Graduate Texts inMathematics Robert Goldblatt Lectures on the Hyperreals An Introduction to Nonstandard Analysis ay SpringerSpringer New York Berlin Heidelberg Barcelona Budapest Hong Kong London Milan Paris Singapore Tokyo Graduate Texts in Mathematics 1 8 8 Editorial Board S. Axler E,W. Gehring K.A. RibetGraduate Texts in Mathematics 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 TAKEUTI/ZARING. Introduction to Axiomatic Set Theory. 2nd ed. Oxtosy. Measure and Category. 2nd ed. SCHAEFER. Topological Vector Spaces. HitTon/STAMMBACH. A Course in Homological Algebra. 2nd ed. Mac Lane. Categories for the Working Mathematician. 2nd ed. HuGHES/PIPER. Projective Planes. SERRE. A Course in Arithmetic. TAKEUTI/ZARING. Axiomatic Set Theory. HuMPHREYS. Introduction to Lie Algebras and Representation Theory. ConEN. A Course in Simple Homotopy Theory. Conway. Functions of One Complex Variable I. 2nd ed. BEALS. Advanced Mathematical Analysis. ANDERSON/FULLER. Rings and Categories of Modules. 2nd ed. GOLUBITSKY/GUILLEMIN. Stable Mappings and Their Singularities. BERBERIAN. Lectures in Functional Analysis and Operator Theory. WINTER. The Structure of Fields. ROSENBLATT. Random Processes. 2nd ed. HALMOS. Measure Theory. Hatwos. A Hilbert Space Problem Book. 2nd ed. [Link]. Fibre Bundles. 3rd ed. Humpureys. Linear Algebraic Groups. BARNES/Mack. An Algebraic Introduction to Mathematical Logic. Greus. Linear Algebra. 4th ed. Hocmes. Geometric Functional Analysis and Its Applications. HEWITT/STROMBERG. Real and Abstract Analysis. Manes. Algebraic Theories. KELLEY. General Topology. ZARISKI/SAMUEL. Commutative Algebra. Vol.I. ZARISKI/SAMUEL, Commutative Algebra. [Link]. JAcoBson, Lectures in Abstract Algebra I. Basic Concepts, JACOBSON. Lectures in Abstract Algebra II. Linear Algebra. Jacosson. Lectures in Abstract Algebra IIT. Theory of Fields and Galois Theory. 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 31 52 33 54 55 56 57 58 59 60 61 Hirscu. Differential Topology. Sprrzer. Principles of Random Walk. 2nd ed. ALEXANDER/WERMER. Several Complex Variables and Banach Algebras. 3rd ed. KELLEY/NAMIOKA et al. Linear Topological Spaces. Monk. Mathematical Logic. GRAUERT/FRITZSCHE. Several Complex Variables. ARVESON. An Invitation to C*-Algebras. KEMENY/SNELL/KnapP. Denumerable Markov Chains. 2nd ed. AposTot. Modular Functions and Dirichlet Series in Number Theory. 2nd ed. SERRE. Linear Representations of Finite Groups. GILLMAN/JERISON, Rings of Continuous Functions. KENpIG. Elementary Algebraic Geometry. Loévs. Probability Theory I. 4th ed. Love. Probability Theory IL. 4th ed. MolsE. Geometric Topology in Dimensions 2 and 3. Sacus/Wu. General Relativity for Mathematicians. GRUENBERG/WEIR. Linear Geometry. 2nd ed. Epwaros. Fermat’s Last Theorem. KLINGENBERG. A Course in Differential Geometry. HaARTSHORNE. Algebraic Geometry. Manin. A Course in Mathematical Logic. GraveR/WATKINS. Combinatorics with Emphasis on the Theory of Graphs. Brown/PeEarcy. Introduction to Operator Theory I: Elements of Functional Analysis. Massey. Algebraic Topology: An Introduction. CROWELL/Fox. Introduction to Knot Theory. Kosirz. p-adic Numbers, p-adic Analysis, and Zeta-Functions. 2nd ed. Lana. Cyclotomic Fields. ARNOLD. Mathematical Methods in Classical Mechanics. 2nd ed. WHITEHEAD. Elements of Homotopy Theory. (continued after index)Robert Goldblatt Lectures on the Hyperreals An Introduction to Nonstandard Analysis @)) SpringerRobert Goldblatt School of Mathematical and Computing Sciences Victoria University Wellington New Zealand Editorial Board 8. Axler F.W. Gehring K.A. Ribet Mathematics Department Mathematics Department Mathematics Department San Francisco State East Hall University of California University University of Michigan at Berkeley San Francisco, CA 94132 Ann Arbor, MI 48109 Berkeley, CA 94720-3840 USA USA USA Mathematics Subject Classification (1991): 26E35, 03H05, 28E05 Library of Congress Cataloging-in-Publication Data Goldblatt, Robert. Lectures on the hyperreals : an introduction to nonstandard analysis / Robert Goldblatt. Pp. cm. — (Graduate texts in mathematics ; 188) Includes bibliographical references and index. ISBN 0-387-98464-X (hardcover : alk. paper) 1. Nonstandard mathematical analysis. I. Title. ID. Series. QA299.82.G65 1998 515—DC21 98-18388 Printed on acid-free paper. © 1998 Springer-Verlag New York, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or here- after developed is forbidden. The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the former are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone. Production managed by Victoria Evarretta; manufacturing supervised by Jacqui Ashri. Photocomposed pages prepared from the author’s LATEX files. Printed and bound by Maple- Vail Book Manufacturing Group, York, PA. Printed in the United States of America. 987654321 ISBN 0-387-98464-X Springer-Verlag New York Berlin Heidelberg SPIN 10659819Preface There are good reasons to believe that nonstandard analysis, in some ver- sion or other, will be the analysis of the future. Kurt GODEL This book is a compilation and development of lecture notes written for a course on nonstandard analysis that I have now taught several times. Students taking the course have typically received previous introductions to standard real analysis and abstract algebra, but few have studied formal logic. Most of the notes have been used several times in class and revised in the light of that experience. The earlier chapters could be used as the basis of a course at the upper undergraduate level, but the work as a whole, including the later applications, may be more suited to a beginning graduate course. This preface describes my motivations and objectives in writing the book. For the most part, these remarks are addressed to the potential instructor. Mathematical understanding develops by a mysterious interplay between intuitive insight and symbolic manipulation. Nonstandard analysis requires an enhanced sensitivity to the particular symbolic form that is used to ex- press our intuitions, and so the subject poses some unique and challenging pedagogical issues. The most fundamental of these is how to turn the trans- fer principle into a working tool of mathematical practice. I have found itvi Preface unproductive to try to give a proof of this principle by introducing the formal Tarskian semantics for first-order languages and working through the proof of Los’s theorem. That has the effect of making the subject seem more difficult and can create an artifical barrier to understanding. But the practical use of transfer is more readily explained informally, and typically involves statements that are no more complicated than the “epsilon-delta” statements used in standard analysis. My approach then has been to illus- trate transfer by many examples, with demonstrations of why those exam- ples work, leading eventually to a situation in which its formulation as a general principle appears quite credible. There is an obvious analogy with standard laws of thought, such as induction. It would be an unwise teacher who attempted to introduce this to the novice by deriving the principle of induction as a theorem from the axioms of set theory. Of course one attempts to describe induction, and explain how it is applied. Eventually after practice with examples the student gets used to using it. So too with transfer. It is sensible to use this approach in many areas of mathematics, for instance beginning a course on standard analysis with a description of the real number system R as a complete ordered field. The student already has well-developed intuitions about real numbers, and the axioms serve to suromarise the essential information needed to proceed. It is rare these days to find a text that begins by explicitly constructing R out of the rationals via Dedekind cuts or Cauchy sequences, before embarking on the theory of limits, convergence, continuity, etc. On the other hand, it is not so clear that such a methodology is ade- quate for the introduction of the hyperreal field *R itself. In view of the controversial history of infinitesimals, and the student’s lack of familiar- ity with them, there is a plausibility problem about simply introducing *R axiomatically as an ordered field that extends R, contains infinitesimals, and has various other properties. I hope that such a descriptive approach will eventually become the norm, but here I have opted to use the founda- tional, or constructive, method of presenting an ultrapower construction of the ordered field structure of *R, and of enlargements of elementary sets, relations, and functions on R, leading to a development of the calculus, analysis, and topology of functions of a single variable. At that point (Part III) the exposition departs from some others by making an early introduc- tion of the notions of internal, external, and hyperfinite subsets of *R, and internal functions from *R to *R, along with the notions of overflow, under- flow, and saturation. It is natural and helpful to develop these important and radically new ideas in this simpler context, rather than waiting to ap- ply them to the more complex objects produced by constructions based on superstructures. As to the use of superstructures themselves, again I have taken a slightly different tack and followed (in Part IV) a more axiomatic path by positing the existence of a universe U containing all the entities (sets, tuples, rela-Preface vii tions, functions, sets of sets of functions, etc., etc.) that might be needed in pursuing a particular piece of mathematical analysis. U is described by set- theoretic closure properties (pairs, unions, powersets, transitive closures). The role of the superstructure construction then becomes the foundational one of showing that universes exist. From the point of view of mathemat- ical practice, enlargements of superstructures seem somewhat artificial (a “gruesome formalism”, according to one author), and the approach taken here is intended to make it clearer as to what exactly is the ontology that we need in order to apply nonstandard methods. Looking to the future, if (one would like to say when) nonstandard analysis becomes as widely recognised as its standard “shadow”, so that a descriptive approach with- out any need for ultrapowers is more amenable, then the kind of axiomatic account developed here on the basis of universes would, I believe, provide an effective and accessible style of exposition of the subject. What does nonstandard analysis offer to our understanding of math- ematics? In writing these notes I have tried to convey that the answer includes the following five features. (1) New definitions of familiar concepts, often simpler and more intu- itively natural Examples to be found here include the definitions of convergence, boundedness, and Cauchy-ness of sequences; continuity, uniform con- tinuity, and differentiability of functions; topological notions of inte- rior, closure, and limit points; and compactness. (2 = New and insightful (often simpler) proofs of familiar theorems In addition to many theorems of basic analysis about convergence and limits of sequences and functions, intermediate and extreme values and fixed points of continuous functions, critical points and inverses of differentiable functions, the Bolzano—Weierstrass and Heine—Borel theorems, the topology of sets of reals, etc., we will see nonstandard proofs of Ramsey’s theorem, the Stone representation theorem for Boolean algebras, and the Hahn—Banach extension theorem on linear functionals. (3) New and insightful constructions of familiar objects For instance, we will obtain integrals as hyperfinite sums; the reals R themselves as a quotient of the hyperrationals *Q; other comple- tions, including the p-adic numbers and standard power series rings as quotients of nonstandard objects; and Lebesgue measure on R by a nonstandard counting process with infinitesimal weights.viii Preface (4) New objects of mathematical interest Here we will exhibit new kinds of number (limited, unlimited, in- finitesimal, appreciable); internal and external sets and functions; shadows; halos; hyperfinite sets; nonstandard hulls; and Loeb mea- sures. (5) Powerful new properties and principles of reasoning These include transfer; internal versions of induction, the least num- ber principle and Dedekind completeness; overflow, underflow, and other principles of permanence; Robinson’s sequential lemma; satu- ration; internal set definition; concurrence; enlargement; hyperfinite approximation; and comprehensiveness. In short, nonstandard analysis provides us with an enlarged view of the mathematical landscape. It represents yet another stage in the emergence of new number systems, which is a significant theme in mathematical history. Its rich conceptual framework will be built on to reveal new systems and new understandings, so its development will itself influence the course of that history.Contents Foundations What Are the Hyperreals? 1.1 Infinitely Small and Large... . 2... . 2.2 1.2 Historical Background ..........-.-..---.. 1.38 What Isa Real Number?..................-. 1.4 Historical References .........-2....2..-0-0-- Large Sets 2.1 Infinitesimals as Variable Quantities ............ 2.2 lLargeness ... 2-2... . 2 eee ee eee ee 2.3 Filters... 2.2... ee ee ee 2.4 Examples of Filters.............-....-2--- 2.5 Facts About Filters... 2.2... 0.0... .00002-0- 2.6 Zorn’sLemma......... 00.0000 ee eee eee 2.7‘ Exercises on Filters... 20-0. ....202.-00-00-. Ultrapower Construction of the Hyperreals 3.1 The Ring of Real-Valued Sequences. ............ 3.2 Equivalence Modulo an Ultrafilter......--......- 3.3. Exercises on Almost-Everywhere Agreement ........ 3.4 A Suggestive Logical Notation. ...-........-.-. 3.5 Exercises on Statement Values. ...-.........-.x Contents 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 The Ultrapower . 2... ee ee Including the Reals in the Hyperreals........-...- Infinitesimals and Unlimited Numbers ........... Enlarging Sets. 2... es Exercises on Enlargement .........-...-.204. Extending Functions .. 2... 0... 0 ee ee eee Exercises on Extensions .. 2... 00. ee eee ee ee Partial Functions and Hypersequences ........... Enlarging Relations. .........-.....20000-. Exercises on Enlarged Relations. ...........-.-- Is the Hyperreal System Unique? .........-.... 4 The Transfer Principle 4.1 4.2 4.3 4.4 4.5 4.6 4.7 Transforming Statements. ........-.....-00-. Relational Structures... .....--0.00.--2.0-000-- The Language of a Relational Structure .......... x-Transforms .. 2... ee ee ee The Transfer Principle ...........-....00-. Justifying Transfer 2.0... ee ee ee Extending Transfer... .........0-.--+0000. 5 Hyperreals Great and Small 5.1 5.2 5.3 5.4 5.5 5.10 5.11 5.12 (Un)limited, Infinitesimal, and Appreciable Numbers . . . Arithmetic of Hyperreals.........-....-.204. On the Use of “Finite” and “Infinite?............ Halos, Galaxies, and Real Comparisons. ..........- Exercises on Halos and Galaxies. .............- Shadows... 2... ee Exercises on Infinite Closeness. . 2... 0-2... ee ae Shadows and Completeness ...........0+.200- Exercise on Dedekind Completeness ............ The Hypernaturals ...............0 00000. Exercises on Hyperintegers and Primes........... On the Existence of Infinitely Many Primes ........ II Basic Analysis 6 Convergence of Sequences and Series 6.1 6.2 6.3 6.4 6.5 6.6 Convergence... 2... ee Monotone Convergence... ........-.-0+ +0008 Limits 2... ee ee Boundedness and Divergence ...........-...4-. Cauchy Sequences ... 2... 0.0. eee ee Cluster Points... 1... ee ee eeContents xi 6.7 Exercises on Limits and Cluster Points. .......... 66 6.8 Limits Superior and Inferior ................. 67 6.9 Exercises on limsup and liminf ............... 70 6.10 Series 2... 2... ee ee 71 6.11 Exercises on Convergence of Series ...-.......-. 71 Continuous Functions 75 7.1 Cauchy’s Account of Continuity... ...........0. 75 7.2 Continuity of the Sine Function .....-.......-4. 77 7.3 Limits of Functions... 2... 2... 00.0002 eee 78 74 Exercises on Limits... 1... ee 78 7.5 The Intermediate Value Theorem .............-. 79 7.6 The Extreme Value Theorem ...........-....-.- 80 7.7 Uniform Continuity... 0.2... 2. ee ee ee 81 7.8 Exercises on Uniform Continuity ...........0.4. 82 7.9 Contraction Mappings and Fixed Points .......... 82 7.10 A First Look at Permanence... ..........000. 84 7.11 Exercises on Permanence of Functions ........... 85 7.12 Sequences of Functions... 2... 0.0.2 200-402 ee 86 7.13 Continuity of a Uniform Limit... ..-..0....004. 87 7.14 Continuity in the Extended Hypersequence ........ 88 7.15 Was Cauchy Right?.... 0.0.02. 2.0000 00000- 90 Differentiation 91 8.1 The Derivative ©... ee 91 8.2 Increments and Differentials ...........-...--. 92 8.3 Rules for Derivatives ...........-..000000. 94 84 ChainRule ...........-.-..- 0000+ eee 94 8.5 Critical Point Theorem... ..-..-..--.--2-.0-. 95 8.6 Inverse Function Theorem .............-..0-4. 96 8.7 Partial Derivatives .....- [Link]... eee eee 97 8.8 Exercises on Partial Derivatives... .........00. 100 8.9 TaylorSeries ......-.. [Link] cee eee eee 100 8.10 Incremental Approximation by Taylor’s Formula ..... 102 8.11 Extending the Incremental Equation ......-..... 103 8.12 Exercises on Increments and Derivatives .......... 104 The Riemann Integral 105 9.1 Riemann Sums ...... 2... 0.0 eee es 105 9.2 The Integral as the Shadow of Riemann Sums ....... 108 9.3 Standard Properties of the Integral... .........-. 110 9.4 Differentiating the Area Function... ........... 111 9.5 Exercise on Average Function Values ......-..... 112xii Contents 10 Topology of the Reals 10.1 Interior, Closure, and Limit Points ....... 10.2 Open and Closed Sets ............0.. 10.3 Compactness ........0..0.00000. 10.4 Compactness and (Uniform) Continuity .... 10.5 Topologies on the Hyperreals .......... III Internal and External Entities 11 Internal and External Sets 11.1 Internal Sets... ...........0-0.00-- 11.2 Algebra of Internal Sets ............. 11.3. Internal Least Number Principle and Induction 114 The Overflow Principle.............. 11.5 Internal Order-Completeness .......... 11.6 External Sets ........0..-..2.---.-. 11.7 Defining Internal Sets ...........-... 11.8 The Underflow Principle.........-... 11.9 Internal Sets and Permanence .......... 11.10 Saturation of Internal Sets. .......-...- 11.11 Saturation Creates Nonstandard Entities... . 11.12 The Size of an Internal Set. ..........0. 11.13 Closure of the Shadow of an Internal Set... . 11.14 Interval Topology and Hyper-Open Sets .... 12 Internal Functions and Hyperfinite Sets 12.1 Internal Functions .........-..-... 12.2 Exercises on Properties of Internal Functions . 12.3 Hyperfinite Sets... .........2.0-040. 12.4 Exercises on Hyperfiniteness........... 12.5 Counting a HyperfiniteSet ........... 12.6 Hyperfinite Pigeonhole Principle ........ 12.7 Integrals as HyperfiniteSums .......... IV Nonstandard Frameworks 13 Universes and Frameworks 13.1 What Do We Need in the Mathematical World? 13.2 Pairs Are Enough. .........--.0-0-. 13.3 Actually, Sets Are Enough. ........... 13.4 Strong Transitivity .............000. 13.5 Universes .. 2.2... ee ee 13.6 Superstructures.........-...0. 0004 113 113 115 116 119 120 123 125 125 127 128 129 130 131 133 136 137 138 140 141 142 143 147 147 148 149 150 151 151 15213.7 13.8 13.9 13.10 13.11 13.12 13.13 13.14 13.15 13.16 13.17 13.18 13.19 13.20 Contents The Language of a Universe... 2... ......2000.-. Nonstandard Frameworks ..........-..2.0005 Standard Entities... 2... ee ee Internal Entities. 2... ee ee Closure Properties of Internal Sets ............. Transformed Power Sets ......-..00.00 00000 G Exercises on Internal Sets and Functions. .........- External Images Are External. .......-.....-. Internal Set Definition Principle. ...........-0.. Internal Function Definition Principle ........... Hyperfiniteness ©... ee Exercises on Hyperfinite Sets and Sizes... ........ Hyperfinite Summation... 2... ee Exercises on HyperfiniteSums.............0.-. 14 The Existence of Nonstandard Entities 14.1 14.2 14.3 14.4 Enlargements 2... ee Concurrence and Hyperfinite Approximation. ......-. Enlargements as Ultrapowers ......-...-.-000. Exercises on the Ultrapower Construction ......... 15 Permanence, Comprehensiveness, Saturation 15.1 15.2 15.3 15.4 15.5 Permanence Principles... 2.2.2... 2.000000 ee Robinson’s Sequential Lemma ..........-.-.-.. Uniformly Converging Sequences of Functions ....... Comprehensiveness .......-.. 00+ ee eee nee Saturation... ee ee ee ee V_ Applications 16 Loeb Measure 16.1 16.2 16.3 16.4 16.5 16.6 16.7 16.8 Rings and Algebras... 1... 2. eee ee ee ee Measures... 1. ee ee es Outer Measures... 0. ee Lebesgue Measure .. 2... ee ee ee ee es Loeb Measures 6... ee ees prApproximability 2... 0... 0. . ee ee ee Loeb Measure as Approximability.....-......... Lebesgue Measure via Loeb Measure ............ 17 Ramsey Theory 17.1 17.2 17.3 Colourings and Monochromatic Sets ............ A Nonstandard Approach ............00..200. Proving Ramsey’s Theorem ...........-..24-. xiiixiv Contents 17.4 The Finite Ramsey Theorem ................ 17.5 The Paris-Harrington Version ...........+.0... 17.6 Reference... 0... ee 18 Completion by Enlargement 18.1 Completing the Rationals ...........-...00.0. 18.2 Metric Space Completion ............2-0005 18.3 Nonstandard Hulls ...........--...0200005 18.4 p-adicIntegers 2.2... 0... 0... eee 18.5 padic Numbers. ............000..00000. 18.6 Power Series... 2... 0... eee ee es 18.7 Hyperfinite Expansions in Basep .........-.... 18.8 Exercises... 0. ee es 19 Hyperfinite Approximation 19.1 Colourings and Graphs. ............-0--04- 19.2 Boolean Algebras... 1.0... ee eee ee 19.3 Atomic Algebras .... 2... 0... 02.0 eee eae 19.4 Hyperfinite Approximating Algebras ....-....... 19.5 Exercises on Generation of Algebras .......-..-.. 19.6 Connecting with the Stone Representation ......... 19.7 Exercises on Filters and Lattices ........-..... 19.8 Hyperfinite-Dimensional Vector Spaces........... 19.9 Exercises on (Hyper) Real Subspaces............ 19.10 The Hahn-Banach Theorem. ...........-....0. 19.11 Exercises on (Hyper) Linear Functionals .......... 20 Books on Nonstandard Analysis Index 257 259 260 262 265 267 269 269 272 273 275 275 278 279 283Part I Foundations1 What Are the Hyperreals? 1.1 Infinitely Small and Large A nonzero number ¢ is defined to be infinitely small, or infinitesimal, if le| < 2 for alln =1,2,3,.... In this case the reciprocal w = i will be infinitely large, or simply infinite, meaning that |w| > n for all n = 1,2,3,.... Conversely, if a number w has this last property, then 2 will be a nonzero infinitesimal. However, in the real number system R there are no such things as nonzero infinitesimals and infinitely large numbers. Our aim here is to study a larger system, the hyperreals, which form an ordered field *R that contains R as a subfield, but also contains infinitely large and small numbers according to these definitions. The new entities in “R, and the relationship between *R and R, provide an intuitively appealing alternative approach to real analysis and topology, and indeed to many other branches of pure and applied mathematics.4 1. What Are the Hyperreals? 1.2 Historical Background Our mathematical heritage owes much to the creative endeavours of people who found it natural to think in terms of the infinite and the-infinitesimal. By examining the words with which they expressed their ideas we can learn much about the origins of our twentieth-century perspective, even if that perspective itself makes it difficult, perhaps impossible, to recapture faithfully the “mind-set” of the past. Archimedes An old idea that has never lost its potency is to think of a geometric object as made up of an “unlimited” number of “indivisible” elements. Thus a curve might be regarded as a polygon with infinitely many sides of infinitesimal length, a plane figure as made up of parallel straight line segments viewed as strips of infinitesimal width, and a solid as composed of infinitely thin plane laminas. The formula A = arc for the area of a circle in terms of its radius and circumference was very likely discovered by regarding the circle as made up of infinitely many segments consisting of isosceles triangles of height r with infinitesimal bases, these bases collectively forming the circle itself. In the third century Bc., Archimedes gave a proof of this formula using the method of exhaustion that had been developed by Eudoxus more than a century earlier. This involved approximating the area arbitrarily closely by regular polygons. From the modern point of view we would say that as the number of sides increases, the sequence of areas of the polygons converges to the area of the circle, but the Greek mathematicians did not develop the idea of taking the limit of an infinite sequence. Instead, they used an indirect reductio ad absurdum argument, showing that if the area was not equal to A = arc, then by taking polygons with sufficiently many sides a contradiction would follow. Archimedes applied this approach to give proofs of many formulae for areas and volumes involving circles, parabolas, ellipses, spirals, spheres, cylinders, and solids of revolution. He wrote a treatise called The Method of Mechanical Theorems in which he explained how he discovered these formulae. His method was to imagine geometrical figures as being connected by a lever that is held in balance as the elements of one figure whose magnitude (area or volume) and centre of gravity is known are weighed against the elements of another whose magnitude is to be determined. These elements are as above: line segments in the case of plane figures, with length as the comparative “weight”; and plane laminas in the case of solids, weighted according to area.’ Archimedes did not regard this procedure as 1A lucid illustration of the “Method” is given on pages 69-70 of the book1.2 Historical Background 5 providing a proof, but said of a result obtained in this way that this has not therefore been proved, but a certain impression has been created that the conclusion is true. The demonstration of its truth was then to be supplied by the method of exhaustion. The lesson of history is that the way in which a mathematical fact is discovered may be very different from the way that it is proven. Indeed Archimedes’ treatise, along with all knowledge of his “method”, was lost for many centuries and found again only in 1906. Newton and Leibniz In the latter part of the seventeenth century the differential and integral calculus was discovered by Isaac Newton and Gottfried Leibniz, indepen- dently. Leibniz created the notation dx for the difference in successive values of a variable x, thinking of this difference as infinitely small or “less than any assignable quantity”. He also introduced the integral sign {, an elon- gated “S” for “sum”, and wrote the expression f ydz to mean the sum of all the infinitely thin rectangles of size y x dz. He expressed what we now know as Leibniz’s rule for the differential of a product xy in the form dry = xdy + ydz. To demonstrate this he first observed that dxzy is the same thing as the difference between two successive xy’s; let one of these be xy, and the other x + dx into y + dy. Then calculating dzy = (x+dx)(y + dy) —xy ady+ydz + dx dy, Il he stated that the desired result follows by the omission of the quantity dx dy, which is infinitely small in comparison with the rest, for it is supposed that dx and dy are infinitely small. Leibniz’s views on the actual existence of infinitesimals make interesting reading. In response to certain criticisms, he drew attention to the fact that Archimedes and others found out their wonderfully elegant theorems by the help of such ideas; these theorems they completed with reductio ad absurdum by C.H. Edwards cited in Section 1.4, showing how it yields the area under the graph of y = x” between 0 and 1.6 1. What Are the Hyperreals? proofs, by which they at the same time provided rigorous demon- strations and also concealed their methods, and went on to write: It will be sufficient if, when we speak of infinitely great (or more strictly unlimited), or of infinitely small quantities (i.e., the very least of those within our knowledge), it is understood that we mean quantities that are indefinitely great or indefinitely small, i.e., as great as you please, or as small as you please, so that the error that one may assign may be less than a certain as- signed quantity ... by infinitely great and infinitely small we un- derstand something indefinitely great, or something indefinitely small, so that each conducts itself as a sort of class, and not merely as the last thing of a class ...it will be sufficient sim- ply to make use of them as a tool that has advantages for the purpose of calculation, just as the algebraists retain imaginary roots with great profit. Further indication of this attitude is found in the following passage from an argument in one of his manuscripts: If dx, ddx ... are by a certain fiction imagined to remain, even when they become evanescent, as if they were infinitely small quantities (and in this there is no danger, since the whole matter can be always referred back to assignable quan- tities), then... Newton’s formulation of the calculus used a different language and had a more dynamic conception of the phenomena under discussion. He consid- ered fluents x,y,... aS quantities varying in a spatial or temporal sense, and their fluzions 2, y,... as the speeds with which they flow and are increased by their gen- erating motion. In modern parlance, the fluxion « is the derivative de of x with respect to time ¢ (or the velocity of «). Newton wrote (1671): The moments of the fluent quantities (that is, their indefinitely small parts, by addition of which they increase during each in- finitely small period of time) are as their speeds of flow ...if the moment of any particular one, say x, be expressed by the product of its speed and an infinitely small quantity o (that is by to) ...it follows that quantities x and y after an infinitely small interval of time will become x + £0 and y + yo. Con- sequently, an equation which expresses a relationship of fluent1.2 Historical Background 7 quantities without variance at all times will express that rela- tionship equally between x + to and y + yo as between x and y; and sox+az0 and y+ yo may be substituted in place of the latter quantities, x and y, in the said equation. In other words, if (x,y) is a point on the curve defined by an equation in xz and y, then (x + #0,y + yo) is also on the curve. But this does not seem right: surely (2 +0, y+ yo) should lie on the tangent to the curve, the line through (z,y) of slope 4/x, rather than on the curve itself? Moreover, in making the proposed substitution and carrying out algebraic calculations, Newton permitted himself to divide by the infinitely small quantity o while at the same time stating that since o is supposed to be infinitely small so that it be able to ex- press the moments of quantities, terms which have it as a factor will be equivalent to nothing in respect of others. I therefore cast them out... which seems to amount to equating o to zero. Such perplexities are typical of the confusions caused by the concepts of infinitesimal calculus. In later writing Newton himself tried to explain his theory of fluxions in terms of limits of ratios of quantities. He wrote that he did not (unlike Leibniz) consider Mathematical Quantities as composed of Parts ea- treamly small, but as generated by a continual motion, and that fluzions are very nearly as the Augments of the Fluents. His conception of limits is conveyed by the following passages: Quantities, and the ratios of quantities, which in any finite time converge continually to equality, and before the end of time ap- proach nearer to each other than by any given difference, become ultimately equal ... Those ultimate ratios with which quantities vanish are not truly the ratios of ultimate quantities, but limits towards which the ratios of quantities decreasing without limit do always converge; and to which they approach nearer than by any given difference, but never go beyond, nor in effect attain to, till the quantities are diminished ad infinitum. Newton considered that the use of limits of ratios provided an adequate basis for his calculus, without ultimately depending on indivisibles: In Finite Quantities so to frame a Calculus, and thus to inves- tigate the Prime and Ultimate Ratios of Nascent or Evanescent Finite Quantities, is agreeable to the Ancients; and I was willing8 1. What Are the Hyperreals? to shew, that in the Method of Fluxions there’s no need of intro- ducing Figures infinitely small into Geometry. For this Analysis may be performed in any Figures whatsoever, whether finite or infinitely small, so they are but imagined to be similar to the Evanescent Figures ... Euler The greatest champion of infinitely small and large numbers was Leonhard Euler, said to be the most prolific of all mathematicians. He simply assumed that such things exist and behave like finite numbers. A good illustration of his approach is to be found in the book Introduction to the Analysis of the Infinite (1748), where he developed infinite series for logarithmic, exponential, and trigonometric functions from the following basis: Let w be an infinitely small number, or a fraction so small that, although not equal to zero, stilla” = 1+, where » is also an infinitely small number ...we let b = kw. Then we have a” = 1+ kw, and with a as the base for the logarithms, we have w = log(1 + kw) ... If now we let j = 2, where z denotes any finite number, since w is infinitely small, then j is infinitely large. Then we have w = 4, where w is represented by a fraction with an infinite denominator, so that w is infinitely small, as it should be. Euler took it for granted that Newton’s formula for the binomial series works for his numbers, and applied it to the expansion of a* = a“) = (1+ kw)? to deduce that kz k?z? 323 i i a ee and hence when z = 1 that - 4 k ke kB a= tirtateato: In fact, since kw = Be , the general term ( 5) kw)” of the binomial series for a* should be . . . iG =-YG—2)--G-ntt) x" nl jr? Since j is infinitely lage, £2 — = 1, and the larger the number we substitute for j, the closer the value of the fraction iS comes to 1. Therefore, f j is a number larger than any assignable number, then 3 i =1, 28 = 1, and so forth.1.2 Historical Background 9 His next step was a natural one: Since we are free to choose the base a for the system of loga- rithms, we now choose a in such a way that k = 1... we obtain the value for a = 2.71828182845904523536028. When. this base is chosen, the logarithms are called natural or hyperbolic. The latter name is used since the quadrature of a hyperbola can be expressed through these logarithms. For the sake of brevity for this number 2.718281828459--- we will use the symbole ... Whereas the modern view is that 1 n e= lim (1+2) , NOS n Euler had obtained it by stipulating that e = (1 + ry, and indeed e* = (1 + 2)’, for infinitely large j. In this way he “proved” that z 22 2A ef = ltgtgtgto log(l+z) = rE yR Ey ., and also showed that ge a eff 4 ee cos = loatac a ; eo eit _ enix sin x x 37 Br a by using the equations cosw = 1, sinw =w, andj =j-l=j—-—2=::- with w infinitely small and j infinitely large. Euler’s demonstration that the function e* is equal to its own derivative employed the practice, which, as we saw, was adopted by Leibniz and New- ton, of “casting out” higher-order infinitesimals like dx dy, (dx)*, (dx)*, ete. Applying his series expansion for the exponential function to e@” he argued that d(e*) = ettt _ et = e*(e#* —1) - 2 (dz)? _ (dx)? = et (de + + ay te) = e*dr.10 1. What Are the Hyperreals? Demise of Infinitesimals The conceptual foundations of the calculus continued to be controversial and to attract criticism, the most famous being that of Berkeley, who wrote (1734) in opposition to the ideas of Newton and his followers: And what are these fluzions? The velocities of evanescent incre- ments? And what are these same evanescent increments? They are neither finite quantities, nor quantities infinitely small, nor yet nothing. May we not call them the ghosts of departed quan- tities? Eventually infinitesimals were expunged from analysis, along with the de- pendence on intuitive geometric concepts and diagrams. The subject was “arithmetised” by the explicit construction of the real numbers out of the rational number system by the work of Dedekind, Cantor, and oth- ers around 1872. Weierstrass provided the purely arithmetical formulation of limits that we use today, defining lim,_,, f(z) = L to mean that (Ve > 0) (36 > 0) such that 0 < |x — a| < 6 implies | f(x) — L| }. The set of all Dedekind cuts of Q can be made into a complete ordered field. (4) A real number is an equivalence class of Cauchy sequences of ratio- nal numbers. A sequence (r1,72,73,... ) is Cauchy if its terms get arbitrarily close to each other as we move along the sequence, i.e., lim [rn —1m| = 0. n,M—oCoO Thus V2 is the limit of the rational Cauchy sequence 1, 1.4, 1.41, 1.414, 1.4142, 1.41421, 1.414213, ... as well as being the limit of any of the subsequences of this sequence, and of other rational sequences besides. Two Cauchy sequences (r1,72,73,... ) and (81,82, 83,... ) are equiv- alent if their corresponding terms approach each other arbitrarily closely: lim |rn — $n| = 0. noo This defines an equivalence relation on the set of rational-valued Cauchy sequences, and the resulting set of equivalence classes forms a complete ordered field. Any two equivalent Cauchy sequences will have the same limit, and so represent the same real number. For ex- ample, /2 corresponds to the equivalence class of the above sequence.1.3 What Is a Real Number? 13 Answer (2) provides the basis for the axiomatic or descriptive approach to the analysis of R. The object of study is simply described as being a complete ordered field, since all its properties derive from that fact. The axioms for a complete ordered field are listed, and everything follows from that. This is by far the favoured approach in introductory texts on real analysis. The constructive approach takes as given only the rational number sys- tem and proceeds to construct R explicitly. There are at least two ways to do this, due respectively to Dedekind (answer (3)) and Cantor (answer (4). It would be possible to develop an axiomatic approach to the hyperreals *R by assuming that we are dealing with an ordered field containing R as well as infinitesimals and satisfying the transfer principle “appropriately formulated”. However, in view of the controversial history of the notion of infinitesimal, one could be forgiven for wondering whether this is an exercise in fantasy, or whether there does exist a number system satisfying the proposed axioms. The constructive approach is needed to resolve this issue. We will be discussing a construction of *R out of R that is analogous to Cantor’s construction of R out of Q. Hyperreal numbers will arise as equivalence classes of real-valued sequences, and the challenge will be to find an equivalence relation on such sequences that produces the desired outcome. To conclude this introduction to our subject, let us examine another putative answer to the question “what is a real number?”—namely, that a real number is a point on the number line: Now, the intuitive geometric idea of a line is an ancient one, much older than the notion of a set of points, let alone an infinite set. The identification of a line with the set of points lying on that line is a perspective that belongs to modern times. For Euclid a line was simply “ length without breadth”, and his diagrams and arguments involved lines with a finite number of points marked on them. By applying the field operations and taking limits of converging sequences we can assign a point to each real number, but the claim that this exhausts all the points on the line is just that: a claim. One could seek to justify it by invoking a principle such as the one attributed to Eudoxus and Archimedes that any two magnitudes are such that the less can be multiplied so as to exceed the other. This entails that for each real number r there is an integer n > r, and that precludes there being any infinitely large or small numbers in R. But then one could say that the Eudoxus—Archimedes principle is just a property of those points on the line that correspond to “assignable” numbers. The14 1. What Are the Hyperreals? hyperreal point of view is that the geometric line is capable of sustaining a much richer and more intricate number set than the real line. 1.4 Historical References Amongst the numerous books available, the following are worth consulting for more details on the historical background we have been discussing. M. E. BARON AND H. J. M. Bos. Newton and Leibniz. Open Uni- versity Press, 1974. J. M. Cuitp. The Early Mathematical Manuscripts of Leibniz. Open Court Publishing Co., 1920. E. J. DIJKSTERHUIS. Archimedes. Princeton University Press, 1987. C. H. Epwarps. The Historical Development of the Calculus, Springer, 1979. LEONHARD EULER. Introduction to the Analysis of the Infinite, Book I, translated by John D. Blanton. Springer, 1988.2 Large Sets 2.1 Infinitesimals as Variable Quantities Cauchy (1789-1857) is regarded as one of the pioneers of the precision that is characteristic of contemporary mathematics. He wrote: My principal aim has been to reconcile rigor, which I have made a law to myself in my Cours d’analyse, with the simplicity which the direct consideration of infinitely small quantities produces. His method was to consider infinitesimals as being variable quantities that vanish: When the successive numerical values of a variable decrease in- definitely so as to be smaller than any given number, this vari- able becomes what is called infinitesimal, or infinitely small quantity .... One says that a variable quantity becomes in- Jinitely small when its value decreases numerically so as to con- verge to the limit zero. Even today there are textbooks containing statements to the effect that a sequence satisfying lim r, = 0 NCO is an infinitesimal, while one satisfying lim rp = co no16 2. Large Sets is an infinitely large magnitude. Can we then construct a number system in which such sequences represent, infinitely small and large numbers respec- tively? According to Cauchy, the sequence 1 pees ble il 99737 is an infinitesimal, as is lPidid 2747967877" ** If these represent infinitely small numbers, perhaps we should regard the second as being half the size of the first because it converges twice as quickly? Similarly, the sequences 1,2,3,4,..., 2,4,6,8,... both represent infinitely large magnitudes, and arguably the second is twice as big as the first because it diverges to oo twice as quickly. On the other hand, the distinct sequences 1,2,3,4,..., 2,2,3,4,... will presumably represent the same infinite number. These ideas are attractive because they suggest the possibility of using infinitely small and large numbers as measures of rates of convergence. But in the construction of real numbers out of Cauchy sequences (Section 1.3), all sequences converging to zero are identified with the number zero itself, while diverging sequences have no role to play at all. Clearly then we need a very different kind of equivalence relation among sequences than the one used in Cantor’s construction of R from Q. 2.2 Largeness Let r = (ri,r2,T3,...) and s = (31, 82, 83,... ) be real-valued sequences. We are going to say that r and s are equivalent if they agree at a “large” number of places, i.e., if their agreement set Eys = {ni Tn = 8n} is large in some sense that is to be determined. Whatever “large” means, there are some properties we will want it to have: e N= {1,2,3,... } must be large, in order to ensure that any sequence will be equivalent to itself.2.2 Largeness 17 e Equivalence is to be a transitive relation, so if E,, and FE; are large, then E,; must be large. Since E,, Fz C E,+, this suggests the following requirement: If A and B are large sets, and AN B C C, then C is large. In particular, this entails that if A and B are large, then so is their intersection AN B, while if A is large, then so is any of its supersets CDA. e The empty set @ is not large, or otherwise by the previous require- ment all subsets of N would be large, and so all sequences would be equivalent. Requiring AN B to be large when A and B are large may seem restrictive, but there are natural situations in which all three requirements are fulfilled. One such is when a set A C N is declared to be large if it is cofinite, ie. its complement N — A is finite. This means that A contains “almost ; all” or “ultimately all” members of N. Although this is a plausible notion * of largeness, it is not adequate to our needs. The number system we are constructing is to be linearly ordered, and a natural way to do this, in terms of our general approach, is to take the equivalence class of sequence r to be less than that of s if the set Lys = {nt Tn < Sn} is large. But consider the sequences r = (1,0,1,0,1,0,...), s = (0,1,0,1,0,1,...). Their agreement set is empty, so they determine distinct equivalence classes, one of which should be less than the other. But L,. (the even numbers) is the complement of L,, (the odds), so both are infinite and neither is cofi- nite. Apparently our definition of largeness is going to require the following condition: e For any subset A of N, one of A and N — A is large. The other requirements imply that A and N— A cannot both be large, or else AM (N— A) = @ would be. Thus the large sets are precisely the complements of the ones that are not large. Either the even numbers form a large set or the odd ones do, but they cannot both do so, so which is it to be? Can there in fact be such a notion of largeness, and if so, how do we show it?18 2. Large Sets 2.3 Filters Let I be a nonempty set. The power set of I is the set PUI) ={A: ACT} of all subsets of J. A filter on I is a nonempty collection F C P(Z) of subsets of I satisfying the following axioms: e Intersections: if A,B € F, then ANBEF. e Supersets: if AE F and AC BCI, then BE F. Thus to show B € f, it suffices to show Ai NN An CB, for some n and some Aj,...,An € F. A filter F contains the empty set @ iff F = P(I). We say that F is proper if @ ¢ F. Every filter contains J, and in fact {J} is the smallest filter on J. An ultrafilter is a proper filter that satisfies e for any A CJ, either A € F or A° EF, where AC = I-A. 2.4 Examples of Filters (1) F§ = {ACI:i€ A} is an ultrafilter, called the principal ultrafilter generated by i. If I is finite, then every ultrafilter on J is of the form F’ for some i € I, and so is principal. (2) F° = {ACI:I- Ais finite} is the cofinite, or Fréchet, filter on I, and is proper iff J is infinite. F°° is not an ultrafilter. (3) If 6 # H C P(D, then the filter generated by H, i.e., the smallest filter on J including H, is the collection Ft={ACI:ADB,N---NB, for some n and some B; € H} (cf. Exercise 2.7(4)). For 71 = @ we put F* = {I}. If 1 has a single member B, then F” = {A C I: AD B}, which is called the principal filter generated by B. The ultrafilter F' of Exam- ple (1) is the special case of this when B = {i}. (4 we If {F,, : x € X} is a collection of filters on J that is linearly ordered by set inclusion, in the sense that F, € Fy or Fy C Fz for any x,y € X, then Unex Fe = {A: dee X(A€F,)} is a filter on J.2.6 Zorn’s Lemma 19 2.5 Facts About Filters (1) The filter axioms are equivalent to the requirement that ANBeF iff A,BEF. (2) If F C P(J) satisfies the superset axiom, then F 4 @ iff I ¢ F. Hence {I} C F for any filter F. (3) An ultrafilter F satisfies ANBeF if Ac€Fand Bef, AUBEF if AE€ForBef, AceF iff AGF. (4) Let F be an ultrafilter and {Ai,...,A,} a finite collection of pairwise disjoint (A; A; = @) sets such that A, U---UA, €F. Then A; € F for exactly one i such that 1 p,, and so on. Now, this whole construction cannot go on forever, because eventually we will “run out of” elements of P. At some point we must finish with the desired maximal element. This argument shows what is going on behind the scenes when Zorn’s lemma is applied. Of course the part about running out of elements is vague, and to make it precise we would need to introduce the theory of infinite “ordinal” numbers and “well-orderings” in order to show that we can generate a list of all the elements of P. In many applications, appeal- ing directly to Zorn’s lemma itself allows us to avoid such machinery. For example: Theorem 2.6.1 Any collection of subsets of I that has the finite intersec- tion property can be extended to an ultrafilter on I. Proof. If ‘H has the fip, then the filter 7 generated by F is proper (2.5(7)). Let P be the collection of all proper filters on J that include F™, partially ordered by set inclusion C. Then every linearly ordered subset of P has an upper bound in P, since by 2.4(4) the union of this chain is in P. Hence by Zorn’s lemma P has a maximal element, which is thereby a maximal proper filter on J and thus an ultrafilter by 2.5(6). Oo2.7 Exercises on Filters 21 Corollary 2.6.2 Any infinite set has a nonprincipal ultrafilter on it. Proof. WU I is infinite, the cofinite filter F°° is proper and has the finite intersection property, and so is included in an ultrafilter F. But for any i € I we have I — {i} € F C F, so {i} ¢ F, whereas {i} € F’. Hence F #F*. Thus F is nonprincipal. oO This result is the key fact we need to begin our construction of the hy- perreal number system. We could have simply taken it as an assumption, but there is insight to be gained in showing how it derives from more gen- eral principles like Zorn’s lemma. In fact, a deeper set-theoretic analysis proves that there are as many nonprincipal ultrafilters on an infinite set I as there possibly could be: an ultrafilter is a member of the double power set P(P(I)), and there is a one-to-one correspondence between the set of all nonprincipal ultrafilters on I and P(P(J)) itself. 2.7 Exercises on Filters (1) @AACT, there is an ultrafilter F on J with AE F. (2) There exists a nonprincipal ultrafilter on N containing the set of even numbers, and another containing the set of odd numbers. (3) An ultrafilter on a finite set. must be principal. (4) For H C P(J), let #7 be as defined in Example 2.4(3). (i) Show that F” is a filter that includes H, i.e, HC F™. (ii) Show that 7” is included in any other filter that includes H. (5) Let ¥ be a proper filter on J. (i) Show that FU{A*} has the finite intersection property iff A ¢ F. (ii) Use (i) to deduce that F is an ultrafilter iff it is a maximal proper filter on J.3 Ultrapower Construction of the Hyperreals 3.1 The Ring of Real-Valued Sequences Let N = {1,2,... }, and let R® be the set of all sequences of real numbers. A typical member of R™ has the form r = (r1,r2,r3,... ), which may be denoted more briefly as (r, : 2 € N) or just (rn). For r = (rp) and s = (8,), put rO@s = (H+s8,:nEN), rOs (Tn + 8n:2EN). ll Then (R¥, @, ©) is a commutative ring with zero 0 = (0,0,0,...) and unity 1= (1,1,... ), and additive inverses given by —r=(-r,:n EN). It is not, however, a field, since {1,0,1,0,1,...) © (0,1,0,1,0,...) =0, so the two sequences on the left of this equation are nonzero elements of R® with a zero product; hence neither can have a multiplicative inverse. Indeed, no sequence that has at least one zero term can have such an inverse in RN,