ebook img

Hypernatural Numbers as Ultrafilters PDF

0.35 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Hypernatural Numbers as Ultrafilters

HYPERNATURAL NUMBERS AS ULTRAFILTERS 5 MAURO DI NASSO 1 0 2 Abstract. Inthispaperwepresentauseofnonstandardmethods in the theory of ultrafilters and in related applications to combi- n natorics of numbers. a J 3 2 1. Introduction. ] O Ultrafilters are really peculiar and multifaced mathematical objects, L whose study turned out a fascinating and often elusive subject. Re- . h searchers may have diverse intuitions about ultrafilters, but they seem t a to agree on the subtlety of this concept; e.g., read the following quo- m tations: “The space βω is a monster having three heads” (J. van Mill [ [41]); “...the somewhat esoteric, but fascinating and very useful object 1 βN” (V. Bergelson [5]). v 5 The notion of ultrafilter can be formulated in diverse languages of 5 mathematics: in set theory, ultrafilters are maximal families of sets 7 that are closed under supersets and intersections; in measure theory, 5 0 they are described as {0,1}-valued finitely additive measures defined . 1 on the family of all subsets of a given space; in algebra, they exactly 0 correspond to maximal ideals in rings of functions FI where I is a set 5 1 and F is a field. Ultrafilters and the corresponding construction of : ultraproduct are a common tool in mathematical logic, but they also v i have many applications in other fields of mathematics, most notably X in topology (the notion of limit along an ultrafilter, the Stone-Cˇech r a compactification βX of a discrete space X, etc.), and in Banach spaces (the so-called ultraproduct technique). In 1975, F. Galvin and S. Glazer found a beautiful ultrafilter proof of Hindman’s theorem, namely the property that for every finite partition of the natural numbers N = C ∪...∪C , there exists an infinite set X 1 r and a piece C such that all sums of distinct elements from X belong i to C . Since this time, ultrafilters on N have been successfully used i also in combinatorial number theory and in Ramsey theory. The key fact is that the compact space βN of ultrafilters on N can be equipped 2000 Mathematics Subject Classification. 03H05, 03E05,54D80. Key words and phrases. Nonstandard analysis, Ultrafilters, Algebra on βN. 1 2 MAURODI NASSO with a pseudo-sum operation, so that the resulting structure (βN,⊕) is a compact topological left semigroup. Such a space satisfies really intriguing properties thathave direct applicationsinthe study ofstruc- tural properties of sets of integers (See the monography [27], where the extensive research originated from that approach is surveyed.) Nonstandard analysis and ultrafilters are intimately connected. In one direction, ultrapowers are the basic ingredient for the usual con- structionsofmodelsofnonstandardanalysissince W.A.J.Luxemburg’s lecture notes [39] of 1962. Actually, by a classic result of H.J. Keisler, the models of nonstandard analysis are characterized up to isomor- phisms as limit ultrapowers, a class of elementary submodels of ultra- powers which correspond to direct limits of ultrapowers (see [32] and [11, §4.4]). In the other direction, the idea that elements of a nonstandard ex- tension ∗X correspond to ultrafilters on X goes back to the golden years of nonstandard analysis, starting from the seminal paper [40] by W.A.J. Luxemburg appeared in 1969. This idea was then systemati- cally pursued by C. Puritz in [43] and by G. Cherlin and J. Hirschfeld in [12]. In those papers, as well as in Puritz’ follow-up [44], new results about the Rudin-Keisler ordering were proved by nonstandard meth- ods, along with new characterizations of special ultrafilters, such as P-points and selective ultrafilters. (See also [42], where the study of similar properties as in Puritz’ papers was continued.) In [7], A. Blass pushed that approach further and provided a comprehensive treatment of ultrafilter properties as reflected by the nonstandard numbers of the associated ultrapowers. Several years later, J. Hirschfeld [28] showed that hypernatural num- berscanalsobeusedasaconvenient tooltoinvestigatecertainRamsey- like properties. In the last years, a new nonstandard technique based on the use of iterated hyper-extensions has been developed to study partition regularity of equations (see [18, 36]). This paper aims at providing a self-contained introduction to a non- standard theory of ultrafilters; several examples are also included to illustrate the use of such a theory in applications. Forgentleintroductions toultrafilters, see thepapers[33,22]; acom- prehensive reference is the monography [14]. Recent surveys on appli- cations of ultrafilters across mathematics can be found in the book [1]. As for nonstandard analysis, a short but rigorous presentation can be found in [11, §4.4]; organic expositions covering virtually all aspects of nonstandard methods are provided in the books [1, 35, 16]. We remark that here we adopt the so-called external approach, based on the existence of a star-map ∗ that associates an hyper-extension (or HYPERNATURAL NUMBERS AS ULTRAFILTERS 3 nonstandard extension) ∗A to each object A under study, and satisfies the transfer principle. This is to be confronted to the internal view- point as formalized by E. Nelson’s Internal Set Theory IST or by K. Hrb´a˘cek’s Nonstandard Set Theories. (See [31] for a thorough treatise of nonstandard set theories.) Let us recall here the saturation property. A family F has the finite intersection property (FIP for short) if A ∩...∩A 6= ∅ for any choice 1 n of finitely many elements A ∈ F. i Definition 1.1. Let κ be an infinite cardinal. A model of nonstandard analysis is κ-saturated if it satisfies the property: • Let F be a family of internal sets with cardinality |F| < κ. If F has the FIP then A 6= ∅. A∈F T When κ-saturation holds, then every infinite internal set A has a cardinality|A| ≥ κ. Indeed, thefamilyofinternalsets{A\{a} | a ∈ A} has the FIP, and has the same cardinality as A. If by contradiction |A| < κ, then by κ-saturation we would obtain A\{a} =6 ∅, which a∈A is absurd. T With the exceptions of Sections 3 and 4, throughout this paper we will work in a fixed c+-saturated model of nonstandard analysis, where c is the cardinality of the continuum. (We recall that κ+ denotes the successor cardinal of κ. So, κ+-saturation applies to families |F| ≤ κ.) In consequence, our hypernatural numbers will have cardinality |∗N| ≥ c+. 2. The u-equivalence There is a canonical way of associating an ultrafilter on N to each hypernatural number. Definition 2.1. The ultrafilter generated by a hypernatural number α ∈ ∗N is the family U = {X ⊆ N | α ∈ ∗X}. α It is easily verified that U actually satisfies the properties of ultra- α filter. Notice that U is principal if and only if α ∈ N is finite. α Definition 2.2. We say that α,β ∈ ∗N are u-equivalent, and write α∼β, if they generate the same ultrafilter, i.e. U = U . The equiva- u α β lence classes u(α) = {β | β∼α} are called u-monads. u Notice that α and β are u-equivalent if and only if they cannot be separated by any hyper-extension, i.e. if α ∈ ∗X ⇔ β ∈ ∗X for every 4 MAURODI NASSO X ⊆ N. In consequence, the equivalence classes u(α) are characterized as follows: u(α) = {∗X | X ∈ U }. α \ (The notion of filter monad µ(F) = {∗F | F ∈ F} of a filter F was first introduced by W.A.J. LuxembuTrg in [40].) For every ultrafilter U on N, the family {∗X | X ∈ U} is a fam- ily of cardinality c with the FIP and so, by c+-saturation, there exist hypernatural numbers α ∈ ∗N such that U = U. (Actually, the c+- α enlargement property suffices: see Definition 4.1.) In consequence, βN = {U | α ∈ N}. α Thus one can identify βN with the quotient set ∗N/∼ of the u-monads. u Example 2.3. Let f : N → R be bounded. If α∼β are u-equivalent u then ∗f(α) ≈ ∗f(β) are at infinitesimal distance. To see this, for every real number r ∈ R consider the set Γ(r) = {n ∈ N | f(n) < r}. Then, by the hypothesis, one has α ∈ ∗Γ(r) ⇔ β ∈ ∗Γ(r), i.e. ∗f(α) < r ⇔ ∗f(β) < r. As this holds for all r ∈ R, it follows that the bounded hyperreal numbers ∗f(α) ≈ ∗f(β) are infinitely close. (This example was suggested to the author by E. Gordon.) Proposition 2.4. ∗f (u(α)) = u(∗f(α)). Indeed: (1) If α∼β then ∗f(α)∼∗f(β). u u (2) If ∗f(α)∼γ then γ = ∗f(β) for some β∼α. u u Proof. (1). For every A ⊆ N, one has the following chain of equiva- lences: ∗f(α) ∈ ∗A ⇔ α ∈ ∗{n | f(n) ∈ A} ⇔ ⇔ β ∈ ∗{n | f(n) ∈ A} ⇔ ∗f(β) ∈ ∗A. (2). For every A ⊆ N, α ∈ ∗A ⇒ ∗f(α) ∈ ∗(f(A)) ⇔ γ ∈ ∗(f(A)), i.e. γ = ∗f(β) for some β ∈ ∗A. But then the family of internal sets {∗f−1(γ)∩ ∗A | α ∈ ∗A} has the finite intersection property. By c+-saturation, there exists an element β in the intersection of that family. Clearly, ∗f(β) = γ and β∼α. (cid:3) u Before starting to develop our nonstandard theory, let us consider a well-known combinatorial property which constitutes a fundamen- tal preliminary step in the theory of ultrafilters. The proof given be- low consists of two steps: we first show a finite version of the desired property, and then use a non-principal ultrafilter to obtain the global HYPERNATURAL NUMBERS AS ULTRAFILTERS 5 version. Although the result is well-known, this particular argument seems to be new in the literature. Lemma 2.5. Let f : N → N be such that f(n) 6= n for all n. Then there exists a 3-coloring χ : N → {1,2,3} such that χ(n) 6= χ(f(n)) for all n. Proof. We begin by showing the following “finite approximation” to the desired result. • For every finite F ⊂ N there exists χ : F → {1,2,3} such that F χ (x) 6= χ (f(x)) whenever both x and f(x) belong to F. F F We proceed by induction on the cardinality of F. The basis is trivial, because if |F| = 1 then it is never the case that both x,f(x) ∈ F. For the inductive step, notice that by the pigeonhole principle there must be at least one element x ∈ F which is the image under f of at most one element in F, i.e. |{y ∈ F | f(y) = x}| ≤ 1. Now let F′ = F \{x} and let χ′ : F′ → {1,2,3} be a 3-coloring as given by the inductive hypothesis. We want to extend χ′ to a 3-coloring χ of F. To this end, define χ(x) in such a way that χ(x) 6= χ′(f(x)) if f(x) ∈ F, and χ(x) 6= χ′(y) if f(y) = x. This is always possible because there is at most one such element y, and because we have 3 colors at disposal. We now have to glue together the finite 3-colorings so as to obtain a 3-coloring of the whole set N. (Of course, this cannot be done directly, because two 3-colorings do not necessarily agree on the intersection of their domains.) One possible way is the following. For every n ∈ N, fix a 3-coloring χ : {1,...,n} → {1,2,3} such that χ (x) 6= χ (f(x)) n n n whenever both x,f(x) ∈ {1,...,n}. Then pick any non-principal ul- trafilter U on N and define the map χ : N → {1,2,3} by putting χ(k) = i ⇐⇒ Γ (k) = {n ≥ k | χ (k) = i} ∈ U. i n The definition is well-posed because for every k the disjoint union Γ (k)∪Γ (k)∪Γ (k) = {n ∈ N | n ≥ k} ∈ U, 1 2 3 andso exactly one set Γ (k) belongs to U. The function χ is the desired i 3-coloring. In fact, if by contradiction χ(k) = χ(f(k)) = i for some k, thenwecouldpickn ∈ Γ (k)∩Γ (f(k)) ∈ U andhaveχ (k) = χ (f(k)), i i n n against the hypothesis on χ . (The same argument could be used to n extend this lemma to functions f : I → I over arbitrary infinite sets I.) (cid:3) Remark 2.6. The second part of the above proof could also be eas- ily carried out by using nonstandard methods. Indeed, by saturation one can pick a hyperfinite set H ⊂ ∗N containing all (finite) natural 6 MAURODI NASSO numbers. By transfer from the “finite approximation” result proved above, there exists an internal 3-coloring Φ : H → {1,2,3} such that Φ(ξ) 6= Φ(∗f(ξ)) whenever both ξ,∗f(ξ) ∈ H. Then the restriction χ = Φ↿ : N → {1,2,3} gives the desired 3-coloring. N As a corollary, we obtain the Theorem 2.7. Let f : N → N and α ∈ ∗N. If ∗f(α)∼α then ∗f(α) = u α. Proof. If ∗f(α) 6= α, then α ∈ ∗A where A = {n | f(n) 6= n}. Pick any function g : N → N that agrees with f on A and such that g(n) 6= n for all n ∈ N. Since α ∈ ∗A ⊆ ∗{n | g(n) = f(n)}, we have that ∗g(α) = ∗f(α). Apply the previous theorem to g and pick a 3-coloring χ : N → {1,2,3} such that χ(n) 6= χ(g(n)) for all n. Then ∗χ(∗f(α)) = ∗χ(∗g(α)) 6= ∗χ(α). Now let X = {n ∈ N | χ(n) = i} where i = ∗χ(α). Clearly, α ∈ ∗X but ∗f(α) ∈/ ∗X, and hence ∗f(α)6∼α. (cid:3) u Two important properties of u-equivalence are the following. Proposition 2.8. Let α ∈ ∗A, and let f be 1-1 when restricted to A. Then (1) There exists a bijection ϕ such that ∗f(α) = ∗ϕ(α); (2) For every g : N → N, ∗f(α)∼∗g(α) ⇒ ∗f(α) = ∗g(α). u Proof. (1). We can assume that α ∈ ∗N \N infinite, as otherwise the thesis is trivial. Then α ∈ ∗A implies that A is infinite, and so we can partition A = B∪C into two disjoint infinite sets B and C where, say, α ∈ ∗B. Since f is 1-1, we can pick a bijection ϕ that agrees with f on B, so that ∗ϕ(α) = ∗f(α) as desired. (2). By the previous point, ∗f(α) = ∗ϕ(α) for some bijection ϕ. Then ∗g(α)∼∗ϕ(α) ⇒ ∗ϕ−1(∗g(α))∼∗ϕ−1(∗ϕ(α)) = α ⇒ ∗ϕ−1(∗g(α)) = α, u u and hence ∗g(α) = ∗ϕ(α) = ∗f(α). (cid:3) We remark that property (2) of the above proposition does not hold if we drop the hypothesis that f is 1-1. (In Section 3 we shall address thequestion of theexistence of infinite points α ∈ ∗N with the property that ∗f(α)∼∗g(α) ⇒ ∗f(α) = ∗g(α) for all f,g : N → N.) u Proposition 2.9. If ∗f(α)∼β and ∗g(β)∼α for suitable f and g, then u u ∗ϕ(α)∼β for some bijection ϕ. u Proof. By the hypotheses, ∗g(∗f(α))∼∗g(β)∼α and so ∗g(∗f(α)) = α. u u If A = {n | g(f(n)) = n}, then α ∈ ∗A and f is 1-1 on A. By the previous proposition, there exists a bijectionϕ such that ∗f(α) = ∗ϕ(α), and hence ∗ϕ(α)∼β. (cid:3) u HYPERNATURAL NUMBERS AS ULTRAFILTERS 7 We recall that the image of an ultrafilter U under a function f : N → N is the ultrafilter f(U) = {A ⊆ N | f−1(A) ∈ U}. Notice that if f ≡ g, i.e. if {n | f(n) = g(n)} ∈ U, then f(U) = g(U). U Proposition 2.10. For every f : N → N and α ∈ ∗N, the image ultrafilter f(Uα) = U∗f(α). Proof. For every A ⊆ N, one has the chain of equivalences: A ∈ U∗f(α) ⇔ ∗f(α) ∈ ∗A ⇔ α ∈ ∗(f−1(A)) ⇔ ⇔ f−1(A) ∈ U ⇔ A ∈ f(U ). α α (cid:3) Let us now show how the above results about u-equivalence are just reformulation in a nonstandard context of fundamental properties of ultrafilter theory. Theorem 2.11. Let f : N → N and let U be an ultrafilter on N. If f(U) = U then {n | f(n) = n} ∈ U. Proof. Let α ∈ ∗N be such that U = U . By the hypothesis, U = α α f(Uα) = U∗f(α), i.e. α∼u ∗f(α)andso, bythe previous theorem, ∗f(α) = α. But then {n | f(n) = n} ∈ U because α ∈ ∗{n | f(n) = n}. (cid:3) Recall the Rudin-Keisler pre-ordering ≤ on ultrafilters: RK V ≤ U ⇐⇒ f(U) = V for some function f. RK In this case, we say that V is Rudin-Keisler below U (or U is Rudin- Keislerabove V). Itisreadilyverifiedthatg(f(U)) = (g◦f)(U),so≤ RK satisfies the transitivity property, and ≤ is actually a pre-ordering. RK Notice that U ≤ U means that ∗f(β)∼α for some function f. α RK β u Proposition 2.12. U ≤ V and V ≤ U if and only if U ∼= V are RK RK isomorphic, i.e. there exists a bijection ϕ : N → N such that ϕ(U) = V. Proof. Let U = U and V = U . If U ≤ V and V ≤ U, then there α β RK RK exist functions f,g : N → N such that ∗f(α)∼β and ∗g(β)∼α. But u u then, by Proposition 2.9, there exists a bijection ϕ : N → N such that ∗ϕ(α)∼u β, and hence ϕ(U) = U∗ϕ(α) = Uβ = V, as desired. The other implication is trivial. (cid:3) We close this section by showing that all infinite numbers α have “large” and “spaced” u-monads, in the sense that u(α) is both a left and a right unbounded subset of the infinite numbers ∗N\N, and that different elements of u(α)areplaced atinfinite distance. (The property of c+-saturation is essential here.) 8 MAURODI NASSO Theorem 2.13. [43, 44] Let α ∈ ∗N\N be infinite. Then: (1) For every ξ ∈ ∗N, there exists an internal 1-1 map ϕ : ∗N → u(α)∩(ξ,+∞). In consequence, the set u(α)∩(ξ,+∞) contains |∗N|-many elements and it is unbounded in ∗N. (2) For every infinite ξ ∈ ∗N \ N, the set u(α) ∩ [0,ξ) contains at least c+-many elements. In consequence, u(α) is unbounded leftward in ∗N\N. (3) If α∼β and α 6= β, then the distance |α−β| ∈ ∗N\N is infinite. u Proof. (1). Since α is infinite, every X ∈ U is an infinite set and so α for each k ∈ N there exists a 1-1 function f : N → X ∩ (k,+∞). By transfer, for every ξ ∈ ∗N the following internal set is non-empty: Γ(X) = {ϕ : ∗N → ∗X ∩(ξ,+∞) | ϕ internal 1-1}. Notice that Γ(X ) ∩ ··· ∩ Γ(X ) = Γ(X ∩ ··· ∩ X ), and hence the 1 n 1 n family {Γ(X) | X ∈ U } has the finite intersection property. By α c+-saturation, we can pick ϕ ∈ Γ(X). Clearly, range(ϕ) is an X∈Uα internal subset ofu(α)∩(ξ,+∞)Twiththesame cardinalityas∗N. Since range(ϕ) is internal and hyperinfinite, it is necessarily unbounded in ∗N. (2). For any given ξ ∈ ∗N \ N, the family {∗X ∩ [0,ξ) | X ∈ U } α is closed under finite intersections, and all its elements are non-empty. So, by c+-saturation, there exists ζ ∈ ∗X ∩[0,ξ). X\∈Uα Clearlyζ ∈ u(α)∩[0,ξ),andthisshowsthatu(α)isunboundedleftward in ∗N\N. Now fix ξ infinite. By what we have just proved, the family of open intervals G = {(k,ζ) | k ∈ N and ζ ∈ u(α)∩[0,ξ)} hasemptyintersection. SinceG satisfiesthefiniteintersectionproperty, and c+-saturation holds, it must be |G| ≥ c+, and hence also |u(α)∩ [0,ξ)| ≥ c+. (3).Foreveryn ≥ 2,letk betheremainderoftheEuclideandivision n of α by n, and consider the set X = {x·n+k | x ∈ N}. Then α ∈ ∗X n n n and α∼β implies that also β ∈ ∗X , so α−β is a multiple of n. Since u n β 6= α, it must be |α−β| ≥ n. As this is true for all n ≥ 2, we conclude that α and β have infinite distance. (cid:3) HYPERNATURAL NUMBERS AS ULTRAFILTERS 9 3. Hausdorff S-topologies and Hausdorff ultrafilters It is natural to ask about properties of the ultrafilter map: U : ∗N → βN where U : α 7→ U . α We already noticed that if one assumes c+-saturation then U is onto βN, i.e. every ultrafilter on N is of the form U for a suitable α ∈ ∗N. α However, in this section no saturation property will be assumed. As a first (negative) result, let us show that the ultrafilter map is never a bijection. Proposition 3.1. In any model of nonstandard analysis, if the ultra- filter map U : ∗N ։ βN is onto then, for every non-principal U ∈ βN, the set {α ∈ ∗N | U = U} contains at least c-many elements. α Proof. Given a non-principal ultrafilter U on N, for X ∈ U and k ∈ N let Λ(X,k) = {F ∈ Fin(N) | F ⊂ X & |F| ≥ k}, where we denoted by Fin(N) = {F ⊂ N | F is finite}. Notice that the family of sets F = {Λ(X,k) | X ∈ U , k ∈ N} has the finite intersection property. Indeed, Λ(X ,k ) ∩ ...∩ Λ(X ,k ) = Λ(X,k) 1 1 m m where X = X ∩ ... ∩ X ∈ U and k = max{k ,...,k }; and every 1 m 1 m set Λ(X,k) 6= ∅ since all X ∈ U are infinite. Now fix a bijection Φ : Fin(N) → N, and let Γ(X,k) = {Ψ(F) | F ∈ Λ(X,k)}. Then also the family {Γ(X,k) | X ∈ U , k ∈ N} ⊆ P(N) has the FIP, and so we can extend it to an ultrafilter V on N. By the hy- pothesis on the ultrafilter map there exists β ∈ ∗N such that U = V; β in particular, β ∈ ∗Γ(X,k), and so β = ∗(Ψ(G)) for a suitable X,k G ∈ ∗Λ(X,k).TThen G ⊆ ∗X for all X ∈ U, and hence U = U X,k γ for alTl γ ∈ G. Moreover, |G| ≥ k for all k ∈ N, and so G is an in- finite internal set. Finally, we use the following general fact: “Every infinite internal set has at least the cardinality of the continuum”. To prove this last property, notice that if A is infinite and internal then there exists a (internal) 1-1 map f : {1,...,ν} → A for some infi- nite ν ∈ ∗N \N. Now, consider the unit real interval [0,1] and define Ψ : [0,1] → {1,...,ν} by putting Ψ(r) = min{1 ≤ i ≤ ν | r ≤ i/ν}. The map Ψ is 1-1 because Ψ(r) = Ψ(r′) ⇒ |r−r′| ≤ 1/ν ≈ 0 ⇒ r = r′, and so we conclude that c = |[0,1]| ≤ |{1,...,ν}| ≤ |A|, as desired. (Whenc+-saturation holds, then|{α ∈ ∗N | U = U}| ≥ c+ byTheorem α 2.13.) (cid:3) 10 MAURODI NASSO We now show that the ultrafilter map is tied up with a topology that is naturally considered in a nonstandard setting. (The notion of S-topology was introduced by A. Robinson himself, the “inventor” of nonstandard analysis.) Definition 3.2. For every set X, the S-topology on ∗X is the topology having the family {∗A | A ⊆ X} as a basis of open sets. The capital letter “S” stands for “standard”, and in fact hyper- extensions ∗Aare oftencalled standard sets inthe literature ofnonstan- dardanalysis. Theadjective“standard”originatedfromthedistinction between a standard universe and a nonstandard universe, according to the most used approaches to nonstandard analysis. However, such a distinction is not needed, and indeed one can adopt a foundational framework where there is a single mathematical universe, and take hyper-extensions of any object under study (see, e.g., [2]). Every basic open set ∗A is also closed because ∗X \ ∗A = ∗(X \ A), and so the S-topologies are totally disconnected. A first relationship between S-topology and ultrafilter map is the following. Proposition 3.3. The S-topology on ∗N is compact if and only if the ultrafilter map U : ∗N → βN is onto. Proof. According to one of the equivalent definitions of compactness, the S-topology is compact if and only if every non-empty family C of closed sets with the FIP has non-empty intersection C 6= ∅. C∈C Without loss of generality, one can assume that C is a famTily of hyper- extensions. Notice that C = {∗A | i ∈ I} has the FIP if and only if i C′ = {A | i ∈ I} ⊂ P(N) has the FIP, and so we can extend C′ to an i ultrafilter V on N. If the ultrafilter map is onto βN, then V = U for a α suitable α, and therefore α ∈ ∗A 6= ∅. i∈I i Conversely, if U is an ultrafiTlter on N, then C = {∗X | X ∈ U} is a familyofclosed sets withtheFIP.If α isanyelement intheintersection of C, then U = U. (cid:3) α In consequence of the above proposition, the S-topology on ∗N is compact when the c+-saturation property holds. More generally, κ- saturation implies that the S-topology is compact on every hyper- extension ∗X where 2|X| < κ. (Actually, the κ-enlarging property suffices: see Definition 4.1.) A natural question that one may ask is whether the S-topologies are Hausdorff or not. This depends on the considered model, and giving a complete answer turns out to be a difficult issue involving deep set- theoretic matters, which will be briefly discussed below. So, it is not

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.