for which
N [= (P
1
P
n
)(x
1
/v
1
(n), x
2
/v
2
(n), . . .).
Note that if the variable x
i
does not appear in P
1
P
n
, then the choice
of v
i
(n) is completely arbitrary. Our embedding h : M N

/T is given
by:
m
i
[n.v
i
(n)].
In the gure below, the i-th row is the solution in integers to P
1
P
n
,
and the i-th column is h(m
i
).
m
1
m
2
. . . m
n
. . .
P
1
v
1
(1) v
2
(1) . . . v
n
(1) . . .
P
2
v
1
(2) v
2
(2) . . . v
n
(2) . . .
.
.
.
.
.
.
.
.
.
.
.
.
P
n
v
1
(n) v
2
(n) . . . v
n
(n) . . .
.
.
.
.
.
.
.
.
.
.
.
.
Note that if m
e
is the element 0
M
of M, then the polynomial equation
x
e
= 0 appears as one of the Ps. It follows that, for n suciently large,
v
e
(n) = 0. Thus h(0
M
) is the equivalence class of the zero function. Similarly,
h maps every standard integer of M to the class of the corresponding constant
function.
We show that h is a homomorphism. Suppose M [= m
i
+m
j
= m
k
. Then
the polynomial x
i
+x
j
= x
k
must be one of the Ps, say P
r
. If n r, then by
construction v
i
(n) +v
j
(n) = v
k
(n). Hence N

/T [= h(m
i
) +h(m
j
) = h(m
k
),
as required. A similar argument works for multiplication. Suppose M [=
m
i
m
j
. By an axiom of PA

, for some k, M [= m
i
+ m
k
= m
j
. Thus, as
we have shown, h(m
i
) +h(m
k
) = h(m
j
). It follows from the denition of the
relation in ^ that h(m
i
) h(m
j
).
To see that h is one to one, suppose that m
i
,= m
j
. Since in models of
PA

the order relation is total, we may assume that m

i
< m
j
. Again by the
axioms of PA

, we can choose m
k
such that m
i
+m
k
+1 = m
j
. As we have
shown, N

/T [= h(m
i
) + h(m
k
) + h(1) = h(m
j
). Since h(1) is the class of
the constant function 1, it follows that h(m
i
) ,= h(m
j
). 2
3
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


Corollary 2 Let M be a countable model of the
1
-Th(N). Then M can be
embedded in ^.
Proof. The models of the
1
-Th(N) are precisely the substructures of models
of Th(N). Thus, M extends to a model of PA

, which can be embedded in

^ as in Theorem 1. 2
Before turning to the theorem for the non-Diophantine correct case, we
observe rst that the given embedding depends upon a particular choice
of enumeration m
1
, m
2
, . . . of M, since dierent enumerations will in general
produce dierent polynomials. We also note that dierent choices of solution
yield dierent embeddings. Also, as we shall see below, we need not restrict
ourselves to Diophantine formulas: we can carry out the construction for LA
formulas of any complexity which hold in M.
We state the non-Diophantine correct case of the theorem:
Theorem 3 (Tennenbaum) Let M be a countable model of PA

. Then M
can be embedded in A

/T.
Proof. Given an enumeration m
1
, m
2
, . . . of M, we form conjunctions of
polynomial equations P
n
exactly as before. We wish to produce solutions
of P
1
P
n
in the nonnegative algebraic reals for each n. We proceed as
follows: The model M can be embedded in a real closed eld F by a stan-
dard construction. (Embed M in an ordered integral domain, then form the
(ordered) quotient eld, and then the real closure.) Choose k so large that
x
1
, . . . , x
k
are all the variables that occur in the conjunction P
1
P
n
.
The sentence x
1
. . . x
k
(P
1
P
n
x
1
0 x
2
0 x
k
0) is true in
M, hence in F. It is a theorem of Tarski that the theory of real closed elds is
complete. Thus, this same sentence must be true in the eld of real algebraic
numbers. This means we can choose nonnegative algebraic real numbers
v
1
(n), v
2
(n) . . . satisfying the conjunction P
1
P
n
. Let h : M A

/T
be given by
m
i
[n.v
i
(n)].
The proof that h is a homomorphism, and furthermore an embedding, pro-
ceeds exactly as before, once we note that the equivalence classes all consist
of nonnegative sequences of real algebraic numbers. 2
4
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


Remark 4 Under any of the embeddings given above, if M [= PA

then
nonstandard elements of M are mapped to equivalence classes of functions
tending to innity. Why? If f is a function in the image of M, and f
does not tend to innity, then choose an integer k such that f is less than k
innitely often. Since M [= PA

, either [f] [k] or [k] [f]. The second

alternative contradicts the denition of in ^. Hence [f] [k], i.e., [f] is
standard.
Remark 5 Let F be a countable ordered eld. Then F is embedded in R

/T,
where R is the eld of real algebraic numbers. The proof is mutatis mutandis
the same as in Theorem 3, except that due to the presence of negative elements
we must demonstrate dierently that the mapping obtained is one to one. But
this must be the case, since every homomorphism of elds has this property.
Remark 6 For any pair of LA structures A and B satisfying PA

, if A is
countable and if A satises the
1
-Th(B) then there is an embedding of A
into B

/T. In particular, if M is a model of PA

, then every countable

extension of M satisfying the
1
-Th(M) can be embedded in M

/T.
Remark 7 Given Theorem 1, one can ask, what is a necessary and sucient
condition for a function to belong to a model of arithmetic inside ^? For a
partial solution to this question, see [3].
3 The Uncountable Case
We now show that some of the restrictions of Theorem 1 can be to some
extent relaxed, i.e. we will prove Theorem 1 for models M of cardinality
1
and with an arbitrary regular lter D in place of the conite lter. We note
that for lters D on for which B

/D is
1
-saturated, where B is the two
element Boolean algebra, this follows from the result of Shelah in [5], that
the reduced power N

/D is
1
-saturated.
Our strategy is similar to the strategy of the proof of Theorem 1, in
that we give an inductive proof on larger and larger initial segments of the
elementary diagram of M. However we must now consider formulas whose
variables are taken from a set of
1
variables. This requires representing each
ordinal < in terms of nite sets u

n
, which sets determine the variables
handled at each stage of the construction. The other technicality we require
5
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


is the use of the following function g(x, y), which bounds the size of the
formulas handled at each stage of the induction.
Let h(n, m) = the total number of non-equivalent Diophantine formulas
(x
1
, . . . x
m
) of length n. Dene
g(n, n) = h(n, 0)
g(n, m1) = 2 +h(g(n, m), m) (g(n, m) + 3), for m n.
Theorem 8 Let M be a model of PA

of cardinality
1
which is Diophantine
correct and let D be a regular lter on . Then M can be embedded in N

/D.
Proof. Let M = a

: <
1
. Let A
n

n
be a family witnessing the
regularity of D. We dene, for each <
1
a function f

N
N
. Our
embedding is then a

[f

]. We need rst a lemma:

Lemma 9 There exists a family of sets u

n
, with <
1
, and n N, such
that for each n,
(i) [u

n
[ < n + 1
(ii) u

n
u

n+1
(iii)

n
u

n
= + 1
(iv) u

n
u

n
= u

n
( + 1)
(v) lim
n
|u

n
|
n+1
= 0
Suppose we have the lemma and suppose we have dened f

for all < .

We choose f

N
N
componentwise, i.e. we choose f

(n) for each n separately

so that f

(n) satises the following condition:

()
,n
: If = (..., x

, ...)
u

n
is a Diophantine formula, such that the length
of is g(n, [u

n
[), then
M [= (..., a

, ...)
u

n
N [= (..., f

(n), ...)
u

n
.
6
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


(A formula is said to be Diophantine if it has the form
x
0
, . . . , x
n1
(t
1
(x
0
, . . . , x
n1
) = t
2
(x
0
, . . . , x
n1
)),
where t
1
and t
2
are LA-terms.) Now suppose = 0, n > 0. By (ii) and (iii),
u
0
n
= 0 for all n. We claim that ()
0,n
holds for each n. To see this, x n
and let
= (x) [ M [= (a
0
), where is Diophantine with [[ g(n, 1).
This is up to equivalence a nite set of formulas. Also, M [= x

(x) [
and therefore by Diophantine correctness N [= x

(x) [ .
If k witnesses this formula, set f
0
(n) = k. Clearly now ()
0,n
holds.
Now assume ()
,n
holds for all < and for each n. We choose f

(n)
for each n as follows. Fix n < and let
= (x
0
, x
1
, . . . x
k
) [ M [= (a

, ..., a

, ...)
u

n
\{}
),
where is Diophantine and [[ g(n, [u

n
[),
for k = [u

n
[ 1 (the case that [u

n
[ 1 reduces to the previous case). This
is again a nite set of formulas, up to equivalence. Now
M [=

(a

, a

)
u

n
\{}
)[
and therefore
M [= x

(x, a

)
u

n
\{}
)[ .
Let

(x
1
, . . . , x
k
) x

(x, x
1
, . . . , x
k
)[ . Note that
[

[ 2 +h(g(n, [u

n
[), [u

n
[) (g(n, [u

n
[) + 3) = g(n, [u

n
[ 1).
Now let = max (u

n
). Then u

n
= u

n
, by (iv) of the lemma.
Since ()
,n
holds for each n and we know that M [=

(a

)
u

n
) then we
know by the induction hypothesis that N [=

(f

(n))
u

n
). Thus N [=
x

(x, ..., f

(n), ...)
u

n
)[ . Let k witness this formula and set
f

(n) = k. Then ()
,n
holds with a

[f

(n))].
7
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


This mapping is an embedding of M into N

/D. To see this, suppose

M [= (a

, a

, a

) where is the formula x

0
+ x
1
= x
2
. Let n = n
0
be
large enough so that [[ g(n
0
, [u

n
0
[), where is chosen so that u

n
0
contains
, , . Then for all n n
0
, ()
,n
holds and N [= f

(n) +f

(n) = f

(n), for
all n n
0
. Let A
n
1
be an element of the chosen regular family of D such
that A
n
1
0, . . . , n
0
= . Then since f

(n) + f

(n) = f

(n), for n n
0
,
this holds also for n A
n
1
. The proof that multiplication is preserved is
the same so we omit it. Finally, as in the countable case, we note that our
mapping is one-to-one and hence an embedding.
We now prove the lemma, by induction on . Let u
0
n
= 0 for all n < .
Case 1. is a successor ordinal, i.e. = +1. Let n
0
be such that n n
0
implies
|u

n
|
n
<
1
2
. then we set
u

n
=

n < n
0
u

n
n n
0
.
Then (i), (ii) and (iii) are trivial. Proof of (iv): Suppose u

n
= u

n
.
Case 1.1. = . Then
u

n
= u

n
= u

n
+ 1
= u

n
( + 1).
Case 1.2. u

n
. Then by the induction hypothesis
u

n
= u

n
( + 1)
= u

n
( + 1),
since u

n
implies .
Proof of (v):
lim
n
[u

n
[
n
lim
n
[u

n
[ + 1
n
= 0,
by the induction hypothesis.
8
7
2
8


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
1
1


Case 2. is a limit ordinal > 0. Let
n
be an increasing conal -sequence
converging to , for all <
1
. Let us choose natural numbers n
0
, n
1
, ... such
that
i
u

i+1
n
i
and n n
i+1
implies n
i
[u

i+1
n
[ < n. Now we let
u

n
= u

i
n
, if n
i
n < n
i+1
.
To prove (iv), let u

n
. We wish to show that u

n
= u

n
( + 1). Suppose
n
i
n < n
i+1
. Then u

n
= u

i
n
. But then u

n
implies u

i
n
.
If = , then as before, u

n
= u

n
( + 1). So suppose u

i
n
. Then
u

n
= u

i
( + 1) = u

n
( + 1).
Finally, we prove (iii): Let u

n
, a limit. Let n
i
be such that u

n
=
u

i
n
i
. We have u

i
n
i
= u

i+1
n
i
(
i
+ 1). Therefore u

i+1
n
i
= u

n+1
.
To prove (v), let > 0 and choose i so that n
i
> 2. We observe that n
i
n
implies
lim
n
[u

n
[
n
= lim
n
[u

i
n
[
n
+
1
n
<
1
n
i
+
1
n
< .
2
In [4] models of higher cardinality are considered, and embedding theo-
rems are obtained under a set theoretic assumption.
References
[1] B. J onsson and P. Olin, Almost direct products and saturation, Com-
positio Math., 20, 1968, 125132
[2] R. Kaye. Models of Peano Arithmetic. Oxford Logic Guides. Oxford:
Oxford University Press, 1991.
[3] J. Kennedy. On embedding models of arithmetic into reduced powers.
Ph.D. thesis, City University of New York Graduate Center, 1996.
[4] J. Kennedy and S. Shelah, On regular reduced products, to appear.
[5] S. Shelah, For what lters is every reduced product saturated?, Israel J.
Math., 12, 1972, 2331
9