Introduction to Nonstandard Analysis
by
Bashir Abdel-Fattah [email protected]
Abstract The basic idea of nonstandard analysis is to extend the reals to a field that includes ”infinite” and ”infinitesimal” elements in order to simplify proofs and con- cepts by replacing limits and epsilon-delta proofs by expressions involving infinites- imals. This paper discusses the construction and properties of the hyperreals and some basic results results in differential calculus before moving on to logic and cul- minating in theL´oˆsTheorem/transfer principle. The contents are largely based on the papers (Davis, 2009) and (Rayo, 2015), with lesser contributions from (Fletcher et al., 2017), (Claassens, 2016), (Marker, 2010), (Keef and Guichard, 2015), and (Murnaghan, 2015). Also note that this paper assumes little background knowledge outside of analysis (the first section does begin with a brief high-level overview in terms of abstract algebra, but if the reader is unfamiliar with that material they can safely proceed knowing that everything else they need will be introduced throughout the paper). Contents
1 Basics of Nonstandard Analysis3 1.1 Constructing Infinitesimals...... 3 1.1.1 Definitions and Properties of Filters...... 3 1.1.2 Ultrafilter Construction of the Hyperreals...... 6 1.1.3 Properties of the Hyperreals...... 9 1.1.4 Enriching Sets and Functions to the Hyperreal Setting.... 12 1.2 Basic Analysis using the Hyperreals...... 13
2 The Transfer Principle 16 2.1 Formal Language and First-Order Logic...... 16 2.2 Basics of Transfer...... 19 2.2.1 Ultraproducts and Ultrapowers...... 19 2.2.2L´oˆs’Theorem ...... 21
References 25
2 1
Basics of Nonstandard Analysis 1.1 Constructing Infinitesimals
Although there are several different schemes of varying degrees of sophistication and utility in rigorously introducing infinite and infinitesimal elements to analysis, in this paper we will focus on the ultrafilter construction developed by Abraham Robinson in the 1960s. Robinson’s method, being both one of the oldest nonstandard schemes and relatively straightforward, is well developed in the mathematical literature, and the basic idea of the method is as follows: for any set X and ring R, the set RX of all function f : X → R is a ring under pointwise addition and multiplication
(f + g)(x) = f(x) + g(x)(f · g)(x) = f(x) · g(x)
However, even if R is an integral domain or a field, RX won’t be an integral domain in general due to the presence of zero divisors. For example, let X be any set and let R be any ring. Then, for any subset S ⊆ X, let χS(x) denote the characteristic function ( 1R if x ∈ S χS(x) = 0R if x 6∈ S
Then, letting A and B are two disjoint nonempty subsets of X, neither χA(x) nor χB(x) are uniformly zero, yet (χA · χB)(x) = χA(x) · χB(x) = χA∩B(x) = χ∅(x) is uniformly zero. In the case of the hyperreals ∗R, our aim is the quotient the ring RN of real-valued sequences by a maximal ideal MAX so that RN/MAX is a field. In a heuristic sense, this amounts to finding an equivalence relation =MAX that declares two sequences hani and hbni equal if the set {n ∈ N : an = bn} is ”too large”, so that if hani · hbni = hanbni =MAX h0i, then at least one of hani and hbni contained ”enough” zeros to be equivalent to zero to begin with. This scheme of eliminating zero divisors is accomplished with the notion of filters.
1.1.1 Definitions and Properties of Filters We begin with a definition of filters in terms of their properties. For many of the properties below, there are several equivalent definitions, many of which are more general or more sophisticated than those stated here, but we shall proceed with following in the interests of simplicity.
Definition 1.1.1. A filter on a set X is a subset F of the power set P(X) satisfying the following properties:
3 1. Proper Filter: ∅ 6∈ F
2. Finite Intersection Property: If A, B ∈ F, then A ∩ B ∈ F
3. Superset Property: If A ∈ F and B ⊃ A, then B ∈ F
Additionally, F is said to be a ultrafilter if it also satisfies:
4. Maximality: For any A ⊆ X, either A ∈ F or X \ A ∈ F
F is further said to be a free ultrafilter if it satisfies:
5. Freeness: F contains no finite subsets of X
Note that ultrafilters actually obey a stronger version of the maximality property (of which the above definition is a special case), which we prove below in order to familiarize the reader with the style of reasoning associated with filters:
Lemma 1.1.1. Let U be an ultrafilter on a set X, and let {A1,A2,...,An} be a finite Sn collection of disjoint subsets of S such that i=1 Ai = X. Then there is exactly one set Aj ∈ {A1,A2,...,An} such that Aj ∈ U.
Proof. First, suppose that U contains none of the sets A1,A2,...An. Then, by the maximality property of ultrafilters, (X \ A1), (X \ A2),..., (X \ An) ∈ U, so by the finite intersection property
n n \ [ (X \ Am) = X \ ( Am) = X \ X = ∅ m=1 m=1 is an element of U. This is a contradiction since U must be a proper filter, so we must have that at least one of the sets A1,A2,...An is in U. Now suppose that U contains more than one of the sets A1,A2,...,An. Let Ai and Aj be two distinct sets such that Ai,Aj ∈ U. Then, by the intersection property, we have
Ai ∩ Aj = ∅ ∈ U
This is also a contradiction, so we must have that U contains at most one of the sets A1,A2,...,An. Therefore U contains exactly one of the sets A1,A2,...,An. Examples of filters on a set X include:
1. The trivial filter F = X.
2. The principle filter FA, which is defined in terms of any set A ⊂ X as
FA = {Y ⊂ X : Y ⊃ A}
In the case that A = {a} contains only a single element, then the principle filter Fa is in fact an ultrafilter.
3. The cofinite/Fr´echet filter F co = {Y ⊂ X : X \ Y is finite}. Fr´echet filters are non-principal.
4 All of the above examples are relatively simple and uninteresting for our pur- poses, and it turns out that every free ultrafilter is non-principle, and in fact cannot be explicitly constructed. One might then doubt whether any collection of sets that obeys all five of the given properties even exists, but we can prove the existence of free ultrafilters on any infinite set using Zorn’s Lemma. We will proceed to that proof in short order, but first we need the following helpful lemma.
Lemma 1.1.2. Let F be a filter on a set X, and let A ⊂ X be a set such that A 6∈ F and (X \ A) 6∈ F. Then F can be extended to a filter F 0 on X containing both F and A.
0 Proof. To begin with, assume F 6= ∅ (if F = ∅, then we can just take F = FA, the principle filter generated by the set A, and the result follows trivially), and consider the collection
F 0 = {Y 0 ⊂ X : there exists some Y ∈ F such that Y 0 ⊃ (Y ∩ A)}
We will now verify that F 0 satisfies each of the properties of a filter in order.
1. Proper Filter: First, note that X ∈ F by the superset property (given any Y ∈ F, then X ⊃ Y , so X ∈ F). This means that A 6= ∅, because in the case that A = ∅ we have (X \ A) = X ∈ F, which contradicts the hypotheses of the lemma. Furthermore, for any Y ∈ F we have Y ∩ A 6= ∅, because if Y ∩ A = ∅ then (X \ A) ⊃ Y , which implies that (X \ A) ∈ F by the superset property of F, a contradiction. Therefore, given any Y 0 ∈ F 0, then there is some Y ∈ F such that Y 0 ⊃ (Y ∩ A) 6= ∅, thus Y 0 6= ∅.
0 0 0 2. Finite Intersection Property: Suppose Y1 ,Y2 ∈ F . Then there exist some 0 0 sets Y1,Y2 ∈ F such that Y1 ⊃ (Y1 ∩ A) and Y2 ⊃ (Y2 ∩ A). Since Y1 ∩ Y2 ∈ F 0 0 by the finite intersection property of F, then Y1 ∩ Y2 ⊃ (Y1 ∩ Y2) ∩ A is an element of F 0 by definition.
0 0 3. Superset Property: This follows fairly trivially. Suppose Y1 ∈ F and that 0 0 0 Y2 ⊃ Y1 . Since there exists some Y1 ∈ F such that Y1 ⊃ (Y1 ∩ A), in which 0 0 0 case Y2 ⊃ (Y1 ∩ A) as well, so Y2 ∈ F by definition. Therefore F 0 is a filter, and we can readily verify that A ∈ F 0 and F ⊂ F 0: for any 0 Y ∈ F we have that Y,A ⊃ (Y ∩ A), hence Y,A ∈ F by definition. Note that the above lemma also justifies the choice of terminology for that max- imality property: any set that does not obey the maximality property of ultrafilters is not maximal in the sense that there is a strictly larger filter containing it. We will prove in the following theorem that a maximal filter (in the sense that there is no strictly larger filter containing it) obeys the maximality property of ultrafilters, and thus the two notions are equivalent.
Theorem 1.1.3 (Ultrafilter Lemma). Let F be a filter on the set X. Then F can be extended to an ultrafilter U on X.
Proof. In this proof, we will first show that there is a maximal filter U on X contain- ing F, then demonstrate that U is an ultrafilter. The first step will be accomplished using Zorn’s Lemma, which is equivalent to the Axiom of Choice:
5 Let S be a family of sets. If for each chain C ⊂ S there exists a member of S that contains all members of C, then S contains a maximal member. Recall that a chain is a collection of sets C such for any pair of distinct sets A, B ∈ C, either A ⊂ B or B ⊂ A (thus a chain can be written ... ⊂ A ⊂ B ⊂ C ⊂ ..., justifying the terminology). Then let Φ be the set of filters on X containing F, and let C be any chain of filters F0 = F ⊂ F1 ⊂ F2 ⊂ ... in Φ. Then we claim that ∞ [ G = Fn n=0 is a filter in Φ that contains every member C. We will now verify the properties of filters in order:
1. Proper Filter: This follows trivially, given that ∅ 6∈ Fn for all n ≥ 0 (since each filter Fn is proper), and hence ∅ 6∈ G.
2. Finite Intersection Property: Given any Y1,Y2 ∈ G, then there must exist
some filters Fn1 , Fn2 ∈ C such that Y1 ∈ Fn1 and Y2 ∈ Fn2 . Suppose without
loss of generality that n2 ≤ n1. Then Fn2 ⊂ Fn1 , so both Y1 and Y2 are
elements of Fn1 . By the finite intersection property of Fn1 , we have that
(Y1 ∩ Y2) ∈ Fn1 ⊂ G.
3. Superset Property: Suppose that Y1 ∈ G and that Y2 ⊃ Y1. Since Y1 ∈ Fn for some Fn ∈ C, then Y2 ∈ Fn ⊂ G by the superset property of Fn. Thus G is a indeed a filter, and G contains both F and every element of C, so G is an element of Φ that contains every element of C. Since such a filter can be constructed for every chain in Φ, then by Zorn’s Lemma Φ must contain a maximal member, which we denote U. Since U is an element of Φ, then U is a filter containing F, and it is also clear that U satisfies the finite intersection property: for any set A ⊂ X, if A 6∈ U and (X \ A) 6∈ U then U can be extented to a filter U 0 ⊃ (U ∪ {A}) by Lemma 1.1.2, contradicting the maximality of U. Therefore U must contain either A of (X \ A) for any A ⊂ X (it cannot contain both because then by the finite intersection property U would also contain the empty set, contradicting the fact that U is a proper filter), thus U is an ultrafilter on X. Given that any filter can be extended to an ultrafilter, we can also demonstrate the existence of a free ultrafilter U on an infinite set X by extending the cofinite filter F co on X. For any finite set A ⊂ X, F co contains (X \ A) because (X \ A) is infinite (X is infinite and A is finite), and hence (X \ A) ∈ U because U ⊃ F co. In particular, U does not contain A (as before, it cannot contain both X and (X \ A) because then U would also contain the empty set, a contradiction), so U cannot contain any finite sets.
1.1.2 Ultrafilter Construction of the Hyperreals Definition 1.1.2 (Equivalence Modulo an Ultrafilter). Given an ultrafilter U on N N, then define an equivalence relation ≡U on R as follows: given any real-valued N sequences hani, hbni ∈ R ,
hani ≡U hbni if and only if {n ∈ N : an = bn} ∈ U
6 Proof. Recall that a relation ∼ on a set X is an equivalence relation if it satisfies the following three properties: 1. Reflexivity: (∀x ∈ X): x ∼ x
2. Symmetry: (∀x, y ∈ X): x ∼ y =⇒ y ∼ x
3. Transitivity: (∀x, y, z ∈ X):(x ∼ y and y ∼ z) =⇒ x ∼ z
One can readily verify the reflexivity, symmetry, and transitivity of the relation ≡U N N on R . Given any sequence hani ∈ R , then the set {n ∈ N : an = an} = N is an element of the ultrafilter U by the maximality principle (either N or N \ N = ∅ is an element of U, but ∅ 6∈ U because U is a proper filter), so hani ≡U hani. For symmetry, if hani ≡U hbni, then {n ∈ N : an = bn} = {n ∈ N : bn = an} ∈ U, so hbni ≡U hani trivially. Finally, suppose hani ≡U hbni and hbni ≡U hcni. Then {n ∈ N : an = bn} and {n ∈ N : bn = cn} are both elements of the ultrafilter U, so {n ∈ N : an = bn and bn = cn} = {n ∈ N : an = bn} ∩ {n ∈ N : bn = cn} ∈ U by the finite intersection property. Since {n ∈ N : an = cn} ⊃ {n ∈ N : an = bn and bn = cn}, then {n ∈ N : an = cn} ∈ U by the superset property, so hani ≡U hcni. Therefore the given relation is indeed a valid equivalence relation. Recall that the equivalence class of an element a of a set X equipped under an equivalence relation ∼ is the set of all elements b ∈ X such that b ∼ a. Then going N forwards we will use [hani] to denote the equivalence class of the sequence hani ∈ R under ultrafilter equivalence, where [hani] = [hbni] if and only if hani ≡U hbni (i.e., if and only if {n ∈ N : an = nn} ∈ U). We will also denote the set of all such N equivalence classes by R / ≡U . Lemma 1.1.4. The operations of pointwise addition and multiplication
[hani] + [hbni] = [han + bni]
[hani] ∗ [hbni] = [hanbni]
N are well-defined binary operations on R / ≡U .
N Proof. Let hani, hαni, hbni, hβni ∈ R be real valued sequences such that [hani] = [hαni] and [hbni] = [hβni]. Then {n ∈ N : an = αn} and {n ∈ N : bn = βn} are both elements of U, so {n ∈ N : an = αn and bn = βn} = {n ∈ N : an = αn} ∩ {n ∈ N : bn = βn} ∈ U by the finite intersection property. The sets {n ∈ N : an+bn = αn+βn} and {n ∈ N : anbn = αnβn} are both supersets of {n ∈ N : an = αn and bn = βn}, so both are also elements of U by the superset property. Therefore
[hani] + [hbni] = [han + bni] = [hαn + βni] = [hαni] + [hβni]
[hani] ∗ [hbni] = [hanbni] = [hαnβni] = [hαni] ∗ [hβni] Thus the results of addition and multiplication as defined are independent of which representatives are chosen for each equivalence class, and hence the given operations are well defined.
N Theorem 1.1.5. Given an ultrafilter U on N, then R / ≡U is a field under pointwise addition and multiplication.
7 Proof. Perhaps the most concrete definition of a field is the following: a field is a set F equipped with binary operations + and · such that for all x, y, z ∈ F 1. x + y = y + x (commutativity of addition) 2.( x + y) + z = x + (y + z) (associativity of addition) 3. There exists some element 0 ∈ F such that x + 0 = x for all x ∈ F (existence of an additive identity) 4. For each x ∈ F there is some element −x ∈ F such that x + (−x) = 0 (existence of additive inverses) 5. x · y = y · x (commutativity of multiplication) 6.( x · y) · z = x · (y · z) (associativity of multiplication) 7. There esists some element 1 ∈ F such that x · 1 = x for all x ∈ F (existence of a multiplicative identity) 8. For each nonzero element x ∈ F there exists some element x−1 ∈ F such that x · x−1 = 1 (existence of multiplicative inverses) 9.( x + y) · z = x · z + y · z (distributivity)
N Although most of the above properties are evident in the case of R / ≡U (for in- stance, commutativity and associativity of addition and multiplication are inherited directly from the corresponding properties of the field R), we will nonetheless show each property in order for definiteness.
1.[ hani] + [hbni] = [han + bni] = [hbn + ani] = [hbni] + [hani] 2. [hani] + [hbni] + [hcni] = [han + bni] + [hcni] = [h(an + bn) + cni]
= [han + (bn + cn)i] = [hani] + [hbn + cni] = [hani] + [hbni] + [hcni]
3. Consider the element [h0i], the equivalence class of the sequence that is iden- N tically zero. Then for any [hani] ∈ R / ≡U ,
[hani] + [h0i] = [han + 0i] = [hani]
N Thus [h0i] is the additive identity in R / ≡U .
N 4. For each [hani] ∈ R / ≡U , define −[hani] = [h−ani]. Then [hani] + − [hani] = [hani] + [h−ani] = [han − ani] = [h0i]
5.[ hani] ∗ [hbni] = [hanbni] = [hbnani] = [hbni] ∗ [hani] 6. [hani] ∗ [hbni] ∗ [hcni] = [hanbni] ∗ [hcni] = [h(anbn)cni]
= [han(bncn)i] = [hani] ∗ [hbncni] = [hani] ∗ [hbni] ∗ [hcni]
8 7. Consider the element [h1i], the equivalence class of the sequence of all ones. N Then for any [hani] ∈ R / ≡U ,
[hani] ∗ [h1i] = [han · 1i] = [hani]
N Thus [h1i] is the multiplicative identity in R / ≡U .
8. For any equivalence class [hani] 6= [h0i], since we would have [hani] = [h0i] if {n ∈ N : an = 0} ∈ U, then it must be true that {n ∈ N : an = 0} 6∈ U. By the maximality property, this means that {n ∈ N : an 6= 0} = N \{n ∈ N : an = 0} ∈ U. Then consider, for instance, the real-valued sequence defined by ( an if an 6= 0 αn = 1 if an = 0
Since {n ∈ N : αn = an} = {n ∈ N : an 6= 0} ∈ U, then [hαni] = [hani]. Because αn 6= 0 for all n ∈ N, then we can define the inverse
−1 −1 [hαni] = [hαn i]
with the property
−1 −1 −1 −1 [hani] ∗ [hαni] = [hαni] ∗ [hαni] = [hαni] ∗ [hαn i] = [hαnαn i] = [h1i]
9. [hani] + [hbni] ∗ [hcni] = [han + bni] ∗ [hcni] = [h(an + bn)cni]
= [hancn + bncni] = [hancni] + [hbncni]
= [hani] ∗ [hcni] + [hbni] ∗ [hcni]
1.1.3 Properties of the Hyperreals In order to show that this newly-constructed field contain ”infinite” and ”infinitesi- N mal” numbers, we first need some notion of size on R / ≡U . We accomplish this in N analogy to the definition of equality on R / ≡U :
Definition 1.1.3 (Inequality Modulo an Ultrafilter). Given an ultrafilter U on N, N define the relation ≤U on R / ≡U as follows: given any two elements [hani], [hbni] ∈ N R / ≡U , [hani] ≤U [hbni] if and only if {n ∈ N : an ≤ bn} ∈ U N Then ≤U is a total ordering on R / ≡U . Proof. Recall that a relation on a set X is a (partial) ordering of X if it is
1. Reflexive: (∀x ∈ X): x x
2. Anti-symmetric: (∀x, y ∈ X):(x y and y x) =⇒ x = y
3. Transitive: (∀x, y, z ∈ X):(x y and y z) =⇒ x z
9 Additionally, is a total ordering on X if it has the additional property that for all x, y ∈ X, either x y or y x. The proof that ≤U is a partial ordering proceeds analogously to the case of N equality modulo and ultrafilter. Given element [hani] ∈ R / ≡U , then the set {n ∈ N : an ≤ an} = N is an element of the ultrafilter U by the maximality principle (either N or N \ N = ∅ is an element of U, but ∅ 6∈ U because U is a proper filter), so [hani] ≤U [hani]. For transitivity, suppose [hani] ≤U [hbni] and [hbni] ≤U [hcni]. Then {n ∈ N : an ≤ bn} and {n ∈ N : bn ≤ cn} are both elements of the ultrafilter U, so {n ∈ N : an ≤ bn and bn ≤ cn} = {n ∈ N : an ≤ bn} ∩ {n ∈ N : bn ≤ cn} ∈ U by the finite intersection property. Since {n ∈ N : an ≤ cn} ⊃ {n ∈ N : an ≤ bn and bn ≤ cn}, then {n ∈ N : an ≤ cn} ∈ U by the superset property, so [hani] ≤U [hcni]. To demonstrate that ≤U is anti-symmetric, suppose [hani] ≤U [hbni] and [hbni] ≤U [hani]. Then {n ∈ N : an ≤ bn} and {n ∈ N : bn ≤ an} are both elements of U, so {n ∈ N : an = bn} = {n ∈ N : an ≤ bn and bn ≤ an} = {n ∈ N : an ≤ bn} ∩ {n ∈ N : bn ≤ an} ∈ U by the finite intersection property. Therefore [hani] = [hbni]. Finally, to show that ≤U is a total ordering, consider any two elements [hani] and N [hbni] in R / ≡U . If {n ∈ N : an ≤ bn} ∈ U, then [hani] ≤U [hbni] and we are done. If {n ∈ N : an ≤ bn} 6∈ U, then {n ∈ N : an > bn} = N \{n ∈ N : an ≤ bn} ∈ U by the maximality property, and {n ∈ N : an ≥ bn} ⊃ {n ∈ N : an > bn} is an element of U by the superset property. Therefore [hbni] ≤U [hani].
N Now that we know that R / ≡U is an ordered field, we will suppress the details ∗ N of its construction when not necessary going forward. We define R = R / ≡U to be the set of hyperreals, and we will write, for instance,
∗ (∃ a, b ∈ R): a ≤ b instead of N (∃ [hani], [hbni] ∈ R / ≡U ):[hani] ≤U [hbni] Note that we now take U to be a free ultrafilter, although none of the properties of ∗R that we have demonstrated thus far rely on freeness. We also introduce the notation σ N R for the set of equivalence classes of constant-valued sequences in R / ≡U , which is an embedding of the standard real numbers R into the hyperreals ∗R that is naturally isomorphic to R:
N a ∈ R ←→ [hai] = [ha, a, a, . . .i] ∈ R / ≡U We can now proceed to state the definition of infinite and infinitesimal numbers in ∗R in terms of the function ( a if a ≥ −a |a| = max(a, −a) = −a if a ≤ −a
∗ which is well-defined on R (if a ≤ −a and −a ≤ a, then {n ∈ N : an ≤ −an} ∈ U and {n ∈ N : −an ≤ an} ∈ U, so {n ∈ N : an = −an} = {n ∈ N : an = 0} = {n ∈ N : an ≤ −an} ∩ {n ∈ N : −an ≤ an} is an element of U by the finite intersection property, and hence a = −a = 0).
Definition 1.1.4 (Infinite and Infinitesimal Numbers). Let σR+ denote the embed- ding of R+ = {r ∈ R : r > 0} into ∗R. Then a hyperreal number a ∈ ∗R is finite
10 if there exists some r ∈ σR+ such that |a| ≤ r, infinitesimal if |a| ≤ r for every r ∈ σR+, and infinite if |a| ≥ r for every r ∈ σR+. We denote the set of all finite numbers in ∗R by O, and the set of all infinitesimal numbers of ∗R by ϑ (which is a proper subset of O).
σ ∼ σ Note that the only infinitesimal in the standard reals R = R is 0, and R doesn’t contain any infinite elements. In order to demonstrate the existence of nonzero infinitesimals and infinite numbers in the hyperreals, consider the elements 1 1 1 1 σ + [hni] = [h1, 2, 3, 4,...i] and [h n i] = [h1, 2 , 3 , 4 ,...i]. Then for any [hri] ∈ R , there are only finitely many terms such that n < r and finitely many terms such 1 1 that n > r, so {n ∈ N : n < r} and {n ∈ N : n > r} are not elements of U because U is free (that is, it contains no finite sets). Therefore {n ∈ N : n ≥ r} 1 and {n ∈ N : n ≤ r} are elements of U by the maximality property of ultrafilters, 1 ∗ + 1 and hence [hni] ≥U [hri] and [h n i] ≤U [hri] for every [hri] ∈ R . Thus [h n i] is infinitesimal and [hni] is infinite. Definition 1.1.5 (Infinitesimal Closeness). Two finite hyperreals a, b ∈ O are said to be infinitesimally close if a − b ∈ ϑ, which we denote a ≈ b. The relation ≈ is an equivalence relation. Proof. The reflexivity of ≈ is clear: for all a ∈ O, a − a = 0 ∈ ϑ. That the relation is symmetric and transitive is also clear from the definition of an infinitesimal, since the negative of an infinitesimal is also infinitesimal and the sum of two infinitesi- mals is also infinitesimal (these assertions follow trivially from the definition of an infinitesimal). For any a, b ∈ O, if a − b ∈ ϑ, then b − a = −(a − b) ∈ ϑ, so a ≈ b =⇒ b ≈ a. Similarly, for any a, b, c ∈ O, if a − b ∈ ϑ and b − c ∈ ϑ, then a − c = (a − b) + (b − c) ∈ ϑ, so if a ≈ b and b ≈ c, then a ≈ c.
Definition 1.1.6. Every finite hyperreal a ∈ O is infinitesimally close to a unique standard real number in σR, which we call the standard part of a and denote st(a). Proof. The existence of such a standard real number is perhaps intuitive, but the proof requires results from abstract algebra and hence is beyond the scope of this paper. Instead, we shall demonstrate the uniqueness of such a number. Suppose that c, c0 ∈ σR are such that c ≈ a and c0 ≈ a. Then, by transitivity, c ≈ c0 and hence c−c0 ∈ ϑ. Since c and c0 are both standard real numbers, then their difference is also a standard real number, and because there is only one standard real number 0 in ϑ, it follows that c − c = 0. Note that there are some drawbacks to the ultrafilter construction of the hyper- reals that later methods have sought to rectify, the foremost issue being that it is not possible to determine any free ultrafilters on N (the best we can do is demonstrate their existence). As a result, the order relation on ∗R is not explicitly known, and one might puzzle over trying to order sequences such as the following:
h1, 0, 1, 0, 1, 0,...i
h0, 1, 0, 1, 0, 1,...i 1 1 1 1 h1, , 3, , 5, , 7, ,...i 2 4 6 8
11 For the first two sequences above, one must be equivalent to h1i modulo U and the other must be equivalent to h0i modulo U, but it is impossible to tell which one is which. For the third sequence, one might wonder whether the sequence represents an infinite number, an infinitesimal, or something in between. Also of concern is that different ultrafilters result in distinct fields, and it is an open problem whether or not these field will turn out to be isomorphic. However, in the section on transfer, we will find that the results that hold in these fields don’t depend on the ultrafilter, because any two such fields are so-called elementary equivalent.
1.1.4 Enriching Sets and Functions to the Hyperreal Setting
∗ N For every set A ⊂ R, we can associate the natural extension A ⊂ R / ≡U by
∗ [hxni] ∈ A if and only if {n ∈ N : xn ∈ A} ∈ U
The above definition can be readily extended to Cartesian products: given sequences 1 2 m xn, xn, . . . , xn (where it is understood that the superscripts index the sequence in this case and do not denote exponentiation) and a set A ∈ Rm, we define