This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

**MAURO DI NASSO AND MARCO FORTI
**

Abstract. We investigate order-theoretic properties of the nonstandard real

line, by isolating the basic order-types involved, and the possible relations

among them. We focus on order-isomorphisms between the inﬁnitesimals, the

inﬁnites, and the whole hyperreal line. We also compute the main cardinal

invariants of the order-topology of the nonstandard real line.

Introduction

An interest in the structural properties of the nonstandard (or hyper-) real line

∗

R arose since the beginnings of nonstandard analysis. In the seminal paper [30],

E. Zakon extensively studied

∗

R, both from the viewpoint of ordering and that of

topology. Some relevant problems were left open, and in the following years many

papers appeared were the subject was further investigated (see e.g. [23], [19], [17],

[11], [9], [7], [21], [13], and [28]).

In this paper we concentrate on the order-theoretic side. We isolate the basic

order-types originated by the structure of

∗

R, and we study the possible relations

among them. In particular, some implications will be shown not to hold in general

by constructing suitable models as counter-examples.

The notion of hyperreal line we are considering here follows the usual “external”

viewpoint to nonstandard analysis, e.g. as given by the so-called superstructure

approach.

1

In a general set-theoretic setting, all the results given here about the

hyperreal line, could be reformulated as (external) properties of the real line in some

elementary extension of V

ω+ω

, the “true” model of ZFC−Replacement. However,

for convenience, in this paper we shall stick to the most popular superstructure

approach to nonstandard analysis. Let us brieﬂy recall the basics of this approach.

The superstructure over a set X is the union V (X) =

n∈N

V

n

(X) where V

0

(X) =

X, and V

n+1

(X) = V

n

(X)∪℘(V

n

(X)), where ℘ is the powerset operator. The stan-

dard model of analysis is the superstructure V (R). A nonstandard model of analysis

is a superstructure V (

∗

R) which is a bounded elementary extension of the standard

model via a nonstandard embedding denoted by the symbol ∗

∗ : V (R) −→ V (

∗

R).

As usual in the literature, we are making a few blanket assumptions here, namely

1991 Mathematics Subject Classiﬁcation. 03H05; 03E04; 03C20; 06F30.

Work partially supported by Fondi di Ateneo 2000 grants.

1

We remark that the notion of hyperreal line in the Internal Set Theory IST of [22] is diﬀerent.

In fact, it is unique because of the axiomatic approach, and its order-properties are of a special

nature, due to the strong saturation given by the Idealization axiom. Order-properties of the

hyperreal line of BST (a “bounded version” of IST) are studied in [18] (see also [24]).

1

2 MAURO DI NASSO AND MARCO FORTI

• R and

∗

R are sets of atoms w.r.t. their respective superstructures, i.e. x ∩

V (R) = ∅ = y ∩ V (

∗

R) for any x ∈ R and y ∈

∗

R;

•

∗

R is the image of R under ∗;

•

∗

x = x for all x ∈ R.

Recall that a bounded elementary extension is a map where the usual elementary

extension property is assumed only for those formulae where all quantiﬁers occur

in the bounded forms ∀x ∈ y or ∃x ∈ y.

An element x ∈ V (

∗

R) is called internal if x ∈

∗

y for some y ∈ V (R), or,

equivalently, if x ∈ 1 =

n∈N

∗

V

n

(R).

Let κ be an inﬁnite cardinal. We say that a nonstandard model is κ-saturated if

every nonempty family / ⊂ 1 of internal sets of size [/[ < κ satisfying the ﬁnite

intersection property (FIP) has nonempty intersection. Recall that / has the FIP

if, for ﬁnitely many A

1

, . . . , A

n

∈ /, the intersection A

1

∩ . . . ∩ A

n

= ∅.

For more on the foundations and the basics of nonstandard analysis, including

the κ-saturation property, we refer to [1] Section 4.4. For an elementary algebraic

approach see [4]. Throughout the paper, by model we always mean a (standard or)

nonstandard model of analysis, as deﬁned above.

The paper is organized as follows. In Section 1, we introduce the basic order-

types involved in the study of the ordering of the hyperreal line, and we state

those relations among them which hold in any nonstandard model. This section

also contains a short treatment of the connections between these order-types and

the main cardinal invariants of the natural group topology of

∗

R. In Section 2 we

introduce the technical model-theoretic constructions that will be needed in the

sequel. In Section 3 we investigate those relations which depend on the choice of

the model. Final remarks and suggestions for further research are contained in

Section 4.

The authors are deeply indebted to Furio Honsell, Karel Hrb`a˘cek and Renling

Jin for many useful discussions and suggestions. The authors are also grateful to

the referee for several useful remarks.

1. The basic order types

We denote as usual by N, R, R

+

the standard natural, real, and positive real

numbers and by

∗

N,

∗

R,

∗

R

+

the nonstandard (or hyper-) natural, real, and positive

real numbers. When considering several models simultaneously, we adopt the usual

notation with the name of the model as a superscript. In the study of the order-

theoretic properties of the nonstandard real line, we shall consider order-types of

various basic internal and external sets. We list them below, in order to ﬁx our

notation:

• ω, λ are the order-types of N and R, respectively;

• ρ is the order-type of

∗

R;

• µ is the order-type of M = ¦x ∈

∗

R [ ∀r ∈ R

+

[x[ < r¦ (the monad of 0);

• γ is the order-type of G = ¦x ∈

∗

R [ ∃r ∈ R

+

[x[ < r¦ (the galaxy of 0).

ON THE ORDERING OF THE NONSTANDARD REAL LINE 3

Notice that G and M are convex additive subgroups of

∗

R, hence for any x ∈

∗

R

the galaxy of x

G

x

= ¦t ∈

∗

R [ ∃r ∈ R

+

[t −x[ < r¦ = x +G

and the monad of x

M

x

= ¦t ∈

∗

R [ ∀r ∈ R

+

[t −x[ < r¦ = x +M

are order-isomorphic to G and M, respectively.

Moreover the quotient groups

∗

R/G, G/M,

∗

R/M are ordered groups w.r.t. the

induced orderings, since Gand M are convex. We say that x, y ∈

∗

R

+

have the same

rank if x/y is neither inﬁnite nor inﬁnitesimal. Notice that the (nonstandard) ex-

ponential function provides an isomorphism from the ordered additive group

∗

R/G

onto the the ordered multiplicative group of ranks

∗

R

+

/(G ` M)

+

. In particular,

the order-type of the positive part of

∗

R/G can be seen as coding the ranks (orders)

of inﬁnite, and similarly, the negative part of

∗

R/G can be seen as coding the ranks

(orders) of inﬁnitesimal. Therefore we introduce:

• θ is the order-type of the orders of inﬁnite, or equivalently of (

∗

R/G)

+

.

We recall some basic facts about order-types that will be used in the sequel

without explicit reference.

Let σ and τ be the order-types of (S, <) and (T, <) respectively.

• the sum σ + τ is the order-type of the disjoint union (S ¬ T, <), where

the ordering between pairs of elements in S and pairs of elements in T is

preserved, and where each element of S precedes all elements of T.

• The product στ is the order-type of the Cartesian product (S T, ≺) with

the antilexicographic order, i.e. (a, b) ≺ (a

, b

) if either b < b

, or b = b

and a < a

.

• the converse order-type σ

∗

of σ is the order-type of (S, >).

2

We say that σ is symmetric if σ

∗

= σ.

• For all order-types σ, τ one has:

σ(τ +τ

) = στ +στ

, (σ +τ)

∗

= τ

∗

+σ

∗

, and (στ)

∗

= σ

∗

τ

∗

.

In the following proposition we list some well-known properties and relations

among the order-types that we have introduced (see e.g. [30]).

Proposition 1.1.

(1) The order-types λ, ρ, γ, µ are symmetric and dense without endpoints; θ is

dense without endpoints.

(2) λ and ρ are the order-types of any open interval in R and

∗

R, respectively.

(3) The order-type of G/M is λ, hence γ = µλ.

(4) The order-type of

∗

R/G is θ

∗

+ 1 + θ, hence ρ = γ(θ

∗

+ 1 + θ) and the

order-type of

∗

N is ω + (ω

∗

+ω)θ.

(5) θ is the order-type of ¦x ∈

∗

R

+

/(G` M)

+

[ x > 1¦ (the ranks of inﬁnite).

(6) θ

∗

is the order-type of ¦x ∈

∗

R

+

/(G` M)

+

[ x < 1¦ (the ranks of inﬁnites-

imal).

(7) The order-type of

∗

R/M is λ(θ

∗

+ 1 +θ).

2

This asterisk symbol is not to be confused with the asterisk symbol denoting the nonstandard

embedding (the ﬁrst is placed up right, the latter is placed up left).

4 MAURO DI NASSO AND MARCO FORTI

(8) µ = γθ + 1 +γθ

∗

.

Let us now turn to the order-types of intervals in

∗

R/G. To this end we introduce:

• θ

x

is the order-type of ¦G

y

∈ (

∗

R/G)

+

[ G

y

< G

x

¦, the initial segment of

θ determined by any inﬁnite x ∈

∗

R

+

.

Some basic facts are the following:

Proposition 1.2.

(1) The order-type θ

x

is symmetric for all x.

(2) Any open interval (G

y

, G

y+x

) in

∗

R/G has order-type θ

x

.

(3) The order-type of any end-interval (G

x

, +∞) of

∗

R/G is θ, hence θ =

θ

x

+ 1 +θ.

Proof. The translations and the change-of-sign map are isometries of

∗

R, hence

they induce order-isomorphisms on the quotient modulo G. Properties 2 and 3

follow by using suitable translations, and Property 1 is obtained by composing the

change-of-sign map with a suitable translation.

The problem of the symmetry of the order-type θ arose since the beginnings of

nonstandard analysis, and it has been posed explicitly by E. Zakon in the quoted

paper [30], together with the connected question as to whether θ = θ

x

for all (some)

x. We consider the following possible identities:

(A1) θ = θ

∗

, i.e. the orders of inﬁnite and of inﬁnitesimal are isomorphic.

(A2) ∀x(θ = θ

x

), i.e. any initial interval of (

∗

R/G)

+

is isomorphic to (

∗

R/G)

+

.

(A3) ∃x(θ = θ

x

), i.e. some initial interval of (

∗

R/G)

+

is isomorphic to (

∗

R/G)

+

.

In this vein, following the analogy with well-known properties of the ordering of

the standard and nonstandard real lines, we also consider the identities:

(A4) θ = θ

∗

+ 1 +θ, i.e.

∗

R/G is order-isomorphic to its positive part.

(A5) θ = θ + 1 + θ, i.e. initial intervals and end-intervals of

∗

R/G are order-

isomorphic.

Theorem 1.3. The conditions (A1), (A2), (A3), (A4), and (A5) are equivalent in

any nonstandard model.

Proof. The implication (A2)⇒(A3) is trivial. The implications (A3)⇒(A1), (A3)⇒(A4),

and (A3)⇒(A5) are immediate consequences of the symmetry of θ

x

and of the iden-

tity θ = θ

x

+ 1 +θ of Proposition 1.2.

Condition (A5) means essentially that there is an initial segment of θ which is

order-isomorphic to θ, hence (A3). Since (A4) yields immediately the symmetry of

θ, we are left to show that (A1)⇒(A2). So, assume (A1). Then, for any x,

θ = θ

∗

= (θ

x

+ 1 +θ)

∗

= θ

∗

+ 1 +θ

∗

x

= θ + 1 +θ

x

.

Hence θ

x

is an open end-interval of θ, which is of type θ by Proposition 1.2, and

(A2) follows.

ON THE ORDERING OF THE NONSTANDARD REAL LINE 5

In the cited paper [30], the question is raised as to whether the monads are order-

isomorphic to the whole nonstandard real line. In order to better investigate this

question, we introduce the following identities, which parallel (A1)–(A5) above.

(B1) µ = ρ, i.e. the monad is isomorphic to the whole line.

(B2) µ = µ + 1 + µ, i.e. the monad is isomorphic both to its positive and its

negative part.

(B3) µ = γθ, i.e. the monad is isomorphic to the positive inﬁnites.

(B4) ρ = γθ, i.e. the hyperreal line is isomorphic to the positive inﬁnites.

(B5) γθ = γθ

∗

, i.e. the positive and the negative inﬁnites are isomorphic.

Theorem 1.4. The conditions (B1), (B2), (B3), (B4), and (B5) are equivalent in

any nonstandard model.

Proof. The implications (B1)⇒(B2) and (B3)⇒(B5) are immediate.

(B2)⇒(B3) follows from point 8 of Proposition 1.1, because every initial interval

of µ has order-type γθ.

Moreover, by the same point, (B4) implies µ = γθ + 1 + γθ

∗

= ρ + 1 + ρ

∗

= ρ,

that is (B1).

Finally, every open end-interval of γθ has order-type ρ. So, if γθ is symmetric,

then also its initial intervals have order-type ρ, hence γθ = ρ + 1 + ρ = ρ, and

(B5)⇒(B4) follows.

As a straightforward consequence of the above theorems, we get the

Corollary 1.5. The following implications hold in any nonstandard model:

(1) θ = θ

∗

=⇒ µ = ρ = γθ.

(2) µ = ρ =⇒ cof θ = cof θ

∗

.

In Section 3, we shall show that neither of the above implications can be reversed,

in general. Since coﬁnalities have particular relevance for our investigation, we state

here the following straightforward identities:

Remark 1.6.

(1) cof λ = cof γ = ω.

(2) cof ρ = cof

∗

N = cof θ.

(3) cof µ = cof θ

∗

= cof θ

x

.

We conclude this section with some interesting topological properties of the hy-

perreal line. The hyperreal line

∗

R carries a natural order topology which coincides

with the natural group topology having the convex subgroups as a neighborhood ba-

sis of 0. This topology on

∗

R is usually called the Q-topology. A detailed study of

the Q-topology lies outside the scope of this paper (see, e.g. [25, 30]). We restrict

ourselves here to consider its main cardinal invariants, as introduced in [5]. We

shall see that all of them are explicitly computable in terms of the order-type θ.

Let X be a topological space of cardinality k. We consider the following topo-

logical invariants of X:

6 MAURO DI NASSO AND MARCO FORTI

χ : the character, i.e. the least cardinal κ s.t. any point has neighborhood basis

of size κ.

ψ : the pseudocharacter, i.e. the least cardinal κ s.t. any point is the intersection

of a family of neighborhoods of size κ.

τ : the tightness, i.e. the least cardinal κ s.t. for any set C and any point x ∈ C,

x lies in the closure of some subset of C of size κ.

∆ : the diagonal weight, i.e. the least size of a family of open subsets of X

2

whose intersection is the diagonal.

d : the density, i.e. the least size of a dense subset.

s : the spread, i.e. the supremum of the sizes of discrete subsets.

e : the extent, i.e. the supremum of the sizes of discrete closed subsets.

l : the Lindel¨of number, i.e. the least cardinal κ s.t. any open cover has a

subcover of size κ.

c : the cellularity, i.e. the supremum of the sizes of disjoint families of open

sets.

w : the weight, i.e. the least size of an open basis.

n : the net weight, i.e. the least size of a family generating all open sets.

u : (only for Tychonoﬀ spaces) the uniform weight, i.e. the least size of a uni-

formity basis inducing the topology of X.

In many cases also the corresponding hereditary invariants are of interest. Thus,

if x is any of the above cardinals, then hx denotes the “hereditary x”, i.e.

hx(X) = sup¦x(Y ) [ Y ⊆ X¦.

We are now ready to compute all these cardinals in terms of θ, namely

Theorem 1.7. The following equalities hold in any nonstandard model:

(1) χ = ψ = τ = u = ∆ = cof θ.

(2) w = n = d = hd = s = hs = l = hl = c = hc = e = he = [θ[.

(3) [θ[ ≤ k ≤ [θ[

cof θ

.

Proof. In any topological space one has hc = he = hs = s (see [5]). Moreover the

following inequalities are stated in Problem 3.12.4 of [5] as holding in any linearly

ordered topological space:

w = n ≥ d = hd ≥ c = hl = hc ≥ l ≥ e and χ = ψ = τ.

Now [θ[ is both the cardinality of the hypernatural and of the hyperrational

numbers. Therefore [θ[ ≥ w since the open intervals with hyperrational endpoints

are a basis of the order-topology of

∗

R. Moreover e ≥ [θ[ since

∗

N is a closed

discrete subspace of

∗

R. All equalities of Point 2 are thus proved.

In order to get Point 1, observe ﬁrst that the hypermetric entourages U

=

¦(x, y) [ [x−y[ < ¦, as runs over a coinitial subset of

∗

R

+

, are a nested uniformity

basis inducing the group topology of

∗

R. Hence

∗

R is a κ-metric space in the sense

of [10] (with κ = cof θ), and so u ≤ cof θ.

ON THE ORDERING OF THE NONSTANDARD REAL LINE 7

Moreover no family of neighborhoods of a point x can have intersection ¦x¦ if its

size is less than cof θ. It follows that ∆ ≥ ψ ≥ cof θ. Combining these inequalities

with ∆ ≤ u, which holds in any uniform space, one gets all of Point 1.

Finally, the ﬁrst inequality of Point 3 is obvious, and the second one follows from

the fact that any point of

∗

R is the intersection of a suitable κ-sequence of open

intervals with hyperrational endpoints.

2. Constructing models of nonstandard analysis

In this section we introduce the technical machinery that will be used in the

next section. All our constructions of models will be obtained by means of suitable

iterations of two special kinds of elementary extensions, namely ultrapowers and

internal ultrapowers.

We assume that the reader is familiar with the ultrapower construction (see e.g.

[1] Section 4.1). If D is an ultraﬁlter over the cardinal κ, we denote by ´

κ

D

the

ultrapower of ´ modulo D.

We shall also consider the more general notion of limit ultrapower.

Deﬁnition 2.1. Let F be a ﬁlter of equivalences over κ. The limit ultrapower

´

κ

D

[F is the submodel of ´

κ

D

whose universe consists of the D-equivalence classes

f

D

of functions f : κ → ´ which are “compatible with F”, i.e. such that

¦(i, j) ∈ κ κ [ f(i) = f(j)¦ ∈ F.

The diagonal embedding d : ´ → ´

κ

D

[F is still deﬁned, and it satisﬁes Los

theorem (see e.g. [1] Section 6.4).

The notion of limit ultrapower is crucial in our context because of H.J. Keisler’s

characterization theorem:

Theorem 2.2 (see [20]). The map ∗ : V (R) → V (

∗

R) is a nonstandard embedding

if and only if there exist

• an ultraﬁlter D over some cardinal κ;

• a ﬁlter F of equivalences over κ;

• an isomorphism π of the bounded limit ultrapower

n∈N

V

n

(R)

κ

D

[F onto

the internal submodel 1 ⊂ V (

∗

R)

such that the following diagram commutes:

V (R) 1

n∈N

V

n

(R)

κ

D

[F

d

π

→ V (

∗

R)

∗

d

d

d

E

In the sequel we shall directly identify nonstandard models with limit ultrapowers

of V (R). This can be done without loss of generality. In fact, on the one hand,

every nonstandard embedding is (up to isomorphism) the diagonal embedding of

the standard model into a bounded limit ultrapower by the above theorem. On

the other hand, given a limit ultrapower V (R)

κ

D

[F, the corresponding nonstandard

embedding is the composition ∗ = π ◦ d : V (R) → V (

∗

R), where

• d : V (R) →

n∈N

V

n

(R)

κ

D

[F is the diagonal embedding;

8 MAURO DI NASSO AND MARCO FORTI

• π :

n∈N

V

n

(R)

κ

D

[F → T is a Mostowski collapse such that

– π induces a bijection of R

κ

D

[F onto

∗

R,

–

∗

R is a set of atoms w.r.t. its superstructure,

– π(d(r)) = r for every r ∈ R.

Notice that T ⊂ V (

∗

R) is transitive up to elements in

∗

R, i.e. t ⊂ T for all t ∈ T

with t / ∈

∗

R.

Deﬁnition 2.3. Let ´ = V (R)

ν

E

[F be a nonstandard model, and let D be an

ultraﬁlter over a cardinal κ. The internal ultrapower of ´ modulo D is the model

[´

κ

D

] = ( V (R)

κ

D

)

ν

E

[F.

The internal embedding

e : ´ −→ [´

κ

D

]

maps the E-equivalence class f

E

of any function f : ν → V (R) (compatible with

F) to the class ϕ

E

of the function ϕ = d ◦ f, where d : V (R) → V (R)

κ

D

is the

diagonal embedding.

It is immediate to verify that e is an elementary embedding. When no ambiguity

can arise, we identify every nonstandard real x ∈

∗

R

M

with its diagonal image

d(x) ∈

∗

R

U

in the ultrapower | = ´

κ

D

, so as to have

∗

R

M

⊂

∗

R

U

. Similarly, we

identify each x ∈

∗

R

M

with e(x) ∈

∗

R

V

in the internal ultrapower 1 = [´

κ

D

].

In the next lemma we itemize a few basic properties that will be useful in the

sequel. A detailed study of this kind of extentions can be found in [23].

Lemma 2.4. Let D be a countably incomplete ultraﬁlter over a cardinal κ and let

´ be a model. Let | = ´

κ

D

be the ultrapower of ´ modulo D and let 1 = [´

κ

D

]

be the corresponding internal ultrapower.

(1) If cof θ

M

> κ then

∗

N

M

is coﬁnal in

∗

N

U

, hence cof θ

M

= cof θ

U

.

(2) Let S be an initial cut of

∗

N

M

(or

∗

R

M

). Then S

κ

D

is an initial cut of

∗

N

U

(or

∗

R

U

, respectively).

(3) θ

M

is bounded from below in θ

U

, hence there are elements in the monad

x ∈ M

U

such that x > x

for all x

∈ M

M

.

(4) There are elements x ∈ θ

U

such that [θ

x

U

[ ≤ 2

κ

, hence cof θ

∗U

≤ 2

κ

.

(5)

∗

N

M

is an initial cut of

∗

N

V

, hence cof θ

∗M

= cof θ

∗V

, and the monad

M

M

is coﬁnal in M

V

.

In order to get models with the desired structural properties, we concentrate on

iterated ultrapowers and iterated internal ultrapowers given by regular ultraﬁlters.

Let ( be any ﬁxed (standard or nonstandard) model. Let δ be a limit ordinal

and let T = '(D

α

, κ

α

)`

α<δ

be a δ-sequence where every D

α

is a regular ultraﬁlter

over the cardinal κ

α

.

Deﬁnition 2.5. The models |((, δ, T) and 1((, δ, T) are the direct limits of the

systems '|

α

; d

α

`

α<δ

and respectively '1

α

; e

α

`

α<δ

, where:

• |

0

= 1

0

= (;

• |

α+1

= |

α

κ

α

D

α

is the ultrapower of |

α

modulo D

α

.

• d

α

: |

α

→ |

α+1

is the diagonal embedding.

ON THE ORDERING OF THE NONSTANDARD REAL LINE 9

• 1

α+1

= [1

α

κ

α

D

α

] is the internal ultrapower of 1

α

modulo D

α

.

• e

α

: 1

α

→ 1

α+1

is the internal embedding.

• For limit β < δ, |

β

and 1

β

are the direct limits of '|

α

; d

α

`

α<β

and

'1

α

; e

α

`

α<β

, respectively.

Notice that the models obtained at each limit step are isomorphic to limit ul-

trapowers, hence the internal ultrapower is well-deﬁned. We summarize in the

following lemmata some basic properties of these models. The proofs require only

straightforward applications of Lemma 2.4 and Deﬁnition 2.5.

Lemma 2.6. Let | = |((, δ, T) be the δ-iterated ultrapower of ( modulo T =

'(D

α

, κ

α

)`

α<δ

.

(1) cof θ

∗U

= cof δ.

(2) If cof θ

G

> κ

α

for all α, then

∗

N

G

is coﬁnal in

∗

N

U

, hence cof θ

U

= cof θ

G

.

(3) If κ

α

≥ cof θ

U

α

for all α, then cof θ

U

= cof δ. In particular this holds when

κ

α

=

α

(ν),

3

with ν ≥ cof θ

G

.

(4) Let ν ≤ cof δ. If ¦α [ |

α

is ν-saturated ¦ is coﬁnal in δ, then | is ν-

saturated.

In the next section we shall use good ultraﬁlters in order to obtain saturated

models. For the deﬁnition we refer to [1], where also the following fundamental

facts are proved:

• Every ultrapower modulo a countably incomplete ν-good ultraﬁlter is ν-

saturated (provided the size of the language is not larger than ν).

• Over any cardinal κ there exist κ

+

-good countably incomplete ultraﬁlters.

Lemma 2.7. Let 1 = 1((, δ, T) be the δ-iterated internal ultrapower of ( modulo

T = '(D

α

, κ

α

)`

α<δ

. Then

(1) cof θ

V

= cof δ.

(2)

∗

N

V

α

is an initial segment of

∗

N

V

for all α < δ, hence θ

x

V

= θ

x

V

α

for all

x ∈ 1

α

. In particular cof θ

∗V

= cof θ

∗V

α

= cof θ

∗G

for all α < δ.

3. Model-sensitive order-theoretic properties

In this section we study order-theoretic properties which depend on the choice

of the model. To this aim we use the construction techniques introduced above.

First of all we build symmetric models of arbitrary coﬁnality.

Theorem 3.1. Let κ be a regular cardinal. There exist κ-saturated models where

θ = ρ and cof θ = κ. In particular θ = θ

∗

is compatible with arbitrary values of

cof θ.

Proof. Let ( be a model and put κ

α

=

α

([([). Let | = |((, κ, T) be the κ-

iterated ultrapower of ( modulo T = '(D

α

, κ

α

)`

α<κ

, where D

α

is a countably

incomplete κ

+

α

-good ultraﬁlter over κ

α

. Then cof θ = cof θ

∗

= κ and | is κ-

saturated by Lemma 2.6.

3

Recall that

α

(ν) is inductively deﬁned by

0

(ν) = ν,

α+1

(ν) = 2

α

(ν)

and, for limit β,

β

(ν) = sup{

α

(ν) | α < β}.

10 MAURO DI NASSO AND MARCO FORTI

We are left to prove that θ = ρ. We want to show that there exists an order-

isomorphism Ψ : X →

∗

R, where X ⊂

∗

R is a set of representatives for θ (i.e.

G

x

= G

y

for all x = y in X, and θ = ¦G

x

[ x ∈ X¦).

For each α < κ, put

∗

R

α

=

∗

R

U

α

. By transﬁnite induction, we deﬁne:

• sets X

α

⊂

∗

R

α

for even α,

• order-preserving maps f

α

: X

α

→

∗

R

α+1

for even α, and

• order-preserving maps g

α

:

∗

R

α

→ X

α+1

for odd α

in such a way that the following properties are satisﬁed:

(1) X

α

is a set of representatives for θ

U

α

;

(2) g

α+1

(f

α

(b)) = b for all b ∈ X

α

(α even);

(3) f

α+1

(g

α

(a)) = a for all a ∈

∗

R

α

(α odd).

If α = 0, pick X

0

a set of representative for θ

U

0

and let f

0

: X

0

→

∗

R

1

be the

identity map.

Now assume α + 1 is odd. Let F(

∗

R) be the (internal) collection of all internal

order-preserving partial functions from

∗

R into itself, and, for ﬁxed a, b ∈

∗

R and

n ∈ N, put:

• Λ(a) = ¦g ∈ F(

∗

R) [ f

α

(a) ∈ dom g ⇒g(f

α

(a)) = a¦;

• Γ(b) = ¦g ∈ F(

∗

R) [ b ∈ dom g¦;

• ∆(n) = ¦g ∈ F(

∗

R) [ ∀x, y ∈ dom g (x = y ⇒[g(x) −g(y)[ > n)¦.

Since both

∗

R and

∗

R/G are dense without endpoints, the family

T = ¦Λ(a) [ a ∈ X

α

¦ ∪ ¦Γ(b) [ b ∈

∗

R

α+1

¦ ∪ ¦∆(n) [ n ∈ N¦

satisﬁes the FIP. Moreover [T[ ≤ [|

α+1

[ = κ

α+1

, hence, by κ

+

α+1

-saturation, there

is a function g in |

α+2

s.t. g ∈

**T. Since g is deﬁned for all b ∈
**

∗

R

α+1

, we can

take as g

α+1

the restriction g

∗

R

α+1

. Such a g

α+1

satisﬁes the desired property.

Now, assume that α + 2 is even. First, pick X

α+2

⊃ X

α

a set of representatives

for θ

U

α+2

, and put

Λ

(a) = ¦f ∈ F(

∗

R) [ g

α+1

(a) ∈ dom f ⇒ f(g

α+1

(a)) = a¦

and

T

= ¦Λ

(a) [ a ∈

∗

R

α+1

¦ ∪ ¦Γ(b) [ b ∈ X

α+2

¦

Proceeding as above we can see that the family T

has the FIP. Since [T

[ ≤

[|

α+2

[ = κ

α+2

, by κ

+

α+2

-saturation we can pick a function f in |

α+3

s.t. f ∈

T

.

Then dom f includes X

α+2

, and f

α+2

can be deﬁned as the restriction of f to

X

α+2

, thus satisfying the required conditions.

If β is a limit (hence even), let

X

β

=

¦X

α

[ α even < β ¦ and f

β

=

¦f

α

[ α even < β ¦.

By induction hypothesis X

β

is a set of representatives for θ

U

β

, and the functions

f

α

are coherent, hence f

β

satisﬁes the desired properties. Clearly, conditions 1-3

above imply that for each limit β the function f

β

is an order-isomorphism of X

β

onto

∗

R

β

, whose inverse is

¦g

α

[ α odd < β ¦.

It follows at once that X =

¦X

α

[ α even < κ ¦ is a set of representatives for

θ and that Ψ =

¦f

α

[ α even < κ¦ is an order-isomorphism from X onto

∗

R,

whose inverse is Ψ

−1

=

¦g

α

[ α odd < κ¦.

**Next we show that coinitiality and coﬁnality of θ can be arbitrarily chosen.
**

ON THE ORDERING OF THE NONSTANDARD REAL LINE 11

Theorem 3.2. Let κ

1

and κ

2

be regular cardinals, and put κ = min¦κ

1

, κ

2

¦.

There exist κ-saturated models such that:

(i) cof θ

∗

= κ

1

and cof θ = κ

2

.

Moreover, if ξ, η are cardinals such that η > ξ ≥ κ

1

, then the models can be chosen

so as to satisfy

(ii) [θ

x

[ = ξ

<κ

1

and [θ

y

[ = η

<κ

1

for suitable x, y. (ξ

ω

and respectively η

ω

if

κ

1

= ω).

Proof. Assume ﬁrst that κ = ζ

+

is a successor cardinal. Take ( a ground model

where cof θ = κ

2

. Consider | = |((, δ, T) where cof δ = κ

1

and T is constantly

equal to (D, ζ), D a ﬁxed ζ

+

-good ultraﬁlter over ζ. Then | satisﬁes the conditions

required by Lemma 2.6, hence it is κ saturated and satisﬁes (i).

In order to satisfy condition (ii), it suﬃces to choose δ = η + ξ + κ

1

(ordinal

sum). Simply pick a positive inﬁnite x in |

η+1

preceding all those in |

η

and an

inﬁnite y ∈

∗

R

U

1

preceding all those in (. Then

[θ

x

[ = sup¦[α[

ζ

[ η +α < δ¦ = ξ

ζ

and [θ

y

[ = sup¦[α[

ζ

[ α < δ¦ = η

ζ

.

Let us now assume that κ =

α<κ

ζ

α

is a regular uncountable limit cardinal

(i.e. weakly inaccessible). We slightly modify the preceding construction. Take (

a ground model where cof θ = κ

2

. Let δ be a limit ordinal with cof δ = κ

1

and let

ϕ : δ → κ be a function such that ϕ

−1

(α) is coﬁnal in δ for all α < κ. For α < κ

let D

α

be a ζ

+

α

-good ultraﬁlter over ζ

α

and put T = '(D

ϕ(β)

, ζ

ϕ(β)

)`

β<δ

.

Then | = |((, δ, T) is κ-saturated and satisﬁes conditions (i) by Lemma 2.6.

Condition (ii) can be met by taking ( and δ exactly as in the successor case.

We are left with κ = ω. We do not need to care about saturation, since every

nonstandard model is ℵ

0

-saturated.

Suppose ﬁrst that κ

2

> κ

1

= ω. Then we can take ζ = ω and proceed exactly

as in the successor case.

Finally suppose that κ

1

≥ κ

2

= ω. Then we can start with a model ( where

cof θ

∗

= κ

1

and property (ii) is satisﬁed. In order to adjust the coﬁnality of θ,

we take an iteration of ω internal ultrapowers 1 = 1((, ω, T) where T is any

sequecence. In this way, by Lemma 2.7, we preserve the initial part of θ and get

cof θ = ω.

**Remark that κ-saturation is best possible in the above theorem, because neither
**

cof θ nor cof θ

∗

can be less than κ in a κ-saturated model. The goal of the construc-

tions above was precisely that of getting the maximal saturation with prescribed

coﬁnalities. Giving up with saturation, we could have chosen any model of the form

1(|((, δ

1

, T

1

), δ

2

, T

2

) with cof δ

i

= κ

i

.

The importance of κ-saturation in our context is witnessed by the following

lemma, which provides a useful suﬃcient condition for the identity µ = ρ.

Lemma 3.3. Assume that cof θ = cof θ

∗

= κ. If the model is κ-saturated, then

µ = ρ.

Proof. It is enough to show that there is an order-isomorphism between the set of

positive inﬁnitesimals M

+

and

∗

R

+

. Let ¦a

α

[ α < κ¦ ⊂ M

+

and ¦b

α

[ α < κ¦ ⊂

∗

R

+

be increasing coﬁnal κ-sequences. By transﬁnite induction, we deﬁne a family

¦f

α

[ α < κ¦ of internal functions such that for every α:

12 MAURO DI NASSO AND MARCO FORTI

(1) f

α

: (0, a

α

] → (0, b

α

] is an order-isomorphism.

(2) f

α

extends f

β

for all β < α.

Pick f

0

an internal order-isomorphism between the intervals (0, a

0

] and (0, b

0

]

(say f

0

(x) = x

b

0

a

0

).

At the inductive step α > 0, proceed as follows. For each β < α, let

F

β

= ¦f : (0, a

α

] → (0, b

α

] [ f is an internal order-isomorphism extending f

β

¦.

Notice that all F

β

are internal and nonempty. For instance, the function ϕ

β

ϕ

β

(x) =

f

β

(x) if x ∈ (0, a

β

]

(x −a

β

)

b

α

−b

β

a

α

−a

β

+b

β

if x ∈ (a

β

, a

α

]

belongs to F

β

.

Notice also that F

β

⊆ F

β

for β

< β < α, because f

β

extends f

β

by the

inductive hypothesis. In particular, F

β

1

∩. . . ∩F

β

n

= F

max {β

1

,...,β

n

}

, and the FIP

holds. Moreover, since [¦F

β

: β < α¦[ = [α[ < κ, by κ-saturation we can pick

f

α

∈

β<α

F

β

. Clearly f

α

satisﬁes the conditions 1 and 2.

The union f =

α<κ

f

α

yields the desired order-isomorphism.

Corollary 3.4. Independently of the value of cof θ = cof θ

∗

,

µ = ρ = γθ =⇒ θ = θ

∗

.

Proof. By Lemma 3.3, the models of (ii) Theorem 3.2 with κ

1

= κ

2

= κ satisfy the

equality µ = ρ together with ∃x∃y(θ

x

= θ

y

), which contradicts θ = θ

∗

, by Theorem

1.3.

The next theorem will yield that also the second implication of Corollary 1.5

cannot be reversed. In the proof of the theorem we shall need the notion of type of

a gap. Therefore we recall the following deﬁnition

Deﬁnition 3.5. Let (X, Y ) be a partition of the ordered set (S, <).

• The partition (X, Y ) is a gap of S if X has no greatest element, Y has no

least element, and X < Y (i.e. x < y for all x ∈ X and y ∈ Y ).

• The gap (X, Y ) has type (κ, ζ) if cof X = κ and cof Y

∗

= ζ.

• The gap (X, Y ) is generated by A ⊂ S if

X = ¦x ∈ S [ x ≤ a for some a ∈ A¦.

Theorem 3.6. Let κ be a regular cardinal and assume κ > 2

ζ

. Then there exist

ζ

+

-saturated nonstandard models where cof θ = cof θ

∗

= κ, but ρ = µ.

ON THE ORDERING OF THE NONSTANDARD REAL LINE 13

Proof. Let ( be a (2

ζ

)

+

-saturated model where cof θ = κ. (Such a model exists for

κ > 2

ζ

.) Let D be a countably incomplete ζ

+

-good ultraﬁlter over ζ. Let T be the

κ-sequence constantly equal to (D, ζ). We claim that the model | = |((, κ, T)

satisﬁes the thesis.

By Lemma 2.6, the model | is ζ

+

-saturated, and satisﬁes cof θ

∗

= cof θ = κ.

We are left to show that ρ = µ in |. By contradiction, assume that there is an

order-isomorphism Ψ :

∗

R → M between the nonstandard real line and the monad

of zero in |. In the following we denote by

∗

R

α

and M

α

the hyperreal line and

the monad in |

α

, respectively. Now

∗

R

0

is coﬁnal in

∗

R and, for every α < κ, M

α

is not coﬁnal in M

α+1

, by Lemmata 2.4 and 2.6. Therefore we can pick increasing

ζ

+

-sequences ¦x

α

[ α < ζ

+

¦ in

∗

R

0

and ¦β

α

[ α < ζ

+

¦ in κ such that

Ψ(x

α+1

) = y

α+1

∈ M

β

α+1

is greater than all elements of M

β

α

In

∗

R

0

the ζ

+

-sequence ¦x

α

[ α < ζ

+

¦ generates a gap (X, X

) of type (ζ

+

, ξ),

with ξ > 2

ζ

, by (2

ζ

)

+

-saturation. Since ζ < ζ

+

< ξ, the ζ

+

-sequence ¦x

α

[ α < ζ

+

¦

generates a gap of the same type in the iterated ultrapower | as well.

On the other hand, the ζ

+

-sequence ¦y

α

[ α < ζ

+

¦ is coﬁnal in M

β

, where

β = sup¦β

α

[ α < ζ

+

¦ < κ. So it generates a gap of type (ζ

+

, ω) on

∗

R

β

and

cosequently one of type (ζ

+

, η) with ζ < η ≤ 2

ζ

in

∗

R

β+1

. The latter type remains

unchanged up to

∗

R, by the argument above. This is a contradiction because, by

order-isomorphism, both gaps must have the same type.

4

**We are now ready to deal with the converse of the second implication of Corollary
**

1.5

Corollary 3.7.

(1) If cof θ = cof θ

∗

= ω, then µ = ρ = γθ.

(2) If cof θ > 2

ω

, then cof θ = cof θ

∗

=⇒ µ = ρ.

Proof. Point 1 follows from Lemma 3.3, since all nonstandard models are ω-saturated.

Point 2 is an immediate consequence of the theorem above for ζ = ω.

4. Final remarks

The consistency and the mutual relations of the order-theoretic properties (A1)−

(A5) and (B1) −(B5) of Section 1, have been studied by several authors since the

problem has been explicitly posed in the paper of E. Zakon [30]. We recall here

only the main achievements which are known to us.

The existence of models where coﬁnality and coinitiality of the inﬁnites can be

arbitrarily ﬁxed has been ﬁrst proved by K. Potthoﬀ in [23], where construction

techniques substantially equivalent to those of Section 2 are developed.

The consistency of θ = θ

∗

has been ﬁrst proved, in [17] and [29], only for fully

saturated models, hence under heavy set-theoretic assumptions and with strong

limitations on size and coﬁnality of θ. It is worth mentioning that Lemma 3.2 of

[17] claims that cof θ = cof θ

∗

is equivalent to ρ = µ, contradicting Point 2 of our

Corollary 3.7. But the purported inductive proof works only for successor steps,

4

The argument of this proof has been suggested to the ﬁrst author by Renling Jin in a personal

communication.

14 MAURO DI NASSO AND MARCO FORTI

and so it proves in fact only Point 1 of our corollary. Notice that this fault aﬀects

also the proofs of some true equivalences among the properties (B1) −(B5), which

are derived there from this false lemma.

The ﬁrst general constructions allowing for symmetric inﬁnites with arbitrary

coﬁnalities can be found in [9], where also several implications among the properties

(A1) −(A5), (B1) −(B5) are proved (see also [7], where Theorems 1.3, 3.1 and 3.2

of the present paper are stated without proof).

We think that the main questions concerning the above order-theoretic properties

which remain open are the following:

5

(1) Are there models where θ = θ

∗

= ρ?

(2) Are there models where ω < cof θ = cof θ

∗

≤ 2

ω

and µ = ρ?

(3) Are there models where ∀x∀y(θ

x

= θ

y

) but θ = θ

∗

?

We conjecture that the answers to these three questions are all in the positive.

For instance, it seems that the model 1(|((, κ, T), ω, c), where |((, κ, T) is the

model in the proof of Theorem 3.1 and c is constantly equal to some nonprincipal

ultraﬁlter over ω, should satisfy

∀x∀y(θ

x

= θ

y

) together with cof θ

∗

= κ and cof θ = ω.

Another interesting topic, stemming from [30] and not considered here, is the

study of the types of the gaps that can occur in the hyperreal line and their con-

nections with the order-type θ. This topic is strictly connected to properties of

the Q-topology and of other natural non-Hausdorﬀ topologies of the hyperreal line.

Several interesting results in this direction can be found in the papers cited above,

and, among others, also in [16],[9],[21],[13],[14],[27]. (We are considering these ques-

tions in a paper under construction.)

In this paper, all models have been explicitly deﬁned by means of the ultrapower

construction. The idea of using iterations of ultrapowers and internal ultrapowers

has been used extensively in the literature. See e.g. the excellent paper [21] where

these techniques are exploited in order to solve the diﬃcult problem of ﬁnding

suﬃciently saturated models where the hyperreal line is Scott complete.

After the pioneering article of Henson [12], where the Isomorphism Property was

introduced, a new interesting approach has been developed, namely that of isolating

“special” principles which hold in suitable classes of models. Among them, we single

out the the Special Model Axiom of Ross [26], the Resplendency Property of Jin

and Shelah [15], and the recent combinatorial ∆-principles of the ﬁrst author and

Hrb`a˘cek [2, 3]. These “special” properties have been showed fruitful to show the

consistency of several properties in nonstandard analysis, and we think that they

could be successfully applied also to the topics treated in this paper.

5

Notice that in this paper we are considering only results that are provable in ZFC. The

relative consistency of θ = θ

∗

= ρ with ZFC+“there exists a huge cardinal” is a consequence of

the consistency of Foreman’s nonregular ultraﬁlter D over ω

2

, see [6]. In fact the ultrapowers ω

ω

2

D

and ω

1

ω

2

D

have sizes ℵ

1

and ℵ

2

, respectively. Therefore, in the corresponding nonstandard model,

θ = θ

∗

holds because θ is a ℵ

1

-saturated order-type of size ℵ

1

. On the other hand θ = ρ since

|ρ| ≥ ℵ

2

.

ON THE ORDERING OF THE NONSTANDARD REAL LINE 15

In particular, the ∆-principles have the advantage of being formulated in purely

combinatorial terms, and they can be directly applied without any use of model-

theoretic constructions, such as ultrapowers. For instance, by simple combinatorial

arguments, it is shown in [3] that ∆

2

proves θ = θ

∗

.

References

[1] C.C. Chang and H.J. Keisler, Model Theory (3rd edition), North-Holland, Amsterdam, 1990.

[2] M. Di Nasso, The generic ﬁlter property in nonstandard analysis, Ann. of Pure and Appl.

Logic 111 (2001), 23–37.

[3] M. Di Nasso and K. Hrb`a˘cek, Combinatorial principles in nonstandard analysis, Ann. of

Pure and Appl. Logic, to appear.

[4] M. Di Nasso and Y. Zhang, Nonstandard analysis and an application to the symmetric group

on natural numbers, this volume.

[5] R. Engelking, General Topology, Polish Scientiﬁc Publishers, Warszawa 1977.

[6] M. Foreman, An ℵ

1

-dense ideal on ℵ

2

, Isr. J. Math. 108 (1998), 253–290.

[7] M. Forti, On the order structure of the nonstandard real axis, in Atti degli incontri di Logica

Matematica, vol.3 (R. Ferro and A. Zanardo, eds.), Padova 1987, 199-205.

[8] M. Forti, On the order structure of the nonstandard real axis (abstract), J. of Symb. Logic

52 (1987), 317–318.

[9] M. Forti and F. Honsell, Sull’ordinamento dei numeri reali non-standard (italian), Atti Acc.

Naz. Lincei Rend. Cl. Sci. Fis. Mat. Nat. S. 8 74 (1985), 1–8.

[10] M. Forti and F. Honsell, Choice Principles in Hyperuniverses, Ann. Pure Appl. Logic 77

(1996), 35–52.

[11] W.S. Hatcher and C. Laﬂamme, On the order structure of the hyperreal line, Zeitschr. fur

Math. Logik und Grund. der Math. 29 (1983), 197–202.

[12] C.W. Henson, The isomorphism property in nonstandard analysis and its use in the theory

of Banach space, J. of Symb. Logic 39 (1974), 717–731.

[13] R. Jin, Cuts in hyperﬁnite time lines, J. of Symb. Logic 57 (1992), 522–527.

[14] R. Jin, U-monad topologies of hyperﬁnite time lines, J. of Symb. Logic 57 (1992), 534–539.

[15] R. Jin and S. Shelah, The strength of the isomorphism property, J. of Symb. Logic 59 (1994),

292–301.

[16] S. Kamo, Non-standard real number systems with regular gaps, Tsukuba J. Math. 5 (1981),

21–24.

[17] S. Kamo, Nonstandard natural number systems, J. of Symb. Logic 46 (1981), 365–376.

[18] V. Kanovei, M. Reeken, Mathematics in a nonstandard world. I, Math. Jap. 45 (1997),

369–408.

[19] H.J. Keisler, Monotone complete ﬁelds, in Victoria Symposium on Nonstandard Analysis

(A. Hurd, P. Loeb, eds.), Berlin 1974, 113–115.

[20] H.J. Keisler, Foundations of Inﬁnitesimal Calculus, Prindle, Weber and Schmidt, Boston,

1976.

[21] H.J. Keisler and J.H. Schmerl, Making the hyperreal line both saturated and complete, J. of

Symb. Logic 56 (1991), 1016–1025.

[22] E. Nelson, Internal Set Theory; a new approach to nonstandard analysis, Bull. Amer. Math.

Soc. 83 (1977), 1165–1198.

[23] K. Potthoﬀ, Ordnungseigenschaften von nichstandardmodellen, in Theory of Sets and Topol-

ogy (in honour of Felix Hausdorﬀ) (G. Asser, J. Flaschmeyer, and W. Rinow, eds.), Berlin

1972, 403–426.

[24] M. Prokhorova, External set properties in IST (abstract), Logic Colloquium 2001, Volume

of abstracts, Vienna 2001.

[25] A. Robinson, Non-standard Analysis, North Holland, Amsterdam 1966.

[26] D. Ross, The special model axiom in nonstandard analysis, J. of Symb. Logic 55 (1990),

1233–1242.

[27] S. Shelah, Vive la diﬀ´erence II. The Ax-Kochen isomorphism theorem, Israel J. of Math. 85

(1994), 351–390.

[28] N. Vakil, Nonstandard topologies with bases that consist only of standard sets, Proc. Amer.

Math. Soc. 129 (2001), 2075–2083.

16 MAURO DI NASSO AND MARCO FORTI

[29] S. Yang, D. Wang and J. Wang, On the order structure of the ω-regular extension hyperreal

numbers (chinese), Acta Math. Sin. 23 (1980), 658–667.

[30] E. Zakon, On the order structure of the nonstandard real axis, in Applications of Model

Theory to Algebra, Analysis and Probability (W.A.J. Luxemburg, ed.), New York 1969, 195–

227.

Dipartimento di Matematica Applicata “U. Dini”, Universit` a di Pisa, Italy.

E-mail address: dinasso@dma.unipi.it

Dipartimento di Matematica Applicata “U. Dini”, Universit` a di Pisa, Italy.

E-mail address: forti@dma.unipi.it

This section also contains a short treatment of the connections between these order-types and the main cardinal invariants of the natural group topology of ∗ R. . . equivalently. • ∗ x = x for all x ∈ R. In Section 2 we introduce the technical model-theoretic constructions that will be needed in the sequel. Let κ be an inﬁnite cardinal. . we introduce the basic ordertypes involved in the study of the ordering of the hyperreal line.r. • ∗ R is the image of R under ∗. In the study of the ordertheoretic properties of the nonstandard real line. ∩ An = ∅. . The authors are also grateful to the referee for several useful remarks. or. including the κ-saturation property. if x ∈ I = n∈N ∗ Vn (R).2 MAURO DI NASSO AND MARCO FORTI • R and ∗ R are sets of atoms w. their respective superstructures. . for ﬁnitely many A1 . Karel Hrb`˘ek and Renling ac Jin for many useful discussions and suggestions. In Section 3 we investigate those relations which depend on the choice of the model. • ρ is the order-type of ∗ R. Throughout the paper. and we state those relations among them which hold in any nonstandard model. and positive real numbers. For more on the foundations and the basics of nonstandard analysis. We say that a nonstandard model is κ-saturated if every nonempty family A ⊂ I of internal sets of size |A| < κ satisfying the ﬁnite intersection property (FIP) has nonempty intersection. ∗ R+ the nonstandard (or hyper-) natural. Recall that A has the FIP if. When considering several models simultaneously. . x ∩ V (R) = ∅ = y ∩ V (∗ R) for any x ∈ R and y ∈ ∗ R. We list them below. The basic order types We denote as usual by N.t. The authors are deeply indebted to Furio Honsell. An element x ∈ V (∗ R) is called internal if x ∈ ∗ y for some y ∈ V (R). Recall that a bounded elementary extension is a map where the usual elementary extension property is assumed only for those formulae where all quantiﬁers occur in the bounded forms ∀x ∈ y or ∃x ∈ y. • µ is the order-type of M = {x ∈ ∗ R | ∀r ∈ R+ |x| < r} (the monad of 0). ∗ R. For an elementary algebraic approach see [4]. In Section 1. i. in order to ﬁx our notation: • ω. respectively. Final remarks and suggestions for further research are contained in Section 4. The paper is organized as follows. by model we always mean a (standard or) nonstandard model of analysis. R+ the standard natural. R. we adopt the usual notation with the name of the model as a superscript. An ∈ A. . λ are the order-types of N and R. real. the intersection A1 ∩ .4.e. we shall consider order-types of various basic internal and external sets. we refer to [1] Section 4. as deﬁned above. • γ is the order-type of G = {x ∈ ∗ R | ∃r ∈ R+ |x| < r} (the galaxy of 0). and positive real numbers and by ∗ N. real. 1.

b) (a . b ) if either b < b . We recall some basic facts about order-types that will be used in the sequel without explicit reference. (5) θ is the order-type of {x ∈ ∗ R+ /(G \ M )+ | x > 1} (the ranks of inﬁnite). <) and (T.t. the order-type of the positive part of ∗ R/G can be seen as coding the ranks (orders) of inﬁnite.r. ) with the antilexicographic order. and (στ )∗ = σ ∗ τ ∗ . θ is dense without endpoints. (4) The order-type of ∗ R/G is θ∗ + 1 + θ. where the ordering between pairs of elements in S and pairs of elements in T is preserved. respectively. 2This asterisk symbol is not to be confused with the asterisk symbol denoting the nonstandard embedding (the ﬁrst is placed up right. Notice that the (nonstandard) exponential function provides an isomorphism from the ordered additive group ∗ R/G onto the the ordered multiplicative group of ranks ∗ R+ /(G \ M )+ . Therefore we introduce: • θ is the order-type of the orders of inﬁnite. Let σ and τ be the order-types of (S.1. <) respectively. the latter is placed up left). i. [30]). • the converse order-type σ ∗ of σ is the order-type of (S. τ one has: σ(τ + τ ) = στ + στ . hence ρ = γ(θ∗ + 1 + θ) and the order-type of ∗ N is ω + (ω ∗ + ω)θ. <). ρ. γ. Moreover the quotient groups ∗ R/G. In the following proposition we list some well-known properties and relations among the order-types that we have introduced (see e. y ∈ ∗ R+ have the same rank if x/y is neither inﬁnite nor inﬁnitesimal.ON THE ORDERING OF THE NONSTANDARD REAL LINE 3 Notice that G and M are convex additive subgroups of ∗ R. the induced orderings.e. We say that x. G/M . >). since G and M are convex. µ are symmetric and dense without endpoints.g. In particular. or equivalently of (∗ R/G)+ . Proposition 1. hence γ = µλ. and where each element of S precedes all elements of T . • the sum σ + τ is the order-type of the disjoint union (S T. ∗ R/M are ordered groups w. (7) The order-type of ∗ R/M is λ(θ∗ + 1 + θ). or b = b and a < a . • For all order-types σ.2 We say that σ is symmetric if σ ∗ = σ. (6) θ∗ is the order-type of {x ∈ ∗ R+ /(G \ M )+ | x < 1} (the ranks of inﬁnitesimal). • The product στ is the order-type of the Cartesian product (S × T. (3) The order-type of G/M is λ. (2) λ and ρ are the order-types of any open interval in R and ∗ R. . and similarly. respectively. (1) The order-types λ. (a. hence for any x ∈ ∗ R the galaxy of x Gx = {t ∈ ∗ R | ∃r ∈ R+ |t − x| < r} = x + G and the monad of x Mx = {t ∈ ∗ R | ∀r ∈ R+ |t − x| < r} = x + M are order-isomorphic to G and M . the negative part of ∗ R/G can be seen as coding the ranks (orders) of inﬁnitesimal. (σ + τ )∗ = τ ∗ + σ ∗ .

we are left to show that (A1)⇒(A2). . Proof. hence they induce order-isomorphisms on the quotient modulo G.2. hence (A3). To this end we introduce: • θx is the order-type of {Gy ∈ (∗ R/G)+ | Gy < Gx }. The implication (A2)⇒(A3) is trivial. i. (A2).2.e. and Property 1 is obtained by composing the change-of-sign map with a suitable translation. Condition (A5) means essentially that there is an initial segment of θ which is order-isomorphic to θ. So. and (A3)⇒(A5) are immediate consequences of the symmetry of θx and of the identity θ = θx + 1 + θ of Proposition 1. Since (A4) yields immediately the symmetry of θ. (A3) ∃x(θ = θx ). In this vein.4 MAURO DI NASSO AND MARCO FORTI (8) µ = γθ + 1 + γθ∗ . which is of type θ by Proposition 1. Hence θx is an open end-interval of θ. The implications (A3)⇒(A1). Proof. initial intervals and end-intervals of ∗ R/G are orderisomorphic. and (A2) follows. (2) Any open interval (Gy . (1) The order-type θx is symmetric for all x. (A3). The problem of the symmetry of the order-type θ arose since the beginnings of nonstandard analysis.3. i. i. (A5) θ = θ + 1 + θ.2. i. and it has been posed explicitly by E. The translations and the change-of-sign map are isometries of ∗ R. (A3)⇒(A4). (3) The order-type of any end-interval (Gx .e. ∗ R/G is order-isomorphic to its positive part. Properties 2 and 3 follow by using suitable translations. We consider the following possible identities: (A1) θ = θ∗ . Theorem 1. the orders of inﬁnite and of inﬁnitesimal are isomorphic. Some basic facts are the following: Proposition 1. Let us now turn to the order-types of intervals in ∗ R/G. together with the connected question as to whether θ = θx for all (some) x. we also consider the identities: (A4) θ = θ∗ + 1 + θ. i. assume (A1).e. (A2) ∀x(θ = θx ). ∗ θ = θ∗ = (θx + 1 + θ)∗ = θ∗ + 1 + θx = θ + 1 + θx .e. some initial interval of (∗ R/G)+ is isomorphic to (∗ R/G)+ . The conditions (A1). Gy+x ) in ∗ R/G has order-type θx . for any x. hence θ = θx + 1 + θ. any initial interval of (∗ R/G)+ is isomorphic to (∗ R/G)+ .e. Zakon in the quoted paper [30]. following the analogy with well-known properties of the ordering of the standard and nonstandard real lines. Then. and (A5) are equivalent in any nonstandard model. +∞) of ∗ R/G is θ. the initial segment of θ determined by any inﬁnite x ∈ ∗ R+ . (A4).

(B1) µ = ρ. (B4) implies µ = γθ + 1 + γθ∗ = ρ + 1 + ρ∗ = ρ. because every initial interval of µ has order-type γθ.e.ON THE ORDERING OF THE NONSTANDARD REAL LINE 5 In the cited paper [30].g. A detailed study of the Q-topology lies outside the scope of this paper (see. As a straightforward consequence of the above theorems. i. the positive and the negative inﬁnites are isomorphic. (B2). [25.e. The implications (B1)⇒(B2) and (B3)⇒(B5) are immediate. then also its initial intervals have order-type ρ. The hyperreal line ∗ R carries a natural order topology which coincides with the natural group topology having the convex subgroups as a neighborhood basis of 0. the monad is isomorphic to the positive inﬁnites. if γθ is symmetric. i.4. in general. (B3) µ = γθ.5. (B3). (3) cof µ = cof θ∗ = cof θx . i. the monad is isomorphic to the whole line. e. which parallel (A1)–(A5) above. Moreover. Proof. i. (2) µ = ρ =⇒ cof θ = cof θ∗ . We consider the following topological invariants of X: . The following implications hold in any nonstandard model: (1) θ = θ∗ =⇒ µ = ρ = γθ. that is (B1). the question is raised as to whether the monads are orderisomorphic to the whole nonstandard real line.e.e. We shall see that all of them are explicitly computable in terms of the order-type θ. the monad is isomorphic both to its positive and its negative part. (2) cof ρ = cof ∗ N = cof θ. as introduced in [5]. In Section 3. and (B5)⇒(B4) follows. we introduce the following identities. (1) cof λ = cof γ = ω. Let X be a topological space of cardinality k. This topology on ∗ R is usually called the Q-topology.1. We restrict ourselves here to consider its main cardinal invariants. Since coﬁnalities have particular relevance for our investigation. (B2) µ = µ + 1 + µ. So. we shall show that neither of the above implications can be reversed. (B2)⇒(B3) follows from point 8 of Proposition 1. Theorem 1. we state here the following straightforward identities: Remark 1. 30]). (B5) γθ = γθ∗ . every open end-interval of γθ has order-type ρ. we get the Corollary 1. The conditions (B1). by the same point. i. and (B5) are equivalent in any nonstandard model. In order to better investigate this question. (B4). Finally. hence γθ = ρ + 1 + ρ = ρ. (B4) ρ = γθ. We conclude this section with some interesting topological properties of the hyperreal line.6. the hyperreal line is isomorphic to the positive inﬁnites.e.

e. the least cardinal κ s. the least cardinal κ s. the least size of a family generating all open sets. i.e.e. i. Thus. ψ : the pseudocharacter. then hx denotes the “hereditary x”. the least size of a uniformity basis inducing the topology of X.e. Proof. . i. the supremum of the sizes of discrete closed subsets. Now |θ| is both the cardinality of the hypernatural and of the hyperrational numbers. (3) |θ| ≤ k ≤ |θ| cof θ . any point has neighborhood basis of size κ. if x is any of the above cardinals. We are now ready to compute all these cardinals in terms of θ. All equalities of Point 2 are thus proved.t. any open cover has a o subcover of size κ. i.e. the least size of a family of open subsets of X 2 whose intersection is the diagonal. ∆ : the diagonal weight. y) | |x−y| < }. as runs over a coinitial subset of ∗ R+ . observe ﬁrst that the hypermetric entourages U = {(x.e. the least cardinal κ s.e. Moreover the following inequalities are stated in Problem 3.e. (2) w = n = d = hd = s = hs = l = hl = c = hc = e = he = |θ|.e.e.12. i. hx(X) = sup{x(Y ) | Y ⊆ X}. d : the density.t.t. s : the spread. In order to get Point 1. i.6 MAURO DI NASSO AND MARCO FORTI χ : the character.e. e : the extent. i. Therefore |θ| ≥ w since the open intervals with hyperrational endpoints are a basis of the order-topology of ∗ R. i. The following equalities hold in any nonstandard model: (1) χ = ψ = τ = u = ∆ = cof θ. i. and so u ≤ cof θ. Hence ∗ R is a κ-metric space in the sense of [10] (with κ = cof θ). for any set C and any point x ∈ C. u : (only for Tychonoﬀ spaces) the uniform weight.e. any point is the intersection of a family of neighborhoods of size κ. Moreover e ≥ |θ| since ∗ N is a closed discrete subspace of ∗ R. x lies in the closure of some subset of C of size κ. τ : the tightness. the least size of an open basis. In any topological space one has hc = he = hs = s (see [5]). namely Theorem 1.e.7. the least size of a dense subset. w : the weight. i.4 of [5] as holding in any linearly ordered topological space: w = n ≥ d = hd ≥ c = hl = hc ≥ l ≥ e and χ = ψ = τ. i. i. the supremum of the sizes of disjoint families of open sets.t. are a nested uniformity basis inducing the group topology of ∗ R. the least cardinal κ s. In many cases also the corresponding hereditary invariants are of interest. i. the supremum of the sizes of discrete subsets. n : the net weight. l : the Lindel¨f number. c : the cellularity.

The map ∗ : V (R) → V (∗ R) is a nonstandard embedding if and only if there exist • an ultraﬁlter D over some cardinal κ.g. If D is an ultraﬁlter over the cardinal κ. Combining these inequalities with ∆ ≤ u. On the other hand. one gets all of Point 1. Deﬁnition 2. such that {(i. Finally. on the one hand. every nonstandard embedding is (up to isomorphism) the diagonal embedding of the standard model into a bounded limit ultrapower by the above theorem. This can be done without loss of generality. the ﬁrst inequality of Point 3 is obvious.1. Keisler’s characterization theorem: Theorem 2. given a limit ultrapower V (R)κ |F .g. We assume that the reader is familiar with the ultrapower construction (see e. Let F be a ﬁlter of equivalences over κ. and the second one follows from the fact that any point of ∗ R is the intersection of a suitable κ-sequence of open intervals with hyperrational endpoints. [1] Section 4. [1] Section 6. All our constructions of models will be obtained by means of suitable iterations of two special kinds of elementary extensions. where • d : V (R) → n∈N Vn (R)κ |F is the diagonal embedding. D .4). The diagonal embedding d : M → Mκ |F is still deﬁned. The limit ultrapower Mκ |F is the submodel of Mκ whose universe consists of the D-equivalence classes D D fD of functions f : κ → M which are “compatible with F ”.J. the corresponding nonstandard D embedding is the composition ∗ = π ◦ d : V (R) → V (∗ R). which holds in any uniform space. we denote by Mκ the D ultrapower of M modulo D. • an isomorphism π of the bounded limit ultrapower n∈N Vn (R)κ |F onto D the internal submodel I ⊂ V (∗ R) such that the following diagram commutes: n∈N Vn (R)κ |F D d π d d ∗ E I → V (∗ R) d V (R) In the sequel we shall directly identify nonstandard models with limit ultrapowers of V (R). namely ultrapowers and internal ultrapowers. • a ﬁlter F of equivalences over κ. In fact.2 (see [20]). It follows that ∆ ≥ ψ ≥ cof θ.e. and it satisﬁes Los D theorem (see e. i. We shall also consider the more general notion of limit ultrapower.ON THE ORDERING OF THE NONSTANDARD REAL LINE 7 Moreover no family of neighborhoods of a point x can have intersection {x} if its size is less than cof θ. The notion of limit ultrapower is crucial in our context because of H. 2. j) ∈ κ × κ | f (i) = f (j)} ∈ F. Constructing models of nonstandard analysis In this section we introduce the technical machinery that will be used in the next section.1).

• Uα+1 = Uα καα is the ultrapower of Uα modulo Dα . A detailed study of this kind of extentions can be found in [23]. Similarly. When no ambiguity can arise. Deﬁnition 2. t ⊂ T for all t ∈ T with t ∈ ∗ R. (3) θM is bounded from below in θU . D – ∗ R is a set of atoms w. hence there are elements in the monad x ∈ M U such that x > x for all x ∈ M M . we D identify each x ∈ ∗ RM with e(x) ∈ ∗ RV in the internal ultrapower V = [Mκ ].5. and let D be an ultraﬁlter over a cardinal κ. Let U = Mκ be the ultrapower of M modulo D and let V = [Mκ ] D D be the corresponding internal ultrapower. The internal ultrapower of M modulo D is the model [Mκ ] = ( V (R)κ )νE |F. i. Let δ be a limit ordinal and let D = (Dα . its superstructure. Notice that T ⊂ V (∗ R) is transitive up to elements in ∗ R. D) are the direct limits of the systems Uα . (5) ∗ NM is an initial cut of ∗ NV . (4) There are elements x ∈ θU such that |θx U | ≤ 2κ . so as to have ∗ RM ⊂ ∗ RU . where: • U0 = V0 = G. δ. Then S κ is an initial cut of D ∗ U N (or ∗ RU .8 MAURO DI NASSO AND MARCO FORTI • π: n∈N Vn (R)κ |F → T is a Mostowski collapse such that D – π induces a bijection of Rκ |F onto ∗ R. / Deﬁnition 2. The models U(G. hence cof θ∗ M = cof θ∗ V . δ.t.3. hence cof θM = cof θU . In order to get models with the desired structural properties. Let D be a countably incomplete ultraﬁlter over a cardinal κ and let M be a model. (2) Let S be an initial cut of ∗ NM (or ∗ RM ). hence cof θ∗ U ≤ 2κ . respectively). κα ) α<δ be a δ-sequence where every Dα is a regular ultraﬁlter over the cardinal κα . we concentrate on iterated ultrapowers and iterated internal ultrapowers given by regular ultraﬁlters. dα α<δ and respectively Vα . we identify every nonstandard real x ∈ ∗ RM with its diagonal image d(x) ∈ ∗ RU in the ultrapower U = Mκ . (1) If cof θM > κ then ∗ NM is coﬁnal in ∗ NU .4. Let M = V (R)νE |F be a nonstandard model. eα α<δ . D In the next lemma we itemize a few basic properties that will be useful in the sequel. The internal embedding . Let G be any ﬁxed (standard or nonstandard) model. D D e : M −→ [Mκ ] D maps the E-equivalence class fE of any function f : ν → V (R) (compatible with F ) to the class ϕE of the function ϕ = d ◦ f . Lemma 2. where d : V (R) → V (R)κ is the D diagonal embedding. D • dα : Uα → Uα+1 is the diagonal embedding. D) and V(G. It is immediate to verify that e is an elementary embedding. – π(d(r)) = r for every r ∈ R. and the monad M M is coﬁnal in M V .e.r.

6. δ. α<β and Notice that the models obtained at each limit step are isomorphic to limit ultrapowers. We summarize in the following lemmata some basic properties of these models. First of all we build symmetric models of arbitrary coﬁnality. κ. In particular this holds when κα = α (ν). .4 and Deﬁnition 2. α+1 (ν) =2 α (ν) and. • For limit β < δ. There exist κ-saturated models where θ = ρ and cof θ = κ. for limit β. D) be the κiterated ultrapower of G modulo D = (Dα . In the next section we shall use good ultraﬁlters in order to obtain saturated models. then U is νsaturated. D • eα : Vα → Vα+1 is the internal embedding. (2) If cof θG > κα for all α.ON THE ORDERING OF THE NONSTANDARD REAL LINE 9 • Vα+1 = [Vα καα ] is the internal ultrapower of Vα modulo Dα . Let V = V(G. Uβ and Vβ are the direct limits of Uα . Model-sensitive order-theoretic properties In this section we study order-theoretic properties which depend on the choice of the model. (2) ∗ NVα is an initial segment of ∗ NV for all α < δ. then cof θU = cof δ. (3) If κα ≥ cof θUα for all α. To this aim we use the construction techniques introduced above. Then (1) cof θV = cof δ. Let U = U(G. Then cof θ = cof θ∗ = κ and U is κα saturated by Lemma 2. hence the internal ultrapower is well-deﬁned. Let G be a model and put κα = α (|G|). Let U = U (G. 0 (ν) = ν. eα α<β . κα ) α<δ . (1) cof θ∗ U = cof δ. Lemma 2. • Over any cardinal κ there exist κ+ -good countably incomplete ultraﬁlters. κα ) α<κ . D) be the δ-iterated internal ultrapower of G modulo D = (Dα . δ.1.6. (4) Let ν ≤ cof δ. respectively.3 with ν ≥ cof θG . Let κ be a regular cardinal. where Dα is a countably incomplete κ+ -good ultraﬁlter over κα . Theorem 3. 3. For the deﬁnition we refer to [1]. In particular cof θ∗ V = cof θ∗ Vα = cof θ∗ G for all α < δ. In particular θ = θ∗ is compatible with arbitrary values of cof θ. D) be the δ-iterated ultrapower of G modulo D = (Dα . where also the following fundamental facts are proved: • Every ultrapower modulo a countably incomplete ν-good ultraﬁlter is νsaturated (provided the size of the language is not larger than ν). hence cof θU = cof θG .7. Lemma 2. dα Vα . If {α | Uα is ν-saturated } is coﬁnal in δ. hence θx V = θx Vα for all x ∈ Vα . The proofs require only straightforward applications of Lemma 2. 3 Recall that β (ν) α (ν) = sup{ α (ν) is inductively deﬁned by | α < β}. then ∗ NG is coﬁnal in ∗ NU . κα ) α<δ . Proof.5.

Now assume α + 1 is odd. • order-preserving maps fα : Xα → ∗ Rα+1 for even α. by κ+ -saturation we can pick a function f in Uα+3 s. and θ = {Gx | x ∈ X}). hence fβ satisﬁes the desired properties. put ∗ Rα = ∗ RUα . b ∈ ∗ R and n ∈ N. f ∈ F . y ∈ dom g (x = y ⇒|g(x) − g(y)| > n)}. . α+2 Then dom f includes Xα+2 . whose inverse is {gα | α odd < β }. there α+1 is a function g in Uα+2 s. by κ+ -saturation. Clearly. thus satisfying the required conditions. Since g is deﬁned for all b ∈ ∗ Rα+1 . It follows at once that X = {Xα | α even < κ } is a set of representatives for θ and that Ψ = {fα | α even < κ} is an order-isomorphism from X onto ∗ R. • Γ(b) = {g ∈ F(∗ R) | b ∈ dom g}. the family F = {Λ(a) | a ∈ Xα } ∪ {Γ(b) | b ∈ ∗ Rα+1 } ∪ {∆(n) | n ∈ N} satisﬁes the FIP. and put Λ (a) = {f ∈ F(∗ R) | gα+1 (a) ∈ dom f ⇒ f (gα+1 (a)) = a} and F = {Λ (a) | a ∈ ∗ Rα+1 } ∪ {Γ(b) | b ∈ Xα+2 } Proceeding as above we can see that the family F has the FIP. whose inverse is Ψ−1 = {gα | α odd < κ}. where X ⊂ ∗ R is a set of representatives for θ (i. First. If β is a limit (hence even). Next we show that coinitiality and coﬁnality of θ can be arbitrarily chosen. By induction hypothesis Xβ is a set of representatives for θ Uβ . If α = 0.10 MAURO DI NASSO AND MARCO FORTI We are left to prove that θ = ρ. and fα+2 can be deﬁned as the restriction of f to Xα+2 . We want to show that there exists an orderisomorphism Ψ : X → ∗ R.t. hence. Since |F | ≤ |Uα+2 | = κα+2 . we deﬁne: • sets Xα ⊂ ∗ Rα for even α. • ∆(n) = {g ∈ F(∗ R) | ∀x. conditions 1-3 above imply that for each limit β the function fβ is an order-isomorphism of Xβ onto ∗ Rβ .e. and the functions fα are coherent. For each α < κ. (3) fα+1 (gα (a)) = a for all a ∈ ∗ Rα (α odd). Moreover |F| ≤ |Uα+1 | = κα+1 . assume that α + 2 is even. pick Xα+2 ⊃ Xα a set of representatives for θ Uα+2 . for ﬁxed a. put: • Λ(a) = {g ∈ F(∗ R) | fα (a) ∈ dom g ⇒ g(fα (a)) = a}. let Xβ = {Xα | α even < β } and fβ = {fα | α even < β }. By transﬁnite induction. Since both ∗ R and ∗ R/G are dense without endpoints. Now. Gx = Gy for all x = y in X. pick X0 a set of representative for θ U0 and let f0 : X0 → ∗ R1 be the identity map. we can take as gα+1 the restriction g ∗ Rα+1 . Let F(∗ R) be the (internal) collection of all internal order-preserving partial functions from ∗ R into itself. and. and • order-preserving maps gα : ∗ Rα → Xα+1 for odd α in such a way that the following properties are satisﬁed: (1) Xα is a set of representatives for θ Uα .t. Such a gα+1 satisﬁes the desired property. (2) gα+1 (fα (b)) = b for all b ∈ Xα (α even). g ∈ F.

δ.7. δ1 . Assume that cof θ = cof θ∗ = κ. D1 ). Let κ1 and κ2 be regular cardinals. by Lemma 2. Consider U = U(G. Let us now assume that κ = α<κ ζα is a regular uncountable limit cardinal (i. then µ = ρ. Proof.3. The importance of κ-saturation in our context is witnessed by the following lemma. D) where D is any sequecence. Proof. because neither cof θ nor cof θ∗ can be less than κ in a κ-saturated model. which provides a useful suﬃcient condition for the identity µ = ρ. Take G a ground model where cof θ = κ2 .2. since every nonstandard model is ℵ0 -saturated. D2 ) with cof δi = κi . Take G a ground model where cof θ = κ2 . η are cardinals such that η > ξ ≥ κ1 . weakly inaccessible). D) where cof δ = κ1 and D is constantly equal to (D. In order to satisfy condition (ii). Finally suppose that κ1 ≥ κ2 = ω. it suﬃces to choose δ = η + ξ + κ1 (ordinal sum). we could have chosen any model of the form V(U (G. By transﬁnite induction. For α < κ + let Dα be a ζα -good ultraﬁlter over ζα and put D = (Dϕ(β) . Moreover. Let {aα | α < κ} ⊂ M + and {bα | α < κ} ⊂ ∗ + R be increasing coﬁnal κ-sequences. There exist κ-saturated models such that: (i) cof θ∗ = κ1 and cof θ = κ2 . Simply pick a positive inﬁnite x in Uη+1 preceding all those in Uη and an inﬁnite y ∈ ∗ RU1 preceding all those in G. In order to adjust the coﬁnality of θ. It is enough to show that there is an order-isomorphism between the set of positive inﬁnitesimals M + and ∗ R+ . δ2 . (ξ ω and respectively η ω if κ1 = ω). κ2 }. we preserve the initial part of θ and get cof θ = ω. ζϕ(β) ) β<δ . D a ﬁxed ζ + -good ultraﬁlter over ζ. ζ). We are left with κ = ω. δ. Giving up with saturation.ON THE ORDERING OF THE NONSTANDARD REAL LINE 11 Theorem 3. We do not need to care about saturation. Condition (ii) can be met by taking G and δ exactly as in the successor case. then the models can be chosen so as to satisfy (ii) |θx | = ξ <κ1 and |θy | = η <κ1 for suitable x. D) is κ-saturated and satisﬁes conditions (i) by Lemma 2. Lemma 3. y. The goal of the constructions above was precisely that of getting the maximal saturation with prescribed coﬁnalities.6. Remark that κ-saturation is best possible in the above theorem. we deﬁne a family {fα | α < κ} of internal functions such that for every α: . We slightly modify the preceding construction. and put κ = min{κ1 . if ξ. we take an iteration of ω internal ultrapowers V = V(G. Then U satisﬁes the conditions required by Lemma 2. Then U = U(G. Let δ be a limit ordinal with cof δ = κ1 and let ϕ : δ → κ be a function such that ϕ−1 (α) is coﬁnal in δ for all α < κ. ω. Assume ﬁrst that κ = ζ + is a successor cardinal. hence it is κ saturated and satisﬁes (i). Then we can take ζ = ω and proceed exactly as in the successor case. If the model is κ-saturated.e. Then |θx | = sup{|α|ζ | η + α < δ} = ξ ζ and |θy | = sup{|α|ζ | α < δ} = η ζ . In this way.6. Suppose ﬁrst that κ2 > κ1 = ω. Then we can start with a model G where cof θ∗ = κ1 and property (ii) is satisﬁed.

the models of (ii) Theorem 3. The union f = α<κ fβ (x) (x − aβ ) aα −aβ + bβ α β b −b if x ∈ (0.. Theorem 3. aβ ] if x ∈ (aβ . aα ] fα yields the desired order-isomorphism. Fβ1 ∩ . .4. ∩ Fβn = Fmax {β1 . Notice also that Fβ ⊆ Fβ for β < β < α. bα ] | f is an internal order-isomorphism extending fβ }. Proof. In the proof of the theorem we shall need the notion of type of a gap. the function ϕβ ϕβ (x) = belongs to Fβ . but ρ = µ.βn } .2 with κ1 = κ2 = κ satisfy the equality µ = ρ together with ∃x∃y(θx = θy ). bα ] is an order-isomorphism. For each β < α. proceed as follows. by Theorem 1. x < y for all x ∈ X and y ∈ Y ). Moreover. and the FIP holds. The next theorem will yield that also the second implication of Corollary 1. µ = ρ = γθ =⇒ θ = θ∗ . Clearly fα satisﬁes the conditions 1 and 2. Pick f0 an internal order-isomorphism between the intervals (0.6. Corollary 3. Y ) has type (κ. by κ-saturation we can pick fα ∈ β<α Fβ . • The partition (X.3. because fβ extends fβ by the inductive hypothesis.12 MAURO DI NASSO AND MARCO FORTI (1) fα : (0. Independently of the value of cof θ = cof θ∗ . Let κ be a regular cardinal and assume κ > 2ζ . let Fβ = {f : (0. since |{Fβ : β < α}| = |α| < κ. Y has no least element. a0 ] and (0. aα ] → (0.. Y ) be a partition of the ordered set (S. (2) fα extends fβ for all β < α.5 cannot be reversed. aα ] → (0. and X < Y (i. ζ) if cof X = κ and cof Y ∗ = ζ. Y ) is generated by A ⊂ S if X = {x ∈ S | x ≤ a for some a ∈ A}. By Lemma 3. b0 ] b (say f0 (x) = x · a0 ). Y ) is a gap of S if X has no greatest element. 0 At the inductive step α > 0. which contradicts θ = θ∗ . .e.. Then there exist ζ + -saturated nonstandard models where cof θ = cof θ∗ = κ. Let (X.5. Notice that all Fβ are internal and nonempty. • The gap (X.3. For instance. . • The gap (X. In particular.. <). Therefore we recall the following deﬁnition Deﬁnition 3.

We claim that the model U = U(G. by Lemmata 2.2 of [17] claims that cof θ = cof θ∗ is equivalent to ρ = µ. hence under heavy set-theoretic assumptions and with strong limitations on size and coﬁnality of θ. since all nonstandard models are ω-saturated. Point 1 follows from Lemma 3. X ) of type (ζ + . contradicting Point 2 of our Corollary 3.7. (1) If cof θ = cof θ∗ = ω. the ζ + -sequence {xα | α < ζ + } generates a gap of the same type in the iterated ultrapower U as well. The consistency of θ = θ∗ has been ﬁrst proved.4 and 2. with ξ > 2ζ . Point 2 is an immediate consequence of the theorem above for ζ = ω. D) satisﬁes the thesis.) Let D be a countably incomplete ζ + -good ultraﬁlter over ζ. Potthoﬀ in [23]. by order-isomorphism. (Such a model exists for κ > 2ζ . assume that there is an order-isomorphism Ψ : ∗ R → M between the nonstandard real line and the monad of zero in U. Let G be a (2ζ )+ -saturated model where cof θ = κ. This is a contradiction because. where construction techniques substantially equivalent to those of Section 2 are developed. in [17] and [29]. the model U is ζ + -saturated. On the other hand.6. We are left to show that ρ = µ in U. ∗ . for every α < κ. ζ). The existence of models where coﬁnality and coinitiality of the inﬁnites can be arbitrarily ﬁxed has been ﬁrst proved by K. both gaps must have the same type. where β = sup{βα | α < ζ + } < κ. Therefore we can pick increasing ζ + -sequences {xα | α < ζ + } in ∗ R0 and {βα | α < ζ + } in κ such that Ψ(xα+1 ) = yα+1 ∈ Mβα+1 is greater than all elements of Mβα In R0 the ζ + -sequence {xα | α < ζ + } generates a gap (X. η) with ζ < η ≤ 2ζ in ∗ Rβ+1 . then µ = ρ = γθ. 4. then cof θ = cof θ∗ =⇒ µ = ρ. In the following we denote by ∗ Rα and Mα the hyperreal line and the monad in Uα . by the argument above. We recall here only the main achievements which are known to us. ξ). Since ζ < ζ + < ξ. the ζ + -sequence {yα | α < ζ + } is coﬁnal in Mβ . 4The argument of this proof has been suggested to the ﬁrst author by Renling Jin in a personal communication. (2) If cof θ > 2ω . But the purported inductive proof works only for successor steps. By Lemma 2. have been studied by several authors since the problem has been explicitly posed in the paper of E. Let D be the κ-sequence constantly equal to (D. It is worth mentioning that Lemma 3. only for fully saturated models.6. By contradiction. Mα is not coﬁnal in Mα+1 . κ. Final remarks The consistency and the mutual relations of the order-theoretic properties (A1)− (A5) and (B1) − (B5) of Section 1. and satisﬁes cof θ∗ = cof θ = κ.3. Now ∗ R0 is coﬁnal in ∗ R and.4 We are now ready to deal with the converse of the second implication of Corollary 1.7. respectively. So it generates a gap of type (ζ + . The latter type remains unchanged up to ∗ R.5 Corollary 3.ON THE ORDERING OF THE NONSTANDARD REAL LINE 13 Proof. Proof. by (2ζ )+ -saturation. Zakon [30]. ω) on ∗ Rβ and cosequently one of type (ζ + .

[27]. which are derived there from this false lemma. where the Isomorphism Property was introduced. .[13]. We think that the main questions concerning the above order-theoretic properties which remain open are the following:5 (1) Are there models where θ = θ∗ = ρ? (2) Are there models where ω < cof θ = cof θ∗ ≤ 2ω and µ = ρ? (3) Are there models where ∀x∀y(θx = θy ) but θ = θ∗ ? We conjecture that the answers to these three questions are all in the positive. the Resplendency Property of Jin and Shelah [15].1 and 3. Among them. a new interesting approach has been developed. 3. 3]. ω. respectively. In fact the ultrapowers ω ω2 D and ω1 ω2 have sizes ℵ1 and ℵ2 . κ. Several interesting results in this direction can be found in the papers cited above.) In this paper. among others.[14]. D).1 and E is constantly equal to some nonprincipal ultraﬁlter over ω. D θ = θ∗ holds because θ is a ℵ1 -saturated order-type of size ℵ1 . See e. For instance. The ﬁrst general constructions allowing for symmetric inﬁnites with arbitrary coﬁnalities can be found in [9]. namely that of isolating “special” principles which hold in suitable classes of models. in the corresponding nonstandard model.3.[9]. where U(G.2 of the present paper are stated without proof). (B1) − (B5) are proved (see also [7]. all models have been explicitly deﬁned by means of the ultrapower construction. where Theorems 1. Therefore. After the pioneering article of Henson [12]. also in [16]. and we think that they could be successfully applied also to the topics treated in this paper. 5Notice that in this paper we are considering only results that are provable in ZFC. The relative consistency of θ = θ ∗ = ρ with ZFC+“there exists a huge cardinal” is a consequence of the consistency of Foreman’s nonregular ultraﬁlter D over ω2 . E). This topic is strictly connected to properties of the Q-topology and of other natural non-Hausdorﬀ topologies of the hyperreal line. The idea of using iterations of ultrapowers and internal ultrapowers has been used extensively in the literature. κ. we single out the the Special Model Axiom of Ross [26]. (We are considering these questions in a paper under construction. D) is the model in the proof of Theorem 3.[21]. and. the excellent paper [21] where these techniques are exploited in order to solve the diﬃcult problem of ﬁnding suﬃciently saturated models where the hyperreal line is Scott complete. where also several implications among the properties (A1) − (A5). These “special” properties have been showed fruitful to show the ac consistency of several properties in nonstandard analysis. see [6]. Another interesting topic. is the study of the types of the gaps that can occur in the hyperreal line and their connections with the order-type θ. should satisfy ∀x∀y(θx = θy ) together with cof θ ∗ = κ and cof θ = ω. stemming from [30] and not considered here. Notice that this fault aﬀects also the proofs of some true equivalences among the properties (B1) − (B5). On the other hand θ = ρ since |ρ| ≥ ℵ2 .g. it seems that the model V(U(G.14 MAURO DI NASSO AND MARCO FORTI and so it proves in fact only Point 1 of our corollary. and the recent combinatorial ∆-principles of the ﬁrst author and Hrb`˘ek [2.

J. Proc. Combinatorial principles in nonstandard analysis. 35–52. [15] R. I. Logik und Grund. [5] R. Logic 52 (1987). On the order structure of the nonstandard real axis. 534–539. Soc. 292–301. Kanovei. Keisler. Isr. in Victoria Symposium on Nonstandard Analysis (A. Math. Weber and Schmidt. North Holland. U-monad topologies of hyperﬁnite time lines. The generic ﬁlter property in nonstandard analysis.). der Math. 1165–1198. [14] R. References [1] C. 23–37. 522–527. J. 85 (1994). Logic 111 (2001). Logic. 403–426. Keisler. and W.W. 45 (1997). Logic 59 (1994). External set properties in IST (abstract). 29 (1983). in Atti degli incontri di Logica Matematica. Logic Colloquium 2001. [16] S. Robinson.C. Berlin 1974. Keisler and J. Logic 56 (1991). 1016–1025. 1–8. [10] M. to appear. of Math. of Symb. by simple combinatorial arguments. Pure Appl. Amsterdam 1966. Ann. it is shown in [3] that ∆2 proves θ = θ∗ . Ross. Amsterdam. On the order structure of the nonstandard real axis (abstract). Di Nasso. Logic 55 (1990). The strength of the isomorphism property. Ann. 21–24. [2] M. Prokhorova.J. On the order structure of the hyperreal line. Non-standard real number systems with regular gaps. Kamo. Nonstandard topologies with bases that consist only of standard sets. Ann. such as ultrapowers. ac [3] M. Berlin 1972. Warszawa 1977. Math. Nat.J. Forti. Logic 46 (1981). of Symb.ON THE ORDERING OF THE NONSTANDARD REAL LINE 15 In particular. Vive la diﬀ´rence II.). vol. Volume of abstracts. 717–731. in Theory of Sets and Topology (in honour of Felix Hausdorﬀ ) (G. fur Math. [22] E. Boston. Loeb. Chang and H. eds. of Symb. Zanardo. Keisler. Engelking. [9] M. M. Forti and F. the ∆-principles have the advantage of being formulated in purely combinatorial terms. Nonstandard natural number systems. [6] M. Math. Cuts in hyperﬁnite time lines. Fis. 113–115. S. [25] A. Soc. Monotone complete ﬁelds. The Ax-Kochen isomorphism theorem. 365–376. J. Asser. [4] M. Mat. The isomorphism property in nonstandard analysis and its use in the theory of Banach space. Mathematics in a nonstandard world. Hrb`˘ek. Forti and F. J. Ferro and A. Tsukuba J. [20] H.J. Jap. Model Theory (3rd edition). [28] N. Padova 1987. Zeitschr. Honsell. Cl. 199-205. [13] R. An ℵ1 -dense ideal on ℵ2 .3 (R. of Pure and Appl. The special model axiom in nonstandard analysis. J. [26] D. 129 (2001). [11] W. Flaschmeyer. . eds. 108 (1998). Logic 57 (1992). a new approach to nonstandard analysis. Honsell. Zhang. Henson. Nelson.H. Naz. J. Lincei Rend. Di Nasso and K. Nonstandard analysis and an application to the symmetric group on natural numbers. J. Vakil. 197–202. General Topology. 8 74 (1985). [21] H. Hatcher and C. J. [24] M. J. Internal Set Theory. Math. of Symb. Logic 57 (1992). Rinow.S. Bull. North-Holland. 83 (1977). [17] S. [23] K. 351–390. Ordnungseigenschaften von nichstandardmodellen. Schmerl. Di Nasso and Y. [12] C. P. Foreman. Kamo. Atti Acc. Logic 77 (1996). Reeken. [8] M. Potthoﬀ. Jin. and they can be directly applied without any use of modeltheoretic constructions. Amer. 5 (1981). 369–408. 253–290. For instance.J. Logic 39 (1974).). Sull’ordinamento dei numeri reali non-standard (italian). Amer. [19] H. Forti. of Symb. 1976. Sci. e [27] S. Vienna 2001. 1233–1242. Choice Principles in Hyperuniverses. Shelah. 2075–2083. Prindle. Polish Scientiﬁc Publishers. Israel J. of Symb. of Symb. Hurd. Math. Non-standard Analysis. [7] M. Jin. of Pure and Appl. of Symb. Making the hyperreal line both saturated and complete. Laﬂamme. 1990. Foundations of Inﬁnitesimal Calculus. Shelah. eds. J. [18] V. this volume. Jin and S. 317–318.

E-mail address: dinasso@dma. E-mail address: forti@dma. Dini”.J. Sin. [30] E. Analysis and Probability (W.A. New York 1969. Dini”.16 MAURO DI NASSO AND MARCO FORTI [29] S. Italy. Universita di Pisa. Acta Math. 658–667. Luxemburg.it . 23 (1980). Wang and J. ed.unipi.it ` Dipartimento di Matematica Applicata “U. Wang.). Italy. D. in Applications of Model Theory to Algebra. Yang. 195– 227. ` Dipartimento di Matematica Applicata “U.unipi. On the order structure of the nonstandard real axis. On the order structure of the ω-regular extension hyperreal numbers (chinese). Universita di Pisa. Zakon.

- Mathematical Analysis II
- Algebra
- L2b
- 08hw3
- Hasse Minkowski - Dicker
- Glossary of Set Theory 2
- Louis H. Kauffman- Math 215 Assignments
- Pinch - Social Studies of Science
- MATH1505_TT3_2013S.pdf
- Sequences and Series–First View
- Picture Books an Automated Story Generator
- On Smarandache Rings II
- Problems on Group Theory
- Types of Functions_3
- Mathematics Notes and Formula for Class 12 Chapter 1. Relations and Functions
- pmf
- Cumulative Distribution Function - Wikipedia, The Free Encyclopedia
- Vakil Note on Algebraic Geometry is a very authentic source of academic study materials in the corresponding field.
- 3 Inconsistent Countable Set
- Research Inventy
- 304
- Engleza Romana
- ch16-NonparametricStatistics
- Art of the Motor Paul Virilio
- 1
- Number Theory
- lectures.pdf
- Seminar 1 - Labor Demand and Supply - Search Theory
- Metrics Eng
- Math 104 Final Notes

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd