15.08.2013 Views

Ω-Theory: Mathematics with Infinite and Infinitesimal Numbers - SELP

Ω-Theory: Mathematics with Infinite and Infinitesimal Numbers - SELP

Ω-Theory: Mathematics with Infinite and Infinitesimal Numbers - SELP

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

UNIVERSITÀ DEGLI STUDI DI PAVIA<br />

FACOLTÀ DI SCIENZE MATEMATICHE, FISICHE E NATURALI<br />

CORSO DI LAUREA IN MATEMATICA<br />

Ω-<strong>Theory</strong>:<br />

<strong>Mathematics</strong> <strong>with</strong> <strong>Infinite</strong> <strong>and</strong><br />

<strong>Infinite</strong>simal <strong>Numbers</strong><br />

Relatore<br />

Chiar.mo Prof. Vieri Benci<br />

Correlatore<br />

Chiar.mo Prof. Mauro Di Nasso<br />

Correlatore<br />

Chiar.mo Prof. Enrico Vitali<br />

Anno Accademico 2010/2011<br />

Tesi di laurea di<br />

Emanuele Bottazzi


To Eleonora


Contents<br />

1 Ω-Calculus 1<br />

1.1 <strong>Infinite</strong>simal <strong>and</strong> infinite numbers . . . . . . . . . . . . . . . . . . . . 2<br />

1.2 Superreal Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4<br />

1.3 The Axioms of Ω-Calculus . . . . . . . . . . . . . . . . . . . . . . . . 7<br />

1.4 First properties of Ω-limit . . . . . . . . . . . . . . . . . . . . . . . . 9<br />

1.5 The Transfer Principle . . . . . . . . . . . . . . . . . . . . . . . . . . 14<br />

1.6 Numerosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />

1.7 Extensions of sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />

1.8 Extensions of functions . . . . . . . . . . . . . . . . . . . . . . . . . . 24<br />

1.9 Towards a model for Ω-Calculus . . . . . . . . . . . . . . . . . . . . . 27<br />

1.10 Filters <strong>and</strong> ultrafilters . . . . . . . . . . . . . . . . . . . . . . . . . . 31<br />

1.11 A model for Ω-Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . 36<br />

1.12 Reasoning in the model . . . . . . . . . . . . . . . . . . . . . . . . . . 39<br />

2 Selected topics of Ω-Calculus 43<br />

2.1 Convergence <strong>and</strong> continuity . . . . . . . . . . . . . . . . . . . . . . . 44<br />

2.2 Topology by Ω-Calculus . . . . . . . . . . . . . . . . . . . . . . . . . 49<br />

3 Integrals by Ω-Calculus 54<br />

3.1 Hyperfinite sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />

3.2 Ω-integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56<br />

3.3 Ω-integral <strong>and</strong> Riemann integral . . . . . . . . . . . . . . . . . . . . . 59<br />

3.4 Numerosity <strong>and</strong> Lebesgue measure . . . . . . . . . . . . . . . . . . . 61<br />

4 Ω-<strong>Theory</strong> 65<br />

4.1 Foundational framework . . . . . . . . . . . . . . . . . . . . . . . . . 65<br />

4.2 The Axioms of Ω-<strong>Theory</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 66<br />

4.3 Generalizing concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . 68<br />

4.4 Transfer Principle by Ω-<strong>Theory</strong> . . . . . . . . . . . . . . . . . . . . . 70<br />

4.5 A model for Ω-<strong>Theory</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . 75<br />

5 The formal version of Transfer Principle 79<br />

5.1 Logic formalism for set theory . . . . . . . . . . . . . . . . . . . . . . 79<br />

5.2 Transfer Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84<br />

v


6 An application: Non-Archimedean Probability 87<br />

6.1 Kolmogorov’s axioms of probability . . . . . . . . . . . . . . . . . . . 87<br />

6.2 Non-Archimedean probability . . . . . . . . . . . . . . . . . . . . . . 89<br />

6.3 First consequences of NAP axioms . . . . . . . . . . . . . . . . . . . 91<br />

6.4 Fair lotteries by NAP . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />

6.5 <strong>Infinite</strong> sequence of coin tosses by NAP . . . . . . . . . . . . . . . . . 97<br />

7 Further ideas 99<br />

7.1 The Cartesian Product Axiom for Numerosity . . . . . . . . . . . . . 99<br />

7.2 Dimension by Ω-<strong>Theory</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 101<br />

7.3 Numerosity <strong>and</strong> Lebesgue measure . . . . . . . . . . . . . . . . . . . 103<br />

7.4 Functional Analysis by Ω-<strong>Theory</strong> . . . . . . . . . . . . . . . . . . . . 104<br />

7.5 R ∗∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104<br />

7.6 Ω-integral of hyperreal functions . . . . . . . . . . . . . . . . . . . . . 106<br />

7.7 The Dirac distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 107<br />

7.8 The natural extension of numerosity . . . . . . . . . . . . . . . . . . 108<br />

vi


Introduction<br />

Borknagar<br />

La mia vita in quest’ultimo anno è<br />

stata un susseguirsi di peripezie.<br />

[...] Ma sono contento di tutto<br />

quello che ho fatto, del capitale di<br />

esperienze che ho accumulato, anzi<br />

avrei voluto fare di più.<br />

Italo Calvino<br />

We believe that there are many ways to do mathematics, <strong>and</strong> this belief is<br />

motivated by a number of examples. One of the simplest but nonetheless<br />

significative is given by the field of complex numbers: historically, it has<br />

been presented as the splitting field of the polynomials <strong>with</strong> real coefficients, but in<br />

the language of modern algebra this field is defined as the quotient of the polynomial<br />

ring R[x] by the principal ideal generated by x2 + 1. Of course, it can be proved<br />

that the two definitions are equivalent, but each of them has its own role, both in<br />

the history <strong>and</strong> in the practice of mathematics.<br />

We find also very interesting that a theory that has been given many different<br />

formalizations during the centuries is the Euclidean geometry. In particular,<br />

during the XX century, many mathematicians have given their own axiomatics for<br />

Euclidean geometry: here, we want to focus on Hilbert’s presentation introduced<br />

in the Grundlagen der Geometrie <strong>and</strong> on Prodi <strong>and</strong> Choquet formulations. The<br />

intent of Hilbert’s axiomatic is clear: to give a correct <strong>and</strong> elegant foundation of Euclidean<br />

geometry <strong>with</strong> the precision of “modern” mathematics. In particular, one of<br />

the characteristics of this approach is that, for the working mathematician, nothing<br />

more than the axioms of the theory is needed to develop all Euclidean geometry.<br />

On the other h<strong>and</strong>, both Prodi <strong>and</strong> Choquet aim to give an axiomatic system for<br />

Euclidean geometry simple enough to be taught to high school students. Hence,<br />

their axiomatics are less tecnical than Hilbert’s one, <strong>and</strong> rely heavily on the structure<br />

of the real numbers. In this case, it’s clear that some previous mathematical<br />

knowledge of the real numbers is necessary for the development of the theory.<br />

vii


INTRODUCTION<br />

It’s a well known fact that even calculus, whose classical approach based on<br />

the “ε-δ” definition of limit has been formulated in the XIX century, can be given<br />

a different formalization based on the concepts of infinite <strong>and</strong> infinitesimal numbers.<br />

The resulting theory, usually called Nonst<strong>and</strong>ard Analysis, has proved to give<br />

means to develop calculus in a way closer to the mathematical intuition than the<br />

“st<strong>and</strong>ard” one, but <strong>with</strong>out sacrificing precision or mathematical rigor. While we<br />

do agree that intuition shouldn’t be used as a valid inference tool of mathematics,<br />

we are equally convinced that many nonst<strong>and</strong>ard definitions linked to the topics of<br />

convergence <strong>and</strong> continuity could help math students (<strong>and</strong> even some teachers) to<br />

better underst<strong>and</strong> the underlying notions <strong>with</strong>out hiding them under a layer of cryptic<br />

formalism. In this regard, it’s also been argued that the “quantifier complexity”<br />

of some nonst<strong>and</strong>ard definitions is indeed reduced <strong>with</strong> respect to st<strong>and</strong>ard ones.<br />

The main weakpoint of Nonst<strong>and</strong>ard Analysis is that the formalism of Robinson’s<br />

original presentation appeared too technical to many, <strong>and</strong> not directly usable by<br />

those mathematicians <strong>with</strong>out a good background in logic. Over the last forty<br />

years, many attempts have been made in order to simplify the foundational matters<br />

<strong>and</strong> popularize nonst<strong>and</strong>ard analysis by means of “easy to grasp” presentations.<br />

Following those attempts, the mathematical theories of Ω-Calculus <strong>and</strong> Ω-<strong>Theory</strong><br />

presented in this paper are two simplified version of Nonst<strong>and</strong>ard Analysis expressed<br />

in the everyday language of mathematics, in a way to still allow for a complete <strong>and</strong><br />

rigorous treatment of all the basics of calculus. An important <strong>and</strong> fascinating feature<br />

of those theories lies in the fact that, in contrast <strong>with</strong> the original formulation of<br />

Nonst<strong>and</strong>ard Analysis, their presentation only assumes a few basic notions from<br />

algebra, so that they need no more prerequisites than the usual Calculus does. For<br />

those reasons, we do believe that Ω-Calculus <strong>and</strong> its extension given by Ω-<strong>Theory</strong><br />

are two beautiful <strong>and</strong> relatively simple theories which could potentially bring many<br />

mathematicians, especially those who have no background in logic, closer to the<br />

methods of Nonst<strong>and</strong>ard Analysis.<br />

Since, in analogy to what happens for Nonst<strong>and</strong>ard Analysis, Ω-Calculus <strong>and</strong> Ω-<br />

<strong>Theory</strong> are both grounded on the use of infinite <strong>and</strong> infinitesimal numbers, the first<br />

chapter of this paper is dedicated to the introduction of the basic definitions of those<br />

concepts. After that, we will introduce the axiomatical theory of Ω-Calculus <strong>and</strong> the<br />

fundamental operation of “Ω-limit”, <strong>and</strong> we will study in detail their properties <strong>and</strong><br />

some of the more significative applications in extending real functions to particular<br />

hyperreal sets. Once the reader is acquainted <strong>with</strong> those results, there shouldn’t be<br />

more difficulties in following the rest of the paper. It’s interesting to remark that, in<br />

order to give the axioms of Ω-Calculus <strong>and</strong> to settle their most useful consequences,<br />

only a little knowledge of algebra is required.<br />

The last three sections of the first chapter are dedicated to the explicit construction<br />

of an algebraic model for Ω-Calculus: the main goal is to show that the<br />

axioms of Ω-Calculus are consistent but, in the course of this paper, we will see that<br />

the algebraic models of Ω-Calculus can be also used to give a nice interpretation<br />

to the operation of Ω-limit <strong>and</strong> are very useful in view of some applications, most<br />

notably to probability. We signal that some notions about rings, ideals <strong>and</strong> fields are<br />

necessary to fully underst<strong>and</strong> the model building process, but the necessary ideas<br />

don’t exceed what is usually taught during an introductive course in algebra. Some<br />

viii


INTRODUCTION<br />

notions of set theory are also required, but we will give a brief presentation of the<br />

main concepts <strong>and</strong> results we will use.<br />

Chapters 2 <strong>and</strong> 3 are dedicated to show the Ω-Calculus “in action”. During the<br />

first part of Chapter 2, we will show the genesis of the concept of convergence for<br />

real sequences <strong>and</strong> for real functions; in the second part, we will focus on how the<br />

usual topology of the real line can be interpreted by using Ω-Calculus, <strong>and</strong> how<br />

this new interpretation seems to simplify the proofs of some famous theorems, the<br />

most important of them being the Banach-Caccioppoli Fixed Point Theorem. In the<br />

course of Chapter 3, we will introduce an interesting notion of Ω-integral that makes<br />

sense for all real functions, <strong>and</strong> then we will show its relations between Riemann<br />

<strong>and</strong> Lebesgue integral.<br />

The theory introduced so far will be generalized in Chapter 4, where the concepts<br />

introduced in Chapter 1 will be extended to functions <strong>with</strong> arbitrary range. The<br />

resulting theory, that will be called Ω-<strong>Theory</strong>, will be interesting on its own, but<br />

can also be applied in a number of situations. One of the possible applications<br />

of Ω-<strong>Theory</strong> will be given in Chapter 6, where we will present a new approach to<br />

probability that overcomes some of the weakpoints of the classical Kolmogorovian<br />

approach, namely the fact that the union of elementary events is not always an event<br />

<strong>and</strong> the vague interpretation of probability measures for which there exist nonempty<br />

events <strong>with</strong> zero probability. This chapter will be rich of examples, in order to focus<br />

on some “tangible” results <strong>and</strong> not only on the abstract theory.<br />

The possible applications of Ω-<strong>Theory</strong> to everyday mathematics seems to be very<br />

broad, <strong>and</strong> surely exceed the scope of this paper. In the conclusive chapter, we have<br />

selected a few of the most interesting ideas that couldn’t be properly developed <strong>and</strong>,<br />

for each of them, we have given a brief presentation. We hope that, one day, many<br />

of those open questions will be given an answer.<br />

Chapter 5 is planned as a “logic intermission”, where we will show that the Ω-<br />

<strong>Theory</strong> proves a very important logic theorem, namely the Transfer Principle. This is<br />

the only chapter where a little logic knowledge is required to fully master the results.<br />

We remark that the main instances of Transfer that will be used in this paper can<br />

be directly shown <strong>with</strong>out any logic knowledge, so our claim that Ω-<strong>Theory</strong> can be<br />

used as a more advanced theory to introduce those mathematicians who have little<br />

or no background in logic close to the methods of Nonst<strong>and</strong>ard Analysis still holds.<br />

A few words about notation<br />

We believe that uniformity in<br />

graphical terminology will never be<br />

attained, <strong>and</strong> is not necessarely<br />

desirable.<br />

Frank Harary<br />

If A <strong>and</strong> B are two sets, we will use the symbol A ⊆ B when A is a subset of B<br />

<strong>and</strong> we admit the possibility that A = B, or we don’t know yet if A = B is true or<br />

ix


INTRODUCTION<br />

not. We will write A ⊂ B if A is a subset of B <strong>and</strong> we don’t admit the possibility<br />

that A = B, or we already know that A = B.<br />

If A <strong>and</strong> B are sets, we will denote by A × B the cartesian product of A <strong>and</strong> B.<br />

For a matter of commodity, if A is a set <strong>and</strong> n ∈ N, in the last chapter of this paper<br />

we will sometimes denote by An the cartesian product A<br />

<br />

× .<br />

<br />

. . × A<br />

<br />

.<br />

n times<br />

We will denote by | · | the cardinality of a set.<br />

If A is a set, we will denote by P(A) its powerset. Moreover, we define once <strong>and</strong><br />

for all the set Pfin(A) = {λ ⊆ A : 0 < |λ| < +∞}. Clearly, Pfin(A) ⊂ P(A), <strong>and</strong><br />

the two differ only for the empty set if <strong>and</strong> only if |A| is finite.<br />

If A ⊆ Ω, by Ac we will denote the set Ω \ A. We will use the same notation<br />

even in other cases, but it will always be clear what is the set B that has the same<br />

role Ω had in the previous case.<br />

For all sets A <strong>and</strong> B, for all b ∈ B we will denote (<strong>with</strong> abuse of notation) by cb<br />

the function cb : A → B defined by cb(a) = b for all a ∈ A. We will denote by χA<br />

the function χA : A → {0, 1} defined by<br />

<br />

1 if x ∈ A<br />

χA(x) =<br />

0 if x ∈ A<br />

<strong>and</strong> we will call it the characteristic function of A.<br />

We will denote by N, Z, Q <strong>and</strong> R the sets of, respectively, natural numbers,<br />

integer numbers, rational numbers <strong>and</strong> real numbers. For a matter of commodity,<br />

we will agree that N = {1, 2, 3, . . .}, so that 0 ∈ N.<br />

If F is an ordered field <strong>and</strong> a, b ∈ F, we will write [a, b)F = {x ∈ F : a ≤ x < b}.<br />

We will use the notations [a, b]F, (a, b)F <strong>and</strong> (a, b]F accordingly. When there will be<br />

littke risk of misunderst<strong>and</strong>ing, we will omit the indication of the field F. If x ∈ F,<br />

we will denote by |x| its absolute value:<br />

<br />

x if x ≥ 0<br />

|x| =<br />

−x if x < 0<br />

x


Chapter 1<br />

Ω-Calculus<br />

<strong>Mathematics</strong> is not a careful march<br />

down a well-cleared highway, but a<br />

journey into a strange wilderness,<br />

where the explorers often get lost.<br />

W. S. Anglin<br />

The intuitive notions of “infinitely small” <strong>and</strong> “infinitely large” quantities<br />

are central in the development of differential calculus. The current “ε-δ<br />

approach”, as elaborated by Weierstrass in the second half of the XIX century,<br />

incorporates those notions in an indirect manner that is grounded on the notion<br />

of limit. This, however, is not the only way to give a mathematical formalization<br />

of those ideas, as most notably shown by Abraham Robinson <strong>with</strong> the introduction<br />

of Nonst<strong>and</strong>ard Analysis. Once given a mathematical theory for infinitesimal <strong>and</strong><br />

infinite numbers, it’s only natural to wonder if calculus can be grounded on those<br />

notions, instead of on the use of limits. It’s a well known fact that this is indeed the<br />

case, <strong>and</strong> the Ω-Calculus that will be presented in this chapter is a new elementary<br />

method for developing calculus in such a way.<br />

In the first two sections of this chapter, we will formally define what infinitesimal<br />

<strong>and</strong> infinite numbers are <strong>and</strong> how they behave under the basic algebraics operations.<br />

In those sections, we will give some fundamental definitions <strong>and</strong> results that will be<br />

used through all the paper, <strong>and</strong> most notably we will introduce the concept of<br />

superreal fields <strong>and</strong> of st<strong>and</strong>ard part of a superreal number.<br />

In Section 1.3, we will give the axioms of Ω-Calculus, that will rule the existence<br />

<strong>and</strong> the behaviour of a new kind of “limit”, that we will call “Ω-limit”. This “Ωlimit”<br />

will allow a construction of a particular superreal field R∗ that, as we will see<br />

Chapters 2 <strong>and</strong> 3, will be the natural setting where basic calculus can be developed.<br />

The central part of this chapter is dedicated to show the main properties of<br />

the Ω-limit <strong>and</strong> how to use it to introduce some interesting mathematical concepts,<br />

namely numerosity <strong>and</strong> the natural <strong>and</strong> hyperfinite extension of sets <strong>and</strong> functions.<br />

The most notable characteristic of those extension is the fact that they preserve all<br />

“elementary properties” of the mathematical entities involved, in a way that will be<br />

precised in Section 1.5.<br />

1


CHAPTER 1. Ω-CALCULUS<br />

Finally, in Sections 1.9, 1.10 <strong>and</strong> 1.11, we will address the problem of the consistency<br />

of the axioms of Ω-Calculus, <strong>and</strong> we will show that they admit an algebraic<br />

model.<br />

In the last section of this chapter, we will show some examples of how the model<br />

can be used to infer some properties of the theory that are not ruled out by the<br />

axioms.<br />

1.1 <strong>Infinite</strong>simal <strong>and</strong> infinite numbers<br />

Infinity is not a big number.<br />

John D. Barrow<br />

In our approach to calculus, we are interested in number fields containing infinitesimal<br />

<strong>and</strong> infinite numbers. Before we define them in a rigorous way, we need<br />

to recall some basic notions about ordered fields.<br />

Definition 1.1.1. An ordered field is a couple (F, F + ) where F is a field, <strong>and</strong> F + ⊂ F<br />

satisfies:<br />

• for all x, y ∈ F + , then x + y <strong>and</strong> xy ∈ F;<br />

• F = −F + ∪ {0} ∪ F + , where −F + = {−x : x ∈ F + }, <strong>and</strong> the union is disjoint.<br />

Then, we can define an order on F by setting<br />

x < y ⇐⇒ y − x ∈ F +<br />

We say that the elements of F + are positive, <strong>and</strong> the elements of −F + are negative.<br />

Following the common practice, we identify each positive n ∈ N <strong>with</strong> the corresponding<br />

iterated sum of the neutral element 1 ∈ F:<br />

n = 1 + . . . + 1<br />

<br />

n times<br />

Moreover, by considering opposites <strong>and</strong> reciprocals, we can suppose Q ⊆ F.<br />

Now, we want to introduce the concept of infinitesimal numbers in a formal way.<br />

The basic idea of infinitesimal numbers is that they are numbers whose absolute value<br />

is “smaller than every other (natural) number”. Other than that, we want to be<br />

able to perform field operations (sum, product, opposite <strong>and</strong> inverse) on infinitesimal<br />

numbers, so we’ll require that infinitesimal numbers are contained in a field F. Since<br />

the informal definition of infinitesimal numbers involves ordering, we require also<br />

that F is an ordered field. Putting everything together, we can rewrite the intuitive<br />

idea of infinitesimal number in the following way:<br />

Definition 1.1.2. Let F be an ordered field. A number ξ ∈ F is called infinitesimal<br />

if, for all n ∈ N, |ξ| < 1/n.<br />

2


1.1. INFINITESIMAL AND INFINITE NUMBERS<br />

Is this definition satisfied by some numbers? Let’s think about the field of rational<br />

numbers <strong>and</strong> the field of real numbers: in those fields, there exists exactly one<br />

infinitesimal number: the number 0.<br />

Up to now, we don’t know if there exist some number fields containing infinitesimal<br />

numbers different from 0 but, in section 1.11, we will give an explicit construction<br />

of one of such fields. This result allows us to study ordered fields F ⊃ R that contains<br />

nonzero infinitesimal numbers, knowing in advance that we’re not reasoning<br />

about nonexistant objects.<br />

The notion of infinitesimal numbers allows a classification of numbers according<br />

to their “size”.<br />

Definition 1.1.3. Let ξ ∈ F. We say that:<br />

• ξ is infinitesimal if for all n ∈ N, |ξ| < 1<br />

n ;<br />

• ξ is finite if there exists n ∈ N such as |ξ| < n;<br />

• ξ is infinite if, for all n ∈ N, |ξ| > n (equivalently, if ξ is not finite).<br />

It’s easily seen that the inverse of a nonzero infinitesimal number is infinite, <strong>and</strong><br />

the inverse of an infinite number is infinitesimal. Clearly, all infinitesimal numbers<br />

are finite. However, we remark that in any non-Archimedean field there are plenty of<br />

finite numbers that are neither infinitesimal nor rational. In fact, for every non-zero<br />

q ∈ Q <strong>and</strong> for every nonzero infinitesimal ξ, the number q + ξ ∈ Q is finite <strong>and</strong><br />

non-infinitesimal.<br />

In the next proposition, we itemize a list of simple algebraic properties that<br />

match the intuition of “small”, “finite”, <strong>and</strong> “infinite” quantities.<br />

Proposition 1.1.4. 1. If ξ <strong>and</strong> ζ are finite, then ξ + ζ <strong>and</strong> ξ · ζ are finite;<br />

2. If ξ <strong>and</strong> ζ are infinitesimal, then ξ + ζ <strong>and</strong> ξ · ζ are infinitesimal;<br />

3. If ξ is infinitesimal <strong>and</strong> ζ is finite, then ξ · ζ is infinitesimal;<br />

4. If ξ is infinite <strong>and</strong> ζ is not infinitesimal, then ξ · ζ is infinite;<br />

5. If ξ = 0 is infinitesimal <strong>and</strong> ζ is not infinitesimal, ζ/ξ is infinite;<br />

6. If ξ is infinite <strong>and</strong> ζ = 0 is finite, then ζ/ξ is infinitesimal.<br />

Proof. All the proofs are elementary, <strong>and</strong> therefore left to the interested reader as<br />

exercises. As an example, we will give the proof of part 5. We already know that, if<br />

ξ = 0 is infinitesimal, 1/ξ is infinite. By definition of infinite number, we have that,<br />

for all n ∈ N, |1/ξ| > n. Now, since ζ is not infinitesimal, there exists k ∈ N such<br />

that |ζ| > 1/k. We can conclude |ζ/ξ| > n/k for all n ∈ N <strong>and</strong> this implies that ζ/ξ<br />

is infinite.<br />

Note that a non-Archimedean field cannot be complete. For instance, the set of<br />

all infinitesimal numbers is upper bounded, but it has no least upper bound (if x is<br />

an upper bound for this set, then also x/2 < x is).<br />

The request that a field F contains infinitesimal number appears to be a fair<br />

simple one, but it’s deeply connected to some other basic properties of the field, as<br />

we’ll see in the next Proposition.<br />

3


CHAPTER 1. Ω-CALCULUS<br />

Proposition 1.1.5. An ordered field is Archimedean if <strong>and</strong> only if its only infinitesimal<br />

number is 0.<br />

Before we give the proof of this proposition, we recall that an ordered field is<br />

Archimedean if the following property holds:<br />

∀x > 0 ∀y > 0 ∃n ∈ N : ny > x<br />

or, in other words, the iterated sums of any positive element y surpass any given<br />

number x.<br />

Proof of Proposition 1.1.5. Let ξ be an infinitesimal number. Without loss of generality,<br />

we can suppose ξ > 0 (otherwise, choose −ξ). Since we have, by definition,<br />

ξ < 1/n for all n ∈ N, if we consider the inverses we have n < 1/ξ for all n ∈ N or,<br />

in other words, the field is not Archimedean.<br />

Now, suppose that the field is not Archimedean. Then, by definition, we have<br />

an element ξ that is a counterexample for the Archimedean property, that is n < ξ<br />

for all n ∈ N. Then 1/ξ is a nonzero infinitesimal number, since 0 < 1/ξ < 1/n for<br />

all n ∈ N.<br />

Proposition 1.1.5 gives a very interesting characterization of non-Archimedean<br />

fields, <strong>and</strong> shows that fields containing infinitesimal numbers don’t satisfy some<br />

common properties we are used to take for granted.<br />

1.2 Superreal Fields<br />

In the mathematical practice <strong>and</strong> especially in analysis, the most used extension of<br />

the real field is given by the complex numbers C. Instead, we want to concentrate<br />

on those extensions of R that are still ordered fields.<br />

Definition 1.2.1. A superreal field is an ordered field F that properly extends R<br />

(e.g. R ⊂ F, R + ⊂ F + , <strong>and</strong> the inclusions are strict).<br />

It turns out that any superreal field contains infinitesimal <strong>and</strong> infinite numbers,<br />

as shown in the next Theorem.<br />

Theorem 1.2.2. Any superreal field F is non-Archimedean.<br />

Proof. Thanks to Proposition 1.1.5, we only need to show that F contains an infinite<br />

or infinitesimal number. Choose ξ ∈ F\R. Without loss of generality, we can suppose<br />

ξ > 0. If ξ > n for all n ∈ N, there’s nothing to prove. Let’s suppose this is not the<br />

case, <strong>and</strong> 0 < ξ < n for some n ∈ N. Now, consider the set<br />

X = {x ∈ R : ξ < x}<br />

of the real upper bounds of ξ. By hypothesis, X is nonempty <strong>and</strong> bounded below.<br />

Using the completeness property of R, the element r = inf X is a well defined real<br />

number. We claim that ζ = ξ − r is an infinitesimal number.<br />

To prove our claim, let’s pick n ∈ N. By the properties of least upper bound, we<br />

have that ξ ≥ r − 1/n, so that ζ ≥ −1/n, <strong>and</strong> ξ < r + 1/n, so that ζ < 1/n. Since<br />

both inequalities hold for all n ∈ N, we conclude that ζ is a nonzero infinitesimal<br />

number.<br />

4


1.2. SUPERREAL FIELDS<br />

Thanks to infinitesimal numbers, in the superreal fields we can formalize a new<br />

notion of “closeness”.<br />

Definition 1.2.3. We say that two numbers ξ <strong>and</strong> ζ are infinitely close if ξ − ζ is<br />

infinitesimal. In this case, we will write ξ ∼ ζ.<br />

It’s easy to see that the relation ∼ if infinite closeness is an equivalence relation.<br />

By using this relation, we can give an idea of the order-structure of superreal<br />

fields. This, in turn, will allow a canonical representation of the finite numbers.<br />

Theorem 1.2.4. Every finite number ξ is infinitely close to a unique real number<br />

r ∼ ξ, called the shadow or the st<strong>and</strong>ard part of ξ. We will write r = sh(ξ).<br />

Proof. The proof is similar to that of Theorem 1.2.2: let’s consider the set of the<br />

real upper bounds of ξ<br />

X = {x ∈ R : ξ < x}<br />

<strong>and</strong> let r = inf X. We have seen in the proof of Theorem 1.2.2 that ξ − r is<br />

infinitesimal. This settles the existence of sh(ξ). To prove its uniqueness, let’s<br />

suppose that there exists r ′ ∈ R such as ξ ∼ r ∼ r ′ <strong>and</strong> r = r ′ . Since r <strong>and</strong> r ′ are<br />

real numbers, this implies r ∼ r ′ , but this contradicts our hypothesis.<br />

From this Theorem, we deduce immediately that every finite number can be<br />

written as sum of one real <strong>and</strong> one infinitesimal number in a unique way.<br />

Corollary 1.2.5. Any finite number ξ ∈ F has a unique representation in the following<br />

form:<br />

ξ = sh(ξ) + ɛ<br />

where sh(ξ) ∈ R <strong>and</strong> ɛ = ξ − sh(ξ) is infinitesimal.<br />

The notion of shadow can be extended to infinite numbers by setting:<br />

• sh(ξ) = +∞ if ξ is infinite <strong>and</strong> positive;<br />

• sh(ξ) = −∞ if ξ is infinite <strong>and</strong> negative.<br />

Accordingly, we will adopt the usual conventional algebra of the extended real<br />

line R ∪ {−∞, +∞}. As a result, we obtain that shadows satisfy compatibility<br />

properties <strong>with</strong> respect to sums, products <strong>and</strong> quotients.<br />

Proposition 1.2.6. For all numbers ξ, ζ, the following equalities hold:<br />

1. sh(ξ + ζ) = sh(ξ) + sh(ζ);<br />

2. sh(ξ · ζ) = sh(ξ) · sh(ζ);<br />

3. sh(1/ξ) = 1/sh(ξ) whenever ξ is not infinitesimal.<br />

The only exceptions are the following indeterminate forms:<br />

(+∞) + (−∞); ∞ · 0; ∞<br />

∞ ; 0<br />

0<br />

5


CHAPTER 1. Ω-CALCULUS<br />

Proof. To prove part 1, if both numbers are finite, Corollary 1.2.5 ensures that we<br />

can write<br />

ξ + ζ = (sh(ξ) + ɛξ) + (sh(ζ) + ɛζ) = (sh(ξ) + sh(ζ)) + (ɛξ + ɛζ)<br />

where both ɛξ <strong>and</strong> ɛζ are infinitesimal. Since sh(ξ) <strong>and</strong> sh(ζ) are real numbers <strong>and</strong>,<br />

by Proposition 1.1.4, ɛξ + ɛζ is infinitesimal, by Corollary 1.2.5 we can conclude<br />

sh(ξ + ζ) = sh(ξ) + sh(ζ)<br />

If ξ is infinite <strong>and</strong> ζ is finite or if both ξ <strong>and</strong> ζ are infinite <strong>and</strong> have the same sign,<br />

then the proposition is trivial.<br />

To prove part 2 <strong>and</strong> 3, we can proceed in a similar way.<br />

Similarly as <strong>with</strong> “classic” limits, in the indeterminate forms above, the resulting<br />

shadows could be any element l ∈ R ∪ {−∞, +∞}, <strong>with</strong> the only restriction of sign<br />

compatibility. In fact, for every given l, one can easily find ξi, ζi ∈ F such that<br />

• sh(ξ1) = +∞, sh(ζ1) = −∞ <strong>and</strong> sh(ξ1 + ζ1) = l;<br />

• sh(ξ2) = ∞, sh(ζ2) = 0 <strong>and</strong> sh(ξ2 · ζ2) = l;<br />

• sh(ξ3) = ∞, sh(ζ3) = ∞ <strong>and</strong> sh( ξ3 ) = l; ζ3<br />

• sh(ξ4) = 0, sh(ζ4) = 0 <strong>and</strong> sh( ξ4 ) = l. ζ4<br />

If we extend the order relation to R ∪ {−∞, +∞} in the obvious way, by setting<br />

−∞ ≤ l ≤ +∞ for all l ∈ R ∪ {−∞, +∞}, then shadows also satisfy a compatibility<br />

property <strong>with</strong> respect to the order. More precisely,<br />

Proposition 1.2.7. For all numbers ξ, ζ, sh(ξ) < sh(ζ) implies ξ < ζ.<br />

The easy proof of this Proposition is left as exercise for the interested reader.<br />

Remark 1.2.8. The above implication cannot be reversed: if ξ is finite <strong>and</strong> ɛ > 0 is<br />

an infinitesimal number, ξ < ξ+ɛ, but sh(ξ) = sh(ξ+ɛ). On the other h<strong>and</strong>, the weak<br />

inequality is preserved, <strong>and</strong> one can prove that ξ ≤ ζ if <strong>and</strong> only if sh(ξ) ≤ sh(ζ).<br />

Up to now, we only considered the equivalence relation ∼ given by the notion<br />

of being “infinitely close” (see Definition 1.2.3). Similarly, we can also consider the<br />

relation of “finite closeness”:<br />

ξ ∼f ζ if <strong>and</strong> only if ξ − ζ is finite.<br />

It is readily seen that also ∼f is an equivalence relation. In the literature, the<br />

equivalence classes relative to the two relations of closeness ∼ <strong>and</strong> ∼f, are called<br />

monads <strong>and</strong> galaxies, respectively.<br />

Definition 1.2.9. The monad of a number ξ is the set of all numbers that are<br />

infinitely close to it:<br />

mon(ξ) = {ζ ∈ F : ξ ∼ ζ}<br />

6<br />

The galaxy of a number ξ is the set of all numbers that are finitely close to it:<br />

gal(ξ) = {ζ ∈ F : ξ ∼f ζ}


1.3. THE AXIOMS OF Ω-CALCULUS<br />

In other words, mon(0) is the set of all infinitesimal numbers in F <strong>and</strong> gal(0) is<br />

the set of all finite numbers.<br />

In the following proposition, we itemize two of the basic properties of monads<br />

<strong>and</strong> galaxies.<br />

Proposition 1.2.10. 1. Any two monads (or galaxies) are either equal or disjoint;<br />

2. The monad (or the galaxy) of a point ξ is the coset of ξ modulo mon(0) (modulo<br />

gal(0), respectively). In other words, mon(ξ) = {ξ + ɛ : ɛ ∈ mon(0)} <strong>and</strong><br />

gal(ξ) = {ξ + ζ : ζ ∈ gal(0)}.<br />

Proof. Part 1 is true because monads <strong>and</strong> galaxies are equivalence classes, while part<br />

2 follows from the definition.<br />

1.3 The Axioms of Ω-Calculus<br />

Work out what the most beautiful<br />

rules are.<br />

Terry Pratchett<br />

Now that all the necessary concepts are formally defined, we can introduce the<br />

theory of Ω-Calculus. We choose to follow an axiomatic way: we start by giving<br />

all the axioms of Ω-Calculus <strong>and</strong> then, in the following sections, we will study<br />

thoroughly their properties <strong>and</strong> their consequences.<br />

Our axioms begin <strong>with</strong> the fundamental request of the existence of a superreal<br />

field R ∗ , whose existence is deeply connected to a new mathematical operation, that<br />

we will call “Ω-limit”. R ∗ will be the natural setting for the study of calculus <strong>with</strong><br />

infinite <strong>and</strong> infinitesimal numbers.<br />

Axiom 1 (Existence Axiom). Given a set Ω ⊃ R, there exists a (non-Archimedean)<br />

field<br />

R ∗ ⊃ R<br />

such that every function ϕ : Pfin(Ω) → R has a unique “ Ω-limit” in R ∗ . This limit<br />

is denoted by<br />

lim<br />

λ↑Ω ϕ(λ)<br />

Moreover, we assume that every ξ ∈ R ∗ is the Ω-limit of some real function ϕ :<br />

Pfin(Ω) → R.<br />

The word “limit” bears a big cultural <strong>and</strong> emotive baggage, having been used<br />

in mathematics for nearly two centuries as a fertile instrument in various fields of<br />

research. We have chosen to use the same word for the concept of Ω-limit because<br />

it encompasses the idea of evaluating the expression ϕ(λ) as the variable becomes<br />

“bigger <strong>and</strong> bigger” in Ω. The readers should be aware that this limit bears few<br />

similarities <strong>with</strong> the well known “ε-δ” limit used in calculus, <strong>and</strong> that even the basic<br />

7


CHAPTER 1. Ω-CALCULUS<br />

properties of Ω-limit must be deduced by the axioms of Ω-Calculus before they can<br />

be used. Sometimes, those properties will be in open contrast <strong>with</strong> the ones of<br />

classical limit, but this should not upset or discourage anyone: the two concepts are<br />

fairly different, despite the similarity of their name.<br />

Now, we proceed <strong>with</strong> the axioms of Ω-Calculus by asking that Ω-limit satisfies<br />

some basic properties: the Ω-limit of eventually constant functions must assume the<br />

expected value, <strong>and</strong> the field operations of R <strong>and</strong> R ∗ must be compatible.<br />

Axiom 2 (Real <strong>Numbers</strong> Axiom). If ϕ is eventually constant, namely ∃λ0 ∈ Pfin(Ω)<br />

such as ϕ(λ) = r for all λ ⊇ λ0, then<br />

lim ϕ(λ) = r<br />

λ↑Ω<br />

Axiom 3 (Sum <strong>and</strong> Product Axiom). For all ϕ, ψ : Pfin(Ω) → R:<br />

limλ↑Ω ϕ(λ) + limλ↑Ω ψ(λ) = limλ↑Ω(ϕ(λ) + ψ(λ))<br />

limλ↑Ω ϕ(λ) · limλ↑Ω ψ(λ) = limλ↑Ω(ϕ(λ) · ψ(λ))<br />

In the next definitions, we want to introduce a concept of “measure” of a set<br />

E ⊆ R, that we will call the numerosity of E. The idea that lies behind numerosity<br />

is that if E is finite, then the numerosity of E is the number of elements of E.<br />

Then, if E is infinite, we will define its numerosity by extending this idea using the<br />

Ω-limit. This new concept of “measure” will give rise to interesting results, that will<br />

be thoroughly studied in Section 1.6.<br />

Before we define numerosity, we find useful to introduce the following convention:<br />

Definition 1.3.1. If E ⊆ Ω <strong>and</strong> ϕ : Pfin(Ω) → R, then we set<br />

lim ϕ(λ) = lim ϕ(λ ∩ E)<br />

λ↑E λ↑Ω<br />

This definition allows us to simplify the notation of Ω-limit in a number of<br />

situations, the first of which is the definition of numerosity.<br />

Definition 1.3.2. The numerosity n(E) of a set E ⊆ Ω is defined by<br />

n(E) = lim<br />

λ↑E |λ|<br />

We can think of numerosity as a way to “extend” the concept of “number of<br />

elements” to infinite sets. In Section 1.6, we will give a precise meaning to this<br />

intuitive statement. Now, we want to ensure that numerosity has at least one familiar<br />

property, <strong>and</strong> the only way to do so is by introducing a new axiom. We will<br />

distinguish this axiom from the axioms regarding Ω-limit by calling it by the name<br />

of “numerosity axiom”.<br />

Numerosity Axiom 1. If a, b ∈ Q, then<br />

n([a, b))<br />

n([0, 1))<br />

= (b − a)<br />

This axiom rules that, for intervals <strong>with</strong> rational extrems, once we “normalize”<br />

their numerosity by a meaningful constant factor, the result is the length of the<br />

interval. We signal that we couldn’t have ruled a more general axiom that imposes<br />

the equality for arbitrary intervals, but the proof of this statement requires the<br />

introduction of new concepts depending upon Ω-limit. The interested reader could<br />

easily deduce it as a consequence of Proposition 6.4.5.<br />

8


1.4 First properties of Ω-limit<br />

1.4. FIRST PROPERTIES OF Ω-LIMIT<br />

In the previous section, we have introduced the axioms of Ω-Calculus: most of<br />

them regard a new mathematical operation, called Ω-limit, which is quite different<br />

from the limit used in classical calculus. In this section, we will begin to study the<br />

behaviour of this Ω-limit, highlighting some of its main properties.<br />

The first property we will show is that, if two functions are different for all<br />

λ ∈ Pfin(Ω), then their Ω-limits will be different, too. This is quite in contrast <strong>with</strong><br />

what happens to usual limits of functions.<br />

Proposition 1.4.1. If ϕ(λ) = ψ(λ) for all λ ∈ Ω, then limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ).<br />

Proof. Begin by noticing that, by hypothesis,<br />

for all λ ∈ Pfin(Ω), so that the function<br />

ϕ(λ) − ψ(λ) = 0<br />

η(λ) =<br />

1<br />

ϕ(λ) − ψ(λ)<br />

is well defined for all λ ∈ Pfin(Ω). Moreover, the equality<br />

η(λ)(ϕ(λ) − ψ(λ)) = 1<br />

holds for all λ ∈ Pfin(Ω). Axiom 2 <strong>and</strong> 3 ensure that<br />

lim η(λ) · lim(ϕ(λ)<br />

− ψ(λ)) = lim η(λ)(ϕ(λ) − ψ(λ)) = 1<br />

λ↑Ω λ↑Ω λ↑Ω<br />

<strong>and</strong> this implies, in particular, that<br />

so the thesis follows.<br />

lim(ϕ(λ)<br />

− ψ(λ)) = 0<br />

λ↑Ω<br />

The second property is that if a function assumes only a finite number of values,<br />

then its Ω-limit must be one of those values.<br />

Proposition 1.4.2. If E = {x1, . . . , xn} ⊂ R is finite <strong>and</strong> ϕ(λ) ∈ E for all λ ∈<br />

Pfin(Ω), then limλ↑Ω ϕ(λ) ∈ E.<br />

Proof. By hypothesis,<br />

lim<br />

λ↑Ω<br />

n<br />

(ϕ(λ) − xi) = 0<br />

i=1<br />

Explicitating the formula, we have that exists i such as limλ↑Ω ϕ(λ) = xi, as we<br />

wanted.<br />

For further development of the theory, for every couple of functions ϕ <strong>and</strong> ψ :<br />

Pfin(Ω) → R, we find useful to define a set Qϕ,ψ that contains all λ ∈ Pfin(Ω) such<br />

as ϕ(λ) = ψ(λ). These sets will play a major role in underst<strong>and</strong>ing the behaviour<br />

of Ω-limit.<br />

9


CHAPTER 1. Ω-CALCULUS<br />

Definition 1.4.3. For all ϕ, ψ : Pfin(Ω) → R, we define<br />

Clearly, Qϕ,ψ ⊆ Pfin(Ω).<br />

Qϕ,ψ = {λ ∈ Pfin(Ω) : ϕ(λ) = ψ(λ)}<br />

In the next proposition, we will explicitely show the relation between the sets<br />

Qϕ,ψ <strong>and</strong> Ω-limit.<br />

Proposition 1.4.4. For all ϕ, ψ : Pfin(Ω) → R,<br />

1. if limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ), then Qϕ,ψ = ∅;<br />

2. If limλ↑Ω ϕ(λ) = 0 <strong>and</strong> Qϕ,c0 ⊆ Qψ,c0 (e.g. ψ vanishes where ϕ does), then<br />

limλ↑Ω ψ(λ) = 0;<br />

3. If limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ), then for all η, γ : Pfin(Ω) → R, if Qϕ,ψ ⊆ Qη,γ,<br />

then limλ↑Ω η = limλ↑Ω γ.<br />

Proof. Part 1 of this Proposition is the contrapositive of Proposition 1.4.1.<br />

To prove part 2, define<br />

<br />

ϕ(λ) if ϕ(λ) = 0<br />

δ(λ) =<br />

1 if ϕ(λ) = 0<br />

Since δ(λ) = 0 for all λ ∈ Pfin(Ω), limλ↑Ω δ(λ) = ξ = 0. It’s easy to verify that<br />

for all λ ∈ Pfin(Ω). This implies<br />

(δ(λ) − ϕ(λ))ψ(λ) = 0<br />

0 = lim<br />

λ↑Ω (δ(λ) − ϕ(λ))ψ(λ) = lim<br />

λ↑Ω (δ(λ) − ϕ(λ)) · lim<br />

λ↑Ω ψ(λ) = ξ · lim<br />

λ↑Ω ψ(λ)<br />

<strong>and</strong>, since ξ = 0, then limλ↑Ω ψ(λ) = 0.<br />

To prove part 3, define the function ϕ ′ (λ) = ϕ(λ) − ψ(λ). By hypothesis,<br />

limλ↑Ω ϕ ′ (λ) = 0, <strong>and</strong> it’s clear that Qϕ ′ ,c0 = Qϕ,ψ. Now, define η ′ (λ) = η(λ) − γ(λ).<br />

It’s easy to verify that Qϕ ′ ,η ′ = Qϕ,ψ <strong>and</strong>, using part 2 of this proposition, we conclude<br />

that limλ↑Ω η ′ (λ) = 0 or, in other words,<br />

The thesis follows.<br />

lim η<br />

λ↑Ω ′ (λ) = lim(η(λ)<br />

− γ(λ)) = lim η(λ) − lim γ(λ) = 0<br />

λ↑Ω λ↑Ω λ↑Ω<br />

Remark 1.4.5. We want to spend a few words on the content of Proposition 1.4.4.<br />

For any two functions ϕ <strong>and</strong> ψ, we have defined the set Qϕ,ψ <strong>and</strong> we have shown<br />

that this set plays a fundamental role in determining the Ω-limit of these functions:<br />

if we know, for instance, that ϕ <strong>and</strong> ψ have the same Ω-limit, then every other pair<br />

of functions η <strong>and</strong> γ who coincide at least on every λ ∈ Qϕ,ψ satisfy<br />

10<br />

lim η(λ) = lim γ(λ)<br />

λ↑Ω λ↑Ω


1.4. FIRST PROPERTIES OF Ω-LIMIT<br />

A consequence of this result is that, if ϕ <strong>and</strong> ψ have the same Ω-limit, we can<br />

change the values of ϕ(λ) even for all λ ∈ Qϕ,ψ, but the function obtained will have<br />

the same Ω-limit as ϕ.<br />

We can go even further: if a function ϕ is well defined only for all λ in a qualified<br />

set Q, this is sufficient to ensure that ϕ has an Ω-limit. In fact, we can define<br />

arbitrarily the values of ϕ(λ) for all λ ∈ Q c , <strong>and</strong> the corresponding functions will<br />

always have the same Ω-limit. This feature presents a strong analogy to the case of<br />

real, convergent sequences {xn}n∈N: even if we change a finite number of the xn, the<br />

limit of the sequence doesn’t change <strong>and</strong>, if there is m ∈ N such that the sequence is<br />

defined only for n ≥ m, then this is sufficient to ensure that the limit of the sequence<br />

is well defined.<br />

Since the sets Qϕ,ψ satisfying the hypothesis that ϕ <strong>and</strong> ψ have the same Ω-limit<br />

play such an important role in the theory, we will give them a name.<br />

Definition 1.4.6. We will call Q ⊆ P(Pfin(Ω)) the set of all Qϕ,ψ satisfying<br />

limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ):<br />

<br />

Q =<br />

Qϕ,ψ : lim<br />

λ↑Ω ϕ(λ) = lim<br />

λ↑Ω ψ(λ)<br />

We will say that a set Q ∈ Q is qualified, <strong>and</strong> we will call Q the set of qualified<br />

sets.<br />

All the sets Q ∈ Q are, so to speak, the sets that “count” for the Ω-limit or, in<br />

other words, the Ω-limit of a functon depends only on its behaviour on the elements<br />

of a set Q ∈ Q. This idea is precised in the following Corollary.<br />

Corollary 1.4.7. For all ϕ, ψ : Pfin(Ω) → R, limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ) if <strong>and</strong> only<br />

if Qϕ,ψ ∈ Q.<br />

Proof. One of the implications is true by definition, so we will focus on the other.<br />

If Qϕ,ψ ∈ Q, then there are two functions η <strong>and</strong> γ such as limλ↑Ω η = limλ↑Ω γ <strong>and</strong><br />

Qη,γ = Qϕ,ψ. Now we can apply part 3 of Proposition 1.4.4 to obtain limλ↑Ω ϕ(λ) =<br />

limλ↑Ω ψ(λ).<br />

The family of sets Q has some additional properties <strong>with</strong> respect to set operations:<br />

they are summarized in the next Proposition.<br />

Proposition 1.4.8. Q satisfies the following properties:<br />

1. If A ∈ Q <strong>and</strong> B ⊇ A, then B ∈ Q;<br />

2. If A, B ∈ Q, then A ∩ B ∈ Q;<br />

3. If A ∪ B ∈ Q, then A ∈ Q or B ∈ Q;<br />

4. A ∈ Q ⇔ A c ∈ Q;<br />

5. If F is finite (e.g. F = {λ1, . . . , λn}), then F ∈ Q;<br />

<br />

11


CHAPTER 1. Ω-CALCULUS<br />

6. For all r ∈ R, Fr = {λ ∈ Pfin(Ω) : r ∈ λ} ∈ Q;<br />

7. ∅ ∈ Q <strong>and</strong> Pfin(Ω) ∈ Q.<br />

Proof. The first part of this Proposition follows from Proposition 1.4.4.<br />

To prove part 2, we will show that, if A = Qϕ,ψ <strong>and</strong> B = Qη,γ, then A∩B = Qδ,φ,<br />

<strong>and</strong> δ <strong>and</strong> φ have the same Ω-limit. If A ⊆ B or B ⊆ A, the thesis follows from<br />

part 1. Let’s suppose that this is not the case. Let’s call ξ = limλ↑Ω ϕ(λ) <strong>and</strong><br />

ζ = limλ↑Ω η(λ). If ξ = ζ, Axiom 1 ensures that there is a function ν such as<br />

limλ↑Ω ν(λ) = ξ −ζ. If we define η ′ = η +ν <strong>and</strong> γ ′ = γ +ν, we have that Qη ′ ,γ ′ = Qη,γ<br />

<strong>and</strong><br />

lim η<br />

λ↑Ω ′ = lim γ<br />

λ↑Ω ′ = ζ + ξ − ζ = ξ<br />

We have shown that, <strong>with</strong>out loss of generality, we can suppose<br />

Now, it’s clear that<br />

<strong>and</strong><br />

lim ϕ(λ) = lim ψ(λ) = lim η(λ) = lim γ(λ)<br />

λ↑Ω λ↑Ω λ↑Ω λ↑Ω<br />

Qϕ−η,ψ−γ = Qϕ,ψ ∩ Qη,γ<br />

lim(ϕ(λ)<br />

− η(λ)) = lim(ψ(λ)<br />

− γ(λ)) = 0<br />

λ↑Ω λ↑Ω<br />

<strong>and</strong> that’s what we wanted to prove.<br />

The proof of part 3 is by converse. Let’s suppose A ∈ Q <strong>and</strong> B ∈ Q. From<br />

the definition we have that, for all functions ϕ <strong>and</strong> ψ, the hypothesis Qϕ,ψ = A or<br />

Qϕ,ψ = B implies limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ). Since the implication holds if both the<br />

premises are true, we have that, if for ϕ <strong>and</strong> ψ the equality Qϕ,ψ = A ∪ B holds,<br />

then limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ), <strong>and</strong> so we conclude A ∪ B ∈ Q, as we wanted.<br />

To prove the first implication of part 4, let’s suppose A ∈ Q, so that there exist<br />

ϕ, ψ such as A = Qϕ,ψ <strong>and</strong> limλ↑Ω ϕ = limλ↑Ω ψ. Let’s define the functions:<br />

<strong>and</strong><br />

η(λ) =<br />

γ(λ) =<br />

ϕ(λ) if λ ∈ A<br />

1 if λ ∈ A c<br />

ϕ + 1 if λ ∈ A<br />

1 if λ ∈ A c<br />

Clearly, Qη,γ = A c <strong>and</strong>, by part 3 of Proposition 1.4.4, limλ↑Ω η(λ) = limλ↑Ω ϕ(λ),<br />

while limλ↑Ω γ(λ) = limλ↑Ω ϕ(λ)+1. This shows that A c ∈ Q. Conversely, if A c ∈ Q,<br />

we can apply the same argument to show that A = (A c ) c ∈ Q.<br />

To prove part 5, we can consider the function c0 <strong>and</strong> the function<br />

<br />

c<br />

0 if λ ∈ F<br />

ϕ(λ) =<br />

1 if λ ∈ F<br />

By definition, we have Qc0,ϕ = F c , so ϕ is eventually constant (take λ0 = n<br />

i=1 λi)<br />

<strong>and</strong>, by Axiom 2, limλ↑Ω ϕ(λ) = 0. This implies F c ∈ Q <strong>and</strong>, by part 4, F ∈ Q.<br />

12


1.4. FIRST PROPERTIES OF Ω-LIMIT<br />

To prove part 6, we have that, if we have two functions ϕ <strong>and</strong> ψ such that Qϕ,ψ =<br />

Fr, then ϕ(λ)−ψ(λ) = 0 is eventually true (take λ0 = {r}), so Axiom 2 ensures that<br />

limλ↑Ω(ϕ(λ)−ψ(λ)) = 0, <strong>and</strong> Axiom 3 allows to conclude limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ).<br />

An application of Corollary 1.4.7 gives us the thesis.<br />

Part 7 follows from Axiom 2 <strong>and</strong> from part 4 of this Proposition.<br />

A family of subsets satisfying the seven properties itemized in this proposition<br />

is called a nonprincipal fine ultrafilter on Pfin(Ω). This important notion is crucial<br />

in constructing models for Ω-Calculus, <strong>and</strong> will be further studied in Sections 1.10<br />

<strong>and</strong> 1.11.<br />

As final result of this section, we will show that R ∗ has a natural structure of<br />

ordered field whose order extends the order on R.<br />

Proposition 1.4.9. R ∗ is an ordered field whose positive part (R ∗ ) + is<br />

(R ∗ ) + =<br />

<br />

<br />

lim ϕ(λ) : ϕ(λ) > 0 ∀λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

Proof. From Axiom 3, we have that (R ∗ ) + is closed under sum <strong>and</strong> products. More-<br />

over, if we set<br />

we have<br />

(R ∗ ) − =<br />

(R ∗ ) − =<br />

<br />

<br />

lim ϕ(λ) : −ϕ(λ) > 0 ∀λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

<br />

<br />

lim ϕ(λ) : ϕ(λ) < 0 ∀λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

We need only to show that (R∗ ) − , {0} <strong>and</strong> (R∗ ) + form a partition of R∗ . First<br />

of all, notice that Proposition 1.4.1 ensures that those sets are pairwise disjoint.<br />

Now, for every ϕ : Pfin(Ω) → R, define<br />

ϕ + <br />

ϕ(λ) if ϕ(λ) > 0<br />

(λ) =<br />

1 if ϕ(λ) ≤ 0<br />

<strong>and</strong><br />

It’s easy to see that<br />

ϕ − (λ) =<br />

ϕ(λ) if ϕ(λ) < 0<br />

−1 if ϕ(λ) ≥ 0<br />

ϕ(λ)(ϕ(λ) − ϕ + (λ))(ϕ(λ) − ϕ − (λ)) = 0<br />

for all λ ∈ Pfin(Ω), so that limλ↑Ω ϕ(λ) = 0 implies<br />

or<br />

as we wanted.<br />

lim ϕ(λ) = lim ϕ<br />

λ↑Ω λ↑Ω + (λ) ∈ (R ∗ ) +<br />

lim ϕ(λ) = lim ϕ<br />

λ↑Ω λ↑Ω − (λ) ∈ (R ∗ ) −<br />

13


CHAPTER 1. Ω-CALCULUS<br />

1.5 The Transfer Principle<br />

In this section, we want to formulate a transfer principle for the notion of Ω-limit.<br />

Since a precise formulation of the transfer principle is possible only after the introduction<br />

of some logic formalism, we settle for an informal version where the<br />

important concept of of “elementary property” is left undefined.<br />

Transfer Principle (Informal version). An “elementary property” P is satisfied by<br />

real numbers ϕ1(λ), . . . , ϕk(λ) for all λ in a qualified set if <strong>and</strong> only if P is satisfied<br />

by the Ω-limits limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕk(λ).<br />

In other words, Transfer Principle tells us that for all “elementary properties”<br />

P , the set QP = {λ ∈ Pfin(Ω) : P (ϕ1(λ), . . . , ϕk(λ)) holds} ∈ Q if <strong>and</strong> only if<br />

P (limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕk(λ)) holds. If P is a property such as QP ∈ Q, we will<br />

say that P holds in a qualified set or that P holds almost everywere. Sometimes,<br />

we will also use the abbreviation “P holds a.e.”<br />

For now, the Transfer Principle can be taken only at an an informal, intuitive<br />

level, because we have no formal definition of an “elementary property” (<strong>and</strong> this<br />

is the reason why we always put “elementary property” between quotation marks).<br />

The content of Transfer Principle is made precise only when a rigorous definition of<br />

“elementary property” is given, <strong>and</strong> this will be done in Chapter 5 of this paper.<br />

Now we want to prove the Transfer Principle for some fundamental “elementary<br />

properties” we will use quite often.<br />

Proposition 1.5.1 (Transfer of inequalities). Let ϕ, ψ : Pfin(Ω) → R. Then<br />

1. ϕ(λ) = ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ);<br />

2. ϕ(λ) < ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω ϕ(λ) < limλ↑Ω ψ(λ);<br />

3. ϕ(λ) ≤ ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω ϕ(λ) ≤ limλ↑Ω ψ(λ).<br />

Proof. To prove part 1, first notice that that, in both cases, Q = {λ ∈ Pfin(Ω) :<br />

ϕ(λ) = ψ(λ)} ∈ Q, <strong>and</strong> Proposition 1.4.8 ensures that Qϕ,ψ = Q c ∈ Q. Now, the<br />

thesis follows from Corollary 1.4.7.<br />

To prove part 2, suppose that ϕ(λ) < ψ(λ) a.e. We can define<br />

δ(λ) = ψ(λ) − ϕ(λ)<br />

<strong>and</strong> notice that, by hypothesis, Q = {λ ∈ Pfin(Ω) : δ(λ) > 0} ∈ Q. If limλ↑Ω δ(λ) ≤<br />

0, then we would have Q c ∈ Q, but once again this contradicts Proposition 1.4.8.<br />

The proof of part 3 can be obtained in a similar way.<br />

The reader should not think that all the properties of real numbers are elementary.<br />

Indeed, we can even show explicit counterexamples of properties that are not<br />

elementary.<br />

Example 1.5.2. As a property of real numbers, the property P (x) saying “x is a<br />

natural number” is not elementary, since |λ| ∈ N for all λ ∈ Pfin(Ω), but clearly<br />

14


1.6. NUMEROSITY<br />

limλ↑Ω |λ| ∈ N. More generally, we will soon see in Proposition 1.7.6 that the property<br />

P (x) saying “x ∈ E” is elementary if <strong>and</strong> only if E is finite. We remark that,<br />

in this case, the property x ∈ E is equivalent to “given the real numbers x1, . . . , xn,<br />

then x = x1 or x = x2 . . . or x = xn”.<br />

We signal that the property ϕ(λ) ∈ N is elementary if considered as a property<br />

of numbers <strong>and</strong> sets. In fact, in the context of a theory that rules the Ω-limit of<br />

sets, <strong>and</strong> not only of real numbers, it can be proved that the property ϕ(λ) ∈ N,<br />

considered as a property of numbers <strong>and</strong> sets, transfers to limλ↑Ω ϕ(λ) ∈ limλ↑Ω cN(λ).<br />

Of course, this statement has no meaning until we define <strong>with</strong>in the theory what is<br />

the Ω-limit of a function ϕ : Pfin(Ω) → P(R). Even if this will be done in Section<br />

1.7, we rem<strong>and</strong> the study of a more general version of Transfer Principle to Section<br />

4.4 of this paper, where we will have a more general theory of Ω-limit <strong>and</strong> we will<br />

be able to prove the validity of Transfer of membership for arbitrary functions.<br />

1.6 Numerosity<br />

Numerosity is a new way to “count” the number of elements of a set by using Ωlimit.<br />

In this section, we will show some of its properties <strong>and</strong> highlight the most<br />

important of them.<br />

First of all, we will show that, if F ⊂ R is a finite set, then n(F ) = |F |. This idea<br />

has already been expressed informally in Section 1.3, while introducing the definition<br />

of numerosity.<br />

Proposition 1.6.1. If F ⊂ R is a finite set, then n(F ) = |F |.<br />

Proof. Since F is finite, F ∈ Pfin(Ω). This ensures that, for all λ ⊇ F , F ∩ λ = F .<br />

By definition of numerosity, we obtain that the equality |F ∩ λ| = |F | is eventually<br />

true, so the thesis follows.<br />

Proposition 1.6.1 ensures that, when F is finite, then numerosity satisfies the<br />

basic intuitive properties about counting: finite additivity <strong>and</strong> monotony. We already<br />

know that, when dealing <strong>with</strong> infinite sets, cardinality loses the property of<br />

monotony: for example, N ⊂ Z but |N| = |Z|. It’s only natural to ask whether this<br />

happens also for numerosity. It turns out that, thanks to the Ω-limit <strong>and</strong> Transfer<br />

Principle, numerosity is always finitely additive <strong>and</strong> monotone.<br />

Proposition 1.6.2. If E1, E2 ⊆ R are two disjoint sets, then n(E1 ∪ E2) = n(E1) +<br />

n(E2).<br />

Proof. By hypothesis,<br />

|λ ∩ (E1 ∪ E2)| = |λ ∩ E1| + |λ ∩ E2|<br />

holds for all λ ∈ Pfin(Ω), so we can apply Transfer Principle <strong>and</strong> conclude.<br />

This result can be easily generalized to the case of finite unions of disjoint sets.<br />

15


CHAPTER 1. Ω-CALCULUS<br />

Corollary 1.6.3 (Finite additivity of numerosity). For all finite sequences of real<br />

subsets {Ei}i∈{1,...,n} satisfying Ei ∩ Ej = ∅ whenever i = j,<br />

<br />

n<br />

<br />

n<br />

n = n(Ei)<br />

i=1<br />

Ei<br />

In the next Proposition, we will prove that numerosity is monotone.<br />

Proposition 1.6.4 (Monotony of numerosity). For all E, F ⊆ R, if E ⊂ F , then<br />

n(E) < n(F ).<br />

Proof. We can write F = E ∪ (F \ E) <strong>and</strong>, by hypothesis, F \ E = ∅. By Corollary<br />

1.6.3, we have that n(F ) = n(E) + n(F \ E). If F \ E is a finite set, we already<br />

know n(F \ E) > 0; if F \ E is infinite, then pick x ∈ F \ E: clearly, n(F \ E) ≥<br />

|(F \ E) ∩ {x}| > 0. This shows that in both cases n(F \ E) > 0, so the proof is<br />

concluded.<br />

As a consequence of monotony, we deduce that infinite sets have infinite numerosity.<br />

Corollary 1.6.5. If E is infinite, then n(E) is an infinite number.<br />

Proof. Thanks to Proposition 1.6.4, n(E) > n(F ) for all finite F ⊂ E <strong>and</strong>, by the<br />

fact that E contains finite subsets <strong>with</strong> arbitrary number of elements, this means<br />

n(E) > n for all n ∈ N.<br />

We want to remark that the facts that numerosity is always finitely additive<br />

<strong>and</strong> monotone can be seen as consequences of the Transfer Principle for equalities<br />

<strong>and</strong> inequalities: thanks to the Ω-limit, the properties of the finite cardinality are<br />

extended to numerosity of arbitrary sets.<br />

The monotony of numerosity seems to be close to Euclid’s idea that the whole is<br />

greater than the parts, an idea that has been somehow challenged by the counting<br />

methods of cardinality <strong>and</strong> of ordinals. In this regard, we find appropriate the<br />

following quotation:<br />

The possibility that whole <strong>and</strong> part may have the same number of terms is, it must<br />

be confessed, shocking to common sense. (Russell, 1903, Principles of <strong>Mathematics</strong>,<br />

p. 358)<br />

We conclude this discussion <strong>with</strong> a well known example:<br />

Example 1.6.6. The set of natural numbers is the disjoint union of the set E<br />

of even numbers <strong>and</strong> the set O of odd numbers. By Proposition 1.6.2, we know<br />

that n(N) = n(E) + n(O), but our theory doesn’t tell if, for instance n(E) = n(O).<br />

This difficulty is due to the fact that, in the axiom of numerosity, we focused on<br />

a property that links the numerosity to the metric of R. If we were interested in<br />

other applications of numerosity, we could have chosen different axioms to ensure<br />

the validity of some other properties useful for our purposes.<br />

We signal that, once we have shown a model for Ω-Calculus, we will be able to<br />

calculate the exact numbers n(E) <strong>and</strong> n(O).<br />

16<br />

i=1


1.6. NUMEROSITY<br />

In the previous part of this Section, we have shown some properties of numerosity<br />

that are independent from Numerosity Axiom 1. Now, we want to focus on the<br />

behaviour of numerosity in the case of real subsets. At first, thanks to Proposition<br />

1.6.4, we will show how the equation of Numerosity Axiom 1 can be extended to all<br />

a, b ∈ R.<br />

Proposition 1.6.7. For all a, b ∈ R, then<br />

n([a, b))<br />

n([0, 1))<br />

∼ b − a<br />

Proof. If a <strong>and</strong> b ∈ Q, then there’s nothing to prove. Otherwise, for all n ∈ N we<br />

can find a + n <strong>and</strong> b + n ∈ Q such that a + n < a <strong>and</strong> b + n > b <strong>and</strong><br />

0 ≤ (b + n − a + n ) − (b − a) < 1<br />

n<br />

(1.1)<br />

On the other h<strong>and</strong>, we can also find a− n <strong>and</strong> b− n ∈ Q such that a− n < a <strong>and</strong> b− n > b<br />

<strong>and</strong><br />

0 ≤ (b − a) − (b − n − a − n ) < 1<br />

(1.2)<br />

n<br />

Putting together equations 1.1 <strong>and</strong> 1.2, we obtain<br />

By Proposition 1.6.4, we deduce<br />

lim(b<br />

λ↑Ω +<br />

|λ| − a+<br />

|λ| ) ∼ lim<br />

λ↑Ω (b− |λ| − a−<br />

|λ| ) ∼ b − a (1.3)<br />

n([a − n , b − n ))<br />

n([0, 1))<br />

n([a, b))<br />

<<br />

n([0, 1)) < n([a−n , b− n ))<br />

n([0, 1))<br />

for all n ∈ N <strong>and</strong>, by equation 1.3, we can conclude<br />

n([a, b))<br />

n([0, 1))<br />

∼ b − a<br />

This proposition, while helpful to determine the numerosity of arbitrary intervals,<br />

has also an interesting consequence:<br />

Corollary 1.6.8. Numerosity is not invariant <strong>with</strong> respect to translations.<br />

This consideration inspires the following question: are there some transformations<br />

of the real line that preserve numerosity? If the answer is yes, then how can<br />

they be characterized? In other words, besides the identity, what are the “isometries”<br />

of R <strong>with</strong> respect to numerosity?<br />

Indeed, we can find some characterizations of such functions, but they are rather<br />

nondescriptive <strong>and</strong> don’t tell us wether there are some functions that satisfy them.<br />

One of such characterizations is presented in the following Proposition.<br />

17


CHAPTER 1. Ω-CALCULUS<br />

Proposition 1.6.9. Let f : R → R satisfy |f(F )| = |F | for all finite subsets of R.<br />

If for all E ⊆ R the set {λ ∈ Pfin(Ω) : |f(E ∩ λ)| = |f(E) ∩ λ|} is qualified, then n<br />

is invariant <strong>with</strong> respect to f.<br />

Proof. By hypothesis, |E ∩ λ| = |f(E ∩ λ)| <strong>and</strong>, since |f(E ∩ λ)| = |f(E) ∩ λ| almost<br />

everywhere, by transfer of equality we conclude<br />

Hence the thesis.<br />

lim |f(E) ∩ λ| = lim |f(E ∩ λ)| = lim |E ∩ λ|<br />

λ↑Ω λ↑Ω λ↑Ω<br />

It’s clear that the functions characterized by Proposition 1.6.9 depend on the<br />

ultrafilter Q, hence the difficulty in determining their existence.<br />

Numerosity presents some other interesting challenges <strong>and</strong> open questions, but<br />

we will rem<strong>and</strong> them to Section 3.4 of this paper, when we will have at our disposal<br />

more mathematical tools.<br />

1.7 Extensions of sets<br />

The Ω-limit is a mathematical instrument that allows the construction of hyperreal<br />

numbers out of real numbers. In the same spirit, a notion of Ω-limit can also be<br />

given for functions ϕ : Pfin(Ω) → P(R): as a result, we will be able to “extend” the<br />

subsets of R into subsets of R ∗ by using the Ω-limit.<br />

We start by defining what we intend as the Ω-limit of a sequence of sets.<br />

Definition 1.7.1. Let {Eλ}λ∈Pfin(Ω) be a family of nonempty subsets of R. We<br />

define<br />

<br />

<br />

lim Eλ =<br />

λ↑Ω<br />

lim ϕ(λ) : ϕ(λ) ∈ Eλ<br />

λ↑Ω<br />

<strong>and</strong><br />

lim c∅(λ) = ∅<br />

λ↑Ω<br />

In the same spirit, if E is a subset of Ω, we define<br />

<br />

<br />

lim Eλ = lim ϕ(λ ∩ E) : ϕ(λ ∩ E) ∈ Eλ<br />

λ↑E λ↑Ω<br />

Thanks to Transfer Principle, the definition of limλ↑Ω Eλ can be given by an<br />

“almost everywhere” formulation. More precisely, we will show that<br />

Proposition 1.7.2. If {Eλ}λ∈Pfin(Ω) is a family of nonempty subsets of R, then<br />

18<br />

<br />

<br />

lim Eλ = lim ϕ(λ) : ϕ(λ) ∈ Eλ a.e.<br />

λ↑Ω λ↑Ω


1.7. EXTENSIONS OF SETS<br />

Proof. By hypothesis, for all λ ∈ Pfin(Ω) we can pick eλ ∈ Eλ. Now, suppose that<br />

ϕ(λ) ∈ Eλ a.e.: then, if we define a function<br />

ϕ ′ <br />

ϕ(λ) if ϕ(λ) ∈ Eλ<br />

(λ) =<br />

eλ if ϕ(λ) ∈ Eλ<br />

we have that ϕ ′ (λ) = ϕ(λ) almost everywhere <strong>and</strong>, by Transfer of equality, we conclude<br />

limλ↑Ω ϕ ′ (λ) = limλ↑Ω ϕ(λ). Now observe that, by construction, limλ↑Ω ϕ ′ (λ) ∈<br />

limλ↑Ω Eλ, so the thesis follows.<br />

Thanks to Definition 1.7.1, we can now define two kinds of extensions of sets:<br />

the first, more generic, is the natural extension.<br />

Definition 1.7.3. The natural extension of a set E ⊆ R is given by<br />

E ∗ <br />

<br />

= lim cE(λ) = lim ϕ(λ) : ϕ(λ) ∈ E<br />

λ↑Ω λ↑Ω<br />

where cE(λ) is the function identically equal to E.<br />

Then, we can define another extension, more restrictive than the one we’ve already<br />

seen, but <strong>with</strong> more interesting properties: the hyperfinite extension.<br />

Definition 1.7.4. The hyperfinite extension of a set E ⊆ R is given by<br />

E ◦ = lim<br />

λ↑E λ<br />

It’s worthy to explicitate this definition:<br />

E ◦ <br />

<br />

= lim ϕ(λ ∩ E) : ϕ(λ ∩ E) ∈ λ ∩ E<br />

λ↑Ω<br />

From the two definitions above, it’s clear that E ⊆ E ◦ ⊆ E ∗ .<br />

Now, we want to study some elementar properties of those extensions of sets,<br />

<strong>and</strong> their behaviour under some basic set operations. The first property we want to<br />

show is that equality is preserved under natural <strong>and</strong> hyperfinite extension.<br />

Proposition 1.7.5. Let E, F ⊆ R. E = F if <strong>and</strong> only if E ∗ = F ∗ if <strong>and</strong> only if<br />

E ◦ = F ◦ .<br />

Proof. If E = F , then it’s trivial that E ∗ = F ∗ <strong>and</strong> E ◦ = F ◦ .<br />

Now, we want to show that, if E = F , then E ∗ = F ∗ <strong>and</strong> E ◦ = F ◦ . Let’s suppose<br />

that E ⊆ F , so that we can choose x satisfying x ∈ E <strong>and</strong> x ∈ F . In particular, we<br />

have that, if ϕ(λ) ∈ F for all λ ∈ Pfin(Ω), then ϕ(λ) = x <strong>and</strong> so x ∈ F ◦ <strong>and</strong> x ∈ F ∗ ,<br />

so that E ◦ ⊆ F ◦ <strong>and</strong> E ∗ ⊆ F ∗ . This allows to conclude E ◦ = F ◦ , as desired.<br />

In the next two Propositions, we will see that the only sets for which it holds the<br />

equality chain E = E ◦ = E ∗ are finite sets. We will split the proof of this statement<br />

in two parts: at first, we will prove that E = E ∗ if <strong>and</strong> only if E is finite, <strong>and</strong> then<br />

that E ◦ = E ∗ if <strong>and</strong> only if E is finite.<br />

19


CHAPTER 1. Ω-CALCULUS<br />

Proposition 1.7.6. Let E ⊆ R. E = E ∗ if <strong>and</strong> only if E is finite.<br />

Proof. If E is finite, we can apply Proposition 1.4.2 <strong>and</strong> obtain immediately E ∗ = E.<br />

If E is infinite, then it’s not bounded or it’s bounded <strong>and</strong> have an accumulation<br />

point p ∈ R. In the first case, we can find a function ϕ : Pfin(Ω) → E such as<br />

for all n ∈ N one of the relations ϕ(λ) > n or −ϕ(λ) > n eventually holds, <strong>and</strong> so<br />

limλ↑Ω ϕ(λ) is an infinite number that doesn’t belong to E.<br />

In the second case, we can find a function ϕ : Pfin(Ω) → E such as |ϕ(λ) − p| <<br />

1/|λ| <strong>and</strong> ϕ(λ) = p. Taking the Ω-limit of both sides <strong>and</strong> calling ξ = limλ↑Ω ϕ(λ),<br />

we have that |ξ − p| < 1/k for all k ∈ N, so that |ξ − p| is infinitesimal or, in other<br />

words, ξ ∈ mon(p). Now, ϕ(λ) = p for all λ ∈ Pfin(Ω), so ξ = p, <strong>and</strong> this implies<br />

ξ ∈ R. Since ξ ∈ E ∗ <strong>and</strong> ξ ∈ E, we conclude E ∗ = E, as we wanted.<br />

Proposition 1.7.7. Let E ⊆ R. E ◦ = E ∗ if <strong>and</strong> only if E is finite.<br />

Proof. If E is finite, then, by Proposition 1.7.6, we have E = E ∗ ⊇ E ◦ ⊇ E, <strong>and</strong><br />

the thesis follows.<br />

If E is infinite, we want to show that E ∗ \ E ◦ = ∅ or, in other words, we want to<br />

show that there is a function ψ : Pfin(Ω) → E such as for every ϕ : Pfin(Ω) → E we<br />

have ζ = limλ↑Ω ψ(λ) ∈ E ◦ . To do so, first notice that, for all λ ∈ Pfin(Ω), E \λ = ∅,<br />

because E infinite <strong>and</strong> λ is finite. This implies that there exists a function ϕ such as<br />

ϕ(λ) ∈ E \ (λ ∩ E) for all λ ∈ Pfin(Ω). Now, since every ξ ∈ E ◦ is the Ω-limit of a<br />

function ϕ such as ϕ(λ ∩ E) ∈ λ ∩ E, it’s clear that, by definition, for all such ϕ we<br />

have ψ(λ) = ϕ(λ ∩ E). This <strong>and</strong> Proposition 1.4.1 ensures that ζ = limλ↑Ω ψ(λ) = ξ<br />

for all ξ ∈ E ◦ . By definition, we have also ζ ∈ E ∗ , so that ζ ∈ E ∗ \ E ◦ , as we<br />

wanted.<br />

It turns out that the relation between E ◦ <strong>and</strong> E ∗ can be further characterized:<br />

Proposition 1.7.8. for all E ⊆ R, E ◦ = E ∗ ∩ R ◦ .<br />

Proof. First, let’s suppose that ξ ∈ E ∗ ∩R ◦ . We can find a function ϕ : Pfin(Ω) → E<br />

such that limλ↑Ω ϕ(λ) = ξ <strong>and</strong> a function ψ : Pfin(Ω) → R such that ψ(λ) ∈ λ <strong>and</strong><br />

limλ↑Ω ψ(λ) = ξ. Since ϕ <strong>and</strong> ψ have the same Ω-limit, we have that Qϕ,ψ ∈ Q <strong>and</strong><br />

so ψ(λ) ∈ E for all λ ∈ Qϕ,ψ, but this implies ψ(λ) ∈ E ∩ λ for those λ. Thanks to<br />

Proposition 1.7.2, we can safely conclude that ξ ∈ E ◦ . This proves E ◦ ⊇ E ∗ ∩ R ◦ .<br />

The other inclusion is straightforward: if ξ ∈ E ◦ , then by definition we have<br />

ξ ∈ E ∗ <strong>and</strong> ξ ∈ R ◦ , so that ξ ∈ E ∗ ∩ R ◦ .<br />

Thanks to this Proposition, a lot of results about natural extension of sets will<br />

apply also to hyperfinite extension. Now, we will show that the operation of natural<br />

extension commutes <strong>with</strong> set operations <strong>and</strong>, as a consequence, we will obtain that<br />

the same result holds for hyperfinite extension.<br />

Proposition 1.7.9. for all E, F ⊆ R, the following are true:<br />

20<br />

1. E ∗ ∩ F ∗ = (E ∩ F ) ∗ ;<br />

2. E ⊆ F implies E ∗ ⊆ F ∗ ;


3. E ∗ ∪ F ∗ = (E ∪ F ) ∗ ;<br />

4. (E c ) ∗ = (E ∗ ) c ;<br />

5. E ∗ \ F ∗ = (E \ F ) ∗ .<br />

Proof. Part 1 is straightforward, once we notice that<br />

1.7. EXTENSIONS OF SETS<br />

{ϕ : Pfin(Ω) → E ∩ F } = {ϕ : Pfin(Ω) → E} ∩ {ϕ : Pfin(Ω) → F }<br />

To prove part 2, notice that, by part 1 <strong>and</strong> Proposition 1.7.6,<br />

E ∗ ⊆ F ∗ ⇔ E ∗ = E ∗ ∩ F ∗ = (E ∩ F ) ∗ ⇔ E = E ∩ F ⇔ E ⊆ F<br />

As for part 3, if E or F is empty, the thesis follows. Let’s suppose that both sets<br />

are nonempty, <strong>and</strong> pick e ∈ E <strong>and</strong> f ∈ F . For every function ϕ : Pfin(Ω) → E ∪ F ,<br />

define:<br />

<strong>and</strong><br />

ϕ ′ (λ) =<br />

ϕ ′′ (λ) =<br />

ϕ(λ) if ϕ(λ) ∈ E<br />

e if ϕ(λ) ∈ E<br />

ϕ(λ) if ϕ(λ) ∈ F<br />

f if ϕ(λ) ∈ F<br />

Since (ϕ(λ) − ϕ ′ (λ)) · (ϕ(λ) − ϕ ′′ (λ)) = 0, we have that limλ↑Ω ϕ(λ) = limλ↑Ω ϕ ′ (λ)<br />

or limλ↑Ω ϕ(λ) = limλ↑Ω ϕ ′′ (λ). In both cases, limλ↑Ω ϕ(λ) ∈ E ∗ ∪ F ∗ , <strong>and</strong> so (E ∪<br />

F ) ∗ ⊆ E ∗ ∪ F ∗ . The reverse inclusion follows from the fact that, by part 2, both<br />

E ∗ ⊆ (E ∪ F ) ∗ <strong>and</strong> F ∗ ⊆ (E ∪ F ) ∗ .<br />

To prove part 4, we can use that (E ∗ ) c is the only set X such as E ∗ ∩ X = ∅<br />

<strong>and</strong> E ∗ ∪ X = R ∗ . By using part 1 <strong>and</strong> 3 of this Proposition, it’s easily checked<br />

that X = (E c ) ∗ satisfies both conditions: E ∗ ∩ (E c ) ∗ = (E ∩ E c ) ∗ = ∅ ∗ = ∅ <strong>and</strong><br />

E ∗ ∪ (E c ) ∗ = (E ∪ E c ) ∗ = R ∗ .<br />

Corollary 1.7.10. for all E, F ⊆ R, the following are true:<br />

1. E ◦ ∩ F ◦ = (E ∩ F ) ◦ ;<br />

2. E ⊂ F implies E ◦ ⊆ F ◦ ;<br />

3. E ◦ ∪ F ◦ = (E ∪ F ) ◦ ;<br />

4. (E c ) ◦ = (E ◦ ) c ;<br />

5. E ◦ \ F ◦ = (E \ F ) ◦ .<br />

Proof. This is a consequence of Proposition 1.7.9 <strong>and</strong> Proposition 1.7.8.<br />

Now, we want to show how natural extension behaves <strong>with</strong> respect to countable<br />

union.<br />

21


CHAPTER 1. Ω-CALCULUS<br />

Proposition 1.7.11. Let {En}n∈N be disjoint subsets of R. Then,<br />

<br />

En ⊆ lim<br />

<br />

<br />

<br />

∗ n∈N<br />

En ⊆<br />

λ↑Ω<br />

n≤|λ|<br />

Moreover, the first inclusion is an equality if <strong>and</strong> only if <br />

n∈N En is finite, <strong>and</strong> the<br />

second inclusion is an equality if <strong>and</strong> only if En = ∅ is eventually true.<br />

Proof. If <br />

n∈N En is a finite set, then the thesis follows from Proposition 1.7.6 <strong>and</strong>,<br />

if En = ∅ is eventually true, then the thesis is trivial.<br />

Suppose that those are not the cases. Omitting redundant terms, we can suppose<br />

that |λ1| < |λ2| implies<br />

<br />

En ⊂ <br />

n≤|λ1|<br />

n≤|λ2|<br />

We can then define a function ϕ : Pfin(Ω) → R in such a way that<br />

⎛<br />

ϕ(λ) ∈ ⎝ <br />

En \ <br />

⎞<br />

⎠<br />

n≤|λ|+1<br />

for all λ ∈ Pfin(Ω). By definition, limλ↑Ω ϕ(λ) ∈ En for all n ∈ N, <strong>and</strong> this concludes<br />

the first part of the proof.<br />

As for the second part, we can define a function ψ : Pfin(Ω) → R in such a way<br />

that<br />

⎛<br />

ψ(λ) ∈ ⎝ <br />

En \ <br />

⎞<br />

⎠<br />

n∈N<br />

n∈N<br />

En<br />

n≤|λ|<br />

n≤|λ|<br />

for all λ ∈ Pfin(Ω). For such a ψ it holds limλ↑Ω ψ(λ) ∈ limλ↑Ω<br />

thesis follows.<br />

En<br />

En<br />

En<br />

<br />

n≤|λ| En, so the<br />

Now, we want to focus on the properties of hyperfinite sets that are not shared by<br />

the natural extension of sets. Maybe, the most important of them is the following:<br />

Proposition 1.7.12. for all E ⊆ R, E ◦ has maximum <strong>and</strong> minimum.<br />

Proof. The two functions max <strong>and</strong> min are well defined for all λ ∈ Pfin(Ω). If we<br />

take their limit, by definition we have<br />

<strong>and</strong><br />

η = lim<br />

λ↑E max(λ) ∈ E ◦<br />

ζ = lim<br />

λ↑E min(λ) ∈ E ◦<br />

By Transfer Principle, ζ ≤ ξ ≤ η is true for all ξ ∈ E ◦ , so that ζ <strong>and</strong> ξ are,<br />

respectively, the minimum <strong>and</strong> the maximum of E ◦ .<br />

We find opportune to give an explicit example of the property ruled by the<br />

previous Proposition.<br />

22


Example 1.7.13. Let E = R: by definition,<br />

R ◦ = {ξ ∈ R ∗ : ξ = lim<br />

λ↑Ω ϕ(λ), <strong>with</strong> ϕ(λ) ∈ λ}<br />

1.7. EXTENSIONS OF SETS<br />

From this set we are excluding, for instance, all the Ω-limits of functions that grows<br />

“too fast”. If we call α = max(R ◦ ), then we have, for example, that α 2 ∈ R ◦ . In<br />

this case, R ⊂ R ◦ ⊂ R ∗ , <strong>and</strong> all inclusions are strict.<br />

On the other h<strong>and</strong>, it’s clear that R ∗ has no maximum <strong>and</strong> no minimum: for<br />

every ϕ : Pfin(Ω) → R, it’s trivial to find ψ ′ <strong>and</strong> ψ ′′ : Pfin(Ω) → R satisfying the<br />

inequalities<br />

ψ ′ (λ) < ϕ(λ) < ψ ′′ (λ)<br />

for all λ ∈ Pfin(Ω). Thanks to Transfer of inequalities, we deduce that R ∗ has no<br />

maximum <strong>and</strong> no minimum.<br />

Since the number α = max(R ◦ ) plays a central role in our theory, we find appropriate<br />

to fix this notation through the rest of this paper <strong>with</strong> the following definition.<br />

Definition 1.7.14. We define α = max(R ◦ ).<br />

Another interesting property of hyperfinite sets is that every element of R ◦ other<br />

than ±α has an “immediate predecessor” <strong>and</strong> an “immediate successor”. Formally,<br />

this property can be written in the following way:<br />

Proposition 1.7.15. If ξ ∈ R ◦ is different than ±α, there exists ξ + <strong>and</strong> ξ − ∈ R ◦<br />

such that<br />

1. ξ + > ξ <strong>and</strong>, if ζ ∈ R ◦ <strong>and</strong> ζ > ξ, then ζ ≥ ξ + ;<br />

2. ξ − < ξ <strong>and</strong>, if ζ ∈ R ◦ <strong>and</strong> ζ < ξ, then ζ ≤ ξ − .<br />

Proof. By hypothesis, there exists ϕ : Pfin(Ω) → R such that limλ↑R ϕ(λ) = ξ <strong>and</strong><br />

ϕ(λ ∩ R) ∈ λ ∩ R for all λ ∈ Ω. From this function, we can define<br />

<strong>and</strong><br />

Sλ = {x ∈ R : x > ϕ(λ ∩ R)}<br />

Iλ = {x ∈ R : x < ϕ(λ ∩ R)}<br />

Notice that Sλ <strong>and</strong> Iλ are nonempty for a.e. λ ∈ Ω. If we call ψ + (λ) = min(λ ∩ Sλ)<br />

<strong>and</strong> ψ − (λ) = max(λ ∩ Iλ) then, by the choiche of ξ, ψ ± are well defined for almost<br />

every λ ∈ Ω. We can conclude that, if we call ξ ± = limλ↑R ψ ± (λ), then ξ ± ∈ R ◦ .<br />

Now, we only need to show that ξ ± satisfy the thesis of the proposition. Notice<br />

that, by construction <strong>and</strong> by the hypothesis ξ = ±α, the chain of inequalities<br />

ψ + (λ ∩ R) > ϕ(λ ∩ R) > ψ − (λ ∩ R)<br />

is true almost everywhere. Let’s now suppose that ζ ∈ R ◦ satisfies ζ > ξ. If we<br />

write ζ = limλ↑R δ(λ), Transfer Principle ensures that δ(λ ∩ R) > ϕ(λ ∩ R) a.e. By<br />

definition of ψ + , this also implies that δ(λ∩R) ≥ ψ + (λ∩R) holds almost everywhere,<br />

so part 1 follows by an application of Transfer Principle. The second part of the<br />

proposition can be proved in a similar way.<br />

23


CHAPTER 1. Ω-CALCULUS<br />

The properties ruled by Proposition 1.7.12 <strong>and</strong> by Proposition 1.7.15 can be<br />

interpreted as particular instances of “Transfer Principle for sets”, where some class<br />

of properties of finite sets are preserved under Ω-limit. This is the main reason of<br />

our interest in hyperfinite extension of sets: they share some properties <strong>with</strong> finite<br />

sets, <strong>and</strong> those properties may be used in interesting ways during the development<br />

of calculus.<br />

1.8 Extensions of functions<br />

In the previous section, we have defined the natural <strong>and</strong> the hyperfinite extension<br />

of subsets of R. Now, we wish to define a similar extension of functions <strong>with</strong> range<br />

<strong>and</strong> domain contained in R.<br />

Let’s consider a function f : A ⊆ R → R. By composing <strong>with</strong> f, we can turn<br />

every function ϕ : Pfin(Ω) → A in a function f ◦ ϕ : Pfin(Ω) → R. Moreover, this<br />

composition is coherent <strong>with</strong> the operation of Ω-limit.<br />

Proposition 1.8.1. Let f : A ⊆ R → R <strong>and</strong> ϕ, ψ : Pfin(Ω) → A.<br />

lim ϕ(λ) = lim ψ(λ) ⇒ lim f(ϕ(λ)) = lim f(ψ(λ))<br />

λ↑Ω λ↑Ω λ↑Ω λ↑Ω<br />

Proof. By definition, we have Qϕ,ψ ∈ Q <strong>and</strong> Qϕ,ψ = Qf(ϕ),f(ψ). The thesis follows<br />

from Corollary 1.4.7.<br />

Thanks to this result, we are allowed to define the extension of functions in<br />

analogy to what we’ve done in the previous Section in the case of sets. In particular,<br />

we will define two kind of extensions of functions: the natural extension <strong>and</strong> the<br />

hyperfinite extension.<br />

Definition 1.8.2. Let f : A ⊆ R → R. The natural extension of f, which we<br />

denote f ∗ , is defined by:<br />

f ∗<br />

<br />

lim ϕ(λ)<br />

λ↑Ω<br />

= lim f(ϕ(λ))<br />

λ↑Ω<br />

Notice that, by definition, f ∗ : A ∗ → R ∗ .<br />

The hyperfinite extension of a function f : A → R is defined as the restriction<br />

of f ∗ to A ◦ :<br />

f ◦ = f ∗<br />

|A ◦ : A◦ → R ∗<br />

Proposition 1.8.1 ensures that Definition 1.8.2 is well posed. Moreover, we have<br />

that<br />

∀a ∈ A, f ∗ <br />

(a) = f lim ca(λ)<br />

λ↑Ω<br />

= lim (f ◦ ca) (λ) = lim cf(a)(λ) = f(a)<br />

λ↑Ω λ↑Ω<br />

so the function f ∗ (<strong>and</strong>, as a consequence, the function f ◦ ), is an actual extension<br />

of the function f.<br />

Now, we want to show some properties of extension of functions. In the beginning,<br />

we will focus on natural extension of functions because, by definition, the<br />

24


1.8. EXTENSIONS OF FUNCTIONS<br />

results will apply also to their hyperfinite extension. Then, in analogy to what happened<br />

to the case of hyperfinite extension of sets, we will prove some properties of<br />

hyperfinite extension of functions that are not satisfied by natural extension.<br />

At first, we will show that the basic properties of functions are preserved under<br />

natural <strong>and</strong> hyperfinite extension.<br />

Proposition 1.8.3. Let f : A ⊆ R → B ⊆ R <strong>and</strong> g : B ⊆ R → C ⊆ R.<br />

1. [f(A)] ∗ = f ∗ (A ∗ );<br />

2. f is injective if <strong>and</strong> only if f ∗ is injective;<br />

3. f is surjective if <strong>and</strong> only if f ∗ is surjective;<br />

4. f is bijective if <strong>and</strong> only if f ∗ is bijective;<br />

5. (g ◦ f) ∗ = g ∗ ◦ f ∗ .<br />

Proof. To prove part 1, for a matter of commodity define B = [f(A)]. If ξ ∈ B ∗ ,<br />

then ξ = limλ↑Ω ϕ(λ), where ϕ(λ) ∈ B for all λ ∈ Pfin(Ω). In particular, for every<br />

λ, there is an element ψ(λ) ∈ A satisfying the equality f(ψ(λ)) = ϕ(λ). If we call<br />

ζ = limλ↑Ω ψ(λ), then we have<br />

f ∗ (ζ) = lim<br />

λ↑Ω f(ψ(λ)) = lim<br />

λ↑Ω ϕ(λ) = ξ<br />

<strong>and</strong> this proves [f(A)] ∗ ⊆ f ∗ (A ∗ ). Conversely, if ζ = limλ↑Ω ψ(λ) for some ψ satisfying<br />

ψ(λ) ∈ A for all λ ∈ Pfin(Ω), then f(ψ(λ)) ∈ B for all λ ∈ Pfin(Ω), hence<br />

<strong>and</strong> this proves the other inclusion.<br />

f ∗ (ζ) = lim<br />

λ↑Ω f(ψ(λ)) ∈ B ∗<br />

is injective. To prove the<br />

For part 2, notice that if f ∗ is injective, then also f = f ∗<br />

|A<br />

other implication, suppose that there are ξ = limλ↑Ω ϕ(λ) <strong>and</strong> ζ = limλ↑Ω ψ(λ) satisfying<br />

f ∗ (ξ) = f ∗ (ζ). This equality can be rewritten limλ↑Ω f(ϕ(λ)) = limλ↑Ω f(ψ(λ)).<br />

By Transfer of equalities, we conclude f(ϕ(λ)) = f(ψ(λ)) a.e. <strong>and</strong>, by the hypothesis<br />

that f is injective, we deduce ϕ(λ) = ψ(λ) almost everywhere. In particular,<br />

this implies ξ = ζ.<br />

Parts 3 <strong>and</strong> 4 of the Proposition follow from the equality stated in part 1.<br />

To prove part 5, let ϕ : Pfin(Ω) → A. Then,<br />

(g ◦ f) ∗<br />

<br />

lim ϕ(λ)<br />

λ↑Ω<br />

= lim((g<br />

◦ f) ◦ ϕ)(λ)<br />

λ↑Ω<br />

as we wanted.<br />

= lim(g<br />

◦ (f ◦ ϕ)(λ)<br />

λ↑Ω<br />

= g ∗<br />

<br />

<br />

lim(f<br />

◦ ϕ)(λ)<br />

λ↑Ω<br />

= g ∗<br />

<br />

f ∗<br />

<br />

lim ϕ(λ)<br />

λ↑Ω<br />

25


CHAPTER 1. Ω-CALCULUS<br />

Corollary 1.8.4. Let f : A ⊆ R → B ⊆ R <strong>and</strong> g : B ⊆ R → C ⊆ R.<br />

1. f is injective if <strong>and</strong> only if f ◦ is injective;<br />

2. f is surjective if <strong>and</strong> only if f ◦ is surjective;<br />

3. f is bijective if <strong>and</strong> only if f ◦ is bijective.<br />

Proof. This is a consequence of Proposition 1.8.3 <strong>and</strong> the definition of f ◦ .<br />

Notice that parts 1 <strong>and</strong> 5 of Proposition 1.8.3 don’t apply also to hyperfinite<br />

extension of functions. We need to settle one more property of hyperfinite extension<br />

of functions before we can show some explicit counterexamples. The property we<br />

need is that monotony is preserved under natural <strong>and</strong> hyperfinite extension.<br />

Proposition 1.8.5. Let f : A ⊆ R → R. f is strictly increasing if <strong>and</strong> only if f ∗<br />

is strictly increasing. The same holds replacing “strictly increasing” <strong>with</strong> “strictly<br />

decreasing”, “not increasing” <strong>and</strong> “not decreasing”.<br />

Proof. Let’s suppose that f is strictly increasing, that is, x < y implies f(x) < f(y).<br />

Let’s now consider ξ <strong>and</strong> ζ ∈ R ∗ , <strong>with</strong> ξ < ζ. If ξ = limλ↑Ω ϕ(λ) <strong>and</strong> ζ = limλ↑Ω ψ(λ),<br />

by hypothesis we have Q ∋ {λ ∈ Pfin(Ω) : ϕ(λ) < ψ(λ)} = {λ ∈ Pfin(Ω) :<br />

f(ϕ(λ)) < f(ψ(λ))}. Applying Transfer Principle, we conclude f(ξ) < f(ζ).<br />

For the other parts of the proposition, the proof is similar: if f is not decreasing,<br />

substitute “


1.9. TOWARDS A MODEL FOR Ω-CALCULUS<br />

Example 1.8.8. Let A be the open set (0, 1) <strong>and</strong> f : A → R be the function<br />

f(x) = 1/x 2 . Clearly, f(A) = (1, ∞), so [f(A)] ◦ = (1, ∞) ◦ . On the other h<strong>and</strong>, if<br />

we define 1/β = min((0, 1) ◦ ), we know that there’s a function ϕ : Pfin(A) → R such<br />

as limλ↑Ω ϕ(λ) = 1/β. If we compose ϕ <strong>with</strong> f ◦ , we deduce limλ↑A f ◦ (ϕ(λ)) = β 2 ,<br />

<strong>and</strong> we don’t know if β 2 ∈ (1, ∞) ◦ .<br />

In our theory, we have no way to show wether R ◦ is closed under square root or<br />

wether β 2 ∈ (1, ∞) ◦ is true or false. However, we anticipate that, in Section 1.12,<br />

we will show a model of Ω-Calculus where the above inclusion is false.<br />

As for the case of composition of functions, since it’s generally false that f ◦ (A ◦ ) =<br />

[f(A)] ◦ , the function g ◦ ◦ f ◦ may not even be well defined. On the other h<strong>and</strong>,<br />

it’s always well defined the function g ∗ ◦ f ◦ , <strong>and</strong> the interested reader could easily<br />

prove, <strong>with</strong> an argument similar to the proof of part 5 of Proposition 1.8.3, that<br />

(g ◦ f) ◦ = g ∗ ◦ f ◦ .<br />

1.9 Towards a model for Ω-Calculus<br />

In the previous sections, we began to develop axiomatically a new mathematical<br />

theory, <strong>with</strong>out addressing the problem of its consistency. Now, we want to give a<br />

proof of relative consistency for Ω-Calculus: in particular, we’ll show that Ω-Calculus<br />

admits a model. This is a well known procedure, most notably used in the second half<br />

of the XIX century to create euclidean models of non-euclidean geometries to show<br />

that they were “at least as consistent as” euclidean geometry. For an introduction<br />

about model theory <strong>and</strong> for a presentation of more recent applications, we rem<strong>and</strong><br />

to [10] <strong>and</strong> to [11].<br />

We begin by defining what a model for Ω-Calculus is.<br />

Definition 1.9.1. A model for Ω-Calculus consists of a map<br />

ϕ ↦→ lim<br />

λ↑Ω ϕ(λ)<br />

that assigns a Ω-limit limλ↑Ω ϕ(λ) to every function ϕ : Pfin(Ω) → R, in such a way<br />

that the axioms of Ω-Calculus are realized.<br />

In the next three sections, we will exhibit an algebraic model for Ω-Calculus:<br />

in particular, every axiom of Ω-Calculus will correspond to an algebraic theorem,<br />

whose truth can be proved by field theory. For this reason, we find useful to recall<br />

the basic notions from Algebra that will be used during the construction of the<br />

model. For an introduction to those topics, complete <strong>with</strong> proofs <strong>and</strong> details we<br />

have omitted, see for instance [8] or [9].<br />

Definition 1.9.2. Let A be a commutative ring <strong>with</strong> an identity element (from now<br />

on, we will simply say “let A be a ring”, but we’ll also require that the ring is<br />

commutative <strong>and</strong> has an identity element).<br />

A set i ⊂ A is an ideal of A if it’s a subgroup of the additive group A <strong>and</strong> Ai = i<br />

(e.g. for all a ∈ A <strong>and</strong> for all j ∈ i, aj ∈ i).<br />

An ideal i ⊂ A is principal if there exists x ∈ i such as i = (x) (e.g. for all j ∈ i<br />

exists a ∈ A satisfying ax = j).<br />

27


CHAPTER 1. Ω-CALCULUS<br />

An ideal i ⊂ A is prime if xy ∈ i implies x ∈ i or y ∈ i (or both).<br />

An ideal i ⊂ A is maximal if for all ideals j ⊂ A, i ⊆ j implies i = j.<br />

Definition 1.9.3. If i is an ideal of a ring A, we can form the quotient ring A/i.<br />

First of all, we give an equivalence relation ≡i on A:<br />

a ≡i b ⇔ a − b ∈ i<br />

<strong>and</strong> then we consider the cosets [a]i = {x ∈ A : x − a ∈ i} induced by this relation.<br />

If we call A/i = {[a]i : a ∈ A}, we can define a sum <strong>and</strong> a product on A/i based upon<br />

those in A:<br />

[a]i + [b]i = [a + b]i<br />

[a]i · [b]i = [ab]i<br />

It can be easily verified that these operations are well defined <strong>and</strong> independent of the<br />

choice of representatives. With these operations, A/i has a ring structure inherited<br />

from A.<br />

If i is an ideal of A, we have the following characterizations: i is a prime ideal<br />

if <strong>and</strong> only if A/i is an integral domain (that is, A/i has no zero divisors); i is a<br />

maximal ideal if <strong>and</strong> only if A/i is a field. We can immediately deduce that every<br />

maximal ideal is prime, because a field is an integral domain. The easy proof of<br />

those statements can be found in [8] or in chapter 11 of [9].<br />

Now, we’ll introduce a particular set, the set of real valued functions from<br />

Pfin(Ω).<br />

Fun(Pfin(Ω), R) = {ϕ : Pfin(Ω) → R}<br />

This set has a natural structure of commutative ring <strong>with</strong> the following operations:<br />

<strong>and</strong><br />

(ϕ + ψ)(λ) = ϕ(λ) + ψ(λ)<br />

(ϕ · ψ)(λ) = ϕ(λ) · ψ(λ)<br />

This ring is interesting to us, because we want to obtain a model for Ω-Calculus<br />

from a a quotient of this ring. With this goal in mind, we find useful to show some<br />

examples of ideals in Fun(Pfin(Ω), R).<br />

Example 1.9.4. Let λ0 ∈ Pfin(Ω) <strong>and</strong> iλ0 = {ϕ : Pfin(Ω) → R : ϕ(λ0) = 0}. We<br />

will show that iλ0 is a maximal, principal ideal of A = Fun(Pfin(Ω), R).<br />

Clearly, iλ0 is an additive subgroup of A. Now, let’s choose once <strong>and</strong> for all<br />

ϕ ∈ iλ0 such as ϕ(λ) = 0 whenever λ = λ0. We have that, for all η ∈ A<br />

(η · ϕ)(λ0) = η(λ0) · ϕ(λ0) = 0<br />

so that the inclusion ϕA ⊆ iλ0 is true. This shows also that Aiλ0 = iλ0, from which<br />

we can conclude that iλ0 is indeed an ideal in A.<br />

Now, we want to show that iλ0 = (ϕ) or, in other words, that every function<br />

ψ ∈ iλ0 can be written as ϕ · η for a suitable η ∈ A. Let’s pick ψ ∈ iλ0 <strong>and</strong> define a<br />

new function<br />

<br />

ψ(λ)/ϕ(λ) if λ = λ0<br />

η(λ) =<br />

1 if λ = λ0<br />

28


1.9. TOWARDS A MODEL FOR Ω-CALCULUS<br />

By definition, η ∈ A <strong>and</strong> (ϕ · η)(λ) = ψ(λ) for all λ ∈ Pfin(Ω). This, combined <strong>with</strong><br />

the inclusion we have shown above, allows to conclude iλ0 = (ϕ).<br />

To see that iλ0 is maximal, we will prove that Fun(Pfin(Ω), R)/iλ0 is a field. Let’s<br />

choose [ϑ] ∈ Fun(Pfin(Ω), R)/iλ0 <strong>with</strong> ϑ ∈ iλ0. We want to show that we can find an<br />

invertible representative for [ϑ].<br />

If ϑ(λ) = 0 for all λ ∈ Pfin(Ω), then is invertible in Fun(Pfin(Ω), R) (its inverse<br />

being the well defined function 1/ϑ), <strong>and</strong> then, passing to the quotient map, we see<br />

that [1/ϑ] is the inverse of [ϑ] in Fun(Pfin(Ω), R)/iλ0.<br />

Let’s now suppose that Qϑ,c0 = ∅ (e.g. ϑ(λ) = 0 for some λ ∈ Pfin(Ω)). Since<br />

[ϑ] = 0 in Fun(Pfin(Ω), R)/iλ0, we can assume that ϑ(λ0) = ξ = 0. If we define the<br />

function<br />

ϑ ′ <br />

ξ if λ = λ0<br />

(λ) =<br />

ϑ(λ) 2 + 1 if λ = λ0<br />

we have that ϑ − ϑ ′ ∈ iλ0 <strong>and</strong> this, by definition, means that [ϑ ′ ] = [ϑ]. But it’s<br />

easily seen that ϑ ′ (λ) = 0 for all λ ∈ Pfin(Ω), <strong>and</strong> so ϑ ′ is invertible, <strong>and</strong> so it is<br />

[ϑ ′ ]. This shows that [ϑ] has an invertible representative, as we wanted.<br />

One of the most important ideals that will be useful in giving a model for Ω-<br />

Calculus is the ideal of the functions eventually equal to zero.<br />

Definition 1.9.5. Let’s call<br />

i0 = {ϕ ∈ Fun(Pfin(Ω), R) : ∃λ0 ∈ Pfin(Ω) such as ϕ(λ) = 0 ∀λ ⊇ λ0}<br />

the set of all functions eventually equal to zero.<br />

Proposition 1.9.6. The set i0 is a nonprincipal ideal of Fun(Pfin(Ω), R), <strong>and</strong> is<br />

not prime (<strong>and</strong>, therefore, is also not maximal).<br />

Proof. We omit the routine checkings of the fact that i0 is an ideal.<br />

To see that i0 is nonprincipal, we want to show that, for every ϕ ∈ i0, there exists<br />

ψ ∈ i0 such that, for all η ∈ Fun(Pfin(Ω), R), ϕ · η = ψ. So let’s choose ϕ ∈ i0. By<br />

definition of i0, there exists λ0 ∈ Pfin(Ω) such as ϕ(λ) = 0 for all λ ⊇ λ0. Now, let<br />

λ1 ⊃ λ0. If we define<br />

<br />

1 if λ ⊆ λ1<br />

ψ(λ) =<br />

0 if λ ⊃ λ1<br />

it’s easily checked that ψ ∈ i0. What we want to show is that, for all η ∈<br />

Fun(Pfin(Ω), R), ϕ · η = ψ. But this is easily seen: for all η ∈ Fun(Pfin(Ω), R),<br />

the following relation holds:<br />

(ϕ · η)(λ1) = ϕ(λ1) · η(λ1) = 0 · η(λ1) = 0<br />

while, on the other h<strong>and</strong>, ψ(λ1) = 1.<br />

To see that i0 is not prime, we will find ϕ <strong>and</strong> ψ ∈ Fun(Pfin(Ω), R) such as<br />

ϕ, ψ ∈ i0, but ϕ · ψ ∈ i0. In order to achieve this goal, we need to construct a strictly<br />

increasing sequence of subsets in Pfin(Ω).<br />

First of all, let’s choose λ0 ∈ Pfin(Ω). It’s easy to see that the set A0 =<br />

{λ ∈ Pfin(Ω) : λ ⊃ λ0} is nonempty, so we can choose λ1 ∈ A0. In the same spirit,<br />

29


CHAPTER 1. Ω-CALCULUS<br />

we define A1 = {λ ∈ Pfin(Ω) : λ ⊃ λ1} <strong>and</strong>, once again, A1 = ∅. As we did before,<br />

we can pick an arbitrary λ2 ∈ A2: proceding this way, we obtain a sequence {λi} i∈N<br />

such as λi ⊂ λj whenever i < j.<br />

Now, we define ϕ <strong>and</strong> ψ in the following ways:<br />

<br />

1 if λ ∈ {λi}<br />

ϕ(λ) =<br />

i∈N<br />

0 if λ ∈ {λi} i∈N<br />

<strong>and</strong><br />

<br />

0 if λ ∈ {λi}<br />

ψ(λ) =<br />

i∈N<br />

1 if λ ∈ {λi} i∈N<br />

It’s clear that both ϕ <strong>and</strong> ψ are not eventually equal to zero, but their product is<br />

c0, <strong>and</strong> c0 ∈ i0.<br />

It’s a straightforward application of Zorn’s Lemma (an equivalent formulation of<br />

the Axiom of Choice) to see that<br />

Proposition 1.9.7. If i ⊂ A is an ideal of A, there exists a maximal ideal of A<br />

containing i.<br />

The proof, that requires Zorn’s Lemma, an equivalent formulation of the Axiom<br />

of Choice, can be found on every algebra textbook (once again we rem<strong>and</strong> to [8] or<br />

to Theorem 11.17 of [9]).<br />

Proposition 1.9.7 ensures that there is a maximal ideal m ⊃ i0. Moreover, this<br />

ideal is not principal.<br />

Proposition 1.9.8. If m is a maximal ideal containing i0, then m is not principal.<br />

Proof. Suppose m is principal, so that there is ϕ ∈ m satisfying m = (ϕ). Since m<br />

is an ideal, m = Fun(Pfin(Ω), R) <strong>and</strong> so ϕ is not invertible. This implies that there<br />

is λ0 ∈ Pfin(Ω) such as ϕ(λ0) = 0.<br />

If we define a function<br />

<br />

1 if λ ⊆ λ0<br />

ψ(λ) =<br />

0 if λ ⊃ λ0<br />

we have clearly ψ ∈ i0. Since for all η ∈ Fun(Pfin(Ω), R), (ϕ · η)(λ0) = 0, we<br />

can conclude that ψ = ϕ · η for all η ∈ Fun(Pfin(Ω), R), but this contradicts the<br />

hypothesis that m is principal. We conclude that m is not principal.<br />

Combining Propositions 1.9.7 <strong>and</strong> 1.9.8, we can deduce the existence of a nonprincipal<br />

maximal ideal in the ring Fun(Pfin(Ω), R). Moreover, this ideal can be<br />

characterized as a maximal ideal extending i0. This allows us to show that, in order<br />

to have a model in which Axioms 1, 2 <strong>and</strong> 3 of Ω-Calculus hold, it’s sufficient to<br />

quotient Fun(Pfin(Ω), R) <strong>with</strong> a nonprincipal maximal ideal.<br />

Theorem 1.9.9. Let m be a nonprincipal maximal ideal of the ring Fun(Pfin(Ω), R)<br />

<strong>and</strong> let π : Fun(Pfin(Ω), R) → Fun(Pfin(Ω), R)/m the canonical projection<br />

π(ϕ) = [ϕ]m<br />

where we agree that cosets [cr]m of constant sequences are identified <strong>with</strong> the corresponding<br />

real numbers. Then, Axioms 1, 2 <strong>and</strong> 3 of Ω-Calculus are satisfied in<br />

Fun(Pfin(Ω), R)/m.<br />

30


1.10. FILTERS AND ULTRAFILTERS<br />

Proof. First of all, we observe that<br />

R ∗ <br />

<br />

= lim ϕ(λ) : λ : Pfin(Ω) → R<br />

λ↑Ω<br />

= Fun(Pfin(Ω), R)/m<br />

<strong>and</strong> Fun(Pfin(Ω), R)/m is a field because m is maximal. This is enough to satisfy<br />

Axiom 1.<br />

Axiom 2 is verified, since we identified limλ↑Ω cr = [cr]m <strong>with</strong> r ∈ R.<br />

By construction, sums <strong>and</strong> products in Fun(Pfin(Ω), R)/m commute <strong>with</strong> cosets,<br />

so that Axiom 3 is verified.<br />

1.10 Filters <strong>and</strong> ultrafilters<br />

Constructing a model in which Numerosity Axiom 1 holds is a little trickier. We<br />

can’t achieve this resulyt by using maximal ideals alone, but we’ll need to use another<br />

algebraic tool, namely filters <strong>and</strong> ultrafilters.<br />

Definition 1.10.1. A filter F on a set A is a nonempty family of subsets of A<br />

satisfying<br />

1. ∅ ∈ F<br />

2. if X ∈ F <strong>and</strong> Y ⊇ X, then Y ∈ F<br />

3. if X, Y ∈ F, then X ∩ Y ∈ F<br />

Remark 1.10.2. We want to observe that the notion of a filter F on a set A can<br />

be reinterpreted <strong>with</strong> the language of algebra. Consider the set P(A) <strong>with</strong> the<br />

two operations ∩ <strong>and</strong> ∪: we can think of ∩ as the “sum” in P(A) <strong>and</strong> ∪ as the<br />

“product”. With this interpretation, we have that A is the identity element of ∩<br />

<strong>and</strong> the absorbing element of ∪ (like 0 ∈ Z is the identity element of the sum <strong>and</strong><br />

the absorbing element of product) <strong>and</strong> ∅ is the identity element of ∪. It should be<br />

clear that P(A), endowed <strong>with</strong> the operations ∩ <strong>and</strong> ∪, becomes a commutative<br />

ring. Whenever we will talk about this ring, we will denote by 0A <strong>and</strong> 1A the sets<br />

A <strong>and</strong> ∅ respectively.<br />

A filter F on A can then be interpreted as an ideal of the ring P(A): condition 1<br />

of Definition 1.10.1 says that 1A ∈ F, condition 2 says that AF = F <strong>and</strong> condition 3<br />

says that F is closed under sums <strong>and</strong> hence a subgroup of the additive group P(A).<br />

Definition 1.10.3. We say that a family of sets F has the finite intersection property<br />

(FIP) if<br />

k<br />

∀X1 . . . Xk ∈ F, Xk = ∅<br />

With the interpretation of Remark 1.10.2, we say that F has the FIP if the<br />

additive subgroup of A generated by F does not contain 1A (hence is different from<br />

the whole P(A)).<br />

i=1<br />

31


CHAPTER 1. Ω-CALCULUS<br />

Definition 1.10.4. If G has the FIP, then<br />

<br />

〈G〉 =<br />

is the filter generated by G.<br />

Y : ∃X1, . . . , Xk ∈ G so that Y ⊇<br />

The name “filter generated by G” is justified by the fact that 〈G〉 is indeed a<br />

filter, <strong>and</strong> it’s the smallest filter containing G. We won’t give the easy proof of this<br />

statement.<br />

Example 1.10.5. For all X ⊆ A, FX = {Y ⊆ A : Y ⊇ X} is a filter. The filters of<br />

the form FX <strong>with</strong> X ⊆ A are called principal filters. It’s clear that all the principal<br />

filters have the FIP.<br />

With the interpretation of Remark 1.10.2, a principal filter FX corresponds to<br />

the (principal) ideal generated by X.<br />

It’s important not to think all filters are principal. In fact there is one example<br />

of an important nonprincipal filter: the Frechet filter.<br />

Definition 1.10.6. The Frechet filter Fr(A) over a set A is the family of cofinite<br />

subsets of A:<br />

Fr(A) = {X ⊆ A : X c is finite}<br />

This filter will play an important role in the construction of models for Ω-<br />

Calculus.<br />

In analogy to the case of ideals, we can define filters that are maximal <strong>with</strong> respect<br />

to inclusion. This condition turns out to be equivalent to other two characterizations.<br />

Proposition 1.10.7. Let F be a filter over a set A. The following properties are<br />

equivalent:<br />

1. X ∈ F ⇒ X c ∈ F;<br />

2. k<br />

i=1 Xk ∈ F ⇒ ∃i : Xi ∈ F;<br />

3. F is a maximal filter on A <strong>with</strong> respect to inclusion.<br />

Proof. All the proofs are by reductio ad absurdum.<br />

To prove that condition 1 implies condition 2, suppose that condition 1 holds <strong>and</strong><br />

that k i=1 Xk ∈ F, but Xi ∈ F for all i = 1, . . . , k. Then, by hypothesis, Xc i ∈ F<br />

for all i = 1, . . . , k; <strong>and</strong> the same holds for the intersection<br />

k<br />

X c c k<br />

i =<br />

∈ F<br />

i=1<br />

i=1<br />

Since we are assuming that k<br />

i=1 Xk ∈ F, this is a contradiction: otherwise, we<br />

would deduce that ∅ ∈ F, <strong>and</strong> conclude that F is not a filter.<br />

Now, we will prove that condition 2 implies condition 3. If F was not maximal,<br />

then there would be a filter F ′ that properly includes it. Now, pick any X ∈ F ′ \ F:<br />

32<br />

Xk<br />

k<br />

i=1<br />

Xk


1.10. FILTERS AND ULTRAFILTERS<br />

clearly, X c ∈ F ′ <strong>and</strong>, as a consequence, X c ∈ F. We conclude that A = X ∪X c ∈ F<br />

is the union of two sets neither belonging to F, <strong>and</strong> this is against the hypothesis.<br />

In order to prove that condition 3 implies condition 1, suppose that, for some<br />

set X, both X <strong>and</strong> X c ∈ F. We want to show that the family<br />

G = {Y ∩ X : Y ∈ F}<br />

has the FIP. Notice that, by definition Y ∩ X = ∅ for all Y ∈ F: otherwise, we<br />

would deduce Y ⊆ X c , <strong>and</strong> this would imply X c ∈ F against the hypothesis. Now,<br />

suppose that (Y1 ∩X), . . . , (Yk ∩X) ∈ G: then, Y1, . . . , Yk ∈ F, <strong>and</strong> also k<br />

i=1 Yi ∈ F.<br />

Since<br />

(Y1 ∩ X) ∩ . . . ∩ (Yk ∩ X) =<br />

k<br />

we deduce that G has the FIP. Now, if we consider the filter F ′ = 〈G〉, we claim<br />

that F ′ properly includes F. In fact, for every Y ∈ F, Y ⊇ Y ∩ X, <strong>and</strong> so Y ∈ F ′ .<br />

Moreover, X ∈ F but X = A∩X ∈ F ′ . This is in contradiction <strong>with</strong> the maximality<br />

of F.<br />

Definition 1.10.8. A filter satisfying one (hence all) of the properties itemized in<br />

Proposition 1.10.7 is called an ultrafilter.<br />

We want to explicitely observe that, if A is infinite, the Frechet filter Fr(A) is<br />

not an ultrafilter. In fact, we can easily find a set X ⊆ A such that both X <strong>and</strong> X c<br />

are infinite. It’s clear that X ∪ X c = A ∈ Fr(A), while both X, X c ∈ Fr(A).<br />

Now, we want to give a characterization of principal ultrafilters.<br />

Proposition 1.10.9. A principal filter FX is an ultrafilter if <strong>and</strong> only if X = {a}<br />

is a singleton.<br />

Proof. If X contains at least two elements, we can partition X = X1 ∪ X2 into two<br />

disjoint nonempty set. By the property characterizing ultrafilters, we would have<br />

that either X1 ∈ FX, or X2 ∈ FX. Both cases lead to a contradiction, because X1<br />

<strong>and</strong> X2 are proper subsets of X, while FX only contains supersets of X. The reverse<br />

implication is trivial, because<br />

i=1<br />

Yk<br />

<br />

∩ X<br />

X ∈ F{a} ⇔ a ∈ X ⇔ a ∈ X c ⇔ X c ∈ F{a}<br />

We will call Ua the principal ultrafilters generated by {a} <strong>and</strong>, <strong>with</strong> abuse of<br />

notation, we will say that Ua is the ultrafilter generated by a.<br />

Now, we proceed by giving a characterization of ultrafilters over finite sets.<br />

Proposition 1.10.10. All ultrafilters on finite sets are principal.<br />

Proof. Let U be an ultrafilter on a finite set A. Then A = {a1}∪. . .∪{ak} ∈ U <strong>and</strong>,<br />

using the property of ultrafilters, we deduce that one of the {ai} ∈ U: we know that<br />

{a1} ∪ ({a2} ∪ . . . ∪ {ak}) ∈ U, <strong>and</strong> this implies {a1} ∈ U or {a2} ∪ . . . ∪ {ak} ∈ U. In<br />

the first case, we’ve concluded. Otherwise, we can proceed in a similar way, until,<br />

by the finiteness of A, we find some {ai} ∈ U. If there was j = i such that {aj} ∈ U,<br />

this would imply {ai} ∩ {aj} = ∅ ∈ U, in contradiction <strong>with</strong> the fact that U is a<br />

filter. We conclude U = Uai .<br />

33


CHAPTER 1. Ω-CALCULUS<br />

In the same spirit, we can give a characterization of nonprincipal ultrafilters over<br />

infinite sets.<br />

Proposition 1.10.11. An ultrafilter on an infinite set A is nonprincipal if <strong>and</strong> only<br />

if it contains no finite sets, hence if <strong>and</strong> only if it contains the Frechet filter Fr(A).<br />

Proof. Let’s suppose Fr(A) ⊂ U. This implies that, for all a ∈ A, {a} c ∈ U, <strong>and</strong> so<br />

{a} ∈ U, from which we conclude that U is not principal.<br />

Now, let’s suppose that Fr(A) ⊆ U. By definition, we can find a cofinite set<br />

X ∈ U. Since U is an ultrafilter, we have that X c = {a1} ∪ . . . ∪ {ak} ∈ U <strong>and</strong> then,<br />

applying the argument of proof of Proposition 1.10.10, we conclude that {ai} ∈ U<br />

for exactly one i. We conclude that U = Uai is principal, <strong>and</strong> so the proof is<br />

concluded.<br />

It’s time to settle a result about the existence of ultrafilters. In analogy to the<br />

algebraic proof of the existence of maximal ideals, this proof is a st<strong>and</strong>ard reasoning<br />

involving Zorn’s Lemma, so we will omit some simple verifications.<br />

Proposition 1.10.12. Every filter F over a set A can be extended to an ultrafilter.<br />

Proof. Let’s consider the set<br />

G = {G filter on A : G ⊇ F}<br />

<strong>with</strong> the partial order given by inclusion.<br />

It’s straightforward to verify that every chain in G has an upper bound (just<br />

take the union of all filters in the chain <strong>and</strong> verify it’s a filter containing F). We can<br />

apply Zorn’s Lemma, that guarantees the existence of a maximal element U ∈ G.<br />

Clearly, U ⊇ F, <strong>and</strong> it’s easy to verify that U is an ultrafilter.<br />

As a consequence, we obtain the following<br />

Corollary 1.10.13. For every infinite set A, there exist a nonprincipal ultrafilters<br />

on A.<br />

Proof. Proposition 1.10.11 ensures that every ultrafilter containing the Frechet filter<br />

is nonprincipal. The existence of such ultrafilters is guaranteed by Proposition<br />

1.10.12.<br />

The importance of ultrafilters is that there’s a relation between nonprincipal<br />

maximal ideals in the ring Fun(Pfin(Ω), R) <strong>and</strong> nonprincipal ultrafilters over the<br />

set Pfin(Ω). This is a consequence of a more general fact, that we’ll show before<br />

focusing on the case of our interest. We want to warn the reader that, <strong>with</strong> abuse<br />

of notation, if ϕ ∈ Fun(A, B), we will use the notation Qϕ,c0 to denote the set<br />

{a ∈ A : ϕ(a) = 0}.<br />

Proposition 1.10.14. If Fun(A, B) <strong>with</strong> the two operations<br />

34<br />

(ϕ + ψ)(a) = ϕ(a) + ψ(a)<br />

(ϕ · ψ)(a) = ϕ(a) · ψ(a)


is a commutative ring <strong>and</strong> F is an ultrafilter on A, then<br />

mF = {ϕ ∈ Fun(A, B) : Qϕ,c0 ∈ F}<br />

is a maximal ideal of Fun(A, B).<br />

Conversely, if m is a maximal ideal of Fun(A, B), then<br />

Fm = {Qϕ,c0 : ϕ ∈ m}<br />

1.10. FILTERS AND ULTRAFILTERS<br />

is an ultrafilter on A.<br />

The two operations F ↦→ mF <strong>and</strong> m ↦→ Fm are one the inverse of the other, <strong>and</strong><br />

in this corrispondence the principal (resp. non principal) ultrafilters corresponds to<br />

the principal (resp. non principal) maximal ideals.<br />

Proof. The identities FmF = F <strong>and</strong> mFm can be easily checked using the definitions.<br />

Let m be a maximal ideal of Fun(A, B) <strong>and</strong> Fm the corresponding family of sets.<br />

Since for all ϕ, ψ ∈ Fun(A, B) we have Qϕ+ψ,c0 ⊇ Qϕ,c0 ∩ Qψ,c0, the fact that m is<br />

closed under sums implies that Fm is closed under finite intersections. In the same<br />

spirit, if ϕ ∈ m <strong>and</strong> ψ ∈ Fun(A, B), then Qϕ·ψ,c0 ⊇ Qϕ,c0, <strong>and</strong> this, coupled <strong>with</strong> the<br />

fact that Fun(A, B)m = m, implies that Fm is closed under superset. Now, choose<br />

X ⊆ A in a way that X ∈ Fm, <strong>and</strong> then let χ be its characteristic function. Clearly,<br />

χ · (1 − χ) = c0 ∈ m <strong>and</strong>, because m is prime, χ ∈ m or 1 − χ ∈ m. But the latter<br />

is false by definition, since Q1−χ,c0 = X ∈ Fm. We conclude χ ∈ m, <strong>and</strong> this implies<br />

Qχ,c0 = Xc ∈ Fm. This <strong>and</strong> the fact that c0 ∈ m ensures that ∅ ∈ Fm, <strong>and</strong> so Fm is<br />

an ultrafilter.<br />

Conversely, let F be an ultrafilter on A. To see that mF is a proper subset of<br />

Fun(A, B), consider that Qc1,c0 = ∅ <strong>and</strong>, since ∅ ∈ F, c1 ∈ mF, so mF = Fun(A, B).<br />

Now, suppose that ϕ, ψ ∈ mF: we have already seen that Qϕ+ψ,c0 ⊇ Qϕ,c0 ∩ Qψ,c0<br />

<strong>and</strong>, since F is a filter, then Qϕ,c0 ∩ Qψ,c0 ∈ F, so we can conclude ϕ + ψ ∈ mF. In<br />

the same spirit, for all ϕ ∈ mF <strong>and</strong> for all ψ ∈ Fun(A, B), we deduce Qϕ·ψ,c0 ∈ F,<br />

<strong>and</strong> this implies Fun(A, B)mF = mF, completing the proof that mF is an ideal.<br />

Now, by contradiction, suppose that mF is not maximal: then, by the first part<br />

of the proposition, we would have that FmF<br />

= F is not an ultrafilter, against the<br />

hypothesis. We can conclude that mF is a maximal ideal.<br />

It can be directly verified that, for every a ∈ A, Fm is the principal ultrafilter<br />

generated by a if <strong>and</strong> only if mF is the principal maximal ideal generated by a.<br />

As a consequence of the previous Proposition, if we take A = Pfin(Ω) <strong>and</strong> B = R,<br />

we obtain immediately the following result:<br />

Proposition 1.10.15. If F is an ultrafilter on Pfin(Ω), then<br />

mF = {ϕ ∈ Fun(Pfin(Ω), R) : Qϕ,c0 ∈ F}<br />

is a maximal ideal on Fun(Pfin(Ω), R).<br />

Conversely, if m is a maximal ideal of Fun(Pfin(Ω), R), then<br />

is an ultrafilter on Pfin(Ω).<br />

Fm = {Qϕ,c0 : ϕ ∈ m}<br />

35


CHAPTER 1. Ω-CALCULUS<br />

The two operations F ↦→ mF <strong>and</strong> m ↦→ Fm are one the inverse of the other, <strong>and</strong><br />

in this corrispondence the principal (resp. non principal) ultrafilters corresponds to<br />

the principal (resp. non principal) maximal ideals.<br />

We are interested in ultrafilters that are in correspondence <strong>with</strong> maximal ideals<br />

m ⊃ i0. We explicitely observe that those ultrafilters have an additional, interesting<br />

property: they are fine.<br />

Definition 1.10.16. A filter F on Pfin(A) is fine if, for all a ∈ A, {X ∈ Pfin(A) :<br />

{a} ⊆ X} ∈ F.<br />

This property has particular importance, since the proof of part 2 of Theorem<br />

1.9.9 depends on the identification of eventually equal functions, <strong>and</strong> this identification<br />

is possible because the nonprincipal maximal ideals extend i0. This hypothesis<br />

is essential in order to have a model in which Axiom 2 of Ω-Calculus holds.<br />

1.11 A model for Ω-Calculus<br />

Thanks to the results of the previous section, we can focus in constructing a nonprincipal<br />

fine ultrafilter on Pfin(Ω) in a way that Numerosity Axiom 1 holds. Then,<br />

we’ll use the corrispondence of Proposition 1.10.15 to obtain a nonprincipal maximal<br />

ideal that will give rise to a model of Ω-Calculus in which Numerosity Axiom 1<br />

holds.<br />

We start by constructing a particular family of sets ΛR ⊂ Pfin(R), which will be<br />

essential to define the desired ultrafilter. In order to define ΛR, we will start <strong>with</strong> a<br />

subfamily ΛQ, <strong>and</strong> then we’ll define ΛR in a way that ΛQ = ΛR ∩ P(Q).<br />

We begin by posing <br />

p<br />

<br />

λn = : p ∈ Z, |p| ≤ n2<br />

n<br />

or, more explicitely,<br />

<br />

λn = −n, − n2 − 1 1 1<br />

, . . . , − , 0,<br />

n n n , . . . , n2 <br />

− 1<br />

, n<br />

n<br />

Now, we define the family ΛQ in the following way:<br />

ΛQ = {λn : n = m!, m ∈ N}<br />

We remark that the restriction n = m! ensures that, for every λn ′, λn ′′ ∈ ΛQ, n ′ ≤ n ′′<br />

implies λn ′ ⊆ λn ′′, <strong>and</strong> so ⊆ is a total order on ΛQ.<br />

Before we define ΛR, we need to construct another family of sets: we call<br />

Θ = Pfin([0, 1] \ Q)<br />

In other words, Θ is the family of finite sets of irrational numbers between 0 <strong>and</strong> 1.<br />

Now, for all λn ∈ ΛQ <strong>and</strong> θ ∈ Θ, we set<br />

<br />

p + a<br />

λn,θ = λn ∪ :<br />

n<br />

p<br />

n ∈ λn<br />

<br />

\ {n}, a ∈ θ<br />

36


1.11. A MODEL FOR Ω-CALCULUS<br />

We can think that λn,θ is constructed in the following way: we started <strong>with</strong> the<br />

segment [−n, n] <strong>and</strong> divided it in n 2 parts of lenght 1/n, so we got a subdivision<br />

into segments of the form p/n, (p + 1)/n , where p = −n 2 , −n 2 + 1, . . . , n 2 . In each<br />

of those segments, we put a “rescaled” copy of θ. Thus, the set λn,θ contains n 2 + 1<br />

rational points <strong>and</strong> n 2 · |θ| irrational points.<br />

Finally, we set<br />

ΛR = {λn,θ : n = m!, m ∈ N, θ ∈ Θ}<br />

Now, we want to show that the numerosity axiom holds or, in a more formal<br />

way:<br />

Proposition 1.11.1. If a, b ∈ Q,<br />

lim<br />

λn,θ↑R |[a, b) ∩ λn,θ| = (b − a) · n([0, 1))<br />

Proof. Let’s take an interval [a, b) where a, b ∈ Q. If n is sufficiently large, [a, b)∩λn,θ<br />

contains n(b − a) rational numbers <strong>and</strong> n|θ|(b − a) irrational numbers, so that<br />

Then,<br />

|[a, b) ∩ λn,θ| = n(b − a)(|θ| + 1)<br />

n([a, b)) = lim<br />

λn,θ↑R |[a, b) ∩ λn,θ| = lim |n(b − a)(|θ| + 1)|<br />

λn,θ↑R<br />

= (b − a) · lim n(|θ| + 1) = (b − a) · n([0, 1))<br />

λn,θ↑R<br />

So far, we have built a family of sets ΛR in a way that, if we take the limit <strong>with</strong><br />

respect to the sets of that family, the numerosity axiom holds. Now, we want to<br />

create an ultrafilter UR over Pfin(Ω) that contains all elements of ΛR. Let’s pause<br />

for a moment <strong>and</strong> think about what properties we want UR to satisfy:<br />

1. UR must be nonprincipal <strong>and</strong> fine<br />

2. Since the function |[a, b) ∩ λ| is mapped to its limit limλ↑Ω |λ ∩ [a, b)|, we want<br />

this limit to be equal to limλn,θ↑Ω |[a, b) ∩ λn,θ|<br />

To satisfy the second condition, it is sufficient that the function |[a, b) ∩ λ| is in<br />

the same equivalence class of |[a, b) ∩ λn,θ| modulo the nonprincipal maximal ideal<br />

we want to determine. A little thought shows that it’s enough that UR contains the<br />

filter FΛR = {Λ ⊆ Pfin(Ω) : Λ ⊇ ΛR}. We want to explicitely note that, if Λ ∈ FΛR ,<br />

then Λ is infinite (because ΛR is).<br />

Putting everything together, we conclude that, in order to obtain our model, we<br />

need to find a nonprincipal fine ultrafilter containing FΛR .<br />

Before we achieve this result, we will prove that the family FΛR ∪ Fr(Pfin(Ω))<br />

has the FIP.<br />

Lemma 1.11.2. FΛR ∪ Fr(Pfin(Ω)) has the FIP.<br />

37


CHAPTER 1. Ω-CALCULUS<br />

Proof. Since both FΛR <strong>and</strong> Fr(Pfin(Ω)) have the FIP, it’s sufficient to prove that,<br />

for all Λ ∈ FΛR <strong>and</strong> for all X ∈ Fr(Pfin(Ω)), Λ ∩ X = ∅. But this is readily seen:<br />

since X c is finite <strong>and</strong> Λ ⊇ ΛR is infinite, Λ ⊆ X c , hence Λ ∩ X = ∅.<br />

As a consequence of this Lemma, we deduce that 〈FΛR ∪ Fr(Pfin(Ω))〉 is a well<br />

defined filter. This fact will be useful during the proof of the next Proposition.<br />

Proposition 1.11.3. There exists a nonprincipal fine ultrafilter on Pfin(Ω) that<br />

extends FΛR .<br />

Proof. We will modify the proof of 1.10.12 to obtain the desired ultrafilter.<br />

Let’s consider the set<br />

G = {G filter on Pfin(Ω) : G ⊇ FΛR<br />

<strong>and</strong> G is fine <strong>and</strong> contains no finite set}<br />

<strong>with</strong> the partial order given by inclusion. If we prove that G = ∅, then the rest<br />

of the proof is a straightforward application of Zorn’s Lemma. But, thanks to<br />

Lemma 1.11.2, we can immediately see that G contains the filter generated by FΛR ∪<br />

Fr(Pfin(Ω)), so G = ∅.<br />

Once again, we leave to the reader the checkings that every chain in G has an<br />

upper bound (as in the proof of Proposition 1.10.12, take the union of all filters in<br />

the chain <strong>and</strong> verify it’s a fine filter that contains no finite sets <strong>and</strong> extends FΛR ).<br />

We can apply Zorn’s Lemma <strong>and</strong> obtain the ultrafilter UR ∈ G.<br />

Observe that, since UR ∈ G, UR is fine <strong>and</strong> contains no finite set. By Proposition<br />

1.10.11, this <strong>and</strong> the fact that Pfin(Ω) is infinite imply that UR is nonprincipal.<br />

Finally, we can apply Proposition 1.10.15 to UR <strong>and</strong> find the corresponding nonprincipal<br />

maximal ideal mR. Now, we will prove that Fun(Pfin(Ω), R)/mR is a model<br />

for Ω-Calculus.<br />

Theorem 1.11.4. Let mR be the nonprincipal maximal ideal defined above <strong>and</strong> let<br />

π : Fun(Pfin(Ω), R) → Fun(Pfin(Ω), R)/mR the canonical projection<br />

π(ϕ) = [ϕ]mR<br />

where we agree that cosets [cr]mR of constant sequences are identified <strong>with</strong> the corresponding<br />

real numbers. Then, all the axioms of Ω-Calculus are satisfied.<br />

Proof. Theorem 1.9.9 ensures that Axioms 1 to 3 are verified.<br />

Propositions 1.11.1 <strong>and</strong> 1.11.3 ensure that the function |[a, b) ∩ λ| is mapped to<br />

(b − a) if a, b ∈ Q, <strong>and</strong> so Numerosity Axiom 1 holds as well.<br />

This shows that Ω-Calculus has a model.<br />

Before we move on, we want to observe that this process of model building is<br />

general <strong>and</strong> can be applied to other situations. If we want to develop a theory based<br />

on the analogue of Axioms 1, 2 <strong>and</strong> 3, we can build a model in a manner similar<br />

to the one shown in this section. Moreover, we can even change domain D <strong>and</strong><br />

codomain C for our functions, as long as Fun(D, C) is a commutative ring.<br />

We conclude by anticipating that, in Section 4.5, we will generalize further this<br />

model building method to give a model for an extension of the theory presented in<br />

this chapter.<br />

38


1.12. REASONING IN THE MODEL<br />

Remark 1.11.5. We want to note explicitely that, in this model we’ve built, the<br />

family of qualified sets Q of Definition 1.4.3 is nothing but our ultrafilter UR. What<br />

we’ve said in Remark 1.4.5 correspond precisely to the idea that the Ω-limit of<br />

functions is defined modulo being equal in a qualified set.<br />

This example suggests a generalized notion of limit, where we consider an arbitrary<br />

filter over a set I. Precisely, suppose that {xi}i∈I is an I-sequence of elements<br />

in a given topological space X, <strong>and</strong> F is a filter on I. We can define the F -limit of<br />

{xi}i∈I by setting:<br />

F − lim xi = x ⇔ {i ∈ I : xi ∈ A} ∈ F for all neighborhoods A of x<br />

It turns out that, <strong>with</strong> some hypothesis over X, the notion of F -limit has some<br />

“good” properties. Since the study of this generalization of the concept of limit is<br />

beyond the scope of this paper, for further references we rem<strong>and</strong> to Chapter 7 of [1].<br />

1.12 Reasoning in the model<br />

The existence of a model for Ω-Calculus has a number of interesting consequences<br />

that go way beyond the consistency of the theory. In particular, having explicitely<br />

determined a family of “qualified” sets, namely the family ΛR ⊂ UR, allows us to<br />

explicitely calculate some Ω-limits that were otherwise impossible to determine. As a<br />

consequence, by reasoning in the model of Ω-Calculus, we can show some interesting<br />

results whose validity depend on the ultrafilter UR.<br />

We begin by showing some properties of the number α.<br />

Proposition 1.12.1. In our model, α = max(N ◦ ). Moreover, for all k ∈ N, α is a<br />

multiple of k.<br />

Proof. We recall that α = max(R ◦ ) <strong>and</strong>, in our model,<br />

α = max(R ◦ ) = lim<br />

λ↑Ω max(λn,θ) = lim<br />

λ↑Ω n!<br />

Since, by definition, n! ∈ N, we conclude that α ∈ N ◦ . Moreover, if we pick k ∈ N,<br />

we have that n! is eventually multiple of k, <strong>and</strong> the second part of the thesis follows<br />

by applying Transfer Principle.<br />

This property has some interesting consequences. First of all, it shows that, by<br />

an appropriate choiche of the model of Ω-Calculus <strong>and</strong> thanks to Transfer Principle,<br />

we can find a model where α satisfies any finite number of elementary properties. If<br />

we needed, for instance, that n√ α ∈ N ∗ for all n ∈ N, we would have posed<br />

ΛQ = λn : n = m k , m, k ∈ N <br />

Notice that this new definition of ΛQ would have determined a different choice of<br />

the ultrafilter UR.<br />

A second consequence of Proposition 1.12.1 is that there is a relation between α<br />

<strong>and</strong> n(N), as shown by the following example.<br />

39


CHAPTER 1. Ω-CALCULUS<br />

Example 1.12.2. We want to calculate n(N). To do so, we use that, in our model,<br />

limλ↑N |λ| = limλn,θ↑N |λn,θ|. Since<br />

<strong>and</strong><br />

we deduce that, by definition of α,<br />

λn,θ ∩ N = {1, . . . , n}<br />

n = |λn,θ ∩ N| = max(λn,θ)<br />

n(N) = lim<br />

λ↑N |λ| = lim<br />

λn,θ↑R max(λn,θ) = α (1.4)<br />

Remark 1.12.3. Equation 1.4 shows us another interesting property of the number<br />

α:<br />

α = lim<br />

λ↑R max(λ) = lim<br />

λ↑N |λ|<br />

We explicitely remark that this equality is true in the model, but is not entailed by<br />

any axiom of Ω-Calculus. In other words: this equality is generally false, unless we<br />

require that it holds by adding an appropriate axiom or by choosing a particular<br />

model of Ω-Calculus.<br />

We continue by calculating the numerosities of the sets E <strong>and</strong> O of even <strong>and</strong> odd<br />

numbers, <strong>and</strong> those of Z <strong>and</strong> Q.<br />

Example 1.12.4. We have already observed that, since N = E ∪ O, then, by finite<br />

additivity of numerosity, n(N) = n(E) + n(O). Now, we will determine n(E) <strong>and</strong><br />

n(O). As in the previous example, n(E) = limλn,θ↑E |λn,θ|. If n > 1, then<br />

|λn,θ ∩ E| = {2, . . . , n}<br />

from which we deduce |λn,θ ∩ E| = n/2 is eventually true. We conclude n(E) = α/2<br />

<strong>and</strong> n(O) = n(N) − n(E) = α/2.<br />

Example 1.12.5. We want to calculate n(Z). Since Z = N ∪ −N ∪ {0}, by using<br />

Corollary 1.6.3, we conclude n(Z) = 2n(N) + 1 = 2α + 1.<br />

Example 1.12.6. We want to calculate n(Q). As in Example 1.12.2, we have<br />

limλ↑Q |λ| = limλn,θ↑R |λn,θ ∩ Q|. We recall that, by definition,<br />

<br />

λn,θ ∩ Q = λn = −n, −n2 + 1<br />

, . . . , −<br />

n<br />

1 1<br />

, 0,<br />

n n , . . . , n2 <br />

− 1<br />

, n<br />

n<br />

so |λn| = 2n 2 + 1. Taking the Ω-limit, we obtain n(Q) = 2α 2 + 1.<br />

In our model, we can also answer the open question of Example 1.8.8. In particular,<br />

we will show that min((0, 1) ◦ ) < 1/α.<br />

Proposition 1.12.7. min((0, 1) ◦ ) < 1/α<br />

40


1.12. REASONING IN THE MODEL<br />

Proof. By the definition of the ultrafilter UR, we have that, if we define xn,θ =<br />

min(λn,θ ∩ (0, 1)), then xn,θ is an irrational number satisfying the inequalities 0 <<br />

xn,θ < 1/n, where n = m! ∈ N. By Transfer of inequalities, the chain of inequalities<br />

holds for the Ω-limits<br />

0 < lim xn,θ < lim<br />

λ↑Ω λ↑Ω<br />

from which we immediately conclude.<br />

1<br />

max(λn,θ)<br />

= 1<br />

α<br />

Thanks to this result, we can go back to Example 1.8.8 <strong>and</strong> conclude the reasoning.<br />

Since f|(0,1) is strictly decreasing, then so is f ◦ by Proposition 1.8.5. Thanks<br />

to Proposition 1.12.7, we know that β = min((0, 1) ◦ ) < 1/α, so we can conclude<br />

f ◦ (1/β) > f ◦ (1/α) = α 2 . Since α 2 ∈ (1, +∞) ◦ , we deduce that also f ◦ (β) ∈<br />

(1, +∞) ◦ .<br />

We explicitely remark that, had we chosen another ultrafilter, some (or maybe<br />

all!) of these results could have been different. From now on, we will choose this<br />

model for Ω-Calculus <strong>and</strong> we will feel free to use some of its peculiar properties for<br />

further developments of the theory.<br />

Remark 1.12.8. We observe that, in the proof of Proposition 1.12.1, we wrote an<br />

expression of the form limλ↑Ω f(n). The meaning of this writing can be rendered<br />

precise in a number of ways, one of them being<br />

In our model, we could have also used<br />

lim f(n) = lim f(|λ|)<br />

λ↑Ω λ↑Ω<br />

lim f(n) = lim<br />

λ↑Ω λn,θ↑R f(max(λn,θ))<br />

as we did in the proof of Proposition 1.12.1. Observe that, in our model, max(λn,θ)<br />

is indeed a natural number <strong>and</strong>, since {λn,θ}n∈N ∈ Q, this limit is well defined. In<br />

the future, when the need arises, we will choose whatever way we will think more<br />

appropriate for the specific situation.<br />

We conclude this chapter by showing that there are some relations between<br />

Ω-Calculus <strong>and</strong> Alpha-Calculus, a similar theory that rules the “Alpha-limit” of<br />

sequences instead of more general functions. For a complete presentation of this<br />

theory, we rem<strong>and</strong> to Part 1 of [1].<br />

Remark 1.12.9. In the setting of Ω-Calculus, we can derive the axioms of Alpha-<br />

Calculus:<br />

• ACT0 <strong>and</strong> ACT1 are consequences of Axioms 1 <strong>and</strong> 2. In particular, real<br />

sequences {xn}n∈N can be interpreted as functions f : N → R <strong>and</strong>, by extension<br />

of functions, we can think of their Alpha-limit as the evaluation of f ∗ at some<br />

infinite point in N ∗ . A good, somehow “natural” c<strong>and</strong>idate is max(N ◦ ) that,<br />

in our model, coincides <strong>with</strong> the number α = max(R ◦ ).<br />

41


CHAPTER 1. Ω-CALCULUS<br />

• It can be easily seen that, if we consider the identity function ι : N → N, then<br />

the number ι ◦ (max(N ◦ )) is indeed a “new” number in the sense of ACT2.<br />

• Finally, ACT3 holds because of Axiom 3.<br />

We can conclude that Ω-Calculus is an “extension” of Alpha <strong>Theory</strong>: we have all<br />

the tools from Alpha Calculus, plus some original notions <strong>and</strong> instruments.<br />

42


Chapter 2<br />

Selected topics of Ω-Calculus<br />

Decisive progress in science often<br />

depends on a change of direction.<br />

Kurt Gödel<br />

Since the main purpose of Ω-Calculus is to provide a theory strong enough<br />

to allow the development of calculus, we find opportune to show how some<br />

selected topics of calculus can be studied inside this theory. In the first<br />

section of this chapter, we will see how the important notions of continuity <strong>and</strong> convergence<br />

can be given by Ω-Calculus <strong>and</strong>, in the second section, we will characterize<br />

the usual topology of the real line by the use of hyperfinite extensions of sets. This<br />

characterization will then be used in the proof of some famous theorems of calculus<br />

by the means of Ω-Calculus.<br />

Our attention won’t be excessively focused on how “classical” mathematics can<br />

be developed <strong>with</strong>out the concept of classical limit <strong>and</strong> by using infinite <strong>and</strong> infinitesimal<br />

numbers, since this is the realm of non st<strong>and</strong>ard analysis. Instead, our<br />

goal is to show how some topics of “classical” mathematics can be interpreted by<br />

Ω-Calculus, <strong>with</strong> a particular interest about how the use of Ω-limit <strong>and</strong> other instruments<br />

presented in the previous Chapter can be used to prove some well known<br />

theorems of calculus.<br />

Of course, since there isn’t only one way to do mathematics, the reader could<br />

wonder why do we think this approach is more interesting than the classical one.<br />

The answer is rather simple: Ω-Calculus combines the intuitive aspects of nonst<strong>and</strong>ard<br />

analysis <strong>with</strong> other tools (namely, the Ω-limit <strong>and</strong> the hyperfinite extension<br />

of sets) that, in our opinion, simplify a lot the complexity of the proofs. Clearly,<br />

this simplification comes <strong>with</strong> a price: the necessity of at least basics knowledge of<br />

Ω-Calculus. We will not discuss further wether the benefits exceed the downsides,<br />

<strong>and</strong> will leave the reader <strong>with</strong> this open question.<br />

In the course of this chapter, we will assume that the reader is familiar <strong>with</strong><br />

the st<strong>and</strong>ard wiewpoint of the topics treated. In particular, we will assume that<br />

the reader knows the fundamental results of st<strong>and</strong>ard analysis regarding sequences,<br />

series, limits, continuity <strong>and</strong> basic topology of the real line. We will often use [3] as<br />

reference.<br />

43


CHAPTER 2. SELECTED TOPICS OF Ω-CALCULUS<br />

2.1 Convergence <strong>and</strong> continuity<br />

Calculus has its limits.<br />

Anonymous<br />

The most powerful <strong>and</strong> characterizing tool of st<strong>and</strong>ard analysis is <strong>with</strong>out any<br />

doubt the notion of limit, <strong>and</strong> in particular that of limit of sequences <strong>and</strong> functions.<br />

In this section, we will approach the idea of “classical limit” from the viewpoint<br />

of Ω-Calculus. More precisely, we will study how the concept of convergence can<br />

be developed <strong>with</strong> the instruments of Ω-Calculus. We begin <strong>with</strong> the study of<br />

convergence in real sequences, <strong>and</strong> then we move on to the case of real, real valued<br />

functions.<br />

To fix the notation, we start <strong>with</strong> two definitions, that should be familiar to the<br />

readers.<br />

Definition 2.1.1. A real sequence is a function f : N → R. Sometimes we will<br />

denote <strong>with</strong> {xi}i∈N the sequence f defined by f(i) = xi.<br />

Definition 2.1.2. Let {xi}i∈N be a sequence <strong>and</strong> y ∈ R. The sequence converges to<br />

y if<br />

∀ɛ > 0 ∃m : ∀n ≥ m |xn − y| ≤ ɛ<br />

In the context of Ω-Calculus, we want to introduce a new definition of convergence<br />

consistent <strong>with</strong> Definition 2.1.2, that doesn’t use the idea of classical limit.<br />

The basic idea is to check the behaviour of the sequence f “near infinity”. Since a<br />

sequence is a function from N to R, we can consider its hyperfinite extension<br />

f ◦ : N ◦ → R ∗<br />

The intuitive idea of checking the behaviour of f near infiniy can be then translated<br />

to evaluating the function f ◦ at some infinite hyperreal point. A somehow “natural”<br />

c<strong>and</strong>idate is the number max(N ◦ ) that, in our model, coincides <strong>with</strong> the number α,<br />

<strong>and</strong> we explicitely observe that we will feel free to use this property in the next<br />

definitions. Taking this into account, the first attempt to formalize the vague idea<br />

of convergence expressed above gives rise to the following definition:<br />

Definition 2.1.3. Let f be a sequence <strong>and</strong> ξ ∈ R ∗ a finite number. The sequence<br />

converges to ξ if<br />

f ◦ (α) ∼ ξ<br />

The idea of this definition is to see the value that f ◦ assumes when calculated<br />

in α, an infinite number. Of course, we need now to check if this definition is well<br />

posed <strong>and</strong> if it is “a good definition of convergence”, in the sense that it has all the<br />

properties we expect convergence to have. We begin to address this problem <strong>with</strong><br />

an example.<br />

44


Example 2.1.4. Let’s consider the sequence<br />

f(i) =<br />

2.1. CONVERGENCE AND CONTINUITY<br />

1 if i ∈ E<br />

0 if i ∈ O<br />

In the st<strong>and</strong>ard sense, this function doesn’t converge. In the sense of definition 2.1.3,<br />

the function converges to 1 or to 0.<br />

A definition of convergence so different than the st<strong>and</strong>ard one is not so useful: we<br />

want the new definition to agree <strong>with</strong> the old one in deciding if a function converges<br />

or doesn’t converge. Let’s try again: the idea is still the same, to check the behaviour<br />

of f ◦ when calculated in an infinite number, but this time we want to avoid situations<br />

like the one of example 2.1.4. To do so, we will check the behaviour of f ◦ at all<br />

infinite points ζ ∈ N ◦ .<br />

Definition 2.1.5. Let f be a sequence <strong>and</strong> ξ ∈ R ∗ a finite number. The sequence<br />

converges to ξ if<br />

f ◦ (ζ) ∼ ξ ∀ζ ∈ N ◦ : sh(ζ) = +∞<br />

What happens if we consider again the sequence f of example 2.1.4? We can<br />

find a function ϕ : Pfin(Ω) → N such as ϕ(λ) ∈ E for all λ ∈ Pfin(Ω) <strong>and</strong>, if<br />

ξ = limλ↑Ω ϕ(λ), sh(ξ) = +∞. Analogously, we can find a function ψ : Pfin(Ω) → N<br />

such as ψ(λ) ∈ O for all λ ∈ Pfin(Ω) <strong>and</strong>, if ζ = limλ↑Ω ψ(λ), sh(ζ) = +∞. From<br />

the definition, we have: ξ <strong>and</strong> ζ ∈ N ◦ ,<br />

<strong>and</strong><br />

f(ϕ(λ)) = 1 ∀λ ∈ Pfin(Ω)<br />

f(ψ(λ)) = 0 ∀λ ∈ Pfin(Ω)<br />

Now, Axiom 2 ensures that f(ξ) = 1 <strong>and</strong> f(ζ) = 0. According to definition 2.1.5,<br />

f doesn’t converge neither to 0 nor to 1, so we can safely conclude that f doesn’t<br />

converge. With a similar argument, we can show that if a sequence doesn’t converge<br />

in the sense of definition 2.1.2, it doesn’t converge in the sense of definition 2.1.5.<br />

Now, let’s check what happens to convergent sequences.<br />

Example 2.1.6. Let’s consider the sequence f(i) = 1/i. We know that this sequence<br />

converges to 0 in the st<strong>and</strong>ard sense. To see what happens <strong>with</strong> the new definition,<br />

let’s take ϕ <strong>and</strong> ψ : Pfin(Ω) → N so that ξ = limλ↑Ω ϕ(λ), ζ = limλ↑Ω ψ(λ) <strong>and</strong><br />

sh(ξ) = sh(ζ) = +∞. From the definition of f, we have that both f(ξ) = 1/ξ <strong>and</strong><br />

f(ζ) = 1/ζ are infinitesimal numbers. Their difference is infinitesimal, too.<br />

What can we conclude from the previous example? Simply: if f converges to ξ<br />

according to Definition 2.1.5, then f converges to every ζ ∈ mon(ξ).<br />

We now have all elements to formulate a definitive, “good” definition of convergent<br />

sequence, that we will complete <strong>with</strong> the concepts of diverging sequence <strong>and</strong><br />

not converging sequence.<br />

45


CHAPTER 2. SELECTED TOPICS OF Ω-CALCULUS<br />

Definition 2.1.7. Let f be a sequence <strong>and</strong> y ∈ R. We will say that the sequence f<br />

converges to y if<br />

f ◦ (ζ) ∈ mon(y) ∀ζ ∈ N ◦ : sh(ζ) = +∞<br />

We will say that f diverges positively if<br />

sh (f ◦ (ζ)) = +∞ ∀ζ ∈ N ◦ : sh(ζ) = +∞<br />

<strong>and</strong> we will say that f diverges negatively if −f diverges positively. If f behaves<br />

differently, we will say that f does not converge.<br />

Now, all the aberrant behaviours are out of the way, <strong>and</strong> we can work <strong>with</strong> this<br />

final definition. We have chosen to show its genesis to give the reader an example of<br />

the process behind the formation of a new mathematical concept: first, we started<br />

<strong>with</strong> an intuitive, imperfect idea, which we improved during the development of the<br />

theory. This process was cleverly noticed <strong>and</strong> clearly exposed for the first time by<br />

Imre Lakatos in [16].<br />

Before we move on, we want to remark that, thanks to Definition 2.1.7, if f<br />

converges, there aren’t any doubts that f converges to a unique number: this will<br />

be the only real number y satisfying the condition<br />

f ◦ (ζ) ∈ mon(y) ∀ζ ∈ N ◦ : sh(ζ) = +∞<br />

For the sake of completeness, we observe explicitely that, if there are two distinct<br />

real numbers y1 <strong>and</strong> y2 such as<br />

<strong>and</strong><br />

f ◦ (ζ) ∈ mon(y1) for some ζ ∈ N ◦ : sh(ζ) = +∞<br />

f ◦ (ξ) ∈ mon(y2) for some other ξ ∈ N ◦ : sh(ξ) = +∞<br />

then, by definition, f wouldn’t converge neither to y1 nor to y2. Moreover, if f <strong>and</strong><br />

g are sequences, Definition 2.1.7 <strong>and</strong> the axioms of Ω-Calculus ensure also that, if f<br />

converges to y <strong>and</strong> g converges to z, then the sequence f + g converges to y + z, <strong>and</strong><br />

the same holds for the other algebraic operations. We recall that, in the st<strong>and</strong>ard<br />

approach to convergence given by classical limit, those are theorems one needs to<br />

prove.<br />

In the next part of this section, we want to study the case of real, real valued<br />

functions. First of all, we recall the st<strong>and</strong>ard definition of limit:<br />

Definition 2.1.8. Let A ⊆ R, x0 an accumulation point for A <strong>and</strong> f : A → R. We<br />

say that f converges to y as x approaches x0 if<br />

∀ɛ > 0 ∃δ > 0 : ∀x ∈ A, x = x0, |x − x0| ≤ δ ⇒ |f(x) − y| ≤ ɛ<br />

This definition sums up in a rigorous manner the informal idea of f(x) being<br />

“very close” to y whenever x is “very close” to x0, <strong>with</strong>out being equal to it.<br />

Let’s pause for a moment <strong>and</strong> think about what we have written. We have<br />

described informally the notion of convergence using the words “very close” but, in<br />

our theory, we have a precise definition of “infinitely close” that express the same<br />

idea formally. This allows us to translate the underlying concept of Definition 2.1.8<br />

to the language of infinitesimal numbers <strong>and</strong> Ω-Calculus.<br />

46


2.1. CONVERGENCE AND CONTINUITY<br />

Definition 2.1.9. Let A ⊆ R, x0 an accumulation point for A <strong>and</strong> f : A → R. We<br />

say that f converges to y as x approaches x0 if<br />

∀ξ ∈ mon(x0) ∩ A ◦ , ξ = x0 ⇒ sh(f ◦ (ξ)) = y<br />

We will say that f diverges positively as x approaches x0 if<br />

∀ξ ∈ mon(x0) ∩ A ◦ , ξ = x0 ⇒ sh(f ◦ (ξ)) = +∞<br />

<strong>and</strong> we will say that f diverges negatively as x approaches x0 if −f diverges positively.<br />

If f behaves differently, we will say that f doesn’t converge as x that approaches x0.<br />

We leave to the reader the easy task to extend Definition 2.1.9 to the cases where<br />

x approaches ±∞.<br />

It’s not immediate to see that the classical definition is equivalent to Definition<br />

2.1.9, but it’s easy to recognize in the latter the intuitive idea of limit. Moreover,<br />

this second definition seems easier to underst<strong>and</strong> <strong>and</strong> remember than the classical<br />

one, even by maths students. For further reference about this brief consideration, we<br />

rem<strong>and</strong> to [17], where the author studied the relationship between formal definitions<br />

of mathematical concepts <strong>and</strong> the mental representation of those concepts by a<br />

statistical population of undergraduate students in mathematics. In particular, some<br />

of her research involved the mental representation of the concept of classical limit:<br />

during some interviews <strong>with</strong> students who learned the definition of limit as stated in<br />

Definition 2.1.8, she discovered that the formal definition was forgotten <strong>and</strong> replaced<br />

<strong>with</strong> the intuitive idea “f has limit l as x approaches x0 if we get nearer <strong>and</strong> nearer<br />

to l, but we never reach it”. In the words of some student, “When the variable,<br />

the x approaches a certain point, the image of the function approaches a certain l”.<br />

It seems to us that this intuitive idea of limit is closer to Definition 2.1.9 than to<br />

Definition 2.1.8. This is the main reason of our interest in nonst<strong>and</strong>ard presentations<br />

of calculus: we are deeply fascinated by ways of do mathematics that combine the<br />

formal precision of the discipline <strong>with</strong>out putting to the test the basic intuitions that<br />

seems rooted in the way humans think.<br />

Back to the theory, we now want to show a useful result: to check if a function<br />

converges when x approaches +∞, we can check the value of f in all infinite points<br />

of R ∗ (<strong>and</strong> not only those in R ◦ ). Formally, we have:<br />

Proposition 2.1.10. Let f : R → R, y ∈ R. f converges to y as x approaches +∞<br />

if <strong>and</strong> only if for all ξ ∈ R ∗ : sh(ξ) = +∞, sh(f ∗ (ξ)) = y. The same holds if we<br />

replace y <strong>with</strong> +∞ or −∞.<br />

Proof. One of the implications is trivial. Let’s focus on the other.<br />

Suppose that there is ξ ∈ R ∗ \ R ◦ <strong>with</strong> sh(ξ) = +∞ <strong>and</strong> f ∗ (ξ) ∼ y. If we find a<br />

function ψ : Pfin(Ω) → R <strong>with</strong> ψ(λ) ∈ λ such as limλ↑Ω f(ψ(λ)) ∼ y, then the proof<br />

would be concluded.<br />

By hypothesis, ξ = limλ↑Ω ϕ(λ) for some ϕ : Pfin(Ω) → R. Without loss of<br />

generality, we can also assume that, for some n ∈ N, |f(ϕ(λ)) − y| > 1/n is true for<br />

all λ ∈ Pfin(Ω). If we define<br />

A = {x ∈ R : x = ϕ(λ) for some λ ∈ Pfin(R)}<br />

47


CHAPTER 2. SELECTED TOPICS OF Ω-CALCULUS<br />

it’s clear that A contains arbitrarily large numbers: otherwise, we would have that<br />

ξ < n for some n ∈ N, against the hypothesis. Moreover, by construction, we have<br />

At this point, define<br />

ψ(λ) =<br />

f ∗ (A ∗ ) ∩ mon(y) = ∅ (2.1)<br />

max(A ∩ λ) if A ∩ λ = ∅<br />

max(λ \ {y}) otherwise<br />

By definition, ψ(λ) ∈ λ for all λ ∈ Pfin(Ω), so that ζ = limλ↑Ω ψ(λ) ∈ R ◦ . It’s also<br />

clear that ζ is an infinite number, thanks to the fact that A has no upper bound.<br />

If we can prove that f ◦ (ζ) ∈ f ∗ (A ∗ ), then the thesis would follow by Equation<br />

2.1. But this is easily shown: since the hypothesis λ ∩ A = ∅ is eventually true, we<br />

deduce that f(ψ(λ)) ∈ f(A) is eventually true <strong>and</strong>, thanks to Proposition 1.7.2, we<br />

obtain f ◦ (ζ) ∈ [f(A)] ∗ . We conclude by part 1 of Proposition 1.8.3.<br />

In the case where f has limit ±∞, mutatis mut<strong>and</strong>is, the same proof can be<br />

applied.<br />

In the next example, we will see Definition 2.1.9 at work.<br />

Example 2.1.11. Let’s consider the function f : (0, +∞) → R defined by f(x) =<br />

sin(1/x). We want to calculate if f(x) converges for x that approaches 0. The idea<br />

is quite easy: we know that sin(x) = 1 when x = 4k+1π,<br />

k ∈ N, <strong>and</strong> sin(x) = −1<br />

2<br />

π, k ∈ N. Now, if we consider the two functions<br />

when x = 4k+3<br />

2<br />

ϕ(λ) =<br />

2<br />

(4|λ| + 1)π<br />

<strong>and</strong> ψ(λ) =<br />

2<br />

(4|λ| + 3)π<br />

we have that limλ↑Ω ϕ(λ) ∼ limλ↑Ω ψ(λ) ∼ 0, but f(ϕ(λ)) = 1 <strong>and</strong> f(ψ(λ)) = −1 for<br />

all λ ∈ Pfin(Ω). This implies<br />

1 = lim<br />

λ↑Ω f(ϕ(λ)) = lim<br />

λ↑Ω f(ψ(λ)) = −1<br />

so we can conclude that sin(1/x) doesn’t converges as x approaches 0.<br />

Now, we will give the definitions of continuity <strong>and</strong> of uniform continuity. The<br />

definitions are quite straightforward, <strong>and</strong> once again we will not prove that they are<br />

equivalent to the st<strong>and</strong>ard ones. We can see that also those definitions are quite<br />

similar to the respective intuitive ideas.<br />

Definition 2.1.12. Let A ⊆ R, x0 ∈ A <strong>and</strong> f : A → R. We say that f is continuous<br />

in x0 if<br />

∀ξ ∈ A ◦ , ξ ∼ x0 ⇒ sh(f ◦ (ξ)) = y<br />

We say that f is continuous in A if f is continuous at every point of A.<br />

We say that f is uniformly continuous in A if f is continuous in A <strong>and</strong><br />

48<br />

∀ξ, ζ ∈ A ◦ , ξ ∼ ζ ⇒ f ◦ (ξ) ∼ f ◦ (ζ)


2.2. TOPOLOGY BY Ω-CALCULUS<br />

Now, we want to provide some examples of reasonings about uniform continuity.<br />

Once again, we remark that the methods used appear to be fair simpler than the<br />

st<strong>and</strong>ard ones.<br />

Example 2.1.13. The same reasoning as in Example 2.1.11 shows that sin(1/x) is<br />

not uniformly continuous.<br />

Example 2.1.14. Let’s consider the function f : R → R defined by f(x) = x 2 . We<br />

want to show that f is not uniformly continuous. Let’s consider the two numbers<br />

α <strong>and</strong> α − 1/α: in our model, we know that both belong to R ◦ . By definition,<br />

f ◦ (α) = α 2 , while f ◦ (α − 1/α) = α 2 − 2 + 1/α 2 . Their difference is 2 − 1/α 2 , <strong>and</strong><br />

this is a finite, non infinitesimal number, so we can conclude that f is not uniformly<br />

continuous.<br />

Now, consider f|[a,b), <strong>with</strong> a <strong>and</strong> b ∈ R. This time, let’s pick ξ ∈ [a, b) ◦ <strong>and</strong> ɛ<br />

infinitesimal such that ξ − ɛ ∈ [a, b) ◦ . Let’s calculate the difference between their<br />

images:<br />

f(ξ) − f(ξ − ɛ) = ξ 2 − (ξ 2 − 2ξɛ + ɛ 2 ) = 2ξɛ − ɛ 2<br />

Since ξ is finite, the last term of the equality chain is infinitesimal. We want to<br />

explicitely remark that this result is independent of the choice of both ξ <strong>and</strong> ɛ, so we<br />

obtained a (quite straightforward) proof that f|[a,b) is uniformly continuous.<br />

Remark 2.1.15. The previous examples give us the chance to explicitely state the<br />

following, crucial consideration:<br />

Studying calculus <strong>with</strong> the tools of Ω-Calculus, we don’t need a “classical” idea of<br />

limit. We can replace the operation of classical limit <strong>with</strong> evaluation of<br />

hyperextended functions at some hyperreal points.<br />

It’s important not to understimate this difference. We think that evaluate a<br />

function at a point is way easier than fully master the “classical” definition of limit.<br />

While our definition isn’t a new <strong>and</strong> more powerful tool than the “classical” one, it<br />

has the potential to give easy access to the complex, yet intuitive idea behind the<br />

concept of “classical” limit, <strong>with</strong>out sacrificing formal rigor.<br />

2.2 Topology by Ω-Calculus<br />

Calculus required continuity, <strong>and</strong><br />

continuity was supposed to require<br />

the infinitely little; but nobody<br />

could discover what the infinitely<br />

little might be.<br />

Bertr<strong>and</strong> Russel<br />

In the previous section, we have introduced the concept of continuity for functions.<br />

The least requirement to define a notion of continuity that “makes some<br />

sense” for real, real valued functions is a topology on R. A privileged choice of<br />

49


CHAPTER 2. SELECTED TOPICS OF Ω-CALCULUS<br />

topology on R which induces the known notion of continuity is the topology induced<br />

by the distance d(x, y) = |x−y|, which we will denote (R, Td). With the instruments<br />

of Ω-Calculus, however, we have based the notion of continuity upon the notion of<br />

“two points being infinitely close”, as formally described in Definition 1.2.3. With<br />

this idea in mind, we want to give R a topology that takes into account this notion.<br />

To obtain this goal, we will follow [1] in endowing R <strong>with</strong> an uniform topology.<br />

Before we proceed, we want to remark that the notion of uniform topology is<br />

interesting on its own, because it can be generalized to arbitrary sets in which<br />

we have defined somehow a notion of “infinite closeness”, <strong>and</strong> this topology alone<br />

is sufficient to define a lot of interesting tool of calculus (e.g. various types of<br />

convergence of function sequences) <strong>and</strong> to prove <strong>with</strong> agility some basic theorems<br />

(e.g. Weierstrass theorem of maximum <strong>and</strong> minimum <strong>and</strong> Bolzano’s theorem of<br />

zeros). For an introduction on this topic, we rem<strong>and</strong> to [1].<br />

We begin <strong>with</strong> the abstract definition of a uniform topology for R, <strong>and</strong> then<br />

specialize it to our concrete case.<br />

Definition 2.2.1. A Hausdorff uniform topology for R is an equivalence relation<br />

∼τ over R ◦ , such that x ∼τ y for all x, y ∈ R. Given that, we define<br />

<strong>and</strong> the family of open <strong>and</strong> closed sets:<br />

monτ(x) = {ξ ∈ R ◦ : ξ ∼τ x}<br />

• A ⊆ R is an open set if <strong>and</strong> only if x ∈ A implies monτ(x) ∈ A ◦ ;<br />

• C ⊆ R is a closed set if <strong>and</strong> only if x ∈ C c implies monτ(x) ∩ A ◦ = ∅<br />

It’s straightforward to verify that the two families defined satisfy all the usual<br />

properties of open <strong>and</strong> closed sets, so that the definition is well posed.<br />

In our case, a good c<strong>and</strong>idate for ∼τ is the relation ∼ of being “infinitely close”,<br />

introduced in Definition 1.2.3. Of course, we will need to restrict ∼, that is defined<br />

on all R ∗ , only to the set R ◦ . With abuse of notation, we will often denote this<br />

relation by ∼ <strong>and</strong>, in the same spirit, we will call mon(x) = {ξ ∈ R ◦ : ξ ∼ x}. When<br />

the need of clarity arises, we will use the symbols ∼ ◦ <strong>and</strong> mon ◦ (x) instead.<br />

With this choice for the relation ∼, we obtain the topology defined by<br />

• A ⊆ R is an open set if <strong>and</strong> only if x ∈ A implies mon(x) ⊆ A ◦ ;<br />

• C ⊆ R is a closed set if <strong>and</strong> only if x ∈ C c implies mon(x) ∩ A ◦ = ∅<br />

We will denote this topology by (R, T ).<br />

Now, we will show that this topology is equivalent to (R, Td).<br />

Proposition 2.2.2. A ⊆ R is open (respectively, closed) in the topology (R, T ) if<br />

<strong>and</strong> only if A is open (respectively, closed) in the topology (R, Td).<br />

Proof. Let’s suppose that A is open in (R, T ): then, for all x ∈ A, mon(x) ⊆ A ◦ .<br />

We want to show that there exists some r ∈ R such that the set Br(x) = {y ∈ R :<br />

|x − y| < r} ⊆ A. Let’s suppose that this is not the case, so that, for all r ∈ R,<br />

Br(x) ⊆ A. This implies that there exists r0 ∈ R <strong>and</strong> a point z ∈ R such that<br />

50


2.2. TOPOLOGY BY Ω-CALCULUS<br />

z ∈ Br \ A for all r > r0. Without loss of generality, we can suppose z > x, so we<br />

can deduce A ∩ (x, z] = ∅. This implies that, for instance, x + 1/α ∈ A ◦ but, since<br />

x + 1/α ∈ mon(x), this is a contradiction.<br />

On the other h<strong>and</strong>, if A is open in (R, Td), then, if we choose x ∈ A, there exists<br />

r0 such that Br(x) ⊆ A for all r < r0. For every ξ ∈ mon(x), there exists a function<br />

ϕξ such that limλ↑Ω ϕξ(λ) = ξ, <strong>and</strong> this hypothesis ensures that ϕξ(λ) ∈ Br(x) is<br />

eventually true for all r ∈ R. Considering only r < r0, we get that ϕξ(λ) ∈ A is<br />

eventually true <strong>and</strong>, by the arbitrariety of ξ, this implies mon(x) ⊆ A ◦ .<br />

If A is a closed set, then use the fact that A c is an open set <strong>and</strong> the previous<br />

part of the Proposition.<br />

With this result at h<strong>and</strong>, it’s natural to wonder why have we chosed to introduce<br />

the usual topology (R, Td) in such a convoluted way. The answer will be clear after<br />

we’ve seen the proofs of some classical results stated in the language of this topology.<br />

Before we do so, we proceed by introducing some other topological concepts.<br />

Now we will define compact sets. Instead of giving the usual characterization<br />

involving finite coverings, we take another route: we begin by introducing the idea<br />

of a near-st<strong>and</strong>ard point for a set, <strong>and</strong> then we will define compact sets using nearst<strong>and</strong>ard<br />

points.<br />

Definition 2.2.3. We say that ξ ∈ A ◦ is near-st<strong>and</strong>ard in A if sh(ξ) ∈ A.<br />

Definition 2.2.4. We say that K ⊆ R is compact if <strong>and</strong> only if every ξ ∈ K ◦ is<br />

near st<strong>and</strong>ard in K.<br />

Using the characterization of compacts <strong>with</strong> respect to the topology induced by<br />

the metric in R (namely, K ⊆ R is compact if <strong>and</strong> only if K is closed <strong>and</strong> bounded),<br />

it’s easy to see that K is compact in the usual sense if <strong>and</strong> only if K is compact<br />

according to Definition 2.2.4.<br />

The concept of near-st<strong>and</strong>ard points is also useful in characterizing connected<br />

sets.<br />

Definition 2.2.5. We say that C ⊆ R is connected if <strong>and</strong> only if, for every decomposition<br />

{C1, C2} of C, there exists ξ1 ∈ (C1) ◦ <strong>and</strong> ξ2 ∈ (C2) ◦ , both near st<strong>and</strong>ard in<br />

C, such that ξ1 ∼ ξ2.<br />

In order to conclude the premises of this section, we will give the definitions of<br />

accumulation point for a set, internal point <strong>and</strong> frontier point.<br />

Definition 2.2.6. Let A ⊆ R <strong>and</strong> x ∈ A. We say that<br />

• x is an accumulation point for A if {x} ⊂ mon(x) ∩ A ◦<br />

• x is an internal point of A if mon(x) ⊆ A ◦<br />

• x is a frontier point for A if mon(x) ⊂ A but mon(x) ∩ A = ∅<br />

It’s easily seen that all definitions given agree <strong>with</strong> the usual ones, so we won’t<br />

give the easy proofs of the equivalence. Instead, we move on <strong>and</strong> show how these<br />

new definitions allow to prove some classical results <strong>with</strong> ease.<br />

51


CHAPTER 2. SELECTED TOPICS OF Ω-CALCULUS<br />

Theorem 2.2.7 (Weierstrass Theorem). If K ⊂ R is a nonempty compact set <strong>and</strong><br />

f : K → R is continuous, then f has maximum <strong>and</strong> minimum.<br />

Proof. For all λ ∈ Pfin(Ω), if λ ∩ K = ∅, f|λ has maximum <strong>and</strong> minimum. Since<br />

this hypothesis is eventually true, let xλ ∈ λ such that f(xλ) is a maximum of<br />

f|λ. Now, let’s consider ξ = limλ↑Ω xλ. By definition, ξ ∈ K ◦ <strong>and</strong> it’s easy to see<br />

that f ◦ (ξ) ≥ f(z) for all z ∈ K. Since K is compact, ξ is near st<strong>and</strong>ard in K, so<br />

sh(ξ) = x ∈ K. The continuity of f ensures that f(x) ∼ f ◦ (ξ), <strong>and</strong> from this we<br />

conclude f(x) ≥ f(z) for all z ∈ K.<br />

The existence of a minimum follows by the fact that −f has a maximum.<br />

Theorem 2.2.8 (Bolzano). If C ⊆ R is connected <strong>and</strong> f : C → R is a continuous<br />

function, then f(C) is connected.<br />

Proof. Suppose that f(C) = C1 ∪ C2: we want to show that there exist two nearst<strong>and</strong>ard<br />

points ζ1 ∈ (C1) ◦ <strong>and</strong> ζ2 ∈ (C2) ◦ such that ζ1 ∼ ζ2. If C1 ∩C2 = ∅, then the<br />

thesis is trivial. Let’s suppose C1 ∩C2 = ∅: clearly, this hypothesis implies f −1 (C1)∩<br />

f −1 (C2) = ∅. By Corollary 1.7.10, we have also that [f −1 (C1)] ◦ ∩ [f −1 (C2)] ◦ = ∅.<br />

Since C is connected, there exists ξ1 ∈ [f −1 (C1)] ◦ <strong>and</strong> ξ2 ∈ [f −1 (C2)] ◦ <strong>and</strong> x ∈ C<br />

such that ξ1 ∼ x ∼ ξ2. Without loss of generality, we can assume f(x) ∈ C1, so we<br />

can choose ζ1 = f(x). If mon(f(x)) ∩ (C2) ◦ = ∅, we can conclude by choosing ζ2 in<br />

that intersection. On the other h<strong>and</strong>, the hypothesis mon(f(x)) ∩ (C2) ◦ = ∅ implies<br />

that f(x) is an internal point of C1, <strong>and</strong> so, because f is continuous, x is an internal<br />

point of f −1 (C1). This would imply that ξ2 ∈ [f −1 (C1)] ◦ , but this in contradiction<br />

<strong>with</strong> the fact that f −1 (C1) <strong>and</strong> f −1 (C2) are disjoint.<br />

Corollary 2.2.9 (Bolzano Theorem of Zeroes). Let f : [a, b] → R such that f(a) ·<br />

f(b) < 0. Then, there exists c ∈ [a, b] such that f(c) = 0.<br />

Proof. Without loss of generality, let’s suppose that f(a) < 0 < f(b). By Theorem<br />

2.2.8, f([a, b]) is connected, <strong>and</strong> this implies that [f(a), f(b)] ⊆ f([a, b]). In<br />

particular, 0 ∈ f([a, b]).<br />

Theorem 2.2.10 (Heine). If K ⊂ R is a compact set <strong>and</strong> f : K → R is continuous,<br />

then f is uniformly continuous.<br />

Proof. Let ξ <strong>and</strong> ζ ∈ K ◦ <strong>with</strong> ξ ∼ ζ. Since K is compact, there exists x ∈ K<br />

verifying ξ ∼ x ∼ ζ. Then, by the continuity of f at x, we deduce f(ξ) ∼ f(x) ∼<br />

f(ζ).<br />

We encourage the reader to compare the proof of this proposition <strong>with</strong> some<br />

st<strong>and</strong>ard ones (for an example, see Theorem 6.3 of [3]).<br />

Now, we want to prove Banach-Caccioppoli Fixed Point Theorem. We recall<br />

that f : R → R is a contraction if there exists 0 < k < 1 such that, for any x,<br />

y ∈ R, |f(x) − f(y)| ≤ k|x − y|; <strong>and</strong> that x0 ∈ R is a fixed point for f if f(x0) = x0.<br />

Observe that, by definition, a contraction is continuous.<br />

Theorem 2.2.11 (Banach-Caccioppoli Fixed Point Theorem). Every contraction<br />

f : R → R has a unique fixed point.<br />

52


2.2. TOPOLOGY BY Ω-CALCULUS<br />

Proof. The unicity of the fixed point is clear, so we will only prove its existence.<br />

Let’s pick x ∈ R <strong>and</strong> define ξ = limλ↑Ω f |λ| (x) (f |λ| is the |λ|-th iteration of f).<br />

Clearly, ξ ∈ R ∗ . By hypothesis,<br />

|f |λ|+1 (x) − f |λ| (x)| ≤ k|f |λ| (x) − f |λ|−1 (x)|<br />

is true for all λ ∈ Pfin(Ω) <strong>and</strong>, iterating the inequality, we get<br />

|f |λ|+1 (x) − f |λ| (x)| ≤ k |λ| |x − f(x)|<br />

Since f ∗ (ξ) = limλ↑Ω f |λ|+1 (x), this result implies that f ∗ (ξ) ∼ ξ. If we show that ξ<br />

is finite, then, using the continuity of f <strong>and</strong> Proposition 2.1.10, from f ∗ (ξ) ∼ ξ we<br />

deduce that sh(ξ) is the fixed point for f.<br />

Now, we will prove the finiteness of ξ by reductio ad absurdum. Suppose that ξ<br />

is infinite: then, for any x ∈ R, we would have<br />

|ξ + ɛ − f(x)| = |f(ξ) − f(x)| ≤ k|ξ − x|<br />

where ɛ is an infinitesimal number. Since ξ is infinite, clearly ξ > (ɛ − f(x)) <strong>and</strong><br />

ξ > x. From those inequalities, we deduce<br />

that immediately leads to<br />

ξ + ɛ − f(x) ≤ k(ξ − x)<br />

(1 − k)ξ ≤ kx + f(x) − ɛ<br />

<strong>and</strong> this is a contradiction, since the first member of the inequality is an infinite<br />

number <strong>and</strong> the second is finite. We conclude that ξ is finite, as we wanted.<br />

We want to explicitely observe some differences between this proof <strong>and</strong> the classical<br />

one. In the classical proof, one picks an x ∈ R <strong>and</strong> defines the sequence<br />

{f n (x)}n∈N; then, the proof revolves around showing that this is a Cauchy sequence<br />

<strong>and</strong> therefore has a finite limit. In our proof, the existence of ξ is guaranteed by<br />

the axioms of the theory, <strong>and</strong> the only thing we need to show is that ξ is finite. It<br />

turns out that showing that ξ is finite involves only basic algebraics operations <strong>and</strong><br />

a little familiarity <strong>with</strong> the concept of finite <strong>and</strong> infinite numbers.<br />

Overall, we believe that the proofs of Banach-Caccioppoli Fixed Point Theorem<br />

<strong>and</strong> many other famous theorems are simplified in the mathematical framework of<br />

Ω-Calculus. It could be interesting to inquire if this happens only in a few cases or<br />

if it’s a general feature of Ω-Calculus. In the second case, then, it would arise the<br />

question if Ω-Calculus could be used as a didactic tool to help students underst<strong>and</strong><br />

mathematics. We hope that, once one gets confidence <strong>with</strong> the operation of Ω-limit,<br />

then the development of the theory could follow an easier route. It’s still an open<br />

question, <strong>and</strong> a very interesting one, to find out if this is the case.<br />

53


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

Chapter 3<br />

Integrals by Ω-Calculus<br />

Conjectures <strong>and</strong> concepts both<br />

have to pass through the purgatory<br />

of proofs <strong>and</strong> refutations. Naive<br />

conjectures <strong>and</strong> naive concepts are<br />

superseded by improved conjectures<br />

<strong>and</strong> concepts growing out of the<br />

method of proofs <strong>and</strong> refutations.<br />

Imre Lakatos<br />

Historically, the notion of integral is deeply connected <strong>with</strong> the attempts to<br />

calculate the area underneath curves. One of the first techniques developed<br />

for this goal was Cavalieri’s method of the “indivisibles”, where the<br />

Italian mathematician assumed that an area was composed by an infinite number of<br />

lines <strong>with</strong> no width. In the practice, Cavalieri’s method consisted in approximating<br />

the area underneath a curve <strong>with</strong> the area of a certain number of rectangles <strong>and</strong><br />

then evaluating the area of those rectangles as their number increases. As we can<br />

see, the underlying ideas of Cavalieri’s method somehow survived in the definition<br />

of Riemann integral, where the “area underneath a curve”, when well defined, is<br />

given as the infimum of the area of finite unions of rectangles that approximate the<br />

curve from above or as the supremum of the area of finite unions of rectangles that<br />

approximate the curve from below.<br />

In our approach to integrals, we think that the original idea expressed by Cavalieri<br />

can be formalized by Ω-Calculus: if we can give a meaningful definition of “area of<br />

a line” then, by a process of “summing all the lines”, we hope to get an interesting<br />

definition of integral.<br />

To do so, at first we find opportune to define a concept of infinite sums that<br />

have meaning for all subsets of R: in the first section of this chapter, we will briefly<br />

present this idea, that will be a generalization by Ω-limit of the concept finite sums.<br />

Then, we will define the “Ω-integral” by formalizing the intuitive ideas expressed<br />

above, <strong>and</strong> subsequently we will show some of its properties. We will see that, thanks<br />

to Ω-limit, the Ω-integral will have many of the good properties we expect integrals<br />

to satisfy.<br />

At this point, a very natural question arises: has the Ω-integral some meaningful<br />

54


3.1. HYPERFINITE SUMS<br />

relations <strong>with</strong> the “classical” integrals over R, namely the Riemann integral <strong>and</strong><br />

Lebesgue integral? It turns out that, when the Riemann integral of a function is<br />

well defined, then the Ω-integral is infinitely close to it. In the last section, we<br />

will try to estabilish some results towards the proof of a similar relation between<br />

Ω-integral <strong>and</strong> Lebesgue integral. Unfortunately, we will encounter some difficulties,<br />

so we will leave the open question if Numerosity Axiom 1 is sufficient to entail such<br />

a result.<br />

3.1 Hyperfinite sums<br />

In order to define the Ω-integral, we want to give a meaningful definition of “infinite<br />

sums” over arbitrary subsets of R. The natural idea is to try to extend the concept<br />

of finite sums by using Ω-limit. Following this intuition, we define:<br />

Definition 3.1.1. For all E ⊆ R <strong>and</strong> for all f : E → R, we set<br />

<br />

<br />

<br />

f(x)<br />

x∈E ◦<br />

f ◦ (x) = lim<br />

λ↑E<br />

<strong>and</strong> we will call it the hyperfinite sum over E◦ . We warn the reader that, <strong>with</strong> abuse<br />

of notation, we will often write<br />

<br />

f(x)<br />

instead of the correct<br />

x∈E ◦<br />

<br />

f ◦ (x)<br />

x∈E ◦<br />

It naturally arises the question wether, if we choose E = N, there are some<br />

differences between the hyperfinite sum over N◦ <strong>and</strong> the usual concept of serie.<br />

We will address this question <strong>with</strong> an example: we know that, <strong>with</strong> the classical<br />

definition of serie, the writing <br />

k∈N (−1)k is meaningless, but, according to definition<br />

3.1.1, <br />

k∈N◦(−1) k is well defined <strong>and</strong>, in our model, we can even calculate it.<br />

Example 3.1.2. We want to calculate <br />

x∈N◦(−1) k . In our model, we have that<br />

<br />

(−1) k <br />

<br />

= lim (−1) k<br />

⎛<br />

= lim ⎝ <br />

(−1) k<br />

⎞<br />

⎠<br />

k∈N ◦<br />

But we know that<br />

λ↑N<br />

<br />

k∈λ<br />

k∈λn,θ∩N<br />

(−1) k =<br />

x∈λ<br />

λn,θ↑N<br />

n<br />

(−1) k<br />

k=1<br />

k∈λn,θ<br />

<strong>and</strong>, since n = m! for some m ∈ N, n is even whenever n = 1. This ensures that<br />

<br />

(−1) k = 0<br />

k∈λn,θ∩N<br />

55


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

is eventually true. We can apply transfer principle <strong>and</strong> conclude that, in our model<br />

<br />

(−1) k = 0<br />

x∈N ◦<br />

We will not study in further details this concept of serie. Instead, we will only<br />

state that, if <br />

x∈E f(x) is well defined <strong>and</strong> absolutely convergent in the st<strong>and</strong>ard<br />

sense, then<br />

<br />

<br />

sh f(x) = <br />

f(x)<br />

x∈E ◦<br />

<strong>and</strong> the proof of this statement is left as an exercise for the interested reader. The<br />

request that <br />

x∈E f(x) must be absolutely convergent is due to the fact that the<br />

definition of <br />

x∈E◦ f(x) involves reordering the terms of the original sum, <strong>and</strong> we<br />

know from basic calculus that, if the serie is not absolutely convergent, then its value<br />

is not independent from the order of its terms.<br />

3.2 Ω-integral<br />

As we have anticipated in the introduction of this chapter, we would like to give<br />

a definition of integral of a function f over a set E ⊆ R as an infinite sum of the<br />

“areas” of the lines <strong>with</strong> base all the singletons {x} ⊆ E <strong>and</strong> height f(x). In the<br />

previous Section, we have defined the kind of infinite sums we’d like to use, namely<br />

the hyperfinite sums. The only ingredient we need to define is this “area”.<br />

For a matter of commodity, suppose that E = [a, b) <strong>and</strong> that the function f<br />

is the constant c1. In this case, the integral of c1 over [a, b) assumes the meaning<br />

of a “measure” for the set [a, b), <strong>and</strong> our wish is that this “measure” is at least<br />

infinitesimally close to the expected value b − a. Thanks to Numerosity Axiom 1,<br />

to ensure the validity of this result, it’s sufficient that the integral of the constant<br />

function c1 over a set [a, b) is equal to n([a, b)) · n([0, 1)) −1 . To ensure that this<br />

equality holds, we can define the integral in the following way:<br />

x∈E<br />

Definition 3.2.1. Let f : E ⊆ R → R. We define<br />

<br />

f ◦ 1 <br />

(x)dΩx =<br />

f<br />

n([0, 1))<br />

◦ (x)<br />

E ◦<br />

This definition is well posed for all functions, because <br />

x∈E◦ f ◦ (x) is well defined<br />

for all functions <strong>and</strong> for all subsets of R. In the future, <strong>with</strong> abuse of notation, we<br />

will often write <br />

f(x)dΩx<br />

instead of the correct <br />

E ◦<br />

E ◦<br />

f ◦ (x)dΩx<br />

Now, we need to check if this integral shares at least some of the good properties<br />

of the Riemann <strong>and</strong> Lebesgue integrals. We will start the study of the properties of<br />

56<br />

x∈E ◦


the Ω-integral by showing that<br />

<br />

[a,b) ◦<br />

c1(x)dΩx =<br />

n([a, b))<br />

n([0, 1))<br />

More precisely, we will show that, for all E ⊆ R,<br />

<br />

c1(x)dΩx = n(E)<br />

n([0, 1))<br />

since this is sufficient to entail the validity of Equation 3.1.<br />

Proposition 3.2.2. For all E ⊆ R,<br />

<br />

c1(x)dΩx = n(E)<br />

n([0, 1))<br />

E ◦<br />

E ◦<br />

Proof. By definition, we have that<br />

<br />

E ◦<br />

c1(x)dΩx =<br />

1<br />

n([0, 1)) lim<br />

<br />

<br />

c1(x)<br />

λ↑E<br />

x∈λ<br />

3.2. Ω-INTEGRAL<br />

Since λ is finite, it’s straightforward to see that the following equality holds:<br />

<br />

c1(x) = |λ|<br />

x∈λ<br />

Taking the Ω-limit <strong>and</strong> multypling by n([0, 1)) −1 , we obtain the thesis.<br />

(3.1)<br />

Since, thanks to the relation given by Proposition 3.2.2, we will make a lot of<br />

use of the ratio n(E) · n([0, 1)) −1 , for a matter of commodity we will denote it by a<br />

specific symbol.<br />

Definition 3.2.3. If E ⊆ R, we will call the normalized numerosity of order 1 of<br />

the set E the ratio<br />

n(E)<br />

n([0, 1))<br />

<strong>and</strong> we will denote it by nn1(E).<br />

We find useful to show a simple example of the use of this new notation.<br />

Example 3.2.4. With the notation of Definition 7.2.2, Numerosity Axiom 1 says<br />

that, for all a, b ∈ Q, nn1([a, b)) = b − a, <strong>and</strong> Proposition 3.2.2 says that, for all<br />

E ⊆ R, <br />

E ◦<br />

c1(x)dΩx = nn1(E)<br />

Now that we have defined the basic concepts, the remaining part of this section<br />

is dedicated to the study some basic properties of Ω-integral. From now on, unless<br />

we say otherwise, E will be an arbitrary subset of R. First of all, we check that the<br />

Ω-integral is linear.<br />

57


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

Proposition 3.2.5. For all f, g : E → R <strong>and</strong> for all a, b ∈ R,<br />

<br />

<br />

<br />

(af + bg)(x)dΩx = a · f(x)dΩx + b · g(x)dΩx<br />

E ◦<br />

Proof. By definition, we have<br />

<br />

(af + bg)(x)dΩx =<br />

E ◦<br />

<strong>and</strong>, by Corollary 1.8.4,<br />

1<br />

n([0, 1))<br />

<br />

(af + bg) ◦ (x) =<br />

x∈E ◦<br />

E ◦<br />

1<br />

n([0, 1))<br />

1<br />

n([0, 1))<br />

E ◦<br />

<br />

(af + bg) ◦ (x)<br />

x∈E ◦<br />

<br />

x∈E ◦<br />

(af ◦ (x) + bg ◦ (x))<br />

From now on, we will omit the term n([0, 1)) −1 . Again by definition, we have<br />

<br />

(af ◦ (x) + bg ◦ <br />

<br />

(x)) = lim (af(x) + bg(x))<br />

x∈λ<br />

x∈E ◦<br />

but, since the sum is finite, we have<br />

<br />

<br />

lim (af(x) + bg(x)) = lim a ·<br />

λ↑E<br />

λ↑E<br />

<br />

<br />

f(x) + b · <br />

<br />

g(x)<br />

<strong>and</strong>, using Ω-Axiom 3 several times, we get<br />

<br />

lim a ·<br />

λ↑E<br />

<br />

<br />

f(x) + b · <br />

<br />

<br />

g(x) = a · lim f(x)<br />

λ↑E<br />

x∈λ<br />

x∈λ<br />

λ↑E<br />

x∈λ<br />

x∈λ<br />

x∈λ<br />

x∈λ<br />

+ b · lim<br />

λ↑E<br />

<br />

<br />

g(x)<br />

Now, multiplying the last member of the equality chain by n([0, 1)) −1 , we obtain, by<br />

definition<br />

<br />

<br />

a · f(x)dΩx + b · g(x)dΩx<br />

as we wanted.<br />

E ◦<br />

As expected, the Ω-integral of a function over the union of disjoint sets is equal<br />

to the sum of the integrals of the function over those sets. Formally:<br />

Proposition 3.2.6. For all f : E → R <strong>and</strong> for all E1, E2 ⊆ E such as E1 ∩ E2 = ∅,<br />

<br />

<br />

<br />

f(x)dΩx = f(x)dΩx + f(x)dΩx<br />

E1∪E2<br />

Proof. Since the relation<br />

<br />

x∈λ∩(E1∪E2)<br />

E1<br />

f(x) = <br />

x∈λ∩E1<br />

E ◦<br />

E2<br />

f(x) + <br />

x∈λ∩E2<br />

f(x)<br />

holds for all λ ∈ Pfin(Ω), we can apply Transfer Principle <strong>and</strong> conclude.<br />

58<br />

x∈λ


3.3. Ω-INTEGRAL AND RIEMANN INTEGRAL<br />

As a consequence, we deduce that the same result holds for a finite union of<br />

disjoint sets.<br />

Corollary 3.2.7. For all f : E → R <strong>and</strong> for all finite families {Ei}i∈{1,...,n} such as<br />

Ei ⊆ E for all i ∈ {1, . . . , n} <strong>and</strong> Ei ∩ Ej = ∅ whenever i = j,<br />

<br />

n<br />

i=1 Ei<br />

f(x)dΩx =<br />

n<br />

<br />

i=1<br />

Ei<br />

f(x)dΩx<br />

The result of Corollary 3.2.7 shouldn’t come as a surprise, since it’s very similar<br />

to the result of finite additivity of numerosity stated by Corollary 1.6.3.<br />

3.3 Ω-integral <strong>and</strong> Riemann integral<br />

The definition of the Ω-integral reflects an intuitive, yet naive idea of area behind a<br />

curve. It’s only natural to wonder if this integral has some mathematical interest <strong>and</strong><br />

has some relations <strong>with</strong> Riemann <strong>and</strong> Lebesgue integrals. In this section, we want<br />

to show that there is indeed a close similarity between the Ω-integral <strong>and</strong> Riemann<br />

integral. In order to distinguish the different integrals, we will denote Riemann<br />

integral over a set E <strong>with</strong> the symbol R<br />

<br />

, <strong>and</strong> Lebesgue integral <strong>with</strong> E E .<br />

Since the notion of a function being Riemann-integrable is given considering step<br />

functions, we begin by recalling that definition.<br />

Definition 3.3.1. A function s : [a, b) ⊂ R → R is a step function if there are<br />

x0, x1, . . . , xn ∈ [a, b] satisfying x0 = a, xn = b <strong>and</strong> xi < xj whenever i < j <strong>and</strong> if<br />

there are y1, . . . , yn ∈ R such as the following equality holds:<br />

s(x) = yi whenever x ∈ [xi−1, xi)<br />

Thanks to Proposition 1.6.7, it’s easy to see that the Ω-integral of step functions<br />

is infinitesimally close to their Lebesgue integral.<br />

Proposition 3.3.2. Let s : [a, b) ⊂ R → R be a step function. Then<br />

<br />

<br />

s(x)dΩx ∼ s(x)dx<br />

[a,b) ◦<br />

Proof. Since the domain is bounded <strong>and</strong> s assumes only a finite number of values<br />

over [a, b), by linearity we can assume that s is constant over [a, b). Then, there<br />

exists y ∈ R such as<br />

<br />

<br />

<br />

s(x)dΩx = y · c1(x)dΩx = y · c1(x)dΩx<br />

[a,b) ◦<br />

[a,b) ◦<br />

[a,b)<br />

is true. By Proposition 3.2.2, we deduce that<br />

<br />

y · c1(x)dΩx = y · nn1([a, b))<br />

[a,b) ◦<br />

[a,b) ◦<br />

59


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

<strong>and</strong>, using Proposition 1.6.7, we can conclude<br />

<br />

y · nn1([a, b)) ∼ y (b − a) =<br />

as we wanted.<br />

[a,b)<br />

s(x)dx<br />

We recall that a function f is Riemann-integrable over the set [a, b) if there<br />

exists a real number l <strong>with</strong> the following property: for every positive ɛ ∈ R there<br />

exists a positive δ ∈ R such that for every partition P = a = x0 < . . . < xs = b <strong>with</strong><br />

defined by<br />

maxi=1,...,s(|xi − xi−1|) < δ, the step functions s ±<br />

P<br />

<strong>and</strong><br />

verify R<br />

s +<br />

P (x) = max<br />

x∈[xi,xi+1) f(x) if x ∈ [xi, xi+1)<br />

s −<br />

P (x) = min f(x) if x ∈ [xi, xi+1)<br />

x∈[xi,xi+1)<br />

[a,b)<br />

s ±<br />

P<br />

<br />

<br />

(x)dx − l<br />

< ɛ<br />

If such a number exists, we say that l is the integral of f over the set [a, b), <strong>and</strong> we<br />

write<br />

l =<br />

R<br />

[a,b)<br />

f(x)dx<br />

Proposition 3.3.3. If f : [a, b) → R is Riemann-integrable, then<br />

<br />

R<br />

f(x)dΩx ∼ f(x)dx<br />

[a,b) ◦<br />

Proof. If we pick k ∈ N, then there exists λk ∈ Pfin(Ω) such as the partition<br />

Pλk = λk ∩ [a, b) verifies<br />

<br />

R<br />

<br />

<br />

<br />

<br />

(x)dx − l<br />

< 1/k<br />

[a,b)<br />

[a,b)<br />

s ±<br />

Pλ k<br />

<strong>and</strong> the formula is still true if we replace λk <strong>with</strong> another λ ⊇ λk. Now, Proposition<br />

3.3.2 ensures that, for all λ ∈ Pfin(Ω), the following equality holds:<br />

R<br />

s ±<br />

<br />

(x)dx = sh<br />

Pλ s ±<br />

Pλ (x)dΩx<br />

<br />

<strong>and</strong>, in our case, this means that<br />

<br />

<br />

<br />

sh <br />

[a,b) ◦<br />

[a,b)<br />

[a,b) ◦<br />

s ±<br />

<br />

(x)dΩx<br />

Pλk <br />

<br />

− l<br />

< 1/k<br />

is eventually true for all k ∈ N. Applying Transfer principle, we can conclude<br />

<br />

lim sh<br />

λ↑Ω<br />

s ±<br />

<br />

(x)dΩx<br />

Pλk ∼ l<br />

60<br />

[a,b) ◦


[a,b) ◦<br />

3.4. NUMEROSITY AND LEBESGUE MEASURE<br />

Another application of Transfer principle ensures that<br />

<br />

lim sh<br />

λ↑Ω<br />

s −<br />

<br />

<br />

(x)dΩx<br />

Pλk ≤ f(x)dx ≤ lim sh<br />

λ↑Ω<br />

Combining the two results, we get<br />

<br />

as we wanted.<br />

[a,b) ◦<br />

[a,b) ◦<br />

f(x)dΩx ∼ l<br />

[a,b) ◦<br />

3.4 Numerosity <strong>and</strong> Lebesgue measure<br />

s +<br />

<br />

(x)dΩx<br />

Pλk The result that allowed to relate Ω-integral to Riemann integral was given by Proposition<br />

3.3.2, whose validity depends upon Proposition 3.2.2. We recall that this<br />

Proposition states the equality between the Ω-integral of the characteristic function<br />

of a set <strong>with</strong> the normalized numerosity of that set: in the case of intervals, Numerosity<br />

Axiom 1 ensures that this value is the Peano-Jordan measure of the interval. For<br />

this reason, we find interesting to begin our study of the relation between Ω-integral<br />

<strong>and</strong> Lebesgue integral by studying the relations between numerosity <strong>and</strong> Lebesgue<br />

measure. In particular, we begin by addressing a specific problem of numerosity:<br />

given Numerosity Axiom 1, it’s hard to determine if numerosity is countably additive<br />

or even countably subadditive. Indeed, it’s easier to show that the other inequality<br />

is more likely to hold.<br />

Example 3.4.1. Suppose that {En}n∈N are nonempty subsets of R <strong>and</strong> that Ei∩Ej =<br />

∅ whenever i = j. If we call E = <br />

n∈N En, then, by the trivial inclusion<br />

we get that, for all k ∈ N,<br />

k<br />

En ⊂ E<br />

n=1<br />

k<br />

n(En) < n(E)<br />

n=1<br />

If we have the additional hypothesis that nn1(En) ∈ R for all n ∈ N, we can go<br />

further: by transfer of inequalities, we deduce<br />

<br />

nn1(En) ≤ sh (nn1(E))<br />

n∈N ◦<br />

The best we can do is trying to get an additivity result under some hypotheses<br />

of finiteness. We summarize those hypotheses in the next definitions.<br />

Definition 3.4.2. We will call E + the family of all sets E = <br />

n∈N [an, bn) satisfying<br />

the hypothesis<br />

1. E ⊆ [−N, N) for some N ∈ N;<br />

61


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

2. {[an, bn)}n∈N are pairwise disjoint;<br />

3. bn ≤ an+1 for all n ∈ N.<br />

<strong>and</strong> we will call E − the family of all sets E = <br />

n∈N [an, bn) satisfying hypothesis 1, 2<br />

<strong>and</strong><br />

3’ bn+1 ≤ an for all n ∈ N.<br />

Notice that, for all E ∈ E ± , the serie <br />

n∈N (bn − an) is well defined in the<br />

classical sense <strong>and</strong> assumes a finite value. Notice also that <br />

n∈N (bn − an) is exactly<br />

the Lebesgue measure of E. For a matter of convenience, we will give a name to<br />

this measure.<br />

Definition 3.4.3. We will denote by µ the Lebesgue measure over R. We remark<br />

explicitely that, for all bounded E = <br />

n∈N [an, bn), where the union is disjoint,<br />

µ(E) = <br />

n∈N (bn − an).<br />

The idea is to give a result of countable additivity <strong>with</strong> respect to all sets E ∈ E ± .<br />

There is only one problem: under those hypothesis, nn1([an, bn)) ∈ R ∗ , so we won’t<br />

be able to calculate their hyperfinite sum. The solution is to consider only the sets<br />

E ∈ E ± for which nn1([an, bn)) ∈ R for all n ∈ N: thanks to Numerosity Axiom 1,<br />

we know that this request means that an <strong>and</strong> bn must be rational numbers.<br />

Definition 3.4.4. We will call E ±<br />

Q the family of all sets E ∈ E ± satisfying the<br />

additional hypothesis an, bn ∈ Q for all n ∈ N.<br />

We can now give a result of countable additivity for all E ∈ E ±<br />

Q . Then, we will<br />

be able to generalize it to all sets of E ± thanks to monotony of numerosity.<br />

Proposition 3.4.5. For all E ∈ E +<br />

Q , it holds<br />

sh (nn1(E)) ∼ <br />

(bn − an)<br />

n∈N ◦<br />

Proof. Observe that, by hypothesis, for the classical limits it holds<br />

lim<br />

n→∞ an = lim bn<br />

n→∞<br />

Now, define b = limn→∞ bn <strong>and</strong>, for all k ∈ N, choose Ak <strong>and</strong> Bk ∈ Q such that<br />

<strong>and</strong><br />

0 ≤ ak+1 − Ak < 1/k<br />

0 ≤ Bk − b < 1/k<br />

Explicitely, Ak <strong>and</strong> Bk are two rational points that satisfy<br />

<strong>and</strong><br />

62<br />

lim<br />

n→∞ Ak = lim Bk = b<br />

n→∞<br />

[Ak, Bk) ⊇ [ak+1, b) ⊇ <br />

[an, bn)<br />

n>k


3.4. NUMEROSITY AND LEBESGUE MEASURE<br />

From the inclusion above, we deduce<br />

<br />

<br />

<br />

[an, bn) ⊆ E ⊆ [an, bn) ∪ [Ak, Bk)<br />

n≤k<br />

hence, by monotony <strong>and</strong> finite additivity of numerosity,<br />

<br />

(bn − an) ≤ nn1(E) ≤ <br />

(bn − an) + (Bk − Ak)<br />

n≤k<br />

n≤k<br />

Notice that, by hypothesis, the first <strong>and</strong> the last term of the chain of inequalities<br />

are rational numbers. We can then take the shadow of all the terms <strong>and</strong> obtain:<br />

<br />

(bn − an) ≤ sh(nn1(E)) ≤ <br />

(bn − an) + (Bk − Ak)<br />

n≤k<br />

At this point, taking the Ω-limit of the previous equation <strong>with</strong> respect to k, we<br />

obtain<br />

<br />

(bn − an) ≤ sh(nn1(E)) ≤ <br />

<br />

(bn − an) − lim B|λ| − A|λ|<br />

n∈N ◦<br />

n≤k<br />

n∈N ◦<br />

<br />

If we prove that limλ↑Ω B|λ| − A|λ| ∼ 0, then the thesis will follow.<br />

Suppose that limλ↑Ω B|λ| − A|λ| ∼ l: by the equivalence <strong>with</strong> classical limit, the<br />

following equality holds in the st<strong>and</strong>ard sense:<br />

n≤k<br />

λ↑Ω<br />

l = lim<br />

n→∞ (Bn − An) = lim<br />

n→∞ Bn − lim<br />

n→∞ An = b − b = 0<br />

We deduce l = 0, <strong>and</strong> this concludes the proof.<br />

Corollary 3.4.6. For all E ∈ E +<br />

Q , sh (nn1(E)) = µ(E).<br />

Proof. This result follows by Proposition 3.4.5 <strong>and</strong> the relation<br />

<br />

(bn − an) ∼ µ(E)<br />

n∈N ◦<br />

Putting everything together, we obtain<br />

sh (nn1(E)) ∼ µ(E)<br />

but, since both are real numbers, they must be equal.<br />

With the next proposition, we will extend this result to all the sets in E + .<br />

Proposition 3.4.7. For all E ∈ E + , sh (nn1(E)) = µ(E).<br />

Proof. Thanks to Corollary 3.4.6, we only need to prove the thesis for all the sets<br />

E ∈ E + \ E +<br />

Q . To do so, choose such a set E <strong>and</strong>, notice that for all n ∈ N, we can<br />

easily find two sets E + n <strong>and</strong> E − n ∈ E +<br />

Q satisfying<br />

E − n ⊆ E ⊆ E + n<br />

63


CHAPTER 3. INTEGRALS BY Ω-CALCULUS<br />

<strong>and</strong><br />

|µ(E ± n ) − µ(E)| < 1<br />

(3.2)<br />

n<br />

By the first chain of inclusions <strong>and</strong> by monotony of numerosity, we deduce that the<br />

following inequality<br />

holds for all n ∈ N or, in other words,<br />

sh(nn1(E − n )) ≤ sh(nn1(E)) ≤ sh(nn1(E + n ))<br />

µ(E − n ) ≤ sh(nn1(E)) ≤ µ(E + n )<br />

Taking the Ω-limit <strong>with</strong> respect to n <strong>and</strong> thanks to equation 3.2, we conclude<br />

sh(nn1(E)) ∼ µ(E)<br />

Now, notice that both numbers are real numbers, hence the desired equality.<br />

Mutatis mut<strong>and</strong>is, the same arguments can be applied to prove the analog of<br />

Proposition 3.4.5 <strong>and</strong> Corollary 3.4.6 where E ∈ E −<br />

Q , <strong>and</strong> of Proposition 3.4.7 for<br />

the case where E ∈ E − . For this reason, we will omit the proof of those statements,<br />

that are left as exercises for the interested reader. Instead, we jump directly to the<br />

final result, obtained from the previous ones by finite additivity of numerosity.<br />

Proposition 3.4.8. If E = k<br />

n=1 Ek <strong>and</strong> En ∈ E + ∪ E − for all n ≤ k, then<br />

sh(nn1(E)) =<br />

k<br />

µ(Ek)<br />

A few words of comment are in order. The final result stated in Proposition 3.4.8<br />

is very narrow, <strong>and</strong> it’s very hard to use it to link numerosity to Lebesgue measure.<br />

The main difficulty is that Lebesgue measure is obtained by considering arbitrary<br />

unions of intervals, whereas in our case we can show how numerosity behaves only in<br />

a very specific case. The restrictive hypotheses on the union considered don’t allow<br />

an easy extension of this result to arbitrary unions, <strong>and</strong> it’s still an open question if<br />

Numerosity Axiom 1 is sufficient to estabilish a meaningful relation between numerosity<br />

<strong>and</strong> Lebesgue measure. Given the difficulties, we choose to leave this question<br />

open for further research. The interested reader could find some more ideas toward<br />

this direction in Sections 7.3 <strong>and</strong> 7.8 of this paper.<br />

64<br />

n=1


Chapter 4<br />

Ω-<strong>Theory</strong><br />

A mathematician, like a painter or<br />

a poet, is a maker of patterns.<br />

G. H. Hardy<br />

As we’ve seen in the previous chapters of this paper, the scope of Ω-Calculus<br />

is limited to real functions, subsets of real numbers <strong>and</strong> their hyperextensions,<br />

so that theory alone is insufficient for many applications. To<br />

solve this issue, we can generalize the axiomatics of Ω-Calculus by simply extending<br />

the range of the postulated properties from functions <strong>with</strong> range in the real numbers<br />

to arbitrary functions. Clearly, we will require that, when we focus on real valued<br />

functions, the Ω-limit will coincide to the one we studied in the previous chapters.<br />

We will obtain a new theory, that we will call Ω-<strong>Theory</strong>, that extends Ω-Calculus<br />

<strong>and</strong> gives more mathematical tools that can be applied in many different situations.<br />

The only difference <strong>with</strong> Ω-Calculus will be that, in the basis of Ω-<strong>Theory</strong>, we<br />

will omit the numerosity axioms <strong>and</strong> we will rule only those regarding Ω-limit. This<br />

choice is due to the fact that different applications could require different axioms for<br />

numerosity, <strong>and</strong> we want to be free to choose them as the need arises. Moreover, it<br />

seems to us that numerosity axioms are more arbitrary <strong>and</strong> less “intrinsic” to the<br />

theory as those of Ω-limit.<br />

After a brief section about the foundational framework of Ω-<strong>Theory</strong>, we will follow<br />

the same structure as Chapter 1: at first, we will present the axioms of the theory,<br />

<strong>and</strong> then we will study their consequences, generalizing concepts already introduced<br />

in the previous chapters <strong>and</strong> exploring new ideas. At last, we will exhibit a model<br />

for Ω-<strong>Theory</strong>.<br />

4.1 Foundational framework<br />

The axioms of Ω-Calculus will have a really broad range, in the sense that they will<br />

rule the Ω-limits of all sequences of mathematical objects. Here, by “mathematical<br />

objects” we mean all the entities that are used in the ordinary practice of mathematics,<br />

namely numbers, sets, functions, relations, ordered tuples, <strong>and</strong> so forth.<br />

As suggested by the common use, in our approach we distinguish between sets as<br />

65


CHAPTER 4. Ω-THEORY<br />

legitimate collections of objects, <strong>and</strong> atoms as mathematical objects that are not<br />

sets. We remark that, in the practice, the “status” of atom or set may depend on<br />

the context. For instance, in the well-known construction of the real numbers as<br />

Dedekind cuts, real numbers are defined as suitable subsets of rational numbers. On<br />

the other h<strong>and</strong>, in the practice of analysis, one hardly considers a real number as a<br />

set. Similarly, when studying a topological space, one takes its elements as atoms,<br />

even in the cases when they are in fact sets (e.g. in the case of space of functions).<br />

As a general rule, we shall make the assumption that numbers are atoms, <strong>and</strong><br />

that ordered tuples, relations <strong>and</strong> functions are sets. In fact, each of these notions<br />

admits a representation as a set, <strong>and</strong> can be formally defined <strong>with</strong>in the framework<br />

of axiomatic set theory. In particular, we agree on the following:<br />

• Ordered pairs are given by the Kuratowski pairs:<br />

(a, b) = {{a}, {a, b}}<br />

Inductively, ordered n + 1-tuples are defined by putting<br />

(a1, . . . an+1) = ((a1, . . . , an), an+1)<br />

• A binary relation R is identified <strong>with</strong> the set of ordered pairs that satisfy it:<br />

R = {(a, b) : aRb}<br />

In the same spirit, a n-place relation is the set of ordered n-tuples that satisfy<br />

it.<br />

• A function f is identified <strong>with</strong> its graph:<br />

f = {(a, b) : b = f(a)}<br />

At this stage, it is safe to say that when applying the Ω-<strong>Theory</strong> in everyday<br />

practice, one does not need to know about the specific foundational framework that<br />

is adopted. However, we signal that ZFC set theory is not the appropriate framework<br />

for Ω-<strong>Theory</strong>: in particular, the axiom of foundation is incompatible <strong>with</strong> Ω-<strong>Theory</strong>,<br />

<strong>and</strong> in Section 4.4 we will give an intuitive idea of why it is so. A convenient<br />

framework to construct models of the Ω-<strong>Theory</strong> is the so-called Zermelo-Fraenkel-<br />

Boffa set theory ZFBC which, roughly speaking, is a non-wellfounded variant of<br />

ZFC where the axiom of foundation is replaced by other basic principles. Since<br />

the discussion of the details of this theory is beyond the scope of this paper, we<br />

rem<strong>and</strong> to Chapter 7 of [1] for further references about the foundational framework<br />

of Ω-<strong>Theory</strong>.<br />

4.2 The Axioms of Ω-<strong>Theory</strong><br />

As we did in the case of Ω-Calculus, we begin the presentation of Ω-<strong>Theory</strong> by<br />

stating all the axioms of the theory. In analogy to the first axiom of Ω-Calculus,<br />

the first axiom of Ω-<strong>Theory</strong> is an existence axiom that ensures the existence of the<br />

Ω-limit of every function that takes values in Ω.<br />

66


4.2. THE AXIOMS OF Ω-THEORY<br />

Ω-Axiom 1 (Existence Axiom). Every function ϕ : Pfin(Ω) → Ω has a unique<br />

Ω-limit. This limit is denoted by<br />

lim<br />

λ↑Ω ϕ(λ)<br />

Since we are no longer restricting the range of functions we consider, it’s useful to<br />

introduce an axiom that helps us distinguish between limits of numbers <strong>and</strong> limits of<br />

sets <strong>and</strong> that explicitely rules how the latter behave. In particular, we require that<br />

the Ω-limits of functions whose values are atoms is an atom, <strong>and</strong> that the Ω-limit<br />

of functions whose values are sets behave in analogy to Definition 1.7.1.<br />

Ω-Axiom 2 (Atoms <strong>and</strong> Sets Axiom). If ϕ(λ) is an atom for all λ ∈ Pfin(Ω), then<br />

limλ↑Ω ϕ(λ) is an atom. If ϕ(λ) is a nonempty set for all λ ∈ Pfin(Ω), then<br />

<br />

<br />

lim ϕ(λ) = lim ψ(λ) : ψ(λ) ∈ ϕ(λ) ∀λ ∈ Pfin(Ω)<br />

λ↑Ω λ↑Ω<br />

(4.1)<br />

Moreover, we ask that the Ω-limit of the function <strong>with</strong> constant value the empty set<br />

is the empty set:<br />

lim<br />

λ↑Ω c∅(λ) = ∅<br />

In analogy to Ω-Calculus, we want the Ω-limit to be compatible <strong>with</strong> the field<br />

operations of R. To do so, we impose it <strong>with</strong> a new axiom.<br />

Ω-Axiom 3 (Real <strong>Numbers</strong> Axiom). The set<br />

R ∗ <br />

<br />

= lim ψ(λ) : ψ(λ) ∈ R ∀λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

is a field <strong>and</strong>, if ϕ : Pfin(Ω) → R is eventually constant, then<br />

Moreover, for all ϕ, ψ : Pfin(Ω) → R:<br />

lim ϕ(λ) = r<br />

λ↑Ω<br />

limλ↑Ω ϕ(λ) + limλ↑Ω ψ(λ) = limλ↑Ω(ϕ(λ) + ψ(λ))<br />

limλ↑Ω ϕ(λ) · limλ↑Ω ψ(λ) = limλ↑Ω(ϕ(λ) · ψ(λ))<br />

As last axiom regarding Ω-limt, we need to introduce an axiom to rule that<br />

Ω-limit is coherent <strong>with</strong> respect to composition of functions.<br />

Ω-Axiom 4 (Composition Axiom). If ϕ, ψ : Pfin(Ω) → Ω satisfy<br />

then also<br />

lim ϕ(λ) = lim ψ(λ)<br />

λ↑Ω λ↑Ω<br />

lim f(ϕ(λ)) = lim f(ψ(λ))<br />

λ↑Ω λ↑Ω<br />

for all f for which are well defined the compositions f ◦ ϕ <strong>and</strong> f ◦ ψ.<br />

67


CHAPTER 4. Ω-THEORY<br />

4.3 Generalizing concepts<br />

Once the axioms of the theory are given, we can generalize to Ω-<strong>Theory</strong> the definitions<br />

<strong>and</strong> concepts introduced in Chapter 1 for the case of real subsets <strong>and</strong> functions.<br />

A lot of new definitions will follow closely their counterparts, so we will give them<br />

<strong>with</strong>out many comments.<br />

Definition 4.3.1. If E ⊆ Ω, we define<br />

lim ϕ(λ) = lim ϕ(λ ∩ E)<br />

λ↑E λ↑Ω<br />

Definition 4.3.2. The natural extension of a set E ⊆ Ω is given by<br />

E ∗ = lim<br />

λ↑Ω cE(λ)<br />

<strong>and</strong>, by Ω-Axiom 2, it’s equal to<br />

E ∗ <br />

<br />

= lim ψ(λ) : ψ(λ) ∈ E ∀λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

The hyperfinite extension of a set E ⊆ Ω is given by<br />

E ◦ = lim<br />

λ↑E λ<br />

Definition 4.3.3. Let f : E → F . The natural extension of f, which we denote<br />

f ∗ , is defined by:<br />

f ∗ (lim<br />

λ↑Ω ϕ(λ)) = lim<br />

λ↑Ω f(ϕ(λ))<br />

Notice that, by definition, f ∗ : E ∗ → F ∗ .<br />

The hyperfinite extension of a function f : E → F is defined as the restriction<br />

of f ∗ to E ◦ :<br />

f ◦ = f ∗<br />

|E ◦ : E◦ → F ∗<br />

Notice that, thanks to Ω-Axiom 4, the natural extension of a function f is well<br />

defined. It’s easy to see that, if E <strong>and</strong> F ⊆ R, definition 4.3.2 coincide <strong>with</strong> definitions<br />

1.7.3 <strong>and</strong> 1.7.4, <strong>and</strong> definition 4.3.3 coincide <strong>with</strong> definition 1.8.2.<br />

In analogy to Definition 1.4.3, we give the following<br />

Definition 4.3.4. For all ϕ, ψ : Pfin(Ω) → Ω, we define Qϕ,ψ = {λ ∈ Pfin(Ω) :<br />

ϕ(λ) = ψ(λ)}. We define also<br />

<br />

<br />

Q =<br />

Qϕ,ψ : lim<br />

λ↑Ω ϕ(λ) = lim<br />

λ↑Ω ψ(λ)<br />

<strong>and</strong>, as we did before, we will say that a set Q ∈ Q is qualified, <strong>and</strong> we will call Q<br />

the set of qualified sets.<br />

Once again, it’s straightforward to check that, if ϕ <strong>and</strong> ψ assume real values,<br />

then the two definitions coincide, so the common notation is justified. Now, we want<br />

to show that Corollary 1.4.7 apply to arbitrary functions.<br />

68


4.3. GENERALIZING CONCEPTS<br />

Proposition 4.3.5. For all ϕ, ψ : Pfin(Ω) → Ω, limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ) if <strong>and</strong><br />

only if Qϕ,ψ ∈ Q.<br />

Proof. One of the implications is true by definition, so we will focus on the other.<br />

Notice also that, if ϕ <strong>and</strong> ψ assume real values almost everywhere, then the thesis<br />

holds by Corollary 1.4.7.<br />

Now, suppose that Qϕ,ψ ∈ Q. If we define the function f : Pfin(Ω) → {0, 1} in<br />

the following way:<br />

<br />

2 if λ ∈ Qϕ,ψ<br />

f(λ) =<br />

0 if λ ∈ Qϕ,ψ<br />

then, by Corollary 1.4.7, we deduce that limλ↑Ω f(λ) = 0. Notice that, if the function<br />

“{ }” defined by {}(x) = {x} is internal to Ω, then f can also be defined by the<br />

formula<br />

f(λ) = |{ϕ(λ)}△{ψ(λ)}|<br />

<strong>and</strong>, by repeated use of Ω-Axiom 4,<br />

lim f(λ) = lim (|{ϕ(λ)}△{ψ(λ)}|) = |{lim ϕ(λ)}△{lim ψ(λ)}|<br />

λ↑Ω λ↑Ω λ↑Ω λ↑Ω<br />

From this equality we conclude ϕ(λ) = ψ(λ).<br />

If {} is not internal to Ω, call A the domain of {} <strong>and</strong> B = Ω\A. Moreover, pose<br />

Ba = {x ∈ B : x is an atom} <strong>and</strong> Bs = {x ∈ B : x is a set}. With those definitions,<br />

f is almost everywhere equal to the function g : Pfin(λ) → R defined a.e. as follows:<br />

⎧<br />

⎪⎨<br />

|{ϕ(λ)}△{ψ(λ)}| if ϕ(λ) <strong>and</strong> ψ(λ) ∈ A<br />

ϕ(λ) − ψ(λ) if ϕ(λ) <strong>and</strong> ψ(λ) ∈ Ba<br />

g(λ) =<br />

⎪⎩<br />

|ϕ(λ)△ψ(λ)| if ϕ(λ) <strong>and</strong> ψ(λ) ∈ Bs<br />

2 otherwise<br />

Using Corollary 1.4.7 <strong>and</strong> <strong>with</strong> an argument similar to the previous part of the proof,<br />

we can conclude.<br />

This Proposition has several consequences, that we present here in the form of<br />

corollaries.<br />

Corollary 4.3.6. For all ϕ : Pfin(Ω) → Ω, limλ↑Ω ϕ(λ) is an atom or is a set.<br />

Proof. Since we assume that the atoms are real numbers, define the function<br />

<br />

0 if ϕ(λ) ∈ R<br />

ψa(λ) =<br />

1 if ϕ(λ) ∈ Rc Now, if limλ↑Ω ψa(λ) = 0 then the set A = {λ ∈ Pfin(Ω) : ϕ(λ) ∈ R} is qualified,<br />

hence we can exhibit a function ψ : Pfin(Ω) → R such that Qϕ,ψ = A. This <strong>and</strong><br />

Proposition 4.3.5 is sufficient to conclude that limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ) ∈ R ∗ is an<br />

atom. In the same spirit, if ψa(λ) = 1, then the set S = {λ ∈ Pfin(Ω) : ϕ(λ) ∈ R}<br />

is qualified <strong>and</strong>, <strong>with</strong> the same reasoning, then limλ↑Ω ϕ(λ) is a set.<br />

Thanks to this Corollary <strong>and</strong> to Proposition 4.3.5, we can easily deduce another<br />

characterization of Q.<br />

69


CHAPTER 4. Ω-THEORY<br />

Corollary 4.3.7. If we call<br />

<br />

A =<br />

<strong>and</strong><br />

B =<br />

<br />

then Q = A = B.<br />

Qϕ,ψ : lim<br />

λ↑Ω ϕ(λ) = lim<br />

λ↑Ω ψ(λ) <strong>and</strong> ϕ(λ), ψ(λ) ∈ R ∀λ ∈ Pfin(Ω)<br />

Qϕ,ψ : lim<br />

λ↑Ω ϕ(λ) = lim<br />

λ↑Ω ψ(λ) <strong>and</strong> ϕ(λ), ψ(λ) ∈ R c ∀λ ∈ Pfin(Ω)<br />

Proof. By Corollary 4.3.6, it should be clear that Q = A ∪ B. By Proposition 4.3.5,<br />

it’s easy to see that A = B.<br />

The last, <strong>and</strong> maybe more important consequence of Proposition 4.3.5 is that,<br />

in analogy to what happened in the case of Ω-Calculus, Q is a nonprincipal fine<br />

ultrafilter on Pfin(Ω).<br />

Corollary 4.3.8. Q is a nonprincipal fine ultrafilter on Pfin(Ω).<br />

Proof. The proof follows easily by the characterization of Q given by Corollary 4.3.7<br />

<strong>and</strong> <strong>with</strong> the same arguments as in the proof of Proposition 1.4.8.<br />

Notice that, by Ω-<strong>Theory</strong>, we could prove all the results about extensions of sets<br />

<strong>and</strong> functions of Sections 1.7 <strong>and</strong> 1.8 for the case of arbitrary subsets of Ω <strong>and</strong> of<br />

arbitrary functions f : A ⊆ Ω → Ω. Since the proofs are almost identical to those<br />

already seen, we will omit them. The reader is warned that, when the need arises,<br />

we will feel free to use the properties ruled in Sections 1.7 <strong>and</strong> 1.8 in this more<br />

general interpretation.<br />

4.4 Transfer Principle by Ω-<strong>Theory</strong><br />

The fundamental feature of the Ω-<strong>Theory</strong> is that all “elementary properties” are<br />

preserved under Ω-limits. In fact, <strong>with</strong> respect to Ω-Calculus, a more general version<br />

of the transfer principle holds, where arbitrary sequences of mathematical objects<br />

(that is, numbers <strong>and</strong> sets) are considered.<br />

Transfer Principle (general form). An “elementary property” P is satisfied by<br />

elements ϕ1(λ), . . . ϕn(λ) for almost all λ if <strong>and</strong> only if P is satisfied by the Ω-limits<br />

limλ↑Ω ϕ1(λ), . . . , ϕn(λ).<br />

The proof of this result requires some logic knowledge, <strong>and</strong> for this reason, we<br />

will give it in Chapter 5, after we’ve introduced the necessary logic formalism. In<br />

this section, we want to show that, using the theory developed so far, we can prove<br />

directly several important instances of Transfer. In particular, we will show that<br />

equality <strong>and</strong> membership transfer under Ω-limit, as well as the set operations <strong>and</strong><br />

the notions of ordered tuples, relations <strong>and</strong> functions. As a consequence, if one<br />

is not interested by the logical aspect of Ω-<strong>Theory</strong>, the results contained in this<br />

70


4.4. TRANSFER PRINCIPLE BY Ω-THEORY<br />

section should give sufficient instances of Transfer Principle to practice everyday<br />

mathematics.<br />

As a first fundamental result of Ω-<strong>Theory</strong>, we will prove that the above transfer<br />

principle apply to both the “elementary” properties of equality <strong>and</strong> membership.<br />

Theorem 4.4.1 (Transfer of equality <strong>and</strong> membership). For all ϕ, ψ : Pfin(Ω) → Ω,<br />

1. ϕ(λ) = ψ(λ) a.e. ⇐⇒ limλ↑Ω ϕ(λ) = limλ↑Ω ψ(λ)<br />

2. ϕ(λ) ∈ ψ(λ) a.e. ⇐⇒ limλ↑Ω ϕ(λ) ∈ limλ↑Ω ψ(λ)<br />

The first part of the Theorem has already been proven in Proposition 4.3.5. To<br />

prove this important theorem, we will need some preliminary results. At first, we<br />

will prove one of the implications of transfer of membership: if ϕ(λ) ∈ ψ(λ) a.e.,<br />

then limλ↑Ω ϕ(λ) ∈ limλ↑Ω ψ(λ).<br />

Proposition 4.4.2. For all ϕ, ψ : Pfin(Ω) → Ω, if ϕ(λ) ∈ ψ(λ) a.e., then<br />

limλ↑Ω ϕ(λ) ∈ limλ↑Ω ψ(λ).<br />

Proof. At first, define<br />

ϕ ′ (λ) =<br />

ϕ(λ) if ϕ(λ) ∈ ψ(λ)<br />

0 if ϕ(λ) ∈ ψ(λ)<br />

Since ϕ = ϕ ′ a.e., if we prove that limλ↑Ω ϕ ′ (λ) ∈ limλ↑Ω ψ(λ) then, by transfer of<br />

equality, we can conclude that the thesis holds. To prove this result, define the<br />

function<br />

ψ ′ <br />

ψ(λ)<br />

(λ) =<br />

{0}<br />

if ϕ(λ) ∈ ψ(λ)<br />

if ϕ(λ) ∈ ψ(λ)<br />

By definition, ϕ ′ (λ) ∈ ψ ′ (λ) for all λ ∈ Pfin(Ω), <strong>and</strong> this implies limλ↑Ω ϕ ′ (λ) ∈<br />

limλ↑Ω ψ ′ (λ). On the other h<strong>and</strong>, ψ ′ (λ) = ψ(λ) a.e. <strong>and</strong>, by transfer of equality,<br />

limλ↑Ω ψ ′ (λ) = limλ↑Ω ψ(λ), so the thesis follows.<br />

Thanks to Proposition 4.4.2, we will now show that Ω-Axiom 2 allows a weaker<br />

formulation. This is the last result we need to settle before we prove the second part<br />

of Theorem 4.4.1.<br />

Lemma 4.4.3 (Sets Axiom - a.e. version). If ϕ(λ) is a set a.e., then<br />

<br />

<br />

lim ϕ(λ) = lim ψ(λ) : ψ(λ) ∈ ϕ(λ) a.e.<br />

λ↑Ω λ↑Ω<br />

Proof. Choose a function η such that η is a set for all λ ∈ Pfin(Ω) <strong>and</strong> Qϕ,η ∈ Q. If<br />

ψ(λ) ∈ ϕ(λ) a.e., then we have the relation<br />

{λ ∈ Pfin(Ω) : ψ(λ) ∈ η(λ)} ⊇ Qϕ,η ∩ {λ ∈ Pfin(Ω) : ψ(λ) ∈ ϕ(λ)}<br />

<strong>and</strong>, since the last two sets are qualified by hypothesis, by Corollary 4.3.8 we conclude<br />

that also ψ(λ) ∈ η(λ) a.e. Thanks to Proposition 4.4.2, we conclude that<br />

limλ↑Ω ψ(λ) ∈ limλ↑Ω η(λ) = limλ↑Ω ϕ(λ).<br />

Conversely, let x ∈ limλ↑Ω ϕ(λ) = limλ↑Ω η(λ). Then, by Ω-Axiom 2, x =<br />

limλ↑Ω ψ(λ) for a suitable function ψ such that ψ(λ) ∈ η(λ) for all λ ∈ Pfin(Ω).<br />

Since Qϕ,η ∈ Q, it follows that ψ(λ) ∈ ϕ(λ) is true a.e., as we wanted.<br />

71


CHAPTER 4. Ω-THEORY<br />

Now, we are ready to prove Theorem 4.4.1.<br />

Proof of Theorem 4.4.1. We have already observed that the proof of the Transfer of<br />

equality is the content of Proposition 4.3.5.<br />

The first implication of Transfer of membership has been proven in Proposition<br />

4.4.2. To show that the other implication holds, suppose that limλ↑Ω ϕ(λ) ∈<br />

limλ↑Ω ψ(λ). This implies that limλ↑Ω ψ(λ) is a nonempty set, hence Proposition 4.3.5<br />

ensures that ψ(λ) is a nonempty set a.e. By Lemma 4.4.3, it follows the existence<br />

of a function η(λ) satisfying η(λ) ∈ ψ(λ) a.e. <strong>and</strong> limλ↑Ω η(λ) = limλ↑Ω ϕ(λ). By<br />

Proposition 4.3.5, this implies Qϕ,η ∈ Q, hence ϕ(λ) ∈ ψ(λ) a.e., <strong>and</strong> this concludes<br />

the proof.<br />

Example 4.4.4. In Example 1.5.2 we have seen that the numerical property P (x)<br />

saying “x ∈ E” does not transfer when E is infinite. Now, if we consider the<br />

relation R(x, E) saying “x ∈ E”, then Theorem 4.4.1 ensures that R(ϕ(λ), Eλ) a.e.<br />

if <strong>and</strong> only if R(limλ↑Ω ϕ(λ), limλ↑Ω Eλ). In particular, this implies that the relation<br />

“x ∈ E” transfers to the relation “x ∗ ∈ E ∗ ”. The key consideration is that we need<br />

to take the Ω-limit of all elements involved in the relation, not only numbers.<br />

Remark 4.4.5. Thanks to Theorem 4.4.1, we can give an intuitive idea of why we<br />

require the existence of non-wellfounded sets. We start by defining inductively the<br />

von Neumann natural numbers<br />

• 0 = ∅<br />

• n + 1 = n ∪ {n} = {0, . . . , n}<br />

From this definition, it’s easy to verify that limλ↑Ω |λ| = n(R) by showing that<br />

lim{0,<br />

. . . , |λ|} = {0, . . . , n(R)}<br />

λ↑Ω<br />

At this point, for all k ∈ N, we can set n(R) − k = limλ↑Ω |λ| − k, <strong>with</strong> the convention<br />

that n − k = ∅ whenever k > n. Now, we can exhibit an infinite ∈-descending chain:<br />

n(R) ∋ n(R) − 1 ∋ . . . ∋ n(R) − k ∋ n(R) − (k + 1) ∋ . . .<br />

<strong>and</strong> the existence of this chain is in open contrast <strong>with</strong> foundation axiom of ZFC set<br />

theory.<br />

We now collect in the next theorem several instances of the Transfer principle of<br />

basic set operations.<br />

Theorem 4.4.6. For all ϕ, ψ : Pfin(Ω) → Ω<br />

72<br />

1. ϕ(λ) ⊆ ψ(λ) a.e. ⇐⇒ limλ↑Ω ϕ(λ) ⊆ ψ(λ)<br />

2. η(λ) = ϕ(λ) ∪ ψ(λ) a.e. ⇐⇒ limλ↑Ω η(λ) = limλ↑Ω ϕ(λ) ∪ limλ↑Ω ψ(λ)<br />

3. η(λ) = ϕ(λ) ∩ ψ(λ) a.e. ⇐⇒ limλ↑Ω η(λ) = limλ↑Ω ϕ(λ) ∩ limλ↑Ω ψ(λ)<br />

4. η(λ) = ϕ(λ) \ ψ(λ) a.e. ⇐⇒ limλ↑Ω η(λ) = limλ↑Ω ϕ(λ) \ limλ↑Ω ψ(λ)


4.4. TRANSFER PRINCIPLE BY Ω-THEORY<br />

5. η(λ) = (ϕ(λ), ψ(λ)) a.e. ⇐⇒ limλ↑Ω η(λ) = (limλ↑Ω ϕ(λ), limλ↑Ω ψ(λ))<br />

6. η(λ) = ϕ(λ) × ψ(λ) a.e. ⇐⇒ limλ↑Ω η(λ) = limλ↑Ω ϕ(λ) × limλ↑Ω ψ(λ)<br />

Proof. Part 2 follows from the following chain of implications: if η(λ) = ϕ(λ) ∪ ψ(λ)<br />

a.e., then limλ↑Ω γ(λ) ∈ η(λ) if <strong>and</strong> only if γ(λ) ∈ η(λ) a.e. if <strong>and</strong> only if γ(λ) ∈ ϕ(λ)<br />

a.e. or γ(λ) ∈ ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω γ(λ) ∈ limλ↑Ω ϕ(λ) or limλ↑Ω γ(λ) ∈<br />

limλ↑Ω ψ(λ).<br />

Similarly, if η(λ) = ϕ(λ) ∩ ψ(λ) a.e., then limλ↑Ω γ(λ) ∈ η(λ) if <strong>and</strong> only if<br />

γ(λ) ∈ η(λ) a.e. if <strong>and</strong> only if γ(λ) ∈ ϕ(λ) a.e. <strong>and</strong> γ(λ) ∈ ψ(λ) a.e. if <strong>and</strong> only if<br />

limλ↑Ω γ(λ) ∈ limλ↑Ω ϕ(λ) <strong>and</strong> limλ↑Ω γ(λ) ∈ limλ↑Ω ψ(λ), <strong>and</strong> this proves part 3.<br />

The same argument can be used in the proof of part 4: if η(λ) = ϕ(λ)\ψ(λ) a.e.,<br />

then limλ↑Ω γ(λ) ∈ η(λ) if <strong>and</strong> only if γ(λ) ∈ η(λ) a.e. if <strong>and</strong> only if γ(λ) ∈ ϕ(λ)<br />

a.e. <strong>and</strong> γ(λ) ∈ ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω γ(λ) ∈ limλ↑Ω ϕ(λ) <strong>and</strong> limλ↑Ω γ(λ) ∈<br />

limλ↑Ω ψ(λ).<br />

To see that part 1 holds, notice that ϕ(λ) ⊆ ψ(λ) a.e. if <strong>and</strong> only if ϕ(λ)∪ψ(λ) =<br />

ψ(λ) a.e. if <strong>and</strong> only if limλ↑Ω ψ(λ) = limλ↑Ω ϕ(λ) ∪ limλ↑Ω ψ(λ) if <strong>and</strong> only if<br />

limλ↑Ω ϕ(λ) ⊆ limλ↑Ω ψ(λ).<br />

Part 5 is a consequence of the following chain of equalities<br />

lim(ϕ(λ),<br />

ψ(λ))<br />

λ↑Ω<br />

= lim{{ϕ(λ)},<br />

{ϕ(λ), ψ(λ)}}<br />

λ↑Ω<br />

<br />

<br />

= lim{ϕ(λ)},<br />

lim{ϕ(λ),<br />

ψ(λ)}<br />

λ↑Ω λ↑Ω<br />

<br />

<br />

= lim ϕ(λ)<br />

λ↑Ω<br />

, lim ϕ(λ), lim ψ(λ)<br />

λ↑Ω λ↑Ω<br />

=<br />

Part 6 follows from part 5 <strong>and</strong> by Lemma 4.4.3.<br />

<br />

<br />

lim ϕ(λ), lim ψ(λ)<br />

λ↑Ω λ↑Ω<br />

As a consequence of part 5 of Theorem 4.4.6, we deduce that Ω-limit is coherent<br />

<strong>with</strong> n-tuples.<br />

Corollary 4.4.7. For all n ∈ N,<br />

lim(ϕ1(λ),<br />

. . . , ϕn(λ)) =<br />

λ↑Ω<br />

<br />

<br />

lim ϕ1(λ), . . . , lim ϕn(λ)<br />

λ↑Ω λ↑Ω<br />

Proof. The proof is obtained by induction <strong>and</strong> by using part 5 of Theorem 4.4.6.<br />

Thanks to the previous results, we can show that n-ary relations <strong>and</strong> functions<br />

transfer under Ω-limit. Recall that a binary relation R is defined as a set of ordered<br />

pairs, <strong>and</strong> the writing R(a, b) or aRb st<strong>and</strong>s for (a, b) ∈ R. The domain <strong>and</strong> range<br />

of R are defined by<br />

dom(R) = {a : ∃b R(a, b)} <strong>and</strong> ran(R) = {b : ∃a R(a, b)}<br />

Theorem 4.4.8 (Transfer of binary relations). Rλ is a binary relation satisfying<br />

dom(Rλ) = Aλ <strong>and</strong> ran(Rλ) = Bλ for almost all λ ∈ Pfin(Ω) if <strong>and</strong> only if R =<br />

limλ↑Ω Rλ is a binary relation <strong>with</strong> dom(R) = A = limλ↑Ω Aλ <strong>and</strong> ran(R) = B =<br />

limλ↑Ω Bλ.<br />

73


CHAPTER 4. Ω-THEORY<br />

Proof. Without loss of generality, we can suppose Rλ = ∅, Aλ = ∅ <strong>and</strong> Bλ = ∅ for<br />

all λ ∈ Pfin(Ω).<br />

To prove the first implication, given an arbitrary function η(λ) ∈ Rλ for all<br />

λ ∈ Pfin(Ω), we have that η(λ) = (ϕ(λ), ψ(λ)) ∈ Aλ × Bλ for all λ ∈ Pfin(Ω)<br />

such that Rλ is a relation, hence for almost all λ. By transfer of ordered pairs,<br />

this implies limλ↑Ω η(λ) ∈ A × B. This proves that R is a binary relation <strong>with</strong><br />

dom(R) ⊆ A <strong>and</strong> ran(B) ⊆ B. To see that the inclusions are equalities, consider<br />

a function ϕ(λ) ∈ Aλ <strong>and</strong> a function ϕ(λ) ∈ Bλ for all λ such that Rλ is a binary<br />

relation. For those functions, we have<br />

<br />

<br />

lim ϕ(λ), lim ψ(λ)<br />

λ↑Ω λ↑Ω<br />

= lim<br />

λ↑Ω (ϕ(λ), ψ(λ)) ∈ R<br />

In particular, limλ↑Ω ϕ(λ) ∈ dom(R), hence the inclusion A ⊆ dom(R). The inclusion<br />

B ⊆ ran(R) can be proven <strong>with</strong> a similar argument.<br />

To show that the second implication holds, define Q = {λ ∈ Pfin(Ω) : Rλ is<br />

a binary relation}. For every λ ∈ Q, pick an element η(λ) ∈ Rλ which is not an<br />

ordered pair. If Q was not qualified, then by Theorem 4.4.6 limλ↑Ω η(λ) would not<br />

be an ordered pair, hence R would not be a binary relation. This proves that Q is<br />

qualified, <strong>and</strong> so Rλ is a binary relation a.e.<br />

Now, fix the family of sets {A ′ λ }λ∈Pfin(Ω) <strong>and</strong> {B ′ λ }λ∈Pfin(Ω) in a way such that<br />

A ′ λ = dom(Rλ) <strong>and</strong> B ′ λ = ran(Rλ) whenever λ ∈ Q. By the first implication of<br />

the theorem, we get dom(R) = limλ↑Ω A ′ λ <strong>and</strong> ran(R) = B′ λ<br />

but, by hypothesis, we<br />

already know dom(R) = limλ↑Ω Aλ <strong>and</strong> ran(R) = Bλ. By Transfer of equalities,<br />

this implies Aλ = A ′ λ <strong>and</strong> Bλ = B ′ λ a.e., so we conclude that dom(Rλ) = Aλ <strong>and</strong><br />

ran(Rλ) = Bλ is true almost everywhere, as desired.<br />

As the Transfer of ordered tuples follows from Transfer of ordered pairs, from<br />

Theorem 4.4.8 we deduce that n-ary relations transfer under Ω-limit.<br />

Corollary 4.4.9 (Transfer of n-ary relations). For all n ∈ N, Rλ is a n-ary relation<br />

<strong>with</strong> dom(Rλ) = Aλ <strong>and</strong> ran(Rλ) = Bλ for almost all λ ∈ Pfin(Ω) if <strong>and</strong> only if<br />

R = limλ↑Ω Rλ is a n-ary relation <strong>with</strong> dom(R) = limλ↑Ω Aλ <strong>and</strong> ran(R) = limλ↑Ω Bλ.<br />

Proof. The proof is obtained by induction from the one of Theorem 4.4.8.<br />

Another consequence of Theorem 4.4.8 is that functions transfer under Ω-limits.<br />

Theorem 4.4.10 (Transfer of functions). fλ is a function fλ : Aλ → Bλ for almost<br />

all λ ∈ Pfin(Ω) if <strong>and</strong> only if f = limλ↑Ω fλ : limλ↑Ω Aλ → limλ↑Ω Bλ is a function.<br />

Proof. For a matter of commodity, define A = limλ↑Ω Aλ <strong>and</strong> B = limλ↑Ω Bλ.<br />

Since a function can be seen as a particular relation, by Theorem 4.4.8 we<br />

conclude that, if fλ : Aλ → Bλ is a function for almost all λ ∈ Pfin(Ω), then<br />

f = limλ↑Ω fλ is a relation <strong>with</strong> dom(f) = limλ↑Ω Aλ <strong>and</strong> ran(f) = limλ↑Ω Bλ. It<br />

remains only to check that, for all a ∈ A, there exists only one b ∈ B such that<br />

(a, b) ∈ f. To do so, for a given a ∈ A consider a function ϕ(λ) ∈ Aλ for all λ<br />

such that fλ is a function, <strong>and</strong> satisfying limλ↑Ω ϕ(λ) = a. Now, suppose there are<br />

74


4.5. A MODEL FOR Ω-THEORY<br />

b = limλ↑Ω ψ(λ) <strong>and</strong> b ′ = limλ↑Ω ψ ′ (λ) such that (a, b) ∈ f <strong>and</strong> (a, b ′ ) ∈ f. By Theorems<br />

4.4.1 <strong>and</strong> 4.4.6, this means that both (ϕ(λ), ψ(λ)) ∈ fλ <strong>and</strong> (ϕ(λ), ψ ′ (λ)) ∈ fλ<br />

are true almost everywhere. By the fact that fλ are functions almost everywhere, we<br />

conclude ψ(λ) = ψ ′ (λ) almost everywere <strong>and</strong>, by Theorem 4.4.1, this ensures b = b ′ ,<br />

as we wanted.<br />

To prove the other implication, thanks to Theorem 4.4.8 it’s sufficient to prove<br />

that for almost all λ, if (aλ, bλ) ∈ fλ <strong>and</strong> (aλ, b ′ λ ) ∈ fλ, then bλ = b ′ λ . Suppose that<br />

this is not the case, <strong>and</strong> that for almost every λ there are aλ ∈ Aλ <strong>and</strong> bλ = b ′ λ ∈ Bλ<br />

satisfying (aλ, bλ) ∈ fλ <strong>and</strong> (aλ, b ′ λ ) ∈ fλ. By Theorem 4.4.8, we would then have<br />

(limλ↑Ω aλ, limλ↑Ω bλ) ∈ f <strong>and</strong> (limλ↑Ω aλ, limλ↑Ω b ′ λ ) ∈ f. Since bλ = b ′ λ a.e., also<br />

, but this contradicts the fact that f is a function.<br />

limλ↑Ω bλ = limλ↑Ω b ′ λ<br />

Notice that Theorem 4.4.10 ensures that Definition 4.3.3 of f ∗ is coherent <strong>with</strong><br />

the definition of f ∗ as the Ω-limit of its graph. With similar arguments as in the<br />

proof of Theorem 4.4.10, it can also be shown that f is injective (resp. surjective)<br />

if <strong>and</strong> only if fλ is injective (resp. surjective) for almost all λ.<br />

We want to explicitely observe that the only basic set operation that is not<br />

preserved under Ω-limits is the powerset. In Example 5.1.6, after we’ve formally<br />

defined what an “elementary property” is, we will give an intuitive explanation of<br />

why it is so.<br />

4.5 A model for Ω-<strong>Theory</strong><br />

Once again, in order to prove the consistency of Ω-<strong>Theory</strong>, we find opportune to<br />

exhibit a model for Ω-<strong>Theory</strong>. We recall that in Section 1.11 we obtained a model<br />

for Ω-<strong>Theory</strong> from a quotient of the ring Fun(Pfin(Ω), R) by an appropriate maximal<br />

ideal. To give a model for Ω-<strong>Theory</strong>, we want to extend this process to the<br />

set Fun(Pfin(Ω), Ω). This set doesn’t have always a natural structure of ring, like<br />

Fun(Pfin(Ω), R) had, so we will need to use a different approach that takes into<br />

account those differences. The idea is to use ultrapowers, an algebraic construction<br />

based on ultrafilters. Ultrapowers are, so to speak, the generalization to arbitrary<br />

sets of the concept of quotient of a ring by a maximal ideal. Before we can give a<br />

precise meaning to this intuitive statement, we find opportune to define a relation<br />

that extends the notion of two functions ϕ, ψ : Pfin(Ω) → R being equal on a<br />

qualified set. More formally, we give the following definition:<br />

Definition 4.5.1. Let E, F be two sets <strong>and</strong> U an ultrafilter on E. We define a<br />

relation ∼U on Fun(E, F ) in the following way: ϕ ∼U ψ if <strong>and</strong> only if Qϕ,ψ ∈ U.<br />

Notice that this definition is independent on the algebraic structure of Fun(E, F ).<br />

As usual, we leave to the reading the easy checkings that ∼U is an equivalence<br />

relation.<br />

In the same way as Definition 4.5.1 extends the notion of two functions being<br />

equal on a qualified set, the notion of ultrapower extends the notion of the quotient<br />

of a ring by a maximal ideal.<br />

75


CHAPTER 4. Ω-THEORY<br />

Definition 4.5.2. Let E, F be sets <strong>and</strong> U an ultrafilter on E. The ultrapower of F<br />

modulo U on E is the quotient set<br />

F E U = Fun(E, F )/ ∼U<br />

Sometimes, when the set E is fixed or when there will be no ambiguity, we will use<br />

the notation FU instead of F E U .<br />

Remark 4.5.3. Given an ultrapower F E U<br />

defined in the following way:<br />

J : Fun(E, F ) → F E U<br />

, there exists a “natural” map<br />

J(ϕ) = [ϕ] = {ψ ∈ Fun(E, F ) : ϕ ∼U ψ}<br />

In other words, J is the canonical projection from Fun(E, F ) to F E U<br />

surjective.<br />

. Clearly, J is<br />

If Fun(E, F ) is a ring, then the notion of ultrapower of F modulo U is equivalent<br />

to the one of quotient of Fun(E, F ) modulo the maximal ideal mU introduced in<br />

Proposition 1.10.14.<br />

Proposition 4.5.4. If Fun(E, F ) is a commutative ring <strong>and</strong> U is an ultrafilter on<br />

E, then<br />

F E U ≡ Fun(E, F )/mU<br />

Proof. By definition, F E U<br />

= {[ϕ] : ϕ ∈ Fun(E, F )}, where<br />

[ϕ] = {ψ ∈ Fun(E, F ) : ϕ ∼U ψ}<br />

On the other h<strong>and</strong>, Fun(E, F )/mU = {[[ϕ]] : ϕ ∈ Fun(E, F )}, where<br />

[[ϕ]] = {ψ ∈ Fun(E, F ) : ϕ − ψ ∈ mU}<br />

Explicitating the definition of mU, we see that<br />

[[ϕ]] = {ψ ∈ Fun(E, F ) : Qϕ−ψ,c0 ∈ U} = {ψ ∈ Fun(E, F ) : ϕ ∼U ψ} = [ϕ]<br />

<strong>and</strong> this proves that F E U = Fun(E, F )/mU as sets. Now, we only need to prove that,<br />

if ϕ, ψ ∈ Fun(E, F ), then the operations in F E U <strong>and</strong> in Fun(E, F )/mU are compatible<br />

or, in other words, that in F E U , [ϕ + ψ] = [ϕ] + [ψ]. By definition,<br />

[ϕ] + [ψ] = {ξ + ζ : Qξ,ϕ <strong>and</strong> Qζ,ψ ∈ U}<br />

Since U is an ultrafilter, we have that Qξ+ζ,ϕ+ψ = Qξ,ϕ∩Qζ,ψ ∈ U, so ξ+ζ ∈ [ϕ+ψ] or,<br />

in other words, [ϕ]+[ψ] ⊆ [ϕ+ψ]. On the other h<strong>and</strong>, it’s clear that ϕ+ψ ∈ [ϕ]+[ψ],<br />

<strong>and</strong> this implies [ϕ + ψ] ⊆ [ϕ] + [ψ]. We can conclude [ϕ + ψ] = [ϕ] + [ψ], as we<br />

wanted. A similar argument can be used to prove that [ϕ · ψ] = [ϕ] · [ψ], <strong>and</strong> this<br />

concludes the proof.<br />

76


4.5. A MODEL FOR Ω-THEORY<br />

In our concrete case, we want to obtain a model for Ω-<strong>Theory</strong> by considering an<br />

appropriate ultrapower of Ω modulo an ultrafilter on Pfin(Ω). To do so, we only<br />

need to define a suitable ultrafilter UΩ over Pfin(Ω) in a way that all the axioms of<br />

Ω-<strong>Theory</strong> are satisfied by the map J in the corresponding ultrapower ΩUΩ . We find<br />

opportune to summarize all the properties of Ω-<strong>Theory</strong> we need to satisfy, <strong>and</strong> the<br />

relative restrictions imposed on UΩ:<br />

• To satisfy Ω-Axiom 1, we don’t have to impose any property on UΩ: since the<br />

projection from Fun(Pfin(Ω), Ω) to ΩUΩ is surjective, the axiom will hold.<br />

• The same applies for Ω-Axiom 2: if ϕ(λ) is an atom for all λ ∈ Pfin(Ω), this<br />

means that ϕ : Pfin(Ω) → R, so its Ω-limit will belong to R ∗ , as desired. The<br />

same reasoning applies if ϕ(λ) is a set for all λ ∈ Pfin(Ω). It only remains<br />

to show that equation 4.1 holds, <strong>and</strong> we will impose its validity during the<br />

construction of the model. For a thorough presentation of the proof method<br />

we will use to show that Ω-Axiom 2 holds in the model, we rem<strong>and</strong> to Chapter<br />

6 of [1].<br />

• As for Ω-Axiom 3, by Theorem 1.9.9 it’s sufficient that UΩ is a nonprincipal<br />

fine ultrafilter.<br />

• In order to satisfy Ω-Axiom 4, it’s sufficient that, for all f ∈ Fun(E, F ) <strong>and</strong><br />

for all ϕ, ψ : Pfin(Ω) → E, if limλ↑Ω ϕ = limλ↑Ω ψ then [f ◦ ϕ] = [f ◦ ψ] as<br />

elements of ΩUΩ . This request is satisfied once we prove that Qf◦ϕ,f◦ψ ∈ Q,<br />

<strong>and</strong> this will follow from the hypothesis Qϕ,ψ ∈ Q.<br />

Putting everything together, we require only that UΩ is a nonprincipal fine ultrafilter.<br />

Theorem 4.5.5. Ω-Axioms 1, 2, 3 <strong>and</strong> 4 hold in the ultrapower ΩUΩ .<br />

Proof. Since the map J : Fun(Pfin(Ω), Ω) → ΩUΩ is surjective, Ω-Axiom 1 holds.<br />

Ω-Axiom 3 holds as a consequence of Theorem 1.9.9, Proposition 1.10.14 <strong>and</strong><br />

Proposition 4.5.4.<br />

As for Ω-Axiom 4, pick E, F ⊆ Ω <strong>and</strong> a function f : E → F . We need to prove<br />

that, if ϕ <strong>and</strong> ψ ∈ Fun(Pfin(Ω), E) satisfy [ϕ] = [ψ] as elements of ΩUΩ , then it’s also<br />

true that [f ◦ ϕ] = [f ◦ ψ] as elements of ΩUΩ . The hypothesis [ϕ] = [ψ] is equivalent<br />

to Qϕ,ψ ∈ UΩ, but this implies that also Qf◦ϕ,f◦ψ ∈ UΩ, so when we consider their<br />

class of equivalence in the ultrafilter, we obtain [f ◦ ϕ] = [f ◦ ψ].<br />

Now, we will impose that Ω-Axiom 2 holds in the model. We start by defining<br />

A = {x ∈ Ω : x is an atom} <strong>and</strong> by considering the projection<br />

J0 : Fun(Pfin(Ω), A) → AUΩ<br />

defined so that J0([cr]) = r for all r ∈ R. Notice that the request that the Ω-limit of<br />

an atom must be an atom translates to limλ↑Ω cA(λ) = A. This, in particular, means<br />

that the Ω-limit of constant functions must be a bijection between A <strong>and</strong> AUΩ , <strong>and</strong><br />

that every r ∈ R is a fixed point of the Ω-limit. To fulfill this condition, we impose<br />

that A satisfy the additional hypothesis |A| |Ω| = |A| (notice that this can be done<br />

77


CHAPTER 4. Ω-THEORY<br />

<strong>with</strong>out loss of generality by adding some “dummy” atoms that won’t be used in<br />

the working practice). From this hypothesis we deduce the following equation:<br />

|A| ≤ |AUΩ | ≤ |Fun(Pfin(Ω), A)| = |A| |Ω| = |A|<br />

<strong>and</strong>, in particular, that |A| = |AUΩ |. As a consequence, we can pick a bijection<br />

ψ0 : AUΩ → A such that ψ([ϕ]) = J0(ϕ) for all ϕ ∈ Fun(Pfin(Ω), R). This is<br />

sufficient to satisfy the first part of Ω-Axiom 2.<br />

At this point, for all ϕ ∈ Fun(Pfin(Ω), A), set limλ↑Ω ϕ(λ) = ψ0(ϕ). Now, in the<br />

model, we define<br />

lim ∅ = ∅<br />

λ↑Ω<br />

<strong>and</strong>, for all Bλ ⊆ A,<br />

lim Bλ =<br />

λ↑Ω<br />

<br />

<br />

lim ϕ(λ) : ϕ(λ) ∈ Bλ for all λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

<strong>and</strong> we continue defining the Ω-limit of sets by (possibly transfinite) induction. This<br />

definition ensures that the second part of Ω-Axiom 2 holds.<br />

78


Chapter 5<br />

The formal version of Transfer<br />

Principle<br />

It seems that if we want to be able<br />

to communicate at all, we have to<br />

adopt some common base, <strong>and</strong> it<br />

pretty well has to include logic.<br />

Douglas R. Hofstadter<br />

Probably, Transfer Principle is the most important theorem that Ω-<strong>Theory</strong><br />

can prove. For this reason, we find interesting to give it a precise formalization<br />

inside first order logic for set theory. To do so, we need to introduce<br />

the logic formalism for set theory <strong>and</strong> to give a rigorous definition the concept of<br />

“elementary property”. Then, after some interesting examples, we will finally show<br />

that Ω-<strong>Theory</strong> proves that every elementary property is preserved under Ω-limit.<br />

5.1 Logic formalism for set theory<br />

It’s a well known fact that the theory of real numbers can be developed inside set<br />

theory: all we need to describe their properties <strong>and</strong> behaviours is the language<br />

of set theory <strong>and</strong>, in particular, the language of first-order logic plus the equality<br />

<strong>and</strong> membership relations. We begin the introduction of the logic formalism for set<br />

theory by specifying the “alphabet” of symbols of the first-order language of set<br />

theory, which will be used to construct our formulas.<br />

• First of all, we require that our alphabet has a denumerable supply of variables,<br />

that we will denote by x, y, z, . . . , x1, x2, . . .<br />

• Then, we ask that our language has all the logic connectives of predicate logic:<br />

– Negation: ¬ (not).<br />

– Conjunction: ∧ (<strong>and</strong>).<br />

– Disjunction: ∨ (or).<br />

79


CHAPTER 5. THE FORMAL VERSION OF TRANSFER PRINCIPLE<br />

– Implication: ⇒ (if . . . then).<br />

– Double implication: ⇔ (if <strong>and</strong> only if).<br />

• In the same spirit, we ask that our language has all the quantifiers of first<br />

order logic:<br />

– Existential quantifier: ∃ (there exists).<br />

– Universal quantifier: ∀ (for all).<br />

• We require also that our language has the equality symbol =.<br />

• Since this is the alphabet of the language of set theory, we need also the symbol<br />

∈ of membership.<br />

• At last, we include the parentheses “(” <strong>and</strong> “)”.<br />

Using this language, we can now give the formal definition of an elementary<br />

formulas. As <strong>with</strong> many other definitions in logic, the elementary formulas will be<br />

defined as the transitive closure of the elementary atomic formulas under certain<br />

operations.<br />

Definition 5.1.1. An elementary formula is a finite string of symbols in the above<br />

language in which variables are distinguished into free variables <strong>and</strong> bound variables,<br />

<strong>and</strong> which are obtained according to the following rules:<br />

EF 1 (Atomic Formulas). Let x <strong>and</strong> y be variables. Then<br />

(x = y), (x ∈ y)<br />

are elementary formulas, named atomic formulas, where:<br />

• the free variables are x <strong>and</strong> y;<br />

• there are no bound variables.<br />

EF 2 (Restricted quantifiers). Let σ be an elementary formula <strong>and</strong> assume that<br />

1. the variable x is free in σ,<br />

2. the variable y is not bound in σ.<br />

Then also<br />

are elementary formulas, where<br />

80<br />

(∀x ∈ y) σ, (∃x ∈ y) σ<br />

• the free variables are y along <strong>with</strong> the free variables of σ;<br />

• the bound variables are x along <strong>with</strong> the bound variables of σ.


5.1. LOGIC FORMALISM FOR SET THEORY<br />

EF 3 (Negation). Let σ be an elementary formula. Then also<br />

is an elementary formula, where<br />

(¬σ)<br />

• the free variables are the same as in σ;<br />

• the bound variables are the same as in σ.<br />

EF 4 (Binary logic connectives). Let σ <strong>and</strong> τ be elementary formulas, <strong>and</strong> assume<br />

that:<br />

1. every free variable in σ is not bound in τ;<br />

2. every free variable in τ is not bound in σ.<br />

Then also<br />

are elementary formulas, where<br />

(σ ∧ τ), (σ ∨ τ), (σ ⇒ τ), (σ ⇔ τ)<br />

• the free variables are union of the free variables of σ <strong>and</strong> the free variables of<br />

τ;<br />

• the free variables are union of the bound variables of σ <strong>and</strong> the bound variables<br />

of τ.<br />

A few comments are in order. Clearly, in the inductive process of forming an<br />

elementary formula, the first rule to be applied is EF1: in other words, every formula<br />

is built on atomic formulas (<strong>and</strong> this justifies the name “atomic”). An arbitrary<br />

formula is then obtained from atomic formulas by finitely many iterations of the<br />

other three rules, in whatever order.<br />

The restricted quantifiers rule EF2 is the only one that produces bound variables.<br />

The idea is that a variable is bound when it is quantified. We remark that a variable<br />

can be quantified only if it is free in the given formula, i.e. only if it actually appears<br />

<strong>and</strong> it has been not quantified already.<br />

It is worth remarking that quantifications are only permitted in the restricted<br />

forms (∀x ∈ y) or (∃x ∈ y), where the “scope” of the quantified variable x is<br />

“restricted” by another variable y. In order to avoid potential ambiguities, it is also<br />

required that the “bounding” variable y do not appear bound itself in the given<br />

formula.<br />

The negation rule EF3 is the simplest one, as it applies to all formulas <strong>with</strong>out<br />

any restrictions, <strong>and</strong> it produces new formulas having the same free <strong>and</strong> bound<br />

variables as the starting ones.<br />

Also the binary connectives rule EF4 is quite simple, <strong>and</strong> it produces formulas<br />

where the free <strong>and</strong> the bound variables are simply obtained by putting together the<br />

free <strong>and</strong> the bound variables of the two given formulas, respectively. The provisions<br />

in rule EF4 are given to prevent “connecting” two formulas where a same variable<br />

appears free in one of them, <strong>and</strong> bound in the other.<br />

For a matter of commodity, we will follow the common use <strong>and</strong> adopt familiar<br />

short-h<strong>and</strong>s to simplify notation. For instance:<br />

81


CHAPTER 5. THE FORMAL VERSION OF TRANSFER PRINCIPLE<br />

• we will write “x = y” instead of “¬(x = y)”;<br />

• we will write “x ∈ y” instead of “¬(x ∈ y)”;<br />

• we will write “∀x1, . . . , xn ∈ y σ” instead of “(∀x1 ∈ y) . . . (∀xn ∈ y) σ”;<br />

• we will write “∃x1, . . . , xn ∈ y σ” instead of “(∃x1 ∈ y) . . . (∃xn ∈ y) σ”.<br />

Moreover, we will take the freedom of using parentheses informally, <strong>and</strong> omit some<br />

of them whenever confusion is unlikely. So, we may write “∀x ∈ y σ” instead of the<br />

correct “(∀x ∈ y) σ”; or “σ ∧ τ” instead of “(σ ∧ τ)”, <strong>and</strong> so on.<br />

Another agreement is that the negations ¬ bind more strongly than conjunctions<br />

∧ <strong>and</strong> disjunctions ∨, which in turn bind more strongly than implications ⇒ <strong>and</strong><br />

double implications ⇔. So, for example,<br />

• we may write “¬σ ∧ τ” instead of “((¬σ) ∧ τ)”;<br />

• we may write “¬σ ∧ τ ⇒ ρ” instead of “(((¬σ) ∧ τ) ⇒ ρ)”;<br />

• we may write “σ ⇒ τ ∧ ρ” instead of “(σ ⇒ (τ ∧ ρ))”.<br />

An important notation is the following: when writing<br />

σ(x1, . . . , xn)<br />

we will mean that x1, . . . , xn are all <strong>and</strong> only the free variables that appear in the<br />

formula σ. The intuition is that the truth or falsity of a formula depends only on the<br />

values given to its free variables, whereas bound variables can be renamed <strong>with</strong>out<br />

changing the meaning of the formula.<br />

We are now ready for the fundamental definition:<br />

Definition 5.1.2 (Property expressed in elementary form). A property of mathematical<br />

objects A1, . . . , An is expressed in elementary form if it is written down by<br />

taking an elementary formula σ(x1, . . . , xn) <strong>and</strong> <strong>and</strong> by replacing all occurrences of<br />

each free variable xi <strong>with</strong> Ai. In this case, we will denote<br />

σ(A1, . . . , An)<br />

<strong>and</strong> the objects A1, . . . , An are referred to as constants.<br />

By a slight abuse, we will often say elementary property instead of the correct<br />

“property expressed in elementary form”.<br />

The motivation of our definition is the well-known fact that virtually all properties<br />

considered in mathematics can be formulated in elementary form. Below is a<br />

list of examples that include the fundamental ones.<br />

Example 5.1.3. Each property is followed by one of its possible expressions in<br />

elementary form. For simplicity, in each item we use short-h<strong>and</strong>s for properties that<br />

have been already expressed in elementary form in a previous case.<br />

82<br />

• “A ⊆ B”: (∀x ∈ A)(x ∈ B);


5.1. LOGIC FORMALISM FOR SET THEORY<br />

• “C = A ∪ B”: (A ⊆ C) ∧ (B ⊆ C) ∧ (∀x ∈ C)(x ∈ A ∨ x ∈ B);<br />

• “C = A ∩ B”: (C ⊆ A) ∧ (∀x ∈ A)(x ∈ B ⇔ x ∈ C);<br />

• “C = A \ B”: (C ⊆ A) ∧ (∀x ∈ A)(x ∈ B ⇔ x ∈ C);<br />

• “C = {a1, . . . , an}”: (a1 ∈ C)∧. . .∧(an ∈ C)∧(∀x ∈ C)(x = a1∨. . .∨x = an);<br />

• “{a1, . . . , an} ∈ C”: (∃x ∈ C)(x = {a1, . . . , an});<br />

• “C = (a, b)”: ({a} ∈ C) ∧ ({a, b} ∈ C) ∧ (∀x ∈ C)(x = {a} ∨ x = {a, b});<br />

• In the same way, the formula for “C = (a1, . . . , an)” can be obtained from the<br />

previous one by using the recursive definition of ordered n-tuple;<br />

• “(a1, . . . , an) ∈ C”: (∃x ∈ C)(x = (a1, . . . , an));<br />

• “C = A1 × . . . × An”: (∀x1 ∈ A) . . . (∀xn ∈ An)((x1, . . . , xn) ∈ C) ∧ (∀z ∈<br />

C)(∃x1 ∈ A1) . . . (∃xn ∈ An)(z = (x1, . . . , xn));<br />

• “R is a n-place relation on A”: (∀z ∈ R)(∃x1, . . . , xn ∈ A)(z = (x1, . . . , xn));<br />

• “f : A → B”: (f ⊆ A × B) ∧ (∀a ∈ A)(∃b ∈ B)((a, b) ∈ f) ∧ (∀a ∈ A)(∀b, b ′ ∈<br />

B)((a, b), (a, b ′ ) ∈ f ⇒ b = b ′ );<br />

• “f(a1, . . . , an) = b”: ((a1, . . . , an), b) = (a1, . . . , an, b) ∈ f;<br />

• “x < y in R”: (x, y) ∈ R where R ⊆ R × R is the order relation on R.<br />

Remark 5.1.4. When expressing a property in elementary form, quantifiers can<br />

only be used in the restricted forms:<br />

“(∀x ∈ A) σ(x, . . .)” <strong>and</strong> “(∃x ∈ X) σ(x, . . .)”<br />

So, whenever quantifying on a variable x, it must be always specified the set A where<br />

the variable x is ranging. Two crucial examples are the following:<br />

• Quantification ranging on subsets:<br />

must be reformulated<br />

respectively;<br />

• Quantification ranging on functions:<br />

must be reformulated<br />

respectively.<br />

“∀x ⊆ X . . . ” or “∃x ⊆ X . . . ”<br />

“∀x ∈ P(X) . . . ” <strong>and</strong> “∃x ∈ P(X) . . . ”<br />

“∀f : A → B . . . ” or “∃f : A → B . . . ”<br />

“∀f ∈ Fun(A, B) . . . ” <strong>and</strong> “∃f ∈ Fun(A, B) . . . ”<br />

83


CHAPTER 5. THE FORMAL VERSION OF TRANSFER PRINCIPLE<br />

Example 5.1.5. An elementary formulation of the completeness property of the<br />

reals requires the powerset P(R) as a constant:<br />

∀X ∈ P(R)(“X bounded above” ⇒ (∃x ∈ R)“x = sup A”)<br />

Where “X bounded above” <strong>and</strong> “x = sup A” are the short-h<strong>and</strong>s for the corresponding<br />

expressions in elementary form:<br />

• (∃x ∈ R)(∀y ∈ X)(y < x), <strong>and</strong><br />

• (∀y ∈ X)(y ≤ x) ∧ (∀z ∈ R)((∀t ∈ X)(t ≤ z)) ⇒ x ≤ z,<br />

respectively.<br />

It is worth remarking that a same property may be expressed both in an elementary<br />

form <strong>and</strong> in a non-elementary form. The typical example involves the powerset<br />

operation.<br />

Example 5.1.6. “P(A) = B” is trivially an elementary property of constants P(A)<br />

<strong>and</strong> B, but cannot be formulated as an elementary property of constants A <strong>and</strong><br />

B. In fact, while the inclusion B ⊆ P(A) is formalized in elementary form by<br />

“(∀x ∈ B)(∀y ∈ x)(y ∈ A)”, the other inclusion P(A) ⊆ B does not admit any<br />

elementary formulation <strong>with</strong> A <strong>and</strong> B as constants. The point is that, under our<br />

rules, quantifications over subsets “(∀x ⊆ A)(x ∈ B)” are not allowed.<br />

5.2 Transfer Principle<br />

Finally, we have everything we need to show that the Transfer Principle holds in the<br />

Ω-<strong>Theory</strong> or, in other words, that all properties expressed in elementary form are<br />

preserved under Ω-limits. This theorem is the fundamental result of Ω-<strong>Theory</strong>.<br />

Theorem 5.2.1 (Transfer Principle). Let σ(x1, . . . , xn) be an elementary formula<br />

<strong>and</strong> ϕ1, . . . , ϕn : Pfin(Ω) → Ω. Then<br />

<br />

<br />

σ(ϕ1(λ), . . . ϕn(λ)) a.e. ⇐⇒ σ lim ϕ1(λ), . . . , lim ϕn(λ)<br />

λ↑Ω λ↑Ω<br />

Proof. We proceed by induction on the complexity of the formulas. For atomic<br />

formulas (x1 = x2) <strong>and</strong> (x1 ∈ x2), the thesis is given by Theorem 4.4.1.<br />

The negation case ¬σ follows from the inductive hypothesis on σ <strong>and</strong> the property<br />

that a set is qualified if <strong>and</strong> only if its complement is not qualified. Formally, we<br />

have the following chain of equivalences:<br />

¬σ(ϕ1(λ), . . . , ϕn(λ)) a.e. ⇐⇒ σ(ϕ1(λ), . . . , ϕn(λ)) not a.e.<br />

⇐⇒ not σ (limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕn(λ))<br />

⇐⇒ ¬σ (limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕn(λ))<br />

The conjunction case (σ ∧ σ ′ ) is proved similarly as above, by using the property<br />

that two sets are both qualified if <strong>and</strong> only if their intersection is qualified. The<br />

cases of the other binary connectives (σ ∨ σ ′ ) <strong>and</strong> (σ ⇒ σ ′ ) <strong>and</strong> (σ ⇔ σ ′ ) can then<br />

be proved by using the tautologies<br />

84


• (σ ∨ σ ′ ) ≡ ¬(¬σ ∧ ¬σ ′ );<br />

• (σ ⇒ σ ′ ) ≡ (¬σ ∨ σ ′ );<br />

• (σ ⇔ σ ′ ) ≡ (σ ⇒ σ ′ ) ∧ (σ ′ ⇒ σ).<br />

5.2. TRANSFER PRINCIPLE<br />

The only cases left are the existential <strong>and</strong> unversal quantifiers. We will prove<br />

the inductive step for the former, since the latter can be treated by considering the<br />

logic equivalence<br />

(∀x ∈ y)σ ≡ ¬(∃x ∈ y)¬σ<br />

First, assume that the set Q = {λ : ∃x ∈ ψ(λ) τ(x, ϕ1(λ), . . . , ϕn(λ))} is qualified.<br />

We can then find a function η : Pfin(Ω) → Ω satisfying the conditions that for<br />

all λ ∈ Q, both η(λ) ∈ ψ(λ) <strong>and</strong> τ(η(λ), ϕ1(λ), . . . , ϕn(λ)) are true. Then, by the<br />

inductive hypothesis, we have<br />

lim η(λ) ∈ lim ψ(λ) <strong>and</strong> τ<br />

λ↑Ω λ↑Ω<br />

from which we deduce immediately<br />

∃x ∈ lim<br />

λ↑Ω ψ(λ) τ<br />

<br />

<br />

lim η(λ), lim ϕ1(λ), . . . , lim ϕn(λ)<br />

λ↑Ω λ↑Ω λ↑Ω<br />

<br />

x, lim<br />

λ↑Ω ϕ1(λ), . . . , lim<br />

λ↑Ω ϕn(λ)<br />

Conversely, if ∃x ∈ limλ↑Ω ψ(λ) τ (x, limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕn(λ)) then, by Transfer<br />

of membership, there exists η : Pfin(Ω) → Ω such that η(λ) ∈ ψ(λ) a.e. <strong>and</strong><br />

τ (limλ↑Ω η(λ), limλ↑Ω ϕ1(λ), . . . , limλ↑Ω ϕn(λ)). By the inductive hypothesis, we deduce<br />

τ(η(λ), ϕ1(λ), . . . , ϕn(λ)) a.e., <strong>and</strong> this concludes the proof.<br />

As a direct consequence, we obtain that every property expressed in elementary<br />

form holds for given objects A1, . . . An if <strong>and</strong> only if it holds for the corresponding<br />

Ω-limits limλ↑Ω cA1(λ), . . . , limλ↑Ω cAn(λ). For a matter of commodity, if we define<br />

limλ↑Ω cA(λ) = A ∗ , then this result can be stated as follows:<br />

Corollary 5.2.2 (Transfer principle for hyper-images). Let σ(x1, . . . , xn) an elementary<br />

formula. Then, for all A1, . . . , An,<br />

σ(A1, . . . , An) ⇐⇒ σ(A ∗ 1, . . . , A ∗ n)<br />

Proof. This result follows from Theorem 5.2.1 <strong>and</strong> by noticing that, for all λ ∈<br />

Pfin(Ω),<br />

σ(A1, . . . , An) ⇐⇒ σ (cA1(λ), . . . , cAn(λ))<br />

In Example 5.1.3, we have seen that C = A ∪ B, C = A ∩ B, C = A \ B,<br />

C = (a, b), C = A × B can all be expressed in elementary form. So, by Transfer<br />

Principle, we obtain again the thesis of Theorem 4.4.6.<br />

Similarly, the reader could re-prove all preservation properties under Ω-limits<br />

presented in Chapter 4, by simply putting them in elementary form <strong>and</strong> then applying<br />

Transfer. It is worth remarking that restricting to elementary sentences is not<br />

<br />

85


CHAPTER 5. THE FORMAL VERSION OF TRANSFER PRINCIPLE<br />

a limitation, because all mathematical properties can be rephrased in elementary<br />

terms.<br />

Plenty of applications of Transfer Principle can be easily conceived by the reader.<br />

Some of them are given below as examples.<br />

Example 5.2.3. If (A, 1”. Notice that the above sentence does not express the Archimedean<br />

property of R ∗ , because the hypernatural number ν could be infinite.<br />

By using correctly Transfer Principle, the interested reader could easily prove<br />

the following Lemma, that will be used in Section 7.5:<br />

Lemma 5.2.6. If ξ ∈ R ∗ , ξ ∼ 0 <strong>and</strong> ξ = 0, then there exists ν ∈ N ∗ satisfying<br />

1/ν < ξ.<br />

An interesting example of the use of this Lemma is given by the density of Q ∗<br />

in R ∗ :<br />

Example 5.2.7. Thanks to Lemma 5.2.6, one can deduce the density of Q ∗ in R ∗<br />

<strong>with</strong> respect to the metric given by d ∗ (x, y) = |x − y|. Notice that this metric is the<br />

natural extension of the metric d : R × R → R.<br />

86


Chapter 6<br />

An application: Non-Archimedean<br />

Probability<br />

The theory of probability as a<br />

mathematical discipline can <strong>and</strong><br />

should be developed from axioms<br />

in exactly the same way as<br />

geometry <strong>and</strong> algebra.<br />

A. Kolmogorov<br />

One among the many interesting application of the Ω-<strong>Theory</strong> seems to be<br />

that of Non-Archimedean Probability (that sometimes we will call NAP).<br />

There are many approaches to the topic, <strong>and</strong> we signal [12], that use<br />

the tools of Ω-limit to construct suitable NAP spaces that model some interesting<br />

theorical situations (e.g. a fair lottery over N, Q <strong>and</strong> R among the others).<br />

We already have a theory of Ω-limit, so we will follow a slightly different route:<br />

in the first section, we will recall the axiomatic formulation of Kolmogorov’s probability<br />

theory <strong>and</strong> then, after a brief discussion of its weaknesses, we will give an<br />

axiomatization of Non-Archimedean Probability that uses the Ω-limit as a replacement<br />

for the Conditional Probability Principle. Once we have defined the axioms<br />

of NAP, we will begin to study their consequences <strong>and</strong> show some interesting examples.<br />

In particular, we will show that we can interpret the idea of NAP spaces inside<br />

Ω-<strong>Theory</strong>.<br />

6.1 Kolmogorov’s axioms of probability<br />

We start by recalling the definition of σ-algebra <strong>and</strong> then by giving an axiomatic<br />

formulation of a probability over a generic set Υ that is equivalent to Kolmogorov’s<br />

original axioms. The set Υ is usually called the sample space, <strong>and</strong> its elements<br />

represent “elementary events”. Contrary to common use, we will not denote the<br />

sample space by the letter Ω, since in our theory Ω is already used <strong>with</strong> a different<br />

meaning.<br />

Definition 6.1.1. A(Υ) is a σ-algebra over Υ if <strong>and</strong> only if<br />

87


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

• A(Υ) ⊆ P(Υ);<br />

• Υ ∈ A(Υ);<br />

• A ∈ A(Υ) implies A c ∈ A(Υ);<br />

• For every family {Ai}i∈N, Ai ∈ A(Υ) for all i ∈ N implies <br />

i∈I Ai ∈ A(Υ);<br />

• If A1, A2 ∈ A(Υ), then A1 ∩ A2 ∈ A(Υ).<br />

An element A ∈ A(Υ) is called an event. If A is a singleton, then it’s called an<br />

elementary event.<br />

KA 1 (Domain <strong>and</strong> Range). The events are the elements of A(Υ), a σ-algebra over<br />

Υ, <strong>and</strong> the probability is a function<br />

P : A(Υ) → R<br />

KA 2 (Positivity). For all A ∈ A(Υ), P (A) ≥ 0.<br />

KA 3 (Normalization). P (Υ) = 1.<br />

KA 4 (Additivity). For all A, B ∈ A(Υ), if A ∩ B = ∅, then P (A ∪ B) = P (A) +<br />

P (B).<br />

Definition 6.1.2. For all A, B ∈ A(Υ), B = ∅, we define the conditional probability<br />

P (A|B) by<br />

P (A ∩ B)<br />

P (A|B) =<br />

P (B)<br />

KA 5 (Conditional Probability Principle). Let {An}n∈N be a family of events satisfying<br />

the two conditions An ⊆ An+1 for all n ∈ N <strong>and</strong> Υ = <br />

n∈N An. Then,<br />

eventually P (An) > 0 <strong>and</strong>, for any B ∈ A(Υ),<br />

P (B) = lim<br />

n→∞ P (B|An)<br />

An immediate consequence of those axioms are, for instance, the fact that 0 ≤<br />

P (A) ≤ 1 for all A ∈ A(Υ), the monotony of probability <strong>and</strong> countable additivity.<br />

Example 6.1.3. If the sample space is finite, then it’s usual to assume A(Υ) =<br />

P(Υ). In this case, to define a probability over P(Υ) it’s sufficient to define a<br />

normalized probability function on the elementary events, namely a function p :<br />

Υ → R satisfying <br />

pυ = 1<br />

υ∈Υ<br />

Once p is given, we can define a probability function P : P(Υ) → [0, 1] by posing<br />

P (A) = <br />

p(υ) (6.1)<br />

υ∈A<br />

It’s trivial to check that P satisfies all of Kolmogorov’s axioms.<br />

88


6.2. NON-ARCHIMEDEAN PROBABILITY<br />

Remark 6.1.4. Equation 6.1 cannot be generalized to the infinite case. In fact, if<br />

Υ = R, then an infinite sum is well defined only if the set {x ∈ R : p(x) = 0} is<br />

denumerable or finite.<br />

Remark 6.1.5. The choiche of the range of probability together <strong>with</strong> countable<br />

additivity may lead to some problems when the sample space is infinite.<br />

A peculiarity of Kolmogorov’s probability is that, in general, A(Υ) = P(Υ). The<br />

most known example is the probability measure given by Lebesgue measure, which<br />

cannot be defined for all the sets in P(Υ). This seems an innocent feature, but it’s<br />

equivalent to say that there are sets in P(R) which are not events (namely elements<br />

of A(R)), even when they are the union of elementary events.<br />

Another weak point for Kolmogorov’s probability is the interpretation of P (A) =<br />

0 <strong>and</strong> P (A) = 1. Explicitely, there are plenty of situations in which we can find<br />

some sets {Ai}i∈I such that, for all i ∈ I,<br />

<strong>and</strong><br />

P<br />

P (Ai) = 0 (6.2)<br />

<br />

i∈I<br />

Ai<br />

<br />

= 1 (6.3)<br />

This situation is very common when I is not denumerable. In this case, it looks as<br />

if equation 6.2 states that each event Ai is impossible, whereas equation 6.3 states<br />

that one of them will certainly occur.<br />

This situation requires further reflection: Kolmogorov’s probability theory works<br />

well as a mahematical theory, but the direct interpretation of its language leads to<br />

counterintuitive results such as the one described above. An obvious solution is to<br />

discard the the intuitive ideas “if P (A) = 0, then A is an impossible event” <strong>and</strong> “if<br />

P (A) = 1, then A is a certain event”. Instead, we could say that “if P (A) = 0, then<br />

A is a very unlikely event”, <strong>and</strong> “if P (A) = 1, then A is an almost certain event”, but<br />

this solution has as a consequence that the correspondence between mathematical<br />

formulas <strong>and</strong> their interpretation is now quite vague <strong>and</strong> far from intuition.<br />

6.2 Non-Archimedean probability<br />

There is a chance for everyone.<br />

Helloween<br />

Up to now, we have only identified some problems of Kolmogorov’s probability,<br />

but we don’t have any ideas about how to overcome them. A useful insight will<br />

come by studying the example of fair lotteries.<br />

Definition 6.2.1 (Fair Lotteries). A fair lottery over Υ is a probability function P<br />

that mirrors the intuitive properties we expect:<br />

1. since every “ticket” must have a probability to be extracted, P must be defined<br />

over all P(Υ);<br />

89


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

2. the probability to extract every “ticket” is the same, so P ({υ}) does not depend<br />

on the choiche of υ ∈ Υ;<br />

3. we expect to calculate the probability of an arbitrary event by a process of<br />

summing over the individual tickets: P (A) = <br />

υ∈A P ({υ}).<br />

Example 6.2.2 (Finite fair lotteries). Suppose that Υ = {1, . . . , n} ⊂ N. Then, a<br />

fair lottery on Υ is described by a function P satisfying<br />

<strong>and</strong> therefore<br />

P ({υ}) = 1<br />

|Υ|<br />

P (A) = |A|<br />

n<br />

= 1<br />

n<br />

Let’s now focus on the fair lottery over N. If we want to follow an approach similar<br />

to the one used in the finite case, we need to assume that the range of the probability<br />

function has to include infinitesimals. Moreover, if we want to extrapolate the<br />

intuitions concerning finite lotteries to infinite ones, we need a tool that allows to<br />

generalize to the infinite case properties that hold for the finite case. In our theory,<br />

we have already seen that, thanks to Transfer Principle, Ω-limit has such properties,<br />

so we will use it in order to define the behaviour of Non-Archimedean Probability.<br />

Now, we want to suggest a new approach at probability, that solves both problems<br />

mentioned in Remark 6.1.5. The two main ideas are the following:<br />

1. we use a non-Archimedean field as the range of probabilities: this allows to<br />

make clear the ideas of “very unlikely” <strong>and</strong> “almost certain” events;<br />

2. we use Ω-limit to give sense to probabilities of infinite sets. In particular, this<br />

will ensure that the probability will be defined on all subsets of Ω <strong>and</strong> will<br />

satisfy some good properties that hold in the finite case.<br />

More formally, we will begin by giving new axioms for the Non-Archimedean Probability<br />

<strong>with</strong>in the context of Ω-<strong>Theory</strong>, <strong>and</strong> then we will show how those axioms<br />

allow to solve the mentioned problems of Kolmogorov probability. In what follows,<br />

we will always suppose Ω ⊇ Υ.<br />

NAP 1 (Domain <strong>and</strong> Range). The events are all the events of P(Υ) <strong>and</strong> the probability<br />

is a function<br />

P : P(Υ) → R ∗<br />

NAP 2 (Positivity). For all A ∈ P(Υ), P (A) ≥ 0.<br />

NAP 3 (Normalization). For all A ∈ P(Υ),<br />

P (A) = 1 ⇔ A = Υ<br />

NAP 4 (Additivity). For all A, B ∈ P(Υ), if A ∩ B = ∅, then P (A ∪ B) =<br />

P (A) + P (B).<br />

90


6.3. FIRST CONSEQUENCES OF NAP AXIOMS<br />

NAP 5 (Non-Archimedean Continuity). For all λ ∈ Pfin(Υ), λ = ∅, P (A|λ) ∈ R,<br />

<strong>and</strong><br />

P (A) = lim<br />

λ↑Υ P (A|λ)<br />

Before we continue, we want to remark that the axioms of non-Archimedean<br />

probability imply that P ({υ}) = 0 for all υ ∈ Υ. In fact, let’s suppose that this is<br />

not the case, so that there is υ0 ∈ Υ such that P ({υ0}) = 0. NAP3 ensures that<br />

P (Υ) = 1, NAP4 implies that P ({υ0}) + P ({υ0} c ) = 1, <strong>and</strong> we immediately deduce<br />

that P ({υ0} c ) = 1, but this is in contradiction <strong>with</strong> NAP3. What is the meaning<br />

of the condition P ({υ}) > 0 for all υ ∈ Υ? Let’s pause for a moment <strong>and</strong> think<br />

about an example: suppose we want to calculate the probability to roll a specific<br />

number on a six sided die. In this case, Υ = {1, 2, 3, 4, 5, 6}, as we already know that,<br />

for instance, number 7 will never occurr. This simple example can be extended to<br />

arbitrary probability. What we impose <strong>with</strong> NAP3 is that every elementary event<br />

of Υ has a positive probability to occurr, <strong>and</strong> we choose to ignore all events we<br />

already know as impossible. In this way, NAP allows to regain the intuition that,<br />

if “P (A) = 0”, then A is truly an impossible event (<strong>and</strong> hence can be omitted from<br />

the sample space).<br />

This is but one of the differences between NAP <strong>and</strong> Kolmogorov’s probability.<br />

If the reader is interest in a thorough comparison between Kolmogorov axioms <strong>and</strong><br />

NAP axioms, we rem<strong>and</strong> to [12].<br />

6.3 First consequences of NAP axioms<br />

As a first consequence of NAP axioms, we want to show how a probability function<br />

P over an arbitrary set Υ can be interpreted. The goal is to achieve a formula that<br />

extends in some sense equation 6.1.<br />

Given a Non-Archimedean probability function P : P(Υ) → R ∗ , we begin by<br />

defining a function p : Υ → R ∗ as follows:<br />

p(υ) = P ({υ})<br />

then, we choose an arbitrary point υ0 ∈ Υ <strong>and</strong> define a weight function w : Υ → R ∗<br />

in the following way:<br />

w(υ) = p(υ)<br />

p(υ0)<br />

This weight function will allow us to get the desired interpretation of P . The first<br />

step toward this result is to prove that the range of w is a subset of R:<br />

Lemma 6.3.1. The function w takes values in R.<br />

Proof. Take υ ∈ Υ arbitrarily <strong>and</strong> set r = P ({υ}|{υ, υ0}). By NAP 5, r ∈ R <strong>and</strong><br />

r < 1. By definition of conditional probability, we have also<br />

r =<br />

p(υ)<br />

p(υ) + p(υ0)<br />

= w(υ)<br />

w(υ) + 1<br />

91


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

so we conclude<br />

w(υ) = r<br />

1 − r<br />

The previous Lemma allows to express the probability function P by using the<br />

weight function w.<br />

∈ R<br />

Lemma 6.3.2. For all λ ∈ Pfin(Ω) such that λ ∩ Υ = ∅,<br />

P (A|λ ∩ Υ) =<br />

<br />

υ∈λ∩A w(υ)<br />

<br />

υ∈λ∩Υ w(υ)<br />

Proof. Notice that, by the finiteness of λ <strong>and</strong> by definition of w, we have<br />

P (A|λ ∩ Υ) =<br />

<br />

υ∈λ∩A p(υ)<br />

<br />

υ∈λ∩Υ<br />

p(υ) =<br />

<br />

υ∈λ∩A p(υ0)w(υ)<br />

<br />

υ∈λ∩Υ<br />

p(υ0)w(υ) =<br />

<br />

υ∈λ∩A w(υ)<br />

<br />

υ∈λ∩Υ w(υ)<br />

(6.4)<br />

Now, we’d like to take the Ω-limit of equation<br />

<br />

6.4 to get the desired result. The<br />

only thing that’s left to determine is limλ↑Υ υ∈λ w(υ).<br />

Lemma 6.3.3.<br />

lim<br />

λ↑Υ<br />

<br />

υ∈λ<br />

w(υ) = 1<br />

p(υ0)<br />

Proof. By Lemma 6.3.2 <strong>and</strong> the fact that w(υ0) = 1, we have<br />

p(υ0) = lim P ({υ0}|λ) = lim<br />

λ↑Υ λ↑Υ<br />

<br />

w(υ0)χλ(υ0)<br />

<br />

υ∈λ w(υ)<br />

<br />

= limλ↑Υ χλ(υ0)<br />

<br />

limλ↑Υ υ∈λ w(υ)<br />

Since p(υ0) > 0 by NAP 1 <strong>and</strong> χλ(υ0) is eventually equal to 1 (take λ = {υ0}), we<br />

conclude.<br />

Putting everything together, we obtain the desired result:<br />

Proposition 6.3.4.<br />

P (A) = p(υ0) · <br />

w(υ) (6.5)<br />

Proof. By Lemma 6.3.2, Lemma 6.3.3 <strong>and</strong> NAP 5, we have the following chain of<br />

equalities:<br />

92<br />

P (A) = lim<br />

P (A|λ) =<br />

λ↑Υ limλ↑Υ<br />

<br />

limλ↑Υ<br />

υ∈A ◦<br />

υ∈λ∩A w(υ)<br />

<br />

υ∈λ w(υ) = p(υ0) · <br />

w(υ)<br />

υ∈A ◦


6.4. FAIR LOTTERIES BY NAP<br />

Remark 6.3.5. Equation 6.5 is an important result, since it generalizes equation<br />

6.1 to the case of arbitrary sample spaces. If Υ is finite, then it’s rather easy to see<br />

that NAP <strong>and</strong> Kolmogorov’s probability coincide. If Υ is infinite, notice that this is<br />

the best result we can get, since the equation<br />

P (A) = <br />

p(υ)<br />

υ∈A ◦<br />

is not well defined, because it’s generally false that p(υ) ∈ R for all υ ∈ Υ, <strong>and</strong> the<br />

sum over hyperextended sets is well defined only in this case.<br />

The next Proposition replaces σ-additivity:<br />

Proposition 6.3.6. Let {Ai}i∈I be pairwise disjoint sets <strong>and</strong> define A = <br />

i∈I Ai.<br />

Then<br />

<br />

P (A) = lim P (Ai|λ)<br />

i∈I ◦<br />

λ↑Υ<br />

i∈I◦ Proof. For all λ ∈ Pfin(Ω), we have<br />

<br />

<br />

i∈I<br />

P (Ai|λ) =<br />

◦<br />

<br />

υ∈λ∩Ai w(υ)<br />

<br />

υ∈λ∩Υ w(υ)<br />

=<br />

Taking the Ω-limit, we conclude.<br />

<br />

υ∈A w(υ)<br />

<br />

υ∈λ∩Υ<br />

w(υ) = P (A|λ)<br />

Remark 6.3.7. The reader should be warned that Proposition 6.3.6 does not say<br />

that P (A) = <br />

i∈I◦ P (Ai). The equality stated by Proposition 6.3.6 is a rather<br />

trivial result that actually doesn’t tell us more information than what we already<br />

have.<br />

Remark 6.3.8. Given any NAP P , we can define from it an Archimedean probability<br />

Parch by the position<br />

Parch(A) = sh(P (A))<br />

This probability Parch is defined on all subsets of Ω <strong>and</strong> is always finitely additive.<br />

Moreover, from case to case, it has some interesting properties: regarding fair<br />

lotteries, we rem<strong>and</strong> to the considerations expressed in Remark 25 of [12] .<br />

6.4 Fair lotteries by NAP<br />

Thanks to the results of the previous section, we can show some examples of NAP<br />

at work. We begin by studying the fair lotteries over infinite sets.<br />

Example 6.4.1 (Fair lottery on N). Since the lottery is fair, we have p(n) = p(m) =<br />

p for all n, m ∈ N. In this case, the normalization axiom requires that<br />

<br />

p = 1<br />

n∈N ◦<br />

93


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

It’s easy to see that <br />

n∈N ◦<br />

p = lim<br />

λ↑N<br />

<strong>and</strong>, taking the Ω-limit, we conclude<br />

from which we deduce<br />

<br />

p(n) = |λ ∩ N| · p<br />

n∈λ<br />

n(N) · p = 1<br />

p(n) = 1<br />

n(N)<br />

for all n ∈ N. Since n(N) is an infinite number, this result is consistent <strong>with</strong> the<br />

intuition that one ticket will be extracted, but every ticket has a very small probability<br />

of being extracted. In the same spirit, we observe that<br />

P (A) = n(A)<br />

n(N)<br />

(6.6)<br />

so that the probability of the event A is a (rigorous <strong>and</strong> formal) generalization of the<br />

intuitive concept “the number of the elementary events in A divided by the number<br />

all the possible cases”.<br />

Equation 6.6 shows us an important link between NAP <strong>and</strong> numerosity. In fact,<br />

if we want the fair lottery on N to satisfy additional properties, such as the fact<br />

that P (E) = P (O), then we need to introduce this fact as an opportune axiom<br />

of numerosity. In turn, this axiom will correspond to a particular choiche of the<br />

ultrafilter UΩ <strong>and</strong> to a particular model of Ω-<strong>Theory</strong>. This is the approach to NAP<br />

followed in [12].<br />

In our case, it turns out that we already have an axiom of numerosity that entails<br />

that the fair lottery on N satisfies P (E) = P (O): it’s Numerosity Axiom 1. In fact,<br />

when studying the model of Ω-Calculus, we have already seen in Example 1.12.4 that,<br />

under Numerosity Axiom 1, we can deduce the equality P (E) = P (O) = 1/2α. We<br />

signal that, assuming this axiom, the fair lottery on N also satisfies the asymptotic<br />

limit property, for which we rem<strong>and</strong> to [12].<br />

Now, we will give an example about how NAP solves the problem of “almost<br />

impossible” <strong>and</strong> “almost certain” events.<br />

Example 6.4.2 (Fair lottery on N, continuation). Let A = E ∪ {1}. Clearly,<br />

A ⊃ E, so we expect P (A) > P (E). Using the obvious decomposition of A, we<br />

have immediately<br />

P (A) = P (E) + 1<br />

α<br />

<strong>and</strong> the desired inequality holds. If we had chosen B = E ∪ {1, 3, . . . , 57}, we would<br />

have found that<br />

P (B) = P (E) + 29<br />

> P (A) > P (E)<br />

α<br />

<strong>and</strong> this rigorously express not only the idea that B is more likely to happen than A,<br />

but also allows to compute precisely how much more likely is B to happen.<br />

94


6.4. FAIR LOTTERIES BY NAP<br />

Using Numerosity Axiom 1, we can generalize the reasonings to fair lotteries over<br />

other number sets.<br />

Example 6.4.3 (Fair lottery on Z). Reasoning as in the previous examples, we have<br />

that, for all n ∈ Z,<br />

1<br />

P (n) =<br />

2α + 1<br />

<strong>and</strong><br />

P (A) = n(A)<br />

2α + 1<br />

This time, if we compute P (2Z), the probability of extracting an even integer, we get<br />

|2Z ∩ λn,θ| = |Z ∩ λn,θ| + 1<br />

2<br />

(the term +1 is due to the presence of 0). We deduce<br />

so, taking the Ω-limit, we get<br />

P (2Z ∩ λn,θ)<br />

P (Z ∩ λn,θ) = |Z ∩ λn,θ| + 1<br />

2|Z ∩ λn,θ|<br />

P (2Z) = 1 1<br />

+<br />

2 4α + 2<br />

1<br />

=<br />

2 +<br />

1<br />

2|Z ∩ λn,θ|<br />

while, on the other h<strong>and</strong>, the probability of extracting an odd number is<br />

P (Z \ 2Z) = 1 1<br />

−<br />

2 4α + 2<br />

This difference is due to the presence of the zero. If we consider the set 2Z \ {0},<br />

we get<br />

P (2Z \ {0}) = P (2Z) − P ({0}) = 1 1 1<br />

+ −<br />

2 4α + 2 2α + 1<br />

1 1<br />

= −<br />

2 4α + 2<br />

as expected. We can conclude that Numerosity Axiom 1 entails that even numbers<br />

are “exactly one more more” than odd numbers.<br />

Example 6.4.4 (Fair lottery on Q). In the case of the fair lottery on Q, we can spend<br />

some words regarding conditional probability. For instance, if [a0, b0)Q ⊂ [a1, b1)Q,<br />

we would expect the conditional probability to satisfy the formula<br />

This equation is equivalent to<br />

P ([a0, b0)Q|[a1, b1)Q) = b0 − a0<br />

b1 − a1<br />

(6.7)<br />

n([a, b)Q) = (b − a) · n([0, 1)Q) (6.8)<br />

since, given this equality, we can easily prove equation 6.7: from<br />

P ([a, b)Q) =<br />

n([a, b)Q)<br />

n(Q)<br />

= (b − a) · n([0, 1)Q)<br />

n(Q)<br />

95


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

we deduce<br />

P ([a0, b0)Q|[a1, b1)Q) = P ([a0, b0)Q)<br />

P ([a1, b1)Q<br />

To see that equation 6.8 holds, it’s sufficient to compute<br />

lim |[a, b)Q ∩ λ|<br />

λ↑Ω<br />

= b0 − a0<br />

b1 − a1<br />

Thanks to Numerosity Axiom 1 <strong>and</strong> by assuming the model of Ω-Calculus given in<br />

section 1.11, we have<br />

lim |[a, b)Q ∩ λ| = lim |[a, b)Q ∩ λn|<br />

λ↑Ω λn↑Q<br />

<strong>and</strong>, reasoning as in Proposition 1.11.1, we obtain that, eventually,<br />

|[a, b)Q ∩ λn| = (b − a) · n<br />

From here, it’s easy to deduce equation 6.8.<br />

For the fair lottery on R, we would like to ask a similar condition:<br />

P ([a0, b0)|[a1, b1)) = b0 − a0<br />

b1 − a1<br />

but this is not possible. To see why, consider the intervals [0, 1) ⊂ [0, π): for those<br />

intervals, we would have<br />

P ([0, 1)|[0, π)) = 1<br />

π<br />

but we can prove that, for all fair lotteries, P (A|B) ∈ Q ∗ .<br />

Proposition 6.4.5. If (Υ, P ) is a fair lottery <strong>and</strong> A, B ⊆ Υ, then P (A|B) ∈ Q ∗ .<br />

Proof. By definition,<br />

Since eventually<br />

P (A|B) =<br />

P (A ∩ B)<br />

P (B)<br />

P (A ∩ B ∩ λ)<br />

= lim<br />

λ↑Υ P (B ∩ λ)<br />

|A ∩ B ∩ λ|<br />

|B ∩ λ|<br />

the thesis follows by transfer of membership.<br />

∈ Q<br />

For the fair lottery on R, the best we can do is require<br />

P ([a0, b0)|[a1, b1)) ∼ b0 − a0<br />

b1 − a1<br />

|A ∩ B ∩ λ|<br />

= lim<br />

λ↑Υ |B ∩ λ|<br />

<strong>and</strong> this is satisfied in our model of Ω-<strong>Theory</strong> thanks to Numerosity Axiom 1. It’s<br />

also easy to see that the fair lotteries over N, Z, Q <strong>and</strong> R are all compatible, in<br />

the sense that, if we call PN, PZ, PQ <strong>and</strong> PR the respective fair lotteries <strong>and</strong> S =<br />

{N, Z, Q, R}, we have<br />

PS1(A) = PS2(A|S1)<br />

for all S1 ⊂ S2 <strong>and</strong> Si ∈ S.<br />

96


6.5. INFINITE SEQUENCE OF COIN TOSSES BY NAP<br />

6.5 <strong>Infinite</strong> sequence of coin tosses by NAP<br />

In this section, we want to consider the modelization of an infinite sequence of coin<br />

tosses <strong>with</strong> a fair coin. In the Kolmogorovian framework, the infinite sequence of<br />

coin tosses is modeled by the triple (Υ, U, µ). The sample space Υ = {H, T } N is the<br />

space of sequence <strong>with</strong> values in the set {H, T } (namely, heads <strong>and</strong> tails). If υ ∈ Υ,<br />

we will denote υ = (υ1, . . . , υn, . . .).<br />

The σ-algebra U is the σ-algebra generated by the “cylindrical sets”: given a nple<br />

of indices (i1, . . . , in) <strong>and</strong> a n-ple (t1, . . . , tk) of elements in {H, T }, a cylindrical<br />

set of condimension n is defined as follows:<br />

C (i1,...,in)<br />

(t1,...,tk)<br />

= {υ ∈ Υ : υik = tk}<br />

represents the event that the ik-th<br />

coin toss gives tk as outcome for k = 1, . . . , n.<br />

The probability measure on the generic cylindrical set is given by<br />

<br />

µ<br />

= 2 −n<br />

(6.9)<br />

From the probabilistic point of view, C (i1,...,in)<br />

(t1,...,tk)<br />

C (i1,...,in)<br />

(t1,...,tk)<br />

Thanks to Caratheodory’s Theorem, this measure can be extended in a unique way<br />

to all U.<br />

In this model, we can clealy see the problems <strong>with</strong> the Kolmogorovian approach<br />

discussed in section 6.1:<br />

• every event {υ} ∈ U has probability 0 to occurr, but the union of all these<br />

“seemingly impossible” events has probability 1;<br />

• if F is a finite set <strong>and</strong> {υ} ⊂ F , then the conditional probability µ({υ}|F ) is<br />

not defined. On the other h<strong>and</strong>, we know that {υ}|F it’s not the empty event,<br />

so the conditional probability makes sense <strong>and</strong> we expect it to be 1/|F |;<br />

• there are subsets of Υ for which the probability is not defined (namely, the<br />

non-measurable sets).<br />

Now, we will show how the same situation can be modeled by non-Archimedean<br />

probability. In particular, we want that the NAP P over Υ satisfies the following<br />

assumptions:<br />

1. if F ⊂ Υ is finite, then<br />

2. P satisfies equation 6.9:<br />

P (A|F ) =<br />

<br />

P<br />

C (i1,...,in)<br />

(t1,...,tk)<br />

|A ∩ F |<br />

|F |<br />

<br />

= 2 −n<br />

(6.10)<br />

Experimentally, we can only observe a finite number of outcomes: both cylindrical<br />

events <strong>and</strong> finite conditional probability are based on a finite number of observations.<br />

In some sense, the two assumptions above are the “experimental data” on<br />

which we want to construct our model.<br />

97


CHAPTER 6. AN APPLICATION: NON-ARCHIMEDEAN PROBABILITY<br />

The first property implies that the probability is fair, so every infinite sequence<br />

of coin tosses in Υ will have probability 1/n(Υ). Equation 6.10 is the counterpart<br />

of equation 6.8 in the case of a fair lottery on R: equation 6.10 implies that for any<br />

µ-measurable set E, we will have P (E) ∼ µ(E), in analogy to what implies equation<br />

6.8 in the fair lottery on R.<br />

To show that we can find a non-Archimedean probability measure P over P(Υ)<br />

that satisfies conditions 1 <strong>and</strong> 2, we only need to show an appropriate model for<br />

Ω-<strong>Theory</strong> in which equation 6.10 holds. This coincides <strong>with</strong> the choiche of an appropriate<br />

ultrafilter UCT over Pfin(Ω).<br />

In order to define the ultrafilter UCT , we need to introduce some notation. If<br />

υ 1 = (υ 1 1, . . . , υ 1 n) ∈ {H, T } n is a finite string <strong>and</strong> υ 2 = (υ 2 1, . . . , υ 2 k , . . .) ∈ {H, T }N ,<br />

then we define<br />

υ 1 ⋆ υ 2 = (υ 1 1, . . . , υ 1 n, υ 2 1, . . . υ 2 k, . . .)<br />

In other words, the sequence υ 1 ⋆ υ 2 is obtained by the sequence υ 1 followed by the<br />

sequence υ2 . <br />

N Now, if σ ∈ Pfin {H, T } <strong>and</strong> n ∈ N, we define<br />

λn,σ = {υ 1 ⋆ υ 2 : υ 1 ∈ {H, T } n <strong>and</strong> υ 2 ∈ σ}<br />

<strong>and</strong><br />

ΛCT = N<br />

λn,σ : n ∈ N <strong>and</strong> σ ∈ Pfin {H, T } <br />

Now, observe that ΛCT can be extended to a filter over Pfin(Ω), that in turn can be<br />

extended to a nonprincipal ultrafilter UCT . Now, we only need to show that, in the<br />

corresponding model of Ω-<strong>Theory</strong>, equation 6.10 holds.<br />

Proposition 6.5.1. Equation 6.10 holds in the model of Ω-<strong>Theory</strong> obtained by the<br />

ultrafilter UCT .<br />

Proof. Consider the cylinder C (i1,...,in)<br />

(t1,...,tk) <strong>and</strong> take N = max in. Then, for every σ ∈<br />

<br />

N {H, T } , we have<br />

Pfin<br />

λN,σ ∩ C (i1,...,in)<br />

(t1,...,tk) = {υ ∈ λN,σ : υik = tk}<br />

Then <br />

λN,σ ∩ C (i1,...,in)<br />

<br />

<br />

(t1,...,tk) = |{υ ∈ λN,σ : υik = tk}| = |λN,σ|<br />

2n Since<br />

|λN,σ| = 2 N · |σ|<br />

we conclude <br />

λN,σ ∩ C (i1,...,in)<br />

<br />

<br />

= 2 N−n · |σ|<br />

Taking the Ω-limit, we obtain<br />

<br />

n<br />

C (i1,...,in)<br />

(t1,...,tk)<br />

(t1,...,tk)<br />

= lim<br />

λN,σ↑Υ 2N−n · |σ| = 2 −n · lim<br />

λN,σ↑Υ 2N · |σ| = 2 −n · n(Υ)<br />

Thanks to the fact that P is fair, we conclude<br />

<br />

P C (i1,...,in)<br />

<br />

n<br />

(t1,...,tk) =<br />

n(Υ)<br />

as we wanted.<br />

98<br />

C (i1,...,in)<br />

(t1,...,tk)<br />

<br />

= 2 −n


Chapter 7<br />

Further ideas<br />

We have many intuitions in our<br />

life, <strong>and</strong> the point is that many of<br />

those intuitions are wrong. The<br />

question is: are we going to test<br />

those intuitions?<br />

Dan Ariely<br />

Due to the finiteness of this paper, we couldn’t develop properly all the ideas<br />

that arose during the study of Ω-<strong>Theory</strong>. While we hope to deepen their<br />

study in the future, we have decided to give a brief sketch of what we think<br />

as the most interesting topics that could be studied <strong>with</strong>in Ω-<strong>Theory</strong>.<br />

This chapter will be somehow different from the rest of this paper: we will often<br />

omit proofs <strong>and</strong> the reader won’t find interesting, polished results, but only rough<br />

ideas <strong>and</strong> intuitions. We do believe that some of those ideas will yeld interesting<br />

results, but the truth of this statement is still to be proved.<br />

7.1 The Cartesian Product Axiom for Numerosity<br />

Since we can assume that Ω ⊃ R N for all N ∈ N, it would be interesting to extend<br />

the results given in Sections 1.6 <strong>and</strong> 3.4 to subsets of R N . With this goal in mind, we<br />

should assume as axioms for numerosity the Numerosity Axiom 1 first introduced<br />

in Section 1.3 <strong>and</strong> a “cartesian product axiom” stating that the numerosity of the<br />

cartesian product A × B is the product of the numerosity of A by the numerosity of<br />

B.<br />

Numerosity Axiom 2 (Cartesian Product Axiom). For all A ⊆ R N , B ⊆ R M<br />

satisfying A × B ⊆ Ω, n(A × B) = n(A) · n(B).<br />

This new axiom allows the desired generalization, but has the major flaw that<br />

it doesn’t help us overcome the difficulties we’ve encountered in Section 3.4. In<br />

particular, it’s extremely difficulty to prove some results about countable additivity,<br />

even if if we settle for infinitely closeness instead of equality.<br />

99


CHAPTER 7. FURTHER IDEAS<br />

Now, we will show that the axioms of Ω-<strong>Theory</strong> plus Numerosity Axioms 1 <strong>and</strong><br />

2 admit a model. We have already seen in Theorem 4.5.5 that a model of Ω-<strong>Theory</strong><br />

is obtained by the ultrapower of Ω <strong>with</strong> respect to a nonprincipal fine ultrafilter U.<br />

To show a model in which Numerosity Axioms 1 <strong>and</strong> 2 hold, we need to impose<br />

further conditions on the ultrafilter:<br />

• If U ⊃ ΛR, then Numerosity Axiom 1 holds as a consequence of Theorem 1.11.4<br />

<strong>and</strong> Proposition 4.5.4.<br />

• In the same spirit, if U ⊃ ΛN R<br />

Numerosity Axiom 2 holds.<br />

for all N ∈ N, then we could show that also<br />

Unfortunately, there is a problem <strong>with</strong> the last request since, if N = M, then<br />

ΛN R <strong>and</strong> ΛMR don’t have the FIP. We circumvent this inconvenient by defining a new<br />

family of sets ΛC in a way that the corresponding ultrafilter satisfies all axioms of<br />

Ω-<strong>Theory</strong> plus Numerosity Axioms 1 <strong>and</strong> 2.<br />

We start by considering the sets λn,θ ∈ ΛR, <strong>and</strong> then, for all N ∈ N, we pose<br />

λn,θ,N = λn,θ ∪ λn,θ × λn,θ ∪ . . . ∪ λn,θ × . . . × λn,θ =<br />

<br />

N times<br />

With this notation, λn,θ = λn,θ,1, <strong>and</strong> we will often omit the third index when<br />

referring to them. Now, we can set<br />

ΛC = {λn,θ,N : n = m!, m ∈ N, θ ∈ Θ, N ∈ N}<br />

It’s now easy to show that, if we take the Ω-limit <strong>with</strong> respect to λn,θ,N, then<br />

Numerosity Axiom 2 holds:<br />

Proposition 7.1.1. If A ⊆ RM M ′<br />

<strong>and</strong> B ⊆ R , then<br />

N<br />

k=1<br />

λ k n,θ<br />

lim<br />

λn,θ,N ↑Ω |(A × B) ∩ λn,θ,N| = lim<br />

λn,θ,N ↑Ω |A ∩ λn,θ,N| · lim |B ∩ λn,θ,N|<br />

λn,θ,N ↑Ω<br />

Proof. If N ≥ M + M ′ , then<br />

|(A × B) ∩ λn,θ,N| = |(A × B) ∩ λ<br />

but, using the properties of cartesian products,<br />

M+M ′<br />

(A × B) ∩ λn,θ M+M ′<br />

n,θ<br />

= (A ∩ λ M M ′<br />

n,θ) × (B ∩ λn,θ) At this point, by the finiteness of λn,θ, we easily conclude<br />

On the other h<strong>and</strong>,<br />

|(A ∩ λ M M ′<br />

n,θ) × (B ∩ λ<br />

n,θ)| = |A ∩ λ M M ′<br />

n,θ| · |B ∩ λn,θ| |A ∩ λn,θ,N| = |A ∩ λ M M ′<br />

n,θ| , |B ∩ λn,θ,N| = |B ∩ λn,θ| so, putting everything together, the thesis follows.<br />

100<br />

|


7.2. DIMENSION BY Ω-THEORY<br />

We’ve seen that ΛC seems to be the right choice of sets if we want to satisfy all the<br />

desired numerosity. Moreover, it’s quite easy to see that ΛC has the FIP: if we take<br />

λn,θ,N <strong>and</strong> λn ′ ,θ ′ ,N ′ ∈ ΛC then, by definition, λn,θ,1 ⊆ λn,θ,N <strong>and</strong> λn ′ ,θ ′ ,1 ⊆ λn ′ ,θ ′ ,N ′,<br />

<strong>and</strong> we already know that λn,θ,1 ∩ λn ′ ,θ ′ ,1 = ∅. We can then consider the filter<br />

FΛC generated by ΛC <strong>and</strong>, by another application of Zorn’s Lemma, we obtain a<br />

nonprincipal ultrafilter UC that extends FΛC . Now, we will show that the Numerosity<br />

Axioms hold in the ultrapower of Ω modulo UC.<br />

Theorem 7.1.2. Ω-Axioms 1, 2, 3, 4 <strong>and</strong> Numerosity Axioms 1 <strong>and</strong> 2 hold in the<br />

ultrapower ΩUC .<br />

Proof. The axioms of Ω-<strong>Theory</strong> hold by Theorem 4.5.5.<br />

Numerosity Axiom 1 holds as a consequence of Theorem 1.11.4 <strong>and</strong> Proposition<br />

4.5.4.<br />

Numerosity Axiom 2 holds as a consequence of Proposition 7.1.1.<br />

7.2 Dimension by Ω-<strong>Theory</strong><br />

Thanks to the Cartesian Product Axiom, we can generalize some results of Section<br />

1.6 <strong>and</strong> of Section 3.4 to subsets of R N . Since we are interested in the development<br />

of a concept of measure for subsets of R N , we begin by showing explicitely the<br />

generalization of Proposition 1.6.7 to arbitrary dimension.<br />

Proposition 7.2.1. Let E = N n=1 [an, bn), <strong>with</strong> an <strong>and</strong> bn ∈ R. Then Numerosity<br />

Axiom 2 implies<br />

N<br />

n(E)<br />

∼ (bn − an)<br />

n([0, 1)) N<br />

Proof. By Numerosity Axiom 2,<br />

n(E)<br />

=<br />

n([0, 1)) N<br />

n=1<br />

N<br />

n=1<br />

At this point, we conclude by Proposition 1.6.7.<br />

n([an, bn))<br />

n([0, 1))<br />

Given a set E ⊆ R N , the ratio n(E) · n([0, 1)) −N plays an important role in the<br />

development of a concept of dimension. For this reason, we find useful to fix it in a<br />

definition.<br />

Definition 7.2.2. If E ⊆ R N <strong>and</strong> r ∈ R + , we will call the normalized numerosity<br />

of order r of the set E the ratio<br />

n(E)<br />

n([0, 1)) r<br />

<strong>and</strong> we will denote it by nnr(E). When there will be no confusion about r, we will<br />

often write nn(E) instead of nnr(E).<br />

As an example of the use of normalized numerosity, we will rephrase some known<br />

results by using this concept.<br />

101


CHAPTER 7. FURTHER IDEAS<br />

Example 7.2.3. Notice that, if r = 1, then the normalized numerosity of order 1<br />

coincides <strong>with</strong> the notion given in Definition 3.2.3.<br />

Now, suppose that a, b ∈ Q. We already know that, thanks to Numerosity Axiom<br />

1, nn1([a, b)) = b − a. It’s also clear that, if r > 1, then nnr([a, b)) ∼ 0, since<br />

nnr([a, b)) = nn1([a, b)) ·<br />

1<br />

1<br />

= (b − a) ·<br />

n([0, 1)) r n([0, 1)) r−1<br />

<strong>and</strong> n([0, 1)) −r+1 is infinitesimal for all r > 1.<br />

If we consider the set E = [a, b) × [c, d) ⊂ R 2 , then Numerosity Axiom 2 entails<br />

nn2(E) ∼ (b − a) · (d − c). In the same spirit as the previous example, it’s also<br />

easy to see that nn1(E) = nn2(E) · n([0, 1)) is an infinite number, while nnr(E) is<br />

infinitesimal for all r > 2.<br />

The definition of the normalized numerosity of order r <strong>and</strong> the ideas shown in<br />

Example 7.2.3 bear some similarities <strong>with</strong> the definition of Hausdorff measure <strong>and</strong><br />

some of its consequences. Moreover, the same results show that, in analogy to the<br />

definition of Hausdorff dimension from Hausdorff measure, it could be interesting to<br />

give a notion of dimension based on normalized numerosity. Since we want to give<br />

this definition for subsets of R N , we will use the setting of Ω-<strong>Theory</strong> plus Numerosity<br />

Axioms 1 <strong>and</strong> 2. We also assume that the reader is familiar <strong>with</strong> Hausdorff measure<br />

<strong>and</strong> dimension. For an introduction about those topics, we rem<strong>and</strong> to [7].<br />

We begin by defining the dimension of a set in the following way:<br />

Definition 7.2.4. If E ⊆ R N , we set<br />

dim(E) = inf{r ∈ R + : nnr(E) ∼ 0}<br />

Given this definition, we can generalize <strong>and</strong> make precise the results found in<br />

Example 7.2.3:<br />

Proposition 7.2.5. If E = N<br />

n=1 [an, bn) <strong>and</strong> we don’t exclude that an = bn for some<br />

n ≤ N, then<br />

dim(E) ≤ N<br />

More precisely, if we call A = {n ≤ N : an = bn}, we have<br />

Proof. By Numerosity Axiom 2, we have<br />

hence, for all r ∈ R + ,<br />

n(E) =<br />

dim(E) = |A|<br />

N<br />

n([an, bn)) = <br />

n([an, bn))<br />

n=1<br />

n∈A<br />

n∈A<br />

nnr(E) = <br />

1<br />

nn1([an, bn)) ·<br />

n([0, 1)) r−|A|<br />

Observe that, by hypothesis, <br />

n∈A nn1([an, bn)) is a finite, positive number. We<br />

immediately deduce nnr(E) ∼ 0 for all r > |A| <strong>and</strong> nn|A|(E) = <br />

n∈A nn1([an, bn)) ><br />

0, as we wanted.<br />

102


7.3. NUMEROSITY AND LEBESGUE MEASURE<br />

The next step would be to generalize those ideas to arbitrary subsets of RN .<br />

Unfortunately, there is a problem <strong>with</strong> subsets whose normalized numerosity is an<br />

infinite number. As an interesting <strong>and</strong> fundamental example, take the case of the<br />

set R: the normalized numerosity of order 1 of R is an infinite number <strong>and</strong>, when<br />

r > 1, we don’t know how to extimate the ratio n(R) · n([0, 1)) −r . This is essential<br />

in order to proceed since, if we find that there exists some positive s ∈ R satisfying<br />

<br />

n(R)<br />

sh<br />

n([0, 1)) s<br />

<br />

< +∞ (7.1)<br />

then the dimension of R would be s. Unfortunately, it’s not straightforward to<br />

determine if such a s exists <strong>and</strong>, even if the existence of s is settled, one needs also<br />

to calculate its exact value. Moreover, there is the possibility that the validity of<br />

equation 7.1 depends upon the choiche of the model of Ω-<strong>Theory</strong>, <strong>and</strong> so a new<br />

axiom could be necessary to rule the dimension of R.<br />

Then, it would naturally arise also the question if there could be some mathematical<br />

interest in giving a definition of dimension where the range of all possible<br />

dimension is all R ∗ . In this case, we should find out what could be a good definition,<br />

<strong>and</strong> then study its consequences <strong>and</strong> possible applications.<br />

7.3 Numerosity <strong>and</strong> Lebesgue measure<br />

I teoremi non si dimostrano a<br />

com<strong>and</strong>o.<br />

Gianni Gilardi<br />

During the study of Ω-<strong>Theory</strong>, we have spent quite some time in the attempt<br />

to find some axioms of numerosity that would entail the equivalence between numerosity<br />

<strong>and</strong> Lebesgue measure (<strong>and</strong>, as a consequence, between Ω-integral <strong>and</strong><br />

Lebesgue integral). At first, we tried in many ways to give a countable additivity<br />

result for numerosity assuming only the validity of Numerosity Axiom 1, <strong>with</strong>out<br />

success. After those failures, we tried to impose the desired result by a new axiom<br />

of numerosity: it turned out that one of such axioms exists, but it isn’t powerful<br />

enough to entail the equivalence of numerosity <strong>and</strong> Lebesgue measure. In particular,<br />

this axiom didn’t entail the equality sh(nn1([0, 1)Q)) = 0, against the well known fact<br />

that µ([0, 1)Q) = 0.<br />

Our more recent attempt is to prove the coherence <strong>with</strong> Ω-<strong>Theory</strong> of the following<br />

axioms:<br />

1. Numerosity Axiom 1 <strong>and</strong><br />

2. if E <strong>and</strong> F are two Lebesgue measurable sets satisfying µ(E) < µ(F ), then<br />

also nn1(E) < nn1(F ).<br />

By a simple argument, we can easily show that the validity of both requests would<br />

immediately imply that nn1(E) ∼ µ(E) . In fact, for all q ∈ Q satisfying µ(E) < q,<br />

103


CHAPTER 7. FURTHER IDEAS<br />

<br />

we can argue that nn1(E) < nn1 [0, q) = q, <strong>and</strong> the same idea can be used to show<br />

the validity of the other inequality.<br />

Unfortunately, it’s still an open question wether those two axioms are compatible<br />

<strong>with</strong> Ω-<strong>Theory</strong>. If this turns out to be the case, then also the desired equivalence<br />

between the Ω-integral <strong>and</strong> Lebesgue integral would easily follow from the result<br />

shown above. On the other h<strong>and</strong>, if the two axioms are not compatible <strong>with</strong> Ω-<br />

<strong>Theory</strong>, it would be a very interesting research field to determine if there can exists<br />

an axiom that entails the equivalence between numerosity <strong>and</strong> Lebesgue measure,<br />

or if such equivalence is incompatible <strong>with</strong> the theory.<br />

We signal that another possible route toward a result of equivalence between<br />

numerosity <strong>and</strong> Lebesgue measure is given by the natural extension of numerosity,<br />

a topic that will be briefly presented in the last section of this chapter.<br />

7.4 Functional Analysis by Ω-<strong>Theory</strong><br />

We want to spend a few words about how the notions of functional analysis can be<br />

given <strong>with</strong>in Ω-<strong>Theory</strong>. Since we know that, by Transfer Principle, the notions of<br />

R-vector space, metric, seminorm, norm, bilinear form <strong>and</strong> scalar product can be<br />

extended by using Ω-limit in a way that their elementary properties are preserved,<br />

we could easily give the definitions of R ∗ -vector spaces <strong>and</strong> of metrics, seminorms,<br />

norms, bilinear forms <strong>and</strong> scalar products that take values in R ∗ <strong>and</strong> study their<br />

behaviour, <strong>with</strong> some applications of functional analysis in mind. Moreover, the<br />

definition of uniform topology, introduced in the real case by Definition 2.2.1, can<br />

be extended to arbitrary sets, <strong>and</strong> we believe that many other notions of topology<br />

<strong>and</strong> differential geometry can be given in the same way. It could be an interesting<br />

research topic to study if those notions together <strong>with</strong> Ω-limit could be used to obtain<br />

some interesting results of functional analysis, topology <strong>and</strong> differential geometry.<br />

7.5 R ∗∗<br />

Given the axioms of Ω-<strong>Theory</strong>, one can wonder wether the Ω-limit of functions<br />

ϕ : Fun(Pfin(Ω), R∗ ) can be defined. The answer is yes, <strong>and</strong> we can indeed develop<br />

a theory for R∗∗ = (R∗ ) ∗ = limλ↑Ω cR∗(λ). Since a thorough study of this topic runs<br />

beyond the scope of this paper, in this section we will only give a few examples of<br />

the behaviour of Ω-limit of hyperreal numbers. For this reason, the main “inference<br />

tool” we will use throughout the section will be Transfer Principle since, if used<br />

correctly, it allows to easily deduce some meaningful inequalities.<br />

We find interesting to begin the studt of R∗∗ by showing some examples of the<br />

behaviour of Ω-limit of constant functions.<br />

Example 7.5.1. Let’s pick ξ ∈ R ∗ , ξ ∼ 0 <strong>and</strong> consider the function cξ(λ). By<br />

definition of infinitesimal number,<br />

104<br />

cξ(λ) < 1<br />

n<br />

for all n ∈ N


<strong>and</strong>, by Transfer of inequalities, we deduce<br />

lim cξ(λ) <<br />

λ↑Ω 1<br />

ν<br />

for all ν ∈ N∗<br />

7.5. R ∗∗<br />

By Lemma 5.2.6, we conclude immediately that limλ↑Ω cξ(λ) = ξ.<br />

In the same spirit, if ξ is infinite, then limλ↑Ω cξ(λ) > ν for all ν ∈ N ∗ <strong>and</strong>, <strong>with</strong><br />

a similar argument, we conclude that also limλ↑Ω cξ(λ) = ξ.<br />

If ξ is finite, then ξ = sh(ξ) + ɛ, where ɛ is infinitesimal, hence<br />

lim cξ(λ) = sh(ξ) + lim cɛ(λ)<br />

λ↑Ω λ↑Ω<br />

<strong>and</strong>, by the first part of this example, we deduce limλ↑Ω cξ(λ) ∼ ξ.<br />

Motivated by this example, we will introduce a new definition, that will be a<br />

useful short-h<strong>and</strong>.<br />

Definition 7.5.2. For all ξ ∈ R ∗ , we will call ξ ∗ = limλ↑Ω cξ(λ).<br />

The reasonings shown in the previous example should be sufficient to convince<br />

the reader of the validity of the following statements:<br />

• ξ ∗ = ξ if <strong>and</strong> only if ξ ∈ R;<br />

• sh(ξ) = sh(ξ ∗ ).<br />

We can now generalize the ideas of Example 7.5.1 to arbitrary functions.<br />

Example 7.5.3. Consider the function ϕ(λ) = |λ|/n(R). Clearly, this is a strictly<br />

increasing function in the sense that, if λ2 ⊃ λ1, then ϕ(λ2) > ϕ(λ1). The intuition<br />

could suggest that the Ω-limit of this function must be 1 but, by using correctly<br />

transfer principle, we deduce that<br />

lim ϕ(λ) =<br />

λ↑Ω n(R)<br />

n(R) ∗<br />

Reasoning as in Example 7.5.1, we also deduce<br />

n(R) 1<br />

<<br />

n(R) ∗ ν<br />

for all ν ∈ N∗<br />

In the same way, one can prove that, if ϕ(λ) is infinitesimal a.e., then<br />

lim ϕ(λ) <<br />

λ↑Ω 1<br />

ν<br />

for all ν ∈ N∗<br />

This property is somehow in contrast <strong>with</strong> the intuition that some infinitesimal functions<br />

could have finite Ω-limit.<br />

105


CHAPTER 7. FURTHER IDEAS<br />

Example 7.5.4. Consider the sequence of hyperreal numbers {ξn}n∈N, <strong>and</strong> suppose<br />

that ξn is finite for all n ∈ N. We could then define<br />

<br />

lim ξn ∈ R<br />

λ↑N<br />

∗∗<br />

n∈λ<br />

<strong>and</strong> wonder if this definition has some good properties. For every λ ∈ Pfin(Ω), we<br />

can write <br />

ξn = <br />

sh(ξn) + ɛλ<br />

n∈λ∩N<br />

n∈λ<br />

n∈λ∩N<br />

where ɛλ is infinitesimal. Taking the Ω-limit, we obtain<br />

<br />

lim ξn = lim sh(ξn) + lim ɛλ<br />

λ↑N<br />

λ↑N<br />

λ↑Ω<br />

<strong>and</strong>, by Example 7.5.3, we already know that limλ↑Ω ɛλ ∼ 0, so we can conclude<br />

<br />

lim ξn ∼<br />

λ↑N<br />

<br />

sh(ξn)<br />

n∈λ<br />

n∈λ<br />

n∈N ◦<br />

7.6 Ω-integral of hyperreal functions<br />

Once we have defined the field R ∗∗ , one of the first temptations is to extend the<br />

definition of Ω-integral to functions <strong>with</strong> hyperreal range. More precisely, we can<br />

take Definition 3.2.1 <strong>and</strong> simply extend the class of functions for which the Ω-integral<br />

is defined. As a result, we would get the following definition:<br />

Definition 7.6.1. For any f : E ⊆ R → R∗ we pose<br />

<br />

f ◦ 1<br />

(x)dΩx =<br />

n([0, 1)) lim<br />

<br />

f(x)<br />

λ↑E<br />

E ◦<br />

Unfortunately, the integral defined above isn’t linear over R ∗ . In fact, it’s easy<br />

to show that linearity is replaced by another property:<br />

Proposition 7.6.2. According to Definition 7.6.1, if f : E → R <strong>and</strong> ξ ∈ R∗ , then<br />

<br />

ξ · f ◦ (x)dΩx = ξ ∗ <br />

· f ◦ (x)dΩx<br />

E ◦<br />

Proof. By the properties of Ω-limit, we have the following chain of equalities:<br />

<br />

ξ · f ◦ (x)dΩx =<br />

1<br />

n([0, 1)) lim<br />

<br />

ξ ·<br />

λ↑E<br />

<br />

<br />

f(x)<br />

E ◦<br />

so the thesis follows.<br />

106<br />

E ◦<br />

x∈λ<br />

x∈λ<br />

=<br />

ξ∗ n([0, 1)) lim<br />

<br />

f(x)<br />

λ↑E<br />

x∈λ<br />

= ξ ∗ <br />

· f ◦ (x)dΩx<br />

E ◦


7.7. THE DIRAC DISTRIBUTION<br />

In order to avoid this problem, a completely new definition is necessary. First of<br />

all notice that, if f : R → R ∗ then, for every x ∈ R, we can find ϕx(λ) : Pfin(Ω) → R<br />

satisfying limλ↑Ω ϕx(λ) = f(x). Moreover, if f(x) ∈ R we can suppose that ϕx(λ) =<br />

cf(x)(λ). Since f(x) = limλ↑Ω ϕx(λ) for all x ∈ R, we can define the Ω-integral of f<br />

in the following way:<br />

Definition 7.6.3. For any f : E ⊆ R → R∗ , we pose<br />

<br />

f ◦ 1<br />

(x)dΩx =<br />

n([0, 1)) lim<br />

λ↑Ω<br />

<br />

E ◦<br />

x∈λ∩E<br />

As we did before, we will usually denote this integral by <br />

We remark that, according to Definition 7.6.3,<br />

<br />

f(x)dΩx ∈ R ∗<br />

E ◦<br />

ϕx(λ)<br />

E ◦ f(x)dΩx.<br />

so we discovered that the field R ∗∗ isn’t really necessary to define the Ω-integral of<br />

functions that take value in R ∗ .<br />

It’s easily verified that this integral is independent on the choiche of the functions<br />

ϕx <strong>and</strong>, if f : R → R, then Definition 3.2.1 <strong>and</strong> Definition 7.6.3 give rise to the same<br />

notion of Ω-integral, since it’s sufficient to take ϕx(λ) = f(x) for all λ ∈ Pfin(Ω)<br />

<strong>and</strong> for all x ∈ R. Moreover, it’s rather easy to see that linearity over R ∗ holds.<br />

Proposition 7.6.4. According to Definition 7.6.3, if f : E → R∗ <strong>and</strong> ξ ∈ R∗ , then<br />

<br />

<br />

ξ · f(x)dΩx = ξ · f(x)dΩx<br />

E ◦<br />

Proof. Pick a function ψ : Pfin(Ω) → R satisfying limλ↑Ω ψ(λ) = ξ. Then, we have<br />

the following chain of equalities<br />

as we wanted.<br />

<br />

E ◦<br />

ξ · f(x)dΩx =<br />

=<br />

1<br />

n([0, 1)) lim<br />

λ↑Ω<br />

1<br />

n([0, 1))<br />

<br />

E ◦<br />

<br />

x∈λ∩E<br />

lim ψ(λ) · lim<br />

λ↑Ω λ↑E<br />

=<br />

ξ<br />

n([0, 1)) lim<br />

<br />

ϕx(λ)<br />

λ↑E<br />

x∈λ<br />

<br />

= ξ · f(x)dΩx<br />

7.7 The Dirac distribution<br />

E ◦<br />

(ψ(λ) · ϕx(λ))<br />

<br />

<br />

ϕx(λ)<br />

In this section, we want to show how the well known distribution of the Dirac mass<br />

can be defined by Ω-<strong>Theory</strong>. We suppose that the reader is familiar <strong>with</strong> the topic<br />

x∈λ<br />

107


CHAPTER 7. FURTHER IDEAS<br />

<strong>and</strong> we rem<strong>and</strong> to Example 4.1 of Chapter 3 of [4] for a reference. We also signal that<br />

we will oversimplify the presentation of this topic, <strong>and</strong> we will omit some important<br />

details <strong>and</strong> definitions in order to focus on the original approach given by Ω-<strong>Theory</strong>.<br />

We begin by recalling briefly the definition of the Dirac mass.<br />

Definition 7.7.1. For a matter of simplicity, suppose that K ⊂ R is a compact<br />

set <strong>and</strong> v ∈ C∞ (K). Given x0 ∈ K, we define the Dirac mass centered at x0 the<br />

distribution satisfying <br />

δx0v = v(x0) (7.2)<br />

where the integral is intended in the sense of distributions.<br />

K<br />

In the context of Ω-<strong>Theory</strong>, we are tempted to define a function ∂x0 : R → R∗ such that <br />

K◦ (∂x0 · v)(x)dΩx = v(x0)<br />

Using the Ω-integral given by Definition 7.6.3, it’s immediate to see that, if we pose<br />

∂x0 = n([0, 1)) · χ{x0}, then<br />

<br />

K◦ (∂x0 · v)(x)dΩx =<br />

n([0, 1)) · v(x0)<br />

n([0, 1))<br />

= v(x0)<br />

It seems to us that this approach shows a number of interesting attributes. First<br />

of all, we find very appealing that the Dirac mass can be defined as a function<br />

whose Ω-integral yelds the desired result. This is in open contrast <strong>with</strong> the classical<br />

theory, where one needs to extend the meaning of the integral to take into account<br />

the fact that a function that is equal to 0 almost everywhere <strong>with</strong> respect to µ has<br />

nevertheless a nonzero integral. Moreover, we think that the definition of the Dirac<br />

mass in the context of Ω-<strong>Theory</strong> is simpler <strong>and</strong> entails in an intuitive way the sense<br />

of the distribution that, in the classical theory, has to be retraced in the alternate<br />

definition of δx0 as limit of simple functions.<br />

It’s still an open question if this approach can be extended <strong>with</strong> similar benefits<br />

to the whole theory of distributions, or if the conceptual simplification happens only<br />

in the case of the Dirac mass.<br />

7.8 The natural extension of numerosity<br />

In the setting of Ω-<strong>Theory</strong> <strong>and</strong> <strong>with</strong>out assuming any axiom of numerosity, we’ve<br />

seen in Section 1.6 that numerosity is a function n : P(Ω) → N ∗ that’s always<br />

monotone <strong>and</strong> finitely additive. By Transfer Principle, we conclude that the natural<br />

extension of numerosity is monotone <strong>and</strong> hyperfinitely additive in a sense that can<br />

be precised. It’s only natural to wonder if this new function can replace numerosity<br />

as basic tool for “counting” the number of elements of sets.<br />

For a matter of commodity, we will start by focus on the restriction of n to P(R):<br />

from the definition<br />

n : P(R) → N ∗<br />

108


we obtain that<br />

7.8. THE NATURAL EXTENSION OF NUMEROSITY<br />

n ∗ : P(R) ∗ → N ∗∗<br />

Explicitely, n∗ is defined on the set<br />

P(R) ∗ <br />

<br />

= lim ψ(λ) : ψ(λ) ∈ P(R) for all λ ∈ Pfin(Ω)<br />

λ↑Ω<br />

From this definition we can already deduce some properties of n ∗ .<br />

Proposition 7.8.1. For all A ⊆ R, n ∗ (A ∗ ) = (n(A)) ∗ .<br />

Proof. The proof is almost trivial: the following chain of inequalities hold by definition<br />

n ∗ (A ∗ ) = lim n(cA(λ)) = lim cn(A)(λ) = n(A)<br />

λ↑Ω λ↑Ω ∗<br />

so the thesis follows.<br />

With a similar argument, we can show that Numerosity Axiom 1 implies that a<br />

similar property holds for n ∗ <strong>and</strong> for all ξ, ζ ∈ Q ∗ . We begin by showing that, for<br />

all a, b ∈ Q, then the ratio n ∗ ([a, b) ∗ ) · n ∗ ([0, 1) ∗ ) assumes the expected value.<br />

Proposition 7.8.2. By assuming Numerosity Axiom 1, for all a, b ∈ Q,<br />

n ∗ ([a, b) ∗ )<br />

n ∗ ([0, 1) ∗ )<br />

= b − a<br />

Proof. With an argument similar to the proof of Proposition 7.8.1,<br />

n ∗ ([a, b) ∗ )<br />

n ∗ ([0, 1) ∗ )<br />

n(c[a,b)(λ))<br />

= lim<br />

λ↑Ω n(c[0,1)(λ))<br />

= lim(b<br />

− a) = b − a<br />

λ↑Ω<br />

In the same spirit, we can easily extend the equality of Proposition 7.8.2 to all<br />

ξ, ζ ∈ Q ∗ .<br />

Proposition 7.8.3. By assuming Numerosity Axiom 1, for all ξ, ζ ∈ Q∗ , if ξ =<br />

limλ↑Ω ϕ(λ) <strong>and</strong> ζ = limλ↑Ω ψ(λ) <strong>and</strong> by calling [ξ, ζ)R∗ = limλ↑Ω [ϕ(λ), ψ(λ)), we<br />

have<br />

n∗ ([ξ, ζ)R∗) n∗ ([0, 1) ∗ = ξ − ζ<br />

)<br />

Proof. By definition,<br />

n ∗ ([ξ, ζ)R ∗)<br />

n ∗ ([0, 1) ∗ )<br />

n([ϕ(λ), ψ(λ)))<br />

= lim<br />

λ↑Ω n([0, 1))<br />

= lim<br />

λ↑Ω ϕ(λ) − lim<br />

λ↑Ω ψ(λ) = ξ − ζ<br />

Corollary 7.8.4. By assuming Numerosity Axiom 1, for all ξ, ζ ∈ R∗ , if ξ =<br />

limλ↑Ω ϕ(λ) <strong>and</strong> ζ = limλ↑Ω ψ(λ) <strong>and</strong> by calling [ξ, ζ)R∗ = limλ↑Ω [ϕ(λ), ψ(λ)), the<br />

following equality holds<br />

n∗ ([ξ, ζ)R∗) n∗ ([0, 1) ∗ ∼ ξ − ζ<br />

)<br />

109


CHAPTER 7. FURTHER IDEAS<br />

Proof. This result follows from Proposition 7.8.3 <strong>and</strong> by the fact that the Ω-limit of<br />

functions that take only infinitesimal values is infinitesimal.<br />

Now, we will make precise the statement that n∗ is hyperfinitely additive. If<br />

{An}n∈N are pairwise disjointed subsets of R <strong>and</strong> µ = limλ↑Ω ϕ(λ) ∈ N∗ , then the<br />

following equality holds by definition:<br />

n ∗<br />

⎛ ⎛ ⎞⎞<br />

ϕ(λ) <br />

⎝lim ⎝<br />

ϕ(λ) <br />

⎠⎠<br />

= lim n(An) (7.3)<br />

λ↑Ω<br />

n=1<br />

An<br />

λ↑Ω<br />

n=1<br />

<br />

∗ Now, it would be nice to relate the first member of equation 7.3 to n n∈N An<br />

∗ .<br />

From Proposition 1.7.11, we know that<br />

⎛<br />

ϕ(λ) <br />

lim ⎝<br />

⎞ <br />

<br />

⎠ ⊆<br />

∗ (7.4)<br />

λ↑Ω<br />

n=1<br />

An<br />

<strong>and</strong> the equality holds only if the union is finite. Since we are interested in the case<br />

where the union is infinite, we already know that the two sets can’t be equal.<br />

It seems that we are facing the same difficulties we encountered when studying<br />

numerosity: even if, in this case, we have equation 7.3, we don’t know if it’s of<br />

any help. In fact, in the next Example, we will show that this impression is wrong<br />

<strong>and</strong>, by assuming Numerosity Axiom 1, we can already use equation 7.3 to get some<br />

interesting results.<br />

Example 7.8.5. Consider the sets {[1/2 n+1 , 1/2 n )}n≥0. It’s clear that<br />

<br />

n≥0<br />

n∈N<br />

An<br />

[1/2 n+1 , 1/2 n ) = (0, 1)<br />

Now, pick µ = max(N◦ ) <strong>and</strong>, thanks to equation 7.3, consider the equality chain<br />

n∗ max(λ) limλ↑N n=0 [1/2n+1 , 1/2n <br />

)<br />

n∗ ([0, 1) ∗ max(λ) 1 1<br />

= lim = 1 − ∼ 1<br />

)<br />

λ↑N 2n+1 2 µ+1<br />

n=0<br />

On the other h<strong>and</strong>, in this case,<br />

from which we deduce<br />

(0, 1) ∗ \<br />

(0, 1) \<br />

⎛<br />

⎝lim<br />

λ↑N<br />

k<br />

n=0<br />

<br />

[1/2 n+1 , 1/2 n ) = (0, 1/2 n+1 )<br />

max(λ)<br />

[1/2 n+1 , 1/2 n ) ⎠ = (0, 1/2 µ+1 )R∗ n=0<br />

<strong>and</strong>, thanks to Corollary 7.8.4, we already knew that<br />

110<br />

n ∗ ((0, 1/2 µ+1 )R ∗)<br />

n ∗ ([0, 1) ∗ )<br />

⎞<br />

∼ 1/2 µ+1 ∼ 0


7.8. THE NATURAL EXTENSION OF NUMEROSITY<br />

We do believe that the ideas used in the previous example could be extended to<br />

other, more general cases <strong>with</strong>out much effort. We want to explicitely observe that,<br />

thanks to Proposition 7.8.1, some results about natural extension of numerosity can<br />

be used to infer similar properties of numerosity, <strong>and</strong> this applies in particular for<br />

any result about countable additivity.<br />

It’s our hope to deepen the study of this topic in the near future.<br />

111


BIBLIOGRAPHY<br />

Bibliography<br />

[1] Vieri Benci, Mauro Di Nasso, Alpha <strong>Theory</strong>: <strong>Mathematics</strong> <strong>with</strong> <strong>Infinite</strong> <strong>and</strong><br />

<strong>Infinite</strong>simal <strong>Numbers</strong> (paper in preparation, 2012)<br />

[2] Vieri Benci, Mauro Di Nasso, Alpha <strong>Theory</strong>: An Elementary Axiomatics for<br />

Nonst<strong>and</strong>ard Analysis (Expositiones Mathematicae, 21-2003)<br />

[3] Gianni Gilardi, Analisi Matematica di Base (McGraw-Hill, 2001)<br />

[4] Gianni Gilardi, Analisi Tre (McGraw-Hill, 1994)<br />

[5] Gianni Gilardi, Analisi Funzionale, available for free download at http://wwwdimat.unipv.it/gilardi/WEBGG/PSPDF/analfunz1011.pdf<br />

[6] Haim Brezis, Carlo Sbordone, Analisi Funzionale, Teoria ed applicazioni - Con<br />

un’appendice su Integrazione Astratta di Carlo Sbordone (Liguori Editore, 1986)<br />

[7] Gianni Gilardi, Misure di Hausdorff e applicazioni, available for free download<br />

at http://www-dimat.unipv.it/gilardi/WEBGG/PSPDF/hausd.pdf<br />

[8] I. N. Herstein, Algebra (Editori Riuniti, Roma 1994)<br />

[9] R. Schoof, B. van Geemen, Algebra, available for free download at http://wwwdimat.unipv.it/canonaco/notealgebra.pdf<br />

[10] Andrea Cantini, Pierluigi Minari, Introduzione alla logica - parte<br />

II, available under the authorization of the authors at the page<br />

http://www.philos.unifi.it/CMpro-v-p-60.html<br />

[11] Dirk Van Dalen, Logic <strong>and</strong> Structure (Springer, 2008)<br />

[12] Vieri Benci, Leon Horsten, Sylvia Wenmackers, Non-Archimedean Probability,<br />

available for free download at the page http://arxiv.org/abs/1106.1524<br />

[13] Appunti delle lezioni di Probabilità e Statistica tenute dal professor<br />

Eugenio Regazzini, available for free download at http://wwwdimat.unipv.it/∼bassetti/didattica/probI/probabilitaNEW.pdf<br />

[14] Jean Jacod, Philip Protter, Probability Essentials (Springer, 2003)<br />

[15] Abraham Robinson, Non-St<strong>and</strong>ard Analysis (North-Holl<strong>and</strong>, 1966)<br />

112


BIBLIOGRAPHY<br />

[16] Imre Lakatos, Proofs <strong>and</strong> Refutations - The logic of mathematical discovery<br />

(Cambridge University Press, 1976)<br />

[17] Roberta De Giambattista, Concetti e definizioni dell’analisi matematica dal<br />

punto di vista didattico (Thesis, 2010)<br />

[18] Dan Ginsburg, Brian Groose, Josh Taylor, Bogdan Vernescu, The History of<br />

the Calculus <strong>and</strong> the Development of Computer Algebra Systems, available for<br />

free consultation at http://www.math.wpi.edu/IQP/BVCalcHist/index.html<br />

113


ACKNOWLEDGMENTS AND THANKS<br />

Acknowledgments <strong>and</strong> Thanks<br />

What I’m trying to say is that<br />

there is always something that<br />

makes your day. Just look. And if<br />

you don’t find it, look again. Look<br />

closer.<br />

Iris B.<br />

La possibilità di studiare matematica è stata una delle tre occasioni più<br />

gr<strong>and</strong>i che io abbia mai avuto. Se sono riuscito ad arrivare fino a qui, il<br />

merito non è solamente mio, ma anche di tutte le persone che, in un modo<br />

o nell’altro, mi hanno insegnato ed aiutato a crescere e ad affrontare le sfide di tutti<br />

i giorni. A ciascuno di loro, dedico qualche parola di ricordo e ringraziamento.<br />

Acknowledgment 1. To Eleonora.<br />

Like an autumn day, you are wonderful <strong>and</strong> fascinating.<br />

Like a star, you are bright <strong>and</strong> splendid.<br />

Like a flame that never dies, you are warm <strong>and</strong> welcoming.<br />

Like π, you are real, <strong>and</strong> yet transcendental.<br />

Fortunately, the proof of those statements runs way beyond the scope of this paper.<br />

Ringraziamento 2. Questa tesi non sarebbe mai stata realizzata senza il supporto<br />

costante e significativo della mia famiglia. Sono molto grato ai miei genitori per<br />

avermi appoggiato in ogni momento della mia vita universitaria e non, agevol<strong>and</strong>omi<br />

in ogni necessità, compresa quella di visitare periodicamente Pisa per poter studiare<br />

l’affascinante argomento presentato in questa tesi. Vi ringrazio anche per la fiducia<br />

che mi avete dimostrato in questi anni, che si è sempre manifestata nelle piccole e<br />

nelle gr<strong>and</strong>i cose.<br />

Ringrazio moltissimo anche la Zia, soprattutto per la sua disponibilità di condividere<br />

insieme alcuni momenti impegnativi (come <strong>and</strong>are a comperare i vestiti) e<br />

di riflessione, e per il suo atteggiamento positivo nei confronti del mondo. La tua<br />

presenza serena e il tuo buon senso sono ammirevoli, soprattutto per me che sono<br />

carente di certe qualità.<br />

Ringrazio anche mio fratello Angelo e mia sorella Paola, per tutte le occasioni<br />

di crescita che abbiamo vissuto insieme. Un pensiero particolare è riservato anche a<br />

Marie, che ha agevolato molte di queste occasioni. Mes meilleurs souhaits pour un<br />

avenir heureux!<br />

114


RINGRAZIAMENTI<br />

Ricordo con riconoscenza e gratitudine anche la famiglia Tiengo che, in tutti<br />

questi anni, si è sempre dimostrata molto accogliente e disponibile nei miei confronti.<br />

Vi sono molto grato per tutto quello che avete fatto per me, e per tutte le attenzioni<br />

e le gentilezze che mi avete sempre riservato.<br />

Ringraziamento 3. Il momento è quanto mai opportuno per ricordare e ringraziare<br />

tutti gli amici presenti e passati. Tra questi, un posto di riguardo è dedicato<br />

a Gianluca: la nostra lunga e sfaccettata amicizia è un tesoro da custodire e da<br />

coltivare, a qualsiasi costo. Un ringraziamento speciale anche a tutta la famiglia<br />

Lanati, per avermi sempre accolto come loro “secondo figlio”.<br />

Un ringraziamento particolare è riservato anche ad Andrea, con il quale ho sempre<br />

condiviso degli ottimi momenti di svago e di riflessione. Mi auguro che possano<br />

ripetersi a lungo nel nostro futuro.<br />

Ricordo con piacere anche Maria Grazia Toso e il nostro complesso rapporto<br />

tra “creativo” e “ricettivo” (Ch’ien e K’un), la cui ultima espressione si trova nella<br />

dimostrazione di un Lemma di questa tesi.<br />

Ringrazio anche Giulia, Caterina, Dario Merzi, Dario Mazzoleni, i miei compagni<br />

di corso, tutti gli amici del <strong>SELP</strong> e della Scuola Estiva di Logica, Giovanna, Lorenzo,<br />

Giulio, Nello, Lame, Johnny, Mich, Luca, Andrea Colonna, Florenc e tutti coloro<br />

che in qualche modo hanno lasciato un segno positivo nella mia vita.<br />

Ringraziamento 4. Il mio interesse nei confronti dell’Alpha-calcolo e dell’Alphateoria<br />

è nato durante l’edizione del 2010 della Scuola Estiva di Logica organizzata<br />

dall’AILA. Sono molto grato al professor Pierluigi Minari per avermi dato la possibilità<br />

di conoscere questo evento e a Caterina per averlo condiviso insieme a tante<br />

altre occasioni di studio.<br />

La mia gratitudine va anche ai professori Vieri Benci e Mauro Di Nasso, per<br />

avermi reso partecipe del loro lavoro e per avermi accompagnato con gr<strong>and</strong>issima<br />

disponibilità e cura durante lo studio di questo argomento così affascinante. Ringrazio<br />

molto anche il professor Vitali che, nonostante gli impegni, ha seguito con<br />

impegno, passione e pazienza l’evoluzione di questa tesi. Gli sono particolarmente<br />

riconoscente per i suoi preziosi spunti di riflessione, che hanno sempre portato buoni<br />

frutti.<br />

Sono molto grato al <strong>SELP</strong>, a Giorgio Venturi e a Samuele Maschio per avermi<br />

dato la possibilità di raccontare le basi di questa tesi durante un seminario <strong>SELP</strong> a<br />

Pavia nel maggio 2011.<br />

Ringraziamento 5. Ringrazio in modo particolare il professor Gilardi, il professor<br />

Boffi, la professoressa Reggiani, il professor Antonini e tutti gli altri professori che<br />

ho incontrato durante i miei studi universitari. Vi sono molto grato per la vostra<br />

gr<strong>and</strong>issima disponibilità e per la passione e competenza con cui vi dedicate all’insegnamento.<br />

Grazie a voi, durante i miei studi universitari sono cresciuto non solo<br />

nella mia matematica.<br />

Ricordo anche tutte le maestre e le professoresse di matematica di cui sono stato<br />

alunno, ed in particolare la maestra Lucia, la maestra Elena, la prof. Ghini, la<br />

prof. Bianchi e la prof. Cazzola. Ringrazio anche la professoressa Sartirana e la<br />

professoressa Cova per avermi guidato, valorizzato e fatto crescere nel mio tutt’ora<br />

prezioso rapporto con la letteratura.<br />

115


ACKNOWLEDGMENTS AND THANKS<br />

Ringraziamento 6. Infine, un pensiero di gratitudine e riconoscenza è sempre rivolto<br />

al Professor Marni, al Dottor Carlo Campana, al Dottor Ghio, a Milena e<br />

a tutte le infermiere e specializz<strong>and</strong>e dell’ambulatorio scompensi e trapianti della<br />

Cardiologia, al Professor Mario Viganò, al Dottor Carlo Pellegrini, a tutta la Cardiochirurgia,<br />

alla Dottoressa Barbara Petracci, alla Dottoressa Angela Di Matteo e<br />

a tutte le persone che ho dimenticato ma che mi hanno permesso di essere qui dopo<br />

così tanto tempo e dopo così tante avventure.<br />

116<br />

06-06-2001 − 07-02-2012<br />

Ten thous<strong>and</strong> miles are quite long. . .<br />

. . . but not too far!

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!